]>
Commit | Line | Data |
---|---|---|
6cbd5570 CM |
1 | /* |
2 | * Copyright (C) 2007 Oracle. All rights reserved. | |
3 | * | |
4 | * This program is free software; you can redistribute it and/or | |
5 | * modify it under the terms of the GNU General Public | |
6 | * License v2 as published by the Free Software Foundation. | |
7 | * | |
8 | * This program is distributed in the hope that it will be useful, | |
9 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
10 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
11 | * General Public License for more details. | |
12 | * | |
13 | * You should have received a copy of the GNU General Public | |
14 | * License along with this program; if not, write to the | |
15 | * Free Software Foundation, Inc., 59 Temple Place - Suite 330, | |
16 | * Boston, MA 021110-1307, USA. | |
17 | */ | |
18 | ||
ec6b910f | 19 | #include <linux/sched.h> |
7bb86316 | 20 | #include <linux/crc32c.h> |
edbd8d4e | 21 | #include <linux/pagemap.h> |
74493f7a | 22 | #include "hash.h" |
fec577fb CM |
23 | #include "ctree.h" |
24 | #include "disk-io.h" | |
25 | #include "print-tree.h" | |
e089f05c | 26 | #include "transaction.h" |
fec577fb | 27 | |
96b5179d CM |
28 | #define BLOCK_GROUP_DATA EXTENT_WRITEBACK |
29 | #define BLOCK_GROUP_METADATA EXTENT_UPTODATE | |
30 | #define BLOCK_GROUP_DIRTY EXTENT_DIRTY | |
31 | ||
e089f05c CM |
32 | static int finish_current_insert(struct btrfs_trans_handle *trans, struct |
33 | btrfs_root *extent_root); | |
e20d96d6 CM |
34 | static int del_pending_extents(struct btrfs_trans_handle *trans, struct |
35 | btrfs_root *extent_root); | |
d548ee51 Y |
36 | static int find_previous_extent(struct btrfs_root *root, |
37 | struct btrfs_path *path) | |
38 | { | |
39 | struct btrfs_key found_key; | |
40 | struct extent_buffer *leaf; | |
41 | int ret; | |
42 | ||
43 | while(1) { | |
44 | if (path->slots[0] == 0) { | |
45 | ret = btrfs_prev_leaf(root, path); | |
46 | if (ret != 0) | |
47 | return ret; | |
48 | } else { | |
49 | path->slots[0]--; | |
50 | } | |
51 | leaf = path->nodes[0]; | |
52 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
53 | if (found_key.type == BTRFS_EXTENT_ITEM_KEY) | |
54 | return 0; | |
55 | } | |
56 | return 1; | |
57 | } | |
fec577fb | 58 | |
e37c9e69 CM |
59 | static int cache_block_group(struct btrfs_root *root, |
60 | struct btrfs_block_group_cache *block_group) | |
61 | { | |
62 | struct btrfs_path *path; | |
63 | int ret; | |
64 | struct btrfs_key key; | |
5f39d397 | 65 | struct extent_buffer *leaf; |
d1310b2e | 66 | struct extent_io_tree *free_space_cache; |
e37c9e69 | 67 | int slot; |
e37c9e69 CM |
68 | u64 last = 0; |
69 | u64 hole_size; | |
7d7d6068 | 70 | u64 first_free; |
e37c9e69 CM |
71 | int found = 0; |
72 | ||
00f5c795 CM |
73 | if (!block_group) |
74 | return 0; | |
75 | ||
e37c9e69 | 76 | root = root->fs_info->extent_root; |
f510cfec | 77 | free_space_cache = &root->fs_info->free_space_cache; |
e37c9e69 CM |
78 | |
79 | if (block_group->cached) | |
80 | return 0; | |
f510cfec | 81 | |
e37c9e69 CM |
82 | path = btrfs_alloc_path(); |
83 | if (!path) | |
84 | return -ENOMEM; | |
7d7d6068 | 85 | |
2cc58cf2 | 86 | path->reada = 2; |
7d7d6068 | 87 | first_free = block_group->key.objectid; |
e37c9e69 | 88 | key.objectid = block_group->key.objectid; |
e37c9e69 CM |
89 | key.offset = 0; |
90 | btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY); | |
91 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
92 | if (ret < 0) | |
93 | return ret; | |
d548ee51 Y |
94 | ret = find_previous_extent(root, path); |
95 | if (ret < 0) | |
96 | return ret; | |
97 | if (ret == 0) { | |
98 | leaf = path->nodes[0]; | |
99 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
100 | if (key.objectid + key.offset > first_free) | |
101 | first_free = key.objectid + key.offset; | |
102 | } | |
e37c9e69 | 103 | while(1) { |
5f39d397 | 104 | leaf = path->nodes[0]; |
e37c9e69 | 105 | slot = path->slots[0]; |
5f39d397 | 106 | if (slot >= btrfs_header_nritems(leaf)) { |
e37c9e69 | 107 | ret = btrfs_next_leaf(root, path); |
54aa1f4d CM |
108 | if (ret < 0) |
109 | goto err; | |
de428b63 | 110 | if (ret == 0) { |
e37c9e69 | 111 | continue; |
de428b63 | 112 | } else { |
e37c9e69 CM |
113 | break; |
114 | } | |
115 | } | |
5f39d397 | 116 | btrfs_item_key_to_cpu(leaf, &key, slot); |
7d7d6068 | 117 | if (key.objectid < block_group->key.objectid) { |
7d7d6068 Y |
118 | goto next; |
119 | } | |
e37c9e69 CM |
120 | if (key.objectid >= block_group->key.objectid + |
121 | block_group->key.offset) { | |
e37c9e69 CM |
122 | break; |
123 | } | |
7d7d6068 | 124 | |
e37c9e69 CM |
125 | if (btrfs_key_type(&key) == BTRFS_EXTENT_ITEM_KEY) { |
126 | if (!found) { | |
7d7d6068 | 127 | last = first_free; |
e37c9e69 | 128 | found = 1; |
e37c9e69 | 129 | } |
f510cfec CM |
130 | if (key.objectid > last) { |
131 | hole_size = key.objectid - last; | |
132 | set_extent_dirty(free_space_cache, last, | |
133 | last + hole_size - 1, | |
134 | GFP_NOFS); | |
7d7d6068 Y |
135 | } |
136 | last = key.objectid + key.offset; | |
e37c9e69 | 137 | } |
7d7d6068 | 138 | next: |
e37c9e69 CM |
139 | path->slots[0]++; |
140 | } | |
141 | ||
7d7d6068 Y |
142 | if (!found) |
143 | last = first_free; | |
144 | if (block_group->key.objectid + | |
145 | block_group->key.offset > last) { | |
146 | hole_size = block_group->key.objectid + | |
147 | block_group->key.offset - last; | |
f510cfec CM |
148 | set_extent_dirty(free_space_cache, last, |
149 | last + hole_size - 1, GFP_NOFS); | |
7d7d6068 | 150 | } |
e37c9e69 | 151 | block_group->cached = 1; |
54aa1f4d | 152 | err: |
e37c9e69 CM |
153 | btrfs_free_path(path); |
154 | return 0; | |
155 | } | |
156 | ||
5276aeda CM |
157 | struct btrfs_block_group_cache *btrfs_lookup_block_group(struct |
158 | btrfs_fs_info *info, | |
db94535d | 159 | u64 bytenr) |
be744175 | 160 | { |
d1310b2e | 161 | struct extent_io_tree *block_group_cache; |
96b5179d CM |
162 | struct btrfs_block_group_cache *block_group = NULL; |
163 | u64 ptr; | |
164 | u64 start; | |
165 | u64 end; | |
be744175 CM |
166 | int ret; |
167 | ||
96b5179d CM |
168 | block_group_cache = &info->block_group_cache; |
169 | ret = find_first_extent_bit(block_group_cache, | |
db94535d | 170 | bytenr, &start, &end, |
96b5179d | 171 | BLOCK_GROUP_DATA | BLOCK_GROUP_METADATA); |
be744175 | 172 | if (ret) { |
96b5179d | 173 | return NULL; |
be744175 | 174 | } |
96b5179d CM |
175 | ret = get_state_private(block_group_cache, start, &ptr); |
176 | if (ret) | |
177 | return NULL; | |
178 | ||
ae2f5411 | 179 | block_group = (struct btrfs_block_group_cache *)(unsigned long)ptr; |
5cf66426 | 180 | if (block_group->key.objectid <= bytenr && bytenr < |
96b5179d CM |
181 | block_group->key.objectid + block_group->key.offset) |
182 | return block_group; | |
be744175 CM |
183 | return NULL; |
184 | } | |
98ed5174 CM |
185 | static u64 noinline find_search_start(struct btrfs_root *root, |
186 | struct btrfs_block_group_cache **cache_ret, | |
d548ee51 | 187 | u64 search_start, int num, int data) |
e37c9e69 | 188 | { |
e37c9e69 CM |
189 | int ret; |
190 | struct btrfs_block_group_cache *cache = *cache_ret; | |
d7fc640e CM |
191 | struct extent_io_tree *free_space_cache; |
192 | struct extent_state *state; | |
e19caa5f | 193 | u64 last; |
f510cfec | 194 | u64 start = 0; |
257d0ce3 | 195 | u64 cache_miss = 0; |
c31f8830 | 196 | u64 total_fs_bytes; |
f84a8b36 | 197 | int wrapped = 0; |
e37c9e69 | 198 | |
00f5c795 | 199 | if (!cache) { |
1a2b2ac7 | 200 | goto out; |
00f5c795 | 201 | } |
c31f8830 | 202 | total_fs_bytes = btrfs_super_total_bytes(&root->fs_info->super_copy); |
d7fc640e CM |
203 | free_space_cache = &root->fs_info->free_space_cache; |
204 | ||
e37c9e69 | 205 | again: |
54aa1f4d CM |
206 | ret = cache_block_group(root, cache); |
207 | if (ret) | |
208 | goto out; | |
f84a8b36 | 209 | |
e19caa5f CM |
210 | last = max(search_start, cache->key.objectid); |
211 | ||
d7fc640e CM |
212 | spin_lock_irq(&free_space_cache->lock); |
213 | state = find_first_extent_bit_state(free_space_cache, last, EXTENT_DIRTY); | |
e37c9e69 | 214 | while(1) { |
d7fc640e | 215 | if (!state) { |
257d0ce3 CM |
216 | if (!cache_miss) |
217 | cache_miss = last; | |
d7fc640e | 218 | spin_unlock_irq(&free_space_cache->lock); |
e19caa5f CM |
219 | goto new_group; |
220 | } | |
f510cfec | 221 | |
d7fc640e CM |
222 | start = max(last, state->start); |
223 | last = state->end + 1; | |
257d0ce3 CM |
224 | if (last - start < num) { |
225 | if (last == cache->key.objectid + cache->key.offset) | |
226 | cache_miss = start; | |
d7fc640e CM |
227 | do { |
228 | state = extent_state_next(state); | |
229 | } while(state && !(state->state & EXTENT_DIRTY)); | |
f510cfec | 230 | continue; |
257d0ce3 | 231 | } |
d7fc640e | 232 | spin_unlock_irq(&free_space_cache->lock); |
257d0ce3 | 233 | if (data != BTRFS_BLOCK_GROUP_MIXED && |
5cf66426 | 234 | start + num > cache->key.objectid + cache->key.offset) |
e37c9e69 | 235 | goto new_group; |
c31f8830 CM |
236 | if (start + num > total_fs_bytes) |
237 | goto new_group; | |
f510cfec | 238 | return start; |
e37c9e69 CM |
239 | } |
240 | out: | |
1a2b2ac7 CM |
241 | cache = btrfs_lookup_block_group(root->fs_info, search_start); |
242 | if (!cache) { | |
243 | printk("Unable to find block group for %Lu\n", | |
244 | search_start); | |
245 | WARN_ON(1); | |
246 | return search_start; | |
247 | } | |
1a5bc167 | 248 | return search_start; |
e37c9e69 CM |
249 | |
250 | new_group: | |
e19caa5f | 251 | last = cache->key.objectid + cache->key.offset; |
f84a8b36 | 252 | wrapped: |
e19caa5f | 253 | cache = btrfs_lookup_block_group(root->fs_info, last); |
c31f8830 | 254 | if (!cache || cache->key.objectid >= total_fs_bytes) { |
0e4de584 | 255 | no_cache: |
f84a8b36 CM |
256 | if (!wrapped) { |
257 | wrapped = 1; | |
258 | last = search_start; | |
259 | data = BTRFS_BLOCK_GROUP_MIXED; | |
260 | goto wrapped; | |
261 | } | |
1a2b2ac7 | 262 | goto out; |
e37c9e69 | 263 | } |
257d0ce3 CM |
264 | if (cache_miss && !cache->cached) { |
265 | cache_block_group(root, cache); | |
266 | last = cache_miss; | |
257d0ce3 CM |
267 | cache = btrfs_lookup_block_group(root->fs_info, last); |
268 | } | |
1a2b2ac7 | 269 | cache = btrfs_find_block_group(root, cache, last, data, 0); |
0e4de584 CM |
270 | if (!cache) |
271 | goto no_cache; | |
e37c9e69 | 272 | *cache_ret = cache; |
257d0ce3 | 273 | cache_miss = 0; |
e37c9e69 CM |
274 | goto again; |
275 | } | |
276 | ||
84f54cfa CM |
277 | static u64 div_factor(u64 num, int factor) |
278 | { | |
257d0ce3 CM |
279 | if (factor == 10) |
280 | return num; | |
84f54cfa CM |
281 | num *= factor; |
282 | do_div(num, 10); | |
283 | return num; | |
284 | } | |
285 | ||
31f3c99b CM |
286 | struct btrfs_block_group_cache *btrfs_find_block_group(struct btrfs_root *root, |
287 | struct btrfs_block_group_cache | |
be744175 | 288 | *hint, u64 search_start, |
de428b63 | 289 | int data, int owner) |
cd1bc465 | 290 | { |
96b5179d | 291 | struct btrfs_block_group_cache *cache; |
d1310b2e | 292 | struct extent_io_tree *block_group_cache; |
31f3c99b | 293 | struct btrfs_block_group_cache *found_group = NULL; |
cd1bc465 CM |
294 | struct btrfs_fs_info *info = root->fs_info; |
295 | u64 used; | |
31f3c99b CM |
296 | u64 last = 0; |
297 | u64 hint_last; | |
96b5179d CM |
298 | u64 start; |
299 | u64 end; | |
300 | u64 free_check; | |
301 | u64 ptr; | |
c31f8830 | 302 | u64 total_fs_bytes; |
96b5179d | 303 | int bit; |
cd1bc465 | 304 | int ret; |
31f3c99b | 305 | int full_search = 0; |
de428b63 | 306 | int factor = 8; |
1e2677e0 | 307 | int data_swap = 0; |
de428b63 | 308 | |
96b5179d | 309 | block_group_cache = &info->block_group_cache; |
c31f8830 | 310 | total_fs_bytes = btrfs_super_total_bytes(&root->fs_info->super_copy); |
96b5179d | 311 | |
de428b63 | 312 | if (!owner) |
f84a8b36 | 313 | factor = 8; |
be744175 | 314 | |
257d0ce3 | 315 | if (data == BTRFS_BLOCK_GROUP_MIXED) { |
f84a8b36 | 316 | bit = BLOCK_GROUP_DATA | BLOCK_GROUP_METADATA; |
257d0ce3 CM |
317 | factor = 10; |
318 | } else if (data) | |
96b5179d CM |
319 | bit = BLOCK_GROUP_DATA; |
320 | else | |
321 | bit = BLOCK_GROUP_METADATA; | |
be744175 | 322 | |
c31f8830 | 323 | if (search_start && search_start < total_fs_bytes) { |
be744175 | 324 | struct btrfs_block_group_cache *shint; |
5276aeda | 325 | shint = btrfs_lookup_block_group(info, search_start); |
f84a8b36 CM |
326 | if (shint && (shint->data == data || |
327 | shint->data == BTRFS_BLOCK_GROUP_MIXED)) { | |
be744175 | 328 | used = btrfs_block_group_used(&shint->item); |
324ae4df Y |
329 | if (used + shint->pinned < |
330 | div_factor(shint->key.offset, factor)) { | |
be744175 CM |
331 | return shint; |
332 | } | |
333 | } | |
334 | } | |
c31f8830 CM |
335 | if (hint && hint->key.objectid < total_fs_bytes && |
336 | (hint->data == data || hint->data == BTRFS_BLOCK_GROUP_MIXED)) { | |
31f3c99b | 337 | used = btrfs_block_group_used(&hint->item); |
324ae4df Y |
338 | if (used + hint->pinned < |
339 | div_factor(hint->key.offset, factor)) { | |
31f3c99b CM |
340 | return hint; |
341 | } | |
e19caa5f | 342 | last = hint->key.objectid + hint->key.offset; |
31f3c99b CM |
343 | hint_last = last; |
344 | } else { | |
e37c9e69 CM |
345 | if (hint) |
346 | hint_last = max(hint->key.objectid, search_start); | |
347 | else | |
348 | hint_last = search_start; | |
349 | ||
c31f8830 CM |
350 | if (hint_last >= total_fs_bytes) |
351 | hint_last = search_start; | |
e37c9e69 | 352 | last = hint_last; |
31f3c99b | 353 | } |
31f3c99b | 354 | again: |
cd1bc465 | 355 | while(1) { |
96b5179d CM |
356 | ret = find_first_extent_bit(block_group_cache, last, |
357 | &start, &end, bit); | |
358 | if (ret) | |
cd1bc465 | 359 | break; |
96b5179d CM |
360 | |
361 | ret = get_state_private(block_group_cache, start, &ptr); | |
362 | if (ret) | |
363 | break; | |
364 | ||
ae2f5411 | 365 | cache = (struct btrfs_block_group_cache *)(unsigned long)ptr; |
96b5179d CM |
366 | last = cache->key.objectid + cache->key.offset; |
367 | used = btrfs_block_group_used(&cache->item); | |
368 | ||
c31f8830 CM |
369 | if (cache->key.objectid > total_fs_bytes) |
370 | break; | |
371 | ||
96b5179d CM |
372 | if (full_search) |
373 | free_check = cache->key.offset; | |
374 | else | |
375 | free_check = div_factor(cache->key.offset, factor); | |
324ae4df | 376 | if (used + cache->pinned < free_check) { |
96b5179d CM |
377 | found_group = cache; |
378 | goto found; | |
cd1bc465 | 379 | } |
de428b63 | 380 | cond_resched(); |
cd1bc465 | 381 | } |
31f3c99b | 382 | if (!full_search) { |
be744175 | 383 | last = search_start; |
31f3c99b CM |
384 | full_search = 1; |
385 | goto again; | |
386 | } | |
1e2677e0 | 387 | if (!data_swap) { |
1e2677e0 | 388 | data_swap = 1; |
96b5179d | 389 | bit = BLOCK_GROUP_DATA | BLOCK_GROUP_METADATA; |
1e2677e0 CM |
390 | last = search_start; |
391 | goto again; | |
392 | } | |
be744175 | 393 | found: |
31f3c99b | 394 | return found_group; |
cd1bc465 CM |
395 | } |
396 | ||
7bb86316 | 397 | static u64 hash_extent_ref(u64 root_objectid, u64 ref_generation, |
74493f7a CM |
398 | u64 owner, u64 owner_offset) |
399 | { | |
400 | u32 high_crc = ~(u32)0; | |
401 | u32 low_crc = ~(u32)0; | |
402 | __le64 lenum; | |
403 | ||
404 | lenum = cpu_to_le64(root_objectid); | |
405 | high_crc = crc32c(high_crc, &lenum, sizeof(lenum)); | |
7bb86316 CM |
406 | lenum = cpu_to_le64(ref_generation); |
407 | low_crc = crc32c(low_crc, &lenum, sizeof(lenum)); | |
21a4989d CM |
408 | if (owner >= BTRFS_FIRST_FREE_OBJECTID) { |
409 | lenum = cpu_to_le64(owner); | |
410 | low_crc = crc32c(low_crc, &lenum, sizeof(lenum)); | |
411 | lenum = cpu_to_le64(owner_offset); | |
412 | low_crc = crc32c(low_crc, &lenum, sizeof(lenum)); | |
413 | } | |
74493f7a CM |
414 | return ((u64)high_crc << 32) | (u64)low_crc; |
415 | } | |
416 | ||
7bb86316 CM |
417 | static int match_extent_ref(struct extent_buffer *leaf, |
418 | struct btrfs_extent_ref *disk_ref, | |
419 | struct btrfs_extent_ref *cpu_ref) | |
420 | { | |
421 | int ret; | |
422 | int len; | |
423 | ||
424 | if (cpu_ref->objectid) | |
425 | len = sizeof(*cpu_ref); | |
426 | else | |
427 | len = 2 * sizeof(u64); | |
428 | ret = memcmp_extent_buffer(leaf, cpu_ref, (unsigned long)disk_ref, | |
429 | len); | |
430 | return ret == 0; | |
431 | } | |
432 | ||
98ed5174 CM |
433 | static int noinline lookup_extent_backref(struct btrfs_trans_handle *trans, |
434 | struct btrfs_root *root, | |
435 | struct btrfs_path *path, u64 bytenr, | |
436 | u64 root_objectid, | |
437 | u64 ref_generation, u64 owner, | |
438 | u64 owner_offset, int del) | |
74493f7a CM |
439 | { |
440 | u64 hash; | |
441 | struct btrfs_key key; | |
7bb86316 | 442 | struct btrfs_key found_key; |
74493f7a | 443 | struct btrfs_extent_ref ref; |
7bb86316 CM |
444 | struct extent_buffer *leaf; |
445 | struct btrfs_extent_ref *disk_ref; | |
446 | int ret; | |
447 | int ret2; | |
448 | ||
449 | btrfs_set_stack_ref_root(&ref, root_objectid); | |
450 | btrfs_set_stack_ref_generation(&ref, ref_generation); | |
451 | btrfs_set_stack_ref_objectid(&ref, owner); | |
452 | btrfs_set_stack_ref_offset(&ref, owner_offset); | |
453 | ||
454 | hash = hash_extent_ref(root_objectid, ref_generation, owner, | |
455 | owner_offset); | |
456 | key.offset = hash; | |
457 | key.objectid = bytenr; | |
458 | key.type = BTRFS_EXTENT_REF_KEY; | |
459 | ||
460 | while (1) { | |
461 | ret = btrfs_search_slot(trans, root, &key, path, | |
462 | del ? -1 : 0, del); | |
463 | if (ret < 0) | |
464 | goto out; | |
465 | leaf = path->nodes[0]; | |
466 | if (ret != 0) { | |
467 | u32 nritems = btrfs_header_nritems(leaf); | |
468 | if (path->slots[0] >= nritems) { | |
469 | ret2 = btrfs_next_leaf(root, path); | |
470 | if (ret2) | |
471 | goto out; | |
472 | leaf = path->nodes[0]; | |
473 | } | |
474 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
475 | if (found_key.objectid != bytenr || | |
476 | found_key.type != BTRFS_EXTENT_REF_KEY) | |
477 | goto out; | |
478 | key.offset = found_key.offset; | |
479 | if (del) { | |
480 | btrfs_release_path(root, path); | |
481 | continue; | |
482 | } | |
483 | } | |
484 | disk_ref = btrfs_item_ptr(path->nodes[0], | |
485 | path->slots[0], | |
486 | struct btrfs_extent_ref); | |
487 | if (match_extent_ref(path->nodes[0], disk_ref, &ref)) { | |
488 | ret = 0; | |
489 | goto out; | |
490 | } | |
491 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
492 | key.offset = found_key.offset + 1; | |
493 | btrfs_release_path(root, path); | |
494 | } | |
495 | out: | |
496 | return ret; | |
497 | } | |
498 | ||
d8d5f3e1 CM |
499 | /* |
500 | * Back reference rules. Back refs have three main goals: | |
501 | * | |
502 | * 1) differentiate between all holders of references to an extent so that | |
503 | * when a reference is dropped we can make sure it was a valid reference | |
504 | * before freeing the extent. | |
505 | * | |
506 | * 2) Provide enough information to quickly find the holders of an extent | |
507 | * if we notice a given block is corrupted or bad. | |
508 | * | |
509 | * 3) Make it easy to migrate blocks for FS shrinking or storage pool | |
510 | * maintenance. This is actually the same as #2, but with a slightly | |
511 | * different use case. | |
512 | * | |
513 | * File extents can be referenced by: | |
514 | * | |
515 | * - multiple snapshots, subvolumes, or different generations in one subvol | |
516 | * - different files inside a single subvolume (in theory, not implemented yet) | |
517 | * - different offsets inside a file (bookend extents in file.c) | |
518 | * | |
519 | * The extent ref structure has fields for: | |
520 | * | |
521 | * - Objectid of the subvolume root | |
522 | * - Generation number of the tree holding the reference | |
523 | * - objectid of the file holding the reference | |
524 | * - offset in the file corresponding to the key holding the reference | |
525 | * | |
526 | * When a file extent is allocated the fields are filled in: | |
527 | * (root_key.objectid, trans->transid, inode objectid, offset in file) | |
528 | * | |
529 | * When a leaf is cow'd new references are added for every file extent found | |
530 | * in the leaf. It looks the same as the create case, but trans->transid | |
531 | * will be different when the block is cow'd. | |
532 | * | |
533 | * (root_key.objectid, trans->transid, inode objectid, offset in file) | |
534 | * | |
535 | * When a file extent is removed either during snapshot deletion or file | |
536 | * truncation, the corresponding back reference is found | |
537 | * by searching for: | |
538 | * | |
539 | * (btrfs_header_owner(leaf), btrfs_header_generation(leaf), | |
540 | * inode objectid, offset in file) | |
541 | * | |
542 | * Btree extents can be referenced by: | |
543 | * | |
544 | * - Different subvolumes | |
545 | * - Different generations of the same subvolume | |
546 | * | |
547 | * Storing sufficient information for a full reverse mapping of a btree | |
548 | * block would require storing the lowest key of the block in the backref, | |
549 | * and it would require updating that lowest key either before write out or | |
550 | * every time it changed. Instead, the objectid of the lowest key is stored | |
551 | * along with the level of the tree block. This provides a hint | |
552 | * about where in the btree the block can be found. Searches through the | |
553 | * btree only need to look for a pointer to that block, so they stop one | |
554 | * level higher than the level recorded in the backref. | |
555 | * | |
556 | * Some btrees do not do reference counting on their extents. These | |
557 | * include the extent tree and the tree of tree roots. Backrefs for these | |
558 | * trees always have a generation of zero. | |
559 | * | |
560 | * When a tree block is created, back references are inserted: | |
561 | * | |
f6dbff55 | 562 | * (root->root_key.objectid, trans->transid or zero, level, lowest_key_objectid) |
d8d5f3e1 CM |
563 | * |
564 | * When a tree block is cow'd in a reference counted root, | |
565 | * new back references are added for all the blocks it points to. | |
566 | * These are of the form (trans->transid will have increased since creation): | |
567 | * | |
f6dbff55 | 568 | * (root->root_key.objectid, trans->transid, level, lowest_key_objectid) |
d8d5f3e1 CM |
569 | * |
570 | * Because the lowest_key_objectid and the level are just hints | |
571 | * they are not used when backrefs are deleted. When a backref is deleted: | |
572 | * | |
573 | * if backref was for a tree root: | |
574 | * root_objectid = root->root_key.objectid | |
575 | * else | |
576 | * root_objectid = btrfs_header_owner(parent) | |
577 | * | |
578 | * (root_objectid, btrfs_header_generation(parent) or zero, 0, 0) | |
579 | * | |
580 | * Back Reference Key hashing: | |
581 | * | |
582 | * Back references have four fields, each 64 bits long. Unfortunately, | |
583 | * This is hashed into a single 64 bit number and placed into the key offset. | |
584 | * The key objectid corresponds to the first byte in the extent, and the | |
585 | * key type is set to BTRFS_EXTENT_REF_KEY | |
586 | */ | |
7bb86316 CM |
587 | int btrfs_insert_extent_backref(struct btrfs_trans_handle *trans, |
588 | struct btrfs_root *root, | |
589 | struct btrfs_path *path, u64 bytenr, | |
590 | u64 root_objectid, u64 ref_generation, | |
591 | u64 owner, u64 owner_offset) | |
592 | { | |
593 | u64 hash; | |
594 | struct btrfs_key key; | |
595 | struct btrfs_extent_ref ref; | |
596 | struct btrfs_extent_ref *disk_ref; | |
74493f7a CM |
597 | int ret; |
598 | ||
599 | btrfs_set_stack_ref_root(&ref, root_objectid); | |
7bb86316 | 600 | btrfs_set_stack_ref_generation(&ref, ref_generation); |
74493f7a CM |
601 | btrfs_set_stack_ref_objectid(&ref, owner); |
602 | btrfs_set_stack_ref_offset(&ref, owner_offset); | |
603 | ||
7bb86316 CM |
604 | hash = hash_extent_ref(root_objectid, ref_generation, owner, |
605 | owner_offset); | |
74493f7a CM |
606 | key.offset = hash; |
607 | key.objectid = bytenr; | |
608 | key.type = BTRFS_EXTENT_REF_KEY; | |
609 | ||
610 | ret = btrfs_insert_empty_item(trans, root, path, &key, sizeof(ref)); | |
611 | while (ret == -EEXIST) { | |
7bb86316 CM |
612 | disk_ref = btrfs_item_ptr(path->nodes[0], path->slots[0], |
613 | struct btrfs_extent_ref); | |
614 | if (match_extent_ref(path->nodes[0], disk_ref, &ref)) | |
615 | goto out; | |
616 | key.offset++; | |
617 | btrfs_release_path(root, path); | |
618 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
619 | sizeof(ref)); | |
74493f7a | 620 | } |
7bb86316 CM |
621 | if (ret) |
622 | goto out; | |
623 | disk_ref = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
624 | struct btrfs_extent_ref); | |
625 | write_extent_buffer(path->nodes[0], &ref, (unsigned long)disk_ref, | |
626 | sizeof(ref)); | |
627 | btrfs_mark_buffer_dirty(path->nodes[0]); | |
628 | out: | |
629 | btrfs_release_path(root, path); | |
630 | return ret; | |
74493f7a CM |
631 | } |
632 | ||
b18c6685 CM |
633 | int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans, |
634 | struct btrfs_root *root, | |
74493f7a | 635 | u64 bytenr, u64 num_bytes, |
7bb86316 | 636 | u64 root_objectid, u64 ref_generation, |
74493f7a | 637 | u64 owner, u64 owner_offset) |
02217ed2 | 638 | { |
5caf2a00 | 639 | struct btrfs_path *path; |
02217ed2 | 640 | int ret; |
e2fa7227 | 641 | struct btrfs_key key; |
5f39d397 | 642 | struct extent_buffer *l; |
234b63a0 | 643 | struct btrfs_extent_item *item; |
cf27e1ee | 644 | u32 refs; |
037e6390 | 645 | |
db94535d | 646 | WARN_ON(num_bytes < root->sectorsize); |
5caf2a00 | 647 | path = btrfs_alloc_path(); |
54aa1f4d CM |
648 | if (!path) |
649 | return -ENOMEM; | |
26b8003f | 650 | |
b0331a4c | 651 | path->reada = 0; |
db94535d | 652 | key.objectid = bytenr; |
62e2749e | 653 | btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY); |
db94535d | 654 | key.offset = num_bytes; |
5caf2a00 | 655 | ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path, |
9f5fae2f | 656 | 0, 1); |
54aa1f4d CM |
657 | if (ret < 0) |
658 | return ret; | |
a429e513 | 659 | if (ret != 0) { |
a28ec197 | 660 | BUG(); |
a429e513 | 661 | } |
02217ed2 | 662 | BUG_ON(ret != 0); |
5f39d397 | 663 | l = path->nodes[0]; |
5caf2a00 | 664 | item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item); |
5f39d397 CM |
665 | refs = btrfs_extent_refs(l, item); |
666 | btrfs_set_extent_refs(l, item, refs + 1); | |
5caf2a00 | 667 | btrfs_mark_buffer_dirty(path->nodes[0]); |
a28ec197 | 668 | |
5caf2a00 | 669 | btrfs_release_path(root->fs_info->extent_root, path); |
7bb86316 | 670 | |
b0331a4c | 671 | path->reada = 0; |
7bb86316 CM |
672 | ret = btrfs_insert_extent_backref(trans, root->fs_info->extent_root, |
673 | path, bytenr, root_objectid, | |
674 | ref_generation, owner, owner_offset); | |
675 | BUG_ON(ret); | |
9f5fae2f | 676 | finish_current_insert(trans, root->fs_info->extent_root); |
e20d96d6 | 677 | del_pending_extents(trans, root->fs_info->extent_root); |
74493f7a CM |
678 | |
679 | btrfs_free_path(path); | |
02217ed2 CM |
680 | return 0; |
681 | } | |
682 | ||
e9d0b13b CM |
683 | int btrfs_extent_post_op(struct btrfs_trans_handle *trans, |
684 | struct btrfs_root *root) | |
685 | { | |
686 | finish_current_insert(trans, root->fs_info->extent_root); | |
687 | del_pending_extents(trans, root->fs_info->extent_root); | |
688 | return 0; | |
689 | } | |
690 | ||
b18c6685 | 691 | static int lookup_extent_ref(struct btrfs_trans_handle *trans, |
db94535d CM |
692 | struct btrfs_root *root, u64 bytenr, |
693 | u64 num_bytes, u32 *refs) | |
a28ec197 | 694 | { |
5caf2a00 | 695 | struct btrfs_path *path; |
a28ec197 | 696 | int ret; |
e2fa7227 | 697 | struct btrfs_key key; |
5f39d397 | 698 | struct extent_buffer *l; |
234b63a0 | 699 | struct btrfs_extent_item *item; |
5caf2a00 | 700 | |
db94535d | 701 | WARN_ON(num_bytes < root->sectorsize); |
5caf2a00 | 702 | path = btrfs_alloc_path(); |
b0331a4c | 703 | path->reada = 0; |
db94535d CM |
704 | key.objectid = bytenr; |
705 | key.offset = num_bytes; | |
62e2749e | 706 | btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY); |
5caf2a00 | 707 | ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path, |
9f5fae2f | 708 | 0, 0); |
54aa1f4d CM |
709 | if (ret < 0) |
710 | goto out; | |
5f39d397 CM |
711 | if (ret != 0) { |
712 | btrfs_print_leaf(root, path->nodes[0]); | |
db94535d | 713 | printk("failed to find block number %Lu\n", bytenr); |
a28ec197 | 714 | BUG(); |
5f39d397 CM |
715 | } |
716 | l = path->nodes[0]; | |
5caf2a00 | 717 | item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item); |
5f39d397 | 718 | *refs = btrfs_extent_refs(l, item); |
54aa1f4d | 719 | out: |
5caf2a00 | 720 | btrfs_free_path(path); |
a28ec197 CM |
721 | return 0; |
722 | } | |
723 | ||
be20aa9d CM |
724 | u32 btrfs_count_snapshots_in_path(struct btrfs_root *root, |
725 | struct btrfs_path *count_path, | |
726 | u64 first_extent) | |
727 | { | |
728 | struct btrfs_root *extent_root = root->fs_info->extent_root; | |
729 | struct btrfs_path *path; | |
730 | u64 bytenr; | |
731 | u64 found_objectid; | |
56b453c9 | 732 | u64 root_objectid = root->root_key.objectid; |
be20aa9d CM |
733 | u32 total_count = 0; |
734 | u32 cur_count; | |
be20aa9d CM |
735 | u32 nritems; |
736 | int ret; | |
737 | struct btrfs_key key; | |
738 | struct btrfs_key found_key; | |
739 | struct extent_buffer *l; | |
740 | struct btrfs_extent_item *item; | |
741 | struct btrfs_extent_ref *ref_item; | |
742 | int level = -1; | |
743 | ||
744 | path = btrfs_alloc_path(); | |
745 | again: | |
746 | if (level == -1) | |
747 | bytenr = first_extent; | |
748 | else | |
749 | bytenr = count_path->nodes[level]->start; | |
750 | ||
751 | cur_count = 0; | |
752 | key.objectid = bytenr; | |
753 | key.offset = 0; | |
754 | ||
755 | btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY); | |
756 | ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0); | |
757 | if (ret < 0) | |
758 | goto out; | |
759 | BUG_ON(ret == 0); | |
760 | ||
761 | l = path->nodes[0]; | |
762 | btrfs_item_key_to_cpu(l, &found_key, path->slots[0]); | |
763 | ||
764 | if (found_key.objectid != bytenr || | |
765 | found_key.type != BTRFS_EXTENT_ITEM_KEY) { | |
766 | goto out; | |
767 | } | |
768 | ||
769 | item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item); | |
be20aa9d | 770 | while (1) { |
bd09835d | 771 | l = path->nodes[0]; |
be20aa9d CM |
772 | nritems = btrfs_header_nritems(l); |
773 | if (path->slots[0] >= nritems) { | |
774 | ret = btrfs_next_leaf(extent_root, path); | |
775 | if (ret == 0) | |
776 | continue; | |
777 | break; | |
778 | } | |
779 | btrfs_item_key_to_cpu(l, &found_key, path->slots[0]); | |
780 | if (found_key.objectid != bytenr) | |
781 | break; | |
bd09835d | 782 | |
be20aa9d CM |
783 | if (found_key.type != BTRFS_EXTENT_REF_KEY) { |
784 | path->slots[0]++; | |
785 | continue; | |
786 | } | |
787 | ||
788 | cur_count++; | |
789 | ref_item = btrfs_item_ptr(l, path->slots[0], | |
790 | struct btrfs_extent_ref); | |
791 | found_objectid = btrfs_ref_root(l, ref_item); | |
792 | ||
56b453c9 CM |
793 | if (found_objectid != root_objectid) { |
794 | total_count = 2; | |
4313b399 | 795 | goto out; |
56b453c9 CM |
796 | } |
797 | total_count = 1; | |
be20aa9d CM |
798 | path->slots[0]++; |
799 | } | |
800 | if (cur_count == 0) { | |
801 | total_count = 0; | |
802 | goto out; | |
803 | } | |
be20aa9d CM |
804 | if (level >= 0 && root->node == count_path->nodes[level]) |
805 | goto out; | |
806 | level++; | |
807 | btrfs_release_path(root, path); | |
808 | goto again; | |
809 | ||
810 | out: | |
811 | btrfs_free_path(path); | |
812 | return total_count; | |
be20aa9d | 813 | } |
c5739bba | 814 | int btrfs_inc_root_ref(struct btrfs_trans_handle *trans, |
7bb86316 | 815 | struct btrfs_root *root, u64 owner_objectid) |
c5739bba | 816 | { |
7bb86316 CM |
817 | u64 generation; |
818 | u64 key_objectid; | |
819 | u64 level; | |
820 | u32 nritems; | |
821 | struct btrfs_disk_key disk_key; | |
822 | ||
823 | level = btrfs_header_level(root->node); | |
824 | generation = trans->transid; | |
825 | nritems = btrfs_header_nritems(root->node); | |
826 | if (nritems > 0) { | |
827 | if (level == 0) | |
828 | btrfs_item_key(root->node, &disk_key, 0); | |
829 | else | |
830 | btrfs_node_key(root->node, &disk_key, 0); | |
831 | key_objectid = btrfs_disk_key_objectid(&disk_key); | |
832 | } else { | |
833 | key_objectid = 0; | |
834 | } | |
db94535d | 835 | return btrfs_inc_extent_ref(trans, root, root->node->start, |
7bb86316 | 836 | root->node->len, owner_objectid, |
f6dbff55 | 837 | generation, level, key_objectid); |
c5739bba CM |
838 | } |
839 | ||
e089f05c | 840 | int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root, |
5f39d397 | 841 | struct extent_buffer *buf) |
02217ed2 | 842 | { |
db94535d | 843 | u64 bytenr; |
5f39d397 CM |
844 | u32 nritems; |
845 | struct btrfs_key key; | |
6407bf6d | 846 | struct btrfs_file_extent_item *fi; |
02217ed2 | 847 | int i; |
db94535d | 848 | int level; |
6407bf6d | 849 | int ret; |
54aa1f4d | 850 | int faili; |
a28ec197 | 851 | |
3768f368 | 852 | if (!root->ref_cows) |
a28ec197 | 853 | return 0; |
5f39d397 | 854 | |
db94535d | 855 | level = btrfs_header_level(buf); |
5f39d397 CM |
856 | nritems = btrfs_header_nritems(buf); |
857 | for (i = 0; i < nritems; i++) { | |
db94535d CM |
858 | if (level == 0) { |
859 | u64 disk_bytenr; | |
5f39d397 CM |
860 | btrfs_item_key_to_cpu(buf, &key, i); |
861 | if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY) | |
6407bf6d | 862 | continue; |
5f39d397 | 863 | fi = btrfs_item_ptr(buf, i, |
6407bf6d | 864 | struct btrfs_file_extent_item); |
5f39d397 | 865 | if (btrfs_file_extent_type(buf, fi) == |
236454df CM |
866 | BTRFS_FILE_EXTENT_INLINE) |
867 | continue; | |
db94535d CM |
868 | disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi); |
869 | if (disk_bytenr == 0) | |
3a686375 | 870 | continue; |
db94535d | 871 | ret = btrfs_inc_extent_ref(trans, root, disk_bytenr, |
7bb86316 CM |
872 | btrfs_file_extent_disk_num_bytes(buf, fi), |
873 | root->root_key.objectid, trans->transid, | |
874 | key.objectid, key.offset); | |
54aa1f4d CM |
875 | if (ret) { |
876 | faili = i; | |
877 | goto fail; | |
878 | } | |
6407bf6d | 879 | } else { |
db94535d | 880 | bytenr = btrfs_node_blockptr(buf, i); |
6caab489 | 881 | btrfs_node_key_to_cpu(buf, &key, i); |
db94535d | 882 | ret = btrfs_inc_extent_ref(trans, root, bytenr, |
7bb86316 CM |
883 | btrfs_level_size(root, level - 1), |
884 | root->root_key.objectid, | |
f6dbff55 CM |
885 | trans->transid, |
886 | level - 1, key.objectid); | |
54aa1f4d CM |
887 | if (ret) { |
888 | faili = i; | |
889 | goto fail; | |
890 | } | |
6407bf6d | 891 | } |
02217ed2 CM |
892 | } |
893 | return 0; | |
54aa1f4d | 894 | fail: |
ccd467d6 | 895 | WARN_ON(1); |
7bb86316 | 896 | #if 0 |
54aa1f4d | 897 | for (i =0; i < faili; i++) { |
db94535d CM |
898 | if (level == 0) { |
899 | u64 disk_bytenr; | |
5f39d397 CM |
900 | btrfs_item_key_to_cpu(buf, &key, i); |
901 | if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY) | |
54aa1f4d | 902 | continue; |
5f39d397 | 903 | fi = btrfs_item_ptr(buf, i, |
54aa1f4d | 904 | struct btrfs_file_extent_item); |
5f39d397 | 905 | if (btrfs_file_extent_type(buf, fi) == |
54aa1f4d CM |
906 | BTRFS_FILE_EXTENT_INLINE) |
907 | continue; | |
db94535d CM |
908 | disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi); |
909 | if (disk_bytenr == 0) | |
54aa1f4d | 910 | continue; |
db94535d CM |
911 | err = btrfs_free_extent(trans, root, disk_bytenr, |
912 | btrfs_file_extent_disk_num_bytes(buf, | |
5f39d397 | 913 | fi), 0); |
54aa1f4d CM |
914 | BUG_ON(err); |
915 | } else { | |
db94535d CM |
916 | bytenr = btrfs_node_blockptr(buf, i); |
917 | err = btrfs_free_extent(trans, root, bytenr, | |
918 | btrfs_level_size(root, level - 1), 0); | |
54aa1f4d CM |
919 | BUG_ON(err); |
920 | } | |
921 | } | |
7bb86316 | 922 | #endif |
54aa1f4d | 923 | return ret; |
02217ed2 CM |
924 | } |
925 | ||
9078a3e1 CM |
926 | static int write_one_cache_group(struct btrfs_trans_handle *trans, |
927 | struct btrfs_root *root, | |
928 | struct btrfs_path *path, | |
929 | struct btrfs_block_group_cache *cache) | |
930 | { | |
931 | int ret; | |
932 | int pending_ret; | |
933 | struct btrfs_root *extent_root = root->fs_info->extent_root; | |
5f39d397 CM |
934 | unsigned long bi; |
935 | struct extent_buffer *leaf; | |
9078a3e1 | 936 | |
9078a3e1 | 937 | ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1); |
54aa1f4d CM |
938 | if (ret < 0) |
939 | goto fail; | |
9078a3e1 | 940 | BUG_ON(ret); |
5f39d397 CM |
941 | |
942 | leaf = path->nodes[0]; | |
943 | bi = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
944 | write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item)); | |
945 | btrfs_mark_buffer_dirty(leaf); | |
9078a3e1 | 946 | btrfs_release_path(extent_root, path); |
54aa1f4d | 947 | fail: |
9078a3e1 CM |
948 | finish_current_insert(trans, extent_root); |
949 | pending_ret = del_pending_extents(trans, extent_root); | |
950 | if (ret) | |
951 | return ret; | |
952 | if (pending_ret) | |
953 | return pending_ret; | |
954 | return 0; | |
955 | ||
956 | } | |
957 | ||
96b5179d CM |
958 | int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans, |
959 | struct btrfs_root *root) | |
9078a3e1 | 960 | { |
d1310b2e | 961 | struct extent_io_tree *block_group_cache; |
96b5179d | 962 | struct btrfs_block_group_cache *cache; |
9078a3e1 CM |
963 | int ret; |
964 | int err = 0; | |
965 | int werr = 0; | |
9078a3e1 | 966 | struct btrfs_path *path; |
96b5179d CM |
967 | u64 last = 0; |
968 | u64 start; | |
969 | u64 end; | |
970 | u64 ptr; | |
9078a3e1 | 971 | |
96b5179d | 972 | block_group_cache = &root->fs_info->block_group_cache; |
9078a3e1 CM |
973 | path = btrfs_alloc_path(); |
974 | if (!path) | |
975 | return -ENOMEM; | |
976 | ||
977 | while(1) { | |
96b5179d CM |
978 | ret = find_first_extent_bit(block_group_cache, last, |
979 | &start, &end, BLOCK_GROUP_DIRTY); | |
980 | if (ret) | |
9078a3e1 | 981 | break; |
54aa1f4d | 982 | |
96b5179d CM |
983 | last = end + 1; |
984 | ret = get_state_private(block_group_cache, start, &ptr); | |
985 | if (ret) | |
986 | break; | |
987 | ||
ae2f5411 | 988 | cache = (struct btrfs_block_group_cache *)(unsigned long)ptr; |
96b5179d CM |
989 | err = write_one_cache_group(trans, root, |
990 | path, cache); | |
991 | /* | |
992 | * if we fail to write the cache group, we want | |
993 | * to keep it marked dirty in hopes that a later | |
994 | * write will work | |
995 | */ | |
996 | if (err) { | |
997 | werr = err; | |
998 | continue; | |
9078a3e1 | 999 | } |
96b5179d CM |
1000 | clear_extent_bits(block_group_cache, start, end, |
1001 | BLOCK_GROUP_DIRTY, GFP_NOFS); | |
9078a3e1 CM |
1002 | } |
1003 | btrfs_free_path(path); | |
1004 | return werr; | |
1005 | } | |
1006 | ||
1007 | static int update_block_group(struct btrfs_trans_handle *trans, | |
1008 | struct btrfs_root *root, | |
db94535d CM |
1009 | u64 bytenr, u64 num_bytes, int alloc, |
1010 | int mark_free, int data) | |
9078a3e1 CM |
1011 | { |
1012 | struct btrfs_block_group_cache *cache; | |
1013 | struct btrfs_fs_info *info = root->fs_info; | |
db94535d | 1014 | u64 total = num_bytes; |
9078a3e1 | 1015 | u64 old_val; |
db94535d | 1016 | u64 byte_in_group; |
96b5179d CM |
1017 | u64 start; |
1018 | u64 end; | |
3e1ad54f | 1019 | |
9078a3e1 | 1020 | while(total) { |
db94535d | 1021 | cache = btrfs_lookup_block_group(info, bytenr); |
3e1ad54f | 1022 | if (!cache) { |
9078a3e1 | 1023 | return -1; |
cd1bc465 | 1024 | } |
db94535d CM |
1025 | byte_in_group = bytenr - cache->key.objectid; |
1026 | WARN_ON(byte_in_group > cache->key.offset); | |
96b5179d CM |
1027 | start = cache->key.objectid; |
1028 | end = start + cache->key.offset - 1; | |
1029 | set_extent_bits(&info->block_group_cache, start, end, | |
1030 | BLOCK_GROUP_DIRTY, GFP_NOFS); | |
9078a3e1 CM |
1031 | |
1032 | old_val = btrfs_block_group_used(&cache->item); | |
db94535d | 1033 | num_bytes = min(total, cache->key.offset - byte_in_group); |
cd1bc465 | 1034 | if (alloc) { |
1e2677e0 | 1035 | if (cache->data != data && |
84f54cfa | 1036 | old_val < (cache->key.offset >> 1)) { |
96b5179d CM |
1037 | int bit_to_clear; |
1038 | int bit_to_set; | |
96b5179d | 1039 | cache->data = data; |
1e2677e0 | 1040 | if (data) { |
b97f9203 Y |
1041 | bit_to_clear = BLOCK_GROUP_METADATA; |
1042 | bit_to_set = BLOCK_GROUP_DATA; | |
f84a8b36 CM |
1043 | cache->item.flags &= |
1044 | ~BTRFS_BLOCK_GROUP_MIXED; | |
1e2677e0 CM |
1045 | cache->item.flags |= |
1046 | BTRFS_BLOCK_GROUP_DATA; | |
1047 | } else { | |
b97f9203 Y |
1048 | bit_to_clear = BLOCK_GROUP_DATA; |
1049 | bit_to_set = BLOCK_GROUP_METADATA; | |
f84a8b36 CM |
1050 | cache->item.flags &= |
1051 | ~BTRFS_BLOCK_GROUP_MIXED; | |
1e2677e0 CM |
1052 | cache->item.flags &= |
1053 | ~BTRFS_BLOCK_GROUP_DATA; | |
1054 | } | |
96b5179d CM |
1055 | clear_extent_bits(&info->block_group_cache, |
1056 | start, end, bit_to_clear, | |
1057 | GFP_NOFS); | |
1058 | set_extent_bits(&info->block_group_cache, | |
1059 | start, end, bit_to_set, | |
1060 | GFP_NOFS); | |
f84a8b36 CM |
1061 | } else if (cache->data != data && |
1062 | cache->data != BTRFS_BLOCK_GROUP_MIXED) { | |
1063 | cache->data = BTRFS_BLOCK_GROUP_MIXED; | |
1064 | set_extent_bits(&info->block_group_cache, | |
1065 | start, end, | |
1066 | BLOCK_GROUP_DATA | | |
1067 | BLOCK_GROUP_METADATA, | |
1068 | GFP_NOFS); | |
1e2677e0 | 1069 | } |
db94535d | 1070 | old_val += num_bytes; |
cd1bc465 | 1071 | } else { |
db94535d | 1072 | old_val -= num_bytes; |
f510cfec CM |
1073 | if (mark_free) { |
1074 | set_extent_dirty(&info->free_space_cache, | |
db94535d | 1075 | bytenr, bytenr + num_bytes - 1, |
f510cfec | 1076 | GFP_NOFS); |
e37c9e69 | 1077 | } |
cd1bc465 | 1078 | } |
9078a3e1 | 1079 | btrfs_set_block_group_used(&cache->item, old_val); |
db94535d CM |
1080 | total -= num_bytes; |
1081 | bytenr += num_bytes; | |
9078a3e1 CM |
1082 | } |
1083 | return 0; | |
1084 | } | |
324ae4df Y |
1085 | static int update_pinned_extents(struct btrfs_root *root, |
1086 | u64 bytenr, u64 num, int pin) | |
1087 | { | |
1088 | u64 len; | |
1089 | struct btrfs_block_group_cache *cache; | |
1090 | struct btrfs_fs_info *fs_info = root->fs_info; | |
1091 | ||
1092 | if (pin) { | |
1093 | set_extent_dirty(&fs_info->pinned_extents, | |
1094 | bytenr, bytenr + num - 1, GFP_NOFS); | |
1095 | } else { | |
1096 | clear_extent_dirty(&fs_info->pinned_extents, | |
1097 | bytenr, bytenr + num - 1, GFP_NOFS); | |
1098 | } | |
1099 | while (num > 0) { | |
1100 | cache = btrfs_lookup_block_group(fs_info, bytenr); | |
1101 | WARN_ON(!cache); | |
1102 | len = min(num, cache->key.offset - | |
1103 | (bytenr - cache->key.objectid)); | |
1104 | if (pin) { | |
1105 | cache->pinned += len; | |
1106 | fs_info->total_pinned += len; | |
1107 | } else { | |
1108 | cache->pinned -= len; | |
1109 | fs_info->total_pinned -= len; | |
1110 | } | |
1111 | bytenr += len; | |
1112 | num -= len; | |
1113 | } | |
1114 | return 0; | |
1115 | } | |
9078a3e1 | 1116 | |
d1310b2e | 1117 | int btrfs_copy_pinned(struct btrfs_root *root, struct extent_io_tree *copy) |
ccd467d6 | 1118 | { |
ccd467d6 | 1119 | u64 last = 0; |
1a5bc167 CM |
1120 | u64 start; |
1121 | u64 end; | |
d1310b2e | 1122 | struct extent_io_tree *pinned_extents = &root->fs_info->pinned_extents; |
ccd467d6 | 1123 | int ret; |
ccd467d6 CM |
1124 | |
1125 | while(1) { | |
1a5bc167 CM |
1126 | ret = find_first_extent_bit(pinned_extents, last, |
1127 | &start, &end, EXTENT_DIRTY); | |
1128 | if (ret) | |
ccd467d6 | 1129 | break; |
1a5bc167 CM |
1130 | set_extent_dirty(copy, start, end, GFP_NOFS); |
1131 | last = end + 1; | |
ccd467d6 CM |
1132 | } |
1133 | return 0; | |
1134 | } | |
1135 | ||
1136 | int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans, | |
1137 | struct btrfs_root *root, | |
d1310b2e | 1138 | struct extent_io_tree *unpin) |
a28ec197 | 1139 | { |
1a5bc167 CM |
1140 | u64 start; |
1141 | u64 end; | |
a28ec197 | 1142 | int ret; |
d1310b2e | 1143 | struct extent_io_tree *free_space_cache; |
f510cfec | 1144 | free_space_cache = &root->fs_info->free_space_cache; |
a28ec197 CM |
1145 | |
1146 | while(1) { | |
1a5bc167 CM |
1147 | ret = find_first_extent_bit(unpin, 0, &start, &end, |
1148 | EXTENT_DIRTY); | |
1149 | if (ret) | |
a28ec197 | 1150 | break; |
324ae4df | 1151 | update_pinned_extents(root, start, end + 1 - start, 0); |
1a5bc167 CM |
1152 | clear_extent_dirty(unpin, start, end, GFP_NOFS); |
1153 | set_extent_dirty(free_space_cache, start, end, GFP_NOFS); | |
a28ec197 CM |
1154 | } |
1155 | return 0; | |
1156 | } | |
1157 | ||
98ed5174 CM |
1158 | static int finish_current_insert(struct btrfs_trans_handle *trans, |
1159 | struct btrfs_root *extent_root) | |
037e6390 | 1160 | { |
7bb86316 CM |
1161 | u64 start; |
1162 | u64 end; | |
1163 | struct btrfs_fs_info *info = extent_root->fs_info; | |
d8d5f3e1 | 1164 | struct extent_buffer *eb; |
7bb86316 | 1165 | struct btrfs_path *path; |
e2fa7227 | 1166 | struct btrfs_key ins; |
d8d5f3e1 | 1167 | struct btrfs_disk_key first; |
234b63a0 | 1168 | struct btrfs_extent_item extent_item; |
037e6390 | 1169 | int ret; |
d8d5f3e1 | 1170 | int level; |
1a5bc167 | 1171 | int err = 0; |
037e6390 | 1172 | |
5f39d397 | 1173 | btrfs_set_stack_extent_refs(&extent_item, 1); |
62e2749e | 1174 | btrfs_set_key_type(&ins, BTRFS_EXTENT_ITEM_KEY); |
7bb86316 | 1175 | path = btrfs_alloc_path(); |
037e6390 | 1176 | |
26b8003f | 1177 | while(1) { |
1a5bc167 CM |
1178 | ret = find_first_extent_bit(&info->extent_ins, 0, &start, |
1179 | &end, EXTENT_LOCKED); | |
1180 | if (ret) | |
26b8003f CM |
1181 | break; |
1182 | ||
1a5bc167 CM |
1183 | ins.objectid = start; |
1184 | ins.offset = end + 1 - start; | |
1185 | err = btrfs_insert_item(trans, extent_root, &ins, | |
1186 | &extent_item, sizeof(extent_item)); | |
1187 | clear_extent_bits(&info->extent_ins, start, end, EXTENT_LOCKED, | |
1188 | GFP_NOFS); | |
d8d5f3e1 CM |
1189 | eb = read_tree_block(extent_root, ins.objectid, ins.offset); |
1190 | level = btrfs_header_level(eb); | |
1191 | if (level == 0) { | |
1192 | btrfs_item_key(eb, &first, 0); | |
1193 | } else { | |
1194 | btrfs_node_key(eb, &first, 0); | |
1195 | } | |
7bb86316 CM |
1196 | err = btrfs_insert_extent_backref(trans, extent_root, path, |
1197 | start, extent_root->root_key.objectid, | |
f6dbff55 CM |
1198 | 0, level, |
1199 | btrfs_disk_key_objectid(&first)); | |
7bb86316 | 1200 | BUG_ON(err); |
d8d5f3e1 | 1201 | free_extent_buffer(eb); |
037e6390 | 1202 | } |
7bb86316 | 1203 | btrfs_free_path(path); |
037e6390 CM |
1204 | return 0; |
1205 | } | |
1206 | ||
db94535d CM |
1207 | static int pin_down_bytes(struct btrfs_root *root, u64 bytenr, u32 num_bytes, |
1208 | int pending) | |
e20d96d6 | 1209 | { |
1a5bc167 | 1210 | int err = 0; |
5f39d397 | 1211 | struct extent_buffer *buf; |
8ef97622 | 1212 | |
f4b9aa8d | 1213 | if (!pending) { |
db94535d | 1214 | buf = btrfs_find_tree_block(root, bytenr, num_bytes); |
5f39d397 CM |
1215 | if (buf) { |
1216 | if (btrfs_buffer_uptodate(buf)) { | |
2c90e5d6 CM |
1217 | u64 transid = |
1218 | root->fs_info->running_transaction->transid; | |
dc17ff8f CM |
1219 | u64 header_transid = |
1220 | btrfs_header_generation(buf); | |
1221 | if (header_transid == transid) { | |
55c69072 | 1222 | clean_tree_block(NULL, root, buf); |
5f39d397 | 1223 | free_extent_buffer(buf); |
c549228f | 1224 | return 1; |
2c90e5d6 | 1225 | } |
f4b9aa8d | 1226 | } |
5f39d397 | 1227 | free_extent_buffer(buf); |
8ef97622 | 1228 | } |
324ae4df | 1229 | update_pinned_extents(root, bytenr, num_bytes, 1); |
f4b9aa8d | 1230 | } else { |
1a5bc167 | 1231 | set_extent_bits(&root->fs_info->pending_del, |
db94535d CM |
1232 | bytenr, bytenr + num_bytes - 1, |
1233 | EXTENT_LOCKED, GFP_NOFS); | |
f4b9aa8d | 1234 | } |
be744175 | 1235 | BUG_ON(err < 0); |
e20d96d6 CM |
1236 | return 0; |
1237 | } | |
1238 | ||
fec577fb | 1239 | /* |
a28ec197 | 1240 | * remove an extent from the root, returns 0 on success |
fec577fb | 1241 | */ |
e089f05c | 1242 | static int __free_extent(struct btrfs_trans_handle *trans, struct btrfs_root |
7bb86316 CM |
1243 | *root, u64 bytenr, u64 num_bytes, |
1244 | u64 root_objectid, u64 ref_generation, | |
1245 | u64 owner_objectid, u64 owner_offset, int pin, | |
e37c9e69 | 1246 | int mark_free) |
a28ec197 | 1247 | { |
5caf2a00 | 1248 | struct btrfs_path *path; |
e2fa7227 | 1249 | struct btrfs_key key; |
1261ec42 CM |
1250 | struct btrfs_fs_info *info = root->fs_info; |
1251 | struct btrfs_root *extent_root = info->extent_root; | |
5f39d397 | 1252 | struct extent_buffer *leaf; |
a28ec197 | 1253 | int ret; |
952fccac CM |
1254 | int extent_slot = 0; |
1255 | int found_extent = 0; | |
1256 | int num_to_del = 1; | |
234b63a0 | 1257 | struct btrfs_extent_item *ei; |
cf27e1ee | 1258 | u32 refs; |
037e6390 | 1259 | |
db94535d | 1260 | key.objectid = bytenr; |
62e2749e | 1261 | btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY); |
db94535d | 1262 | key.offset = num_bytes; |
5caf2a00 | 1263 | path = btrfs_alloc_path(); |
54aa1f4d CM |
1264 | if (!path) |
1265 | return -ENOMEM; | |
5f26f772 | 1266 | |
b0331a4c | 1267 | path->reada = 0; |
7bb86316 CM |
1268 | ret = lookup_extent_backref(trans, extent_root, path, |
1269 | bytenr, root_objectid, | |
1270 | ref_generation, | |
1271 | owner_objectid, owner_offset, 1); | |
1272 | if (ret == 0) { | |
952fccac CM |
1273 | struct btrfs_key found_key; |
1274 | extent_slot = path->slots[0]; | |
1275 | while(extent_slot > 0) { | |
1276 | extent_slot--; | |
1277 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, | |
1278 | extent_slot); | |
1279 | if (found_key.objectid != bytenr) | |
1280 | break; | |
1281 | if (found_key.type == BTRFS_EXTENT_ITEM_KEY && | |
1282 | found_key.offset == num_bytes) { | |
1283 | found_extent = 1; | |
1284 | break; | |
1285 | } | |
1286 | if (path->slots[0] - extent_slot > 5) | |
1287 | break; | |
1288 | } | |
1289 | if (!found_extent) | |
1290 | ret = btrfs_del_item(trans, extent_root, path); | |
7bb86316 CM |
1291 | } else { |
1292 | btrfs_print_leaf(extent_root, path->nodes[0]); | |
1293 | WARN_ON(1); | |
1294 | printk("Unable to find ref byte nr %Lu root %Lu " | |
1295 | " gen %Lu owner %Lu offset %Lu\n", bytenr, | |
1296 | root_objectid, ref_generation, owner_objectid, | |
1297 | owner_offset); | |
1298 | } | |
952fccac CM |
1299 | if (!found_extent) { |
1300 | btrfs_release_path(extent_root, path); | |
1301 | ret = btrfs_search_slot(trans, extent_root, &key, path, -1, 1); | |
1302 | if (ret < 0) | |
1303 | return ret; | |
1304 | BUG_ON(ret); | |
1305 | extent_slot = path->slots[0]; | |
1306 | } | |
5f39d397 CM |
1307 | |
1308 | leaf = path->nodes[0]; | |
952fccac | 1309 | ei = btrfs_item_ptr(leaf, extent_slot, |
123abc88 | 1310 | struct btrfs_extent_item); |
5f39d397 CM |
1311 | refs = btrfs_extent_refs(leaf, ei); |
1312 | BUG_ON(refs == 0); | |
1313 | refs -= 1; | |
1314 | btrfs_set_extent_refs(leaf, ei, refs); | |
952fccac | 1315 | |
5f39d397 CM |
1316 | btrfs_mark_buffer_dirty(leaf); |
1317 | ||
952fccac CM |
1318 | if (refs == 0 && found_extent && path->slots[0] == extent_slot + 1) { |
1319 | /* if the back ref and the extent are next to each other | |
1320 | * they get deleted below in one shot | |
1321 | */ | |
1322 | path->slots[0] = extent_slot; | |
1323 | num_to_del = 2; | |
1324 | } else if (found_extent) { | |
1325 | /* otherwise delete the extent back ref */ | |
1326 | ret = btrfs_del_item(trans, extent_root, path); | |
1327 | BUG_ON(ret); | |
1328 | /* if refs are 0, we need to setup the path for deletion */ | |
1329 | if (refs == 0) { | |
1330 | btrfs_release_path(extent_root, path); | |
1331 | ret = btrfs_search_slot(trans, extent_root, &key, path, | |
1332 | -1, 1); | |
1333 | if (ret < 0) | |
1334 | return ret; | |
1335 | BUG_ON(ret); | |
1336 | } | |
1337 | } | |
1338 | ||
cf27e1ee | 1339 | if (refs == 0) { |
db94535d CM |
1340 | u64 super_used; |
1341 | u64 root_used; | |
78fae27e CM |
1342 | |
1343 | if (pin) { | |
db94535d | 1344 | ret = pin_down_bytes(root, bytenr, num_bytes, 0); |
c549228f Y |
1345 | if (ret > 0) |
1346 | mark_free = 1; | |
1347 | BUG_ON(ret < 0); | |
78fae27e CM |
1348 | } |
1349 | ||
58176a96 | 1350 | /* block accounting for super block */ |
db94535d CM |
1351 | super_used = btrfs_super_bytes_used(&info->super_copy); |
1352 | btrfs_set_super_bytes_used(&info->super_copy, | |
1353 | super_used - num_bytes); | |
58176a96 JB |
1354 | |
1355 | /* block accounting for root item */ | |
db94535d | 1356 | root_used = btrfs_root_used(&root->root_item); |
5f39d397 | 1357 | btrfs_set_root_used(&root->root_item, |
db94535d | 1358 | root_used - num_bytes); |
952fccac CM |
1359 | ret = btrfs_del_items(trans, extent_root, path, path->slots[0], |
1360 | num_to_del); | |
54aa1f4d CM |
1361 | if (ret) { |
1362 | return ret; | |
1363 | } | |
db94535d | 1364 | ret = update_block_group(trans, root, bytenr, num_bytes, 0, |
1e2677e0 | 1365 | mark_free, 0); |
9078a3e1 | 1366 | BUG_ON(ret); |
a28ec197 | 1367 | } |
5caf2a00 | 1368 | btrfs_free_path(path); |
e089f05c | 1369 | finish_current_insert(trans, extent_root); |
a28ec197 CM |
1370 | return ret; |
1371 | } | |
1372 | ||
a28ec197 CM |
1373 | /* |
1374 | * find all the blocks marked as pending in the radix tree and remove | |
1375 | * them from the extent map | |
1376 | */ | |
e089f05c CM |
1377 | static int del_pending_extents(struct btrfs_trans_handle *trans, struct |
1378 | btrfs_root *extent_root) | |
a28ec197 CM |
1379 | { |
1380 | int ret; | |
e20d96d6 | 1381 | int err = 0; |
1a5bc167 CM |
1382 | u64 start; |
1383 | u64 end; | |
d1310b2e CM |
1384 | struct extent_io_tree *pending_del; |
1385 | struct extent_io_tree *pinned_extents; | |
8ef97622 | 1386 | |
1a5bc167 CM |
1387 | pending_del = &extent_root->fs_info->pending_del; |
1388 | pinned_extents = &extent_root->fs_info->pinned_extents; | |
a28ec197 CM |
1389 | |
1390 | while(1) { | |
1a5bc167 CM |
1391 | ret = find_first_extent_bit(pending_del, 0, &start, &end, |
1392 | EXTENT_LOCKED); | |
1393 | if (ret) | |
a28ec197 | 1394 | break; |
324ae4df | 1395 | update_pinned_extents(extent_root, start, end + 1 - start, 1); |
1a5bc167 CM |
1396 | clear_extent_bits(pending_del, start, end, EXTENT_LOCKED, |
1397 | GFP_NOFS); | |
1398 | ret = __free_extent(trans, extent_root, | |
7bb86316 CM |
1399 | start, end + 1 - start, |
1400 | extent_root->root_key.objectid, | |
1401 | 0, 0, 0, 0, 0); | |
1a5bc167 CM |
1402 | if (ret) |
1403 | err = ret; | |
fec577fb | 1404 | } |
e20d96d6 | 1405 | return err; |
fec577fb CM |
1406 | } |
1407 | ||
1408 | /* | |
1409 | * remove an extent from the root, returns 0 on success | |
1410 | */ | |
e089f05c | 1411 | int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root |
7bb86316 CM |
1412 | *root, u64 bytenr, u64 num_bytes, |
1413 | u64 root_objectid, u64 ref_generation, | |
1414 | u64 owner_objectid, u64 owner_offset, int pin) | |
fec577fb | 1415 | { |
9f5fae2f | 1416 | struct btrfs_root *extent_root = root->fs_info->extent_root; |
fec577fb CM |
1417 | int pending_ret; |
1418 | int ret; | |
a28ec197 | 1419 | |
db94535d | 1420 | WARN_ON(num_bytes < root->sectorsize); |
7bb86316 CM |
1421 | if (!root->ref_cows) |
1422 | ref_generation = 0; | |
1423 | ||
fec577fb | 1424 | if (root == extent_root) { |
db94535d | 1425 | pin_down_bytes(root, bytenr, num_bytes, 1); |
fec577fb CM |
1426 | return 0; |
1427 | } | |
7bb86316 CM |
1428 | ret = __free_extent(trans, root, bytenr, num_bytes, root_objectid, |
1429 | ref_generation, owner_objectid, owner_offset, | |
1430 | pin, pin == 0); | |
e20d96d6 | 1431 | pending_ret = del_pending_extents(trans, root->fs_info->extent_root); |
fec577fb CM |
1432 | return ret ? ret : pending_ret; |
1433 | } | |
1434 | ||
87ee04eb CM |
1435 | static u64 stripe_align(struct btrfs_root *root, u64 val) |
1436 | { | |
1437 | u64 mask = ((u64)root->stripesize - 1); | |
1438 | u64 ret = (val + mask) & ~mask; | |
1439 | return ret; | |
1440 | } | |
1441 | ||
fec577fb CM |
1442 | /* |
1443 | * walks the btree of allocated extents and find a hole of a given size. | |
1444 | * The key ins is changed to record the hole: | |
1445 | * ins->objectid == block start | |
62e2749e | 1446 | * ins->flags = BTRFS_EXTENT_ITEM_KEY |
fec577fb CM |
1447 | * ins->offset == number of blocks |
1448 | * Any available blocks before search_start are skipped. | |
1449 | */ | |
98ed5174 CM |
1450 | static int noinline find_free_extent(struct btrfs_trans_handle *trans, |
1451 | struct btrfs_root *orig_root, | |
1452 | u64 num_bytes, u64 empty_size, | |
1453 | u64 search_start, u64 search_end, | |
1454 | u64 hint_byte, struct btrfs_key *ins, | |
1455 | u64 exclude_start, u64 exclude_nr, | |
1456 | int data) | |
fec577fb | 1457 | { |
5caf2a00 | 1458 | struct btrfs_path *path; |
e2fa7227 | 1459 | struct btrfs_key key; |
fec577fb | 1460 | u64 hole_size = 0; |
87ee04eb CM |
1461 | u64 aligned; |
1462 | int ret; | |
fec577fb | 1463 | int slot = 0; |
db94535d | 1464 | u64 last_byte = 0; |
4529ba49 | 1465 | u64 *last_ptr = NULL; |
be744175 | 1466 | u64 orig_search_start = search_start; |
fec577fb | 1467 | int start_found; |
5f39d397 | 1468 | struct extent_buffer *l; |
9f5fae2f | 1469 | struct btrfs_root * root = orig_root->fs_info->extent_root; |
f2458e1d | 1470 | struct btrfs_fs_info *info = root->fs_info; |
db94535d | 1471 | u64 total_needed = num_bytes; |
e20d96d6 | 1472 | int level; |
be08c1b9 | 1473 | struct btrfs_block_group_cache *block_group; |
be744175 | 1474 | int full_scan = 0; |
fbdc762b | 1475 | int wrapped = 0; |
d7fc640e | 1476 | int empty_cluster; |
f84a8b36 | 1477 | u64 cached_start; |
fec577fb | 1478 | |
db94535d | 1479 | WARN_ON(num_bytes < root->sectorsize); |
b1a4d965 CM |
1480 | btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY); |
1481 | ||
5f39d397 CM |
1482 | level = btrfs_header_level(root->node); |
1483 | ||
015a739c | 1484 | if (num_bytes >= 32 * 1024 * 1024 && hint_byte) { |
257d0ce3 CM |
1485 | data = BTRFS_BLOCK_GROUP_MIXED; |
1486 | } | |
1487 | ||
d7fc640e | 1488 | if (!data) { |
60cde612 | 1489 | last_ptr = &root->fs_info->last_alloc; |
d7fc640e CM |
1490 | empty_cluster = 128 * 1024; |
1491 | } | |
60cde612 | 1492 | |
d7fc640e | 1493 | if (data && btrfs_test_opt(root, SSD)) { |
60cde612 | 1494 | last_ptr = &root->fs_info->last_data_alloc; |
d7fc640e CM |
1495 | empty_cluster = 2 * 1024 * 1024; |
1496 | } | |
60cde612 CM |
1497 | |
1498 | if (last_ptr) { | |
1499 | if (*last_ptr) | |
4529ba49 | 1500 | hint_byte = *last_ptr; |
4529ba49 CM |
1501 | else { |
1502 | hint_byte = hint_byte & | |
1503 | ~((u64)BTRFS_BLOCK_GROUP_SIZE - 1); | |
d7fc640e | 1504 | empty_size += empty_cluster; |
e18e4809 | 1505 | } |
d7fc640e | 1506 | search_start = max(search_start, hint_byte); |
e18e4809 CM |
1507 | } |
1508 | ||
c31f8830 CM |
1509 | search_end = min(search_end, |
1510 | btrfs_super_total_bytes(&info->super_copy)); | |
db94535d CM |
1511 | if (hint_byte) { |
1512 | block_group = btrfs_lookup_block_group(info, hint_byte); | |
1a2b2ac7 CM |
1513 | if (!block_group) |
1514 | hint_byte = search_start; | |
be744175 | 1515 | block_group = btrfs_find_block_group(root, block_group, |
db94535d | 1516 | hint_byte, data, 1); |
be744175 CM |
1517 | } else { |
1518 | block_group = btrfs_find_block_group(root, | |
1a2b2ac7 CM |
1519 | trans->block_group, |
1520 | search_start, data, 1); | |
be744175 CM |
1521 | } |
1522 | ||
6702ed49 | 1523 | total_needed += empty_size; |
e011599b | 1524 | path = btrfs_alloc_path(); |
be744175 | 1525 | check_failed: |
70b043f0 CM |
1526 | if (!block_group) { |
1527 | block_group = btrfs_lookup_block_group(info, search_start); | |
1528 | if (!block_group) | |
1529 | block_group = btrfs_lookup_block_group(info, | |
1530 | orig_search_start); | |
1531 | } | |
5e5745dc | 1532 | search_start = find_search_start(root, &block_group, search_start, |
d548ee51 | 1533 | total_needed, data); |
e18e4809 | 1534 | |
60cde612 CM |
1535 | if (last_ptr && *last_ptr && search_start != *last_ptr) { |
1536 | *last_ptr = 0; | |
e18e4809 | 1537 | if (!empty_size) { |
d7fc640e | 1538 | empty_size += empty_cluster; |
e18e4809 CM |
1539 | total_needed += empty_size; |
1540 | } | |
1541 | search_start = find_search_start(root, &block_group, | |
1542 | search_start, total_needed, | |
1543 | data); | |
1544 | } | |
1545 | ||
87ee04eb | 1546 | search_start = stripe_align(root, search_start); |
f84a8b36 | 1547 | cached_start = search_start; |
5caf2a00 | 1548 | btrfs_init_path(path); |
fec577fb CM |
1549 | ins->objectid = search_start; |
1550 | ins->offset = 0; | |
fec577fb | 1551 | start_found = 0; |
2cc58cf2 | 1552 | path->reada = 2; |
e37c9e69 | 1553 | |
5caf2a00 | 1554 | ret = btrfs_search_slot(trans, root, ins, path, 0, 0); |
0f70abe2 CM |
1555 | if (ret < 0) |
1556 | goto error; | |
d548ee51 Y |
1557 | ret = find_previous_extent(root, path); |
1558 | if (ret < 0) | |
1559 | goto error; | |
5f39d397 CM |
1560 | l = path->nodes[0]; |
1561 | btrfs_item_key_to_cpu(l, &key, path->slots[0]); | |
fec577fb | 1562 | while (1) { |
5f39d397 | 1563 | l = path->nodes[0]; |
5caf2a00 | 1564 | slot = path->slots[0]; |
5f39d397 | 1565 | if (slot >= btrfs_header_nritems(l)) { |
5caf2a00 | 1566 | ret = btrfs_next_leaf(root, path); |
fec577fb CM |
1567 | if (ret == 0) |
1568 | continue; | |
0f70abe2 CM |
1569 | if (ret < 0) |
1570 | goto error; | |
e19caa5f CM |
1571 | |
1572 | search_start = max(search_start, | |
1573 | block_group->key.objectid); | |
fec577fb | 1574 | if (!start_found) { |
87ee04eb CM |
1575 | aligned = stripe_align(root, search_start); |
1576 | ins->objectid = aligned; | |
1577 | if (aligned >= search_end) { | |
1578 | ret = -ENOSPC; | |
1579 | goto error; | |
1580 | } | |
1581 | ins->offset = search_end - aligned; | |
fec577fb CM |
1582 | start_found = 1; |
1583 | goto check_pending; | |
1584 | } | |
87ee04eb CM |
1585 | ins->objectid = stripe_align(root, |
1586 | last_byte > search_start ? | |
1587 | last_byte : search_start); | |
1588 | if (search_end <= ins->objectid) { | |
1589 | ret = -ENOSPC; | |
1590 | goto error; | |
1591 | } | |
3e1ad54f | 1592 | ins->offset = search_end - ins->objectid; |
e19caa5f | 1593 | BUG_ON(ins->objectid >= search_end); |
fec577fb CM |
1594 | goto check_pending; |
1595 | } | |
5f39d397 | 1596 | btrfs_item_key_to_cpu(l, &key, slot); |
96b5179d | 1597 | |
db94535d | 1598 | if (key.objectid >= search_start && key.objectid > last_byte && |
e37c9e69 | 1599 | start_found) { |
db94535d CM |
1600 | if (last_byte < search_start) |
1601 | last_byte = search_start; | |
87ee04eb CM |
1602 | aligned = stripe_align(root, last_byte); |
1603 | hole_size = key.objectid - aligned; | |
1604 | if (key.objectid > aligned && hole_size >= num_bytes) { | |
1605 | ins->objectid = aligned; | |
e37c9e69 CM |
1606 | ins->offset = hole_size; |
1607 | goto check_pending; | |
0579da42 | 1608 | } |
fec577fb | 1609 | } |
96b5179d | 1610 | if (btrfs_key_type(&key) != BTRFS_EXTENT_ITEM_KEY) { |
7bb86316 CM |
1611 | if (!start_found && btrfs_key_type(&key) == |
1612 | BTRFS_BLOCK_GROUP_ITEM_KEY) { | |
db94535d | 1613 | last_byte = key.objectid; |
96b5179d CM |
1614 | start_found = 1; |
1615 | } | |
e37c9e69 | 1616 | goto next; |
96b5179d CM |
1617 | } |
1618 | ||
e37c9e69 | 1619 | |
0579da42 | 1620 | start_found = 1; |
db94535d | 1621 | last_byte = key.objectid + key.offset; |
f510cfec | 1622 | |
257d0ce3 CM |
1623 | if (!full_scan && data != BTRFS_BLOCK_GROUP_MIXED && |
1624 | last_byte >= block_group->key.objectid + | |
be744175 CM |
1625 | block_group->key.offset) { |
1626 | btrfs_release_path(root, path); | |
1627 | search_start = block_group->key.objectid + | |
e19caa5f | 1628 | block_group->key.offset; |
be744175 CM |
1629 | goto new_group; |
1630 | } | |
9078a3e1 | 1631 | next: |
5caf2a00 | 1632 | path->slots[0]++; |
de428b63 | 1633 | cond_resched(); |
fec577fb | 1634 | } |
fec577fb CM |
1635 | check_pending: |
1636 | /* we have to make sure we didn't find an extent that has already | |
1637 | * been allocated by the map tree or the original allocation | |
1638 | */ | |
5caf2a00 | 1639 | btrfs_release_path(root, path); |
fec577fb | 1640 | BUG_ON(ins->objectid < search_start); |
e37c9e69 | 1641 | |
db94535d | 1642 | if (ins->objectid + num_bytes >= search_end) |
cf67582b | 1643 | goto enospc; |
257d0ce3 | 1644 | if (!full_scan && data != BTRFS_BLOCK_GROUP_MIXED && |
5cf66426 | 1645 | ins->objectid + num_bytes > block_group-> |
e19caa5f CM |
1646 | key.objectid + block_group->key.offset) { |
1647 | search_start = block_group->key.objectid + | |
1648 | block_group->key.offset; | |
1649 | goto new_group; | |
1650 | } | |
1a5bc167 | 1651 | if (test_range_bit(&info->extent_ins, ins->objectid, |
db94535d CM |
1652 | ins->objectid + num_bytes -1, EXTENT_LOCKED, 0)) { |
1653 | search_start = ins->objectid + num_bytes; | |
1a5bc167 CM |
1654 | goto new_group; |
1655 | } | |
1656 | if (test_range_bit(&info->pinned_extents, ins->objectid, | |
db94535d CM |
1657 | ins->objectid + num_bytes -1, EXTENT_DIRTY, 0)) { |
1658 | search_start = ins->objectid + num_bytes; | |
1a5bc167 | 1659 | goto new_group; |
fec577fb | 1660 | } |
db94535d | 1661 | if (exclude_nr > 0 && (ins->objectid + num_bytes > exclude_start && |
f2654de4 CM |
1662 | ins->objectid < exclude_start + exclude_nr)) { |
1663 | search_start = exclude_start + exclude_nr; | |
1664 | goto new_group; | |
1665 | } | |
e37c9e69 | 1666 | if (!data) { |
5276aeda | 1667 | block_group = btrfs_lookup_block_group(info, ins->objectid); |
26b8003f CM |
1668 | if (block_group) |
1669 | trans->block_group = block_group; | |
f2458e1d | 1670 | } |
db94535d | 1671 | ins->offset = num_bytes; |
5caf2a00 | 1672 | btrfs_free_path(path); |
068fe39f | 1673 | if (last_ptr) { |
4529ba49 | 1674 | *last_ptr = ins->objectid + ins->offset; |
068fe39f CM |
1675 | if (*last_ptr == |
1676 | btrfs_super_total_bytes(&root->fs_info->super_copy)) { | |
1677 | *last_ptr = 0; | |
1678 | } | |
1679 | } | |
fec577fb | 1680 | return 0; |
be744175 CM |
1681 | |
1682 | new_group: | |
db94535d | 1683 | if (search_start + num_bytes >= search_end) { |
cf67582b | 1684 | enospc: |
be744175 | 1685 | search_start = orig_search_start; |
fbdc762b CM |
1686 | if (full_scan) { |
1687 | ret = -ENOSPC; | |
1688 | goto error; | |
1689 | } | |
6702ed49 CM |
1690 | if (wrapped) { |
1691 | if (!full_scan) | |
1692 | total_needed -= empty_size; | |
fbdc762b | 1693 | full_scan = 1; |
1a2b2ac7 | 1694 | data = BTRFS_BLOCK_GROUP_MIXED; |
6702ed49 | 1695 | } else |
fbdc762b | 1696 | wrapped = 1; |
be744175 | 1697 | } |
5276aeda | 1698 | block_group = btrfs_lookup_block_group(info, search_start); |
fbdc762b | 1699 | cond_resched(); |
1a2b2ac7 CM |
1700 | block_group = btrfs_find_block_group(root, block_group, |
1701 | search_start, data, 0); | |
be744175 CM |
1702 | goto check_failed; |
1703 | ||
0f70abe2 | 1704 | error: |
5caf2a00 CM |
1705 | btrfs_release_path(root, path); |
1706 | btrfs_free_path(path); | |
0f70abe2 | 1707 | return ret; |
fec577fb | 1708 | } |
fec577fb CM |
1709 | /* |
1710 | * finds a free extent and does all the dirty work required for allocation | |
1711 | * returns the key for the extent through ins, and a tree buffer for | |
1712 | * the first block of the extent through buf. | |
1713 | * | |
1714 | * returns 0 if everything worked, non-zero otherwise. | |
1715 | */ | |
4d775673 | 1716 | int btrfs_alloc_extent(struct btrfs_trans_handle *trans, |
7bb86316 CM |
1717 | struct btrfs_root *root, |
1718 | u64 num_bytes, u64 root_objectid, u64 ref_generation, | |
1719 | u64 owner, u64 owner_offset, | |
1720 | u64 empty_size, u64 hint_byte, | |
be08c1b9 | 1721 | u64 search_end, struct btrfs_key *ins, int data) |
fec577fb CM |
1722 | { |
1723 | int ret; | |
1724 | int pending_ret; | |
c31f8830 CM |
1725 | u64 super_used; |
1726 | u64 root_used; | |
fbdc762b | 1727 | u64 search_start = 0; |
edbd8d4e | 1728 | u64 new_hint; |
47e4bb98 | 1729 | u32 sizes[2]; |
1261ec42 CM |
1730 | struct btrfs_fs_info *info = root->fs_info; |
1731 | struct btrfs_root *extent_root = info->extent_root; | |
47e4bb98 CM |
1732 | struct btrfs_extent_item *extent_item; |
1733 | struct btrfs_extent_ref *ref; | |
7bb86316 | 1734 | struct btrfs_path *path; |
47e4bb98 | 1735 | struct btrfs_key keys[2]; |
8f662a76 CM |
1736 | |
1737 | new_hint = max(hint_byte, root->fs_info->alloc_start); | |
edbd8d4e CM |
1738 | if (new_hint < btrfs_super_total_bytes(&info->super_copy)) |
1739 | hint_byte = new_hint; | |
8f662a76 | 1740 | |
db94535d CM |
1741 | WARN_ON(num_bytes < root->sectorsize); |
1742 | ret = find_free_extent(trans, root, num_bytes, empty_size, | |
1743 | search_start, search_end, hint_byte, ins, | |
26b8003f CM |
1744 | trans->alloc_exclude_start, |
1745 | trans->alloc_exclude_nr, data); | |
ccd467d6 | 1746 | BUG_ON(ret); |
f2654de4 CM |
1747 | if (ret) |
1748 | return ret; | |
fec577fb | 1749 | |
58176a96 | 1750 | /* block accounting for super block */ |
db94535d CM |
1751 | super_used = btrfs_super_bytes_used(&info->super_copy); |
1752 | btrfs_set_super_bytes_used(&info->super_copy, super_used + num_bytes); | |
26b8003f | 1753 | |
58176a96 | 1754 | /* block accounting for root item */ |
db94535d CM |
1755 | root_used = btrfs_root_used(&root->root_item); |
1756 | btrfs_set_root_used(&root->root_item, root_used + num_bytes); | |
58176a96 | 1757 | |
f510cfec CM |
1758 | clear_extent_dirty(&root->fs_info->free_space_cache, |
1759 | ins->objectid, ins->objectid + ins->offset - 1, | |
1760 | GFP_NOFS); | |
1761 | ||
26b8003f | 1762 | if (root == extent_root) { |
1a5bc167 CM |
1763 | set_extent_bits(&root->fs_info->extent_ins, ins->objectid, |
1764 | ins->objectid + ins->offset - 1, | |
1765 | EXTENT_LOCKED, GFP_NOFS); | |
e19caa5f | 1766 | WARN_ON(data == 1); |
26b8003f CM |
1767 | goto update_block; |
1768 | } | |
1769 | ||
1770 | WARN_ON(trans->alloc_exclude_nr); | |
1771 | trans->alloc_exclude_start = ins->objectid; | |
1772 | trans->alloc_exclude_nr = ins->offset; | |
037e6390 | 1773 | |
47e4bb98 CM |
1774 | memcpy(&keys[0], ins, sizeof(*ins)); |
1775 | keys[1].offset = hash_extent_ref(root_objectid, ref_generation, | |
1776 | owner, owner_offset); | |
1777 | keys[1].objectid = ins->objectid; | |
1778 | keys[1].type = BTRFS_EXTENT_REF_KEY; | |
1779 | sizes[0] = sizeof(*extent_item); | |
1780 | sizes[1] = sizeof(*ref); | |
7bb86316 CM |
1781 | |
1782 | path = btrfs_alloc_path(); | |
1783 | BUG_ON(!path); | |
47e4bb98 CM |
1784 | |
1785 | ret = btrfs_insert_empty_items(trans, extent_root, path, keys, | |
1786 | sizes, 2); | |
26b8003f | 1787 | |
ccd467d6 | 1788 | BUG_ON(ret); |
47e4bb98 CM |
1789 | extent_item = btrfs_item_ptr(path->nodes[0], path->slots[0], |
1790 | struct btrfs_extent_item); | |
1791 | btrfs_set_extent_refs(path->nodes[0], extent_item, 1); | |
1792 | ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1, | |
1793 | struct btrfs_extent_ref); | |
1794 | ||
1795 | btrfs_set_ref_root(path->nodes[0], ref, root_objectid); | |
1796 | btrfs_set_ref_generation(path->nodes[0], ref, ref_generation); | |
1797 | btrfs_set_ref_objectid(path->nodes[0], ref, owner); | |
1798 | btrfs_set_ref_offset(path->nodes[0], ref, owner_offset); | |
1799 | ||
1800 | btrfs_mark_buffer_dirty(path->nodes[0]); | |
1801 | ||
1802 | trans->alloc_exclude_start = 0; | |
1803 | trans->alloc_exclude_nr = 0; | |
7bb86316 | 1804 | btrfs_free_path(path); |
e089f05c | 1805 | finish_current_insert(trans, extent_root); |
e20d96d6 | 1806 | pending_ret = del_pending_extents(trans, extent_root); |
f510cfec | 1807 | |
e37c9e69 | 1808 | if (ret) { |
037e6390 | 1809 | return ret; |
e37c9e69 CM |
1810 | } |
1811 | if (pending_ret) { | |
037e6390 | 1812 | return pending_ret; |
e37c9e69 | 1813 | } |
26b8003f CM |
1814 | |
1815 | update_block: | |
1e2677e0 CM |
1816 | ret = update_block_group(trans, root, ins->objectid, ins->offset, 1, 0, |
1817 | data); | |
f5947066 CM |
1818 | if (ret) { |
1819 | printk("update block group failed for %Lu %Lu\n", | |
1820 | ins->objectid, ins->offset); | |
1821 | BUG(); | |
1822 | } | |
037e6390 | 1823 | return 0; |
fec577fb CM |
1824 | } |
1825 | ||
1826 | /* | |
1827 | * helper function to allocate a block for a given tree | |
1828 | * returns the tree buffer or NULL. | |
1829 | */ | |
5f39d397 | 1830 | struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans, |
db94535d | 1831 | struct btrfs_root *root, |
7bb86316 CM |
1832 | u32 blocksize, |
1833 | u64 root_objectid, u64 hint, | |
1834 | u64 empty_size) | |
1835 | { | |
1836 | u64 ref_generation; | |
1837 | ||
1838 | if (root->ref_cows) | |
1839 | ref_generation = trans->transid; | |
1840 | else | |
1841 | ref_generation = 0; | |
1842 | ||
1843 | ||
1844 | return __btrfs_alloc_free_block(trans, root, blocksize, root_objectid, | |
1845 | ref_generation, 0, 0, hint, empty_size); | |
1846 | } | |
1847 | ||
1848 | /* | |
1849 | * helper function to allocate a block for a given tree | |
1850 | * returns the tree buffer or NULL. | |
1851 | */ | |
1852 | struct extent_buffer *__btrfs_alloc_free_block(struct btrfs_trans_handle *trans, | |
1853 | struct btrfs_root *root, | |
1854 | u32 blocksize, | |
1855 | u64 root_objectid, | |
1856 | u64 ref_generation, | |
1857 | u64 first_objectid, | |
1858 | int level, | |
1859 | u64 hint, | |
5f39d397 | 1860 | u64 empty_size) |
fec577fb | 1861 | { |
e2fa7227 | 1862 | struct btrfs_key ins; |
fec577fb | 1863 | int ret; |
5f39d397 | 1864 | struct extent_buffer *buf; |
fec577fb | 1865 | |
7bb86316 CM |
1866 | ret = btrfs_alloc_extent(trans, root, blocksize, |
1867 | root_objectid, ref_generation, | |
f6dbff55 | 1868 | level, first_objectid, empty_size, hint, |
db94535d | 1869 | (u64)-1, &ins, 0); |
fec577fb | 1870 | if (ret) { |
54aa1f4d CM |
1871 | BUG_ON(ret > 0); |
1872 | return ERR_PTR(ret); | |
fec577fb | 1873 | } |
db94535d | 1874 | buf = btrfs_find_create_tree_block(root, ins.objectid, blocksize); |
54aa1f4d | 1875 | if (!buf) { |
7bb86316 CM |
1876 | btrfs_free_extent(trans, root, ins.objectid, blocksize, |
1877 | root->root_key.objectid, ref_generation, | |
1878 | 0, 0, 0); | |
54aa1f4d CM |
1879 | return ERR_PTR(-ENOMEM); |
1880 | } | |
55c69072 CM |
1881 | btrfs_set_header_generation(buf, trans->transid); |
1882 | clean_tree_block(trans, root, buf); | |
1883 | wait_on_tree_block_writeback(root, buf); | |
5f39d397 | 1884 | btrfs_set_buffer_uptodate(buf); |
55c69072 CM |
1885 | |
1886 | if (PageDirty(buf->first_page)) { | |
1887 | printk("page %lu dirty\n", buf->first_page->index); | |
1888 | WARN_ON(1); | |
1889 | } | |
1890 | ||
5f39d397 CM |
1891 | set_extent_dirty(&trans->transaction->dirty_pages, buf->start, |
1892 | buf->start + buf->len - 1, GFP_NOFS); | |
d1310b2e | 1893 | set_extent_bits(&BTRFS_I(root->fs_info->btree_inode)->io_tree, |
19c00ddc CM |
1894 | buf->start, buf->start + buf->len - 1, |
1895 | EXTENT_CSUM, GFP_NOFS); | |
1896 | buf->flags |= EXTENT_CSUM; | |
9afbb0b7 CM |
1897 | if (!btrfs_test_opt(root, SSD)) |
1898 | btrfs_set_buffer_defrag(buf); | |
d3c2fdcf | 1899 | trans->blocks_used++; |
fec577fb CM |
1900 | return buf; |
1901 | } | |
a28ec197 | 1902 | |
98ed5174 CM |
1903 | static int noinline drop_leaf_ref(struct btrfs_trans_handle *trans, |
1904 | struct btrfs_root *root, | |
1905 | struct extent_buffer *leaf) | |
6407bf6d | 1906 | { |
7bb86316 CM |
1907 | u64 leaf_owner; |
1908 | u64 leaf_generation; | |
5f39d397 | 1909 | struct btrfs_key key; |
6407bf6d CM |
1910 | struct btrfs_file_extent_item *fi; |
1911 | int i; | |
1912 | int nritems; | |
1913 | int ret; | |
1914 | ||
5f39d397 CM |
1915 | BUG_ON(!btrfs_is_leaf(leaf)); |
1916 | nritems = btrfs_header_nritems(leaf); | |
7bb86316 CM |
1917 | leaf_owner = btrfs_header_owner(leaf); |
1918 | leaf_generation = btrfs_header_generation(leaf); | |
1919 | ||
6407bf6d | 1920 | for (i = 0; i < nritems; i++) { |
db94535d | 1921 | u64 disk_bytenr; |
5f39d397 CM |
1922 | |
1923 | btrfs_item_key_to_cpu(leaf, &key, i); | |
1924 | if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY) | |
6407bf6d CM |
1925 | continue; |
1926 | fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item); | |
5f39d397 CM |
1927 | if (btrfs_file_extent_type(leaf, fi) == |
1928 | BTRFS_FILE_EXTENT_INLINE) | |
236454df | 1929 | continue; |
6407bf6d CM |
1930 | /* |
1931 | * FIXME make sure to insert a trans record that | |
1932 | * repeats the snapshot del on crash | |
1933 | */ | |
db94535d CM |
1934 | disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); |
1935 | if (disk_bytenr == 0) | |
3a686375 | 1936 | continue; |
db94535d | 1937 | ret = btrfs_free_extent(trans, root, disk_bytenr, |
7bb86316 CM |
1938 | btrfs_file_extent_disk_num_bytes(leaf, fi), |
1939 | leaf_owner, leaf_generation, | |
1940 | key.objectid, key.offset, 0); | |
6407bf6d CM |
1941 | BUG_ON(ret); |
1942 | } | |
1943 | return 0; | |
1944 | } | |
1945 | ||
98ed5174 | 1946 | static void noinline reada_walk_down(struct btrfs_root *root, |
bea495e5 CM |
1947 | struct extent_buffer *node, |
1948 | int slot) | |
e011599b | 1949 | { |
db94535d | 1950 | u64 bytenr; |
bea495e5 CM |
1951 | u64 last = 0; |
1952 | u32 nritems; | |
e011599b | 1953 | u32 refs; |
db94535d | 1954 | u32 blocksize; |
bea495e5 CM |
1955 | int ret; |
1956 | int i; | |
1957 | int level; | |
1958 | int skipped = 0; | |
e011599b | 1959 | |
5f39d397 | 1960 | nritems = btrfs_header_nritems(node); |
db94535d | 1961 | level = btrfs_header_level(node); |
bea495e5 CM |
1962 | if (level) |
1963 | return; | |
1964 | ||
1965 | for (i = slot; i < nritems && skipped < 32; i++) { | |
db94535d | 1966 | bytenr = btrfs_node_blockptr(node, i); |
bea495e5 CM |
1967 | if (last && ((bytenr > last && bytenr - last > 32 * 1024) || |
1968 | (last > bytenr && last - bytenr > 32 * 1024))) { | |
1969 | skipped++; | |
e011599b | 1970 | continue; |
bea495e5 CM |
1971 | } |
1972 | blocksize = btrfs_level_size(root, level - 1); | |
1973 | if (i != slot) { | |
1974 | ret = lookup_extent_ref(NULL, root, bytenr, | |
1975 | blocksize, &refs); | |
1976 | BUG_ON(ret); | |
1977 | if (refs != 1) { | |
1978 | skipped++; | |
1979 | continue; | |
1980 | } | |
1981 | } | |
409eb95d | 1982 | mutex_unlock(&root->fs_info->fs_mutex); |
db94535d | 1983 | ret = readahead_tree_block(root, bytenr, blocksize); |
bea495e5 | 1984 | last = bytenr + blocksize; |
409eb95d CM |
1985 | cond_resched(); |
1986 | mutex_lock(&root->fs_info->fs_mutex); | |
e011599b CM |
1987 | if (ret) |
1988 | break; | |
1989 | } | |
1990 | } | |
1991 | ||
9aca1d51 CM |
1992 | /* |
1993 | * helper function for drop_snapshot, this walks down the tree dropping ref | |
1994 | * counts as it goes. | |
1995 | */ | |
98ed5174 CM |
1996 | static int noinline walk_down_tree(struct btrfs_trans_handle *trans, |
1997 | struct btrfs_root *root, | |
1998 | struct btrfs_path *path, int *level) | |
20524f02 | 1999 | { |
7bb86316 CM |
2000 | u64 root_owner; |
2001 | u64 root_gen; | |
2002 | u64 bytenr; | |
5f39d397 CM |
2003 | struct extent_buffer *next; |
2004 | struct extent_buffer *cur; | |
7bb86316 | 2005 | struct extent_buffer *parent; |
db94535d | 2006 | u32 blocksize; |
20524f02 CM |
2007 | int ret; |
2008 | u32 refs; | |
2009 | ||
5caf2a00 CM |
2010 | WARN_ON(*level < 0); |
2011 | WARN_ON(*level >= BTRFS_MAX_LEVEL); | |
5f39d397 | 2012 | ret = lookup_extent_ref(trans, root, |
db94535d CM |
2013 | path->nodes[*level]->start, |
2014 | path->nodes[*level]->len, &refs); | |
20524f02 CM |
2015 | BUG_ON(ret); |
2016 | if (refs > 1) | |
2017 | goto out; | |
e011599b | 2018 | |
9aca1d51 CM |
2019 | /* |
2020 | * walk down to the last node level and free all the leaves | |
2021 | */ | |
6407bf6d | 2022 | while(*level >= 0) { |
5caf2a00 CM |
2023 | WARN_ON(*level < 0); |
2024 | WARN_ON(*level >= BTRFS_MAX_LEVEL); | |
20524f02 | 2025 | cur = path->nodes[*level]; |
e011599b | 2026 | |
5f39d397 | 2027 | if (btrfs_header_level(cur) != *level) |
2c90e5d6 | 2028 | WARN_ON(1); |
e011599b | 2029 | |
7518a238 | 2030 | if (path->slots[*level] >= |
5f39d397 | 2031 | btrfs_header_nritems(cur)) |
20524f02 | 2032 | break; |
6407bf6d CM |
2033 | if (*level == 0) { |
2034 | ret = drop_leaf_ref(trans, root, cur); | |
2035 | BUG_ON(ret); | |
2036 | break; | |
2037 | } | |
db94535d CM |
2038 | bytenr = btrfs_node_blockptr(cur, path->slots[*level]); |
2039 | blocksize = btrfs_level_size(root, *level - 1); | |
2040 | ret = lookup_extent_ref(trans, root, bytenr, blocksize, &refs); | |
6407bf6d CM |
2041 | BUG_ON(ret); |
2042 | if (refs != 1) { | |
7bb86316 CM |
2043 | parent = path->nodes[*level]; |
2044 | root_owner = btrfs_header_owner(parent); | |
2045 | root_gen = btrfs_header_generation(parent); | |
20524f02 | 2046 | path->slots[*level]++; |
db94535d | 2047 | ret = btrfs_free_extent(trans, root, bytenr, |
7bb86316 CM |
2048 | blocksize, root_owner, |
2049 | root_gen, 0, 0, 1); | |
20524f02 CM |
2050 | BUG_ON(ret); |
2051 | continue; | |
2052 | } | |
db94535d | 2053 | next = btrfs_find_tree_block(root, bytenr, blocksize); |
5f39d397 CM |
2054 | if (!next || !btrfs_buffer_uptodate(next)) { |
2055 | free_extent_buffer(next); | |
bea495e5 | 2056 | reada_walk_down(root, cur, path->slots[*level]); |
e9d0b13b | 2057 | mutex_unlock(&root->fs_info->fs_mutex); |
db94535d | 2058 | next = read_tree_block(root, bytenr, blocksize); |
e9d0b13b CM |
2059 | mutex_lock(&root->fs_info->fs_mutex); |
2060 | ||
2061 | /* we dropped the lock, check one more time */ | |
db94535d CM |
2062 | ret = lookup_extent_ref(trans, root, bytenr, |
2063 | blocksize, &refs); | |
e9d0b13b CM |
2064 | BUG_ON(ret); |
2065 | if (refs != 1) { | |
7bb86316 CM |
2066 | parent = path->nodes[*level]; |
2067 | root_owner = btrfs_header_owner(parent); | |
2068 | root_gen = btrfs_header_generation(parent); | |
2069 | ||
e9d0b13b | 2070 | path->slots[*level]++; |
5f39d397 | 2071 | free_extent_buffer(next); |
7bb86316 CM |
2072 | ret = btrfs_free_extent(trans, root, bytenr, |
2073 | blocksize, | |
2074 | root_owner, | |
2075 | root_gen, 0, 0, 1); | |
e9d0b13b CM |
2076 | BUG_ON(ret); |
2077 | continue; | |
2078 | } | |
2079 | } | |
5caf2a00 | 2080 | WARN_ON(*level <= 0); |
83e15a28 | 2081 | if (path->nodes[*level-1]) |
5f39d397 | 2082 | free_extent_buffer(path->nodes[*level-1]); |
20524f02 | 2083 | path->nodes[*level-1] = next; |
5f39d397 | 2084 | *level = btrfs_header_level(next); |
20524f02 CM |
2085 | path->slots[*level] = 0; |
2086 | } | |
2087 | out: | |
5caf2a00 CM |
2088 | WARN_ON(*level < 0); |
2089 | WARN_ON(*level >= BTRFS_MAX_LEVEL); | |
7bb86316 CM |
2090 | |
2091 | if (path->nodes[*level] == root->node) { | |
2092 | root_owner = root->root_key.objectid; | |
2093 | parent = path->nodes[*level]; | |
2094 | } else { | |
2095 | parent = path->nodes[*level + 1]; | |
2096 | root_owner = btrfs_header_owner(parent); | |
2097 | } | |
2098 | ||
2099 | root_gen = btrfs_header_generation(parent); | |
db94535d | 2100 | ret = btrfs_free_extent(trans, root, path->nodes[*level]->start, |
7bb86316 CM |
2101 | path->nodes[*level]->len, |
2102 | root_owner, root_gen, 0, 0, 1); | |
5f39d397 | 2103 | free_extent_buffer(path->nodes[*level]); |
20524f02 CM |
2104 | path->nodes[*level] = NULL; |
2105 | *level += 1; | |
2106 | BUG_ON(ret); | |
2107 | return 0; | |
2108 | } | |
2109 | ||
9aca1d51 CM |
2110 | /* |
2111 | * helper for dropping snapshots. This walks back up the tree in the path | |
2112 | * to find the first node higher up where we haven't yet gone through | |
2113 | * all the slots | |
2114 | */ | |
98ed5174 CM |
2115 | static int noinline walk_up_tree(struct btrfs_trans_handle *trans, |
2116 | struct btrfs_root *root, | |
2117 | struct btrfs_path *path, int *level) | |
20524f02 | 2118 | { |
7bb86316 CM |
2119 | u64 root_owner; |
2120 | u64 root_gen; | |
2121 | struct btrfs_root_item *root_item = &root->root_item; | |
20524f02 CM |
2122 | int i; |
2123 | int slot; | |
2124 | int ret; | |
9f3a7427 | 2125 | |
234b63a0 | 2126 | for(i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) { |
20524f02 | 2127 | slot = path->slots[i]; |
5f39d397 CM |
2128 | if (slot < btrfs_header_nritems(path->nodes[i]) - 1) { |
2129 | struct extent_buffer *node; | |
2130 | struct btrfs_disk_key disk_key; | |
2131 | node = path->nodes[i]; | |
20524f02 CM |
2132 | path->slots[i]++; |
2133 | *level = i; | |
9f3a7427 | 2134 | WARN_ON(*level == 0); |
5f39d397 | 2135 | btrfs_node_key(node, &disk_key, path->slots[i]); |
9f3a7427 | 2136 | memcpy(&root_item->drop_progress, |
5f39d397 | 2137 | &disk_key, sizeof(disk_key)); |
9f3a7427 | 2138 | root_item->drop_level = i; |
20524f02 CM |
2139 | return 0; |
2140 | } else { | |
7bb86316 CM |
2141 | if (path->nodes[*level] == root->node) { |
2142 | root_owner = root->root_key.objectid; | |
2143 | root_gen = | |
2144 | btrfs_header_generation(path->nodes[*level]); | |
2145 | } else { | |
2146 | struct extent_buffer *node; | |
2147 | node = path->nodes[*level + 1]; | |
2148 | root_owner = btrfs_header_owner(node); | |
2149 | root_gen = btrfs_header_generation(node); | |
2150 | } | |
e089f05c | 2151 | ret = btrfs_free_extent(trans, root, |
db94535d | 2152 | path->nodes[*level]->start, |
7bb86316 CM |
2153 | path->nodes[*level]->len, |
2154 | root_owner, root_gen, 0, 0, 1); | |
6407bf6d | 2155 | BUG_ON(ret); |
5f39d397 | 2156 | free_extent_buffer(path->nodes[*level]); |
83e15a28 | 2157 | path->nodes[*level] = NULL; |
20524f02 | 2158 | *level = i + 1; |
20524f02 CM |
2159 | } |
2160 | } | |
2161 | return 1; | |
2162 | } | |
2163 | ||
9aca1d51 CM |
2164 | /* |
2165 | * drop the reference count on the tree rooted at 'snap'. This traverses | |
2166 | * the tree freeing any blocks that have a ref count of zero after being | |
2167 | * decremented. | |
2168 | */ | |
e089f05c | 2169 | int btrfs_drop_snapshot(struct btrfs_trans_handle *trans, struct btrfs_root |
9f3a7427 | 2170 | *root) |
20524f02 | 2171 | { |
3768f368 | 2172 | int ret = 0; |
9aca1d51 | 2173 | int wret; |
20524f02 | 2174 | int level; |
5caf2a00 | 2175 | struct btrfs_path *path; |
20524f02 CM |
2176 | int i; |
2177 | int orig_level; | |
9f3a7427 | 2178 | struct btrfs_root_item *root_item = &root->root_item; |
20524f02 | 2179 | |
5caf2a00 CM |
2180 | path = btrfs_alloc_path(); |
2181 | BUG_ON(!path); | |
20524f02 | 2182 | |
5f39d397 | 2183 | level = btrfs_header_level(root->node); |
20524f02 | 2184 | orig_level = level; |
9f3a7427 CM |
2185 | if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) { |
2186 | path->nodes[level] = root->node; | |
f510cfec | 2187 | extent_buffer_get(root->node); |
9f3a7427 CM |
2188 | path->slots[level] = 0; |
2189 | } else { | |
2190 | struct btrfs_key key; | |
5f39d397 CM |
2191 | struct btrfs_disk_key found_key; |
2192 | struct extent_buffer *node; | |
6702ed49 | 2193 | |
9f3a7427 | 2194 | btrfs_disk_key_to_cpu(&key, &root_item->drop_progress); |
6702ed49 CM |
2195 | level = root_item->drop_level; |
2196 | path->lowest_level = level; | |
9f3a7427 | 2197 | wret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
6702ed49 | 2198 | if (wret < 0) { |
9f3a7427 CM |
2199 | ret = wret; |
2200 | goto out; | |
2201 | } | |
5f39d397 CM |
2202 | node = path->nodes[level]; |
2203 | btrfs_node_key(node, &found_key, path->slots[level]); | |
2204 | WARN_ON(memcmp(&found_key, &root_item->drop_progress, | |
2205 | sizeof(found_key))); | |
9f3a7427 | 2206 | } |
20524f02 | 2207 | while(1) { |
5caf2a00 | 2208 | wret = walk_down_tree(trans, root, path, &level); |
9aca1d51 | 2209 | if (wret > 0) |
20524f02 | 2210 | break; |
9aca1d51 CM |
2211 | if (wret < 0) |
2212 | ret = wret; | |
2213 | ||
5caf2a00 | 2214 | wret = walk_up_tree(trans, root, path, &level); |
9aca1d51 | 2215 | if (wret > 0) |
20524f02 | 2216 | break; |
9aca1d51 CM |
2217 | if (wret < 0) |
2218 | ret = wret; | |
409eb95d | 2219 | ret = -EAGAIN; |
409eb95d | 2220 | break; |
20524f02 | 2221 | } |
83e15a28 | 2222 | for (i = 0; i <= orig_level; i++) { |
5caf2a00 | 2223 | if (path->nodes[i]) { |
5f39d397 | 2224 | free_extent_buffer(path->nodes[i]); |
0f82731f | 2225 | path->nodes[i] = NULL; |
83e15a28 | 2226 | } |
20524f02 | 2227 | } |
9f3a7427 | 2228 | out: |
5caf2a00 | 2229 | btrfs_free_path(path); |
9aca1d51 | 2230 | return ret; |
20524f02 | 2231 | } |
9078a3e1 | 2232 | |
96b5179d | 2233 | int btrfs_free_block_groups(struct btrfs_fs_info *info) |
9078a3e1 | 2234 | { |
96b5179d CM |
2235 | u64 start; |
2236 | u64 end; | |
b97f9203 | 2237 | u64 ptr; |
9078a3e1 | 2238 | int ret; |
9078a3e1 | 2239 | while(1) { |
96b5179d CM |
2240 | ret = find_first_extent_bit(&info->block_group_cache, 0, |
2241 | &start, &end, (unsigned int)-1); | |
2242 | if (ret) | |
9078a3e1 | 2243 | break; |
b97f9203 Y |
2244 | ret = get_state_private(&info->block_group_cache, start, &ptr); |
2245 | if (!ret) | |
2246 | kfree((void *)(unsigned long)ptr); | |
96b5179d CM |
2247 | clear_extent_bits(&info->block_group_cache, start, |
2248 | end, (unsigned int)-1, GFP_NOFS); | |
9078a3e1 | 2249 | } |
e37c9e69 | 2250 | while(1) { |
f510cfec CM |
2251 | ret = find_first_extent_bit(&info->free_space_cache, 0, |
2252 | &start, &end, EXTENT_DIRTY); | |
2253 | if (ret) | |
e37c9e69 | 2254 | break; |
f510cfec CM |
2255 | clear_extent_dirty(&info->free_space_cache, start, |
2256 | end, GFP_NOFS); | |
e37c9e69 | 2257 | } |
be744175 CM |
2258 | return 0; |
2259 | } | |
2260 | ||
98ed5174 CM |
2261 | static int noinline relocate_inode_pages(struct inode *inode, u64 start, |
2262 | u64 len) | |
edbd8d4e CM |
2263 | { |
2264 | u64 page_start; | |
2265 | u64 page_end; | |
2266 | u64 delalloc_start; | |
2267 | u64 existing_delalloc; | |
2268 | unsigned long last_index; | |
edbd8d4e CM |
2269 | unsigned long i; |
2270 | struct page *page; | |
d1310b2e | 2271 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
4313b399 CM |
2272 | struct file_ra_state *ra; |
2273 | ||
2274 | ra = kzalloc(sizeof(*ra), GFP_NOFS); | |
edbd8d4e CM |
2275 | |
2276 | mutex_lock(&inode->i_mutex); | |
4313b399 | 2277 | i = start >> PAGE_CACHE_SHIFT; |
edbd8d4e CM |
2278 | last_index = (start + len - 1) >> PAGE_CACHE_SHIFT; |
2279 | ||
4313b399 CM |
2280 | file_ra_state_init(ra, inode->i_mapping); |
2281 | btrfs_force_ra(inode->i_mapping, ra, NULL, i, last_index); | |
2282 | kfree(ra); | |
edbd8d4e | 2283 | |
4313b399 | 2284 | for (; i <= last_index; i++) { |
edbd8d4e CM |
2285 | page = grab_cache_page(inode->i_mapping, i); |
2286 | if (!page) | |
2287 | goto out_unlock; | |
2288 | if (!PageUptodate(page)) { | |
2289 | btrfs_readpage(NULL, page); | |
2290 | lock_page(page); | |
2291 | if (!PageUptodate(page)) { | |
2292 | unlock_page(page); | |
2293 | page_cache_release(page); | |
2294 | goto out_unlock; | |
2295 | } | |
2296 | } | |
2297 | page_start = (u64)page->index << PAGE_CACHE_SHIFT; | |
2298 | page_end = page_start + PAGE_CACHE_SIZE - 1; | |
2299 | ||
d1310b2e | 2300 | lock_extent(io_tree, page_start, page_end, GFP_NOFS); |
edbd8d4e CM |
2301 | |
2302 | delalloc_start = page_start; | |
d1310b2e CM |
2303 | existing_delalloc = count_range_bits(io_tree, |
2304 | &delalloc_start, page_end, | |
2305 | PAGE_CACHE_SIZE, EXTENT_DELALLOC); | |
edbd8d4e | 2306 | |
d1310b2e | 2307 | set_extent_delalloc(io_tree, page_start, |
edbd8d4e CM |
2308 | page_end, GFP_NOFS); |
2309 | ||
d1310b2e | 2310 | unlock_extent(io_tree, page_start, page_end, GFP_NOFS); |
edbd8d4e CM |
2311 | set_page_dirty(page); |
2312 | unlock_page(page); | |
2313 | page_cache_release(page); | |
2314 | } | |
2315 | ||
2316 | out_unlock: | |
2317 | mutex_unlock(&inode->i_mutex); | |
2318 | return 0; | |
2319 | } | |
2320 | ||
4313b399 CM |
2321 | /* |
2322 | * note, this releases the path | |
2323 | */ | |
98ed5174 | 2324 | static int noinline relocate_one_reference(struct btrfs_root *extent_root, |
edbd8d4e | 2325 | struct btrfs_path *path, |
4313b399 | 2326 | struct btrfs_key *extent_key) |
edbd8d4e CM |
2327 | { |
2328 | struct inode *inode; | |
2329 | struct btrfs_root *found_root; | |
4313b399 CM |
2330 | struct btrfs_key *root_location; |
2331 | struct btrfs_extent_ref *ref; | |
2332 | u64 ref_root; | |
2333 | u64 ref_gen; | |
2334 | u64 ref_objectid; | |
2335 | u64 ref_offset; | |
edbd8d4e CM |
2336 | int ret; |
2337 | ||
4313b399 CM |
2338 | ref = btrfs_item_ptr(path->nodes[0], path->slots[0], |
2339 | struct btrfs_extent_ref); | |
2340 | ref_root = btrfs_ref_root(path->nodes[0], ref); | |
2341 | ref_gen = btrfs_ref_generation(path->nodes[0], ref); | |
2342 | ref_objectid = btrfs_ref_objectid(path->nodes[0], ref); | |
2343 | ref_offset = btrfs_ref_offset(path->nodes[0], ref); | |
2344 | btrfs_release_path(extent_root, path); | |
2345 | ||
2346 | root_location = kmalloc(sizeof(*root_location), GFP_NOFS); | |
2347 | root_location->objectid = ref_root; | |
edbd8d4e | 2348 | if (ref_gen == 0) |
4313b399 | 2349 | root_location->offset = 0; |
edbd8d4e | 2350 | else |
4313b399 CM |
2351 | root_location->offset = (u64)-1; |
2352 | root_location->type = BTRFS_ROOT_ITEM_KEY; | |
edbd8d4e CM |
2353 | |
2354 | found_root = btrfs_read_fs_root_no_name(extent_root->fs_info, | |
4313b399 | 2355 | root_location); |
edbd8d4e | 2356 | BUG_ON(!found_root); |
4313b399 | 2357 | kfree(root_location); |
edbd8d4e CM |
2358 | |
2359 | if (ref_objectid >= BTRFS_FIRST_FREE_OBJECTID) { | |
2360 | mutex_unlock(&extent_root->fs_info->fs_mutex); | |
2361 | inode = btrfs_iget_locked(extent_root->fs_info->sb, | |
2362 | ref_objectid, found_root); | |
2363 | if (inode->i_state & I_NEW) { | |
2364 | /* the inode and parent dir are two different roots */ | |
2365 | BTRFS_I(inode)->root = found_root; | |
2366 | BTRFS_I(inode)->location.objectid = ref_objectid; | |
2367 | BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY; | |
2368 | BTRFS_I(inode)->location.offset = 0; | |
2369 | btrfs_read_locked_inode(inode); | |
2370 | unlock_new_inode(inode); | |
2371 | ||
2372 | } | |
2373 | /* this can happen if the reference is not against | |
2374 | * the latest version of the tree root | |
2375 | */ | |
2376 | if (is_bad_inode(inode)) { | |
2377 | mutex_lock(&extent_root->fs_info->fs_mutex); | |
2378 | goto out; | |
2379 | } | |
2380 | relocate_inode_pages(inode, ref_offset, extent_key->offset); | |
2381 | /* FIXME, data=ordered will help get rid of this */ | |
2382 | filemap_fdatawrite(inode->i_mapping); | |
2383 | iput(inode); | |
2384 | mutex_lock(&extent_root->fs_info->fs_mutex); | |
2385 | } else { | |
2386 | struct btrfs_trans_handle *trans; | |
2387 | struct btrfs_key found_key; | |
2388 | struct extent_buffer *eb; | |
2389 | int level; | |
2390 | int i; | |
2391 | ||
2392 | trans = btrfs_start_transaction(found_root, 1); | |
2393 | eb = read_tree_block(found_root, extent_key->objectid, | |
2394 | extent_key->offset); | |
2395 | level = btrfs_header_level(eb); | |
2396 | ||
2397 | if (level == 0) | |
2398 | btrfs_item_key_to_cpu(eb, &found_key, 0); | |
2399 | else | |
2400 | btrfs_node_key_to_cpu(eb, &found_key, 0); | |
2401 | ||
2402 | free_extent_buffer(eb); | |
2403 | ||
2404 | path->lowest_level = level; | |
8f662a76 | 2405 | path->reada = 2; |
edbd8d4e CM |
2406 | ret = btrfs_search_slot(trans, found_root, &found_key, path, |
2407 | 0, 1); | |
2408 | path->lowest_level = 0; | |
2409 | for (i = level; i < BTRFS_MAX_LEVEL; i++) { | |
2410 | if (!path->nodes[i]) | |
2411 | break; | |
2412 | free_extent_buffer(path->nodes[i]); | |
2413 | path->nodes[i] = NULL; | |
2414 | } | |
2415 | btrfs_release_path(found_root, path); | |
2416 | btrfs_end_transaction(trans, found_root); | |
2417 | } | |
2418 | ||
2419 | out: | |
2420 | return 0; | |
2421 | } | |
2422 | ||
98ed5174 CM |
2423 | static int noinline relocate_one_extent(struct btrfs_root *extent_root, |
2424 | struct btrfs_path *path, | |
2425 | struct btrfs_key *extent_key) | |
edbd8d4e CM |
2426 | { |
2427 | struct btrfs_key key; | |
2428 | struct btrfs_key found_key; | |
edbd8d4e | 2429 | struct extent_buffer *leaf; |
edbd8d4e CM |
2430 | u32 nritems; |
2431 | u32 item_size; | |
2432 | int ret = 0; | |
2433 | ||
2434 | key.objectid = extent_key->objectid; | |
2435 | key.type = BTRFS_EXTENT_REF_KEY; | |
2436 | key.offset = 0; | |
2437 | ||
2438 | while(1) { | |
2439 | ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0); | |
2440 | ||
edbd8d4e CM |
2441 | if (ret < 0) |
2442 | goto out; | |
2443 | ||
2444 | ret = 0; | |
2445 | leaf = path->nodes[0]; | |
2446 | nritems = btrfs_header_nritems(leaf); | |
2447 | if (path->slots[0] == nritems) | |
2448 | goto out; | |
2449 | ||
2450 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
2451 | if (found_key.objectid != extent_key->objectid) | |
2452 | break; | |
2453 | ||
2454 | if (found_key.type != BTRFS_EXTENT_REF_KEY) | |
2455 | break; | |
2456 | ||
2457 | key.offset = found_key.offset + 1; | |
2458 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
2459 | ||
4313b399 | 2460 | ret = relocate_one_reference(extent_root, path, extent_key); |
edbd8d4e CM |
2461 | if (ret) |
2462 | goto out; | |
2463 | } | |
2464 | ret = 0; | |
2465 | out: | |
2466 | btrfs_release_path(extent_root, path); | |
2467 | return ret; | |
2468 | } | |
2469 | ||
edbd8d4e CM |
2470 | int btrfs_shrink_extent_tree(struct btrfs_root *root, u64 new_size) |
2471 | { | |
2472 | struct btrfs_trans_handle *trans; | |
2473 | struct btrfs_root *tree_root = root->fs_info->tree_root; | |
2474 | struct btrfs_path *path; | |
2475 | u64 cur_byte; | |
2476 | u64 total_found; | |
edbd8d4e | 2477 | struct btrfs_fs_info *info = root->fs_info; |
d1310b2e | 2478 | struct extent_io_tree *block_group_cache; |
edbd8d4e | 2479 | struct btrfs_key key; |
73e48b27 | 2480 | struct btrfs_key found_key; |
edbd8d4e CM |
2481 | struct extent_buffer *leaf; |
2482 | u32 nritems; | |
2483 | int ret; | |
725c8463 | 2484 | int progress = 0; |
edbd8d4e CM |
2485 | |
2486 | btrfs_set_super_total_bytes(&info->super_copy, new_size); | |
c31f8830 CM |
2487 | clear_extent_dirty(&info->free_space_cache, new_size, (u64)-1, |
2488 | GFP_NOFS); | |
edbd8d4e CM |
2489 | block_group_cache = &info->block_group_cache; |
2490 | path = btrfs_alloc_path(); | |
2491 | root = root->fs_info->extent_root; | |
8f662a76 | 2492 | path->reada = 2; |
edbd8d4e CM |
2493 | |
2494 | again: | |
2495 | total_found = 0; | |
2496 | key.objectid = new_size; | |
edbd8d4e CM |
2497 | key.offset = 0; |
2498 | key.type = 0; | |
73e48b27 | 2499 | cur_byte = key.objectid; |
4313b399 | 2500 | |
73e48b27 Y |
2501 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
2502 | if (ret < 0) | |
2503 | goto out; | |
2504 | ||
2505 | ret = find_previous_extent(root, path); | |
2506 | if (ret < 0) | |
2507 | goto out; | |
2508 | if (ret == 0) { | |
2509 | leaf = path->nodes[0]; | |
2510 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
2511 | if (found_key.objectid + found_key.offset > new_size) { | |
2512 | cur_byte = found_key.objectid; | |
2513 | key.objectid = cur_byte; | |
2514 | } | |
2515 | } | |
2516 | btrfs_release_path(root, path); | |
2517 | ||
2518 | while(1) { | |
edbd8d4e CM |
2519 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
2520 | if (ret < 0) | |
2521 | goto out; | |
73e48b27 | 2522 | |
edbd8d4e | 2523 | leaf = path->nodes[0]; |
73e48b27 Y |
2524 | nritems = btrfs_header_nritems(leaf); |
2525 | next: | |
2526 | if (path->slots[0] >= nritems) { | |
2527 | ret = btrfs_next_leaf(root, path); | |
2528 | if (ret < 0) | |
2529 | goto out; | |
2530 | if (ret == 1) { | |
2531 | ret = 0; | |
2532 | break; | |
edbd8d4e | 2533 | } |
73e48b27 Y |
2534 | leaf = path->nodes[0]; |
2535 | nritems = btrfs_header_nritems(leaf); | |
edbd8d4e | 2536 | } |
73e48b27 Y |
2537 | |
2538 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
725c8463 CM |
2539 | |
2540 | if (progress && need_resched()) { | |
2541 | memcpy(&key, &found_key, sizeof(key)); | |
2542 | mutex_unlock(&root->fs_info->fs_mutex); | |
2543 | cond_resched(); | |
2544 | mutex_lock(&root->fs_info->fs_mutex); | |
2545 | btrfs_release_path(root, path); | |
2546 | btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
2547 | progress = 0; | |
2548 | goto next; | |
2549 | } | |
2550 | progress = 1; | |
2551 | ||
73e48b27 Y |
2552 | if (btrfs_key_type(&found_key) != BTRFS_EXTENT_ITEM_KEY || |
2553 | found_key.objectid + found_key.offset <= cur_byte) { | |
edbd8d4e | 2554 | path->slots[0]++; |
edbd8d4e CM |
2555 | goto next; |
2556 | } | |
73e48b27 | 2557 | |
edbd8d4e CM |
2558 | total_found++; |
2559 | cur_byte = found_key.objectid + found_key.offset; | |
2560 | key.objectid = cur_byte; | |
2561 | btrfs_release_path(root, path); | |
2562 | ret = relocate_one_extent(root, path, &found_key); | |
2563 | } | |
2564 | ||
2565 | btrfs_release_path(root, path); | |
2566 | ||
2567 | if (total_found > 0) { | |
2568 | trans = btrfs_start_transaction(tree_root, 1); | |
2569 | btrfs_commit_transaction(trans, tree_root); | |
2570 | ||
2571 | mutex_unlock(&root->fs_info->fs_mutex); | |
2572 | btrfs_clean_old_snapshots(tree_root); | |
2573 | mutex_lock(&root->fs_info->fs_mutex); | |
2574 | ||
2575 | trans = btrfs_start_transaction(tree_root, 1); | |
2576 | btrfs_commit_transaction(trans, tree_root); | |
2577 | goto again; | |
2578 | } | |
2579 | ||
2580 | trans = btrfs_start_transaction(root, 1); | |
2581 | key.objectid = new_size; | |
2582 | key.offset = 0; | |
2583 | key.type = 0; | |
2584 | while(1) { | |
4313b399 CM |
2585 | u64 ptr; |
2586 | ||
edbd8d4e CM |
2587 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
2588 | if (ret < 0) | |
2589 | goto out; | |
73e48b27 | 2590 | |
edbd8d4e CM |
2591 | leaf = path->nodes[0]; |
2592 | nritems = btrfs_header_nritems(leaf); | |
73e48b27 Y |
2593 | bg_next: |
2594 | if (path->slots[0] >= nritems) { | |
2595 | ret = btrfs_next_leaf(root, path); | |
2596 | if (ret < 0) | |
2597 | break; | |
2598 | if (ret == 1) { | |
2599 | ret = 0; | |
2600 | break; | |
edbd8d4e | 2601 | } |
73e48b27 | 2602 | leaf = path->nodes[0]; |
1372f8e6 CM |
2603 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
2604 | ||
2605 | /* | |
2606 | * btrfs_next_leaf doesn't cow buffers, we have to | |
2607 | * do the search again | |
2608 | */ | |
2609 | memcpy(&key, &found_key, sizeof(key)); | |
2610 | btrfs_release_path(root, path); | |
725c8463 | 2611 | goto resched_check; |
73e48b27 Y |
2612 | } |
2613 | ||
2614 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
2615 | if (btrfs_key_type(&found_key) != BTRFS_BLOCK_GROUP_ITEM_KEY) { | |
2616 | printk("shrinker found key %Lu %u %Lu\n", | |
2617 | found_key.objectid, found_key.type, | |
2618 | found_key.offset); | |
2619 | path->slots[0]++; | |
edbd8d4e CM |
2620 | goto bg_next; |
2621 | } | |
edbd8d4e CM |
2622 | ret = get_state_private(&info->block_group_cache, |
2623 | found_key.objectid, &ptr); | |
2624 | if (!ret) | |
2625 | kfree((void *)(unsigned long)ptr); | |
2626 | ||
2627 | clear_extent_bits(&info->block_group_cache, found_key.objectid, | |
2628 | found_key.objectid + found_key.offset - 1, | |
2629 | (unsigned int)-1, GFP_NOFS); | |
2630 | ||
2631 | key.objectid = found_key.objectid + 1; | |
2632 | btrfs_del_item(trans, root, path); | |
2633 | btrfs_release_path(root, path); | |
725c8463 CM |
2634 | resched_check: |
2635 | if (need_resched()) { | |
2636 | mutex_unlock(&root->fs_info->fs_mutex); | |
2637 | cond_resched(); | |
2638 | mutex_lock(&root->fs_info->fs_mutex); | |
2639 | } | |
edbd8d4e CM |
2640 | } |
2641 | clear_extent_dirty(&info->free_space_cache, new_size, (u64)-1, | |
2642 | GFP_NOFS); | |
2643 | btrfs_commit_transaction(trans, root); | |
2644 | out: | |
2645 | btrfs_free_path(path); | |
2646 | return ret; | |
2647 | } | |
2648 | ||
2649 | int btrfs_grow_extent_tree(struct btrfs_trans_handle *trans, | |
2650 | struct btrfs_root *root, u64 new_size) | |
2651 | { | |
2652 | struct btrfs_path *path; | |
2653 | u64 nr = 0; | |
2654 | u64 cur_byte; | |
2655 | u64 old_size; | |
f9ef6604 | 2656 | unsigned long rem; |
edbd8d4e CM |
2657 | struct btrfs_block_group_cache *cache; |
2658 | struct btrfs_block_group_item *item; | |
2659 | struct btrfs_fs_info *info = root->fs_info; | |
d1310b2e | 2660 | struct extent_io_tree *block_group_cache; |
edbd8d4e CM |
2661 | struct btrfs_key key; |
2662 | struct extent_buffer *leaf; | |
2663 | int ret; | |
2664 | int bit; | |
2665 | ||
2666 | old_size = btrfs_super_total_bytes(&info->super_copy); | |
2667 | block_group_cache = &info->block_group_cache; | |
2668 | ||
2669 | root = info->extent_root; | |
2670 | ||
2671 | cache = btrfs_lookup_block_group(root->fs_info, old_size - 1); | |
2672 | ||
2673 | cur_byte = cache->key.objectid + cache->key.offset; | |
2674 | if (cur_byte >= new_size) | |
2675 | goto set_size; | |
2676 | ||
2677 | key.offset = BTRFS_BLOCK_GROUP_SIZE; | |
2678 | btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY); | |
2679 | ||
2680 | path = btrfs_alloc_path(); | |
2681 | if (!path) | |
2682 | return -ENOMEM; | |
2683 | ||
2684 | while(cur_byte < new_size) { | |
2685 | key.objectid = cur_byte; | |
2686 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
2687 | sizeof(struct btrfs_block_group_item)); | |
2688 | BUG_ON(ret); | |
2689 | leaf = path->nodes[0]; | |
2690 | item = btrfs_item_ptr(leaf, path->slots[0], | |
2691 | struct btrfs_block_group_item); | |
2692 | ||
2693 | btrfs_set_disk_block_group_used(leaf, item, 0); | |
f9ef6604 CM |
2694 | div_long_long_rem(nr, 3, &rem); |
2695 | if (rem) { | |
edbd8d4e CM |
2696 | btrfs_set_disk_block_group_flags(leaf, item, |
2697 | BTRFS_BLOCK_GROUP_DATA); | |
2698 | } else { | |
2699 | btrfs_set_disk_block_group_flags(leaf, item, 0); | |
2700 | } | |
2701 | nr++; | |
2702 | ||
2703 | cache = kmalloc(sizeof(*cache), GFP_NOFS); | |
2704 | BUG_ON(!cache); | |
2705 | ||
2706 | read_extent_buffer(leaf, &cache->item, (unsigned long)item, | |
2707 | sizeof(cache->item)); | |
2708 | ||
2709 | memcpy(&cache->key, &key, sizeof(key)); | |
2710 | cache->cached = 0; | |
2711 | cache->pinned = 0; | |
2712 | cur_byte = key.objectid + key.offset; | |
2713 | btrfs_release_path(root, path); | |
2714 | ||
2715 | if (cache->item.flags & BTRFS_BLOCK_GROUP_DATA) { | |
2716 | bit = BLOCK_GROUP_DATA; | |
2717 | cache->data = BTRFS_BLOCK_GROUP_DATA; | |
2718 | } else { | |
2719 | bit = BLOCK_GROUP_METADATA; | |
2720 | cache->data = 0; | |
2721 | } | |
2722 | ||
2723 | /* use EXTENT_LOCKED to prevent merging */ | |
2724 | set_extent_bits(block_group_cache, key.objectid, | |
2725 | key.objectid + key.offset - 1, | |
2726 | bit | EXTENT_LOCKED, GFP_NOFS); | |
2727 | set_state_private(block_group_cache, key.objectid, | |
2728 | (unsigned long)cache); | |
2729 | } | |
2730 | btrfs_free_path(path); | |
2731 | set_size: | |
2732 | btrfs_set_super_total_bytes(&info->super_copy, new_size); | |
2733 | return 0; | |
2734 | } | |
2735 | ||
9078a3e1 CM |
2736 | int btrfs_read_block_groups(struct btrfs_root *root) |
2737 | { | |
2738 | struct btrfs_path *path; | |
2739 | int ret; | |
2740 | int err = 0; | |
96b5179d | 2741 | int bit; |
9078a3e1 | 2742 | struct btrfs_block_group_cache *cache; |
be744175 | 2743 | struct btrfs_fs_info *info = root->fs_info; |
d1310b2e | 2744 | struct extent_io_tree *block_group_cache; |
9078a3e1 CM |
2745 | struct btrfs_key key; |
2746 | struct btrfs_key found_key; | |
5f39d397 | 2747 | struct extent_buffer *leaf; |
96b5179d CM |
2748 | |
2749 | block_group_cache = &info->block_group_cache; | |
9078a3e1 | 2750 | |
be744175 | 2751 | root = info->extent_root; |
9078a3e1 | 2752 | key.objectid = 0; |
db94535d | 2753 | key.offset = BTRFS_BLOCK_GROUP_SIZE; |
9078a3e1 CM |
2754 | btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY); |
2755 | ||
2756 | path = btrfs_alloc_path(); | |
2757 | if (!path) | |
2758 | return -ENOMEM; | |
2759 | ||
2760 | while(1) { | |
be744175 | 2761 | ret = btrfs_search_slot(NULL, info->extent_root, |
9078a3e1 CM |
2762 | &key, path, 0, 0); |
2763 | if (ret != 0) { | |
2764 | err = ret; | |
2765 | break; | |
2766 | } | |
5f39d397 CM |
2767 | leaf = path->nodes[0]; |
2768 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
9078a3e1 CM |
2769 | cache = kmalloc(sizeof(*cache), GFP_NOFS); |
2770 | if (!cache) { | |
2771 | err = -1; | |
2772 | break; | |
2773 | } | |
3e1ad54f | 2774 | |
5f39d397 CM |
2775 | read_extent_buffer(leaf, &cache->item, |
2776 | btrfs_item_ptr_offset(leaf, path->slots[0]), | |
2777 | sizeof(cache->item)); | |
9078a3e1 | 2778 | memcpy(&cache->key, &found_key, sizeof(found_key)); |
e37c9e69 | 2779 | cache->cached = 0; |
324ae4df | 2780 | cache->pinned = 0; |
9078a3e1 CM |
2781 | key.objectid = found_key.objectid + found_key.offset; |
2782 | btrfs_release_path(root, path); | |
5f39d397 | 2783 | |
f84a8b36 CM |
2784 | if (cache->item.flags & BTRFS_BLOCK_GROUP_MIXED) { |
2785 | bit = BLOCK_GROUP_DATA | BLOCK_GROUP_METADATA; | |
2786 | cache->data = BTRFS_BLOCK_GROUP_MIXED; | |
2787 | } else if (cache->item.flags & BTRFS_BLOCK_GROUP_DATA) { | |
96b5179d | 2788 | bit = BLOCK_GROUP_DATA; |
f84a8b36 | 2789 | cache->data = BTRFS_BLOCK_GROUP_DATA; |
96b5179d CM |
2790 | } else { |
2791 | bit = BLOCK_GROUP_METADATA; | |
2792 | cache->data = 0; | |
31f3c99b | 2793 | } |
96b5179d CM |
2794 | |
2795 | /* use EXTENT_LOCKED to prevent merging */ | |
2796 | set_extent_bits(block_group_cache, found_key.objectid, | |
2797 | found_key.objectid + found_key.offset - 1, | |
2798 | bit | EXTENT_LOCKED, GFP_NOFS); | |
2799 | set_state_private(block_group_cache, found_key.objectid, | |
ae2f5411 | 2800 | (unsigned long)cache); |
96b5179d | 2801 | |
9078a3e1 | 2802 | if (key.objectid >= |
db94535d | 2803 | btrfs_super_total_bytes(&info->super_copy)) |
9078a3e1 CM |
2804 | break; |
2805 | } | |
2806 | ||
2807 | btrfs_free_path(path); | |
2808 | return 0; | |
2809 | } |