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