xref: /openbmc/linux/fs/btrfs/extent-tree.c (revision 051e1b9f748ae673b7325d3fc049bb838606cffa)
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 	first_free = block_group->key.objectid;
92 	key.objectid = block_group->key.objectid;
93 	key.offset = 0;
94 	btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
95 	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
96 	if (ret < 0)
97 		return ret;
98 	ret = btrfs_previous_item(root, path, 0, BTRFS_EXTENT_ITEM_KEY);
99 	if (ret < 0)
100 		return ret;
101 	if (ret == 0) {
102 		leaf = path->nodes[0];
103 		btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
104 		if (key.objectid + key.offset > first_free)
105 			first_free = key.objectid + key.offset;
106 	}
107 	while(1) {
108 		leaf = path->nodes[0];
109 		slot = path->slots[0];
110 		if (slot >= btrfs_header_nritems(leaf)) {
111 			ret = btrfs_next_leaf(root, path);
112 			if (ret < 0)
113 				goto err;
114 			if (ret == 0) {
115 				continue;
116 			} else {
117 				break;
118 			}
119 		}
120 		btrfs_item_key_to_cpu(leaf, &key, slot);
121 		if (key.objectid < block_group->key.objectid) {
122 			goto next;
123 		}
124 		if (key.objectid >= block_group->key.objectid +
125 		    block_group->key.offset) {
126 			break;
127 		}
128 
129 		if (btrfs_key_type(&key) == BTRFS_EXTENT_ITEM_KEY) {
130 			if (!found) {
131 				last = first_free;
132 				found = 1;
133 			}
134 			if (key.objectid > last) {
135 				hole_size = key.objectid - last;
136 				set_extent_dirty(free_space_cache, last,
137 						 last + hole_size - 1,
138 						 GFP_NOFS);
139 			}
140 			last = key.objectid + key.offset;
141 		}
142 next:
143 		path->slots[0]++;
144 	}
145 
146 	if (!found)
147 		last = first_free;
148 	if (block_group->key.objectid +
149 	    block_group->key.offset > last) {
150 		hole_size = block_group->key.objectid +
151 			block_group->key.offset - last;
152 		set_extent_dirty(free_space_cache, last,
153 				 last + hole_size - 1, GFP_NOFS);
154 	}
155 	block_group->cached = 1;
156 err:
157 	btrfs_free_path(path);
158 	return 0;
159 }
160 
161 struct btrfs_block_group_cache *btrfs_lookup_first_block_group(struct
162 						       btrfs_fs_info *info,
163 							 u64 bytenr)
164 {
165 	struct extent_io_tree *block_group_cache;
166 	struct btrfs_block_group_cache *block_group = NULL;
167 	u64 ptr;
168 	u64 start;
169 	u64 end;
170 	int ret;
171 
172 	bytenr = max_t(u64, bytenr,
173 		       BTRFS_SUPER_INFO_OFFSET + BTRFS_SUPER_INFO_SIZE);
174 	block_group_cache = &info->block_group_cache;
175 	ret = find_first_extent_bit(block_group_cache,
176 				    bytenr, &start, &end,
177 				    BLOCK_GROUP_DATA | BLOCK_GROUP_METADATA |
178 				    BLOCK_GROUP_SYSTEM);
179 	if (ret) {
180 		return NULL;
181 	}
182 	ret = get_state_private(block_group_cache, start, &ptr);
183 	if (ret)
184 		return NULL;
185 
186 	block_group = (struct btrfs_block_group_cache *)(unsigned long)ptr;
187 	return block_group;
188 }
189 
190 struct btrfs_block_group_cache *btrfs_lookup_block_group(struct
191 							 btrfs_fs_info *info,
192 							 u64 bytenr)
193 {
194 	struct extent_io_tree *block_group_cache;
195 	struct btrfs_block_group_cache *block_group = NULL;
196 	u64 ptr;
197 	u64 start;
198 	u64 end;
199 	int ret;
200 
201 	bytenr = max_t(u64, bytenr,
202 		       BTRFS_SUPER_INFO_OFFSET + BTRFS_SUPER_INFO_SIZE);
203 	block_group_cache = &info->block_group_cache;
204 	ret = find_first_extent_bit(block_group_cache,
205 				    bytenr, &start, &end,
206 				    BLOCK_GROUP_DATA | BLOCK_GROUP_METADATA |
207 				    BLOCK_GROUP_SYSTEM);
208 	if (ret) {
209 		return NULL;
210 	}
211 	ret = get_state_private(block_group_cache, start, &ptr);
212 	if (ret)
213 		return NULL;
214 
215 	block_group = (struct btrfs_block_group_cache *)(unsigned long)ptr;
216 	if (block_group->key.objectid <= bytenr && bytenr <
217 	    block_group->key.objectid + block_group->key.offset)
218 		return block_group;
219 	return NULL;
220 }
221 
222 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
223 {
224 	return (cache->flags & bits) == bits;
225 }
226 
227 static int noinline find_search_start(struct btrfs_root *root,
228 			      struct btrfs_block_group_cache **cache_ret,
229 			      u64 *start_ret, u64 num, int data)
230 {
231 	int ret;
232 	struct btrfs_block_group_cache *cache = *cache_ret;
233 	struct extent_io_tree *free_space_cache;
234 	struct extent_state *state;
235 	u64 last;
236 	u64 start = 0;
237 	u64 cache_miss = 0;
238 	u64 total_fs_bytes;
239 	u64 search_start = *start_ret;
240 	int wrapped = 0;
241 
242 	total_fs_bytes = btrfs_super_total_bytes(&root->fs_info->super_copy);
243 	free_space_cache = &root->fs_info->free_space_cache;
244 
245 	if (!cache)
246 		goto out;
247 
248 again:
249 	ret = cache_block_group(root, cache);
250 	if (ret) {
251 		goto out;
252 	}
253 
254 	last = max(search_start, cache->key.objectid);
255 	if (!block_group_bits(cache, data) || cache->ro)
256 		goto new_group;
257 
258 	spin_lock_irq(&free_space_cache->lock);
259 	state = find_first_extent_bit_state(free_space_cache, last, EXTENT_DIRTY);
260 	while(1) {
261 		if (!state) {
262 			if (!cache_miss)
263 				cache_miss = last;
264 			spin_unlock_irq(&free_space_cache->lock);
265 			goto new_group;
266 		}
267 
268 		start = max(last, state->start);
269 		last = state->end + 1;
270 		if (last - start < num) {
271 			do {
272 				state = extent_state_next(state);
273 			} while(state && !(state->state & EXTENT_DIRTY));
274 			continue;
275 		}
276 		spin_unlock_irq(&free_space_cache->lock);
277 		if (cache->ro) {
278 			goto new_group;
279 		}
280 		if (start + num > cache->key.objectid + cache->key.offset)
281 			goto new_group;
282 		if (!block_group_bits(cache, data)) {
283 			printk("block group bits don't match %Lu %d\n", cache->flags, data);
284 		}
285 		*start_ret = start;
286 		return 0;
287 	}
288 out:
289 	cache = btrfs_lookup_block_group(root->fs_info, search_start);
290 	if (!cache) {
291 		printk("Unable to find block group for %Lu\n", search_start);
292 		WARN_ON(1);
293 	}
294 	return -ENOSPC;
295 
296 new_group:
297 	last = cache->key.objectid + cache->key.offset;
298 wrapped:
299 	cache = btrfs_lookup_first_block_group(root->fs_info, last);
300 	if (!cache || cache->key.objectid >= total_fs_bytes) {
301 no_cache:
302 		if (!wrapped) {
303 			wrapped = 1;
304 			last = search_start;
305 			goto wrapped;
306 		}
307 		goto out;
308 	}
309 	if (cache_miss && !cache->cached) {
310 		cache_block_group(root, cache);
311 		last = cache_miss;
312 		cache = btrfs_lookup_first_block_group(root->fs_info, last);
313 	}
314 	cache_miss = 0;
315 	cache = __btrfs_find_block_group(root, cache, last, data, 0);
316 	if (!cache)
317 		goto no_cache;
318 	*cache_ret = cache;
319 	goto again;
320 }
321 
322 static u64 div_factor(u64 num, int factor)
323 {
324 	if (factor == 10)
325 		return num;
326 	num *= factor;
327 	do_div(num, 10);
328 	return num;
329 }
330 
331 static int block_group_state_bits(u64 flags)
332 {
333 	int bits = 0;
334 	if (flags & BTRFS_BLOCK_GROUP_DATA)
335 		bits |= BLOCK_GROUP_DATA;
336 	if (flags & BTRFS_BLOCK_GROUP_METADATA)
337 		bits |= BLOCK_GROUP_METADATA;
338 	if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
339 		bits |= BLOCK_GROUP_SYSTEM;
340 	return bits;
341 }
342 
343 static struct btrfs_block_group_cache *
344 __btrfs_find_block_group(struct btrfs_root *root,
345 			 struct btrfs_block_group_cache *hint,
346 			 u64 search_start, int data, int owner)
347 {
348 	struct btrfs_block_group_cache *cache;
349 	struct extent_io_tree *block_group_cache;
350 	struct btrfs_block_group_cache *found_group = NULL;
351 	struct btrfs_fs_info *info = root->fs_info;
352 	u64 used;
353 	u64 last = 0;
354 	u64 start;
355 	u64 end;
356 	u64 free_check;
357 	u64 ptr;
358 	int bit;
359 	int ret;
360 	int full_search = 0;
361 	int factor = 10;
362 	int wrapped = 0;
363 
364 	block_group_cache = &info->block_group_cache;
365 
366 	if (data & BTRFS_BLOCK_GROUP_METADATA)
367 		factor = 9;
368 
369 	bit = block_group_state_bits(data);
370 
371 	if (search_start) {
372 		struct btrfs_block_group_cache *shint;
373 		shint = btrfs_lookup_first_block_group(info, search_start);
374 		if (shint && block_group_bits(shint, data) && !shint->ro) {
375 			used = btrfs_block_group_used(&shint->item);
376 			if (used + shint->pinned <
377 			    div_factor(shint->key.offset, factor)) {
378 				return shint;
379 			}
380 		}
381 	}
382 	if (hint && !hint->ro && block_group_bits(hint, data)) {
383 		used = btrfs_block_group_used(&hint->item);
384 		if (used + hint->pinned <
385 		    div_factor(hint->key.offset, factor)) {
386 			return hint;
387 		}
388 		last = hint->key.objectid + hint->key.offset;
389 	} else {
390 		if (hint)
391 			last = max(hint->key.objectid, search_start);
392 		else
393 			last = search_start;
394 	}
395 again:
396 	while(1) {
397 		ret = find_first_extent_bit(block_group_cache, last,
398 					    &start, &end, bit);
399 		if (ret)
400 			break;
401 
402 		ret = get_state_private(block_group_cache, start, &ptr);
403 		if (ret) {
404 			last = end + 1;
405 			continue;
406 		}
407 
408 		cache = (struct btrfs_block_group_cache *)(unsigned long)ptr;
409 		last = cache->key.objectid + cache->key.offset;
410 		used = btrfs_block_group_used(&cache->item);
411 
412 		if (!cache->ro && block_group_bits(cache, data)) {
413 			free_check = div_factor(cache->key.offset, factor);
414 			if (used + cache->pinned < free_check) {
415 				found_group = cache;
416 				goto found;
417 			}
418 		}
419 		cond_resched();
420 	}
421 	if (!wrapped) {
422 		last = search_start;
423 		wrapped = 1;
424 		goto again;
425 	}
426 	if (!full_search && factor < 10) {
427 		last = search_start;
428 		full_search = 1;
429 		factor = 10;
430 		goto again;
431 	}
432 found:
433 	return found_group;
434 }
435 
436 struct btrfs_block_group_cache *btrfs_find_block_group(struct btrfs_root *root,
437 						 struct btrfs_block_group_cache
438 						 *hint, u64 search_start,
439 						 int data, int owner)
440 {
441 
442 	struct btrfs_block_group_cache *ret;
443 	mutex_lock(&root->fs_info->alloc_mutex);
444 	ret = __btrfs_find_block_group(root, hint, search_start, data, owner);
445 	mutex_unlock(&root->fs_info->alloc_mutex);
446 	return ret;
447 }
448 static u64 hash_extent_ref(u64 root_objectid, u64 ref_generation,
449 			   u64 owner, u64 owner_offset)
450 {
451 	u32 high_crc = ~(u32)0;
452 	u32 low_crc = ~(u32)0;
453 	__le64 lenum;
454 	lenum = cpu_to_le64(root_objectid);
455 	high_crc = btrfs_crc32c(high_crc, &lenum, sizeof(lenum));
456 	lenum = cpu_to_le64(ref_generation);
457 	low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum));
458 	if (owner >= BTRFS_FIRST_FREE_OBJECTID) {
459 		lenum = cpu_to_le64(owner);
460 		low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum));
461 		lenum = cpu_to_le64(owner_offset);
462 		low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum));
463 	}
464 	return ((u64)high_crc << 32) | (u64)low_crc;
465 }
466 
467 static int match_extent_ref(struct extent_buffer *leaf,
468 			    struct btrfs_extent_ref *disk_ref,
469 			    struct btrfs_extent_ref *cpu_ref)
470 {
471 	int ret;
472 	int len;
473 
474 	if (cpu_ref->objectid)
475 		len = sizeof(*cpu_ref);
476 	else
477 		len = 2 * sizeof(u64);
478 	ret = memcmp_extent_buffer(leaf, cpu_ref, (unsigned long)disk_ref,
479 				   len);
480 	return ret == 0;
481 }
482 
483 static int noinline lookup_extent_backref(struct btrfs_trans_handle *trans,
484 					  struct btrfs_root *root,
485 					  struct btrfs_path *path, u64 bytenr,
486 					  u64 root_objectid,
487 					  u64 ref_generation, u64 owner,
488 					  u64 owner_offset, int del)
489 {
490 	u64 hash;
491 	struct btrfs_key key;
492 	struct btrfs_key found_key;
493 	struct btrfs_extent_ref ref;
494 	struct extent_buffer *leaf;
495 	struct btrfs_extent_ref *disk_ref;
496 	int ret;
497 	int ret2;
498 
499 	btrfs_set_stack_ref_root(&ref, root_objectid);
500 	btrfs_set_stack_ref_generation(&ref, ref_generation);
501 	btrfs_set_stack_ref_objectid(&ref, owner);
502 	btrfs_set_stack_ref_offset(&ref, owner_offset);
503 
504 	hash = hash_extent_ref(root_objectid, ref_generation, owner,
505 			       owner_offset);
506 	key.offset = hash;
507 	key.objectid = bytenr;
508 	key.type = BTRFS_EXTENT_REF_KEY;
509 
510 	while (1) {
511 		ret = btrfs_search_slot(trans, root, &key, path,
512 					del ? -1 : 0, del);
513 		if (ret < 0)
514 			goto out;
515 		leaf = path->nodes[0];
516 		if (ret != 0) {
517 			u32 nritems = btrfs_header_nritems(leaf);
518 			if (path->slots[0] >= nritems) {
519 				ret2 = btrfs_next_leaf(root, path);
520 				if (ret2)
521 					goto out;
522 				leaf = path->nodes[0];
523 			}
524 			btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
525 			if (found_key.objectid != bytenr ||
526 			    found_key.type != BTRFS_EXTENT_REF_KEY)
527 				goto out;
528 			key.offset = found_key.offset;
529 			if (del) {
530 				btrfs_release_path(root, path);
531 				continue;
532 			}
533 		}
534 		disk_ref = btrfs_item_ptr(path->nodes[0],
535 					  path->slots[0],
536 					  struct btrfs_extent_ref);
537 		if (match_extent_ref(path->nodes[0], disk_ref, &ref)) {
538 			ret = 0;
539 			goto out;
540 		}
541 		btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
542 		key.offset = found_key.offset + 1;
543 		btrfs_release_path(root, path);
544 	}
545 out:
546 	return ret;
547 }
548 
549 /*
550  * Back reference rules.  Back refs have three main goals:
551  *
552  * 1) differentiate between all holders of references to an extent so that
553  *    when a reference is dropped we can make sure it was a valid reference
554  *    before freeing the extent.
555  *
556  * 2) Provide enough information to quickly find the holders of an extent
557  *    if we notice a given block is corrupted or bad.
558  *
559  * 3) Make it easy to migrate blocks for FS shrinking or storage pool
560  *    maintenance.  This is actually the same as #2, but with a slightly
561  *    different use case.
562  *
563  * File extents can be referenced by:
564  *
565  * - multiple snapshots, subvolumes, or different generations in one subvol
566  * - different files inside a single subvolume (in theory, not implemented yet)
567  * - different offsets inside a file (bookend extents in file.c)
568  *
569  * The extent ref structure has fields for:
570  *
571  * - Objectid of the subvolume root
572  * - Generation number of the tree holding the reference
573  * - objectid of the file holding the reference
574  * - offset in the file corresponding to the key holding the reference
575  *
576  * When a file extent is allocated the fields are filled in:
577  *     (root_key.objectid, trans->transid, inode objectid, offset in file)
578  *
579  * When a leaf is cow'd new references are added for every file extent found
580  * in the leaf.  It looks the same as the create case, but trans->transid
581  * will be different when the block is cow'd.
582  *
583  *     (root_key.objectid, trans->transid, inode objectid, offset in file)
584  *
585  * When a file extent is removed either during snapshot deletion or file
586  * truncation, the corresponding back reference is found
587  * by searching for:
588  *
589  *     (btrfs_header_owner(leaf), btrfs_header_generation(leaf),
590  *      inode objectid, offset in file)
591  *
592  * Btree extents can be referenced by:
593  *
594  * - Different subvolumes
595  * - Different generations of the same subvolume
596  *
597  * Storing sufficient information for a full reverse mapping of a btree
598  * block would require storing the lowest key of the block in the backref,
599  * and it would require updating that lowest key either before write out or
600  * every time it changed.  Instead, the objectid of the lowest key is stored
601  * along with the level of the tree block.  This provides a hint
602  * about where in the btree the block can be found.  Searches through the
603  * btree only need to look for a pointer to that block, so they stop one
604  * level higher than the level recorded in the backref.
605  *
606  * Some btrees do not do reference counting on their extents.  These
607  * include the extent tree and the tree of tree roots.  Backrefs for these
608  * trees always have a generation of zero.
609  *
610  * When a tree block is created, back references are inserted:
611  *
612  * (root->root_key.objectid, trans->transid or zero, level, lowest_key_objectid)
613  *
614  * When a tree block is cow'd in a reference counted root,
615  * new back references are added for all the blocks it points to.
616  * These are of the form (trans->transid will have increased since creation):
617  *
618  * (root->root_key.objectid, trans->transid, level, lowest_key_objectid)
619  *
620  * Because the lowest_key_objectid and the level are just hints
621  * they are not used when backrefs are deleted.  When a backref is deleted:
622  *
623  * if backref was for a tree root:
624  *     root_objectid = root->root_key.objectid
625  * else
626  *     root_objectid = btrfs_header_owner(parent)
627  *
628  * (root_objectid, btrfs_header_generation(parent) or zero, 0, 0)
629  *
630  * Back Reference Key hashing:
631  *
632  * Back references have four fields, each 64 bits long.  Unfortunately,
633  * This is hashed into a single 64 bit number and placed into the key offset.
634  * The key objectid corresponds to the first byte in the extent, and the
635  * key type is set to BTRFS_EXTENT_REF_KEY
636  */
637 int btrfs_insert_extent_backref(struct btrfs_trans_handle *trans,
638 				 struct btrfs_root *root,
639 				 struct btrfs_path *path, u64 bytenr,
640 				 u64 root_objectid, u64 ref_generation,
641 				 u64 owner, u64 owner_offset)
642 {
643 	u64 hash;
644 	struct btrfs_key key;
645 	struct btrfs_extent_ref ref;
646 	struct btrfs_extent_ref *disk_ref;
647 	int ret;
648 
649 	btrfs_set_stack_ref_root(&ref, root_objectid);
650 	btrfs_set_stack_ref_generation(&ref, ref_generation);
651 	btrfs_set_stack_ref_objectid(&ref, owner);
652 	btrfs_set_stack_ref_offset(&ref, owner_offset);
653 
654 	hash = hash_extent_ref(root_objectid, ref_generation, owner,
655 			       owner_offset);
656 	key.offset = hash;
657 	key.objectid = bytenr;
658 	key.type = BTRFS_EXTENT_REF_KEY;
659 
660 	ret = btrfs_insert_empty_item(trans, root, path, &key, sizeof(ref));
661 	while (ret == -EEXIST) {
662 		disk_ref = btrfs_item_ptr(path->nodes[0], path->slots[0],
663 					  struct btrfs_extent_ref);
664 		if (match_extent_ref(path->nodes[0], disk_ref, &ref))
665 			goto out;
666 		key.offset++;
667 		btrfs_release_path(root, path);
668 		ret = btrfs_insert_empty_item(trans, root, path, &key,
669 					      sizeof(ref));
670 	}
671 	if (ret)
672 		goto out;
673 	disk_ref = btrfs_item_ptr(path->nodes[0], path->slots[0],
674 				  struct btrfs_extent_ref);
675 	write_extent_buffer(path->nodes[0], &ref, (unsigned long)disk_ref,
676 			    sizeof(ref));
677 	btrfs_mark_buffer_dirty(path->nodes[0]);
678 out:
679 	btrfs_release_path(root, path);
680 	return ret;
681 }
682 
683 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
684 				struct btrfs_root *root,
685 				u64 bytenr, u64 num_bytes,
686 				u64 root_objectid, u64 ref_generation,
687 				u64 owner, u64 owner_offset)
688 {
689 	struct btrfs_path *path;
690 	int ret;
691 	struct btrfs_key key;
692 	struct extent_buffer *l;
693 	struct btrfs_extent_item *item;
694 	u32 refs;
695 
696 	WARN_ON(num_bytes < root->sectorsize);
697 	path = btrfs_alloc_path();
698 	if (!path)
699 		return -ENOMEM;
700 
701 	path->reada = 1;
702 	key.objectid = bytenr;
703 	btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
704 	key.offset = num_bytes;
705 	ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path,
706 				0, 1);
707 	if (ret < 0)
708 		return ret;
709 	if (ret != 0) {
710 		BUG();
711 	}
712 	BUG_ON(ret != 0);
713 	l = path->nodes[0];
714 	item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
715 	refs = btrfs_extent_refs(l, item);
716 	btrfs_set_extent_refs(l, item, refs + 1);
717 	btrfs_mark_buffer_dirty(path->nodes[0]);
718 
719 	btrfs_release_path(root->fs_info->extent_root, path);
720 
721 	path->reada = 1;
722 	ret = btrfs_insert_extent_backref(trans, root->fs_info->extent_root,
723 					  path, bytenr, root_objectid,
724 					  ref_generation, owner, owner_offset);
725 	BUG_ON(ret);
726 	finish_current_insert(trans, root->fs_info->extent_root);
727 	del_pending_extents(trans, root->fs_info->extent_root);
728 
729 	btrfs_free_path(path);
730 	return 0;
731 }
732 
733 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
734 				struct btrfs_root *root,
735 				u64 bytenr, u64 num_bytes,
736 				u64 root_objectid, u64 ref_generation,
737 				u64 owner, u64 owner_offset)
738 {
739 	int ret;
740 
741 	mutex_lock(&root->fs_info->alloc_mutex);
742 	ret = __btrfs_inc_extent_ref(trans, root, bytenr, num_bytes,
743 				     root_objectid, ref_generation,
744 				     owner, owner_offset);
745 	mutex_unlock(&root->fs_info->alloc_mutex);
746 	return ret;
747 }
748 
749 int btrfs_extent_post_op(struct btrfs_trans_handle *trans,
750 			 struct btrfs_root *root)
751 {
752 	finish_current_insert(trans, root->fs_info->extent_root);
753 	del_pending_extents(trans, root->fs_info->extent_root);
754 	return 0;
755 }
756 
757 static int lookup_extent_ref(struct btrfs_trans_handle *trans,
758 			     struct btrfs_root *root, u64 bytenr,
759 			     u64 num_bytes, u32 *refs)
760 {
761 	struct btrfs_path *path;
762 	int ret;
763 	struct btrfs_key key;
764 	struct extent_buffer *l;
765 	struct btrfs_extent_item *item;
766 
767 	WARN_ON(num_bytes < root->sectorsize);
768 	path = btrfs_alloc_path();
769 	path->reada = 1;
770 	key.objectid = bytenr;
771 	key.offset = num_bytes;
772 	btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
773 	ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path,
774 				0, 0);
775 	if (ret < 0)
776 		goto out;
777 	if (ret != 0) {
778 		btrfs_print_leaf(root, path->nodes[0]);
779 		printk("failed to find block number %Lu\n", bytenr);
780 		BUG();
781 	}
782 	l = path->nodes[0];
783 	item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
784 	*refs = btrfs_extent_refs(l, item);
785 out:
786 	btrfs_free_path(path);
787 	return 0;
788 }
789 
790 u32 btrfs_count_snapshots_in_path(struct btrfs_root *root,
791 				  struct btrfs_path *count_path,
792 				  u64 expected_owner,
793 				  u64 first_extent)
794 {
795 	struct btrfs_root *extent_root = root->fs_info->extent_root;
796 	struct btrfs_path *path;
797 	u64 bytenr;
798 	u64 found_objectid;
799 	u64 found_owner;
800 	u64 root_objectid = root->root_key.objectid;
801 	u32 total_count = 0;
802 	u32 extent_refs;
803 	u32 cur_count;
804 	u32 nritems;
805 	int ret;
806 	struct btrfs_key key;
807 	struct btrfs_key found_key;
808 	struct extent_buffer *l;
809 	struct btrfs_extent_item *item;
810 	struct btrfs_extent_ref *ref_item;
811 	int level = -1;
812 
813 	/* FIXME, needs locking */
814 	BUG();
815 
816 	mutex_lock(&root->fs_info->alloc_mutex);
817 	path = btrfs_alloc_path();
818 again:
819 	if (level == -1)
820 		bytenr = first_extent;
821 	else
822 		bytenr = count_path->nodes[level]->start;
823 
824 	cur_count = 0;
825 	key.objectid = bytenr;
826 	key.offset = 0;
827 
828 	btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
829 	ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
830 	if (ret < 0)
831 		goto out;
832 	BUG_ON(ret == 0);
833 
834 	l = path->nodes[0];
835 	btrfs_item_key_to_cpu(l, &found_key, path->slots[0]);
836 
837 	if (found_key.objectid != bytenr ||
838 	    found_key.type != BTRFS_EXTENT_ITEM_KEY) {
839 		goto out;
840 	}
841 
842 	item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
843 	extent_refs = btrfs_extent_refs(l, item);
844 	while (1) {
845 		l = path->nodes[0];
846 		nritems = btrfs_header_nritems(l);
847 		if (path->slots[0] >= nritems) {
848 			ret = btrfs_next_leaf(extent_root, path);
849 			if (ret == 0)
850 				continue;
851 			break;
852 		}
853 		btrfs_item_key_to_cpu(l, &found_key, path->slots[0]);
854 		if (found_key.objectid != bytenr)
855 			break;
856 
857 		if (found_key.type != BTRFS_EXTENT_REF_KEY) {
858 			path->slots[0]++;
859 			continue;
860 		}
861 
862 		cur_count++;
863 		ref_item = btrfs_item_ptr(l, path->slots[0],
864 					  struct btrfs_extent_ref);
865 		found_objectid = btrfs_ref_root(l, ref_item);
866 
867 		if (found_objectid != root_objectid) {
868 			total_count = 2;
869 			goto out;
870 		}
871 		if (level == -1) {
872 			found_owner = btrfs_ref_objectid(l, ref_item);
873 			if (found_owner != expected_owner) {
874 				total_count = 2;
875 				goto out;
876 			}
877 			/*
878 			 * nasty.  we don't count a reference held by
879 			 * the running transaction.  This allows nodatacow
880 			 * to avoid cow most of the time
881 			 */
882 			if (found_owner >= BTRFS_FIRST_FREE_OBJECTID &&
883 			    btrfs_ref_generation(l, ref_item) ==
884 			    root->fs_info->generation) {
885 				extent_refs--;
886 			}
887 		}
888 		total_count = 1;
889 		path->slots[0]++;
890 	}
891 	/*
892 	 * if there is more than one reference against a data extent,
893 	 * we have to assume the other ref is another snapshot
894 	 */
895 	if (level == -1 && extent_refs > 1) {
896 		total_count = 2;
897 		goto out;
898 	}
899 	if (cur_count == 0) {
900 		total_count = 0;
901 		goto out;
902 	}
903 	if (level >= 0 && root->node == count_path->nodes[level])
904 		goto out;
905 	level++;
906 	btrfs_release_path(root, path);
907 	goto again;
908 
909 out:
910 	btrfs_free_path(path);
911 	mutex_unlock(&root->fs_info->alloc_mutex);
912 	return total_count;
913 }
914 
915 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
916 		  struct extent_buffer *buf)
917 {
918 	u64 bytenr;
919 	u32 nritems;
920 	struct btrfs_key key;
921 	struct btrfs_file_extent_item *fi;
922 	int i;
923 	int level;
924 	int ret;
925 	int faili;
926 
927 	if (!root->ref_cows)
928 		return 0;
929 
930 	mutex_lock(&root->fs_info->alloc_mutex);
931 	level = btrfs_header_level(buf);
932 	nritems = btrfs_header_nritems(buf);
933 	for (i = 0; i < nritems; i++) {
934 		if (level == 0) {
935 			u64 disk_bytenr;
936 			btrfs_item_key_to_cpu(buf, &key, i);
937 			if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
938 				continue;
939 			fi = btrfs_item_ptr(buf, i,
940 					    struct btrfs_file_extent_item);
941 			if (btrfs_file_extent_type(buf, fi) ==
942 			    BTRFS_FILE_EXTENT_INLINE)
943 				continue;
944 			disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
945 			if (disk_bytenr == 0)
946 				continue;
947 			ret = __btrfs_inc_extent_ref(trans, root, disk_bytenr,
948 				    btrfs_file_extent_disk_num_bytes(buf, fi),
949 				    root->root_key.objectid, trans->transid,
950 				    key.objectid, key.offset);
951 			if (ret) {
952 				faili = i;
953 				goto fail;
954 			}
955 		} else {
956 			bytenr = btrfs_node_blockptr(buf, i);
957 			btrfs_node_key_to_cpu(buf, &key, i);
958 			ret = __btrfs_inc_extent_ref(trans, root, bytenr,
959 					   btrfs_level_size(root, level - 1),
960 					   root->root_key.objectid,
961 					   trans->transid,
962 					   level - 1, key.objectid);
963 			if (ret) {
964 				faili = i;
965 				goto fail;
966 			}
967 		}
968 	}
969 	mutex_unlock(&root->fs_info->alloc_mutex);
970 	return 0;
971 fail:
972 	WARN_ON(1);
973 #if 0
974 	for (i =0; i < faili; i++) {
975 		if (level == 0) {
976 			u64 disk_bytenr;
977 			btrfs_item_key_to_cpu(buf, &key, i);
978 			if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
979 				continue;
980 			fi = btrfs_item_ptr(buf, i,
981 					    struct btrfs_file_extent_item);
982 			if (btrfs_file_extent_type(buf, fi) ==
983 			    BTRFS_FILE_EXTENT_INLINE)
984 				continue;
985 			disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
986 			if (disk_bytenr == 0)
987 				continue;
988 			err = btrfs_free_extent(trans, root, disk_bytenr,
989 				    btrfs_file_extent_disk_num_bytes(buf,
990 								      fi), 0);
991 			BUG_ON(err);
992 		} else {
993 			bytenr = btrfs_node_blockptr(buf, i);
994 			err = btrfs_free_extent(trans, root, bytenr,
995 					btrfs_level_size(root, level - 1), 0);
996 			BUG_ON(err);
997 		}
998 	}
999 #endif
1000 	mutex_unlock(&root->fs_info->alloc_mutex);
1001 	return ret;
1002 }
1003 
1004 static int write_one_cache_group(struct btrfs_trans_handle *trans,
1005 				 struct btrfs_root *root,
1006 				 struct btrfs_path *path,
1007 				 struct btrfs_block_group_cache *cache)
1008 {
1009 	int ret;
1010 	int pending_ret;
1011 	struct btrfs_root *extent_root = root->fs_info->extent_root;
1012 	unsigned long bi;
1013 	struct extent_buffer *leaf;
1014 
1015 	ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
1016 	if (ret < 0)
1017 		goto fail;
1018 	BUG_ON(ret);
1019 
1020 	leaf = path->nodes[0];
1021 	bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
1022 	write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
1023 	btrfs_mark_buffer_dirty(leaf);
1024 	btrfs_release_path(extent_root, path);
1025 fail:
1026 	finish_current_insert(trans, extent_root);
1027 	pending_ret = del_pending_extents(trans, extent_root);
1028 	if (ret)
1029 		return ret;
1030 	if (pending_ret)
1031 		return pending_ret;
1032 	return 0;
1033 
1034 }
1035 
1036 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
1037 				   struct btrfs_root *root)
1038 {
1039 	struct extent_io_tree *block_group_cache;
1040 	struct btrfs_block_group_cache *cache;
1041 	int ret;
1042 	int err = 0;
1043 	int werr = 0;
1044 	struct btrfs_path *path;
1045 	u64 last = 0;
1046 	u64 start;
1047 	u64 end;
1048 	u64 ptr;
1049 
1050 	block_group_cache = &root->fs_info->block_group_cache;
1051 	path = btrfs_alloc_path();
1052 	if (!path)
1053 		return -ENOMEM;
1054 
1055 	mutex_lock(&root->fs_info->alloc_mutex);
1056 	while(1) {
1057 		ret = find_first_extent_bit(block_group_cache, last,
1058 					    &start, &end, BLOCK_GROUP_DIRTY);
1059 		if (ret)
1060 			break;
1061 
1062 		last = end + 1;
1063 		ret = get_state_private(block_group_cache, start, &ptr);
1064 		if (ret)
1065 			break;
1066 		cache = (struct btrfs_block_group_cache *)(unsigned long)ptr;
1067 		err = write_one_cache_group(trans, root,
1068 					    path, cache);
1069 		/*
1070 		 * if we fail to write the cache group, we want
1071 		 * to keep it marked dirty in hopes that a later
1072 		 * write will work
1073 		 */
1074 		if (err) {
1075 			werr = err;
1076 			continue;
1077 		}
1078 		clear_extent_bits(block_group_cache, start, end,
1079 				  BLOCK_GROUP_DIRTY, GFP_NOFS);
1080 	}
1081 	btrfs_free_path(path);
1082 	mutex_unlock(&root->fs_info->alloc_mutex);
1083 	return werr;
1084 }
1085 
1086 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
1087 						  u64 flags)
1088 {
1089 	struct list_head *head = &info->space_info;
1090 	struct list_head *cur;
1091 	struct btrfs_space_info *found;
1092 	list_for_each(cur, head) {
1093 		found = list_entry(cur, struct btrfs_space_info, list);
1094 		if (found->flags == flags)
1095 			return found;
1096 	}
1097 	return NULL;
1098 
1099 }
1100 
1101 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
1102 			     u64 total_bytes, u64 bytes_used,
1103 			     struct btrfs_space_info **space_info)
1104 {
1105 	struct btrfs_space_info *found;
1106 
1107 	found = __find_space_info(info, flags);
1108 	if (found) {
1109 		found->total_bytes += total_bytes;
1110 		found->bytes_used += bytes_used;
1111 		found->full = 0;
1112 		WARN_ON(found->total_bytes < found->bytes_used);
1113 		*space_info = found;
1114 		return 0;
1115 	}
1116 	found = kmalloc(sizeof(*found), GFP_NOFS);
1117 	if (!found)
1118 		return -ENOMEM;
1119 
1120 	list_add(&found->list, &info->space_info);
1121 	found->flags = flags;
1122 	found->total_bytes = total_bytes;
1123 	found->bytes_used = bytes_used;
1124 	found->bytes_pinned = 0;
1125 	found->full = 0;
1126 	found->force_alloc = 0;
1127 	*space_info = found;
1128 	return 0;
1129 }
1130 
1131 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
1132 {
1133 	u64 extra_flags = flags & (BTRFS_BLOCK_GROUP_RAID0 |
1134 				   BTRFS_BLOCK_GROUP_RAID1 |
1135 				   BTRFS_BLOCK_GROUP_RAID10 |
1136 				   BTRFS_BLOCK_GROUP_DUP);
1137 	if (extra_flags) {
1138 		if (flags & BTRFS_BLOCK_GROUP_DATA)
1139 			fs_info->avail_data_alloc_bits |= extra_flags;
1140 		if (flags & BTRFS_BLOCK_GROUP_METADATA)
1141 			fs_info->avail_metadata_alloc_bits |= extra_flags;
1142 		if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
1143 			fs_info->avail_system_alloc_bits |= extra_flags;
1144 	}
1145 }
1146 
1147 static u64 reduce_alloc_profile(struct btrfs_root *root, u64 flags)
1148 {
1149 	u64 num_devices = root->fs_info->fs_devices->num_devices;
1150 
1151 	if (num_devices == 1)
1152 		flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0);
1153 	if (num_devices < 4)
1154 		flags &= ~BTRFS_BLOCK_GROUP_RAID10;
1155 
1156 	if ((flags & BTRFS_BLOCK_GROUP_DUP) &&
1157 	    (flags & (BTRFS_BLOCK_GROUP_RAID1 |
1158 		      BTRFS_BLOCK_GROUP_RAID10))) {
1159 		flags &= ~BTRFS_BLOCK_GROUP_DUP;
1160 	}
1161 
1162 	if ((flags & BTRFS_BLOCK_GROUP_RAID1) &&
1163 	    (flags & BTRFS_BLOCK_GROUP_RAID10)) {
1164 		flags &= ~BTRFS_BLOCK_GROUP_RAID1;
1165 	}
1166 
1167 	if ((flags & BTRFS_BLOCK_GROUP_RAID0) &&
1168 	    ((flags & BTRFS_BLOCK_GROUP_RAID1) |
1169 	     (flags & BTRFS_BLOCK_GROUP_RAID10) |
1170 	     (flags & BTRFS_BLOCK_GROUP_DUP)))
1171 		flags &= ~BTRFS_BLOCK_GROUP_RAID0;
1172 	return flags;
1173 }
1174 
1175 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
1176 			  struct btrfs_root *extent_root, u64 alloc_bytes,
1177 			  u64 flags, int force)
1178 {
1179 	struct btrfs_space_info *space_info;
1180 	u64 thresh;
1181 	u64 start;
1182 	u64 num_bytes;
1183 	int ret;
1184 
1185 	flags = reduce_alloc_profile(extent_root, flags);
1186 
1187 	space_info = __find_space_info(extent_root->fs_info, flags);
1188 	if (!space_info) {
1189 		ret = update_space_info(extent_root->fs_info, flags,
1190 					0, 0, &space_info);
1191 		BUG_ON(ret);
1192 	}
1193 	BUG_ON(!space_info);
1194 
1195 	if (space_info->force_alloc) {
1196 		force = 1;
1197 		space_info->force_alloc = 0;
1198 	}
1199 	if (space_info->full)
1200 		goto out;
1201 
1202 	thresh = div_factor(space_info->total_bytes, 6);
1203 	if (!force &&
1204 	   (space_info->bytes_used + space_info->bytes_pinned + alloc_bytes) <
1205 	    thresh)
1206 		goto out;
1207 
1208 	mutex_lock(&extent_root->fs_info->chunk_mutex);
1209 	ret = btrfs_alloc_chunk(trans, extent_root, &start, &num_bytes, flags);
1210 	if (ret == -ENOSPC) {
1211 printk("space info full %Lu\n", flags);
1212 		space_info->full = 1;
1213 		goto out;
1214 	}
1215 	BUG_ON(ret);
1216 
1217 	ret = btrfs_make_block_group(trans, extent_root, 0, flags,
1218 		     BTRFS_FIRST_CHUNK_TREE_OBJECTID, start, num_bytes);
1219 	BUG_ON(ret);
1220 	mutex_unlock(&extent_root->fs_info->chunk_mutex);
1221 out:
1222 	return 0;
1223 }
1224 
1225 static int update_block_group(struct btrfs_trans_handle *trans,
1226 			      struct btrfs_root *root,
1227 			      u64 bytenr, u64 num_bytes, int alloc,
1228 			      int mark_free)
1229 {
1230 	struct btrfs_block_group_cache *cache;
1231 	struct btrfs_fs_info *info = root->fs_info;
1232 	u64 total = num_bytes;
1233 	u64 old_val;
1234 	u64 byte_in_group;
1235 	u64 start;
1236 	u64 end;
1237 
1238 	while(total) {
1239 		cache = btrfs_lookup_block_group(info, bytenr);
1240 		if (!cache) {
1241 			return -1;
1242 		}
1243 		byte_in_group = bytenr - cache->key.objectid;
1244 		WARN_ON(byte_in_group > cache->key.offset);
1245 		start = cache->key.objectid;
1246 		end = start + cache->key.offset - 1;
1247 		set_extent_bits(&info->block_group_cache, start, end,
1248 				BLOCK_GROUP_DIRTY, GFP_NOFS);
1249 
1250 		old_val = btrfs_block_group_used(&cache->item);
1251 		num_bytes = min(total, cache->key.offset - byte_in_group);
1252 		if (alloc) {
1253 			old_val += num_bytes;
1254 			cache->space_info->bytes_used += num_bytes;
1255 		} else {
1256 			old_val -= num_bytes;
1257 			cache->space_info->bytes_used -= num_bytes;
1258 			if (mark_free) {
1259 				set_extent_dirty(&info->free_space_cache,
1260 						 bytenr, bytenr + num_bytes - 1,
1261 						 GFP_NOFS);
1262 			}
1263 		}
1264 		btrfs_set_block_group_used(&cache->item, old_val);
1265 		total -= num_bytes;
1266 		bytenr += num_bytes;
1267 	}
1268 	return 0;
1269 }
1270 
1271 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
1272 {
1273 	u64 start;
1274 	u64 end;
1275 	int ret;
1276 	ret = find_first_extent_bit(&root->fs_info->block_group_cache,
1277 				    search_start, &start, &end,
1278 				    BLOCK_GROUP_DATA | BLOCK_GROUP_METADATA |
1279 				    BLOCK_GROUP_SYSTEM);
1280 	if (ret)
1281 		return 0;
1282 	return start;
1283 }
1284 
1285 
1286 static int update_pinned_extents(struct btrfs_root *root,
1287 				u64 bytenr, u64 num, int pin)
1288 {
1289 	u64 len;
1290 	struct btrfs_block_group_cache *cache;
1291 	struct btrfs_fs_info *fs_info = root->fs_info;
1292 
1293 	if (pin) {
1294 		set_extent_dirty(&fs_info->pinned_extents,
1295 				bytenr, bytenr + num - 1, GFP_NOFS);
1296 	} else {
1297 		clear_extent_dirty(&fs_info->pinned_extents,
1298 				bytenr, bytenr + num - 1, GFP_NOFS);
1299 	}
1300 	while (num > 0) {
1301 		cache = btrfs_lookup_block_group(fs_info, bytenr);
1302 		if (!cache) {
1303 			u64 first = first_logical_byte(root, bytenr);
1304 			WARN_ON(first < bytenr);
1305 			len = min(first - bytenr, num);
1306 		} else {
1307 			len = min(num, cache->key.offset -
1308 				  (bytenr - cache->key.objectid));
1309 		}
1310 		if (pin) {
1311 			if (cache) {
1312 				cache->pinned += len;
1313 				cache->space_info->bytes_pinned += len;
1314 			}
1315 			fs_info->total_pinned += len;
1316 		} else {
1317 			if (cache) {
1318 				cache->pinned -= len;
1319 				cache->space_info->bytes_pinned -= len;
1320 			}
1321 			fs_info->total_pinned -= len;
1322 		}
1323 		bytenr += len;
1324 		num -= len;
1325 	}
1326 	return 0;
1327 }
1328 
1329 int btrfs_copy_pinned(struct btrfs_root *root, struct extent_io_tree *copy)
1330 {
1331 	u64 last = 0;
1332 	u64 start;
1333 	u64 end;
1334 	struct extent_io_tree *pinned_extents = &root->fs_info->pinned_extents;
1335 	int ret;
1336 
1337 	while(1) {
1338 		ret = find_first_extent_bit(pinned_extents, last,
1339 					    &start, &end, EXTENT_DIRTY);
1340 		if (ret)
1341 			break;
1342 		set_extent_dirty(copy, start, end, GFP_NOFS);
1343 		last = end + 1;
1344 	}
1345 	return 0;
1346 }
1347 
1348 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
1349 			       struct btrfs_root *root,
1350 			       struct extent_io_tree *unpin)
1351 {
1352 	u64 start;
1353 	u64 end;
1354 	int ret;
1355 	struct extent_io_tree *free_space_cache;
1356 	free_space_cache = &root->fs_info->free_space_cache;
1357 
1358 	mutex_lock(&root->fs_info->alloc_mutex);
1359 	while(1) {
1360 		ret = find_first_extent_bit(unpin, 0, &start, &end,
1361 					    EXTENT_DIRTY);
1362 		if (ret)
1363 			break;
1364 		update_pinned_extents(root, start, end + 1 - start, 0);
1365 		clear_extent_dirty(unpin, start, end, GFP_NOFS);
1366 		set_extent_dirty(free_space_cache, start, end, GFP_NOFS);
1367 	}
1368 	mutex_unlock(&root->fs_info->alloc_mutex);
1369 	return 0;
1370 }
1371 
1372 static int finish_current_insert(struct btrfs_trans_handle *trans,
1373 				 struct btrfs_root *extent_root)
1374 {
1375 	u64 start;
1376 	u64 end;
1377 	struct btrfs_fs_info *info = extent_root->fs_info;
1378 	struct extent_buffer *eb;
1379 	struct btrfs_path *path;
1380 	struct btrfs_key ins;
1381 	struct btrfs_disk_key first;
1382 	struct btrfs_extent_item extent_item;
1383 	int ret;
1384 	int level;
1385 	int err = 0;
1386 
1387 	btrfs_set_stack_extent_refs(&extent_item, 1);
1388 	btrfs_set_key_type(&ins, BTRFS_EXTENT_ITEM_KEY);
1389 	path = btrfs_alloc_path();
1390 
1391 	while(1) {
1392 		ret = find_first_extent_bit(&info->extent_ins, 0, &start,
1393 					    &end, EXTENT_LOCKED);
1394 		if (ret)
1395 			break;
1396 
1397 		ins.objectid = start;
1398 		ins.offset = end + 1 - start;
1399 		err = btrfs_insert_item(trans, extent_root, &ins,
1400 					&extent_item, sizeof(extent_item));
1401 		clear_extent_bits(&info->extent_ins, start, end, EXTENT_LOCKED,
1402 				  GFP_NOFS);
1403 		eb = read_tree_block(extent_root, ins.objectid, ins.offset,
1404 				     trans->transid);
1405 		btrfs_tree_lock(eb);
1406 		level = btrfs_header_level(eb);
1407 		if (level == 0) {
1408 			btrfs_item_key(eb, &first, 0);
1409 		} else {
1410 			btrfs_node_key(eb, &first, 0);
1411 		}
1412 		btrfs_tree_unlock(eb);
1413 		free_extent_buffer(eb);
1414 		/*
1415 		 * the first key is just a hint, so the race we've created
1416 		 * against reading it is fine
1417 		 */
1418 		err = btrfs_insert_extent_backref(trans, extent_root, path,
1419 					  start, extent_root->root_key.objectid,
1420 					  0, level,
1421 					  btrfs_disk_key_objectid(&first));
1422 		BUG_ON(err);
1423 	}
1424 	btrfs_free_path(path);
1425 	return 0;
1426 }
1427 
1428 static int pin_down_bytes(struct btrfs_root *root, u64 bytenr, u32 num_bytes,
1429 			  int pending)
1430 {
1431 	int err = 0;
1432 
1433 	if (!pending) {
1434 #if 0
1435 		struct extent_buffer *buf;
1436 		buf = btrfs_find_tree_block(root, bytenr, num_bytes);
1437 		if (buf) {
1438 			if (!btrfs_try_tree_lock(buf) &&
1439 			    btrfs_buffer_uptodate(buf, 0)) {
1440 				u64 transid =
1441 				    root->fs_info->running_transaction->transid;
1442 				u64 header_transid =
1443 					btrfs_header_generation(buf);
1444 				if (header_transid == transid &&
1445 				    !btrfs_header_flag(buf,
1446 					       BTRFS_HEADER_FLAG_WRITTEN)) {
1447 					clean_tree_block(NULL, root, buf);
1448 					btrfs_tree_unlock(buf);
1449 					free_extent_buffer(buf);
1450 					return 1;
1451 				}
1452 				btrfs_tree_unlock(buf);
1453 			}
1454 			free_extent_buffer(buf);
1455 		}
1456 #endif
1457 		update_pinned_extents(root, bytenr, num_bytes, 1);
1458 	} else {
1459 		set_extent_bits(&root->fs_info->pending_del,
1460 				bytenr, bytenr + num_bytes - 1,
1461 				EXTENT_LOCKED, GFP_NOFS);
1462 	}
1463 	BUG_ON(err < 0);
1464 	return 0;
1465 }
1466 
1467 /*
1468  * remove an extent from the root, returns 0 on success
1469  */
1470 static int __free_extent(struct btrfs_trans_handle *trans, struct btrfs_root
1471 			 *root, u64 bytenr, u64 num_bytes,
1472 			 u64 root_objectid, u64 ref_generation,
1473 			 u64 owner_objectid, u64 owner_offset, int pin,
1474 			 int mark_free)
1475 {
1476 	struct btrfs_path *path;
1477 	struct btrfs_key key;
1478 	struct btrfs_fs_info *info = root->fs_info;
1479 	struct btrfs_root *extent_root = info->extent_root;
1480 	struct extent_buffer *leaf;
1481 	int ret;
1482 	int extent_slot = 0;
1483 	int found_extent = 0;
1484 	int num_to_del = 1;
1485 	struct btrfs_extent_item *ei;
1486 	u32 refs;
1487 
1488 	key.objectid = bytenr;
1489 	btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
1490 	key.offset = num_bytes;
1491 	path = btrfs_alloc_path();
1492 	if (!path)
1493 		return -ENOMEM;
1494 
1495 	path->reada = 1;
1496 	ret = lookup_extent_backref(trans, extent_root, path,
1497 				    bytenr, root_objectid,
1498 				    ref_generation,
1499 				    owner_objectid, owner_offset, 1);
1500 	if (ret == 0) {
1501 		struct btrfs_key found_key;
1502 		extent_slot = path->slots[0];
1503 		while(extent_slot > 0) {
1504 			extent_slot--;
1505 			btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1506 					      extent_slot);
1507 			if (found_key.objectid != bytenr)
1508 				break;
1509 			if (found_key.type == BTRFS_EXTENT_ITEM_KEY &&
1510 			    found_key.offset == num_bytes) {
1511 				found_extent = 1;
1512 				break;
1513 			}
1514 			if (path->slots[0] - extent_slot > 5)
1515 				break;
1516 		}
1517 		if (!found_extent)
1518 			ret = btrfs_del_item(trans, extent_root, path);
1519 	} else {
1520 		btrfs_print_leaf(extent_root, path->nodes[0]);
1521 		WARN_ON(1);
1522 		printk("Unable to find ref byte nr %Lu root %Lu "
1523 		       " gen %Lu owner %Lu offset %Lu\n", bytenr,
1524 		       root_objectid, ref_generation, owner_objectid,
1525 		       owner_offset);
1526 	}
1527 	if (!found_extent) {
1528 		btrfs_release_path(extent_root, path);
1529 		ret = btrfs_search_slot(trans, extent_root, &key, path, -1, 1);
1530 		if (ret < 0)
1531 			return ret;
1532 		BUG_ON(ret);
1533 		extent_slot = path->slots[0];
1534 	}
1535 
1536 	leaf = path->nodes[0];
1537 	ei = btrfs_item_ptr(leaf, extent_slot,
1538 			    struct btrfs_extent_item);
1539 	refs = btrfs_extent_refs(leaf, ei);
1540 	BUG_ON(refs == 0);
1541 	refs -= 1;
1542 	btrfs_set_extent_refs(leaf, ei, refs);
1543 
1544 	btrfs_mark_buffer_dirty(leaf);
1545 
1546 	if (refs == 0 && found_extent && path->slots[0] == extent_slot + 1) {
1547 		/* if the back ref and the extent are next to each other
1548 		 * they get deleted below in one shot
1549 		 */
1550 		path->slots[0] = extent_slot;
1551 		num_to_del = 2;
1552 	} else if (found_extent) {
1553 		/* otherwise delete the extent back ref */
1554 		ret = btrfs_del_item(trans, extent_root, path);
1555 		BUG_ON(ret);
1556 		/* if refs are 0, we need to setup the path for deletion */
1557 		if (refs == 0) {
1558 			btrfs_release_path(extent_root, path);
1559 			ret = btrfs_search_slot(trans, extent_root, &key, path,
1560 						-1, 1);
1561 			if (ret < 0)
1562 				return ret;
1563 			BUG_ON(ret);
1564 		}
1565 	}
1566 
1567 	if (refs == 0) {
1568 		u64 super_used;
1569 		u64 root_used;
1570 
1571 		if (pin) {
1572 			ret = pin_down_bytes(root, bytenr, num_bytes, 0);
1573 			if (ret > 0)
1574 				mark_free = 1;
1575 			BUG_ON(ret < 0);
1576 		}
1577 
1578 		/* block accounting for super block */
1579 		spin_lock_irq(&info->delalloc_lock);
1580 		super_used = btrfs_super_bytes_used(&info->super_copy);
1581 		btrfs_set_super_bytes_used(&info->super_copy,
1582 					   super_used - num_bytes);
1583 		spin_unlock_irq(&info->delalloc_lock);
1584 
1585 		/* block accounting for root item */
1586 		root_used = btrfs_root_used(&root->root_item);
1587 		btrfs_set_root_used(&root->root_item,
1588 					   root_used - num_bytes);
1589 		ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
1590 				      num_to_del);
1591 		if (ret) {
1592 			return ret;
1593 		}
1594 		ret = update_block_group(trans, root, bytenr, num_bytes, 0,
1595 					 mark_free);
1596 		BUG_ON(ret);
1597 	}
1598 	btrfs_free_path(path);
1599 	finish_current_insert(trans, extent_root);
1600 	return ret;
1601 }
1602 
1603 /*
1604  * find all the blocks marked as pending in the radix tree and remove
1605  * them from the extent map
1606  */
1607 static int del_pending_extents(struct btrfs_trans_handle *trans, struct
1608 			       btrfs_root *extent_root)
1609 {
1610 	int ret;
1611 	int err = 0;
1612 	u64 start;
1613 	u64 end;
1614 	struct extent_io_tree *pending_del;
1615 	struct extent_io_tree *pinned_extents;
1616 
1617 	pending_del = &extent_root->fs_info->pending_del;
1618 	pinned_extents = &extent_root->fs_info->pinned_extents;
1619 
1620 	while(1) {
1621 		ret = find_first_extent_bit(pending_del, 0, &start, &end,
1622 					    EXTENT_LOCKED);
1623 		if (ret)
1624 			break;
1625 		update_pinned_extents(extent_root, start, end + 1 - start, 1);
1626 		clear_extent_bits(pending_del, start, end, EXTENT_LOCKED,
1627 				  GFP_NOFS);
1628 		ret = __free_extent(trans, extent_root,
1629 				     start, end + 1 - start,
1630 				     extent_root->root_key.objectid,
1631 				     0, 0, 0, 0, 0);
1632 		if (ret)
1633 			err = ret;
1634 	}
1635 	return err;
1636 }
1637 
1638 /*
1639  * remove an extent from the root, returns 0 on success
1640  */
1641 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
1642 			       struct btrfs_root *root, u64 bytenr,
1643 			       u64 num_bytes, u64 root_objectid,
1644 			       u64 ref_generation, u64 owner_objectid,
1645 			       u64 owner_offset, int pin)
1646 {
1647 	struct btrfs_root *extent_root = root->fs_info->extent_root;
1648 	int pending_ret;
1649 	int ret;
1650 
1651 	WARN_ON(num_bytes < root->sectorsize);
1652 	if (!root->ref_cows)
1653 		ref_generation = 0;
1654 
1655 	if (root == extent_root) {
1656 		pin_down_bytes(root, bytenr, num_bytes, 1);
1657 		return 0;
1658 	}
1659 	ret = __free_extent(trans, root, bytenr, num_bytes, root_objectid,
1660 			    ref_generation, owner_objectid, owner_offset,
1661 			    pin, pin == 0);
1662 	pending_ret = del_pending_extents(trans, root->fs_info->extent_root);
1663 	return ret ? ret : pending_ret;
1664 }
1665 
1666 int btrfs_free_extent(struct btrfs_trans_handle *trans,
1667 		      struct btrfs_root *root, u64 bytenr,
1668 		      u64 num_bytes, u64 root_objectid,
1669 		      u64 ref_generation, u64 owner_objectid,
1670 		      u64 owner_offset, int pin)
1671 {
1672 	int ret;
1673 
1674 	maybe_lock_mutex(root);
1675 	ret = __btrfs_free_extent(trans, root, bytenr, num_bytes,
1676 				  root_objectid, ref_generation,
1677 				  owner_objectid, owner_offset, pin);
1678 	maybe_unlock_mutex(root);
1679 	return ret;
1680 }
1681 
1682 static u64 stripe_align(struct btrfs_root *root, u64 val)
1683 {
1684 	u64 mask = ((u64)root->stripesize - 1);
1685 	u64 ret = (val + mask) & ~mask;
1686 	return ret;
1687 }
1688 
1689 /*
1690  * walks the btree of allocated extents and find a hole of a given size.
1691  * The key ins is changed to record the hole:
1692  * ins->objectid == block start
1693  * ins->flags = BTRFS_EXTENT_ITEM_KEY
1694  * ins->offset == number of blocks
1695  * Any available blocks before search_start are skipped.
1696  */
1697 static int noinline find_free_extent(struct btrfs_trans_handle *trans,
1698 				     struct btrfs_root *orig_root,
1699 				     u64 num_bytes, u64 empty_size,
1700 				     u64 search_start, u64 search_end,
1701 				     u64 hint_byte, struct btrfs_key *ins,
1702 				     u64 exclude_start, u64 exclude_nr,
1703 				     int data)
1704 {
1705 	int ret;
1706 	u64 orig_search_start;
1707 	struct btrfs_root * root = orig_root->fs_info->extent_root;
1708 	struct btrfs_fs_info *info = root->fs_info;
1709 	u64 total_needed = num_bytes;
1710 	u64 *last_ptr = NULL;
1711 	struct btrfs_block_group_cache *block_group;
1712 	int full_scan = 0;
1713 	int wrapped = 0;
1714 	int chunk_alloc_done = 0;
1715 	int empty_cluster = 2 * 1024 * 1024;
1716 	int allowed_chunk_alloc = 0;
1717 
1718 	WARN_ON(num_bytes < root->sectorsize);
1719 	btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
1720 
1721 	if (orig_root->ref_cows || empty_size)
1722 		allowed_chunk_alloc = 1;
1723 
1724 	if (data & BTRFS_BLOCK_GROUP_METADATA) {
1725 		last_ptr = &root->fs_info->last_alloc;
1726 		empty_cluster = 256 * 1024;
1727 	}
1728 
1729 	if ((data & BTRFS_BLOCK_GROUP_DATA) && btrfs_test_opt(root, SSD)) {
1730 		last_ptr = &root->fs_info->last_data_alloc;
1731 	}
1732 
1733 	if (last_ptr) {
1734 		if (*last_ptr)
1735 			hint_byte = *last_ptr;
1736 		else {
1737 			empty_size += empty_cluster;
1738 		}
1739 	}
1740 
1741 	search_start = max(search_start, first_logical_byte(root, 0));
1742 	orig_search_start = search_start;
1743 
1744 	if (search_end == (u64)-1)
1745 		search_end = btrfs_super_total_bytes(&info->super_copy);
1746 
1747 	if (hint_byte) {
1748 		block_group = btrfs_lookup_first_block_group(info, hint_byte);
1749 		if (!block_group)
1750 			hint_byte = search_start;
1751 		block_group = __btrfs_find_block_group(root, block_group,
1752 						     hint_byte, data, 1);
1753 		if (last_ptr && *last_ptr == 0 && block_group)
1754 			hint_byte = block_group->key.objectid;
1755 	} else {
1756 		block_group = __btrfs_find_block_group(root,
1757 						     trans->block_group,
1758 						     search_start, data, 1);
1759 	}
1760 	search_start = max(search_start, hint_byte);
1761 
1762 	total_needed += empty_size;
1763 
1764 check_failed:
1765 	if (!block_group) {
1766 		block_group = btrfs_lookup_first_block_group(info,
1767 							     search_start);
1768 		if (!block_group)
1769 			block_group = btrfs_lookup_first_block_group(info,
1770 						       orig_search_start);
1771 	}
1772 	if (full_scan && !chunk_alloc_done) {
1773 		if (allowed_chunk_alloc) {
1774 			do_chunk_alloc(trans, root,
1775 				     num_bytes + 2 * 1024 * 1024, data, 1);
1776 			allowed_chunk_alloc = 0;
1777 		} else if (block_group && block_group_bits(block_group, data)) {
1778 			block_group->space_info->force_alloc = 1;
1779 		}
1780 		chunk_alloc_done = 1;
1781 	}
1782 	ret = find_search_start(root, &block_group, &search_start,
1783 				total_needed, data);
1784 	if (ret == -ENOSPC && last_ptr && *last_ptr) {
1785 		*last_ptr = 0;
1786 		block_group = btrfs_lookup_first_block_group(info,
1787 							     orig_search_start);
1788 		search_start = orig_search_start;
1789 		ret = find_search_start(root, &block_group, &search_start,
1790 					total_needed, data);
1791 	}
1792 	if (ret == -ENOSPC)
1793 		goto enospc;
1794 	if (ret)
1795 		goto error;
1796 
1797 	if (last_ptr && *last_ptr && search_start != *last_ptr) {
1798 		*last_ptr = 0;
1799 		if (!empty_size) {
1800 			empty_size += empty_cluster;
1801 			total_needed += empty_size;
1802 		}
1803 		block_group = btrfs_lookup_first_block_group(info,
1804 						       orig_search_start);
1805 		search_start = orig_search_start;
1806 		ret = find_search_start(root, &block_group,
1807 					&search_start, total_needed, data);
1808 		if (ret == -ENOSPC)
1809 			goto enospc;
1810 		if (ret)
1811 			goto error;
1812 	}
1813 
1814 	search_start = stripe_align(root, search_start);
1815 	ins->objectid = search_start;
1816 	ins->offset = num_bytes;
1817 
1818 	if (ins->objectid + num_bytes >= search_end)
1819 		goto enospc;
1820 
1821 	if (ins->objectid + num_bytes >
1822 	    block_group->key.objectid + block_group->key.offset) {
1823 		search_start = block_group->key.objectid +
1824 			block_group->key.offset;
1825 		goto new_group;
1826 	}
1827 
1828 	if (test_range_bit(&info->extent_ins, ins->objectid,
1829 			   ins->objectid + num_bytes -1, EXTENT_LOCKED, 0)) {
1830 		search_start = ins->objectid + num_bytes;
1831 		goto new_group;
1832 	}
1833 
1834 	if (test_range_bit(&info->pinned_extents, ins->objectid,
1835 			   ins->objectid + num_bytes -1, EXTENT_DIRTY, 0)) {
1836 		search_start = ins->objectid + num_bytes;
1837 		goto new_group;
1838 	}
1839 
1840 	if (exclude_nr > 0 && (ins->objectid + num_bytes > exclude_start &&
1841 	    ins->objectid < exclude_start + exclude_nr)) {
1842 		search_start = exclude_start + exclude_nr;
1843 		goto new_group;
1844 	}
1845 
1846 	if (!(data & BTRFS_BLOCK_GROUP_DATA)) {
1847 		block_group = btrfs_lookup_block_group(info, ins->objectid);
1848 		if (block_group)
1849 			trans->block_group = block_group;
1850 	}
1851 	ins->offset = num_bytes;
1852 	if (last_ptr) {
1853 		*last_ptr = ins->objectid + ins->offset;
1854 		if (*last_ptr ==
1855 		    btrfs_super_total_bytes(&root->fs_info->super_copy)) {
1856 			*last_ptr = 0;
1857 		}
1858 	}
1859 	return 0;
1860 
1861 new_group:
1862 	if (search_start + num_bytes >= search_end) {
1863 enospc:
1864 		search_start = orig_search_start;
1865 		if (full_scan) {
1866 			ret = -ENOSPC;
1867 			goto error;
1868 		}
1869 		if (wrapped) {
1870 			if (!full_scan)
1871 				total_needed -= empty_size;
1872 			full_scan = 1;
1873 		} else
1874 			wrapped = 1;
1875 	}
1876 	block_group = btrfs_lookup_first_block_group(info, search_start);
1877 	cond_resched();
1878 	block_group = __btrfs_find_block_group(root, block_group,
1879 					     search_start, data, 0);
1880 	goto check_failed;
1881 
1882 error:
1883 	return ret;
1884 }
1885 
1886 /*
1887  * finds a free extent and does all the dirty work required for allocation
1888  * returns the key for the extent through ins, and a tree buffer for
1889  * the first block of the extent through buf.
1890  *
1891  * returns 0 if everything worked, non-zero otherwise.
1892  */
1893 int btrfs_alloc_extent(struct btrfs_trans_handle *trans,
1894 		       struct btrfs_root *root,
1895 		       u64 num_bytes, u64 min_alloc_size,
1896 		       u64 root_objectid, u64 ref_generation,
1897 		       u64 owner, u64 owner_offset,
1898 		       u64 empty_size, u64 hint_byte,
1899 		       u64 search_end, struct btrfs_key *ins, u64 data)
1900 {
1901 	int ret;
1902 	int pending_ret;
1903 	u64 super_used;
1904 	u64 root_used;
1905 	u64 search_start = 0;
1906 	u64 alloc_profile;
1907 	u32 sizes[2];
1908 	struct btrfs_fs_info *info = root->fs_info;
1909 	struct btrfs_root *extent_root = info->extent_root;
1910 	struct btrfs_extent_item *extent_item;
1911 	struct btrfs_extent_ref *ref;
1912 	struct btrfs_path *path;
1913 	struct btrfs_key keys[2];
1914 
1915 	maybe_lock_mutex(root);
1916 
1917 	if (data) {
1918 		alloc_profile = info->avail_data_alloc_bits &
1919 			        info->data_alloc_profile;
1920 		data = BTRFS_BLOCK_GROUP_DATA | alloc_profile;
1921 	} else if (root == root->fs_info->chunk_root) {
1922 		alloc_profile = info->avail_system_alloc_bits &
1923 			        info->system_alloc_profile;
1924 		data = BTRFS_BLOCK_GROUP_SYSTEM | alloc_profile;
1925 	} else {
1926 		alloc_profile = info->avail_metadata_alloc_bits &
1927 			        info->metadata_alloc_profile;
1928 		data = BTRFS_BLOCK_GROUP_METADATA | alloc_profile;
1929 	}
1930 again:
1931 	data = reduce_alloc_profile(root, data);
1932 	/*
1933 	 * the only place that sets empty_size is btrfs_realloc_node, which
1934 	 * is not called recursively on allocations
1935 	 */
1936 	if (empty_size || root->ref_cows) {
1937 		if (!(data & BTRFS_BLOCK_GROUP_METADATA)) {
1938 			ret = do_chunk_alloc(trans, root->fs_info->extent_root,
1939 				     2 * 1024 * 1024,
1940 				     BTRFS_BLOCK_GROUP_METADATA |
1941 				     (info->metadata_alloc_profile &
1942 				      info->avail_metadata_alloc_bits), 0);
1943 			BUG_ON(ret);
1944 		}
1945 		ret = do_chunk_alloc(trans, root->fs_info->extent_root,
1946 				     num_bytes + 2 * 1024 * 1024, data, 0);
1947 		BUG_ON(ret);
1948 	}
1949 
1950 	WARN_ON(num_bytes < root->sectorsize);
1951 	ret = find_free_extent(trans, root, num_bytes, empty_size,
1952 			       search_start, search_end, hint_byte, ins,
1953 			       trans->alloc_exclude_start,
1954 			       trans->alloc_exclude_nr, data);
1955 
1956 	if (ret == -ENOSPC && num_bytes > min_alloc_size) {
1957 		num_bytes = num_bytes >> 1;
1958 		num_bytes = max(num_bytes, min_alloc_size);
1959 		do_chunk_alloc(trans, root->fs_info->extent_root,
1960 			       num_bytes, data, 1);
1961 		goto again;
1962 	}
1963 	if (ret) {
1964 		printk("allocation failed flags %Lu\n", data);
1965 	}
1966 	if (ret) {
1967 		BUG();
1968 		goto out;
1969 	}
1970 
1971 	/* block accounting for super block */
1972 	spin_lock_irq(&info->delalloc_lock);
1973 	super_used = btrfs_super_bytes_used(&info->super_copy);
1974 	btrfs_set_super_bytes_used(&info->super_copy, super_used + num_bytes);
1975 	spin_unlock_irq(&info->delalloc_lock);
1976 
1977 	/* block accounting for root item */
1978 	root_used = btrfs_root_used(&root->root_item);
1979 	btrfs_set_root_used(&root->root_item, root_used + num_bytes);
1980 
1981 	clear_extent_dirty(&root->fs_info->free_space_cache,
1982 			   ins->objectid, ins->objectid + ins->offset - 1,
1983 			   GFP_NOFS);
1984 
1985 	if (root == extent_root) {
1986 		set_extent_bits(&root->fs_info->extent_ins, ins->objectid,
1987 				ins->objectid + ins->offset - 1,
1988 				EXTENT_LOCKED, GFP_NOFS);
1989 		goto update_block;
1990 	}
1991 
1992 	WARN_ON(trans->alloc_exclude_nr);
1993 	trans->alloc_exclude_start = ins->objectid;
1994 	trans->alloc_exclude_nr = ins->offset;
1995 
1996 	memcpy(&keys[0], ins, sizeof(*ins));
1997 	keys[1].offset = hash_extent_ref(root_objectid, ref_generation,
1998 					 owner, owner_offset);
1999 	keys[1].objectid = ins->objectid;
2000 	keys[1].type = BTRFS_EXTENT_REF_KEY;
2001 	sizes[0] = sizeof(*extent_item);
2002 	sizes[1] = sizeof(*ref);
2003 
2004 	path = btrfs_alloc_path();
2005 	BUG_ON(!path);
2006 
2007 	ret = btrfs_insert_empty_items(trans, extent_root, path, keys,
2008 				       sizes, 2);
2009 
2010 	BUG_ON(ret);
2011 	extent_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
2012 				     struct btrfs_extent_item);
2013 	btrfs_set_extent_refs(path->nodes[0], extent_item, 1);
2014 	ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
2015 			     struct btrfs_extent_ref);
2016 
2017 	btrfs_set_ref_root(path->nodes[0], ref, root_objectid);
2018 	btrfs_set_ref_generation(path->nodes[0], ref, ref_generation);
2019 	btrfs_set_ref_objectid(path->nodes[0], ref, owner);
2020 	btrfs_set_ref_offset(path->nodes[0], ref, owner_offset);
2021 
2022 	btrfs_mark_buffer_dirty(path->nodes[0]);
2023 
2024 	trans->alloc_exclude_start = 0;
2025 	trans->alloc_exclude_nr = 0;
2026 	btrfs_free_path(path);
2027 	finish_current_insert(trans, extent_root);
2028 	pending_ret = del_pending_extents(trans, extent_root);
2029 
2030 	if (ret)
2031 		goto out;
2032 	if (pending_ret) {
2033 		ret = pending_ret;
2034 		goto out;
2035 	}
2036 
2037 update_block:
2038 	ret = update_block_group(trans, root, ins->objectid, ins->offset, 1, 0);
2039 	if (ret) {
2040 		printk("update block group failed for %Lu %Lu\n",
2041 		       ins->objectid, ins->offset);
2042 		BUG();
2043 	}
2044 out:
2045 	maybe_unlock_mutex(root);
2046 	return ret;
2047 }
2048 /*
2049  * helper function to allocate a block for a given tree
2050  * returns the tree buffer or NULL.
2051  */
2052 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
2053 					     struct btrfs_root *root,
2054 					     u32 blocksize,
2055 					     u64 root_objectid,
2056 					     u64 ref_generation,
2057 					     u64 first_objectid,
2058 					     int level,
2059 					     u64 hint,
2060 					     u64 empty_size)
2061 {
2062 	struct btrfs_key ins;
2063 	int ret;
2064 	struct extent_buffer *buf;
2065 
2066 	ret = btrfs_alloc_extent(trans, root, blocksize, blocksize,
2067 				 root_objectid, ref_generation,
2068 				 level, first_objectid, empty_size, hint,
2069 				 (u64)-1, &ins, 0);
2070 	if (ret) {
2071 		BUG_ON(ret > 0);
2072 		return ERR_PTR(ret);
2073 	}
2074 	buf = btrfs_find_create_tree_block(root, ins.objectid, blocksize);
2075 	if (!buf) {
2076 		btrfs_free_extent(trans, root, ins.objectid, blocksize,
2077 				  root->root_key.objectid, ref_generation,
2078 				  0, 0, 0);
2079 		return ERR_PTR(-ENOMEM);
2080 	}
2081 	btrfs_set_header_generation(buf, trans->transid);
2082 	btrfs_tree_lock(buf);
2083 	clean_tree_block(trans, root, buf);
2084 	btrfs_set_buffer_uptodate(buf);
2085 
2086 	if (PageDirty(buf->first_page)) {
2087 		printk("page %lu dirty\n", buf->first_page->index);
2088 		WARN_ON(1);
2089 	}
2090 
2091 	set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
2092 			 buf->start + buf->len - 1, GFP_NOFS);
2093 	if (!btrfs_test_opt(root, SSD))
2094 		btrfs_set_buffer_defrag(buf);
2095 	trans->blocks_used++;
2096 	return buf;
2097 }
2098 
2099 static int noinline drop_leaf_ref(struct btrfs_trans_handle *trans,
2100 				  struct btrfs_root *root,
2101 				  struct extent_buffer *leaf)
2102 {
2103 	u64 leaf_owner;
2104 	u64 leaf_generation;
2105 	struct btrfs_key key;
2106 	struct btrfs_file_extent_item *fi;
2107 	int i;
2108 	int nritems;
2109 	int ret;
2110 
2111 	BUG_ON(!btrfs_is_leaf(leaf));
2112 	nritems = btrfs_header_nritems(leaf);
2113 	leaf_owner = btrfs_header_owner(leaf);
2114 	leaf_generation = btrfs_header_generation(leaf);
2115 
2116 	for (i = 0; i < nritems; i++) {
2117 		u64 disk_bytenr;
2118 
2119 		btrfs_item_key_to_cpu(leaf, &key, i);
2120 		if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2121 			continue;
2122 		fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
2123 		if (btrfs_file_extent_type(leaf, fi) ==
2124 		    BTRFS_FILE_EXTENT_INLINE)
2125 			continue;
2126 		/*
2127 		 * FIXME make sure to insert a trans record that
2128 		 * repeats the snapshot del on crash
2129 		 */
2130 		disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
2131 		if (disk_bytenr == 0)
2132 			continue;
2133 		ret = __btrfs_free_extent(trans, root, disk_bytenr,
2134 				btrfs_file_extent_disk_num_bytes(leaf, fi),
2135 				leaf_owner, leaf_generation,
2136 				key.objectid, key.offset, 0);
2137 		BUG_ON(ret);
2138 	}
2139 	return 0;
2140 }
2141 
2142 static void noinline reada_walk_down(struct btrfs_root *root,
2143 				     struct extent_buffer *node,
2144 				     int slot)
2145 {
2146 	u64 bytenr;
2147 	u64 last = 0;
2148 	u32 nritems;
2149 	u32 refs;
2150 	u32 blocksize;
2151 	int ret;
2152 	int i;
2153 	int level;
2154 	int skipped = 0;
2155 
2156 	nritems = btrfs_header_nritems(node);
2157 	level = btrfs_header_level(node);
2158 	if (level)
2159 		return;
2160 
2161 	for (i = slot; i < nritems && skipped < 32; i++) {
2162 		bytenr = btrfs_node_blockptr(node, i);
2163 		if (last && ((bytenr > last && bytenr - last > 32 * 1024) ||
2164 			     (last > bytenr && last - bytenr > 32 * 1024))) {
2165 			skipped++;
2166 			continue;
2167 		}
2168 		blocksize = btrfs_level_size(root, level - 1);
2169 		if (i != slot) {
2170 			ret = lookup_extent_ref(NULL, root, bytenr,
2171 						blocksize, &refs);
2172 			BUG_ON(ret);
2173 			if (refs != 1) {
2174 				skipped++;
2175 				continue;
2176 			}
2177 		}
2178 		mutex_unlock(&root->fs_info->alloc_mutex);
2179 		ret = readahead_tree_block(root, bytenr, blocksize,
2180 					   btrfs_node_ptr_generation(node, i));
2181 		last = bytenr + blocksize;
2182 		cond_resched();
2183 		mutex_lock(&root->fs_info->alloc_mutex);
2184 		if (ret)
2185 			break;
2186 	}
2187 }
2188 
2189 /*
2190  * helper function for drop_snapshot, this walks down the tree dropping ref
2191  * counts as it goes.
2192  */
2193 static int noinline walk_down_tree(struct btrfs_trans_handle *trans,
2194 				   struct btrfs_root *root,
2195 				   struct btrfs_path *path, int *level)
2196 {
2197 	u64 root_owner;
2198 	u64 root_gen;
2199 	u64 bytenr;
2200 	u64 ptr_gen;
2201 	struct extent_buffer *next;
2202 	struct extent_buffer *cur;
2203 	struct extent_buffer *parent;
2204 	u32 blocksize;
2205 	int ret;
2206 	u32 refs;
2207 
2208 	mutex_lock(&root->fs_info->alloc_mutex);
2209 
2210 	WARN_ON(*level < 0);
2211 	WARN_ON(*level >= BTRFS_MAX_LEVEL);
2212 	ret = lookup_extent_ref(trans, root,
2213 				path->nodes[*level]->start,
2214 				path->nodes[*level]->len, &refs);
2215 	BUG_ON(ret);
2216 	if (refs > 1)
2217 		goto out;
2218 
2219 	/*
2220 	 * walk down to the last node level and free all the leaves
2221 	 */
2222 	while(*level >= 0) {
2223 		WARN_ON(*level < 0);
2224 		WARN_ON(*level >= BTRFS_MAX_LEVEL);
2225 		cur = path->nodes[*level];
2226 
2227 		if (btrfs_header_level(cur) != *level)
2228 			WARN_ON(1);
2229 
2230 		if (path->slots[*level] >=
2231 		    btrfs_header_nritems(cur))
2232 			break;
2233 		if (*level == 0) {
2234 			ret = drop_leaf_ref(trans, root, cur);
2235 			BUG_ON(ret);
2236 			break;
2237 		}
2238 		bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
2239 		ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
2240 		blocksize = btrfs_level_size(root, *level - 1);
2241 
2242 		ret = lookup_extent_ref(trans, root, bytenr, blocksize, &refs);
2243 		BUG_ON(ret);
2244 		if (refs != 1) {
2245 			parent = path->nodes[*level];
2246 			root_owner = btrfs_header_owner(parent);
2247 			root_gen = btrfs_header_generation(parent);
2248 			path->slots[*level]++;
2249 			ret = __btrfs_free_extent(trans, root, bytenr,
2250 						blocksize, root_owner,
2251 						root_gen, 0, 0, 1);
2252 			BUG_ON(ret);
2253 			continue;
2254 		}
2255 		next = btrfs_find_tree_block(root, bytenr, blocksize);
2256 		if (!next || !btrfs_buffer_uptodate(next, ptr_gen)) {
2257 			free_extent_buffer(next);
2258 			reada_walk_down(root, cur, path->slots[*level]);
2259 
2260 			mutex_unlock(&root->fs_info->alloc_mutex);
2261 			next = read_tree_block(root, bytenr, blocksize,
2262 					       ptr_gen);
2263 			mutex_lock(&root->fs_info->alloc_mutex);
2264 
2265 			/* we've dropped the lock, double check */
2266 			ret = lookup_extent_ref(trans, root, bytenr,
2267 						blocksize, &refs);
2268 			BUG_ON(ret);
2269 			if (refs != 1) {
2270 				parent = path->nodes[*level];
2271 				root_owner = btrfs_header_owner(parent);
2272 				root_gen = btrfs_header_generation(parent);
2273 
2274 				path->slots[*level]++;
2275 				free_extent_buffer(next);
2276 				ret = __btrfs_free_extent(trans, root, bytenr,
2277 							blocksize,
2278 							root_owner,
2279 							root_gen, 0, 0, 1);
2280 				BUG_ON(ret);
2281 				continue;
2282 			}
2283 		}
2284 		WARN_ON(*level <= 0);
2285 		if (path->nodes[*level-1])
2286 			free_extent_buffer(path->nodes[*level-1]);
2287 		path->nodes[*level-1] = next;
2288 		*level = btrfs_header_level(next);
2289 		path->slots[*level] = 0;
2290 	}
2291 out:
2292 	WARN_ON(*level < 0);
2293 	WARN_ON(*level >= BTRFS_MAX_LEVEL);
2294 
2295 	if (path->nodes[*level] == root->node) {
2296 		root_owner = root->root_key.objectid;
2297 		parent = path->nodes[*level];
2298 	} else {
2299 		parent = path->nodes[*level + 1];
2300 		root_owner = btrfs_header_owner(parent);
2301 	}
2302 
2303 	root_gen = btrfs_header_generation(parent);
2304 	ret = __btrfs_free_extent(trans, root, path->nodes[*level]->start,
2305 				path->nodes[*level]->len,
2306 				root_owner, root_gen, 0, 0, 1);
2307 	free_extent_buffer(path->nodes[*level]);
2308 	path->nodes[*level] = NULL;
2309 	*level += 1;
2310 	BUG_ON(ret);
2311 	mutex_unlock(&root->fs_info->alloc_mutex);
2312 	return 0;
2313 }
2314 
2315 /*
2316  * helper for dropping snapshots.  This walks back up the tree in the path
2317  * to find the first node higher up where we haven't yet gone through
2318  * all the slots
2319  */
2320 static int noinline walk_up_tree(struct btrfs_trans_handle *trans,
2321 				 struct btrfs_root *root,
2322 				 struct btrfs_path *path, int *level)
2323 {
2324 	u64 root_owner;
2325 	u64 root_gen;
2326 	struct btrfs_root_item *root_item = &root->root_item;
2327 	int i;
2328 	int slot;
2329 	int ret;
2330 
2331 	for(i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) {
2332 		slot = path->slots[i];
2333 		if (slot < btrfs_header_nritems(path->nodes[i]) - 1) {
2334 			struct extent_buffer *node;
2335 			struct btrfs_disk_key disk_key;
2336 			node = path->nodes[i];
2337 			path->slots[i]++;
2338 			*level = i;
2339 			WARN_ON(*level == 0);
2340 			btrfs_node_key(node, &disk_key, path->slots[i]);
2341 			memcpy(&root_item->drop_progress,
2342 			       &disk_key, sizeof(disk_key));
2343 			root_item->drop_level = i;
2344 			return 0;
2345 		} else {
2346 			if (path->nodes[*level] == root->node) {
2347 				root_owner = root->root_key.objectid;
2348 				root_gen =
2349 				   btrfs_header_generation(path->nodes[*level]);
2350 			} else {
2351 				struct extent_buffer *node;
2352 				node = path->nodes[*level + 1];
2353 				root_owner = btrfs_header_owner(node);
2354 				root_gen = btrfs_header_generation(node);
2355 			}
2356 			ret = btrfs_free_extent(trans, root,
2357 						path->nodes[*level]->start,
2358 						path->nodes[*level]->len,
2359 						root_owner, root_gen, 0, 0, 1);
2360 			BUG_ON(ret);
2361 			free_extent_buffer(path->nodes[*level]);
2362 			path->nodes[*level] = NULL;
2363 			*level = i + 1;
2364 		}
2365 	}
2366 	return 1;
2367 }
2368 
2369 /*
2370  * drop the reference count on the tree rooted at 'snap'.  This traverses
2371  * the tree freeing any blocks that have a ref count of zero after being
2372  * decremented.
2373  */
2374 int btrfs_drop_snapshot(struct btrfs_trans_handle *trans, struct btrfs_root
2375 			*root)
2376 {
2377 	int ret = 0;
2378 	int wret;
2379 	int level;
2380 	struct btrfs_path *path;
2381 	int i;
2382 	int orig_level;
2383 	struct btrfs_root_item *root_item = &root->root_item;
2384 
2385 	WARN_ON(!mutex_is_locked(&root->fs_info->drop_mutex));
2386 	path = btrfs_alloc_path();
2387 	BUG_ON(!path);
2388 
2389 	level = btrfs_header_level(root->node);
2390 	orig_level = level;
2391 	if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
2392 		path->nodes[level] = root->node;
2393 		extent_buffer_get(root->node);
2394 		path->slots[level] = 0;
2395 	} else {
2396 		struct btrfs_key key;
2397 		struct btrfs_disk_key found_key;
2398 		struct extent_buffer *node;
2399 
2400 		btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
2401 		level = root_item->drop_level;
2402 		path->lowest_level = level;
2403 		wret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2404 		if (wret < 0) {
2405 			ret = wret;
2406 			goto out;
2407 		}
2408 		node = path->nodes[level];
2409 		btrfs_node_key(node, &found_key, path->slots[level]);
2410 		WARN_ON(memcmp(&found_key, &root_item->drop_progress,
2411 			       sizeof(found_key)));
2412 		for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
2413 			if (path->nodes[i] && path->locks[i]) {
2414 				path->locks[i] = 0;
2415 				btrfs_tree_unlock(path->nodes[i]);
2416 			}
2417 		}
2418 	}
2419 	while(1) {
2420 		wret = walk_down_tree(trans, root, path, &level);
2421 		if (wret > 0)
2422 			break;
2423 		if (wret < 0)
2424 			ret = wret;
2425 
2426 		wret = walk_up_tree(trans, root, path, &level);
2427 		if (wret > 0)
2428 			break;
2429 		if (wret < 0)
2430 			ret = wret;
2431 		ret = -EAGAIN;
2432 		break;
2433 	}
2434 	for (i = 0; i <= orig_level; i++) {
2435 		if (path->nodes[i]) {
2436 			free_extent_buffer(path->nodes[i]);
2437 			path->nodes[i] = NULL;
2438 		}
2439 	}
2440 out:
2441 	btrfs_free_path(path);
2442 	return ret;
2443 }
2444 
2445 int btrfs_free_block_groups(struct btrfs_fs_info *info)
2446 {
2447 	u64 start;
2448 	u64 end;
2449 	u64 ptr;
2450 	int ret;
2451 
2452 	mutex_lock(&info->alloc_mutex);
2453 	while(1) {
2454 		ret = find_first_extent_bit(&info->block_group_cache, 0,
2455 					    &start, &end, (unsigned int)-1);
2456 		if (ret)
2457 			break;
2458 		ret = get_state_private(&info->block_group_cache, start, &ptr);
2459 		if (!ret)
2460 			kfree((void *)(unsigned long)ptr);
2461 		clear_extent_bits(&info->block_group_cache, start,
2462 				  end, (unsigned int)-1, GFP_NOFS);
2463 	}
2464 	while(1) {
2465 		ret = find_first_extent_bit(&info->free_space_cache, 0,
2466 					    &start, &end, EXTENT_DIRTY);
2467 		if (ret)
2468 			break;
2469 		clear_extent_dirty(&info->free_space_cache, start,
2470 				   end, GFP_NOFS);
2471 	}
2472 	mutex_unlock(&info->alloc_mutex);
2473 	return 0;
2474 }
2475 
2476 static unsigned long calc_ra(unsigned long start, unsigned long last,
2477 			     unsigned long nr)
2478 {
2479 	return min(last, start + nr - 1);
2480 }
2481 
2482 static int noinline relocate_inode_pages(struct inode *inode, u64 start,
2483 					 u64 len)
2484 {
2485 	u64 page_start;
2486 	u64 page_end;
2487 	unsigned long last_index;
2488 	unsigned long i;
2489 	struct page *page;
2490 	struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2491 	struct file_ra_state *ra;
2492 	unsigned long total_read = 0;
2493 	unsigned long ra_pages;
2494 	struct btrfs_trans_handle *trans;
2495 
2496 	ra = kzalloc(sizeof(*ra), GFP_NOFS);
2497 
2498 	mutex_lock(&inode->i_mutex);
2499 	i = start >> PAGE_CACHE_SHIFT;
2500 	last_index = (start + len - 1) >> PAGE_CACHE_SHIFT;
2501 
2502 	ra_pages = BTRFS_I(inode)->root->fs_info->bdi.ra_pages;
2503 
2504 	file_ra_state_init(ra, inode->i_mapping);
2505 
2506 	for (; i <= last_index; i++) {
2507 		if (total_read % ra_pages == 0) {
2508 			btrfs_force_ra(inode->i_mapping, ra, NULL, i,
2509 				       calc_ra(i, last_index, ra_pages));
2510 		}
2511 		total_read++;
2512 		if (((u64)i << PAGE_CACHE_SHIFT) > inode->i_size)
2513 			goto truncate_racing;
2514 
2515 		page = grab_cache_page(inode->i_mapping, i);
2516 		if (!page) {
2517 			goto out_unlock;
2518 		}
2519 		if (!PageUptodate(page)) {
2520 			btrfs_readpage(NULL, page);
2521 			lock_page(page);
2522 			if (!PageUptodate(page)) {
2523 				unlock_page(page);
2524 				page_cache_release(page);
2525 				goto out_unlock;
2526 			}
2527 		}
2528 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
2529 		ClearPageDirty(page);
2530 #else
2531 		cancel_dirty_page(page, PAGE_CACHE_SIZE);
2532 #endif
2533 		wait_on_page_writeback(page);
2534 		set_page_extent_mapped(page);
2535 		page_start = (u64)page->index << PAGE_CACHE_SHIFT;
2536 		page_end = page_start + PAGE_CACHE_SIZE - 1;
2537 
2538 		lock_extent(io_tree, page_start, page_end, GFP_NOFS);
2539 
2540 		set_extent_delalloc(io_tree, page_start,
2541 				    page_end, GFP_NOFS);
2542 		set_page_dirty(page);
2543 
2544 		unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
2545 		unlock_page(page);
2546 		page_cache_release(page);
2547 	}
2548 	balance_dirty_pages_ratelimited_nr(inode->i_mapping,
2549 					   total_read);
2550 
2551 out_unlock:
2552 	kfree(ra);
2553 	trans = btrfs_start_transaction(BTRFS_I(inode)->root, 1);
2554 	if (trans) {
2555 		btrfs_add_ordered_inode(inode);
2556 		btrfs_end_transaction(trans, BTRFS_I(inode)->root);
2557 		mark_inode_dirty(inode);
2558 	}
2559 	mutex_unlock(&inode->i_mutex);
2560 	return 0;
2561 
2562 truncate_racing:
2563 	vmtruncate(inode, inode->i_size);
2564 	balance_dirty_pages_ratelimited_nr(inode->i_mapping,
2565 					   total_read);
2566 	goto out_unlock;
2567 }
2568 
2569 /*
2570  * The back references tell us which tree holds a ref on a block,
2571  * but it is possible for the tree root field in the reference to
2572  * reflect the original root before a snapshot was made.  In this
2573  * case we should search through all the children of a given root
2574  * to find potential holders of references on a block.
2575  *
2576  * Instead, we do something a little less fancy and just search
2577  * all the roots for a given key/block combination.
2578  */
2579 static int find_root_for_ref(struct btrfs_root *root,
2580 			     struct btrfs_path *path,
2581 			     struct btrfs_key *key0,
2582 			     int level,
2583 			     int file_key,
2584 			     struct btrfs_root **found_root,
2585 			     u64 bytenr)
2586 {
2587 	struct btrfs_key root_location;
2588 	struct btrfs_root *cur_root = *found_root;
2589 	struct btrfs_file_extent_item *file_extent;
2590 	u64 root_search_start = BTRFS_FS_TREE_OBJECTID;
2591 	u64 found_bytenr;
2592 	int ret;
2593 	int i;
2594 
2595 	root_location.offset = (u64)-1;
2596 	root_location.type = BTRFS_ROOT_ITEM_KEY;
2597 	path->lowest_level = level;
2598 	path->reada = 0;
2599 	while(1) {
2600 		ret = btrfs_search_slot(NULL, cur_root, key0, path, 0, 0);
2601 		found_bytenr = 0;
2602 		if (ret == 0 && file_key) {
2603 			struct extent_buffer *leaf = path->nodes[0];
2604 			file_extent = btrfs_item_ptr(leaf, path->slots[0],
2605 					     struct btrfs_file_extent_item);
2606 			if (btrfs_file_extent_type(leaf, file_extent) ==
2607 			    BTRFS_FILE_EXTENT_REG) {
2608 				found_bytenr =
2609 					btrfs_file_extent_disk_bytenr(leaf,
2610 							       file_extent);
2611 		       }
2612 		} else if (!file_key) {
2613 			if (path->nodes[level])
2614 				found_bytenr = path->nodes[level]->start;
2615 		}
2616 
2617 		for (i = level; i < BTRFS_MAX_LEVEL; i++) {
2618 			if (!path->nodes[i])
2619 				break;
2620 			free_extent_buffer(path->nodes[i]);
2621 			path->nodes[i] = NULL;
2622 		}
2623 		btrfs_release_path(cur_root, path);
2624 
2625 		if (found_bytenr == bytenr) {
2626 			*found_root = cur_root;
2627 			ret = 0;
2628 			goto out;
2629 		}
2630 		ret = btrfs_search_root(root->fs_info->tree_root,
2631 					root_search_start, &root_search_start);
2632 		if (ret)
2633 			break;
2634 
2635 		root_location.objectid = root_search_start;
2636 		cur_root = btrfs_read_fs_root_no_name(root->fs_info,
2637 						      &root_location);
2638 		if (!cur_root) {
2639 			ret = 1;
2640 			break;
2641 		}
2642 	}
2643 out:
2644 	path->lowest_level = 0;
2645 	return ret;
2646 }
2647 
2648 /*
2649  * note, this releases the path
2650  */
2651 static int noinline relocate_one_reference(struct btrfs_root *extent_root,
2652 				  struct btrfs_path *path,
2653 				  struct btrfs_key *extent_key,
2654 				  u64 *last_file_objectid,
2655 				  u64 *last_file_offset,
2656 				  u64 *last_file_root,
2657 				  u64 last_extent)
2658 {
2659 	struct inode *inode;
2660 	struct btrfs_root *found_root;
2661 	struct btrfs_key root_location;
2662 	struct btrfs_key found_key;
2663 	struct btrfs_extent_ref *ref;
2664 	u64 ref_root;
2665 	u64 ref_gen;
2666 	u64 ref_objectid;
2667 	u64 ref_offset;
2668 	int ret;
2669 	int level;
2670 
2671 	ref = btrfs_item_ptr(path->nodes[0], path->slots[0],
2672 			     struct btrfs_extent_ref);
2673 	ref_root = btrfs_ref_root(path->nodes[0], ref);
2674 	ref_gen = btrfs_ref_generation(path->nodes[0], ref);
2675 	ref_objectid = btrfs_ref_objectid(path->nodes[0], ref);
2676 	ref_offset = btrfs_ref_offset(path->nodes[0], ref);
2677 	btrfs_release_path(extent_root, path);
2678 
2679 	root_location.objectid = ref_root;
2680 	if (ref_gen == 0)
2681 		root_location.offset = 0;
2682 	else
2683 		root_location.offset = (u64)-1;
2684 	root_location.type = BTRFS_ROOT_ITEM_KEY;
2685 
2686 	found_root = btrfs_read_fs_root_no_name(extent_root->fs_info,
2687 						&root_location);
2688 	BUG_ON(!found_root);
2689 
2690 	if (ref_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
2691 		found_key.objectid = ref_objectid;
2692 		found_key.type = BTRFS_EXTENT_DATA_KEY;
2693 		found_key.offset = ref_offset;
2694 		level = 0;
2695 
2696 		if (last_extent == extent_key->objectid &&
2697 		    *last_file_objectid == ref_objectid &&
2698 		    *last_file_offset == ref_offset &&
2699 		    *last_file_root == ref_root)
2700 			goto out;
2701 
2702 		ret = find_root_for_ref(extent_root, path, &found_key,
2703 					level, 1, &found_root,
2704 					extent_key->objectid);
2705 
2706 		if (ret)
2707 			goto out;
2708 
2709 		if (last_extent == extent_key->objectid &&
2710 		    *last_file_objectid == ref_objectid &&
2711 		    *last_file_offset == ref_offset &&
2712 		    *last_file_root == ref_root)
2713 			goto out;
2714 
2715 		inode = btrfs_iget_locked(extent_root->fs_info->sb,
2716 					  ref_objectid, found_root);
2717 		if (inode->i_state & I_NEW) {
2718 			/* the inode and parent dir are two different roots */
2719 			BTRFS_I(inode)->root = found_root;
2720 			BTRFS_I(inode)->location.objectid = ref_objectid;
2721 			BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
2722 			BTRFS_I(inode)->location.offset = 0;
2723 			btrfs_read_locked_inode(inode);
2724 			unlock_new_inode(inode);
2725 
2726 		}
2727 		/* this can happen if the reference is not against
2728 		 * the latest version of the tree root
2729 		 */
2730 		if (is_bad_inode(inode)) {
2731 			goto out;
2732 		}
2733 		*last_file_objectid = inode->i_ino;
2734 		*last_file_root = found_root->root_key.objectid;
2735 		*last_file_offset = ref_offset;
2736 
2737 		relocate_inode_pages(inode, ref_offset, extent_key->offset);
2738 		iput(inode);
2739 	} else {
2740 		struct btrfs_trans_handle *trans;
2741 		struct extent_buffer *eb;
2742 		int i;
2743 
2744 		eb = read_tree_block(found_root, extent_key->objectid,
2745 				     extent_key->offset, 0);
2746 		btrfs_tree_lock(eb);
2747 		level = btrfs_header_level(eb);
2748 
2749 		if (level == 0)
2750 			btrfs_item_key_to_cpu(eb, &found_key, 0);
2751 		else
2752 			btrfs_node_key_to_cpu(eb, &found_key, 0);
2753 
2754 		btrfs_tree_unlock(eb);
2755 		free_extent_buffer(eb);
2756 
2757 		ret = find_root_for_ref(extent_root, path, &found_key,
2758 					level, 0, &found_root,
2759 					extent_key->objectid);
2760 
2761 		if (ret)
2762 			goto out;
2763 
2764 		trans = btrfs_start_transaction(found_root, 1);
2765 
2766 		path->lowest_level = level;
2767 		path->reada = 2;
2768 		ret = btrfs_search_slot(trans, found_root, &found_key, path,
2769 					0, 1);
2770 		path->lowest_level = 0;
2771 		for (i = level; i < BTRFS_MAX_LEVEL; i++) {
2772 			if (!path->nodes[i])
2773 				break;
2774 			free_extent_buffer(path->nodes[i]);
2775 			path->nodes[i] = NULL;
2776 		}
2777 		btrfs_release_path(found_root, path);
2778 		if (found_root == found_root->fs_info->extent_root)
2779 			btrfs_extent_post_op(trans, found_root);
2780 		btrfs_end_transaction(trans, found_root);
2781 	}
2782 
2783 out:
2784 	return 0;
2785 }
2786 
2787 static int noinline del_extent_zero(struct btrfs_root *extent_root,
2788 				    struct btrfs_path *path,
2789 				    struct btrfs_key *extent_key)
2790 {
2791 	int ret;
2792 	struct btrfs_trans_handle *trans;
2793 
2794 	trans = btrfs_start_transaction(extent_root, 1);
2795 	ret = btrfs_search_slot(trans, extent_root, extent_key, path, -1, 1);
2796 	if (ret > 0) {
2797 		ret = -EIO;
2798 		goto out;
2799 	}
2800 	if (ret < 0)
2801 		goto out;
2802 	ret = btrfs_del_item(trans, extent_root, path);
2803 out:
2804 	btrfs_end_transaction(trans, extent_root);
2805 	return ret;
2806 }
2807 
2808 static int noinline relocate_one_extent(struct btrfs_root *extent_root,
2809 					struct btrfs_path *path,
2810 					struct btrfs_key *extent_key)
2811 {
2812 	struct btrfs_key key;
2813 	struct btrfs_key found_key;
2814 	struct extent_buffer *leaf;
2815 	u64 last_file_objectid = 0;
2816 	u64 last_file_root = 0;
2817 	u64 last_file_offset = (u64)-1;
2818 	u64 last_extent = 0;
2819 	u32 nritems;
2820 	u32 item_size;
2821 	int ret = 0;
2822 
2823 	if (extent_key->objectid == 0) {
2824 		ret = del_extent_zero(extent_root, path, extent_key);
2825 		goto out;
2826 	}
2827 	key.objectid = extent_key->objectid;
2828 	key.type = BTRFS_EXTENT_REF_KEY;
2829 	key.offset = 0;
2830 
2831 	while(1) {
2832 		ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2833 
2834 		if (ret < 0)
2835 			goto out;
2836 
2837 		ret = 0;
2838 		leaf = path->nodes[0];
2839 		nritems = btrfs_header_nritems(leaf);
2840 		if (path->slots[0] == nritems) {
2841 			ret = btrfs_next_leaf(extent_root, path);
2842 			if (ret > 0) {
2843 				ret = 0;
2844 				goto out;
2845 			}
2846 			if (ret < 0)
2847 				goto out;
2848 			leaf = path->nodes[0];
2849 		}
2850 
2851 		btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2852 		if (found_key.objectid != extent_key->objectid) {
2853 			break;
2854 		}
2855 
2856 		if (found_key.type != BTRFS_EXTENT_REF_KEY) {
2857 			break;
2858 		}
2859 
2860 		key.offset = found_key.offset + 1;
2861 		item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2862 
2863 		ret = relocate_one_reference(extent_root, path, extent_key,
2864 					     &last_file_objectid,
2865 					     &last_file_offset,
2866 					     &last_file_root, last_extent);
2867 		if (ret)
2868 			goto out;
2869 		last_extent = extent_key->objectid;
2870 	}
2871 	ret = 0;
2872 out:
2873 	btrfs_release_path(extent_root, path);
2874 	return ret;
2875 }
2876 
2877 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
2878 {
2879 	u64 num_devices;
2880 	u64 stripped = BTRFS_BLOCK_GROUP_RAID0 |
2881 		BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
2882 
2883 	num_devices = root->fs_info->fs_devices->num_devices;
2884 	if (num_devices == 1) {
2885 		stripped |= BTRFS_BLOCK_GROUP_DUP;
2886 		stripped = flags & ~stripped;
2887 
2888 		/* turn raid0 into single device chunks */
2889 		if (flags & BTRFS_BLOCK_GROUP_RAID0)
2890 			return stripped;
2891 
2892 		/* turn mirroring into duplication */
2893 		if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
2894 			     BTRFS_BLOCK_GROUP_RAID10))
2895 			return stripped | BTRFS_BLOCK_GROUP_DUP;
2896 		return flags;
2897 	} else {
2898 		/* they already had raid on here, just return */
2899 		if (flags & stripped)
2900 			return flags;
2901 
2902 		stripped |= BTRFS_BLOCK_GROUP_DUP;
2903 		stripped = flags & ~stripped;
2904 
2905 		/* switch duplicated blocks with raid1 */
2906 		if (flags & BTRFS_BLOCK_GROUP_DUP)
2907 			return stripped | BTRFS_BLOCK_GROUP_RAID1;
2908 
2909 		/* turn single device chunks into raid0 */
2910 		return stripped | BTRFS_BLOCK_GROUP_RAID0;
2911 	}
2912 	return flags;
2913 }
2914 
2915 int __alloc_chunk_for_shrink(struct btrfs_root *root,
2916 		     struct btrfs_block_group_cache *shrink_block_group,
2917 		     int force)
2918 {
2919 	struct btrfs_trans_handle *trans;
2920 	u64 new_alloc_flags;
2921 	u64 calc;
2922 
2923 	if (btrfs_block_group_used(&shrink_block_group->item) > 0) {
2924 
2925 		trans = btrfs_start_transaction(root, 1);
2926 		new_alloc_flags = update_block_group_flags(root,
2927 						   shrink_block_group->flags);
2928 		if (new_alloc_flags != shrink_block_group->flags) {
2929 			calc =
2930 			     btrfs_block_group_used(&shrink_block_group->item);
2931 		} else {
2932 			calc = shrink_block_group->key.offset;
2933 		}
2934 		do_chunk_alloc(trans, root->fs_info->extent_root,
2935 			       calc + 2 * 1024 * 1024, new_alloc_flags, force);
2936 		btrfs_end_transaction(trans, root);
2937 	}
2938 	return 0;
2939 }
2940 
2941 int btrfs_shrink_extent_tree(struct btrfs_root *root, u64 shrink_start)
2942 {
2943 	struct btrfs_trans_handle *trans;
2944 	struct btrfs_root *tree_root = root->fs_info->tree_root;
2945 	struct btrfs_path *path;
2946 	u64 cur_byte;
2947 	u64 total_found;
2948 	u64 shrink_last_byte;
2949 	struct btrfs_block_group_cache *shrink_block_group;
2950 	struct btrfs_fs_info *info = root->fs_info;
2951 	struct btrfs_key key;
2952 	struct btrfs_key found_key;
2953 	struct extent_buffer *leaf;
2954 	u32 nritems;
2955 	int ret;
2956 	int progress;
2957 
2958 	mutex_lock(&root->fs_info->alloc_mutex);
2959 	shrink_block_group = btrfs_lookup_block_group(root->fs_info,
2960 						      shrink_start);
2961 	BUG_ON(!shrink_block_group);
2962 
2963 	shrink_last_byte = shrink_block_group->key.objectid +
2964 		shrink_block_group->key.offset;
2965 
2966 	shrink_block_group->space_info->total_bytes -=
2967 		shrink_block_group->key.offset;
2968 	path = btrfs_alloc_path();
2969 	root = root->fs_info->extent_root;
2970 	path->reada = 2;
2971 
2972 	printk("btrfs relocating block group %llu flags %llu\n",
2973 	       (unsigned long long)shrink_start,
2974 	       (unsigned long long)shrink_block_group->flags);
2975 
2976 	__alloc_chunk_for_shrink(root, shrink_block_group, 1);
2977 
2978 again:
2979 
2980 	shrink_block_group->ro = 1;
2981 
2982 	total_found = 0;
2983 	progress = 0;
2984 	key.objectid = shrink_start;
2985 	key.offset = 0;
2986 	key.type = 0;
2987 	cur_byte = key.objectid;
2988 
2989 	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2990 	if (ret < 0)
2991 		goto out;
2992 
2993 	ret = btrfs_previous_item(root, path, 0, BTRFS_EXTENT_ITEM_KEY);
2994 	if (ret < 0)
2995 		goto out;
2996 
2997 	if (ret == 0) {
2998 		leaf = path->nodes[0];
2999 		btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3000 		if (found_key.objectid + found_key.offset > shrink_start &&
3001 		    found_key.objectid < shrink_last_byte) {
3002 			cur_byte = found_key.objectid;
3003 			key.objectid = cur_byte;
3004 		}
3005 	}
3006 	btrfs_release_path(root, path);
3007 
3008 	while(1) {
3009 		ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3010 		if (ret < 0)
3011 			goto out;
3012 
3013 		leaf = path->nodes[0];
3014 		nritems = btrfs_header_nritems(leaf);
3015 next:
3016 		if (path->slots[0] >= nritems) {
3017 			ret = btrfs_next_leaf(root, path);
3018 			if (ret < 0)
3019 				goto out;
3020 			if (ret == 1) {
3021 				ret = 0;
3022 				break;
3023 			}
3024 			leaf = path->nodes[0];
3025 			nritems = btrfs_header_nritems(leaf);
3026 		}
3027 
3028 		btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3029 
3030 		if (found_key.objectid >= shrink_last_byte)
3031 			break;
3032 
3033 		if (progress && need_resched()) {
3034 			memcpy(&key, &found_key, sizeof(key));
3035 			cond_resched();
3036 			btrfs_release_path(root, path);
3037 			btrfs_search_slot(NULL, root, &key, path, 0, 0);
3038 			progress = 0;
3039 			goto next;
3040 		}
3041 		progress = 1;
3042 
3043 		if (btrfs_key_type(&found_key) != BTRFS_EXTENT_ITEM_KEY ||
3044 		    found_key.objectid + found_key.offset <= cur_byte) {
3045 			memcpy(&key, &found_key, sizeof(key));
3046 			key.offset++;
3047 			path->slots[0]++;
3048 			goto next;
3049 		}
3050 
3051 		total_found++;
3052 		cur_byte = found_key.objectid + found_key.offset;
3053 		key.objectid = cur_byte;
3054 		btrfs_release_path(root, path);
3055 		ret = relocate_one_extent(root, path, &found_key);
3056 		__alloc_chunk_for_shrink(root, shrink_block_group, 0);
3057 	}
3058 
3059 	btrfs_release_path(root, path);
3060 
3061 	if (total_found > 0) {
3062 		printk("btrfs relocate found %llu last extent was %llu\n",
3063 		       (unsigned long long)total_found,
3064 		       (unsigned long long)found_key.objectid);
3065 		trans = btrfs_start_transaction(tree_root, 1);
3066 		btrfs_commit_transaction(trans, tree_root);
3067 
3068 		btrfs_clean_old_snapshots(tree_root);
3069 
3070 		trans = btrfs_start_transaction(tree_root, 1);
3071 		btrfs_commit_transaction(trans, tree_root);
3072 		goto again;
3073 	}
3074 
3075 	/*
3076 	 * we've freed all the extents, now remove the block
3077 	 * group item from the tree
3078 	 */
3079 	trans = btrfs_start_transaction(root, 1);
3080 	memcpy(&key, &shrink_block_group->key, sizeof(key));
3081 
3082 	ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3083 	if (ret > 0)
3084 		ret = -EIO;
3085 	if (ret < 0)
3086 		goto out;
3087 
3088 	clear_extent_bits(&info->block_group_cache, key.objectid,
3089 			  key.objectid + key.offset - 1,
3090 			  (unsigned int)-1, GFP_NOFS);
3091 
3092 
3093 	clear_extent_bits(&info->free_space_cache,
3094 			   key.objectid, key.objectid + key.offset - 1,
3095 			   (unsigned int)-1, GFP_NOFS);
3096 
3097 	memset(shrink_block_group, 0, sizeof(*shrink_block_group));
3098 	kfree(shrink_block_group);
3099 
3100 	btrfs_del_item(trans, root, path);
3101 	btrfs_commit_transaction(trans, root);
3102 
3103 	/* the code to unpin extents might set a few bits in the free
3104 	 * space cache for this range again
3105 	 */
3106 	clear_extent_bits(&info->free_space_cache,
3107 			   key.objectid, key.objectid + key.offset - 1,
3108 			   (unsigned int)-1, GFP_NOFS);
3109 out:
3110 	btrfs_free_path(path);
3111 	mutex_unlock(&root->fs_info->alloc_mutex);
3112 	return ret;
3113 }
3114 
3115 int find_first_block_group(struct btrfs_root *root, struct btrfs_path *path,
3116 			   struct btrfs_key *key)
3117 {
3118 	int ret = 0;
3119 	struct btrfs_key found_key;
3120 	struct extent_buffer *leaf;
3121 	int slot;
3122 
3123 	ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
3124 	if (ret < 0)
3125 		goto out;
3126 
3127 	while(1) {
3128 		slot = path->slots[0];
3129 		leaf = path->nodes[0];
3130 		if (slot >= btrfs_header_nritems(leaf)) {
3131 			ret = btrfs_next_leaf(root, path);
3132 			if (ret == 0)
3133 				continue;
3134 			if (ret < 0)
3135 				goto out;
3136 			break;
3137 		}
3138 		btrfs_item_key_to_cpu(leaf, &found_key, slot);
3139 
3140 		if (found_key.objectid >= key->objectid &&
3141 		    found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
3142 			ret = 0;
3143 			goto out;
3144 		}
3145 		path->slots[0]++;
3146 	}
3147 	ret = -ENOENT;
3148 out:
3149 	return ret;
3150 }
3151 
3152 int btrfs_read_block_groups(struct btrfs_root *root)
3153 {
3154 	struct btrfs_path *path;
3155 	int ret;
3156 	int bit;
3157 	struct btrfs_block_group_cache *cache;
3158 	struct btrfs_fs_info *info = root->fs_info;
3159 	struct btrfs_space_info *space_info;
3160 	struct extent_io_tree *block_group_cache;
3161 	struct btrfs_key key;
3162 	struct btrfs_key found_key;
3163 	struct extent_buffer *leaf;
3164 
3165 	block_group_cache = &info->block_group_cache;
3166 	root = info->extent_root;
3167 	key.objectid = 0;
3168 	key.offset = 0;
3169 	btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
3170 	path = btrfs_alloc_path();
3171 	if (!path)
3172 		return -ENOMEM;
3173 
3174 	mutex_lock(&root->fs_info->alloc_mutex);
3175 	while(1) {
3176 		ret = find_first_block_group(root, path, &key);
3177 		if (ret > 0) {
3178 			ret = 0;
3179 			goto error;
3180 		}
3181 		if (ret != 0)
3182 			goto error;
3183 
3184 		leaf = path->nodes[0];
3185 		btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3186 		cache = kzalloc(sizeof(*cache), GFP_NOFS);
3187 		if (!cache) {
3188 			ret = -ENOMEM;
3189 			break;
3190 		}
3191 
3192 		read_extent_buffer(leaf, &cache->item,
3193 				   btrfs_item_ptr_offset(leaf, path->slots[0]),
3194 				   sizeof(cache->item));
3195 		memcpy(&cache->key, &found_key, sizeof(found_key));
3196 
3197 		key.objectid = found_key.objectid + found_key.offset;
3198 		btrfs_release_path(root, path);
3199 		cache->flags = btrfs_block_group_flags(&cache->item);
3200 		bit = 0;
3201 		if (cache->flags & BTRFS_BLOCK_GROUP_DATA) {
3202 			bit = BLOCK_GROUP_DATA;
3203 		} else if (cache->flags & BTRFS_BLOCK_GROUP_SYSTEM) {
3204 			bit = BLOCK_GROUP_SYSTEM;
3205 		} else if (cache->flags & BTRFS_BLOCK_GROUP_METADATA) {
3206 			bit = BLOCK_GROUP_METADATA;
3207 		}
3208 		set_avail_alloc_bits(info, cache->flags);
3209 
3210 		ret = update_space_info(info, cache->flags, found_key.offset,
3211 					btrfs_block_group_used(&cache->item),
3212 					&space_info);
3213 		BUG_ON(ret);
3214 		cache->space_info = space_info;
3215 
3216 		/* use EXTENT_LOCKED to prevent merging */
3217 		set_extent_bits(block_group_cache, found_key.objectid,
3218 				found_key.objectid + found_key.offset - 1,
3219 				bit | EXTENT_LOCKED, GFP_NOFS);
3220 		set_state_private(block_group_cache, found_key.objectid,
3221 				  (unsigned long)cache);
3222 
3223 		if (key.objectid >=
3224 		    btrfs_super_total_bytes(&info->super_copy))
3225 			break;
3226 	}
3227 	ret = 0;
3228 error:
3229 	btrfs_free_path(path);
3230 	mutex_unlock(&root->fs_info->alloc_mutex);
3231 	return ret;
3232 }
3233 
3234 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
3235 			   struct btrfs_root *root, u64 bytes_used,
3236 			   u64 type, u64 chunk_objectid, u64 chunk_offset,
3237 			   u64 size)
3238 {
3239 	int ret;
3240 	int bit = 0;
3241 	struct btrfs_root *extent_root;
3242 	struct btrfs_block_group_cache *cache;
3243 	struct extent_io_tree *block_group_cache;
3244 
3245 	extent_root = root->fs_info->extent_root;
3246 	block_group_cache = &root->fs_info->block_group_cache;
3247 
3248 	cache = kzalloc(sizeof(*cache), GFP_NOFS);
3249 	BUG_ON(!cache);
3250 	cache->key.objectid = chunk_offset;
3251 	cache->key.offset = size;
3252 	btrfs_set_key_type(&cache->key, BTRFS_BLOCK_GROUP_ITEM_KEY);
3253 
3254 	btrfs_set_block_group_used(&cache->item, bytes_used);
3255 	btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
3256 	cache->flags = type;
3257 	btrfs_set_block_group_flags(&cache->item, type);
3258 
3259 	ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
3260 				&cache->space_info);
3261 	BUG_ON(ret);
3262 
3263 	bit = block_group_state_bits(type);
3264 	set_extent_bits(block_group_cache, chunk_offset,
3265 			chunk_offset + size - 1,
3266 			bit | EXTENT_LOCKED, GFP_NOFS);
3267 
3268 	set_state_private(block_group_cache, chunk_offset,
3269 			  (unsigned long)cache);
3270 	ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
3271 				sizeof(cache->item));
3272 	BUG_ON(ret);
3273 
3274 	finish_current_insert(trans, extent_root);
3275 	ret = del_pending_extents(trans, extent_root);
3276 	BUG_ON(ret);
3277 	set_avail_alloc_bits(extent_root->fs_info, type);
3278 
3279 	return 0;
3280 }
3281