projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Btrfs: mnt_drop_write in ioctl_trans_end
[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>
4b4e25f2
CM		 22#include <linux/version.h>
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"
15916de8 33#include "compat.h"
fec577fb 34
31840ae1
ZY		 35#define PENDING_EXTENT_INSERT 0
36#define PENDING_EXTENT_DELETE 1
37#define PENDING_BACKREF_UPDATE 2
38
39struct pending_extent_op {
40 int type;
41 u64 bytenr;
42 u64 num_bytes;
43 u64 parent;
44 u64 orig_parent;
45 u64 generation;
46 u64 orig_generation;
47 int level;
f3465ca4
JB		 48 struct list_head list;
49 int del;
31840ae1
ZY		 50};
51
e089f05c 52static int finish_current_insert(struct btrfs_trans_handle *trans, struct
87ef2bb4 53 btrfs_root *extent_root, int all);
e20d96d6 54static int del_pending_extents(struct btrfs_trans_handle *trans, struct
87ef2bb4 55 btrfs_root *extent_root, int all);
925baedd
CM		 56static struct btrfs_block_group_cache *
57__btrfs_find_block_group(struct btrfs_root *root,
58 struct btrfs_block_group_cache *hint,
59 u64 search_start, int data, int owner);
f3465ca4
JB		 60static int pin_down_bytes(struct btrfs_trans_handle *trans,
61 struct btrfs_root *root,
62 u64 bytenr, u64 num_bytes, int is_data);
63static int update_block_group(struct btrfs_trans_handle *trans,
64 struct btrfs_root *root,
65 u64 bytenr, u64 num_bytes, int alloc,
66 int mark_free);
d548ee51 67
0f9dd46c
JB		 68static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
69{
70 return (cache->flags & bits) == bits;
71}
72
73/*
74 * this adds the block group to the fs_info rb tree for the block group
75 * cache
76 */
b2950863 77static int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
0f9dd46c
JB		 78 struct btrfs_block_group_cache *block_group)
79{
80 struct rb_node **p;
81 struct rb_node *parent = NULL;
82 struct btrfs_block_group_cache *cache;
83
84 spin_lock(&info->block_group_cache_lock);
85 p = &info->block_group_cache_tree.rb_node;
86
87 while (*p) {
88 parent = *p;
89 cache = rb_entry(parent, struct btrfs_block_group_cache,
90 cache_node);
91 if (block_group->key.objectid < cache->key.objectid) {
92 p = &(*p)->rb_left;
93 } else if (block_group->key.objectid > cache->key.objectid) {
94 p = &(*p)->rb_right;
95 } else {
96 spin_unlock(&info->block_group_cache_lock);
97 return -EEXIST;
98 }
99 }
100
101 rb_link_node(&block_group->cache_node, parent, p);
102 rb_insert_color(&block_group->cache_node,
103 &info->block_group_cache_tree);
104 spin_unlock(&info->block_group_cache_lock);
105
106 return 0;
107}
108
109/*
110 * This will return the block group at or after bytenr if contains is 0, else
111 * it will return the block group that contains the bytenr
112 */
113static struct btrfs_block_group_cache *
114block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
115 int contains)
116{
117 struct btrfs_block_group_cache *cache, *ret = NULL;
118 struct rb_node *n;
119 u64 end, start;
120
121 spin_lock(&info->block_group_cache_lock);
122 n = info->block_group_cache_tree.rb_node;
123
124 while (n) {
125 cache = rb_entry(n, struct btrfs_block_group_cache,
126 cache_node);
127 end = cache->key.objectid + cache->key.offset - 1;
128 start = cache->key.objectid;
129
130 if (bytenr < start) {
131 if (!contains && (!ret || start < ret->key.objectid))
132 ret = cache;
133 n = n->rb_left;
134 } else if (bytenr > start) {
135 if (contains && bytenr <= end) {
136 ret = cache;
137 break;
138 }
139 n = n->rb_right;
140 } else {
141 ret = cache;
142 break;
143 }
144 }
145 spin_unlock(&info->block_group_cache_lock);
146
147 return ret;
148}
149
150/*
151 * this is only called by cache_block_group, since we could have freed extents
152 * we need to check the pinned_extents for any extents that can't be used yet
153 * since their free space will be released as soon as the transaction commits.
154 */
155static int add_new_free_space(struct btrfs_block_group_cache *block_group,
156 struct btrfs_fs_info *info, u64 start, u64 end)
157{
158 u64 extent_start, extent_end, size;
159 int ret;
160
25179201 161 mutex_lock(&info->pinned_mutex);
0f9dd46c
JB		 162 while (start < end) {
163 ret = find_first_extent_bit(&info->pinned_extents, start,
164 &extent_start, &extent_end,
165 EXTENT_DIRTY);
166 if (ret)
167 break;
168
169 if (extent_start == start) {
170 start = extent_end + 1;
171 } else if (extent_start > start && extent_start < end) {
172 size = extent_start - start;
ea6a478e
JB		 173 ret = btrfs_add_free_space(block_group, start,
174 size);
0f9dd46c
JB		 175 BUG_ON(ret);
176 start = extent_end + 1;
177 } else {
178 break;
179 }
180 }
181
182 if (start < end) {
183 size = end - start;
ea6a478e 184 ret = btrfs_add_free_space(block_group, start, size);
0f9dd46c
JB		 185 BUG_ON(ret);
186 }
25179201 187 mutex_unlock(&info->pinned_mutex);
0f9dd46c
JB		 188
189 return 0;
190}
191
a512bbf8
YZ		 192static int remove_sb_from_cache(struct btrfs_root *root,
193 struct btrfs_block_group_cache *cache)
194{
195 u64 bytenr;
196 u64 *logical;
197 int stripe_len;
198 int i, nr, ret;
199
200 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
201 bytenr = btrfs_sb_offset(i);
202 ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
203 cache->key.objectid, bytenr, 0,
204 &logical, &nr, &stripe_len);
205 BUG_ON(ret);
206 while (nr--) {
207 btrfs_remove_free_space(cache, logical[nr],
208 stripe_len);
209 }
210 kfree(logical);
211 }
212 return 0;
213}
214
e37c9e69
CM		 215static int cache_block_group(struct btrfs_root *root,
216 struct btrfs_block_group_cache *block_group)
217{
218 struct btrfs_path *path;
ef8bbdfe 219 int ret = 0;
e37c9e69 220 struct btrfs_key key;
5f39d397 221 struct extent_buffer *leaf;
e37c9e69 222 int slot;
a512bbf8 223 u64 last = block_group->key.objectid;
e37c9e69 224
00f5c795
CM		 225 if (!block_group)
226 return 0;
227
e37c9e69 228 root = root->fs_info->extent_root;
e37c9e69
CM		 229
230 if (block_group->cached)
231 return 0;
f510cfec 232
e37c9e69
CM		 233 path = btrfs_alloc_path();
234 if (!path)
235 return -ENOMEM;
7d7d6068 236
2cc58cf2 237 path->reada = 2;
5cd57b2c
CM		 238 /*
239 * we get into deadlocks with paths held by callers of this function.
240 * since the alloc_mutex is protecting things right now, just
241 * skip the locking here
242 */
243 path->skip_locking = 1;
0403e47e 244 key.objectid = max_t(u64, last, BTRFS_SUPER_INFO_OFFSET);
e37c9e69
CM		 245 key.offset = 0;
246 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
247 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
248 if (ret < 0)
ef8bbdfe 249 goto err;
a512bbf8 250
e37c9e69 251 while(1) {
5f39d397 252 leaf = path->nodes[0];
e37c9e69 253 slot = path->slots[0];
5f39d397 254 if (slot >= btrfs_header_nritems(leaf)) {
e37c9e69 255 ret = btrfs_next_leaf(root, path);
54aa1f4d
CM		 256 if (ret < 0)
257 goto err;
0f9dd46c 258 if (ret == 0)
e37c9e69 259 continue;
0f9dd46c 260 else
e37c9e69 261 break;
e37c9e69 262 }
5f39d397 263 btrfs_item_key_to_cpu(leaf, &key, slot);
0f9dd46c 264 if (key.objectid < block_group->key.objectid)
7d7d6068 265 goto next;
0f9dd46c 266
e37c9e69 267 if (key.objectid >= block_group->key.objectid +
0f9dd46c 268 block_group->key.offset)
e37c9e69 269 break;
7d7d6068 270
e37c9e69 271 if (btrfs_key_type(&key) == BTRFS_EXTENT_ITEM_KEY) {
0f9dd46c
JB		 272 add_new_free_space(block_group, root->fs_info, last,
273 key.objectid);
274
7d7d6068 275 last = key.objectid + key.offset;
e37c9e69 276 }
7d7d6068 277next:
e37c9e69
CM		 278 path->slots[0]++;
279 }
280
0f9dd46c
JB		 281 add_new_free_space(block_group, root->fs_info, last,
282 block_group->key.objectid +
283 block_group->key.offset);
284
a512bbf8 285 remove_sb_from_cache(root, block_group);
e37c9e69 286 block_group->cached = 1;
ef8bbdfe 287 ret = 0;
54aa1f4d 288err:
e37c9e69 289 btrfs_free_path(path);
ef8bbdfe 290 return ret;
e37c9e69
CM		 291}
292
0f9dd46c
JB		 293/*
294 * return the block group that starts at or after bytenr
295 */
b2950863 296static struct btrfs_block_group_cache *btrfs_lookup_first_block_group(struct
0ef3e66b
CM		 297 btrfs_fs_info *info,
298 u64 bytenr)
299{
0f9dd46c 300 struct btrfs_block_group_cache *cache;
0ef3e66b 301
0f9dd46c 302 cache = block_group_cache_tree_search(info, bytenr, 0);
0ef3e66b 303
0f9dd46c 304 return cache;
0ef3e66b
CM		 305}
306
0f9dd46c
JB		 307/*
308 * return the block group that contains teh given bytenr
309 */
5276aeda
CM		 310struct btrfs_block_group_cache *btrfs_lookup_block_group(struct
311 btrfs_fs_info *info,
db94535d 312 u64 bytenr)
be744175 313{
0f9dd46c 314 struct btrfs_block_group_cache *cache;
be744175 315
0f9dd46c 316 cache = block_group_cache_tree_search(info, bytenr, 1);
96b5179d 317
0f9dd46c 318 return cache;
be744175 319}
0b86a832 320
0f9dd46c
JB		 321static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
322 u64 flags)
6324fbf3 323{
0f9dd46c
JB		 324 struct list_head *head = &info->space_info;
325 struct list_head *cur;
326 struct btrfs_space_info *found;
327 list_for_each(cur, head) {
328 found = list_entry(cur, struct btrfs_space_info, list);
329 if (found->flags == flags)
330 return found;
331 }
332 return NULL;
6324fbf3
CM		 333}
334
80eb234a
JB		 335static u64 div_factor(u64 num, int factor)
336{
337 if (factor == 10)
338 return num;
339 num *= factor;
340 do_div(num, 10);
341 return num;
342}
343
925baedd
CM		 344static struct btrfs_block_group_cache *
345__btrfs_find_block_group(struct btrfs_root *root,
346 struct btrfs_block_group_cache *hint,
347 u64 search_start, int data, int owner)
cd1bc465 348{
96b5179d 349 struct btrfs_block_group_cache *cache;
31f3c99b 350 struct btrfs_block_group_cache *found_group = NULL;
cd1bc465
CM		 351 struct btrfs_fs_info *info = root->fs_info;
352 u64 used;
31f3c99b 353 u64 last = 0;
96b5179d 354 u64 free_check;
31f3c99b 355 int full_search = 0;
bce4eae9 356 int factor = 10;
0ef3e66b 357 int wrapped = 0;
de428b63 358
a236aed1
CM		 359 if (data & BTRFS_BLOCK_GROUP_METADATA)
360 factor = 9;
be744175 361
0ef3e66b 362 if (search_start) {
be744175 363 struct btrfs_block_group_cache *shint;
0ef3e66b 364 shint = btrfs_lookup_first_block_group(info, search_start);
2b82032c 365 if (shint && block_group_bits(shint, data)) {
c286ac48 366 spin_lock(&shint->lock);
be744175 367 used = btrfs_block_group_used(&shint->item);
e8569813 368 if (used + shint->pinned + shint->reserved <
324ae4df 369 div_factor(shint->key.offset, factor)) {
c286ac48 370 spin_unlock(&shint->lock);
be744175
CM		 371 return shint;
372 }
c286ac48 373 spin_unlock(&shint->lock);
be744175
CM		 374 }
375 }
2b82032c 376 if (hint && block_group_bits(hint, data)) {
c286ac48 377 spin_lock(&hint->lock);
31f3c99b 378 used = btrfs_block_group_used(&hint->item);
e8569813 379 if (used + hint->pinned + hint->reserved <
324ae4df 380 div_factor(hint->key.offset, factor)) {
c286ac48 381 spin_unlock(&hint->lock);
31f3c99b
CM		 382 return hint;
383 }
c286ac48 384 spin_unlock(&hint->lock);
e19caa5f 385 last = hint->key.objectid + hint->key.offset;
31f3c99b 386 } else {
e37c9e69 387 if (hint)
0ef3e66b 388 last = max(hint->key.objectid, search_start);
e37c9e69 389 else
0ef3e66b 390 last = search_start;
31f3c99b 391 }
31f3c99b 392again:
e8569813
ZY		 393 while (1) {
394 cache = btrfs_lookup_first_block_group(root->fs_info, last);
0f9dd46c
JB		 395 if (!cache)
396 break;
96b5179d 397
c286ac48 398 spin_lock(&cache->lock);
96b5179d
CM		 399 last = cache->key.objectid + cache->key.offset;
400 used = btrfs_block_group_used(&cache->item);
401
2b82032c 402 if (block_group_bits(cache, data)) {
0ef3e66b 403 free_check = div_factor(cache->key.offset, factor);
e8569813
ZY		 404 if (used + cache->pinned + cache->reserved <
405 free_check) {
8790d502 406 found_group = cache;
c286ac48 407 spin_unlock(&cache->lock);
8790d502
CM		 408 goto found;
409 }
6324fbf3 410 }
c286ac48 411 spin_unlock(&cache->lock);
de428b63 412 cond_resched();
cd1bc465 413 }
0ef3e66b
CM		 414 if (!wrapped) {
415 last = search_start;
416 wrapped = 1;
417 goto again;
418 }
419 if (!full_search && factor < 10) {
be744175 420 last = search_start;
31f3c99b 421 full_search = 1;
0ef3e66b 422 factor = 10;
31f3c99b
CM		 423 goto again;
424 }
be744175 425found:
31f3c99b 426 return found_group;
cd1bc465
CM		 427}
428
925baedd
CM		 429struct btrfs_block_group_cache *btrfs_find_block_group(struct btrfs_root *root,
430 struct btrfs_block_group_cache
431 *hint, u64 search_start,
432 int data, int owner)
433{
434
435 struct btrfs_block_group_cache *ret;
925baedd 436 ret = __btrfs_find_block_group(root, hint, search_start, data, owner);
925baedd
CM		 437 return ret;
438}
0f9dd46c 439
e02119d5 440/* simple helper to search for an existing extent at a given offset */
31840ae1 441int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len)
e02119d5
CM		 442{
443 int ret;
444 struct btrfs_key key;
31840ae1 445 struct btrfs_path *path;
e02119d5 446
31840ae1
ZY		 447 path = btrfs_alloc_path();
448 BUG_ON(!path);
e02119d5
CM		 449 key.objectid = start;
450 key.offset = len;
451 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
452 ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
453 0, 0);
31840ae1 454 btrfs_free_path(path);
7bb86316
CM		 455 return ret;
456}
457
d8d5f3e1
CM		 458/*
459 * Back reference rules. Back refs have three main goals:
460 *
461 * 1) differentiate between all holders of references to an extent so that
462 * when a reference is dropped we can make sure it was a valid reference
463 * before freeing the extent.
464 *
465 * 2) Provide enough information to quickly find the holders of an extent
466 * if we notice a given block is corrupted or bad.
467 *
468 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
469 * maintenance. This is actually the same as #2, but with a slightly
470 * different use case.
471 *
472 * File extents can be referenced by:
473 *
474 * - multiple snapshots, subvolumes, or different generations in one subvol
31840ae1 475 * - different files inside a single subvolume
d8d5f3e1
CM		 476 * - different offsets inside a file (bookend extents in file.c)
477 *
478 * The extent ref structure has fields for:
479 *
480 * - Objectid of the subvolume root
481 * - Generation number of the tree holding the reference
482 * - objectid of the file holding the reference
31840ae1
ZY		 483 * - number of references holding by parent node (alway 1 for tree blocks)
484 *
485 * Btree leaf may hold multiple references to a file extent. In most cases,
486 * these references are from same file and the corresponding offsets inside
3bb1a1bc 487 * the file are close together.
d8d5f3e1
CM		 488 *
489 * When a file extent is allocated the fields are filled in:
3bb1a1bc 490 * (root_key.objectid, trans->transid, inode objectid, 1)
d8d5f3e1
CM		 491 *
492 * When a leaf is cow'd new references are added for every file extent found
31840ae1
ZY		 493 * in the leaf. It looks similar to the create case, but trans->transid will
494 * be different when the block is cow'd.
d8d5f3e1 495 *
3bb1a1bc 496 * (root_key.objectid, trans->transid, inode objectid,
31840ae1 497 * number of references in the leaf)
d8d5f3e1 498 *
3bb1a1bc
YZ		 499 * When a file extent is removed either during snapshot deletion or
500 * file truncation, we find the corresponding back reference and check
501 * the following fields:
d8d5f3e1 502 *
3bb1a1bc
YZ		 503 * (btrfs_header_owner(leaf), btrfs_header_generation(leaf),
504 * inode objectid)
d8d5f3e1
CM		 505 *
506 * Btree extents can be referenced by:
507 *
508 * - Different subvolumes
509 * - Different generations of the same subvolume
510 *
d8d5f3e1
CM		 511 * When a tree block is created, back references are inserted:
512 *
3bb1a1bc 513 * (root->root_key.objectid, trans->transid, level, 1)
d8d5f3e1 514 *
31840ae1
ZY		 515 * When a tree block is cow'd, new back references are added for all the
516 * blocks it points to. If the tree block isn't in reference counted root,
517 * the old back references are removed. These new back references are of
518 * the form (trans->transid will have increased since creation):
d8d5f3e1 519 *
3bb1a1bc 520 * (root->root_key.objectid, trans->transid, level, 1)
d8d5f3e1 521 *
31840ae1 522 * When a backref is in deleting, the following fields are checked:
d8d5f3e1
CM		 523 *
524 * if backref was for a tree root:
3bb1a1bc 525 * (btrfs_header_owner(itself), btrfs_header_generation(itself), level)
d8d5f3e1 526 * else
3bb1a1bc 527 * (btrfs_header_owner(parent), btrfs_header_generation(parent), level)
d8d5f3e1 528 *
31840ae1 529 * Back Reference Key composing:
d8d5f3e1 530 *
31840ae1
ZY		 531 * The key objectid corresponds to the first byte in the extent, the key
532 * type is set to BTRFS_EXTENT_REF_KEY, and the key offset is the first
533 * byte of parent extent. If a extent is tree root, the key offset is set
534 * to the key objectid.
d8d5f3e1 535 */
31840ae1
ZY		 536
537static int noinline lookup_extent_backref(struct btrfs_trans_handle *trans,
538 struct btrfs_root *root,
3bb1a1bc
YZ		 539 struct btrfs_path *path,
540 u64 bytenr, u64 parent,
541 u64 ref_root, u64 ref_generation,
542 u64 owner_objectid, int del)
7bb86316 543{
7bb86316 544 struct btrfs_key key;
31840ae1
ZY		 545 struct btrfs_extent_ref *ref;
546 struct extent_buffer *leaf;
3bb1a1bc 547 u64 ref_objectid;
74493f7a
CM		 548 int ret;
549
31840ae1
ZY		 550 key.objectid = bytenr;
551 key.type = BTRFS_EXTENT_REF_KEY;
552 key.offset = parent;
553
554 ret = btrfs_search_slot(trans, root, &key, path, del ? -1 : 0, 1);
555 if (ret < 0)
556 goto out;
557 if (ret > 0) {
558 ret = -ENOENT;
559 goto out;
560 }
561
562 leaf = path->nodes[0];
563 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_ref);
3bb1a1bc 564 ref_objectid = btrfs_ref_objectid(leaf, ref);
31840ae1 565 if (btrfs_ref_root(leaf, ref) != ref_root ||
3bb1a1bc
YZ		 566 btrfs_ref_generation(leaf, ref) != ref_generation ||
567 (ref_objectid != owner_objectid &&
568 ref_objectid != BTRFS_MULTIPLE_OBJECTIDS)) {
31840ae1
ZY		 569 ret = -EIO;
570 WARN_ON(1);
571 goto out;
572 }
573 ret = 0;
574out:
575 return ret;
576}
577
f3465ca4
JB		 578/*
579 * updates all the backrefs that are pending on update_list for the
580 * extent_root
581 */
582static int noinline update_backrefs(struct btrfs_trans_handle *trans,
583 struct btrfs_root *extent_root,
584 struct btrfs_path *path,
585 struct list_head *update_list)
586{
587 struct btrfs_key key;
588 struct btrfs_extent_ref *ref;
589 struct btrfs_fs_info *info = extent_root->fs_info;
590 struct pending_extent_op *op;
591 struct extent_buffer *leaf;
592 int ret = 0;
593 struct list_head *cur = update_list->next;
594 u64 ref_objectid;
595 u64 ref_root = extent_root->root_key.objectid;
596
597 op = list_entry(cur, struct pending_extent_op, list);
598
599search:
600 key.objectid = op->bytenr;
601 key.type = BTRFS_EXTENT_REF_KEY;
602 key.offset = op->orig_parent;
603
604 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 1);
605 BUG_ON(ret);
606
607 leaf = path->nodes[0];
608
609loop:
610 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_ref);
611
612 ref_objectid = btrfs_ref_objectid(leaf, ref);
613
614 if (btrfs_ref_root(leaf, ref) != ref_root ||
615 btrfs_ref_generation(leaf, ref) != op->orig_generation ||
616 (ref_objectid != op->level &&
617 ref_objectid != BTRFS_MULTIPLE_OBJECTIDS)) {
618 printk(KERN_ERR "couldn't find %Lu, parent %Lu, root %Lu, "
619 "owner %u\n", op->bytenr, op->orig_parent,
620 ref_root, op->level);
621 btrfs_print_leaf(extent_root, leaf);
622 BUG();
623 }
624
625 key.objectid = op->bytenr;
626 key.offset = op->parent;
627 key.type = BTRFS_EXTENT_REF_KEY;
628 ret = btrfs_set_item_key_safe(trans, extent_root, path, &key);
629 BUG_ON(ret);
630 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_ref);
631 btrfs_set_ref_generation(leaf, ref, op->generation);
632
633 cur = cur->next;
634
635 list_del_init(&op->list);
636 unlock_extent(&info->extent_ins, op->bytenr,
637 op->bytenr + op->num_bytes - 1, GFP_NOFS);
638 kfree(op);
639
640 if (cur == update_list) {
641 btrfs_mark_buffer_dirty(path->nodes[0]);
642 btrfs_release_path(extent_root, path);
643 goto out;
644 }
645
646 op = list_entry(cur, struct pending_extent_op, list);
647
648 path->slots[0]++;
649 while (path->slots[0] < btrfs_header_nritems(leaf)) {
650 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
651 if (key.objectid == op->bytenr &&
652 key.type == BTRFS_EXTENT_REF_KEY)
653 goto loop;
654 path->slots[0]++;
655 }
656
657 btrfs_mark_buffer_dirty(path->nodes[0]);
658 btrfs_release_path(extent_root, path);
659 goto search;
660
661out:
662 return 0;
663}
664
665static int noinline insert_extents(struct btrfs_trans_handle *trans,
666 struct btrfs_root *extent_root,
667 struct btrfs_path *path,
668 struct list_head *insert_list, int nr)
669{
670 struct btrfs_key *keys;
671 u32 *data_size;
672 struct pending_extent_op *op;
673 struct extent_buffer *leaf;
674 struct list_head *cur = insert_list->next;
675 struct btrfs_fs_info *info = extent_root->fs_info;
676 u64 ref_root = extent_root->root_key.objectid;
677 int i = 0, last = 0, ret;
678 int total = nr * 2;
679
680 if (!nr)
681 return 0;
682
683 keys = kzalloc(total * sizeof(struct btrfs_key), GFP_NOFS);
684 if (!keys)
685 return -ENOMEM;
686
687 data_size = kzalloc(total * sizeof(u32), GFP_NOFS);
688 if (!data_size) {
689 kfree(keys);
690 return -ENOMEM;
691 }
692
693 list_for_each_entry(op, insert_list, list) {
694 keys[i].objectid = op->bytenr;
695 keys[i].offset = op->num_bytes;
696 keys[i].type = BTRFS_EXTENT_ITEM_KEY;
697 data_size[i] = sizeof(struct btrfs_extent_item);
698 i++;
699
700 keys[i].objectid = op->bytenr;
701 keys[i].offset = op->parent;
702 keys[i].type = BTRFS_EXTENT_REF_KEY;
703 data_size[i] = sizeof(struct btrfs_extent_ref);
704 i++;
705 }
706
707 op = list_entry(cur, struct pending_extent_op, list);
708 i = 0;
709 while (i < total) {
710 int c;
711 ret = btrfs_insert_some_items(trans, extent_root, path,
712 keys+i, data_size+i, total-i);
713 BUG_ON(ret < 0);
714
715 if (last && ret > 1)
716 BUG();
717
718 leaf = path->nodes[0];
719 for (c = 0; c < ret; c++) {
720 int ref_first = keys[i].type == BTRFS_EXTENT_REF_KEY;
721
722 /*
723 * if the first item we inserted was a backref, then
724 * the EXTENT_ITEM will be the odd c's, else it will
725 * be the even c's
726 */
727 if ((ref_first && (c % 2)) ||
728 (!ref_first && !(c % 2))) {
729 struct btrfs_extent_item *itm;
730
731 itm = btrfs_item_ptr(leaf, path->slots[0] + c,
732 struct btrfs_extent_item);
733 btrfs_set_extent_refs(path->nodes[0], itm, 1);
734 op->del++;
735 } else {
736 struct btrfs_extent_ref *ref;
737
738 ref = btrfs_item_ptr(leaf, path->slots[0] + c,
739 struct btrfs_extent_ref);
740 btrfs_set_ref_root(leaf, ref, ref_root);
741 btrfs_set_ref_generation(leaf, ref,
742 op->generation);
743 btrfs_set_ref_objectid(leaf, ref, op->level);
744 btrfs_set_ref_num_refs(leaf, ref, 1);
745 op->del++;
746 }
747
748 /*
749 * using del to see when its ok to free up the
750 * pending_extent_op. In the case where we insert the
751 * last item on the list in order to help do batching
752 * we need to not free the extent op until we actually
753 * insert the extent_item
754 */
755 if (op->del == 2) {
756 unlock_extent(&info->extent_ins, op->bytenr,
757 op->bytenr + op->num_bytes - 1,
758 GFP_NOFS);
759 cur = cur->next;
760 list_del_init(&op->list);
761 kfree(op);
762 if (cur != insert_list)
763 op = list_entry(cur,
764 struct pending_extent_op,
765 list);
766 }
767 }
768 btrfs_mark_buffer_dirty(leaf);
769 btrfs_release_path(extent_root, path);
770
771 /*
772 * Ok backref's and items usually go right next to eachother,
773 * but if we could only insert 1 item that means that we
774 * inserted on the end of a leaf, and we have no idea what may
775 * be on the next leaf so we just play it safe. In order to
776 * try and help this case we insert the last thing on our
777 * insert list so hopefully it will end up being the last
778 * thing on the leaf and everything else will be before it,
779 * which will let us insert a whole bunch of items at the same
780 * time.
781 */
782 if (ret == 1 && !last && (i + ret < total)) {
783 /*
784 * last: where we will pick up the next time around
785 * i: our current key to insert, will be total - 1
786 * cur: the current op we are screwing with
787 * op: duh
788 */
789 last = i + ret;
790 i = total - 1;
791 cur = insert_list->prev;
792 op = list_entry(cur, struct pending_extent_op, list);
793 } else if (last) {
794 /*
795 * ok we successfully inserted the last item on the
796 * list, lets reset everything
797 *
798 * i: our current key to insert, so where we left off
799 * last time
800 * last: done with this
801 * cur: the op we are messing with
802 * op: duh
803 * total: since we inserted the last key, we need to
804 * decrement total so we dont overflow
805 */
806 i = last;
807 last = 0;
f3465ca4 808 total--;
b4eec2ca
LH		 809 if (i < total) {
810 cur = insert_list->next;
811 op = list_entry(cur, struct pending_extent_op,
812 list);
813 }
f3465ca4
JB		 814 } else {
815 i += ret;
816 }
817
818 cond_resched();
819 }
820 ret = 0;
821 kfree(keys);
822 kfree(data_size);
823 return ret;
824}
825
31840ae1
ZY		 826static int noinline insert_extent_backref(struct btrfs_trans_handle *trans,
827 struct btrfs_root *root,
828 struct btrfs_path *path,
829 u64 bytenr, u64 parent,
830 u64 ref_root, u64 ref_generation,
3bb1a1bc 831 u64 owner_objectid)
31840ae1
ZY		 832{
833 struct btrfs_key key;
834 struct extent_buffer *leaf;
835 struct btrfs_extent_ref *ref;
836 u32 num_refs;
837 int ret;
74493f7a 838
74493f7a
CM		 839 key.objectid = bytenr;
840 key.type = BTRFS_EXTENT_REF_KEY;
31840ae1 841 key.offset = parent;
74493f7a 842
31840ae1
ZY		 843 ret = btrfs_insert_empty_item(trans, root, path, &key, sizeof(*ref));
844 if (ret == 0) {
845 leaf = path->nodes[0];
846 ref = btrfs_item_ptr(leaf, path->slots[0],
847 struct btrfs_extent_ref);
848 btrfs_set_ref_root(leaf, ref, ref_root);
849 btrfs_set_ref_generation(leaf, ref, ref_generation);
850 btrfs_set_ref_objectid(leaf, ref, owner_objectid);
31840ae1
ZY		 851 btrfs_set_ref_num_refs(leaf, ref, 1);
852 } else if (ret == -EEXIST) {
853 u64 existing_owner;
854 BUG_ON(owner_objectid < BTRFS_FIRST_FREE_OBJECTID);
855 leaf = path->nodes[0];
856 ref = btrfs_item_ptr(leaf, path->slots[0],
857 struct btrfs_extent_ref);
858 if (btrfs_ref_root(leaf, ref) != ref_root ||
859 btrfs_ref_generation(leaf, ref) != ref_generation) {
860 ret = -EIO;
861 WARN_ON(1);
7bb86316 862 goto out;
31840ae1
ZY		 863 }
864
865 num_refs = btrfs_ref_num_refs(leaf, ref);
866 BUG_ON(num_refs == 0);
867 btrfs_set_ref_num_refs(leaf, ref, num_refs + 1);
868
869 existing_owner = btrfs_ref_objectid(leaf, ref);
3bb1a1bc
YZ		 870 if (existing_owner != owner_objectid &&
871 existing_owner != BTRFS_MULTIPLE_OBJECTIDS) {
31840ae1
ZY		 872 btrfs_set_ref_objectid(leaf, ref,
873 BTRFS_MULTIPLE_OBJECTIDS);
31840ae1
ZY		 874 }
875 ret = 0;
876 } else {
7bb86316 877 goto out;
31840ae1 878 }
7bb86316
CM		 879 btrfs_mark_buffer_dirty(path->nodes[0]);
880out:
881 btrfs_release_path(root, path);
882 return ret;
74493f7a
CM		 883}
884
31840ae1
ZY		 885static int noinline remove_extent_backref(struct btrfs_trans_handle *trans,
886 struct btrfs_root *root,
887 struct btrfs_path *path)
888{
889 struct extent_buffer *leaf;
890 struct btrfs_extent_ref *ref;
891 u32 num_refs;
892 int ret = 0;
893
894 leaf = path->nodes[0];
895 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_ref);
896 num_refs = btrfs_ref_num_refs(leaf, ref);
897 BUG_ON(num_refs == 0);
898 num_refs -= 1;
899 if (num_refs == 0) {
900 ret = btrfs_del_item(trans, root, path);
901 } else {
902 btrfs_set_ref_num_refs(leaf, ref, num_refs);
903 btrfs_mark_buffer_dirty(leaf);
904 }
905 btrfs_release_path(root, path);
906 return ret;
907}
908
4b4e25f2 909#ifdef BIO_RW_DISCARD
15916de8
CM		 910static void btrfs_issue_discard(struct block_device *bdev,
911 u64 start, u64 len)
912{
913#if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,28)
914 blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_KERNEL);
915#else
916 blkdev_issue_discard(bdev, start >> 9, len >> 9);
917#endif
918}
4b4e25f2 919#endif
15916de8 920
f3465ca4
JB		 921static int noinline free_extents(struct btrfs_trans_handle *trans,
922 struct btrfs_root *extent_root,
923 struct list_head *del_list)
924{
925 struct btrfs_fs_info *info = extent_root->fs_info;
926 struct btrfs_path *path;
927 struct btrfs_key key, found_key;
928 struct extent_buffer *leaf;
929 struct list_head *cur;
930 struct pending_extent_op *op;
931 struct btrfs_extent_item *ei;
932 int ret, num_to_del, extent_slot = 0, found_extent = 0;
933 u32 refs;
934 u64 bytes_freed = 0;
935
936 path = btrfs_alloc_path();
937 if (!path)
938 return -ENOMEM;
939 path->reada = 1;
940
941search:
942 /* search for the backref for the current ref we want to delete */
943 cur = del_list->next;
944 op = list_entry(cur, struct pending_extent_op, list);
945 ret = lookup_extent_backref(trans, extent_root, path, op->bytenr,
946 op->orig_parent,
947 extent_root->root_key.objectid,
948 op->orig_generation, op->level, 1);
949 if (ret) {
950 printk("Unable to find backref byte nr %Lu root %Lu gen %Lu "
951 "owner %u\n", op->bytenr,
952 extent_root->root_key.objectid, op->orig_generation,
953 op->level);
954 btrfs_print_leaf(extent_root, path->nodes[0]);
955 WARN_ON(1);
956 goto out;
957 }
958
959 extent_slot = path->slots[0];
960 num_to_del = 1;
961 found_extent = 0;
962
963 /*
964 * if we aren't the first item on the leaf we can move back one and see
965 * if our ref is right next to our extent item
966 */
967 if (likely(extent_slot)) {
968 extent_slot--;
969 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
970 extent_slot);
971 if (found_key.objectid == op->bytenr &&
972 found_key.type == BTRFS_EXTENT_ITEM_KEY &&
973 found_key.offset == op->num_bytes) {
974 num_to_del++;
975 found_extent = 1;
976 }
977 }
978
979 /*
980 * if we didn't find the extent we need to delete the backref and then
981 * search for the extent item key so we can update its ref count
982 */
983 if (!found_extent) {
984 key.objectid = op->bytenr;
985 key.type = BTRFS_EXTENT_ITEM_KEY;
986 key.offset = op->num_bytes;
987
988 ret = remove_extent_backref(trans, extent_root, path);
989 BUG_ON(ret);
990 btrfs_release_path(extent_root, path);
991 ret = btrfs_search_slot(trans, extent_root, &key, path, -1, 1);
992 BUG_ON(ret);
993 extent_slot = path->slots[0];
994 }
995
996 /* this is where we update the ref count for the extent */
997 leaf = path->nodes[0];
998 ei = btrfs_item_ptr(leaf, extent_slot, struct btrfs_extent_item);
999 refs = btrfs_extent_refs(leaf, ei);
1000 BUG_ON(refs == 0);
1001 refs--;
1002 btrfs_set_extent_refs(leaf, ei, refs);
1003
1004 btrfs_mark_buffer_dirty(leaf);
1005
1006 /*
1007 * This extent needs deleting. The reason cur_slot is extent_slot +
1008 * num_to_del is because extent_slot points to the slot where the extent
1009 * is, and if the backref was not right next to the extent we will be
1010 * deleting at least 1 item, and will want to start searching at the
1011 * slot directly next to extent_slot. However if we did find the
1012 * backref next to the extent item them we will be deleting at least 2
1013 * items and will want to start searching directly after the ref slot
1014 */
1015 if (!refs) {
1016 struct list_head *pos, *n, *end;
1017 int cur_slot = extent_slot+num_to_del;
1018 u64 super_used;
1019 u64 root_used;
1020
1021 path->slots[0] = extent_slot;
1022 bytes_freed = op->num_bytes;
1023
e3e469f8
JB		 1024 mutex_lock(&info->pinned_mutex);
1025 ret = pin_down_bytes(trans, extent_root, op->bytenr,
1026 op->num_bytes, op->level >=
1027 BTRFS_FIRST_FREE_OBJECTID);
1028 mutex_unlock(&info->pinned_mutex);
1029 BUG_ON(ret < 0);
1030 op->del = ret;
1031
f3465ca4
JB		 1032 /*
1033 * we need to see if we can delete multiple things at once, so
1034 * start looping through the list of extents we are wanting to
1035 * delete and see if their extent/backref's are right next to
1036 * eachother and the extents only have 1 ref
1037 */
1038 for (pos = cur->next; pos != del_list; pos = pos->next) {
1039 struct pending_extent_op *tmp;
1040
1041 tmp = list_entry(pos, struct pending_extent_op, list);
1042
1043 /* we only want to delete extent+ref at this stage */
1044 if (cur_slot >= btrfs_header_nritems(leaf) - 1)
1045 break;
1046
1047 btrfs_item_key_to_cpu(leaf, &found_key, cur_slot);
1048 if (found_key.objectid != tmp->bytenr ||
1049 found_key.type != BTRFS_EXTENT_ITEM_KEY ||
1050 found_key.offset != tmp->num_bytes)
1051 break;
1052
1053 /* check to make sure this extent only has one ref */
1054 ei = btrfs_item_ptr(leaf, cur_slot,
1055 struct btrfs_extent_item);
1056 if (btrfs_extent_refs(leaf, ei) != 1)
1057 break;
1058
1059 btrfs_item_key_to_cpu(leaf, &found_key, cur_slot+1);
1060 if (found_key.objectid != tmp->bytenr ||
1061 found_key.type != BTRFS_EXTENT_REF_KEY ||
1062 found_key.offset != tmp->orig_parent)
1063 break;
1064
1065 /*
1066 * the ref is right next to the extent, we can set the
1067 * ref count to 0 since we will delete them both now
1068 */
1069 btrfs_set_extent_refs(leaf, ei, 0);
1070
1071 /* pin down the bytes for this extent */
1072 mutex_lock(&info->pinned_mutex);
1073 ret = pin_down_bytes(trans, extent_root, tmp->bytenr,
1074 tmp->num_bytes, tmp->level >=
1075 BTRFS_FIRST_FREE_OBJECTID);
1076 mutex_unlock(&info->pinned_mutex);
1077 BUG_ON(ret < 0);
1078
1079 /*
1080 * use the del field to tell if we need to go ahead and
1081 * free up the extent when we delete the item or not.
1082 */
1083 tmp->del = ret;
1084 bytes_freed += tmp->num_bytes;
1085
1086 num_to_del += 2;
1087 cur_slot += 2;
1088 }
1089 end = pos;
1090
1091 /* update the free space counters */
1092 spin_lock_irq(&info->delalloc_lock);
1093 super_used = btrfs_super_bytes_used(&info->super_copy);
1094 btrfs_set_super_bytes_used(&info->super_copy,
1095 super_used - bytes_freed);
1096 spin_unlock_irq(&info->delalloc_lock);
1097
1098 root_used = btrfs_root_used(&extent_root->root_item);
1099 btrfs_set_root_used(&extent_root->root_item,
1100 root_used - bytes_freed);
1101
1102 /* delete the items */
1103 ret = btrfs_del_items(trans, extent_root, path,
1104 path->slots[0], num_to_del);
1105 BUG_ON(ret);
1106
1107 /*
1108 * loop through the extents we deleted and do the cleanup work
1109 * on them
1110 */
1111 for (pos = cur, n = pos->next; pos != end;
1112 pos = n, n = pos->next) {
1113 struct pending_extent_op *tmp;
1114#ifdef BIO_RW_DISCARD
1115 u64 map_length;
1116 struct btrfs_multi_bio *multi = NULL;
1117#endif
1118 tmp = list_entry(pos, struct pending_extent_op, list);
1119
1120 /*
1121 * remember tmp->del tells us wether or not we pinned
1122 * down the extent
1123 */
1124 ret = update_block_group(trans, extent_root,
1125 tmp->bytenr, tmp->num_bytes, 0,
1126 tmp->del);
1127 BUG_ON(ret);
1128
1129#ifdef BIO_RW_DISCARD
15916de8 1130 map_length = tmp->num_bytes;
f3465ca4
JB		 1131 ret = btrfs_map_block(&info->mapping_tree, READ,
1132 tmp->bytenr, &map_length, &multi,
1133 0);
1134 if (!ret) {
1135 struct btrfs_bio_stripe *stripe;
1136 int i;
1137
15916de8 1138 stripe = multi->stripes;
f3465ca4
JB		 1139
1140 if (map_length > tmp->num_bytes)
1141 map_length = tmp->num_bytes;
1142
1143 for (i = 0; i < multi->num_stripes;
1144 i++, stripe++)
15916de8
CM		 1145 btrfs_issue_discard(stripe->dev->bdev,
1146 stripe->physical,
1147 map_length);
f3465ca4
JB		 1148 kfree(multi);
1149 }
1150#endif
1151 list_del_init(&tmp->list);
1152 unlock_extent(&info->extent_ins, tmp->bytenr,
1153 tmp->bytenr + tmp->num_bytes - 1,
1154 GFP_NOFS);
1155 kfree(tmp);
1156 }
1157 } else if (refs && found_extent) {
1158 /*
1159 * the ref and extent were right next to eachother, but the
1160 * extent still has a ref, so just free the backref and keep
1161 * going
1162 */
1163 ret = remove_extent_backref(trans, extent_root, path);
1164 BUG_ON(ret);
1165
1166 list_del_init(&op->list);
1167 unlock_extent(&info->extent_ins, op->bytenr,
1168 op->bytenr + op->num_bytes - 1, GFP_NOFS);
1169 kfree(op);
1170 } else {
1171 /*
1172 * the extent has multiple refs and the backref we were looking
1173 * for was not right next to it, so just unlock and go next,
1174 * we're good to go
1175 */
1176 list_del_init(&op->list);
1177 unlock_extent(&info->extent_ins, op->bytenr,
1178 op->bytenr + op->num_bytes - 1, GFP_NOFS);
1179 kfree(op);
1180 }
1181
1182 btrfs_release_path(extent_root, path);
1183 if (!list_empty(del_list))
1184 goto search;
1185
1186out:
1187 btrfs_free_path(path);
1188 return ret;
1189}
1190
31840ae1
ZY		 1191static int __btrfs_update_extent_ref(struct btrfs_trans_handle *trans,
1192 struct btrfs_root *root, u64 bytenr,
1193 u64 orig_parent, u64 parent,
1194 u64 orig_root, u64 ref_root,
1195 u64 orig_generation, u64 ref_generation,
3bb1a1bc 1196 u64 owner_objectid)
31840ae1
ZY		 1197{
1198 int ret;
1199 struct btrfs_root *extent_root = root->fs_info->extent_root;
1200 struct btrfs_path *path;
1201
1202 if (root == root->fs_info->extent_root) {
1203 struct pending_extent_op *extent_op;
1204 u64 num_bytes;
1205
1206 BUG_ON(owner_objectid >= BTRFS_MAX_LEVEL);
1207 num_bytes = btrfs_level_size(root, (int)owner_objectid);
25179201 1208 mutex_lock(&root->fs_info->extent_ins_mutex);
31840ae1 1209 if (test_range_bit(&root->fs_info->extent_ins, bytenr,
25179201 1210 bytenr + num_bytes - 1, EXTENT_WRITEBACK, 0)) {
31840ae1
ZY		 1211 u64 priv;
1212 ret = get_state_private(&root->fs_info->extent_ins,
1213 bytenr, &priv);
1214 BUG_ON(ret);
1215 extent_op = (struct pending_extent_op *)
1216 (unsigned long)priv;
1217 BUG_ON(extent_op->parent != orig_parent);
1218 BUG_ON(extent_op->generation != orig_generation);
25179201 1219
31840ae1
ZY		 1220 extent_op->parent = parent;
1221 extent_op->generation = ref_generation;
1222 } else {
1223 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
1224 BUG_ON(!extent_op);
1225
1226 extent_op->type = PENDING_BACKREF_UPDATE;
1227 extent_op->bytenr = bytenr;
1228 extent_op->num_bytes = num_bytes;
1229 extent_op->parent = parent;
1230 extent_op->orig_parent = orig_parent;
1231 extent_op->generation = ref_generation;
1232 extent_op->orig_generation = orig_generation;
1233 extent_op->level = (int)owner_objectid;
f3465ca4
JB		 1234 INIT_LIST_HEAD(&extent_op->list);
1235 extent_op->del = 0;
31840ae1
ZY		 1236
1237 set_extent_bits(&root->fs_info->extent_ins,
1238 bytenr, bytenr + num_bytes - 1,
25179201 1239 EXTENT_WRITEBACK, GFP_NOFS);
31840ae1
ZY		 1240 set_state_private(&root->fs_info->extent_ins,
1241 bytenr, (unsigned long)extent_op);
1242 }
25179201 1243 mutex_unlock(&root->fs_info->extent_ins_mutex);
31840ae1
ZY		 1244 return 0;
1245 }
1246
1247 path = btrfs_alloc_path();
1248 if (!path)
1249 return -ENOMEM;
1250 ret = lookup_extent_backref(trans, extent_root, path,
1251 bytenr, orig_parent, orig_root,
3bb1a1bc 1252 orig_generation, owner_objectid, 1);
31840ae1
ZY		 1253 if (ret)
1254 goto out;
1255 ret = remove_extent_backref(trans, extent_root, path);
1256 if (ret)
1257 goto out;
1258 ret = insert_extent_backref(trans, extent_root, path, bytenr,
1259 parent, ref_root, ref_generation,
3bb1a1bc 1260 owner_objectid);
31840ae1 1261 BUG_ON(ret);
87ef2bb4
CM		 1262 finish_current_insert(trans, extent_root, 0);
1263 del_pending_extents(trans, extent_root, 0);
31840ae1
ZY		 1264out:
1265 btrfs_free_path(path);
1266 return ret;
1267}
1268
1269int btrfs_update_extent_ref(struct btrfs_trans_handle *trans,
1270 struct btrfs_root *root, u64 bytenr,
1271 u64 orig_parent, u64 parent,
1272 u64 ref_root, u64 ref_generation,
3bb1a1bc 1273 u64 owner_objectid)
31840ae1
ZY		 1274{
1275 int ret;
1276 if (ref_root == BTRFS_TREE_LOG_OBJECTID &&
1277 owner_objectid < BTRFS_FIRST_FREE_OBJECTID)
1278 return 0;
31840ae1
ZY		 1279 ret = __btrfs_update_extent_ref(trans, root, bytenr, orig_parent,
1280 parent, ref_root, ref_root,
1281 ref_generation, ref_generation,
3bb1a1bc 1282 owner_objectid);
31840ae1
ZY		 1283 return ret;
1284}
1285
925baedd 1286static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
31840ae1
ZY		 1287 struct btrfs_root *root, u64 bytenr,
1288 u64 orig_parent, u64 parent,
1289 u64 orig_root, u64 ref_root,
1290 u64 orig_generation, u64 ref_generation,
3bb1a1bc 1291 u64 owner_objectid)
02217ed2 1292{
5caf2a00 1293 struct btrfs_path *path;
02217ed2 1294 int ret;
e2fa7227 1295 struct btrfs_key key;
5f39d397 1296 struct extent_buffer *l;
234b63a0 1297 struct btrfs_extent_item *item;
cf27e1ee 1298 u32 refs;
037e6390 1299
5caf2a00 1300 path = btrfs_alloc_path();
54aa1f4d
CM		 1301 if (!path)
1302 return -ENOMEM;
26b8003f 1303
3c12ac72 1304 path->reada = 1;
db94535d 1305 key.objectid = bytenr;
31840ae1
ZY		 1306 key.type = BTRFS_EXTENT_ITEM_KEY;
1307 key.offset = (u64)-1;
1308
5caf2a00 1309 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path,
9f5fae2f 1310 0, 1);
54aa1f4d
CM		 1311 if (ret < 0)
1312 return ret;
31840ae1
ZY		 1313 BUG_ON(ret == 0 || path->slots[0] == 0);
1314
1315 path->slots[0]--;
5f39d397 1316 l = path->nodes[0];
31840ae1
ZY		 1317
1318 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
771ed689
CM		 1319 if (key.objectid != bytenr) {
1320 btrfs_print_leaf(root->fs_info->extent_root, path->nodes[0]);
1321 printk("wanted %Lu found %Lu\n", bytenr, key.objectid);
1322 BUG();
1323 }
31840ae1
ZY		 1324 BUG_ON(key.type != BTRFS_EXTENT_ITEM_KEY);
1325
5caf2a00 1326 item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
5f39d397
CM		 1327 refs = btrfs_extent_refs(l, item);
1328 btrfs_set_extent_refs(l, item, refs + 1);
5caf2a00 1329 btrfs_mark_buffer_dirty(path->nodes[0]);
a28ec197 1330
5caf2a00 1331 btrfs_release_path(root->fs_info->extent_root, path);
7bb86316 1332
3c12ac72 1333 path->reada = 1;
31840ae1
ZY		 1334 ret = insert_extent_backref(trans, root->fs_info->extent_root,
1335 path, bytenr, parent,
1336 ref_root, ref_generation,
3bb1a1bc 1337 owner_objectid);
7bb86316 1338 BUG_ON(ret);
87ef2bb4
CM		 1339 finish_current_insert(trans, root->fs_info->extent_root, 0);
1340 del_pending_extents(trans, root->fs_info->extent_root, 0);
74493f7a
CM		 1341
1342 btrfs_free_path(path);
02217ed2
CM		 1343 return 0;
1344}
1345
925baedd 1346int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
31840ae1
ZY		 1347 struct btrfs_root *root,
1348 u64 bytenr, u64 num_bytes, u64 parent,
1349 u64 ref_root, u64 ref_generation,
3bb1a1bc 1350 u64 owner_objectid)
925baedd
CM		 1351{
1352 int ret;
31840ae1
ZY		 1353 if (ref_root == BTRFS_TREE_LOG_OBJECTID &&
1354 owner_objectid < BTRFS_FIRST_FREE_OBJECTID)
1355 return 0;
31840ae1
ZY		 1356 ret = __btrfs_inc_extent_ref(trans, root, bytenr, 0, parent,
1357 0, ref_root, 0, ref_generation,
3bb1a1bc 1358 owner_objectid);
925baedd
CM		 1359 return ret;
1360}
1361
e9d0b13b
CM		 1362int btrfs_extent_post_op(struct btrfs_trans_handle *trans,
1363 struct btrfs_root *root)
1364{
87ef2bb4
CM		 1365 finish_current_insert(trans, root->fs_info->extent_root, 1);
1366 del_pending_extents(trans, root->fs_info->extent_root, 1);
e9d0b13b
CM		 1367 return 0;
1368}
1369
31840ae1
ZY		 1370int btrfs_lookup_extent_ref(struct btrfs_trans_handle *trans,
1371 struct btrfs_root *root, u64 bytenr,
1372 u64 num_bytes, u32 *refs)
a28ec197 1373{
5caf2a00 1374 struct btrfs_path *path;
a28ec197 1375 int ret;
e2fa7227 1376 struct btrfs_key key;
5f39d397 1377 struct extent_buffer *l;
234b63a0 1378 struct btrfs_extent_item *item;
5caf2a00 1379
db94535d 1380 WARN_ON(num_bytes < root->sectorsize);
5caf2a00 1381 path = btrfs_alloc_path();
3c12ac72 1382 path->reada = 1;
db94535d
CM		 1383 key.objectid = bytenr;
1384 key.offset = num_bytes;
62e2749e 1385 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
5caf2a00 1386 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path,
9f5fae2f 1387 0, 0);
54aa1f4d
CM		 1388 if (ret < 0)
1389 goto out;
5f39d397
CM		 1390 if (ret != 0) {
1391 btrfs_print_leaf(root, path->nodes[0]);
db94535d 1392 printk("failed to find block number %Lu\n", bytenr);
a28ec197 1393 BUG();
5f39d397
CM		 1394 }
1395 l = path->nodes[0];
5caf2a00 1396 item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
5f39d397 1397 *refs = btrfs_extent_refs(l, item);
54aa1f4d 1398out:
5caf2a00 1399 btrfs_free_path(path);
a28ec197
CM		 1400 return 0;
1401}
1402
80ff3856
YZ		 1403int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
1404 struct btrfs_root *root, u64 bytenr)
be20aa9d
CM		 1405{
1406 struct btrfs_root *extent_root = root->fs_info->extent_root;
1407 struct btrfs_path *path;
f321e491
YZ		 1408 struct extent_buffer *leaf;
1409 struct btrfs_extent_ref *ref_item;
1410 struct btrfs_key key;
1411 struct btrfs_key found_key;
80ff3856
YZ		 1412 u64 ref_root;
1413 u64 last_snapshot;
be20aa9d
CM		 1414 u32 nritems;
1415 int ret;
925baedd 1416
be20aa9d 1417 key.objectid = bytenr;
31840ae1 1418 key.offset = (u64)-1;
f321e491 1419 key.type = BTRFS_EXTENT_ITEM_KEY;
be20aa9d 1420
f321e491 1421 path = btrfs_alloc_path();
be20aa9d
CM		 1422 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
1423 if (ret < 0)
1424 goto out;
1425 BUG_ON(ret == 0);
80ff3856
YZ		 1426
1427 ret = -ENOENT;
1428 if (path->slots[0] == 0)
31840ae1 1429 goto out;
be20aa9d 1430
31840ae1 1431 path->slots[0]--;
f321e491
YZ		 1432 leaf = path->nodes[0];
1433 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
be20aa9d
CM		 1434
1435 if (found_key.objectid != bytenr ||
80ff3856 1436 found_key.type != BTRFS_EXTENT_ITEM_KEY)
be20aa9d 1437 goto out;
f321e491 1438
80ff3856 1439 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
be20aa9d 1440 while (1) {
f321e491
YZ		 1441 leaf = path->nodes[0];
1442 nritems = btrfs_header_nritems(leaf);
be20aa9d
CM		 1443 if (path->slots[0] >= nritems) {
1444 ret = btrfs_next_leaf(extent_root, path);
f321e491
YZ		 1445 if (ret < 0)
1446 goto out;
be20aa9d
CM		 1447 if (ret == 0)
1448 continue;
1449 break;
1450 }
f321e491 1451 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
be20aa9d
CM		 1452 if (found_key.objectid != bytenr)
1453 break;
bd09835d 1454
be20aa9d
CM		 1455 if (found_key.type != BTRFS_EXTENT_REF_KEY) {
1456 path->slots[0]++;
1457 continue;
1458 }
1459
f321e491 1460 ref_item = btrfs_item_ptr(leaf, path->slots[0],
be20aa9d 1461 struct btrfs_extent_ref);
80ff3856
YZ		 1462 ref_root = btrfs_ref_root(leaf, ref_item);
1463 if (ref_root != root->root_key.objectid &&
1464 ref_root != BTRFS_TREE_LOG_OBJECTID) {
1465 ret = 1;
f321e491 1466 goto out;
80ff3856
YZ		 1467 }
1468 if (btrfs_ref_generation(leaf, ref_item) <= last_snapshot) {
f321e491
YZ		 1469 ret = 1;
1470 goto out;
1471 }
80ff3856
YZ		 1472
1473 path->slots[0]++;
f321e491
YZ		 1474 }
1475 ret = 0;
be20aa9d 1476out:
80ff3856 1477 btrfs_free_path(path);
f321e491 1478 return ret;
be20aa9d 1479}
c5739bba 1480
31840ae1
ZY		 1481int btrfs_cache_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1482 struct extent_buffer *buf, u32 nr_extents)
02217ed2 1483{
5f39d397 1484 struct btrfs_key key;
6407bf6d 1485 struct btrfs_file_extent_item *fi;
e4657689
ZY		 1486 u64 root_gen;
1487 u32 nritems;
02217ed2 1488 int i;
db94535d 1489 int level;
31840ae1 1490 int ret = 0;
e4657689 1491 int shared = 0;
a28ec197 1492
3768f368 1493 if (!root->ref_cows)
a28ec197 1494 return 0;
5f39d397 1495
e4657689
ZY		 1496 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
1497 shared = 0;
1498 root_gen = root->root_key.offset;
1499 } else {
1500 shared = 1;
1501 root_gen = trans->transid - 1;
1502 }
1503
db94535d 1504 level = btrfs_header_level(buf);
5f39d397 1505 nritems = btrfs_header_nritems(buf);
4a096752 1506
31840ae1 1507 if (level == 0) {
31153d81
YZ		 1508 struct btrfs_leaf_ref *ref;
1509 struct btrfs_extent_info *info;
1510
31840ae1 1511 ref = btrfs_alloc_leaf_ref(root, nr_extents);
31153d81 1512 if (!ref) {
31840ae1 1513 ret = -ENOMEM;
31153d81
YZ		 1514 goto out;
1515 }
1516
e4657689 1517 ref->root_gen = root_gen;
31153d81
YZ		 1518 ref->bytenr = buf->start;
1519 ref->owner = btrfs_header_owner(buf);
1520 ref->generation = btrfs_header_generation(buf);
31840ae1 1521 ref->nritems = nr_extents;
31153d81 1522 info = ref->extents;
bcc63abb 1523
31840ae1 1524 for (i = 0; nr_extents > 0 && i < nritems; i++) {
31153d81
YZ		 1525 u64 disk_bytenr;
1526 btrfs_item_key_to_cpu(buf, &key, i);
1527 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
1528 continue;
1529 fi = btrfs_item_ptr(buf, i,
1530 struct btrfs_file_extent_item);
1531 if (btrfs_file_extent_type(buf, fi) ==
1532 BTRFS_FILE_EXTENT_INLINE)
1533 continue;
1534 disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
1535 if (disk_bytenr == 0)
1536 continue;
1537
1538 info->bytenr = disk_bytenr;
1539 info->num_bytes =
1540 btrfs_file_extent_disk_num_bytes(buf, fi);
1541 info->objectid = key.objectid;
1542 info->offset = key.offset;
1543 info++;
1544 }
1545
e4657689 1546 ret = btrfs_add_leaf_ref(root, ref, shared);
5b84e8d6
YZ		 1547 if (ret == -EEXIST && shared) {
1548 struct btrfs_leaf_ref *old;
1549 old = btrfs_lookup_leaf_ref(root, ref->bytenr);
1550 BUG_ON(!old);
1551 btrfs_remove_leaf_ref(root, old);
1552 btrfs_free_leaf_ref(root, old);
1553 ret = btrfs_add_leaf_ref(root, ref, shared);
1554 }
31153d81 1555 WARN_ON(ret);
bcc63abb 1556 btrfs_free_leaf_ref(root, ref);
31153d81
YZ		 1557 }
1558out:
31840ae1
ZY		 1559 return ret;
1560}
1561
1562int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1563 struct extent_buffer *orig_buf, struct extent_buffer *buf,
1564 u32 *nr_extents)
1565{
1566 u64 bytenr;
1567 u64 ref_root;
1568 u64 orig_root;
1569 u64 ref_generation;
1570 u64 orig_generation;
1571 u32 nritems;
1572 u32 nr_file_extents = 0;
1573 struct btrfs_key key;
1574 struct btrfs_file_extent_item *fi;
1575 int i;
1576 int level;
1577 int ret = 0;
1578 int faili = 0;
1579 int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
3bb1a1bc 1580 u64, u64, u64, u64, u64, u64, u64, u64);
31840ae1
ZY		 1581
1582 ref_root = btrfs_header_owner(buf);
1583 ref_generation = btrfs_header_generation(buf);
1584 orig_root = btrfs_header_owner(orig_buf);
1585 orig_generation = btrfs_header_generation(orig_buf);
1586
1587 nritems = btrfs_header_nritems(buf);
1588 level = btrfs_header_level(buf);
1589
1590 if (root->ref_cows) {
1591 process_func = __btrfs_inc_extent_ref;
1592 } else {
1593 if (level == 0 &&
1594 root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
1595 goto out;
1596 if (level != 0 &&
1597 root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID)
1598 goto out;
1599 process_func = __btrfs_update_extent_ref;
1600 }
1601
1602 for (i = 0; i < nritems; i++) {
1603 cond_resched();
db94535d 1604 if (level == 0) {
5f39d397
CM		 1605 btrfs_item_key_to_cpu(buf, &key, i);
1606 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
54aa1f4d 1607 continue;
5f39d397 1608 fi = btrfs_item_ptr(buf, i,
54aa1f4d 1609 struct btrfs_file_extent_item);
5f39d397 1610 if (btrfs_file_extent_type(buf, fi) ==
54aa1f4d
CM		 1611 BTRFS_FILE_EXTENT_INLINE)
1612 continue;
31840ae1
ZY		 1613 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
1614 if (bytenr == 0)
54aa1f4d 1615 continue;
31840ae1
ZY		 1616
1617 nr_file_extents++;
1618
31840ae1
ZY		 1619 ret = process_func(trans, root, bytenr,
1620 orig_buf->start, buf->start,
1621 orig_root, ref_root,
1622 orig_generation, ref_generation,
3bb1a1bc 1623 key.objectid);
31840ae1
ZY		 1624
1625 if (ret) {
1626 faili = i;
1627 WARN_ON(1);
1628 goto fail;
1629 }
54aa1f4d 1630 } else {
db94535d 1631 bytenr = btrfs_node_blockptr(buf, i);
31840ae1
ZY		 1632 ret = process_func(trans, root, bytenr,
1633 orig_buf->start, buf->start,
1634 orig_root, ref_root,
1635 orig_generation, ref_generation,
3bb1a1bc 1636 level - 1);
31840ae1
ZY		 1637 if (ret) {
1638 faili = i;
1639 WARN_ON(1);
1640 goto fail;
1641 }
54aa1f4d
CM		 1642 }
1643 }
31840ae1
ZY		 1644out:
1645 if (nr_extents) {
1646 if (level == 0)
1647 *nr_extents = nr_file_extents;
1648 else
1649 *nr_extents = nritems;
1650 }
1651 return 0;
1652fail:
1653 WARN_ON(1);
54aa1f4d 1654 return ret;
02217ed2
CM		 1655}
1656
31840ae1
ZY		 1657int btrfs_update_ref(struct btrfs_trans_handle *trans,
1658 struct btrfs_root *root, struct extent_buffer *orig_buf,
1659 struct extent_buffer *buf, int start_slot, int nr)
1660
1661{
1662 u64 bytenr;
1663 u64 ref_root;
1664 u64 orig_root;
1665 u64 ref_generation;
1666 u64 orig_generation;
1667 struct btrfs_key key;
1668 struct btrfs_file_extent_item *fi;
1669 int i;
1670 int ret;
1671 int slot;
1672 int level;
1673
1674 BUG_ON(start_slot < 0);
1675 BUG_ON(start_slot + nr > btrfs_header_nritems(buf));
1676
1677 ref_root = btrfs_header_owner(buf);
1678 ref_generation = btrfs_header_generation(buf);
1679 orig_root = btrfs_header_owner(orig_buf);
1680 orig_generation = btrfs_header_generation(orig_buf);
1681 level = btrfs_header_level(buf);
1682
1683 if (!root->ref_cows) {
1684 if (level == 0 &&
1685 root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
1686 return 0;
1687 if (level != 0 &&
1688 root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID)
1689 return 0;
1690 }
1691
1692 for (i = 0, slot = start_slot; i < nr; i++, slot++) {
1693 cond_resched();
1694 if (level == 0) {
1695 btrfs_item_key_to_cpu(buf, &key, slot);
1696 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
1697 continue;
1698 fi = btrfs_item_ptr(buf, slot,
1699 struct btrfs_file_extent_item);
1700 if (btrfs_file_extent_type(buf, fi) ==
1701 BTRFS_FILE_EXTENT_INLINE)
1702 continue;
1703 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
1704 if (bytenr == 0)
1705 continue;
31840ae1
ZY		 1706 ret = __btrfs_update_extent_ref(trans, root, bytenr,
1707 orig_buf->start, buf->start,
1708 orig_root, ref_root,
1709 orig_generation, ref_generation,
3bb1a1bc 1710 key.objectid);
31840ae1
ZY		 1711 if (ret)
1712 goto fail;
1713 } else {
1714 bytenr = btrfs_node_blockptr(buf, slot);
31840ae1
ZY		 1715 ret = __btrfs_update_extent_ref(trans, root, bytenr,
1716 orig_buf->start, buf->start,
1717 orig_root, ref_root,
1718 orig_generation, ref_generation,
3bb1a1bc 1719 level - 1);
31840ae1
ZY		 1720 if (ret)
1721 goto fail;
1722 }
1723 }
1724 return 0;
1725fail:
1726 WARN_ON(1);
1727 return -1;
1728}
1729
9078a3e1
CM		 1730static int write_one_cache_group(struct btrfs_trans_handle *trans,
1731 struct btrfs_root *root,
1732 struct btrfs_path *path,
1733 struct btrfs_block_group_cache *cache)
1734{
1735 int ret;
1736 int pending_ret;
1737 struct btrfs_root *extent_root = root->fs_info->extent_root;
5f39d397
CM		 1738 unsigned long bi;
1739 struct extent_buffer *leaf;
9078a3e1 1740
9078a3e1 1741 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
54aa1f4d
CM		 1742 if (ret < 0)
1743 goto fail;
9078a3e1 1744 BUG_ON(ret);
5f39d397
CM		 1745
1746 leaf = path->nodes[0];
1747 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
1748 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
1749 btrfs_mark_buffer_dirty(leaf);
9078a3e1 1750 btrfs_release_path(extent_root, path);
54aa1f4d 1751fail:
87ef2bb4
CM		 1752 finish_current_insert(trans, extent_root, 0);
1753 pending_ret = del_pending_extents(trans, extent_root, 0);
9078a3e1
CM		 1754 if (ret)
1755 return ret;
1756 if (pending_ret)
1757 return pending_ret;
1758 return 0;
1759
1760}
1761
96b5179d
CM		 1762int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
1763 struct btrfs_root *root)
9078a3e1 1764{
0f9dd46c
JB		 1765 struct btrfs_block_group_cache *cache, *entry;
1766 struct rb_node *n;
9078a3e1
CM		 1767 int err = 0;
1768 int werr = 0;
9078a3e1 1769 struct btrfs_path *path;
96b5179d 1770 u64 last = 0;
9078a3e1
CM		 1771
1772 path = btrfs_alloc_path();
1773 if (!path)
1774 return -ENOMEM;
1775
1776 while(1) {
0f9dd46c
JB		 1777 cache = NULL;
1778 spin_lock(&root->fs_info->block_group_cache_lock);
1779 for (n = rb_first(&root->fs_info->block_group_cache_tree);
1780 n; n = rb_next(n)) {
1781 entry = rb_entry(n, struct btrfs_block_group_cache,
1782 cache_node);
1783 if (entry->dirty) {
1784 cache = entry;
1785 break;
1786 }
1787 }
1788 spin_unlock(&root->fs_info->block_group_cache_lock);
54aa1f4d 1789
0f9dd46c 1790 if (!cache)
96b5179d 1791 break;
0f9dd46c 1792
e8569813 1793 cache->dirty = 0;
0f9dd46c
JB		 1794 last += cache->key.offset;
1795
96b5179d
CM		 1796 err = write_one_cache_group(trans, root,
1797 path, cache);
1798 /*
1799 * if we fail to write the cache group, we want
1800 * to keep it marked dirty in hopes that a later
1801 * write will work
1802 */
1803 if (err) {
1804 werr = err;
1805 continue;
9078a3e1
CM		 1806 }
1807 }
1808 btrfs_free_path(path);
1809 return werr;
1810}
1811
593060d7
CM		 1812static int update_space_info(struct btrfs_fs_info *info, u64 flags,
1813 u64 total_bytes, u64 bytes_used,
1814 struct btrfs_space_info **space_info)
1815{
1816 struct btrfs_space_info *found;
1817
1818 found = __find_space_info(info, flags);
1819 if (found) {
25179201 1820 spin_lock(&found->lock);
593060d7
CM		 1821 found->total_bytes += total_bytes;
1822 found->bytes_used += bytes_used;
8f18cf13 1823 found->full = 0;
25179201 1824 spin_unlock(&found->lock);
593060d7
CM		 1825 *space_info = found;
1826 return 0;
1827 }
c146afad 1828 found = kzalloc(sizeof(*found), GFP_NOFS);
593060d7
CM		 1829 if (!found)
1830 return -ENOMEM;
1831
1832 list_add(&found->list, &info->space_info);
0f9dd46c 1833 INIT_LIST_HEAD(&found->block_groups);
80eb234a 1834 init_rwsem(&found->groups_sem);
0f9dd46c 1835 spin_lock_init(&found->lock);
593060d7
CM		 1836 found->flags = flags;
1837 found->total_bytes = total_bytes;
1838 found->bytes_used = bytes_used;
1839 found->bytes_pinned = 0;
e8569813 1840 found->bytes_reserved = 0;
c146afad 1841 found->bytes_readonly = 0;
593060d7 1842 found->full = 0;
0ef3e66b 1843 found->force_alloc = 0;
593060d7
CM		 1844 *space_info = found;
1845 return 0;
1846}
1847
8790d502
CM		 1848static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
1849{
1850 u64 extra_flags = flags & (BTRFS_BLOCK_GROUP_RAID0 |
611f0e00 1851 BTRFS_BLOCK_GROUP_RAID1 |
321aecc6 1852 BTRFS_BLOCK_GROUP_RAID10 |
611f0e00 1853 BTRFS_BLOCK_GROUP_DUP);
8790d502
CM		 1854 if (extra_flags) {
1855 if (flags & BTRFS_BLOCK_GROUP_DATA)
1856 fs_info->avail_data_alloc_bits |= extra_flags;
1857 if (flags & BTRFS_BLOCK_GROUP_METADATA)
1858 fs_info->avail_metadata_alloc_bits |= extra_flags;
1859 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
1860 fs_info->avail_system_alloc_bits |= extra_flags;
1861 }
1862}
593060d7 1863
c146afad
YZ		 1864static void set_block_group_readonly(struct btrfs_block_group_cache *cache)
1865{
1866 spin_lock(&cache->space_info->lock);
1867 spin_lock(&cache->lock);
1868 if (!cache->ro) {
1869 cache->space_info->bytes_readonly += cache->key.offset -
1870 btrfs_block_group_used(&cache->item);
1871 cache->ro = 1;
1872 }
1873 spin_unlock(&cache->lock);
1874 spin_unlock(&cache->space_info->lock);
1875}
1876
2b82032c 1877u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
ec44a35c 1878{
2b82032c 1879 u64 num_devices = root->fs_info->fs_devices->rw_devices;
a061fc8d
CM		 1880
1881 if (num_devices == 1)
1882 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0);
1883 if (num_devices < 4)
1884 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
1885
ec44a35c
CM		 1886 if ((flags & BTRFS_BLOCK_GROUP_DUP) &&
1887 (flags & (BTRFS_BLOCK_GROUP_RAID1 |
a061fc8d 1888 BTRFS_BLOCK_GROUP_RAID10))) {
ec44a35c 1889 flags &= ~BTRFS_BLOCK_GROUP_DUP;
a061fc8d 1890 }
ec44a35c
CM		 1891
1892 if ((flags & BTRFS_BLOCK_GROUP_RAID1) &&
a061fc8d 1893 (flags & BTRFS_BLOCK_GROUP_RAID10)) {
ec44a35c 1894 flags &= ~BTRFS_BLOCK_GROUP_RAID1;
a061fc8d 1895 }
ec44a35c
CM		 1896
1897 if ((flags & BTRFS_BLOCK_GROUP_RAID0) &&
1898 ((flags & BTRFS_BLOCK_GROUP_RAID1) |
1899 (flags & BTRFS_BLOCK_GROUP_RAID10) |
1900 (flags & BTRFS_BLOCK_GROUP_DUP)))
1901 flags &= ~BTRFS_BLOCK_GROUP_RAID0;
1902 return flags;
1903}
1904
6324fbf3
CM		 1905static int do_chunk_alloc(struct btrfs_trans_handle *trans,
1906 struct btrfs_root *extent_root, u64 alloc_bytes,
0ef3e66b 1907 u64 flags, int force)
6324fbf3
CM		 1908{
1909 struct btrfs_space_info *space_info;
1910 u64 thresh;
c146afad
YZ		 1911 int ret = 0;
1912
1913 mutex_lock(&extent_root->fs_info->chunk_mutex);
6324fbf3 1914
2b82032c 1915 flags = btrfs_reduce_alloc_profile(extent_root, flags);
ec44a35c 1916
6324fbf3 1917 space_info = __find_space_info(extent_root->fs_info, flags);
593060d7
CM		 1918 if (!space_info) {
1919 ret = update_space_info(extent_root->fs_info, flags,
1920 0, 0, &space_info);
1921 BUG_ON(ret);
1922 }
6324fbf3
CM		 1923 BUG_ON(!space_info);
1924
25179201 1925 spin_lock(&space_info->lock);
0ef3e66b
CM		 1926 if (space_info->force_alloc) {
1927 force = 1;
1928 space_info->force_alloc = 0;
1929 }
25179201
JB		 1930 if (space_info->full) {
1931 spin_unlock(&space_info->lock);
925baedd 1932 goto out;
25179201 1933 }
6324fbf3 1934
c146afad
YZ		 1935 thresh = space_info->total_bytes - space_info->bytes_readonly;
1936 thresh = div_factor(thresh, 6);
0ef3e66b 1937 if (!force &&
e8569813 1938 (space_info->bytes_used + space_info->bytes_pinned +
25179201
JB		 1939 space_info->bytes_reserved + alloc_bytes) < thresh) {
1940 spin_unlock(&space_info->lock);
925baedd 1941 goto out;
25179201 1942 }
25179201
JB		 1943 spin_unlock(&space_info->lock);
1944
2b82032c 1945 ret = btrfs_alloc_chunk(trans, extent_root, flags);
25179201 1946 if (ret) {
6324fbf3
CM		 1947printk("space info full %Lu\n", flags);
1948 space_info->full = 1;
6324fbf3 1949 }
a74a4b97 1950out:
c146afad 1951 mutex_unlock(&extent_root->fs_info->chunk_mutex);
0f9dd46c 1952 return ret;
6324fbf3
CM		 1953}
1954
9078a3e1
CM		 1955static int update_block_group(struct btrfs_trans_handle *trans,
1956 struct btrfs_root *root,
db94535d 1957 u64 bytenr, u64 num_bytes, int alloc,
0b86a832 1958 int mark_free)
9078a3e1
CM		 1959{
1960 struct btrfs_block_group_cache *cache;
1961 struct btrfs_fs_info *info = root->fs_info;
db94535d 1962 u64 total = num_bytes;
9078a3e1 1963 u64 old_val;
db94535d 1964 u64 byte_in_group;
3e1ad54f 1965
9078a3e1 1966 while(total) {
db94535d 1967 cache = btrfs_lookup_block_group(info, bytenr);
f3465ca4 1968 if (!cache)
9078a3e1 1969 return -1;
db94535d
CM		 1970 byte_in_group = bytenr - cache->key.objectid;
1971 WARN_ON(byte_in_group > cache->key.offset);
9078a3e1 1972
25179201 1973 spin_lock(&cache->space_info->lock);
c286ac48 1974 spin_lock(&cache->lock);
0f9dd46c 1975 cache->dirty = 1;
9078a3e1 1976 old_val = btrfs_block_group_used(&cache->item);
db94535d 1977 num_bytes = min(total, cache->key.offset - byte_in_group);
cd1bc465 1978 if (alloc) {
db94535d 1979 old_val += num_bytes;
6324fbf3 1980 cache->space_info->bytes_used += num_bytes;
a512bbf8 1981 if (cache->ro)
c146afad 1982 cache->space_info->bytes_readonly -= num_bytes;
c286ac48
CM		 1983 btrfs_set_block_group_used(&cache->item, old_val);
1984 spin_unlock(&cache->lock);
25179201 1985 spin_unlock(&cache->space_info->lock);
cd1bc465 1986 } else {
db94535d 1987 old_val -= num_bytes;
6324fbf3 1988 cache->space_info->bytes_used -= num_bytes;
c146afad
YZ		 1989 if (cache->ro)
1990 cache->space_info->bytes_readonly += num_bytes;
c286ac48
CM		 1991 btrfs_set_block_group_used(&cache->item, old_val);
1992 spin_unlock(&cache->lock);
25179201 1993 spin_unlock(&cache->space_info->lock);
f510cfec 1994 if (mark_free) {
0f9dd46c
JB		 1995 int ret;
1996 ret = btrfs_add_free_space(cache, bytenr,
1997 num_bytes);
1998 if (ret)
1999 return -1;
e37c9e69 2000 }
cd1bc465 2001 }
db94535d
CM		 2002 total -= num_bytes;
2003 bytenr += num_bytes;
9078a3e1
CM		 2004 }
2005 return 0;
2006}
6324fbf3 2007
a061fc8d
CM		 2008static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
2009{
0f9dd46c
JB		 2010 struct btrfs_block_group_cache *cache;
2011
2012 cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
2013 if (!cache)
a061fc8d 2014 return 0;
0f9dd46c
JB		 2015
2016 return cache->key.objectid;
a061fc8d
CM		 2017}
2018
e02119d5 2019int btrfs_update_pinned_extents(struct btrfs_root *root,
324ae4df
Y		 2020 u64 bytenr, u64 num, int pin)
2021{
2022 u64 len;
2023 struct btrfs_block_group_cache *cache;
2024 struct btrfs_fs_info *fs_info = root->fs_info;
2025
25179201 2026 WARN_ON(!mutex_is_locked(&root->fs_info->pinned_mutex));
324ae4df
Y		 2027 if (pin) {
2028 set_extent_dirty(&fs_info->pinned_extents,
2029 bytenr, bytenr + num - 1, GFP_NOFS);
2030 } else {
2031 clear_extent_dirty(&fs_info->pinned_extents,
2032 bytenr, bytenr + num - 1, GFP_NOFS);
2033 }
2034 while (num > 0) {
2035 cache = btrfs_lookup_block_group(fs_info, bytenr);
e8569813
ZY		 2036 BUG_ON(!cache);
2037 len = min(num, cache->key.offset -
2038 (bytenr - cache->key.objectid));
324ae4df 2039 if (pin) {
25179201 2040 spin_lock(&cache->space_info->lock);
e8569813
ZY		 2041 spin_lock(&cache->lock);
2042 cache->pinned += len;
2043 cache->space_info->bytes_pinned += len;
2044 spin_unlock(&cache->lock);
25179201 2045 spin_unlock(&cache->space_info->lock);
324ae4df
Y		 2046 fs_info->total_pinned += len;
2047 } else {
25179201 2048 spin_lock(&cache->space_info->lock);
e8569813
ZY		 2049 spin_lock(&cache->lock);
2050 cache->pinned -= len;
2051 cache->space_info->bytes_pinned -= len;
2052 spin_unlock(&cache->lock);
25179201 2053 spin_unlock(&cache->space_info->lock);
324ae4df 2054 fs_info->total_pinned -= len;
07103a3c
JB		 2055 if (cache->cached)
2056 btrfs_add_free_space(cache, bytenr, len);
324ae4df
Y		 2057 }
2058 bytenr += len;
2059 num -= len;
2060 }
2061 return 0;
2062}
9078a3e1 2063
e8569813
ZY		 2064static int update_reserved_extents(struct btrfs_root *root,
2065 u64 bytenr, u64 num, int reserve)
2066{
2067 u64 len;
2068 struct btrfs_block_group_cache *cache;
2069 struct btrfs_fs_info *fs_info = root->fs_info;
2070
e8569813
ZY		 2071 while (num > 0) {
2072 cache = btrfs_lookup_block_group(fs_info, bytenr);
2073 BUG_ON(!cache);
2074 len = min(num, cache->key.offset -
2075 (bytenr - cache->key.objectid));
25179201
JB		 2076
2077 spin_lock(&cache->space_info->lock);
2078 spin_lock(&cache->lock);
e8569813 2079 if (reserve) {
e8569813
ZY		 2080 cache->reserved += len;
2081 cache->space_info->bytes_reserved += len;
e8569813 2082 } else {
e8569813
ZY		 2083 cache->reserved -= len;
2084 cache->space_info->bytes_reserved -= len;
e8569813 2085 }
25179201
JB		 2086 spin_unlock(&cache->lock);
2087 spin_unlock(&cache->space_info->lock);
e8569813
ZY		 2088 bytenr += len;
2089 num -= len;
2090 }
2091 return 0;
2092}
2093
d1310b2e 2094int btrfs_copy_pinned(struct btrfs_root *root, struct extent_io_tree *copy)
ccd467d6 2095{
ccd467d6 2096 u64 last = 0;
1a5bc167
CM		 2097 u64 start;
2098 u64 end;
d1310b2e 2099 struct extent_io_tree *pinned_extents = &root->fs_info->pinned_extents;
ccd467d6 2100 int ret;
ccd467d6 2101
25179201 2102 mutex_lock(&root->fs_info->pinned_mutex);
ccd467d6 2103 while(1) {
1a5bc167
CM		 2104 ret = find_first_extent_bit(pinned_extents, last,
2105 &start, &end, EXTENT_DIRTY);
2106 if (ret)
ccd467d6 2107 break;
1a5bc167
CM		 2108 set_extent_dirty(copy, start, end, GFP_NOFS);
2109 last = end + 1;
ccd467d6 2110 }
25179201 2111 mutex_unlock(&root->fs_info->pinned_mutex);
ccd467d6
CM		 2112 return 0;
2113}
2114
2115int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
2116 struct btrfs_root *root,
d1310b2e 2117 struct extent_io_tree *unpin)
a28ec197 2118{
1a5bc167
CM		 2119 u64 start;
2120 u64 end;
a28ec197 2121 int ret;
a28ec197 2122
25179201 2123 mutex_lock(&root->fs_info->pinned_mutex);
a28ec197 2124 while(1) {
1a5bc167
CM		 2125 ret = find_first_extent_bit(unpin, 0, &start, &end,
2126 EXTENT_DIRTY);
2127 if (ret)
a28ec197 2128 break;
e02119d5 2129 btrfs_update_pinned_extents(root, start, end + 1 - start, 0);
1a5bc167 2130 clear_extent_dirty(unpin, start, end, GFP_NOFS);
c286ac48 2131 if (need_resched()) {
25179201 2132 mutex_unlock(&root->fs_info->pinned_mutex);
c286ac48 2133 cond_resched();
25179201 2134 mutex_lock(&root->fs_info->pinned_mutex);
c286ac48 2135 }
a28ec197 2136 }
25179201 2137 mutex_unlock(&root->fs_info->pinned_mutex);
a28ec197
CM		 2138 return 0;
2139}
2140
98ed5174 2141static int finish_current_insert(struct btrfs_trans_handle *trans,
87ef2bb4 2142 struct btrfs_root *extent_root, int all)
037e6390 2143{
7bb86316
CM		 2144 u64 start;
2145 u64 end;
31840ae1 2146 u64 priv;
25179201 2147 u64 search = 0;
f3465ca4 2148 u64 skipped = 0;
7bb86316
CM		 2149 struct btrfs_fs_info *info = extent_root->fs_info;
2150 struct btrfs_path *path;
f3465ca4
JB		 2151 struct pending_extent_op *extent_op, *tmp;
2152 struct list_head insert_list, update_list;
037e6390 2153 int ret;
f3465ca4 2154 int num_inserts = 0, max_inserts;
037e6390 2155
7bb86316 2156 path = btrfs_alloc_path();
f3465ca4
JB		 2157 INIT_LIST_HEAD(&insert_list);
2158 INIT_LIST_HEAD(&update_list);
037e6390 2159
f3465ca4
JB		 2160 max_inserts = extent_root->leafsize /
2161 (2 * sizeof(struct btrfs_key) + 2 * sizeof(struct btrfs_item) +
2162 sizeof(struct btrfs_extent_ref) +
2163 sizeof(struct btrfs_extent_item));
2164again:
2165 mutex_lock(&info->extent_ins_mutex);
2166 while (1) {
25179201
JB		 2167 ret = find_first_extent_bit(&info->extent_ins, search, &start,
2168 &end, EXTENT_WRITEBACK);
2169 if (ret) {
f3465ca4
JB		 2170 if (skipped && all && !num_inserts) {
2171 skipped = 0;
b4eec2ca 2172 search = 0;
25179201
JB		 2173 continue;
2174 }
f3465ca4 2175 mutex_unlock(&info->extent_ins_mutex);
26b8003f 2176 break;
25179201
JB		 2177 }
2178
2179 ret = try_lock_extent(&info->extent_ins, start, end, GFP_NOFS);
2180 if (!ret) {
f3465ca4 2181 skipped = 1;
87ef2bb4 2182 search = end + 1;
f3465ca4
JB		 2183 if (need_resched()) {
2184 mutex_unlock(&info->extent_ins_mutex);
2185 cond_resched();
2186 mutex_lock(&info->extent_ins_mutex);
2187 }
25179201
JB		 2188 continue;
2189 }
26b8003f 2190
31840ae1
ZY		 2191 ret = get_state_private(&info->extent_ins, start, &priv);
2192 BUG_ON(ret);
f3465ca4 2193 extent_op = (struct pending_extent_op *)(unsigned long) priv;
25179201 2194
31840ae1 2195 if (extent_op->type == PENDING_EXTENT_INSERT) {
f3465ca4
JB		 2196 num_inserts++;
2197 list_add_tail(&extent_op->list, &insert_list);
2198 search = end + 1;
2199 if (num_inserts == max_inserts) {
2200 mutex_unlock(&info->extent_ins_mutex);
2201 break;
2202 }
2203 } else if (extent_op->type == PENDING_BACKREF_UPDATE) {
2204 list_add_tail(&extent_op->list, &update_list);
2205 search = end + 1;
2206 } else {
2207 BUG();
2208 }
2209 }
c286ac48 2210
f3465ca4 2211 /*
b4eec2ca 2212 * process the update list, clear the writeback bit for it, and if
f3465ca4
JB		 2213 * somebody marked this thing for deletion then just unlock it and be
2214 * done, the free_extents will handle it
2215 */
2216 mutex_lock(&info->extent_ins_mutex);
2217 list_for_each_entry_safe(extent_op, tmp, &update_list, list) {
2218 clear_extent_bits(&info->extent_ins, extent_op->bytenr,
2219 extent_op->bytenr + extent_op->num_bytes - 1,
2220 EXTENT_WRITEBACK, GFP_NOFS);
2221 if (extent_op->del) {
2222 list_del_init(&extent_op->list);
2223 unlock_extent(&info->extent_ins, extent_op->bytenr,
2224 extent_op->bytenr + extent_op->num_bytes
2225 - 1, GFP_NOFS);
2226 kfree(extent_op);
2227 }
2228 }
2229 mutex_unlock(&info->extent_ins_mutex);
c286ac48 2230
f3465ca4
JB		 2231 /*
2232 * still have things left on the update list, go ahead an update
2233 * everything
2234 */
2235 if (!list_empty(&update_list)) {
2236 ret = update_backrefs(trans, extent_root, path, &update_list);
2237 BUG_ON(ret);
2238 }
c286ac48 2239
f3465ca4
JB		 2240 /*
2241 * if no inserts need to be done, but we skipped some extents and we
2242 * need to make sure everything is cleaned then reset everything and
2243 * go back to the beginning
2244 */
2245 if (!num_inserts && all && skipped) {
2246 search = 0;
2247 skipped = 0;
2248 INIT_LIST_HEAD(&update_list);
2249 INIT_LIST_HEAD(&insert_list);
2250 goto again;
2251 } else if (!num_inserts) {
2252 goto out;
2253 }
31840ae1 2254
f3465ca4
JB		 2255 /*
2256 * process the insert extents list. Again if we are deleting this
2257 * extent, then just unlock it, pin down the bytes if need be, and be
2258 * done with it. Saves us from having to actually insert the extent
2259 * into the tree and then subsequently come along and delete it
2260 */
2261 mutex_lock(&info->extent_ins_mutex);
2262 list_for_each_entry_safe(extent_op, tmp, &insert_list, list) {
2263 clear_extent_bits(&info->extent_ins, extent_op->bytenr,
2264 extent_op->bytenr + extent_op->num_bytes - 1,
2265 EXTENT_WRITEBACK, GFP_NOFS);
2266 if (extent_op->del) {
2267 list_del_init(&extent_op->list);
2268 unlock_extent(&info->extent_ins, extent_op->bytenr,
2269 extent_op->bytenr + extent_op->num_bytes
2270 - 1, GFP_NOFS);
2271
2272 mutex_lock(&extent_root->fs_info->pinned_mutex);
2273 ret = pin_down_bytes(trans, extent_root,
2274 extent_op->bytenr,
2275 extent_op->num_bytes, 0);
2276 mutex_unlock(&extent_root->fs_info->pinned_mutex);
2277
2278 ret = update_block_group(trans, extent_root,
2279 extent_op->bytenr,
2280 extent_op->num_bytes,
2281 0, ret > 0);
2282 BUG_ON(ret);
2283 kfree(extent_op);
2284 num_inserts--;
d8d5f3e1 2285 }
f3465ca4
JB		 2286 }
2287 mutex_unlock(&info->extent_ins_mutex);
31840ae1 2288
f3465ca4
JB		 2289 ret = insert_extents(trans, extent_root, path, &insert_list,
2290 num_inserts);
2291 BUG_ON(ret);
2292
2293 /*
2294 * if we broke out of the loop in order to insert stuff because we hit
2295 * the maximum number of inserts at a time we can handle, then loop
2296 * back and pick up where we left off
2297 */
2298 if (num_inserts == max_inserts) {
2299 INIT_LIST_HEAD(&insert_list);
2300 INIT_LIST_HEAD(&update_list);
2301 num_inserts = 0;
2302 goto again;
2303 }
2304
2305 /*
2306 * again, if we need to make absolutely sure there are no more pending
2307 * extent operations left and we know that we skipped some, go back to
2308 * the beginning and do it all again
2309 */
2310 if (all && skipped) {
2311 INIT_LIST_HEAD(&insert_list);
2312 INIT_LIST_HEAD(&update_list);
2313 search = 0;
2314 skipped = 0;
2315 num_inserts = 0;
2316 goto again;
037e6390 2317 }
f3465ca4 2318out:
7bb86316 2319 btrfs_free_path(path);
037e6390
CM		 2320 return 0;
2321}
2322
31840ae1
ZY		 2323static int pin_down_bytes(struct btrfs_trans_handle *trans,
2324 struct btrfs_root *root,
2325 u64 bytenr, u64 num_bytes, int is_data)
e20d96d6 2326{
1a5bc167 2327 int err = 0;
31840ae1 2328 struct extent_buffer *buf;
8ef97622 2329
31840ae1
ZY		 2330 if (is_data)
2331 goto pinit;
2332
2333 buf = btrfs_find_tree_block(root, bytenr, num_bytes);
2334 if (!buf)
2335 goto pinit;
2336
2337 /* we can reuse a block if it hasn't been written
2338 * and it is from this transaction. We can't
2339 * reuse anything from the tree log root because
2340 * it has tiny sub-transactions.
2341 */
2342 if (btrfs_buffer_uptodate(buf, 0) &&
2343 btrfs_try_tree_lock(buf)) {
2344 u64 header_owner = btrfs_header_owner(buf);
2345 u64 header_transid = btrfs_header_generation(buf);
2346 if (header_owner != BTRFS_TREE_LOG_OBJECTID &&
1a40e23b 2347 header_owner != BTRFS_TREE_RELOC_OBJECTID &&
31840ae1
ZY		 2348 header_transid == trans->transid &&
2349 !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
2350 clean_tree_block(NULL, root, buf);
2351 btrfs_tree_unlock(buf);
5f39d397 2352 free_extent_buffer(buf);
31840ae1 2353 return 1;
8ef97622 2354 }
31840ae1 2355 btrfs_tree_unlock(buf);
f4b9aa8d 2356 }
31840ae1
ZY		 2357 free_extent_buffer(buf);
2358pinit:
2359 btrfs_update_pinned_extents(root, bytenr, num_bytes, 1);
2360
be744175 2361 BUG_ON(err < 0);
e20d96d6
CM		 2362 return 0;
2363}
2364
fec577fb 2365/*
a28ec197 2366 * remove an extent from the root, returns 0 on success
fec577fb 2367 */
31840ae1
ZY		 2368static int __free_extent(struct btrfs_trans_handle *trans,
2369 struct btrfs_root *root,
2370 u64 bytenr, u64 num_bytes, u64 parent,
7bb86316 2371 u64 root_objectid, u64 ref_generation,
3bb1a1bc 2372 u64 owner_objectid, int pin, int mark_free)
a28ec197 2373{
5caf2a00 2374 struct btrfs_path *path;
e2fa7227 2375 struct btrfs_key key;
1261ec42
CM		 2376 struct btrfs_fs_info *info = root->fs_info;
2377 struct btrfs_root *extent_root = info->extent_root;
5f39d397 2378 struct extent_buffer *leaf;
a28ec197 2379 int ret;
952fccac
CM		 2380 int extent_slot = 0;
2381 int found_extent = 0;
2382 int num_to_del = 1;
234b63a0 2383 struct btrfs_extent_item *ei;
cf27e1ee 2384 u32 refs;
037e6390 2385
db94535d 2386 key.objectid = bytenr;
62e2749e 2387 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
db94535d 2388 key.offset = num_bytes;
5caf2a00 2389 path = btrfs_alloc_path();
54aa1f4d
CM		 2390 if (!path)
2391 return -ENOMEM;
5f26f772 2392
3c12ac72 2393 path->reada = 1;
3bb1a1bc
YZ		 2394 ret = lookup_extent_backref(trans, extent_root, path,
2395 bytenr, parent, root_objectid,
2396 ref_generation, owner_objectid, 1);
7bb86316 2397 if (ret == 0) {
952fccac
CM		 2398 struct btrfs_key found_key;
2399 extent_slot = path->slots[0];
2400 while(extent_slot > 0) {
2401 extent_slot--;
2402 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
2403 extent_slot);
2404 if (found_key.objectid != bytenr)
2405 break;
2406 if (found_key.type == BTRFS_EXTENT_ITEM_KEY &&
2407 found_key.offset == num_bytes) {
2408 found_extent = 1;
2409 break;
2410 }
2411 if (path->slots[0] - extent_slot > 5)
2412 break;
2413 }
31840ae1
ZY		 2414 if (!found_extent) {
2415 ret = remove_extent_backref(trans, extent_root, path);
2416 BUG_ON(ret);
2417 btrfs_release_path(extent_root, path);
2418 ret = btrfs_search_slot(trans, extent_root,
2419 &key, path, -1, 1);
f3465ca4
JB		 2420 if (ret) {
2421 printk(KERN_ERR "umm, got %d back from search"
2422 ", was looking for %Lu\n", ret,
2423 bytenr);
2424 btrfs_print_leaf(extent_root, path->nodes[0]);
2425 }
31840ae1
ZY		 2426 BUG_ON(ret);
2427 extent_slot = path->slots[0];
2428 }
7bb86316
CM		 2429 } else {
2430 btrfs_print_leaf(extent_root, path->nodes[0]);
2431 WARN_ON(1);
2432 printk("Unable to find ref byte nr %Lu root %Lu "
3bb1a1bc
YZ		 2433 "gen %Lu owner %Lu\n", bytenr,
2434 root_objectid, ref_generation, owner_objectid);
7bb86316 2435 }
5f39d397
CM		 2436
2437 leaf = path->nodes[0];
952fccac 2438 ei = btrfs_item_ptr(leaf, extent_slot,
123abc88 2439 struct btrfs_extent_item);
5f39d397
CM		 2440 refs = btrfs_extent_refs(leaf, ei);
2441 BUG_ON(refs == 0);
2442 refs -= 1;
2443 btrfs_set_extent_refs(leaf, ei, refs);
952fccac 2444
5f39d397
CM		 2445 btrfs_mark_buffer_dirty(leaf);
2446
952fccac 2447 if (refs == 0 && found_extent && path->slots[0] == extent_slot + 1) {
31840ae1
ZY		 2448 struct btrfs_extent_ref *ref;
2449 ref = btrfs_item_ptr(leaf, path->slots[0],
2450 struct btrfs_extent_ref);
2451 BUG_ON(btrfs_ref_num_refs(leaf, ref) != 1);
952fccac
CM		 2452 /* if the back ref and the extent are next to each other
2453 * they get deleted below in one shot
2454 */
2455 path->slots[0] = extent_slot;
2456 num_to_del = 2;
2457 } else if (found_extent) {
2458 /* otherwise delete the extent back ref */
31840ae1 2459 ret = remove_extent_backref(trans, extent_root, path);
952fccac
CM		 2460 BUG_ON(ret);
2461 /* if refs are 0, we need to setup the path for deletion */
2462 if (refs == 0) {
2463 btrfs_release_path(extent_root, path);
2464 ret = btrfs_search_slot(trans, extent_root, &key, path,
2465 -1, 1);
952fccac
CM		 2466 BUG_ON(ret);
2467 }
2468 }
2469
cf27e1ee 2470 if (refs == 0) {
db94535d
CM		 2471 u64 super_used;
2472 u64 root_used;
21af804c
DW		 2473#ifdef BIO_RW_DISCARD
2474 u64 map_length = num_bytes;
2475 struct btrfs_multi_bio *multi = NULL;
2476#endif
78fae27e
CM		 2477
2478 if (pin) {
25179201 2479 mutex_lock(&root->fs_info->pinned_mutex);
31840ae1
ZY		 2480 ret = pin_down_bytes(trans, root, bytenr, num_bytes,
2481 owner_objectid >= BTRFS_FIRST_FREE_OBJECTID);
25179201 2482 mutex_unlock(&root->fs_info->pinned_mutex);
c549228f
Y		 2483 if (ret > 0)
2484 mark_free = 1;
2485 BUG_ON(ret < 0);
78fae27e 2486 }
58176a96 2487 /* block accounting for super block */
a2135011 2488 spin_lock_irq(&info->delalloc_lock);
db94535d
CM		 2489 super_used = btrfs_super_bytes_used(&info->super_copy);
2490 btrfs_set_super_bytes_used(&info->super_copy,
2491 super_used - num_bytes);
a2135011 2492 spin_unlock_irq(&info->delalloc_lock);
58176a96
JB		 2493
2494 /* block accounting for root item */
db94535d 2495 root_used = btrfs_root_used(&root->root_item);
5f39d397 2496 btrfs_set_root_used(&root->root_item,
db94535d 2497 root_used - num_bytes);
952fccac
CM		 2498 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
2499 num_to_del);
31840ae1 2500 BUG_ON(ret);
25179201 2501 btrfs_release_path(extent_root, path);
db94535d 2502 ret = update_block_group(trans, root, bytenr, num_bytes, 0,
0b86a832 2503 mark_free);
9078a3e1 2504 BUG_ON(ret);
21af804c 2505
459931ec
CM		 2506 if (owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
2507 ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
2508 BUG_ON(ret);
2509 }
2510
21af804c
DW		 2511#ifdef BIO_RW_DISCARD
2512 /* Tell the block device(s) that the sectors can be discarded */
2513 ret = btrfs_map_block(&root->fs_info->mapping_tree, READ,
2514 bytenr, &map_length, &multi, 0);
2515 if (!ret) {
2516 struct btrfs_bio_stripe *stripe = multi->stripes;
2517 int i;
2518
2519 if (map_length > num_bytes)
2520 map_length = num_bytes;
2521
2522 for (i = 0; i < multi->num_stripes; i++, stripe++) {
15916de8
CM		 2523 btrfs_issue_discard(stripe->dev->bdev,
2524 stripe->physical,
2525 map_length);
21af804c
DW		 2526 }
2527 kfree(multi);
2528 }
2529#endif
a28ec197 2530 }
5caf2a00 2531 btrfs_free_path(path);
87ef2bb4 2532 finish_current_insert(trans, extent_root, 0);
a28ec197
CM		 2533 return ret;
2534}
2535
a28ec197
CM		 2536/*
2537 * find all the blocks marked as pending in the radix tree and remove
2538 * them from the extent map
2539 */
e089f05c 2540static int del_pending_extents(struct btrfs_trans_handle *trans, struct
87ef2bb4 2541 btrfs_root *extent_root, int all)
a28ec197
CM		 2542{
2543 int ret;
e20d96d6 2544 int err = 0;
1a5bc167
CM		 2545 u64 start;
2546 u64 end;
31840ae1 2547 u64 priv;
25179201 2548 u64 search = 0;
f3465ca4 2549 int nr = 0, skipped = 0;
d1310b2e 2550 struct extent_io_tree *pending_del;
31840ae1
ZY		 2551 struct extent_io_tree *extent_ins;
2552 struct pending_extent_op *extent_op;
25179201 2553 struct btrfs_fs_info *info = extent_root->fs_info;
f3465ca4 2554 struct list_head delete_list;
8ef97622 2555
f3465ca4 2556 INIT_LIST_HEAD(&delete_list);
31840ae1 2557 extent_ins = &extent_root->fs_info->extent_ins;
1a5bc167 2558 pending_del = &extent_root->fs_info->pending_del;
a28ec197 2559
f3465ca4
JB		 2560again:
2561 mutex_lock(&info->extent_ins_mutex);
a28ec197 2562 while(1) {
25179201
JB		 2563 ret = find_first_extent_bit(pending_del, search, &start, &end,
2564 EXTENT_WRITEBACK);
2565 if (ret) {
f3465ca4 2566 if (all && skipped && !nr) {
25179201
JB		 2567 search = 0;
2568 continue;
2569 }
f3465ca4 2570 mutex_unlock(&info->extent_ins_mutex);
a28ec197 2571 break;
25179201
JB		 2572 }
2573
2574 ret = try_lock_extent(extent_ins, start, end, GFP_NOFS);
2575 if (!ret) {
2576 search = end+1;
f3465ca4
JB		 2577 skipped = 1;
2578
2579 if (need_resched()) {
2580 mutex_unlock(&info->extent_ins_mutex);
2581 cond_resched();
2582 mutex_lock(&info->extent_ins_mutex);
2583 }
2584
25179201
JB		 2585 continue;
2586 }
2587 BUG_ON(ret < 0);
31840ae1
ZY		 2588
2589 ret = get_state_private(pending_del, start, &priv);
2590 BUG_ON(ret);
2591 extent_op = (struct pending_extent_op *)(unsigned long)priv;
2592
25179201 2593 clear_extent_bits(pending_del, start, end, EXTENT_WRITEBACK,
1a5bc167 2594 GFP_NOFS);
31840ae1 2595 if (!test_range_bit(extent_ins, start, end,
25179201 2596 EXTENT_WRITEBACK, 0)) {
f3465ca4
JB		 2597 list_add_tail(&extent_op->list, &delete_list);
2598 nr++;
c286ac48 2599 } else {
31840ae1 2600 kfree(extent_op);
25179201
JB		 2601
2602 ret = get_state_private(&info->extent_ins, start,
2603 &priv);
31840ae1
ZY		 2604 BUG_ON(ret);
2605 extent_op = (struct pending_extent_op *)
25179201
JB		 2606 (unsigned long)priv;
2607
2608 clear_extent_bits(&info->extent_ins, start, end,
2609 EXTENT_WRITEBACK, GFP_NOFS);
31840ae1 2610
f3465ca4
JB		 2611 if (extent_op->type == PENDING_BACKREF_UPDATE) {
2612 list_add_tail(&extent_op->list, &delete_list);
2613 search = end + 1;
2614 nr++;
2615 continue;
2616 }
31840ae1 2617
25179201
JB		 2618 mutex_lock(&extent_root->fs_info->pinned_mutex);
2619 ret = pin_down_bytes(trans, extent_root, start,
2620 end + 1 - start, 0);
2621 mutex_unlock(&extent_root->fs_info->pinned_mutex);
2622
31840ae1 2623 ret = update_block_group(trans, extent_root, start,
25179201
JB		 2624 end + 1 - start, 0, ret > 0);
2625
f3465ca4 2626 unlock_extent(extent_ins, start, end, GFP_NOFS);
31840ae1
ZY		 2627 BUG_ON(ret);
2628 kfree(extent_op);
c286ac48 2629 }
1a5bc167
CM		 2630 if (ret)
2631 err = ret;
c286ac48 2632
f3465ca4
JB		 2633 search = end + 1;
2634
2635 if (need_resched()) {
2636 mutex_unlock(&info->extent_ins_mutex);
2637 cond_resched();
2638 mutex_lock(&info->extent_ins_mutex);
2639 }
2640 }
2641
2642 if (nr) {
2643 ret = free_extents(trans, extent_root, &delete_list);
2644 BUG_ON(ret);
2645 }
2646
2647 if (all && skipped) {
2648 INIT_LIST_HEAD(&delete_list);
2649 search = 0;
2650 nr = 0;
2651 goto again;
fec577fb 2652 }
f3465ca4 2653
e20d96d6 2654 return err;
fec577fb
CM		 2655}
2656
2657/*
2658 * remove an extent from the root, returns 0 on success
2659 */
925baedd 2660static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
31840ae1
ZY		 2661 struct btrfs_root *root,
2662 u64 bytenr, u64 num_bytes, u64 parent,
2663 u64 root_objectid, u64 ref_generation,
3bb1a1bc 2664 u64 owner_objectid, int pin)
fec577fb 2665{
9f5fae2f 2666 struct btrfs_root *extent_root = root->fs_info->extent_root;
fec577fb
CM		 2667 int pending_ret;
2668 int ret;
a28ec197 2669
db94535d 2670 WARN_ON(num_bytes < root->sectorsize);
fec577fb 2671 if (root == extent_root) {
f3465ca4
JB		 2672 struct pending_extent_op *extent_op = NULL;
2673
2674 mutex_lock(&root->fs_info->extent_ins_mutex);
2675 if (test_range_bit(&root->fs_info->extent_ins, bytenr,
2676 bytenr + num_bytes - 1, EXTENT_WRITEBACK, 0)) {
2677 u64 priv;
2678 ret = get_state_private(&root->fs_info->extent_ins,
2679 bytenr, &priv);
2680 BUG_ON(ret);
2681 extent_op = (struct pending_extent_op *)
2682 (unsigned long)priv;
2683
2684 extent_op->del = 1;
2685 if (extent_op->type == PENDING_EXTENT_INSERT) {
2686 mutex_unlock(&root->fs_info->extent_ins_mutex);
2687 return 0;
2688 }
2689 }
2690
2691 if (extent_op) {
2692 ref_generation = extent_op->orig_generation;
2693 parent = extent_op->orig_parent;
2694 }
31840ae1
ZY		 2695
2696 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
2697 BUG_ON(!extent_op);
2698
2699 extent_op->type = PENDING_EXTENT_DELETE;
2700 extent_op->bytenr = bytenr;
2701 extent_op->num_bytes = num_bytes;
2702 extent_op->parent = parent;
2703 extent_op->orig_parent = parent;
2704 extent_op->generation = ref_generation;
2705 extent_op->orig_generation = ref_generation;
2706 extent_op->level = (int)owner_objectid;
f3465ca4
JB		 2707 INIT_LIST_HEAD(&extent_op->list);
2708 extent_op->del = 0;
31840ae1
ZY		 2709
2710 set_extent_bits(&root->fs_info->pending_del,
2711 bytenr, bytenr + num_bytes - 1,
25179201 2712 EXTENT_WRITEBACK, GFP_NOFS);
31840ae1
ZY		 2713 set_state_private(&root->fs_info->pending_del,
2714 bytenr, (unsigned long)extent_op);
25179201 2715 mutex_unlock(&root->fs_info->extent_ins_mutex);
fec577fb
CM		 2716 return 0;
2717 }
4bef0848 2718 /* if metadata always pin */
d00aff00
CM		 2719 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) {
2720 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
0f9dd46c
JB		 2721 struct btrfs_block_group_cache *cache;
2722
d00aff00 2723 /* btrfs_free_reserved_extent */
0f9dd46c
JB		 2724 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
2725 BUG_ON(!cache);
2726 btrfs_add_free_space(cache, bytenr, num_bytes);
e8569813 2727 update_reserved_extents(root, bytenr, num_bytes, 0);
d00aff00
CM		 2728 return 0;
2729 }
4bef0848 2730 pin = 1;
d00aff00 2731 }
4bef0848
CM		 2732
2733 /* if data pin when any transaction has committed this */
2734 if (ref_generation != trans->transid)
2735 pin = 1;
2736
31840ae1 2737 ret = __free_extent(trans, root, bytenr, num_bytes, parent,
3bb1a1bc
YZ		 2738 root_objectid, ref_generation,
2739 owner_objectid, pin, pin == 0);
ee6e6504 2740
87ef2bb4
CM		 2741 finish_current_insert(trans, root->fs_info->extent_root, 0);
2742 pending_ret = del_pending_extents(trans, root->fs_info->extent_root, 0);
fec577fb
CM		 2743 return ret ? ret : pending_ret;
2744}
2745
925baedd 2746int btrfs_free_extent(struct btrfs_trans_handle *trans,
31840ae1
ZY		 2747 struct btrfs_root *root,
2748 u64 bytenr, u64 num_bytes, u64 parent,
2749 u64 root_objectid, u64 ref_generation,
3bb1a1bc 2750 u64 owner_objectid, int pin)
925baedd
CM		 2751{
2752 int ret;
2753
31840ae1 2754 ret = __btrfs_free_extent(trans, root, bytenr, num_bytes, parent,
925baedd 2755 root_objectid, ref_generation,
3bb1a1bc 2756 owner_objectid, pin);
925baedd
CM		 2757 return ret;
2758}
2759
87ee04eb
CM		 2760static u64 stripe_align(struct btrfs_root *root, u64 val)
2761{
2762 u64 mask = ((u64)root->stripesize - 1);
2763 u64 ret = (val + mask) & ~mask;
2764 return ret;
2765}
2766
fec577fb
CM		 2767/*
2768 * walks the btree of allocated extents and find a hole of a given size.
2769 * The key ins is changed to record the hole:
2770 * ins->objectid == block start
62e2749e 2771 * ins->flags = BTRFS_EXTENT_ITEM_KEY
fec577fb
CM		 2772 * ins->offset == number of blocks
2773 * Any available blocks before search_start are skipped.
2774 */
98ed5174
CM		 2775static int noinline find_free_extent(struct btrfs_trans_handle *trans,
2776 struct btrfs_root *orig_root,
2777 u64 num_bytes, u64 empty_size,
2778 u64 search_start, u64 search_end,
2779 u64 hint_byte, struct btrfs_key *ins,
2780 u64 exclude_start, u64 exclude_nr,
2781 int data)
fec577fb 2782{
80eb234a 2783 int ret = 0;
9f5fae2f 2784 struct btrfs_root * root = orig_root->fs_info->extent_root;
db94535d 2785 u64 total_needed = num_bytes;
239b14b3 2786 u64 *last_ptr = NULL;
4366211c 2787 u64 last_wanted = 0;
80eb234a 2788 struct btrfs_block_group_cache *block_group = NULL;
0ef3e66b 2789 int chunk_alloc_done = 0;
239b14b3 2790 int empty_cluster = 2 * 1024 * 1024;
0ef3e66b 2791 int allowed_chunk_alloc = 0;
80eb234a
JB		 2792 struct list_head *head = NULL, *cur = NULL;
2793 int loop = 0;
f5a31e16 2794 int extra_loop = 0;
80eb234a 2795 struct btrfs_space_info *space_info;
fec577fb 2796
db94535d 2797 WARN_ON(num_bytes < root->sectorsize);
b1a4d965 2798 btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
80eb234a
JB		 2799 ins->objectid = 0;
2800 ins->offset = 0;
b1a4d965 2801
0ef3e66b
CM		 2802 if (orig_root->ref_cows || empty_size)
2803 allowed_chunk_alloc = 1;
2804
239b14b3
CM		 2805 if (data & BTRFS_BLOCK_GROUP_METADATA) {
2806 last_ptr = &root->fs_info->last_alloc;
4366211c 2807 empty_cluster = 64 * 1024;
239b14b3
CM		 2808 }
2809
0f9dd46c 2810 if ((data & BTRFS_BLOCK_GROUP_DATA) && btrfs_test_opt(root, SSD))
239b14b3 2811 last_ptr = &root->fs_info->last_data_alloc;
0f9dd46c 2812
239b14b3 2813 if (last_ptr) {
4366211c 2814 if (*last_ptr) {
239b14b3 2815 hint_byte = *last_ptr;
4366211c
CM		 2816 last_wanted = *last_ptr;
2817 } else
239b14b3 2818 empty_size += empty_cluster;
4366211c
CM		 2819 } else {
2820 empty_cluster = 0;
239b14b3 2821 }
a061fc8d 2822 search_start = max(search_start, first_logical_byte(root, 0));
239b14b3 2823 search_start = max(search_start, hint_byte);
0b86a832 2824
42e70e7a 2825 if (last_wanted && search_start != last_wanted) {
4366211c 2826 last_wanted = 0;
42e70e7a
CM		 2827 empty_size += empty_cluster;
2828 }
4366211c 2829
42e70e7a 2830 total_needed += empty_size;
80eb234a 2831 block_group = btrfs_lookup_block_group(root->fs_info, search_start);
d899e052
YZ		 2832 if (!block_group)
2833 block_group = btrfs_lookup_first_block_group(root->fs_info,
2834 search_start);
80eb234a 2835 space_info = __find_space_info(root->fs_info, data);
0f9dd46c 2836
80eb234a
JB		 2837 down_read(&space_info->groups_sem);
2838 while (1) {
2839 struct btrfs_free_space *free_space;
0f9dd46c 2840 /*
80eb234a
JB		 2841 * the only way this happens if our hint points to a block
2842 * group thats not of the proper type, while looping this
2843 * should never happen
0f9dd46c 2844 */
8a1413a2
CM		 2845 if (empty_size)
2846 extra_loop = 1;
2847
42e70e7a
CM		 2848 if (!block_group)
2849 goto new_group_no_lock;
2850
ea6a478e
JB		 2851 if (unlikely(!block_group->cached)) {
2852 mutex_lock(&block_group->cache_mutex);
2853 ret = cache_block_group(root, block_group);
2854 mutex_unlock(&block_group->cache_mutex);
2855 if (ret)
2856 break;
2857 }
2858
25179201 2859 mutex_lock(&block_group->alloc_mutex);
80eb234a 2860 if (unlikely(!block_group_bits(block_group, data)))
0f9dd46c 2861 goto new_group;
0f9dd46c 2862
ea6a478e 2863 if (unlikely(block_group->ro))
80eb234a 2864 goto new_group;
0f9dd46c 2865
80eb234a
JB		 2866 free_space = btrfs_find_free_space(block_group, search_start,
2867 total_needed);
2868 if (free_space) {
2869 u64 start = block_group->key.objectid;
2870 u64 end = block_group->key.objectid +
0f9dd46c 2871 block_group->key.offset;
239b14b3 2872
80eb234a 2873 search_start = stripe_align(root, free_space->offset);
0f9dd46c 2874
80eb234a
JB		 2875 /* move on to the next group */
2876 if (search_start + num_bytes >= search_end)
2877 goto new_group;
e37c9e69 2878
80eb234a
JB		 2879 /* move on to the next group */
2880 if (search_start + num_bytes > end)
2881 goto new_group;
2882
4366211c 2883 if (last_wanted && search_start != last_wanted) {
3b7885bf 2884 total_needed += empty_cluster;
8a1413a2 2885 empty_size += empty_cluster;
4366211c 2886 last_wanted = 0;
3b7885bf
CM		 2887 /*
2888 * if search_start is still in this block group
2889 * then we just re-search this block group
2890 */
2891 if (search_start >= start &&
2892 search_start < end) {
2893 mutex_unlock(&block_group->alloc_mutex);
2894 continue;
2895 }
2896
2897 /* else we go to the next block group */
2898 goto new_group;
2899 }
2900
80eb234a
JB		 2901 if (exclude_nr > 0 &&
2902 (search_start + num_bytes > exclude_start &&
2903 search_start < exclude_start + exclude_nr)) {
2904 search_start = exclude_start + exclude_nr;
2905 /*
2906 * if search_start is still in this block group
2907 * then we just re-search this block group
2908 */
2909 if (search_start >= start &&
25179201
JB		 2910 search_start < end) {
2911 mutex_unlock(&block_group->alloc_mutex);
4366211c 2912 last_wanted = 0;
80eb234a 2913 continue;
25179201 2914 }
80eb234a
JB		 2915
2916 /* else we go to the next block group */
2917 goto new_group;
2918 }
2919
2920 ins->objectid = search_start;
2921 ins->offset = num_bytes;
25179201
JB		 2922
2923 btrfs_remove_free_space_lock(block_group, search_start,
2924 num_bytes);
80eb234a 2925 /* we are all good, lets return */
25179201 2926 mutex_unlock(&block_group->alloc_mutex);
80eb234a 2927 break;
0f9dd46c 2928 }
80eb234a 2929new_group:
42e70e7a
CM		 2930 mutex_unlock(&block_group->alloc_mutex);
2931new_group_no_lock:
f5a31e16
CM		 2932 /* don't try to compare new allocations against the
2933 * last allocation any more
2934 */
4366211c 2935 last_wanted = 0;
f5a31e16 2936
80eb234a
JB		 2937 /*
2938 * Here's how this works.
2939 * loop == 0: we were searching a block group via a hint
2940 * and didn't find anything, so we start at
2941 * the head of the block groups and keep searching
2942 * loop == 1: we're searching through all of the block groups
2943 * if we hit the head again we have searched
2944 * all of the block groups for this space and we
2945 * need to try and allocate, if we cant error out.
2946 * loop == 2: we allocated more space and are looping through
2947 * all of the block groups again.
2948 */
2949 if (loop == 0) {
2950 head = &space_info->block_groups;
2951 cur = head->next;
80eb234a
JB		 2952 loop++;
2953 } else if (loop == 1 && cur == head) {
f5a31e16
CM		 2954 int keep_going;
2955
2956 /* at this point we give up on the empty_size
2957 * allocations and just try to allocate the min
2958 * space.
2959 *
2960 * The extra_loop field was set if an empty_size
2961 * allocation was attempted above, and if this
2962 * is try we need to try the loop again without
2963 * the additional empty_size.
2964 */
5b7c3fcc
CM		 2965 total_needed -= empty_size;
2966 empty_size = 0;
f5a31e16
CM		 2967 keep_going = extra_loop;
2968 loop++;
42e70e7a 2969
80eb234a
JB		 2970 if (allowed_chunk_alloc && !chunk_alloc_done) {
2971 up_read(&space_info->groups_sem);
2972 ret = do_chunk_alloc(trans, root, num_bytes +
2973 2 * 1024 * 1024, data, 1);
80eb234a 2974 down_read(&space_info->groups_sem);
2ed6d664
CM		 2975 if (ret < 0)
2976 goto loop_check;
80eb234a 2977 head = &space_info->block_groups;
f5a31e16
CM		 2978 /*
2979 * we've allocated a new chunk, keep
2980 * trying
2981 */
2982 keep_going = 1;
80eb234a
JB		 2983 chunk_alloc_done = 1;
2984 } else if (!allowed_chunk_alloc) {
2985 space_info->force_alloc = 1;
f5a31e16 2986 }
2ed6d664 2987loop_check:
f5a31e16
CM		 2988 if (keep_going) {
2989 cur = head->next;
2990 extra_loop = 0;
80eb234a
JB		 2991 } else {
2992 break;
2993 }
2994 } else if (cur == head) {
2995 break;
0f9dd46c 2996 }
0b86a832 2997
80eb234a
JB		 2998 block_group = list_entry(cur, struct btrfs_block_group_cache,
2999 list);
3000 search_start = block_group->key.objectid;
3001 cur = cur->next;
f2654de4 3002 }
0b86a832 3003
80eb234a
JB		 3004 /* we found what we needed */
3005 if (ins->objectid) {
3006 if (!(data & BTRFS_BLOCK_GROUP_DATA))
3007 trans->block_group = block_group;
0f9dd46c 3008
80eb234a
JB		 3009 if (last_ptr)
3010 *last_ptr = ins->objectid + ins->offset;
3011 ret = 0;
3012 } else if (!ret) {
4ce4cb52
JB		 3013 printk(KERN_ERR "we were searching for %Lu bytes, num_bytes %Lu,"
3014 " loop %d, allowed_alloc %d\n", total_needed, num_bytes,
3015 loop, allowed_chunk_alloc);
80eb234a 3016 ret = -ENOSPC;
be744175 3017 }
be744175 3018
80eb234a 3019 up_read(&space_info->groups_sem);
0f70abe2 3020 return ret;
fec577fb 3021}
ec44a35c 3022
0f9dd46c
JB		 3023static void dump_space_info(struct btrfs_space_info *info, u64 bytes)
3024{
3025 struct btrfs_block_group_cache *cache;
3026 struct list_head *l;
3027
3028 printk(KERN_INFO "space_info has %Lu free, is %sfull\n",
e8569813
ZY		 3029 info->total_bytes - info->bytes_used - info->bytes_pinned -
3030 info->bytes_reserved, (info->full) ? "" : "not ");
0f9dd46c 3031
80eb234a 3032 down_read(&info->groups_sem);
0f9dd46c
JB		 3033 list_for_each(l, &info->block_groups) {
3034 cache = list_entry(l, struct btrfs_block_group_cache, list);
3035 spin_lock(&cache->lock);
3036 printk(KERN_INFO "block group %Lu has %Lu bytes, %Lu used "
e8569813 3037 "%Lu pinned %Lu reserved\n",
0f9dd46c 3038 cache->key.objectid, cache->key.offset,
e8569813
ZY		 3039 btrfs_block_group_used(&cache->item),
3040 cache->pinned, cache->reserved);
0f9dd46c
JB		 3041 btrfs_dump_free_space(cache, bytes);
3042 spin_unlock(&cache->lock);
3043 }
80eb234a 3044 up_read(&info->groups_sem);
0f9dd46c 3045}
e8569813 3046
e6dcd2dc
CM		 3047static int __btrfs_reserve_extent(struct btrfs_trans_handle *trans,
3048 struct btrfs_root *root,
3049 u64 num_bytes, u64 min_alloc_size,
3050 u64 empty_size, u64 hint_byte,
3051 u64 search_end, struct btrfs_key *ins,
3052 u64 data)
fec577fb
CM		 3053{
3054 int ret;
fbdc762b 3055 u64 search_start = 0;
8790d502 3056 u64 alloc_profile;
1261ec42 3057 struct btrfs_fs_info *info = root->fs_info;
925baedd 3058
6324fbf3 3059 if (data) {
8790d502
CM		 3060 alloc_profile = info->avail_data_alloc_bits &
3061 info->data_alloc_profile;
3062 data = BTRFS_BLOCK_GROUP_DATA | alloc_profile;
6324fbf3 3063 } else if (root == root->fs_info->chunk_root) {
8790d502
CM		 3064 alloc_profile = info->avail_system_alloc_bits &
3065 info->system_alloc_profile;
3066 data = BTRFS_BLOCK_GROUP_SYSTEM | alloc_profile;
6324fbf3 3067 } else {
8790d502
CM		 3068 alloc_profile = info->avail_metadata_alloc_bits &
3069 info->metadata_alloc_profile;
3070 data = BTRFS_BLOCK_GROUP_METADATA | alloc_profile;
6324fbf3 3071 }
98d20f67 3072again:
2b82032c 3073 data = btrfs_reduce_alloc_profile(root, data);
0ef3e66b
CM		 3074 /*
3075 * the only place that sets empty_size is btrfs_realloc_node, which
3076 * is not called recursively on allocations
3077 */
3078 if (empty_size || root->ref_cows) {
593060d7 3079 if (!(data & BTRFS_BLOCK_GROUP_METADATA)) {
6324fbf3 3080 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
0ef3e66b
CM		 3081 2 * 1024 * 1024,
3082 BTRFS_BLOCK_GROUP_METADATA |
3083 (info->metadata_alloc_profile &
3084 info->avail_metadata_alloc_bits), 0);
6324fbf3
CM		 3085 }
3086 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
0ef3e66b 3087 num_bytes + 2 * 1024 * 1024, data, 0);
6324fbf3 3088 }
0b86a832 3089
db94535d
CM		 3090 WARN_ON(num_bytes < root->sectorsize);
3091 ret = find_free_extent(trans, root, num_bytes, empty_size,
3092 search_start, search_end, hint_byte, ins,
26b8003f
CM		 3093 trans->alloc_exclude_start,
3094 trans->alloc_exclude_nr, data);
3b951516 3095
98d20f67
CM		 3096 if (ret == -ENOSPC && num_bytes > min_alloc_size) {
3097 num_bytes = num_bytes >> 1;
0f9dd46c 3098 num_bytes = num_bytes & ~(root->sectorsize - 1);
98d20f67 3099 num_bytes = max(num_bytes, min_alloc_size);
0ef3e66b
CM		 3100 do_chunk_alloc(trans, root->fs_info->extent_root,
3101 num_bytes, data, 1);
98d20f67
CM		 3102 goto again;
3103 }
ec44a35c 3104 if (ret) {
0f9dd46c
JB		 3105 struct btrfs_space_info *sinfo;
3106
3107 sinfo = __find_space_info(root->fs_info, data);
3108 printk("allocation failed flags %Lu, wanted %Lu\n",
3109 data, num_bytes);
3110 dump_space_info(sinfo, num_bytes);
925baedd 3111 BUG();
925baedd 3112 }
0f9dd46c
JB		 3113
3114 return ret;
e6dcd2dc
CM		 3115}
3116
65b51a00
CM		 3117int btrfs_free_reserved_extent(struct btrfs_root *root, u64 start, u64 len)
3118{
0f9dd46c
JB		 3119 struct btrfs_block_group_cache *cache;
3120
0f9dd46c
JB		 3121 cache = btrfs_lookup_block_group(root->fs_info, start);
3122 if (!cache) {
3123 printk(KERN_ERR "Unable to find block group for %Lu\n", start);
0f9dd46c
JB		 3124 return -ENOSPC;
3125 }
3126 btrfs_add_free_space(cache, start, len);
1a40e23b 3127 update_reserved_extents(root, start, len, 0);
65b51a00
CM		 3128 return 0;
3129}
3130
e6dcd2dc
CM		 3131int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
3132 struct btrfs_root *root,
3133 u64 num_bytes, u64 min_alloc_size,
3134 u64 empty_size, u64 hint_byte,
3135 u64 search_end, struct btrfs_key *ins,
3136 u64 data)
3137{
3138 int ret;
e6dcd2dc
CM		 3139 ret = __btrfs_reserve_extent(trans, root, num_bytes, min_alloc_size,
3140 empty_size, hint_byte, search_end, ins,
3141 data);
e8569813 3142 update_reserved_extents(root, ins->objectid, ins->offset, 1);
e6dcd2dc
CM		 3143 return ret;
3144}
3145
3146static int __btrfs_alloc_reserved_extent(struct btrfs_trans_handle *trans,
31840ae1 3147 struct btrfs_root *root, u64 parent,
e6dcd2dc 3148 u64 root_objectid, u64 ref_generation,
3bb1a1bc 3149 u64 owner, struct btrfs_key *ins)
e6dcd2dc
CM		 3150{
3151 int ret;
3152 int pending_ret;
3153 u64 super_used;
3154 u64 root_used;
3155 u64 num_bytes = ins->offset;
3156 u32 sizes[2];
3157 struct btrfs_fs_info *info = root->fs_info;
3158 struct btrfs_root *extent_root = info->extent_root;
3159 struct btrfs_extent_item *extent_item;
3160 struct btrfs_extent_ref *ref;
3161 struct btrfs_path *path;
3162 struct btrfs_key keys[2];
fec577fb 3163
31840ae1
ZY		 3164 if (parent == 0)
3165 parent = ins->objectid;
3166
58176a96 3167 /* block accounting for super block */
a2135011 3168 spin_lock_irq(&info->delalloc_lock);
db94535d
CM		 3169 super_used = btrfs_super_bytes_used(&info->super_copy);
3170 btrfs_set_super_bytes_used(&info->super_copy, super_used + num_bytes);
a2135011 3171 spin_unlock_irq(&info->delalloc_lock);
26b8003f 3172
58176a96 3173 /* block accounting for root item */
db94535d
CM		 3174 root_used = btrfs_root_used(&root->root_item);
3175 btrfs_set_root_used(&root->root_item, root_used + num_bytes);
58176a96 3176
26b8003f 3177 if (root == extent_root) {
31840ae1
ZY		 3178 struct pending_extent_op *extent_op;
3179
3180 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
3181 BUG_ON(!extent_op);
3182
3183 extent_op->type = PENDING_EXTENT_INSERT;
3184 extent_op->bytenr = ins->objectid;
3185 extent_op->num_bytes = ins->offset;
3186 extent_op->parent = parent;
3187 extent_op->orig_parent = 0;
3188 extent_op->generation = ref_generation;
3189 extent_op->orig_generation = 0;
3190 extent_op->level = (int)owner;
f3465ca4
JB		 3191 INIT_LIST_HEAD(&extent_op->list);
3192 extent_op->del = 0;
31840ae1 3193
25179201 3194 mutex_lock(&root->fs_info->extent_ins_mutex);
1a5bc167
CM		 3195 set_extent_bits(&root->fs_info->extent_ins, ins->objectid,
3196 ins->objectid + ins->offset - 1,
25179201 3197 EXTENT_WRITEBACK, GFP_NOFS);
31840ae1
ZY		 3198 set_state_private(&root->fs_info->extent_ins,
3199 ins->objectid, (unsigned long)extent_op);
25179201 3200 mutex_unlock(&root->fs_info->extent_ins_mutex);
26b8003f
CM		 3201 goto update_block;
3202 }
3203
47e4bb98 3204 memcpy(&keys[0], ins, sizeof(*ins));
47e4bb98
CM		 3205 keys[1].objectid = ins->objectid;
3206 keys[1].type = BTRFS_EXTENT_REF_KEY;
31840ae1 3207 keys[1].offset = parent;
47e4bb98
CM		 3208 sizes[0] = sizeof(*extent_item);
3209 sizes[1] = sizeof(*ref);
7bb86316
CM		 3210
3211 path = btrfs_alloc_path();
3212 BUG_ON(!path);
47e4bb98
CM		 3213
3214 ret = btrfs_insert_empty_items(trans, extent_root, path, keys,
3215 sizes, 2);
ccd467d6 3216 BUG_ON(ret);
0f9dd46c 3217
47e4bb98
CM		 3218 extent_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3219 struct btrfs_extent_item);
3220 btrfs_set_extent_refs(path->nodes[0], extent_item, 1);
3221 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
3222 struct btrfs_extent_ref);
3223
3224 btrfs_set_ref_root(path->nodes[0], ref, root_objectid);
3225 btrfs_set_ref_generation(path->nodes[0], ref, ref_generation);
3226 btrfs_set_ref_objectid(path->nodes[0], ref, owner);
31840ae1 3227 btrfs_set_ref_num_refs(path->nodes[0], ref, 1);
47e4bb98
CM		 3228
3229 btrfs_mark_buffer_dirty(path->nodes[0]);
3230
3231 trans->alloc_exclude_start = 0;
3232 trans->alloc_exclude_nr = 0;
7bb86316 3233 btrfs_free_path(path);
87ef2bb4
CM		 3234 finish_current_insert(trans, extent_root, 0);
3235 pending_ret = del_pending_extents(trans, extent_root, 0);
f510cfec 3236
925baedd
CM		 3237 if (ret)
3238 goto out;
e37c9e69 3239 if (pending_ret) {
925baedd
CM		 3240 ret = pending_ret;
3241 goto out;
e37c9e69 3242 }
26b8003f
CM		 3243
3244update_block:
0b86a832 3245 ret = update_block_group(trans, root, ins->objectid, ins->offset, 1, 0);
f5947066
CM		 3246 if (ret) {
3247 printk("update block group failed for %Lu %Lu\n",
3248 ins->objectid, ins->offset);
3249 BUG();
3250 }
925baedd 3251out:
e6dcd2dc
CM		 3252 return ret;
3253}
3254
3255int btrfs_alloc_reserved_extent(struct btrfs_trans_handle *trans,
31840ae1 3256 struct btrfs_root *root, u64 parent,
e6dcd2dc 3257 u64 root_objectid, u64 ref_generation,
3bb1a1bc 3258 u64 owner, struct btrfs_key *ins)
e6dcd2dc
CM		 3259{
3260 int ret;
1c2308f8
CM		 3261
3262 if (root_objectid == BTRFS_TREE_LOG_OBJECTID)
3263 return 0;
3bb1a1bc
YZ		 3264 ret = __btrfs_alloc_reserved_extent(trans, root, parent, root_objectid,
3265 ref_generation, owner, ins);
e8569813 3266 update_reserved_extents(root, ins->objectid, ins->offset, 0);
e6dcd2dc
CM		 3267 return ret;
3268}
e02119d5
CM		 3269
3270/*
3271 * this is used by the tree logging recovery code. It records that
3272 * an extent has been allocated and makes sure to clear the free
3273 * space cache bits as well
3274 */
3275int btrfs_alloc_logged_extent(struct btrfs_trans_handle *trans,
31840ae1 3276 struct btrfs_root *root, u64 parent,
e02119d5 3277 u64 root_objectid, u64 ref_generation,
3bb1a1bc 3278 u64 owner, struct btrfs_key *ins)
e02119d5
CM		 3279{
3280 int ret;
3281 struct btrfs_block_group_cache *block_group;
3282
e02119d5 3283 block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
ea6a478e 3284 mutex_lock(&block_group->cache_mutex);
e02119d5 3285 cache_block_group(root, block_group);
ea6a478e 3286 mutex_unlock(&block_group->cache_mutex);
e02119d5 3287
ea6a478e
JB		 3288 ret = btrfs_remove_free_space(block_group, ins->objectid,
3289 ins->offset);
0f9dd46c 3290 BUG_ON(ret);
3bb1a1bc
YZ		 3291 ret = __btrfs_alloc_reserved_extent(trans, root, parent, root_objectid,
3292 ref_generation, owner, ins);
e02119d5
CM		 3293 return ret;
3294}
3295
e6dcd2dc
CM		 3296/*
3297 * finds a free extent and does all the dirty work required for allocation
3298 * returns the key for the extent through ins, and a tree buffer for
3299 * the first block of the extent through buf.
3300 *
3301 * returns 0 if everything worked, non-zero otherwise.
3302 */
3303int btrfs_alloc_extent(struct btrfs_trans_handle *trans,
3304 struct btrfs_root *root,
31840ae1 3305 u64 num_bytes, u64 parent, u64 min_alloc_size,
e6dcd2dc 3306 u64 root_objectid, u64 ref_generation,
3bb1a1bc 3307 u64 owner_objectid, u64 empty_size, u64 hint_byte,
e6dcd2dc
CM		 3308 u64 search_end, struct btrfs_key *ins, u64 data)
3309{
3310 int ret;
3311
e6dcd2dc
CM		 3312 ret = __btrfs_reserve_extent(trans, root, num_bytes,
3313 min_alloc_size, empty_size, hint_byte,
3314 search_end, ins, data);
3315 BUG_ON(ret);
d00aff00 3316 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
31840ae1
ZY		 3317 ret = __btrfs_alloc_reserved_extent(trans, root, parent,
3318 root_objectid, ref_generation,
3bb1a1bc 3319 owner_objectid, ins);
d00aff00 3320 BUG_ON(ret);
e6dcd2dc 3321
e8569813
ZY		 3322 } else {
3323 update_reserved_extents(root, ins->objectid, ins->offset, 1);
d00aff00 3324 }
925baedd 3325 return ret;
fec577fb 3326}
65b51a00
CM		 3327
3328struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans,
3329 struct btrfs_root *root,
3330 u64 bytenr, u32 blocksize)
3331{
3332 struct extent_buffer *buf;
3333
3334 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
3335 if (!buf)
3336 return ERR_PTR(-ENOMEM);
3337 btrfs_set_header_generation(buf, trans->transid);
3338 btrfs_tree_lock(buf);
3339 clean_tree_block(trans, root, buf);
3340 btrfs_set_buffer_uptodate(buf);
d0c803c4
CM		 3341 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
3342 set_extent_dirty(&root->dirty_log_pages, buf->start,
3343 buf->start + buf->len - 1, GFP_NOFS);
3344 } else {
3345 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
65b51a00 3346 buf->start + buf->len - 1, GFP_NOFS);
d0c803c4 3347 }
65b51a00
CM		 3348 trans->blocks_used++;
3349 return buf;
3350}
3351
fec577fb
CM		 3352/*
3353 * helper function to allocate a block for a given tree
3354 * returns the tree buffer or NULL.
3355 */
5f39d397 3356struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
7bb86316 3357 struct btrfs_root *root,
31840ae1 3358 u32 blocksize, u64 parent,
7bb86316
CM		 3359 u64 root_objectid,
3360 u64 ref_generation,
7bb86316
CM		 3361 int level,
3362 u64 hint,
5f39d397 3363 u64 empty_size)
fec577fb 3364{
e2fa7227 3365 struct btrfs_key ins;
fec577fb 3366 int ret;
5f39d397 3367 struct extent_buffer *buf;
fec577fb 3368
31840ae1 3369 ret = btrfs_alloc_extent(trans, root, blocksize, parent, blocksize,
3bb1a1bc 3370 root_objectid, ref_generation, level,
31840ae1 3371 empty_size, hint, (u64)-1, &ins, 0);
fec577fb 3372 if (ret) {
54aa1f4d
CM		 3373 BUG_ON(ret > 0);
3374 return ERR_PTR(ret);
fec577fb 3375 }
55c69072 3376
65b51a00 3377 buf = btrfs_init_new_buffer(trans, root, ins.objectid, blocksize);
fec577fb
CM		 3378 return buf;
3379}
a28ec197 3380
e02119d5
CM		 3381int btrfs_drop_leaf_ref(struct btrfs_trans_handle *trans,
3382 struct btrfs_root *root, struct extent_buffer *leaf)
6407bf6d 3383{
7bb86316
CM		 3384 u64 leaf_owner;
3385 u64 leaf_generation;
5f39d397 3386 struct btrfs_key key;
6407bf6d
CM		 3387 struct btrfs_file_extent_item *fi;
3388 int i;
3389 int nritems;
3390 int ret;
3391
5f39d397
CM		 3392 BUG_ON(!btrfs_is_leaf(leaf));
3393 nritems = btrfs_header_nritems(leaf);
7bb86316
CM		 3394 leaf_owner = btrfs_header_owner(leaf);
3395 leaf_generation = btrfs_header_generation(leaf);
3396
6407bf6d 3397 for (i = 0; i < nritems; i++) {
db94535d 3398 u64 disk_bytenr;
e34a5b4f 3399 cond_resched();
5f39d397
CM		 3400
3401 btrfs_item_key_to_cpu(leaf, &key, i);
3402 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
6407bf6d
CM		 3403 continue;
3404 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
5f39d397
CM		 3405 if (btrfs_file_extent_type(leaf, fi) ==
3406 BTRFS_FILE_EXTENT_INLINE)
236454df 3407 continue;
6407bf6d
CM		 3408 /*
3409 * FIXME make sure to insert a trans record that
3410 * repeats the snapshot del on crash
3411 */
db94535d
CM		 3412 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
3413 if (disk_bytenr == 0)
3a686375 3414 continue;
4a096752 3415
925baedd 3416 ret = __btrfs_free_extent(trans, root, disk_bytenr,
7bb86316 3417 btrfs_file_extent_disk_num_bytes(leaf, fi),
31840ae1 3418 leaf->start, leaf_owner, leaf_generation,
3bb1a1bc 3419 key.objectid, 0);
31840ae1 3420 BUG_ON(ret);
2dd3e67b
CM		 3421
3422 atomic_inc(&root->fs_info->throttle_gen);
3423 wake_up(&root->fs_info->transaction_throttle);
3424 cond_resched();
6407bf6d
CM		 3425 }
3426 return 0;
3427}
3428
e02119d5
CM		 3429static int noinline cache_drop_leaf_ref(struct btrfs_trans_handle *trans,
3430 struct btrfs_root *root,
3431 struct btrfs_leaf_ref *ref)
31153d81
YZ		 3432{
3433 int i;
3434 int ret;
3435 struct btrfs_extent_info *info = ref->extents;
3436
31153d81 3437 for (i = 0; i < ref->nritems; i++) {
31840ae1
ZY		 3438 ret = __btrfs_free_extent(trans, root, info->bytenr,
3439 info->num_bytes, ref->bytenr,
3440 ref->owner, ref->generation,
3bb1a1bc 3441 info->objectid, 0);
2dd3e67b
CM		 3442
3443 atomic_inc(&root->fs_info->throttle_gen);
3444 wake_up(&root->fs_info->transaction_throttle);
3445 cond_resched();
3446
31153d81
YZ		 3447 BUG_ON(ret);
3448 info++;
3449 }
31153d81
YZ		 3450
3451 return 0;
3452}
3453
b2950863 3454static int drop_snap_lookup_refcount(struct btrfs_root *root, u64 start, u64 len,
333db94c
CM		 3455 u32 *refs)
3456{
017e5369 3457 int ret;
f87f057b 3458
31840ae1 3459 ret = btrfs_lookup_extent_ref(NULL, root, start, len, refs);
f87f057b
CM		 3460 BUG_ON(ret);
3461
3462#if 0 // some debugging code in case we see problems here
3463 /* if the refs count is one, it won't get increased again. But
3464 * if the ref count is > 1, someone may be decreasing it at
3465 * the same time we are.
3466 */
3467 if (*refs != 1) {
3468 struct extent_buffer *eb = NULL;
3469 eb = btrfs_find_create_tree_block(root, start, len);
3470 if (eb)
3471 btrfs_tree_lock(eb);
3472
3473 mutex_lock(&root->fs_info->alloc_mutex);
3474 ret = lookup_extent_ref(NULL, root, start, len, refs);
3475 BUG_ON(ret);
3476 mutex_unlock(&root->fs_info->alloc_mutex);
3477
3478 if (eb) {
3479 btrfs_tree_unlock(eb);
3480 free_extent_buffer(eb);
3481 }
3482 if (*refs == 1) {
3483 printk("block %llu went down to one during drop_snap\n",
3484 (unsigned long long)start);
3485 }
3486
3487 }
3488#endif
3489
e7a84565 3490 cond_resched();
017e5369 3491 return ret;
333db94c
CM		 3492}
3493
9aca1d51
CM		 3494/*
3495 * helper function for drop_snapshot, this walks down the tree dropping ref
3496 * counts as it goes.
3497 */
98ed5174
CM		 3498static int noinline walk_down_tree(struct btrfs_trans_handle *trans,
3499 struct btrfs_root *root,
3500 struct btrfs_path *path, int *level)
20524f02 3501{
7bb86316
CM		 3502 u64 root_owner;
3503 u64 root_gen;
3504 u64 bytenr;
ca7a79ad 3505 u64 ptr_gen;
5f39d397
CM		 3506 struct extent_buffer *next;
3507 struct extent_buffer *cur;
7bb86316 3508 struct extent_buffer *parent;
31153d81 3509 struct btrfs_leaf_ref *ref;
db94535d 3510 u32 blocksize;
20524f02
CM		 3511 int ret;
3512 u32 refs;
3513
5caf2a00
CM		 3514 WARN_ON(*level < 0);
3515 WARN_ON(*level >= BTRFS_MAX_LEVEL);
333db94c 3516 ret = drop_snap_lookup_refcount(root, path->nodes[*level]->start,
db94535d 3517 path->nodes[*level]->len, &refs);
20524f02
CM		 3518 BUG_ON(ret);
3519 if (refs > 1)
3520 goto out;
e011599b 3521
9aca1d51
CM		 3522 /*
3523 * walk down to the last node level and free all the leaves
3524 */
6407bf6d 3525 while(*level >= 0) {
5caf2a00
CM		 3526 WARN_ON(*level < 0);
3527 WARN_ON(*level >= BTRFS_MAX_LEVEL);
20524f02 3528 cur = path->nodes[*level];
e011599b 3529
5f39d397 3530 if (btrfs_header_level(cur) != *level)
2c90e5d6 3531 WARN_ON(1);
e011599b 3532
7518a238 3533 if (path->slots[*level] >=
5f39d397 3534 btrfs_header_nritems(cur))
20524f02 3535 break;
6407bf6d 3536 if (*level == 0) {
e02119d5 3537 ret = btrfs_drop_leaf_ref(trans, root, cur);
6407bf6d
CM		 3538 BUG_ON(ret);
3539 break;
3540 }
db94535d 3541 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
ca7a79ad 3542 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
db94535d 3543 blocksize = btrfs_level_size(root, *level - 1);
925baedd 3544
333db94c 3545 ret = drop_snap_lookup_refcount(root, bytenr, blocksize, &refs);
6407bf6d
CM		 3546 BUG_ON(ret);
3547 if (refs != 1) {
7bb86316
CM		 3548 parent = path->nodes[*level];
3549 root_owner = btrfs_header_owner(parent);
3550 root_gen = btrfs_header_generation(parent);
20524f02 3551 path->slots[*level]++;
f87f057b 3552
925baedd 3553 ret = __btrfs_free_extent(trans, root, bytenr,
31840ae1 3554 blocksize, parent->start,
3bb1a1bc
YZ		 3555 root_owner, root_gen,
3556 *level - 1, 1);
20524f02 3557 BUG_ON(ret);
18e35e0a
CM		 3558
3559 atomic_inc(&root->fs_info->throttle_gen);
3560 wake_up(&root->fs_info->transaction_throttle);
2dd3e67b 3561 cond_resched();
18e35e0a 3562
20524f02
CM		 3563 continue;
3564 }
f87f057b
CM		 3565 /*
3566 * at this point, we have a single ref, and since the
3567 * only place referencing this extent is a dead root
3568 * the reference count should never go higher.
3569 * So, we don't need to check it again
3570 */
31153d81 3571 if (*level == 1) {
017e5369 3572 ref = btrfs_lookup_leaf_ref(root, bytenr);
1a40e23b
ZY		 3573 if (ref && ref->generation != ptr_gen) {
3574 btrfs_free_leaf_ref(root, ref);
3575 ref = NULL;
3576 }
31153d81 3577 if (ref) {
e02119d5 3578 ret = cache_drop_leaf_ref(trans, root, ref);
31153d81
YZ		 3579 BUG_ON(ret);
3580 btrfs_remove_leaf_ref(root, ref);
bcc63abb 3581 btrfs_free_leaf_ref(root, ref);
31153d81
YZ		 3582 *level = 0;
3583 break;
3584 }
3bb1a1bc 3585 if (printk_ratelimit()) {
37d1aeee
CM		 3586 printk("leaf ref miss for bytenr %llu\n",
3587 (unsigned long long)bytenr);
3bb1a1bc 3588 }
31153d81 3589 }
db94535d 3590 next = btrfs_find_tree_block(root, bytenr, blocksize);
1259ab75 3591 if (!next || !btrfs_buffer_uptodate(next, ptr_gen)) {
5f39d397 3592 free_extent_buffer(next);
333db94c 3593
ca7a79ad
CM		 3594 next = read_tree_block(root, bytenr, blocksize,
3595 ptr_gen);
e7a84565 3596 cond_resched();
f87f057b
CM		 3597#if 0
3598 /*
3599 * this is a debugging check and can go away
3600 * the ref should never go all the way down to 1
3601 * at this point
3602 */
e6dcd2dc
CM		 3603 ret = lookup_extent_ref(NULL, root, bytenr, blocksize,
3604 &refs);
e9d0b13b 3605 BUG_ON(ret);
f87f057b
CM		 3606 WARN_ON(refs != 1);
3607#endif
e9d0b13b 3608 }
5caf2a00 3609 WARN_ON(*level <= 0);
83e15a28 3610 if (path->nodes[*level-1])
5f39d397 3611 free_extent_buffer(path->nodes[*level-1]);
20524f02 3612 path->nodes[*level-1] = next;
5f39d397 3613 *level = btrfs_header_level(next);
20524f02 3614 path->slots[*level] = 0;
2dd3e67b 3615 cond_resched();
20524f02
CM		 3616 }
3617out:
5caf2a00
CM		 3618 WARN_ON(*level < 0);
3619 WARN_ON(*level >= BTRFS_MAX_LEVEL);
7bb86316
CM		 3620
3621 if (path->nodes[*level] == root->node) {
7bb86316 3622 parent = path->nodes[*level];
31153d81 3623 bytenr = path->nodes[*level]->start;
7bb86316
CM		 3624 } else {
3625 parent = path->nodes[*level + 1];
31153d81 3626 bytenr = btrfs_node_blockptr(parent, path->slots[*level + 1]);
7bb86316
CM		 3627 }
3628
31153d81
YZ		 3629 blocksize = btrfs_level_size(root, *level);
3630 root_owner = btrfs_header_owner(parent);
7bb86316 3631 root_gen = btrfs_header_generation(parent);
31153d81
YZ		 3632
3633 ret = __btrfs_free_extent(trans, root, bytenr, blocksize,
31840ae1 3634 parent->start, root_owner, root_gen,
3bb1a1bc 3635 *level, 1);
5f39d397 3636 free_extent_buffer(path->nodes[*level]);
20524f02
CM		 3637 path->nodes[*level] = NULL;
3638 *level += 1;
3639 BUG_ON(ret);
f87f057b 3640
e7a84565 3641 cond_resched();
20524f02
CM		 3642 return 0;
3643}
3644
f82d02d9
YZ		 3645/*
3646 * helper function for drop_subtree, this function is similar to
3647 * walk_down_tree. The main difference is that it checks reference
3648 * counts while tree blocks are locked.
3649 */
3650static int noinline walk_down_subtree(struct btrfs_trans_handle *trans,
3651 struct btrfs_root *root,
3652 struct btrfs_path *path, int *level)
3653{
3654 struct extent_buffer *next;
3655 struct extent_buffer *cur;
3656 struct extent_buffer *parent;
3657 u64 bytenr;
3658 u64 ptr_gen;
3659 u32 blocksize;
3660 u32 refs;
3661 int ret;
3662
3663 cur = path->nodes[*level];
3664 ret = btrfs_lookup_extent_ref(trans, root, cur->start, cur->len,
3665 &refs);
3666 BUG_ON(ret);
3667 if (refs > 1)
3668 goto out;
3669
3670 while (*level >= 0) {
3671 cur = path->nodes[*level];
3672 if (*level == 0) {
3673 ret = btrfs_drop_leaf_ref(trans, root, cur);
3674 BUG_ON(ret);
3675 clean_tree_block(trans, root, cur);
3676 break;
3677 }
3678 if (path->slots[*level] >= btrfs_header_nritems(cur)) {
3679 clean_tree_block(trans, root, cur);
3680 break;
3681 }
3682
3683 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
3684 blocksize = btrfs_level_size(root, *level - 1);
3685 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
3686
3687 next = read_tree_block(root, bytenr, blocksize, ptr_gen);
3688 btrfs_tree_lock(next);
3689
3690 ret = btrfs_lookup_extent_ref(trans, root, bytenr, blocksize,
3691 &refs);
3692 BUG_ON(ret);
3693 if (refs > 1) {
3694 parent = path->nodes[*level];
3695 ret = btrfs_free_extent(trans, root, bytenr,
3696 blocksize, parent->start,
3697 btrfs_header_owner(parent),
3698 btrfs_header_generation(parent),
3699 *level - 1, 1);
3700 BUG_ON(ret);
3701 path->slots[*level]++;
3702 btrfs_tree_unlock(next);
3703 free_extent_buffer(next);
3704 continue;
3705 }
3706
3707 *level = btrfs_header_level(next);
3708 path->nodes[*level] = next;
3709 path->slots[*level] = 0;
3710 path->locks[*level] = 1;
3711 cond_resched();
3712 }
3713out:
3714 parent = path->nodes[*level + 1];
3715 bytenr = path->nodes[*level]->start;
3716 blocksize = path->nodes[*level]->len;
3717
3718 ret = btrfs_free_extent(trans, root, bytenr, blocksize,
3719 parent->start, btrfs_header_owner(parent),
3720 btrfs_header_generation(parent), *level, 1);
3721 BUG_ON(ret);
3722
3723 if (path->locks[*level]) {
3724 btrfs_tree_unlock(path->nodes[*level]);
3725 path->locks[*level] = 0;
3726 }
3727 free_extent_buffer(path->nodes[*level]);
3728 path->nodes[*level] = NULL;
3729 *level += 1;
3730 cond_resched();
3731 return 0;
3732}
3733
9aca1d51
CM		 3734/*
3735 * helper for dropping snapshots. This walks back up the tree in the path
3736 * to find the first node higher up where we haven't yet gone through
3737 * all the slots
3738 */
98ed5174
CM		 3739static int noinline walk_up_tree(struct btrfs_trans_handle *trans,
3740 struct btrfs_root *root,
f82d02d9
YZ		 3741 struct btrfs_path *path,
3742 int *level, int max_level)
20524f02 3743{
7bb86316
CM		 3744 u64 root_owner;
3745 u64 root_gen;
3746 struct btrfs_root_item *root_item = &root->root_item;
20524f02
CM		 3747 int i;
3748 int slot;
3749 int ret;
9f3a7427 3750
f82d02d9 3751 for (i = *level; i < max_level && path->nodes[i]; i++) {
20524f02 3752 slot = path->slots[i];
5f39d397
CM		 3753 if (slot < btrfs_header_nritems(path->nodes[i]) - 1) {
3754 struct extent_buffer *node;
3755 struct btrfs_disk_key disk_key;
3756 node = path->nodes[i];
20524f02
CM		 3757 path->slots[i]++;
3758 *level = i;
9f3a7427 3759 WARN_ON(*level == 0);
5f39d397 3760 btrfs_node_key(node, &disk_key, path->slots[i]);
9f3a7427 3761 memcpy(&root_item->drop_progress,
5f39d397 3762 &disk_key, sizeof(disk_key));
9f3a7427 3763 root_item->drop_level = i;
20524f02
CM		 3764 return 0;
3765 } else {
31840ae1
ZY		 3766 struct extent_buffer *parent;
3767 if (path->nodes[*level] == root->node)
3768 parent = path->nodes[*level];
3769 else
3770 parent = path->nodes[*level + 1];
3771
3772 root_owner = btrfs_header_owner(parent);
3773 root_gen = btrfs_header_generation(parent);
f82d02d9
YZ		 3774
3775 clean_tree_block(trans, root, path->nodes[*level]);
e089f05c 3776 ret = btrfs_free_extent(trans, root,
db94535d 3777 path->nodes[*level]->start,
7bb86316 3778 path->nodes[*level]->len,
3bb1a1bc
YZ		 3779 parent->start, root_owner,
3780 root_gen, *level, 1);
6407bf6d 3781 BUG_ON(ret);
f82d02d9
YZ		 3782 if (path->locks[*level]) {
3783 btrfs_tree_unlock(path->nodes[*level]);
3784 path->locks[*level] = 0;
3785 }
5f39d397 3786 free_extent_buffer(path->nodes[*level]);
83e15a28 3787 path->nodes[*level] = NULL;
20524f02 3788 *level = i + 1;
20524f02
CM		 3789 }
3790 }
3791 return 1;
3792}
3793
9aca1d51
CM		 3794/*
3795 * drop the reference count on the tree rooted at 'snap'. This traverses
3796 * the tree freeing any blocks that have a ref count of zero after being
3797 * decremented.
3798 */
e089f05c 3799int btrfs_drop_snapshot(struct btrfs_trans_handle *trans, struct btrfs_root
9f3a7427 3800 *root)
20524f02 3801{
3768f368 3802 int ret = 0;
9aca1d51 3803 int wret;
20524f02 3804 int level;
5caf2a00 3805 struct btrfs_path *path;
20524f02
CM		 3806 int i;
3807 int orig_level;
9f3a7427 3808 struct btrfs_root_item *root_item = &root->root_item;
20524f02 3809
a2135011 3810 WARN_ON(!mutex_is_locked(&root->fs_info->drop_mutex));
5caf2a00
CM		 3811 path = btrfs_alloc_path();
3812 BUG_ON(!path);
20524f02 3813
5f39d397 3814 level = btrfs_header_level(root->node);
20524f02 3815 orig_level = level;
9f3a7427
CM		 3816 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
3817 path->nodes[level] = root->node;
f510cfec 3818 extent_buffer_get(root->node);
9f3a7427
CM		 3819 path->slots[level] = 0;
3820 } else {
3821 struct btrfs_key key;
5f39d397
CM		 3822 struct btrfs_disk_key found_key;
3823 struct extent_buffer *node;
6702ed49 3824
9f3a7427 3825 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
6702ed49
CM		 3826 level = root_item->drop_level;
3827 path->lowest_level = level;
9f3a7427 3828 wret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6702ed49 3829 if (wret < 0) {
9f3a7427
CM		 3830 ret = wret;
3831 goto out;
3832 }
5f39d397
CM		 3833 node = path->nodes[level];
3834 btrfs_node_key(node, &found_key, path->slots[level]);
3835 WARN_ON(memcmp(&found_key, &root_item->drop_progress,
3836 sizeof(found_key)));
7d9eb12c
CM		 3837 /*
3838 * unlock our path, this is safe because only this
3839 * function is allowed to delete this snapshot
3840 */
925baedd
CM		 3841 for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
3842 if (path->nodes[i] && path->locks[i]) {
3843 path->locks[i] = 0;
3844 btrfs_tree_unlock(path->nodes[i]);
3845 }
3846 }
9f3a7427 3847 }
20524f02 3848 while(1) {
5caf2a00 3849 wret = walk_down_tree(trans, root, path, &level);
9aca1d51 3850 if (wret > 0)
20524f02 3851 break;
9aca1d51
CM		 3852 if (wret < 0)
3853 ret = wret;
3854
f82d02d9
YZ		 3855 wret = walk_up_tree(trans, root, path, &level,
3856 BTRFS_MAX_LEVEL);
9aca1d51 3857 if (wret > 0)
20524f02 3858 break;
9aca1d51
CM		 3859 if (wret < 0)
3860 ret = wret;
e7a84565
CM		 3861 if (trans->transaction->in_commit) {
3862 ret = -EAGAIN;
3863 break;
3864 }
18e35e0a 3865 atomic_inc(&root->fs_info->throttle_gen);
017e5369 3866 wake_up(&root->fs_info->transaction_throttle);
20524f02 3867 }
83e15a28 3868 for (i = 0; i <= orig_level; i++) {
5caf2a00 3869 if (path->nodes[i]) {
5f39d397 3870 free_extent_buffer(path->nodes[i]);
0f82731f 3871 path->nodes[i] = NULL;
83e15a28 3872 }
20524f02 3873 }
9f3a7427 3874out:
5caf2a00 3875 btrfs_free_path(path);
9aca1d51 3876 return ret;
20524f02 3877}
9078a3e1 3878
f82d02d9
YZ		 3879int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
3880 struct btrfs_root *root,
3881 struct extent_buffer *node,
3882 struct extent_buffer *parent)
3883{
3884 struct btrfs_path *path;
3885 int level;
3886 int parent_level;
3887 int ret = 0;
3888 int wret;
3889
3890 path = btrfs_alloc_path();
3891 BUG_ON(!path);
3892
3893 BUG_ON(!btrfs_tree_locked(parent));
3894 parent_level = btrfs_header_level(parent);
3895 extent_buffer_get(parent);
3896 path->nodes[parent_level] = parent;
3897 path->slots[parent_level] = btrfs_header_nritems(parent);
3898
3899 BUG_ON(!btrfs_tree_locked(node));
3900 level = btrfs_header_level(node);
3901 extent_buffer_get(node);
3902 path->nodes[level] = node;
3903 path->slots[level] = 0;
3904
3905 while (1) {
3906 wret = walk_down_subtree(trans, root, path, &level);
3907 if (wret < 0)
3908 ret = wret;
3909 if (wret != 0)
3910 break;
3911
3912 wret = walk_up_tree(trans, root, path, &level, parent_level);
3913 if (wret < 0)
3914 ret = wret;
3915 if (wret != 0)
3916 break;
3917 }
3918
3919 btrfs_free_path(path);
3920 return ret;
3921}
3922
8e7bf94f
CM		 3923static unsigned long calc_ra(unsigned long start, unsigned long last,
3924 unsigned long nr)
3925{
3926 return min(last, start + nr - 1);
3927}
3928
98ed5174
CM		 3929static int noinline relocate_inode_pages(struct inode *inode, u64 start,
3930 u64 len)
edbd8d4e
CM		 3931{
3932 u64 page_start;
3933 u64 page_end;
1a40e23b 3934 unsigned long first_index;
edbd8d4e 3935 unsigned long last_index;
edbd8d4e
CM		 3936 unsigned long i;
3937 struct page *page;
d1310b2e 3938 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4313b399 3939 struct file_ra_state *ra;
3eaa2885 3940 struct btrfs_ordered_extent *ordered;
1a40e23b
ZY		 3941 unsigned int total_read = 0;
3942 unsigned int total_dirty = 0;
3943 int ret = 0;
4313b399
CM		 3944
3945 ra = kzalloc(sizeof(*ra), GFP_NOFS);
edbd8d4e
CM		 3946
3947 mutex_lock(&inode->i_mutex);
1a40e23b 3948 first_index = start >> PAGE_CACHE_SHIFT;
edbd8d4e
CM		 3949 last_index = (start + len - 1) >> PAGE_CACHE_SHIFT;
3950
1a40e23b
ZY		 3951 /* make sure the dirty trick played by the caller work */
3952 ret = invalidate_inode_pages2_range(inode->i_mapping,
3953 first_index, last_index);
3954 if (ret)
3955 goto out_unlock;
8e7bf94f 3956
4313b399 3957 file_ra_state_init(ra, inode->i_mapping);
edbd8d4e 3958
1a40e23b
ZY		 3959 for (i = first_index ; i <= last_index; i++) {
3960 if (total_read % ra->ra_pages == 0) {
8e7bf94f 3961 btrfs_force_ra(inode->i_mapping, ra, NULL, i,
1a40e23b 3962 calc_ra(i, last_index, ra->ra_pages));
8e7bf94f
CM		 3963 }
3964 total_read++;
3eaa2885
CM		 3965again:
3966 if (((u64)i << PAGE_CACHE_SHIFT) > i_size_read(inode))
1a40e23b 3967 BUG_ON(1);
edbd8d4e 3968 page = grab_cache_page(inode->i_mapping, i);
a061fc8d 3969 if (!page) {
1a40e23b 3970 ret = -ENOMEM;
edbd8d4e 3971 goto out_unlock;
a061fc8d 3972 }
edbd8d4e
CM		 3973 if (!PageUptodate(page)) {
3974 btrfs_readpage(NULL, page);
3975 lock_page(page);
3976 if (!PageUptodate(page)) {
3977 unlock_page(page);
3978 page_cache_release(page);
1a40e23b 3979 ret = -EIO;
edbd8d4e
CM		 3980 goto out_unlock;
3981 }
3982 }
ec44a35c 3983 wait_on_page_writeback(page);
3eaa2885 3984
edbd8d4e
CM		 3985 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
3986 page_end = page_start + PAGE_CACHE_SIZE - 1;
d1310b2e 3987 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
edbd8d4e 3988
3eaa2885
CM		 3989 ordered = btrfs_lookup_ordered_extent(inode, page_start);
3990 if (ordered) {
3991 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
3992 unlock_page(page);
3993 page_cache_release(page);
3994 btrfs_start_ordered_extent(inode, ordered, 1);
3995 btrfs_put_ordered_extent(ordered);
3996 goto again;
3997 }
3998 set_page_extent_mapped(page);
3999
ea8c2819 4000 btrfs_set_extent_delalloc(inode, page_start, page_end);
1a40e23b
ZY		 4001 if (i == first_index)
4002 set_extent_bits(io_tree, page_start, page_end,
4003 EXTENT_BOUNDARY, GFP_NOFS);
4004
a061fc8d 4005 set_page_dirty(page);
1a40e23b 4006 total_dirty++;
edbd8d4e 4007
d1310b2e 4008 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
edbd8d4e
CM		 4009 unlock_page(page);
4010 page_cache_release(page);
4011 }
4012
4013out_unlock:
ec44a35c 4014 kfree(ra);
edbd8d4e 4015 mutex_unlock(&inode->i_mutex);
1a40e23b
ZY		 4016 balance_dirty_pages_ratelimited_nr(inode->i_mapping, total_dirty);
4017 return ret;
edbd8d4e
CM		 4018}
4019
1a40e23b
ZY		 4020static int noinline relocate_data_extent(struct inode *reloc_inode,
4021 struct btrfs_key *extent_key,
4022 u64 offset)
4023{
4024 struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
4025 struct extent_map_tree *em_tree = &BTRFS_I(reloc_inode)->extent_tree;
4026 struct extent_map *em;
6643558d
YZ		 4027 u64 start = extent_key->objectid - offset;
4028 u64 end = start + extent_key->offset - 1;
bf4ef679 4029
1a40e23b
ZY		 4030 em = alloc_extent_map(GFP_NOFS);
4031 BUG_ON(!em || IS_ERR(em));
bf4ef679 4032
6643558d 4033 em->start = start;
1a40e23b 4034 em->len = extent_key->offset;
c8b97818 4035 em->block_len = extent_key->offset;
1a40e23b
ZY		 4036 em->block_start = extent_key->objectid;
4037 em->bdev = root->fs_info->fs_devices->latest_bdev;
4038 set_bit(EXTENT_FLAG_PINNED, &em->flags);
4039
4040 /* setup extent map to cheat btrfs_readpage */
6643558d 4041 lock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
1a40e23b
ZY		 4042 while (1) {
4043 int ret;
4044 spin_lock(&em_tree->lock);
4045 ret = add_extent_mapping(em_tree, em);
4046 spin_unlock(&em_tree->lock);
4047 if (ret != -EEXIST) {
4048 free_extent_map(em);
bf4ef679
CM		 4049 break;
4050 }
6643558d 4051 btrfs_drop_extent_cache(reloc_inode, start, end, 0);
bf4ef679 4052 }
6643558d 4053 unlock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
bf4ef679 4054
6643558d 4055 return relocate_inode_pages(reloc_inode, start, extent_key->offset);
1a40e23b 4056}
edbd8d4e 4057
1a40e23b
ZY		 4058struct btrfs_ref_path {
4059 u64 extent_start;
4060 u64 nodes[BTRFS_MAX_LEVEL];
4061 u64 root_objectid;
4062 u64 root_generation;
4063 u64 owner_objectid;
1a40e23b
ZY		 4064 u32 num_refs;
4065 int lowest_level;
4066 int current_level;
f82d02d9
YZ		 4067 int shared_level;
4068
4069 struct btrfs_key node_keys[BTRFS_MAX_LEVEL];
4070 u64 new_nodes[BTRFS_MAX_LEVEL];
1a40e23b 4071};
7d9eb12c 4072
1a40e23b 4073struct disk_extent {
c8b97818 4074 u64 ram_bytes;
1a40e23b
ZY		 4075 u64 disk_bytenr;
4076 u64 disk_num_bytes;
4077 u64 offset;
4078 u64 num_bytes;
c8b97818
CM		 4079 u8 compression;
4080 u8 encryption;
4081 u16 other_encoding;
1a40e23b 4082};
4313b399 4083
1a40e23b
ZY		 4084static int is_cowonly_root(u64 root_objectid)
4085{
4086 if (root_objectid == BTRFS_ROOT_TREE_OBJECTID ||
4087 root_objectid == BTRFS_EXTENT_TREE_OBJECTID ||
4088 root_objectid == BTRFS_CHUNK_TREE_OBJECTID ||
4089 root_objectid == BTRFS_DEV_TREE_OBJECTID ||
0403e47e
YZ		 4090 root_objectid == BTRFS_TREE_LOG_OBJECTID ||
4091 root_objectid == BTRFS_CSUM_TREE_OBJECTID)
1a40e23b
ZY		 4092 return 1;
4093 return 0;
4094}
edbd8d4e 4095
1a40e23b
ZY		 4096static int noinline __next_ref_path(struct btrfs_trans_handle *trans,
4097 struct btrfs_root *extent_root,
4098 struct btrfs_ref_path *ref_path,
4099 int first_time)
4100{
4101 struct extent_buffer *leaf;
4102 struct btrfs_path *path;
4103 struct btrfs_extent_ref *ref;
4104 struct btrfs_key key;
4105 struct btrfs_key found_key;
4106 u64 bytenr;
4107 u32 nritems;
4108 int level;
4109 int ret = 1;
edbd8d4e 4110
1a40e23b
ZY		 4111 path = btrfs_alloc_path();
4112 if (!path)
4113 return -ENOMEM;
bf4ef679 4114
1a40e23b
ZY		 4115 if (first_time) {
4116 ref_path->lowest_level = -1;
4117 ref_path->current_level = -1;
f82d02d9 4118 ref_path->shared_level = -1;
1a40e23b
ZY		 4119 goto walk_up;
4120 }
4121walk_down:
4122 level = ref_path->current_level - 1;
4123 while (level >= -1) {
4124 u64 parent;
4125 if (level < ref_path->lowest_level)
4126 break;
bf4ef679 4127
1a40e23b
ZY		 4128 if (level >= 0) {
4129 bytenr = ref_path->nodes[level];
4130 } else {
4131 bytenr = ref_path->extent_start;
4132 }
4133 BUG_ON(bytenr == 0);
bf4ef679 4134
1a40e23b
ZY		 4135 parent = ref_path->nodes[level + 1];
4136 ref_path->nodes[level + 1] = 0;
4137 ref_path->current_level = level;
4138 BUG_ON(parent == 0);
0ef3e66b 4139
1a40e23b
ZY		 4140 key.objectid = bytenr;
4141 key.offset = parent + 1;
4142 key.type = BTRFS_EXTENT_REF_KEY;
edbd8d4e 4143
1a40e23b
ZY		 4144 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
4145 if (ret < 0)
edbd8d4e 4146 goto out;
1a40e23b 4147 BUG_ON(ret == 0);
7d9eb12c 4148
1a40e23b
ZY		 4149 leaf = path->nodes[0];
4150 nritems = btrfs_header_nritems(leaf);
4151 if (path->slots[0] >= nritems) {
4152 ret = btrfs_next_leaf(extent_root, path);
4153 if (ret < 0)
4154 goto out;
4155 if (ret > 0)
4156 goto next;
4157 leaf = path->nodes[0];
4158 }
0ef3e66b 4159
1a40e23b
ZY		 4160 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4161 if (found_key.objectid == bytenr &&
f82d02d9
YZ		 4162 found_key.type == BTRFS_EXTENT_REF_KEY) {
4163 if (level < ref_path->shared_level)
4164 ref_path->shared_level = level;
1a40e23b 4165 goto found;
f82d02d9 4166 }
1a40e23b
ZY		 4167next:
4168 level--;
4169 btrfs_release_path(extent_root, path);
d899e052 4170 cond_resched();
1a40e23b
ZY		 4171 }
4172 /* reached lowest level */
4173 ret = 1;
4174 goto out;
4175walk_up:
4176 level = ref_path->current_level;
4177 while (level < BTRFS_MAX_LEVEL - 1) {
4178 u64 ref_objectid;
4179 if (level >= 0) {
4180 bytenr = ref_path->nodes[level];
4181 } else {
4182 bytenr = ref_path->extent_start;
4183 }
4184 BUG_ON(bytenr == 0);
edbd8d4e 4185
1a40e23b
ZY		 4186 key.objectid = bytenr;
4187 key.offset = 0;
4188 key.type = BTRFS_EXTENT_REF_KEY;
edbd8d4e 4189
1a40e23b
ZY		 4190 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
4191 if (ret < 0)
4192 goto out;
edbd8d4e 4193
1a40e23b
ZY		 4194 leaf = path->nodes[0];
4195 nritems = btrfs_header_nritems(leaf);
4196 if (path->slots[0] >= nritems) {
4197 ret = btrfs_next_leaf(extent_root, path);
4198 if (ret < 0)
4199 goto out;
4200 if (ret > 0) {
4201 /* the extent was freed by someone */
4202 if (ref_path->lowest_level == level)
4203 goto out;
4204 btrfs_release_path(extent_root, path);
4205 goto walk_down;
4206 }
4207 leaf = path->nodes[0];
4208 }
edbd8d4e 4209
1a40e23b
ZY		 4210 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4211 if (found_key.objectid != bytenr ||
4212 found_key.type != BTRFS_EXTENT_REF_KEY) {
4213 /* the extent was freed by someone */
4214 if (ref_path->lowest_level == level) {
4215 ret = 1;
4216 goto out;
4217 }
4218 btrfs_release_path(extent_root, path);
4219 goto walk_down;
4220 }
4221found:
4222 ref = btrfs_item_ptr(leaf, path->slots[0],
4223 struct btrfs_extent_ref);
4224 ref_objectid = btrfs_ref_objectid(leaf, ref);
4225 if (ref_objectid < BTRFS_FIRST_FREE_OBJECTID) {
4226 if (first_time) {
4227 level = (int)ref_objectid;
4228 BUG_ON(level >= BTRFS_MAX_LEVEL);
4229 ref_path->lowest_level = level;
4230 ref_path->current_level = level;
4231 ref_path->nodes[level] = bytenr;
4232 } else {
4233 WARN_ON(ref_objectid != level);
4234 }
4235 } else {
4236 WARN_ON(level != -1);
4237 }
4238 first_time = 0;
bf4ef679 4239
1a40e23b
ZY		 4240 if (ref_path->lowest_level == level) {
4241 ref_path->owner_objectid = ref_objectid;
1a40e23b
ZY		 4242 ref_path->num_refs = btrfs_ref_num_refs(leaf, ref);
4243 }
bf4ef679 4244
7d9eb12c 4245 /*
1a40e23b
ZY		 4246 * the block is tree root or the block isn't in reference
4247 * counted tree.
7d9eb12c 4248 */
1a40e23b
ZY		 4249 if (found_key.objectid == found_key.offset ||
4250 is_cowonly_root(btrfs_ref_root(leaf, ref))) {
4251 ref_path->root_objectid = btrfs_ref_root(leaf, ref);
4252 ref_path->root_generation =
4253 btrfs_ref_generation(leaf, ref);
4254 if (level < 0) {
4255 /* special reference from the tree log */
4256 ref_path->nodes[0] = found_key.offset;
4257 ref_path->current_level = 0;
4258 }
4259 ret = 0;
4260 goto out;
4261 }
7d9eb12c 4262
1a40e23b
ZY		 4263 level++;
4264 BUG_ON(ref_path->nodes[level] != 0);
4265 ref_path->nodes[level] = found_key.offset;
4266 ref_path->current_level = level;
bf4ef679 4267
1a40e23b
ZY		 4268 /*
4269 * the reference was created in the running transaction,
4270 * no need to continue walking up.
4271 */
4272 if (btrfs_ref_generation(leaf, ref) == trans->transid) {
4273 ref_path->root_objectid = btrfs_ref_root(leaf, ref);
4274 ref_path->root_generation =
4275 btrfs_ref_generation(leaf, ref);
4276 ret = 0;
4277 goto out;
7d9eb12c
CM		 4278 }
4279
1a40e23b 4280 btrfs_release_path(extent_root, path);
d899e052 4281 cond_resched();
7d9eb12c 4282 }
1a40e23b
ZY		 4283 /* reached max tree level, but no tree root found. */
4284 BUG();
edbd8d4e 4285out:
1a40e23b
ZY		 4286 btrfs_free_path(path);
4287 return ret;
edbd8d4e
CM		 4288}
4289
1a40e23b
ZY		 4290static int btrfs_first_ref_path(struct btrfs_trans_handle *trans,
4291 struct btrfs_root *extent_root,
4292 struct btrfs_ref_path *ref_path,
4293 u64 extent_start)
a061fc8d 4294{
1a40e23b
ZY		 4295 memset(ref_path, 0, sizeof(*ref_path));
4296 ref_path->extent_start = extent_start;
a061fc8d 4297
1a40e23b 4298 return __next_ref_path(trans, extent_root, ref_path, 1);
a061fc8d
CM		 4299}
4300
1a40e23b
ZY		 4301static int btrfs_next_ref_path(struct btrfs_trans_handle *trans,
4302 struct btrfs_root *extent_root,
4303 struct btrfs_ref_path *ref_path)
edbd8d4e 4304{
1a40e23b
ZY		 4305 return __next_ref_path(trans, extent_root, ref_path, 0);
4306}
4307
4308static int noinline get_new_locations(struct inode *reloc_inode,
4309 struct btrfs_key *extent_key,
4310 u64 offset, int no_fragment,
4311 struct disk_extent **extents,
4312 int *nr_extents)
4313{
4314 struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
4315 struct btrfs_path *path;
4316 struct btrfs_file_extent_item *fi;
edbd8d4e 4317 struct extent_buffer *leaf;
1a40e23b
ZY		 4318 struct disk_extent *exts = *extents;
4319 struct btrfs_key found_key;
4320 u64 cur_pos;
4321 u64 last_byte;
edbd8d4e 4322 u32 nritems;
1a40e23b
ZY		 4323 int nr = 0;
4324 int max = *nr_extents;
4325 int ret;
edbd8d4e 4326
1a40e23b
ZY		 4327 WARN_ON(!no_fragment && *extents);
4328 if (!exts) {
4329 max = 1;
4330 exts = kmalloc(sizeof(*exts) * max, GFP_NOFS);
4331 if (!exts)
4332 return -ENOMEM;
a061fc8d 4333 }
edbd8d4e 4334
1a40e23b
ZY		 4335 path = btrfs_alloc_path();
4336 BUG_ON(!path);
edbd8d4e 4337
1a40e23b
ZY		 4338 cur_pos = extent_key->objectid - offset;
4339 last_byte = extent_key->objectid + extent_key->offset;
4340 ret = btrfs_lookup_file_extent(NULL, root, path, reloc_inode->i_ino,
4341 cur_pos, 0);
4342 if (ret < 0)
4343 goto out;
4344 if (ret > 0) {
4345 ret = -ENOENT;
4346 goto out;
4347 }
edbd8d4e 4348
1a40e23b 4349 while (1) {
edbd8d4e
CM		 4350 leaf = path->nodes[0];
4351 nritems = btrfs_header_nritems(leaf);
1a40e23b
ZY		 4352 if (path->slots[0] >= nritems) {
4353 ret = btrfs_next_leaf(root, path);
a061fc8d
CM		 4354 if (ret < 0)
4355 goto out;
1a40e23b
ZY		 4356 if (ret > 0)
4357 break;
bf4ef679 4358 leaf = path->nodes[0];
a061fc8d 4359 }
edbd8d4e
CM		 4360
4361 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1a40e23b
ZY		 4362 if (found_key.offset != cur_pos ||
4363 found_key.type != BTRFS_EXTENT_DATA_KEY ||
4364 found_key.objectid != reloc_inode->i_ino)
edbd8d4e
CM		 4365 break;
4366
1a40e23b
ZY		 4367 fi = btrfs_item_ptr(leaf, path->slots[0],
4368 struct btrfs_file_extent_item);
4369 if (btrfs_file_extent_type(leaf, fi) !=
4370 BTRFS_FILE_EXTENT_REG ||
4371 btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
edbd8d4e 4372 break;
1a40e23b
ZY		 4373
4374 if (nr == max) {
4375 struct disk_extent *old = exts;
4376 max *= 2;
4377 exts = kzalloc(sizeof(*exts) * max, GFP_NOFS);
4378 memcpy(exts, old, sizeof(*exts) * nr);
4379 if (old != *extents)
4380 kfree(old);
a061fc8d 4381 }
edbd8d4e 4382
1a40e23b
ZY		 4383 exts[nr].disk_bytenr =
4384 btrfs_file_extent_disk_bytenr(leaf, fi);
4385 exts[nr].disk_num_bytes =
4386 btrfs_file_extent_disk_num_bytes(leaf, fi);
4387 exts[nr].offset = btrfs_file_extent_offset(leaf, fi);
4388 exts[nr].num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
c8b97818
CM		 4389 exts[nr].ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
4390 exts[nr].compression = btrfs_file_extent_compression(leaf, fi);
4391 exts[nr].encryption = btrfs_file_extent_encryption(leaf, fi);
4392 exts[nr].other_encoding = btrfs_file_extent_other_encoding(leaf,
4393 fi);
d899e052
YZ		 4394 BUG_ON(exts[nr].offset > 0);
4395 BUG_ON(exts[nr].compression || exts[nr].encryption);
4396 BUG_ON(exts[nr].num_bytes != exts[nr].disk_num_bytes);
edbd8d4e 4397
1a40e23b
ZY		 4398 cur_pos += exts[nr].num_bytes;
4399 nr++;
4400
4401 if (cur_pos + offset >= last_byte)
4402 break;
4403
4404 if (no_fragment) {
4405 ret = 1;
edbd8d4e 4406 goto out;
1a40e23b
ZY		 4407 }
4408 path->slots[0]++;
4409 }
4410
4411 WARN_ON(cur_pos + offset > last_byte);
4412 if (cur_pos + offset < last_byte) {
4413 ret = -ENOENT;
4414 goto out;
edbd8d4e
CM		 4415 }
4416 ret = 0;
4417out:
1a40e23b
ZY		 4418 btrfs_free_path(path);
4419 if (ret) {
4420 if (exts != *extents)
4421 kfree(exts);
4422 } else {
4423 *extents = exts;
4424 *nr_extents = nr;
4425 }
4426 return ret;
4427}
4428
4429static int noinline replace_one_extent(struct btrfs_trans_handle *trans,
4430 struct btrfs_root *root,
4431 struct btrfs_path *path,
4432 struct btrfs_key *extent_key,
4433 struct btrfs_key *leaf_key,
4434 struct btrfs_ref_path *ref_path,
4435 struct disk_extent *new_extents,
4436 int nr_extents)
4437{
4438 struct extent_buffer *leaf;
4439 struct btrfs_file_extent_item *fi;
4440 struct inode *inode = NULL;
4441 struct btrfs_key key;
4442 u64 lock_start = 0;
4443 u64 lock_end = 0;
4444 u64 num_bytes;
4445 u64 ext_offset;
4446 u64 first_pos;
4447 u32 nritems;
3bb1a1bc 4448 int nr_scaned = 0;
1a40e23b 4449 int extent_locked = 0;
d899e052 4450 int extent_type;
1a40e23b
ZY		 4451 int ret;
4452
3bb1a1bc
YZ		 4453 memcpy(&key, leaf_key, sizeof(key));
4454 first_pos = INT_LIMIT(loff_t) - extent_key->offset;
1a40e23b 4455 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
3bb1a1bc
YZ		 4456 if (key.objectid < ref_path->owner_objectid ||
4457 (key.objectid == ref_path->owner_objectid &&
4458 key.type < BTRFS_EXTENT_DATA_KEY)) {
4459 key.objectid = ref_path->owner_objectid;
4460 key.type = BTRFS_EXTENT_DATA_KEY;
4461 key.offset = 0;
4462 }
1a40e23b
ZY		 4463 }
4464
4465 while (1) {
4466 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
4467 if (ret < 0)
4468 goto out;
4469
4470 leaf = path->nodes[0];
4471 nritems = btrfs_header_nritems(leaf);
4472next:
4473 if (extent_locked && ret > 0) {
4474 /*
4475 * the file extent item was modified by someone
4476 * before the extent got locked.
4477 */
1a40e23b
ZY		 4478 unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
4479 lock_end, GFP_NOFS);
4480 extent_locked = 0;
4481 }
4482
4483 if (path->slots[0] >= nritems) {
3bb1a1bc 4484 if (++nr_scaned > 2)
1a40e23b
ZY		 4485 break;
4486
4487 BUG_ON(extent_locked);
4488 ret = btrfs_next_leaf(root, path);
4489 if (ret < 0)
4490 goto out;
4491 if (ret > 0)
4492 break;
4493 leaf = path->nodes[0];
4494 nritems = btrfs_header_nritems(leaf);
4495 }
4496
4497 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4498
4499 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
4500 if ((key.objectid > ref_path->owner_objectid) ||
4501 (key.objectid == ref_path->owner_objectid &&
4502 key.type > BTRFS_EXTENT_DATA_KEY) ||
4503 (key.offset >= first_pos + extent_key->offset))
4504 break;
4505 }
4506
4507 if (inode && key.objectid != inode->i_ino) {
4508 BUG_ON(extent_locked);
4509 btrfs_release_path(root, path);
4510 mutex_unlock(&inode->i_mutex);
4511 iput(inode);
4512 inode = NULL;
4513 continue;
4514 }
4515
4516 if (key.type != BTRFS_EXTENT_DATA_KEY) {
4517 path->slots[0]++;
4518 ret = 1;
4519 goto next;
4520 }
4521 fi = btrfs_item_ptr(leaf, path->slots[0],
4522 struct btrfs_file_extent_item);
d899e052
YZ		 4523 extent_type = btrfs_file_extent_type(leaf, fi);
4524 if ((extent_type != BTRFS_FILE_EXTENT_REG &&
4525 extent_type != BTRFS_FILE_EXTENT_PREALLOC) ||
1a40e23b
ZY		 4526 (btrfs_file_extent_disk_bytenr(leaf, fi) !=
4527 extent_key->objectid)) {
4528 path->slots[0]++;
4529 ret = 1;
4530 goto next;
4531 }
4532
4533 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
4534 ext_offset = btrfs_file_extent_offset(leaf, fi);
4535
4536 if (first_pos > key.offset - ext_offset)
4537 first_pos = key.offset - ext_offset;
4538
4539 if (!extent_locked) {
4540 lock_start = key.offset;
4541 lock_end = lock_start + num_bytes - 1;
4542 } else {
6643558d
YZ		 4543 if (lock_start > key.offset ||
4544 lock_end + 1 < key.offset + num_bytes) {
4545 unlock_extent(&BTRFS_I(inode)->io_tree,
4546 lock_start, lock_end, GFP_NOFS);
4547 extent_locked = 0;
4548 }
1a40e23b
ZY		 4549 }
4550
4551 if (!inode) {
4552 btrfs_release_path(root, path);
4553
4554 inode = btrfs_iget_locked(root->fs_info->sb,
4555 key.objectid, root);
4556 if (inode->i_state & I_NEW) {
4557 BTRFS_I(inode)->root = root;
4558 BTRFS_I(inode)->location.objectid =
4559 key.objectid;
4560 BTRFS_I(inode)->location.type =
4561 BTRFS_INODE_ITEM_KEY;
4562 BTRFS_I(inode)->location.offset = 0;
4563 btrfs_read_locked_inode(inode);
4564 unlock_new_inode(inode);
4565 }
4566 /*
4567 * some code call btrfs_commit_transaction while
4568 * holding the i_mutex, so we can't use mutex_lock
4569 * here.
4570 */
4571 if (is_bad_inode(inode) ||
4572 !mutex_trylock(&inode->i_mutex)) {
4573 iput(inode);
4574 inode = NULL;
4575 key.offset = (u64)-1;
4576 goto skip;
4577 }
4578 }
4579
4580 if (!extent_locked) {
4581 struct btrfs_ordered_extent *ordered;
4582
4583 btrfs_release_path(root, path);
4584
4585 lock_extent(&BTRFS_I(inode)->io_tree, lock_start,
4586 lock_end, GFP_NOFS);
4587 ordered = btrfs_lookup_first_ordered_extent(inode,
4588 lock_end);
4589 if (ordered &&
4590 ordered->file_offset <= lock_end &&
4591 ordered->file_offset + ordered->len > lock_start) {
4592 unlock_extent(&BTRFS_I(inode)->io_tree,
4593 lock_start, lock_end, GFP_NOFS);
4594 btrfs_start_ordered_extent(inode, ordered, 1);
4595 btrfs_put_ordered_extent(ordered);
4596 key.offset += num_bytes;
4597 goto skip;
4598 }
4599 if (ordered)
4600 btrfs_put_ordered_extent(ordered);
4601
1a40e23b
ZY		 4602 extent_locked = 1;
4603 continue;
4604 }
4605
4606 if (nr_extents == 1) {
4607 /* update extent pointer in place */
1a40e23b
ZY		 4608 btrfs_set_file_extent_disk_bytenr(leaf, fi,
4609 new_extents[0].disk_bytenr);
4610 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
4611 new_extents[0].disk_num_bytes);
1a40e23b
ZY		 4612 btrfs_mark_buffer_dirty(leaf);
4613
4614 btrfs_drop_extent_cache(inode, key.offset,
4615 key.offset + num_bytes - 1, 0);
4616
4617 ret = btrfs_inc_extent_ref(trans, root,
4618 new_extents[0].disk_bytenr,
4619 new_extents[0].disk_num_bytes,
4620 leaf->start,
4621 root->root_key.objectid,
4622 trans->transid,
3bb1a1bc 4623 key.objectid);
1a40e23b
ZY		 4624 BUG_ON(ret);
4625
4626 ret = btrfs_free_extent(trans, root,
4627 extent_key->objectid,
4628 extent_key->offset,
4629 leaf->start,
4630 btrfs_header_owner(leaf),
4631 btrfs_header_generation(leaf),
3bb1a1bc 4632 key.objectid, 0);
1a40e23b
ZY		 4633 BUG_ON(ret);
4634
4635 btrfs_release_path(root, path);
4636 key.offset += num_bytes;
4637 } else {
d899e052
YZ		 4638 BUG_ON(1);
4639#if 0
1a40e23b
ZY		 4640 u64 alloc_hint;
4641 u64 extent_len;
4642 int i;
4643 /*
4644 * drop old extent pointer at first, then insert the
4645 * new pointers one bye one
4646 */
4647 btrfs_release_path(root, path);
4648 ret = btrfs_drop_extents(trans, root, inode, key.offset,
4649 key.offset + num_bytes,
4650 key.offset, &alloc_hint);
4651 BUG_ON(ret);
4652
4653 for (i = 0; i < nr_extents; i++) {
4654 if (ext_offset >= new_extents[i].num_bytes) {
4655 ext_offset -= new_extents[i].num_bytes;
4656 continue;
4657 }
4658 extent_len = min(new_extents[i].num_bytes -
4659 ext_offset, num_bytes);
4660
4661 ret = btrfs_insert_empty_item(trans, root,
4662 path, &key,
4663 sizeof(*fi));
4664 BUG_ON(ret);
4665
4666 leaf = path->nodes[0];
4667 fi = btrfs_item_ptr(leaf, path->slots[0],
4668 struct btrfs_file_extent_item);
4669 btrfs_set_file_extent_generation(leaf, fi,
4670 trans->transid);
4671 btrfs_set_file_extent_type(leaf, fi,
4672 BTRFS_FILE_EXTENT_REG);
4673 btrfs_set_file_extent_disk_bytenr(leaf, fi,
4674 new_extents[i].disk_bytenr);
4675 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
4676 new_extents[i].disk_num_bytes);
c8b97818
CM		 4677 btrfs_set_file_extent_ram_bytes(leaf, fi,
4678 new_extents[i].ram_bytes);
4679
4680 btrfs_set_file_extent_compression(leaf, fi,
4681 new_extents[i].compression);
4682 btrfs_set_file_extent_encryption(leaf, fi,
4683 new_extents[i].encryption);
4684 btrfs_set_file_extent_other_encoding(leaf, fi,
4685 new_extents[i].other_encoding);
4686
1a40e23b
ZY		 4687 btrfs_set_file_extent_num_bytes(leaf, fi,
4688 extent_len);
4689 ext_offset += new_extents[i].offset;
4690 btrfs_set_file_extent_offset(leaf, fi,
4691 ext_offset);
4692 btrfs_mark_buffer_dirty(leaf);
4693
4694 btrfs_drop_extent_cache(inode, key.offset,
4695 key.offset + extent_len - 1, 0);
4696
4697 ret = btrfs_inc_extent_ref(trans, root,
4698 new_extents[i].disk_bytenr,
4699 new_extents[i].disk_num_bytes,
4700 leaf->start,
4701 root->root_key.objectid,
3bb1a1bc 4702 trans->transid, key.objectid);
1a40e23b
ZY		 4703 BUG_ON(ret);
4704 btrfs_release_path(root, path);
4705
a76a3cd4 4706 inode_add_bytes(inode, extent_len);
1a40e23b
ZY		 4707
4708 ext_offset = 0;
4709 num_bytes -= extent_len;
4710 key.offset += extent_len;
4711
4712 if (num_bytes == 0)
4713 break;
4714 }
4715 BUG_ON(i >= nr_extents);
d899e052 4716#endif
1a40e23b
ZY		 4717 }
4718
4719 if (extent_locked) {
1a40e23b
ZY		 4720 unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
4721 lock_end, GFP_NOFS);
4722 extent_locked = 0;
4723 }
4724skip:
4725 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS &&
4726 key.offset >= first_pos + extent_key->offset)
4727 break;
4728
4729 cond_resched();
4730 }
4731 ret = 0;
4732out:
4733 btrfs_release_path(root, path);
4734 if (inode) {
4735 mutex_unlock(&inode->i_mutex);
4736 if (extent_locked) {
1a40e23b
ZY		 4737 unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
4738 lock_end, GFP_NOFS);
4739 }
4740 iput(inode);
4741 }
4742 return ret;
4743}
4744
1a40e23b
ZY		 4745int btrfs_reloc_tree_cache_ref(struct btrfs_trans_handle *trans,
4746 struct btrfs_root *root,
4747 struct extent_buffer *buf, u64 orig_start)
4748{
4749 int level;
4750 int ret;
4751
4752 BUG_ON(btrfs_header_generation(buf) != trans->transid);
4753 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
4754
4755 level = btrfs_header_level(buf);
4756 if (level == 0) {
4757 struct btrfs_leaf_ref *ref;
4758 struct btrfs_leaf_ref *orig_ref;
4759
4760 orig_ref = btrfs_lookup_leaf_ref(root, orig_start);
4761 if (!orig_ref)
4762 return -ENOENT;
4763
4764 ref = btrfs_alloc_leaf_ref(root, orig_ref->nritems);
4765 if (!ref) {
4766 btrfs_free_leaf_ref(root, orig_ref);
4767 return -ENOMEM;
4768 }
4769
4770 ref->nritems = orig_ref->nritems;
4771 memcpy(ref->extents, orig_ref->extents,
4772 sizeof(ref->extents[0]) * ref->nritems);
4773
4774 btrfs_free_leaf_ref(root, orig_ref);
4775
4776 ref->root_gen = trans->transid;
4777 ref->bytenr = buf->start;
4778 ref->owner = btrfs_header_owner(buf);
4779 ref->generation = btrfs_header_generation(buf);
4780 ret = btrfs_add_leaf_ref(root, ref, 0);
4781 WARN_ON(ret);
4782 btrfs_free_leaf_ref(root, ref);
4783 }
4784 return 0;
4785}
4786
4787static int noinline invalidate_extent_cache(struct btrfs_root *root,
4788 struct extent_buffer *leaf,
4789 struct btrfs_block_group_cache *group,
4790 struct btrfs_root *target_root)
4791{
4792 struct btrfs_key key;
4793 struct inode *inode = NULL;
4794 struct btrfs_file_extent_item *fi;
4795 u64 num_bytes;
4796 u64 skip_objectid = 0;
4797 u32 nritems;
4798 u32 i;
4799
4800 nritems = btrfs_header_nritems(leaf);
4801 for (i = 0; i < nritems; i++) {
4802 btrfs_item_key_to_cpu(leaf, &key, i);
4803 if (key.objectid == skip_objectid ||
4804 key.type != BTRFS_EXTENT_DATA_KEY)
4805 continue;
4806 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
4807 if (btrfs_file_extent_type(leaf, fi) ==
4808 BTRFS_FILE_EXTENT_INLINE)
4809 continue;
4810 if (btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
4811 continue;
4812 if (!inode || inode->i_ino != key.objectid) {
4813 iput(inode);
4814 inode = btrfs_ilookup(target_root->fs_info->sb,
4815 key.objectid, target_root, 1);
4816 }
4817 if (!inode) {
4818 skip_objectid = key.objectid;
4819 continue;
4820 }
4821 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
4822
4823 lock_extent(&BTRFS_I(inode)->io_tree, key.offset,
4824 key.offset + num_bytes - 1, GFP_NOFS);
1a40e23b
ZY		 4825 btrfs_drop_extent_cache(inode, key.offset,
4826 key.offset + num_bytes - 1, 1);
1a40e23b
ZY		 4827 unlock_extent(&BTRFS_I(inode)->io_tree, key.offset,
4828 key.offset + num_bytes - 1, GFP_NOFS);
4829 cond_resched();
4830 }
4831 iput(inode);
4832 return 0;
4833}
4834
4835static int noinline replace_extents_in_leaf(struct btrfs_trans_handle *trans,
4836 struct btrfs_root *root,
4837 struct extent_buffer *leaf,
4838 struct btrfs_block_group_cache *group,
4839 struct inode *reloc_inode)
4840{
4841 struct btrfs_key key;
4842 struct btrfs_key extent_key;
4843 struct btrfs_file_extent_item *fi;
4844 struct btrfs_leaf_ref *ref;
4845 struct disk_extent *new_extent;
4846 u64 bytenr;
4847 u64 num_bytes;
4848 u32 nritems;
4849 u32 i;
4850 int ext_index;
4851 int nr_extent;
4852 int ret;
4853
4854 new_extent = kmalloc(sizeof(*new_extent), GFP_NOFS);
4855 BUG_ON(!new_extent);
4856
4857 ref = btrfs_lookup_leaf_ref(root, leaf->start);
4858 BUG_ON(!ref);
4859
4860 ext_index = -1;
4861 nritems = btrfs_header_nritems(leaf);
4862 for (i = 0; i < nritems; i++) {
4863 btrfs_item_key_to_cpu(leaf, &key, i);
4864 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
4865 continue;
4866 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
4867 if (btrfs_file_extent_type(leaf, fi) ==
4868 BTRFS_FILE_EXTENT_INLINE)
4869 continue;
4870 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
4871 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
4872 if (bytenr == 0)
4873 continue;
4874
4875 ext_index++;
4876 if (bytenr >= group->key.objectid + group->key.offset ||
4877 bytenr + num_bytes <= group->key.objectid)
4878 continue;
4879
4880 extent_key.objectid = bytenr;
4881 extent_key.offset = num_bytes;
4882 extent_key.type = BTRFS_EXTENT_ITEM_KEY;
4883 nr_extent = 1;
4884 ret = get_new_locations(reloc_inode, &extent_key,
4885 group->key.objectid, 1,
4886 &new_extent, &nr_extent);
4887 if (ret > 0)
4888 continue;
4889 BUG_ON(ret < 0);
4890
4891 BUG_ON(ref->extents[ext_index].bytenr != bytenr);
4892 BUG_ON(ref->extents[ext_index].num_bytes != num_bytes);
4893 ref->extents[ext_index].bytenr = new_extent->disk_bytenr;
4894 ref->extents[ext_index].num_bytes = new_extent->disk_num_bytes;
4895
1a40e23b
ZY		 4896 btrfs_set_file_extent_disk_bytenr(leaf, fi,
4897 new_extent->disk_bytenr);
4898 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
4899 new_extent->disk_num_bytes);
1a40e23b
ZY		 4900 btrfs_mark_buffer_dirty(leaf);
4901
4902 ret = btrfs_inc_extent_ref(trans, root,
4903 new_extent->disk_bytenr,
4904 new_extent->disk_num_bytes,
4905 leaf->start,
4906 root->root_key.objectid,
3bb1a1bc 4907 trans->transid, key.objectid);
1a40e23b
ZY		 4908 BUG_ON(ret);
4909 ret = btrfs_free_extent(trans, root,
4910 bytenr, num_bytes, leaf->start,
4911 btrfs_header_owner(leaf),
4912 btrfs_header_generation(leaf),
3bb1a1bc 4913 key.objectid, 0);
1a40e23b
ZY		 4914 BUG_ON(ret);
4915 cond_resched();
4916 }
4917 kfree(new_extent);
4918 BUG_ON(ext_index + 1 != ref->nritems);
4919 btrfs_free_leaf_ref(root, ref);
4920 return 0;
4921}
4922
f82d02d9
YZ		 4923int btrfs_free_reloc_root(struct btrfs_trans_handle *trans,
4924 struct btrfs_root *root)
1a40e23b
ZY		 4925{
4926 struct btrfs_root *reloc_root;
f82d02d9 4927 int ret;
1a40e23b
ZY		 4928
4929 if (root->reloc_root) {
4930 reloc_root = root->reloc_root;
4931 root->reloc_root = NULL;
4932 list_add(&reloc_root->dead_list,
4933 &root->fs_info->dead_reloc_roots);
f82d02d9
YZ		 4934
4935 btrfs_set_root_bytenr(&reloc_root->root_item,
4936 reloc_root->node->start);
4937 btrfs_set_root_level(&root->root_item,
4938 btrfs_header_level(reloc_root->node));
4939 memset(&reloc_root->root_item.drop_progress, 0,
4940 sizeof(struct btrfs_disk_key));
4941 reloc_root->root_item.drop_level = 0;
4942
4943 ret = btrfs_update_root(trans, root->fs_info->tree_root,
4944 &reloc_root->root_key,
4945 &reloc_root->root_item);
4946 BUG_ON(ret);
1a40e23b
ZY		 4947 }
4948 return 0;
4949}
4950
4951int btrfs_drop_dead_reloc_roots(struct btrfs_root *root)
4952{
4953 struct btrfs_trans_handle *trans;
4954 struct btrfs_root *reloc_root;
4955 struct btrfs_root *prev_root = NULL;
4956 struct list_head dead_roots;
4957 int ret;
4958 unsigned long nr;
4959
4960 INIT_LIST_HEAD(&dead_roots);
4961 list_splice_init(&root->fs_info->dead_reloc_roots, &dead_roots);
4962
4963 while (!list_empty(&dead_roots)) {
4964 reloc_root = list_entry(dead_roots.prev,
4965 struct btrfs_root, dead_list);
4966 list_del_init(&reloc_root->dead_list);
4967
4968 BUG_ON(reloc_root->commit_root != NULL);
4969 while (1) {
4970 trans = btrfs_join_transaction(root, 1);
4971 BUG_ON(!trans);
4972
4973 mutex_lock(&root->fs_info->drop_mutex);
4974 ret = btrfs_drop_snapshot(trans, reloc_root);
4975 if (ret != -EAGAIN)
4976 break;
4977 mutex_unlock(&root->fs_info->drop_mutex);
4978
4979 nr = trans->blocks_used;
4980 ret = btrfs_end_transaction(trans, root);
4981 BUG_ON(ret);
4982 btrfs_btree_balance_dirty(root, nr);
4983 }
4984
4985 free_extent_buffer(reloc_root->node);
4986
4987 ret = btrfs_del_root(trans, root->fs_info->tree_root,
4988 &reloc_root->root_key);
4989 BUG_ON(ret);
4990 mutex_unlock(&root->fs_info->drop_mutex);
4991
4992 nr = trans->blocks_used;
4993 ret = btrfs_end_transaction(trans, root);
4994 BUG_ON(ret);
4995 btrfs_btree_balance_dirty(root, nr);
4996
4997 kfree(prev_root);
4998 prev_root = reloc_root;
4999 }
5000 if (prev_root) {
5001 btrfs_remove_leaf_refs(prev_root, (u64)-1, 0);
5002 kfree(prev_root);
5003 }
5004 return 0;
5005}
5006
5007int btrfs_add_dead_reloc_root(struct btrfs_root *root)
5008{
5009 list_add(&root->dead_list, &root->fs_info->dead_reloc_roots);
5010 return 0;
5011}
5012
5013int btrfs_cleanup_reloc_trees(struct btrfs_root *root)
5014{
5015 struct btrfs_root *reloc_root;
5016 struct btrfs_trans_handle *trans;
5017 struct btrfs_key location;
5018 int found;
5019 int ret;
5020
5021 mutex_lock(&root->fs_info->tree_reloc_mutex);
5022 ret = btrfs_find_dead_roots(root, BTRFS_TREE_RELOC_OBJECTID, NULL);
5023 BUG_ON(ret);
5024 found = !list_empty(&root->fs_info->dead_reloc_roots);
5025 mutex_unlock(&root->fs_info->tree_reloc_mutex);
5026
5027 if (found) {
5028 trans = btrfs_start_transaction(root, 1);
5029 BUG_ON(!trans);
5030 ret = btrfs_commit_transaction(trans, root);
5031 BUG_ON(ret);
5032 }
5033
5034 location.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
5035 location.offset = (u64)-1;
5036 location.type = BTRFS_ROOT_ITEM_KEY;
5037
5038 reloc_root = btrfs_read_fs_root_no_name(root->fs_info, &location);
5039 BUG_ON(!reloc_root);
5040 btrfs_orphan_cleanup(reloc_root);
5041 return 0;
5042}
5043
5044static int noinline init_reloc_tree(struct btrfs_trans_handle *trans,
5045 struct btrfs_root *root)
5046{
5047 struct btrfs_root *reloc_root;
5048 struct extent_buffer *eb;
5049 struct btrfs_root_item *root_item;
5050 struct btrfs_key root_key;
5051 int ret;
5052
5053 BUG_ON(!root->ref_cows);
5054 if (root->reloc_root)
5055 return 0;
5056
5057 root_item = kmalloc(sizeof(*root_item), GFP_NOFS);
5058 BUG_ON(!root_item);
5059
5060 ret = btrfs_copy_root(trans, root, root->commit_root,
5061 &eb, BTRFS_TREE_RELOC_OBJECTID);
5062 BUG_ON(ret);
5063
5064 root_key.objectid = BTRFS_TREE_RELOC_OBJECTID;
5065 root_key.offset = root->root_key.objectid;
5066 root_key.type = BTRFS_ROOT_ITEM_KEY;
5067
5068 memcpy(root_item, &root->root_item, sizeof(root_item));
5069 btrfs_set_root_refs(root_item, 0);
5070 btrfs_set_root_bytenr(root_item, eb->start);
5071 btrfs_set_root_level(root_item, btrfs_header_level(eb));
84234f3a 5072 btrfs_set_root_generation(root_item, trans->transid);
1a40e23b
ZY		 5073
5074 btrfs_tree_unlock(eb);
5075 free_extent_buffer(eb);
5076
5077 ret = btrfs_insert_root(trans, root->fs_info->tree_root,
5078 &root_key, root_item);
5079 BUG_ON(ret);
5080 kfree(root_item);
5081
5082 reloc_root = btrfs_read_fs_root_no_radix(root->fs_info->tree_root,
5083 &root_key);
5084 BUG_ON(!reloc_root);
5085 reloc_root->last_trans = trans->transid;
5086 reloc_root->commit_root = NULL;
5087 reloc_root->ref_tree = &root->fs_info->reloc_ref_tree;
5088
5089 root->reloc_root = reloc_root;
5090 return 0;
5091}
5092
5093/*
5094 * Core function of space balance.
5095 *
5096 * The idea is using reloc trees to relocate tree blocks in reference
f82d02d9
YZ		 5097 * counted roots. There is one reloc tree for each subvol, and all
5098 * reloc trees share same root key objectid. Reloc trees are snapshots
5099 * of the latest committed roots of subvols (root->commit_root).
5100 *
5101 * To relocate a tree block referenced by a subvol, there are two steps.
5102 * COW the block through subvol's reloc tree, then update block pointer
5103 * in the subvol to point to the new block. Since all reloc trees share
5104 * same root key objectid, doing special handing for tree blocks owned
5105 * by them is easy. Once a tree block has been COWed in one reloc tree,
5106 * we can use the resulting new block directly when the same block is
5107 * required to COW again through other reloc trees. By this way, relocated
5108 * tree blocks are shared between reloc trees, so they are also shared
5109 * between subvols.
1a40e23b
ZY		 5110 */
5111static int noinline relocate_one_path(struct btrfs_trans_handle *trans,
5112 struct btrfs_root *root,
5113 struct btrfs_path *path,
5114 struct btrfs_key *first_key,
5115 struct btrfs_ref_path *ref_path,
5116 struct btrfs_block_group_cache *group,
5117 struct inode *reloc_inode)
5118{
5119 struct btrfs_root *reloc_root;
5120 struct extent_buffer *eb = NULL;
5121 struct btrfs_key *keys;
5122 u64 *nodes;
5123 int level;
f82d02d9 5124 int shared_level;
1a40e23b 5125 int lowest_level = 0;
1a40e23b
ZY		 5126 int ret;
5127
5128 if (ref_path->owner_objectid < BTRFS_FIRST_FREE_OBJECTID)
5129 lowest_level = ref_path->owner_objectid;
5130
f82d02d9 5131 if (!root->ref_cows) {
1a40e23b
ZY		 5132 path->lowest_level = lowest_level;
5133 ret = btrfs_search_slot(trans, root, first_key, path, 0, 1);
5134 BUG_ON(ret < 0);
5135 path->lowest_level = 0;
5136 btrfs_release_path(root, path);
5137 return 0;
5138 }
5139
1a40e23b
ZY		 5140 mutex_lock(&root->fs_info->tree_reloc_mutex);
5141 ret = init_reloc_tree(trans, root);
5142 BUG_ON(ret);
5143 reloc_root = root->reloc_root;
5144
f82d02d9
YZ		 5145 shared_level = ref_path->shared_level;
5146 ref_path->shared_level = BTRFS_MAX_LEVEL - 1;
1a40e23b 5147
f82d02d9
YZ		 5148 keys = ref_path->node_keys;
5149 nodes = ref_path->new_nodes;
5150 memset(&keys[shared_level + 1], 0,
5151 sizeof(*keys) * (BTRFS_MAX_LEVEL - shared_level - 1));
5152 memset(&nodes[shared_level + 1], 0,
5153 sizeof(*nodes) * (BTRFS_MAX_LEVEL - shared_level - 1));
1a40e23b 5154
f82d02d9
YZ		 5155 if (nodes[lowest_level] == 0) {
5156 path->lowest_level = lowest_level;
5157 ret = btrfs_search_slot(trans, reloc_root, first_key, path,
5158 0, 1);
5159 BUG_ON(ret);
5160 for (level = lowest_level; level < BTRFS_MAX_LEVEL; level++) {
5161 eb = path->nodes[level];
5162 if (!eb || eb == reloc_root->node)
5163 break;
5164 nodes[level] = eb->start;
5165 if (level == 0)
5166 btrfs_item_key_to_cpu(eb, &keys[level], 0);
5167 else
5168 btrfs_node_key_to_cpu(eb, &keys[level], 0);
5169 }
2b82032c
YZ		 5170 if (nodes[0] &&
5171 ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
f82d02d9
YZ		 5172 eb = path->nodes[0];
5173 ret = replace_extents_in_leaf(trans, reloc_root, eb,
5174 group, reloc_inode);
5175 BUG_ON(ret);
5176 }
5177 btrfs_release_path(reloc_root, path);
5178 } else {
1a40e23b 5179 ret = btrfs_merge_path(trans, reloc_root, keys, nodes,
f82d02d9 5180 lowest_level);
1a40e23b
ZY		 5181 BUG_ON(ret);
5182 }
5183
1a40e23b
ZY		 5184 /*
5185 * replace tree blocks in the fs tree with tree blocks in
5186 * the reloc tree.
5187 */
5188 ret = btrfs_merge_path(trans, root, keys, nodes, lowest_level);
5189 BUG_ON(ret < 0);
5190
5191 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
f82d02d9
YZ		 5192 ret = btrfs_search_slot(trans, reloc_root, first_key, path,
5193 0, 0);
5194 BUG_ON(ret);
5195 extent_buffer_get(path->nodes[0]);
5196 eb = path->nodes[0];
5197 btrfs_release_path(reloc_root, path);
1a40e23b
ZY		 5198 ret = invalidate_extent_cache(reloc_root, eb, group, root);
5199 BUG_ON(ret);
5200 free_extent_buffer(eb);
5201 }
1a40e23b 5202
f82d02d9 5203 mutex_unlock(&root->fs_info->tree_reloc_mutex);
1a40e23b 5204 path->lowest_level = 0;
1a40e23b
ZY		 5205 return 0;
5206}
5207
5208static int noinline relocate_tree_block(struct btrfs_trans_handle *trans,
5209 struct btrfs_root *root,
5210 struct btrfs_path *path,
5211 struct btrfs_key *first_key,
5212 struct btrfs_ref_path *ref_path)
5213{
5214 int ret;
1a40e23b
ZY		 5215
5216 ret = relocate_one_path(trans, root, path, first_key,
5217 ref_path, NULL, NULL);
5218 BUG_ON(ret);
5219
5220 if (root == root->fs_info->extent_root)
5221 btrfs_extent_post_op(trans, root);
1a40e23b
ZY		 5222
5223 return 0;
5224}
5225
5226static int noinline del_extent_zero(struct btrfs_trans_handle *trans,
5227 struct btrfs_root *extent_root,
5228 struct btrfs_path *path,
5229 struct btrfs_key *extent_key)
5230{
5231 int ret;
5232
1a40e23b
ZY		 5233 ret = btrfs_search_slot(trans, extent_root, extent_key, path, -1, 1);
5234 if (ret)
5235 goto out;
5236 ret = btrfs_del_item(trans, extent_root, path);
5237out:
5238 btrfs_release_path(extent_root, path);
1a40e23b
ZY		 5239 return ret;
5240}
5241
5242static struct btrfs_root noinline *read_ref_root(struct btrfs_fs_info *fs_info,
5243 struct btrfs_ref_path *ref_path)
5244{
5245 struct btrfs_key root_key;
5246
5247 root_key.objectid = ref_path->root_objectid;
5248 root_key.type = BTRFS_ROOT_ITEM_KEY;
5249 if (is_cowonly_root(ref_path->root_objectid))
5250 root_key.offset = 0;
5251 else
5252 root_key.offset = (u64)-1;
5253
5254 return btrfs_read_fs_root_no_name(fs_info, &root_key);
5255}
5256
5257static int noinline relocate_one_extent(struct btrfs_root *extent_root,
5258 struct btrfs_path *path,
5259 struct btrfs_key *extent_key,
5260 struct btrfs_block_group_cache *group,
5261 struct inode *reloc_inode, int pass)
5262{
5263 struct btrfs_trans_handle *trans;
5264 struct btrfs_root *found_root;
5265 struct btrfs_ref_path *ref_path = NULL;
5266 struct disk_extent *new_extents = NULL;
5267 int nr_extents = 0;
5268 int loops;
5269 int ret;
5270 int level;
5271 struct btrfs_key first_key;
5272 u64 prev_block = 0;
5273
1a40e23b
ZY		 5274
5275 trans = btrfs_start_transaction(extent_root, 1);
5276 BUG_ON(!trans);
5277
5278 if (extent_key->objectid == 0) {
5279 ret = del_extent_zero(trans, extent_root, path, extent_key);
5280 goto out;
5281 }
5282
5283 ref_path = kmalloc(sizeof(*ref_path), GFP_NOFS);
5284 if (!ref_path) {
5285 ret = -ENOMEM;
5286 goto out;
5287 }
5288
5289 for (loops = 0; ; loops++) {
5290 if (loops == 0) {
5291 ret = btrfs_first_ref_path(trans, extent_root, ref_path,
5292 extent_key->objectid);
5293 } else {
5294 ret = btrfs_next_ref_path(trans, extent_root, ref_path);
5295 }
5296 if (ret < 0)
5297 goto out;
5298 if (ret > 0)
5299 break;
5300
5301 if (ref_path->root_objectid == BTRFS_TREE_LOG_OBJECTID ||
5302 ref_path->root_objectid == BTRFS_TREE_RELOC_OBJECTID)
5303 continue;
5304
5305 found_root = read_ref_root(extent_root->fs_info, ref_path);
5306 BUG_ON(!found_root);
5307 /*
5308 * for reference counted tree, only process reference paths
5309 * rooted at the latest committed root.
5310 */
5311 if (found_root->ref_cows &&
5312 ref_path->root_generation != found_root->root_key.offset)
5313 continue;
5314
5315 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
5316 if (pass == 0) {
5317 /*
5318 * copy data extents to new locations
5319 */
5320 u64 group_start = group->key.objectid;
5321 ret = relocate_data_extent(reloc_inode,
5322 extent_key,
5323 group_start);
5324 if (ret < 0)
5325 goto out;
5326 break;
5327 }
5328 level = 0;
5329 } else {
5330 level = ref_path->owner_objectid;
5331 }
5332
5333 if (prev_block != ref_path->nodes[level]) {
5334 struct extent_buffer *eb;
5335 u64 block_start = ref_path->nodes[level];
5336 u64 block_size = btrfs_level_size(found_root, level);
5337
5338 eb = read_tree_block(found_root, block_start,
5339 block_size, 0);
5340 btrfs_tree_lock(eb);
5341 BUG_ON(level != btrfs_header_level(eb));
5342
5343 if (level == 0)
5344 btrfs_item_key_to_cpu(eb, &first_key, 0);
5345 else
5346 btrfs_node_key_to_cpu(eb, &first_key, 0);
5347
5348 btrfs_tree_unlock(eb);
5349 free_extent_buffer(eb);
5350 prev_block = block_start;
5351 }
5352
5353 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID &&
5354 pass >= 2) {
5355 /*
5356 * use fallback method to process the remaining
5357 * references.
5358 */
5359 if (!new_extents) {
5360 u64 group_start = group->key.objectid;
d899e052
YZ		 5361 new_extents = kmalloc(sizeof(*new_extents),
5362 GFP_NOFS);
5363 nr_extents = 1;
1a40e23b
ZY		 5364 ret = get_new_locations(reloc_inode,
5365 extent_key,
d899e052 5366 group_start, 1,
1a40e23b
ZY		 5367 &new_extents,
5368 &nr_extents);
d899e052 5369 if (ret)
1a40e23b
ZY		 5370 goto out;
5371 }
5372 btrfs_record_root_in_trans(found_root);
5373 ret = replace_one_extent(trans, found_root,
5374 path, extent_key,
5375 &first_key, ref_path,
5376 new_extents, nr_extents);
5377 if (ret < 0)
5378 goto out;
5379 continue;
5380 }
5381
5382 btrfs_record_root_in_trans(found_root);
5383 if (ref_path->owner_objectid < BTRFS_FIRST_FREE_OBJECTID) {
5384 ret = relocate_tree_block(trans, found_root, path,
5385 &first_key, ref_path);
5386 } else {
5387 /*
5388 * try to update data extent references while
5389 * keeping metadata shared between snapshots.
5390 */
5391 ret = relocate_one_path(trans, found_root, path,
5392 &first_key, ref_path,
5393 group, reloc_inode);
5394 }
5395 if (ret < 0)
5396 goto out;
5397 }
5398 ret = 0;
5399out:
5400 btrfs_end_transaction(trans, extent_root);
5401 kfree(new_extents);
5402 kfree(ref_path);
1a40e23b
ZY		 5403 return ret;
5404}
5405
ec44a35c
CM		 5406static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
5407{
5408 u64 num_devices;
5409 u64 stripped = BTRFS_BLOCK_GROUP_RAID0 |
5410 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
5411
2b82032c 5412 num_devices = root->fs_info->fs_devices->rw_devices;
ec44a35c
CM		 5413 if (num_devices == 1) {
5414 stripped |= BTRFS_BLOCK_GROUP_DUP;
5415 stripped = flags & ~stripped;
5416
5417 /* turn raid0 into single device chunks */
5418 if (flags & BTRFS_BLOCK_GROUP_RAID0)
5419 return stripped;
5420
5421 /* turn mirroring into duplication */
5422 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
5423 BTRFS_BLOCK_GROUP_RAID10))
5424 return stripped | BTRFS_BLOCK_GROUP_DUP;
5425 return flags;
5426 } else {
5427 /* they already had raid on here, just return */
ec44a35c
CM		 5428 if (flags & stripped)
5429 return flags;
5430
5431 stripped |= BTRFS_BLOCK_GROUP_DUP;
5432 stripped = flags & ~stripped;
5433
5434 /* switch duplicated blocks with raid1 */
5435 if (flags & BTRFS_BLOCK_GROUP_DUP)
5436 return stripped | BTRFS_BLOCK_GROUP_RAID1;
5437
5438 /* turn single device chunks into raid0 */
5439 return stripped | BTRFS_BLOCK_GROUP_RAID0;
5440 }
5441 return flags;
5442}
5443
b2950863 5444static int __alloc_chunk_for_shrink(struct btrfs_root *root,
0ef3e66b
CM		 5445 struct btrfs_block_group_cache *shrink_block_group,
5446 int force)
5447{
5448 struct btrfs_trans_handle *trans;
5449 u64 new_alloc_flags;
5450 u64 calc;
5451
c286ac48 5452 spin_lock(&shrink_block_group->lock);
0ef3e66b 5453 if (btrfs_block_group_used(&shrink_block_group->item) > 0) {
c286ac48 5454 spin_unlock(&shrink_block_group->lock);
c286ac48 5455
0ef3e66b 5456 trans = btrfs_start_transaction(root, 1);
c286ac48 5457 spin_lock(&shrink_block_group->lock);
7d9eb12c 5458
0ef3e66b
CM		 5459 new_alloc_flags = update_block_group_flags(root,
5460 shrink_block_group->flags);
5461 if (new_alloc_flags != shrink_block_group->flags) {
5462 calc =
5463 btrfs_block_group_used(&shrink_block_group->item);
5464 } else {
5465 calc = shrink_block_group->key.offset;
5466 }
c286ac48
CM		 5467 spin_unlock(&shrink_block_group->lock);
5468
0ef3e66b
CM		 5469 do_chunk_alloc(trans, root->fs_info->extent_root,
5470 calc + 2 * 1024 * 1024, new_alloc_flags, force);
7d9eb12c 5471
0ef3e66b 5472 btrfs_end_transaction(trans, root);
c286ac48
CM		 5473 } else
5474 spin_unlock(&shrink_block_group->lock);
0ef3e66b
CM		 5475 return 0;
5476}
5477
1a40e23b
ZY		 5478static int __insert_orphan_inode(struct btrfs_trans_handle *trans,
5479 struct btrfs_root *root,
5480 u64 objectid, u64 size)
5481{
5482 struct btrfs_path *path;
5483 struct btrfs_inode_item *item;
5484 struct extent_buffer *leaf;
5485 int ret;
5486
5487 path = btrfs_alloc_path();
5488 if (!path)
5489 return -ENOMEM;
5490
5491 ret = btrfs_insert_empty_inode(trans, root, path, objectid);
5492 if (ret)
5493 goto out;
5494
5495 leaf = path->nodes[0];
5496 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_inode_item);
5497 memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item));
5498 btrfs_set_inode_generation(leaf, item, 1);
5499 btrfs_set_inode_size(leaf, item, size);
5500 btrfs_set_inode_mode(leaf, item, S_IFREG | 0600);
0403e47e 5501 btrfs_set_inode_flags(leaf, item, BTRFS_INODE_NOCOMPRESS);
1a40e23b
ZY		 5502 btrfs_mark_buffer_dirty(leaf);
5503 btrfs_release_path(root, path);
5504out:
5505 btrfs_free_path(path);
5506 return ret;
5507}
5508
5509static struct inode noinline *create_reloc_inode(struct btrfs_fs_info *fs_info,
5510 struct btrfs_block_group_cache *group)
5511{
5512 struct inode *inode = NULL;
5513 struct btrfs_trans_handle *trans;
5514 struct btrfs_root *root;
5515 struct btrfs_key root_key;
5516 u64 objectid = BTRFS_FIRST_FREE_OBJECTID;
5517 int err = 0;
5518
5519 root_key.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
5520 root_key.type = BTRFS_ROOT_ITEM_KEY;
5521 root_key.offset = (u64)-1;
5522 root = btrfs_read_fs_root_no_name(fs_info, &root_key);
5523 if (IS_ERR(root))
5524 return ERR_CAST(root);
5525
5526 trans = btrfs_start_transaction(root, 1);
5527 BUG_ON(!trans);
5528
5529 err = btrfs_find_free_objectid(trans, root, objectid, &objectid);
5530 if (err)
5531 goto out;
5532
5533 err = __insert_orphan_inode(trans, root, objectid, group->key.offset);
5534 BUG_ON(err);
5535
5536 err = btrfs_insert_file_extent(trans, root, objectid, 0, 0, 0,
c8b97818
CM		 5537 group->key.offset, 0, group->key.offset,
5538 0, 0, 0);
1a40e23b
ZY		 5539 BUG_ON(err);
5540
5541 inode = btrfs_iget_locked(root->fs_info->sb, objectid, root);
5542 if (inode->i_state & I_NEW) {
5543 BTRFS_I(inode)->root = root;
5544 BTRFS_I(inode)->location.objectid = objectid;
5545 BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
5546 BTRFS_I(inode)->location.offset = 0;
5547 btrfs_read_locked_inode(inode);
5548 unlock_new_inode(inode);
5549 BUG_ON(is_bad_inode(inode));
5550 } else {
5551 BUG_ON(1);
5552 }
5553
5554 err = btrfs_orphan_add(trans, inode);
5555out:
5556 btrfs_end_transaction(trans, root);
5557 if (err) {
5558 if (inode)
5559 iput(inode);
5560 inode = ERR_PTR(err);
5561 }
5562 return inode;
5563}
5564
5565int btrfs_relocate_block_group(struct btrfs_root *root, u64 group_start)
edbd8d4e
CM		 5566{
5567 struct btrfs_trans_handle *trans;
edbd8d4e 5568 struct btrfs_path *path;
1a40e23b
ZY		 5569 struct btrfs_fs_info *info = root->fs_info;
5570 struct extent_buffer *leaf;
5571 struct inode *reloc_inode;
5572 struct btrfs_block_group_cache *block_group;
5573 struct btrfs_key key;
d899e052 5574 u64 skipped;
edbd8d4e
CM		 5575 u64 cur_byte;
5576 u64 total_found;
edbd8d4e
CM		 5577 u32 nritems;
5578 int ret;
a061fc8d 5579 int progress;
1a40e23b 5580 int pass = 0;
edbd8d4e 5581
1a40e23b
ZY		 5582 root = root->fs_info->extent_root;
5583
5584 block_group = btrfs_lookup_block_group(info, group_start);
5585 BUG_ON(!block_group);
8f18cf13 5586
1a40e23b
ZY		 5587 printk("btrfs relocating block group %llu flags %llu\n",
5588 (unsigned long long)block_group->key.objectid,
5589 (unsigned long long)block_group->flags);
8f18cf13 5590
edbd8d4e 5591 path = btrfs_alloc_path();
1a40e23b 5592 BUG_ON(!path);
edbd8d4e 5593
1a40e23b
ZY		 5594 reloc_inode = create_reloc_inode(info, block_group);
5595 BUG_ON(IS_ERR(reloc_inode));
323da79c 5596
1a40e23b 5597 __alloc_chunk_for_shrink(root, block_group, 1);
c146afad 5598 set_block_group_readonly(block_group);
323da79c 5599
1a40e23b
ZY		 5600 btrfs_start_delalloc_inodes(info->tree_root);
5601 btrfs_wait_ordered_extents(info->tree_root, 0);
5602again:
d899e052 5603 skipped = 0;
edbd8d4e 5604 total_found = 0;
a061fc8d 5605 progress = 0;
1a40e23b 5606 key.objectid = block_group->key.objectid;
edbd8d4e
CM		 5607 key.offset = 0;
5608 key.type = 0;
73e48b27 5609 cur_byte = key.objectid;
4313b399 5610
1a40e23b
ZY		 5611 trans = btrfs_start_transaction(info->tree_root, 1);
5612 btrfs_commit_transaction(trans, info->tree_root);
ea8c2819 5613
1a40e23b
ZY		 5614 mutex_lock(&root->fs_info->cleaner_mutex);
5615 btrfs_clean_old_snapshots(info->tree_root);
5616 btrfs_remove_leaf_refs(info->tree_root, (u64)-1, 1);
5617 mutex_unlock(&root->fs_info->cleaner_mutex);
ea8c2819 5618
73e48b27 5619 while(1) {
edbd8d4e
CM		 5620 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5621 if (ret < 0)
5622 goto out;
7d9eb12c 5623next:
edbd8d4e 5624 leaf = path->nodes[0];
73e48b27 5625 nritems = btrfs_header_nritems(leaf);
73e48b27
Y		 5626 if (path->slots[0] >= nritems) {
5627 ret = btrfs_next_leaf(root, path);
5628 if (ret < 0)
5629 goto out;
5630 if (ret == 1) {
5631 ret = 0;
5632 break;
edbd8d4e 5633 }
73e48b27
Y		 5634 leaf = path->nodes[0];
5635 nritems = btrfs_header_nritems(leaf);
edbd8d4e 5636 }
73e48b27 5637
1a40e23b 5638 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
725c8463 5639
1a40e23b
ZY		 5640 if (key.objectid >= block_group->key.objectid +
5641 block_group->key.offset)
8f18cf13
CM		 5642 break;
5643
725c8463 5644 if (progress && need_resched()) {
725c8463 5645 btrfs_release_path(root, path);
1a40e23b 5646 cond_resched();
725c8463 5647 progress = 0;
1a40e23b 5648 continue;
725c8463
CM		 5649 }
5650 progress = 1;
5651
1a40e23b
ZY		 5652 if (btrfs_key_type(&key) != BTRFS_EXTENT_ITEM_KEY ||
5653 key.objectid + key.offset <= cur_byte) {
edbd8d4e 5654 path->slots[0]++;
edbd8d4e
CM		 5655 goto next;
5656 }
73e48b27 5657
edbd8d4e 5658 total_found++;
1a40e23b 5659 cur_byte = key.objectid + key.offset;
edbd8d4e 5660 btrfs_release_path(root, path);
edbd8d4e 5661
1a40e23b
ZY		 5662 __alloc_chunk_for_shrink(root, block_group, 0);
5663 ret = relocate_one_extent(root, path, &key, block_group,
5664 reloc_inode, pass);
5665 BUG_ON(ret < 0);
d899e052
YZ		 5666 if (ret > 0)
5667 skipped++;
edbd8d4e 5668
1a40e23b
ZY		 5669 key.objectid = cur_byte;
5670 key.type = 0;
5671 key.offset = 0;
edbd8d4e
CM		 5672 }
5673
1a40e23b 5674 btrfs_release_path(root, path);
7d9eb12c 5675
1a40e23b
ZY		 5676 if (pass == 0) {
5677 btrfs_wait_ordered_range(reloc_inode, 0, (u64)-1);
5678 invalidate_mapping_pages(reloc_inode->i_mapping, 0, -1);
5679 WARN_ON(reloc_inode->i_mapping->nrpages);
8e8a1e31 5680 }
73e48b27 5681
1a40e23b
ZY		 5682 if (total_found > 0) {
5683 printk("btrfs found %llu extents in pass %d\n",
5684 (unsigned long long)total_found, pass);
5685 pass++;
d899e052
YZ		 5686 if (total_found == skipped && pass > 2) {
5687 iput(reloc_inode);
5688 reloc_inode = create_reloc_inode(info, block_group);
5689 pass = 0;
5690 }
1a40e23b 5691 goto again;
0f9dd46c 5692 }
0ef3e66b 5693
1a40e23b
ZY		 5694 /* delete reloc_inode */
5695 iput(reloc_inode);
5696
5697 /* unpin extents in this range */
5698 trans = btrfs_start_transaction(info->tree_root, 1);
5699 btrfs_commit_transaction(trans, info->tree_root);
0ef3e66b 5700
1a40e23b
ZY		 5701 spin_lock(&block_group->lock);
5702 WARN_ON(block_group->pinned > 0);
5703 WARN_ON(block_group->reserved > 0);
5704 WARN_ON(btrfs_block_group_used(&block_group->item) > 0);
5705 spin_unlock(&block_group->lock);
5706 ret = 0;
edbd8d4e 5707out:
1a40e23b 5708 btrfs_free_path(path);
edbd8d4e
CM		 5709 return ret;
5710}
5711
b2950863
CH		 5712static int find_first_block_group(struct btrfs_root *root,
5713 struct btrfs_path *path, struct btrfs_key *key)
0b86a832 5714{
925baedd 5715 int ret = 0;
0b86a832
CM		 5716 struct btrfs_key found_key;
5717 struct extent_buffer *leaf;
5718 int slot;
edbd8d4e 5719
0b86a832
CM		 5720 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
5721 if (ret < 0)
925baedd
CM		 5722 goto out;
5723
0b86a832
CM		 5724 while(1) {
5725 slot = path->slots[0];
edbd8d4e 5726 leaf = path->nodes[0];
0b86a832
CM		 5727 if (slot >= btrfs_header_nritems(leaf)) {
5728 ret = btrfs_next_leaf(root, path);
5729 if (ret == 0)
5730 continue;
5731 if (ret < 0)
925baedd 5732 goto out;
0b86a832 5733 break;
edbd8d4e 5734 }
0b86a832 5735 btrfs_item_key_to_cpu(leaf, &found_key, slot);
edbd8d4e 5736
0b86a832 5737 if (found_key.objectid >= key->objectid &&
925baedd
CM		 5738 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
5739 ret = 0;
5740 goto out;
5741 }
0b86a832 5742 path->slots[0]++;
edbd8d4e 5743 }
0b86a832 5744 ret = -ENOENT;
925baedd 5745out:
0b86a832 5746 return ret;
edbd8d4e
CM		 5747}
5748
1a40e23b
ZY		 5749int btrfs_free_block_groups(struct btrfs_fs_info *info)
5750{
5751 struct btrfs_block_group_cache *block_group;
5752 struct rb_node *n;
5753
1a40e23b
ZY		 5754 spin_lock(&info->block_group_cache_lock);
5755 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
5756 block_group = rb_entry(n, struct btrfs_block_group_cache,
5757 cache_node);
1a40e23b
ZY		 5758 rb_erase(&block_group->cache_node,
5759 &info->block_group_cache_tree);
d899e052
YZ		 5760 spin_unlock(&info->block_group_cache_lock);
5761
5762 btrfs_remove_free_space_cache(block_group);
80eb234a 5763 down_write(&block_group->space_info->groups_sem);
1a40e23b 5764 list_del(&block_group->list);
80eb234a 5765 up_write(&block_group->space_info->groups_sem);
1a40e23b 5766 kfree(block_group);
d899e052
YZ		 5767
5768 spin_lock(&info->block_group_cache_lock);
1a40e23b
ZY		 5769 }
5770 spin_unlock(&info->block_group_cache_lock);
1a40e23b
ZY		 5771 return 0;
5772}
5773
9078a3e1
CM		 5774int btrfs_read_block_groups(struct btrfs_root *root)
5775{
5776 struct btrfs_path *path;
5777 int ret;
9078a3e1 5778 struct btrfs_block_group_cache *cache;
be744175 5779 struct btrfs_fs_info *info = root->fs_info;
6324fbf3 5780 struct btrfs_space_info *space_info;
9078a3e1
CM		 5781 struct btrfs_key key;
5782 struct btrfs_key found_key;
5f39d397 5783 struct extent_buffer *leaf;
96b5179d 5784
be744175 5785 root = info->extent_root;
9078a3e1 5786 key.objectid = 0;
0b86a832 5787 key.offset = 0;
9078a3e1 5788 btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
9078a3e1
CM		 5789 path = btrfs_alloc_path();
5790 if (!path)
5791 return -ENOMEM;
5792
5793 while(1) {
0b86a832
CM		 5794 ret = find_first_block_group(root, path, &key);
5795 if (ret > 0) {
5796 ret = 0;
5797 goto error;
9078a3e1 5798 }
0b86a832
CM		 5799 if (ret != 0)
5800 goto error;
5801
5f39d397
CM		 5802 leaf = path->nodes[0];
5803 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
8f18cf13 5804 cache = kzalloc(sizeof(*cache), GFP_NOFS);
9078a3e1 5805 if (!cache) {
0b86a832 5806 ret = -ENOMEM;
9078a3e1
CM		 5807 break;
5808 }
3e1ad54f 5809
c286ac48 5810 spin_lock_init(&cache->lock);
25179201 5811 mutex_init(&cache->alloc_mutex);
ea6a478e 5812 mutex_init(&cache->cache_mutex);
0f9dd46c 5813 INIT_LIST_HEAD(&cache->list);
5f39d397
CM		 5814 read_extent_buffer(leaf, &cache->item,
5815 btrfs_item_ptr_offset(leaf, path->slots[0]),
5816 sizeof(cache->item));
9078a3e1 5817 memcpy(&cache->key, &found_key, sizeof(found_key));
0b86a832 5818
9078a3e1
CM		 5819 key.objectid = found_key.objectid + found_key.offset;
5820 btrfs_release_path(root, path);
0b86a832 5821 cache->flags = btrfs_block_group_flags(&cache->item);
96b5179d 5822
6324fbf3
CM		 5823 ret = update_space_info(info, cache->flags, found_key.offset,
5824 btrfs_block_group_used(&cache->item),
5825 &space_info);
5826 BUG_ON(ret);
5827 cache->space_info = space_info;
80eb234a
JB		 5828 down_write(&space_info->groups_sem);
5829 list_add_tail(&cache->list, &space_info->block_groups);
5830 up_write(&space_info->groups_sem);
0f9dd46c
JB		 5831
5832 ret = btrfs_add_block_group_cache(root->fs_info, cache);
5833 BUG_ON(ret);
75ccf47d
CM		 5834
5835 set_avail_alloc_bits(root->fs_info, cache->flags);
2b82032c
YZ		 5836 if (btrfs_chunk_readonly(root, cache->key.objectid))
5837 set_block_group_readonly(cache);
9078a3e1 5838 }
0b86a832
CM		 5839 ret = 0;
5840error:
9078a3e1 5841 btrfs_free_path(path);
0b86a832 5842 return ret;
9078a3e1 5843}
6324fbf3
CM		 5844
5845int btrfs_make_block_group(struct btrfs_trans_handle *trans,
5846 struct btrfs_root *root, u64 bytes_used,
e17cade2 5847 u64 type, u64 chunk_objectid, u64 chunk_offset,
6324fbf3
CM		 5848 u64 size)
5849{
5850 int ret;
6324fbf3
CM		 5851 struct btrfs_root *extent_root;
5852 struct btrfs_block_group_cache *cache;
6324fbf3
CM		 5853
5854 extent_root = root->fs_info->extent_root;
6324fbf3 5855
e02119d5
CM		 5856 root->fs_info->last_trans_new_blockgroup = trans->transid;
5857
8f18cf13 5858 cache = kzalloc(sizeof(*cache), GFP_NOFS);
0f9dd46c
JB		 5859 if (!cache)
5860 return -ENOMEM;
5861
e17cade2 5862 cache->key.objectid = chunk_offset;
6324fbf3 5863 cache->key.offset = size;
c286ac48 5864 spin_lock_init(&cache->lock);
25179201 5865 mutex_init(&cache->alloc_mutex);
ea6a478e 5866 mutex_init(&cache->cache_mutex);
0f9dd46c 5867 INIT_LIST_HEAD(&cache->list);
6324fbf3 5868 btrfs_set_key_type(&cache->key, BTRFS_BLOCK_GROUP_ITEM_KEY);
0ef3e66b 5869
6324fbf3 5870 btrfs_set_block_group_used(&cache->item, bytes_used);
6324fbf3
CM		 5871 btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
5872 cache->flags = type;
5873 btrfs_set_block_group_flags(&cache->item, type);
5874
5875 ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
5876 &cache->space_info);
5877 BUG_ON(ret);
80eb234a
JB		 5878 down_write(&cache->space_info->groups_sem);
5879 list_add_tail(&cache->list, &cache->space_info->block_groups);
5880 up_write(&cache->space_info->groups_sem);
6324fbf3 5881
0f9dd46c
JB		 5882 ret = btrfs_add_block_group_cache(root->fs_info, cache);
5883 BUG_ON(ret);
c286ac48 5884
6324fbf3
CM		 5885 ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
5886 sizeof(cache->item));
5887 BUG_ON(ret);
5888
87ef2bb4
CM		 5889 finish_current_insert(trans, extent_root, 0);
5890 ret = del_pending_extents(trans, extent_root, 0);
6324fbf3 5891 BUG_ON(ret);
d18a2c44 5892 set_avail_alloc_bits(extent_root->fs_info, type);
925baedd 5893
6324fbf3
CM		 5894 return 0;
5895}
1a40e23b
ZY		 5896
5897int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
5898 struct btrfs_root *root, u64 group_start)
5899{
5900 struct btrfs_path *path;
5901 struct btrfs_block_group_cache *block_group;
5902 struct btrfs_key key;
5903 int ret;
5904
1a40e23b
ZY		 5905 root = root->fs_info->extent_root;
5906
5907 block_group = btrfs_lookup_block_group(root->fs_info, group_start);
5908 BUG_ON(!block_group);
c146afad 5909 BUG_ON(!block_group->ro);
1a40e23b
ZY		 5910
5911 memcpy(&key, &block_group->key, sizeof(key));
5912
5913 path = btrfs_alloc_path();
5914 BUG_ON(!path);
5915
5916 btrfs_remove_free_space_cache(block_group);
5917 rb_erase(&block_group->cache_node,
5918 &root->fs_info->block_group_cache_tree);
80eb234a 5919 down_write(&block_group->space_info->groups_sem);
1a40e23b 5920 list_del(&block_group->list);
80eb234a 5921 up_write(&block_group->space_info->groups_sem);
1a40e23b 5922
c146afad
YZ		 5923 spin_lock(&block_group->space_info->lock);
5924 block_group->space_info->total_bytes -= block_group->key.offset;
5925 block_group->space_info->bytes_readonly -= block_group->key.offset;
5926 spin_unlock(&block_group->space_info->lock);
2b82032c 5927 block_group->space_info->full = 0;
c146afad 5928
1a40e23b
ZY		 5929 /*
5930 memset(shrink_block_group, 0, sizeof(*shrink_block_group));
5931 kfree(shrink_block_group);
5932 */
5933
5934 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
5935 if (ret > 0)
5936 ret = -EIO;
5937 if (ret < 0)
5938 goto out;
5939
5940 ret = btrfs_del_item(trans, root, path);
5941out:
5942 btrfs_free_path(path);
5943 return ret;
5944}
Linux 6.x block layer and io_uring tree(s)