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