projects / deliverable / linux.git / blame_incremental
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (non-incremental) | history | HEAD
Btrfs: Replace the transaction work queue with kthreads
[deliverable/linux.git] / fs / btrfs / extent-tree.c
... / ...
CommitLineData
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#include <linux/sched.h>
19#include <linux/pagemap.h>
20#include <linux/writeback.h>
21#include "hash.h"
22#include "crc32c.h"
23#include "ctree.h"
24#include "disk-io.h"
25#include "print-tree.h"
26#include "transaction.h"
27#include "volumes.h"
28#include "locking.h"
29
30#define BLOCK_GROUP_DATA EXTENT_WRITEBACK
31#define BLOCK_GROUP_METADATA EXTENT_UPTODATE
32#define BLOCK_GROUP_SYSTEM EXTENT_NEW
33
34#define BLOCK_GROUP_DIRTY EXTENT_DIRTY
35
36static int finish_current_insert(struct btrfs_trans_handle *trans, struct
37 btrfs_root *extent_root);
38static int del_pending_extents(struct btrfs_trans_handle *trans, struct
39 btrfs_root *extent_root);
40static struct btrfs_block_group_cache *
41__btrfs_find_block_group(struct btrfs_root *root,
42 struct btrfs_block_group_cache *hint,
43 u64 search_start, int data, int owner);
44
45void maybe_lock_mutex(struct btrfs_root *root)
46{
47 if (root != root->fs_info->extent_root &&
48 root != root->fs_info->chunk_root &&
49 root != root->fs_info->dev_root) {
50 mutex_lock(&root->fs_info->alloc_mutex);
51 }
52}
53
54void maybe_unlock_mutex(struct btrfs_root *root)
55{
56 if (root != root->fs_info->extent_root &&
57 root != root->fs_info->chunk_root &&
58 root != root->fs_info->dev_root) {
59 mutex_unlock(&root->fs_info->alloc_mutex);
60 }
61}
62
63static int cache_block_group(struct btrfs_root *root,
64 struct btrfs_block_group_cache *block_group)
65{
66 struct btrfs_path *path;
67 int ret;
68 struct btrfs_key key;
69 struct extent_buffer *leaf;
70 struct extent_io_tree *free_space_cache;
71 int slot;
72 u64 last = 0;
73 u64 hole_size;
74 u64 first_free;
75 int found = 0;
76
77 if (!block_group)
78 return 0;
79
80 root = root->fs_info->extent_root;
81 free_space_cache = &root->fs_info->free_space_cache;
82
83 if (block_group->cached)
84 return 0;
85
86 path = btrfs_alloc_path();
87 if (!path)
88 return -ENOMEM;
89
90 path->reada = 2;
91 /*
92 * we get into deadlocks with paths held by callers of this function.
93 * since the alloc_mutex is protecting things right now, just
94 * skip the locking here
95 */
96 path->skip_locking = 1;
97 first_free = block_group->key.objectid;
98 key.objectid = block_group->key.objectid;
99 key.offset = 0;
100 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
101 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
102 if (ret < 0)
103 return ret;
104 ret = btrfs_previous_item(root, path, 0, BTRFS_EXTENT_ITEM_KEY);
105 if (ret < 0)
106 return ret;
107 if (ret == 0) {
108 leaf = path->nodes[0];
109 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
110 if (key.objectid + key.offset > first_free)
111 first_free = key.objectid + key.offset;
112 }
113 while(1) {
114 leaf = path->nodes[0];
115 slot = path->slots[0];
116 if (slot >= btrfs_header_nritems(leaf)) {
117 ret = btrfs_next_leaf(root, path);
118 if (ret < 0)
119 goto err;
120 if (ret == 0) {
121 continue;
122 } else {
123 break;
124 }
125 }
126 btrfs_item_key_to_cpu(leaf, &key, slot);
127 if (key.objectid < block_group->key.objectid) {
128 goto next;
129 }
130 if (key.objectid >= block_group->key.objectid +
131 block_group->key.offset) {
132 break;
133 }
134
135 if (btrfs_key_type(&key) == BTRFS_EXTENT_ITEM_KEY) {
136 if (!found) {
137 last = first_free;
138 found = 1;
139 }
140 if (key.objectid > last) {
141 hole_size = key.objectid - last;
142 set_extent_dirty(free_space_cache, last,
143 last + hole_size - 1,
144 GFP_NOFS);
145 }
146 last = key.objectid + key.offset;
147 }
148next:
149 path->slots[0]++;
150 }
151
152 if (!found)
153 last = first_free;
154 if (block_group->key.objectid +
155 block_group->key.offset > last) {
156 hole_size = block_group->key.objectid +
157 block_group->key.offset - last;
158 set_extent_dirty(free_space_cache, last,
159 last + hole_size - 1, GFP_NOFS);
160 }
161 block_group->cached = 1;
162err:
163 btrfs_free_path(path);
164 return 0;
165}
166
167struct btrfs_block_group_cache *btrfs_lookup_first_block_group(struct
168 btrfs_fs_info *info,
169 u64 bytenr)
170{
171 struct extent_io_tree *block_group_cache;
172 struct btrfs_block_group_cache *block_group = NULL;
173 u64 ptr;
174 u64 start;
175 u64 end;
176 int ret;
177
178 bytenr = max_t(u64, bytenr,
179 BTRFS_SUPER_INFO_OFFSET + BTRFS_SUPER_INFO_SIZE);
180 block_group_cache = &info->block_group_cache;
181 ret = find_first_extent_bit(block_group_cache,
182 bytenr, &start, &end,
183 BLOCK_GROUP_DATA | BLOCK_GROUP_METADATA |
184 BLOCK_GROUP_SYSTEM);
185 if (ret) {
186 return NULL;
187 }
188 ret = get_state_private(block_group_cache, start, &ptr);
189 if (ret)
190 return NULL;
191
192 block_group = (struct btrfs_block_group_cache *)(unsigned long)ptr;
193 return block_group;
194}
195
196struct btrfs_block_group_cache *btrfs_lookup_block_group(struct
197 btrfs_fs_info *info,
198 u64 bytenr)
199{
200 struct extent_io_tree *block_group_cache;
201 struct btrfs_block_group_cache *block_group = NULL;
202 u64 ptr;
203 u64 start;
204 u64 end;
205 int ret;
206
207 bytenr = max_t(u64, bytenr,
208 BTRFS_SUPER_INFO_OFFSET + BTRFS_SUPER_INFO_SIZE);
209 block_group_cache = &info->block_group_cache;
210 ret = find_first_extent_bit(block_group_cache,
211 bytenr, &start, &end,
212 BLOCK_GROUP_DATA | BLOCK_GROUP_METADATA |
213 BLOCK_GROUP_SYSTEM);
214 if (ret) {
215 return NULL;
216 }
217 ret = get_state_private(block_group_cache, start, &ptr);
218 if (ret)
219 return NULL;
220
221 block_group = (struct btrfs_block_group_cache *)(unsigned long)ptr;
222 if (block_group->key.objectid <= bytenr && bytenr <
223 block_group->key.objectid + block_group->key.offset)
224 return block_group;
225 return NULL;
226}
227
228static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
229{
230 return (cache->flags & bits) == bits;
231}
232
233static int noinline find_search_start(struct btrfs_root *root,
234 struct btrfs_block_group_cache **cache_ret,
235 u64 *start_ret, u64 num, int data)
236{
237 int ret;
238 struct btrfs_block_group_cache *cache = *cache_ret;
239 struct extent_io_tree *free_space_cache;
240 struct extent_state *state;
241 u64 last;
242 u64 start = 0;
243 u64 cache_miss = 0;
244 u64 total_fs_bytes;
245 u64 search_start = *start_ret;
246 int wrapped = 0;
247
248 total_fs_bytes = btrfs_super_total_bytes(&root->fs_info->super_copy);
249 free_space_cache = &root->fs_info->free_space_cache;
250
251 if (!cache)
252 goto out;
253
254again:
255 ret = cache_block_group(root, cache);
256 if (ret) {
257 goto out;
258 }
259
260 last = max(search_start, cache->key.objectid);
261 if (!block_group_bits(cache, data) || cache->ro)
262 goto new_group;
263
264 spin_lock_irq(&free_space_cache->lock);
265 state = find_first_extent_bit_state(free_space_cache, last, EXTENT_DIRTY);
266 while(1) {
267 if (!state) {
268 if (!cache_miss)
269 cache_miss = last;
270 spin_unlock_irq(&free_space_cache->lock);
271 goto new_group;
272 }
273
274 start = max(last, state->start);
275 last = state->end + 1;
276 if (last - start < num) {
277 do {
278 state = extent_state_next(state);
279 } while(state && !(state->state & EXTENT_DIRTY));
280 continue;
281 }
282 spin_unlock_irq(&free_space_cache->lock);
283 if (cache->ro) {
284 goto new_group;
285 }
286 if (start + num > cache->key.objectid + cache->key.offset)
287 goto new_group;
288 if (!block_group_bits(cache, data)) {
289 printk("block group bits don't match %Lu %d\n", cache->flags, data);
290 }
291 *start_ret = start;
292 return 0;
293 }
294out:
295 cache = btrfs_lookup_block_group(root->fs_info, search_start);
296 if (!cache) {
297 printk("Unable to find block group for %Lu\n", search_start);
298 WARN_ON(1);
299 }
300 return -ENOSPC;
301
302new_group:
303 last = cache->key.objectid + cache->key.offset;
304wrapped:
305 cache = btrfs_lookup_first_block_group(root->fs_info, last);
306 if (!cache || cache->key.objectid >= total_fs_bytes) {
307no_cache:
308 if (!wrapped) {
309 wrapped = 1;
310 last = search_start;
311 goto wrapped;
312 }
313 goto out;
314 }
315 if (cache_miss && !cache->cached) {
316 cache_block_group(root, cache);
317 last = cache_miss;
318 cache = btrfs_lookup_first_block_group(root->fs_info, last);
319 }
320 cache_miss = 0;
321 cache = __btrfs_find_block_group(root, cache, last, data, 0);
322 if (!cache)
323 goto no_cache;
324 *cache_ret = cache;
325 goto again;
326}
327
328static u64 div_factor(u64 num, int factor)
329{
330 if (factor == 10)
331 return num;
332 num *= factor;
333 do_div(num, 10);
334 return num;
335}
336
337static int block_group_state_bits(u64 flags)
338{
339 int bits = 0;
340 if (flags & BTRFS_BLOCK_GROUP_DATA)
341 bits |= BLOCK_GROUP_DATA;
342 if (flags & BTRFS_BLOCK_GROUP_METADATA)
343 bits |= BLOCK_GROUP_METADATA;
344 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
345 bits |= BLOCK_GROUP_SYSTEM;
346 return bits;
347}
348
349static struct btrfs_block_group_cache *
350__btrfs_find_block_group(struct btrfs_root *root,
351 struct btrfs_block_group_cache *hint,
352 u64 search_start, int data, int owner)
353{
354 struct btrfs_block_group_cache *cache;
355 struct extent_io_tree *block_group_cache;
356 struct btrfs_block_group_cache *found_group = NULL;
357 struct btrfs_fs_info *info = root->fs_info;
358 u64 used;
359 u64 last = 0;
360 u64 start;
361 u64 end;
362 u64 free_check;
363 u64 ptr;
364 int bit;
365 int ret;
366 int full_search = 0;
367 int factor = 10;
368 int wrapped = 0;
369
370 block_group_cache = &info->block_group_cache;
371
372 if (data & BTRFS_BLOCK_GROUP_METADATA)
373 factor = 9;
374
375 bit = block_group_state_bits(data);
376
377 if (search_start) {
378 struct btrfs_block_group_cache *shint;
379 shint = btrfs_lookup_first_block_group(info, search_start);
380 if (shint && block_group_bits(shint, data) && !shint->ro) {
381 used = btrfs_block_group_used(&shint->item);
382 if (used + shint->pinned <
383 div_factor(shint->key.offset, factor)) {
384 return shint;
385 }
386 }
387 }
388 if (hint && !hint->ro && block_group_bits(hint, data)) {
389 used = btrfs_block_group_used(&hint->item);
390 if (used + hint->pinned <
391 div_factor(hint->key.offset, factor)) {
392 return hint;
393 }
394 last = hint->key.objectid + hint->key.offset;
395 } else {
396 if (hint)
397 last = max(hint->key.objectid, search_start);
398 else
399 last = search_start;
400 }
401again:
402 while(1) {
403 ret = find_first_extent_bit(block_group_cache, last,
404 &start, &end, bit);
405 if (ret)
406 break;
407
408 ret = get_state_private(block_group_cache, start, &ptr);
409 if (ret) {
410 last = end + 1;
411 continue;
412 }
413
414 cache = (struct btrfs_block_group_cache *)(unsigned long)ptr;
415 last = cache->key.objectid + cache->key.offset;
416 used = btrfs_block_group_used(&cache->item);
417
418 if (!cache->ro && block_group_bits(cache, data)) {
419 free_check = div_factor(cache->key.offset, factor);
420 if (used + cache->pinned < free_check) {
421 found_group = cache;
422 goto found;
423 }
424 }
425 cond_resched();
426 }
427 if (!wrapped) {
428 last = search_start;
429 wrapped = 1;
430 goto again;
431 }
432 if (!full_search && factor < 10) {
433 last = search_start;
434 full_search = 1;
435 factor = 10;
436 goto again;
437 }
438found:
439 return found_group;
440}
441
442struct btrfs_block_group_cache *btrfs_find_block_group(struct btrfs_root *root,
443 struct btrfs_block_group_cache
444 *hint, u64 search_start,
445 int data, int owner)
446{
447
448 struct btrfs_block_group_cache *ret;
449 mutex_lock(&root->fs_info->alloc_mutex);
450 ret = __btrfs_find_block_group(root, hint, search_start, data, owner);
451 mutex_unlock(&root->fs_info->alloc_mutex);
452 return ret;
453}
454static u64 hash_extent_ref(u64 root_objectid, u64 ref_generation,
455 u64 owner, u64 owner_offset)
456{
457 u32 high_crc = ~(u32)0;
458 u32 low_crc = ~(u32)0;
459 __le64 lenum;
460 lenum = cpu_to_le64(root_objectid);
461 high_crc = btrfs_crc32c(high_crc, &lenum, sizeof(lenum));
462 lenum = cpu_to_le64(ref_generation);
463 low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum));
464 if (owner >= BTRFS_FIRST_FREE_OBJECTID) {
465 lenum = cpu_to_le64(owner);
466 low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum));
467 lenum = cpu_to_le64(owner_offset);
468 low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum));
469 }
470 return ((u64)high_crc << 32) | (u64)low_crc;
471}
472
473static int match_extent_ref(struct extent_buffer *leaf,
474 struct btrfs_extent_ref *disk_ref,
475 struct btrfs_extent_ref *cpu_ref)
476{
477 int ret;
478 int len;
479
480 if (cpu_ref->objectid)
481 len = sizeof(*cpu_ref);
482 else
483 len = 2 * sizeof(u64);
484 ret = memcmp_extent_buffer(leaf, cpu_ref, (unsigned long)disk_ref,
485 len);
486 return ret == 0;
487}
488
489static int noinline lookup_extent_backref(struct btrfs_trans_handle *trans,
490 struct btrfs_root *root,
491 struct btrfs_path *path, u64 bytenr,
492 u64 root_objectid,
493 u64 ref_generation, u64 owner,
494 u64 owner_offset, int del)
495{
496 u64 hash;
497 struct btrfs_key key;
498 struct btrfs_key found_key;
499 struct btrfs_extent_ref ref;
500 struct extent_buffer *leaf;
501 struct btrfs_extent_ref *disk_ref;
502 int ret;
503 int ret2;
504
505 btrfs_set_stack_ref_root(&ref, root_objectid);
506 btrfs_set_stack_ref_generation(&ref, ref_generation);
507 btrfs_set_stack_ref_objectid(&ref, owner);
508 btrfs_set_stack_ref_offset(&ref, owner_offset);
509
510 hash = hash_extent_ref(root_objectid, ref_generation, owner,
511 owner_offset);
512 key.offset = hash;
513 key.objectid = bytenr;
514 key.type = BTRFS_EXTENT_REF_KEY;
515
516 while (1) {
517 ret = btrfs_search_slot(trans, root, &key, path,
518 del ? -1 : 0, del);
519 if (ret < 0)
520 goto out;
521 leaf = path->nodes[0];
522 if (ret != 0) {
523 u32 nritems = btrfs_header_nritems(leaf);
524 if (path->slots[0] >= nritems) {
525 ret2 = btrfs_next_leaf(root, path);
526 if (ret2)
527 goto out;
528 leaf = path->nodes[0];
529 }
530 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
531 if (found_key.objectid != bytenr ||
532 found_key.type != BTRFS_EXTENT_REF_KEY)
533 goto out;
534 key.offset = found_key.offset;
535 if (del) {
536 btrfs_release_path(root, path);
537 continue;
538 }
539 }
540 disk_ref = btrfs_item_ptr(path->nodes[0],
541 path->slots[0],
542 struct btrfs_extent_ref);
543 if (match_extent_ref(path->nodes[0], disk_ref, &ref)) {
544 ret = 0;
545 goto out;
546 }
547 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
548 key.offset = found_key.offset + 1;
549 btrfs_release_path(root, path);
550 }
551out:
552 return ret;
553}
554
555/*
556 * Back reference rules. Back refs have three main goals:
557 *
558 * 1) differentiate between all holders of references to an extent so that
559 * when a reference is dropped we can make sure it was a valid reference
560 * before freeing the extent.
561 *
562 * 2) Provide enough information to quickly find the holders of an extent
563 * if we notice a given block is corrupted or bad.
564 *
565 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
566 * maintenance. This is actually the same as #2, but with a slightly
567 * different use case.
568 *
569 * File extents can be referenced by:
570 *
571 * - multiple snapshots, subvolumes, or different generations in one subvol
572 * - different files inside a single subvolume (in theory, not implemented yet)
573 * - different offsets inside a file (bookend extents in file.c)
574 *
575 * The extent ref structure has fields for:
576 *
577 * - Objectid of the subvolume root
578 * - Generation number of the tree holding the reference
579 * - objectid of the file holding the reference
580 * - offset in the file corresponding to the key holding the reference
581 *
582 * When a file extent is allocated the fields are filled in:
583 * (root_key.objectid, trans->transid, inode objectid, offset in file)
584 *
585 * When a leaf is cow'd new references are added for every file extent found
586 * in the leaf. It looks the same as the create case, but trans->transid
587 * will be different when the block is cow'd.
588 *
589 * (root_key.objectid, trans->transid, inode objectid, offset in file)
590 *
591 * When a file extent is removed either during snapshot deletion or file
592 * truncation, the corresponding back reference is found
593 * by searching for:
594 *
595 * (btrfs_header_owner(leaf), btrfs_header_generation(leaf),
596 * inode objectid, offset in file)
597 *
598 * Btree extents can be referenced by:
599 *
600 * - Different subvolumes
601 * - Different generations of the same subvolume
602 *
603 * Storing sufficient information for a full reverse mapping of a btree
604 * block would require storing the lowest key of the block in the backref,
605 * and it would require updating that lowest key either before write out or
606 * every time it changed. Instead, the objectid of the lowest key is stored
607 * along with the level of the tree block. This provides a hint
608 * about where in the btree the block can be found. Searches through the
609 * btree only need to look for a pointer to that block, so they stop one
610 * level higher than the level recorded in the backref.
611 *
612 * Some btrees do not do reference counting on their extents. These
613 * include the extent tree and the tree of tree roots. Backrefs for these
614 * trees always have a generation of zero.
615 *
616 * When a tree block is created, back references are inserted:
617 *
618 * (root->root_key.objectid, trans->transid or zero, level, lowest_key_objectid)
619 *
620 * When a tree block is cow'd in a reference counted root,
621 * new back references are added for all the blocks it points to.
622 * These are of the form (trans->transid will have increased since creation):
623 *
624 * (root->root_key.objectid, trans->transid, level, lowest_key_objectid)
625 *
626 * Because the lowest_key_objectid and the level are just hints
627 * they are not used when backrefs are deleted. When a backref is deleted:
628 *
629 * if backref was for a tree root:
630 * root_objectid = root->root_key.objectid
631 * else
632 * root_objectid = btrfs_header_owner(parent)
633 *
634 * (root_objectid, btrfs_header_generation(parent) or zero, 0, 0)
635 *
636 * Back Reference Key hashing:
637 *
638 * Back references have four fields, each 64 bits long. Unfortunately,
639 * This is hashed into a single 64 bit number and placed into the key offset.
640 * The key objectid corresponds to the first byte in the extent, and the
641 * key type is set to BTRFS_EXTENT_REF_KEY
642 */
643int btrfs_insert_extent_backref(struct btrfs_trans_handle *trans,
644 struct btrfs_root *root,
645 struct btrfs_path *path, u64 bytenr,
646 u64 root_objectid, u64 ref_generation,
647 u64 owner, u64 owner_offset)
648{
649 u64 hash;
650 struct btrfs_key key;
651 struct btrfs_extent_ref ref;
652 struct btrfs_extent_ref *disk_ref;
653 int ret;
654
655 btrfs_set_stack_ref_root(&ref, root_objectid);
656 btrfs_set_stack_ref_generation(&ref, ref_generation);
657 btrfs_set_stack_ref_objectid(&ref, owner);
658 btrfs_set_stack_ref_offset(&ref, owner_offset);
659
660 hash = hash_extent_ref(root_objectid, ref_generation, owner,
661 owner_offset);
662 key.offset = hash;
663 key.objectid = bytenr;
664 key.type = BTRFS_EXTENT_REF_KEY;
665
666 ret = btrfs_insert_empty_item(trans, root, path, &key, sizeof(ref));
667 while (ret == -EEXIST) {
668 disk_ref = btrfs_item_ptr(path->nodes[0], path->slots[0],
669 struct btrfs_extent_ref);
670 if (match_extent_ref(path->nodes[0], disk_ref, &ref))
671 goto out;
672 key.offset++;
673 btrfs_release_path(root, path);
674 ret = btrfs_insert_empty_item(trans, root, path, &key,
675 sizeof(ref));
676 }
677 if (ret)
678 goto out;
679 disk_ref = btrfs_item_ptr(path->nodes[0], path->slots[0],
680 struct btrfs_extent_ref);
681 write_extent_buffer(path->nodes[0], &ref, (unsigned long)disk_ref,
682 sizeof(ref));
683 btrfs_mark_buffer_dirty(path->nodes[0]);
684out:
685 btrfs_release_path(root, path);
686 return ret;
687}
688
689static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
690 struct btrfs_root *root,
691 u64 bytenr, u64 num_bytes,
692 u64 root_objectid, u64 ref_generation,
693 u64 owner, u64 owner_offset)
694{
695 struct btrfs_path *path;
696 int ret;
697 struct btrfs_key key;
698 struct extent_buffer *l;
699 struct btrfs_extent_item *item;
700 u32 refs;
701
702 WARN_ON(num_bytes < root->sectorsize);
703 path = btrfs_alloc_path();
704 if (!path)
705 return -ENOMEM;
706
707 path->reada = 1;
708 key.objectid = bytenr;
709 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
710 key.offset = num_bytes;
711 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path,
712 0, 1);
713 if (ret < 0)
714 return ret;
715 if (ret != 0) {
716 BUG();
717 }
718 BUG_ON(ret != 0);
719 l = path->nodes[0];
720 item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
721 refs = btrfs_extent_refs(l, item);
722 btrfs_set_extent_refs(l, item, refs + 1);
723 btrfs_mark_buffer_dirty(path->nodes[0]);
724
725 btrfs_release_path(root->fs_info->extent_root, path);
726
727 path->reada = 1;
728 ret = btrfs_insert_extent_backref(trans, root->fs_info->extent_root,
729 path, bytenr, root_objectid,
730 ref_generation, owner, owner_offset);
731 BUG_ON(ret);
732 finish_current_insert(trans, root->fs_info->extent_root);
733 del_pending_extents(trans, root->fs_info->extent_root);
734
735 btrfs_free_path(path);
736 return 0;
737}
738
739int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
740 struct btrfs_root *root,
741 u64 bytenr, u64 num_bytes,
742 u64 root_objectid, u64 ref_generation,
743 u64 owner, u64 owner_offset)
744{
745 int ret;
746
747 mutex_lock(&root->fs_info->alloc_mutex);
748 ret = __btrfs_inc_extent_ref(trans, root, bytenr, num_bytes,
749 root_objectid, ref_generation,
750 owner, owner_offset);
751 mutex_unlock(&root->fs_info->alloc_mutex);
752 return ret;
753}
754
755int btrfs_extent_post_op(struct btrfs_trans_handle *trans,
756 struct btrfs_root *root)
757{
758 finish_current_insert(trans, root->fs_info->extent_root);
759 del_pending_extents(trans, root->fs_info->extent_root);
760 return 0;
761}
762
763static int lookup_extent_ref(struct btrfs_trans_handle *trans,
764 struct btrfs_root *root, u64 bytenr,
765 u64 num_bytes, u32 *refs)
766{
767 struct btrfs_path *path;
768 int ret;
769 struct btrfs_key key;
770 struct extent_buffer *l;
771 struct btrfs_extent_item *item;
772
773 WARN_ON(num_bytes < root->sectorsize);
774 path = btrfs_alloc_path();
775 path->reada = 1;
776 key.objectid = bytenr;
777 key.offset = num_bytes;
778 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
779 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path,
780 0, 0);
781 if (ret < 0)
782 goto out;
783 if (ret != 0) {
784 btrfs_print_leaf(root, path->nodes[0]);
785 printk("failed to find block number %Lu\n", bytenr);
786 BUG();
787 }
788 l = path->nodes[0];
789 item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
790 *refs = btrfs_extent_refs(l, item);
791out:
792 btrfs_free_path(path);
793 return 0;
794}
795
796u32 btrfs_count_snapshots_in_path(struct btrfs_root *root,
797 struct btrfs_path *count_path,
798 u64 expected_owner,
799 u64 first_extent)
800{
801 struct btrfs_root *extent_root = root->fs_info->extent_root;
802 struct btrfs_path *path;
803 u64 bytenr;
804 u64 found_objectid;
805 u64 found_owner;
806 u64 root_objectid = root->root_key.objectid;
807 u32 total_count = 0;
808 u32 extent_refs;
809 u32 cur_count;
810 u32 nritems;
811 int ret;
812 struct btrfs_key key;
813 struct btrfs_key found_key;
814 struct extent_buffer *l;
815 struct btrfs_extent_item *item;
816 struct btrfs_extent_ref *ref_item;
817 int level = -1;
818
819 /* FIXME, needs locking */
820 BUG();
821
822 mutex_lock(&root->fs_info->alloc_mutex);
823 path = btrfs_alloc_path();
824again:
825 if (level == -1)
826 bytenr = first_extent;
827 else
828 bytenr = count_path->nodes[level]->start;
829
830 cur_count = 0;
831 key.objectid = bytenr;
832 key.offset = 0;
833
834 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
835 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
836 if (ret < 0)
837 goto out;
838 BUG_ON(ret == 0);
839
840 l = path->nodes[0];
841 btrfs_item_key_to_cpu(l, &found_key, path->slots[0]);
842
843 if (found_key.objectid != bytenr ||
844 found_key.type != BTRFS_EXTENT_ITEM_KEY) {
845 goto out;
846 }
847
848 item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
849 extent_refs = btrfs_extent_refs(l, item);
850 while (1) {
851 l = path->nodes[0];
852 nritems = btrfs_header_nritems(l);
853 if (path->slots[0] >= nritems) {
854 ret = btrfs_next_leaf(extent_root, path);
855 if (ret == 0)
856 continue;
857 break;
858 }
859 btrfs_item_key_to_cpu(l, &found_key, path->slots[0]);
860 if (found_key.objectid != bytenr)
861 break;
862
863 if (found_key.type != BTRFS_EXTENT_REF_KEY) {
864 path->slots[0]++;
865 continue;
866 }
867
868 cur_count++;
869 ref_item = btrfs_item_ptr(l, path->slots[0],
870 struct btrfs_extent_ref);
871 found_objectid = btrfs_ref_root(l, ref_item);
872
873 if (found_objectid != root_objectid) {
874 total_count = 2;
875 goto out;
876 }
877 if (level == -1) {
878 found_owner = btrfs_ref_objectid(l, ref_item);
879 if (found_owner != expected_owner) {
880 total_count = 2;
881 goto out;
882 }
883 /*
884 * nasty. we don't count a reference held by
885 * the running transaction. This allows nodatacow
886 * to avoid cow most of the time
887 */
888 if (found_owner >= BTRFS_FIRST_FREE_OBJECTID &&
889 btrfs_ref_generation(l, ref_item) ==
890 root->fs_info->generation) {
891 extent_refs--;
892 }
893 }
894 total_count = 1;
895 path->slots[0]++;
896 }
897 /*
898 * if there is more than one reference against a data extent,
899 * we have to assume the other ref is another snapshot
900 */
901 if (level == -1 && extent_refs > 1) {
902 total_count = 2;
903 goto out;
904 }
905 if (cur_count == 0) {
906 total_count = 0;
907 goto out;
908 }
909 if (level >= 0 && root->node == count_path->nodes[level])
910 goto out;
911 level++;
912 btrfs_release_path(root, path);
913 goto again;
914
915out:
916 btrfs_free_path(path);
917 mutex_unlock(&root->fs_info->alloc_mutex);
918 return total_count;
919}
920
921int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
922 struct extent_buffer *buf)
923{
924 u64 bytenr;
925 u32 nritems;
926 struct btrfs_key key;
927 struct btrfs_file_extent_item *fi;
928 int i;
929 int level;
930 int ret;
931 int faili;
932
933 if (!root->ref_cows)
934 return 0;
935
936 mutex_lock(&root->fs_info->alloc_mutex);
937 level = btrfs_header_level(buf);
938 nritems = btrfs_header_nritems(buf);
939 for (i = 0; i < nritems; i++) {
940 if (level == 0) {
941 u64 disk_bytenr;
942 btrfs_item_key_to_cpu(buf, &key, i);
943 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
944 continue;
945 fi = btrfs_item_ptr(buf, i,
946 struct btrfs_file_extent_item);
947 if (btrfs_file_extent_type(buf, fi) ==
948 BTRFS_FILE_EXTENT_INLINE)
949 continue;
950 disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
951 if (disk_bytenr == 0)
952 continue;
953 ret = __btrfs_inc_extent_ref(trans, root, disk_bytenr,
954 btrfs_file_extent_disk_num_bytes(buf, fi),
955 root->root_key.objectid, trans->transid,
956 key.objectid, key.offset);
957 if (ret) {
958 faili = i;
959 goto fail;
960 }
961 } else {
962 bytenr = btrfs_node_blockptr(buf, i);
963 btrfs_node_key_to_cpu(buf, &key, i);
964 ret = __btrfs_inc_extent_ref(trans, root, bytenr,
965 btrfs_level_size(root, level - 1),
966 root->root_key.objectid,
967 trans->transid,
968 level - 1, key.objectid);
969 if (ret) {
970 faili = i;
971 goto fail;
972 }
973 }
974 }
975 mutex_unlock(&root->fs_info->alloc_mutex);
976 return 0;
977fail:
978 WARN_ON(1);
979#if 0
980 for (i =0; i < faili; i++) {
981 if (level == 0) {
982 u64 disk_bytenr;
983 btrfs_item_key_to_cpu(buf, &key, i);
984 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
985 continue;
986 fi = btrfs_item_ptr(buf, i,
987 struct btrfs_file_extent_item);
988 if (btrfs_file_extent_type(buf, fi) ==
989 BTRFS_FILE_EXTENT_INLINE)
990 continue;
991 disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
992 if (disk_bytenr == 0)
993 continue;
994 err = btrfs_free_extent(trans, root, disk_bytenr,
995 btrfs_file_extent_disk_num_bytes(buf,
996 fi), 0);
997 BUG_ON(err);
998 } else {
999 bytenr = btrfs_node_blockptr(buf, i);
1000 err = btrfs_free_extent(trans, root, bytenr,
1001 btrfs_level_size(root, level - 1), 0);
1002 BUG_ON(err);
1003 }
1004 }
1005#endif
1006 mutex_unlock(&root->fs_info->alloc_mutex);
1007 return ret;
1008}
1009
1010static int write_one_cache_group(struct btrfs_trans_handle *trans,
1011 struct btrfs_root *root,
1012 struct btrfs_path *path,
1013 struct btrfs_block_group_cache *cache)
1014{
1015 int ret;
1016 int pending_ret;
1017 struct btrfs_root *extent_root = root->fs_info->extent_root;
1018 unsigned long bi;
1019 struct extent_buffer *leaf;
1020
1021 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
1022 if (ret < 0)
1023 goto fail;
1024 BUG_ON(ret);
1025
1026 leaf = path->nodes[0];
1027 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
1028 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
1029 btrfs_mark_buffer_dirty(leaf);
1030 btrfs_release_path(extent_root, path);
1031fail:
1032 finish_current_insert(trans, extent_root);
1033 pending_ret = del_pending_extents(trans, extent_root);
1034 if (ret)
1035 return ret;
1036 if (pending_ret)
1037 return pending_ret;
1038 return 0;
1039
1040}
1041
1042int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
1043 struct btrfs_root *root)
1044{
1045 struct extent_io_tree *block_group_cache;
1046 struct btrfs_block_group_cache *cache;
1047 int ret;
1048 int err = 0;
1049 int werr = 0;
1050 struct btrfs_path *path;
1051 u64 last = 0;
1052 u64 start;
1053 u64 end;
1054 u64 ptr;
1055
1056 block_group_cache = &root->fs_info->block_group_cache;
1057 path = btrfs_alloc_path();
1058 if (!path)
1059 return -ENOMEM;
1060
1061 mutex_lock(&root->fs_info->alloc_mutex);
1062 while(1) {
1063 ret = find_first_extent_bit(block_group_cache, last,
1064 &start, &end, BLOCK_GROUP_DIRTY);
1065 if (ret)
1066 break;
1067
1068 last = end + 1;
1069 ret = get_state_private(block_group_cache, start, &ptr);
1070 if (ret)
1071 break;
1072 cache = (struct btrfs_block_group_cache *)(unsigned long)ptr;
1073 err = write_one_cache_group(trans, root,
1074 path, cache);
1075 /*
1076 * if we fail to write the cache group, we want
1077 * to keep it marked dirty in hopes that a later
1078 * write will work
1079 */
1080 if (err) {
1081 werr = err;
1082 continue;
1083 }
1084 clear_extent_bits(block_group_cache, start, end,
1085 BLOCK_GROUP_DIRTY, GFP_NOFS);
1086 }
1087 btrfs_free_path(path);
1088 mutex_unlock(&root->fs_info->alloc_mutex);
1089 return werr;
1090}
1091
1092static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
1093 u64 flags)
1094{
1095 struct list_head *head = &info->space_info;
1096 struct list_head *cur;
1097 struct btrfs_space_info *found;
1098 list_for_each(cur, head) {
1099 found = list_entry(cur, struct btrfs_space_info, list);
1100 if (found->flags == flags)
1101 return found;
1102 }
1103 return NULL;
1104
1105}
1106
1107static int update_space_info(struct btrfs_fs_info *info, u64 flags,
1108 u64 total_bytes, u64 bytes_used,
1109 struct btrfs_space_info **space_info)
1110{
1111 struct btrfs_space_info *found;
1112
1113 found = __find_space_info(info, flags);
1114 if (found) {
1115 found->total_bytes += total_bytes;
1116 found->bytes_used += bytes_used;
1117 found->full = 0;
1118 WARN_ON(found->total_bytes < found->bytes_used);
1119 *space_info = found;
1120 return 0;
1121 }
1122 found = kmalloc(sizeof(*found), GFP_NOFS);
1123 if (!found)
1124 return -ENOMEM;
1125
1126 list_add(&found->list, &info->space_info);
1127 found->flags = flags;
1128 found->total_bytes = total_bytes;
1129 found->bytes_used = bytes_used;
1130 found->bytes_pinned = 0;
1131 found->full = 0;
1132 found->force_alloc = 0;
1133 *space_info = found;
1134 return 0;
1135}
1136
1137static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
1138{
1139 u64 extra_flags = flags & (BTRFS_BLOCK_GROUP_RAID0 |
1140 BTRFS_BLOCK_GROUP_RAID1 |
1141 BTRFS_BLOCK_GROUP_RAID10 |
1142 BTRFS_BLOCK_GROUP_DUP);
1143 if (extra_flags) {
1144 if (flags & BTRFS_BLOCK_GROUP_DATA)
1145 fs_info->avail_data_alloc_bits |= extra_flags;
1146 if (flags & BTRFS_BLOCK_GROUP_METADATA)
1147 fs_info->avail_metadata_alloc_bits |= extra_flags;
1148 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
1149 fs_info->avail_system_alloc_bits |= extra_flags;
1150 }
1151}
1152
1153static u64 reduce_alloc_profile(struct btrfs_root *root, u64 flags)
1154{
1155 u64 num_devices = root->fs_info->fs_devices->num_devices;
1156
1157 if (num_devices == 1)
1158 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0);
1159 if (num_devices < 4)
1160 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
1161
1162 if ((flags & BTRFS_BLOCK_GROUP_DUP) &&
1163 (flags & (BTRFS_BLOCK_GROUP_RAID1 |
1164 BTRFS_BLOCK_GROUP_RAID10))) {
1165 flags &= ~BTRFS_BLOCK_GROUP_DUP;
1166 }
1167
1168 if ((flags & BTRFS_BLOCK_GROUP_RAID1) &&
1169 (flags & BTRFS_BLOCK_GROUP_RAID10)) {
1170 flags &= ~BTRFS_BLOCK_GROUP_RAID1;
1171 }
1172
1173 if ((flags & BTRFS_BLOCK_GROUP_RAID0) &&
1174 ((flags & BTRFS_BLOCK_GROUP_RAID1) |
1175 (flags & BTRFS_BLOCK_GROUP_RAID10) |
1176 (flags & BTRFS_BLOCK_GROUP_DUP)))
1177 flags &= ~BTRFS_BLOCK_GROUP_RAID0;
1178 return flags;
1179}
1180
1181static int do_chunk_alloc(struct btrfs_trans_handle *trans,
1182 struct btrfs_root *extent_root, u64 alloc_bytes,
1183 u64 flags, int force)
1184{
1185 struct btrfs_space_info *space_info;
1186 u64 thresh;
1187 u64 start;
1188 u64 num_bytes;
1189 int ret;
1190
1191 flags = reduce_alloc_profile(extent_root, flags);
1192
1193 space_info = __find_space_info(extent_root->fs_info, flags);
1194 if (!space_info) {
1195 ret = update_space_info(extent_root->fs_info, flags,
1196 0, 0, &space_info);
1197 BUG_ON(ret);
1198 }
1199 BUG_ON(!space_info);
1200
1201 if (space_info->force_alloc) {
1202 force = 1;
1203 space_info->force_alloc = 0;
1204 }
1205 if (space_info->full)
1206 goto out;
1207
1208 thresh = div_factor(space_info->total_bytes, 6);
1209 if (!force &&
1210 (space_info->bytes_used + space_info->bytes_pinned + alloc_bytes) <
1211 thresh)
1212 goto out;
1213
1214 mutex_lock(&extent_root->fs_info->chunk_mutex);
1215 ret = btrfs_alloc_chunk(trans, extent_root, &start, &num_bytes, flags);
1216 if (ret == -ENOSPC) {
1217printk("space info full %Lu\n", flags);
1218 space_info->full = 1;
1219 goto out_unlock;
1220 }
1221 BUG_ON(ret);
1222
1223 ret = btrfs_make_block_group(trans, extent_root, 0, flags,
1224 BTRFS_FIRST_CHUNK_TREE_OBJECTID, start, num_bytes);
1225 BUG_ON(ret);
1226out_unlock:
1227 mutex_unlock(&extent_root->fs_info->chunk_mutex);
1228out:
1229 return 0;
1230}
1231
1232static int update_block_group(struct btrfs_trans_handle *trans,
1233 struct btrfs_root *root,
1234 u64 bytenr, u64 num_bytes, int alloc,
1235 int mark_free)
1236{
1237 struct btrfs_block_group_cache *cache;
1238 struct btrfs_fs_info *info = root->fs_info;
1239 u64 total = num_bytes;
1240 u64 old_val;
1241 u64 byte_in_group;
1242 u64 start;
1243 u64 end;
1244
1245 while(total) {
1246 cache = btrfs_lookup_block_group(info, bytenr);
1247 if (!cache) {
1248 return -1;
1249 }
1250 byte_in_group = bytenr - cache->key.objectid;
1251 WARN_ON(byte_in_group > cache->key.offset);
1252 start = cache->key.objectid;
1253 end = start + cache->key.offset - 1;
1254 set_extent_bits(&info->block_group_cache, start, end,
1255 BLOCK_GROUP_DIRTY, GFP_NOFS);
1256
1257 old_val = btrfs_block_group_used(&cache->item);
1258 num_bytes = min(total, cache->key.offset - byte_in_group);
1259 if (alloc) {
1260 old_val += num_bytes;
1261 cache->space_info->bytes_used += num_bytes;
1262 } else {
1263 old_val -= num_bytes;
1264 cache->space_info->bytes_used -= num_bytes;
1265 if (mark_free) {
1266 set_extent_dirty(&info->free_space_cache,
1267 bytenr, bytenr + num_bytes - 1,
1268 GFP_NOFS);
1269 }
1270 }
1271 btrfs_set_block_group_used(&cache->item, old_val);
1272 total -= num_bytes;
1273 bytenr += num_bytes;
1274 }
1275 return 0;
1276}
1277
1278static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
1279{
1280 u64 start;
1281 u64 end;
1282 int ret;
1283 ret = find_first_extent_bit(&root->fs_info->block_group_cache,
1284 search_start, &start, &end,
1285 BLOCK_GROUP_DATA | BLOCK_GROUP_METADATA |
1286 BLOCK_GROUP_SYSTEM);
1287 if (ret)
1288 return 0;
1289 return start;
1290}
1291
1292
1293static int update_pinned_extents(struct btrfs_root *root,
1294 u64 bytenr, u64 num, int pin)
1295{
1296 u64 len;
1297 struct btrfs_block_group_cache *cache;
1298 struct btrfs_fs_info *fs_info = root->fs_info;
1299
1300 if (pin) {
1301 set_extent_dirty(&fs_info->pinned_extents,
1302 bytenr, bytenr + num - 1, GFP_NOFS);
1303 } else {
1304 clear_extent_dirty(&fs_info->pinned_extents,
1305 bytenr, bytenr + num - 1, GFP_NOFS);
1306 }
1307 while (num > 0) {
1308 cache = btrfs_lookup_block_group(fs_info, bytenr);
1309 if (!cache) {
1310 u64 first = first_logical_byte(root, bytenr);
1311 WARN_ON(first < bytenr);
1312 len = min(first - bytenr, num);
1313 } else {
1314 len = min(num, cache->key.offset -
1315 (bytenr - cache->key.objectid));
1316 }
1317 if (pin) {
1318 if (cache) {
1319 cache->pinned += len;
1320 cache->space_info->bytes_pinned += len;
1321 }
1322 fs_info->total_pinned += len;
1323 } else {
1324 if (cache) {
1325 cache->pinned -= len;
1326 cache->space_info->bytes_pinned -= len;
1327 }
1328 fs_info->total_pinned -= len;
1329 }
1330 bytenr += len;
1331 num -= len;
1332 }
1333 return 0;
1334}
1335
1336int btrfs_copy_pinned(struct btrfs_root *root, struct extent_io_tree *copy)
1337{
1338 u64 last = 0;
1339 u64 start;
1340 u64 end;
1341 struct extent_io_tree *pinned_extents = &root->fs_info->pinned_extents;
1342 int ret;
1343
1344 while(1) {
1345 ret = find_first_extent_bit(pinned_extents, last,
1346 &start, &end, EXTENT_DIRTY);
1347 if (ret)
1348 break;
1349 set_extent_dirty(copy, start, end, GFP_NOFS);
1350 last = end + 1;
1351 }
1352 return 0;
1353}
1354
1355int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
1356 struct btrfs_root *root,
1357 struct extent_io_tree *unpin)
1358{
1359 u64 start;
1360 u64 end;
1361 int ret;
1362 struct extent_io_tree *free_space_cache;
1363 free_space_cache = &root->fs_info->free_space_cache;
1364
1365 mutex_lock(&root->fs_info->alloc_mutex);
1366 while(1) {
1367 ret = find_first_extent_bit(unpin, 0, &start, &end,
1368 EXTENT_DIRTY);
1369 if (ret)
1370 break;
1371 update_pinned_extents(root, start, end + 1 - start, 0);
1372 clear_extent_dirty(unpin, start, end, GFP_NOFS);
1373 set_extent_dirty(free_space_cache, start, end, GFP_NOFS);
1374 }
1375 mutex_unlock(&root->fs_info->alloc_mutex);
1376 return 0;
1377}
1378
1379static int finish_current_insert(struct btrfs_trans_handle *trans,
1380 struct btrfs_root *extent_root)
1381{
1382 u64 start;
1383 u64 end;
1384 struct btrfs_fs_info *info = extent_root->fs_info;
1385 struct extent_buffer *eb;
1386 struct btrfs_path *path;
1387 struct btrfs_key ins;
1388 struct btrfs_disk_key first;
1389 struct btrfs_extent_item extent_item;
1390 int ret;
1391 int level;
1392 int err = 0;
1393
1394 btrfs_set_stack_extent_refs(&extent_item, 1);
1395 btrfs_set_key_type(&ins, BTRFS_EXTENT_ITEM_KEY);
1396 path = btrfs_alloc_path();
1397
1398 while(1) {
1399 ret = find_first_extent_bit(&info->extent_ins, 0, &start,
1400 &end, EXTENT_LOCKED);
1401 if (ret)
1402 break;
1403
1404 ins.objectid = start;
1405 ins.offset = end + 1 - start;
1406 err = btrfs_insert_item(trans, extent_root, &ins,
1407 &extent_item, sizeof(extent_item));
1408 clear_extent_bits(&info->extent_ins, start, end, EXTENT_LOCKED,
1409 GFP_NOFS);
1410 eb = read_tree_block(extent_root, ins.objectid, ins.offset,
1411 trans->transid);
1412 btrfs_tree_lock(eb);
1413 level = btrfs_header_level(eb);
1414 if (level == 0) {
1415 btrfs_item_key(eb, &first, 0);
1416 } else {
1417 btrfs_node_key(eb, &first, 0);
1418 }
1419 btrfs_tree_unlock(eb);
1420 free_extent_buffer(eb);
1421 /*
1422 * the first key is just a hint, so the race we've created
1423 * against reading it is fine
1424 */
1425 err = btrfs_insert_extent_backref(trans, extent_root, path,
1426 start, extent_root->root_key.objectid,
1427 0, level,
1428 btrfs_disk_key_objectid(&first));
1429 BUG_ON(err);
1430 }
1431 btrfs_free_path(path);
1432 return 0;
1433}
1434
1435static int pin_down_bytes(struct btrfs_root *root, u64 bytenr, u32 num_bytes,
1436 int pending)
1437{
1438 int err = 0;
1439
1440 if (!pending) {
1441#if 0
1442 struct extent_buffer *buf;
1443 buf = btrfs_find_tree_block(root, bytenr, num_bytes);
1444 if (buf) {
1445 if (!btrfs_try_tree_lock(buf) &&
1446 btrfs_buffer_uptodate(buf, 0)) {
1447 u64 transid =
1448 root->fs_info->running_transaction->transid;
1449 u64 header_transid =
1450 btrfs_header_generation(buf);
1451 if (header_transid == transid &&
1452 !btrfs_header_flag(buf,
1453 BTRFS_HEADER_FLAG_WRITTEN)) {
1454 clean_tree_block(NULL, root, buf);
1455 btrfs_tree_unlock(buf);
1456 free_extent_buffer(buf);
1457 return 1;
1458 }
1459 btrfs_tree_unlock(buf);
1460 }
1461 free_extent_buffer(buf);
1462 }
1463#endif
1464 update_pinned_extents(root, bytenr, num_bytes, 1);
1465 } else {
1466 set_extent_bits(&root->fs_info->pending_del,
1467 bytenr, bytenr + num_bytes - 1,
1468 EXTENT_LOCKED, GFP_NOFS);
1469 }
1470 BUG_ON(err < 0);
1471 return 0;
1472}
1473
1474/*
1475 * remove an extent from the root, returns 0 on success
1476 */
1477static int __free_extent(struct btrfs_trans_handle *trans, struct btrfs_root
1478 *root, u64 bytenr, u64 num_bytes,
1479 u64 root_objectid, u64 ref_generation,
1480 u64 owner_objectid, u64 owner_offset, int pin,
1481 int mark_free)
1482{
1483 struct btrfs_path *path;
1484 struct btrfs_key key;
1485 struct btrfs_fs_info *info = root->fs_info;
1486 struct btrfs_root *extent_root = info->extent_root;
1487 struct extent_buffer *leaf;
1488 int ret;
1489 int extent_slot = 0;
1490 int found_extent = 0;
1491 int num_to_del = 1;
1492 struct btrfs_extent_item *ei;
1493 u32 refs;
1494
1495 key.objectid = bytenr;
1496 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
1497 key.offset = num_bytes;
1498 path = btrfs_alloc_path();
1499 if (!path)
1500 return -ENOMEM;
1501
1502 path->reada = 1;
1503 ret = lookup_extent_backref(trans, extent_root, path,
1504 bytenr, root_objectid,
1505 ref_generation,
1506 owner_objectid, owner_offset, 1);
1507 if (ret == 0) {
1508 struct btrfs_key found_key;
1509 extent_slot = path->slots[0];
1510 while(extent_slot > 0) {
1511 extent_slot--;
1512 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1513 extent_slot);
1514 if (found_key.objectid != bytenr)
1515 break;
1516 if (found_key.type == BTRFS_EXTENT_ITEM_KEY &&
1517 found_key.offset == num_bytes) {
1518 found_extent = 1;
1519 break;
1520 }
1521 if (path->slots[0] - extent_slot > 5)
1522 break;
1523 }
1524 if (!found_extent)
1525 ret = btrfs_del_item(trans, extent_root, path);
1526 } else {
1527 btrfs_print_leaf(extent_root, path->nodes[0]);
1528 WARN_ON(1);
1529 printk("Unable to find ref byte nr %Lu root %Lu "
1530 " gen %Lu owner %Lu offset %Lu\n", bytenr,
1531 root_objectid, ref_generation, owner_objectid,
1532 owner_offset);
1533 }
1534 if (!found_extent) {
1535 btrfs_release_path(extent_root, path);
1536 ret = btrfs_search_slot(trans, extent_root, &key, path, -1, 1);
1537 if (ret < 0)
1538 return ret;
1539 BUG_ON(ret);
1540 extent_slot = path->slots[0];
1541 }
1542
1543 leaf = path->nodes[0];
1544 ei = btrfs_item_ptr(leaf, extent_slot,
1545 struct btrfs_extent_item);
1546 refs = btrfs_extent_refs(leaf, ei);
1547 BUG_ON(refs == 0);
1548 refs -= 1;
1549 btrfs_set_extent_refs(leaf, ei, refs);
1550
1551 btrfs_mark_buffer_dirty(leaf);
1552
1553 if (refs == 0 && found_extent && path->slots[0] == extent_slot + 1) {
1554 /* if the back ref and the extent are next to each other
1555 * they get deleted below in one shot
1556 */
1557 path->slots[0] = extent_slot;
1558 num_to_del = 2;
1559 } else if (found_extent) {
1560 /* otherwise delete the extent back ref */
1561 ret = btrfs_del_item(trans, extent_root, path);
1562 BUG_ON(ret);
1563 /* if refs are 0, we need to setup the path for deletion */
1564 if (refs == 0) {
1565 btrfs_release_path(extent_root, path);
1566 ret = btrfs_search_slot(trans, extent_root, &key, path,
1567 -1, 1);
1568 if (ret < 0)
1569 return ret;
1570 BUG_ON(ret);
1571 }
1572 }
1573
1574 if (refs == 0) {
1575 u64 super_used;
1576 u64 root_used;
1577
1578 if (pin) {
1579 ret = pin_down_bytes(root, bytenr, num_bytes, 0);
1580 if (ret > 0)
1581 mark_free = 1;
1582 BUG_ON(ret < 0);
1583 }
1584
1585 /* block accounting for super block */
1586 spin_lock_irq(&info->delalloc_lock);
1587 super_used = btrfs_super_bytes_used(&info->super_copy);
1588 btrfs_set_super_bytes_used(&info->super_copy,
1589 super_used - num_bytes);
1590 spin_unlock_irq(&info->delalloc_lock);
1591
1592 /* block accounting for root item */
1593 root_used = btrfs_root_used(&root->root_item);
1594 btrfs_set_root_used(&root->root_item,
1595 root_used - num_bytes);
1596 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
1597 num_to_del);
1598 if (ret) {
1599 return ret;
1600 }
1601 ret = update_block_group(trans, root, bytenr, num_bytes, 0,
1602 mark_free);
1603 BUG_ON(ret);
1604 }
1605 btrfs_free_path(path);
1606 finish_current_insert(trans, extent_root);
1607 return ret;
1608}
1609
1610/*
1611 * find all the blocks marked as pending in the radix tree and remove
1612 * them from the extent map
1613 */
1614static int del_pending_extents(struct btrfs_trans_handle *trans, struct
1615 btrfs_root *extent_root)
1616{
1617 int ret;
1618 int err = 0;
1619 u64 start;
1620 u64 end;
1621 struct extent_io_tree *pending_del;
1622 struct extent_io_tree *pinned_extents;
1623
1624 pending_del = &extent_root->fs_info->pending_del;
1625 pinned_extents = &extent_root->fs_info->pinned_extents;
1626
1627 while(1) {
1628 ret = find_first_extent_bit(pending_del, 0, &start, &end,
1629 EXTENT_LOCKED);
1630 if (ret)
1631 break;
1632 update_pinned_extents(extent_root, start, end + 1 - start, 1);
1633 clear_extent_bits(pending_del, start, end, EXTENT_LOCKED,
1634 GFP_NOFS);
1635 ret = __free_extent(trans, extent_root,
1636 start, end + 1 - start,
1637 extent_root->root_key.objectid,
1638 0, 0, 0, 0, 0);
1639 if (ret)
1640 err = ret;
1641 }
1642 return err;
1643}
1644
1645/*
1646 * remove an extent from the root, returns 0 on success
1647 */
1648static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
1649 struct btrfs_root *root, u64 bytenr,
1650 u64 num_bytes, u64 root_objectid,
1651 u64 ref_generation, u64 owner_objectid,
1652 u64 owner_offset, int pin)
1653{
1654 struct btrfs_root *extent_root = root->fs_info->extent_root;
1655 int pending_ret;
1656 int ret;
1657
1658 WARN_ON(num_bytes < root->sectorsize);
1659 if (!root->ref_cows)
1660 ref_generation = 0;
1661
1662 if (root == extent_root) {
1663 pin_down_bytes(root, bytenr, num_bytes, 1);
1664 return 0;
1665 }
1666 ret = __free_extent(trans, root, bytenr, num_bytes, root_objectid,
1667 ref_generation, owner_objectid, owner_offset,
1668 pin, pin == 0);
1669 pending_ret = del_pending_extents(trans, root->fs_info->extent_root);
1670 return ret ? ret : pending_ret;
1671}
1672
1673int btrfs_free_extent(struct btrfs_trans_handle *trans,
1674 struct btrfs_root *root, u64 bytenr,
1675 u64 num_bytes, u64 root_objectid,
1676 u64 ref_generation, u64 owner_objectid,
1677 u64 owner_offset, int pin)
1678{
1679 int ret;
1680
1681 maybe_lock_mutex(root);
1682 ret = __btrfs_free_extent(trans, root, bytenr, num_bytes,
1683 root_objectid, ref_generation,
1684 owner_objectid, owner_offset, pin);
1685 maybe_unlock_mutex(root);
1686 return ret;
1687}
1688
1689static u64 stripe_align(struct btrfs_root *root, u64 val)
1690{
1691 u64 mask = ((u64)root->stripesize - 1);
1692 u64 ret = (val + mask) & ~mask;
1693 return ret;
1694}
1695
1696/*
1697 * walks the btree of allocated extents and find a hole of a given size.
1698 * The key ins is changed to record the hole:
1699 * ins->objectid == block start
1700 * ins->flags = BTRFS_EXTENT_ITEM_KEY
1701 * ins->offset == number of blocks
1702 * Any available blocks before search_start are skipped.
1703 */
1704static int noinline find_free_extent(struct btrfs_trans_handle *trans,
1705 struct btrfs_root *orig_root,
1706 u64 num_bytes, u64 empty_size,
1707 u64 search_start, u64 search_end,
1708 u64 hint_byte, struct btrfs_key *ins,
1709 u64 exclude_start, u64 exclude_nr,
1710 int data)
1711{
1712 int ret;
1713 u64 orig_search_start;
1714 struct btrfs_root * root = orig_root->fs_info->extent_root;
1715 struct btrfs_fs_info *info = root->fs_info;
1716 u64 total_needed = num_bytes;
1717 u64 *last_ptr = NULL;
1718 struct btrfs_block_group_cache *block_group;
1719 int full_scan = 0;
1720 int wrapped = 0;
1721 int chunk_alloc_done = 0;
1722 int empty_cluster = 2 * 1024 * 1024;
1723 int allowed_chunk_alloc = 0;
1724
1725 WARN_ON(num_bytes < root->sectorsize);
1726 btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
1727
1728 if (orig_root->ref_cows || empty_size)
1729 allowed_chunk_alloc = 1;
1730
1731 if (data & BTRFS_BLOCK_GROUP_METADATA) {
1732 last_ptr = &root->fs_info->last_alloc;
1733 empty_cluster = 256 * 1024;
1734 }
1735
1736 if ((data & BTRFS_BLOCK_GROUP_DATA) && btrfs_test_opt(root, SSD)) {
1737 last_ptr = &root->fs_info->last_data_alloc;
1738 }
1739
1740 if (last_ptr) {
1741 if (*last_ptr)
1742 hint_byte = *last_ptr;
1743 else {
1744 empty_size += empty_cluster;
1745 }
1746 }
1747
1748 search_start = max(search_start, first_logical_byte(root, 0));
1749 orig_search_start = search_start;
1750
1751 if (search_end == (u64)-1)
1752 search_end = btrfs_super_total_bytes(&info->super_copy);
1753
1754 if (hint_byte) {
1755 block_group = btrfs_lookup_first_block_group(info, hint_byte);
1756 if (!block_group)
1757 hint_byte = search_start;
1758 block_group = __btrfs_find_block_group(root, block_group,
1759 hint_byte, data, 1);
1760 if (last_ptr && *last_ptr == 0 && block_group)
1761 hint_byte = block_group->key.objectid;
1762 } else {
1763 block_group = __btrfs_find_block_group(root,
1764 trans->block_group,
1765 search_start, data, 1);
1766 }
1767 search_start = max(search_start, hint_byte);
1768
1769 total_needed += empty_size;
1770
1771check_failed:
1772 if (!block_group) {
1773 block_group = btrfs_lookup_first_block_group(info,
1774 search_start);
1775 if (!block_group)
1776 block_group = btrfs_lookup_first_block_group(info,
1777 orig_search_start);
1778 }
1779 if (full_scan && !chunk_alloc_done) {
1780 if (allowed_chunk_alloc) {
1781 do_chunk_alloc(trans, root,
1782 num_bytes + 2 * 1024 * 1024, data, 1);
1783 allowed_chunk_alloc = 0;
1784 } else if (block_group && block_group_bits(block_group, data)) {
1785 block_group->space_info->force_alloc = 1;
1786 }
1787 chunk_alloc_done = 1;
1788 }
1789 ret = find_search_start(root, &block_group, &search_start,
1790 total_needed, data);
1791 if (ret == -ENOSPC && last_ptr && *last_ptr) {
1792 *last_ptr = 0;
1793 block_group = btrfs_lookup_first_block_group(info,
1794 orig_search_start);
1795 search_start = orig_search_start;
1796 ret = find_search_start(root, &block_group, &search_start,
1797 total_needed, data);
1798 }
1799 if (ret == -ENOSPC)
1800 goto enospc;
1801 if (ret)
1802 goto error;
1803
1804 if (last_ptr && *last_ptr && search_start != *last_ptr) {
1805 *last_ptr = 0;
1806 if (!empty_size) {
1807 empty_size += empty_cluster;
1808 total_needed += empty_size;
1809 }
1810 block_group = btrfs_lookup_first_block_group(info,
1811 orig_search_start);
1812 search_start = orig_search_start;
1813 ret = find_search_start(root, &block_group,
1814 &search_start, total_needed, data);
1815 if (ret == -ENOSPC)
1816 goto enospc;
1817 if (ret)
1818 goto error;
1819 }
1820
1821 search_start = stripe_align(root, search_start);
1822 ins->objectid = search_start;
1823 ins->offset = num_bytes;
1824
1825 if (ins->objectid + num_bytes >= search_end)
1826 goto enospc;
1827
1828 if (ins->objectid + num_bytes >
1829 block_group->key.objectid + block_group->key.offset) {
1830 search_start = block_group->key.objectid +
1831 block_group->key.offset;
1832 goto new_group;
1833 }
1834
1835 if (test_range_bit(&info->extent_ins, ins->objectid,
1836 ins->objectid + num_bytes -1, EXTENT_LOCKED, 0)) {
1837 search_start = ins->objectid + num_bytes;
1838 goto new_group;
1839 }
1840
1841 if (test_range_bit(&info->pinned_extents, ins->objectid,
1842 ins->objectid + num_bytes -1, EXTENT_DIRTY, 0)) {
1843 search_start = ins->objectid + num_bytes;
1844 goto new_group;
1845 }
1846
1847 if (exclude_nr > 0 && (ins->objectid + num_bytes > exclude_start &&
1848 ins->objectid < exclude_start + exclude_nr)) {
1849 search_start = exclude_start + exclude_nr;
1850 goto new_group;
1851 }
1852
1853 if (!(data & BTRFS_BLOCK_GROUP_DATA)) {
1854 block_group = btrfs_lookup_block_group(info, ins->objectid);
1855 if (block_group)
1856 trans->block_group = block_group;
1857 }
1858 ins->offset = num_bytes;
1859 if (last_ptr) {
1860 *last_ptr = ins->objectid + ins->offset;
1861 if (*last_ptr ==
1862 btrfs_super_total_bytes(&root->fs_info->super_copy)) {
1863 *last_ptr = 0;
1864 }
1865 }
1866 return 0;
1867
1868new_group:
1869 if (search_start + num_bytes >= search_end) {
1870enospc:
1871 search_start = orig_search_start;
1872 if (full_scan) {
1873 ret = -ENOSPC;
1874 goto error;
1875 }
1876 if (wrapped) {
1877 if (!full_scan)
1878 total_needed -= empty_size;
1879 full_scan = 1;
1880 } else
1881 wrapped = 1;
1882 }
1883 block_group = btrfs_lookup_first_block_group(info, search_start);
1884 cond_resched();
1885 block_group = __btrfs_find_block_group(root, block_group,
1886 search_start, data, 0);
1887 goto check_failed;
1888
1889error:
1890 return ret;
1891}
1892
1893/*
1894 * finds a free extent and does all the dirty work required for allocation
1895 * returns the key for the extent through ins, and a tree buffer for
1896 * the first block of the extent through buf.
1897 *
1898 * returns 0 if everything worked, non-zero otherwise.
1899 */
1900int btrfs_alloc_extent(struct btrfs_trans_handle *trans,
1901 struct btrfs_root *root,
1902 u64 num_bytes, u64 min_alloc_size,
1903 u64 root_objectid, u64 ref_generation,
1904 u64 owner, u64 owner_offset,
1905 u64 empty_size, u64 hint_byte,
1906 u64 search_end, struct btrfs_key *ins, u64 data)
1907{
1908 int ret;
1909 int pending_ret;
1910 u64 super_used;
1911 u64 root_used;
1912 u64 search_start = 0;
1913 u64 alloc_profile;
1914 u32 sizes[2];
1915 struct btrfs_fs_info *info = root->fs_info;
1916 struct btrfs_root *extent_root = info->extent_root;
1917 struct btrfs_extent_item *extent_item;
1918 struct btrfs_extent_ref *ref;
1919 struct btrfs_path *path;
1920 struct btrfs_key keys[2];
1921
1922 maybe_lock_mutex(root);
1923
1924 if (data) {
1925 alloc_profile = info->avail_data_alloc_bits &
1926 info->data_alloc_profile;
1927 data = BTRFS_BLOCK_GROUP_DATA | alloc_profile;
1928 } else if (root == root->fs_info->chunk_root) {
1929 alloc_profile = info->avail_system_alloc_bits &
1930 info->system_alloc_profile;
1931 data = BTRFS_BLOCK_GROUP_SYSTEM | alloc_profile;
1932 } else {
1933 alloc_profile = info->avail_metadata_alloc_bits &
1934 info->metadata_alloc_profile;
1935 data = BTRFS_BLOCK_GROUP_METADATA | alloc_profile;
1936 }
1937again:
1938 data = reduce_alloc_profile(root, data);
1939 /*
1940 * the only place that sets empty_size is btrfs_realloc_node, which
1941 * is not called recursively on allocations
1942 */
1943 if (empty_size || root->ref_cows) {
1944 if (!(data & BTRFS_BLOCK_GROUP_METADATA)) {
1945 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
1946 2 * 1024 * 1024,
1947 BTRFS_BLOCK_GROUP_METADATA |
1948 (info->metadata_alloc_profile &
1949 info->avail_metadata_alloc_bits), 0);
1950 BUG_ON(ret);
1951 }
1952 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
1953 num_bytes + 2 * 1024 * 1024, data, 0);
1954 BUG_ON(ret);
1955 }
1956
1957 WARN_ON(num_bytes < root->sectorsize);
1958 ret = find_free_extent(trans, root, num_bytes, empty_size,
1959 search_start, search_end, hint_byte, ins,
1960 trans->alloc_exclude_start,
1961 trans->alloc_exclude_nr, data);
1962
1963 if (ret == -ENOSPC && num_bytes > min_alloc_size) {
1964 num_bytes = num_bytes >> 1;
1965 num_bytes = max(num_bytes, min_alloc_size);
1966 do_chunk_alloc(trans, root->fs_info->extent_root,
1967 num_bytes, data, 1);
1968 goto again;
1969 }
1970 if (ret) {
1971 printk("allocation failed flags %Lu\n", data);
1972 }
1973 if (ret) {
1974 BUG();
1975 goto out;
1976 }
1977
1978 /* block accounting for super block */
1979 spin_lock_irq(&info->delalloc_lock);
1980 super_used = btrfs_super_bytes_used(&info->super_copy);
1981 btrfs_set_super_bytes_used(&info->super_copy, super_used + num_bytes);
1982 spin_unlock_irq(&info->delalloc_lock);
1983
1984 /* block accounting for root item */
1985 root_used = btrfs_root_used(&root->root_item);
1986 btrfs_set_root_used(&root->root_item, root_used + num_bytes);
1987
1988 clear_extent_dirty(&root->fs_info->free_space_cache,
1989 ins->objectid, ins->objectid + ins->offset - 1,
1990 GFP_NOFS);
1991
1992 if (root == extent_root) {
1993 set_extent_bits(&root->fs_info->extent_ins, ins->objectid,
1994 ins->objectid + ins->offset - 1,
1995 EXTENT_LOCKED, GFP_NOFS);
1996 goto update_block;
1997 }
1998
1999 WARN_ON(trans->alloc_exclude_nr);
2000 trans->alloc_exclude_start = ins->objectid;
2001 trans->alloc_exclude_nr = ins->offset;
2002
2003 memcpy(&keys[0], ins, sizeof(*ins));
2004 keys[1].offset = hash_extent_ref(root_objectid, ref_generation,
2005 owner, owner_offset);
2006 keys[1].objectid = ins->objectid;
2007 keys[1].type = BTRFS_EXTENT_REF_KEY;
2008 sizes[0] = sizeof(*extent_item);
2009 sizes[1] = sizeof(*ref);
2010
2011 path = btrfs_alloc_path();
2012 BUG_ON(!path);
2013
2014 ret = btrfs_insert_empty_items(trans, extent_root, path, keys,
2015 sizes, 2);
2016
2017 BUG_ON(ret);
2018 extent_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
2019 struct btrfs_extent_item);
2020 btrfs_set_extent_refs(path->nodes[0], extent_item, 1);
2021 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
2022 struct btrfs_extent_ref);
2023
2024 btrfs_set_ref_root(path->nodes[0], ref, root_objectid);
2025 btrfs_set_ref_generation(path->nodes[0], ref, ref_generation);
2026 btrfs_set_ref_objectid(path->nodes[0], ref, owner);
2027 btrfs_set_ref_offset(path->nodes[0], ref, owner_offset);
2028
2029 btrfs_mark_buffer_dirty(path->nodes[0]);
2030
2031 trans->alloc_exclude_start = 0;
2032 trans->alloc_exclude_nr = 0;
2033 btrfs_free_path(path);
2034 finish_current_insert(trans, extent_root);
2035 pending_ret = del_pending_extents(trans, extent_root);
2036
2037 if (ret)
2038 goto out;
2039 if (pending_ret) {
2040 ret = pending_ret;
2041 goto out;
2042 }
2043
2044update_block:
2045 ret = update_block_group(trans, root, ins->objectid, ins->offset, 1, 0);
2046 if (ret) {
2047 printk("update block group failed for %Lu %Lu\n",
2048 ins->objectid, ins->offset);
2049 BUG();
2050 }
2051out:
2052 maybe_unlock_mutex(root);
2053 return ret;
2054}
2055/*
2056 * helper function to allocate a block for a given tree
2057 * returns the tree buffer or NULL.
2058 */
2059struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
2060 struct btrfs_root *root,
2061 u32 blocksize,
2062 u64 root_objectid,
2063 u64 ref_generation,
2064 u64 first_objectid,
2065 int level,
2066 u64 hint,
2067 u64 empty_size)
2068{
2069 struct btrfs_key ins;
2070 int ret;
2071 struct extent_buffer *buf;
2072
2073 ret = btrfs_alloc_extent(trans, root, blocksize, blocksize,
2074 root_objectid, ref_generation,
2075 level, first_objectid, empty_size, hint,
2076 (u64)-1, &ins, 0);
2077 if (ret) {
2078 BUG_ON(ret > 0);
2079 return ERR_PTR(ret);
2080 }
2081 buf = btrfs_find_create_tree_block(root, ins.objectid, blocksize);
2082 if (!buf) {
2083 btrfs_free_extent(trans, root, ins.objectid, blocksize,
2084 root->root_key.objectid, ref_generation,
2085 0, 0, 0);
2086 return ERR_PTR(-ENOMEM);
2087 }
2088 btrfs_set_header_generation(buf, trans->transid);
2089 btrfs_tree_lock(buf);
2090 clean_tree_block(trans, root, buf);
2091 btrfs_set_buffer_uptodate(buf);
2092
2093 if (PageDirty(buf->first_page)) {
2094 printk("page %lu dirty\n", buf->first_page->index);
2095 WARN_ON(1);
2096 }
2097
2098 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
2099 buf->start + buf->len - 1, GFP_NOFS);
2100 if (!btrfs_test_opt(root, SSD))
2101 btrfs_set_buffer_defrag(buf);
2102 trans->blocks_used++;
2103 return buf;
2104}
2105
2106static int noinline drop_leaf_ref(struct btrfs_trans_handle *trans,
2107 struct btrfs_root *root,
2108 struct extent_buffer *leaf)
2109{
2110 u64 leaf_owner;
2111 u64 leaf_generation;
2112 struct btrfs_key key;
2113 struct btrfs_file_extent_item *fi;
2114 int i;
2115 int nritems;
2116 int ret;
2117
2118 BUG_ON(!btrfs_is_leaf(leaf));
2119 nritems = btrfs_header_nritems(leaf);
2120 leaf_owner = btrfs_header_owner(leaf);
2121 leaf_generation = btrfs_header_generation(leaf);
2122
2123 for (i = 0; i < nritems; i++) {
2124 u64 disk_bytenr;
2125
2126 btrfs_item_key_to_cpu(leaf, &key, i);
2127 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2128 continue;
2129 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
2130 if (btrfs_file_extent_type(leaf, fi) ==
2131 BTRFS_FILE_EXTENT_INLINE)
2132 continue;
2133 /*
2134 * FIXME make sure to insert a trans record that
2135 * repeats the snapshot del on crash
2136 */
2137 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
2138 if (disk_bytenr == 0)
2139 continue;
2140 ret = __btrfs_free_extent(trans, root, disk_bytenr,
2141 btrfs_file_extent_disk_num_bytes(leaf, fi),
2142 leaf_owner, leaf_generation,
2143 key.objectid, key.offset, 0);
2144 BUG_ON(ret);
2145 }
2146 return 0;
2147}
2148
2149static void noinline reada_walk_down(struct btrfs_root *root,
2150 struct extent_buffer *node,
2151 int slot)
2152{
2153 u64 bytenr;
2154 u64 last = 0;
2155 u32 nritems;
2156 u32 refs;
2157 u32 blocksize;
2158 int ret;
2159 int i;
2160 int level;
2161 int skipped = 0;
2162
2163 nritems = btrfs_header_nritems(node);
2164 level = btrfs_header_level(node);
2165 if (level)
2166 return;
2167
2168 for (i = slot; i < nritems && skipped < 32; i++) {
2169 bytenr = btrfs_node_blockptr(node, i);
2170 if (last && ((bytenr > last && bytenr - last > 32 * 1024) ||
2171 (last > bytenr && last - bytenr > 32 * 1024))) {
2172 skipped++;
2173 continue;
2174 }
2175 blocksize = btrfs_level_size(root, level - 1);
2176 if (i != slot) {
2177 ret = lookup_extent_ref(NULL, root, bytenr,
2178 blocksize, &refs);
2179 BUG_ON(ret);
2180 if (refs != 1) {
2181 skipped++;
2182 continue;
2183 }
2184 }
2185 ret = readahead_tree_block(root, bytenr, blocksize,
2186 btrfs_node_ptr_generation(node, i));
2187 last = bytenr + blocksize;
2188 cond_resched();
2189 if (ret)
2190 break;
2191 }
2192}
2193
2194/*
2195 * we want to avoid as much random IO as we can with the alloc mutex
2196 * held, so drop the lock and do the lookup, then do it again with the
2197 * lock held.
2198 */
2199int drop_snap_lookup_refcount(struct btrfs_root *root, u64 start, u64 len,
2200 u32 *refs)
2201{
2202 mutex_unlock(&root->fs_info->alloc_mutex);
2203 lookup_extent_ref(NULL, root, start, len, refs);
2204 mutex_lock(&root->fs_info->alloc_mutex);
2205 return lookup_extent_ref(NULL, root, start, len, refs);
2206}
2207
2208/*
2209 * helper function for drop_snapshot, this walks down the tree dropping ref
2210 * counts as it goes.
2211 */
2212static int noinline walk_down_tree(struct btrfs_trans_handle *trans,
2213 struct btrfs_root *root,
2214 struct btrfs_path *path, int *level)
2215{
2216 u64 root_owner;
2217 u64 root_gen;
2218 u64 bytenr;
2219 u64 ptr_gen;
2220 struct extent_buffer *next;
2221 struct extent_buffer *cur;
2222 struct extent_buffer *parent;
2223 u32 blocksize;
2224 int ret;
2225 u32 refs;
2226
2227 mutex_lock(&root->fs_info->alloc_mutex);
2228
2229 WARN_ON(*level < 0);
2230 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2231 ret = drop_snap_lookup_refcount(root, path->nodes[*level]->start,
2232 path->nodes[*level]->len, &refs);
2233 BUG_ON(ret);
2234 if (refs > 1)
2235 goto out;
2236
2237 /*
2238 * walk down to the last node level and free all the leaves
2239 */
2240 while(*level >= 0) {
2241 WARN_ON(*level < 0);
2242 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2243 cur = path->nodes[*level];
2244
2245 if (btrfs_header_level(cur) != *level)
2246 WARN_ON(1);
2247
2248 if (path->slots[*level] >=
2249 btrfs_header_nritems(cur))
2250 break;
2251 if (*level == 0) {
2252 ret = drop_leaf_ref(trans, root, cur);
2253 BUG_ON(ret);
2254 break;
2255 }
2256 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
2257 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
2258 blocksize = btrfs_level_size(root, *level - 1);
2259
2260 ret = drop_snap_lookup_refcount(root, bytenr, blocksize, &refs);
2261 BUG_ON(ret);
2262 if (refs != 1) {
2263 parent = path->nodes[*level];
2264 root_owner = btrfs_header_owner(parent);
2265 root_gen = btrfs_header_generation(parent);
2266 path->slots[*level]++;
2267 ret = __btrfs_free_extent(trans, root, bytenr,
2268 blocksize, root_owner,
2269 root_gen, 0, 0, 1);
2270 BUG_ON(ret);
2271 continue;
2272 }
2273 next = btrfs_find_tree_block(root, bytenr, blocksize);
2274 if (!next || !btrfs_buffer_uptodate(next, ptr_gen)) {
2275 free_extent_buffer(next);
2276 mutex_unlock(&root->fs_info->alloc_mutex);
2277
2278 if (path->slots[*level] == 0)
2279 reada_walk_down(root, cur, path->slots[*level]);
2280
2281 next = read_tree_block(root, bytenr, blocksize,
2282 ptr_gen);
2283 mutex_lock(&root->fs_info->alloc_mutex);
2284
2285 /* we've dropped the lock, double check */
2286 ret = drop_snap_lookup_refcount(root, bytenr,
2287 blocksize, &refs);
2288 BUG_ON(ret);
2289 if (refs != 1) {
2290 parent = path->nodes[*level];
2291 root_owner = btrfs_header_owner(parent);
2292 root_gen = btrfs_header_generation(parent);
2293
2294 path->slots[*level]++;
2295 free_extent_buffer(next);
2296 ret = __btrfs_free_extent(trans, root, bytenr,
2297 blocksize,
2298 root_owner,
2299 root_gen, 0, 0, 1);
2300 BUG_ON(ret);
2301 continue;
2302 }
2303 }
2304 WARN_ON(*level <= 0);
2305 if (path->nodes[*level-1])
2306 free_extent_buffer(path->nodes[*level-1]);
2307 path->nodes[*level-1] = next;
2308 *level = btrfs_header_level(next);
2309 path->slots[*level] = 0;
2310 }
2311out:
2312 WARN_ON(*level < 0);
2313 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2314
2315 if (path->nodes[*level] == root->node) {
2316 root_owner = root->root_key.objectid;
2317 parent = path->nodes[*level];
2318 } else {
2319 parent = path->nodes[*level + 1];
2320 root_owner = btrfs_header_owner(parent);
2321 }
2322
2323 root_gen = btrfs_header_generation(parent);
2324 ret = __btrfs_free_extent(trans, root, path->nodes[*level]->start,
2325 path->nodes[*level]->len,
2326 root_owner, root_gen, 0, 0, 1);
2327 free_extent_buffer(path->nodes[*level]);
2328 path->nodes[*level] = NULL;
2329 *level += 1;
2330 BUG_ON(ret);
2331 mutex_unlock(&root->fs_info->alloc_mutex);
2332 return 0;
2333}
2334
2335/*
2336 * helper for dropping snapshots. This walks back up the tree in the path
2337 * to find the first node higher up where we haven't yet gone through
2338 * all the slots
2339 */
2340static int noinline walk_up_tree(struct btrfs_trans_handle *trans,
2341 struct btrfs_root *root,
2342 struct btrfs_path *path, int *level)
2343{
2344 u64 root_owner;
2345 u64 root_gen;
2346 struct btrfs_root_item *root_item = &root->root_item;
2347 int i;
2348 int slot;
2349 int ret;
2350
2351 for(i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) {
2352 slot = path->slots[i];
2353 if (slot < btrfs_header_nritems(path->nodes[i]) - 1) {
2354 struct extent_buffer *node;
2355 struct btrfs_disk_key disk_key;
2356 node = path->nodes[i];
2357 path->slots[i]++;
2358 *level = i;
2359 WARN_ON(*level == 0);
2360 btrfs_node_key(node, &disk_key, path->slots[i]);
2361 memcpy(&root_item->drop_progress,
2362 &disk_key, sizeof(disk_key));
2363 root_item->drop_level = i;
2364 return 0;
2365 } else {
2366 if (path->nodes[*level] == root->node) {
2367 root_owner = root->root_key.objectid;
2368 root_gen =
2369 btrfs_header_generation(path->nodes[*level]);
2370 } else {
2371 struct extent_buffer *node;
2372 node = path->nodes[*level + 1];
2373 root_owner = btrfs_header_owner(node);
2374 root_gen = btrfs_header_generation(node);
2375 }
2376 ret = btrfs_free_extent(trans, root,
2377 path->nodes[*level]->start,
2378 path->nodes[*level]->len,
2379 root_owner, root_gen, 0, 0, 1);
2380 BUG_ON(ret);
2381 free_extent_buffer(path->nodes[*level]);
2382 path->nodes[*level] = NULL;
2383 *level = i + 1;
2384 }
2385 }
2386 return 1;
2387}
2388
2389/*
2390 * drop the reference count on the tree rooted at 'snap'. This traverses
2391 * the tree freeing any blocks that have a ref count of zero after being
2392 * decremented.
2393 */
2394int btrfs_drop_snapshot(struct btrfs_trans_handle *trans, struct btrfs_root
2395 *root)
2396{
2397 int ret = 0;
2398 int wret;
2399 int level;
2400 struct btrfs_path *path;
2401 int i;
2402 int orig_level;
2403 struct btrfs_root_item *root_item = &root->root_item;
2404
2405 WARN_ON(!mutex_is_locked(&root->fs_info->drop_mutex));
2406 path = btrfs_alloc_path();
2407 BUG_ON(!path);
2408
2409 level = btrfs_header_level(root->node);
2410 orig_level = level;
2411 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
2412 path->nodes[level] = root->node;
2413 extent_buffer_get(root->node);
2414 path->slots[level] = 0;
2415 } else {
2416 struct btrfs_key key;
2417 struct btrfs_disk_key found_key;
2418 struct extent_buffer *node;
2419
2420 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
2421 level = root_item->drop_level;
2422 path->lowest_level = level;
2423 wret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2424 if (wret < 0) {
2425 ret = wret;
2426 goto out;
2427 }
2428 node = path->nodes[level];
2429 btrfs_node_key(node, &found_key, path->slots[level]);
2430 WARN_ON(memcmp(&found_key, &root_item->drop_progress,
2431 sizeof(found_key)));
2432 for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
2433 if (path->nodes[i] && path->locks[i]) {
2434 path->locks[i] = 0;
2435 btrfs_tree_unlock(path->nodes[i]);
2436 }
2437 }
2438 }
2439 while(1) {
2440 wret = walk_down_tree(trans, root, path, &level);
2441 if (wret > 0)
2442 break;
2443 if (wret < 0)
2444 ret = wret;
2445
2446 wret = walk_up_tree(trans, root, path, &level);
2447 if (wret > 0)
2448 break;
2449 if (wret < 0)
2450 ret = wret;
2451 }
2452 for (i = 0; i <= orig_level; i++) {
2453 if (path->nodes[i]) {
2454 free_extent_buffer(path->nodes[i]);
2455 path->nodes[i] = NULL;
2456 }
2457 }
2458out:
2459 btrfs_free_path(path);
2460 return ret;
2461}
2462
2463int btrfs_free_block_groups(struct btrfs_fs_info *info)
2464{
2465 u64 start;
2466 u64 end;
2467 u64 ptr;
2468 int ret;
2469
2470 mutex_lock(&info->alloc_mutex);
2471 while(1) {
2472 ret = find_first_extent_bit(&info->block_group_cache, 0,
2473 &start, &end, (unsigned int)-1);
2474 if (ret)
2475 break;
2476 ret = get_state_private(&info->block_group_cache, start, &ptr);
2477 if (!ret)
2478 kfree((void *)(unsigned long)ptr);
2479 clear_extent_bits(&info->block_group_cache, start,
2480 end, (unsigned int)-1, GFP_NOFS);
2481 }
2482 while(1) {
2483 ret = find_first_extent_bit(&info->free_space_cache, 0,
2484 &start, &end, EXTENT_DIRTY);
2485 if (ret)
2486 break;
2487 clear_extent_dirty(&info->free_space_cache, start,
2488 end, GFP_NOFS);
2489 }
2490 mutex_unlock(&info->alloc_mutex);
2491 return 0;
2492}
2493
2494static unsigned long calc_ra(unsigned long start, unsigned long last,
2495 unsigned long nr)
2496{
2497 return min(last, start + nr - 1);
2498}
2499
2500static int noinline relocate_inode_pages(struct inode *inode, u64 start,
2501 u64 len)
2502{
2503 u64 page_start;
2504 u64 page_end;
2505 unsigned long last_index;
2506 unsigned long i;
2507 struct page *page;
2508 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2509 struct file_ra_state *ra;
2510 unsigned long total_read = 0;
2511 unsigned long ra_pages;
2512 struct btrfs_trans_handle *trans;
2513
2514 ra = kzalloc(sizeof(*ra), GFP_NOFS);
2515
2516 mutex_lock(&inode->i_mutex);
2517 i = start >> PAGE_CACHE_SHIFT;
2518 last_index = (start + len - 1) >> PAGE_CACHE_SHIFT;
2519
2520 ra_pages = BTRFS_I(inode)->root->fs_info->bdi.ra_pages;
2521
2522 file_ra_state_init(ra, inode->i_mapping);
2523
2524 for (; i <= last_index; i++) {
2525 if (total_read % ra_pages == 0) {
2526 btrfs_force_ra(inode->i_mapping, ra, NULL, i,
2527 calc_ra(i, last_index, ra_pages));
2528 }
2529 total_read++;
2530 if (((u64)i << PAGE_CACHE_SHIFT) > inode->i_size)
2531 goto truncate_racing;
2532
2533 page = grab_cache_page(inode->i_mapping, i);
2534 if (!page) {
2535 goto out_unlock;
2536 }
2537 if (!PageUptodate(page)) {
2538 btrfs_readpage(NULL, page);
2539 lock_page(page);
2540 if (!PageUptodate(page)) {
2541 unlock_page(page);
2542 page_cache_release(page);
2543 goto out_unlock;
2544 }
2545 }
2546#if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
2547 ClearPageDirty(page);
2548#else
2549 cancel_dirty_page(page, PAGE_CACHE_SIZE);
2550#endif
2551 wait_on_page_writeback(page);
2552 set_page_extent_mapped(page);
2553 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
2554 page_end = page_start + PAGE_CACHE_SIZE - 1;
2555
2556 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
2557
2558 set_extent_delalloc(io_tree, page_start,
2559 page_end, GFP_NOFS);
2560 set_page_dirty(page);
2561
2562 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
2563 unlock_page(page);
2564 page_cache_release(page);
2565 }
2566 balance_dirty_pages_ratelimited_nr(inode->i_mapping,
2567 total_read);
2568
2569out_unlock:
2570 kfree(ra);
2571 trans = btrfs_start_transaction(BTRFS_I(inode)->root, 1);
2572 if (trans) {
2573 btrfs_add_ordered_inode(inode);
2574 btrfs_end_transaction(trans, BTRFS_I(inode)->root);
2575 mark_inode_dirty(inode);
2576 }
2577 mutex_unlock(&inode->i_mutex);
2578 return 0;
2579
2580truncate_racing:
2581 vmtruncate(inode, inode->i_size);
2582 balance_dirty_pages_ratelimited_nr(inode->i_mapping,
2583 total_read);
2584 goto out_unlock;
2585}
2586
2587/*
2588 * The back references tell us which tree holds a ref on a block,
2589 * but it is possible for the tree root field in the reference to
2590 * reflect the original root before a snapshot was made. In this
2591 * case we should search through all the children of a given root
2592 * to find potential holders of references on a block.
2593 *
2594 * Instead, we do something a little less fancy and just search
2595 * all the roots for a given key/block combination.
2596 */
2597static int find_root_for_ref(struct btrfs_root *root,
2598 struct btrfs_path *path,
2599 struct btrfs_key *key0,
2600 int level,
2601 int file_key,
2602 struct btrfs_root **found_root,
2603 u64 bytenr)
2604{
2605 struct btrfs_key root_location;
2606 struct btrfs_root *cur_root = *found_root;
2607 struct btrfs_file_extent_item *file_extent;
2608 u64 root_search_start = BTRFS_FS_TREE_OBJECTID;
2609 u64 found_bytenr;
2610 int ret;
2611 int i;
2612
2613 root_location.offset = (u64)-1;
2614 root_location.type = BTRFS_ROOT_ITEM_KEY;
2615 path->lowest_level = level;
2616 path->reada = 0;
2617 while(1) {
2618 ret = btrfs_search_slot(NULL, cur_root, key0, path, 0, 0);
2619 found_bytenr = 0;
2620 if (ret == 0 && file_key) {
2621 struct extent_buffer *leaf = path->nodes[0];
2622 file_extent = btrfs_item_ptr(leaf, path->slots[0],
2623 struct btrfs_file_extent_item);
2624 if (btrfs_file_extent_type(leaf, file_extent) ==
2625 BTRFS_FILE_EXTENT_REG) {
2626 found_bytenr =
2627 btrfs_file_extent_disk_bytenr(leaf,
2628 file_extent);
2629 }
2630 } else if (!file_key) {
2631 if (path->nodes[level])
2632 found_bytenr = path->nodes[level]->start;
2633 }
2634
2635 for (i = level; i < BTRFS_MAX_LEVEL; i++) {
2636 if (!path->nodes[i])
2637 break;
2638 free_extent_buffer(path->nodes[i]);
2639 path->nodes[i] = NULL;
2640 }
2641 btrfs_release_path(cur_root, path);
2642
2643 if (found_bytenr == bytenr) {
2644 *found_root = cur_root;
2645 ret = 0;
2646 goto out;
2647 }
2648 ret = btrfs_search_root(root->fs_info->tree_root,
2649 root_search_start, &root_search_start);
2650 if (ret)
2651 break;
2652
2653 root_location.objectid = root_search_start;
2654 cur_root = btrfs_read_fs_root_no_name(root->fs_info,
2655 &root_location);
2656 if (!cur_root) {
2657 ret = 1;
2658 break;
2659 }
2660 }
2661out:
2662 path->lowest_level = 0;
2663 return ret;
2664}
2665
2666/*
2667 * note, this releases the path
2668 */
2669static int noinline relocate_one_reference(struct btrfs_root *extent_root,
2670 struct btrfs_path *path,
2671 struct btrfs_key *extent_key,
2672 u64 *last_file_objectid,
2673 u64 *last_file_offset,
2674 u64 *last_file_root,
2675 u64 last_extent)
2676{
2677 struct inode *inode;
2678 struct btrfs_root *found_root;
2679 struct btrfs_key root_location;
2680 struct btrfs_key found_key;
2681 struct btrfs_extent_ref *ref;
2682 u64 ref_root;
2683 u64 ref_gen;
2684 u64 ref_objectid;
2685 u64 ref_offset;
2686 int ret;
2687 int level;
2688
2689 ref = btrfs_item_ptr(path->nodes[0], path->slots[0],
2690 struct btrfs_extent_ref);
2691 ref_root = btrfs_ref_root(path->nodes[0], ref);
2692 ref_gen = btrfs_ref_generation(path->nodes[0], ref);
2693 ref_objectid = btrfs_ref_objectid(path->nodes[0], ref);
2694 ref_offset = btrfs_ref_offset(path->nodes[0], ref);
2695 btrfs_release_path(extent_root, path);
2696
2697 root_location.objectid = ref_root;
2698 if (ref_gen == 0)
2699 root_location.offset = 0;
2700 else
2701 root_location.offset = (u64)-1;
2702 root_location.type = BTRFS_ROOT_ITEM_KEY;
2703
2704 found_root = btrfs_read_fs_root_no_name(extent_root->fs_info,
2705 &root_location);
2706 BUG_ON(!found_root);
2707
2708 if (ref_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
2709 found_key.objectid = ref_objectid;
2710 found_key.type = BTRFS_EXTENT_DATA_KEY;
2711 found_key.offset = ref_offset;
2712 level = 0;
2713
2714 if (last_extent == extent_key->objectid &&
2715 *last_file_objectid == ref_objectid &&
2716 *last_file_offset == ref_offset &&
2717 *last_file_root == ref_root)
2718 goto out;
2719
2720 ret = find_root_for_ref(extent_root, path, &found_key,
2721 level, 1, &found_root,
2722 extent_key->objectid);
2723
2724 if (ret)
2725 goto out;
2726
2727 if (last_extent == extent_key->objectid &&
2728 *last_file_objectid == ref_objectid &&
2729 *last_file_offset == ref_offset &&
2730 *last_file_root == ref_root)
2731 goto out;
2732
2733 inode = btrfs_iget_locked(extent_root->fs_info->sb,
2734 ref_objectid, found_root);
2735 if (inode->i_state & I_NEW) {
2736 /* the inode and parent dir are two different roots */
2737 BTRFS_I(inode)->root = found_root;
2738 BTRFS_I(inode)->location.objectid = ref_objectid;
2739 BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
2740 BTRFS_I(inode)->location.offset = 0;
2741 btrfs_read_locked_inode(inode);
2742 unlock_new_inode(inode);
2743
2744 }
2745 /* this can happen if the reference is not against
2746 * the latest version of the tree root
2747 */
2748 if (is_bad_inode(inode)) {
2749 goto out;
2750 }
2751 *last_file_objectid = inode->i_ino;
2752 *last_file_root = found_root->root_key.objectid;
2753 *last_file_offset = ref_offset;
2754
2755 relocate_inode_pages(inode, ref_offset, extent_key->offset);
2756 iput(inode);
2757 } else {
2758 struct btrfs_trans_handle *trans;
2759 struct extent_buffer *eb;
2760 int i;
2761
2762 eb = read_tree_block(found_root, extent_key->objectid,
2763 extent_key->offset, 0);
2764 btrfs_tree_lock(eb);
2765 level = btrfs_header_level(eb);
2766
2767 if (level == 0)
2768 btrfs_item_key_to_cpu(eb, &found_key, 0);
2769 else
2770 btrfs_node_key_to_cpu(eb, &found_key, 0);
2771
2772 btrfs_tree_unlock(eb);
2773 free_extent_buffer(eb);
2774
2775 ret = find_root_for_ref(extent_root, path, &found_key,
2776 level, 0, &found_root,
2777 extent_key->objectid);
2778
2779 if (ret)
2780 goto out;
2781
2782 trans = btrfs_start_transaction(found_root, 1);
2783
2784 path->lowest_level = level;
2785 path->reada = 2;
2786 ret = btrfs_search_slot(trans, found_root, &found_key, path,
2787 0, 1);
2788 path->lowest_level = 0;
2789 for (i = level; i < BTRFS_MAX_LEVEL; i++) {
2790 if (!path->nodes[i])
2791 break;
2792 free_extent_buffer(path->nodes[i]);
2793 path->nodes[i] = NULL;
2794 }
2795 btrfs_release_path(found_root, path);
2796 if (found_root == found_root->fs_info->extent_root)
2797 btrfs_extent_post_op(trans, found_root);
2798 btrfs_end_transaction(trans, found_root);
2799 }
2800
2801out:
2802 return 0;
2803}
2804
2805static int noinline del_extent_zero(struct btrfs_root *extent_root,
2806 struct btrfs_path *path,
2807 struct btrfs_key *extent_key)
2808{
2809 int ret;
2810 struct btrfs_trans_handle *trans;
2811
2812 trans = btrfs_start_transaction(extent_root, 1);
2813 ret = btrfs_search_slot(trans, extent_root, extent_key, path, -1, 1);
2814 if (ret > 0) {
2815 ret = -EIO;
2816 goto out;
2817 }
2818 if (ret < 0)
2819 goto out;
2820 ret = btrfs_del_item(trans, extent_root, path);
2821out:
2822 btrfs_end_transaction(trans, extent_root);
2823 return ret;
2824}
2825
2826static int noinline relocate_one_extent(struct btrfs_root *extent_root,
2827 struct btrfs_path *path,
2828 struct btrfs_key *extent_key)
2829{
2830 struct btrfs_key key;
2831 struct btrfs_key found_key;
2832 struct extent_buffer *leaf;
2833 u64 last_file_objectid = 0;
2834 u64 last_file_root = 0;
2835 u64 last_file_offset = (u64)-1;
2836 u64 last_extent = 0;
2837 u32 nritems;
2838 u32 item_size;
2839 int ret = 0;
2840
2841 if (extent_key->objectid == 0) {
2842 ret = del_extent_zero(extent_root, path, extent_key);
2843 goto out;
2844 }
2845 key.objectid = extent_key->objectid;
2846 key.type = BTRFS_EXTENT_REF_KEY;
2847 key.offset = 0;
2848
2849 while(1) {
2850 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2851
2852 if (ret < 0)
2853 goto out;
2854
2855 ret = 0;
2856 leaf = path->nodes[0];
2857 nritems = btrfs_header_nritems(leaf);
2858 if (path->slots[0] == nritems) {
2859 ret = btrfs_next_leaf(extent_root, path);
2860 if (ret > 0) {
2861 ret = 0;
2862 goto out;
2863 }
2864 if (ret < 0)
2865 goto out;
2866 leaf = path->nodes[0];
2867 }
2868
2869 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2870 if (found_key.objectid != extent_key->objectid) {
2871 break;
2872 }
2873
2874 if (found_key.type != BTRFS_EXTENT_REF_KEY) {
2875 break;
2876 }
2877
2878 key.offset = found_key.offset + 1;
2879 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2880
2881 ret = relocate_one_reference(extent_root, path, extent_key,
2882 &last_file_objectid,
2883 &last_file_offset,
2884 &last_file_root, last_extent);
2885 if (ret)
2886 goto out;
2887 last_extent = extent_key->objectid;
2888 }
2889 ret = 0;
2890out:
2891 btrfs_release_path(extent_root, path);
2892 return ret;
2893}
2894
2895static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
2896{
2897 u64 num_devices;
2898 u64 stripped = BTRFS_BLOCK_GROUP_RAID0 |
2899 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
2900
2901 num_devices = root->fs_info->fs_devices->num_devices;
2902 if (num_devices == 1) {
2903 stripped |= BTRFS_BLOCK_GROUP_DUP;
2904 stripped = flags & ~stripped;
2905
2906 /* turn raid0 into single device chunks */
2907 if (flags & BTRFS_BLOCK_GROUP_RAID0)
2908 return stripped;
2909
2910 /* turn mirroring into duplication */
2911 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
2912 BTRFS_BLOCK_GROUP_RAID10))
2913 return stripped | BTRFS_BLOCK_GROUP_DUP;
2914 return flags;
2915 } else {
2916 /* they already had raid on here, just return */
2917 if (flags & stripped)
2918 return flags;
2919
2920 stripped |= BTRFS_BLOCK_GROUP_DUP;
2921 stripped = flags & ~stripped;
2922
2923 /* switch duplicated blocks with raid1 */
2924 if (flags & BTRFS_BLOCK_GROUP_DUP)
2925 return stripped | BTRFS_BLOCK_GROUP_RAID1;
2926
2927 /* turn single device chunks into raid0 */
2928 return stripped | BTRFS_BLOCK_GROUP_RAID0;
2929 }
2930 return flags;
2931}
2932
2933int __alloc_chunk_for_shrink(struct btrfs_root *root,
2934 struct btrfs_block_group_cache *shrink_block_group,
2935 int force)
2936{
2937 struct btrfs_trans_handle *trans;
2938 u64 new_alloc_flags;
2939 u64 calc;
2940
2941 if (btrfs_block_group_used(&shrink_block_group->item) > 0) {
2942
2943 trans = btrfs_start_transaction(root, 1);
2944 new_alloc_flags = update_block_group_flags(root,
2945 shrink_block_group->flags);
2946 if (new_alloc_flags != shrink_block_group->flags) {
2947 calc =
2948 btrfs_block_group_used(&shrink_block_group->item);
2949 } else {
2950 calc = shrink_block_group->key.offset;
2951 }
2952 do_chunk_alloc(trans, root->fs_info->extent_root,
2953 calc + 2 * 1024 * 1024, new_alloc_flags, force);
2954 btrfs_end_transaction(trans, root);
2955 }
2956 return 0;
2957}
2958
2959int btrfs_shrink_extent_tree(struct btrfs_root *root, u64 shrink_start)
2960{
2961 struct btrfs_trans_handle *trans;
2962 struct btrfs_root *tree_root = root->fs_info->tree_root;
2963 struct btrfs_path *path;
2964 u64 cur_byte;
2965 u64 total_found;
2966 u64 shrink_last_byte;
2967 struct btrfs_block_group_cache *shrink_block_group;
2968 struct btrfs_fs_info *info = root->fs_info;
2969 struct btrfs_key key;
2970 struct btrfs_key found_key;
2971 struct extent_buffer *leaf;
2972 u32 nritems;
2973 int ret;
2974 int progress;
2975
2976 mutex_lock(&root->fs_info->alloc_mutex);
2977 shrink_block_group = btrfs_lookup_block_group(root->fs_info,
2978 shrink_start);
2979 BUG_ON(!shrink_block_group);
2980
2981 shrink_last_byte = shrink_block_group->key.objectid +
2982 shrink_block_group->key.offset;
2983
2984 shrink_block_group->space_info->total_bytes -=
2985 shrink_block_group->key.offset;
2986 path = btrfs_alloc_path();
2987 root = root->fs_info->extent_root;
2988 path->reada = 2;
2989
2990 printk("btrfs relocating block group %llu flags %llu\n",
2991 (unsigned long long)shrink_start,
2992 (unsigned long long)shrink_block_group->flags);
2993
2994 __alloc_chunk_for_shrink(root, shrink_block_group, 1);
2995
2996again:
2997
2998 shrink_block_group->ro = 1;
2999
3000 total_found = 0;
3001 progress = 0;
3002 key.objectid = shrink_start;
3003 key.offset = 0;
3004 key.type = 0;
3005 cur_byte = key.objectid;
3006
3007 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3008 if (ret < 0)
3009 goto out;
3010
3011 ret = btrfs_previous_item(root, path, 0, BTRFS_EXTENT_ITEM_KEY);
3012 if (ret < 0)
3013 goto out;
3014
3015 if (ret == 0) {
3016 leaf = path->nodes[0];
3017 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3018 if (found_key.objectid + found_key.offset > shrink_start &&
3019 found_key.objectid < shrink_last_byte) {
3020 cur_byte = found_key.objectid;
3021 key.objectid = cur_byte;
3022 }
3023 }
3024 btrfs_release_path(root, path);
3025
3026 while(1) {
3027 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3028 if (ret < 0)
3029 goto out;
3030
3031 leaf = path->nodes[0];
3032 nritems = btrfs_header_nritems(leaf);
3033next:
3034 if (path->slots[0] >= nritems) {
3035 ret = btrfs_next_leaf(root, path);
3036 if (ret < 0)
3037 goto out;
3038 if (ret == 1) {
3039 ret = 0;
3040 break;
3041 }
3042 leaf = path->nodes[0];
3043 nritems = btrfs_header_nritems(leaf);
3044 }
3045
3046 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3047
3048 if (found_key.objectid >= shrink_last_byte)
3049 break;
3050
3051 if (progress && need_resched()) {
3052 memcpy(&key, &found_key, sizeof(key));
3053 cond_resched();
3054 btrfs_release_path(root, path);
3055 btrfs_search_slot(NULL, root, &key, path, 0, 0);
3056 progress = 0;
3057 goto next;
3058 }
3059 progress = 1;
3060
3061 if (btrfs_key_type(&found_key) != BTRFS_EXTENT_ITEM_KEY ||
3062 found_key.objectid + found_key.offset <= cur_byte) {
3063 memcpy(&key, &found_key, sizeof(key));
3064 key.offset++;
3065 path->slots[0]++;
3066 goto next;
3067 }
3068
3069 total_found++;
3070 cur_byte = found_key.objectid + found_key.offset;
3071 key.objectid = cur_byte;
3072 btrfs_release_path(root, path);
3073 ret = relocate_one_extent(root, path, &found_key);
3074 __alloc_chunk_for_shrink(root, shrink_block_group, 0);
3075 }
3076
3077 btrfs_release_path(root, path);
3078
3079 if (total_found > 0) {
3080 printk("btrfs relocate found %llu last extent was %llu\n",
3081 (unsigned long long)total_found,
3082 (unsigned long long)found_key.objectid);
3083 trans = btrfs_start_transaction(tree_root, 1);
3084 btrfs_commit_transaction(trans, tree_root);
3085
3086 btrfs_clean_old_snapshots(tree_root);
3087
3088 trans = btrfs_start_transaction(tree_root, 1);
3089 btrfs_commit_transaction(trans, tree_root);
3090 goto again;
3091 }
3092
3093 /*
3094 * we've freed all the extents, now remove the block
3095 * group item from the tree
3096 */
3097 trans = btrfs_start_transaction(root, 1);
3098 memcpy(&key, &shrink_block_group->key, sizeof(key));
3099
3100 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3101 if (ret > 0)
3102 ret = -EIO;
3103 if (ret < 0)
3104 goto out;
3105
3106 clear_extent_bits(&info->block_group_cache, key.objectid,
3107 key.objectid + key.offset - 1,
3108 (unsigned int)-1, GFP_NOFS);
3109
3110
3111 clear_extent_bits(&info->free_space_cache,
3112 key.objectid, key.objectid + key.offset - 1,
3113 (unsigned int)-1, GFP_NOFS);
3114
3115 memset(shrink_block_group, 0, sizeof(*shrink_block_group));
3116 kfree(shrink_block_group);
3117
3118 btrfs_del_item(trans, root, path);
3119 btrfs_commit_transaction(trans, root);
3120
3121 /* the code to unpin extents might set a few bits in the free
3122 * space cache for this range again
3123 */
3124 clear_extent_bits(&info->free_space_cache,
3125 key.objectid, key.objectid + key.offset - 1,
3126 (unsigned int)-1, GFP_NOFS);
3127out:
3128 btrfs_free_path(path);
3129 mutex_unlock(&root->fs_info->alloc_mutex);
3130 return ret;
3131}
3132
3133int find_first_block_group(struct btrfs_root *root, struct btrfs_path *path,
3134 struct btrfs_key *key)
3135{
3136 int ret = 0;
3137 struct btrfs_key found_key;
3138 struct extent_buffer *leaf;
3139 int slot;
3140
3141 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
3142 if (ret < 0)
3143 goto out;
3144
3145 while(1) {
3146 slot = path->slots[0];
3147 leaf = path->nodes[0];
3148 if (slot >= btrfs_header_nritems(leaf)) {
3149 ret = btrfs_next_leaf(root, path);
3150 if (ret == 0)
3151 continue;
3152 if (ret < 0)
3153 goto out;
3154 break;
3155 }
3156 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3157
3158 if (found_key.objectid >= key->objectid &&
3159 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
3160 ret = 0;
3161 goto out;
3162 }
3163 path->slots[0]++;
3164 }
3165 ret = -ENOENT;
3166out:
3167 return ret;
3168}
3169
3170int btrfs_read_block_groups(struct btrfs_root *root)
3171{
3172 struct btrfs_path *path;
3173 int ret;
3174 int bit;
3175 struct btrfs_block_group_cache *cache;
3176 struct btrfs_fs_info *info = root->fs_info;
3177 struct btrfs_space_info *space_info;
3178 struct extent_io_tree *block_group_cache;
3179 struct btrfs_key key;
3180 struct btrfs_key found_key;
3181 struct extent_buffer *leaf;
3182
3183 block_group_cache = &info->block_group_cache;
3184 root = info->extent_root;
3185 key.objectid = 0;
3186 key.offset = 0;
3187 btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
3188 path = btrfs_alloc_path();
3189 if (!path)
3190 return -ENOMEM;
3191
3192 mutex_lock(&root->fs_info->alloc_mutex);
3193 while(1) {
3194 ret = find_first_block_group(root, path, &key);
3195 if (ret > 0) {
3196 ret = 0;
3197 goto error;
3198 }
3199 if (ret != 0)
3200 goto error;
3201
3202 leaf = path->nodes[0];
3203 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3204 cache = kzalloc(sizeof(*cache), GFP_NOFS);
3205 if (!cache) {
3206 ret = -ENOMEM;
3207 break;
3208 }
3209
3210 read_extent_buffer(leaf, &cache->item,
3211 btrfs_item_ptr_offset(leaf, path->slots[0]),
3212 sizeof(cache->item));
3213 memcpy(&cache->key, &found_key, sizeof(found_key));
3214
3215 key.objectid = found_key.objectid + found_key.offset;
3216 btrfs_release_path(root, path);
3217 cache->flags = btrfs_block_group_flags(&cache->item);
3218 bit = 0;
3219 if (cache->flags & BTRFS_BLOCK_GROUP_DATA) {
3220 bit = BLOCK_GROUP_DATA;
3221 } else if (cache->flags & BTRFS_BLOCK_GROUP_SYSTEM) {
3222 bit = BLOCK_GROUP_SYSTEM;
3223 } else if (cache->flags & BTRFS_BLOCK_GROUP_METADATA) {
3224 bit = BLOCK_GROUP_METADATA;
3225 }
3226 set_avail_alloc_bits(info, cache->flags);
3227
3228 ret = update_space_info(info, cache->flags, found_key.offset,
3229 btrfs_block_group_used(&cache->item),
3230 &space_info);
3231 BUG_ON(ret);
3232 cache->space_info = space_info;
3233
3234 /* use EXTENT_LOCKED to prevent merging */
3235 set_extent_bits(block_group_cache, found_key.objectid,
3236 found_key.objectid + found_key.offset - 1,
3237 bit | EXTENT_LOCKED, GFP_NOFS);
3238 set_state_private(block_group_cache, found_key.objectid,
3239 (unsigned long)cache);
3240
3241 if (key.objectid >=
3242 btrfs_super_total_bytes(&info->super_copy))
3243 break;
3244 }
3245 ret = 0;
3246error:
3247 btrfs_free_path(path);
3248 mutex_unlock(&root->fs_info->alloc_mutex);
3249 return ret;
3250}
3251
3252int btrfs_make_block_group(struct btrfs_trans_handle *trans,
3253 struct btrfs_root *root, u64 bytes_used,
3254 u64 type, u64 chunk_objectid, u64 chunk_offset,
3255 u64 size)
3256{
3257 int ret;
3258 int bit = 0;
3259 struct btrfs_root *extent_root;
3260 struct btrfs_block_group_cache *cache;
3261 struct extent_io_tree *block_group_cache;
3262
3263 extent_root = root->fs_info->extent_root;
3264 block_group_cache = &root->fs_info->block_group_cache;
3265
3266 cache = kzalloc(sizeof(*cache), GFP_NOFS);
3267 BUG_ON(!cache);
3268 cache->key.objectid = chunk_offset;
3269 cache->key.offset = size;
3270 btrfs_set_key_type(&cache->key, BTRFS_BLOCK_GROUP_ITEM_KEY);
3271
3272 btrfs_set_block_group_used(&cache->item, bytes_used);
3273 btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
3274 cache->flags = type;
3275 btrfs_set_block_group_flags(&cache->item, type);
3276
3277 ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
3278 &cache->space_info);
3279 BUG_ON(ret);
3280
3281 bit = block_group_state_bits(type);
3282 set_extent_bits(block_group_cache, chunk_offset,
3283 chunk_offset + size - 1,
3284 bit | EXTENT_LOCKED, GFP_NOFS);
3285
3286 set_state_private(block_group_cache, chunk_offset,
3287 (unsigned long)cache);
3288 ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
3289 sizeof(cache->item));
3290 BUG_ON(ret);
3291
3292 finish_current_insert(trans, extent_root);
3293 ret = del_pending_extents(trans, extent_root);
3294 BUG_ON(ret);
3295 set_avail_alloc_bits(extent_root->fs_info, type);
3296
3297 return 0;
3298}
Linux kernel - Deliverables
This page took 0.038257 seconds and 5 git commands to generate.