projects / deliverable / linux.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Btrfs: fix leaking block group on balance
[deliverable/linux.git] / fs / btrfs / extent-tree.c
CommitLineData
6cbd5570
CM		 1/*
2 * Copyright (C) 2007 Oracle. All rights reserved.
3 *
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
12 *
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
17 */
ec6b910f 18#include <linux/sched.h>
edbd8d4e 19#include <linux/pagemap.h>
ec44a35c 20#include <linux/writeback.h>
21af804c 21#include <linux/blkdev.h>
4b4e25f2
CM		 22#include <linux/version.h>
23#include "compat.h"
74493f7a 24#include "hash.h"
a5eb62e3 25#include "crc32c.h"
fec577fb
CM		 26#include "ctree.h"
27#include "disk-io.h"
28#include "print-tree.h"
e089f05c 29#include "transaction.h"
0b86a832 30#include "volumes.h"
925baedd 31#include "locking.h"
31153d81 32#include "ref-cache.h"
15916de8 33#include "compat.h"
fec577fb 34
31840ae1
ZY		 35#define PENDING_EXTENT_INSERT 0
36#define PENDING_EXTENT_DELETE 1
37#define PENDING_BACKREF_UPDATE 2
38
39struct pending_extent_op {
40 int type;
41 u64 bytenr;
42 u64 num_bytes;
43 u64 parent;
44 u64 orig_parent;
45 u64 generation;
46 u64 orig_generation;
47 int level;
f3465ca4
JB		 48 struct list_head list;
49 int del;
31840ae1
ZY		 50};
51
e089f05c 52static int finish_current_insert(struct btrfs_trans_handle *trans, struct
87ef2bb4 53 btrfs_root *extent_root, int all);
e20d96d6 54static int del_pending_extents(struct btrfs_trans_handle *trans, struct
87ef2bb4 55 btrfs_root *extent_root, int all);
f3465ca4
JB		 56static int pin_down_bytes(struct btrfs_trans_handle *trans,
57 struct btrfs_root *root,
58 u64 bytenr, u64 num_bytes, int is_data);
59static int update_block_group(struct btrfs_trans_handle *trans,
60 struct btrfs_root *root,
61 u64 bytenr, u64 num_bytes, int alloc,
62 int mark_free);
d548ee51 63
0f9dd46c
JB		 64static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
65{
66 return (cache->flags & bits) == bits;
67}
68
69/*
70 * this adds the block group to the fs_info rb tree for the block group
71 * cache
72 */
b2950863 73static int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
0f9dd46c
JB		 74 struct btrfs_block_group_cache *block_group)
75{
76 struct rb_node **p;
77 struct rb_node *parent = NULL;
78 struct btrfs_block_group_cache *cache;
79
80 spin_lock(&info->block_group_cache_lock);
81 p = &info->block_group_cache_tree.rb_node;
82
83 while (*p) {
84 parent = *p;
85 cache = rb_entry(parent, struct btrfs_block_group_cache,
86 cache_node);
87 if (block_group->key.objectid < cache->key.objectid) {
88 p = &(*p)->rb_left;
89 } else if (block_group->key.objectid > cache->key.objectid) {
90 p = &(*p)->rb_right;
91 } else {
92 spin_unlock(&info->block_group_cache_lock);
93 return -EEXIST;
94 }
95 }
96
97 rb_link_node(&block_group->cache_node, parent, p);
98 rb_insert_color(&block_group->cache_node,
99 &info->block_group_cache_tree);
100 spin_unlock(&info->block_group_cache_lock);
101
102 return 0;
103}
104
105/*
106 * This will return the block group at or after bytenr if contains is 0, else
107 * it will return the block group that contains the bytenr
108 */
109static struct btrfs_block_group_cache *
110block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
111 int contains)
112{
113 struct btrfs_block_group_cache *cache, *ret = NULL;
114 struct rb_node *n;
115 u64 end, start;
116
117 spin_lock(&info->block_group_cache_lock);
118 n = info->block_group_cache_tree.rb_node;
119
120 while (n) {
121 cache = rb_entry(n, struct btrfs_block_group_cache,
122 cache_node);
123 end = cache->key.objectid + cache->key.offset - 1;
124 start = cache->key.objectid;
125
126 if (bytenr < start) {
127 if (!contains && (!ret || start < ret->key.objectid))
128 ret = cache;
129 n = n->rb_left;
130 } else if (bytenr > start) {
131 if (contains && bytenr <= end) {
132 ret = cache;
133 break;
134 }
135 n = n->rb_right;
136 } else {
137 ret = cache;
138 break;
139 }
140 }
d2fb3437
YZ		 141 if (ret)
142 atomic_inc(&ret->count);
0f9dd46c
JB		 143 spin_unlock(&info->block_group_cache_lock);
144
145 return ret;
146}
147
148/*
149 * this is only called by cache_block_group, since we could have freed extents
150 * we need to check the pinned_extents for any extents that can't be used yet
151 * since their free space will be released as soon as the transaction commits.
152 */
153static int add_new_free_space(struct btrfs_block_group_cache *block_group,
154 struct btrfs_fs_info *info, u64 start, u64 end)
155{
156 u64 extent_start, extent_end, size;
157 int ret;
158
25179201 159 mutex_lock(&info->pinned_mutex);
0f9dd46c
JB		 160 while (start < end) {
161 ret = find_first_extent_bit(&info->pinned_extents, start,
162 &extent_start, &extent_end,
163 EXTENT_DIRTY);
164 if (ret)
165 break;
166
167 if (extent_start == start) {
168 start = extent_end + 1;
169 } else if (extent_start > start && extent_start < end) {
170 size = extent_start - start;
ea6a478e
JB		 171 ret = btrfs_add_free_space(block_group, start,
172 size);
0f9dd46c
JB		 173 BUG_ON(ret);
174 start = extent_end + 1;
175 } else {
176 break;
177 }
178 }
179
180 if (start < end) {
181 size = end - start;
ea6a478e 182 ret = btrfs_add_free_space(block_group, start, size);
0f9dd46c
JB		 183 BUG_ON(ret);
184 }
25179201 185 mutex_unlock(&info->pinned_mutex);
0f9dd46c
JB		 186
187 return 0;
188}
189
a512bbf8
YZ		 190static int remove_sb_from_cache(struct btrfs_root *root,
191 struct btrfs_block_group_cache *cache)
192{
193 u64 bytenr;
194 u64 *logical;
195 int stripe_len;
196 int i, nr, ret;
197
198 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
199 bytenr = btrfs_sb_offset(i);
200 ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
201 cache->key.objectid, bytenr, 0,
202 &logical, &nr, &stripe_len);
203 BUG_ON(ret);
204 while (nr--) {
205 btrfs_remove_free_space(cache, logical[nr],
206 stripe_len);
207 }
208 kfree(logical);
209 }
210 return 0;
211}
212
e37c9e69
CM		 213static int cache_block_group(struct btrfs_root *root,
214 struct btrfs_block_group_cache *block_group)
215{
216 struct btrfs_path *path;
ef8bbdfe 217 int ret = 0;
e37c9e69 218 struct btrfs_key key;
5f39d397 219 struct extent_buffer *leaf;
e37c9e69 220 int slot;
a512bbf8 221 u64 last = block_group->key.objectid;
e37c9e69 222
00f5c795
CM		 223 if (!block_group)
224 return 0;
225
e37c9e69 226 root = root->fs_info->extent_root;
e37c9e69
CM		 227
228 if (block_group->cached)
229 return 0;
f510cfec 230
e37c9e69
CM		 231 path = btrfs_alloc_path();
232 if (!path)
233 return -ENOMEM;
7d7d6068 234
2cc58cf2 235 path->reada = 2;
5cd57b2c
CM		 236 /*
237 * we get into deadlocks with paths held by callers of this function.
238 * since the alloc_mutex is protecting things right now, just
239 * skip the locking here
240 */
241 path->skip_locking = 1;
0403e47e 242 key.objectid = max_t(u64, last, BTRFS_SUPER_INFO_OFFSET);
e37c9e69
CM		 243 key.offset = 0;
244 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
245 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
246 if (ret < 0)
ef8bbdfe 247 goto err;
a512bbf8 248
e37c9e69 249 while(1) {
5f39d397 250 leaf = path->nodes[0];
e37c9e69 251 slot = path->slots[0];
5f39d397 252 if (slot >= btrfs_header_nritems(leaf)) {
e37c9e69 253 ret = btrfs_next_leaf(root, path);
54aa1f4d
CM		 254 if (ret < 0)
255 goto err;
0f9dd46c 256 if (ret == 0)
e37c9e69 257 continue;
0f9dd46c 258 else
e37c9e69 259 break;
e37c9e69 260 }
5f39d397 261 btrfs_item_key_to_cpu(leaf, &key, slot);
0f9dd46c 262 if (key.objectid < block_group->key.objectid)
7d7d6068 263 goto next;
0f9dd46c 264
e37c9e69 265 if (key.objectid >= block_group->key.objectid +
0f9dd46c 266 block_group->key.offset)
e37c9e69 267 break;
7d7d6068 268
e37c9e69 269 if (btrfs_key_type(&key) == BTRFS_EXTENT_ITEM_KEY) {
0f9dd46c
JB		 270 add_new_free_space(block_group, root->fs_info, last,
271 key.objectid);
272
7d7d6068 273 last = key.objectid + key.offset;
e37c9e69 274 }
7d7d6068 275next:
e37c9e69
CM		 276 path->slots[0]++;
277 }
278
0f9dd46c
JB		 279 add_new_free_space(block_group, root->fs_info, last,
280 block_group->key.objectid +
281 block_group->key.offset);
282
a512bbf8 283 remove_sb_from_cache(root, block_group);
e37c9e69 284 block_group->cached = 1;
ef8bbdfe 285 ret = 0;
54aa1f4d 286err:
e37c9e69 287 btrfs_free_path(path);
ef8bbdfe 288 return ret;
e37c9e69
CM		 289}
290
0f9dd46c
JB		 291/*
292 * return the block group that starts at or after bytenr
293 */
b2950863 294static struct btrfs_block_group_cache *btrfs_lookup_first_block_group(struct
0ef3e66b
CM		 295 btrfs_fs_info *info,
296 u64 bytenr)
297{
0f9dd46c 298 struct btrfs_block_group_cache *cache;
0ef3e66b 299
0f9dd46c 300 cache = block_group_cache_tree_search(info, bytenr, 0);
0ef3e66b 301
0f9dd46c 302 return cache;
0ef3e66b
CM		 303}
304
0f9dd46c
JB		 305/*
306 * return the block group that contains teh given bytenr
307 */
5276aeda
CM		 308struct btrfs_block_group_cache *btrfs_lookup_block_group(struct
309 btrfs_fs_info *info,
db94535d 310 u64 bytenr)
be744175 311{
0f9dd46c 312 struct btrfs_block_group_cache *cache;
be744175 313
0f9dd46c 314 cache = block_group_cache_tree_search(info, bytenr, 1);
96b5179d 315
0f9dd46c 316 return cache;
be744175 317}
0b86a832 318
d2fb3437
YZ		 319static inline void put_block_group(struct btrfs_block_group_cache *cache)
320{
321 if (atomic_dec_and_test(&cache->count))
322 kfree(cache);
323}
324
0f9dd46c
JB		 325static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
326 u64 flags)
6324fbf3 327{
0f9dd46c
JB		 328 struct list_head *head = &info->space_info;
329 struct list_head *cur;
330 struct btrfs_space_info *found;
331 list_for_each(cur, head) {
332 found = list_entry(cur, struct btrfs_space_info, list);
333 if (found->flags == flags)
334 return found;
335 }
336 return NULL;
6324fbf3
CM		 337}
338
80eb234a
JB		 339static u64 div_factor(u64 num, int factor)
340{
341 if (factor == 10)
342 return num;
343 num *= factor;
344 do_div(num, 10);
345 return num;
346}
347
d2fb3437
YZ		 348u64 btrfs_find_block_group(struct btrfs_root *root,
349 u64 search_start, u64 search_hint, int owner)
cd1bc465 350{
96b5179d 351 struct btrfs_block_group_cache *cache;
cd1bc465 352 u64 used;
d2fb3437
YZ		 353 u64 last = max(search_hint, search_start);
354 u64 group_start = 0;
31f3c99b 355 int full_search = 0;
d2fb3437 356 int factor = 9;
0ef3e66b 357 int wrapped = 0;
31f3c99b 358again:
e8569813
ZY		 359 while (1) {
360 cache = btrfs_lookup_first_block_group(root->fs_info, last);
0f9dd46c
JB		 361 if (!cache)
362 break;
96b5179d 363
c286ac48 364 spin_lock(&cache->lock);
96b5179d
CM		 365 last = cache->key.objectid + cache->key.offset;
366 used = btrfs_block_group_used(&cache->item);
367
d2fb3437
YZ		 368 if ((full_search || !cache->ro) &&
369 block_group_bits(cache, BTRFS_BLOCK_GROUP_METADATA)) {
e8569813 370 if (used + cache->pinned + cache->reserved <
d2fb3437
YZ		 371 div_factor(cache->key.offset, factor)) {
372 group_start = cache->key.objectid;
c286ac48 373 spin_unlock(&cache->lock);
d2fb3437 374 put_block_group(cache);
8790d502
CM		 375 goto found;
376 }
6324fbf3 377 }
c286ac48 378 spin_unlock(&cache->lock);
d2fb3437 379 put_block_group(cache);
de428b63 380 cond_resched();
cd1bc465 381 }
0ef3e66b
CM		 382 if (!wrapped) {
383 last = search_start;
384 wrapped = 1;
385 goto again;
386 }
387 if (!full_search && factor < 10) {
be744175 388 last = search_start;
31f3c99b 389 full_search = 1;
0ef3e66b 390 factor = 10;
31f3c99b
CM		 391 goto again;
392 }
be744175 393found:
d2fb3437 394 return group_start;
925baedd 395}
0f9dd46c 396
e02119d5 397/* simple helper to search for an existing extent at a given offset */
31840ae1 398int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len)
e02119d5
CM		 399{
400 int ret;
401 struct btrfs_key key;
31840ae1 402 struct btrfs_path *path;
e02119d5 403
31840ae1
ZY		 404 path = btrfs_alloc_path();
405 BUG_ON(!path);
e02119d5
CM		 406 key.objectid = start;
407 key.offset = len;
408 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
409 ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
410 0, 0);
31840ae1 411 btrfs_free_path(path);
7bb86316
CM		 412 return ret;
413}
414
d8d5f3e1
CM		 415/*
416 * Back reference rules. Back refs have three main goals:
417 *
418 * 1) differentiate between all holders of references to an extent so that
419 * when a reference is dropped we can make sure it was a valid reference
420 * before freeing the extent.
421 *
422 * 2) Provide enough information to quickly find the holders of an extent
423 * if we notice a given block is corrupted or bad.
424 *
425 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
426 * maintenance. This is actually the same as #2, but with a slightly
427 * different use case.
428 *
429 * File extents can be referenced by:
430 *
431 * - multiple snapshots, subvolumes, or different generations in one subvol
31840ae1 432 * - different files inside a single subvolume
d8d5f3e1
CM		 433 * - different offsets inside a file (bookend extents in file.c)
434 *
435 * The extent ref structure has fields for:
436 *
437 * - Objectid of the subvolume root
438 * - Generation number of the tree holding the reference
439 * - objectid of the file holding the reference
31840ae1
ZY		 440 * - number of references holding by parent node (alway 1 for tree blocks)
441 *
442 * Btree leaf may hold multiple references to a file extent. In most cases,
443 * these references are from same file and the corresponding offsets inside
3bb1a1bc 444 * the file are close together.
d8d5f3e1
CM		 445 *
446 * When a file extent is allocated the fields are filled in:
3bb1a1bc 447 * (root_key.objectid, trans->transid, inode objectid, 1)
d8d5f3e1
CM		 448 *
449 * When a leaf is cow'd new references are added for every file extent found
31840ae1
ZY		 450 * in the leaf. It looks similar to the create case, but trans->transid will
451 * be different when the block is cow'd.
d8d5f3e1 452 *
3bb1a1bc 453 * (root_key.objectid, trans->transid, inode objectid,
31840ae1 454 * number of references in the leaf)
d8d5f3e1 455 *
3bb1a1bc
YZ		 456 * When a file extent is removed either during snapshot deletion or
457 * file truncation, we find the corresponding back reference and check
458 * the following fields:
d8d5f3e1 459 *
3bb1a1bc
YZ		 460 * (btrfs_header_owner(leaf), btrfs_header_generation(leaf),
461 * inode objectid)
d8d5f3e1
CM		 462 *
463 * Btree extents can be referenced by:
464 *
465 * - Different subvolumes
466 * - Different generations of the same subvolume
467 *
d8d5f3e1
CM		 468 * When a tree block is created, back references are inserted:
469 *
3bb1a1bc 470 * (root->root_key.objectid, trans->transid, level, 1)
d8d5f3e1 471 *
31840ae1
ZY		 472 * When a tree block is cow'd, new back references are added for all the
473 * blocks it points to. If the tree block isn't in reference counted root,
474 * the old back references are removed. These new back references are of
475 * the form (trans->transid will have increased since creation):
d8d5f3e1 476 *
3bb1a1bc 477 * (root->root_key.objectid, trans->transid, level, 1)
d8d5f3e1 478 *
31840ae1 479 * When a backref is in deleting, the following fields are checked:
d8d5f3e1
CM		 480 *
481 * if backref was for a tree root:
3bb1a1bc 482 * (btrfs_header_owner(itself), btrfs_header_generation(itself), level)
d8d5f3e1 483 * else
3bb1a1bc 484 * (btrfs_header_owner(parent), btrfs_header_generation(parent), level)
d8d5f3e1 485 *
31840ae1 486 * Back Reference Key composing:
d8d5f3e1 487 *
31840ae1
ZY		 488 * The key objectid corresponds to the first byte in the extent, the key
489 * type is set to BTRFS_EXTENT_REF_KEY, and the key offset is the first
490 * byte of parent extent. If a extent is tree root, the key offset is set
491 * to the key objectid.
d8d5f3e1 492 */
31840ae1
ZY		 493
494static int noinline lookup_extent_backref(struct btrfs_trans_handle *trans,
495 struct btrfs_root *root,
3bb1a1bc
YZ		 496 struct btrfs_path *path,
497 u64 bytenr, u64 parent,
498 u64 ref_root, u64 ref_generation,
499 u64 owner_objectid, int del)
7bb86316 500{
7bb86316 501 struct btrfs_key key;
31840ae1
ZY		 502 struct btrfs_extent_ref *ref;
503 struct extent_buffer *leaf;
3bb1a1bc 504 u64 ref_objectid;
74493f7a
CM		 505 int ret;
506
31840ae1
ZY		 507 key.objectid = bytenr;
508 key.type = BTRFS_EXTENT_REF_KEY;
509 key.offset = parent;
510
511 ret = btrfs_search_slot(trans, root, &key, path, del ? -1 : 0, 1);
512 if (ret < 0)
513 goto out;
514 if (ret > 0) {
515 ret = -ENOENT;
516 goto out;
517 }
518
519 leaf = path->nodes[0];
520 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_ref);
3bb1a1bc 521 ref_objectid = btrfs_ref_objectid(leaf, ref);
31840ae1 522 if (btrfs_ref_root(leaf, ref) != ref_root ||
3bb1a1bc
YZ		 523 btrfs_ref_generation(leaf, ref) != ref_generation ||
524 (ref_objectid != owner_objectid &&
525 ref_objectid != BTRFS_MULTIPLE_OBJECTIDS)) {
31840ae1
ZY		 526 ret = -EIO;
527 WARN_ON(1);
528 goto out;
529 }
530 ret = 0;
531out:
532 return ret;
533}
534
f3465ca4
JB		 535/*
536 * updates all the backrefs that are pending on update_list for the
537 * extent_root
538 */
539static int noinline update_backrefs(struct btrfs_trans_handle *trans,
540 struct btrfs_root *extent_root,
541 struct btrfs_path *path,
542 struct list_head *update_list)
543{
544 struct btrfs_key key;
545 struct btrfs_extent_ref *ref;
546 struct btrfs_fs_info *info = extent_root->fs_info;
547 struct pending_extent_op *op;
548 struct extent_buffer *leaf;
549 int ret = 0;
550 struct list_head *cur = update_list->next;
551 u64 ref_objectid;
552 u64 ref_root = extent_root->root_key.objectid;
553
554 op = list_entry(cur, struct pending_extent_op, list);
555
556search:
557 key.objectid = op->bytenr;
558 key.type = BTRFS_EXTENT_REF_KEY;
559 key.offset = op->orig_parent;
560
561 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 1);
562 BUG_ON(ret);
563
564 leaf = path->nodes[0];
565
566loop:
567 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_ref);
568
569 ref_objectid = btrfs_ref_objectid(leaf, ref);
570
571 if (btrfs_ref_root(leaf, ref) != ref_root ||
572 btrfs_ref_generation(leaf, ref) != op->orig_generation ||
573 (ref_objectid != op->level &&
574 ref_objectid != BTRFS_MULTIPLE_OBJECTIDS)) {
575 printk(KERN_ERR "couldn't find %Lu, parent %Lu, root %Lu, "
576 "owner %u\n", op->bytenr, op->orig_parent,
577 ref_root, op->level);
578 btrfs_print_leaf(extent_root, leaf);
579 BUG();
580 }
581
582 key.objectid = op->bytenr;
583 key.offset = op->parent;
584 key.type = BTRFS_EXTENT_REF_KEY;
585 ret = btrfs_set_item_key_safe(trans, extent_root, path, &key);
586 BUG_ON(ret);
587 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_ref);
588 btrfs_set_ref_generation(leaf, ref, op->generation);
589
590 cur = cur->next;
591
592 list_del_init(&op->list);
593 unlock_extent(&info->extent_ins, op->bytenr,
594 op->bytenr + op->num_bytes - 1, GFP_NOFS);
595 kfree(op);
596
597 if (cur == update_list) {
598 btrfs_mark_buffer_dirty(path->nodes[0]);
599 btrfs_release_path(extent_root, path);
600 goto out;
601 }
602
603 op = list_entry(cur, struct pending_extent_op, list);
604
605 path->slots[0]++;
606 while (path->slots[0] < btrfs_header_nritems(leaf)) {
607 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
608 if (key.objectid == op->bytenr &&
609 key.type == BTRFS_EXTENT_REF_KEY)
610 goto loop;
611 path->slots[0]++;
612 }
613
614 btrfs_mark_buffer_dirty(path->nodes[0]);
615 btrfs_release_path(extent_root, path);
616 goto search;
617
618out:
619 return 0;
620}
621
622static int noinline insert_extents(struct btrfs_trans_handle *trans,
623 struct btrfs_root *extent_root,
624 struct btrfs_path *path,
625 struct list_head *insert_list, int nr)
626{
627 struct btrfs_key *keys;
628 u32 *data_size;
629 struct pending_extent_op *op;
630 struct extent_buffer *leaf;
631 struct list_head *cur = insert_list->next;
632 struct btrfs_fs_info *info = extent_root->fs_info;
633 u64 ref_root = extent_root->root_key.objectid;
634 int i = 0, last = 0, ret;
635 int total = nr * 2;
636
637 if (!nr)
638 return 0;
639
640 keys = kzalloc(total * sizeof(struct btrfs_key), GFP_NOFS);
641 if (!keys)
642 return -ENOMEM;
643
644 data_size = kzalloc(total * sizeof(u32), GFP_NOFS);
645 if (!data_size) {
646 kfree(keys);
647 return -ENOMEM;
648 }
649
650 list_for_each_entry(op, insert_list, list) {
651 keys[i].objectid = op->bytenr;
652 keys[i].offset = op->num_bytes;
653 keys[i].type = BTRFS_EXTENT_ITEM_KEY;
654 data_size[i] = sizeof(struct btrfs_extent_item);
655 i++;
656
657 keys[i].objectid = op->bytenr;
658 keys[i].offset = op->parent;
659 keys[i].type = BTRFS_EXTENT_REF_KEY;
660 data_size[i] = sizeof(struct btrfs_extent_ref);
661 i++;
662 }
663
664 op = list_entry(cur, struct pending_extent_op, list);
665 i = 0;
666 while (i < total) {
667 int c;
668 ret = btrfs_insert_some_items(trans, extent_root, path,
669 keys+i, data_size+i, total-i);
670 BUG_ON(ret < 0);
671
672 if (last && ret > 1)
673 BUG();
674
675 leaf = path->nodes[0];
676 for (c = 0; c < ret; c++) {
677 int ref_first = keys[i].type == BTRFS_EXTENT_REF_KEY;
678
679 /*
680 * if the first item we inserted was a backref, then
681 * the EXTENT_ITEM will be the odd c's, else it will
682 * be the even c's
683 */
684 if ((ref_first && (c % 2)) ||
685 (!ref_first && !(c % 2))) {
686 struct btrfs_extent_item *itm;
687
688 itm = btrfs_item_ptr(leaf, path->slots[0] + c,
689 struct btrfs_extent_item);
690 btrfs_set_extent_refs(path->nodes[0], itm, 1);
691 op->del++;
692 } else {
693 struct btrfs_extent_ref *ref;
694
695 ref = btrfs_item_ptr(leaf, path->slots[0] + c,
696 struct btrfs_extent_ref);
697 btrfs_set_ref_root(leaf, ref, ref_root);
698 btrfs_set_ref_generation(leaf, ref,
699 op->generation);
700 btrfs_set_ref_objectid(leaf, ref, op->level);
701 btrfs_set_ref_num_refs(leaf, ref, 1);
702 op->del++;
703 }
704
705 /*
706 * using del to see when its ok to free up the
707 * pending_extent_op. In the case where we insert the
708 * last item on the list in order to help do batching
709 * we need to not free the extent op until we actually
710 * insert the extent_item
711 */
712 if (op->del == 2) {
713 unlock_extent(&info->extent_ins, op->bytenr,
714 op->bytenr + op->num_bytes - 1,
715 GFP_NOFS);
716 cur = cur->next;
717 list_del_init(&op->list);
718 kfree(op);
719 if (cur != insert_list)
720 op = list_entry(cur,
721 struct pending_extent_op,
722 list);
723 }
724 }
725 btrfs_mark_buffer_dirty(leaf);
726 btrfs_release_path(extent_root, path);
727
728 /*
729 * Ok backref's and items usually go right next to eachother,
730 * but if we could only insert 1 item that means that we
731 * inserted on the end of a leaf, and we have no idea what may
732 * be on the next leaf so we just play it safe. In order to
733 * try and help this case we insert the last thing on our
734 * insert list so hopefully it will end up being the last
735 * thing on the leaf and everything else will be before it,
736 * which will let us insert a whole bunch of items at the same
737 * time.
738 */
739 if (ret == 1 && !last && (i + ret < total)) {
740 /*
741 * last: where we will pick up the next time around
742 * i: our current key to insert, will be total - 1
743 * cur: the current op we are screwing with
744 * op: duh
745 */
746 last = i + ret;
747 i = total - 1;
748 cur = insert_list->prev;
749 op = list_entry(cur, struct pending_extent_op, list);
750 } else if (last) {
751 /*
752 * ok we successfully inserted the last item on the
753 * list, lets reset everything
754 *
755 * i: our current key to insert, so where we left off
756 * last time
757 * last: done with this
758 * cur: the op we are messing with
759 * op: duh
760 * total: since we inserted the last key, we need to
761 * decrement total so we dont overflow
762 */
763 i = last;
764 last = 0;
f3465ca4 765 total--;
b4eec2ca
LH		 766 if (i < total) {
767 cur = insert_list->next;
768 op = list_entry(cur, struct pending_extent_op,
769 list);
770 }
f3465ca4
JB		 771 } else {
772 i += ret;
773 }
774
775 cond_resched();
776 }
777 ret = 0;
778 kfree(keys);
779 kfree(data_size);
780 return ret;
781}
782
31840ae1
ZY		 783static int noinline insert_extent_backref(struct btrfs_trans_handle *trans,
784 struct btrfs_root *root,
785 struct btrfs_path *path,
786 u64 bytenr, u64 parent,
787 u64 ref_root, u64 ref_generation,
3bb1a1bc 788 u64 owner_objectid)
31840ae1
ZY		 789{
790 struct btrfs_key key;
791 struct extent_buffer *leaf;
792 struct btrfs_extent_ref *ref;
793 u32 num_refs;
794 int ret;
74493f7a 795
74493f7a
CM		 796 key.objectid = bytenr;
797 key.type = BTRFS_EXTENT_REF_KEY;
31840ae1 798 key.offset = parent;
74493f7a 799
31840ae1
ZY		 800 ret = btrfs_insert_empty_item(trans, root, path, &key, sizeof(*ref));
801 if (ret == 0) {
802 leaf = path->nodes[0];
803 ref = btrfs_item_ptr(leaf, path->slots[0],
804 struct btrfs_extent_ref);
805 btrfs_set_ref_root(leaf, ref, ref_root);
806 btrfs_set_ref_generation(leaf, ref, ref_generation);
807 btrfs_set_ref_objectid(leaf, ref, owner_objectid);
31840ae1
ZY		 808 btrfs_set_ref_num_refs(leaf, ref, 1);
809 } else if (ret == -EEXIST) {
810 u64 existing_owner;
811 BUG_ON(owner_objectid < BTRFS_FIRST_FREE_OBJECTID);
812 leaf = path->nodes[0];
813 ref = btrfs_item_ptr(leaf, path->slots[0],
814 struct btrfs_extent_ref);
815 if (btrfs_ref_root(leaf, ref) != ref_root ||
816 btrfs_ref_generation(leaf, ref) != ref_generation) {
817 ret = -EIO;
818 WARN_ON(1);
7bb86316 819 goto out;
31840ae1
ZY		 820 }
821
822 num_refs = btrfs_ref_num_refs(leaf, ref);
823 BUG_ON(num_refs == 0);
824 btrfs_set_ref_num_refs(leaf, ref, num_refs + 1);
825
826 existing_owner = btrfs_ref_objectid(leaf, ref);
3bb1a1bc
YZ		 827 if (existing_owner != owner_objectid &&
828 existing_owner != BTRFS_MULTIPLE_OBJECTIDS) {
31840ae1
ZY		 829 btrfs_set_ref_objectid(leaf, ref,
830 BTRFS_MULTIPLE_OBJECTIDS);
31840ae1
ZY		 831 }
832 ret = 0;
833 } else {
7bb86316 834 goto out;
31840ae1 835 }
7bb86316
CM		 836 btrfs_mark_buffer_dirty(path->nodes[0]);
837out:
838 btrfs_release_path(root, path);
839 return ret;
74493f7a
CM		 840}
841
31840ae1
ZY		 842static int noinline remove_extent_backref(struct btrfs_trans_handle *trans,
843 struct btrfs_root *root,
844 struct btrfs_path *path)
845{
846 struct extent_buffer *leaf;
847 struct btrfs_extent_ref *ref;
848 u32 num_refs;
849 int ret = 0;
850
851 leaf = path->nodes[0];
852 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_ref);
853 num_refs = btrfs_ref_num_refs(leaf, ref);
854 BUG_ON(num_refs == 0);
855 num_refs -= 1;
856 if (num_refs == 0) {
857 ret = btrfs_del_item(trans, root, path);
858 } else {
859 btrfs_set_ref_num_refs(leaf, ref, num_refs);
860 btrfs_mark_buffer_dirty(leaf);
861 }
862 btrfs_release_path(root, path);
863 return ret;
864}
865
4b4e25f2 866#ifdef BIO_RW_DISCARD
15916de8
CM		 867static void btrfs_issue_discard(struct block_device *bdev,
868 u64 start, u64 len)
869{
870#if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,28)
871 blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_KERNEL);
872#else
873 blkdev_issue_discard(bdev, start >> 9, len >> 9);
874#endif
875}
4b4e25f2 876#endif
15916de8 877
f3465ca4
JB		 878static int noinline free_extents(struct btrfs_trans_handle *trans,
879 struct btrfs_root *extent_root,
880 struct list_head *del_list)
881{
882 struct btrfs_fs_info *info = extent_root->fs_info;
883 struct btrfs_path *path;
884 struct btrfs_key key, found_key;
885 struct extent_buffer *leaf;
886 struct list_head *cur;
887 struct pending_extent_op *op;
888 struct btrfs_extent_item *ei;
889 int ret, num_to_del, extent_slot = 0, found_extent = 0;
890 u32 refs;
891 u64 bytes_freed = 0;
892
893 path = btrfs_alloc_path();
894 if (!path)
895 return -ENOMEM;
896 path->reada = 1;
897
898search:
899 /* search for the backref for the current ref we want to delete */
900 cur = del_list->next;
901 op = list_entry(cur, struct pending_extent_op, list);
902 ret = lookup_extent_backref(trans, extent_root, path, op->bytenr,
903 op->orig_parent,
904 extent_root->root_key.objectid,
905 op->orig_generation, op->level, 1);
906 if (ret) {
907 printk("Unable to find backref byte nr %Lu root %Lu gen %Lu "
908 "owner %u\n", op->bytenr,
909 extent_root->root_key.objectid, op->orig_generation,
910 op->level);
911 btrfs_print_leaf(extent_root, path->nodes[0]);
912 WARN_ON(1);
913 goto out;
914 }
915
916 extent_slot = path->slots[0];
917 num_to_del = 1;
918 found_extent = 0;
919
920 /*
921 * if we aren't the first item on the leaf we can move back one and see
922 * if our ref is right next to our extent item
923 */
924 if (likely(extent_slot)) {
925 extent_slot--;
926 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
927 extent_slot);
928 if (found_key.objectid == op->bytenr &&
929 found_key.type == BTRFS_EXTENT_ITEM_KEY &&
930 found_key.offset == op->num_bytes) {
931 num_to_del++;
932 found_extent = 1;
933 }
934 }
935
936 /*
937 * if we didn't find the extent we need to delete the backref and then
938 * search for the extent item key so we can update its ref count
939 */
940 if (!found_extent) {
941 key.objectid = op->bytenr;
942 key.type = BTRFS_EXTENT_ITEM_KEY;
943 key.offset = op->num_bytes;
944
945 ret = remove_extent_backref(trans, extent_root, path);
946 BUG_ON(ret);
947 btrfs_release_path(extent_root, path);
948 ret = btrfs_search_slot(trans, extent_root, &key, path, -1, 1);
949 BUG_ON(ret);
950 extent_slot = path->slots[0];
951 }
952
953 /* this is where we update the ref count for the extent */
954 leaf = path->nodes[0];
955 ei = btrfs_item_ptr(leaf, extent_slot, struct btrfs_extent_item);
956 refs = btrfs_extent_refs(leaf, ei);
957 BUG_ON(refs == 0);
958 refs--;
959 btrfs_set_extent_refs(leaf, ei, refs);
960
961 btrfs_mark_buffer_dirty(leaf);
962
963 /*
964 * This extent needs deleting. The reason cur_slot is extent_slot +
965 * num_to_del is because extent_slot points to the slot where the extent
966 * is, and if the backref was not right next to the extent we will be
967 * deleting at least 1 item, and will want to start searching at the
968 * slot directly next to extent_slot. However if we did find the
969 * backref next to the extent item them we will be deleting at least 2
970 * items and will want to start searching directly after the ref slot
971 */
972 if (!refs) {
973 struct list_head *pos, *n, *end;
974 int cur_slot = extent_slot+num_to_del;
975 u64 super_used;
976 u64 root_used;
977
978 path->slots[0] = extent_slot;
979 bytes_freed = op->num_bytes;
980
e3e469f8
JB		 981 mutex_lock(&info->pinned_mutex);
982 ret = pin_down_bytes(trans, extent_root, op->bytenr,
983 op->num_bytes, op->level >=
984 BTRFS_FIRST_FREE_OBJECTID);
985 mutex_unlock(&info->pinned_mutex);
986 BUG_ON(ret < 0);
987 op->del = ret;
988
f3465ca4
JB		 989 /*
990 * we need to see if we can delete multiple things at once, so
991 * start looping through the list of extents we are wanting to
992 * delete and see if their extent/backref's are right next to
993 * eachother and the extents only have 1 ref
994 */
995 for (pos = cur->next; pos != del_list; pos = pos->next) {
996 struct pending_extent_op *tmp;
997
998 tmp = list_entry(pos, struct pending_extent_op, list);
999
1000 /* we only want to delete extent+ref at this stage */
1001 if (cur_slot >= btrfs_header_nritems(leaf) - 1)
1002 break;
1003
1004 btrfs_item_key_to_cpu(leaf, &found_key, cur_slot);
1005 if (found_key.objectid != tmp->bytenr ||
1006 found_key.type != BTRFS_EXTENT_ITEM_KEY ||
1007 found_key.offset != tmp->num_bytes)
1008 break;
1009
1010 /* check to make sure this extent only has one ref */
1011 ei = btrfs_item_ptr(leaf, cur_slot,
1012 struct btrfs_extent_item);
1013 if (btrfs_extent_refs(leaf, ei) != 1)
1014 break;
1015
1016 btrfs_item_key_to_cpu(leaf, &found_key, cur_slot+1);
1017 if (found_key.objectid != tmp->bytenr ||
1018 found_key.type != BTRFS_EXTENT_REF_KEY ||
1019 found_key.offset != tmp->orig_parent)
1020 break;
1021
1022 /*
1023 * the ref is right next to the extent, we can set the
1024 * ref count to 0 since we will delete them both now
1025 */
1026 btrfs_set_extent_refs(leaf, ei, 0);
1027
1028 /* pin down the bytes for this extent */
1029 mutex_lock(&info->pinned_mutex);
1030 ret = pin_down_bytes(trans, extent_root, tmp->bytenr,
1031 tmp->num_bytes, tmp->level >=
1032 BTRFS_FIRST_FREE_OBJECTID);
1033 mutex_unlock(&info->pinned_mutex);
1034 BUG_ON(ret < 0);
1035
1036 /*
1037 * use the del field to tell if we need to go ahead and
1038 * free up the extent when we delete the item or not.
1039 */
1040 tmp->del = ret;
1041 bytes_freed += tmp->num_bytes;
1042
1043 num_to_del += 2;
1044 cur_slot += 2;
1045 }
1046 end = pos;
1047
1048 /* update the free space counters */
1049 spin_lock_irq(&info->delalloc_lock);
1050 super_used = btrfs_super_bytes_used(&info->super_copy);
1051 btrfs_set_super_bytes_used(&info->super_copy,
1052 super_used - bytes_freed);
1053 spin_unlock_irq(&info->delalloc_lock);
1054
1055 root_used = btrfs_root_used(&extent_root->root_item);
1056 btrfs_set_root_used(&extent_root->root_item,
1057 root_used - bytes_freed);
1058
1059 /* delete the items */
1060 ret = btrfs_del_items(trans, extent_root, path,
1061 path->slots[0], num_to_del);
1062 BUG_ON(ret);
1063
1064 /*
1065 * loop through the extents we deleted and do the cleanup work
1066 * on them
1067 */
1068 for (pos = cur, n = pos->next; pos != end;
1069 pos = n, n = pos->next) {
1070 struct pending_extent_op *tmp;
1071#ifdef BIO_RW_DISCARD
1072 u64 map_length;
1073 struct btrfs_multi_bio *multi = NULL;
1074#endif
1075 tmp = list_entry(pos, struct pending_extent_op, list);
1076
1077 /*
1078 * remember tmp->del tells us wether or not we pinned
1079 * down the extent
1080 */
1081 ret = update_block_group(trans, extent_root,
1082 tmp->bytenr, tmp->num_bytes, 0,
1083 tmp->del);
1084 BUG_ON(ret);
1085
1086#ifdef BIO_RW_DISCARD
15916de8 1087 map_length = tmp->num_bytes;
f3465ca4
JB		 1088 ret = btrfs_map_block(&info->mapping_tree, READ,
1089 tmp->bytenr, &map_length, &multi,
1090 0);
1091 if (!ret) {
1092 struct btrfs_bio_stripe *stripe;
1093 int i;
1094
15916de8 1095 stripe = multi->stripes;
f3465ca4
JB		 1096
1097 if (map_length > tmp->num_bytes)
1098 map_length = tmp->num_bytes;
1099
1100 for (i = 0; i < multi->num_stripes;
1101 i++, stripe++)
15916de8
CM		 1102 btrfs_issue_discard(stripe->dev->bdev,
1103 stripe->physical,
1104 map_length);
f3465ca4
JB		 1105 kfree(multi);
1106 }
1107#endif
1108 list_del_init(&tmp->list);
1109 unlock_extent(&info->extent_ins, tmp->bytenr,
1110 tmp->bytenr + tmp->num_bytes - 1,
1111 GFP_NOFS);
1112 kfree(tmp);
1113 }
1114 } else if (refs && found_extent) {
1115 /*
1116 * the ref and extent were right next to eachother, but the
1117 * extent still has a ref, so just free the backref and keep
1118 * going
1119 */
1120 ret = remove_extent_backref(trans, extent_root, path);
1121 BUG_ON(ret);
1122
1123 list_del_init(&op->list);
1124 unlock_extent(&info->extent_ins, op->bytenr,
1125 op->bytenr + op->num_bytes - 1, GFP_NOFS);
1126 kfree(op);
1127 } else {
1128 /*
1129 * the extent has multiple refs and the backref we were looking
1130 * for was not right next to it, so just unlock and go next,
1131 * we're good to go
1132 */
1133 list_del_init(&op->list);
1134 unlock_extent(&info->extent_ins, op->bytenr,
1135 op->bytenr + op->num_bytes - 1, GFP_NOFS);
1136 kfree(op);
1137 }
1138
1139 btrfs_release_path(extent_root, path);
1140 if (!list_empty(del_list))
1141 goto search;
1142
1143out:
1144 btrfs_free_path(path);
1145 return ret;
1146}
1147
31840ae1
ZY		 1148static int __btrfs_update_extent_ref(struct btrfs_trans_handle *trans,
1149 struct btrfs_root *root, u64 bytenr,
1150 u64 orig_parent, u64 parent,
1151 u64 orig_root, u64 ref_root,
1152 u64 orig_generation, u64 ref_generation,
3bb1a1bc 1153 u64 owner_objectid)
31840ae1
ZY		 1154{
1155 int ret;
1156 struct btrfs_root *extent_root = root->fs_info->extent_root;
1157 struct btrfs_path *path;
1158
1159 if (root == root->fs_info->extent_root) {
1160 struct pending_extent_op *extent_op;
1161 u64 num_bytes;
1162
1163 BUG_ON(owner_objectid >= BTRFS_MAX_LEVEL);
1164 num_bytes = btrfs_level_size(root, (int)owner_objectid);
25179201 1165 mutex_lock(&root->fs_info->extent_ins_mutex);
31840ae1 1166 if (test_range_bit(&root->fs_info->extent_ins, bytenr,
25179201 1167 bytenr + num_bytes - 1, EXTENT_WRITEBACK, 0)) {
31840ae1
ZY		 1168 u64 priv;
1169 ret = get_state_private(&root->fs_info->extent_ins,
1170 bytenr, &priv);
1171 BUG_ON(ret);
1172 extent_op = (struct pending_extent_op *)
1173 (unsigned long)priv;
1174 BUG_ON(extent_op->parent != orig_parent);
1175 BUG_ON(extent_op->generation != orig_generation);
25179201 1176
31840ae1
ZY		 1177 extent_op->parent = parent;
1178 extent_op->generation = ref_generation;
1179 } else {
1180 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
1181 BUG_ON(!extent_op);
1182
1183 extent_op->type = PENDING_BACKREF_UPDATE;
1184 extent_op->bytenr = bytenr;
1185 extent_op->num_bytes = num_bytes;
1186 extent_op->parent = parent;
1187 extent_op->orig_parent = orig_parent;
1188 extent_op->generation = ref_generation;
1189 extent_op->orig_generation = orig_generation;
1190 extent_op->level = (int)owner_objectid;
f3465ca4
JB		 1191 INIT_LIST_HEAD(&extent_op->list);
1192 extent_op->del = 0;
31840ae1
ZY		 1193
1194 set_extent_bits(&root->fs_info->extent_ins,
1195 bytenr, bytenr + num_bytes - 1,
25179201 1196 EXTENT_WRITEBACK, GFP_NOFS);
31840ae1
ZY		 1197 set_state_private(&root->fs_info->extent_ins,
1198 bytenr, (unsigned long)extent_op);
1199 }
25179201 1200 mutex_unlock(&root->fs_info->extent_ins_mutex);
31840ae1
ZY		 1201 return 0;
1202 }
1203
1204 path = btrfs_alloc_path();
1205 if (!path)
1206 return -ENOMEM;
1207 ret = lookup_extent_backref(trans, extent_root, path,
1208 bytenr, orig_parent, orig_root,
3bb1a1bc 1209 orig_generation, owner_objectid, 1);
31840ae1
ZY		 1210 if (ret)
1211 goto out;
1212 ret = remove_extent_backref(trans, extent_root, path);
1213 if (ret)
1214 goto out;
1215 ret = insert_extent_backref(trans, extent_root, path, bytenr,
1216 parent, ref_root, ref_generation,
3bb1a1bc 1217 owner_objectid);
31840ae1 1218 BUG_ON(ret);
87ef2bb4
CM		 1219 finish_current_insert(trans, extent_root, 0);
1220 del_pending_extents(trans, extent_root, 0);
31840ae1
ZY		 1221out:
1222 btrfs_free_path(path);
1223 return ret;
1224}
1225
1226int btrfs_update_extent_ref(struct btrfs_trans_handle *trans,
1227 struct btrfs_root *root, u64 bytenr,
1228 u64 orig_parent, u64 parent,
1229 u64 ref_root, u64 ref_generation,
3bb1a1bc 1230 u64 owner_objectid)
31840ae1
ZY		 1231{
1232 int ret;
1233 if (ref_root == BTRFS_TREE_LOG_OBJECTID &&
1234 owner_objectid < BTRFS_FIRST_FREE_OBJECTID)
1235 return 0;
31840ae1
ZY		 1236 ret = __btrfs_update_extent_ref(trans, root, bytenr, orig_parent,
1237 parent, ref_root, ref_root,
1238 ref_generation, ref_generation,
3bb1a1bc 1239 owner_objectid);
31840ae1
ZY		 1240 return ret;
1241}
1242
925baedd 1243static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
31840ae1
ZY		 1244 struct btrfs_root *root, u64 bytenr,
1245 u64 orig_parent, u64 parent,
1246 u64 orig_root, u64 ref_root,
1247 u64 orig_generation, u64 ref_generation,
3bb1a1bc 1248 u64 owner_objectid)
02217ed2 1249{
5caf2a00 1250 struct btrfs_path *path;
02217ed2 1251 int ret;
e2fa7227 1252 struct btrfs_key key;
5f39d397 1253 struct extent_buffer *l;
234b63a0 1254 struct btrfs_extent_item *item;
cf27e1ee 1255 u32 refs;
037e6390 1256
5caf2a00 1257 path = btrfs_alloc_path();
54aa1f4d
CM		 1258 if (!path)
1259 return -ENOMEM;
26b8003f 1260
3c12ac72 1261 path->reada = 1;
db94535d 1262 key.objectid = bytenr;
31840ae1
ZY		 1263 key.type = BTRFS_EXTENT_ITEM_KEY;
1264 key.offset = (u64)-1;
1265
5caf2a00 1266 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path,
9f5fae2f 1267 0, 1);
54aa1f4d
CM		 1268 if (ret < 0)
1269 return ret;
31840ae1
ZY		 1270 BUG_ON(ret == 0 || path->slots[0] == 0);
1271
1272 path->slots[0]--;
5f39d397 1273 l = path->nodes[0];
31840ae1
ZY		 1274
1275 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
771ed689
CM		 1276 if (key.objectid != bytenr) {
1277 btrfs_print_leaf(root->fs_info->extent_root, path->nodes[0]);
1278 printk("wanted %Lu found %Lu\n", bytenr, key.objectid);
1279 BUG();
1280 }
31840ae1
ZY		 1281 BUG_ON(key.type != BTRFS_EXTENT_ITEM_KEY);
1282
5caf2a00 1283 item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
5f39d397
CM		 1284 refs = btrfs_extent_refs(l, item);
1285 btrfs_set_extent_refs(l, item, refs + 1);
5caf2a00 1286 btrfs_mark_buffer_dirty(path->nodes[0]);
a28ec197 1287
5caf2a00 1288 btrfs_release_path(root->fs_info->extent_root, path);
7bb86316 1289
3c12ac72 1290 path->reada = 1;
31840ae1
ZY		 1291 ret = insert_extent_backref(trans, root->fs_info->extent_root,
1292 path, bytenr, parent,
1293 ref_root, ref_generation,
3bb1a1bc 1294 owner_objectid);
7bb86316 1295 BUG_ON(ret);
87ef2bb4
CM		 1296 finish_current_insert(trans, root->fs_info->extent_root, 0);
1297 del_pending_extents(trans, root->fs_info->extent_root, 0);
74493f7a
CM		 1298
1299 btrfs_free_path(path);
02217ed2
CM		 1300 return 0;
1301}
1302
925baedd 1303int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
31840ae1
ZY		 1304 struct btrfs_root *root,
1305 u64 bytenr, u64 num_bytes, u64 parent,
1306 u64 ref_root, u64 ref_generation,
3bb1a1bc 1307 u64 owner_objectid)
925baedd
CM		 1308{
1309 int ret;
31840ae1
ZY		 1310 if (ref_root == BTRFS_TREE_LOG_OBJECTID &&
1311 owner_objectid < BTRFS_FIRST_FREE_OBJECTID)
1312 return 0;
31840ae1
ZY		 1313 ret = __btrfs_inc_extent_ref(trans, root, bytenr, 0, parent,
1314 0, ref_root, 0, ref_generation,
3bb1a1bc 1315 owner_objectid);
925baedd
CM		 1316 return ret;
1317}
1318
e9d0b13b
CM		 1319int btrfs_extent_post_op(struct btrfs_trans_handle *trans,
1320 struct btrfs_root *root)
1321{
87ef2bb4
CM		 1322 finish_current_insert(trans, root->fs_info->extent_root, 1);
1323 del_pending_extents(trans, root->fs_info->extent_root, 1);
e9d0b13b
CM		 1324 return 0;
1325}
1326
31840ae1
ZY		 1327int btrfs_lookup_extent_ref(struct btrfs_trans_handle *trans,
1328 struct btrfs_root *root, u64 bytenr,
1329 u64 num_bytes, u32 *refs)
a28ec197 1330{
5caf2a00 1331 struct btrfs_path *path;
a28ec197 1332 int ret;
e2fa7227 1333 struct btrfs_key key;
5f39d397 1334 struct extent_buffer *l;
234b63a0 1335 struct btrfs_extent_item *item;
5caf2a00 1336
db94535d 1337 WARN_ON(num_bytes < root->sectorsize);
5caf2a00 1338 path = btrfs_alloc_path();
3c12ac72 1339 path->reada = 1;
db94535d
CM		 1340 key.objectid = bytenr;
1341 key.offset = num_bytes;
62e2749e 1342 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
5caf2a00 1343 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path,
9f5fae2f 1344 0, 0);
54aa1f4d
CM		 1345 if (ret < 0)
1346 goto out;
5f39d397
CM		 1347 if (ret != 0) {
1348 btrfs_print_leaf(root, path->nodes[0]);
db94535d 1349 printk("failed to find block number %Lu\n", bytenr);
a28ec197 1350 BUG();
5f39d397
CM		 1351 }
1352 l = path->nodes[0];
5caf2a00 1353 item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
5f39d397 1354 *refs = btrfs_extent_refs(l, item);
54aa1f4d 1355out:
5caf2a00 1356 btrfs_free_path(path);
a28ec197
CM		 1357 return 0;
1358}
1359
80ff3856
YZ		 1360int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
1361 struct btrfs_root *root, u64 bytenr)
be20aa9d
CM		 1362{
1363 struct btrfs_root *extent_root = root->fs_info->extent_root;
1364 struct btrfs_path *path;
f321e491
YZ		 1365 struct extent_buffer *leaf;
1366 struct btrfs_extent_ref *ref_item;
1367 struct btrfs_key key;
1368 struct btrfs_key found_key;
80ff3856
YZ		 1369 u64 ref_root;
1370 u64 last_snapshot;
be20aa9d
CM		 1371 u32 nritems;
1372 int ret;
925baedd 1373
be20aa9d 1374 key.objectid = bytenr;
31840ae1 1375 key.offset = (u64)-1;
f321e491 1376 key.type = BTRFS_EXTENT_ITEM_KEY;
be20aa9d 1377
f321e491 1378 path = btrfs_alloc_path();
be20aa9d
CM		 1379 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
1380 if (ret < 0)
1381 goto out;
1382 BUG_ON(ret == 0);
80ff3856
YZ		 1383
1384 ret = -ENOENT;
1385 if (path->slots[0] == 0)
31840ae1 1386 goto out;
be20aa9d 1387
31840ae1 1388 path->slots[0]--;
f321e491
YZ		 1389 leaf = path->nodes[0];
1390 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
be20aa9d
CM		 1391
1392 if (found_key.objectid != bytenr ||
80ff3856 1393 found_key.type != BTRFS_EXTENT_ITEM_KEY)
be20aa9d 1394 goto out;
f321e491 1395
80ff3856 1396 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
be20aa9d 1397 while (1) {
f321e491
YZ		 1398 leaf = path->nodes[0];
1399 nritems = btrfs_header_nritems(leaf);
be20aa9d
CM		 1400 if (path->slots[0] >= nritems) {
1401 ret = btrfs_next_leaf(extent_root, path);
f321e491
YZ		 1402 if (ret < 0)
1403 goto out;
be20aa9d
CM		 1404 if (ret == 0)
1405 continue;
1406 break;
1407 }
f321e491 1408 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
be20aa9d
CM		 1409 if (found_key.objectid != bytenr)
1410 break;
bd09835d 1411
be20aa9d
CM		 1412 if (found_key.type != BTRFS_EXTENT_REF_KEY) {
1413 path->slots[0]++;
1414 continue;
1415 }
1416
f321e491 1417 ref_item = btrfs_item_ptr(leaf, path->slots[0],
be20aa9d 1418 struct btrfs_extent_ref);
80ff3856
YZ		 1419 ref_root = btrfs_ref_root(leaf, ref_item);
1420 if (ref_root != root->root_key.objectid &&
1421 ref_root != BTRFS_TREE_LOG_OBJECTID) {
1422 ret = 1;
f321e491 1423 goto out;
80ff3856
YZ		 1424 }
1425 if (btrfs_ref_generation(leaf, ref_item) <= last_snapshot) {
f321e491
YZ		 1426 ret = 1;
1427 goto out;
1428 }
80ff3856
YZ		 1429
1430 path->slots[0]++;
f321e491
YZ		 1431 }
1432 ret = 0;
be20aa9d 1433out:
80ff3856 1434 btrfs_free_path(path);
f321e491 1435 return ret;
be20aa9d 1436}
c5739bba 1437
31840ae1
ZY		 1438int btrfs_cache_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1439 struct extent_buffer *buf, u32 nr_extents)
02217ed2 1440{
5f39d397 1441 struct btrfs_key key;
6407bf6d 1442 struct btrfs_file_extent_item *fi;
e4657689
ZY		 1443 u64 root_gen;
1444 u32 nritems;
02217ed2 1445 int i;
db94535d 1446 int level;
31840ae1 1447 int ret = 0;
e4657689 1448 int shared = 0;
a28ec197 1449
3768f368 1450 if (!root->ref_cows)
a28ec197 1451 return 0;
5f39d397 1452
e4657689
ZY		 1453 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
1454 shared = 0;
1455 root_gen = root->root_key.offset;
1456 } else {
1457 shared = 1;
1458 root_gen = trans->transid - 1;
1459 }
1460
db94535d 1461 level = btrfs_header_level(buf);
5f39d397 1462 nritems = btrfs_header_nritems(buf);
4a096752 1463
31840ae1 1464 if (level == 0) {
31153d81
YZ		 1465 struct btrfs_leaf_ref *ref;
1466 struct btrfs_extent_info *info;
1467
31840ae1 1468 ref = btrfs_alloc_leaf_ref(root, nr_extents);
31153d81 1469 if (!ref) {
31840ae1 1470 ret = -ENOMEM;
31153d81
YZ		 1471 goto out;
1472 }
1473
e4657689 1474 ref->root_gen = root_gen;
31153d81
YZ		 1475 ref->bytenr = buf->start;
1476 ref->owner = btrfs_header_owner(buf);
1477 ref->generation = btrfs_header_generation(buf);
31840ae1 1478 ref->nritems = nr_extents;
31153d81 1479 info = ref->extents;
bcc63abb 1480
31840ae1 1481 for (i = 0; nr_extents > 0 && i < nritems; i++) {
31153d81
YZ		 1482 u64 disk_bytenr;
1483 btrfs_item_key_to_cpu(buf, &key, i);
1484 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
1485 continue;
1486 fi = btrfs_item_ptr(buf, i,
1487 struct btrfs_file_extent_item);
1488 if (btrfs_file_extent_type(buf, fi) ==
1489 BTRFS_FILE_EXTENT_INLINE)
1490 continue;
1491 disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
1492 if (disk_bytenr == 0)
1493 continue;
1494
1495 info->bytenr = disk_bytenr;
1496 info->num_bytes =
1497 btrfs_file_extent_disk_num_bytes(buf, fi);
1498 info->objectid = key.objectid;
1499 info->offset = key.offset;
1500 info++;
1501 }
1502
e4657689 1503 ret = btrfs_add_leaf_ref(root, ref, shared);
5b84e8d6
YZ		 1504 if (ret == -EEXIST && shared) {
1505 struct btrfs_leaf_ref *old;
1506 old = btrfs_lookup_leaf_ref(root, ref->bytenr);
1507 BUG_ON(!old);
1508 btrfs_remove_leaf_ref(root, old);
1509 btrfs_free_leaf_ref(root, old);
1510 ret = btrfs_add_leaf_ref(root, ref, shared);
1511 }
31153d81 1512 WARN_ON(ret);
bcc63abb 1513 btrfs_free_leaf_ref(root, ref);
31153d81
YZ		 1514 }
1515out:
31840ae1
ZY		 1516 return ret;
1517}
1518
1519int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1520 struct extent_buffer *orig_buf, struct extent_buffer *buf,
1521 u32 *nr_extents)
1522{
1523 u64 bytenr;
1524 u64 ref_root;
1525 u64 orig_root;
1526 u64 ref_generation;
1527 u64 orig_generation;
1528 u32 nritems;
1529 u32 nr_file_extents = 0;
1530 struct btrfs_key key;
1531 struct btrfs_file_extent_item *fi;
1532 int i;
1533 int level;
1534 int ret = 0;
1535 int faili = 0;
1536 int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
3bb1a1bc 1537 u64, u64, u64, u64, u64, u64, u64, u64);
31840ae1
ZY		 1538
1539 ref_root = btrfs_header_owner(buf);
1540 ref_generation = btrfs_header_generation(buf);
1541 orig_root = btrfs_header_owner(orig_buf);
1542 orig_generation = btrfs_header_generation(orig_buf);
1543
1544 nritems = btrfs_header_nritems(buf);
1545 level = btrfs_header_level(buf);
1546
1547 if (root->ref_cows) {
1548 process_func = __btrfs_inc_extent_ref;
1549 } else {
1550 if (level == 0 &&
1551 root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
1552 goto out;
1553 if (level != 0 &&
1554 root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID)
1555 goto out;
1556 process_func = __btrfs_update_extent_ref;
1557 }
1558
1559 for (i = 0; i < nritems; i++) {
1560 cond_resched();
db94535d 1561 if (level == 0) {
5f39d397
CM		 1562 btrfs_item_key_to_cpu(buf, &key, i);
1563 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
54aa1f4d 1564 continue;
5f39d397 1565 fi = btrfs_item_ptr(buf, i,
54aa1f4d 1566 struct btrfs_file_extent_item);
5f39d397 1567 if (btrfs_file_extent_type(buf, fi) ==
54aa1f4d
CM		 1568 BTRFS_FILE_EXTENT_INLINE)
1569 continue;
31840ae1
ZY		 1570 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
1571 if (bytenr == 0)
54aa1f4d 1572 continue;
31840ae1
ZY		 1573
1574 nr_file_extents++;
1575
31840ae1
ZY		 1576 ret = process_func(trans, root, bytenr,
1577 orig_buf->start, buf->start,
1578 orig_root, ref_root,
1579 orig_generation, ref_generation,
3bb1a1bc 1580 key.objectid);
31840ae1
ZY		 1581
1582 if (ret) {
1583 faili = i;
1584 WARN_ON(1);
1585 goto fail;
1586 }
54aa1f4d 1587 } else {
db94535d 1588 bytenr = btrfs_node_blockptr(buf, i);
31840ae1
ZY		 1589 ret = process_func(trans, root, bytenr,
1590 orig_buf->start, buf->start,
1591 orig_root, ref_root,
1592 orig_generation, ref_generation,
3bb1a1bc 1593 level - 1);
31840ae1
ZY		 1594 if (ret) {
1595 faili = i;
1596 WARN_ON(1);
1597 goto fail;
1598 }
54aa1f4d
CM		 1599 }
1600 }
31840ae1
ZY		 1601out:
1602 if (nr_extents) {
1603 if (level == 0)
1604 *nr_extents = nr_file_extents;
1605 else
1606 *nr_extents = nritems;
1607 }
1608 return 0;
1609fail:
1610 WARN_ON(1);
54aa1f4d 1611 return ret;
02217ed2
CM		 1612}
1613
31840ae1
ZY		 1614int btrfs_update_ref(struct btrfs_trans_handle *trans,
1615 struct btrfs_root *root, struct extent_buffer *orig_buf,
1616 struct extent_buffer *buf, int start_slot, int nr)
1617
1618{
1619 u64 bytenr;
1620 u64 ref_root;
1621 u64 orig_root;
1622 u64 ref_generation;
1623 u64 orig_generation;
1624 struct btrfs_key key;
1625 struct btrfs_file_extent_item *fi;
1626 int i;
1627 int ret;
1628 int slot;
1629 int level;
1630
1631 BUG_ON(start_slot < 0);
1632 BUG_ON(start_slot + nr > btrfs_header_nritems(buf));
1633
1634 ref_root = btrfs_header_owner(buf);
1635 ref_generation = btrfs_header_generation(buf);
1636 orig_root = btrfs_header_owner(orig_buf);
1637 orig_generation = btrfs_header_generation(orig_buf);
1638 level = btrfs_header_level(buf);
1639
1640 if (!root->ref_cows) {
1641 if (level == 0 &&
1642 root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
1643 return 0;
1644 if (level != 0 &&
1645 root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID)
1646 return 0;
1647 }
1648
1649 for (i = 0, slot = start_slot; i < nr; i++, slot++) {
1650 cond_resched();
1651 if (level == 0) {
1652 btrfs_item_key_to_cpu(buf, &key, slot);
1653 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
1654 continue;
1655 fi = btrfs_item_ptr(buf, slot,
1656 struct btrfs_file_extent_item);
1657 if (btrfs_file_extent_type(buf, fi) ==
1658 BTRFS_FILE_EXTENT_INLINE)
1659 continue;
1660 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
1661 if (bytenr == 0)
1662 continue;
31840ae1
ZY		 1663 ret = __btrfs_update_extent_ref(trans, root, bytenr,
1664 orig_buf->start, buf->start,
1665 orig_root, ref_root,
1666 orig_generation, ref_generation,
3bb1a1bc 1667 key.objectid);
31840ae1
ZY		 1668 if (ret)
1669 goto fail;
1670 } else {
1671 bytenr = btrfs_node_blockptr(buf, slot);
31840ae1
ZY		 1672 ret = __btrfs_update_extent_ref(trans, root, bytenr,
1673 orig_buf->start, buf->start,
1674 orig_root, ref_root,
1675 orig_generation, ref_generation,
3bb1a1bc 1676 level - 1);
31840ae1
ZY		 1677 if (ret)
1678 goto fail;
1679 }
1680 }
1681 return 0;
1682fail:
1683 WARN_ON(1);
1684 return -1;
1685}
1686
9078a3e1
CM		 1687static int write_one_cache_group(struct btrfs_trans_handle *trans,
1688 struct btrfs_root *root,
1689 struct btrfs_path *path,
1690 struct btrfs_block_group_cache *cache)
1691{
1692 int ret;
1693 int pending_ret;
1694 struct btrfs_root *extent_root = root->fs_info->extent_root;
5f39d397
CM		 1695 unsigned long bi;
1696 struct extent_buffer *leaf;
9078a3e1 1697
9078a3e1 1698 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
54aa1f4d
CM		 1699 if (ret < 0)
1700 goto fail;
9078a3e1 1701 BUG_ON(ret);
5f39d397
CM		 1702
1703 leaf = path->nodes[0];
1704 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
1705 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
1706 btrfs_mark_buffer_dirty(leaf);
9078a3e1 1707 btrfs_release_path(extent_root, path);
54aa1f4d 1708fail:
87ef2bb4
CM		 1709 finish_current_insert(trans, extent_root, 0);
1710 pending_ret = del_pending_extents(trans, extent_root, 0);
9078a3e1
CM		 1711 if (ret)
1712 return ret;
1713 if (pending_ret)
1714 return pending_ret;
1715 return 0;
1716
1717}
1718
96b5179d
CM		 1719int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
1720 struct btrfs_root *root)
9078a3e1 1721{
0f9dd46c
JB		 1722 struct btrfs_block_group_cache *cache, *entry;
1723 struct rb_node *n;
9078a3e1
CM		 1724 int err = 0;
1725 int werr = 0;
9078a3e1 1726 struct btrfs_path *path;
96b5179d 1727 u64 last = 0;
9078a3e1
CM		 1728
1729 path = btrfs_alloc_path();
1730 if (!path)
1731 return -ENOMEM;
1732
1733 while(1) {
0f9dd46c
JB		 1734 cache = NULL;
1735 spin_lock(&root->fs_info->block_group_cache_lock);
1736 for (n = rb_first(&root->fs_info->block_group_cache_tree);
1737 n; n = rb_next(n)) {
1738 entry = rb_entry(n, struct btrfs_block_group_cache,
1739 cache_node);
1740 if (entry->dirty) {
1741 cache = entry;
1742 break;
1743 }
1744 }
1745 spin_unlock(&root->fs_info->block_group_cache_lock);
54aa1f4d 1746
0f9dd46c 1747 if (!cache)
96b5179d 1748 break;
0f9dd46c 1749
e8569813 1750 cache->dirty = 0;
0f9dd46c
JB		 1751 last += cache->key.offset;
1752
96b5179d
CM		 1753 err = write_one_cache_group(trans, root,
1754 path, cache);
1755 /*
1756 * if we fail to write the cache group, we want
1757 * to keep it marked dirty in hopes that a later
1758 * write will work
1759 */
1760 if (err) {
1761 werr = err;
1762 continue;
9078a3e1
CM		 1763 }
1764 }
1765 btrfs_free_path(path);
1766 return werr;
1767}
1768
d2fb3437
YZ		 1769int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
1770{
1771 struct btrfs_block_group_cache *block_group;
1772 int readonly = 0;
1773
1774 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
1775 if (!block_group || block_group->ro)
1776 readonly = 1;
1777 if (block_group)
1778 put_block_group(block_group);
1779 return readonly;
1780}
1781
593060d7
CM		 1782static int update_space_info(struct btrfs_fs_info *info, u64 flags,
1783 u64 total_bytes, u64 bytes_used,
1784 struct btrfs_space_info **space_info)
1785{
1786 struct btrfs_space_info *found;
1787
1788 found = __find_space_info(info, flags);
1789 if (found) {
25179201 1790 spin_lock(&found->lock);
593060d7
CM		 1791 found->total_bytes += total_bytes;
1792 found->bytes_used += bytes_used;
8f18cf13 1793 found->full = 0;
25179201 1794 spin_unlock(&found->lock);
593060d7
CM		 1795 *space_info = found;
1796 return 0;
1797 }
c146afad 1798 found = kzalloc(sizeof(*found), GFP_NOFS);
593060d7
CM		 1799 if (!found)
1800 return -ENOMEM;
1801
1802 list_add(&found->list, &info->space_info);
0f9dd46c 1803 INIT_LIST_HEAD(&found->block_groups);
80eb234a 1804 init_rwsem(&found->groups_sem);
0f9dd46c 1805 spin_lock_init(&found->lock);
593060d7
CM		 1806 found->flags = flags;
1807 found->total_bytes = total_bytes;
1808 found->bytes_used = bytes_used;
1809 found->bytes_pinned = 0;
e8569813 1810 found->bytes_reserved = 0;
c146afad 1811 found->bytes_readonly = 0;
593060d7 1812 found->full = 0;
0ef3e66b 1813 found->force_alloc = 0;
593060d7
CM		 1814 *space_info = found;
1815 return 0;
1816}
1817
8790d502
CM		 1818static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
1819{
1820 u64 extra_flags = flags & (BTRFS_BLOCK_GROUP_RAID0 |
611f0e00 1821 BTRFS_BLOCK_GROUP_RAID1 |
321aecc6 1822 BTRFS_BLOCK_GROUP_RAID10 |
611f0e00 1823 BTRFS_BLOCK_GROUP_DUP);
8790d502
CM		 1824 if (extra_flags) {
1825 if (flags & BTRFS_BLOCK_GROUP_DATA)
1826 fs_info->avail_data_alloc_bits |= extra_flags;
1827 if (flags & BTRFS_BLOCK_GROUP_METADATA)
1828 fs_info->avail_metadata_alloc_bits |= extra_flags;
1829 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
1830 fs_info->avail_system_alloc_bits |= extra_flags;
1831 }
1832}
593060d7 1833
c146afad
YZ		 1834static void set_block_group_readonly(struct btrfs_block_group_cache *cache)
1835{
1836 spin_lock(&cache->space_info->lock);
1837 spin_lock(&cache->lock);
1838 if (!cache->ro) {
1839 cache->space_info->bytes_readonly += cache->key.offset -
1840 btrfs_block_group_used(&cache->item);
1841 cache->ro = 1;
1842 }
1843 spin_unlock(&cache->lock);
1844 spin_unlock(&cache->space_info->lock);
1845}
1846
2b82032c 1847u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
ec44a35c 1848{
2b82032c 1849 u64 num_devices = root->fs_info->fs_devices->rw_devices;
a061fc8d
CM		 1850
1851 if (num_devices == 1)
1852 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0);
1853 if (num_devices < 4)
1854 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
1855
ec44a35c
CM		 1856 if ((flags & BTRFS_BLOCK_GROUP_DUP) &&
1857 (flags & (BTRFS_BLOCK_GROUP_RAID1 |
a061fc8d 1858 BTRFS_BLOCK_GROUP_RAID10))) {
ec44a35c 1859 flags &= ~BTRFS_BLOCK_GROUP_DUP;
a061fc8d 1860 }
ec44a35c
CM		 1861
1862 if ((flags & BTRFS_BLOCK_GROUP_RAID1) &&
a061fc8d 1863 (flags & BTRFS_BLOCK_GROUP_RAID10)) {
ec44a35c 1864 flags &= ~BTRFS_BLOCK_GROUP_RAID1;
a061fc8d 1865 }
ec44a35c
CM		 1866
1867 if ((flags & BTRFS_BLOCK_GROUP_RAID0) &&
1868 ((flags & BTRFS_BLOCK_GROUP_RAID1) |
1869 (flags & BTRFS_BLOCK_GROUP_RAID10) |
1870 (flags & BTRFS_BLOCK_GROUP_DUP)))
1871 flags &= ~BTRFS_BLOCK_GROUP_RAID0;
1872 return flags;
1873}
1874
6324fbf3
CM		 1875static int do_chunk_alloc(struct btrfs_trans_handle *trans,
1876 struct btrfs_root *extent_root, u64 alloc_bytes,
0ef3e66b 1877 u64 flags, int force)
6324fbf3
CM		 1878{
1879 struct btrfs_space_info *space_info;
1880 u64 thresh;
c146afad
YZ		 1881 int ret = 0;
1882
1883 mutex_lock(&extent_root->fs_info->chunk_mutex);
6324fbf3 1884
2b82032c 1885 flags = btrfs_reduce_alloc_profile(extent_root, flags);
ec44a35c 1886
6324fbf3 1887 space_info = __find_space_info(extent_root->fs_info, flags);
593060d7
CM		 1888 if (!space_info) {
1889 ret = update_space_info(extent_root->fs_info, flags,
1890 0, 0, &space_info);
1891 BUG_ON(ret);
1892 }
6324fbf3
CM		 1893 BUG_ON(!space_info);
1894
25179201 1895 spin_lock(&space_info->lock);
0ef3e66b
CM		 1896 if (space_info->force_alloc) {
1897 force = 1;
1898 space_info->force_alloc = 0;
1899 }
25179201
JB		 1900 if (space_info->full) {
1901 spin_unlock(&space_info->lock);
925baedd 1902 goto out;
25179201 1903 }
6324fbf3 1904
c146afad
YZ		 1905 thresh = space_info->total_bytes - space_info->bytes_readonly;
1906 thresh = div_factor(thresh, 6);
0ef3e66b 1907 if (!force &&
e8569813 1908 (space_info->bytes_used + space_info->bytes_pinned +
25179201
JB		 1909 space_info->bytes_reserved + alloc_bytes) < thresh) {
1910 spin_unlock(&space_info->lock);
925baedd 1911 goto out;
25179201 1912 }
25179201
JB		 1913 spin_unlock(&space_info->lock);
1914
2b82032c 1915 ret = btrfs_alloc_chunk(trans, extent_root, flags);
25179201 1916 if (ret) {
6324fbf3
CM		 1917printk("space info full %Lu\n", flags);
1918 space_info->full = 1;
6324fbf3 1919 }
a74a4b97 1920out:
c146afad 1921 mutex_unlock(&extent_root->fs_info->chunk_mutex);
0f9dd46c 1922 return ret;
6324fbf3
CM		 1923}
1924
9078a3e1
CM		 1925static int update_block_group(struct btrfs_trans_handle *trans,
1926 struct btrfs_root *root,
db94535d 1927 u64 bytenr, u64 num_bytes, int alloc,
0b86a832 1928 int mark_free)
9078a3e1
CM		 1929{
1930 struct btrfs_block_group_cache *cache;
1931 struct btrfs_fs_info *info = root->fs_info;
db94535d 1932 u64 total = num_bytes;
9078a3e1 1933 u64 old_val;
db94535d 1934 u64 byte_in_group;
3e1ad54f 1935
9078a3e1 1936 while(total) {
db94535d 1937 cache = btrfs_lookup_block_group(info, bytenr);
f3465ca4 1938 if (!cache)
9078a3e1 1939 return -1;
db94535d
CM		 1940 byte_in_group = bytenr - cache->key.objectid;
1941 WARN_ON(byte_in_group > cache->key.offset);
9078a3e1 1942
25179201 1943 spin_lock(&cache->space_info->lock);
c286ac48 1944 spin_lock(&cache->lock);
0f9dd46c 1945 cache->dirty = 1;
9078a3e1 1946 old_val = btrfs_block_group_used(&cache->item);
db94535d 1947 num_bytes = min(total, cache->key.offset - byte_in_group);
cd1bc465 1948 if (alloc) {
db94535d 1949 old_val += num_bytes;
6324fbf3 1950 cache->space_info->bytes_used += num_bytes;
a512bbf8 1951 if (cache->ro)
c146afad 1952 cache->space_info->bytes_readonly -= num_bytes;
c286ac48
CM		 1953 btrfs_set_block_group_used(&cache->item, old_val);
1954 spin_unlock(&cache->lock);
25179201 1955 spin_unlock(&cache->space_info->lock);
cd1bc465 1956 } else {
db94535d 1957 old_val -= num_bytes;
6324fbf3 1958 cache->space_info->bytes_used -= num_bytes;
c146afad
YZ		 1959 if (cache->ro)
1960 cache->space_info->bytes_readonly += num_bytes;
c286ac48
CM		 1961 btrfs_set_block_group_used(&cache->item, old_val);
1962 spin_unlock(&cache->lock);
25179201 1963 spin_unlock(&cache->space_info->lock);
f510cfec 1964 if (mark_free) {
0f9dd46c
JB		 1965 int ret;
1966 ret = btrfs_add_free_space(cache, bytenr,
1967 num_bytes);
d2fb3437 1968 WARN_ON(ret);
e37c9e69 1969 }
cd1bc465 1970 }
d2fb3437 1971 put_block_group(cache);
db94535d
CM		 1972 total -= num_bytes;
1973 bytenr += num_bytes;
9078a3e1
CM		 1974 }
1975 return 0;
1976}
6324fbf3 1977
a061fc8d
CM		 1978static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
1979{
0f9dd46c 1980 struct btrfs_block_group_cache *cache;
d2fb3437 1981 u64 bytenr;
0f9dd46c
JB		 1982
1983 cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
1984 if (!cache)
a061fc8d 1985 return 0;
0f9dd46c 1986
d2fb3437
YZ		 1987 bytenr = cache->key.objectid;
1988 put_block_group(cache);
1989
1990 return bytenr;
a061fc8d
CM		 1991}
1992
e02119d5 1993int btrfs_update_pinned_extents(struct btrfs_root *root,
324ae4df
Y		 1994 u64 bytenr, u64 num, int pin)
1995{
1996 u64 len;
1997 struct btrfs_block_group_cache *cache;
1998 struct btrfs_fs_info *fs_info = root->fs_info;
1999
25179201 2000 WARN_ON(!mutex_is_locked(&root->fs_info->pinned_mutex));
324ae4df
Y		 2001 if (pin) {
2002 set_extent_dirty(&fs_info->pinned_extents,
2003 bytenr, bytenr + num - 1, GFP_NOFS);
2004 } else {
2005 clear_extent_dirty(&fs_info->pinned_extents,
2006 bytenr, bytenr + num - 1, GFP_NOFS);
2007 }
2008 while (num > 0) {
2009 cache = btrfs_lookup_block_group(fs_info, bytenr);
e8569813
ZY		 2010 BUG_ON(!cache);
2011 len = min(num, cache->key.offset -
2012 (bytenr - cache->key.objectid));
324ae4df 2013 if (pin) {
25179201 2014 spin_lock(&cache->space_info->lock);
e8569813
ZY		 2015 spin_lock(&cache->lock);
2016 cache->pinned += len;
2017 cache->space_info->bytes_pinned += len;
2018 spin_unlock(&cache->lock);
25179201 2019 spin_unlock(&cache->space_info->lock);
324ae4df
Y		 2020 fs_info->total_pinned += len;
2021 } else {
25179201 2022 spin_lock(&cache->space_info->lock);
e8569813
ZY		 2023 spin_lock(&cache->lock);
2024 cache->pinned -= len;
2025 cache->space_info->bytes_pinned -= len;
2026 spin_unlock(&cache->lock);
25179201 2027 spin_unlock(&cache->space_info->lock);
324ae4df 2028 fs_info->total_pinned -= len;
07103a3c
JB		 2029 if (cache->cached)
2030 btrfs_add_free_space(cache, bytenr, len);
324ae4df 2031 }
d2fb3437 2032 put_block_group(cache);
324ae4df
Y		 2033 bytenr += len;
2034 num -= len;
2035 }
2036 return 0;
2037}
9078a3e1 2038
e8569813
ZY		 2039static int update_reserved_extents(struct btrfs_root *root,
2040 u64 bytenr, u64 num, int reserve)
2041{
2042 u64 len;
2043 struct btrfs_block_group_cache *cache;
2044 struct btrfs_fs_info *fs_info = root->fs_info;
2045
e8569813
ZY		 2046 while (num > 0) {
2047 cache = btrfs_lookup_block_group(fs_info, bytenr);
2048 BUG_ON(!cache);
2049 len = min(num, cache->key.offset -
2050 (bytenr - cache->key.objectid));
25179201
JB		 2051
2052 spin_lock(&cache->space_info->lock);
2053 spin_lock(&cache->lock);
e8569813 2054 if (reserve) {
e8569813
ZY		 2055 cache->reserved += len;
2056 cache->space_info->bytes_reserved += len;
e8569813 2057 } else {
e8569813
ZY		 2058 cache->reserved -= len;
2059 cache->space_info->bytes_reserved -= len;
e8569813 2060 }
25179201
JB		 2061 spin_unlock(&cache->lock);
2062 spin_unlock(&cache->space_info->lock);
d2fb3437 2063 put_block_group(cache);
e8569813
ZY		 2064 bytenr += len;
2065 num -= len;
2066 }
2067 return 0;
2068}
2069
d1310b2e 2070int btrfs_copy_pinned(struct btrfs_root *root, struct extent_io_tree *copy)
ccd467d6 2071{
ccd467d6 2072 u64 last = 0;
1a5bc167
CM		 2073 u64 start;
2074 u64 end;
d1310b2e 2075 struct extent_io_tree *pinned_extents = &root->fs_info->pinned_extents;
ccd467d6 2076 int ret;
ccd467d6 2077
25179201 2078 mutex_lock(&root->fs_info->pinned_mutex);
ccd467d6 2079 while(1) {
1a5bc167
CM		 2080 ret = find_first_extent_bit(pinned_extents, last,
2081 &start, &end, EXTENT_DIRTY);
2082 if (ret)
ccd467d6 2083 break;
1a5bc167
CM		 2084 set_extent_dirty(copy, start, end, GFP_NOFS);
2085 last = end + 1;
ccd467d6 2086 }
25179201 2087 mutex_unlock(&root->fs_info->pinned_mutex);
ccd467d6
CM		 2088 return 0;
2089}
2090
2091int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
2092 struct btrfs_root *root,
d1310b2e 2093 struct extent_io_tree *unpin)
a28ec197 2094{
1a5bc167
CM		 2095 u64 start;
2096 u64 end;
a28ec197 2097 int ret;
a28ec197 2098
25179201 2099 mutex_lock(&root->fs_info->pinned_mutex);
a28ec197 2100 while(1) {
1a5bc167
CM		 2101 ret = find_first_extent_bit(unpin, 0, &start, &end,
2102 EXTENT_DIRTY);
2103 if (ret)
a28ec197 2104 break;
e02119d5 2105 btrfs_update_pinned_extents(root, start, end + 1 - start, 0);
1a5bc167 2106 clear_extent_dirty(unpin, start, end, GFP_NOFS);
c286ac48 2107 if (need_resched()) {
25179201 2108 mutex_unlock(&root->fs_info->pinned_mutex);
c286ac48 2109 cond_resched();
25179201 2110 mutex_lock(&root->fs_info->pinned_mutex);
c286ac48 2111 }
a28ec197 2112 }
25179201 2113 mutex_unlock(&root->fs_info->pinned_mutex);
a28ec197
CM		 2114 return 0;
2115}
2116
98ed5174 2117static int finish_current_insert(struct btrfs_trans_handle *trans,
87ef2bb4 2118 struct btrfs_root *extent_root, int all)
037e6390 2119{
7bb86316
CM		 2120 u64 start;
2121 u64 end;
31840ae1 2122 u64 priv;
25179201 2123 u64 search = 0;
f3465ca4 2124 u64 skipped = 0;
7bb86316
CM		 2125 struct btrfs_fs_info *info = extent_root->fs_info;
2126 struct btrfs_path *path;
f3465ca4
JB		 2127 struct pending_extent_op *extent_op, *tmp;
2128 struct list_head insert_list, update_list;
037e6390 2129 int ret;
f3465ca4 2130 int num_inserts = 0, max_inserts;
037e6390 2131
7bb86316 2132 path = btrfs_alloc_path();
f3465ca4
JB		 2133 INIT_LIST_HEAD(&insert_list);
2134 INIT_LIST_HEAD(&update_list);
037e6390 2135
f3465ca4
JB		 2136 max_inserts = extent_root->leafsize /
2137 (2 * sizeof(struct btrfs_key) + 2 * sizeof(struct btrfs_item) +
2138 sizeof(struct btrfs_extent_ref) +
2139 sizeof(struct btrfs_extent_item));
2140again:
2141 mutex_lock(&info->extent_ins_mutex);
2142 while (1) {
25179201
JB		 2143 ret = find_first_extent_bit(&info->extent_ins, search, &start,
2144 &end, EXTENT_WRITEBACK);
2145 if (ret) {
f3465ca4
JB		 2146 if (skipped && all && !num_inserts) {
2147 skipped = 0;
b4eec2ca 2148 search = 0;
25179201
JB		 2149 continue;
2150 }
f3465ca4 2151 mutex_unlock(&info->extent_ins_mutex);
26b8003f 2152 break;
25179201
JB		 2153 }
2154
2155 ret = try_lock_extent(&info->extent_ins, start, end, GFP_NOFS);
2156 if (!ret) {
f3465ca4 2157 skipped = 1;
87ef2bb4 2158 search = end + 1;
f3465ca4
JB		 2159 if (need_resched()) {
2160 mutex_unlock(&info->extent_ins_mutex);
2161 cond_resched();
2162 mutex_lock(&info->extent_ins_mutex);
2163 }
25179201
JB		 2164 continue;
2165 }
26b8003f 2166
31840ae1
ZY		 2167 ret = get_state_private(&info->extent_ins, start, &priv);
2168 BUG_ON(ret);
f3465ca4 2169 extent_op = (struct pending_extent_op *)(unsigned long) priv;
25179201 2170
31840ae1 2171 if (extent_op->type == PENDING_EXTENT_INSERT) {
f3465ca4
JB		 2172 num_inserts++;
2173 list_add_tail(&extent_op->list, &insert_list);
2174 search = end + 1;
2175 if (num_inserts == max_inserts) {
2176 mutex_unlock(&info->extent_ins_mutex);
2177 break;
2178 }
2179 } else if (extent_op->type == PENDING_BACKREF_UPDATE) {
2180 list_add_tail(&extent_op->list, &update_list);
2181 search = end + 1;
2182 } else {
2183 BUG();
2184 }
2185 }
c286ac48 2186
f3465ca4 2187 /*
b4eec2ca 2188 * process the update list, clear the writeback bit for it, and if
f3465ca4
JB		 2189 * somebody marked this thing for deletion then just unlock it and be
2190 * done, the free_extents will handle it
2191 */
2192 mutex_lock(&info->extent_ins_mutex);
2193 list_for_each_entry_safe(extent_op, tmp, &update_list, list) {
2194 clear_extent_bits(&info->extent_ins, extent_op->bytenr,
2195 extent_op->bytenr + extent_op->num_bytes - 1,
2196 EXTENT_WRITEBACK, GFP_NOFS);
2197 if (extent_op->del) {
2198 list_del_init(&extent_op->list);
2199 unlock_extent(&info->extent_ins, extent_op->bytenr,
2200 extent_op->bytenr + extent_op->num_bytes
2201 - 1, GFP_NOFS);
2202 kfree(extent_op);
2203 }
2204 }
2205 mutex_unlock(&info->extent_ins_mutex);
c286ac48 2206
f3465ca4
JB		 2207 /*
2208 * still have things left on the update list, go ahead an update
2209 * everything
2210 */
2211 if (!list_empty(&update_list)) {
2212 ret = update_backrefs(trans, extent_root, path, &update_list);
2213 BUG_ON(ret);
2214 }
c286ac48 2215
f3465ca4
JB		 2216 /*
2217 * if no inserts need to be done, but we skipped some extents and we
2218 * need to make sure everything is cleaned then reset everything and
2219 * go back to the beginning
2220 */
2221 if (!num_inserts && all && skipped) {
2222 search = 0;
2223 skipped = 0;
2224 INIT_LIST_HEAD(&update_list);
2225 INIT_LIST_HEAD(&insert_list);
2226 goto again;
2227 } else if (!num_inserts) {
2228 goto out;
2229 }
31840ae1 2230
f3465ca4
JB		 2231 /*
2232 * process the insert extents list. Again if we are deleting this
2233 * extent, then just unlock it, pin down the bytes if need be, and be
2234 * done with it. Saves us from having to actually insert the extent
2235 * into the tree and then subsequently come along and delete it
2236 */
2237 mutex_lock(&info->extent_ins_mutex);
2238 list_for_each_entry_safe(extent_op, tmp, &insert_list, list) {
2239 clear_extent_bits(&info->extent_ins, extent_op->bytenr,
2240 extent_op->bytenr + extent_op->num_bytes - 1,
2241 EXTENT_WRITEBACK, GFP_NOFS);
2242 if (extent_op->del) {
2243 list_del_init(&extent_op->list);
2244 unlock_extent(&info->extent_ins, extent_op->bytenr,
2245 extent_op->bytenr + extent_op->num_bytes
2246 - 1, GFP_NOFS);
2247
2248 mutex_lock(&extent_root->fs_info->pinned_mutex);
2249 ret = pin_down_bytes(trans, extent_root,
2250 extent_op->bytenr,
2251 extent_op->num_bytes, 0);
2252 mutex_unlock(&extent_root->fs_info->pinned_mutex);
2253
2254 ret = update_block_group(trans, extent_root,
2255 extent_op->bytenr,
2256 extent_op->num_bytes,
2257 0, ret > 0);
2258 BUG_ON(ret);
2259 kfree(extent_op);
2260 num_inserts--;
d8d5f3e1 2261 }
f3465ca4
JB		 2262 }
2263 mutex_unlock(&info->extent_ins_mutex);
31840ae1 2264
f3465ca4
JB		 2265 ret = insert_extents(trans, extent_root, path, &insert_list,
2266 num_inserts);
2267 BUG_ON(ret);
2268
2269 /*
2270 * if we broke out of the loop in order to insert stuff because we hit
2271 * the maximum number of inserts at a time we can handle, then loop
2272 * back and pick up where we left off
2273 */
2274 if (num_inserts == max_inserts) {
2275 INIT_LIST_HEAD(&insert_list);
2276 INIT_LIST_HEAD(&update_list);
2277 num_inserts = 0;
2278 goto again;
2279 }
2280
2281 /*
2282 * again, if we need to make absolutely sure there are no more pending
2283 * extent operations left and we know that we skipped some, go back to
2284 * the beginning and do it all again
2285 */
2286 if (all && skipped) {
2287 INIT_LIST_HEAD(&insert_list);
2288 INIT_LIST_HEAD(&update_list);
2289 search = 0;
2290 skipped = 0;
2291 num_inserts = 0;
2292 goto again;
037e6390 2293 }
f3465ca4 2294out:
7bb86316 2295 btrfs_free_path(path);
037e6390
CM		 2296 return 0;
2297}
2298
31840ae1
ZY		 2299static int pin_down_bytes(struct btrfs_trans_handle *trans,
2300 struct btrfs_root *root,
2301 u64 bytenr, u64 num_bytes, int is_data)
e20d96d6 2302{
1a5bc167 2303 int err = 0;
31840ae1 2304 struct extent_buffer *buf;
8ef97622 2305
31840ae1
ZY		 2306 if (is_data)
2307 goto pinit;
2308
2309 buf = btrfs_find_tree_block(root, bytenr, num_bytes);
2310 if (!buf)
2311 goto pinit;
2312
2313 /* we can reuse a block if it hasn't been written
2314 * and it is from this transaction. We can't
2315 * reuse anything from the tree log root because
2316 * it has tiny sub-transactions.
2317 */
2318 if (btrfs_buffer_uptodate(buf, 0) &&
2319 btrfs_try_tree_lock(buf)) {
2320 u64 header_owner = btrfs_header_owner(buf);
2321 u64 header_transid = btrfs_header_generation(buf);
2322 if (header_owner != BTRFS_TREE_LOG_OBJECTID &&
1a40e23b 2323 header_owner != BTRFS_TREE_RELOC_OBJECTID &&
31840ae1
ZY		 2324 header_transid == trans->transid &&
2325 !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
2326 clean_tree_block(NULL, root, buf);
2327 btrfs_tree_unlock(buf);
5f39d397 2328 free_extent_buffer(buf);
31840ae1 2329 return 1;
8ef97622 2330 }
31840ae1 2331 btrfs_tree_unlock(buf);
f4b9aa8d 2332 }
31840ae1
ZY		 2333 free_extent_buffer(buf);
2334pinit:
2335 btrfs_update_pinned_extents(root, bytenr, num_bytes, 1);
2336
be744175 2337 BUG_ON(err < 0);
e20d96d6
CM		 2338 return 0;
2339}
2340
fec577fb 2341/*
a28ec197 2342 * remove an extent from the root, returns 0 on success
fec577fb 2343 */
31840ae1
ZY		 2344static int __free_extent(struct btrfs_trans_handle *trans,
2345 struct btrfs_root *root,
2346 u64 bytenr, u64 num_bytes, u64 parent,
7bb86316 2347 u64 root_objectid, u64 ref_generation,
3bb1a1bc 2348 u64 owner_objectid, int pin, int mark_free)
a28ec197 2349{
5caf2a00 2350 struct btrfs_path *path;
e2fa7227 2351 struct btrfs_key key;
1261ec42
CM		 2352 struct btrfs_fs_info *info = root->fs_info;
2353 struct btrfs_root *extent_root = info->extent_root;
5f39d397 2354 struct extent_buffer *leaf;
a28ec197 2355 int ret;
952fccac
CM		 2356 int extent_slot = 0;
2357 int found_extent = 0;
2358 int num_to_del = 1;
234b63a0 2359 struct btrfs_extent_item *ei;
cf27e1ee 2360 u32 refs;
037e6390 2361
db94535d 2362 key.objectid = bytenr;
62e2749e 2363 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
db94535d 2364 key.offset = num_bytes;
5caf2a00 2365 path = btrfs_alloc_path();
54aa1f4d
CM		 2366 if (!path)
2367 return -ENOMEM;
5f26f772 2368
3c12ac72 2369 path->reada = 1;
3bb1a1bc
YZ		 2370 ret = lookup_extent_backref(trans, extent_root, path,
2371 bytenr, parent, root_objectid,
2372 ref_generation, owner_objectid, 1);
7bb86316 2373 if (ret == 0) {
952fccac
CM		 2374 struct btrfs_key found_key;
2375 extent_slot = path->slots[0];
2376 while(extent_slot > 0) {
2377 extent_slot--;
2378 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
2379 extent_slot);
2380 if (found_key.objectid != bytenr)
2381 break;
2382 if (found_key.type == BTRFS_EXTENT_ITEM_KEY &&
2383 found_key.offset == num_bytes) {
2384 found_extent = 1;
2385 break;
2386 }
2387 if (path->slots[0] - extent_slot > 5)
2388 break;
2389 }
31840ae1
ZY		 2390 if (!found_extent) {
2391 ret = remove_extent_backref(trans, extent_root, path);
2392 BUG_ON(ret);
2393 btrfs_release_path(extent_root, path);
2394 ret = btrfs_search_slot(trans, extent_root,
2395 &key, path, -1, 1);
f3465ca4
JB		 2396 if (ret) {
2397 printk(KERN_ERR "umm, got %d back from search"
2398 ", was looking for %Lu\n", ret,
2399 bytenr);
2400 btrfs_print_leaf(extent_root, path->nodes[0]);
2401 }
31840ae1
ZY		 2402 BUG_ON(ret);
2403 extent_slot = path->slots[0];
2404 }
7bb86316
CM		 2405 } else {
2406 btrfs_print_leaf(extent_root, path->nodes[0]);
2407 WARN_ON(1);
2408 printk("Unable to find ref byte nr %Lu root %Lu "
3bb1a1bc
YZ		 2409 "gen %Lu owner %Lu\n", bytenr,
2410 root_objectid, ref_generation, owner_objectid);
7bb86316 2411 }
5f39d397
CM		 2412
2413 leaf = path->nodes[0];
952fccac 2414 ei = btrfs_item_ptr(leaf, extent_slot,
123abc88 2415 struct btrfs_extent_item);
5f39d397
CM		 2416 refs = btrfs_extent_refs(leaf, ei);
2417 BUG_ON(refs == 0);
2418 refs -= 1;
2419 btrfs_set_extent_refs(leaf, ei, refs);
952fccac 2420
5f39d397
CM		 2421 btrfs_mark_buffer_dirty(leaf);
2422
952fccac 2423 if (refs == 0 && found_extent && path->slots[0] == extent_slot + 1) {
31840ae1
ZY		 2424 struct btrfs_extent_ref *ref;
2425 ref = btrfs_item_ptr(leaf, path->slots[0],
2426 struct btrfs_extent_ref);
2427 BUG_ON(btrfs_ref_num_refs(leaf, ref) != 1);
952fccac
CM		 2428 /* if the back ref and the extent are next to each other
2429 * they get deleted below in one shot
2430 */
2431 path->slots[0] = extent_slot;
2432 num_to_del = 2;
2433 } else if (found_extent) {
2434 /* otherwise delete the extent back ref */
31840ae1 2435 ret = remove_extent_backref(trans, extent_root, path);
952fccac
CM		 2436 BUG_ON(ret);
2437 /* if refs are 0, we need to setup the path for deletion */
2438 if (refs == 0) {
2439 btrfs_release_path(extent_root, path);
2440 ret = btrfs_search_slot(trans, extent_root, &key, path,
2441 -1, 1);
952fccac
CM		 2442 BUG_ON(ret);
2443 }
2444 }
2445
cf27e1ee 2446 if (refs == 0) {
db94535d
CM		 2447 u64 super_used;
2448 u64 root_used;
21af804c
DW		 2449#ifdef BIO_RW_DISCARD
2450 u64 map_length = num_bytes;
2451 struct btrfs_multi_bio *multi = NULL;
2452#endif
78fae27e
CM		 2453
2454 if (pin) {
25179201 2455 mutex_lock(&root->fs_info->pinned_mutex);
31840ae1
ZY		 2456 ret = pin_down_bytes(trans, root, bytenr, num_bytes,
2457 owner_objectid >= BTRFS_FIRST_FREE_OBJECTID);
25179201 2458 mutex_unlock(&root->fs_info->pinned_mutex);
c549228f
Y		 2459 if (ret > 0)
2460 mark_free = 1;
2461 BUG_ON(ret < 0);
78fae27e 2462 }
58176a96 2463 /* block accounting for super block */
a2135011 2464 spin_lock_irq(&info->delalloc_lock);
db94535d
CM		 2465 super_used = btrfs_super_bytes_used(&info->super_copy);
2466 btrfs_set_super_bytes_used(&info->super_copy,
2467 super_used - num_bytes);
a2135011 2468 spin_unlock_irq(&info->delalloc_lock);
58176a96
JB		 2469
2470 /* block accounting for root item */
db94535d 2471 root_used = btrfs_root_used(&root->root_item);
5f39d397 2472 btrfs_set_root_used(&root->root_item,
db94535d 2473 root_used - num_bytes);
952fccac
CM		 2474 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
2475 num_to_del);
31840ae1 2476 BUG_ON(ret);
25179201 2477 btrfs_release_path(extent_root, path);
db94535d 2478 ret = update_block_group(trans, root, bytenr, num_bytes, 0,
0b86a832 2479 mark_free);
9078a3e1 2480 BUG_ON(ret);
21af804c 2481
459931ec
CM		 2482 if (owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
2483 ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
2484 BUG_ON(ret);
2485 }
2486
21af804c
DW		 2487#ifdef BIO_RW_DISCARD
2488 /* Tell the block device(s) that the sectors can be discarded */
2489 ret = btrfs_map_block(&root->fs_info->mapping_tree, READ,
2490 bytenr, &map_length, &multi, 0);
2491 if (!ret) {
2492 struct btrfs_bio_stripe *stripe = multi->stripes;
2493 int i;
2494
2495 if (map_length > num_bytes)
2496 map_length = num_bytes;
2497
2498 for (i = 0; i < multi->num_stripes; i++, stripe++) {
15916de8
CM		 2499 btrfs_issue_discard(stripe->dev->bdev,
2500 stripe->physical,
2501 map_length);
21af804c
DW		 2502 }
2503 kfree(multi);
2504 }
2505#endif
a28ec197 2506 }
5caf2a00 2507 btrfs_free_path(path);
87ef2bb4 2508 finish_current_insert(trans, extent_root, 0);
a28ec197
CM		 2509 return ret;
2510}
2511
a28ec197
CM		 2512/*
2513 * find all the blocks marked as pending in the radix tree and remove
2514 * them from the extent map
2515 */
e089f05c 2516static int del_pending_extents(struct btrfs_trans_handle *trans, struct
87ef2bb4 2517 btrfs_root *extent_root, int all)
a28ec197
CM		 2518{
2519 int ret;
e20d96d6 2520 int err = 0;
1a5bc167
CM		 2521 u64 start;
2522 u64 end;
31840ae1 2523 u64 priv;
25179201 2524 u64 search = 0;
f3465ca4 2525 int nr = 0, skipped = 0;
d1310b2e 2526 struct extent_io_tree *pending_del;
31840ae1
ZY		 2527 struct extent_io_tree *extent_ins;
2528 struct pending_extent_op *extent_op;
25179201 2529 struct btrfs_fs_info *info = extent_root->fs_info;
f3465ca4 2530 struct list_head delete_list;
8ef97622 2531
f3465ca4 2532 INIT_LIST_HEAD(&delete_list);
31840ae1 2533 extent_ins = &extent_root->fs_info->extent_ins;
1a5bc167 2534 pending_del = &extent_root->fs_info->pending_del;
a28ec197 2535
f3465ca4
JB		 2536again:
2537 mutex_lock(&info->extent_ins_mutex);
a28ec197 2538 while(1) {
25179201
JB		 2539 ret = find_first_extent_bit(pending_del, search, &start, &end,
2540 EXTENT_WRITEBACK);
2541 if (ret) {
f3465ca4 2542 if (all && skipped && !nr) {
25179201
JB		 2543 search = 0;
2544 continue;
2545 }
f3465ca4 2546 mutex_unlock(&info->extent_ins_mutex);
a28ec197 2547 break;
25179201
JB		 2548 }
2549
2550 ret = try_lock_extent(extent_ins, start, end, GFP_NOFS);
2551 if (!ret) {
2552 search = end+1;
f3465ca4
JB		 2553 skipped = 1;
2554
2555 if (need_resched()) {
2556 mutex_unlock(&info->extent_ins_mutex);
2557 cond_resched();
2558 mutex_lock(&info->extent_ins_mutex);
2559 }
2560
25179201
JB		 2561 continue;
2562 }
2563 BUG_ON(ret < 0);
31840ae1
ZY		 2564
2565 ret = get_state_private(pending_del, start, &priv);
2566 BUG_ON(ret);
2567 extent_op = (struct pending_extent_op *)(unsigned long)priv;
2568
25179201 2569 clear_extent_bits(pending_del, start, end, EXTENT_WRITEBACK,
1a5bc167 2570 GFP_NOFS);
31840ae1 2571 if (!test_range_bit(extent_ins, start, end,
25179201 2572 EXTENT_WRITEBACK, 0)) {
f3465ca4
JB		 2573 list_add_tail(&extent_op->list, &delete_list);
2574 nr++;
c286ac48 2575 } else {
31840ae1 2576 kfree(extent_op);
25179201
JB		 2577
2578 ret = get_state_private(&info->extent_ins, start,
2579 &priv);
31840ae1
ZY		 2580 BUG_ON(ret);
2581 extent_op = (struct pending_extent_op *)
25179201
JB		 2582 (unsigned long)priv;
2583
2584 clear_extent_bits(&info->extent_ins, start, end,
2585 EXTENT_WRITEBACK, GFP_NOFS);
31840ae1 2586
f3465ca4
JB		 2587 if (extent_op->type == PENDING_BACKREF_UPDATE) {
2588 list_add_tail(&extent_op->list, &delete_list);
2589 search = end + 1;
2590 nr++;
2591 continue;
2592 }
31840ae1 2593
25179201
JB		 2594 mutex_lock(&extent_root->fs_info->pinned_mutex);
2595 ret = pin_down_bytes(trans, extent_root, start,
2596 end + 1 - start, 0);
2597 mutex_unlock(&extent_root->fs_info->pinned_mutex);
2598
31840ae1 2599 ret = update_block_group(trans, extent_root, start,
25179201
JB		 2600 end + 1 - start, 0, ret > 0);
2601
f3465ca4 2602 unlock_extent(extent_ins, start, end, GFP_NOFS);
31840ae1
ZY		 2603 BUG_ON(ret);
2604 kfree(extent_op);
c286ac48 2605 }
1a5bc167
CM		 2606 if (ret)
2607 err = ret;
c286ac48 2608
f3465ca4
JB		 2609 search = end + 1;
2610
2611 if (need_resched()) {
2612 mutex_unlock(&info->extent_ins_mutex);
2613 cond_resched();
2614 mutex_lock(&info->extent_ins_mutex);
2615 }
2616 }
2617
2618 if (nr) {
2619 ret = free_extents(trans, extent_root, &delete_list);
2620 BUG_ON(ret);
2621 }
2622
2623 if (all && skipped) {
2624 INIT_LIST_HEAD(&delete_list);
2625 search = 0;
2626 nr = 0;
2627 goto again;
fec577fb 2628 }
f3465ca4 2629
e20d96d6 2630 return err;
fec577fb
CM		 2631}
2632
2633/*
2634 * remove an extent from the root, returns 0 on success
2635 */
925baedd 2636static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
31840ae1
ZY		 2637 struct btrfs_root *root,
2638 u64 bytenr, u64 num_bytes, u64 parent,
2639 u64 root_objectid, u64 ref_generation,
3bb1a1bc 2640 u64 owner_objectid, int pin)
fec577fb 2641{
9f5fae2f 2642 struct btrfs_root *extent_root = root->fs_info->extent_root;
fec577fb
CM		 2643 int pending_ret;
2644 int ret;
a28ec197 2645
db94535d 2646 WARN_ON(num_bytes < root->sectorsize);
fec577fb 2647 if (root == extent_root) {
f3465ca4
JB		 2648 struct pending_extent_op *extent_op = NULL;
2649
2650 mutex_lock(&root->fs_info->extent_ins_mutex);
2651 if (test_range_bit(&root->fs_info->extent_ins, bytenr,
2652 bytenr + num_bytes - 1, EXTENT_WRITEBACK, 0)) {
2653 u64 priv;
2654 ret = get_state_private(&root->fs_info->extent_ins,
2655 bytenr, &priv);
2656 BUG_ON(ret);
2657 extent_op = (struct pending_extent_op *)
2658 (unsigned long)priv;
2659
2660 extent_op->del = 1;
2661 if (extent_op->type == PENDING_EXTENT_INSERT) {
2662 mutex_unlock(&root->fs_info->extent_ins_mutex);
2663 return 0;
2664 }
2665 }
2666
2667 if (extent_op) {
2668 ref_generation = extent_op->orig_generation;
2669 parent = extent_op->orig_parent;
2670 }
31840ae1
ZY		 2671
2672 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
2673 BUG_ON(!extent_op);
2674
2675 extent_op->type = PENDING_EXTENT_DELETE;
2676 extent_op->bytenr = bytenr;
2677 extent_op->num_bytes = num_bytes;
2678 extent_op->parent = parent;
2679 extent_op->orig_parent = parent;
2680 extent_op->generation = ref_generation;
2681 extent_op->orig_generation = ref_generation;
2682 extent_op->level = (int)owner_objectid;
f3465ca4
JB		 2683 INIT_LIST_HEAD(&extent_op->list);
2684 extent_op->del = 0;
31840ae1
ZY		 2685
2686 set_extent_bits(&root->fs_info->pending_del,
2687 bytenr, bytenr + num_bytes - 1,
25179201 2688 EXTENT_WRITEBACK, GFP_NOFS);
31840ae1
ZY		 2689 set_state_private(&root->fs_info->pending_del,
2690 bytenr, (unsigned long)extent_op);
25179201 2691 mutex_unlock(&root->fs_info->extent_ins_mutex);
fec577fb
CM		 2692 return 0;
2693 }
4bef0848 2694 /* if metadata always pin */
d00aff00
CM		 2695 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) {
2696 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
0f9dd46c
JB		 2697 struct btrfs_block_group_cache *cache;
2698
d00aff00 2699 /* btrfs_free_reserved_extent */
0f9dd46c
JB		 2700 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
2701 BUG_ON(!cache);
2702 btrfs_add_free_space(cache, bytenr, num_bytes);
d2fb3437 2703 put_block_group(cache);
e8569813 2704 update_reserved_extents(root, bytenr, num_bytes, 0);
d00aff00
CM		 2705 return 0;
2706 }
4bef0848 2707 pin = 1;
d00aff00 2708 }
4bef0848
CM		 2709
2710 /* if data pin when any transaction has committed this */
2711 if (ref_generation != trans->transid)
2712 pin = 1;
2713
31840ae1 2714 ret = __free_extent(trans, root, bytenr, num_bytes, parent,
3bb1a1bc
YZ		 2715 root_objectid, ref_generation,
2716 owner_objectid, pin, pin == 0);
ee6e6504 2717
87ef2bb4
CM		 2718 finish_current_insert(trans, root->fs_info->extent_root, 0);
2719 pending_ret = del_pending_extents(trans, root->fs_info->extent_root, 0);
fec577fb
CM		 2720 return ret ? ret : pending_ret;
2721}
2722
925baedd 2723int btrfs_free_extent(struct btrfs_trans_handle *trans,
31840ae1
ZY		 2724 struct btrfs_root *root,
2725 u64 bytenr, u64 num_bytes, u64 parent,
2726 u64 root_objectid, u64 ref_generation,
3bb1a1bc 2727 u64 owner_objectid, int pin)
925baedd
CM		 2728{
2729 int ret;
2730
31840ae1 2731 ret = __btrfs_free_extent(trans, root, bytenr, num_bytes, parent,
925baedd 2732 root_objectid, ref_generation,
3bb1a1bc 2733 owner_objectid, pin);
925baedd
CM		 2734 return ret;
2735}
2736
87ee04eb
CM		 2737static u64 stripe_align(struct btrfs_root *root, u64 val)
2738{
2739 u64 mask = ((u64)root->stripesize - 1);
2740 u64 ret = (val + mask) & ~mask;
2741 return ret;
2742}
2743
fec577fb
CM		 2744/*
2745 * walks the btree of allocated extents and find a hole of a given size.
2746 * The key ins is changed to record the hole:
2747 * ins->objectid == block start
62e2749e 2748 * ins->flags = BTRFS_EXTENT_ITEM_KEY
fec577fb
CM		 2749 * ins->offset == number of blocks
2750 * Any available blocks before search_start are skipped.
2751 */
98ed5174
CM		 2752static int noinline find_free_extent(struct btrfs_trans_handle *trans,
2753 struct btrfs_root *orig_root,
2754 u64 num_bytes, u64 empty_size,
2755 u64 search_start, u64 search_end,
2756 u64 hint_byte, struct btrfs_key *ins,
2757 u64 exclude_start, u64 exclude_nr,
2758 int data)
fec577fb 2759{
80eb234a 2760 int ret = 0;
9f5fae2f 2761 struct btrfs_root * root = orig_root->fs_info->extent_root;
db94535d 2762 u64 total_needed = num_bytes;
239b14b3 2763 u64 *last_ptr = NULL;
4366211c 2764 u64 last_wanted = 0;
80eb234a 2765 struct btrfs_block_group_cache *block_group = NULL;
0ef3e66b 2766 int chunk_alloc_done = 0;
239b14b3 2767 int empty_cluster = 2 * 1024 * 1024;
0ef3e66b 2768 int allowed_chunk_alloc = 0;
80eb234a
JB		 2769 struct list_head *head = NULL, *cur = NULL;
2770 int loop = 0;
f5a31e16 2771 int extra_loop = 0;
80eb234a 2772 struct btrfs_space_info *space_info;
fec577fb 2773
db94535d 2774 WARN_ON(num_bytes < root->sectorsize);
b1a4d965 2775 btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
80eb234a
JB		 2776 ins->objectid = 0;
2777 ins->offset = 0;
b1a4d965 2778
0ef3e66b
CM		 2779 if (orig_root->ref_cows || empty_size)
2780 allowed_chunk_alloc = 1;
2781
239b14b3
CM		 2782 if (data & BTRFS_BLOCK_GROUP_METADATA) {
2783 last_ptr = &root->fs_info->last_alloc;
4366211c 2784 empty_cluster = 64 * 1024;
239b14b3
CM		 2785 }
2786
0f9dd46c 2787 if ((data & BTRFS_BLOCK_GROUP_DATA) && btrfs_test_opt(root, SSD))
239b14b3 2788 last_ptr = &root->fs_info->last_data_alloc;
0f9dd46c 2789
239b14b3 2790 if (last_ptr) {
4366211c 2791 if (*last_ptr) {
239b14b3 2792 hint_byte = *last_ptr;
4366211c
CM		 2793 last_wanted = *last_ptr;
2794 } else
239b14b3 2795 empty_size += empty_cluster;
4366211c
CM		 2796 } else {
2797 empty_cluster = 0;
239b14b3 2798 }
a061fc8d 2799 search_start = max(search_start, first_logical_byte(root, 0));
239b14b3 2800 search_start = max(search_start, hint_byte);
0b86a832 2801
42e70e7a 2802 if (last_wanted && search_start != last_wanted) {
4366211c 2803 last_wanted = 0;
42e70e7a
CM		 2804 empty_size += empty_cluster;
2805 }
4366211c 2806
42e70e7a 2807 total_needed += empty_size;
80eb234a 2808 block_group = btrfs_lookup_block_group(root->fs_info, search_start);
d899e052
YZ		 2809 if (!block_group)
2810 block_group = btrfs_lookup_first_block_group(root->fs_info,
2811 search_start);
80eb234a 2812 space_info = __find_space_info(root->fs_info, data);
0f9dd46c 2813
80eb234a
JB		 2814 down_read(&space_info->groups_sem);
2815 while (1) {
2816 struct btrfs_free_space *free_space;
0f9dd46c 2817 /*
80eb234a
JB		 2818 * the only way this happens if our hint points to a block
2819 * group thats not of the proper type, while looping this
2820 * should never happen
0f9dd46c 2821 */
8a1413a2
CM		 2822 if (empty_size)
2823 extra_loop = 1;
2824
42e70e7a
CM		 2825 if (!block_group)
2826 goto new_group_no_lock;
2827
ea6a478e
JB		 2828 if (unlikely(!block_group->cached)) {
2829 mutex_lock(&block_group->cache_mutex);
2830 ret = cache_block_group(root, block_group);
2831 mutex_unlock(&block_group->cache_mutex);
2832 if (ret)
2833 break;
2834 }
2835
25179201 2836 mutex_lock(&block_group->alloc_mutex);
80eb234a 2837 if (unlikely(!block_group_bits(block_group, data)))
0f9dd46c 2838 goto new_group;
0f9dd46c 2839
ea6a478e 2840 if (unlikely(block_group->ro))
80eb234a 2841 goto new_group;
0f9dd46c 2842
80eb234a
JB		 2843 free_space = btrfs_find_free_space(block_group, search_start,
2844 total_needed);
2845 if (free_space) {
2846 u64 start = block_group->key.objectid;
2847 u64 end = block_group->key.objectid +
0f9dd46c 2848 block_group->key.offset;
239b14b3 2849
80eb234a 2850 search_start = stripe_align(root, free_space->offset);
0f9dd46c 2851
80eb234a
JB		 2852 /* move on to the next group */
2853 if (search_start + num_bytes >= search_end)
2854 goto new_group;
e37c9e69 2855
80eb234a
JB		 2856 /* move on to the next group */
2857 if (search_start + num_bytes > end)
2858 goto new_group;
2859
4366211c 2860 if (last_wanted && search_start != last_wanted) {
3b7885bf 2861 total_needed += empty_cluster;
8a1413a2 2862 empty_size += empty_cluster;
4366211c 2863 last_wanted = 0;
3b7885bf
CM		 2864 /*
2865 * if search_start is still in this block group
2866 * then we just re-search this block group
2867 */
2868 if (search_start >= start &&
2869 search_start < end) {
2870 mutex_unlock(&block_group->alloc_mutex);
2871 continue;
2872 }
2873
2874 /* else we go to the next block group */
2875 goto new_group;
2876 }
2877
80eb234a
JB		 2878 if (exclude_nr > 0 &&
2879 (search_start + num_bytes > exclude_start &&
2880 search_start < exclude_start + exclude_nr)) {
2881 search_start = exclude_start + exclude_nr;
2882 /*
2883 * if search_start is still in this block group
2884 * then we just re-search this block group
2885 */
2886 if (search_start >= start &&
25179201
JB		 2887 search_start < end) {
2888 mutex_unlock(&block_group->alloc_mutex);
4366211c 2889 last_wanted = 0;
80eb234a 2890 continue;
25179201 2891 }
80eb234a
JB		 2892
2893 /* else we go to the next block group */
2894 goto new_group;
2895 }
2896
2897 ins->objectid = search_start;
2898 ins->offset = num_bytes;
25179201
JB		 2899
2900 btrfs_remove_free_space_lock(block_group, search_start,
2901 num_bytes);
80eb234a 2902 /* we are all good, lets return */
25179201 2903 mutex_unlock(&block_group->alloc_mutex);
80eb234a 2904 break;
0f9dd46c 2905 }
80eb234a 2906new_group:
42e70e7a 2907 mutex_unlock(&block_group->alloc_mutex);
d2fb3437
YZ		 2908 put_block_group(block_group);
2909 block_group = NULL;
42e70e7a 2910new_group_no_lock:
f5a31e16
CM		 2911 /* don't try to compare new allocations against the
2912 * last allocation any more
2913 */
4366211c 2914 last_wanted = 0;
f5a31e16 2915
80eb234a
JB		 2916 /*
2917 * Here's how this works.
2918 * loop == 0: we were searching a block group via a hint
2919 * and didn't find anything, so we start at
2920 * the head of the block groups and keep searching
2921 * loop == 1: we're searching through all of the block groups
2922 * if we hit the head again we have searched
2923 * all of the block groups for this space and we
2924 * need to try and allocate, if we cant error out.
2925 * loop == 2: we allocated more space and are looping through
2926 * all of the block groups again.
2927 */
2928 if (loop == 0) {
2929 head = &space_info->block_groups;
2930 cur = head->next;
80eb234a
JB		 2931 loop++;
2932 } else if (loop == 1 && cur == head) {
f5a31e16
CM		 2933 int keep_going;
2934
2935 /* at this point we give up on the empty_size
2936 * allocations and just try to allocate the min
2937 * space.
2938 *
2939 * The extra_loop field was set if an empty_size
2940 * allocation was attempted above, and if this
2941 * is try we need to try the loop again without
2942 * the additional empty_size.
2943 */
5b7c3fcc
CM		 2944 total_needed -= empty_size;
2945 empty_size = 0;
f5a31e16
CM		 2946 keep_going = extra_loop;
2947 loop++;
42e70e7a 2948
80eb234a
JB		 2949 if (allowed_chunk_alloc && !chunk_alloc_done) {
2950 up_read(&space_info->groups_sem);
2951 ret = do_chunk_alloc(trans, root, num_bytes +
2952 2 * 1024 * 1024, data, 1);
80eb234a 2953 down_read(&space_info->groups_sem);
2ed6d664
CM		 2954 if (ret < 0)
2955 goto loop_check;
80eb234a 2956 head = &space_info->block_groups;
f5a31e16
CM		 2957 /*
2958 * we've allocated a new chunk, keep
2959 * trying
2960 */
2961 keep_going = 1;
80eb234a
JB		 2962 chunk_alloc_done = 1;
2963 } else if (!allowed_chunk_alloc) {
2964 space_info->force_alloc = 1;
f5a31e16 2965 }
2ed6d664 2966loop_check:
f5a31e16
CM		 2967 if (keep_going) {
2968 cur = head->next;
2969 extra_loop = 0;
80eb234a
JB		 2970 } else {
2971 break;
2972 }
2973 } else if (cur == head) {
2974 break;
0f9dd46c 2975 }
0b86a832 2976
80eb234a
JB		 2977 block_group = list_entry(cur, struct btrfs_block_group_cache,
2978 list);
d2fb3437
YZ		 2979 atomic_inc(&block_group->count);
2980
80eb234a
JB		 2981 search_start = block_group->key.objectid;
2982 cur = cur->next;
f2654de4 2983 }
0b86a832 2984
80eb234a
JB		 2985 /* we found what we needed */
2986 if (ins->objectid) {
2987 if (!(data & BTRFS_BLOCK_GROUP_DATA))
d2fb3437 2988 trans->block_group = block_group->key.objectid;
0f9dd46c 2989
80eb234a
JB		 2990 if (last_ptr)
2991 *last_ptr = ins->objectid + ins->offset;
2992 ret = 0;
2993 } else if (!ret) {
4ce4cb52
JB		 2994 printk(KERN_ERR "we were searching for %Lu bytes, num_bytes %Lu,"
2995 " loop %d, allowed_alloc %d\n", total_needed, num_bytes,
2996 loop, allowed_chunk_alloc);
80eb234a 2997 ret = -ENOSPC;
be744175 2998 }
d2fb3437
YZ		 2999 if (block_group)
3000 put_block_group(block_group);
be744175 3001
80eb234a 3002 up_read(&space_info->groups_sem);
0f70abe2 3003 return ret;
fec577fb 3004}
ec44a35c 3005
0f9dd46c
JB		 3006static void dump_space_info(struct btrfs_space_info *info, u64 bytes)
3007{
3008 struct btrfs_block_group_cache *cache;
3009 struct list_head *l;
3010
3011 printk(KERN_INFO "space_info has %Lu free, is %sfull\n",
e8569813
ZY		 3012 info->total_bytes - info->bytes_used - info->bytes_pinned -
3013 info->bytes_reserved, (info->full) ? "" : "not ");
0f9dd46c 3014
80eb234a 3015 down_read(&info->groups_sem);
0f9dd46c
JB		 3016 list_for_each(l, &info->block_groups) {
3017 cache = list_entry(l, struct btrfs_block_group_cache, list);
3018 spin_lock(&cache->lock);
3019 printk(KERN_INFO "block group %Lu has %Lu bytes, %Lu used "
e8569813 3020 "%Lu pinned %Lu reserved\n",
0f9dd46c 3021 cache->key.objectid, cache->key.offset,
e8569813
ZY		 3022 btrfs_block_group_used(&cache->item),
3023 cache->pinned, cache->reserved);
0f9dd46c
JB		 3024 btrfs_dump_free_space(cache, bytes);
3025 spin_unlock(&cache->lock);
3026 }
80eb234a 3027 up_read(&info->groups_sem);
0f9dd46c 3028}
e8569813 3029
e6dcd2dc
CM		 3030static int __btrfs_reserve_extent(struct btrfs_trans_handle *trans,
3031 struct btrfs_root *root,
3032 u64 num_bytes, u64 min_alloc_size,
3033 u64 empty_size, u64 hint_byte,
3034 u64 search_end, struct btrfs_key *ins,
3035 u64 data)
fec577fb
CM		 3036{
3037 int ret;
fbdc762b 3038 u64 search_start = 0;
8790d502 3039 u64 alloc_profile;
1261ec42 3040 struct btrfs_fs_info *info = root->fs_info;
925baedd 3041
6324fbf3 3042 if (data) {
8790d502
CM		 3043 alloc_profile = info->avail_data_alloc_bits &
3044 info->data_alloc_profile;
3045 data = BTRFS_BLOCK_GROUP_DATA | alloc_profile;
6324fbf3 3046 } else if (root == root->fs_info->chunk_root) {
8790d502
CM		 3047 alloc_profile = info->avail_system_alloc_bits &
3048 info->system_alloc_profile;
3049 data = BTRFS_BLOCK_GROUP_SYSTEM | alloc_profile;
6324fbf3 3050 } else {
8790d502
CM		 3051 alloc_profile = info->avail_metadata_alloc_bits &
3052 info->metadata_alloc_profile;
3053 data = BTRFS_BLOCK_GROUP_METADATA | alloc_profile;
6324fbf3 3054 }
98d20f67 3055again:
2b82032c 3056 data = btrfs_reduce_alloc_profile(root, data);
0ef3e66b
CM		 3057 /*
3058 * the only place that sets empty_size is btrfs_realloc_node, which
3059 * is not called recursively on allocations
3060 */
3061 if (empty_size || root->ref_cows) {
593060d7 3062 if (!(data & BTRFS_BLOCK_GROUP_METADATA)) {
6324fbf3 3063 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
0ef3e66b
CM		 3064 2 * 1024 * 1024,
3065 BTRFS_BLOCK_GROUP_METADATA |
3066 (info->metadata_alloc_profile &
3067 info->avail_metadata_alloc_bits), 0);
6324fbf3
CM		 3068 }
3069 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
0ef3e66b 3070 num_bytes + 2 * 1024 * 1024, data, 0);
6324fbf3 3071 }
0b86a832 3072
db94535d
CM		 3073 WARN_ON(num_bytes < root->sectorsize);
3074 ret = find_free_extent(trans, root, num_bytes, empty_size,
3075 search_start, search_end, hint_byte, ins,
26b8003f
CM		 3076 trans->alloc_exclude_start,
3077 trans->alloc_exclude_nr, data);
3b951516 3078
98d20f67
CM		 3079 if (ret == -ENOSPC && num_bytes > min_alloc_size) {
3080 num_bytes = num_bytes >> 1;
0f9dd46c 3081 num_bytes = num_bytes & ~(root->sectorsize - 1);
98d20f67 3082 num_bytes = max(num_bytes, min_alloc_size);
0ef3e66b
CM		 3083 do_chunk_alloc(trans, root->fs_info->extent_root,
3084 num_bytes, data, 1);
98d20f67
CM		 3085 goto again;
3086 }
ec44a35c 3087 if (ret) {
0f9dd46c
JB		 3088 struct btrfs_space_info *sinfo;
3089
3090 sinfo = __find_space_info(root->fs_info, data);
3091 printk("allocation failed flags %Lu, wanted %Lu\n",
3092 data, num_bytes);
3093 dump_space_info(sinfo, num_bytes);
925baedd 3094 BUG();
925baedd 3095 }
0f9dd46c
JB		 3096
3097 return ret;
e6dcd2dc
CM		 3098}
3099
65b51a00
CM		 3100int btrfs_free_reserved_extent(struct btrfs_root *root, u64 start, u64 len)
3101{
0f9dd46c
JB		 3102 struct btrfs_block_group_cache *cache;
3103
0f9dd46c
JB		 3104 cache = btrfs_lookup_block_group(root->fs_info, start);
3105 if (!cache) {
3106 printk(KERN_ERR "Unable to find block group for %Lu\n", start);
0f9dd46c
JB		 3107 return -ENOSPC;
3108 }
3109 btrfs_add_free_space(cache, start, len);
d2fb3437 3110 put_block_group(cache);
1a40e23b 3111 update_reserved_extents(root, start, len, 0);
65b51a00
CM		 3112 return 0;
3113}
3114
e6dcd2dc
CM		 3115int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
3116 struct btrfs_root *root,
3117 u64 num_bytes, u64 min_alloc_size,
3118 u64 empty_size, u64 hint_byte,
3119 u64 search_end, struct btrfs_key *ins,
3120 u64 data)
3121{
3122 int ret;
e6dcd2dc
CM		 3123 ret = __btrfs_reserve_extent(trans, root, num_bytes, min_alloc_size,
3124 empty_size, hint_byte, search_end, ins,
3125 data);
e8569813 3126 update_reserved_extents(root, ins->objectid, ins->offset, 1);
e6dcd2dc
CM		 3127 return ret;
3128}
3129
3130static int __btrfs_alloc_reserved_extent(struct btrfs_trans_handle *trans,
31840ae1 3131 struct btrfs_root *root, u64 parent,
e6dcd2dc 3132 u64 root_objectid, u64 ref_generation,
3bb1a1bc 3133 u64 owner, struct btrfs_key *ins)
e6dcd2dc
CM		 3134{
3135 int ret;
3136 int pending_ret;
3137 u64 super_used;
3138 u64 root_used;
3139 u64 num_bytes = ins->offset;
3140 u32 sizes[2];
3141 struct btrfs_fs_info *info = root->fs_info;
3142 struct btrfs_root *extent_root = info->extent_root;
3143 struct btrfs_extent_item *extent_item;
3144 struct btrfs_extent_ref *ref;
3145 struct btrfs_path *path;
3146 struct btrfs_key keys[2];
fec577fb 3147
31840ae1
ZY		 3148 if (parent == 0)
3149 parent = ins->objectid;
3150
58176a96 3151 /* block accounting for super block */
a2135011 3152 spin_lock_irq(&info->delalloc_lock);
db94535d
CM		 3153 super_used = btrfs_super_bytes_used(&info->super_copy);
3154 btrfs_set_super_bytes_used(&info->super_copy, super_used + num_bytes);
a2135011 3155 spin_unlock_irq(&info->delalloc_lock);
26b8003f 3156
58176a96 3157 /* block accounting for root item */
db94535d
CM		 3158 root_used = btrfs_root_used(&root->root_item);
3159 btrfs_set_root_used(&root->root_item, root_used + num_bytes);
58176a96 3160
26b8003f 3161 if (root == extent_root) {
31840ae1
ZY		 3162 struct pending_extent_op *extent_op;
3163
3164 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
3165 BUG_ON(!extent_op);
3166
3167 extent_op->type = PENDING_EXTENT_INSERT;
3168 extent_op->bytenr = ins->objectid;
3169 extent_op->num_bytes = ins->offset;
3170 extent_op->parent = parent;
3171 extent_op->orig_parent = 0;
3172 extent_op->generation = ref_generation;
3173 extent_op->orig_generation = 0;
3174 extent_op->level = (int)owner;
f3465ca4
JB		 3175 INIT_LIST_HEAD(&extent_op->list);
3176 extent_op->del = 0;
31840ae1 3177
25179201 3178 mutex_lock(&root->fs_info->extent_ins_mutex);
1a5bc167
CM		 3179 set_extent_bits(&root->fs_info->extent_ins, ins->objectid,
3180 ins->objectid + ins->offset - 1,
25179201 3181 EXTENT_WRITEBACK, GFP_NOFS);
31840ae1
ZY		 3182 set_state_private(&root->fs_info->extent_ins,
3183 ins->objectid, (unsigned long)extent_op);
25179201 3184 mutex_unlock(&root->fs_info->extent_ins_mutex);
26b8003f
CM		 3185 goto update_block;
3186 }
3187
47e4bb98 3188 memcpy(&keys[0], ins, sizeof(*ins));
47e4bb98
CM		 3189 keys[1].objectid = ins->objectid;
3190 keys[1].type = BTRFS_EXTENT_REF_KEY;
31840ae1 3191 keys[1].offset = parent;
47e4bb98
CM		 3192 sizes[0] = sizeof(*extent_item);
3193 sizes[1] = sizeof(*ref);
7bb86316
CM		 3194
3195 path = btrfs_alloc_path();
3196 BUG_ON(!path);
47e4bb98
CM		 3197
3198 ret = btrfs_insert_empty_items(trans, extent_root, path, keys,
3199 sizes, 2);
ccd467d6 3200 BUG_ON(ret);
0f9dd46c 3201
47e4bb98
CM		 3202 extent_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3203 struct btrfs_extent_item);
3204 btrfs_set_extent_refs(path->nodes[0], extent_item, 1);
3205 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
3206 struct btrfs_extent_ref);
3207
3208 btrfs_set_ref_root(path->nodes[0], ref, root_objectid);
3209 btrfs_set_ref_generation(path->nodes[0], ref, ref_generation);
3210 btrfs_set_ref_objectid(path->nodes[0], ref, owner);
31840ae1 3211 btrfs_set_ref_num_refs(path->nodes[0], ref, 1);
47e4bb98
CM		 3212
3213 btrfs_mark_buffer_dirty(path->nodes[0]);
3214
3215 trans->alloc_exclude_start = 0;
3216 trans->alloc_exclude_nr = 0;
7bb86316 3217 btrfs_free_path(path);
87ef2bb4
CM		 3218 finish_current_insert(trans, extent_root, 0);
3219 pending_ret = del_pending_extents(trans, extent_root, 0);
f510cfec 3220
925baedd
CM		 3221 if (ret)
3222 goto out;
e37c9e69 3223 if (pending_ret) {
925baedd
CM		 3224 ret = pending_ret;
3225 goto out;
e37c9e69 3226 }
26b8003f
CM		 3227
3228update_block:
0b86a832 3229 ret = update_block_group(trans, root, ins->objectid, ins->offset, 1, 0);
f5947066
CM		 3230 if (ret) {
3231 printk("update block group failed for %Lu %Lu\n",
3232 ins->objectid, ins->offset);
3233 BUG();
3234 }
925baedd 3235out:
e6dcd2dc
CM		 3236 return ret;
3237}
3238
3239int btrfs_alloc_reserved_extent(struct btrfs_trans_handle *trans,
31840ae1 3240 struct btrfs_root *root, u64 parent,
e6dcd2dc 3241 u64 root_objectid, u64 ref_generation,
3bb1a1bc 3242 u64 owner, struct btrfs_key *ins)
e6dcd2dc
CM		 3243{
3244 int ret;
1c2308f8
CM		 3245
3246 if (root_objectid == BTRFS_TREE_LOG_OBJECTID)
3247 return 0;
3bb1a1bc
YZ		 3248 ret = __btrfs_alloc_reserved_extent(trans, root, parent, root_objectid,
3249 ref_generation, owner, ins);
e8569813 3250 update_reserved_extents(root, ins->objectid, ins->offset, 0);
e6dcd2dc
CM		 3251 return ret;
3252}
e02119d5
CM		 3253
3254/*
3255 * this is used by the tree logging recovery code. It records that
3256 * an extent has been allocated and makes sure to clear the free
3257 * space cache bits as well
3258 */
3259int btrfs_alloc_logged_extent(struct btrfs_trans_handle *trans,
31840ae1 3260 struct btrfs_root *root, u64 parent,
e02119d5 3261 u64 root_objectid, u64 ref_generation,
3bb1a1bc 3262 u64 owner, struct btrfs_key *ins)
e02119d5
CM		 3263{
3264 int ret;
3265 struct btrfs_block_group_cache *block_group;
3266
e02119d5 3267 block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
ea6a478e 3268 mutex_lock(&block_group->cache_mutex);
e02119d5 3269 cache_block_group(root, block_group);
ea6a478e 3270 mutex_unlock(&block_group->cache_mutex);
e02119d5 3271
ea6a478e
JB		 3272 ret = btrfs_remove_free_space(block_group, ins->objectid,
3273 ins->offset);
0f9dd46c 3274 BUG_ON(ret);
d2fb3437 3275 put_block_group(block_group);
3bb1a1bc
YZ		 3276 ret = __btrfs_alloc_reserved_extent(trans, root, parent, root_objectid,
3277 ref_generation, owner, ins);
e02119d5
CM		 3278 return ret;
3279}
3280
e6dcd2dc
CM		 3281/*
3282 * finds a free extent and does all the dirty work required for allocation
3283 * returns the key for the extent through ins, and a tree buffer for
3284 * the first block of the extent through buf.
3285 *
3286 * returns 0 if everything worked, non-zero otherwise.
3287 */
3288int btrfs_alloc_extent(struct btrfs_trans_handle *trans,
3289 struct btrfs_root *root,
31840ae1 3290 u64 num_bytes, u64 parent, u64 min_alloc_size,
e6dcd2dc 3291 u64 root_objectid, u64 ref_generation,
3bb1a1bc 3292 u64 owner_objectid, u64 empty_size, u64 hint_byte,
e6dcd2dc
CM		 3293 u64 search_end, struct btrfs_key *ins, u64 data)
3294{
3295 int ret;
3296
e6dcd2dc
CM		 3297 ret = __btrfs_reserve_extent(trans, root, num_bytes,
3298 min_alloc_size, empty_size, hint_byte,
3299 search_end, ins, data);
3300 BUG_ON(ret);
d00aff00 3301 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
31840ae1
ZY		 3302 ret = __btrfs_alloc_reserved_extent(trans, root, parent,
3303 root_objectid, ref_generation,
3bb1a1bc 3304 owner_objectid, ins);
d00aff00 3305 BUG_ON(ret);
e6dcd2dc 3306
e8569813
ZY		 3307 } else {
3308 update_reserved_extents(root, ins->objectid, ins->offset, 1);
d00aff00 3309 }
925baedd 3310 return ret;
fec577fb 3311}
65b51a00
CM		 3312
3313struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans,
3314 struct btrfs_root *root,
3315 u64 bytenr, u32 blocksize)
3316{
3317 struct extent_buffer *buf;
3318
3319 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
3320 if (!buf)
3321 return ERR_PTR(-ENOMEM);
3322 btrfs_set_header_generation(buf, trans->transid);
3323 btrfs_tree_lock(buf);
3324 clean_tree_block(trans, root, buf);
3325 btrfs_set_buffer_uptodate(buf);
d0c803c4
CM		 3326 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
3327 set_extent_dirty(&root->dirty_log_pages, buf->start,
3328 buf->start + buf->len - 1, GFP_NOFS);
3329 } else {
3330 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
65b51a00 3331 buf->start + buf->len - 1, GFP_NOFS);
d0c803c4 3332 }
65b51a00
CM		 3333 trans->blocks_used++;
3334 return buf;
3335}
3336
fec577fb
CM		 3337/*
3338 * helper function to allocate a block for a given tree
3339 * returns the tree buffer or NULL.
3340 */
5f39d397 3341struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
7bb86316 3342 struct btrfs_root *root,
31840ae1 3343 u32 blocksize, u64 parent,
7bb86316
CM		 3344 u64 root_objectid,
3345 u64 ref_generation,
7bb86316
CM		 3346 int level,
3347 u64 hint,
5f39d397 3348 u64 empty_size)
fec577fb 3349{
e2fa7227 3350 struct btrfs_key ins;
fec577fb 3351 int ret;
5f39d397 3352 struct extent_buffer *buf;
fec577fb 3353
31840ae1 3354 ret = btrfs_alloc_extent(trans, root, blocksize, parent, blocksize,
3bb1a1bc 3355 root_objectid, ref_generation, level,
31840ae1 3356 empty_size, hint, (u64)-1, &ins, 0);
fec577fb 3357 if (ret) {
54aa1f4d
CM		 3358 BUG_ON(ret > 0);
3359 return ERR_PTR(ret);
fec577fb 3360 }
55c69072 3361
65b51a00 3362 buf = btrfs_init_new_buffer(trans, root, ins.objectid, blocksize);
fec577fb
CM		 3363 return buf;
3364}
a28ec197 3365
e02119d5
CM		 3366int btrfs_drop_leaf_ref(struct btrfs_trans_handle *trans,
3367 struct btrfs_root *root, struct extent_buffer *leaf)
6407bf6d 3368{
7bb86316
CM		 3369 u64 leaf_owner;
3370 u64 leaf_generation;
5f39d397 3371 struct btrfs_key key;
6407bf6d
CM		 3372 struct btrfs_file_extent_item *fi;
3373 int i;
3374 int nritems;
3375 int ret;
3376
5f39d397
CM		 3377 BUG_ON(!btrfs_is_leaf(leaf));
3378 nritems = btrfs_header_nritems(leaf);
7bb86316
CM		 3379 leaf_owner = btrfs_header_owner(leaf);
3380 leaf_generation = btrfs_header_generation(leaf);
3381
6407bf6d 3382 for (i = 0; i < nritems; i++) {
db94535d 3383 u64 disk_bytenr;
e34a5b4f 3384 cond_resched();
5f39d397
CM		 3385
3386 btrfs_item_key_to_cpu(leaf, &key, i);
3387 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
6407bf6d
CM		 3388 continue;
3389 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
5f39d397
CM		 3390 if (btrfs_file_extent_type(leaf, fi) ==
3391 BTRFS_FILE_EXTENT_INLINE)
236454df 3392 continue;
6407bf6d
CM		 3393 /*
3394 * FIXME make sure to insert a trans record that
3395 * repeats the snapshot del on crash
3396 */
db94535d
CM		 3397 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
3398 if (disk_bytenr == 0)
3a686375 3399 continue;
4a096752 3400
925baedd 3401 ret = __btrfs_free_extent(trans, root, disk_bytenr,
7bb86316 3402 btrfs_file_extent_disk_num_bytes(leaf, fi),
31840ae1 3403 leaf->start, leaf_owner, leaf_generation,
3bb1a1bc 3404 key.objectid, 0);
31840ae1 3405 BUG_ON(ret);
2dd3e67b
CM		 3406
3407 atomic_inc(&root->fs_info->throttle_gen);
3408 wake_up(&root->fs_info->transaction_throttle);
3409 cond_resched();
6407bf6d
CM		 3410 }
3411 return 0;
3412}
3413
e02119d5
CM		 3414static int noinline cache_drop_leaf_ref(struct btrfs_trans_handle *trans,
3415 struct btrfs_root *root,
3416 struct btrfs_leaf_ref *ref)
31153d81
YZ		 3417{
3418 int i;
3419 int ret;
3420 struct btrfs_extent_info *info = ref->extents;
3421
31153d81 3422 for (i = 0; i < ref->nritems; i++) {
31840ae1
ZY		 3423 ret = __btrfs_free_extent(trans, root, info->bytenr,
3424 info->num_bytes, ref->bytenr,
3425 ref->owner, ref->generation,
3bb1a1bc 3426 info->objectid, 0);
2dd3e67b
CM		 3427
3428 atomic_inc(&root->fs_info->throttle_gen);
3429 wake_up(&root->fs_info->transaction_throttle);
3430 cond_resched();
3431
31153d81
YZ		 3432 BUG_ON(ret);
3433 info++;
3434 }
31153d81
YZ		 3435
3436 return 0;
3437}
3438
b2950863 3439static int drop_snap_lookup_refcount(struct btrfs_root *root, u64 start, u64 len,
333db94c
CM		 3440 u32 *refs)
3441{
017e5369 3442 int ret;
f87f057b 3443
31840ae1 3444 ret = btrfs_lookup_extent_ref(NULL, root, start, len, refs);
f87f057b
CM		 3445 BUG_ON(ret);
3446
3447#if 0 // some debugging code in case we see problems here
3448 /* if the refs count is one, it won't get increased again. But
3449 * if the ref count is > 1, someone may be decreasing it at
3450 * the same time we are.
3451 */
3452 if (*refs != 1) {
3453 struct extent_buffer *eb = NULL;
3454 eb = btrfs_find_create_tree_block(root, start, len);
3455 if (eb)
3456 btrfs_tree_lock(eb);
3457
3458 mutex_lock(&root->fs_info->alloc_mutex);
3459 ret = lookup_extent_ref(NULL, root, start, len, refs);
3460 BUG_ON(ret);
3461 mutex_unlock(&root->fs_info->alloc_mutex);
3462
3463 if (eb) {
3464 btrfs_tree_unlock(eb);
3465 free_extent_buffer(eb);
3466 }
3467 if (*refs == 1) {
3468 printk("block %llu went down to one during drop_snap\n",
3469 (unsigned long long)start);
3470 }
3471
3472 }
3473#endif
3474
e7a84565 3475 cond_resched();
017e5369 3476 return ret;
333db94c
CM		 3477}
3478
9aca1d51
CM		 3479/*
3480 * helper function for drop_snapshot, this walks down the tree dropping ref
3481 * counts as it goes.
3482 */
98ed5174
CM		 3483static int noinline walk_down_tree(struct btrfs_trans_handle *trans,
3484 struct btrfs_root *root,
3485 struct btrfs_path *path, int *level)
20524f02 3486{
7bb86316
CM		 3487 u64 root_owner;
3488 u64 root_gen;
3489 u64 bytenr;
ca7a79ad 3490 u64 ptr_gen;
5f39d397
CM		 3491 struct extent_buffer *next;
3492 struct extent_buffer *cur;
7bb86316 3493 struct extent_buffer *parent;
31153d81 3494 struct btrfs_leaf_ref *ref;
db94535d 3495 u32 blocksize;
20524f02
CM		 3496 int ret;
3497 u32 refs;
3498
5caf2a00
CM		 3499 WARN_ON(*level < 0);
3500 WARN_ON(*level >= BTRFS_MAX_LEVEL);
333db94c 3501 ret = drop_snap_lookup_refcount(root, path->nodes[*level]->start,
db94535d 3502 path->nodes[*level]->len, &refs);
20524f02
CM		 3503 BUG_ON(ret);
3504 if (refs > 1)
3505 goto out;
e011599b 3506
9aca1d51
CM		 3507 /*
3508 * walk down to the last node level and free all the leaves
3509 */
6407bf6d 3510 while(*level >= 0) {
5caf2a00
CM		 3511 WARN_ON(*level < 0);
3512 WARN_ON(*level >= BTRFS_MAX_LEVEL);
20524f02 3513 cur = path->nodes[*level];
e011599b 3514
5f39d397 3515 if (btrfs_header_level(cur) != *level)
2c90e5d6 3516 WARN_ON(1);
e011599b 3517
7518a238 3518 if (path->slots[*level] >=
5f39d397 3519 btrfs_header_nritems(cur))
20524f02 3520 break;
6407bf6d 3521 if (*level == 0) {
e02119d5 3522 ret = btrfs_drop_leaf_ref(trans, root, cur);
6407bf6d
CM		 3523 BUG_ON(ret);
3524 break;
3525 }
db94535d 3526 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
ca7a79ad 3527 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
db94535d 3528 blocksize = btrfs_level_size(root, *level - 1);
925baedd 3529
333db94c 3530 ret = drop_snap_lookup_refcount(root, bytenr, blocksize, &refs);
6407bf6d
CM		 3531 BUG_ON(ret);
3532 if (refs != 1) {
7bb86316
CM		 3533 parent = path->nodes[*level];
3534 root_owner = btrfs_header_owner(parent);
3535 root_gen = btrfs_header_generation(parent);
20524f02 3536 path->slots[*level]++;
f87f057b 3537
925baedd 3538 ret = __btrfs_free_extent(trans, root, bytenr,
31840ae1 3539 blocksize, parent->start,
3bb1a1bc
YZ		 3540 root_owner, root_gen,
3541 *level - 1, 1);
20524f02 3542 BUG_ON(ret);
18e35e0a
CM		 3543
3544 atomic_inc(&root->fs_info->throttle_gen);
3545 wake_up(&root->fs_info->transaction_throttle);
2dd3e67b 3546 cond_resched();
18e35e0a 3547
20524f02
CM		 3548 continue;
3549 }
f87f057b
CM		 3550 /*
3551 * at this point, we have a single ref, and since the
3552 * only place referencing this extent is a dead root
3553 * the reference count should never go higher.
3554 * So, we don't need to check it again
3555 */
31153d81 3556 if (*level == 1) {
017e5369 3557 ref = btrfs_lookup_leaf_ref(root, bytenr);
1a40e23b
ZY		 3558 if (ref && ref->generation != ptr_gen) {
3559 btrfs_free_leaf_ref(root, ref);
3560 ref = NULL;
3561 }
31153d81 3562 if (ref) {
e02119d5 3563 ret = cache_drop_leaf_ref(trans, root, ref);
31153d81
YZ		 3564 BUG_ON(ret);
3565 btrfs_remove_leaf_ref(root, ref);
bcc63abb 3566 btrfs_free_leaf_ref(root, ref);
31153d81
YZ		 3567 *level = 0;
3568 break;
3569 }
3bb1a1bc 3570 if (printk_ratelimit()) {
37d1aeee
CM		 3571 printk("leaf ref miss for bytenr %llu\n",
3572 (unsigned long long)bytenr);
3bb1a1bc 3573 }
31153d81 3574 }
db94535d 3575 next = btrfs_find_tree_block(root, bytenr, blocksize);
1259ab75 3576 if (!next || !btrfs_buffer_uptodate(next, ptr_gen)) {
5f39d397 3577 free_extent_buffer(next);
333db94c 3578
ca7a79ad
CM		 3579 next = read_tree_block(root, bytenr, blocksize,
3580 ptr_gen);
e7a84565 3581 cond_resched();
f87f057b
CM		 3582#if 0
3583 /*
3584 * this is a debugging check and can go away
3585 * the ref should never go all the way down to 1
3586 * at this point
3587 */
e6dcd2dc
CM		 3588 ret = lookup_extent_ref(NULL, root, bytenr, blocksize,
3589 &refs);
e9d0b13b 3590 BUG_ON(ret);
f87f057b
CM		 3591 WARN_ON(refs != 1);
3592#endif
e9d0b13b 3593 }
5caf2a00 3594 WARN_ON(*level <= 0);
83e15a28 3595 if (path->nodes[*level-1])
5f39d397 3596 free_extent_buffer(path->nodes[*level-1]);
20524f02 3597 path->nodes[*level-1] = next;
5f39d397 3598 *level = btrfs_header_level(next);
20524f02 3599 path->slots[*level] = 0;
2dd3e67b 3600 cond_resched();
20524f02
CM		 3601 }
3602out:
5caf2a00
CM		 3603 WARN_ON(*level < 0);
3604 WARN_ON(*level >= BTRFS_MAX_LEVEL);
7bb86316
CM		 3605
3606 if (path->nodes[*level] == root->node) {
7bb86316 3607 parent = path->nodes[*level];
31153d81 3608 bytenr = path->nodes[*level]->start;
7bb86316
CM		 3609 } else {
3610 parent = path->nodes[*level + 1];
31153d81 3611 bytenr = btrfs_node_blockptr(parent, path->slots[*level + 1]);
7bb86316
CM		 3612 }
3613
31153d81
YZ		 3614 blocksize = btrfs_level_size(root, *level);
3615 root_owner = btrfs_header_owner(parent);
7bb86316 3616 root_gen = btrfs_header_generation(parent);
31153d81
YZ		 3617
3618 ret = __btrfs_free_extent(trans, root, bytenr, blocksize,
31840ae1 3619 parent->start, root_owner, root_gen,
3bb1a1bc 3620 *level, 1);
5f39d397 3621 free_extent_buffer(path->nodes[*level]);
20524f02
CM		 3622 path->nodes[*level] = NULL;
3623 *level += 1;
3624 BUG_ON(ret);
f87f057b 3625
e7a84565 3626 cond_resched();
20524f02
CM		 3627 return 0;
3628}
3629
f82d02d9
YZ		 3630/*
3631 * helper function for drop_subtree, this function is similar to
3632 * walk_down_tree. The main difference is that it checks reference
3633 * counts while tree blocks are locked.
3634 */
3635static int noinline walk_down_subtree(struct btrfs_trans_handle *trans,
3636 struct btrfs_root *root,
3637 struct btrfs_path *path, int *level)
3638{
3639 struct extent_buffer *next;
3640 struct extent_buffer *cur;
3641 struct extent_buffer *parent;
3642 u64 bytenr;
3643 u64 ptr_gen;
3644 u32 blocksize;
3645 u32 refs;
3646 int ret;
3647
3648 cur = path->nodes[*level];
3649 ret = btrfs_lookup_extent_ref(trans, root, cur->start, cur->len,
3650 &refs);
3651 BUG_ON(ret);
3652 if (refs > 1)
3653 goto out;
3654
3655 while (*level >= 0) {
3656 cur = path->nodes[*level];
3657 if (*level == 0) {
3658 ret = btrfs_drop_leaf_ref(trans, root, cur);
3659 BUG_ON(ret);
3660 clean_tree_block(trans, root, cur);
3661 break;
3662 }
3663 if (path->slots[*level] >= btrfs_header_nritems(cur)) {
3664 clean_tree_block(trans, root, cur);
3665 break;
3666 }
3667
3668 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
3669 blocksize = btrfs_level_size(root, *level - 1);
3670 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
3671
3672 next = read_tree_block(root, bytenr, blocksize, ptr_gen);
3673 btrfs_tree_lock(next);
3674
3675 ret = btrfs_lookup_extent_ref(trans, root, bytenr, blocksize,
3676 &refs);
3677 BUG_ON(ret);
3678 if (refs > 1) {
3679 parent = path->nodes[*level];
3680 ret = btrfs_free_extent(trans, root, bytenr,
3681 blocksize, parent->start,
3682 btrfs_header_owner(parent),
3683 btrfs_header_generation(parent),
3684 *level - 1, 1);
3685 BUG_ON(ret);
3686 path->slots[*level]++;
3687 btrfs_tree_unlock(next);
3688 free_extent_buffer(next);
3689 continue;
3690 }
3691
3692 *level = btrfs_header_level(next);
3693 path->nodes[*level] = next;
3694 path->slots[*level] = 0;
3695 path->locks[*level] = 1;
3696 cond_resched();
3697 }
3698out:
3699 parent = path->nodes[*level + 1];
3700 bytenr = path->nodes[*level]->start;
3701 blocksize = path->nodes[*level]->len;
3702
3703 ret = btrfs_free_extent(trans, root, bytenr, blocksize,
3704 parent->start, btrfs_header_owner(parent),
3705 btrfs_header_generation(parent), *level, 1);
3706 BUG_ON(ret);
3707
3708 if (path->locks[*level]) {
3709 btrfs_tree_unlock(path->nodes[*level]);
3710 path->locks[*level] = 0;
3711 }
3712 free_extent_buffer(path->nodes[*level]);
3713 path->nodes[*level] = NULL;
3714 *level += 1;
3715 cond_resched();
3716 return 0;
3717}
3718
9aca1d51
CM		 3719/*
3720 * helper for dropping snapshots. This walks back up the tree in the path
3721 * to find the first node higher up where we haven't yet gone through
3722 * all the slots
3723 */
98ed5174
CM		 3724static int noinline walk_up_tree(struct btrfs_trans_handle *trans,
3725 struct btrfs_root *root,
f82d02d9
YZ		 3726 struct btrfs_path *path,
3727 int *level, int max_level)
20524f02 3728{
7bb86316
CM		 3729 u64 root_owner;
3730 u64 root_gen;
3731 struct btrfs_root_item *root_item = &root->root_item;
20524f02
CM		 3732 int i;
3733 int slot;
3734 int ret;
9f3a7427 3735
f82d02d9 3736 for (i = *level; i < max_level && path->nodes[i]; i++) {
20524f02 3737 slot = path->slots[i];
5f39d397
CM		 3738 if (slot < btrfs_header_nritems(path->nodes[i]) - 1) {
3739 struct extent_buffer *node;
3740 struct btrfs_disk_key disk_key;
3741 node = path->nodes[i];
20524f02
CM		 3742 path->slots[i]++;
3743 *level = i;
9f3a7427 3744 WARN_ON(*level == 0);
5f39d397 3745 btrfs_node_key(node, &disk_key, path->slots[i]);
9f3a7427 3746 memcpy(&root_item->drop_progress,
5f39d397 3747 &disk_key, sizeof(disk_key));
9f3a7427 3748 root_item->drop_level = i;
20524f02
CM		 3749 return 0;
3750 } else {
31840ae1
ZY		 3751 struct extent_buffer *parent;
3752 if (path->nodes[*level] == root->node)
3753 parent = path->nodes[*level];
3754 else
3755 parent = path->nodes[*level + 1];
3756
3757 root_owner = btrfs_header_owner(parent);
3758 root_gen = btrfs_header_generation(parent);
f82d02d9
YZ		 3759
3760 clean_tree_block(trans, root, path->nodes[*level]);
e089f05c 3761 ret = btrfs_free_extent(trans, root,
db94535d 3762 path->nodes[*level]->start,
7bb86316 3763 path->nodes[*level]->len,
3bb1a1bc
YZ		 3764 parent->start, root_owner,
3765 root_gen, *level, 1);
6407bf6d 3766 BUG_ON(ret);
f82d02d9
YZ		 3767 if (path->locks[*level]) {
3768 btrfs_tree_unlock(path->nodes[*level]);
3769 path->locks[*level] = 0;
3770 }
5f39d397 3771 free_extent_buffer(path->nodes[*level]);
83e15a28 3772 path->nodes[*level] = NULL;
20524f02 3773 *level = i + 1;
20524f02
CM		 3774 }
3775 }
3776 return 1;
3777}
3778
9aca1d51
CM		 3779/*
3780 * drop the reference count on the tree rooted at 'snap'. This traverses
3781 * the tree freeing any blocks that have a ref count of zero after being
3782 * decremented.
3783 */
e089f05c 3784int btrfs_drop_snapshot(struct btrfs_trans_handle *trans, struct btrfs_root
9f3a7427 3785 *root)
20524f02 3786{
3768f368 3787 int ret = 0;
9aca1d51 3788 int wret;
20524f02 3789 int level;
5caf2a00 3790 struct btrfs_path *path;
20524f02
CM		 3791 int i;
3792 int orig_level;
9f3a7427 3793 struct btrfs_root_item *root_item = &root->root_item;
20524f02 3794
a2135011 3795 WARN_ON(!mutex_is_locked(&root->fs_info->drop_mutex));
5caf2a00
CM		 3796 path = btrfs_alloc_path();
3797 BUG_ON(!path);
20524f02 3798
5f39d397 3799 level = btrfs_header_level(root->node);
20524f02 3800 orig_level = level;
9f3a7427
CM		 3801 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
3802 path->nodes[level] = root->node;
f510cfec 3803 extent_buffer_get(root->node);
9f3a7427
CM		 3804 path->slots[level] = 0;
3805 } else {
3806 struct btrfs_key key;
5f39d397
CM		 3807 struct btrfs_disk_key found_key;
3808 struct extent_buffer *node;
6702ed49 3809
9f3a7427 3810 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
6702ed49
CM		 3811 level = root_item->drop_level;
3812 path->lowest_level = level;
9f3a7427 3813 wret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6702ed49 3814 if (wret < 0) {
9f3a7427
CM		 3815 ret = wret;
3816 goto out;
3817 }
5f39d397
CM		 3818 node = path->nodes[level];
3819 btrfs_node_key(node, &found_key, path->slots[level]);
3820 WARN_ON(memcmp(&found_key, &root_item->drop_progress,
3821 sizeof(found_key)));
7d9eb12c
CM		 3822 /*
3823 * unlock our path, this is safe because only this
3824 * function is allowed to delete this snapshot
3825 */
925baedd
CM		 3826 for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
3827 if (path->nodes[i] && path->locks[i]) {
3828 path->locks[i] = 0;
3829 btrfs_tree_unlock(path->nodes[i]);
3830 }
3831 }
9f3a7427 3832 }
20524f02 3833 while(1) {
5caf2a00 3834 wret = walk_down_tree(trans, root, path, &level);
9aca1d51 3835 if (wret > 0)
20524f02 3836 break;
9aca1d51
CM		 3837 if (wret < 0)
3838 ret = wret;
3839
f82d02d9
YZ		 3840 wret = walk_up_tree(trans, root, path, &level,
3841 BTRFS_MAX_LEVEL);
9aca1d51 3842 if (wret > 0)
20524f02 3843 break;
9aca1d51
CM		 3844 if (wret < 0)
3845 ret = wret;
e7a84565
CM		 3846 if (trans->transaction->in_commit) {
3847 ret = -EAGAIN;
3848 break;
3849 }
18e35e0a 3850 atomic_inc(&root->fs_info->throttle_gen);
017e5369 3851 wake_up(&root->fs_info->transaction_throttle);
20524f02 3852 }
83e15a28 3853 for (i = 0; i <= orig_level; i++) {
5caf2a00 3854 if (path->nodes[i]) {
5f39d397 3855 free_extent_buffer(path->nodes[i]);
0f82731f 3856 path->nodes[i] = NULL;
83e15a28 3857 }
20524f02 3858 }
9f3a7427 3859out:
5caf2a00 3860 btrfs_free_path(path);
9aca1d51 3861 return ret;
20524f02 3862}
9078a3e1 3863
f82d02d9
YZ		 3864int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
3865 struct btrfs_root *root,
3866 struct extent_buffer *node,
3867 struct extent_buffer *parent)
3868{
3869 struct btrfs_path *path;
3870 int level;
3871 int parent_level;
3872 int ret = 0;
3873 int wret;
3874
3875 path = btrfs_alloc_path();
3876 BUG_ON(!path);
3877
3878 BUG_ON(!btrfs_tree_locked(parent));
3879 parent_level = btrfs_header_level(parent);
3880 extent_buffer_get(parent);
3881 path->nodes[parent_level] = parent;
3882 path->slots[parent_level] = btrfs_header_nritems(parent);
3883
3884 BUG_ON(!btrfs_tree_locked(node));
3885 level = btrfs_header_level(node);
3886 extent_buffer_get(node);
3887 path->nodes[level] = node;
3888 path->slots[level] = 0;
3889
3890 while (1) {
3891 wret = walk_down_subtree(trans, root, path, &level);
3892 if (wret < 0)
3893 ret = wret;
3894 if (wret != 0)
3895 break;
3896
3897 wret = walk_up_tree(trans, root, path, &level, parent_level);
3898 if (wret < 0)
3899 ret = wret;
3900 if (wret != 0)
3901 break;
3902 }
3903
3904 btrfs_free_path(path);
3905 return ret;
3906}
3907
8e7bf94f
CM		 3908static unsigned long calc_ra(unsigned long start, unsigned long last,
3909 unsigned long nr)
3910{
3911 return min(last, start + nr - 1);
3912}
3913
98ed5174
CM		 3914static int noinline relocate_inode_pages(struct inode *inode, u64 start,
3915 u64 len)
edbd8d4e
CM		 3916{
3917 u64 page_start;
3918 u64 page_end;
1a40e23b 3919 unsigned long first_index;
edbd8d4e 3920 unsigned long last_index;
edbd8d4e
CM		 3921 unsigned long i;
3922 struct page *page;
d1310b2e 3923 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4313b399 3924 struct file_ra_state *ra;
3eaa2885 3925 struct btrfs_ordered_extent *ordered;
1a40e23b
ZY		 3926 unsigned int total_read = 0;
3927 unsigned int total_dirty = 0;
3928 int ret = 0;
4313b399
CM		 3929
3930 ra = kzalloc(sizeof(*ra), GFP_NOFS);
edbd8d4e
CM		 3931
3932 mutex_lock(&inode->i_mutex);
1a40e23b 3933 first_index = start >> PAGE_CACHE_SHIFT;
edbd8d4e
CM		 3934 last_index = (start + len - 1) >> PAGE_CACHE_SHIFT;
3935
1a40e23b
ZY		 3936 /* make sure the dirty trick played by the caller work */
3937 ret = invalidate_inode_pages2_range(inode->i_mapping,
3938 first_index, last_index);
3939 if (ret)
3940 goto out_unlock;
8e7bf94f 3941
4313b399 3942 file_ra_state_init(ra, inode->i_mapping);
edbd8d4e 3943
1a40e23b
ZY		 3944 for (i = first_index ; i <= last_index; i++) {
3945 if (total_read % ra->ra_pages == 0) {
8e7bf94f 3946 btrfs_force_ra(inode->i_mapping, ra, NULL, i,
1a40e23b 3947 calc_ra(i, last_index, ra->ra_pages));
8e7bf94f
CM		 3948 }
3949 total_read++;
3eaa2885
CM		 3950again:
3951 if (((u64)i << PAGE_CACHE_SHIFT) > i_size_read(inode))
1a40e23b 3952 BUG_ON(1);
edbd8d4e 3953 page = grab_cache_page(inode->i_mapping, i);
a061fc8d 3954 if (!page) {
1a40e23b 3955 ret = -ENOMEM;
edbd8d4e 3956 goto out_unlock;
a061fc8d 3957 }
edbd8d4e
CM		 3958 if (!PageUptodate(page)) {
3959 btrfs_readpage(NULL, page);
3960 lock_page(page);
3961 if (!PageUptodate(page)) {
3962 unlock_page(page);
3963 page_cache_release(page);
1a40e23b 3964 ret = -EIO;
edbd8d4e
CM		 3965 goto out_unlock;
3966 }
3967 }
ec44a35c 3968 wait_on_page_writeback(page);
3eaa2885 3969
edbd8d4e
CM		 3970 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
3971 page_end = page_start + PAGE_CACHE_SIZE - 1;
d1310b2e 3972 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
edbd8d4e 3973
3eaa2885
CM		 3974 ordered = btrfs_lookup_ordered_extent(inode, page_start);
3975 if (ordered) {
3976 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
3977 unlock_page(page);
3978 page_cache_release(page);
3979 btrfs_start_ordered_extent(inode, ordered, 1);
3980 btrfs_put_ordered_extent(ordered);
3981 goto again;
3982 }
3983 set_page_extent_mapped(page);
3984
ea8c2819 3985 btrfs_set_extent_delalloc(inode, page_start, page_end);
1a40e23b
ZY		 3986 if (i == first_index)
3987 set_extent_bits(io_tree, page_start, page_end,
3988 EXTENT_BOUNDARY, GFP_NOFS);
3989
a061fc8d 3990 set_page_dirty(page);
1a40e23b 3991 total_dirty++;
edbd8d4e 3992
d1310b2e 3993 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
edbd8d4e
CM		 3994 unlock_page(page);
3995 page_cache_release(page);
3996 }
3997
3998out_unlock:
ec44a35c 3999 kfree(ra);
edbd8d4e 4000 mutex_unlock(&inode->i_mutex);
1a40e23b
ZY		 4001 balance_dirty_pages_ratelimited_nr(inode->i_mapping, total_dirty);
4002 return ret;
edbd8d4e
CM		 4003}
4004
1a40e23b
ZY		 4005static int noinline relocate_data_extent(struct inode *reloc_inode,
4006 struct btrfs_key *extent_key,
4007 u64 offset)
4008{
4009 struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
4010 struct extent_map_tree *em_tree = &BTRFS_I(reloc_inode)->extent_tree;
4011 struct extent_map *em;
6643558d
YZ		 4012 u64 start = extent_key->objectid - offset;
4013 u64 end = start + extent_key->offset - 1;
bf4ef679 4014
1a40e23b
ZY		 4015 em = alloc_extent_map(GFP_NOFS);
4016 BUG_ON(!em || IS_ERR(em));
bf4ef679 4017
6643558d 4018 em->start = start;
1a40e23b 4019 em->len = extent_key->offset;
c8b97818 4020 em->block_len = extent_key->offset;
1a40e23b
ZY		 4021 em->block_start = extent_key->objectid;
4022 em->bdev = root->fs_info->fs_devices->latest_bdev;
4023 set_bit(EXTENT_FLAG_PINNED, &em->flags);
4024
4025 /* setup extent map to cheat btrfs_readpage */
6643558d 4026 lock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
1a40e23b
ZY		 4027 while (1) {
4028 int ret;
4029 spin_lock(&em_tree->lock);
4030 ret = add_extent_mapping(em_tree, em);
4031 spin_unlock(&em_tree->lock);
4032 if (ret != -EEXIST) {
4033 free_extent_map(em);
bf4ef679
CM		 4034 break;
4035 }
6643558d 4036 btrfs_drop_extent_cache(reloc_inode, start, end, 0);
bf4ef679 4037 }
6643558d 4038 unlock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
bf4ef679 4039
6643558d 4040 return relocate_inode_pages(reloc_inode, start, extent_key->offset);
1a40e23b 4041}
edbd8d4e 4042
1a40e23b
ZY		 4043struct btrfs_ref_path {
4044 u64 extent_start;
4045 u64 nodes[BTRFS_MAX_LEVEL];
4046 u64 root_objectid;
4047 u64 root_generation;
4048 u64 owner_objectid;
1a40e23b
ZY		 4049 u32 num_refs;
4050 int lowest_level;
4051 int current_level;
f82d02d9
YZ		 4052 int shared_level;
4053
4054 struct btrfs_key node_keys[BTRFS_MAX_LEVEL];
4055 u64 new_nodes[BTRFS_MAX_LEVEL];
1a40e23b 4056};
7d9eb12c 4057
1a40e23b 4058struct disk_extent {
c8b97818 4059 u64 ram_bytes;
1a40e23b
ZY		 4060 u64 disk_bytenr;
4061 u64 disk_num_bytes;
4062 u64 offset;
4063 u64 num_bytes;
c8b97818
CM		 4064 u8 compression;
4065 u8 encryption;
4066 u16 other_encoding;
1a40e23b 4067};
4313b399 4068
1a40e23b
ZY		 4069static int is_cowonly_root(u64 root_objectid)
4070{
4071 if (root_objectid == BTRFS_ROOT_TREE_OBJECTID ||
4072 root_objectid == BTRFS_EXTENT_TREE_OBJECTID ||
4073 root_objectid == BTRFS_CHUNK_TREE_OBJECTID ||
4074 root_objectid == BTRFS_DEV_TREE_OBJECTID ||
0403e47e
YZ		 4075 root_objectid == BTRFS_TREE_LOG_OBJECTID ||
4076 root_objectid == BTRFS_CSUM_TREE_OBJECTID)
1a40e23b
ZY		 4077 return 1;
4078 return 0;
4079}
edbd8d4e 4080
1a40e23b
ZY		 4081static int noinline __next_ref_path(struct btrfs_trans_handle *trans,
4082 struct btrfs_root *extent_root,
4083 struct btrfs_ref_path *ref_path,
4084 int first_time)
4085{
4086 struct extent_buffer *leaf;
4087 struct btrfs_path *path;
4088 struct btrfs_extent_ref *ref;
4089 struct btrfs_key key;
4090 struct btrfs_key found_key;
4091 u64 bytenr;
4092 u32 nritems;
4093 int level;
4094 int ret = 1;
edbd8d4e 4095
1a40e23b
ZY		 4096 path = btrfs_alloc_path();
4097 if (!path)
4098 return -ENOMEM;
bf4ef679 4099
1a40e23b
ZY		 4100 if (first_time) {
4101 ref_path->lowest_level = -1;
4102 ref_path->current_level = -1;
f82d02d9 4103 ref_path->shared_level = -1;
1a40e23b
ZY		 4104 goto walk_up;
4105 }
4106walk_down:
4107 level = ref_path->current_level - 1;
4108 while (level >= -1) {
4109 u64 parent;
4110 if (level < ref_path->lowest_level)
4111 break;
bf4ef679 4112
1a40e23b
ZY		 4113 if (level >= 0) {
4114 bytenr = ref_path->nodes[level];
4115 } else {
4116 bytenr = ref_path->extent_start;
4117 }
4118 BUG_ON(bytenr == 0);
bf4ef679 4119
1a40e23b
ZY		 4120 parent = ref_path->nodes[level + 1];
4121 ref_path->nodes[level + 1] = 0;
4122 ref_path->current_level = level;
4123 BUG_ON(parent == 0);
0ef3e66b 4124
1a40e23b
ZY		 4125 key.objectid = bytenr;
4126 key.offset = parent + 1;
4127 key.type = BTRFS_EXTENT_REF_KEY;
edbd8d4e 4128
1a40e23b
ZY		 4129 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
4130 if (ret < 0)
edbd8d4e 4131 goto out;
1a40e23b 4132 BUG_ON(ret == 0);
7d9eb12c 4133
1a40e23b
ZY		 4134 leaf = path->nodes[0];
4135 nritems = btrfs_header_nritems(leaf);
4136 if (path->slots[0] >= nritems) {
4137 ret = btrfs_next_leaf(extent_root, path);
4138 if (ret < 0)
4139 goto out;
4140 if (ret > 0)
4141 goto next;
4142 leaf = path->nodes[0];
4143 }
0ef3e66b 4144
1a40e23b
ZY		 4145 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4146 if (found_key.objectid == bytenr &&
f82d02d9
YZ		 4147 found_key.type == BTRFS_EXTENT_REF_KEY) {
4148 if (level < ref_path->shared_level)
4149 ref_path->shared_level = level;
1a40e23b 4150 goto found;
f82d02d9 4151 }
1a40e23b
ZY		 4152next:
4153 level--;
4154 btrfs_release_path(extent_root, path);
d899e052 4155 cond_resched();
1a40e23b
ZY		 4156 }
4157 /* reached lowest level */
4158 ret = 1;
4159 goto out;
4160walk_up:
4161 level = ref_path->current_level;
4162 while (level < BTRFS_MAX_LEVEL - 1) {
4163 u64 ref_objectid;
4164 if (level >= 0) {
4165 bytenr = ref_path->nodes[level];
4166 } else {
4167 bytenr = ref_path->extent_start;
4168 }
4169 BUG_ON(bytenr == 0);
edbd8d4e 4170
1a40e23b
ZY		 4171 key.objectid = bytenr;
4172 key.offset = 0;
4173 key.type = BTRFS_EXTENT_REF_KEY;
edbd8d4e 4174
1a40e23b
ZY		 4175 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
4176 if (ret < 0)
4177 goto out;
edbd8d4e 4178
1a40e23b
ZY		 4179 leaf = path->nodes[0];
4180 nritems = btrfs_header_nritems(leaf);
4181 if (path->slots[0] >= nritems) {
4182 ret = btrfs_next_leaf(extent_root, path);
4183 if (ret < 0)
4184 goto out;
4185 if (ret > 0) {
4186 /* the extent was freed by someone */
4187 if (ref_path->lowest_level == level)
4188 goto out;
4189 btrfs_release_path(extent_root, path);
4190 goto walk_down;
4191 }
4192 leaf = path->nodes[0];
4193 }
edbd8d4e 4194
1a40e23b
ZY		 4195 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4196 if (found_key.objectid != bytenr ||
4197 found_key.type != BTRFS_EXTENT_REF_KEY) {
4198 /* the extent was freed by someone */
4199 if (ref_path->lowest_level == level) {
4200 ret = 1;
4201 goto out;
4202 }
4203 btrfs_release_path(extent_root, path);
4204 goto walk_down;
4205 }
4206found:
4207 ref = btrfs_item_ptr(leaf, path->slots[0],
4208 struct btrfs_extent_ref);
4209 ref_objectid = btrfs_ref_objectid(leaf, ref);
4210 if (ref_objectid < BTRFS_FIRST_FREE_OBJECTID) {
4211 if (first_time) {
4212 level = (int)ref_objectid;
4213 BUG_ON(level >= BTRFS_MAX_LEVEL);
4214 ref_path->lowest_level = level;
4215 ref_path->current_level = level;
4216 ref_path->nodes[level] = bytenr;
4217 } else {
4218 WARN_ON(ref_objectid != level);
4219 }
4220 } else {
4221 WARN_ON(level != -1);
4222 }
4223 first_time = 0;
bf4ef679 4224
1a40e23b
ZY		 4225 if (ref_path->lowest_level == level) {
4226 ref_path->owner_objectid = ref_objectid;
1a40e23b
ZY		 4227 ref_path->num_refs = btrfs_ref_num_refs(leaf, ref);
4228 }
bf4ef679 4229
7d9eb12c 4230 /*
1a40e23b
ZY		 4231 * the block is tree root or the block isn't in reference
4232 * counted tree.
7d9eb12c 4233 */
1a40e23b
ZY		 4234 if (found_key.objectid == found_key.offset ||
4235 is_cowonly_root(btrfs_ref_root(leaf, ref))) {
4236 ref_path->root_objectid = btrfs_ref_root(leaf, ref);
4237 ref_path->root_generation =
4238 btrfs_ref_generation(leaf, ref);
4239 if (level < 0) {
4240 /* special reference from the tree log */
4241 ref_path->nodes[0] = found_key.offset;
4242 ref_path->current_level = 0;
4243 }
4244 ret = 0;
4245 goto out;
4246 }
7d9eb12c 4247
1a40e23b
ZY		 4248 level++;
4249 BUG_ON(ref_path->nodes[level] != 0);
4250 ref_path->nodes[level] = found_key.offset;
4251 ref_path->current_level = level;
bf4ef679 4252
1a40e23b
ZY		 4253 /*
4254 * the reference was created in the running transaction,
4255 * no need to continue walking up.
4256 */
4257 if (btrfs_ref_generation(leaf, ref) == trans->transid) {
4258 ref_path->root_objectid = btrfs_ref_root(leaf, ref);
4259 ref_path->root_generation =
4260 btrfs_ref_generation(leaf, ref);
4261 ret = 0;
4262 goto out;
7d9eb12c
CM		 4263 }
4264
1a40e23b 4265 btrfs_release_path(extent_root, path);
d899e052 4266 cond_resched();
7d9eb12c 4267 }
1a40e23b
ZY		 4268 /* reached max tree level, but no tree root found. */
4269 BUG();
edbd8d4e 4270out:
1a40e23b
ZY		 4271 btrfs_free_path(path);
4272 return ret;
edbd8d4e
CM		 4273}
4274
1a40e23b
ZY		 4275static int btrfs_first_ref_path(struct btrfs_trans_handle *trans,
4276 struct btrfs_root *extent_root,
4277 struct btrfs_ref_path *ref_path,
4278 u64 extent_start)
a061fc8d 4279{
1a40e23b
ZY		 4280 memset(ref_path, 0, sizeof(*ref_path));
4281 ref_path->extent_start = extent_start;
a061fc8d 4282
1a40e23b 4283 return __next_ref_path(trans, extent_root, ref_path, 1);
a061fc8d
CM		 4284}
4285
1a40e23b
ZY		 4286static int btrfs_next_ref_path(struct btrfs_trans_handle *trans,
4287 struct btrfs_root *extent_root,
4288 struct btrfs_ref_path *ref_path)
edbd8d4e 4289{
1a40e23b
ZY		 4290 return __next_ref_path(trans, extent_root, ref_path, 0);
4291}
4292
4293static int noinline get_new_locations(struct inode *reloc_inode,
4294 struct btrfs_key *extent_key,
4295 u64 offset, int no_fragment,
4296 struct disk_extent **extents,
4297 int *nr_extents)
4298{
4299 struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
4300 struct btrfs_path *path;
4301 struct btrfs_file_extent_item *fi;
edbd8d4e 4302 struct extent_buffer *leaf;
1a40e23b
ZY		 4303 struct disk_extent *exts = *extents;
4304 struct btrfs_key found_key;
4305 u64 cur_pos;
4306 u64 last_byte;
edbd8d4e 4307 u32 nritems;
1a40e23b
ZY		 4308 int nr = 0;
4309 int max = *nr_extents;
4310 int ret;
edbd8d4e 4311
1a40e23b
ZY		 4312 WARN_ON(!no_fragment && *extents);
4313 if (!exts) {
4314 max = 1;
4315 exts = kmalloc(sizeof(*exts) * max, GFP_NOFS);
4316 if (!exts)
4317 return -ENOMEM;
a061fc8d 4318 }
edbd8d4e 4319
1a40e23b
ZY		 4320 path = btrfs_alloc_path();
4321 BUG_ON(!path);
edbd8d4e 4322
1a40e23b
ZY		 4323 cur_pos = extent_key->objectid - offset;
4324 last_byte = extent_key->objectid + extent_key->offset;
4325 ret = btrfs_lookup_file_extent(NULL, root, path, reloc_inode->i_ino,
4326 cur_pos, 0);
4327 if (ret < 0)
4328 goto out;
4329 if (ret > 0) {
4330 ret = -ENOENT;
4331 goto out;
4332 }
edbd8d4e 4333
1a40e23b 4334 while (1) {
edbd8d4e
CM		 4335 leaf = path->nodes[0];
4336 nritems = btrfs_header_nritems(leaf);
1a40e23b
ZY		 4337 if (path->slots[0] >= nritems) {
4338 ret = btrfs_next_leaf(root, path);
a061fc8d
CM		 4339 if (ret < 0)
4340 goto out;
1a40e23b
ZY		 4341 if (ret > 0)
4342 break;
bf4ef679 4343 leaf = path->nodes[0];
a061fc8d 4344 }
edbd8d4e
CM		 4345
4346 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1a40e23b
ZY		 4347 if (found_key.offset != cur_pos ||
4348 found_key.type != BTRFS_EXTENT_DATA_KEY ||
4349 found_key.objectid != reloc_inode->i_ino)
edbd8d4e
CM		 4350 break;
4351
1a40e23b
ZY		 4352 fi = btrfs_item_ptr(leaf, path->slots[0],
4353 struct btrfs_file_extent_item);
4354 if (btrfs_file_extent_type(leaf, fi) !=
4355 BTRFS_FILE_EXTENT_REG ||
4356 btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
edbd8d4e 4357 break;
1a40e23b
ZY		 4358
4359 if (nr == max) {
4360 struct disk_extent *old = exts;
4361 max *= 2;
4362 exts = kzalloc(sizeof(*exts) * max, GFP_NOFS);
4363 memcpy(exts, old, sizeof(*exts) * nr);
4364 if (old != *extents)
4365 kfree(old);
a061fc8d 4366 }
edbd8d4e 4367
1a40e23b
ZY		 4368 exts[nr].disk_bytenr =
4369 btrfs_file_extent_disk_bytenr(leaf, fi);
4370 exts[nr].disk_num_bytes =
4371 btrfs_file_extent_disk_num_bytes(leaf, fi);
4372 exts[nr].offset = btrfs_file_extent_offset(leaf, fi);
4373 exts[nr].num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
c8b97818
CM		 4374 exts[nr].ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
4375 exts[nr].compression = btrfs_file_extent_compression(leaf, fi);
4376 exts[nr].encryption = btrfs_file_extent_encryption(leaf, fi);
4377 exts[nr].other_encoding = btrfs_file_extent_other_encoding(leaf,
4378 fi);
d899e052
YZ		 4379 BUG_ON(exts[nr].offset > 0);
4380 BUG_ON(exts[nr].compression || exts[nr].encryption);
4381 BUG_ON(exts[nr].num_bytes != exts[nr].disk_num_bytes);
edbd8d4e 4382
1a40e23b
ZY		 4383 cur_pos += exts[nr].num_bytes;
4384 nr++;
4385
4386 if (cur_pos + offset >= last_byte)
4387 break;
4388
4389 if (no_fragment) {
4390 ret = 1;
edbd8d4e 4391 goto out;
1a40e23b
ZY		 4392 }
4393 path->slots[0]++;
4394 }
4395
4396 WARN_ON(cur_pos + offset > last_byte);
4397 if (cur_pos + offset < last_byte) {
4398 ret = -ENOENT;
4399 goto out;
edbd8d4e
CM		 4400 }
4401 ret = 0;
4402out:
1a40e23b
ZY		 4403 btrfs_free_path(path);
4404 if (ret) {
4405 if (exts != *extents)
4406 kfree(exts);
4407 } else {
4408 *extents = exts;
4409 *nr_extents = nr;
4410 }
4411 return ret;
4412}
4413
4414static int noinline replace_one_extent(struct btrfs_trans_handle *trans,
4415 struct btrfs_root *root,
4416 struct btrfs_path *path,
4417 struct btrfs_key *extent_key,
4418 struct btrfs_key *leaf_key,
4419 struct btrfs_ref_path *ref_path,
4420 struct disk_extent *new_extents,
4421 int nr_extents)
4422{
4423 struct extent_buffer *leaf;
4424 struct btrfs_file_extent_item *fi;
4425 struct inode *inode = NULL;
4426 struct btrfs_key key;
4427 u64 lock_start = 0;
4428 u64 lock_end = 0;
4429 u64 num_bytes;
4430 u64 ext_offset;
4431 u64 first_pos;
4432 u32 nritems;
3bb1a1bc 4433 int nr_scaned = 0;
1a40e23b 4434 int extent_locked = 0;
d899e052 4435 int extent_type;
1a40e23b
ZY		 4436 int ret;
4437
3bb1a1bc
YZ		 4438 memcpy(&key, leaf_key, sizeof(key));
4439 first_pos = INT_LIMIT(loff_t) - extent_key->offset;
1a40e23b 4440 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
3bb1a1bc
YZ		 4441 if (key.objectid < ref_path->owner_objectid ||
4442 (key.objectid == ref_path->owner_objectid &&
4443 key.type < BTRFS_EXTENT_DATA_KEY)) {
4444 key.objectid = ref_path->owner_objectid;
4445 key.type = BTRFS_EXTENT_DATA_KEY;
4446 key.offset = 0;
4447 }
1a40e23b
ZY		 4448 }
4449
4450 while (1) {
4451 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
4452 if (ret < 0)
4453 goto out;
4454
4455 leaf = path->nodes[0];
4456 nritems = btrfs_header_nritems(leaf);
4457next:
4458 if (extent_locked && ret > 0) {
4459 /*
4460 * the file extent item was modified by someone
4461 * before the extent got locked.
4462 */
1a40e23b
ZY		 4463 unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
4464 lock_end, GFP_NOFS);
4465 extent_locked = 0;
4466 }
4467
4468 if (path->slots[0] >= nritems) {
3bb1a1bc 4469 if (++nr_scaned > 2)
1a40e23b
ZY		 4470 break;
4471
4472 BUG_ON(extent_locked);
4473 ret = btrfs_next_leaf(root, path);
4474 if (ret < 0)
4475 goto out;
4476 if (ret > 0)
4477 break;
4478 leaf = path->nodes[0];
4479 nritems = btrfs_header_nritems(leaf);
4480 }
4481
4482 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4483
4484 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
4485 if ((key.objectid > ref_path->owner_objectid) ||
4486 (key.objectid == ref_path->owner_objectid &&
4487 key.type > BTRFS_EXTENT_DATA_KEY) ||
4488 (key.offset >= first_pos + extent_key->offset))
4489 break;
4490 }
4491
4492 if (inode && key.objectid != inode->i_ino) {
4493 BUG_ON(extent_locked);
4494 btrfs_release_path(root, path);
4495 mutex_unlock(&inode->i_mutex);
4496 iput(inode);
4497 inode = NULL;
4498 continue;
4499 }
4500
4501 if (key.type != BTRFS_EXTENT_DATA_KEY) {
4502 path->slots[0]++;
4503 ret = 1;
4504 goto next;
4505 }
4506 fi = btrfs_item_ptr(leaf, path->slots[0],
4507 struct btrfs_file_extent_item);
d899e052
YZ		 4508 extent_type = btrfs_file_extent_type(leaf, fi);
4509 if ((extent_type != BTRFS_FILE_EXTENT_REG &&
4510 extent_type != BTRFS_FILE_EXTENT_PREALLOC) ||
1a40e23b
ZY		 4511 (btrfs_file_extent_disk_bytenr(leaf, fi) !=
4512 extent_key->objectid)) {
4513 path->slots[0]++;
4514 ret = 1;
4515 goto next;
4516 }
4517
4518 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
4519 ext_offset = btrfs_file_extent_offset(leaf, fi);
4520
4521 if (first_pos > key.offset - ext_offset)
4522 first_pos = key.offset - ext_offset;
4523
4524 if (!extent_locked) {
4525 lock_start = key.offset;
4526 lock_end = lock_start + num_bytes - 1;
4527 } else {
6643558d
YZ		 4528 if (lock_start > key.offset ||
4529 lock_end + 1 < key.offset + num_bytes) {
4530 unlock_extent(&BTRFS_I(inode)->io_tree,
4531 lock_start, lock_end, GFP_NOFS);
4532 extent_locked = 0;
4533 }
1a40e23b
ZY		 4534 }
4535
4536 if (!inode) {
4537 btrfs_release_path(root, path);
4538
4539 inode = btrfs_iget_locked(root->fs_info->sb,
4540 key.objectid, root);
4541 if (inode->i_state & I_NEW) {
4542 BTRFS_I(inode)->root = root;
4543 BTRFS_I(inode)->location.objectid =
4544 key.objectid;
4545 BTRFS_I(inode)->location.type =
4546 BTRFS_INODE_ITEM_KEY;
4547 BTRFS_I(inode)->location.offset = 0;
4548 btrfs_read_locked_inode(inode);
4549 unlock_new_inode(inode);
4550 }
4551 /*
4552 * some code call btrfs_commit_transaction while
4553 * holding the i_mutex, so we can't use mutex_lock
4554 * here.
4555 */
4556 if (is_bad_inode(inode) ||
4557 !mutex_trylock(&inode->i_mutex)) {
4558 iput(inode);
4559 inode = NULL;
4560 key.offset = (u64)-1;
4561 goto skip;
4562 }
4563 }
4564
4565 if (!extent_locked) {
4566 struct btrfs_ordered_extent *ordered;
4567
4568 btrfs_release_path(root, path);
4569
4570 lock_extent(&BTRFS_I(inode)->io_tree, lock_start,
4571 lock_end, GFP_NOFS);
4572 ordered = btrfs_lookup_first_ordered_extent(inode,
4573 lock_end);
4574 if (ordered &&
4575 ordered->file_offset <= lock_end &&
4576 ordered->file_offset + ordered->len > lock_start) {
4577 unlock_extent(&BTRFS_I(inode)->io_tree,
4578 lock_start, lock_end, GFP_NOFS);
4579 btrfs_start_ordered_extent(inode, ordered, 1);
4580 btrfs_put_ordered_extent(ordered);
4581 key.offset += num_bytes;
4582 goto skip;
4583 }
4584 if (ordered)
4585 btrfs_put_ordered_extent(ordered);
4586
1a40e23b
ZY		 4587 extent_locked = 1;
4588 continue;
4589 }
4590
4591 if (nr_extents == 1) {
4592 /* update extent pointer in place */
1a40e23b
ZY		 4593 btrfs_set_file_extent_disk_bytenr(leaf, fi,
4594 new_extents[0].disk_bytenr);
4595 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
4596 new_extents[0].disk_num_bytes);
1a40e23b
ZY		 4597 btrfs_mark_buffer_dirty(leaf);
4598
4599 btrfs_drop_extent_cache(inode, key.offset,
4600 key.offset + num_bytes - 1, 0);
4601
4602 ret = btrfs_inc_extent_ref(trans, root,
4603 new_extents[0].disk_bytenr,
4604 new_extents[0].disk_num_bytes,
4605 leaf->start,
4606 root->root_key.objectid,
4607 trans->transid,
3bb1a1bc 4608 key.objectid);
1a40e23b
ZY		 4609 BUG_ON(ret);
4610
4611 ret = btrfs_free_extent(trans, root,
4612 extent_key->objectid,
4613 extent_key->offset,
4614 leaf->start,
4615 btrfs_header_owner(leaf),
4616 btrfs_header_generation(leaf),
3bb1a1bc 4617 key.objectid, 0);
1a40e23b
ZY		 4618 BUG_ON(ret);
4619
4620 btrfs_release_path(root, path);
4621 key.offset += num_bytes;
4622 } else {
d899e052
YZ		 4623 BUG_ON(1);
4624#if 0
1a40e23b
ZY		 4625 u64 alloc_hint;
4626 u64 extent_len;
4627 int i;
4628 /*
4629 * drop old extent pointer at first, then insert the
4630 * new pointers one bye one
4631 */
4632 btrfs_release_path(root, path);
4633 ret = btrfs_drop_extents(trans, root, inode, key.offset,
4634 key.offset + num_bytes,
4635 key.offset, &alloc_hint);
4636 BUG_ON(ret);
4637
4638 for (i = 0; i < nr_extents; i++) {
4639 if (ext_offset >= new_extents[i].num_bytes) {
4640 ext_offset -= new_extents[i].num_bytes;
4641 continue;
4642 }
4643 extent_len = min(new_extents[i].num_bytes -
4644 ext_offset, num_bytes);
4645
4646 ret = btrfs_insert_empty_item(trans, root,
4647 path, &key,
4648 sizeof(*fi));
4649 BUG_ON(ret);
4650
4651 leaf = path->nodes[0];
4652 fi = btrfs_item_ptr(leaf, path->slots[0],
4653 struct btrfs_file_extent_item);
4654 btrfs_set_file_extent_generation(leaf, fi,
4655 trans->transid);
4656 btrfs_set_file_extent_type(leaf, fi,
4657 BTRFS_FILE_EXTENT_REG);
4658 btrfs_set_file_extent_disk_bytenr(leaf, fi,
4659 new_extents[i].disk_bytenr);
4660 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
4661 new_extents[i].disk_num_bytes);
c8b97818
CM		 4662 btrfs_set_file_extent_ram_bytes(leaf, fi,
4663 new_extents[i].ram_bytes);
4664
4665 btrfs_set_file_extent_compression(leaf, fi,
4666 new_extents[i].compression);
4667 btrfs_set_file_extent_encryption(leaf, fi,
4668 new_extents[i].encryption);
4669 btrfs_set_file_extent_other_encoding(leaf, fi,
4670 new_extents[i].other_encoding);
4671
1a40e23b
ZY		 4672 btrfs_set_file_extent_num_bytes(leaf, fi,
4673 extent_len);
4674 ext_offset += new_extents[i].offset;
4675 btrfs_set_file_extent_offset(leaf, fi,
4676 ext_offset);
4677 btrfs_mark_buffer_dirty(leaf);
4678
4679 btrfs_drop_extent_cache(inode, key.offset,
4680 key.offset + extent_len - 1, 0);
4681
4682 ret = btrfs_inc_extent_ref(trans, root,
4683 new_extents[i].disk_bytenr,
4684 new_extents[i].disk_num_bytes,
4685 leaf->start,
4686 root->root_key.objectid,
3bb1a1bc 4687 trans->transid, key.objectid);
1a40e23b
ZY		 4688 BUG_ON(ret);
4689 btrfs_release_path(root, path);
4690
a76a3cd4 4691 inode_add_bytes(inode, extent_len);
1a40e23b
ZY		 4692
4693 ext_offset = 0;
4694 num_bytes -= extent_len;
4695 key.offset += extent_len;
4696
4697 if (num_bytes == 0)
4698 break;
4699 }
4700 BUG_ON(i >= nr_extents);
d899e052 4701#endif
1a40e23b
ZY		 4702 }
4703
4704 if (extent_locked) {
1a40e23b
ZY		 4705 unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
4706 lock_end, GFP_NOFS);
4707 extent_locked = 0;
4708 }
4709skip:
4710 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS &&
4711 key.offset >= first_pos + extent_key->offset)
4712 break;
4713
4714 cond_resched();
4715 }
4716 ret = 0;
4717out:
4718 btrfs_release_path(root, path);
4719 if (inode) {
4720 mutex_unlock(&inode->i_mutex);
4721 if (extent_locked) {
1a40e23b
ZY		 4722 unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
4723 lock_end, GFP_NOFS);
4724 }
4725 iput(inode);
4726 }
4727 return ret;
4728}
4729
1a40e23b
ZY		 4730int btrfs_reloc_tree_cache_ref(struct btrfs_trans_handle *trans,
4731 struct btrfs_root *root,
4732 struct extent_buffer *buf, u64 orig_start)
4733{
4734 int level;
4735 int ret;
4736
4737 BUG_ON(btrfs_header_generation(buf) != trans->transid);
4738 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
4739
4740 level = btrfs_header_level(buf);
4741 if (level == 0) {
4742 struct btrfs_leaf_ref *ref;
4743 struct btrfs_leaf_ref *orig_ref;
4744
4745 orig_ref = btrfs_lookup_leaf_ref(root, orig_start);
4746 if (!orig_ref)
4747 return -ENOENT;
4748
4749 ref = btrfs_alloc_leaf_ref(root, orig_ref->nritems);
4750 if (!ref) {
4751 btrfs_free_leaf_ref(root, orig_ref);
4752 return -ENOMEM;
4753 }
4754
4755 ref->nritems = orig_ref->nritems;
4756 memcpy(ref->extents, orig_ref->extents,
4757 sizeof(ref->extents[0]) * ref->nritems);
4758
4759 btrfs_free_leaf_ref(root, orig_ref);
4760
4761 ref->root_gen = trans->transid;
4762 ref->bytenr = buf->start;
4763 ref->owner = btrfs_header_owner(buf);
4764 ref->generation = btrfs_header_generation(buf);
4765 ret = btrfs_add_leaf_ref(root, ref, 0);
4766 WARN_ON(ret);
4767 btrfs_free_leaf_ref(root, ref);
4768 }
4769 return 0;
4770}
4771
4772static int noinline invalidate_extent_cache(struct btrfs_root *root,
4773 struct extent_buffer *leaf,
4774 struct btrfs_block_group_cache *group,
4775 struct btrfs_root *target_root)
4776{
4777 struct btrfs_key key;
4778 struct inode *inode = NULL;
4779 struct btrfs_file_extent_item *fi;
4780 u64 num_bytes;
4781 u64 skip_objectid = 0;
4782 u32 nritems;
4783 u32 i;
4784
4785 nritems = btrfs_header_nritems(leaf);
4786 for (i = 0; i < nritems; i++) {
4787 btrfs_item_key_to_cpu(leaf, &key, i);
4788 if (key.objectid == skip_objectid ||
4789 key.type != BTRFS_EXTENT_DATA_KEY)
4790 continue;
4791 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
4792 if (btrfs_file_extent_type(leaf, fi) ==
4793 BTRFS_FILE_EXTENT_INLINE)
4794 continue;
4795 if (btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
4796 continue;
4797 if (!inode || inode->i_ino != key.objectid) {
4798 iput(inode);
4799 inode = btrfs_ilookup(target_root->fs_info->sb,
4800 key.objectid, target_root, 1);
4801 }
4802 if (!inode) {
4803 skip_objectid = key.objectid;
4804 continue;
4805 }
4806 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
4807
4808 lock_extent(&BTRFS_I(inode)->io_tree, key.offset,
4809 key.offset + num_bytes - 1, GFP_NOFS);
1a40e23b
ZY		 4810 btrfs_drop_extent_cache(inode, key.offset,
4811 key.offset + num_bytes - 1, 1);
1a40e23b
ZY		 4812 unlock_extent(&BTRFS_I(inode)->io_tree, key.offset,
4813 key.offset + num_bytes - 1, GFP_NOFS);
4814 cond_resched();
4815 }
4816 iput(inode);
4817 return 0;
4818}
4819
4820static int noinline replace_extents_in_leaf(struct btrfs_trans_handle *trans,
4821 struct btrfs_root *root,
4822 struct extent_buffer *leaf,
4823 struct btrfs_block_group_cache *group,
4824 struct inode *reloc_inode)
4825{
4826 struct btrfs_key key;
4827 struct btrfs_key extent_key;
4828 struct btrfs_file_extent_item *fi;
4829 struct btrfs_leaf_ref *ref;
4830 struct disk_extent *new_extent;
4831 u64 bytenr;
4832 u64 num_bytes;
4833 u32 nritems;
4834 u32 i;
4835 int ext_index;
4836 int nr_extent;
4837 int ret;
4838
4839 new_extent = kmalloc(sizeof(*new_extent), GFP_NOFS);
4840 BUG_ON(!new_extent);
4841
4842 ref = btrfs_lookup_leaf_ref(root, leaf->start);
4843 BUG_ON(!ref);
4844
4845 ext_index = -1;
4846 nritems = btrfs_header_nritems(leaf);
4847 for (i = 0; i < nritems; i++) {
4848 btrfs_item_key_to_cpu(leaf, &key, i);
4849 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
4850 continue;
4851 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
4852 if (btrfs_file_extent_type(leaf, fi) ==
4853 BTRFS_FILE_EXTENT_INLINE)
4854 continue;
4855 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
4856 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
4857 if (bytenr == 0)
4858 continue;
4859
4860 ext_index++;
4861 if (bytenr >= group->key.objectid + group->key.offset ||
4862 bytenr + num_bytes <= group->key.objectid)
4863 continue;
4864
4865 extent_key.objectid = bytenr;
4866 extent_key.offset = num_bytes;
4867 extent_key.type = BTRFS_EXTENT_ITEM_KEY;
4868 nr_extent = 1;
4869 ret = get_new_locations(reloc_inode, &extent_key,
4870 group->key.objectid, 1,
4871 &new_extent, &nr_extent);
4872 if (ret > 0)
4873 continue;
4874 BUG_ON(ret < 0);
4875
4876 BUG_ON(ref->extents[ext_index].bytenr != bytenr);
4877 BUG_ON(ref->extents[ext_index].num_bytes != num_bytes);
4878 ref->extents[ext_index].bytenr = new_extent->disk_bytenr;
4879 ref->extents[ext_index].num_bytes = new_extent->disk_num_bytes;
4880
1a40e23b
ZY		 4881 btrfs_set_file_extent_disk_bytenr(leaf, fi,
4882 new_extent->disk_bytenr);
4883 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
4884 new_extent->disk_num_bytes);
1a40e23b
ZY		 4885 btrfs_mark_buffer_dirty(leaf);
4886
4887 ret = btrfs_inc_extent_ref(trans, root,
4888 new_extent->disk_bytenr,
4889 new_extent->disk_num_bytes,
4890 leaf->start,
4891 root->root_key.objectid,
3bb1a1bc 4892 trans->transid, key.objectid);
1a40e23b
ZY		 4893 BUG_ON(ret);
4894 ret = btrfs_free_extent(trans, root,
4895 bytenr, num_bytes, leaf->start,
4896 btrfs_header_owner(leaf),
4897 btrfs_header_generation(leaf),
3bb1a1bc 4898 key.objectid, 0);
1a40e23b
ZY		 4899 BUG_ON(ret);
4900 cond_resched();
4901 }
4902 kfree(new_extent);
4903 BUG_ON(ext_index + 1 != ref->nritems);
4904 btrfs_free_leaf_ref(root, ref);
4905 return 0;
4906}
4907
f82d02d9
YZ		 4908int btrfs_free_reloc_root(struct btrfs_trans_handle *trans,
4909 struct btrfs_root *root)
1a40e23b
ZY		 4910{
4911 struct btrfs_root *reloc_root;
f82d02d9 4912 int ret;
1a40e23b
ZY		 4913
4914 if (root->reloc_root) {
4915 reloc_root = root->reloc_root;
4916 root->reloc_root = NULL;
4917 list_add(&reloc_root->dead_list,
4918 &root->fs_info->dead_reloc_roots);
f82d02d9
YZ		 4919
4920 btrfs_set_root_bytenr(&reloc_root->root_item,
4921 reloc_root->node->start);
4922 btrfs_set_root_level(&root->root_item,
4923 btrfs_header_level(reloc_root->node));
4924 memset(&reloc_root->root_item.drop_progress, 0,
4925 sizeof(struct btrfs_disk_key));
4926 reloc_root->root_item.drop_level = 0;
4927
4928 ret = btrfs_update_root(trans, root->fs_info->tree_root,
4929 &reloc_root->root_key,
4930 &reloc_root->root_item);
4931 BUG_ON(ret);
1a40e23b
ZY		 4932 }
4933 return 0;
4934}
4935
4936int btrfs_drop_dead_reloc_roots(struct btrfs_root *root)
4937{
4938 struct btrfs_trans_handle *trans;
4939 struct btrfs_root *reloc_root;
4940 struct btrfs_root *prev_root = NULL;
4941 struct list_head dead_roots;
4942 int ret;
4943 unsigned long nr;
4944
4945 INIT_LIST_HEAD(&dead_roots);
4946 list_splice_init(&root->fs_info->dead_reloc_roots, &dead_roots);
4947
4948 while (!list_empty(&dead_roots)) {
4949 reloc_root = list_entry(dead_roots.prev,
4950 struct btrfs_root, dead_list);
4951 list_del_init(&reloc_root->dead_list);
4952
4953 BUG_ON(reloc_root->commit_root != NULL);
4954 while (1) {
4955 trans = btrfs_join_transaction(root, 1);
4956 BUG_ON(!trans);
4957
4958 mutex_lock(&root->fs_info->drop_mutex);
4959 ret = btrfs_drop_snapshot(trans, reloc_root);
4960 if (ret != -EAGAIN)
4961 break;
4962 mutex_unlock(&root->fs_info->drop_mutex);
4963
4964 nr = trans->blocks_used;
4965 ret = btrfs_end_transaction(trans, root);
4966 BUG_ON(ret);
4967 btrfs_btree_balance_dirty(root, nr);
4968 }
4969
4970 free_extent_buffer(reloc_root->node);
4971
4972 ret = btrfs_del_root(trans, root->fs_info->tree_root,
4973 &reloc_root->root_key);
4974 BUG_ON(ret);
4975 mutex_unlock(&root->fs_info->drop_mutex);
4976
4977 nr = trans->blocks_used;
4978 ret = btrfs_end_transaction(trans, root);
4979 BUG_ON(ret);
4980 btrfs_btree_balance_dirty(root, nr);
4981
4982 kfree(prev_root);
4983 prev_root = reloc_root;
4984 }
4985 if (prev_root) {
4986 btrfs_remove_leaf_refs(prev_root, (u64)-1, 0);
4987 kfree(prev_root);
4988 }
4989 return 0;
4990}
4991
4992int btrfs_add_dead_reloc_root(struct btrfs_root *root)
4993{
4994 list_add(&root->dead_list, &root->fs_info->dead_reloc_roots);
4995 return 0;
4996}
4997
4998int btrfs_cleanup_reloc_trees(struct btrfs_root *root)
4999{
5000 struct btrfs_root *reloc_root;
5001 struct btrfs_trans_handle *trans;
5002 struct btrfs_key location;
5003 int found;
5004 int ret;
5005
5006 mutex_lock(&root->fs_info->tree_reloc_mutex);
5007 ret = btrfs_find_dead_roots(root, BTRFS_TREE_RELOC_OBJECTID, NULL);
5008 BUG_ON(ret);
5009 found = !list_empty(&root->fs_info->dead_reloc_roots);
5010 mutex_unlock(&root->fs_info->tree_reloc_mutex);
5011
5012 if (found) {
5013 trans = btrfs_start_transaction(root, 1);
5014 BUG_ON(!trans);
5015 ret = btrfs_commit_transaction(trans, root);
5016 BUG_ON(ret);
5017 }
5018
5019 location.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
5020 location.offset = (u64)-1;
5021 location.type = BTRFS_ROOT_ITEM_KEY;
5022
5023 reloc_root = btrfs_read_fs_root_no_name(root->fs_info, &location);
5024 BUG_ON(!reloc_root);
5025 btrfs_orphan_cleanup(reloc_root);
5026 return 0;
5027}
5028
5029static int noinline init_reloc_tree(struct btrfs_trans_handle *trans,
5030 struct btrfs_root *root)
5031{
5032 struct btrfs_root *reloc_root;
5033 struct extent_buffer *eb;
5034 struct btrfs_root_item *root_item;
5035 struct btrfs_key root_key;
5036 int ret;
5037
5038 BUG_ON(!root->ref_cows);
5039 if (root->reloc_root)
5040 return 0;
5041
5042 root_item = kmalloc(sizeof(*root_item), GFP_NOFS);
5043 BUG_ON(!root_item);
5044
5045 ret = btrfs_copy_root(trans, root, root->commit_root,
5046 &eb, BTRFS_TREE_RELOC_OBJECTID);
5047 BUG_ON(ret);
5048
5049 root_key.objectid = BTRFS_TREE_RELOC_OBJECTID;
5050 root_key.offset = root->root_key.objectid;
5051 root_key.type = BTRFS_ROOT_ITEM_KEY;
5052
5053 memcpy(root_item, &root->root_item, sizeof(root_item));
5054 btrfs_set_root_refs(root_item, 0);
5055 btrfs_set_root_bytenr(root_item, eb->start);
5056 btrfs_set_root_level(root_item, btrfs_header_level(eb));
84234f3a 5057 btrfs_set_root_generation(root_item, trans->transid);
1a40e23b
ZY		 5058
5059 btrfs_tree_unlock(eb);
5060 free_extent_buffer(eb);
5061
5062 ret = btrfs_insert_root(trans, root->fs_info->tree_root,
5063 &root_key, root_item);
5064 BUG_ON(ret);
5065 kfree(root_item);
5066
5067 reloc_root = btrfs_read_fs_root_no_radix(root->fs_info->tree_root,
5068 &root_key);
5069 BUG_ON(!reloc_root);
5070 reloc_root->last_trans = trans->transid;
5071 reloc_root->commit_root = NULL;
5072 reloc_root->ref_tree = &root->fs_info->reloc_ref_tree;
5073
5074 root->reloc_root = reloc_root;
5075 return 0;
5076}
5077
5078/*
5079 * Core function of space balance.
5080 *
5081 * The idea is using reloc trees to relocate tree blocks in reference
f82d02d9
YZ		 5082 * counted roots. There is one reloc tree for each subvol, and all
5083 * reloc trees share same root key objectid. Reloc trees are snapshots
5084 * of the latest committed roots of subvols (root->commit_root).
5085 *
5086 * To relocate a tree block referenced by a subvol, there are two steps.
5087 * COW the block through subvol's reloc tree, then update block pointer
5088 * in the subvol to point to the new block. Since all reloc trees share
5089 * same root key objectid, doing special handing for tree blocks owned
5090 * by them is easy. Once a tree block has been COWed in one reloc tree,
5091 * we can use the resulting new block directly when the same block is
5092 * required to COW again through other reloc trees. By this way, relocated
5093 * tree blocks are shared between reloc trees, so they are also shared
5094 * between subvols.
1a40e23b
ZY		 5095 */
5096static int noinline relocate_one_path(struct btrfs_trans_handle *trans,
5097 struct btrfs_root *root,
5098 struct btrfs_path *path,
5099 struct btrfs_key *first_key,
5100 struct btrfs_ref_path *ref_path,
5101 struct btrfs_block_group_cache *group,
5102 struct inode *reloc_inode)
5103{
5104 struct btrfs_root *reloc_root;
5105 struct extent_buffer *eb = NULL;
5106 struct btrfs_key *keys;
5107 u64 *nodes;
5108 int level;
f82d02d9 5109 int shared_level;
1a40e23b 5110 int lowest_level = 0;
1a40e23b
ZY		 5111 int ret;
5112
5113 if (ref_path->owner_objectid < BTRFS_FIRST_FREE_OBJECTID)
5114 lowest_level = ref_path->owner_objectid;
5115
f82d02d9 5116 if (!root->ref_cows) {
1a40e23b
ZY		 5117 path->lowest_level = lowest_level;
5118 ret = btrfs_search_slot(trans, root, first_key, path, 0, 1);
5119 BUG_ON(ret < 0);
5120 path->lowest_level = 0;
5121 btrfs_release_path(root, path);
5122 return 0;
5123 }
5124
1a40e23b
ZY		 5125 mutex_lock(&root->fs_info->tree_reloc_mutex);
5126 ret = init_reloc_tree(trans, root);
5127 BUG_ON(ret);
5128 reloc_root = root->reloc_root;
5129
f82d02d9
YZ		 5130 shared_level = ref_path->shared_level;
5131 ref_path->shared_level = BTRFS_MAX_LEVEL - 1;
1a40e23b 5132
f82d02d9
YZ		 5133 keys = ref_path->node_keys;
5134 nodes = ref_path->new_nodes;
5135 memset(&keys[shared_level + 1], 0,
5136 sizeof(*keys) * (BTRFS_MAX_LEVEL - shared_level - 1));
5137 memset(&nodes[shared_level + 1], 0,
5138 sizeof(*nodes) * (BTRFS_MAX_LEVEL - shared_level - 1));
1a40e23b 5139
f82d02d9
YZ		 5140 if (nodes[lowest_level] == 0) {
5141 path->lowest_level = lowest_level;
5142 ret = btrfs_search_slot(trans, reloc_root, first_key, path,
5143 0, 1);
5144 BUG_ON(ret);
5145 for (level = lowest_level; level < BTRFS_MAX_LEVEL; level++) {
5146 eb = path->nodes[level];
5147 if (!eb || eb == reloc_root->node)
5148 break;
5149 nodes[level] = eb->start;
5150 if (level == 0)
5151 btrfs_item_key_to_cpu(eb, &keys[level], 0);
5152 else
5153 btrfs_node_key_to_cpu(eb, &keys[level], 0);
5154 }
2b82032c
YZ		 5155 if (nodes[0] &&
5156 ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
f82d02d9
YZ		 5157 eb = path->nodes[0];
5158 ret = replace_extents_in_leaf(trans, reloc_root, eb,
5159 group, reloc_inode);
5160 BUG_ON(ret);
5161 }
5162 btrfs_release_path(reloc_root, path);
5163 } else {
1a40e23b 5164 ret = btrfs_merge_path(trans, reloc_root, keys, nodes,
f82d02d9 5165 lowest_level);
1a40e23b
ZY		 5166 BUG_ON(ret);
5167 }
5168
1a40e23b
ZY		 5169 /*
5170 * replace tree blocks in the fs tree with tree blocks in
5171 * the reloc tree.
5172 */
5173 ret = btrfs_merge_path(trans, root, keys, nodes, lowest_level);
5174 BUG_ON(ret < 0);
5175
5176 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
f82d02d9
YZ		 5177 ret = btrfs_search_slot(trans, reloc_root, first_key, path,
5178 0, 0);
5179 BUG_ON(ret);
5180 extent_buffer_get(path->nodes[0]);
5181 eb = path->nodes[0];
5182 btrfs_release_path(reloc_root, path);
1a40e23b
ZY		 5183 ret = invalidate_extent_cache(reloc_root, eb, group, root);
5184 BUG_ON(ret);
5185 free_extent_buffer(eb);
5186 }
1a40e23b 5187
f82d02d9 5188 mutex_unlock(&root->fs_info->tree_reloc_mutex);
1a40e23b 5189 path->lowest_level = 0;
1a40e23b
ZY		 5190 return 0;
5191}
5192
5193static int noinline relocate_tree_block(struct btrfs_trans_handle *trans,
5194 struct btrfs_root *root,
5195 struct btrfs_path *path,
5196 struct btrfs_key *first_key,
5197 struct btrfs_ref_path *ref_path)
5198{
5199 int ret;
1a40e23b
ZY		 5200
5201 ret = relocate_one_path(trans, root, path, first_key,
5202 ref_path, NULL, NULL);
5203 BUG_ON(ret);
5204
5205 if (root == root->fs_info->extent_root)
5206 btrfs_extent_post_op(trans, root);
1a40e23b
ZY		 5207
5208 return 0;
5209}
5210
5211static int noinline del_extent_zero(struct btrfs_trans_handle *trans,
5212 struct btrfs_root *extent_root,
5213 struct btrfs_path *path,
5214 struct btrfs_key *extent_key)
5215{
5216 int ret;
5217
1a40e23b
ZY		 5218 ret = btrfs_search_slot(trans, extent_root, extent_key, path, -1, 1);
5219 if (ret)
5220 goto out;
5221 ret = btrfs_del_item(trans, extent_root, path);
5222out:
5223 btrfs_release_path(extent_root, path);
1a40e23b
ZY		 5224 return ret;
5225}
5226
5227static struct btrfs_root noinline *read_ref_root(struct btrfs_fs_info *fs_info,
5228 struct btrfs_ref_path *ref_path)
5229{
5230 struct btrfs_key root_key;
5231
5232 root_key.objectid = ref_path->root_objectid;
5233 root_key.type = BTRFS_ROOT_ITEM_KEY;
5234 if (is_cowonly_root(ref_path->root_objectid))
5235 root_key.offset = 0;
5236 else
5237 root_key.offset = (u64)-1;
5238
5239 return btrfs_read_fs_root_no_name(fs_info, &root_key);
5240}
5241
5242static int noinline relocate_one_extent(struct btrfs_root *extent_root,
5243 struct btrfs_path *path,
5244 struct btrfs_key *extent_key,
5245 struct btrfs_block_group_cache *group,
5246 struct inode *reloc_inode, int pass)
5247{
5248 struct btrfs_trans_handle *trans;
5249 struct btrfs_root *found_root;
5250 struct btrfs_ref_path *ref_path = NULL;
5251 struct disk_extent *new_extents = NULL;
5252 int nr_extents = 0;
5253 int loops;
5254 int ret;
5255 int level;
5256 struct btrfs_key first_key;
5257 u64 prev_block = 0;
5258
1a40e23b
ZY		 5259
5260 trans = btrfs_start_transaction(extent_root, 1);
5261 BUG_ON(!trans);
5262
5263 if (extent_key->objectid == 0) {
5264 ret = del_extent_zero(trans, extent_root, path, extent_key);
5265 goto out;
5266 }
5267
5268 ref_path = kmalloc(sizeof(*ref_path), GFP_NOFS);
5269 if (!ref_path) {
5270 ret = -ENOMEM;
5271 goto out;
5272 }
5273
5274 for (loops = 0; ; loops++) {
5275 if (loops == 0) {
5276 ret = btrfs_first_ref_path(trans, extent_root, ref_path,
5277 extent_key->objectid);
5278 } else {
5279 ret = btrfs_next_ref_path(trans, extent_root, ref_path);
5280 }
5281 if (ret < 0)
5282 goto out;
5283 if (ret > 0)
5284 break;
5285
5286 if (ref_path->root_objectid == BTRFS_TREE_LOG_OBJECTID ||
5287 ref_path->root_objectid == BTRFS_TREE_RELOC_OBJECTID)
5288 continue;
5289
5290 found_root = read_ref_root(extent_root->fs_info, ref_path);
5291 BUG_ON(!found_root);
5292 /*
5293 * for reference counted tree, only process reference paths
5294 * rooted at the latest committed root.
5295 */
5296 if (found_root->ref_cows &&
5297 ref_path->root_generation != found_root->root_key.offset)
5298 continue;
5299
5300 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
5301 if (pass == 0) {
5302 /*
5303 * copy data extents to new locations
5304 */
5305 u64 group_start = group->key.objectid;
5306 ret = relocate_data_extent(reloc_inode,
5307 extent_key,
5308 group_start);
5309 if (ret < 0)
5310 goto out;
5311 break;
5312 }
5313 level = 0;
5314 } else {
5315 level = ref_path->owner_objectid;
5316 }
5317
5318 if (prev_block != ref_path->nodes[level]) {
5319 struct extent_buffer *eb;
5320 u64 block_start = ref_path->nodes[level];
5321 u64 block_size = btrfs_level_size(found_root, level);
5322
5323 eb = read_tree_block(found_root, block_start,
5324 block_size, 0);
5325 btrfs_tree_lock(eb);
5326 BUG_ON(level != btrfs_header_level(eb));
5327
5328 if (level == 0)
5329 btrfs_item_key_to_cpu(eb, &first_key, 0);
5330 else
5331 btrfs_node_key_to_cpu(eb, &first_key, 0);
5332
5333 btrfs_tree_unlock(eb);
5334 free_extent_buffer(eb);
5335 prev_block = block_start;
5336 }
5337
5338 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID &&
5339 pass >= 2) {
5340 /*
5341 * use fallback method to process the remaining
5342 * references.
5343 */
5344 if (!new_extents) {
5345 u64 group_start = group->key.objectid;
d899e052
YZ		 5346 new_extents = kmalloc(sizeof(*new_extents),
5347 GFP_NOFS);
5348 nr_extents = 1;
1a40e23b
ZY		 5349 ret = get_new_locations(reloc_inode,
5350 extent_key,
d899e052 5351 group_start, 1,
1a40e23b
ZY		 5352 &new_extents,
5353 &nr_extents);
d899e052 5354 if (ret)
1a40e23b
ZY		 5355 goto out;
5356 }
5357 btrfs_record_root_in_trans(found_root);
5358 ret = replace_one_extent(trans, found_root,
5359 path, extent_key,
5360 &first_key, ref_path,
5361 new_extents, nr_extents);
5362 if (ret < 0)
5363 goto out;
5364 continue;
5365 }
5366
5367 btrfs_record_root_in_trans(found_root);
5368 if (ref_path->owner_objectid < BTRFS_FIRST_FREE_OBJECTID) {
5369 ret = relocate_tree_block(trans, found_root, path,
5370 &first_key, ref_path);
5371 } else {
5372 /*
5373 * try to update data extent references while
5374 * keeping metadata shared between snapshots.
5375 */
5376 ret = relocate_one_path(trans, found_root, path,
5377 &first_key, ref_path,
5378 group, reloc_inode);
5379 }
5380 if (ret < 0)
5381 goto out;
5382 }
5383 ret = 0;
5384out:
5385 btrfs_end_transaction(trans, extent_root);
5386 kfree(new_extents);
5387 kfree(ref_path);
1a40e23b
ZY		 5388 return ret;
5389}
5390
ec44a35c
CM		 5391static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
5392{
5393 u64 num_devices;
5394 u64 stripped = BTRFS_BLOCK_GROUP_RAID0 |
5395 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
5396
2b82032c 5397 num_devices = root->fs_info->fs_devices->rw_devices;
ec44a35c
CM		 5398 if (num_devices == 1) {
5399 stripped |= BTRFS_BLOCK_GROUP_DUP;
5400 stripped = flags & ~stripped;
5401
5402 /* turn raid0 into single device chunks */
5403 if (flags & BTRFS_BLOCK_GROUP_RAID0)
5404 return stripped;
5405
5406 /* turn mirroring into duplication */
5407 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
5408 BTRFS_BLOCK_GROUP_RAID10))
5409 return stripped | BTRFS_BLOCK_GROUP_DUP;
5410 return flags;
5411 } else {
5412 /* they already had raid on here, just return */
ec44a35c
CM		 5413 if (flags & stripped)
5414 return flags;
5415
5416 stripped |= BTRFS_BLOCK_GROUP_DUP;
5417 stripped = flags & ~stripped;
5418
5419 /* switch duplicated blocks with raid1 */
5420 if (flags & BTRFS_BLOCK_GROUP_DUP)
5421 return stripped | BTRFS_BLOCK_GROUP_RAID1;
5422
5423 /* turn single device chunks into raid0 */
5424 return stripped | BTRFS_BLOCK_GROUP_RAID0;
5425 }
5426 return flags;
5427}
5428
b2950863 5429static int __alloc_chunk_for_shrink(struct btrfs_root *root,
0ef3e66b
CM		 5430 struct btrfs_block_group_cache *shrink_block_group,
5431 int force)
5432{
5433 struct btrfs_trans_handle *trans;
5434 u64 new_alloc_flags;
5435 u64 calc;
5436
c286ac48 5437 spin_lock(&shrink_block_group->lock);
0ef3e66b 5438 if (btrfs_block_group_used(&shrink_block_group->item) > 0) {
c286ac48 5439 spin_unlock(&shrink_block_group->lock);
c286ac48 5440
0ef3e66b 5441 trans = btrfs_start_transaction(root, 1);
c286ac48 5442 spin_lock(&shrink_block_group->lock);
7d9eb12c 5443
0ef3e66b
CM		 5444 new_alloc_flags = update_block_group_flags(root,
5445 shrink_block_group->flags);
5446 if (new_alloc_flags != shrink_block_group->flags) {
5447 calc =
5448 btrfs_block_group_used(&shrink_block_group->item);
5449 } else {
5450 calc = shrink_block_group->key.offset;
5451 }
c286ac48
CM		 5452 spin_unlock(&shrink_block_group->lock);
5453
0ef3e66b
CM		 5454 do_chunk_alloc(trans, root->fs_info->extent_root,
5455 calc + 2 * 1024 * 1024, new_alloc_flags, force);
7d9eb12c 5456
0ef3e66b 5457 btrfs_end_transaction(trans, root);
c286ac48
CM		 5458 } else
5459 spin_unlock(&shrink_block_group->lock);
0ef3e66b
CM		 5460 return 0;
5461}
5462
1a40e23b
ZY		 5463static int __insert_orphan_inode(struct btrfs_trans_handle *trans,
5464 struct btrfs_root *root,
5465 u64 objectid, u64 size)
5466{
5467 struct btrfs_path *path;
5468 struct btrfs_inode_item *item;
5469 struct extent_buffer *leaf;
5470 int ret;
5471
5472 path = btrfs_alloc_path();
5473 if (!path)
5474 return -ENOMEM;
5475
5476 ret = btrfs_insert_empty_inode(trans, root, path, objectid);
5477 if (ret)
5478 goto out;
5479
5480 leaf = path->nodes[0];
5481 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_inode_item);
5482 memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item));
5483 btrfs_set_inode_generation(leaf, item, 1);
5484 btrfs_set_inode_size(leaf, item, size);
5485 btrfs_set_inode_mode(leaf, item, S_IFREG | 0600);
0403e47e 5486 btrfs_set_inode_flags(leaf, item, BTRFS_INODE_NOCOMPRESS);
1a40e23b
ZY		 5487 btrfs_mark_buffer_dirty(leaf);
5488 btrfs_release_path(root, path);
5489out:
5490 btrfs_free_path(path);
5491 return ret;
5492}
5493
5494static struct inode noinline *create_reloc_inode(struct btrfs_fs_info *fs_info,
5495 struct btrfs_block_group_cache *group)
5496{
5497 struct inode *inode = NULL;
5498 struct btrfs_trans_handle *trans;
5499 struct btrfs_root *root;
5500 struct btrfs_key root_key;
5501 u64 objectid = BTRFS_FIRST_FREE_OBJECTID;
5502 int err = 0;
5503
5504 root_key.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
5505 root_key.type = BTRFS_ROOT_ITEM_KEY;
5506 root_key.offset = (u64)-1;
5507 root = btrfs_read_fs_root_no_name(fs_info, &root_key);
5508 if (IS_ERR(root))
5509 return ERR_CAST(root);
5510
5511 trans = btrfs_start_transaction(root, 1);
5512 BUG_ON(!trans);
5513
5514 err = btrfs_find_free_objectid(trans, root, objectid, &objectid);
5515 if (err)
5516 goto out;
5517
5518 err = __insert_orphan_inode(trans, root, objectid, group->key.offset);
5519 BUG_ON(err);
5520
5521 err = btrfs_insert_file_extent(trans, root, objectid, 0, 0, 0,
c8b97818
CM		 5522 group->key.offset, 0, group->key.offset,
5523 0, 0, 0);
1a40e23b
ZY		 5524 BUG_ON(err);
5525
5526 inode = btrfs_iget_locked(root->fs_info->sb, objectid, root);
5527 if (inode->i_state & I_NEW) {
5528 BTRFS_I(inode)->root = root;
5529 BTRFS_I(inode)->location.objectid = objectid;
5530 BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
5531 BTRFS_I(inode)->location.offset = 0;
5532 btrfs_read_locked_inode(inode);
5533 unlock_new_inode(inode);
5534 BUG_ON(is_bad_inode(inode));
5535 } else {
5536 BUG_ON(1);
5537 }
5538
5539 err = btrfs_orphan_add(trans, inode);
5540out:
5541 btrfs_end_transaction(trans, root);
5542 if (err) {
5543 if (inode)
5544 iput(inode);
5545 inode = ERR_PTR(err);
5546 }
5547 return inode;
5548}
5549
5550int btrfs_relocate_block_group(struct btrfs_root *root, u64 group_start)
edbd8d4e
CM		 5551{
5552 struct btrfs_trans_handle *trans;
edbd8d4e 5553 struct btrfs_path *path;
1a40e23b
ZY		 5554 struct btrfs_fs_info *info = root->fs_info;
5555 struct extent_buffer *leaf;
5556 struct inode *reloc_inode;
5557 struct btrfs_block_group_cache *block_group;
5558 struct btrfs_key key;
d899e052 5559 u64 skipped;
edbd8d4e
CM		 5560 u64 cur_byte;
5561 u64 total_found;
edbd8d4e
CM		 5562 u32 nritems;
5563 int ret;
a061fc8d 5564 int progress;
1a40e23b 5565 int pass = 0;
edbd8d4e 5566
1a40e23b
ZY		 5567 root = root->fs_info->extent_root;
5568
5569 block_group = btrfs_lookup_block_group(info, group_start);
5570 BUG_ON(!block_group);
8f18cf13 5571
1a40e23b
ZY		 5572 printk("btrfs relocating block group %llu flags %llu\n",
5573 (unsigned long long)block_group->key.objectid,
5574 (unsigned long long)block_group->flags);
8f18cf13 5575
edbd8d4e 5576 path = btrfs_alloc_path();
1a40e23b 5577 BUG_ON(!path);
edbd8d4e 5578
1a40e23b
ZY		 5579 reloc_inode = create_reloc_inode(info, block_group);
5580 BUG_ON(IS_ERR(reloc_inode));
323da79c 5581
1a40e23b 5582 __alloc_chunk_for_shrink(root, block_group, 1);
c146afad 5583 set_block_group_readonly(block_group);
323da79c 5584
1a40e23b
ZY		 5585 btrfs_start_delalloc_inodes(info->tree_root);
5586 btrfs_wait_ordered_extents(info->tree_root, 0);
5587again:
d899e052 5588 skipped = 0;
edbd8d4e 5589 total_found = 0;
a061fc8d 5590 progress = 0;
1a40e23b 5591 key.objectid = block_group->key.objectid;
edbd8d4e
CM		 5592 key.offset = 0;
5593 key.type = 0;
73e48b27 5594 cur_byte = key.objectid;
4313b399 5595
1a40e23b
ZY		 5596 trans = btrfs_start_transaction(info->tree_root, 1);
5597 btrfs_commit_transaction(trans, info->tree_root);
ea8c2819 5598
1a40e23b
ZY		 5599 mutex_lock(&root->fs_info->cleaner_mutex);
5600 btrfs_clean_old_snapshots(info->tree_root);
5601 btrfs_remove_leaf_refs(info->tree_root, (u64)-1, 1);
5602 mutex_unlock(&root->fs_info->cleaner_mutex);
ea8c2819 5603
73e48b27 5604 while(1) {
edbd8d4e
CM		 5605 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5606 if (ret < 0)
5607 goto out;
7d9eb12c 5608next:
edbd8d4e 5609 leaf = path->nodes[0];
73e48b27 5610 nritems = btrfs_header_nritems(leaf);
73e48b27
Y		 5611 if (path->slots[0] >= nritems) {
5612 ret = btrfs_next_leaf(root, path);
5613 if (ret < 0)
5614 goto out;
5615 if (ret == 1) {
5616 ret = 0;
5617 break;
edbd8d4e 5618 }
73e48b27
Y		 5619 leaf = path->nodes[0];
5620 nritems = btrfs_header_nritems(leaf);
edbd8d4e 5621 }
73e48b27 5622
1a40e23b 5623 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
725c8463 5624
1a40e23b
ZY		 5625 if (key.objectid >= block_group->key.objectid +
5626 block_group->key.offset)
8f18cf13
CM		 5627 break;
5628
725c8463 5629 if (progress && need_resched()) {
725c8463 5630 btrfs_release_path(root, path);
1a40e23b 5631 cond_resched();
725c8463 5632 progress = 0;
1a40e23b 5633 continue;
725c8463
CM		 5634 }
5635 progress = 1;
5636
1a40e23b
ZY		 5637 if (btrfs_key_type(&key) != BTRFS_EXTENT_ITEM_KEY ||
5638 key.objectid + key.offset <= cur_byte) {
edbd8d4e 5639 path->slots[0]++;
edbd8d4e
CM		 5640 goto next;
5641 }
73e48b27 5642
edbd8d4e 5643 total_found++;
1a40e23b 5644 cur_byte = key.objectid + key.offset;
edbd8d4e 5645 btrfs_release_path(root, path);
edbd8d4e 5646
1a40e23b
ZY		 5647 __alloc_chunk_for_shrink(root, block_group, 0);
5648 ret = relocate_one_extent(root, path, &key, block_group,
5649 reloc_inode, pass);
5650 BUG_ON(ret < 0);
d899e052
YZ		 5651 if (ret > 0)
5652 skipped++;
edbd8d4e 5653
1a40e23b
ZY		 5654 key.objectid = cur_byte;
5655 key.type = 0;
5656 key.offset = 0;
edbd8d4e
CM		 5657 }
5658
1a40e23b 5659 btrfs_release_path(root, path);
7d9eb12c 5660
1a40e23b
ZY		 5661 if (pass == 0) {
5662 btrfs_wait_ordered_range(reloc_inode, 0, (u64)-1);
5663 invalidate_mapping_pages(reloc_inode->i_mapping, 0, -1);
5664 WARN_ON(reloc_inode->i_mapping->nrpages);
8e8a1e31 5665 }
73e48b27 5666
1a40e23b
ZY		 5667 if (total_found > 0) {
5668 printk("btrfs found %llu extents in pass %d\n",
5669 (unsigned long long)total_found, pass);
5670 pass++;
d899e052
YZ		 5671 if (total_found == skipped && pass > 2) {
5672 iput(reloc_inode);
5673 reloc_inode = create_reloc_inode(info, block_group);
5674 pass = 0;
5675 }
1a40e23b 5676 goto again;
0f9dd46c 5677 }
0ef3e66b 5678
1a40e23b
ZY		 5679 /* delete reloc_inode */
5680 iput(reloc_inode);
5681
5682 /* unpin extents in this range */
5683 trans = btrfs_start_transaction(info->tree_root, 1);
5684 btrfs_commit_transaction(trans, info->tree_root);
0ef3e66b 5685
1a40e23b
ZY		 5686 spin_lock(&block_group->lock);
5687 WARN_ON(block_group->pinned > 0);
5688 WARN_ON(block_group->reserved > 0);
5689 WARN_ON(btrfs_block_group_used(&block_group->item) > 0);
5690 spin_unlock(&block_group->lock);
d2fb3437 5691 put_block_group(block_group);
1a40e23b 5692 ret = 0;
edbd8d4e 5693out:
1a40e23b 5694 btrfs_free_path(path);
edbd8d4e
CM		 5695 return ret;
5696}
5697
b2950863
CH		 5698static int find_first_block_group(struct btrfs_root *root,
5699 struct btrfs_path *path, struct btrfs_key *key)
0b86a832 5700{
925baedd 5701 int ret = 0;
0b86a832
CM		 5702 struct btrfs_key found_key;
5703 struct extent_buffer *leaf;
5704 int slot;
edbd8d4e 5705
0b86a832
CM		 5706 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
5707 if (ret < 0)
925baedd
CM		 5708 goto out;
5709
0b86a832
CM		 5710 while(1) {
5711 slot = path->slots[0];
edbd8d4e 5712 leaf = path->nodes[0];
0b86a832
CM		 5713 if (slot >= btrfs_header_nritems(leaf)) {
5714 ret = btrfs_next_leaf(root, path);
5715 if (ret == 0)
5716 continue;
5717 if (ret < 0)
925baedd 5718 goto out;
0b86a832 5719 break;
edbd8d4e 5720 }
0b86a832 5721 btrfs_item_key_to_cpu(leaf, &found_key, slot);
edbd8d4e 5722
0b86a832 5723 if (found_key.objectid >= key->objectid &&
925baedd
CM		 5724 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
5725 ret = 0;
5726 goto out;
5727 }
0b86a832 5728 path->slots[0]++;
edbd8d4e 5729 }
0b86a832 5730 ret = -ENOENT;
925baedd 5731out:
0b86a832 5732 return ret;
edbd8d4e
CM		 5733}
5734
1a40e23b
ZY		 5735int btrfs_free_block_groups(struct btrfs_fs_info *info)
5736{
5737 struct btrfs_block_group_cache *block_group;
5738 struct rb_node *n;
5739
1a40e23b
ZY		 5740 spin_lock(&info->block_group_cache_lock);
5741 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
5742 block_group = rb_entry(n, struct btrfs_block_group_cache,
5743 cache_node);
1a40e23b
ZY		 5744 rb_erase(&block_group->cache_node,
5745 &info->block_group_cache_tree);
d899e052
YZ		 5746 spin_unlock(&info->block_group_cache_lock);
5747
5748 btrfs_remove_free_space_cache(block_group);
80eb234a 5749 down_write(&block_group->space_info->groups_sem);
1a40e23b 5750 list_del(&block_group->list);
80eb234a 5751 up_write(&block_group->space_info->groups_sem);
d2fb3437
YZ		 5752
5753 WARN_ON(atomic_read(&block_group->count) != 1);
1a40e23b 5754 kfree(block_group);
d899e052
YZ		 5755
5756 spin_lock(&info->block_group_cache_lock);
1a40e23b
ZY		 5757 }
5758 spin_unlock(&info->block_group_cache_lock);
1a40e23b
ZY		 5759 return 0;
5760}
5761
9078a3e1
CM		 5762int btrfs_read_block_groups(struct btrfs_root *root)
5763{
5764 struct btrfs_path *path;
5765 int ret;
9078a3e1 5766 struct btrfs_block_group_cache *cache;
be744175 5767 struct btrfs_fs_info *info = root->fs_info;
6324fbf3 5768 struct btrfs_space_info *space_info;
9078a3e1
CM		 5769 struct btrfs_key key;
5770 struct btrfs_key found_key;
5f39d397 5771 struct extent_buffer *leaf;
96b5179d 5772
be744175 5773 root = info->extent_root;
9078a3e1 5774 key.objectid = 0;
0b86a832 5775 key.offset = 0;
9078a3e1 5776 btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
9078a3e1
CM		 5777 path = btrfs_alloc_path();
5778 if (!path)
5779 return -ENOMEM;
5780
5781 while(1) {
0b86a832
CM		 5782 ret = find_first_block_group(root, path, &key);
5783 if (ret > 0) {
5784 ret = 0;
5785 goto error;
9078a3e1 5786 }
0b86a832
CM		 5787 if (ret != 0)
5788 goto error;
5789
5f39d397
CM		 5790 leaf = path->nodes[0];
5791 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
8f18cf13 5792 cache = kzalloc(sizeof(*cache), GFP_NOFS);
9078a3e1 5793 if (!cache) {
0b86a832 5794 ret = -ENOMEM;
9078a3e1
CM		 5795 break;
5796 }
3e1ad54f 5797
d2fb3437 5798 atomic_set(&cache->count, 1);
c286ac48 5799 spin_lock_init(&cache->lock);
25179201 5800 mutex_init(&cache->alloc_mutex);
ea6a478e 5801 mutex_init(&cache->cache_mutex);
0f9dd46c 5802 INIT_LIST_HEAD(&cache->list);
5f39d397
CM		 5803 read_extent_buffer(leaf, &cache->item,
5804 btrfs_item_ptr_offset(leaf, path->slots[0]),
5805 sizeof(cache->item));
9078a3e1 5806 memcpy(&cache->key, &found_key, sizeof(found_key));
0b86a832 5807
9078a3e1
CM		 5808 key.objectid = found_key.objectid + found_key.offset;
5809 btrfs_release_path(root, path);
0b86a832 5810 cache->flags = btrfs_block_group_flags(&cache->item);
96b5179d 5811
6324fbf3
CM		 5812 ret = update_space_info(info, cache->flags, found_key.offset,
5813 btrfs_block_group_used(&cache->item),
5814 &space_info);
5815 BUG_ON(ret);
5816 cache->space_info = space_info;
80eb234a
JB		 5817 down_write(&space_info->groups_sem);
5818 list_add_tail(&cache->list, &space_info->block_groups);
5819 up_write(&space_info->groups_sem);
0f9dd46c
JB		 5820
5821 ret = btrfs_add_block_group_cache(root->fs_info, cache);
5822 BUG_ON(ret);
75ccf47d
CM		 5823
5824 set_avail_alloc_bits(root->fs_info, cache->flags);
2b82032c
YZ		 5825 if (btrfs_chunk_readonly(root, cache->key.objectid))
5826 set_block_group_readonly(cache);
9078a3e1 5827 }
0b86a832
CM		 5828 ret = 0;
5829error:
9078a3e1 5830 btrfs_free_path(path);
0b86a832 5831 return ret;
9078a3e1 5832}
6324fbf3
CM		 5833
5834int btrfs_make_block_group(struct btrfs_trans_handle *trans,
5835 struct btrfs_root *root, u64 bytes_used,
e17cade2 5836 u64 type, u64 chunk_objectid, u64 chunk_offset,
6324fbf3
CM		 5837 u64 size)
5838{
5839 int ret;
6324fbf3
CM		 5840 struct btrfs_root *extent_root;
5841 struct btrfs_block_group_cache *cache;
6324fbf3
CM		 5842
5843 extent_root = root->fs_info->extent_root;
6324fbf3 5844
e02119d5
CM		 5845 root->fs_info->last_trans_new_blockgroup = trans->transid;
5846
8f18cf13 5847 cache = kzalloc(sizeof(*cache), GFP_NOFS);
0f9dd46c
JB		 5848 if (!cache)
5849 return -ENOMEM;
5850
e17cade2 5851 cache->key.objectid = chunk_offset;
6324fbf3 5852 cache->key.offset = size;
d2fb3437
YZ		 5853 cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
5854 atomic_set(&cache->count, 1);
c286ac48 5855 spin_lock_init(&cache->lock);
25179201 5856 mutex_init(&cache->alloc_mutex);
ea6a478e 5857 mutex_init(&cache->cache_mutex);
0f9dd46c 5858 INIT_LIST_HEAD(&cache->list);
0ef3e66b 5859
6324fbf3 5860 btrfs_set_block_group_used(&cache->item, bytes_used);
6324fbf3
CM		 5861 btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
5862 cache->flags = type;
5863 btrfs_set_block_group_flags(&cache->item, type);
5864
5865 ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
5866 &cache->space_info);
5867 BUG_ON(ret);
80eb234a
JB		 5868 down_write(&cache->space_info->groups_sem);
5869 list_add_tail(&cache->list, &cache->space_info->block_groups);
5870 up_write(&cache->space_info->groups_sem);
6324fbf3 5871
0f9dd46c
JB		 5872 ret = btrfs_add_block_group_cache(root->fs_info, cache);
5873 BUG_ON(ret);
c286ac48 5874
6324fbf3
CM		 5875 ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
5876 sizeof(cache->item));
5877 BUG_ON(ret);
5878
87ef2bb4
CM		 5879 finish_current_insert(trans, extent_root, 0);
5880 ret = del_pending_extents(trans, extent_root, 0);
6324fbf3 5881 BUG_ON(ret);
d18a2c44 5882 set_avail_alloc_bits(extent_root->fs_info, type);
925baedd 5883
6324fbf3
CM		 5884 return 0;
5885}
1a40e23b
ZY		 5886
5887int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
5888 struct btrfs_root *root, u64 group_start)
5889{
5890 struct btrfs_path *path;
5891 struct btrfs_block_group_cache *block_group;
5892 struct btrfs_key key;
5893 int ret;
5894
1a40e23b
ZY		 5895 root = root->fs_info->extent_root;
5896
5897 block_group = btrfs_lookup_block_group(root->fs_info, group_start);
5898 BUG_ON(!block_group);
c146afad 5899 BUG_ON(!block_group->ro);
1a40e23b
ZY		 5900
5901 memcpy(&key, &block_group->key, sizeof(key));
5902
5903 path = btrfs_alloc_path();
5904 BUG_ON(!path);
5905
5906 btrfs_remove_free_space_cache(block_group);
5907 rb_erase(&block_group->cache_node,
5908 &root->fs_info->block_group_cache_tree);
80eb234a 5909 down_write(&block_group->space_info->groups_sem);
1a40e23b 5910 list_del(&block_group->list);
80eb234a 5911 up_write(&block_group->space_info->groups_sem);
1a40e23b 5912
c146afad
YZ		 5913 spin_lock(&block_group->space_info->lock);
5914 block_group->space_info->total_bytes -= block_group->key.offset;
5915 block_group->space_info->bytes_readonly -= block_group->key.offset;
5916 spin_unlock(&block_group->space_info->lock);
2b82032c 5917 block_group->space_info->full = 0;
c146afad 5918
d2fb3437
YZ		 5919 put_block_group(block_group);
5920 put_block_group(block_group);
1a40e23b
ZY		 5921
5922 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
5923 if (ret > 0)
5924 ret = -EIO;
5925 if (ret < 0)
5926 goto out;
5927
5928 ret = btrfs_del_item(trans, root, path);
5929out:
5930 btrfs_free_path(path);
5931 return ret;
5932}
Linux kernel - Deliverables
This page took 0.421468 seconds and 5 git commands to generate.