xref: /openbmc/linux/fs/btrfs/extent-tree.c (revision a213501153fd66e2359e091b1612841305ba6551)
1 /*
2  * Copyright (C) 2007 Oracle.  All rights reserved.
3  *
4  * This program is free software; you can redistribute it and/or
5  * modify it under the terms of the GNU General Public
6  * License v2 as published by the Free Software Foundation.
7  *
8  * This program is distributed in the hope that it will be useful,
9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
11  * General Public License for more details.
12  *
13  * You should have received a copy of the GNU General Public
14  * License along with this program; if not, write to the
15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16  * Boston, MA 021110-1307, USA.
17  */
18 #include <linux/sched.h>
19 #include <linux/pagemap.h>
20 #include <linux/writeback.h>
21 #include "hash.h"
22 #include "crc32c.h"
23 #include "ctree.h"
24 #include "disk-io.h"
25 #include "print-tree.h"
26 #include "transaction.h"
27 #include "volumes.h"
28 #include "locking.h"
29 
30 #define BLOCK_GROUP_DATA     EXTENT_WRITEBACK
31 #define BLOCK_GROUP_METADATA EXTENT_UPTODATE
32 #define BLOCK_GROUP_SYSTEM   EXTENT_NEW
33 
34 #define BLOCK_GROUP_DIRTY EXTENT_DIRTY
35 
36 static int finish_current_insert(struct btrfs_trans_handle *trans, struct
37 				 btrfs_root *extent_root);
38 static int del_pending_extents(struct btrfs_trans_handle *trans, struct
39 			       btrfs_root *extent_root);
40 static struct btrfs_block_group_cache *
41 __btrfs_find_block_group(struct btrfs_root *root,
42 			 struct btrfs_block_group_cache *hint,
43 			 u64 search_start, int data, int owner);
44 
45 void maybe_lock_mutex(struct btrfs_root *root)
46 {
47 	if (root != root->fs_info->extent_root &&
48 	    root != root->fs_info->chunk_root &&
49 	    root != root->fs_info->dev_root) {
50 		mutex_lock(&root->fs_info->alloc_mutex);
51 	}
52 }
53 
54 void maybe_unlock_mutex(struct btrfs_root *root)
55 {
56 	if (root != root->fs_info->extent_root &&
57 	    root != root->fs_info->chunk_root &&
58 	    root != root->fs_info->dev_root) {
59 		mutex_unlock(&root->fs_info->alloc_mutex);
60 	}
61 }
62 
63 static int cache_block_group(struct btrfs_root *root,
64 			     struct btrfs_block_group_cache *block_group)
65 {
66 	struct btrfs_path *path;
67 	int ret;
68 	struct btrfs_key key;
69 	struct extent_buffer *leaf;
70 	struct extent_io_tree *free_space_cache;
71 	int slot;
72 	u64 last = 0;
73 	u64 hole_size;
74 	u64 first_free;
75 	int found = 0;
76 
77 	if (!block_group)
78 		return 0;
79 
80 	root = root->fs_info->extent_root;
81 	free_space_cache = &root->fs_info->free_space_cache;
82 
83 	if (block_group->cached)
84 		return 0;
85 
86 	path = btrfs_alloc_path();
87 	if (!path)
88 		return -ENOMEM;
89 
90 	path->reada = 2;
91 	path->skip_locking = 1;
92 	first_free = block_group->key.objectid;
93 	key.objectid = block_group->key.objectid;
94 	key.offset = 0;
95 	btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
96 	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
97 	if (ret < 0)
98 		return ret;
99 	ret = btrfs_previous_item(root, path, 0, BTRFS_EXTENT_ITEM_KEY);
100 	if (ret < 0)
101 		return ret;
102 	if (ret == 0) {
103 		leaf = path->nodes[0];
104 		btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
105 		if (key.objectid + key.offset > first_free)
106 			first_free = key.objectid + key.offset;
107 	}
108 	while(1) {
109 		leaf = path->nodes[0];
110 		slot = path->slots[0];
111 		if (slot >= btrfs_header_nritems(leaf)) {
112 			ret = btrfs_next_leaf(root, path);
113 			if (ret < 0)
114 				goto err;
115 			if (ret == 0) {
116 				continue;
117 			} else {
118 				break;
119 			}
120 		}
121 		btrfs_item_key_to_cpu(leaf, &key, slot);
122 		if (key.objectid < block_group->key.objectid) {
123 			goto next;
124 		}
125 		if (key.objectid >= block_group->key.objectid +
126 		    block_group->key.offset) {
127 			break;
128 		}
129 
130 		if (btrfs_key_type(&key) == BTRFS_EXTENT_ITEM_KEY) {
131 			if (!found) {
132 				last = first_free;
133 				found = 1;
134 			}
135 			if (key.objectid > last) {
136 				hole_size = key.objectid - last;
137 				set_extent_dirty(free_space_cache, last,
138 						 last + hole_size - 1,
139 						 GFP_NOFS);
140 			}
141 			last = key.objectid + key.offset;
142 		}
143 next:
144 		path->slots[0]++;
145 	}
146 
147 	if (!found)
148 		last = first_free;
149 	if (block_group->key.objectid +
150 	    block_group->key.offset > last) {
151 		hole_size = block_group->key.objectid +
152 			block_group->key.offset - last;
153 		set_extent_dirty(free_space_cache, last,
154 				 last + hole_size - 1, GFP_NOFS);
155 	}
156 	block_group->cached = 1;
157 err:
158 	btrfs_free_path(path);
159 	return 0;
160 }
161 
162 struct btrfs_block_group_cache *btrfs_lookup_first_block_group(struct
163 						       btrfs_fs_info *info,
164 							 u64 bytenr)
165 {
166 	struct extent_io_tree *block_group_cache;
167 	struct btrfs_block_group_cache *block_group = NULL;
168 	u64 ptr;
169 	u64 start;
170 	u64 end;
171 	int ret;
172 
173 	bytenr = max_t(u64, bytenr,
174 		       BTRFS_SUPER_INFO_OFFSET + BTRFS_SUPER_INFO_SIZE);
175 	block_group_cache = &info->block_group_cache;
176 	ret = find_first_extent_bit(block_group_cache,
177 				    bytenr, &start, &end,
178 				    BLOCK_GROUP_DATA | BLOCK_GROUP_METADATA |
179 				    BLOCK_GROUP_SYSTEM);
180 	if (ret) {
181 		return NULL;
182 	}
183 	ret = get_state_private(block_group_cache, start, &ptr);
184 	if (ret)
185 		return NULL;
186 
187 	block_group = (struct btrfs_block_group_cache *)(unsigned long)ptr;
188 	return block_group;
189 }
190 
191 struct btrfs_block_group_cache *btrfs_lookup_block_group(struct
192 							 btrfs_fs_info *info,
193 							 u64 bytenr)
194 {
195 	struct extent_io_tree *block_group_cache;
196 	struct btrfs_block_group_cache *block_group = NULL;
197 	u64 ptr;
198 	u64 start;
199 	u64 end;
200 	int ret;
201 
202 	bytenr = max_t(u64, bytenr,
203 		       BTRFS_SUPER_INFO_OFFSET + BTRFS_SUPER_INFO_SIZE);
204 	block_group_cache = &info->block_group_cache;
205 	ret = find_first_extent_bit(block_group_cache,
206 				    bytenr, &start, &end,
207 				    BLOCK_GROUP_DATA | BLOCK_GROUP_METADATA |
208 				    BLOCK_GROUP_SYSTEM);
209 	if (ret) {
210 		return NULL;
211 	}
212 	ret = get_state_private(block_group_cache, start, &ptr);
213 	if (ret)
214 		return NULL;
215 
216 	block_group = (struct btrfs_block_group_cache *)(unsigned long)ptr;
217 	if (block_group->key.objectid <= bytenr && bytenr <
218 	    block_group->key.objectid + block_group->key.offset)
219 		return block_group;
220 	return NULL;
221 }
222 
223 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
224 {
225 	return (cache->flags & bits) == bits;
226 }
227 
228 static int noinline find_search_start(struct btrfs_root *root,
229 			      struct btrfs_block_group_cache **cache_ret,
230 			      u64 *start_ret, u64 num, int data)
231 {
232 	int ret;
233 	struct btrfs_block_group_cache *cache = *cache_ret;
234 	struct extent_io_tree *free_space_cache;
235 	struct extent_state *state;
236 	u64 last;
237 	u64 start = 0;
238 	u64 cache_miss = 0;
239 	u64 total_fs_bytes;
240 	u64 search_start = *start_ret;
241 	int wrapped = 0;
242 
243 	total_fs_bytes = btrfs_super_total_bytes(&root->fs_info->super_copy);
244 	free_space_cache = &root->fs_info->free_space_cache;
245 
246 	if (!cache)
247 		goto out;
248 
249 again:
250 	ret = cache_block_group(root, cache);
251 	if (ret) {
252 		goto out;
253 	}
254 
255 	last = max(search_start, cache->key.objectid);
256 	if (!block_group_bits(cache, data) || cache->ro)
257 		goto new_group;
258 
259 	spin_lock_irq(&free_space_cache->lock);
260 	state = find_first_extent_bit_state(free_space_cache, last, EXTENT_DIRTY);
261 	while(1) {
262 		if (!state) {
263 			if (!cache_miss)
264 				cache_miss = last;
265 			spin_unlock_irq(&free_space_cache->lock);
266 			goto new_group;
267 		}
268 
269 		start = max(last, state->start);
270 		last = state->end + 1;
271 		if (last - start < num) {
272 			do {
273 				state = extent_state_next(state);
274 			} while(state && !(state->state & EXTENT_DIRTY));
275 			continue;
276 		}
277 		spin_unlock_irq(&free_space_cache->lock);
278 		if (cache->ro) {
279 			goto new_group;
280 		}
281 		if (start + num > cache->key.objectid + cache->key.offset)
282 			goto new_group;
283 		if (!block_group_bits(cache, data)) {
284 			printk("block group bits don't match %Lu %d\n", cache->flags, data);
285 		}
286 		*start_ret = start;
287 		return 0;
288 	}
289 out:
290 	cache = btrfs_lookup_block_group(root->fs_info, search_start);
291 	if (!cache) {
292 		printk("Unable to find block group for %Lu\n", search_start);
293 		WARN_ON(1);
294 	}
295 	return -ENOSPC;
296 
297 new_group:
298 	last = cache->key.objectid + cache->key.offset;
299 wrapped:
300 	cache = btrfs_lookup_first_block_group(root->fs_info, last);
301 	if (!cache || cache->key.objectid >= total_fs_bytes) {
302 no_cache:
303 		if (!wrapped) {
304 			wrapped = 1;
305 			last = search_start;
306 			goto wrapped;
307 		}
308 		goto out;
309 	}
310 	if (cache_miss && !cache->cached) {
311 		cache_block_group(root, cache);
312 		last = cache_miss;
313 		cache = btrfs_lookup_first_block_group(root->fs_info, last);
314 	}
315 	cache_miss = 0;
316 	cache = __btrfs_find_block_group(root, cache, last, data, 0);
317 	if (!cache)
318 		goto no_cache;
319 	*cache_ret = cache;
320 	goto again;
321 }
322 
323 static u64 div_factor(u64 num, int factor)
324 {
325 	if (factor == 10)
326 		return num;
327 	num *= factor;
328 	do_div(num, 10);
329 	return num;
330 }
331 
332 static int block_group_state_bits(u64 flags)
333 {
334 	int bits = 0;
335 	if (flags & BTRFS_BLOCK_GROUP_DATA)
336 		bits |= BLOCK_GROUP_DATA;
337 	if (flags & BTRFS_BLOCK_GROUP_METADATA)
338 		bits |= BLOCK_GROUP_METADATA;
339 	if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
340 		bits |= BLOCK_GROUP_SYSTEM;
341 	return bits;
342 }
343 
344 static struct btrfs_block_group_cache *
345 __btrfs_find_block_group(struct btrfs_root *root,
346 			 struct btrfs_block_group_cache *hint,
347 			 u64 search_start, int data, int owner)
348 {
349 	struct btrfs_block_group_cache *cache;
350 	struct extent_io_tree *block_group_cache;
351 	struct btrfs_block_group_cache *found_group = NULL;
352 	struct btrfs_fs_info *info = root->fs_info;
353 	u64 used;
354 	u64 last = 0;
355 	u64 start;
356 	u64 end;
357 	u64 free_check;
358 	u64 ptr;
359 	int bit;
360 	int ret;
361 	int full_search = 0;
362 	int factor = 10;
363 	int wrapped = 0;
364 
365 	block_group_cache = &info->block_group_cache;
366 
367 	if (data & BTRFS_BLOCK_GROUP_METADATA)
368 		factor = 9;
369 
370 	bit = block_group_state_bits(data);
371 
372 	if (search_start) {
373 		struct btrfs_block_group_cache *shint;
374 		shint = btrfs_lookup_first_block_group(info, search_start);
375 		if (shint && block_group_bits(shint, data) && !shint->ro) {
376 			used = btrfs_block_group_used(&shint->item);
377 			if (used + shint->pinned <
378 			    div_factor(shint->key.offset, factor)) {
379 				return shint;
380 			}
381 		}
382 	}
383 	if (hint && !hint->ro && block_group_bits(hint, data)) {
384 		used = btrfs_block_group_used(&hint->item);
385 		if (used + hint->pinned <
386 		    div_factor(hint->key.offset, factor)) {
387 			return hint;
388 		}
389 		last = hint->key.objectid + hint->key.offset;
390 	} else {
391 		if (hint)
392 			last = max(hint->key.objectid, search_start);
393 		else
394 			last = search_start;
395 	}
396 again:
397 	while(1) {
398 		ret = find_first_extent_bit(block_group_cache, last,
399 					    &start, &end, bit);
400 		if (ret)
401 			break;
402 
403 		ret = get_state_private(block_group_cache, start, &ptr);
404 		if (ret) {
405 			last = end + 1;
406 			continue;
407 		}
408 
409 		cache = (struct btrfs_block_group_cache *)(unsigned long)ptr;
410 		last = cache->key.objectid + cache->key.offset;
411 		used = btrfs_block_group_used(&cache->item);
412 
413 		if (!cache->ro && block_group_bits(cache, data)) {
414 			free_check = div_factor(cache->key.offset, factor);
415 			if (used + cache->pinned < free_check) {
416 				found_group = cache;
417 				goto found;
418 			}
419 		}
420 		cond_resched();
421 	}
422 	if (!wrapped) {
423 		last = search_start;
424 		wrapped = 1;
425 		goto again;
426 	}
427 	if (!full_search && factor < 10) {
428 		last = search_start;
429 		full_search = 1;
430 		factor = 10;
431 		goto again;
432 	}
433 found:
434 	return found_group;
435 }
436 
437 struct btrfs_block_group_cache *btrfs_find_block_group(struct btrfs_root *root,
438 						 struct btrfs_block_group_cache
439 						 *hint, u64 search_start,
440 						 int data, int owner)
441 {
442 
443 	struct btrfs_block_group_cache *ret;
444 	mutex_lock(&root->fs_info->alloc_mutex);
445 	ret = __btrfs_find_block_group(root, hint, search_start, data, owner);
446 	mutex_unlock(&root->fs_info->alloc_mutex);
447 	return ret;
448 }
449 static u64 hash_extent_ref(u64 root_objectid, u64 ref_generation,
450 			   u64 owner, u64 owner_offset)
451 {
452 	u32 high_crc = ~(u32)0;
453 	u32 low_crc = ~(u32)0;
454 	__le64 lenum;
455 	lenum = cpu_to_le64(root_objectid);
456 	high_crc = btrfs_crc32c(high_crc, &lenum, sizeof(lenum));
457 	lenum = cpu_to_le64(ref_generation);
458 	low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum));
459 	if (owner >= BTRFS_FIRST_FREE_OBJECTID) {
460 		lenum = cpu_to_le64(owner);
461 		low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum));
462 		lenum = cpu_to_le64(owner_offset);
463 		low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum));
464 	}
465 	return ((u64)high_crc << 32) | (u64)low_crc;
466 }
467 
468 static int match_extent_ref(struct extent_buffer *leaf,
469 			    struct btrfs_extent_ref *disk_ref,
470 			    struct btrfs_extent_ref *cpu_ref)
471 {
472 	int ret;
473 	int len;
474 
475 	if (cpu_ref->objectid)
476 		len = sizeof(*cpu_ref);
477 	else
478 		len = 2 * sizeof(u64);
479 	ret = memcmp_extent_buffer(leaf, cpu_ref, (unsigned long)disk_ref,
480 				   len);
481 	return ret == 0;
482 }
483 
484 static int noinline lookup_extent_backref(struct btrfs_trans_handle *trans,
485 					  struct btrfs_root *root,
486 					  struct btrfs_path *path, u64 bytenr,
487 					  u64 root_objectid,
488 					  u64 ref_generation, u64 owner,
489 					  u64 owner_offset, int del)
490 {
491 	u64 hash;
492 	struct btrfs_key key;
493 	struct btrfs_key found_key;
494 	struct btrfs_extent_ref ref;
495 	struct extent_buffer *leaf;
496 	struct btrfs_extent_ref *disk_ref;
497 	int ret;
498 	int ret2;
499 
500 	btrfs_set_stack_ref_root(&ref, root_objectid);
501 	btrfs_set_stack_ref_generation(&ref, ref_generation);
502 	btrfs_set_stack_ref_objectid(&ref, owner);
503 	btrfs_set_stack_ref_offset(&ref, owner_offset);
504 
505 	hash = hash_extent_ref(root_objectid, ref_generation, owner,
506 			       owner_offset);
507 	key.offset = hash;
508 	key.objectid = bytenr;
509 	key.type = BTRFS_EXTENT_REF_KEY;
510 
511 	while (1) {
512 		ret = btrfs_search_slot(trans, root, &key, path,
513 					del ? -1 : 0, del);
514 		if (ret < 0)
515 			goto out;
516 		leaf = path->nodes[0];
517 		if (ret != 0) {
518 			u32 nritems = btrfs_header_nritems(leaf);
519 			if (path->slots[0] >= nritems) {
520 				ret2 = btrfs_next_leaf(root, path);
521 				if (ret2)
522 					goto out;
523 				leaf = path->nodes[0];
524 			}
525 			btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
526 			if (found_key.objectid != bytenr ||
527 			    found_key.type != BTRFS_EXTENT_REF_KEY)
528 				goto out;
529 			key.offset = found_key.offset;
530 			if (del) {
531 				btrfs_release_path(root, path);
532 				continue;
533 			}
534 		}
535 		disk_ref = btrfs_item_ptr(path->nodes[0],
536 					  path->slots[0],
537 					  struct btrfs_extent_ref);
538 		if (match_extent_ref(path->nodes[0], disk_ref, &ref)) {
539 			ret = 0;
540 			goto out;
541 		}
542 		btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
543 		key.offset = found_key.offset + 1;
544 		btrfs_release_path(root, path);
545 	}
546 out:
547 	return ret;
548 }
549 
550 /*
551  * Back reference rules.  Back refs have three main goals:
552  *
553  * 1) differentiate between all holders of references to an extent so that
554  *    when a reference is dropped we can make sure it was a valid reference
555  *    before freeing the extent.
556  *
557  * 2) Provide enough information to quickly find the holders of an extent
558  *    if we notice a given block is corrupted or bad.
559  *
560  * 3) Make it easy to migrate blocks for FS shrinking or storage pool
561  *    maintenance.  This is actually the same as #2, but with a slightly
562  *    different use case.
563  *
564  * File extents can be referenced by:
565  *
566  * - multiple snapshots, subvolumes, or different generations in one subvol
567  * - different files inside a single subvolume (in theory, not implemented yet)
568  * - different offsets inside a file (bookend extents in file.c)
569  *
570  * The extent ref structure has fields for:
571  *
572  * - Objectid of the subvolume root
573  * - Generation number of the tree holding the reference
574  * - objectid of the file holding the reference
575  * - offset in the file corresponding to the key holding the reference
576  *
577  * When a file extent is allocated the fields are filled in:
578  *     (root_key.objectid, trans->transid, inode objectid, offset in file)
579  *
580  * When a leaf is cow'd new references are added for every file extent found
581  * in the leaf.  It looks the same as the create case, but trans->transid
582  * will be different when the block is cow'd.
583  *
584  *     (root_key.objectid, trans->transid, inode objectid, offset in file)
585  *
586  * When a file extent is removed either during snapshot deletion or file
587  * truncation, the corresponding back reference is found
588  * by searching for:
589  *
590  *     (btrfs_header_owner(leaf), btrfs_header_generation(leaf),
591  *      inode objectid, offset in file)
592  *
593  * Btree extents can be referenced by:
594  *
595  * - Different subvolumes
596  * - Different generations of the same subvolume
597  *
598  * Storing sufficient information for a full reverse mapping of a btree
599  * block would require storing the lowest key of the block in the backref,
600  * and it would require updating that lowest key either before write out or
601  * every time it changed.  Instead, the objectid of the lowest key is stored
602  * along with the level of the tree block.  This provides a hint
603  * about where in the btree the block can be found.  Searches through the
604  * btree only need to look for a pointer to that block, so they stop one
605  * level higher than the level recorded in the backref.
606  *
607  * Some btrees do not do reference counting on their extents.  These
608  * include the extent tree and the tree of tree roots.  Backrefs for these
609  * trees always have a generation of zero.
610  *
611  * When a tree block is created, back references are inserted:
612  *
613  * (root->root_key.objectid, trans->transid or zero, level, lowest_key_objectid)
614  *
615  * When a tree block is cow'd in a reference counted root,
616  * new back references are added for all the blocks it points to.
617  * These are of the form (trans->transid will have increased since creation):
618  *
619  * (root->root_key.objectid, trans->transid, level, lowest_key_objectid)
620  *
621  * Because the lowest_key_objectid and the level are just hints
622  * they are not used when backrefs are deleted.  When a backref is deleted:
623  *
624  * if backref was for a tree root:
625  *     root_objectid = root->root_key.objectid
626  * else
627  *     root_objectid = btrfs_header_owner(parent)
628  *
629  * (root_objectid, btrfs_header_generation(parent) or zero, 0, 0)
630  *
631  * Back Reference Key hashing:
632  *
633  * Back references have four fields, each 64 bits long.  Unfortunately,
634  * This is hashed into a single 64 bit number and placed into the key offset.
635  * The key objectid corresponds to the first byte in the extent, and the
636  * key type is set to BTRFS_EXTENT_REF_KEY
637  */
638 int btrfs_insert_extent_backref(struct btrfs_trans_handle *trans,
639 				 struct btrfs_root *root,
640 				 struct btrfs_path *path, u64 bytenr,
641 				 u64 root_objectid, u64 ref_generation,
642 				 u64 owner, u64 owner_offset)
643 {
644 	u64 hash;
645 	struct btrfs_key key;
646 	struct btrfs_extent_ref ref;
647 	struct btrfs_extent_ref *disk_ref;
648 	int ret;
649 
650 	btrfs_set_stack_ref_root(&ref, root_objectid);
651 	btrfs_set_stack_ref_generation(&ref, ref_generation);
652 	btrfs_set_stack_ref_objectid(&ref, owner);
653 	btrfs_set_stack_ref_offset(&ref, owner_offset);
654 
655 	hash = hash_extent_ref(root_objectid, ref_generation, owner,
656 			       owner_offset);
657 	key.offset = hash;
658 	key.objectid = bytenr;
659 	key.type = BTRFS_EXTENT_REF_KEY;
660 
661 	ret = btrfs_insert_empty_item(trans, root, path, &key, sizeof(ref));
662 	while (ret == -EEXIST) {
663 		disk_ref = btrfs_item_ptr(path->nodes[0], path->slots[0],
664 					  struct btrfs_extent_ref);
665 		if (match_extent_ref(path->nodes[0], disk_ref, &ref))
666 			goto out;
667 		key.offset++;
668 		btrfs_release_path(root, path);
669 		ret = btrfs_insert_empty_item(trans, root, path, &key,
670 					      sizeof(ref));
671 	}
672 	if (ret)
673 		goto out;
674 	disk_ref = btrfs_item_ptr(path->nodes[0], path->slots[0],
675 				  struct btrfs_extent_ref);
676 	write_extent_buffer(path->nodes[0], &ref, (unsigned long)disk_ref,
677 			    sizeof(ref));
678 	btrfs_mark_buffer_dirty(path->nodes[0]);
679 out:
680 	btrfs_release_path(root, path);
681 	return ret;
682 }
683 
684 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
685 				struct btrfs_root *root,
686 				u64 bytenr, u64 num_bytes,
687 				u64 root_objectid, u64 ref_generation,
688 				u64 owner, u64 owner_offset)
689 {
690 	struct btrfs_path *path;
691 	int ret;
692 	struct btrfs_key key;
693 	struct extent_buffer *l;
694 	struct btrfs_extent_item *item;
695 	u32 refs;
696 
697 	WARN_ON(num_bytes < root->sectorsize);
698 	path = btrfs_alloc_path();
699 	if (!path)
700 		return -ENOMEM;
701 
702 	path->reada = 1;
703 	key.objectid = bytenr;
704 	btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
705 	key.offset = num_bytes;
706 	ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path,
707 				0, 1);
708 	if (ret < 0)
709 		return ret;
710 	if (ret != 0) {
711 		BUG();
712 	}
713 	BUG_ON(ret != 0);
714 	l = path->nodes[0];
715 	item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
716 	refs = btrfs_extent_refs(l, item);
717 	btrfs_set_extent_refs(l, item, refs + 1);
718 	btrfs_mark_buffer_dirty(path->nodes[0]);
719 
720 	btrfs_release_path(root->fs_info->extent_root, path);
721 
722 	path->reada = 1;
723 	ret = btrfs_insert_extent_backref(trans, root->fs_info->extent_root,
724 					  path, bytenr, root_objectid,
725 					  ref_generation, owner, owner_offset);
726 	BUG_ON(ret);
727 	finish_current_insert(trans, root->fs_info->extent_root);
728 	del_pending_extents(trans, root->fs_info->extent_root);
729 
730 	btrfs_free_path(path);
731 	return 0;
732 }
733 
734 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
735 				struct btrfs_root *root,
736 				u64 bytenr, u64 num_bytes,
737 				u64 root_objectid, u64 ref_generation,
738 				u64 owner, u64 owner_offset)
739 {
740 	int ret;
741 
742 	mutex_lock(&root->fs_info->alloc_mutex);
743 	ret = __btrfs_inc_extent_ref(trans, root, bytenr, num_bytes,
744 				     root_objectid, ref_generation,
745 				     owner, owner_offset);
746 	mutex_unlock(&root->fs_info->alloc_mutex);
747 	return ret;
748 }
749 
750 int btrfs_extent_post_op(struct btrfs_trans_handle *trans,
751 			 struct btrfs_root *root)
752 {
753 	finish_current_insert(trans, root->fs_info->extent_root);
754 	del_pending_extents(trans, root->fs_info->extent_root);
755 	return 0;
756 }
757 
758 static int lookup_extent_ref(struct btrfs_trans_handle *trans,
759 			     struct btrfs_root *root, u64 bytenr,
760 			     u64 num_bytes, u32 *refs)
761 {
762 	struct btrfs_path *path;
763 	int ret;
764 	struct btrfs_key key;
765 	struct extent_buffer *l;
766 	struct btrfs_extent_item *item;
767 
768 	WARN_ON(num_bytes < root->sectorsize);
769 	path = btrfs_alloc_path();
770 	path->reada = 1;
771 	key.objectid = bytenr;
772 	key.offset = num_bytes;
773 	btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
774 	ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path,
775 				0, 0);
776 	if (ret < 0)
777 		goto out;
778 	if (ret != 0) {
779 		btrfs_print_leaf(root, path->nodes[0]);
780 		printk("failed to find block number %Lu\n", bytenr);
781 		BUG();
782 	}
783 	l = path->nodes[0];
784 	item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
785 	*refs = btrfs_extent_refs(l, item);
786 out:
787 	btrfs_free_path(path);
788 	return 0;
789 }
790 
791 u32 btrfs_count_snapshots_in_path(struct btrfs_root *root,
792 				  struct btrfs_path *count_path,
793 				  u64 expected_owner,
794 				  u64 first_extent)
795 {
796 	struct btrfs_root *extent_root = root->fs_info->extent_root;
797 	struct btrfs_path *path;
798 	u64 bytenr;
799 	u64 found_objectid;
800 	u64 found_owner;
801 	u64 root_objectid = root->root_key.objectid;
802 	u32 total_count = 0;
803 	u32 extent_refs;
804 	u32 cur_count;
805 	u32 nritems;
806 	int ret;
807 	struct btrfs_key key;
808 	struct btrfs_key found_key;
809 	struct extent_buffer *l;
810 	struct btrfs_extent_item *item;
811 	struct btrfs_extent_ref *ref_item;
812 	int level = -1;
813 
814 	/* FIXME, needs locking */
815 	BUG();
816 
817 	mutex_lock(&root->fs_info->alloc_mutex);
818 	path = btrfs_alloc_path();
819 again:
820 	if (level == -1)
821 		bytenr = first_extent;
822 	else
823 		bytenr = count_path->nodes[level]->start;
824 
825 	cur_count = 0;
826 	key.objectid = bytenr;
827 	key.offset = 0;
828 
829 	btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
830 	ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
831 	if (ret < 0)
832 		goto out;
833 	BUG_ON(ret == 0);
834 
835 	l = path->nodes[0];
836 	btrfs_item_key_to_cpu(l, &found_key, path->slots[0]);
837 
838 	if (found_key.objectid != bytenr ||
839 	    found_key.type != BTRFS_EXTENT_ITEM_KEY) {
840 		goto out;
841 	}
842 
843 	item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
844 	extent_refs = btrfs_extent_refs(l, item);
845 	while (1) {
846 		l = path->nodes[0];
847 		nritems = btrfs_header_nritems(l);
848 		if (path->slots[0] >= nritems) {
849 			ret = btrfs_next_leaf(extent_root, path);
850 			if (ret == 0)
851 				continue;
852 			break;
853 		}
854 		btrfs_item_key_to_cpu(l, &found_key, path->slots[0]);
855 		if (found_key.objectid != bytenr)
856 			break;
857 
858 		if (found_key.type != BTRFS_EXTENT_REF_KEY) {
859 			path->slots[0]++;
860 			continue;
861 		}
862 
863 		cur_count++;
864 		ref_item = btrfs_item_ptr(l, path->slots[0],
865 					  struct btrfs_extent_ref);
866 		found_objectid = btrfs_ref_root(l, ref_item);
867 
868 		if (found_objectid != root_objectid) {
869 			total_count = 2;
870 			goto out;
871 		}
872 		if (level == -1) {
873 			found_owner = btrfs_ref_objectid(l, ref_item);
874 			if (found_owner != expected_owner) {
875 				total_count = 2;
876 				goto out;
877 			}
878 			/*
879 			 * nasty.  we don't count a reference held by
880 			 * the running transaction.  This allows nodatacow
881 			 * to avoid cow most of the time
882 			 */
883 			if (found_owner >= BTRFS_FIRST_FREE_OBJECTID &&
884 			    btrfs_ref_generation(l, ref_item) ==
885 			    root->fs_info->generation) {
886 				extent_refs--;
887 			}
888 		}
889 		total_count = 1;
890 		path->slots[0]++;
891 	}
892 	/*
893 	 * if there is more than one reference against a data extent,
894 	 * we have to assume the other ref is another snapshot
895 	 */
896 	if (level == -1 && extent_refs > 1) {
897 		total_count = 2;
898 		goto out;
899 	}
900 	if (cur_count == 0) {
901 		total_count = 0;
902 		goto out;
903 	}
904 	if (level >= 0 && root->node == count_path->nodes[level])
905 		goto out;
906 	level++;
907 	btrfs_release_path(root, path);
908 	goto again;
909 
910 out:
911 	btrfs_free_path(path);
912 	mutex_unlock(&root->fs_info->alloc_mutex);
913 	return total_count;
914 }
915 
916 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
917 		  struct extent_buffer *buf)
918 {
919 	u64 bytenr;
920 	u32 nritems;
921 	struct btrfs_key key;
922 	struct btrfs_file_extent_item *fi;
923 	int i;
924 	int level;
925 	int ret;
926 	int faili;
927 
928 	if (!root->ref_cows)
929 		return 0;
930 
931 	mutex_lock(&root->fs_info->alloc_mutex);
932 	level = btrfs_header_level(buf);
933 	nritems = btrfs_header_nritems(buf);
934 	for (i = 0; i < nritems; i++) {
935 		if (level == 0) {
936 			u64 disk_bytenr;
937 			btrfs_item_key_to_cpu(buf, &key, i);
938 			if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
939 				continue;
940 			fi = btrfs_item_ptr(buf, i,
941 					    struct btrfs_file_extent_item);
942 			if (btrfs_file_extent_type(buf, fi) ==
943 			    BTRFS_FILE_EXTENT_INLINE)
944 				continue;
945 			disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
946 			if (disk_bytenr == 0)
947 				continue;
948 			ret = __btrfs_inc_extent_ref(trans, root, disk_bytenr,
949 				    btrfs_file_extent_disk_num_bytes(buf, fi),
950 				    root->root_key.objectid, trans->transid,
951 				    key.objectid, key.offset);
952 			if (ret) {
953 				faili = i;
954 				goto fail;
955 			}
956 		} else {
957 			bytenr = btrfs_node_blockptr(buf, i);
958 			btrfs_node_key_to_cpu(buf, &key, i);
959 			ret = __btrfs_inc_extent_ref(trans, root, bytenr,
960 					   btrfs_level_size(root, level - 1),
961 					   root->root_key.objectid,
962 					   trans->transid,
963 					   level - 1, key.objectid);
964 			if (ret) {
965 				faili = i;
966 				goto fail;
967 			}
968 		}
969 	}
970 	mutex_unlock(&root->fs_info->alloc_mutex);
971 	return 0;
972 fail:
973 	WARN_ON(1);
974 #if 0
975 	for (i =0; i < faili; i++) {
976 		if (level == 0) {
977 			u64 disk_bytenr;
978 			btrfs_item_key_to_cpu(buf, &key, i);
979 			if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
980 				continue;
981 			fi = btrfs_item_ptr(buf, i,
982 					    struct btrfs_file_extent_item);
983 			if (btrfs_file_extent_type(buf, fi) ==
984 			    BTRFS_FILE_EXTENT_INLINE)
985 				continue;
986 			disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
987 			if (disk_bytenr == 0)
988 				continue;
989 			err = btrfs_free_extent(trans, root, disk_bytenr,
990 				    btrfs_file_extent_disk_num_bytes(buf,
991 								      fi), 0);
992 			BUG_ON(err);
993 		} else {
994 			bytenr = btrfs_node_blockptr(buf, i);
995 			err = btrfs_free_extent(trans, root, bytenr,
996 					btrfs_level_size(root, level - 1), 0);
997 			BUG_ON(err);
998 		}
999 	}
1000 #endif
1001 	mutex_unlock(&root->fs_info->alloc_mutex);
1002 	return ret;
1003 }
1004 
1005 static int write_one_cache_group(struct btrfs_trans_handle *trans,
1006 				 struct btrfs_root *root,
1007 				 struct btrfs_path *path,
1008 				 struct btrfs_block_group_cache *cache)
1009 {
1010 	int ret;
1011 	int pending_ret;
1012 	struct btrfs_root *extent_root = root->fs_info->extent_root;
1013 	unsigned long bi;
1014 	struct extent_buffer *leaf;
1015 
1016 	ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
1017 	if (ret < 0)
1018 		goto fail;
1019 	BUG_ON(ret);
1020 
1021 	leaf = path->nodes[0];
1022 	bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
1023 	write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
1024 	btrfs_mark_buffer_dirty(leaf);
1025 	btrfs_release_path(extent_root, path);
1026 fail:
1027 	finish_current_insert(trans, extent_root);
1028 	pending_ret = del_pending_extents(trans, extent_root);
1029 	if (ret)
1030 		return ret;
1031 	if (pending_ret)
1032 		return pending_ret;
1033 	return 0;
1034 
1035 }
1036 
1037 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
1038 				   struct btrfs_root *root)
1039 {
1040 	struct extent_io_tree *block_group_cache;
1041 	struct btrfs_block_group_cache *cache;
1042 	int ret;
1043 	int err = 0;
1044 	int werr = 0;
1045 	struct btrfs_path *path;
1046 	u64 last = 0;
1047 	u64 start;
1048 	u64 end;
1049 	u64 ptr;
1050 
1051 	block_group_cache = &root->fs_info->block_group_cache;
1052 	path = btrfs_alloc_path();
1053 	if (!path)
1054 		return -ENOMEM;
1055 
1056 	mutex_lock(&root->fs_info->alloc_mutex);
1057 	while(1) {
1058 		ret = find_first_extent_bit(block_group_cache, last,
1059 					    &start, &end, BLOCK_GROUP_DIRTY);
1060 		if (ret)
1061 			break;
1062 
1063 		last = end + 1;
1064 		ret = get_state_private(block_group_cache, start, &ptr);
1065 		if (ret)
1066 			break;
1067 		cache = (struct btrfs_block_group_cache *)(unsigned long)ptr;
1068 		err = write_one_cache_group(trans, root,
1069 					    path, cache);
1070 		/*
1071 		 * if we fail to write the cache group, we want
1072 		 * to keep it marked dirty in hopes that a later
1073 		 * write will work
1074 		 */
1075 		if (err) {
1076 			werr = err;
1077 			continue;
1078 		}
1079 		clear_extent_bits(block_group_cache, start, end,
1080 				  BLOCK_GROUP_DIRTY, GFP_NOFS);
1081 	}
1082 	btrfs_free_path(path);
1083 	mutex_unlock(&root->fs_info->alloc_mutex);
1084 	return werr;
1085 }
1086 
1087 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
1088 						  u64 flags)
1089 {
1090 	struct list_head *head = &info->space_info;
1091 	struct list_head *cur;
1092 	struct btrfs_space_info *found;
1093 	list_for_each(cur, head) {
1094 		found = list_entry(cur, struct btrfs_space_info, list);
1095 		if (found->flags == flags)
1096 			return found;
1097 	}
1098 	return NULL;
1099 
1100 }
1101 
1102 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
1103 			     u64 total_bytes, u64 bytes_used,
1104 			     struct btrfs_space_info **space_info)
1105 {
1106 	struct btrfs_space_info *found;
1107 
1108 	found = __find_space_info(info, flags);
1109 	if (found) {
1110 		found->total_bytes += total_bytes;
1111 		found->bytes_used += bytes_used;
1112 		found->full = 0;
1113 		WARN_ON(found->total_bytes < found->bytes_used);
1114 		*space_info = found;
1115 		return 0;
1116 	}
1117 	found = kmalloc(sizeof(*found), GFP_NOFS);
1118 	if (!found)
1119 		return -ENOMEM;
1120 
1121 	list_add(&found->list, &info->space_info);
1122 	found->flags = flags;
1123 	found->total_bytes = total_bytes;
1124 	found->bytes_used = bytes_used;
1125 	found->bytes_pinned = 0;
1126 	found->full = 0;
1127 	found->force_alloc = 0;
1128 	*space_info = found;
1129 	return 0;
1130 }
1131 
1132 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
1133 {
1134 	u64 extra_flags = flags & (BTRFS_BLOCK_GROUP_RAID0 |
1135 				   BTRFS_BLOCK_GROUP_RAID1 |
1136 				   BTRFS_BLOCK_GROUP_RAID10 |
1137 				   BTRFS_BLOCK_GROUP_DUP);
1138 	if (extra_flags) {
1139 		if (flags & BTRFS_BLOCK_GROUP_DATA)
1140 			fs_info->avail_data_alloc_bits |= extra_flags;
1141 		if (flags & BTRFS_BLOCK_GROUP_METADATA)
1142 			fs_info->avail_metadata_alloc_bits |= extra_flags;
1143 		if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
1144 			fs_info->avail_system_alloc_bits |= extra_flags;
1145 	}
1146 }
1147 
1148 static u64 reduce_alloc_profile(struct btrfs_root *root, u64 flags)
1149 {
1150 	u64 num_devices = root->fs_info->fs_devices->num_devices;
1151 
1152 	if (num_devices == 1)
1153 		flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0);
1154 	if (num_devices < 4)
1155 		flags &= ~BTRFS_BLOCK_GROUP_RAID10;
1156 
1157 	if ((flags & BTRFS_BLOCK_GROUP_DUP) &&
1158 	    (flags & (BTRFS_BLOCK_GROUP_RAID1 |
1159 		      BTRFS_BLOCK_GROUP_RAID10))) {
1160 		flags &= ~BTRFS_BLOCK_GROUP_DUP;
1161 	}
1162 
1163 	if ((flags & BTRFS_BLOCK_GROUP_RAID1) &&
1164 	    (flags & BTRFS_BLOCK_GROUP_RAID10)) {
1165 		flags &= ~BTRFS_BLOCK_GROUP_RAID1;
1166 	}
1167 
1168 	if ((flags & BTRFS_BLOCK_GROUP_RAID0) &&
1169 	    ((flags & BTRFS_BLOCK_GROUP_RAID1) |
1170 	     (flags & BTRFS_BLOCK_GROUP_RAID10) |
1171 	     (flags & BTRFS_BLOCK_GROUP_DUP)))
1172 		flags &= ~BTRFS_BLOCK_GROUP_RAID0;
1173 	return flags;
1174 }
1175 
1176 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
1177 			  struct btrfs_root *extent_root, u64 alloc_bytes,
1178 			  u64 flags, int force)
1179 {
1180 	struct btrfs_space_info *space_info;
1181 	u64 thresh;
1182 	u64 start;
1183 	u64 num_bytes;
1184 	int ret;
1185 
1186 	flags = reduce_alloc_profile(extent_root, flags);
1187 
1188 	space_info = __find_space_info(extent_root->fs_info, flags);
1189 	if (!space_info) {
1190 		ret = update_space_info(extent_root->fs_info, flags,
1191 					0, 0, &space_info);
1192 		BUG_ON(ret);
1193 	}
1194 	BUG_ON(!space_info);
1195 
1196 	if (space_info->force_alloc) {
1197 		force = 1;
1198 		space_info->force_alloc = 0;
1199 	}
1200 	if (space_info->full)
1201 		goto out;
1202 
1203 	thresh = div_factor(space_info->total_bytes, 6);
1204 	if (!force &&
1205 	   (space_info->bytes_used + space_info->bytes_pinned + alloc_bytes) <
1206 	    thresh)
1207 		goto out;
1208 
1209 	mutex_lock(&extent_root->fs_info->chunk_mutex);
1210 	ret = btrfs_alloc_chunk(trans, extent_root, &start, &num_bytes, flags);
1211 	if (ret == -ENOSPC) {
1212 printk("space info full %Lu\n", flags);
1213 		space_info->full = 1;
1214 		goto out;
1215 	}
1216 	BUG_ON(ret);
1217 
1218 	ret = btrfs_make_block_group(trans, extent_root, 0, flags,
1219 		     BTRFS_FIRST_CHUNK_TREE_OBJECTID, start, num_bytes);
1220 	BUG_ON(ret);
1221 	mutex_unlock(&extent_root->fs_info->chunk_mutex);
1222 out:
1223 	return 0;
1224 }
1225 
1226 static int update_block_group(struct btrfs_trans_handle *trans,
1227 			      struct btrfs_root *root,
1228 			      u64 bytenr, u64 num_bytes, int alloc,
1229 			      int mark_free)
1230 {
1231 	struct btrfs_block_group_cache *cache;
1232 	struct btrfs_fs_info *info = root->fs_info;
1233 	u64 total = num_bytes;
1234 	u64 old_val;
1235 	u64 byte_in_group;
1236 	u64 start;
1237 	u64 end;
1238 
1239 	while(total) {
1240 		cache = btrfs_lookup_block_group(info, bytenr);
1241 		if (!cache) {
1242 			return -1;
1243 		}
1244 		byte_in_group = bytenr - cache->key.objectid;
1245 		WARN_ON(byte_in_group > cache->key.offset);
1246 		start = cache->key.objectid;
1247 		end = start + cache->key.offset - 1;
1248 		set_extent_bits(&info->block_group_cache, start, end,
1249 				BLOCK_GROUP_DIRTY, GFP_NOFS);
1250 
1251 		old_val = btrfs_block_group_used(&cache->item);
1252 		num_bytes = min(total, cache->key.offset - byte_in_group);
1253 		if (alloc) {
1254 			old_val += num_bytes;
1255 			cache->space_info->bytes_used += num_bytes;
1256 		} else {
1257 			old_val -= num_bytes;
1258 			cache->space_info->bytes_used -= num_bytes;
1259 			if (mark_free) {
1260 				set_extent_dirty(&info->free_space_cache,
1261 						 bytenr, bytenr + num_bytes - 1,
1262 						 GFP_NOFS);
1263 			}
1264 		}
1265 		btrfs_set_block_group_used(&cache->item, old_val);
1266 		total -= num_bytes;
1267 		bytenr += num_bytes;
1268 	}
1269 	return 0;
1270 }
1271 
1272 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
1273 {
1274 	u64 start;
1275 	u64 end;
1276 	int ret;
1277 	ret = find_first_extent_bit(&root->fs_info->block_group_cache,
1278 				    search_start, &start, &end,
1279 				    BLOCK_GROUP_DATA | BLOCK_GROUP_METADATA |
1280 				    BLOCK_GROUP_SYSTEM);
1281 	if (ret)
1282 		return 0;
1283 	return start;
1284 }
1285 
1286 
1287 static int update_pinned_extents(struct btrfs_root *root,
1288 				u64 bytenr, u64 num, int pin)
1289 {
1290 	u64 len;
1291 	struct btrfs_block_group_cache *cache;
1292 	struct btrfs_fs_info *fs_info = root->fs_info;
1293 
1294 	if (pin) {
1295 		set_extent_dirty(&fs_info->pinned_extents,
1296 				bytenr, bytenr + num - 1, GFP_NOFS);
1297 	} else {
1298 		clear_extent_dirty(&fs_info->pinned_extents,
1299 				bytenr, bytenr + num - 1, GFP_NOFS);
1300 	}
1301 	while (num > 0) {
1302 		cache = btrfs_lookup_block_group(fs_info, bytenr);
1303 		if (!cache) {
1304 			u64 first = first_logical_byte(root, bytenr);
1305 			WARN_ON(first < bytenr);
1306 			len = min(first - bytenr, num);
1307 		} else {
1308 			len = min(num, cache->key.offset -
1309 				  (bytenr - cache->key.objectid));
1310 		}
1311 		if (pin) {
1312 			if (cache) {
1313 				cache->pinned += len;
1314 				cache->space_info->bytes_pinned += len;
1315 			}
1316 			fs_info->total_pinned += len;
1317 		} else {
1318 			if (cache) {
1319 				cache->pinned -= len;
1320 				cache->space_info->bytes_pinned -= len;
1321 			}
1322 			fs_info->total_pinned -= len;
1323 		}
1324 		bytenr += len;
1325 		num -= len;
1326 	}
1327 	return 0;
1328 }
1329 
1330 int btrfs_copy_pinned(struct btrfs_root *root, struct extent_io_tree *copy)
1331 {
1332 	u64 last = 0;
1333 	u64 start;
1334 	u64 end;
1335 	struct extent_io_tree *pinned_extents = &root->fs_info->pinned_extents;
1336 	int ret;
1337 
1338 	while(1) {
1339 		ret = find_first_extent_bit(pinned_extents, last,
1340 					    &start, &end, EXTENT_DIRTY);
1341 		if (ret)
1342 			break;
1343 		set_extent_dirty(copy, start, end, GFP_NOFS);
1344 		last = end + 1;
1345 	}
1346 	return 0;
1347 }
1348 
1349 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
1350 			       struct btrfs_root *root,
1351 			       struct extent_io_tree *unpin)
1352 {
1353 	u64 start;
1354 	u64 end;
1355 	int ret;
1356 	struct extent_io_tree *free_space_cache;
1357 	free_space_cache = &root->fs_info->free_space_cache;
1358 
1359 	mutex_lock(&root->fs_info->alloc_mutex);
1360 	while(1) {
1361 		ret = find_first_extent_bit(unpin, 0, &start, &end,
1362 					    EXTENT_DIRTY);
1363 		if (ret)
1364 			break;
1365 		update_pinned_extents(root, start, end + 1 - start, 0);
1366 		clear_extent_dirty(unpin, start, end, GFP_NOFS);
1367 		set_extent_dirty(free_space_cache, start, end, GFP_NOFS);
1368 	}
1369 	mutex_unlock(&root->fs_info->alloc_mutex);
1370 	return 0;
1371 }
1372 
1373 static int finish_current_insert(struct btrfs_trans_handle *trans,
1374 				 struct btrfs_root *extent_root)
1375 {
1376 	u64 start;
1377 	u64 end;
1378 	struct btrfs_fs_info *info = extent_root->fs_info;
1379 	struct extent_buffer *eb;
1380 	struct btrfs_path *path;
1381 	struct btrfs_key ins;
1382 	struct btrfs_disk_key first;
1383 	struct btrfs_extent_item extent_item;
1384 	int ret;
1385 	int level;
1386 	int err = 0;
1387 
1388 	btrfs_set_stack_extent_refs(&extent_item, 1);
1389 	btrfs_set_key_type(&ins, BTRFS_EXTENT_ITEM_KEY);
1390 	path = btrfs_alloc_path();
1391 
1392 	while(1) {
1393 		ret = find_first_extent_bit(&info->extent_ins, 0, &start,
1394 					    &end, EXTENT_LOCKED);
1395 		if (ret)
1396 			break;
1397 
1398 		ins.objectid = start;
1399 		ins.offset = end + 1 - start;
1400 		err = btrfs_insert_item(trans, extent_root, &ins,
1401 					&extent_item, sizeof(extent_item));
1402 		clear_extent_bits(&info->extent_ins, start, end, EXTENT_LOCKED,
1403 				  GFP_NOFS);
1404 		eb = read_tree_block(extent_root, ins.objectid, ins.offset,
1405 				     trans->transid);
1406 		btrfs_tree_lock(eb);
1407 		level = btrfs_header_level(eb);
1408 		if (level == 0) {
1409 			btrfs_item_key(eb, &first, 0);
1410 		} else {
1411 			btrfs_node_key(eb, &first, 0);
1412 		}
1413 		btrfs_tree_unlock(eb);
1414 		free_extent_buffer(eb);
1415 		/*
1416 		 * the first key is just a hint, so the race we've created
1417 		 * against reading it is fine
1418 		 */
1419 		err = btrfs_insert_extent_backref(trans, extent_root, path,
1420 					  start, extent_root->root_key.objectid,
1421 					  0, level,
1422 					  btrfs_disk_key_objectid(&first));
1423 		BUG_ON(err);
1424 	}
1425 	btrfs_free_path(path);
1426 	return 0;
1427 }
1428 
1429 static int pin_down_bytes(struct btrfs_root *root, u64 bytenr, u32 num_bytes,
1430 			  int pending)
1431 {
1432 	int err = 0;
1433 
1434 	if (!pending) {
1435 #if 0
1436 		struct extent_buffer *buf;
1437 		buf = btrfs_find_tree_block(root, bytenr, num_bytes);
1438 		if (buf) {
1439 			if (!btrfs_try_tree_lock(buf) &&
1440 			    btrfs_buffer_uptodate(buf, 0)) {
1441 				u64 transid =
1442 				    root->fs_info->running_transaction->transid;
1443 				u64 header_transid =
1444 					btrfs_header_generation(buf);
1445 				if (header_transid == transid &&
1446 				    !btrfs_header_flag(buf,
1447 					       BTRFS_HEADER_FLAG_WRITTEN)) {
1448 					clean_tree_block(NULL, root, buf);
1449 					btrfs_tree_unlock(buf);
1450 					free_extent_buffer(buf);
1451 					return 1;
1452 				}
1453 				btrfs_tree_unlock(buf);
1454 			}
1455 			free_extent_buffer(buf);
1456 		}
1457 #endif
1458 		update_pinned_extents(root, bytenr, num_bytes, 1);
1459 	} else {
1460 		set_extent_bits(&root->fs_info->pending_del,
1461 				bytenr, bytenr + num_bytes - 1,
1462 				EXTENT_LOCKED, GFP_NOFS);
1463 	}
1464 	BUG_ON(err < 0);
1465 	return 0;
1466 }
1467 
1468 /*
1469  * remove an extent from the root, returns 0 on success
1470  */
1471 static int __free_extent(struct btrfs_trans_handle *trans, struct btrfs_root
1472 			 *root, u64 bytenr, u64 num_bytes,
1473 			 u64 root_objectid, u64 ref_generation,
1474 			 u64 owner_objectid, u64 owner_offset, int pin,
1475 			 int mark_free)
1476 {
1477 	struct btrfs_path *path;
1478 	struct btrfs_key key;
1479 	struct btrfs_fs_info *info = root->fs_info;
1480 	struct btrfs_root *extent_root = info->extent_root;
1481 	struct extent_buffer *leaf;
1482 	int ret;
1483 	int extent_slot = 0;
1484 	int found_extent = 0;
1485 	int num_to_del = 1;
1486 	struct btrfs_extent_item *ei;
1487 	u32 refs;
1488 
1489 	key.objectid = bytenr;
1490 	btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
1491 	key.offset = num_bytes;
1492 	path = btrfs_alloc_path();
1493 	if (!path)
1494 		return -ENOMEM;
1495 
1496 	path->reada = 1;
1497 	ret = lookup_extent_backref(trans, extent_root, path,
1498 				    bytenr, root_objectid,
1499 				    ref_generation,
1500 				    owner_objectid, owner_offset, 1);
1501 	if (ret == 0) {
1502 		struct btrfs_key found_key;
1503 		extent_slot = path->slots[0];
1504 		while(extent_slot > 0) {
1505 			extent_slot--;
1506 			btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1507 					      extent_slot);
1508 			if (found_key.objectid != bytenr)
1509 				break;
1510 			if (found_key.type == BTRFS_EXTENT_ITEM_KEY &&
1511 			    found_key.offset == num_bytes) {
1512 				found_extent = 1;
1513 				break;
1514 			}
1515 			if (path->slots[0] - extent_slot > 5)
1516 				break;
1517 		}
1518 		if (!found_extent)
1519 			ret = btrfs_del_item(trans, extent_root, path);
1520 	} else {
1521 		btrfs_print_leaf(extent_root, path->nodes[0]);
1522 		WARN_ON(1);
1523 		printk("Unable to find ref byte nr %Lu root %Lu "
1524 		       " gen %Lu owner %Lu offset %Lu\n", bytenr,
1525 		       root_objectid, ref_generation, owner_objectid,
1526 		       owner_offset);
1527 	}
1528 	if (!found_extent) {
1529 		btrfs_release_path(extent_root, path);
1530 		ret = btrfs_search_slot(trans, extent_root, &key, path, -1, 1);
1531 		if (ret < 0)
1532 			return ret;
1533 		BUG_ON(ret);
1534 		extent_slot = path->slots[0];
1535 	}
1536 
1537 	leaf = path->nodes[0];
1538 	ei = btrfs_item_ptr(leaf, extent_slot,
1539 			    struct btrfs_extent_item);
1540 	refs = btrfs_extent_refs(leaf, ei);
1541 	BUG_ON(refs == 0);
1542 	refs -= 1;
1543 	btrfs_set_extent_refs(leaf, ei, refs);
1544 
1545 	btrfs_mark_buffer_dirty(leaf);
1546 
1547 	if (refs == 0 && found_extent && path->slots[0] == extent_slot + 1) {
1548 		/* if the back ref and the extent are next to each other
1549 		 * they get deleted below in one shot
1550 		 */
1551 		path->slots[0] = extent_slot;
1552 		num_to_del = 2;
1553 	} else if (found_extent) {
1554 		/* otherwise delete the extent back ref */
1555 		ret = btrfs_del_item(trans, extent_root, path);
1556 		BUG_ON(ret);
1557 		/* if refs are 0, we need to setup the path for deletion */
1558 		if (refs == 0) {
1559 			btrfs_release_path(extent_root, path);
1560 			ret = btrfs_search_slot(trans, extent_root, &key, path,
1561 						-1, 1);
1562 			if (ret < 0)
1563 				return ret;
1564 			BUG_ON(ret);
1565 		}
1566 	}
1567 
1568 	if (refs == 0) {
1569 		u64 super_used;
1570 		u64 root_used;
1571 
1572 		if (pin) {
1573 			ret = pin_down_bytes(root, bytenr, num_bytes, 0);
1574 			if (ret > 0)
1575 				mark_free = 1;
1576 			BUG_ON(ret < 0);
1577 		}
1578 
1579 		/* block accounting for super block */
1580 		spin_lock_irq(&info->delalloc_lock);
1581 		super_used = btrfs_super_bytes_used(&info->super_copy);
1582 		btrfs_set_super_bytes_used(&info->super_copy,
1583 					   super_used - num_bytes);
1584 		spin_unlock_irq(&info->delalloc_lock);
1585 
1586 		/* block accounting for root item */
1587 		root_used = btrfs_root_used(&root->root_item);
1588 		btrfs_set_root_used(&root->root_item,
1589 					   root_used - num_bytes);
1590 		ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
1591 				      num_to_del);
1592 		if (ret) {
1593 			return ret;
1594 		}
1595 		ret = update_block_group(trans, root, bytenr, num_bytes, 0,
1596 					 mark_free);
1597 		BUG_ON(ret);
1598 	}
1599 	btrfs_free_path(path);
1600 	finish_current_insert(trans, extent_root);
1601 	return ret;
1602 }
1603 
1604 /*
1605  * find all the blocks marked as pending in the radix tree and remove
1606  * them from the extent map
1607  */
1608 static int del_pending_extents(struct btrfs_trans_handle *trans, struct
1609 			       btrfs_root *extent_root)
1610 {
1611 	int ret;
1612 	int err = 0;
1613 	u64 start;
1614 	u64 end;
1615 	struct extent_io_tree *pending_del;
1616 	struct extent_io_tree *pinned_extents;
1617 
1618 	pending_del = &extent_root->fs_info->pending_del;
1619 	pinned_extents = &extent_root->fs_info->pinned_extents;
1620 
1621 	while(1) {
1622 		ret = find_first_extent_bit(pending_del, 0, &start, &end,
1623 					    EXTENT_LOCKED);
1624 		if (ret)
1625 			break;
1626 		update_pinned_extents(extent_root, start, end + 1 - start, 1);
1627 		clear_extent_bits(pending_del, start, end, EXTENT_LOCKED,
1628 				  GFP_NOFS);
1629 		ret = __free_extent(trans, extent_root,
1630 				     start, end + 1 - start,
1631 				     extent_root->root_key.objectid,
1632 				     0, 0, 0, 0, 0);
1633 		if (ret)
1634 			err = ret;
1635 	}
1636 	return err;
1637 }
1638 
1639 /*
1640  * remove an extent from the root, returns 0 on success
1641  */
1642 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
1643 			       struct btrfs_root *root, u64 bytenr,
1644 			       u64 num_bytes, u64 root_objectid,
1645 			       u64 ref_generation, u64 owner_objectid,
1646 			       u64 owner_offset, int pin)
1647 {
1648 	struct btrfs_root *extent_root = root->fs_info->extent_root;
1649 	int pending_ret;
1650 	int ret;
1651 
1652 	WARN_ON(num_bytes < root->sectorsize);
1653 	if (!root->ref_cows)
1654 		ref_generation = 0;
1655 
1656 	if (root == extent_root) {
1657 		pin_down_bytes(root, bytenr, num_bytes, 1);
1658 		return 0;
1659 	}
1660 	ret = __free_extent(trans, root, bytenr, num_bytes, root_objectid,
1661 			    ref_generation, owner_objectid, owner_offset,
1662 			    pin, pin == 0);
1663 	pending_ret = del_pending_extents(trans, root->fs_info->extent_root);
1664 	return ret ? ret : pending_ret;
1665 }
1666 
1667 int btrfs_free_extent(struct btrfs_trans_handle *trans,
1668 		      struct btrfs_root *root, u64 bytenr,
1669 		      u64 num_bytes, u64 root_objectid,
1670 		      u64 ref_generation, u64 owner_objectid,
1671 		      u64 owner_offset, int pin)
1672 {
1673 	int ret;
1674 
1675 	maybe_lock_mutex(root);
1676 	ret = __btrfs_free_extent(trans, root, bytenr, num_bytes,
1677 				  root_objectid, ref_generation,
1678 				  owner_objectid, owner_offset, pin);
1679 	maybe_unlock_mutex(root);
1680 	return ret;
1681 }
1682 
1683 static u64 stripe_align(struct btrfs_root *root, u64 val)
1684 {
1685 	u64 mask = ((u64)root->stripesize - 1);
1686 	u64 ret = (val + mask) & ~mask;
1687 	return ret;
1688 }
1689 
1690 /*
1691  * walks the btree of allocated extents and find a hole of a given size.
1692  * The key ins is changed to record the hole:
1693  * ins->objectid == block start
1694  * ins->flags = BTRFS_EXTENT_ITEM_KEY
1695  * ins->offset == number of blocks
1696  * Any available blocks before search_start are skipped.
1697  */
1698 static int noinline find_free_extent(struct btrfs_trans_handle *trans,
1699 				     struct btrfs_root *orig_root,
1700 				     u64 num_bytes, u64 empty_size,
1701 				     u64 search_start, u64 search_end,
1702 				     u64 hint_byte, struct btrfs_key *ins,
1703 				     u64 exclude_start, u64 exclude_nr,
1704 				     int data)
1705 {
1706 	int ret;
1707 	u64 orig_search_start;
1708 	struct btrfs_root * root = orig_root->fs_info->extent_root;
1709 	struct btrfs_fs_info *info = root->fs_info;
1710 	u64 total_needed = num_bytes;
1711 	u64 *last_ptr = NULL;
1712 	struct btrfs_block_group_cache *block_group;
1713 	int full_scan = 0;
1714 	int wrapped = 0;
1715 	int chunk_alloc_done = 0;
1716 	int empty_cluster = 2 * 1024 * 1024;
1717 	int allowed_chunk_alloc = 0;
1718 
1719 	WARN_ON(num_bytes < root->sectorsize);
1720 	btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
1721 
1722 	if (orig_root->ref_cows || empty_size)
1723 		allowed_chunk_alloc = 1;
1724 
1725 	if (data & BTRFS_BLOCK_GROUP_METADATA) {
1726 		last_ptr = &root->fs_info->last_alloc;
1727 		empty_cluster = 256 * 1024;
1728 	}
1729 
1730 	if ((data & BTRFS_BLOCK_GROUP_DATA) && btrfs_test_opt(root, SSD)) {
1731 		last_ptr = &root->fs_info->last_data_alloc;
1732 	}
1733 
1734 	if (last_ptr) {
1735 		if (*last_ptr)
1736 			hint_byte = *last_ptr;
1737 		else {
1738 			empty_size += empty_cluster;
1739 		}
1740 	}
1741 
1742 	search_start = max(search_start, first_logical_byte(root, 0));
1743 	orig_search_start = search_start;
1744 
1745 	if (search_end == (u64)-1)
1746 		search_end = btrfs_super_total_bytes(&info->super_copy);
1747 
1748 	if (hint_byte) {
1749 		block_group = btrfs_lookup_first_block_group(info, hint_byte);
1750 		if (!block_group)
1751 			hint_byte = search_start;
1752 		block_group = __btrfs_find_block_group(root, block_group,
1753 						     hint_byte, data, 1);
1754 		if (last_ptr && *last_ptr == 0 && block_group)
1755 			hint_byte = block_group->key.objectid;
1756 	} else {
1757 		block_group = __btrfs_find_block_group(root,
1758 						     trans->block_group,
1759 						     search_start, data, 1);
1760 	}
1761 	search_start = max(search_start, hint_byte);
1762 
1763 	total_needed += empty_size;
1764 
1765 check_failed:
1766 	if (!block_group) {
1767 		block_group = btrfs_lookup_first_block_group(info,
1768 							     search_start);
1769 		if (!block_group)
1770 			block_group = btrfs_lookup_first_block_group(info,
1771 						       orig_search_start);
1772 	}
1773 	if (full_scan && !chunk_alloc_done) {
1774 		if (allowed_chunk_alloc) {
1775 			do_chunk_alloc(trans, root,
1776 				     num_bytes + 2 * 1024 * 1024, data, 1);
1777 			allowed_chunk_alloc = 0;
1778 		} else if (block_group && block_group_bits(block_group, data)) {
1779 			block_group->space_info->force_alloc = 1;
1780 		}
1781 		chunk_alloc_done = 1;
1782 	}
1783 	ret = find_search_start(root, &block_group, &search_start,
1784 				total_needed, data);
1785 	if (ret == -ENOSPC && last_ptr && *last_ptr) {
1786 		*last_ptr = 0;
1787 		block_group = btrfs_lookup_first_block_group(info,
1788 							     orig_search_start);
1789 		search_start = orig_search_start;
1790 		ret = find_search_start(root, &block_group, &search_start,
1791 					total_needed, data);
1792 	}
1793 	if (ret == -ENOSPC)
1794 		goto enospc;
1795 	if (ret)
1796 		goto error;
1797 
1798 	if (last_ptr && *last_ptr && search_start != *last_ptr) {
1799 		*last_ptr = 0;
1800 		if (!empty_size) {
1801 			empty_size += empty_cluster;
1802 			total_needed += empty_size;
1803 		}
1804 		block_group = btrfs_lookup_first_block_group(info,
1805 						       orig_search_start);
1806 		search_start = orig_search_start;
1807 		ret = find_search_start(root, &block_group,
1808 					&search_start, total_needed, data);
1809 		if (ret == -ENOSPC)
1810 			goto enospc;
1811 		if (ret)
1812 			goto error;
1813 	}
1814 
1815 	search_start = stripe_align(root, search_start);
1816 	ins->objectid = search_start;
1817 	ins->offset = num_bytes;
1818 
1819 	if (ins->objectid + num_bytes >= search_end)
1820 		goto enospc;
1821 
1822 	if (ins->objectid + num_bytes >
1823 	    block_group->key.objectid + block_group->key.offset) {
1824 		search_start = block_group->key.objectid +
1825 			block_group->key.offset;
1826 		goto new_group;
1827 	}
1828 
1829 	if (test_range_bit(&info->extent_ins, ins->objectid,
1830 			   ins->objectid + num_bytes -1, EXTENT_LOCKED, 0)) {
1831 		search_start = ins->objectid + num_bytes;
1832 		goto new_group;
1833 	}
1834 
1835 	if (test_range_bit(&info->pinned_extents, ins->objectid,
1836 			   ins->objectid + num_bytes -1, EXTENT_DIRTY, 0)) {
1837 		search_start = ins->objectid + num_bytes;
1838 		goto new_group;
1839 	}
1840 
1841 	if (exclude_nr > 0 && (ins->objectid + num_bytes > exclude_start &&
1842 	    ins->objectid < exclude_start + exclude_nr)) {
1843 		search_start = exclude_start + exclude_nr;
1844 		goto new_group;
1845 	}
1846 
1847 	if (!(data & BTRFS_BLOCK_GROUP_DATA)) {
1848 		block_group = btrfs_lookup_block_group(info, ins->objectid);
1849 		if (block_group)
1850 			trans->block_group = block_group;
1851 	}
1852 	ins->offset = num_bytes;
1853 	if (last_ptr) {
1854 		*last_ptr = ins->objectid + ins->offset;
1855 		if (*last_ptr ==
1856 		    btrfs_super_total_bytes(&root->fs_info->super_copy)) {
1857 			*last_ptr = 0;
1858 		}
1859 	}
1860 	return 0;
1861 
1862 new_group:
1863 	if (search_start + num_bytes >= search_end) {
1864 enospc:
1865 		search_start = orig_search_start;
1866 		if (full_scan) {
1867 			ret = -ENOSPC;
1868 			goto error;
1869 		}
1870 		if (wrapped) {
1871 			if (!full_scan)
1872 				total_needed -= empty_size;
1873 			full_scan = 1;
1874 		} else
1875 			wrapped = 1;
1876 	}
1877 	block_group = btrfs_lookup_first_block_group(info, search_start);
1878 	cond_resched();
1879 	block_group = __btrfs_find_block_group(root, block_group,
1880 					     search_start, data, 0);
1881 	goto check_failed;
1882 
1883 error:
1884 	return ret;
1885 }
1886 
1887 /*
1888  * finds a free extent and does all the dirty work required for allocation
1889  * returns the key for the extent through ins, and a tree buffer for
1890  * the first block of the extent through buf.
1891  *
1892  * returns 0 if everything worked, non-zero otherwise.
1893  */
1894 int btrfs_alloc_extent(struct btrfs_trans_handle *trans,
1895 		       struct btrfs_root *root,
1896 		       u64 num_bytes, u64 min_alloc_size,
1897 		       u64 root_objectid, u64 ref_generation,
1898 		       u64 owner, u64 owner_offset,
1899 		       u64 empty_size, u64 hint_byte,
1900 		       u64 search_end, struct btrfs_key *ins, u64 data)
1901 {
1902 	int ret;
1903 	int pending_ret;
1904 	u64 super_used;
1905 	u64 root_used;
1906 	u64 search_start = 0;
1907 	u64 alloc_profile;
1908 	u32 sizes[2];
1909 	struct btrfs_fs_info *info = root->fs_info;
1910 	struct btrfs_root *extent_root = info->extent_root;
1911 	struct btrfs_extent_item *extent_item;
1912 	struct btrfs_extent_ref *ref;
1913 	struct btrfs_path *path;
1914 	struct btrfs_key keys[2];
1915 
1916 	maybe_lock_mutex(root);
1917 
1918 	if (data) {
1919 		alloc_profile = info->avail_data_alloc_bits &
1920 			        info->data_alloc_profile;
1921 		data = BTRFS_BLOCK_GROUP_DATA | alloc_profile;
1922 	} else if (root == root->fs_info->chunk_root) {
1923 		alloc_profile = info->avail_system_alloc_bits &
1924 			        info->system_alloc_profile;
1925 		data = BTRFS_BLOCK_GROUP_SYSTEM | alloc_profile;
1926 	} else {
1927 		alloc_profile = info->avail_metadata_alloc_bits &
1928 			        info->metadata_alloc_profile;
1929 		data = BTRFS_BLOCK_GROUP_METADATA | alloc_profile;
1930 	}
1931 again:
1932 	data = reduce_alloc_profile(root, data);
1933 	/*
1934 	 * the only place that sets empty_size is btrfs_realloc_node, which
1935 	 * is not called recursively on allocations
1936 	 */
1937 	if (empty_size || root->ref_cows) {
1938 		if (!(data & BTRFS_BLOCK_GROUP_METADATA)) {
1939 			ret = do_chunk_alloc(trans, root->fs_info->extent_root,
1940 				     2 * 1024 * 1024,
1941 				     BTRFS_BLOCK_GROUP_METADATA |
1942 				     (info->metadata_alloc_profile &
1943 				      info->avail_metadata_alloc_bits), 0);
1944 			BUG_ON(ret);
1945 		}
1946 		ret = do_chunk_alloc(trans, root->fs_info->extent_root,
1947 				     num_bytes + 2 * 1024 * 1024, data, 0);
1948 		BUG_ON(ret);
1949 	}
1950 
1951 	WARN_ON(num_bytes < root->sectorsize);
1952 	ret = find_free_extent(trans, root, num_bytes, empty_size,
1953 			       search_start, search_end, hint_byte, ins,
1954 			       trans->alloc_exclude_start,
1955 			       trans->alloc_exclude_nr, data);
1956 
1957 	if (ret == -ENOSPC && num_bytes > min_alloc_size) {
1958 		num_bytes = num_bytes >> 1;
1959 		num_bytes = max(num_bytes, min_alloc_size);
1960 		do_chunk_alloc(trans, root->fs_info->extent_root,
1961 			       num_bytes, data, 1);
1962 		goto again;
1963 	}
1964 	if (ret) {
1965 		printk("allocation failed flags %Lu\n", data);
1966 	}
1967 	if (ret) {
1968 		BUG();
1969 		goto out;
1970 	}
1971 
1972 	/* block accounting for super block */
1973 	spin_lock_irq(&info->delalloc_lock);
1974 	super_used = btrfs_super_bytes_used(&info->super_copy);
1975 	btrfs_set_super_bytes_used(&info->super_copy, super_used + num_bytes);
1976 	spin_unlock_irq(&info->delalloc_lock);
1977 
1978 	/* block accounting for root item */
1979 	root_used = btrfs_root_used(&root->root_item);
1980 	btrfs_set_root_used(&root->root_item, root_used + num_bytes);
1981 
1982 	clear_extent_dirty(&root->fs_info->free_space_cache,
1983 			   ins->objectid, ins->objectid + ins->offset - 1,
1984 			   GFP_NOFS);
1985 
1986 	if (root == extent_root) {
1987 		set_extent_bits(&root->fs_info->extent_ins, ins->objectid,
1988 				ins->objectid + ins->offset - 1,
1989 				EXTENT_LOCKED, GFP_NOFS);
1990 		goto update_block;
1991 	}
1992 
1993 	WARN_ON(trans->alloc_exclude_nr);
1994 	trans->alloc_exclude_start = ins->objectid;
1995 	trans->alloc_exclude_nr = ins->offset;
1996 
1997 	memcpy(&keys[0], ins, sizeof(*ins));
1998 	keys[1].offset = hash_extent_ref(root_objectid, ref_generation,
1999 					 owner, owner_offset);
2000 	keys[1].objectid = ins->objectid;
2001 	keys[1].type = BTRFS_EXTENT_REF_KEY;
2002 	sizes[0] = sizeof(*extent_item);
2003 	sizes[1] = sizeof(*ref);
2004 
2005 	path = btrfs_alloc_path();
2006 	BUG_ON(!path);
2007 
2008 	ret = btrfs_insert_empty_items(trans, extent_root, path, keys,
2009 				       sizes, 2);
2010 
2011 	BUG_ON(ret);
2012 	extent_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
2013 				     struct btrfs_extent_item);
2014 	btrfs_set_extent_refs(path->nodes[0], extent_item, 1);
2015 	ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
2016 			     struct btrfs_extent_ref);
2017 
2018 	btrfs_set_ref_root(path->nodes[0], ref, root_objectid);
2019 	btrfs_set_ref_generation(path->nodes[0], ref, ref_generation);
2020 	btrfs_set_ref_objectid(path->nodes[0], ref, owner);
2021 	btrfs_set_ref_offset(path->nodes[0], ref, owner_offset);
2022 
2023 	btrfs_mark_buffer_dirty(path->nodes[0]);
2024 
2025 	trans->alloc_exclude_start = 0;
2026 	trans->alloc_exclude_nr = 0;
2027 	btrfs_free_path(path);
2028 	finish_current_insert(trans, extent_root);
2029 	pending_ret = del_pending_extents(trans, extent_root);
2030 
2031 	if (ret)
2032 		goto out;
2033 	if (pending_ret) {
2034 		ret = pending_ret;
2035 		goto out;
2036 	}
2037 
2038 update_block:
2039 	ret = update_block_group(trans, root, ins->objectid, ins->offset, 1, 0);
2040 	if (ret) {
2041 		printk("update block group failed for %Lu %Lu\n",
2042 		       ins->objectid, ins->offset);
2043 		BUG();
2044 	}
2045 out:
2046 	maybe_unlock_mutex(root);
2047 	return ret;
2048 }
2049 /*
2050  * helper function to allocate a block for a given tree
2051  * returns the tree buffer or NULL.
2052  */
2053 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
2054 					     struct btrfs_root *root,
2055 					     u32 blocksize,
2056 					     u64 root_objectid,
2057 					     u64 ref_generation,
2058 					     u64 first_objectid,
2059 					     int level,
2060 					     u64 hint,
2061 					     u64 empty_size)
2062 {
2063 	struct btrfs_key ins;
2064 	int ret;
2065 	struct extent_buffer *buf;
2066 
2067 	ret = btrfs_alloc_extent(trans, root, blocksize, blocksize,
2068 				 root_objectid, ref_generation,
2069 				 level, first_objectid, empty_size, hint,
2070 				 (u64)-1, &ins, 0);
2071 	if (ret) {
2072 		BUG_ON(ret > 0);
2073 		return ERR_PTR(ret);
2074 	}
2075 	buf = btrfs_find_create_tree_block(root, ins.objectid, blocksize);
2076 	if (!buf) {
2077 		btrfs_free_extent(trans, root, ins.objectid, blocksize,
2078 				  root->root_key.objectid, ref_generation,
2079 				  0, 0, 0);
2080 		return ERR_PTR(-ENOMEM);
2081 	}
2082 	btrfs_set_header_generation(buf, trans->transid);
2083 	btrfs_tree_lock(buf);
2084 	clean_tree_block(trans, root, buf);
2085 	btrfs_set_buffer_uptodate(buf);
2086 
2087 	if (PageDirty(buf->first_page)) {
2088 		printk("page %lu dirty\n", buf->first_page->index);
2089 		WARN_ON(1);
2090 	}
2091 
2092 	set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
2093 			 buf->start + buf->len - 1, GFP_NOFS);
2094 	if (!btrfs_test_opt(root, SSD))
2095 		btrfs_set_buffer_defrag(buf);
2096 	trans->blocks_used++;
2097 	return buf;
2098 }
2099 
2100 static int noinline drop_leaf_ref(struct btrfs_trans_handle *trans,
2101 				  struct btrfs_root *root,
2102 				  struct extent_buffer *leaf)
2103 {
2104 	u64 leaf_owner;
2105 	u64 leaf_generation;
2106 	struct btrfs_key key;
2107 	struct btrfs_file_extent_item *fi;
2108 	int i;
2109 	int nritems;
2110 	int ret;
2111 
2112 	BUG_ON(!btrfs_is_leaf(leaf));
2113 	nritems = btrfs_header_nritems(leaf);
2114 	leaf_owner = btrfs_header_owner(leaf);
2115 	leaf_generation = btrfs_header_generation(leaf);
2116 
2117 	for (i = 0; i < nritems; i++) {
2118 		u64 disk_bytenr;
2119 
2120 		btrfs_item_key_to_cpu(leaf, &key, i);
2121 		if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2122 			continue;
2123 		fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
2124 		if (btrfs_file_extent_type(leaf, fi) ==
2125 		    BTRFS_FILE_EXTENT_INLINE)
2126 			continue;
2127 		/*
2128 		 * FIXME make sure to insert a trans record that
2129 		 * repeats the snapshot del on crash
2130 		 */
2131 		disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
2132 		if (disk_bytenr == 0)
2133 			continue;
2134 		ret = __btrfs_free_extent(trans, root, disk_bytenr,
2135 				btrfs_file_extent_disk_num_bytes(leaf, fi),
2136 				leaf_owner, leaf_generation,
2137 				key.objectid, key.offset, 0);
2138 		BUG_ON(ret);
2139 	}
2140 	return 0;
2141 }
2142 
2143 static void noinline reada_walk_down(struct btrfs_root *root,
2144 				     struct extent_buffer *node,
2145 				     int slot)
2146 {
2147 	u64 bytenr;
2148 	u64 last = 0;
2149 	u32 nritems;
2150 	u32 refs;
2151 	u32 blocksize;
2152 	int ret;
2153 	int i;
2154 	int level;
2155 	int skipped = 0;
2156 
2157 	nritems = btrfs_header_nritems(node);
2158 	level = btrfs_header_level(node);
2159 	if (level)
2160 		return;
2161 
2162 	for (i = slot; i < nritems && skipped < 32; i++) {
2163 		bytenr = btrfs_node_blockptr(node, i);
2164 		if (last && ((bytenr > last && bytenr - last > 32 * 1024) ||
2165 			     (last > bytenr && last - bytenr > 32 * 1024))) {
2166 			skipped++;
2167 			continue;
2168 		}
2169 		blocksize = btrfs_level_size(root, level - 1);
2170 		if (i != slot) {
2171 			ret = lookup_extent_ref(NULL, root, bytenr,
2172 						blocksize, &refs);
2173 			BUG_ON(ret);
2174 			if (refs != 1) {
2175 				skipped++;
2176 				continue;
2177 			}
2178 		}
2179 		mutex_unlock(&root->fs_info->alloc_mutex);
2180 		ret = readahead_tree_block(root, bytenr, blocksize,
2181 					   btrfs_node_ptr_generation(node, i));
2182 		last = bytenr + blocksize;
2183 		cond_resched();
2184 		mutex_lock(&root->fs_info->alloc_mutex);
2185 		if (ret)
2186 			break;
2187 	}
2188 }
2189 
2190 /*
2191  * helper function for drop_snapshot, this walks down the tree dropping ref
2192  * counts as it goes.
2193  */
2194 static int noinline walk_down_tree(struct btrfs_trans_handle *trans,
2195 				   struct btrfs_root *root,
2196 				   struct btrfs_path *path, int *level)
2197 {
2198 	u64 root_owner;
2199 	u64 root_gen;
2200 	u64 bytenr;
2201 	u64 ptr_gen;
2202 	struct extent_buffer *next;
2203 	struct extent_buffer *cur;
2204 	struct extent_buffer *parent;
2205 	u32 blocksize;
2206 	int ret;
2207 	u32 refs;
2208 
2209 	mutex_lock(&root->fs_info->alloc_mutex);
2210 
2211 	WARN_ON(*level < 0);
2212 	WARN_ON(*level >= BTRFS_MAX_LEVEL);
2213 	ret = lookup_extent_ref(trans, root,
2214 				path->nodes[*level]->start,
2215 				path->nodes[*level]->len, &refs);
2216 	BUG_ON(ret);
2217 	if (refs > 1)
2218 		goto out;
2219 
2220 	/*
2221 	 * walk down to the last node level and free all the leaves
2222 	 */
2223 	while(*level >= 0) {
2224 		WARN_ON(*level < 0);
2225 		WARN_ON(*level >= BTRFS_MAX_LEVEL);
2226 		cur = path->nodes[*level];
2227 
2228 		if (btrfs_header_level(cur) != *level)
2229 			WARN_ON(1);
2230 
2231 		if (path->slots[*level] >=
2232 		    btrfs_header_nritems(cur))
2233 			break;
2234 		if (*level == 0) {
2235 			ret = drop_leaf_ref(trans, root, cur);
2236 			BUG_ON(ret);
2237 			break;
2238 		}
2239 		bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
2240 		ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
2241 		blocksize = btrfs_level_size(root, *level - 1);
2242 
2243 		ret = lookup_extent_ref(trans, root, bytenr, blocksize, &refs);
2244 		BUG_ON(ret);
2245 		if (refs != 1) {
2246 			parent = path->nodes[*level];
2247 			root_owner = btrfs_header_owner(parent);
2248 			root_gen = btrfs_header_generation(parent);
2249 			path->slots[*level]++;
2250 			ret = __btrfs_free_extent(trans, root, bytenr,
2251 						blocksize, root_owner,
2252 						root_gen, 0, 0, 1);
2253 			BUG_ON(ret);
2254 			continue;
2255 		}
2256 		next = btrfs_find_tree_block(root, bytenr, blocksize);
2257 		if (!next || !btrfs_buffer_uptodate(next, ptr_gen)) {
2258 			free_extent_buffer(next);
2259 			reada_walk_down(root, cur, path->slots[*level]);
2260 
2261 			mutex_unlock(&root->fs_info->alloc_mutex);
2262 			next = read_tree_block(root, bytenr, blocksize,
2263 					       ptr_gen);
2264 			mutex_lock(&root->fs_info->alloc_mutex);
2265 
2266 			/* we've dropped the lock, double check */
2267 			ret = lookup_extent_ref(trans, root, bytenr,
2268 						blocksize, &refs);
2269 			BUG_ON(ret);
2270 			if (refs != 1) {
2271 				parent = path->nodes[*level];
2272 				root_owner = btrfs_header_owner(parent);
2273 				root_gen = btrfs_header_generation(parent);
2274 
2275 				path->slots[*level]++;
2276 				free_extent_buffer(next);
2277 				ret = __btrfs_free_extent(trans, root, bytenr,
2278 							blocksize,
2279 							root_owner,
2280 							root_gen, 0, 0, 1);
2281 				BUG_ON(ret);
2282 				continue;
2283 			}
2284 		}
2285 		WARN_ON(*level <= 0);
2286 		if (path->nodes[*level-1])
2287 			free_extent_buffer(path->nodes[*level-1]);
2288 		path->nodes[*level-1] = next;
2289 		*level = btrfs_header_level(next);
2290 		path->slots[*level] = 0;
2291 	}
2292 out:
2293 	WARN_ON(*level < 0);
2294 	WARN_ON(*level >= BTRFS_MAX_LEVEL);
2295 
2296 	if (path->nodes[*level] == root->node) {
2297 		root_owner = root->root_key.objectid;
2298 		parent = path->nodes[*level];
2299 	} else {
2300 		parent = path->nodes[*level + 1];
2301 		root_owner = btrfs_header_owner(parent);
2302 	}
2303 
2304 	root_gen = btrfs_header_generation(parent);
2305 	ret = __btrfs_free_extent(trans, root, path->nodes[*level]->start,
2306 				path->nodes[*level]->len,
2307 				root_owner, root_gen, 0, 0, 1);
2308 	free_extent_buffer(path->nodes[*level]);
2309 	path->nodes[*level] = NULL;
2310 	*level += 1;
2311 	BUG_ON(ret);
2312 	mutex_unlock(&root->fs_info->alloc_mutex);
2313 	return 0;
2314 }
2315 
2316 /*
2317  * helper for dropping snapshots.  This walks back up the tree in the path
2318  * to find the first node higher up where we haven't yet gone through
2319  * all the slots
2320  */
2321 static int noinline walk_up_tree(struct btrfs_trans_handle *trans,
2322 				 struct btrfs_root *root,
2323 				 struct btrfs_path *path, int *level)
2324 {
2325 	u64 root_owner;
2326 	u64 root_gen;
2327 	struct btrfs_root_item *root_item = &root->root_item;
2328 	int i;
2329 	int slot;
2330 	int ret;
2331 
2332 	for(i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) {
2333 		slot = path->slots[i];
2334 		if (slot < btrfs_header_nritems(path->nodes[i]) - 1) {
2335 			struct extent_buffer *node;
2336 			struct btrfs_disk_key disk_key;
2337 			node = path->nodes[i];
2338 			path->slots[i]++;
2339 			*level = i;
2340 			WARN_ON(*level == 0);
2341 			btrfs_node_key(node, &disk_key, path->slots[i]);
2342 			memcpy(&root_item->drop_progress,
2343 			       &disk_key, sizeof(disk_key));
2344 			root_item->drop_level = i;
2345 			return 0;
2346 		} else {
2347 			if (path->nodes[*level] == root->node) {
2348 				root_owner = root->root_key.objectid;
2349 				root_gen =
2350 				   btrfs_header_generation(path->nodes[*level]);
2351 			} else {
2352 				struct extent_buffer *node;
2353 				node = path->nodes[*level + 1];
2354 				root_owner = btrfs_header_owner(node);
2355 				root_gen = btrfs_header_generation(node);
2356 			}
2357 			ret = btrfs_free_extent(trans, root,
2358 						path->nodes[*level]->start,
2359 						path->nodes[*level]->len,
2360 						root_owner, root_gen, 0, 0, 1);
2361 			BUG_ON(ret);
2362 			free_extent_buffer(path->nodes[*level]);
2363 			path->nodes[*level] = NULL;
2364 			*level = i + 1;
2365 		}
2366 	}
2367 	return 1;
2368 }
2369 
2370 /*
2371  * drop the reference count on the tree rooted at 'snap'.  This traverses
2372  * the tree freeing any blocks that have a ref count of zero after being
2373  * decremented.
2374  */
2375 int btrfs_drop_snapshot(struct btrfs_trans_handle *trans, struct btrfs_root
2376 			*root)
2377 {
2378 	int ret = 0;
2379 	int wret;
2380 	int level;
2381 	struct btrfs_path *path;
2382 	int i;
2383 	int orig_level;
2384 	struct btrfs_root_item *root_item = &root->root_item;
2385 
2386 	WARN_ON(!mutex_is_locked(&root->fs_info->drop_mutex));
2387 	path = btrfs_alloc_path();
2388 	BUG_ON(!path);
2389 
2390 	level = btrfs_header_level(root->node);
2391 	orig_level = level;
2392 	if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
2393 		path->nodes[level] = root->node;
2394 		extent_buffer_get(root->node);
2395 		path->slots[level] = 0;
2396 	} else {
2397 		struct btrfs_key key;
2398 		struct btrfs_disk_key found_key;
2399 		struct extent_buffer *node;
2400 
2401 		btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
2402 		level = root_item->drop_level;
2403 		path->lowest_level = level;
2404 		wret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2405 		if (wret < 0) {
2406 			ret = wret;
2407 			goto out;
2408 		}
2409 		node = path->nodes[level];
2410 		btrfs_node_key(node, &found_key, path->slots[level]);
2411 		WARN_ON(memcmp(&found_key, &root_item->drop_progress,
2412 			       sizeof(found_key)));
2413 		for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
2414 			if (path->nodes[i] && path->locks[i]) {
2415 				path->locks[i] = 0;
2416 				btrfs_tree_unlock(path->nodes[i]);
2417 			}
2418 		}
2419 	}
2420 	while(1) {
2421 		wret = walk_down_tree(trans, root, path, &level);
2422 		if (wret > 0)
2423 			break;
2424 		if (wret < 0)
2425 			ret = wret;
2426 
2427 		wret = walk_up_tree(trans, root, path, &level);
2428 		if (wret > 0)
2429 			break;
2430 		if (wret < 0)
2431 			ret = wret;
2432 		ret = -EAGAIN;
2433 		break;
2434 	}
2435 	for (i = 0; i <= orig_level; i++) {
2436 		if (path->nodes[i]) {
2437 			free_extent_buffer(path->nodes[i]);
2438 			path->nodes[i] = NULL;
2439 		}
2440 	}
2441 out:
2442 	btrfs_free_path(path);
2443 	return ret;
2444 }
2445 
2446 int btrfs_free_block_groups(struct btrfs_fs_info *info)
2447 {
2448 	u64 start;
2449 	u64 end;
2450 	u64 ptr;
2451 	int ret;
2452 
2453 	mutex_lock(&info->alloc_mutex);
2454 	while(1) {
2455 		ret = find_first_extent_bit(&info->block_group_cache, 0,
2456 					    &start, &end, (unsigned int)-1);
2457 		if (ret)
2458 			break;
2459 		ret = get_state_private(&info->block_group_cache, start, &ptr);
2460 		if (!ret)
2461 			kfree((void *)(unsigned long)ptr);
2462 		clear_extent_bits(&info->block_group_cache, start,
2463 				  end, (unsigned int)-1, GFP_NOFS);
2464 	}
2465 	while(1) {
2466 		ret = find_first_extent_bit(&info->free_space_cache, 0,
2467 					    &start, &end, EXTENT_DIRTY);
2468 		if (ret)
2469 			break;
2470 		clear_extent_dirty(&info->free_space_cache, start,
2471 				   end, GFP_NOFS);
2472 	}
2473 	mutex_unlock(&info->alloc_mutex);
2474 	return 0;
2475 }
2476 
2477 static unsigned long calc_ra(unsigned long start, unsigned long last,
2478 			     unsigned long nr)
2479 {
2480 	return min(last, start + nr - 1);
2481 }
2482 
2483 static int noinline relocate_inode_pages(struct inode *inode, u64 start,
2484 					 u64 len)
2485 {
2486 	u64 page_start;
2487 	u64 page_end;
2488 	unsigned long last_index;
2489 	unsigned long i;
2490 	struct page *page;
2491 	struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2492 	struct file_ra_state *ra;
2493 	unsigned long total_read = 0;
2494 	unsigned long ra_pages;
2495 	struct btrfs_trans_handle *trans;
2496 
2497 	ra = kzalloc(sizeof(*ra), GFP_NOFS);
2498 
2499 	mutex_lock(&inode->i_mutex);
2500 	i = start >> PAGE_CACHE_SHIFT;
2501 	last_index = (start + len - 1) >> PAGE_CACHE_SHIFT;
2502 
2503 	ra_pages = BTRFS_I(inode)->root->fs_info->bdi.ra_pages;
2504 
2505 	file_ra_state_init(ra, inode->i_mapping);
2506 
2507 	for (; i <= last_index; i++) {
2508 		if (total_read % ra_pages == 0) {
2509 			btrfs_force_ra(inode->i_mapping, ra, NULL, i,
2510 				       calc_ra(i, last_index, ra_pages));
2511 		}
2512 		total_read++;
2513 		if (((u64)i << PAGE_CACHE_SHIFT) > inode->i_size)
2514 			goto truncate_racing;
2515 
2516 		page = grab_cache_page(inode->i_mapping, i);
2517 		if (!page) {
2518 			goto out_unlock;
2519 		}
2520 		if (!PageUptodate(page)) {
2521 			btrfs_readpage(NULL, page);
2522 			lock_page(page);
2523 			if (!PageUptodate(page)) {
2524 				unlock_page(page);
2525 				page_cache_release(page);
2526 				goto out_unlock;
2527 			}
2528 		}
2529 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
2530 		ClearPageDirty(page);
2531 #else
2532 		cancel_dirty_page(page, PAGE_CACHE_SIZE);
2533 #endif
2534 		wait_on_page_writeback(page);
2535 		set_page_extent_mapped(page);
2536 		page_start = (u64)page->index << PAGE_CACHE_SHIFT;
2537 		page_end = page_start + PAGE_CACHE_SIZE - 1;
2538 
2539 		lock_extent(io_tree, page_start, page_end, GFP_NOFS);
2540 
2541 		set_extent_delalloc(io_tree, page_start,
2542 				    page_end, GFP_NOFS);
2543 		set_page_dirty(page);
2544 
2545 		unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
2546 		unlock_page(page);
2547 		page_cache_release(page);
2548 	}
2549 	balance_dirty_pages_ratelimited_nr(inode->i_mapping,
2550 					   total_read);
2551 
2552 out_unlock:
2553 	kfree(ra);
2554 	trans = btrfs_start_transaction(BTRFS_I(inode)->root, 1);
2555 	if (trans) {
2556 		btrfs_add_ordered_inode(inode);
2557 		btrfs_end_transaction(trans, BTRFS_I(inode)->root);
2558 		mark_inode_dirty(inode);
2559 	}
2560 	mutex_unlock(&inode->i_mutex);
2561 	return 0;
2562 
2563 truncate_racing:
2564 	vmtruncate(inode, inode->i_size);
2565 	balance_dirty_pages_ratelimited_nr(inode->i_mapping,
2566 					   total_read);
2567 	goto out_unlock;
2568 }
2569 
2570 /*
2571  * The back references tell us which tree holds a ref on a block,
2572  * but it is possible for the tree root field in the reference to
2573  * reflect the original root before a snapshot was made.  In this
2574  * case we should search through all the children of a given root
2575  * to find potential holders of references on a block.
2576  *
2577  * Instead, we do something a little less fancy and just search
2578  * all the roots for a given key/block combination.
2579  */
2580 static int find_root_for_ref(struct btrfs_root *root,
2581 			     struct btrfs_path *path,
2582 			     struct btrfs_key *key0,
2583 			     int level,
2584 			     int file_key,
2585 			     struct btrfs_root **found_root,
2586 			     u64 bytenr)
2587 {
2588 	struct btrfs_key root_location;
2589 	struct btrfs_root *cur_root = *found_root;
2590 	struct btrfs_file_extent_item *file_extent;
2591 	u64 root_search_start = BTRFS_FS_TREE_OBJECTID;
2592 	u64 found_bytenr;
2593 	int ret;
2594 	int i;
2595 
2596 	root_location.offset = (u64)-1;
2597 	root_location.type = BTRFS_ROOT_ITEM_KEY;
2598 	path->lowest_level = level;
2599 	path->reada = 0;
2600 	while(1) {
2601 		ret = btrfs_search_slot(NULL, cur_root, key0, path, 0, 0);
2602 		found_bytenr = 0;
2603 		if (ret == 0 && file_key) {
2604 			struct extent_buffer *leaf = path->nodes[0];
2605 			file_extent = btrfs_item_ptr(leaf, path->slots[0],
2606 					     struct btrfs_file_extent_item);
2607 			if (btrfs_file_extent_type(leaf, file_extent) ==
2608 			    BTRFS_FILE_EXTENT_REG) {
2609 				found_bytenr =
2610 					btrfs_file_extent_disk_bytenr(leaf,
2611 							       file_extent);
2612 		       }
2613 		} else if (!file_key) {
2614 			if (path->nodes[level])
2615 				found_bytenr = path->nodes[level]->start;
2616 		}
2617 
2618 		for (i = level; i < BTRFS_MAX_LEVEL; i++) {
2619 			if (!path->nodes[i])
2620 				break;
2621 			free_extent_buffer(path->nodes[i]);
2622 			path->nodes[i] = NULL;
2623 		}
2624 		btrfs_release_path(cur_root, path);
2625 
2626 		if (found_bytenr == bytenr) {
2627 			*found_root = cur_root;
2628 			ret = 0;
2629 			goto out;
2630 		}
2631 		ret = btrfs_search_root(root->fs_info->tree_root,
2632 					root_search_start, &root_search_start);
2633 		if (ret)
2634 			break;
2635 
2636 		root_location.objectid = root_search_start;
2637 		cur_root = btrfs_read_fs_root_no_name(root->fs_info,
2638 						      &root_location);
2639 		if (!cur_root) {
2640 			ret = 1;
2641 			break;
2642 		}
2643 	}
2644 out:
2645 	path->lowest_level = 0;
2646 	return ret;
2647 }
2648 
2649 /*
2650  * note, this releases the path
2651  */
2652 static int noinline relocate_one_reference(struct btrfs_root *extent_root,
2653 				  struct btrfs_path *path,
2654 				  struct btrfs_key *extent_key,
2655 				  u64 *last_file_objectid,
2656 				  u64 *last_file_offset,
2657 				  u64 *last_file_root,
2658 				  u64 last_extent)
2659 {
2660 	struct inode *inode;
2661 	struct btrfs_root *found_root;
2662 	struct btrfs_key root_location;
2663 	struct btrfs_key found_key;
2664 	struct btrfs_extent_ref *ref;
2665 	u64 ref_root;
2666 	u64 ref_gen;
2667 	u64 ref_objectid;
2668 	u64 ref_offset;
2669 	int ret;
2670 	int level;
2671 
2672 	ref = btrfs_item_ptr(path->nodes[0], path->slots[0],
2673 			     struct btrfs_extent_ref);
2674 	ref_root = btrfs_ref_root(path->nodes[0], ref);
2675 	ref_gen = btrfs_ref_generation(path->nodes[0], ref);
2676 	ref_objectid = btrfs_ref_objectid(path->nodes[0], ref);
2677 	ref_offset = btrfs_ref_offset(path->nodes[0], ref);
2678 	btrfs_release_path(extent_root, path);
2679 
2680 	root_location.objectid = ref_root;
2681 	if (ref_gen == 0)
2682 		root_location.offset = 0;
2683 	else
2684 		root_location.offset = (u64)-1;
2685 	root_location.type = BTRFS_ROOT_ITEM_KEY;
2686 
2687 	found_root = btrfs_read_fs_root_no_name(extent_root->fs_info,
2688 						&root_location);
2689 	BUG_ON(!found_root);
2690 
2691 	if (ref_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
2692 		found_key.objectid = ref_objectid;
2693 		found_key.type = BTRFS_EXTENT_DATA_KEY;
2694 		found_key.offset = ref_offset;
2695 		level = 0;
2696 
2697 		if (last_extent == extent_key->objectid &&
2698 		    *last_file_objectid == ref_objectid &&
2699 		    *last_file_offset == ref_offset &&
2700 		    *last_file_root == ref_root)
2701 			goto out;
2702 
2703 		ret = find_root_for_ref(extent_root, path, &found_key,
2704 					level, 1, &found_root,
2705 					extent_key->objectid);
2706 
2707 		if (ret)
2708 			goto out;
2709 
2710 		if (last_extent == extent_key->objectid &&
2711 		    *last_file_objectid == ref_objectid &&
2712 		    *last_file_offset == ref_offset &&
2713 		    *last_file_root == ref_root)
2714 			goto out;
2715 
2716 		inode = btrfs_iget_locked(extent_root->fs_info->sb,
2717 					  ref_objectid, found_root);
2718 		if (inode->i_state & I_NEW) {
2719 			/* the inode and parent dir are two different roots */
2720 			BTRFS_I(inode)->root = found_root;
2721 			BTRFS_I(inode)->location.objectid = ref_objectid;
2722 			BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
2723 			BTRFS_I(inode)->location.offset = 0;
2724 			btrfs_read_locked_inode(inode);
2725 			unlock_new_inode(inode);
2726 
2727 		}
2728 		/* this can happen if the reference is not against
2729 		 * the latest version of the tree root
2730 		 */
2731 		if (is_bad_inode(inode)) {
2732 			goto out;
2733 		}
2734 		*last_file_objectid = inode->i_ino;
2735 		*last_file_root = found_root->root_key.objectid;
2736 		*last_file_offset = ref_offset;
2737 
2738 		relocate_inode_pages(inode, ref_offset, extent_key->offset);
2739 		iput(inode);
2740 	} else {
2741 		struct btrfs_trans_handle *trans;
2742 		struct extent_buffer *eb;
2743 		int i;
2744 
2745 		eb = read_tree_block(found_root, extent_key->objectid,
2746 				     extent_key->offset, 0);
2747 		btrfs_tree_lock(eb);
2748 		level = btrfs_header_level(eb);
2749 
2750 		if (level == 0)
2751 			btrfs_item_key_to_cpu(eb, &found_key, 0);
2752 		else
2753 			btrfs_node_key_to_cpu(eb, &found_key, 0);
2754 
2755 		btrfs_tree_unlock(eb);
2756 		free_extent_buffer(eb);
2757 
2758 		ret = find_root_for_ref(extent_root, path, &found_key,
2759 					level, 0, &found_root,
2760 					extent_key->objectid);
2761 
2762 		if (ret)
2763 			goto out;
2764 
2765 		trans = btrfs_start_transaction(found_root, 1);
2766 
2767 		path->lowest_level = level;
2768 		path->reada = 2;
2769 		ret = btrfs_search_slot(trans, found_root, &found_key, path,
2770 					0, 1);
2771 		path->lowest_level = 0;
2772 		for (i = level; i < BTRFS_MAX_LEVEL; i++) {
2773 			if (!path->nodes[i])
2774 				break;
2775 			free_extent_buffer(path->nodes[i]);
2776 			path->nodes[i] = NULL;
2777 		}
2778 		btrfs_release_path(found_root, path);
2779 		if (found_root == found_root->fs_info->extent_root)
2780 			btrfs_extent_post_op(trans, found_root);
2781 		btrfs_end_transaction(trans, found_root);
2782 	}
2783 
2784 out:
2785 	return 0;
2786 }
2787 
2788 static int noinline del_extent_zero(struct btrfs_root *extent_root,
2789 				    struct btrfs_path *path,
2790 				    struct btrfs_key *extent_key)
2791 {
2792 	int ret;
2793 	struct btrfs_trans_handle *trans;
2794 
2795 	trans = btrfs_start_transaction(extent_root, 1);
2796 	ret = btrfs_search_slot(trans, extent_root, extent_key, path, -1, 1);
2797 	if (ret > 0) {
2798 		ret = -EIO;
2799 		goto out;
2800 	}
2801 	if (ret < 0)
2802 		goto out;
2803 	ret = btrfs_del_item(trans, extent_root, path);
2804 out:
2805 	btrfs_end_transaction(trans, extent_root);
2806 	return ret;
2807 }
2808 
2809 static int noinline relocate_one_extent(struct btrfs_root *extent_root,
2810 					struct btrfs_path *path,
2811 					struct btrfs_key *extent_key)
2812 {
2813 	struct btrfs_key key;
2814 	struct btrfs_key found_key;
2815 	struct extent_buffer *leaf;
2816 	u64 last_file_objectid = 0;
2817 	u64 last_file_root = 0;
2818 	u64 last_file_offset = (u64)-1;
2819 	u64 last_extent = 0;
2820 	u32 nritems;
2821 	u32 item_size;
2822 	int ret = 0;
2823 
2824 	if (extent_key->objectid == 0) {
2825 		ret = del_extent_zero(extent_root, path, extent_key);
2826 		goto out;
2827 	}
2828 	key.objectid = extent_key->objectid;
2829 	key.type = BTRFS_EXTENT_REF_KEY;
2830 	key.offset = 0;
2831 
2832 	while(1) {
2833 		ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2834 
2835 		if (ret < 0)
2836 			goto out;
2837 
2838 		ret = 0;
2839 		leaf = path->nodes[0];
2840 		nritems = btrfs_header_nritems(leaf);
2841 		if (path->slots[0] == nritems) {
2842 			ret = btrfs_next_leaf(extent_root, path);
2843 			if (ret > 0) {
2844 				ret = 0;
2845 				goto out;
2846 			}
2847 			if (ret < 0)
2848 				goto out;
2849 			leaf = path->nodes[0];
2850 		}
2851 
2852 		btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2853 		if (found_key.objectid != extent_key->objectid) {
2854 			break;
2855 		}
2856 
2857 		if (found_key.type != BTRFS_EXTENT_REF_KEY) {
2858 			break;
2859 		}
2860 
2861 		key.offset = found_key.offset + 1;
2862 		item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2863 
2864 		ret = relocate_one_reference(extent_root, path, extent_key,
2865 					     &last_file_objectid,
2866 					     &last_file_offset,
2867 					     &last_file_root, last_extent);
2868 		if (ret)
2869 			goto out;
2870 		last_extent = extent_key->objectid;
2871 	}
2872 	ret = 0;
2873 out:
2874 	btrfs_release_path(extent_root, path);
2875 	return ret;
2876 }
2877 
2878 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
2879 {
2880 	u64 num_devices;
2881 	u64 stripped = BTRFS_BLOCK_GROUP_RAID0 |
2882 		BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
2883 
2884 	num_devices = root->fs_info->fs_devices->num_devices;
2885 	if (num_devices == 1) {
2886 		stripped |= BTRFS_BLOCK_GROUP_DUP;
2887 		stripped = flags & ~stripped;
2888 
2889 		/* turn raid0 into single device chunks */
2890 		if (flags & BTRFS_BLOCK_GROUP_RAID0)
2891 			return stripped;
2892 
2893 		/* turn mirroring into duplication */
2894 		if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
2895 			     BTRFS_BLOCK_GROUP_RAID10))
2896 			return stripped | BTRFS_BLOCK_GROUP_DUP;
2897 		return flags;
2898 	} else {
2899 		/* they already had raid on here, just return */
2900 		if (flags & stripped)
2901 			return flags;
2902 
2903 		stripped |= BTRFS_BLOCK_GROUP_DUP;
2904 		stripped = flags & ~stripped;
2905 
2906 		/* switch duplicated blocks with raid1 */
2907 		if (flags & BTRFS_BLOCK_GROUP_DUP)
2908 			return stripped | BTRFS_BLOCK_GROUP_RAID1;
2909 
2910 		/* turn single device chunks into raid0 */
2911 		return stripped | BTRFS_BLOCK_GROUP_RAID0;
2912 	}
2913 	return flags;
2914 }
2915 
2916 int __alloc_chunk_for_shrink(struct btrfs_root *root,
2917 		     struct btrfs_block_group_cache *shrink_block_group,
2918 		     int force)
2919 {
2920 	struct btrfs_trans_handle *trans;
2921 	u64 new_alloc_flags;
2922 	u64 calc;
2923 
2924 	if (btrfs_block_group_used(&shrink_block_group->item) > 0) {
2925 
2926 		trans = btrfs_start_transaction(root, 1);
2927 		new_alloc_flags = update_block_group_flags(root,
2928 						   shrink_block_group->flags);
2929 		if (new_alloc_flags != shrink_block_group->flags) {
2930 			calc =
2931 			     btrfs_block_group_used(&shrink_block_group->item);
2932 		} else {
2933 			calc = shrink_block_group->key.offset;
2934 		}
2935 		do_chunk_alloc(trans, root->fs_info->extent_root,
2936 			       calc + 2 * 1024 * 1024, new_alloc_flags, force);
2937 		btrfs_end_transaction(trans, root);
2938 	}
2939 	return 0;
2940 }
2941 
2942 int btrfs_shrink_extent_tree(struct btrfs_root *root, u64 shrink_start)
2943 {
2944 	struct btrfs_trans_handle *trans;
2945 	struct btrfs_root *tree_root = root->fs_info->tree_root;
2946 	struct btrfs_path *path;
2947 	u64 cur_byte;
2948 	u64 total_found;
2949 	u64 shrink_last_byte;
2950 	struct btrfs_block_group_cache *shrink_block_group;
2951 	struct btrfs_fs_info *info = root->fs_info;
2952 	struct btrfs_key key;
2953 	struct btrfs_key found_key;
2954 	struct extent_buffer *leaf;
2955 	u32 nritems;
2956 	int ret;
2957 	int progress;
2958 
2959 	mutex_lock(&root->fs_info->alloc_mutex);
2960 	shrink_block_group = btrfs_lookup_block_group(root->fs_info,
2961 						      shrink_start);
2962 	BUG_ON(!shrink_block_group);
2963 
2964 	shrink_last_byte = shrink_block_group->key.objectid +
2965 		shrink_block_group->key.offset;
2966 
2967 	shrink_block_group->space_info->total_bytes -=
2968 		shrink_block_group->key.offset;
2969 	path = btrfs_alloc_path();
2970 	root = root->fs_info->extent_root;
2971 	path->reada = 2;
2972 
2973 	printk("btrfs relocating block group %llu flags %llu\n",
2974 	       (unsigned long long)shrink_start,
2975 	       (unsigned long long)shrink_block_group->flags);
2976 
2977 	__alloc_chunk_for_shrink(root, shrink_block_group, 1);
2978 
2979 again:
2980 
2981 	shrink_block_group->ro = 1;
2982 
2983 	total_found = 0;
2984 	progress = 0;
2985 	key.objectid = shrink_start;
2986 	key.offset = 0;
2987 	key.type = 0;
2988 	cur_byte = key.objectid;
2989 
2990 	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2991 	if (ret < 0)
2992 		goto out;
2993 
2994 	ret = btrfs_previous_item(root, path, 0, BTRFS_EXTENT_ITEM_KEY);
2995 	if (ret < 0)
2996 		goto out;
2997 
2998 	if (ret == 0) {
2999 		leaf = path->nodes[0];
3000 		btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3001 		if (found_key.objectid + found_key.offset > shrink_start &&
3002 		    found_key.objectid < shrink_last_byte) {
3003 			cur_byte = found_key.objectid;
3004 			key.objectid = cur_byte;
3005 		}
3006 	}
3007 	btrfs_release_path(root, path);
3008 
3009 	while(1) {
3010 		ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3011 		if (ret < 0)
3012 			goto out;
3013 
3014 		leaf = path->nodes[0];
3015 		nritems = btrfs_header_nritems(leaf);
3016 next:
3017 		if (path->slots[0] >= nritems) {
3018 			ret = btrfs_next_leaf(root, path);
3019 			if (ret < 0)
3020 				goto out;
3021 			if (ret == 1) {
3022 				ret = 0;
3023 				break;
3024 			}
3025 			leaf = path->nodes[0];
3026 			nritems = btrfs_header_nritems(leaf);
3027 		}
3028 
3029 		btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3030 
3031 		if (found_key.objectid >= shrink_last_byte)
3032 			break;
3033 
3034 		if (progress && need_resched()) {
3035 			memcpy(&key, &found_key, sizeof(key));
3036 			cond_resched();
3037 			btrfs_release_path(root, path);
3038 			btrfs_search_slot(NULL, root, &key, path, 0, 0);
3039 			progress = 0;
3040 			goto next;
3041 		}
3042 		progress = 1;
3043 
3044 		if (btrfs_key_type(&found_key) != BTRFS_EXTENT_ITEM_KEY ||
3045 		    found_key.objectid + found_key.offset <= cur_byte) {
3046 			memcpy(&key, &found_key, sizeof(key));
3047 			key.offset++;
3048 			path->slots[0]++;
3049 			goto next;
3050 		}
3051 
3052 		total_found++;
3053 		cur_byte = found_key.objectid + found_key.offset;
3054 		key.objectid = cur_byte;
3055 		btrfs_release_path(root, path);
3056 		ret = relocate_one_extent(root, path, &found_key);
3057 		__alloc_chunk_for_shrink(root, shrink_block_group, 0);
3058 	}
3059 
3060 	btrfs_release_path(root, path);
3061 
3062 	if (total_found > 0) {
3063 		printk("btrfs relocate found %llu last extent was %llu\n",
3064 		       (unsigned long long)total_found,
3065 		       (unsigned long long)found_key.objectid);
3066 		trans = btrfs_start_transaction(tree_root, 1);
3067 		btrfs_commit_transaction(trans, tree_root);
3068 
3069 		btrfs_clean_old_snapshots(tree_root);
3070 
3071 		trans = btrfs_start_transaction(tree_root, 1);
3072 		btrfs_commit_transaction(trans, tree_root);
3073 		goto again;
3074 	}
3075 
3076 	/*
3077 	 * we've freed all the extents, now remove the block
3078 	 * group item from the tree
3079 	 */
3080 	trans = btrfs_start_transaction(root, 1);
3081 	memcpy(&key, &shrink_block_group->key, sizeof(key));
3082 
3083 	ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3084 	if (ret > 0)
3085 		ret = -EIO;
3086 	if (ret < 0)
3087 		goto out;
3088 
3089 	clear_extent_bits(&info->block_group_cache, key.objectid,
3090 			  key.objectid + key.offset - 1,
3091 			  (unsigned int)-1, GFP_NOFS);
3092 
3093 
3094 	clear_extent_bits(&info->free_space_cache,
3095 			   key.objectid, key.objectid + key.offset - 1,
3096 			   (unsigned int)-1, GFP_NOFS);
3097 
3098 	memset(shrink_block_group, 0, sizeof(*shrink_block_group));
3099 	kfree(shrink_block_group);
3100 
3101 	btrfs_del_item(trans, root, path);
3102 	btrfs_commit_transaction(trans, root);
3103 
3104 	/* the code to unpin extents might set a few bits in the free
3105 	 * space cache for this range again
3106 	 */
3107 	clear_extent_bits(&info->free_space_cache,
3108 			   key.objectid, key.objectid + key.offset - 1,
3109 			   (unsigned int)-1, GFP_NOFS);
3110 out:
3111 	btrfs_free_path(path);
3112 	mutex_unlock(&root->fs_info->alloc_mutex);
3113 	return ret;
3114 }
3115 
3116 int find_first_block_group(struct btrfs_root *root, struct btrfs_path *path,
3117 			   struct btrfs_key *key)
3118 {
3119 	int ret = 0;
3120 	struct btrfs_key found_key;
3121 	struct extent_buffer *leaf;
3122 	int slot;
3123 
3124 	ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
3125 	if (ret < 0)
3126 		goto out;
3127 
3128 	while(1) {
3129 		slot = path->slots[0];
3130 		leaf = path->nodes[0];
3131 		if (slot >= btrfs_header_nritems(leaf)) {
3132 			ret = btrfs_next_leaf(root, path);
3133 			if (ret == 0)
3134 				continue;
3135 			if (ret < 0)
3136 				goto out;
3137 			break;
3138 		}
3139 		btrfs_item_key_to_cpu(leaf, &found_key, slot);
3140 
3141 		if (found_key.objectid >= key->objectid &&
3142 		    found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
3143 			ret = 0;
3144 			goto out;
3145 		}
3146 		path->slots[0]++;
3147 	}
3148 	ret = -ENOENT;
3149 out:
3150 	return ret;
3151 }
3152 
3153 int btrfs_read_block_groups(struct btrfs_root *root)
3154 {
3155 	struct btrfs_path *path;
3156 	int ret;
3157 	int bit;
3158 	struct btrfs_block_group_cache *cache;
3159 	struct btrfs_fs_info *info = root->fs_info;
3160 	struct btrfs_space_info *space_info;
3161 	struct extent_io_tree *block_group_cache;
3162 	struct btrfs_key key;
3163 	struct btrfs_key found_key;
3164 	struct extent_buffer *leaf;
3165 
3166 	block_group_cache = &info->block_group_cache;
3167 	root = info->extent_root;
3168 	key.objectid = 0;
3169 	key.offset = 0;
3170 	btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
3171 	path = btrfs_alloc_path();
3172 	if (!path)
3173 		return -ENOMEM;
3174 
3175 	mutex_lock(&root->fs_info->alloc_mutex);
3176 	while(1) {
3177 		ret = find_first_block_group(root, path, &key);
3178 		if (ret > 0) {
3179 			ret = 0;
3180 			goto error;
3181 		}
3182 		if (ret != 0)
3183 			goto error;
3184 
3185 		leaf = path->nodes[0];
3186 		btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3187 		cache = kzalloc(sizeof(*cache), GFP_NOFS);
3188 		if (!cache) {
3189 			ret = -ENOMEM;
3190 			break;
3191 		}
3192 
3193 		read_extent_buffer(leaf, &cache->item,
3194 				   btrfs_item_ptr_offset(leaf, path->slots[0]),
3195 				   sizeof(cache->item));
3196 		memcpy(&cache->key, &found_key, sizeof(found_key));
3197 
3198 		key.objectid = found_key.objectid + found_key.offset;
3199 		btrfs_release_path(root, path);
3200 		cache->flags = btrfs_block_group_flags(&cache->item);
3201 		bit = 0;
3202 		if (cache->flags & BTRFS_BLOCK_GROUP_DATA) {
3203 			bit = BLOCK_GROUP_DATA;
3204 		} else if (cache->flags & BTRFS_BLOCK_GROUP_SYSTEM) {
3205 			bit = BLOCK_GROUP_SYSTEM;
3206 		} else if (cache->flags & BTRFS_BLOCK_GROUP_METADATA) {
3207 			bit = BLOCK_GROUP_METADATA;
3208 		}
3209 		set_avail_alloc_bits(info, cache->flags);
3210 
3211 		ret = update_space_info(info, cache->flags, found_key.offset,
3212 					btrfs_block_group_used(&cache->item),
3213 					&space_info);
3214 		BUG_ON(ret);
3215 		cache->space_info = space_info;
3216 
3217 		/* use EXTENT_LOCKED to prevent merging */
3218 		set_extent_bits(block_group_cache, found_key.objectid,
3219 				found_key.objectid + found_key.offset - 1,
3220 				bit | EXTENT_LOCKED, GFP_NOFS);
3221 		set_state_private(block_group_cache, found_key.objectid,
3222 				  (unsigned long)cache);
3223 
3224 		if (key.objectid >=
3225 		    btrfs_super_total_bytes(&info->super_copy))
3226 			break;
3227 	}
3228 	ret = 0;
3229 error:
3230 	btrfs_free_path(path);
3231 	mutex_unlock(&root->fs_info->alloc_mutex);
3232 	return ret;
3233 }
3234 
3235 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
3236 			   struct btrfs_root *root, u64 bytes_used,
3237 			   u64 type, u64 chunk_objectid, u64 chunk_offset,
3238 			   u64 size)
3239 {
3240 	int ret;
3241 	int bit = 0;
3242 	struct btrfs_root *extent_root;
3243 	struct btrfs_block_group_cache *cache;
3244 	struct extent_io_tree *block_group_cache;
3245 
3246 	extent_root = root->fs_info->extent_root;
3247 	block_group_cache = &root->fs_info->block_group_cache;
3248 
3249 	cache = kzalloc(sizeof(*cache), GFP_NOFS);
3250 	BUG_ON(!cache);
3251 	cache->key.objectid = chunk_offset;
3252 	cache->key.offset = size;
3253 	btrfs_set_key_type(&cache->key, BTRFS_BLOCK_GROUP_ITEM_KEY);
3254 
3255 	btrfs_set_block_group_used(&cache->item, bytes_used);
3256 	btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
3257 	cache->flags = type;
3258 	btrfs_set_block_group_flags(&cache->item, type);
3259 
3260 	ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
3261 				&cache->space_info);
3262 	BUG_ON(ret);
3263 
3264 	bit = block_group_state_bits(type);
3265 	set_extent_bits(block_group_cache, chunk_offset,
3266 			chunk_offset + size - 1,
3267 			bit | EXTENT_LOCKED, GFP_NOFS);
3268 
3269 	set_state_private(block_group_cache, chunk_offset,
3270 			  (unsigned long)cache);
3271 	ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
3272 				sizeof(cache->item));
3273 	BUG_ON(ret);
3274 
3275 	finish_current_insert(trans, extent_root);
3276 	ret = del_pending_extents(trans, extent_root);
3277 	BUG_ON(ret);
3278 	set_avail_alloc_bits(extent_root->fs_info, type);
3279 
3280 	return 0;
3281 }
3282