Commit | Line | Data |
---|---|---|
6cbd5570 CM |
1 | /* |
2 | * Copyright (C) 2007 Oracle. All rights reserved. | |
3 | * | |
4 | * This program is free software; you can redistribute it and/or | |
5 | * modify it under the terms of the GNU General Public | |
6 | * License v2 as published by the Free Software Foundation. | |
7 | * | |
8 | * This program is distributed in the hope that it will be useful, | |
9 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
10 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
11 | * General Public License for more details. | |
12 | * | |
13 | * You should have received a copy of the GNU General Public | |
14 | * License along with this program; if not, write to the | |
15 | * Free Software Foundation, Inc., 59 Temple Place - Suite 330, | |
16 | * Boston, MA 021110-1307, USA. | |
17 | */ | |
ec6b910f | 18 | #include <linux/sched.h> |
edbd8d4e | 19 | #include <linux/pagemap.h> |
ec44a35c | 20 | #include <linux/writeback.h> |
74493f7a | 21 | #include "hash.h" |
a5eb62e3 | 22 | #include "crc32c.h" |
fec577fb CM |
23 | #include "ctree.h" |
24 | #include "disk-io.h" | |
25 | #include "print-tree.h" | |
e089f05c | 26 | #include "transaction.h" |
0b86a832 | 27 | #include "volumes.h" |
925baedd | 28 | #include "locking.h" |
fec577fb | 29 | |
0b86a832 | 30 | #define BLOCK_GROUP_DATA EXTENT_WRITEBACK |
96b5179d | 31 | #define BLOCK_GROUP_METADATA EXTENT_UPTODATE |
0b86a832 CM |
32 | #define BLOCK_GROUP_SYSTEM EXTENT_NEW |
33 | ||
96b5179d CM |
34 | #define BLOCK_GROUP_DIRTY EXTENT_DIRTY |
35 | ||
e089f05c CM |
36 | static int finish_current_insert(struct btrfs_trans_handle *trans, struct |
37 | btrfs_root *extent_root); | |
e20d96d6 CM |
38 | static int del_pending_extents(struct btrfs_trans_handle *trans, struct |
39 | btrfs_root *extent_root); | |
925baedd CM |
40 | static struct btrfs_block_group_cache * |
41 | __btrfs_find_block_group(struct btrfs_root *root, | |
42 | struct btrfs_block_group_cache *hint, | |
43 | u64 search_start, int data, int owner); | |
d548ee51 | 44 | |
925baedd CM |
45 | void maybe_lock_mutex(struct btrfs_root *root) |
46 | { | |
47 | if (root != root->fs_info->extent_root && | |
48 | root != root->fs_info->chunk_root && | |
49 | root != root->fs_info->dev_root) { | |
50 | mutex_lock(&root->fs_info->alloc_mutex); | |
51 | } | |
52 | } | |
53 | ||
54 | void maybe_unlock_mutex(struct btrfs_root *root) | |
55 | { | |
56 | if (root != root->fs_info->extent_root && | |
57 | root != root->fs_info->chunk_root && | |
58 | root != root->fs_info->dev_root) { | |
59 | mutex_unlock(&root->fs_info->alloc_mutex); | |
60 | } | |
61 | } | |
fec577fb | 62 | |
e37c9e69 CM |
63 | static int cache_block_group(struct btrfs_root *root, |
64 | struct btrfs_block_group_cache *block_group) | |
65 | { | |
66 | struct btrfs_path *path; | |
67 | int ret; | |
68 | struct btrfs_key key; | |
5f39d397 | 69 | struct extent_buffer *leaf; |
d1310b2e | 70 | struct extent_io_tree *free_space_cache; |
e37c9e69 | 71 | int slot; |
e37c9e69 CM |
72 | u64 last = 0; |
73 | u64 hole_size; | |
7d7d6068 | 74 | u64 first_free; |
e37c9e69 CM |
75 | int found = 0; |
76 | ||
00f5c795 CM |
77 | if (!block_group) |
78 | return 0; | |
79 | ||
e37c9e69 | 80 | root = root->fs_info->extent_root; |
f510cfec | 81 | free_space_cache = &root->fs_info->free_space_cache; |
e37c9e69 CM |
82 | |
83 | if (block_group->cached) | |
84 | return 0; | |
f510cfec | 85 | |
e37c9e69 CM |
86 | path = btrfs_alloc_path(); |
87 | if (!path) | |
88 | return -ENOMEM; | |
7d7d6068 | 89 | |
2cc58cf2 | 90 | path->reada = 2; |
5cd57b2c CM |
91 | /* |
92 | * we get into deadlocks with paths held by callers of this function. | |
93 | * since the alloc_mutex is protecting things right now, just | |
94 | * skip the locking here | |
95 | */ | |
96 | path->skip_locking = 1; | |
7d7d6068 | 97 | first_free = block_group->key.objectid; |
e37c9e69 | 98 | key.objectid = block_group->key.objectid; |
e37c9e69 CM |
99 | key.offset = 0; |
100 | btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY); | |
101 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
102 | if (ret < 0) | |
103 | return ret; | |
0b86a832 | 104 | ret = btrfs_previous_item(root, path, 0, BTRFS_EXTENT_ITEM_KEY); |
d548ee51 Y |
105 | if (ret < 0) |
106 | return ret; | |
107 | if (ret == 0) { | |
108 | leaf = path->nodes[0]; | |
109 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
110 | if (key.objectid + key.offset > first_free) | |
111 | first_free = key.objectid + key.offset; | |
112 | } | |
e37c9e69 | 113 | while(1) { |
5f39d397 | 114 | leaf = path->nodes[0]; |
e37c9e69 | 115 | slot = path->slots[0]; |
5f39d397 | 116 | if (slot >= btrfs_header_nritems(leaf)) { |
e37c9e69 | 117 | ret = btrfs_next_leaf(root, path); |
54aa1f4d CM |
118 | if (ret < 0) |
119 | goto err; | |
de428b63 | 120 | if (ret == 0) { |
e37c9e69 | 121 | continue; |
de428b63 | 122 | } else { |
e37c9e69 CM |
123 | break; |
124 | } | |
125 | } | |
5f39d397 | 126 | btrfs_item_key_to_cpu(leaf, &key, slot); |
7d7d6068 | 127 | if (key.objectid < block_group->key.objectid) { |
7d7d6068 Y |
128 | goto next; |
129 | } | |
e37c9e69 CM |
130 | if (key.objectid >= block_group->key.objectid + |
131 | block_group->key.offset) { | |
e37c9e69 CM |
132 | break; |
133 | } | |
7d7d6068 | 134 | |
e37c9e69 CM |
135 | if (btrfs_key_type(&key) == BTRFS_EXTENT_ITEM_KEY) { |
136 | if (!found) { | |
7d7d6068 | 137 | last = first_free; |
e37c9e69 | 138 | found = 1; |
e37c9e69 | 139 | } |
f510cfec CM |
140 | if (key.objectid > last) { |
141 | hole_size = key.objectid - last; | |
142 | set_extent_dirty(free_space_cache, last, | |
143 | last + hole_size - 1, | |
144 | GFP_NOFS); | |
7d7d6068 Y |
145 | } |
146 | last = key.objectid + key.offset; | |
e37c9e69 | 147 | } |
7d7d6068 | 148 | next: |
e37c9e69 CM |
149 | path->slots[0]++; |
150 | } | |
151 | ||
7d7d6068 Y |
152 | if (!found) |
153 | last = first_free; | |
154 | if (block_group->key.objectid + | |
155 | block_group->key.offset > last) { | |
156 | hole_size = block_group->key.objectid + | |
157 | block_group->key.offset - last; | |
f510cfec CM |
158 | set_extent_dirty(free_space_cache, last, |
159 | last + hole_size - 1, GFP_NOFS); | |
7d7d6068 | 160 | } |
e37c9e69 | 161 | block_group->cached = 1; |
54aa1f4d | 162 | err: |
e37c9e69 CM |
163 | btrfs_free_path(path); |
164 | return 0; | |
165 | } | |
166 | ||
0ef3e66b CM |
167 | struct btrfs_block_group_cache *btrfs_lookup_first_block_group(struct |
168 | btrfs_fs_info *info, | |
169 | u64 bytenr) | |
170 | { | |
171 | struct extent_io_tree *block_group_cache; | |
172 | struct btrfs_block_group_cache *block_group = NULL; | |
173 | u64 ptr; | |
174 | u64 start; | |
175 | u64 end; | |
176 | int ret; | |
177 | ||
178 | bytenr = max_t(u64, bytenr, | |
179 | BTRFS_SUPER_INFO_OFFSET + BTRFS_SUPER_INFO_SIZE); | |
180 | block_group_cache = &info->block_group_cache; | |
181 | ret = find_first_extent_bit(block_group_cache, | |
182 | bytenr, &start, &end, | |
183 | BLOCK_GROUP_DATA | BLOCK_GROUP_METADATA | | |
184 | BLOCK_GROUP_SYSTEM); | |
185 | if (ret) { | |
186 | return NULL; | |
187 | } | |
188 | ret = get_state_private(block_group_cache, start, &ptr); | |
189 | if (ret) | |
190 | return NULL; | |
191 | ||
192 | block_group = (struct btrfs_block_group_cache *)(unsigned long)ptr; | |
193 | return block_group; | |
194 | } | |
195 | ||
5276aeda CM |
196 | struct btrfs_block_group_cache *btrfs_lookup_block_group(struct |
197 | btrfs_fs_info *info, | |
db94535d | 198 | u64 bytenr) |
be744175 | 199 | { |
d1310b2e | 200 | struct extent_io_tree *block_group_cache; |
96b5179d CM |
201 | struct btrfs_block_group_cache *block_group = NULL; |
202 | u64 ptr; | |
203 | u64 start; | |
204 | u64 end; | |
be744175 CM |
205 | int ret; |
206 | ||
a061fc8d CM |
207 | bytenr = max_t(u64, bytenr, |
208 | BTRFS_SUPER_INFO_OFFSET + BTRFS_SUPER_INFO_SIZE); | |
96b5179d CM |
209 | block_group_cache = &info->block_group_cache; |
210 | ret = find_first_extent_bit(block_group_cache, | |
db94535d | 211 | bytenr, &start, &end, |
0b86a832 CM |
212 | BLOCK_GROUP_DATA | BLOCK_GROUP_METADATA | |
213 | BLOCK_GROUP_SYSTEM); | |
be744175 | 214 | if (ret) { |
96b5179d | 215 | return NULL; |
be744175 | 216 | } |
96b5179d CM |
217 | ret = get_state_private(block_group_cache, start, &ptr); |
218 | if (ret) | |
219 | return NULL; | |
220 | ||
ae2f5411 | 221 | block_group = (struct btrfs_block_group_cache *)(unsigned long)ptr; |
5cf66426 | 222 | if (block_group->key.objectid <= bytenr && bytenr < |
96b5179d CM |
223 | block_group->key.objectid + block_group->key.offset) |
224 | return block_group; | |
be744175 CM |
225 | return NULL; |
226 | } | |
0b86a832 CM |
227 | |
228 | static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits) | |
229 | { | |
593060d7 | 230 | return (cache->flags & bits) == bits; |
0b86a832 CM |
231 | } |
232 | ||
233 | static int noinline find_search_start(struct btrfs_root *root, | |
98ed5174 | 234 | struct btrfs_block_group_cache **cache_ret, |
0ef3e66b | 235 | u64 *start_ret, u64 num, int data) |
e37c9e69 | 236 | { |
e37c9e69 CM |
237 | int ret; |
238 | struct btrfs_block_group_cache *cache = *cache_ret; | |
d7fc640e | 239 | struct extent_io_tree *free_space_cache; |
7d1660d4 | 240 | struct extent_state *state; |
e19caa5f | 241 | u64 last; |
f510cfec | 242 | u64 start = 0; |
257d0ce3 | 243 | u64 cache_miss = 0; |
c31f8830 | 244 | u64 total_fs_bytes; |
0b86a832 | 245 | u64 search_start = *start_ret; |
f84a8b36 | 246 | int wrapped = 0; |
e37c9e69 | 247 | |
7d9eb12c | 248 | WARN_ON(!mutex_is_locked(&root->fs_info->alloc_mutex)); |
c31f8830 | 249 | total_fs_bytes = btrfs_super_total_bytes(&root->fs_info->super_copy); |
d7fc640e CM |
250 | free_space_cache = &root->fs_info->free_space_cache; |
251 | ||
0ef3e66b CM |
252 | if (!cache) |
253 | goto out; | |
254 | ||
e37c9e69 | 255 | again: |
54aa1f4d | 256 | ret = cache_block_group(root, cache); |
0ef3e66b | 257 | if (ret) { |
54aa1f4d | 258 | goto out; |
0ef3e66b | 259 | } |
f84a8b36 | 260 | |
e19caa5f | 261 | last = max(search_start, cache->key.objectid); |
0ef3e66b | 262 | if (!block_group_bits(cache, data) || cache->ro) |
0b86a832 | 263 | goto new_group; |
e19caa5f | 264 | |
7d1660d4 CM |
265 | spin_lock_irq(&free_space_cache->lock); |
266 | state = find_first_extent_bit_state(free_space_cache, last, EXTENT_DIRTY); | |
e37c9e69 | 267 | while(1) { |
7d1660d4 | 268 | if (!state) { |
257d0ce3 CM |
269 | if (!cache_miss) |
270 | cache_miss = last; | |
7d1660d4 | 271 | spin_unlock_irq(&free_space_cache->lock); |
e19caa5f CM |
272 | goto new_group; |
273 | } | |
f510cfec | 274 | |
7d1660d4 CM |
275 | start = max(last, state->start); |
276 | last = state->end + 1; | |
257d0ce3 | 277 | if (last - start < num) { |
7d1660d4 CM |
278 | do { |
279 | state = extent_state_next(state); | |
280 | } while(state && !(state->state & EXTENT_DIRTY)); | |
f510cfec | 281 | continue; |
257d0ce3 | 282 | } |
7d1660d4 | 283 | spin_unlock_irq(&free_space_cache->lock); |
0ef3e66b | 284 | if (cache->ro) { |
8f18cf13 | 285 | goto new_group; |
0ef3e66b | 286 | } |
0b86a832 | 287 | if (start + num > cache->key.objectid + cache->key.offset) |
e37c9e69 | 288 | goto new_group; |
8790d502 | 289 | if (!block_group_bits(cache, data)) { |
611f0e00 | 290 | printk("block group bits don't match %Lu %d\n", cache->flags, data); |
8790d502 | 291 | } |
0b86a832 CM |
292 | *start_ret = start; |
293 | return 0; | |
8790d502 CM |
294 | } |
295 | out: | |
1a2b2ac7 CM |
296 | cache = btrfs_lookup_block_group(root->fs_info, search_start); |
297 | if (!cache) { | |
0b86a832 | 298 | printk("Unable to find block group for %Lu\n", search_start); |
1a2b2ac7 | 299 | WARN_ON(1); |
1a2b2ac7 | 300 | } |
0b86a832 | 301 | return -ENOSPC; |
e37c9e69 CM |
302 | |
303 | new_group: | |
e19caa5f | 304 | last = cache->key.objectid + cache->key.offset; |
f84a8b36 | 305 | wrapped: |
0ef3e66b | 306 | cache = btrfs_lookup_first_block_group(root->fs_info, last); |
c31f8830 | 307 | if (!cache || cache->key.objectid >= total_fs_bytes) { |
0e4de584 | 308 | no_cache: |
f84a8b36 CM |
309 | if (!wrapped) { |
310 | wrapped = 1; | |
311 | last = search_start; | |
f84a8b36 CM |
312 | goto wrapped; |
313 | } | |
1a2b2ac7 | 314 | goto out; |
e37c9e69 | 315 | } |
257d0ce3 CM |
316 | if (cache_miss && !cache->cached) { |
317 | cache_block_group(root, cache); | |
318 | last = cache_miss; | |
0ef3e66b | 319 | cache = btrfs_lookup_first_block_group(root->fs_info, last); |
257d0ce3 | 320 | } |
0ef3e66b | 321 | cache_miss = 0; |
925baedd | 322 | cache = __btrfs_find_block_group(root, cache, last, data, 0); |
0e4de584 CM |
323 | if (!cache) |
324 | goto no_cache; | |
e37c9e69 CM |
325 | *cache_ret = cache; |
326 | goto again; | |
327 | } | |
328 | ||
84f54cfa CM |
329 | static u64 div_factor(u64 num, int factor) |
330 | { | |
257d0ce3 CM |
331 | if (factor == 10) |
332 | return num; | |
84f54cfa CM |
333 | num *= factor; |
334 | do_div(num, 10); | |
335 | return num; | |
336 | } | |
337 | ||
6324fbf3 CM |
338 | static int block_group_state_bits(u64 flags) |
339 | { | |
340 | int bits = 0; | |
341 | if (flags & BTRFS_BLOCK_GROUP_DATA) | |
342 | bits |= BLOCK_GROUP_DATA; | |
343 | if (flags & BTRFS_BLOCK_GROUP_METADATA) | |
344 | bits |= BLOCK_GROUP_METADATA; | |
345 | if (flags & BTRFS_BLOCK_GROUP_SYSTEM) | |
346 | bits |= BLOCK_GROUP_SYSTEM; | |
347 | return bits; | |
348 | } | |
349 | ||
925baedd CM |
350 | static struct btrfs_block_group_cache * |
351 | __btrfs_find_block_group(struct btrfs_root *root, | |
352 | struct btrfs_block_group_cache *hint, | |
353 | u64 search_start, int data, int owner) | |
cd1bc465 | 354 | { |
96b5179d | 355 | struct btrfs_block_group_cache *cache; |
d1310b2e | 356 | struct extent_io_tree *block_group_cache; |
31f3c99b | 357 | struct btrfs_block_group_cache *found_group = NULL; |
cd1bc465 CM |
358 | struct btrfs_fs_info *info = root->fs_info; |
359 | u64 used; | |
31f3c99b | 360 | u64 last = 0; |
96b5179d CM |
361 | u64 start; |
362 | u64 end; | |
363 | u64 free_check; | |
364 | u64 ptr; | |
365 | int bit; | |
cd1bc465 | 366 | int ret; |
31f3c99b | 367 | int full_search = 0; |
bce4eae9 | 368 | int factor = 10; |
0ef3e66b | 369 | int wrapped = 0; |
de428b63 | 370 | |
96b5179d CM |
371 | block_group_cache = &info->block_group_cache; |
372 | ||
a236aed1 CM |
373 | if (data & BTRFS_BLOCK_GROUP_METADATA) |
374 | factor = 9; | |
be744175 | 375 | |
6324fbf3 | 376 | bit = block_group_state_bits(data); |
be744175 | 377 | |
0ef3e66b | 378 | if (search_start) { |
be744175 | 379 | struct btrfs_block_group_cache *shint; |
0ef3e66b | 380 | shint = btrfs_lookup_first_block_group(info, search_start); |
8f18cf13 | 381 | if (shint && block_group_bits(shint, data) && !shint->ro) { |
be744175 | 382 | used = btrfs_block_group_used(&shint->item); |
324ae4df Y |
383 | if (used + shint->pinned < |
384 | div_factor(shint->key.offset, factor)) { | |
be744175 CM |
385 | return shint; |
386 | } | |
387 | } | |
388 | } | |
0ef3e66b | 389 | if (hint && !hint->ro && block_group_bits(hint, data)) { |
31f3c99b | 390 | used = btrfs_block_group_used(&hint->item); |
324ae4df Y |
391 | if (used + hint->pinned < |
392 | div_factor(hint->key.offset, factor)) { | |
31f3c99b CM |
393 | return hint; |
394 | } | |
e19caa5f | 395 | last = hint->key.objectid + hint->key.offset; |
31f3c99b | 396 | } else { |
e37c9e69 | 397 | if (hint) |
0ef3e66b | 398 | last = max(hint->key.objectid, search_start); |
e37c9e69 | 399 | else |
0ef3e66b | 400 | last = search_start; |
31f3c99b | 401 | } |
31f3c99b | 402 | again: |
cd1bc465 | 403 | while(1) { |
96b5179d CM |
404 | ret = find_first_extent_bit(block_group_cache, last, |
405 | &start, &end, bit); | |
406 | if (ret) | |
cd1bc465 | 407 | break; |
96b5179d CM |
408 | |
409 | ret = get_state_private(block_group_cache, start, &ptr); | |
0ef3e66b CM |
410 | if (ret) { |
411 | last = end + 1; | |
412 | continue; | |
413 | } | |
96b5179d | 414 | |
ae2f5411 | 415 | cache = (struct btrfs_block_group_cache *)(unsigned long)ptr; |
96b5179d CM |
416 | last = cache->key.objectid + cache->key.offset; |
417 | used = btrfs_block_group_used(&cache->item); | |
418 | ||
8f18cf13 | 419 | if (!cache->ro && block_group_bits(cache, data)) { |
0ef3e66b | 420 | free_check = div_factor(cache->key.offset, factor); |
8790d502 CM |
421 | if (used + cache->pinned < free_check) { |
422 | found_group = cache; | |
423 | goto found; | |
424 | } | |
6324fbf3 | 425 | } |
de428b63 | 426 | cond_resched(); |
cd1bc465 | 427 | } |
0ef3e66b CM |
428 | if (!wrapped) { |
429 | last = search_start; | |
430 | wrapped = 1; | |
431 | goto again; | |
432 | } | |
433 | if (!full_search && factor < 10) { | |
be744175 | 434 | last = search_start; |
31f3c99b | 435 | full_search = 1; |
0ef3e66b | 436 | factor = 10; |
31f3c99b CM |
437 | goto again; |
438 | } | |
be744175 | 439 | found: |
31f3c99b | 440 | return found_group; |
cd1bc465 CM |
441 | } |
442 | ||
925baedd CM |
443 | struct btrfs_block_group_cache *btrfs_find_block_group(struct btrfs_root *root, |
444 | struct btrfs_block_group_cache | |
445 | *hint, u64 search_start, | |
446 | int data, int owner) | |
447 | { | |
448 | ||
449 | struct btrfs_block_group_cache *ret; | |
450 | mutex_lock(&root->fs_info->alloc_mutex); | |
451 | ret = __btrfs_find_block_group(root, hint, search_start, data, owner); | |
452 | mutex_unlock(&root->fs_info->alloc_mutex); | |
453 | return ret; | |
454 | } | |
7bb86316 | 455 | static u64 hash_extent_ref(u64 root_objectid, u64 ref_generation, |
74493f7a CM |
456 | u64 owner, u64 owner_offset) |
457 | { | |
458 | u32 high_crc = ~(u32)0; | |
459 | u32 low_crc = ~(u32)0; | |
460 | __le64 lenum; | |
74493f7a | 461 | lenum = cpu_to_le64(root_objectid); |
a5eb62e3 | 462 | high_crc = btrfs_crc32c(high_crc, &lenum, sizeof(lenum)); |
7bb86316 | 463 | lenum = cpu_to_le64(ref_generation); |
a5eb62e3 | 464 | low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum)); |
21a4989d CM |
465 | if (owner >= BTRFS_FIRST_FREE_OBJECTID) { |
466 | lenum = cpu_to_le64(owner); | |
a5eb62e3 | 467 | low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum)); |
21a4989d | 468 | lenum = cpu_to_le64(owner_offset); |
a5eb62e3 | 469 | low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum)); |
21a4989d | 470 | } |
74493f7a CM |
471 | return ((u64)high_crc << 32) | (u64)low_crc; |
472 | } | |
473 | ||
7bb86316 CM |
474 | static int match_extent_ref(struct extent_buffer *leaf, |
475 | struct btrfs_extent_ref *disk_ref, | |
476 | struct btrfs_extent_ref *cpu_ref) | |
477 | { | |
478 | int ret; | |
479 | int len; | |
480 | ||
481 | if (cpu_ref->objectid) | |
482 | len = sizeof(*cpu_ref); | |
483 | else | |
484 | len = 2 * sizeof(u64); | |
485 | ret = memcmp_extent_buffer(leaf, cpu_ref, (unsigned long)disk_ref, | |
486 | len); | |
487 | return ret == 0; | |
488 | } | |
489 | ||
98ed5174 CM |
490 | static int noinline lookup_extent_backref(struct btrfs_trans_handle *trans, |
491 | struct btrfs_root *root, | |
492 | struct btrfs_path *path, u64 bytenr, | |
493 | u64 root_objectid, | |
494 | u64 ref_generation, u64 owner, | |
495 | u64 owner_offset, int del) | |
74493f7a CM |
496 | { |
497 | u64 hash; | |
498 | struct btrfs_key key; | |
7bb86316 | 499 | struct btrfs_key found_key; |
74493f7a | 500 | struct btrfs_extent_ref ref; |
7bb86316 CM |
501 | struct extent_buffer *leaf; |
502 | struct btrfs_extent_ref *disk_ref; | |
503 | int ret; | |
504 | int ret2; | |
505 | ||
506 | btrfs_set_stack_ref_root(&ref, root_objectid); | |
507 | btrfs_set_stack_ref_generation(&ref, ref_generation); | |
508 | btrfs_set_stack_ref_objectid(&ref, owner); | |
509 | btrfs_set_stack_ref_offset(&ref, owner_offset); | |
510 | ||
511 | hash = hash_extent_ref(root_objectid, ref_generation, owner, | |
512 | owner_offset); | |
513 | key.offset = hash; | |
514 | key.objectid = bytenr; | |
515 | key.type = BTRFS_EXTENT_REF_KEY; | |
516 | ||
517 | while (1) { | |
518 | ret = btrfs_search_slot(trans, root, &key, path, | |
519 | del ? -1 : 0, del); | |
520 | if (ret < 0) | |
521 | goto out; | |
522 | leaf = path->nodes[0]; | |
523 | if (ret != 0) { | |
524 | u32 nritems = btrfs_header_nritems(leaf); | |
525 | if (path->slots[0] >= nritems) { | |
526 | ret2 = btrfs_next_leaf(root, path); | |
527 | if (ret2) | |
528 | goto out; | |
529 | leaf = path->nodes[0]; | |
530 | } | |
531 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
532 | if (found_key.objectid != bytenr || | |
533 | found_key.type != BTRFS_EXTENT_REF_KEY) | |
534 | goto out; | |
535 | key.offset = found_key.offset; | |
536 | if (del) { | |
537 | btrfs_release_path(root, path); | |
538 | continue; | |
539 | } | |
540 | } | |
541 | disk_ref = btrfs_item_ptr(path->nodes[0], | |
542 | path->slots[0], | |
543 | struct btrfs_extent_ref); | |
544 | if (match_extent_ref(path->nodes[0], disk_ref, &ref)) { | |
545 | ret = 0; | |
546 | goto out; | |
547 | } | |
548 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
549 | key.offset = found_key.offset + 1; | |
550 | btrfs_release_path(root, path); | |
551 | } | |
552 | out: | |
553 | return ret; | |
554 | } | |
555 | ||
d8d5f3e1 CM |
556 | /* |
557 | * Back reference rules. Back refs have three main goals: | |
558 | * | |
559 | * 1) differentiate between all holders of references to an extent so that | |
560 | * when a reference is dropped we can make sure it was a valid reference | |
561 | * before freeing the extent. | |
562 | * | |
563 | * 2) Provide enough information to quickly find the holders of an extent | |
564 | * if we notice a given block is corrupted or bad. | |
565 | * | |
566 | * 3) Make it easy to migrate blocks for FS shrinking or storage pool | |
567 | * maintenance. This is actually the same as #2, but with a slightly | |
568 | * different use case. | |
569 | * | |
570 | * File extents can be referenced by: | |
571 | * | |
572 | * - multiple snapshots, subvolumes, or different generations in one subvol | |
573 | * - different files inside a single subvolume (in theory, not implemented yet) | |
574 | * - different offsets inside a file (bookend extents in file.c) | |
575 | * | |
576 | * The extent ref structure has fields for: | |
577 | * | |
578 | * - Objectid of the subvolume root | |
579 | * - Generation number of the tree holding the reference | |
580 | * - objectid of the file holding the reference | |
581 | * - offset in the file corresponding to the key holding the reference | |
582 | * | |
583 | * When a file extent is allocated the fields are filled in: | |
584 | * (root_key.objectid, trans->transid, inode objectid, offset in file) | |
585 | * | |
586 | * When a leaf is cow'd new references are added for every file extent found | |
587 | * in the leaf. It looks the same as the create case, but trans->transid | |
588 | * will be different when the block is cow'd. | |
589 | * | |
590 | * (root_key.objectid, trans->transid, inode objectid, offset in file) | |
591 | * | |
592 | * When a file extent is removed either during snapshot deletion or file | |
593 | * truncation, the corresponding back reference is found | |
594 | * by searching for: | |
595 | * | |
596 | * (btrfs_header_owner(leaf), btrfs_header_generation(leaf), | |
597 | * inode objectid, offset in file) | |
598 | * | |
599 | * Btree extents can be referenced by: | |
600 | * | |
601 | * - Different subvolumes | |
602 | * - Different generations of the same subvolume | |
603 | * | |
604 | * Storing sufficient information for a full reverse mapping of a btree | |
605 | * block would require storing the lowest key of the block in the backref, | |
606 | * and it would require updating that lowest key either before write out or | |
607 | * every time it changed. Instead, the objectid of the lowest key is stored | |
608 | * along with the level of the tree block. This provides a hint | |
609 | * about where in the btree the block can be found. Searches through the | |
610 | * btree only need to look for a pointer to that block, so they stop one | |
611 | * level higher than the level recorded in the backref. | |
612 | * | |
613 | * Some btrees do not do reference counting on their extents. These | |
614 | * include the extent tree and the tree of tree roots. Backrefs for these | |
615 | * trees always have a generation of zero. | |
616 | * | |
617 | * When a tree block is created, back references are inserted: | |
618 | * | |
f6dbff55 | 619 | * (root->root_key.objectid, trans->transid or zero, level, lowest_key_objectid) |
d8d5f3e1 CM |
620 | * |
621 | * When a tree block is cow'd in a reference counted root, | |
622 | * new back references are added for all the blocks it points to. | |
623 | * These are of the form (trans->transid will have increased since creation): | |
624 | * | |
f6dbff55 | 625 | * (root->root_key.objectid, trans->transid, level, lowest_key_objectid) |
d8d5f3e1 CM |
626 | * |
627 | * Because the lowest_key_objectid and the level are just hints | |
628 | * they are not used when backrefs are deleted. When a backref is deleted: | |
629 | * | |
630 | * if backref was for a tree root: | |
631 | * root_objectid = root->root_key.objectid | |
632 | * else | |
633 | * root_objectid = btrfs_header_owner(parent) | |
634 | * | |
635 | * (root_objectid, btrfs_header_generation(parent) or zero, 0, 0) | |
636 | * | |
637 | * Back Reference Key hashing: | |
638 | * | |
639 | * Back references have four fields, each 64 bits long. Unfortunately, | |
640 | * This is hashed into a single 64 bit number and placed into the key offset. | |
641 | * The key objectid corresponds to the first byte in the extent, and the | |
642 | * key type is set to BTRFS_EXTENT_REF_KEY | |
643 | */ | |
7bb86316 CM |
644 | int btrfs_insert_extent_backref(struct btrfs_trans_handle *trans, |
645 | struct btrfs_root *root, | |
646 | struct btrfs_path *path, u64 bytenr, | |
647 | u64 root_objectid, u64 ref_generation, | |
648 | u64 owner, u64 owner_offset) | |
649 | { | |
650 | u64 hash; | |
651 | struct btrfs_key key; | |
652 | struct btrfs_extent_ref ref; | |
653 | struct btrfs_extent_ref *disk_ref; | |
74493f7a CM |
654 | int ret; |
655 | ||
656 | btrfs_set_stack_ref_root(&ref, root_objectid); | |
7bb86316 | 657 | btrfs_set_stack_ref_generation(&ref, ref_generation); |
74493f7a CM |
658 | btrfs_set_stack_ref_objectid(&ref, owner); |
659 | btrfs_set_stack_ref_offset(&ref, owner_offset); | |
660 | ||
7bb86316 CM |
661 | hash = hash_extent_ref(root_objectid, ref_generation, owner, |
662 | owner_offset); | |
74493f7a CM |
663 | key.offset = hash; |
664 | key.objectid = bytenr; | |
665 | key.type = BTRFS_EXTENT_REF_KEY; | |
666 | ||
667 | ret = btrfs_insert_empty_item(trans, root, path, &key, sizeof(ref)); | |
668 | while (ret == -EEXIST) { | |
7bb86316 CM |
669 | disk_ref = btrfs_item_ptr(path->nodes[0], path->slots[0], |
670 | struct btrfs_extent_ref); | |
671 | if (match_extent_ref(path->nodes[0], disk_ref, &ref)) | |
672 | goto out; | |
673 | key.offset++; | |
674 | btrfs_release_path(root, path); | |
675 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
676 | sizeof(ref)); | |
74493f7a | 677 | } |
7bb86316 CM |
678 | if (ret) |
679 | goto out; | |
680 | disk_ref = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
681 | struct btrfs_extent_ref); | |
682 | write_extent_buffer(path->nodes[0], &ref, (unsigned long)disk_ref, | |
683 | sizeof(ref)); | |
684 | btrfs_mark_buffer_dirty(path->nodes[0]); | |
685 | out: | |
686 | btrfs_release_path(root, path); | |
687 | return ret; | |
74493f7a CM |
688 | } |
689 | ||
925baedd | 690 | static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans, |
b18c6685 | 691 | struct btrfs_root *root, |
74493f7a | 692 | u64 bytenr, u64 num_bytes, |
7bb86316 | 693 | u64 root_objectid, u64 ref_generation, |
74493f7a | 694 | u64 owner, u64 owner_offset) |
02217ed2 | 695 | { |
5caf2a00 | 696 | struct btrfs_path *path; |
02217ed2 | 697 | int ret; |
e2fa7227 | 698 | struct btrfs_key key; |
5f39d397 | 699 | struct extent_buffer *l; |
234b63a0 | 700 | struct btrfs_extent_item *item; |
cf27e1ee | 701 | u32 refs; |
037e6390 | 702 | |
db94535d | 703 | WARN_ON(num_bytes < root->sectorsize); |
5caf2a00 | 704 | path = btrfs_alloc_path(); |
54aa1f4d CM |
705 | if (!path) |
706 | return -ENOMEM; | |
26b8003f | 707 | |
3c12ac72 | 708 | path->reada = 1; |
db94535d | 709 | key.objectid = bytenr; |
62e2749e | 710 | btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY); |
db94535d | 711 | key.offset = num_bytes; |
5caf2a00 | 712 | ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path, |
9f5fae2f | 713 | 0, 1); |
54aa1f4d CM |
714 | if (ret < 0) |
715 | return ret; | |
a429e513 | 716 | if (ret != 0) { |
a28ec197 | 717 | BUG(); |
a429e513 | 718 | } |
02217ed2 | 719 | BUG_ON(ret != 0); |
5f39d397 | 720 | l = path->nodes[0]; |
5caf2a00 | 721 | item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item); |
5f39d397 CM |
722 | refs = btrfs_extent_refs(l, item); |
723 | btrfs_set_extent_refs(l, item, refs + 1); | |
5caf2a00 | 724 | btrfs_mark_buffer_dirty(path->nodes[0]); |
a28ec197 | 725 | |
5caf2a00 | 726 | btrfs_release_path(root->fs_info->extent_root, path); |
7bb86316 | 727 | |
3c12ac72 | 728 | path->reada = 1; |
7bb86316 CM |
729 | ret = btrfs_insert_extent_backref(trans, root->fs_info->extent_root, |
730 | path, bytenr, root_objectid, | |
731 | ref_generation, owner, owner_offset); | |
732 | BUG_ON(ret); | |
9f5fae2f | 733 | finish_current_insert(trans, root->fs_info->extent_root); |
e20d96d6 | 734 | del_pending_extents(trans, root->fs_info->extent_root); |
74493f7a CM |
735 | |
736 | btrfs_free_path(path); | |
02217ed2 CM |
737 | return 0; |
738 | } | |
739 | ||
925baedd CM |
740 | int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans, |
741 | struct btrfs_root *root, | |
742 | u64 bytenr, u64 num_bytes, | |
743 | u64 root_objectid, u64 ref_generation, | |
744 | u64 owner, u64 owner_offset) | |
745 | { | |
746 | int ret; | |
747 | ||
748 | mutex_lock(&root->fs_info->alloc_mutex); | |
749 | ret = __btrfs_inc_extent_ref(trans, root, bytenr, num_bytes, | |
750 | root_objectid, ref_generation, | |
751 | owner, owner_offset); | |
752 | mutex_unlock(&root->fs_info->alloc_mutex); | |
753 | return ret; | |
754 | } | |
755 | ||
e9d0b13b CM |
756 | int btrfs_extent_post_op(struct btrfs_trans_handle *trans, |
757 | struct btrfs_root *root) | |
758 | { | |
759 | finish_current_insert(trans, root->fs_info->extent_root); | |
760 | del_pending_extents(trans, root->fs_info->extent_root); | |
761 | return 0; | |
762 | } | |
763 | ||
b18c6685 | 764 | static int lookup_extent_ref(struct btrfs_trans_handle *trans, |
db94535d CM |
765 | struct btrfs_root *root, u64 bytenr, |
766 | u64 num_bytes, u32 *refs) | |
a28ec197 | 767 | { |
5caf2a00 | 768 | struct btrfs_path *path; |
a28ec197 | 769 | int ret; |
e2fa7227 | 770 | struct btrfs_key key; |
5f39d397 | 771 | struct extent_buffer *l; |
234b63a0 | 772 | struct btrfs_extent_item *item; |
5caf2a00 | 773 | |
db94535d | 774 | WARN_ON(num_bytes < root->sectorsize); |
5caf2a00 | 775 | path = btrfs_alloc_path(); |
3c12ac72 | 776 | path->reada = 1; |
db94535d CM |
777 | key.objectid = bytenr; |
778 | key.offset = num_bytes; | |
62e2749e | 779 | btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY); |
5caf2a00 | 780 | ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path, |
9f5fae2f | 781 | 0, 0); |
54aa1f4d CM |
782 | if (ret < 0) |
783 | goto out; | |
5f39d397 CM |
784 | if (ret != 0) { |
785 | btrfs_print_leaf(root, path->nodes[0]); | |
db94535d | 786 | printk("failed to find block number %Lu\n", bytenr); |
a28ec197 | 787 | BUG(); |
5f39d397 CM |
788 | } |
789 | l = path->nodes[0]; | |
5caf2a00 | 790 | item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item); |
5f39d397 | 791 | *refs = btrfs_extent_refs(l, item); |
54aa1f4d | 792 | out: |
5caf2a00 | 793 | btrfs_free_path(path); |
a28ec197 CM |
794 | return 0; |
795 | } | |
796 | ||
be20aa9d CM |
797 | u32 btrfs_count_snapshots_in_path(struct btrfs_root *root, |
798 | struct btrfs_path *count_path, | |
a68d5933 | 799 | u64 expected_owner, |
be20aa9d CM |
800 | u64 first_extent) |
801 | { | |
802 | struct btrfs_root *extent_root = root->fs_info->extent_root; | |
803 | struct btrfs_path *path; | |
804 | u64 bytenr; | |
805 | u64 found_objectid; | |
a68d5933 | 806 | u64 found_owner; |
56b453c9 | 807 | u64 root_objectid = root->root_key.objectid; |
be20aa9d | 808 | u32 total_count = 0; |
bbaf549e | 809 | u32 extent_refs; |
be20aa9d | 810 | u32 cur_count; |
be20aa9d CM |
811 | u32 nritems; |
812 | int ret; | |
813 | struct btrfs_key key; | |
814 | struct btrfs_key found_key; | |
815 | struct extent_buffer *l; | |
816 | struct btrfs_extent_item *item; | |
817 | struct btrfs_extent_ref *ref_item; | |
818 | int level = -1; | |
819 | ||
925baedd CM |
820 | /* FIXME, needs locking */ |
821 | BUG(); | |
822 | ||
823 | mutex_lock(&root->fs_info->alloc_mutex); | |
be20aa9d CM |
824 | path = btrfs_alloc_path(); |
825 | again: | |
826 | if (level == -1) | |
827 | bytenr = first_extent; | |
828 | else | |
829 | bytenr = count_path->nodes[level]->start; | |
830 | ||
831 | cur_count = 0; | |
832 | key.objectid = bytenr; | |
833 | key.offset = 0; | |
834 | ||
835 | btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY); | |
836 | ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0); | |
837 | if (ret < 0) | |
838 | goto out; | |
839 | BUG_ON(ret == 0); | |
840 | ||
841 | l = path->nodes[0]; | |
842 | btrfs_item_key_to_cpu(l, &found_key, path->slots[0]); | |
843 | ||
844 | if (found_key.objectid != bytenr || | |
845 | found_key.type != BTRFS_EXTENT_ITEM_KEY) { | |
846 | goto out; | |
847 | } | |
848 | ||
849 | item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item); | |
bbaf549e | 850 | extent_refs = btrfs_extent_refs(l, item); |
be20aa9d | 851 | while (1) { |
bd09835d | 852 | l = path->nodes[0]; |
be20aa9d CM |
853 | nritems = btrfs_header_nritems(l); |
854 | if (path->slots[0] >= nritems) { | |
855 | ret = btrfs_next_leaf(extent_root, path); | |
856 | if (ret == 0) | |
857 | continue; | |
858 | break; | |
859 | } | |
860 | btrfs_item_key_to_cpu(l, &found_key, path->slots[0]); | |
861 | if (found_key.objectid != bytenr) | |
862 | break; | |
bd09835d | 863 | |
be20aa9d CM |
864 | if (found_key.type != BTRFS_EXTENT_REF_KEY) { |
865 | path->slots[0]++; | |
866 | continue; | |
867 | } | |
868 | ||
869 | cur_count++; | |
870 | ref_item = btrfs_item_ptr(l, path->slots[0], | |
871 | struct btrfs_extent_ref); | |
872 | found_objectid = btrfs_ref_root(l, ref_item); | |
873 | ||
56b453c9 CM |
874 | if (found_objectid != root_objectid) { |
875 | total_count = 2; | |
4313b399 | 876 | goto out; |
56b453c9 | 877 | } |
a68d5933 CM |
878 | if (level == -1) { |
879 | found_owner = btrfs_ref_objectid(l, ref_item); | |
880 | if (found_owner != expected_owner) { | |
881 | total_count = 2; | |
882 | goto out; | |
883 | } | |
bbaf549e CM |
884 | /* |
885 | * nasty. we don't count a reference held by | |
886 | * the running transaction. This allows nodatacow | |
887 | * to avoid cow most of the time | |
888 | */ | |
889 | if (found_owner >= BTRFS_FIRST_FREE_OBJECTID && | |
890 | btrfs_ref_generation(l, ref_item) == | |
891 | root->fs_info->generation) { | |
892 | extent_refs--; | |
893 | } | |
a68d5933 | 894 | } |
56b453c9 | 895 | total_count = 1; |
be20aa9d CM |
896 | path->slots[0]++; |
897 | } | |
bbaf549e CM |
898 | /* |
899 | * if there is more than one reference against a data extent, | |
900 | * we have to assume the other ref is another snapshot | |
901 | */ | |
902 | if (level == -1 && extent_refs > 1) { | |
903 | total_count = 2; | |
904 | goto out; | |
905 | } | |
be20aa9d CM |
906 | if (cur_count == 0) { |
907 | total_count = 0; | |
908 | goto out; | |
909 | } | |
be20aa9d CM |
910 | if (level >= 0 && root->node == count_path->nodes[level]) |
911 | goto out; | |
912 | level++; | |
913 | btrfs_release_path(root, path); | |
914 | goto again; | |
915 | ||
916 | out: | |
917 | btrfs_free_path(path); | |
925baedd | 918 | mutex_unlock(&root->fs_info->alloc_mutex); |
be20aa9d | 919 | return total_count; |
be20aa9d | 920 | } |
c5739bba | 921 | |
e089f05c | 922 | int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root, |
5f39d397 | 923 | struct extent_buffer *buf) |
02217ed2 | 924 | { |
db94535d | 925 | u64 bytenr; |
5f39d397 CM |
926 | u32 nritems; |
927 | struct btrfs_key key; | |
6407bf6d | 928 | struct btrfs_file_extent_item *fi; |
02217ed2 | 929 | int i; |
db94535d | 930 | int level; |
6407bf6d | 931 | int ret; |
54aa1f4d | 932 | int faili; |
a28ec197 | 933 | |
3768f368 | 934 | if (!root->ref_cows) |
a28ec197 | 935 | return 0; |
5f39d397 | 936 | |
db94535d | 937 | level = btrfs_header_level(buf); |
5f39d397 CM |
938 | nritems = btrfs_header_nritems(buf); |
939 | for (i = 0; i < nritems; i++) { | |
e34a5b4f | 940 | cond_resched(); |
db94535d CM |
941 | if (level == 0) { |
942 | u64 disk_bytenr; | |
5f39d397 CM |
943 | btrfs_item_key_to_cpu(buf, &key, i); |
944 | if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY) | |
6407bf6d | 945 | continue; |
5f39d397 | 946 | fi = btrfs_item_ptr(buf, i, |
6407bf6d | 947 | struct btrfs_file_extent_item); |
5f39d397 | 948 | if (btrfs_file_extent_type(buf, fi) == |
236454df CM |
949 | BTRFS_FILE_EXTENT_INLINE) |
950 | continue; | |
db94535d CM |
951 | disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi); |
952 | if (disk_bytenr == 0) | |
3a686375 | 953 | continue; |
4a096752 CM |
954 | |
955 | mutex_lock(&root->fs_info->alloc_mutex); | |
925baedd | 956 | ret = __btrfs_inc_extent_ref(trans, root, disk_bytenr, |
7bb86316 CM |
957 | btrfs_file_extent_disk_num_bytes(buf, fi), |
958 | root->root_key.objectid, trans->transid, | |
959 | key.objectid, key.offset); | |
4a096752 | 960 | mutex_unlock(&root->fs_info->alloc_mutex); |
54aa1f4d CM |
961 | if (ret) { |
962 | faili = i; | |
4a096752 | 963 | WARN_ON(1); |
54aa1f4d CM |
964 | goto fail; |
965 | } | |
6407bf6d | 966 | } else { |
db94535d | 967 | bytenr = btrfs_node_blockptr(buf, i); |
6caab489 | 968 | btrfs_node_key_to_cpu(buf, &key, i); |
4a096752 CM |
969 | |
970 | mutex_lock(&root->fs_info->alloc_mutex); | |
925baedd | 971 | ret = __btrfs_inc_extent_ref(trans, root, bytenr, |
7bb86316 CM |
972 | btrfs_level_size(root, level - 1), |
973 | root->root_key.objectid, | |
f6dbff55 CM |
974 | trans->transid, |
975 | level - 1, key.objectid); | |
4a096752 | 976 | mutex_unlock(&root->fs_info->alloc_mutex); |
54aa1f4d CM |
977 | if (ret) { |
978 | faili = i; | |
4a096752 | 979 | WARN_ON(1); |
54aa1f4d CM |
980 | goto fail; |
981 | } | |
6407bf6d | 982 | } |
02217ed2 CM |
983 | } |
984 | return 0; | |
54aa1f4d | 985 | fail: |
ccd467d6 | 986 | WARN_ON(1); |
7bb86316 | 987 | #if 0 |
54aa1f4d | 988 | for (i =0; i < faili; i++) { |
db94535d CM |
989 | if (level == 0) { |
990 | u64 disk_bytenr; | |
5f39d397 CM |
991 | btrfs_item_key_to_cpu(buf, &key, i); |
992 | if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY) | |
54aa1f4d | 993 | continue; |
5f39d397 | 994 | fi = btrfs_item_ptr(buf, i, |
54aa1f4d | 995 | struct btrfs_file_extent_item); |
5f39d397 | 996 | if (btrfs_file_extent_type(buf, fi) == |
54aa1f4d CM |
997 | BTRFS_FILE_EXTENT_INLINE) |
998 | continue; | |
db94535d CM |
999 | disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi); |
1000 | if (disk_bytenr == 0) | |
54aa1f4d | 1001 | continue; |
db94535d CM |
1002 | err = btrfs_free_extent(trans, root, disk_bytenr, |
1003 | btrfs_file_extent_disk_num_bytes(buf, | |
5f39d397 | 1004 | fi), 0); |
54aa1f4d CM |
1005 | BUG_ON(err); |
1006 | } else { | |
db94535d CM |
1007 | bytenr = btrfs_node_blockptr(buf, i); |
1008 | err = btrfs_free_extent(trans, root, bytenr, | |
1009 | btrfs_level_size(root, level - 1), 0); | |
54aa1f4d CM |
1010 | BUG_ON(err); |
1011 | } | |
1012 | } | |
7bb86316 | 1013 | #endif |
54aa1f4d | 1014 | return ret; |
02217ed2 CM |
1015 | } |
1016 | ||
9078a3e1 CM |
1017 | static int write_one_cache_group(struct btrfs_trans_handle *trans, |
1018 | struct btrfs_root *root, | |
1019 | struct btrfs_path *path, | |
1020 | struct btrfs_block_group_cache *cache) | |
1021 | { | |
1022 | int ret; | |
1023 | int pending_ret; | |
1024 | struct btrfs_root *extent_root = root->fs_info->extent_root; | |
5f39d397 CM |
1025 | unsigned long bi; |
1026 | struct extent_buffer *leaf; | |
9078a3e1 | 1027 | |
9078a3e1 | 1028 | ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1); |
54aa1f4d CM |
1029 | if (ret < 0) |
1030 | goto fail; | |
9078a3e1 | 1031 | BUG_ON(ret); |
5f39d397 CM |
1032 | |
1033 | leaf = path->nodes[0]; | |
1034 | bi = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
1035 | write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item)); | |
1036 | btrfs_mark_buffer_dirty(leaf); | |
9078a3e1 | 1037 | btrfs_release_path(extent_root, path); |
54aa1f4d | 1038 | fail: |
9078a3e1 CM |
1039 | finish_current_insert(trans, extent_root); |
1040 | pending_ret = del_pending_extents(trans, extent_root); | |
1041 | if (ret) | |
1042 | return ret; | |
1043 | if (pending_ret) | |
1044 | return pending_ret; | |
1045 | return 0; | |
1046 | ||
1047 | } | |
1048 | ||
96b5179d CM |
1049 | int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans, |
1050 | struct btrfs_root *root) | |
9078a3e1 | 1051 | { |
d1310b2e | 1052 | struct extent_io_tree *block_group_cache; |
96b5179d | 1053 | struct btrfs_block_group_cache *cache; |
9078a3e1 CM |
1054 | int ret; |
1055 | int err = 0; | |
1056 | int werr = 0; | |
9078a3e1 | 1057 | struct btrfs_path *path; |
96b5179d CM |
1058 | u64 last = 0; |
1059 | u64 start; | |
1060 | u64 end; | |
1061 | u64 ptr; | |
9078a3e1 | 1062 | |
96b5179d | 1063 | block_group_cache = &root->fs_info->block_group_cache; |
9078a3e1 CM |
1064 | path = btrfs_alloc_path(); |
1065 | if (!path) | |
1066 | return -ENOMEM; | |
1067 | ||
925baedd | 1068 | mutex_lock(&root->fs_info->alloc_mutex); |
9078a3e1 | 1069 | while(1) { |
96b5179d CM |
1070 | ret = find_first_extent_bit(block_group_cache, last, |
1071 | &start, &end, BLOCK_GROUP_DIRTY); | |
1072 | if (ret) | |
9078a3e1 | 1073 | break; |
54aa1f4d | 1074 | |
96b5179d CM |
1075 | last = end + 1; |
1076 | ret = get_state_private(block_group_cache, start, &ptr); | |
1077 | if (ret) | |
1078 | break; | |
ae2f5411 | 1079 | cache = (struct btrfs_block_group_cache *)(unsigned long)ptr; |
96b5179d CM |
1080 | err = write_one_cache_group(trans, root, |
1081 | path, cache); | |
1082 | /* | |
1083 | * if we fail to write the cache group, we want | |
1084 | * to keep it marked dirty in hopes that a later | |
1085 | * write will work | |
1086 | */ | |
1087 | if (err) { | |
1088 | werr = err; | |
1089 | continue; | |
9078a3e1 | 1090 | } |
96b5179d CM |
1091 | clear_extent_bits(block_group_cache, start, end, |
1092 | BLOCK_GROUP_DIRTY, GFP_NOFS); | |
9078a3e1 CM |
1093 | } |
1094 | btrfs_free_path(path); | |
925baedd | 1095 | mutex_unlock(&root->fs_info->alloc_mutex); |
9078a3e1 CM |
1096 | return werr; |
1097 | } | |
1098 | ||
6324fbf3 CM |
1099 | static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info, |
1100 | u64 flags) | |
1101 | { | |
1102 | struct list_head *head = &info->space_info; | |
1103 | struct list_head *cur; | |
1104 | struct btrfs_space_info *found; | |
1105 | list_for_each(cur, head) { | |
1106 | found = list_entry(cur, struct btrfs_space_info, list); | |
1107 | if (found->flags == flags) | |
1108 | return found; | |
1109 | } | |
1110 | return NULL; | |
1111 | ||
1112 | } | |
1113 | ||
593060d7 CM |
1114 | static int update_space_info(struct btrfs_fs_info *info, u64 flags, |
1115 | u64 total_bytes, u64 bytes_used, | |
1116 | struct btrfs_space_info **space_info) | |
1117 | { | |
1118 | struct btrfs_space_info *found; | |
1119 | ||
1120 | found = __find_space_info(info, flags); | |
1121 | if (found) { | |
1122 | found->total_bytes += total_bytes; | |
1123 | found->bytes_used += bytes_used; | |
8f18cf13 | 1124 | found->full = 0; |
593060d7 CM |
1125 | WARN_ON(found->total_bytes < found->bytes_used); |
1126 | *space_info = found; | |
1127 | return 0; | |
1128 | } | |
1129 | found = kmalloc(sizeof(*found), GFP_NOFS); | |
1130 | if (!found) | |
1131 | return -ENOMEM; | |
1132 | ||
1133 | list_add(&found->list, &info->space_info); | |
1134 | found->flags = flags; | |
1135 | found->total_bytes = total_bytes; | |
1136 | found->bytes_used = bytes_used; | |
1137 | found->bytes_pinned = 0; | |
1138 | found->full = 0; | |
0ef3e66b | 1139 | found->force_alloc = 0; |
593060d7 CM |
1140 | *space_info = found; |
1141 | return 0; | |
1142 | } | |
1143 | ||
8790d502 CM |
1144 | static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags) |
1145 | { | |
1146 | u64 extra_flags = flags & (BTRFS_BLOCK_GROUP_RAID0 | | |
611f0e00 | 1147 | BTRFS_BLOCK_GROUP_RAID1 | |
321aecc6 | 1148 | BTRFS_BLOCK_GROUP_RAID10 | |
611f0e00 | 1149 | BTRFS_BLOCK_GROUP_DUP); |
8790d502 CM |
1150 | if (extra_flags) { |
1151 | if (flags & BTRFS_BLOCK_GROUP_DATA) | |
1152 | fs_info->avail_data_alloc_bits |= extra_flags; | |
1153 | if (flags & BTRFS_BLOCK_GROUP_METADATA) | |
1154 | fs_info->avail_metadata_alloc_bits |= extra_flags; | |
1155 | if (flags & BTRFS_BLOCK_GROUP_SYSTEM) | |
1156 | fs_info->avail_system_alloc_bits |= extra_flags; | |
1157 | } | |
1158 | } | |
593060d7 | 1159 | |
a061fc8d | 1160 | static u64 reduce_alloc_profile(struct btrfs_root *root, u64 flags) |
ec44a35c | 1161 | { |
a061fc8d CM |
1162 | u64 num_devices = root->fs_info->fs_devices->num_devices; |
1163 | ||
1164 | if (num_devices == 1) | |
1165 | flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0); | |
1166 | if (num_devices < 4) | |
1167 | flags &= ~BTRFS_BLOCK_GROUP_RAID10; | |
1168 | ||
ec44a35c CM |
1169 | if ((flags & BTRFS_BLOCK_GROUP_DUP) && |
1170 | (flags & (BTRFS_BLOCK_GROUP_RAID1 | | |
a061fc8d | 1171 | BTRFS_BLOCK_GROUP_RAID10))) { |
ec44a35c | 1172 | flags &= ~BTRFS_BLOCK_GROUP_DUP; |
a061fc8d | 1173 | } |
ec44a35c CM |
1174 | |
1175 | if ((flags & BTRFS_BLOCK_GROUP_RAID1) && | |
a061fc8d | 1176 | (flags & BTRFS_BLOCK_GROUP_RAID10)) { |
ec44a35c | 1177 | flags &= ~BTRFS_BLOCK_GROUP_RAID1; |
a061fc8d | 1178 | } |
ec44a35c CM |
1179 | |
1180 | if ((flags & BTRFS_BLOCK_GROUP_RAID0) && | |
1181 | ((flags & BTRFS_BLOCK_GROUP_RAID1) | | |
1182 | (flags & BTRFS_BLOCK_GROUP_RAID10) | | |
1183 | (flags & BTRFS_BLOCK_GROUP_DUP))) | |
1184 | flags &= ~BTRFS_BLOCK_GROUP_RAID0; | |
1185 | return flags; | |
1186 | } | |
1187 | ||
6324fbf3 CM |
1188 | static int do_chunk_alloc(struct btrfs_trans_handle *trans, |
1189 | struct btrfs_root *extent_root, u64 alloc_bytes, | |
0ef3e66b | 1190 | u64 flags, int force) |
6324fbf3 CM |
1191 | { |
1192 | struct btrfs_space_info *space_info; | |
1193 | u64 thresh; | |
1194 | u64 start; | |
1195 | u64 num_bytes; | |
1196 | int ret; | |
1197 | ||
a061fc8d | 1198 | flags = reduce_alloc_profile(extent_root, flags); |
ec44a35c | 1199 | |
6324fbf3 | 1200 | space_info = __find_space_info(extent_root->fs_info, flags); |
593060d7 CM |
1201 | if (!space_info) { |
1202 | ret = update_space_info(extent_root->fs_info, flags, | |
1203 | 0, 0, &space_info); | |
1204 | BUG_ON(ret); | |
1205 | } | |
6324fbf3 CM |
1206 | BUG_ON(!space_info); |
1207 | ||
0ef3e66b CM |
1208 | if (space_info->force_alloc) { |
1209 | force = 1; | |
1210 | space_info->force_alloc = 0; | |
1211 | } | |
6324fbf3 | 1212 | if (space_info->full) |
925baedd | 1213 | goto out; |
6324fbf3 | 1214 | |
8790d502 | 1215 | thresh = div_factor(space_info->total_bytes, 6); |
0ef3e66b CM |
1216 | if (!force && |
1217 | (space_info->bytes_used + space_info->bytes_pinned + alloc_bytes) < | |
6324fbf3 | 1218 | thresh) |
925baedd | 1219 | goto out; |
6324fbf3 | 1220 | |
925baedd | 1221 | mutex_lock(&extent_root->fs_info->chunk_mutex); |
6324fbf3 CM |
1222 | ret = btrfs_alloc_chunk(trans, extent_root, &start, &num_bytes, flags); |
1223 | if (ret == -ENOSPC) { | |
1224 | printk("space info full %Lu\n", flags); | |
1225 | space_info->full = 1; | |
a74a4b97 | 1226 | goto out_unlock; |
6324fbf3 | 1227 | } |
6324fbf3 CM |
1228 | BUG_ON(ret); |
1229 | ||
1230 | ret = btrfs_make_block_group(trans, extent_root, 0, flags, | |
e17cade2 | 1231 | BTRFS_FIRST_CHUNK_TREE_OBJECTID, start, num_bytes); |
6324fbf3 | 1232 | BUG_ON(ret); |
a74a4b97 | 1233 | out_unlock: |
333db94c | 1234 | mutex_unlock(&extent_root->fs_info->chunk_mutex); |
a74a4b97 | 1235 | out: |
6324fbf3 CM |
1236 | return 0; |
1237 | } | |
1238 | ||
9078a3e1 CM |
1239 | static int update_block_group(struct btrfs_trans_handle *trans, |
1240 | struct btrfs_root *root, | |
db94535d | 1241 | u64 bytenr, u64 num_bytes, int alloc, |
0b86a832 | 1242 | int mark_free) |
9078a3e1 CM |
1243 | { |
1244 | struct btrfs_block_group_cache *cache; | |
1245 | struct btrfs_fs_info *info = root->fs_info; | |
db94535d | 1246 | u64 total = num_bytes; |
9078a3e1 | 1247 | u64 old_val; |
db94535d | 1248 | u64 byte_in_group; |
96b5179d CM |
1249 | u64 start; |
1250 | u64 end; | |
3e1ad54f | 1251 | |
7d9eb12c | 1252 | WARN_ON(!mutex_is_locked(&root->fs_info->alloc_mutex)); |
9078a3e1 | 1253 | while(total) { |
db94535d | 1254 | cache = btrfs_lookup_block_group(info, bytenr); |
3e1ad54f | 1255 | if (!cache) { |
9078a3e1 | 1256 | return -1; |
cd1bc465 | 1257 | } |
db94535d CM |
1258 | byte_in_group = bytenr - cache->key.objectid; |
1259 | WARN_ON(byte_in_group > cache->key.offset); | |
96b5179d CM |
1260 | start = cache->key.objectid; |
1261 | end = start + cache->key.offset - 1; | |
1262 | set_extent_bits(&info->block_group_cache, start, end, | |
1263 | BLOCK_GROUP_DIRTY, GFP_NOFS); | |
9078a3e1 CM |
1264 | |
1265 | old_val = btrfs_block_group_used(&cache->item); | |
db94535d | 1266 | num_bytes = min(total, cache->key.offset - byte_in_group); |
cd1bc465 | 1267 | if (alloc) { |
db94535d | 1268 | old_val += num_bytes; |
6324fbf3 | 1269 | cache->space_info->bytes_used += num_bytes; |
cd1bc465 | 1270 | } else { |
db94535d | 1271 | old_val -= num_bytes; |
6324fbf3 | 1272 | cache->space_info->bytes_used -= num_bytes; |
f510cfec CM |
1273 | if (mark_free) { |
1274 | set_extent_dirty(&info->free_space_cache, | |
db94535d | 1275 | bytenr, bytenr + num_bytes - 1, |
f510cfec | 1276 | GFP_NOFS); |
e37c9e69 | 1277 | } |
cd1bc465 | 1278 | } |
9078a3e1 | 1279 | btrfs_set_block_group_used(&cache->item, old_val); |
db94535d CM |
1280 | total -= num_bytes; |
1281 | bytenr += num_bytes; | |
9078a3e1 CM |
1282 | } |
1283 | return 0; | |
1284 | } | |
6324fbf3 | 1285 | |
a061fc8d CM |
1286 | static u64 first_logical_byte(struct btrfs_root *root, u64 search_start) |
1287 | { | |
1288 | u64 start; | |
1289 | u64 end; | |
1290 | int ret; | |
1291 | ret = find_first_extent_bit(&root->fs_info->block_group_cache, | |
1292 | search_start, &start, &end, | |
1293 | BLOCK_GROUP_DATA | BLOCK_GROUP_METADATA | | |
1294 | BLOCK_GROUP_SYSTEM); | |
1295 | if (ret) | |
1296 | return 0; | |
1297 | return start; | |
1298 | } | |
1299 | ||
1300 | ||
324ae4df Y |
1301 | static int update_pinned_extents(struct btrfs_root *root, |
1302 | u64 bytenr, u64 num, int pin) | |
1303 | { | |
1304 | u64 len; | |
1305 | struct btrfs_block_group_cache *cache; | |
1306 | struct btrfs_fs_info *fs_info = root->fs_info; | |
1307 | ||
7d9eb12c | 1308 | WARN_ON(!mutex_is_locked(&root->fs_info->alloc_mutex)); |
324ae4df Y |
1309 | if (pin) { |
1310 | set_extent_dirty(&fs_info->pinned_extents, | |
1311 | bytenr, bytenr + num - 1, GFP_NOFS); | |
1312 | } else { | |
1313 | clear_extent_dirty(&fs_info->pinned_extents, | |
1314 | bytenr, bytenr + num - 1, GFP_NOFS); | |
1315 | } | |
1316 | while (num > 0) { | |
1317 | cache = btrfs_lookup_block_group(fs_info, bytenr); | |
a061fc8d CM |
1318 | if (!cache) { |
1319 | u64 first = first_logical_byte(root, bytenr); | |
1320 | WARN_ON(first < bytenr); | |
1321 | len = min(first - bytenr, num); | |
1322 | } else { | |
1323 | len = min(num, cache->key.offset - | |
1324 | (bytenr - cache->key.objectid)); | |
1325 | } | |
324ae4df | 1326 | if (pin) { |
a061fc8d CM |
1327 | if (cache) { |
1328 | cache->pinned += len; | |
1329 | cache->space_info->bytes_pinned += len; | |
1330 | } | |
324ae4df Y |
1331 | fs_info->total_pinned += len; |
1332 | } else { | |
a061fc8d CM |
1333 | if (cache) { |
1334 | cache->pinned -= len; | |
1335 | cache->space_info->bytes_pinned -= len; | |
1336 | } | |
324ae4df Y |
1337 | fs_info->total_pinned -= len; |
1338 | } | |
1339 | bytenr += len; | |
1340 | num -= len; | |
1341 | } | |
1342 | return 0; | |
1343 | } | |
9078a3e1 | 1344 | |
d1310b2e | 1345 | int btrfs_copy_pinned(struct btrfs_root *root, struct extent_io_tree *copy) |
ccd467d6 | 1346 | { |
ccd467d6 | 1347 | u64 last = 0; |
1a5bc167 CM |
1348 | u64 start; |
1349 | u64 end; | |
d1310b2e | 1350 | struct extent_io_tree *pinned_extents = &root->fs_info->pinned_extents; |
ccd467d6 | 1351 | int ret; |
ccd467d6 CM |
1352 | |
1353 | while(1) { | |
1a5bc167 CM |
1354 | ret = find_first_extent_bit(pinned_extents, last, |
1355 | &start, &end, EXTENT_DIRTY); | |
1356 | if (ret) | |
ccd467d6 | 1357 | break; |
1a5bc167 CM |
1358 | set_extent_dirty(copy, start, end, GFP_NOFS); |
1359 | last = end + 1; | |
ccd467d6 CM |
1360 | } |
1361 | return 0; | |
1362 | } | |
1363 | ||
1364 | int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans, | |
1365 | struct btrfs_root *root, | |
d1310b2e | 1366 | struct extent_io_tree *unpin) |
a28ec197 | 1367 | { |
1a5bc167 CM |
1368 | u64 start; |
1369 | u64 end; | |
a28ec197 | 1370 | int ret; |
d1310b2e | 1371 | struct extent_io_tree *free_space_cache; |
f510cfec | 1372 | free_space_cache = &root->fs_info->free_space_cache; |
a28ec197 | 1373 | |
925baedd | 1374 | mutex_lock(&root->fs_info->alloc_mutex); |
a28ec197 | 1375 | while(1) { |
1a5bc167 CM |
1376 | ret = find_first_extent_bit(unpin, 0, &start, &end, |
1377 | EXTENT_DIRTY); | |
1378 | if (ret) | |
a28ec197 | 1379 | break; |
324ae4df | 1380 | update_pinned_extents(root, start, end + 1 - start, 0); |
1a5bc167 CM |
1381 | clear_extent_dirty(unpin, start, end, GFP_NOFS); |
1382 | set_extent_dirty(free_space_cache, start, end, GFP_NOFS); | |
a28ec197 | 1383 | } |
925baedd | 1384 | mutex_unlock(&root->fs_info->alloc_mutex); |
a28ec197 CM |
1385 | return 0; |
1386 | } | |
1387 | ||
98ed5174 CM |
1388 | static int finish_current_insert(struct btrfs_trans_handle *trans, |
1389 | struct btrfs_root *extent_root) | |
037e6390 | 1390 | { |
7bb86316 CM |
1391 | u64 start; |
1392 | u64 end; | |
1393 | struct btrfs_fs_info *info = extent_root->fs_info; | |
d8d5f3e1 | 1394 | struct extent_buffer *eb; |
7bb86316 | 1395 | struct btrfs_path *path; |
e2fa7227 | 1396 | struct btrfs_key ins; |
d8d5f3e1 | 1397 | struct btrfs_disk_key first; |
234b63a0 | 1398 | struct btrfs_extent_item extent_item; |
037e6390 | 1399 | int ret; |
d8d5f3e1 | 1400 | int level; |
1a5bc167 | 1401 | int err = 0; |
037e6390 | 1402 | |
7d9eb12c | 1403 | WARN_ON(!mutex_is_locked(&extent_root->fs_info->alloc_mutex)); |
5f39d397 | 1404 | btrfs_set_stack_extent_refs(&extent_item, 1); |
62e2749e | 1405 | btrfs_set_key_type(&ins, BTRFS_EXTENT_ITEM_KEY); |
7bb86316 | 1406 | path = btrfs_alloc_path(); |
037e6390 | 1407 | |
26b8003f | 1408 | while(1) { |
1a5bc167 CM |
1409 | ret = find_first_extent_bit(&info->extent_ins, 0, &start, |
1410 | &end, EXTENT_LOCKED); | |
1411 | if (ret) | |
26b8003f CM |
1412 | break; |
1413 | ||
1a5bc167 CM |
1414 | ins.objectid = start; |
1415 | ins.offset = end + 1 - start; | |
1416 | err = btrfs_insert_item(trans, extent_root, &ins, | |
1417 | &extent_item, sizeof(extent_item)); | |
1418 | clear_extent_bits(&info->extent_ins, start, end, EXTENT_LOCKED, | |
1419 | GFP_NOFS); | |
ca7a79ad CM |
1420 | eb = read_tree_block(extent_root, ins.objectid, ins.offset, |
1421 | trans->transid); | |
925baedd | 1422 | btrfs_tree_lock(eb); |
d8d5f3e1 CM |
1423 | level = btrfs_header_level(eb); |
1424 | if (level == 0) { | |
1425 | btrfs_item_key(eb, &first, 0); | |
1426 | } else { | |
1427 | btrfs_node_key(eb, &first, 0); | |
1428 | } | |
925baedd CM |
1429 | btrfs_tree_unlock(eb); |
1430 | free_extent_buffer(eb); | |
1431 | /* | |
1432 | * the first key is just a hint, so the race we've created | |
1433 | * against reading it is fine | |
1434 | */ | |
7bb86316 CM |
1435 | err = btrfs_insert_extent_backref(trans, extent_root, path, |
1436 | start, extent_root->root_key.objectid, | |
f6dbff55 CM |
1437 | 0, level, |
1438 | btrfs_disk_key_objectid(&first)); | |
7bb86316 | 1439 | BUG_ON(err); |
037e6390 | 1440 | } |
7bb86316 | 1441 | btrfs_free_path(path); |
037e6390 CM |
1442 | return 0; |
1443 | } | |
1444 | ||
db94535d CM |
1445 | static int pin_down_bytes(struct btrfs_root *root, u64 bytenr, u32 num_bytes, |
1446 | int pending) | |
e20d96d6 | 1447 | { |
1a5bc167 | 1448 | int err = 0; |
8ef97622 | 1449 | |
7d9eb12c | 1450 | WARN_ON(!mutex_is_locked(&root->fs_info->alloc_mutex)); |
f4b9aa8d | 1451 | if (!pending) { |
925baedd | 1452 | struct extent_buffer *buf; |
db94535d | 1453 | buf = btrfs_find_tree_block(root, bytenr, num_bytes); |
5f39d397 | 1454 | if (buf) { |
a61e6f29 | 1455 | if (btrfs_try_tree_lock(buf) && |
925baedd | 1456 | btrfs_buffer_uptodate(buf, 0)) { |
2c90e5d6 CM |
1457 | u64 transid = |
1458 | root->fs_info->running_transaction->transid; | |
dc17ff8f CM |
1459 | u64 header_transid = |
1460 | btrfs_header_generation(buf); | |
6bc34676 CM |
1461 | if (header_transid == transid && |
1462 | !btrfs_header_flag(buf, | |
1463 | BTRFS_HEADER_FLAG_WRITTEN)) { | |
55c69072 | 1464 | clean_tree_block(NULL, root, buf); |
925baedd | 1465 | btrfs_tree_unlock(buf); |
5f39d397 | 1466 | free_extent_buffer(buf); |
c549228f | 1467 | return 1; |
2c90e5d6 | 1468 | } |
925baedd | 1469 | btrfs_tree_unlock(buf); |
f4b9aa8d | 1470 | } |
5f39d397 | 1471 | free_extent_buffer(buf); |
8ef97622 | 1472 | } |
324ae4df | 1473 | update_pinned_extents(root, bytenr, num_bytes, 1); |
f4b9aa8d | 1474 | } else { |
1a5bc167 | 1475 | set_extent_bits(&root->fs_info->pending_del, |
db94535d CM |
1476 | bytenr, bytenr + num_bytes - 1, |
1477 | EXTENT_LOCKED, GFP_NOFS); | |
f4b9aa8d | 1478 | } |
be744175 | 1479 | BUG_ON(err < 0); |
e20d96d6 CM |
1480 | return 0; |
1481 | } | |
1482 | ||
fec577fb | 1483 | /* |
a28ec197 | 1484 | * remove an extent from the root, returns 0 on success |
fec577fb | 1485 | */ |
e089f05c | 1486 | static int __free_extent(struct btrfs_trans_handle *trans, struct btrfs_root |
7bb86316 CM |
1487 | *root, u64 bytenr, u64 num_bytes, |
1488 | u64 root_objectid, u64 ref_generation, | |
1489 | u64 owner_objectid, u64 owner_offset, int pin, | |
e37c9e69 | 1490 | int mark_free) |
a28ec197 | 1491 | { |
5caf2a00 | 1492 | struct btrfs_path *path; |
e2fa7227 | 1493 | struct btrfs_key key; |
1261ec42 CM |
1494 | struct btrfs_fs_info *info = root->fs_info; |
1495 | struct btrfs_root *extent_root = info->extent_root; | |
5f39d397 | 1496 | struct extent_buffer *leaf; |
a28ec197 | 1497 | int ret; |
952fccac CM |
1498 | int extent_slot = 0; |
1499 | int found_extent = 0; | |
1500 | int num_to_del = 1; | |
234b63a0 | 1501 | struct btrfs_extent_item *ei; |
cf27e1ee | 1502 | u32 refs; |
037e6390 | 1503 | |
7d9eb12c | 1504 | WARN_ON(!mutex_is_locked(&root->fs_info->alloc_mutex)); |
db94535d | 1505 | key.objectid = bytenr; |
62e2749e | 1506 | btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY); |
db94535d | 1507 | key.offset = num_bytes; |
5caf2a00 | 1508 | path = btrfs_alloc_path(); |
54aa1f4d CM |
1509 | if (!path) |
1510 | return -ENOMEM; | |
5f26f772 | 1511 | |
3c12ac72 | 1512 | path->reada = 1; |
7bb86316 CM |
1513 | ret = lookup_extent_backref(trans, extent_root, path, |
1514 | bytenr, root_objectid, | |
1515 | ref_generation, | |
1516 | owner_objectid, owner_offset, 1); | |
1517 | if (ret == 0) { | |
952fccac CM |
1518 | struct btrfs_key found_key; |
1519 | extent_slot = path->slots[0]; | |
1520 | while(extent_slot > 0) { | |
1521 | extent_slot--; | |
1522 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, | |
1523 | extent_slot); | |
1524 | if (found_key.objectid != bytenr) | |
1525 | break; | |
1526 | if (found_key.type == BTRFS_EXTENT_ITEM_KEY && | |
1527 | found_key.offset == num_bytes) { | |
1528 | found_extent = 1; | |
1529 | break; | |
1530 | } | |
1531 | if (path->slots[0] - extent_slot > 5) | |
1532 | break; | |
1533 | } | |
1534 | if (!found_extent) | |
1535 | ret = btrfs_del_item(trans, extent_root, path); | |
7bb86316 CM |
1536 | } else { |
1537 | btrfs_print_leaf(extent_root, path->nodes[0]); | |
1538 | WARN_ON(1); | |
1539 | printk("Unable to find ref byte nr %Lu root %Lu " | |
1540 | " gen %Lu owner %Lu offset %Lu\n", bytenr, | |
1541 | root_objectid, ref_generation, owner_objectid, | |
1542 | owner_offset); | |
1543 | } | |
952fccac CM |
1544 | if (!found_extent) { |
1545 | btrfs_release_path(extent_root, path); | |
1546 | ret = btrfs_search_slot(trans, extent_root, &key, path, -1, 1); | |
1547 | if (ret < 0) | |
1548 | return ret; | |
1549 | BUG_ON(ret); | |
1550 | extent_slot = path->slots[0]; | |
1551 | } | |
5f39d397 CM |
1552 | |
1553 | leaf = path->nodes[0]; | |
952fccac | 1554 | ei = btrfs_item_ptr(leaf, extent_slot, |
123abc88 | 1555 | struct btrfs_extent_item); |
5f39d397 CM |
1556 | refs = btrfs_extent_refs(leaf, ei); |
1557 | BUG_ON(refs == 0); | |
1558 | refs -= 1; | |
1559 | btrfs_set_extent_refs(leaf, ei, refs); | |
952fccac | 1560 | |
5f39d397 CM |
1561 | btrfs_mark_buffer_dirty(leaf); |
1562 | ||
952fccac CM |
1563 | if (refs == 0 && found_extent && path->slots[0] == extent_slot + 1) { |
1564 | /* if the back ref and the extent are next to each other | |
1565 | * they get deleted below in one shot | |
1566 | */ | |
1567 | path->slots[0] = extent_slot; | |
1568 | num_to_del = 2; | |
1569 | } else if (found_extent) { | |
1570 | /* otherwise delete the extent back ref */ | |
1571 | ret = btrfs_del_item(trans, extent_root, path); | |
1572 | BUG_ON(ret); | |
1573 | /* if refs are 0, we need to setup the path for deletion */ | |
1574 | if (refs == 0) { | |
1575 | btrfs_release_path(extent_root, path); | |
1576 | ret = btrfs_search_slot(trans, extent_root, &key, path, | |
1577 | -1, 1); | |
1578 | if (ret < 0) | |
1579 | return ret; | |
1580 | BUG_ON(ret); | |
1581 | } | |
1582 | } | |
1583 | ||
cf27e1ee | 1584 | if (refs == 0) { |
db94535d CM |
1585 | u64 super_used; |
1586 | u64 root_used; | |
78fae27e CM |
1587 | |
1588 | if (pin) { | |
db94535d | 1589 | ret = pin_down_bytes(root, bytenr, num_bytes, 0); |
c549228f Y |
1590 | if (ret > 0) |
1591 | mark_free = 1; | |
1592 | BUG_ON(ret < 0); | |
78fae27e CM |
1593 | } |
1594 | ||
58176a96 | 1595 | /* block accounting for super block */ |
a2135011 | 1596 | spin_lock_irq(&info->delalloc_lock); |
db94535d CM |
1597 | super_used = btrfs_super_bytes_used(&info->super_copy); |
1598 | btrfs_set_super_bytes_used(&info->super_copy, | |
1599 | super_used - num_bytes); | |
a2135011 | 1600 | spin_unlock_irq(&info->delalloc_lock); |
58176a96 JB |
1601 | |
1602 | /* block accounting for root item */ | |
db94535d | 1603 | root_used = btrfs_root_used(&root->root_item); |
5f39d397 | 1604 | btrfs_set_root_used(&root->root_item, |
db94535d | 1605 | root_used - num_bytes); |
952fccac CM |
1606 | ret = btrfs_del_items(trans, extent_root, path, path->slots[0], |
1607 | num_to_del); | |
54aa1f4d CM |
1608 | if (ret) { |
1609 | return ret; | |
1610 | } | |
db94535d | 1611 | ret = update_block_group(trans, root, bytenr, num_bytes, 0, |
0b86a832 | 1612 | mark_free); |
9078a3e1 | 1613 | BUG_ON(ret); |
a28ec197 | 1614 | } |
5caf2a00 | 1615 | btrfs_free_path(path); |
e089f05c | 1616 | finish_current_insert(trans, extent_root); |
a28ec197 CM |
1617 | return ret; |
1618 | } | |
1619 | ||
a28ec197 CM |
1620 | /* |
1621 | * find all the blocks marked as pending in the radix tree and remove | |
1622 | * them from the extent map | |
1623 | */ | |
e089f05c CM |
1624 | static int del_pending_extents(struct btrfs_trans_handle *trans, struct |
1625 | btrfs_root *extent_root) | |
a28ec197 CM |
1626 | { |
1627 | int ret; | |
e20d96d6 | 1628 | int err = 0; |
1a5bc167 CM |
1629 | u64 start; |
1630 | u64 end; | |
d1310b2e CM |
1631 | struct extent_io_tree *pending_del; |
1632 | struct extent_io_tree *pinned_extents; | |
8ef97622 | 1633 | |
7d9eb12c | 1634 | WARN_ON(!mutex_is_locked(&extent_root->fs_info->alloc_mutex)); |
1a5bc167 CM |
1635 | pending_del = &extent_root->fs_info->pending_del; |
1636 | pinned_extents = &extent_root->fs_info->pinned_extents; | |
a28ec197 CM |
1637 | |
1638 | while(1) { | |
1a5bc167 CM |
1639 | ret = find_first_extent_bit(pending_del, 0, &start, &end, |
1640 | EXTENT_LOCKED); | |
1641 | if (ret) | |
a28ec197 | 1642 | break; |
324ae4df | 1643 | update_pinned_extents(extent_root, start, end + 1 - start, 1); |
1a5bc167 CM |
1644 | clear_extent_bits(pending_del, start, end, EXTENT_LOCKED, |
1645 | GFP_NOFS); | |
1646 | ret = __free_extent(trans, extent_root, | |
7bb86316 CM |
1647 | start, end + 1 - start, |
1648 | extent_root->root_key.objectid, | |
1649 | 0, 0, 0, 0, 0); | |
1a5bc167 CM |
1650 | if (ret) |
1651 | err = ret; | |
fec577fb | 1652 | } |
e20d96d6 | 1653 | return err; |
fec577fb CM |
1654 | } |
1655 | ||
1656 | /* | |
1657 | * remove an extent from the root, returns 0 on success | |
1658 | */ | |
925baedd CM |
1659 | static int __btrfs_free_extent(struct btrfs_trans_handle *trans, |
1660 | struct btrfs_root *root, u64 bytenr, | |
1661 | u64 num_bytes, u64 root_objectid, | |
1662 | u64 ref_generation, u64 owner_objectid, | |
1663 | u64 owner_offset, int pin) | |
fec577fb | 1664 | { |
9f5fae2f | 1665 | struct btrfs_root *extent_root = root->fs_info->extent_root; |
fec577fb CM |
1666 | int pending_ret; |
1667 | int ret; | |
a28ec197 | 1668 | |
db94535d | 1669 | WARN_ON(num_bytes < root->sectorsize); |
7bb86316 CM |
1670 | if (!root->ref_cows) |
1671 | ref_generation = 0; | |
1672 | ||
fec577fb | 1673 | if (root == extent_root) { |
db94535d | 1674 | pin_down_bytes(root, bytenr, num_bytes, 1); |
fec577fb CM |
1675 | return 0; |
1676 | } | |
7bb86316 CM |
1677 | ret = __free_extent(trans, root, bytenr, num_bytes, root_objectid, |
1678 | ref_generation, owner_objectid, owner_offset, | |
1679 | pin, pin == 0); | |
ee6e6504 CM |
1680 | |
1681 | finish_current_insert(trans, root->fs_info->extent_root); | |
e20d96d6 | 1682 | pending_ret = del_pending_extents(trans, root->fs_info->extent_root); |
fec577fb CM |
1683 | return ret ? ret : pending_ret; |
1684 | } | |
1685 | ||
925baedd CM |
1686 | int btrfs_free_extent(struct btrfs_trans_handle *trans, |
1687 | struct btrfs_root *root, u64 bytenr, | |
1688 | u64 num_bytes, u64 root_objectid, | |
1689 | u64 ref_generation, u64 owner_objectid, | |
1690 | u64 owner_offset, int pin) | |
1691 | { | |
1692 | int ret; | |
1693 | ||
1694 | maybe_lock_mutex(root); | |
1695 | ret = __btrfs_free_extent(trans, root, bytenr, num_bytes, | |
1696 | root_objectid, ref_generation, | |
1697 | owner_objectid, owner_offset, pin); | |
1698 | maybe_unlock_mutex(root); | |
1699 | return ret; | |
1700 | } | |
1701 | ||
87ee04eb CM |
1702 | static u64 stripe_align(struct btrfs_root *root, u64 val) |
1703 | { | |
1704 | u64 mask = ((u64)root->stripesize - 1); | |
1705 | u64 ret = (val + mask) & ~mask; | |
1706 | return ret; | |
1707 | } | |
1708 | ||
fec577fb CM |
1709 | /* |
1710 | * walks the btree of allocated extents and find a hole of a given size. | |
1711 | * The key ins is changed to record the hole: | |
1712 | * ins->objectid == block start | |
62e2749e | 1713 | * ins->flags = BTRFS_EXTENT_ITEM_KEY |
fec577fb CM |
1714 | * ins->offset == number of blocks |
1715 | * Any available blocks before search_start are skipped. | |
1716 | */ | |
98ed5174 CM |
1717 | static int noinline find_free_extent(struct btrfs_trans_handle *trans, |
1718 | struct btrfs_root *orig_root, | |
1719 | u64 num_bytes, u64 empty_size, | |
1720 | u64 search_start, u64 search_end, | |
1721 | u64 hint_byte, struct btrfs_key *ins, | |
1722 | u64 exclude_start, u64 exclude_nr, | |
1723 | int data) | |
fec577fb | 1724 | { |
87ee04eb | 1725 | int ret; |
a061fc8d | 1726 | u64 orig_search_start; |
9f5fae2f | 1727 | struct btrfs_root * root = orig_root->fs_info->extent_root; |
f2458e1d | 1728 | struct btrfs_fs_info *info = root->fs_info; |
db94535d | 1729 | u64 total_needed = num_bytes; |
239b14b3 | 1730 | u64 *last_ptr = NULL; |
be08c1b9 | 1731 | struct btrfs_block_group_cache *block_group; |
be744175 | 1732 | int full_scan = 0; |
fbdc762b | 1733 | int wrapped = 0; |
0ef3e66b | 1734 | int chunk_alloc_done = 0; |
239b14b3 | 1735 | int empty_cluster = 2 * 1024 * 1024; |
0ef3e66b | 1736 | int allowed_chunk_alloc = 0; |
fec577fb | 1737 | |
db94535d | 1738 | WARN_ON(num_bytes < root->sectorsize); |
b1a4d965 CM |
1739 | btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY); |
1740 | ||
0ef3e66b CM |
1741 | if (orig_root->ref_cows || empty_size) |
1742 | allowed_chunk_alloc = 1; | |
1743 | ||
239b14b3 CM |
1744 | if (data & BTRFS_BLOCK_GROUP_METADATA) { |
1745 | last_ptr = &root->fs_info->last_alloc; | |
8790d502 | 1746 | empty_cluster = 256 * 1024; |
239b14b3 CM |
1747 | } |
1748 | ||
1749 | if ((data & BTRFS_BLOCK_GROUP_DATA) && btrfs_test_opt(root, SSD)) { | |
1750 | last_ptr = &root->fs_info->last_data_alloc; | |
1751 | } | |
1752 | ||
1753 | if (last_ptr) { | |
1754 | if (*last_ptr) | |
1755 | hint_byte = *last_ptr; | |
1756 | else { | |
1757 | empty_size += empty_cluster; | |
1758 | } | |
1759 | } | |
1760 | ||
a061fc8d CM |
1761 | search_start = max(search_start, first_logical_byte(root, 0)); |
1762 | orig_search_start = search_start; | |
1763 | ||
7f93bf8d CM |
1764 | if (search_end == (u64)-1) |
1765 | search_end = btrfs_super_total_bytes(&info->super_copy); | |
0b86a832 | 1766 | |
db94535d | 1767 | if (hint_byte) { |
0ef3e66b | 1768 | block_group = btrfs_lookup_first_block_group(info, hint_byte); |
1a2b2ac7 CM |
1769 | if (!block_group) |
1770 | hint_byte = search_start; | |
925baedd | 1771 | block_group = __btrfs_find_block_group(root, block_group, |
db94535d | 1772 | hint_byte, data, 1); |
239b14b3 CM |
1773 | if (last_ptr && *last_ptr == 0 && block_group) |
1774 | hint_byte = block_group->key.objectid; | |
be744175 | 1775 | } else { |
925baedd | 1776 | block_group = __btrfs_find_block_group(root, |
1a2b2ac7 CM |
1777 | trans->block_group, |
1778 | search_start, data, 1); | |
be744175 | 1779 | } |
239b14b3 | 1780 | search_start = max(search_start, hint_byte); |
be744175 | 1781 | |
6702ed49 | 1782 | total_needed += empty_size; |
0b86a832 | 1783 | |
be744175 | 1784 | check_failed: |
70b043f0 | 1785 | if (!block_group) { |
0ef3e66b CM |
1786 | block_group = btrfs_lookup_first_block_group(info, |
1787 | search_start); | |
70b043f0 | 1788 | if (!block_group) |
0ef3e66b | 1789 | block_group = btrfs_lookup_first_block_group(info, |
70b043f0 CM |
1790 | orig_search_start); |
1791 | } | |
0ef3e66b CM |
1792 | if (full_scan && !chunk_alloc_done) { |
1793 | if (allowed_chunk_alloc) { | |
1794 | do_chunk_alloc(trans, root, | |
1795 | num_bytes + 2 * 1024 * 1024, data, 1); | |
1796 | allowed_chunk_alloc = 0; | |
1797 | } else if (block_group && block_group_bits(block_group, data)) { | |
1798 | block_group->space_info->force_alloc = 1; | |
1799 | } | |
1800 | chunk_alloc_done = 1; | |
1801 | } | |
0b86a832 CM |
1802 | ret = find_search_start(root, &block_group, &search_start, |
1803 | total_needed, data); | |
239b14b3 CM |
1804 | if (ret == -ENOSPC && last_ptr && *last_ptr) { |
1805 | *last_ptr = 0; | |
0ef3e66b CM |
1806 | block_group = btrfs_lookup_first_block_group(info, |
1807 | orig_search_start); | |
239b14b3 CM |
1808 | search_start = orig_search_start; |
1809 | ret = find_search_start(root, &block_group, &search_start, | |
1810 | total_needed, data); | |
1811 | } | |
1812 | if (ret == -ENOSPC) | |
1813 | goto enospc; | |
0b86a832 | 1814 | if (ret) |
d548ee51 | 1815 | goto error; |
e19caa5f | 1816 | |
239b14b3 CM |
1817 | if (last_ptr && *last_ptr && search_start != *last_ptr) { |
1818 | *last_ptr = 0; | |
1819 | if (!empty_size) { | |
1820 | empty_size += empty_cluster; | |
1821 | total_needed += empty_size; | |
1822 | } | |
0ef3e66b | 1823 | block_group = btrfs_lookup_first_block_group(info, |
239b14b3 CM |
1824 | orig_search_start); |
1825 | search_start = orig_search_start; | |
1826 | ret = find_search_start(root, &block_group, | |
1827 | &search_start, total_needed, data); | |
1828 | if (ret == -ENOSPC) | |
1829 | goto enospc; | |
1830 | if (ret) | |
1831 | goto error; | |
1832 | } | |
1833 | ||
0b86a832 CM |
1834 | search_start = stripe_align(root, search_start); |
1835 | ins->objectid = search_start; | |
1836 | ins->offset = num_bytes; | |
e37c9e69 | 1837 | |
db94535d | 1838 | if (ins->objectid + num_bytes >= search_end) |
cf67582b | 1839 | goto enospc; |
0b86a832 CM |
1840 | |
1841 | if (ins->objectid + num_bytes > | |
1842 | block_group->key.objectid + block_group->key.offset) { | |
e19caa5f CM |
1843 | search_start = block_group->key.objectid + |
1844 | block_group->key.offset; | |
1845 | goto new_group; | |
1846 | } | |
0b86a832 | 1847 | |
1a5bc167 | 1848 | if (test_range_bit(&info->extent_ins, ins->objectid, |
db94535d CM |
1849 | ins->objectid + num_bytes -1, EXTENT_LOCKED, 0)) { |
1850 | search_start = ins->objectid + num_bytes; | |
1a5bc167 CM |
1851 | goto new_group; |
1852 | } | |
0b86a832 | 1853 | |
1a5bc167 | 1854 | if (test_range_bit(&info->pinned_extents, ins->objectid, |
db94535d CM |
1855 | ins->objectid + num_bytes -1, EXTENT_DIRTY, 0)) { |
1856 | search_start = ins->objectid + num_bytes; | |
1a5bc167 | 1857 | goto new_group; |
fec577fb | 1858 | } |
0b86a832 | 1859 | |
db94535d | 1860 | if (exclude_nr > 0 && (ins->objectid + num_bytes > exclude_start && |
f2654de4 CM |
1861 | ins->objectid < exclude_start + exclude_nr)) { |
1862 | search_start = exclude_start + exclude_nr; | |
1863 | goto new_group; | |
1864 | } | |
0b86a832 | 1865 | |
6324fbf3 | 1866 | if (!(data & BTRFS_BLOCK_GROUP_DATA)) { |
5276aeda | 1867 | block_group = btrfs_lookup_block_group(info, ins->objectid); |
26b8003f CM |
1868 | if (block_group) |
1869 | trans->block_group = block_group; | |
f2458e1d | 1870 | } |
db94535d | 1871 | ins->offset = num_bytes; |
239b14b3 CM |
1872 | if (last_ptr) { |
1873 | *last_ptr = ins->objectid + ins->offset; | |
1874 | if (*last_ptr == | |
1875 | btrfs_super_total_bytes(&root->fs_info->super_copy)) { | |
1876 | *last_ptr = 0; | |
1877 | } | |
1878 | } | |
fec577fb | 1879 | return 0; |
be744175 CM |
1880 | |
1881 | new_group: | |
db94535d | 1882 | if (search_start + num_bytes >= search_end) { |
cf67582b | 1883 | enospc: |
be744175 | 1884 | search_start = orig_search_start; |
fbdc762b CM |
1885 | if (full_scan) { |
1886 | ret = -ENOSPC; | |
1887 | goto error; | |
1888 | } | |
6702ed49 CM |
1889 | if (wrapped) { |
1890 | if (!full_scan) | |
1891 | total_needed -= empty_size; | |
fbdc762b | 1892 | full_scan = 1; |
6702ed49 | 1893 | } else |
fbdc762b | 1894 | wrapped = 1; |
be744175 | 1895 | } |
0ef3e66b | 1896 | block_group = btrfs_lookup_first_block_group(info, search_start); |
fbdc762b | 1897 | cond_resched(); |
925baedd | 1898 | block_group = __btrfs_find_block_group(root, block_group, |
1a2b2ac7 | 1899 | search_start, data, 0); |
be744175 CM |
1900 | goto check_failed; |
1901 | ||
0f70abe2 | 1902 | error: |
0f70abe2 | 1903 | return ret; |
fec577fb | 1904 | } |
ec44a35c | 1905 | |
e6dcd2dc CM |
1906 | static int __btrfs_reserve_extent(struct btrfs_trans_handle *trans, |
1907 | struct btrfs_root *root, | |
1908 | u64 num_bytes, u64 min_alloc_size, | |
1909 | u64 empty_size, u64 hint_byte, | |
1910 | u64 search_end, struct btrfs_key *ins, | |
1911 | u64 data) | |
fec577fb CM |
1912 | { |
1913 | int ret; | |
fbdc762b | 1914 | u64 search_start = 0; |
8790d502 | 1915 | u64 alloc_profile; |
1261ec42 | 1916 | struct btrfs_fs_info *info = root->fs_info; |
925baedd | 1917 | |
6324fbf3 | 1918 | if (data) { |
8790d502 CM |
1919 | alloc_profile = info->avail_data_alloc_bits & |
1920 | info->data_alloc_profile; | |
1921 | data = BTRFS_BLOCK_GROUP_DATA | alloc_profile; | |
6324fbf3 | 1922 | } else if (root == root->fs_info->chunk_root) { |
8790d502 CM |
1923 | alloc_profile = info->avail_system_alloc_bits & |
1924 | info->system_alloc_profile; | |
1925 | data = BTRFS_BLOCK_GROUP_SYSTEM | alloc_profile; | |
6324fbf3 | 1926 | } else { |
8790d502 CM |
1927 | alloc_profile = info->avail_metadata_alloc_bits & |
1928 | info->metadata_alloc_profile; | |
1929 | data = BTRFS_BLOCK_GROUP_METADATA | alloc_profile; | |
6324fbf3 | 1930 | } |
98d20f67 | 1931 | again: |
a061fc8d | 1932 | data = reduce_alloc_profile(root, data); |
0ef3e66b CM |
1933 | /* |
1934 | * the only place that sets empty_size is btrfs_realloc_node, which | |
1935 | * is not called recursively on allocations | |
1936 | */ | |
1937 | if (empty_size || root->ref_cows) { | |
593060d7 | 1938 | if (!(data & BTRFS_BLOCK_GROUP_METADATA)) { |
6324fbf3 | 1939 | ret = do_chunk_alloc(trans, root->fs_info->extent_root, |
0ef3e66b CM |
1940 | 2 * 1024 * 1024, |
1941 | BTRFS_BLOCK_GROUP_METADATA | | |
1942 | (info->metadata_alloc_profile & | |
1943 | info->avail_metadata_alloc_bits), 0); | |
6324fbf3 CM |
1944 | BUG_ON(ret); |
1945 | } | |
1946 | ret = do_chunk_alloc(trans, root->fs_info->extent_root, | |
0ef3e66b | 1947 | num_bytes + 2 * 1024 * 1024, data, 0); |
6324fbf3 CM |
1948 | BUG_ON(ret); |
1949 | } | |
0b86a832 | 1950 | |
db94535d CM |
1951 | WARN_ON(num_bytes < root->sectorsize); |
1952 | ret = find_free_extent(trans, root, num_bytes, empty_size, | |
1953 | search_start, search_end, hint_byte, ins, | |
26b8003f CM |
1954 | trans->alloc_exclude_start, |
1955 | trans->alloc_exclude_nr, data); | |
3b951516 | 1956 | |
98d20f67 CM |
1957 | if (ret == -ENOSPC && num_bytes > min_alloc_size) { |
1958 | num_bytes = num_bytes >> 1; | |
1959 | num_bytes = max(num_bytes, min_alloc_size); | |
0ef3e66b CM |
1960 | do_chunk_alloc(trans, root->fs_info->extent_root, |
1961 | num_bytes, data, 1); | |
98d20f67 CM |
1962 | goto again; |
1963 | } | |
ec44a35c CM |
1964 | if (ret) { |
1965 | printk("allocation failed flags %Lu\n", data); | |
925baedd | 1966 | BUG(); |
925baedd | 1967 | } |
e6dcd2dc CM |
1968 | clear_extent_dirty(&root->fs_info->free_space_cache, |
1969 | ins->objectid, ins->objectid + ins->offset - 1, | |
1970 | GFP_NOFS); | |
1971 | return 0; | |
1972 | } | |
1973 | ||
1974 | int btrfs_reserve_extent(struct btrfs_trans_handle *trans, | |
1975 | struct btrfs_root *root, | |
1976 | u64 num_bytes, u64 min_alloc_size, | |
1977 | u64 empty_size, u64 hint_byte, | |
1978 | u64 search_end, struct btrfs_key *ins, | |
1979 | u64 data) | |
1980 | { | |
1981 | int ret; | |
1982 | maybe_lock_mutex(root); | |
1983 | ret = __btrfs_reserve_extent(trans, root, num_bytes, min_alloc_size, | |
1984 | empty_size, hint_byte, search_end, ins, | |
1985 | data); | |
1986 | maybe_unlock_mutex(root); | |
1987 | return ret; | |
1988 | } | |
1989 | ||
1990 | static int __btrfs_alloc_reserved_extent(struct btrfs_trans_handle *trans, | |
1991 | struct btrfs_root *root, | |
1992 | u64 root_objectid, u64 ref_generation, | |
1993 | u64 owner, u64 owner_offset, | |
1994 | struct btrfs_key *ins) | |
1995 | { | |
1996 | int ret; | |
1997 | int pending_ret; | |
1998 | u64 super_used; | |
1999 | u64 root_used; | |
2000 | u64 num_bytes = ins->offset; | |
2001 | u32 sizes[2]; | |
2002 | struct btrfs_fs_info *info = root->fs_info; | |
2003 | struct btrfs_root *extent_root = info->extent_root; | |
2004 | struct btrfs_extent_item *extent_item; | |
2005 | struct btrfs_extent_ref *ref; | |
2006 | struct btrfs_path *path; | |
2007 | struct btrfs_key keys[2]; | |
fec577fb | 2008 | |
58176a96 | 2009 | /* block accounting for super block */ |
a2135011 | 2010 | spin_lock_irq(&info->delalloc_lock); |
db94535d CM |
2011 | super_used = btrfs_super_bytes_used(&info->super_copy); |
2012 | btrfs_set_super_bytes_used(&info->super_copy, super_used + num_bytes); | |
a2135011 | 2013 | spin_unlock_irq(&info->delalloc_lock); |
26b8003f | 2014 | |
58176a96 | 2015 | /* block accounting for root item */ |
db94535d CM |
2016 | root_used = btrfs_root_used(&root->root_item); |
2017 | btrfs_set_root_used(&root->root_item, root_used + num_bytes); | |
58176a96 | 2018 | |
26b8003f | 2019 | if (root == extent_root) { |
1a5bc167 CM |
2020 | set_extent_bits(&root->fs_info->extent_ins, ins->objectid, |
2021 | ins->objectid + ins->offset - 1, | |
2022 | EXTENT_LOCKED, GFP_NOFS); | |
26b8003f CM |
2023 | goto update_block; |
2024 | } | |
2025 | ||
47e4bb98 CM |
2026 | memcpy(&keys[0], ins, sizeof(*ins)); |
2027 | keys[1].offset = hash_extent_ref(root_objectid, ref_generation, | |
2028 | owner, owner_offset); | |
2029 | keys[1].objectid = ins->objectid; | |
2030 | keys[1].type = BTRFS_EXTENT_REF_KEY; | |
2031 | sizes[0] = sizeof(*extent_item); | |
2032 | sizes[1] = sizeof(*ref); | |
7bb86316 CM |
2033 | |
2034 | path = btrfs_alloc_path(); | |
2035 | BUG_ON(!path); | |
47e4bb98 CM |
2036 | |
2037 | ret = btrfs_insert_empty_items(trans, extent_root, path, keys, | |
2038 | sizes, 2); | |
26b8003f | 2039 | |
ccd467d6 | 2040 | BUG_ON(ret); |
47e4bb98 CM |
2041 | extent_item = btrfs_item_ptr(path->nodes[0], path->slots[0], |
2042 | struct btrfs_extent_item); | |
2043 | btrfs_set_extent_refs(path->nodes[0], extent_item, 1); | |
2044 | ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1, | |
2045 | struct btrfs_extent_ref); | |
2046 | ||
2047 | btrfs_set_ref_root(path->nodes[0], ref, root_objectid); | |
2048 | btrfs_set_ref_generation(path->nodes[0], ref, ref_generation); | |
2049 | btrfs_set_ref_objectid(path->nodes[0], ref, owner); | |
2050 | btrfs_set_ref_offset(path->nodes[0], ref, owner_offset); | |
2051 | ||
2052 | btrfs_mark_buffer_dirty(path->nodes[0]); | |
2053 | ||
2054 | trans->alloc_exclude_start = 0; | |
2055 | trans->alloc_exclude_nr = 0; | |
7bb86316 | 2056 | btrfs_free_path(path); |
e089f05c | 2057 | finish_current_insert(trans, extent_root); |
e20d96d6 | 2058 | pending_ret = del_pending_extents(trans, extent_root); |
f510cfec | 2059 | |
925baedd CM |
2060 | if (ret) |
2061 | goto out; | |
e37c9e69 | 2062 | if (pending_ret) { |
925baedd CM |
2063 | ret = pending_ret; |
2064 | goto out; | |
e37c9e69 | 2065 | } |
26b8003f CM |
2066 | |
2067 | update_block: | |
0b86a832 | 2068 | ret = update_block_group(trans, root, ins->objectid, ins->offset, 1, 0); |
f5947066 CM |
2069 | if (ret) { |
2070 | printk("update block group failed for %Lu %Lu\n", | |
2071 | ins->objectid, ins->offset); | |
2072 | BUG(); | |
2073 | } | |
925baedd | 2074 | out: |
e6dcd2dc CM |
2075 | return ret; |
2076 | } | |
2077 | ||
2078 | int btrfs_alloc_reserved_extent(struct btrfs_trans_handle *trans, | |
2079 | struct btrfs_root *root, | |
2080 | u64 root_objectid, u64 ref_generation, | |
2081 | u64 owner, u64 owner_offset, | |
2082 | struct btrfs_key *ins) | |
2083 | { | |
2084 | int ret; | |
2085 | maybe_lock_mutex(root); | |
2086 | ret = __btrfs_alloc_reserved_extent(trans, root, root_objectid, | |
2087 | ref_generation, owner, | |
2088 | owner_offset, ins); | |
2089 | maybe_unlock_mutex(root); | |
2090 | return ret; | |
2091 | } | |
2092 | /* | |
2093 | * finds a free extent and does all the dirty work required for allocation | |
2094 | * returns the key for the extent through ins, and a tree buffer for | |
2095 | * the first block of the extent through buf. | |
2096 | * | |
2097 | * returns 0 if everything worked, non-zero otherwise. | |
2098 | */ | |
2099 | int btrfs_alloc_extent(struct btrfs_trans_handle *trans, | |
2100 | struct btrfs_root *root, | |
2101 | u64 num_bytes, u64 min_alloc_size, | |
2102 | u64 root_objectid, u64 ref_generation, | |
2103 | u64 owner, u64 owner_offset, | |
2104 | u64 empty_size, u64 hint_byte, | |
2105 | u64 search_end, struct btrfs_key *ins, u64 data) | |
2106 | { | |
2107 | int ret; | |
2108 | ||
2109 | maybe_lock_mutex(root); | |
2110 | ||
2111 | ret = __btrfs_reserve_extent(trans, root, num_bytes, | |
2112 | min_alloc_size, empty_size, hint_byte, | |
2113 | search_end, ins, data); | |
2114 | BUG_ON(ret); | |
2115 | ret = __btrfs_alloc_reserved_extent(trans, root, root_objectid, | |
2116 | ref_generation, owner, | |
2117 | owner_offset, ins); | |
2118 | BUG_ON(ret); | |
2119 | ||
925baedd CM |
2120 | maybe_unlock_mutex(root); |
2121 | return ret; | |
fec577fb | 2122 | } |
fec577fb CM |
2123 | /* |
2124 | * helper function to allocate a block for a given tree | |
2125 | * returns the tree buffer or NULL. | |
2126 | */ | |
5f39d397 | 2127 | struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans, |
7bb86316 CM |
2128 | struct btrfs_root *root, |
2129 | u32 blocksize, | |
2130 | u64 root_objectid, | |
2131 | u64 ref_generation, | |
2132 | u64 first_objectid, | |
2133 | int level, | |
2134 | u64 hint, | |
5f39d397 | 2135 | u64 empty_size) |
fec577fb | 2136 | { |
e2fa7227 | 2137 | struct btrfs_key ins; |
fec577fb | 2138 | int ret; |
5f39d397 | 2139 | struct extent_buffer *buf; |
fec577fb | 2140 | |
98d20f67 | 2141 | ret = btrfs_alloc_extent(trans, root, blocksize, blocksize, |
7bb86316 | 2142 | root_objectid, ref_generation, |
f6dbff55 | 2143 | level, first_objectid, empty_size, hint, |
db94535d | 2144 | (u64)-1, &ins, 0); |
fec577fb | 2145 | if (ret) { |
54aa1f4d CM |
2146 | BUG_ON(ret > 0); |
2147 | return ERR_PTR(ret); | |
fec577fb | 2148 | } |
db94535d | 2149 | buf = btrfs_find_create_tree_block(root, ins.objectid, blocksize); |
54aa1f4d | 2150 | if (!buf) { |
7bb86316 CM |
2151 | btrfs_free_extent(trans, root, ins.objectid, blocksize, |
2152 | root->root_key.objectid, ref_generation, | |
2153 | 0, 0, 0); | |
54aa1f4d CM |
2154 | return ERR_PTR(-ENOMEM); |
2155 | } | |
55c69072 | 2156 | btrfs_set_header_generation(buf, trans->transid); |
925baedd | 2157 | btrfs_tree_lock(buf); |
55c69072 | 2158 | clean_tree_block(trans, root, buf); |
5f39d397 | 2159 | btrfs_set_buffer_uptodate(buf); |
55c69072 CM |
2160 | |
2161 | if (PageDirty(buf->first_page)) { | |
2162 | printk("page %lu dirty\n", buf->first_page->index); | |
2163 | WARN_ON(1); | |
2164 | } | |
2165 | ||
5f39d397 CM |
2166 | set_extent_dirty(&trans->transaction->dirty_pages, buf->start, |
2167 | buf->start + buf->len - 1, GFP_NOFS); | |
d3c2fdcf | 2168 | trans->blocks_used++; |
fec577fb CM |
2169 | return buf; |
2170 | } | |
a28ec197 | 2171 | |
98ed5174 CM |
2172 | static int noinline drop_leaf_ref(struct btrfs_trans_handle *trans, |
2173 | struct btrfs_root *root, | |
2174 | struct extent_buffer *leaf) | |
6407bf6d | 2175 | { |
7bb86316 CM |
2176 | u64 leaf_owner; |
2177 | u64 leaf_generation; | |
5f39d397 | 2178 | struct btrfs_key key; |
6407bf6d CM |
2179 | struct btrfs_file_extent_item *fi; |
2180 | int i; | |
2181 | int nritems; | |
2182 | int ret; | |
2183 | ||
5f39d397 CM |
2184 | BUG_ON(!btrfs_is_leaf(leaf)); |
2185 | nritems = btrfs_header_nritems(leaf); | |
7bb86316 CM |
2186 | leaf_owner = btrfs_header_owner(leaf); |
2187 | leaf_generation = btrfs_header_generation(leaf); | |
2188 | ||
4a096752 CM |
2189 | mutex_unlock(&root->fs_info->alloc_mutex); |
2190 | ||
6407bf6d | 2191 | for (i = 0; i < nritems; i++) { |
db94535d | 2192 | u64 disk_bytenr; |
e34a5b4f | 2193 | cond_resched(); |
5f39d397 CM |
2194 | |
2195 | btrfs_item_key_to_cpu(leaf, &key, i); | |
2196 | if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY) | |
6407bf6d CM |
2197 | continue; |
2198 | fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item); | |
5f39d397 CM |
2199 | if (btrfs_file_extent_type(leaf, fi) == |
2200 | BTRFS_FILE_EXTENT_INLINE) | |
236454df | 2201 | continue; |
6407bf6d CM |
2202 | /* |
2203 | * FIXME make sure to insert a trans record that | |
2204 | * repeats the snapshot del on crash | |
2205 | */ | |
db94535d CM |
2206 | disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); |
2207 | if (disk_bytenr == 0) | |
3a686375 | 2208 | continue; |
4a096752 CM |
2209 | |
2210 | mutex_lock(&root->fs_info->alloc_mutex); | |
925baedd | 2211 | ret = __btrfs_free_extent(trans, root, disk_bytenr, |
7bb86316 CM |
2212 | btrfs_file_extent_disk_num_bytes(leaf, fi), |
2213 | leaf_owner, leaf_generation, | |
2214 | key.objectid, key.offset, 0); | |
4a096752 | 2215 | mutex_unlock(&root->fs_info->alloc_mutex); |
6407bf6d CM |
2216 | BUG_ON(ret); |
2217 | } | |
4a096752 CM |
2218 | |
2219 | mutex_lock(&root->fs_info->alloc_mutex); | |
6407bf6d CM |
2220 | return 0; |
2221 | } | |
2222 | ||
98ed5174 | 2223 | static void noinline reada_walk_down(struct btrfs_root *root, |
bea495e5 CM |
2224 | struct extent_buffer *node, |
2225 | int slot) | |
e011599b | 2226 | { |
db94535d | 2227 | u64 bytenr; |
bea495e5 CM |
2228 | u64 last = 0; |
2229 | u32 nritems; | |
e011599b | 2230 | u32 refs; |
db94535d | 2231 | u32 blocksize; |
bea495e5 CM |
2232 | int ret; |
2233 | int i; | |
2234 | int level; | |
2235 | int skipped = 0; | |
e011599b | 2236 | |
5f39d397 | 2237 | nritems = btrfs_header_nritems(node); |
db94535d | 2238 | level = btrfs_header_level(node); |
bea495e5 CM |
2239 | if (level) |
2240 | return; | |
2241 | ||
2242 | for (i = slot; i < nritems && skipped < 32; i++) { | |
db94535d | 2243 | bytenr = btrfs_node_blockptr(node, i); |
bea495e5 CM |
2244 | if (last && ((bytenr > last && bytenr - last > 32 * 1024) || |
2245 | (last > bytenr && last - bytenr > 32 * 1024))) { | |
2246 | skipped++; | |
e011599b | 2247 | continue; |
bea495e5 CM |
2248 | } |
2249 | blocksize = btrfs_level_size(root, level - 1); | |
2250 | if (i != slot) { | |
2251 | ret = lookup_extent_ref(NULL, root, bytenr, | |
2252 | blocksize, &refs); | |
2253 | BUG_ON(ret); | |
2254 | if (refs != 1) { | |
2255 | skipped++; | |
2256 | continue; | |
2257 | } | |
2258 | } | |
ca7a79ad CM |
2259 | ret = readahead_tree_block(root, bytenr, blocksize, |
2260 | btrfs_node_ptr_generation(node, i)); | |
bea495e5 | 2261 | last = bytenr + blocksize; |
409eb95d | 2262 | cond_resched(); |
e011599b CM |
2263 | if (ret) |
2264 | break; | |
2265 | } | |
2266 | } | |
2267 | ||
333db94c CM |
2268 | /* |
2269 | * we want to avoid as much random IO as we can with the alloc mutex | |
2270 | * held, so drop the lock and do the lookup, then do it again with the | |
2271 | * lock held. | |
2272 | */ | |
2273 | int drop_snap_lookup_refcount(struct btrfs_root *root, u64 start, u64 len, | |
2274 | u32 *refs) | |
2275 | { | |
2276 | mutex_unlock(&root->fs_info->alloc_mutex); | |
2277 | lookup_extent_ref(NULL, root, start, len, refs); | |
e7a84565 | 2278 | cond_resched(); |
333db94c CM |
2279 | mutex_lock(&root->fs_info->alloc_mutex); |
2280 | return lookup_extent_ref(NULL, root, start, len, refs); | |
2281 | } | |
2282 | ||
9aca1d51 CM |
2283 | /* |
2284 | * helper function for drop_snapshot, this walks down the tree dropping ref | |
2285 | * counts as it goes. | |
2286 | */ | |
98ed5174 CM |
2287 | static int noinline walk_down_tree(struct btrfs_trans_handle *trans, |
2288 | struct btrfs_root *root, | |
2289 | struct btrfs_path *path, int *level) | |
20524f02 | 2290 | { |
7bb86316 CM |
2291 | u64 root_owner; |
2292 | u64 root_gen; | |
2293 | u64 bytenr; | |
ca7a79ad | 2294 | u64 ptr_gen; |
5f39d397 CM |
2295 | struct extent_buffer *next; |
2296 | struct extent_buffer *cur; | |
7bb86316 | 2297 | struct extent_buffer *parent; |
db94535d | 2298 | u32 blocksize; |
20524f02 CM |
2299 | int ret; |
2300 | u32 refs; | |
2301 | ||
925baedd CM |
2302 | mutex_lock(&root->fs_info->alloc_mutex); |
2303 | ||
5caf2a00 CM |
2304 | WARN_ON(*level < 0); |
2305 | WARN_ON(*level >= BTRFS_MAX_LEVEL); | |
333db94c | 2306 | ret = drop_snap_lookup_refcount(root, path->nodes[*level]->start, |
db94535d | 2307 | path->nodes[*level]->len, &refs); |
20524f02 CM |
2308 | BUG_ON(ret); |
2309 | if (refs > 1) | |
2310 | goto out; | |
e011599b | 2311 | |
9aca1d51 CM |
2312 | /* |
2313 | * walk down to the last node level and free all the leaves | |
2314 | */ | |
6407bf6d | 2315 | while(*level >= 0) { |
5caf2a00 CM |
2316 | WARN_ON(*level < 0); |
2317 | WARN_ON(*level >= BTRFS_MAX_LEVEL); | |
20524f02 | 2318 | cur = path->nodes[*level]; |
e011599b | 2319 | |
5f39d397 | 2320 | if (btrfs_header_level(cur) != *level) |
2c90e5d6 | 2321 | WARN_ON(1); |
e011599b | 2322 | |
7518a238 | 2323 | if (path->slots[*level] >= |
5f39d397 | 2324 | btrfs_header_nritems(cur)) |
20524f02 | 2325 | break; |
6407bf6d CM |
2326 | if (*level == 0) { |
2327 | ret = drop_leaf_ref(trans, root, cur); | |
2328 | BUG_ON(ret); | |
2329 | break; | |
2330 | } | |
db94535d | 2331 | bytenr = btrfs_node_blockptr(cur, path->slots[*level]); |
ca7a79ad | 2332 | ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]); |
db94535d | 2333 | blocksize = btrfs_level_size(root, *level - 1); |
925baedd | 2334 | |
333db94c | 2335 | ret = drop_snap_lookup_refcount(root, bytenr, blocksize, &refs); |
6407bf6d CM |
2336 | BUG_ON(ret); |
2337 | if (refs != 1) { | |
7bb86316 CM |
2338 | parent = path->nodes[*level]; |
2339 | root_owner = btrfs_header_owner(parent); | |
2340 | root_gen = btrfs_header_generation(parent); | |
20524f02 | 2341 | path->slots[*level]++; |
925baedd | 2342 | ret = __btrfs_free_extent(trans, root, bytenr, |
7bb86316 CM |
2343 | blocksize, root_owner, |
2344 | root_gen, 0, 0, 1); | |
20524f02 CM |
2345 | BUG_ON(ret); |
2346 | continue; | |
2347 | } | |
db94535d | 2348 | next = btrfs_find_tree_block(root, bytenr, blocksize); |
1259ab75 | 2349 | if (!next || !btrfs_buffer_uptodate(next, ptr_gen)) { |
5f39d397 | 2350 | free_extent_buffer(next); |
333db94c CM |
2351 | mutex_unlock(&root->fs_info->alloc_mutex); |
2352 | ||
a74a4b97 CM |
2353 | if (path->slots[*level] == 0) |
2354 | reada_walk_down(root, cur, path->slots[*level]); | |
8790d502 | 2355 | |
ca7a79ad CM |
2356 | next = read_tree_block(root, bytenr, blocksize, |
2357 | ptr_gen); | |
e7a84565 | 2358 | cond_resched(); |
925baedd | 2359 | mutex_lock(&root->fs_info->alloc_mutex); |
e9d0b13b | 2360 | |
8790d502 | 2361 | /* we've dropped the lock, double check */ |
e6dcd2dc CM |
2362 | ret = lookup_extent_ref(NULL, root, bytenr, blocksize, |
2363 | &refs); | |
e9d0b13b CM |
2364 | BUG_ON(ret); |
2365 | if (refs != 1) { | |
7bb86316 CM |
2366 | parent = path->nodes[*level]; |
2367 | root_owner = btrfs_header_owner(parent); | |
2368 | root_gen = btrfs_header_generation(parent); | |
2369 | ||
e9d0b13b | 2370 | path->slots[*level]++; |
5f39d397 | 2371 | free_extent_buffer(next); |
925baedd | 2372 | ret = __btrfs_free_extent(trans, root, bytenr, |
7bb86316 CM |
2373 | blocksize, |
2374 | root_owner, | |
2375 | root_gen, 0, 0, 1); | |
e9d0b13b CM |
2376 | BUG_ON(ret); |
2377 | continue; | |
2378 | } | |
2379 | } | |
5caf2a00 | 2380 | WARN_ON(*level <= 0); |
83e15a28 | 2381 | if (path->nodes[*level-1]) |
5f39d397 | 2382 | free_extent_buffer(path->nodes[*level-1]); |
20524f02 | 2383 | path->nodes[*level-1] = next; |
5f39d397 | 2384 | *level = btrfs_header_level(next); |
20524f02 CM |
2385 | path->slots[*level] = 0; |
2386 | } | |
2387 | out: | |
5caf2a00 CM |
2388 | WARN_ON(*level < 0); |
2389 | WARN_ON(*level >= BTRFS_MAX_LEVEL); | |
7bb86316 CM |
2390 | |
2391 | if (path->nodes[*level] == root->node) { | |
2392 | root_owner = root->root_key.objectid; | |
2393 | parent = path->nodes[*level]; | |
2394 | } else { | |
2395 | parent = path->nodes[*level + 1]; | |
2396 | root_owner = btrfs_header_owner(parent); | |
2397 | } | |
2398 | ||
2399 | root_gen = btrfs_header_generation(parent); | |
925baedd | 2400 | ret = __btrfs_free_extent(trans, root, path->nodes[*level]->start, |
7bb86316 CM |
2401 | path->nodes[*level]->len, |
2402 | root_owner, root_gen, 0, 0, 1); | |
5f39d397 | 2403 | free_extent_buffer(path->nodes[*level]); |
20524f02 CM |
2404 | path->nodes[*level] = NULL; |
2405 | *level += 1; | |
2406 | BUG_ON(ret); | |
925baedd | 2407 | mutex_unlock(&root->fs_info->alloc_mutex); |
e7a84565 | 2408 | cond_resched(); |
20524f02 CM |
2409 | return 0; |
2410 | } | |
2411 | ||
9aca1d51 CM |
2412 | /* |
2413 | * helper for dropping snapshots. This walks back up the tree in the path | |
2414 | * to find the first node higher up where we haven't yet gone through | |
2415 | * all the slots | |
2416 | */ | |
98ed5174 CM |
2417 | static int noinline walk_up_tree(struct btrfs_trans_handle *trans, |
2418 | struct btrfs_root *root, | |
2419 | struct btrfs_path *path, int *level) | |
20524f02 | 2420 | { |
7bb86316 CM |
2421 | u64 root_owner; |
2422 | u64 root_gen; | |
2423 | struct btrfs_root_item *root_item = &root->root_item; | |
20524f02 CM |
2424 | int i; |
2425 | int slot; | |
2426 | int ret; | |
9f3a7427 | 2427 | |
234b63a0 | 2428 | for(i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) { |
20524f02 | 2429 | slot = path->slots[i]; |
5f39d397 CM |
2430 | if (slot < btrfs_header_nritems(path->nodes[i]) - 1) { |
2431 | struct extent_buffer *node; | |
2432 | struct btrfs_disk_key disk_key; | |
2433 | node = path->nodes[i]; | |
20524f02 CM |
2434 | path->slots[i]++; |
2435 | *level = i; | |
9f3a7427 | 2436 | WARN_ON(*level == 0); |
5f39d397 | 2437 | btrfs_node_key(node, &disk_key, path->slots[i]); |
9f3a7427 | 2438 | memcpy(&root_item->drop_progress, |
5f39d397 | 2439 | &disk_key, sizeof(disk_key)); |
9f3a7427 | 2440 | root_item->drop_level = i; |
20524f02 CM |
2441 | return 0; |
2442 | } else { | |
7bb86316 CM |
2443 | if (path->nodes[*level] == root->node) { |
2444 | root_owner = root->root_key.objectid; | |
2445 | root_gen = | |
2446 | btrfs_header_generation(path->nodes[*level]); | |
2447 | } else { | |
2448 | struct extent_buffer *node; | |
2449 | node = path->nodes[*level + 1]; | |
2450 | root_owner = btrfs_header_owner(node); | |
2451 | root_gen = btrfs_header_generation(node); | |
2452 | } | |
e089f05c | 2453 | ret = btrfs_free_extent(trans, root, |
db94535d | 2454 | path->nodes[*level]->start, |
7bb86316 CM |
2455 | path->nodes[*level]->len, |
2456 | root_owner, root_gen, 0, 0, 1); | |
6407bf6d | 2457 | BUG_ON(ret); |
5f39d397 | 2458 | free_extent_buffer(path->nodes[*level]); |
83e15a28 | 2459 | path->nodes[*level] = NULL; |
20524f02 | 2460 | *level = i + 1; |
20524f02 CM |
2461 | } |
2462 | } | |
2463 | return 1; | |
2464 | } | |
2465 | ||
9aca1d51 CM |
2466 | /* |
2467 | * drop the reference count on the tree rooted at 'snap'. This traverses | |
2468 | * the tree freeing any blocks that have a ref count of zero after being | |
2469 | * decremented. | |
2470 | */ | |
e089f05c | 2471 | int btrfs_drop_snapshot(struct btrfs_trans_handle *trans, struct btrfs_root |
9f3a7427 | 2472 | *root) |
20524f02 | 2473 | { |
3768f368 | 2474 | int ret = 0; |
9aca1d51 | 2475 | int wret; |
20524f02 | 2476 | int level; |
5caf2a00 | 2477 | struct btrfs_path *path; |
20524f02 CM |
2478 | int i; |
2479 | int orig_level; | |
9f3a7427 | 2480 | struct btrfs_root_item *root_item = &root->root_item; |
20524f02 | 2481 | |
a2135011 | 2482 | WARN_ON(!mutex_is_locked(&root->fs_info->drop_mutex)); |
5caf2a00 CM |
2483 | path = btrfs_alloc_path(); |
2484 | BUG_ON(!path); | |
20524f02 | 2485 | |
5f39d397 | 2486 | level = btrfs_header_level(root->node); |
20524f02 | 2487 | orig_level = level; |
9f3a7427 CM |
2488 | if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) { |
2489 | path->nodes[level] = root->node; | |
f510cfec | 2490 | extent_buffer_get(root->node); |
9f3a7427 CM |
2491 | path->slots[level] = 0; |
2492 | } else { | |
2493 | struct btrfs_key key; | |
5f39d397 CM |
2494 | struct btrfs_disk_key found_key; |
2495 | struct extent_buffer *node; | |
6702ed49 | 2496 | |
9f3a7427 | 2497 | btrfs_disk_key_to_cpu(&key, &root_item->drop_progress); |
6702ed49 CM |
2498 | level = root_item->drop_level; |
2499 | path->lowest_level = level; | |
9f3a7427 | 2500 | wret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
6702ed49 | 2501 | if (wret < 0) { |
9f3a7427 CM |
2502 | ret = wret; |
2503 | goto out; | |
2504 | } | |
5f39d397 CM |
2505 | node = path->nodes[level]; |
2506 | btrfs_node_key(node, &found_key, path->slots[level]); | |
2507 | WARN_ON(memcmp(&found_key, &root_item->drop_progress, | |
2508 | sizeof(found_key))); | |
7d9eb12c CM |
2509 | /* |
2510 | * unlock our path, this is safe because only this | |
2511 | * function is allowed to delete this snapshot | |
2512 | */ | |
925baedd CM |
2513 | for (i = 0; i < BTRFS_MAX_LEVEL; i++) { |
2514 | if (path->nodes[i] && path->locks[i]) { | |
2515 | path->locks[i] = 0; | |
2516 | btrfs_tree_unlock(path->nodes[i]); | |
2517 | } | |
2518 | } | |
9f3a7427 | 2519 | } |
20524f02 | 2520 | while(1) { |
5caf2a00 | 2521 | wret = walk_down_tree(trans, root, path, &level); |
9aca1d51 | 2522 | if (wret > 0) |
20524f02 | 2523 | break; |
9aca1d51 CM |
2524 | if (wret < 0) |
2525 | ret = wret; | |
2526 | ||
5caf2a00 | 2527 | wret = walk_up_tree(trans, root, path, &level); |
9aca1d51 | 2528 | if (wret > 0) |
20524f02 | 2529 | break; |
9aca1d51 CM |
2530 | if (wret < 0) |
2531 | ret = wret; | |
e7a84565 CM |
2532 | if (trans->transaction->in_commit) { |
2533 | ret = -EAGAIN; | |
2534 | break; | |
2535 | } | |
20524f02 | 2536 | } |
83e15a28 | 2537 | for (i = 0; i <= orig_level; i++) { |
5caf2a00 | 2538 | if (path->nodes[i]) { |
5f39d397 | 2539 | free_extent_buffer(path->nodes[i]); |
0f82731f | 2540 | path->nodes[i] = NULL; |
83e15a28 | 2541 | } |
20524f02 | 2542 | } |
9f3a7427 | 2543 | out: |
5caf2a00 | 2544 | btrfs_free_path(path); |
9aca1d51 | 2545 | return ret; |
20524f02 | 2546 | } |
9078a3e1 | 2547 | |
96b5179d | 2548 | int btrfs_free_block_groups(struct btrfs_fs_info *info) |
9078a3e1 | 2549 | { |
96b5179d CM |
2550 | u64 start; |
2551 | u64 end; | |
b97f9203 | 2552 | u64 ptr; |
9078a3e1 | 2553 | int ret; |
925baedd CM |
2554 | |
2555 | mutex_lock(&info->alloc_mutex); | |
9078a3e1 | 2556 | while(1) { |
96b5179d CM |
2557 | ret = find_first_extent_bit(&info->block_group_cache, 0, |
2558 | &start, &end, (unsigned int)-1); | |
2559 | if (ret) | |
9078a3e1 | 2560 | break; |
b97f9203 Y |
2561 | ret = get_state_private(&info->block_group_cache, start, &ptr); |
2562 | if (!ret) | |
2563 | kfree((void *)(unsigned long)ptr); | |
96b5179d CM |
2564 | clear_extent_bits(&info->block_group_cache, start, |
2565 | end, (unsigned int)-1, GFP_NOFS); | |
9078a3e1 | 2566 | } |
e37c9e69 | 2567 | while(1) { |
f510cfec CM |
2568 | ret = find_first_extent_bit(&info->free_space_cache, 0, |
2569 | &start, &end, EXTENT_DIRTY); | |
2570 | if (ret) | |
e37c9e69 | 2571 | break; |
f510cfec CM |
2572 | clear_extent_dirty(&info->free_space_cache, start, |
2573 | end, GFP_NOFS); | |
e37c9e69 | 2574 | } |
925baedd | 2575 | mutex_unlock(&info->alloc_mutex); |
be744175 CM |
2576 | return 0; |
2577 | } | |
2578 | ||
8e7bf94f CM |
2579 | static unsigned long calc_ra(unsigned long start, unsigned long last, |
2580 | unsigned long nr) | |
2581 | { | |
2582 | return min(last, start + nr - 1); | |
2583 | } | |
2584 | ||
98ed5174 CM |
2585 | static int noinline relocate_inode_pages(struct inode *inode, u64 start, |
2586 | u64 len) | |
edbd8d4e CM |
2587 | { |
2588 | u64 page_start; | |
2589 | u64 page_end; | |
edbd8d4e | 2590 | unsigned long last_index; |
edbd8d4e CM |
2591 | unsigned long i; |
2592 | struct page *page; | |
d1310b2e | 2593 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
4313b399 | 2594 | struct file_ra_state *ra; |
8e7bf94f CM |
2595 | unsigned long total_read = 0; |
2596 | unsigned long ra_pages; | |
a061fc8d | 2597 | struct btrfs_trans_handle *trans; |
4313b399 CM |
2598 | |
2599 | ra = kzalloc(sizeof(*ra), GFP_NOFS); | |
edbd8d4e CM |
2600 | |
2601 | mutex_lock(&inode->i_mutex); | |
4313b399 | 2602 | i = start >> PAGE_CACHE_SHIFT; |
edbd8d4e CM |
2603 | last_index = (start + len - 1) >> PAGE_CACHE_SHIFT; |
2604 | ||
8e7bf94f CM |
2605 | ra_pages = BTRFS_I(inode)->root->fs_info->bdi.ra_pages; |
2606 | ||
4313b399 | 2607 | file_ra_state_init(ra, inode->i_mapping); |
edbd8d4e | 2608 | |
4313b399 | 2609 | for (; i <= last_index; i++) { |
8e7bf94f CM |
2610 | if (total_read % ra_pages == 0) { |
2611 | btrfs_force_ra(inode->i_mapping, ra, NULL, i, | |
2612 | calc_ra(i, last_index, ra_pages)); | |
2613 | } | |
2614 | total_read++; | |
a061fc8d CM |
2615 | if (((u64)i << PAGE_CACHE_SHIFT) > inode->i_size) |
2616 | goto truncate_racing; | |
2617 | ||
edbd8d4e | 2618 | page = grab_cache_page(inode->i_mapping, i); |
a061fc8d | 2619 | if (!page) { |
edbd8d4e | 2620 | goto out_unlock; |
a061fc8d | 2621 | } |
edbd8d4e CM |
2622 | if (!PageUptodate(page)) { |
2623 | btrfs_readpage(NULL, page); | |
2624 | lock_page(page); | |
2625 | if (!PageUptodate(page)) { | |
2626 | unlock_page(page); | |
2627 | page_cache_release(page); | |
2628 | goto out_unlock; | |
2629 | } | |
2630 | } | |
ec44a35c CM |
2631 | #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18) |
2632 | ClearPageDirty(page); | |
2633 | #else | |
2634 | cancel_dirty_page(page, PAGE_CACHE_SIZE); | |
2635 | #endif | |
2636 | wait_on_page_writeback(page); | |
2637 | set_page_extent_mapped(page); | |
edbd8d4e CM |
2638 | page_start = (u64)page->index << PAGE_CACHE_SHIFT; |
2639 | page_end = page_start + PAGE_CACHE_SIZE - 1; | |
2640 | ||
d1310b2e | 2641 | lock_extent(io_tree, page_start, page_end, GFP_NOFS); |
edbd8d4e | 2642 | |
d1310b2e | 2643 | set_extent_delalloc(io_tree, page_start, |
edbd8d4e | 2644 | page_end, GFP_NOFS); |
a061fc8d | 2645 | set_page_dirty(page); |
edbd8d4e | 2646 | |
d1310b2e | 2647 | unlock_extent(io_tree, page_start, page_end, GFP_NOFS); |
edbd8d4e CM |
2648 | unlock_page(page); |
2649 | page_cache_release(page); | |
2650 | } | |
a061fc8d CM |
2651 | balance_dirty_pages_ratelimited_nr(inode->i_mapping, |
2652 | total_read); | |
edbd8d4e CM |
2653 | |
2654 | out_unlock: | |
ec44a35c | 2655 | kfree(ra); |
a061fc8d CM |
2656 | trans = btrfs_start_transaction(BTRFS_I(inode)->root, 1); |
2657 | if (trans) { | |
a061fc8d CM |
2658 | btrfs_end_transaction(trans, BTRFS_I(inode)->root); |
2659 | mark_inode_dirty(inode); | |
2660 | } | |
edbd8d4e CM |
2661 | mutex_unlock(&inode->i_mutex); |
2662 | return 0; | |
a061fc8d CM |
2663 | |
2664 | truncate_racing: | |
2665 | vmtruncate(inode, inode->i_size); | |
2666 | balance_dirty_pages_ratelimited_nr(inode->i_mapping, | |
2667 | total_read); | |
2668 | goto out_unlock; | |
edbd8d4e CM |
2669 | } |
2670 | ||
bf4ef679 CM |
2671 | /* |
2672 | * The back references tell us which tree holds a ref on a block, | |
2673 | * but it is possible for the tree root field in the reference to | |
2674 | * reflect the original root before a snapshot was made. In this | |
2675 | * case we should search through all the children of a given root | |
2676 | * to find potential holders of references on a block. | |
2677 | * | |
2678 | * Instead, we do something a little less fancy and just search | |
2679 | * all the roots for a given key/block combination. | |
2680 | */ | |
2681 | static int find_root_for_ref(struct btrfs_root *root, | |
2682 | struct btrfs_path *path, | |
2683 | struct btrfs_key *key0, | |
2684 | int level, | |
2685 | int file_key, | |
2686 | struct btrfs_root **found_root, | |
2687 | u64 bytenr) | |
2688 | { | |
2689 | struct btrfs_key root_location; | |
2690 | struct btrfs_root *cur_root = *found_root; | |
2691 | struct btrfs_file_extent_item *file_extent; | |
2692 | u64 root_search_start = BTRFS_FS_TREE_OBJECTID; | |
2693 | u64 found_bytenr; | |
2694 | int ret; | |
bf4ef679 CM |
2695 | |
2696 | root_location.offset = (u64)-1; | |
2697 | root_location.type = BTRFS_ROOT_ITEM_KEY; | |
2698 | path->lowest_level = level; | |
2699 | path->reada = 0; | |
2700 | while(1) { | |
2701 | ret = btrfs_search_slot(NULL, cur_root, key0, path, 0, 0); | |
2702 | found_bytenr = 0; | |
2703 | if (ret == 0 && file_key) { | |
2704 | struct extent_buffer *leaf = path->nodes[0]; | |
2705 | file_extent = btrfs_item_ptr(leaf, path->slots[0], | |
2706 | struct btrfs_file_extent_item); | |
2707 | if (btrfs_file_extent_type(leaf, file_extent) == | |
2708 | BTRFS_FILE_EXTENT_REG) { | |
2709 | found_bytenr = | |
2710 | btrfs_file_extent_disk_bytenr(leaf, | |
2711 | file_extent); | |
2712 | } | |
323da79c | 2713 | } else if (!file_key) { |
bf4ef679 CM |
2714 | if (path->nodes[level]) |
2715 | found_bytenr = path->nodes[level]->start; | |
2716 | } | |
2717 | ||
bf4ef679 CM |
2718 | btrfs_release_path(cur_root, path); |
2719 | ||
2720 | if (found_bytenr == bytenr) { | |
2721 | *found_root = cur_root; | |
2722 | ret = 0; | |
2723 | goto out; | |
2724 | } | |
2725 | ret = btrfs_search_root(root->fs_info->tree_root, | |
2726 | root_search_start, &root_search_start); | |
2727 | if (ret) | |
2728 | break; | |
2729 | ||
2730 | root_location.objectid = root_search_start; | |
2731 | cur_root = btrfs_read_fs_root_no_name(root->fs_info, | |
2732 | &root_location); | |
2733 | if (!cur_root) { | |
2734 | ret = 1; | |
2735 | break; | |
2736 | } | |
2737 | } | |
2738 | out: | |
2739 | path->lowest_level = 0; | |
2740 | return ret; | |
2741 | } | |
2742 | ||
4313b399 CM |
2743 | /* |
2744 | * note, this releases the path | |
2745 | */ | |
98ed5174 | 2746 | static int noinline relocate_one_reference(struct btrfs_root *extent_root, |
edbd8d4e | 2747 | struct btrfs_path *path, |
0ef3e66b CM |
2748 | struct btrfs_key *extent_key, |
2749 | u64 *last_file_objectid, | |
2750 | u64 *last_file_offset, | |
2751 | u64 *last_file_root, | |
2752 | u64 last_extent) | |
edbd8d4e CM |
2753 | { |
2754 | struct inode *inode; | |
2755 | struct btrfs_root *found_root; | |
bf4ef679 CM |
2756 | struct btrfs_key root_location; |
2757 | struct btrfs_key found_key; | |
4313b399 CM |
2758 | struct btrfs_extent_ref *ref; |
2759 | u64 ref_root; | |
2760 | u64 ref_gen; | |
2761 | u64 ref_objectid; | |
2762 | u64 ref_offset; | |
edbd8d4e | 2763 | int ret; |
bf4ef679 | 2764 | int level; |
edbd8d4e | 2765 | |
7d9eb12c CM |
2766 | WARN_ON(!mutex_is_locked(&extent_root->fs_info->alloc_mutex)); |
2767 | ||
4313b399 CM |
2768 | ref = btrfs_item_ptr(path->nodes[0], path->slots[0], |
2769 | struct btrfs_extent_ref); | |
2770 | ref_root = btrfs_ref_root(path->nodes[0], ref); | |
2771 | ref_gen = btrfs_ref_generation(path->nodes[0], ref); | |
2772 | ref_objectid = btrfs_ref_objectid(path->nodes[0], ref); | |
2773 | ref_offset = btrfs_ref_offset(path->nodes[0], ref); | |
2774 | btrfs_release_path(extent_root, path); | |
2775 | ||
bf4ef679 | 2776 | root_location.objectid = ref_root; |
edbd8d4e | 2777 | if (ref_gen == 0) |
bf4ef679 | 2778 | root_location.offset = 0; |
edbd8d4e | 2779 | else |
bf4ef679 CM |
2780 | root_location.offset = (u64)-1; |
2781 | root_location.type = BTRFS_ROOT_ITEM_KEY; | |
edbd8d4e CM |
2782 | |
2783 | found_root = btrfs_read_fs_root_no_name(extent_root->fs_info, | |
bf4ef679 | 2784 | &root_location); |
edbd8d4e | 2785 | BUG_ON(!found_root); |
7d9eb12c | 2786 | mutex_unlock(&extent_root->fs_info->alloc_mutex); |
edbd8d4e CM |
2787 | |
2788 | if (ref_objectid >= BTRFS_FIRST_FREE_OBJECTID) { | |
bf4ef679 CM |
2789 | found_key.objectid = ref_objectid; |
2790 | found_key.type = BTRFS_EXTENT_DATA_KEY; | |
2791 | found_key.offset = ref_offset; | |
2792 | level = 0; | |
2793 | ||
0ef3e66b CM |
2794 | if (last_extent == extent_key->objectid && |
2795 | *last_file_objectid == ref_objectid && | |
2796 | *last_file_offset == ref_offset && | |
2797 | *last_file_root == ref_root) | |
2798 | goto out; | |
2799 | ||
bf4ef679 CM |
2800 | ret = find_root_for_ref(extent_root, path, &found_key, |
2801 | level, 1, &found_root, | |
2802 | extent_key->objectid); | |
2803 | ||
2804 | if (ret) | |
2805 | goto out; | |
2806 | ||
0ef3e66b CM |
2807 | if (last_extent == extent_key->objectid && |
2808 | *last_file_objectid == ref_objectid && | |
2809 | *last_file_offset == ref_offset && | |
2810 | *last_file_root == ref_root) | |
2811 | goto out; | |
2812 | ||
edbd8d4e CM |
2813 | inode = btrfs_iget_locked(extent_root->fs_info->sb, |
2814 | ref_objectid, found_root); | |
2815 | if (inode->i_state & I_NEW) { | |
2816 | /* the inode and parent dir are two different roots */ | |
2817 | BTRFS_I(inode)->root = found_root; | |
2818 | BTRFS_I(inode)->location.objectid = ref_objectid; | |
2819 | BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY; | |
2820 | BTRFS_I(inode)->location.offset = 0; | |
2821 | btrfs_read_locked_inode(inode); | |
2822 | unlock_new_inode(inode); | |
2823 | ||
2824 | } | |
2825 | /* this can happen if the reference is not against | |
2826 | * the latest version of the tree root | |
2827 | */ | |
7d9eb12c | 2828 | if (is_bad_inode(inode)) |
edbd8d4e | 2829 | goto out; |
7d9eb12c | 2830 | |
0ef3e66b CM |
2831 | *last_file_objectid = inode->i_ino; |
2832 | *last_file_root = found_root->root_key.objectid; | |
2833 | *last_file_offset = ref_offset; | |
2834 | ||
edbd8d4e | 2835 | relocate_inode_pages(inode, ref_offset, extent_key->offset); |
edbd8d4e | 2836 | iput(inode); |
edbd8d4e CM |
2837 | } else { |
2838 | struct btrfs_trans_handle *trans; | |
edbd8d4e | 2839 | struct extent_buffer *eb; |
7d9eb12c | 2840 | int needs_lock = 0; |
edbd8d4e | 2841 | |
edbd8d4e | 2842 | eb = read_tree_block(found_root, extent_key->objectid, |
ca7a79ad | 2843 | extent_key->offset, 0); |
925baedd | 2844 | btrfs_tree_lock(eb); |
edbd8d4e CM |
2845 | level = btrfs_header_level(eb); |
2846 | ||
2847 | if (level == 0) | |
2848 | btrfs_item_key_to_cpu(eb, &found_key, 0); | |
2849 | else | |
2850 | btrfs_node_key_to_cpu(eb, &found_key, 0); | |
2851 | ||
925baedd | 2852 | btrfs_tree_unlock(eb); |
edbd8d4e CM |
2853 | free_extent_buffer(eb); |
2854 | ||
bf4ef679 CM |
2855 | ret = find_root_for_ref(extent_root, path, &found_key, |
2856 | level, 0, &found_root, | |
2857 | extent_key->objectid); | |
2858 | ||
2859 | if (ret) | |
2860 | goto out; | |
2861 | ||
7d9eb12c CM |
2862 | /* |
2863 | * right here almost anything could happen to our key, | |
2864 | * but that's ok. The cow below will either relocate it | |
2865 | * or someone else will have relocated it. Either way, | |
2866 | * it is in a different spot than it was before and | |
2867 | * we're happy. | |
2868 | */ | |
2869 | ||
bf4ef679 CM |
2870 | trans = btrfs_start_transaction(found_root, 1); |
2871 | ||
7d9eb12c CM |
2872 | if (found_root == extent_root->fs_info->extent_root || |
2873 | found_root == extent_root->fs_info->chunk_root || | |
2874 | found_root == extent_root->fs_info->dev_root) { | |
2875 | needs_lock = 1; | |
2876 | mutex_lock(&extent_root->fs_info->alloc_mutex); | |
2877 | } | |
2878 | ||
edbd8d4e | 2879 | path->lowest_level = level; |
8f662a76 | 2880 | path->reada = 2; |
edbd8d4e CM |
2881 | ret = btrfs_search_slot(trans, found_root, &found_key, path, |
2882 | 0, 1); | |
2883 | path->lowest_level = 0; | |
edbd8d4e | 2884 | btrfs_release_path(found_root, path); |
7d9eb12c | 2885 | |
0ef3e66b CM |
2886 | if (found_root == found_root->fs_info->extent_root) |
2887 | btrfs_extent_post_op(trans, found_root); | |
7d9eb12c CM |
2888 | if (needs_lock) |
2889 | mutex_unlock(&extent_root->fs_info->alloc_mutex); | |
2890 | ||
edbd8d4e | 2891 | btrfs_end_transaction(trans, found_root); |
edbd8d4e | 2892 | |
7d9eb12c | 2893 | } |
edbd8d4e | 2894 | out: |
7d9eb12c | 2895 | mutex_lock(&extent_root->fs_info->alloc_mutex); |
edbd8d4e CM |
2896 | return 0; |
2897 | } | |
2898 | ||
a061fc8d CM |
2899 | static int noinline del_extent_zero(struct btrfs_root *extent_root, |
2900 | struct btrfs_path *path, | |
2901 | struct btrfs_key *extent_key) | |
2902 | { | |
2903 | int ret; | |
2904 | struct btrfs_trans_handle *trans; | |
2905 | ||
2906 | trans = btrfs_start_transaction(extent_root, 1); | |
2907 | ret = btrfs_search_slot(trans, extent_root, extent_key, path, -1, 1); | |
2908 | if (ret > 0) { | |
2909 | ret = -EIO; | |
2910 | goto out; | |
2911 | } | |
2912 | if (ret < 0) | |
2913 | goto out; | |
2914 | ret = btrfs_del_item(trans, extent_root, path); | |
2915 | out: | |
2916 | btrfs_end_transaction(trans, extent_root); | |
2917 | return ret; | |
2918 | } | |
2919 | ||
98ed5174 CM |
2920 | static int noinline relocate_one_extent(struct btrfs_root *extent_root, |
2921 | struct btrfs_path *path, | |
2922 | struct btrfs_key *extent_key) | |
edbd8d4e CM |
2923 | { |
2924 | struct btrfs_key key; | |
2925 | struct btrfs_key found_key; | |
edbd8d4e | 2926 | struct extent_buffer *leaf; |
0ef3e66b CM |
2927 | u64 last_file_objectid = 0; |
2928 | u64 last_file_root = 0; | |
2929 | u64 last_file_offset = (u64)-1; | |
2930 | u64 last_extent = 0; | |
edbd8d4e CM |
2931 | u32 nritems; |
2932 | u32 item_size; | |
2933 | int ret = 0; | |
2934 | ||
a061fc8d CM |
2935 | if (extent_key->objectid == 0) { |
2936 | ret = del_extent_zero(extent_root, path, extent_key); | |
2937 | goto out; | |
2938 | } | |
edbd8d4e CM |
2939 | key.objectid = extent_key->objectid; |
2940 | key.type = BTRFS_EXTENT_REF_KEY; | |
2941 | key.offset = 0; | |
2942 | ||
2943 | while(1) { | |
2944 | ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0); | |
2945 | ||
edbd8d4e CM |
2946 | if (ret < 0) |
2947 | goto out; | |
2948 | ||
2949 | ret = 0; | |
2950 | leaf = path->nodes[0]; | |
2951 | nritems = btrfs_header_nritems(leaf); | |
a061fc8d CM |
2952 | if (path->slots[0] == nritems) { |
2953 | ret = btrfs_next_leaf(extent_root, path); | |
2954 | if (ret > 0) { | |
2955 | ret = 0; | |
2956 | goto out; | |
2957 | } | |
2958 | if (ret < 0) | |
2959 | goto out; | |
bf4ef679 | 2960 | leaf = path->nodes[0]; |
a061fc8d | 2961 | } |
edbd8d4e CM |
2962 | |
2963 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
a061fc8d | 2964 | if (found_key.objectid != extent_key->objectid) { |
edbd8d4e | 2965 | break; |
a061fc8d | 2966 | } |
edbd8d4e | 2967 | |
a061fc8d | 2968 | if (found_key.type != BTRFS_EXTENT_REF_KEY) { |
edbd8d4e | 2969 | break; |
a061fc8d | 2970 | } |
edbd8d4e CM |
2971 | |
2972 | key.offset = found_key.offset + 1; | |
2973 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
2974 | ||
0ef3e66b CM |
2975 | ret = relocate_one_reference(extent_root, path, extent_key, |
2976 | &last_file_objectid, | |
2977 | &last_file_offset, | |
2978 | &last_file_root, last_extent); | |
edbd8d4e CM |
2979 | if (ret) |
2980 | goto out; | |
0ef3e66b | 2981 | last_extent = extent_key->objectid; |
edbd8d4e CM |
2982 | } |
2983 | ret = 0; | |
2984 | out: | |
2985 | btrfs_release_path(extent_root, path); | |
2986 | return ret; | |
2987 | } | |
2988 | ||
ec44a35c CM |
2989 | static u64 update_block_group_flags(struct btrfs_root *root, u64 flags) |
2990 | { | |
2991 | u64 num_devices; | |
2992 | u64 stripped = BTRFS_BLOCK_GROUP_RAID0 | | |
2993 | BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10; | |
2994 | ||
a061fc8d | 2995 | num_devices = root->fs_info->fs_devices->num_devices; |
ec44a35c CM |
2996 | if (num_devices == 1) { |
2997 | stripped |= BTRFS_BLOCK_GROUP_DUP; | |
2998 | stripped = flags & ~stripped; | |
2999 | ||
3000 | /* turn raid0 into single device chunks */ | |
3001 | if (flags & BTRFS_BLOCK_GROUP_RAID0) | |
3002 | return stripped; | |
3003 | ||
3004 | /* turn mirroring into duplication */ | |
3005 | if (flags & (BTRFS_BLOCK_GROUP_RAID1 | | |
3006 | BTRFS_BLOCK_GROUP_RAID10)) | |
3007 | return stripped | BTRFS_BLOCK_GROUP_DUP; | |
3008 | return flags; | |
3009 | } else { | |
3010 | /* they already had raid on here, just return */ | |
ec44a35c CM |
3011 | if (flags & stripped) |
3012 | return flags; | |
3013 | ||
3014 | stripped |= BTRFS_BLOCK_GROUP_DUP; | |
3015 | stripped = flags & ~stripped; | |
3016 | ||
3017 | /* switch duplicated blocks with raid1 */ | |
3018 | if (flags & BTRFS_BLOCK_GROUP_DUP) | |
3019 | return stripped | BTRFS_BLOCK_GROUP_RAID1; | |
3020 | ||
3021 | /* turn single device chunks into raid0 */ | |
3022 | return stripped | BTRFS_BLOCK_GROUP_RAID0; | |
3023 | } | |
3024 | return flags; | |
3025 | } | |
3026 | ||
0ef3e66b CM |
3027 | int __alloc_chunk_for_shrink(struct btrfs_root *root, |
3028 | struct btrfs_block_group_cache *shrink_block_group, | |
3029 | int force) | |
3030 | { | |
3031 | struct btrfs_trans_handle *trans; | |
3032 | u64 new_alloc_flags; | |
3033 | u64 calc; | |
3034 | ||
3035 | if (btrfs_block_group_used(&shrink_block_group->item) > 0) { | |
3036 | ||
7d9eb12c | 3037 | mutex_unlock(&root->fs_info->alloc_mutex); |
0ef3e66b | 3038 | trans = btrfs_start_transaction(root, 1); |
7d9eb12c CM |
3039 | mutex_lock(&root->fs_info->alloc_mutex); |
3040 | ||
0ef3e66b CM |
3041 | new_alloc_flags = update_block_group_flags(root, |
3042 | shrink_block_group->flags); | |
3043 | if (new_alloc_flags != shrink_block_group->flags) { | |
3044 | calc = | |
3045 | btrfs_block_group_used(&shrink_block_group->item); | |
3046 | } else { | |
3047 | calc = shrink_block_group->key.offset; | |
3048 | } | |
3049 | do_chunk_alloc(trans, root->fs_info->extent_root, | |
3050 | calc + 2 * 1024 * 1024, new_alloc_flags, force); | |
7d9eb12c CM |
3051 | |
3052 | mutex_unlock(&root->fs_info->alloc_mutex); | |
0ef3e66b | 3053 | btrfs_end_transaction(trans, root); |
7d9eb12c | 3054 | mutex_lock(&root->fs_info->alloc_mutex); |
0ef3e66b CM |
3055 | } |
3056 | return 0; | |
3057 | } | |
3058 | ||
8f18cf13 | 3059 | int btrfs_shrink_extent_tree(struct btrfs_root *root, u64 shrink_start) |
edbd8d4e CM |
3060 | { |
3061 | struct btrfs_trans_handle *trans; | |
3062 | struct btrfs_root *tree_root = root->fs_info->tree_root; | |
3063 | struct btrfs_path *path; | |
3064 | u64 cur_byte; | |
3065 | u64 total_found; | |
8f18cf13 CM |
3066 | u64 shrink_last_byte; |
3067 | struct btrfs_block_group_cache *shrink_block_group; | |
edbd8d4e | 3068 | struct btrfs_fs_info *info = root->fs_info; |
edbd8d4e | 3069 | struct btrfs_key key; |
73e48b27 | 3070 | struct btrfs_key found_key; |
edbd8d4e CM |
3071 | struct extent_buffer *leaf; |
3072 | u32 nritems; | |
3073 | int ret; | |
a061fc8d | 3074 | int progress; |
edbd8d4e | 3075 | |
925baedd | 3076 | mutex_lock(&root->fs_info->alloc_mutex); |
8f18cf13 CM |
3077 | shrink_block_group = btrfs_lookup_block_group(root->fs_info, |
3078 | shrink_start); | |
3079 | BUG_ON(!shrink_block_group); | |
3080 | ||
0ef3e66b CM |
3081 | shrink_last_byte = shrink_block_group->key.objectid + |
3082 | shrink_block_group->key.offset; | |
8f18cf13 CM |
3083 | |
3084 | shrink_block_group->space_info->total_bytes -= | |
3085 | shrink_block_group->key.offset; | |
edbd8d4e CM |
3086 | path = btrfs_alloc_path(); |
3087 | root = root->fs_info->extent_root; | |
8f662a76 | 3088 | path->reada = 2; |
edbd8d4e | 3089 | |
323da79c CM |
3090 | printk("btrfs relocating block group %llu flags %llu\n", |
3091 | (unsigned long long)shrink_start, | |
3092 | (unsigned long long)shrink_block_group->flags); | |
3093 | ||
0ef3e66b CM |
3094 | __alloc_chunk_for_shrink(root, shrink_block_group, 1); |
3095 | ||
edbd8d4e | 3096 | again: |
323da79c | 3097 | |
8f18cf13 CM |
3098 | shrink_block_group->ro = 1; |
3099 | ||
edbd8d4e | 3100 | total_found = 0; |
a061fc8d | 3101 | progress = 0; |
8f18cf13 | 3102 | key.objectid = shrink_start; |
edbd8d4e CM |
3103 | key.offset = 0; |
3104 | key.type = 0; | |
73e48b27 | 3105 | cur_byte = key.objectid; |
4313b399 | 3106 | |
73e48b27 Y |
3107 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
3108 | if (ret < 0) | |
3109 | goto out; | |
3110 | ||
0b86a832 | 3111 | ret = btrfs_previous_item(root, path, 0, BTRFS_EXTENT_ITEM_KEY); |
73e48b27 Y |
3112 | if (ret < 0) |
3113 | goto out; | |
8f18cf13 | 3114 | |
73e48b27 Y |
3115 | if (ret == 0) { |
3116 | leaf = path->nodes[0]; | |
3117 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
8f18cf13 CM |
3118 | if (found_key.objectid + found_key.offset > shrink_start && |
3119 | found_key.objectid < shrink_last_byte) { | |
73e48b27 Y |
3120 | cur_byte = found_key.objectid; |
3121 | key.objectid = cur_byte; | |
3122 | } | |
3123 | } | |
3124 | btrfs_release_path(root, path); | |
3125 | ||
3126 | while(1) { | |
edbd8d4e CM |
3127 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
3128 | if (ret < 0) | |
3129 | goto out; | |
73e48b27 | 3130 | |
7d9eb12c | 3131 | next: |
edbd8d4e | 3132 | leaf = path->nodes[0]; |
73e48b27 | 3133 | nritems = btrfs_header_nritems(leaf); |
73e48b27 Y |
3134 | if (path->slots[0] >= nritems) { |
3135 | ret = btrfs_next_leaf(root, path); | |
3136 | if (ret < 0) | |
3137 | goto out; | |
3138 | if (ret == 1) { | |
3139 | ret = 0; | |
3140 | break; | |
edbd8d4e | 3141 | } |
73e48b27 Y |
3142 | leaf = path->nodes[0]; |
3143 | nritems = btrfs_header_nritems(leaf); | |
edbd8d4e | 3144 | } |
73e48b27 Y |
3145 | |
3146 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
725c8463 | 3147 | |
8f18cf13 CM |
3148 | if (found_key.objectid >= shrink_last_byte) |
3149 | break; | |
3150 | ||
725c8463 CM |
3151 | if (progress && need_resched()) { |
3152 | memcpy(&key, &found_key, sizeof(key)); | |
725c8463 | 3153 | cond_resched(); |
725c8463 CM |
3154 | btrfs_release_path(root, path); |
3155 | btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
3156 | progress = 0; | |
3157 | goto next; | |
3158 | } | |
3159 | progress = 1; | |
3160 | ||
73e48b27 Y |
3161 | if (btrfs_key_type(&found_key) != BTRFS_EXTENT_ITEM_KEY || |
3162 | found_key.objectid + found_key.offset <= cur_byte) { | |
0ef3e66b CM |
3163 | memcpy(&key, &found_key, sizeof(key)); |
3164 | key.offset++; | |
edbd8d4e | 3165 | path->slots[0]++; |
edbd8d4e CM |
3166 | goto next; |
3167 | } | |
73e48b27 | 3168 | |
edbd8d4e CM |
3169 | total_found++; |
3170 | cur_byte = found_key.objectid + found_key.offset; | |
3171 | key.objectid = cur_byte; | |
3172 | btrfs_release_path(root, path); | |
3173 | ret = relocate_one_extent(root, path, &found_key); | |
0ef3e66b | 3174 | __alloc_chunk_for_shrink(root, shrink_block_group, 0); |
edbd8d4e CM |
3175 | } |
3176 | ||
3177 | btrfs_release_path(root, path); | |
3178 | ||
3179 | if (total_found > 0) { | |
323da79c CM |
3180 | printk("btrfs relocate found %llu last extent was %llu\n", |
3181 | (unsigned long long)total_found, | |
3182 | (unsigned long long)found_key.objectid); | |
7d9eb12c | 3183 | mutex_unlock(&root->fs_info->alloc_mutex); |
edbd8d4e CM |
3184 | trans = btrfs_start_transaction(tree_root, 1); |
3185 | btrfs_commit_transaction(trans, tree_root); | |
3186 | ||
edbd8d4e | 3187 | btrfs_clean_old_snapshots(tree_root); |
edbd8d4e CM |
3188 | |
3189 | trans = btrfs_start_transaction(tree_root, 1); | |
3190 | btrfs_commit_transaction(trans, tree_root); | |
7d9eb12c | 3191 | mutex_lock(&root->fs_info->alloc_mutex); |
edbd8d4e CM |
3192 | goto again; |
3193 | } | |
3194 | ||
8f18cf13 CM |
3195 | /* |
3196 | * we've freed all the extents, now remove the block | |
3197 | * group item from the tree | |
3198 | */ | |
7d9eb12c CM |
3199 | mutex_unlock(&root->fs_info->alloc_mutex); |
3200 | ||
edbd8d4e | 3201 | trans = btrfs_start_transaction(root, 1); |
7d9eb12c | 3202 | mutex_lock(&root->fs_info->alloc_mutex); |
8f18cf13 | 3203 | memcpy(&key, &shrink_block_group->key, sizeof(key)); |
1372f8e6 | 3204 | |
8f18cf13 CM |
3205 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
3206 | if (ret > 0) | |
3207 | ret = -EIO; | |
3208 | if (ret < 0) | |
3209 | goto out; | |
73e48b27 | 3210 | |
0ef3e66b CM |
3211 | clear_extent_bits(&info->block_group_cache, key.objectid, |
3212 | key.objectid + key.offset - 1, | |
8f18cf13 | 3213 | (unsigned int)-1, GFP_NOFS); |
edbd8d4e | 3214 | |
0ef3e66b CM |
3215 | |
3216 | clear_extent_bits(&info->free_space_cache, | |
3217 | key.objectid, key.objectid + key.offset - 1, | |
3218 | (unsigned int)-1, GFP_NOFS); | |
3219 | ||
3220 | memset(shrink_block_group, 0, sizeof(*shrink_block_group)); | |
3221 | kfree(shrink_block_group); | |
3222 | ||
8f18cf13 | 3223 | btrfs_del_item(trans, root, path); |
7d9eb12c CM |
3224 | btrfs_release_path(root, path); |
3225 | mutex_unlock(&root->fs_info->alloc_mutex); | |
edbd8d4e | 3226 | btrfs_commit_transaction(trans, root); |
0ef3e66b | 3227 | |
7d9eb12c CM |
3228 | mutex_lock(&root->fs_info->alloc_mutex); |
3229 | ||
0ef3e66b CM |
3230 | /* the code to unpin extents might set a few bits in the free |
3231 | * space cache for this range again | |
3232 | */ | |
3233 | clear_extent_bits(&info->free_space_cache, | |
3234 | key.objectid, key.objectid + key.offset - 1, | |
3235 | (unsigned int)-1, GFP_NOFS); | |
edbd8d4e CM |
3236 | out: |
3237 | btrfs_free_path(path); | |
925baedd | 3238 | mutex_unlock(&root->fs_info->alloc_mutex); |
edbd8d4e CM |
3239 | return ret; |
3240 | } | |
3241 | ||
0b86a832 CM |
3242 | int find_first_block_group(struct btrfs_root *root, struct btrfs_path *path, |
3243 | struct btrfs_key *key) | |
3244 | { | |
925baedd | 3245 | int ret = 0; |
0b86a832 CM |
3246 | struct btrfs_key found_key; |
3247 | struct extent_buffer *leaf; | |
3248 | int slot; | |
edbd8d4e | 3249 | |
0b86a832 CM |
3250 | ret = btrfs_search_slot(NULL, root, key, path, 0, 0); |
3251 | if (ret < 0) | |
925baedd CM |
3252 | goto out; |
3253 | ||
0b86a832 CM |
3254 | while(1) { |
3255 | slot = path->slots[0]; | |
edbd8d4e | 3256 | leaf = path->nodes[0]; |
0b86a832 CM |
3257 | if (slot >= btrfs_header_nritems(leaf)) { |
3258 | ret = btrfs_next_leaf(root, path); | |
3259 | if (ret == 0) | |
3260 | continue; | |
3261 | if (ret < 0) | |
925baedd | 3262 | goto out; |
0b86a832 | 3263 | break; |
edbd8d4e | 3264 | } |
0b86a832 | 3265 | btrfs_item_key_to_cpu(leaf, &found_key, slot); |
edbd8d4e | 3266 | |
0b86a832 | 3267 | if (found_key.objectid >= key->objectid && |
925baedd CM |
3268 | found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) { |
3269 | ret = 0; | |
3270 | goto out; | |
3271 | } | |
0b86a832 | 3272 | path->slots[0]++; |
edbd8d4e | 3273 | } |
0b86a832 | 3274 | ret = -ENOENT; |
925baedd | 3275 | out: |
0b86a832 | 3276 | return ret; |
edbd8d4e CM |
3277 | } |
3278 | ||
9078a3e1 CM |
3279 | int btrfs_read_block_groups(struct btrfs_root *root) |
3280 | { | |
3281 | struct btrfs_path *path; | |
3282 | int ret; | |
96b5179d | 3283 | int bit; |
9078a3e1 | 3284 | struct btrfs_block_group_cache *cache; |
be744175 | 3285 | struct btrfs_fs_info *info = root->fs_info; |
6324fbf3 | 3286 | struct btrfs_space_info *space_info; |
d1310b2e | 3287 | struct extent_io_tree *block_group_cache; |
9078a3e1 CM |
3288 | struct btrfs_key key; |
3289 | struct btrfs_key found_key; | |
5f39d397 | 3290 | struct extent_buffer *leaf; |
96b5179d CM |
3291 | |
3292 | block_group_cache = &info->block_group_cache; | |
be744175 | 3293 | root = info->extent_root; |
9078a3e1 | 3294 | key.objectid = 0; |
0b86a832 | 3295 | key.offset = 0; |
9078a3e1 | 3296 | btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY); |
9078a3e1 CM |
3297 | path = btrfs_alloc_path(); |
3298 | if (!path) | |
3299 | return -ENOMEM; | |
3300 | ||
925baedd | 3301 | mutex_lock(&root->fs_info->alloc_mutex); |
9078a3e1 | 3302 | while(1) { |
0b86a832 CM |
3303 | ret = find_first_block_group(root, path, &key); |
3304 | if (ret > 0) { | |
3305 | ret = 0; | |
3306 | goto error; | |
9078a3e1 | 3307 | } |
0b86a832 CM |
3308 | if (ret != 0) |
3309 | goto error; | |
3310 | ||
5f39d397 CM |
3311 | leaf = path->nodes[0]; |
3312 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
8f18cf13 | 3313 | cache = kzalloc(sizeof(*cache), GFP_NOFS); |
9078a3e1 | 3314 | if (!cache) { |
0b86a832 | 3315 | ret = -ENOMEM; |
9078a3e1 CM |
3316 | break; |
3317 | } | |
3e1ad54f | 3318 | |
5f39d397 CM |
3319 | read_extent_buffer(leaf, &cache->item, |
3320 | btrfs_item_ptr_offset(leaf, path->slots[0]), | |
3321 | sizeof(cache->item)); | |
9078a3e1 | 3322 | memcpy(&cache->key, &found_key, sizeof(found_key)); |
0b86a832 | 3323 | |
9078a3e1 CM |
3324 | key.objectid = found_key.objectid + found_key.offset; |
3325 | btrfs_release_path(root, path); | |
0b86a832 CM |
3326 | cache->flags = btrfs_block_group_flags(&cache->item); |
3327 | bit = 0; | |
3328 | if (cache->flags & BTRFS_BLOCK_GROUP_DATA) { | |
96b5179d | 3329 | bit = BLOCK_GROUP_DATA; |
0b86a832 CM |
3330 | } else if (cache->flags & BTRFS_BLOCK_GROUP_SYSTEM) { |
3331 | bit = BLOCK_GROUP_SYSTEM; | |
3332 | } else if (cache->flags & BTRFS_BLOCK_GROUP_METADATA) { | |
96b5179d | 3333 | bit = BLOCK_GROUP_METADATA; |
31f3c99b | 3334 | } |
8790d502 | 3335 | set_avail_alloc_bits(info, cache->flags); |
96b5179d | 3336 | |
6324fbf3 CM |
3337 | ret = update_space_info(info, cache->flags, found_key.offset, |
3338 | btrfs_block_group_used(&cache->item), | |
3339 | &space_info); | |
3340 | BUG_ON(ret); | |
3341 | cache->space_info = space_info; | |
3342 | ||
96b5179d CM |
3343 | /* use EXTENT_LOCKED to prevent merging */ |
3344 | set_extent_bits(block_group_cache, found_key.objectid, | |
3345 | found_key.objectid + found_key.offset - 1, | |
3346 | bit | EXTENT_LOCKED, GFP_NOFS); | |
3347 | set_state_private(block_group_cache, found_key.objectid, | |
ae2f5411 | 3348 | (unsigned long)cache); |
96b5179d | 3349 | |
9078a3e1 | 3350 | if (key.objectid >= |
db94535d | 3351 | btrfs_super_total_bytes(&info->super_copy)) |
9078a3e1 CM |
3352 | break; |
3353 | } | |
0b86a832 CM |
3354 | ret = 0; |
3355 | error: | |
9078a3e1 | 3356 | btrfs_free_path(path); |
925baedd | 3357 | mutex_unlock(&root->fs_info->alloc_mutex); |
0b86a832 | 3358 | return ret; |
9078a3e1 | 3359 | } |
6324fbf3 CM |
3360 | |
3361 | int btrfs_make_block_group(struct btrfs_trans_handle *trans, | |
3362 | struct btrfs_root *root, u64 bytes_used, | |
e17cade2 | 3363 | u64 type, u64 chunk_objectid, u64 chunk_offset, |
6324fbf3 CM |
3364 | u64 size) |
3365 | { | |
3366 | int ret; | |
3367 | int bit = 0; | |
3368 | struct btrfs_root *extent_root; | |
3369 | struct btrfs_block_group_cache *cache; | |
3370 | struct extent_io_tree *block_group_cache; | |
3371 | ||
7d9eb12c | 3372 | WARN_ON(!mutex_is_locked(&root->fs_info->alloc_mutex)); |
6324fbf3 CM |
3373 | extent_root = root->fs_info->extent_root; |
3374 | block_group_cache = &root->fs_info->block_group_cache; | |
3375 | ||
8f18cf13 | 3376 | cache = kzalloc(sizeof(*cache), GFP_NOFS); |
6324fbf3 | 3377 | BUG_ON(!cache); |
e17cade2 | 3378 | cache->key.objectid = chunk_offset; |
6324fbf3 | 3379 | cache->key.offset = size; |
6324fbf3 | 3380 | btrfs_set_key_type(&cache->key, BTRFS_BLOCK_GROUP_ITEM_KEY); |
0ef3e66b | 3381 | |
6324fbf3 | 3382 | btrfs_set_block_group_used(&cache->item, bytes_used); |
6324fbf3 CM |
3383 | btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid); |
3384 | cache->flags = type; | |
3385 | btrfs_set_block_group_flags(&cache->item, type); | |
3386 | ||
3387 | ret = update_space_info(root->fs_info, cache->flags, size, bytes_used, | |
3388 | &cache->space_info); | |
3389 | BUG_ON(ret); | |
3390 | ||
d18a2c44 | 3391 | bit = block_group_state_bits(type); |
e17cade2 CM |
3392 | set_extent_bits(block_group_cache, chunk_offset, |
3393 | chunk_offset + size - 1, | |
6324fbf3 | 3394 | bit | EXTENT_LOCKED, GFP_NOFS); |
6324fbf3 | 3395 | |
e17cade2 CM |
3396 | set_state_private(block_group_cache, chunk_offset, |
3397 | (unsigned long)cache); | |
6324fbf3 CM |
3398 | ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item, |
3399 | sizeof(cache->item)); | |
3400 | BUG_ON(ret); | |
3401 | ||
3402 | finish_current_insert(trans, extent_root); | |
3403 | ret = del_pending_extents(trans, extent_root); | |
3404 | BUG_ON(ret); | |
d18a2c44 | 3405 | set_avail_alloc_bits(extent_root->fs_info, type); |
925baedd | 3406 | |
6324fbf3 CM |
3407 | return 0; |
3408 | } |