projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Btrfs: try committing transaction before returning ENOSPC
[linux-2.6-block.git] / fs / btrfs / extent-tree.c
CommitLineData
6cbd5570
CM		 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 */
ec6b910f 18#include <linux/sched.h>
edbd8d4e 19#include <linux/pagemap.h>
ec44a35c 20#include <linux/writeback.h>
21af804c 21#include <linux/blkdev.h>
b7a9f29f 22#include <linux/sort.h>
4b4e25f2 23#include "compat.h"
74493f7a 24#include "hash.h"
a5eb62e3 25#include "crc32c.h"
fec577fb
CM		 26#include "ctree.h"
27#include "disk-io.h"
28#include "print-tree.h"
e089f05c 29#include "transaction.h"
0b86a832 30#include "volumes.h"
925baedd 31#include "locking.h"
31153d81 32#include "ref-cache.h"
fec577fb 33
31840ae1
ZY		 34#define PENDING_EXTENT_INSERT 0
35#define PENDING_EXTENT_DELETE 1
36#define PENDING_BACKREF_UPDATE 2
37
38struct pending_extent_op {
39 int type;
40 u64 bytenr;
41 u64 num_bytes;
42 u64 parent;
43 u64 orig_parent;
44 u64 generation;
45 u64 orig_generation;
46 int level;
f3465ca4
JB		 47 struct list_head list;
48 int del;
31840ae1
ZY		 49};
50
d397712b
CM		 51static int finish_current_insert(struct btrfs_trans_handle *trans,
52 struct btrfs_root *extent_root, int all);
53static int del_pending_extents(struct btrfs_trans_handle *trans,
54 struct btrfs_root *extent_root, int all);
f3465ca4
JB		 55static int pin_down_bytes(struct btrfs_trans_handle *trans,
56 struct btrfs_root *root,
57 u64 bytenr, u64 num_bytes, int is_data);
58static int update_block_group(struct btrfs_trans_handle *trans,
59 struct btrfs_root *root,
60 u64 bytenr, u64 num_bytes, int alloc,
61 int mark_free);
d548ee51 62
6a63209f
JB		 63static int do_chunk_alloc(struct btrfs_trans_handle *trans,
64 struct btrfs_root *extent_root, u64 alloc_bytes,
65 u64 flags, int force);
66
0f9dd46c
JB		 67static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
68{
69 return (cache->flags & bits) == bits;
70}
71
72/*
73 * this adds the block group to the fs_info rb tree for the block group
74 * cache
75 */
b2950863 76static int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
0f9dd46c
JB		 77 struct btrfs_block_group_cache *block_group)
78{
79 struct rb_node **p;
80 struct rb_node *parent = NULL;
81 struct btrfs_block_group_cache *cache;
82
83 spin_lock(&info->block_group_cache_lock);
84 p = &info->block_group_cache_tree.rb_node;
85
86 while (*p) {
87 parent = *p;
88 cache = rb_entry(parent, struct btrfs_block_group_cache,
89 cache_node);
90 if (block_group->key.objectid < cache->key.objectid) {
91 p = &(*p)->rb_left;
92 } else if (block_group->key.objectid > cache->key.objectid) {
93 p = &(*p)->rb_right;
94 } else {
95 spin_unlock(&info->block_group_cache_lock);
96 return -EEXIST;
97 }
98 }
99
100 rb_link_node(&block_group->cache_node, parent, p);
101 rb_insert_color(&block_group->cache_node,
102 &info->block_group_cache_tree);
103 spin_unlock(&info->block_group_cache_lock);
104
105 return 0;
106}
107
108/*
109 * This will return the block group at or after bytenr if contains is 0, else
110 * it will return the block group that contains the bytenr
111 */
112static struct btrfs_block_group_cache *
113block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
114 int contains)
115{
116 struct btrfs_block_group_cache *cache, *ret = NULL;
117 struct rb_node *n;
118 u64 end, start;
119
120 spin_lock(&info->block_group_cache_lock);
121 n = info->block_group_cache_tree.rb_node;
122
123 while (n) {
124 cache = rb_entry(n, struct btrfs_block_group_cache,
125 cache_node);
126 end = cache->key.objectid + cache->key.offset - 1;
127 start = cache->key.objectid;
128
129 if (bytenr < start) {
130 if (!contains && (!ret || start < ret->key.objectid))
131 ret = cache;
132 n = n->rb_left;
133 } else if (bytenr > start) {
134 if (contains && bytenr <= end) {
135 ret = cache;
136 break;
137 }
138 n = n->rb_right;
139 } else {
140 ret = cache;
141 break;
142 }
143 }
d2fb3437
YZ		 144 if (ret)
145 atomic_inc(&ret->count);
0f9dd46c
JB		 146 spin_unlock(&info->block_group_cache_lock);
147
148 return ret;
149}
150
151/*
152 * this is only called by cache_block_group, since we could have freed extents
153 * we need to check the pinned_extents for any extents that can't be used yet
154 * since their free space will be released as soon as the transaction commits.
155 */
156static int add_new_free_space(struct btrfs_block_group_cache *block_group,
157 struct btrfs_fs_info *info, u64 start, u64 end)
158{
159 u64 extent_start, extent_end, size;
160 int ret;
161
25179201 162 mutex_lock(&info->pinned_mutex);
0f9dd46c
JB		 163 while (start < end) {
164 ret = find_first_extent_bit(&info->pinned_extents, start,
165 &extent_start, &extent_end,
166 EXTENT_DIRTY);
167 if (ret)
168 break;
169
170 if (extent_start == start) {
171 start = extent_end + 1;
172 } else if (extent_start > start && extent_start < end) {
173 size = extent_start - start;
ea6a478e
JB		 174 ret = btrfs_add_free_space(block_group, start,
175 size);
0f9dd46c
JB		 176 BUG_ON(ret);
177 start = extent_end + 1;
178 } else {
179 break;
180 }
181 }
182
183 if (start < end) {
184 size = end - start;
ea6a478e 185 ret = btrfs_add_free_space(block_group, start, size);
0f9dd46c
JB		 186 BUG_ON(ret);
187 }
25179201 188 mutex_unlock(&info->pinned_mutex);
0f9dd46c
JB		 189
190 return 0;
191}
192
a512bbf8
YZ		 193static int remove_sb_from_cache(struct btrfs_root *root,
194 struct btrfs_block_group_cache *cache)
195{
196 u64 bytenr;
197 u64 *logical;
198 int stripe_len;
199 int i, nr, ret;
200
201 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
202 bytenr = btrfs_sb_offset(i);
203 ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
204 cache->key.objectid, bytenr, 0,
205 &logical, &nr, &stripe_len);
206 BUG_ON(ret);
207 while (nr--) {
208 btrfs_remove_free_space(cache, logical[nr],
209 stripe_len);
210 }
211 kfree(logical);
212 }
213 return 0;
214}
215
e37c9e69
CM		 216static int cache_block_group(struct btrfs_root *root,
217 struct btrfs_block_group_cache *block_group)
218{
219 struct btrfs_path *path;
ef8bbdfe 220 int ret = 0;
e37c9e69 221 struct btrfs_key key;
5f39d397 222 struct extent_buffer *leaf;
e37c9e69 223 int slot;
e4404d6e 224 u64 last;
e37c9e69 225
00f5c795
CM		 226 if (!block_group)
227 return 0;
228
e37c9e69 229 root = root->fs_info->extent_root;
e37c9e69
CM		 230
231 if (block_group->cached)
232 return 0;
f510cfec 233
e37c9e69
CM		 234 path = btrfs_alloc_path();
235 if (!path)
236 return -ENOMEM;
7d7d6068 237
2cc58cf2 238 path->reada = 2;
5cd57b2c
CM		 239 /*
240 * we get into deadlocks with paths held by callers of this function.
241 * since the alloc_mutex is protecting things right now, just
242 * skip the locking here
243 */
244 path->skip_locking = 1;
e4404d6e
YZ		 245 last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
246 key.objectid = last;
e37c9e69
CM		 247 key.offset = 0;
248 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
249 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
250 if (ret < 0)
ef8bbdfe 251 goto err;
a512bbf8 252
d397712b 253 while (1) {
5f39d397 254 leaf = path->nodes[0];
e37c9e69 255 slot = path->slots[0];
5f39d397 256 if (slot >= btrfs_header_nritems(leaf)) {
e37c9e69 257 ret = btrfs_next_leaf(root, path);
54aa1f4d
CM		 258 if (ret < 0)
259 goto err;
0f9dd46c 260 if (ret == 0)
e37c9e69 261 continue;
0f9dd46c 262 else
e37c9e69 263 break;
e37c9e69 264 }
5f39d397 265 btrfs_item_key_to_cpu(leaf, &key, slot);
0f9dd46c 266 if (key.objectid < block_group->key.objectid)
7d7d6068 267 goto next;
0f9dd46c 268
e37c9e69 269 if (key.objectid >= block_group->key.objectid +
0f9dd46c 270 block_group->key.offset)
e37c9e69 271 break;
7d7d6068 272
e37c9e69 273 if (btrfs_key_type(&key) == BTRFS_EXTENT_ITEM_KEY) {
0f9dd46c
JB		 274 add_new_free_space(block_group, root->fs_info, last,
275 key.objectid);
276
7d7d6068 277 last = key.objectid + key.offset;
e37c9e69 278 }
7d7d6068 279next:
e37c9e69
CM		 280 path->slots[0]++;
281 }
282
0f9dd46c
JB		 283 add_new_free_space(block_group, root->fs_info, last,
284 block_group->key.objectid +
285 block_group->key.offset);
286
a512bbf8 287 remove_sb_from_cache(root, block_group);
e37c9e69 288 block_group->cached = 1;
ef8bbdfe 289 ret = 0;
54aa1f4d 290err:
e37c9e69 291 btrfs_free_path(path);
ef8bbdfe 292 return ret;
e37c9e69
CM		 293}
294
0f9dd46c
JB		 295/*
296 * return the block group that starts at or after bytenr
297 */
d397712b
CM		 298static struct btrfs_block_group_cache *
299btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr)
0ef3e66b 300{
0f9dd46c 301 struct btrfs_block_group_cache *cache;
0ef3e66b 302
0f9dd46c 303 cache = block_group_cache_tree_search(info, bytenr, 0);
0ef3e66b 304
0f9dd46c 305 return cache;
0ef3e66b
CM		 306}
307
0f9dd46c
JB		 308/*
309 * return the block group that contains teh given bytenr
310 */
d397712b
CM		 311struct btrfs_block_group_cache *btrfs_lookup_block_group(
312 struct btrfs_fs_info *info,
313 u64 bytenr)
be744175 314{
0f9dd46c 315 struct btrfs_block_group_cache *cache;
be744175 316
0f9dd46c 317 cache = block_group_cache_tree_search(info, bytenr, 1);
96b5179d 318
0f9dd46c 319 return cache;
be744175 320}
0b86a832 321
d2fb3437
YZ		 322static inline void put_block_group(struct btrfs_block_group_cache *cache)
323{
324 if (atomic_dec_and_test(&cache->count))
325 kfree(cache);
326}
327
0f9dd46c
JB		 328static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
329 u64 flags)
6324fbf3 330{
0f9dd46c 331 struct list_head *head = &info->space_info;
0f9dd46c 332 struct btrfs_space_info *found;
c6e30871 333 list_for_each_entry(found, head, list) {
0f9dd46c
JB		 334 if (found->flags == flags)
335 return found;
336 }
337 return NULL;
6324fbf3
CM		 338}
339
80eb234a
JB		 340static u64 div_factor(u64 num, int factor)
341{
342 if (factor == 10)
343 return num;
344 num *= factor;
345 do_div(num, 10);
346 return num;
347}
348
d2fb3437
YZ		 349u64 btrfs_find_block_group(struct btrfs_root *root,
350 u64 search_start, u64 search_hint, int owner)
cd1bc465 351{
96b5179d 352 struct btrfs_block_group_cache *cache;
cd1bc465 353 u64 used;
d2fb3437
YZ		 354 u64 last = max(search_hint, search_start);
355 u64 group_start = 0;
31f3c99b 356 int full_search = 0;
d2fb3437 357 int factor = 9;
0ef3e66b 358 int wrapped = 0;
31f3c99b 359again:
e8569813
ZY		 360 while (1) {
361 cache = btrfs_lookup_first_block_group(root->fs_info, last);
0f9dd46c
JB		 362 if (!cache)
363 break;
96b5179d 364
c286ac48 365 spin_lock(&cache->lock);
96b5179d
CM		 366 last = cache->key.objectid + cache->key.offset;
367 used = btrfs_block_group_used(&cache->item);
368
d2fb3437
YZ		 369 if ((full_search || !cache->ro) &&
370 block_group_bits(cache, BTRFS_BLOCK_GROUP_METADATA)) {
e8569813 371 if (used + cache->pinned + cache->reserved <
d2fb3437
YZ		 372 div_factor(cache->key.offset, factor)) {
373 group_start = cache->key.objectid;
c286ac48 374 spin_unlock(&cache->lock);
d2fb3437 375 put_block_group(cache);
8790d502
CM		 376 goto found;
377 }
6324fbf3 378 }
c286ac48 379 spin_unlock(&cache->lock);
d2fb3437 380 put_block_group(cache);
de428b63 381 cond_resched();
cd1bc465 382 }
0ef3e66b
CM		 383 if (!wrapped) {
384 last = search_start;
385 wrapped = 1;
386 goto again;
387 }
388 if (!full_search && factor < 10) {
be744175 389 last = search_start;
31f3c99b 390 full_search = 1;
0ef3e66b 391 factor = 10;
31f3c99b
CM		 392 goto again;
393 }
be744175 394found:
d2fb3437 395 return group_start;
925baedd 396}
0f9dd46c 397
e02119d5 398/* simple helper to search for an existing extent at a given offset */
31840ae1 399int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len)
e02119d5
CM		 400{
401 int ret;
402 struct btrfs_key key;
31840ae1 403 struct btrfs_path *path;
e02119d5 404
31840ae1
ZY		 405 path = btrfs_alloc_path();
406 BUG_ON(!path);
e02119d5
CM		 407 key.objectid = start;
408 key.offset = len;
409 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
410 ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
411 0, 0);
31840ae1 412 btrfs_free_path(path);
7bb86316
CM		 413 return ret;
414}
415
d8d5f3e1
CM		 416/*
417 * Back reference rules. Back refs have three main goals:
418 *
419 * 1) differentiate between all holders of references to an extent so that
420 * when a reference is dropped we can make sure it was a valid reference
421 * before freeing the extent.
422 *
423 * 2) Provide enough information to quickly find the holders of an extent
424 * if we notice a given block is corrupted or bad.
425 *
426 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
427 * maintenance. This is actually the same as #2, but with a slightly
428 * different use case.
429 *
430 * File extents can be referenced by:
431 *
432 * - multiple snapshots, subvolumes, or different generations in one subvol
31840ae1 433 * - different files inside a single subvolume
d8d5f3e1
CM		 434 * - different offsets inside a file (bookend extents in file.c)
435 *
436 * The extent ref structure has fields for:
437 *
438 * - Objectid of the subvolume root
439 * - Generation number of the tree holding the reference
440 * - objectid of the file holding the reference
31840ae1
ZY		 441 * - number of references holding by parent node (alway 1 for tree blocks)
442 *
443 * Btree leaf may hold multiple references to a file extent. In most cases,
444 * these references are from same file and the corresponding offsets inside
3bb1a1bc 445 * the file are close together.
d8d5f3e1
CM		 446 *
447 * When a file extent is allocated the fields are filled in:
3bb1a1bc 448 * (root_key.objectid, trans->transid, inode objectid, 1)
d8d5f3e1
CM		 449 *
450 * When a leaf is cow'd new references are added for every file extent found
31840ae1
ZY		 451 * in the leaf. It looks similar to the create case, but trans->transid will
452 * be different when the block is cow'd.
d8d5f3e1 453 *
3bb1a1bc 454 * (root_key.objectid, trans->transid, inode objectid,
31840ae1 455 * number of references in the leaf)
d8d5f3e1 456 *
3bb1a1bc
YZ		 457 * When a file extent is removed either during snapshot deletion or
458 * file truncation, we find the corresponding back reference and check
459 * the following fields:
d8d5f3e1 460 *
3bb1a1bc
YZ		 461 * (btrfs_header_owner(leaf), btrfs_header_generation(leaf),
462 * inode objectid)
d8d5f3e1
CM		 463 *
464 * Btree extents can be referenced by:
465 *
466 * - Different subvolumes
467 * - Different generations of the same subvolume
468 *
d8d5f3e1
CM		 469 * When a tree block is created, back references are inserted:
470 *
3bb1a1bc 471 * (root->root_key.objectid, trans->transid, level, 1)
d8d5f3e1 472 *
31840ae1
ZY		 473 * When a tree block is cow'd, new back references are added for all the
474 * blocks it points to. If the tree block isn't in reference counted root,
475 * the old back references are removed. These new back references are of
476 * the form (trans->transid will have increased since creation):
d8d5f3e1 477 *
3bb1a1bc 478 * (root->root_key.objectid, trans->transid, level, 1)
d8d5f3e1 479 *
31840ae1 480 * When a backref is in deleting, the following fields are checked:
d8d5f3e1
CM		 481 *
482 * if backref was for a tree root:
3bb1a1bc 483 * (btrfs_header_owner(itself), btrfs_header_generation(itself), level)
d8d5f3e1 484 * else
3bb1a1bc 485 * (btrfs_header_owner(parent), btrfs_header_generation(parent), level)
d8d5f3e1 486 *
31840ae1 487 * Back Reference Key composing:
d8d5f3e1 488 *
31840ae1
ZY		 489 * The key objectid corresponds to the first byte in the extent, the key
490 * type is set to BTRFS_EXTENT_REF_KEY, and the key offset is the first
491 * byte of parent extent. If a extent is tree root, the key offset is set
492 * to the key objectid.
d8d5f3e1 493 */
31840ae1 494
d397712b 495static noinline int lookup_extent_backref(struct btrfs_trans_handle *trans,
31840ae1 496 struct btrfs_root *root,
3bb1a1bc
YZ		 497 struct btrfs_path *path,
498 u64 bytenr, u64 parent,
499 u64 ref_root, u64 ref_generation,
500 u64 owner_objectid, int del)
7bb86316 501{
7bb86316 502 struct btrfs_key key;
31840ae1
ZY		 503 struct btrfs_extent_ref *ref;
504 struct extent_buffer *leaf;
3bb1a1bc 505 u64 ref_objectid;
74493f7a
CM		 506 int ret;
507
31840ae1
ZY		 508 key.objectid = bytenr;
509 key.type = BTRFS_EXTENT_REF_KEY;
510 key.offset = parent;
511
512 ret = btrfs_search_slot(trans, root, &key, path, del ? -1 : 0, 1);
513 if (ret < 0)
514 goto out;
515 if (ret > 0) {
516 ret = -ENOENT;
517 goto out;
518 }
519
520 leaf = path->nodes[0];
521 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_ref);
3bb1a1bc 522 ref_objectid = btrfs_ref_objectid(leaf, ref);
31840ae1 523 if (btrfs_ref_root(leaf, ref) != ref_root ||
3bb1a1bc
YZ		 524 btrfs_ref_generation(leaf, ref) != ref_generation ||
525 (ref_objectid != owner_objectid &&
526 ref_objectid != BTRFS_MULTIPLE_OBJECTIDS)) {
31840ae1
ZY		 527 ret = -EIO;
528 WARN_ON(1);
529 goto out;
530 }
531 ret = 0;
532out:
533 return ret;
534}
535
f3465ca4
JB		 536/*
537 * updates all the backrefs that are pending on update_list for the
538 * extent_root
539 */
d397712b 540static noinline int update_backrefs(struct btrfs_trans_handle *trans,
f3465ca4
JB		 541 struct btrfs_root *extent_root,
542 struct btrfs_path *path,
543 struct list_head *update_list)
544{
545 struct btrfs_key key;
546 struct btrfs_extent_ref *ref;
547 struct btrfs_fs_info *info = extent_root->fs_info;
548 struct pending_extent_op *op;
549 struct extent_buffer *leaf;
550 int ret = 0;
551 struct list_head *cur = update_list->next;
552 u64 ref_objectid;
553 u64 ref_root = extent_root->root_key.objectid;
554
555 op = list_entry(cur, struct pending_extent_op, list);
556
557search:
558 key.objectid = op->bytenr;
559 key.type = BTRFS_EXTENT_REF_KEY;
560 key.offset = op->orig_parent;
561
562 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 1);
563 BUG_ON(ret);
564
565 leaf = path->nodes[0];
566
567loop:
568 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_ref);
569
570 ref_objectid = btrfs_ref_objectid(leaf, ref);
571
572 if (btrfs_ref_root(leaf, ref) != ref_root ||
573 btrfs_ref_generation(leaf, ref) != op->orig_generation ||
574 (ref_objectid != op->level &&
575 ref_objectid != BTRFS_MULTIPLE_OBJECTIDS)) {
d397712b
CM		 576 printk(KERN_ERR "btrfs couldn't find %llu, parent %llu, "
577 "root %llu, owner %u\n",
578 (unsigned long long)op->bytenr,
579 (unsigned long long)op->orig_parent,
580 (unsigned long long)ref_root, op->level);
f3465ca4
JB		 581 btrfs_print_leaf(extent_root, leaf);
582 BUG();
583 }
584
585 key.objectid = op->bytenr;
586 key.offset = op->parent;
587 key.type = BTRFS_EXTENT_REF_KEY;
588 ret = btrfs_set_item_key_safe(trans, extent_root, path, &key);
589 BUG_ON(ret);
590 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_ref);
591 btrfs_set_ref_generation(leaf, ref, op->generation);
592
593 cur = cur->next;
594
595 list_del_init(&op->list);
596 unlock_extent(&info->extent_ins, op->bytenr,
597 op->bytenr + op->num_bytes - 1, GFP_NOFS);
598 kfree(op);
599
600 if (cur == update_list) {
601 btrfs_mark_buffer_dirty(path->nodes[0]);
602 btrfs_release_path(extent_root, path);
603 goto out;
604 }
605
606 op = list_entry(cur, struct pending_extent_op, list);
607
608 path->slots[0]++;
609 while (path->slots[0] < btrfs_header_nritems(leaf)) {
610 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
611 if (key.objectid == op->bytenr &&
612 key.type == BTRFS_EXTENT_REF_KEY)
613 goto loop;
614 path->slots[0]++;
615 }
616
617 btrfs_mark_buffer_dirty(path->nodes[0]);
618 btrfs_release_path(extent_root, path);
619 goto search;
620
621out:
622 return 0;
623}
624
d397712b 625static noinline int insert_extents(struct btrfs_trans_handle *trans,
f3465ca4
JB		 626 struct btrfs_root *extent_root,
627 struct btrfs_path *path,
628 struct list_head *insert_list, int nr)
629{
630 struct btrfs_key *keys;
631 u32 *data_size;
632 struct pending_extent_op *op;
633 struct extent_buffer *leaf;
634 struct list_head *cur = insert_list->next;
635 struct btrfs_fs_info *info = extent_root->fs_info;
636 u64 ref_root = extent_root->root_key.objectid;
637 int i = 0, last = 0, ret;
638 int total = nr * 2;
639
640 if (!nr)
641 return 0;
642
643 keys = kzalloc(total * sizeof(struct btrfs_key), GFP_NOFS);
644 if (!keys)
645 return -ENOMEM;
646
647 data_size = kzalloc(total * sizeof(u32), GFP_NOFS);
648 if (!data_size) {
649 kfree(keys);
650 return -ENOMEM;
651 }
652
653 list_for_each_entry(op, insert_list, list) {
654 keys[i].objectid = op->bytenr;
655 keys[i].offset = op->num_bytes;
656 keys[i].type = BTRFS_EXTENT_ITEM_KEY;
657 data_size[i] = sizeof(struct btrfs_extent_item);
658 i++;
659
660 keys[i].objectid = op->bytenr;
661 keys[i].offset = op->parent;
662 keys[i].type = BTRFS_EXTENT_REF_KEY;
663 data_size[i] = sizeof(struct btrfs_extent_ref);
664 i++;
665 }
666
667 op = list_entry(cur, struct pending_extent_op, list);
668 i = 0;
669 while (i < total) {
670 int c;
671 ret = btrfs_insert_some_items(trans, extent_root, path,
672 keys+i, data_size+i, total-i);
673 BUG_ON(ret < 0);
674
675 if (last && ret > 1)
676 BUG();
677
678 leaf = path->nodes[0];
679 for (c = 0; c < ret; c++) {
680 int ref_first = keys[i].type == BTRFS_EXTENT_REF_KEY;
681
682 /*
683 * if the first item we inserted was a backref, then
684 * the EXTENT_ITEM will be the odd c's, else it will
685 * be the even c's
686 */
687 if ((ref_first && (c % 2)) ||
688 (!ref_first && !(c % 2))) {
689 struct btrfs_extent_item *itm;
690
691 itm = btrfs_item_ptr(leaf, path->slots[0] + c,
692 struct btrfs_extent_item);
693 btrfs_set_extent_refs(path->nodes[0], itm, 1);
694 op->del++;
695 } else {
696 struct btrfs_extent_ref *ref;
697
698 ref = btrfs_item_ptr(leaf, path->slots[0] + c,
699 struct btrfs_extent_ref);
700 btrfs_set_ref_root(leaf, ref, ref_root);
701 btrfs_set_ref_generation(leaf, ref,
702 op->generation);
703 btrfs_set_ref_objectid(leaf, ref, op->level);
704 btrfs_set_ref_num_refs(leaf, ref, 1);
705 op->del++;
706 }
707
708 /*
709 * using del to see when its ok to free up the
710 * pending_extent_op. In the case where we insert the
711 * last item on the list in order to help do batching
712 * we need to not free the extent op until we actually
713 * insert the extent_item
714 */
715 if (op->del == 2) {
716 unlock_extent(&info->extent_ins, op->bytenr,
717 op->bytenr + op->num_bytes - 1,
718 GFP_NOFS);
719 cur = cur->next;
720 list_del_init(&op->list);
721 kfree(op);
722 if (cur != insert_list)
723 op = list_entry(cur,
724 struct pending_extent_op,
725 list);
726 }
727 }
728 btrfs_mark_buffer_dirty(leaf);
729 btrfs_release_path(extent_root, path);
730
731 /*
732 * Ok backref's and items usually go right next to eachother,
733 * but if we could only insert 1 item that means that we
734 * inserted on the end of a leaf, and we have no idea what may
735 * be on the next leaf so we just play it safe. In order to
736 * try and help this case we insert the last thing on our
737 * insert list so hopefully it will end up being the last
738 * thing on the leaf and everything else will be before it,
739 * which will let us insert a whole bunch of items at the same
740 * time.
741 */
742 if (ret == 1 && !last && (i + ret < total)) {
743 /*
744 * last: where we will pick up the next time around
745 * i: our current key to insert, will be total - 1
746 * cur: the current op we are screwing with
747 * op: duh
748 */
749 last = i + ret;
750 i = total - 1;
751 cur = insert_list->prev;
752 op = list_entry(cur, struct pending_extent_op, list);
753 } else if (last) {
754 /*
755 * ok we successfully inserted the last item on the
756 * list, lets reset everything
757 *
758 * i: our current key to insert, so where we left off
759 * last time
760 * last: done with this
761 * cur: the op we are messing with
762 * op: duh
763 * total: since we inserted the last key, we need to
764 * decrement total so we dont overflow
765 */
766 i = last;
767 last = 0;
f3465ca4 768 total--;
b4eec2ca
LH		 769 if (i < total) {
770 cur = insert_list->next;
771 op = list_entry(cur, struct pending_extent_op,
772 list);
773 }
f3465ca4
JB		 774 } else {
775 i += ret;
776 }
777
778 cond_resched();
779 }
780 ret = 0;
781 kfree(keys);
782 kfree(data_size);
783 return ret;
784}
785
d397712b 786static noinline int insert_extent_backref(struct btrfs_trans_handle *trans,
31840ae1
ZY		 787 struct btrfs_root *root,
788 struct btrfs_path *path,
789 u64 bytenr, u64 parent,
790 u64 ref_root, u64 ref_generation,
3bb1a1bc 791 u64 owner_objectid)
31840ae1
ZY		 792{
793 struct btrfs_key key;
794 struct extent_buffer *leaf;
795 struct btrfs_extent_ref *ref;
796 u32 num_refs;
797 int ret;
74493f7a 798
74493f7a
CM		 799 key.objectid = bytenr;
800 key.type = BTRFS_EXTENT_REF_KEY;
31840ae1 801 key.offset = parent;
74493f7a 802
31840ae1
ZY		 803 ret = btrfs_insert_empty_item(trans, root, path, &key, sizeof(*ref));
804 if (ret == 0) {
805 leaf = path->nodes[0];
806 ref = btrfs_item_ptr(leaf, path->slots[0],
807 struct btrfs_extent_ref);
808 btrfs_set_ref_root(leaf, ref, ref_root);
809 btrfs_set_ref_generation(leaf, ref, ref_generation);
810 btrfs_set_ref_objectid(leaf, ref, owner_objectid);
31840ae1
ZY		 811 btrfs_set_ref_num_refs(leaf, ref, 1);
812 } else if (ret == -EEXIST) {
813 u64 existing_owner;
814 BUG_ON(owner_objectid < BTRFS_FIRST_FREE_OBJECTID);
815 leaf = path->nodes[0];
816 ref = btrfs_item_ptr(leaf, path->slots[0],
817 struct btrfs_extent_ref);
818 if (btrfs_ref_root(leaf, ref) != ref_root ||
819 btrfs_ref_generation(leaf, ref) != ref_generation) {
820 ret = -EIO;
821 WARN_ON(1);
7bb86316 822 goto out;
31840ae1
ZY		 823 }
824
825 num_refs = btrfs_ref_num_refs(leaf, ref);
826 BUG_ON(num_refs == 0);
827 btrfs_set_ref_num_refs(leaf, ref, num_refs + 1);
828
829 existing_owner = btrfs_ref_objectid(leaf, ref);
3bb1a1bc
YZ		 830 if (existing_owner != owner_objectid &&
831 existing_owner != BTRFS_MULTIPLE_OBJECTIDS) {
31840ae1
ZY		 832 btrfs_set_ref_objectid(leaf, ref,
833 BTRFS_MULTIPLE_OBJECTIDS);
31840ae1
ZY		 834 }
835 ret = 0;
836 } else {
7bb86316 837 goto out;
31840ae1 838 }
7bb86316
CM		 839 btrfs_mark_buffer_dirty(path->nodes[0]);
840out:
841 btrfs_release_path(root, path);
842 return ret;
74493f7a
CM		 843}
844
d397712b 845static noinline int remove_extent_backref(struct btrfs_trans_handle *trans,
31840ae1
ZY		 846 struct btrfs_root *root,
847 struct btrfs_path *path)
848{
849 struct extent_buffer *leaf;
850 struct btrfs_extent_ref *ref;
851 u32 num_refs;
852 int ret = 0;
853
854 leaf = path->nodes[0];
855 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_ref);
856 num_refs = btrfs_ref_num_refs(leaf, ref);
857 BUG_ON(num_refs == 0);
858 num_refs -= 1;
859 if (num_refs == 0) {
860 ret = btrfs_del_item(trans, root, path);
861 } else {
862 btrfs_set_ref_num_refs(leaf, ref, num_refs);
863 btrfs_mark_buffer_dirty(leaf);
864 }
865 btrfs_release_path(root, path);
866 return ret;
867}
868
4b4e25f2 869#ifdef BIO_RW_DISCARD
15916de8
CM		 870static void btrfs_issue_discard(struct block_device *bdev,
871 u64 start, u64 len)
872{
15916de8 873 blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_KERNEL);
15916de8 874}
4b4e25f2 875#endif
15916de8 876
1f3c79a2
LH		 877static int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr,
878 u64 num_bytes)
879{
880#ifdef BIO_RW_DISCARD
881 int ret;
882 u64 map_length = num_bytes;
883 struct btrfs_multi_bio *multi = NULL;
884
885 /* Tell the block device(s) that the sectors can be discarded */
886 ret = btrfs_map_block(&root->fs_info->mapping_tree, READ,
887 bytenr, &map_length, &multi, 0);
888 if (!ret) {
889 struct btrfs_bio_stripe *stripe = multi->stripes;
890 int i;
891
892 if (map_length > num_bytes)
893 map_length = num_bytes;
894
895 for (i = 0; i < multi->num_stripes; i++, stripe++) {
896 btrfs_issue_discard(stripe->dev->bdev,
897 stripe->physical,
898 map_length);
899 }
900 kfree(multi);
901 }
902
903 return ret;
904#else
905 return 0;
906#endif
907}
908
d397712b 909static noinline int free_extents(struct btrfs_trans_handle *trans,
f3465ca4
JB		 910 struct btrfs_root *extent_root,
911 struct list_head *del_list)
912{
913 struct btrfs_fs_info *info = extent_root->fs_info;
914 struct btrfs_path *path;
915 struct btrfs_key key, found_key;
916 struct extent_buffer *leaf;
917 struct list_head *cur;
918 struct pending_extent_op *op;
919 struct btrfs_extent_item *ei;
920 int ret, num_to_del, extent_slot = 0, found_extent = 0;
921 u32 refs;
922 u64 bytes_freed = 0;
923
924 path = btrfs_alloc_path();
925 if (!path)
926 return -ENOMEM;
927 path->reada = 1;
928
929search:
930 /* search for the backref for the current ref we want to delete */
931 cur = del_list->next;
932 op = list_entry(cur, struct pending_extent_op, list);
933 ret = lookup_extent_backref(trans, extent_root, path, op->bytenr,
934 op->orig_parent,
935 extent_root->root_key.objectid,
936 op->orig_generation, op->level, 1);
937 if (ret) {
d397712b
CM		 938 printk(KERN_ERR "btrfs unable to find backref byte nr %llu "
939 "root %llu gen %llu owner %u\n",
940 (unsigned long long)op->bytenr,
941 (unsigned long long)extent_root->root_key.objectid,
942 (unsigned long long)op->orig_generation, op->level);
f3465ca4
JB		 943 btrfs_print_leaf(extent_root, path->nodes[0]);
944 WARN_ON(1);
945 goto out;
946 }
947
948 extent_slot = path->slots[0];
949 num_to_del = 1;
950 found_extent = 0;
951
952 /*
953 * if we aren't the first item on the leaf we can move back one and see
954 * if our ref is right next to our extent item
955 */
956 if (likely(extent_slot)) {
957 extent_slot--;
958 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
959 extent_slot);
960 if (found_key.objectid == op->bytenr &&
961 found_key.type == BTRFS_EXTENT_ITEM_KEY &&
962 found_key.offset == op->num_bytes) {
963 num_to_del++;
964 found_extent = 1;
965 }
966 }
967
968 /*
969 * if we didn't find the extent we need to delete the backref and then
970 * search for the extent item key so we can update its ref count
971 */
972 if (!found_extent) {
973 key.objectid = op->bytenr;
974 key.type = BTRFS_EXTENT_ITEM_KEY;
975 key.offset = op->num_bytes;
976
977 ret = remove_extent_backref(trans, extent_root, path);
978 BUG_ON(ret);
979 btrfs_release_path(extent_root, path);
980 ret = btrfs_search_slot(trans, extent_root, &key, path, -1, 1);
981 BUG_ON(ret);
982 extent_slot = path->slots[0];
983 }
984
985 /* this is where we update the ref count for the extent */
986 leaf = path->nodes[0];
987 ei = btrfs_item_ptr(leaf, extent_slot, struct btrfs_extent_item);
988 refs = btrfs_extent_refs(leaf, ei);
989 BUG_ON(refs == 0);
990 refs--;
991 btrfs_set_extent_refs(leaf, ei, refs);
992
993 btrfs_mark_buffer_dirty(leaf);
994
995 /*
996 * This extent needs deleting. The reason cur_slot is extent_slot +
997 * num_to_del is because extent_slot points to the slot where the extent
998 * is, and if the backref was not right next to the extent we will be
999 * deleting at least 1 item, and will want to start searching at the
1000 * slot directly next to extent_slot. However if we did find the
1001 * backref next to the extent item them we will be deleting at least 2
1002 * items and will want to start searching directly after the ref slot
1003 */
1004 if (!refs) {
1005 struct list_head *pos, *n, *end;
1006 int cur_slot = extent_slot+num_to_del;
1007 u64 super_used;
1008 u64 root_used;
1009
1010 path->slots[0] = extent_slot;
1011 bytes_freed = op->num_bytes;
1012
e3e469f8
JB		 1013 mutex_lock(&info->pinned_mutex);
1014 ret = pin_down_bytes(trans, extent_root, op->bytenr,
1015 op->num_bytes, op->level >=
1016 BTRFS_FIRST_FREE_OBJECTID);
1017 mutex_unlock(&info->pinned_mutex);
1018 BUG_ON(ret < 0);
1019 op->del = ret;
1020
f3465ca4
JB		 1021 /*
1022 * we need to see if we can delete multiple things at once, so
1023 * start looping through the list of extents we are wanting to
1024 * delete and see if their extent/backref's are right next to
1025 * eachother and the extents only have 1 ref
1026 */
1027 for (pos = cur->next; pos != del_list; pos = pos->next) {
1028 struct pending_extent_op *tmp;
1029
1030 tmp = list_entry(pos, struct pending_extent_op, list);
1031
1032 /* we only want to delete extent+ref at this stage */
1033 if (cur_slot >= btrfs_header_nritems(leaf) - 1)
1034 break;
1035
1036 btrfs_item_key_to_cpu(leaf, &found_key, cur_slot);
1037 if (found_key.objectid != tmp->bytenr ||
1038 found_key.type != BTRFS_EXTENT_ITEM_KEY ||
1039 found_key.offset != tmp->num_bytes)
1040 break;
1041
1042 /* check to make sure this extent only has one ref */
1043 ei = btrfs_item_ptr(leaf, cur_slot,
1044 struct btrfs_extent_item);
1045 if (btrfs_extent_refs(leaf, ei) != 1)
1046 break;
1047
1048 btrfs_item_key_to_cpu(leaf, &found_key, cur_slot+1);
1049 if (found_key.objectid != tmp->bytenr ||
1050 found_key.type != BTRFS_EXTENT_REF_KEY ||
1051 found_key.offset != tmp->orig_parent)
1052 break;
1053
1054 /*
1055 * the ref is right next to the extent, we can set the
1056 * ref count to 0 since we will delete them both now
1057 */
1058 btrfs_set_extent_refs(leaf, ei, 0);
1059
1060 /* pin down the bytes for this extent */
1061 mutex_lock(&info->pinned_mutex);
1062 ret = pin_down_bytes(trans, extent_root, tmp->bytenr,
1063 tmp->num_bytes, tmp->level >=
1064 BTRFS_FIRST_FREE_OBJECTID);
1065 mutex_unlock(&info->pinned_mutex);
1066 BUG_ON(ret < 0);
1067
1068 /*
1069 * use the del field to tell if we need to go ahead and
1070 * free up the extent when we delete the item or not.
1071 */
1072 tmp->del = ret;
1073 bytes_freed += tmp->num_bytes;
1074
1075 num_to_del += 2;
1076 cur_slot += 2;
1077 }
1078 end = pos;
1079
1080 /* update the free space counters */
75eff68e 1081 spin_lock(&info->delalloc_lock);
f3465ca4
JB		 1082 super_used = btrfs_super_bytes_used(&info->super_copy);
1083 btrfs_set_super_bytes_used(&info->super_copy,
1084 super_used - bytes_freed);
f3465ca4
JB		 1085
1086 root_used = btrfs_root_used(&extent_root->root_item);
1087 btrfs_set_root_used(&extent_root->root_item,
1088 root_used - bytes_freed);
34bf63c4 1089 spin_unlock(&info->delalloc_lock);
f3465ca4
JB		 1090
1091 /* delete the items */
1092 ret = btrfs_del_items(trans, extent_root, path,
1093 path->slots[0], num_to_del);
1094 BUG_ON(ret);
1095
1096 /*
1097 * loop through the extents we deleted and do the cleanup work
1098 * on them
1099 */
1100 for (pos = cur, n = pos->next; pos != end;
1101 pos = n, n = pos->next) {
1102 struct pending_extent_op *tmp;
f3465ca4
JB		 1103 tmp = list_entry(pos, struct pending_extent_op, list);
1104
1105 /*
1106 * remember tmp->del tells us wether or not we pinned
1107 * down the extent
1108 */
1109 ret = update_block_group(trans, extent_root,
1110 tmp->bytenr, tmp->num_bytes, 0,
1111 tmp->del);
1112 BUG_ON(ret);
1113
f3465ca4
JB		 1114 list_del_init(&tmp->list);
1115 unlock_extent(&info->extent_ins, tmp->bytenr,
1116 tmp->bytenr + tmp->num_bytes - 1,
1117 GFP_NOFS);
1118 kfree(tmp);
1119 }
1120 } else if (refs && found_extent) {
1121 /*
1122 * the ref and extent were right next to eachother, but the
1123 * extent still has a ref, so just free the backref and keep
1124 * going
1125 */
1126 ret = remove_extent_backref(trans, extent_root, path);
1127 BUG_ON(ret);
1128
1129 list_del_init(&op->list);
1130 unlock_extent(&info->extent_ins, op->bytenr,
1131 op->bytenr + op->num_bytes - 1, GFP_NOFS);
1132 kfree(op);
1133 } else {
1134 /*
1135 * the extent has multiple refs and the backref we were looking
1136 * for was not right next to it, so just unlock and go next,
1137 * we're good to go
1138 */
1139 list_del_init(&op->list);
1140 unlock_extent(&info->extent_ins, op->bytenr,
1141 op->bytenr + op->num_bytes - 1, GFP_NOFS);
1142 kfree(op);
1143 }
1144
1145 btrfs_release_path(extent_root, path);
1146 if (!list_empty(del_list))
1147 goto search;
1148
1149out:
1150 btrfs_free_path(path);
1151 return ret;
1152}
1153
31840ae1
ZY		 1154static int __btrfs_update_extent_ref(struct btrfs_trans_handle *trans,
1155 struct btrfs_root *root, u64 bytenr,
1156 u64 orig_parent, u64 parent,
1157 u64 orig_root, u64 ref_root,
1158 u64 orig_generation, u64 ref_generation,
3bb1a1bc 1159 u64 owner_objectid)
31840ae1
ZY		 1160{
1161 int ret;
1162 struct btrfs_root *extent_root = root->fs_info->extent_root;
1163 struct btrfs_path *path;
1164
1165 if (root == root->fs_info->extent_root) {
1166 struct pending_extent_op *extent_op;
1167 u64 num_bytes;
1168
1169 BUG_ON(owner_objectid >= BTRFS_MAX_LEVEL);
1170 num_bytes = btrfs_level_size(root, (int)owner_objectid);
25179201 1171 mutex_lock(&root->fs_info->extent_ins_mutex);
31840ae1 1172 if (test_range_bit(&root->fs_info->extent_ins, bytenr,
25179201 1173 bytenr + num_bytes - 1, EXTENT_WRITEBACK, 0)) {
31840ae1
ZY		 1174 u64 priv;
1175 ret = get_state_private(&root->fs_info->extent_ins,
1176 bytenr, &priv);
1177 BUG_ON(ret);
1178 extent_op = (struct pending_extent_op *)
1179 (unsigned long)priv;
1180 BUG_ON(extent_op->parent != orig_parent);
1181 BUG_ON(extent_op->generation != orig_generation);
25179201 1182
31840ae1
ZY		 1183 extent_op->parent = parent;
1184 extent_op->generation = ref_generation;
1185 } else {
1186 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
1187 BUG_ON(!extent_op);
1188
1189 extent_op->type = PENDING_BACKREF_UPDATE;
1190 extent_op->bytenr = bytenr;
1191 extent_op->num_bytes = num_bytes;
1192 extent_op->parent = parent;
1193 extent_op->orig_parent = orig_parent;
1194 extent_op->generation = ref_generation;
1195 extent_op->orig_generation = orig_generation;
1196 extent_op->level = (int)owner_objectid;
f3465ca4
JB		 1197 INIT_LIST_HEAD(&extent_op->list);
1198 extent_op->del = 0;
31840ae1
ZY		 1199
1200 set_extent_bits(&root->fs_info->extent_ins,
1201 bytenr, bytenr + num_bytes - 1,
25179201 1202 EXTENT_WRITEBACK, GFP_NOFS);
31840ae1
ZY		 1203 set_state_private(&root->fs_info->extent_ins,
1204 bytenr, (unsigned long)extent_op);
1205 }
25179201 1206 mutex_unlock(&root->fs_info->extent_ins_mutex);
31840ae1
ZY		 1207 return 0;
1208 }
1209
1210 path = btrfs_alloc_path();
1211 if (!path)
1212 return -ENOMEM;
1213 ret = lookup_extent_backref(trans, extent_root, path,
1214 bytenr, orig_parent, orig_root,
3bb1a1bc 1215 orig_generation, owner_objectid, 1);
31840ae1
ZY		 1216 if (ret)
1217 goto out;
1218 ret = remove_extent_backref(trans, extent_root, path);
1219 if (ret)
1220 goto out;
1221 ret = insert_extent_backref(trans, extent_root, path, bytenr,
1222 parent, ref_root, ref_generation,
3bb1a1bc 1223 owner_objectid);
31840ae1 1224 BUG_ON(ret);
87ef2bb4
CM		 1225 finish_current_insert(trans, extent_root, 0);
1226 del_pending_extents(trans, extent_root, 0);
31840ae1
ZY		 1227out:
1228 btrfs_free_path(path);
1229 return ret;
1230}
1231
1232int btrfs_update_extent_ref(struct btrfs_trans_handle *trans,
1233 struct btrfs_root *root, u64 bytenr,
1234 u64 orig_parent, u64 parent,
1235 u64 ref_root, u64 ref_generation,
3bb1a1bc 1236 u64 owner_objectid)
31840ae1
ZY		 1237{
1238 int ret;
1239 if (ref_root == BTRFS_TREE_LOG_OBJECTID &&
1240 owner_objectid < BTRFS_FIRST_FREE_OBJECTID)
1241 return 0;
31840ae1
ZY		 1242 ret = __btrfs_update_extent_ref(trans, root, bytenr, orig_parent,
1243 parent, ref_root, ref_root,
1244 ref_generation, ref_generation,
3bb1a1bc 1245 owner_objectid);
31840ae1
ZY		 1246 return ret;
1247}
1248
925baedd 1249static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
31840ae1
ZY		 1250 struct btrfs_root *root, u64 bytenr,
1251 u64 orig_parent, u64 parent,
1252 u64 orig_root, u64 ref_root,
1253 u64 orig_generation, u64 ref_generation,
3bb1a1bc 1254 u64 owner_objectid)
02217ed2 1255{
5caf2a00 1256 struct btrfs_path *path;
02217ed2 1257 int ret;
e2fa7227 1258 struct btrfs_key key;
5f39d397 1259 struct extent_buffer *l;
234b63a0 1260 struct btrfs_extent_item *item;
cf27e1ee 1261 u32 refs;
037e6390 1262
5caf2a00 1263 path = btrfs_alloc_path();
54aa1f4d
CM		 1264 if (!path)
1265 return -ENOMEM;
26b8003f 1266
3c12ac72 1267 path->reada = 1;
db94535d 1268 key.objectid = bytenr;
31840ae1
ZY		 1269 key.type = BTRFS_EXTENT_ITEM_KEY;
1270 key.offset = (u64)-1;
1271
5caf2a00 1272 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path,
9f5fae2f 1273 0, 1);
54aa1f4d
CM		 1274 if (ret < 0)
1275 return ret;
31840ae1
ZY		 1276 BUG_ON(ret == 0 || path->slots[0] == 0);
1277
1278 path->slots[0]--;
5f39d397 1279 l = path->nodes[0];
31840ae1
ZY		 1280
1281 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
771ed689
CM		 1282 if (key.objectid != bytenr) {
1283 btrfs_print_leaf(root->fs_info->extent_root, path->nodes[0]);
d397712b
CM		 1284 printk(KERN_ERR "btrfs wanted %llu found %llu\n",
1285 (unsigned long long)bytenr,
1286 (unsigned long long)key.objectid);
771ed689
CM		 1287 BUG();
1288 }
31840ae1
ZY		 1289 BUG_ON(key.type != BTRFS_EXTENT_ITEM_KEY);
1290
5caf2a00 1291 item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
5f39d397
CM		 1292 refs = btrfs_extent_refs(l, item);
1293 btrfs_set_extent_refs(l, item, refs + 1);
5caf2a00 1294 btrfs_mark_buffer_dirty(path->nodes[0]);
a28ec197 1295
5caf2a00 1296 btrfs_release_path(root->fs_info->extent_root, path);
7bb86316 1297
3c12ac72 1298 path->reada = 1;
31840ae1
ZY		 1299 ret = insert_extent_backref(trans, root->fs_info->extent_root,
1300 path, bytenr, parent,
1301 ref_root, ref_generation,
3bb1a1bc 1302 owner_objectid);
7bb86316 1303 BUG_ON(ret);
87ef2bb4
CM		 1304 finish_current_insert(trans, root->fs_info->extent_root, 0);
1305 del_pending_extents(trans, root->fs_info->extent_root, 0);
74493f7a
CM		 1306
1307 btrfs_free_path(path);
02217ed2
CM		 1308 return 0;
1309}
1310
925baedd 1311int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
31840ae1
ZY		 1312 struct btrfs_root *root,
1313 u64 bytenr, u64 num_bytes, u64 parent,
1314 u64 ref_root, u64 ref_generation,
3bb1a1bc 1315 u64 owner_objectid)
925baedd
CM		 1316{
1317 int ret;
31840ae1
ZY		 1318 if (ref_root == BTRFS_TREE_LOG_OBJECTID &&
1319 owner_objectid < BTRFS_FIRST_FREE_OBJECTID)
1320 return 0;
31840ae1
ZY		 1321 ret = __btrfs_inc_extent_ref(trans, root, bytenr, 0, parent,
1322 0, ref_root, 0, ref_generation,
3bb1a1bc 1323 owner_objectid);
925baedd
CM		 1324 return ret;
1325}
1326
e9d0b13b
CM		 1327int btrfs_extent_post_op(struct btrfs_trans_handle *trans,
1328 struct btrfs_root *root)
1329{
eb099670
JB		 1330 u64 start;
1331 u64 end;
1332 int ret;
1333
1334 while(1) {
1335 finish_current_insert(trans, root->fs_info->extent_root, 1);
1336 del_pending_extents(trans, root->fs_info->extent_root, 1);
1337
1338 /* is there more work to do? */
1339 ret = find_first_extent_bit(&root->fs_info->pending_del,
1340 0, &start, &end, EXTENT_WRITEBACK);
1341 if (!ret)
1342 continue;
1343 ret = find_first_extent_bit(&root->fs_info->extent_ins,
1344 0, &start, &end, EXTENT_WRITEBACK);
1345 if (!ret)
1346 continue;
1347 break;
1348 }
e9d0b13b
CM		 1349 return 0;
1350}
1351
31840ae1
ZY		 1352int btrfs_lookup_extent_ref(struct btrfs_trans_handle *trans,
1353 struct btrfs_root *root, u64 bytenr,
1354 u64 num_bytes, u32 *refs)
a28ec197 1355{
5caf2a00 1356 struct btrfs_path *path;
a28ec197 1357 int ret;
e2fa7227 1358 struct btrfs_key key;
5f39d397 1359 struct extent_buffer *l;
234b63a0 1360 struct btrfs_extent_item *item;
5caf2a00 1361
db94535d 1362 WARN_ON(num_bytes < root->sectorsize);
5caf2a00 1363 path = btrfs_alloc_path();
3c12ac72 1364 path->reada = 1;
db94535d
CM		 1365 key.objectid = bytenr;
1366 key.offset = num_bytes;
62e2749e 1367 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
5caf2a00 1368 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path,
9f5fae2f 1369 0, 0);
54aa1f4d
CM		 1370 if (ret < 0)
1371 goto out;
5f39d397
CM		 1372 if (ret != 0) {
1373 btrfs_print_leaf(root, path->nodes[0]);
d397712b
CM		 1374 printk(KERN_INFO "btrfs failed to find block number %llu\n",
1375 (unsigned long long)bytenr);
a28ec197 1376 BUG();
5f39d397
CM		 1377 }
1378 l = path->nodes[0];
5caf2a00 1379 item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
5f39d397 1380 *refs = btrfs_extent_refs(l, item);
54aa1f4d 1381out:
5caf2a00 1382 btrfs_free_path(path);
a28ec197
CM		 1383 return 0;
1384}
1385
80ff3856 1386int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
17d217fe 1387 struct btrfs_root *root, u64 objectid, u64 bytenr)
be20aa9d
CM		 1388{
1389 struct btrfs_root *extent_root = root->fs_info->extent_root;
1390 struct btrfs_path *path;
f321e491
YZ		 1391 struct extent_buffer *leaf;
1392 struct btrfs_extent_ref *ref_item;
1393 struct btrfs_key key;
1394 struct btrfs_key found_key;
80ff3856
YZ		 1395 u64 ref_root;
1396 u64 last_snapshot;
be20aa9d
CM		 1397 u32 nritems;
1398 int ret;
925baedd 1399
be20aa9d 1400 key.objectid = bytenr;
31840ae1 1401 key.offset = (u64)-1;
f321e491 1402 key.type = BTRFS_EXTENT_ITEM_KEY;
be20aa9d 1403
f321e491 1404 path = btrfs_alloc_path();
be20aa9d
CM		 1405 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
1406 if (ret < 0)
1407 goto out;
1408 BUG_ON(ret == 0);
80ff3856
YZ		 1409
1410 ret = -ENOENT;
1411 if (path->slots[0] == 0)
31840ae1 1412 goto out;
be20aa9d 1413
31840ae1 1414 path->slots[0]--;
f321e491
YZ		 1415 leaf = path->nodes[0];
1416 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
be20aa9d
CM		 1417
1418 if (found_key.objectid != bytenr ||
80ff3856 1419 found_key.type != BTRFS_EXTENT_ITEM_KEY)
be20aa9d 1420 goto out;
f321e491 1421
80ff3856 1422 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
be20aa9d 1423 while (1) {
f321e491
YZ		 1424 leaf = path->nodes[0];
1425 nritems = btrfs_header_nritems(leaf);
be20aa9d
CM		 1426 if (path->slots[0] >= nritems) {
1427 ret = btrfs_next_leaf(extent_root, path);
f321e491
YZ		 1428 if (ret < 0)
1429 goto out;
be20aa9d
CM		 1430 if (ret == 0)
1431 continue;
1432 break;
1433 }
f321e491 1434 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
be20aa9d
CM		 1435 if (found_key.objectid != bytenr)
1436 break;
bd09835d 1437
be20aa9d
CM		 1438 if (found_key.type != BTRFS_EXTENT_REF_KEY) {
1439 path->slots[0]++;
1440 continue;
1441 }
1442
f321e491 1443 ref_item = btrfs_item_ptr(leaf, path->slots[0],
be20aa9d 1444 struct btrfs_extent_ref);
80ff3856 1445 ref_root = btrfs_ref_root(leaf, ref_item);
17d217fe
YZ		 1446 if ((ref_root != root->root_key.objectid &&
1447 ref_root != BTRFS_TREE_LOG_OBJECTID) ||
1448 objectid != btrfs_ref_objectid(leaf, ref_item)) {
80ff3856 1449 ret = 1;
f321e491 1450 goto out;
80ff3856
YZ		 1451 }
1452 if (btrfs_ref_generation(leaf, ref_item) <= last_snapshot) {
f321e491
YZ		 1453 ret = 1;
1454 goto out;
1455 }
80ff3856
YZ		 1456
1457 path->slots[0]++;
f321e491
YZ		 1458 }
1459 ret = 0;
be20aa9d 1460out:
80ff3856 1461 btrfs_free_path(path);
f321e491 1462 return ret;
be20aa9d 1463}
c5739bba 1464
31840ae1
ZY		 1465int btrfs_cache_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1466 struct extent_buffer *buf, u32 nr_extents)
02217ed2 1467{
5f39d397 1468 struct btrfs_key key;
6407bf6d 1469 struct btrfs_file_extent_item *fi;
e4657689
ZY		 1470 u64 root_gen;
1471 u32 nritems;
02217ed2 1472 int i;
db94535d 1473 int level;
31840ae1 1474 int ret = 0;
e4657689 1475 int shared = 0;
a28ec197 1476
3768f368 1477 if (!root->ref_cows)
a28ec197 1478 return 0;
5f39d397 1479
e4657689
ZY		 1480 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
1481 shared = 0;
1482 root_gen = root->root_key.offset;
1483 } else {
1484 shared = 1;
1485 root_gen = trans->transid - 1;
1486 }
1487
db94535d 1488 level = btrfs_header_level(buf);
5f39d397 1489 nritems = btrfs_header_nritems(buf);
4a096752 1490
31840ae1 1491 if (level == 0) {
31153d81
YZ		 1492 struct btrfs_leaf_ref *ref;
1493 struct btrfs_extent_info *info;
1494
31840ae1 1495 ref = btrfs_alloc_leaf_ref(root, nr_extents);
31153d81 1496 if (!ref) {
31840ae1 1497 ret = -ENOMEM;
31153d81
YZ		 1498 goto out;
1499 }
1500
e4657689 1501 ref->root_gen = root_gen;
31153d81
YZ		 1502 ref->bytenr = buf->start;
1503 ref->owner = btrfs_header_owner(buf);
1504 ref->generation = btrfs_header_generation(buf);
31840ae1 1505 ref->nritems = nr_extents;
31153d81 1506 info = ref->extents;
bcc63abb 1507
31840ae1 1508 for (i = 0; nr_extents > 0 && i < nritems; i++) {
31153d81
YZ		 1509 u64 disk_bytenr;
1510 btrfs_item_key_to_cpu(buf, &key, i);
1511 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
1512 continue;
1513 fi = btrfs_item_ptr(buf, i,
1514 struct btrfs_file_extent_item);
1515 if (btrfs_file_extent_type(buf, fi) ==
1516 BTRFS_FILE_EXTENT_INLINE)
1517 continue;
1518 disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
1519 if (disk_bytenr == 0)
1520 continue;
1521
1522 info->bytenr = disk_bytenr;
1523 info->num_bytes =
1524 btrfs_file_extent_disk_num_bytes(buf, fi);
1525 info->objectid = key.objectid;
1526 info->offset = key.offset;
1527 info++;
1528 }
1529
e4657689 1530 ret = btrfs_add_leaf_ref(root, ref, shared);
5b84e8d6
YZ		 1531 if (ret == -EEXIST && shared) {
1532 struct btrfs_leaf_ref *old;
1533 old = btrfs_lookup_leaf_ref(root, ref->bytenr);
1534 BUG_ON(!old);
1535 btrfs_remove_leaf_ref(root, old);
1536 btrfs_free_leaf_ref(root, old);
1537 ret = btrfs_add_leaf_ref(root, ref, shared);
1538 }
31153d81 1539 WARN_ON(ret);
bcc63abb 1540 btrfs_free_leaf_ref(root, ref);
31153d81
YZ		 1541 }
1542out:
31840ae1
ZY		 1543 return ret;
1544}
1545
b7a9f29f
CM		 1546/* when a block goes through cow, we update the reference counts of
1547 * everything that block points to. The internal pointers of the block
1548 * can be in just about any order, and it is likely to have clusters of
1549 * things that are close together and clusters of things that are not.
1550 *
1551 * To help reduce the seeks that come with updating all of these reference
1552 * counts, sort them by byte number before actual updates are done.
1553 *
1554 * struct refsort is used to match byte number to slot in the btree block.
1555 * we sort based on the byte number and then use the slot to actually
1556 * find the item.
bd56b302
CM		 1557 *
1558 * struct refsort is smaller than strcut btrfs_item and smaller than
1559 * struct btrfs_key_ptr. Since we're currently limited to the page size
1560 * for a btree block, there's no way for a kmalloc of refsorts for a
1561 * single node to be bigger than a page.
b7a9f29f
CM		 1562 */
1563struct refsort {
1564 u64 bytenr;
1565 u32 slot;
1566};
1567
1568/*
1569 * for passing into sort()
1570 */
1571static int refsort_cmp(const void *a_void, const void *b_void)
1572{
1573 const struct refsort *a = a_void;
1574 const struct refsort *b = b_void;
1575
1576 if (a->bytenr < b->bytenr)
1577 return -1;
1578 if (a->bytenr > b->bytenr)
1579 return 1;
1580 return 0;
1581}
1582
1583
1584noinline int btrfs_inc_ref(struct btrfs_trans_handle *trans,
1585 struct btrfs_root *root,
1586 struct extent_buffer *orig_buf,
1587 struct extent_buffer *buf, u32 *nr_extents)
31840ae1
ZY		 1588{
1589 u64 bytenr;
1590 u64 ref_root;
1591 u64 orig_root;
1592 u64 ref_generation;
1593 u64 orig_generation;
b7a9f29f 1594 struct refsort *sorted;
31840ae1
ZY		 1595 u32 nritems;
1596 u32 nr_file_extents = 0;
1597 struct btrfs_key key;
1598 struct btrfs_file_extent_item *fi;
1599 int i;
1600 int level;
1601 int ret = 0;
1602 int faili = 0;
b7a9f29f
CM		 1603 int refi = 0;
1604 int slot;
31840ae1 1605 int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
3bb1a1bc 1606 u64, u64, u64, u64, u64, u64, u64, u64);
31840ae1
ZY		 1607
1608 ref_root = btrfs_header_owner(buf);
1609 ref_generation = btrfs_header_generation(buf);
1610 orig_root = btrfs_header_owner(orig_buf);
1611 orig_generation = btrfs_header_generation(orig_buf);
1612
1613 nritems = btrfs_header_nritems(buf);
1614 level = btrfs_header_level(buf);
1615
b7a9f29f
CM		 1616 sorted = kmalloc(sizeof(struct refsort) * nritems, GFP_NOFS);
1617 BUG_ON(!sorted);
1618
31840ae1
ZY		 1619 if (root->ref_cows) {
1620 process_func = __btrfs_inc_extent_ref;
1621 } else {
1622 if (level == 0 &&
1623 root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
1624 goto out;
1625 if (level != 0 &&
1626 root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID)
1627 goto out;
1628 process_func = __btrfs_update_extent_ref;
1629 }
1630
b7a9f29f
CM		 1631 /*
1632 * we make two passes through the items. In the first pass we
1633 * only record the byte number and slot. Then we sort based on
1634 * byte number and do the actual work based on the sorted results
1635 */
31840ae1
ZY		 1636 for (i = 0; i < nritems; i++) {
1637 cond_resched();
db94535d 1638 if (level == 0) {
5f39d397
CM		 1639 btrfs_item_key_to_cpu(buf, &key, i);
1640 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
54aa1f4d 1641 continue;
5f39d397 1642 fi = btrfs_item_ptr(buf, i,
54aa1f4d 1643 struct btrfs_file_extent_item);
5f39d397 1644 if (btrfs_file_extent_type(buf, fi) ==
54aa1f4d
CM		 1645 BTRFS_FILE_EXTENT_INLINE)
1646 continue;
31840ae1
ZY		 1647 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
1648 if (bytenr == 0)
54aa1f4d 1649 continue;
31840ae1
ZY		 1650
1651 nr_file_extents++;
b7a9f29f
CM		 1652 sorted[refi].bytenr = bytenr;
1653 sorted[refi].slot = i;
1654 refi++;
1655 } else {
1656 bytenr = btrfs_node_blockptr(buf, i);
1657 sorted[refi].bytenr = bytenr;
1658 sorted[refi].slot = i;
1659 refi++;
1660 }
1661 }
1662 /*
1663 * if refi == 0, we didn't actually put anything into the sorted
1664 * array and we're done
1665 */
1666 if (refi == 0)
1667 goto out;
1668
1669 sort(sorted, refi, sizeof(struct refsort), refsort_cmp, NULL);
1670
1671 for (i = 0; i < refi; i++) {
1672 cond_resched();
1673 slot = sorted[i].slot;
1674 bytenr = sorted[i].bytenr;
1675
1676 if (level == 0) {
1677 btrfs_item_key_to_cpu(buf, &key, slot);
31840ae1 1678
31840ae1
ZY		 1679 ret = process_func(trans, root, bytenr,
1680 orig_buf->start, buf->start,
1681 orig_root, ref_root,
1682 orig_generation, ref_generation,
3bb1a1bc 1683 key.objectid);
31840ae1
ZY		 1684
1685 if (ret) {
b7a9f29f 1686 faili = slot;
31840ae1
ZY		 1687 WARN_ON(1);
1688 goto fail;
1689 }
54aa1f4d 1690 } else {
31840ae1
ZY		 1691 ret = process_func(trans, root, bytenr,
1692 orig_buf->start, buf->start,
1693 orig_root, ref_root,
1694 orig_generation, ref_generation,
3bb1a1bc 1695 level - 1);
31840ae1 1696 if (ret) {
b7a9f29f 1697 faili = slot;
31840ae1
ZY		 1698 WARN_ON(1);
1699 goto fail;
1700 }
54aa1f4d
CM		 1701 }
1702 }
31840ae1 1703out:
b7a9f29f 1704 kfree(sorted);
31840ae1
ZY		 1705 if (nr_extents) {
1706 if (level == 0)
1707 *nr_extents = nr_file_extents;
1708 else
1709 *nr_extents = nritems;
1710 }
1711 return 0;
1712fail:
b7a9f29f 1713 kfree(sorted);
31840ae1 1714 WARN_ON(1);
54aa1f4d 1715 return ret;
02217ed2
CM		 1716}
1717
31840ae1
ZY		 1718int btrfs_update_ref(struct btrfs_trans_handle *trans,
1719 struct btrfs_root *root, struct extent_buffer *orig_buf,
1720 struct extent_buffer *buf, int start_slot, int nr)
1721
1722{
1723 u64 bytenr;
1724 u64 ref_root;
1725 u64 orig_root;
1726 u64 ref_generation;
1727 u64 orig_generation;
1728 struct btrfs_key key;
1729 struct btrfs_file_extent_item *fi;
1730 int i;
1731 int ret;
1732 int slot;
1733 int level;
1734
1735 BUG_ON(start_slot < 0);
1736 BUG_ON(start_slot + nr > btrfs_header_nritems(buf));
1737
1738 ref_root = btrfs_header_owner(buf);
1739 ref_generation = btrfs_header_generation(buf);
1740 orig_root = btrfs_header_owner(orig_buf);
1741 orig_generation = btrfs_header_generation(orig_buf);
1742 level = btrfs_header_level(buf);
1743
1744 if (!root->ref_cows) {
1745 if (level == 0 &&
1746 root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
1747 return 0;
1748 if (level != 0 &&
1749 root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID)
1750 return 0;
1751 }
1752
1753 for (i = 0, slot = start_slot; i < nr; i++, slot++) {
1754 cond_resched();
1755 if (level == 0) {
1756 btrfs_item_key_to_cpu(buf, &key, slot);
1757 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
1758 continue;
1759 fi = btrfs_item_ptr(buf, slot,
1760 struct btrfs_file_extent_item);
1761 if (btrfs_file_extent_type(buf, fi) ==
1762 BTRFS_FILE_EXTENT_INLINE)
1763 continue;
1764 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
1765 if (bytenr == 0)
1766 continue;
31840ae1
ZY		 1767 ret = __btrfs_update_extent_ref(trans, root, bytenr,
1768 orig_buf->start, buf->start,
1769 orig_root, ref_root,
1770 orig_generation, ref_generation,
3bb1a1bc 1771 key.objectid);
31840ae1
ZY		 1772 if (ret)
1773 goto fail;
1774 } else {
1775 bytenr = btrfs_node_blockptr(buf, slot);
31840ae1
ZY		 1776 ret = __btrfs_update_extent_ref(trans, root, bytenr,
1777 orig_buf->start, buf->start,
1778 orig_root, ref_root,
1779 orig_generation, ref_generation,
3bb1a1bc 1780 level - 1);
31840ae1
ZY		 1781 if (ret)
1782 goto fail;
1783 }
1784 }
1785 return 0;
1786fail:
1787 WARN_ON(1);
1788 return -1;
1789}
1790
9078a3e1
CM		 1791static int write_one_cache_group(struct btrfs_trans_handle *trans,
1792 struct btrfs_root *root,
1793 struct btrfs_path *path,
1794 struct btrfs_block_group_cache *cache)
1795{
1796 int ret;
1797 int pending_ret;
1798 struct btrfs_root *extent_root = root->fs_info->extent_root;
5f39d397
CM		 1799 unsigned long bi;
1800 struct extent_buffer *leaf;
9078a3e1 1801
9078a3e1 1802 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
54aa1f4d
CM		 1803 if (ret < 0)
1804 goto fail;
9078a3e1 1805 BUG_ON(ret);
5f39d397
CM		 1806
1807 leaf = path->nodes[0];
1808 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
1809 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
1810 btrfs_mark_buffer_dirty(leaf);
9078a3e1 1811 btrfs_release_path(extent_root, path);
54aa1f4d 1812fail:
87ef2bb4
CM		 1813 finish_current_insert(trans, extent_root, 0);
1814 pending_ret = del_pending_extents(trans, extent_root, 0);
9078a3e1
CM		 1815 if (ret)
1816 return ret;
1817 if (pending_ret)
1818 return pending_ret;
1819 return 0;
1820
1821}
1822
96b5179d
CM		 1823int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
1824 struct btrfs_root *root)
9078a3e1 1825{
0f9dd46c
JB		 1826 struct btrfs_block_group_cache *cache, *entry;
1827 struct rb_node *n;
9078a3e1
CM		 1828 int err = 0;
1829 int werr = 0;
9078a3e1 1830 struct btrfs_path *path;
96b5179d 1831 u64 last = 0;
9078a3e1
CM		 1832
1833 path = btrfs_alloc_path();
1834 if (!path)
1835 return -ENOMEM;
1836
d397712b 1837 while (1) {
0f9dd46c
JB		 1838 cache = NULL;
1839 spin_lock(&root->fs_info->block_group_cache_lock);
1840 for (n = rb_first(&root->fs_info->block_group_cache_tree);
1841 n; n = rb_next(n)) {
1842 entry = rb_entry(n, struct btrfs_block_group_cache,
1843 cache_node);
1844 if (entry->dirty) {
1845 cache = entry;
1846 break;
1847 }
1848 }
1849 spin_unlock(&root->fs_info->block_group_cache_lock);
54aa1f4d 1850
0f9dd46c 1851 if (!cache)
96b5179d 1852 break;
0f9dd46c 1853
e8569813 1854 cache->dirty = 0;
0f9dd46c
JB		 1855 last += cache->key.offset;
1856
96b5179d
CM		 1857 err = write_one_cache_group(trans, root,
1858 path, cache);
1859 /*
1860 * if we fail to write the cache group, we want
1861 * to keep it marked dirty in hopes that a later
1862 * write will work
1863 */
1864 if (err) {
1865 werr = err;
1866 continue;
9078a3e1
CM		 1867 }
1868 }
1869 btrfs_free_path(path);
1870 return werr;
1871}
1872
d2fb3437
YZ		 1873int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
1874{
1875 struct btrfs_block_group_cache *block_group;
1876 int readonly = 0;
1877
1878 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
1879 if (!block_group || block_group->ro)
1880 readonly = 1;
1881 if (block_group)
1882 put_block_group(block_group);
1883 return readonly;
1884}
1885
593060d7
CM		 1886static int update_space_info(struct btrfs_fs_info *info, u64 flags,
1887 u64 total_bytes, u64 bytes_used,
1888 struct btrfs_space_info **space_info)
1889{
1890 struct btrfs_space_info *found;
1891
1892 found = __find_space_info(info, flags);
1893 if (found) {
25179201 1894 spin_lock(&found->lock);
593060d7
CM		 1895 found->total_bytes += total_bytes;
1896 found->bytes_used += bytes_used;
8f18cf13 1897 found->full = 0;
25179201 1898 spin_unlock(&found->lock);
593060d7
CM		 1899 *space_info = found;
1900 return 0;
1901 }
c146afad 1902 found = kzalloc(sizeof(*found), GFP_NOFS);
593060d7
CM		 1903 if (!found)
1904 return -ENOMEM;
1905
1906 list_add(&found->list, &info->space_info);
0f9dd46c 1907 INIT_LIST_HEAD(&found->block_groups);
80eb234a 1908 init_rwsem(&found->groups_sem);
0f9dd46c 1909 spin_lock_init(&found->lock);
593060d7
CM		 1910 found->flags = flags;
1911 found->total_bytes = total_bytes;
1912 found->bytes_used = bytes_used;
1913 found->bytes_pinned = 0;
e8569813 1914 found->bytes_reserved = 0;
c146afad 1915 found->bytes_readonly = 0;
6a63209f 1916 found->bytes_delalloc = 0;
593060d7 1917 found->full = 0;
0ef3e66b 1918 found->force_alloc = 0;
593060d7
CM		 1919 *space_info = found;
1920 return 0;
1921}
1922
8790d502
CM		 1923static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
1924{
1925 u64 extra_flags = flags & (BTRFS_BLOCK_GROUP_RAID0 |
611f0e00 1926 BTRFS_BLOCK_GROUP_RAID1 |
321aecc6 1927 BTRFS_BLOCK_GROUP_RAID10 |
611f0e00 1928 BTRFS_BLOCK_GROUP_DUP);
8790d502
CM		 1929 if (extra_flags) {
1930 if (flags & BTRFS_BLOCK_GROUP_DATA)
1931 fs_info->avail_data_alloc_bits |= extra_flags;
1932 if (flags & BTRFS_BLOCK_GROUP_METADATA)
1933 fs_info->avail_metadata_alloc_bits |= extra_flags;
1934 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
1935 fs_info->avail_system_alloc_bits |= extra_flags;
1936 }
1937}
593060d7 1938
c146afad
YZ		 1939static void set_block_group_readonly(struct btrfs_block_group_cache *cache)
1940{
1941 spin_lock(&cache->space_info->lock);
1942 spin_lock(&cache->lock);
1943 if (!cache->ro) {
1944 cache->space_info->bytes_readonly += cache->key.offset -
1945 btrfs_block_group_used(&cache->item);
1946 cache->ro = 1;
1947 }
1948 spin_unlock(&cache->lock);
1949 spin_unlock(&cache->space_info->lock);
1950}
1951
2b82032c 1952u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
ec44a35c 1953{
2b82032c 1954 u64 num_devices = root->fs_info->fs_devices->rw_devices;
a061fc8d
CM		 1955
1956 if (num_devices == 1)
1957 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0);
1958 if (num_devices < 4)
1959 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
1960
ec44a35c
CM		 1961 if ((flags & BTRFS_BLOCK_GROUP_DUP) &&
1962 (flags & (BTRFS_BLOCK_GROUP_RAID1 |
a061fc8d 1963 BTRFS_BLOCK_GROUP_RAID10))) {
ec44a35c 1964 flags &= ~BTRFS_BLOCK_GROUP_DUP;
a061fc8d 1965 }
ec44a35c
CM		 1966
1967 if ((flags & BTRFS_BLOCK_GROUP_RAID1) &&
a061fc8d 1968 (flags & BTRFS_BLOCK_GROUP_RAID10)) {
ec44a35c 1969 flags &= ~BTRFS_BLOCK_GROUP_RAID1;
a061fc8d 1970 }
ec44a35c
CM		 1971
1972 if ((flags & BTRFS_BLOCK_GROUP_RAID0) &&
1973 ((flags & BTRFS_BLOCK_GROUP_RAID1) |
1974 (flags & BTRFS_BLOCK_GROUP_RAID10) |
1975 (flags & BTRFS_BLOCK_GROUP_DUP)))
1976 flags &= ~BTRFS_BLOCK_GROUP_RAID0;
1977 return flags;
1978}
1979
6a63209f
JB		 1980static u64 btrfs_get_alloc_profile(struct btrfs_root *root, u64 data)
1981{
1982 struct btrfs_fs_info *info = root->fs_info;
1983 u64 alloc_profile;
1984
1985 if (data) {
1986 alloc_profile = info->avail_data_alloc_bits &
1987 info->data_alloc_profile;
1988 data = BTRFS_BLOCK_GROUP_DATA | alloc_profile;
1989 } else if (root == root->fs_info->chunk_root) {
1990 alloc_profile = info->avail_system_alloc_bits &
1991 info->system_alloc_profile;
1992 data = BTRFS_BLOCK_GROUP_SYSTEM | alloc_profile;
1993 } else {
1994 alloc_profile = info->avail_metadata_alloc_bits &
1995 info->metadata_alloc_profile;
1996 data = BTRFS_BLOCK_GROUP_METADATA | alloc_profile;
1997 }
1998
1999 return btrfs_reduce_alloc_profile(root, data);
2000}
2001
2002void btrfs_set_inode_space_info(struct btrfs_root *root, struct inode *inode)
2003{
2004 u64 alloc_target;
2005
2006 alloc_target = btrfs_get_alloc_profile(root, 1);
2007 BTRFS_I(inode)->space_info = __find_space_info(root->fs_info,
2008 alloc_target);
2009}
2010
2011/*
2012 * for now this just makes sure we have at least 5% of our metadata space free
2013 * for use.
2014 */
2015int btrfs_check_metadata_free_space(struct btrfs_root *root)
2016{
2017 struct btrfs_fs_info *info = root->fs_info;
2018 struct btrfs_space_info *meta_sinfo;
2019 u64 alloc_target, thresh;
4e06bdd6 2020 int committed = 0, ret;
6a63209f
JB		 2021
2022 /* get the space info for where the metadata will live */
2023 alloc_target = btrfs_get_alloc_profile(root, 0);
2024 meta_sinfo = __find_space_info(info, alloc_target);
2025
4e06bdd6 2026again:
6a63209f 2027 spin_lock(&meta_sinfo->lock);
4e06bdd6
JB		 2028 if (!meta_sinfo->full)
2029 thresh = meta_sinfo->total_bytes * 80;
2030 else
2031 thresh = meta_sinfo->total_bytes * 95;
6a63209f
JB		 2032
2033 do_div(thresh, 100);
2034
2035 if (meta_sinfo->bytes_used + meta_sinfo->bytes_reserved +
2036 meta_sinfo->bytes_pinned + meta_sinfo->bytes_readonly > thresh) {
4e06bdd6
JB		 2037 struct btrfs_trans_handle *trans;
2038 if (!meta_sinfo->full) {
2039 meta_sinfo->force_alloc = 1;
2040 spin_unlock(&meta_sinfo->lock);
2041
2042 trans = btrfs_start_transaction(root, 1);
2043 if (!trans)
2044 return -ENOMEM;
2045
2046 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
2047 2 * 1024 * 1024, alloc_target, 0);
2048 btrfs_end_transaction(trans, root);
2049 goto again;
2050 }
6a63209f 2051 spin_unlock(&meta_sinfo->lock);
4e06bdd6
JB		 2052
2053 if (!committed) {
2054 committed = 1;
2055 trans = btrfs_join_transaction(root, 1);
2056 if (!trans)
2057 return -ENOMEM;
2058 ret = btrfs_commit_transaction(trans, root);
2059 if (ret)
2060 return ret;
2061 goto again;
2062 }
6a63209f
JB		 2063 return -ENOSPC;
2064 }
2065 spin_unlock(&meta_sinfo->lock);
2066
2067 return 0;
2068}
2069
2070/*
2071 * This will check the space that the inode allocates from to make sure we have
2072 * enough space for bytes.
2073 */
2074int btrfs_check_data_free_space(struct btrfs_root *root, struct inode *inode,
2075 u64 bytes)
2076{
2077 struct btrfs_space_info *data_sinfo;
4e06bdd6 2078 int ret = 0, committed = 0;
6a63209f
JB		 2079
2080 /* make sure bytes are sectorsize aligned */
2081 bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
2082
2083 data_sinfo = BTRFS_I(inode)->space_info;
2084again:
2085 /* make sure we have enough space to handle the data first */
2086 spin_lock(&data_sinfo->lock);
2087 if (data_sinfo->total_bytes - data_sinfo->bytes_used -
2088 data_sinfo->bytes_delalloc - data_sinfo->bytes_reserved -
2089 data_sinfo->bytes_pinned - data_sinfo->bytes_readonly -
2090 data_sinfo->bytes_may_use < bytes) {
4e06bdd6
JB		 2091 struct btrfs_trans_handle *trans;
2092
6a63209f
JB		 2093 /*
2094 * if we don't have enough free bytes in this space then we need
2095 * to alloc a new chunk.
2096 */
2097 if (!data_sinfo->full) {
2098 u64 alloc_target;
6a63209f
JB		 2099
2100 data_sinfo->force_alloc = 1;
2101 spin_unlock(&data_sinfo->lock);
2102
2103 alloc_target = btrfs_get_alloc_profile(root, 1);
2104 trans = btrfs_start_transaction(root, 1);
2105 if (!trans)
2106 return -ENOMEM;
2107
2108 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
2109 bytes + 2 * 1024 * 1024,
2110 alloc_target, 0);
2111 btrfs_end_transaction(trans, root);
2112 if (ret)
2113 return ret;
2114 goto again;
2115 }
2116 spin_unlock(&data_sinfo->lock);
4e06bdd6
JB		 2117
2118 /* commit the current transaction and try again */
2119 if (!committed) {
2120 committed = 1;
2121 trans = btrfs_join_transaction(root, 1);
2122 if (!trans)
2123 return -ENOMEM;
2124 ret = btrfs_commit_transaction(trans, root);
2125 if (ret)
2126 return ret;
2127 goto again;
2128 }
2129
6a63209f
JB		 2130 printk(KERN_ERR "no space left, need %llu, %llu delalloc bytes"
2131 ", %llu bytes_used, %llu bytes_reserved, "
2132 "%llu bytes_pinned, %llu bytes_readonly, %llu may use"
2133 "%llu total\n", bytes, data_sinfo->bytes_delalloc,
2134 data_sinfo->bytes_used, data_sinfo->bytes_reserved,
2135 data_sinfo->bytes_pinned, data_sinfo->bytes_readonly,
2136 data_sinfo->bytes_may_use, data_sinfo->total_bytes);
2137 return -ENOSPC;
2138 }
2139 data_sinfo->bytes_may_use += bytes;
2140 BTRFS_I(inode)->reserved_bytes += bytes;
2141 spin_unlock(&data_sinfo->lock);
2142
2143 return btrfs_check_metadata_free_space(root);
2144}
2145
2146/*
2147 * if there was an error for whatever reason after calling
2148 * btrfs_check_data_free_space, call this so we can cleanup the counters.
2149 */
2150void btrfs_free_reserved_data_space(struct btrfs_root *root,
2151 struct inode *inode, u64 bytes)
2152{
2153 struct btrfs_space_info *data_sinfo;
2154
2155 /* make sure bytes are sectorsize aligned */
2156 bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
2157
2158 data_sinfo = BTRFS_I(inode)->space_info;
2159 spin_lock(&data_sinfo->lock);
2160 data_sinfo->bytes_may_use -= bytes;
2161 BTRFS_I(inode)->reserved_bytes -= bytes;
2162 spin_unlock(&data_sinfo->lock);
2163}
2164
2165/* called when we are adding a delalloc extent to the inode's io_tree */
2166void btrfs_delalloc_reserve_space(struct btrfs_root *root, struct inode *inode,
2167 u64 bytes)
2168{
2169 struct btrfs_space_info *data_sinfo;
2170
2171 /* get the space info for where this inode will be storing its data */
2172 data_sinfo = BTRFS_I(inode)->space_info;
2173
2174 /* make sure we have enough space to handle the data first */
2175 spin_lock(&data_sinfo->lock);
2176 data_sinfo->bytes_delalloc += bytes;
2177
2178 /*
2179 * we are adding a delalloc extent without calling
2180 * btrfs_check_data_free_space first. This happens on a weird
2181 * writepage condition, but shouldn't hurt our accounting
2182 */
2183 if (unlikely(bytes > BTRFS_I(inode)->reserved_bytes)) {
2184 data_sinfo->bytes_may_use -= BTRFS_I(inode)->reserved_bytes;
2185 BTRFS_I(inode)->reserved_bytes = 0;
2186 } else {
2187 data_sinfo->bytes_may_use -= bytes;
2188 BTRFS_I(inode)->reserved_bytes -= bytes;
2189 }
2190
2191 spin_unlock(&data_sinfo->lock);
2192}
2193
2194/* called when we are clearing an delalloc extent from the inode's io_tree */
2195void btrfs_delalloc_free_space(struct btrfs_root *root, struct inode *inode,
2196 u64 bytes)
2197{
2198 struct btrfs_space_info *info;
2199
2200 info = BTRFS_I(inode)->space_info;
2201
2202 spin_lock(&info->lock);
2203 info->bytes_delalloc -= bytes;
2204 spin_unlock(&info->lock);
2205}
2206
6324fbf3
CM		 2207static int do_chunk_alloc(struct btrfs_trans_handle *trans,
2208 struct btrfs_root *extent_root, u64 alloc_bytes,
0ef3e66b 2209 u64 flags, int force)
6324fbf3
CM		 2210{
2211 struct btrfs_space_info *space_info;
2212 u64 thresh;
c146afad
YZ		 2213 int ret = 0;
2214
2215 mutex_lock(&extent_root->fs_info->chunk_mutex);
6324fbf3 2216
2b82032c 2217 flags = btrfs_reduce_alloc_profile(extent_root, flags);
ec44a35c 2218
6324fbf3 2219 space_info = __find_space_info(extent_root->fs_info, flags);
593060d7
CM		 2220 if (!space_info) {
2221 ret = update_space_info(extent_root->fs_info, flags,
2222 0, 0, &space_info);
2223 BUG_ON(ret);
2224 }
6324fbf3
CM		 2225 BUG_ON(!space_info);
2226
25179201 2227 spin_lock(&space_info->lock);
0ef3e66b
CM		 2228 if (space_info->force_alloc) {
2229 force = 1;
2230 space_info->force_alloc = 0;
2231 }
25179201
JB		 2232 if (space_info->full) {
2233 spin_unlock(&space_info->lock);
925baedd 2234 goto out;
25179201 2235 }
6324fbf3 2236
c146afad
YZ		 2237 thresh = space_info->total_bytes - space_info->bytes_readonly;
2238 thresh = div_factor(thresh, 6);
0ef3e66b 2239 if (!force &&
e8569813 2240 (space_info->bytes_used + space_info->bytes_pinned +
25179201
JB		 2241 space_info->bytes_reserved + alloc_bytes) < thresh) {
2242 spin_unlock(&space_info->lock);
925baedd 2243 goto out;
25179201 2244 }
25179201
JB		 2245 spin_unlock(&space_info->lock);
2246
2b82032c 2247 ret = btrfs_alloc_chunk(trans, extent_root, flags);
d397712b 2248 if (ret)
6324fbf3 2249 space_info->full = 1;
a74a4b97 2250out:
c146afad 2251 mutex_unlock(&extent_root->fs_info->chunk_mutex);
0f9dd46c 2252 return ret;
6324fbf3
CM		 2253}
2254
9078a3e1
CM		 2255static int update_block_group(struct btrfs_trans_handle *trans,
2256 struct btrfs_root *root,
db94535d 2257 u64 bytenr, u64 num_bytes, int alloc,
0b86a832 2258 int mark_free)
9078a3e1
CM		 2259{
2260 struct btrfs_block_group_cache *cache;
2261 struct btrfs_fs_info *info = root->fs_info;
db94535d 2262 u64 total = num_bytes;
9078a3e1 2263 u64 old_val;
db94535d 2264 u64 byte_in_group;
3e1ad54f 2265
d397712b 2266 while (total) {
db94535d 2267 cache = btrfs_lookup_block_group(info, bytenr);
f3465ca4 2268 if (!cache)
9078a3e1 2269 return -1;
db94535d
CM		 2270 byte_in_group = bytenr - cache->key.objectid;
2271 WARN_ON(byte_in_group > cache->key.offset);
9078a3e1 2272
25179201 2273 spin_lock(&cache->space_info->lock);
c286ac48 2274 spin_lock(&cache->lock);
0f9dd46c 2275 cache->dirty = 1;
9078a3e1 2276 old_val = btrfs_block_group_used(&cache->item);
db94535d 2277 num_bytes = min(total, cache->key.offset - byte_in_group);
cd1bc465 2278 if (alloc) {
db94535d 2279 old_val += num_bytes;
6324fbf3 2280 cache->space_info->bytes_used += num_bytes;
a512bbf8 2281 if (cache->ro)
c146afad 2282 cache->space_info->bytes_readonly -= num_bytes;
c286ac48
CM		 2283 btrfs_set_block_group_used(&cache->item, old_val);
2284 spin_unlock(&cache->lock);
25179201 2285 spin_unlock(&cache->space_info->lock);
cd1bc465 2286 } else {
db94535d 2287 old_val -= num_bytes;
6324fbf3 2288 cache->space_info->bytes_used -= num_bytes;
c146afad
YZ		 2289 if (cache->ro)
2290 cache->space_info->bytes_readonly += num_bytes;
c286ac48
CM		 2291 btrfs_set_block_group_used(&cache->item, old_val);
2292 spin_unlock(&cache->lock);
25179201 2293 spin_unlock(&cache->space_info->lock);
f510cfec 2294 if (mark_free) {
0f9dd46c 2295 int ret;
1f3c79a2
LH		 2296
2297 ret = btrfs_discard_extent(root, bytenr,
2298 num_bytes);
2299 WARN_ON(ret);
2300
0f9dd46c
JB		 2301 ret = btrfs_add_free_space(cache, bytenr,
2302 num_bytes);
d2fb3437 2303 WARN_ON(ret);
e37c9e69 2304 }
cd1bc465 2305 }
d2fb3437 2306 put_block_group(cache);
db94535d
CM		 2307 total -= num_bytes;
2308 bytenr += num_bytes;
9078a3e1
CM		 2309 }
2310 return 0;
2311}
6324fbf3 2312
a061fc8d
CM		 2313static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
2314{
0f9dd46c 2315 struct btrfs_block_group_cache *cache;
d2fb3437 2316 u64 bytenr;
0f9dd46c
JB		 2317
2318 cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
2319 if (!cache)
a061fc8d 2320 return 0;
0f9dd46c 2321
d2fb3437
YZ		 2322 bytenr = cache->key.objectid;
2323 put_block_group(cache);
2324
2325 return bytenr;
a061fc8d
CM		 2326}
2327
e02119d5 2328int btrfs_update_pinned_extents(struct btrfs_root *root,
324ae4df
Y		 2329 u64 bytenr, u64 num, int pin)
2330{
2331 u64 len;
2332 struct btrfs_block_group_cache *cache;
2333 struct btrfs_fs_info *fs_info = root->fs_info;
2334
25179201 2335 WARN_ON(!mutex_is_locked(&root->fs_info->pinned_mutex));
324ae4df
Y		 2336 if (pin) {
2337 set_extent_dirty(&fs_info->pinned_extents,
2338 bytenr, bytenr + num - 1, GFP_NOFS);
2339 } else {
2340 clear_extent_dirty(&fs_info->pinned_extents,
2341 bytenr, bytenr + num - 1, GFP_NOFS);
2342 }
2343 while (num > 0) {
2344 cache = btrfs_lookup_block_group(fs_info, bytenr);
e8569813
ZY		 2345 BUG_ON(!cache);
2346 len = min(num, cache->key.offset -
2347 (bytenr - cache->key.objectid));
324ae4df 2348 if (pin) {
25179201 2349 spin_lock(&cache->space_info->lock);
e8569813
ZY		 2350 spin_lock(&cache->lock);
2351 cache->pinned += len;
2352 cache->space_info->bytes_pinned += len;
2353 spin_unlock(&cache->lock);
25179201 2354 spin_unlock(&cache->space_info->lock);
324ae4df
Y		 2355 fs_info->total_pinned += len;
2356 } else {
25179201 2357 spin_lock(&cache->space_info->lock);
e8569813
ZY		 2358 spin_lock(&cache->lock);
2359 cache->pinned -= len;
2360 cache->space_info->bytes_pinned -= len;
2361 spin_unlock(&cache->lock);
25179201 2362 spin_unlock(&cache->space_info->lock);
324ae4df 2363 fs_info->total_pinned -= len;
07103a3c
JB		 2364 if (cache->cached)
2365 btrfs_add_free_space(cache, bytenr, len);
324ae4df 2366 }
d2fb3437 2367 put_block_group(cache);
324ae4df
Y		 2368 bytenr += len;
2369 num -= len;
2370 }
2371 return 0;
2372}
9078a3e1 2373
e8569813
ZY		 2374static int update_reserved_extents(struct btrfs_root *root,
2375 u64 bytenr, u64 num, int reserve)
2376{
2377 u64 len;
2378 struct btrfs_block_group_cache *cache;
2379 struct btrfs_fs_info *fs_info = root->fs_info;
2380
e8569813
ZY		 2381 while (num > 0) {
2382 cache = btrfs_lookup_block_group(fs_info, bytenr);
2383 BUG_ON(!cache);
2384 len = min(num, cache->key.offset -
2385 (bytenr - cache->key.objectid));
25179201
JB		 2386
2387 spin_lock(&cache->space_info->lock);
2388 spin_lock(&cache->lock);
e8569813 2389 if (reserve) {
e8569813
ZY		 2390 cache->reserved += len;
2391 cache->space_info->bytes_reserved += len;
e8569813 2392 } else {
e8569813
ZY		 2393 cache->reserved -= len;
2394 cache->space_info->bytes_reserved -= len;
e8569813 2395 }
25179201
JB		 2396 spin_unlock(&cache->lock);
2397 spin_unlock(&cache->space_info->lock);
d2fb3437 2398 put_block_group(cache);
e8569813
ZY		 2399 bytenr += len;
2400 num -= len;
2401 }
2402 return 0;
2403}
2404
d1310b2e 2405int btrfs_copy_pinned(struct btrfs_root *root, struct extent_io_tree *copy)
ccd467d6 2406{
ccd467d6 2407 u64 last = 0;
1a5bc167
CM		 2408 u64 start;
2409 u64 end;
d1310b2e 2410 struct extent_io_tree *pinned_extents = &root->fs_info->pinned_extents;
ccd467d6 2411 int ret;
ccd467d6 2412
25179201 2413 mutex_lock(&root->fs_info->pinned_mutex);
d397712b 2414 while (1) {
1a5bc167
CM		 2415 ret = find_first_extent_bit(pinned_extents, last,
2416 &start, &end, EXTENT_DIRTY);
2417 if (ret)
ccd467d6 2418 break;
1a5bc167
CM		 2419 set_extent_dirty(copy, start, end, GFP_NOFS);
2420 last = end + 1;
ccd467d6 2421 }
25179201 2422 mutex_unlock(&root->fs_info->pinned_mutex);
ccd467d6
CM		 2423 return 0;
2424}
2425
2426int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
2427 struct btrfs_root *root,
d1310b2e 2428 struct extent_io_tree *unpin)
a28ec197 2429{
1a5bc167
CM		 2430 u64 start;
2431 u64 end;
a28ec197 2432 int ret;
a28ec197 2433
25179201 2434 mutex_lock(&root->fs_info->pinned_mutex);
d397712b 2435 while (1) {
1a5bc167
CM		 2436 ret = find_first_extent_bit(unpin, 0, &start, &end,
2437 EXTENT_DIRTY);
2438 if (ret)
a28ec197 2439 break;
1f3c79a2
LH		 2440
2441 ret = btrfs_discard_extent(root, start, end + 1 - start);
2442
e02119d5 2443 btrfs_update_pinned_extents(root, start, end + 1 - start, 0);
1a5bc167 2444 clear_extent_dirty(unpin, start, end, GFP_NOFS);
1f3c79a2 2445
c286ac48 2446 if (need_resched()) {
25179201 2447 mutex_unlock(&root->fs_info->pinned_mutex);
c286ac48 2448 cond_resched();
25179201 2449 mutex_lock(&root->fs_info->pinned_mutex);
c286ac48 2450 }
a28ec197 2451 }
25179201 2452 mutex_unlock(&root->fs_info->pinned_mutex);
1f3c79a2 2453 return ret;
a28ec197
CM		 2454}
2455
98ed5174 2456static int finish_current_insert(struct btrfs_trans_handle *trans,
87ef2bb4 2457 struct btrfs_root *extent_root, int all)
037e6390 2458{
7bb86316
CM		 2459 u64 start;
2460 u64 end;
31840ae1 2461 u64 priv;
25179201 2462 u64 search = 0;
7bb86316
CM		 2463 struct btrfs_fs_info *info = extent_root->fs_info;
2464 struct btrfs_path *path;
f3465ca4
JB		 2465 struct pending_extent_op *extent_op, *tmp;
2466 struct list_head insert_list, update_list;
037e6390 2467 int ret;
eb099670 2468 int num_inserts = 0, max_inserts, restart = 0;
037e6390 2469
7bb86316 2470 path = btrfs_alloc_path();
f3465ca4
JB		 2471 INIT_LIST_HEAD(&insert_list);
2472 INIT_LIST_HEAD(&update_list);
037e6390 2473
f3465ca4
JB		 2474 max_inserts = extent_root->leafsize /
2475 (2 * sizeof(struct btrfs_key) + 2 * sizeof(struct btrfs_item) +
2476 sizeof(struct btrfs_extent_ref) +
2477 sizeof(struct btrfs_extent_item));
2478again:
2479 mutex_lock(&info->extent_ins_mutex);
2480 while (1) {
25179201
JB		 2481 ret = find_first_extent_bit(&info->extent_ins, search, &start,
2482 &end, EXTENT_WRITEBACK);
2483 if (ret) {
eb099670 2484 if (restart && !num_inserts &&
5a7be515 2485 list_empty(&update_list)) {
eb099670 2486 restart = 0;
b4eec2ca 2487 search = 0;
25179201
JB		 2488 continue;
2489 }
26b8003f 2490 break;
25179201
JB		 2491 }
2492
2493 ret = try_lock_extent(&info->extent_ins, start, end, GFP_NOFS);
2494 if (!ret) {
eb099670
JB		 2495 if (all)
2496 restart = 1;
87ef2bb4 2497 search = end + 1;
f3465ca4
JB		 2498 if (need_resched()) {
2499 mutex_unlock(&info->extent_ins_mutex);
2500 cond_resched();
2501 mutex_lock(&info->extent_ins_mutex);
2502 }
25179201
JB		 2503 continue;
2504 }
26b8003f 2505
31840ae1
ZY		 2506 ret = get_state_private(&info->extent_ins, start, &priv);
2507 BUG_ON(ret);
f3465ca4 2508 extent_op = (struct pending_extent_op *)(unsigned long) priv;
25179201 2509
31840ae1 2510 if (extent_op->type == PENDING_EXTENT_INSERT) {
f3465ca4
JB		 2511 num_inserts++;
2512 list_add_tail(&extent_op->list, &insert_list);
2513 search = end + 1;
2514 if (num_inserts == max_inserts) {
eb099670 2515 restart = 1;
f3465ca4
JB		 2516 break;
2517 }
2518 } else if (extent_op->type == PENDING_BACKREF_UPDATE) {
2519 list_add_tail(&extent_op->list, &update_list);
2520 search = end + 1;
2521 } else {
2522 BUG();
2523 }
2524 }
c286ac48 2525
f3465ca4 2526 /*
b4eec2ca 2527 * process the update list, clear the writeback bit for it, and if
f3465ca4
JB		 2528 * somebody marked this thing for deletion then just unlock it and be
2529 * done, the free_extents will handle it
2530 */
f3465ca4
JB		 2531 list_for_each_entry_safe(extent_op, tmp, &update_list, list) {
2532 clear_extent_bits(&info->extent_ins, extent_op->bytenr,
2533 extent_op->bytenr + extent_op->num_bytes - 1,
2534 EXTENT_WRITEBACK, GFP_NOFS);
2535 if (extent_op->del) {
2536 list_del_init(&extent_op->list);
2537 unlock_extent(&info->extent_ins, extent_op->bytenr,
2538 extent_op->bytenr + extent_op->num_bytes
2539 - 1, GFP_NOFS);
2540 kfree(extent_op);
2541 }
2542 }
2543 mutex_unlock(&info->extent_ins_mutex);
c286ac48 2544
f3465ca4
JB		 2545 /*
2546 * still have things left on the update list, go ahead an update
2547 * everything
2548 */
2549 if (!list_empty(&update_list)) {
2550 ret = update_backrefs(trans, extent_root, path, &update_list);
2551 BUG_ON(ret);
eb099670
JB		 2552
2553 /* we may have COW'ed new blocks, so lets start over */
2554 if (all)
2555 restart = 1;
f3465ca4 2556 }
c286ac48 2557
f3465ca4
JB		 2558 /*
2559 * if no inserts need to be done, but we skipped some extents and we
2560 * need to make sure everything is cleaned then reset everything and
2561 * go back to the beginning
2562 */
eb099670 2563 if (!num_inserts && restart) {
f3465ca4 2564 search = 0;
eb099670 2565 restart = 0;
f3465ca4
JB		 2566 INIT_LIST_HEAD(&update_list);
2567 INIT_LIST_HEAD(&insert_list);
2568 goto again;
2569 } else if (!num_inserts) {
2570 goto out;
2571 }
31840ae1 2572
f3465ca4
JB		 2573 /*
2574 * process the insert extents list. Again if we are deleting this
2575 * extent, then just unlock it, pin down the bytes if need be, and be
2576 * done with it. Saves us from having to actually insert the extent
2577 * into the tree and then subsequently come along and delete it
2578 */
2579 mutex_lock(&info->extent_ins_mutex);
2580 list_for_each_entry_safe(extent_op, tmp, &insert_list, list) {
2581 clear_extent_bits(&info->extent_ins, extent_op->bytenr,
2582 extent_op->bytenr + extent_op->num_bytes - 1,
2583 EXTENT_WRITEBACK, GFP_NOFS);
2584 if (extent_op->del) {
34bf63c4 2585 u64 used;
f3465ca4
JB		 2586 list_del_init(&extent_op->list);
2587 unlock_extent(&info->extent_ins, extent_op->bytenr,
2588 extent_op->bytenr + extent_op->num_bytes
2589 - 1, GFP_NOFS);
2590
2591 mutex_lock(&extent_root->fs_info->pinned_mutex);
2592 ret = pin_down_bytes(trans, extent_root,
2593 extent_op->bytenr,
2594 extent_op->num_bytes, 0);
2595 mutex_unlock(&extent_root->fs_info->pinned_mutex);
2596
34bf63c4
YZ		 2597 spin_lock(&info->delalloc_lock);
2598 used = btrfs_super_bytes_used(&info->super_copy);
2599 btrfs_set_super_bytes_used(&info->super_copy,
2600 used - extent_op->num_bytes);
2601 used = btrfs_root_used(&extent_root->root_item);
2602 btrfs_set_root_used(&extent_root->root_item,
2603 used - extent_op->num_bytes);
2604 spin_unlock(&info->delalloc_lock);
2605
f3465ca4
JB		 2606 ret = update_block_group(trans, extent_root,
2607 extent_op->bytenr,
2608 extent_op->num_bytes,
2609 0, ret > 0);
2610 BUG_ON(ret);
2611 kfree(extent_op);
2612 num_inserts--;
d8d5f3e1 2613 }
f3465ca4
JB		 2614 }
2615 mutex_unlock(&info->extent_ins_mutex);
31840ae1 2616
f3465ca4
JB		 2617 ret = insert_extents(trans, extent_root, path, &insert_list,
2618 num_inserts);
2619 BUG_ON(ret);
2620
2621 /*
eb099670
JB		 2622 * if restart is set for whatever reason we need to go back and start
2623 * searching through the pending list again.
2624 *
2625 * We just inserted some extents, which could have resulted in new
2626 * blocks being allocated, which would result in new blocks needing
2627 * updates, so if all is set we _must_ restart to get the updated
2628 * blocks.
f3465ca4 2629 */
eb099670 2630 if (restart || all) {
f3465ca4
JB		 2631 INIT_LIST_HEAD(&insert_list);
2632 INIT_LIST_HEAD(&update_list);
2633 search = 0;
eb099670 2634 restart = 0;
f3465ca4
JB		 2635 num_inserts = 0;
2636 goto again;
037e6390 2637 }
f3465ca4 2638out:
7bb86316 2639 btrfs_free_path(path);
037e6390
CM		 2640 return 0;
2641}
2642
31840ae1
ZY		 2643static int pin_down_bytes(struct btrfs_trans_handle *trans,
2644 struct btrfs_root *root,
2645 u64 bytenr, u64 num_bytes, int is_data)
e20d96d6 2646{
1a5bc167 2647 int err = 0;
31840ae1 2648 struct extent_buffer *buf;
8ef97622 2649
31840ae1
ZY		 2650 if (is_data)
2651 goto pinit;
2652
2653 buf = btrfs_find_tree_block(root, bytenr, num_bytes);
2654 if (!buf)
2655 goto pinit;
2656
2657 /* we can reuse a block if it hasn't been written
2658 * and it is from this transaction. We can't
2659 * reuse anything from the tree log root because
2660 * it has tiny sub-transactions.
2661 */
2662 if (btrfs_buffer_uptodate(buf, 0) &&
2663 btrfs_try_tree_lock(buf)) {
2664 u64 header_owner = btrfs_header_owner(buf);
2665 u64 header_transid = btrfs_header_generation(buf);
2666 if (header_owner != BTRFS_TREE_LOG_OBJECTID &&
1a40e23b 2667 header_owner != BTRFS_TREE_RELOC_OBJECTID &&
31840ae1
ZY		 2668 header_transid == trans->transid &&
2669 !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
2670 clean_tree_block(NULL, root, buf);
2671 btrfs_tree_unlock(buf);
5f39d397 2672 free_extent_buffer(buf);
31840ae1 2673 return 1;
8ef97622 2674 }
31840ae1 2675 btrfs_tree_unlock(buf);
f4b9aa8d 2676 }
31840ae1
ZY		 2677 free_extent_buffer(buf);
2678pinit:
2679 btrfs_update_pinned_extents(root, bytenr, num_bytes, 1);
2680
be744175 2681 BUG_ON(err < 0);
e20d96d6
CM		 2682 return 0;
2683}
2684
fec577fb 2685/*
a28ec197 2686 * remove an extent from the root, returns 0 on success
fec577fb 2687 */
31840ae1
ZY		 2688static int __free_extent(struct btrfs_trans_handle *trans,
2689 struct btrfs_root *root,
2690 u64 bytenr, u64 num_bytes, u64 parent,
7bb86316 2691 u64 root_objectid, u64 ref_generation,
3bb1a1bc 2692 u64 owner_objectid, int pin, int mark_free)
a28ec197 2693{
5caf2a00 2694 struct btrfs_path *path;
e2fa7227 2695 struct btrfs_key key;
1261ec42
CM		 2696 struct btrfs_fs_info *info = root->fs_info;
2697 struct btrfs_root *extent_root = info->extent_root;
5f39d397 2698 struct extent_buffer *leaf;
a28ec197 2699 int ret;
952fccac
CM		 2700 int extent_slot = 0;
2701 int found_extent = 0;
2702 int num_to_del = 1;
234b63a0 2703 struct btrfs_extent_item *ei;
cf27e1ee 2704 u32 refs;
037e6390 2705
db94535d 2706 key.objectid = bytenr;
62e2749e 2707 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
db94535d 2708 key.offset = num_bytes;
5caf2a00 2709 path = btrfs_alloc_path();
54aa1f4d
CM		 2710 if (!path)
2711 return -ENOMEM;
5f26f772 2712
3c12ac72 2713 path->reada = 1;
3bb1a1bc
YZ		 2714 ret = lookup_extent_backref(trans, extent_root, path,
2715 bytenr, parent, root_objectid,
2716 ref_generation, owner_objectid, 1);
7bb86316 2717 if (ret == 0) {
952fccac
CM		 2718 struct btrfs_key found_key;
2719 extent_slot = path->slots[0];
d397712b 2720 while (extent_slot > 0) {
952fccac
CM		 2721 extent_slot--;
2722 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
2723 extent_slot);
2724 if (found_key.objectid != bytenr)
2725 break;
2726 if (found_key.type == BTRFS_EXTENT_ITEM_KEY &&
2727 found_key.offset == num_bytes) {
2728 found_extent = 1;
2729 break;
2730 }
2731 if (path->slots[0] - extent_slot > 5)
2732 break;
2733 }
31840ae1
ZY		 2734 if (!found_extent) {
2735 ret = remove_extent_backref(trans, extent_root, path);
2736 BUG_ON(ret);
2737 btrfs_release_path(extent_root, path);
2738 ret = btrfs_search_slot(trans, extent_root,
2739 &key, path, -1, 1);
f3465ca4
JB		 2740 if (ret) {
2741 printk(KERN_ERR "umm, got %d back from search"
d397712b
CM		 2742 ", was looking for %llu\n", ret,
2743 (unsigned long long)bytenr);
f3465ca4
JB		 2744 btrfs_print_leaf(extent_root, path->nodes[0]);
2745 }
31840ae1
ZY		 2746 BUG_ON(ret);
2747 extent_slot = path->slots[0];
2748 }
7bb86316
CM		 2749 } else {
2750 btrfs_print_leaf(extent_root, path->nodes[0]);
2751 WARN_ON(1);
d397712b
CM		 2752 printk(KERN_ERR "btrfs unable to find ref byte nr %llu "
2753 "root %llu gen %llu owner %llu\n",
2754 (unsigned long long)bytenr,
2755 (unsigned long long)root_objectid,
2756 (unsigned long long)ref_generation,
2757 (unsigned long long)owner_objectid);
7bb86316 2758 }
5f39d397
CM		 2759
2760 leaf = path->nodes[0];
952fccac 2761 ei = btrfs_item_ptr(leaf, extent_slot,
123abc88 2762 struct btrfs_extent_item);
5f39d397
CM		 2763 refs = btrfs_extent_refs(leaf, ei);
2764 BUG_ON(refs == 0);
2765 refs -= 1;
2766 btrfs_set_extent_refs(leaf, ei, refs);
952fccac 2767
5f39d397
CM		 2768 btrfs_mark_buffer_dirty(leaf);
2769
952fccac 2770 if (refs == 0 && found_extent && path->slots[0] == extent_slot + 1) {
31840ae1
ZY		 2771 struct btrfs_extent_ref *ref;
2772 ref = btrfs_item_ptr(leaf, path->slots[0],
2773 struct btrfs_extent_ref);
2774 BUG_ON(btrfs_ref_num_refs(leaf, ref) != 1);
952fccac
CM		 2775 /* if the back ref and the extent are next to each other
2776 * they get deleted below in one shot
2777 */
2778 path->slots[0] = extent_slot;
2779 num_to_del = 2;
2780 } else if (found_extent) {
2781 /* otherwise delete the extent back ref */
31840ae1 2782 ret = remove_extent_backref(trans, extent_root, path);
952fccac
CM		 2783 BUG_ON(ret);
2784 /* if refs are 0, we need to setup the path for deletion */
2785 if (refs == 0) {
2786 btrfs_release_path(extent_root, path);
2787 ret = btrfs_search_slot(trans, extent_root, &key, path,
2788 -1, 1);
952fccac
CM		 2789 BUG_ON(ret);
2790 }
2791 }
2792
cf27e1ee 2793 if (refs == 0) {
db94535d
CM		 2794 u64 super_used;
2795 u64 root_used;
78fae27e
CM		 2796
2797 if (pin) {
25179201 2798 mutex_lock(&root->fs_info->pinned_mutex);
31840ae1
ZY		 2799 ret = pin_down_bytes(trans, root, bytenr, num_bytes,
2800 owner_objectid >= BTRFS_FIRST_FREE_OBJECTID);
25179201 2801 mutex_unlock(&root->fs_info->pinned_mutex);
c549228f
Y		 2802 if (ret > 0)
2803 mark_free = 1;
2804 BUG_ON(ret < 0);
78fae27e 2805 }
58176a96 2806 /* block accounting for super block */
75eff68e 2807 spin_lock(&info->delalloc_lock);
db94535d
CM		 2808 super_used = btrfs_super_bytes_used(&info->super_copy);
2809 btrfs_set_super_bytes_used(&info->super_copy,
2810 super_used - num_bytes);
58176a96
JB		 2811
2812 /* block accounting for root item */
db94535d 2813 root_used = btrfs_root_used(&root->root_item);
5f39d397 2814 btrfs_set_root_used(&root->root_item,
db94535d 2815 root_used - num_bytes);
34bf63c4 2816 spin_unlock(&info->delalloc_lock);
952fccac
CM		 2817 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
2818 num_to_del);
31840ae1 2819 BUG_ON(ret);
25179201 2820 btrfs_release_path(extent_root, path);
21af804c 2821
459931ec
CM		 2822 if (owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
2823 ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
2824 BUG_ON(ret);
2825 }
2826
dcbdd4dc
CM		 2827 ret = update_block_group(trans, root, bytenr, num_bytes, 0,
2828 mark_free);
2829 BUG_ON(ret);
a28ec197 2830 }
5caf2a00 2831 btrfs_free_path(path);
87ef2bb4 2832 finish_current_insert(trans, extent_root, 0);
a28ec197
CM		 2833 return ret;
2834}
2835
a28ec197
CM		 2836/*
2837 * find all the blocks marked as pending in the radix tree and remove
2838 * them from the extent map
2839 */
d397712b
CM		 2840static int del_pending_extents(struct btrfs_trans_handle *trans,
2841 struct btrfs_root *extent_root, int all)
a28ec197
CM		 2842{
2843 int ret;
e20d96d6 2844 int err = 0;
1a5bc167
CM		 2845 u64 start;
2846 u64 end;
31840ae1 2847 u64 priv;
25179201 2848 u64 search = 0;
f3465ca4 2849 int nr = 0, skipped = 0;
d1310b2e 2850 struct extent_io_tree *pending_del;
31840ae1
ZY		 2851 struct extent_io_tree *extent_ins;
2852 struct pending_extent_op *extent_op;
25179201 2853 struct btrfs_fs_info *info = extent_root->fs_info;
f3465ca4 2854 struct list_head delete_list;
8ef97622 2855
f3465ca4 2856 INIT_LIST_HEAD(&delete_list);
31840ae1 2857 extent_ins = &extent_root->fs_info->extent_ins;
1a5bc167 2858 pending_del = &extent_root->fs_info->pending_del;
a28ec197 2859
f3465ca4
JB		 2860again:
2861 mutex_lock(&info->extent_ins_mutex);
d397712b 2862 while (1) {
25179201
JB		 2863 ret = find_first_extent_bit(pending_del, search, &start, &end,
2864 EXTENT_WRITEBACK);
2865 if (ret) {
f3465ca4 2866 if (all && skipped && !nr) {
25179201 2867 search = 0;
5a7be515 2868 skipped = 0;
25179201
JB		 2869 continue;
2870 }
f3465ca4 2871 mutex_unlock(&info->extent_ins_mutex);
a28ec197 2872 break;
25179201
JB		 2873 }
2874
2875 ret = try_lock_extent(extent_ins, start, end, GFP_NOFS);
2876 if (!ret) {
2877 search = end+1;
f3465ca4
JB		 2878 skipped = 1;
2879
2880 if (need_resched()) {
2881 mutex_unlock(&info->extent_ins_mutex);
2882 cond_resched();
2883 mutex_lock(&info->extent_ins_mutex);
2884 }
2885
25179201
JB		 2886 continue;
2887 }
2888 BUG_ON(ret < 0);
31840ae1
ZY		 2889
2890 ret = get_state_private(pending_del, start, &priv);
2891 BUG_ON(ret);
2892 extent_op = (struct pending_extent_op *)(unsigned long)priv;
2893
25179201 2894 clear_extent_bits(pending_del, start, end, EXTENT_WRITEBACK,
1a5bc167 2895 GFP_NOFS);
31840ae1 2896 if (!test_range_bit(extent_ins, start, end,
25179201 2897 EXTENT_WRITEBACK, 0)) {
f3465ca4
JB		 2898 list_add_tail(&extent_op->list, &delete_list);
2899 nr++;
c286ac48 2900 } else {
31840ae1 2901 kfree(extent_op);
25179201
JB		 2902
2903 ret = get_state_private(&info->extent_ins, start,
2904 &priv);
31840ae1
ZY		 2905 BUG_ON(ret);
2906 extent_op = (struct pending_extent_op *)
25179201
JB		 2907 (unsigned long)priv;
2908
2909 clear_extent_bits(&info->extent_ins, start, end,
2910 EXTENT_WRITEBACK, GFP_NOFS);
31840ae1 2911
f3465ca4
JB		 2912 if (extent_op->type == PENDING_BACKREF_UPDATE) {
2913 list_add_tail(&extent_op->list, &delete_list);
2914 search = end + 1;
2915 nr++;
2916 continue;
2917 }
31840ae1 2918
25179201
JB		 2919 mutex_lock(&extent_root->fs_info->pinned_mutex);
2920 ret = pin_down_bytes(trans, extent_root, start,
2921 end + 1 - start, 0);
2922 mutex_unlock(&extent_root->fs_info->pinned_mutex);
2923
31840ae1 2924 ret = update_block_group(trans, extent_root, start,
25179201
JB		 2925 end + 1 - start, 0, ret > 0);
2926
f3465ca4 2927 unlock_extent(extent_ins, start, end, GFP_NOFS);
31840ae1
ZY		 2928 BUG_ON(ret);
2929 kfree(extent_op);
c286ac48 2930 }
1a5bc167
CM		 2931 if (ret)
2932 err = ret;
c286ac48 2933
f3465ca4
JB		 2934 search = end + 1;
2935
2936 if (need_resched()) {
2937 mutex_unlock(&info->extent_ins_mutex);
2938 cond_resched();
2939 mutex_lock(&info->extent_ins_mutex);
2940 }
2941 }
2942
2943 if (nr) {
2944 ret = free_extents(trans, extent_root, &delete_list);
2945 BUG_ON(ret);
2946 }
2947
2948 if (all && skipped) {
2949 INIT_LIST_HEAD(&delete_list);
2950 search = 0;
2951 nr = 0;
2952 goto again;
fec577fb 2953 }
f3465ca4 2954
eb099670
JB		 2955 if (!err)
2956 finish_current_insert(trans, extent_root, 0);
e20d96d6 2957 return err;
fec577fb
CM		 2958}
2959
2960/*
2961 * remove an extent from the root, returns 0 on success
2962 */
925baedd 2963static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
31840ae1
ZY		 2964 struct btrfs_root *root,
2965 u64 bytenr, u64 num_bytes, u64 parent,
2966 u64 root_objectid, u64 ref_generation,
3bb1a1bc 2967 u64 owner_objectid, int pin)
fec577fb 2968{
9f5fae2f 2969 struct btrfs_root *extent_root = root->fs_info->extent_root;
fec577fb
CM		 2970 int pending_ret;
2971 int ret;
a28ec197 2972
db94535d 2973 WARN_ON(num_bytes < root->sectorsize);
fec577fb 2974 if (root == extent_root) {
f3465ca4
JB		 2975 struct pending_extent_op *extent_op = NULL;
2976
2977 mutex_lock(&root->fs_info->extent_ins_mutex);
2978 if (test_range_bit(&root->fs_info->extent_ins, bytenr,
2979 bytenr + num_bytes - 1, EXTENT_WRITEBACK, 0)) {
2980 u64 priv;
2981 ret = get_state_private(&root->fs_info->extent_ins,
2982 bytenr, &priv);
2983 BUG_ON(ret);
2984 extent_op = (struct pending_extent_op *)
2985 (unsigned long)priv;
2986
2987 extent_op->del = 1;
2988 if (extent_op->type == PENDING_EXTENT_INSERT) {
2989 mutex_unlock(&root->fs_info->extent_ins_mutex);
2990 return 0;
2991 }
2992 }
2993
2994 if (extent_op) {
2995 ref_generation = extent_op->orig_generation;
2996 parent = extent_op->orig_parent;
2997 }
31840ae1
ZY		 2998
2999 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
3000 BUG_ON(!extent_op);
3001
3002 extent_op->type = PENDING_EXTENT_DELETE;
3003 extent_op->bytenr = bytenr;
3004 extent_op->num_bytes = num_bytes;
3005 extent_op->parent = parent;
3006 extent_op->orig_parent = parent;
3007 extent_op->generation = ref_generation;
3008 extent_op->orig_generation = ref_generation;
3009 extent_op->level = (int)owner_objectid;
f3465ca4
JB		 3010 INIT_LIST_HEAD(&extent_op->list);
3011 extent_op->del = 0;
31840ae1
ZY		 3012
3013 set_extent_bits(&root->fs_info->pending_del,
3014 bytenr, bytenr + num_bytes - 1,
25179201 3015 EXTENT_WRITEBACK, GFP_NOFS);
31840ae1
ZY		 3016 set_state_private(&root->fs_info->pending_del,
3017 bytenr, (unsigned long)extent_op);
25179201 3018 mutex_unlock(&root->fs_info->extent_ins_mutex);
fec577fb
CM		 3019 return 0;
3020 }
4bef0848 3021 /* if metadata always pin */
d00aff00
CM		 3022 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) {
3023 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
7237f183
YZ		 3024 mutex_lock(&root->fs_info->pinned_mutex);
3025 btrfs_update_pinned_extents(root, bytenr, num_bytes, 1);
3026 mutex_unlock(&root->fs_info->pinned_mutex);
e8569813 3027 update_reserved_extents(root, bytenr, num_bytes, 0);
d00aff00
CM		 3028 return 0;
3029 }
4bef0848 3030 pin = 1;
d00aff00 3031 }
4bef0848
CM		 3032
3033 /* if data pin when any transaction has committed this */
3034 if (ref_generation != trans->transid)
3035 pin = 1;
3036
31840ae1 3037 ret = __free_extent(trans, root, bytenr, num_bytes, parent,
3bb1a1bc
YZ		 3038 root_objectid, ref_generation,
3039 owner_objectid, pin, pin == 0);
ee6e6504 3040
87ef2bb4
CM		 3041 finish_current_insert(trans, root->fs_info->extent_root, 0);
3042 pending_ret = del_pending_extents(trans, root->fs_info->extent_root, 0);
fec577fb
CM		 3043 return ret ? ret : pending_ret;
3044}
3045
925baedd 3046int btrfs_free_extent(struct btrfs_trans_handle *trans,
31840ae1
ZY		 3047 struct btrfs_root *root,
3048 u64 bytenr, u64 num_bytes, u64 parent,
3049 u64 root_objectid, u64 ref_generation,
3bb1a1bc 3050 u64 owner_objectid, int pin)
925baedd
CM		 3051{
3052 int ret;
3053
31840ae1 3054 ret = __btrfs_free_extent(trans, root, bytenr, num_bytes, parent,
925baedd 3055 root_objectid, ref_generation,
3bb1a1bc 3056 owner_objectid, pin);
925baedd
CM		 3057 return ret;
3058}
3059
87ee04eb
CM		 3060static u64 stripe_align(struct btrfs_root *root, u64 val)
3061{
3062 u64 mask = ((u64)root->stripesize - 1);
3063 u64 ret = (val + mask) & ~mask;
3064 return ret;
3065}
3066
fec577fb
CM		 3067/*
3068 * walks the btree of allocated extents and find a hole of a given size.
3069 * The key ins is changed to record the hole:
3070 * ins->objectid == block start
62e2749e 3071 * ins->flags = BTRFS_EXTENT_ITEM_KEY
fec577fb
CM		 3072 * ins->offset == number of blocks
3073 * Any available blocks before search_start are skipped.
3074 */
d397712b 3075static noinline int find_free_extent(struct btrfs_trans_handle *trans,
98ed5174
CM		 3076 struct btrfs_root *orig_root,
3077 u64 num_bytes, u64 empty_size,
3078 u64 search_start, u64 search_end,
3079 u64 hint_byte, struct btrfs_key *ins,
3080 u64 exclude_start, u64 exclude_nr,
3081 int data)
fec577fb 3082{
80eb234a 3083 int ret = 0;
d397712b 3084 struct btrfs_root *root = orig_root->fs_info->extent_root;
db94535d 3085 u64 total_needed = num_bytes;
239b14b3 3086 u64 *last_ptr = NULL;
4366211c 3087 u64 last_wanted = 0;
80eb234a 3088 struct btrfs_block_group_cache *block_group = NULL;
0ef3e66b 3089 int chunk_alloc_done = 0;
239b14b3 3090 int empty_cluster = 2 * 1024 * 1024;
0ef3e66b 3091 int allowed_chunk_alloc = 0;
80eb234a
JB		 3092 struct list_head *head = NULL, *cur = NULL;
3093 int loop = 0;
f5a31e16 3094 int extra_loop = 0;
80eb234a 3095 struct btrfs_space_info *space_info;
fec577fb 3096
db94535d 3097 WARN_ON(num_bytes < root->sectorsize);
b1a4d965 3098 btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
80eb234a
JB		 3099 ins->objectid = 0;
3100 ins->offset = 0;
b1a4d965 3101
0ef3e66b
CM		 3102 if (orig_root->ref_cows || empty_size)
3103 allowed_chunk_alloc = 1;
3104
239b14b3
CM		 3105 if (data & BTRFS_BLOCK_GROUP_METADATA) {
3106 last_ptr = &root->fs_info->last_alloc;
536ac8ae
CM		 3107 if (!btrfs_test_opt(root, SSD))
3108 empty_cluster = 64 * 1024;
239b14b3
CM		 3109 }
3110
0f9dd46c 3111 if ((data & BTRFS_BLOCK_GROUP_DATA) && btrfs_test_opt(root, SSD))
239b14b3 3112 last_ptr = &root->fs_info->last_data_alloc;
0f9dd46c 3113
239b14b3 3114 if (last_ptr) {
4366211c 3115 if (*last_ptr) {
239b14b3 3116 hint_byte = *last_ptr;
4366211c
CM		 3117 last_wanted = *last_ptr;
3118 } else
239b14b3 3119 empty_size += empty_cluster;
4366211c
CM		 3120 } else {
3121 empty_cluster = 0;
239b14b3 3122 }
a061fc8d 3123 search_start = max(search_start, first_logical_byte(root, 0));
239b14b3 3124 search_start = max(search_start, hint_byte);
0b86a832 3125
42e70e7a 3126 if (last_wanted && search_start != last_wanted) {
4366211c 3127 last_wanted = 0;
42e70e7a
CM		 3128 empty_size += empty_cluster;
3129 }
4366211c 3130
42e70e7a 3131 total_needed += empty_size;
80eb234a 3132 block_group = btrfs_lookup_block_group(root->fs_info, search_start);
d899e052
YZ		 3133 if (!block_group)
3134 block_group = btrfs_lookup_first_block_group(root->fs_info,
3135 search_start);
80eb234a 3136 space_info = __find_space_info(root->fs_info, data);
0f9dd46c 3137
80eb234a
JB		 3138 down_read(&space_info->groups_sem);
3139 while (1) {
3140 struct btrfs_free_space *free_space;
0f9dd46c 3141 /*
80eb234a
JB		 3142 * the only way this happens if our hint points to a block
3143 * group thats not of the proper type, while looping this
3144 * should never happen
0f9dd46c 3145 */
8a1413a2
CM		 3146 if (empty_size)
3147 extra_loop = 1;
3148
42e70e7a
CM		 3149 if (!block_group)
3150 goto new_group_no_lock;
3151
ea6a478e
JB		 3152 if (unlikely(!block_group->cached)) {
3153 mutex_lock(&block_group->cache_mutex);
3154 ret = cache_block_group(root, block_group);
3155 mutex_unlock(&block_group->cache_mutex);
3156 if (ret)
3157 break;
3158 }
3159
25179201 3160 mutex_lock(&block_group->alloc_mutex);
80eb234a 3161 if (unlikely(!block_group_bits(block_group, data)))
0f9dd46c 3162 goto new_group;
0f9dd46c 3163
ea6a478e 3164 if (unlikely(block_group->ro))
80eb234a 3165 goto new_group;
0f9dd46c 3166
80eb234a
JB		 3167 free_space = btrfs_find_free_space(block_group, search_start,
3168 total_needed);
3169 if (free_space) {
3170 u64 start = block_group->key.objectid;
3171 u64 end = block_group->key.objectid +
0f9dd46c 3172 block_group->key.offset;
239b14b3 3173
80eb234a 3174 search_start = stripe_align(root, free_space->offset);
0f9dd46c 3175
80eb234a
JB		 3176 /* move on to the next group */
3177 if (search_start + num_bytes >= search_end)
3178 goto new_group;
e37c9e69 3179
80eb234a
JB		 3180 /* move on to the next group */
3181 if (search_start + num_bytes > end)
3182 goto new_group;
3183
4366211c 3184 if (last_wanted && search_start != last_wanted) {
3b7885bf 3185 total_needed += empty_cluster;
8a1413a2 3186 empty_size += empty_cluster;
4366211c 3187 last_wanted = 0;
3b7885bf
CM		 3188 /*
3189 * if search_start is still in this block group
3190 * then we just re-search this block group
3191 */
3192 if (search_start >= start &&
3193 search_start < end) {
3194 mutex_unlock(&block_group->alloc_mutex);
3195 continue;
3196 }
3197
3198 /* else we go to the next block group */
3199 goto new_group;
3200 }
3201
80eb234a
JB		 3202 if (exclude_nr > 0 &&
3203 (search_start + num_bytes > exclude_start &&
3204 search_start < exclude_start + exclude_nr)) {
3205 search_start = exclude_start + exclude_nr;
3206 /*
3207 * if search_start is still in this block group
3208 * then we just re-search this block group
3209 */
3210 if (search_start >= start &&
25179201
JB		 3211 search_start < end) {
3212 mutex_unlock(&block_group->alloc_mutex);
4366211c 3213 last_wanted = 0;
80eb234a 3214 continue;
25179201 3215 }
80eb234a
JB		 3216
3217 /* else we go to the next block group */
3218 goto new_group;
3219 }
3220
3221 ins->objectid = search_start;
3222 ins->offset = num_bytes;
25179201
JB		 3223
3224 btrfs_remove_free_space_lock(block_group, search_start,
3225 num_bytes);
80eb234a 3226 /* we are all good, lets return */
25179201 3227 mutex_unlock(&block_group->alloc_mutex);
80eb234a 3228 break;
0f9dd46c 3229 }
80eb234a 3230new_group:
42e70e7a 3231 mutex_unlock(&block_group->alloc_mutex);
d2fb3437
YZ		 3232 put_block_group(block_group);
3233 block_group = NULL;
42e70e7a 3234new_group_no_lock:
f5a31e16
CM		 3235 /* don't try to compare new allocations against the
3236 * last allocation any more
3237 */
4366211c 3238 last_wanted = 0;
f5a31e16 3239
80eb234a
JB		 3240 /*
3241 * Here's how this works.
3242 * loop == 0: we were searching a block group via a hint
3243 * and didn't find anything, so we start at
3244 * the head of the block groups and keep searching
3245 * loop == 1: we're searching through all of the block groups
3246 * if we hit the head again we have searched
3247 * all of the block groups for this space and we
3248 * need to try and allocate, if we cant error out.
3249 * loop == 2: we allocated more space and are looping through
3250 * all of the block groups again.
3251 */
3252 if (loop == 0) {
3253 head = &space_info->block_groups;
3254 cur = head->next;
80eb234a
JB		 3255 loop++;
3256 } else if (loop == 1 && cur == head) {
f5a31e16
CM		 3257 int keep_going;
3258
3259 /* at this point we give up on the empty_size
3260 * allocations and just try to allocate the min
3261 * space.
3262 *
3263 * The extra_loop field was set if an empty_size
3264 * allocation was attempted above, and if this
3265 * is try we need to try the loop again without
3266 * the additional empty_size.
3267 */
5b7c3fcc
CM		 3268 total_needed -= empty_size;
3269 empty_size = 0;
f5a31e16
CM		 3270 keep_going = extra_loop;
3271 loop++;
42e70e7a 3272
80eb234a
JB		 3273 if (allowed_chunk_alloc && !chunk_alloc_done) {
3274 up_read(&space_info->groups_sem);
3275 ret = do_chunk_alloc(trans, root, num_bytes +
3276 2 * 1024 * 1024, data, 1);
80eb234a 3277 down_read(&space_info->groups_sem);
2ed6d664
CM		 3278 if (ret < 0)
3279 goto loop_check;
80eb234a 3280 head = &space_info->block_groups;
f5a31e16
CM		 3281 /*
3282 * we've allocated a new chunk, keep
3283 * trying
3284 */
3285 keep_going = 1;
80eb234a
JB		 3286 chunk_alloc_done = 1;
3287 } else if (!allowed_chunk_alloc) {
3288 space_info->force_alloc = 1;
f5a31e16 3289 }
2ed6d664 3290loop_check:
f5a31e16
CM		 3291 if (keep_going) {
3292 cur = head->next;
3293 extra_loop = 0;
80eb234a
JB		 3294 } else {
3295 break;
3296 }
3297 } else if (cur == head) {
3298 break;
0f9dd46c 3299 }
0b86a832 3300
80eb234a
JB		 3301 block_group = list_entry(cur, struct btrfs_block_group_cache,
3302 list);
d2fb3437
YZ		 3303 atomic_inc(&block_group->count);
3304
80eb234a
JB		 3305 search_start = block_group->key.objectid;
3306 cur = cur->next;
f2654de4 3307 }
0b86a832 3308
80eb234a
JB		 3309 /* we found what we needed */
3310 if (ins->objectid) {
3311 if (!(data & BTRFS_BLOCK_GROUP_DATA))
d2fb3437 3312 trans->block_group = block_group->key.objectid;
0f9dd46c 3313
80eb234a
JB		 3314 if (last_ptr)
3315 *last_ptr = ins->objectid + ins->offset;
3316 ret = 0;
3317 } else if (!ret) {
d397712b
CM		 3318 printk(KERN_ERR "btrfs searching for %llu bytes, "
3319 "num_bytes %llu, loop %d, allowed_alloc %d\n",
3320 (unsigned long long)total_needed,
3321 (unsigned long long)num_bytes,
4ce4cb52 3322 loop, allowed_chunk_alloc);
80eb234a 3323 ret = -ENOSPC;
be744175 3324 }
d2fb3437
YZ		 3325 if (block_group)
3326 put_block_group(block_group);
be744175 3327
80eb234a 3328 up_read(&space_info->groups_sem);
0f70abe2 3329 return ret;
fec577fb 3330}
ec44a35c 3331
0f9dd46c
JB		 3332static void dump_space_info(struct btrfs_space_info *info, u64 bytes)
3333{
3334 struct btrfs_block_group_cache *cache;
0f9dd46c 3335
d397712b
CM		 3336 printk(KERN_INFO "space_info has %llu free, is %sfull\n",
3337 (unsigned long long)(info->total_bytes - info->bytes_used -
3338 info->bytes_pinned - info->bytes_reserved),
3339 (info->full) ? "" : "not ");
6a63209f
JB		 3340 printk(KERN_INFO "space_info total=%llu, pinned=%llu, delalloc=%llu,"
3341 " may_use=%llu, used=%llu\n", info->total_bytes,
3342 info->bytes_pinned, info->bytes_delalloc, info->bytes_may_use,
3343 info->bytes_used);
0f9dd46c 3344
80eb234a 3345 down_read(&info->groups_sem);
c6e30871 3346 list_for_each_entry(cache, &info->block_groups, list) {
0f9dd46c 3347 spin_lock(&cache->lock);
d397712b
CM		 3348 printk(KERN_INFO "block group %llu has %llu bytes, %llu used "
3349 "%llu pinned %llu reserved\n",
3350 (unsigned long long)cache->key.objectid,
3351 (unsigned long long)cache->key.offset,
3352 (unsigned long long)btrfs_block_group_used(&cache->item),
3353 (unsigned long long)cache->pinned,
3354 (unsigned long long)cache->reserved);
0f9dd46c
JB		 3355 btrfs_dump_free_space(cache, bytes);
3356 spin_unlock(&cache->lock);
3357 }
80eb234a 3358 up_read(&info->groups_sem);
0f9dd46c 3359}
e8569813 3360
e6dcd2dc
CM		 3361static int __btrfs_reserve_extent(struct btrfs_trans_handle *trans,
3362 struct btrfs_root *root,
3363 u64 num_bytes, u64 min_alloc_size,
3364 u64 empty_size, u64 hint_byte,
3365 u64 search_end, struct btrfs_key *ins,
3366 u64 data)
fec577fb
CM		 3367{
3368 int ret;
fbdc762b 3369 u64 search_start = 0;
1261ec42 3370 struct btrfs_fs_info *info = root->fs_info;
925baedd 3371
6a63209f 3372 data = btrfs_get_alloc_profile(root, data);
98d20f67 3373again:
0ef3e66b
CM		 3374 /*
3375 * the only place that sets empty_size is btrfs_realloc_node, which
3376 * is not called recursively on allocations
3377 */
3378 if (empty_size || root->ref_cows) {
593060d7 3379 if (!(data & BTRFS_BLOCK_GROUP_METADATA)) {
6324fbf3 3380 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
0ef3e66b
CM		 3381 2 * 1024 * 1024,
3382 BTRFS_BLOCK_GROUP_METADATA |
3383 (info->metadata_alloc_profile &
3384 info->avail_metadata_alloc_bits), 0);
6324fbf3
CM		 3385 }
3386 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
0ef3e66b 3387 num_bytes + 2 * 1024 * 1024, data, 0);
6324fbf3 3388 }
0b86a832 3389
db94535d
CM		 3390 WARN_ON(num_bytes < root->sectorsize);
3391 ret = find_free_extent(trans, root, num_bytes, empty_size,
3392 search_start, search_end, hint_byte, ins,
26b8003f
CM		 3393 trans->alloc_exclude_start,
3394 trans->alloc_exclude_nr, data);
3b951516 3395
98d20f67
CM		 3396 if (ret == -ENOSPC && num_bytes > min_alloc_size) {
3397 num_bytes = num_bytes >> 1;
0f9dd46c 3398 num_bytes = num_bytes & ~(root->sectorsize - 1);
98d20f67 3399 num_bytes = max(num_bytes, min_alloc_size);
0ef3e66b
CM		 3400 do_chunk_alloc(trans, root->fs_info->extent_root,
3401 num_bytes, data, 1);
98d20f67
CM		 3402 goto again;
3403 }
ec44a35c 3404 if (ret) {
0f9dd46c
JB		 3405 struct btrfs_space_info *sinfo;
3406
3407 sinfo = __find_space_info(root->fs_info, data);
d397712b
CM		 3408 printk(KERN_ERR "btrfs allocation failed flags %llu, "
3409 "wanted %llu\n", (unsigned long long)data,
3410 (unsigned long long)num_bytes);
0f9dd46c 3411 dump_space_info(sinfo, num_bytes);
925baedd 3412 BUG();
925baedd 3413 }
0f9dd46c
JB		 3414
3415 return ret;
e6dcd2dc
CM		 3416}
3417
65b51a00
CM		 3418int btrfs_free_reserved_extent(struct btrfs_root *root, u64 start, u64 len)
3419{
0f9dd46c 3420 struct btrfs_block_group_cache *cache;
1f3c79a2 3421 int ret = 0;
0f9dd46c 3422
0f9dd46c
JB		 3423 cache = btrfs_lookup_block_group(root->fs_info, start);
3424 if (!cache) {
d397712b
CM		 3425 printk(KERN_ERR "Unable to find block group for %llu\n",
3426 (unsigned long long)start);
0f9dd46c
JB		 3427 return -ENOSPC;
3428 }
1f3c79a2
LH		 3429
3430 ret = btrfs_discard_extent(root, start, len);
3431
0f9dd46c 3432 btrfs_add_free_space(cache, start, len);
d2fb3437 3433 put_block_group(cache);
1a40e23b 3434 update_reserved_extents(root, start, len, 0);
1f3c79a2
LH		 3435
3436 return ret;
65b51a00
CM		 3437}
3438
e6dcd2dc
CM		 3439int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
3440 struct btrfs_root *root,
3441 u64 num_bytes, u64 min_alloc_size,
3442 u64 empty_size, u64 hint_byte,
3443 u64 search_end, struct btrfs_key *ins,
3444 u64 data)
3445{
3446 int ret;
e6dcd2dc
CM		 3447 ret = __btrfs_reserve_extent(trans, root, num_bytes, min_alloc_size,
3448 empty_size, hint_byte, search_end, ins,
3449 data);
e8569813 3450 update_reserved_extents(root, ins->objectid, ins->offset, 1);
e6dcd2dc
CM		 3451 return ret;
3452}
3453
3454static int __btrfs_alloc_reserved_extent(struct btrfs_trans_handle *trans,
31840ae1 3455 struct btrfs_root *root, u64 parent,
e6dcd2dc 3456 u64 root_objectid, u64 ref_generation,
3bb1a1bc 3457 u64 owner, struct btrfs_key *ins)
e6dcd2dc
CM		 3458{
3459 int ret;
3460 int pending_ret;
3461 u64 super_used;
3462 u64 root_used;
3463 u64 num_bytes = ins->offset;
3464 u32 sizes[2];
3465 struct btrfs_fs_info *info = root->fs_info;
3466 struct btrfs_root *extent_root = info->extent_root;
3467 struct btrfs_extent_item *extent_item;
3468 struct btrfs_extent_ref *ref;
3469 struct btrfs_path *path;
3470 struct btrfs_key keys[2];
fec577fb 3471
31840ae1
ZY		 3472 if (parent == 0)
3473 parent = ins->objectid;
3474
58176a96 3475 /* block accounting for super block */
75eff68e 3476 spin_lock(&info->delalloc_lock);
db94535d
CM		 3477 super_used = btrfs_super_bytes_used(&info->super_copy);
3478 btrfs_set_super_bytes_used(&info->super_copy, super_used + num_bytes);
26b8003f 3479
58176a96 3480 /* block accounting for root item */
db94535d
CM		 3481 root_used = btrfs_root_used(&root->root_item);
3482 btrfs_set_root_used(&root->root_item, root_used + num_bytes);
34bf63c4 3483 spin_unlock(&info->delalloc_lock);
58176a96 3484
26b8003f 3485 if (root == extent_root) {
31840ae1
ZY		 3486 struct pending_extent_op *extent_op;
3487
3488 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
3489 BUG_ON(!extent_op);
3490
3491 extent_op->type = PENDING_EXTENT_INSERT;
3492 extent_op->bytenr = ins->objectid;
3493 extent_op->num_bytes = ins->offset;
3494 extent_op->parent = parent;
3495 extent_op->orig_parent = 0;
3496 extent_op->generation = ref_generation;
3497 extent_op->orig_generation = 0;
3498 extent_op->level = (int)owner;
f3465ca4
JB		 3499 INIT_LIST_HEAD(&extent_op->list);
3500 extent_op->del = 0;
31840ae1 3501
25179201 3502 mutex_lock(&root->fs_info->extent_ins_mutex);
1a5bc167
CM		 3503 set_extent_bits(&root->fs_info->extent_ins, ins->objectid,
3504 ins->objectid + ins->offset - 1,
25179201 3505 EXTENT_WRITEBACK, GFP_NOFS);
31840ae1
ZY		 3506 set_state_private(&root->fs_info->extent_ins,
3507 ins->objectid, (unsigned long)extent_op);
25179201 3508 mutex_unlock(&root->fs_info->extent_ins_mutex);
26b8003f
CM		 3509 goto update_block;
3510 }
3511
47e4bb98 3512 memcpy(&keys[0], ins, sizeof(*ins));
47e4bb98
CM		 3513 keys[1].objectid = ins->objectid;
3514 keys[1].type = BTRFS_EXTENT_REF_KEY;
31840ae1 3515 keys[1].offset = parent;
47e4bb98
CM		 3516 sizes[0] = sizeof(*extent_item);
3517 sizes[1] = sizeof(*ref);
7bb86316
CM		 3518
3519 path = btrfs_alloc_path();
3520 BUG_ON(!path);
47e4bb98
CM		 3521
3522 ret = btrfs_insert_empty_items(trans, extent_root, path, keys,
3523 sizes, 2);
ccd467d6 3524 BUG_ON(ret);
0f9dd46c 3525
47e4bb98
CM		 3526 extent_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3527 struct btrfs_extent_item);
3528 btrfs_set_extent_refs(path->nodes[0], extent_item, 1);
3529 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
3530 struct btrfs_extent_ref);
3531
3532 btrfs_set_ref_root(path->nodes[0], ref, root_objectid);
3533 btrfs_set_ref_generation(path->nodes[0], ref, ref_generation);
3534 btrfs_set_ref_objectid(path->nodes[0], ref, owner);
31840ae1 3535 btrfs_set_ref_num_refs(path->nodes[0], ref, 1);
47e4bb98
CM		 3536
3537 btrfs_mark_buffer_dirty(path->nodes[0]);
3538
3539 trans->alloc_exclude_start = 0;
3540 trans->alloc_exclude_nr = 0;
7bb86316 3541 btrfs_free_path(path);
87ef2bb4
CM		 3542 finish_current_insert(trans, extent_root, 0);
3543 pending_ret = del_pending_extents(trans, extent_root, 0);
f510cfec 3544
925baedd
CM		 3545 if (ret)
3546 goto out;
e37c9e69 3547 if (pending_ret) {
925baedd
CM		 3548 ret = pending_ret;
3549 goto out;
e37c9e69 3550 }
26b8003f
CM		 3551
3552update_block:
d397712b
CM		 3553 ret = update_block_group(trans, root, ins->objectid,
3554 ins->offset, 1, 0);
f5947066 3555 if (ret) {
d397712b
CM		 3556 printk(KERN_ERR "btrfs update block group failed for %llu "
3557 "%llu\n", (unsigned long long)ins->objectid,
3558 (unsigned long long)ins->offset);
f5947066
CM		 3559 BUG();
3560 }
925baedd 3561out:
e6dcd2dc
CM		 3562 return ret;
3563}
3564
3565int btrfs_alloc_reserved_extent(struct btrfs_trans_handle *trans,
31840ae1 3566 struct btrfs_root *root, u64 parent,
e6dcd2dc 3567 u64 root_objectid, u64 ref_generation,
3bb1a1bc 3568 u64 owner, struct btrfs_key *ins)
e6dcd2dc
CM		 3569{
3570 int ret;
1c2308f8
CM		 3571
3572 if (root_objectid == BTRFS_TREE_LOG_OBJECTID)
3573 return 0;
3bb1a1bc
YZ		 3574 ret = __btrfs_alloc_reserved_extent(trans, root, parent, root_objectid,
3575 ref_generation, owner, ins);
e8569813 3576 update_reserved_extents(root, ins->objectid, ins->offset, 0);
e6dcd2dc
CM		 3577 return ret;
3578}
e02119d5
CM		 3579
3580/*
3581 * this is used by the tree logging recovery code. It records that
3582 * an extent has been allocated and makes sure to clear the free
3583 * space cache bits as well
3584 */
3585int btrfs_alloc_logged_extent(struct btrfs_trans_handle *trans,
31840ae1 3586 struct btrfs_root *root, u64 parent,
e02119d5 3587 u64 root_objectid, u64 ref_generation,
3bb1a1bc 3588 u64 owner, struct btrfs_key *ins)
e02119d5
CM		 3589{
3590 int ret;
3591 struct btrfs_block_group_cache *block_group;
3592
e02119d5 3593 block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
ea6a478e 3594 mutex_lock(&block_group->cache_mutex);
e02119d5 3595 cache_block_group(root, block_group);
ea6a478e 3596 mutex_unlock(&block_group->cache_mutex);
e02119d5 3597
ea6a478e
JB		 3598 ret = btrfs_remove_free_space(block_group, ins->objectid,
3599 ins->offset);
0f9dd46c 3600 BUG_ON(ret);
d2fb3437 3601 put_block_group(block_group);
3bb1a1bc
YZ		 3602 ret = __btrfs_alloc_reserved_extent(trans, root, parent, root_objectid,
3603 ref_generation, owner, ins);
e02119d5
CM		 3604 return ret;
3605}
3606
e6dcd2dc
CM		 3607/*
3608 * finds a free extent and does all the dirty work required for allocation
3609 * returns the key for the extent through ins, and a tree buffer for
3610 * the first block of the extent through buf.
3611 *
3612 * returns 0 if everything worked, non-zero otherwise.
3613 */
3614int btrfs_alloc_extent(struct btrfs_trans_handle *trans,
3615 struct btrfs_root *root,
31840ae1 3616 u64 num_bytes, u64 parent, u64 min_alloc_size,
e6dcd2dc 3617 u64 root_objectid, u64 ref_generation,
3bb1a1bc 3618 u64 owner_objectid, u64 empty_size, u64 hint_byte,
e6dcd2dc
CM		 3619 u64 search_end, struct btrfs_key *ins, u64 data)
3620{
3621 int ret;
3622
e6dcd2dc
CM		 3623 ret = __btrfs_reserve_extent(trans, root, num_bytes,
3624 min_alloc_size, empty_size, hint_byte,
3625 search_end, ins, data);
3626 BUG_ON(ret);
d00aff00 3627 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
31840ae1
ZY		 3628 ret = __btrfs_alloc_reserved_extent(trans, root, parent,
3629 root_objectid, ref_generation,
3bb1a1bc 3630 owner_objectid, ins);
d00aff00 3631 BUG_ON(ret);
e6dcd2dc 3632
e8569813
ZY		 3633 } else {
3634 update_reserved_extents(root, ins->objectid, ins->offset, 1);
d00aff00 3635 }
925baedd 3636 return ret;
fec577fb 3637}
65b51a00
CM		 3638
3639struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans,
3640 struct btrfs_root *root,
4008c04a
CM		 3641 u64 bytenr, u32 blocksize,
3642 int level)
65b51a00
CM		 3643{
3644 struct extent_buffer *buf;
3645
3646 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
3647 if (!buf)
3648 return ERR_PTR(-ENOMEM);
3649 btrfs_set_header_generation(buf, trans->transid);
4008c04a 3650 btrfs_set_buffer_lockdep_class(buf, level);
65b51a00
CM		 3651 btrfs_tree_lock(buf);
3652 clean_tree_block(trans, root, buf);
b4ce94de
CM		 3653
3654 btrfs_set_lock_blocking(buf);
65b51a00 3655 btrfs_set_buffer_uptodate(buf);
b4ce94de 3656
d0c803c4
CM		 3657 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
3658 set_extent_dirty(&root->dirty_log_pages, buf->start,
3659 buf->start + buf->len - 1, GFP_NOFS);
3660 } else {
3661 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
65b51a00 3662 buf->start + buf->len - 1, GFP_NOFS);
d0c803c4 3663 }
65b51a00 3664 trans->blocks_used++;
b4ce94de 3665 /* this returns a buffer locked for blocking */
65b51a00
CM		 3666 return buf;
3667}
3668
fec577fb
CM		 3669/*
3670 * helper function to allocate a block for a given tree
3671 * returns the tree buffer or NULL.
3672 */
5f39d397 3673struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
7bb86316 3674 struct btrfs_root *root,
31840ae1 3675 u32 blocksize, u64 parent,
7bb86316
CM		 3676 u64 root_objectid,
3677 u64 ref_generation,
7bb86316
CM		 3678 int level,
3679 u64 hint,
5f39d397 3680 u64 empty_size)
fec577fb 3681{
e2fa7227 3682 struct btrfs_key ins;
fec577fb 3683 int ret;
5f39d397 3684 struct extent_buffer *buf;
fec577fb 3685
31840ae1 3686 ret = btrfs_alloc_extent(trans, root, blocksize, parent, blocksize,
3bb1a1bc 3687 root_objectid, ref_generation, level,
31840ae1 3688 empty_size, hint, (u64)-1, &ins, 0);
fec577fb 3689 if (ret) {
54aa1f4d
CM		 3690 BUG_ON(ret > 0);
3691 return ERR_PTR(ret);
fec577fb 3692 }
55c69072 3693
4008c04a
CM		 3694 buf = btrfs_init_new_buffer(trans, root, ins.objectid,
3695 blocksize, level);
fec577fb
CM		 3696 return buf;
3697}
a28ec197 3698
e02119d5
CM		 3699int btrfs_drop_leaf_ref(struct btrfs_trans_handle *trans,
3700 struct btrfs_root *root, struct extent_buffer *leaf)
6407bf6d 3701{
7bb86316
CM		 3702 u64 leaf_owner;
3703 u64 leaf_generation;
bd56b302 3704 struct refsort *sorted;
5f39d397 3705 struct btrfs_key key;
6407bf6d
CM		 3706 struct btrfs_file_extent_item *fi;
3707 int i;
3708 int nritems;
3709 int ret;
bd56b302
CM		 3710 int refi = 0;
3711 int slot;
6407bf6d 3712
5f39d397
CM		 3713 BUG_ON(!btrfs_is_leaf(leaf));
3714 nritems = btrfs_header_nritems(leaf);
7bb86316
CM		 3715 leaf_owner = btrfs_header_owner(leaf);
3716 leaf_generation = btrfs_header_generation(leaf);
3717
bd56b302
CM		 3718 sorted = kmalloc(sizeof(*sorted) * nritems, GFP_NOFS);
3719 /* we do this loop twice. The first time we build a list
3720 * of the extents we have a reference on, then we sort the list
3721 * by bytenr. The second time around we actually do the
3722 * extent freeing.
3723 */
6407bf6d 3724 for (i = 0; i < nritems; i++) {
db94535d 3725 u64 disk_bytenr;
e34a5b4f 3726 cond_resched();
5f39d397
CM		 3727
3728 btrfs_item_key_to_cpu(leaf, &key, i);
bd56b302
CM		 3729
3730 /* only extents have references, skip everything else */
5f39d397 3731 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
6407bf6d 3732 continue;
bd56b302 3733
6407bf6d 3734 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
bd56b302
CM		 3735
3736 /* inline extents live in the btree, they don't have refs */
5f39d397
CM		 3737 if (btrfs_file_extent_type(leaf, fi) ==
3738 BTRFS_FILE_EXTENT_INLINE)
236454df 3739 continue;
bd56b302 3740
db94535d 3741 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
bd56b302
CM		 3742
3743 /* holes don't have refs */
db94535d 3744 if (disk_bytenr == 0)
3a686375 3745 continue;
4a096752 3746
bd56b302
CM		 3747 sorted[refi].bytenr = disk_bytenr;
3748 sorted[refi].slot = i;
3749 refi++;
3750 }
3751
3752 if (refi == 0)
3753 goto out;
3754
3755 sort(sorted, refi, sizeof(struct refsort), refsort_cmp, NULL);
3756
3757 for (i = 0; i < refi; i++) {
3758 u64 disk_bytenr;
3759
3760 disk_bytenr = sorted[i].bytenr;
3761 slot = sorted[i].slot;
3762
3763 cond_resched();
3764
3765 btrfs_item_key_to_cpu(leaf, &key, slot);
3766 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
3767 continue;
3768
3769 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
3770
925baedd 3771 ret = __btrfs_free_extent(trans, root, disk_bytenr,
7bb86316 3772 btrfs_file_extent_disk_num_bytes(leaf, fi),
31840ae1 3773 leaf->start, leaf_owner, leaf_generation,
3bb1a1bc 3774 key.objectid, 0);
31840ae1 3775 BUG_ON(ret);
2dd3e67b
CM		 3776
3777 atomic_inc(&root->fs_info->throttle_gen);
3778 wake_up(&root->fs_info->transaction_throttle);
3779 cond_resched();
6407bf6d 3780 }
bd56b302
CM		 3781out:
3782 kfree(sorted);
6407bf6d
CM		 3783 return 0;
3784}
3785
d397712b 3786static noinline int cache_drop_leaf_ref(struct btrfs_trans_handle *trans,
e02119d5
CM		 3787 struct btrfs_root *root,
3788 struct btrfs_leaf_ref *ref)
31153d81
YZ		 3789{
3790 int i;
3791 int ret;
bd56b302
CM		 3792 struct btrfs_extent_info *info;
3793 struct refsort *sorted;
3794
3795 if (ref->nritems == 0)
3796 return 0;
31153d81 3797
bd56b302
CM		 3798 sorted = kmalloc(sizeof(*sorted) * ref->nritems, GFP_NOFS);
3799 for (i = 0; i < ref->nritems; i++) {
3800 sorted[i].bytenr = ref->extents[i].bytenr;
3801 sorted[i].slot = i;
3802 }
3803 sort(sorted, ref->nritems, sizeof(struct refsort), refsort_cmp, NULL);
3804
3805 /*
3806 * the items in the ref were sorted when the ref was inserted
3807 * into the ref cache, so this is already in order
3808 */
31153d81 3809 for (i = 0; i < ref->nritems; i++) {
bd56b302 3810 info = ref->extents + sorted[i].slot;
31840ae1
ZY		 3811 ret = __btrfs_free_extent(trans, root, info->bytenr,
3812 info->num_bytes, ref->bytenr,
3813 ref->owner, ref->generation,
3bb1a1bc 3814 info->objectid, 0);
2dd3e67b
CM		 3815
3816 atomic_inc(&root->fs_info->throttle_gen);
3817 wake_up(&root->fs_info->transaction_throttle);
3818 cond_resched();
3819
31153d81
YZ		 3820 BUG_ON(ret);
3821 info++;
3822 }
31153d81 3823
806638bc 3824 kfree(sorted);
31153d81
YZ		 3825 return 0;
3826}
3827
d397712b
CM		 3828static int drop_snap_lookup_refcount(struct btrfs_root *root, u64 start,
3829 u64 len, u32 *refs)
333db94c 3830{
017e5369 3831 int ret;
f87f057b 3832
31840ae1 3833 ret = btrfs_lookup_extent_ref(NULL, root, start, len, refs);
f87f057b
CM		 3834 BUG_ON(ret);
3835
d397712b 3836#if 0 /* some debugging code in case we see problems here */
f87f057b
CM		 3837 /* if the refs count is one, it won't get increased again. But
3838 * if the ref count is > 1, someone may be decreasing it at
3839 * the same time we are.
3840 */
3841 if (*refs != 1) {
3842 struct extent_buffer *eb = NULL;
3843 eb = btrfs_find_create_tree_block(root, start, len);
3844 if (eb)
3845 btrfs_tree_lock(eb);
3846
3847 mutex_lock(&root->fs_info->alloc_mutex);
3848 ret = lookup_extent_ref(NULL, root, start, len, refs);
3849 BUG_ON(ret);
3850 mutex_unlock(&root->fs_info->alloc_mutex);
3851
3852 if (eb) {
3853 btrfs_tree_unlock(eb);
3854 free_extent_buffer(eb);
3855 }
3856 if (*refs == 1) {
d397712b
CM		 3857 printk(KERN_ERR "btrfs block %llu went down to one "
3858 "during drop_snap\n", (unsigned long long)start);
f87f057b
CM		 3859 }
3860
3861 }
3862#endif
3863
e7a84565 3864 cond_resched();
017e5369 3865 return ret;
333db94c
CM		 3866}
3867
bd56b302
CM		 3868/*
3869 * this is used while deleting old snapshots, and it drops the refs
3870 * on a whole subtree starting from a level 1 node.
3871 *
3872 * The idea is to sort all the leaf pointers, and then drop the
3873 * ref on all the leaves in order. Most of the time the leaves
3874 * will have ref cache entries, so no leaf IOs will be required to
3875 * find the extents they have references on.
3876 *
3877 * For each leaf, any references it has are also dropped in order
3878 *
3879 * This ends up dropping the references in something close to optimal
3880 * order for reading and modifying the extent allocation tree.
3881 */
3882static noinline int drop_level_one_refs(struct btrfs_trans_handle *trans,
3883 struct btrfs_root *root,
3884 struct btrfs_path *path)
3885{
3886 u64 bytenr;
3887 u64 root_owner;
3888 u64 root_gen;
3889 struct extent_buffer *eb = path->nodes[1];
3890 struct extent_buffer *leaf;
3891 struct btrfs_leaf_ref *ref;
3892 struct refsort *sorted = NULL;
3893 int nritems = btrfs_header_nritems(eb);
3894 int ret;
3895 int i;
3896 int refi = 0;
3897 int slot = path->slots[1];
3898 u32 blocksize = btrfs_level_size(root, 0);
3899 u32 refs;
3900
3901 if (nritems == 0)
3902 goto out;
3903
3904 root_owner = btrfs_header_owner(eb);
3905 root_gen = btrfs_header_generation(eb);
3906 sorted = kmalloc(sizeof(*sorted) * nritems, GFP_NOFS);
3907
3908 /*
3909 * step one, sort all the leaf pointers so we don't scribble
3910 * randomly into the extent allocation tree
3911 */
3912 for (i = slot; i < nritems; i++) {
3913 sorted[refi].bytenr = btrfs_node_blockptr(eb, i);
3914 sorted[refi].slot = i;
3915 refi++;
3916 }
3917
3918 /*
3919 * nritems won't be zero, but if we're picking up drop_snapshot
3920 * after a crash, slot might be > 0, so double check things
3921 * just in case.
3922 */
3923 if (refi == 0)
3924 goto out;
3925
3926 sort(sorted, refi, sizeof(struct refsort), refsort_cmp, NULL);
3927
3928 /*
3929 * the first loop frees everything the leaves point to
3930 */
3931 for (i = 0; i < refi; i++) {
3932 u64 ptr_gen;
3933
3934 bytenr = sorted[i].bytenr;
3935
3936 /*
3937 * check the reference count on this leaf. If it is > 1
3938 * we just decrement it below and don't update any
3939 * of the refs the leaf points to.
3940 */
3941 ret = drop_snap_lookup_refcount(root, bytenr, blocksize, &refs);
3942 BUG_ON(ret);
3943 if (refs != 1)
3944 continue;
3945
3946 ptr_gen = btrfs_node_ptr_generation(eb, sorted[i].slot);
3947
3948 /*
3949 * the leaf only had one reference, which means the
3950 * only thing pointing to this leaf is the snapshot
3951 * we're deleting. It isn't possible for the reference
3952 * count to increase again later
3953 *
3954 * The reference cache is checked for the leaf,
3955 * and if found we'll be able to drop any refs held by
3956 * the leaf without needing to read it in.
3957 */
3958 ref = btrfs_lookup_leaf_ref(root, bytenr);
3959 if (ref && ref->generation != ptr_gen) {
3960 btrfs_free_leaf_ref(root, ref);
3961 ref = NULL;
3962 }
3963 if (ref) {
3964 ret = cache_drop_leaf_ref(trans, root, ref);
3965 BUG_ON(ret);
3966 btrfs_remove_leaf_ref(root, ref);
3967 btrfs_free_leaf_ref(root, ref);
3968 } else {
3969 /*
3970 * the leaf wasn't in the reference cache, so
3971 * we have to read it.
3972 */
3973 leaf = read_tree_block(root, bytenr, blocksize,
3974 ptr_gen);
3975 ret = btrfs_drop_leaf_ref(trans, root, leaf);
3976 BUG_ON(ret);
3977 free_extent_buffer(leaf);
3978 }
3979 atomic_inc(&root->fs_info->throttle_gen);
3980 wake_up(&root->fs_info->transaction_throttle);
3981 cond_resched();
3982 }
3983
3984 /*
3985 * run through the loop again to free the refs on the leaves.
3986 * This is faster than doing it in the loop above because
3987 * the leaves are likely to be clustered together. We end up
3988 * working in nice chunks on the extent allocation tree.
3989 */
3990 for (i = 0; i < refi; i++) {
3991 bytenr = sorted[i].bytenr;
3992 ret = __btrfs_free_extent(trans, root, bytenr,
3993 blocksize, eb->start,
3994 root_owner, root_gen, 0, 1);
3995 BUG_ON(ret);
3996
3997 atomic_inc(&root->fs_info->throttle_gen);
3998 wake_up(&root->fs_info->transaction_throttle);
3999 cond_resched();
4000 }
4001out:
4002 kfree(sorted);
4003
4004 /*
4005 * update the path to show we've processed the entire level 1
4006 * node. This will get saved into the root's drop_snapshot_progress
4007 * field so these drops are not repeated again if this transaction
4008 * commits.
4009 */
4010 path->slots[1] = nritems;
4011 return 0;
4012}
4013
9aca1d51
CM		 4014/*
4015 * helper function for drop_snapshot, this walks down the tree dropping ref
4016 * counts as it goes.
4017 */
d397712b 4018static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
98ed5174
CM		 4019 struct btrfs_root *root,
4020 struct btrfs_path *path, int *level)
20524f02 4021{
7bb86316
CM		 4022 u64 root_owner;
4023 u64 root_gen;
4024 u64 bytenr;
ca7a79ad 4025 u64 ptr_gen;
5f39d397
CM		 4026 struct extent_buffer *next;
4027 struct extent_buffer *cur;
7bb86316 4028 struct extent_buffer *parent;
db94535d 4029 u32 blocksize;
20524f02
CM		 4030 int ret;
4031 u32 refs;
4032
5caf2a00
CM		 4033 WARN_ON(*level < 0);
4034 WARN_ON(*level >= BTRFS_MAX_LEVEL);
333db94c 4035 ret = drop_snap_lookup_refcount(root, path->nodes[*level]->start,
db94535d 4036 path->nodes[*level]->len, &refs);
20524f02
CM		 4037 BUG_ON(ret);
4038 if (refs > 1)
4039 goto out;
e011599b 4040
9aca1d51
CM		 4041 /*
4042 * walk down to the last node level and free all the leaves
4043 */
d397712b 4044 while (*level >= 0) {
5caf2a00
CM		 4045 WARN_ON(*level < 0);
4046 WARN_ON(*level >= BTRFS_MAX_LEVEL);
20524f02 4047 cur = path->nodes[*level];
e011599b 4048
5f39d397 4049 if (btrfs_header_level(cur) != *level)
2c90e5d6 4050 WARN_ON(1);
e011599b 4051
7518a238 4052 if (path->slots[*level] >=
5f39d397 4053 btrfs_header_nritems(cur))
20524f02 4054 break;
bd56b302
CM		 4055
4056 /* the new code goes down to level 1 and does all the
4057 * leaves pointed to that node in bulk. So, this check
4058 * for level 0 will always be false.
4059 *
4060 * But, the disk format allows the drop_snapshot_progress
4061 * field in the root to leave things in a state where
4062 * a leaf will need cleaning up here. If someone crashes
4063 * with the old code and then boots with the new code,
4064 * we might find a leaf here.
4065 */
6407bf6d 4066 if (*level == 0) {
e02119d5 4067 ret = btrfs_drop_leaf_ref(trans, root, cur);
6407bf6d
CM		 4068 BUG_ON(ret);
4069 break;
4070 }
bd56b302
CM		 4071
4072 /*
4073 * once we get to level one, process the whole node
4074 * at once, including everything below it.
4075 */
4076 if (*level == 1) {
4077 ret = drop_level_one_refs(trans, root, path);
4078 BUG_ON(ret);
4079 break;
4080 }
4081
db94535d 4082 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
ca7a79ad 4083 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
db94535d 4084 blocksize = btrfs_level_size(root, *level - 1);
925baedd 4085
333db94c 4086 ret = drop_snap_lookup_refcount(root, bytenr, blocksize, &refs);
6407bf6d 4087 BUG_ON(ret);
bd56b302
CM		 4088
4089 /*
4090 * if there is more than one reference, we don't need
4091 * to read that node to drop any references it has. We
4092 * just drop the ref we hold on that node and move on to the
4093 * next slot in this level.
4094 */
6407bf6d 4095 if (refs != 1) {
7bb86316
CM		 4096 parent = path->nodes[*level];
4097 root_owner = btrfs_header_owner(parent);
4098 root_gen = btrfs_header_generation(parent);
20524f02 4099 path->slots[*level]++;
f87f057b 4100
925baedd 4101 ret = __btrfs_free_extent(trans, root, bytenr,
31840ae1 4102 blocksize, parent->start,
3bb1a1bc
YZ		 4103 root_owner, root_gen,
4104 *level - 1, 1);
20524f02 4105 BUG_ON(ret);
18e35e0a
CM		 4106
4107 atomic_inc(&root->fs_info->throttle_gen);
4108 wake_up(&root->fs_info->transaction_throttle);
2dd3e67b 4109 cond_resched();
18e35e0a 4110
20524f02
CM		 4111 continue;
4112 }
bd56b302 4113
f87f057b 4114 /*
bd56b302
CM		 4115 * we need to keep freeing things in the next level down.
4116 * read the block and loop around to process it
f87f057b 4117 */
bd56b302 4118 next = read_tree_block(root, bytenr, blocksize, ptr_gen);
5caf2a00 4119 WARN_ON(*level <= 0);
83e15a28 4120 if (path->nodes[*level-1])
5f39d397 4121 free_extent_buffer(path->nodes[*level-1]);
20524f02 4122 path->nodes[*level-1] = next;
5f39d397 4123 *level = btrfs_header_level(next);
20524f02 4124 path->slots[*level] = 0;
2dd3e67b 4125 cond_resched();
20524f02
CM		 4126 }
4127out:
5caf2a00
CM		 4128 WARN_ON(*level < 0);
4129 WARN_ON(*level >= BTRFS_MAX_LEVEL);
7bb86316
CM		 4130
4131 if (path->nodes[*level] == root->node) {
7bb86316 4132 parent = path->nodes[*level];
31153d81 4133 bytenr = path->nodes[*level]->start;
7bb86316
CM		 4134 } else {
4135 parent = path->nodes[*level + 1];
31153d81 4136 bytenr = btrfs_node_blockptr(parent, path->slots[*level + 1]);
7bb86316
CM		 4137 }
4138
31153d81
YZ		 4139 blocksize = btrfs_level_size(root, *level);
4140 root_owner = btrfs_header_owner(parent);
7bb86316 4141 root_gen = btrfs_header_generation(parent);
31153d81 4142
bd56b302
CM		 4143 /*
4144 * cleanup and free the reference on the last node
4145 * we processed
4146 */
31153d81 4147 ret = __btrfs_free_extent(trans, root, bytenr, blocksize,
31840ae1 4148 parent->start, root_owner, root_gen,
3bb1a1bc 4149 *level, 1);
5f39d397 4150 free_extent_buffer(path->nodes[*level]);
20524f02 4151 path->nodes[*level] = NULL;
bd56b302 4152
20524f02
CM		 4153 *level += 1;
4154 BUG_ON(ret);
f87f057b 4155
e7a84565 4156 cond_resched();
20524f02
CM		 4157 return 0;
4158}
4159
f82d02d9
YZ		 4160/*
4161 * helper function for drop_subtree, this function is similar to
4162 * walk_down_tree. The main difference is that it checks reference
4163 * counts while tree blocks are locked.
4164 */
d397712b 4165static noinline int walk_down_subtree(struct btrfs_trans_handle *trans,
f82d02d9
YZ		 4166 struct btrfs_root *root,
4167 struct btrfs_path *path, int *level)
4168{
4169 struct extent_buffer *next;
4170 struct extent_buffer *cur;
4171 struct extent_buffer *parent;
4172 u64 bytenr;
4173 u64 ptr_gen;
4174 u32 blocksize;
4175 u32 refs;
4176 int ret;
4177
4178 cur = path->nodes[*level];
4179 ret = btrfs_lookup_extent_ref(trans, root, cur->start, cur->len,
4180 &refs);
4181 BUG_ON(ret);
4182 if (refs > 1)
4183 goto out;
4184
4185 while (*level >= 0) {
4186 cur = path->nodes[*level];
4187 if (*level == 0) {
4188 ret = btrfs_drop_leaf_ref(trans, root, cur);
4189 BUG_ON(ret);
4190 clean_tree_block(trans, root, cur);
4191 break;
4192 }
4193 if (path->slots[*level] >= btrfs_header_nritems(cur)) {
4194 clean_tree_block(trans, root, cur);
4195 break;
4196 }
4197
4198 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
4199 blocksize = btrfs_level_size(root, *level - 1);
4200 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
4201
4202 next = read_tree_block(root, bytenr, blocksize, ptr_gen);
4203 btrfs_tree_lock(next);
b4ce94de 4204 btrfs_set_lock_blocking(next);
f82d02d9
YZ		 4205
4206 ret = btrfs_lookup_extent_ref(trans, root, bytenr, blocksize,
4207 &refs);
4208 BUG_ON(ret);
4209 if (refs > 1) {
4210 parent = path->nodes[*level];
4211 ret = btrfs_free_extent(trans, root, bytenr,
4212 blocksize, parent->start,
4213 btrfs_header_owner(parent),
4214 btrfs_header_generation(parent),
4215 *level - 1, 1);
4216 BUG_ON(ret);
4217 path->slots[*level]++;
4218 btrfs_tree_unlock(next);
4219 free_extent_buffer(next);
4220 continue;
4221 }
4222
4223 *level = btrfs_header_level(next);
4224 path->nodes[*level] = next;
4225 path->slots[*level] = 0;
4226 path->locks[*level] = 1;
4227 cond_resched();
4228 }
4229out:
4230 parent = path->nodes[*level + 1];
4231 bytenr = path->nodes[*level]->start;
4232 blocksize = path->nodes[*level]->len;
4233
4234 ret = btrfs_free_extent(trans, root, bytenr, blocksize,
4235 parent->start, btrfs_header_owner(parent),
4236 btrfs_header_generation(parent), *level, 1);
4237 BUG_ON(ret);
4238
4239 if (path->locks[*level]) {
4240 btrfs_tree_unlock(path->nodes[*level]);
4241 path->locks[*level] = 0;
4242 }
4243 free_extent_buffer(path->nodes[*level]);
4244 path->nodes[*level] = NULL;
4245 *level += 1;
4246 cond_resched();
4247 return 0;
4248}
4249
9aca1d51
CM		 4250/*
4251 * helper for dropping snapshots. This walks back up the tree in the path
4252 * to find the first node higher up where we haven't yet gone through
4253 * all the slots
4254 */
d397712b 4255static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
98ed5174 4256 struct btrfs_root *root,
f82d02d9
YZ		 4257 struct btrfs_path *path,
4258 int *level, int max_level)
20524f02 4259{
7bb86316
CM		 4260 u64 root_owner;
4261 u64 root_gen;
4262 struct btrfs_root_item *root_item = &root->root_item;
20524f02
CM		 4263 int i;
4264 int slot;
4265 int ret;
9f3a7427 4266
f82d02d9 4267 for (i = *level; i < max_level && path->nodes[i]; i++) {
20524f02 4268 slot = path->slots[i];
5f39d397
CM		 4269 if (slot < btrfs_header_nritems(path->nodes[i]) - 1) {
4270 struct extent_buffer *node;
4271 struct btrfs_disk_key disk_key;
bd56b302
CM		 4272
4273 /*
4274 * there is more work to do in this level.
4275 * Update the drop_progress marker to reflect
4276 * the work we've done so far, and then bump
4277 * the slot number
4278 */
5f39d397 4279 node = path->nodes[i];
20524f02
CM		 4280 path->slots[i]++;
4281 *level = i;
9f3a7427 4282 WARN_ON(*level == 0);
5f39d397 4283 btrfs_node_key(node, &disk_key, path->slots[i]);
9f3a7427 4284 memcpy(&root_item->drop_progress,
5f39d397 4285 &disk_key, sizeof(disk_key));
9f3a7427 4286 root_item->drop_level = i;
20524f02
CM		 4287 return 0;
4288 } else {
31840ae1 4289 struct extent_buffer *parent;
bd56b302
CM		 4290
4291 /*
4292 * this whole node is done, free our reference
4293 * on it and go up one level
4294 */
31840ae1
ZY		 4295 if (path->nodes[*level] == root->node)
4296 parent = path->nodes[*level];
4297 else
4298 parent = path->nodes[*level + 1];
4299
4300 root_owner = btrfs_header_owner(parent);
4301 root_gen = btrfs_header_generation(parent);
f82d02d9
YZ		 4302
4303 clean_tree_block(trans, root, path->nodes[*level]);
e089f05c 4304 ret = btrfs_free_extent(trans, root,
db94535d 4305 path->nodes[*level]->start,
7bb86316 4306 path->nodes[*level]->len,
3bb1a1bc
YZ		 4307 parent->start, root_owner,
4308 root_gen, *level, 1);
6407bf6d 4309 BUG_ON(ret);
f82d02d9
YZ		 4310 if (path->locks[*level]) {
4311 btrfs_tree_unlock(path->nodes[*level]);
4312 path->locks[*level] = 0;
4313 }
5f39d397 4314 free_extent_buffer(path->nodes[*level]);
83e15a28 4315 path->nodes[*level] = NULL;
20524f02 4316 *level = i + 1;
20524f02
CM		 4317 }
4318 }
4319 return 1;
4320}
4321
9aca1d51
CM		 4322/*
4323 * drop the reference count on the tree rooted at 'snap'. This traverses
4324 * the tree freeing any blocks that have a ref count of zero after being
4325 * decremented.
4326 */
e089f05c 4327int btrfs_drop_snapshot(struct btrfs_trans_handle *trans, struct btrfs_root
9f3a7427 4328 *root)
20524f02 4329{
3768f368 4330 int ret = 0;
9aca1d51 4331 int wret;
20524f02 4332 int level;
5caf2a00 4333 struct btrfs_path *path;
20524f02
CM		 4334 int i;
4335 int orig_level;
9f3a7427 4336 struct btrfs_root_item *root_item = &root->root_item;
20524f02 4337
a2135011 4338 WARN_ON(!mutex_is_locked(&root->fs_info->drop_mutex));
5caf2a00
CM		 4339 path = btrfs_alloc_path();
4340 BUG_ON(!path);
20524f02 4341
5f39d397 4342 level = btrfs_header_level(root->node);
20524f02 4343 orig_level = level;
9f3a7427
CM		 4344 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
4345 path->nodes[level] = root->node;
f510cfec 4346 extent_buffer_get(root->node);
9f3a7427
CM		 4347 path->slots[level] = 0;
4348 } else {
4349 struct btrfs_key key;
5f39d397
CM		 4350 struct btrfs_disk_key found_key;
4351 struct extent_buffer *node;
6702ed49 4352
9f3a7427 4353 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
6702ed49
CM		 4354 level = root_item->drop_level;
4355 path->lowest_level = level;
9f3a7427 4356 wret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6702ed49 4357 if (wret < 0) {
9f3a7427
CM		 4358 ret = wret;
4359 goto out;
4360 }
5f39d397
CM		 4361 node = path->nodes[level];
4362 btrfs_node_key(node, &found_key, path->slots[level]);
4363 WARN_ON(memcmp(&found_key, &root_item->drop_progress,
4364 sizeof(found_key)));
7d9eb12c
CM		 4365 /*
4366 * unlock our path, this is safe because only this
4367 * function is allowed to delete this snapshot
4368 */
925baedd
CM		 4369 for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
4370 if (path->nodes[i] && path->locks[i]) {
4371 path->locks[i] = 0;
4372 btrfs_tree_unlock(path->nodes[i]);
4373 }
4374 }
9f3a7427 4375 }
d397712b 4376 while (1) {
5caf2a00 4377 wret = walk_down_tree(trans, root, path, &level);
9aca1d51 4378 if (wret > 0)
20524f02 4379 break;
9aca1d51
CM		 4380 if (wret < 0)
4381 ret = wret;
4382
f82d02d9
YZ		 4383 wret = walk_up_tree(trans, root, path, &level,
4384 BTRFS_MAX_LEVEL);
9aca1d51 4385 if (wret > 0)
20524f02 4386 break;
9aca1d51
CM		 4387 if (wret < 0)
4388 ret = wret;
e7a84565
CM		 4389 if (trans->transaction->in_commit) {
4390 ret = -EAGAIN;
4391 break;
4392 }
18e35e0a 4393 atomic_inc(&root->fs_info->throttle_gen);
017e5369 4394 wake_up(&root->fs_info->transaction_throttle);
20524f02 4395 }
83e15a28 4396 for (i = 0; i <= orig_level; i++) {
5caf2a00 4397 if (path->nodes[i]) {
5f39d397 4398 free_extent_buffer(path->nodes[i]);
0f82731f 4399 path->nodes[i] = NULL;
83e15a28 4400 }
20524f02 4401 }
9f3a7427 4402out:
5caf2a00 4403 btrfs_free_path(path);
9aca1d51 4404 return ret;
20524f02 4405}
9078a3e1 4406
f82d02d9
YZ		 4407int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
4408 struct btrfs_root *root,
4409 struct extent_buffer *node,
4410 struct extent_buffer *parent)
4411{
4412 struct btrfs_path *path;
4413 int level;
4414 int parent_level;
4415 int ret = 0;
4416 int wret;
4417
4418 path = btrfs_alloc_path();
4419 BUG_ON(!path);
4420
4421 BUG_ON(!btrfs_tree_locked(parent));
4422 parent_level = btrfs_header_level(parent);
4423 extent_buffer_get(parent);
4424 path->nodes[parent_level] = parent;
4425 path->slots[parent_level] = btrfs_header_nritems(parent);
4426
4427 BUG_ON(!btrfs_tree_locked(node));
4428 level = btrfs_header_level(node);
4429 extent_buffer_get(node);
4430 path->nodes[level] = node;
4431 path->slots[level] = 0;
4432
4433 while (1) {
4434 wret = walk_down_subtree(trans, root, path, &level);
4435 if (wret < 0)
4436 ret = wret;
4437 if (wret != 0)
4438 break;
4439
4440 wret = walk_up_tree(trans, root, path, &level, parent_level);
4441 if (wret < 0)
4442 ret = wret;
4443 if (wret != 0)
4444 break;
4445 }
4446
4447 btrfs_free_path(path);
4448 return ret;
4449}
4450
8e7bf94f
CM		 4451static unsigned long calc_ra(unsigned long start, unsigned long last,
4452 unsigned long nr)
4453{
4454 return min(last, start + nr - 1);
4455}
4456
d397712b 4457static noinline int relocate_inode_pages(struct inode *inode, u64 start,
98ed5174 4458 u64 len)
edbd8d4e
CM		 4459{
4460 u64 page_start;
4461 u64 page_end;
1a40e23b 4462 unsigned long first_index;
edbd8d4e 4463 unsigned long last_index;
edbd8d4e
CM		 4464 unsigned long i;
4465 struct page *page;
d1310b2e 4466 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4313b399 4467 struct file_ra_state *ra;
3eaa2885 4468 struct btrfs_ordered_extent *ordered;
1a40e23b
ZY		 4469 unsigned int total_read = 0;
4470 unsigned int total_dirty = 0;
4471 int ret = 0;
4313b399
CM		 4472
4473 ra = kzalloc(sizeof(*ra), GFP_NOFS);
edbd8d4e
CM		 4474
4475 mutex_lock(&inode->i_mutex);
1a40e23b 4476 first_index = start >> PAGE_CACHE_SHIFT;
edbd8d4e
CM		 4477 last_index = (start + len - 1) >> PAGE_CACHE_SHIFT;
4478
1a40e23b
ZY		 4479 /* make sure the dirty trick played by the caller work */
4480 ret = invalidate_inode_pages2_range(inode->i_mapping,
4481 first_index, last_index);
4482 if (ret)
4483 goto out_unlock;
8e7bf94f 4484
4313b399 4485 file_ra_state_init(ra, inode->i_mapping);
edbd8d4e 4486
1a40e23b
ZY		 4487 for (i = first_index ; i <= last_index; i++) {
4488 if (total_read % ra->ra_pages == 0) {
8e7bf94f 4489 btrfs_force_ra(inode->i_mapping, ra, NULL, i,
1a40e23b 4490 calc_ra(i, last_index, ra->ra_pages));
8e7bf94f
CM		 4491 }
4492 total_read++;
3eaa2885
CM		 4493again:
4494 if (((u64)i << PAGE_CACHE_SHIFT) > i_size_read(inode))
1a40e23b 4495 BUG_ON(1);
edbd8d4e 4496 page = grab_cache_page(inode->i_mapping, i);
a061fc8d 4497 if (!page) {
1a40e23b 4498 ret = -ENOMEM;
edbd8d4e 4499 goto out_unlock;
a061fc8d 4500 }
edbd8d4e
CM		 4501 if (!PageUptodate(page)) {
4502 btrfs_readpage(NULL, page);
4503 lock_page(page);
4504 if (!PageUptodate(page)) {
4505 unlock_page(page);
4506 page_cache_release(page);
1a40e23b 4507 ret = -EIO;
edbd8d4e
CM		 4508 goto out_unlock;
4509 }
4510 }
ec44a35c 4511 wait_on_page_writeback(page);
3eaa2885 4512
edbd8d4e
CM		 4513 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
4514 page_end = page_start + PAGE_CACHE_SIZE - 1;
d1310b2e 4515 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
edbd8d4e 4516
3eaa2885
CM		 4517 ordered = btrfs_lookup_ordered_extent(inode, page_start);
4518 if (ordered) {
4519 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
4520 unlock_page(page);
4521 page_cache_release(page);
4522 btrfs_start_ordered_extent(inode, ordered, 1);
4523 btrfs_put_ordered_extent(ordered);
4524 goto again;
4525 }
4526 set_page_extent_mapped(page);
4527
1a40e23b
ZY		 4528 if (i == first_index)
4529 set_extent_bits(io_tree, page_start, page_end,
4530 EXTENT_BOUNDARY, GFP_NOFS);
1f80e4db 4531 btrfs_set_extent_delalloc(inode, page_start, page_end);
1a40e23b 4532
a061fc8d 4533 set_page_dirty(page);
1a40e23b 4534 total_dirty++;
edbd8d4e 4535
d1310b2e 4536 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
edbd8d4e
CM		 4537 unlock_page(page);
4538 page_cache_release(page);
4539 }
4540
4541out_unlock:
ec44a35c 4542 kfree(ra);
edbd8d4e 4543 mutex_unlock(&inode->i_mutex);
1a40e23b
ZY		 4544 balance_dirty_pages_ratelimited_nr(inode->i_mapping, total_dirty);
4545 return ret;
edbd8d4e
CM		 4546}
4547
d397712b 4548static noinline int relocate_data_extent(struct inode *reloc_inode,
1a40e23b
ZY		 4549 struct btrfs_key *extent_key,
4550 u64 offset)
4551{
4552 struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
4553 struct extent_map_tree *em_tree = &BTRFS_I(reloc_inode)->extent_tree;
4554 struct extent_map *em;
6643558d
YZ		 4555 u64 start = extent_key->objectid - offset;
4556 u64 end = start + extent_key->offset - 1;
bf4ef679 4557
1a40e23b
ZY		 4558 em = alloc_extent_map(GFP_NOFS);
4559 BUG_ON(!em || IS_ERR(em));
bf4ef679 4560
6643558d 4561 em->start = start;
1a40e23b 4562 em->len = extent_key->offset;
c8b97818 4563 em->block_len = extent_key->offset;
1a40e23b
ZY		 4564 em->block_start = extent_key->objectid;
4565 em->bdev = root->fs_info->fs_devices->latest_bdev;
4566 set_bit(EXTENT_FLAG_PINNED, &em->flags);
4567
4568 /* setup extent map to cheat btrfs_readpage */
6643558d 4569 lock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
1a40e23b
ZY		 4570 while (1) {
4571 int ret;
4572 spin_lock(&em_tree->lock);
4573 ret = add_extent_mapping(em_tree, em);
4574 spin_unlock(&em_tree->lock);
4575 if (ret != -EEXIST) {
4576 free_extent_map(em);
bf4ef679
CM		 4577 break;
4578 }
6643558d 4579 btrfs_drop_extent_cache(reloc_inode, start, end, 0);
bf4ef679 4580 }
6643558d 4581 unlock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
bf4ef679 4582
6643558d 4583 return relocate_inode_pages(reloc_inode, start, extent_key->offset);
1a40e23b 4584}
edbd8d4e 4585
1a40e23b
ZY		 4586struct btrfs_ref_path {
4587 u64 extent_start;
4588 u64 nodes[BTRFS_MAX_LEVEL];
4589 u64 root_objectid;
4590 u64 root_generation;
4591 u64 owner_objectid;
1a40e23b
ZY		 4592 u32 num_refs;
4593 int lowest_level;
4594 int current_level;
f82d02d9
YZ		 4595 int shared_level;
4596
4597 struct btrfs_key node_keys[BTRFS_MAX_LEVEL];
4598 u64 new_nodes[BTRFS_MAX_LEVEL];
1a40e23b 4599};
7d9eb12c 4600
1a40e23b 4601struct disk_extent {
c8b97818 4602 u64 ram_bytes;
1a40e23b
ZY		 4603 u64 disk_bytenr;
4604 u64 disk_num_bytes;
4605 u64 offset;
4606 u64 num_bytes;
c8b97818
CM		 4607 u8 compression;
4608 u8 encryption;
4609 u16 other_encoding;
1a40e23b 4610};
4313b399 4611
1a40e23b
ZY		 4612static int is_cowonly_root(u64 root_objectid)
4613{
4614 if (root_objectid == BTRFS_ROOT_TREE_OBJECTID ||
4615 root_objectid == BTRFS_EXTENT_TREE_OBJECTID ||
4616 root_objectid == BTRFS_CHUNK_TREE_OBJECTID ||
4617 root_objectid == BTRFS_DEV_TREE_OBJECTID ||
0403e47e
YZ		 4618 root_objectid == BTRFS_TREE_LOG_OBJECTID ||
4619 root_objectid == BTRFS_CSUM_TREE_OBJECTID)
1a40e23b
ZY		 4620 return 1;
4621 return 0;
4622}
edbd8d4e 4623
d397712b 4624static noinline int __next_ref_path(struct btrfs_trans_handle *trans,
1a40e23b
ZY		 4625 struct btrfs_root *extent_root,
4626 struct btrfs_ref_path *ref_path,
4627 int first_time)
4628{
4629 struct extent_buffer *leaf;
4630 struct btrfs_path *path;
4631 struct btrfs_extent_ref *ref;
4632 struct btrfs_key key;
4633 struct btrfs_key found_key;
4634 u64 bytenr;
4635 u32 nritems;
4636 int level;
4637 int ret = 1;
edbd8d4e 4638
1a40e23b
ZY		 4639 path = btrfs_alloc_path();
4640 if (!path)
4641 return -ENOMEM;
bf4ef679 4642
1a40e23b
ZY		 4643 if (first_time) {
4644 ref_path->lowest_level = -1;
4645 ref_path->current_level = -1;
f82d02d9 4646 ref_path->shared_level = -1;
1a40e23b
ZY		 4647 goto walk_up;
4648 }
4649walk_down:
4650 level = ref_path->current_level - 1;
4651 while (level >= -1) {
4652 u64 parent;
4653 if (level < ref_path->lowest_level)
4654 break;
bf4ef679 4655
d397712b 4656 if (level >= 0)
1a40e23b 4657 bytenr = ref_path->nodes[level];
d397712b 4658 else
1a40e23b 4659 bytenr = ref_path->extent_start;
1a40e23b 4660 BUG_ON(bytenr == 0);
bf4ef679 4661
1a40e23b
ZY		 4662 parent = ref_path->nodes[level + 1];
4663 ref_path->nodes[level + 1] = 0;
4664 ref_path->current_level = level;
4665 BUG_ON(parent == 0);
0ef3e66b 4666
1a40e23b
ZY		 4667 key.objectid = bytenr;
4668 key.offset = parent + 1;
4669 key.type = BTRFS_EXTENT_REF_KEY;
edbd8d4e 4670
1a40e23b
ZY		 4671 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
4672 if (ret < 0)
edbd8d4e 4673 goto out;
1a40e23b 4674 BUG_ON(ret == 0);
7d9eb12c 4675
1a40e23b
ZY		 4676 leaf = path->nodes[0];
4677 nritems = btrfs_header_nritems(leaf);
4678 if (path->slots[0] >= nritems) {
4679 ret = btrfs_next_leaf(extent_root, path);
4680 if (ret < 0)
4681 goto out;
4682 if (ret > 0)
4683 goto next;
4684 leaf = path->nodes[0];
4685 }
0ef3e66b 4686
1a40e23b
ZY		 4687 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4688 if (found_key.objectid == bytenr &&
f82d02d9
YZ		 4689 found_key.type == BTRFS_EXTENT_REF_KEY) {
4690 if (level < ref_path->shared_level)
4691 ref_path->shared_level = level;
1a40e23b 4692 goto found;
f82d02d9 4693 }
1a40e23b
ZY		 4694next:
4695 level--;
4696 btrfs_release_path(extent_root, path);
d899e052 4697 cond_resched();
1a40e23b
ZY		 4698 }
4699 /* reached lowest level */
4700 ret = 1;
4701 goto out;
4702walk_up:
4703 level = ref_path->current_level;
4704 while (level < BTRFS_MAX_LEVEL - 1) {
4705 u64 ref_objectid;
d397712b
CM		 4706
4707 if (level >= 0)
1a40e23b 4708 bytenr = ref_path->nodes[level];
d397712b 4709 else
1a40e23b 4710 bytenr = ref_path->extent_start;
d397712b 4711
1a40e23b 4712 BUG_ON(bytenr == 0);
edbd8d4e 4713
1a40e23b
ZY		 4714 key.objectid = bytenr;
4715 key.offset = 0;
4716 key.type = BTRFS_EXTENT_REF_KEY;
edbd8d4e 4717
1a40e23b
ZY		 4718 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
4719 if (ret < 0)
4720 goto out;
edbd8d4e 4721
1a40e23b
ZY		 4722 leaf = path->nodes[0];
4723 nritems = btrfs_header_nritems(leaf);
4724 if (path->slots[0] >= nritems) {
4725 ret = btrfs_next_leaf(extent_root, path);
4726 if (ret < 0)
4727 goto out;
4728 if (ret > 0) {
4729 /* the extent was freed by someone */
4730 if (ref_path->lowest_level == level)
4731 goto out;
4732 btrfs_release_path(extent_root, path);
4733 goto walk_down;
4734 }
4735 leaf = path->nodes[0];
4736 }
edbd8d4e 4737
1a40e23b
ZY		 4738 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4739 if (found_key.objectid != bytenr ||
4740 found_key.type != BTRFS_EXTENT_REF_KEY) {
4741 /* the extent was freed by someone */
4742 if (ref_path->lowest_level == level) {
4743 ret = 1;
4744 goto out;
4745 }
4746 btrfs_release_path(extent_root, path);
4747 goto walk_down;
4748 }
4749found:
4750 ref = btrfs_item_ptr(leaf, path->slots[0],
4751 struct btrfs_extent_ref);
4752 ref_objectid = btrfs_ref_objectid(leaf, ref);
4753 if (ref_objectid < BTRFS_FIRST_FREE_OBJECTID) {
4754 if (first_time) {
4755 level = (int)ref_objectid;
4756 BUG_ON(level >= BTRFS_MAX_LEVEL);
4757 ref_path->lowest_level = level;
4758 ref_path->current_level = level;
4759 ref_path->nodes[level] = bytenr;
4760 } else {
4761 WARN_ON(ref_objectid != level);
4762 }
4763 } else {
4764 WARN_ON(level != -1);
4765 }
4766 first_time = 0;
bf4ef679 4767
1a40e23b
ZY		 4768 if (ref_path->lowest_level == level) {
4769 ref_path->owner_objectid = ref_objectid;
1a40e23b
ZY		 4770 ref_path->num_refs = btrfs_ref_num_refs(leaf, ref);
4771 }
bf4ef679 4772
7d9eb12c 4773 /*
1a40e23b
ZY		 4774 * the block is tree root or the block isn't in reference
4775 * counted tree.
7d9eb12c 4776 */
1a40e23b
ZY		 4777 if (found_key.objectid == found_key.offset ||
4778 is_cowonly_root(btrfs_ref_root(leaf, ref))) {
4779 ref_path->root_objectid = btrfs_ref_root(leaf, ref);
4780 ref_path->root_generation =
4781 btrfs_ref_generation(leaf, ref);
4782 if (level < 0) {
4783 /* special reference from the tree log */
4784 ref_path->nodes[0] = found_key.offset;
4785 ref_path->current_level = 0;
4786 }
4787 ret = 0;
4788 goto out;
4789 }
7d9eb12c 4790
1a40e23b
ZY		 4791 level++;
4792 BUG_ON(ref_path->nodes[level] != 0);
4793 ref_path->nodes[level] = found_key.offset;
4794 ref_path->current_level = level;
bf4ef679 4795
1a40e23b
ZY		 4796 /*
4797 * the reference was created in the running transaction,
4798 * no need to continue walking up.
4799 */
4800 if (btrfs_ref_generation(leaf, ref) == trans->transid) {
4801 ref_path->root_objectid = btrfs_ref_root(leaf, ref);
4802 ref_path->root_generation =
4803 btrfs_ref_generation(leaf, ref);
4804 ret = 0;
4805 goto out;
7d9eb12c
CM		 4806 }
4807
1a40e23b 4808 btrfs_release_path(extent_root, path);
d899e052 4809 cond_resched();
7d9eb12c 4810 }
1a40e23b
ZY		 4811 /* reached max tree level, but no tree root found. */
4812 BUG();
edbd8d4e 4813out:
1a40e23b
ZY		 4814 btrfs_free_path(path);
4815 return ret;
edbd8d4e
CM		 4816}
4817
1a40e23b
ZY		 4818static int btrfs_first_ref_path(struct btrfs_trans_handle *trans,
4819 struct btrfs_root *extent_root,
4820 struct btrfs_ref_path *ref_path,
4821 u64 extent_start)
a061fc8d 4822{
1a40e23b
ZY		 4823 memset(ref_path, 0, sizeof(*ref_path));
4824 ref_path->extent_start = extent_start;
a061fc8d 4825
1a40e23b 4826 return __next_ref_path(trans, extent_root, ref_path, 1);
a061fc8d
CM		 4827}
4828
1a40e23b
ZY		 4829static int btrfs_next_ref_path(struct btrfs_trans_handle *trans,
4830 struct btrfs_root *extent_root,
4831 struct btrfs_ref_path *ref_path)
edbd8d4e 4832{
1a40e23b
ZY		 4833 return __next_ref_path(trans, extent_root, ref_path, 0);
4834}
4835
d397712b 4836static noinline int get_new_locations(struct inode *reloc_inode,
1a40e23b
ZY		 4837 struct btrfs_key *extent_key,
4838 u64 offset, int no_fragment,
4839 struct disk_extent **extents,
4840 int *nr_extents)
4841{
4842 struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
4843 struct btrfs_path *path;
4844 struct btrfs_file_extent_item *fi;
edbd8d4e 4845 struct extent_buffer *leaf;
1a40e23b
ZY		 4846 struct disk_extent *exts = *extents;
4847 struct btrfs_key found_key;
4848 u64 cur_pos;
4849 u64 last_byte;
edbd8d4e 4850 u32 nritems;
1a40e23b
ZY		 4851 int nr = 0;
4852 int max = *nr_extents;
4853 int ret;
edbd8d4e 4854
1a40e23b
ZY		 4855 WARN_ON(!no_fragment && *extents);
4856 if (!exts) {
4857 max = 1;
4858 exts = kmalloc(sizeof(*exts) * max, GFP_NOFS);
4859 if (!exts)
4860 return -ENOMEM;
a061fc8d 4861 }
edbd8d4e 4862
1a40e23b
ZY		 4863 path = btrfs_alloc_path();
4864 BUG_ON(!path);
edbd8d4e 4865
1a40e23b
ZY		 4866 cur_pos = extent_key->objectid - offset;
4867 last_byte = extent_key->objectid + extent_key->offset;
4868 ret = btrfs_lookup_file_extent(NULL, root, path, reloc_inode->i_ino,
4869 cur_pos, 0);
4870 if (ret < 0)
4871 goto out;
4872 if (ret > 0) {
4873 ret = -ENOENT;
4874 goto out;
4875 }
edbd8d4e 4876
1a40e23b 4877 while (1) {
edbd8d4e
CM		 4878 leaf = path->nodes[0];
4879 nritems = btrfs_header_nritems(leaf);
1a40e23b
ZY		 4880 if (path->slots[0] >= nritems) {
4881 ret = btrfs_next_leaf(root, path);
a061fc8d
CM		 4882 if (ret < 0)
4883 goto out;
1a40e23b
ZY		 4884 if (ret > 0)
4885 break;
bf4ef679 4886 leaf = path->nodes[0];
a061fc8d 4887 }
edbd8d4e
CM		 4888
4889 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1a40e23b
ZY		 4890 if (found_key.offset != cur_pos ||
4891 found_key.type != BTRFS_EXTENT_DATA_KEY ||
4892 found_key.objectid != reloc_inode->i_ino)
edbd8d4e
CM		 4893 break;
4894
1a40e23b
ZY		 4895 fi = btrfs_item_ptr(leaf, path->slots[0],
4896 struct btrfs_file_extent_item);
4897 if (btrfs_file_extent_type(leaf, fi) !=
4898 BTRFS_FILE_EXTENT_REG ||
4899 btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
edbd8d4e 4900 break;
1a40e23b
ZY		 4901
4902 if (nr == max) {
4903 struct disk_extent *old = exts;
4904 max *= 2;
4905 exts = kzalloc(sizeof(*exts) * max, GFP_NOFS);
4906 memcpy(exts, old, sizeof(*exts) * nr);
4907 if (old != *extents)
4908 kfree(old);
a061fc8d 4909 }
edbd8d4e 4910
1a40e23b
ZY		 4911 exts[nr].disk_bytenr =
4912 btrfs_file_extent_disk_bytenr(leaf, fi);
4913 exts[nr].disk_num_bytes =
4914 btrfs_file_extent_disk_num_bytes(leaf, fi);
4915 exts[nr].offset = btrfs_file_extent_offset(leaf, fi);
4916 exts[nr].num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
c8b97818
CM		 4917 exts[nr].ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
4918 exts[nr].compression = btrfs_file_extent_compression(leaf, fi);
4919 exts[nr].encryption = btrfs_file_extent_encryption(leaf, fi);
4920 exts[nr].other_encoding = btrfs_file_extent_other_encoding(leaf,
4921 fi);
d899e052
YZ		 4922 BUG_ON(exts[nr].offset > 0);
4923 BUG_ON(exts[nr].compression || exts[nr].encryption);
4924 BUG_ON(exts[nr].num_bytes != exts[nr].disk_num_bytes);
edbd8d4e 4925
1a40e23b
ZY		 4926 cur_pos += exts[nr].num_bytes;
4927 nr++;
4928
4929 if (cur_pos + offset >= last_byte)
4930 break;
4931
4932 if (no_fragment) {
4933 ret = 1;
edbd8d4e 4934 goto out;
1a40e23b
ZY		 4935 }
4936 path->slots[0]++;
4937 }
4938
1f80e4db 4939 BUG_ON(cur_pos + offset > last_byte);
1a40e23b
ZY		 4940 if (cur_pos + offset < last_byte) {
4941 ret = -ENOENT;
4942 goto out;
edbd8d4e
CM		 4943 }
4944 ret = 0;
4945out:
1a40e23b
ZY		 4946 btrfs_free_path(path);
4947 if (ret) {
4948 if (exts != *extents)
4949 kfree(exts);
4950 } else {
4951 *extents = exts;
4952 *nr_extents = nr;
4953 }
4954 return ret;
4955}
4956
d397712b 4957static noinline int replace_one_extent(struct btrfs_trans_handle *trans,
1a40e23b
ZY		 4958 struct btrfs_root *root,
4959 struct btrfs_path *path,
4960 struct btrfs_key *extent_key,
4961 struct btrfs_key *leaf_key,
4962 struct btrfs_ref_path *ref_path,
4963 struct disk_extent *new_extents,
4964 int nr_extents)
4965{
4966 struct extent_buffer *leaf;
4967 struct btrfs_file_extent_item *fi;
4968 struct inode *inode = NULL;
4969 struct btrfs_key key;
4970 u64 lock_start = 0;
4971 u64 lock_end = 0;
4972 u64 num_bytes;
4973 u64 ext_offset;
86288a19 4974 u64 search_end = (u64)-1;
1a40e23b 4975 u32 nritems;
3bb1a1bc 4976 int nr_scaned = 0;
1a40e23b 4977 int extent_locked = 0;
d899e052 4978 int extent_type;
1a40e23b
ZY		 4979 int ret;
4980
3bb1a1bc 4981 memcpy(&key, leaf_key, sizeof(key));
1a40e23b 4982 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
3bb1a1bc
YZ		 4983 if (key.objectid < ref_path->owner_objectid ||
4984 (key.objectid == ref_path->owner_objectid &&
4985 key.type < BTRFS_EXTENT_DATA_KEY)) {
4986 key.objectid = ref_path->owner_objectid;
4987 key.type = BTRFS_EXTENT_DATA_KEY;
4988 key.offset = 0;
4989 }
1a40e23b
ZY		 4990 }
4991
4992 while (1) {
4993 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
4994 if (ret < 0)
4995 goto out;
4996
4997 leaf = path->nodes[0];
4998 nritems = btrfs_header_nritems(leaf);
4999next:
5000 if (extent_locked && ret > 0) {
5001 /*
5002 * the file extent item was modified by someone
5003 * before the extent got locked.
5004 */
1a40e23b
ZY		 5005 unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
5006 lock_end, GFP_NOFS);
5007 extent_locked = 0;
5008 }
5009
5010 if (path->slots[0] >= nritems) {
3bb1a1bc 5011 if (++nr_scaned > 2)
1a40e23b
ZY		 5012 break;
5013
5014 BUG_ON(extent_locked);
5015 ret = btrfs_next_leaf(root, path);
5016 if (ret < 0)
5017 goto out;
5018 if (ret > 0)
5019 break;
5020 leaf = path->nodes[0];
5021 nritems = btrfs_header_nritems(leaf);
5022 }
5023
5024 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
5025
5026 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
5027 if ((key.objectid > ref_path->owner_objectid) ||
5028 (key.objectid == ref_path->owner_objectid &&
5029 key.type > BTRFS_EXTENT_DATA_KEY) ||
86288a19 5030 key.offset >= search_end)
1a40e23b
ZY		 5031 break;
5032 }
5033
5034 if (inode && key.objectid != inode->i_ino) {
5035 BUG_ON(extent_locked);
5036 btrfs_release_path(root, path);
5037 mutex_unlock(&inode->i_mutex);
5038 iput(inode);
5039 inode = NULL;
5040 continue;
5041 }
5042
5043 if (key.type != BTRFS_EXTENT_DATA_KEY) {
5044 path->slots[0]++;
5045 ret = 1;
5046 goto next;
5047 }
5048 fi = btrfs_item_ptr(leaf, path->slots[0],
5049 struct btrfs_file_extent_item);
d899e052
YZ		 5050 extent_type = btrfs_file_extent_type(leaf, fi);
5051 if ((extent_type != BTRFS_FILE_EXTENT_REG &&
5052 extent_type != BTRFS_FILE_EXTENT_PREALLOC) ||
1a40e23b
ZY		 5053 (btrfs_file_extent_disk_bytenr(leaf, fi) !=
5054 extent_key->objectid)) {
5055 path->slots[0]++;
5056 ret = 1;
5057 goto next;
5058 }
5059
5060 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
5061 ext_offset = btrfs_file_extent_offset(leaf, fi);
5062
86288a19
YZ		 5063 if (search_end == (u64)-1) {
5064 search_end = key.offset - ext_offset +
5065 btrfs_file_extent_ram_bytes(leaf, fi);
5066 }
1a40e23b
ZY		 5067
5068 if (!extent_locked) {
5069 lock_start = key.offset;
5070 lock_end = lock_start + num_bytes - 1;
5071 } else {
6643558d
YZ		 5072 if (lock_start > key.offset ||
5073 lock_end + 1 < key.offset + num_bytes) {
5074 unlock_extent(&BTRFS_I(inode)->io_tree,
5075 lock_start, lock_end, GFP_NOFS);
5076 extent_locked = 0;
5077 }
1a40e23b
ZY		 5078 }
5079
5080 if (!inode) {
5081 btrfs_release_path(root, path);
5082
5083 inode = btrfs_iget_locked(root->fs_info->sb,
5084 key.objectid, root);
5085 if (inode->i_state & I_NEW) {
5086 BTRFS_I(inode)->root = root;
5087 BTRFS_I(inode)->location.objectid =
5088 key.objectid;
5089 BTRFS_I(inode)->location.type =
5090 BTRFS_INODE_ITEM_KEY;
5091 BTRFS_I(inode)->location.offset = 0;
5092 btrfs_read_locked_inode(inode);
5093 unlock_new_inode(inode);
5094 }
5095 /*
5096 * some code call btrfs_commit_transaction while
5097 * holding the i_mutex, so we can't use mutex_lock
5098 * here.
5099 */
5100 if (is_bad_inode(inode) ||
5101 !mutex_trylock(&inode->i_mutex)) {
5102 iput(inode);
5103 inode = NULL;
5104 key.offset = (u64)-1;
5105 goto skip;
5106 }
5107 }
5108
5109 if (!extent_locked) {
5110 struct btrfs_ordered_extent *ordered;
5111
5112 btrfs_release_path(root, path);
5113
5114 lock_extent(&BTRFS_I(inode)->io_tree, lock_start,
5115 lock_end, GFP_NOFS);
5116 ordered = btrfs_lookup_first_ordered_extent(inode,
5117 lock_end);
5118 if (ordered &&
5119 ordered->file_offset <= lock_end &&
5120 ordered->file_offset + ordered->len > lock_start) {
5121 unlock_extent(&BTRFS_I(inode)->io_tree,
5122 lock_start, lock_end, GFP_NOFS);
5123 btrfs_start_ordered_extent(inode, ordered, 1);
5124 btrfs_put_ordered_extent(ordered);
5125 key.offset += num_bytes;
5126 goto skip;
5127 }
5128 if (ordered)
5129 btrfs_put_ordered_extent(ordered);
5130
1a40e23b
ZY		 5131 extent_locked = 1;
5132 continue;
5133 }
5134
5135 if (nr_extents == 1) {
5136 /* update extent pointer in place */
1a40e23b
ZY		 5137 btrfs_set_file_extent_disk_bytenr(leaf, fi,
5138 new_extents[0].disk_bytenr);
5139 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
5140 new_extents[0].disk_num_bytes);
1a40e23b
ZY		 5141 btrfs_mark_buffer_dirty(leaf);
5142
5143 btrfs_drop_extent_cache(inode, key.offset,
5144 key.offset + num_bytes - 1, 0);
5145
5146 ret = btrfs_inc_extent_ref(trans, root,
5147 new_extents[0].disk_bytenr,
5148 new_extents[0].disk_num_bytes,
5149 leaf->start,
5150 root->root_key.objectid,
5151 trans->transid,
3bb1a1bc 5152 key.objectid);
1a40e23b
ZY		 5153 BUG_ON(ret);
5154
5155 ret = btrfs_free_extent(trans, root,
5156 extent_key->objectid,
5157 extent_key->offset,
5158 leaf->start,
5159 btrfs_header_owner(leaf),
5160 btrfs_header_generation(leaf),
3bb1a1bc 5161 key.objectid, 0);
1a40e23b
ZY		 5162 BUG_ON(ret);
5163
5164 btrfs_release_path(root, path);
5165 key.offset += num_bytes;
5166 } else {
d899e052
YZ		 5167 BUG_ON(1);
5168#if 0
1a40e23b
ZY		 5169 u64 alloc_hint;
5170 u64 extent_len;
5171 int i;
5172 /*
5173 * drop old extent pointer at first, then insert the
5174 * new pointers one bye one
5175 */
5176 btrfs_release_path(root, path);
5177 ret = btrfs_drop_extents(trans, root, inode, key.offset,
5178 key.offset + num_bytes,
5179 key.offset, &alloc_hint);
5180 BUG_ON(ret);
5181
5182 for (i = 0; i < nr_extents; i++) {
5183 if (ext_offset >= new_extents[i].num_bytes) {
5184 ext_offset -= new_extents[i].num_bytes;
5185 continue;
5186 }
5187 extent_len = min(new_extents[i].num_bytes -
5188 ext_offset, num_bytes);
5189
5190 ret = btrfs_insert_empty_item(trans, root,
5191 path, &key,
5192 sizeof(*fi));
5193 BUG_ON(ret);
5194
5195 leaf = path->nodes[0];
5196 fi = btrfs_item_ptr(leaf, path->slots[0],
5197 struct btrfs_file_extent_item);
5198 btrfs_set_file_extent_generation(leaf, fi,
5199 trans->transid);
5200 btrfs_set_file_extent_type(leaf, fi,
5201 BTRFS_FILE_EXTENT_REG);
5202 btrfs_set_file_extent_disk_bytenr(leaf, fi,
5203 new_extents[i].disk_bytenr);
5204 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
5205 new_extents[i].disk_num_bytes);
c8b97818
CM		 5206 btrfs_set_file_extent_ram_bytes(leaf, fi,
5207 new_extents[i].ram_bytes);
5208
5209 btrfs_set_file_extent_compression(leaf, fi,
5210 new_extents[i].compression);
5211 btrfs_set_file_extent_encryption(leaf, fi,
5212 new_extents[i].encryption);
5213 btrfs_set_file_extent_other_encoding(leaf, fi,
5214 new_extents[i].other_encoding);
5215
1a40e23b
ZY		 5216 btrfs_set_file_extent_num_bytes(leaf, fi,
5217 extent_len);
5218 ext_offset += new_extents[i].offset;
5219 btrfs_set_file_extent_offset(leaf, fi,
5220 ext_offset);
5221 btrfs_mark_buffer_dirty(leaf);
5222
5223 btrfs_drop_extent_cache(inode, key.offset,
5224 key.offset + extent_len - 1, 0);
5225
5226 ret = btrfs_inc_extent_ref(trans, root,
5227 new_extents[i].disk_bytenr,
5228 new_extents[i].disk_num_bytes,
5229 leaf->start,
5230 root->root_key.objectid,
3bb1a1bc 5231 trans->transid, key.objectid);
1a40e23b
ZY		 5232 BUG_ON(ret);
5233 btrfs_release_path(root, path);
5234
a76a3cd4 5235 inode_add_bytes(inode, extent_len);
1a40e23b
ZY		 5236
5237 ext_offset = 0;
5238 num_bytes -= extent_len;
5239 key.offset += extent_len;
5240
5241 if (num_bytes == 0)
5242 break;
5243 }
5244 BUG_ON(i >= nr_extents);
d899e052 5245#endif
1a40e23b
ZY		 5246 }
5247
5248 if (extent_locked) {
1a40e23b
ZY		 5249 unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
5250 lock_end, GFP_NOFS);
5251 extent_locked = 0;
5252 }
5253skip:
5254 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS &&
86288a19 5255 key.offset >= search_end)
1a40e23b
ZY		 5256 break;
5257
5258 cond_resched();
5259 }
5260 ret = 0;
5261out:
5262 btrfs_release_path(root, path);
5263 if (inode) {
5264 mutex_unlock(&inode->i_mutex);
5265 if (extent_locked) {
1a40e23b
ZY		 5266 unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
5267 lock_end, GFP_NOFS);
5268 }
5269 iput(inode);
5270 }
5271 return ret;
5272}
5273
1a40e23b
ZY		 5274int btrfs_reloc_tree_cache_ref(struct btrfs_trans_handle *trans,
5275 struct btrfs_root *root,
5276 struct extent_buffer *buf, u64 orig_start)
5277{
5278 int level;
5279 int ret;
5280
5281 BUG_ON(btrfs_header_generation(buf) != trans->transid);
5282 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
5283
5284 level = btrfs_header_level(buf);
5285 if (level == 0) {
5286 struct btrfs_leaf_ref *ref;
5287 struct btrfs_leaf_ref *orig_ref;
5288
5289 orig_ref = btrfs_lookup_leaf_ref(root, orig_start);
5290 if (!orig_ref)
5291 return -ENOENT;
5292
5293 ref = btrfs_alloc_leaf_ref(root, orig_ref->nritems);
5294 if (!ref) {
5295 btrfs_free_leaf_ref(root, orig_ref);
5296 return -ENOMEM;
5297 }
5298
5299 ref->nritems = orig_ref->nritems;
5300 memcpy(ref->extents, orig_ref->extents,
5301 sizeof(ref->extents[0]) * ref->nritems);
5302
5303 btrfs_free_leaf_ref(root, orig_ref);
5304
5305 ref->root_gen = trans->transid;
5306 ref->bytenr = buf->start;
5307 ref->owner = btrfs_header_owner(buf);
5308 ref->generation = btrfs_header_generation(buf);
bd56b302 5309
1a40e23b
ZY		 5310 ret = btrfs_add_leaf_ref(root, ref, 0);
5311 WARN_ON(ret);
5312 btrfs_free_leaf_ref(root, ref);
5313 }
5314 return 0;
5315}
5316
d397712b 5317static noinline int invalidate_extent_cache(struct btrfs_root *root,
1a40e23b
ZY		 5318 struct extent_buffer *leaf,
5319 struct btrfs_block_group_cache *group,
5320 struct btrfs_root *target_root)
5321{
5322 struct btrfs_key key;
5323 struct inode *inode = NULL;
5324 struct btrfs_file_extent_item *fi;
5325 u64 num_bytes;
5326 u64 skip_objectid = 0;
5327 u32 nritems;
5328 u32 i;
5329
5330 nritems = btrfs_header_nritems(leaf);
5331 for (i = 0; i < nritems; i++) {
5332 btrfs_item_key_to_cpu(leaf, &key, i);
5333 if (key.objectid == skip_objectid ||
5334 key.type != BTRFS_EXTENT_DATA_KEY)
5335 continue;
5336 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
5337 if (btrfs_file_extent_type(leaf, fi) ==
5338 BTRFS_FILE_EXTENT_INLINE)
5339 continue;
5340 if (btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
5341 continue;
5342 if (!inode || inode->i_ino != key.objectid) {
5343 iput(inode);
5344 inode = btrfs_ilookup(target_root->fs_info->sb,
5345 key.objectid, target_root, 1);
5346 }
5347 if (!inode) {
5348 skip_objectid = key.objectid;
5349 continue;
5350 }
5351 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
5352
5353 lock_extent(&BTRFS_I(inode)->io_tree, key.offset,
5354 key.offset + num_bytes - 1, GFP_NOFS);
1a40e23b
ZY		 5355 btrfs_drop_extent_cache(inode, key.offset,
5356 key.offset + num_bytes - 1, 1);
1a40e23b
ZY		 5357 unlock_extent(&BTRFS_I(inode)->io_tree, key.offset,
5358 key.offset + num_bytes - 1, GFP_NOFS);
5359 cond_resched();
5360 }
5361 iput(inode);
5362 return 0;
5363}
5364
d397712b 5365static noinline int replace_extents_in_leaf(struct btrfs_trans_handle *trans,
1a40e23b
ZY		 5366 struct btrfs_root *root,
5367 struct extent_buffer *leaf,
5368 struct btrfs_block_group_cache *group,
5369 struct inode *reloc_inode)
5370{
5371 struct btrfs_key key;
5372 struct btrfs_key extent_key;
5373 struct btrfs_file_extent_item *fi;
5374 struct btrfs_leaf_ref *ref;
5375 struct disk_extent *new_extent;
5376 u64 bytenr;
5377 u64 num_bytes;
5378 u32 nritems;
5379 u32 i;
5380 int ext_index;
5381 int nr_extent;
5382 int ret;
5383
5384 new_extent = kmalloc(sizeof(*new_extent), GFP_NOFS);
5385 BUG_ON(!new_extent);
5386
5387 ref = btrfs_lookup_leaf_ref(root, leaf->start);
5388 BUG_ON(!ref);
5389
5390 ext_index = -1;
5391 nritems = btrfs_header_nritems(leaf);
5392 for (i = 0; i < nritems; i++) {
5393 btrfs_item_key_to_cpu(leaf, &key, i);
5394 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
5395 continue;
5396 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
5397 if (btrfs_file_extent_type(leaf, fi) ==
5398 BTRFS_FILE_EXTENT_INLINE)
5399 continue;
5400 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
5401 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
5402 if (bytenr == 0)
5403 continue;
5404
5405 ext_index++;
5406 if (bytenr >= group->key.objectid + group->key.offset ||
5407 bytenr + num_bytes <= group->key.objectid)
5408 continue;
5409
5410 extent_key.objectid = bytenr;
5411 extent_key.offset = num_bytes;
5412 extent_key.type = BTRFS_EXTENT_ITEM_KEY;
5413 nr_extent = 1;
5414 ret = get_new_locations(reloc_inode, &extent_key,
5415 group->key.objectid, 1,
5416 &new_extent, &nr_extent);
5417 if (ret > 0)
5418 continue;
5419 BUG_ON(ret < 0);
5420
5421 BUG_ON(ref->extents[ext_index].bytenr != bytenr);
5422 BUG_ON(ref->extents[ext_index].num_bytes != num_bytes);
5423 ref->extents[ext_index].bytenr = new_extent->disk_bytenr;
5424 ref->extents[ext_index].num_bytes = new_extent->disk_num_bytes;
5425
1a40e23b
ZY		 5426 btrfs_set_file_extent_disk_bytenr(leaf, fi,
5427 new_extent->disk_bytenr);
5428 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
5429 new_extent->disk_num_bytes);
1a40e23b
ZY		 5430 btrfs_mark_buffer_dirty(leaf);
5431
5432 ret = btrfs_inc_extent_ref(trans, root,
5433 new_extent->disk_bytenr,
5434 new_extent->disk_num_bytes,
5435 leaf->start,
5436 root->root_key.objectid,
3bb1a1bc 5437 trans->transid, key.objectid);
1a40e23b
ZY		 5438 BUG_ON(ret);
5439 ret = btrfs_free_extent(trans, root,
5440 bytenr, num_bytes, leaf->start,
5441 btrfs_header_owner(leaf),
5442 btrfs_header_generation(leaf),
3bb1a1bc 5443 key.objectid, 0);
1a40e23b
ZY		 5444 BUG_ON(ret);
5445 cond_resched();
5446 }
5447 kfree(new_extent);
5448 BUG_ON(ext_index + 1 != ref->nritems);
5449 btrfs_free_leaf_ref(root, ref);
5450 return 0;
5451}
5452
f82d02d9
YZ		 5453int btrfs_free_reloc_root(struct btrfs_trans_handle *trans,
5454 struct btrfs_root *root)
1a40e23b
ZY		 5455{
5456 struct btrfs_root *reloc_root;
f82d02d9 5457 int ret;
1a40e23b
ZY		 5458
5459 if (root->reloc_root) {
5460 reloc_root = root->reloc_root;
5461 root->reloc_root = NULL;
5462 list_add(&reloc_root->dead_list,
5463 &root->fs_info->dead_reloc_roots);
f82d02d9
YZ		 5464
5465 btrfs_set_root_bytenr(&reloc_root->root_item,
5466 reloc_root->node->start);
5467 btrfs_set_root_level(&root->root_item,
5468 btrfs_header_level(reloc_root->node));
5469 memset(&reloc_root->root_item.drop_progress, 0,
5470 sizeof(struct btrfs_disk_key));
5471 reloc_root->root_item.drop_level = 0;
5472
5473 ret = btrfs_update_root(trans, root->fs_info->tree_root,
5474 &reloc_root->root_key,
5475 &reloc_root->root_item);
5476 BUG_ON(ret);
1a40e23b
ZY		 5477 }
5478 return 0;
5479}
5480
5481int btrfs_drop_dead_reloc_roots(struct btrfs_root *root)
5482{
5483 struct btrfs_trans_handle *trans;
5484 struct btrfs_root *reloc_root;
5485 struct btrfs_root *prev_root = NULL;
5486 struct list_head dead_roots;
5487 int ret;
5488 unsigned long nr;
5489
5490 INIT_LIST_HEAD(&dead_roots);
5491 list_splice_init(&root->fs_info->dead_reloc_roots, &dead_roots);
5492
5493 while (!list_empty(&dead_roots)) {
5494 reloc_root = list_entry(dead_roots.prev,
5495 struct btrfs_root, dead_list);
5496 list_del_init(&reloc_root->dead_list);
5497
5498 BUG_ON(reloc_root->commit_root != NULL);
5499 while (1) {
5500 trans = btrfs_join_transaction(root, 1);
5501 BUG_ON(!trans);
5502
5503 mutex_lock(&root->fs_info->drop_mutex);
5504 ret = btrfs_drop_snapshot(trans, reloc_root);
5505 if (ret != -EAGAIN)
5506 break;
5507 mutex_unlock(&root->fs_info->drop_mutex);
5508
5509 nr = trans->blocks_used;
5510 ret = btrfs_end_transaction(trans, root);
5511 BUG_ON(ret);
5512 btrfs_btree_balance_dirty(root, nr);
5513 }
5514
5515 free_extent_buffer(reloc_root->node);
5516
5517 ret = btrfs_del_root(trans, root->fs_info->tree_root,
5518 &reloc_root->root_key);
5519 BUG_ON(ret);
5520 mutex_unlock(&root->fs_info->drop_mutex);
5521
5522 nr = trans->blocks_used;
5523 ret = btrfs_end_transaction(trans, root);
5524 BUG_ON(ret);
5525 btrfs_btree_balance_dirty(root, nr);
5526
5527 kfree(prev_root);
5528 prev_root = reloc_root;
5529 }
5530 if (prev_root) {
5531 btrfs_remove_leaf_refs(prev_root, (u64)-1, 0);
5532 kfree(prev_root);
5533 }
5534 return 0;
5535}
5536
5537int btrfs_add_dead_reloc_root(struct btrfs_root *root)
5538{
5539 list_add(&root->dead_list, &root->fs_info->dead_reloc_roots);
5540 return 0;
5541}
5542
5543int btrfs_cleanup_reloc_trees(struct btrfs_root *root)
5544{
5545 struct btrfs_root *reloc_root;
5546 struct btrfs_trans_handle *trans;
5547 struct btrfs_key location;
5548 int found;
5549 int ret;
5550
5551 mutex_lock(&root->fs_info->tree_reloc_mutex);
5552 ret = btrfs_find_dead_roots(root, BTRFS_TREE_RELOC_OBJECTID, NULL);
5553 BUG_ON(ret);
5554 found = !list_empty(&root->fs_info->dead_reloc_roots);
5555 mutex_unlock(&root->fs_info->tree_reloc_mutex);
5556
5557 if (found) {
5558 trans = btrfs_start_transaction(root, 1);
5559 BUG_ON(!trans);
5560 ret = btrfs_commit_transaction(trans, root);
5561 BUG_ON(ret);
5562 }
5563
5564 location.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
5565 location.offset = (u64)-1;
5566 location.type = BTRFS_ROOT_ITEM_KEY;
5567
5568 reloc_root = btrfs_read_fs_root_no_name(root->fs_info, &location);
5569 BUG_ON(!reloc_root);
5570 btrfs_orphan_cleanup(reloc_root);
5571 return 0;
5572}
5573
d397712b 5574static noinline int init_reloc_tree(struct btrfs_trans_handle *trans,
1a40e23b
ZY		 5575 struct btrfs_root *root)
5576{
5577 struct btrfs_root *reloc_root;
5578 struct extent_buffer *eb;
5579 struct btrfs_root_item *root_item;
5580 struct btrfs_key root_key;
5581 int ret;
5582
5583 BUG_ON(!root->ref_cows);
5584 if (root->reloc_root)
5585 return 0;
5586
5587 root_item = kmalloc(sizeof(*root_item), GFP_NOFS);
5588 BUG_ON(!root_item);
5589
5590 ret = btrfs_copy_root(trans, root, root->commit_root,
5591 &eb, BTRFS_TREE_RELOC_OBJECTID);
5592 BUG_ON(ret);
5593
5594 root_key.objectid = BTRFS_TREE_RELOC_OBJECTID;
5595 root_key.offset = root->root_key.objectid;
5596 root_key.type = BTRFS_ROOT_ITEM_KEY;
5597
5598 memcpy(root_item, &root->root_item, sizeof(root_item));
5599 btrfs_set_root_refs(root_item, 0);
5600 btrfs_set_root_bytenr(root_item, eb->start);
5601 btrfs_set_root_level(root_item, btrfs_header_level(eb));
84234f3a 5602 btrfs_set_root_generation(root_item, trans->transid);
1a40e23b
ZY		 5603
5604 btrfs_tree_unlock(eb);
5605 free_extent_buffer(eb);
5606
5607 ret = btrfs_insert_root(trans, root->fs_info->tree_root,
5608 &root_key, root_item);
5609 BUG_ON(ret);
5610 kfree(root_item);
5611
5612 reloc_root = btrfs_read_fs_root_no_radix(root->fs_info->tree_root,
5613 &root_key);
5614 BUG_ON(!reloc_root);
5615 reloc_root->last_trans = trans->transid;
5616 reloc_root->commit_root = NULL;
5617 reloc_root->ref_tree = &root->fs_info->reloc_ref_tree;
5618
5619 root->reloc_root = reloc_root;
5620 return 0;
5621}
5622
5623/*
5624 * Core function of space balance.
5625 *
5626 * The idea is using reloc trees to relocate tree blocks in reference
f82d02d9
YZ		 5627 * counted roots. There is one reloc tree for each subvol, and all
5628 * reloc trees share same root key objectid. Reloc trees are snapshots
5629 * of the latest committed roots of subvols (root->commit_root).
5630 *
5631 * To relocate a tree block referenced by a subvol, there are two steps.
5632 * COW the block through subvol's reloc tree, then update block pointer
5633 * in the subvol to point to the new block. Since all reloc trees share
5634 * same root key objectid, doing special handing for tree blocks owned
5635 * by them is easy. Once a tree block has been COWed in one reloc tree,
5636 * we can use the resulting new block directly when the same block is
5637 * required to COW again through other reloc trees. By this way, relocated
5638 * tree blocks are shared between reloc trees, so they are also shared
5639 * between subvols.
1a40e23b 5640 */
d397712b 5641static noinline int relocate_one_path(struct btrfs_trans_handle *trans,
1a40e23b
ZY		 5642 struct btrfs_root *root,
5643 struct btrfs_path *path,
5644 struct btrfs_key *first_key,
5645 struct btrfs_ref_path *ref_path,
5646 struct btrfs_block_group_cache *group,
5647 struct inode *reloc_inode)
5648{
5649 struct btrfs_root *reloc_root;
5650 struct extent_buffer *eb = NULL;
5651 struct btrfs_key *keys;
5652 u64 *nodes;
5653 int level;
f82d02d9 5654 int shared_level;
1a40e23b 5655 int lowest_level = 0;
1a40e23b
ZY		 5656 int ret;
5657
5658 if (ref_path->owner_objectid < BTRFS_FIRST_FREE_OBJECTID)
5659 lowest_level = ref_path->owner_objectid;
5660
f82d02d9 5661 if (!root->ref_cows) {
1a40e23b
ZY		 5662 path->lowest_level = lowest_level;
5663 ret = btrfs_search_slot(trans, root, first_key, path, 0, 1);
5664 BUG_ON(ret < 0);
5665 path->lowest_level = 0;
5666 btrfs_release_path(root, path);
5667 return 0;
5668 }
5669
1a40e23b
ZY		 5670 mutex_lock(&root->fs_info->tree_reloc_mutex);
5671 ret = init_reloc_tree(trans, root);
5672 BUG_ON(ret);
5673 reloc_root = root->reloc_root;
5674
f82d02d9
YZ		 5675 shared_level = ref_path->shared_level;
5676 ref_path->shared_level = BTRFS_MAX_LEVEL - 1;
1a40e23b 5677
f82d02d9
YZ		 5678 keys = ref_path->node_keys;
5679 nodes = ref_path->new_nodes;
5680 memset(&keys[shared_level + 1], 0,
5681 sizeof(*keys) * (BTRFS_MAX_LEVEL - shared_level - 1));
5682 memset(&nodes[shared_level + 1], 0,
5683 sizeof(*nodes) * (BTRFS_MAX_LEVEL - shared_level - 1));
1a40e23b 5684
f82d02d9
YZ		 5685 if (nodes[lowest_level] == 0) {
5686 path->lowest_level = lowest_level;
5687 ret = btrfs_search_slot(trans, reloc_root, first_key, path,
5688 0, 1);
5689 BUG_ON(ret);
5690 for (level = lowest_level; level < BTRFS_MAX_LEVEL; level++) {
5691 eb = path->nodes[level];
5692 if (!eb || eb == reloc_root->node)
5693 break;
5694 nodes[level] = eb->start;
5695 if (level == 0)
5696 btrfs_item_key_to_cpu(eb, &keys[level], 0);
5697 else
5698 btrfs_node_key_to_cpu(eb, &keys[level], 0);
5699 }
2b82032c
YZ		 5700 if (nodes[0] &&
5701 ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
f82d02d9
YZ		 5702 eb = path->nodes[0];
5703 ret = replace_extents_in_leaf(trans, reloc_root, eb,
5704 group, reloc_inode);
5705 BUG_ON(ret);
5706 }
5707 btrfs_release_path(reloc_root, path);
5708 } else {
1a40e23b 5709 ret = btrfs_merge_path(trans, reloc_root, keys, nodes,
f82d02d9 5710 lowest_level);
1a40e23b
ZY		 5711 BUG_ON(ret);
5712 }
5713
1a40e23b
ZY		 5714 /*
5715 * replace tree blocks in the fs tree with tree blocks in
5716 * the reloc tree.
5717 */
5718 ret = btrfs_merge_path(trans, root, keys, nodes, lowest_level);
5719 BUG_ON(ret < 0);
5720
5721 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
f82d02d9
YZ		 5722 ret = btrfs_search_slot(trans, reloc_root, first_key, path,
5723 0, 0);
5724 BUG_ON(ret);
5725 extent_buffer_get(path->nodes[0]);
5726 eb = path->nodes[0];
5727 btrfs_release_path(reloc_root, path);
1a40e23b
ZY		 5728 ret = invalidate_extent_cache(reloc_root, eb, group, root);
5729 BUG_ON(ret);
5730 free_extent_buffer(eb);
5731 }
1a40e23b 5732
f82d02d9 5733 mutex_unlock(&root->fs_info->tree_reloc_mutex);
1a40e23b 5734 path->lowest_level = 0;
1a40e23b
ZY		 5735 return 0;
5736}
5737
d397712b 5738static noinline int relocate_tree_block(struct btrfs_trans_handle *trans,
1a40e23b
ZY		 5739 struct btrfs_root *root,
5740 struct btrfs_path *path,
5741 struct btrfs_key *first_key,
5742 struct btrfs_ref_path *ref_path)
5743{
5744 int ret;
1a40e23b
ZY		 5745
5746 ret = relocate_one_path(trans, root, path, first_key,
5747 ref_path, NULL, NULL);
5748 BUG_ON(ret);
5749
5750 if (root == root->fs_info->extent_root)
5751 btrfs_extent_post_op(trans, root);
1a40e23b
ZY		 5752
5753 return 0;
5754}
5755
d397712b 5756static noinline int del_extent_zero(struct btrfs_trans_handle *trans,
1a40e23b
ZY		 5757 struct btrfs_root *extent_root,
5758 struct btrfs_path *path,
5759 struct btrfs_key *extent_key)
5760{
5761 int ret;
5762
1a40e23b
ZY		 5763 ret = btrfs_search_slot(trans, extent_root, extent_key, path, -1, 1);
5764 if (ret)
5765 goto out;
5766 ret = btrfs_del_item(trans, extent_root, path);
5767out:
5768 btrfs_release_path(extent_root, path);
1a40e23b
ZY		 5769 return ret;
5770}
5771
d397712b 5772static noinline struct btrfs_root *read_ref_root(struct btrfs_fs_info *fs_info,
1a40e23b
ZY		 5773 struct btrfs_ref_path *ref_path)
5774{
5775 struct btrfs_key root_key;
5776
5777 root_key.objectid = ref_path->root_objectid;
5778 root_key.type = BTRFS_ROOT_ITEM_KEY;
5779 if (is_cowonly_root(ref_path->root_objectid))
5780 root_key.offset = 0;
5781 else
5782 root_key.offset = (u64)-1;
5783
5784 return btrfs_read_fs_root_no_name(fs_info, &root_key);
5785}
5786
d397712b 5787static noinline int relocate_one_extent(struct btrfs_root *extent_root,
1a40e23b
ZY		 5788 struct btrfs_path *path,
5789 struct btrfs_key *extent_key,
5790 struct btrfs_block_group_cache *group,
5791 struct inode *reloc_inode, int pass)
5792{
5793 struct btrfs_trans_handle *trans;
5794 struct btrfs_root *found_root;
5795 struct btrfs_ref_path *ref_path = NULL;
5796 struct disk_extent *new_extents = NULL;
5797 int nr_extents = 0;
5798 int loops;
5799 int ret;
5800 int level;
5801 struct btrfs_key first_key;
5802 u64 prev_block = 0;
5803
1a40e23b
ZY		 5804
5805 trans = btrfs_start_transaction(extent_root, 1);
5806 BUG_ON(!trans);
5807
5808 if (extent_key->objectid == 0) {
5809 ret = del_extent_zero(trans, extent_root, path, extent_key);
5810 goto out;
5811 }
5812
5813 ref_path = kmalloc(sizeof(*ref_path), GFP_NOFS);
5814 if (!ref_path) {
d397712b
CM		 5815 ret = -ENOMEM;
5816 goto out;
1a40e23b
ZY		 5817 }
5818
5819 for (loops = 0; ; loops++) {
5820 if (loops == 0) {
5821 ret = btrfs_first_ref_path(trans, extent_root, ref_path,
5822 extent_key->objectid);
5823 } else {
5824 ret = btrfs_next_ref_path(trans, extent_root, ref_path);
5825 }
5826 if (ret < 0)
5827 goto out;
5828 if (ret > 0)
5829 break;
5830
5831 if (ref_path->root_objectid == BTRFS_TREE_LOG_OBJECTID ||
5832 ref_path->root_objectid == BTRFS_TREE_RELOC_OBJECTID)
5833 continue;
5834
5835 found_root = read_ref_root(extent_root->fs_info, ref_path);
5836 BUG_ON(!found_root);
5837 /*
5838 * for reference counted tree, only process reference paths
5839 * rooted at the latest committed root.
5840 */
5841 if (found_root->ref_cows &&
5842 ref_path->root_generation != found_root->root_key.offset)
5843 continue;
5844
5845 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
5846 if (pass == 0) {
5847 /*
5848 * copy data extents to new locations
5849 */
5850 u64 group_start = group->key.objectid;
5851 ret = relocate_data_extent(reloc_inode,
5852 extent_key,
5853 group_start);
5854 if (ret < 0)
5855 goto out;
5856 break;
5857 }
5858 level = 0;
5859 } else {
5860 level = ref_path->owner_objectid;
5861 }
5862
5863 if (prev_block != ref_path->nodes[level]) {
5864 struct extent_buffer *eb;
5865 u64 block_start = ref_path->nodes[level];
5866 u64 block_size = btrfs_level_size(found_root, level);
5867
5868 eb = read_tree_block(found_root, block_start,
5869 block_size, 0);
5870 btrfs_tree_lock(eb);
5871 BUG_ON(level != btrfs_header_level(eb));
5872
5873 if (level == 0)
5874 btrfs_item_key_to_cpu(eb, &first_key, 0);
5875 else
5876 btrfs_node_key_to_cpu(eb, &first_key, 0);
5877
5878 btrfs_tree_unlock(eb);
5879 free_extent_buffer(eb);
5880 prev_block = block_start;
5881 }
5882
24562425 5883 mutex_lock(&extent_root->fs_info->trans_mutex);
e4404d6e 5884 btrfs_record_root_in_trans(found_root);
24562425 5885 mutex_unlock(&extent_root->fs_info->trans_mutex);
e4404d6e
YZ		 5886 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
5887 /*
5888 * try to update data extent references while
5889 * keeping metadata shared between snapshots.
5890 */
5891 if (pass == 1) {
5892 ret = relocate_one_path(trans, found_root,
5893 path, &first_key, ref_path,
5894 group, reloc_inode);
5895 if (ret < 0)
5896 goto out;
5897 continue;
5898 }
1a40e23b
ZY		 5899 /*
5900 * use fallback method to process the remaining
5901 * references.
5902 */
5903 if (!new_extents) {
5904 u64 group_start = group->key.objectid;
d899e052
YZ		 5905 new_extents = kmalloc(sizeof(*new_extents),
5906 GFP_NOFS);
5907 nr_extents = 1;
1a40e23b
ZY		 5908 ret = get_new_locations(reloc_inode,
5909 extent_key,
d899e052 5910 group_start, 1,
1a40e23b
ZY		 5911 &new_extents,
5912 &nr_extents);
d899e052 5913 if (ret)
1a40e23b
ZY		 5914 goto out;
5915 }
1a40e23b
ZY		 5916 ret = replace_one_extent(trans, found_root,
5917 path, extent_key,
5918 &first_key, ref_path,
5919 new_extents, nr_extents);
e4404d6e 5920 } else {
1a40e23b
ZY		 5921 ret = relocate_tree_block(trans, found_root, path,
5922 &first_key, ref_path);
1a40e23b
ZY		 5923 }
5924 if (ret < 0)
5925 goto out;
5926 }
5927 ret = 0;
5928out:
5929 btrfs_end_transaction(trans, extent_root);
5930 kfree(new_extents);
5931 kfree(ref_path);
1a40e23b
ZY		 5932 return ret;
5933}
5934
ec44a35c
CM		 5935static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
5936{
5937 u64 num_devices;
5938 u64 stripped = BTRFS_BLOCK_GROUP_RAID0 |
5939 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
5940
2b82032c 5941 num_devices = root->fs_info->fs_devices->rw_devices;
ec44a35c
CM		 5942 if (num_devices == 1) {
5943 stripped |= BTRFS_BLOCK_GROUP_DUP;
5944 stripped = flags & ~stripped;
5945
5946 /* turn raid0 into single device chunks */
5947 if (flags & BTRFS_BLOCK_GROUP_RAID0)
5948 return stripped;
5949
5950 /* turn mirroring into duplication */
5951 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
5952 BTRFS_BLOCK_GROUP_RAID10))
5953 return stripped | BTRFS_BLOCK_GROUP_DUP;
5954 return flags;
5955 } else {
5956 /* they already had raid on here, just return */
ec44a35c
CM		 5957 if (flags & stripped)
5958 return flags;
5959
5960 stripped |= BTRFS_BLOCK_GROUP_DUP;
5961 stripped = flags & ~stripped;
5962
5963 /* switch duplicated blocks with raid1 */
5964 if (flags & BTRFS_BLOCK_GROUP_DUP)
5965 return stripped | BTRFS_BLOCK_GROUP_RAID1;
5966
5967 /* turn single device chunks into raid0 */
5968 return stripped | BTRFS_BLOCK_GROUP_RAID0;
5969 }
5970 return flags;
5971}
5972
b2950863 5973static int __alloc_chunk_for_shrink(struct btrfs_root *root,
0ef3e66b
CM		 5974 struct btrfs_block_group_cache *shrink_block_group,
5975 int force)
5976{
5977 struct btrfs_trans_handle *trans;
5978 u64 new_alloc_flags;
5979 u64 calc;
5980
c286ac48 5981 spin_lock(&shrink_block_group->lock);
0ef3e66b 5982 if (btrfs_block_group_used(&shrink_block_group->item) > 0) {
c286ac48 5983 spin_unlock(&shrink_block_group->lock);
c286ac48 5984
0ef3e66b 5985 trans = btrfs_start_transaction(root, 1);
c286ac48 5986 spin_lock(&shrink_block_group->lock);
7d9eb12c 5987
0ef3e66b
CM		 5988 new_alloc_flags = update_block_group_flags(root,
5989 shrink_block_group->flags);
5990 if (new_alloc_flags != shrink_block_group->flags) {
5991 calc =
5992 btrfs_block_group_used(&shrink_block_group->item);
5993 } else {
5994 calc = shrink_block_group->key.offset;
5995 }
c286ac48
CM		 5996 spin_unlock(&shrink_block_group->lock);
5997
0ef3e66b
CM		 5998 do_chunk_alloc(trans, root->fs_info->extent_root,
5999 calc + 2 * 1024 * 1024, new_alloc_flags, force);
7d9eb12c 6000
0ef3e66b 6001 btrfs_end_transaction(trans, root);
c286ac48
CM		 6002 } else
6003 spin_unlock(&shrink_block_group->lock);
0ef3e66b
CM		 6004 return 0;
6005}
6006
1a40e23b
ZY		 6007static int __insert_orphan_inode(struct btrfs_trans_handle *trans,
6008 struct btrfs_root *root,
6009 u64 objectid, u64 size)
6010{
6011 struct btrfs_path *path;
6012 struct btrfs_inode_item *item;
6013 struct extent_buffer *leaf;
6014 int ret;
6015
6016 path = btrfs_alloc_path();
6017 if (!path)
6018 return -ENOMEM;
6019
6020 ret = btrfs_insert_empty_inode(trans, root, path, objectid);
6021 if (ret)
6022 goto out;
6023
6024 leaf = path->nodes[0];
6025 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_inode_item);
6026 memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item));
6027 btrfs_set_inode_generation(leaf, item, 1);
6028 btrfs_set_inode_size(leaf, item, size);
6029 btrfs_set_inode_mode(leaf, item, S_IFREG | 0600);
0403e47e 6030 btrfs_set_inode_flags(leaf, item, BTRFS_INODE_NOCOMPRESS);
1a40e23b
ZY		 6031 btrfs_mark_buffer_dirty(leaf);
6032 btrfs_release_path(root, path);
6033out:
6034 btrfs_free_path(path);
6035 return ret;
6036}
6037
d397712b 6038static noinline struct inode *create_reloc_inode(struct btrfs_fs_info *fs_info,
1a40e23b
ZY		 6039 struct btrfs_block_group_cache *group)
6040{
6041 struct inode *inode = NULL;
6042 struct btrfs_trans_handle *trans;
6043 struct btrfs_root *root;
6044 struct btrfs_key root_key;
6045 u64 objectid = BTRFS_FIRST_FREE_OBJECTID;
6046 int err = 0;
6047
6048 root_key.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
6049 root_key.type = BTRFS_ROOT_ITEM_KEY;
6050 root_key.offset = (u64)-1;
6051 root = btrfs_read_fs_root_no_name(fs_info, &root_key);
6052 if (IS_ERR(root))
6053 return ERR_CAST(root);
6054
6055 trans = btrfs_start_transaction(root, 1);
6056 BUG_ON(!trans);
6057
6058 err = btrfs_find_free_objectid(trans, root, objectid, &objectid);
6059 if (err)
6060 goto out;
6061
6062 err = __insert_orphan_inode(trans, root, objectid, group->key.offset);
6063 BUG_ON(err);
6064
6065 err = btrfs_insert_file_extent(trans, root, objectid, 0, 0, 0,
c8b97818
CM		 6066 group->key.offset, 0, group->key.offset,
6067 0, 0, 0);
1a40e23b
ZY		 6068 BUG_ON(err);
6069
6070 inode = btrfs_iget_locked(root->fs_info->sb, objectid, root);
6071 if (inode->i_state & I_NEW) {
6072 BTRFS_I(inode)->root = root;
6073 BTRFS_I(inode)->location.objectid = objectid;
6074 BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
6075 BTRFS_I(inode)->location.offset = 0;
6076 btrfs_read_locked_inode(inode);
6077 unlock_new_inode(inode);
6078 BUG_ON(is_bad_inode(inode));
6079 } else {
6080 BUG_ON(1);
6081 }
17d217fe 6082 BTRFS_I(inode)->index_cnt = group->key.objectid;
1a40e23b
ZY		 6083
6084 err = btrfs_orphan_add(trans, inode);
6085out:
6086 btrfs_end_transaction(trans, root);
6087 if (err) {
6088 if (inode)
6089 iput(inode);
6090 inode = ERR_PTR(err);
6091 }
6092 return inode;
6093}
6094
17d217fe
YZ		 6095int btrfs_reloc_clone_csums(struct inode *inode, u64 file_pos, u64 len)
6096{
6097
6098 struct btrfs_ordered_sum *sums;
6099 struct btrfs_sector_sum *sector_sum;
6100 struct btrfs_ordered_extent *ordered;
6101 struct btrfs_root *root = BTRFS_I(inode)->root;
6102 struct list_head list;
6103 size_t offset;
6104 int ret;
6105 u64 disk_bytenr;
6106
6107 INIT_LIST_HEAD(&list);
6108
6109 ordered = btrfs_lookup_ordered_extent(inode, file_pos);
6110 BUG_ON(ordered->file_offset != file_pos || ordered->len != len);
6111
6112 disk_bytenr = file_pos + BTRFS_I(inode)->index_cnt;
07d400a6 6113 ret = btrfs_lookup_csums_range(root->fs_info->csum_root, disk_bytenr,
17d217fe
YZ		 6114 disk_bytenr + len - 1, &list);
6115
6116 while (!list_empty(&list)) {
6117 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
6118 list_del_init(&sums->list);
6119
6120 sector_sum = sums->sums;
6121 sums->bytenr = ordered->start;
6122
6123 offset = 0;
6124 while (offset < sums->len) {
6125 sector_sum->bytenr += ordered->start - disk_bytenr;
6126 sector_sum++;
6127 offset += root->sectorsize;
6128 }
6129
6130 btrfs_add_ordered_sum(inode, ordered, sums);
6131 }
6132 btrfs_put_ordered_extent(ordered);
6133 return 0;
6134}
6135
1a40e23b 6136int btrfs_relocate_block_group(struct btrfs_root *root, u64 group_start)
edbd8d4e
CM		 6137{
6138 struct btrfs_trans_handle *trans;
edbd8d4e 6139 struct btrfs_path *path;
1a40e23b
ZY		 6140 struct btrfs_fs_info *info = root->fs_info;
6141 struct extent_buffer *leaf;
6142 struct inode *reloc_inode;
6143 struct btrfs_block_group_cache *block_group;
6144 struct btrfs_key key;
d899e052 6145 u64 skipped;
edbd8d4e
CM		 6146 u64 cur_byte;
6147 u64 total_found;
edbd8d4e
CM		 6148 u32 nritems;
6149 int ret;
a061fc8d 6150 int progress;
1a40e23b 6151 int pass = 0;
edbd8d4e 6152
1a40e23b
ZY		 6153 root = root->fs_info->extent_root;
6154
6155 block_group = btrfs_lookup_block_group(info, group_start);
6156 BUG_ON(!block_group);
8f18cf13 6157
d397712b 6158 printk(KERN_INFO "btrfs relocating block group %llu flags %llu\n",
1a40e23b
ZY		 6159 (unsigned long long)block_group->key.objectid,
6160 (unsigned long long)block_group->flags);
8f18cf13 6161
edbd8d4e 6162 path = btrfs_alloc_path();
1a40e23b 6163 BUG_ON(!path);
edbd8d4e 6164
1a40e23b
ZY		 6165 reloc_inode = create_reloc_inode(info, block_group);
6166 BUG_ON(IS_ERR(reloc_inode));
323da79c 6167
1a40e23b 6168 __alloc_chunk_for_shrink(root, block_group, 1);
c146afad 6169 set_block_group_readonly(block_group);
323da79c 6170
1a40e23b
ZY		 6171 btrfs_start_delalloc_inodes(info->tree_root);
6172 btrfs_wait_ordered_extents(info->tree_root, 0);
6173again:
d899e052 6174 skipped = 0;
edbd8d4e 6175 total_found = 0;
a061fc8d 6176 progress = 0;
1a40e23b 6177 key.objectid = block_group->key.objectid;
edbd8d4e
CM		 6178 key.offset = 0;
6179 key.type = 0;
73e48b27 6180 cur_byte = key.objectid;
4313b399 6181
1a40e23b
ZY		 6182 trans = btrfs_start_transaction(info->tree_root, 1);
6183 btrfs_commit_transaction(trans, info->tree_root);
ea8c2819 6184
1a40e23b
ZY		 6185 mutex_lock(&root->fs_info->cleaner_mutex);
6186 btrfs_clean_old_snapshots(info->tree_root);
6187 btrfs_remove_leaf_refs(info->tree_root, (u64)-1, 1);
6188 mutex_unlock(&root->fs_info->cleaner_mutex);
ea8c2819 6189
d397712b 6190 while (1) {
edbd8d4e
CM		 6191 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6192 if (ret < 0)
6193 goto out;
7d9eb12c 6194next:
edbd8d4e 6195 leaf = path->nodes[0];
73e48b27 6196 nritems = btrfs_header_nritems(leaf);
73e48b27
Y		 6197 if (path->slots[0] >= nritems) {
6198 ret = btrfs_next_leaf(root, path);
6199 if (ret < 0)
6200 goto out;
6201 if (ret == 1) {
6202 ret = 0;
6203 break;
edbd8d4e 6204 }
73e48b27
Y		 6205 leaf = path->nodes[0];
6206 nritems = btrfs_header_nritems(leaf);
edbd8d4e 6207 }
73e48b27 6208
1a40e23b 6209 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
725c8463 6210
1a40e23b
ZY		 6211 if (key.objectid >= block_group->key.objectid +
6212 block_group->key.offset)
8f18cf13
CM		 6213 break;
6214
725c8463 6215 if (progress && need_resched()) {
725c8463 6216 btrfs_release_path(root, path);
1a40e23b 6217 cond_resched();
725c8463 6218 progress = 0;
1a40e23b 6219 continue;
725c8463
CM		 6220 }
6221 progress = 1;
6222
1a40e23b
ZY		 6223 if (btrfs_key_type(&key) != BTRFS_EXTENT_ITEM_KEY ||
6224 key.objectid + key.offset <= cur_byte) {
edbd8d4e 6225 path->slots[0]++;
edbd8d4e
CM		 6226 goto next;
6227 }
73e48b27 6228
edbd8d4e 6229 total_found++;
1a40e23b 6230 cur_byte = key.objectid + key.offset;
edbd8d4e 6231 btrfs_release_path(root, path);
edbd8d4e 6232
1a40e23b
ZY		 6233 __alloc_chunk_for_shrink(root, block_group, 0);
6234 ret = relocate_one_extent(root, path, &key, block_group,
6235 reloc_inode, pass);
6236 BUG_ON(ret < 0);
d899e052
YZ		 6237 if (ret > 0)
6238 skipped++;
edbd8d4e 6239
1a40e23b
ZY		 6240 key.objectid = cur_byte;
6241 key.type = 0;
6242 key.offset = 0;
edbd8d4e
CM		 6243 }
6244
1a40e23b 6245 btrfs_release_path(root, path);
7d9eb12c 6246
1a40e23b
ZY		 6247 if (pass == 0) {
6248 btrfs_wait_ordered_range(reloc_inode, 0, (u64)-1);
6249 invalidate_mapping_pages(reloc_inode->i_mapping, 0, -1);
8e8a1e31 6250 }
73e48b27 6251
1a40e23b 6252 if (total_found > 0) {
d397712b 6253 printk(KERN_INFO "btrfs found %llu extents in pass %d\n",
1a40e23b
ZY		 6254 (unsigned long long)total_found, pass);
6255 pass++;
d899e052
YZ		 6256 if (total_found == skipped && pass > 2) {
6257 iput(reloc_inode);
6258 reloc_inode = create_reloc_inode(info, block_group);
6259 pass = 0;
6260 }
1a40e23b 6261 goto again;
0f9dd46c 6262 }
0ef3e66b 6263
1a40e23b
ZY		 6264 /* delete reloc_inode */
6265 iput(reloc_inode);
6266
6267 /* unpin extents in this range */
6268 trans = btrfs_start_transaction(info->tree_root, 1);
6269 btrfs_commit_transaction(trans, info->tree_root);
0ef3e66b 6270
1a40e23b
ZY		 6271 spin_lock(&block_group->lock);
6272 WARN_ON(block_group->pinned > 0);
6273 WARN_ON(block_group->reserved > 0);
6274 WARN_ON(btrfs_block_group_used(&block_group->item) > 0);
6275 spin_unlock(&block_group->lock);
d2fb3437 6276 put_block_group(block_group);
1a40e23b 6277 ret = 0;
edbd8d4e 6278out:
1a40e23b 6279 btrfs_free_path(path);
edbd8d4e
CM		 6280 return ret;
6281}
6282
b2950863
CH		 6283static int find_first_block_group(struct btrfs_root *root,
6284 struct btrfs_path *path, struct btrfs_key *key)
0b86a832 6285{
925baedd 6286 int ret = 0;
0b86a832
CM		 6287 struct btrfs_key found_key;
6288 struct extent_buffer *leaf;
6289 int slot;
edbd8d4e 6290
0b86a832
CM		 6291 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
6292 if (ret < 0)
925baedd
CM		 6293 goto out;
6294
d397712b 6295 while (1) {
0b86a832 6296 slot = path->slots[0];
edbd8d4e 6297 leaf = path->nodes[0];
0b86a832
CM		 6298 if (slot >= btrfs_header_nritems(leaf)) {
6299 ret = btrfs_next_leaf(root, path);
6300 if (ret == 0)
6301 continue;
6302 if (ret < 0)
925baedd 6303 goto out;
0b86a832 6304 break;
edbd8d4e 6305 }
0b86a832 6306 btrfs_item_key_to_cpu(leaf, &found_key, slot);
edbd8d4e 6307
0b86a832 6308 if (found_key.objectid >= key->objectid &&
925baedd
CM		 6309 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
6310 ret = 0;
6311 goto out;
6312 }
0b86a832 6313 path->slots[0]++;
edbd8d4e 6314 }
0b86a832 6315 ret = -ENOENT;
925baedd 6316out:
0b86a832 6317 return ret;
edbd8d4e
CM		 6318}
6319
1a40e23b
ZY		 6320int btrfs_free_block_groups(struct btrfs_fs_info *info)
6321{
6322 struct btrfs_block_group_cache *block_group;
6323 struct rb_node *n;
6324
1a40e23b
ZY		 6325 spin_lock(&info->block_group_cache_lock);
6326 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
6327 block_group = rb_entry(n, struct btrfs_block_group_cache,
6328 cache_node);
1a40e23b
ZY		 6329 rb_erase(&block_group->cache_node,
6330 &info->block_group_cache_tree);
d899e052
YZ		 6331 spin_unlock(&info->block_group_cache_lock);
6332
6333 btrfs_remove_free_space_cache(block_group);
80eb234a 6334 down_write(&block_group->space_info->groups_sem);
1a40e23b 6335 list_del(&block_group->list);
80eb234a 6336 up_write(&block_group->space_info->groups_sem);
d2fb3437
YZ		 6337
6338 WARN_ON(atomic_read(&block_group->count) != 1);
1a40e23b 6339 kfree(block_group);
d899e052
YZ		 6340
6341 spin_lock(&info->block_group_cache_lock);
1a40e23b
ZY		 6342 }
6343 spin_unlock(&info->block_group_cache_lock);
1a40e23b
ZY		 6344 return 0;
6345}
6346
9078a3e1
CM		 6347int btrfs_read_block_groups(struct btrfs_root *root)
6348{
6349 struct btrfs_path *path;
6350 int ret;
9078a3e1 6351 struct btrfs_block_group_cache *cache;
be744175 6352 struct btrfs_fs_info *info = root->fs_info;
6324fbf3 6353 struct btrfs_space_info *space_info;
9078a3e1
CM		 6354 struct btrfs_key key;
6355 struct btrfs_key found_key;
5f39d397 6356 struct extent_buffer *leaf;
96b5179d 6357
be744175 6358 root = info->extent_root;
9078a3e1 6359 key.objectid = 0;
0b86a832 6360 key.offset = 0;
9078a3e1 6361 btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
9078a3e1
CM		 6362 path = btrfs_alloc_path();
6363 if (!path)
6364 return -ENOMEM;
6365
d397712b 6366 while (1) {
0b86a832
CM		 6367 ret = find_first_block_group(root, path, &key);
6368 if (ret > 0) {
6369 ret = 0;
6370 goto error;
9078a3e1 6371 }
0b86a832
CM		 6372 if (ret != 0)
6373 goto error;
6374
5f39d397
CM		 6375 leaf = path->nodes[0];
6376 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
8f18cf13 6377 cache = kzalloc(sizeof(*cache), GFP_NOFS);
9078a3e1 6378 if (!cache) {
0b86a832 6379 ret = -ENOMEM;
9078a3e1
CM		 6380 break;
6381 }
3e1ad54f 6382
d2fb3437 6383 atomic_set(&cache->count, 1);
c286ac48 6384 spin_lock_init(&cache->lock);
25179201 6385 mutex_init(&cache->alloc_mutex);
ea6a478e 6386 mutex_init(&cache->cache_mutex);
0f9dd46c 6387 INIT_LIST_HEAD(&cache->list);
5f39d397
CM		 6388 read_extent_buffer(leaf, &cache->item,
6389 btrfs_item_ptr_offset(leaf, path->slots[0]),
6390 sizeof(cache->item));
9078a3e1 6391 memcpy(&cache->key, &found_key, sizeof(found_key));
0b86a832 6392
9078a3e1
CM		 6393 key.objectid = found_key.objectid + found_key.offset;
6394 btrfs_release_path(root, path);
0b86a832 6395 cache->flags = btrfs_block_group_flags(&cache->item);
96b5179d 6396
6324fbf3
CM		 6397 ret = update_space_info(info, cache->flags, found_key.offset,
6398 btrfs_block_group_used(&cache->item),
6399 &space_info);
6400 BUG_ON(ret);
6401 cache->space_info = space_info;
80eb234a
JB		 6402 down_write(&space_info->groups_sem);
6403 list_add_tail(&cache->list, &space_info->block_groups);
6404 up_write(&space_info->groups_sem);
0f9dd46c
JB		 6405
6406 ret = btrfs_add_block_group_cache(root->fs_info, cache);
6407 BUG_ON(ret);
75ccf47d
CM		 6408
6409 set_avail_alloc_bits(root->fs_info, cache->flags);
2b82032c
YZ		 6410 if (btrfs_chunk_readonly(root, cache->key.objectid))
6411 set_block_group_readonly(cache);
9078a3e1 6412 }
0b86a832
CM		 6413 ret = 0;
6414error:
9078a3e1 6415 btrfs_free_path(path);
0b86a832 6416 return ret;
9078a3e1 6417}
6324fbf3
CM		 6418
6419int btrfs_make_block_group(struct btrfs_trans_handle *trans,
6420 struct btrfs_root *root, u64 bytes_used,
e17cade2 6421 u64 type, u64 chunk_objectid, u64 chunk_offset,
6324fbf3
CM		 6422 u64 size)
6423{
6424 int ret;
6324fbf3
CM		 6425 struct btrfs_root *extent_root;
6426 struct btrfs_block_group_cache *cache;
6324fbf3
CM		 6427
6428 extent_root = root->fs_info->extent_root;
6324fbf3 6429
e02119d5
CM		 6430 root->fs_info->last_trans_new_blockgroup = trans->transid;
6431
8f18cf13 6432 cache = kzalloc(sizeof(*cache), GFP_NOFS);
0f9dd46c
JB		 6433 if (!cache)
6434 return -ENOMEM;
6435
e17cade2 6436 cache->key.objectid = chunk_offset;
6324fbf3 6437 cache->key.offset = size;
d2fb3437
YZ		 6438 cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
6439 atomic_set(&cache->count, 1);
c286ac48 6440 spin_lock_init(&cache->lock);
25179201 6441 mutex_init(&cache->alloc_mutex);
ea6a478e 6442 mutex_init(&cache->cache_mutex);
0f9dd46c 6443 INIT_LIST_HEAD(&cache->list);
0ef3e66b 6444
6324fbf3 6445 btrfs_set_block_group_used(&cache->item, bytes_used);
6324fbf3
CM		 6446 btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
6447 cache->flags = type;
6448 btrfs_set_block_group_flags(&cache->item, type);
6449
6450 ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
6451 &cache->space_info);
6452 BUG_ON(ret);
80eb234a
JB		 6453 down_write(&cache->space_info->groups_sem);
6454 list_add_tail(&cache->list, &cache->space_info->block_groups);
6455 up_write(&cache->space_info->groups_sem);
6324fbf3 6456
0f9dd46c
JB		 6457 ret = btrfs_add_block_group_cache(root->fs_info, cache);
6458 BUG_ON(ret);
c286ac48 6459
6324fbf3
CM		 6460 ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
6461 sizeof(cache->item));
6462 BUG_ON(ret);
6463
87ef2bb4
CM		 6464 finish_current_insert(trans, extent_root, 0);
6465 ret = del_pending_extents(trans, extent_root, 0);
6324fbf3 6466 BUG_ON(ret);
d18a2c44 6467 set_avail_alloc_bits(extent_root->fs_info, type);
925baedd 6468
6324fbf3
CM		 6469 return 0;
6470}
1a40e23b
ZY		 6471
6472int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
6473 struct btrfs_root *root, u64 group_start)
6474{
6475 struct btrfs_path *path;
6476 struct btrfs_block_group_cache *block_group;
6477 struct btrfs_key key;
6478 int ret;
6479
1a40e23b
ZY		 6480 root = root->fs_info->extent_root;
6481
6482 block_group = btrfs_lookup_block_group(root->fs_info, group_start);
6483 BUG_ON(!block_group);
c146afad 6484 BUG_ON(!block_group->ro);
1a40e23b
ZY		 6485
6486 memcpy(&key, &block_group->key, sizeof(key));
6487
6488 path = btrfs_alloc_path();
6489 BUG_ON(!path);
6490
3dfdb934 6491 spin_lock(&root->fs_info->block_group_cache_lock);
1a40e23b
ZY		 6492 rb_erase(&block_group->cache_node,
6493 &root->fs_info->block_group_cache_tree);
3dfdb934
YZ		 6494 spin_unlock(&root->fs_info->block_group_cache_lock);
6495 btrfs_remove_free_space_cache(block_group);
80eb234a 6496 down_write(&block_group->space_info->groups_sem);
1a40e23b 6497 list_del(&block_group->list);
80eb234a 6498 up_write(&block_group->space_info->groups_sem);
1a40e23b 6499
c146afad
YZ		 6500 spin_lock(&block_group->space_info->lock);
6501 block_group->space_info->total_bytes -= block_group->key.offset;
6502 block_group->space_info->bytes_readonly -= block_group->key.offset;
6503 spin_unlock(&block_group->space_info->lock);
2b82032c 6504 block_group->space_info->full = 0;
c146afad 6505
d2fb3437
YZ		 6506 put_block_group(block_group);
6507 put_block_group(block_group);
1a40e23b
ZY		 6508
6509 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
6510 if (ret > 0)
6511 ret = -EIO;
6512 if (ret < 0)
6513 goto out;
6514
6515 ret = btrfs_del_item(trans, root, path);
6516out:
6517 btrfs_free_path(path);
6518 return ret;
6519}
Linux 6.x block layer and io_uring tree(s)