projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Btrfs: Avoid unplug storms during commit
[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>
74493f7a 22#include "hash.h"
a5eb62e3 23#include "crc32c.h"
fec577fb
CM		 24#include "ctree.h"
25#include "disk-io.h"
26#include "print-tree.h"
e089f05c 27#include "transaction.h"
0b86a832 28#include "volumes.h"
925baedd 29#include "locking.h"
31153d81 30#include "ref-cache.h"
fec577fb 31
31840ae1
ZY		 32#define PENDING_EXTENT_INSERT 0
33#define PENDING_EXTENT_DELETE 1
34#define PENDING_BACKREF_UPDATE 2
35
36struct pending_extent_op {
37 int type;
38 u64 bytenr;
39 u64 num_bytes;
40 u64 parent;
41 u64 orig_parent;
42 u64 generation;
43 u64 orig_generation;
44 int level;
45};
46
e089f05c 47static int finish_current_insert(struct btrfs_trans_handle *trans, struct
87ef2bb4 48 btrfs_root *extent_root, int all);
e20d96d6 49static int del_pending_extents(struct btrfs_trans_handle *trans, struct
87ef2bb4 50 btrfs_root *extent_root, int all);
925baedd
CM		 51static struct btrfs_block_group_cache *
52__btrfs_find_block_group(struct btrfs_root *root,
53 struct btrfs_block_group_cache *hint,
54 u64 search_start, int data, int owner);
d548ee51 55
0f9dd46c
JB		 56static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
57{
58 return (cache->flags & bits) == bits;
59}
60
61/*
62 * this adds the block group to the fs_info rb tree for the block group
63 * cache
64 */
65int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
66 struct btrfs_block_group_cache *block_group)
67{
68 struct rb_node **p;
69 struct rb_node *parent = NULL;
70 struct btrfs_block_group_cache *cache;
71
72 spin_lock(&info->block_group_cache_lock);
73 p = &info->block_group_cache_tree.rb_node;
74
75 while (*p) {
76 parent = *p;
77 cache = rb_entry(parent, struct btrfs_block_group_cache,
78 cache_node);
79 if (block_group->key.objectid < cache->key.objectid) {
80 p = &(*p)->rb_left;
81 } else if (block_group->key.objectid > cache->key.objectid) {
82 p = &(*p)->rb_right;
83 } else {
84 spin_unlock(&info->block_group_cache_lock);
85 return -EEXIST;
86 }
87 }
88
89 rb_link_node(&block_group->cache_node, parent, p);
90 rb_insert_color(&block_group->cache_node,
91 &info->block_group_cache_tree);
92 spin_unlock(&info->block_group_cache_lock);
93
94 return 0;
95}
96
97/*
98 * This will return the block group at or after bytenr if contains is 0, else
99 * it will return the block group that contains the bytenr
100 */
101static struct btrfs_block_group_cache *
102block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
103 int contains)
104{
105 struct btrfs_block_group_cache *cache, *ret = NULL;
106 struct rb_node *n;
107 u64 end, start;
108
109 spin_lock(&info->block_group_cache_lock);
110 n = info->block_group_cache_tree.rb_node;
111
112 while (n) {
113 cache = rb_entry(n, struct btrfs_block_group_cache,
114 cache_node);
115 end = cache->key.objectid + cache->key.offset - 1;
116 start = cache->key.objectid;
117
118 if (bytenr < start) {
119 if (!contains && (!ret || start < ret->key.objectid))
120 ret = cache;
121 n = n->rb_left;
122 } else if (bytenr > start) {
123 if (contains && bytenr <= end) {
124 ret = cache;
125 break;
126 }
127 n = n->rb_right;
128 } else {
129 ret = cache;
130 break;
131 }
132 }
133 spin_unlock(&info->block_group_cache_lock);
134
135 return ret;
136}
137
138/*
139 * this is only called by cache_block_group, since we could have freed extents
140 * we need to check the pinned_extents for any extents that can't be used yet
141 * since their free space will be released as soon as the transaction commits.
142 */
143static int add_new_free_space(struct btrfs_block_group_cache *block_group,
144 struct btrfs_fs_info *info, u64 start, u64 end)
145{
146 u64 extent_start, extent_end, size;
147 int ret;
148
25179201 149 mutex_lock(&info->pinned_mutex);
0f9dd46c
JB		 150 while (start < end) {
151 ret = find_first_extent_bit(&info->pinned_extents, start,
152 &extent_start, &extent_end,
153 EXTENT_DIRTY);
154 if (ret)
155 break;
156
157 if (extent_start == start) {
158 start = extent_end + 1;
159 } else if (extent_start > start && extent_start < end) {
160 size = extent_start - start;
25179201
JB		 161 ret = btrfs_add_free_space_lock(block_group, start,
162 size);
0f9dd46c
JB		 163 BUG_ON(ret);
164 start = extent_end + 1;
165 } else {
166 break;
167 }
168 }
169
170 if (start < end) {
171 size = end - start;
25179201 172 ret = btrfs_add_free_space_lock(block_group, start, size);
0f9dd46c
JB		 173 BUG_ON(ret);
174 }
25179201 175 mutex_unlock(&info->pinned_mutex);
0f9dd46c
JB		 176
177 return 0;
178}
179
e37c9e69
CM		 180static int cache_block_group(struct btrfs_root *root,
181 struct btrfs_block_group_cache *block_group)
182{
183 struct btrfs_path *path;
ef8bbdfe 184 int ret = 0;
e37c9e69 185 struct btrfs_key key;
5f39d397 186 struct extent_buffer *leaf;
e37c9e69 187 int slot;
e37c9e69 188 u64 last = 0;
7d7d6068 189 u64 first_free;
e37c9e69
CM		 190 int found = 0;
191
00f5c795
CM		 192 if (!block_group)
193 return 0;
194
e37c9e69 195 root = root->fs_info->extent_root;
e37c9e69
CM		 196
197 if (block_group->cached)
198 return 0;
f510cfec 199
e37c9e69
CM		 200 path = btrfs_alloc_path();
201 if (!path)
202 return -ENOMEM;
7d7d6068 203
2cc58cf2 204 path->reada = 2;
5cd57b2c
CM		 205 /*
206 * we get into deadlocks with paths held by callers of this function.
207 * since the alloc_mutex is protecting things right now, just
208 * skip the locking here
209 */
210 path->skip_locking = 1;
0f9dd46c
JB		 211 first_free = max_t(u64, block_group->key.objectid,
212 BTRFS_SUPER_INFO_OFFSET + BTRFS_SUPER_INFO_SIZE);
e37c9e69 213 key.objectid = block_group->key.objectid;
e37c9e69
CM		 214 key.offset = 0;
215 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
216 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
217 if (ret < 0)
ef8bbdfe 218 goto err;
0b86a832 219 ret = btrfs_previous_item(root, path, 0, BTRFS_EXTENT_ITEM_KEY);
d548ee51 220 if (ret < 0)
ef8bbdfe 221 goto err;
d548ee51
Y		 222 if (ret == 0) {
223 leaf = path->nodes[0];
224 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
225 if (key.objectid + key.offset > first_free)
226 first_free = key.objectid + key.offset;
227 }
e37c9e69 228 while(1) {
5f39d397 229 leaf = path->nodes[0];
e37c9e69 230 slot = path->slots[0];
5f39d397 231 if (slot >= btrfs_header_nritems(leaf)) {
e37c9e69 232 ret = btrfs_next_leaf(root, path);
54aa1f4d
CM		 233 if (ret < 0)
234 goto err;
0f9dd46c 235 if (ret == 0)
e37c9e69 236 continue;
0f9dd46c 237 else
e37c9e69 238 break;
e37c9e69 239 }
5f39d397 240 btrfs_item_key_to_cpu(leaf, &key, slot);
0f9dd46c 241 if (key.objectid < block_group->key.objectid)
7d7d6068 242 goto next;
0f9dd46c 243
e37c9e69 244 if (key.objectid >= block_group->key.objectid +
0f9dd46c 245 block_group->key.offset)
e37c9e69 246 break;
7d7d6068 247
e37c9e69
CM		 248 if (btrfs_key_type(&key) == BTRFS_EXTENT_ITEM_KEY) {
249 if (!found) {
7d7d6068 250 last = first_free;
e37c9e69 251 found = 1;
e37c9e69 252 }
0f9dd46c
JB		 253
254 add_new_free_space(block_group, root->fs_info, last,
255 key.objectid);
256
7d7d6068 257 last = key.objectid + key.offset;
e37c9e69 258 }
7d7d6068 259next:
e37c9e69
CM		 260 path->slots[0]++;
261 }
262
7d7d6068
Y		 263 if (!found)
264 last = first_free;
0f9dd46c
JB		 265
266 add_new_free_space(block_group, root->fs_info, last,
267 block_group->key.objectid +
268 block_group->key.offset);
269
e37c9e69 270 block_group->cached = 1;
ef8bbdfe 271 ret = 0;
54aa1f4d 272err:
e37c9e69 273 btrfs_free_path(path);
ef8bbdfe 274 return ret;
e37c9e69
CM		 275}
276
0f9dd46c
JB		 277/*
278 * return the block group that starts at or after bytenr
279 */
0ef3e66b
CM		 280struct btrfs_block_group_cache *btrfs_lookup_first_block_group(struct
281 btrfs_fs_info *info,
282 u64 bytenr)
283{
0f9dd46c 284 struct btrfs_block_group_cache *cache;
0ef3e66b 285
0f9dd46c 286 cache = block_group_cache_tree_search(info, bytenr, 0);
0ef3e66b 287
0f9dd46c 288 return cache;
0ef3e66b
CM		 289}
290
0f9dd46c
JB		 291/*
292 * return the block group that contains teh given bytenr
293 */
5276aeda
CM		 294struct btrfs_block_group_cache *btrfs_lookup_block_group(struct
295 btrfs_fs_info *info,
db94535d 296 u64 bytenr)
be744175 297{
0f9dd46c 298 struct btrfs_block_group_cache *cache;
be744175 299
0f9dd46c 300 cache = block_group_cache_tree_search(info, bytenr, 1);
96b5179d 301
0f9dd46c 302 return cache;
be744175 303}
0b86a832 304
0f9dd46c
JB		 305static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
306 u64 flags)
6324fbf3 307{
0f9dd46c
JB		 308 struct list_head *head = &info->space_info;
309 struct list_head *cur;
310 struct btrfs_space_info *found;
311 list_for_each(cur, head) {
312 found = list_entry(cur, struct btrfs_space_info, list);
313 if (found->flags == flags)
314 return found;
315 }
316 return NULL;
6324fbf3
CM		 317}
318
80eb234a
JB		 319static u64 div_factor(u64 num, int factor)
320{
321 if (factor == 10)
322 return num;
323 num *= factor;
324 do_div(num, 10);
325 return num;
326}
327
925baedd
CM		 328static struct btrfs_block_group_cache *
329__btrfs_find_block_group(struct btrfs_root *root,
330 struct btrfs_block_group_cache *hint,
331 u64 search_start, int data, int owner)
cd1bc465 332{
96b5179d 333 struct btrfs_block_group_cache *cache;
31f3c99b 334 struct btrfs_block_group_cache *found_group = NULL;
cd1bc465
CM		 335 struct btrfs_fs_info *info = root->fs_info;
336 u64 used;
31f3c99b 337 u64 last = 0;
96b5179d 338 u64 free_check;
31f3c99b 339 int full_search = 0;
bce4eae9 340 int factor = 10;
0ef3e66b 341 int wrapped = 0;
de428b63 342
a236aed1
CM		 343 if (data & BTRFS_BLOCK_GROUP_METADATA)
344 factor = 9;
be744175 345
0ef3e66b 346 if (search_start) {
be744175 347 struct btrfs_block_group_cache *shint;
0ef3e66b 348 shint = btrfs_lookup_first_block_group(info, search_start);
8f18cf13 349 if (shint && block_group_bits(shint, data) && !shint->ro) {
c286ac48 350 spin_lock(&shint->lock);
be744175 351 used = btrfs_block_group_used(&shint->item);
e8569813 352 if (used + shint->pinned + shint->reserved <
324ae4df 353 div_factor(shint->key.offset, factor)) {
c286ac48 354 spin_unlock(&shint->lock);
be744175
CM		 355 return shint;
356 }
c286ac48 357 spin_unlock(&shint->lock);
be744175
CM		 358 }
359 }
0ef3e66b 360 if (hint && !hint->ro && block_group_bits(hint, data)) {
c286ac48 361 spin_lock(&hint->lock);
31f3c99b 362 used = btrfs_block_group_used(&hint->item);
e8569813 363 if (used + hint->pinned + hint->reserved <
324ae4df 364 div_factor(hint->key.offset, factor)) {
c286ac48 365 spin_unlock(&hint->lock);
31f3c99b
CM		 366 return hint;
367 }
c286ac48 368 spin_unlock(&hint->lock);
e19caa5f 369 last = hint->key.objectid + hint->key.offset;
31f3c99b 370 } else {
e37c9e69 371 if (hint)
0ef3e66b 372 last = max(hint->key.objectid, search_start);
e37c9e69 373 else
0ef3e66b 374 last = search_start;
31f3c99b 375 }
31f3c99b 376again:
e8569813
ZY		 377 while (1) {
378 cache = btrfs_lookup_first_block_group(root->fs_info, last);
0f9dd46c
JB		 379 if (!cache)
380 break;
96b5179d 381
c286ac48 382 spin_lock(&cache->lock);
96b5179d
CM		 383 last = cache->key.objectid + cache->key.offset;
384 used = btrfs_block_group_used(&cache->item);
385
8f18cf13 386 if (!cache->ro && block_group_bits(cache, data)) {
0ef3e66b 387 free_check = div_factor(cache->key.offset, factor);
e8569813
ZY		 388 if (used + cache->pinned + cache->reserved <
389 free_check) {
8790d502 390 found_group = cache;
c286ac48 391 spin_unlock(&cache->lock);
8790d502
CM		 392 goto found;
393 }
6324fbf3 394 }
c286ac48 395 spin_unlock(&cache->lock);
de428b63 396 cond_resched();
cd1bc465 397 }
0ef3e66b
CM		 398 if (!wrapped) {
399 last = search_start;
400 wrapped = 1;
401 goto again;
402 }
403 if (!full_search && factor < 10) {
be744175 404 last = search_start;
31f3c99b 405 full_search = 1;
0ef3e66b 406 factor = 10;
31f3c99b
CM		 407 goto again;
408 }
be744175 409found:
31f3c99b 410 return found_group;
cd1bc465
CM		 411}
412
925baedd
CM		 413struct btrfs_block_group_cache *btrfs_find_block_group(struct btrfs_root *root,
414 struct btrfs_block_group_cache
415 *hint, u64 search_start,
416 int data, int owner)
417{
418
419 struct btrfs_block_group_cache *ret;
925baedd 420 ret = __btrfs_find_block_group(root, hint, search_start, data, owner);
925baedd
CM		 421 return ret;
422}
0f9dd46c 423
e02119d5 424/* simple helper to search for an existing extent at a given offset */
31840ae1 425int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len)
e02119d5
CM		 426{
427 int ret;
428 struct btrfs_key key;
31840ae1 429 struct btrfs_path *path;
e02119d5 430
31840ae1
ZY		 431 path = btrfs_alloc_path();
432 BUG_ON(!path);
e02119d5
CM		 433 key.objectid = start;
434 key.offset = len;
435 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
436 ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
437 0, 0);
31840ae1 438 btrfs_free_path(path);
7bb86316
CM		 439 return ret;
440}
441
d8d5f3e1
CM		 442/*
443 * Back reference rules. Back refs have three main goals:
444 *
445 * 1) differentiate between all holders of references to an extent so that
446 * when a reference is dropped we can make sure it was a valid reference
447 * before freeing the extent.
448 *
449 * 2) Provide enough information to quickly find the holders of an extent
450 * if we notice a given block is corrupted or bad.
451 *
452 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
453 * maintenance. This is actually the same as #2, but with a slightly
454 * different use case.
455 *
456 * File extents can be referenced by:
457 *
458 * - multiple snapshots, subvolumes, or different generations in one subvol
31840ae1 459 * - different files inside a single subvolume
d8d5f3e1
CM		 460 * - different offsets inside a file (bookend extents in file.c)
461 *
462 * The extent ref structure has fields for:
463 *
464 * - Objectid of the subvolume root
465 * - Generation number of the tree holding the reference
466 * - objectid of the file holding the reference
31840ae1
ZY		 467 * - number of references holding by parent node (alway 1 for tree blocks)
468 *
469 * Btree leaf may hold multiple references to a file extent. In most cases,
470 * these references are from same file and the corresponding offsets inside
3bb1a1bc 471 * the file are close together.
d8d5f3e1
CM		 472 *
473 * When a file extent is allocated the fields are filled in:
3bb1a1bc 474 * (root_key.objectid, trans->transid, inode objectid, 1)
d8d5f3e1
CM		 475 *
476 * When a leaf is cow'd new references are added for every file extent found
31840ae1
ZY		 477 * in the leaf. It looks similar to the create case, but trans->transid will
478 * be different when the block is cow'd.
d8d5f3e1 479 *
3bb1a1bc 480 * (root_key.objectid, trans->transid, inode objectid,
31840ae1 481 * number of references in the leaf)
d8d5f3e1 482 *
3bb1a1bc
YZ		 483 * When a file extent is removed either during snapshot deletion or
484 * file truncation, we find the corresponding back reference and check
485 * the following fields:
d8d5f3e1 486 *
3bb1a1bc
YZ		 487 * (btrfs_header_owner(leaf), btrfs_header_generation(leaf),
488 * inode objectid)
d8d5f3e1
CM		 489 *
490 * Btree extents can be referenced by:
491 *
492 * - Different subvolumes
493 * - Different generations of the same subvolume
494 *
d8d5f3e1
CM		 495 * When a tree block is created, back references are inserted:
496 *
3bb1a1bc 497 * (root->root_key.objectid, trans->transid, level, 1)
d8d5f3e1 498 *
31840ae1
ZY		 499 * When a tree block is cow'd, new back references are added for all the
500 * blocks it points to. If the tree block isn't in reference counted root,
501 * the old back references are removed. These new back references are of
502 * the form (trans->transid will have increased since creation):
d8d5f3e1 503 *
3bb1a1bc 504 * (root->root_key.objectid, trans->transid, level, 1)
d8d5f3e1 505 *
31840ae1 506 * When a backref is in deleting, the following fields are checked:
d8d5f3e1
CM		 507 *
508 * if backref was for a tree root:
3bb1a1bc 509 * (btrfs_header_owner(itself), btrfs_header_generation(itself), level)
d8d5f3e1 510 * else
3bb1a1bc 511 * (btrfs_header_owner(parent), btrfs_header_generation(parent), level)
d8d5f3e1 512 *
31840ae1 513 * Back Reference Key composing:
d8d5f3e1 514 *
31840ae1
ZY		 515 * The key objectid corresponds to the first byte in the extent, the key
516 * type is set to BTRFS_EXTENT_REF_KEY, and the key offset is the first
517 * byte of parent extent. If a extent is tree root, the key offset is set
518 * to the key objectid.
d8d5f3e1 519 */
31840ae1
ZY		 520
521static int noinline lookup_extent_backref(struct btrfs_trans_handle *trans,
522 struct btrfs_root *root,
3bb1a1bc
YZ		 523 struct btrfs_path *path,
524 u64 bytenr, u64 parent,
525 u64 ref_root, u64 ref_generation,
526 u64 owner_objectid, int del)
7bb86316 527{
7bb86316 528 struct btrfs_key key;
31840ae1
ZY		 529 struct btrfs_extent_ref *ref;
530 struct extent_buffer *leaf;
3bb1a1bc 531 u64 ref_objectid;
74493f7a
CM		 532 int ret;
533
31840ae1
ZY		 534 key.objectid = bytenr;
535 key.type = BTRFS_EXTENT_REF_KEY;
536 key.offset = parent;
537
538 ret = btrfs_search_slot(trans, root, &key, path, del ? -1 : 0, 1);
539 if (ret < 0)
540 goto out;
541 if (ret > 0) {
542 ret = -ENOENT;
543 goto out;
544 }
545
546 leaf = path->nodes[0];
547 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_ref);
3bb1a1bc 548 ref_objectid = btrfs_ref_objectid(leaf, ref);
31840ae1 549 if (btrfs_ref_root(leaf, ref) != ref_root ||
3bb1a1bc
YZ		 550 btrfs_ref_generation(leaf, ref) != ref_generation ||
551 (ref_objectid != owner_objectid &&
552 ref_objectid != BTRFS_MULTIPLE_OBJECTIDS)) {
31840ae1
ZY		 553 ret = -EIO;
554 WARN_ON(1);
555 goto out;
556 }
557 ret = 0;
558out:
559 return ret;
560}
561
562static int noinline insert_extent_backref(struct btrfs_trans_handle *trans,
563 struct btrfs_root *root,
564 struct btrfs_path *path,
565 u64 bytenr, u64 parent,
566 u64 ref_root, u64 ref_generation,
3bb1a1bc 567 u64 owner_objectid)
31840ae1
ZY		 568{
569 struct btrfs_key key;
570 struct extent_buffer *leaf;
571 struct btrfs_extent_ref *ref;
572 u32 num_refs;
573 int ret;
74493f7a 574
74493f7a
CM		 575 key.objectid = bytenr;
576 key.type = BTRFS_EXTENT_REF_KEY;
31840ae1 577 key.offset = parent;
74493f7a 578
31840ae1
ZY		 579 ret = btrfs_insert_empty_item(trans, root, path, &key, sizeof(*ref));
580 if (ret == 0) {
581 leaf = path->nodes[0];
582 ref = btrfs_item_ptr(leaf, path->slots[0],
583 struct btrfs_extent_ref);
584 btrfs_set_ref_root(leaf, ref, ref_root);
585 btrfs_set_ref_generation(leaf, ref, ref_generation);
586 btrfs_set_ref_objectid(leaf, ref, owner_objectid);
31840ae1
ZY		 587 btrfs_set_ref_num_refs(leaf, ref, 1);
588 } else if (ret == -EEXIST) {
589 u64 existing_owner;
590 BUG_ON(owner_objectid < BTRFS_FIRST_FREE_OBJECTID);
591 leaf = path->nodes[0];
592 ref = btrfs_item_ptr(leaf, path->slots[0],
593 struct btrfs_extent_ref);
594 if (btrfs_ref_root(leaf, ref) != ref_root ||
595 btrfs_ref_generation(leaf, ref) != ref_generation) {
596 ret = -EIO;
597 WARN_ON(1);
7bb86316 598 goto out;
31840ae1
ZY		 599 }
600
601 num_refs = btrfs_ref_num_refs(leaf, ref);
602 BUG_ON(num_refs == 0);
603 btrfs_set_ref_num_refs(leaf, ref, num_refs + 1);
604
605 existing_owner = btrfs_ref_objectid(leaf, ref);
3bb1a1bc
YZ		 606 if (existing_owner != owner_objectid &&
607 existing_owner != BTRFS_MULTIPLE_OBJECTIDS) {
31840ae1
ZY		 608 btrfs_set_ref_objectid(leaf, ref,
609 BTRFS_MULTIPLE_OBJECTIDS);
31840ae1
ZY		 610 }
611 ret = 0;
612 } else {
7bb86316 613 goto out;
31840ae1 614 }
7bb86316
CM		 615 btrfs_mark_buffer_dirty(path->nodes[0]);
616out:
617 btrfs_release_path(root, path);
618 return ret;
74493f7a
CM		 619}
620
31840ae1
ZY		 621static int noinline remove_extent_backref(struct btrfs_trans_handle *trans,
622 struct btrfs_root *root,
623 struct btrfs_path *path)
624{
625 struct extent_buffer *leaf;
626 struct btrfs_extent_ref *ref;
627 u32 num_refs;
628 int ret = 0;
629
630 leaf = path->nodes[0];
631 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_ref);
632 num_refs = btrfs_ref_num_refs(leaf, ref);
633 BUG_ON(num_refs == 0);
634 num_refs -= 1;
635 if (num_refs == 0) {
636 ret = btrfs_del_item(trans, root, path);
637 } else {
638 btrfs_set_ref_num_refs(leaf, ref, num_refs);
639 btrfs_mark_buffer_dirty(leaf);
640 }
641 btrfs_release_path(root, path);
642 return ret;
643}
644
645static int __btrfs_update_extent_ref(struct btrfs_trans_handle *trans,
646 struct btrfs_root *root, u64 bytenr,
647 u64 orig_parent, u64 parent,
648 u64 orig_root, u64 ref_root,
649 u64 orig_generation, u64 ref_generation,
3bb1a1bc 650 u64 owner_objectid)
31840ae1
ZY		 651{
652 int ret;
653 struct btrfs_root *extent_root = root->fs_info->extent_root;
654 struct btrfs_path *path;
655
656 if (root == root->fs_info->extent_root) {
657 struct pending_extent_op *extent_op;
658 u64 num_bytes;
659
660 BUG_ON(owner_objectid >= BTRFS_MAX_LEVEL);
661 num_bytes = btrfs_level_size(root, (int)owner_objectid);
25179201 662 mutex_lock(&root->fs_info->extent_ins_mutex);
31840ae1 663 if (test_range_bit(&root->fs_info->extent_ins, bytenr,
25179201 664 bytenr + num_bytes - 1, EXTENT_WRITEBACK, 0)) {
31840ae1
ZY		 665 u64 priv;
666 ret = get_state_private(&root->fs_info->extent_ins,
667 bytenr, &priv);
668 BUG_ON(ret);
669 extent_op = (struct pending_extent_op *)
670 (unsigned long)priv;
671 BUG_ON(extent_op->parent != orig_parent);
672 BUG_ON(extent_op->generation != orig_generation);
25179201 673
31840ae1
ZY		 674 extent_op->parent = parent;
675 extent_op->generation = ref_generation;
676 } else {
677 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
678 BUG_ON(!extent_op);
679
680 extent_op->type = PENDING_BACKREF_UPDATE;
681 extent_op->bytenr = bytenr;
682 extent_op->num_bytes = num_bytes;
683 extent_op->parent = parent;
684 extent_op->orig_parent = orig_parent;
685 extent_op->generation = ref_generation;
686 extent_op->orig_generation = orig_generation;
687 extent_op->level = (int)owner_objectid;
688
689 set_extent_bits(&root->fs_info->extent_ins,
690 bytenr, bytenr + num_bytes - 1,
25179201 691 EXTENT_WRITEBACK, GFP_NOFS);
31840ae1
ZY		 692 set_state_private(&root->fs_info->extent_ins,
693 bytenr, (unsigned long)extent_op);
694 }
25179201 695 mutex_unlock(&root->fs_info->extent_ins_mutex);
31840ae1
ZY		 696 return 0;
697 }
698
699 path = btrfs_alloc_path();
700 if (!path)
701 return -ENOMEM;
702 ret = lookup_extent_backref(trans, extent_root, path,
703 bytenr, orig_parent, orig_root,
3bb1a1bc 704 orig_generation, owner_objectid, 1);
31840ae1
ZY		 705 if (ret)
706 goto out;
707 ret = remove_extent_backref(trans, extent_root, path);
708 if (ret)
709 goto out;
710 ret = insert_extent_backref(trans, extent_root, path, bytenr,
711 parent, ref_root, ref_generation,
3bb1a1bc 712 owner_objectid);
31840ae1 713 BUG_ON(ret);
87ef2bb4
CM		 714 finish_current_insert(trans, extent_root, 0);
715 del_pending_extents(trans, extent_root, 0);
31840ae1
ZY		 716out:
717 btrfs_free_path(path);
718 return ret;
719}
720
721int btrfs_update_extent_ref(struct btrfs_trans_handle *trans,
722 struct btrfs_root *root, u64 bytenr,
723 u64 orig_parent, u64 parent,
724 u64 ref_root, u64 ref_generation,
3bb1a1bc 725 u64 owner_objectid)
31840ae1
ZY		 726{
727 int ret;
728 if (ref_root == BTRFS_TREE_LOG_OBJECTID &&
729 owner_objectid < BTRFS_FIRST_FREE_OBJECTID)
730 return 0;
31840ae1
ZY		 731 ret = __btrfs_update_extent_ref(trans, root, bytenr, orig_parent,
732 parent, ref_root, ref_root,
733 ref_generation, ref_generation,
3bb1a1bc 734 owner_objectid);
31840ae1
ZY		 735 return ret;
736}
737
925baedd 738static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
31840ae1
ZY		 739 struct btrfs_root *root, u64 bytenr,
740 u64 orig_parent, u64 parent,
741 u64 orig_root, u64 ref_root,
742 u64 orig_generation, u64 ref_generation,
3bb1a1bc 743 u64 owner_objectid)
02217ed2 744{
5caf2a00 745 struct btrfs_path *path;
02217ed2 746 int ret;
e2fa7227 747 struct btrfs_key key;
5f39d397 748 struct extent_buffer *l;
234b63a0 749 struct btrfs_extent_item *item;
cf27e1ee 750 u32 refs;
037e6390 751
5caf2a00 752 path = btrfs_alloc_path();
54aa1f4d
CM		 753 if (!path)
754 return -ENOMEM;
26b8003f 755
3c12ac72 756 path->reada = 1;
db94535d 757 key.objectid = bytenr;
31840ae1
ZY		 758 key.type = BTRFS_EXTENT_ITEM_KEY;
759 key.offset = (u64)-1;
760
5caf2a00 761 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path,
9f5fae2f 762 0, 1);
54aa1f4d
CM		 763 if (ret < 0)
764 return ret;
31840ae1
ZY		 765 BUG_ON(ret == 0 || path->slots[0] == 0);
766
767 path->slots[0]--;
5f39d397 768 l = path->nodes[0];
31840ae1
ZY		 769
770 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
771ed689
CM		 771 if (key.objectid != bytenr) {
772 btrfs_print_leaf(root->fs_info->extent_root, path->nodes[0]);
773 printk("wanted %Lu found %Lu\n", bytenr, key.objectid);
774 BUG();
775 }
31840ae1
ZY		 776 BUG_ON(key.type != BTRFS_EXTENT_ITEM_KEY);
777
5caf2a00 778 item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
5f39d397
CM		 779 refs = btrfs_extent_refs(l, item);
780 btrfs_set_extent_refs(l, item, refs + 1);
5caf2a00 781 btrfs_mark_buffer_dirty(path->nodes[0]);
a28ec197 782
5caf2a00 783 btrfs_release_path(root->fs_info->extent_root, path);
7bb86316 784
3c12ac72 785 path->reada = 1;
31840ae1
ZY		 786 ret = insert_extent_backref(trans, root->fs_info->extent_root,
787 path, bytenr, parent,
788 ref_root, ref_generation,
3bb1a1bc 789 owner_objectid);
7bb86316 790 BUG_ON(ret);
87ef2bb4
CM		 791 finish_current_insert(trans, root->fs_info->extent_root, 0);
792 del_pending_extents(trans, root->fs_info->extent_root, 0);
74493f7a
CM		 793
794 btrfs_free_path(path);
02217ed2
CM		 795 return 0;
796}
797
925baedd 798int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
31840ae1
ZY		 799 struct btrfs_root *root,
800 u64 bytenr, u64 num_bytes, u64 parent,
801 u64 ref_root, u64 ref_generation,
3bb1a1bc 802 u64 owner_objectid)
925baedd
CM		 803{
804 int ret;
31840ae1
ZY		 805 if (ref_root == BTRFS_TREE_LOG_OBJECTID &&
806 owner_objectid < BTRFS_FIRST_FREE_OBJECTID)
807 return 0;
31840ae1
ZY		 808 ret = __btrfs_inc_extent_ref(trans, root, bytenr, 0, parent,
809 0, ref_root, 0, ref_generation,
3bb1a1bc 810 owner_objectid);
925baedd
CM		 811 return ret;
812}
813
e9d0b13b
CM		 814int btrfs_extent_post_op(struct btrfs_trans_handle *trans,
815 struct btrfs_root *root)
816{
87ef2bb4
CM		 817 finish_current_insert(trans, root->fs_info->extent_root, 1);
818 del_pending_extents(trans, root->fs_info->extent_root, 1);
e9d0b13b
CM		 819 return 0;
820}
821
31840ae1
ZY		 822int btrfs_lookup_extent_ref(struct btrfs_trans_handle *trans,
823 struct btrfs_root *root, u64 bytenr,
824 u64 num_bytes, u32 *refs)
a28ec197 825{
5caf2a00 826 struct btrfs_path *path;
a28ec197 827 int ret;
e2fa7227 828 struct btrfs_key key;
5f39d397 829 struct extent_buffer *l;
234b63a0 830 struct btrfs_extent_item *item;
5caf2a00 831
db94535d 832 WARN_ON(num_bytes < root->sectorsize);
5caf2a00 833 path = btrfs_alloc_path();
3c12ac72 834 path->reada = 1;
db94535d
CM		 835 key.objectid = bytenr;
836 key.offset = num_bytes;
62e2749e 837 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
5caf2a00 838 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path,
9f5fae2f 839 0, 0);
54aa1f4d
CM		 840 if (ret < 0)
841 goto out;
5f39d397
CM		 842 if (ret != 0) {
843 btrfs_print_leaf(root, path->nodes[0]);
db94535d 844 printk("failed to find block number %Lu\n", bytenr);
a28ec197 845 BUG();
5f39d397
CM		 846 }
847 l = path->nodes[0];
5caf2a00 848 item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
5f39d397 849 *refs = btrfs_extent_refs(l, item);
54aa1f4d 850out:
5caf2a00 851 btrfs_free_path(path);
a28ec197
CM		 852 return 0;
853}
854
80ff3856
YZ		 855int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
856 struct btrfs_root *root, u64 bytenr)
be20aa9d
CM		 857{
858 struct btrfs_root *extent_root = root->fs_info->extent_root;
859 struct btrfs_path *path;
f321e491
YZ		 860 struct extent_buffer *leaf;
861 struct btrfs_extent_ref *ref_item;
862 struct btrfs_key key;
863 struct btrfs_key found_key;
80ff3856
YZ		 864 u64 ref_root;
865 u64 last_snapshot;
be20aa9d
CM		 866 u32 nritems;
867 int ret;
925baedd 868
be20aa9d 869 key.objectid = bytenr;
31840ae1 870 key.offset = (u64)-1;
f321e491 871 key.type = BTRFS_EXTENT_ITEM_KEY;
be20aa9d 872
f321e491 873 path = btrfs_alloc_path();
be20aa9d
CM		 874 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
875 if (ret < 0)
876 goto out;
877 BUG_ON(ret == 0);
80ff3856
YZ		 878
879 ret = -ENOENT;
880 if (path->slots[0] == 0)
31840ae1 881 goto out;
be20aa9d 882
31840ae1 883 path->slots[0]--;
f321e491
YZ		 884 leaf = path->nodes[0];
885 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
be20aa9d
CM		 886
887 if (found_key.objectid != bytenr ||
80ff3856 888 found_key.type != BTRFS_EXTENT_ITEM_KEY)
be20aa9d 889 goto out;
f321e491 890
80ff3856 891 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
be20aa9d 892 while (1) {
f321e491
YZ		 893 leaf = path->nodes[0];
894 nritems = btrfs_header_nritems(leaf);
be20aa9d
CM		 895 if (path->slots[0] >= nritems) {
896 ret = btrfs_next_leaf(extent_root, path);
f321e491
YZ		 897 if (ret < 0)
898 goto out;
be20aa9d
CM		 899 if (ret == 0)
900 continue;
901 break;
902 }
f321e491 903 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
be20aa9d
CM		 904 if (found_key.objectid != bytenr)
905 break;
bd09835d 906
be20aa9d
CM		 907 if (found_key.type != BTRFS_EXTENT_REF_KEY) {
908 path->slots[0]++;
909 continue;
910 }
911
f321e491 912 ref_item = btrfs_item_ptr(leaf, path->slots[0],
be20aa9d 913 struct btrfs_extent_ref);
80ff3856
YZ		 914 ref_root = btrfs_ref_root(leaf, ref_item);
915 if (ref_root != root->root_key.objectid &&
916 ref_root != BTRFS_TREE_LOG_OBJECTID) {
917 ret = 1;
f321e491 918 goto out;
80ff3856
YZ		 919 }
920 if (btrfs_ref_generation(leaf, ref_item) <= last_snapshot) {
f321e491
YZ		 921 ret = 1;
922 goto out;
923 }
80ff3856
YZ		 924
925 path->slots[0]++;
f321e491
YZ		 926 }
927 ret = 0;
be20aa9d 928out:
80ff3856 929 btrfs_free_path(path);
f321e491 930 return ret;
be20aa9d 931}
c5739bba 932
31840ae1
ZY		 933int btrfs_cache_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
934 struct extent_buffer *buf, u32 nr_extents)
02217ed2 935{
5f39d397 936 struct btrfs_key key;
6407bf6d 937 struct btrfs_file_extent_item *fi;
e4657689
ZY		 938 u64 root_gen;
939 u32 nritems;
02217ed2 940 int i;
db94535d 941 int level;
31840ae1 942 int ret = 0;
e4657689 943 int shared = 0;
a28ec197 944
3768f368 945 if (!root->ref_cows)
a28ec197 946 return 0;
5f39d397 947
e4657689
ZY		 948 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
949 shared = 0;
950 root_gen = root->root_key.offset;
951 } else {
952 shared = 1;
953 root_gen = trans->transid - 1;
954 }
955
db94535d 956 level = btrfs_header_level(buf);
5f39d397 957 nritems = btrfs_header_nritems(buf);
4a096752 958
31840ae1 959 if (level == 0) {
31153d81
YZ		 960 struct btrfs_leaf_ref *ref;
961 struct btrfs_extent_info *info;
962
31840ae1 963 ref = btrfs_alloc_leaf_ref(root, nr_extents);
31153d81 964 if (!ref) {
31840ae1 965 ret = -ENOMEM;
31153d81
YZ		 966 goto out;
967 }
968
e4657689 969 ref->root_gen = root_gen;
31153d81
YZ		 970 ref->bytenr = buf->start;
971 ref->owner = btrfs_header_owner(buf);
972 ref->generation = btrfs_header_generation(buf);
31840ae1 973 ref->nritems = nr_extents;
31153d81 974 info = ref->extents;
bcc63abb 975
31840ae1 976 for (i = 0; nr_extents > 0 && i < nritems; i++) {
31153d81
YZ		 977 u64 disk_bytenr;
978 btrfs_item_key_to_cpu(buf, &key, i);
979 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
980 continue;
981 fi = btrfs_item_ptr(buf, i,
982 struct btrfs_file_extent_item);
983 if (btrfs_file_extent_type(buf, fi) ==
984 BTRFS_FILE_EXTENT_INLINE)
985 continue;
986 disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
987 if (disk_bytenr == 0)
988 continue;
989
990 info->bytenr = disk_bytenr;
991 info->num_bytes =
992 btrfs_file_extent_disk_num_bytes(buf, fi);
993 info->objectid = key.objectid;
994 info->offset = key.offset;
995 info++;
996 }
997
e4657689 998 ret = btrfs_add_leaf_ref(root, ref, shared);
5b84e8d6
YZ		 999 if (ret == -EEXIST && shared) {
1000 struct btrfs_leaf_ref *old;
1001 old = btrfs_lookup_leaf_ref(root, ref->bytenr);
1002 BUG_ON(!old);
1003 btrfs_remove_leaf_ref(root, old);
1004 btrfs_free_leaf_ref(root, old);
1005 ret = btrfs_add_leaf_ref(root, ref, shared);
1006 }
31153d81 1007 WARN_ON(ret);
bcc63abb 1008 btrfs_free_leaf_ref(root, ref);
31153d81
YZ		 1009 }
1010out:
31840ae1
ZY		 1011 return ret;
1012}
1013
1014int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1015 struct extent_buffer *orig_buf, struct extent_buffer *buf,
1016 u32 *nr_extents)
1017{
1018 u64 bytenr;
1019 u64 ref_root;
1020 u64 orig_root;
1021 u64 ref_generation;
1022 u64 orig_generation;
1023 u32 nritems;
1024 u32 nr_file_extents = 0;
1025 struct btrfs_key key;
1026 struct btrfs_file_extent_item *fi;
1027 int i;
1028 int level;
1029 int ret = 0;
1030 int faili = 0;
1031 int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
3bb1a1bc 1032 u64, u64, u64, u64, u64, u64, u64, u64);
31840ae1
ZY		 1033
1034 ref_root = btrfs_header_owner(buf);
1035 ref_generation = btrfs_header_generation(buf);
1036 orig_root = btrfs_header_owner(orig_buf);
1037 orig_generation = btrfs_header_generation(orig_buf);
1038
1039 nritems = btrfs_header_nritems(buf);
1040 level = btrfs_header_level(buf);
1041
1042 if (root->ref_cows) {
1043 process_func = __btrfs_inc_extent_ref;
1044 } else {
1045 if (level == 0 &&
1046 root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
1047 goto out;
1048 if (level != 0 &&
1049 root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID)
1050 goto out;
1051 process_func = __btrfs_update_extent_ref;
1052 }
1053
1054 for (i = 0; i < nritems; i++) {
1055 cond_resched();
db94535d 1056 if (level == 0) {
5f39d397
CM		 1057 btrfs_item_key_to_cpu(buf, &key, i);
1058 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
54aa1f4d 1059 continue;
5f39d397 1060 fi = btrfs_item_ptr(buf, i,
54aa1f4d 1061 struct btrfs_file_extent_item);
5f39d397 1062 if (btrfs_file_extent_type(buf, fi) ==
54aa1f4d
CM		 1063 BTRFS_FILE_EXTENT_INLINE)
1064 continue;
31840ae1
ZY		 1065 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
1066 if (bytenr == 0)
54aa1f4d 1067 continue;
31840ae1
ZY		 1068
1069 nr_file_extents++;
1070
31840ae1
ZY		 1071 ret = process_func(trans, root, bytenr,
1072 orig_buf->start, buf->start,
1073 orig_root, ref_root,
1074 orig_generation, ref_generation,
3bb1a1bc 1075 key.objectid);
31840ae1
ZY		 1076
1077 if (ret) {
1078 faili = i;
1079 WARN_ON(1);
1080 goto fail;
1081 }
54aa1f4d 1082 } else {
db94535d 1083 bytenr = btrfs_node_blockptr(buf, i);
31840ae1
ZY		 1084 ret = process_func(trans, root, bytenr,
1085 orig_buf->start, buf->start,
1086 orig_root, ref_root,
1087 orig_generation, ref_generation,
3bb1a1bc 1088 level - 1);
31840ae1
ZY		 1089 if (ret) {
1090 faili = i;
1091 WARN_ON(1);
1092 goto fail;
1093 }
54aa1f4d
CM		 1094 }
1095 }
31840ae1
ZY		 1096out:
1097 if (nr_extents) {
1098 if (level == 0)
1099 *nr_extents = nr_file_extents;
1100 else
1101 *nr_extents = nritems;
1102 }
1103 return 0;
1104fail:
1105 WARN_ON(1);
54aa1f4d 1106 return ret;
02217ed2
CM		 1107}
1108
31840ae1
ZY		 1109int btrfs_update_ref(struct btrfs_trans_handle *trans,
1110 struct btrfs_root *root, struct extent_buffer *orig_buf,
1111 struct extent_buffer *buf, int start_slot, int nr)
1112
1113{
1114 u64 bytenr;
1115 u64 ref_root;
1116 u64 orig_root;
1117 u64 ref_generation;
1118 u64 orig_generation;
1119 struct btrfs_key key;
1120 struct btrfs_file_extent_item *fi;
1121 int i;
1122 int ret;
1123 int slot;
1124 int level;
1125
1126 BUG_ON(start_slot < 0);
1127 BUG_ON(start_slot + nr > btrfs_header_nritems(buf));
1128
1129 ref_root = btrfs_header_owner(buf);
1130 ref_generation = btrfs_header_generation(buf);
1131 orig_root = btrfs_header_owner(orig_buf);
1132 orig_generation = btrfs_header_generation(orig_buf);
1133 level = btrfs_header_level(buf);
1134
1135 if (!root->ref_cows) {
1136 if (level == 0 &&
1137 root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
1138 return 0;
1139 if (level != 0 &&
1140 root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID)
1141 return 0;
1142 }
1143
1144 for (i = 0, slot = start_slot; i < nr; i++, slot++) {
1145 cond_resched();
1146 if (level == 0) {
1147 btrfs_item_key_to_cpu(buf, &key, slot);
1148 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
1149 continue;
1150 fi = btrfs_item_ptr(buf, slot,
1151 struct btrfs_file_extent_item);
1152 if (btrfs_file_extent_type(buf, fi) ==
1153 BTRFS_FILE_EXTENT_INLINE)
1154 continue;
1155 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
1156 if (bytenr == 0)
1157 continue;
31840ae1
ZY		 1158 ret = __btrfs_update_extent_ref(trans, root, bytenr,
1159 orig_buf->start, buf->start,
1160 orig_root, ref_root,
1161 orig_generation, ref_generation,
3bb1a1bc 1162 key.objectid);
31840ae1
ZY		 1163 if (ret)
1164 goto fail;
1165 } else {
1166 bytenr = btrfs_node_blockptr(buf, slot);
31840ae1
ZY		 1167 ret = __btrfs_update_extent_ref(trans, root, bytenr,
1168 orig_buf->start, buf->start,
1169 orig_root, ref_root,
1170 orig_generation, ref_generation,
3bb1a1bc 1171 level - 1);
31840ae1
ZY		 1172 if (ret)
1173 goto fail;
1174 }
1175 }
1176 return 0;
1177fail:
1178 WARN_ON(1);
1179 return -1;
1180}
1181
9078a3e1
CM		 1182static int write_one_cache_group(struct btrfs_trans_handle *trans,
1183 struct btrfs_root *root,
1184 struct btrfs_path *path,
1185 struct btrfs_block_group_cache *cache)
1186{
1187 int ret;
1188 int pending_ret;
1189 struct btrfs_root *extent_root = root->fs_info->extent_root;
5f39d397
CM		 1190 unsigned long bi;
1191 struct extent_buffer *leaf;
9078a3e1 1192
9078a3e1 1193 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
54aa1f4d
CM		 1194 if (ret < 0)
1195 goto fail;
9078a3e1 1196 BUG_ON(ret);
5f39d397
CM		 1197
1198 leaf = path->nodes[0];
1199 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
1200 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
1201 btrfs_mark_buffer_dirty(leaf);
9078a3e1 1202 btrfs_release_path(extent_root, path);
54aa1f4d 1203fail:
87ef2bb4
CM		 1204 finish_current_insert(trans, extent_root, 0);
1205 pending_ret = del_pending_extents(trans, extent_root, 0);
9078a3e1
CM		 1206 if (ret)
1207 return ret;
1208 if (pending_ret)
1209 return pending_ret;
1210 return 0;
1211
1212}
1213
96b5179d
CM		 1214int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
1215 struct btrfs_root *root)
9078a3e1 1216{
0f9dd46c
JB		 1217 struct btrfs_block_group_cache *cache, *entry;
1218 struct rb_node *n;
9078a3e1
CM		 1219 int err = 0;
1220 int werr = 0;
9078a3e1 1221 struct btrfs_path *path;
96b5179d 1222 u64 last = 0;
9078a3e1
CM		 1223
1224 path = btrfs_alloc_path();
1225 if (!path)
1226 return -ENOMEM;
1227
1228 while(1) {
0f9dd46c
JB		 1229 cache = NULL;
1230 spin_lock(&root->fs_info->block_group_cache_lock);
1231 for (n = rb_first(&root->fs_info->block_group_cache_tree);
1232 n; n = rb_next(n)) {
1233 entry = rb_entry(n, struct btrfs_block_group_cache,
1234 cache_node);
1235 if (entry->dirty) {
1236 cache = entry;
1237 break;
1238 }
1239 }
1240 spin_unlock(&root->fs_info->block_group_cache_lock);
54aa1f4d 1241
0f9dd46c 1242 if (!cache)
96b5179d 1243 break;
0f9dd46c 1244
e8569813 1245 cache->dirty = 0;
0f9dd46c
JB		 1246 last += cache->key.offset;
1247
96b5179d
CM		 1248 err = write_one_cache_group(trans, root,
1249 path, cache);
1250 /*
1251 * if we fail to write the cache group, we want
1252 * to keep it marked dirty in hopes that a later
1253 * write will work
1254 */
1255 if (err) {
1256 werr = err;
1257 continue;
9078a3e1
CM		 1258 }
1259 }
1260 btrfs_free_path(path);
1261 return werr;
1262}
1263
593060d7
CM		 1264static int update_space_info(struct btrfs_fs_info *info, u64 flags,
1265 u64 total_bytes, u64 bytes_used,
1266 struct btrfs_space_info **space_info)
1267{
1268 struct btrfs_space_info *found;
1269
1270 found = __find_space_info(info, flags);
1271 if (found) {
25179201 1272 spin_lock(&found->lock);
593060d7
CM		 1273 found->total_bytes += total_bytes;
1274 found->bytes_used += bytes_used;
8f18cf13 1275 found->full = 0;
25179201 1276 spin_unlock(&found->lock);
593060d7
CM		 1277 *space_info = found;
1278 return 0;
1279 }
1280 found = kmalloc(sizeof(*found), GFP_NOFS);
1281 if (!found)
1282 return -ENOMEM;
1283
1284 list_add(&found->list, &info->space_info);
0f9dd46c 1285 INIT_LIST_HEAD(&found->block_groups);
80eb234a 1286 init_rwsem(&found->groups_sem);
0f9dd46c 1287 spin_lock_init(&found->lock);
593060d7
CM		 1288 found->flags = flags;
1289 found->total_bytes = total_bytes;
1290 found->bytes_used = bytes_used;
1291 found->bytes_pinned = 0;
e8569813 1292 found->bytes_reserved = 0;
593060d7 1293 found->full = 0;
0ef3e66b 1294 found->force_alloc = 0;
593060d7
CM		 1295 *space_info = found;
1296 return 0;
1297}
1298
8790d502
CM		 1299static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
1300{
1301 u64 extra_flags = flags & (BTRFS_BLOCK_GROUP_RAID0 |
611f0e00 1302 BTRFS_BLOCK_GROUP_RAID1 |
321aecc6 1303 BTRFS_BLOCK_GROUP_RAID10 |
611f0e00 1304 BTRFS_BLOCK_GROUP_DUP);
8790d502
CM		 1305 if (extra_flags) {
1306 if (flags & BTRFS_BLOCK_GROUP_DATA)
1307 fs_info->avail_data_alloc_bits |= extra_flags;
1308 if (flags & BTRFS_BLOCK_GROUP_METADATA)
1309 fs_info->avail_metadata_alloc_bits |= extra_flags;
1310 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
1311 fs_info->avail_system_alloc_bits |= extra_flags;
1312 }
1313}
593060d7 1314
a061fc8d 1315static u64 reduce_alloc_profile(struct btrfs_root *root, u64 flags)
ec44a35c 1316{
a061fc8d
CM		 1317 u64 num_devices = root->fs_info->fs_devices->num_devices;
1318
1319 if (num_devices == 1)
1320 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0);
1321 if (num_devices < 4)
1322 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
1323
ec44a35c
CM		 1324 if ((flags & BTRFS_BLOCK_GROUP_DUP) &&
1325 (flags & (BTRFS_BLOCK_GROUP_RAID1 |
a061fc8d 1326 BTRFS_BLOCK_GROUP_RAID10))) {
ec44a35c 1327 flags &= ~BTRFS_BLOCK_GROUP_DUP;
a061fc8d 1328 }
ec44a35c
CM		 1329
1330 if ((flags & BTRFS_BLOCK_GROUP_RAID1) &&
a061fc8d 1331 (flags & BTRFS_BLOCK_GROUP_RAID10)) {
ec44a35c 1332 flags &= ~BTRFS_BLOCK_GROUP_RAID1;
a061fc8d 1333 }
ec44a35c
CM		 1334
1335 if ((flags & BTRFS_BLOCK_GROUP_RAID0) &&
1336 ((flags & BTRFS_BLOCK_GROUP_RAID1) |
1337 (flags & BTRFS_BLOCK_GROUP_RAID10) |
1338 (flags & BTRFS_BLOCK_GROUP_DUP)))
1339 flags &= ~BTRFS_BLOCK_GROUP_RAID0;
1340 return flags;
1341}
1342
6324fbf3
CM		 1343static int do_chunk_alloc(struct btrfs_trans_handle *trans,
1344 struct btrfs_root *extent_root, u64 alloc_bytes,
0ef3e66b 1345 u64 flags, int force)
6324fbf3
CM		 1346{
1347 struct btrfs_space_info *space_info;
1348 u64 thresh;
1349 u64 start;
1350 u64 num_bytes;
cf749823 1351 int ret = 0, waited = 0;
6324fbf3 1352
a061fc8d 1353 flags = reduce_alloc_profile(extent_root, flags);
ec44a35c 1354
6324fbf3 1355 space_info = __find_space_info(extent_root->fs_info, flags);
593060d7
CM		 1356 if (!space_info) {
1357 ret = update_space_info(extent_root->fs_info, flags,
1358 0, 0, &space_info);
1359 BUG_ON(ret);
1360 }
6324fbf3
CM		 1361 BUG_ON(!space_info);
1362
25179201 1363 spin_lock(&space_info->lock);
0ef3e66b
CM		 1364 if (space_info->force_alloc) {
1365 force = 1;
1366 space_info->force_alloc = 0;
1367 }
25179201
JB		 1368 if (space_info->full) {
1369 spin_unlock(&space_info->lock);
925baedd 1370 goto out;
25179201 1371 }
6324fbf3 1372
8790d502 1373 thresh = div_factor(space_info->total_bytes, 6);
0ef3e66b 1374 if (!force &&
e8569813 1375 (space_info->bytes_used + space_info->bytes_pinned +
25179201
JB		 1376 space_info->bytes_reserved + alloc_bytes) < thresh) {
1377 spin_unlock(&space_info->lock);
925baedd 1378 goto out;
25179201 1379 }
6324fbf3 1380
25179201
JB		 1381 spin_unlock(&space_info->lock);
1382
1383 ret = mutex_trylock(&extent_root->fs_info->chunk_mutex);
1384 if (!ret && !force) {
1385 goto out;
1386 } else if (!ret) {
1387 mutex_lock(&extent_root->fs_info->chunk_mutex);
cf749823
JB		 1388 waited = 1;
1389 }
1390
25179201
JB		 1391 if (waited) {
1392 spin_lock(&space_info->lock);
1393 if (space_info->full) {
1394 spin_unlock(&space_info->lock);
1395 goto out_unlock;
1396 }
1397 spin_unlock(&space_info->lock);
1398 }
cf749823 1399
6324fbf3 1400 ret = btrfs_alloc_chunk(trans, extent_root, &start, &num_bytes, flags);
25179201 1401 if (ret) {
6324fbf3
CM		 1402printk("space info full %Lu\n", flags);
1403 space_info->full = 1;
a74a4b97 1404 goto out_unlock;
6324fbf3 1405 }
6324fbf3
CM		 1406
1407 ret = btrfs_make_block_group(trans, extent_root, 0, flags,
e17cade2 1408 BTRFS_FIRST_CHUNK_TREE_OBJECTID, start, num_bytes);
6324fbf3 1409 BUG_ON(ret);
a74a4b97 1410out_unlock:
333db94c 1411 mutex_unlock(&extent_root->fs_info->chunk_mutex);
a74a4b97 1412out:
0f9dd46c 1413 return ret;
6324fbf3
CM		 1414}
1415
9078a3e1
CM		 1416static int update_block_group(struct btrfs_trans_handle *trans,
1417 struct btrfs_root *root,
db94535d 1418 u64 bytenr, u64 num_bytes, int alloc,
0b86a832 1419 int mark_free)
9078a3e1
CM		 1420{
1421 struct btrfs_block_group_cache *cache;
1422 struct btrfs_fs_info *info = root->fs_info;
db94535d 1423 u64 total = num_bytes;
9078a3e1 1424 u64 old_val;
db94535d 1425 u64 byte_in_group;
3e1ad54f 1426
9078a3e1 1427 while(total) {
db94535d 1428 cache = btrfs_lookup_block_group(info, bytenr);
3e1ad54f 1429 if (!cache) {
9078a3e1 1430 return -1;
cd1bc465 1431 }
db94535d
CM		 1432 byte_in_group = bytenr - cache->key.objectid;
1433 WARN_ON(byte_in_group > cache->key.offset);
9078a3e1 1434
25179201 1435 spin_lock(&cache->space_info->lock);
c286ac48 1436 spin_lock(&cache->lock);
0f9dd46c 1437 cache->dirty = 1;
9078a3e1 1438 old_val = btrfs_block_group_used(&cache->item);
db94535d 1439 num_bytes = min(total, cache->key.offset - byte_in_group);
cd1bc465 1440 if (alloc) {
db94535d 1441 old_val += num_bytes;
6324fbf3 1442 cache->space_info->bytes_used += num_bytes;
c286ac48
CM		 1443 btrfs_set_block_group_used(&cache->item, old_val);
1444 spin_unlock(&cache->lock);
25179201 1445 spin_unlock(&cache->space_info->lock);
cd1bc465 1446 } else {
db94535d 1447 old_val -= num_bytes;
6324fbf3 1448 cache->space_info->bytes_used -= num_bytes;
c286ac48
CM		 1449 btrfs_set_block_group_used(&cache->item, old_val);
1450 spin_unlock(&cache->lock);
25179201 1451 spin_unlock(&cache->space_info->lock);
f510cfec 1452 if (mark_free) {
0f9dd46c
JB		 1453 int ret;
1454 ret = btrfs_add_free_space(cache, bytenr,
1455 num_bytes);
1456 if (ret)
1457 return -1;
e37c9e69 1458 }
cd1bc465 1459 }
db94535d
CM		 1460 total -= num_bytes;
1461 bytenr += num_bytes;
9078a3e1
CM		 1462 }
1463 return 0;
1464}
6324fbf3 1465
a061fc8d
CM		 1466static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
1467{
0f9dd46c
JB		 1468 struct btrfs_block_group_cache *cache;
1469
1470 cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
1471 if (!cache)
a061fc8d 1472 return 0;
0f9dd46c
JB		 1473
1474 return cache->key.objectid;
a061fc8d
CM		 1475}
1476
e02119d5 1477int btrfs_update_pinned_extents(struct btrfs_root *root,
324ae4df
Y		 1478 u64 bytenr, u64 num, int pin)
1479{
1480 u64 len;
1481 struct btrfs_block_group_cache *cache;
1482 struct btrfs_fs_info *fs_info = root->fs_info;
1483
25179201 1484 WARN_ON(!mutex_is_locked(&root->fs_info->pinned_mutex));
324ae4df
Y		 1485 if (pin) {
1486 set_extent_dirty(&fs_info->pinned_extents,
1487 bytenr, bytenr + num - 1, GFP_NOFS);
1488 } else {
1489 clear_extent_dirty(&fs_info->pinned_extents,
1490 bytenr, bytenr + num - 1, GFP_NOFS);
1491 }
1492 while (num > 0) {
1493 cache = btrfs_lookup_block_group(fs_info, bytenr);
e8569813
ZY		 1494 BUG_ON(!cache);
1495 len = min(num, cache->key.offset -
1496 (bytenr - cache->key.objectid));
324ae4df 1497 if (pin) {
25179201 1498 spin_lock(&cache->space_info->lock);
e8569813
ZY		 1499 spin_lock(&cache->lock);
1500 cache->pinned += len;
1501 cache->space_info->bytes_pinned += len;
1502 spin_unlock(&cache->lock);
25179201 1503 spin_unlock(&cache->space_info->lock);
324ae4df
Y		 1504 fs_info->total_pinned += len;
1505 } else {
25179201 1506 spin_lock(&cache->space_info->lock);
e8569813
ZY		 1507 spin_lock(&cache->lock);
1508 cache->pinned -= len;
1509 cache->space_info->bytes_pinned -= len;
1510 spin_unlock(&cache->lock);
25179201 1511 spin_unlock(&cache->space_info->lock);
324ae4df
Y		 1512 fs_info->total_pinned -= len;
1513 }
1514 bytenr += len;
1515 num -= len;
1516 }
1517 return 0;
1518}
9078a3e1 1519
e8569813
ZY		 1520static int update_reserved_extents(struct btrfs_root *root,
1521 u64 bytenr, u64 num, int reserve)
1522{
1523 u64 len;
1524 struct btrfs_block_group_cache *cache;
1525 struct btrfs_fs_info *fs_info = root->fs_info;
1526
e8569813
ZY		 1527 while (num > 0) {
1528 cache = btrfs_lookup_block_group(fs_info, bytenr);
1529 BUG_ON(!cache);
1530 len = min(num, cache->key.offset -
1531 (bytenr - cache->key.objectid));
25179201
JB		 1532
1533 spin_lock(&cache->space_info->lock);
1534 spin_lock(&cache->lock);
e8569813 1535 if (reserve) {
e8569813
ZY		 1536 cache->reserved += len;
1537 cache->space_info->bytes_reserved += len;
e8569813 1538 } else {
e8569813
ZY		 1539 cache->reserved -= len;
1540 cache->space_info->bytes_reserved -= len;
e8569813 1541 }
25179201
JB		 1542 spin_unlock(&cache->lock);
1543 spin_unlock(&cache->space_info->lock);
e8569813
ZY		 1544 bytenr += len;
1545 num -= len;
1546 }
1547 return 0;
1548}
1549
d1310b2e 1550int btrfs_copy_pinned(struct btrfs_root *root, struct extent_io_tree *copy)
ccd467d6 1551{
ccd467d6 1552 u64 last = 0;
1a5bc167
CM		 1553 u64 start;
1554 u64 end;
d1310b2e 1555 struct extent_io_tree *pinned_extents = &root->fs_info->pinned_extents;
ccd467d6 1556 int ret;
ccd467d6 1557
25179201 1558 mutex_lock(&root->fs_info->pinned_mutex);
ccd467d6 1559 while(1) {
1a5bc167
CM		 1560 ret = find_first_extent_bit(pinned_extents, last,
1561 &start, &end, EXTENT_DIRTY);
1562 if (ret)
ccd467d6 1563 break;
1a5bc167
CM		 1564 set_extent_dirty(copy, start, end, GFP_NOFS);
1565 last = end + 1;
ccd467d6 1566 }
25179201 1567 mutex_unlock(&root->fs_info->pinned_mutex);
ccd467d6
CM		 1568 return 0;
1569}
1570
1571int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
1572 struct btrfs_root *root,
d1310b2e 1573 struct extent_io_tree *unpin)
a28ec197 1574{
1a5bc167
CM		 1575 u64 start;
1576 u64 end;
a28ec197 1577 int ret;
0f9dd46c 1578 struct btrfs_block_group_cache *cache;
a28ec197 1579
25179201 1580 mutex_lock(&root->fs_info->pinned_mutex);
a28ec197 1581 while(1) {
1a5bc167
CM		 1582 ret = find_first_extent_bit(unpin, 0, &start, &end,
1583 EXTENT_DIRTY);
1584 if (ret)
a28ec197 1585 break;
e02119d5 1586 btrfs_update_pinned_extents(root, start, end + 1 - start, 0);
1a5bc167 1587 clear_extent_dirty(unpin, start, end, GFP_NOFS);
0f9dd46c
JB		 1588 cache = btrfs_lookup_block_group(root->fs_info, start);
1589 if (cache->cached)
1590 btrfs_add_free_space(cache, start, end - start + 1);
c286ac48 1591 if (need_resched()) {
25179201 1592 mutex_unlock(&root->fs_info->pinned_mutex);
c286ac48 1593 cond_resched();
25179201 1594 mutex_lock(&root->fs_info->pinned_mutex);
c286ac48 1595 }
a28ec197 1596 }
25179201 1597 mutex_unlock(&root->fs_info->pinned_mutex);
a28ec197
CM		 1598 return 0;
1599}
1600
98ed5174 1601static int finish_current_insert(struct btrfs_trans_handle *trans,
87ef2bb4 1602 struct btrfs_root *extent_root, int all)
037e6390 1603{
7bb86316
CM		 1604 u64 start;
1605 u64 end;
31840ae1 1606 u64 priv;
25179201 1607 u64 search = 0;
7bb86316
CM		 1608 struct btrfs_fs_info *info = extent_root->fs_info;
1609 struct btrfs_path *path;
31840ae1
ZY		 1610 struct btrfs_extent_ref *ref;
1611 struct pending_extent_op *extent_op;
1612 struct btrfs_key key;
234b63a0 1613 struct btrfs_extent_item extent_item;
037e6390 1614 int ret;
1a5bc167 1615 int err = 0;
037e6390 1616
5f39d397 1617 btrfs_set_stack_extent_refs(&extent_item, 1);
7bb86316 1618 path = btrfs_alloc_path();
037e6390 1619
26b8003f 1620 while(1) {
25179201
JB		 1621 mutex_lock(&info->extent_ins_mutex);
1622 ret = find_first_extent_bit(&info->extent_ins, search, &start,
1623 &end, EXTENT_WRITEBACK);
1624 if (ret) {
1625 mutex_unlock(&info->extent_ins_mutex);
87ef2bb4 1626 if (search && all) {
25179201
JB		 1627 search = 0;
1628 continue;
1629 }
26b8003f 1630 break;
25179201
JB		 1631 }
1632
1633 ret = try_lock_extent(&info->extent_ins, start, end, GFP_NOFS);
1634 if (!ret) {
87ef2bb4 1635 search = end + 1;
25179201
JB		 1636 mutex_unlock(&info->extent_ins_mutex);
1637 cond_resched();
1638 continue;
1639 }
1640 BUG_ON(ret < 0);
26b8003f 1641
31840ae1
ZY		 1642 ret = get_state_private(&info->extent_ins, start, &priv);
1643 BUG_ON(ret);
1644 extent_op = (struct pending_extent_op *)(unsigned long)priv;
1645
25179201
JB		 1646 mutex_unlock(&info->extent_ins_mutex);
1647
31840ae1
ZY		 1648 if (extent_op->type == PENDING_EXTENT_INSERT) {
1649 key.objectid = start;
1650 key.offset = end + 1 - start;
1651 key.type = BTRFS_EXTENT_ITEM_KEY;
1652 err = btrfs_insert_item(trans, extent_root, &key,
1a5bc167 1653 &extent_item, sizeof(extent_item));
31840ae1 1654 BUG_ON(err);
c286ac48 1655
25179201 1656 mutex_lock(&info->extent_ins_mutex);
31840ae1 1657 clear_extent_bits(&info->extent_ins, start, end,
25179201
JB		 1658 EXTENT_WRITEBACK, GFP_NOFS);
1659 mutex_unlock(&info->extent_ins_mutex);
c286ac48 1660
31840ae1
ZY		 1661 err = insert_extent_backref(trans, extent_root, path,
1662 start, extent_op->parent,
1663 extent_root->root_key.objectid,
1664 extent_op->generation,
3bb1a1bc 1665 extent_op->level);
31840ae1
ZY		 1666 BUG_ON(err);
1667 } else if (extent_op->type == PENDING_BACKREF_UPDATE) {
1668 err = lookup_extent_backref(trans, extent_root, path,
1669 start, extent_op->orig_parent,
1670 extent_root->root_key.objectid,
3bb1a1bc
YZ		 1671 extent_op->orig_generation,
1672 extent_op->level, 0);
31840ae1 1673 BUG_ON(err);
c286ac48 1674
25179201 1675 mutex_lock(&info->extent_ins_mutex);
31840ae1 1676 clear_extent_bits(&info->extent_ins, start, end,
25179201
JB		 1677 EXTENT_WRITEBACK, GFP_NOFS);
1678 mutex_unlock(&info->extent_ins_mutex);
31840ae1
ZY		 1679
1680 key.objectid = start;
1681 key.offset = extent_op->parent;
1682 key.type = BTRFS_EXTENT_REF_KEY;
1683 err = btrfs_set_item_key_safe(trans, extent_root, path,
1684 &key);
1685 BUG_ON(err);
1686 ref = btrfs_item_ptr(path->nodes[0], path->slots[0],
1687 struct btrfs_extent_ref);
1688 btrfs_set_ref_generation(path->nodes[0], ref,
1689 extent_op->generation);
1690 btrfs_mark_buffer_dirty(path->nodes[0]);
1691 btrfs_release_path(extent_root, path);
d8d5f3e1 1692 } else {
31840ae1 1693 BUG_ON(1);
d8d5f3e1 1694 }
31840ae1 1695 kfree(extent_op);
25179201 1696 unlock_extent(&info->extent_ins, start, end, GFP_NOFS);
87ef2bb4
CM		 1697 if (all)
1698 search = 0;
1699 else
1700 search = end + 1;
31840ae1 1701
25179201 1702 cond_resched();
037e6390 1703 }
7bb86316 1704 btrfs_free_path(path);
037e6390
CM		 1705 return 0;
1706}
1707
31840ae1
ZY		 1708static int pin_down_bytes(struct btrfs_trans_handle *trans,
1709 struct btrfs_root *root,
1710 u64 bytenr, u64 num_bytes, int is_data)
e20d96d6 1711{
1a5bc167 1712 int err = 0;
31840ae1 1713 struct extent_buffer *buf;
8ef97622 1714
31840ae1
ZY		 1715 if (is_data)
1716 goto pinit;
1717
1718 buf = btrfs_find_tree_block(root, bytenr, num_bytes);
1719 if (!buf)
1720 goto pinit;
1721
1722 /* we can reuse a block if it hasn't been written
1723 * and it is from this transaction. We can't
1724 * reuse anything from the tree log root because
1725 * it has tiny sub-transactions.
1726 */
1727 if (btrfs_buffer_uptodate(buf, 0) &&
1728 btrfs_try_tree_lock(buf)) {
1729 u64 header_owner = btrfs_header_owner(buf);
1730 u64 header_transid = btrfs_header_generation(buf);
1731 if (header_owner != BTRFS_TREE_LOG_OBJECTID &&
1a40e23b 1732 header_owner != BTRFS_TREE_RELOC_OBJECTID &&
31840ae1
ZY		 1733 header_transid == trans->transid &&
1734 !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
1735 clean_tree_block(NULL, root, buf);
1736 btrfs_tree_unlock(buf);
5f39d397 1737 free_extent_buffer(buf);
31840ae1 1738 return 1;
8ef97622 1739 }
31840ae1 1740 btrfs_tree_unlock(buf);
f4b9aa8d 1741 }
31840ae1
ZY		 1742 free_extent_buffer(buf);
1743pinit:
1744 btrfs_update_pinned_extents(root, bytenr, num_bytes, 1);
1745
be744175 1746 BUG_ON(err < 0);
e20d96d6
CM		 1747 return 0;
1748}
1749
fec577fb 1750/*
a28ec197 1751 * remove an extent from the root, returns 0 on success
fec577fb 1752 */
31840ae1
ZY		 1753static int __free_extent(struct btrfs_trans_handle *trans,
1754 struct btrfs_root *root,
1755 u64 bytenr, u64 num_bytes, u64 parent,
7bb86316 1756 u64 root_objectid, u64 ref_generation,
3bb1a1bc 1757 u64 owner_objectid, int pin, int mark_free)
a28ec197 1758{
5caf2a00 1759 struct btrfs_path *path;
e2fa7227 1760 struct btrfs_key key;
1261ec42
CM		 1761 struct btrfs_fs_info *info = root->fs_info;
1762 struct btrfs_root *extent_root = info->extent_root;
5f39d397 1763 struct extent_buffer *leaf;
a28ec197 1764 int ret;
952fccac
CM		 1765 int extent_slot = 0;
1766 int found_extent = 0;
1767 int num_to_del = 1;
234b63a0 1768 struct btrfs_extent_item *ei;
cf27e1ee 1769 u32 refs;
037e6390 1770
db94535d 1771 key.objectid = bytenr;
62e2749e 1772 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
db94535d 1773 key.offset = num_bytes;
5caf2a00 1774 path = btrfs_alloc_path();
54aa1f4d
CM		 1775 if (!path)
1776 return -ENOMEM;
5f26f772 1777
3c12ac72 1778 path->reada = 1;
3bb1a1bc
YZ		 1779 ret = lookup_extent_backref(trans, extent_root, path,
1780 bytenr, parent, root_objectid,
1781 ref_generation, owner_objectid, 1);
7bb86316 1782 if (ret == 0) {
952fccac
CM		 1783 struct btrfs_key found_key;
1784 extent_slot = path->slots[0];
1785 while(extent_slot > 0) {
1786 extent_slot--;
1787 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1788 extent_slot);
1789 if (found_key.objectid != bytenr)
1790 break;
1791 if (found_key.type == BTRFS_EXTENT_ITEM_KEY &&
1792 found_key.offset == num_bytes) {
1793 found_extent = 1;
1794 break;
1795 }
1796 if (path->slots[0] - extent_slot > 5)
1797 break;
1798 }
31840ae1
ZY		 1799 if (!found_extent) {
1800 ret = remove_extent_backref(trans, extent_root, path);
1801 BUG_ON(ret);
1802 btrfs_release_path(extent_root, path);
1803 ret = btrfs_search_slot(trans, extent_root,
1804 &key, path, -1, 1);
1805 BUG_ON(ret);
1806 extent_slot = path->slots[0];
1807 }
7bb86316
CM		 1808 } else {
1809 btrfs_print_leaf(extent_root, path->nodes[0]);
1810 WARN_ON(1);
1811 printk("Unable to find ref byte nr %Lu root %Lu "
3bb1a1bc
YZ		 1812 "gen %Lu owner %Lu\n", bytenr,
1813 root_objectid, ref_generation, owner_objectid);
7bb86316 1814 }
5f39d397
CM		 1815
1816 leaf = path->nodes[0];
952fccac 1817 ei = btrfs_item_ptr(leaf, extent_slot,
123abc88 1818 struct btrfs_extent_item);
5f39d397
CM		 1819 refs = btrfs_extent_refs(leaf, ei);
1820 BUG_ON(refs == 0);
1821 refs -= 1;
1822 btrfs_set_extent_refs(leaf, ei, refs);
952fccac 1823
5f39d397
CM		 1824 btrfs_mark_buffer_dirty(leaf);
1825
952fccac 1826 if (refs == 0 && found_extent && path->slots[0] == extent_slot + 1) {
31840ae1
ZY		 1827 struct btrfs_extent_ref *ref;
1828 ref = btrfs_item_ptr(leaf, path->slots[0],
1829 struct btrfs_extent_ref);
1830 BUG_ON(btrfs_ref_num_refs(leaf, ref) != 1);
952fccac
CM		 1831 /* if the back ref and the extent are next to each other
1832 * they get deleted below in one shot
1833 */
1834 path->slots[0] = extent_slot;
1835 num_to_del = 2;
1836 } else if (found_extent) {
1837 /* otherwise delete the extent back ref */
31840ae1 1838 ret = remove_extent_backref(trans, extent_root, path);
952fccac
CM		 1839 BUG_ON(ret);
1840 /* if refs are 0, we need to setup the path for deletion */
1841 if (refs == 0) {
1842 btrfs_release_path(extent_root, path);
1843 ret = btrfs_search_slot(trans, extent_root, &key, path,
1844 -1, 1);
952fccac
CM		 1845 BUG_ON(ret);
1846 }
1847 }
1848
cf27e1ee 1849 if (refs == 0) {
db94535d
CM		 1850 u64 super_used;
1851 u64 root_used;
21af804c
DW		 1852#ifdef BIO_RW_DISCARD
1853 u64 map_length = num_bytes;
1854 struct btrfs_multi_bio *multi = NULL;
1855#endif
78fae27e
CM		 1856
1857 if (pin) {
25179201 1858 mutex_lock(&root->fs_info->pinned_mutex);
31840ae1
ZY		 1859 ret = pin_down_bytes(trans, root, bytenr, num_bytes,
1860 owner_objectid >= BTRFS_FIRST_FREE_OBJECTID);
25179201 1861 mutex_unlock(&root->fs_info->pinned_mutex);
c549228f
Y		 1862 if (ret > 0)
1863 mark_free = 1;
1864 BUG_ON(ret < 0);
78fae27e
CM		 1865 }
1866
58176a96 1867 /* block accounting for super block */
a2135011 1868 spin_lock_irq(&info->delalloc_lock);
db94535d
CM		 1869 super_used = btrfs_super_bytes_used(&info->super_copy);
1870 btrfs_set_super_bytes_used(&info->super_copy,
1871 super_used - num_bytes);
a2135011 1872 spin_unlock_irq(&info->delalloc_lock);
58176a96
JB		 1873
1874 /* block accounting for root item */
db94535d 1875 root_used = btrfs_root_used(&root->root_item);
5f39d397 1876 btrfs_set_root_used(&root->root_item,
db94535d 1877 root_used - num_bytes);
952fccac
CM		 1878 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
1879 num_to_del);
31840ae1 1880 BUG_ON(ret);
25179201 1881 btrfs_release_path(extent_root, path);
db94535d 1882 ret = update_block_group(trans, root, bytenr, num_bytes, 0,
0b86a832 1883 mark_free);
9078a3e1 1884 BUG_ON(ret);
21af804c
DW		 1885
1886#ifdef BIO_RW_DISCARD
1887 /* Tell the block device(s) that the sectors can be discarded */
1888 ret = btrfs_map_block(&root->fs_info->mapping_tree, READ,
1889 bytenr, &map_length, &multi, 0);
1890 if (!ret) {
1891 struct btrfs_bio_stripe *stripe = multi->stripes;
1892 int i;
1893
1894 if (map_length > num_bytes)
1895 map_length = num_bytes;
1896
1897 for (i = 0; i < multi->num_stripes; i++, stripe++) {
1898 blkdev_issue_discard(stripe->dev->bdev,
1899 stripe->physical >> 9,
1900 map_length >> 9);
1901 }
1902 kfree(multi);
1903 }
1904#endif
a28ec197 1905 }
5caf2a00 1906 btrfs_free_path(path);
87ef2bb4 1907 finish_current_insert(trans, extent_root, 0);
a28ec197
CM		 1908 return ret;
1909}
1910
a28ec197
CM		 1911/*
1912 * find all the blocks marked as pending in the radix tree and remove
1913 * them from the extent map
1914 */
e089f05c 1915static int del_pending_extents(struct btrfs_trans_handle *trans, struct
87ef2bb4 1916 btrfs_root *extent_root, int all)
a28ec197
CM		 1917{
1918 int ret;
e20d96d6 1919 int err = 0;
1a5bc167
CM		 1920 u64 start;
1921 u64 end;
31840ae1 1922 u64 priv;
25179201 1923 u64 search = 0;
d1310b2e 1924 struct extent_io_tree *pending_del;
31840ae1
ZY		 1925 struct extent_io_tree *extent_ins;
1926 struct pending_extent_op *extent_op;
25179201 1927 struct btrfs_fs_info *info = extent_root->fs_info;
8ef97622 1928
31840ae1 1929 extent_ins = &extent_root->fs_info->extent_ins;
1a5bc167 1930 pending_del = &extent_root->fs_info->pending_del;
a28ec197
CM		 1931
1932 while(1) {
25179201
JB		 1933 mutex_lock(&info->extent_ins_mutex);
1934 ret = find_first_extent_bit(pending_del, search, &start, &end,
1935 EXTENT_WRITEBACK);
1936 if (ret) {
1937 mutex_unlock(&info->extent_ins_mutex);
87ef2bb4 1938 if (all && search) {
25179201
JB		 1939 search = 0;
1940 continue;
1941 }
a28ec197 1942 break;
25179201
JB		 1943 }
1944
1945 ret = try_lock_extent(extent_ins, start, end, GFP_NOFS);
1946 if (!ret) {
1947 search = end+1;
1948 mutex_unlock(&info->extent_ins_mutex);
1949 cond_resched();
1950 continue;
1951 }
1952 BUG_ON(ret < 0);
31840ae1
ZY		 1953
1954 ret = get_state_private(pending_del, start, &priv);
1955 BUG_ON(ret);
1956 extent_op = (struct pending_extent_op *)(unsigned long)priv;
1957
25179201 1958 clear_extent_bits(pending_del, start, end, EXTENT_WRITEBACK,
1a5bc167 1959 GFP_NOFS);
31840ae1 1960 if (!test_range_bit(extent_ins, start, end,
25179201
JB		 1961 EXTENT_WRITEBACK, 0)) {
1962 mutex_unlock(&info->extent_ins_mutex);
31840ae1 1963free_extent:
c286ac48 1964 ret = __free_extent(trans, extent_root,
31840ae1
ZY		 1965 start, end + 1 - start,
1966 extent_op->orig_parent,
1967 extent_root->root_key.objectid,
1968 extent_op->orig_generation,
25179201 1969 extent_op->level, 1, 0);
31840ae1 1970 kfree(extent_op);
c286ac48 1971 } else {
31840ae1 1972 kfree(extent_op);
25179201
JB		 1973
1974 ret = get_state_private(&info->extent_ins, start,
1975 &priv);
31840ae1
ZY		 1976 BUG_ON(ret);
1977 extent_op = (struct pending_extent_op *)
25179201
JB		 1978 (unsigned long)priv;
1979
1980 clear_extent_bits(&info->extent_ins, start, end,
1981 EXTENT_WRITEBACK, GFP_NOFS);
31840ae1 1982
25179201 1983 mutex_unlock(&info->extent_ins_mutex);
31840ae1
ZY		 1984
1985 if (extent_op->type == PENDING_BACKREF_UPDATE)
1986 goto free_extent;
1987
25179201
JB		 1988 mutex_lock(&extent_root->fs_info->pinned_mutex);
1989 ret = pin_down_bytes(trans, extent_root, start,
1990 end + 1 - start, 0);
1991 mutex_unlock(&extent_root->fs_info->pinned_mutex);
1992
31840ae1 1993 ret = update_block_group(trans, extent_root, start,
25179201
JB		 1994 end + 1 - start, 0, ret > 0);
1995
31840ae1
ZY		 1996 BUG_ON(ret);
1997 kfree(extent_op);
c286ac48 1998 }
1a5bc167
CM		 1999 if (ret)
2000 err = ret;
25179201 2001 unlock_extent(extent_ins, start, end, GFP_NOFS);
c286ac48 2002
87ef2bb4
CM		 2003 if (all)
2004 search = 0;
2005 else
2006 search = end + 1;
25179201 2007 cond_resched();
fec577fb 2008 }
e20d96d6 2009 return err;
fec577fb
CM		 2010}
2011
2012/*
2013 * remove an extent from the root, returns 0 on success
2014 */
925baedd 2015static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
31840ae1
ZY		 2016 struct btrfs_root *root,
2017 u64 bytenr, u64 num_bytes, u64 parent,
2018 u64 root_objectid, u64 ref_generation,
3bb1a1bc 2019 u64 owner_objectid, int pin)
fec577fb 2020{
9f5fae2f 2021 struct btrfs_root *extent_root = root->fs_info->extent_root;
fec577fb
CM		 2022 int pending_ret;
2023 int ret;
a28ec197 2024
db94535d 2025 WARN_ON(num_bytes < root->sectorsize);
fec577fb 2026 if (root == extent_root) {
31840ae1
ZY		 2027 struct pending_extent_op *extent_op;
2028
2029 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
2030 BUG_ON(!extent_op);
2031
2032 extent_op->type = PENDING_EXTENT_DELETE;
2033 extent_op->bytenr = bytenr;
2034 extent_op->num_bytes = num_bytes;
2035 extent_op->parent = parent;
2036 extent_op->orig_parent = parent;
2037 extent_op->generation = ref_generation;
2038 extent_op->orig_generation = ref_generation;
2039 extent_op->level = (int)owner_objectid;
2040
25179201 2041 mutex_lock(&root->fs_info->extent_ins_mutex);
31840ae1
ZY		 2042 set_extent_bits(&root->fs_info->pending_del,
2043 bytenr, bytenr + num_bytes - 1,
25179201 2044 EXTENT_WRITEBACK, GFP_NOFS);
31840ae1
ZY		 2045 set_state_private(&root->fs_info->pending_del,
2046 bytenr, (unsigned long)extent_op);
25179201 2047 mutex_unlock(&root->fs_info->extent_ins_mutex);
fec577fb
CM		 2048 return 0;
2049 }
4bef0848 2050 /* if metadata always pin */
d00aff00
CM		 2051 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) {
2052 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
0f9dd46c
JB		 2053 struct btrfs_block_group_cache *cache;
2054
d00aff00 2055 /* btrfs_free_reserved_extent */
0f9dd46c
JB		 2056 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
2057 BUG_ON(!cache);
2058 btrfs_add_free_space(cache, bytenr, num_bytes);
e8569813 2059 update_reserved_extents(root, bytenr, num_bytes, 0);
d00aff00
CM		 2060 return 0;
2061 }
4bef0848 2062 pin = 1;
d00aff00 2063 }
4bef0848
CM		 2064
2065 /* if data pin when any transaction has committed this */
2066 if (ref_generation != trans->transid)
2067 pin = 1;
2068
31840ae1 2069 ret = __free_extent(trans, root, bytenr, num_bytes, parent,
3bb1a1bc
YZ		 2070 root_objectid, ref_generation,
2071 owner_objectid, pin, pin == 0);
ee6e6504 2072
87ef2bb4
CM		 2073 finish_current_insert(trans, root->fs_info->extent_root, 0);
2074 pending_ret = del_pending_extents(trans, root->fs_info->extent_root, 0);
fec577fb
CM		 2075 return ret ? ret : pending_ret;
2076}
2077
925baedd 2078int btrfs_free_extent(struct btrfs_trans_handle *trans,
31840ae1
ZY		 2079 struct btrfs_root *root,
2080 u64 bytenr, u64 num_bytes, u64 parent,
2081 u64 root_objectid, u64 ref_generation,
3bb1a1bc 2082 u64 owner_objectid, int pin)
925baedd
CM		 2083{
2084 int ret;
2085
31840ae1 2086 ret = __btrfs_free_extent(trans, root, bytenr, num_bytes, parent,
925baedd 2087 root_objectid, ref_generation,
3bb1a1bc 2088 owner_objectid, pin);
925baedd
CM		 2089 return ret;
2090}
2091
87ee04eb
CM		 2092static u64 stripe_align(struct btrfs_root *root, u64 val)
2093{
2094 u64 mask = ((u64)root->stripesize - 1);
2095 u64 ret = (val + mask) & ~mask;
2096 return ret;
2097}
2098
fec577fb
CM		 2099/*
2100 * walks the btree of allocated extents and find a hole of a given size.
2101 * The key ins is changed to record the hole:
2102 * ins->objectid == block start
62e2749e 2103 * ins->flags = BTRFS_EXTENT_ITEM_KEY
fec577fb
CM		 2104 * ins->offset == number of blocks
2105 * Any available blocks before search_start are skipped.
2106 */
98ed5174
CM		 2107static int noinline find_free_extent(struct btrfs_trans_handle *trans,
2108 struct btrfs_root *orig_root,
2109 u64 num_bytes, u64 empty_size,
2110 u64 search_start, u64 search_end,
2111 u64 hint_byte, struct btrfs_key *ins,
2112 u64 exclude_start, u64 exclude_nr,
2113 int data)
fec577fb 2114{
80eb234a 2115 int ret = 0;
9f5fae2f 2116 struct btrfs_root * root = orig_root->fs_info->extent_root;
db94535d 2117 u64 total_needed = num_bytes;
239b14b3 2118 u64 *last_ptr = NULL;
4366211c 2119 u64 last_wanted = 0;
80eb234a 2120 struct btrfs_block_group_cache *block_group = NULL;
0ef3e66b 2121 int chunk_alloc_done = 0;
239b14b3 2122 int empty_cluster = 2 * 1024 * 1024;
0ef3e66b 2123 int allowed_chunk_alloc = 0;
80eb234a
JB		 2124 struct list_head *head = NULL, *cur = NULL;
2125 int loop = 0;
2126 struct btrfs_space_info *space_info;
fec577fb 2127
db94535d 2128 WARN_ON(num_bytes < root->sectorsize);
b1a4d965 2129 btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
80eb234a
JB		 2130 ins->objectid = 0;
2131 ins->offset = 0;
b1a4d965 2132
0ef3e66b
CM		 2133 if (orig_root->ref_cows || empty_size)
2134 allowed_chunk_alloc = 1;
2135
239b14b3
CM		 2136 if (data & BTRFS_BLOCK_GROUP_METADATA) {
2137 last_ptr = &root->fs_info->last_alloc;
4366211c 2138 empty_cluster = 64 * 1024;
239b14b3
CM		 2139 }
2140
0f9dd46c 2141 if ((data & BTRFS_BLOCK_GROUP_DATA) && btrfs_test_opt(root, SSD))
239b14b3 2142 last_ptr = &root->fs_info->last_data_alloc;
0f9dd46c 2143
239b14b3 2144 if (last_ptr) {
4366211c 2145 if (*last_ptr) {
239b14b3 2146 hint_byte = *last_ptr;
4366211c
CM		 2147 last_wanted = *last_ptr;
2148 } else
239b14b3 2149 empty_size += empty_cluster;
4366211c
CM		 2150 } else {
2151 empty_cluster = 0;
239b14b3 2152 }
a061fc8d 2153 search_start = max(search_start, first_logical_byte(root, 0));
239b14b3 2154 search_start = max(search_start, hint_byte);
0b86a832 2155
42e70e7a 2156 if (last_wanted && search_start != last_wanted) {
4366211c 2157 last_wanted = 0;
42e70e7a
CM		 2158 empty_size += empty_cluster;
2159 }
4366211c 2160
42e70e7a 2161 total_needed += empty_size;
80eb234a 2162 block_group = btrfs_lookup_block_group(root->fs_info, search_start);
d899e052
YZ		 2163 if (!block_group)
2164 block_group = btrfs_lookup_first_block_group(root->fs_info,
2165 search_start);
80eb234a 2166 space_info = __find_space_info(root->fs_info, data);
0f9dd46c 2167
80eb234a
JB		 2168 down_read(&space_info->groups_sem);
2169 while (1) {
2170 struct btrfs_free_space *free_space;
0f9dd46c 2171 /*
80eb234a
JB		 2172 * the only way this happens if our hint points to a block
2173 * group thats not of the proper type, while looping this
2174 * should never happen
0f9dd46c 2175 */
42e70e7a
CM		 2176 if (!block_group)
2177 goto new_group_no_lock;
2178
25179201 2179 mutex_lock(&block_group->alloc_mutex);
80eb234a 2180 if (unlikely(!block_group_bits(block_group, data)))
0f9dd46c 2181 goto new_group;
0f9dd46c 2182
80eb234a 2183 ret = cache_block_group(root, block_group);
25179201
JB		 2184 if (ret) {
2185 mutex_unlock(&block_group->alloc_mutex);
80eb234a 2186 break;
25179201 2187 }
0f9dd46c 2188
80eb234a
JB		 2189 if (block_group->ro)
2190 goto new_group;
0f9dd46c 2191
80eb234a
JB		 2192 free_space = btrfs_find_free_space(block_group, search_start,
2193 total_needed);
2194 if (free_space) {
2195 u64 start = block_group->key.objectid;
2196 u64 end = block_group->key.objectid +
0f9dd46c 2197 block_group->key.offset;
239b14b3 2198
80eb234a 2199 search_start = stripe_align(root, free_space->offset);
0f9dd46c 2200
80eb234a
JB		 2201 /* move on to the next group */
2202 if (search_start + num_bytes >= search_end)
2203 goto new_group;
e37c9e69 2204
80eb234a
JB		 2205 /* move on to the next group */
2206 if (search_start + num_bytes > end)
2207 goto new_group;
2208
4366211c 2209 if (last_wanted && search_start != last_wanted) {
3b7885bf 2210 total_needed += empty_cluster;
4366211c 2211 last_wanted = 0;
3b7885bf
CM		 2212 /*
2213 * if search_start is still in this block group
2214 * then we just re-search this block group
2215 */
2216 if (search_start >= start &&
2217 search_start < end) {
2218 mutex_unlock(&block_group->alloc_mutex);
2219 continue;
2220 }
2221
2222 /* else we go to the next block group */
2223 goto new_group;
2224 }
2225
80eb234a
JB		 2226 if (exclude_nr > 0 &&
2227 (search_start + num_bytes > exclude_start &&
2228 search_start < exclude_start + exclude_nr)) {
2229 search_start = exclude_start + exclude_nr;
2230 /*
2231 * if search_start is still in this block group
2232 * then we just re-search this block group
2233 */
2234 if (search_start >= start &&
25179201
JB		 2235 search_start < end) {
2236 mutex_unlock(&block_group->alloc_mutex);
4366211c 2237 last_wanted = 0;
80eb234a 2238 continue;
25179201 2239 }
80eb234a
JB		 2240
2241 /* else we go to the next block group */
2242 goto new_group;
2243 }
2244
2245 ins->objectid = search_start;
2246 ins->offset = num_bytes;
25179201
JB		 2247
2248 btrfs_remove_free_space_lock(block_group, search_start,
2249 num_bytes);
80eb234a 2250 /* we are all good, lets return */
25179201 2251 mutex_unlock(&block_group->alloc_mutex);
80eb234a 2252 break;
0f9dd46c 2253 }
80eb234a 2254new_group:
42e70e7a
CM		 2255 mutex_unlock(&block_group->alloc_mutex);
2256new_group_no_lock:
4366211c 2257 last_wanted = 0;
42e70e7a 2258 if (!allowed_chunk_alloc && loop > 0) {
4366211c
CM		 2259 total_needed -= empty_cluster;
2260 empty_cluster = 0;
2261 }
80eb234a
JB		 2262 /*
2263 * Here's how this works.
2264 * loop == 0: we were searching a block group via a hint
2265 * and didn't find anything, so we start at
2266 * the head of the block groups and keep searching
2267 * loop == 1: we're searching through all of the block groups
2268 * if we hit the head again we have searched
2269 * all of the block groups for this space and we
2270 * need to try and allocate, if we cant error out.
2271 * loop == 2: we allocated more space and are looping through
2272 * all of the block groups again.
2273 */
2274 if (loop == 0) {
2275 head = &space_info->block_groups;
2276 cur = head->next;
80eb234a
JB		 2277 loop++;
2278 } else if (loop == 1 && cur == head) {
42e70e7a
CM		 2279
2280 total_needed -= empty_cluster;
2281 empty_cluster = 0;
2282
80eb234a
JB		 2283 if (allowed_chunk_alloc && !chunk_alloc_done) {
2284 up_read(&space_info->groups_sem);
2285 ret = do_chunk_alloc(trans, root, num_bytes +
2286 2 * 1024 * 1024, data, 1);
2287 if (ret < 0)
2288 break;
2289 down_read(&space_info->groups_sem);
2290 loop++;
2291 head = &space_info->block_groups;
2292 cur = head->next;
2293 chunk_alloc_done = 1;
2294 } else if (!allowed_chunk_alloc) {
2295 space_info->force_alloc = 1;
2296 break;
2297 } else {
2298 break;
2299 }
2300 } else if (cur == head) {
2301 break;
0f9dd46c 2302 }
0b86a832 2303
80eb234a
JB		 2304 block_group = list_entry(cur, struct btrfs_block_group_cache,
2305 list);
2306 search_start = block_group->key.objectid;
2307 cur = cur->next;
f2654de4 2308 }
0b86a832 2309
80eb234a
JB		 2310 /* we found what we needed */
2311 if (ins->objectid) {
2312 if (!(data & BTRFS_BLOCK_GROUP_DATA))
2313 trans->block_group = block_group;
0f9dd46c 2314
80eb234a
JB		 2315 if (last_ptr)
2316 *last_ptr = ins->objectid + ins->offset;
2317 ret = 0;
2318 } else if (!ret) {
2319 ret = -ENOSPC;
be744175 2320 }
be744175 2321
80eb234a 2322 up_read(&space_info->groups_sem);
0f70abe2 2323 return ret;
fec577fb 2324}
ec44a35c 2325
0f9dd46c
JB		 2326static void dump_space_info(struct btrfs_space_info *info, u64 bytes)
2327{
2328 struct btrfs_block_group_cache *cache;
2329 struct list_head *l;
2330
2331 printk(KERN_INFO "space_info has %Lu free, is %sfull\n",
e8569813
ZY		 2332 info->total_bytes - info->bytes_used - info->bytes_pinned -
2333 info->bytes_reserved, (info->full) ? "" : "not ");
0f9dd46c 2334
80eb234a 2335 down_read(&info->groups_sem);
0f9dd46c
JB		 2336 list_for_each(l, &info->block_groups) {
2337 cache = list_entry(l, struct btrfs_block_group_cache, list);
2338 spin_lock(&cache->lock);
2339 printk(KERN_INFO "block group %Lu has %Lu bytes, %Lu used "
e8569813 2340 "%Lu pinned %Lu reserved\n",
0f9dd46c 2341 cache->key.objectid, cache->key.offset,
e8569813
ZY		 2342 btrfs_block_group_used(&cache->item),
2343 cache->pinned, cache->reserved);
0f9dd46c
JB		 2344 btrfs_dump_free_space(cache, bytes);
2345 spin_unlock(&cache->lock);
2346 }
80eb234a 2347 up_read(&info->groups_sem);
0f9dd46c 2348}
e8569813 2349
e6dcd2dc
CM		 2350static int __btrfs_reserve_extent(struct btrfs_trans_handle *trans,
2351 struct btrfs_root *root,
2352 u64 num_bytes, u64 min_alloc_size,
2353 u64 empty_size, u64 hint_byte,
2354 u64 search_end, struct btrfs_key *ins,
2355 u64 data)
fec577fb
CM		 2356{
2357 int ret;
fbdc762b 2358 u64 search_start = 0;
8790d502 2359 u64 alloc_profile;
1261ec42 2360 struct btrfs_fs_info *info = root->fs_info;
925baedd 2361
6324fbf3 2362 if (data) {
8790d502
CM		 2363 alloc_profile = info->avail_data_alloc_bits &
2364 info->data_alloc_profile;
2365 data = BTRFS_BLOCK_GROUP_DATA | alloc_profile;
6324fbf3 2366 } else if (root == root->fs_info->chunk_root) {
8790d502
CM		 2367 alloc_profile = info->avail_system_alloc_bits &
2368 info->system_alloc_profile;
2369 data = BTRFS_BLOCK_GROUP_SYSTEM | alloc_profile;
6324fbf3 2370 } else {
8790d502
CM		 2371 alloc_profile = info->avail_metadata_alloc_bits &
2372 info->metadata_alloc_profile;
2373 data = BTRFS_BLOCK_GROUP_METADATA | alloc_profile;
6324fbf3 2374 }
98d20f67 2375again:
a061fc8d 2376 data = reduce_alloc_profile(root, data);
0ef3e66b
CM		 2377 /*
2378 * the only place that sets empty_size is btrfs_realloc_node, which
2379 * is not called recursively on allocations
2380 */
2381 if (empty_size || root->ref_cows) {
593060d7 2382 if (!(data & BTRFS_BLOCK_GROUP_METADATA)) {
6324fbf3 2383 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
0ef3e66b
CM		 2384 2 * 1024 * 1024,
2385 BTRFS_BLOCK_GROUP_METADATA |
2386 (info->metadata_alloc_profile &
2387 info->avail_metadata_alloc_bits), 0);
6324fbf3
CM		 2388 }
2389 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
0ef3e66b 2390 num_bytes + 2 * 1024 * 1024, data, 0);
6324fbf3 2391 }
0b86a832 2392
db94535d
CM		 2393 WARN_ON(num_bytes < root->sectorsize);
2394 ret = find_free_extent(trans, root, num_bytes, empty_size,
2395 search_start, search_end, hint_byte, ins,
26b8003f
CM		 2396 trans->alloc_exclude_start,
2397 trans->alloc_exclude_nr, data);
3b951516 2398
98d20f67
CM		 2399 if (ret == -ENOSPC && num_bytes > min_alloc_size) {
2400 num_bytes = num_bytes >> 1;
0f9dd46c 2401 num_bytes = num_bytes & ~(root->sectorsize - 1);
98d20f67 2402 num_bytes = max(num_bytes, min_alloc_size);
0ef3e66b
CM		 2403 do_chunk_alloc(trans, root->fs_info->extent_root,
2404 num_bytes, data, 1);
98d20f67
CM		 2405 goto again;
2406 }
ec44a35c 2407 if (ret) {
0f9dd46c
JB		 2408 struct btrfs_space_info *sinfo;
2409
2410 sinfo = __find_space_info(root->fs_info, data);
2411 printk("allocation failed flags %Lu, wanted %Lu\n",
2412 data, num_bytes);
2413 dump_space_info(sinfo, num_bytes);
925baedd 2414 BUG();
925baedd 2415 }
0f9dd46c
JB		 2416
2417 return ret;
e6dcd2dc
CM		 2418}
2419
65b51a00
CM		 2420int btrfs_free_reserved_extent(struct btrfs_root *root, u64 start, u64 len)
2421{
0f9dd46c
JB		 2422 struct btrfs_block_group_cache *cache;
2423
0f9dd46c
JB		 2424 cache = btrfs_lookup_block_group(root->fs_info, start);
2425 if (!cache) {
2426 printk(KERN_ERR "Unable to find block group for %Lu\n", start);
0f9dd46c
JB		 2427 return -ENOSPC;
2428 }
2429 btrfs_add_free_space(cache, start, len);
1a40e23b 2430 update_reserved_extents(root, start, len, 0);
65b51a00
CM		 2431 return 0;
2432}
2433
e6dcd2dc
CM		 2434int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
2435 struct btrfs_root *root,
2436 u64 num_bytes, u64 min_alloc_size,
2437 u64 empty_size, u64 hint_byte,
2438 u64 search_end, struct btrfs_key *ins,
2439 u64 data)
2440{
2441 int ret;
e6dcd2dc
CM		 2442 ret = __btrfs_reserve_extent(trans, root, num_bytes, min_alloc_size,
2443 empty_size, hint_byte, search_end, ins,
2444 data);
e8569813 2445 update_reserved_extents(root, ins->objectid, ins->offset, 1);
e6dcd2dc
CM		 2446 return ret;
2447}
2448
2449static int __btrfs_alloc_reserved_extent(struct btrfs_trans_handle *trans,
31840ae1 2450 struct btrfs_root *root, u64 parent,
e6dcd2dc 2451 u64 root_objectid, u64 ref_generation,
3bb1a1bc 2452 u64 owner, struct btrfs_key *ins)
e6dcd2dc
CM		 2453{
2454 int ret;
2455 int pending_ret;
2456 u64 super_used;
2457 u64 root_used;
2458 u64 num_bytes = ins->offset;
2459 u32 sizes[2];
2460 struct btrfs_fs_info *info = root->fs_info;
2461 struct btrfs_root *extent_root = info->extent_root;
2462 struct btrfs_extent_item *extent_item;
2463 struct btrfs_extent_ref *ref;
2464 struct btrfs_path *path;
2465 struct btrfs_key keys[2];
fec577fb 2466
31840ae1
ZY		 2467 if (parent == 0)
2468 parent = ins->objectid;
2469
58176a96 2470 /* block accounting for super block */
a2135011 2471 spin_lock_irq(&info->delalloc_lock);
db94535d
CM		 2472 super_used = btrfs_super_bytes_used(&info->super_copy);
2473 btrfs_set_super_bytes_used(&info->super_copy, super_used + num_bytes);
a2135011 2474 spin_unlock_irq(&info->delalloc_lock);
26b8003f 2475
58176a96 2476 /* block accounting for root item */
db94535d
CM		 2477 root_used = btrfs_root_used(&root->root_item);
2478 btrfs_set_root_used(&root->root_item, root_used + num_bytes);
58176a96 2479
26b8003f 2480 if (root == extent_root) {
31840ae1
ZY		 2481 struct pending_extent_op *extent_op;
2482
2483 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
2484 BUG_ON(!extent_op);
2485
2486 extent_op->type = PENDING_EXTENT_INSERT;
2487 extent_op->bytenr = ins->objectid;
2488 extent_op->num_bytes = ins->offset;
2489 extent_op->parent = parent;
2490 extent_op->orig_parent = 0;
2491 extent_op->generation = ref_generation;
2492 extent_op->orig_generation = 0;
2493 extent_op->level = (int)owner;
2494
25179201 2495 mutex_lock(&root->fs_info->extent_ins_mutex);
1a5bc167
CM		 2496 set_extent_bits(&root->fs_info->extent_ins, ins->objectid,
2497 ins->objectid + ins->offset - 1,
25179201 2498 EXTENT_WRITEBACK, GFP_NOFS);
31840ae1
ZY		 2499 set_state_private(&root->fs_info->extent_ins,
2500 ins->objectid, (unsigned long)extent_op);
25179201 2501 mutex_unlock(&root->fs_info->extent_ins_mutex);
26b8003f
CM		 2502 goto update_block;
2503 }
2504
47e4bb98 2505 memcpy(&keys[0], ins, sizeof(*ins));
47e4bb98
CM		 2506 keys[1].objectid = ins->objectid;
2507 keys[1].type = BTRFS_EXTENT_REF_KEY;
31840ae1 2508 keys[1].offset = parent;
47e4bb98
CM		 2509 sizes[0] = sizeof(*extent_item);
2510 sizes[1] = sizeof(*ref);
7bb86316
CM		 2511
2512 path = btrfs_alloc_path();
2513 BUG_ON(!path);
47e4bb98
CM		 2514
2515 ret = btrfs_insert_empty_items(trans, extent_root, path, keys,
2516 sizes, 2);
ccd467d6 2517 BUG_ON(ret);
0f9dd46c 2518
47e4bb98
CM		 2519 extent_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
2520 struct btrfs_extent_item);
2521 btrfs_set_extent_refs(path->nodes[0], extent_item, 1);
2522 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
2523 struct btrfs_extent_ref);
2524
2525 btrfs_set_ref_root(path->nodes[0], ref, root_objectid);
2526 btrfs_set_ref_generation(path->nodes[0], ref, ref_generation);
2527 btrfs_set_ref_objectid(path->nodes[0], ref, owner);
31840ae1 2528 btrfs_set_ref_num_refs(path->nodes[0], ref, 1);
47e4bb98
CM		 2529
2530 btrfs_mark_buffer_dirty(path->nodes[0]);
2531
2532 trans->alloc_exclude_start = 0;
2533 trans->alloc_exclude_nr = 0;
7bb86316 2534 btrfs_free_path(path);
87ef2bb4
CM		 2535 finish_current_insert(trans, extent_root, 0);
2536 pending_ret = del_pending_extents(trans, extent_root, 0);
f510cfec 2537
925baedd
CM		 2538 if (ret)
2539 goto out;
e37c9e69 2540 if (pending_ret) {
925baedd
CM		 2541 ret = pending_ret;
2542 goto out;
e37c9e69 2543 }
26b8003f
CM		 2544
2545update_block:
0b86a832 2546 ret = update_block_group(trans, root, ins->objectid, ins->offset, 1, 0);
f5947066
CM		 2547 if (ret) {
2548 printk("update block group failed for %Lu %Lu\n",
2549 ins->objectid, ins->offset);
2550 BUG();
2551 }
925baedd 2552out:
e6dcd2dc
CM		 2553 return ret;
2554}
2555
2556int btrfs_alloc_reserved_extent(struct btrfs_trans_handle *trans,
31840ae1 2557 struct btrfs_root *root, u64 parent,
e6dcd2dc 2558 u64 root_objectid, u64 ref_generation,
3bb1a1bc 2559 u64 owner, struct btrfs_key *ins)
e6dcd2dc
CM		 2560{
2561 int ret;
1c2308f8
CM		 2562
2563 if (root_objectid == BTRFS_TREE_LOG_OBJECTID)
2564 return 0;
3bb1a1bc
YZ		 2565 ret = __btrfs_alloc_reserved_extent(trans, root, parent, root_objectid,
2566 ref_generation, owner, ins);
e8569813 2567 update_reserved_extents(root, ins->objectid, ins->offset, 0);
e6dcd2dc
CM		 2568 return ret;
2569}
e02119d5
CM		 2570
2571/*
2572 * this is used by the tree logging recovery code. It records that
2573 * an extent has been allocated and makes sure to clear the free
2574 * space cache bits as well
2575 */
2576int btrfs_alloc_logged_extent(struct btrfs_trans_handle *trans,
31840ae1 2577 struct btrfs_root *root, u64 parent,
e02119d5 2578 u64 root_objectid, u64 ref_generation,
3bb1a1bc 2579 u64 owner, struct btrfs_key *ins)
e02119d5
CM		 2580{
2581 int ret;
2582 struct btrfs_block_group_cache *block_group;
2583
e02119d5 2584 block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
25179201 2585 mutex_lock(&block_group->alloc_mutex);
e02119d5
CM		 2586 cache_block_group(root, block_group);
2587
25179201
JB		 2588 ret = btrfs_remove_free_space_lock(block_group, ins->objectid,
2589 ins->offset);
2590 mutex_unlock(&block_group->alloc_mutex);
0f9dd46c 2591 BUG_ON(ret);
3bb1a1bc
YZ		 2592 ret = __btrfs_alloc_reserved_extent(trans, root, parent, root_objectid,
2593 ref_generation, owner, ins);
e02119d5
CM		 2594 return ret;
2595}
2596
e6dcd2dc
CM		 2597/*
2598 * finds a free extent and does all the dirty work required for allocation
2599 * returns the key for the extent through ins, and a tree buffer for
2600 * the first block of the extent through buf.
2601 *
2602 * returns 0 if everything worked, non-zero otherwise.
2603 */
2604int btrfs_alloc_extent(struct btrfs_trans_handle *trans,
2605 struct btrfs_root *root,
31840ae1 2606 u64 num_bytes, u64 parent, u64 min_alloc_size,
e6dcd2dc 2607 u64 root_objectid, u64 ref_generation,
3bb1a1bc 2608 u64 owner_objectid, u64 empty_size, u64 hint_byte,
e6dcd2dc
CM		 2609 u64 search_end, struct btrfs_key *ins, u64 data)
2610{
2611 int ret;
2612
e6dcd2dc
CM		 2613 ret = __btrfs_reserve_extent(trans, root, num_bytes,
2614 min_alloc_size, empty_size, hint_byte,
2615 search_end, ins, data);
2616 BUG_ON(ret);
d00aff00 2617 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
31840ae1
ZY		 2618 ret = __btrfs_alloc_reserved_extent(trans, root, parent,
2619 root_objectid, ref_generation,
3bb1a1bc 2620 owner_objectid, ins);
d00aff00 2621 BUG_ON(ret);
e6dcd2dc 2622
e8569813
ZY		 2623 } else {
2624 update_reserved_extents(root, ins->objectid, ins->offset, 1);
d00aff00 2625 }
925baedd 2626 return ret;
fec577fb 2627}
65b51a00
CM		 2628
2629struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans,
2630 struct btrfs_root *root,
2631 u64 bytenr, u32 blocksize)
2632{
2633 struct extent_buffer *buf;
2634
2635 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
2636 if (!buf)
2637 return ERR_PTR(-ENOMEM);
2638 btrfs_set_header_generation(buf, trans->transid);
2639 btrfs_tree_lock(buf);
2640 clean_tree_block(trans, root, buf);
2641 btrfs_set_buffer_uptodate(buf);
d0c803c4
CM		 2642 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
2643 set_extent_dirty(&root->dirty_log_pages, buf->start,
2644 buf->start + buf->len - 1, GFP_NOFS);
2645 } else {
2646 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
65b51a00 2647 buf->start + buf->len - 1, GFP_NOFS);
d0c803c4 2648 }
65b51a00
CM		 2649 trans->blocks_used++;
2650 return buf;
2651}
2652
fec577fb
CM		 2653/*
2654 * helper function to allocate a block for a given tree
2655 * returns the tree buffer or NULL.
2656 */
5f39d397 2657struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
7bb86316 2658 struct btrfs_root *root,
31840ae1 2659 u32 blocksize, u64 parent,
7bb86316
CM		 2660 u64 root_objectid,
2661 u64 ref_generation,
7bb86316
CM		 2662 int level,
2663 u64 hint,
5f39d397 2664 u64 empty_size)
fec577fb 2665{
e2fa7227 2666 struct btrfs_key ins;
fec577fb 2667 int ret;
5f39d397 2668 struct extent_buffer *buf;
fec577fb 2669
31840ae1 2670 ret = btrfs_alloc_extent(trans, root, blocksize, parent, blocksize,
3bb1a1bc 2671 root_objectid, ref_generation, level,
31840ae1 2672 empty_size, hint, (u64)-1, &ins, 0);
fec577fb 2673 if (ret) {
54aa1f4d
CM		 2674 BUG_ON(ret > 0);
2675 return ERR_PTR(ret);
fec577fb 2676 }
55c69072 2677
65b51a00 2678 buf = btrfs_init_new_buffer(trans, root, ins.objectid, blocksize);
fec577fb
CM		 2679 return buf;
2680}
a28ec197 2681
e02119d5
CM		 2682int btrfs_drop_leaf_ref(struct btrfs_trans_handle *trans,
2683 struct btrfs_root *root, struct extent_buffer *leaf)
6407bf6d 2684{
7bb86316
CM		 2685 u64 leaf_owner;
2686 u64 leaf_generation;
5f39d397 2687 struct btrfs_key key;
6407bf6d
CM		 2688 struct btrfs_file_extent_item *fi;
2689 int i;
2690 int nritems;
2691 int ret;
2692
5f39d397
CM		 2693 BUG_ON(!btrfs_is_leaf(leaf));
2694 nritems = btrfs_header_nritems(leaf);
7bb86316
CM		 2695 leaf_owner = btrfs_header_owner(leaf);
2696 leaf_generation = btrfs_header_generation(leaf);
2697
6407bf6d 2698 for (i = 0; i < nritems; i++) {
db94535d 2699 u64 disk_bytenr;
e34a5b4f 2700 cond_resched();
5f39d397
CM		 2701
2702 btrfs_item_key_to_cpu(leaf, &key, i);
2703 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
6407bf6d
CM		 2704 continue;
2705 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
5f39d397
CM		 2706 if (btrfs_file_extent_type(leaf, fi) ==
2707 BTRFS_FILE_EXTENT_INLINE)
236454df 2708 continue;
6407bf6d
CM		 2709 /*
2710 * FIXME make sure to insert a trans record that
2711 * repeats the snapshot del on crash
2712 */
db94535d
CM		 2713 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
2714 if (disk_bytenr == 0)
3a686375 2715 continue;
4a096752 2716
925baedd 2717 ret = __btrfs_free_extent(trans, root, disk_bytenr,
7bb86316 2718 btrfs_file_extent_disk_num_bytes(leaf, fi),
31840ae1 2719 leaf->start, leaf_owner, leaf_generation,
3bb1a1bc 2720 key.objectid, 0);
31840ae1 2721 BUG_ON(ret);
2dd3e67b
CM		 2722
2723 atomic_inc(&root->fs_info->throttle_gen);
2724 wake_up(&root->fs_info->transaction_throttle);
2725 cond_resched();
6407bf6d
CM		 2726 }
2727 return 0;
2728}
2729
e02119d5
CM		 2730static int noinline cache_drop_leaf_ref(struct btrfs_trans_handle *trans,
2731 struct btrfs_root *root,
2732 struct btrfs_leaf_ref *ref)
31153d81
YZ		 2733{
2734 int i;
2735 int ret;
2736 struct btrfs_extent_info *info = ref->extents;
2737
31153d81 2738 for (i = 0; i < ref->nritems; i++) {
31840ae1
ZY		 2739 ret = __btrfs_free_extent(trans, root, info->bytenr,
2740 info->num_bytes, ref->bytenr,
2741 ref->owner, ref->generation,
3bb1a1bc 2742 info->objectid, 0);
2dd3e67b
CM		 2743
2744 atomic_inc(&root->fs_info->throttle_gen);
2745 wake_up(&root->fs_info->transaction_throttle);
2746 cond_resched();
2747
31153d81
YZ		 2748 BUG_ON(ret);
2749 info++;
2750 }
31153d81
YZ		 2751
2752 return 0;
2753}
2754
333db94c
CM		 2755int drop_snap_lookup_refcount(struct btrfs_root *root, u64 start, u64 len,
2756 u32 *refs)
2757{
017e5369 2758 int ret;
f87f057b 2759
31840ae1 2760 ret = btrfs_lookup_extent_ref(NULL, root, start, len, refs);
f87f057b
CM		 2761 BUG_ON(ret);
2762
2763#if 0 // some debugging code in case we see problems here
2764 /* if the refs count is one, it won't get increased again. But
2765 * if the ref count is > 1, someone may be decreasing it at
2766 * the same time we are.
2767 */
2768 if (*refs != 1) {
2769 struct extent_buffer *eb = NULL;
2770 eb = btrfs_find_create_tree_block(root, start, len);
2771 if (eb)
2772 btrfs_tree_lock(eb);
2773
2774 mutex_lock(&root->fs_info->alloc_mutex);
2775 ret = lookup_extent_ref(NULL, root, start, len, refs);
2776 BUG_ON(ret);
2777 mutex_unlock(&root->fs_info->alloc_mutex);
2778
2779 if (eb) {
2780 btrfs_tree_unlock(eb);
2781 free_extent_buffer(eb);
2782 }
2783 if (*refs == 1) {
2784 printk("block %llu went down to one during drop_snap\n",
2785 (unsigned long long)start);
2786 }
2787
2788 }
2789#endif
2790
e7a84565 2791 cond_resched();
017e5369 2792 return ret;
333db94c
CM		 2793}
2794
9aca1d51
CM		 2795/*
2796 * helper function for drop_snapshot, this walks down the tree dropping ref
2797 * counts as it goes.
2798 */
98ed5174
CM		 2799static int noinline walk_down_tree(struct btrfs_trans_handle *trans,
2800 struct btrfs_root *root,
2801 struct btrfs_path *path, int *level)
20524f02 2802{
7bb86316
CM		 2803 u64 root_owner;
2804 u64 root_gen;
2805 u64 bytenr;
ca7a79ad 2806 u64 ptr_gen;
5f39d397
CM		 2807 struct extent_buffer *next;
2808 struct extent_buffer *cur;
7bb86316 2809 struct extent_buffer *parent;
31153d81 2810 struct btrfs_leaf_ref *ref;
db94535d 2811 u32 blocksize;
20524f02
CM		 2812 int ret;
2813 u32 refs;
2814
5caf2a00
CM		 2815 WARN_ON(*level < 0);
2816 WARN_ON(*level >= BTRFS_MAX_LEVEL);
333db94c 2817 ret = drop_snap_lookup_refcount(root, path->nodes[*level]->start,
db94535d 2818 path->nodes[*level]->len, &refs);
20524f02
CM		 2819 BUG_ON(ret);
2820 if (refs > 1)
2821 goto out;
e011599b 2822
9aca1d51
CM		 2823 /*
2824 * walk down to the last node level and free all the leaves
2825 */
6407bf6d 2826 while(*level >= 0) {
5caf2a00
CM		 2827 WARN_ON(*level < 0);
2828 WARN_ON(*level >= BTRFS_MAX_LEVEL);
20524f02 2829 cur = path->nodes[*level];
e011599b 2830
5f39d397 2831 if (btrfs_header_level(cur) != *level)
2c90e5d6 2832 WARN_ON(1);
e011599b 2833
7518a238 2834 if (path->slots[*level] >=
5f39d397 2835 btrfs_header_nritems(cur))
20524f02 2836 break;
6407bf6d 2837 if (*level == 0) {
e02119d5 2838 ret = btrfs_drop_leaf_ref(trans, root, cur);
6407bf6d
CM		 2839 BUG_ON(ret);
2840 break;
2841 }
db94535d 2842 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
ca7a79ad 2843 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
db94535d 2844 blocksize = btrfs_level_size(root, *level - 1);
925baedd 2845
333db94c 2846 ret = drop_snap_lookup_refcount(root, bytenr, blocksize, &refs);
6407bf6d
CM		 2847 BUG_ON(ret);
2848 if (refs != 1) {
7bb86316
CM		 2849 parent = path->nodes[*level];
2850 root_owner = btrfs_header_owner(parent);
2851 root_gen = btrfs_header_generation(parent);
20524f02 2852 path->slots[*level]++;
f87f057b 2853
925baedd 2854 ret = __btrfs_free_extent(trans, root, bytenr,
31840ae1 2855 blocksize, parent->start,
3bb1a1bc
YZ		 2856 root_owner, root_gen,
2857 *level - 1, 1);
20524f02 2858 BUG_ON(ret);
18e35e0a
CM		 2859
2860 atomic_inc(&root->fs_info->throttle_gen);
2861 wake_up(&root->fs_info->transaction_throttle);
2dd3e67b 2862 cond_resched();
18e35e0a 2863
20524f02
CM		 2864 continue;
2865 }
f87f057b
CM		 2866 /*
2867 * at this point, we have a single ref, and since the
2868 * only place referencing this extent is a dead root
2869 * the reference count should never go higher.
2870 * So, we don't need to check it again
2871 */
31153d81 2872 if (*level == 1) {
017e5369 2873 ref = btrfs_lookup_leaf_ref(root, bytenr);
1a40e23b
ZY		 2874 if (ref && ref->generation != ptr_gen) {
2875 btrfs_free_leaf_ref(root, ref);
2876 ref = NULL;
2877 }
31153d81 2878 if (ref) {
e02119d5 2879 ret = cache_drop_leaf_ref(trans, root, ref);
31153d81
YZ		 2880 BUG_ON(ret);
2881 btrfs_remove_leaf_ref(root, ref);
bcc63abb 2882 btrfs_free_leaf_ref(root, ref);
31153d81
YZ		 2883 *level = 0;
2884 break;
2885 }
3bb1a1bc 2886 if (printk_ratelimit()) {
37d1aeee
CM		 2887 printk("leaf ref miss for bytenr %llu\n",
2888 (unsigned long long)bytenr);
3bb1a1bc 2889 }
31153d81 2890 }
db94535d 2891 next = btrfs_find_tree_block(root, bytenr, blocksize);
1259ab75 2892 if (!next || !btrfs_buffer_uptodate(next, ptr_gen)) {
5f39d397 2893 free_extent_buffer(next);
333db94c 2894
ca7a79ad
CM		 2895 next = read_tree_block(root, bytenr, blocksize,
2896 ptr_gen);
e7a84565 2897 cond_resched();
f87f057b
CM		 2898#if 0
2899 /*
2900 * this is a debugging check and can go away
2901 * the ref should never go all the way down to 1
2902 * at this point
2903 */
e6dcd2dc
CM		 2904 ret = lookup_extent_ref(NULL, root, bytenr, blocksize,
2905 &refs);
e9d0b13b 2906 BUG_ON(ret);
f87f057b
CM		 2907 WARN_ON(refs != 1);
2908#endif
e9d0b13b 2909 }
5caf2a00 2910 WARN_ON(*level <= 0);
83e15a28 2911 if (path->nodes[*level-1])
5f39d397 2912 free_extent_buffer(path->nodes[*level-1]);
20524f02 2913 path->nodes[*level-1] = next;
5f39d397 2914 *level = btrfs_header_level(next);
20524f02 2915 path->slots[*level] = 0;
2dd3e67b 2916 cond_resched();
20524f02
CM		 2917 }
2918out:
5caf2a00
CM		 2919 WARN_ON(*level < 0);
2920 WARN_ON(*level >= BTRFS_MAX_LEVEL);
7bb86316
CM		 2921
2922 if (path->nodes[*level] == root->node) {
7bb86316 2923 parent = path->nodes[*level];
31153d81 2924 bytenr = path->nodes[*level]->start;
7bb86316
CM		 2925 } else {
2926 parent = path->nodes[*level + 1];
31153d81 2927 bytenr = btrfs_node_blockptr(parent, path->slots[*level + 1]);
7bb86316
CM		 2928 }
2929
31153d81
YZ		 2930 blocksize = btrfs_level_size(root, *level);
2931 root_owner = btrfs_header_owner(parent);
7bb86316 2932 root_gen = btrfs_header_generation(parent);
31153d81
YZ		 2933
2934 ret = __btrfs_free_extent(trans, root, bytenr, blocksize,
31840ae1 2935 parent->start, root_owner, root_gen,
3bb1a1bc 2936 *level, 1);
5f39d397 2937 free_extent_buffer(path->nodes[*level]);
20524f02
CM		 2938 path->nodes[*level] = NULL;
2939 *level += 1;
2940 BUG_ON(ret);
f87f057b 2941
e7a84565 2942 cond_resched();
20524f02
CM		 2943 return 0;
2944}
2945
f82d02d9
YZ		 2946/*
2947 * helper function for drop_subtree, this function is similar to
2948 * walk_down_tree. The main difference is that it checks reference
2949 * counts while tree blocks are locked.
2950 */
2951static int noinline walk_down_subtree(struct btrfs_trans_handle *trans,
2952 struct btrfs_root *root,
2953 struct btrfs_path *path, int *level)
2954{
2955 struct extent_buffer *next;
2956 struct extent_buffer *cur;
2957 struct extent_buffer *parent;
2958 u64 bytenr;
2959 u64 ptr_gen;
2960 u32 blocksize;
2961 u32 refs;
2962 int ret;
2963
2964 cur = path->nodes[*level];
2965 ret = btrfs_lookup_extent_ref(trans, root, cur->start, cur->len,
2966 &refs);
2967 BUG_ON(ret);
2968 if (refs > 1)
2969 goto out;
2970
2971 while (*level >= 0) {
2972 cur = path->nodes[*level];
2973 if (*level == 0) {
2974 ret = btrfs_drop_leaf_ref(trans, root, cur);
2975 BUG_ON(ret);
2976 clean_tree_block(trans, root, cur);
2977 break;
2978 }
2979 if (path->slots[*level] >= btrfs_header_nritems(cur)) {
2980 clean_tree_block(trans, root, cur);
2981 break;
2982 }
2983
2984 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
2985 blocksize = btrfs_level_size(root, *level - 1);
2986 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
2987
2988 next = read_tree_block(root, bytenr, blocksize, ptr_gen);
2989 btrfs_tree_lock(next);
2990
2991 ret = btrfs_lookup_extent_ref(trans, root, bytenr, blocksize,
2992 &refs);
2993 BUG_ON(ret);
2994 if (refs > 1) {
2995 parent = path->nodes[*level];
2996 ret = btrfs_free_extent(trans, root, bytenr,
2997 blocksize, parent->start,
2998 btrfs_header_owner(parent),
2999 btrfs_header_generation(parent),
3000 *level - 1, 1);
3001 BUG_ON(ret);
3002 path->slots[*level]++;
3003 btrfs_tree_unlock(next);
3004 free_extent_buffer(next);
3005 continue;
3006 }
3007
3008 *level = btrfs_header_level(next);
3009 path->nodes[*level] = next;
3010 path->slots[*level] = 0;
3011 path->locks[*level] = 1;
3012 cond_resched();
3013 }
3014out:
3015 parent = path->nodes[*level + 1];
3016 bytenr = path->nodes[*level]->start;
3017 blocksize = path->nodes[*level]->len;
3018
3019 ret = btrfs_free_extent(trans, root, bytenr, blocksize,
3020 parent->start, btrfs_header_owner(parent),
3021 btrfs_header_generation(parent), *level, 1);
3022 BUG_ON(ret);
3023
3024 if (path->locks[*level]) {
3025 btrfs_tree_unlock(path->nodes[*level]);
3026 path->locks[*level] = 0;
3027 }
3028 free_extent_buffer(path->nodes[*level]);
3029 path->nodes[*level] = NULL;
3030 *level += 1;
3031 cond_resched();
3032 return 0;
3033}
3034
9aca1d51
CM		 3035/*
3036 * helper for dropping snapshots. This walks back up the tree in the path
3037 * to find the first node higher up where we haven't yet gone through
3038 * all the slots
3039 */
98ed5174
CM		 3040static int noinline walk_up_tree(struct btrfs_trans_handle *trans,
3041 struct btrfs_root *root,
f82d02d9
YZ		 3042 struct btrfs_path *path,
3043 int *level, int max_level)
20524f02 3044{
7bb86316
CM		 3045 u64 root_owner;
3046 u64 root_gen;
3047 struct btrfs_root_item *root_item = &root->root_item;
20524f02
CM		 3048 int i;
3049 int slot;
3050 int ret;
9f3a7427 3051
f82d02d9 3052 for (i = *level; i < max_level && path->nodes[i]; i++) {
20524f02 3053 slot = path->slots[i];
5f39d397
CM		 3054 if (slot < btrfs_header_nritems(path->nodes[i]) - 1) {
3055 struct extent_buffer *node;
3056 struct btrfs_disk_key disk_key;
3057 node = path->nodes[i];
20524f02
CM		 3058 path->slots[i]++;
3059 *level = i;
9f3a7427 3060 WARN_ON(*level == 0);
5f39d397 3061 btrfs_node_key(node, &disk_key, path->slots[i]);
9f3a7427 3062 memcpy(&root_item->drop_progress,
5f39d397 3063 &disk_key, sizeof(disk_key));
9f3a7427 3064 root_item->drop_level = i;
20524f02
CM		 3065 return 0;
3066 } else {
31840ae1
ZY		 3067 struct extent_buffer *parent;
3068 if (path->nodes[*level] == root->node)
3069 parent = path->nodes[*level];
3070 else
3071 parent = path->nodes[*level + 1];
3072
3073 root_owner = btrfs_header_owner(parent);
3074 root_gen = btrfs_header_generation(parent);
f82d02d9
YZ		 3075
3076 clean_tree_block(trans, root, path->nodes[*level]);
e089f05c 3077 ret = btrfs_free_extent(trans, root,
db94535d 3078 path->nodes[*level]->start,
7bb86316 3079 path->nodes[*level]->len,
3bb1a1bc
YZ		 3080 parent->start, root_owner,
3081 root_gen, *level, 1);
6407bf6d 3082 BUG_ON(ret);
f82d02d9
YZ		 3083 if (path->locks[*level]) {
3084 btrfs_tree_unlock(path->nodes[*level]);
3085 path->locks[*level] = 0;
3086 }
5f39d397 3087 free_extent_buffer(path->nodes[*level]);
83e15a28 3088 path->nodes[*level] = NULL;
20524f02 3089 *level = i + 1;
20524f02
CM		 3090 }
3091 }
3092 return 1;
3093}
3094
9aca1d51
CM		 3095/*
3096 * drop the reference count on the tree rooted at 'snap'. This traverses
3097 * the tree freeing any blocks that have a ref count of zero after being
3098 * decremented.
3099 */
e089f05c 3100int btrfs_drop_snapshot(struct btrfs_trans_handle *trans, struct btrfs_root
9f3a7427 3101 *root)
20524f02 3102{
3768f368 3103 int ret = 0;
9aca1d51 3104 int wret;
20524f02 3105 int level;
5caf2a00 3106 struct btrfs_path *path;
20524f02
CM		 3107 int i;
3108 int orig_level;
9f3a7427 3109 struct btrfs_root_item *root_item = &root->root_item;
20524f02 3110
a2135011 3111 WARN_ON(!mutex_is_locked(&root->fs_info->drop_mutex));
5caf2a00
CM		 3112 path = btrfs_alloc_path();
3113 BUG_ON(!path);
20524f02 3114
5f39d397 3115 level = btrfs_header_level(root->node);
20524f02 3116 orig_level = level;
9f3a7427
CM		 3117 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
3118 path->nodes[level] = root->node;
f510cfec 3119 extent_buffer_get(root->node);
9f3a7427
CM		 3120 path->slots[level] = 0;
3121 } else {
3122 struct btrfs_key key;
5f39d397
CM		 3123 struct btrfs_disk_key found_key;
3124 struct extent_buffer *node;
6702ed49 3125
9f3a7427 3126 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
6702ed49
CM		 3127 level = root_item->drop_level;
3128 path->lowest_level = level;
9f3a7427 3129 wret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6702ed49 3130 if (wret < 0) {
9f3a7427
CM		 3131 ret = wret;
3132 goto out;
3133 }
5f39d397
CM		 3134 node = path->nodes[level];
3135 btrfs_node_key(node, &found_key, path->slots[level]);
3136 WARN_ON(memcmp(&found_key, &root_item->drop_progress,
3137 sizeof(found_key)));
7d9eb12c
CM		 3138 /*
3139 * unlock our path, this is safe because only this
3140 * function is allowed to delete this snapshot
3141 */
925baedd
CM		 3142 for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
3143 if (path->nodes[i] && path->locks[i]) {
3144 path->locks[i] = 0;
3145 btrfs_tree_unlock(path->nodes[i]);
3146 }
3147 }
9f3a7427 3148 }
20524f02 3149 while(1) {
5caf2a00 3150 wret = walk_down_tree(trans, root, path, &level);
9aca1d51 3151 if (wret > 0)
20524f02 3152 break;
9aca1d51
CM		 3153 if (wret < 0)
3154 ret = wret;
3155
f82d02d9
YZ		 3156 wret = walk_up_tree(trans, root, path, &level,
3157 BTRFS_MAX_LEVEL);
9aca1d51 3158 if (wret > 0)
20524f02 3159 break;
9aca1d51
CM		 3160 if (wret < 0)
3161 ret = wret;
e7a84565
CM		 3162 if (trans->transaction->in_commit) {
3163 ret = -EAGAIN;
3164 break;
3165 }
18e35e0a 3166 atomic_inc(&root->fs_info->throttle_gen);
017e5369 3167 wake_up(&root->fs_info->transaction_throttle);
20524f02 3168 }
83e15a28 3169 for (i = 0; i <= orig_level; i++) {
5caf2a00 3170 if (path->nodes[i]) {
5f39d397 3171 free_extent_buffer(path->nodes[i]);
0f82731f 3172 path->nodes[i] = NULL;
83e15a28 3173 }
20524f02 3174 }
9f3a7427 3175out:
5caf2a00 3176 btrfs_free_path(path);
9aca1d51 3177 return ret;
20524f02 3178}
9078a3e1 3179
f82d02d9
YZ		 3180int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
3181 struct btrfs_root *root,
3182 struct extent_buffer *node,
3183 struct extent_buffer *parent)
3184{
3185 struct btrfs_path *path;
3186 int level;
3187 int parent_level;
3188 int ret = 0;
3189 int wret;
3190
3191 path = btrfs_alloc_path();
3192 BUG_ON(!path);
3193
3194 BUG_ON(!btrfs_tree_locked(parent));
3195 parent_level = btrfs_header_level(parent);
3196 extent_buffer_get(parent);
3197 path->nodes[parent_level] = parent;
3198 path->slots[parent_level] = btrfs_header_nritems(parent);
3199
3200 BUG_ON(!btrfs_tree_locked(node));
3201 level = btrfs_header_level(node);
3202 extent_buffer_get(node);
3203 path->nodes[level] = node;
3204 path->slots[level] = 0;
3205
3206 while (1) {
3207 wret = walk_down_subtree(trans, root, path, &level);
3208 if (wret < 0)
3209 ret = wret;
3210 if (wret != 0)
3211 break;
3212
3213 wret = walk_up_tree(trans, root, path, &level, parent_level);
3214 if (wret < 0)
3215 ret = wret;
3216 if (wret != 0)
3217 break;
3218 }
3219
3220 btrfs_free_path(path);
3221 return ret;
3222}
3223
8e7bf94f
CM		 3224static unsigned long calc_ra(unsigned long start, unsigned long last,
3225 unsigned long nr)
3226{
3227 return min(last, start + nr - 1);
3228}
3229
98ed5174
CM		 3230static int noinline relocate_inode_pages(struct inode *inode, u64 start,
3231 u64 len)
edbd8d4e
CM		 3232{
3233 u64 page_start;
3234 u64 page_end;
1a40e23b 3235 unsigned long first_index;
edbd8d4e 3236 unsigned long last_index;
edbd8d4e
CM		 3237 unsigned long i;
3238 struct page *page;
d1310b2e 3239 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4313b399 3240 struct file_ra_state *ra;
3eaa2885 3241 struct btrfs_ordered_extent *ordered;
1a40e23b
ZY		 3242 unsigned int total_read = 0;
3243 unsigned int total_dirty = 0;
3244 int ret = 0;
4313b399
CM		 3245
3246 ra = kzalloc(sizeof(*ra), GFP_NOFS);
edbd8d4e
CM		 3247
3248 mutex_lock(&inode->i_mutex);
1a40e23b 3249 first_index = start >> PAGE_CACHE_SHIFT;
edbd8d4e
CM		 3250 last_index = (start + len - 1) >> PAGE_CACHE_SHIFT;
3251
1a40e23b
ZY		 3252 /* make sure the dirty trick played by the caller work */
3253 ret = invalidate_inode_pages2_range(inode->i_mapping,
3254 first_index, last_index);
3255 if (ret)
3256 goto out_unlock;
8e7bf94f 3257
4313b399 3258 file_ra_state_init(ra, inode->i_mapping);
edbd8d4e 3259
1a40e23b
ZY		 3260 for (i = first_index ; i <= last_index; i++) {
3261 if (total_read % ra->ra_pages == 0) {
8e7bf94f 3262 btrfs_force_ra(inode->i_mapping, ra, NULL, i,
1a40e23b 3263 calc_ra(i, last_index, ra->ra_pages));
8e7bf94f
CM		 3264 }
3265 total_read++;
3eaa2885
CM		 3266again:
3267 if (((u64)i << PAGE_CACHE_SHIFT) > i_size_read(inode))
1a40e23b 3268 BUG_ON(1);
edbd8d4e 3269 page = grab_cache_page(inode->i_mapping, i);
a061fc8d 3270 if (!page) {
1a40e23b 3271 ret = -ENOMEM;
edbd8d4e 3272 goto out_unlock;
a061fc8d 3273 }
edbd8d4e
CM		 3274 if (!PageUptodate(page)) {
3275 btrfs_readpage(NULL, page);
3276 lock_page(page);
3277 if (!PageUptodate(page)) {
3278 unlock_page(page);
3279 page_cache_release(page);
1a40e23b 3280 ret = -EIO;
edbd8d4e
CM		 3281 goto out_unlock;
3282 }
3283 }
ec44a35c 3284 wait_on_page_writeback(page);
3eaa2885 3285
edbd8d4e
CM		 3286 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
3287 page_end = page_start + PAGE_CACHE_SIZE - 1;
d1310b2e 3288 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
edbd8d4e 3289
3eaa2885
CM		 3290 ordered = btrfs_lookup_ordered_extent(inode, page_start);
3291 if (ordered) {
3292 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
3293 unlock_page(page);
3294 page_cache_release(page);
3295 btrfs_start_ordered_extent(inode, ordered, 1);
3296 btrfs_put_ordered_extent(ordered);
3297 goto again;
3298 }
3299 set_page_extent_mapped(page);
3300
ea8c2819 3301 btrfs_set_extent_delalloc(inode, page_start, page_end);
1a40e23b
ZY		 3302 if (i == first_index)
3303 set_extent_bits(io_tree, page_start, page_end,
3304 EXTENT_BOUNDARY, GFP_NOFS);
3305
a061fc8d 3306 set_page_dirty(page);
1a40e23b 3307 total_dirty++;
edbd8d4e 3308
d1310b2e 3309 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
edbd8d4e
CM		 3310 unlock_page(page);
3311 page_cache_release(page);
3312 }
3313
3314out_unlock:
ec44a35c 3315 kfree(ra);
edbd8d4e 3316 mutex_unlock(&inode->i_mutex);
1a40e23b
ZY		 3317 balance_dirty_pages_ratelimited_nr(inode->i_mapping, total_dirty);
3318 return ret;
edbd8d4e
CM		 3319}
3320
1a40e23b
ZY		 3321static int noinline relocate_data_extent(struct inode *reloc_inode,
3322 struct btrfs_key *extent_key,
3323 u64 offset)
3324{
3325 struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
3326 struct extent_map_tree *em_tree = &BTRFS_I(reloc_inode)->extent_tree;
3327 struct extent_map *em;
6643558d
YZ		 3328 u64 start = extent_key->objectid - offset;
3329 u64 end = start + extent_key->offset - 1;
bf4ef679 3330
1a40e23b
ZY		 3331 em = alloc_extent_map(GFP_NOFS);
3332 BUG_ON(!em || IS_ERR(em));
bf4ef679 3333
6643558d 3334 em->start = start;
1a40e23b 3335 em->len = extent_key->offset;
c8b97818 3336 em->block_len = extent_key->offset;
1a40e23b
ZY		 3337 em->block_start = extent_key->objectid;
3338 em->bdev = root->fs_info->fs_devices->latest_bdev;
3339 set_bit(EXTENT_FLAG_PINNED, &em->flags);
3340
3341 /* setup extent map to cheat btrfs_readpage */
6643558d 3342 lock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
1a40e23b
ZY		 3343 while (1) {
3344 int ret;
3345 spin_lock(&em_tree->lock);
3346 ret = add_extent_mapping(em_tree, em);
3347 spin_unlock(&em_tree->lock);
3348 if (ret != -EEXIST) {
3349 free_extent_map(em);
bf4ef679
CM		 3350 break;
3351 }
6643558d 3352 btrfs_drop_extent_cache(reloc_inode, start, end, 0);
bf4ef679 3353 }
6643558d 3354 unlock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
bf4ef679 3355
6643558d 3356 return relocate_inode_pages(reloc_inode, start, extent_key->offset);
1a40e23b 3357}
edbd8d4e 3358
1a40e23b
ZY		 3359struct btrfs_ref_path {
3360 u64 extent_start;
3361 u64 nodes[BTRFS_MAX_LEVEL];
3362 u64 root_objectid;
3363 u64 root_generation;
3364 u64 owner_objectid;
1a40e23b
ZY		 3365 u32 num_refs;
3366 int lowest_level;
3367 int current_level;
f82d02d9
YZ		 3368 int shared_level;
3369
3370 struct btrfs_key node_keys[BTRFS_MAX_LEVEL];
3371 u64 new_nodes[BTRFS_MAX_LEVEL];
1a40e23b 3372};
7d9eb12c 3373
1a40e23b 3374struct disk_extent {
c8b97818 3375 u64 ram_bytes;
1a40e23b
ZY		 3376 u64 disk_bytenr;
3377 u64 disk_num_bytes;
3378 u64 offset;
3379 u64 num_bytes;
c8b97818
CM		 3380 u8 compression;
3381 u8 encryption;
3382 u16 other_encoding;
1a40e23b 3383};
4313b399 3384
1a40e23b
ZY		 3385static int is_cowonly_root(u64 root_objectid)
3386{
3387 if (root_objectid == BTRFS_ROOT_TREE_OBJECTID ||
3388 root_objectid == BTRFS_EXTENT_TREE_OBJECTID ||
3389 root_objectid == BTRFS_CHUNK_TREE_OBJECTID ||
3390 root_objectid == BTRFS_DEV_TREE_OBJECTID ||
3391 root_objectid == BTRFS_TREE_LOG_OBJECTID)
3392 return 1;
3393 return 0;
3394}
edbd8d4e 3395
1a40e23b
ZY		 3396static int noinline __next_ref_path(struct btrfs_trans_handle *trans,
3397 struct btrfs_root *extent_root,
3398 struct btrfs_ref_path *ref_path,
3399 int first_time)
3400{
3401 struct extent_buffer *leaf;
3402 struct btrfs_path *path;
3403 struct btrfs_extent_ref *ref;
3404 struct btrfs_key key;
3405 struct btrfs_key found_key;
3406 u64 bytenr;
3407 u32 nritems;
3408 int level;
3409 int ret = 1;
edbd8d4e 3410
1a40e23b
ZY		 3411 path = btrfs_alloc_path();
3412 if (!path)
3413 return -ENOMEM;
bf4ef679 3414
1a40e23b
ZY		 3415 if (first_time) {
3416 ref_path->lowest_level = -1;
3417 ref_path->current_level = -1;
f82d02d9 3418 ref_path->shared_level = -1;
1a40e23b
ZY		 3419 goto walk_up;
3420 }
3421walk_down:
3422 level = ref_path->current_level - 1;
3423 while (level >= -1) {
3424 u64 parent;
3425 if (level < ref_path->lowest_level)
3426 break;
bf4ef679 3427
1a40e23b
ZY		 3428 if (level >= 0) {
3429 bytenr = ref_path->nodes[level];
3430 } else {
3431 bytenr = ref_path->extent_start;
3432 }
3433 BUG_ON(bytenr == 0);
bf4ef679 3434
1a40e23b
ZY		 3435 parent = ref_path->nodes[level + 1];
3436 ref_path->nodes[level + 1] = 0;
3437 ref_path->current_level = level;
3438 BUG_ON(parent == 0);
0ef3e66b 3439
1a40e23b
ZY		 3440 key.objectid = bytenr;
3441 key.offset = parent + 1;
3442 key.type = BTRFS_EXTENT_REF_KEY;
edbd8d4e 3443
1a40e23b
ZY		 3444 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
3445 if (ret < 0)
edbd8d4e 3446 goto out;
1a40e23b 3447 BUG_ON(ret == 0);
7d9eb12c 3448
1a40e23b
ZY		 3449 leaf = path->nodes[0];
3450 nritems = btrfs_header_nritems(leaf);
3451 if (path->slots[0] >= nritems) {
3452 ret = btrfs_next_leaf(extent_root, path);
3453 if (ret < 0)
3454 goto out;
3455 if (ret > 0)
3456 goto next;
3457 leaf = path->nodes[0];
3458 }
0ef3e66b 3459
1a40e23b
ZY		 3460 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3461 if (found_key.objectid == bytenr &&
f82d02d9
YZ		 3462 found_key.type == BTRFS_EXTENT_REF_KEY) {
3463 if (level < ref_path->shared_level)
3464 ref_path->shared_level = level;
1a40e23b 3465 goto found;
f82d02d9 3466 }
1a40e23b
ZY		 3467next:
3468 level--;
3469 btrfs_release_path(extent_root, path);
d899e052 3470 cond_resched();
1a40e23b
ZY		 3471 }
3472 /* reached lowest level */
3473 ret = 1;
3474 goto out;
3475walk_up:
3476 level = ref_path->current_level;
3477 while (level < BTRFS_MAX_LEVEL - 1) {
3478 u64 ref_objectid;
3479 if (level >= 0) {
3480 bytenr = ref_path->nodes[level];
3481 } else {
3482 bytenr = ref_path->extent_start;
3483 }
3484 BUG_ON(bytenr == 0);
edbd8d4e 3485
1a40e23b
ZY		 3486 key.objectid = bytenr;
3487 key.offset = 0;
3488 key.type = BTRFS_EXTENT_REF_KEY;
edbd8d4e 3489
1a40e23b
ZY		 3490 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
3491 if (ret < 0)
3492 goto out;
edbd8d4e 3493
1a40e23b
ZY		 3494 leaf = path->nodes[0];
3495 nritems = btrfs_header_nritems(leaf);
3496 if (path->slots[0] >= nritems) {
3497 ret = btrfs_next_leaf(extent_root, path);
3498 if (ret < 0)
3499 goto out;
3500 if (ret > 0) {
3501 /* the extent was freed by someone */
3502 if (ref_path->lowest_level == level)
3503 goto out;
3504 btrfs_release_path(extent_root, path);
3505 goto walk_down;
3506 }
3507 leaf = path->nodes[0];
3508 }
edbd8d4e 3509
1a40e23b
ZY		 3510 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3511 if (found_key.objectid != bytenr ||
3512 found_key.type != BTRFS_EXTENT_REF_KEY) {
3513 /* the extent was freed by someone */
3514 if (ref_path->lowest_level == level) {
3515 ret = 1;
3516 goto out;
3517 }
3518 btrfs_release_path(extent_root, path);
3519 goto walk_down;
3520 }
3521found:
3522 ref = btrfs_item_ptr(leaf, path->slots[0],
3523 struct btrfs_extent_ref);
3524 ref_objectid = btrfs_ref_objectid(leaf, ref);
3525 if (ref_objectid < BTRFS_FIRST_FREE_OBJECTID) {
3526 if (first_time) {
3527 level = (int)ref_objectid;
3528 BUG_ON(level >= BTRFS_MAX_LEVEL);
3529 ref_path->lowest_level = level;
3530 ref_path->current_level = level;
3531 ref_path->nodes[level] = bytenr;
3532 } else {
3533 WARN_ON(ref_objectid != level);
3534 }
3535 } else {
3536 WARN_ON(level != -1);
3537 }
3538 first_time = 0;
bf4ef679 3539
1a40e23b
ZY		 3540 if (ref_path->lowest_level == level) {
3541 ref_path->owner_objectid = ref_objectid;
1a40e23b
ZY		 3542 ref_path->num_refs = btrfs_ref_num_refs(leaf, ref);
3543 }
bf4ef679 3544
7d9eb12c 3545 /*
1a40e23b
ZY		 3546 * the block is tree root or the block isn't in reference
3547 * counted tree.
7d9eb12c 3548 */
1a40e23b
ZY		 3549 if (found_key.objectid == found_key.offset ||
3550 is_cowonly_root(btrfs_ref_root(leaf, ref))) {
3551 ref_path->root_objectid = btrfs_ref_root(leaf, ref);
3552 ref_path->root_generation =
3553 btrfs_ref_generation(leaf, ref);
3554 if (level < 0) {
3555 /* special reference from the tree log */
3556 ref_path->nodes[0] = found_key.offset;
3557 ref_path->current_level = 0;
3558 }
3559 ret = 0;
3560 goto out;
3561 }
7d9eb12c 3562
1a40e23b
ZY		 3563 level++;
3564 BUG_ON(ref_path->nodes[level] != 0);
3565 ref_path->nodes[level] = found_key.offset;
3566 ref_path->current_level = level;
bf4ef679 3567
1a40e23b
ZY		 3568 /*
3569 * the reference was created in the running transaction,
3570 * no need to continue walking up.
3571 */
3572 if (btrfs_ref_generation(leaf, ref) == trans->transid) {
3573 ref_path->root_objectid = btrfs_ref_root(leaf, ref);
3574 ref_path->root_generation =
3575 btrfs_ref_generation(leaf, ref);
3576 ret = 0;
3577 goto out;
7d9eb12c
CM		 3578 }
3579
1a40e23b 3580 btrfs_release_path(extent_root, path);
d899e052 3581 cond_resched();
7d9eb12c 3582 }
1a40e23b
ZY		 3583 /* reached max tree level, but no tree root found. */
3584 BUG();
edbd8d4e 3585out:
1a40e23b
ZY		 3586 btrfs_free_path(path);
3587 return ret;
edbd8d4e
CM		 3588}
3589
1a40e23b
ZY		 3590static int btrfs_first_ref_path(struct btrfs_trans_handle *trans,
3591 struct btrfs_root *extent_root,
3592 struct btrfs_ref_path *ref_path,
3593 u64 extent_start)
a061fc8d 3594{
1a40e23b
ZY		 3595 memset(ref_path, 0, sizeof(*ref_path));
3596 ref_path->extent_start = extent_start;
a061fc8d 3597
1a40e23b 3598 return __next_ref_path(trans, extent_root, ref_path, 1);
a061fc8d
CM		 3599}
3600
1a40e23b
ZY		 3601static int btrfs_next_ref_path(struct btrfs_trans_handle *trans,
3602 struct btrfs_root *extent_root,
3603 struct btrfs_ref_path *ref_path)
edbd8d4e 3604{
1a40e23b
ZY		 3605 return __next_ref_path(trans, extent_root, ref_path, 0);
3606}
3607
3608static int noinline get_new_locations(struct inode *reloc_inode,
3609 struct btrfs_key *extent_key,
3610 u64 offset, int no_fragment,
3611 struct disk_extent **extents,
3612 int *nr_extents)
3613{
3614 struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
3615 struct btrfs_path *path;
3616 struct btrfs_file_extent_item *fi;
edbd8d4e 3617 struct extent_buffer *leaf;
1a40e23b
ZY		 3618 struct disk_extent *exts = *extents;
3619 struct btrfs_key found_key;
3620 u64 cur_pos;
3621 u64 last_byte;
edbd8d4e 3622 u32 nritems;
1a40e23b
ZY		 3623 int nr = 0;
3624 int max = *nr_extents;
3625 int ret;
edbd8d4e 3626
1a40e23b
ZY		 3627 WARN_ON(!no_fragment && *extents);
3628 if (!exts) {
3629 max = 1;
3630 exts = kmalloc(sizeof(*exts) * max, GFP_NOFS);
3631 if (!exts)
3632 return -ENOMEM;
a061fc8d 3633 }
edbd8d4e 3634
1a40e23b
ZY		 3635 path = btrfs_alloc_path();
3636 BUG_ON(!path);
edbd8d4e 3637
1a40e23b
ZY		 3638 cur_pos = extent_key->objectid - offset;
3639 last_byte = extent_key->objectid + extent_key->offset;
3640 ret = btrfs_lookup_file_extent(NULL, root, path, reloc_inode->i_ino,
3641 cur_pos, 0);
3642 if (ret < 0)
3643 goto out;
3644 if (ret > 0) {
3645 ret = -ENOENT;
3646 goto out;
3647 }
edbd8d4e 3648
1a40e23b 3649 while (1) {
edbd8d4e
CM		 3650 leaf = path->nodes[0];
3651 nritems = btrfs_header_nritems(leaf);
1a40e23b
ZY		 3652 if (path->slots[0] >= nritems) {
3653 ret = btrfs_next_leaf(root, path);
a061fc8d
CM		 3654 if (ret < 0)
3655 goto out;
1a40e23b
ZY		 3656 if (ret > 0)
3657 break;
bf4ef679 3658 leaf = path->nodes[0];
a061fc8d 3659 }
edbd8d4e
CM		 3660
3661 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1a40e23b
ZY		 3662 if (found_key.offset != cur_pos ||
3663 found_key.type != BTRFS_EXTENT_DATA_KEY ||
3664 found_key.objectid != reloc_inode->i_ino)
edbd8d4e
CM		 3665 break;
3666
1a40e23b
ZY		 3667 fi = btrfs_item_ptr(leaf, path->slots[0],
3668 struct btrfs_file_extent_item);
3669 if (btrfs_file_extent_type(leaf, fi) !=
3670 BTRFS_FILE_EXTENT_REG ||
3671 btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
edbd8d4e 3672 break;
1a40e23b
ZY		 3673
3674 if (nr == max) {
3675 struct disk_extent *old = exts;
3676 max *= 2;
3677 exts = kzalloc(sizeof(*exts) * max, GFP_NOFS);
3678 memcpy(exts, old, sizeof(*exts) * nr);
3679 if (old != *extents)
3680 kfree(old);
a061fc8d 3681 }
edbd8d4e 3682
1a40e23b
ZY		 3683 exts[nr].disk_bytenr =
3684 btrfs_file_extent_disk_bytenr(leaf, fi);
3685 exts[nr].disk_num_bytes =
3686 btrfs_file_extent_disk_num_bytes(leaf, fi);
3687 exts[nr].offset = btrfs_file_extent_offset(leaf, fi);
3688 exts[nr].num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
c8b97818
CM		 3689 exts[nr].ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
3690 exts[nr].compression = btrfs_file_extent_compression(leaf, fi);
3691 exts[nr].encryption = btrfs_file_extent_encryption(leaf, fi);
3692 exts[nr].other_encoding = btrfs_file_extent_other_encoding(leaf,
3693 fi);
d899e052
YZ		 3694 BUG_ON(exts[nr].offset > 0);
3695 BUG_ON(exts[nr].compression || exts[nr].encryption);
3696 BUG_ON(exts[nr].num_bytes != exts[nr].disk_num_bytes);
edbd8d4e 3697
1a40e23b
ZY		 3698 cur_pos += exts[nr].num_bytes;
3699 nr++;
3700
3701 if (cur_pos + offset >= last_byte)
3702 break;
3703
3704 if (no_fragment) {
3705 ret = 1;
edbd8d4e 3706 goto out;
1a40e23b
ZY		 3707 }
3708 path->slots[0]++;
3709 }
3710
3711 WARN_ON(cur_pos + offset > last_byte);
3712 if (cur_pos + offset < last_byte) {
3713 ret = -ENOENT;
3714 goto out;
edbd8d4e
CM		 3715 }
3716 ret = 0;
3717out:
1a40e23b
ZY		 3718 btrfs_free_path(path);
3719 if (ret) {
3720 if (exts != *extents)
3721 kfree(exts);
3722 } else {
3723 *extents = exts;
3724 *nr_extents = nr;
3725 }
3726 return ret;
3727}
3728
3729static int noinline replace_one_extent(struct btrfs_trans_handle *trans,
3730 struct btrfs_root *root,
3731 struct btrfs_path *path,
3732 struct btrfs_key *extent_key,
3733 struct btrfs_key *leaf_key,
3734 struct btrfs_ref_path *ref_path,
3735 struct disk_extent *new_extents,
3736 int nr_extents)
3737{
3738 struct extent_buffer *leaf;
3739 struct btrfs_file_extent_item *fi;
3740 struct inode *inode = NULL;
3741 struct btrfs_key key;
3742 u64 lock_start = 0;
3743 u64 lock_end = 0;
3744 u64 num_bytes;
3745 u64 ext_offset;
3746 u64 first_pos;
3747 u32 nritems;
3bb1a1bc 3748 int nr_scaned = 0;
1a40e23b 3749 int extent_locked = 0;
d899e052 3750 int extent_type;
1a40e23b
ZY		 3751 int ret;
3752
3bb1a1bc
YZ		 3753 memcpy(&key, leaf_key, sizeof(key));
3754 first_pos = INT_LIMIT(loff_t) - extent_key->offset;
1a40e23b 3755 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
3bb1a1bc
YZ		 3756 if (key.objectid < ref_path->owner_objectid ||
3757 (key.objectid == ref_path->owner_objectid &&
3758 key.type < BTRFS_EXTENT_DATA_KEY)) {
3759 key.objectid = ref_path->owner_objectid;
3760 key.type = BTRFS_EXTENT_DATA_KEY;
3761 key.offset = 0;
3762 }
1a40e23b
ZY		 3763 }
3764
3765 while (1) {
3766 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
3767 if (ret < 0)
3768 goto out;
3769
3770 leaf = path->nodes[0];
3771 nritems = btrfs_header_nritems(leaf);
3772next:
3773 if (extent_locked && ret > 0) {
3774 /*
3775 * the file extent item was modified by someone
3776 * before the extent got locked.
3777 */
1a40e23b
ZY		 3778 unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
3779 lock_end, GFP_NOFS);
3780 extent_locked = 0;
3781 }
3782
3783 if (path->slots[0] >= nritems) {
3bb1a1bc 3784 if (++nr_scaned > 2)
1a40e23b
ZY		 3785 break;
3786
3787 BUG_ON(extent_locked);
3788 ret = btrfs_next_leaf(root, path);
3789 if (ret < 0)
3790 goto out;
3791 if (ret > 0)
3792 break;
3793 leaf = path->nodes[0];
3794 nritems = btrfs_header_nritems(leaf);
3795 }
3796
3797 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3798
3799 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
3800 if ((key.objectid > ref_path->owner_objectid) ||
3801 (key.objectid == ref_path->owner_objectid &&
3802 key.type > BTRFS_EXTENT_DATA_KEY) ||
3803 (key.offset >= first_pos + extent_key->offset))
3804 break;
3805 }
3806
3807 if (inode && key.objectid != inode->i_ino) {
3808 BUG_ON(extent_locked);
3809 btrfs_release_path(root, path);
3810 mutex_unlock(&inode->i_mutex);
3811 iput(inode);
3812 inode = NULL;
3813 continue;
3814 }
3815
3816 if (key.type != BTRFS_EXTENT_DATA_KEY) {
3817 path->slots[0]++;
3818 ret = 1;
3819 goto next;
3820 }
3821 fi = btrfs_item_ptr(leaf, path->slots[0],
3822 struct btrfs_file_extent_item);
d899e052
YZ		 3823 extent_type = btrfs_file_extent_type(leaf, fi);
3824 if ((extent_type != BTRFS_FILE_EXTENT_REG &&
3825 extent_type != BTRFS_FILE_EXTENT_PREALLOC) ||
1a40e23b
ZY		 3826 (btrfs_file_extent_disk_bytenr(leaf, fi) !=
3827 extent_key->objectid)) {
3828 path->slots[0]++;
3829 ret = 1;
3830 goto next;
3831 }
3832
3833 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
3834 ext_offset = btrfs_file_extent_offset(leaf, fi);
3835
3836 if (first_pos > key.offset - ext_offset)
3837 first_pos = key.offset - ext_offset;
3838
3839 if (!extent_locked) {
3840 lock_start = key.offset;
3841 lock_end = lock_start + num_bytes - 1;
3842 } else {
6643558d
YZ		 3843 if (lock_start > key.offset ||
3844 lock_end + 1 < key.offset + num_bytes) {
3845 unlock_extent(&BTRFS_I(inode)->io_tree,
3846 lock_start, lock_end, GFP_NOFS);
3847 extent_locked = 0;
3848 }
1a40e23b
ZY		 3849 }
3850
3851 if (!inode) {
3852 btrfs_release_path(root, path);
3853
3854 inode = btrfs_iget_locked(root->fs_info->sb,
3855 key.objectid, root);
3856 if (inode->i_state & I_NEW) {
3857 BTRFS_I(inode)->root = root;
3858 BTRFS_I(inode)->location.objectid =
3859 key.objectid;
3860 BTRFS_I(inode)->location.type =
3861 BTRFS_INODE_ITEM_KEY;
3862 BTRFS_I(inode)->location.offset = 0;
3863 btrfs_read_locked_inode(inode);
3864 unlock_new_inode(inode);
3865 }
3866 /*
3867 * some code call btrfs_commit_transaction while
3868 * holding the i_mutex, so we can't use mutex_lock
3869 * here.
3870 */
3871 if (is_bad_inode(inode) ||
3872 !mutex_trylock(&inode->i_mutex)) {
3873 iput(inode);
3874 inode = NULL;
3875 key.offset = (u64)-1;
3876 goto skip;
3877 }
3878 }
3879
3880 if (!extent_locked) {
3881 struct btrfs_ordered_extent *ordered;
3882
3883 btrfs_release_path(root, path);
3884
3885 lock_extent(&BTRFS_I(inode)->io_tree, lock_start,
3886 lock_end, GFP_NOFS);
3887 ordered = btrfs_lookup_first_ordered_extent(inode,
3888 lock_end);
3889 if (ordered &&
3890 ordered->file_offset <= lock_end &&
3891 ordered->file_offset + ordered->len > lock_start) {
3892 unlock_extent(&BTRFS_I(inode)->io_tree,
3893 lock_start, lock_end, GFP_NOFS);
3894 btrfs_start_ordered_extent(inode, ordered, 1);
3895 btrfs_put_ordered_extent(ordered);
3896 key.offset += num_bytes;
3897 goto skip;
3898 }
3899 if (ordered)
3900 btrfs_put_ordered_extent(ordered);
3901
1a40e23b
ZY		 3902 extent_locked = 1;
3903 continue;
3904 }
3905
3906 if (nr_extents == 1) {
3907 /* update extent pointer in place */
1a40e23b
ZY		 3908 btrfs_set_file_extent_disk_bytenr(leaf, fi,
3909 new_extents[0].disk_bytenr);
3910 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
3911 new_extents[0].disk_num_bytes);
1a40e23b
ZY		 3912 btrfs_mark_buffer_dirty(leaf);
3913
3914 btrfs_drop_extent_cache(inode, key.offset,
3915 key.offset + num_bytes - 1, 0);
3916
3917 ret = btrfs_inc_extent_ref(trans, root,
3918 new_extents[0].disk_bytenr,
3919 new_extents[0].disk_num_bytes,
3920 leaf->start,
3921 root->root_key.objectid,
3922 trans->transid,
3bb1a1bc 3923 key.objectid);
1a40e23b
ZY		 3924 BUG_ON(ret);
3925
3926 ret = btrfs_free_extent(trans, root,
3927 extent_key->objectid,
3928 extent_key->offset,
3929 leaf->start,
3930 btrfs_header_owner(leaf),
3931 btrfs_header_generation(leaf),
3bb1a1bc 3932 key.objectid, 0);
1a40e23b
ZY		 3933 BUG_ON(ret);
3934
3935 btrfs_release_path(root, path);
3936 key.offset += num_bytes;
3937 } else {
d899e052
YZ		 3938 BUG_ON(1);
3939#if 0
1a40e23b
ZY		 3940 u64 alloc_hint;
3941 u64 extent_len;
3942 int i;
3943 /*
3944 * drop old extent pointer at first, then insert the
3945 * new pointers one bye one
3946 */
3947 btrfs_release_path(root, path);
3948 ret = btrfs_drop_extents(trans, root, inode, key.offset,
3949 key.offset + num_bytes,
3950 key.offset, &alloc_hint);
3951 BUG_ON(ret);
3952
3953 for (i = 0; i < nr_extents; i++) {
3954 if (ext_offset >= new_extents[i].num_bytes) {
3955 ext_offset -= new_extents[i].num_bytes;
3956 continue;
3957 }
3958 extent_len = min(new_extents[i].num_bytes -
3959 ext_offset, num_bytes);
3960
3961 ret = btrfs_insert_empty_item(trans, root,
3962 path, &key,
3963 sizeof(*fi));
3964 BUG_ON(ret);
3965
3966 leaf = path->nodes[0];
3967 fi = btrfs_item_ptr(leaf, path->slots[0],
3968 struct btrfs_file_extent_item);
3969 btrfs_set_file_extent_generation(leaf, fi,
3970 trans->transid);
3971 btrfs_set_file_extent_type(leaf, fi,
3972 BTRFS_FILE_EXTENT_REG);
3973 btrfs_set_file_extent_disk_bytenr(leaf, fi,
3974 new_extents[i].disk_bytenr);
3975 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
3976 new_extents[i].disk_num_bytes);
c8b97818
CM		 3977 btrfs_set_file_extent_ram_bytes(leaf, fi,
3978 new_extents[i].ram_bytes);
3979
3980 btrfs_set_file_extent_compression(leaf, fi,
3981 new_extents[i].compression);
3982 btrfs_set_file_extent_encryption(leaf, fi,
3983 new_extents[i].encryption);
3984 btrfs_set_file_extent_other_encoding(leaf, fi,
3985 new_extents[i].other_encoding);
3986
1a40e23b
ZY		 3987 btrfs_set_file_extent_num_bytes(leaf, fi,
3988 extent_len);
3989 ext_offset += new_extents[i].offset;
3990 btrfs_set_file_extent_offset(leaf, fi,
3991 ext_offset);
3992 btrfs_mark_buffer_dirty(leaf);
3993
3994 btrfs_drop_extent_cache(inode, key.offset,
3995 key.offset + extent_len - 1, 0);
3996
3997 ret = btrfs_inc_extent_ref(trans, root,
3998 new_extents[i].disk_bytenr,
3999 new_extents[i].disk_num_bytes,
4000 leaf->start,
4001 root->root_key.objectid,
3bb1a1bc 4002 trans->transid, key.objectid);
1a40e23b
ZY		 4003 BUG_ON(ret);
4004 btrfs_release_path(root, path);
4005
a76a3cd4 4006 inode_add_bytes(inode, extent_len);
1a40e23b
ZY		 4007
4008 ext_offset = 0;
4009 num_bytes -= extent_len;
4010 key.offset += extent_len;
4011
4012 if (num_bytes == 0)
4013 break;
4014 }
4015 BUG_ON(i >= nr_extents);
d899e052 4016#endif
1a40e23b
ZY		 4017 }
4018
4019 if (extent_locked) {
1a40e23b
ZY		 4020 unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
4021 lock_end, GFP_NOFS);
4022 extent_locked = 0;
4023 }
4024skip:
4025 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS &&
4026 key.offset >= first_pos + extent_key->offset)
4027 break;
4028
4029 cond_resched();
4030 }
4031 ret = 0;
4032out:
4033 btrfs_release_path(root, path);
4034 if (inode) {
4035 mutex_unlock(&inode->i_mutex);
4036 if (extent_locked) {
1a40e23b
ZY		 4037 unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
4038 lock_end, GFP_NOFS);
4039 }
4040 iput(inode);
4041 }
4042 return ret;
4043}
4044
1a40e23b
ZY		 4045int btrfs_reloc_tree_cache_ref(struct btrfs_trans_handle *trans,
4046 struct btrfs_root *root,
4047 struct extent_buffer *buf, u64 orig_start)
4048{
4049 int level;
4050 int ret;
4051
4052 BUG_ON(btrfs_header_generation(buf) != trans->transid);
4053 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
4054
4055 level = btrfs_header_level(buf);
4056 if (level == 0) {
4057 struct btrfs_leaf_ref *ref;
4058 struct btrfs_leaf_ref *orig_ref;
4059
4060 orig_ref = btrfs_lookup_leaf_ref(root, orig_start);
4061 if (!orig_ref)
4062 return -ENOENT;
4063
4064 ref = btrfs_alloc_leaf_ref(root, orig_ref->nritems);
4065 if (!ref) {
4066 btrfs_free_leaf_ref(root, orig_ref);
4067 return -ENOMEM;
4068 }
4069
4070 ref->nritems = orig_ref->nritems;
4071 memcpy(ref->extents, orig_ref->extents,
4072 sizeof(ref->extents[0]) * ref->nritems);
4073
4074 btrfs_free_leaf_ref(root, orig_ref);
4075
4076 ref->root_gen = trans->transid;
4077 ref->bytenr = buf->start;
4078 ref->owner = btrfs_header_owner(buf);
4079 ref->generation = btrfs_header_generation(buf);
4080 ret = btrfs_add_leaf_ref(root, ref, 0);
4081 WARN_ON(ret);
4082 btrfs_free_leaf_ref(root, ref);
4083 }
4084 return 0;
4085}
4086
4087static int noinline invalidate_extent_cache(struct btrfs_root *root,
4088 struct extent_buffer *leaf,
4089 struct btrfs_block_group_cache *group,
4090 struct btrfs_root *target_root)
4091{
4092 struct btrfs_key key;
4093 struct inode *inode = NULL;
4094 struct btrfs_file_extent_item *fi;
4095 u64 num_bytes;
4096 u64 skip_objectid = 0;
4097 u32 nritems;
4098 u32 i;
4099
4100 nritems = btrfs_header_nritems(leaf);
4101 for (i = 0; i < nritems; i++) {
4102 btrfs_item_key_to_cpu(leaf, &key, i);
4103 if (key.objectid == skip_objectid ||
4104 key.type != BTRFS_EXTENT_DATA_KEY)
4105 continue;
4106 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
4107 if (btrfs_file_extent_type(leaf, fi) ==
4108 BTRFS_FILE_EXTENT_INLINE)
4109 continue;
4110 if (btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
4111 continue;
4112 if (!inode || inode->i_ino != key.objectid) {
4113 iput(inode);
4114 inode = btrfs_ilookup(target_root->fs_info->sb,
4115 key.objectid, target_root, 1);
4116 }
4117 if (!inode) {
4118 skip_objectid = key.objectid;
4119 continue;
4120 }
4121 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
4122
4123 lock_extent(&BTRFS_I(inode)->io_tree, key.offset,
4124 key.offset + num_bytes - 1, GFP_NOFS);
1a40e23b
ZY		 4125 btrfs_drop_extent_cache(inode, key.offset,
4126 key.offset + num_bytes - 1, 1);
1a40e23b
ZY		 4127 unlock_extent(&BTRFS_I(inode)->io_tree, key.offset,
4128 key.offset + num_bytes - 1, GFP_NOFS);
4129 cond_resched();
4130 }
4131 iput(inode);
4132 return 0;
4133}
4134
4135static int noinline replace_extents_in_leaf(struct btrfs_trans_handle *trans,
4136 struct btrfs_root *root,
4137 struct extent_buffer *leaf,
4138 struct btrfs_block_group_cache *group,
4139 struct inode *reloc_inode)
4140{
4141 struct btrfs_key key;
4142 struct btrfs_key extent_key;
4143 struct btrfs_file_extent_item *fi;
4144 struct btrfs_leaf_ref *ref;
4145 struct disk_extent *new_extent;
4146 u64 bytenr;
4147 u64 num_bytes;
4148 u32 nritems;
4149 u32 i;
4150 int ext_index;
4151 int nr_extent;
4152 int ret;
4153
4154 new_extent = kmalloc(sizeof(*new_extent), GFP_NOFS);
4155 BUG_ON(!new_extent);
4156
4157 ref = btrfs_lookup_leaf_ref(root, leaf->start);
4158 BUG_ON(!ref);
4159
4160 ext_index = -1;
4161 nritems = btrfs_header_nritems(leaf);
4162 for (i = 0; i < nritems; i++) {
4163 btrfs_item_key_to_cpu(leaf, &key, i);
4164 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
4165 continue;
4166 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
4167 if (btrfs_file_extent_type(leaf, fi) ==
4168 BTRFS_FILE_EXTENT_INLINE)
4169 continue;
4170 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
4171 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
4172 if (bytenr == 0)
4173 continue;
4174
4175 ext_index++;
4176 if (bytenr >= group->key.objectid + group->key.offset ||
4177 bytenr + num_bytes <= group->key.objectid)
4178 continue;
4179
4180 extent_key.objectid = bytenr;
4181 extent_key.offset = num_bytes;
4182 extent_key.type = BTRFS_EXTENT_ITEM_KEY;
4183 nr_extent = 1;
4184 ret = get_new_locations(reloc_inode, &extent_key,
4185 group->key.objectid, 1,
4186 &new_extent, &nr_extent);
4187 if (ret > 0)
4188 continue;
4189 BUG_ON(ret < 0);
4190
4191 BUG_ON(ref->extents[ext_index].bytenr != bytenr);
4192 BUG_ON(ref->extents[ext_index].num_bytes != num_bytes);
4193 ref->extents[ext_index].bytenr = new_extent->disk_bytenr;
4194 ref->extents[ext_index].num_bytes = new_extent->disk_num_bytes;
4195
1a40e23b
ZY		 4196 btrfs_set_file_extent_disk_bytenr(leaf, fi,
4197 new_extent->disk_bytenr);
4198 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
4199 new_extent->disk_num_bytes);
1a40e23b
ZY		 4200 btrfs_mark_buffer_dirty(leaf);
4201
4202 ret = btrfs_inc_extent_ref(trans, root,
4203 new_extent->disk_bytenr,
4204 new_extent->disk_num_bytes,
4205 leaf->start,
4206 root->root_key.objectid,
3bb1a1bc 4207 trans->transid, key.objectid);
1a40e23b
ZY		 4208 BUG_ON(ret);
4209 ret = btrfs_free_extent(trans, root,
4210 bytenr, num_bytes, leaf->start,
4211 btrfs_header_owner(leaf),
4212 btrfs_header_generation(leaf),
3bb1a1bc 4213 key.objectid, 0);
1a40e23b
ZY		 4214 BUG_ON(ret);
4215 cond_resched();
4216 }
4217 kfree(new_extent);
4218 BUG_ON(ext_index + 1 != ref->nritems);
4219 btrfs_free_leaf_ref(root, ref);
4220 return 0;
4221}
4222
f82d02d9
YZ		 4223int btrfs_free_reloc_root(struct btrfs_trans_handle *trans,
4224 struct btrfs_root *root)
1a40e23b
ZY		 4225{
4226 struct btrfs_root *reloc_root;
f82d02d9 4227 int ret;
1a40e23b
ZY		 4228
4229 if (root->reloc_root) {
4230 reloc_root = root->reloc_root;
4231 root->reloc_root = NULL;
4232 list_add(&reloc_root->dead_list,
4233 &root->fs_info->dead_reloc_roots);
f82d02d9
YZ		 4234
4235 btrfs_set_root_bytenr(&reloc_root->root_item,
4236 reloc_root->node->start);
4237 btrfs_set_root_level(&root->root_item,
4238 btrfs_header_level(reloc_root->node));
4239 memset(&reloc_root->root_item.drop_progress, 0,
4240 sizeof(struct btrfs_disk_key));
4241 reloc_root->root_item.drop_level = 0;
4242
4243 ret = btrfs_update_root(trans, root->fs_info->tree_root,
4244 &reloc_root->root_key,
4245 &reloc_root->root_item);
4246 BUG_ON(ret);
1a40e23b
ZY		 4247 }
4248 return 0;
4249}
4250
4251int btrfs_drop_dead_reloc_roots(struct btrfs_root *root)
4252{
4253 struct btrfs_trans_handle *trans;
4254 struct btrfs_root *reloc_root;
4255 struct btrfs_root *prev_root = NULL;
4256 struct list_head dead_roots;
4257 int ret;
4258 unsigned long nr;
4259
4260 INIT_LIST_HEAD(&dead_roots);
4261 list_splice_init(&root->fs_info->dead_reloc_roots, &dead_roots);
4262
4263 while (!list_empty(&dead_roots)) {
4264 reloc_root = list_entry(dead_roots.prev,
4265 struct btrfs_root, dead_list);
4266 list_del_init(&reloc_root->dead_list);
4267
4268 BUG_ON(reloc_root->commit_root != NULL);
4269 while (1) {
4270 trans = btrfs_join_transaction(root, 1);
4271 BUG_ON(!trans);
4272
4273 mutex_lock(&root->fs_info->drop_mutex);
4274 ret = btrfs_drop_snapshot(trans, reloc_root);
4275 if (ret != -EAGAIN)
4276 break;
4277 mutex_unlock(&root->fs_info->drop_mutex);
4278
4279 nr = trans->blocks_used;
4280 ret = btrfs_end_transaction(trans, root);
4281 BUG_ON(ret);
4282 btrfs_btree_balance_dirty(root, nr);
4283 }
4284
4285 free_extent_buffer(reloc_root->node);
4286
4287 ret = btrfs_del_root(trans, root->fs_info->tree_root,
4288 &reloc_root->root_key);
4289 BUG_ON(ret);
4290 mutex_unlock(&root->fs_info->drop_mutex);
4291
4292 nr = trans->blocks_used;
4293 ret = btrfs_end_transaction(trans, root);
4294 BUG_ON(ret);
4295 btrfs_btree_balance_dirty(root, nr);
4296
4297 kfree(prev_root);
4298 prev_root = reloc_root;
4299 }
4300 if (prev_root) {
4301 btrfs_remove_leaf_refs(prev_root, (u64)-1, 0);
4302 kfree(prev_root);
4303 }
4304 return 0;
4305}
4306
4307int btrfs_add_dead_reloc_root(struct btrfs_root *root)
4308{
4309 list_add(&root->dead_list, &root->fs_info->dead_reloc_roots);
4310 return 0;
4311}
4312
4313int btrfs_cleanup_reloc_trees(struct btrfs_root *root)
4314{
4315 struct btrfs_root *reloc_root;
4316 struct btrfs_trans_handle *trans;
4317 struct btrfs_key location;
4318 int found;
4319 int ret;
4320
4321 mutex_lock(&root->fs_info->tree_reloc_mutex);
4322 ret = btrfs_find_dead_roots(root, BTRFS_TREE_RELOC_OBJECTID, NULL);
4323 BUG_ON(ret);
4324 found = !list_empty(&root->fs_info->dead_reloc_roots);
4325 mutex_unlock(&root->fs_info->tree_reloc_mutex);
4326
4327 if (found) {
4328 trans = btrfs_start_transaction(root, 1);
4329 BUG_ON(!trans);
4330 ret = btrfs_commit_transaction(trans, root);
4331 BUG_ON(ret);
4332 }
4333
4334 location.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
4335 location.offset = (u64)-1;
4336 location.type = BTRFS_ROOT_ITEM_KEY;
4337
4338 reloc_root = btrfs_read_fs_root_no_name(root->fs_info, &location);
4339 BUG_ON(!reloc_root);
4340 btrfs_orphan_cleanup(reloc_root);
4341 return 0;
4342}
4343
4344static int noinline init_reloc_tree(struct btrfs_trans_handle *trans,
4345 struct btrfs_root *root)
4346{
4347 struct btrfs_root *reloc_root;
4348 struct extent_buffer *eb;
4349 struct btrfs_root_item *root_item;
4350 struct btrfs_key root_key;
4351 int ret;
4352
4353 BUG_ON(!root->ref_cows);
4354 if (root->reloc_root)
4355 return 0;
4356
4357 root_item = kmalloc(sizeof(*root_item), GFP_NOFS);
4358 BUG_ON(!root_item);
4359
4360 ret = btrfs_copy_root(trans, root, root->commit_root,
4361 &eb, BTRFS_TREE_RELOC_OBJECTID);
4362 BUG_ON(ret);
4363
4364 root_key.objectid = BTRFS_TREE_RELOC_OBJECTID;
4365 root_key.offset = root->root_key.objectid;
4366 root_key.type = BTRFS_ROOT_ITEM_KEY;
4367
4368 memcpy(root_item, &root->root_item, sizeof(root_item));
4369 btrfs_set_root_refs(root_item, 0);
4370 btrfs_set_root_bytenr(root_item, eb->start);
4371 btrfs_set_root_level(root_item, btrfs_header_level(eb));
84234f3a 4372 btrfs_set_root_generation(root_item, trans->transid);
1a40e23b
ZY		 4373
4374 btrfs_tree_unlock(eb);
4375 free_extent_buffer(eb);
4376
4377 ret = btrfs_insert_root(trans, root->fs_info->tree_root,
4378 &root_key, root_item);
4379 BUG_ON(ret);
4380 kfree(root_item);
4381
4382 reloc_root = btrfs_read_fs_root_no_radix(root->fs_info->tree_root,
4383 &root_key);
4384 BUG_ON(!reloc_root);
4385 reloc_root->last_trans = trans->transid;
4386 reloc_root->commit_root = NULL;
4387 reloc_root->ref_tree = &root->fs_info->reloc_ref_tree;
4388
4389 root->reloc_root = reloc_root;
4390 return 0;
4391}
4392
4393/*
4394 * Core function of space balance.
4395 *
4396 * The idea is using reloc trees to relocate tree blocks in reference
f82d02d9
YZ		 4397 * counted roots. There is one reloc tree for each subvol, and all
4398 * reloc trees share same root key objectid. Reloc trees are snapshots
4399 * of the latest committed roots of subvols (root->commit_root).
4400 *
4401 * To relocate a tree block referenced by a subvol, there are two steps.
4402 * COW the block through subvol's reloc tree, then update block pointer
4403 * in the subvol to point to the new block. Since all reloc trees share
4404 * same root key objectid, doing special handing for tree blocks owned
4405 * by them is easy. Once a tree block has been COWed in one reloc tree,
4406 * we can use the resulting new block directly when the same block is
4407 * required to COW again through other reloc trees. By this way, relocated
4408 * tree blocks are shared between reloc trees, so they are also shared
4409 * between subvols.
1a40e23b
ZY		 4410 */
4411static int noinline relocate_one_path(struct btrfs_trans_handle *trans,
4412 struct btrfs_root *root,
4413 struct btrfs_path *path,
4414 struct btrfs_key *first_key,
4415 struct btrfs_ref_path *ref_path,
4416 struct btrfs_block_group_cache *group,
4417 struct inode *reloc_inode)
4418{
4419 struct btrfs_root *reloc_root;
4420 struct extent_buffer *eb = NULL;
4421 struct btrfs_key *keys;
4422 u64 *nodes;
4423 int level;
f82d02d9 4424 int shared_level;
1a40e23b 4425 int lowest_level = 0;
1a40e23b
ZY		 4426 int ret;
4427
4428 if (ref_path->owner_objectid < BTRFS_FIRST_FREE_OBJECTID)
4429 lowest_level = ref_path->owner_objectid;
4430
f82d02d9 4431 if (!root->ref_cows) {
1a40e23b
ZY		 4432 path->lowest_level = lowest_level;
4433 ret = btrfs_search_slot(trans, root, first_key, path, 0, 1);
4434 BUG_ON(ret < 0);
4435 path->lowest_level = 0;
4436 btrfs_release_path(root, path);
4437 return 0;
4438 }
4439
1a40e23b
ZY		 4440 mutex_lock(&root->fs_info->tree_reloc_mutex);
4441 ret = init_reloc_tree(trans, root);
4442 BUG_ON(ret);
4443 reloc_root = root->reloc_root;
4444
f82d02d9
YZ		 4445 shared_level = ref_path->shared_level;
4446 ref_path->shared_level = BTRFS_MAX_LEVEL - 1;
1a40e23b 4447
f82d02d9
YZ		 4448 keys = ref_path->node_keys;
4449 nodes = ref_path->new_nodes;
4450 memset(&keys[shared_level + 1], 0,
4451 sizeof(*keys) * (BTRFS_MAX_LEVEL - shared_level - 1));
4452 memset(&nodes[shared_level + 1], 0,
4453 sizeof(*nodes) * (BTRFS_MAX_LEVEL - shared_level - 1));
1a40e23b 4454
f82d02d9
YZ		 4455 if (nodes[lowest_level] == 0) {
4456 path->lowest_level = lowest_level;
4457 ret = btrfs_search_slot(trans, reloc_root, first_key, path,
4458 0, 1);
4459 BUG_ON(ret);
4460 for (level = lowest_level; level < BTRFS_MAX_LEVEL; level++) {
4461 eb = path->nodes[level];
4462 if (!eb || eb == reloc_root->node)
4463 break;
4464 nodes[level] = eb->start;
4465 if (level == 0)
4466 btrfs_item_key_to_cpu(eb, &keys[level], 0);
4467 else
4468 btrfs_node_key_to_cpu(eb, &keys[level], 0);
4469 }
4470 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
4471 eb = path->nodes[0];
4472 ret = replace_extents_in_leaf(trans, reloc_root, eb,
4473 group, reloc_inode);
4474 BUG_ON(ret);
4475 }
4476 btrfs_release_path(reloc_root, path);
4477 } else {
1a40e23b 4478 ret = btrfs_merge_path(trans, reloc_root, keys, nodes,
f82d02d9 4479 lowest_level);
1a40e23b
ZY		 4480 BUG_ON(ret);
4481 }
4482
1a40e23b
ZY		 4483 /*
4484 * replace tree blocks in the fs tree with tree blocks in
4485 * the reloc tree.
4486 */
4487 ret = btrfs_merge_path(trans, root, keys, nodes, lowest_level);
4488 BUG_ON(ret < 0);
4489
4490 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
f82d02d9
YZ		 4491 ret = btrfs_search_slot(trans, reloc_root, first_key, path,
4492 0, 0);
4493 BUG_ON(ret);
4494 extent_buffer_get(path->nodes[0]);
4495 eb = path->nodes[0];
4496 btrfs_release_path(reloc_root, path);
1a40e23b
ZY		 4497 ret = invalidate_extent_cache(reloc_root, eb, group, root);
4498 BUG_ON(ret);
4499 free_extent_buffer(eb);
4500 }
1a40e23b 4501
f82d02d9 4502 mutex_unlock(&root->fs_info->tree_reloc_mutex);
1a40e23b 4503 path->lowest_level = 0;
1a40e23b
ZY		 4504 return 0;
4505}
4506
4507static int noinline relocate_tree_block(struct btrfs_trans_handle *trans,
4508 struct btrfs_root *root,
4509 struct btrfs_path *path,
4510 struct btrfs_key *first_key,
4511 struct btrfs_ref_path *ref_path)
4512{
4513 int ret;
1a40e23b
ZY		 4514
4515 ret = relocate_one_path(trans, root, path, first_key,
4516 ref_path, NULL, NULL);
4517 BUG_ON(ret);
4518
4519 if (root == root->fs_info->extent_root)
4520 btrfs_extent_post_op(trans, root);
1a40e23b
ZY		 4521
4522 return 0;
4523}
4524
4525static int noinline del_extent_zero(struct btrfs_trans_handle *trans,
4526 struct btrfs_root *extent_root,
4527 struct btrfs_path *path,
4528 struct btrfs_key *extent_key)
4529{
4530 int ret;
4531
1a40e23b
ZY		 4532 ret = btrfs_search_slot(trans, extent_root, extent_key, path, -1, 1);
4533 if (ret)
4534 goto out;
4535 ret = btrfs_del_item(trans, extent_root, path);
4536out:
4537 btrfs_release_path(extent_root, path);
1a40e23b
ZY		 4538 return ret;
4539}
4540
4541static struct btrfs_root noinline *read_ref_root(struct btrfs_fs_info *fs_info,
4542 struct btrfs_ref_path *ref_path)
4543{
4544 struct btrfs_key root_key;
4545
4546 root_key.objectid = ref_path->root_objectid;
4547 root_key.type = BTRFS_ROOT_ITEM_KEY;
4548 if (is_cowonly_root(ref_path->root_objectid))
4549 root_key.offset = 0;
4550 else
4551 root_key.offset = (u64)-1;
4552
4553 return btrfs_read_fs_root_no_name(fs_info, &root_key);
4554}
4555
4556static int noinline relocate_one_extent(struct btrfs_root *extent_root,
4557 struct btrfs_path *path,
4558 struct btrfs_key *extent_key,
4559 struct btrfs_block_group_cache *group,
4560 struct inode *reloc_inode, int pass)
4561{
4562 struct btrfs_trans_handle *trans;
4563 struct btrfs_root *found_root;
4564 struct btrfs_ref_path *ref_path = NULL;
4565 struct disk_extent *new_extents = NULL;
4566 int nr_extents = 0;
4567 int loops;
4568 int ret;
4569 int level;
4570 struct btrfs_key first_key;
4571 u64 prev_block = 0;
4572
1a40e23b
ZY		 4573
4574 trans = btrfs_start_transaction(extent_root, 1);
4575 BUG_ON(!trans);
4576
4577 if (extent_key->objectid == 0) {
4578 ret = del_extent_zero(trans, extent_root, path, extent_key);
4579 goto out;
4580 }
4581
4582 ref_path = kmalloc(sizeof(*ref_path), GFP_NOFS);
4583 if (!ref_path) {
4584 ret = -ENOMEM;
4585 goto out;
4586 }
4587
4588 for (loops = 0; ; loops++) {
4589 if (loops == 0) {
4590 ret = btrfs_first_ref_path(trans, extent_root, ref_path,
4591 extent_key->objectid);
4592 } else {
4593 ret = btrfs_next_ref_path(trans, extent_root, ref_path);
4594 }
4595 if (ret < 0)
4596 goto out;
4597 if (ret > 0)
4598 break;
4599
4600 if (ref_path->root_objectid == BTRFS_TREE_LOG_OBJECTID ||
4601 ref_path->root_objectid == BTRFS_TREE_RELOC_OBJECTID)
4602 continue;
4603
4604 found_root = read_ref_root(extent_root->fs_info, ref_path);
4605 BUG_ON(!found_root);
4606 /*
4607 * for reference counted tree, only process reference paths
4608 * rooted at the latest committed root.
4609 */
4610 if (found_root->ref_cows &&
4611 ref_path->root_generation != found_root->root_key.offset)
4612 continue;
4613
4614 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
4615 if (pass == 0) {
4616 /*
4617 * copy data extents to new locations
4618 */
4619 u64 group_start = group->key.objectid;
4620 ret = relocate_data_extent(reloc_inode,
4621 extent_key,
4622 group_start);
4623 if (ret < 0)
4624 goto out;
4625 break;
4626 }
4627 level = 0;
4628 } else {
4629 level = ref_path->owner_objectid;
4630 }
4631
4632 if (prev_block != ref_path->nodes[level]) {
4633 struct extent_buffer *eb;
4634 u64 block_start = ref_path->nodes[level];
4635 u64 block_size = btrfs_level_size(found_root, level);
4636
4637 eb = read_tree_block(found_root, block_start,
4638 block_size, 0);
4639 btrfs_tree_lock(eb);
4640 BUG_ON(level != btrfs_header_level(eb));
4641
4642 if (level == 0)
4643 btrfs_item_key_to_cpu(eb, &first_key, 0);
4644 else
4645 btrfs_node_key_to_cpu(eb, &first_key, 0);
4646
4647 btrfs_tree_unlock(eb);
4648 free_extent_buffer(eb);
4649 prev_block = block_start;
4650 }
4651
4652 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID &&
4653 pass >= 2) {
4654 /*
4655 * use fallback method to process the remaining
4656 * references.
4657 */
4658 if (!new_extents) {
4659 u64 group_start = group->key.objectid;
d899e052
YZ		 4660 new_extents = kmalloc(sizeof(*new_extents),
4661 GFP_NOFS);
4662 nr_extents = 1;
1a40e23b
ZY		 4663 ret = get_new_locations(reloc_inode,
4664 extent_key,
d899e052 4665 group_start, 1,
1a40e23b
ZY		 4666 &new_extents,
4667 &nr_extents);
d899e052 4668 if (ret)
1a40e23b
ZY		 4669 goto out;
4670 }
4671 btrfs_record_root_in_trans(found_root);
4672 ret = replace_one_extent(trans, found_root,
4673 path, extent_key,
4674 &first_key, ref_path,
4675 new_extents, nr_extents);
4676 if (ret < 0)
4677 goto out;
4678 continue;
4679 }
4680
4681 btrfs_record_root_in_trans(found_root);
4682 if (ref_path->owner_objectid < BTRFS_FIRST_FREE_OBJECTID) {
4683 ret = relocate_tree_block(trans, found_root, path,
4684 &first_key, ref_path);
4685 } else {
4686 /*
4687 * try to update data extent references while
4688 * keeping metadata shared between snapshots.
4689 */
4690 ret = relocate_one_path(trans, found_root, path,
4691 &first_key, ref_path,
4692 group, reloc_inode);
4693 }
4694 if (ret < 0)
4695 goto out;
4696 }
4697 ret = 0;
4698out:
4699 btrfs_end_transaction(trans, extent_root);
4700 kfree(new_extents);
4701 kfree(ref_path);
1a40e23b
ZY		 4702 return ret;
4703}
4704
ec44a35c
CM		 4705static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
4706{
4707 u64 num_devices;
4708 u64 stripped = BTRFS_BLOCK_GROUP_RAID0 |
4709 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
4710
a061fc8d 4711 num_devices = root->fs_info->fs_devices->num_devices;
ec44a35c
CM		 4712 if (num_devices == 1) {
4713 stripped |= BTRFS_BLOCK_GROUP_DUP;
4714 stripped = flags & ~stripped;
4715
4716 /* turn raid0 into single device chunks */
4717 if (flags & BTRFS_BLOCK_GROUP_RAID0)
4718 return stripped;
4719
4720 /* turn mirroring into duplication */
4721 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
4722 BTRFS_BLOCK_GROUP_RAID10))
4723 return stripped | BTRFS_BLOCK_GROUP_DUP;
4724 return flags;
4725 } else {
4726 /* they already had raid on here, just return */
ec44a35c
CM		 4727 if (flags & stripped)
4728 return flags;
4729
4730 stripped |= BTRFS_BLOCK_GROUP_DUP;
4731 stripped = flags & ~stripped;
4732
4733 /* switch duplicated blocks with raid1 */
4734 if (flags & BTRFS_BLOCK_GROUP_DUP)
4735 return stripped | BTRFS_BLOCK_GROUP_RAID1;
4736
4737 /* turn single device chunks into raid0 */
4738 return stripped | BTRFS_BLOCK_GROUP_RAID0;
4739 }
4740 return flags;
4741}
4742
0ef3e66b
CM		 4743int __alloc_chunk_for_shrink(struct btrfs_root *root,
4744 struct btrfs_block_group_cache *shrink_block_group,
4745 int force)
4746{
4747 struct btrfs_trans_handle *trans;
4748 u64 new_alloc_flags;
4749 u64 calc;
4750
c286ac48 4751 spin_lock(&shrink_block_group->lock);
0ef3e66b 4752 if (btrfs_block_group_used(&shrink_block_group->item) > 0) {
c286ac48 4753 spin_unlock(&shrink_block_group->lock);
c286ac48 4754
0ef3e66b 4755 trans = btrfs_start_transaction(root, 1);
c286ac48 4756 spin_lock(&shrink_block_group->lock);
7d9eb12c 4757
0ef3e66b
CM		 4758 new_alloc_flags = update_block_group_flags(root,
4759 shrink_block_group->flags);
4760 if (new_alloc_flags != shrink_block_group->flags) {
4761 calc =
4762 btrfs_block_group_used(&shrink_block_group->item);
4763 } else {
4764 calc = shrink_block_group->key.offset;
4765 }
c286ac48
CM		 4766 spin_unlock(&shrink_block_group->lock);
4767
0ef3e66b
CM		 4768 do_chunk_alloc(trans, root->fs_info->extent_root,
4769 calc + 2 * 1024 * 1024, new_alloc_flags, force);
7d9eb12c 4770
0ef3e66b 4771 btrfs_end_transaction(trans, root);
c286ac48
CM		 4772 } else
4773 spin_unlock(&shrink_block_group->lock);
0ef3e66b
CM		 4774 return 0;
4775}
4776
1a40e23b
ZY		 4777static int __insert_orphan_inode(struct btrfs_trans_handle *trans,
4778 struct btrfs_root *root,
4779 u64 objectid, u64 size)
4780{
4781 struct btrfs_path *path;
4782 struct btrfs_inode_item *item;
4783 struct extent_buffer *leaf;
4784 int ret;
4785
4786 path = btrfs_alloc_path();
4787 if (!path)
4788 return -ENOMEM;
4789
4790 ret = btrfs_insert_empty_inode(trans, root, path, objectid);
4791 if (ret)
4792 goto out;
4793
4794 leaf = path->nodes[0];
4795 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_inode_item);
4796 memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item));
4797 btrfs_set_inode_generation(leaf, item, 1);
4798 btrfs_set_inode_size(leaf, item, size);
4799 btrfs_set_inode_mode(leaf, item, S_IFREG | 0600);
d899e052
YZ		 4800 btrfs_set_inode_flags(leaf, item, BTRFS_INODE_NODATASUM |
4801 BTRFS_INODE_NOCOMPRESS);
1a40e23b
ZY		 4802 btrfs_mark_buffer_dirty(leaf);
4803 btrfs_release_path(root, path);
4804out:
4805 btrfs_free_path(path);
4806 return ret;
4807}
4808
4809static struct inode noinline *create_reloc_inode(struct btrfs_fs_info *fs_info,
4810 struct btrfs_block_group_cache *group)
4811{
4812 struct inode *inode = NULL;
4813 struct btrfs_trans_handle *trans;
4814 struct btrfs_root *root;
4815 struct btrfs_key root_key;
4816 u64 objectid = BTRFS_FIRST_FREE_OBJECTID;
4817 int err = 0;
4818
4819 root_key.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
4820 root_key.type = BTRFS_ROOT_ITEM_KEY;
4821 root_key.offset = (u64)-1;
4822 root = btrfs_read_fs_root_no_name(fs_info, &root_key);
4823 if (IS_ERR(root))
4824 return ERR_CAST(root);
4825
4826 trans = btrfs_start_transaction(root, 1);
4827 BUG_ON(!trans);
4828
4829 err = btrfs_find_free_objectid(trans, root, objectid, &objectid);
4830 if (err)
4831 goto out;
4832
4833 err = __insert_orphan_inode(trans, root, objectid, group->key.offset);
4834 BUG_ON(err);
4835
4836 err = btrfs_insert_file_extent(trans, root, objectid, 0, 0, 0,
c8b97818
CM		 4837 group->key.offset, 0, group->key.offset,
4838 0, 0, 0);
1a40e23b
ZY		 4839 BUG_ON(err);
4840
4841 inode = btrfs_iget_locked(root->fs_info->sb, objectid, root);
4842 if (inode->i_state & I_NEW) {
4843 BTRFS_I(inode)->root = root;
4844 BTRFS_I(inode)->location.objectid = objectid;
4845 BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
4846 BTRFS_I(inode)->location.offset = 0;
4847 btrfs_read_locked_inode(inode);
4848 unlock_new_inode(inode);
4849 BUG_ON(is_bad_inode(inode));
4850 } else {
4851 BUG_ON(1);
4852 }
4853
4854 err = btrfs_orphan_add(trans, inode);
4855out:
4856 btrfs_end_transaction(trans, root);
4857 if (err) {
4858 if (inode)
4859 iput(inode);
4860 inode = ERR_PTR(err);
4861 }
4862 return inode;
4863}
4864
4865int btrfs_relocate_block_group(struct btrfs_root *root, u64 group_start)
edbd8d4e
CM		 4866{
4867 struct btrfs_trans_handle *trans;
edbd8d4e 4868 struct btrfs_path *path;
1a40e23b
ZY		 4869 struct btrfs_fs_info *info = root->fs_info;
4870 struct extent_buffer *leaf;
4871 struct inode *reloc_inode;
4872 struct btrfs_block_group_cache *block_group;
4873 struct btrfs_key key;
d899e052 4874 u64 skipped;
edbd8d4e
CM		 4875 u64 cur_byte;
4876 u64 total_found;
edbd8d4e
CM		 4877 u32 nritems;
4878 int ret;
a061fc8d 4879 int progress;
1a40e23b 4880 int pass = 0;
edbd8d4e 4881
1a40e23b
ZY		 4882 root = root->fs_info->extent_root;
4883
4884 block_group = btrfs_lookup_block_group(info, group_start);
4885 BUG_ON(!block_group);
8f18cf13 4886
1a40e23b
ZY		 4887 printk("btrfs relocating block group %llu flags %llu\n",
4888 (unsigned long long)block_group->key.objectid,
4889 (unsigned long long)block_group->flags);
8f18cf13 4890
edbd8d4e 4891 path = btrfs_alloc_path();
1a40e23b 4892 BUG_ON(!path);
edbd8d4e 4893
1a40e23b
ZY		 4894 reloc_inode = create_reloc_inode(info, block_group);
4895 BUG_ON(IS_ERR(reloc_inode));
323da79c 4896
1a40e23b
ZY		 4897 __alloc_chunk_for_shrink(root, block_group, 1);
4898 block_group->ro = 1;
4899 block_group->space_info->total_bytes -= block_group->key.offset;
323da79c 4900
1a40e23b
ZY		 4901 btrfs_start_delalloc_inodes(info->tree_root);
4902 btrfs_wait_ordered_extents(info->tree_root, 0);
4903again:
d899e052 4904 skipped = 0;
edbd8d4e 4905 total_found = 0;
a061fc8d 4906 progress = 0;
1a40e23b 4907 key.objectid = block_group->key.objectid;
edbd8d4e
CM		 4908 key.offset = 0;
4909 key.type = 0;
73e48b27 4910 cur_byte = key.objectid;
4313b399 4911
1a40e23b
ZY		 4912 trans = btrfs_start_transaction(info->tree_root, 1);
4913 btrfs_commit_transaction(trans, info->tree_root);
ea8c2819 4914
1a40e23b
ZY		 4915 mutex_lock(&root->fs_info->cleaner_mutex);
4916 btrfs_clean_old_snapshots(info->tree_root);
4917 btrfs_remove_leaf_refs(info->tree_root, (u64)-1, 1);
4918 mutex_unlock(&root->fs_info->cleaner_mutex);
ea8c2819 4919
73e48b27 4920 while(1) {
edbd8d4e
CM		 4921 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4922 if (ret < 0)
4923 goto out;
7d9eb12c 4924next:
edbd8d4e 4925 leaf = path->nodes[0];
73e48b27 4926 nritems = btrfs_header_nritems(leaf);
73e48b27
Y		 4927 if (path->slots[0] >= nritems) {
4928 ret = btrfs_next_leaf(root, path);
4929 if (ret < 0)
4930 goto out;
4931 if (ret == 1) {
4932 ret = 0;
4933 break;
edbd8d4e 4934 }
73e48b27
Y		 4935 leaf = path->nodes[0];
4936 nritems = btrfs_header_nritems(leaf);
edbd8d4e 4937 }
73e48b27 4938
1a40e23b 4939 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
725c8463 4940
1a40e23b
ZY		 4941 if (key.objectid >= block_group->key.objectid +
4942 block_group->key.offset)
8f18cf13
CM		 4943 break;
4944
725c8463 4945 if (progress && need_resched()) {
725c8463 4946 btrfs_release_path(root, path);
1a40e23b 4947 cond_resched();
725c8463 4948 progress = 0;
1a40e23b 4949 continue;
725c8463
CM		 4950 }
4951 progress = 1;
4952
1a40e23b
ZY		 4953 if (btrfs_key_type(&key) != BTRFS_EXTENT_ITEM_KEY ||
4954 key.objectid + key.offset <= cur_byte) {
edbd8d4e 4955 path->slots[0]++;
edbd8d4e
CM		 4956 goto next;
4957 }
73e48b27 4958
edbd8d4e 4959 total_found++;
1a40e23b 4960 cur_byte = key.objectid + key.offset;
edbd8d4e 4961 btrfs_release_path(root, path);
edbd8d4e 4962
1a40e23b
ZY		 4963 __alloc_chunk_for_shrink(root, block_group, 0);
4964 ret = relocate_one_extent(root, path, &key, block_group,
4965 reloc_inode, pass);
4966 BUG_ON(ret < 0);
d899e052
YZ		 4967 if (ret > 0)
4968 skipped++;
edbd8d4e 4969
1a40e23b
ZY		 4970 key.objectid = cur_byte;
4971 key.type = 0;
4972 key.offset = 0;
edbd8d4e
CM		 4973 }
4974
1a40e23b 4975 btrfs_release_path(root, path);
7d9eb12c 4976
1a40e23b
ZY		 4977 if (pass == 0) {
4978 btrfs_wait_ordered_range(reloc_inode, 0, (u64)-1);
4979 invalidate_mapping_pages(reloc_inode->i_mapping, 0, -1);
4980 WARN_ON(reloc_inode->i_mapping->nrpages);
8e8a1e31 4981 }
73e48b27 4982
1a40e23b
ZY		 4983 if (total_found > 0) {
4984 printk("btrfs found %llu extents in pass %d\n",
4985 (unsigned long long)total_found, pass);
4986 pass++;
d899e052
YZ		 4987 if (total_found == skipped && pass > 2) {
4988 iput(reloc_inode);
4989 reloc_inode = create_reloc_inode(info, block_group);
4990 pass = 0;
4991 }
1a40e23b 4992 goto again;
0f9dd46c 4993 }
0ef3e66b 4994
1a40e23b
ZY		 4995 /* delete reloc_inode */
4996 iput(reloc_inode);
4997
4998 /* unpin extents in this range */
4999 trans = btrfs_start_transaction(info->tree_root, 1);
5000 btrfs_commit_transaction(trans, info->tree_root);
0ef3e66b 5001
1a40e23b
ZY		 5002 spin_lock(&block_group->lock);
5003 WARN_ON(block_group->pinned > 0);
5004 WARN_ON(block_group->reserved > 0);
5005 WARN_ON(btrfs_block_group_used(&block_group->item) > 0);
5006 spin_unlock(&block_group->lock);
5007 ret = 0;
edbd8d4e 5008out:
1a40e23b 5009 btrfs_free_path(path);
edbd8d4e
CM		 5010 return ret;
5011}
5012
0b86a832
CM		 5013int find_first_block_group(struct btrfs_root *root, struct btrfs_path *path,
5014 struct btrfs_key *key)
5015{
925baedd 5016 int ret = 0;
0b86a832
CM		 5017 struct btrfs_key found_key;
5018 struct extent_buffer *leaf;
5019 int slot;
edbd8d4e 5020
0b86a832
CM		 5021 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
5022 if (ret < 0)
925baedd
CM		 5023 goto out;
5024
0b86a832
CM		 5025 while(1) {
5026 slot = path->slots[0];
edbd8d4e 5027 leaf = path->nodes[0];
0b86a832
CM		 5028 if (slot >= btrfs_header_nritems(leaf)) {
5029 ret = btrfs_next_leaf(root, path);
5030 if (ret == 0)
5031 continue;
5032 if (ret < 0)
925baedd 5033 goto out;
0b86a832 5034 break;
edbd8d4e 5035 }
0b86a832 5036 btrfs_item_key_to_cpu(leaf, &found_key, slot);
edbd8d4e 5037
0b86a832 5038 if (found_key.objectid >= key->objectid &&
925baedd
CM		 5039 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
5040 ret = 0;
5041 goto out;
5042 }
0b86a832 5043 path->slots[0]++;
edbd8d4e 5044 }
0b86a832 5045 ret = -ENOENT;
925baedd 5046out:
0b86a832 5047 return ret;
edbd8d4e
CM		 5048}
5049
1a40e23b
ZY		 5050int btrfs_free_block_groups(struct btrfs_fs_info *info)
5051{
5052 struct btrfs_block_group_cache *block_group;
5053 struct rb_node *n;
5054
1a40e23b
ZY		 5055 spin_lock(&info->block_group_cache_lock);
5056 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
5057 block_group = rb_entry(n, struct btrfs_block_group_cache,
5058 cache_node);
1a40e23b
ZY		 5059 rb_erase(&block_group->cache_node,
5060 &info->block_group_cache_tree);
d899e052
YZ		 5061 spin_unlock(&info->block_group_cache_lock);
5062
5063 btrfs_remove_free_space_cache(block_group);
80eb234a 5064 down_write(&block_group->space_info->groups_sem);
1a40e23b 5065 list_del(&block_group->list);
80eb234a 5066 up_write(&block_group->space_info->groups_sem);
1a40e23b 5067 kfree(block_group);
d899e052
YZ		 5068
5069 spin_lock(&info->block_group_cache_lock);
1a40e23b
ZY		 5070 }
5071 spin_unlock(&info->block_group_cache_lock);
1a40e23b
ZY		 5072 return 0;
5073}
5074
9078a3e1
CM		 5075int btrfs_read_block_groups(struct btrfs_root *root)
5076{
5077 struct btrfs_path *path;
5078 int ret;
9078a3e1 5079 struct btrfs_block_group_cache *cache;
be744175 5080 struct btrfs_fs_info *info = root->fs_info;
6324fbf3 5081 struct btrfs_space_info *space_info;
9078a3e1
CM		 5082 struct btrfs_key key;
5083 struct btrfs_key found_key;
5f39d397 5084 struct extent_buffer *leaf;
96b5179d 5085
be744175 5086 root = info->extent_root;
9078a3e1 5087 key.objectid = 0;
0b86a832 5088 key.offset = 0;
9078a3e1 5089 btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
9078a3e1
CM		 5090 path = btrfs_alloc_path();
5091 if (!path)
5092 return -ENOMEM;
5093
5094 while(1) {
0b86a832
CM		 5095 ret = find_first_block_group(root, path, &key);
5096 if (ret > 0) {
5097 ret = 0;
5098 goto error;
9078a3e1 5099 }
0b86a832
CM		 5100 if (ret != 0)
5101 goto error;
5102
5f39d397
CM		 5103 leaf = path->nodes[0];
5104 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
8f18cf13 5105 cache = kzalloc(sizeof(*cache), GFP_NOFS);
9078a3e1 5106 if (!cache) {
0b86a832 5107 ret = -ENOMEM;
9078a3e1
CM		 5108 break;
5109 }
3e1ad54f 5110
c286ac48 5111 spin_lock_init(&cache->lock);
25179201 5112 mutex_init(&cache->alloc_mutex);
0f9dd46c 5113 INIT_LIST_HEAD(&cache->list);
5f39d397
CM		 5114 read_extent_buffer(leaf, &cache->item,
5115 btrfs_item_ptr_offset(leaf, path->slots[0]),
5116 sizeof(cache->item));
9078a3e1 5117 memcpy(&cache->key, &found_key, sizeof(found_key));
0b86a832 5118
9078a3e1
CM		 5119 key.objectid = found_key.objectid + found_key.offset;
5120 btrfs_release_path(root, path);
0b86a832 5121 cache->flags = btrfs_block_group_flags(&cache->item);
96b5179d 5122
6324fbf3
CM		 5123 ret = update_space_info(info, cache->flags, found_key.offset,
5124 btrfs_block_group_used(&cache->item),
5125 &space_info);
5126 BUG_ON(ret);
5127 cache->space_info = space_info;
80eb234a
JB		 5128 down_write(&space_info->groups_sem);
5129 list_add_tail(&cache->list, &space_info->block_groups);
5130 up_write(&space_info->groups_sem);
0f9dd46c
JB		 5131
5132 ret = btrfs_add_block_group_cache(root->fs_info, cache);
5133 BUG_ON(ret);
75ccf47d
CM		 5134
5135 set_avail_alloc_bits(root->fs_info, cache->flags);
9078a3e1 5136 }
0b86a832
CM		 5137 ret = 0;
5138error:
9078a3e1 5139 btrfs_free_path(path);
0b86a832 5140 return ret;
9078a3e1 5141}
6324fbf3
CM		 5142
5143int btrfs_make_block_group(struct btrfs_trans_handle *trans,
5144 struct btrfs_root *root, u64 bytes_used,
e17cade2 5145 u64 type, u64 chunk_objectid, u64 chunk_offset,
6324fbf3
CM		 5146 u64 size)
5147{
5148 int ret;
6324fbf3
CM		 5149 struct btrfs_root *extent_root;
5150 struct btrfs_block_group_cache *cache;
6324fbf3
CM		 5151
5152 extent_root = root->fs_info->extent_root;
6324fbf3 5153
e02119d5
CM		 5154 root->fs_info->last_trans_new_blockgroup = trans->transid;
5155
8f18cf13 5156 cache = kzalloc(sizeof(*cache), GFP_NOFS);
0f9dd46c
JB		 5157 if (!cache)
5158 return -ENOMEM;
5159
e17cade2 5160 cache->key.objectid = chunk_offset;
6324fbf3 5161 cache->key.offset = size;
c286ac48 5162 spin_lock_init(&cache->lock);
25179201 5163 mutex_init(&cache->alloc_mutex);
0f9dd46c 5164 INIT_LIST_HEAD(&cache->list);
6324fbf3 5165 btrfs_set_key_type(&cache->key, BTRFS_BLOCK_GROUP_ITEM_KEY);
0ef3e66b 5166
6324fbf3 5167 btrfs_set_block_group_used(&cache->item, bytes_used);
6324fbf3
CM		 5168 btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
5169 cache->flags = type;
5170 btrfs_set_block_group_flags(&cache->item, type);
5171
5172 ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
5173 &cache->space_info);
5174 BUG_ON(ret);
80eb234a
JB		 5175 down_write(&cache->space_info->groups_sem);
5176 list_add_tail(&cache->list, &cache->space_info->block_groups);
5177 up_write(&cache->space_info->groups_sem);
6324fbf3 5178
0f9dd46c
JB		 5179 ret = btrfs_add_block_group_cache(root->fs_info, cache);
5180 BUG_ON(ret);
c286ac48 5181
6324fbf3
CM		 5182 ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
5183 sizeof(cache->item));
5184 BUG_ON(ret);
5185
87ef2bb4
CM		 5186 finish_current_insert(trans, extent_root, 0);
5187 ret = del_pending_extents(trans, extent_root, 0);
6324fbf3 5188 BUG_ON(ret);
d18a2c44 5189 set_avail_alloc_bits(extent_root->fs_info, type);
925baedd 5190
6324fbf3
CM		 5191 return 0;
5192}
1a40e23b
ZY		 5193
5194int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
5195 struct btrfs_root *root, u64 group_start)
5196{
5197 struct btrfs_path *path;
5198 struct btrfs_block_group_cache *block_group;
5199 struct btrfs_key key;
5200 int ret;
5201
1a40e23b
ZY		 5202 root = root->fs_info->extent_root;
5203
5204 block_group = btrfs_lookup_block_group(root->fs_info, group_start);
5205 BUG_ON(!block_group);
5206
5207 memcpy(&key, &block_group->key, sizeof(key));
5208
5209 path = btrfs_alloc_path();
5210 BUG_ON(!path);
5211
5212 btrfs_remove_free_space_cache(block_group);
5213 rb_erase(&block_group->cache_node,
5214 &root->fs_info->block_group_cache_tree);
80eb234a 5215 down_write(&block_group->space_info->groups_sem);
1a40e23b 5216 list_del(&block_group->list);
80eb234a 5217 up_write(&block_group->space_info->groups_sem);
1a40e23b
ZY		 5218
5219 /*
5220 memset(shrink_block_group, 0, sizeof(*shrink_block_group));
5221 kfree(shrink_block_group);
5222 */
5223
5224 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
5225 if (ret > 0)
5226 ret = -EIO;
5227 if (ret < 0)
5228 goto out;
5229
5230 ret = btrfs_del_item(trans, root, path);
5231out:
5232 btrfs_free_path(path);
5233 return ret;
5234}
Linux 6.x block layer and io_uring tree(s)