]>
Commit | Line | Data |
---|---|---|
6cbd5570 CM |
1 | /* |
2 | * Copyright (C) 2007 Oracle. All rights reserved. | |
3 | * | |
4 | * This program is free software; you can redistribute it and/or | |
5 | * modify it under the terms of the GNU General Public | |
6 | * License v2 as published by the Free Software Foundation. | |
7 | * | |
8 | * This program is distributed in the hope that it will be useful, | |
9 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
10 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
11 | * General Public License for more details. | |
12 | * | |
13 | * You should have received a copy of the GNU General Public | |
14 | * License along with this program; if not, write to the | |
15 | * Free Software Foundation, Inc., 59 Temple Place - Suite 330, | |
16 | * Boston, MA 021110-1307, USA. | |
17 | */ | |
ec6b910f | 18 | #include <linux/sched.h> |
edbd8d4e | 19 | #include <linux/pagemap.h> |
ec44a35c | 20 | #include <linux/writeback.h> |
74493f7a | 21 | #include "hash.h" |
a5eb62e3 | 22 | #include "crc32c.h" |
fec577fb CM |
23 | #include "ctree.h" |
24 | #include "disk-io.h" | |
25 | #include "print-tree.h" | |
e089f05c | 26 | #include "transaction.h" |
0b86a832 | 27 | #include "volumes.h" |
925baedd | 28 | #include "locking.h" |
fec577fb | 29 | |
0b86a832 | 30 | #define BLOCK_GROUP_DATA EXTENT_WRITEBACK |
96b5179d | 31 | #define BLOCK_GROUP_METADATA EXTENT_UPTODATE |
0b86a832 CM |
32 | #define BLOCK_GROUP_SYSTEM EXTENT_NEW |
33 | ||
96b5179d CM |
34 | #define BLOCK_GROUP_DIRTY EXTENT_DIRTY |
35 | ||
e089f05c CM |
36 | static int finish_current_insert(struct btrfs_trans_handle *trans, struct |
37 | btrfs_root *extent_root); | |
e20d96d6 CM |
38 | static int del_pending_extents(struct btrfs_trans_handle *trans, struct |
39 | btrfs_root *extent_root); | |
925baedd CM |
40 | static struct btrfs_block_group_cache * |
41 | __btrfs_find_block_group(struct btrfs_root *root, | |
42 | struct btrfs_block_group_cache *hint, | |
43 | u64 search_start, int data, int owner); | |
d548ee51 | 44 | |
925baedd CM |
45 | void maybe_lock_mutex(struct btrfs_root *root) |
46 | { | |
47 | if (root != root->fs_info->extent_root && | |
48 | root != root->fs_info->chunk_root && | |
49 | root != root->fs_info->dev_root) { | |
50 | mutex_lock(&root->fs_info->alloc_mutex); | |
51 | } | |
52 | } | |
53 | ||
54 | void maybe_unlock_mutex(struct btrfs_root *root) | |
55 | { | |
56 | if (root != root->fs_info->extent_root && | |
57 | root != root->fs_info->chunk_root && | |
58 | root != root->fs_info->dev_root) { | |
59 | mutex_unlock(&root->fs_info->alloc_mutex); | |
60 | } | |
61 | } | |
fec577fb | 62 | |
e37c9e69 CM |
63 | static int cache_block_group(struct btrfs_root *root, |
64 | struct btrfs_block_group_cache *block_group) | |
65 | { | |
66 | struct btrfs_path *path; | |
67 | int ret; | |
68 | struct btrfs_key key; | |
5f39d397 | 69 | struct extent_buffer *leaf; |
d1310b2e | 70 | struct extent_io_tree *free_space_cache; |
e37c9e69 | 71 | int slot; |
e37c9e69 CM |
72 | u64 last = 0; |
73 | u64 hole_size; | |
7d7d6068 | 74 | u64 first_free; |
e37c9e69 CM |
75 | int found = 0; |
76 | ||
00f5c795 CM |
77 | if (!block_group) |
78 | return 0; | |
79 | ||
e37c9e69 | 80 | root = root->fs_info->extent_root; |
f510cfec | 81 | free_space_cache = &root->fs_info->free_space_cache; |
e37c9e69 CM |
82 | |
83 | if (block_group->cached) | |
84 | return 0; | |
f510cfec | 85 | |
e37c9e69 CM |
86 | path = btrfs_alloc_path(); |
87 | if (!path) | |
88 | return -ENOMEM; | |
7d7d6068 | 89 | |
2cc58cf2 | 90 | path->reada = 2; |
5cd57b2c CM |
91 | /* |
92 | * we get into deadlocks with paths held by callers of this function. | |
93 | * since the alloc_mutex is protecting things right now, just | |
94 | * skip the locking here | |
95 | */ | |
96 | path->skip_locking = 1; | |
7d7d6068 | 97 | first_free = block_group->key.objectid; |
e37c9e69 | 98 | key.objectid = block_group->key.objectid; |
e37c9e69 CM |
99 | key.offset = 0; |
100 | btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY); | |
101 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
102 | if (ret < 0) | |
103 | return ret; | |
0b86a832 | 104 | ret = btrfs_previous_item(root, path, 0, BTRFS_EXTENT_ITEM_KEY); |
d548ee51 Y |
105 | if (ret < 0) |
106 | return ret; | |
107 | if (ret == 0) { | |
108 | leaf = path->nodes[0]; | |
109 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
110 | if (key.objectid + key.offset > first_free) | |
111 | first_free = key.objectid + key.offset; | |
112 | } | |
e37c9e69 | 113 | while(1) { |
5f39d397 | 114 | leaf = path->nodes[0]; |
e37c9e69 | 115 | slot = path->slots[0]; |
5f39d397 | 116 | if (slot >= btrfs_header_nritems(leaf)) { |
e37c9e69 | 117 | ret = btrfs_next_leaf(root, path); |
54aa1f4d CM |
118 | if (ret < 0) |
119 | goto err; | |
de428b63 | 120 | if (ret == 0) { |
e37c9e69 | 121 | continue; |
de428b63 | 122 | } else { |
e37c9e69 CM |
123 | break; |
124 | } | |
125 | } | |
5f39d397 | 126 | btrfs_item_key_to_cpu(leaf, &key, slot); |
7d7d6068 | 127 | if (key.objectid < block_group->key.objectid) { |
7d7d6068 Y |
128 | goto next; |
129 | } | |
e37c9e69 CM |
130 | if (key.objectid >= block_group->key.objectid + |
131 | block_group->key.offset) { | |
e37c9e69 CM |
132 | break; |
133 | } | |
7d7d6068 | 134 | |
e37c9e69 CM |
135 | if (btrfs_key_type(&key) == BTRFS_EXTENT_ITEM_KEY) { |
136 | if (!found) { | |
7d7d6068 | 137 | last = first_free; |
e37c9e69 | 138 | found = 1; |
e37c9e69 | 139 | } |
f510cfec CM |
140 | if (key.objectid > last) { |
141 | hole_size = key.objectid - last; | |
142 | set_extent_dirty(free_space_cache, last, | |
143 | last + hole_size - 1, | |
144 | GFP_NOFS); | |
7d7d6068 Y |
145 | } |
146 | last = key.objectid + key.offset; | |
e37c9e69 | 147 | } |
7d7d6068 | 148 | next: |
e37c9e69 CM |
149 | path->slots[0]++; |
150 | } | |
151 | ||
7d7d6068 Y |
152 | if (!found) |
153 | last = first_free; | |
154 | if (block_group->key.objectid + | |
155 | block_group->key.offset > last) { | |
156 | hole_size = block_group->key.objectid + | |
157 | block_group->key.offset - last; | |
f510cfec CM |
158 | set_extent_dirty(free_space_cache, last, |
159 | last + hole_size - 1, GFP_NOFS); | |
7d7d6068 | 160 | } |
e37c9e69 | 161 | block_group->cached = 1; |
54aa1f4d | 162 | err: |
e37c9e69 CM |
163 | btrfs_free_path(path); |
164 | return 0; | |
165 | } | |
166 | ||
0ef3e66b CM |
167 | struct btrfs_block_group_cache *btrfs_lookup_first_block_group(struct |
168 | btrfs_fs_info *info, | |
169 | u64 bytenr) | |
170 | { | |
171 | struct extent_io_tree *block_group_cache; | |
172 | struct btrfs_block_group_cache *block_group = NULL; | |
173 | u64 ptr; | |
174 | u64 start; | |
175 | u64 end; | |
176 | int ret; | |
177 | ||
178 | bytenr = max_t(u64, bytenr, | |
179 | BTRFS_SUPER_INFO_OFFSET + BTRFS_SUPER_INFO_SIZE); | |
180 | block_group_cache = &info->block_group_cache; | |
181 | ret = find_first_extent_bit(block_group_cache, | |
182 | bytenr, &start, &end, | |
183 | BLOCK_GROUP_DATA | BLOCK_GROUP_METADATA | | |
184 | BLOCK_GROUP_SYSTEM); | |
185 | if (ret) { | |
186 | return NULL; | |
187 | } | |
188 | ret = get_state_private(block_group_cache, start, &ptr); | |
189 | if (ret) | |
190 | return NULL; | |
191 | ||
192 | block_group = (struct btrfs_block_group_cache *)(unsigned long)ptr; | |
193 | return block_group; | |
194 | } | |
195 | ||
5276aeda CM |
196 | struct btrfs_block_group_cache *btrfs_lookup_block_group(struct |
197 | btrfs_fs_info *info, | |
db94535d | 198 | u64 bytenr) |
be744175 | 199 | { |
d1310b2e | 200 | struct extent_io_tree *block_group_cache; |
96b5179d CM |
201 | struct btrfs_block_group_cache *block_group = NULL; |
202 | u64 ptr; | |
203 | u64 start; | |
204 | u64 end; | |
be744175 CM |
205 | int ret; |
206 | ||
a061fc8d CM |
207 | bytenr = max_t(u64, bytenr, |
208 | BTRFS_SUPER_INFO_OFFSET + BTRFS_SUPER_INFO_SIZE); | |
96b5179d CM |
209 | block_group_cache = &info->block_group_cache; |
210 | ret = find_first_extent_bit(block_group_cache, | |
db94535d | 211 | bytenr, &start, &end, |
0b86a832 CM |
212 | BLOCK_GROUP_DATA | BLOCK_GROUP_METADATA | |
213 | BLOCK_GROUP_SYSTEM); | |
be744175 | 214 | if (ret) { |
96b5179d | 215 | return NULL; |
be744175 | 216 | } |
96b5179d CM |
217 | ret = get_state_private(block_group_cache, start, &ptr); |
218 | if (ret) | |
219 | return NULL; | |
220 | ||
ae2f5411 | 221 | block_group = (struct btrfs_block_group_cache *)(unsigned long)ptr; |
5cf66426 | 222 | if (block_group->key.objectid <= bytenr && bytenr < |
96b5179d CM |
223 | block_group->key.objectid + block_group->key.offset) |
224 | return block_group; | |
be744175 CM |
225 | return NULL; |
226 | } | |
0b86a832 CM |
227 | |
228 | static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits) | |
229 | { | |
593060d7 | 230 | return (cache->flags & bits) == bits; |
0b86a832 CM |
231 | } |
232 | ||
233 | static int noinline find_search_start(struct btrfs_root *root, | |
98ed5174 | 234 | struct btrfs_block_group_cache **cache_ret, |
0ef3e66b | 235 | u64 *start_ret, u64 num, int data) |
e37c9e69 | 236 | { |
e37c9e69 CM |
237 | int ret; |
238 | struct btrfs_block_group_cache *cache = *cache_ret; | |
d7fc640e | 239 | struct extent_io_tree *free_space_cache; |
7d1660d4 | 240 | struct extent_state *state; |
e19caa5f | 241 | u64 last; |
f510cfec | 242 | u64 start = 0; |
257d0ce3 | 243 | u64 cache_miss = 0; |
c31f8830 | 244 | u64 total_fs_bytes; |
0b86a832 | 245 | u64 search_start = *start_ret; |
f84a8b36 | 246 | int wrapped = 0; |
e37c9e69 | 247 | |
c31f8830 | 248 | total_fs_bytes = btrfs_super_total_bytes(&root->fs_info->super_copy); |
d7fc640e CM |
249 | free_space_cache = &root->fs_info->free_space_cache; |
250 | ||
0ef3e66b CM |
251 | if (!cache) |
252 | goto out; | |
253 | ||
e37c9e69 | 254 | again: |
54aa1f4d | 255 | ret = cache_block_group(root, cache); |
0ef3e66b | 256 | if (ret) { |
54aa1f4d | 257 | goto out; |
0ef3e66b | 258 | } |
f84a8b36 | 259 | |
e19caa5f | 260 | last = max(search_start, cache->key.objectid); |
0ef3e66b | 261 | if (!block_group_bits(cache, data) || cache->ro) |
0b86a832 | 262 | goto new_group; |
e19caa5f | 263 | |
7d1660d4 CM |
264 | spin_lock_irq(&free_space_cache->lock); |
265 | state = find_first_extent_bit_state(free_space_cache, last, EXTENT_DIRTY); | |
e37c9e69 | 266 | while(1) { |
7d1660d4 | 267 | if (!state) { |
257d0ce3 CM |
268 | if (!cache_miss) |
269 | cache_miss = last; | |
7d1660d4 | 270 | spin_unlock_irq(&free_space_cache->lock); |
e19caa5f CM |
271 | goto new_group; |
272 | } | |
f510cfec | 273 | |
7d1660d4 CM |
274 | start = max(last, state->start); |
275 | last = state->end + 1; | |
257d0ce3 | 276 | if (last - start < num) { |
7d1660d4 CM |
277 | do { |
278 | state = extent_state_next(state); | |
279 | } while(state && !(state->state & EXTENT_DIRTY)); | |
f510cfec | 280 | continue; |
257d0ce3 | 281 | } |
7d1660d4 | 282 | spin_unlock_irq(&free_space_cache->lock); |
0ef3e66b | 283 | if (cache->ro) { |
8f18cf13 | 284 | goto new_group; |
0ef3e66b | 285 | } |
0b86a832 | 286 | if (start + num > cache->key.objectid + cache->key.offset) |
e37c9e69 | 287 | goto new_group; |
8790d502 | 288 | if (!block_group_bits(cache, data)) { |
611f0e00 | 289 | printk("block group bits don't match %Lu %d\n", cache->flags, data); |
8790d502 | 290 | } |
0b86a832 CM |
291 | *start_ret = start; |
292 | return 0; | |
8790d502 CM |
293 | } |
294 | out: | |
1a2b2ac7 CM |
295 | cache = btrfs_lookup_block_group(root->fs_info, search_start); |
296 | if (!cache) { | |
0b86a832 | 297 | printk("Unable to find block group for %Lu\n", search_start); |
1a2b2ac7 | 298 | WARN_ON(1); |
1a2b2ac7 | 299 | } |
0b86a832 | 300 | return -ENOSPC; |
e37c9e69 CM |
301 | |
302 | new_group: | |
e19caa5f | 303 | last = cache->key.objectid + cache->key.offset; |
f84a8b36 | 304 | wrapped: |
0ef3e66b | 305 | cache = btrfs_lookup_first_block_group(root->fs_info, last); |
c31f8830 | 306 | if (!cache || cache->key.objectid >= total_fs_bytes) { |
0e4de584 | 307 | no_cache: |
f84a8b36 CM |
308 | if (!wrapped) { |
309 | wrapped = 1; | |
310 | last = search_start; | |
f84a8b36 CM |
311 | goto wrapped; |
312 | } | |
1a2b2ac7 | 313 | goto out; |
e37c9e69 | 314 | } |
257d0ce3 CM |
315 | if (cache_miss && !cache->cached) { |
316 | cache_block_group(root, cache); | |
317 | last = cache_miss; | |
0ef3e66b | 318 | cache = btrfs_lookup_first_block_group(root->fs_info, last); |
257d0ce3 | 319 | } |
0ef3e66b | 320 | cache_miss = 0; |
925baedd | 321 | cache = __btrfs_find_block_group(root, cache, last, data, 0); |
0e4de584 CM |
322 | if (!cache) |
323 | goto no_cache; | |
e37c9e69 CM |
324 | *cache_ret = cache; |
325 | goto again; | |
326 | } | |
327 | ||
84f54cfa CM |
328 | static u64 div_factor(u64 num, int factor) |
329 | { | |
257d0ce3 CM |
330 | if (factor == 10) |
331 | return num; | |
84f54cfa CM |
332 | num *= factor; |
333 | do_div(num, 10); | |
334 | return num; | |
335 | } | |
336 | ||
6324fbf3 CM |
337 | static int block_group_state_bits(u64 flags) |
338 | { | |
339 | int bits = 0; | |
340 | if (flags & BTRFS_BLOCK_GROUP_DATA) | |
341 | bits |= BLOCK_GROUP_DATA; | |
342 | if (flags & BTRFS_BLOCK_GROUP_METADATA) | |
343 | bits |= BLOCK_GROUP_METADATA; | |
344 | if (flags & BTRFS_BLOCK_GROUP_SYSTEM) | |
345 | bits |= BLOCK_GROUP_SYSTEM; | |
346 | return bits; | |
347 | } | |
348 | ||
925baedd CM |
349 | static struct btrfs_block_group_cache * |
350 | __btrfs_find_block_group(struct btrfs_root *root, | |
351 | struct btrfs_block_group_cache *hint, | |
352 | u64 search_start, int data, int owner) | |
cd1bc465 | 353 | { |
96b5179d | 354 | struct btrfs_block_group_cache *cache; |
d1310b2e | 355 | struct extent_io_tree *block_group_cache; |
31f3c99b | 356 | struct btrfs_block_group_cache *found_group = NULL; |
cd1bc465 CM |
357 | struct btrfs_fs_info *info = root->fs_info; |
358 | u64 used; | |
31f3c99b | 359 | u64 last = 0; |
96b5179d CM |
360 | u64 start; |
361 | u64 end; | |
362 | u64 free_check; | |
363 | u64 ptr; | |
364 | int bit; | |
cd1bc465 | 365 | int ret; |
31f3c99b | 366 | int full_search = 0; |
bce4eae9 | 367 | int factor = 10; |
0ef3e66b | 368 | int wrapped = 0; |
de428b63 | 369 | |
96b5179d CM |
370 | block_group_cache = &info->block_group_cache; |
371 | ||
a236aed1 CM |
372 | if (data & BTRFS_BLOCK_GROUP_METADATA) |
373 | factor = 9; | |
be744175 | 374 | |
6324fbf3 | 375 | bit = block_group_state_bits(data); |
be744175 | 376 | |
0ef3e66b | 377 | if (search_start) { |
be744175 | 378 | struct btrfs_block_group_cache *shint; |
0ef3e66b | 379 | shint = btrfs_lookup_first_block_group(info, search_start); |
8f18cf13 | 380 | if (shint && block_group_bits(shint, data) && !shint->ro) { |
be744175 | 381 | used = btrfs_block_group_used(&shint->item); |
324ae4df Y |
382 | if (used + shint->pinned < |
383 | div_factor(shint->key.offset, factor)) { | |
be744175 CM |
384 | return shint; |
385 | } | |
386 | } | |
387 | } | |
0ef3e66b | 388 | if (hint && !hint->ro && block_group_bits(hint, data)) { |
31f3c99b | 389 | used = btrfs_block_group_used(&hint->item); |
324ae4df Y |
390 | if (used + hint->pinned < |
391 | div_factor(hint->key.offset, factor)) { | |
31f3c99b CM |
392 | return hint; |
393 | } | |
e19caa5f | 394 | last = hint->key.objectid + hint->key.offset; |
31f3c99b | 395 | } else { |
e37c9e69 | 396 | if (hint) |
0ef3e66b | 397 | last = max(hint->key.objectid, search_start); |
e37c9e69 | 398 | else |
0ef3e66b | 399 | last = search_start; |
31f3c99b | 400 | } |
31f3c99b | 401 | again: |
cd1bc465 | 402 | while(1) { |
96b5179d CM |
403 | ret = find_first_extent_bit(block_group_cache, last, |
404 | &start, &end, bit); | |
405 | if (ret) | |
cd1bc465 | 406 | break; |
96b5179d CM |
407 | |
408 | ret = get_state_private(block_group_cache, start, &ptr); | |
0ef3e66b CM |
409 | if (ret) { |
410 | last = end + 1; | |
411 | continue; | |
412 | } | |
96b5179d | 413 | |
ae2f5411 | 414 | cache = (struct btrfs_block_group_cache *)(unsigned long)ptr; |
96b5179d CM |
415 | last = cache->key.objectid + cache->key.offset; |
416 | used = btrfs_block_group_used(&cache->item); | |
417 | ||
8f18cf13 | 418 | if (!cache->ro && block_group_bits(cache, data)) { |
0ef3e66b | 419 | free_check = div_factor(cache->key.offset, factor); |
8790d502 CM |
420 | if (used + cache->pinned < free_check) { |
421 | found_group = cache; | |
422 | goto found; | |
423 | } | |
6324fbf3 | 424 | } |
de428b63 | 425 | cond_resched(); |
cd1bc465 | 426 | } |
0ef3e66b CM |
427 | if (!wrapped) { |
428 | last = search_start; | |
429 | wrapped = 1; | |
430 | goto again; | |
431 | } | |
432 | if (!full_search && factor < 10) { | |
be744175 | 433 | last = search_start; |
31f3c99b | 434 | full_search = 1; |
0ef3e66b | 435 | factor = 10; |
31f3c99b CM |
436 | goto again; |
437 | } | |
be744175 | 438 | found: |
31f3c99b | 439 | return found_group; |
cd1bc465 CM |
440 | } |
441 | ||
925baedd CM |
442 | struct btrfs_block_group_cache *btrfs_find_block_group(struct btrfs_root *root, |
443 | struct btrfs_block_group_cache | |
444 | *hint, u64 search_start, | |
445 | int data, int owner) | |
446 | { | |
447 | ||
448 | struct btrfs_block_group_cache *ret; | |
449 | mutex_lock(&root->fs_info->alloc_mutex); | |
450 | ret = __btrfs_find_block_group(root, hint, search_start, data, owner); | |
451 | mutex_unlock(&root->fs_info->alloc_mutex); | |
452 | return ret; | |
453 | } | |
7bb86316 | 454 | static u64 hash_extent_ref(u64 root_objectid, u64 ref_generation, |
74493f7a CM |
455 | u64 owner, u64 owner_offset) |
456 | { | |
457 | u32 high_crc = ~(u32)0; | |
458 | u32 low_crc = ~(u32)0; | |
459 | __le64 lenum; | |
74493f7a | 460 | lenum = cpu_to_le64(root_objectid); |
a5eb62e3 | 461 | high_crc = btrfs_crc32c(high_crc, &lenum, sizeof(lenum)); |
7bb86316 | 462 | lenum = cpu_to_le64(ref_generation); |
a5eb62e3 | 463 | low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum)); |
21a4989d CM |
464 | if (owner >= BTRFS_FIRST_FREE_OBJECTID) { |
465 | lenum = cpu_to_le64(owner); | |
a5eb62e3 | 466 | low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum)); |
21a4989d | 467 | lenum = cpu_to_le64(owner_offset); |
a5eb62e3 | 468 | low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum)); |
21a4989d | 469 | } |
74493f7a CM |
470 | return ((u64)high_crc << 32) | (u64)low_crc; |
471 | } | |
472 | ||
7bb86316 CM |
473 | static int match_extent_ref(struct extent_buffer *leaf, |
474 | struct btrfs_extent_ref *disk_ref, | |
475 | struct btrfs_extent_ref *cpu_ref) | |
476 | { | |
477 | int ret; | |
478 | int len; | |
479 | ||
480 | if (cpu_ref->objectid) | |
481 | len = sizeof(*cpu_ref); | |
482 | else | |
483 | len = 2 * sizeof(u64); | |
484 | ret = memcmp_extent_buffer(leaf, cpu_ref, (unsigned long)disk_ref, | |
485 | len); | |
486 | return ret == 0; | |
487 | } | |
488 | ||
98ed5174 CM |
489 | static int noinline lookup_extent_backref(struct btrfs_trans_handle *trans, |
490 | struct btrfs_root *root, | |
491 | struct btrfs_path *path, u64 bytenr, | |
492 | u64 root_objectid, | |
493 | u64 ref_generation, u64 owner, | |
494 | u64 owner_offset, int del) | |
74493f7a CM |
495 | { |
496 | u64 hash; | |
497 | struct btrfs_key key; | |
7bb86316 | 498 | struct btrfs_key found_key; |
74493f7a | 499 | struct btrfs_extent_ref ref; |
7bb86316 CM |
500 | struct extent_buffer *leaf; |
501 | struct btrfs_extent_ref *disk_ref; | |
502 | int ret; | |
503 | int ret2; | |
504 | ||
505 | btrfs_set_stack_ref_root(&ref, root_objectid); | |
506 | btrfs_set_stack_ref_generation(&ref, ref_generation); | |
507 | btrfs_set_stack_ref_objectid(&ref, owner); | |
508 | btrfs_set_stack_ref_offset(&ref, owner_offset); | |
509 | ||
510 | hash = hash_extent_ref(root_objectid, ref_generation, owner, | |
511 | owner_offset); | |
512 | key.offset = hash; | |
513 | key.objectid = bytenr; | |
514 | key.type = BTRFS_EXTENT_REF_KEY; | |
515 | ||
516 | while (1) { | |
517 | ret = btrfs_search_slot(trans, root, &key, path, | |
518 | del ? -1 : 0, del); | |
519 | if (ret < 0) | |
520 | goto out; | |
521 | leaf = path->nodes[0]; | |
522 | if (ret != 0) { | |
523 | u32 nritems = btrfs_header_nritems(leaf); | |
524 | if (path->slots[0] >= nritems) { | |
525 | ret2 = btrfs_next_leaf(root, path); | |
526 | if (ret2) | |
527 | goto out; | |
528 | leaf = path->nodes[0]; | |
529 | } | |
530 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
531 | if (found_key.objectid != bytenr || | |
532 | found_key.type != BTRFS_EXTENT_REF_KEY) | |
533 | goto out; | |
534 | key.offset = found_key.offset; | |
535 | if (del) { | |
536 | btrfs_release_path(root, path); | |
537 | continue; | |
538 | } | |
539 | } | |
540 | disk_ref = btrfs_item_ptr(path->nodes[0], | |
541 | path->slots[0], | |
542 | struct btrfs_extent_ref); | |
543 | if (match_extent_ref(path->nodes[0], disk_ref, &ref)) { | |
544 | ret = 0; | |
545 | goto out; | |
546 | } | |
547 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
548 | key.offset = found_key.offset + 1; | |
549 | btrfs_release_path(root, path); | |
550 | } | |
551 | out: | |
552 | return ret; | |
553 | } | |
554 | ||
d8d5f3e1 CM |
555 | /* |
556 | * Back reference rules. Back refs have three main goals: | |
557 | * | |
558 | * 1) differentiate between all holders of references to an extent so that | |
559 | * when a reference is dropped we can make sure it was a valid reference | |
560 | * before freeing the extent. | |
561 | * | |
562 | * 2) Provide enough information to quickly find the holders of an extent | |
563 | * if we notice a given block is corrupted or bad. | |
564 | * | |
565 | * 3) Make it easy to migrate blocks for FS shrinking or storage pool | |
566 | * maintenance. This is actually the same as #2, but with a slightly | |
567 | * different use case. | |
568 | * | |
569 | * File extents can be referenced by: | |
570 | * | |
571 | * - multiple snapshots, subvolumes, or different generations in one subvol | |
572 | * - different files inside a single subvolume (in theory, not implemented yet) | |
573 | * - different offsets inside a file (bookend extents in file.c) | |
574 | * | |
575 | * The extent ref structure has fields for: | |
576 | * | |
577 | * - Objectid of the subvolume root | |
578 | * - Generation number of the tree holding the reference | |
579 | * - objectid of the file holding the reference | |
580 | * - offset in the file corresponding to the key holding the reference | |
581 | * | |
582 | * When a file extent is allocated the fields are filled in: | |
583 | * (root_key.objectid, trans->transid, inode objectid, offset in file) | |
584 | * | |
585 | * When a leaf is cow'd new references are added for every file extent found | |
586 | * in the leaf. It looks the same as the create case, but trans->transid | |
587 | * will be different when the block is cow'd. | |
588 | * | |
589 | * (root_key.objectid, trans->transid, inode objectid, offset in file) | |
590 | * | |
591 | * When a file extent is removed either during snapshot deletion or file | |
592 | * truncation, the corresponding back reference is found | |
593 | * by searching for: | |
594 | * | |
595 | * (btrfs_header_owner(leaf), btrfs_header_generation(leaf), | |
596 | * inode objectid, offset in file) | |
597 | * | |
598 | * Btree extents can be referenced by: | |
599 | * | |
600 | * - Different subvolumes | |
601 | * - Different generations of the same subvolume | |
602 | * | |
603 | * Storing sufficient information for a full reverse mapping of a btree | |
604 | * block would require storing the lowest key of the block in the backref, | |
605 | * and it would require updating that lowest key either before write out or | |
606 | * every time it changed. Instead, the objectid of the lowest key is stored | |
607 | * along with the level of the tree block. This provides a hint | |
608 | * about where in the btree the block can be found. Searches through the | |
609 | * btree only need to look for a pointer to that block, so they stop one | |
610 | * level higher than the level recorded in the backref. | |
611 | * | |
612 | * Some btrees do not do reference counting on their extents. These | |
613 | * include the extent tree and the tree of tree roots. Backrefs for these | |
614 | * trees always have a generation of zero. | |
615 | * | |
616 | * When a tree block is created, back references are inserted: | |
617 | * | |
f6dbff55 | 618 | * (root->root_key.objectid, trans->transid or zero, level, lowest_key_objectid) |
d8d5f3e1 CM |
619 | * |
620 | * When a tree block is cow'd in a reference counted root, | |
621 | * new back references are added for all the blocks it points to. | |
622 | * These are of the form (trans->transid will have increased since creation): | |
623 | * | |
f6dbff55 | 624 | * (root->root_key.objectid, trans->transid, level, lowest_key_objectid) |
d8d5f3e1 CM |
625 | * |
626 | * Because the lowest_key_objectid and the level are just hints | |
627 | * they are not used when backrefs are deleted. When a backref is deleted: | |
628 | * | |
629 | * if backref was for a tree root: | |
630 | * root_objectid = root->root_key.objectid | |
631 | * else | |
632 | * root_objectid = btrfs_header_owner(parent) | |
633 | * | |
634 | * (root_objectid, btrfs_header_generation(parent) or zero, 0, 0) | |
635 | * | |
636 | * Back Reference Key hashing: | |
637 | * | |
638 | * Back references have four fields, each 64 bits long. Unfortunately, | |
639 | * This is hashed into a single 64 bit number and placed into the key offset. | |
640 | * The key objectid corresponds to the first byte in the extent, and the | |
641 | * key type is set to BTRFS_EXTENT_REF_KEY | |
642 | */ | |
7bb86316 CM |
643 | int btrfs_insert_extent_backref(struct btrfs_trans_handle *trans, |
644 | struct btrfs_root *root, | |
645 | struct btrfs_path *path, u64 bytenr, | |
646 | u64 root_objectid, u64 ref_generation, | |
647 | u64 owner, u64 owner_offset) | |
648 | { | |
649 | u64 hash; | |
650 | struct btrfs_key key; | |
651 | struct btrfs_extent_ref ref; | |
652 | struct btrfs_extent_ref *disk_ref; | |
74493f7a CM |
653 | int ret; |
654 | ||
655 | btrfs_set_stack_ref_root(&ref, root_objectid); | |
7bb86316 | 656 | btrfs_set_stack_ref_generation(&ref, ref_generation); |
74493f7a CM |
657 | btrfs_set_stack_ref_objectid(&ref, owner); |
658 | btrfs_set_stack_ref_offset(&ref, owner_offset); | |
659 | ||
7bb86316 CM |
660 | hash = hash_extent_ref(root_objectid, ref_generation, owner, |
661 | owner_offset); | |
74493f7a CM |
662 | key.offset = hash; |
663 | key.objectid = bytenr; | |
664 | key.type = BTRFS_EXTENT_REF_KEY; | |
665 | ||
666 | ret = btrfs_insert_empty_item(trans, root, path, &key, sizeof(ref)); | |
667 | while (ret == -EEXIST) { | |
7bb86316 CM |
668 | disk_ref = btrfs_item_ptr(path->nodes[0], path->slots[0], |
669 | struct btrfs_extent_ref); | |
670 | if (match_extent_ref(path->nodes[0], disk_ref, &ref)) | |
671 | goto out; | |
672 | key.offset++; | |
673 | btrfs_release_path(root, path); | |
674 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
675 | sizeof(ref)); | |
74493f7a | 676 | } |
7bb86316 CM |
677 | if (ret) |
678 | goto out; | |
679 | disk_ref = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
680 | struct btrfs_extent_ref); | |
681 | write_extent_buffer(path->nodes[0], &ref, (unsigned long)disk_ref, | |
682 | sizeof(ref)); | |
683 | btrfs_mark_buffer_dirty(path->nodes[0]); | |
684 | out: | |
685 | btrfs_release_path(root, path); | |
686 | return ret; | |
74493f7a CM |
687 | } |
688 | ||
925baedd | 689 | static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans, |
b18c6685 | 690 | struct btrfs_root *root, |
74493f7a | 691 | u64 bytenr, u64 num_bytes, |
7bb86316 | 692 | u64 root_objectid, u64 ref_generation, |
74493f7a | 693 | u64 owner, u64 owner_offset) |
02217ed2 | 694 | { |
5caf2a00 | 695 | struct btrfs_path *path; |
02217ed2 | 696 | int ret; |
e2fa7227 | 697 | struct btrfs_key key; |
5f39d397 | 698 | struct extent_buffer *l; |
234b63a0 | 699 | struct btrfs_extent_item *item; |
cf27e1ee | 700 | u32 refs; |
037e6390 | 701 | |
db94535d | 702 | WARN_ON(num_bytes < root->sectorsize); |
5caf2a00 | 703 | path = btrfs_alloc_path(); |
54aa1f4d CM |
704 | if (!path) |
705 | return -ENOMEM; | |
26b8003f | 706 | |
3c12ac72 | 707 | path->reada = 1; |
db94535d | 708 | key.objectid = bytenr; |
62e2749e | 709 | btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY); |
db94535d | 710 | key.offset = num_bytes; |
5caf2a00 | 711 | ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path, |
9f5fae2f | 712 | 0, 1); |
54aa1f4d CM |
713 | if (ret < 0) |
714 | return ret; | |
a429e513 | 715 | if (ret != 0) { |
a28ec197 | 716 | BUG(); |
a429e513 | 717 | } |
02217ed2 | 718 | BUG_ON(ret != 0); |
5f39d397 | 719 | l = path->nodes[0]; |
5caf2a00 | 720 | item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item); |
5f39d397 CM |
721 | refs = btrfs_extent_refs(l, item); |
722 | btrfs_set_extent_refs(l, item, refs + 1); | |
5caf2a00 | 723 | btrfs_mark_buffer_dirty(path->nodes[0]); |
a28ec197 | 724 | |
5caf2a00 | 725 | btrfs_release_path(root->fs_info->extent_root, path); |
7bb86316 | 726 | |
3c12ac72 | 727 | path->reada = 1; |
7bb86316 CM |
728 | ret = btrfs_insert_extent_backref(trans, root->fs_info->extent_root, |
729 | path, bytenr, root_objectid, | |
730 | ref_generation, owner, owner_offset); | |
731 | BUG_ON(ret); | |
9f5fae2f | 732 | finish_current_insert(trans, root->fs_info->extent_root); |
e20d96d6 | 733 | del_pending_extents(trans, root->fs_info->extent_root); |
74493f7a CM |
734 | |
735 | btrfs_free_path(path); | |
02217ed2 CM |
736 | return 0; |
737 | } | |
738 | ||
925baedd CM |
739 | int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans, |
740 | struct btrfs_root *root, | |
741 | u64 bytenr, u64 num_bytes, | |
742 | u64 root_objectid, u64 ref_generation, | |
743 | u64 owner, u64 owner_offset) | |
744 | { | |
745 | int ret; | |
746 | ||
747 | mutex_lock(&root->fs_info->alloc_mutex); | |
748 | ret = __btrfs_inc_extent_ref(trans, root, bytenr, num_bytes, | |
749 | root_objectid, ref_generation, | |
750 | owner, owner_offset); | |
751 | mutex_unlock(&root->fs_info->alloc_mutex); | |
752 | return ret; | |
753 | } | |
754 | ||
e9d0b13b CM |
755 | int btrfs_extent_post_op(struct btrfs_trans_handle *trans, |
756 | struct btrfs_root *root) | |
757 | { | |
758 | finish_current_insert(trans, root->fs_info->extent_root); | |
759 | del_pending_extents(trans, root->fs_info->extent_root); | |
760 | return 0; | |
761 | } | |
762 | ||
b18c6685 | 763 | static int lookup_extent_ref(struct btrfs_trans_handle *trans, |
db94535d CM |
764 | struct btrfs_root *root, u64 bytenr, |
765 | u64 num_bytes, u32 *refs) | |
a28ec197 | 766 | { |
5caf2a00 | 767 | struct btrfs_path *path; |
a28ec197 | 768 | int ret; |
e2fa7227 | 769 | struct btrfs_key key; |
5f39d397 | 770 | struct extent_buffer *l; |
234b63a0 | 771 | struct btrfs_extent_item *item; |
5caf2a00 | 772 | |
db94535d | 773 | WARN_ON(num_bytes < root->sectorsize); |
5caf2a00 | 774 | path = btrfs_alloc_path(); |
3c12ac72 | 775 | path->reada = 1; |
db94535d CM |
776 | key.objectid = bytenr; |
777 | key.offset = num_bytes; | |
62e2749e | 778 | btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY); |
5caf2a00 | 779 | ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path, |
9f5fae2f | 780 | 0, 0); |
54aa1f4d CM |
781 | if (ret < 0) |
782 | goto out; | |
5f39d397 CM |
783 | if (ret != 0) { |
784 | btrfs_print_leaf(root, path->nodes[0]); | |
db94535d | 785 | printk("failed to find block number %Lu\n", bytenr); |
a28ec197 | 786 | BUG(); |
5f39d397 CM |
787 | } |
788 | l = path->nodes[0]; | |
5caf2a00 | 789 | item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item); |
5f39d397 | 790 | *refs = btrfs_extent_refs(l, item); |
54aa1f4d | 791 | out: |
5caf2a00 | 792 | btrfs_free_path(path); |
a28ec197 CM |
793 | return 0; |
794 | } | |
795 | ||
be20aa9d CM |
796 | u32 btrfs_count_snapshots_in_path(struct btrfs_root *root, |
797 | struct btrfs_path *count_path, | |
a68d5933 | 798 | u64 expected_owner, |
be20aa9d CM |
799 | u64 first_extent) |
800 | { | |
801 | struct btrfs_root *extent_root = root->fs_info->extent_root; | |
802 | struct btrfs_path *path; | |
803 | u64 bytenr; | |
804 | u64 found_objectid; | |
a68d5933 | 805 | u64 found_owner; |
56b453c9 | 806 | u64 root_objectid = root->root_key.objectid; |
be20aa9d | 807 | u32 total_count = 0; |
bbaf549e | 808 | u32 extent_refs; |
be20aa9d | 809 | u32 cur_count; |
be20aa9d CM |
810 | u32 nritems; |
811 | int ret; | |
812 | struct btrfs_key key; | |
813 | struct btrfs_key found_key; | |
814 | struct extent_buffer *l; | |
815 | struct btrfs_extent_item *item; | |
816 | struct btrfs_extent_ref *ref_item; | |
817 | int level = -1; | |
818 | ||
925baedd CM |
819 | /* FIXME, needs locking */ |
820 | BUG(); | |
821 | ||
822 | mutex_lock(&root->fs_info->alloc_mutex); | |
be20aa9d CM |
823 | path = btrfs_alloc_path(); |
824 | again: | |
825 | if (level == -1) | |
826 | bytenr = first_extent; | |
827 | else | |
828 | bytenr = count_path->nodes[level]->start; | |
829 | ||
830 | cur_count = 0; | |
831 | key.objectid = bytenr; | |
832 | key.offset = 0; | |
833 | ||
834 | btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY); | |
835 | ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0); | |
836 | if (ret < 0) | |
837 | goto out; | |
838 | BUG_ON(ret == 0); | |
839 | ||
840 | l = path->nodes[0]; | |
841 | btrfs_item_key_to_cpu(l, &found_key, path->slots[0]); | |
842 | ||
843 | if (found_key.objectid != bytenr || | |
844 | found_key.type != BTRFS_EXTENT_ITEM_KEY) { | |
845 | goto out; | |
846 | } | |
847 | ||
848 | item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item); | |
bbaf549e | 849 | extent_refs = btrfs_extent_refs(l, item); |
be20aa9d | 850 | while (1) { |
bd09835d | 851 | l = path->nodes[0]; |
be20aa9d CM |
852 | nritems = btrfs_header_nritems(l); |
853 | if (path->slots[0] >= nritems) { | |
854 | ret = btrfs_next_leaf(extent_root, path); | |
855 | if (ret == 0) | |
856 | continue; | |
857 | break; | |
858 | } | |
859 | btrfs_item_key_to_cpu(l, &found_key, path->slots[0]); | |
860 | if (found_key.objectid != bytenr) | |
861 | break; | |
bd09835d | 862 | |
be20aa9d CM |
863 | if (found_key.type != BTRFS_EXTENT_REF_KEY) { |
864 | path->slots[0]++; | |
865 | continue; | |
866 | } | |
867 | ||
868 | cur_count++; | |
869 | ref_item = btrfs_item_ptr(l, path->slots[0], | |
870 | struct btrfs_extent_ref); | |
871 | found_objectid = btrfs_ref_root(l, ref_item); | |
872 | ||
56b453c9 CM |
873 | if (found_objectid != root_objectid) { |
874 | total_count = 2; | |
4313b399 | 875 | goto out; |
56b453c9 | 876 | } |
a68d5933 CM |
877 | if (level == -1) { |
878 | found_owner = btrfs_ref_objectid(l, ref_item); | |
879 | if (found_owner != expected_owner) { | |
880 | total_count = 2; | |
881 | goto out; | |
882 | } | |
bbaf549e CM |
883 | /* |
884 | * nasty. we don't count a reference held by | |
885 | * the running transaction. This allows nodatacow | |
886 | * to avoid cow most of the time | |
887 | */ | |
888 | if (found_owner >= BTRFS_FIRST_FREE_OBJECTID && | |
889 | btrfs_ref_generation(l, ref_item) == | |
890 | root->fs_info->generation) { | |
891 | extent_refs--; | |
892 | } | |
a68d5933 | 893 | } |
56b453c9 | 894 | total_count = 1; |
be20aa9d CM |
895 | path->slots[0]++; |
896 | } | |
bbaf549e CM |
897 | /* |
898 | * if there is more than one reference against a data extent, | |
899 | * we have to assume the other ref is another snapshot | |
900 | */ | |
901 | if (level == -1 && extent_refs > 1) { | |
902 | total_count = 2; | |
903 | goto out; | |
904 | } | |
be20aa9d CM |
905 | if (cur_count == 0) { |
906 | total_count = 0; | |
907 | goto out; | |
908 | } | |
be20aa9d CM |
909 | if (level >= 0 && root->node == count_path->nodes[level]) |
910 | goto out; | |
911 | level++; | |
912 | btrfs_release_path(root, path); | |
913 | goto again; | |
914 | ||
915 | out: | |
916 | btrfs_free_path(path); | |
925baedd | 917 | mutex_unlock(&root->fs_info->alloc_mutex); |
be20aa9d | 918 | return total_count; |
be20aa9d | 919 | } |
c5739bba | 920 | |
e089f05c | 921 | int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root, |
5f39d397 | 922 | struct extent_buffer *buf) |
02217ed2 | 923 | { |
db94535d | 924 | u64 bytenr; |
5f39d397 CM |
925 | u32 nritems; |
926 | struct btrfs_key key; | |
6407bf6d | 927 | struct btrfs_file_extent_item *fi; |
02217ed2 | 928 | int i; |
db94535d | 929 | int level; |
6407bf6d | 930 | int ret; |
54aa1f4d | 931 | int faili; |
a28ec197 | 932 | |
3768f368 | 933 | if (!root->ref_cows) |
a28ec197 | 934 | return 0; |
5f39d397 | 935 | |
925baedd | 936 | mutex_lock(&root->fs_info->alloc_mutex); |
db94535d | 937 | level = btrfs_header_level(buf); |
5f39d397 CM |
938 | nritems = btrfs_header_nritems(buf); |
939 | for (i = 0; i < nritems; i++) { | |
db94535d CM |
940 | if (level == 0) { |
941 | u64 disk_bytenr; | |
5f39d397 CM |
942 | btrfs_item_key_to_cpu(buf, &key, i); |
943 | if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY) | |
6407bf6d | 944 | continue; |
5f39d397 | 945 | fi = btrfs_item_ptr(buf, i, |
6407bf6d | 946 | struct btrfs_file_extent_item); |
5f39d397 | 947 | if (btrfs_file_extent_type(buf, fi) == |
236454df CM |
948 | BTRFS_FILE_EXTENT_INLINE) |
949 | continue; | |
db94535d CM |
950 | disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi); |
951 | if (disk_bytenr == 0) | |
3a686375 | 952 | continue; |
925baedd | 953 | ret = __btrfs_inc_extent_ref(trans, root, disk_bytenr, |
7bb86316 CM |
954 | btrfs_file_extent_disk_num_bytes(buf, fi), |
955 | root->root_key.objectid, trans->transid, | |
956 | key.objectid, key.offset); | |
54aa1f4d CM |
957 | if (ret) { |
958 | faili = i; | |
959 | goto fail; | |
960 | } | |
6407bf6d | 961 | } else { |
db94535d | 962 | bytenr = btrfs_node_blockptr(buf, i); |
6caab489 | 963 | btrfs_node_key_to_cpu(buf, &key, i); |
925baedd | 964 | ret = __btrfs_inc_extent_ref(trans, root, bytenr, |
7bb86316 CM |
965 | btrfs_level_size(root, level - 1), |
966 | root->root_key.objectid, | |
f6dbff55 CM |
967 | trans->transid, |
968 | level - 1, key.objectid); | |
54aa1f4d CM |
969 | if (ret) { |
970 | faili = i; | |
971 | goto fail; | |
972 | } | |
6407bf6d | 973 | } |
02217ed2 | 974 | } |
925baedd | 975 | mutex_unlock(&root->fs_info->alloc_mutex); |
02217ed2 | 976 | return 0; |
54aa1f4d | 977 | fail: |
ccd467d6 | 978 | WARN_ON(1); |
7bb86316 | 979 | #if 0 |
54aa1f4d | 980 | for (i =0; i < faili; i++) { |
db94535d CM |
981 | if (level == 0) { |
982 | u64 disk_bytenr; | |
5f39d397 CM |
983 | btrfs_item_key_to_cpu(buf, &key, i); |
984 | if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY) | |
54aa1f4d | 985 | continue; |
5f39d397 | 986 | fi = btrfs_item_ptr(buf, i, |
54aa1f4d | 987 | struct btrfs_file_extent_item); |
5f39d397 | 988 | if (btrfs_file_extent_type(buf, fi) == |
54aa1f4d CM |
989 | BTRFS_FILE_EXTENT_INLINE) |
990 | continue; | |
db94535d CM |
991 | disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi); |
992 | if (disk_bytenr == 0) | |
54aa1f4d | 993 | continue; |
db94535d CM |
994 | err = btrfs_free_extent(trans, root, disk_bytenr, |
995 | btrfs_file_extent_disk_num_bytes(buf, | |
5f39d397 | 996 | fi), 0); |
54aa1f4d CM |
997 | BUG_ON(err); |
998 | } else { | |
db94535d CM |
999 | bytenr = btrfs_node_blockptr(buf, i); |
1000 | err = btrfs_free_extent(trans, root, bytenr, | |
1001 | btrfs_level_size(root, level - 1), 0); | |
54aa1f4d CM |
1002 | BUG_ON(err); |
1003 | } | |
1004 | } | |
7bb86316 | 1005 | #endif |
925baedd | 1006 | mutex_unlock(&root->fs_info->alloc_mutex); |
54aa1f4d | 1007 | return ret; |
02217ed2 CM |
1008 | } |
1009 | ||
9078a3e1 CM |
1010 | static int write_one_cache_group(struct btrfs_trans_handle *trans, |
1011 | struct btrfs_root *root, | |
1012 | struct btrfs_path *path, | |
1013 | struct btrfs_block_group_cache *cache) | |
1014 | { | |
1015 | int ret; | |
1016 | int pending_ret; | |
1017 | struct btrfs_root *extent_root = root->fs_info->extent_root; | |
5f39d397 CM |
1018 | unsigned long bi; |
1019 | struct extent_buffer *leaf; | |
9078a3e1 | 1020 | |
9078a3e1 | 1021 | ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1); |
54aa1f4d CM |
1022 | if (ret < 0) |
1023 | goto fail; | |
9078a3e1 | 1024 | BUG_ON(ret); |
5f39d397 CM |
1025 | |
1026 | leaf = path->nodes[0]; | |
1027 | bi = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
1028 | write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item)); | |
1029 | btrfs_mark_buffer_dirty(leaf); | |
9078a3e1 | 1030 | btrfs_release_path(extent_root, path); |
54aa1f4d | 1031 | fail: |
9078a3e1 CM |
1032 | finish_current_insert(trans, extent_root); |
1033 | pending_ret = del_pending_extents(trans, extent_root); | |
1034 | if (ret) | |
1035 | return ret; | |
1036 | if (pending_ret) | |
1037 | return pending_ret; | |
1038 | return 0; | |
1039 | ||
1040 | } | |
1041 | ||
96b5179d CM |
1042 | int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans, |
1043 | struct btrfs_root *root) | |
9078a3e1 | 1044 | { |
d1310b2e | 1045 | struct extent_io_tree *block_group_cache; |
96b5179d | 1046 | struct btrfs_block_group_cache *cache; |
9078a3e1 CM |
1047 | int ret; |
1048 | int err = 0; | |
1049 | int werr = 0; | |
9078a3e1 | 1050 | struct btrfs_path *path; |
96b5179d CM |
1051 | u64 last = 0; |
1052 | u64 start; | |
1053 | u64 end; | |
1054 | u64 ptr; | |
9078a3e1 | 1055 | |
96b5179d | 1056 | block_group_cache = &root->fs_info->block_group_cache; |
9078a3e1 CM |
1057 | path = btrfs_alloc_path(); |
1058 | if (!path) | |
1059 | return -ENOMEM; | |
1060 | ||
925baedd | 1061 | mutex_lock(&root->fs_info->alloc_mutex); |
9078a3e1 | 1062 | while(1) { |
96b5179d CM |
1063 | ret = find_first_extent_bit(block_group_cache, last, |
1064 | &start, &end, BLOCK_GROUP_DIRTY); | |
1065 | if (ret) | |
9078a3e1 | 1066 | break; |
54aa1f4d | 1067 | |
96b5179d CM |
1068 | last = end + 1; |
1069 | ret = get_state_private(block_group_cache, start, &ptr); | |
1070 | if (ret) | |
1071 | break; | |
ae2f5411 | 1072 | cache = (struct btrfs_block_group_cache *)(unsigned long)ptr; |
96b5179d CM |
1073 | err = write_one_cache_group(trans, root, |
1074 | path, cache); | |
1075 | /* | |
1076 | * if we fail to write the cache group, we want | |
1077 | * to keep it marked dirty in hopes that a later | |
1078 | * write will work | |
1079 | */ | |
1080 | if (err) { | |
1081 | werr = err; | |
1082 | continue; | |
9078a3e1 | 1083 | } |
96b5179d CM |
1084 | clear_extent_bits(block_group_cache, start, end, |
1085 | BLOCK_GROUP_DIRTY, GFP_NOFS); | |
9078a3e1 CM |
1086 | } |
1087 | btrfs_free_path(path); | |
925baedd | 1088 | mutex_unlock(&root->fs_info->alloc_mutex); |
9078a3e1 CM |
1089 | return werr; |
1090 | } | |
1091 | ||
6324fbf3 CM |
1092 | static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info, |
1093 | u64 flags) | |
1094 | { | |
1095 | struct list_head *head = &info->space_info; | |
1096 | struct list_head *cur; | |
1097 | struct btrfs_space_info *found; | |
1098 | list_for_each(cur, head) { | |
1099 | found = list_entry(cur, struct btrfs_space_info, list); | |
1100 | if (found->flags == flags) | |
1101 | return found; | |
1102 | } | |
1103 | return NULL; | |
1104 | ||
1105 | } | |
1106 | ||
593060d7 CM |
1107 | static int update_space_info(struct btrfs_fs_info *info, u64 flags, |
1108 | u64 total_bytes, u64 bytes_used, | |
1109 | struct btrfs_space_info **space_info) | |
1110 | { | |
1111 | struct btrfs_space_info *found; | |
1112 | ||
1113 | found = __find_space_info(info, flags); | |
1114 | if (found) { | |
1115 | found->total_bytes += total_bytes; | |
1116 | found->bytes_used += bytes_used; | |
8f18cf13 | 1117 | found->full = 0; |
593060d7 CM |
1118 | WARN_ON(found->total_bytes < found->bytes_used); |
1119 | *space_info = found; | |
1120 | return 0; | |
1121 | } | |
1122 | found = kmalloc(sizeof(*found), GFP_NOFS); | |
1123 | if (!found) | |
1124 | return -ENOMEM; | |
1125 | ||
1126 | list_add(&found->list, &info->space_info); | |
1127 | found->flags = flags; | |
1128 | found->total_bytes = total_bytes; | |
1129 | found->bytes_used = bytes_used; | |
1130 | found->bytes_pinned = 0; | |
1131 | found->full = 0; | |
0ef3e66b | 1132 | found->force_alloc = 0; |
593060d7 CM |
1133 | *space_info = found; |
1134 | return 0; | |
1135 | } | |
1136 | ||
8790d502 CM |
1137 | static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags) |
1138 | { | |
1139 | u64 extra_flags = flags & (BTRFS_BLOCK_GROUP_RAID0 | | |
611f0e00 | 1140 | BTRFS_BLOCK_GROUP_RAID1 | |
321aecc6 | 1141 | BTRFS_BLOCK_GROUP_RAID10 | |
611f0e00 | 1142 | BTRFS_BLOCK_GROUP_DUP); |
8790d502 CM |
1143 | if (extra_flags) { |
1144 | if (flags & BTRFS_BLOCK_GROUP_DATA) | |
1145 | fs_info->avail_data_alloc_bits |= extra_flags; | |
1146 | if (flags & BTRFS_BLOCK_GROUP_METADATA) | |
1147 | fs_info->avail_metadata_alloc_bits |= extra_flags; | |
1148 | if (flags & BTRFS_BLOCK_GROUP_SYSTEM) | |
1149 | fs_info->avail_system_alloc_bits |= extra_flags; | |
1150 | } | |
1151 | } | |
593060d7 | 1152 | |
a061fc8d | 1153 | static u64 reduce_alloc_profile(struct btrfs_root *root, u64 flags) |
ec44a35c | 1154 | { |
a061fc8d CM |
1155 | u64 num_devices = root->fs_info->fs_devices->num_devices; |
1156 | ||
1157 | if (num_devices == 1) | |
1158 | flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0); | |
1159 | if (num_devices < 4) | |
1160 | flags &= ~BTRFS_BLOCK_GROUP_RAID10; | |
1161 | ||
ec44a35c CM |
1162 | if ((flags & BTRFS_BLOCK_GROUP_DUP) && |
1163 | (flags & (BTRFS_BLOCK_GROUP_RAID1 | | |
a061fc8d | 1164 | BTRFS_BLOCK_GROUP_RAID10))) { |
ec44a35c | 1165 | flags &= ~BTRFS_BLOCK_GROUP_DUP; |
a061fc8d | 1166 | } |
ec44a35c CM |
1167 | |
1168 | if ((flags & BTRFS_BLOCK_GROUP_RAID1) && | |
a061fc8d | 1169 | (flags & BTRFS_BLOCK_GROUP_RAID10)) { |
ec44a35c | 1170 | flags &= ~BTRFS_BLOCK_GROUP_RAID1; |
a061fc8d | 1171 | } |
ec44a35c CM |
1172 | |
1173 | if ((flags & BTRFS_BLOCK_GROUP_RAID0) && | |
1174 | ((flags & BTRFS_BLOCK_GROUP_RAID1) | | |
1175 | (flags & BTRFS_BLOCK_GROUP_RAID10) | | |
1176 | (flags & BTRFS_BLOCK_GROUP_DUP))) | |
1177 | flags &= ~BTRFS_BLOCK_GROUP_RAID0; | |
1178 | return flags; | |
1179 | } | |
1180 | ||
6324fbf3 CM |
1181 | static int do_chunk_alloc(struct btrfs_trans_handle *trans, |
1182 | struct btrfs_root *extent_root, u64 alloc_bytes, | |
0ef3e66b | 1183 | u64 flags, int force) |
6324fbf3 CM |
1184 | { |
1185 | struct btrfs_space_info *space_info; | |
1186 | u64 thresh; | |
1187 | u64 start; | |
1188 | u64 num_bytes; | |
1189 | int ret; | |
1190 | ||
a061fc8d | 1191 | flags = reduce_alloc_profile(extent_root, flags); |
ec44a35c | 1192 | |
6324fbf3 | 1193 | space_info = __find_space_info(extent_root->fs_info, flags); |
593060d7 CM |
1194 | if (!space_info) { |
1195 | ret = update_space_info(extent_root->fs_info, flags, | |
1196 | 0, 0, &space_info); | |
1197 | BUG_ON(ret); | |
1198 | } | |
6324fbf3 CM |
1199 | BUG_ON(!space_info); |
1200 | ||
0ef3e66b CM |
1201 | if (space_info->force_alloc) { |
1202 | force = 1; | |
1203 | space_info->force_alloc = 0; | |
1204 | } | |
6324fbf3 | 1205 | if (space_info->full) |
925baedd | 1206 | goto out; |
6324fbf3 | 1207 | |
8790d502 | 1208 | thresh = div_factor(space_info->total_bytes, 6); |
0ef3e66b CM |
1209 | if (!force && |
1210 | (space_info->bytes_used + space_info->bytes_pinned + alloc_bytes) < | |
6324fbf3 | 1211 | thresh) |
925baedd | 1212 | goto out; |
6324fbf3 | 1213 | |
925baedd | 1214 | mutex_lock(&extent_root->fs_info->chunk_mutex); |
6324fbf3 CM |
1215 | ret = btrfs_alloc_chunk(trans, extent_root, &start, &num_bytes, flags); |
1216 | if (ret == -ENOSPC) { | |
1217 | printk("space info full %Lu\n", flags); | |
1218 | space_info->full = 1; | |
925baedd | 1219 | goto out; |
6324fbf3 | 1220 | } |
6324fbf3 CM |
1221 | BUG_ON(ret); |
1222 | ||
1223 | ret = btrfs_make_block_group(trans, extent_root, 0, flags, | |
e17cade2 | 1224 | BTRFS_FIRST_CHUNK_TREE_OBJECTID, start, num_bytes); |
6324fbf3 | 1225 | BUG_ON(ret); |
925baedd | 1226 | out: |
333db94c | 1227 | mutex_unlock(&extent_root->fs_info->chunk_mutex); |
6324fbf3 CM |
1228 | return 0; |
1229 | } | |
1230 | ||
9078a3e1 CM |
1231 | static int update_block_group(struct btrfs_trans_handle *trans, |
1232 | struct btrfs_root *root, | |
db94535d | 1233 | u64 bytenr, u64 num_bytes, int alloc, |
0b86a832 | 1234 | int mark_free) |
9078a3e1 CM |
1235 | { |
1236 | struct btrfs_block_group_cache *cache; | |
1237 | struct btrfs_fs_info *info = root->fs_info; | |
db94535d | 1238 | u64 total = num_bytes; |
9078a3e1 | 1239 | u64 old_val; |
db94535d | 1240 | u64 byte_in_group; |
96b5179d CM |
1241 | u64 start; |
1242 | u64 end; | |
3e1ad54f | 1243 | |
9078a3e1 | 1244 | while(total) { |
db94535d | 1245 | cache = btrfs_lookup_block_group(info, bytenr); |
3e1ad54f | 1246 | if (!cache) { |
9078a3e1 | 1247 | return -1; |
cd1bc465 | 1248 | } |
db94535d CM |
1249 | byte_in_group = bytenr - cache->key.objectid; |
1250 | WARN_ON(byte_in_group > cache->key.offset); | |
96b5179d CM |
1251 | start = cache->key.objectid; |
1252 | end = start + cache->key.offset - 1; | |
1253 | set_extent_bits(&info->block_group_cache, start, end, | |
1254 | BLOCK_GROUP_DIRTY, GFP_NOFS); | |
9078a3e1 CM |
1255 | |
1256 | old_val = btrfs_block_group_used(&cache->item); | |
db94535d | 1257 | num_bytes = min(total, cache->key.offset - byte_in_group); |
cd1bc465 | 1258 | if (alloc) { |
db94535d | 1259 | old_val += num_bytes; |
6324fbf3 | 1260 | cache->space_info->bytes_used += num_bytes; |
cd1bc465 | 1261 | } else { |
db94535d | 1262 | old_val -= num_bytes; |
6324fbf3 | 1263 | cache->space_info->bytes_used -= num_bytes; |
f510cfec CM |
1264 | if (mark_free) { |
1265 | set_extent_dirty(&info->free_space_cache, | |
db94535d | 1266 | bytenr, bytenr + num_bytes - 1, |
f510cfec | 1267 | GFP_NOFS); |
e37c9e69 | 1268 | } |
cd1bc465 | 1269 | } |
9078a3e1 | 1270 | btrfs_set_block_group_used(&cache->item, old_val); |
db94535d CM |
1271 | total -= num_bytes; |
1272 | bytenr += num_bytes; | |
9078a3e1 CM |
1273 | } |
1274 | return 0; | |
1275 | } | |
6324fbf3 | 1276 | |
a061fc8d CM |
1277 | static u64 first_logical_byte(struct btrfs_root *root, u64 search_start) |
1278 | { | |
1279 | u64 start; | |
1280 | u64 end; | |
1281 | int ret; | |
1282 | ret = find_first_extent_bit(&root->fs_info->block_group_cache, | |
1283 | search_start, &start, &end, | |
1284 | BLOCK_GROUP_DATA | BLOCK_GROUP_METADATA | | |
1285 | BLOCK_GROUP_SYSTEM); | |
1286 | if (ret) | |
1287 | return 0; | |
1288 | return start; | |
1289 | } | |
1290 | ||
1291 | ||
324ae4df Y |
1292 | static int update_pinned_extents(struct btrfs_root *root, |
1293 | u64 bytenr, u64 num, int pin) | |
1294 | { | |
1295 | u64 len; | |
1296 | struct btrfs_block_group_cache *cache; | |
1297 | struct btrfs_fs_info *fs_info = root->fs_info; | |
1298 | ||
1299 | if (pin) { | |
1300 | set_extent_dirty(&fs_info->pinned_extents, | |
1301 | bytenr, bytenr + num - 1, GFP_NOFS); | |
1302 | } else { | |
1303 | clear_extent_dirty(&fs_info->pinned_extents, | |
1304 | bytenr, bytenr + num - 1, GFP_NOFS); | |
1305 | } | |
1306 | while (num > 0) { | |
1307 | cache = btrfs_lookup_block_group(fs_info, bytenr); | |
a061fc8d CM |
1308 | if (!cache) { |
1309 | u64 first = first_logical_byte(root, bytenr); | |
1310 | WARN_ON(first < bytenr); | |
1311 | len = min(first - bytenr, num); | |
1312 | } else { | |
1313 | len = min(num, cache->key.offset - | |
1314 | (bytenr - cache->key.objectid)); | |
1315 | } | |
324ae4df | 1316 | if (pin) { |
a061fc8d CM |
1317 | if (cache) { |
1318 | cache->pinned += len; | |
1319 | cache->space_info->bytes_pinned += len; | |
1320 | } | |
324ae4df Y |
1321 | fs_info->total_pinned += len; |
1322 | } else { | |
a061fc8d CM |
1323 | if (cache) { |
1324 | cache->pinned -= len; | |
1325 | cache->space_info->bytes_pinned -= len; | |
1326 | } | |
324ae4df Y |
1327 | fs_info->total_pinned -= len; |
1328 | } | |
1329 | bytenr += len; | |
1330 | num -= len; | |
1331 | } | |
1332 | return 0; | |
1333 | } | |
9078a3e1 | 1334 | |
d1310b2e | 1335 | int btrfs_copy_pinned(struct btrfs_root *root, struct extent_io_tree *copy) |
ccd467d6 | 1336 | { |
ccd467d6 | 1337 | u64 last = 0; |
1a5bc167 CM |
1338 | u64 start; |
1339 | u64 end; | |
d1310b2e | 1340 | struct extent_io_tree *pinned_extents = &root->fs_info->pinned_extents; |
ccd467d6 | 1341 | int ret; |
ccd467d6 CM |
1342 | |
1343 | while(1) { | |
1a5bc167 CM |
1344 | ret = find_first_extent_bit(pinned_extents, last, |
1345 | &start, &end, EXTENT_DIRTY); | |
1346 | if (ret) | |
ccd467d6 | 1347 | break; |
1a5bc167 CM |
1348 | set_extent_dirty(copy, start, end, GFP_NOFS); |
1349 | last = end + 1; | |
ccd467d6 CM |
1350 | } |
1351 | return 0; | |
1352 | } | |
1353 | ||
1354 | int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans, | |
1355 | struct btrfs_root *root, | |
d1310b2e | 1356 | struct extent_io_tree *unpin) |
a28ec197 | 1357 | { |
1a5bc167 CM |
1358 | u64 start; |
1359 | u64 end; | |
a28ec197 | 1360 | int ret; |
d1310b2e | 1361 | struct extent_io_tree *free_space_cache; |
f510cfec | 1362 | free_space_cache = &root->fs_info->free_space_cache; |
a28ec197 | 1363 | |
925baedd | 1364 | mutex_lock(&root->fs_info->alloc_mutex); |
a28ec197 | 1365 | while(1) { |
1a5bc167 CM |
1366 | ret = find_first_extent_bit(unpin, 0, &start, &end, |
1367 | EXTENT_DIRTY); | |
1368 | if (ret) | |
a28ec197 | 1369 | break; |
324ae4df | 1370 | update_pinned_extents(root, start, end + 1 - start, 0); |
1a5bc167 CM |
1371 | clear_extent_dirty(unpin, start, end, GFP_NOFS); |
1372 | set_extent_dirty(free_space_cache, start, end, GFP_NOFS); | |
a28ec197 | 1373 | } |
925baedd | 1374 | mutex_unlock(&root->fs_info->alloc_mutex); |
a28ec197 CM |
1375 | return 0; |
1376 | } | |
1377 | ||
98ed5174 CM |
1378 | static int finish_current_insert(struct btrfs_trans_handle *trans, |
1379 | struct btrfs_root *extent_root) | |
037e6390 | 1380 | { |
7bb86316 CM |
1381 | u64 start; |
1382 | u64 end; | |
1383 | struct btrfs_fs_info *info = extent_root->fs_info; | |
d8d5f3e1 | 1384 | struct extent_buffer *eb; |
7bb86316 | 1385 | struct btrfs_path *path; |
e2fa7227 | 1386 | struct btrfs_key ins; |
d8d5f3e1 | 1387 | struct btrfs_disk_key first; |
234b63a0 | 1388 | struct btrfs_extent_item extent_item; |
037e6390 | 1389 | int ret; |
d8d5f3e1 | 1390 | int level; |
1a5bc167 | 1391 | int err = 0; |
037e6390 | 1392 | |
5f39d397 | 1393 | btrfs_set_stack_extent_refs(&extent_item, 1); |
62e2749e | 1394 | btrfs_set_key_type(&ins, BTRFS_EXTENT_ITEM_KEY); |
7bb86316 | 1395 | path = btrfs_alloc_path(); |
037e6390 | 1396 | |
26b8003f | 1397 | while(1) { |
1a5bc167 CM |
1398 | ret = find_first_extent_bit(&info->extent_ins, 0, &start, |
1399 | &end, EXTENT_LOCKED); | |
1400 | if (ret) | |
26b8003f CM |
1401 | break; |
1402 | ||
1a5bc167 CM |
1403 | ins.objectid = start; |
1404 | ins.offset = end + 1 - start; | |
1405 | err = btrfs_insert_item(trans, extent_root, &ins, | |
1406 | &extent_item, sizeof(extent_item)); | |
1407 | clear_extent_bits(&info->extent_ins, start, end, EXTENT_LOCKED, | |
1408 | GFP_NOFS); | |
ca7a79ad CM |
1409 | eb = read_tree_block(extent_root, ins.objectid, ins.offset, |
1410 | trans->transid); | |
925baedd | 1411 | btrfs_tree_lock(eb); |
d8d5f3e1 CM |
1412 | level = btrfs_header_level(eb); |
1413 | if (level == 0) { | |
1414 | btrfs_item_key(eb, &first, 0); | |
1415 | } else { | |
1416 | btrfs_node_key(eb, &first, 0); | |
1417 | } | |
925baedd CM |
1418 | btrfs_tree_unlock(eb); |
1419 | free_extent_buffer(eb); | |
1420 | /* | |
1421 | * the first key is just a hint, so the race we've created | |
1422 | * against reading it is fine | |
1423 | */ | |
7bb86316 CM |
1424 | err = btrfs_insert_extent_backref(trans, extent_root, path, |
1425 | start, extent_root->root_key.objectid, | |
f6dbff55 CM |
1426 | 0, level, |
1427 | btrfs_disk_key_objectid(&first)); | |
7bb86316 | 1428 | BUG_ON(err); |
037e6390 | 1429 | } |
7bb86316 | 1430 | btrfs_free_path(path); |
037e6390 CM |
1431 | return 0; |
1432 | } | |
1433 | ||
db94535d CM |
1434 | static int pin_down_bytes(struct btrfs_root *root, u64 bytenr, u32 num_bytes, |
1435 | int pending) | |
e20d96d6 | 1436 | { |
1a5bc167 | 1437 | int err = 0; |
8ef97622 | 1438 | |
f4b9aa8d | 1439 | if (!pending) { |
925baedd CM |
1440 | #if 0 |
1441 | struct extent_buffer *buf; | |
db94535d | 1442 | buf = btrfs_find_tree_block(root, bytenr, num_bytes); |
5f39d397 | 1443 | if (buf) { |
925baedd CM |
1444 | if (!btrfs_try_tree_lock(buf) && |
1445 | btrfs_buffer_uptodate(buf, 0)) { | |
2c90e5d6 CM |
1446 | u64 transid = |
1447 | root->fs_info->running_transaction->transid; | |
dc17ff8f CM |
1448 | u64 header_transid = |
1449 | btrfs_header_generation(buf); | |
6bc34676 CM |
1450 | if (header_transid == transid && |
1451 | !btrfs_header_flag(buf, | |
1452 | BTRFS_HEADER_FLAG_WRITTEN)) { | |
55c69072 | 1453 | clean_tree_block(NULL, root, buf); |
925baedd | 1454 | btrfs_tree_unlock(buf); |
5f39d397 | 1455 | free_extent_buffer(buf); |
c549228f | 1456 | return 1; |
2c90e5d6 | 1457 | } |
925baedd | 1458 | btrfs_tree_unlock(buf); |
f4b9aa8d | 1459 | } |
5f39d397 | 1460 | free_extent_buffer(buf); |
8ef97622 | 1461 | } |
925baedd | 1462 | #endif |
324ae4df | 1463 | update_pinned_extents(root, bytenr, num_bytes, 1); |
f4b9aa8d | 1464 | } else { |
1a5bc167 | 1465 | set_extent_bits(&root->fs_info->pending_del, |
db94535d CM |
1466 | bytenr, bytenr + num_bytes - 1, |
1467 | EXTENT_LOCKED, GFP_NOFS); | |
f4b9aa8d | 1468 | } |
be744175 | 1469 | BUG_ON(err < 0); |
e20d96d6 CM |
1470 | return 0; |
1471 | } | |
1472 | ||
fec577fb | 1473 | /* |
a28ec197 | 1474 | * remove an extent from the root, returns 0 on success |
fec577fb | 1475 | */ |
e089f05c | 1476 | static int __free_extent(struct btrfs_trans_handle *trans, struct btrfs_root |
7bb86316 CM |
1477 | *root, u64 bytenr, u64 num_bytes, |
1478 | u64 root_objectid, u64 ref_generation, | |
1479 | u64 owner_objectid, u64 owner_offset, int pin, | |
e37c9e69 | 1480 | int mark_free) |
a28ec197 | 1481 | { |
5caf2a00 | 1482 | struct btrfs_path *path; |
e2fa7227 | 1483 | struct btrfs_key key; |
1261ec42 CM |
1484 | struct btrfs_fs_info *info = root->fs_info; |
1485 | struct btrfs_root *extent_root = info->extent_root; | |
5f39d397 | 1486 | struct extent_buffer *leaf; |
a28ec197 | 1487 | int ret; |
952fccac CM |
1488 | int extent_slot = 0; |
1489 | int found_extent = 0; | |
1490 | int num_to_del = 1; | |
234b63a0 | 1491 | struct btrfs_extent_item *ei; |
cf27e1ee | 1492 | u32 refs; |
037e6390 | 1493 | |
db94535d | 1494 | key.objectid = bytenr; |
62e2749e | 1495 | btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY); |
db94535d | 1496 | key.offset = num_bytes; |
5caf2a00 | 1497 | path = btrfs_alloc_path(); |
54aa1f4d CM |
1498 | if (!path) |
1499 | return -ENOMEM; | |
5f26f772 | 1500 | |
3c12ac72 | 1501 | path->reada = 1; |
7bb86316 CM |
1502 | ret = lookup_extent_backref(trans, extent_root, path, |
1503 | bytenr, root_objectid, | |
1504 | ref_generation, | |
1505 | owner_objectid, owner_offset, 1); | |
1506 | if (ret == 0) { | |
952fccac CM |
1507 | struct btrfs_key found_key; |
1508 | extent_slot = path->slots[0]; | |
1509 | while(extent_slot > 0) { | |
1510 | extent_slot--; | |
1511 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, | |
1512 | extent_slot); | |
1513 | if (found_key.objectid != bytenr) | |
1514 | break; | |
1515 | if (found_key.type == BTRFS_EXTENT_ITEM_KEY && | |
1516 | found_key.offset == num_bytes) { | |
1517 | found_extent = 1; | |
1518 | break; | |
1519 | } | |
1520 | if (path->slots[0] - extent_slot > 5) | |
1521 | break; | |
1522 | } | |
1523 | if (!found_extent) | |
1524 | ret = btrfs_del_item(trans, extent_root, path); | |
7bb86316 CM |
1525 | } else { |
1526 | btrfs_print_leaf(extent_root, path->nodes[0]); | |
1527 | WARN_ON(1); | |
1528 | printk("Unable to find ref byte nr %Lu root %Lu " | |
1529 | " gen %Lu owner %Lu offset %Lu\n", bytenr, | |
1530 | root_objectid, ref_generation, owner_objectid, | |
1531 | owner_offset); | |
1532 | } | |
952fccac CM |
1533 | if (!found_extent) { |
1534 | btrfs_release_path(extent_root, path); | |
1535 | ret = btrfs_search_slot(trans, extent_root, &key, path, -1, 1); | |
1536 | if (ret < 0) | |
1537 | return ret; | |
1538 | BUG_ON(ret); | |
1539 | extent_slot = path->slots[0]; | |
1540 | } | |
5f39d397 CM |
1541 | |
1542 | leaf = path->nodes[0]; | |
952fccac | 1543 | ei = btrfs_item_ptr(leaf, extent_slot, |
123abc88 | 1544 | struct btrfs_extent_item); |
5f39d397 CM |
1545 | refs = btrfs_extent_refs(leaf, ei); |
1546 | BUG_ON(refs == 0); | |
1547 | refs -= 1; | |
1548 | btrfs_set_extent_refs(leaf, ei, refs); | |
952fccac | 1549 | |
5f39d397 CM |
1550 | btrfs_mark_buffer_dirty(leaf); |
1551 | ||
952fccac CM |
1552 | if (refs == 0 && found_extent && path->slots[0] == extent_slot + 1) { |
1553 | /* if the back ref and the extent are next to each other | |
1554 | * they get deleted below in one shot | |
1555 | */ | |
1556 | path->slots[0] = extent_slot; | |
1557 | num_to_del = 2; | |
1558 | } else if (found_extent) { | |
1559 | /* otherwise delete the extent back ref */ | |
1560 | ret = btrfs_del_item(trans, extent_root, path); | |
1561 | BUG_ON(ret); | |
1562 | /* if refs are 0, we need to setup the path for deletion */ | |
1563 | if (refs == 0) { | |
1564 | btrfs_release_path(extent_root, path); | |
1565 | ret = btrfs_search_slot(trans, extent_root, &key, path, | |
1566 | -1, 1); | |
1567 | if (ret < 0) | |
1568 | return ret; | |
1569 | BUG_ON(ret); | |
1570 | } | |
1571 | } | |
1572 | ||
cf27e1ee | 1573 | if (refs == 0) { |
db94535d CM |
1574 | u64 super_used; |
1575 | u64 root_used; | |
78fae27e CM |
1576 | |
1577 | if (pin) { | |
db94535d | 1578 | ret = pin_down_bytes(root, bytenr, num_bytes, 0); |
c549228f Y |
1579 | if (ret > 0) |
1580 | mark_free = 1; | |
1581 | BUG_ON(ret < 0); | |
78fae27e CM |
1582 | } |
1583 | ||
58176a96 | 1584 | /* block accounting for super block */ |
a2135011 | 1585 | spin_lock_irq(&info->delalloc_lock); |
db94535d CM |
1586 | super_used = btrfs_super_bytes_used(&info->super_copy); |
1587 | btrfs_set_super_bytes_used(&info->super_copy, | |
1588 | super_used - num_bytes); | |
a2135011 | 1589 | spin_unlock_irq(&info->delalloc_lock); |
58176a96 JB |
1590 | |
1591 | /* block accounting for root item */ | |
db94535d | 1592 | root_used = btrfs_root_used(&root->root_item); |
5f39d397 | 1593 | btrfs_set_root_used(&root->root_item, |
db94535d | 1594 | root_used - num_bytes); |
952fccac CM |
1595 | ret = btrfs_del_items(trans, extent_root, path, path->slots[0], |
1596 | num_to_del); | |
54aa1f4d CM |
1597 | if (ret) { |
1598 | return ret; | |
1599 | } | |
db94535d | 1600 | ret = update_block_group(trans, root, bytenr, num_bytes, 0, |
0b86a832 | 1601 | mark_free); |
9078a3e1 | 1602 | BUG_ON(ret); |
a28ec197 | 1603 | } |
5caf2a00 | 1604 | btrfs_free_path(path); |
e089f05c | 1605 | finish_current_insert(trans, extent_root); |
a28ec197 CM |
1606 | return ret; |
1607 | } | |
1608 | ||
a28ec197 CM |
1609 | /* |
1610 | * find all the blocks marked as pending in the radix tree and remove | |
1611 | * them from the extent map | |
1612 | */ | |
e089f05c CM |
1613 | static int del_pending_extents(struct btrfs_trans_handle *trans, struct |
1614 | btrfs_root *extent_root) | |
a28ec197 CM |
1615 | { |
1616 | int ret; | |
e20d96d6 | 1617 | int err = 0; |
1a5bc167 CM |
1618 | u64 start; |
1619 | u64 end; | |
d1310b2e CM |
1620 | struct extent_io_tree *pending_del; |
1621 | struct extent_io_tree *pinned_extents; | |
8ef97622 | 1622 | |
1a5bc167 CM |
1623 | pending_del = &extent_root->fs_info->pending_del; |
1624 | pinned_extents = &extent_root->fs_info->pinned_extents; | |
a28ec197 CM |
1625 | |
1626 | while(1) { | |
1a5bc167 CM |
1627 | ret = find_first_extent_bit(pending_del, 0, &start, &end, |
1628 | EXTENT_LOCKED); | |
1629 | if (ret) | |
a28ec197 | 1630 | break; |
324ae4df | 1631 | update_pinned_extents(extent_root, start, end + 1 - start, 1); |
1a5bc167 CM |
1632 | clear_extent_bits(pending_del, start, end, EXTENT_LOCKED, |
1633 | GFP_NOFS); | |
1634 | ret = __free_extent(trans, extent_root, | |
7bb86316 CM |
1635 | start, end + 1 - start, |
1636 | extent_root->root_key.objectid, | |
1637 | 0, 0, 0, 0, 0); | |
1a5bc167 CM |
1638 | if (ret) |
1639 | err = ret; | |
fec577fb | 1640 | } |
e20d96d6 | 1641 | return err; |
fec577fb CM |
1642 | } |
1643 | ||
1644 | /* | |
1645 | * remove an extent from the root, returns 0 on success | |
1646 | */ | |
925baedd CM |
1647 | static int __btrfs_free_extent(struct btrfs_trans_handle *trans, |
1648 | struct btrfs_root *root, u64 bytenr, | |
1649 | u64 num_bytes, u64 root_objectid, | |
1650 | u64 ref_generation, u64 owner_objectid, | |
1651 | u64 owner_offset, int pin) | |
fec577fb | 1652 | { |
9f5fae2f | 1653 | struct btrfs_root *extent_root = root->fs_info->extent_root; |
fec577fb CM |
1654 | int pending_ret; |
1655 | int ret; | |
a28ec197 | 1656 | |
db94535d | 1657 | WARN_ON(num_bytes < root->sectorsize); |
7bb86316 CM |
1658 | if (!root->ref_cows) |
1659 | ref_generation = 0; | |
1660 | ||
fec577fb | 1661 | if (root == extent_root) { |
db94535d | 1662 | pin_down_bytes(root, bytenr, num_bytes, 1); |
fec577fb CM |
1663 | return 0; |
1664 | } | |
7bb86316 CM |
1665 | ret = __free_extent(trans, root, bytenr, num_bytes, root_objectid, |
1666 | ref_generation, owner_objectid, owner_offset, | |
1667 | pin, pin == 0); | |
e20d96d6 | 1668 | pending_ret = del_pending_extents(trans, root->fs_info->extent_root); |
fec577fb CM |
1669 | return ret ? ret : pending_ret; |
1670 | } | |
1671 | ||
925baedd CM |
1672 | int btrfs_free_extent(struct btrfs_trans_handle *trans, |
1673 | struct btrfs_root *root, u64 bytenr, | |
1674 | u64 num_bytes, u64 root_objectid, | |
1675 | u64 ref_generation, u64 owner_objectid, | |
1676 | u64 owner_offset, int pin) | |
1677 | { | |
1678 | int ret; | |
1679 | ||
1680 | maybe_lock_mutex(root); | |
1681 | ret = __btrfs_free_extent(trans, root, bytenr, num_bytes, | |
1682 | root_objectid, ref_generation, | |
1683 | owner_objectid, owner_offset, pin); | |
1684 | maybe_unlock_mutex(root); | |
1685 | return ret; | |
1686 | } | |
1687 | ||
87ee04eb CM |
1688 | static u64 stripe_align(struct btrfs_root *root, u64 val) |
1689 | { | |
1690 | u64 mask = ((u64)root->stripesize - 1); | |
1691 | u64 ret = (val + mask) & ~mask; | |
1692 | return ret; | |
1693 | } | |
1694 | ||
fec577fb CM |
1695 | /* |
1696 | * walks the btree of allocated extents and find a hole of a given size. | |
1697 | * The key ins is changed to record the hole: | |
1698 | * ins->objectid == block start | |
62e2749e | 1699 | * ins->flags = BTRFS_EXTENT_ITEM_KEY |
fec577fb CM |
1700 | * ins->offset == number of blocks |
1701 | * Any available blocks before search_start are skipped. | |
1702 | */ | |
98ed5174 CM |
1703 | static int noinline find_free_extent(struct btrfs_trans_handle *trans, |
1704 | struct btrfs_root *orig_root, | |
1705 | u64 num_bytes, u64 empty_size, | |
1706 | u64 search_start, u64 search_end, | |
1707 | u64 hint_byte, struct btrfs_key *ins, | |
1708 | u64 exclude_start, u64 exclude_nr, | |
1709 | int data) | |
fec577fb | 1710 | { |
87ee04eb | 1711 | int ret; |
a061fc8d | 1712 | u64 orig_search_start; |
9f5fae2f | 1713 | struct btrfs_root * root = orig_root->fs_info->extent_root; |
f2458e1d | 1714 | struct btrfs_fs_info *info = root->fs_info; |
db94535d | 1715 | u64 total_needed = num_bytes; |
239b14b3 | 1716 | u64 *last_ptr = NULL; |
be08c1b9 | 1717 | struct btrfs_block_group_cache *block_group; |
be744175 | 1718 | int full_scan = 0; |
fbdc762b | 1719 | int wrapped = 0; |
0ef3e66b | 1720 | int chunk_alloc_done = 0; |
239b14b3 | 1721 | int empty_cluster = 2 * 1024 * 1024; |
0ef3e66b | 1722 | int allowed_chunk_alloc = 0; |
fec577fb | 1723 | |
db94535d | 1724 | WARN_ON(num_bytes < root->sectorsize); |
b1a4d965 CM |
1725 | btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY); |
1726 | ||
0ef3e66b CM |
1727 | if (orig_root->ref_cows || empty_size) |
1728 | allowed_chunk_alloc = 1; | |
1729 | ||
239b14b3 CM |
1730 | if (data & BTRFS_BLOCK_GROUP_METADATA) { |
1731 | last_ptr = &root->fs_info->last_alloc; | |
8790d502 | 1732 | empty_cluster = 256 * 1024; |
239b14b3 CM |
1733 | } |
1734 | ||
1735 | if ((data & BTRFS_BLOCK_GROUP_DATA) && btrfs_test_opt(root, SSD)) { | |
1736 | last_ptr = &root->fs_info->last_data_alloc; | |
1737 | } | |
1738 | ||
1739 | if (last_ptr) { | |
1740 | if (*last_ptr) | |
1741 | hint_byte = *last_ptr; | |
1742 | else { | |
1743 | empty_size += empty_cluster; | |
1744 | } | |
1745 | } | |
1746 | ||
a061fc8d CM |
1747 | search_start = max(search_start, first_logical_byte(root, 0)); |
1748 | orig_search_start = search_start; | |
1749 | ||
7f93bf8d CM |
1750 | if (search_end == (u64)-1) |
1751 | search_end = btrfs_super_total_bytes(&info->super_copy); | |
0b86a832 | 1752 | |
db94535d | 1753 | if (hint_byte) { |
0ef3e66b | 1754 | block_group = btrfs_lookup_first_block_group(info, hint_byte); |
1a2b2ac7 CM |
1755 | if (!block_group) |
1756 | hint_byte = search_start; | |
925baedd | 1757 | block_group = __btrfs_find_block_group(root, block_group, |
db94535d | 1758 | hint_byte, data, 1); |
239b14b3 CM |
1759 | if (last_ptr && *last_ptr == 0 && block_group) |
1760 | hint_byte = block_group->key.objectid; | |
be744175 | 1761 | } else { |
925baedd | 1762 | block_group = __btrfs_find_block_group(root, |
1a2b2ac7 CM |
1763 | trans->block_group, |
1764 | search_start, data, 1); | |
be744175 | 1765 | } |
239b14b3 | 1766 | search_start = max(search_start, hint_byte); |
be744175 | 1767 | |
6702ed49 | 1768 | total_needed += empty_size; |
0b86a832 | 1769 | |
be744175 | 1770 | check_failed: |
70b043f0 | 1771 | if (!block_group) { |
0ef3e66b CM |
1772 | block_group = btrfs_lookup_first_block_group(info, |
1773 | search_start); | |
70b043f0 | 1774 | if (!block_group) |
0ef3e66b | 1775 | block_group = btrfs_lookup_first_block_group(info, |
70b043f0 CM |
1776 | orig_search_start); |
1777 | } | |
0ef3e66b CM |
1778 | if (full_scan && !chunk_alloc_done) { |
1779 | if (allowed_chunk_alloc) { | |
1780 | do_chunk_alloc(trans, root, | |
1781 | num_bytes + 2 * 1024 * 1024, data, 1); | |
1782 | allowed_chunk_alloc = 0; | |
1783 | } else if (block_group && block_group_bits(block_group, data)) { | |
1784 | block_group->space_info->force_alloc = 1; | |
1785 | } | |
1786 | chunk_alloc_done = 1; | |
1787 | } | |
0b86a832 CM |
1788 | ret = find_search_start(root, &block_group, &search_start, |
1789 | total_needed, data); | |
239b14b3 CM |
1790 | if (ret == -ENOSPC && last_ptr && *last_ptr) { |
1791 | *last_ptr = 0; | |
0ef3e66b CM |
1792 | block_group = btrfs_lookup_first_block_group(info, |
1793 | orig_search_start); | |
239b14b3 CM |
1794 | search_start = orig_search_start; |
1795 | ret = find_search_start(root, &block_group, &search_start, | |
1796 | total_needed, data); | |
1797 | } | |
1798 | if (ret == -ENOSPC) | |
1799 | goto enospc; | |
0b86a832 | 1800 | if (ret) |
d548ee51 | 1801 | goto error; |
e19caa5f | 1802 | |
239b14b3 CM |
1803 | if (last_ptr && *last_ptr && search_start != *last_ptr) { |
1804 | *last_ptr = 0; | |
1805 | if (!empty_size) { | |
1806 | empty_size += empty_cluster; | |
1807 | total_needed += empty_size; | |
1808 | } | |
0ef3e66b | 1809 | block_group = btrfs_lookup_first_block_group(info, |
239b14b3 CM |
1810 | orig_search_start); |
1811 | search_start = orig_search_start; | |
1812 | ret = find_search_start(root, &block_group, | |
1813 | &search_start, total_needed, data); | |
1814 | if (ret == -ENOSPC) | |
1815 | goto enospc; | |
1816 | if (ret) | |
1817 | goto error; | |
1818 | } | |
1819 | ||
0b86a832 CM |
1820 | search_start = stripe_align(root, search_start); |
1821 | ins->objectid = search_start; | |
1822 | ins->offset = num_bytes; | |
e37c9e69 | 1823 | |
db94535d | 1824 | if (ins->objectid + num_bytes >= search_end) |
cf67582b | 1825 | goto enospc; |
0b86a832 CM |
1826 | |
1827 | if (ins->objectid + num_bytes > | |
1828 | block_group->key.objectid + block_group->key.offset) { | |
e19caa5f CM |
1829 | search_start = block_group->key.objectid + |
1830 | block_group->key.offset; | |
1831 | goto new_group; | |
1832 | } | |
0b86a832 | 1833 | |
1a5bc167 | 1834 | if (test_range_bit(&info->extent_ins, ins->objectid, |
db94535d CM |
1835 | ins->objectid + num_bytes -1, EXTENT_LOCKED, 0)) { |
1836 | search_start = ins->objectid + num_bytes; | |
1a5bc167 CM |
1837 | goto new_group; |
1838 | } | |
0b86a832 | 1839 | |
1a5bc167 | 1840 | if (test_range_bit(&info->pinned_extents, ins->objectid, |
db94535d CM |
1841 | ins->objectid + num_bytes -1, EXTENT_DIRTY, 0)) { |
1842 | search_start = ins->objectid + num_bytes; | |
1a5bc167 | 1843 | goto new_group; |
fec577fb | 1844 | } |
0b86a832 | 1845 | |
db94535d | 1846 | if (exclude_nr > 0 && (ins->objectid + num_bytes > exclude_start && |
f2654de4 CM |
1847 | ins->objectid < exclude_start + exclude_nr)) { |
1848 | search_start = exclude_start + exclude_nr; | |
1849 | goto new_group; | |
1850 | } | |
0b86a832 | 1851 | |
6324fbf3 | 1852 | if (!(data & BTRFS_BLOCK_GROUP_DATA)) { |
5276aeda | 1853 | block_group = btrfs_lookup_block_group(info, ins->objectid); |
26b8003f CM |
1854 | if (block_group) |
1855 | trans->block_group = block_group; | |
f2458e1d | 1856 | } |
db94535d | 1857 | ins->offset = num_bytes; |
239b14b3 CM |
1858 | if (last_ptr) { |
1859 | *last_ptr = ins->objectid + ins->offset; | |
1860 | if (*last_ptr == | |
1861 | btrfs_super_total_bytes(&root->fs_info->super_copy)) { | |
1862 | *last_ptr = 0; | |
1863 | } | |
1864 | } | |
fec577fb | 1865 | return 0; |
be744175 CM |
1866 | |
1867 | new_group: | |
db94535d | 1868 | if (search_start + num_bytes >= search_end) { |
cf67582b | 1869 | enospc: |
be744175 | 1870 | search_start = orig_search_start; |
fbdc762b CM |
1871 | if (full_scan) { |
1872 | ret = -ENOSPC; | |
1873 | goto error; | |
1874 | } | |
6702ed49 CM |
1875 | if (wrapped) { |
1876 | if (!full_scan) | |
1877 | total_needed -= empty_size; | |
fbdc762b | 1878 | full_scan = 1; |
6702ed49 | 1879 | } else |
fbdc762b | 1880 | wrapped = 1; |
be744175 | 1881 | } |
0ef3e66b | 1882 | block_group = btrfs_lookup_first_block_group(info, search_start); |
fbdc762b | 1883 | cond_resched(); |
925baedd | 1884 | block_group = __btrfs_find_block_group(root, block_group, |
1a2b2ac7 | 1885 | search_start, data, 0); |
be744175 CM |
1886 | goto check_failed; |
1887 | ||
0f70abe2 | 1888 | error: |
0f70abe2 | 1889 | return ret; |
fec577fb | 1890 | } |
ec44a35c | 1891 | |
fec577fb CM |
1892 | /* |
1893 | * finds a free extent and does all the dirty work required for allocation | |
1894 | * returns the key for the extent through ins, and a tree buffer for | |
1895 | * the first block of the extent through buf. | |
1896 | * | |
1897 | * returns 0 if everything worked, non-zero otherwise. | |
1898 | */ | |
4d775673 | 1899 | int btrfs_alloc_extent(struct btrfs_trans_handle *trans, |
7bb86316 | 1900 | struct btrfs_root *root, |
98d20f67 CM |
1901 | u64 num_bytes, u64 min_alloc_size, |
1902 | u64 root_objectid, u64 ref_generation, | |
7bb86316 CM |
1903 | u64 owner, u64 owner_offset, |
1904 | u64 empty_size, u64 hint_byte, | |
ec44a35c | 1905 | u64 search_end, struct btrfs_key *ins, u64 data) |
fec577fb CM |
1906 | { |
1907 | int ret; | |
1908 | int pending_ret; | |
c31f8830 CM |
1909 | u64 super_used; |
1910 | u64 root_used; | |
fbdc762b | 1911 | u64 search_start = 0; |
8790d502 | 1912 | u64 alloc_profile; |
47e4bb98 | 1913 | u32 sizes[2]; |
1261ec42 CM |
1914 | struct btrfs_fs_info *info = root->fs_info; |
1915 | struct btrfs_root *extent_root = info->extent_root; | |
47e4bb98 CM |
1916 | struct btrfs_extent_item *extent_item; |
1917 | struct btrfs_extent_ref *ref; | |
7bb86316 | 1918 | struct btrfs_path *path; |
47e4bb98 | 1919 | struct btrfs_key keys[2]; |
8f662a76 | 1920 | |
925baedd CM |
1921 | maybe_lock_mutex(root); |
1922 | ||
6324fbf3 | 1923 | if (data) { |
8790d502 CM |
1924 | alloc_profile = info->avail_data_alloc_bits & |
1925 | info->data_alloc_profile; | |
1926 | data = BTRFS_BLOCK_GROUP_DATA | alloc_profile; | |
6324fbf3 | 1927 | } else if (root == root->fs_info->chunk_root) { |
8790d502 CM |
1928 | alloc_profile = info->avail_system_alloc_bits & |
1929 | info->system_alloc_profile; | |
1930 | data = BTRFS_BLOCK_GROUP_SYSTEM | alloc_profile; | |
6324fbf3 | 1931 | } else { |
8790d502 CM |
1932 | alloc_profile = info->avail_metadata_alloc_bits & |
1933 | info->metadata_alloc_profile; | |
1934 | data = BTRFS_BLOCK_GROUP_METADATA | alloc_profile; | |
6324fbf3 | 1935 | } |
98d20f67 | 1936 | again: |
a061fc8d | 1937 | data = reduce_alloc_profile(root, data); |
0ef3e66b CM |
1938 | /* |
1939 | * the only place that sets empty_size is btrfs_realloc_node, which | |
1940 | * is not called recursively on allocations | |
1941 | */ | |
1942 | if (empty_size || root->ref_cows) { | |
593060d7 | 1943 | if (!(data & BTRFS_BLOCK_GROUP_METADATA)) { |
6324fbf3 | 1944 | ret = do_chunk_alloc(trans, root->fs_info->extent_root, |
0ef3e66b CM |
1945 | 2 * 1024 * 1024, |
1946 | BTRFS_BLOCK_GROUP_METADATA | | |
1947 | (info->metadata_alloc_profile & | |
1948 | info->avail_metadata_alloc_bits), 0); | |
6324fbf3 CM |
1949 | BUG_ON(ret); |
1950 | } | |
1951 | ret = do_chunk_alloc(trans, root->fs_info->extent_root, | |
0ef3e66b | 1952 | num_bytes + 2 * 1024 * 1024, data, 0); |
6324fbf3 CM |
1953 | BUG_ON(ret); |
1954 | } | |
0b86a832 | 1955 | |
db94535d CM |
1956 | WARN_ON(num_bytes < root->sectorsize); |
1957 | ret = find_free_extent(trans, root, num_bytes, empty_size, | |
1958 | search_start, search_end, hint_byte, ins, | |
26b8003f CM |
1959 | trans->alloc_exclude_start, |
1960 | trans->alloc_exclude_nr, data); | |
3b951516 | 1961 | |
98d20f67 CM |
1962 | if (ret == -ENOSPC && num_bytes > min_alloc_size) { |
1963 | num_bytes = num_bytes >> 1; | |
1964 | num_bytes = max(num_bytes, min_alloc_size); | |
0ef3e66b CM |
1965 | do_chunk_alloc(trans, root->fs_info->extent_root, |
1966 | num_bytes, data, 1); | |
98d20f67 CM |
1967 | goto again; |
1968 | } | |
ec44a35c CM |
1969 | if (ret) { |
1970 | printk("allocation failed flags %Lu\n", data); | |
1971 | } | |
925baedd CM |
1972 | if (ret) { |
1973 | BUG(); | |
1974 | goto out; | |
1975 | } | |
fec577fb | 1976 | |
58176a96 | 1977 | /* block accounting for super block */ |
a2135011 | 1978 | spin_lock_irq(&info->delalloc_lock); |
db94535d CM |
1979 | super_used = btrfs_super_bytes_used(&info->super_copy); |
1980 | btrfs_set_super_bytes_used(&info->super_copy, super_used + num_bytes); | |
a2135011 | 1981 | spin_unlock_irq(&info->delalloc_lock); |
26b8003f | 1982 | |
58176a96 | 1983 | /* block accounting for root item */ |
db94535d CM |
1984 | root_used = btrfs_root_used(&root->root_item); |
1985 | btrfs_set_root_used(&root->root_item, root_used + num_bytes); | |
58176a96 | 1986 | |
f510cfec CM |
1987 | clear_extent_dirty(&root->fs_info->free_space_cache, |
1988 | ins->objectid, ins->objectid + ins->offset - 1, | |
1989 | GFP_NOFS); | |
1990 | ||
26b8003f | 1991 | if (root == extent_root) { |
1a5bc167 CM |
1992 | set_extent_bits(&root->fs_info->extent_ins, ins->objectid, |
1993 | ins->objectid + ins->offset - 1, | |
1994 | EXTENT_LOCKED, GFP_NOFS); | |
26b8003f CM |
1995 | goto update_block; |
1996 | } | |
1997 | ||
1998 | WARN_ON(trans->alloc_exclude_nr); | |
1999 | trans->alloc_exclude_start = ins->objectid; | |
2000 | trans->alloc_exclude_nr = ins->offset; | |
037e6390 | 2001 | |
47e4bb98 CM |
2002 | memcpy(&keys[0], ins, sizeof(*ins)); |
2003 | keys[1].offset = hash_extent_ref(root_objectid, ref_generation, | |
2004 | owner, owner_offset); | |
2005 | keys[1].objectid = ins->objectid; | |
2006 | keys[1].type = BTRFS_EXTENT_REF_KEY; | |
2007 | sizes[0] = sizeof(*extent_item); | |
2008 | sizes[1] = sizeof(*ref); | |
7bb86316 CM |
2009 | |
2010 | path = btrfs_alloc_path(); | |
2011 | BUG_ON(!path); | |
47e4bb98 CM |
2012 | |
2013 | ret = btrfs_insert_empty_items(trans, extent_root, path, keys, | |
2014 | sizes, 2); | |
26b8003f | 2015 | |
ccd467d6 | 2016 | BUG_ON(ret); |
47e4bb98 CM |
2017 | extent_item = btrfs_item_ptr(path->nodes[0], path->slots[0], |
2018 | struct btrfs_extent_item); | |
2019 | btrfs_set_extent_refs(path->nodes[0], extent_item, 1); | |
2020 | ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1, | |
2021 | struct btrfs_extent_ref); | |
2022 | ||
2023 | btrfs_set_ref_root(path->nodes[0], ref, root_objectid); | |
2024 | btrfs_set_ref_generation(path->nodes[0], ref, ref_generation); | |
2025 | btrfs_set_ref_objectid(path->nodes[0], ref, owner); | |
2026 | btrfs_set_ref_offset(path->nodes[0], ref, owner_offset); | |
2027 | ||
2028 | btrfs_mark_buffer_dirty(path->nodes[0]); | |
2029 | ||
2030 | trans->alloc_exclude_start = 0; | |
2031 | trans->alloc_exclude_nr = 0; | |
7bb86316 | 2032 | btrfs_free_path(path); |
e089f05c | 2033 | finish_current_insert(trans, extent_root); |
e20d96d6 | 2034 | pending_ret = del_pending_extents(trans, extent_root); |
f510cfec | 2035 | |
925baedd CM |
2036 | if (ret) |
2037 | goto out; | |
e37c9e69 | 2038 | if (pending_ret) { |
925baedd CM |
2039 | ret = pending_ret; |
2040 | goto out; | |
e37c9e69 | 2041 | } |
26b8003f CM |
2042 | |
2043 | update_block: | |
0b86a832 | 2044 | ret = update_block_group(trans, root, ins->objectid, ins->offset, 1, 0); |
f5947066 CM |
2045 | if (ret) { |
2046 | printk("update block group failed for %Lu %Lu\n", | |
2047 | ins->objectid, ins->offset); | |
2048 | BUG(); | |
2049 | } | |
925baedd CM |
2050 | out: |
2051 | maybe_unlock_mutex(root); | |
2052 | return ret; | |
fec577fb | 2053 | } |
fec577fb CM |
2054 | /* |
2055 | * helper function to allocate a block for a given tree | |
2056 | * returns the tree buffer or NULL. | |
2057 | */ | |
5f39d397 | 2058 | struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans, |
7bb86316 CM |
2059 | struct btrfs_root *root, |
2060 | u32 blocksize, | |
2061 | u64 root_objectid, | |
2062 | u64 ref_generation, | |
2063 | u64 first_objectid, | |
2064 | int level, | |
2065 | u64 hint, | |
5f39d397 | 2066 | u64 empty_size) |
fec577fb | 2067 | { |
e2fa7227 | 2068 | struct btrfs_key ins; |
fec577fb | 2069 | int ret; |
5f39d397 | 2070 | struct extent_buffer *buf; |
fec577fb | 2071 | |
98d20f67 | 2072 | ret = btrfs_alloc_extent(trans, root, blocksize, blocksize, |
7bb86316 | 2073 | root_objectid, ref_generation, |
f6dbff55 | 2074 | level, first_objectid, empty_size, hint, |
db94535d | 2075 | (u64)-1, &ins, 0); |
fec577fb | 2076 | if (ret) { |
54aa1f4d CM |
2077 | BUG_ON(ret > 0); |
2078 | return ERR_PTR(ret); | |
fec577fb | 2079 | } |
db94535d | 2080 | buf = btrfs_find_create_tree_block(root, ins.objectid, blocksize); |
54aa1f4d | 2081 | if (!buf) { |
7bb86316 CM |
2082 | btrfs_free_extent(trans, root, ins.objectid, blocksize, |
2083 | root->root_key.objectid, ref_generation, | |
2084 | 0, 0, 0); | |
54aa1f4d CM |
2085 | return ERR_PTR(-ENOMEM); |
2086 | } | |
55c69072 | 2087 | btrfs_set_header_generation(buf, trans->transid); |
925baedd | 2088 | btrfs_tree_lock(buf); |
55c69072 | 2089 | clean_tree_block(trans, root, buf); |
5f39d397 | 2090 | btrfs_set_buffer_uptodate(buf); |
55c69072 CM |
2091 | |
2092 | if (PageDirty(buf->first_page)) { | |
2093 | printk("page %lu dirty\n", buf->first_page->index); | |
2094 | WARN_ON(1); | |
2095 | } | |
2096 | ||
5f39d397 CM |
2097 | set_extent_dirty(&trans->transaction->dirty_pages, buf->start, |
2098 | buf->start + buf->len - 1, GFP_NOFS); | |
9afbb0b7 CM |
2099 | if (!btrfs_test_opt(root, SSD)) |
2100 | btrfs_set_buffer_defrag(buf); | |
d3c2fdcf | 2101 | trans->blocks_used++; |
fec577fb CM |
2102 | return buf; |
2103 | } | |
a28ec197 | 2104 | |
98ed5174 CM |
2105 | static int noinline drop_leaf_ref(struct btrfs_trans_handle *trans, |
2106 | struct btrfs_root *root, | |
2107 | struct extent_buffer *leaf) | |
6407bf6d | 2108 | { |
7bb86316 CM |
2109 | u64 leaf_owner; |
2110 | u64 leaf_generation; | |
5f39d397 | 2111 | struct btrfs_key key; |
6407bf6d CM |
2112 | struct btrfs_file_extent_item *fi; |
2113 | int i; | |
2114 | int nritems; | |
2115 | int ret; | |
2116 | ||
5f39d397 CM |
2117 | BUG_ON(!btrfs_is_leaf(leaf)); |
2118 | nritems = btrfs_header_nritems(leaf); | |
7bb86316 CM |
2119 | leaf_owner = btrfs_header_owner(leaf); |
2120 | leaf_generation = btrfs_header_generation(leaf); | |
2121 | ||
6407bf6d | 2122 | for (i = 0; i < nritems; i++) { |
db94535d | 2123 | u64 disk_bytenr; |
5f39d397 CM |
2124 | |
2125 | btrfs_item_key_to_cpu(leaf, &key, i); | |
2126 | if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY) | |
6407bf6d CM |
2127 | continue; |
2128 | fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item); | |
5f39d397 CM |
2129 | if (btrfs_file_extent_type(leaf, fi) == |
2130 | BTRFS_FILE_EXTENT_INLINE) | |
236454df | 2131 | continue; |
6407bf6d CM |
2132 | /* |
2133 | * FIXME make sure to insert a trans record that | |
2134 | * repeats the snapshot del on crash | |
2135 | */ | |
db94535d CM |
2136 | disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); |
2137 | if (disk_bytenr == 0) | |
3a686375 | 2138 | continue; |
925baedd | 2139 | ret = __btrfs_free_extent(trans, root, disk_bytenr, |
7bb86316 CM |
2140 | btrfs_file_extent_disk_num_bytes(leaf, fi), |
2141 | leaf_owner, leaf_generation, | |
2142 | key.objectid, key.offset, 0); | |
6407bf6d CM |
2143 | BUG_ON(ret); |
2144 | } | |
2145 | return 0; | |
2146 | } | |
2147 | ||
98ed5174 | 2148 | static void noinline reada_walk_down(struct btrfs_root *root, |
bea495e5 CM |
2149 | struct extent_buffer *node, |
2150 | int slot) | |
e011599b | 2151 | { |
db94535d | 2152 | u64 bytenr; |
bea495e5 CM |
2153 | u64 last = 0; |
2154 | u32 nritems; | |
e011599b | 2155 | u32 refs; |
db94535d | 2156 | u32 blocksize; |
bea495e5 CM |
2157 | int ret; |
2158 | int i; | |
2159 | int level; | |
2160 | int skipped = 0; | |
e011599b | 2161 | |
5f39d397 | 2162 | nritems = btrfs_header_nritems(node); |
db94535d | 2163 | level = btrfs_header_level(node); |
bea495e5 CM |
2164 | if (level) |
2165 | return; | |
2166 | ||
2167 | for (i = slot; i < nritems && skipped < 32; i++) { | |
db94535d | 2168 | bytenr = btrfs_node_blockptr(node, i); |
bea495e5 CM |
2169 | if (last && ((bytenr > last && bytenr - last > 32 * 1024) || |
2170 | (last > bytenr && last - bytenr > 32 * 1024))) { | |
2171 | skipped++; | |
e011599b | 2172 | continue; |
bea495e5 CM |
2173 | } |
2174 | blocksize = btrfs_level_size(root, level - 1); | |
2175 | if (i != slot) { | |
2176 | ret = lookup_extent_ref(NULL, root, bytenr, | |
2177 | blocksize, &refs); | |
2178 | BUG_ON(ret); | |
2179 | if (refs != 1) { | |
2180 | skipped++; | |
2181 | continue; | |
2182 | } | |
2183 | } | |
ca7a79ad CM |
2184 | ret = readahead_tree_block(root, bytenr, blocksize, |
2185 | btrfs_node_ptr_generation(node, i)); | |
bea495e5 | 2186 | last = bytenr + blocksize; |
409eb95d | 2187 | cond_resched(); |
e011599b CM |
2188 | if (ret) |
2189 | break; | |
2190 | } | |
2191 | } | |
2192 | ||
333db94c CM |
2193 | /* |
2194 | * we want to avoid as much random IO as we can with the alloc mutex | |
2195 | * held, so drop the lock and do the lookup, then do it again with the | |
2196 | * lock held. | |
2197 | */ | |
2198 | int drop_snap_lookup_refcount(struct btrfs_root *root, u64 start, u64 len, | |
2199 | u32 *refs) | |
2200 | { | |
2201 | mutex_unlock(&root->fs_info->alloc_mutex); | |
2202 | lookup_extent_ref(NULL, root, start, len, refs); | |
2203 | mutex_lock(&root->fs_info->alloc_mutex); | |
2204 | return lookup_extent_ref(NULL, root, start, len, refs); | |
2205 | } | |
2206 | ||
9aca1d51 CM |
2207 | /* |
2208 | * helper function for drop_snapshot, this walks down the tree dropping ref | |
2209 | * counts as it goes. | |
2210 | */ | |
98ed5174 CM |
2211 | static int noinline walk_down_tree(struct btrfs_trans_handle *trans, |
2212 | struct btrfs_root *root, | |
2213 | struct btrfs_path *path, int *level) | |
20524f02 | 2214 | { |
7bb86316 CM |
2215 | u64 root_owner; |
2216 | u64 root_gen; | |
2217 | u64 bytenr; | |
ca7a79ad | 2218 | u64 ptr_gen; |
5f39d397 CM |
2219 | struct extent_buffer *next; |
2220 | struct extent_buffer *cur; | |
7bb86316 | 2221 | struct extent_buffer *parent; |
db94535d | 2222 | u32 blocksize; |
20524f02 CM |
2223 | int ret; |
2224 | u32 refs; | |
2225 | ||
925baedd CM |
2226 | mutex_lock(&root->fs_info->alloc_mutex); |
2227 | ||
5caf2a00 CM |
2228 | WARN_ON(*level < 0); |
2229 | WARN_ON(*level >= BTRFS_MAX_LEVEL); | |
333db94c | 2230 | ret = drop_snap_lookup_refcount(root, path->nodes[*level]->start, |
db94535d | 2231 | path->nodes[*level]->len, &refs); |
20524f02 CM |
2232 | BUG_ON(ret); |
2233 | if (refs > 1) | |
2234 | goto out; | |
e011599b | 2235 | |
9aca1d51 CM |
2236 | /* |
2237 | * walk down to the last node level and free all the leaves | |
2238 | */ | |
6407bf6d | 2239 | while(*level >= 0) { |
5caf2a00 CM |
2240 | WARN_ON(*level < 0); |
2241 | WARN_ON(*level >= BTRFS_MAX_LEVEL); | |
20524f02 | 2242 | cur = path->nodes[*level]; |
e011599b | 2243 | |
5f39d397 | 2244 | if (btrfs_header_level(cur) != *level) |
2c90e5d6 | 2245 | WARN_ON(1); |
e011599b | 2246 | |
7518a238 | 2247 | if (path->slots[*level] >= |
5f39d397 | 2248 | btrfs_header_nritems(cur)) |
20524f02 | 2249 | break; |
6407bf6d CM |
2250 | if (*level == 0) { |
2251 | ret = drop_leaf_ref(trans, root, cur); | |
2252 | BUG_ON(ret); | |
2253 | break; | |
2254 | } | |
db94535d | 2255 | bytenr = btrfs_node_blockptr(cur, path->slots[*level]); |
ca7a79ad | 2256 | ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]); |
db94535d | 2257 | blocksize = btrfs_level_size(root, *level - 1); |
925baedd | 2258 | |
333db94c | 2259 | ret = drop_snap_lookup_refcount(root, bytenr, blocksize, &refs); |
6407bf6d CM |
2260 | BUG_ON(ret); |
2261 | if (refs != 1) { | |
7bb86316 CM |
2262 | parent = path->nodes[*level]; |
2263 | root_owner = btrfs_header_owner(parent); | |
2264 | root_gen = btrfs_header_generation(parent); | |
20524f02 | 2265 | path->slots[*level]++; |
925baedd | 2266 | ret = __btrfs_free_extent(trans, root, bytenr, |
7bb86316 CM |
2267 | blocksize, root_owner, |
2268 | root_gen, 0, 0, 1); | |
20524f02 CM |
2269 | BUG_ON(ret); |
2270 | continue; | |
2271 | } | |
db94535d | 2272 | next = btrfs_find_tree_block(root, bytenr, blocksize); |
1259ab75 | 2273 | if (!next || !btrfs_buffer_uptodate(next, ptr_gen)) { |
5f39d397 | 2274 | free_extent_buffer(next); |
333db94c CM |
2275 | mutex_unlock(&root->fs_info->alloc_mutex); |
2276 | ||
bea495e5 | 2277 | reada_walk_down(root, cur, path->slots[*level]); |
8790d502 | 2278 | |
ca7a79ad CM |
2279 | next = read_tree_block(root, bytenr, blocksize, |
2280 | ptr_gen); | |
925baedd | 2281 | mutex_lock(&root->fs_info->alloc_mutex); |
e9d0b13b | 2282 | |
8790d502 | 2283 | /* we've dropped the lock, double check */ |
333db94c | 2284 | ret = drop_snap_lookup_refcount(root, bytenr, |
db94535d | 2285 | blocksize, &refs); |
e9d0b13b CM |
2286 | BUG_ON(ret); |
2287 | if (refs != 1) { | |
7bb86316 CM |
2288 | parent = path->nodes[*level]; |
2289 | root_owner = btrfs_header_owner(parent); | |
2290 | root_gen = btrfs_header_generation(parent); | |
2291 | ||
e9d0b13b | 2292 | path->slots[*level]++; |
5f39d397 | 2293 | free_extent_buffer(next); |
925baedd | 2294 | ret = __btrfs_free_extent(trans, root, bytenr, |
7bb86316 CM |
2295 | blocksize, |
2296 | root_owner, | |
2297 | root_gen, 0, 0, 1); | |
e9d0b13b CM |
2298 | BUG_ON(ret); |
2299 | continue; | |
2300 | } | |
2301 | } | |
5caf2a00 | 2302 | WARN_ON(*level <= 0); |
83e15a28 | 2303 | if (path->nodes[*level-1]) |
5f39d397 | 2304 | free_extent_buffer(path->nodes[*level-1]); |
20524f02 | 2305 | path->nodes[*level-1] = next; |
5f39d397 | 2306 | *level = btrfs_header_level(next); |
20524f02 CM |
2307 | path->slots[*level] = 0; |
2308 | } | |
2309 | out: | |
5caf2a00 CM |
2310 | WARN_ON(*level < 0); |
2311 | WARN_ON(*level >= BTRFS_MAX_LEVEL); | |
7bb86316 CM |
2312 | |
2313 | if (path->nodes[*level] == root->node) { | |
2314 | root_owner = root->root_key.objectid; | |
2315 | parent = path->nodes[*level]; | |
2316 | } else { | |
2317 | parent = path->nodes[*level + 1]; | |
2318 | root_owner = btrfs_header_owner(parent); | |
2319 | } | |
2320 | ||
2321 | root_gen = btrfs_header_generation(parent); | |
925baedd | 2322 | ret = __btrfs_free_extent(trans, root, path->nodes[*level]->start, |
7bb86316 CM |
2323 | path->nodes[*level]->len, |
2324 | root_owner, root_gen, 0, 0, 1); | |
5f39d397 | 2325 | free_extent_buffer(path->nodes[*level]); |
20524f02 CM |
2326 | path->nodes[*level] = NULL; |
2327 | *level += 1; | |
2328 | BUG_ON(ret); | |
925baedd | 2329 | mutex_unlock(&root->fs_info->alloc_mutex); |
20524f02 CM |
2330 | return 0; |
2331 | } | |
2332 | ||
9aca1d51 CM |
2333 | /* |
2334 | * helper for dropping snapshots. This walks back up the tree in the path | |
2335 | * to find the first node higher up where we haven't yet gone through | |
2336 | * all the slots | |
2337 | */ | |
98ed5174 CM |
2338 | static int noinline walk_up_tree(struct btrfs_trans_handle *trans, |
2339 | struct btrfs_root *root, | |
2340 | struct btrfs_path *path, int *level) | |
20524f02 | 2341 | { |
7bb86316 CM |
2342 | u64 root_owner; |
2343 | u64 root_gen; | |
2344 | struct btrfs_root_item *root_item = &root->root_item; | |
20524f02 CM |
2345 | int i; |
2346 | int slot; | |
2347 | int ret; | |
9f3a7427 | 2348 | |
234b63a0 | 2349 | for(i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) { |
20524f02 | 2350 | slot = path->slots[i]; |
5f39d397 CM |
2351 | if (slot < btrfs_header_nritems(path->nodes[i]) - 1) { |
2352 | struct extent_buffer *node; | |
2353 | struct btrfs_disk_key disk_key; | |
2354 | node = path->nodes[i]; | |
20524f02 CM |
2355 | path->slots[i]++; |
2356 | *level = i; | |
9f3a7427 | 2357 | WARN_ON(*level == 0); |
5f39d397 | 2358 | btrfs_node_key(node, &disk_key, path->slots[i]); |
9f3a7427 | 2359 | memcpy(&root_item->drop_progress, |
5f39d397 | 2360 | &disk_key, sizeof(disk_key)); |
9f3a7427 | 2361 | root_item->drop_level = i; |
20524f02 CM |
2362 | return 0; |
2363 | } else { | |
7bb86316 CM |
2364 | if (path->nodes[*level] == root->node) { |
2365 | root_owner = root->root_key.objectid; | |
2366 | root_gen = | |
2367 | btrfs_header_generation(path->nodes[*level]); | |
2368 | } else { | |
2369 | struct extent_buffer *node; | |
2370 | node = path->nodes[*level + 1]; | |
2371 | root_owner = btrfs_header_owner(node); | |
2372 | root_gen = btrfs_header_generation(node); | |
2373 | } | |
e089f05c | 2374 | ret = btrfs_free_extent(trans, root, |
db94535d | 2375 | path->nodes[*level]->start, |
7bb86316 CM |
2376 | path->nodes[*level]->len, |
2377 | root_owner, root_gen, 0, 0, 1); | |
6407bf6d | 2378 | BUG_ON(ret); |
5f39d397 | 2379 | free_extent_buffer(path->nodes[*level]); |
83e15a28 | 2380 | path->nodes[*level] = NULL; |
20524f02 | 2381 | *level = i + 1; |
20524f02 CM |
2382 | } |
2383 | } | |
2384 | return 1; | |
2385 | } | |
2386 | ||
9aca1d51 CM |
2387 | /* |
2388 | * drop the reference count on the tree rooted at 'snap'. This traverses | |
2389 | * the tree freeing any blocks that have a ref count of zero after being | |
2390 | * decremented. | |
2391 | */ | |
e089f05c | 2392 | int btrfs_drop_snapshot(struct btrfs_trans_handle *trans, struct btrfs_root |
9f3a7427 | 2393 | *root) |
20524f02 | 2394 | { |
3768f368 | 2395 | int ret = 0; |
9aca1d51 | 2396 | int wret; |
20524f02 | 2397 | int level; |
5caf2a00 | 2398 | struct btrfs_path *path; |
20524f02 CM |
2399 | int i; |
2400 | int orig_level; | |
9f3a7427 | 2401 | struct btrfs_root_item *root_item = &root->root_item; |
20524f02 | 2402 | |
a2135011 | 2403 | WARN_ON(!mutex_is_locked(&root->fs_info->drop_mutex)); |
5caf2a00 CM |
2404 | path = btrfs_alloc_path(); |
2405 | BUG_ON(!path); | |
20524f02 | 2406 | |
5f39d397 | 2407 | level = btrfs_header_level(root->node); |
20524f02 | 2408 | orig_level = level; |
9f3a7427 CM |
2409 | if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) { |
2410 | path->nodes[level] = root->node; | |
f510cfec | 2411 | extent_buffer_get(root->node); |
9f3a7427 CM |
2412 | path->slots[level] = 0; |
2413 | } else { | |
2414 | struct btrfs_key key; | |
5f39d397 CM |
2415 | struct btrfs_disk_key found_key; |
2416 | struct extent_buffer *node; | |
6702ed49 | 2417 | |
9f3a7427 | 2418 | btrfs_disk_key_to_cpu(&key, &root_item->drop_progress); |
6702ed49 CM |
2419 | level = root_item->drop_level; |
2420 | path->lowest_level = level; | |
9f3a7427 | 2421 | wret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
6702ed49 | 2422 | if (wret < 0) { |
9f3a7427 CM |
2423 | ret = wret; |
2424 | goto out; | |
2425 | } | |
5f39d397 CM |
2426 | node = path->nodes[level]; |
2427 | btrfs_node_key(node, &found_key, path->slots[level]); | |
2428 | WARN_ON(memcmp(&found_key, &root_item->drop_progress, | |
2429 | sizeof(found_key))); | |
925baedd CM |
2430 | for (i = 0; i < BTRFS_MAX_LEVEL; i++) { |
2431 | if (path->nodes[i] && path->locks[i]) { | |
2432 | path->locks[i] = 0; | |
2433 | btrfs_tree_unlock(path->nodes[i]); | |
2434 | } | |
2435 | } | |
9f3a7427 | 2436 | } |
20524f02 | 2437 | while(1) { |
5caf2a00 | 2438 | wret = walk_down_tree(trans, root, path, &level); |
9aca1d51 | 2439 | if (wret > 0) |
20524f02 | 2440 | break; |
9aca1d51 CM |
2441 | if (wret < 0) |
2442 | ret = wret; | |
2443 | ||
5caf2a00 | 2444 | wret = walk_up_tree(trans, root, path, &level); |
9aca1d51 | 2445 | if (wret > 0) |
20524f02 | 2446 | break; |
9aca1d51 CM |
2447 | if (wret < 0) |
2448 | ret = wret; | |
409eb95d | 2449 | ret = -EAGAIN; |
409eb95d | 2450 | break; |
20524f02 | 2451 | } |
83e15a28 | 2452 | for (i = 0; i <= orig_level; i++) { |
5caf2a00 | 2453 | if (path->nodes[i]) { |
5f39d397 | 2454 | free_extent_buffer(path->nodes[i]); |
0f82731f | 2455 | path->nodes[i] = NULL; |
83e15a28 | 2456 | } |
20524f02 | 2457 | } |
9f3a7427 | 2458 | out: |
5caf2a00 | 2459 | btrfs_free_path(path); |
9aca1d51 | 2460 | return ret; |
20524f02 | 2461 | } |
9078a3e1 | 2462 | |
96b5179d | 2463 | int btrfs_free_block_groups(struct btrfs_fs_info *info) |
9078a3e1 | 2464 | { |
96b5179d CM |
2465 | u64 start; |
2466 | u64 end; | |
b97f9203 | 2467 | u64 ptr; |
9078a3e1 | 2468 | int ret; |
925baedd CM |
2469 | |
2470 | mutex_lock(&info->alloc_mutex); | |
9078a3e1 | 2471 | while(1) { |
96b5179d CM |
2472 | ret = find_first_extent_bit(&info->block_group_cache, 0, |
2473 | &start, &end, (unsigned int)-1); | |
2474 | if (ret) | |
9078a3e1 | 2475 | break; |
b97f9203 Y |
2476 | ret = get_state_private(&info->block_group_cache, start, &ptr); |
2477 | if (!ret) | |
2478 | kfree((void *)(unsigned long)ptr); | |
96b5179d CM |
2479 | clear_extent_bits(&info->block_group_cache, start, |
2480 | end, (unsigned int)-1, GFP_NOFS); | |
9078a3e1 | 2481 | } |
e37c9e69 | 2482 | while(1) { |
f510cfec CM |
2483 | ret = find_first_extent_bit(&info->free_space_cache, 0, |
2484 | &start, &end, EXTENT_DIRTY); | |
2485 | if (ret) | |
e37c9e69 | 2486 | break; |
f510cfec CM |
2487 | clear_extent_dirty(&info->free_space_cache, start, |
2488 | end, GFP_NOFS); | |
e37c9e69 | 2489 | } |
925baedd | 2490 | mutex_unlock(&info->alloc_mutex); |
be744175 CM |
2491 | return 0; |
2492 | } | |
2493 | ||
8e7bf94f CM |
2494 | static unsigned long calc_ra(unsigned long start, unsigned long last, |
2495 | unsigned long nr) | |
2496 | { | |
2497 | return min(last, start + nr - 1); | |
2498 | } | |
2499 | ||
98ed5174 CM |
2500 | static int noinline relocate_inode_pages(struct inode *inode, u64 start, |
2501 | u64 len) | |
edbd8d4e CM |
2502 | { |
2503 | u64 page_start; | |
2504 | u64 page_end; | |
edbd8d4e | 2505 | unsigned long last_index; |
edbd8d4e CM |
2506 | unsigned long i; |
2507 | struct page *page; | |
d1310b2e | 2508 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
4313b399 | 2509 | struct file_ra_state *ra; |
8e7bf94f CM |
2510 | unsigned long total_read = 0; |
2511 | unsigned long ra_pages; | |
a061fc8d | 2512 | struct btrfs_trans_handle *trans; |
4313b399 CM |
2513 | |
2514 | ra = kzalloc(sizeof(*ra), GFP_NOFS); | |
edbd8d4e CM |
2515 | |
2516 | mutex_lock(&inode->i_mutex); | |
4313b399 | 2517 | i = start >> PAGE_CACHE_SHIFT; |
edbd8d4e CM |
2518 | last_index = (start + len - 1) >> PAGE_CACHE_SHIFT; |
2519 | ||
8e7bf94f CM |
2520 | ra_pages = BTRFS_I(inode)->root->fs_info->bdi.ra_pages; |
2521 | ||
4313b399 | 2522 | file_ra_state_init(ra, inode->i_mapping); |
edbd8d4e | 2523 | |
4313b399 | 2524 | for (; i <= last_index; i++) { |
8e7bf94f CM |
2525 | if (total_read % ra_pages == 0) { |
2526 | btrfs_force_ra(inode->i_mapping, ra, NULL, i, | |
2527 | calc_ra(i, last_index, ra_pages)); | |
2528 | } | |
2529 | total_read++; | |
a061fc8d CM |
2530 | if (((u64)i << PAGE_CACHE_SHIFT) > inode->i_size) |
2531 | goto truncate_racing; | |
2532 | ||
edbd8d4e | 2533 | page = grab_cache_page(inode->i_mapping, i); |
a061fc8d | 2534 | if (!page) { |
edbd8d4e | 2535 | goto out_unlock; |
a061fc8d | 2536 | } |
edbd8d4e CM |
2537 | if (!PageUptodate(page)) { |
2538 | btrfs_readpage(NULL, page); | |
2539 | lock_page(page); | |
2540 | if (!PageUptodate(page)) { | |
2541 | unlock_page(page); | |
2542 | page_cache_release(page); | |
2543 | goto out_unlock; | |
2544 | } | |
2545 | } | |
ec44a35c CM |
2546 | #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18) |
2547 | ClearPageDirty(page); | |
2548 | #else | |
2549 | cancel_dirty_page(page, PAGE_CACHE_SIZE); | |
2550 | #endif | |
2551 | wait_on_page_writeback(page); | |
2552 | set_page_extent_mapped(page); | |
edbd8d4e CM |
2553 | page_start = (u64)page->index << PAGE_CACHE_SHIFT; |
2554 | page_end = page_start + PAGE_CACHE_SIZE - 1; | |
2555 | ||
d1310b2e | 2556 | lock_extent(io_tree, page_start, page_end, GFP_NOFS); |
edbd8d4e | 2557 | |
d1310b2e | 2558 | set_extent_delalloc(io_tree, page_start, |
edbd8d4e | 2559 | page_end, GFP_NOFS); |
a061fc8d | 2560 | set_page_dirty(page); |
edbd8d4e | 2561 | |
d1310b2e | 2562 | unlock_extent(io_tree, page_start, page_end, GFP_NOFS); |
edbd8d4e CM |
2563 | unlock_page(page); |
2564 | page_cache_release(page); | |
2565 | } | |
a061fc8d CM |
2566 | balance_dirty_pages_ratelimited_nr(inode->i_mapping, |
2567 | total_read); | |
edbd8d4e CM |
2568 | |
2569 | out_unlock: | |
ec44a35c | 2570 | kfree(ra); |
a061fc8d CM |
2571 | trans = btrfs_start_transaction(BTRFS_I(inode)->root, 1); |
2572 | if (trans) { | |
2573 | btrfs_add_ordered_inode(inode); | |
2574 | btrfs_end_transaction(trans, BTRFS_I(inode)->root); | |
2575 | mark_inode_dirty(inode); | |
2576 | } | |
edbd8d4e CM |
2577 | mutex_unlock(&inode->i_mutex); |
2578 | return 0; | |
a061fc8d CM |
2579 | |
2580 | truncate_racing: | |
2581 | vmtruncate(inode, inode->i_size); | |
2582 | balance_dirty_pages_ratelimited_nr(inode->i_mapping, | |
2583 | total_read); | |
2584 | goto out_unlock; | |
edbd8d4e CM |
2585 | } |
2586 | ||
bf4ef679 CM |
2587 | /* |
2588 | * The back references tell us which tree holds a ref on a block, | |
2589 | * but it is possible for the tree root field in the reference to | |
2590 | * reflect the original root before a snapshot was made. In this | |
2591 | * case we should search through all the children of a given root | |
2592 | * to find potential holders of references on a block. | |
2593 | * | |
2594 | * Instead, we do something a little less fancy and just search | |
2595 | * all the roots for a given key/block combination. | |
2596 | */ | |
2597 | static int find_root_for_ref(struct btrfs_root *root, | |
2598 | struct btrfs_path *path, | |
2599 | struct btrfs_key *key0, | |
2600 | int level, | |
2601 | int file_key, | |
2602 | struct btrfs_root **found_root, | |
2603 | u64 bytenr) | |
2604 | { | |
2605 | struct btrfs_key root_location; | |
2606 | struct btrfs_root *cur_root = *found_root; | |
2607 | struct btrfs_file_extent_item *file_extent; | |
2608 | u64 root_search_start = BTRFS_FS_TREE_OBJECTID; | |
2609 | u64 found_bytenr; | |
2610 | int ret; | |
2611 | int i; | |
2612 | ||
2613 | root_location.offset = (u64)-1; | |
2614 | root_location.type = BTRFS_ROOT_ITEM_KEY; | |
2615 | path->lowest_level = level; | |
2616 | path->reada = 0; | |
2617 | while(1) { | |
2618 | ret = btrfs_search_slot(NULL, cur_root, key0, path, 0, 0); | |
2619 | found_bytenr = 0; | |
2620 | if (ret == 0 && file_key) { | |
2621 | struct extent_buffer *leaf = path->nodes[0]; | |
2622 | file_extent = btrfs_item_ptr(leaf, path->slots[0], | |
2623 | struct btrfs_file_extent_item); | |
2624 | if (btrfs_file_extent_type(leaf, file_extent) == | |
2625 | BTRFS_FILE_EXTENT_REG) { | |
2626 | found_bytenr = | |
2627 | btrfs_file_extent_disk_bytenr(leaf, | |
2628 | file_extent); | |
2629 | } | |
323da79c | 2630 | } else if (!file_key) { |
bf4ef679 CM |
2631 | if (path->nodes[level]) |
2632 | found_bytenr = path->nodes[level]->start; | |
2633 | } | |
2634 | ||
2635 | for (i = level; i < BTRFS_MAX_LEVEL; i++) { | |
2636 | if (!path->nodes[i]) | |
2637 | break; | |
2638 | free_extent_buffer(path->nodes[i]); | |
2639 | path->nodes[i] = NULL; | |
2640 | } | |
2641 | btrfs_release_path(cur_root, path); | |
2642 | ||
2643 | if (found_bytenr == bytenr) { | |
2644 | *found_root = cur_root; | |
2645 | ret = 0; | |
2646 | goto out; | |
2647 | } | |
2648 | ret = btrfs_search_root(root->fs_info->tree_root, | |
2649 | root_search_start, &root_search_start); | |
2650 | if (ret) | |
2651 | break; | |
2652 | ||
2653 | root_location.objectid = root_search_start; | |
2654 | cur_root = btrfs_read_fs_root_no_name(root->fs_info, | |
2655 | &root_location); | |
2656 | if (!cur_root) { | |
2657 | ret = 1; | |
2658 | break; | |
2659 | } | |
2660 | } | |
2661 | out: | |
2662 | path->lowest_level = 0; | |
2663 | return ret; | |
2664 | } | |
2665 | ||
4313b399 CM |
2666 | /* |
2667 | * note, this releases the path | |
2668 | */ | |
98ed5174 | 2669 | static int noinline relocate_one_reference(struct btrfs_root *extent_root, |
edbd8d4e | 2670 | struct btrfs_path *path, |
0ef3e66b CM |
2671 | struct btrfs_key *extent_key, |
2672 | u64 *last_file_objectid, | |
2673 | u64 *last_file_offset, | |
2674 | u64 *last_file_root, | |
2675 | u64 last_extent) | |
edbd8d4e CM |
2676 | { |
2677 | struct inode *inode; | |
2678 | struct btrfs_root *found_root; | |
bf4ef679 CM |
2679 | struct btrfs_key root_location; |
2680 | struct btrfs_key found_key; | |
4313b399 CM |
2681 | struct btrfs_extent_ref *ref; |
2682 | u64 ref_root; | |
2683 | u64 ref_gen; | |
2684 | u64 ref_objectid; | |
2685 | u64 ref_offset; | |
edbd8d4e | 2686 | int ret; |
bf4ef679 | 2687 | int level; |
edbd8d4e | 2688 | |
4313b399 CM |
2689 | ref = btrfs_item_ptr(path->nodes[0], path->slots[0], |
2690 | struct btrfs_extent_ref); | |
2691 | ref_root = btrfs_ref_root(path->nodes[0], ref); | |
2692 | ref_gen = btrfs_ref_generation(path->nodes[0], ref); | |
2693 | ref_objectid = btrfs_ref_objectid(path->nodes[0], ref); | |
2694 | ref_offset = btrfs_ref_offset(path->nodes[0], ref); | |
2695 | btrfs_release_path(extent_root, path); | |
2696 | ||
bf4ef679 | 2697 | root_location.objectid = ref_root; |
edbd8d4e | 2698 | if (ref_gen == 0) |
bf4ef679 | 2699 | root_location.offset = 0; |
edbd8d4e | 2700 | else |
bf4ef679 CM |
2701 | root_location.offset = (u64)-1; |
2702 | root_location.type = BTRFS_ROOT_ITEM_KEY; | |
edbd8d4e CM |
2703 | |
2704 | found_root = btrfs_read_fs_root_no_name(extent_root->fs_info, | |
bf4ef679 | 2705 | &root_location); |
edbd8d4e CM |
2706 | BUG_ON(!found_root); |
2707 | ||
2708 | if (ref_objectid >= BTRFS_FIRST_FREE_OBJECTID) { | |
bf4ef679 CM |
2709 | found_key.objectid = ref_objectid; |
2710 | found_key.type = BTRFS_EXTENT_DATA_KEY; | |
2711 | found_key.offset = ref_offset; | |
2712 | level = 0; | |
2713 | ||
0ef3e66b CM |
2714 | if (last_extent == extent_key->objectid && |
2715 | *last_file_objectid == ref_objectid && | |
2716 | *last_file_offset == ref_offset && | |
2717 | *last_file_root == ref_root) | |
2718 | goto out; | |
2719 | ||
bf4ef679 CM |
2720 | ret = find_root_for_ref(extent_root, path, &found_key, |
2721 | level, 1, &found_root, | |
2722 | extent_key->objectid); | |
2723 | ||
2724 | if (ret) | |
2725 | goto out; | |
2726 | ||
0ef3e66b CM |
2727 | if (last_extent == extent_key->objectid && |
2728 | *last_file_objectid == ref_objectid && | |
2729 | *last_file_offset == ref_offset && | |
2730 | *last_file_root == ref_root) | |
2731 | goto out; | |
2732 | ||
edbd8d4e CM |
2733 | inode = btrfs_iget_locked(extent_root->fs_info->sb, |
2734 | ref_objectid, found_root); | |
2735 | if (inode->i_state & I_NEW) { | |
2736 | /* the inode and parent dir are two different roots */ | |
2737 | BTRFS_I(inode)->root = found_root; | |
2738 | BTRFS_I(inode)->location.objectid = ref_objectid; | |
2739 | BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY; | |
2740 | BTRFS_I(inode)->location.offset = 0; | |
2741 | btrfs_read_locked_inode(inode); | |
2742 | unlock_new_inode(inode); | |
2743 | ||
2744 | } | |
2745 | /* this can happen if the reference is not against | |
2746 | * the latest version of the tree root | |
2747 | */ | |
2748 | if (is_bad_inode(inode)) { | |
edbd8d4e CM |
2749 | goto out; |
2750 | } | |
0ef3e66b CM |
2751 | *last_file_objectid = inode->i_ino; |
2752 | *last_file_root = found_root->root_key.objectid; | |
2753 | *last_file_offset = ref_offset; | |
2754 | ||
edbd8d4e | 2755 | relocate_inode_pages(inode, ref_offset, extent_key->offset); |
edbd8d4e | 2756 | iput(inode); |
edbd8d4e CM |
2757 | } else { |
2758 | struct btrfs_trans_handle *trans; | |
edbd8d4e | 2759 | struct extent_buffer *eb; |
edbd8d4e CM |
2760 | int i; |
2761 | ||
edbd8d4e | 2762 | eb = read_tree_block(found_root, extent_key->objectid, |
ca7a79ad | 2763 | extent_key->offset, 0); |
925baedd | 2764 | btrfs_tree_lock(eb); |
edbd8d4e CM |
2765 | level = btrfs_header_level(eb); |
2766 | ||
2767 | if (level == 0) | |
2768 | btrfs_item_key_to_cpu(eb, &found_key, 0); | |
2769 | else | |
2770 | btrfs_node_key_to_cpu(eb, &found_key, 0); | |
2771 | ||
925baedd | 2772 | btrfs_tree_unlock(eb); |
edbd8d4e CM |
2773 | free_extent_buffer(eb); |
2774 | ||
bf4ef679 CM |
2775 | ret = find_root_for_ref(extent_root, path, &found_key, |
2776 | level, 0, &found_root, | |
2777 | extent_key->objectid); | |
2778 | ||
2779 | if (ret) | |
2780 | goto out; | |
2781 | ||
2782 | trans = btrfs_start_transaction(found_root, 1); | |
2783 | ||
edbd8d4e | 2784 | path->lowest_level = level; |
8f662a76 | 2785 | path->reada = 2; |
edbd8d4e CM |
2786 | ret = btrfs_search_slot(trans, found_root, &found_key, path, |
2787 | 0, 1); | |
2788 | path->lowest_level = 0; | |
2789 | for (i = level; i < BTRFS_MAX_LEVEL; i++) { | |
2790 | if (!path->nodes[i]) | |
2791 | break; | |
2792 | free_extent_buffer(path->nodes[i]); | |
2793 | path->nodes[i] = NULL; | |
2794 | } | |
2795 | btrfs_release_path(found_root, path); | |
0ef3e66b CM |
2796 | if (found_root == found_root->fs_info->extent_root) |
2797 | btrfs_extent_post_op(trans, found_root); | |
edbd8d4e CM |
2798 | btrfs_end_transaction(trans, found_root); |
2799 | } | |
2800 | ||
2801 | out: | |
2802 | return 0; | |
2803 | } | |
2804 | ||
a061fc8d CM |
2805 | static int noinline del_extent_zero(struct btrfs_root *extent_root, |
2806 | struct btrfs_path *path, | |
2807 | struct btrfs_key *extent_key) | |
2808 | { | |
2809 | int ret; | |
2810 | struct btrfs_trans_handle *trans; | |
2811 | ||
2812 | trans = btrfs_start_transaction(extent_root, 1); | |
2813 | ret = btrfs_search_slot(trans, extent_root, extent_key, path, -1, 1); | |
2814 | if (ret > 0) { | |
2815 | ret = -EIO; | |
2816 | goto out; | |
2817 | } | |
2818 | if (ret < 0) | |
2819 | goto out; | |
2820 | ret = btrfs_del_item(trans, extent_root, path); | |
2821 | out: | |
2822 | btrfs_end_transaction(trans, extent_root); | |
2823 | return ret; | |
2824 | } | |
2825 | ||
98ed5174 CM |
2826 | static int noinline relocate_one_extent(struct btrfs_root *extent_root, |
2827 | struct btrfs_path *path, | |
2828 | struct btrfs_key *extent_key) | |
edbd8d4e CM |
2829 | { |
2830 | struct btrfs_key key; | |
2831 | struct btrfs_key found_key; | |
edbd8d4e | 2832 | struct extent_buffer *leaf; |
0ef3e66b CM |
2833 | u64 last_file_objectid = 0; |
2834 | u64 last_file_root = 0; | |
2835 | u64 last_file_offset = (u64)-1; | |
2836 | u64 last_extent = 0; | |
edbd8d4e CM |
2837 | u32 nritems; |
2838 | u32 item_size; | |
2839 | int ret = 0; | |
2840 | ||
a061fc8d CM |
2841 | if (extent_key->objectid == 0) { |
2842 | ret = del_extent_zero(extent_root, path, extent_key); | |
2843 | goto out; | |
2844 | } | |
edbd8d4e CM |
2845 | key.objectid = extent_key->objectid; |
2846 | key.type = BTRFS_EXTENT_REF_KEY; | |
2847 | key.offset = 0; | |
2848 | ||
2849 | while(1) { | |
2850 | ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0); | |
2851 | ||
edbd8d4e CM |
2852 | if (ret < 0) |
2853 | goto out; | |
2854 | ||
2855 | ret = 0; | |
2856 | leaf = path->nodes[0]; | |
2857 | nritems = btrfs_header_nritems(leaf); | |
a061fc8d CM |
2858 | if (path->slots[0] == nritems) { |
2859 | ret = btrfs_next_leaf(extent_root, path); | |
2860 | if (ret > 0) { | |
2861 | ret = 0; | |
2862 | goto out; | |
2863 | } | |
2864 | if (ret < 0) | |
2865 | goto out; | |
bf4ef679 | 2866 | leaf = path->nodes[0]; |
a061fc8d | 2867 | } |
edbd8d4e CM |
2868 | |
2869 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
a061fc8d | 2870 | if (found_key.objectid != extent_key->objectid) { |
edbd8d4e | 2871 | break; |
a061fc8d | 2872 | } |
edbd8d4e | 2873 | |
a061fc8d | 2874 | if (found_key.type != BTRFS_EXTENT_REF_KEY) { |
edbd8d4e | 2875 | break; |
a061fc8d | 2876 | } |
edbd8d4e CM |
2877 | |
2878 | key.offset = found_key.offset + 1; | |
2879 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
2880 | ||
0ef3e66b CM |
2881 | ret = relocate_one_reference(extent_root, path, extent_key, |
2882 | &last_file_objectid, | |
2883 | &last_file_offset, | |
2884 | &last_file_root, last_extent); | |
edbd8d4e CM |
2885 | if (ret) |
2886 | goto out; | |
0ef3e66b | 2887 | last_extent = extent_key->objectid; |
edbd8d4e CM |
2888 | } |
2889 | ret = 0; | |
2890 | out: | |
2891 | btrfs_release_path(extent_root, path); | |
2892 | return ret; | |
2893 | } | |
2894 | ||
ec44a35c CM |
2895 | static u64 update_block_group_flags(struct btrfs_root *root, u64 flags) |
2896 | { | |
2897 | u64 num_devices; | |
2898 | u64 stripped = BTRFS_BLOCK_GROUP_RAID0 | | |
2899 | BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10; | |
2900 | ||
a061fc8d | 2901 | num_devices = root->fs_info->fs_devices->num_devices; |
ec44a35c CM |
2902 | if (num_devices == 1) { |
2903 | stripped |= BTRFS_BLOCK_GROUP_DUP; | |
2904 | stripped = flags & ~stripped; | |
2905 | ||
2906 | /* turn raid0 into single device chunks */ | |
2907 | if (flags & BTRFS_BLOCK_GROUP_RAID0) | |
2908 | return stripped; | |
2909 | ||
2910 | /* turn mirroring into duplication */ | |
2911 | if (flags & (BTRFS_BLOCK_GROUP_RAID1 | | |
2912 | BTRFS_BLOCK_GROUP_RAID10)) | |
2913 | return stripped | BTRFS_BLOCK_GROUP_DUP; | |
2914 | return flags; | |
2915 | } else { | |
2916 | /* they already had raid on here, just return */ | |
ec44a35c CM |
2917 | if (flags & stripped) |
2918 | return flags; | |
2919 | ||
2920 | stripped |= BTRFS_BLOCK_GROUP_DUP; | |
2921 | stripped = flags & ~stripped; | |
2922 | ||
2923 | /* switch duplicated blocks with raid1 */ | |
2924 | if (flags & BTRFS_BLOCK_GROUP_DUP) | |
2925 | return stripped | BTRFS_BLOCK_GROUP_RAID1; | |
2926 | ||
2927 | /* turn single device chunks into raid0 */ | |
2928 | return stripped | BTRFS_BLOCK_GROUP_RAID0; | |
2929 | } | |
2930 | return flags; | |
2931 | } | |
2932 | ||
0ef3e66b CM |
2933 | int __alloc_chunk_for_shrink(struct btrfs_root *root, |
2934 | struct btrfs_block_group_cache *shrink_block_group, | |
2935 | int force) | |
2936 | { | |
2937 | struct btrfs_trans_handle *trans; | |
2938 | u64 new_alloc_flags; | |
2939 | u64 calc; | |
2940 | ||
2941 | if (btrfs_block_group_used(&shrink_block_group->item) > 0) { | |
2942 | ||
2943 | trans = btrfs_start_transaction(root, 1); | |
2944 | new_alloc_flags = update_block_group_flags(root, | |
2945 | shrink_block_group->flags); | |
2946 | if (new_alloc_flags != shrink_block_group->flags) { | |
2947 | calc = | |
2948 | btrfs_block_group_used(&shrink_block_group->item); | |
2949 | } else { | |
2950 | calc = shrink_block_group->key.offset; | |
2951 | } | |
2952 | do_chunk_alloc(trans, root->fs_info->extent_root, | |
2953 | calc + 2 * 1024 * 1024, new_alloc_flags, force); | |
2954 | btrfs_end_transaction(trans, root); | |
2955 | } | |
2956 | return 0; | |
2957 | } | |
2958 | ||
8f18cf13 | 2959 | int btrfs_shrink_extent_tree(struct btrfs_root *root, u64 shrink_start) |
edbd8d4e CM |
2960 | { |
2961 | struct btrfs_trans_handle *trans; | |
2962 | struct btrfs_root *tree_root = root->fs_info->tree_root; | |
2963 | struct btrfs_path *path; | |
2964 | u64 cur_byte; | |
2965 | u64 total_found; | |
8f18cf13 CM |
2966 | u64 shrink_last_byte; |
2967 | struct btrfs_block_group_cache *shrink_block_group; | |
edbd8d4e | 2968 | struct btrfs_fs_info *info = root->fs_info; |
edbd8d4e | 2969 | struct btrfs_key key; |
73e48b27 | 2970 | struct btrfs_key found_key; |
edbd8d4e CM |
2971 | struct extent_buffer *leaf; |
2972 | u32 nritems; | |
2973 | int ret; | |
a061fc8d | 2974 | int progress; |
edbd8d4e | 2975 | |
925baedd | 2976 | mutex_lock(&root->fs_info->alloc_mutex); |
8f18cf13 CM |
2977 | shrink_block_group = btrfs_lookup_block_group(root->fs_info, |
2978 | shrink_start); | |
2979 | BUG_ON(!shrink_block_group); | |
2980 | ||
0ef3e66b CM |
2981 | shrink_last_byte = shrink_block_group->key.objectid + |
2982 | shrink_block_group->key.offset; | |
8f18cf13 CM |
2983 | |
2984 | shrink_block_group->space_info->total_bytes -= | |
2985 | shrink_block_group->key.offset; | |
edbd8d4e CM |
2986 | path = btrfs_alloc_path(); |
2987 | root = root->fs_info->extent_root; | |
8f662a76 | 2988 | path->reada = 2; |
edbd8d4e | 2989 | |
323da79c CM |
2990 | printk("btrfs relocating block group %llu flags %llu\n", |
2991 | (unsigned long long)shrink_start, | |
2992 | (unsigned long long)shrink_block_group->flags); | |
2993 | ||
0ef3e66b CM |
2994 | __alloc_chunk_for_shrink(root, shrink_block_group, 1); |
2995 | ||
edbd8d4e | 2996 | again: |
323da79c | 2997 | |
8f18cf13 CM |
2998 | shrink_block_group->ro = 1; |
2999 | ||
edbd8d4e | 3000 | total_found = 0; |
a061fc8d | 3001 | progress = 0; |
8f18cf13 | 3002 | key.objectid = shrink_start; |
edbd8d4e CM |
3003 | key.offset = 0; |
3004 | key.type = 0; | |
73e48b27 | 3005 | cur_byte = key.objectid; |
4313b399 | 3006 | |
73e48b27 Y |
3007 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
3008 | if (ret < 0) | |
3009 | goto out; | |
3010 | ||
0b86a832 | 3011 | ret = btrfs_previous_item(root, path, 0, BTRFS_EXTENT_ITEM_KEY); |
73e48b27 Y |
3012 | if (ret < 0) |
3013 | goto out; | |
8f18cf13 | 3014 | |
73e48b27 Y |
3015 | if (ret == 0) { |
3016 | leaf = path->nodes[0]; | |
3017 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
8f18cf13 CM |
3018 | if (found_key.objectid + found_key.offset > shrink_start && |
3019 | found_key.objectid < shrink_last_byte) { | |
73e48b27 Y |
3020 | cur_byte = found_key.objectid; |
3021 | key.objectid = cur_byte; | |
3022 | } | |
3023 | } | |
3024 | btrfs_release_path(root, path); | |
3025 | ||
3026 | while(1) { | |
edbd8d4e CM |
3027 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
3028 | if (ret < 0) | |
3029 | goto out; | |
73e48b27 | 3030 | |
edbd8d4e | 3031 | leaf = path->nodes[0]; |
73e48b27 Y |
3032 | nritems = btrfs_header_nritems(leaf); |
3033 | next: | |
3034 | if (path->slots[0] >= nritems) { | |
3035 | ret = btrfs_next_leaf(root, path); | |
3036 | if (ret < 0) | |
3037 | goto out; | |
3038 | if (ret == 1) { | |
3039 | ret = 0; | |
3040 | break; | |
edbd8d4e | 3041 | } |
73e48b27 Y |
3042 | leaf = path->nodes[0]; |
3043 | nritems = btrfs_header_nritems(leaf); | |
edbd8d4e | 3044 | } |
73e48b27 Y |
3045 | |
3046 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
725c8463 | 3047 | |
8f18cf13 CM |
3048 | if (found_key.objectid >= shrink_last_byte) |
3049 | break; | |
3050 | ||
725c8463 CM |
3051 | if (progress && need_resched()) { |
3052 | memcpy(&key, &found_key, sizeof(key)); | |
725c8463 | 3053 | cond_resched(); |
725c8463 CM |
3054 | btrfs_release_path(root, path); |
3055 | btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
3056 | progress = 0; | |
3057 | goto next; | |
3058 | } | |
3059 | progress = 1; | |
3060 | ||
73e48b27 Y |
3061 | if (btrfs_key_type(&found_key) != BTRFS_EXTENT_ITEM_KEY || |
3062 | found_key.objectid + found_key.offset <= cur_byte) { | |
0ef3e66b CM |
3063 | memcpy(&key, &found_key, sizeof(key)); |
3064 | key.offset++; | |
edbd8d4e | 3065 | path->slots[0]++; |
edbd8d4e CM |
3066 | goto next; |
3067 | } | |
73e48b27 | 3068 | |
edbd8d4e CM |
3069 | total_found++; |
3070 | cur_byte = found_key.objectid + found_key.offset; | |
3071 | key.objectid = cur_byte; | |
3072 | btrfs_release_path(root, path); | |
3073 | ret = relocate_one_extent(root, path, &found_key); | |
0ef3e66b | 3074 | __alloc_chunk_for_shrink(root, shrink_block_group, 0); |
edbd8d4e CM |
3075 | } |
3076 | ||
3077 | btrfs_release_path(root, path); | |
3078 | ||
3079 | if (total_found > 0) { | |
323da79c CM |
3080 | printk("btrfs relocate found %llu last extent was %llu\n", |
3081 | (unsigned long long)total_found, | |
3082 | (unsigned long long)found_key.objectid); | |
edbd8d4e CM |
3083 | trans = btrfs_start_transaction(tree_root, 1); |
3084 | btrfs_commit_transaction(trans, tree_root); | |
3085 | ||
edbd8d4e | 3086 | btrfs_clean_old_snapshots(tree_root); |
edbd8d4e CM |
3087 | |
3088 | trans = btrfs_start_transaction(tree_root, 1); | |
3089 | btrfs_commit_transaction(trans, tree_root); | |
3090 | goto again; | |
3091 | } | |
3092 | ||
8f18cf13 CM |
3093 | /* |
3094 | * we've freed all the extents, now remove the block | |
3095 | * group item from the tree | |
3096 | */ | |
edbd8d4e | 3097 | trans = btrfs_start_transaction(root, 1); |
8f18cf13 | 3098 | memcpy(&key, &shrink_block_group->key, sizeof(key)); |
1372f8e6 | 3099 | |
8f18cf13 CM |
3100 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
3101 | if (ret > 0) | |
3102 | ret = -EIO; | |
3103 | if (ret < 0) | |
3104 | goto out; | |
73e48b27 | 3105 | |
0ef3e66b CM |
3106 | clear_extent_bits(&info->block_group_cache, key.objectid, |
3107 | key.objectid + key.offset - 1, | |
8f18cf13 | 3108 | (unsigned int)-1, GFP_NOFS); |
edbd8d4e | 3109 | |
0ef3e66b CM |
3110 | |
3111 | clear_extent_bits(&info->free_space_cache, | |
3112 | key.objectid, key.objectid + key.offset - 1, | |
3113 | (unsigned int)-1, GFP_NOFS); | |
3114 | ||
3115 | memset(shrink_block_group, 0, sizeof(*shrink_block_group)); | |
3116 | kfree(shrink_block_group); | |
3117 | ||
8f18cf13 | 3118 | btrfs_del_item(trans, root, path); |
edbd8d4e | 3119 | btrfs_commit_transaction(trans, root); |
0ef3e66b CM |
3120 | |
3121 | /* the code to unpin extents might set a few bits in the free | |
3122 | * space cache for this range again | |
3123 | */ | |
3124 | clear_extent_bits(&info->free_space_cache, | |
3125 | key.objectid, key.objectid + key.offset - 1, | |
3126 | (unsigned int)-1, GFP_NOFS); | |
edbd8d4e CM |
3127 | out: |
3128 | btrfs_free_path(path); | |
925baedd | 3129 | mutex_unlock(&root->fs_info->alloc_mutex); |
edbd8d4e CM |
3130 | return ret; |
3131 | } | |
3132 | ||
0b86a832 CM |
3133 | int find_first_block_group(struct btrfs_root *root, struct btrfs_path *path, |
3134 | struct btrfs_key *key) | |
3135 | { | |
925baedd | 3136 | int ret = 0; |
0b86a832 CM |
3137 | struct btrfs_key found_key; |
3138 | struct extent_buffer *leaf; | |
3139 | int slot; | |
edbd8d4e | 3140 | |
0b86a832 CM |
3141 | ret = btrfs_search_slot(NULL, root, key, path, 0, 0); |
3142 | if (ret < 0) | |
925baedd CM |
3143 | goto out; |
3144 | ||
0b86a832 CM |
3145 | while(1) { |
3146 | slot = path->slots[0]; | |
edbd8d4e | 3147 | leaf = path->nodes[0]; |
0b86a832 CM |
3148 | if (slot >= btrfs_header_nritems(leaf)) { |
3149 | ret = btrfs_next_leaf(root, path); | |
3150 | if (ret == 0) | |
3151 | continue; | |
3152 | if (ret < 0) | |
925baedd | 3153 | goto out; |
0b86a832 | 3154 | break; |
edbd8d4e | 3155 | } |
0b86a832 | 3156 | btrfs_item_key_to_cpu(leaf, &found_key, slot); |
edbd8d4e | 3157 | |
0b86a832 | 3158 | if (found_key.objectid >= key->objectid && |
925baedd CM |
3159 | found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) { |
3160 | ret = 0; | |
3161 | goto out; | |
3162 | } | |
0b86a832 | 3163 | path->slots[0]++; |
edbd8d4e | 3164 | } |
0b86a832 | 3165 | ret = -ENOENT; |
925baedd | 3166 | out: |
0b86a832 | 3167 | return ret; |
edbd8d4e CM |
3168 | } |
3169 | ||
9078a3e1 CM |
3170 | int btrfs_read_block_groups(struct btrfs_root *root) |
3171 | { | |
3172 | struct btrfs_path *path; | |
3173 | int ret; | |
96b5179d | 3174 | int bit; |
9078a3e1 | 3175 | struct btrfs_block_group_cache *cache; |
be744175 | 3176 | struct btrfs_fs_info *info = root->fs_info; |
6324fbf3 | 3177 | struct btrfs_space_info *space_info; |
d1310b2e | 3178 | struct extent_io_tree *block_group_cache; |
9078a3e1 CM |
3179 | struct btrfs_key key; |
3180 | struct btrfs_key found_key; | |
5f39d397 | 3181 | struct extent_buffer *leaf; |
96b5179d CM |
3182 | |
3183 | block_group_cache = &info->block_group_cache; | |
be744175 | 3184 | root = info->extent_root; |
9078a3e1 | 3185 | key.objectid = 0; |
0b86a832 | 3186 | key.offset = 0; |
9078a3e1 | 3187 | btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY); |
9078a3e1 CM |
3188 | path = btrfs_alloc_path(); |
3189 | if (!path) | |
3190 | return -ENOMEM; | |
3191 | ||
925baedd | 3192 | mutex_lock(&root->fs_info->alloc_mutex); |
9078a3e1 | 3193 | while(1) { |
0b86a832 CM |
3194 | ret = find_first_block_group(root, path, &key); |
3195 | if (ret > 0) { | |
3196 | ret = 0; | |
3197 | goto error; | |
9078a3e1 | 3198 | } |
0b86a832 CM |
3199 | if (ret != 0) |
3200 | goto error; | |
3201 | ||
5f39d397 CM |
3202 | leaf = path->nodes[0]; |
3203 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
8f18cf13 | 3204 | cache = kzalloc(sizeof(*cache), GFP_NOFS); |
9078a3e1 | 3205 | if (!cache) { |
0b86a832 | 3206 | ret = -ENOMEM; |
9078a3e1 CM |
3207 | break; |
3208 | } | |
3e1ad54f | 3209 | |
5f39d397 CM |
3210 | read_extent_buffer(leaf, &cache->item, |
3211 | btrfs_item_ptr_offset(leaf, path->slots[0]), | |
3212 | sizeof(cache->item)); | |
9078a3e1 | 3213 | memcpy(&cache->key, &found_key, sizeof(found_key)); |
0b86a832 | 3214 | |
9078a3e1 CM |
3215 | key.objectid = found_key.objectid + found_key.offset; |
3216 | btrfs_release_path(root, path); | |
0b86a832 CM |
3217 | cache->flags = btrfs_block_group_flags(&cache->item); |
3218 | bit = 0; | |
3219 | if (cache->flags & BTRFS_BLOCK_GROUP_DATA) { | |
96b5179d | 3220 | bit = BLOCK_GROUP_DATA; |
0b86a832 CM |
3221 | } else if (cache->flags & BTRFS_BLOCK_GROUP_SYSTEM) { |
3222 | bit = BLOCK_GROUP_SYSTEM; | |
3223 | } else if (cache->flags & BTRFS_BLOCK_GROUP_METADATA) { | |
96b5179d | 3224 | bit = BLOCK_GROUP_METADATA; |
31f3c99b | 3225 | } |
8790d502 | 3226 | set_avail_alloc_bits(info, cache->flags); |
96b5179d | 3227 | |
6324fbf3 CM |
3228 | ret = update_space_info(info, cache->flags, found_key.offset, |
3229 | btrfs_block_group_used(&cache->item), | |
3230 | &space_info); | |
3231 | BUG_ON(ret); | |
3232 | cache->space_info = space_info; | |
3233 | ||
96b5179d CM |
3234 | /* use EXTENT_LOCKED to prevent merging */ |
3235 | set_extent_bits(block_group_cache, found_key.objectid, | |
3236 | found_key.objectid + found_key.offset - 1, | |
3237 | bit | EXTENT_LOCKED, GFP_NOFS); | |
3238 | set_state_private(block_group_cache, found_key.objectid, | |
ae2f5411 | 3239 | (unsigned long)cache); |
96b5179d | 3240 | |
9078a3e1 | 3241 | if (key.objectid >= |
db94535d | 3242 | btrfs_super_total_bytes(&info->super_copy)) |
9078a3e1 CM |
3243 | break; |
3244 | } | |
0b86a832 CM |
3245 | ret = 0; |
3246 | error: | |
9078a3e1 | 3247 | btrfs_free_path(path); |
925baedd | 3248 | mutex_unlock(&root->fs_info->alloc_mutex); |
0b86a832 | 3249 | return ret; |
9078a3e1 | 3250 | } |
6324fbf3 CM |
3251 | |
3252 | int btrfs_make_block_group(struct btrfs_trans_handle *trans, | |
3253 | struct btrfs_root *root, u64 bytes_used, | |
e17cade2 | 3254 | u64 type, u64 chunk_objectid, u64 chunk_offset, |
6324fbf3 CM |
3255 | u64 size) |
3256 | { | |
3257 | int ret; | |
3258 | int bit = 0; | |
3259 | struct btrfs_root *extent_root; | |
3260 | struct btrfs_block_group_cache *cache; | |
3261 | struct extent_io_tree *block_group_cache; | |
3262 | ||
3263 | extent_root = root->fs_info->extent_root; | |
3264 | block_group_cache = &root->fs_info->block_group_cache; | |
3265 | ||
8f18cf13 | 3266 | cache = kzalloc(sizeof(*cache), GFP_NOFS); |
6324fbf3 | 3267 | BUG_ON(!cache); |
e17cade2 | 3268 | cache->key.objectid = chunk_offset; |
6324fbf3 | 3269 | cache->key.offset = size; |
6324fbf3 | 3270 | btrfs_set_key_type(&cache->key, BTRFS_BLOCK_GROUP_ITEM_KEY); |
0ef3e66b | 3271 | |
6324fbf3 | 3272 | btrfs_set_block_group_used(&cache->item, bytes_used); |
6324fbf3 CM |
3273 | btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid); |
3274 | cache->flags = type; | |
3275 | btrfs_set_block_group_flags(&cache->item, type); | |
3276 | ||
3277 | ret = update_space_info(root->fs_info, cache->flags, size, bytes_used, | |
3278 | &cache->space_info); | |
3279 | BUG_ON(ret); | |
3280 | ||
d18a2c44 | 3281 | bit = block_group_state_bits(type); |
e17cade2 CM |
3282 | set_extent_bits(block_group_cache, chunk_offset, |
3283 | chunk_offset + size - 1, | |
6324fbf3 | 3284 | bit | EXTENT_LOCKED, GFP_NOFS); |
6324fbf3 | 3285 | |
e17cade2 CM |
3286 | set_state_private(block_group_cache, chunk_offset, |
3287 | (unsigned long)cache); | |
6324fbf3 CM |
3288 | ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item, |
3289 | sizeof(cache->item)); | |
3290 | BUG_ON(ret); | |
3291 | ||
3292 | finish_current_insert(trans, extent_root); | |
3293 | ret = del_pending_extents(trans, extent_root); | |
3294 | BUG_ON(ret); | |
d18a2c44 | 3295 | set_avail_alloc_bits(extent_root->fs_info, type); |
925baedd | 3296 | |
6324fbf3 CM |
3297 | return 0; |
3298 | } |