projects / deliverable / linux.git / blob
? search:


62d49705d14047890996a965123fd630ef53735c
[deliverable/linux.git] / fs / btrfs / extent-tree.c
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 <linux/blkdev.h>
22 #include "hash.h"
23 #include "crc32c.h"
24 #include "ctree.h"
25 #include "disk-io.h"
26 #include "print-tree.h"
27 #include "transaction.h"
28 #include "volumes.h"
29 #include "locking.h"
30 #include "ref-cache.h"
31 #include "compat.h"
32
33 #define PENDING_EXTENT_INSERT 0
34 #define PENDING_EXTENT_DELETE 1
35 #define PENDING_BACKREF_UPDATE 2
36
37 struct pending_extent_op {
38 int type;
39 u64 bytenr;
40 u64 num_bytes;
41 u64 parent;
42 u64 orig_parent;
43 u64 generation;
44 u64 orig_generation;
45 int level;
46 struct list_head list;
47 int del;
48 };
49
50 static int finish_current_insert(struct btrfs_trans_handle *trans, struct
51 btrfs_root *extent_root, int all);
52 static int del_pending_extents(struct btrfs_trans_handle *trans, struct
53 btrfs_root *extent_root, int all);
54 static struct btrfs_block_group_cache *
55 __btrfs_find_block_group(struct btrfs_root *root,
56 struct btrfs_block_group_cache *hint,
57 u64 search_start, int data, int owner);
58 static int pin_down_bytes(struct btrfs_trans_handle *trans,
59 struct btrfs_root *root,
60 u64 bytenr, u64 num_bytes, int is_data);
61 static int update_block_group(struct btrfs_trans_handle *trans,
62 struct btrfs_root *root,
63 u64 bytenr, u64 num_bytes, int alloc,
64 int mark_free);
65
66 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
67 {
68 return (cache->flags & bits) == bits;
69 }
70
71 /*
72 * this adds the block group to the fs_info rb tree for the block group
73 * cache
74 */
75 int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
76 struct btrfs_block_group_cache *block_group)
77 {
78 struct rb_node **p;
79 struct rb_node *parent = NULL;
80 struct btrfs_block_group_cache *cache;
81
82 spin_lock(&info->block_group_cache_lock);
83 p = &info->block_group_cache_tree.rb_node;
84
85 while (*p) {
86 parent = *p;
87 cache = rb_entry(parent, struct btrfs_block_group_cache,
88 cache_node);
89 if (block_group->key.objectid < cache->key.objectid) {
90 p = &(*p)->rb_left;
91 } else if (block_group->key.objectid > cache->key.objectid) {
92 p = &(*p)->rb_right;
93 } else {
94 spin_unlock(&info->block_group_cache_lock);
95 return -EEXIST;
96 }
97 }
98
99 rb_link_node(&block_group->cache_node, parent, p);
100 rb_insert_color(&block_group->cache_node,
101 &info->block_group_cache_tree);
102 spin_unlock(&info->block_group_cache_lock);
103
104 return 0;
105 }
106
107 /*
108 * This will return the block group at or after bytenr if contains is 0, else
109 * it will return the block group that contains the bytenr
110 */
111 static struct btrfs_block_group_cache *
112 block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
113 int contains)
114 {
115 struct btrfs_block_group_cache *cache, *ret = NULL;
116 struct rb_node *n;
117 u64 end, start;
118
119 spin_lock(&info->block_group_cache_lock);
120 n = info->block_group_cache_tree.rb_node;
121
122 while (n) {
123 cache = rb_entry(n, struct btrfs_block_group_cache,
124 cache_node);
125 end = cache->key.objectid + cache->key.offset - 1;
126 start = cache->key.objectid;
127
128 if (bytenr < start) {
129 if (!contains && (!ret || start < ret->key.objectid))
130 ret = cache;
131 n = n->rb_left;
132 } else if (bytenr > start) {
133 if (contains && bytenr <= end) {
134 ret = cache;
135 break;
136 }
137 n = n->rb_right;
138 } else {
139 ret = cache;
140 break;
141 }
142 }
143 spin_unlock(&info->block_group_cache_lock);
144
145 return ret;
146 }
147
148 /*
149 * this is only called by cache_block_group, since we could have freed extents
150 * we need to check the pinned_extents for any extents that can't be used yet
151 * since their free space will be released as soon as the transaction commits.
152 */
153 static int add_new_free_space(struct btrfs_block_group_cache *block_group,
154 struct btrfs_fs_info *info, u64 start, u64 end)
155 {
156 u64 extent_start, extent_end, size;
157 int ret;
158
159 mutex_lock(&info->pinned_mutex);
160 while (start < end) {
161 ret = find_first_extent_bit(&info->pinned_extents, start,
162 &extent_start, &extent_end,
163 EXTENT_DIRTY);
164 if (ret)
165 break;
166
167 if (extent_start == start) {
168 start = extent_end + 1;
169 } else if (extent_start > start && extent_start < end) {
170 size = extent_start - start;
171 ret = btrfs_add_free_space_lock(block_group, start,
172 size);
173 BUG_ON(ret);
174 start = extent_end + 1;
175 } else {
176 break;
177 }
178 }
179
180 if (start < end) {
181 size = end - start;
182 ret = btrfs_add_free_space_lock(block_group, start, size);
183 BUG_ON(ret);
184 }
185 mutex_unlock(&info->pinned_mutex);
186
187 return 0;
188 }
189
190 static int cache_block_group(struct btrfs_root *root,
191 struct btrfs_block_group_cache *block_group)
192 {
193 struct btrfs_path *path;
194 int ret = 0;
195 struct btrfs_key key;
196 struct extent_buffer *leaf;
197 int slot;
198 u64 last = 0;
199 u64 first_free;
200 int found = 0;
201
202 if (!block_group)
203 return 0;
204
205 root = root->fs_info->extent_root;
206
207 if (block_group->cached)
208 return 0;
209
210 path = btrfs_alloc_path();
211 if (!path)
212 return -ENOMEM;
213
214 path->reada = 2;
215 /*
216 * we get into deadlocks with paths held by callers of this function.
217 * since the alloc_mutex is protecting things right now, just
218 * skip the locking here
219 */
220 path->skip_locking = 1;
221 first_free = max_t(u64, block_group->key.objectid,
222 BTRFS_SUPER_INFO_OFFSET + BTRFS_SUPER_INFO_SIZE);
223 key.objectid = block_group->key.objectid;
224 key.offset = 0;
225 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
226 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
227 if (ret < 0)
228 goto err;
229 ret = btrfs_previous_item(root, path, 0, BTRFS_EXTENT_ITEM_KEY);
230 if (ret < 0)
231 goto err;
232 if (ret == 0) {
233 leaf = path->nodes[0];
234 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
235 if (key.objectid + key.offset > first_free)
236 first_free = key.objectid + key.offset;
237 }
238 while(1) {
239 leaf = path->nodes[0];
240 slot = path->slots[0];
241 if (slot >= btrfs_header_nritems(leaf)) {
242 ret = btrfs_next_leaf(root, path);
243 if (ret < 0)
244 goto err;
245 if (ret == 0)
246 continue;
247 else
248 break;
249 }
250 btrfs_item_key_to_cpu(leaf, &key, slot);
251 if (key.objectid < block_group->key.objectid)
252 goto next;
253
254 if (key.objectid >= block_group->key.objectid +
255 block_group->key.offset)
256 break;
257
258 if (btrfs_key_type(&key) == BTRFS_EXTENT_ITEM_KEY) {
259 if (!found) {
260 last = first_free;
261 found = 1;
262 }
263
264 add_new_free_space(block_group, root->fs_info, last,
265 key.objectid);
266
267 last = key.objectid + key.offset;
268 }
269 next:
270 path->slots[0]++;
271 }
272
273 if (!found)
274 last = first_free;
275
276 add_new_free_space(block_group, root->fs_info, last,
277 block_group->key.objectid +
278 block_group->key.offset);
279
280 block_group->cached = 1;
281 ret = 0;
282 err:
283 btrfs_free_path(path);
284 return ret;
285 }
286
287 /*
288 * return the block group that starts at or after bytenr
289 */
290 struct btrfs_block_group_cache *btrfs_lookup_first_block_group(struct
291 btrfs_fs_info *info,
292 u64 bytenr)
293 {
294 struct btrfs_block_group_cache *cache;
295
296 cache = block_group_cache_tree_search(info, bytenr, 0);
297
298 return cache;
299 }
300
301 /*
302 * return the block group that contains teh given bytenr
303 */
304 struct btrfs_block_group_cache *btrfs_lookup_block_group(struct
305 btrfs_fs_info *info,
306 u64 bytenr)
307 {
308 struct btrfs_block_group_cache *cache;
309
310 cache = block_group_cache_tree_search(info, bytenr, 1);
311
312 return cache;
313 }
314
315 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
316 u64 flags)
317 {
318 struct list_head *head = &info->space_info;
319 struct list_head *cur;
320 struct btrfs_space_info *found;
321 list_for_each(cur, head) {
322 found = list_entry(cur, struct btrfs_space_info, list);
323 if (found->flags == flags)
324 return found;
325 }
326 return NULL;
327 }
328
329 static u64 div_factor(u64 num, int factor)
330 {
331 if (factor == 10)
332 return num;
333 num *= factor;
334 do_div(num, 10);
335 return num;
336 }
337
338 static struct btrfs_block_group_cache *
339 __btrfs_find_block_group(struct btrfs_root *root,
340 struct btrfs_block_group_cache *hint,
341 u64 search_start, int data, int owner)
342 {
343 struct btrfs_block_group_cache *cache;
344 struct btrfs_block_group_cache *found_group = NULL;
345 struct btrfs_fs_info *info = root->fs_info;
346 u64 used;
347 u64 last = 0;
348 u64 free_check;
349 int full_search = 0;
350 int factor = 10;
351 int wrapped = 0;
352
353 if (data & BTRFS_BLOCK_GROUP_METADATA)
354 factor = 9;
355
356 if (search_start) {
357 struct btrfs_block_group_cache *shint;
358 shint = btrfs_lookup_first_block_group(info, search_start);
359 if (shint && block_group_bits(shint, data)) {
360 spin_lock(&shint->lock);
361 used = btrfs_block_group_used(&shint->item);
362 if (used + shint->pinned + shint->reserved <
363 div_factor(shint->key.offset, factor)) {
364 spin_unlock(&shint->lock);
365 return shint;
366 }
367 spin_unlock(&shint->lock);
368 }
369 }
370 if (hint && block_group_bits(hint, data)) {
371 spin_lock(&hint->lock);
372 used = btrfs_block_group_used(&hint->item);
373 if (used + hint->pinned + hint->reserved <
374 div_factor(hint->key.offset, factor)) {
375 spin_unlock(&hint->lock);
376 return hint;
377 }
378 spin_unlock(&hint->lock);
379 last = hint->key.objectid + hint->key.offset;
380 } else {
381 if (hint)
382 last = max(hint->key.objectid, search_start);
383 else
384 last = search_start;
385 }
386 again:
387 while (1) {
388 cache = btrfs_lookup_first_block_group(root->fs_info, last);
389 if (!cache)
390 break;
391
392 spin_lock(&cache->lock);
393 last = cache->key.objectid + cache->key.offset;
394 used = btrfs_block_group_used(&cache->item);
395
396 if (block_group_bits(cache, data)) {
397 free_check = div_factor(cache->key.offset, factor);
398 if (used + cache->pinned + cache->reserved <
399 free_check) {
400 found_group = cache;
401 spin_unlock(&cache->lock);
402 goto found;
403 }
404 }
405 spin_unlock(&cache->lock);
406 cond_resched();
407 }
408 if (!wrapped) {
409 last = search_start;
410 wrapped = 1;
411 goto again;
412 }
413 if (!full_search && factor < 10) {
414 last = search_start;
415 full_search = 1;
416 factor = 10;
417 goto again;
418 }
419 found:
420 return found_group;
421 }
422
423 struct btrfs_block_group_cache *btrfs_find_block_group(struct btrfs_root *root,
424 struct btrfs_block_group_cache
425 *hint, u64 search_start,
426 int data, int owner)
427 {
428
429 struct btrfs_block_group_cache *ret;
430 ret = __btrfs_find_block_group(root, hint, search_start, data, owner);
431 return ret;
432 }
433
434 /* simple helper to search for an existing extent at a given offset */
435 int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len)
436 {
437 int ret;
438 struct btrfs_key key;
439 struct btrfs_path *path;
440
441 path = btrfs_alloc_path();
442 BUG_ON(!path);
443 key.objectid = start;
444 key.offset = len;
445 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
446 ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
447 0, 0);
448 btrfs_free_path(path);
449 return ret;
450 }
451
452 /*
453 * Back reference rules. Back refs have three main goals:
454 *
455 * 1) differentiate between all holders of references to an extent so that
456 * when a reference is dropped we can make sure it was a valid reference
457 * before freeing the extent.
458 *
459 * 2) Provide enough information to quickly find the holders of an extent
460 * if we notice a given block is corrupted or bad.
461 *
462 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
463 * maintenance. This is actually the same as #2, but with a slightly
464 * different use case.
465 *
466 * File extents can be referenced by:
467 *
468 * - multiple snapshots, subvolumes, or different generations in one subvol
469 * - different files inside a single subvolume
470 * - different offsets inside a file (bookend extents in file.c)
471 *
472 * The extent ref structure has fields for:
473 *
474 * - Objectid of the subvolume root
475 * - Generation number of the tree holding the reference
476 * - objectid of the file holding the reference
477 * - number of references holding by parent node (alway 1 for tree blocks)
478 *
479 * Btree leaf may hold multiple references to a file extent. In most cases,
480 * these references are from same file and the corresponding offsets inside
481 * the file are close together.
482 *
483 * When a file extent is allocated the fields are filled in:
484 * (root_key.objectid, trans->transid, inode objectid, 1)
485 *
486 * When a leaf is cow'd new references are added for every file extent found
487 * in the leaf. It looks similar to the create case, but trans->transid will
488 * be different when the block is cow'd.
489 *
490 * (root_key.objectid, trans->transid, inode objectid,
491 * number of references in the leaf)
492 *
493 * When a file extent is removed either during snapshot deletion or
494 * file truncation, we find the corresponding back reference and check
495 * the following fields:
496 *
497 * (btrfs_header_owner(leaf), btrfs_header_generation(leaf),
498 * inode objectid)
499 *
500 * Btree extents can be referenced by:
501 *
502 * - Different subvolumes
503 * - Different generations of the same subvolume
504 *
505 * When a tree block is created, back references are inserted:
506 *
507 * (root->root_key.objectid, trans->transid, level, 1)
508 *
509 * When a tree block is cow'd, new back references are added for all the
510 * blocks it points to. If the tree block isn't in reference counted root,
511 * the old back references are removed. These new back references are of
512 * the form (trans->transid will have increased since creation):
513 *
514 * (root->root_key.objectid, trans->transid, level, 1)
515 *
516 * When a backref is in deleting, the following fields are checked:
517 *
518 * if backref was for a tree root:
519 * (btrfs_header_owner(itself), btrfs_header_generation(itself), level)
520 * else
521 * (btrfs_header_owner(parent), btrfs_header_generation(parent), level)
522 *
523 * Back Reference Key composing:
524 *
525 * The key objectid corresponds to the first byte in the extent, the key
526 * type is set to BTRFS_EXTENT_REF_KEY, and the key offset is the first
527 * byte of parent extent. If a extent is tree root, the key offset is set
528 * to the key objectid.
529 */
530
531 static int noinline lookup_extent_backref(struct btrfs_trans_handle *trans,
532 struct btrfs_root *root,
533 struct btrfs_path *path,
534 u64 bytenr, u64 parent,
535 u64 ref_root, u64 ref_generation,
536 u64 owner_objectid, int del)
537 {
538 struct btrfs_key key;
539 struct btrfs_extent_ref *ref;
540 struct extent_buffer *leaf;
541 u64 ref_objectid;
542 int ret;
543
544 key.objectid = bytenr;
545 key.type = BTRFS_EXTENT_REF_KEY;
546 key.offset = parent;
547
548 ret = btrfs_search_slot(trans, root, &key, path, del ? -1 : 0, 1);
549 if (ret < 0)
550 goto out;
551 if (ret > 0) {
552 ret = -ENOENT;
553 goto out;
554 }
555
556 leaf = path->nodes[0];
557 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_ref);
558 ref_objectid = btrfs_ref_objectid(leaf, ref);
559 if (btrfs_ref_root(leaf, ref) != ref_root ||
560 btrfs_ref_generation(leaf, ref) != ref_generation ||
561 (ref_objectid != owner_objectid &&
562 ref_objectid != BTRFS_MULTIPLE_OBJECTIDS)) {
563 ret = -EIO;
564 WARN_ON(1);
565 goto out;
566 }
567 ret = 0;
568 out:
569 return ret;
570 }
571
572 /*
573 * updates all the backrefs that are pending on update_list for the
574 * extent_root
575 */
576 static int noinline update_backrefs(struct btrfs_trans_handle *trans,
577 struct btrfs_root *extent_root,
578 struct btrfs_path *path,
579 struct list_head *update_list)
580 {
581 struct btrfs_key key;
582 struct btrfs_extent_ref *ref;
583 struct btrfs_fs_info *info = extent_root->fs_info;
584 struct pending_extent_op *op;
585 struct extent_buffer *leaf;
586 int ret = 0;
587 struct list_head *cur = update_list->next;
588 u64 ref_objectid;
589 u64 ref_root = extent_root->root_key.objectid;
590
591 op = list_entry(cur, struct pending_extent_op, list);
592
593 search:
594 key.objectid = op->bytenr;
595 key.type = BTRFS_EXTENT_REF_KEY;
596 key.offset = op->orig_parent;
597
598 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 1);
599 BUG_ON(ret);
600
601 leaf = path->nodes[0];
602
603 loop:
604 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_ref);
605
606 ref_objectid = btrfs_ref_objectid(leaf, ref);
607
608 if (btrfs_ref_root(leaf, ref) != ref_root ||
609 btrfs_ref_generation(leaf, ref) != op->orig_generation ||
610 (ref_objectid != op->level &&
611 ref_objectid != BTRFS_MULTIPLE_OBJECTIDS)) {
612 printk(KERN_ERR "couldn't find %Lu, parent %Lu, root %Lu, "
613 "owner %u\n", op->bytenr, op->orig_parent,
614 ref_root, op->level);
615 btrfs_print_leaf(extent_root, leaf);
616 BUG();
617 }
618
619 key.objectid = op->bytenr;
620 key.offset = op->parent;
621 key.type = BTRFS_EXTENT_REF_KEY;
622 ret = btrfs_set_item_key_safe(trans, extent_root, path, &key);
623 BUG_ON(ret);
624 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_ref);
625 btrfs_set_ref_generation(leaf, ref, op->generation);
626
627 cur = cur->next;
628
629 list_del_init(&op->list);
630 unlock_extent(&info->extent_ins, op->bytenr,
631 op->bytenr + op->num_bytes - 1, GFP_NOFS);
632 kfree(op);
633
634 if (cur == update_list) {
635 btrfs_mark_buffer_dirty(path->nodes[0]);
636 btrfs_release_path(extent_root, path);
637 goto out;
638 }
639
640 op = list_entry(cur, struct pending_extent_op, list);
641
642 path->slots[0]++;
643 while (path->slots[0] < btrfs_header_nritems(leaf)) {
644 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
645 if (key.objectid == op->bytenr &&
646 key.type == BTRFS_EXTENT_REF_KEY)
647 goto loop;
648 path->slots[0]++;
649 }
650
651 btrfs_mark_buffer_dirty(path->nodes[0]);
652 btrfs_release_path(extent_root, path);
653 goto search;
654
655 out:
656 return 0;
657 }
658
659 static int noinline insert_extents(struct btrfs_trans_handle *trans,
660 struct btrfs_root *extent_root,
661 struct btrfs_path *path,
662 struct list_head *insert_list, int nr)
663 {
664 struct btrfs_key *keys;
665 u32 *data_size;
666 struct pending_extent_op *op;
667 struct extent_buffer *leaf;
668 struct list_head *cur = insert_list->next;
669 struct btrfs_fs_info *info = extent_root->fs_info;
670 u64 ref_root = extent_root->root_key.objectid;
671 int i = 0, last = 0, ret;
672 int total = nr * 2;
673
674 if (!nr)
675 return 0;
676
677 keys = kzalloc(total * sizeof(struct btrfs_key), GFP_NOFS);
678 if (!keys)
679 return -ENOMEM;
680
681 data_size = kzalloc(total * sizeof(u32), GFP_NOFS);
682 if (!data_size) {
683 kfree(keys);
684 return -ENOMEM;
685 }
686
687 list_for_each_entry(op, insert_list, list) {
688 keys[i].objectid = op->bytenr;
689 keys[i].offset = op->num_bytes;
690 keys[i].type = BTRFS_EXTENT_ITEM_KEY;
691 data_size[i] = sizeof(struct btrfs_extent_item);
692 i++;
693
694 keys[i].objectid = op->bytenr;
695 keys[i].offset = op->parent;
696 keys[i].type = BTRFS_EXTENT_REF_KEY;
697 data_size[i] = sizeof(struct btrfs_extent_ref);
698 i++;
699 }
700
701 op = list_entry(cur, struct pending_extent_op, list);
702 i = 0;
703 while (i < total) {
704 int c;
705 ret = btrfs_insert_some_items(trans, extent_root, path,
706 keys+i, data_size+i, total-i);
707 BUG_ON(ret < 0);
708
709 if (last && ret > 1)
710 BUG();
711
712 leaf = path->nodes[0];
713 for (c = 0; c < ret; c++) {
714 int ref_first = keys[i].type == BTRFS_EXTENT_REF_KEY;
715
716 /*
717 * if the first item we inserted was a backref, then
718 * the EXTENT_ITEM will be the odd c's, else it will
719 * be the even c's
720 */
721 if ((ref_first && (c % 2)) ||
722 (!ref_first && !(c % 2))) {
723 struct btrfs_extent_item *itm;
724
725 itm = btrfs_item_ptr(leaf, path->slots[0] + c,
726 struct btrfs_extent_item);
727 btrfs_set_extent_refs(path->nodes[0], itm, 1);
728 op->del++;
729 } else {
730 struct btrfs_extent_ref *ref;
731
732 ref = btrfs_item_ptr(leaf, path->slots[0] + c,
733 struct btrfs_extent_ref);
734 btrfs_set_ref_root(leaf, ref, ref_root);
735 btrfs_set_ref_generation(leaf, ref,
736 op->generation);
737 btrfs_set_ref_objectid(leaf, ref, op->level);
738 btrfs_set_ref_num_refs(leaf, ref, 1);
739 op->del++;
740 }
741
742 /*
743 * using del to see when its ok to free up the
744 * pending_extent_op. In the case where we insert the
745 * last item on the list in order to help do batching
746 * we need to not free the extent op until we actually
747 * insert the extent_item
748 */
749 if (op->del == 2) {
750 unlock_extent(&info->extent_ins, op->bytenr,
751 op->bytenr + op->num_bytes - 1,
752 GFP_NOFS);
753 cur = cur->next;
754 list_del_init(&op->list);
755 kfree(op);
756 if (cur != insert_list)
757 op = list_entry(cur,
758 struct pending_extent_op,
759 list);
760 }
761 }
762 btrfs_mark_buffer_dirty(leaf);
763 btrfs_release_path(extent_root, path);
764
765 /*
766 * Ok backref's and items usually go right next to eachother,
767 * but if we could only insert 1 item that means that we
768 * inserted on the end of a leaf, and we have no idea what may
769 * be on the next leaf so we just play it safe. In order to
770 * try and help this case we insert the last thing on our
771 * insert list so hopefully it will end up being the last
772 * thing on the leaf and everything else will be before it,
773 * which will let us insert a whole bunch of items at the same
774 * time.
775 */
776 if (ret == 1 && !last && (i + ret < total)) {
777 /*
778 * last: where we will pick up the next time around
779 * i: our current key to insert, will be total - 1
780 * cur: the current op we are screwing with
781 * op: duh
782 */
783 last = i + ret;
784 i = total - 1;
785 cur = insert_list->prev;
786 op = list_entry(cur, struct pending_extent_op, list);
787 } else if (last) {
788 /*
789 * ok we successfully inserted the last item on the
790 * list, lets reset everything
791 *
792 * i: our current key to insert, so where we left off
793 * last time
794 * last: done with this
795 * cur: the op we are messing with
796 * op: duh
797 * total: since we inserted the last key, we need to
798 * decrement total so we dont overflow
799 */
800 i = last;
801 last = 0;
802 total--;
803 if (i < total) {
804 cur = insert_list->next;
805 op = list_entry(cur, struct pending_extent_op,
806 list);
807 }
808 } else {
809 i += ret;
810 }
811
812 cond_resched();
813 }
814 ret = 0;
815 kfree(keys);
816 kfree(data_size);
817 return ret;
818 }
819
820 static int noinline insert_extent_backref(struct btrfs_trans_handle *trans,
821 struct btrfs_root *root,
822 struct btrfs_path *path,
823 u64 bytenr, u64 parent,
824 u64 ref_root, u64 ref_generation,
825 u64 owner_objectid)
826 {
827 struct btrfs_key key;
828 struct extent_buffer *leaf;
829 struct btrfs_extent_ref *ref;
830 u32 num_refs;
831 int ret;
832
833 key.objectid = bytenr;
834 key.type = BTRFS_EXTENT_REF_KEY;
835 key.offset = parent;
836
837 ret = btrfs_insert_empty_item(trans, root, path, &key, sizeof(*ref));
838 if (ret == 0) {
839 leaf = path->nodes[0];
840 ref = btrfs_item_ptr(leaf, path->slots[0],
841 struct btrfs_extent_ref);
842 btrfs_set_ref_root(leaf, ref, ref_root);
843 btrfs_set_ref_generation(leaf, ref, ref_generation);
844 btrfs_set_ref_objectid(leaf, ref, owner_objectid);
845 btrfs_set_ref_num_refs(leaf, ref, 1);
846 } else if (ret == -EEXIST) {
847 u64 existing_owner;
848 BUG_ON(owner_objectid < BTRFS_FIRST_FREE_OBJECTID);
849 leaf = path->nodes[0];
850 ref = btrfs_item_ptr(leaf, path->slots[0],
851 struct btrfs_extent_ref);
852 if (btrfs_ref_root(leaf, ref) != ref_root ||
853 btrfs_ref_generation(leaf, ref) != ref_generation) {
854 ret = -EIO;
855 WARN_ON(1);
856 goto out;
857 }
858
859 num_refs = btrfs_ref_num_refs(leaf, ref);
860 BUG_ON(num_refs == 0);
861 btrfs_set_ref_num_refs(leaf, ref, num_refs + 1);
862
863 existing_owner = btrfs_ref_objectid(leaf, ref);
864 if (existing_owner != owner_objectid &&
865 existing_owner != BTRFS_MULTIPLE_OBJECTIDS) {
866 btrfs_set_ref_objectid(leaf, ref,
867 BTRFS_MULTIPLE_OBJECTIDS);
868 }
869 ret = 0;
870 } else {
871 goto out;
872 }
873 btrfs_mark_buffer_dirty(path->nodes[0]);
874 out:
875 btrfs_release_path(root, path);
876 return ret;
877 }
878
879 static int noinline remove_extent_backref(struct btrfs_trans_handle *trans,
880 struct btrfs_root *root,
881 struct btrfs_path *path)
882 {
883 struct extent_buffer *leaf;
884 struct btrfs_extent_ref *ref;
885 u32 num_refs;
886 int ret = 0;
887
888 leaf = path->nodes[0];
889 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_ref);
890 num_refs = btrfs_ref_num_refs(leaf, ref);
891 BUG_ON(num_refs == 0);
892 num_refs -= 1;
893 if (num_refs == 0) {
894 ret = btrfs_del_item(trans, root, path);
895 } else {
896 btrfs_set_ref_num_refs(leaf, ref, num_refs);
897 btrfs_mark_buffer_dirty(leaf);
898 }
899 btrfs_release_path(root, path);
900 return ret;
901 }
902
903 static void btrfs_issue_discard(struct block_device *bdev,
904 u64 start, u64 len)
905 {
906 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,28)
907 blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_KERNEL);
908 #else
909 blkdev_issue_discard(bdev, start >> 9, len >> 9);
910 #endif
911 }
912
913
914 static int noinline free_extents(struct btrfs_trans_handle *trans,
915 struct btrfs_root *extent_root,
916 struct list_head *del_list)
917 {
918 struct btrfs_fs_info *info = extent_root->fs_info;
919 struct btrfs_path *path;
920 struct btrfs_key key, found_key;
921 struct extent_buffer *leaf;
922 struct list_head *cur;
923 struct pending_extent_op *op;
924 struct btrfs_extent_item *ei;
925 int ret, num_to_del, extent_slot = 0, found_extent = 0;
926 u32 refs;
927 u64 bytes_freed = 0;
928
929 path = btrfs_alloc_path();
930 if (!path)
931 return -ENOMEM;
932 path->reada = 1;
933
934 search:
935 /* search for the backref for the current ref we want to delete */
936 cur = del_list->next;
937 op = list_entry(cur, struct pending_extent_op, list);
938 ret = lookup_extent_backref(trans, extent_root, path, op->bytenr,
939 op->orig_parent,
940 extent_root->root_key.objectid,
941 op->orig_generation, op->level, 1);
942 if (ret) {
943 printk("Unable to find backref byte nr %Lu root %Lu gen %Lu "
944 "owner %u\n", op->bytenr,
945 extent_root->root_key.objectid, op->orig_generation,
946 op->level);
947 btrfs_print_leaf(extent_root, path->nodes[0]);
948 WARN_ON(1);
949 goto out;
950 }
951
952 extent_slot = path->slots[0];
953 num_to_del = 1;
954 found_extent = 0;
955
956 /*
957 * if we aren't the first item on the leaf we can move back one and see
958 * if our ref is right next to our extent item
959 */
960 if (likely(extent_slot)) {
961 extent_slot--;
962 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
963 extent_slot);
964 if (found_key.objectid == op->bytenr &&
965 found_key.type == BTRFS_EXTENT_ITEM_KEY &&
966 found_key.offset == op->num_bytes) {
967 num_to_del++;
968 found_extent = 1;
969 }
970 }
971
972 /*
973 * if we didn't find the extent we need to delete the backref and then
974 * search for the extent item key so we can update its ref count
975 */
976 if (!found_extent) {
977 key.objectid = op->bytenr;
978 key.type = BTRFS_EXTENT_ITEM_KEY;
979 key.offset = op->num_bytes;
980
981 ret = remove_extent_backref(trans, extent_root, path);
982 BUG_ON(ret);
983 btrfs_release_path(extent_root, path);
984 ret = btrfs_search_slot(trans, extent_root, &key, path, -1, 1);
985 BUG_ON(ret);
986 extent_slot = path->slots[0];
987 }
988
989 /* this is where we update the ref count for the extent */
990 leaf = path->nodes[0];
991 ei = btrfs_item_ptr(leaf, extent_slot, struct btrfs_extent_item);
992 refs = btrfs_extent_refs(leaf, ei);
993 BUG_ON(refs == 0);
994 refs--;
995 btrfs_set_extent_refs(leaf, ei, refs);
996
997 btrfs_mark_buffer_dirty(leaf);
998
999 /*
1000 * This extent needs deleting. The reason cur_slot is extent_slot +
1001 * num_to_del is because extent_slot points to the slot where the extent
1002 * is, and if the backref was not right next to the extent we will be
1003 * deleting at least 1 item, and will want to start searching at the
1004 * slot directly next to extent_slot. However if we did find the
1005 * backref next to the extent item them we will be deleting at least 2
1006 * items and will want to start searching directly after the ref slot
1007 */
1008 if (!refs) {
1009 struct list_head *pos, *n, *end;
1010 int cur_slot = extent_slot+num_to_del;
1011 u64 super_used;
1012 u64 root_used;
1013
1014 path->slots[0] = extent_slot;
1015 bytes_freed = op->num_bytes;
1016
1017 mutex_lock(&info->pinned_mutex);
1018 ret = pin_down_bytes(trans, extent_root, op->bytenr,
1019 op->num_bytes, op->level >=
1020 BTRFS_FIRST_FREE_OBJECTID);
1021 mutex_unlock(&info->pinned_mutex);
1022 BUG_ON(ret < 0);
1023 op->del = ret;
1024
1025 /*
1026 * we need to see if we can delete multiple things at once, so
1027 * start looping through the list of extents we are wanting to
1028 * delete and see if their extent/backref's are right next to
1029 * eachother and the extents only have 1 ref
1030 */
1031 for (pos = cur->next; pos != del_list; pos = pos->next) {
1032 struct pending_extent_op *tmp;
1033
1034 tmp = list_entry(pos, struct pending_extent_op, list);
1035
1036 /* we only want to delete extent+ref at this stage */
1037 if (cur_slot >= btrfs_header_nritems(leaf) - 1)
1038 break;
1039
1040 btrfs_item_key_to_cpu(leaf, &found_key, cur_slot);
1041 if (found_key.objectid != tmp->bytenr ||
1042 found_key.type != BTRFS_EXTENT_ITEM_KEY ||
1043 found_key.offset != tmp->num_bytes)
1044 break;
1045
1046 /* check to make sure this extent only has one ref */
1047 ei = btrfs_item_ptr(leaf, cur_slot,
1048 struct btrfs_extent_item);
1049 if (btrfs_extent_refs(leaf, ei) != 1)
1050 break;
1051
1052 btrfs_item_key_to_cpu(leaf, &found_key, cur_slot+1);
1053 if (found_key.objectid != tmp->bytenr ||
1054 found_key.type != BTRFS_EXTENT_REF_KEY ||
1055 found_key.offset != tmp->orig_parent)
1056 break;
1057
1058 /*
1059 * the ref is right next to the extent, we can set the
1060 * ref count to 0 since we will delete them both now
1061 */
1062 btrfs_set_extent_refs(leaf, ei, 0);
1063
1064 /* pin down the bytes for this extent */
1065 mutex_lock(&info->pinned_mutex);
1066 ret = pin_down_bytes(trans, extent_root, tmp->bytenr,
1067 tmp->num_bytes, tmp->level >=
1068 BTRFS_FIRST_FREE_OBJECTID);
1069 mutex_unlock(&info->pinned_mutex);
1070 BUG_ON(ret < 0);
1071
1072 /*
1073 * use the del field to tell if we need to go ahead and
1074 * free up the extent when we delete the item or not.
1075 */
1076 tmp->del = ret;
1077 bytes_freed += tmp->num_bytes;
1078
1079 num_to_del += 2;
1080 cur_slot += 2;
1081 }
1082 end = pos;
1083
1084 /* update the free space counters */
1085 spin_lock_irq(&info->delalloc_lock);
1086 super_used = btrfs_super_bytes_used(&info->super_copy);
1087 btrfs_set_super_bytes_used(&info->super_copy,
1088 super_used - bytes_freed);
1089 spin_unlock_irq(&info->delalloc_lock);
1090
1091 root_used = btrfs_root_used(&extent_root->root_item);
1092 btrfs_set_root_used(&extent_root->root_item,
1093 root_used - bytes_freed);
1094
1095 /* delete the items */
1096 ret = btrfs_del_items(trans, extent_root, path,
1097 path->slots[0], num_to_del);
1098 BUG_ON(ret);
1099
1100 /*
1101 * loop through the extents we deleted and do the cleanup work
1102 * on them
1103 */
1104 for (pos = cur, n = pos->next; pos != end;
1105 pos = n, n = pos->next) {
1106 struct pending_extent_op *tmp;
1107 #ifdef BIO_RW_DISCARD
1108 u64 map_length;
1109 struct btrfs_multi_bio *multi = NULL;
1110 #endif
1111 tmp = list_entry(pos, struct pending_extent_op, list);
1112
1113 /*
1114 * remember tmp->del tells us wether or not we pinned
1115 * down the extent
1116 */
1117 ret = update_block_group(trans, extent_root,
1118 tmp->bytenr, tmp->num_bytes, 0,
1119 tmp->del);
1120 BUG_ON(ret);
1121
1122 #ifdef BIO_RW_DISCARD
1123 map_length = tmp->num_bytes;
1124 ret = btrfs_map_block(&info->mapping_tree, READ,
1125 tmp->bytenr, &map_length, &multi,
1126 0);
1127 if (!ret) {
1128 struct btrfs_bio_stripe *stripe;
1129 int i;
1130
1131 stripe = multi->stripes;
1132
1133 if (map_length > tmp->num_bytes)
1134 map_length = tmp->num_bytes;
1135
1136 for (i = 0; i < multi->num_stripes;
1137 i++, stripe++)
1138 btrfs_issue_discard(stripe->dev->bdev,
1139 stripe->physical,
1140 map_length);
1141 kfree(multi);
1142 }
1143 #endif
1144 list_del_init(&tmp->list);
1145 unlock_extent(&info->extent_ins, tmp->bytenr,
1146 tmp->bytenr + tmp->num_bytes - 1,
1147 GFP_NOFS);
1148 kfree(tmp);
1149 }
1150 } else if (refs && found_extent) {
1151 /*
1152 * the ref and extent were right next to eachother, but the
1153 * extent still has a ref, so just free the backref and keep
1154 * going
1155 */
1156 ret = remove_extent_backref(trans, extent_root, path);
1157 BUG_ON(ret);
1158
1159 list_del_init(&op->list);
1160 unlock_extent(&info->extent_ins, op->bytenr,
1161 op->bytenr + op->num_bytes - 1, GFP_NOFS);
1162 kfree(op);
1163 } else {
1164 /*
1165 * the extent has multiple refs and the backref we were looking
1166 * for was not right next to it, so just unlock and go next,
1167 * we're good to go
1168 */
1169 list_del_init(&op->list);
1170 unlock_extent(&info->extent_ins, op->bytenr,
1171 op->bytenr + op->num_bytes - 1, GFP_NOFS);
1172 kfree(op);
1173 }
1174
1175 btrfs_release_path(extent_root, path);
1176 if (!list_empty(del_list))
1177 goto search;
1178
1179 out:
1180 btrfs_free_path(path);
1181 return ret;
1182 }
1183
1184 static int __btrfs_update_extent_ref(struct btrfs_trans_handle *trans,
1185 struct btrfs_root *root, u64 bytenr,
1186 u64 orig_parent, u64 parent,
1187 u64 orig_root, u64 ref_root,
1188 u64 orig_generation, u64 ref_generation,
1189 u64 owner_objectid)
1190 {
1191 int ret;
1192 struct btrfs_root *extent_root = root->fs_info->extent_root;
1193 struct btrfs_path *path;
1194
1195 if (root == root->fs_info->extent_root) {
1196 struct pending_extent_op *extent_op;
1197 u64 num_bytes;
1198
1199 BUG_ON(owner_objectid >= BTRFS_MAX_LEVEL);
1200 num_bytes = btrfs_level_size(root, (int)owner_objectid);
1201 mutex_lock(&root->fs_info->extent_ins_mutex);
1202 if (test_range_bit(&root->fs_info->extent_ins, bytenr,
1203 bytenr + num_bytes - 1, EXTENT_WRITEBACK, 0)) {
1204 u64 priv;
1205 ret = get_state_private(&root->fs_info->extent_ins,
1206 bytenr, &priv);
1207 BUG_ON(ret);
1208 extent_op = (struct pending_extent_op *)
1209 (unsigned long)priv;
1210 BUG_ON(extent_op->parent != orig_parent);
1211 BUG_ON(extent_op->generation != orig_generation);
1212
1213 extent_op->parent = parent;
1214 extent_op->generation = ref_generation;
1215 } else {
1216 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
1217 BUG_ON(!extent_op);
1218
1219 extent_op->type = PENDING_BACKREF_UPDATE;
1220 extent_op->bytenr = bytenr;
1221 extent_op->num_bytes = num_bytes;
1222 extent_op->parent = parent;
1223 extent_op->orig_parent = orig_parent;
1224 extent_op->generation = ref_generation;
1225 extent_op->orig_generation = orig_generation;
1226 extent_op->level = (int)owner_objectid;
1227 INIT_LIST_HEAD(&extent_op->list);
1228 extent_op->del = 0;
1229
1230 set_extent_bits(&root->fs_info->extent_ins,
1231 bytenr, bytenr + num_bytes - 1,
1232 EXTENT_WRITEBACK, GFP_NOFS);
1233 set_state_private(&root->fs_info->extent_ins,
1234 bytenr, (unsigned long)extent_op);
1235 }
1236 mutex_unlock(&root->fs_info->extent_ins_mutex);
1237 return 0;
1238 }
1239
1240 path = btrfs_alloc_path();
1241 if (!path)
1242 return -ENOMEM;
1243 ret = lookup_extent_backref(trans, extent_root, path,
1244 bytenr, orig_parent, orig_root,
1245 orig_generation, owner_objectid, 1);
1246 if (ret)
1247 goto out;
1248 ret = remove_extent_backref(trans, extent_root, path);
1249 if (ret)
1250 goto out;
1251 ret = insert_extent_backref(trans, extent_root, path, bytenr,
1252 parent, ref_root, ref_generation,
1253 owner_objectid);
1254 BUG_ON(ret);
1255 finish_current_insert(trans, extent_root, 0);
1256 del_pending_extents(trans, extent_root, 0);
1257 out:
1258 btrfs_free_path(path);
1259 return ret;
1260 }
1261
1262 int btrfs_update_extent_ref(struct btrfs_trans_handle *trans,
1263 struct btrfs_root *root, u64 bytenr,
1264 u64 orig_parent, u64 parent,
1265 u64 ref_root, u64 ref_generation,
1266 u64 owner_objectid)
1267 {
1268 int ret;
1269 if (ref_root == BTRFS_TREE_LOG_OBJECTID &&
1270 owner_objectid < BTRFS_FIRST_FREE_OBJECTID)
1271 return 0;
1272 ret = __btrfs_update_extent_ref(trans, root, bytenr, orig_parent,
1273 parent, ref_root, ref_root,
1274 ref_generation, ref_generation,
1275 owner_objectid);
1276 return ret;
1277 }
1278
1279 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1280 struct btrfs_root *root, u64 bytenr,
1281 u64 orig_parent, u64 parent,
1282 u64 orig_root, u64 ref_root,
1283 u64 orig_generation, u64 ref_generation,
1284 u64 owner_objectid)
1285 {
1286 struct btrfs_path *path;
1287 int ret;
1288 struct btrfs_key key;
1289 struct extent_buffer *l;
1290 struct btrfs_extent_item *item;
1291 u32 refs;
1292
1293 path = btrfs_alloc_path();
1294 if (!path)
1295 return -ENOMEM;
1296
1297 path->reada = 1;
1298 key.objectid = bytenr;
1299 key.type = BTRFS_EXTENT_ITEM_KEY;
1300 key.offset = (u64)-1;
1301
1302 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path,
1303 0, 1);
1304 if (ret < 0)
1305 return ret;
1306 BUG_ON(ret == 0 || path->slots[0] == 0);
1307
1308 path->slots[0]--;
1309 l = path->nodes[0];
1310
1311 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
1312 if (key.objectid != bytenr) {
1313 btrfs_print_leaf(root->fs_info->extent_root, path->nodes[0]);
1314 printk("wanted %Lu found %Lu\n", bytenr, key.objectid);
1315 BUG();
1316 }
1317 BUG_ON(key.type != BTRFS_EXTENT_ITEM_KEY);
1318
1319 item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
1320 refs = btrfs_extent_refs(l, item);
1321 btrfs_set_extent_refs(l, item, refs + 1);
1322 btrfs_mark_buffer_dirty(path->nodes[0]);
1323
1324 btrfs_release_path(root->fs_info->extent_root, path);
1325
1326 path->reada = 1;
1327 ret = insert_extent_backref(trans, root->fs_info->extent_root,
1328 path, bytenr, parent,
1329 ref_root, ref_generation,
1330 owner_objectid);
1331 BUG_ON(ret);
1332 finish_current_insert(trans, root->fs_info->extent_root, 0);
1333 del_pending_extents(trans, root->fs_info->extent_root, 0);
1334
1335 btrfs_free_path(path);
1336 return 0;
1337 }
1338
1339 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1340 struct btrfs_root *root,
1341 u64 bytenr, u64 num_bytes, u64 parent,
1342 u64 ref_root, u64 ref_generation,
1343 u64 owner_objectid)
1344 {
1345 int ret;
1346 if (ref_root == BTRFS_TREE_LOG_OBJECTID &&
1347 owner_objectid < BTRFS_FIRST_FREE_OBJECTID)
1348 return 0;
1349 ret = __btrfs_inc_extent_ref(trans, root, bytenr, 0, parent,
1350 0, ref_root, 0, ref_generation,
1351 owner_objectid);
1352 return ret;
1353 }
1354
1355 int btrfs_extent_post_op(struct btrfs_trans_handle *trans,
1356 struct btrfs_root *root)
1357 {
1358 finish_current_insert(trans, root->fs_info->extent_root, 1);
1359 del_pending_extents(trans, root->fs_info->extent_root, 1);
1360 return 0;
1361 }
1362
1363 int btrfs_lookup_extent_ref(struct btrfs_trans_handle *trans,
1364 struct btrfs_root *root, u64 bytenr,
1365 u64 num_bytes, u32 *refs)
1366 {
1367 struct btrfs_path *path;
1368 int ret;
1369 struct btrfs_key key;
1370 struct extent_buffer *l;
1371 struct btrfs_extent_item *item;
1372
1373 WARN_ON(num_bytes < root->sectorsize);
1374 path = btrfs_alloc_path();
1375 path->reada = 1;
1376 key.objectid = bytenr;
1377 key.offset = num_bytes;
1378 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
1379 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path,
1380 0, 0);
1381 if (ret < 0)
1382 goto out;
1383 if (ret != 0) {
1384 btrfs_print_leaf(root, path->nodes[0]);
1385 printk("failed to find block number %Lu\n", bytenr);
1386 BUG();
1387 }
1388 l = path->nodes[0];
1389 item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
1390 *refs = btrfs_extent_refs(l, item);
1391 out:
1392 btrfs_free_path(path);
1393 return 0;
1394 }
1395
1396 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
1397 struct btrfs_root *root, u64 bytenr)
1398 {
1399 struct btrfs_root *extent_root = root->fs_info->extent_root;
1400 struct btrfs_path *path;
1401 struct extent_buffer *leaf;
1402 struct btrfs_extent_ref *ref_item;
1403 struct btrfs_key key;
1404 struct btrfs_key found_key;
1405 u64 ref_root;
1406 u64 last_snapshot;
1407 u32 nritems;
1408 int ret;
1409
1410 key.objectid = bytenr;
1411 key.offset = (u64)-1;
1412 key.type = BTRFS_EXTENT_ITEM_KEY;
1413
1414 path = btrfs_alloc_path();
1415 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
1416 if (ret < 0)
1417 goto out;
1418 BUG_ON(ret == 0);
1419
1420 ret = -ENOENT;
1421 if (path->slots[0] == 0)
1422 goto out;
1423
1424 path->slots[0]--;
1425 leaf = path->nodes[0];
1426 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1427
1428 if (found_key.objectid != bytenr ||
1429 found_key.type != BTRFS_EXTENT_ITEM_KEY)
1430 goto out;
1431
1432 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
1433 while (1) {
1434 leaf = path->nodes[0];
1435 nritems = btrfs_header_nritems(leaf);
1436 if (path->slots[0] >= nritems) {
1437 ret = btrfs_next_leaf(extent_root, path);
1438 if (ret < 0)
1439 goto out;
1440 if (ret == 0)
1441 continue;
1442 break;
1443 }
1444 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1445 if (found_key.objectid != bytenr)
1446 break;
1447
1448 if (found_key.type != BTRFS_EXTENT_REF_KEY) {
1449 path->slots[0]++;
1450 continue;
1451 }
1452
1453 ref_item = btrfs_item_ptr(leaf, path->slots[0],
1454 struct btrfs_extent_ref);
1455 ref_root = btrfs_ref_root(leaf, ref_item);
1456 if (ref_root != root->root_key.objectid &&
1457 ref_root != BTRFS_TREE_LOG_OBJECTID) {
1458 ret = 1;
1459 goto out;
1460 }
1461 if (btrfs_ref_generation(leaf, ref_item) <= last_snapshot) {
1462 ret = 1;
1463 goto out;
1464 }
1465
1466 path->slots[0]++;
1467 }
1468 ret = 0;
1469 out:
1470 btrfs_free_path(path);
1471 return ret;
1472 }
1473
1474 int btrfs_cache_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1475 struct extent_buffer *buf, u32 nr_extents)
1476 {
1477 struct btrfs_key key;
1478 struct btrfs_file_extent_item *fi;
1479 u64 root_gen;
1480 u32 nritems;
1481 int i;
1482 int level;
1483 int ret = 0;
1484 int shared = 0;
1485
1486 if (!root->ref_cows)
1487 return 0;
1488
1489 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
1490 shared = 0;
1491 root_gen = root->root_key.offset;
1492 } else {
1493 shared = 1;
1494 root_gen = trans->transid - 1;
1495 }
1496
1497 level = btrfs_header_level(buf);
1498 nritems = btrfs_header_nritems(buf);
1499
1500 if (level == 0) {
1501 struct btrfs_leaf_ref *ref;
1502 struct btrfs_extent_info *info;
1503
1504 ref = btrfs_alloc_leaf_ref(root, nr_extents);
1505 if (!ref) {
1506 ret = -ENOMEM;
1507 goto out;
1508 }
1509
1510 ref->root_gen = root_gen;
1511 ref->bytenr = buf->start;
1512 ref->owner = btrfs_header_owner(buf);
1513 ref->generation = btrfs_header_generation(buf);
1514 ref->nritems = nr_extents;
1515 info = ref->extents;
1516
1517 for (i = 0; nr_extents > 0 && i < nritems; i++) {
1518 u64 disk_bytenr;
1519 btrfs_item_key_to_cpu(buf, &key, i);
1520 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
1521 continue;
1522 fi = btrfs_item_ptr(buf, i,
1523 struct btrfs_file_extent_item);
1524 if (btrfs_file_extent_type(buf, fi) ==
1525 BTRFS_FILE_EXTENT_INLINE)
1526 continue;
1527 disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
1528 if (disk_bytenr == 0)
1529 continue;
1530
1531 info->bytenr = disk_bytenr;
1532 info->num_bytes =
1533 btrfs_file_extent_disk_num_bytes(buf, fi);
1534 info->objectid = key.objectid;
1535 info->offset = key.offset;
1536 info++;
1537 }
1538
1539 ret = btrfs_add_leaf_ref(root, ref, shared);
1540 if (ret == -EEXIST && shared) {
1541 struct btrfs_leaf_ref *old;
1542 old = btrfs_lookup_leaf_ref(root, ref->bytenr);
1543 BUG_ON(!old);
1544 btrfs_remove_leaf_ref(root, old);
1545 btrfs_free_leaf_ref(root, old);
1546 ret = btrfs_add_leaf_ref(root, ref, shared);
1547 }
1548 WARN_ON(ret);
1549 btrfs_free_leaf_ref(root, ref);
1550 }
1551 out:
1552 return ret;
1553 }
1554
1555 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1556 struct extent_buffer *orig_buf, struct extent_buffer *buf,
1557 u32 *nr_extents)
1558 {
1559 u64 bytenr;
1560 u64 ref_root;
1561 u64 orig_root;
1562 u64 ref_generation;
1563 u64 orig_generation;
1564 u32 nritems;
1565 u32 nr_file_extents = 0;
1566 struct btrfs_key key;
1567 struct btrfs_file_extent_item *fi;
1568 int i;
1569 int level;
1570 int ret = 0;
1571 int faili = 0;
1572 int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
1573 u64, u64, u64, u64, u64, u64, u64, u64);
1574
1575 ref_root = btrfs_header_owner(buf);
1576 ref_generation = btrfs_header_generation(buf);
1577 orig_root = btrfs_header_owner(orig_buf);
1578 orig_generation = btrfs_header_generation(orig_buf);
1579
1580 nritems = btrfs_header_nritems(buf);
1581 level = btrfs_header_level(buf);
1582
1583 if (root->ref_cows) {
1584 process_func = __btrfs_inc_extent_ref;
1585 } else {
1586 if (level == 0 &&
1587 root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
1588 goto out;
1589 if (level != 0 &&
1590 root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID)
1591 goto out;
1592 process_func = __btrfs_update_extent_ref;
1593 }
1594
1595 for (i = 0; i < nritems; i++) {
1596 cond_resched();
1597 if (level == 0) {
1598 btrfs_item_key_to_cpu(buf, &key, i);
1599 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
1600 continue;
1601 fi = btrfs_item_ptr(buf, i,
1602 struct btrfs_file_extent_item);
1603 if (btrfs_file_extent_type(buf, fi) ==
1604 BTRFS_FILE_EXTENT_INLINE)
1605 continue;
1606 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
1607 if (bytenr == 0)
1608 continue;
1609
1610 nr_file_extents++;
1611
1612 ret = process_func(trans, root, bytenr,
1613 orig_buf->start, buf->start,
1614 orig_root, ref_root,
1615 orig_generation, ref_generation,
1616 key.objectid);
1617
1618 if (ret) {
1619 faili = i;
1620 WARN_ON(1);
1621 goto fail;
1622 }
1623 } else {
1624 bytenr = btrfs_node_blockptr(buf, i);
1625 ret = process_func(trans, root, bytenr,
1626 orig_buf->start, buf->start,
1627 orig_root, ref_root,
1628 orig_generation, ref_generation,
1629 level - 1);
1630 if (ret) {
1631 faili = i;
1632 WARN_ON(1);
1633 goto fail;
1634 }
1635 }
1636 }
1637 out:
1638 if (nr_extents) {
1639 if (level == 0)
1640 *nr_extents = nr_file_extents;
1641 else
1642 *nr_extents = nritems;
1643 }
1644 return 0;
1645 fail:
1646 WARN_ON(1);
1647 return ret;
1648 }
1649
1650 int btrfs_update_ref(struct btrfs_trans_handle *trans,
1651 struct btrfs_root *root, struct extent_buffer *orig_buf,
1652 struct extent_buffer *buf, int start_slot, int nr)
1653
1654 {
1655 u64 bytenr;
1656 u64 ref_root;
1657 u64 orig_root;
1658 u64 ref_generation;
1659 u64 orig_generation;
1660 struct btrfs_key key;
1661 struct btrfs_file_extent_item *fi;
1662 int i;
1663 int ret;
1664 int slot;
1665 int level;
1666
1667 BUG_ON(start_slot < 0);
1668 BUG_ON(start_slot + nr > btrfs_header_nritems(buf));
1669
1670 ref_root = btrfs_header_owner(buf);
1671 ref_generation = btrfs_header_generation(buf);
1672 orig_root = btrfs_header_owner(orig_buf);
1673 orig_generation = btrfs_header_generation(orig_buf);
1674 level = btrfs_header_level(buf);
1675
1676 if (!root->ref_cows) {
1677 if (level == 0 &&
1678 root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
1679 return 0;
1680 if (level != 0 &&
1681 root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID)
1682 return 0;
1683 }
1684
1685 for (i = 0, slot = start_slot; i < nr; i++, slot++) {
1686 cond_resched();
1687 if (level == 0) {
1688 btrfs_item_key_to_cpu(buf, &key, slot);
1689 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
1690 continue;
1691 fi = btrfs_item_ptr(buf, slot,
1692 struct btrfs_file_extent_item);
1693 if (btrfs_file_extent_type(buf, fi) ==
1694 BTRFS_FILE_EXTENT_INLINE)
1695 continue;
1696 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
1697 if (bytenr == 0)
1698 continue;
1699 ret = __btrfs_update_extent_ref(trans, root, bytenr,
1700 orig_buf->start, buf->start,
1701 orig_root, ref_root,
1702 orig_generation, ref_generation,
1703 key.objectid);
1704 if (ret)
1705 goto fail;
1706 } else {
1707 bytenr = btrfs_node_blockptr(buf, slot);
1708 ret = __btrfs_update_extent_ref(trans, root, bytenr,
1709 orig_buf->start, buf->start,
1710 orig_root, ref_root,
1711 orig_generation, ref_generation,
1712 level - 1);
1713 if (ret)
1714 goto fail;
1715 }
1716 }
1717 return 0;
1718 fail:
1719 WARN_ON(1);
1720 return -1;
1721 }
1722
1723 static int write_one_cache_group(struct btrfs_trans_handle *trans,
1724 struct btrfs_root *root,
1725 struct btrfs_path *path,
1726 struct btrfs_block_group_cache *cache)
1727 {
1728 int ret;
1729 int pending_ret;
1730 struct btrfs_root *extent_root = root->fs_info->extent_root;
1731 unsigned long bi;
1732 struct extent_buffer *leaf;
1733
1734 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
1735 if (ret < 0)
1736 goto fail;
1737 BUG_ON(ret);
1738
1739 leaf = path->nodes[0];
1740 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
1741 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
1742 btrfs_mark_buffer_dirty(leaf);
1743 btrfs_release_path(extent_root, path);
1744 fail:
1745 finish_current_insert(trans, extent_root, 0);
1746 pending_ret = del_pending_extents(trans, extent_root, 0);
1747 if (ret)
1748 return ret;
1749 if (pending_ret)
1750 return pending_ret;
1751 return 0;
1752
1753 }
1754
1755 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
1756 struct btrfs_root *root)
1757 {
1758 struct btrfs_block_group_cache *cache, *entry;
1759 struct rb_node *n;
1760 int err = 0;
1761 int werr = 0;
1762 struct btrfs_path *path;
1763 u64 last = 0;
1764
1765 path = btrfs_alloc_path();
1766 if (!path)
1767 return -ENOMEM;
1768
1769 while(1) {
1770 cache = NULL;
1771 spin_lock(&root->fs_info->block_group_cache_lock);
1772 for (n = rb_first(&root->fs_info->block_group_cache_tree);
1773 n; n = rb_next(n)) {
1774 entry = rb_entry(n, struct btrfs_block_group_cache,
1775 cache_node);
1776 if (entry->dirty) {
1777 cache = entry;
1778 break;
1779 }
1780 }
1781 spin_unlock(&root->fs_info->block_group_cache_lock);
1782
1783 if (!cache)
1784 break;
1785
1786 cache->dirty = 0;
1787 last += cache->key.offset;
1788
1789 err = write_one_cache_group(trans, root,
1790 path, cache);
1791 /*
1792 * if we fail to write the cache group, we want
1793 * to keep it marked dirty in hopes that a later
1794 * write will work
1795 */
1796 if (err) {
1797 werr = err;
1798 continue;
1799 }
1800 }
1801 btrfs_free_path(path);
1802 return werr;
1803 }
1804
1805 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
1806 u64 total_bytes, u64 bytes_used,
1807 struct btrfs_space_info **space_info)
1808 {
1809 struct btrfs_space_info *found;
1810
1811 found = __find_space_info(info, flags);
1812 if (found) {
1813 spin_lock(&found->lock);
1814 found->total_bytes += total_bytes;
1815 found->bytes_used += bytes_used;
1816 found->full = 0;
1817 spin_unlock(&found->lock);
1818 *space_info = found;
1819 return 0;
1820 }
1821 found = kzalloc(sizeof(*found), GFP_NOFS);
1822 if (!found)
1823 return -ENOMEM;
1824
1825 list_add(&found->list, &info->space_info);
1826 INIT_LIST_HEAD(&found->block_groups);
1827 init_rwsem(&found->groups_sem);
1828 spin_lock_init(&found->lock);
1829 found->flags = flags;
1830 found->total_bytes = total_bytes;
1831 found->bytes_used = bytes_used;
1832 found->bytes_pinned = 0;
1833 found->bytes_reserved = 0;
1834 found->bytes_readonly = 0;
1835 found->full = 0;
1836 found->force_alloc = 0;
1837 *space_info = found;
1838 return 0;
1839 }
1840
1841 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
1842 {
1843 u64 extra_flags = flags & (BTRFS_BLOCK_GROUP_RAID0 |
1844 BTRFS_BLOCK_GROUP_RAID1 |
1845 BTRFS_BLOCK_GROUP_RAID10 |
1846 BTRFS_BLOCK_GROUP_DUP);
1847 if (extra_flags) {
1848 if (flags & BTRFS_BLOCK_GROUP_DATA)
1849 fs_info->avail_data_alloc_bits |= extra_flags;
1850 if (flags & BTRFS_BLOCK_GROUP_METADATA)
1851 fs_info->avail_metadata_alloc_bits |= extra_flags;
1852 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
1853 fs_info->avail_system_alloc_bits |= extra_flags;
1854 }
1855 }
1856
1857 static void set_block_group_readonly(struct btrfs_block_group_cache *cache)
1858 {
1859 spin_lock(&cache->space_info->lock);
1860 spin_lock(&cache->lock);
1861 if (!cache->ro) {
1862 cache->space_info->bytes_readonly += cache->key.offset -
1863 btrfs_block_group_used(&cache->item);
1864 cache->ro = 1;
1865 }
1866 spin_unlock(&cache->lock);
1867 spin_unlock(&cache->space_info->lock);
1868 }
1869
1870 u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
1871 {
1872 u64 num_devices = root->fs_info->fs_devices->rw_devices;
1873
1874 if (num_devices == 1)
1875 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0);
1876 if (num_devices < 4)
1877 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
1878
1879 if ((flags & BTRFS_BLOCK_GROUP_DUP) &&
1880 (flags & (BTRFS_BLOCK_GROUP_RAID1 |
1881 BTRFS_BLOCK_GROUP_RAID10))) {
1882 flags &= ~BTRFS_BLOCK_GROUP_DUP;
1883 }
1884
1885 if ((flags & BTRFS_BLOCK_GROUP_RAID1) &&
1886 (flags & BTRFS_BLOCK_GROUP_RAID10)) {
1887 flags &= ~BTRFS_BLOCK_GROUP_RAID1;
1888 }
1889
1890 if ((flags & BTRFS_BLOCK_GROUP_RAID0) &&
1891 ((flags & BTRFS_BLOCK_GROUP_RAID1) |
1892 (flags & BTRFS_BLOCK_GROUP_RAID10) |
1893 (flags & BTRFS_BLOCK_GROUP_DUP)))
1894 flags &= ~BTRFS_BLOCK_GROUP_RAID0;
1895 return flags;
1896 }
1897
1898 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
1899 struct btrfs_root *extent_root, u64 alloc_bytes,
1900 u64 flags, int force)
1901 {
1902 struct btrfs_space_info *space_info;
1903 u64 thresh;
1904 int ret = 0;
1905
1906 mutex_lock(&extent_root->fs_info->chunk_mutex);
1907
1908 flags = btrfs_reduce_alloc_profile(extent_root, flags);
1909
1910 space_info = __find_space_info(extent_root->fs_info, flags);
1911 if (!space_info) {
1912 ret = update_space_info(extent_root->fs_info, flags,
1913 0, 0, &space_info);
1914 BUG_ON(ret);
1915 }
1916 BUG_ON(!space_info);
1917
1918 spin_lock(&space_info->lock);
1919 if (space_info->force_alloc) {
1920 force = 1;
1921 space_info->force_alloc = 0;
1922 }
1923 if (space_info->full) {
1924 spin_unlock(&space_info->lock);
1925 goto out;
1926 }
1927
1928 thresh = space_info->total_bytes - space_info->bytes_readonly;
1929 thresh = div_factor(thresh, 6);
1930 if (!force &&
1931 (space_info->bytes_used + space_info->bytes_pinned +
1932 space_info->bytes_reserved + alloc_bytes) < thresh) {
1933 spin_unlock(&space_info->lock);
1934 goto out;
1935 }
1936 spin_unlock(&space_info->lock);
1937
1938 ret = btrfs_alloc_chunk(trans, extent_root, flags);
1939 if (ret) {
1940 printk("space info full %Lu\n", flags);
1941 space_info->full = 1;
1942 }
1943 out:
1944 mutex_unlock(&extent_root->fs_info->chunk_mutex);
1945 return ret;
1946 }
1947
1948 static int update_block_group(struct btrfs_trans_handle *trans,
1949 struct btrfs_root *root,
1950 u64 bytenr, u64 num_bytes, int alloc,
1951 int mark_free)
1952 {
1953 struct btrfs_block_group_cache *cache;
1954 struct btrfs_fs_info *info = root->fs_info;
1955 u64 total = num_bytes;
1956 u64 old_val;
1957 u64 byte_in_group;
1958
1959 while(total) {
1960 cache = btrfs_lookup_block_group(info, bytenr);
1961 if (!cache)
1962 return -1;
1963 byte_in_group = bytenr - cache->key.objectid;
1964 WARN_ON(byte_in_group > cache->key.offset);
1965
1966 spin_lock(&cache->space_info->lock);
1967 spin_lock(&cache->lock);
1968 cache->dirty = 1;
1969 old_val = btrfs_block_group_used(&cache->item);
1970 num_bytes = min(total, cache->key.offset - byte_in_group);
1971 if (alloc) {
1972 old_val += num_bytes;
1973 cache->space_info->bytes_used += num_bytes;
1974 if (cache->ro) {
1975 cache->space_info->bytes_readonly -= num_bytes;
1976 WARN_ON(1);
1977 }
1978 btrfs_set_block_group_used(&cache->item, old_val);
1979 spin_unlock(&cache->lock);
1980 spin_unlock(&cache->space_info->lock);
1981 } else {
1982 old_val -= num_bytes;
1983 cache->space_info->bytes_used -= num_bytes;
1984 if (cache->ro)
1985 cache->space_info->bytes_readonly += num_bytes;
1986 btrfs_set_block_group_used(&cache->item, old_val);
1987 spin_unlock(&cache->lock);
1988 spin_unlock(&cache->space_info->lock);
1989 if (mark_free) {
1990 int ret;
1991 ret = btrfs_add_free_space(cache, bytenr,
1992 num_bytes);
1993 if (ret)
1994 return -1;
1995 }
1996 }
1997 total -= num_bytes;
1998 bytenr += num_bytes;
1999 }
2000 return 0;
2001 }
2002
2003 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
2004 {
2005 struct btrfs_block_group_cache *cache;
2006
2007 cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
2008 if (!cache)
2009 return 0;
2010
2011 return cache->key.objectid;
2012 }
2013
2014 int btrfs_update_pinned_extents(struct btrfs_root *root,
2015 u64 bytenr, u64 num, int pin)
2016 {
2017 u64 len;
2018 struct btrfs_block_group_cache *cache;
2019 struct btrfs_fs_info *fs_info = root->fs_info;
2020
2021 WARN_ON(!mutex_is_locked(&root->fs_info->pinned_mutex));
2022 if (pin) {
2023 set_extent_dirty(&fs_info->pinned_extents,
2024 bytenr, bytenr + num - 1, GFP_NOFS);
2025 } else {
2026 clear_extent_dirty(&fs_info->pinned_extents,
2027 bytenr, bytenr + num - 1, GFP_NOFS);
2028 }
2029 while (num > 0) {
2030 cache = btrfs_lookup_block_group(fs_info, bytenr);
2031 BUG_ON(!cache);
2032 len = min(num, cache->key.offset -
2033 (bytenr - cache->key.objectid));
2034 if (pin) {
2035 spin_lock(&cache->space_info->lock);
2036 spin_lock(&cache->lock);
2037 cache->pinned += len;
2038 cache->space_info->bytes_pinned += len;
2039 spin_unlock(&cache->lock);
2040 spin_unlock(&cache->space_info->lock);
2041 fs_info->total_pinned += len;
2042 } else {
2043 spin_lock(&cache->space_info->lock);
2044 spin_lock(&cache->lock);
2045 cache->pinned -= len;
2046 cache->space_info->bytes_pinned -= len;
2047 spin_unlock(&cache->lock);
2048 spin_unlock(&cache->space_info->lock);
2049 fs_info->total_pinned -= len;
2050 if (cache->cached)
2051 btrfs_add_free_space(cache, bytenr, len);
2052 }
2053 bytenr += len;
2054 num -= len;
2055 }
2056 return 0;
2057 }
2058
2059 static int update_reserved_extents(struct btrfs_root *root,
2060 u64 bytenr, u64 num, int reserve)
2061 {
2062 u64 len;
2063 struct btrfs_block_group_cache *cache;
2064 struct btrfs_fs_info *fs_info = root->fs_info;
2065
2066 while (num > 0) {
2067 cache = btrfs_lookup_block_group(fs_info, bytenr);
2068 BUG_ON(!cache);
2069 len = min(num, cache->key.offset -
2070 (bytenr - cache->key.objectid));
2071
2072 spin_lock(&cache->space_info->lock);
2073 spin_lock(&cache->lock);
2074 if (reserve) {
2075 cache->reserved += len;
2076 cache->space_info->bytes_reserved += len;
2077 } else {
2078 cache->reserved -= len;
2079 cache->space_info->bytes_reserved -= len;
2080 }
2081 spin_unlock(&cache->lock);
2082 spin_unlock(&cache->space_info->lock);
2083 bytenr += len;
2084 num -= len;
2085 }
2086 return 0;
2087 }
2088
2089 int btrfs_copy_pinned(struct btrfs_root *root, struct extent_io_tree *copy)
2090 {
2091 u64 last = 0;
2092 u64 start;
2093 u64 end;
2094 struct extent_io_tree *pinned_extents = &root->fs_info->pinned_extents;
2095 int ret;
2096
2097 mutex_lock(&root->fs_info->pinned_mutex);
2098 while(1) {
2099 ret = find_first_extent_bit(pinned_extents, last,
2100 &start, &end, EXTENT_DIRTY);
2101 if (ret)
2102 break;
2103 set_extent_dirty(copy, start, end, GFP_NOFS);
2104 last = end + 1;
2105 }
2106 mutex_unlock(&root->fs_info->pinned_mutex);
2107 return 0;
2108 }
2109
2110 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
2111 struct btrfs_root *root,
2112 struct extent_io_tree *unpin)
2113 {
2114 u64 start;
2115 u64 end;
2116 int ret;
2117
2118 mutex_lock(&root->fs_info->pinned_mutex);
2119 while(1) {
2120 ret = find_first_extent_bit(unpin, 0, &start, &end,
2121 EXTENT_DIRTY);
2122 if (ret)
2123 break;
2124 btrfs_update_pinned_extents(root, start, end + 1 - start, 0);
2125 clear_extent_dirty(unpin, start, end, GFP_NOFS);
2126 if (need_resched()) {
2127 mutex_unlock(&root->fs_info->pinned_mutex);
2128 cond_resched();
2129 mutex_lock(&root->fs_info->pinned_mutex);
2130 }
2131 }
2132 mutex_unlock(&root->fs_info->pinned_mutex);
2133 return 0;
2134 }
2135
2136 static int finish_current_insert(struct btrfs_trans_handle *trans,
2137 struct btrfs_root *extent_root, int all)
2138 {
2139 u64 start;
2140 u64 end;
2141 u64 priv;
2142 u64 search = 0;
2143 u64 skipped = 0;
2144 struct btrfs_fs_info *info = extent_root->fs_info;
2145 struct btrfs_path *path;
2146 struct pending_extent_op *extent_op, *tmp;
2147 struct list_head insert_list, update_list;
2148 int ret;
2149 int num_inserts = 0, max_inserts;
2150
2151 path = btrfs_alloc_path();
2152 INIT_LIST_HEAD(&insert_list);
2153 INIT_LIST_HEAD(&update_list);
2154
2155 max_inserts = extent_root->leafsize /
2156 (2 * sizeof(struct btrfs_key) + 2 * sizeof(struct btrfs_item) +
2157 sizeof(struct btrfs_extent_ref) +
2158 sizeof(struct btrfs_extent_item));
2159 again:
2160 mutex_lock(&info->extent_ins_mutex);
2161 while (1) {
2162 ret = find_first_extent_bit(&info->extent_ins, search, &start,
2163 &end, EXTENT_WRITEBACK);
2164 if (ret) {
2165 if (skipped && all && !num_inserts) {
2166 skipped = 0;
2167 search = 0;
2168 continue;
2169 }
2170 mutex_unlock(&info->extent_ins_mutex);
2171 break;
2172 }
2173
2174 ret = try_lock_extent(&info->extent_ins, start, end, GFP_NOFS);
2175 if (!ret) {
2176 skipped = 1;
2177 search = end + 1;
2178 if (need_resched()) {
2179 mutex_unlock(&info->extent_ins_mutex);
2180 cond_resched();
2181 mutex_lock(&info->extent_ins_mutex);
2182 }
2183 continue;
2184 }
2185
2186 ret = get_state_private(&info->extent_ins, start, &priv);
2187 BUG_ON(ret);
2188 extent_op = (struct pending_extent_op *)(unsigned long) priv;
2189
2190 if (extent_op->type == PENDING_EXTENT_INSERT) {
2191 num_inserts++;
2192 list_add_tail(&extent_op->list, &insert_list);
2193 search = end + 1;
2194 if (num_inserts == max_inserts) {
2195 mutex_unlock(&info->extent_ins_mutex);
2196 break;
2197 }
2198 } else if (extent_op->type == PENDING_BACKREF_UPDATE) {
2199 list_add_tail(&extent_op->list, &update_list);
2200 search = end + 1;
2201 } else {
2202 BUG();
2203 }
2204 }
2205
2206 /*
2207 * process the update list, clear the writeback bit for it, and if
2208 * somebody marked this thing for deletion then just unlock it and be
2209 * done, the free_extents will handle it
2210 */
2211 mutex_lock(&info->extent_ins_mutex);
2212 list_for_each_entry_safe(extent_op, tmp, &update_list, list) {
2213 clear_extent_bits(&info->extent_ins, extent_op->bytenr,
2214 extent_op->bytenr + extent_op->num_bytes - 1,
2215 EXTENT_WRITEBACK, GFP_NOFS);
2216 if (extent_op->del) {
2217 list_del_init(&extent_op->list);
2218 unlock_extent(&info->extent_ins, extent_op->bytenr,
2219 extent_op->bytenr + extent_op->num_bytes
2220 - 1, GFP_NOFS);
2221 kfree(extent_op);
2222 }
2223 }
2224 mutex_unlock(&info->extent_ins_mutex);
2225
2226 /*
2227 * still have things left on the update list, go ahead an update
2228 * everything
2229 */
2230 if (!list_empty(&update_list)) {
2231 ret = update_backrefs(trans, extent_root, path, &update_list);
2232 BUG_ON(ret);
2233 }
2234
2235 /*
2236 * if no inserts need to be done, but we skipped some extents and we
2237 * need to make sure everything is cleaned then reset everything and
2238 * go back to the beginning
2239 */
2240 if (!num_inserts && all && skipped) {
2241 search = 0;
2242 skipped = 0;
2243 INIT_LIST_HEAD(&update_list);
2244 INIT_LIST_HEAD(&insert_list);
2245 goto again;
2246 } else if (!num_inserts) {
2247 goto out;
2248 }
2249
2250 /*
2251 * process the insert extents list. Again if we are deleting this
2252 * extent, then just unlock it, pin down the bytes if need be, and be
2253 * done with it. Saves us from having to actually insert the extent
2254 * into the tree and then subsequently come along and delete it
2255 */
2256 mutex_lock(&info->extent_ins_mutex);
2257 list_for_each_entry_safe(extent_op, tmp, &insert_list, list) {
2258 clear_extent_bits(&info->extent_ins, extent_op->bytenr,
2259 extent_op->bytenr + extent_op->num_bytes - 1,
2260 EXTENT_WRITEBACK, GFP_NOFS);
2261 if (extent_op->del) {
2262 list_del_init(&extent_op->list);
2263 unlock_extent(&info->extent_ins, extent_op->bytenr,
2264 extent_op->bytenr + extent_op->num_bytes
2265 - 1, GFP_NOFS);
2266
2267 mutex_lock(&extent_root->fs_info->pinned_mutex);
2268 ret = pin_down_bytes(trans, extent_root,
2269 extent_op->bytenr,
2270 extent_op->num_bytes, 0);
2271 mutex_unlock(&extent_root->fs_info->pinned_mutex);
2272
2273 ret = update_block_group(trans, extent_root,
2274 extent_op->bytenr,
2275 extent_op->num_bytes,
2276 0, ret > 0);
2277 BUG_ON(ret);
2278 kfree(extent_op);
2279 num_inserts--;
2280 }
2281 }
2282 mutex_unlock(&info->extent_ins_mutex);
2283
2284 ret = insert_extents(trans, extent_root, path, &insert_list,
2285 num_inserts);
2286 BUG_ON(ret);
2287
2288 /*
2289 * if we broke out of the loop in order to insert stuff because we hit
2290 * the maximum number of inserts at a time we can handle, then loop
2291 * back and pick up where we left off
2292 */
2293 if (num_inserts == max_inserts) {
2294 INIT_LIST_HEAD(&insert_list);
2295 INIT_LIST_HEAD(&update_list);
2296 num_inserts = 0;
2297 goto again;
2298 }
2299
2300 /*
2301 * again, if we need to make absolutely sure there are no more pending
2302 * extent operations left and we know that we skipped some, go back to
2303 * the beginning and do it all again
2304 */
2305 if (all && skipped) {
2306 INIT_LIST_HEAD(&insert_list);
2307 INIT_LIST_HEAD(&update_list);
2308 search = 0;
2309 skipped = 0;
2310 num_inserts = 0;
2311 goto again;
2312 }
2313 out:
2314 btrfs_free_path(path);
2315 return 0;
2316 }
2317
2318 static int pin_down_bytes(struct btrfs_trans_handle *trans,
2319 struct btrfs_root *root,
2320 u64 bytenr, u64 num_bytes, int is_data)
2321 {
2322 int err = 0;
2323 struct extent_buffer *buf;
2324
2325 if (is_data)
2326 goto pinit;
2327
2328 buf = btrfs_find_tree_block(root, bytenr, num_bytes);
2329 if (!buf)
2330 goto pinit;
2331
2332 /* we can reuse a block if it hasn't been written
2333 * and it is from this transaction. We can't
2334 * reuse anything from the tree log root because
2335 * it has tiny sub-transactions.
2336 */
2337 if (btrfs_buffer_uptodate(buf, 0) &&
2338 btrfs_try_tree_lock(buf)) {
2339 u64 header_owner = btrfs_header_owner(buf);
2340 u64 header_transid = btrfs_header_generation(buf);
2341 if (header_owner != BTRFS_TREE_LOG_OBJECTID &&
2342 header_owner != BTRFS_TREE_RELOC_OBJECTID &&
2343 header_transid == trans->transid &&
2344 !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
2345 clean_tree_block(NULL, root, buf);
2346 btrfs_tree_unlock(buf);
2347 free_extent_buffer(buf);
2348 return 1;
2349 }
2350 btrfs_tree_unlock(buf);
2351 }
2352 free_extent_buffer(buf);
2353 pinit:
2354 btrfs_update_pinned_extents(root, bytenr, num_bytes, 1);
2355
2356 BUG_ON(err < 0);
2357 return 0;
2358 }
2359
2360 /*
2361 * remove an extent from the root, returns 0 on success
2362 */
2363 static int __free_extent(struct btrfs_trans_handle *trans,
2364 struct btrfs_root *root,
2365 u64 bytenr, u64 num_bytes, u64 parent,
2366 u64 root_objectid, u64 ref_generation,
2367 u64 owner_objectid, int pin, int mark_free)
2368 {
2369 struct btrfs_path *path;
2370 struct btrfs_key key;
2371 struct btrfs_fs_info *info = root->fs_info;
2372 struct btrfs_root *extent_root = info->extent_root;
2373 struct extent_buffer *leaf;
2374 int ret;
2375 int extent_slot = 0;
2376 int found_extent = 0;
2377 int num_to_del = 1;
2378 struct btrfs_extent_item *ei;
2379 u32 refs;
2380
2381 key.objectid = bytenr;
2382 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
2383 key.offset = num_bytes;
2384 path = btrfs_alloc_path();
2385 if (!path)
2386 return -ENOMEM;
2387
2388 path->reada = 1;
2389 ret = lookup_extent_backref(trans, extent_root, path,
2390 bytenr, parent, root_objectid,
2391 ref_generation, owner_objectid, 1);
2392 if (ret == 0) {
2393 struct btrfs_key found_key;
2394 extent_slot = path->slots[0];
2395 while(extent_slot > 0) {
2396 extent_slot--;
2397 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
2398 extent_slot);
2399 if (found_key.objectid != bytenr)
2400 break;
2401 if (found_key.type == BTRFS_EXTENT_ITEM_KEY &&
2402 found_key.offset == num_bytes) {
2403 found_extent = 1;
2404 break;
2405 }
2406 if (path->slots[0] - extent_slot > 5)
2407 break;
2408 }
2409 if (!found_extent) {
2410 ret = remove_extent_backref(trans, extent_root, path);
2411 BUG_ON(ret);
2412 btrfs_release_path(extent_root, path);
2413 ret = btrfs_search_slot(trans, extent_root,
2414 &key, path, -1, 1);
2415 if (ret) {
2416 printk(KERN_ERR "umm, got %d back from search"
2417 ", was looking for %Lu\n", ret,
2418 bytenr);
2419 btrfs_print_leaf(extent_root, path->nodes[0]);
2420 }
2421 BUG_ON(ret);
2422 extent_slot = path->slots[0];
2423 }
2424 } else {
2425 btrfs_print_leaf(extent_root, path->nodes[0]);
2426 WARN_ON(1);
2427 printk("Unable to find ref byte nr %Lu root %Lu "
2428 "gen %Lu owner %Lu\n", bytenr,
2429 root_objectid, ref_generation, owner_objectid);
2430 }
2431
2432 leaf = path->nodes[0];
2433 ei = btrfs_item_ptr(leaf, extent_slot,
2434 struct btrfs_extent_item);
2435 refs = btrfs_extent_refs(leaf, ei);
2436 BUG_ON(refs == 0);
2437 refs -= 1;
2438 btrfs_set_extent_refs(leaf, ei, refs);
2439
2440 btrfs_mark_buffer_dirty(leaf);
2441
2442 if (refs == 0 && found_extent && path->slots[0] == extent_slot + 1) {
2443 struct btrfs_extent_ref *ref;
2444 ref = btrfs_item_ptr(leaf, path->slots[0],
2445 struct btrfs_extent_ref);
2446 BUG_ON(btrfs_ref_num_refs(leaf, ref) != 1);
2447 /* if the back ref and the extent are next to each other
2448 * they get deleted below in one shot
2449 */
2450 path->slots[0] = extent_slot;
2451 num_to_del = 2;
2452 } else if (found_extent) {
2453 /* otherwise delete the extent back ref */
2454 ret = remove_extent_backref(trans, extent_root, path);
2455 BUG_ON(ret);
2456 /* if refs are 0, we need to setup the path for deletion */
2457 if (refs == 0) {
2458 btrfs_release_path(extent_root, path);
2459 ret = btrfs_search_slot(trans, extent_root, &key, path,
2460 -1, 1);
2461 BUG_ON(ret);
2462 }
2463 }
2464
2465 if (refs == 0) {
2466 u64 super_used;
2467 u64 root_used;
2468 #ifdef BIO_RW_DISCARD
2469 u64 map_length = num_bytes;
2470 struct btrfs_multi_bio *multi = NULL;
2471 #endif
2472
2473 if (pin) {
2474 mutex_lock(&root->fs_info->pinned_mutex);
2475 ret = pin_down_bytes(trans, root, bytenr, num_bytes,
2476 owner_objectid >= BTRFS_FIRST_FREE_OBJECTID);
2477 mutex_unlock(&root->fs_info->pinned_mutex);
2478 if (ret > 0)
2479 mark_free = 1;
2480 BUG_ON(ret < 0);
2481 }
2482
2483 /* block accounting for super block */
2484 spin_lock_irq(&info->delalloc_lock);
2485 super_used = btrfs_super_bytes_used(&info->super_copy);
2486 btrfs_set_super_bytes_used(&info->super_copy,
2487 super_used - num_bytes);
2488 spin_unlock_irq(&info->delalloc_lock);
2489
2490 /* block accounting for root item */
2491 root_used = btrfs_root_used(&root->root_item);
2492 btrfs_set_root_used(&root->root_item,
2493 root_used - num_bytes);
2494 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
2495 num_to_del);
2496 BUG_ON(ret);
2497 btrfs_release_path(extent_root, path);
2498 ret = update_block_group(trans, root, bytenr, num_bytes, 0,
2499 mark_free);
2500 BUG_ON(ret);
2501
2502 #ifdef BIO_RW_DISCARD
2503 /* Tell the block device(s) that the sectors can be discarded */
2504 ret = btrfs_map_block(&root->fs_info->mapping_tree, READ,
2505 bytenr, &map_length, &multi, 0);
2506 if (!ret) {
2507 struct btrfs_bio_stripe *stripe = multi->stripes;
2508 int i;
2509
2510 if (map_length > num_bytes)
2511 map_length = num_bytes;
2512
2513 for (i = 0; i < multi->num_stripes; i++, stripe++) {
2514 btrfs_issue_discard(stripe->dev->bdev,
2515 stripe->physical,
2516 map_length);
2517 }
2518 kfree(multi);
2519 }
2520 #endif
2521 }
2522 btrfs_free_path(path);
2523 finish_current_insert(trans, extent_root, 0);
2524 return ret;
2525 }
2526
2527 /*
2528 * find all the blocks marked as pending in the radix tree and remove
2529 * them from the extent map
2530 */
2531 static int del_pending_extents(struct btrfs_trans_handle *trans, struct
2532 btrfs_root *extent_root, int all)
2533 {
2534 int ret;
2535 int err = 0;
2536 u64 start;
2537 u64 end;
2538 u64 priv;
2539 u64 search = 0;
2540 int nr = 0, skipped = 0;
2541 struct extent_io_tree *pending_del;
2542 struct extent_io_tree *extent_ins;
2543 struct pending_extent_op *extent_op;
2544 struct btrfs_fs_info *info = extent_root->fs_info;
2545 struct list_head delete_list;
2546
2547 INIT_LIST_HEAD(&delete_list);
2548 extent_ins = &extent_root->fs_info->extent_ins;
2549 pending_del = &extent_root->fs_info->pending_del;
2550
2551 again:
2552 mutex_lock(&info->extent_ins_mutex);
2553 while(1) {
2554 ret = find_first_extent_bit(pending_del, search, &start, &end,
2555 EXTENT_WRITEBACK);
2556 if (ret) {
2557 if (all && skipped && !nr) {
2558 search = 0;
2559 continue;
2560 }
2561 mutex_unlock(&info->extent_ins_mutex);
2562 break;
2563 }
2564
2565 ret = try_lock_extent(extent_ins, start, end, GFP_NOFS);
2566 if (!ret) {
2567 search = end+1;
2568 skipped = 1;
2569
2570 if (need_resched()) {
2571 mutex_unlock(&info->extent_ins_mutex);
2572 cond_resched();
2573 mutex_lock(&info->extent_ins_mutex);
2574 }
2575
2576 continue;
2577 }
2578 BUG_ON(ret < 0);
2579
2580 ret = get_state_private(pending_del, start, &priv);
2581 BUG_ON(ret);
2582 extent_op = (struct pending_extent_op *)(unsigned long)priv;
2583
2584 clear_extent_bits(pending_del, start, end, EXTENT_WRITEBACK,
2585 GFP_NOFS);
2586 if (!test_range_bit(extent_ins, start, end,
2587 EXTENT_WRITEBACK, 0)) {
2588 list_add_tail(&extent_op->list, &delete_list);
2589 nr++;
2590 } else {
2591 kfree(extent_op);
2592
2593 ret = get_state_private(&info->extent_ins, start,
2594 &priv);
2595 BUG_ON(ret);
2596 extent_op = (struct pending_extent_op *)
2597 (unsigned long)priv;
2598
2599 clear_extent_bits(&info->extent_ins, start, end,
2600 EXTENT_WRITEBACK, GFP_NOFS);
2601
2602 if (extent_op->type == PENDING_BACKREF_UPDATE) {
2603 list_add_tail(&extent_op->list, &delete_list);
2604 search = end + 1;
2605 nr++;
2606 continue;
2607 }
2608
2609 mutex_lock(&extent_root->fs_info->pinned_mutex);
2610 ret = pin_down_bytes(trans, extent_root, start,
2611 end + 1 - start, 0);
2612 mutex_unlock(&extent_root->fs_info->pinned_mutex);
2613
2614 ret = update_block_group(trans, extent_root, start,
2615 end + 1 - start, 0, ret > 0);
2616
2617 unlock_extent(extent_ins, start, end, GFP_NOFS);
2618 BUG_ON(ret);
2619 kfree(extent_op);
2620 }
2621 if (ret)
2622 err = ret;
2623
2624 search = end + 1;
2625
2626 if (need_resched()) {
2627 mutex_unlock(&info->extent_ins_mutex);
2628 cond_resched();
2629 mutex_lock(&info->extent_ins_mutex);
2630 }
2631 }
2632
2633 if (nr) {
2634 ret = free_extents(trans, extent_root, &delete_list);
2635 BUG_ON(ret);
2636 }
2637
2638 if (all && skipped) {
2639 INIT_LIST_HEAD(&delete_list);
2640 search = 0;
2641 nr = 0;
2642 goto again;
2643 }
2644
2645 return err;
2646 }
2647
2648 /*
2649 * remove an extent from the root, returns 0 on success
2650 */
2651 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
2652 struct btrfs_root *root,
2653 u64 bytenr, u64 num_bytes, u64 parent,
2654 u64 root_objectid, u64 ref_generation,
2655 u64 owner_objectid, int pin)
2656 {
2657 struct btrfs_root *extent_root = root->fs_info->extent_root;
2658 int pending_ret;
2659 int ret;
2660
2661 WARN_ON(num_bytes < root->sectorsize);
2662 if (root == extent_root) {
2663 struct pending_extent_op *extent_op = NULL;
2664
2665 mutex_lock(&root->fs_info->extent_ins_mutex);
2666 if (test_range_bit(&root->fs_info->extent_ins, bytenr,
2667 bytenr + num_bytes - 1, EXTENT_WRITEBACK, 0)) {
2668 u64 priv;
2669 ret = get_state_private(&root->fs_info->extent_ins,
2670 bytenr, &priv);
2671 BUG_ON(ret);
2672 extent_op = (struct pending_extent_op *)
2673 (unsigned long)priv;
2674
2675 extent_op->del = 1;
2676 if (extent_op->type == PENDING_EXTENT_INSERT) {
2677 mutex_unlock(&root->fs_info->extent_ins_mutex);
2678 return 0;
2679 }
2680 }
2681
2682 if (extent_op) {
2683 ref_generation = extent_op->orig_generation;
2684 parent = extent_op->orig_parent;
2685 }
2686
2687 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
2688 BUG_ON(!extent_op);
2689
2690 extent_op->type = PENDING_EXTENT_DELETE;
2691 extent_op->bytenr = bytenr;
2692 extent_op->num_bytes = num_bytes;
2693 extent_op->parent = parent;
2694 extent_op->orig_parent = parent;
2695 extent_op->generation = ref_generation;
2696 extent_op->orig_generation = ref_generation;
2697 extent_op->level = (int)owner_objectid;
2698 INIT_LIST_HEAD(&extent_op->list);
2699 extent_op->del = 0;
2700
2701 set_extent_bits(&root->fs_info->pending_del,
2702 bytenr, bytenr + num_bytes - 1,
2703 EXTENT_WRITEBACK, GFP_NOFS);
2704 set_state_private(&root->fs_info->pending_del,
2705 bytenr, (unsigned long)extent_op);
2706 mutex_unlock(&root->fs_info->extent_ins_mutex);
2707 return 0;
2708 }
2709 /* if metadata always pin */
2710 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) {
2711 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
2712 struct btrfs_block_group_cache *cache;
2713
2714 /* btrfs_free_reserved_extent */
2715 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
2716 BUG_ON(!cache);
2717 btrfs_add_free_space(cache, bytenr, num_bytes);
2718 update_reserved_extents(root, bytenr, num_bytes, 0);
2719 return 0;
2720 }
2721 pin = 1;
2722 }
2723
2724 /* if data pin when any transaction has committed this */
2725 if (ref_generation != trans->transid)
2726 pin = 1;
2727
2728 ret = __free_extent(trans, root, bytenr, num_bytes, parent,
2729 root_objectid, ref_generation,
2730 owner_objectid, pin, pin == 0);
2731
2732 finish_current_insert(trans, root->fs_info->extent_root, 0);
2733 pending_ret = del_pending_extents(trans, root->fs_info->extent_root, 0);
2734 return ret ? ret : pending_ret;
2735 }
2736
2737 int btrfs_free_extent(struct btrfs_trans_handle *trans,
2738 struct btrfs_root *root,
2739 u64 bytenr, u64 num_bytes, u64 parent,
2740 u64 root_objectid, u64 ref_generation,
2741 u64 owner_objectid, int pin)
2742 {
2743 int ret;
2744
2745 ret = __btrfs_free_extent(trans, root, bytenr, num_bytes, parent,
2746 root_objectid, ref_generation,
2747 owner_objectid, pin);
2748 return ret;
2749 }
2750
2751 static u64 stripe_align(struct btrfs_root *root, u64 val)
2752 {
2753 u64 mask = ((u64)root->stripesize - 1);
2754 u64 ret = (val + mask) & ~mask;
2755 return ret;
2756 }
2757
2758 /*
2759 * walks the btree of allocated extents and find a hole of a given size.
2760 * The key ins is changed to record the hole:
2761 * ins->objectid == block start
2762 * ins->flags = BTRFS_EXTENT_ITEM_KEY
2763 * ins->offset == number of blocks
2764 * Any available blocks before search_start are skipped.
2765 */
2766 static int noinline find_free_extent(struct btrfs_trans_handle *trans,
2767 struct btrfs_root *orig_root,
2768 u64 num_bytes, u64 empty_size,
2769 u64 search_start, u64 search_end,
2770 u64 hint_byte, struct btrfs_key *ins,
2771 u64 exclude_start, u64 exclude_nr,
2772 int data)
2773 {
2774 int ret = 0;
2775 struct btrfs_root * root = orig_root->fs_info->extent_root;
2776 u64 total_needed = num_bytes;
2777 u64 *last_ptr = NULL;
2778 u64 last_wanted = 0;
2779 struct btrfs_block_group_cache *block_group = NULL;
2780 int chunk_alloc_done = 0;
2781 int empty_cluster = 2 * 1024 * 1024;
2782 int allowed_chunk_alloc = 0;
2783 struct list_head *head = NULL, *cur = NULL;
2784 int loop = 0;
2785 int extra_loop = 0;
2786 struct btrfs_space_info *space_info;
2787
2788 WARN_ON(num_bytes < root->sectorsize);
2789 btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
2790 ins->objectid = 0;
2791 ins->offset = 0;
2792
2793 if (orig_root->ref_cows || empty_size)
2794 allowed_chunk_alloc = 1;
2795
2796 if (data & BTRFS_BLOCK_GROUP_METADATA) {
2797 last_ptr = &root->fs_info->last_alloc;
2798 empty_cluster = 64 * 1024;
2799 }
2800
2801 if ((data & BTRFS_BLOCK_GROUP_DATA) && btrfs_test_opt(root, SSD))
2802 last_ptr = &root->fs_info->last_data_alloc;
2803
2804 if (last_ptr) {
2805 if (*last_ptr) {
2806 hint_byte = *last_ptr;
2807 last_wanted = *last_ptr;
2808 } else
2809 empty_size += empty_cluster;
2810 } else {
2811 empty_cluster = 0;
2812 }
2813 search_start = max(search_start, first_logical_byte(root, 0));
2814 search_start = max(search_start, hint_byte);
2815
2816 if (last_wanted && search_start != last_wanted) {
2817 last_wanted = 0;
2818 empty_size += empty_cluster;
2819 }
2820
2821 total_needed += empty_size;
2822 block_group = btrfs_lookup_block_group(root->fs_info, search_start);
2823 if (!block_group)
2824 block_group = btrfs_lookup_first_block_group(root->fs_info,
2825 search_start);
2826 space_info = __find_space_info(root->fs_info, data);
2827
2828 down_read(&space_info->groups_sem);
2829 while (1) {
2830 struct btrfs_free_space *free_space;
2831 /*
2832 * the only way this happens if our hint points to a block
2833 * group thats not of the proper type, while looping this
2834 * should never happen
2835 */
2836 if (empty_size)
2837 extra_loop = 1;
2838
2839 if (!block_group)
2840 goto new_group_no_lock;
2841
2842 mutex_lock(&block_group->alloc_mutex);
2843 if (unlikely(!block_group_bits(block_group, data)))
2844 goto new_group;
2845
2846 ret = cache_block_group(root, block_group);
2847 if (ret) {
2848 mutex_unlock(&block_group->alloc_mutex);
2849 break;
2850 }
2851
2852 if (block_group->ro)
2853 goto new_group;
2854
2855 free_space = btrfs_find_free_space(block_group, search_start,
2856 total_needed);
2857 if (free_space) {
2858 u64 start = block_group->key.objectid;
2859 u64 end = block_group->key.objectid +
2860 block_group->key.offset;
2861
2862 search_start = stripe_align(root, free_space->offset);
2863
2864 /* move on to the next group */
2865 if (search_start + num_bytes >= search_end)
2866 goto new_group;
2867
2868 /* move on to the next group */
2869 if (search_start + num_bytes > end)
2870 goto new_group;
2871
2872 if (last_wanted && search_start != last_wanted) {
2873 total_needed += empty_cluster;
2874 empty_size += empty_cluster;
2875 last_wanted = 0;
2876 /*
2877 * if search_start is still in this block group
2878 * then we just re-search this block group
2879 */
2880 if (search_start >= start &&
2881 search_start < end) {
2882 mutex_unlock(&block_group->alloc_mutex);
2883 continue;
2884 }
2885
2886 /* else we go to the next block group */
2887 goto new_group;
2888 }
2889
2890 if (exclude_nr > 0 &&
2891 (search_start + num_bytes > exclude_start &&
2892 search_start < exclude_start + exclude_nr)) {
2893 search_start = exclude_start + exclude_nr;
2894 /*
2895 * if search_start is still in this block group
2896 * then we just re-search this block group
2897 */
2898 if (search_start >= start &&
2899 search_start < end) {
2900 mutex_unlock(&block_group->alloc_mutex);
2901 last_wanted = 0;
2902 continue;
2903 }
2904
2905 /* else we go to the next block group */
2906 goto new_group;
2907 }
2908
2909 ins->objectid = search_start;
2910 ins->offset = num_bytes;
2911
2912 btrfs_remove_free_space_lock(block_group, search_start,
2913 num_bytes);
2914 /* we are all good, lets return */
2915 mutex_unlock(&block_group->alloc_mutex);
2916 break;
2917 }
2918 new_group:
2919 mutex_unlock(&block_group->alloc_mutex);
2920 new_group_no_lock:
2921 /* don't try to compare new allocations against the
2922 * last allocation any more
2923 */
2924 last_wanted = 0;
2925
2926 /*
2927 * Here's how this works.
2928 * loop == 0: we were searching a block group via a hint
2929 * and didn't find anything, so we start at
2930 * the head of the block groups and keep searching
2931 * loop == 1: we're searching through all of the block groups
2932 * if we hit the head again we have searched
2933 * all of the block groups for this space and we
2934 * need to try and allocate, if we cant error out.
2935 * loop == 2: we allocated more space and are looping through
2936 * all of the block groups again.
2937 */
2938 if (loop == 0) {
2939 head = &space_info->block_groups;
2940 cur = head->next;
2941 loop++;
2942 } else if (loop == 1 && cur == head) {
2943 int keep_going;
2944
2945 /* at this point we give up on the empty_size
2946 * allocations and just try to allocate the min
2947 * space.
2948 *
2949 * The extra_loop field was set if an empty_size
2950 * allocation was attempted above, and if this
2951 * is try we need to try the loop again without
2952 * the additional empty_size.
2953 */
2954 total_needed -= empty_size;
2955 empty_size = 0;
2956 keep_going = extra_loop;
2957 loop++;
2958
2959 if (allowed_chunk_alloc && !chunk_alloc_done) {
2960 up_read(&space_info->groups_sem);
2961 ret = do_chunk_alloc(trans, root, num_bytes +
2962 2 * 1024 * 1024, data, 1);
2963 down_read(&space_info->groups_sem);
2964 if (ret < 0)
2965 goto loop_check;
2966 head = &space_info->block_groups;
2967 /*
2968 * we've allocated a new chunk, keep
2969 * trying
2970 */
2971 keep_going = 1;
2972 chunk_alloc_done = 1;
2973 } else if (!allowed_chunk_alloc) {
2974 space_info->force_alloc = 1;
2975 }
2976 loop_check:
2977 if (keep_going) {
2978 cur = head->next;
2979 extra_loop = 0;
2980 } else {
2981 break;
2982 }
2983 } else if (cur == head) {
2984 break;
2985 }
2986
2987 block_group = list_entry(cur, struct btrfs_block_group_cache,
2988 list);
2989 search_start = block_group->key.objectid;
2990 cur = cur->next;
2991 }
2992
2993 /* we found what we needed */
2994 if (ins->objectid) {
2995 if (!(data & BTRFS_BLOCK_GROUP_DATA))
2996 trans->block_group = block_group;
2997
2998 if (last_ptr)
2999 *last_ptr = ins->objectid + ins->offset;
3000 ret = 0;
3001 } else if (!ret) {
3002 printk(KERN_ERR "we were searching for %Lu bytes, num_bytes %Lu,"
3003 " loop %d, allowed_alloc %d\n", total_needed, num_bytes,
3004 loop, allowed_chunk_alloc);
3005 ret = -ENOSPC;
3006 }
3007
3008 up_read(&space_info->groups_sem);
3009 return ret;
3010 }
3011
3012 static void dump_space_info(struct btrfs_space_info *info, u64 bytes)
3013 {
3014 struct btrfs_block_group_cache *cache;
3015 struct list_head *l;
3016
3017 printk(KERN_INFO "space_info has %Lu free, is %sfull\n",
3018 info->total_bytes - info->bytes_used - info->bytes_pinned -
3019 info->bytes_reserved, (info->full) ? "" : "not ");
3020
3021 down_read(&info->groups_sem);
3022 list_for_each(l, &info->block_groups) {
3023 cache = list_entry(l, struct btrfs_block_group_cache, list);
3024 spin_lock(&cache->lock);
3025 printk(KERN_INFO "block group %Lu has %Lu bytes, %Lu used "
3026 "%Lu pinned %Lu reserved\n",
3027 cache->key.objectid, cache->key.offset,
3028 btrfs_block_group_used(&cache->item),
3029 cache->pinned, cache->reserved);
3030 btrfs_dump_free_space(cache, bytes);
3031 spin_unlock(&cache->lock);
3032 }
3033 up_read(&info->groups_sem);
3034 }
3035
3036 static int __btrfs_reserve_extent(struct btrfs_trans_handle *trans,
3037 struct btrfs_root *root,
3038 u64 num_bytes, u64 min_alloc_size,
3039 u64 empty_size, u64 hint_byte,
3040 u64 search_end, struct btrfs_key *ins,
3041 u64 data)
3042 {
3043 int ret;
3044 u64 search_start = 0;
3045 u64 alloc_profile;
3046 struct btrfs_fs_info *info = root->fs_info;
3047
3048 if (data) {
3049 alloc_profile = info->avail_data_alloc_bits &
3050 info->data_alloc_profile;
3051 data = BTRFS_BLOCK_GROUP_DATA | alloc_profile;
3052 } else if (root == root->fs_info->chunk_root) {
3053 alloc_profile = info->avail_system_alloc_bits &
3054 info->system_alloc_profile;
3055 data = BTRFS_BLOCK_GROUP_SYSTEM | alloc_profile;
3056 } else {
3057 alloc_profile = info->avail_metadata_alloc_bits &
3058 info->metadata_alloc_profile;
3059 data = BTRFS_BLOCK_GROUP_METADATA | alloc_profile;
3060 }
3061 again:
3062 data = btrfs_reduce_alloc_profile(root, data);
3063 /*
3064 * the only place that sets empty_size is btrfs_realloc_node, which
3065 * is not called recursively on allocations
3066 */
3067 if (empty_size || root->ref_cows) {
3068 if (!(data & BTRFS_BLOCK_GROUP_METADATA)) {
3069 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3070 2 * 1024 * 1024,
3071 BTRFS_BLOCK_GROUP_METADATA |
3072 (info->metadata_alloc_profile &
3073 info->avail_metadata_alloc_bits), 0);
3074 }
3075 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3076 num_bytes + 2 * 1024 * 1024, data, 0);
3077 }
3078
3079 WARN_ON(num_bytes < root->sectorsize);
3080 ret = find_free_extent(trans, root, num_bytes, empty_size,
3081 search_start, search_end, hint_byte, ins,
3082 trans->alloc_exclude_start,
3083 trans->alloc_exclude_nr, data);
3084
3085 if (ret == -ENOSPC && num_bytes > min_alloc_size) {
3086 num_bytes = num_bytes >> 1;
3087 num_bytes = num_bytes & ~(root->sectorsize - 1);
3088 num_bytes = max(num_bytes, min_alloc_size);
3089 do_chunk_alloc(trans, root->fs_info->extent_root,
3090 num_bytes, data, 1);
3091 goto again;
3092 }
3093 if (ret) {
3094 struct btrfs_space_info *sinfo;
3095
3096 sinfo = __find_space_info(root->fs_info, data);
3097 printk("allocation failed flags %Lu, wanted %Lu\n",
3098 data, num_bytes);
3099 dump_space_info(sinfo, num_bytes);
3100 BUG();
3101 }
3102
3103 return ret;
3104 }
3105
3106 int btrfs_free_reserved_extent(struct btrfs_root *root, u64 start, u64 len)
3107 {
3108 struct btrfs_block_group_cache *cache;
3109
3110 cache = btrfs_lookup_block_group(root->fs_info, start);
3111 if (!cache) {
3112 printk(KERN_ERR "Unable to find block group for %Lu\n", start);
3113 return -ENOSPC;
3114 }
3115 btrfs_add_free_space(cache, start, len);
3116 update_reserved_extents(root, start, len, 0);
3117 return 0;
3118 }
3119
3120 int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
3121 struct btrfs_root *root,
3122 u64 num_bytes, u64 min_alloc_size,
3123 u64 empty_size, u64 hint_byte,
3124 u64 search_end, struct btrfs_key *ins,
3125 u64 data)
3126 {
3127 int ret;
3128 ret = __btrfs_reserve_extent(trans, root, num_bytes, min_alloc_size,
3129 empty_size, hint_byte, search_end, ins,
3130 data);
3131 update_reserved_extents(root, ins->objectid, ins->offset, 1);
3132 return ret;
3133 }
3134
3135 static int __btrfs_alloc_reserved_extent(struct btrfs_trans_handle *trans,
3136 struct btrfs_root *root, u64 parent,
3137 u64 root_objectid, u64 ref_generation,
3138 u64 owner, struct btrfs_key *ins)
3139 {
3140 int ret;
3141 int pending_ret;
3142 u64 super_used;
3143 u64 root_used;
3144 u64 num_bytes = ins->offset;
3145 u32 sizes[2];
3146 struct btrfs_fs_info *info = root->fs_info;
3147 struct btrfs_root *extent_root = info->extent_root;
3148 struct btrfs_extent_item *extent_item;
3149 struct btrfs_extent_ref *ref;
3150 struct btrfs_path *path;
3151 struct btrfs_key keys[2];
3152
3153 if (parent == 0)
3154 parent = ins->objectid;
3155
3156 /* block accounting for super block */
3157 spin_lock_irq(&info->delalloc_lock);
3158 super_used = btrfs_super_bytes_used(&info->super_copy);
3159 btrfs_set_super_bytes_used(&info->super_copy, super_used + num_bytes);
3160 spin_unlock_irq(&info->delalloc_lock);
3161
3162 /* block accounting for root item */
3163 root_used = btrfs_root_used(&root->root_item);
3164 btrfs_set_root_used(&root->root_item, root_used + num_bytes);
3165
3166 if (root == extent_root) {
3167 struct pending_extent_op *extent_op;
3168
3169 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
3170 BUG_ON(!extent_op);
3171
3172 extent_op->type = PENDING_EXTENT_INSERT;
3173 extent_op->bytenr = ins->objectid;
3174 extent_op->num_bytes = ins->offset;
3175 extent_op->parent = parent;
3176 extent_op->orig_parent = 0;
3177 extent_op->generation = ref_generation;
3178 extent_op->orig_generation = 0;
3179 extent_op->level = (int)owner;
3180 INIT_LIST_HEAD(&extent_op->list);
3181 extent_op->del = 0;
3182
3183 mutex_lock(&root->fs_info->extent_ins_mutex);
3184 set_extent_bits(&root->fs_info->extent_ins, ins->objectid,
3185 ins->objectid + ins->offset - 1,
3186 EXTENT_WRITEBACK, GFP_NOFS);
3187 set_state_private(&root->fs_info->extent_ins,
3188 ins->objectid, (unsigned long)extent_op);
3189 mutex_unlock(&root->fs_info->extent_ins_mutex);
3190 goto update_block;
3191 }
3192
3193 memcpy(&keys[0], ins, sizeof(*ins));
3194 keys[1].objectid = ins->objectid;
3195 keys[1].type = BTRFS_EXTENT_REF_KEY;
3196 keys[1].offset = parent;
3197 sizes[0] = sizeof(*extent_item);
3198 sizes[1] = sizeof(*ref);
3199
3200 path = btrfs_alloc_path();
3201 BUG_ON(!path);
3202
3203 ret = btrfs_insert_empty_items(trans, extent_root, path, keys,
3204 sizes, 2);
3205 BUG_ON(ret);
3206
3207 extent_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3208 struct btrfs_extent_item);
3209 btrfs_set_extent_refs(path->nodes[0], extent_item, 1);
3210 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
3211 struct btrfs_extent_ref);
3212
3213 btrfs_set_ref_root(path->nodes[0], ref, root_objectid);
3214 btrfs_set_ref_generation(path->nodes[0], ref, ref_generation);
3215 btrfs_set_ref_objectid(path->nodes[0], ref, owner);
3216 btrfs_set_ref_num_refs(path->nodes[0], ref, 1);
3217
3218 btrfs_mark_buffer_dirty(path->nodes[0]);
3219
3220 trans->alloc_exclude_start = 0;
3221 trans->alloc_exclude_nr = 0;
3222 btrfs_free_path(path);
3223 finish_current_insert(trans, extent_root, 0);
3224 pending_ret = del_pending_extents(trans, extent_root, 0);
3225
3226 if (ret)
3227 goto out;
3228 if (pending_ret) {
3229 ret = pending_ret;
3230 goto out;
3231 }
3232
3233 update_block:
3234 ret = update_block_group(trans, root, ins->objectid, ins->offset, 1, 0);
3235 if (ret) {
3236 printk("update block group failed for %Lu %Lu\n",
3237 ins->objectid, ins->offset);
3238 BUG();
3239 }
3240 out:
3241 return ret;
3242 }
3243
3244 int btrfs_alloc_reserved_extent(struct btrfs_trans_handle *trans,
3245 struct btrfs_root *root, u64 parent,
3246 u64 root_objectid, u64 ref_generation,
3247 u64 owner, struct btrfs_key *ins)
3248 {
3249 int ret;
3250
3251 if (root_objectid == BTRFS_TREE_LOG_OBJECTID)
3252 return 0;
3253 ret = __btrfs_alloc_reserved_extent(trans, root, parent, root_objectid,
3254 ref_generation, owner, ins);
3255 update_reserved_extents(root, ins->objectid, ins->offset, 0);
3256 return ret;
3257 }
3258
3259 /*
3260 * this is used by the tree logging recovery code. It records that
3261 * an extent has been allocated and makes sure to clear the free
3262 * space cache bits as well
3263 */
3264 int btrfs_alloc_logged_extent(struct btrfs_trans_handle *trans,
3265 struct btrfs_root *root, u64 parent,
3266 u64 root_objectid, u64 ref_generation,
3267 u64 owner, struct btrfs_key *ins)
3268 {
3269 int ret;
3270 struct btrfs_block_group_cache *block_group;
3271
3272 block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
3273 mutex_lock(&block_group->alloc_mutex);
3274 cache_block_group(root, block_group);
3275
3276 ret = btrfs_remove_free_space_lock(block_group, ins->objectid,
3277 ins->offset);
3278 mutex_unlock(&block_group->alloc_mutex);
3279 BUG_ON(ret);
3280 ret = __btrfs_alloc_reserved_extent(trans, root, parent, root_objectid,
3281 ref_generation, owner, ins);
3282 return ret;
3283 }
3284
3285 /*
3286 * finds a free extent and does all the dirty work required for allocation
3287 * returns the key for the extent through ins, and a tree buffer for
3288 * the first block of the extent through buf.
3289 *
3290 * returns 0 if everything worked, non-zero otherwise.
3291 */
3292 int btrfs_alloc_extent(struct btrfs_trans_handle *trans,
3293 struct btrfs_root *root,
3294 u64 num_bytes, u64 parent, u64 min_alloc_size,
3295 u64 root_objectid, u64 ref_generation,
3296 u64 owner_objectid, u64 empty_size, u64 hint_byte,
3297 u64 search_end, struct btrfs_key *ins, u64 data)
3298 {
3299 int ret;
3300
3301 ret = __btrfs_reserve_extent(trans, root, num_bytes,
3302 min_alloc_size, empty_size, hint_byte,
3303 search_end, ins, data);
3304 BUG_ON(ret);
3305 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
3306 ret = __btrfs_alloc_reserved_extent(trans, root, parent,
3307 root_objectid, ref_generation,
3308 owner_objectid, ins);
3309 BUG_ON(ret);
3310
3311 } else {
3312 update_reserved_extents(root, ins->objectid, ins->offset, 1);
3313 }
3314 return ret;
3315 }
3316
3317 struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans,
3318 struct btrfs_root *root,
3319 u64 bytenr, u32 blocksize)
3320 {
3321 struct extent_buffer *buf;
3322
3323 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
3324 if (!buf)
3325 return ERR_PTR(-ENOMEM);
3326 btrfs_set_header_generation(buf, trans->transid);
3327 btrfs_tree_lock(buf);
3328 clean_tree_block(trans, root, buf);
3329 btrfs_set_buffer_uptodate(buf);
3330 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
3331 set_extent_dirty(&root->dirty_log_pages, buf->start,
3332 buf->start + buf->len - 1, GFP_NOFS);
3333 } else {
3334 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
3335 buf->start + buf->len - 1, GFP_NOFS);
3336 }
3337 trans->blocks_used++;
3338 return buf;
3339 }
3340
3341 /*
3342 * helper function to allocate a block for a given tree
3343 * returns the tree buffer or NULL.
3344 */
3345 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
3346 struct btrfs_root *root,
3347 u32 blocksize, u64 parent,
3348 u64 root_objectid,
3349 u64 ref_generation,
3350 int level,
3351 u64 hint,
3352 u64 empty_size)
3353 {
3354 struct btrfs_key ins;
3355 int ret;
3356 struct extent_buffer *buf;
3357
3358 ret = btrfs_alloc_extent(trans, root, blocksize, parent, blocksize,
3359 root_objectid, ref_generation, level,
3360 empty_size, hint, (u64)-1, &ins, 0);
3361 if (ret) {
3362 BUG_ON(ret > 0);
3363 return ERR_PTR(ret);
3364 }
3365
3366 buf = btrfs_init_new_buffer(trans, root, ins.objectid, blocksize);
3367 return buf;
3368 }
3369
3370 int btrfs_drop_leaf_ref(struct btrfs_trans_handle *trans,
3371 struct btrfs_root *root, struct extent_buffer *leaf)
3372 {
3373 u64 leaf_owner;
3374 u64 leaf_generation;
3375 struct btrfs_key key;
3376 struct btrfs_file_extent_item *fi;
3377 int i;
3378 int nritems;
3379 int ret;
3380
3381 BUG_ON(!btrfs_is_leaf(leaf));
3382 nritems = btrfs_header_nritems(leaf);
3383 leaf_owner = btrfs_header_owner(leaf);
3384 leaf_generation = btrfs_header_generation(leaf);
3385
3386 for (i = 0; i < nritems; i++) {
3387 u64 disk_bytenr;
3388 cond_resched();
3389
3390 btrfs_item_key_to_cpu(leaf, &key, i);
3391 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
3392 continue;
3393 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
3394 if (btrfs_file_extent_type(leaf, fi) ==
3395 BTRFS_FILE_EXTENT_INLINE)
3396 continue;
3397 /*
3398 * FIXME make sure to insert a trans record that
3399 * repeats the snapshot del on crash
3400 */
3401 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
3402 if (disk_bytenr == 0)
3403 continue;
3404
3405 ret = __btrfs_free_extent(trans, root, disk_bytenr,
3406 btrfs_file_extent_disk_num_bytes(leaf, fi),
3407 leaf->start, leaf_owner, leaf_generation,
3408 key.objectid, 0);
3409 BUG_ON(ret);
3410
3411 atomic_inc(&root->fs_info->throttle_gen);
3412 wake_up(&root->fs_info->transaction_throttle);
3413 cond_resched();
3414 }
3415 return 0;
3416 }
3417
3418 static int noinline cache_drop_leaf_ref(struct btrfs_trans_handle *trans,
3419 struct btrfs_root *root,
3420 struct btrfs_leaf_ref *ref)
3421 {
3422 int i;
3423 int ret;
3424 struct btrfs_extent_info *info = ref->extents;
3425
3426 for (i = 0; i < ref->nritems; i++) {
3427 ret = __btrfs_free_extent(trans, root, info->bytenr,
3428 info->num_bytes, ref->bytenr,
3429 ref->owner, ref->generation,
3430 info->objectid, 0);
3431
3432 atomic_inc(&root->fs_info->throttle_gen);
3433 wake_up(&root->fs_info->transaction_throttle);
3434 cond_resched();
3435
3436 BUG_ON(ret);
3437 info++;
3438 }
3439
3440 return 0;
3441 }
3442
3443 int drop_snap_lookup_refcount(struct btrfs_root *root, u64 start, u64 len,
3444 u32 *refs)
3445 {
3446 int ret;
3447
3448 ret = btrfs_lookup_extent_ref(NULL, root, start, len, refs);
3449 BUG_ON(ret);
3450
3451 #if 0 // some debugging code in case we see problems here
3452 /* if the refs count is one, it won't get increased again. But
3453 * if the ref count is > 1, someone may be decreasing it at
3454 * the same time we are.
3455 */
3456 if (*refs != 1) {
3457 struct extent_buffer *eb = NULL;
3458 eb = btrfs_find_create_tree_block(root, start, len);
3459 if (eb)
3460 btrfs_tree_lock(eb);
3461
3462 mutex_lock(&root->fs_info->alloc_mutex);
3463 ret = lookup_extent_ref(NULL, root, start, len, refs);
3464 BUG_ON(ret);
3465 mutex_unlock(&root->fs_info->alloc_mutex);
3466
3467 if (eb) {
3468 btrfs_tree_unlock(eb);
3469 free_extent_buffer(eb);
3470 }
3471 if (*refs == 1) {
3472 printk("block %llu went down to one during drop_snap\n",
3473 (unsigned long long)start);
3474 }
3475
3476 }
3477 #endif
3478
3479 cond_resched();
3480 return ret;
3481 }
3482
3483 /*
3484 * helper function for drop_snapshot, this walks down the tree dropping ref
3485 * counts as it goes.
3486 */
3487 static int noinline walk_down_tree(struct btrfs_trans_handle *trans,
3488 struct btrfs_root *root,
3489 struct btrfs_path *path, int *level)
3490 {
3491 u64 root_owner;
3492 u64 root_gen;
3493 u64 bytenr;
3494 u64 ptr_gen;
3495 struct extent_buffer *next;
3496 struct extent_buffer *cur;
3497 struct extent_buffer *parent;
3498 struct btrfs_leaf_ref *ref;
3499 u32 blocksize;
3500 int ret;
3501 u32 refs;
3502
3503 WARN_ON(*level < 0);
3504 WARN_ON(*level >= BTRFS_MAX_LEVEL);
3505 ret = drop_snap_lookup_refcount(root, path->nodes[*level]->start,
3506 path->nodes[*level]->len, &refs);
3507 BUG_ON(ret);
3508 if (refs > 1)
3509 goto out;
3510
3511 /*
3512 * walk down to the last node level and free all the leaves
3513 */
3514 while(*level >= 0) {
3515 WARN_ON(*level < 0);
3516 WARN_ON(*level >= BTRFS_MAX_LEVEL);
3517 cur = path->nodes[*level];
3518
3519 if (btrfs_header_level(cur) != *level)
3520 WARN_ON(1);
3521
3522 if (path->slots[*level] >=
3523 btrfs_header_nritems(cur))
3524 break;
3525 if (*level == 0) {
3526 ret = btrfs_drop_leaf_ref(trans, root, cur);
3527 BUG_ON(ret);
3528 break;
3529 }
3530 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
3531 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
3532 blocksize = btrfs_level_size(root, *level - 1);
3533
3534 ret = drop_snap_lookup_refcount(root, bytenr, blocksize, &refs);
3535 BUG_ON(ret);
3536 if (refs != 1) {
3537 parent = path->nodes[*level];
3538 root_owner = btrfs_header_owner(parent);
3539 root_gen = btrfs_header_generation(parent);
3540 path->slots[*level]++;
3541
3542 ret = __btrfs_free_extent(trans, root, bytenr,
3543 blocksize, parent->start,
3544 root_owner, root_gen,
3545 *level - 1, 1);
3546 BUG_ON(ret);
3547
3548 atomic_inc(&root->fs_info->throttle_gen);
3549 wake_up(&root->fs_info->transaction_throttle);
3550 cond_resched();
3551
3552 continue;
3553 }
3554 /*
3555 * at this point, we have a single ref, and since the
3556 * only place referencing this extent is a dead root
3557 * the reference count should never go higher.
3558 * So, we don't need to check it again
3559 */
3560 if (*level == 1) {
3561 ref = btrfs_lookup_leaf_ref(root, bytenr);
3562 if (ref && ref->generation != ptr_gen) {
3563 btrfs_free_leaf_ref(root, ref);
3564 ref = NULL;
3565 }
3566 if (ref) {
3567 ret = cache_drop_leaf_ref(trans, root, ref);
3568 BUG_ON(ret);
3569 btrfs_remove_leaf_ref(root, ref);
3570 btrfs_free_leaf_ref(root, ref);
3571 *level = 0;
3572 break;
3573 }
3574 if (printk_ratelimit()) {
3575 printk("leaf ref miss for bytenr %llu\n",
3576 (unsigned long long)bytenr);
3577 }
3578 }
3579 next = btrfs_find_tree_block(root, bytenr, blocksize);
3580 if (!next || !btrfs_buffer_uptodate(next, ptr_gen)) {
3581 free_extent_buffer(next);
3582
3583 next = read_tree_block(root, bytenr, blocksize,
3584 ptr_gen);
3585 cond_resched();
3586 #if 0
3587 /*
3588 * this is a debugging check and can go away
3589 * the ref should never go all the way down to 1
3590 * at this point
3591 */
3592 ret = lookup_extent_ref(NULL, root, bytenr, blocksize,
3593 &refs);
3594 BUG_ON(ret);
3595 WARN_ON(refs != 1);
3596 #endif
3597 }
3598 WARN_ON(*level <= 0);
3599 if (path->nodes[*level-1])
3600 free_extent_buffer(path->nodes[*level-1]);
3601 path->nodes[*level-1] = next;
3602 *level = btrfs_header_level(next);
3603 path->slots[*level] = 0;
3604 cond_resched();
3605 }
3606 out:
3607 WARN_ON(*level < 0);
3608 WARN_ON(*level >= BTRFS_MAX_LEVEL);
3609
3610 if (path->nodes[*level] == root->node) {
3611 parent = path->nodes[*level];
3612 bytenr = path->nodes[*level]->start;
3613 } else {
3614 parent = path->nodes[*level + 1];
3615 bytenr = btrfs_node_blockptr(parent, path->slots[*level + 1]);
3616 }
3617
3618 blocksize = btrfs_level_size(root, *level);
3619 root_owner = btrfs_header_owner(parent);
3620 root_gen = btrfs_header_generation(parent);
3621
3622 ret = __btrfs_free_extent(trans, root, bytenr, blocksize,
3623 parent->start, root_owner, root_gen,
3624 *level, 1);
3625 free_extent_buffer(path->nodes[*level]);
3626 path->nodes[*level] = NULL;
3627 *level += 1;
3628 BUG_ON(ret);
3629
3630 cond_resched();
3631 return 0;
3632 }
3633
3634 /*
3635 * helper function for drop_subtree, this function is similar to
3636 * walk_down_tree. The main difference is that it checks reference
3637 * counts while tree blocks are locked.
3638 */
3639 static int noinline walk_down_subtree(struct btrfs_trans_handle *trans,
3640 struct btrfs_root *root,
3641 struct btrfs_path *path, int *level)
3642 {
3643 struct extent_buffer *next;
3644 struct extent_buffer *cur;
3645 struct extent_buffer *parent;
3646 u64 bytenr;
3647 u64 ptr_gen;
3648 u32 blocksize;
3649 u32 refs;
3650 int ret;
3651
3652 cur = path->nodes[*level];
3653 ret = btrfs_lookup_extent_ref(trans, root, cur->start, cur->len,
3654 &refs);
3655 BUG_ON(ret);
3656 if (refs > 1)
3657 goto out;
3658
3659 while (*level >= 0) {
3660 cur = path->nodes[*level];
3661 if (*level == 0) {
3662 ret = btrfs_drop_leaf_ref(trans, root, cur);
3663 BUG_ON(ret);
3664 clean_tree_block(trans, root, cur);
3665 break;
3666 }
3667 if (path->slots[*level] >= btrfs_header_nritems(cur)) {
3668 clean_tree_block(trans, root, cur);
3669 break;
3670 }
3671
3672 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
3673 blocksize = btrfs_level_size(root, *level - 1);
3674 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
3675
3676 next = read_tree_block(root, bytenr, blocksize, ptr_gen);
3677 btrfs_tree_lock(next);
3678
3679 ret = btrfs_lookup_extent_ref(trans, root, bytenr, blocksize,
3680 &refs);
3681 BUG_ON(ret);
3682 if (refs > 1) {
3683 parent = path->nodes[*level];
3684 ret = btrfs_free_extent(trans, root, bytenr,
3685 blocksize, parent->start,
3686 btrfs_header_owner(parent),
3687 btrfs_header_generation(parent),
3688 *level - 1, 1);
3689 BUG_ON(ret);
3690 path->slots[*level]++;
3691 btrfs_tree_unlock(next);
3692 free_extent_buffer(next);
3693 continue;
3694 }
3695
3696 *level = btrfs_header_level(next);
3697 path->nodes[*level] = next;
3698 path->slots[*level] = 0;
3699 path->locks[*level] = 1;
3700 cond_resched();
3701 }
3702 out:
3703 parent = path->nodes[*level + 1];
3704 bytenr = path->nodes[*level]->start;
3705 blocksize = path->nodes[*level]->len;
3706
3707 ret = btrfs_free_extent(trans, root, bytenr, blocksize,
3708 parent->start, btrfs_header_owner(parent),
3709 btrfs_header_generation(parent), *level, 1);
3710 BUG_ON(ret);
3711
3712 if (path->locks[*level]) {
3713 btrfs_tree_unlock(path->nodes[*level]);
3714 path->locks[*level] = 0;
3715 }
3716 free_extent_buffer(path->nodes[*level]);
3717 path->nodes[*level] = NULL;
3718 *level += 1;
3719 cond_resched();
3720 return 0;
3721 }
3722
3723 /*
3724 * helper for dropping snapshots. This walks back up the tree in the path
3725 * to find the first node higher up where we haven't yet gone through
3726 * all the slots
3727 */
3728 static int noinline walk_up_tree(struct btrfs_trans_handle *trans,
3729 struct btrfs_root *root,
3730 struct btrfs_path *path,
3731 int *level, int max_level)
3732 {
3733 u64 root_owner;
3734 u64 root_gen;
3735 struct btrfs_root_item *root_item = &root->root_item;
3736 int i;
3737 int slot;
3738 int ret;
3739
3740 for (i = *level; i < max_level && path->nodes[i]; i++) {
3741 slot = path->slots[i];
3742 if (slot < btrfs_header_nritems(path->nodes[i]) - 1) {
3743 struct extent_buffer *node;
3744 struct btrfs_disk_key disk_key;
3745 node = path->nodes[i];
3746 path->slots[i]++;
3747 *level = i;
3748 WARN_ON(*level == 0);
3749 btrfs_node_key(node, &disk_key, path->slots[i]);
3750 memcpy(&root_item->drop_progress,
3751 &disk_key, sizeof(disk_key));
3752 root_item->drop_level = i;
3753 return 0;
3754 } else {
3755 struct extent_buffer *parent;
3756 if (path->nodes[*level] == root->node)
3757 parent = path->nodes[*level];
3758 else
3759 parent = path->nodes[*level + 1];
3760
3761 root_owner = btrfs_header_owner(parent);
3762 root_gen = btrfs_header_generation(parent);
3763
3764 clean_tree_block(trans, root, path->nodes[*level]);
3765 ret = btrfs_free_extent(trans, root,
3766 path->nodes[*level]->start,
3767 path->nodes[*level]->len,
3768 parent->start, root_owner,
3769 root_gen, *level, 1);
3770 BUG_ON(ret);
3771 if (path->locks[*level]) {
3772 btrfs_tree_unlock(path->nodes[*level]);
3773 path->locks[*level] = 0;
3774 }
3775 free_extent_buffer(path->nodes[*level]);
3776 path->nodes[*level] = NULL;
3777 *level = i + 1;
3778 }
3779 }
3780 return 1;
3781 }
3782
3783 /*
3784 * drop the reference count on the tree rooted at 'snap'. This traverses
3785 * the tree freeing any blocks that have a ref count of zero after being
3786 * decremented.
3787 */
3788 int btrfs_drop_snapshot(struct btrfs_trans_handle *trans, struct btrfs_root
3789 *root)
3790 {
3791 int ret = 0;
3792 int wret;
3793 int level;
3794 struct btrfs_path *path;
3795 int i;
3796 int orig_level;
3797 struct btrfs_root_item *root_item = &root->root_item;
3798
3799 WARN_ON(!mutex_is_locked(&root->fs_info->drop_mutex));
3800 path = btrfs_alloc_path();
3801 BUG_ON(!path);
3802
3803 level = btrfs_header_level(root->node);
3804 orig_level = level;
3805 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
3806 path->nodes[level] = root->node;
3807 extent_buffer_get(root->node);
3808 path->slots[level] = 0;
3809 } else {
3810 struct btrfs_key key;
3811 struct btrfs_disk_key found_key;
3812 struct extent_buffer *node;
3813
3814 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
3815 level = root_item->drop_level;
3816 path->lowest_level = level;
3817 wret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3818 if (wret < 0) {
3819 ret = wret;
3820 goto out;
3821 }
3822 node = path->nodes[level];
3823 btrfs_node_key(node, &found_key, path->slots[level]);
3824 WARN_ON(memcmp(&found_key, &root_item->drop_progress,
3825 sizeof(found_key)));
3826 /*
3827 * unlock our path, this is safe because only this
3828 * function is allowed to delete this snapshot
3829 */
3830 for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
3831 if (path->nodes[i] && path->locks[i]) {
3832 path->locks[i] = 0;
3833 btrfs_tree_unlock(path->nodes[i]);
3834 }
3835 }
3836 }
3837 while(1) {
3838 wret = walk_down_tree(trans, root, path, &level);
3839 if (wret > 0)
3840 break;
3841 if (wret < 0)
3842 ret = wret;
3843
3844 wret = walk_up_tree(trans, root, path, &level,
3845 BTRFS_MAX_LEVEL);
3846 if (wret > 0)
3847 break;
3848 if (wret < 0)
3849 ret = wret;
3850 if (trans->transaction->in_commit) {
3851 ret = -EAGAIN;
3852 break;
3853 }
3854 atomic_inc(&root->fs_info->throttle_gen);
3855 wake_up(&root->fs_info->transaction_throttle);
3856 }
3857 for (i = 0; i <= orig_level; i++) {
3858 if (path->nodes[i]) {
3859 free_extent_buffer(path->nodes[i]);
3860 path->nodes[i] = NULL;
3861 }
3862 }
3863 out:
3864 btrfs_free_path(path);
3865 return ret;
3866 }
3867
3868 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
3869 struct btrfs_root *root,
3870 struct extent_buffer *node,
3871 struct extent_buffer *parent)
3872 {
3873 struct btrfs_path *path;
3874 int level;
3875 int parent_level;
3876 int ret = 0;
3877 int wret;
3878
3879 path = btrfs_alloc_path();
3880 BUG_ON(!path);
3881
3882 BUG_ON(!btrfs_tree_locked(parent));
3883 parent_level = btrfs_header_level(parent);
3884 extent_buffer_get(parent);
3885 path->nodes[parent_level] = parent;
3886 path->slots[parent_level] = btrfs_header_nritems(parent);
3887
3888 BUG_ON(!btrfs_tree_locked(node));
3889 level = btrfs_header_level(node);
3890 extent_buffer_get(node);
3891 path->nodes[level] = node;
3892 path->slots[level] = 0;
3893
3894 while (1) {
3895 wret = walk_down_subtree(trans, root, path, &level);
3896 if (wret < 0)
3897 ret = wret;
3898 if (wret != 0)
3899 break;
3900
3901 wret = walk_up_tree(trans, root, path, &level, parent_level);
3902 if (wret < 0)
3903 ret = wret;
3904 if (wret != 0)
3905 break;
3906 }
3907
3908 btrfs_free_path(path);
3909 return ret;
3910 }
3911
3912 static unsigned long calc_ra(unsigned long start, unsigned long last,
3913 unsigned long nr)
3914 {
3915 return min(last, start + nr - 1);
3916 }
3917
3918 static int noinline relocate_inode_pages(struct inode *inode, u64 start,
3919 u64 len)
3920 {
3921 u64 page_start;
3922 u64 page_end;
3923 unsigned long first_index;
3924 unsigned long last_index;
3925 unsigned long i;
3926 struct page *page;
3927 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3928 struct file_ra_state *ra;
3929 struct btrfs_ordered_extent *ordered;
3930 unsigned int total_read = 0;
3931 unsigned int total_dirty = 0;
3932 int ret = 0;
3933
3934 ra = kzalloc(sizeof(*ra), GFP_NOFS);
3935
3936 mutex_lock(&inode->i_mutex);
3937 first_index = start >> PAGE_CACHE_SHIFT;
3938 last_index = (start + len - 1) >> PAGE_CACHE_SHIFT;
3939
3940 /* make sure the dirty trick played by the caller work */
3941 ret = invalidate_inode_pages2_range(inode->i_mapping,
3942 first_index, last_index);
3943 if (ret)
3944 goto out_unlock;
3945
3946 file_ra_state_init(ra, inode->i_mapping);
3947
3948 for (i = first_index ; i <= last_index; i++) {
3949 if (total_read % ra->ra_pages == 0) {
3950 btrfs_force_ra(inode->i_mapping, ra, NULL, i,
3951 calc_ra(i, last_index, ra->ra_pages));
3952 }
3953 total_read++;
3954 again:
3955 if (((u64)i << PAGE_CACHE_SHIFT) > i_size_read(inode))
3956 BUG_ON(1);
3957 page = grab_cache_page(inode->i_mapping, i);
3958 if (!page) {
3959 ret = -ENOMEM;
3960 goto out_unlock;
3961 }
3962 if (!PageUptodate(page)) {
3963 btrfs_readpage(NULL, page);
3964 lock_page(page);
3965 if (!PageUptodate(page)) {
3966 unlock_page(page);
3967 page_cache_release(page);
3968 ret = -EIO;
3969 goto out_unlock;
3970 }
3971 }
3972 wait_on_page_writeback(page);
3973
3974 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
3975 page_end = page_start + PAGE_CACHE_SIZE - 1;
3976 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
3977
3978 ordered = btrfs_lookup_ordered_extent(inode, page_start);
3979 if (ordered) {
3980 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
3981 unlock_page(page);
3982 page_cache_release(page);
3983 btrfs_start_ordered_extent(inode, ordered, 1);
3984 btrfs_put_ordered_extent(ordered);
3985 goto again;
3986 }
3987 set_page_extent_mapped(page);
3988
3989 btrfs_set_extent_delalloc(inode, page_start, page_end);
3990 if (i == first_index)
3991 set_extent_bits(io_tree, page_start, page_end,
3992 EXTENT_BOUNDARY, GFP_NOFS);
3993
3994 set_page_dirty(page);
3995 total_dirty++;
3996
3997 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
3998 unlock_page(page);
3999 page_cache_release(page);
4000 }
4001
4002 out_unlock:
4003 kfree(ra);
4004 mutex_unlock(&inode->i_mutex);
4005 balance_dirty_pages_ratelimited_nr(inode->i_mapping, total_dirty);
4006 return ret;
4007 }
4008
4009 static int noinline relocate_data_extent(struct inode *reloc_inode,
4010 struct btrfs_key *extent_key,
4011 u64 offset)
4012 {
4013 struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
4014 struct extent_map_tree *em_tree = &BTRFS_I(reloc_inode)->extent_tree;
4015 struct extent_map *em;
4016 u64 start = extent_key->objectid - offset;
4017 u64 end = start + extent_key->offset - 1;
4018
4019 em = alloc_extent_map(GFP_NOFS);
4020 BUG_ON(!em || IS_ERR(em));
4021
4022 em->start = start;
4023 em->len = extent_key->offset;
4024 em->block_len = extent_key->offset;
4025 em->block_start = extent_key->objectid;
4026 em->bdev = root->fs_info->fs_devices->latest_bdev;
4027 set_bit(EXTENT_FLAG_PINNED, &em->flags);
4028
4029 /* setup extent map to cheat btrfs_readpage */
4030 lock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
4031 while (1) {
4032 int ret;
4033 spin_lock(&em_tree->lock);
4034 ret = add_extent_mapping(em_tree, em);
4035 spin_unlock(&em_tree->lock);
4036 if (ret != -EEXIST) {
4037 free_extent_map(em);
4038 break;
4039 }
4040 btrfs_drop_extent_cache(reloc_inode, start, end, 0);
4041 }
4042 unlock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
4043
4044 return relocate_inode_pages(reloc_inode, start, extent_key->offset);
4045 }
4046
4047 struct btrfs_ref_path {
4048 u64 extent_start;
4049 u64 nodes[BTRFS_MAX_LEVEL];
4050 u64 root_objectid;
4051 u64 root_generation;
4052 u64 owner_objectid;
4053 u32 num_refs;
4054 int lowest_level;
4055 int current_level;
4056 int shared_level;
4057
4058 struct btrfs_key node_keys[BTRFS_MAX_LEVEL];
4059 u64 new_nodes[BTRFS_MAX_LEVEL];
4060 };
4061
4062 struct disk_extent {
4063 u64 ram_bytes;
4064 u64 disk_bytenr;
4065 u64 disk_num_bytes;
4066 u64 offset;
4067 u64 num_bytes;
4068 u8 compression;
4069 u8 encryption;
4070 u16 other_encoding;
4071 };
4072
4073 static int is_cowonly_root(u64 root_objectid)
4074 {
4075 if (root_objectid == BTRFS_ROOT_TREE_OBJECTID ||
4076 root_objectid == BTRFS_EXTENT_TREE_OBJECTID ||
4077 root_objectid == BTRFS_CHUNK_TREE_OBJECTID ||
4078 root_objectid == BTRFS_DEV_TREE_OBJECTID ||
4079 root_objectid == BTRFS_TREE_LOG_OBJECTID)
4080 return 1;
4081 return 0;
4082 }
4083
4084 static int noinline __next_ref_path(struct btrfs_trans_handle *trans,
4085 struct btrfs_root *extent_root,
4086 struct btrfs_ref_path *ref_path,
4087 int first_time)
4088 {
4089 struct extent_buffer *leaf;
4090 struct btrfs_path *path;
4091 struct btrfs_extent_ref *ref;
4092 struct btrfs_key key;
4093 struct btrfs_key found_key;
4094 u64 bytenr;
4095 u32 nritems;
4096 int level;
4097 int ret = 1;
4098
4099 path = btrfs_alloc_path();
4100 if (!path)
4101 return -ENOMEM;
4102
4103 if (first_time) {
4104 ref_path->lowest_level = -1;
4105 ref_path->current_level = -1;
4106 ref_path->shared_level = -1;
4107 goto walk_up;
4108 }
4109 walk_down:
4110 level = ref_path->current_level - 1;
4111 while (level >= -1) {
4112 u64 parent;
4113 if (level < ref_path->lowest_level)
4114 break;
4115
4116 if (level >= 0) {
4117 bytenr = ref_path->nodes[level];
4118 } else {
4119 bytenr = ref_path->extent_start;
4120 }
4121 BUG_ON(bytenr == 0);
4122
4123 parent = ref_path->nodes[level + 1];
4124 ref_path->nodes[level + 1] = 0;
4125 ref_path->current_level = level;
4126 BUG_ON(parent == 0);
4127
4128 key.objectid = bytenr;
4129 key.offset = parent + 1;
4130 key.type = BTRFS_EXTENT_REF_KEY;
4131
4132 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
4133 if (ret < 0)
4134 goto out;
4135 BUG_ON(ret == 0);
4136
4137 leaf = path->nodes[0];
4138 nritems = btrfs_header_nritems(leaf);
4139 if (path->slots[0] >= nritems) {
4140 ret = btrfs_next_leaf(extent_root, path);
4141 if (ret < 0)
4142 goto out;
4143 if (ret > 0)
4144 goto next;
4145 leaf = path->nodes[0];
4146 }
4147
4148 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4149 if (found_key.objectid == bytenr &&
4150 found_key.type == BTRFS_EXTENT_REF_KEY) {
4151 if (level < ref_path->shared_level)
4152 ref_path->shared_level = level;
4153 goto found;
4154 }
4155 next:
4156 level--;
4157 btrfs_release_path(extent_root, path);
4158 cond_resched();
4159 }
4160 /* reached lowest level */
4161 ret = 1;
4162 goto out;
4163 walk_up:
4164 level = ref_path->current_level;
4165 while (level < BTRFS_MAX_LEVEL - 1) {
4166 u64 ref_objectid;
4167 if (level >= 0) {
4168 bytenr = ref_path->nodes[level];
4169 } else {
4170 bytenr = ref_path->extent_start;
4171 }
4172 BUG_ON(bytenr == 0);
4173
4174 key.objectid = bytenr;
4175 key.offset = 0;
4176 key.type = BTRFS_EXTENT_REF_KEY;
4177
4178 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
4179 if (ret < 0)
4180 goto out;
4181
4182 leaf = path->nodes[0];
4183 nritems = btrfs_header_nritems(leaf);
4184 if (path->slots[0] >= nritems) {
4185 ret = btrfs_next_leaf(extent_root, path);
4186 if (ret < 0)
4187 goto out;
4188 if (ret > 0) {
4189 /* the extent was freed by someone */
4190 if (ref_path->lowest_level == level)
4191 goto out;
4192 btrfs_release_path(extent_root, path);
4193 goto walk_down;
4194 }
4195 leaf = path->nodes[0];
4196 }
4197
4198 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4199 if (found_key.objectid != bytenr ||
4200 found_key.type != BTRFS_EXTENT_REF_KEY) {
4201 /* the extent was freed by someone */
4202 if (ref_path->lowest_level == level) {
4203 ret = 1;
4204 goto out;
4205 }
4206 btrfs_release_path(extent_root, path);
4207 goto walk_down;
4208 }
4209 found:
4210 ref = btrfs_item_ptr(leaf, path->slots[0],
4211 struct btrfs_extent_ref);
4212 ref_objectid = btrfs_ref_objectid(leaf, ref);
4213 if (ref_objectid < BTRFS_FIRST_FREE_OBJECTID) {
4214 if (first_time) {
4215 level = (int)ref_objectid;
4216 BUG_ON(level >= BTRFS_MAX_LEVEL);
4217 ref_path->lowest_level = level;
4218 ref_path->current_level = level;
4219 ref_path->nodes[level] = bytenr;
4220 } else {
4221 WARN_ON(ref_objectid != level);
4222 }
4223 } else {
4224 WARN_ON(level != -1);
4225 }
4226 first_time = 0;
4227
4228 if (ref_path->lowest_level == level) {
4229 ref_path->owner_objectid = ref_objectid;
4230 ref_path->num_refs = btrfs_ref_num_refs(leaf, ref);
4231 }
4232
4233 /*
4234 * the block is tree root or the block isn't in reference
4235 * counted tree.
4236 */
4237 if (found_key.objectid == found_key.offset ||
4238 is_cowonly_root(btrfs_ref_root(leaf, ref))) {
4239 ref_path->root_objectid = btrfs_ref_root(leaf, ref);
4240 ref_path->root_generation =
4241 btrfs_ref_generation(leaf, ref);
4242 if (level < 0) {
4243 /* special reference from the tree log */
4244 ref_path->nodes[0] = found_key.offset;
4245 ref_path->current_level = 0;
4246 }
4247 ret = 0;
4248 goto out;
4249 }
4250
4251 level++;
4252 BUG_ON(ref_path->nodes[level] != 0);
4253 ref_path->nodes[level] = found_key.offset;
4254 ref_path->current_level = level;
4255
4256 /*
4257 * the reference was created in the running transaction,
4258 * no need to continue walking up.
4259 */
4260 if (btrfs_ref_generation(leaf, ref) == trans->transid) {
4261 ref_path->root_objectid = btrfs_ref_root(leaf, ref);
4262 ref_path->root_generation =
4263 btrfs_ref_generation(leaf, ref);
4264 ret = 0;
4265 goto out;
4266 }
4267
4268 btrfs_release_path(extent_root, path);
4269 cond_resched();
4270 }
4271 /* reached max tree level, but no tree root found. */
4272 BUG();
4273 out:
4274 btrfs_free_path(path);
4275 return ret;
4276 }
4277
4278 static int btrfs_first_ref_path(struct btrfs_trans_handle *trans,
4279 struct btrfs_root *extent_root,
4280 struct btrfs_ref_path *ref_path,
4281 u64 extent_start)
4282 {
4283 memset(ref_path, 0, sizeof(*ref_path));
4284 ref_path->extent_start = extent_start;
4285
4286 return __next_ref_path(trans, extent_root, ref_path, 1);
4287 }
4288
4289 static int btrfs_next_ref_path(struct btrfs_trans_handle *trans,
4290 struct btrfs_root *extent_root,
4291 struct btrfs_ref_path *ref_path)
4292 {
4293 return __next_ref_path(trans, extent_root, ref_path, 0);
4294 }
4295
4296 static int noinline get_new_locations(struct inode *reloc_inode,
4297 struct btrfs_key *extent_key,
4298 u64 offset, int no_fragment,
4299 struct disk_extent **extents,
4300 int *nr_extents)
4301 {
4302 struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
4303 struct btrfs_path *path;
4304 struct btrfs_file_extent_item *fi;
4305 struct extent_buffer *leaf;
4306 struct disk_extent *exts = *extents;
4307 struct btrfs_key found_key;
4308 u64 cur_pos;
4309 u64 last_byte;
4310 u32 nritems;
4311 int nr = 0;
4312 int max = *nr_extents;
4313 int ret;
4314
4315 WARN_ON(!no_fragment && *extents);
4316 if (!exts) {
4317 max = 1;
4318 exts = kmalloc(sizeof(*exts) * max, GFP_NOFS);
4319 if (!exts)
4320 return -ENOMEM;
4321 }
4322
4323 path = btrfs_alloc_path();
4324 BUG_ON(!path);
4325
4326 cur_pos = extent_key->objectid - offset;
4327 last_byte = extent_key->objectid + extent_key->offset;
4328 ret = btrfs_lookup_file_extent(NULL, root, path, reloc_inode->i_ino,
4329 cur_pos, 0);
4330 if (ret < 0)
4331 goto out;
4332 if (ret > 0) {
4333 ret = -ENOENT;
4334 goto out;
4335 }
4336
4337 while (1) {
4338 leaf = path->nodes[0];
4339 nritems = btrfs_header_nritems(leaf);
4340 if (path->slots[0] >= nritems) {
4341 ret = btrfs_next_leaf(root, path);
4342 if (ret < 0)
4343 goto out;
4344 if (ret > 0)
4345 break;
4346 leaf = path->nodes[0];
4347 }
4348
4349 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4350 if (found_key.offset != cur_pos ||
4351 found_key.type != BTRFS_EXTENT_DATA_KEY ||
4352 found_key.objectid != reloc_inode->i_ino)
4353 break;
4354
4355 fi = btrfs_item_ptr(leaf, path->slots[0],
4356 struct btrfs_file_extent_item);
4357 if (btrfs_file_extent_type(leaf, fi) !=
4358 BTRFS_FILE_EXTENT_REG ||
4359 btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
4360 break;
4361
4362 if (nr == max) {
4363 struct disk_extent *old = exts;
4364 max *= 2;
4365 exts = kzalloc(sizeof(*exts) * max, GFP_NOFS);
4366 memcpy(exts, old, sizeof(*exts) * nr);
4367 if (old != *extents)
4368 kfree(old);
4369 }
4370
4371 exts[nr].disk_bytenr =
4372 btrfs_file_extent_disk_bytenr(leaf, fi);
4373 exts[nr].disk_num_bytes =
4374 btrfs_file_extent_disk_num_bytes(leaf, fi);
4375 exts[nr].offset = btrfs_file_extent_offset(leaf, fi);
4376 exts[nr].num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
4377 exts[nr].ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
4378 exts[nr].compression = btrfs_file_extent_compression(leaf, fi);
4379 exts[nr].encryption = btrfs_file_extent_encryption(leaf, fi);
4380 exts[nr].other_encoding = btrfs_file_extent_other_encoding(leaf,
4381 fi);
4382 BUG_ON(exts[nr].offset > 0);
4383 BUG_ON(exts[nr].compression || exts[nr].encryption);
4384 BUG_ON(exts[nr].num_bytes != exts[nr].disk_num_bytes);
4385
4386 cur_pos += exts[nr].num_bytes;
4387 nr++;
4388
4389 if (cur_pos + offset >= last_byte)
4390 break;
4391
4392 if (no_fragment) {
4393 ret = 1;
4394 goto out;
4395 }
4396 path->slots[0]++;
4397 }
4398
4399 WARN_ON(cur_pos + offset > last_byte);
4400 if (cur_pos + offset < last_byte) {
4401 ret = -ENOENT;
4402 goto out;
4403 }
4404 ret = 0;
4405 out:
4406 btrfs_free_path(path);
4407 if (ret) {
4408 if (exts != *extents)
4409 kfree(exts);
4410 } else {
4411 *extents = exts;
4412 *nr_extents = nr;
4413 }
4414 return ret;
4415 }
4416
4417 static int noinline replace_one_extent(struct btrfs_trans_handle *trans,
4418 struct btrfs_root *root,
4419 struct btrfs_path *path,
4420 struct btrfs_key *extent_key,
4421 struct btrfs_key *leaf_key,
4422 struct btrfs_ref_path *ref_path,
4423 struct disk_extent *new_extents,
4424 int nr_extents)
4425 {
4426 struct extent_buffer *leaf;
4427 struct btrfs_file_extent_item *fi;
4428 struct inode *inode = NULL;
4429 struct btrfs_key key;
4430 u64 lock_start = 0;
4431 u64 lock_end = 0;
4432 u64 num_bytes;
4433 u64 ext_offset;
4434 u64 first_pos;
4435 u32 nritems;
4436 int nr_scaned = 0;
4437 int extent_locked = 0;
4438 int extent_type;
4439 int ret;
4440
4441 memcpy(&key, leaf_key, sizeof(key));
4442 first_pos = INT_LIMIT(loff_t) - extent_key->offset;
4443 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
4444 if (key.objectid < ref_path->owner_objectid ||
4445 (key.objectid == ref_path->owner_objectid &&
4446 key.type < BTRFS_EXTENT_DATA_KEY)) {
4447 key.objectid = ref_path->owner_objectid;
4448 key.type = BTRFS_EXTENT_DATA_KEY;
4449 key.offset = 0;
4450 }
4451 }
4452
4453 while (1) {
4454 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
4455 if (ret < 0)
4456 goto out;
4457
4458 leaf = path->nodes[0];
4459 nritems = btrfs_header_nritems(leaf);
4460 next:
4461 if (extent_locked && ret > 0) {
4462 /*
4463 * the file extent item was modified by someone
4464 * before the extent got locked.
4465 */
4466 unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
4467 lock_end, GFP_NOFS);
4468 extent_locked = 0;
4469 }
4470
4471 if (path->slots[0] >= nritems) {
4472 if (++nr_scaned > 2)
4473 break;
4474
4475 BUG_ON(extent_locked);
4476 ret = btrfs_next_leaf(root, path);
4477 if (ret < 0)
4478 goto out;
4479 if (ret > 0)
4480 break;
4481 leaf = path->nodes[0];
4482 nritems = btrfs_header_nritems(leaf);
4483 }
4484
4485 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4486
4487 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
4488 if ((key.objectid > ref_path->owner_objectid) ||
4489 (key.objectid == ref_path->owner_objectid &&
4490 key.type > BTRFS_EXTENT_DATA_KEY) ||
4491 (key.offset >= first_pos + extent_key->offset))
4492 break;
4493 }
4494
4495 if (inode && key.objectid != inode->i_ino) {
4496 BUG_ON(extent_locked);
4497 btrfs_release_path(root, path);
4498 mutex_unlock(&inode->i_mutex);
4499 iput(inode);
4500 inode = NULL;
4501 continue;
4502 }
4503
4504 if (key.type != BTRFS_EXTENT_DATA_KEY) {
4505 path->slots[0]++;
4506 ret = 1;
4507 goto next;
4508 }
4509 fi = btrfs_item_ptr(leaf, path->slots[0],
4510 struct btrfs_file_extent_item);
4511 extent_type = btrfs_file_extent_type(leaf, fi);
4512 if ((extent_type != BTRFS_FILE_EXTENT_REG &&
4513 extent_type != BTRFS_FILE_EXTENT_PREALLOC) ||
4514 (btrfs_file_extent_disk_bytenr(leaf, fi) !=
4515 extent_key->objectid)) {
4516 path->slots[0]++;
4517 ret = 1;
4518 goto next;
4519 }
4520
4521 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
4522 ext_offset = btrfs_file_extent_offset(leaf, fi);
4523
4524 if (first_pos > key.offset - ext_offset)
4525 first_pos = key.offset - ext_offset;
4526
4527 if (!extent_locked) {
4528 lock_start = key.offset;
4529 lock_end = lock_start + num_bytes - 1;
4530 } else {
4531 if (lock_start > key.offset ||
4532 lock_end + 1 < key.offset + num_bytes) {
4533 unlock_extent(&BTRFS_I(inode)->io_tree,
4534 lock_start, lock_end, GFP_NOFS);
4535 extent_locked = 0;
4536 }
4537 }
4538
4539 if (!inode) {
4540 btrfs_release_path(root, path);
4541
4542 inode = btrfs_iget_locked(root->fs_info->sb,
4543 key.objectid, root);
4544 if (inode->i_state & I_NEW) {
4545 BTRFS_I(inode)->root = root;
4546 BTRFS_I(inode)->location.objectid =
4547 key.objectid;
4548 BTRFS_I(inode)->location.type =
4549 BTRFS_INODE_ITEM_KEY;
4550 BTRFS_I(inode)->location.offset = 0;
4551 btrfs_read_locked_inode(inode);
4552 unlock_new_inode(inode);
4553 }
4554 /*
4555 * some code call btrfs_commit_transaction while
4556 * holding the i_mutex, so we can't use mutex_lock
4557 * here.
4558 */
4559 if (is_bad_inode(inode) ||
4560 !mutex_trylock(&inode->i_mutex)) {
4561 iput(inode);
4562 inode = NULL;
4563 key.offset = (u64)-1;
4564 goto skip;
4565 }
4566 }
4567
4568 if (!extent_locked) {
4569 struct btrfs_ordered_extent *ordered;
4570
4571 btrfs_release_path(root, path);
4572
4573 lock_extent(&BTRFS_I(inode)->io_tree, lock_start,
4574 lock_end, GFP_NOFS);
4575 ordered = btrfs_lookup_first_ordered_extent(inode,
4576 lock_end);
4577 if (ordered &&
4578 ordered->file_offset <= lock_end &&
4579 ordered->file_offset + ordered->len > lock_start) {
4580 unlock_extent(&BTRFS_I(inode)->io_tree,
4581 lock_start, lock_end, GFP_NOFS);
4582 btrfs_start_ordered_extent(inode, ordered, 1);
4583 btrfs_put_ordered_extent(ordered);
4584 key.offset += num_bytes;
4585 goto skip;
4586 }
4587 if (ordered)
4588 btrfs_put_ordered_extent(ordered);
4589
4590 extent_locked = 1;
4591 continue;
4592 }
4593
4594 if (nr_extents == 1) {
4595 /* update extent pointer in place */
4596 btrfs_set_file_extent_disk_bytenr(leaf, fi,
4597 new_extents[0].disk_bytenr);
4598 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
4599 new_extents[0].disk_num_bytes);
4600 btrfs_mark_buffer_dirty(leaf);
4601
4602 btrfs_drop_extent_cache(inode, key.offset,
4603 key.offset + num_bytes - 1, 0);
4604
4605 ret = btrfs_inc_extent_ref(trans, root,
4606 new_extents[0].disk_bytenr,
4607 new_extents[0].disk_num_bytes,
4608 leaf->start,
4609 root->root_key.objectid,
4610 trans->transid,
4611 key.objectid);
4612 BUG_ON(ret);
4613
4614 ret = btrfs_free_extent(trans, root,
4615 extent_key->objectid,
4616 extent_key->offset,
4617 leaf->start,
4618 btrfs_header_owner(leaf),
4619 btrfs_header_generation(leaf),
4620 key.objectid, 0);
4621 BUG_ON(ret);
4622
4623 btrfs_release_path(root, path);
4624 key.offset += num_bytes;
4625 } else {
4626 BUG_ON(1);
4627 #if 0
4628 u64 alloc_hint;
4629 u64 extent_len;
4630 int i;
4631 /*
4632 * drop old extent pointer at first, then insert the
4633 * new pointers one bye one
4634 */
4635 btrfs_release_path(root, path);
4636 ret = btrfs_drop_extents(trans, root, inode, key.offset,
4637 key.offset + num_bytes,
4638 key.offset, &alloc_hint);
4639 BUG_ON(ret);
4640
4641 for (i = 0; i < nr_extents; i++) {
4642 if (ext_offset >= new_extents[i].num_bytes) {
4643 ext_offset -= new_extents[i].num_bytes;
4644 continue;
4645 }
4646 extent_len = min(new_extents[i].num_bytes -
4647 ext_offset, num_bytes);
4648
4649 ret = btrfs_insert_empty_item(trans, root,
4650 path, &key,
4651 sizeof(*fi));
4652 BUG_ON(ret);
4653
4654 leaf = path->nodes[0];
4655 fi = btrfs_item_ptr(leaf, path->slots[0],
4656 struct btrfs_file_extent_item);
4657 btrfs_set_file_extent_generation(leaf, fi,
4658 trans->transid);
4659 btrfs_set_file_extent_type(leaf, fi,
4660 BTRFS_FILE_EXTENT_REG);
4661 btrfs_set_file_extent_disk_bytenr(leaf, fi,
4662 new_extents[i].disk_bytenr);
4663 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
4664 new_extents[i].disk_num_bytes);
4665 btrfs_set_file_extent_ram_bytes(leaf, fi,
4666 new_extents[i].ram_bytes);
4667
4668 btrfs_set_file_extent_compression(leaf, fi,
4669 new_extents[i].compression);
4670 btrfs_set_file_extent_encryption(leaf, fi,
4671 new_extents[i].encryption);
4672 btrfs_set_file_extent_other_encoding(leaf, fi,
4673 new_extents[i].other_encoding);
4674
4675 btrfs_set_file_extent_num_bytes(leaf, fi,
4676 extent_len);
4677 ext_offset += new_extents[i].offset;
4678 btrfs_set_file_extent_offset(leaf, fi,
4679 ext_offset);
4680 btrfs_mark_buffer_dirty(leaf);
4681
4682 btrfs_drop_extent_cache(inode, key.offset,
4683 key.offset + extent_len - 1, 0);
4684
4685 ret = btrfs_inc_extent_ref(trans, root,
4686 new_extents[i].disk_bytenr,
4687 new_extents[i].disk_num_bytes,
4688 leaf->start,
4689 root->root_key.objectid,
4690 trans->transid, key.objectid);
4691 BUG_ON(ret);
4692 btrfs_release_path(root, path);
4693
4694 inode_add_bytes(inode, extent_len);
4695
4696 ext_offset = 0;
4697 num_bytes -= extent_len;
4698 key.offset += extent_len;
4699
4700 if (num_bytes == 0)
4701 break;
4702 }
4703 BUG_ON(i >= nr_extents);
4704 #endif
4705 }
4706
4707 if (extent_locked) {
4708 unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
4709 lock_end, GFP_NOFS);
4710 extent_locked = 0;
4711 }
4712 skip:
4713 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS &&
4714 key.offset >= first_pos + extent_key->offset)
4715 break;
4716
4717 cond_resched();
4718 }
4719 ret = 0;
4720 out:
4721 btrfs_release_path(root, path);
4722 if (inode) {
4723 mutex_unlock(&inode->i_mutex);
4724 if (extent_locked) {
4725 unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
4726 lock_end, GFP_NOFS);
4727 }
4728 iput(inode);
4729 }
4730 return ret;
4731 }
4732
4733 int btrfs_reloc_tree_cache_ref(struct btrfs_trans_handle *trans,
4734 struct btrfs_root *root,
4735 struct extent_buffer *buf, u64 orig_start)
4736 {
4737 int level;
4738 int ret;
4739
4740 BUG_ON(btrfs_header_generation(buf) != trans->transid);
4741 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
4742
4743 level = btrfs_header_level(buf);
4744 if (level == 0) {
4745 struct btrfs_leaf_ref *ref;
4746 struct btrfs_leaf_ref *orig_ref;
4747
4748 orig_ref = btrfs_lookup_leaf_ref(root, orig_start);
4749 if (!orig_ref)
4750 return -ENOENT;
4751
4752 ref = btrfs_alloc_leaf_ref(root, orig_ref->nritems);
4753 if (!ref) {
4754 btrfs_free_leaf_ref(root, orig_ref);
4755 return -ENOMEM;
4756 }
4757
4758 ref->nritems = orig_ref->nritems;
4759 memcpy(ref->extents, orig_ref->extents,
4760 sizeof(ref->extents[0]) * ref->nritems);
4761
4762 btrfs_free_leaf_ref(root, orig_ref);
4763
4764 ref->root_gen = trans->transid;
4765 ref->bytenr = buf->start;
4766 ref->owner = btrfs_header_owner(buf);
4767 ref->generation = btrfs_header_generation(buf);
4768 ret = btrfs_add_leaf_ref(root, ref, 0);
4769 WARN_ON(ret);
4770 btrfs_free_leaf_ref(root, ref);
4771 }
4772 return 0;
4773 }
4774
4775 static int noinline invalidate_extent_cache(struct btrfs_root *root,
4776 struct extent_buffer *leaf,
4777 struct btrfs_block_group_cache *group,
4778 struct btrfs_root *target_root)
4779 {
4780 struct btrfs_key key;
4781 struct inode *inode = NULL;
4782 struct btrfs_file_extent_item *fi;
4783 u64 num_bytes;
4784 u64 skip_objectid = 0;
4785 u32 nritems;
4786 u32 i;
4787
4788 nritems = btrfs_header_nritems(leaf);
4789 for (i = 0; i < nritems; i++) {
4790 btrfs_item_key_to_cpu(leaf, &key, i);
4791 if (key.objectid == skip_objectid ||
4792 key.type != BTRFS_EXTENT_DATA_KEY)
4793 continue;
4794 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
4795 if (btrfs_file_extent_type(leaf, fi) ==
4796 BTRFS_FILE_EXTENT_INLINE)
4797 continue;
4798 if (btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
4799 continue;
4800 if (!inode || inode->i_ino != key.objectid) {
4801 iput(inode);
4802 inode = btrfs_ilookup(target_root->fs_info->sb,
4803 key.objectid, target_root, 1);
4804 }
4805 if (!inode) {
4806 skip_objectid = key.objectid;
4807 continue;
4808 }
4809 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
4810
4811 lock_extent(&BTRFS_I(inode)->io_tree, key.offset,
4812 key.offset + num_bytes - 1, GFP_NOFS);
4813 btrfs_drop_extent_cache(inode, key.offset,
4814 key.offset + num_bytes - 1, 1);
4815 unlock_extent(&BTRFS_I(inode)->io_tree, key.offset,
4816 key.offset + num_bytes - 1, GFP_NOFS);
4817 cond_resched();
4818 }
4819 iput(inode);
4820 return 0;
4821 }
4822
4823 static int noinline replace_extents_in_leaf(struct btrfs_trans_handle *trans,
4824 struct btrfs_root *root,
4825 struct extent_buffer *leaf,
4826 struct btrfs_block_group_cache *group,
4827 struct inode *reloc_inode)
4828 {
4829 struct btrfs_key key;
4830 struct btrfs_key extent_key;
4831 struct btrfs_file_extent_item *fi;
4832 struct btrfs_leaf_ref *ref;
4833 struct disk_extent *new_extent;
4834 u64 bytenr;
4835 u64 num_bytes;
4836 u32 nritems;
4837 u32 i;
4838 int ext_index;
4839 int nr_extent;
4840 int ret;
4841
4842 new_extent = kmalloc(sizeof(*new_extent), GFP_NOFS);
4843 BUG_ON(!new_extent);
4844
4845 ref = btrfs_lookup_leaf_ref(root, leaf->start);
4846 BUG_ON(!ref);
4847
4848 ext_index = -1;
4849 nritems = btrfs_header_nritems(leaf);
4850 for (i = 0; i < nritems; i++) {
4851 btrfs_item_key_to_cpu(leaf, &key, i);
4852 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
4853 continue;
4854 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
4855 if (btrfs_file_extent_type(leaf, fi) ==
4856 BTRFS_FILE_EXTENT_INLINE)
4857 continue;
4858 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
4859 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
4860 if (bytenr == 0)
4861 continue;
4862
4863 ext_index++;
4864 if (bytenr >= group->key.objectid + group->key.offset ||
4865 bytenr + num_bytes <= group->key.objectid)
4866 continue;
4867
4868 extent_key.objectid = bytenr;
4869 extent_key.offset = num_bytes;
4870 extent_key.type = BTRFS_EXTENT_ITEM_KEY;
4871 nr_extent = 1;
4872 ret = get_new_locations(reloc_inode, &extent_key,
4873 group->key.objectid, 1,
4874 &new_extent, &nr_extent);
4875 if (ret > 0)
4876 continue;
4877 BUG_ON(ret < 0);
4878
4879 BUG_ON(ref->extents[ext_index].bytenr != bytenr);
4880 BUG_ON(ref->extents[ext_index].num_bytes != num_bytes);
4881 ref->extents[ext_index].bytenr = new_extent->disk_bytenr;
4882 ref->extents[ext_index].num_bytes = new_extent->disk_num_bytes;
4883
4884 btrfs_set_file_extent_disk_bytenr(leaf, fi,
4885 new_extent->disk_bytenr);
4886 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
4887 new_extent->disk_num_bytes);
4888 btrfs_mark_buffer_dirty(leaf);
4889
4890 ret = btrfs_inc_extent_ref(trans, root,
4891 new_extent->disk_bytenr,
4892 new_extent->disk_num_bytes,
4893 leaf->start,
4894 root->root_key.objectid,
4895 trans->transid, key.objectid);
4896 BUG_ON(ret);
4897 ret = btrfs_free_extent(trans, root,
4898 bytenr, num_bytes, leaf->start,
4899 btrfs_header_owner(leaf),
4900 btrfs_header_generation(leaf),
4901 key.objectid, 0);
4902 BUG_ON(ret);
4903 cond_resched();
4904 }
4905 kfree(new_extent);
4906 BUG_ON(ext_index + 1 != ref->nritems);
4907 btrfs_free_leaf_ref(root, ref);
4908 return 0;
4909 }
4910
4911 int btrfs_free_reloc_root(struct btrfs_trans_handle *trans,
4912 struct btrfs_root *root)
4913 {
4914 struct btrfs_root *reloc_root;
4915 int ret;
4916
4917 if (root->reloc_root) {
4918 reloc_root = root->reloc_root;
4919 root->reloc_root = NULL;
4920 list_add(&reloc_root->dead_list,
4921 &root->fs_info->dead_reloc_roots);
4922
4923 btrfs_set_root_bytenr(&reloc_root->root_item,
4924 reloc_root->node->start);
4925 btrfs_set_root_level(&root->root_item,
4926 btrfs_header_level(reloc_root->node));
4927 memset(&reloc_root->root_item.drop_progress, 0,
4928 sizeof(struct btrfs_disk_key));
4929 reloc_root->root_item.drop_level = 0;
4930
4931 ret = btrfs_update_root(trans, root->fs_info->tree_root,
4932 &reloc_root->root_key,
4933 &reloc_root->root_item);
4934 BUG_ON(ret);
4935 }
4936 return 0;
4937 }
4938
4939 int btrfs_drop_dead_reloc_roots(struct btrfs_root *root)
4940 {
4941 struct btrfs_trans_handle *trans;
4942 struct btrfs_root *reloc_root;
4943 struct btrfs_root *prev_root = NULL;
4944 struct list_head dead_roots;
4945 int ret;
4946 unsigned long nr;
4947
4948 INIT_LIST_HEAD(&dead_roots);
4949 list_splice_init(&root->fs_info->dead_reloc_roots, &dead_roots);
4950
4951 while (!list_empty(&dead_roots)) {
4952 reloc_root = list_entry(dead_roots.prev,
4953 struct btrfs_root, dead_list);
4954 list_del_init(&reloc_root->dead_list);
4955
4956 BUG_ON(reloc_root->commit_root != NULL);
4957 while (1) {
4958 trans = btrfs_join_transaction(root, 1);
4959 BUG_ON(!trans);
4960
4961 mutex_lock(&root->fs_info->drop_mutex);
4962 ret = btrfs_drop_snapshot(trans, reloc_root);
4963 if (ret != -EAGAIN)
4964 break;
4965 mutex_unlock(&root->fs_info->drop_mutex);
4966
4967 nr = trans->blocks_used;
4968 ret = btrfs_end_transaction(trans, root);
4969 BUG_ON(ret);
4970 btrfs_btree_balance_dirty(root, nr);
4971 }
4972
4973 free_extent_buffer(reloc_root->node);
4974
4975 ret = btrfs_del_root(trans, root->fs_info->tree_root,
4976 &reloc_root->root_key);
4977 BUG_ON(ret);
4978 mutex_unlock(&root->fs_info->drop_mutex);
4979
4980 nr = trans->blocks_used;
4981 ret = btrfs_end_transaction(trans, root);
4982 BUG_ON(ret);
4983 btrfs_btree_balance_dirty(root, nr);
4984
4985 kfree(prev_root);
4986 prev_root = reloc_root;
4987 }
4988 if (prev_root) {
4989 btrfs_remove_leaf_refs(prev_root, (u64)-1, 0);
4990 kfree(prev_root);
4991 }
4992 return 0;
4993 }
4994
4995 int btrfs_add_dead_reloc_root(struct btrfs_root *root)
4996 {
4997 list_add(&root->dead_list, &root->fs_info->dead_reloc_roots);
4998 return 0;
4999 }
5000
5001 int btrfs_cleanup_reloc_trees(struct btrfs_root *root)
5002 {
5003 struct btrfs_root *reloc_root;
5004 struct btrfs_trans_handle *trans;
5005 struct btrfs_key location;
5006 int found;
5007 int ret;
5008
5009 mutex_lock(&root->fs_info->tree_reloc_mutex);
5010 ret = btrfs_find_dead_roots(root, BTRFS_TREE_RELOC_OBJECTID, NULL);
5011 BUG_ON(ret);
5012 found = !list_empty(&root->fs_info->dead_reloc_roots);
5013 mutex_unlock(&root->fs_info->tree_reloc_mutex);
5014
5015 if (found) {
5016 trans = btrfs_start_transaction(root, 1);
5017 BUG_ON(!trans);
5018 ret = btrfs_commit_transaction(trans, root);
5019 BUG_ON(ret);
5020 }
5021
5022 location.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
5023 location.offset = (u64)-1;
5024 location.type = BTRFS_ROOT_ITEM_KEY;
5025
5026 reloc_root = btrfs_read_fs_root_no_name(root->fs_info, &location);
5027 BUG_ON(!reloc_root);
5028 btrfs_orphan_cleanup(reloc_root);
5029 return 0;
5030 }
5031
5032 static int noinline init_reloc_tree(struct btrfs_trans_handle *trans,
5033 struct btrfs_root *root)
5034 {
5035 struct btrfs_root *reloc_root;
5036 struct extent_buffer *eb;
5037 struct btrfs_root_item *root_item;
5038 struct btrfs_key root_key;
5039 int ret;
5040
5041 BUG_ON(!root->ref_cows);
5042 if (root->reloc_root)
5043 return 0;
5044
5045 root_item = kmalloc(sizeof(*root_item), GFP_NOFS);
5046 BUG_ON(!root_item);
5047
5048 ret = btrfs_copy_root(trans, root, root->commit_root,
5049 &eb, BTRFS_TREE_RELOC_OBJECTID);
5050 BUG_ON(ret);
5051
5052 root_key.objectid = BTRFS_TREE_RELOC_OBJECTID;
5053 root_key.offset = root->root_key.objectid;
5054 root_key.type = BTRFS_ROOT_ITEM_KEY;
5055
5056 memcpy(root_item, &root->root_item, sizeof(root_item));
5057 btrfs_set_root_refs(root_item, 0);
5058 btrfs_set_root_bytenr(root_item, eb->start);
5059 btrfs_set_root_level(root_item, btrfs_header_level(eb));
5060 btrfs_set_root_generation(root_item, trans->transid);
5061
5062 btrfs_tree_unlock(eb);
5063 free_extent_buffer(eb);
5064
5065 ret = btrfs_insert_root(trans, root->fs_info->tree_root,
5066 &root_key, root_item);
5067 BUG_ON(ret);
5068 kfree(root_item);
5069
5070 reloc_root = btrfs_read_fs_root_no_radix(root->fs_info->tree_root,
5071 &root_key);
5072 BUG_ON(!reloc_root);
5073 reloc_root->last_trans = trans->transid;
5074 reloc_root->commit_root = NULL;
5075 reloc_root->ref_tree = &root->fs_info->reloc_ref_tree;
5076
5077 root->reloc_root = reloc_root;
5078 return 0;
5079 }
5080
5081 /*
5082 * Core function of space balance.
5083 *
5084 * The idea is using reloc trees to relocate tree blocks in reference
5085 * counted roots. There is one reloc tree for each subvol, and all
5086 * reloc trees share same root key objectid. Reloc trees are snapshots
5087 * of the latest committed roots of subvols (root->commit_root).
5088 *
5089 * To relocate a tree block referenced by a subvol, there are two steps.
5090 * COW the block through subvol's reloc tree, then update block pointer
5091 * in the subvol to point to the new block. Since all reloc trees share
5092 * same root key objectid, doing special handing for tree blocks owned
5093 * by them is easy. Once a tree block has been COWed in one reloc tree,
5094 * we can use the resulting new block directly when the same block is
5095 * required to COW again through other reloc trees. By this way, relocated
5096 * tree blocks are shared between reloc trees, so they are also shared
5097 * between subvols.
5098 */
5099 static int noinline relocate_one_path(struct btrfs_trans_handle *trans,
5100 struct btrfs_root *root,
5101 struct btrfs_path *path,
5102 struct btrfs_key *first_key,
5103 struct btrfs_ref_path *ref_path,
5104 struct btrfs_block_group_cache *group,
5105 struct inode *reloc_inode)
5106 {
5107 struct btrfs_root *reloc_root;
5108 struct extent_buffer *eb = NULL;
5109 struct btrfs_key *keys;
5110 u64 *nodes;
5111 int level;
5112 int shared_level;
5113 int lowest_level = 0;
5114 int ret;
5115
5116 if (ref_path->owner_objectid < BTRFS_FIRST_FREE_OBJECTID)
5117 lowest_level = ref_path->owner_objectid;
5118
5119 if (!root->ref_cows) {
5120 path->lowest_level = lowest_level;
5121 ret = btrfs_search_slot(trans, root, first_key, path, 0, 1);
5122 BUG_ON(ret < 0);
5123 path->lowest_level = 0;
5124 btrfs_release_path(root, path);
5125 return 0;
5126 }
5127
5128 mutex_lock(&root->fs_info->tree_reloc_mutex);
5129 ret = init_reloc_tree(trans, root);
5130 BUG_ON(ret);
5131 reloc_root = root->reloc_root;
5132
5133 shared_level = ref_path->shared_level;
5134 ref_path->shared_level = BTRFS_MAX_LEVEL - 1;
5135
5136 keys = ref_path->node_keys;
5137 nodes = ref_path->new_nodes;
5138 memset(&keys[shared_level + 1], 0,
5139 sizeof(*keys) * (BTRFS_MAX_LEVEL - shared_level - 1));
5140 memset(&nodes[shared_level + 1], 0,
5141 sizeof(*nodes) * (BTRFS_MAX_LEVEL - shared_level - 1));
5142
5143 if (nodes[lowest_level] == 0) {
5144 path->lowest_level = lowest_level;
5145 ret = btrfs_search_slot(trans, reloc_root, first_key, path,
5146 0, 1);
5147 BUG_ON(ret);
5148 for (level = lowest_level; level < BTRFS_MAX_LEVEL; level++) {
5149 eb = path->nodes[level];
5150 if (!eb || eb == reloc_root->node)
5151 break;
5152 nodes[level] = eb->start;
5153 if (level == 0)
5154 btrfs_item_key_to_cpu(eb, &keys[level], 0);
5155 else
5156 btrfs_node_key_to_cpu(eb, &keys[level], 0);
5157 }
5158 if (nodes[0] &&
5159 ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
5160 eb = path->nodes[0];
5161 ret = replace_extents_in_leaf(trans, reloc_root, eb,
5162 group, reloc_inode);
5163 BUG_ON(ret);
5164 }
5165 btrfs_release_path(reloc_root, path);
5166 } else {
5167 ret = btrfs_merge_path(trans, reloc_root, keys, nodes,
5168 lowest_level);
5169 BUG_ON(ret);
5170 }
5171
5172 /*
5173 * replace tree blocks in the fs tree with tree blocks in
5174 * the reloc tree.
5175 */
5176 ret = btrfs_merge_path(trans, root, keys, nodes, lowest_level);
5177 BUG_ON(ret < 0);
5178
5179 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
5180 ret = btrfs_search_slot(trans, reloc_root, first_key, path,
5181 0, 0);
5182 BUG_ON(ret);
5183 extent_buffer_get(path->nodes[0]);
5184 eb = path->nodes[0];
5185 btrfs_release_path(reloc_root, path);
5186 ret = invalidate_extent_cache(reloc_root, eb, group, root);
5187 BUG_ON(ret);
5188 free_extent_buffer(eb);
5189 }
5190
5191 mutex_unlock(&root->fs_info->tree_reloc_mutex);
5192 path->lowest_level = 0;
5193 return 0;
5194 }
5195
5196 static int noinline relocate_tree_block(struct btrfs_trans_handle *trans,
5197 struct btrfs_root *root,
5198 struct btrfs_path *path,
5199 struct btrfs_key *first_key,
5200 struct btrfs_ref_path *ref_path)
5201 {
5202 int ret;
5203
5204 ret = relocate_one_path(trans, root, path, first_key,
5205 ref_path, NULL, NULL);
5206 BUG_ON(ret);
5207
5208 if (root == root->fs_info->extent_root)
5209 btrfs_extent_post_op(trans, root);
5210
5211 return 0;
5212 }
5213
5214 static int noinline del_extent_zero(struct btrfs_trans_handle *trans,
5215 struct btrfs_root *extent_root,
5216 struct btrfs_path *path,
5217 struct btrfs_key *extent_key)
5218 {
5219 int ret;
5220
5221 ret = btrfs_search_slot(trans, extent_root, extent_key, path, -1, 1);
5222 if (ret)
5223 goto out;
5224 ret = btrfs_del_item(trans, extent_root, path);
5225 out:
5226 btrfs_release_path(extent_root, path);
5227 return ret;
5228 }
5229
5230 static struct btrfs_root noinline *read_ref_root(struct btrfs_fs_info *fs_info,
5231 struct btrfs_ref_path *ref_path)
5232 {
5233 struct btrfs_key root_key;
5234
5235 root_key.objectid = ref_path->root_objectid;
5236 root_key.type = BTRFS_ROOT_ITEM_KEY;
5237 if (is_cowonly_root(ref_path->root_objectid))
5238 root_key.offset = 0;
5239 else
5240 root_key.offset = (u64)-1;
5241
5242 return btrfs_read_fs_root_no_name(fs_info, &root_key);
5243 }
5244
5245 static int noinline relocate_one_extent(struct btrfs_root *extent_root,
5246 struct btrfs_path *path,
5247 struct btrfs_key *extent_key,
5248 struct btrfs_block_group_cache *group,
5249 struct inode *reloc_inode, int pass)
5250 {
5251 struct btrfs_trans_handle *trans;
5252 struct btrfs_root *found_root;
5253 struct btrfs_ref_path *ref_path = NULL;
5254 struct disk_extent *new_extents = NULL;
5255 int nr_extents = 0;
5256 int loops;
5257 int ret;
5258 int level;
5259 struct btrfs_key first_key;
5260 u64 prev_block = 0;
5261
5262
5263 trans = btrfs_start_transaction(extent_root, 1);
5264 BUG_ON(!trans);
5265
5266 if (extent_key->objectid == 0) {
5267 ret = del_extent_zero(trans, extent_root, path, extent_key);
5268 goto out;
5269 }
5270
5271 ref_path = kmalloc(sizeof(*ref_path), GFP_NOFS);
5272 if (!ref_path) {
5273 ret = -ENOMEM;
5274 goto out;
5275 }
5276
5277 for (loops = 0; ; loops++) {
5278 if (loops == 0) {
5279 ret = btrfs_first_ref_path(trans, extent_root, ref_path,
5280 extent_key->objectid);
5281 } else {
5282 ret = btrfs_next_ref_path(trans, extent_root, ref_path);
5283 }
5284 if (ret < 0)
5285 goto out;
5286 if (ret > 0)
5287 break;
5288
5289 if (ref_path->root_objectid == BTRFS_TREE_LOG_OBJECTID ||
5290 ref_path->root_objectid == BTRFS_TREE_RELOC_OBJECTID)
5291 continue;
5292
5293 found_root = read_ref_root(extent_root->fs_info, ref_path);
5294 BUG_ON(!found_root);
5295 /*
5296 * for reference counted tree, only process reference paths
5297 * rooted at the latest committed root.
5298 */
5299 if (found_root->ref_cows &&
5300 ref_path->root_generation != found_root->root_key.offset)
5301 continue;
5302
5303 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
5304 if (pass == 0) {
5305 /*
5306 * copy data extents to new locations
5307 */
5308 u64 group_start = group->key.objectid;
5309 ret = relocate_data_extent(reloc_inode,
5310 extent_key,
5311 group_start);
5312 if (ret < 0)
5313 goto out;
5314 break;
5315 }
5316 level = 0;
5317 } else {
5318 level = ref_path->owner_objectid;
5319 }
5320
5321 if (prev_block != ref_path->nodes[level]) {
5322 struct extent_buffer *eb;
5323 u64 block_start = ref_path->nodes[level];
5324 u64 block_size = btrfs_level_size(found_root, level);
5325
5326 eb = read_tree_block(found_root, block_start,
5327 block_size, 0);
5328 btrfs_tree_lock(eb);
5329 BUG_ON(level != btrfs_header_level(eb));
5330
5331 if (level == 0)
5332 btrfs_item_key_to_cpu(eb, &first_key, 0);
5333 else
5334 btrfs_node_key_to_cpu(eb, &first_key, 0);
5335
5336 btrfs_tree_unlock(eb);
5337 free_extent_buffer(eb);
5338 prev_block = block_start;
5339 }
5340
5341 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID &&
5342 pass >= 2) {
5343 /*
5344 * use fallback method to process the remaining
5345 * references.
5346 */
5347 if (!new_extents) {
5348 u64 group_start = group->key.objectid;
5349 new_extents = kmalloc(sizeof(*new_extents),
5350 GFP_NOFS);
5351 nr_extents = 1;
5352 ret = get_new_locations(reloc_inode,
5353 extent_key,
5354 group_start, 1,
5355 &new_extents,
5356 &nr_extents);
5357 if (ret)
5358 goto out;
5359 }
5360 btrfs_record_root_in_trans(found_root);
5361 ret = replace_one_extent(trans, found_root,
5362 path, extent_key,
5363 &first_key, ref_path,
5364 new_extents, nr_extents);
5365 if (ret < 0)
5366 goto out;
5367 continue;
5368 }
5369
5370 btrfs_record_root_in_trans(found_root);
5371 if (ref_path->owner_objectid < BTRFS_FIRST_FREE_OBJECTID) {
5372 ret = relocate_tree_block(trans, found_root, path,
5373 &first_key, ref_path);
5374 } else {
5375 /*
5376 * try to update data extent references while
5377 * keeping metadata shared between snapshots.
5378 */
5379 ret = relocate_one_path(trans, found_root, path,
5380 &first_key, ref_path,
5381 group, reloc_inode);
5382 }
5383 if (ret < 0)
5384 goto out;
5385 }
5386 ret = 0;
5387 out:
5388 btrfs_end_transaction(trans, extent_root);
5389 kfree(new_extents);
5390 kfree(ref_path);
5391 return ret;
5392 }
5393
5394 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
5395 {
5396 u64 num_devices;
5397 u64 stripped = BTRFS_BLOCK_GROUP_RAID0 |
5398 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
5399
5400 num_devices = root->fs_info->fs_devices->rw_devices;
5401 if (num_devices == 1) {
5402 stripped |= BTRFS_BLOCK_GROUP_DUP;
5403 stripped = flags & ~stripped;
5404
5405 /* turn raid0 into single device chunks */
5406 if (flags & BTRFS_BLOCK_GROUP_RAID0)
5407 return stripped;
5408
5409 /* turn mirroring into duplication */
5410 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
5411 BTRFS_BLOCK_GROUP_RAID10))
5412 return stripped | BTRFS_BLOCK_GROUP_DUP;
5413 return flags;
5414 } else {
5415 /* they already had raid on here, just return */
5416 if (flags & stripped)
5417 return flags;
5418
5419 stripped |= BTRFS_BLOCK_GROUP_DUP;
5420 stripped = flags & ~stripped;
5421
5422 /* switch duplicated blocks with raid1 */
5423 if (flags & BTRFS_BLOCK_GROUP_DUP)
5424 return stripped | BTRFS_BLOCK_GROUP_RAID1;
5425
5426 /* turn single device chunks into raid0 */
5427 return stripped | BTRFS_BLOCK_GROUP_RAID0;
5428 }
5429 return flags;
5430 }
5431
5432 int __alloc_chunk_for_shrink(struct btrfs_root *root,
5433 struct btrfs_block_group_cache *shrink_block_group,
5434 int force)
5435 {
5436 struct btrfs_trans_handle *trans;
5437 u64 new_alloc_flags;
5438 u64 calc;
5439
5440 spin_lock(&shrink_block_group->lock);
5441 if (btrfs_block_group_used(&shrink_block_group->item) > 0) {
5442 spin_unlock(&shrink_block_group->lock);
5443
5444 trans = btrfs_start_transaction(root, 1);
5445 spin_lock(&shrink_block_group->lock);
5446
5447 new_alloc_flags = update_block_group_flags(root,
5448 shrink_block_group->flags);
5449 if (new_alloc_flags != shrink_block_group->flags) {
5450 calc =
5451 btrfs_block_group_used(&shrink_block_group->item);
5452 } else {
5453 calc = shrink_block_group->key.offset;
5454 }
5455 spin_unlock(&shrink_block_group->lock);
5456
5457 do_chunk_alloc(trans, root->fs_info->extent_root,
5458 calc + 2 * 1024 * 1024, new_alloc_flags, force);
5459
5460 btrfs_end_transaction(trans, root);
5461 } else
5462 spin_unlock(&shrink_block_group->lock);
5463 return 0;
5464 }
5465
5466 static int __insert_orphan_inode(struct btrfs_trans_handle *trans,
5467 struct btrfs_root *root,
5468 u64 objectid, u64 size)
5469 {
5470 struct btrfs_path *path;
5471 struct btrfs_inode_item *item;
5472 struct extent_buffer *leaf;
5473 int ret;
5474
5475 path = btrfs_alloc_path();
5476 if (!path)
5477 return -ENOMEM;
5478
5479 ret = btrfs_insert_empty_inode(trans, root, path, objectid);
5480 if (ret)
5481 goto out;
5482
5483 leaf = path->nodes[0];
5484 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_inode_item);
5485 memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item));
5486 btrfs_set_inode_generation(leaf, item, 1);
5487 btrfs_set_inode_size(leaf, item, size);
5488 btrfs_set_inode_mode(leaf, item, S_IFREG | 0600);
5489 btrfs_set_inode_flags(leaf, item, BTRFS_INODE_NODATASUM |
5490 BTRFS_INODE_NOCOMPRESS);
5491 btrfs_mark_buffer_dirty(leaf);
5492 btrfs_release_path(root, path);
5493 out:
5494 btrfs_free_path(path);
5495 return ret;
5496 }
5497
5498 static struct inode noinline *create_reloc_inode(struct btrfs_fs_info *fs_info,
5499 struct btrfs_block_group_cache *group)
5500 {
5501 struct inode *inode = NULL;
5502 struct btrfs_trans_handle *trans;
5503 struct btrfs_root *root;
5504 struct btrfs_key root_key;
5505 u64 objectid = BTRFS_FIRST_FREE_OBJECTID;
5506 int err = 0;
5507
5508 root_key.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
5509 root_key.type = BTRFS_ROOT_ITEM_KEY;
5510 root_key.offset = (u64)-1;
5511 root = btrfs_read_fs_root_no_name(fs_info, &root_key);
5512 if (IS_ERR(root))
5513 return ERR_CAST(root);
5514
5515 trans = btrfs_start_transaction(root, 1);
5516 BUG_ON(!trans);
5517
5518 err = btrfs_find_free_objectid(trans, root, objectid, &objectid);
5519 if (err)
5520 goto out;
5521
5522 err = __insert_orphan_inode(trans, root, objectid, group->key.offset);
5523 BUG_ON(err);
5524
5525 err = btrfs_insert_file_extent(trans, root, objectid, 0, 0, 0,
5526 group->key.offset, 0, group->key.offset,
5527 0, 0, 0);
5528 BUG_ON(err);
5529
5530 inode = btrfs_iget_locked(root->fs_info->sb, objectid, root);
5531 if (inode->i_state & I_NEW) {
5532 BTRFS_I(inode)->root = root;
5533 BTRFS_I(inode)->location.objectid = objectid;
5534 BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
5535 BTRFS_I(inode)->location.offset = 0;
5536 btrfs_read_locked_inode(inode);
5537 unlock_new_inode(inode);
5538 BUG_ON(is_bad_inode(inode));
5539 } else {
5540 BUG_ON(1);
5541 }
5542
5543 err = btrfs_orphan_add(trans, inode);
5544 out:
5545 btrfs_end_transaction(trans, root);
5546 if (err) {
5547 if (inode)
5548 iput(inode);
5549 inode = ERR_PTR(err);
5550 }
5551 return inode;
5552 }
5553
5554 int btrfs_relocate_block_group(struct btrfs_root *root, u64 group_start)
5555 {
5556 struct btrfs_trans_handle *trans;
5557 struct btrfs_path *path;
5558 struct btrfs_fs_info *info = root->fs_info;
5559 struct extent_buffer *leaf;
5560 struct inode *reloc_inode;
5561 struct btrfs_block_group_cache *block_group;
5562 struct btrfs_key key;
5563 u64 skipped;
5564 u64 cur_byte;
5565 u64 total_found;
5566 u32 nritems;
5567 int ret;
5568 int progress;
5569 int pass = 0;
5570
5571 root = root->fs_info->extent_root;
5572
5573 block_group = btrfs_lookup_block_group(info, group_start);
5574 BUG_ON(!block_group);
5575
5576 printk("btrfs relocating block group %llu flags %llu\n",
5577 (unsigned long long)block_group->key.objectid,
5578 (unsigned long long)block_group->flags);
5579
5580 path = btrfs_alloc_path();
5581 BUG_ON(!path);
5582
5583 reloc_inode = create_reloc_inode(info, block_group);
5584 BUG_ON(IS_ERR(reloc_inode));
5585
5586 __alloc_chunk_for_shrink(root, block_group, 1);
5587 set_block_group_readonly(block_group);
5588
5589 btrfs_start_delalloc_inodes(info->tree_root);
5590 btrfs_wait_ordered_extents(info->tree_root, 0);
5591 again:
5592 skipped = 0;
5593 total_found = 0;
5594 progress = 0;
5595 key.objectid = block_group->key.objectid;
5596 key.offset = 0;
5597 key.type = 0;
5598 cur_byte = key.objectid;
5599
5600 trans = btrfs_start_transaction(info->tree_root, 1);
5601 btrfs_commit_transaction(trans, info->tree_root);
5602
5603 mutex_lock(&root->fs_info->cleaner_mutex);
5604 btrfs_clean_old_snapshots(info->tree_root);
5605 btrfs_remove_leaf_refs(info->tree_root, (u64)-1, 1);
5606 mutex_unlock(&root->fs_info->cleaner_mutex);
5607
5608 while(1) {
5609 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5610 if (ret < 0)
5611 goto out;
5612 next:
5613 leaf = path->nodes[0];
5614 nritems = btrfs_header_nritems(leaf);
5615 if (path->slots[0] >= nritems) {
5616 ret = btrfs_next_leaf(root, path);
5617 if (ret < 0)
5618 goto out;
5619 if (ret == 1) {
5620 ret = 0;
5621 break;
5622 }
5623 leaf = path->nodes[0];
5624 nritems = btrfs_header_nritems(leaf);
5625 }
5626
5627 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
5628
5629 if (key.objectid >= block_group->key.objectid +
5630 block_group->key.offset)
5631 break;
5632
5633 if (progress && need_resched()) {
5634 btrfs_release_path(root, path);
5635 cond_resched();
5636 progress = 0;
5637 continue;
5638 }
5639 progress = 1;
5640
5641 if (btrfs_key_type(&key) != BTRFS_EXTENT_ITEM_KEY ||
5642 key.objectid + key.offset <= cur_byte) {
5643 path->slots[0]++;
5644 goto next;
5645 }
5646
5647 total_found++;
5648 cur_byte = key.objectid + key.offset;
5649 btrfs_release_path(root, path);
5650
5651 __alloc_chunk_for_shrink(root, block_group, 0);
5652 ret = relocate_one_extent(root, path, &key, block_group,
5653 reloc_inode, pass);
5654 BUG_ON(ret < 0);
5655 if (ret > 0)
5656 skipped++;
5657
5658 key.objectid = cur_byte;
5659 key.type = 0;
5660 key.offset = 0;
5661 }
5662
5663 btrfs_release_path(root, path);
5664
5665 if (pass == 0) {
5666 btrfs_wait_ordered_range(reloc_inode, 0, (u64)-1);
5667 invalidate_mapping_pages(reloc_inode->i_mapping, 0, -1);
5668 WARN_ON(reloc_inode->i_mapping->nrpages);
5669 }
5670
5671 if (total_found > 0) {
5672 printk("btrfs found %llu extents in pass %d\n",
5673 (unsigned long long)total_found, pass);
5674 pass++;
5675 if (total_found == skipped && pass > 2) {
5676 iput(reloc_inode);
5677 reloc_inode = create_reloc_inode(info, block_group);
5678 pass = 0;
5679 }
5680 goto again;
5681 }
5682
5683 /* delete reloc_inode */
5684 iput(reloc_inode);
5685
5686 /* unpin extents in this range */
5687 trans = btrfs_start_transaction(info->tree_root, 1);
5688 btrfs_commit_transaction(trans, info->tree_root);
5689
5690 spin_lock(&block_group->lock);
5691 WARN_ON(block_group->pinned > 0);
5692 WARN_ON(block_group->reserved > 0);
5693 WARN_ON(btrfs_block_group_used(&block_group->item) > 0);
5694 spin_unlock(&block_group->lock);
5695 ret = 0;
5696 out:
5697 btrfs_free_path(path);
5698 return ret;
5699 }
5700
5701 int find_first_block_group(struct btrfs_root *root, struct btrfs_path *path,
5702 struct btrfs_key *key)
5703 {
5704 int ret = 0;
5705 struct btrfs_key found_key;
5706 struct extent_buffer *leaf;
5707 int slot;
5708
5709 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
5710 if (ret < 0)
5711 goto out;
5712
5713 while(1) {
5714 slot = path->slots[0];
5715 leaf = path->nodes[0];
5716 if (slot >= btrfs_header_nritems(leaf)) {
5717 ret = btrfs_next_leaf(root, path);
5718 if (ret == 0)
5719 continue;
5720 if (ret < 0)
5721 goto out;
5722 break;
5723 }
5724 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5725
5726 if (found_key.objectid >= key->objectid &&
5727 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
5728 ret = 0;
5729 goto out;
5730 }
5731 path->slots[0]++;
5732 }
5733 ret = -ENOENT;
5734 out:
5735 return ret;
5736 }
5737
5738 int btrfs_free_block_groups(struct btrfs_fs_info *info)
5739 {
5740 struct btrfs_block_group_cache *block_group;
5741 struct rb_node *n;
5742
5743 spin_lock(&info->block_group_cache_lock);
5744 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
5745 block_group = rb_entry(n, struct btrfs_block_group_cache,
5746 cache_node);
5747 rb_erase(&block_group->cache_node,
5748 &info->block_group_cache_tree);
5749 spin_unlock(&info->block_group_cache_lock);
5750
5751 btrfs_remove_free_space_cache(block_group);
5752 down_write(&block_group->space_info->groups_sem);
5753 list_del(&block_group->list);
5754 up_write(&block_group->space_info->groups_sem);
5755 kfree(block_group);
5756
5757 spin_lock(&info->block_group_cache_lock);
5758 }
5759 spin_unlock(&info->block_group_cache_lock);
5760 return 0;
5761 }
5762
5763 int btrfs_read_block_groups(struct btrfs_root *root)
5764 {
5765 struct btrfs_path *path;
5766 int ret;
5767 struct btrfs_block_group_cache *cache;
5768 struct btrfs_fs_info *info = root->fs_info;
5769 struct btrfs_space_info *space_info;
5770 struct btrfs_key key;
5771 struct btrfs_key found_key;
5772 struct extent_buffer *leaf;
5773
5774 root = info->extent_root;
5775 key.objectid = 0;
5776 key.offset = 0;
5777 btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
5778 path = btrfs_alloc_path();
5779 if (!path)
5780 return -ENOMEM;
5781
5782 while(1) {
5783 ret = find_first_block_group(root, path, &key);
5784 if (ret > 0) {
5785 ret = 0;
5786 goto error;
5787 }
5788 if (ret != 0)
5789 goto error;
5790
5791 leaf = path->nodes[0];
5792 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5793 cache = kzalloc(sizeof(*cache), GFP_NOFS);
5794 if (!cache) {
5795 ret = -ENOMEM;
5796 break;
5797 }
5798
5799 spin_lock_init(&cache->lock);
5800 mutex_init(&cache->alloc_mutex);
5801 INIT_LIST_HEAD(&cache->list);
5802 read_extent_buffer(leaf, &cache->item,
5803 btrfs_item_ptr_offset(leaf, path->slots[0]),
5804 sizeof(cache->item));
5805 memcpy(&cache->key, &found_key, sizeof(found_key));
5806
5807 key.objectid = found_key.objectid + found_key.offset;
5808 btrfs_release_path(root, path);
5809 cache->flags = btrfs_block_group_flags(&cache->item);
5810
5811 ret = update_space_info(info, cache->flags, found_key.offset,
5812 btrfs_block_group_used(&cache->item),
5813 &space_info);
5814 BUG_ON(ret);
5815 cache->space_info = space_info;
5816 down_write(&space_info->groups_sem);
5817 list_add_tail(&cache->list, &space_info->block_groups);
5818 up_write(&space_info->groups_sem);
5819
5820 ret = btrfs_add_block_group_cache(root->fs_info, cache);
5821 BUG_ON(ret);
5822
5823 set_avail_alloc_bits(root->fs_info, cache->flags);
5824 if (btrfs_chunk_readonly(root, cache->key.objectid))
5825 set_block_group_readonly(cache);
5826 }
5827 ret = 0;
5828 error:
5829 btrfs_free_path(path);
5830 return ret;
5831 }
5832
5833 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
5834 struct btrfs_root *root, u64 bytes_used,
5835 u64 type, u64 chunk_objectid, u64 chunk_offset,
5836 u64 size)
5837 {
5838 int ret;
5839 struct btrfs_root *extent_root;
5840 struct btrfs_block_group_cache *cache;
5841
5842 extent_root = root->fs_info->extent_root;
5843
5844 root->fs_info->last_trans_new_blockgroup = trans->transid;
5845
5846 cache = kzalloc(sizeof(*cache), GFP_NOFS);
5847 if (!cache)
5848 return -ENOMEM;
5849
5850 cache->key.objectid = chunk_offset;
5851 cache->key.offset = size;
5852 spin_lock_init(&cache->lock);
5853 mutex_init(&cache->alloc_mutex);
5854 INIT_LIST_HEAD(&cache->list);
5855 btrfs_set_key_type(&cache->key, BTRFS_BLOCK_GROUP_ITEM_KEY);
5856
5857 btrfs_set_block_group_used(&cache->item, bytes_used);
5858 btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
5859 cache->flags = type;
5860 btrfs_set_block_group_flags(&cache->item, type);
5861
5862 ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
5863 &cache->space_info);
5864 BUG_ON(ret);
5865 down_write(&cache->space_info->groups_sem);
5866 list_add_tail(&cache->list, &cache->space_info->block_groups);
5867 up_write(&cache->space_info->groups_sem);
5868
5869 ret = btrfs_add_block_group_cache(root->fs_info, cache);
5870 BUG_ON(ret);
5871
5872 ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
5873 sizeof(cache->item));
5874 BUG_ON(ret);
5875
5876 finish_current_insert(trans, extent_root, 0);
5877 ret = del_pending_extents(trans, extent_root, 0);
5878 BUG_ON(ret);
5879 set_avail_alloc_bits(extent_root->fs_info, type);
5880
5881 return 0;
5882 }
5883
5884 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
5885 struct btrfs_root *root, u64 group_start)
5886 {
5887 struct btrfs_path *path;
5888 struct btrfs_block_group_cache *block_group;
5889 struct btrfs_key key;
5890 int ret;
5891
5892 root = root->fs_info->extent_root;
5893
5894 block_group = btrfs_lookup_block_group(root->fs_info, group_start);
5895 BUG_ON(!block_group);
5896 BUG_ON(!block_group->ro);
5897
5898 memcpy(&key, &block_group->key, sizeof(key));
5899
5900 path = btrfs_alloc_path();
5901 BUG_ON(!path);
5902
5903 btrfs_remove_free_space_cache(block_group);
5904 rb_erase(&block_group->cache_node,
5905 &root->fs_info->block_group_cache_tree);
5906 down_write(&block_group->space_info->groups_sem);
5907 list_del(&block_group->list);
5908 up_write(&block_group->space_info->groups_sem);
5909
5910 spin_lock(&block_group->space_info->lock);
5911 block_group->space_info->total_bytes -= block_group->key.offset;
5912 block_group->space_info->bytes_readonly -= block_group->key.offset;
5913 spin_unlock(&block_group->space_info->lock);
5914 block_group->space_info->full = 0;
5915
5916 /*
5917 memset(shrink_block_group, 0, sizeof(*shrink_block_group));
5918 kfree(shrink_block_group);
5919 */
5920
5921 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
5922 if (ret > 0)
5923 ret = -EIO;
5924 if (ret < 0)
5925 goto out;
5926
5927 ret = btrfs_del_item(trans, root, path);
5928 out:
5929 btrfs_free_path(path);
5930 return ret;
5931 }
Linux kernel - Deliverables
This page took 0.14766 seconds and 4 git commands to generate.