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 | */ | |
18 | ||
ec6b910f | 19 | #include <linux/sched.h> |
7bb86316 | 20 | #include <linux/crc32c.h> |
edbd8d4e | 21 | #include <linux/pagemap.h> |
74493f7a | 22 | #include "hash.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" |
fec577fb | 28 | |
0b86a832 | 29 | #define BLOCK_GROUP_DATA EXTENT_WRITEBACK |
96b5179d | 30 | #define BLOCK_GROUP_METADATA EXTENT_UPTODATE |
0b86a832 CM |
31 | #define BLOCK_GROUP_SYSTEM EXTENT_NEW |
32 | ||
96b5179d CM |
33 | #define BLOCK_GROUP_DIRTY EXTENT_DIRTY |
34 | ||
e089f05c CM |
35 | static int finish_current_insert(struct btrfs_trans_handle *trans, struct |
36 | btrfs_root *extent_root); | |
e20d96d6 CM |
37 | static int del_pending_extents(struct btrfs_trans_handle *trans, struct |
38 | btrfs_root *extent_root); | |
6324fbf3 CM |
39 | int btrfs_make_block_group(struct btrfs_trans_handle *trans, |
40 | struct btrfs_root *root, u64 bytes_used, | |
41 | u64 type, u64 chunk_tree, u64 chunk_objectid, | |
42 | u64 size); | |
d548ee51 | 43 | |
fec577fb | 44 | |
e37c9e69 CM |
45 | static int cache_block_group(struct btrfs_root *root, |
46 | struct btrfs_block_group_cache *block_group) | |
47 | { | |
48 | struct btrfs_path *path; | |
49 | int ret; | |
50 | struct btrfs_key key; | |
5f39d397 | 51 | struct extent_buffer *leaf; |
d1310b2e | 52 | struct extent_io_tree *free_space_cache; |
e37c9e69 | 53 | int slot; |
e37c9e69 CM |
54 | u64 last = 0; |
55 | u64 hole_size; | |
7d7d6068 | 56 | u64 first_free; |
e37c9e69 CM |
57 | int found = 0; |
58 | ||
00f5c795 CM |
59 | if (!block_group) |
60 | return 0; | |
61 | ||
e37c9e69 | 62 | root = root->fs_info->extent_root; |
f510cfec | 63 | free_space_cache = &root->fs_info->free_space_cache; |
e37c9e69 CM |
64 | |
65 | if (block_group->cached) | |
66 | return 0; | |
f510cfec | 67 | |
e37c9e69 CM |
68 | path = btrfs_alloc_path(); |
69 | if (!path) | |
70 | return -ENOMEM; | |
7d7d6068 | 71 | |
2cc58cf2 | 72 | path->reada = 2; |
7d7d6068 | 73 | first_free = block_group->key.objectid; |
e37c9e69 | 74 | key.objectid = block_group->key.objectid; |
e37c9e69 CM |
75 | key.offset = 0; |
76 | btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY); | |
77 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
78 | if (ret < 0) | |
79 | return ret; | |
0b86a832 | 80 | ret = btrfs_previous_item(root, path, 0, BTRFS_EXTENT_ITEM_KEY); |
d548ee51 Y |
81 | if (ret < 0) |
82 | return ret; | |
83 | if (ret == 0) { | |
84 | leaf = path->nodes[0]; | |
85 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
86 | if (key.objectid + key.offset > first_free) | |
87 | first_free = key.objectid + key.offset; | |
88 | } | |
e37c9e69 | 89 | while(1) { |
5f39d397 | 90 | leaf = path->nodes[0]; |
e37c9e69 | 91 | slot = path->slots[0]; |
5f39d397 | 92 | if (slot >= btrfs_header_nritems(leaf)) { |
e37c9e69 | 93 | ret = btrfs_next_leaf(root, path); |
54aa1f4d CM |
94 | if (ret < 0) |
95 | goto err; | |
de428b63 | 96 | if (ret == 0) { |
e37c9e69 | 97 | continue; |
de428b63 | 98 | } else { |
e37c9e69 CM |
99 | break; |
100 | } | |
101 | } | |
5f39d397 | 102 | btrfs_item_key_to_cpu(leaf, &key, slot); |
7d7d6068 | 103 | if (key.objectid < block_group->key.objectid) { |
7d7d6068 Y |
104 | goto next; |
105 | } | |
e37c9e69 CM |
106 | if (key.objectid >= block_group->key.objectid + |
107 | block_group->key.offset) { | |
e37c9e69 CM |
108 | break; |
109 | } | |
7d7d6068 | 110 | |
e37c9e69 CM |
111 | if (btrfs_key_type(&key) == BTRFS_EXTENT_ITEM_KEY) { |
112 | if (!found) { | |
7d7d6068 | 113 | last = first_free; |
e37c9e69 | 114 | found = 1; |
e37c9e69 | 115 | } |
f510cfec CM |
116 | if (key.objectid > last) { |
117 | hole_size = key.objectid - last; | |
118 | set_extent_dirty(free_space_cache, last, | |
119 | last + hole_size - 1, | |
120 | GFP_NOFS); | |
7d7d6068 Y |
121 | } |
122 | last = key.objectid + key.offset; | |
e37c9e69 | 123 | } |
7d7d6068 | 124 | next: |
e37c9e69 CM |
125 | path->slots[0]++; |
126 | } | |
127 | ||
7d7d6068 Y |
128 | if (!found) |
129 | last = first_free; | |
130 | if (block_group->key.objectid + | |
131 | block_group->key.offset > last) { | |
132 | hole_size = block_group->key.objectid + | |
133 | block_group->key.offset - last; | |
f510cfec CM |
134 | set_extent_dirty(free_space_cache, last, |
135 | last + hole_size - 1, GFP_NOFS); | |
7d7d6068 | 136 | } |
e37c9e69 | 137 | block_group->cached = 1; |
54aa1f4d | 138 | err: |
e37c9e69 CM |
139 | btrfs_free_path(path); |
140 | return 0; | |
141 | } | |
142 | ||
5276aeda CM |
143 | struct btrfs_block_group_cache *btrfs_lookup_block_group(struct |
144 | btrfs_fs_info *info, | |
db94535d | 145 | u64 bytenr) |
be744175 | 146 | { |
d1310b2e | 147 | struct extent_io_tree *block_group_cache; |
96b5179d CM |
148 | struct btrfs_block_group_cache *block_group = NULL; |
149 | u64 ptr; | |
150 | u64 start; | |
151 | u64 end; | |
be744175 CM |
152 | int ret; |
153 | ||
96b5179d CM |
154 | block_group_cache = &info->block_group_cache; |
155 | ret = find_first_extent_bit(block_group_cache, | |
db94535d | 156 | bytenr, &start, &end, |
0b86a832 CM |
157 | BLOCK_GROUP_DATA | BLOCK_GROUP_METADATA | |
158 | BLOCK_GROUP_SYSTEM); | |
be744175 | 159 | if (ret) { |
96b5179d | 160 | return NULL; |
be744175 | 161 | } |
96b5179d CM |
162 | ret = get_state_private(block_group_cache, start, &ptr); |
163 | if (ret) | |
164 | return NULL; | |
165 | ||
ae2f5411 | 166 | block_group = (struct btrfs_block_group_cache *)(unsigned long)ptr; |
5cf66426 | 167 | if (block_group->key.objectid <= bytenr && bytenr < |
96b5179d CM |
168 | block_group->key.objectid + block_group->key.offset) |
169 | return block_group; | |
be744175 CM |
170 | return NULL; |
171 | } | |
0b86a832 CM |
172 | |
173 | static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits) | |
174 | { | |
593060d7 | 175 | return (cache->flags & bits) == bits; |
0b86a832 CM |
176 | } |
177 | ||
178 | static int noinline find_search_start(struct btrfs_root *root, | |
98ed5174 | 179 | struct btrfs_block_group_cache **cache_ret, |
0b86a832 | 180 | u64 *start_ret, int num, int data) |
e37c9e69 | 181 | { |
e37c9e69 CM |
182 | int ret; |
183 | struct btrfs_block_group_cache *cache = *cache_ret; | |
d7fc640e | 184 | struct extent_io_tree *free_space_cache; |
7d1660d4 | 185 | struct extent_state *state; |
e19caa5f | 186 | u64 last; |
f510cfec | 187 | u64 start = 0; |
257d0ce3 | 188 | u64 cache_miss = 0; |
c31f8830 | 189 | u64 total_fs_bytes; |
0b86a832 | 190 | u64 search_start = *start_ret; |
f84a8b36 | 191 | int wrapped = 0; |
e37c9e69 | 192 | |
0b86a832 | 193 | if (!cache) |
1a2b2ac7 | 194 | goto out; |
c31f8830 | 195 | total_fs_bytes = btrfs_super_total_bytes(&root->fs_info->super_copy); |
d7fc640e CM |
196 | free_space_cache = &root->fs_info->free_space_cache; |
197 | ||
e37c9e69 | 198 | again: |
54aa1f4d CM |
199 | ret = cache_block_group(root, cache); |
200 | if (ret) | |
201 | goto out; | |
f84a8b36 | 202 | |
e19caa5f | 203 | last = max(search_start, cache->key.objectid); |
0b86a832 CM |
204 | if (!block_group_bits(cache, data)) { |
205 | goto new_group; | |
206 | } | |
e19caa5f | 207 | |
7d1660d4 CM |
208 | spin_lock_irq(&free_space_cache->lock); |
209 | state = find_first_extent_bit_state(free_space_cache, last, EXTENT_DIRTY); | |
e37c9e69 | 210 | while(1) { |
7d1660d4 | 211 | if (!state) { |
257d0ce3 CM |
212 | if (!cache_miss) |
213 | cache_miss = last; | |
7d1660d4 | 214 | spin_unlock_irq(&free_space_cache->lock); |
e19caa5f CM |
215 | goto new_group; |
216 | } | |
f510cfec | 217 | |
7d1660d4 CM |
218 | start = max(last, state->start); |
219 | last = state->end + 1; | |
257d0ce3 CM |
220 | if (last - start < num) { |
221 | if (last == cache->key.objectid + cache->key.offset) | |
222 | cache_miss = start; | |
7d1660d4 CM |
223 | do { |
224 | state = extent_state_next(state); | |
225 | } while(state && !(state->state & EXTENT_DIRTY)); | |
f510cfec | 226 | continue; |
257d0ce3 | 227 | } |
7d1660d4 | 228 | spin_unlock_irq(&free_space_cache->lock); |
0b86a832 | 229 | if (start + num > cache->key.objectid + cache->key.offset) |
e37c9e69 | 230 | goto new_group; |
c31f8830 CM |
231 | if (start + num > total_fs_bytes) |
232 | goto new_group; | |
8790d502 | 233 | if (!block_group_bits(cache, data)) { |
611f0e00 | 234 | printk("block group bits don't match %Lu %d\n", cache->flags, data); |
8790d502 | 235 | } |
0b86a832 CM |
236 | *start_ret = start; |
237 | return 0; | |
8790d502 CM |
238 | } |
239 | out: | |
1a2b2ac7 CM |
240 | cache = btrfs_lookup_block_group(root->fs_info, search_start); |
241 | if (!cache) { | |
0b86a832 | 242 | printk("Unable to find block group for %Lu\n", search_start); |
1a2b2ac7 | 243 | WARN_ON(1); |
1a2b2ac7 | 244 | } |
0b86a832 | 245 | return -ENOSPC; |
e37c9e69 CM |
246 | |
247 | new_group: | |
e19caa5f | 248 | last = cache->key.objectid + cache->key.offset; |
f84a8b36 | 249 | wrapped: |
e19caa5f | 250 | cache = btrfs_lookup_block_group(root->fs_info, last); |
c31f8830 | 251 | if (!cache || cache->key.objectid >= total_fs_bytes) { |
0e4de584 | 252 | no_cache: |
f84a8b36 CM |
253 | if (!wrapped) { |
254 | wrapped = 1; | |
255 | last = search_start; | |
f84a8b36 CM |
256 | goto wrapped; |
257 | } | |
1a2b2ac7 | 258 | goto out; |
e37c9e69 | 259 | } |
257d0ce3 CM |
260 | if (cache_miss && !cache->cached) { |
261 | cache_block_group(root, cache); | |
262 | last = cache_miss; | |
257d0ce3 CM |
263 | cache = btrfs_lookup_block_group(root->fs_info, last); |
264 | } | |
1a2b2ac7 | 265 | cache = btrfs_find_block_group(root, cache, last, data, 0); |
0e4de584 CM |
266 | if (!cache) |
267 | goto no_cache; | |
e37c9e69 | 268 | *cache_ret = cache; |
257d0ce3 | 269 | cache_miss = 0; |
e37c9e69 CM |
270 | goto again; |
271 | } | |
272 | ||
84f54cfa CM |
273 | static u64 div_factor(u64 num, int factor) |
274 | { | |
257d0ce3 CM |
275 | if (factor == 10) |
276 | return num; | |
84f54cfa CM |
277 | num *= factor; |
278 | do_div(num, 10); | |
279 | return num; | |
280 | } | |
281 | ||
6324fbf3 CM |
282 | static int block_group_state_bits(u64 flags) |
283 | { | |
284 | int bits = 0; | |
285 | if (flags & BTRFS_BLOCK_GROUP_DATA) | |
286 | bits |= BLOCK_GROUP_DATA; | |
287 | if (flags & BTRFS_BLOCK_GROUP_METADATA) | |
288 | bits |= BLOCK_GROUP_METADATA; | |
289 | if (flags & BTRFS_BLOCK_GROUP_SYSTEM) | |
290 | bits |= BLOCK_GROUP_SYSTEM; | |
291 | return bits; | |
292 | } | |
293 | ||
31f3c99b CM |
294 | struct btrfs_block_group_cache *btrfs_find_block_group(struct btrfs_root *root, |
295 | struct btrfs_block_group_cache | |
be744175 | 296 | *hint, u64 search_start, |
de428b63 | 297 | int data, int owner) |
cd1bc465 | 298 | { |
96b5179d | 299 | struct btrfs_block_group_cache *cache; |
d1310b2e | 300 | struct extent_io_tree *block_group_cache; |
31f3c99b | 301 | struct btrfs_block_group_cache *found_group = NULL; |
cd1bc465 CM |
302 | struct btrfs_fs_info *info = root->fs_info; |
303 | u64 used; | |
31f3c99b CM |
304 | u64 last = 0; |
305 | u64 hint_last; | |
96b5179d CM |
306 | u64 start; |
307 | u64 end; | |
308 | u64 free_check; | |
309 | u64 ptr; | |
c31f8830 | 310 | u64 total_fs_bytes; |
96b5179d | 311 | int bit; |
cd1bc465 | 312 | int ret; |
31f3c99b | 313 | int full_search = 0; |
de428b63 CM |
314 | int factor = 8; |
315 | ||
96b5179d | 316 | block_group_cache = &info->block_group_cache; |
c31f8830 | 317 | total_fs_bytes = btrfs_super_total_bytes(&root->fs_info->super_copy); |
96b5179d | 318 | |
de428b63 | 319 | if (!owner) |
f84a8b36 | 320 | factor = 8; |
be744175 | 321 | |
6324fbf3 | 322 | bit = block_group_state_bits(data); |
be744175 | 323 | |
c31f8830 | 324 | if (search_start && search_start < total_fs_bytes) { |
be744175 | 325 | struct btrfs_block_group_cache *shint; |
5276aeda | 326 | shint = btrfs_lookup_block_group(info, search_start); |
0b86a832 | 327 | if (shint && block_group_bits(shint, data)) { |
be744175 | 328 | used = btrfs_block_group_used(&shint->item); |
324ae4df Y |
329 | if (used + shint->pinned < |
330 | div_factor(shint->key.offset, factor)) { | |
be744175 CM |
331 | return shint; |
332 | } | |
333 | } | |
334 | } | |
0b86a832 CM |
335 | if (hint && block_group_bits(hint, data) && |
336 | hint->key.objectid < total_fs_bytes) { | |
31f3c99b | 337 | used = btrfs_block_group_used(&hint->item); |
324ae4df Y |
338 | if (used + hint->pinned < |
339 | div_factor(hint->key.offset, factor)) { | |
31f3c99b CM |
340 | return hint; |
341 | } | |
e19caa5f | 342 | last = hint->key.objectid + hint->key.offset; |
31f3c99b CM |
343 | hint_last = last; |
344 | } else { | |
e37c9e69 CM |
345 | if (hint) |
346 | hint_last = max(hint->key.objectid, search_start); | |
347 | else | |
348 | hint_last = search_start; | |
349 | ||
c31f8830 CM |
350 | if (hint_last >= total_fs_bytes) |
351 | hint_last = search_start; | |
e37c9e69 | 352 | last = hint_last; |
31f3c99b | 353 | } |
31f3c99b | 354 | again: |
cd1bc465 | 355 | while(1) { |
96b5179d CM |
356 | ret = find_first_extent_bit(block_group_cache, last, |
357 | &start, &end, bit); | |
358 | if (ret) | |
cd1bc465 | 359 | break; |
96b5179d CM |
360 | |
361 | ret = get_state_private(block_group_cache, start, &ptr); | |
362 | if (ret) | |
363 | break; | |
364 | ||
ae2f5411 | 365 | cache = (struct btrfs_block_group_cache *)(unsigned long)ptr; |
96b5179d CM |
366 | last = cache->key.objectid + cache->key.offset; |
367 | used = btrfs_block_group_used(&cache->item); | |
368 | ||
c31f8830 CM |
369 | if (cache->key.objectid > total_fs_bytes) |
370 | break; | |
371 | ||
8790d502 CM |
372 | if (block_group_bits(cache, data)) { |
373 | if (full_search) | |
374 | free_check = cache->key.offset; | |
375 | else | |
376 | free_check = div_factor(cache->key.offset, | |
377 | factor); | |
6324fbf3 | 378 | |
8790d502 CM |
379 | if (used + cache->pinned < free_check) { |
380 | found_group = cache; | |
381 | goto found; | |
382 | } | |
6324fbf3 | 383 | } |
de428b63 | 384 | cond_resched(); |
cd1bc465 | 385 | } |
31f3c99b | 386 | if (!full_search) { |
be744175 | 387 | last = search_start; |
31f3c99b CM |
388 | full_search = 1; |
389 | goto again; | |
390 | } | |
be744175 | 391 | found: |
31f3c99b | 392 | return found_group; |
cd1bc465 CM |
393 | } |
394 | ||
7bb86316 | 395 | static u64 hash_extent_ref(u64 root_objectid, u64 ref_generation, |
74493f7a CM |
396 | u64 owner, u64 owner_offset) |
397 | { | |
398 | u32 high_crc = ~(u32)0; | |
399 | u32 low_crc = ~(u32)0; | |
400 | __le64 lenum; | |
401 | ||
402 | lenum = cpu_to_le64(root_objectid); | |
403 | high_crc = crc32c(high_crc, &lenum, sizeof(lenum)); | |
7bb86316 CM |
404 | lenum = cpu_to_le64(ref_generation); |
405 | low_crc = crc32c(low_crc, &lenum, sizeof(lenum)); | |
21a4989d CM |
406 | if (owner >= BTRFS_FIRST_FREE_OBJECTID) { |
407 | lenum = cpu_to_le64(owner); | |
408 | low_crc = crc32c(low_crc, &lenum, sizeof(lenum)); | |
409 | lenum = cpu_to_le64(owner_offset); | |
410 | low_crc = crc32c(low_crc, &lenum, sizeof(lenum)); | |
411 | } | |
74493f7a CM |
412 | return ((u64)high_crc << 32) | (u64)low_crc; |
413 | } | |
414 | ||
7bb86316 CM |
415 | static int match_extent_ref(struct extent_buffer *leaf, |
416 | struct btrfs_extent_ref *disk_ref, | |
417 | struct btrfs_extent_ref *cpu_ref) | |
418 | { | |
419 | int ret; | |
420 | int len; | |
421 | ||
422 | if (cpu_ref->objectid) | |
423 | len = sizeof(*cpu_ref); | |
424 | else | |
425 | len = 2 * sizeof(u64); | |
426 | ret = memcmp_extent_buffer(leaf, cpu_ref, (unsigned long)disk_ref, | |
427 | len); | |
428 | return ret == 0; | |
429 | } | |
430 | ||
98ed5174 CM |
431 | static int noinline lookup_extent_backref(struct btrfs_trans_handle *trans, |
432 | struct btrfs_root *root, | |
433 | struct btrfs_path *path, u64 bytenr, | |
434 | u64 root_objectid, | |
435 | u64 ref_generation, u64 owner, | |
436 | u64 owner_offset, int del) | |
74493f7a CM |
437 | { |
438 | u64 hash; | |
439 | struct btrfs_key key; | |
7bb86316 | 440 | struct btrfs_key found_key; |
74493f7a | 441 | struct btrfs_extent_ref ref; |
7bb86316 CM |
442 | struct extent_buffer *leaf; |
443 | struct btrfs_extent_ref *disk_ref; | |
444 | int ret; | |
445 | int ret2; | |
446 | ||
447 | btrfs_set_stack_ref_root(&ref, root_objectid); | |
448 | btrfs_set_stack_ref_generation(&ref, ref_generation); | |
449 | btrfs_set_stack_ref_objectid(&ref, owner); | |
450 | btrfs_set_stack_ref_offset(&ref, owner_offset); | |
451 | ||
452 | hash = hash_extent_ref(root_objectid, ref_generation, owner, | |
453 | owner_offset); | |
454 | key.offset = hash; | |
455 | key.objectid = bytenr; | |
456 | key.type = BTRFS_EXTENT_REF_KEY; | |
457 | ||
458 | while (1) { | |
459 | ret = btrfs_search_slot(trans, root, &key, path, | |
460 | del ? -1 : 0, del); | |
461 | if (ret < 0) | |
462 | goto out; | |
463 | leaf = path->nodes[0]; | |
464 | if (ret != 0) { | |
465 | u32 nritems = btrfs_header_nritems(leaf); | |
466 | if (path->slots[0] >= nritems) { | |
467 | ret2 = btrfs_next_leaf(root, path); | |
468 | if (ret2) | |
469 | goto out; | |
470 | leaf = path->nodes[0]; | |
471 | } | |
472 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
473 | if (found_key.objectid != bytenr || | |
474 | found_key.type != BTRFS_EXTENT_REF_KEY) | |
475 | goto out; | |
476 | key.offset = found_key.offset; | |
477 | if (del) { | |
478 | btrfs_release_path(root, path); | |
479 | continue; | |
480 | } | |
481 | } | |
482 | disk_ref = btrfs_item_ptr(path->nodes[0], | |
483 | path->slots[0], | |
484 | struct btrfs_extent_ref); | |
485 | if (match_extent_ref(path->nodes[0], disk_ref, &ref)) { | |
486 | ret = 0; | |
487 | goto out; | |
488 | } | |
489 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
490 | key.offset = found_key.offset + 1; | |
491 | btrfs_release_path(root, path); | |
492 | } | |
493 | out: | |
494 | return ret; | |
495 | } | |
496 | ||
d8d5f3e1 CM |
497 | /* |
498 | * Back reference rules. Back refs have three main goals: | |
499 | * | |
500 | * 1) differentiate between all holders of references to an extent so that | |
501 | * when a reference is dropped we can make sure it was a valid reference | |
502 | * before freeing the extent. | |
503 | * | |
504 | * 2) Provide enough information to quickly find the holders of an extent | |
505 | * if we notice a given block is corrupted or bad. | |
506 | * | |
507 | * 3) Make it easy to migrate blocks for FS shrinking or storage pool | |
508 | * maintenance. This is actually the same as #2, but with a slightly | |
509 | * different use case. | |
510 | * | |
511 | * File extents can be referenced by: | |
512 | * | |
513 | * - multiple snapshots, subvolumes, or different generations in one subvol | |
514 | * - different files inside a single subvolume (in theory, not implemented yet) | |
515 | * - different offsets inside a file (bookend extents in file.c) | |
516 | * | |
517 | * The extent ref structure has fields for: | |
518 | * | |
519 | * - Objectid of the subvolume root | |
520 | * - Generation number of the tree holding the reference | |
521 | * - objectid of the file holding the reference | |
522 | * - offset in the file corresponding to the key holding the reference | |
523 | * | |
524 | * When a file extent is allocated the fields are filled in: | |
525 | * (root_key.objectid, trans->transid, inode objectid, offset in file) | |
526 | * | |
527 | * When a leaf is cow'd new references are added for every file extent found | |
528 | * in the leaf. It looks the same as the create case, but trans->transid | |
529 | * will be different when the block is cow'd. | |
530 | * | |
531 | * (root_key.objectid, trans->transid, inode objectid, offset in file) | |
532 | * | |
533 | * When a file extent is removed either during snapshot deletion or file | |
534 | * truncation, the corresponding back reference is found | |
535 | * by searching for: | |
536 | * | |
537 | * (btrfs_header_owner(leaf), btrfs_header_generation(leaf), | |
538 | * inode objectid, offset in file) | |
539 | * | |
540 | * Btree extents can be referenced by: | |
541 | * | |
542 | * - Different subvolumes | |
543 | * - Different generations of the same subvolume | |
544 | * | |
545 | * Storing sufficient information for a full reverse mapping of a btree | |
546 | * block would require storing the lowest key of the block in the backref, | |
547 | * and it would require updating that lowest key either before write out or | |
548 | * every time it changed. Instead, the objectid of the lowest key is stored | |
549 | * along with the level of the tree block. This provides a hint | |
550 | * about where in the btree the block can be found. Searches through the | |
551 | * btree only need to look for a pointer to that block, so they stop one | |
552 | * level higher than the level recorded in the backref. | |
553 | * | |
554 | * Some btrees do not do reference counting on their extents. These | |
555 | * include the extent tree and the tree of tree roots. Backrefs for these | |
556 | * trees always have a generation of zero. | |
557 | * | |
558 | * When a tree block is created, back references are inserted: | |
559 | * | |
f6dbff55 | 560 | * (root->root_key.objectid, trans->transid or zero, level, lowest_key_objectid) |
d8d5f3e1 CM |
561 | * |
562 | * When a tree block is cow'd in a reference counted root, | |
563 | * new back references are added for all the blocks it points to. | |
564 | * These are of the form (trans->transid will have increased since creation): | |
565 | * | |
f6dbff55 | 566 | * (root->root_key.objectid, trans->transid, level, lowest_key_objectid) |
d8d5f3e1 CM |
567 | * |
568 | * Because the lowest_key_objectid and the level are just hints | |
569 | * they are not used when backrefs are deleted. When a backref is deleted: | |
570 | * | |
571 | * if backref was for a tree root: | |
572 | * root_objectid = root->root_key.objectid | |
573 | * else | |
574 | * root_objectid = btrfs_header_owner(parent) | |
575 | * | |
576 | * (root_objectid, btrfs_header_generation(parent) or zero, 0, 0) | |
577 | * | |
578 | * Back Reference Key hashing: | |
579 | * | |
580 | * Back references have four fields, each 64 bits long. Unfortunately, | |
581 | * This is hashed into a single 64 bit number and placed into the key offset. | |
582 | * The key objectid corresponds to the first byte in the extent, and the | |
583 | * key type is set to BTRFS_EXTENT_REF_KEY | |
584 | */ | |
7bb86316 CM |
585 | int btrfs_insert_extent_backref(struct btrfs_trans_handle *trans, |
586 | struct btrfs_root *root, | |
587 | struct btrfs_path *path, u64 bytenr, | |
588 | u64 root_objectid, u64 ref_generation, | |
589 | u64 owner, u64 owner_offset) | |
590 | { | |
591 | u64 hash; | |
592 | struct btrfs_key key; | |
593 | struct btrfs_extent_ref ref; | |
594 | struct btrfs_extent_ref *disk_ref; | |
74493f7a CM |
595 | int ret; |
596 | ||
597 | btrfs_set_stack_ref_root(&ref, root_objectid); | |
7bb86316 | 598 | btrfs_set_stack_ref_generation(&ref, ref_generation); |
74493f7a CM |
599 | btrfs_set_stack_ref_objectid(&ref, owner); |
600 | btrfs_set_stack_ref_offset(&ref, owner_offset); | |
601 | ||
7bb86316 CM |
602 | hash = hash_extent_ref(root_objectid, ref_generation, owner, |
603 | owner_offset); | |
74493f7a CM |
604 | key.offset = hash; |
605 | key.objectid = bytenr; | |
606 | key.type = BTRFS_EXTENT_REF_KEY; | |
607 | ||
608 | ret = btrfs_insert_empty_item(trans, root, path, &key, sizeof(ref)); | |
609 | while (ret == -EEXIST) { | |
7bb86316 CM |
610 | disk_ref = btrfs_item_ptr(path->nodes[0], path->slots[0], |
611 | struct btrfs_extent_ref); | |
612 | if (match_extent_ref(path->nodes[0], disk_ref, &ref)) | |
613 | goto out; | |
614 | key.offset++; | |
615 | btrfs_release_path(root, path); | |
616 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
617 | sizeof(ref)); | |
74493f7a | 618 | } |
7bb86316 CM |
619 | if (ret) |
620 | goto out; | |
621 | disk_ref = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
622 | struct btrfs_extent_ref); | |
623 | write_extent_buffer(path->nodes[0], &ref, (unsigned long)disk_ref, | |
624 | sizeof(ref)); | |
625 | btrfs_mark_buffer_dirty(path->nodes[0]); | |
626 | out: | |
627 | btrfs_release_path(root, path); | |
628 | return ret; | |
74493f7a CM |
629 | } |
630 | ||
b18c6685 CM |
631 | int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans, |
632 | struct btrfs_root *root, | |
74493f7a | 633 | u64 bytenr, u64 num_bytes, |
7bb86316 | 634 | u64 root_objectid, u64 ref_generation, |
74493f7a | 635 | u64 owner, u64 owner_offset) |
02217ed2 | 636 | { |
5caf2a00 | 637 | struct btrfs_path *path; |
02217ed2 | 638 | int ret; |
e2fa7227 | 639 | struct btrfs_key key; |
5f39d397 | 640 | struct extent_buffer *l; |
234b63a0 | 641 | struct btrfs_extent_item *item; |
cf27e1ee | 642 | u32 refs; |
037e6390 | 643 | |
db94535d | 644 | WARN_ON(num_bytes < root->sectorsize); |
5caf2a00 | 645 | path = btrfs_alloc_path(); |
54aa1f4d CM |
646 | if (!path) |
647 | return -ENOMEM; | |
26b8003f | 648 | |
b0331a4c | 649 | path->reada = 0; |
db94535d | 650 | key.objectid = bytenr; |
62e2749e | 651 | btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY); |
db94535d | 652 | key.offset = num_bytes; |
5caf2a00 | 653 | ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path, |
9f5fae2f | 654 | 0, 1); |
54aa1f4d CM |
655 | if (ret < 0) |
656 | return ret; | |
a429e513 | 657 | if (ret != 0) { |
a28ec197 | 658 | BUG(); |
a429e513 | 659 | } |
02217ed2 | 660 | BUG_ON(ret != 0); |
5f39d397 | 661 | l = path->nodes[0]; |
5caf2a00 | 662 | item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item); |
5f39d397 CM |
663 | refs = btrfs_extent_refs(l, item); |
664 | btrfs_set_extent_refs(l, item, refs + 1); | |
5caf2a00 | 665 | btrfs_mark_buffer_dirty(path->nodes[0]); |
a28ec197 | 666 | |
5caf2a00 | 667 | btrfs_release_path(root->fs_info->extent_root, path); |
7bb86316 | 668 | |
b0331a4c | 669 | path->reada = 0; |
7bb86316 CM |
670 | ret = btrfs_insert_extent_backref(trans, root->fs_info->extent_root, |
671 | path, bytenr, root_objectid, | |
672 | ref_generation, owner, owner_offset); | |
673 | BUG_ON(ret); | |
9f5fae2f | 674 | finish_current_insert(trans, root->fs_info->extent_root); |
e20d96d6 | 675 | del_pending_extents(trans, root->fs_info->extent_root); |
74493f7a CM |
676 | |
677 | btrfs_free_path(path); | |
02217ed2 CM |
678 | return 0; |
679 | } | |
680 | ||
e9d0b13b CM |
681 | int btrfs_extent_post_op(struct btrfs_trans_handle *trans, |
682 | struct btrfs_root *root) | |
683 | { | |
684 | finish_current_insert(trans, root->fs_info->extent_root); | |
685 | del_pending_extents(trans, root->fs_info->extent_root); | |
686 | return 0; | |
687 | } | |
688 | ||
b18c6685 | 689 | static int lookup_extent_ref(struct btrfs_trans_handle *trans, |
db94535d CM |
690 | struct btrfs_root *root, u64 bytenr, |
691 | u64 num_bytes, u32 *refs) | |
a28ec197 | 692 | { |
5caf2a00 | 693 | struct btrfs_path *path; |
a28ec197 | 694 | int ret; |
e2fa7227 | 695 | struct btrfs_key key; |
5f39d397 | 696 | struct extent_buffer *l; |
234b63a0 | 697 | struct btrfs_extent_item *item; |
5caf2a00 | 698 | |
db94535d | 699 | WARN_ON(num_bytes < root->sectorsize); |
5caf2a00 | 700 | path = btrfs_alloc_path(); |
b0331a4c | 701 | path->reada = 0; |
db94535d CM |
702 | key.objectid = bytenr; |
703 | key.offset = num_bytes; | |
62e2749e | 704 | btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY); |
5caf2a00 | 705 | ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path, |
9f5fae2f | 706 | 0, 0); |
54aa1f4d CM |
707 | if (ret < 0) |
708 | goto out; | |
5f39d397 CM |
709 | if (ret != 0) { |
710 | btrfs_print_leaf(root, path->nodes[0]); | |
db94535d | 711 | printk("failed to find block number %Lu\n", bytenr); |
a28ec197 | 712 | BUG(); |
5f39d397 CM |
713 | } |
714 | l = path->nodes[0]; | |
5caf2a00 | 715 | item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item); |
5f39d397 | 716 | *refs = btrfs_extent_refs(l, item); |
54aa1f4d | 717 | out: |
5caf2a00 | 718 | btrfs_free_path(path); |
a28ec197 CM |
719 | return 0; |
720 | } | |
721 | ||
be20aa9d CM |
722 | u32 btrfs_count_snapshots_in_path(struct btrfs_root *root, |
723 | struct btrfs_path *count_path, | |
724 | u64 first_extent) | |
725 | { | |
726 | struct btrfs_root *extent_root = root->fs_info->extent_root; | |
727 | struct btrfs_path *path; | |
728 | u64 bytenr; | |
729 | u64 found_objectid; | |
56b453c9 | 730 | u64 root_objectid = root->root_key.objectid; |
be20aa9d CM |
731 | u32 total_count = 0; |
732 | u32 cur_count; | |
be20aa9d CM |
733 | u32 nritems; |
734 | int ret; | |
735 | struct btrfs_key key; | |
736 | struct btrfs_key found_key; | |
737 | struct extent_buffer *l; | |
738 | struct btrfs_extent_item *item; | |
739 | struct btrfs_extent_ref *ref_item; | |
740 | int level = -1; | |
741 | ||
742 | path = btrfs_alloc_path(); | |
743 | again: | |
744 | if (level == -1) | |
745 | bytenr = first_extent; | |
746 | else | |
747 | bytenr = count_path->nodes[level]->start; | |
748 | ||
749 | cur_count = 0; | |
750 | key.objectid = bytenr; | |
751 | key.offset = 0; | |
752 | ||
753 | btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY); | |
754 | ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0); | |
755 | if (ret < 0) | |
756 | goto out; | |
757 | BUG_ON(ret == 0); | |
758 | ||
759 | l = path->nodes[0]; | |
760 | btrfs_item_key_to_cpu(l, &found_key, path->slots[0]); | |
761 | ||
762 | if (found_key.objectid != bytenr || | |
763 | found_key.type != BTRFS_EXTENT_ITEM_KEY) { | |
764 | goto out; | |
765 | } | |
766 | ||
767 | item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item); | |
be20aa9d | 768 | while (1) { |
bd09835d | 769 | l = path->nodes[0]; |
be20aa9d CM |
770 | nritems = btrfs_header_nritems(l); |
771 | if (path->slots[0] >= nritems) { | |
772 | ret = btrfs_next_leaf(extent_root, path); | |
773 | if (ret == 0) | |
774 | continue; | |
775 | break; | |
776 | } | |
777 | btrfs_item_key_to_cpu(l, &found_key, path->slots[0]); | |
778 | if (found_key.objectid != bytenr) | |
779 | break; | |
bd09835d | 780 | |
be20aa9d CM |
781 | if (found_key.type != BTRFS_EXTENT_REF_KEY) { |
782 | path->slots[0]++; | |
783 | continue; | |
784 | } | |
785 | ||
786 | cur_count++; | |
787 | ref_item = btrfs_item_ptr(l, path->slots[0], | |
788 | struct btrfs_extent_ref); | |
789 | found_objectid = btrfs_ref_root(l, ref_item); | |
790 | ||
56b453c9 CM |
791 | if (found_objectid != root_objectid) { |
792 | total_count = 2; | |
4313b399 | 793 | goto out; |
56b453c9 CM |
794 | } |
795 | total_count = 1; | |
be20aa9d CM |
796 | path->slots[0]++; |
797 | } | |
798 | if (cur_count == 0) { | |
799 | total_count = 0; | |
800 | goto out; | |
801 | } | |
be20aa9d CM |
802 | if (level >= 0 && root->node == count_path->nodes[level]) |
803 | goto out; | |
804 | level++; | |
805 | btrfs_release_path(root, path); | |
806 | goto again; | |
807 | ||
808 | out: | |
809 | btrfs_free_path(path); | |
810 | return total_count; | |
be20aa9d | 811 | } |
c5739bba | 812 | int btrfs_inc_root_ref(struct btrfs_trans_handle *trans, |
7bb86316 | 813 | struct btrfs_root *root, u64 owner_objectid) |
c5739bba | 814 | { |
7bb86316 CM |
815 | u64 generation; |
816 | u64 key_objectid; | |
817 | u64 level; | |
818 | u32 nritems; | |
819 | struct btrfs_disk_key disk_key; | |
820 | ||
821 | level = btrfs_header_level(root->node); | |
822 | generation = trans->transid; | |
823 | nritems = btrfs_header_nritems(root->node); | |
824 | if (nritems > 0) { | |
825 | if (level == 0) | |
826 | btrfs_item_key(root->node, &disk_key, 0); | |
827 | else | |
828 | btrfs_node_key(root->node, &disk_key, 0); | |
829 | key_objectid = btrfs_disk_key_objectid(&disk_key); | |
830 | } else { | |
831 | key_objectid = 0; | |
832 | } | |
db94535d | 833 | return btrfs_inc_extent_ref(trans, root, root->node->start, |
7bb86316 | 834 | root->node->len, owner_objectid, |
f6dbff55 | 835 | generation, level, key_objectid); |
c5739bba CM |
836 | } |
837 | ||
e089f05c | 838 | int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root, |
5f39d397 | 839 | struct extent_buffer *buf) |
02217ed2 | 840 | { |
db94535d | 841 | u64 bytenr; |
5f39d397 CM |
842 | u32 nritems; |
843 | struct btrfs_key key; | |
6407bf6d | 844 | struct btrfs_file_extent_item *fi; |
02217ed2 | 845 | int i; |
db94535d | 846 | int level; |
6407bf6d | 847 | int ret; |
54aa1f4d | 848 | int faili; |
a28ec197 | 849 | |
3768f368 | 850 | if (!root->ref_cows) |
a28ec197 | 851 | return 0; |
5f39d397 | 852 | |
db94535d | 853 | level = btrfs_header_level(buf); |
5f39d397 CM |
854 | nritems = btrfs_header_nritems(buf); |
855 | for (i = 0; i < nritems; i++) { | |
db94535d CM |
856 | if (level == 0) { |
857 | u64 disk_bytenr; | |
5f39d397 CM |
858 | btrfs_item_key_to_cpu(buf, &key, i); |
859 | if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY) | |
6407bf6d | 860 | continue; |
5f39d397 | 861 | fi = btrfs_item_ptr(buf, i, |
6407bf6d | 862 | struct btrfs_file_extent_item); |
5f39d397 | 863 | if (btrfs_file_extent_type(buf, fi) == |
236454df CM |
864 | BTRFS_FILE_EXTENT_INLINE) |
865 | continue; | |
db94535d CM |
866 | disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi); |
867 | if (disk_bytenr == 0) | |
3a686375 | 868 | continue; |
db94535d | 869 | ret = btrfs_inc_extent_ref(trans, root, disk_bytenr, |
7bb86316 CM |
870 | btrfs_file_extent_disk_num_bytes(buf, fi), |
871 | root->root_key.objectid, trans->transid, | |
872 | key.objectid, key.offset); | |
54aa1f4d CM |
873 | if (ret) { |
874 | faili = i; | |
875 | goto fail; | |
876 | } | |
6407bf6d | 877 | } else { |
db94535d | 878 | bytenr = btrfs_node_blockptr(buf, i); |
6caab489 | 879 | btrfs_node_key_to_cpu(buf, &key, i); |
db94535d | 880 | ret = btrfs_inc_extent_ref(trans, root, bytenr, |
7bb86316 CM |
881 | btrfs_level_size(root, level - 1), |
882 | root->root_key.objectid, | |
f6dbff55 CM |
883 | trans->transid, |
884 | level - 1, key.objectid); | |
54aa1f4d CM |
885 | if (ret) { |
886 | faili = i; | |
887 | goto fail; | |
888 | } | |
6407bf6d | 889 | } |
02217ed2 CM |
890 | } |
891 | return 0; | |
54aa1f4d | 892 | fail: |
ccd467d6 | 893 | WARN_ON(1); |
7bb86316 | 894 | #if 0 |
54aa1f4d | 895 | for (i =0; i < faili; i++) { |
db94535d CM |
896 | if (level == 0) { |
897 | u64 disk_bytenr; | |
5f39d397 CM |
898 | btrfs_item_key_to_cpu(buf, &key, i); |
899 | if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY) | |
54aa1f4d | 900 | continue; |
5f39d397 | 901 | fi = btrfs_item_ptr(buf, i, |
54aa1f4d | 902 | struct btrfs_file_extent_item); |
5f39d397 | 903 | if (btrfs_file_extent_type(buf, fi) == |
54aa1f4d CM |
904 | BTRFS_FILE_EXTENT_INLINE) |
905 | continue; | |
db94535d CM |
906 | disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi); |
907 | if (disk_bytenr == 0) | |
54aa1f4d | 908 | continue; |
db94535d CM |
909 | err = btrfs_free_extent(trans, root, disk_bytenr, |
910 | btrfs_file_extent_disk_num_bytes(buf, | |
5f39d397 | 911 | fi), 0); |
54aa1f4d CM |
912 | BUG_ON(err); |
913 | } else { | |
db94535d CM |
914 | bytenr = btrfs_node_blockptr(buf, i); |
915 | err = btrfs_free_extent(trans, root, bytenr, | |
916 | btrfs_level_size(root, level - 1), 0); | |
54aa1f4d CM |
917 | BUG_ON(err); |
918 | } | |
919 | } | |
7bb86316 | 920 | #endif |
54aa1f4d | 921 | return ret; |
02217ed2 CM |
922 | } |
923 | ||
9078a3e1 CM |
924 | static int write_one_cache_group(struct btrfs_trans_handle *trans, |
925 | struct btrfs_root *root, | |
926 | struct btrfs_path *path, | |
927 | struct btrfs_block_group_cache *cache) | |
928 | { | |
929 | int ret; | |
930 | int pending_ret; | |
931 | struct btrfs_root *extent_root = root->fs_info->extent_root; | |
5f39d397 CM |
932 | unsigned long bi; |
933 | struct extent_buffer *leaf; | |
9078a3e1 | 934 | |
9078a3e1 | 935 | ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1); |
54aa1f4d CM |
936 | if (ret < 0) |
937 | goto fail; | |
9078a3e1 | 938 | BUG_ON(ret); |
5f39d397 CM |
939 | |
940 | leaf = path->nodes[0]; | |
941 | bi = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
942 | write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item)); | |
943 | btrfs_mark_buffer_dirty(leaf); | |
9078a3e1 | 944 | btrfs_release_path(extent_root, path); |
54aa1f4d | 945 | fail: |
9078a3e1 CM |
946 | finish_current_insert(trans, extent_root); |
947 | pending_ret = del_pending_extents(trans, extent_root); | |
948 | if (ret) | |
949 | return ret; | |
950 | if (pending_ret) | |
951 | return pending_ret; | |
952 | return 0; | |
953 | ||
954 | } | |
955 | ||
96b5179d CM |
956 | int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans, |
957 | struct btrfs_root *root) | |
9078a3e1 | 958 | { |
d1310b2e | 959 | struct extent_io_tree *block_group_cache; |
96b5179d | 960 | struct btrfs_block_group_cache *cache; |
9078a3e1 CM |
961 | int ret; |
962 | int err = 0; | |
963 | int werr = 0; | |
9078a3e1 | 964 | struct btrfs_path *path; |
96b5179d CM |
965 | u64 last = 0; |
966 | u64 start; | |
967 | u64 end; | |
968 | u64 ptr; | |
9078a3e1 | 969 | |
96b5179d | 970 | block_group_cache = &root->fs_info->block_group_cache; |
9078a3e1 CM |
971 | path = btrfs_alloc_path(); |
972 | if (!path) | |
973 | return -ENOMEM; | |
974 | ||
975 | while(1) { | |
96b5179d CM |
976 | ret = find_first_extent_bit(block_group_cache, last, |
977 | &start, &end, BLOCK_GROUP_DIRTY); | |
978 | if (ret) | |
9078a3e1 | 979 | break; |
54aa1f4d | 980 | |
96b5179d CM |
981 | last = end + 1; |
982 | ret = get_state_private(block_group_cache, start, &ptr); | |
983 | if (ret) | |
984 | break; | |
985 | ||
ae2f5411 | 986 | cache = (struct btrfs_block_group_cache *)(unsigned long)ptr; |
96b5179d CM |
987 | err = write_one_cache_group(trans, root, |
988 | path, cache); | |
989 | /* | |
990 | * if we fail to write the cache group, we want | |
991 | * to keep it marked dirty in hopes that a later | |
992 | * write will work | |
993 | */ | |
994 | if (err) { | |
995 | werr = err; | |
996 | continue; | |
9078a3e1 | 997 | } |
96b5179d CM |
998 | clear_extent_bits(block_group_cache, start, end, |
999 | BLOCK_GROUP_DIRTY, GFP_NOFS); | |
9078a3e1 CM |
1000 | } |
1001 | btrfs_free_path(path); | |
1002 | return werr; | |
1003 | } | |
1004 | ||
6324fbf3 CM |
1005 | static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info, |
1006 | u64 flags) | |
1007 | { | |
1008 | struct list_head *head = &info->space_info; | |
1009 | struct list_head *cur; | |
1010 | struct btrfs_space_info *found; | |
1011 | list_for_each(cur, head) { | |
1012 | found = list_entry(cur, struct btrfs_space_info, list); | |
1013 | if (found->flags == flags) | |
1014 | return found; | |
1015 | } | |
1016 | return NULL; | |
1017 | ||
1018 | } | |
1019 | ||
593060d7 CM |
1020 | static int update_space_info(struct btrfs_fs_info *info, u64 flags, |
1021 | u64 total_bytes, u64 bytes_used, | |
1022 | struct btrfs_space_info **space_info) | |
1023 | { | |
1024 | struct btrfs_space_info *found; | |
1025 | ||
1026 | found = __find_space_info(info, flags); | |
1027 | if (found) { | |
1028 | found->total_bytes += total_bytes; | |
1029 | found->bytes_used += bytes_used; | |
1030 | WARN_ON(found->total_bytes < found->bytes_used); | |
1031 | *space_info = found; | |
1032 | return 0; | |
1033 | } | |
1034 | found = kmalloc(sizeof(*found), GFP_NOFS); | |
1035 | if (!found) | |
1036 | return -ENOMEM; | |
1037 | ||
1038 | list_add(&found->list, &info->space_info); | |
1039 | found->flags = flags; | |
1040 | found->total_bytes = total_bytes; | |
1041 | found->bytes_used = bytes_used; | |
1042 | found->bytes_pinned = 0; | |
1043 | found->full = 0; | |
1044 | *space_info = found; | |
1045 | return 0; | |
1046 | } | |
1047 | ||
8790d502 CM |
1048 | static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags) |
1049 | { | |
1050 | u64 extra_flags = flags & (BTRFS_BLOCK_GROUP_RAID0 | | |
611f0e00 CM |
1051 | BTRFS_BLOCK_GROUP_RAID1 | |
1052 | BTRFS_BLOCK_GROUP_DUP); | |
8790d502 CM |
1053 | if (extra_flags) { |
1054 | if (flags & BTRFS_BLOCK_GROUP_DATA) | |
1055 | fs_info->avail_data_alloc_bits |= extra_flags; | |
1056 | if (flags & BTRFS_BLOCK_GROUP_METADATA) | |
1057 | fs_info->avail_metadata_alloc_bits |= extra_flags; | |
1058 | if (flags & BTRFS_BLOCK_GROUP_SYSTEM) | |
1059 | fs_info->avail_system_alloc_bits |= extra_flags; | |
1060 | } | |
1061 | } | |
593060d7 | 1062 | |
6324fbf3 CM |
1063 | static int do_chunk_alloc(struct btrfs_trans_handle *trans, |
1064 | struct btrfs_root *extent_root, u64 alloc_bytes, | |
1065 | u64 flags) | |
1066 | { | |
1067 | struct btrfs_space_info *space_info; | |
1068 | u64 thresh; | |
1069 | u64 start; | |
1070 | u64 num_bytes; | |
1071 | int ret; | |
1072 | ||
1073 | space_info = __find_space_info(extent_root->fs_info, flags); | |
593060d7 CM |
1074 | if (!space_info) { |
1075 | ret = update_space_info(extent_root->fs_info, flags, | |
1076 | 0, 0, &space_info); | |
1077 | BUG_ON(ret); | |
1078 | } | |
6324fbf3 CM |
1079 | BUG_ON(!space_info); |
1080 | ||
1081 | if (space_info->full) | |
1082 | return 0; | |
1083 | ||
8790d502 | 1084 | thresh = div_factor(space_info->total_bytes, 6); |
6324fbf3 CM |
1085 | if ((space_info->bytes_used + space_info->bytes_pinned + alloc_bytes) < |
1086 | thresh) | |
1087 | return 0; | |
1088 | ||
1089 | ret = btrfs_alloc_chunk(trans, extent_root, &start, &num_bytes, flags); | |
1090 | if (ret == -ENOSPC) { | |
1091 | printk("space info full %Lu\n", flags); | |
1092 | space_info->full = 1; | |
1093 | return 0; | |
1094 | } | |
1095 | ||
1096 | BUG_ON(ret); | |
1097 | ||
1098 | ret = btrfs_make_block_group(trans, extent_root, 0, flags, | |
1099 | extent_root->fs_info->chunk_root->root_key.objectid, | |
1100 | start, num_bytes); | |
1101 | BUG_ON(ret); | |
593060d7 | 1102 | |
8790d502 | 1103 | set_avail_alloc_bits(extent_root->fs_info, flags); |
6324fbf3 CM |
1104 | return 0; |
1105 | } | |
1106 | ||
9078a3e1 CM |
1107 | static int update_block_group(struct btrfs_trans_handle *trans, |
1108 | struct btrfs_root *root, | |
db94535d | 1109 | u64 bytenr, u64 num_bytes, int alloc, |
0b86a832 | 1110 | int mark_free) |
9078a3e1 CM |
1111 | { |
1112 | struct btrfs_block_group_cache *cache; | |
1113 | struct btrfs_fs_info *info = root->fs_info; | |
db94535d | 1114 | u64 total = num_bytes; |
9078a3e1 | 1115 | u64 old_val; |
db94535d | 1116 | u64 byte_in_group; |
96b5179d CM |
1117 | u64 start; |
1118 | u64 end; | |
3e1ad54f | 1119 | |
9078a3e1 | 1120 | while(total) { |
db94535d | 1121 | cache = btrfs_lookup_block_group(info, bytenr); |
3e1ad54f | 1122 | if (!cache) { |
9078a3e1 | 1123 | return -1; |
cd1bc465 | 1124 | } |
db94535d CM |
1125 | byte_in_group = bytenr - cache->key.objectid; |
1126 | WARN_ON(byte_in_group > cache->key.offset); | |
96b5179d CM |
1127 | start = cache->key.objectid; |
1128 | end = start + cache->key.offset - 1; | |
1129 | set_extent_bits(&info->block_group_cache, start, end, | |
1130 | BLOCK_GROUP_DIRTY, GFP_NOFS); | |
9078a3e1 CM |
1131 | |
1132 | old_val = btrfs_block_group_used(&cache->item); | |
db94535d | 1133 | num_bytes = min(total, cache->key.offset - byte_in_group); |
cd1bc465 | 1134 | if (alloc) { |
db94535d | 1135 | old_val += num_bytes; |
6324fbf3 | 1136 | cache->space_info->bytes_used += num_bytes; |
cd1bc465 | 1137 | } else { |
db94535d | 1138 | old_val -= num_bytes; |
6324fbf3 | 1139 | cache->space_info->bytes_used -= num_bytes; |
f510cfec CM |
1140 | if (mark_free) { |
1141 | set_extent_dirty(&info->free_space_cache, | |
db94535d | 1142 | bytenr, bytenr + num_bytes - 1, |
f510cfec | 1143 | GFP_NOFS); |
e37c9e69 | 1144 | } |
cd1bc465 | 1145 | } |
9078a3e1 | 1146 | btrfs_set_block_group_used(&cache->item, old_val); |
db94535d CM |
1147 | total -= num_bytes; |
1148 | bytenr += num_bytes; | |
9078a3e1 CM |
1149 | } |
1150 | return 0; | |
1151 | } | |
6324fbf3 | 1152 | |
324ae4df Y |
1153 | static int update_pinned_extents(struct btrfs_root *root, |
1154 | u64 bytenr, u64 num, int pin) | |
1155 | { | |
1156 | u64 len; | |
1157 | struct btrfs_block_group_cache *cache; | |
1158 | struct btrfs_fs_info *fs_info = root->fs_info; | |
1159 | ||
1160 | if (pin) { | |
1161 | set_extent_dirty(&fs_info->pinned_extents, | |
1162 | bytenr, bytenr + num - 1, GFP_NOFS); | |
1163 | } else { | |
1164 | clear_extent_dirty(&fs_info->pinned_extents, | |
1165 | bytenr, bytenr + num - 1, GFP_NOFS); | |
1166 | } | |
1167 | while (num > 0) { | |
1168 | cache = btrfs_lookup_block_group(fs_info, bytenr); | |
1169 | WARN_ON(!cache); | |
1170 | len = min(num, cache->key.offset - | |
1171 | (bytenr - cache->key.objectid)); | |
1172 | if (pin) { | |
1173 | cache->pinned += len; | |
6324fbf3 | 1174 | cache->space_info->bytes_pinned += len; |
324ae4df Y |
1175 | fs_info->total_pinned += len; |
1176 | } else { | |
1177 | cache->pinned -= len; | |
6324fbf3 | 1178 | cache->space_info->bytes_pinned -= len; |
324ae4df Y |
1179 | fs_info->total_pinned -= len; |
1180 | } | |
1181 | bytenr += len; | |
1182 | num -= len; | |
1183 | } | |
1184 | return 0; | |
1185 | } | |
9078a3e1 | 1186 | |
d1310b2e | 1187 | int btrfs_copy_pinned(struct btrfs_root *root, struct extent_io_tree *copy) |
ccd467d6 | 1188 | { |
ccd467d6 | 1189 | u64 last = 0; |
1a5bc167 CM |
1190 | u64 start; |
1191 | u64 end; | |
d1310b2e | 1192 | struct extent_io_tree *pinned_extents = &root->fs_info->pinned_extents; |
ccd467d6 | 1193 | int ret; |
ccd467d6 CM |
1194 | |
1195 | while(1) { | |
1a5bc167 CM |
1196 | ret = find_first_extent_bit(pinned_extents, last, |
1197 | &start, &end, EXTENT_DIRTY); | |
1198 | if (ret) | |
ccd467d6 | 1199 | break; |
1a5bc167 CM |
1200 | set_extent_dirty(copy, start, end, GFP_NOFS); |
1201 | last = end + 1; | |
ccd467d6 CM |
1202 | } |
1203 | return 0; | |
1204 | } | |
1205 | ||
1206 | int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans, | |
1207 | struct btrfs_root *root, | |
d1310b2e | 1208 | struct extent_io_tree *unpin) |
a28ec197 | 1209 | { |
1a5bc167 CM |
1210 | u64 start; |
1211 | u64 end; | |
a28ec197 | 1212 | int ret; |
d1310b2e | 1213 | struct extent_io_tree *free_space_cache; |
f510cfec | 1214 | free_space_cache = &root->fs_info->free_space_cache; |
a28ec197 CM |
1215 | |
1216 | while(1) { | |
1a5bc167 CM |
1217 | ret = find_first_extent_bit(unpin, 0, &start, &end, |
1218 | EXTENT_DIRTY); | |
1219 | if (ret) | |
a28ec197 | 1220 | break; |
324ae4df | 1221 | update_pinned_extents(root, start, end + 1 - start, 0); |
1a5bc167 CM |
1222 | clear_extent_dirty(unpin, start, end, GFP_NOFS); |
1223 | set_extent_dirty(free_space_cache, start, end, GFP_NOFS); | |
a28ec197 CM |
1224 | } |
1225 | return 0; | |
1226 | } | |
1227 | ||
98ed5174 CM |
1228 | static int finish_current_insert(struct btrfs_trans_handle *trans, |
1229 | struct btrfs_root *extent_root) | |
037e6390 | 1230 | { |
7bb86316 CM |
1231 | u64 start; |
1232 | u64 end; | |
1233 | struct btrfs_fs_info *info = extent_root->fs_info; | |
d8d5f3e1 | 1234 | struct extent_buffer *eb; |
7bb86316 | 1235 | struct btrfs_path *path; |
e2fa7227 | 1236 | struct btrfs_key ins; |
d8d5f3e1 | 1237 | struct btrfs_disk_key first; |
234b63a0 | 1238 | struct btrfs_extent_item extent_item; |
037e6390 | 1239 | int ret; |
d8d5f3e1 | 1240 | int level; |
1a5bc167 | 1241 | int err = 0; |
037e6390 | 1242 | |
5f39d397 | 1243 | btrfs_set_stack_extent_refs(&extent_item, 1); |
62e2749e | 1244 | btrfs_set_key_type(&ins, BTRFS_EXTENT_ITEM_KEY); |
7bb86316 | 1245 | path = btrfs_alloc_path(); |
037e6390 | 1246 | |
26b8003f | 1247 | while(1) { |
1a5bc167 CM |
1248 | ret = find_first_extent_bit(&info->extent_ins, 0, &start, |
1249 | &end, EXTENT_LOCKED); | |
1250 | if (ret) | |
26b8003f CM |
1251 | break; |
1252 | ||
1a5bc167 CM |
1253 | ins.objectid = start; |
1254 | ins.offset = end + 1 - start; | |
1255 | err = btrfs_insert_item(trans, extent_root, &ins, | |
1256 | &extent_item, sizeof(extent_item)); | |
1257 | clear_extent_bits(&info->extent_ins, start, end, EXTENT_LOCKED, | |
1258 | GFP_NOFS); | |
d8d5f3e1 CM |
1259 | eb = read_tree_block(extent_root, ins.objectid, ins.offset); |
1260 | level = btrfs_header_level(eb); | |
1261 | if (level == 0) { | |
1262 | btrfs_item_key(eb, &first, 0); | |
1263 | } else { | |
1264 | btrfs_node_key(eb, &first, 0); | |
1265 | } | |
7bb86316 CM |
1266 | err = btrfs_insert_extent_backref(trans, extent_root, path, |
1267 | start, extent_root->root_key.objectid, | |
f6dbff55 CM |
1268 | 0, level, |
1269 | btrfs_disk_key_objectid(&first)); | |
7bb86316 | 1270 | BUG_ON(err); |
d8d5f3e1 | 1271 | free_extent_buffer(eb); |
037e6390 | 1272 | } |
7bb86316 | 1273 | btrfs_free_path(path); |
037e6390 CM |
1274 | return 0; |
1275 | } | |
1276 | ||
db94535d CM |
1277 | static int pin_down_bytes(struct btrfs_root *root, u64 bytenr, u32 num_bytes, |
1278 | int pending) | |
e20d96d6 | 1279 | { |
1a5bc167 | 1280 | int err = 0; |
5f39d397 | 1281 | struct extent_buffer *buf; |
8ef97622 | 1282 | |
f4b9aa8d | 1283 | if (!pending) { |
db94535d | 1284 | buf = btrfs_find_tree_block(root, bytenr, num_bytes); |
5f39d397 CM |
1285 | if (buf) { |
1286 | if (btrfs_buffer_uptodate(buf)) { | |
2c90e5d6 CM |
1287 | u64 transid = |
1288 | root->fs_info->running_transaction->transid; | |
dc17ff8f CM |
1289 | u64 header_transid = |
1290 | btrfs_header_generation(buf); | |
6bc34676 CM |
1291 | if (header_transid == transid && |
1292 | !btrfs_header_flag(buf, | |
1293 | BTRFS_HEADER_FLAG_WRITTEN)) { | |
55c69072 | 1294 | clean_tree_block(NULL, root, buf); |
5f39d397 | 1295 | free_extent_buffer(buf); |
c549228f | 1296 | return 1; |
2c90e5d6 | 1297 | } |
f4b9aa8d | 1298 | } |
5f39d397 | 1299 | free_extent_buffer(buf); |
8ef97622 | 1300 | } |
324ae4df | 1301 | update_pinned_extents(root, bytenr, num_bytes, 1); |
f4b9aa8d | 1302 | } else { |
1a5bc167 | 1303 | set_extent_bits(&root->fs_info->pending_del, |
db94535d CM |
1304 | bytenr, bytenr + num_bytes - 1, |
1305 | EXTENT_LOCKED, GFP_NOFS); | |
f4b9aa8d | 1306 | } |
be744175 | 1307 | BUG_ON(err < 0); |
e20d96d6 CM |
1308 | return 0; |
1309 | } | |
1310 | ||
fec577fb | 1311 | /* |
a28ec197 | 1312 | * remove an extent from the root, returns 0 on success |
fec577fb | 1313 | */ |
e089f05c | 1314 | static int __free_extent(struct btrfs_trans_handle *trans, struct btrfs_root |
7bb86316 CM |
1315 | *root, u64 bytenr, u64 num_bytes, |
1316 | u64 root_objectid, u64 ref_generation, | |
1317 | u64 owner_objectid, u64 owner_offset, int pin, | |
e37c9e69 | 1318 | int mark_free) |
a28ec197 | 1319 | { |
5caf2a00 | 1320 | struct btrfs_path *path; |
e2fa7227 | 1321 | struct btrfs_key key; |
1261ec42 CM |
1322 | struct btrfs_fs_info *info = root->fs_info; |
1323 | struct btrfs_root *extent_root = info->extent_root; | |
5f39d397 | 1324 | struct extent_buffer *leaf; |
a28ec197 | 1325 | int ret; |
952fccac CM |
1326 | int extent_slot = 0; |
1327 | int found_extent = 0; | |
1328 | int num_to_del = 1; | |
234b63a0 | 1329 | struct btrfs_extent_item *ei; |
cf27e1ee | 1330 | u32 refs; |
037e6390 | 1331 | |
db94535d | 1332 | key.objectid = bytenr; |
62e2749e | 1333 | btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY); |
db94535d | 1334 | key.offset = num_bytes; |
5caf2a00 | 1335 | path = btrfs_alloc_path(); |
54aa1f4d CM |
1336 | if (!path) |
1337 | return -ENOMEM; | |
5f26f772 | 1338 | |
b0331a4c | 1339 | path->reada = 0; |
7bb86316 CM |
1340 | ret = lookup_extent_backref(trans, extent_root, path, |
1341 | bytenr, root_objectid, | |
1342 | ref_generation, | |
1343 | owner_objectid, owner_offset, 1); | |
1344 | if (ret == 0) { | |
952fccac CM |
1345 | struct btrfs_key found_key; |
1346 | extent_slot = path->slots[0]; | |
1347 | while(extent_slot > 0) { | |
1348 | extent_slot--; | |
1349 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, | |
1350 | extent_slot); | |
1351 | if (found_key.objectid != bytenr) | |
1352 | break; | |
1353 | if (found_key.type == BTRFS_EXTENT_ITEM_KEY && | |
1354 | found_key.offset == num_bytes) { | |
1355 | found_extent = 1; | |
1356 | break; | |
1357 | } | |
1358 | if (path->slots[0] - extent_slot > 5) | |
1359 | break; | |
1360 | } | |
1361 | if (!found_extent) | |
1362 | ret = btrfs_del_item(trans, extent_root, path); | |
7bb86316 CM |
1363 | } else { |
1364 | btrfs_print_leaf(extent_root, path->nodes[0]); | |
1365 | WARN_ON(1); | |
1366 | printk("Unable to find ref byte nr %Lu root %Lu " | |
1367 | " gen %Lu owner %Lu offset %Lu\n", bytenr, | |
1368 | root_objectid, ref_generation, owner_objectid, | |
1369 | owner_offset); | |
1370 | } | |
952fccac CM |
1371 | if (!found_extent) { |
1372 | btrfs_release_path(extent_root, path); | |
1373 | ret = btrfs_search_slot(trans, extent_root, &key, path, -1, 1); | |
1374 | if (ret < 0) | |
1375 | return ret; | |
1376 | BUG_ON(ret); | |
1377 | extent_slot = path->slots[0]; | |
1378 | } | |
5f39d397 CM |
1379 | |
1380 | leaf = path->nodes[0]; | |
952fccac | 1381 | ei = btrfs_item_ptr(leaf, extent_slot, |
123abc88 | 1382 | struct btrfs_extent_item); |
5f39d397 CM |
1383 | refs = btrfs_extent_refs(leaf, ei); |
1384 | BUG_ON(refs == 0); | |
1385 | refs -= 1; | |
1386 | btrfs_set_extent_refs(leaf, ei, refs); | |
952fccac | 1387 | |
5f39d397 CM |
1388 | btrfs_mark_buffer_dirty(leaf); |
1389 | ||
952fccac CM |
1390 | if (refs == 0 && found_extent && path->slots[0] == extent_slot + 1) { |
1391 | /* if the back ref and the extent are next to each other | |
1392 | * they get deleted below in one shot | |
1393 | */ | |
1394 | path->slots[0] = extent_slot; | |
1395 | num_to_del = 2; | |
1396 | } else if (found_extent) { | |
1397 | /* otherwise delete the extent back ref */ | |
1398 | ret = btrfs_del_item(trans, extent_root, path); | |
1399 | BUG_ON(ret); | |
1400 | /* if refs are 0, we need to setup the path for deletion */ | |
1401 | if (refs == 0) { | |
1402 | btrfs_release_path(extent_root, path); | |
1403 | ret = btrfs_search_slot(trans, extent_root, &key, path, | |
1404 | -1, 1); | |
1405 | if (ret < 0) | |
1406 | return ret; | |
1407 | BUG_ON(ret); | |
1408 | } | |
1409 | } | |
1410 | ||
cf27e1ee | 1411 | if (refs == 0) { |
db94535d CM |
1412 | u64 super_used; |
1413 | u64 root_used; | |
78fae27e CM |
1414 | |
1415 | if (pin) { | |
db94535d | 1416 | ret = pin_down_bytes(root, bytenr, num_bytes, 0); |
c549228f Y |
1417 | if (ret > 0) |
1418 | mark_free = 1; | |
1419 | BUG_ON(ret < 0); | |
78fae27e CM |
1420 | } |
1421 | ||
58176a96 | 1422 | /* block accounting for super block */ |
db94535d CM |
1423 | super_used = btrfs_super_bytes_used(&info->super_copy); |
1424 | btrfs_set_super_bytes_used(&info->super_copy, | |
1425 | super_used - num_bytes); | |
58176a96 JB |
1426 | |
1427 | /* block accounting for root item */ | |
db94535d | 1428 | root_used = btrfs_root_used(&root->root_item); |
5f39d397 | 1429 | btrfs_set_root_used(&root->root_item, |
db94535d | 1430 | root_used - num_bytes); |
952fccac CM |
1431 | ret = btrfs_del_items(trans, extent_root, path, path->slots[0], |
1432 | num_to_del); | |
54aa1f4d CM |
1433 | if (ret) { |
1434 | return ret; | |
1435 | } | |
db94535d | 1436 | ret = update_block_group(trans, root, bytenr, num_bytes, 0, |
0b86a832 | 1437 | mark_free); |
9078a3e1 | 1438 | BUG_ON(ret); |
a28ec197 | 1439 | } |
5caf2a00 | 1440 | btrfs_free_path(path); |
e089f05c | 1441 | finish_current_insert(trans, extent_root); |
a28ec197 CM |
1442 | return ret; |
1443 | } | |
1444 | ||
a28ec197 CM |
1445 | /* |
1446 | * find all the blocks marked as pending in the radix tree and remove | |
1447 | * them from the extent map | |
1448 | */ | |
e089f05c CM |
1449 | static int del_pending_extents(struct btrfs_trans_handle *trans, struct |
1450 | btrfs_root *extent_root) | |
a28ec197 CM |
1451 | { |
1452 | int ret; | |
e20d96d6 | 1453 | int err = 0; |
1a5bc167 CM |
1454 | u64 start; |
1455 | u64 end; | |
d1310b2e CM |
1456 | struct extent_io_tree *pending_del; |
1457 | struct extent_io_tree *pinned_extents; | |
8ef97622 | 1458 | |
1a5bc167 CM |
1459 | pending_del = &extent_root->fs_info->pending_del; |
1460 | pinned_extents = &extent_root->fs_info->pinned_extents; | |
a28ec197 CM |
1461 | |
1462 | while(1) { | |
1a5bc167 CM |
1463 | ret = find_first_extent_bit(pending_del, 0, &start, &end, |
1464 | EXTENT_LOCKED); | |
1465 | if (ret) | |
a28ec197 | 1466 | break; |
324ae4df | 1467 | update_pinned_extents(extent_root, start, end + 1 - start, 1); |
1a5bc167 CM |
1468 | clear_extent_bits(pending_del, start, end, EXTENT_LOCKED, |
1469 | GFP_NOFS); | |
1470 | ret = __free_extent(trans, extent_root, | |
7bb86316 CM |
1471 | start, end + 1 - start, |
1472 | extent_root->root_key.objectid, | |
1473 | 0, 0, 0, 0, 0); | |
1a5bc167 CM |
1474 | if (ret) |
1475 | err = ret; | |
fec577fb | 1476 | } |
e20d96d6 | 1477 | return err; |
fec577fb CM |
1478 | } |
1479 | ||
1480 | /* | |
1481 | * remove an extent from the root, returns 0 on success | |
1482 | */ | |
e089f05c | 1483 | int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root |
7bb86316 CM |
1484 | *root, u64 bytenr, u64 num_bytes, |
1485 | u64 root_objectid, u64 ref_generation, | |
1486 | u64 owner_objectid, u64 owner_offset, int pin) | |
fec577fb | 1487 | { |
9f5fae2f | 1488 | struct btrfs_root *extent_root = root->fs_info->extent_root; |
fec577fb CM |
1489 | int pending_ret; |
1490 | int ret; | |
a28ec197 | 1491 | |
db94535d | 1492 | WARN_ON(num_bytes < root->sectorsize); |
7bb86316 CM |
1493 | if (!root->ref_cows) |
1494 | ref_generation = 0; | |
1495 | ||
fec577fb | 1496 | if (root == extent_root) { |
db94535d | 1497 | pin_down_bytes(root, bytenr, num_bytes, 1); |
fec577fb CM |
1498 | return 0; |
1499 | } | |
7bb86316 CM |
1500 | ret = __free_extent(trans, root, bytenr, num_bytes, root_objectid, |
1501 | ref_generation, owner_objectid, owner_offset, | |
1502 | pin, pin == 0); | |
e20d96d6 | 1503 | pending_ret = del_pending_extents(trans, root->fs_info->extent_root); |
fec577fb CM |
1504 | return ret ? ret : pending_ret; |
1505 | } | |
1506 | ||
87ee04eb CM |
1507 | static u64 stripe_align(struct btrfs_root *root, u64 val) |
1508 | { | |
1509 | u64 mask = ((u64)root->stripesize - 1); | |
1510 | u64 ret = (val + mask) & ~mask; | |
1511 | return ret; | |
1512 | } | |
1513 | ||
fec577fb CM |
1514 | /* |
1515 | * walks the btree of allocated extents and find a hole of a given size. | |
1516 | * The key ins is changed to record the hole: | |
1517 | * ins->objectid == block start | |
62e2749e | 1518 | * ins->flags = BTRFS_EXTENT_ITEM_KEY |
fec577fb CM |
1519 | * ins->offset == number of blocks |
1520 | * Any available blocks before search_start are skipped. | |
1521 | */ | |
98ed5174 CM |
1522 | static int noinline find_free_extent(struct btrfs_trans_handle *trans, |
1523 | struct btrfs_root *orig_root, | |
1524 | u64 num_bytes, u64 empty_size, | |
1525 | u64 search_start, u64 search_end, | |
1526 | u64 hint_byte, struct btrfs_key *ins, | |
1527 | u64 exclude_start, u64 exclude_nr, | |
1528 | int data) | |
fec577fb | 1529 | { |
87ee04eb | 1530 | int ret; |
be744175 | 1531 | u64 orig_search_start = search_start; |
9f5fae2f | 1532 | struct btrfs_root * root = orig_root->fs_info->extent_root; |
f2458e1d | 1533 | struct btrfs_fs_info *info = root->fs_info; |
db94535d | 1534 | u64 total_needed = num_bytes; |
239b14b3 | 1535 | u64 *last_ptr = NULL; |
be08c1b9 | 1536 | struct btrfs_block_group_cache *block_group; |
be744175 | 1537 | int full_scan = 0; |
fbdc762b | 1538 | int wrapped = 0; |
239b14b3 | 1539 | int empty_cluster = 2 * 1024 * 1024; |
fec577fb | 1540 | |
db94535d | 1541 | WARN_ON(num_bytes < root->sectorsize); |
b1a4d965 CM |
1542 | btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY); |
1543 | ||
239b14b3 CM |
1544 | if (data & BTRFS_BLOCK_GROUP_METADATA) { |
1545 | last_ptr = &root->fs_info->last_alloc; | |
8790d502 | 1546 | empty_cluster = 256 * 1024; |
239b14b3 CM |
1547 | } |
1548 | ||
1549 | if ((data & BTRFS_BLOCK_GROUP_DATA) && btrfs_test_opt(root, SSD)) { | |
1550 | last_ptr = &root->fs_info->last_data_alloc; | |
1551 | } | |
1552 | ||
1553 | if (last_ptr) { | |
1554 | if (*last_ptr) | |
1555 | hint_byte = *last_ptr; | |
1556 | else { | |
1557 | empty_size += empty_cluster; | |
1558 | } | |
1559 | } | |
1560 | ||
7f93bf8d CM |
1561 | if (search_end == (u64)-1) |
1562 | search_end = btrfs_super_total_bytes(&info->super_copy); | |
0b86a832 | 1563 | |
db94535d CM |
1564 | if (hint_byte) { |
1565 | block_group = btrfs_lookup_block_group(info, hint_byte); | |
1a2b2ac7 CM |
1566 | if (!block_group) |
1567 | hint_byte = search_start; | |
be744175 | 1568 | block_group = btrfs_find_block_group(root, block_group, |
db94535d | 1569 | hint_byte, data, 1); |
239b14b3 CM |
1570 | if (last_ptr && *last_ptr == 0 && block_group) |
1571 | hint_byte = block_group->key.objectid; | |
be744175 CM |
1572 | } else { |
1573 | block_group = btrfs_find_block_group(root, | |
1a2b2ac7 CM |
1574 | trans->block_group, |
1575 | search_start, data, 1); | |
be744175 | 1576 | } |
239b14b3 | 1577 | search_start = max(search_start, hint_byte); |
be744175 | 1578 | |
6702ed49 | 1579 | total_needed += empty_size; |
0b86a832 | 1580 | |
be744175 | 1581 | check_failed: |
70b043f0 CM |
1582 | if (!block_group) { |
1583 | block_group = btrfs_lookup_block_group(info, search_start); | |
1584 | if (!block_group) | |
1585 | block_group = btrfs_lookup_block_group(info, | |
1586 | orig_search_start); | |
1587 | } | |
0b86a832 CM |
1588 | ret = find_search_start(root, &block_group, &search_start, |
1589 | total_needed, data); | |
239b14b3 CM |
1590 | if (ret == -ENOSPC && last_ptr && *last_ptr) { |
1591 | *last_ptr = 0; | |
1592 | block_group = btrfs_lookup_block_group(info, | |
1593 | orig_search_start); | |
1594 | search_start = orig_search_start; | |
1595 | ret = find_search_start(root, &block_group, &search_start, | |
1596 | total_needed, data); | |
1597 | } | |
1598 | if (ret == -ENOSPC) | |
1599 | goto enospc; | |
0b86a832 | 1600 | if (ret) |
d548ee51 | 1601 | goto error; |
e19caa5f | 1602 | |
239b14b3 CM |
1603 | if (last_ptr && *last_ptr && search_start != *last_ptr) { |
1604 | *last_ptr = 0; | |
1605 | if (!empty_size) { | |
1606 | empty_size += empty_cluster; | |
1607 | total_needed += empty_size; | |
1608 | } | |
1609 | block_group = btrfs_lookup_block_group(info, | |
1610 | orig_search_start); | |
1611 | search_start = orig_search_start; | |
1612 | ret = find_search_start(root, &block_group, | |
1613 | &search_start, total_needed, data); | |
1614 | if (ret == -ENOSPC) | |
1615 | goto enospc; | |
1616 | if (ret) | |
1617 | goto error; | |
1618 | } | |
1619 | ||
0b86a832 CM |
1620 | search_start = stripe_align(root, search_start); |
1621 | ins->objectid = search_start; | |
1622 | ins->offset = num_bytes; | |
e37c9e69 | 1623 | |
db94535d | 1624 | if (ins->objectid + num_bytes >= search_end) |
cf67582b | 1625 | goto enospc; |
0b86a832 CM |
1626 | |
1627 | if (ins->objectid + num_bytes > | |
1628 | block_group->key.objectid + block_group->key.offset) { | |
e19caa5f CM |
1629 | search_start = block_group->key.objectid + |
1630 | block_group->key.offset; | |
1631 | goto new_group; | |
1632 | } | |
0b86a832 | 1633 | |
1a5bc167 | 1634 | if (test_range_bit(&info->extent_ins, ins->objectid, |
db94535d CM |
1635 | ins->objectid + num_bytes -1, EXTENT_LOCKED, 0)) { |
1636 | search_start = ins->objectid + num_bytes; | |
1a5bc167 CM |
1637 | goto new_group; |
1638 | } | |
0b86a832 | 1639 | |
1a5bc167 | 1640 | if (test_range_bit(&info->pinned_extents, ins->objectid, |
db94535d CM |
1641 | ins->objectid + num_bytes -1, EXTENT_DIRTY, 0)) { |
1642 | search_start = ins->objectid + num_bytes; | |
1a5bc167 | 1643 | goto new_group; |
fec577fb | 1644 | } |
0b86a832 | 1645 | |
db94535d | 1646 | if (exclude_nr > 0 && (ins->objectid + num_bytes > exclude_start && |
f2654de4 CM |
1647 | ins->objectid < exclude_start + exclude_nr)) { |
1648 | search_start = exclude_start + exclude_nr; | |
1649 | goto new_group; | |
1650 | } | |
0b86a832 | 1651 | |
6324fbf3 | 1652 | if (!(data & BTRFS_BLOCK_GROUP_DATA)) { |
5276aeda | 1653 | block_group = btrfs_lookup_block_group(info, ins->objectid); |
26b8003f CM |
1654 | if (block_group) |
1655 | trans->block_group = block_group; | |
f2458e1d | 1656 | } |
db94535d | 1657 | ins->offset = num_bytes; |
239b14b3 CM |
1658 | if (last_ptr) { |
1659 | *last_ptr = ins->objectid + ins->offset; | |
1660 | if (*last_ptr == | |
1661 | btrfs_super_total_bytes(&root->fs_info->super_copy)) { | |
1662 | *last_ptr = 0; | |
1663 | } | |
1664 | } | |
fec577fb | 1665 | return 0; |
be744175 CM |
1666 | |
1667 | new_group: | |
db94535d | 1668 | if (search_start + num_bytes >= search_end) { |
cf67582b | 1669 | enospc: |
be744175 | 1670 | search_start = orig_search_start; |
fbdc762b CM |
1671 | if (full_scan) { |
1672 | ret = -ENOSPC; | |
1673 | goto error; | |
1674 | } | |
6702ed49 CM |
1675 | if (wrapped) { |
1676 | if (!full_scan) | |
1677 | total_needed -= empty_size; | |
fbdc762b | 1678 | full_scan = 1; |
6702ed49 | 1679 | } else |
fbdc762b | 1680 | wrapped = 1; |
be744175 | 1681 | } |
5276aeda | 1682 | block_group = btrfs_lookup_block_group(info, search_start); |
fbdc762b | 1683 | cond_resched(); |
1a2b2ac7 CM |
1684 | block_group = btrfs_find_block_group(root, block_group, |
1685 | search_start, data, 0); | |
be744175 CM |
1686 | goto check_failed; |
1687 | ||
0f70abe2 | 1688 | error: |
0f70abe2 | 1689 | return ret; |
fec577fb | 1690 | } |
fec577fb CM |
1691 | /* |
1692 | * finds a free extent and does all the dirty work required for allocation | |
1693 | * returns the key for the extent through ins, and a tree buffer for | |
1694 | * the first block of the extent through buf. | |
1695 | * | |
1696 | * returns 0 if everything worked, non-zero otherwise. | |
1697 | */ | |
4d775673 | 1698 | int btrfs_alloc_extent(struct btrfs_trans_handle *trans, |
7bb86316 CM |
1699 | struct btrfs_root *root, |
1700 | u64 num_bytes, u64 root_objectid, u64 ref_generation, | |
1701 | u64 owner, u64 owner_offset, | |
1702 | u64 empty_size, u64 hint_byte, | |
be08c1b9 | 1703 | u64 search_end, struct btrfs_key *ins, int data) |
fec577fb CM |
1704 | { |
1705 | int ret; | |
1706 | int pending_ret; | |
c31f8830 CM |
1707 | u64 super_used; |
1708 | u64 root_used; | |
fbdc762b | 1709 | u64 search_start = 0; |
edbd8d4e | 1710 | u64 new_hint; |
8790d502 | 1711 | u64 alloc_profile; |
47e4bb98 | 1712 | u32 sizes[2]; |
1261ec42 CM |
1713 | struct btrfs_fs_info *info = root->fs_info; |
1714 | struct btrfs_root *extent_root = info->extent_root; | |
47e4bb98 CM |
1715 | struct btrfs_extent_item *extent_item; |
1716 | struct btrfs_extent_ref *ref; | |
7bb86316 | 1717 | struct btrfs_path *path; |
47e4bb98 | 1718 | struct btrfs_key keys[2]; |
8f662a76 | 1719 | |
6324fbf3 | 1720 | if (data) { |
8790d502 CM |
1721 | alloc_profile = info->avail_data_alloc_bits & |
1722 | info->data_alloc_profile; | |
1723 | data = BTRFS_BLOCK_GROUP_DATA | alloc_profile; | |
6324fbf3 | 1724 | } else if (root == root->fs_info->chunk_root) { |
8790d502 CM |
1725 | alloc_profile = info->avail_system_alloc_bits & |
1726 | info->system_alloc_profile; | |
1727 | data = BTRFS_BLOCK_GROUP_SYSTEM | alloc_profile; | |
6324fbf3 | 1728 | } else { |
8790d502 CM |
1729 | alloc_profile = info->avail_metadata_alloc_bits & |
1730 | info->metadata_alloc_profile; | |
1731 | data = BTRFS_BLOCK_GROUP_METADATA | alloc_profile; | |
6324fbf3 CM |
1732 | } |
1733 | ||
1734 | if (root->ref_cows) { | |
593060d7 | 1735 | if (!(data & BTRFS_BLOCK_GROUP_METADATA)) { |
6324fbf3 | 1736 | ret = do_chunk_alloc(trans, root->fs_info->extent_root, |
239b14b3 | 1737 | 2 * 1024 * 1024, |
593060d7 | 1738 | BTRFS_BLOCK_GROUP_METADATA | |
8790d502 CM |
1739 | (info->metadata_alloc_profile & |
1740 | info->avail_metadata_alloc_bits)); | |
6324fbf3 CM |
1741 | BUG_ON(ret); |
1742 | } | |
1743 | ret = do_chunk_alloc(trans, root->fs_info->extent_root, | |
8790d502 | 1744 | num_bytes + 2 * 1024 * 1024, data); |
6324fbf3 CM |
1745 | BUG_ON(ret); |
1746 | } | |
0b86a832 | 1747 | |
8f662a76 | 1748 | new_hint = max(hint_byte, root->fs_info->alloc_start); |
edbd8d4e CM |
1749 | if (new_hint < btrfs_super_total_bytes(&info->super_copy)) |
1750 | hint_byte = new_hint; | |
8f662a76 | 1751 | |
db94535d CM |
1752 | WARN_ON(num_bytes < root->sectorsize); |
1753 | ret = find_free_extent(trans, root, num_bytes, empty_size, | |
1754 | search_start, search_end, hint_byte, ins, | |
26b8003f CM |
1755 | trans->alloc_exclude_start, |
1756 | trans->alloc_exclude_nr, data); | |
ccd467d6 | 1757 | BUG_ON(ret); |
f2654de4 CM |
1758 | if (ret) |
1759 | return ret; | |
fec577fb | 1760 | |
58176a96 | 1761 | /* block accounting for super block */ |
db94535d CM |
1762 | super_used = btrfs_super_bytes_used(&info->super_copy); |
1763 | btrfs_set_super_bytes_used(&info->super_copy, super_used + num_bytes); | |
26b8003f | 1764 | |
58176a96 | 1765 | /* block accounting for root item */ |
db94535d CM |
1766 | root_used = btrfs_root_used(&root->root_item); |
1767 | btrfs_set_root_used(&root->root_item, root_used + num_bytes); | |
58176a96 | 1768 | |
f510cfec CM |
1769 | clear_extent_dirty(&root->fs_info->free_space_cache, |
1770 | ins->objectid, ins->objectid + ins->offset - 1, | |
1771 | GFP_NOFS); | |
1772 | ||
26b8003f | 1773 | if (root == extent_root) { |
1a5bc167 CM |
1774 | set_extent_bits(&root->fs_info->extent_ins, ins->objectid, |
1775 | ins->objectid + ins->offset - 1, | |
1776 | EXTENT_LOCKED, GFP_NOFS); | |
26b8003f CM |
1777 | goto update_block; |
1778 | } | |
1779 | ||
1780 | WARN_ON(trans->alloc_exclude_nr); | |
1781 | trans->alloc_exclude_start = ins->objectid; | |
1782 | trans->alloc_exclude_nr = ins->offset; | |
037e6390 | 1783 | |
47e4bb98 CM |
1784 | memcpy(&keys[0], ins, sizeof(*ins)); |
1785 | keys[1].offset = hash_extent_ref(root_objectid, ref_generation, | |
1786 | owner, owner_offset); | |
1787 | keys[1].objectid = ins->objectid; | |
1788 | keys[1].type = BTRFS_EXTENT_REF_KEY; | |
1789 | sizes[0] = sizeof(*extent_item); | |
1790 | sizes[1] = sizeof(*ref); | |
7bb86316 CM |
1791 | |
1792 | path = btrfs_alloc_path(); | |
1793 | BUG_ON(!path); | |
47e4bb98 CM |
1794 | |
1795 | ret = btrfs_insert_empty_items(trans, extent_root, path, keys, | |
1796 | sizes, 2); | |
26b8003f | 1797 | |
ccd467d6 | 1798 | BUG_ON(ret); |
47e4bb98 CM |
1799 | extent_item = btrfs_item_ptr(path->nodes[0], path->slots[0], |
1800 | struct btrfs_extent_item); | |
1801 | btrfs_set_extent_refs(path->nodes[0], extent_item, 1); | |
1802 | ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1, | |
1803 | struct btrfs_extent_ref); | |
1804 | ||
1805 | btrfs_set_ref_root(path->nodes[0], ref, root_objectid); | |
1806 | btrfs_set_ref_generation(path->nodes[0], ref, ref_generation); | |
1807 | btrfs_set_ref_objectid(path->nodes[0], ref, owner); | |
1808 | btrfs_set_ref_offset(path->nodes[0], ref, owner_offset); | |
1809 | ||
1810 | btrfs_mark_buffer_dirty(path->nodes[0]); | |
1811 | ||
1812 | trans->alloc_exclude_start = 0; | |
1813 | trans->alloc_exclude_nr = 0; | |
7bb86316 | 1814 | btrfs_free_path(path); |
e089f05c | 1815 | finish_current_insert(trans, extent_root); |
e20d96d6 | 1816 | pending_ret = del_pending_extents(trans, extent_root); |
f510cfec | 1817 | |
e37c9e69 | 1818 | if (ret) { |
037e6390 | 1819 | return ret; |
e37c9e69 CM |
1820 | } |
1821 | if (pending_ret) { | |
037e6390 | 1822 | return pending_ret; |
e37c9e69 | 1823 | } |
26b8003f CM |
1824 | |
1825 | update_block: | |
0b86a832 | 1826 | ret = update_block_group(trans, root, ins->objectid, ins->offset, 1, 0); |
f5947066 CM |
1827 | if (ret) { |
1828 | printk("update block group failed for %Lu %Lu\n", | |
1829 | ins->objectid, ins->offset); | |
1830 | BUG(); | |
1831 | } | |
037e6390 | 1832 | return 0; |
fec577fb CM |
1833 | } |
1834 | ||
1835 | /* | |
1836 | * helper function to allocate a block for a given tree | |
1837 | * returns the tree buffer or NULL. | |
1838 | */ | |
5f39d397 | 1839 | struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans, |
db94535d | 1840 | struct btrfs_root *root, |
7bb86316 CM |
1841 | u32 blocksize, |
1842 | u64 root_objectid, u64 hint, | |
1843 | u64 empty_size) | |
1844 | { | |
1845 | u64 ref_generation; | |
1846 | ||
1847 | if (root->ref_cows) | |
1848 | ref_generation = trans->transid; | |
1849 | else | |
1850 | ref_generation = 0; | |
1851 | ||
1852 | ||
1853 | return __btrfs_alloc_free_block(trans, root, blocksize, root_objectid, | |
1854 | ref_generation, 0, 0, hint, empty_size); | |
1855 | } | |
1856 | ||
1857 | /* | |
1858 | * helper function to allocate a block for a given tree | |
1859 | * returns the tree buffer or NULL. | |
1860 | */ | |
1861 | struct extent_buffer *__btrfs_alloc_free_block(struct btrfs_trans_handle *trans, | |
1862 | struct btrfs_root *root, | |
1863 | u32 blocksize, | |
1864 | u64 root_objectid, | |
1865 | u64 ref_generation, | |
1866 | u64 first_objectid, | |
1867 | int level, | |
1868 | u64 hint, | |
5f39d397 | 1869 | u64 empty_size) |
fec577fb | 1870 | { |
e2fa7227 | 1871 | struct btrfs_key ins; |
fec577fb | 1872 | int ret; |
5f39d397 | 1873 | struct extent_buffer *buf; |
fec577fb | 1874 | |
7bb86316 CM |
1875 | ret = btrfs_alloc_extent(trans, root, blocksize, |
1876 | root_objectid, ref_generation, | |
f6dbff55 | 1877 | level, first_objectid, empty_size, hint, |
db94535d | 1878 | (u64)-1, &ins, 0); |
fec577fb | 1879 | if (ret) { |
54aa1f4d CM |
1880 | BUG_ON(ret > 0); |
1881 | return ERR_PTR(ret); | |
fec577fb | 1882 | } |
db94535d | 1883 | buf = btrfs_find_create_tree_block(root, ins.objectid, blocksize); |
54aa1f4d | 1884 | if (!buf) { |
7bb86316 CM |
1885 | btrfs_free_extent(trans, root, ins.objectid, blocksize, |
1886 | root->root_key.objectid, ref_generation, | |
1887 | 0, 0, 0); | |
54aa1f4d CM |
1888 | return ERR_PTR(-ENOMEM); |
1889 | } | |
55c69072 CM |
1890 | btrfs_set_header_generation(buf, trans->transid); |
1891 | clean_tree_block(trans, root, buf); | |
1892 | wait_on_tree_block_writeback(root, buf); | |
5f39d397 | 1893 | btrfs_set_buffer_uptodate(buf); |
55c69072 CM |
1894 | |
1895 | if (PageDirty(buf->first_page)) { | |
1896 | printk("page %lu dirty\n", buf->first_page->index); | |
1897 | WARN_ON(1); | |
1898 | } | |
1899 | ||
5f39d397 CM |
1900 | set_extent_dirty(&trans->transaction->dirty_pages, buf->start, |
1901 | buf->start + buf->len - 1, GFP_NOFS); | |
d1310b2e | 1902 | set_extent_bits(&BTRFS_I(root->fs_info->btree_inode)->io_tree, |
19c00ddc CM |
1903 | buf->start, buf->start + buf->len - 1, |
1904 | EXTENT_CSUM, GFP_NOFS); | |
1905 | buf->flags |= EXTENT_CSUM; | |
9afbb0b7 CM |
1906 | if (!btrfs_test_opt(root, SSD)) |
1907 | btrfs_set_buffer_defrag(buf); | |
d3c2fdcf | 1908 | trans->blocks_used++; |
fec577fb CM |
1909 | return buf; |
1910 | } | |
a28ec197 | 1911 | |
98ed5174 CM |
1912 | static int noinline drop_leaf_ref(struct btrfs_trans_handle *trans, |
1913 | struct btrfs_root *root, | |
1914 | struct extent_buffer *leaf) | |
6407bf6d | 1915 | { |
7bb86316 CM |
1916 | u64 leaf_owner; |
1917 | u64 leaf_generation; | |
5f39d397 | 1918 | struct btrfs_key key; |
6407bf6d CM |
1919 | struct btrfs_file_extent_item *fi; |
1920 | int i; | |
1921 | int nritems; | |
1922 | int ret; | |
1923 | ||
5f39d397 CM |
1924 | BUG_ON(!btrfs_is_leaf(leaf)); |
1925 | nritems = btrfs_header_nritems(leaf); | |
7bb86316 CM |
1926 | leaf_owner = btrfs_header_owner(leaf); |
1927 | leaf_generation = btrfs_header_generation(leaf); | |
1928 | ||
6407bf6d | 1929 | for (i = 0; i < nritems; i++) { |
db94535d | 1930 | u64 disk_bytenr; |
5f39d397 CM |
1931 | |
1932 | btrfs_item_key_to_cpu(leaf, &key, i); | |
1933 | if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY) | |
6407bf6d CM |
1934 | continue; |
1935 | fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item); | |
5f39d397 CM |
1936 | if (btrfs_file_extent_type(leaf, fi) == |
1937 | BTRFS_FILE_EXTENT_INLINE) | |
236454df | 1938 | continue; |
6407bf6d CM |
1939 | /* |
1940 | * FIXME make sure to insert a trans record that | |
1941 | * repeats the snapshot del on crash | |
1942 | */ | |
db94535d CM |
1943 | disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); |
1944 | if (disk_bytenr == 0) | |
3a686375 | 1945 | continue; |
db94535d | 1946 | ret = btrfs_free_extent(trans, root, disk_bytenr, |
7bb86316 CM |
1947 | btrfs_file_extent_disk_num_bytes(leaf, fi), |
1948 | leaf_owner, leaf_generation, | |
1949 | key.objectid, key.offset, 0); | |
6407bf6d CM |
1950 | BUG_ON(ret); |
1951 | } | |
1952 | return 0; | |
1953 | } | |
1954 | ||
98ed5174 | 1955 | static void noinline reada_walk_down(struct btrfs_root *root, |
bea495e5 CM |
1956 | struct extent_buffer *node, |
1957 | int slot) | |
e011599b | 1958 | { |
db94535d | 1959 | u64 bytenr; |
bea495e5 CM |
1960 | u64 last = 0; |
1961 | u32 nritems; | |
e011599b | 1962 | u32 refs; |
db94535d | 1963 | u32 blocksize; |
bea495e5 CM |
1964 | int ret; |
1965 | int i; | |
1966 | int level; | |
1967 | int skipped = 0; | |
e011599b | 1968 | |
5f39d397 | 1969 | nritems = btrfs_header_nritems(node); |
db94535d | 1970 | level = btrfs_header_level(node); |
bea495e5 CM |
1971 | if (level) |
1972 | return; | |
1973 | ||
1974 | for (i = slot; i < nritems && skipped < 32; i++) { | |
db94535d | 1975 | bytenr = btrfs_node_blockptr(node, i); |
bea495e5 CM |
1976 | if (last && ((bytenr > last && bytenr - last > 32 * 1024) || |
1977 | (last > bytenr && last - bytenr > 32 * 1024))) { | |
1978 | skipped++; | |
e011599b | 1979 | continue; |
bea495e5 CM |
1980 | } |
1981 | blocksize = btrfs_level_size(root, level - 1); | |
1982 | if (i != slot) { | |
1983 | ret = lookup_extent_ref(NULL, root, bytenr, | |
1984 | blocksize, &refs); | |
1985 | BUG_ON(ret); | |
1986 | if (refs != 1) { | |
1987 | skipped++; | |
1988 | continue; | |
1989 | } | |
1990 | } | |
409eb95d | 1991 | mutex_unlock(&root->fs_info->fs_mutex); |
db94535d | 1992 | ret = readahead_tree_block(root, bytenr, blocksize); |
bea495e5 | 1993 | last = bytenr + blocksize; |
409eb95d CM |
1994 | cond_resched(); |
1995 | mutex_lock(&root->fs_info->fs_mutex); | |
e011599b CM |
1996 | if (ret) |
1997 | break; | |
1998 | } | |
1999 | } | |
2000 | ||
9aca1d51 CM |
2001 | /* |
2002 | * helper function for drop_snapshot, this walks down the tree dropping ref | |
2003 | * counts as it goes. | |
2004 | */ | |
98ed5174 CM |
2005 | static int noinline walk_down_tree(struct btrfs_trans_handle *trans, |
2006 | struct btrfs_root *root, | |
2007 | struct btrfs_path *path, int *level) | |
20524f02 | 2008 | { |
7bb86316 CM |
2009 | u64 root_owner; |
2010 | u64 root_gen; | |
2011 | u64 bytenr; | |
5f39d397 CM |
2012 | struct extent_buffer *next; |
2013 | struct extent_buffer *cur; | |
7bb86316 | 2014 | struct extent_buffer *parent; |
db94535d | 2015 | u32 blocksize; |
20524f02 CM |
2016 | int ret; |
2017 | u32 refs; | |
2018 | ||
5caf2a00 CM |
2019 | WARN_ON(*level < 0); |
2020 | WARN_ON(*level >= BTRFS_MAX_LEVEL); | |
5f39d397 | 2021 | ret = lookup_extent_ref(trans, root, |
db94535d CM |
2022 | path->nodes[*level]->start, |
2023 | path->nodes[*level]->len, &refs); | |
20524f02 CM |
2024 | BUG_ON(ret); |
2025 | if (refs > 1) | |
2026 | goto out; | |
e011599b | 2027 | |
9aca1d51 CM |
2028 | /* |
2029 | * walk down to the last node level and free all the leaves | |
2030 | */ | |
6407bf6d | 2031 | while(*level >= 0) { |
5caf2a00 CM |
2032 | WARN_ON(*level < 0); |
2033 | WARN_ON(*level >= BTRFS_MAX_LEVEL); | |
20524f02 | 2034 | cur = path->nodes[*level]; |
e011599b | 2035 | |
5f39d397 | 2036 | if (btrfs_header_level(cur) != *level) |
2c90e5d6 | 2037 | WARN_ON(1); |
e011599b | 2038 | |
7518a238 | 2039 | if (path->slots[*level] >= |
5f39d397 | 2040 | btrfs_header_nritems(cur)) |
20524f02 | 2041 | break; |
6407bf6d CM |
2042 | if (*level == 0) { |
2043 | ret = drop_leaf_ref(trans, root, cur); | |
2044 | BUG_ON(ret); | |
2045 | break; | |
2046 | } | |
db94535d CM |
2047 | bytenr = btrfs_node_blockptr(cur, path->slots[*level]); |
2048 | blocksize = btrfs_level_size(root, *level - 1); | |
2049 | ret = lookup_extent_ref(trans, root, bytenr, blocksize, &refs); | |
6407bf6d CM |
2050 | BUG_ON(ret); |
2051 | if (refs != 1) { | |
7bb86316 CM |
2052 | parent = path->nodes[*level]; |
2053 | root_owner = btrfs_header_owner(parent); | |
2054 | root_gen = btrfs_header_generation(parent); | |
20524f02 | 2055 | path->slots[*level]++; |
db94535d | 2056 | ret = btrfs_free_extent(trans, root, bytenr, |
7bb86316 CM |
2057 | blocksize, root_owner, |
2058 | root_gen, 0, 0, 1); | |
20524f02 CM |
2059 | BUG_ON(ret); |
2060 | continue; | |
2061 | } | |
db94535d | 2062 | next = btrfs_find_tree_block(root, bytenr, blocksize); |
5f39d397 CM |
2063 | if (!next || !btrfs_buffer_uptodate(next)) { |
2064 | free_extent_buffer(next); | |
bea495e5 | 2065 | reada_walk_down(root, cur, path->slots[*level]); |
8790d502 CM |
2066 | |
2067 | mutex_unlock(&root->fs_info->fs_mutex); | |
db94535d | 2068 | next = read_tree_block(root, bytenr, blocksize); |
8790d502 | 2069 | mutex_lock(&root->fs_info->fs_mutex); |
e9d0b13b | 2070 | |
8790d502 | 2071 | /* we've dropped the lock, double check */ |
db94535d CM |
2072 | ret = lookup_extent_ref(trans, root, bytenr, |
2073 | blocksize, &refs); | |
e9d0b13b CM |
2074 | BUG_ON(ret); |
2075 | if (refs != 1) { | |
7bb86316 CM |
2076 | parent = path->nodes[*level]; |
2077 | root_owner = btrfs_header_owner(parent); | |
2078 | root_gen = btrfs_header_generation(parent); | |
2079 | ||
e9d0b13b | 2080 | path->slots[*level]++; |
5f39d397 | 2081 | free_extent_buffer(next); |
7bb86316 CM |
2082 | ret = btrfs_free_extent(trans, root, bytenr, |
2083 | blocksize, | |
2084 | root_owner, | |
2085 | root_gen, 0, 0, 1); | |
e9d0b13b CM |
2086 | BUG_ON(ret); |
2087 | continue; | |
2088 | } | |
0999df54 CM |
2089 | } else if (next) { |
2090 | btrfs_verify_block_csum(root, next); | |
e9d0b13b | 2091 | } |
5caf2a00 | 2092 | WARN_ON(*level <= 0); |
83e15a28 | 2093 | if (path->nodes[*level-1]) |
5f39d397 | 2094 | free_extent_buffer(path->nodes[*level-1]); |
20524f02 | 2095 | path->nodes[*level-1] = next; |
5f39d397 | 2096 | *level = btrfs_header_level(next); |
20524f02 CM |
2097 | path->slots[*level] = 0; |
2098 | } | |
2099 | out: | |
5caf2a00 CM |
2100 | WARN_ON(*level < 0); |
2101 | WARN_ON(*level >= BTRFS_MAX_LEVEL); | |
7bb86316 CM |
2102 | |
2103 | if (path->nodes[*level] == root->node) { | |
2104 | root_owner = root->root_key.objectid; | |
2105 | parent = path->nodes[*level]; | |
2106 | } else { | |
2107 | parent = path->nodes[*level + 1]; | |
2108 | root_owner = btrfs_header_owner(parent); | |
2109 | } | |
2110 | ||
2111 | root_gen = btrfs_header_generation(parent); | |
db94535d | 2112 | ret = btrfs_free_extent(trans, root, path->nodes[*level]->start, |
7bb86316 CM |
2113 | path->nodes[*level]->len, |
2114 | root_owner, root_gen, 0, 0, 1); | |
5f39d397 | 2115 | free_extent_buffer(path->nodes[*level]); |
20524f02 CM |
2116 | path->nodes[*level] = NULL; |
2117 | *level += 1; | |
2118 | BUG_ON(ret); | |
2119 | return 0; | |
2120 | } | |
2121 | ||
9aca1d51 CM |
2122 | /* |
2123 | * helper for dropping snapshots. This walks back up the tree in the path | |
2124 | * to find the first node higher up where we haven't yet gone through | |
2125 | * all the slots | |
2126 | */ | |
98ed5174 CM |
2127 | static int noinline walk_up_tree(struct btrfs_trans_handle *trans, |
2128 | struct btrfs_root *root, | |
2129 | struct btrfs_path *path, int *level) | |
20524f02 | 2130 | { |
7bb86316 CM |
2131 | u64 root_owner; |
2132 | u64 root_gen; | |
2133 | struct btrfs_root_item *root_item = &root->root_item; | |
20524f02 CM |
2134 | int i; |
2135 | int slot; | |
2136 | int ret; | |
9f3a7427 | 2137 | |
234b63a0 | 2138 | for(i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) { |
20524f02 | 2139 | slot = path->slots[i]; |
5f39d397 CM |
2140 | if (slot < btrfs_header_nritems(path->nodes[i]) - 1) { |
2141 | struct extent_buffer *node; | |
2142 | struct btrfs_disk_key disk_key; | |
2143 | node = path->nodes[i]; | |
20524f02 CM |
2144 | path->slots[i]++; |
2145 | *level = i; | |
9f3a7427 | 2146 | WARN_ON(*level == 0); |
5f39d397 | 2147 | btrfs_node_key(node, &disk_key, path->slots[i]); |
9f3a7427 | 2148 | memcpy(&root_item->drop_progress, |
5f39d397 | 2149 | &disk_key, sizeof(disk_key)); |
9f3a7427 | 2150 | root_item->drop_level = i; |
20524f02 CM |
2151 | return 0; |
2152 | } else { | |
7bb86316 CM |
2153 | if (path->nodes[*level] == root->node) { |
2154 | root_owner = root->root_key.objectid; | |
2155 | root_gen = | |
2156 | btrfs_header_generation(path->nodes[*level]); | |
2157 | } else { | |
2158 | struct extent_buffer *node; | |
2159 | node = path->nodes[*level + 1]; | |
2160 | root_owner = btrfs_header_owner(node); | |
2161 | root_gen = btrfs_header_generation(node); | |
2162 | } | |
e089f05c | 2163 | ret = btrfs_free_extent(trans, root, |
db94535d | 2164 | path->nodes[*level]->start, |
7bb86316 CM |
2165 | path->nodes[*level]->len, |
2166 | root_owner, root_gen, 0, 0, 1); | |
6407bf6d | 2167 | BUG_ON(ret); |
5f39d397 | 2168 | free_extent_buffer(path->nodes[*level]); |
83e15a28 | 2169 | path->nodes[*level] = NULL; |
20524f02 | 2170 | *level = i + 1; |
20524f02 CM |
2171 | } |
2172 | } | |
2173 | return 1; | |
2174 | } | |
2175 | ||
9aca1d51 CM |
2176 | /* |
2177 | * drop the reference count on the tree rooted at 'snap'. This traverses | |
2178 | * the tree freeing any blocks that have a ref count of zero after being | |
2179 | * decremented. | |
2180 | */ | |
e089f05c | 2181 | int btrfs_drop_snapshot(struct btrfs_trans_handle *trans, struct btrfs_root |
9f3a7427 | 2182 | *root) |
20524f02 | 2183 | { |
3768f368 | 2184 | int ret = 0; |
9aca1d51 | 2185 | int wret; |
20524f02 | 2186 | int level; |
5caf2a00 | 2187 | struct btrfs_path *path; |
20524f02 CM |
2188 | int i; |
2189 | int orig_level; | |
9f3a7427 | 2190 | struct btrfs_root_item *root_item = &root->root_item; |
20524f02 | 2191 | |
5caf2a00 CM |
2192 | path = btrfs_alloc_path(); |
2193 | BUG_ON(!path); | |
20524f02 | 2194 | |
5f39d397 | 2195 | level = btrfs_header_level(root->node); |
20524f02 | 2196 | orig_level = level; |
9f3a7427 CM |
2197 | if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) { |
2198 | path->nodes[level] = root->node; | |
f510cfec | 2199 | extent_buffer_get(root->node); |
9f3a7427 CM |
2200 | path->slots[level] = 0; |
2201 | } else { | |
2202 | struct btrfs_key key; | |
5f39d397 CM |
2203 | struct btrfs_disk_key found_key; |
2204 | struct extent_buffer *node; | |
6702ed49 | 2205 | |
9f3a7427 | 2206 | btrfs_disk_key_to_cpu(&key, &root_item->drop_progress); |
6702ed49 CM |
2207 | level = root_item->drop_level; |
2208 | path->lowest_level = level; | |
9f3a7427 | 2209 | wret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
6702ed49 | 2210 | if (wret < 0) { |
9f3a7427 CM |
2211 | ret = wret; |
2212 | goto out; | |
2213 | } | |
5f39d397 CM |
2214 | node = path->nodes[level]; |
2215 | btrfs_node_key(node, &found_key, path->slots[level]); | |
2216 | WARN_ON(memcmp(&found_key, &root_item->drop_progress, | |
2217 | sizeof(found_key))); | |
9f3a7427 | 2218 | } |
20524f02 | 2219 | while(1) { |
5caf2a00 | 2220 | wret = walk_down_tree(trans, root, path, &level); |
9aca1d51 | 2221 | if (wret > 0) |
20524f02 | 2222 | break; |
9aca1d51 CM |
2223 | if (wret < 0) |
2224 | ret = wret; | |
2225 | ||
5caf2a00 | 2226 | wret = walk_up_tree(trans, root, path, &level); |
9aca1d51 | 2227 | if (wret > 0) |
20524f02 | 2228 | break; |
9aca1d51 CM |
2229 | if (wret < 0) |
2230 | ret = wret; | |
409eb95d | 2231 | ret = -EAGAIN; |
409eb95d | 2232 | break; |
20524f02 | 2233 | } |
83e15a28 | 2234 | for (i = 0; i <= orig_level; i++) { |
5caf2a00 | 2235 | if (path->nodes[i]) { |
5f39d397 | 2236 | free_extent_buffer(path->nodes[i]); |
0f82731f | 2237 | path->nodes[i] = NULL; |
83e15a28 | 2238 | } |
20524f02 | 2239 | } |
9f3a7427 | 2240 | out: |
5caf2a00 | 2241 | btrfs_free_path(path); |
9aca1d51 | 2242 | return ret; |
20524f02 | 2243 | } |
9078a3e1 | 2244 | |
96b5179d | 2245 | int btrfs_free_block_groups(struct btrfs_fs_info *info) |
9078a3e1 | 2246 | { |
96b5179d CM |
2247 | u64 start; |
2248 | u64 end; | |
b97f9203 | 2249 | u64 ptr; |
9078a3e1 | 2250 | int ret; |
9078a3e1 | 2251 | while(1) { |
96b5179d CM |
2252 | ret = find_first_extent_bit(&info->block_group_cache, 0, |
2253 | &start, &end, (unsigned int)-1); | |
2254 | if (ret) | |
9078a3e1 | 2255 | break; |
b97f9203 Y |
2256 | ret = get_state_private(&info->block_group_cache, start, &ptr); |
2257 | if (!ret) | |
2258 | kfree((void *)(unsigned long)ptr); | |
96b5179d CM |
2259 | clear_extent_bits(&info->block_group_cache, start, |
2260 | end, (unsigned int)-1, GFP_NOFS); | |
9078a3e1 | 2261 | } |
e37c9e69 | 2262 | while(1) { |
f510cfec CM |
2263 | ret = find_first_extent_bit(&info->free_space_cache, 0, |
2264 | &start, &end, EXTENT_DIRTY); | |
2265 | if (ret) | |
e37c9e69 | 2266 | break; |
f510cfec CM |
2267 | clear_extent_dirty(&info->free_space_cache, start, |
2268 | end, GFP_NOFS); | |
e37c9e69 | 2269 | } |
be744175 CM |
2270 | return 0; |
2271 | } | |
2272 | ||
98ed5174 CM |
2273 | static int noinline relocate_inode_pages(struct inode *inode, u64 start, |
2274 | u64 len) | |
edbd8d4e CM |
2275 | { |
2276 | u64 page_start; | |
2277 | u64 page_end; | |
2278 | u64 delalloc_start; | |
2279 | u64 existing_delalloc; | |
2280 | unsigned long last_index; | |
edbd8d4e CM |
2281 | unsigned long i; |
2282 | struct page *page; | |
d1310b2e | 2283 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
4313b399 CM |
2284 | struct file_ra_state *ra; |
2285 | ||
2286 | ra = kzalloc(sizeof(*ra), GFP_NOFS); | |
edbd8d4e CM |
2287 | |
2288 | mutex_lock(&inode->i_mutex); | |
4313b399 | 2289 | i = start >> PAGE_CACHE_SHIFT; |
edbd8d4e CM |
2290 | last_index = (start + len - 1) >> PAGE_CACHE_SHIFT; |
2291 | ||
4313b399 CM |
2292 | file_ra_state_init(ra, inode->i_mapping); |
2293 | btrfs_force_ra(inode->i_mapping, ra, NULL, i, last_index); | |
2294 | kfree(ra); | |
edbd8d4e | 2295 | |
4313b399 | 2296 | for (; i <= last_index; i++) { |
edbd8d4e CM |
2297 | page = grab_cache_page(inode->i_mapping, i); |
2298 | if (!page) | |
2299 | goto out_unlock; | |
2300 | if (!PageUptodate(page)) { | |
2301 | btrfs_readpage(NULL, page); | |
2302 | lock_page(page); | |
2303 | if (!PageUptodate(page)) { | |
2304 | unlock_page(page); | |
2305 | page_cache_release(page); | |
2306 | goto out_unlock; | |
2307 | } | |
2308 | } | |
2309 | page_start = (u64)page->index << PAGE_CACHE_SHIFT; | |
2310 | page_end = page_start + PAGE_CACHE_SIZE - 1; | |
2311 | ||
d1310b2e | 2312 | lock_extent(io_tree, page_start, page_end, GFP_NOFS); |
edbd8d4e CM |
2313 | |
2314 | delalloc_start = page_start; | |
d1310b2e CM |
2315 | existing_delalloc = count_range_bits(io_tree, |
2316 | &delalloc_start, page_end, | |
2317 | PAGE_CACHE_SIZE, EXTENT_DELALLOC); | |
edbd8d4e | 2318 | |
d1310b2e | 2319 | set_extent_delalloc(io_tree, page_start, |
edbd8d4e CM |
2320 | page_end, GFP_NOFS); |
2321 | ||
d1310b2e | 2322 | unlock_extent(io_tree, page_start, page_end, GFP_NOFS); |
edbd8d4e CM |
2323 | set_page_dirty(page); |
2324 | unlock_page(page); | |
2325 | page_cache_release(page); | |
2326 | } | |
2327 | ||
2328 | out_unlock: | |
2329 | mutex_unlock(&inode->i_mutex); | |
2330 | return 0; | |
2331 | } | |
2332 | ||
4313b399 CM |
2333 | /* |
2334 | * note, this releases the path | |
2335 | */ | |
98ed5174 | 2336 | static int noinline relocate_one_reference(struct btrfs_root *extent_root, |
edbd8d4e | 2337 | struct btrfs_path *path, |
4313b399 | 2338 | struct btrfs_key *extent_key) |
edbd8d4e CM |
2339 | { |
2340 | struct inode *inode; | |
2341 | struct btrfs_root *found_root; | |
4313b399 CM |
2342 | struct btrfs_key *root_location; |
2343 | struct btrfs_extent_ref *ref; | |
2344 | u64 ref_root; | |
2345 | u64 ref_gen; | |
2346 | u64 ref_objectid; | |
2347 | u64 ref_offset; | |
edbd8d4e CM |
2348 | int ret; |
2349 | ||
4313b399 CM |
2350 | ref = btrfs_item_ptr(path->nodes[0], path->slots[0], |
2351 | struct btrfs_extent_ref); | |
2352 | ref_root = btrfs_ref_root(path->nodes[0], ref); | |
2353 | ref_gen = btrfs_ref_generation(path->nodes[0], ref); | |
2354 | ref_objectid = btrfs_ref_objectid(path->nodes[0], ref); | |
2355 | ref_offset = btrfs_ref_offset(path->nodes[0], ref); | |
2356 | btrfs_release_path(extent_root, path); | |
2357 | ||
2358 | root_location = kmalloc(sizeof(*root_location), GFP_NOFS); | |
2359 | root_location->objectid = ref_root; | |
edbd8d4e | 2360 | if (ref_gen == 0) |
4313b399 | 2361 | root_location->offset = 0; |
edbd8d4e | 2362 | else |
4313b399 CM |
2363 | root_location->offset = (u64)-1; |
2364 | root_location->type = BTRFS_ROOT_ITEM_KEY; | |
edbd8d4e CM |
2365 | |
2366 | found_root = btrfs_read_fs_root_no_name(extent_root->fs_info, | |
4313b399 | 2367 | root_location); |
edbd8d4e | 2368 | BUG_ON(!found_root); |
4313b399 | 2369 | kfree(root_location); |
edbd8d4e CM |
2370 | |
2371 | if (ref_objectid >= BTRFS_FIRST_FREE_OBJECTID) { | |
2372 | mutex_unlock(&extent_root->fs_info->fs_mutex); | |
2373 | inode = btrfs_iget_locked(extent_root->fs_info->sb, | |
2374 | ref_objectid, found_root); | |
2375 | if (inode->i_state & I_NEW) { | |
2376 | /* the inode and parent dir are two different roots */ | |
2377 | BTRFS_I(inode)->root = found_root; | |
2378 | BTRFS_I(inode)->location.objectid = ref_objectid; | |
2379 | BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY; | |
2380 | BTRFS_I(inode)->location.offset = 0; | |
2381 | btrfs_read_locked_inode(inode); | |
2382 | unlock_new_inode(inode); | |
2383 | ||
2384 | } | |
2385 | /* this can happen if the reference is not against | |
2386 | * the latest version of the tree root | |
2387 | */ | |
2388 | if (is_bad_inode(inode)) { | |
2389 | mutex_lock(&extent_root->fs_info->fs_mutex); | |
2390 | goto out; | |
2391 | } | |
2392 | relocate_inode_pages(inode, ref_offset, extent_key->offset); | |
2393 | /* FIXME, data=ordered will help get rid of this */ | |
2394 | filemap_fdatawrite(inode->i_mapping); | |
2395 | iput(inode); | |
2396 | mutex_lock(&extent_root->fs_info->fs_mutex); | |
2397 | } else { | |
2398 | struct btrfs_trans_handle *trans; | |
2399 | struct btrfs_key found_key; | |
2400 | struct extent_buffer *eb; | |
2401 | int level; | |
2402 | int i; | |
2403 | ||
2404 | trans = btrfs_start_transaction(found_root, 1); | |
2405 | eb = read_tree_block(found_root, extent_key->objectid, | |
2406 | extent_key->offset); | |
2407 | level = btrfs_header_level(eb); | |
2408 | ||
2409 | if (level == 0) | |
2410 | btrfs_item_key_to_cpu(eb, &found_key, 0); | |
2411 | else | |
2412 | btrfs_node_key_to_cpu(eb, &found_key, 0); | |
2413 | ||
2414 | free_extent_buffer(eb); | |
2415 | ||
2416 | path->lowest_level = level; | |
8f662a76 | 2417 | path->reada = 2; |
edbd8d4e CM |
2418 | ret = btrfs_search_slot(trans, found_root, &found_key, path, |
2419 | 0, 1); | |
2420 | path->lowest_level = 0; | |
2421 | for (i = level; i < BTRFS_MAX_LEVEL; i++) { | |
2422 | if (!path->nodes[i]) | |
2423 | break; | |
2424 | free_extent_buffer(path->nodes[i]); | |
2425 | path->nodes[i] = NULL; | |
2426 | } | |
2427 | btrfs_release_path(found_root, path); | |
2428 | btrfs_end_transaction(trans, found_root); | |
2429 | } | |
2430 | ||
2431 | out: | |
2432 | return 0; | |
2433 | } | |
2434 | ||
98ed5174 CM |
2435 | static int noinline relocate_one_extent(struct btrfs_root *extent_root, |
2436 | struct btrfs_path *path, | |
2437 | struct btrfs_key *extent_key) | |
edbd8d4e CM |
2438 | { |
2439 | struct btrfs_key key; | |
2440 | struct btrfs_key found_key; | |
edbd8d4e | 2441 | struct extent_buffer *leaf; |
edbd8d4e CM |
2442 | u32 nritems; |
2443 | u32 item_size; | |
2444 | int ret = 0; | |
2445 | ||
2446 | key.objectid = extent_key->objectid; | |
2447 | key.type = BTRFS_EXTENT_REF_KEY; | |
2448 | key.offset = 0; | |
2449 | ||
2450 | while(1) { | |
2451 | ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0); | |
2452 | ||
edbd8d4e CM |
2453 | if (ret < 0) |
2454 | goto out; | |
2455 | ||
2456 | ret = 0; | |
2457 | leaf = path->nodes[0]; | |
2458 | nritems = btrfs_header_nritems(leaf); | |
2459 | if (path->slots[0] == nritems) | |
2460 | goto out; | |
2461 | ||
2462 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
2463 | if (found_key.objectid != extent_key->objectid) | |
2464 | break; | |
2465 | ||
2466 | if (found_key.type != BTRFS_EXTENT_REF_KEY) | |
2467 | break; | |
2468 | ||
2469 | key.offset = found_key.offset + 1; | |
2470 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
2471 | ||
4313b399 | 2472 | ret = relocate_one_reference(extent_root, path, extent_key); |
edbd8d4e CM |
2473 | if (ret) |
2474 | goto out; | |
2475 | } | |
2476 | ret = 0; | |
2477 | out: | |
2478 | btrfs_release_path(extent_root, path); | |
2479 | return ret; | |
2480 | } | |
2481 | ||
edbd8d4e CM |
2482 | int btrfs_shrink_extent_tree(struct btrfs_root *root, u64 new_size) |
2483 | { | |
2484 | struct btrfs_trans_handle *trans; | |
2485 | struct btrfs_root *tree_root = root->fs_info->tree_root; | |
2486 | struct btrfs_path *path; | |
2487 | u64 cur_byte; | |
2488 | u64 total_found; | |
edbd8d4e | 2489 | struct btrfs_fs_info *info = root->fs_info; |
d1310b2e | 2490 | struct extent_io_tree *block_group_cache; |
edbd8d4e | 2491 | struct btrfs_key key; |
73e48b27 | 2492 | struct btrfs_key found_key; |
edbd8d4e CM |
2493 | struct extent_buffer *leaf; |
2494 | u32 nritems; | |
2495 | int ret; | |
725c8463 | 2496 | int progress = 0; |
edbd8d4e CM |
2497 | |
2498 | btrfs_set_super_total_bytes(&info->super_copy, new_size); | |
c31f8830 CM |
2499 | clear_extent_dirty(&info->free_space_cache, new_size, (u64)-1, |
2500 | GFP_NOFS); | |
edbd8d4e CM |
2501 | block_group_cache = &info->block_group_cache; |
2502 | path = btrfs_alloc_path(); | |
2503 | root = root->fs_info->extent_root; | |
8f662a76 | 2504 | path->reada = 2; |
edbd8d4e CM |
2505 | |
2506 | again: | |
2507 | total_found = 0; | |
2508 | key.objectid = new_size; | |
edbd8d4e CM |
2509 | key.offset = 0; |
2510 | key.type = 0; | |
73e48b27 | 2511 | cur_byte = key.objectid; |
4313b399 | 2512 | |
73e48b27 Y |
2513 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
2514 | if (ret < 0) | |
2515 | goto out; | |
2516 | ||
0b86a832 | 2517 | ret = btrfs_previous_item(root, path, 0, BTRFS_EXTENT_ITEM_KEY); |
73e48b27 Y |
2518 | if (ret < 0) |
2519 | goto out; | |
2520 | if (ret == 0) { | |
2521 | leaf = path->nodes[0]; | |
2522 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
2523 | if (found_key.objectid + found_key.offset > new_size) { | |
2524 | cur_byte = found_key.objectid; | |
2525 | key.objectid = cur_byte; | |
2526 | } | |
2527 | } | |
2528 | btrfs_release_path(root, path); | |
2529 | ||
2530 | while(1) { | |
edbd8d4e CM |
2531 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
2532 | if (ret < 0) | |
2533 | goto out; | |
73e48b27 | 2534 | |
edbd8d4e | 2535 | leaf = path->nodes[0]; |
73e48b27 Y |
2536 | nritems = btrfs_header_nritems(leaf); |
2537 | next: | |
2538 | if (path->slots[0] >= nritems) { | |
2539 | ret = btrfs_next_leaf(root, path); | |
2540 | if (ret < 0) | |
2541 | goto out; | |
2542 | if (ret == 1) { | |
2543 | ret = 0; | |
2544 | break; | |
edbd8d4e | 2545 | } |
73e48b27 Y |
2546 | leaf = path->nodes[0]; |
2547 | nritems = btrfs_header_nritems(leaf); | |
edbd8d4e | 2548 | } |
73e48b27 Y |
2549 | |
2550 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
725c8463 CM |
2551 | |
2552 | if (progress && need_resched()) { | |
2553 | memcpy(&key, &found_key, sizeof(key)); | |
2554 | mutex_unlock(&root->fs_info->fs_mutex); | |
2555 | cond_resched(); | |
2556 | mutex_lock(&root->fs_info->fs_mutex); | |
2557 | btrfs_release_path(root, path); | |
2558 | btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
2559 | progress = 0; | |
2560 | goto next; | |
2561 | } | |
2562 | progress = 1; | |
2563 | ||
73e48b27 Y |
2564 | if (btrfs_key_type(&found_key) != BTRFS_EXTENT_ITEM_KEY || |
2565 | found_key.objectid + found_key.offset <= cur_byte) { | |
edbd8d4e | 2566 | path->slots[0]++; |
edbd8d4e CM |
2567 | goto next; |
2568 | } | |
73e48b27 | 2569 | |
edbd8d4e CM |
2570 | total_found++; |
2571 | cur_byte = found_key.objectid + found_key.offset; | |
2572 | key.objectid = cur_byte; | |
2573 | btrfs_release_path(root, path); | |
2574 | ret = relocate_one_extent(root, path, &found_key); | |
2575 | } | |
2576 | ||
2577 | btrfs_release_path(root, path); | |
2578 | ||
2579 | if (total_found > 0) { | |
2580 | trans = btrfs_start_transaction(tree_root, 1); | |
2581 | btrfs_commit_transaction(trans, tree_root); | |
2582 | ||
2583 | mutex_unlock(&root->fs_info->fs_mutex); | |
2584 | btrfs_clean_old_snapshots(tree_root); | |
2585 | mutex_lock(&root->fs_info->fs_mutex); | |
2586 | ||
2587 | trans = btrfs_start_transaction(tree_root, 1); | |
2588 | btrfs_commit_transaction(trans, tree_root); | |
2589 | goto again; | |
2590 | } | |
2591 | ||
2592 | trans = btrfs_start_transaction(root, 1); | |
2593 | key.objectid = new_size; | |
2594 | key.offset = 0; | |
2595 | key.type = 0; | |
2596 | while(1) { | |
4313b399 CM |
2597 | u64 ptr; |
2598 | ||
edbd8d4e CM |
2599 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
2600 | if (ret < 0) | |
2601 | goto out; | |
73e48b27 | 2602 | |
edbd8d4e CM |
2603 | leaf = path->nodes[0]; |
2604 | nritems = btrfs_header_nritems(leaf); | |
73e48b27 Y |
2605 | bg_next: |
2606 | if (path->slots[0] >= nritems) { | |
2607 | ret = btrfs_next_leaf(root, path); | |
2608 | if (ret < 0) | |
2609 | break; | |
2610 | if (ret == 1) { | |
2611 | ret = 0; | |
2612 | break; | |
edbd8d4e | 2613 | } |
73e48b27 | 2614 | leaf = path->nodes[0]; |
1372f8e6 CM |
2615 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
2616 | ||
2617 | /* | |
2618 | * btrfs_next_leaf doesn't cow buffers, we have to | |
2619 | * do the search again | |
2620 | */ | |
2621 | memcpy(&key, &found_key, sizeof(key)); | |
2622 | btrfs_release_path(root, path); | |
725c8463 | 2623 | goto resched_check; |
73e48b27 Y |
2624 | } |
2625 | ||
2626 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
2627 | if (btrfs_key_type(&found_key) != BTRFS_BLOCK_GROUP_ITEM_KEY) { | |
2628 | printk("shrinker found key %Lu %u %Lu\n", | |
2629 | found_key.objectid, found_key.type, | |
2630 | found_key.offset); | |
2631 | path->slots[0]++; | |
edbd8d4e CM |
2632 | goto bg_next; |
2633 | } | |
edbd8d4e CM |
2634 | ret = get_state_private(&info->block_group_cache, |
2635 | found_key.objectid, &ptr); | |
2636 | if (!ret) | |
2637 | kfree((void *)(unsigned long)ptr); | |
2638 | ||
2639 | clear_extent_bits(&info->block_group_cache, found_key.objectid, | |
2640 | found_key.objectid + found_key.offset - 1, | |
2641 | (unsigned int)-1, GFP_NOFS); | |
2642 | ||
2643 | key.objectid = found_key.objectid + 1; | |
2644 | btrfs_del_item(trans, root, path); | |
2645 | btrfs_release_path(root, path); | |
725c8463 CM |
2646 | resched_check: |
2647 | if (need_resched()) { | |
2648 | mutex_unlock(&root->fs_info->fs_mutex); | |
2649 | cond_resched(); | |
2650 | mutex_lock(&root->fs_info->fs_mutex); | |
2651 | } | |
edbd8d4e CM |
2652 | } |
2653 | clear_extent_dirty(&info->free_space_cache, new_size, (u64)-1, | |
2654 | GFP_NOFS); | |
2655 | btrfs_commit_transaction(trans, root); | |
2656 | out: | |
2657 | btrfs_free_path(path); | |
2658 | return ret; | |
2659 | } | |
2660 | ||
2661 | int btrfs_grow_extent_tree(struct btrfs_trans_handle *trans, | |
2662 | struct btrfs_root *root, u64 new_size) | |
2663 | { | |
0b86a832 CM |
2664 | btrfs_set_super_total_bytes(&root->fs_info->super_copy, new_size); |
2665 | return 0; | |
2666 | } | |
edbd8d4e | 2667 | |
0b86a832 CM |
2668 | int find_first_block_group(struct btrfs_root *root, struct btrfs_path *path, |
2669 | struct btrfs_key *key) | |
2670 | { | |
2671 | int ret; | |
2672 | struct btrfs_key found_key; | |
2673 | struct extent_buffer *leaf; | |
2674 | int slot; | |
edbd8d4e | 2675 | |
0b86a832 CM |
2676 | ret = btrfs_search_slot(NULL, root, key, path, 0, 0); |
2677 | if (ret < 0) | |
2678 | return ret; | |
2679 | while(1) { | |
2680 | slot = path->slots[0]; | |
edbd8d4e | 2681 | leaf = path->nodes[0]; |
0b86a832 CM |
2682 | if (slot >= btrfs_header_nritems(leaf)) { |
2683 | ret = btrfs_next_leaf(root, path); | |
2684 | if (ret == 0) | |
2685 | continue; | |
2686 | if (ret < 0) | |
2687 | goto error; | |
2688 | break; | |
edbd8d4e | 2689 | } |
0b86a832 | 2690 | btrfs_item_key_to_cpu(leaf, &found_key, slot); |
edbd8d4e | 2691 | |
0b86a832 CM |
2692 | if (found_key.objectid >= key->objectid && |
2693 | found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) | |
2694 | return 0; | |
2695 | path->slots[0]++; | |
edbd8d4e | 2696 | } |
0b86a832 CM |
2697 | ret = -ENOENT; |
2698 | error: | |
2699 | return ret; | |
edbd8d4e CM |
2700 | } |
2701 | ||
9078a3e1 CM |
2702 | int btrfs_read_block_groups(struct btrfs_root *root) |
2703 | { | |
2704 | struct btrfs_path *path; | |
2705 | int ret; | |
96b5179d | 2706 | int bit; |
9078a3e1 | 2707 | struct btrfs_block_group_cache *cache; |
be744175 | 2708 | struct btrfs_fs_info *info = root->fs_info; |
6324fbf3 | 2709 | struct btrfs_space_info *space_info; |
d1310b2e | 2710 | struct extent_io_tree *block_group_cache; |
9078a3e1 CM |
2711 | struct btrfs_key key; |
2712 | struct btrfs_key found_key; | |
5f39d397 | 2713 | struct extent_buffer *leaf; |
96b5179d CM |
2714 | |
2715 | block_group_cache = &info->block_group_cache; | |
be744175 | 2716 | root = info->extent_root; |
9078a3e1 | 2717 | key.objectid = 0; |
0b86a832 | 2718 | key.offset = 0; |
9078a3e1 | 2719 | btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY); |
9078a3e1 CM |
2720 | path = btrfs_alloc_path(); |
2721 | if (!path) | |
2722 | return -ENOMEM; | |
2723 | ||
2724 | while(1) { | |
0b86a832 CM |
2725 | ret = find_first_block_group(root, path, &key); |
2726 | if (ret > 0) { | |
2727 | ret = 0; | |
2728 | goto error; | |
9078a3e1 | 2729 | } |
0b86a832 CM |
2730 | if (ret != 0) |
2731 | goto error; | |
2732 | ||
5f39d397 CM |
2733 | leaf = path->nodes[0]; |
2734 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
9078a3e1 CM |
2735 | cache = kmalloc(sizeof(*cache), GFP_NOFS); |
2736 | if (!cache) { | |
0b86a832 | 2737 | ret = -ENOMEM; |
9078a3e1 CM |
2738 | break; |
2739 | } | |
3e1ad54f | 2740 | |
5f39d397 CM |
2741 | read_extent_buffer(leaf, &cache->item, |
2742 | btrfs_item_ptr_offset(leaf, path->slots[0]), | |
2743 | sizeof(cache->item)); | |
9078a3e1 | 2744 | memcpy(&cache->key, &found_key, sizeof(found_key)); |
e37c9e69 | 2745 | cache->cached = 0; |
324ae4df | 2746 | cache->pinned = 0; |
0b86a832 | 2747 | |
9078a3e1 CM |
2748 | key.objectid = found_key.objectid + found_key.offset; |
2749 | btrfs_release_path(root, path); | |
0b86a832 CM |
2750 | cache->flags = btrfs_block_group_flags(&cache->item); |
2751 | bit = 0; | |
2752 | if (cache->flags & BTRFS_BLOCK_GROUP_DATA) { | |
96b5179d | 2753 | bit = BLOCK_GROUP_DATA; |
0b86a832 CM |
2754 | } else if (cache->flags & BTRFS_BLOCK_GROUP_SYSTEM) { |
2755 | bit = BLOCK_GROUP_SYSTEM; | |
2756 | } else if (cache->flags & BTRFS_BLOCK_GROUP_METADATA) { | |
96b5179d | 2757 | bit = BLOCK_GROUP_METADATA; |
31f3c99b | 2758 | } |
8790d502 | 2759 | set_avail_alloc_bits(info, cache->flags); |
96b5179d | 2760 | |
6324fbf3 CM |
2761 | ret = update_space_info(info, cache->flags, found_key.offset, |
2762 | btrfs_block_group_used(&cache->item), | |
2763 | &space_info); | |
2764 | BUG_ON(ret); | |
2765 | cache->space_info = space_info; | |
2766 | ||
96b5179d CM |
2767 | /* use EXTENT_LOCKED to prevent merging */ |
2768 | set_extent_bits(block_group_cache, found_key.objectid, | |
2769 | found_key.objectid + found_key.offset - 1, | |
2770 | bit | EXTENT_LOCKED, GFP_NOFS); | |
2771 | set_state_private(block_group_cache, found_key.objectid, | |
ae2f5411 | 2772 | (unsigned long)cache); |
96b5179d | 2773 | |
9078a3e1 | 2774 | if (key.objectid >= |
db94535d | 2775 | btrfs_super_total_bytes(&info->super_copy)) |
9078a3e1 CM |
2776 | break; |
2777 | } | |
0b86a832 CM |
2778 | ret = 0; |
2779 | error: | |
9078a3e1 | 2780 | btrfs_free_path(path); |
0b86a832 | 2781 | return ret; |
9078a3e1 | 2782 | } |
6324fbf3 CM |
2783 | |
2784 | int btrfs_make_block_group(struct btrfs_trans_handle *trans, | |
2785 | struct btrfs_root *root, u64 bytes_used, | |
2786 | u64 type, u64 chunk_tree, u64 chunk_objectid, | |
2787 | u64 size) | |
2788 | { | |
2789 | int ret; | |
2790 | int bit = 0; | |
2791 | struct btrfs_root *extent_root; | |
2792 | struct btrfs_block_group_cache *cache; | |
2793 | struct extent_io_tree *block_group_cache; | |
2794 | ||
2795 | extent_root = root->fs_info->extent_root; | |
2796 | block_group_cache = &root->fs_info->block_group_cache; | |
2797 | ||
2798 | cache = kmalloc(sizeof(*cache), GFP_NOFS); | |
2799 | BUG_ON(!cache); | |
2800 | cache->key.objectid = chunk_objectid; | |
2801 | cache->key.offset = size; | |
2802 | cache->cached = 0; | |
2803 | cache->pinned = 0; | |
2804 | btrfs_set_key_type(&cache->key, BTRFS_BLOCK_GROUP_ITEM_KEY); | |
2805 | memset(&cache->item, 0, sizeof(cache->item)); | |
2806 | btrfs_set_block_group_used(&cache->item, bytes_used); | |
2807 | btrfs_set_block_group_chunk_tree(&cache->item, chunk_tree); | |
2808 | btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid); | |
2809 | cache->flags = type; | |
2810 | btrfs_set_block_group_flags(&cache->item, type); | |
2811 | ||
2812 | ret = update_space_info(root->fs_info, cache->flags, size, bytes_used, | |
2813 | &cache->space_info); | |
2814 | BUG_ON(ret); | |
2815 | ||
2816 | if (type & BTRFS_BLOCK_GROUP_DATA) { | |
2817 | bit = BLOCK_GROUP_DATA; | |
2818 | } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) { | |
2819 | bit = BLOCK_GROUP_SYSTEM; | |
2820 | } else if (type & BTRFS_BLOCK_GROUP_METADATA) { | |
2821 | bit = BLOCK_GROUP_METADATA; | |
2822 | } | |
2823 | set_extent_bits(block_group_cache, chunk_objectid, | |
2824 | chunk_objectid + size - 1, | |
2825 | bit | EXTENT_LOCKED, GFP_NOFS); | |
2826 | set_state_private(block_group_cache, chunk_objectid, | |
2827 | (unsigned long)cache); | |
2828 | ||
2829 | ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item, | |
2830 | sizeof(cache->item)); | |
2831 | BUG_ON(ret); | |
2832 | ||
2833 | finish_current_insert(trans, extent_root); | |
2834 | ret = del_pending_extents(trans, extent_root); | |
2835 | BUG_ON(ret); | |
2836 | return 0; | |
2837 | } |