blob: 9f5fdd37451deaa1c3b6306e72bc23d1fdc23c1b [file] [log] [blame]
Chris Mason6cbd5572007-06-12 09:07:21 -04001/*
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 */
Zach Brownec6b9102007-07-11 10:00:37 -040018#include <linux/sched.h>
Chris Masonedbd8d42007-12-21 16:27:24 -050019#include <linux/pagemap.h>
Chris Masonec44a352008-04-28 15:29:52 -040020#include <linux/writeback.h>
David Woodhouse21af8042008-08-12 14:13:26 +010021#include <linux/blkdev.h>
Chris Masonb7a9f292009-02-04 09:23:45 -050022#include <linux/sort.h>
Chris Mason4184ea72009-03-10 12:39:20 -040023#include <linux/rcupdate.h>
Josef Bacik817d52f2009-07-13 21:29:25 -040024#include <linux/kthread.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090025#include <linux/slab.h>
Chris Mason4b4e25f2008-11-20 10:22:27 -050026#include "compat.h"
Chris Mason74493f72007-12-11 09:25:06 -050027#include "hash.h"
Chris Masonfec577f2007-02-26 10:40:21 -050028#include "ctree.h"
29#include "disk-io.h"
30#include "print-tree.h"
Chris Masone089f052007-03-16 16:20:31 -040031#include "transaction.h"
Chris Mason0b86a832008-03-24 15:01:56 -040032#include "volumes.h"
Chris Mason925baed2008-06-25 16:01:30 -040033#include "locking.h"
Chris Masonfa9c0d72009-04-03 09:47:43 -040034#include "free-space-cache.h"
Chris Masonfec577f2007-02-26 10:40:21 -050035
Chris Mason0e4f8f82011-04-15 16:05:44 -040036/* control flags for do_chunk_alloc's force field
37 * CHUNK_ALLOC_NO_FORCE means to only allocate a chunk
38 * if we really need one.
39 *
40 * CHUNK_ALLOC_FORCE means it must try to allocate one
41 *
42 * CHUNK_ALLOC_LIMITED means to only try and allocate one
43 * if we have very few chunks already allocated. This is
44 * used as part of the clustering code to help make sure
45 * we have a good pool of storage to cluster in, without
46 * filling the FS with empty chunks
47 *
48 */
49enum {
50 CHUNK_ALLOC_NO_FORCE = 0,
51 CHUNK_ALLOC_FORCE = 1,
52 CHUNK_ALLOC_LIMITED = 2,
53};
54
Josef Bacikf3465ca2008-11-12 14:19:50 -050055static int update_block_group(struct btrfs_trans_handle *trans,
56 struct btrfs_root *root,
Yan, Zhengf0486c62010-05-16 10:46:25 -040057 u64 bytenr, u64 num_bytes, int alloc);
Yan Zheng5d4f98a2009-06-10 10:45:14 -040058static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
59 struct btrfs_root *root,
60 u64 bytenr, u64 num_bytes, u64 parent,
61 u64 root_objectid, u64 owner_objectid,
62 u64 owner_offset, int refs_to_drop,
63 struct btrfs_delayed_extent_op *extra_op);
64static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
65 struct extent_buffer *leaf,
66 struct btrfs_extent_item *ei);
67static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
68 struct btrfs_root *root,
69 u64 parent, u64 root_objectid,
70 u64 flags, u64 owner, u64 offset,
71 struct btrfs_key *ins, int ref_mod);
72static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
73 struct btrfs_root *root,
74 u64 parent, u64 root_objectid,
75 u64 flags, struct btrfs_disk_key *key,
76 int level, struct btrfs_key *ins);
Josef Bacik6a632092009-02-20 11:00:09 -050077static int do_chunk_alloc(struct btrfs_trans_handle *trans,
78 struct btrfs_root *extent_root, u64 alloc_bytes,
79 u64 flags, int force);
Yan Zheng11833d62009-09-11 16:11:19 -040080static int find_next_key(struct btrfs_path *path, int level,
81 struct btrfs_key *key);
Josef Bacik9ed74f22009-09-11 16:12:44 -040082static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
83 int dump_block_groups);
Josef Bacik6a632092009-02-20 11:00:09 -050084
Josef Bacik817d52f2009-07-13 21:29:25 -040085static noinline int
86block_group_cache_done(struct btrfs_block_group_cache *cache)
87{
88 smp_mb();
89 return cache->cached == BTRFS_CACHE_FINISHED;
90}
91
Josef Bacik0f9dd462008-09-23 13:14:11 -040092static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
93{
94 return (cache->flags & bits) == bits;
95}
96
Josef Bacik11dfe352009-11-13 20:12:59 +000097void btrfs_get_block_group(struct btrfs_block_group_cache *cache)
98{
99 atomic_inc(&cache->count);
100}
101
102void btrfs_put_block_group(struct btrfs_block_group_cache *cache)
103{
Yan, Zhengf0486c62010-05-16 10:46:25 -0400104 if (atomic_dec_and_test(&cache->count)) {
105 WARN_ON(cache->pinned > 0);
106 WARN_ON(cache->reserved > 0);
107 WARN_ON(cache->reserved_pinned > 0);
Josef Bacik11dfe352009-11-13 20:12:59 +0000108 kfree(cache);
Yan, Zhengf0486c62010-05-16 10:46:25 -0400109 }
Josef Bacik11dfe352009-11-13 20:12:59 +0000110}
111
Josef Bacik0f9dd462008-09-23 13:14:11 -0400112/*
113 * this adds the block group to the fs_info rb tree for the block group
114 * cache
115 */
Christoph Hellwigb2950862008-12-02 09:54:17 -0500116static int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
Josef Bacik0f9dd462008-09-23 13:14:11 -0400117 struct btrfs_block_group_cache *block_group)
118{
119 struct rb_node **p;
120 struct rb_node *parent = NULL;
121 struct btrfs_block_group_cache *cache;
122
123 spin_lock(&info->block_group_cache_lock);
124 p = &info->block_group_cache_tree.rb_node;
125
126 while (*p) {
127 parent = *p;
128 cache = rb_entry(parent, struct btrfs_block_group_cache,
129 cache_node);
130 if (block_group->key.objectid < cache->key.objectid) {
131 p = &(*p)->rb_left;
132 } else if (block_group->key.objectid > cache->key.objectid) {
133 p = &(*p)->rb_right;
134 } else {
135 spin_unlock(&info->block_group_cache_lock);
136 return -EEXIST;
137 }
138 }
139
140 rb_link_node(&block_group->cache_node, parent, p);
141 rb_insert_color(&block_group->cache_node,
142 &info->block_group_cache_tree);
143 spin_unlock(&info->block_group_cache_lock);
144
145 return 0;
146}
147
148/*
149 * This will return the block group at or after bytenr if contains is 0, else
150 * it will return the block group that contains the bytenr
151 */
152static struct btrfs_block_group_cache *
153block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
154 int contains)
155{
156 struct btrfs_block_group_cache *cache, *ret = NULL;
157 struct rb_node *n;
158 u64 end, start;
159
160 spin_lock(&info->block_group_cache_lock);
161 n = info->block_group_cache_tree.rb_node;
162
163 while (n) {
164 cache = rb_entry(n, struct btrfs_block_group_cache,
165 cache_node);
166 end = cache->key.objectid + cache->key.offset - 1;
167 start = cache->key.objectid;
168
169 if (bytenr < start) {
170 if (!contains && (!ret || start < ret->key.objectid))
171 ret = cache;
172 n = n->rb_left;
173 } else if (bytenr > start) {
174 if (contains && bytenr <= end) {
175 ret = cache;
176 break;
177 }
178 n = n->rb_right;
179 } else {
180 ret = cache;
181 break;
182 }
183 }
Yan Zhengd2fb3432008-12-11 16:30:39 -0500184 if (ret)
Josef Bacik11dfe352009-11-13 20:12:59 +0000185 btrfs_get_block_group(ret);
Josef Bacik0f9dd462008-09-23 13:14:11 -0400186 spin_unlock(&info->block_group_cache_lock);
187
188 return ret;
189}
190
Yan Zheng11833d62009-09-11 16:11:19 -0400191static int add_excluded_extent(struct btrfs_root *root,
192 u64 start, u64 num_bytes)
Josef Bacik817d52f2009-07-13 21:29:25 -0400193{
Yan Zheng11833d62009-09-11 16:11:19 -0400194 u64 end = start + num_bytes - 1;
195 set_extent_bits(&root->fs_info->freed_extents[0],
196 start, end, EXTENT_UPTODATE, GFP_NOFS);
197 set_extent_bits(&root->fs_info->freed_extents[1],
198 start, end, EXTENT_UPTODATE, GFP_NOFS);
199 return 0;
Josef Bacik817d52f2009-07-13 21:29:25 -0400200}
201
Yan Zheng11833d62009-09-11 16:11:19 -0400202static void free_excluded_extents(struct btrfs_root *root,
203 struct btrfs_block_group_cache *cache)
Josef Bacik817d52f2009-07-13 21:29:25 -0400204{
Yan Zheng11833d62009-09-11 16:11:19 -0400205 u64 start, end;
206
207 start = cache->key.objectid;
208 end = start + cache->key.offset - 1;
209
210 clear_extent_bits(&root->fs_info->freed_extents[0],
211 start, end, EXTENT_UPTODATE, GFP_NOFS);
212 clear_extent_bits(&root->fs_info->freed_extents[1],
213 start, end, EXTENT_UPTODATE, GFP_NOFS);
214}
215
216static int exclude_super_stripes(struct btrfs_root *root,
217 struct btrfs_block_group_cache *cache)
218{
Josef Bacik817d52f2009-07-13 21:29:25 -0400219 u64 bytenr;
220 u64 *logical;
221 int stripe_len;
222 int i, nr, ret;
223
Yan, Zheng06b23312009-11-26 09:31:11 +0000224 if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) {
225 stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid;
226 cache->bytes_super += stripe_len;
227 ret = add_excluded_extent(root, cache->key.objectid,
228 stripe_len);
229 BUG_ON(ret);
230 }
231
Josef Bacik817d52f2009-07-13 21:29:25 -0400232 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
233 bytenr = btrfs_sb_offset(i);
234 ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
235 cache->key.objectid, bytenr,
236 0, &logical, &nr, &stripe_len);
237 BUG_ON(ret);
Yan Zheng11833d62009-09-11 16:11:19 -0400238
Josef Bacik817d52f2009-07-13 21:29:25 -0400239 while (nr--) {
Josef Bacik1b2da372009-09-11 16:11:20 -0400240 cache->bytes_super += stripe_len;
Yan Zheng11833d62009-09-11 16:11:19 -0400241 ret = add_excluded_extent(root, logical[nr],
242 stripe_len);
243 BUG_ON(ret);
Josef Bacik817d52f2009-07-13 21:29:25 -0400244 }
Yan Zheng11833d62009-09-11 16:11:19 -0400245
Josef Bacik817d52f2009-07-13 21:29:25 -0400246 kfree(logical);
247 }
Josef Bacik817d52f2009-07-13 21:29:25 -0400248 return 0;
249}
250
Yan Zheng11833d62009-09-11 16:11:19 -0400251static struct btrfs_caching_control *
252get_caching_control(struct btrfs_block_group_cache *cache)
253{
254 struct btrfs_caching_control *ctl;
255
256 spin_lock(&cache->lock);
257 if (cache->cached != BTRFS_CACHE_STARTED) {
258 spin_unlock(&cache->lock);
259 return NULL;
260 }
261
Josef Bacikdde5abe2010-09-16 16:17:03 -0400262 /* We're loading it the fast way, so we don't have a caching_ctl. */
263 if (!cache->caching_ctl) {
264 spin_unlock(&cache->lock);
265 return NULL;
266 }
267
Yan Zheng11833d62009-09-11 16:11:19 -0400268 ctl = cache->caching_ctl;
269 atomic_inc(&ctl->count);
270 spin_unlock(&cache->lock);
271 return ctl;
272}
273
274static void put_caching_control(struct btrfs_caching_control *ctl)
275{
276 if (atomic_dec_and_test(&ctl->count))
277 kfree(ctl);
278}
279
Josef Bacik0f9dd462008-09-23 13:14:11 -0400280/*
281 * this is only called by cache_block_group, since we could have freed extents
282 * we need to check the pinned_extents for any extents that can't be used yet
283 * since their free space will be released as soon as the transaction commits.
284 */
Josef Bacik817d52f2009-07-13 21:29:25 -0400285static u64 add_new_free_space(struct btrfs_block_group_cache *block_group,
Josef Bacik0f9dd462008-09-23 13:14:11 -0400286 struct btrfs_fs_info *info, u64 start, u64 end)
287{
Josef Bacik817d52f2009-07-13 21:29:25 -0400288 u64 extent_start, extent_end, size, total_added = 0;
Josef Bacik0f9dd462008-09-23 13:14:11 -0400289 int ret;
290
291 while (start < end) {
Yan Zheng11833d62009-09-11 16:11:19 -0400292 ret = find_first_extent_bit(info->pinned_extents, start,
Josef Bacik0f9dd462008-09-23 13:14:11 -0400293 &extent_start, &extent_end,
Yan Zheng11833d62009-09-11 16:11:19 -0400294 EXTENT_DIRTY | EXTENT_UPTODATE);
Josef Bacik0f9dd462008-09-23 13:14:11 -0400295 if (ret)
296 break;
297
Yan, Zheng06b23312009-11-26 09:31:11 +0000298 if (extent_start <= start) {
Josef Bacik0f9dd462008-09-23 13:14:11 -0400299 start = extent_end + 1;
300 } else if (extent_start > start && extent_start < end) {
301 size = extent_start - start;
Josef Bacik817d52f2009-07-13 21:29:25 -0400302 total_added += size;
Josef Bacikea6a4782008-11-20 12:16:16 -0500303 ret = btrfs_add_free_space(block_group, start,
304 size);
Josef Bacik0f9dd462008-09-23 13:14:11 -0400305 BUG_ON(ret);
306 start = extent_end + 1;
307 } else {
308 break;
309 }
310 }
311
312 if (start < end) {
313 size = end - start;
Josef Bacik817d52f2009-07-13 21:29:25 -0400314 total_added += size;
Josef Bacikea6a4782008-11-20 12:16:16 -0500315 ret = btrfs_add_free_space(block_group, start, size);
Josef Bacik0f9dd462008-09-23 13:14:11 -0400316 BUG_ON(ret);
317 }
318
Josef Bacik817d52f2009-07-13 21:29:25 -0400319 return total_added;
Josef Bacik0f9dd462008-09-23 13:14:11 -0400320}
321
Josef Bacik817d52f2009-07-13 21:29:25 -0400322static int caching_kthread(void *data)
Chris Masone37c9e62007-05-09 20:13:14 -0400323{
Josef Bacik817d52f2009-07-13 21:29:25 -0400324 struct btrfs_block_group_cache *block_group = data;
325 struct btrfs_fs_info *fs_info = block_group->fs_info;
Yan Zheng11833d62009-09-11 16:11:19 -0400326 struct btrfs_caching_control *caching_ctl = block_group->caching_ctl;
327 struct btrfs_root *extent_root = fs_info->extent_root;
Chris Masone37c9e62007-05-09 20:13:14 -0400328 struct btrfs_path *path;
Chris Mason5f39d392007-10-15 16:14:19 -0400329 struct extent_buffer *leaf;
Yan Zheng11833d62009-09-11 16:11:19 -0400330 struct btrfs_key key;
Josef Bacik817d52f2009-07-13 21:29:25 -0400331 u64 total_found = 0;
Yan Zheng11833d62009-09-11 16:11:19 -0400332 u64 last = 0;
333 u32 nritems;
334 int ret = 0;
Chris Masonf510cfe2007-10-15 16:14:48 -0400335
Chris Masone37c9e62007-05-09 20:13:14 -0400336 path = btrfs_alloc_path();
337 if (!path)
338 return -ENOMEM;
Yan7d7d6062007-09-14 16:15:28 -0400339
Josef Bacik817d52f2009-07-13 21:29:25 -0400340 last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
Yan Zheng11833d62009-09-11 16:11:19 -0400341
Chris Mason5cd57b22008-06-25 16:01:30 -0400342 /*
Josef Bacik817d52f2009-07-13 21:29:25 -0400343 * We don't want to deadlock with somebody trying to allocate a new
344 * extent for the extent root while also trying to search the extent
345 * root to add free space. So we skip locking and search the commit
346 * root, since its read-only
Chris Mason5cd57b22008-06-25 16:01:30 -0400347 */
348 path->skip_locking = 1;
Josef Bacik817d52f2009-07-13 21:29:25 -0400349 path->search_commit_root = 1;
350 path->reada = 2;
351
Yan Zhenge4404d62008-12-12 10:03:26 -0500352 key.objectid = last;
Chris Masone37c9e62007-05-09 20:13:14 -0400353 key.offset = 0;
Yan Zheng11833d62009-09-11 16:11:19 -0400354 key.type = BTRFS_EXTENT_ITEM_KEY;
Chris Mason013f1b12009-07-31 14:57:55 -0400355again:
Yan Zheng11833d62009-09-11 16:11:19 -0400356 mutex_lock(&caching_ctl->mutex);
Chris Mason013f1b12009-07-31 14:57:55 -0400357 /* need to make sure the commit_root doesn't disappear */
358 down_read(&fs_info->extent_commit_sem);
359
Yan Zheng11833d62009-09-11 16:11:19 -0400360 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
Chris Masone37c9e62007-05-09 20:13:14 -0400361 if (ret < 0)
Josef Bacikef8bbdf2008-09-23 13:14:11 -0400362 goto err;
Yan Zhenga512bbf2008-12-08 16:46:26 -0500363
Yan Zheng11833d62009-09-11 16:11:19 -0400364 leaf = path->nodes[0];
365 nritems = btrfs_header_nritems(leaf);
366
Chris Masond3977122009-01-05 21:25:51 -0500367 while (1) {
Josef Bacik817d52f2009-07-13 21:29:25 -0400368 smp_mb();
Yan Zheng11833d62009-09-11 16:11:19 -0400369 if (fs_info->closing > 1) {
Yan Zhengf25784b2009-07-28 08:41:57 -0400370 last = (u64)-1;
Josef Bacik817d52f2009-07-13 21:29:25 -0400371 break;
Yan Zhengf25784b2009-07-28 08:41:57 -0400372 }
Josef Bacik817d52f2009-07-13 21:29:25 -0400373
Yan Zheng11833d62009-09-11 16:11:19 -0400374 if (path->slots[0] < nritems) {
375 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
376 } else {
377 ret = find_next_key(path, 0, &key);
378 if (ret)
Chris Masone37c9e62007-05-09 20:13:14 -0400379 break;
Josef Bacik817d52f2009-07-13 21:29:25 -0400380
Yan Zheng11833d62009-09-11 16:11:19 -0400381 caching_ctl->progress = last;
382 btrfs_release_path(extent_root, path);
383 up_read(&fs_info->extent_commit_sem);
384 mutex_unlock(&caching_ctl->mutex);
385 if (btrfs_transaction_in_commit(fs_info))
Chris Masonf36f3042009-07-30 10:04:48 -0400386 schedule_timeout(1);
Yan Zheng11833d62009-09-11 16:11:19 -0400387 else
388 cond_resched();
389 goto again;
390 }
Josef Bacik817d52f2009-07-13 21:29:25 -0400391
Yan Zheng11833d62009-09-11 16:11:19 -0400392 if (key.objectid < block_group->key.objectid) {
393 path->slots[0]++;
Josef Bacik817d52f2009-07-13 21:29:25 -0400394 continue;
Chris Masone37c9e62007-05-09 20:13:14 -0400395 }
Josef Bacik0f9dd462008-09-23 13:14:11 -0400396
Chris Masone37c9e62007-05-09 20:13:14 -0400397 if (key.objectid >= block_group->key.objectid +
Josef Bacik0f9dd462008-09-23 13:14:11 -0400398 block_group->key.offset)
Yan7d7d6062007-09-14 16:15:28 -0400399 break;
Yan7d7d6062007-09-14 16:15:28 -0400400
Yan Zheng11833d62009-09-11 16:11:19 -0400401 if (key.type == BTRFS_EXTENT_ITEM_KEY) {
Josef Bacik817d52f2009-07-13 21:29:25 -0400402 total_found += add_new_free_space(block_group,
403 fs_info, last,
404 key.objectid);
Yan7d7d6062007-09-14 16:15:28 -0400405 last = key.objectid + key.offset;
Josef Bacik817d52f2009-07-13 21:29:25 -0400406
Yan Zheng11833d62009-09-11 16:11:19 -0400407 if (total_found > (1024 * 1024 * 2)) {
408 total_found = 0;
409 wake_up(&caching_ctl->wait);
410 }
Josef Bacik817d52f2009-07-13 21:29:25 -0400411 }
Chris Masone37c9e62007-05-09 20:13:14 -0400412 path->slots[0]++;
413 }
Josef Bacikef8bbdf2008-09-23 13:14:11 -0400414 ret = 0;
Josef Bacik817d52f2009-07-13 21:29:25 -0400415
416 total_found += add_new_free_space(block_group, fs_info, last,
417 block_group->key.objectid +
418 block_group->key.offset);
Yan Zheng11833d62009-09-11 16:11:19 -0400419 caching_ctl->progress = (u64)-1;
Josef Bacik817d52f2009-07-13 21:29:25 -0400420
421 spin_lock(&block_group->lock);
Yan Zheng11833d62009-09-11 16:11:19 -0400422 block_group->caching_ctl = NULL;
Josef Bacik817d52f2009-07-13 21:29:25 -0400423 block_group->cached = BTRFS_CACHE_FINISHED;
424 spin_unlock(&block_group->lock);
425
Chris Mason54aa1f42007-06-22 14:16:25 -0400426err:
Chris Masone37c9e62007-05-09 20:13:14 -0400427 btrfs_free_path(path);
Yan Zheng276e6802009-07-30 09:40:40 -0400428 up_read(&fs_info->extent_commit_sem);
Josef Bacik817d52f2009-07-13 21:29:25 -0400429
Yan Zheng11833d62009-09-11 16:11:19 -0400430 free_excluded_extents(extent_root, block_group);
431
432 mutex_unlock(&caching_ctl->mutex);
433 wake_up(&caching_ctl->wait);
434
435 put_caching_control(caching_ctl);
436 atomic_dec(&block_group->space_info->caching_threads);
Josef Bacik11dfe352009-11-13 20:12:59 +0000437 btrfs_put_block_group(block_group);
438
Josef Bacik817d52f2009-07-13 21:29:25 -0400439 return 0;
440}
441
Josef Bacik9d66e232010-08-25 16:54:15 -0400442static int cache_block_group(struct btrfs_block_group_cache *cache,
443 struct btrfs_trans_handle *trans,
Josef Bacikb8399de2010-12-08 09:15:11 -0500444 struct btrfs_root *root,
Josef Bacik9d66e232010-08-25 16:54:15 -0400445 int load_cache_only)
Josef Bacik817d52f2009-07-13 21:29:25 -0400446{
Yan Zheng11833d62009-09-11 16:11:19 -0400447 struct btrfs_fs_info *fs_info = cache->fs_info;
448 struct btrfs_caching_control *caching_ctl;
Josef Bacik817d52f2009-07-13 21:29:25 -0400449 struct task_struct *tsk;
450 int ret = 0;
451
Yan Zheng11833d62009-09-11 16:11:19 -0400452 smp_mb();
453 if (cache->cached != BTRFS_CACHE_NO)
454 return 0;
455
Josef Bacik9d66e232010-08-25 16:54:15 -0400456 /*
457 * We can't do the read from on-disk cache during a commit since we need
Josef Bacikb8399de2010-12-08 09:15:11 -0500458 * to have the normal tree locking. Also if we are currently trying to
459 * allocate blocks for the tree root we can't do the fast caching since
460 * we likely hold important locks.
Josef Bacik9d66e232010-08-25 16:54:15 -0400461 */
Li Dongyangf7039b12011-03-24 10:24:28 +0000462 if (trans && (!trans->transaction->in_commit) &&
Josef Bacikb8399de2010-12-08 09:15:11 -0500463 (root && root != root->fs_info->tree_root)) {
Josef Bacik9d66e232010-08-25 16:54:15 -0400464 spin_lock(&cache->lock);
465 if (cache->cached != BTRFS_CACHE_NO) {
466 spin_unlock(&cache->lock);
467 return 0;
468 }
469 cache->cached = BTRFS_CACHE_STARTED;
470 spin_unlock(&cache->lock);
471
472 ret = load_free_space_cache(fs_info, cache);
473
474 spin_lock(&cache->lock);
475 if (ret == 1) {
476 cache->cached = BTRFS_CACHE_FINISHED;
477 cache->last_byte_to_unpin = (u64)-1;
478 } else {
479 cache->cached = BTRFS_CACHE_NO;
480 }
481 spin_unlock(&cache->lock);
Josef Bacik3c148742011-02-02 15:53:47 +0000482 if (ret == 1) {
483 free_excluded_extents(fs_info->extent_root, cache);
Josef Bacik9d66e232010-08-25 16:54:15 -0400484 return 0;
Josef Bacik3c148742011-02-02 15:53:47 +0000485 }
Josef Bacik9d66e232010-08-25 16:54:15 -0400486 }
487
488 if (load_cache_only)
489 return 0;
490
Miao Xiefc0e4a32011-03-24 11:41:21 +0000491 caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_NOFS);
Yan Zheng11833d62009-09-11 16:11:19 -0400492 BUG_ON(!caching_ctl);
493
494 INIT_LIST_HEAD(&caching_ctl->list);
495 mutex_init(&caching_ctl->mutex);
496 init_waitqueue_head(&caching_ctl->wait);
497 caching_ctl->block_group = cache;
498 caching_ctl->progress = cache->key.objectid;
499 /* one for caching kthread, one for caching block group list */
500 atomic_set(&caching_ctl->count, 2);
501
Josef Bacik817d52f2009-07-13 21:29:25 -0400502 spin_lock(&cache->lock);
503 if (cache->cached != BTRFS_CACHE_NO) {
504 spin_unlock(&cache->lock);
Yan Zheng11833d62009-09-11 16:11:19 -0400505 kfree(caching_ctl);
506 return 0;
Josef Bacik817d52f2009-07-13 21:29:25 -0400507 }
Yan Zheng11833d62009-09-11 16:11:19 -0400508 cache->caching_ctl = caching_ctl;
Josef Bacik817d52f2009-07-13 21:29:25 -0400509 cache->cached = BTRFS_CACHE_STARTED;
510 spin_unlock(&cache->lock);
511
Yan Zheng11833d62009-09-11 16:11:19 -0400512 down_write(&fs_info->extent_commit_sem);
513 list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups);
514 up_write(&fs_info->extent_commit_sem);
515
516 atomic_inc(&cache->space_info->caching_threads);
Josef Bacik11dfe352009-11-13 20:12:59 +0000517 btrfs_get_block_group(cache);
Yan Zheng11833d62009-09-11 16:11:19 -0400518
Josef Bacik817d52f2009-07-13 21:29:25 -0400519 tsk = kthread_run(caching_kthread, cache, "btrfs-cache-%llu\n",
520 cache->key.objectid);
521 if (IS_ERR(tsk)) {
522 ret = PTR_ERR(tsk);
523 printk(KERN_ERR "error running thread %d\n", ret);
524 BUG();
525 }
526
Josef Bacikef8bbdf2008-09-23 13:14:11 -0400527 return ret;
Chris Masone37c9e62007-05-09 20:13:14 -0400528}
529
Josef Bacik0f9dd462008-09-23 13:14:11 -0400530/*
531 * return the block group that starts at or after bytenr
532 */
Chris Masond3977122009-01-05 21:25:51 -0500533static struct btrfs_block_group_cache *
534btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr)
Chris Mason0ef3e662008-05-24 14:04:53 -0400535{
Josef Bacik0f9dd462008-09-23 13:14:11 -0400536 struct btrfs_block_group_cache *cache;
Chris Mason0ef3e662008-05-24 14:04:53 -0400537
Josef Bacik0f9dd462008-09-23 13:14:11 -0400538 cache = block_group_cache_tree_search(info, bytenr, 0);
Chris Mason0ef3e662008-05-24 14:04:53 -0400539
Josef Bacik0f9dd462008-09-23 13:14:11 -0400540 return cache;
Chris Mason0ef3e662008-05-24 14:04:53 -0400541}
542
Josef Bacik0f9dd462008-09-23 13:14:11 -0400543/*
Sankar P9f556842009-05-14 13:52:22 -0400544 * return the block group that contains the given bytenr
Josef Bacik0f9dd462008-09-23 13:14:11 -0400545 */
Chris Masond3977122009-01-05 21:25:51 -0500546struct btrfs_block_group_cache *btrfs_lookup_block_group(
547 struct btrfs_fs_info *info,
548 u64 bytenr)
Chris Masonbe744172007-05-06 10:15:01 -0400549{
Josef Bacik0f9dd462008-09-23 13:14:11 -0400550 struct btrfs_block_group_cache *cache;
Chris Masonbe744172007-05-06 10:15:01 -0400551
Josef Bacik0f9dd462008-09-23 13:14:11 -0400552 cache = block_group_cache_tree_search(info, bytenr, 1);
Chris Mason96b51792007-10-15 16:15:19 -0400553
Josef Bacik0f9dd462008-09-23 13:14:11 -0400554 return cache;
Chris Masonbe744172007-05-06 10:15:01 -0400555}
Chris Mason0b86a832008-03-24 15:01:56 -0400556
Josef Bacik0f9dd462008-09-23 13:14:11 -0400557static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
558 u64 flags)
Chris Mason6324fbf2008-03-24 15:01:59 -0400559{
Josef Bacik0f9dd462008-09-23 13:14:11 -0400560 struct list_head *head = &info->space_info;
Josef Bacik0f9dd462008-09-23 13:14:11 -0400561 struct btrfs_space_info *found;
Chris Mason4184ea72009-03-10 12:39:20 -0400562
Yan, Zhengb742bb82010-05-16 10:46:24 -0400563 flags &= BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_SYSTEM |
564 BTRFS_BLOCK_GROUP_METADATA;
565
Chris Mason4184ea72009-03-10 12:39:20 -0400566 rcu_read_lock();
567 list_for_each_entry_rcu(found, head, list) {
Josef Bacik67377732010-09-16 16:19:09 -0400568 if (found->flags & flags) {
Chris Mason4184ea72009-03-10 12:39:20 -0400569 rcu_read_unlock();
Josef Bacik0f9dd462008-09-23 13:14:11 -0400570 return found;
Chris Mason4184ea72009-03-10 12:39:20 -0400571 }
Josef Bacik0f9dd462008-09-23 13:14:11 -0400572 }
Chris Mason4184ea72009-03-10 12:39:20 -0400573 rcu_read_unlock();
Josef Bacik0f9dd462008-09-23 13:14:11 -0400574 return NULL;
Chris Mason6324fbf2008-03-24 15:01:59 -0400575}
576
Chris Mason4184ea72009-03-10 12:39:20 -0400577/*
578 * after adding space to the filesystem, we need to clear the full flags
579 * on all the space infos.
580 */
581void btrfs_clear_space_info_full(struct btrfs_fs_info *info)
582{
583 struct list_head *head = &info->space_info;
584 struct btrfs_space_info *found;
585
586 rcu_read_lock();
587 list_for_each_entry_rcu(found, head, list)
588 found->full = 0;
589 rcu_read_unlock();
590}
591
Josef Bacik80eb2342008-10-29 14:49:05 -0400592static u64 div_factor(u64 num, int factor)
593{
594 if (factor == 10)
595 return num;
596 num *= factor;
597 do_div(num, 10);
598 return num;
599}
600
Chris Masone5bc2452010-10-26 13:37:56 -0400601static u64 div_factor_fine(u64 num, int factor)
602{
603 if (factor == 100)
604 return num;
605 num *= factor;
606 do_div(num, 100);
607 return num;
608}
609
Yan Zhengd2fb3432008-12-11 16:30:39 -0500610u64 btrfs_find_block_group(struct btrfs_root *root,
611 u64 search_start, u64 search_hint, int owner)
Chris Masoncd1bc462007-04-27 10:08:34 -0400612{
Chris Mason96b51792007-10-15 16:15:19 -0400613 struct btrfs_block_group_cache *cache;
Chris Masoncd1bc462007-04-27 10:08:34 -0400614 u64 used;
Yan Zhengd2fb3432008-12-11 16:30:39 -0500615 u64 last = max(search_hint, search_start);
616 u64 group_start = 0;
Chris Mason31f3c992007-04-30 15:25:45 -0400617 int full_search = 0;
Yan Zhengd2fb3432008-12-11 16:30:39 -0500618 int factor = 9;
Chris Mason0ef3e662008-05-24 14:04:53 -0400619 int wrapped = 0;
Chris Mason31f3c992007-04-30 15:25:45 -0400620again:
Zheng Yane8569812008-09-26 10:05:48 -0400621 while (1) {
622 cache = btrfs_lookup_first_block_group(root->fs_info, last);
Josef Bacik0f9dd462008-09-23 13:14:11 -0400623 if (!cache)
Chris Masoncd1bc462007-04-27 10:08:34 -0400624 break;
Chris Mason96b51792007-10-15 16:15:19 -0400625
Chris Masonc286ac42008-07-22 23:06:41 -0400626 spin_lock(&cache->lock);
Chris Mason96b51792007-10-15 16:15:19 -0400627 last = cache->key.objectid + cache->key.offset;
628 used = btrfs_block_group_used(&cache->item);
629
Yan Zhengd2fb3432008-12-11 16:30:39 -0500630 if ((full_search || !cache->ro) &&
631 block_group_bits(cache, BTRFS_BLOCK_GROUP_METADATA)) {
Zheng Yane8569812008-09-26 10:05:48 -0400632 if (used + cache->pinned + cache->reserved <
Yan Zhengd2fb3432008-12-11 16:30:39 -0500633 div_factor(cache->key.offset, factor)) {
634 group_start = cache->key.objectid;
Chris Masonc286ac42008-07-22 23:06:41 -0400635 spin_unlock(&cache->lock);
Chris Masonfa9c0d72009-04-03 09:47:43 -0400636 btrfs_put_block_group(cache);
Chris Mason8790d502008-04-03 16:29:03 -0400637 goto found;
638 }
Chris Masoncd1bc462007-04-27 10:08:34 -0400639 }
Chris Masonc286ac42008-07-22 23:06:41 -0400640 spin_unlock(&cache->lock);
Chris Masonfa9c0d72009-04-03 09:47:43 -0400641 btrfs_put_block_group(cache);
Chris Masonde428b62007-05-18 13:28:27 -0400642 cond_resched();
Chris Masoncd1bc462007-04-27 10:08:34 -0400643 }
Chris Mason0ef3e662008-05-24 14:04:53 -0400644 if (!wrapped) {
645 last = search_start;
646 wrapped = 1;
647 goto again;
648 }
649 if (!full_search && factor < 10) {
Chris Masonbe744172007-05-06 10:15:01 -0400650 last = search_start;
Chris Mason31f3c992007-04-30 15:25:45 -0400651 full_search = 1;
Chris Mason0ef3e662008-05-24 14:04:53 -0400652 factor = 10;
Chris Mason31f3c992007-04-30 15:25:45 -0400653 goto again;
654 }
Chris Masonbe744172007-05-06 10:15:01 -0400655found:
Yan Zhengd2fb3432008-12-11 16:30:39 -0500656 return group_start;
Chris Mason925baed2008-06-25 16:01:30 -0400657}
Josef Bacik0f9dd462008-09-23 13:14:11 -0400658
Chris Masone02119d2008-09-05 16:13:11 -0400659/* simple helper to search for an existing extent at a given offset */
Zheng Yan31840ae2008-09-23 13:14:14 -0400660int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len)
Chris Masone02119d2008-09-05 16:13:11 -0400661{
662 int ret;
663 struct btrfs_key key;
Zheng Yan31840ae2008-09-23 13:14:14 -0400664 struct btrfs_path *path;
Chris Masone02119d2008-09-05 16:13:11 -0400665
Zheng Yan31840ae2008-09-23 13:14:14 -0400666 path = btrfs_alloc_path();
667 BUG_ON(!path);
Chris Masone02119d2008-09-05 16:13:11 -0400668 key.objectid = start;
669 key.offset = len;
670 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
671 ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
672 0, 0);
Zheng Yan31840ae2008-09-23 13:14:14 -0400673 btrfs_free_path(path);
Chris Mason7bb86312007-12-11 09:25:06 -0500674 return ret;
675}
676
Chris Masond8d5f3e2007-12-11 12:42:00 -0500677/*
Yan, Zhenga22285a2010-05-16 10:48:46 -0400678 * helper function to lookup reference count and flags of extent.
679 *
680 * the head node for delayed ref is used to store the sum of all the
681 * reference count modifications queued up in the rbtree. the head
682 * node may also store the extent flags to set. This way you can check
683 * to see what the reference count and extent flags would be if all of
684 * the delayed refs are not processed.
685 */
686int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans,
687 struct btrfs_root *root, u64 bytenr,
688 u64 num_bytes, u64 *refs, u64 *flags)
689{
690 struct btrfs_delayed_ref_head *head;
691 struct btrfs_delayed_ref_root *delayed_refs;
692 struct btrfs_path *path;
693 struct btrfs_extent_item *ei;
694 struct extent_buffer *leaf;
695 struct btrfs_key key;
696 u32 item_size;
697 u64 num_refs;
698 u64 extent_flags;
699 int ret;
700
701 path = btrfs_alloc_path();
702 if (!path)
703 return -ENOMEM;
704
705 key.objectid = bytenr;
706 key.type = BTRFS_EXTENT_ITEM_KEY;
707 key.offset = num_bytes;
708 if (!trans) {
709 path->skip_locking = 1;
710 path->search_commit_root = 1;
711 }
712again:
713 ret = btrfs_search_slot(trans, root->fs_info->extent_root,
714 &key, path, 0, 0);
715 if (ret < 0)
716 goto out_free;
717
718 if (ret == 0) {
719 leaf = path->nodes[0];
720 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
721 if (item_size >= sizeof(*ei)) {
722 ei = btrfs_item_ptr(leaf, path->slots[0],
723 struct btrfs_extent_item);
724 num_refs = btrfs_extent_refs(leaf, ei);
725 extent_flags = btrfs_extent_flags(leaf, ei);
726 } else {
727#ifdef BTRFS_COMPAT_EXTENT_TREE_V0
728 struct btrfs_extent_item_v0 *ei0;
729 BUG_ON(item_size != sizeof(*ei0));
730 ei0 = btrfs_item_ptr(leaf, path->slots[0],
731 struct btrfs_extent_item_v0);
732 num_refs = btrfs_extent_refs_v0(leaf, ei0);
733 /* FIXME: this isn't correct for data */
734 extent_flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
735#else
736 BUG();
737#endif
738 }
739 BUG_ON(num_refs == 0);
740 } else {
741 num_refs = 0;
742 extent_flags = 0;
743 ret = 0;
744 }
745
746 if (!trans)
747 goto out;
748
749 delayed_refs = &trans->transaction->delayed_refs;
750 spin_lock(&delayed_refs->lock);
751 head = btrfs_find_delayed_ref_head(trans, bytenr);
752 if (head) {
753 if (!mutex_trylock(&head->mutex)) {
754 atomic_inc(&head->node.refs);
755 spin_unlock(&delayed_refs->lock);
756
757 btrfs_release_path(root->fs_info->extent_root, path);
758
759 mutex_lock(&head->mutex);
760 mutex_unlock(&head->mutex);
761 btrfs_put_delayed_ref(&head->node);
762 goto again;
763 }
764 if (head->extent_op && head->extent_op->update_flags)
765 extent_flags |= head->extent_op->flags_to_set;
766 else
767 BUG_ON(num_refs == 0);
768
769 num_refs += head->node.ref_mod;
770 mutex_unlock(&head->mutex);
771 }
772 spin_unlock(&delayed_refs->lock);
773out:
774 WARN_ON(num_refs == 0);
775 if (refs)
776 *refs = num_refs;
777 if (flags)
778 *flags = extent_flags;
779out_free:
780 btrfs_free_path(path);
781 return ret;
782}
783
784/*
Chris Masond8d5f3e2007-12-11 12:42:00 -0500785 * Back reference rules. Back refs have three main goals:
786 *
787 * 1) differentiate between all holders of references to an extent so that
788 * when a reference is dropped we can make sure it was a valid reference
789 * before freeing the extent.
790 *
791 * 2) Provide enough information to quickly find the holders of an extent
792 * if we notice a given block is corrupted or bad.
793 *
794 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
795 * maintenance. This is actually the same as #2, but with a slightly
796 * different use case.
797 *
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400798 * There are two kinds of back refs. The implicit back refs is optimized
799 * for pointers in non-shared tree blocks. For a given pointer in a block,
800 * back refs of this kind provide information about the block's owner tree
801 * and the pointer's key. These information allow us to find the block by
802 * b-tree searching. The full back refs is for pointers in tree blocks not
803 * referenced by their owner trees. The location of tree block is recorded
804 * in the back refs. Actually the full back refs is generic, and can be
805 * used in all cases the implicit back refs is used. The major shortcoming
806 * of the full back refs is its overhead. Every time a tree block gets
807 * COWed, we have to update back refs entry for all pointers in it.
808 *
809 * For a newly allocated tree block, we use implicit back refs for
810 * pointers in it. This means most tree related operations only involve
811 * implicit back refs. For a tree block created in old transaction, the
812 * only way to drop a reference to it is COW it. So we can detect the
813 * event that tree block loses its owner tree's reference and do the
814 * back refs conversion.
815 *
816 * When a tree block is COW'd through a tree, there are four cases:
817 *
818 * The reference count of the block is one and the tree is the block's
819 * owner tree. Nothing to do in this case.
820 *
821 * The reference count of the block is one and the tree is not the
822 * block's owner tree. In this case, full back refs is used for pointers
823 * in the block. Remove these full back refs, add implicit back refs for
824 * every pointers in the new block.
825 *
826 * The reference count of the block is greater than one and the tree is
827 * the block's owner tree. In this case, implicit back refs is used for
828 * pointers in the block. Add full back refs for every pointers in the
829 * block, increase lower level extents' reference counts. The original
830 * implicit back refs are entailed to the new block.
831 *
832 * The reference count of the block is greater than one and the tree is
833 * not the block's owner tree. Add implicit back refs for every pointer in
834 * the new block, increase lower level extents' reference count.
835 *
836 * Back Reference Key composing:
837 *
838 * The key objectid corresponds to the first byte in the extent,
839 * The key type is used to differentiate between types of back refs.
840 * There are different meanings of the key offset for different types
841 * of back refs.
842 *
Chris Masond8d5f3e2007-12-11 12:42:00 -0500843 * File extents can be referenced by:
844 *
845 * - multiple snapshots, subvolumes, or different generations in one subvol
Zheng Yan31840ae2008-09-23 13:14:14 -0400846 * - different files inside a single subvolume
Chris Masond8d5f3e2007-12-11 12:42:00 -0500847 * - different offsets inside a file (bookend extents in file.c)
848 *
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400849 * The extent ref structure for the implicit back refs has fields for:
Chris Masond8d5f3e2007-12-11 12:42:00 -0500850 *
851 * - Objectid of the subvolume root
Chris Masond8d5f3e2007-12-11 12:42:00 -0500852 * - objectid of the file holding the reference
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400853 * - original offset in the file
854 * - how many bookend extents
Zheng Yan31840ae2008-09-23 13:14:14 -0400855 *
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400856 * The key offset for the implicit back refs is hash of the first
857 * three fields.
Chris Masond8d5f3e2007-12-11 12:42:00 -0500858 *
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400859 * The extent ref structure for the full back refs has field for:
Chris Masond8d5f3e2007-12-11 12:42:00 -0500860 *
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400861 * - number of pointers in the tree leaf
Chris Masond8d5f3e2007-12-11 12:42:00 -0500862 *
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400863 * The key offset for the implicit back refs is the first byte of
864 * the tree leaf
Chris Masond8d5f3e2007-12-11 12:42:00 -0500865 *
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400866 * When a file extent is allocated, The implicit back refs is used.
867 * the fields are filled in:
Chris Masond8d5f3e2007-12-11 12:42:00 -0500868 *
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400869 * (root_key.objectid, inode objectid, offset in file, 1)
870 *
871 * When a file extent is removed file truncation, we find the
872 * corresponding implicit back refs and check the following fields:
873 *
874 * (btrfs_header_owner(leaf), inode objectid, offset in file)
Chris Masond8d5f3e2007-12-11 12:42:00 -0500875 *
876 * Btree extents can be referenced by:
877 *
878 * - Different subvolumes
Chris Masond8d5f3e2007-12-11 12:42:00 -0500879 *
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400880 * Both the implicit back refs and the full back refs for tree blocks
881 * only consist of key. The key offset for the implicit back refs is
882 * objectid of block's owner tree. The key offset for the full back refs
883 * is the first byte of parent block.
Chris Masond8d5f3e2007-12-11 12:42:00 -0500884 *
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400885 * When implicit back refs is used, information about the lowest key and
886 * level of the tree block are required. These information are stored in
887 * tree block info structure.
Chris Masond8d5f3e2007-12-11 12:42:00 -0500888 */
Zheng Yan31840ae2008-09-23 13:14:14 -0400889
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400890#ifdef BTRFS_COMPAT_EXTENT_TREE_V0
891static int convert_extent_item_v0(struct btrfs_trans_handle *trans,
892 struct btrfs_root *root,
893 struct btrfs_path *path,
894 u64 owner, u32 extra_size)
Chris Mason74493f72007-12-11 09:25:06 -0500895{
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400896 struct btrfs_extent_item *item;
897 struct btrfs_extent_item_v0 *ei0;
898 struct btrfs_extent_ref_v0 *ref0;
899 struct btrfs_tree_block_info *bi;
Zheng Yan31840ae2008-09-23 13:14:14 -0400900 struct extent_buffer *leaf;
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400901 struct btrfs_key key;
902 struct btrfs_key found_key;
903 u32 new_size = sizeof(*item);
904 u64 refs;
Chris Mason74493f72007-12-11 09:25:06 -0500905 int ret;
906
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400907 leaf = path->nodes[0];
908 BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0));
Chris Mason74493f72007-12-11 09:25:06 -0500909
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400910 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
911 ei0 = btrfs_item_ptr(leaf, path->slots[0],
912 struct btrfs_extent_item_v0);
913 refs = btrfs_extent_refs_v0(leaf, ei0);
914
915 if (owner == (u64)-1) {
916 while (1) {
917 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
918 ret = btrfs_next_leaf(root, path);
919 if (ret < 0)
920 return ret;
921 BUG_ON(ret > 0);
922 leaf = path->nodes[0];
923 }
924 btrfs_item_key_to_cpu(leaf, &found_key,
925 path->slots[0]);
926 BUG_ON(key.objectid != found_key.objectid);
927 if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) {
928 path->slots[0]++;
929 continue;
930 }
931 ref0 = btrfs_item_ptr(leaf, path->slots[0],
932 struct btrfs_extent_ref_v0);
933 owner = btrfs_ref_objectid_v0(leaf, ref0);
934 break;
935 }
936 }
937 btrfs_release_path(root, path);
938
939 if (owner < BTRFS_FIRST_FREE_OBJECTID)
940 new_size += sizeof(*bi);
941
942 new_size -= sizeof(*ei0);
943 ret = btrfs_search_slot(trans, root, &key, path,
944 new_size + extra_size, 1);
Zheng Yan31840ae2008-09-23 13:14:14 -0400945 if (ret < 0)
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400946 return ret;
947 BUG_ON(ret);
948
949 ret = btrfs_extend_item(trans, root, path, new_size);
950 BUG_ON(ret);
951
952 leaf = path->nodes[0];
953 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
954 btrfs_set_extent_refs(leaf, item, refs);
955 /* FIXME: get real generation */
956 btrfs_set_extent_generation(leaf, item, 0);
957 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
958 btrfs_set_extent_flags(leaf, item,
959 BTRFS_EXTENT_FLAG_TREE_BLOCK |
960 BTRFS_BLOCK_FLAG_FULL_BACKREF);
961 bi = (struct btrfs_tree_block_info *)(item + 1);
962 /* FIXME: get first key of the block */
963 memset_extent_buffer(leaf, 0, (unsigned long)bi, sizeof(*bi));
964 btrfs_set_tree_block_level(leaf, bi, (int)owner);
965 } else {
966 btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA);
967 }
968 btrfs_mark_buffer_dirty(leaf);
969 return 0;
970}
971#endif
972
973static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset)
974{
975 u32 high_crc = ~(u32)0;
976 u32 low_crc = ~(u32)0;
977 __le64 lenum;
978
979 lenum = cpu_to_le64(root_objectid);
David Woodhouse163e7832009-04-19 13:02:41 +0100980 high_crc = crc32c(high_crc, &lenum, sizeof(lenum));
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400981 lenum = cpu_to_le64(owner);
David Woodhouse163e7832009-04-19 13:02:41 +0100982 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400983 lenum = cpu_to_le64(offset);
David Woodhouse163e7832009-04-19 13:02:41 +0100984 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400985
986 return ((u64)high_crc << 31) ^ (u64)low_crc;
987}
988
989static u64 hash_extent_data_ref_item(struct extent_buffer *leaf,
990 struct btrfs_extent_data_ref *ref)
991{
992 return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref),
993 btrfs_extent_data_ref_objectid(leaf, ref),
994 btrfs_extent_data_ref_offset(leaf, ref));
995}
996
997static int match_extent_data_ref(struct extent_buffer *leaf,
998 struct btrfs_extent_data_ref *ref,
999 u64 root_objectid, u64 owner, u64 offset)
1000{
1001 if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid ||
1002 btrfs_extent_data_ref_objectid(leaf, ref) != owner ||
1003 btrfs_extent_data_ref_offset(leaf, ref) != offset)
1004 return 0;
1005 return 1;
1006}
1007
1008static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans,
1009 struct btrfs_root *root,
1010 struct btrfs_path *path,
1011 u64 bytenr, u64 parent,
1012 u64 root_objectid,
1013 u64 owner, u64 offset)
1014{
1015 struct btrfs_key key;
1016 struct btrfs_extent_data_ref *ref;
1017 struct extent_buffer *leaf;
1018 u32 nritems;
1019 int ret;
1020 int recow;
1021 int err = -ENOENT;
1022
1023 key.objectid = bytenr;
1024 if (parent) {
1025 key.type = BTRFS_SHARED_DATA_REF_KEY;
1026 key.offset = parent;
1027 } else {
1028 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1029 key.offset = hash_extent_data_ref(root_objectid,
1030 owner, offset);
1031 }
1032again:
1033 recow = 0;
1034 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1035 if (ret < 0) {
1036 err = ret;
1037 goto fail;
1038 }
1039
1040 if (parent) {
1041 if (!ret)
1042 return 0;
1043#ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1044 key.type = BTRFS_EXTENT_REF_V0_KEY;
1045 btrfs_release_path(root, path);
1046 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1047 if (ret < 0) {
1048 err = ret;
1049 goto fail;
1050 }
1051 if (!ret)
1052 return 0;
1053#endif
1054 goto fail;
Zheng Yan31840ae2008-09-23 13:14:14 -04001055 }
1056
1057 leaf = path->nodes[0];
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001058 nritems = btrfs_header_nritems(leaf);
1059 while (1) {
1060 if (path->slots[0] >= nritems) {
1061 ret = btrfs_next_leaf(root, path);
1062 if (ret < 0)
1063 err = ret;
1064 if (ret)
1065 goto fail;
1066
1067 leaf = path->nodes[0];
1068 nritems = btrfs_header_nritems(leaf);
1069 recow = 1;
1070 }
1071
1072 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1073 if (key.objectid != bytenr ||
1074 key.type != BTRFS_EXTENT_DATA_REF_KEY)
1075 goto fail;
1076
1077 ref = btrfs_item_ptr(leaf, path->slots[0],
1078 struct btrfs_extent_data_ref);
1079
1080 if (match_extent_data_ref(leaf, ref, root_objectid,
1081 owner, offset)) {
1082 if (recow) {
1083 btrfs_release_path(root, path);
1084 goto again;
1085 }
1086 err = 0;
1087 break;
1088 }
1089 path->slots[0]++;
Zheng Yan31840ae2008-09-23 13:14:14 -04001090 }
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001091fail:
1092 return err;
Zheng Yan31840ae2008-09-23 13:14:14 -04001093}
1094
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001095static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans,
1096 struct btrfs_root *root,
1097 struct btrfs_path *path,
1098 u64 bytenr, u64 parent,
1099 u64 root_objectid, u64 owner,
1100 u64 offset, int refs_to_add)
Zheng Yan31840ae2008-09-23 13:14:14 -04001101{
1102 struct btrfs_key key;
1103 struct extent_buffer *leaf;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001104 u32 size;
Zheng Yan31840ae2008-09-23 13:14:14 -04001105 u32 num_refs;
1106 int ret;
1107
1108 key.objectid = bytenr;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001109 if (parent) {
1110 key.type = BTRFS_SHARED_DATA_REF_KEY;
1111 key.offset = parent;
1112 size = sizeof(struct btrfs_shared_data_ref);
Zheng Yan31840ae2008-09-23 13:14:14 -04001113 } else {
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001114 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1115 key.offset = hash_extent_data_ref(root_objectid,
1116 owner, offset);
1117 size = sizeof(struct btrfs_extent_data_ref);
Zheng Yan31840ae2008-09-23 13:14:14 -04001118 }
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001119
1120 ret = btrfs_insert_empty_item(trans, root, path, &key, size);
1121 if (ret && ret != -EEXIST)
1122 goto fail;
1123
1124 leaf = path->nodes[0];
1125 if (parent) {
1126 struct btrfs_shared_data_ref *ref;
1127 ref = btrfs_item_ptr(leaf, path->slots[0],
1128 struct btrfs_shared_data_ref);
1129 if (ret == 0) {
1130 btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add);
1131 } else {
1132 num_refs = btrfs_shared_data_ref_count(leaf, ref);
1133 num_refs += refs_to_add;
1134 btrfs_set_shared_data_ref_count(leaf, ref, num_refs);
1135 }
1136 } else {
1137 struct btrfs_extent_data_ref *ref;
1138 while (ret == -EEXIST) {
1139 ref = btrfs_item_ptr(leaf, path->slots[0],
1140 struct btrfs_extent_data_ref);
1141 if (match_extent_data_ref(leaf, ref, root_objectid,
1142 owner, offset))
1143 break;
1144 btrfs_release_path(root, path);
1145 key.offset++;
1146 ret = btrfs_insert_empty_item(trans, root, path, &key,
1147 size);
1148 if (ret && ret != -EEXIST)
1149 goto fail;
1150
1151 leaf = path->nodes[0];
1152 }
1153 ref = btrfs_item_ptr(leaf, path->slots[0],
1154 struct btrfs_extent_data_ref);
1155 if (ret == 0) {
1156 btrfs_set_extent_data_ref_root(leaf, ref,
1157 root_objectid);
1158 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
1159 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
1160 btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add);
1161 } else {
1162 num_refs = btrfs_extent_data_ref_count(leaf, ref);
1163 num_refs += refs_to_add;
1164 btrfs_set_extent_data_ref_count(leaf, ref, num_refs);
1165 }
1166 }
1167 btrfs_mark_buffer_dirty(leaf);
1168 ret = 0;
1169fail:
Chris Mason7bb86312007-12-11 09:25:06 -05001170 btrfs_release_path(root, path);
1171 return ret;
Chris Mason74493f72007-12-11 09:25:06 -05001172}
1173
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001174static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans,
1175 struct btrfs_root *root,
1176 struct btrfs_path *path,
1177 int refs_to_drop)
Zheng Yan31840ae2008-09-23 13:14:14 -04001178{
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001179 struct btrfs_key key;
1180 struct btrfs_extent_data_ref *ref1 = NULL;
1181 struct btrfs_shared_data_ref *ref2 = NULL;
Zheng Yan31840ae2008-09-23 13:14:14 -04001182 struct extent_buffer *leaf;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001183 u32 num_refs = 0;
Zheng Yan31840ae2008-09-23 13:14:14 -04001184 int ret = 0;
1185
1186 leaf = path->nodes[0];
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001187 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1188
1189 if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1190 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1191 struct btrfs_extent_data_ref);
1192 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1193 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1194 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1195 struct btrfs_shared_data_ref);
1196 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1197#ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1198 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1199 struct btrfs_extent_ref_v0 *ref0;
1200 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1201 struct btrfs_extent_ref_v0);
1202 num_refs = btrfs_ref_count_v0(leaf, ref0);
1203#endif
1204 } else {
1205 BUG();
1206 }
1207
Chris Mason56bec292009-03-13 10:10:06 -04001208 BUG_ON(num_refs < refs_to_drop);
1209 num_refs -= refs_to_drop;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001210
Zheng Yan31840ae2008-09-23 13:14:14 -04001211 if (num_refs == 0) {
1212 ret = btrfs_del_item(trans, root, path);
1213 } else {
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001214 if (key.type == BTRFS_EXTENT_DATA_REF_KEY)
1215 btrfs_set_extent_data_ref_count(leaf, ref1, num_refs);
1216 else if (key.type == BTRFS_SHARED_DATA_REF_KEY)
1217 btrfs_set_shared_data_ref_count(leaf, ref2, num_refs);
1218#ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1219 else {
1220 struct btrfs_extent_ref_v0 *ref0;
1221 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1222 struct btrfs_extent_ref_v0);
1223 btrfs_set_ref_count_v0(leaf, ref0, num_refs);
1224 }
1225#endif
Zheng Yan31840ae2008-09-23 13:14:14 -04001226 btrfs_mark_buffer_dirty(leaf);
1227 }
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001228 return ret;
1229}
1230
1231static noinline u32 extent_data_ref_count(struct btrfs_root *root,
1232 struct btrfs_path *path,
1233 struct btrfs_extent_inline_ref *iref)
1234{
1235 struct btrfs_key key;
1236 struct extent_buffer *leaf;
1237 struct btrfs_extent_data_ref *ref1;
1238 struct btrfs_shared_data_ref *ref2;
1239 u32 num_refs = 0;
1240
1241 leaf = path->nodes[0];
1242 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1243 if (iref) {
1244 if (btrfs_extent_inline_ref_type(leaf, iref) ==
1245 BTRFS_EXTENT_DATA_REF_KEY) {
1246 ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
1247 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1248 } else {
1249 ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
1250 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1251 }
1252 } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1253 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1254 struct btrfs_extent_data_ref);
1255 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1256 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1257 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1258 struct btrfs_shared_data_ref);
1259 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1260#ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1261 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1262 struct btrfs_extent_ref_v0 *ref0;
1263 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1264 struct btrfs_extent_ref_v0);
1265 num_refs = btrfs_ref_count_v0(leaf, ref0);
1266#endif
1267 } else {
1268 WARN_ON(1);
1269 }
1270 return num_refs;
1271}
1272
1273static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
1274 struct btrfs_root *root,
1275 struct btrfs_path *path,
1276 u64 bytenr, u64 parent,
1277 u64 root_objectid)
1278{
1279 struct btrfs_key key;
1280 int ret;
1281
1282 key.objectid = bytenr;
1283 if (parent) {
1284 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1285 key.offset = parent;
1286 } else {
1287 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1288 key.offset = root_objectid;
1289 }
1290
1291 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1292 if (ret > 0)
1293 ret = -ENOENT;
1294#ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1295 if (ret == -ENOENT && parent) {
1296 btrfs_release_path(root, path);
1297 key.type = BTRFS_EXTENT_REF_V0_KEY;
1298 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1299 if (ret > 0)
1300 ret = -ENOENT;
1301 }
1302#endif
1303 return ret;
1304}
1305
1306static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
1307 struct btrfs_root *root,
1308 struct btrfs_path *path,
1309 u64 bytenr, u64 parent,
1310 u64 root_objectid)
1311{
1312 struct btrfs_key key;
1313 int ret;
1314
1315 key.objectid = bytenr;
1316 if (parent) {
1317 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1318 key.offset = parent;
1319 } else {
1320 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1321 key.offset = root_objectid;
1322 }
1323
1324 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
Zheng Yan31840ae2008-09-23 13:14:14 -04001325 btrfs_release_path(root, path);
1326 return ret;
1327}
1328
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001329static inline int extent_ref_type(u64 parent, u64 owner)
1330{
1331 int type;
1332 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1333 if (parent > 0)
1334 type = BTRFS_SHARED_BLOCK_REF_KEY;
1335 else
1336 type = BTRFS_TREE_BLOCK_REF_KEY;
1337 } else {
1338 if (parent > 0)
1339 type = BTRFS_SHARED_DATA_REF_KEY;
1340 else
1341 type = BTRFS_EXTENT_DATA_REF_KEY;
1342 }
1343 return type;
1344}
1345
Yan Zheng2c47e6052009-06-27 21:07:35 -04001346static int find_next_key(struct btrfs_path *path, int level,
1347 struct btrfs_key *key)
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001348
1349{
Yan Zheng2c47e6052009-06-27 21:07:35 -04001350 for (; level < BTRFS_MAX_LEVEL; level++) {
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001351 if (!path->nodes[level])
1352 break;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001353 if (path->slots[level] + 1 >=
1354 btrfs_header_nritems(path->nodes[level]))
1355 continue;
1356 if (level == 0)
1357 btrfs_item_key_to_cpu(path->nodes[level], key,
1358 path->slots[level] + 1);
1359 else
1360 btrfs_node_key_to_cpu(path->nodes[level], key,
1361 path->slots[level] + 1);
1362 return 0;
1363 }
1364 return 1;
1365}
1366
1367/*
1368 * look for inline back ref. if back ref is found, *ref_ret is set
1369 * to the address of inline back ref, and 0 is returned.
1370 *
1371 * if back ref isn't found, *ref_ret is set to the address where it
1372 * should be inserted, and -ENOENT is returned.
1373 *
1374 * if insert is true and there are too many inline back refs, the path
1375 * points to the extent item, and -EAGAIN is returned.
1376 *
1377 * NOTE: inline back refs are ordered in the same way that back ref
1378 * items in the tree are ordered.
1379 */
1380static noinline_for_stack
1381int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
1382 struct btrfs_root *root,
1383 struct btrfs_path *path,
1384 struct btrfs_extent_inline_ref **ref_ret,
1385 u64 bytenr, u64 num_bytes,
1386 u64 parent, u64 root_objectid,
1387 u64 owner, u64 offset, int insert)
1388{
1389 struct btrfs_key key;
1390 struct extent_buffer *leaf;
1391 struct btrfs_extent_item *ei;
1392 struct btrfs_extent_inline_ref *iref;
1393 u64 flags;
1394 u64 item_size;
1395 unsigned long ptr;
1396 unsigned long end;
1397 int extra_size;
1398 int type;
1399 int want;
1400 int ret;
1401 int err = 0;
1402
1403 key.objectid = bytenr;
1404 key.type = BTRFS_EXTENT_ITEM_KEY;
1405 key.offset = num_bytes;
1406
1407 want = extent_ref_type(parent, owner);
1408 if (insert) {
1409 extra_size = btrfs_extent_inline_ref_size(want);
Yan Zheng85d41982009-06-11 08:51:10 -04001410 path->keep_locks = 1;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001411 } else
1412 extra_size = -1;
1413 ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
1414 if (ret < 0) {
1415 err = ret;
1416 goto out;
1417 }
1418 BUG_ON(ret);
1419
1420 leaf = path->nodes[0];
1421 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1422#ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1423 if (item_size < sizeof(*ei)) {
1424 if (!insert) {
1425 err = -ENOENT;
1426 goto out;
1427 }
1428 ret = convert_extent_item_v0(trans, root, path, owner,
1429 extra_size);
1430 if (ret < 0) {
1431 err = ret;
1432 goto out;
1433 }
1434 leaf = path->nodes[0];
1435 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1436 }
1437#endif
1438 BUG_ON(item_size < sizeof(*ei));
1439
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001440 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1441 flags = btrfs_extent_flags(leaf, ei);
1442
1443 ptr = (unsigned long)(ei + 1);
1444 end = (unsigned long)ei + item_size;
1445
1446 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
1447 ptr += sizeof(struct btrfs_tree_block_info);
1448 BUG_ON(ptr > end);
1449 } else {
1450 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_DATA));
1451 }
1452
1453 err = -ENOENT;
1454 while (1) {
1455 if (ptr >= end) {
1456 WARN_ON(ptr > end);
1457 break;
1458 }
1459 iref = (struct btrfs_extent_inline_ref *)ptr;
1460 type = btrfs_extent_inline_ref_type(leaf, iref);
1461 if (want < type)
1462 break;
1463 if (want > type) {
1464 ptr += btrfs_extent_inline_ref_size(type);
1465 continue;
1466 }
1467
1468 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1469 struct btrfs_extent_data_ref *dref;
1470 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1471 if (match_extent_data_ref(leaf, dref, root_objectid,
1472 owner, offset)) {
1473 err = 0;
1474 break;
1475 }
1476 if (hash_extent_data_ref_item(leaf, dref) <
1477 hash_extent_data_ref(root_objectid, owner, offset))
1478 break;
1479 } else {
1480 u64 ref_offset;
1481 ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1482 if (parent > 0) {
1483 if (parent == ref_offset) {
1484 err = 0;
1485 break;
1486 }
1487 if (ref_offset < parent)
1488 break;
1489 } else {
1490 if (root_objectid == ref_offset) {
1491 err = 0;
1492 break;
1493 }
1494 if (ref_offset < root_objectid)
1495 break;
1496 }
1497 }
1498 ptr += btrfs_extent_inline_ref_size(type);
1499 }
1500 if (err == -ENOENT && insert) {
1501 if (item_size + extra_size >=
1502 BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
1503 err = -EAGAIN;
1504 goto out;
1505 }
1506 /*
1507 * To add new inline back ref, we have to make sure
1508 * there is no corresponding back ref item.
1509 * For simplicity, we just do not add new inline back
1510 * ref if there is any kind of item for this block
1511 */
Yan Zheng2c47e6052009-06-27 21:07:35 -04001512 if (find_next_key(path, 0, &key) == 0 &&
1513 key.objectid == bytenr &&
Yan Zheng85d41982009-06-11 08:51:10 -04001514 key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001515 err = -EAGAIN;
1516 goto out;
1517 }
1518 }
1519 *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
1520out:
Yan Zheng85d41982009-06-11 08:51:10 -04001521 if (insert) {
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001522 path->keep_locks = 0;
1523 btrfs_unlock_up_safe(path, 1);
1524 }
1525 return err;
1526}
1527
1528/*
1529 * helper to add new inline back ref
1530 */
1531static noinline_for_stack
1532int setup_inline_extent_backref(struct btrfs_trans_handle *trans,
1533 struct btrfs_root *root,
1534 struct btrfs_path *path,
1535 struct btrfs_extent_inline_ref *iref,
1536 u64 parent, u64 root_objectid,
1537 u64 owner, u64 offset, int refs_to_add,
1538 struct btrfs_delayed_extent_op *extent_op)
1539{
1540 struct extent_buffer *leaf;
1541 struct btrfs_extent_item *ei;
1542 unsigned long ptr;
1543 unsigned long end;
1544 unsigned long item_offset;
1545 u64 refs;
1546 int size;
1547 int type;
1548 int ret;
1549
1550 leaf = path->nodes[0];
1551 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1552 item_offset = (unsigned long)iref - (unsigned long)ei;
1553
1554 type = extent_ref_type(parent, owner);
1555 size = btrfs_extent_inline_ref_size(type);
1556
1557 ret = btrfs_extend_item(trans, root, path, size);
1558 BUG_ON(ret);
1559
1560 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1561 refs = btrfs_extent_refs(leaf, ei);
1562 refs += refs_to_add;
1563 btrfs_set_extent_refs(leaf, ei, refs);
1564 if (extent_op)
1565 __run_delayed_extent_op(extent_op, leaf, ei);
1566
1567 ptr = (unsigned long)ei + item_offset;
1568 end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
1569 if (ptr < end - size)
1570 memmove_extent_buffer(leaf, ptr + size, ptr,
1571 end - size - ptr);
1572
1573 iref = (struct btrfs_extent_inline_ref *)ptr;
1574 btrfs_set_extent_inline_ref_type(leaf, iref, type);
1575 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1576 struct btrfs_extent_data_ref *dref;
1577 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1578 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1579 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1580 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1581 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1582 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1583 struct btrfs_shared_data_ref *sref;
1584 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1585 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1586 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1587 } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1588 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1589 } else {
1590 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1591 }
1592 btrfs_mark_buffer_dirty(leaf);
1593 return 0;
1594}
1595
1596static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1597 struct btrfs_root *root,
1598 struct btrfs_path *path,
1599 struct btrfs_extent_inline_ref **ref_ret,
1600 u64 bytenr, u64 num_bytes, u64 parent,
1601 u64 root_objectid, u64 owner, u64 offset)
1602{
1603 int ret;
1604
1605 ret = lookup_inline_extent_backref(trans, root, path, ref_ret,
1606 bytenr, num_bytes, parent,
1607 root_objectid, owner, offset, 0);
1608 if (ret != -ENOENT)
1609 return ret;
1610
1611 btrfs_release_path(root, path);
1612 *ref_ret = NULL;
1613
1614 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1615 ret = lookup_tree_block_ref(trans, root, path, bytenr, parent,
1616 root_objectid);
1617 } else {
1618 ret = lookup_extent_data_ref(trans, root, path, bytenr, parent,
1619 root_objectid, owner, offset);
1620 }
1621 return ret;
1622}
1623
1624/*
1625 * helper to update/remove inline back ref
1626 */
1627static noinline_for_stack
1628int update_inline_extent_backref(struct btrfs_trans_handle *trans,
1629 struct btrfs_root *root,
1630 struct btrfs_path *path,
1631 struct btrfs_extent_inline_ref *iref,
1632 int refs_to_mod,
1633 struct btrfs_delayed_extent_op *extent_op)
1634{
1635 struct extent_buffer *leaf;
1636 struct btrfs_extent_item *ei;
1637 struct btrfs_extent_data_ref *dref = NULL;
1638 struct btrfs_shared_data_ref *sref = NULL;
1639 unsigned long ptr;
1640 unsigned long end;
1641 u32 item_size;
1642 int size;
1643 int type;
1644 int ret;
1645 u64 refs;
1646
1647 leaf = path->nodes[0];
1648 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1649 refs = btrfs_extent_refs(leaf, ei);
1650 WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1651 refs += refs_to_mod;
1652 btrfs_set_extent_refs(leaf, ei, refs);
1653 if (extent_op)
1654 __run_delayed_extent_op(extent_op, leaf, ei);
1655
1656 type = btrfs_extent_inline_ref_type(leaf, iref);
1657
1658 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1659 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1660 refs = btrfs_extent_data_ref_count(leaf, dref);
1661 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1662 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1663 refs = btrfs_shared_data_ref_count(leaf, sref);
1664 } else {
1665 refs = 1;
1666 BUG_ON(refs_to_mod != -1);
1667 }
1668
1669 BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1670 refs += refs_to_mod;
1671
1672 if (refs > 0) {
1673 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1674 btrfs_set_extent_data_ref_count(leaf, dref, refs);
1675 else
1676 btrfs_set_shared_data_ref_count(leaf, sref, refs);
1677 } else {
1678 size = btrfs_extent_inline_ref_size(type);
1679 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1680 ptr = (unsigned long)iref;
1681 end = (unsigned long)ei + item_size;
1682 if (ptr + size < end)
1683 memmove_extent_buffer(leaf, ptr, ptr + size,
1684 end - ptr - size);
1685 item_size -= size;
1686 ret = btrfs_truncate_item(trans, root, path, item_size, 1);
1687 BUG_ON(ret);
1688 }
1689 btrfs_mark_buffer_dirty(leaf);
1690 return 0;
1691}
1692
1693static noinline_for_stack
1694int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1695 struct btrfs_root *root,
1696 struct btrfs_path *path,
1697 u64 bytenr, u64 num_bytes, u64 parent,
1698 u64 root_objectid, u64 owner,
1699 u64 offset, int refs_to_add,
1700 struct btrfs_delayed_extent_op *extent_op)
1701{
1702 struct btrfs_extent_inline_ref *iref;
1703 int ret;
1704
1705 ret = lookup_inline_extent_backref(trans, root, path, &iref,
1706 bytenr, num_bytes, parent,
1707 root_objectid, owner, offset, 1);
1708 if (ret == 0) {
1709 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID);
1710 ret = update_inline_extent_backref(trans, root, path, iref,
1711 refs_to_add, extent_op);
1712 } else if (ret == -ENOENT) {
1713 ret = setup_inline_extent_backref(trans, root, path, iref,
1714 parent, root_objectid,
1715 owner, offset, refs_to_add,
1716 extent_op);
1717 }
1718 return ret;
1719}
1720
1721static int insert_extent_backref(struct btrfs_trans_handle *trans,
1722 struct btrfs_root *root,
1723 struct btrfs_path *path,
1724 u64 bytenr, u64 parent, u64 root_objectid,
1725 u64 owner, u64 offset, int refs_to_add)
1726{
1727 int ret;
1728 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1729 BUG_ON(refs_to_add != 1);
1730 ret = insert_tree_block_ref(trans, root, path, bytenr,
1731 parent, root_objectid);
1732 } else {
1733 ret = insert_extent_data_ref(trans, root, path, bytenr,
1734 parent, root_objectid,
1735 owner, offset, refs_to_add);
1736 }
1737 return ret;
1738}
1739
1740static int remove_extent_backref(struct btrfs_trans_handle *trans,
1741 struct btrfs_root *root,
1742 struct btrfs_path *path,
1743 struct btrfs_extent_inline_ref *iref,
1744 int refs_to_drop, int is_data)
1745{
1746 int ret;
1747
1748 BUG_ON(!is_data && refs_to_drop != 1);
1749 if (iref) {
1750 ret = update_inline_extent_backref(trans, root, path, iref,
1751 -refs_to_drop, NULL);
1752 } else if (is_data) {
1753 ret = remove_extent_data_ref(trans, root, path, refs_to_drop);
1754 } else {
1755 ret = btrfs_del_item(trans, root, path);
1756 }
1757 return ret;
1758}
1759
Li Dongyang5378e602011-03-24 10:24:27 +00001760static int btrfs_issue_discard(struct block_device *bdev,
Chris Mason15916de2008-11-19 21:17:22 -05001761 u64 start, u64 len)
1762{
Li Dongyang5378e602011-03-24 10:24:27 +00001763 return blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_NOFS, 0);
Chris Mason15916de2008-11-19 21:17:22 -05001764}
Chris Mason15916de2008-11-19 21:17:22 -05001765
Liu Hui1f3c79a2009-01-05 15:57:51 -05001766static int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr,
Li Dongyang5378e602011-03-24 10:24:27 +00001767 u64 num_bytes, u64 *actual_bytes)
Liu Hui1f3c79a2009-01-05 15:57:51 -05001768{
Liu Hui1f3c79a2009-01-05 15:57:51 -05001769 int ret;
Li Dongyang5378e602011-03-24 10:24:27 +00001770 u64 discarded_bytes = 0;
Liu Hui1f3c79a2009-01-05 15:57:51 -05001771 struct btrfs_multi_bio *multi = NULL;
1772
Christoph Hellwige244a0a2009-10-14 09:24:59 -04001773
Liu Hui1f3c79a2009-01-05 15:57:51 -05001774 /* Tell the block device(s) that the sectors can be discarded */
Li Dongyang5378e602011-03-24 10:24:27 +00001775 ret = btrfs_map_block(&root->fs_info->mapping_tree, REQ_DISCARD,
1776 bytenr, &num_bytes, &multi, 0);
Liu Hui1f3c79a2009-01-05 15:57:51 -05001777 if (!ret) {
1778 struct btrfs_bio_stripe *stripe = multi->stripes;
1779 int i;
1780
Liu Hui1f3c79a2009-01-05 15:57:51 -05001781
1782 for (i = 0; i < multi->num_stripes; i++, stripe++) {
Li Dongyang5378e602011-03-24 10:24:27 +00001783 ret = btrfs_issue_discard(stripe->dev->bdev,
1784 stripe->physical,
1785 stripe->length);
1786 if (!ret)
1787 discarded_bytes += stripe->length;
1788 else if (ret != -EOPNOTSUPP)
1789 break;
Liu Hui1f3c79a2009-01-05 15:57:51 -05001790 }
1791 kfree(multi);
1792 }
Li Dongyang5378e602011-03-24 10:24:27 +00001793 if (discarded_bytes && ret == -EOPNOTSUPP)
1794 ret = 0;
1795
1796 if (actual_bytes)
1797 *actual_bytes = discarded_bytes;
1798
Liu Hui1f3c79a2009-01-05 15:57:51 -05001799
1800 return ret;
Liu Hui1f3c79a2009-01-05 15:57:51 -05001801}
1802
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001803int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1804 struct btrfs_root *root,
1805 u64 bytenr, u64 num_bytes, u64 parent,
1806 u64 root_objectid, u64 owner, u64 offset)
Zheng Yan31840ae2008-09-23 13:14:14 -04001807{
1808 int ret;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001809 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID &&
1810 root_objectid == BTRFS_TREE_LOG_OBJECTID);
Zheng Yan31840ae2008-09-23 13:14:14 -04001811
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001812 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1813 ret = btrfs_add_delayed_tree_ref(trans, bytenr, num_bytes,
1814 parent, root_objectid, (int)owner,
1815 BTRFS_ADD_DELAYED_REF, NULL);
1816 } else {
1817 ret = btrfs_add_delayed_data_ref(trans, bytenr, num_bytes,
1818 parent, root_objectid, owner, offset,
1819 BTRFS_ADD_DELAYED_REF, NULL);
1820 }
Zheng Yan31840ae2008-09-23 13:14:14 -04001821 return ret;
1822}
1823
Chris Mason925baed2008-06-25 16:01:30 -04001824static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001825 struct btrfs_root *root,
1826 u64 bytenr, u64 num_bytes,
1827 u64 parent, u64 root_objectid,
1828 u64 owner, u64 offset, int refs_to_add,
1829 struct btrfs_delayed_extent_op *extent_op)
Chris Mason56bec292009-03-13 10:10:06 -04001830{
Chris Mason5caf2a02007-04-02 11:20:42 -04001831 struct btrfs_path *path;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001832 struct extent_buffer *leaf;
Chris Mason234b63a2007-03-13 10:46:10 -04001833 struct btrfs_extent_item *item;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001834 u64 refs;
1835 int ret;
1836 int err = 0;
Chris Mason037e6392007-03-07 11:50:24 -05001837
Chris Mason5caf2a02007-04-02 11:20:42 -04001838 path = btrfs_alloc_path();
Chris Mason54aa1f42007-06-22 14:16:25 -04001839 if (!path)
1840 return -ENOMEM;
Chris Mason26b80032007-08-08 20:17:12 -04001841
Chris Mason3c12ac72008-04-21 12:01:38 -04001842 path->reada = 1;
Chris Masonb9473432009-03-13 11:00:37 -04001843 path->leave_spinning = 1;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001844 /* this will setup the path even if it fails to insert the back ref */
1845 ret = insert_inline_extent_backref(trans, root->fs_info->extent_root,
1846 path, bytenr, num_bytes, parent,
1847 root_objectid, owner, offset,
1848 refs_to_add, extent_op);
1849 if (ret == 0)
1850 goto out;
Zheng Yan31840ae2008-09-23 13:14:14 -04001851
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001852 if (ret != -EAGAIN) {
1853 err = ret;
1854 goto out;
Chris Masonb9473432009-03-13 11:00:37 -04001855 }
Zheng Yan31840ae2008-09-23 13:14:14 -04001856
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001857 leaf = path->nodes[0];
1858 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1859 refs = btrfs_extent_refs(leaf, item);
1860 btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
1861 if (extent_op)
1862 __run_delayed_extent_op(extent_op, leaf, item);
Zheng Yan31840ae2008-09-23 13:14:14 -04001863
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001864 btrfs_mark_buffer_dirty(leaf);
Chris Mason5caf2a02007-04-02 11:20:42 -04001865 btrfs_release_path(root->fs_info->extent_root, path);
Chris Mason7bb86312007-12-11 09:25:06 -05001866
Chris Mason3c12ac72008-04-21 12:01:38 -04001867 path->reada = 1;
Chris Masonb9473432009-03-13 11:00:37 -04001868 path->leave_spinning = 1;
1869
Chris Mason56bec292009-03-13 10:10:06 -04001870 /* now insert the actual backref */
Zheng Yan31840ae2008-09-23 13:14:14 -04001871 ret = insert_extent_backref(trans, root->fs_info->extent_root,
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001872 path, bytenr, parent, root_objectid,
1873 owner, offset, refs_to_add);
Chris Mason7bb86312007-12-11 09:25:06 -05001874 BUG_ON(ret);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001875out:
Chris Mason74493f72007-12-11 09:25:06 -05001876 btrfs_free_path(path);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001877 return err;
Chris Mason02217ed2007-03-02 16:08:05 -05001878}
1879
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001880static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
1881 struct btrfs_root *root,
1882 struct btrfs_delayed_ref_node *node,
1883 struct btrfs_delayed_extent_op *extent_op,
1884 int insert_reserved)
Chris Masone9d0b132007-08-10 14:06:19 -04001885{
Chris Mason56bec292009-03-13 10:10:06 -04001886 int ret = 0;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001887 struct btrfs_delayed_data_ref *ref;
1888 struct btrfs_key ins;
1889 u64 parent = 0;
1890 u64 ref_root = 0;
1891 u64 flags = 0;
Chris Mason56bec292009-03-13 10:10:06 -04001892
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001893 ins.objectid = node->bytenr;
1894 ins.offset = node->num_bytes;
1895 ins.type = BTRFS_EXTENT_ITEM_KEY;
Chris Mason56bec292009-03-13 10:10:06 -04001896
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001897 ref = btrfs_delayed_node_to_data_ref(node);
1898 if (node->type == BTRFS_SHARED_DATA_REF_KEY)
1899 parent = ref->parent;
1900 else
1901 ref_root = ref->root;
1902
1903 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
1904 if (extent_op) {
1905 BUG_ON(extent_op->update_key);
1906 flags |= extent_op->flags_to_set;
1907 }
1908 ret = alloc_reserved_file_extent(trans, root,
1909 parent, ref_root, flags,
1910 ref->objectid, ref->offset,
1911 &ins, node->ref_mod);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001912 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
1913 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
1914 node->num_bytes, parent,
1915 ref_root, ref->objectid,
1916 ref->offset, node->ref_mod,
1917 extent_op);
1918 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
1919 ret = __btrfs_free_extent(trans, root, node->bytenr,
1920 node->num_bytes, parent,
1921 ref_root, ref->objectid,
1922 ref->offset, node->ref_mod,
1923 extent_op);
1924 } else {
1925 BUG();
1926 }
Chris Mason56bec292009-03-13 10:10:06 -04001927 return ret;
1928}
1929
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001930static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
1931 struct extent_buffer *leaf,
1932 struct btrfs_extent_item *ei)
1933{
1934 u64 flags = btrfs_extent_flags(leaf, ei);
1935 if (extent_op->update_flags) {
1936 flags |= extent_op->flags_to_set;
1937 btrfs_set_extent_flags(leaf, ei, flags);
1938 }
1939
1940 if (extent_op->update_key) {
1941 struct btrfs_tree_block_info *bi;
1942 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
1943 bi = (struct btrfs_tree_block_info *)(ei + 1);
1944 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
1945 }
1946}
1947
1948static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
1949 struct btrfs_root *root,
1950 struct btrfs_delayed_ref_node *node,
1951 struct btrfs_delayed_extent_op *extent_op)
1952{
1953 struct btrfs_key key;
1954 struct btrfs_path *path;
1955 struct btrfs_extent_item *ei;
1956 struct extent_buffer *leaf;
1957 u32 item_size;
1958 int ret;
1959 int err = 0;
1960
1961 path = btrfs_alloc_path();
1962 if (!path)
1963 return -ENOMEM;
1964
1965 key.objectid = node->bytenr;
1966 key.type = BTRFS_EXTENT_ITEM_KEY;
1967 key.offset = node->num_bytes;
1968
1969 path->reada = 1;
1970 path->leave_spinning = 1;
1971 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key,
1972 path, 0, 1);
1973 if (ret < 0) {
1974 err = ret;
1975 goto out;
1976 }
1977 if (ret > 0) {
1978 err = -EIO;
1979 goto out;
1980 }
1981
1982 leaf = path->nodes[0];
1983 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1984#ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1985 if (item_size < sizeof(*ei)) {
1986 ret = convert_extent_item_v0(trans, root->fs_info->extent_root,
1987 path, (u64)-1, 0);
1988 if (ret < 0) {
1989 err = ret;
1990 goto out;
1991 }
1992 leaf = path->nodes[0];
1993 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1994 }
1995#endif
1996 BUG_ON(item_size < sizeof(*ei));
1997 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1998 __run_delayed_extent_op(extent_op, leaf, ei);
1999
2000 btrfs_mark_buffer_dirty(leaf);
2001out:
2002 btrfs_free_path(path);
2003 return err;
2004}
2005
2006static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
2007 struct btrfs_root *root,
2008 struct btrfs_delayed_ref_node *node,
2009 struct btrfs_delayed_extent_op *extent_op,
2010 int insert_reserved)
2011{
2012 int ret = 0;
2013 struct btrfs_delayed_tree_ref *ref;
2014 struct btrfs_key ins;
2015 u64 parent = 0;
2016 u64 ref_root = 0;
2017
2018 ins.objectid = node->bytenr;
2019 ins.offset = node->num_bytes;
2020 ins.type = BTRFS_EXTENT_ITEM_KEY;
2021
2022 ref = btrfs_delayed_node_to_tree_ref(node);
2023 if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2024 parent = ref->parent;
2025 else
2026 ref_root = ref->root;
2027
2028 BUG_ON(node->ref_mod != 1);
2029 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2030 BUG_ON(!extent_op || !extent_op->update_flags ||
2031 !extent_op->update_key);
2032 ret = alloc_reserved_tree_block(trans, root,
2033 parent, ref_root,
2034 extent_op->flags_to_set,
2035 &extent_op->key,
2036 ref->level, &ins);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002037 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2038 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
2039 node->num_bytes, parent, ref_root,
2040 ref->level, 0, 1, extent_op);
2041 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2042 ret = __btrfs_free_extent(trans, root, node->bytenr,
2043 node->num_bytes, parent, ref_root,
2044 ref->level, 0, 1, extent_op);
2045 } else {
2046 BUG();
2047 }
2048 return ret;
2049}
2050
Chris Mason56bec292009-03-13 10:10:06 -04002051/* helper function to actually process a single delayed ref entry */
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002052static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
2053 struct btrfs_root *root,
2054 struct btrfs_delayed_ref_node *node,
2055 struct btrfs_delayed_extent_op *extent_op,
2056 int insert_reserved)
Chris Mason56bec292009-03-13 10:10:06 -04002057{
Josef Bacikeb099672009-02-12 09:27:38 -05002058 int ret;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002059 if (btrfs_delayed_ref_is_head(node)) {
Chris Mason56bec292009-03-13 10:10:06 -04002060 struct btrfs_delayed_ref_head *head;
2061 /*
2062 * we've hit the end of the chain and we were supposed
2063 * to insert this extent into the tree. But, it got
2064 * deleted before we ever needed to insert it, so all
2065 * we have to do is clean up the accounting
2066 */
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002067 BUG_ON(extent_op);
2068 head = btrfs_delayed_node_to_head(node);
Chris Mason56bec292009-03-13 10:10:06 -04002069 if (insert_reserved) {
Yan, Zhengf0486c62010-05-16 10:46:25 -04002070 btrfs_pin_extent(root, node->bytenr,
2071 node->num_bytes, 1);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002072 if (head->is_data) {
2073 ret = btrfs_del_csums(trans, root,
2074 node->bytenr,
2075 node->num_bytes);
2076 BUG_ON(ret);
2077 }
Chris Mason56bec292009-03-13 10:10:06 -04002078 }
Chris Mason56bec292009-03-13 10:10:06 -04002079 mutex_unlock(&head->mutex);
2080 return 0;
2081 }
Josef Bacikeb099672009-02-12 09:27:38 -05002082
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002083 if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
2084 node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2085 ret = run_delayed_tree_ref(trans, root, node, extent_op,
2086 insert_reserved);
2087 else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
2088 node->type == BTRFS_SHARED_DATA_REF_KEY)
2089 ret = run_delayed_data_ref(trans, root, node, extent_op,
2090 insert_reserved);
2091 else
2092 BUG();
2093 return ret;
Chris Masone9d0b132007-08-10 14:06:19 -04002094}
2095
Chris Mason56bec292009-03-13 10:10:06 -04002096static noinline struct btrfs_delayed_ref_node *
2097select_delayed_ref(struct btrfs_delayed_ref_head *head)
Chris Masona28ec192007-03-06 20:08:01 -05002098{
Chris Mason56bec292009-03-13 10:10:06 -04002099 struct rb_node *node;
2100 struct btrfs_delayed_ref_node *ref;
2101 int action = BTRFS_ADD_DELAYED_REF;
2102again:
2103 /*
2104 * select delayed ref of type BTRFS_ADD_DELAYED_REF first.
2105 * this prevents ref count from going down to zero when
2106 * there still are pending delayed ref.
2107 */
2108 node = rb_prev(&head->node.rb_node);
2109 while (1) {
2110 if (!node)
2111 break;
2112 ref = rb_entry(node, struct btrfs_delayed_ref_node,
2113 rb_node);
2114 if (ref->bytenr != head->node.bytenr)
2115 break;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002116 if (ref->action == action)
Chris Mason56bec292009-03-13 10:10:06 -04002117 return ref;
2118 node = rb_prev(node);
Chris Mason5f39d392007-10-15 16:14:19 -04002119 }
Chris Mason56bec292009-03-13 10:10:06 -04002120 if (action == BTRFS_ADD_DELAYED_REF) {
2121 action = BTRFS_DROP_DELAYED_REF;
2122 goto again;
2123 }
2124 return NULL;
2125}
2126
Chris Masonc3e69d52009-03-13 10:17:05 -04002127static noinline int run_clustered_refs(struct btrfs_trans_handle *trans,
2128 struct btrfs_root *root,
2129 struct list_head *cluster)
Chris Mason56bec292009-03-13 10:10:06 -04002130{
Chris Mason56bec292009-03-13 10:10:06 -04002131 struct btrfs_delayed_ref_root *delayed_refs;
2132 struct btrfs_delayed_ref_node *ref;
2133 struct btrfs_delayed_ref_head *locked_ref = NULL;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002134 struct btrfs_delayed_extent_op *extent_op;
Chris Mason56bec292009-03-13 10:10:06 -04002135 int ret;
Chris Masonc3e69d52009-03-13 10:17:05 -04002136 int count = 0;
Chris Mason56bec292009-03-13 10:10:06 -04002137 int must_insert_reserved = 0;
Chris Mason56bec292009-03-13 10:10:06 -04002138
2139 delayed_refs = &trans->transaction->delayed_refs;
Chris Mason56bec292009-03-13 10:10:06 -04002140 while (1) {
2141 if (!locked_ref) {
Chris Masonc3e69d52009-03-13 10:17:05 -04002142 /* pick a new head ref from the cluster list */
2143 if (list_empty(cluster))
Chris Mason56bec292009-03-13 10:10:06 -04002144 break;
Chris Mason56bec292009-03-13 10:10:06 -04002145
Chris Masonc3e69d52009-03-13 10:17:05 -04002146 locked_ref = list_entry(cluster->next,
2147 struct btrfs_delayed_ref_head, cluster);
2148
2149 /* grab the lock that says we are going to process
2150 * all the refs for this head */
2151 ret = btrfs_delayed_ref_lock(trans, locked_ref);
2152
2153 /*
2154 * we may have dropped the spin lock to get the head
2155 * mutex lock, and that might have given someone else
2156 * time to free the head. If that's true, it has been
2157 * removed from our list and we can move on.
2158 */
2159 if (ret == -EAGAIN) {
2160 locked_ref = NULL;
2161 count++;
2162 continue;
Chris Mason56bec292009-03-13 10:10:06 -04002163 }
2164 }
2165
2166 /*
2167 * record the must insert reserved flag before we
2168 * drop the spin lock.
2169 */
2170 must_insert_reserved = locked_ref->must_insert_reserved;
2171 locked_ref->must_insert_reserved = 0;
2172
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002173 extent_op = locked_ref->extent_op;
2174 locked_ref->extent_op = NULL;
2175
Chris Mason56bec292009-03-13 10:10:06 -04002176 /*
2177 * locked_ref is the head node, so we have to go one
2178 * node back for any delayed ref updates
2179 */
Chris Mason56bec292009-03-13 10:10:06 -04002180 ref = select_delayed_ref(locked_ref);
2181 if (!ref) {
2182 /* All delayed refs have been processed, Go ahead
2183 * and send the head node to run_one_delayed_ref,
2184 * so that any accounting fixes can happen
2185 */
2186 ref = &locked_ref->node;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002187
2188 if (extent_op && must_insert_reserved) {
2189 kfree(extent_op);
2190 extent_op = NULL;
2191 }
2192
2193 if (extent_op) {
2194 spin_unlock(&delayed_refs->lock);
2195
2196 ret = run_delayed_extent_op(trans, root,
2197 ref, extent_op);
2198 BUG_ON(ret);
2199 kfree(extent_op);
2200
2201 cond_resched();
2202 spin_lock(&delayed_refs->lock);
2203 continue;
2204 }
2205
Chris Masonc3e69d52009-03-13 10:17:05 -04002206 list_del_init(&locked_ref->cluster);
Chris Mason56bec292009-03-13 10:10:06 -04002207 locked_ref = NULL;
2208 }
2209
2210 ref->in_tree = 0;
2211 rb_erase(&ref->rb_node, &delayed_refs->root);
2212 delayed_refs->num_entries--;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002213
Chris Mason56bec292009-03-13 10:10:06 -04002214 spin_unlock(&delayed_refs->lock);
2215
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002216 ret = run_one_delayed_ref(trans, root, ref, extent_op,
Chris Mason56bec292009-03-13 10:10:06 -04002217 must_insert_reserved);
2218 BUG_ON(ret);
Chris Mason56bec292009-03-13 10:10:06 -04002219
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002220 btrfs_put_delayed_ref(ref);
2221 kfree(extent_op);
Chris Masonc3e69d52009-03-13 10:17:05 -04002222 count++;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002223
Chris Mason1887be62009-03-13 10:11:24 -04002224 cond_resched();
Chris Mason56bec292009-03-13 10:10:06 -04002225 spin_lock(&delayed_refs->lock);
2226 }
Chris Masonc3e69d52009-03-13 10:17:05 -04002227 return count;
2228}
2229
2230/*
2231 * this starts processing the delayed reference count updates and
2232 * extent insertions we have queued up so far. count can be
2233 * 0, which means to process everything in the tree at the start
2234 * of the run (but not newly added entries), or it can be some target
2235 * number you'd like to process.
2236 */
2237int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2238 struct btrfs_root *root, unsigned long count)
2239{
2240 struct rb_node *node;
2241 struct btrfs_delayed_ref_root *delayed_refs;
2242 struct btrfs_delayed_ref_node *ref;
2243 struct list_head cluster;
2244 int ret;
2245 int run_all = count == (unsigned long)-1;
2246 int run_most = 0;
2247
2248 if (root == root->fs_info->extent_root)
2249 root = root->fs_info->tree_root;
2250
2251 delayed_refs = &trans->transaction->delayed_refs;
2252 INIT_LIST_HEAD(&cluster);
2253again:
2254 spin_lock(&delayed_refs->lock);
2255 if (count == 0) {
2256 count = delayed_refs->num_entries * 2;
2257 run_most = 1;
2258 }
2259 while (1) {
2260 if (!(run_all || run_most) &&
2261 delayed_refs->num_heads_ready < 64)
2262 break;
2263
2264 /*
2265 * go find something we can process in the rbtree. We start at
2266 * the beginning of the tree, and then build a cluster
2267 * of refs to process starting at the first one we are able to
2268 * lock
2269 */
2270 ret = btrfs_find_ref_cluster(trans, &cluster,
2271 delayed_refs->run_delayed_start);
2272 if (ret)
2273 break;
2274
2275 ret = run_clustered_refs(trans, root, &cluster);
2276 BUG_ON(ret < 0);
2277
2278 count -= min_t(unsigned long, ret, count);
2279
2280 if (count == 0)
2281 break;
2282 }
2283
Chris Mason56bec292009-03-13 10:10:06 -04002284 if (run_all) {
Chris Mason56bec292009-03-13 10:10:06 -04002285 node = rb_first(&delayed_refs->root);
Chris Masonc3e69d52009-03-13 10:17:05 -04002286 if (!node)
Chris Mason56bec292009-03-13 10:10:06 -04002287 goto out;
Chris Masonc3e69d52009-03-13 10:17:05 -04002288 count = (unsigned long)-1;
Chris Mason56bec292009-03-13 10:10:06 -04002289
2290 while (node) {
2291 ref = rb_entry(node, struct btrfs_delayed_ref_node,
2292 rb_node);
2293 if (btrfs_delayed_ref_is_head(ref)) {
2294 struct btrfs_delayed_ref_head *head;
2295
2296 head = btrfs_delayed_node_to_head(ref);
2297 atomic_inc(&ref->refs);
2298
2299 spin_unlock(&delayed_refs->lock);
2300 mutex_lock(&head->mutex);
2301 mutex_unlock(&head->mutex);
2302
2303 btrfs_put_delayed_ref(ref);
Chris Mason1887be62009-03-13 10:11:24 -04002304 cond_resched();
Chris Mason56bec292009-03-13 10:10:06 -04002305 goto again;
2306 }
2307 node = rb_next(node);
2308 }
2309 spin_unlock(&delayed_refs->lock);
Chris Mason56bec292009-03-13 10:10:06 -04002310 schedule_timeout(1);
2311 goto again;
2312 }
Chris Mason54aa1f42007-06-22 14:16:25 -04002313out:
Chris Masonc3e69d52009-03-13 10:17:05 -04002314 spin_unlock(&delayed_refs->lock);
Chris Masona28ec192007-03-06 20:08:01 -05002315 return 0;
2316}
2317
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002318int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
2319 struct btrfs_root *root,
2320 u64 bytenr, u64 num_bytes, u64 flags,
2321 int is_data)
2322{
2323 struct btrfs_delayed_extent_op *extent_op;
2324 int ret;
2325
2326 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
2327 if (!extent_op)
2328 return -ENOMEM;
2329
2330 extent_op->flags_to_set = flags;
2331 extent_op->update_flags = 1;
2332 extent_op->update_key = 0;
2333 extent_op->is_data = is_data ? 1 : 0;
2334
2335 ret = btrfs_add_delayed_extent_op(trans, bytenr, num_bytes, extent_op);
2336 if (ret)
2337 kfree(extent_op);
2338 return ret;
2339}
2340
2341static noinline int check_delayed_ref(struct btrfs_trans_handle *trans,
2342 struct btrfs_root *root,
2343 struct btrfs_path *path,
2344 u64 objectid, u64 offset, u64 bytenr)
2345{
2346 struct btrfs_delayed_ref_head *head;
2347 struct btrfs_delayed_ref_node *ref;
2348 struct btrfs_delayed_data_ref *data_ref;
2349 struct btrfs_delayed_ref_root *delayed_refs;
2350 struct rb_node *node;
2351 int ret = 0;
2352
2353 ret = -ENOENT;
2354 delayed_refs = &trans->transaction->delayed_refs;
2355 spin_lock(&delayed_refs->lock);
2356 head = btrfs_find_delayed_ref_head(trans, bytenr);
2357 if (!head)
2358 goto out;
2359
2360 if (!mutex_trylock(&head->mutex)) {
2361 atomic_inc(&head->node.refs);
2362 spin_unlock(&delayed_refs->lock);
2363
2364 btrfs_release_path(root->fs_info->extent_root, path);
2365
2366 mutex_lock(&head->mutex);
2367 mutex_unlock(&head->mutex);
2368 btrfs_put_delayed_ref(&head->node);
2369 return -EAGAIN;
2370 }
2371
2372 node = rb_prev(&head->node.rb_node);
2373 if (!node)
2374 goto out_unlock;
2375
2376 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2377
2378 if (ref->bytenr != bytenr)
2379 goto out_unlock;
2380
2381 ret = 1;
2382 if (ref->type != BTRFS_EXTENT_DATA_REF_KEY)
2383 goto out_unlock;
2384
2385 data_ref = btrfs_delayed_node_to_data_ref(ref);
2386
2387 node = rb_prev(node);
2388 if (node) {
2389 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2390 if (ref->bytenr == bytenr)
2391 goto out_unlock;
2392 }
2393
2394 if (data_ref->root != root->root_key.objectid ||
2395 data_ref->objectid != objectid || data_ref->offset != offset)
2396 goto out_unlock;
2397
2398 ret = 0;
2399out_unlock:
2400 mutex_unlock(&head->mutex);
2401out:
2402 spin_unlock(&delayed_refs->lock);
2403 return ret;
2404}
2405
2406static noinline int check_committed_ref(struct btrfs_trans_handle *trans,
2407 struct btrfs_root *root,
2408 struct btrfs_path *path,
2409 u64 objectid, u64 offset, u64 bytenr)
Chris Masonbe20aa92007-12-17 20:14:01 -05002410{
2411 struct btrfs_root *extent_root = root->fs_info->extent_root;
Yan Zhengf321e492008-07-30 09:26:11 -04002412 struct extent_buffer *leaf;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002413 struct btrfs_extent_data_ref *ref;
2414 struct btrfs_extent_inline_ref *iref;
2415 struct btrfs_extent_item *ei;
Chris Masonbe20aa92007-12-17 20:14:01 -05002416 struct btrfs_key key;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002417 u32 item_size;
Yan Zhengf321e492008-07-30 09:26:11 -04002418 int ret;
Chris Masonbe20aa92007-12-17 20:14:01 -05002419
Chris Masonbe20aa92007-12-17 20:14:01 -05002420 key.objectid = bytenr;
Zheng Yan31840ae2008-09-23 13:14:14 -04002421 key.offset = (u64)-1;
Yan Zhengf321e492008-07-30 09:26:11 -04002422 key.type = BTRFS_EXTENT_ITEM_KEY;
Chris Masonbe20aa92007-12-17 20:14:01 -05002423
Chris Masonbe20aa92007-12-17 20:14:01 -05002424 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2425 if (ret < 0)
2426 goto out;
2427 BUG_ON(ret == 0);
Yan Zheng80ff3852008-10-30 14:20:02 -04002428
2429 ret = -ENOENT;
2430 if (path->slots[0] == 0)
Zheng Yan31840ae2008-09-23 13:14:14 -04002431 goto out;
Chris Masonbe20aa92007-12-17 20:14:01 -05002432
Zheng Yan31840ae2008-09-23 13:14:14 -04002433 path->slots[0]--;
Yan Zhengf321e492008-07-30 09:26:11 -04002434 leaf = path->nodes[0];
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002435 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
Chris Masonbe20aa92007-12-17 20:14:01 -05002436
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002437 if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
Chris Masonbe20aa92007-12-17 20:14:01 -05002438 goto out;
Chris Masonbe20aa92007-12-17 20:14:01 -05002439
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002440 ret = 1;
2441 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2442#ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2443 if (item_size < sizeof(*ei)) {
2444 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
2445 goto out;
Chris Masonbe20aa92007-12-17 20:14:01 -05002446 }
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002447#endif
2448 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2449
2450 if (item_size != sizeof(*ei) +
2451 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
2452 goto out;
2453
2454 if (btrfs_extent_generation(leaf, ei) <=
2455 btrfs_root_last_snapshot(&root->root_item))
2456 goto out;
2457
2458 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
2459 if (btrfs_extent_inline_ref_type(leaf, iref) !=
2460 BTRFS_EXTENT_DATA_REF_KEY)
2461 goto out;
2462
2463 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
2464 if (btrfs_extent_refs(leaf, ei) !=
2465 btrfs_extent_data_ref_count(leaf, ref) ||
2466 btrfs_extent_data_ref_root(leaf, ref) !=
2467 root->root_key.objectid ||
2468 btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
2469 btrfs_extent_data_ref_offset(leaf, ref) != offset)
2470 goto out;
2471
Yan Zhengf321e492008-07-30 09:26:11 -04002472 ret = 0;
Chris Masonbe20aa92007-12-17 20:14:01 -05002473out:
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002474 return ret;
2475}
2476
2477int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
2478 struct btrfs_root *root,
2479 u64 objectid, u64 offset, u64 bytenr)
2480{
2481 struct btrfs_path *path;
2482 int ret;
2483 int ret2;
2484
2485 path = btrfs_alloc_path();
2486 if (!path)
2487 return -ENOENT;
2488
2489 do {
2490 ret = check_committed_ref(trans, root, path, objectid,
2491 offset, bytenr);
2492 if (ret && ret != -ENOENT)
2493 goto out;
2494
2495 ret2 = check_delayed_ref(trans, root, path, objectid,
2496 offset, bytenr);
2497 } while (ret2 == -EAGAIN);
2498
2499 if (ret2 && ret2 != -ENOENT) {
2500 ret = ret2;
2501 goto out;
2502 }
2503
2504 if (ret != -ENOENT || ret2 != -ENOENT)
2505 ret = 0;
2506out:
Yan Zhengf321e492008-07-30 09:26:11 -04002507 btrfs_free_path(path);
Yan, Zhengf0486c62010-05-16 10:46:25 -04002508 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
2509 WARN_ON(ret > 0);
Yan Zhengf321e492008-07-30 09:26:11 -04002510 return ret;
2511}
2512
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002513#if 0
Zheng Yan31840ae2008-09-23 13:14:14 -04002514int btrfs_cache_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2515 struct extent_buffer *buf, u32 nr_extents)
Chris Mason02217ed2007-03-02 16:08:05 -05002516{
Chris Mason5f39d392007-10-15 16:14:19 -04002517 struct btrfs_key key;
Chris Mason6407bf62007-03-27 06:33:00 -04002518 struct btrfs_file_extent_item *fi;
Zheng Yane4657682008-09-26 10:04:53 -04002519 u64 root_gen;
2520 u32 nritems;
Chris Mason02217ed2007-03-02 16:08:05 -05002521 int i;
Chris Masondb945352007-10-15 16:15:53 -04002522 int level;
Zheng Yan31840ae2008-09-23 13:14:14 -04002523 int ret = 0;
Zheng Yane4657682008-09-26 10:04:53 -04002524 int shared = 0;
Chris Masona28ec192007-03-06 20:08:01 -05002525
Chris Mason3768f362007-03-13 16:47:54 -04002526 if (!root->ref_cows)
Chris Masona28ec192007-03-06 20:08:01 -05002527 return 0;
Chris Mason5f39d392007-10-15 16:14:19 -04002528
Zheng Yane4657682008-09-26 10:04:53 -04002529 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
2530 shared = 0;
2531 root_gen = root->root_key.offset;
2532 } else {
2533 shared = 1;
2534 root_gen = trans->transid - 1;
2535 }
2536
Chris Masondb945352007-10-15 16:15:53 -04002537 level = btrfs_header_level(buf);
Chris Mason5f39d392007-10-15 16:14:19 -04002538 nritems = btrfs_header_nritems(buf);
Chris Mason4a096752008-07-21 10:29:44 -04002539
Zheng Yan31840ae2008-09-23 13:14:14 -04002540 if (level == 0) {
Yan Zheng31153d82008-07-28 15:32:19 -04002541 struct btrfs_leaf_ref *ref;
2542 struct btrfs_extent_info *info;
2543
Zheng Yan31840ae2008-09-23 13:14:14 -04002544 ref = btrfs_alloc_leaf_ref(root, nr_extents);
Yan Zheng31153d82008-07-28 15:32:19 -04002545 if (!ref) {
Zheng Yan31840ae2008-09-23 13:14:14 -04002546 ret = -ENOMEM;
Yan Zheng31153d82008-07-28 15:32:19 -04002547 goto out;
2548 }
2549
Zheng Yane4657682008-09-26 10:04:53 -04002550 ref->root_gen = root_gen;
Yan Zheng31153d82008-07-28 15:32:19 -04002551 ref->bytenr = buf->start;
2552 ref->owner = btrfs_header_owner(buf);
2553 ref->generation = btrfs_header_generation(buf);
Zheng Yan31840ae2008-09-23 13:14:14 -04002554 ref->nritems = nr_extents;
Yan Zheng31153d82008-07-28 15:32:19 -04002555 info = ref->extents;
Yanbcc63ab2008-07-30 16:29:20 -04002556
Zheng Yan31840ae2008-09-23 13:14:14 -04002557 for (i = 0; nr_extents > 0 && i < nritems; i++) {
Yan Zheng31153d82008-07-28 15:32:19 -04002558 u64 disk_bytenr;
2559 btrfs_item_key_to_cpu(buf, &key, i);
2560 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2561 continue;
2562 fi = btrfs_item_ptr(buf, i,
2563 struct btrfs_file_extent_item);
2564 if (btrfs_file_extent_type(buf, fi) ==
2565 BTRFS_FILE_EXTENT_INLINE)
2566 continue;
2567 disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
2568 if (disk_bytenr == 0)
2569 continue;
2570
2571 info->bytenr = disk_bytenr;
2572 info->num_bytes =
2573 btrfs_file_extent_disk_num_bytes(buf, fi);
2574 info->objectid = key.objectid;
2575 info->offset = key.offset;
2576 info++;
2577 }
2578
Zheng Yane4657682008-09-26 10:04:53 -04002579 ret = btrfs_add_leaf_ref(root, ref, shared);
Yan Zheng5b84e8d2008-10-09 11:46:19 -04002580 if (ret == -EEXIST && shared) {
2581 struct btrfs_leaf_ref *old;
2582 old = btrfs_lookup_leaf_ref(root, ref->bytenr);
2583 BUG_ON(!old);
2584 btrfs_remove_leaf_ref(root, old);
2585 btrfs_free_leaf_ref(root, old);
2586 ret = btrfs_add_leaf_ref(root, ref, shared);
2587 }
Yan Zheng31153d82008-07-28 15:32:19 -04002588 WARN_ON(ret);
Yanbcc63ab2008-07-30 16:29:20 -04002589 btrfs_free_leaf_ref(root, ref);
Yan Zheng31153d82008-07-28 15:32:19 -04002590 }
2591out:
Zheng Yan31840ae2008-09-23 13:14:14 -04002592 return ret;
2593}
2594
Chris Masonb7a9f292009-02-04 09:23:45 -05002595/* when a block goes through cow, we update the reference counts of
2596 * everything that block points to. The internal pointers of the block
2597 * can be in just about any order, and it is likely to have clusters of
2598 * things that are close together and clusters of things that are not.
2599 *
2600 * To help reduce the seeks that come with updating all of these reference
2601 * counts, sort them by byte number before actual updates are done.
2602 *
2603 * struct refsort is used to match byte number to slot in the btree block.
2604 * we sort based on the byte number and then use the slot to actually
2605 * find the item.
Chris Masonbd56b302009-02-04 09:27:02 -05002606 *
2607 * struct refsort is smaller than strcut btrfs_item and smaller than
2608 * struct btrfs_key_ptr. Since we're currently limited to the page size
2609 * for a btree block, there's no way for a kmalloc of refsorts for a
2610 * single node to be bigger than a page.
Chris Masonb7a9f292009-02-04 09:23:45 -05002611 */
2612struct refsort {
2613 u64 bytenr;
2614 u32 slot;
2615};
2616
2617/*
2618 * for passing into sort()
2619 */
2620static int refsort_cmp(const void *a_void, const void *b_void)
2621{
2622 const struct refsort *a = a_void;
2623 const struct refsort *b = b_void;
2624
2625 if (a->bytenr < b->bytenr)
2626 return -1;
2627 if (a->bytenr > b->bytenr)
2628 return 1;
2629 return 0;
2630}
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002631#endif
Chris Masonb7a9f292009-02-04 09:23:45 -05002632
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002633static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
Chris Masonb7a9f292009-02-04 09:23:45 -05002634 struct btrfs_root *root,
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002635 struct extent_buffer *buf,
2636 int full_backref, int inc)
Zheng Yan31840ae2008-09-23 13:14:14 -04002637{
2638 u64 bytenr;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002639 u64 num_bytes;
2640 u64 parent;
Zheng Yan31840ae2008-09-23 13:14:14 -04002641 u64 ref_root;
Zheng Yan31840ae2008-09-23 13:14:14 -04002642 u32 nritems;
Zheng Yan31840ae2008-09-23 13:14:14 -04002643 struct btrfs_key key;
2644 struct btrfs_file_extent_item *fi;
2645 int i;
2646 int level;
2647 int ret = 0;
Zheng Yan31840ae2008-09-23 13:14:14 -04002648 int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002649 u64, u64, u64, u64, u64, u64);
Zheng Yan31840ae2008-09-23 13:14:14 -04002650
2651 ref_root = btrfs_header_owner(buf);
Zheng Yan31840ae2008-09-23 13:14:14 -04002652 nritems = btrfs_header_nritems(buf);
2653 level = btrfs_header_level(buf);
2654
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002655 if (!root->ref_cows && level == 0)
2656 return 0;
Chris Masonb7a9f292009-02-04 09:23:45 -05002657
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002658 if (inc)
2659 process_func = btrfs_inc_extent_ref;
2660 else
2661 process_func = btrfs_free_extent;
Zheng Yan31840ae2008-09-23 13:14:14 -04002662
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002663 if (full_backref)
2664 parent = buf->start;
2665 else
2666 parent = 0;
2667
Zheng Yan31840ae2008-09-23 13:14:14 -04002668 for (i = 0; i < nritems; i++) {
Chris Masondb945352007-10-15 16:15:53 -04002669 if (level == 0) {
Chris Mason5f39d392007-10-15 16:14:19 -04002670 btrfs_item_key_to_cpu(buf, &key, i);
2671 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
Chris Mason54aa1f42007-06-22 14:16:25 -04002672 continue;
Chris Mason5f39d392007-10-15 16:14:19 -04002673 fi = btrfs_item_ptr(buf, i,
Chris Mason54aa1f42007-06-22 14:16:25 -04002674 struct btrfs_file_extent_item);
Chris Mason5f39d392007-10-15 16:14:19 -04002675 if (btrfs_file_extent_type(buf, fi) ==
Chris Mason54aa1f42007-06-22 14:16:25 -04002676 BTRFS_FILE_EXTENT_INLINE)
2677 continue;
Zheng Yan31840ae2008-09-23 13:14:14 -04002678 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
2679 if (bytenr == 0)
Chris Mason54aa1f42007-06-22 14:16:25 -04002680 continue;
Zheng Yan31840ae2008-09-23 13:14:14 -04002681
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002682 num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
2683 key.offset -= btrfs_file_extent_offset(buf, fi);
2684 ret = process_func(trans, root, bytenr, num_bytes,
2685 parent, ref_root, key.objectid,
2686 key.offset);
2687 if (ret)
2688 goto fail;
Chris Masonb7a9f292009-02-04 09:23:45 -05002689 } else {
2690 bytenr = btrfs_node_blockptr(buf, i);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002691 num_bytes = btrfs_level_size(root, level - 1);
2692 ret = process_func(trans, root, bytenr, num_bytes,
2693 parent, ref_root, level - 1, 0);
2694 if (ret)
Zheng Yan31840ae2008-09-23 13:14:14 -04002695 goto fail;
Chris Mason54aa1f42007-06-22 14:16:25 -04002696 }
2697 }
Zheng Yan31840ae2008-09-23 13:14:14 -04002698 return 0;
2699fail:
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002700 BUG();
Chris Mason54aa1f42007-06-22 14:16:25 -04002701 return ret;
Chris Mason02217ed2007-03-02 16:08:05 -05002702}
2703
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002704int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2705 struct extent_buffer *buf, int full_backref)
Zheng Yan31840ae2008-09-23 13:14:14 -04002706{
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002707 return __btrfs_mod_ref(trans, root, buf, full_backref, 1);
2708}
Zheng Yan31840ae2008-09-23 13:14:14 -04002709
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002710int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2711 struct extent_buffer *buf, int full_backref)
2712{
2713 return __btrfs_mod_ref(trans, root, buf, full_backref, 0);
Zheng Yan31840ae2008-09-23 13:14:14 -04002714}
2715
Chris Mason9078a3e2007-04-26 16:46:15 -04002716static int write_one_cache_group(struct btrfs_trans_handle *trans,
2717 struct btrfs_root *root,
2718 struct btrfs_path *path,
2719 struct btrfs_block_group_cache *cache)
2720{
2721 int ret;
Chris Mason9078a3e2007-04-26 16:46:15 -04002722 struct btrfs_root *extent_root = root->fs_info->extent_root;
Chris Mason5f39d392007-10-15 16:14:19 -04002723 unsigned long bi;
2724 struct extent_buffer *leaf;
Chris Mason9078a3e2007-04-26 16:46:15 -04002725
Chris Mason9078a3e2007-04-26 16:46:15 -04002726 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
Chris Mason54aa1f42007-06-22 14:16:25 -04002727 if (ret < 0)
2728 goto fail;
Chris Mason9078a3e2007-04-26 16:46:15 -04002729 BUG_ON(ret);
Chris Mason5f39d392007-10-15 16:14:19 -04002730
2731 leaf = path->nodes[0];
2732 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
2733 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
2734 btrfs_mark_buffer_dirty(leaf);
Chris Mason9078a3e2007-04-26 16:46:15 -04002735 btrfs_release_path(extent_root, path);
Chris Mason54aa1f42007-06-22 14:16:25 -04002736fail:
Chris Mason9078a3e2007-04-26 16:46:15 -04002737 if (ret)
2738 return ret;
Chris Mason9078a3e2007-04-26 16:46:15 -04002739 return 0;
2740
2741}
2742
Yan Zheng4a8c9a62009-07-22 10:07:05 -04002743static struct btrfs_block_group_cache *
2744next_block_group(struct btrfs_root *root,
2745 struct btrfs_block_group_cache *cache)
2746{
2747 struct rb_node *node;
2748 spin_lock(&root->fs_info->block_group_cache_lock);
2749 node = rb_next(&cache->cache_node);
2750 btrfs_put_block_group(cache);
2751 if (node) {
2752 cache = rb_entry(node, struct btrfs_block_group_cache,
2753 cache_node);
Josef Bacik11dfe352009-11-13 20:12:59 +00002754 btrfs_get_block_group(cache);
Yan Zheng4a8c9a62009-07-22 10:07:05 -04002755 } else
2756 cache = NULL;
2757 spin_unlock(&root->fs_info->block_group_cache_lock);
2758 return cache;
2759}
2760
Josef Bacik0af3d002010-06-21 14:48:16 -04002761static int cache_save_setup(struct btrfs_block_group_cache *block_group,
2762 struct btrfs_trans_handle *trans,
2763 struct btrfs_path *path)
2764{
2765 struct btrfs_root *root = block_group->fs_info->tree_root;
2766 struct inode *inode = NULL;
2767 u64 alloc_hint = 0;
Josef Bacik2b209822010-12-03 13:17:53 -05002768 int dcs = BTRFS_DC_ERROR;
Josef Bacik0af3d002010-06-21 14:48:16 -04002769 int num_pages = 0;
2770 int retries = 0;
2771 int ret = 0;
2772
2773 /*
2774 * If this block group is smaller than 100 megs don't bother caching the
2775 * block group.
2776 */
2777 if (block_group->key.offset < (100 * 1024 * 1024)) {
2778 spin_lock(&block_group->lock);
2779 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
2780 spin_unlock(&block_group->lock);
2781 return 0;
2782 }
2783
2784again:
2785 inode = lookup_free_space_inode(root, block_group, path);
2786 if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
2787 ret = PTR_ERR(inode);
2788 btrfs_release_path(root, path);
2789 goto out;
2790 }
2791
2792 if (IS_ERR(inode)) {
2793 BUG_ON(retries);
2794 retries++;
2795
2796 if (block_group->ro)
2797 goto out_free;
2798
2799 ret = create_free_space_inode(root, trans, block_group, path);
2800 if (ret)
2801 goto out_free;
2802 goto again;
2803 }
2804
2805 /*
2806 * We want to set the generation to 0, that way if anything goes wrong
2807 * from here on out we know not to trust this cache when we load up next
2808 * time.
2809 */
2810 BTRFS_I(inode)->generation = 0;
2811 ret = btrfs_update_inode(trans, root, inode);
2812 WARN_ON(ret);
2813
2814 if (i_size_read(inode) > 0) {
2815 ret = btrfs_truncate_free_space_cache(root, trans, path,
2816 inode);
2817 if (ret)
2818 goto out_put;
2819 }
2820
2821 spin_lock(&block_group->lock);
2822 if (block_group->cached != BTRFS_CACHE_FINISHED) {
Josef Bacik2b209822010-12-03 13:17:53 -05002823 /* We're not cached, don't bother trying to write stuff out */
2824 dcs = BTRFS_DC_WRITTEN;
Josef Bacik0af3d002010-06-21 14:48:16 -04002825 spin_unlock(&block_group->lock);
2826 goto out_put;
2827 }
2828 spin_unlock(&block_group->lock);
2829
2830 num_pages = (int)div64_u64(block_group->key.offset, 1024 * 1024 * 1024);
2831 if (!num_pages)
2832 num_pages = 1;
2833
2834 /*
2835 * Just to make absolutely sure we have enough space, we're going to
2836 * preallocate 12 pages worth of space for each block group. In
2837 * practice we ought to use at most 8, but we need extra space so we can
2838 * add our header and have a terminator between the extents and the
2839 * bitmaps.
2840 */
2841 num_pages *= 16;
2842 num_pages *= PAGE_CACHE_SIZE;
2843
2844 ret = btrfs_check_data_free_space(inode, num_pages);
2845 if (ret)
2846 goto out_put;
2847
2848 ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages,
2849 num_pages, num_pages,
2850 &alloc_hint);
Josef Bacik2b209822010-12-03 13:17:53 -05002851 if (!ret)
2852 dcs = BTRFS_DC_SETUP;
Josef Bacik0af3d002010-06-21 14:48:16 -04002853 btrfs_free_reserved_data_space(inode, num_pages);
2854out_put:
2855 iput(inode);
2856out_free:
2857 btrfs_release_path(root, path);
2858out:
2859 spin_lock(&block_group->lock);
Josef Bacik2b209822010-12-03 13:17:53 -05002860 block_group->disk_cache_state = dcs;
Josef Bacik0af3d002010-06-21 14:48:16 -04002861 spin_unlock(&block_group->lock);
2862
2863 return ret;
2864}
2865
Chris Mason96b51792007-10-15 16:15:19 -04002866int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
2867 struct btrfs_root *root)
Chris Mason9078a3e2007-04-26 16:46:15 -04002868{
Yan Zheng4a8c9a62009-07-22 10:07:05 -04002869 struct btrfs_block_group_cache *cache;
Chris Mason9078a3e2007-04-26 16:46:15 -04002870 int err = 0;
Chris Mason9078a3e2007-04-26 16:46:15 -04002871 struct btrfs_path *path;
Chris Mason96b51792007-10-15 16:15:19 -04002872 u64 last = 0;
Chris Mason9078a3e2007-04-26 16:46:15 -04002873
2874 path = btrfs_alloc_path();
2875 if (!path)
2876 return -ENOMEM;
2877
Josef Bacik0af3d002010-06-21 14:48:16 -04002878again:
2879 while (1) {
2880 cache = btrfs_lookup_first_block_group(root->fs_info, last);
2881 while (cache) {
2882 if (cache->disk_cache_state == BTRFS_DC_CLEAR)
2883 break;
2884 cache = next_block_group(root, cache);
2885 }
2886 if (!cache) {
2887 if (last == 0)
2888 break;
2889 last = 0;
2890 continue;
2891 }
2892 err = cache_save_setup(cache, trans, path);
2893 last = cache->key.objectid + cache->key.offset;
2894 btrfs_put_block_group(cache);
2895 }
2896
Chris Masond3977122009-01-05 21:25:51 -05002897 while (1) {
Yan Zheng4a8c9a62009-07-22 10:07:05 -04002898 if (last == 0) {
2899 err = btrfs_run_delayed_refs(trans, root,
2900 (unsigned long)-1);
2901 BUG_ON(err);
Josef Bacik0f9dd462008-09-23 13:14:11 -04002902 }
Josef Bacik0f9dd462008-09-23 13:14:11 -04002903
Yan Zheng4a8c9a62009-07-22 10:07:05 -04002904 cache = btrfs_lookup_first_block_group(root->fs_info, last);
2905 while (cache) {
Josef Bacik0af3d002010-06-21 14:48:16 -04002906 if (cache->disk_cache_state == BTRFS_DC_CLEAR) {
2907 btrfs_put_block_group(cache);
2908 goto again;
2909 }
2910
Yan Zheng4a8c9a62009-07-22 10:07:05 -04002911 if (cache->dirty)
2912 break;
2913 cache = next_block_group(root, cache);
2914 }
2915 if (!cache) {
2916 if (last == 0)
2917 break;
2918 last = 0;
Chris Mason96b51792007-10-15 16:15:19 -04002919 continue;
Chris Mason9078a3e2007-04-26 16:46:15 -04002920 }
Yan Zheng4a8c9a62009-07-22 10:07:05 -04002921
Josef Bacik0cb59c92010-07-02 12:14:14 -04002922 if (cache->disk_cache_state == BTRFS_DC_SETUP)
2923 cache->disk_cache_state = BTRFS_DC_NEED_WRITE;
Yan Zheng4a8c9a62009-07-22 10:07:05 -04002924 cache->dirty = 0;
2925 last = cache->key.objectid + cache->key.offset;
2926
2927 err = write_one_cache_group(trans, root, path, cache);
2928 BUG_ON(err);
2929 btrfs_put_block_group(cache);
Chris Mason9078a3e2007-04-26 16:46:15 -04002930 }
Yan Zheng4a8c9a62009-07-22 10:07:05 -04002931
Josef Bacik0cb59c92010-07-02 12:14:14 -04002932 while (1) {
2933 /*
2934 * I don't think this is needed since we're just marking our
2935 * preallocated extent as written, but just in case it can't
2936 * hurt.
2937 */
2938 if (last == 0) {
2939 err = btrfs_run_delayed_refs(trans, root,
2940 (unsigned long)-1);
2941 BUG_ON(err);
2942 }
2943
2944 cache = btrfs_lookup_first_block_group(root->fs_info, last);
2945 while (cache) {
2946 /*
2947 * Really this shouldn't happen, but it could if we
2948 * couldn't write the entire preallocated extent and
2949 * splitting the extent resulted in a new block.
2950 */
2951 if (cache->dirty) {
2952 btrfs_put_block_group(cache);
2953 goto again;
2954 }
2955 if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
2956 break;
2957 cache = next_block_group(root, cache);
2958 }
2959 if (!cache) {
2960 if (last == 0)
2961 break;
2962 last = 0;
2963 continue;
2964 }
2965
2966 btrfs_write_out_cache(root, trans, cache, path);
2967
2968 /*
2969 * If we didn't have an error then the cache state is still
2970 * NEED_WRITE, so we can set it to WRITTEN.
2971 */
2972 if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
2973 cache->disk_cache_state = BTRFS_DC_WRITTEN;
2974 last = cache->key.objectid + cache->key.offset;
2975 btrfs_put_block_group(cache);
2976 }
2977
Chris Mason9078a3e2007-04-26 16:46:15 -04002978 btrfs_free_path(path);
Yan Zheng4a8c9a62009-07-22 10:07:05 -04002979 return 0;
Chris Mason9078a3e2007-04-26 16:46:15 -04002980}
2981
Yan Zhengd2fb3432008-12-11 16:30:39 -05002982int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
2983{
2984 struct btrfs_block_group_cache *block_group;
2985 int readonly = 0;
2986
2987 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
2988 if (!block_group || block_group->ro)
2989 readonly = 1;
2990 if (block_group)
Chris Masonfa9c0d72009-04-03 09:47:43 -04002991 btrfs_put_block_group(block_group);
Yan Zhengd2fb3432008-12-11 16:30:39 -05002992 return readonly;
2993}
2994
Chris Mason593060d2008-03-25 16:50:33 -04002995static int update_space_info(struct btrfs_fs_info *info, u64 flags,
2996 u64 total_bytes, u64 bytes_used,
2997 struct btrfs_space_info **space_info)
2998{
2999 struct btrfs_space_info *found;
Yan, Zhengb742bb82010-05-16 10:46:24 -04003000 int i;
3001 int factor;
3002
3003 if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3004 BTRFS_BLOCK_GROUP_RAID10))
3005 factor = 2;
3006 else
3007 factor = 1;
Chris Mason593060d2008-03-25 16:50:33 -04003008
3009 found = __find_space_info(info, flags);
3010 if (found) {
Josef Bacik25179202008-10-29 14:49:05 -04003011 spin_lock(&found->lock);
Chris Mason593060d2008-03-25 16:50:33 -04003012 found->total_bytes += total_bytes;
Josef Bacik89a55892010-10-14 14:52:27 -04003013 found->disk_total += total_bytes * factor;
Chris Mason593060d2008-03-25 16:50:33 -04003014 found->bytes_used += bytes_used;
Yan, Zhengb742bb82010-05-16 10:46:24 -04003015 found->disk_used += bytes_used * factor;
Chris Mason8f18cf12008-04-25 16:53:30 -04003016 found->full = 0;
Josef Bacik25179202008-10-29 14:49:05 -04003017 spin_unlock(&found->lock);
Chris Mason593060d2008-03-25 16:50:33 -04003018 *space_info = found;
3019 return 0;
3020 }
Yan Zhengc146afa2008-11-12 14:34:12 -05003021 found = kzalloc(sizeof(*found), GFP_NOFS);
Chris Mason593060d2008-03-25 16:50:33 -04003022 if (!found)
3023 return -ENOMEM;
3024
Yan, Zhengb742bb82010-05-16 10:46:24 -04003025 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
3026 INIT_LIST_HEAD(&found->block_groups[i]);
Josef Bacik80eb2342008-10-29 14:49:05 -04003027 init_rwsem(&found->groups_sem);
Josef Bacik0f9dd462008-09-23 13:14:11 -04003028 spin_lock_init(&found->lock);
Yan, Zhengb742bb82010-05-16 10:46:24 -04003029 found->flags = flags & (BTRFS_BLOCK_GROUP_DATA |
3030 BTRFS_BLOCK_GROUP_SYSTEM |
3031 BTRFS_BLOCK_GROUP_METADATA);
Chris Mason593060d2008-03-25 16:50:33 -04003032 found->total_bytes = total_bytes;
Josef Bacik89a55892010-10-14 14:52:27 -04003033 found->disk_total = total_bytes * factor;
Chris Mason593060d2008-03-25 16:50:33 -04003034 found->bytes_used = bytes_used;
Yan, Zhengb742bb82010-05-16 10:46:24 -04003035 found->disk_used = bytes_used * factor;
Chris Mason593060d2008-03-25 16:50:33 -04003036 found->bytes_pinned = 0;
Zheng Yane8569812008-09-26 10:05:48 -04003037 found->bytes_reserved = 0;
Yan Zhengc146afa2008-11-12 14:34:12 -05003038 found->bytes_readonly = 0;
Yan, Zhengf0486c62010-05-16 10:46:25 -04003039 found->bytes_may_use = 0;
Chris Mason593060d2008-03-25 16:50:33 -04003040 found->full = 0;
Chris Mason0e4f8f82011-04-15 16:05:44 -04003041 found->force_alloc = CHUNK_ALLOC_NO_FORCE;
Josef Bacik6d741192011-04-11 20:20:11 -04003042 found->chunk_alloc = 0;
Chris Mason593060d2008-03-25 16:50:33 -04003043 *space_info = found;
Chris Mason4184ea72009-03-10 12:39:20 -04003044 list_add_rcu(&found->list, &info->space_info);
Josef Bacik817d52f2009-07-13 21:29:25 -04003045 atomic_set(&found->caching_threads, 0);
Chris Mason593060d2008-03-25 16:50:33 -04003046 return 0;
3047}
3048
Chris Mason8790d502008-04-03 16:29:03 -04003049static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
3050{
3051 u64 extra_flags = flags & (BTRFS_BLOCK_GROUP_RAID0 |
Chris Mason611f0e02008-04-03 16:29:03 -04003052 BTRFS_BLOCK_GROUP_RAID1 |
Chris Mason321aecc2008-04-16 10:49:51 -04003053 BTRFS_BLOCK_GROUP_RAID10 |
Chris Mason611f0e02008-04-03 16:29:03 -04003054 BTRFS_BLOCK_GROUP_DUP);
Chris Mason8790d502008-04-03 16:29:03 -04003055 if (extra_flags) {
3056 if (flags & BTRFS_BLOCK_GROUP_DATA)
3057 fs_info->avail_data_alloc_bits |= extra_flags;
3058 if (flags & BTRFS_BLOCK_GROUP_METADATA)
3059 fs_info->avail_metadata_alloc_bits |= extra_flags;
3060 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3061 fs_info->avail_system_alloc_bits |= extra_flags;
3062 }
3063}
Chris Mason593060d2008-03-25 16:50:33 -04003064
Yan Zheng2b820322008-11-17 21:11:30 -05003065u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
Chris Masonec44a352008-04-28 15:29:52 -04003066{
Chris Masoncd02dca2010-12-13 14:56:23 -05003067 /*
3068 * we add in the count of missing devices because we want
3069 * to make sure that any RAID levels on a degraded FS
3070 * continue to be honored.
3071 */
3072 u64 num_devices = root->fs_info->fs_devices->rw_devices +
3073 root->fs_info->fs_devices->missing_devices;
Chris Masona061fc82008-05-07 11:43:44 -04003074
3075 if (num_devices == 1)
3076 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0);
3077 if (num_devices < 4)
3078 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
3079
Chris Masonec44a352008-04-28 15:29:52 -04003080 if ((flags & BTRFS_BLOCK_GROUP_DUP) &&
3081 (flags & (BTRFS_BLOCK_GROUP_RAID1 |
Chris Masona061fc82008-05-07 11:43:44 -04003082 BTRFS_BLOCK_GROUP_RAID10))) {
Chris Masonec44a352008-04-28 15:29:52 -04003083 flags &= ~BTRFS_BLOCK_GROUP_DUP;
Chris Masona061fc82008-05-07 11:43:44 -04003084 }
Chris Masonec44a352008-04-28 15:29:52 -04003085
3086 if ((flags & BTRFS_BLOCK_GROUP_RAID1) &&
Chris Masona061fc82008-05-07 11:43:44 -04003087 (flags & BTRFS_BLOCK_GROUP_RAID10)) {
Chris Masonec44a352008-04-28 15:29:52 -04003088 flags &= ~BTRFS_BLOCK_GROUP_RAID1;
Chris Masona061fc82008-05-07 11:43:44 -04003089 }
Chris Masonec44a352008-04-28 15:29:52 -04003090
3091 if ((flags & BTRFS_BLOCK_GROUP_RAID0) &&
3092 ((flags & BTRFS_BLOCK_GROUP_RAID1) |
3093 (flags & BTRFS_BLOCK_GROUP_RAID10) |
3094 (flags & BTRFS_BLOCK_GROUP_DUP)))
3095 flags &= ~BTRFS_BLOCK_GROUP_RAID0;
3096 return flags;
3097}
3098
Yan, Zhengb742bb82010-05-16 10:46:24 -04003099static u64 get_alloc_profile(struct btrfs_root *root, u64 flags)
Josef Bacik6a632092009-02-20 11:00:09 -05003100{
Yan, Zhengb742bb82010-05-16 10:46:24 -04003101 if (flags & BTRFS_BLOCK_GROUP_DATA)
3102 flags |= root->fs_info->avail_data_alloc_bits &
3103 root->fs_info->data_alloc_profile;
3104 else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3105 flags |= root->fs_info->avail_system_alloc_bits &
3106 root->fs_info->system_alloc_profile;
3107 else if (flags & BTRFS_BLOCK_GROUP_METADATA)
3108 flags |= root->fs_info->avail_metadata_alloc_bits &
3109 root->fs_info->metadata_alloc_profile;
3110 return btrfs_reduce_alloc_profile(root, flags);
3111}
Josef Bacik6a632092009-02-20 11:00:09 -05003112
Miao Xie6d07bce2011-01-05 10:07:31 +00003113u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data)
Yan, Zhengb742bb82010-05-16 10:46:24 -04003114{
3115 u64 flags;
Josef Bacik6a632092009-02-20 11:00:09 -05003116
Yan, Zhengb742bb82010-05-16 10:46:24 -04003117 if (data)
3118 flags = BTRFS_BLOCK_GROUP_DATA;
3119 else if (root == root->fs_info->chunk_root)
3120 flags = BTRFS_BLOCK_GROUP_SYSTEM;
3121 else
3122 flags = BTRFS_BLOCK_GROUP_METADATA;
3123
3124 return get_alloc_profile(root, flags);
Josef Bacik6a632092009-02-20 11:00:09 -05003125}
3126
3127void btrfs_set_inode_space_info(struct btrfs_root *root, struct inode *inode)
3128{
Josef Bacik6a632092009-02-20 11:00:09 -05003129 BTRFS_I(inode)->space_info = __find_space_info(root->fs_info,
Yan, Zhengf0486c62010-05-16 10:46:25 -04003130 BTRFS_BLOCK_GROUP_DATA);
Josef Bacik6a632092009-02-20 11:00:09 -05003131}
3132
3133/*
3134 * This will check the space that the inode allocates from to make sure we have
3135 * enough space for bytes.
3136 */
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04003137int btrfs_check_data_free_space(struct inode *inode, u64 bytes)
Josef Bacik6a632092009-02-20 11:00:09 -05003138{
3139 struct btrfs_space_info *data_sinfo;
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04003140 struct btrfs_root *root = BTRFS_I(inode)->root;
Josef Bacikab6e2412010-03-19 14:38:13 +00003141 u64 used;
Josef Bacik0af3d002010-06-21 14:48:16 -04003142 int ret = 0, committed = 0, alloc_chunk = 1;
Josef Bacik6a632092009-02-20 11:00:09 -05003143
3144 /* make sure bytes are sectorsize aligned */
3145 bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
3146
Josef Bacik0af3d002010-06-21 14:48:16 -04003147 if (root == root->fs_info->tree_root) {
3148 alloc_chunk = 0;
3149 committed = 1;
3150 }
3151
Josef Bacik6a632092009-02-20 11:00:09 -05003152 data_sinfo = BTRFS_I(inode)->space_info;
Chris Mason33b4d472009-09-22 14:45:50 -04003153 if (!data_sinfo)
3154 goto alloc;
3155
Josef Bacik6a632092009-02-20 11:00:09 -05003156again:
3157 /* make sure we have enough space to handle the data first */
3158 spin_lock(&data_sinfo->lock);
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04003159 used = data_sinfo->bytes_used + data_sinfo->bytes_reserved +
3160 data_sinfo->bytes_pinned + data_sinfo->bytes_readonly +
3161 data_sinfo->bytes_may_use;
Josef Bacikab6e2412010-03-19 14:38:13 +00003162
3163 if (used + bytes > data_sinfo->total_bytes) {
Josef Bacik4e06bdd2009-02-20 10:59:53 -05003164 struct btrfs_trans_handle *trans;
3165
Josef Bacik6a632092009-02-20 11:00:09 -05003166 /*
3167 * if we don't have enough free bytes in this space then we need
3168 * to alloc a new chunk.
3169 */
Josef Bacik0af3d002010-06-21 14:48:16 -04003170 if (!data_sinfo->full && alloc_chunk) {
Josef Bacik6a632092009-02-20 11:00:09 -05003171 u64 alloc_target;
Josef Bacik6a632092009-02-20 11:00:09 -05003172
Chris Mason0e4f8f82011-04-15 16:05:44 -04003173 data_sinfo->force_alloc = CHUNK_ALLOC_FORCE;
Josef Bacik6a632092009-02-20 11:00:09 -05003174 spin_unlock(&data_sinfo->lock);
Chris Mason33b4d472009-09-22 14:45:50 -04003175alloc:
Josef Bacik6a632092009-02-20 11:00:09 -05003176 alloc_target = btrfs_get_alloc_profile(root, 1);
Yan, Zhenga22285a2010-05-16 10:48:46 -04003177 trans = btrfs_join_transaction(root, 1);
3178 if (IS_ERR(trans))
3179 return PTR_ERR(trans);
Josef Bacik6a632092009-02-20 11:00:09 -05003180
3181 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3182 bytes + 2 * 1024 * 1024,
Chris Mason0e4f8f82011-04-15 16:05:44 -04003183 alloc_target,
3184 CHUNK_ALLOC_NO_FORCE);
Josef Bacik6a632092009-02-20 11:00:09 -05003185 btrfs_end_transaction(trans, root);
Miao Xied52a5b52011-01-05 10:07:18 +00003186 if (ret < 0) {
3187 if (ret != -ENOSPC)
3188 return ret;
3189 else
3190 goto commit_trans;
3191 }
Chris Mason33b4d472009-09-22 14:45:50 -04003192
3193 if (!data_sinfo) {
3194 btrfs_set_inode_space_info(root, inode);
3195 data_sinfo = BTRFS_I(inode)->space_info;
3196 }
Josef Bacik6a632092009-02-20 11:00:09 -05003197 goto again;
3198 }
3199 spin_unlock(&data_sinfo->lock);
Josef Bacik4e06bdd2009-02-20 10:59:53 -05003200
3201 /* commit the current transaction and try again */
Miao Xied52a5b52011-01-05 10:07:18 +00003202commit_trans:
Sage Weildd7e0b72009-09-29 18:38:44 -04003203 if (!committed && !root->fs_info->open_ioctl_trans) {
Josef Bacik4e06bdd2009-02-20 10:59:53 -05003204 committed = 1;
3205 trans = btrfs_join_transaction(root, 1);
Yan, Zhenga22285a2010-05-16 10:48:46 -04003206 if (IS_ERR(trans))
3207 return PTR_ERR(trans);
Josef Bacik4e06bdd2009-02-20 10:59:53 -05003208 ret = btrfs_commit_transaction(trans, root);
3209 if (ret)
3210 return ret;
3211 goto again;
3212 }
3213
Chris Mason933b5852010-05-26 11:31:00 -04003214#if 0 /* I hope we never need this code again, just in case */
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04003215 printk(KERN_ERR "no space left, need %llu, %llu bytes_used, "
3216 "%llu bytes_reserved, " "%llu bytes_pinned, "
3217 "%llu bytes_readonly, %llu may use %llu total\n",
3218 (unsigned long long)bytes,
Joel Becker21380932009-04-21 12:38:29 -07003219 (unsigned long long)data_sinfo->bytes_used,
3220 (unsigned long long)data_sinfo->bytes_reserved,
3221 (unsigned long long)data_sinfo->bytes_pinned,
3222 (unsigned long long)data_sinfo->bytes_readonly,
3223 (unsigned long long)data_sinfo->bytes_may_use,
3224 (unsigned long long)data_sinfo->total_bytes);
Chris Mason933b5852010-05-26 11:31:00 -04003225#endif
Josef Bacik6a632092009-02-20 11:00:09 -05003226 return -ENOSPC;
3227 }
3228 data_sinfo->bytes_may_use += bytes;
3229 BTRFS_I(inode)->reserved_bytes += bytes;
3230 spin_unlock(&data_sinfo->lock);
3231
Josef Bacik9ed74f22009-09-11 16:12:44 -04003232 return 0;
Josef Bacik6a632092009-02-20 11:00:09 -05003233}
3234
3235/*
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04003236 * called when we are clearing an delalloc extent from the
3237 * inode's io_tree or there was an error for whatever reason
3238 * after calling btrfs_check_data_free_space
Josef Bacik6a632092009-02-20 11:00:09 -05003239 */
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04003240void btrfs_free_reserved_data_space(struct inode *inode, u64 bytes)
Josef Bacik6a632092009-02-20 11:00:09 -05003241{
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04003242 struct btrfs_root *root = BTRFS_I(inode)->root;
Josef Bacik6a632092009-02-20 11:00:09 -05003243 struct btrfs_space_info *data_sinfo;
3244
3245 /* make sure bytes are sectorsize aligned */
3246 bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
3247
3248 data_sinfo = BTRFS_I(inode)->space_info;
3249 spin_lock(&data_sinfo->lock);
3250 data_sinfo->bytes_may_use -= bytes;
3251 BTRFS_I(inode)->reserved_bytes -= bytes;
3252 spin_unlock(&data_sinfo->lock);
3253}
3254
Josef Bacik97e728d2009-04-21 17:40:57 -04003255static void force_metadata_allocation(struct btrfs_fs_info *info)
3256{
3257 struct list_head *head = &info->space_info;
3258 struct btrfs_space_info *found;
3259
3260 rcu_read_lock();
3261 list_for_each_entry_rcu(found, head, list) {
3262 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
Chris Mason0e4f8f82011-04-15 16:05:44 -04003263 found->force_alloc = CHUNK_ALLOC_FORCE;
Josef Bacik97e728d2009-04-21 17:40:57 -04003264 }
3265 rcu_read_unlock();
3266}
3267
Chris Masone5bc2452010-10-26 13:37:56 -04003268static int should_alloc_chunk(struct btrfs_root *root,
Chris Mason0e4f8f82011-04-15 16:05:44 -04003269 struct btrfs_space_info *sinfo, u64 alloc_bytes,
3270 int force)
Yan, Zheng424499d2010-05-16 10:46:25 -04003271{
3272 u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly;
Chris Mason0e4f8f82011-04-15 16:05:44 -04003273 u64 num_allocated = sinfo->bytes_used + sinfo->bytes_reserved;
Chris Masone5bc2452010-10-26 13:37:56 -04003274 u64 thresh;
Yan, Zheng424499d2010-05-16 10:46:25 -04003275
Chris Mason0e4f8f82011-04-15 16:05:44 -04003276 if (force == CHUNK_ALLOC_FORCE)
3277 return 1;
3278
3279 /*
3280 * in limited mode, we want to have some free space up to
3281 * about 1% of the FS size.
3282 */
3283 if (force == CHUNK_ALLOC_LIMITED) {
3284 thresh = btrfs_super_total_bytes(&root->fs_info->super_copy);
3285 thresh = max_t(u64, 64 * 1024 * 1024,
3286 div_factor_fine(thresh, 1));
3287
3288 if (num_bytes - num_allocated < thresh)
3289 return 1;
3290 }
3291
3292 /*
3293 * we have two similar checks here, one based on percentage
3294 * and once based on a hard number of 256MB. The idea
3295 * is that if we have a good amount of free
3296 * room, don't allocate a chunk. A good mount is
3297 * less than 80% utilized of the chunks we have allocated,
3298 * or more than 256MB free
3299 */
3300 if (num_allocated + alloc_bytes + 256 * 1024 * 1024 < num_bytes)
Yan, Zheng424499d2010-05-16 10:46:25 -04003301 return 0;
3302
Chris Mason0e4f8f82011-04-15 16:05:44 -04003303 if (num_allocated + alloc_bytes < div_factor(num_bytes, 8))
Yan, Zheng424499d2010-05-16 10:46:25 -04003304 return 0;
3305
Chris Masone5bc2452010-10-26 13:37:56 -04003306 thresh = btrfs_super_total_bytes(&root->fs_info->super_copy);
Chris Mason0e4f8f82011-04-15 16:05:44 -04003307
3308 /* 256MB or 5% of the FS */
Chris Masone5bc2452010-10-26 13:37:56 -04003309 thresh = max_t(u64, 256 * 1024 * 1024, div_factor_fine(thresh, 5));
3310
3311 if (num_bytes > thresh && sinfo->bytes_used < div_factor(num_bytes, 3))
Josef Bacik14ed0ca2010-10-15 15:23:48 -04003312 return 0;
Yan, Zheng424499d2010-05-16 10:46:25 -04003313 return 1;
3314}
3315
Chris Mason6324fbf2008-03-24 15:01:59 -04003316static int do_chunk_alloc(struct btrfs_trans_handle *trans,
3317 struct btrfs_root *extent_root, u64 alloc_bytes,
Chris Mason0ef3e662008-05-24 14:04:53 -04003318 u64 flags, int force)
Chris Mason6324fbf2008-03-24 15:01:59 -04003319{
3320 struct btrfs_space_info *space_info;
Josef Bacik97e728d2009-04-21 17:40:57 -04003321 struct btrfs_fs_info *fs_info = extent_root->fs_info;
Josef Bacik6d741192011-04-11 20:20:11 -04003322 int wait_for_alloc = 0;
Yan Zhengc146afa2008-11-12 14:34:12 -05003323 int ret = 0;
3324
Yan Zheng2b820322008-11-17 21:11:30 -05003325 flags = btrfs_reduce_alloc_profile(extent_root, flags);
Chris Masonec44a352008-04-28 15:29:52 -04003326
Chris Mason6324fbf2008-03-24 15:01:59 -04003327 space_info = __find_space_info(extent_root->fs_info, flags);
Chris Mason593060d2008-03-25 16:50:33 -04003328 if (!space_info) {
3329 ret = update_space_info(extent_root->fs_info, flags,
3330 0, 0, &space_info);
3331 BUG_ON(ret);
3332 }
Chris Mason6324fbf2008-03-24 15:01:59 -04003333 BUG_ON(!space_info);
3334
Josef Bacik6d741192011-04-11 20:20:11 -04003335again:
Josef Bacik25179202008-10-29 14:49:05 -04003336 spin_lock(&space_info->lock);
Josef Bacik9ed74f22009-09-11 16:12:44 -04003337 if (space_info->force_alloc)
Chris Mason0e4f8f82011-04-15 16:05:44 -04003338 force = space_info->force_alloc;
Josef Bacik25179202008-10-29 14:49:05 -04003339 if (space_info->full) {
3340 spin_unlock(&space_info->lock);
Josef Bacik6d741192011-04-11 20:20:11 -04003341 return 0;
Josef Bacik25179202008-10-29 14:49:05 -04003342 }
Chris Mason6324fbf2008-03-24 15:01:59 -04003343
Chris Mason0e4f8f82011-04-15 16:05:44 -04003344 if (!should_alloc_chunk(extent_root, space_info, alloc_bytes, force)) {
Josef Bacik25179202008-10-29 14:49:05 -04003345 spin_unlock(&space_info->lock);
Josef Bacik6d741192011-04-11 20:20:11 -04003346 return 0;
3347 } else if (space_info->chunk_alloc) {
3348 wait_for_alloc = 1;
3349 } else {
3350 space_info->chunk_alloc = 1;
Josef Bacik25179202008-10-29 14:49:05 -04003351 }
Chris Mason0e4f8f82011-04-15 16:05:44 -04003352
Josef Bacik25179202008-10-29 14:49:05 -04003353 spin_unlock(&space_info->lock);
3354
Josef Bacik6d741192011-04-11 20:20:11 -04003355 mutex_lock(&fs_info->chunk_mutex);
3356
3357 /*
3358 * The chunk_mutex is held throughout the entirety of a chunk
3359 * allocation, so once we've acquired the chunk_mutex we know that the
3360 * other guy is done and we need to recheck and see if we should
3361 * allocate.
3362 */
3363 if (wait_for_alloc) {
3364 mutex_unlock(&fs_info->chunk_mutex);
3365 wait_for_alloc = 0;
3366 goto again;
3367 }
3368
Josef Bacik97e728d2009-04-21 17:40:57 -04003369 /*
Josef Bacik67377732010-09-16 16:19:09 -04003370 * If we have mixed data/metadata chunks we want to make sure we keep
3371 * allocating mixed chunks instead of individual chunks.
3372 */
3373 if (btrfs_mixed_space_info(space_info))
3374 flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
3375
3376 /*
Josef Bacik97e728d2009-04-21 17:40:57 -04003377 * if we're doing a data chunk, go ahead and make sure that
3378 * we keep a reasonable number of metadata chunks allocated in the
3379 * FS as well.
3380 */
Josef Bacik9ed74f22009-09-11 16:12:44 -04003381 if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
Josef Bacik97e728d2009-04-21 17:40:57 -04003382 fs_info->data_chunk_allocations++;
3383 if (!(fs_info->data_chunk_allocations %
3384 fs_info->metadata_ratio))
3385 force_metadata_allocation(fs_info);
3386 }
3387
Yan Zheng2b820322008-11-17 21:11:30 -05003388 ret = btrfs_alloc_chunk(trans, extent_root, flags);
Josef Bacik9ed74f22009-09-11 16:12:44 -04003389 spin_lock(&space_info->lock);
Chris Masond3977122009-01-05 21:25:51 -05003390 if (ret)
Chris Mason6324fbf2008-03-24 15:01:59 -04003391 space_info->full = 1;
Yan, Zheng424499d2010-05-16 10:46:25 -04003392 else
3393 ret = 1;
Josef Bacik6d741192011-04-11 20:20:11 -04003394
Chris Mason0e4f8f82011-04-15 16:05:44 -04003395 space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
Josef Bacik6d741192011-04-11 20:20:11 -04003396 space_info->chunk_alloc = 0;
Josef Bacik9ed74f22009-09-11 16:12:44 -04003397 spin_unlock(&space_info->lock);
Yan Zhengc146afa2008-11-12 14:34:12 -05003398 mutex_unlock(&extent_root->fs_info->chunk_mutex);
Josef Bacik0f9dd462008-09-23 13:14:11 -04003399 return ret;
Chris Mason6324fbf2008-03-24 15:01:59 -04003400}
3401
Yan, Zheng5da9d012010-05-16 10:46:25 -04003402/*
3403 * shrink metadata reservation for delalloc
3404 */
3405static int shrink_delalloc(struct btrfs_trans_handle *trans,
Josef Bacik0019f102010-10-15 15:18:40 -04003406 struct btrfs_root *root, u64 to_reclaim, int sync)
Yan, Zheng5da9d012010-05-16 10:46:25 -04003407{
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04003408 struct btrfs_block_rsv *block_rsv;
Josef Bacik0019f102010-10-15 15:18:40 -04003409 struct btrfs_space_info *space_info;
Yan, Zheng5da9d012010-05-16 10:46:25 -04003410 u64 reserved;
3411 u64 max_reclaim;
3412 u64 reclaimed = 0;
Josef Bacikb1953bc2011-01-21 21:10:01 +00003413 long time_left;
Chris Masonbf9022e2010-10-26 13:40:45 -04003414 int nr_pages = (2 * 1024 * 1024) >> PAGE_CACHE_SHIFT;
Josef Bacikb1953bc2011-01-21 21:10:01 +00003415 int loops = 0;
Chris Mason36e39c42011-03-12 07:08:42 -05003416 unsigned long progress;
Yan, Zheng5da9d012010-05-16 10:46:25 -04003417
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04003418 block_rsv = &root->fs_info->delalloc_block_rsv;
Josef Bacik0019f102010-10-15 15:18:40 -04003419 space_info = block_rsv->space_info;
Chris Masonbf9022e2010-10-26 13:40:45 -04003420
3421 smp_mb();
Josef Bacik0019f102010-10-15 15:18:40 -04003422 reserved = space_info->bytes_reserved;
Chris Mason36e39c42011-03-12 07:08:42 -05003423 progress = space_info->reservation_progress;
Yan, Zheng5da9d012010-05-16 10:46:25 -04003424
3425 if (reserved == 0)
3426 return 0;
3427
Sergei Trofimovichc4f675c2011-05-20 20:20:30 +00003428 /* nothing to shrink - nothing to reclaim */
3429 if (root->fs_info->delalloc_bytes == 0)
3430 return 0;
3431
Yan, Zheng5da9d012010-05-16 10:46:25 -04003432 max_reclaim = min(reserved, to_reclaim);
3433
Josef Bacikb1953bc2011-01-21 21:10:01 +00003434 while (loops < 1024) {
Chris Masonbf9022e2010-10-26 13:40:45 -04003435 /* have the flusher threads jump in and do some IO */
3436 smp_mb();
3437 nr_pages = min_t(unsigned long, nr_pages,
3438 root->fs_info->delalloc_bytes >> PAGE_CACHE_SHIFT);
3439 writeback_inodes_sb_nr_if_idle(root->fs_info->sb, nr_pages);
Yan, Zheng5da9d012010-05-16 10:46:25 -04003440
Josef Bacik0019f102010-10-15 15:18:40 -04003441 spin_lock(&space_info->lock);
Chris Mason36e39c42011-03-12 07:08:42 -05003442 if (reserved > space_info->bytes_reserved)
Josef Bacik0019f102010-10-15 15:18:40 -04003443 reclaimed += reserved - space_info->bytes_reserved;
3444 reserved = space_info->bytes_reserved;
3445 spin_unlock(&space_info->lock);
Yan, Zheng5da9d012010-05-16 10:46:25 -04003446
Chris Mason36e39c42011-03-12 07:08:42 -05003447 loops++;
3448
Yan, Zheng5da9d012010-05-16 10:46:25 -04003449 if (reserved == 0 || reclaimed >= max_reclaim)
3450 break;
3451
3452 if (trans && trans->transaction->blocked)
3453 return -EAGAIN;
Chris Masonbf9022e2010-10-26 13:40:45 -04003454
Chris Mason36e39c42011-03-12 07:08:42 -05003455 time_left = schedule_timeout_interruptible(1);
Josef Bacikb1953bc2011-01-21 21:10:01 +00003456
3457 /* We were interrupted, exit */
3458 if (time_left)
3459 break;
3460
Chris Mason36e39c42011-03-12 07:08:42 -05003461 /* we've kicked the IO a few times, if anything has been freed,
3462 * exit. There is no sense in looping here for a long time
3463 * when we really need to commit the transaction, or there are
3464 * just too many writers without enough free space
3465 */
3466
3467 if (loops > 3) {
3468 smp_mb();
3469 if (progress != space_info->reservation_progress)
3470 break;
3471 }
Chris Masonbf9022e2010-10-26 13:40:45 -04003472
Yan, Zheng5da9d012010-05-16 10:46:25 -04003473 }
3474 return reclaimed >= to_reclaim;
3475}
3476
Josef Bacik8bb8ab22010-10-15 16:52:49 -04003477/*
3478 * Retries tells us how many times we've called reserve_metadata_bytes. The
3479 * idea is if this is the first call (retries == 0) then we will add to our
3480 * reserved count if we can't make the allocation in order to hold our place
3481 * while we go and try and free up space. That way for retries > 1 we don't try
3482 * and add space, we just check to see if the amount of unused space is >= the
3483 * total space, meaning that our reservation is valid.
3484 *
3485 * However if we don't intend to retry this reservation, pass -1 as retries so
3486 * that it short circuits this logic.
3487 */
3488static int reserve_metadata_bytes(struct btrfs_trans_handle *trans,
3489 struct btrfs_root *root,
3490 struct btrfs_block_rsv *block_rsv,
3491 u64 orig_bytes, int flush)
Yan, Zhengf0486c62010-05-16 10:46:25 -04003492{
3493 struct btrfs_space_info *space_info = block_rsv->space_info;
3494 u64 unused;
Josef Bacik8bb8ab22010-10-15 16:52:49 -04003495 u64 num_bytes = orig_bytes;
3496 int retries = 0;
3497 int ret = 0;
3498 bool reserved = false;
Josef Bacik38227932010-10-26 12:52:53 -04003499 bool committed = false;
Josef Bacik8bb8ab22010-10-15 16:52:49 -04003500
3501again:
3502 ret = -ENOSPC;
3503 if (reserved)
3504 num_bytes = 0;
Yan, Zhengf0486c62010-05-16 10:46:25 -04003505
3506 spin_lock(&space_info->lock);
3507 unused = space_info->bytes_used + space_info->bytes_reserved +
Josef Bacik6d487552010-10-15 15:13:32 -04003508 space_info->bytes_pinned + space_info->bytes_readonly +
3509 space_info->bytes_may_use;
Yan, Zhengf0486c62010-05-16 10:46:25 -04003510
Josef Bacik8bb8ab22010-10-15 16:52:49 -04003511 /*
3512 * The idea here is that we've not already over-reserved the block group
3513 * then we can go ahead and save our reservation first and then start
3514 * flushing if we need to. Otherwise if we've already overcommitted
3515 * lets start flushing stuff first and then come back and try to make
3516 * our reservation.
3517 */
3518 if (unused <= space_info->total_bytes) {
Arne Jansen6f334342010-11-12 23:17:56 +00003519 unused = space_info->total_bytes - unused;
Josef Bacik8bb8ab22010-10-15 16:52:49 -04003520 if (unused >= num_bytes) {
3521 if (!reserved)
3522 space_info->bytes_reserved += orig_bytes;
Yan, Zhengf0486c62010-05-16 10:46:25 -04003523 ret = 0;
3524 } else {
Josef Bacik8bb8ab22010-10-15 16:52:49 -04003525 /*
3526 * Ok set num_bytes to orig_bytes since we aren't
3527 * overocmmitted, this way we only try and reclaim what
3528 * we need.
3529 */
3530 num_bytes = orig_bytes;
Yan, Zhengf0486c62010-05-16 10:46:25 -04003531 }
Josef Bacik8bb8ab22010-10-15 16:52:49 -04003532 } else {
3533 /*
3534 * Ok we're over committed, set num_bytes to the overcommitted
3535 * amount plus the amount of bytes that we need for this
3536 * reservation.
3537 */
3538 num_bytes = unused - space_info->total_bytes +
3539 (orig_bytes * (retries + 1));
Yan, Zhengf0486c62010-05-16 10:46:25 -04003540 }
Josef Bacik8bb8ab22010-10-15 16:52:49 -04003541
3542 /*
3543 * Couldn't make our reservation, save our place so while we're trying
3544 * to reclaim space we can actually use it instead of somebody else
3545 * stealing it from us.
3546 */
3547 if (ret && !reserved) {
3548 space_info->bytes_reserved += orig_bytes;
3549 reserved = true;
3550 }
3551
Yan, Zhengf0486c62010-05-16 10:46:25 -04003552 spin_unlock(&space_info->lock);
3553
Josef Bacik8bb8ab22010-10-15 16:52:49 -04003554 if (!ret)
3555 return 0;
3556
3557 if (!flush)
3558 goto out;
3559
3560 /*
3561 * We do synchronous shrinking since we don't actually unreserve
3562 * metadata until after the IO is completed.
3563 */
3564 ret = shrink_delalloc(trans, root, num_bytes, 1);
3565 if (ret > 0)
3566 return 0;
3567 else if (ret < 0)
3568 goto out;
3569
3570 /*
3571 * So if we were overcommitted it's possible that somebody else flushed
3572 * out enough space and we simply didn't have enough space to reclaim,
3573 * so go back around and try again.
3574 */
3575 if (retries < 2) {
3576 retries++;
3577 goto again;
3578 }
3579
3580 spin_lock(&space_info->lock);
3581 /*
3582 * Not enough space to be reclaimed, don't bother committing the
3583 * transaction.
3584 */
3585 if (space_info->bytes_pinned < orig_bytes)
3586 ret = -ENOSPC;
3587 spin_unlock(&space_info->lock);
3588 if (ret)
3589 goto out;
3590
3591 ret = -EAGAIN;
Josef Bacik38227932010-10-26 12:52:53 -04003592 if (trans || committed)
Josef Bacik8bb8ab22010-10-15 16:52:49 -04003593 goto out;
3594
Josef Bacik8bb8ab22010-10-15 16:52:49 -04003595 ret = -ENOSPC;
3596 trans = btrfs_join_transaction(root, 1);
3597 if (IS_ERR(trans))
3598 goto out;
3599 ret = btrfs_commit_transaction(trans, root);
Josef Bacik38227932010-10-26 12:52:53 -04003600 if (!ret) {
3601 trans = NULL;
3602 committed = true;
Josef Bacik8bb8ab22010-10-15 16:52:49 -04003603 goto again;
Josef Bacik38227932010-10-26 12:52:53 -04003604 }
Josef Bacik8bb8ab22010-10-15 16:52:49 -04003605
3606out:
3607 if (reserved) {
3608 spin_lock(&space_info->lock);
3609 space_info->bytes_reserved -= orig_bytes;
3610 spin_unlock(&space_info->lock);
3611 }
3612
Yan, Zhengf0486c62010-05-16 10:46:25 -04003613 return ret;
3614}
3615
3616static struct btrfs_block_rsv *get_block_rsv(struct btrfs_trans_handle *trans,
3617 struct btrfs_root *root)
3618{
3619 struct btrfs_block_rsv *block_rsv;
3620 if (root->ref_cows)
3621 block_rsv = trans->block_rsv;
3622 else
3623 block_rsv = root->block_rsv;
3624
3625 if (!block_rsv)
3626 block_rsv = &root->fs_info->empty_block_rsv;
3627
3628 return block_rsv;
3629}
3630
3631static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
3632 u64 num_bytes)
3633{
3634 int ret = -ENOSPC;
3635 spin_lock(&block_rsv->lock);
3636 if (block_rsv->reserved >= num_bytes) {
3637 block_rsv->reserved -= num_bytes;
3638 if (block_rsv->reserved < block_rsv->size)
3639 block_rsv->full = 0;
3640 ret = 0;
3641 }
3642 spin_unlock(&block_rsv->lock);
3643 return ret;
3644}
3645
3646static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
3647 u64 num_bytes, int update_size)
3648{
3649 spin_lock(&block_rsv->lock);
3650 block_rsv->reserved += num_bytes;
3651 if (update_size)
3652 block_rsv->size += num_bytes;
3653 else if (block_rsv->reserved >= block_rsv->size)
3654 block_rsv->full = 1;
3655 spin_unlock(&block_rsv->lock);
3656}
3657
3658void block_rsv_release_bytes(struct btrfs_block_rsv *block_rsv,
3659 struct btrfs_block_rsv *dest, u64 num_bytes)
3660{
3661 struct btrfs_space_info *space_info = block_rsv->space_info;
3662
3663 spin_lock(&block_rsv->lock);
3664 if (num_bytes == (u64)-1)
3665 num_bytes = block_rsv->size;
3666 block_rsv->size -= num_bytes;
3667 if (block_rsv->reserved >= block_rsv->size) {
3668 num_bytes = block_rsv->reserved - block_rsv->size;
3669 block_rsv->reserved = block_rsv->size;
3670 block_rsv->full = 1;
3671 } else {
3672 num_bytes = 0;
3673 }
3674 spin_unlock(&block_rsv->lock);
3675
3676 if (num_bytes > 0) {
3677 if (dest) {
Josef Bacike9e22892011-01-24 21:43:19 +00003678 spin_lock(&dest->lock);
3679 if (!dest->full) {
3680 u64 bytes_to_add;
3681
3682 bytes_to_add = dest->size - dest->reserved;
3683 bytes_to_add = min(num_bytes, bytes_to_add);
3684 dest->reserved += bytes_to_add;
3685 if (dest->reserved >= dest->size)
3686 dest->full = 1;
3687 num_bytes -= bytes_to_add;
3688 }
3689 spin_unlock(&dest->lock);
3690 }
3691 if (num_bytes) {
Yan, Zhengf0486c62010-05-16 10:46:25 -04003692 spin_lock(&space_info->lock);
3693 space_info->bytes_reserved -= num_bytes;
Chris Mason36e39c42011-03-12 07:08:42 -05003694 space_info->reservation_progress++;
Yan, Zhengf0486c62010-05-16 10:46:25 -04003695 spin_unlock(&space_info->lock);
3696 }
3697 }
3698}
3699
3700static int block_rsv_migrate_bytes(struct btrfs_block_rsv *src,
3701 struct btrfs_block_rsv *dst, u64 num_bytes)
3702{
3703 int ret;
3704
3705 ret = block_rsv_use_bytes(src, num_bytes);
3706 if (ret)
3707 return ret;
3708
3709 block_rsv_add_bytes(dst, num_bytes, 1);
3710 return 0;
3711}
3712
3713void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv)
3714{
3715 memset(rsv, 0, sizeof(*rsv));
3716 spin_lock_init(&rsv->lock);
3717 atomic_set(&rsv->usage, 1);
3718 rsv->priority = 6;
3719 INIT_LIST_HEAD(&rsv->list);
3720}
3721
3722struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root)
3723{
3724 struct btrfs_block_rsv *block_rsv;
3725 struct btrfs_fs_info *fs_info = root->fs_info;
Yan, Zhengf0486c62010-05-16 10:46:25 -04003726
3727 block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
3728 if (!block_rsv)
3729 return NULL;
3730
3731 btrfs_init_block_rsv(block_rsv);
Yan, Zhengf0486c62010-05-16 10:46:25 -04003732 block_rsv->space_info = __find_space_info(fs_info,
3733 BTRFS_BLOCK_GROUP_METADATA);
Yan, Zhengf0486c62010-05-16 10:46:25 -04003734 return block_rsv;
3735}
3736
3737void btrfs_free_block_rsv(struct btrfs_root *root,
3738 struct btrfs_block_rsv *rsv)
3739{
3740 if (rsv && atomic_dec_and_test(&rsv->usage)) {
3741 btrfs_block_rsv_release(root, rsv, (u64)-1);
3742 if (!rsv->durable)
3743 kfree(rsv);
3744 }
3745}
3746
3747/*
3748 * make the block_rsv struct be able to capture freed space.
3749 * the captured space will re-add to the the block_rsv struct
3750 * after transaction commit
3751 */
3752void btrfs_add_durable_block_rsv(struct btrfs_fs_info *fs_info,
3753 struct btrfs_block_rsv *block_rsv)
3754{
3755 block_rsv->durable = 1;
3756 mutex_lock(&fs_info->durable_block_rsv_mutex);
3757 list_add_tail(&block_rsv->list, &fs_info->durable_block_rsv_list);
3758 mutex_unlock(&fs_info->durable_block_rsv_mutex);
3759}
3760
3761int btrfs_block_rsv_add(struct btrfs_trans_handle *trans,
3762 struct btrfs_root *root,
3763 struct btrfs_block_rsv *block_rsv,
Josef Bacik8bb8ab22010-10-15 16:52:49 -04003764 u64 num_bytes)
Yan, Zhengf0486c62010-05-16 10:46:25 -04003765{
3766 int ret;
3767
3768 if (num_bytes == 0)
3769 return 0;
Josef Bacik8bb8ab22010-10-15 16:52:49 -04003770
3771 ret = reserve_metadata_bytes(trans, root, block_rsv, num_bytes, 1);
Yan, Zhengf0486c62010-05-16 10:46:25 -04003772 if (!ret) {
3773 block_rsv_add_bytes(block_rsv, num_bytes, 1);
3774 return 0;
3775 }
3776
Yan, Zhengf0486c62010-05-16 10:46:25 -04003777 return ret;
3778}
3779
3780int btrfs_block_rsv_check(struct btrfs_trans_handle *trans,
3781 struct btrfs_root *root,
3782 struct btrfs_block_rsv *block_rsv,
3783 u64 min_reserved, int min_factor)
3784{
3785 u64 num_bytes = 0;
3786 int commit_trans = 0;
3787 int ret = -ENOSPC;
3788
3789 if (!block_rsv)
3790 return 0;
3791
3792 spin_lock(&block_rsv->lock);
3793 if (min_factor > 0)
3794 num_bytes = div_factor(block_rsv->size, min_factor);
3795 if (min_reserved > num_bytes)
3796 num_bytes = min_reserved;
3797
3798 if (block_rsv->reserved >= num_bytes) {
3799 ret = 0;
3800 } else {
3801 num_bytes -= block_rsv->reserved;
3802 if (block_rsv->durable &&
3803 block_rsv->freed[0] + block_rsv->freed[1] >= num_bytes)
3804 commit_trans = 1;
3805 }
3806 spin_unlock(&block_rsv->lock);
3807 if (!ret)
3808 return 0;
3809
3810 if (block_rsv->refill_used) {
Josef Bacik8bb8ab22010-10-15 16:52:49 -04003811 ret = reserve_metadata_bytes(trans, root, block_rsv,
3812 num_bytes, 0);
Yan, Zhengf0486c62010-05-16 10:46:25 -04003813 if (!ret) {
3814 block_rsv_add_bytes(block_rsv, num_bytes, 0);
3815 return 0;
3816 }
3817 }
3818
3819 if (commit_trans) {
3820 if (trans)
3821 return -EAGAIN;
3822
3823 trans = btrfs_join_transaction(root, 1);
3824 BUG_ON(IS_ERR(trans));
3825 ret = btrfs_commit_transaction(trans, root);
3826 return 0;
3827 }
3828
Yan, Zhengf0486c62010-05-16 10:46:25 -04003829 return -ENOSPC;
3830}
3831
3832int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv,
3833 struct btrfs_block_rsv *dst_rsv,
3834 u64 num_bytes)
3835{
3836 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
3837}
3838
3839void btrfs_block_rsv_release(struct btrfs_root *root,
3840 struct btrfs_block_rsv *block_rsv,
3841 u64 num_bytes)
3842{
3843 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
3844 if (global_rsv->full || global_rsv == block_rsv ||
3845 block_rsv->space_info != global_rsv->space_info)
3846 global_rsv = NULL;
3847 block_rsv_release_bytes(block_rsv, global_rsv, num_bytes);
3848}
3849
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04003850/*
3851 * helper to calculate size of global block reservation.
3852 * the desired value is sum of space used by extent tree,
3853 * checksum tree and root tree
3854 */
3855static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info)
3856{
3857 struct btrfs_space_info *sinfo;
3858 u64 num_bytes;
3859 u64 meta_used;
3860 u64 data_used;
3861 int csum_size = btrfs_super_csum_size(&fs_info->super_copy);
3862#if 0
3863 /*
3864 * per tree used space accounting can be inaccuracy, so we
3865 * can't rely on it.
3866 */
3867 spin_lock(&fs_info->extent_root->accounting_lock);
3868 num_bytes = btrfs_root_used(&fs_info->extent_root->root_item);
3869 spin_unlock(&fs_info->extent_root->accounting_lock);
3870
3871 spin_lock(&fs_info->csum_root->accounting_lock);
3872 num_bytes += btrfs_root_used(&fs_info->csum_root->root_item);
3873 spin_unlock(&fs_info->csum_root->accounting_lock);
3874
3875 spin_lock(&fs_info->tree_root->accounting_lock);
3876 num_bytes += btrfs_root_used(&fs_info->tree_root->root_item);
3877 spin_unlock(&fs_info->tree_root->accounting_lock);
3878#endif
3879 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA);
3880 spin_lock(&sinfo->lock);
3881 data_used = sinfo->bytes_used;
3882 spin_unlock(&sinfo->lock);
3883
3884 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
3885 spin_lock(&sinfo->lock);
Josef Bacik6d487552010-10-15 15:13:32 -04003886 if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA)
3887 data_used = 0;
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04003888 meta_used = sinfo->bytes_used;
3889 spin_unlock(&sinfo->lock);
3890
3891 num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) *
3892 csum_size * 2;
3893 num_bytes += div64_u64(data_used + meta_used, 50);
3894
3895 if (num_bytes * 3 > meta_used)
3896 num_bytes = div64_u64(meta_used, 3);
3897
3898 return ALIGN(num_bytes, fs_info->extent_root->leafsize << 10);
3899}
3900
3901static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
3902{
3903 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
3904 struct btrfs_space_info *sinfo = block_rsv->space_info;
3905 u64 num_bytes;
3906
3907 num_bytes = calc_global_metadata_size(fs_info);
3908
3909 spin_lock(&block_rsv->lock);
3910 spin_lock(&sinfo->lock);
3911
3912 block_rsv->size = num_bytes;
3913
3914 num_bytes = sinfo->bytes_used + sinfo->bytes_pinned +
Josef Bacik6d487552010-10-15 15:13:32 -04003915 sinfo->bytes_reserved + sinfo->bytes_readonly +
3916 sinfo->bytes_may_use;
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04003917
3918 if (sinfo->total_bytes > num_bytes) {
3919 num_bytes = sinfo->total_bytes - num_bytes;
3920 block_rsv->reserved += num_bytes;
3921 sinfo->bytes_reserved += num_bytes;
3922 }
3923
3924 if (block_rsv->reserved >= block_rsv->size) {
3925 num_bytes = block_rsv->reserved - block_rsv->size;
3926 sinfo->bytes_reserved -= num_bytes;
Chris Mason36e39c42011-03-12 07:08:42 -05003927 sinfo->reservation_progress++;
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04003928 block_rsv->reserved = block_rsv->size;
3929 block_rsv->full = 1;
3930 }
3931#if 0
3932 printk(KERN_INFO"global block rsv size %llu reserved %llu\n",
3933 block_rsv->size, block_rsv->reserved);
3934#endif
3935 spin_unlock(&sinfo->lock);
3936 spin_unlock(&block_rsv->lock);
3937}
3938
Yan, Zhengf0486c62010-05-16 10:46:25 -04003939static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
3940{
3941 struct btrfs_space_info *space_info;
3942
3943 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
3944 fs_info->chunk_block_rsv.space_info = space_info;
3945 fs_info->chunk_block_rsv.priority = 10;
3946
3947 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04003948 fs_info->global_block_rsv.space_info = space_info;
3949 fs_info->global_block_rsv.priority = 10;
3950 fs_info->global_block_rsv.refill_used = 1;
3951 fs_info->delalloc_block_rsv.space_info = space_info;
Yan, Zhengf0486c62010-05-16 10:46:25 -04003952 fs_info->trans_block_rsv.space_info = space_info;
3953 fs_info->empty_block_rsv.space_info = space_info;
3954 fs_info->empty_block_rsv.priority = 10;
3955
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04003956 fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
3957 fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
3958 fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
3959 fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
Yan, Zhengf0486c62010-05-16 10:46:25 -04003960 fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04003961
3962 btrfs_add_durable_block_rsv(fs_info, &fs_info->global_block_rsv);
3963
3964 btrfs_add_durable_block_rsv(fs_info, &fs_info->delalloc_block_rsv);
3965
3966 update_global_block_rsv(fs_info);
3967}
3968
3969static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
3970{
3971 block_rsv_release_bytes(&fs_info->global_block_rsv, NULL, (u64)-1);
3972 WARN_ON(fs_info->delalloc_block_rsv.size > 0);
3973 WARN_ON(fs_info->delalloc_block_rsv.reserved > 0);
3974 WARN_ON(fs_info->trans_block_rsv.size > 0);
3975 WARN_ON(fs_info->trans_block_rsv.reserved > 0);
3976 WARN_ON(fs_info->chunk_block_rsv.size > 0);
3977 WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
Yan, Zhengf0486c62010-05-16 10:46:25 -04003978}
3979
Yan, Zhenga22285a2010-05-16 10:48:46 -04003980static u64 calc_trans_metadata_size(struct btrfs_root *root, int num_items)
3981{
3982 return (root->leafsize + root->nodesize * (BTRFS_MAX_LEVEL - 1)) *
3983 3 * num_items;
3984}
3985
3986int btrfs_trans_reserve_metadata(struct btrfs_trans_handle *trans,
3987 struct btrfs_root *root,
Josef Bacik8bb8ab22010-10-15 16:52:49 -04003988 int num_items)
Yan, Zhenga22285a2010-05-16 10:48:46 -04003989{
3990 u64 num_bytes;
3991 int ret;
3992
3993 if (num_items == 0 || root->fs_info->chunk_root == root)
3994 return 0;
3995
3996 num_bytes = calc_trans_metadata_size(root, num_items);
3997 ret = btrfs_block_rsv_add(trans, root, &root->fs_info->trans_block_rsv,
Josef Bacik8bb8ab22010-10-15 16:52:49 -04003998 num_bytes);
Yan, Zhenga22285a2010-05-16 10:48:46 -04003999 if (!ret) {
4000 trans->bytes_reserved += num_bytes;
4001 trans->block_rsv = &root->fs_info->trans_block_rsv;
4002 }
4003 return ret;
4004}
4005
4006void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans,
4007 struct btrfs_root *root)
4008{
4009 if (!trans->bytes_reserved)
4010 return;
4011
4012 BUG_ON(trans->block_rsv != &root->fs_info->trans_block_rsv);
4013 btrfs_block_rsv_release(root, trans->block_rsv,
4014 trans->bytes_reserved);
4015 trans->bytes_reserved = 0;
4016}
4017
Yan, Zhengd68fc572010-05-16 10:49:58 -04004018int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
4019 struct inode *inode)
4020{
4021 struct btrfs_root *root = BTRFS_I(inode)->root;
4022 struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
4023 struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv;
4024
4025 /*
4026 * one for deleting orphan item, one for updating inode and
4027 * two for calling btrfs_truncate_inode_items.
4028 *
4029 * btrfs_truncate_inode_items is a delete operation, it frees
4030 * more space than it uses in most cases. So two units of
4031 * metadata space should be enough for calling it many times.
4032 * If all of the metadata space is used, we can commit
4033 * transaction and use space it freed.
4034 */
4035 u64 num_bytes = calc_trans_metadata_size(root, 4);
4036 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4037}
4038
4039void btrfs_orphan_release_metadata(struct inode *inode)
4040{
4041 struct btrfs_root *root = BTRFS_I(inode)->root;
4042 u64 num_bytes = calc_trans_metadata_size(root, 4);
4043 btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes);
4044}
4045
Yan, Zhenga22285a2010-05-16 10:48:46 -04004046int btrfs_snap_reserve_metadata(struct btrfs_trans_handle *trans,
4047 struct btrfs_pending_snapshot *pending)
4048{
4049 struct btrfs_root *root = pending->root;
4050 struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
4051 struct btrfs_block_rsv *dst_rsv = &pending->block_rsv;
4052 /*
4053 * two for root back/forward refs, two for directory entries
4054 * and one for root of the snapshot.
4055 */
4056 u64 num_bytes = calc_trans_metadata_size(root, 5);
4057 dst_rsv->space_info = src_rsv->space_info;
4058 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4059}
4060
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04004061static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes)
4062{
4063 return num_bytes >>= 3;
4064}
4065
4066int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes)
4067{
4068 struct btrfs_root *root = BTRFS_I(inode)->root;
4069 struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv;
4070 u64 to_reserve;
4071 int nr_extents;
Josef Bacik57a45ced2011-01-25 16:30:38 -05004072 int reserved_extents;
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04004073 int ret;
4074
4075 if (btrfs_transaction_in_commit(root->fs_info))
4076 schedule_timeout(1);
4077
4078 num_bytes = ALIGN(num_bytes, root->sectorsize);
Josef Bacik8bb8ab22010-10-15 16:52:49 -04004079
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04004080 nr_extents = atomic_read(&BTRFS_I(inode)->outstanding_extents) + 1;
Josef Bacik57a45ced2011-01-25 16:30:38 -05004081 reserved_extents = atomic_read(&BTRFS_I(inode)->reserved_extents);
4082
4083 if (nr_extents > reserved_extents) {
4084 nr_extents -= reserved_extents;
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04004085 to_reserve = calc_trans_metadata_size(root, nr_extents);
4086 } else {
4087 nr_extents = 0;
4088 to_reserve = 0;
4089 }
Josef Bacik57a45ced2011-01-25 16:30:38 -05004090
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04004091 to_reserve += calc_csum_metadata_size(inode, num_bytes);
Josef Bacik8bb8ab22010-10-15 16:52:49 -04004092 ret = reserve_metadata_bytes(NULL, root, block_rsv, to_reserve, 1);
4093 if (ret)
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04004094 return ret;
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04004095
Josef Bacik57a45ced2011-01-25 16:30:38 -05004096 atomic_add(nr_extents, &BTRFS_I(inode)->reserved_extents);
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04004097 atomic_inc(&BTRFS_I(inode)->outstanding_extents);
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04004098
4099 block_rsv_add_bytes(block_rsv, to_reserve, 1);
4100
4101 if (block_rsv->size > 512 * 1024 * 1024)
Josef Bacik0019f102010-10-15 15:18:40 -04004102 shrink_delalloc(NULL, root, to_reserve, 0);
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04004103
4104 return 0;
4105}
4106
4107void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes)
4108{
4109 struct btrfs_root *root = BTRFS_I(inode)->root;
4110 u64 to_free;
4111 int nr_extents;
Josef Bacik57a45ced2011-01-25 16:30:38 -05004112 int reserved_extents;
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04004113
4114 num_bytes = ALIGN(num_bytes, root->sectorsize);
4115 atomic_dec(&BTRFS_I(inode)->outstanding_extents);
Josef Bacik3c148742011-02-02 15:53:47 +00004116 WARN_ON(atomic_read(&BTRFS_I(inode)->outstanding_extents) < 0);
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04004117
Josef Bacik57a45ced2011-01-25 16:30:38 -05004118 reserved_extents = atomic_read(&BTRFS_I(inode)->reserved_extents);
4119 do {
4120 int old, new;
4121
4122 nr_extents = atomic_read(&BTRFS_I(inode)->outstanding_extents);
4123 if (nr_extents >= reserved_extents) {
4124 nr_extents = 0;
4125 break;
4126 }
4127 old = reserved_extents;
4128 nr_extents = reserved_extents - nr_extents;
4129 new = reserved_extents - nr_extents;
4130 old = atomic_cmpxchg(&BTRFS_I(inode)->reserved_extents,
4131 reserved_extents, new);
4132 if (likely(old == reserved_extents))
4133 break;
4134 reserved_extents = old;
4135 } while (1);
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04004136
4137 to_free = calc_csum_metadata_size(inode, num_bytes);
4138 if (nr_extents > 0)
4139 to_free += calc_trans_metadata_size(root, nr_extents);
4140
4141 btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv,
4142 to_free);
4143}
4144
4145int btrfs_delalloc_reserve_space(struct inode *inode, u64 num_bytes)
4146{
4147 int ret;
4148
4149 ret = btrfs_check_data_free_space(inode, num_bytes);
4150 if (ret)
4151 return ret;
4152
4153 ret = btrfs_delalloc_reserve_metadata(inode, num_bytes);
4154 if (ret) {
4155 btrfs_free_reserved_data_space(inode, num_bytes);
4156 return ret;
4157 }
4158
4159 return 0;
4160}
4161
4162void btrfs_delalloc_release_space(struct inode *inode, u64 num_bytes)
4163{
4164 btrfs_delalloc_release_metadata(inode, num_bytes);
4165 btrfs_free_reserved_data_space(inode, num_bytes);
4166}
4167
Chris Mason9078a3e2007-04-26 16:46:15 -04004168static int update_block_group(struct btrfs_trans_handle *trans,
4169 struct btrfs_root *root,
Yan, Zhengf0486c62010-05-16 10:46:25 -04004170 u64 bytenr, u64 num_bytes, int alloc)
Chris Mason9078a3e2007-04-26 16:46:15 -04004171{
Josef Bacik0af3d002010-06-21 14:48:16 -04004172 struct btrfs_block_group_cache *cache = NULL;
Chris Mason9078a3e2007-04-26 16:46:15 -04004173 struct btrfs_fs_info *info = root->fs_info;
Chris Masondb945352007-10-15 16:15:53 -04004174 u64 total = num_bytes;
Chris Mason9078a3e2007-04-26 16:46:15 -04004175 u64 old_val;
Chris Masondb945352007-10-15 16:15:53 -04004176 u64 byte_in_group;
Josef Bacik0af3d002010-06-21 14:48:16 -04004177 int factor;
Chris Mason3e1ad542007-05-07 20:03:49 -04004178
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004179 /* block accounting for super block */
4180 spin_lock(&info->delalloc_lock);
4181 old_val = btrfs_super_bytes_used(&info->super_copy);
4182 if (alloc)
4183 old_val += num_bytes;
4184 else
4185 old_val -= num_bytes;
4186 btrfs_set_super_bytes_used(&info->super_copy, old_val);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004187 spin_unlock(&info->delalloc_lock);
4188
Chris Masond3977122009-01-05 21:25:51 -05004189 while (total) {
Chris Masondb945352007-10-15 16:15:53 -04004190 cache = btrfs_lookup_block_group(info, bytenr);
Josef Bacikf3465ca2008-11-12 14:19:50 -05004191 if (!cache)
Chris Mason9078a3e2007-04-26 16:46:15 -04004192 return -1;
Yan, Zhengb742bb82010-05-16 10:46:24 -04004193 if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
4194 BTRFS_BLOCK_GROUP_RAID1 |
4195 BTRFS_BLOCK_GROUP_RAID10))
4196 factor = 2;
4197 else
4198 factor = 1;
Josef Bacik9d66e232010-08-25 16:54:15 -04004199 /*
4200 * If this block group has free space cache written out, we
4201 * need to make sure to load it if we are removing space. This
4202 * is because we need the unpinning stage to actually add the
4203 * space back to the block group, otherwise we will leak space.
4204 */
4205 if (!alloc && cache->cached == BTRFS_CACHE_NO)
Josef Bacikb8399de2010-12-08 09:15:11 -05004206 cache_block_group(cache, trans, NULL, 1);
Josef Bacik0af3d002010-06-21 14:48:16 -04004207
Chris Masondb945352007-10-15 16:15:53 -04004208 byte_in_group = bytenr - cache->key.objectid;
4209 WARN_ON(byte_in_group > cache->key.offset);
Chris Mason9078a3e2007-04-26 16:46:15 -04004210
Josef Bacik25179202008-10-29 14:49:05 -04004211 spin_lock(&cache->space_info->lock);
Chris Masonc286ac42008-07-22 23:06:41 -04004212 spin_lock(&cache->lock);
Josef Bacik0af3d002010-06-21 14:48:16 -04004213
4214 if (btrfs_super_cache_generation(&info->super_copy) != 0 &&
4215 cache->disk_cache_state < BTRFS_DC_CLEAR)
4216 cache->disk_cache_state = BTRFS_DC_CLEAR;
4217
Josef Bacik0f9dd462008-09-23 13:14:11 -04004218 cache->dirty = 1;
Chris Mason9078a3e2007-04-26 16:46:15 -04004219 old_val = btrfs_block_group_used(&cache->item);
Chris Masondb945352007-10-15 16:15:53 -04004220 num_bytes = min(total, cache->key.offset - byte_in_group);
Chris Masoncd1bc462007-04-27 10:08:34 -04004221 if (alloc) {
Chris Masondb945352007-10-15 16:15:53 -04004222 old_val += num_bytes;
Yan Zheng11833d62009-09-11 16:11:19 -04004223 btrfs_set_block_group_used(&cache->item, old_val);
4224 cache->reserved -= num_bytes;
Yan Zheng11833d62009-09-11 16:11:19 -04004225 cache->space_info->bytes_reserved -= num_bytes;
Chris Mason36e39c42011-03-12 07:08:42 -05004226 cache->space_info->reservation_progress++;
Yan, Zhengb742bb82010-05-16 10:46:24 -04004227 cache->space_info->bytes_used += num_bytes;
4228 cache->space_info->disk_used += num_bytes * factor;
Chris Masonc286ac42008-07-22 23:06:41 -04004229 spin_unlock(&cache->lock);
Josef Bacik25179202008-10-29 14:49:05 -04004230 spin_unlock(&cache->space_info->lock);
Chris Masoncd1bc462007-04-27 10:08:34 -04004231 } else {
Chris Masondb945352007-10-15 16:15:53 -04004232 old_val -= num_bytes;
Chris Masonc286ac42008-07-22 23:06:41 -04004233 btrfs_set_block_group_used(&cache->item, old_val);
Yan, Zhengf0486c62010-05-16 10:46:25 -04004234 cache->pinned += num_bytes;
4235 cache->space_info->bytes_pinned += num_bytes;
Chris Mason9078a3e2007-04-26 16:46:15 -04004236 cache->space_info->bytes_used -= num_bytes;
Yan, Zhengb742bb82010-05-16 10:46:24 -04004237 cache->space_info->disk_used -= num_bytes * factor;
Chris Masonc286ac42008-07-22 23:06:41 -04004238 spin_unlock(&cache->lock);
Josef Bacik25179202008-10-29 14:49:05 -04004239 spin_unlock(&cache->space_info->lock);
Liu Hui1f3c79a2009-01-05 15:57:51 -05004240
Yan, Zhengf0486c62010-05-16 10:46:25 -04004241 set_extent_dirty(info->pinned_extents,
4242 bytenr, bytenr + num_bytes - 1,
4243 GFP_NOFS | __GFP_NOFAIL);
Chris Masoncd1bc462007-04-27 10:08:34 -04004244 }
Chris Masonfa9c0d72009-04-03 09:47:43 -04004245 btrfs_put_block_group(cache);
Chris Masondb945352007-10-15 16:15:53 -04004246 total -= num_bytes;
4247 bytenr += num_bytes;
Chris Mason9078a3e2007-04-26 16:46:15 -04004248 }
4249 return 0;
4250}
Chris Mason6324fbf2008-03-24 15:01:59 -04004251
Chris Masona061fc82008-05-07 11:43:44 -04004252static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
4253{
Josef Bacik0f9dd462008-09-23 13:14:11 -04004254 struct btrfs_block_group_cache *cache;
Yan Zhengd2fb3432008-12-11 16:30:39 -05004255 u64 bytenr;
Josef Bacik0f9dd462008-09-23 13:14:11 -04004256
4257 cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
4258 if (!cache)
Chris Masona061fc82008-05-07 11:43:44 -04004259 return 0;
Josef Bacik0f9dd462008-09-23 13:14:11 -04004260
Yan Zhengd2fb3432008-12-11 16:30:39 -05004261 bytenr = cache->key.objectid;
Chris Masonfa9c0d72009-04-03 09:47:43 -04004262 btrfs_put_block_group(cache);
Yan Zhengd2fb3432008-12-11 16:30:39 -05004263
4264 return bytenr;
Chris Masona061fc82008-05-07 11:43:44 -04004265}
4266
Yan, Zhengf0486c62010-05-16 10:46:25 -04004267static int pin_down_extent(struct btrfs_root *root,
4268 struct btrfs_block_group_cache *cache,
4269 u64 bytenr, u64 num_bytes, int reserved)
Yan324ae4d2007-11-16 14:57:08 -05004270{
Yan Zheng11833d62009-09-11 16:11:19 -04004271 spin_lock(&cache->space_info->lock);
4272 spin_lock(&cache->lock);
4273 cache->pinned += num_bytes;
4274 cache->space_info->bytes_pinned += num_bytes;
4275 if (reserved) {
4276 cache->reserved -= num_bytes;
4277 cache->space_info->bytes_reserved -= num_bytes;
Chris Mason36e39c42011-03-12 07:08:42 -05004278 cache->space_info->reservation_progress++;
Yan324ae4d2007-11-16 14:57:08 -05004279 }
Yan Zheng11833d62009-09-11 16:11:19 -04004280 spin_unlock(&cache->lock);
4281 spin_unlock(&cache->space_info->lock);
4282
Yan, Zhengf0486c62010-05-16 10:46:25 -04004283 set_extent_dirty(root->fs_info->pinned_extents, bytenr,
4284 bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
Yan324ae4d2007-11-16 14:57:08 -05004285 return 0;
4286}
Chris Mason9078a3e2007-04-26 16:46:15 -04004287
Yan, Zhengf0486c62010-05-16 10:46:25 -04004288/*
4289 * this function must be called within transaction
4290 */
4291int btrfs_pin_extent(struct btrfs_root *root,
4292 u64 bytenr, u64 num_bytes, int reserved)
Zheng Yane8569812008-09-26 10:05:48 -04004293{
Yan, Zhengf0486c62010-05-16 10:46:25 -04004294 struct btrfs_block_group_cache *cache;
4295
4296 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
4297 BUG_ON(!cache);
4298
4299 pin_down_extent(root, cache, bytenr, num_bytes, reserved);
4300
4301 btrfs_put_block_group(cache);
Yan Zheng11833d62009-09-11 16:11:19 -04004302 return 0;
4303}
Zheng Yane8569812008-09-26 10:05:48 -04004304
Yan, Zhengf0486c62010-05-16 10:46:25 -04004305/*
4306 * update size of reserved extents. this function may return -EAGAIN
4307 * if 'reserve' is true or 'sinfo' is false.
4308 */
Li Dongyangb4d00d52011-03-24 10:24:25 +00004309int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
4310 u64 num_bytes, int reserve, int sinfo)
Yan, Zhengf0486c62010-05-16 10:46:25 -04004311{
4312 int ret = 0;
4313 if (sinfo) {
4314 struct btrfs_space_info *space_info = cache->space_info;
4315 spin_lock(&space_info->lock);
4316 spin_lock(&cache->lock);
4317 if (reserve) {
4318 if (cache->ro) {
4319 ret = -EAGAIN;
4320 } else {
4321 cache->reserved += num_bytes;
4322 space_info->bytes_reserved += num_bytes;
4323 }
4324 } else {
4325 if (cache->ro)
4326 space_info->bytes_readonly += num_bytes;
4327 cache->reserved -= num_bytes;
4328 space_info->bytes_reserved -= num_bytes;
Chris Mason36e39c42011-03-12 07:08:42 -05004329 space_info->reservation_progress++;
Yan, Zhengf0486c62010-05-16 10:46:25 -04004330 }
4331 spin_unlock(&cache->lock);
4332 spin_unlock(&space_info->lock);
4333 } else {
4334 spin_lock(&cache->lock);
4335 if (cache->ro) {
4336 ret = -EAGAIN;
4337 } else {
4338 if (reserve)
4339 cache->reserved += num_bytes;
4340 else
4341 cache->reserved -= num_bytes;
4342 }
4343 spin_unlock(&cache->lock);
4344 }
4345 return ret;
4346}
4347
Yan Zheng11833d62009-09-11 16:11:19 -04004348int btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
4349 struct btrfs_root *root)
4350{
4351 struct btrfs_fs_info *fs_info = root->fs_info;
4352 struct btrfs_caching_control *next;
4353 struct btrfs_caching_control *caching_ctl;
4354 struct btrfs_block_group_cache *cache;
4355
4356 down_write(&fs_info->extent_commit_sem);
4357
4358 list_for_each_entry_safe(caching_ctl, next,
4359 &fs_info->caching_block_groups, list) {
4360 cache = caching_ctl->block_group;
4361 if (block_group_cache_done(cache)) {
4362 cache->last_byte_to_unpin = (u64)-1;
4363 list_del_init(&caching_ctl->list);
4364 put_caching_control(caching_ctl);
4365 } else {
4366 cache->last_byte_to_unpin = caching_ctl->progress;
4367 }
4368 }
4369
4370 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
4371 fs_info->pinned_extents = &fs_info->freed_extents[1];
4372 else
4373 fs_info->pinned_extents = &fs_info->freed_extents[0];
4374
4375 up_write(&fs_info->extent_commit_sem);
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04004376
4377 update_global_block_rsv(fs_info);
Yan Zheng11833d62009-09-11 16:11:19 -04004378 return 0;
4379}
4380
4381static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
4382{
4383 struct btrfs_fs_info *fs_info = root->fs_info;
4384 struct btrfs_block_group_cache *cache = NULL;
4385 u64 len;
4386
4387 while (start <= end) {
4388 if (!cache ||
4389 start >= cache->key.objectid + cache->key.offset) {
4390 if (cache)
4391 btrfs_put_block_group(cache);
4392 cache = btrfs_lookup_block_group(fs_info, start);
4393 BUG_ON(!cache);
4394 }
4395
4396 len = cache->key.objectid + cache->key.offset - start;
4397 len = min(len, end + 1 - start);
4398
4399 if (start < cache->last_byte_to_unpin) {
4400 len = min(len, cache->last_byte_to_unpin - start);
4401 btrfs_add_free_space(cache, start, len);
4402 }
Josef Bacik25179202008-10-29 14:49:05 -04004403
Yan, Zhengf0486c62010-05-16 10:46:25 -04004404 start += len;
4405
Josef Bacik25179202008-10-29 14:49:05 -04004406 spin_lock(&cache->space_info->lock);
4407 spin_lock(&cache->lock);
Yan Zheng11833d62009-09-11 16:11:19 -04004408 cache->pinned -= len;
4409 cache->space_info->bytes_pinned -= len;
Yan, Zhengf0486c62010-05-16 10:46:25 -04004410 if (cache->ro) {
4411 cache->space_info->bytes_readonly += len;
4412 } else if (cache->reserved_pinned > 0) {
4413 len = min(len, cache->reserved_pinned);
4414 cache->reserved_pinned -= len;
4415 cache->space_info->bytes_reserved += len;
4416 }
Josef Bacik25179202008-10-29 14:49:05 -04004417 spin_unlock(&cache->lock);
4418 spin_unlock(&cache->space_info->lock);
Yan Zheng11833d62009-09-11 16:11:19 -04004419 }
4420
4421 if (cache)
Chris Masonfa9c0d72009-04-03 09:47:43 -04004422 btrfs_put_block_group(cache);
Chris Masonccd467d2007-06-28 15:57:36 -04004423 return 0;
4424}
4425
4426int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
Yan Zheng11833d62009-09-11 16:11:19 -04004427 struct btrfs_root *root)
Chris Masona28ec192007-03-06 20:08:01 -05004428{
Yan Zheng11833d62009-09-11 16:11:19 -04004429 struct btrfs_fs_info *fs_info = root->fs_info;
4430 struct extent_io_tree *unpin;
Yan, Zhengf0486c62010-05-16 10:46:25 -04004431 struct btrfs_block_rsv *block_rsv;
4432 struct btrfs_block_rsv *next_rsv;
Chris Mason1a5bc162007-10-15 16:15:26 -04004433 u64 start;
4434 u64 end;
Yan, Zhengf0486c62010-05-16 10:46:25 -04004435 int idx;
Chris Masona28ec192007-03-06 20:08:01 -05004436 int ret;
Chris Masona28ec192007-03-06 20:08:01 -05004437
Yan Zheng11833d62009-09-11 16:11:19 -04004438 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
4439 unpin = &fs_info->freed_extents[1];
4440 else
4441 unpin = &fs_info->freed_extents[0];
4442
Chris Masond3977122009-01-05 21:25:51 -05004443 while (1) {
Chris Mason1a5bc162007-10-15 16:15:26 -04004444 ret = find_first_extent_bit(unpin, 0, &start, &end,
4445 EXTENT_DIRTY);
4446 if (ret)
Chris Masona28ec192007-03-06 20:08:01 -05004447 break;
Liu Hui1f3c79a2009-01-05 15:57:51 -05004448
Li Dongyang5378e602011-03-24 10:24:27 +00004449 if (btrfs_test_opt(root, DISCARD))
4450 ret = btrfs_discard_extent(root, start,
4451 end + 1 - start, NULL);
Liu Hui1f3c79a2009-01-05 15:57:51 -05004452
Chris Mason1a5bc162007-10-15 16:15:26 -04004453 clear_extent_dirty(unpin, start, end, GFP_NOFS);
Yan Zheng11833d62009-09-11 16:11:19 -04004454 unpin_extent_range(root, start, end);
Chris Masonb9473432009-03-13 11:00:37 -04004455 cond_resched();
Chris Masona28ec192007-03-06 20:08:01 -05004456 }
Josef Bacik817d52f2009-07-13 21:29:25 -04004457
Yan, Zhengf0486c62010-05-16 10:46:25 -04004458 mutex_lock(&fs_info->durable_block_rsv_mutex);
4459 list_for_each_entry_safe(block_rsv, next_rsv,
4460 &fs_info->durable_block_rsv_list, list) {
Chris Masona28ec192007-03-06 20:08:01 -05004461
Yan, Zhengf0486c62010-05-16 10:46:25 -04004462 idx = trans->transid & 0x1;
4463 if (block_rsv->freed[idx] > 0) {
4464 block_rsv_add_bytes(block_rsv,
4465 block_rsv->freed[idx], 0);
4466 block_rsv->freed[idx] = 0;
Chris Mason8ef97622007-03-26 10:15:30 -04004467 }
Yan, Zhengf0486c62010-05-16 10:46:25 -04004468 if (atomic_read(&block_rsv->usage) == 0) {
4469 btrfs_block_rsv_release(root, block_rsv, (u64)-1);
Zheng Yan31840ae2008-09-23 13:14:14 -04004470
Yan, Zhengf0486c62010-05-16 10:46:25 -04004471 if (block_rsv->freed[0] == 0 &&
4472 block_rsv->freed[1] == 0) {
4473 list_del_init(&block_rsv->list);
4474 kfree(block_rsv);
4475 }
4476 } else {
4477 btrfs_block_rsv_release(root, block_rsv, 0);
4478 }
4479 }
4480 mutex_unlock(&fs_info->durable_block_rsv_mutex);
4481
Chris Masone20d96d2007-03-22 12:13:20 -04004482 return 0;
4483}
4484
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004485static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
4486 struct btrfs_root *root,
4487 u64 bytenr, u64 num_bytes, u64 parent,
4488 u64 root_objectid, u64 owner_objectid,
4489 u64 owner_offset, int refs_to_drop,
4490 struct btrfs_delayed_extent_op *extent_op)
Chris Masona28ec192007-03-06 20:08:01 -05004491{
Chris Masone2fa7222007-03-12 16:22:34 -04004492 struct btrfs_key key;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004493 struct btrfs_path *path;
Chris Mason1261ec42007-03-20 20:35:03 -04004494 struct btrfs_fs_info *info = root->fs_info;
4495 struct btrfs_root *extent_root = info->extent_root;
Chris Mason5f39d392007-10-15 16:14:19 -04004496 struct extent_buffer *leaf;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004497 struct btrfs_extent_item *ei;
4498 struct btrfs_extent_inline_ref *iref;
Chris Masona28ec192007-03-06 20:08:01 -05004499 int ret;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004500 int is_data;
Chris Mason952fcca2008-02-18 16:33:44 -05004501 int extent_slot = 0;
4502 int found_extent = 0;
4503 int num_to_del = 1;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004504 u32 item_size;
4505 u64 refs;
Chris Mason037e6392007-03-07 11:50:24 -05004506
Chris Mason5caf2a02007-04-02 11:20:42 -04004507 path = btrfs_alloc_path();
Chris Mason54aa1f42007-06-22 14:16:25 -04004508 if (!path)
4509 return -ENOMEM;
4510
Chris Mason3c12ac72008-04-21 12:01:38 -04004511 path->reada = 1;
Chris Masonb9473432009-03-13 11:00:37 -04004512 path->leave_spinning = 1;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004513
4514 is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
4515 BUG_ON(!is_data && refs_to_drop != 1);
4516
4517 ret = lookup_extent_backref(trans, extent_root, path, &iref,
4518 bytenr, num_bytes, parent,
4519 root_objectid, owner_objectid,
4520 owner_offset);
Chris Mason7bb86312007-12-11 09:25:06 -05004521 if (ret == 0) {
Chris Mason952fcca2008-02-18 16:33:44 -05004522 extent_slot = path->slots[0];
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004523 while (extent_slot >= 0) {
4524 btrfs_item_key_to_cpu(path->nodes[0], &key,
Chris Mason952fcca2008-02-18 16:33:44 -05004525 extent_slot);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004526 if (key.objectid != bytenr)
Chris Mason952fcca2008-02-18 16:33:44 -05004527 break;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004528 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
4529 key.offset == num_bytes) {
Chris Mason952fcca2008-02-18 16:33:44 -05004530 found_extent = 1;
4531 break;
4532 }
4533 if (path->slots[0] - extent_slot > 5)
4534 break;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004535 extent_slot--;
Chris Mason952fcca2008-02-18 16:33:44 -05004536 }
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004537#ifdef BTRFS_COMPAT_EXTENT_TREE_V0
4538 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
4539 if (found_extent && item_size < sizeof(*ei))
4540 found_extent = 0;
4541#endif
Zheng Yan31840ae2008-09-23 13:14:14 -04004542 if (!found_extent) {
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004543 BUG_ON(iref);
Chris Mason56bec292009-03-13 10:10:06 -04004544 ret = remove_extent_backref(trans, extent_root, path,
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004545 NULL, refs_to_drop,
4546 is_data);
Zheng Yan31840ae2008-09-23 13:14:14 -04004547 BUG_ON(ret);
4548 btrfs_release_path(extent_root, path);
Chris Masonb9473432009-03-13 11:00:37 -04004549 path->leave_spinning = 1;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004550
4551 key.objectid = bytenr;
4552 key.type = BTRFS_EXTENT_ITEM_KEY;
4553 key.offset = num_bytes;
4554
Zheng Yan31840ae2008-09-23 13:14:14 -04004555 ret = btrfs_search_slot(trans, extent_root,
4556 &key, path, -1, 1);
Josef Bacikf3465ca2008-11-12 14:19:50 -05004557 if (ret) {
4558 printk(KERN_ERR "umm, got %d back from search"
Chris Masond3977122009-01-05 21:25:51 -05004559 ", was looking for %llu\n", ret,
4560 (unsigned long long)bytenr);
Josef Bacikf3465ca2008-11-12 14:19:50 -05004561 btrfs_print_leaf(extent_root, path->nodes[0]);
4562 }
Zheng Yan31840ae2008-09-23 13:14:14 -04004563 BUG_ON(ret);
4564 extent_slot = path->slots[0];
4565 }
Chris Mason7bb86312007-12-11 09:25:06 -05004566 } else {
4567 btrfs_print_leaf(extent_root, path->nodes[0]);
4568 WARN_ON(1);
Chris Masond3977122009-01-05 21:25:51 -05004569 printk(KERN_ERR "btrfs unable to find ref byte nr %llu "
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004570 "parent %llu root %llu owner %llu offset %llu\n",
Chris Masond3977122009-01-05 21:25:51 -05004571 (unsigned long long)bytenr,
Chris Mason56bec292009-03-13 10:10:06 -04004572 (unsigned long long)parent,
Chris Masond3977122009-01-05 21:25:51 -05004573 (unsigned long long)root_objectid,
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004574 (unsigned long long)owner_objectid,
4575 (unsigned long long)owner_offset);
Chris Mason7bb86312007-12-11 09:25:06 -05004576 }
Chris Mason5f39d392007-10-15 16:14:19 -04004577
4578 leaf = path->nodes[0];
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004579 item_size = btrfs_item_size_nr(leaf, extent_slot);
4580#ifdef BTRFS_COMPAT_EXTENT_TREE_V0
4581 if (item_size < sizeof(*ei)) {
4582 BUG_ON(found_extent || extent_slot != path->slots[0]);
4583 ret = convert_extent_item_v0(trans, extent_root, path,
4584 owner_objectid, 0);
4585 BUG_ON(ret < 0);
4586
4587 btrfs_release_path(extent_root, path);
4588 path->leave_spinning = 1;
4589
4590 key.objectid = bytenr;
4591 key.type = BTRFS_EXTENT_ITEM_KEY;
4592 key.offset = num_bytes;
4593
4594 ret = btrfs_search_slot(trans, extent_root, &key, path,
4595 -1, 1);
4596 if (ret) {
4597 printk(KERN_ERR "umm, got %d back from search"
4598 ", was looking for %llu\n", ret,
4599 (unsigned long long)bytenr);
4600 btrfs_print_leaf(extent_root, path->nodes[0]);
4601 }
4602 BUG_ON(ret);
4603 extent_slot = path->slots[0];
4604 leaf = path->nodes[0];
4605 item_size = btrfs_item_size_nr(leaf, extent_slot);
4606 }
4607#endif
4608 BUG_ON(item_size < sizeof(*ei));
Chris Mason952fcca2008-02-18 16:33:44 -05004609 ei = btrfs_item_ptr(leaf, extent_slot,
Chris Mason123abc82007-03-14 14:14:43 -04004610 struct btrfs_extent_item);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004611 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) {
4612 struct btrfs_tree_block_info *bi;
4613 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
4614 bi = (struct btrfs_tree_block_info *)(ei + 1);
4615 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
Chris Mason952fcca2008-02-18 16:33:44 -05004616 }
4617
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004618 refs = btrfs_extent_refs(leaf, ei);
4619 BUG_ON(refs < refs_to_drop);
4620 refs -= refs_to_drop;
4621
4622 if (refs > 0) {
4623 if (extent_op)
4624 __run_delayed_extent_op(extent_op, leaf, ei);
4625 /*
4626 * In the case of inline back ref, reference count will
4627 * be updated by remove_extent_backref
4628 */
4629 if (iref) {
4630 BUG_ON(!found_extent);
4631 } else {
4632 btrfs_set_extent_refs(leaf, ei, refs);
4633 btrfs_mark_buffer_dirty(leaf);
4634 }
4635 if (found_extent) {
4636 ret = remove_extent_backref(trans, extent_root, path,
4637 iref, refs_to_drop,
4638 is_data);
4639 BUG_ON(ret);
4640 }
4641 } else {
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004642 if (found_extent) {
4643 BUG_ON(is_data && refs_to_drop !=
4644 extent_data_ref_count(root, path, iref));
4645 if (iref) {
4646 BUG_ON(path->slots[0] != extent_slot);
4647 } else {
4648 BUG_ON(path->slots[0] != extent_slot + 1);
4649 path->slots[0] = extent_slot;
4650 num_to_del = 2;
4651 }
Chris Mason78fae272007-03-25 11:35:08 -04004652 }
Chris Masonb9473432009-03-13 11:00:37 -04004653
Chris Mason952fcca2008-02-18 16:33:44 -05004654 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
4655 num_to_del);
Zheng Yan31840ae2008-09-23 13:14:14 -04004656 BUG_ON(ret);
Josef Bacik25179202008-10-29 14:49:05 -04004657 btrfs_release_path(extent_root, path);
David Woodhouse21af8042008-08-12 14:13:26 +01004658
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004659 if (is_data) {
Chris Mason459931e2008-12-10 09:10:46 -05004660 ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
4661 BUG_ON(ret);
Chris Masond57e62b2009-03-31 13:47:50 -04004662 } else {
4663 invalidate_mapping_pages(info->btree_inode->i_mapping,
4664 bytenr >> PAGE_CACHE_SHIFT,
4665 (bytenr + num_bytes - 1) >> PAGE_CACHE_SHIFT);
Chris Mason459931e2008-12-10 09:10:46 -05004666 }
4667
Yan, Zhengf0486c62010-05-16 10:46:25 -04004668 ret = update_block_group(trans, root, bytenr, num_bytes, 0);
Chris Masondcbdd4d2008-12-16 13:51:01 -05004669 BUG_ON(ret);
Chris Masona28ec192007-03-06 20:08:01 -05004670 }
Chris Mason5caf2a02007-04-02 11:20:42 -04004671 btrfs_free_path(path);
Chris Masona28ec192007-03-06 20:08:01 -05004672 return ret;
4673}
4674
4675/*
Yan, Zhengf0486c62010-05-16 10:46:25 -04004676 * when we free an block, it is possible (and likely) that we free the last
Chris Mason1887be62009-03-13 10:11:24 -04004677 * delayed ref for that extent as well. This searches the delayed ref tree for
4678 * a given extent, and if there are no other delayed refs to be processed, it
4679 * removes it from the tree.
4680 */
4681static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
4682 struct btrfs_root *root, u64 bytenr)
4683{
4684 struct btrfs_delayed_ref_head *head;
4685 struct btrfs_delayed_ref_root *delayed_refs;
4686 struct btrfs_delayed_ref_node *ref;
4687 struct rb_node *node;
Yan, Zhengf0486c62010-05-16 10:46:25 -04004688 int ret = 0;
Chris Mason1887be62009-03-13 10:11:24 -04004689
4690 delayed_refs = &trans->transaction->delayed_refs;
4691 spin_lock(&delayed_refs->lock);
4692 head = btrfs_find_delayed_ref_head(trans, bytenr);
4693 if (!head)
4694 goto out;
4695
4696 node = rb_prev(&head->node.rb_node);
4697 if (!node)
4698 goto out;
4699
4700 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
4701
4702 /* there are still entries for this ref, we can't drop it */
4703 if (ref->bytenr == bytenr)
4704 goto out;
4705
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004706 if (head->extent_op) {
4707 if (!head->must_insert_reserved)
4708 goto out;
4709 kfree(head->extent_op);
4710 head->extent_op = NULL;
4711 }
4712
Chris Mason1887be62009-03-13 10:11:24 -04004713 /*
4714 * waiting for the lock here would deadlock. If someone else has it
4715 * locked they are already in the process of dropping it anyway
4716 */
4717 if (!mutex_trylock(&head->mutex))
4718 goto out;
4719
4720 /*
4721 * at this point we have a head with no other entries. Go
4722 * ahead and process it.
4723 */
4724 head->node.in_tree = 0;
4725 rb_erase(&head->node.rb_node, &delayed_refs->root);
Chris Masonc3e69d52009-03-13 10:17:05 -04004726
Chris Mason1887be62009-03-13 10:11:24 -04004727 delayed_refs->num_entries--;
4728
4729 /*
4730 * we don't take a ref on the node because we're removing it from the
4731 * tree, so we just steal the ref the tree was holding.
4732 */
Chris Masonc3e69d52009-03-13 10:17:05 -04004733 delayed_refs->num_heads--;
4734 if (list_empty(&head->cluster))
4735 delayed_refs->num_heads_ready--;
4736
4737 list_del_init(&head->cluster);
Chris Mason1887be62009-03-13 10:11:24 -04004738 spin_unlock(&delayed_refs->lock);
4739
Yan, Zhengf0486c62010-05-16 10:46:25 -04004740 BUG_ON(head->extent_op);
4741 if (head->must_insert_reserved)
4742 ret = 1;
4743
4744 mutex_unlock(&head->mutex);
Chris Mason1887be62009-03-13 10:11:24 -04004745 btrfs_put_delayed_ref(&head->node);
Yan, Zhengf0486c62010-05-16 10:46:25 -04004746 return ret;
Chris Mason1887be62009-03-13 10:11:24 -04004747out:
4748 spin_unlock(&delayed_refs->lock);
4749 return 0;
4750}
4751
Yan, Zhengf0486c62010-05-16 10:46:25 -04004752void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
4753 struct btrfs_root *root,
4754 struct extent_buffer *buf,
4755 u64 parent, int last_ref)
4756{
4757 struct btrfs_block_rsv *block_rsv;
4758 struct btrfs_block_group_cache *cache = NULL;
4759 int ret;
4760
4761 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
4762 ret = btrfs_add_delayed_tree_ref(trans, buf->start, buf->len,
4763 parent, root->root_key.objectid,
4764 btrfs_header_level(buf),
4765 BTRFS_DROP_DELAYED_REF, NULL);
4766 BUG_ON(ret);
4767 }
4768
4769 if (!last_ref)
4770 return;
4771
4772 block_rsv = get_block_rsv(trans, root);
4773 cache = btrfs_lookup_block_group(root->fs_info, buf->start);
Yan, Zheng3bf84a52010-05-31 09:04:46 +00004774 if (block_rsv->space_info != cache->space_info)
4775 goto out;
Yan, Zhengf0486c62010-05-16 10:46:25 -04004776
4777 if (btrfs_header_generation(buf) == trans->transid) {
4778 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
4779 ret = check_ref_cleanup(trans, root, buf->start);
4780 if (!ret)
4781 goto pin;
4782 }
4783
4784 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
4785 pin_down_extent(root, cache, buf->start, buf->len, 1);
4786 goto pin;
4787 }
4788
4789 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
4790
4791 btrfs_add_free_space(cache, buf->start, buf->len);
Li Dongyangb4d00d52011-03-24 10:24:25 +00004792 ret = btrfs_update_reserved_bytes(cache, buf->len, 0, 0);
Yan, Zhengf0486c62010-05-16 10:46:25 -04004793 if (ret == -EAGAIN) {
4794 /* block group became read-only */
Li Dongyangb4d00d52011-03-24 10:24:25 +00004795 btrfs_update_reserved_bytes(cache, buf->len, 0, 1);
Yan, Zhengf0486c62010-05-16 10:46:25 -04004796 goto out;
4797 }
4798
4799 ret = 1;
4800 spin_lock(&block_rsv->lock);
4801 if (block_rsv->reserved < block_rsv->size) {
4802 block_rsv->reserved += buf->len;
4803 ret = 0;
4804 }
4805 spin_unlock(&block_rsv->lock);
4806
4807 if (ret) {
4808 spin_lock(&cache->space_info->lock);
4809 cache->space_info->bytes_reserved -= buf->len;
Chris Mason36e39c42011-03-12 07:08:42 -05004810 cache->space_info->reservation_progress++;
Yan, Zhengf0486c62010-05-16 10:46:25 -04004811 spin_unlock(&cache->space_info->lock);
4812 }
4813 goto out;
4814 }
4815pin:
4816 if (block_rsv->durable && !cache->ro) {
4817 ret = 0;
4818 spin_lock(&cache->lock);
4819 if (!cache->ro) {
4820 cache->reserved_pinned += buf->len;
4821 ret = 1;
4822 }
4823 spin_unlock(&cache->lock);
4824
4825 if (ret) {
4826 spin_lock(&block_rsv->lock);
4827 block_rsv->freed[trans->transid & 0x1] += buf->len;
4828 spin_unlock(&block_rsv->lock);
4829 }
4830 }
4831out:
Josef Bacika826d6d2011-03-16 13:42:43 -04004832 /*
4833 * Deleting the buffer, clear the corrupt flag since it doesn't matter
4834 * anymore.
4835 */
4836 clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
Yan, Zhengf0486c62010-05-16 10:46:25 -04004837 btrfs_put_block_group(cache);
4838}
4839
Chris Mason925baed2008-06-25 16:01:30 -04004840int btrfs_free_extent(struct btrfs_trans_handle *trans,
Zheng Yan31840ae2008-09-23 13:14:14 -04004841 struct btrfs_root *root,
4842 u64 bytenr, u64 num_bytes, u64 parent,
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004843 u64 root_objectid, u64 owner, u64 offset)
Chris Mason925baed2008-06-25 16:01:30 -04004844{
4845 int ret;
4846
Chris Mason56bec292009-03-13 10:10:06 -04004847 /*
4848 * tree log blocks never actually go into the extent allocation
4849 * tree, just update pinning info and exit early.
Chris Mason56bec292009-03-13 10:10:06 -04004850 */
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004851 if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
4852 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
Chris Masonb9473432009-03-13 11:00:37 -04004853 /* unlocks the pinned mutex */
Yan Zheng11833d62009-09-11 16:11:19 -04004854 btrfs_pin_extent(root, bytenr, num_bytes, 1);
Chris Mason56bec292009-03-13 10:10:06 -04004855 ret = 0;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004856 } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
4857 ret = btrfs_add_delayed_tree_ref(trans, bytenr, num_bytes,
4858 parent, root_objectid, (int)owner,
4859 BTRFS_DROP_DELAYED_REF, NULL);
Chris Mason1887be62009-03-13 10:11:24 -04004860 BUG_ON(ret);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004861 } else {
4862 ret = btrfs_add_delayed_data_ref(trans, bytenr, num_bytes,
4863 parent, root_objectid, owner,
4864 offset, BTRFS_DROP_DELAYED_REF, NULL);
4865 BUG_ON(ret);
Chris Mason56bec292009-03-13 10:10:06 -04004866 }
Chris Mason925baed2008-06-25 16:01:30 -04004867 return ret;
4868}
4869
Chris Mason87ee04e2007-11-30 11:30:34 -05004870static u64 stripe_align(struct btrfs_root *root, u64 val)
4871{
4872 u64 mask = ((u64)root->stripesize - 1);
4873 u64 ret = (val + mask) & ~mask;
4874 return ret;
4875}
4876
Chris Masonfec577f2007-02-26 10:40:21 -05004877/*
Josef Bacik817d52f2009-07-13 21:29:25 -04004878 * when we wait for progress in the block group caching, its because
4879 * our allocation attempt failed at least once. So, we must sleep
4880 * and let some progress happen before we try again.
4881 *
4882 * This function will sleep at least once waiting for new free space to
4883 * show up, and then it will check the block group free space numbers
4884 * for our min num_bytes. Another option is to have it go ahead
4885 * and look in the rbtree for a free extent of a given size, but this
4886 * is a good start.
4887 */
4888static noinline int
4889wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
4890 u64 num_bytes)
4891{
Yan Zheng11833d62009-09-11 16:11:19 -04004892 struct btrfs_caching_control *caching_ctl;
Josef Bacik817d52f2009-07-13 21:29:25 -04004893 DEFINE_WAIT(wait);
4894
Yan Zheng11833d62009-09-11 16:11:19 -04004895 caching_ctl = get_caching_control(cache);
4896 if (!caching_ctl)
Josef Bacik817d52f2009-07-13 21:29:25 -04004897 return 0;
Josef Bacik817d52f2009-07-13 21:29:25 -04004898
Yan Zheng11833d62009-09-11 16:11:19 -04004899 wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
Josef Bacik817d52f2009-07-13 21:29:25 -04004900 (cache->free_space >= num_bytes));
Yan Zheng11833d62009-09-11 16:11:19 -04004901
4902 put_caching_control(caching_ctl);
4903 return 0;
4904}
4905
4906static noinline int
4907wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
4908{
4909 struct btrfs_caching_control *caching_ctl;
4910 DEFINE_WAIT(wait);
4911
4912 caching_ctl = get_caching_control(cache);
4913 if (!caching_ctl)
4914 return 0;
4915
4916 wait_event(caching_ctl->wait, block_group_cache_done(cache));
4917
4918 put_caching_control(caching_ctl);
Josef Bacik817d52f2009-07-13 21:29:25 -04004919 return 0;
4920}
4921
Yan, Zhengb742bb82010-05-16 10:46:24 -04004922static int get_block_group_index(struct btrfs_block_group_cache *cache)
4923{
4924 int index;
4925 if (cache->flags & BTRFS_BLOCK_GROUP_RAID10)
4926 index = 0;
4927 else if (cache->flags & BTRFS_BLOCK_GROUP_RAID1)
4928 index = 1;
4929 else if (cache->flags & BTRFS_BLOCK_GROUP_DUP)
4930 index = 2;
4931 else if (cache->flags & BTRFS_BLOCK_GROUP_RAID0)
4932 index = 3;
4933 else
4934 index = 4;
4935 return index;
4936}
4937
Josef Bacik817d52f2009-07-13 21:29:25 -04004938enum btrfs_loop_type {
Josef Bacikccf0e722009-11-10 21:23:48 -05004939 LOOP_FIND_IDEAL = 0,
Josef Bacik817d52f2009-07-13 21:29:25 -04004940 LOOP_CACHING_NOWAIT = 1,
4941 LOOP_CACHING_WAIT = 2,
4942 LOOP_ALLOC_CHUNK = 3,
4943 LOOP_NO_EMPTY_SIZE = 4,
4944};
4945
4946/*
Chris Masonfec577f2007-02-26 10:40:21 -05004947 * walks the btree of allocated extents and find a hole of a given size.
4948 * The key ins is changed to record the hole:
4949 * ins->objectid == block start
Chris Mason62e27492007-03-15 12:56:47 -04004950 * ins->flags = BTRFS_EXTENT_ITEM_KEY
Chris Masonfec577f2007-02-26 10:40:21 -05004951 * ins->offset == number of blocks
4952 * Any available blocks before search_start are skipped.
4953 */
Chris Masond3977122009-01-05 21:25:51 -05004954static noinline int find_free_extent(struct btrfs_trans_handle *trans,
Chris Mason98ed5172008-01-03 10:01:48 -05004955 struct btrfs_root *orig_root,
4956 u64 num_bytes, u64 empty_size,
4957 u64 search_start, u64 search_end,
4958 u64 hint_byte, struct btrfs_key *ins,
Chris Mason98ed5172008-01-03 10:01:48 -05004959 int data)
Chris Masonfec577f2007-02-26 10:40:21 -05004960{
Josef Bacik80eb2342008-10-29 14:49:05 -04004961 int ret = 0;
Chris Masond3977122009-01-05 21:25:51 -05004962 struct btrfs_root *root = orig_root->fs_info->extent_root;
Chris Masonfa9c0d72009-04-03 09:47:43 -04004963 struct btrfs_free_cluster *last_ptr = NULL;
Josef Bacik80eb2342008-10-29 14:49:05 -04004964 struct btrfs_block_group_cache *block_group = NULL;
Chris Mason239b14b2008-03-24 15:02:07 -04004965 int empty_cluster = 2 * 1024 * 1024;
Chris Mason0ef3e662008-05-24 14:04:53 -04004966 int allowed_chunk_alloc = 0;
Josef Bacikccf0e722009-11-10 21:23:48 -05004967 int done_chunk_alloc = 0;
Josef Bacik80eb2342008-10-29 14:49:05 -04004968 struct btrfs_space_info *space_info;
Chris Masonfa9c0d72009-04-03 09:47:43 -04004969 int last_ptr_loop = 0;
4970 int loop = 0;
Yan, Zhengf0486c62010-05-16 10:46:25 -04004971 int index = 0;
Josef Bacik817d52f2009-07-13 21:29:25 -04004972 bool found_uncached_bg = false;
Josef Bacik0a243252009-09-11 16:11:20 -04004973 bool failed_cluster_refill = false;
Josef Bacik1cdda9b2009-10-06 10:04:28 -04004974 bool failed_alloc = false;
Josef Bacik67377732010-09-16 16:19:09 -04004975 bool use_cluster = true;
Josef Bacikccf0e722009-11-10 21:23:48 -05004976 u64 ideal_cache_percent = 0;
4977 u64 ideal_cache_offset = 0;
Chris Masonfec577f2007-02-26 10:40:21 -05004978
Chris Masondb945352007-10-15 16:15:53 -04004979 WARN_ON(num_bytes < root->sectorsize);
Chris Masonb1a4d962007-04-04 15:27:52 -04004980 btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
Josef Bacik80eb2342008-10-29 14:49:05 -04004981 ins->objectid = 0;
4982 ins->offset = 0;
Chris Masonb1a4d962007-04-04 15:27:52 -04004983
Josef Bacik2552d172009-04-03 10:14:19 -04004984 space_info = __find_space_info(root->fs_info, data);
Josef Bacik1b1d1f62010-03-19 20:49:55 +00004985 if (!space_info) {
4986 printk(KERN_ERR "No space info for %d\n", data);
4987 return -ENOSPC;
4988 }
Josef Bacik2552d172009-04-03 10:14:19 -04004989
Josef Bacik67377732010-09-16 16:19:09 -04004990 /*
4991 * If the space info is for both data and metadata it means we have a
4992 * small filesystem and we can't use the clustering stuff.
4993 */
4994 if (btrfs_mixed_space_info(space_info))
4995 use_cluster = false;
4996
Chris Mason0ef3e662008-05-24 14:04:53 -04004997 if (orig_root->ref_cows || empty_size)
4998 allowed_chunk_alloc = 1;
4999
Josef Bacik67377732010-09-16 16:19:09 -04005000 if (data & BTRFS_BLOCK_GROUP_METADATA && use_cluster) {
Chris Masonfa9c0d72009-04-03 09:47:43 -04005001 last_ptr = &root->fs_info->meta_alloc_cluster;
Chris Mason536ac8a2009-02-12 09:41:38 -05005002 if (!btrfs_test_opt(root, SSD))
5003 empty_cluster = 64 * 1024;
Chris Mason239b14b2008-03-24 15:02:07 -04005004 }
5005
Josef Bacik67377732010-09-16 16:19:09 -04005006 if ((data & BTRFS_BLOCK_GROUP_DATA) && use_cluster &&
5007 btrfs_test_opt(root, SSD)) {
Chris Masonfa9c0d72009-04-03 09:47:43 -04005008 last_ptr = &root->fs_info->data_alloc_cluster;
5009 }
Josef Bacik0f9dd462008-09-23 13:14:11 -04005010
Chris Mason239b14b2008-03-24 15:02:07 -04005011 if (last_ptr) {
Chris Masonfa9c0d72009-04-03 09:47:43 -04005012 spin_lock(&last_ptr->lock);
5013 if (last_ptr->block_group)
5014 hint_byte = last_ptr->window_start;
5015 spin_unlock(&last_ptr->lock);
Chris Mason239b14b2008-03-24 15:02:07 -04005016 }
Chris Masonfa9c0d72009-04-03 09:47:43 -04005017
Chris Masona061fc82008-05-07 11:43:44 -04005018 search_start = max(search_start, first_logical_byte(root, 0));
Chris Mason239b14b2008-03-24 15:02:07 -04005019 search_start = max(search_start, hint_byte);
Chris Mason0b86a832008-03-24 15:01:56 -04005020
Josef Bacik817d52f2009-07-13 21:29:25 -04005021 if (!last_ptr)
Chris Masonfa9c0d72009-04-03 09:47:43 -04005022 empty_cluster = 0;
Chris Masonfa9c0d72009-04-03 09:47:43 -04005023
Josef Bacik2552d172009-04-03 10:14:19 -04005024 if (search_start == hint_byte) {
Josef Bacikccf0e722009-11-10 21:23:48 -05005025ideal_cache:
Josef Bacik2552d172009-04-03 10:14:19 -04005026 block_group = btrfs_lookup_block_group(root->fs_info,
5027 search_start);
Josef Bacik817d52f2009-07-13 21:29:25 -04005028 /*
5029 * we don't want to use the block group if it doesn't match our
5030 * allocation bits, or if its not cached.
Josef Bacikccf0e722009-11-10 21:23:48 -05005031 *
5032 * However if we are re-searching with an ideal block group
5033 * picked out then we don't care that the block group is cached.
Josef Bacik817d52f2009-07-13 21:29:25 -04005034 */
5035 if (block_group && block_group_bits(block_group, data) &&
Josef Bacikccf0e722009-11-10 21:23:48 -05005036 (block_group->cached != BTRFS_CACHE_NO ||
5037 search_start == ideal_cache_offset)) {
Josef Bacik2552d172009-04-03 10:14:19 -04005038 down_read(&space_info->groups_sem);
Chris Mason44fb5512009-06-04 15:34:51 -04005039 if (list_empty(&block_group->list) ||
5040 block_group->ro) {
5041 /*
5042 * someone is removing this block group,
5043 * we can't jump into the have_block_group
5044 * target because our list pointers are not
5045 * valid
5046 */
5047 btrfs_put_block_group(block_group);
5048 up_read(&space_info->groups_sem);
Josef Bacikccf0e722009-11-10 21:23:48 -05005049 } else {
Yan, Zhengb742bb82010-05-16 10:46:24 -04005050 index = get_block_group_index(block_group);
Chris Mason44fb5512009-06-04 15:34:51 -04005051 goto have_block_group;
Josef Bacikccf0e722009-11-10 21:23:48 -05005052 }
Josef Bacik2552d172009-04-03 10:14:19 -04005053 } else if (block_group) {
Chris Masonfa9c0d72009-04-03 09:47:43 -04005054 btrfs_put_block_group(block_group);
Josef Bacik2552d172009-04-03 10:14:19 -04005055 }
Chris Mason42e70e72008-11-07 18:17:11 -05005056 }
Josef Bacik2552d172009-04-03 10:14:19 -04005057search:
Josef Bacik80eb2342008-10-29 14:49:05 -04005058 down_read(&space_info->groups_sem);
Yan, Zhengb742bb82010-05-16 10:46:24 -04005059 list_for_each_entry(block_group, &space_info->block_groups[index],
5060 list) {
Josef Bacik6226cb02009-04-03 10:14:18 -04005061 u64 offset;
Josef Bacik817d52f2009-07-13 21:29:25 -04005062 int cached;
Chris Mason8a1413a2008-11-10 16:13:54 -05005063
Josef Bacik11dfe352009-11-13 20:12:59 +00005064 btrfs_get_block_group(block_group);
Josef Bacik2552d172009-04-03 10:14:19 -04005065 search_start = block_group->key.objectid;
Chris Mason42e70e72008-11-07 18:17:11 -05005066
Chris Mason83a50de2010-12-13 15:06:46 -05005067 /*
5068 * this can happen if we end up cycling through all the
5069 * raid types, but we want to make sure we only allocate
5070 * for the proper type.
5071 */
5072 if (!block_group_bits(block_group, data)) {
5073 u64 extra = BTRFS_BLOCK_GROUP_DUP |
5074 BTRFS_BLOCK_GROUP_RAID1 |
5075 BTRFS_BLOCK_GROUP_RAID10;
5076
5077 /*
5078 * if they asked for extra copies and this block group
5079 * doesn't provide them, bail. This does allow us to
5080 * fill raid0 from raid1.
5081 */
5082 if ((data & extra) && !(block_group->flags & extra))
5083 goto loop;
5084 }
5085
Josef Bacik2552d172009-04-03 10:14:19 -04005086have_block_group:
Josef Bacik817d52f2009-07-13 21:29:25 -04005087 if (unlikely(block_group->cached == BTRFS_CACHE_NO)) {
Josef Bacikccf0e722009-11-10 21:23:48 -05005088 u64 free_percent;
5089
Josef Bacikb8399de2010-12-08 09:15:11 -05005090 ret = cache_block_group(block_group, trans,
5091 orig_root, 1);
Josef Bacik9d66e232010-08-25 16:54:15 -04005092 if (block_group->cached == BTRFS_CACHE_FINISHED)
5093 goto have_block_group;
5094
Josef Bacikccf0e722009-11-10 21:23:48 -05005095 free_percent = btrfs_block_group_used(&block_group->item);
5096 free_percent *= 100;
5097 free_percent = div64_u64(free_percent,
5098 block_group->key.offset);
5099 free_percent = 100 - free_percent;
5100 if (free_percent > ideal_cache_percent &&
5101 likely(!block_group->ro)) {
5102 ideal_cache_offset = block_group->key.objectid;
5103 ideal_cache_percent = free_percent;
5104 }
5105
Josef Bacik817d52f2009-07-13 21:29:25 -04005106 /*
Josef Bacikccf0e722009-11-10 21:23:48 -05005107 * We only want to start kthread caching if we are at
5108 * the point where we will wait for caching to make
5109 * progress, or if our ideal search is over and we've
5110 * found somebody to start caching.
Josef Bacik817d52f2009-07-13 21:29:25 -04005111 */
5112 if (loop > LOOP_CACHING_NOWAIT ||
Josef Bacikccf0e722009-11-10 21:23:48 -05005113 (loop > LOOP_FIND_IDEAL &&
5114 atomic_read(&space_info->caching_threads) < 2)) {
Josef Bacikb8399de2010-12-08 09:15:11 -05005115 ret = cache_block_group(block_group, trans,
5116 orig_root, 0);
Josef Bacik817d52f2009-07-13 21:29:25 -04005117 BUG_ON(ret);
Josef Bacik2552d172009-04-03 10:14:19 -04005118 }
Josef Bacikccf0e722009-11-10 21:23:48 -05005119 found_uncached_bg = true;
5120
5121 /*
5122 * If loop is set for cached only, try the next block
5123 * group.
5124 */
5125 if (loop == LOOP_FIND_IDEAL)
5126 goto loop;
Josef Bacikea6a4782008-11-20 12:16:16 -05005127 }
5128
Josef Bacik817d52f2009-07-13 21:29:25 -04005129 cached = block_group_cache_done(block_group);
Josef Bacikccf0e722009-11-10 21:23:48 -05005130 if (unlikely(!cached))
Josef Bacik817d52f2009-07-13 21:29:25 -04005131 found_uncached_bg = true;
5132
Josef Bacikea6a4782008-11-20 12:16:16 -05005133 if (unlikely(block_group->ro))
Josef Bacik2552d172009-04-03 10:14:19 -04005134 goto loop;
Josef Bacik0f9dd462008-09-23 13:14:11 -04005135
Josef Bacik0a243252009-09-11 16:11:20 -04005136 /*
5137 * Ok we want to try and use the cluster allocator, so lets look
5138 * there, unless we are on LOOP_NO_EMPTY_SIZE, since we will
5139 * have tried the cluster allocator plenty of times at this
5140 * point and not have found anything, so we are likely way too
5141 * fragmented for the clustering stuff to find anything, so lets
5142 * just skip it and let the allocator find whatever block it can
5143 * find
5144 */
5145 if (last_ptr && loop < LOOP_NO_EMPTY_SIZE) {
Chris Masonfa9c0d72009-04-03 09:47:43 -04005146 /*
5147 * the refill lock keeps out other
5148 * people trying to start a new cluster
5149 */
5150 spin_lock(&last_ptr->refill_lock);
Chris Mason44fb5512009-06-04 15:34:51 -04005151 if (last_ptr->block_group &&
5152 (last_ptr->block_group->ro ||
5153 !block_group_bits(last_ptr->block_group, data))) {
5154 offset = 0;
5155 goto refill_cluster;
5156 }
5157
Chris Masonfa9c0d72009-04-03 09:47:43 -04005158 offset = btrfs_alloc_from_cluster(block_group, last_ptr,
5159 num_bytes, search_start);
5160 if (offset) {
5161 /* we have a block, we're done */
5162 spin_unlock(&last_ptr->refill_lock);
5163 goto checks;
5164 }
5165
5166 spin_lock(&last_ptr->lock);
5167 /*
5168 * whoops, this cluster doesn't actually point to
5169 * this block group. Get a ref on the block
5170 * group is does point to and try again
5171 */
5172 if (!last_ptr_loop && last_ptr->block_group &&
5173 last_ptr->block_group != block_group) {
5174
5175 btrfs_put_block_group(block_group);
5176 block_group = last_ptr->block_group;
Josef Bacik11dfe352009-11-13 20:12:59 +00005177 btrfs_get_block_group(block_group);
Chris Masonfa9c0d72009-04-03 09:47:43 -04005178 spin_unlock(&last_ptr->lock);
5179 spin_unlock(&last_ptr->refill_lock);
5180
5181 last_ptr_loop = 1;
5182 search_start = block_group->key.objectid;
Chris Mason44fb5512009-06-04 15:34:51 -04005183 /*
5184 * we know this block group is properly
5185 * in the list because
5186 * btrfs_remove_block_group, drops the
5187 * cluster before it removes the block
5188 * group from the list
5189 */
Chris Masonfa9c0d72009-04-03 09:47:43 -04005190 goto have_block_group;
5191 }
5192 spin_unlock(&last_ptr->lock);
Chris Mason44fb5512009-06-04 15:34:51 -04005193refill_cluster:
Chris Masonfa9c0d72009-04-03 09:47:43 -04005194 /*
5195 * this cluster didn't work out, free it and
5196 * start over
5197 */
5198 btrfs_return_cluster_to_free_space(NULL, last_ptr);
5199
5200 last_ptr_loop = 0;
5201
5202 /* allocate a cluster in this block group */
Chris Mason451d7582009-06-09 20:28:34 -04005203 ret = btrfs_find_space_cluster(trans, root,
Chris Masonfa9c0d72009-04-03 09:47:43 -04005204 block_group, last_ptr,
5205 offset, num_bytes,
5206 empty_cluster + empty_size);
5207 if (ret == 0) {
5208 /*
5209 * now pull our allocation out of this
5210 * cluster
5211 */
5212 offset = btrfs_alloc_from_cluster(block_group,
5213 last_ptr, num_bytes,
5214 search_start);
5215 if (offset) {
5216 /* we found one, proceed */
5217 spin_unlock(&last_ptr->refill_lock);
5218 goto checks;
5219 }
Josef Bacik0a243252009-09-11 16:11:20 -04005220 } else if (!cached && loop > LOOP_CACHING_NOWAIT
5221 && !failed_cluster_refill) {
Josef Bacik817d52f2009-07-13 21:29:25 -04005222 spin_unlock(&last_ptr->refill_lock);
5223
Josef Bacik0a243252009-09-11 16:11:20 -04005224 failed_cluster_refill = true;
Josef Bacik817d52f2009-07-13 21:29:25 -04005225 wait_block_group_cache_progress(block_group,
5226 num_bytes + empty_cluster + empty_size);
5227 goto have_block_group;
Chris Masonfa9c0d72009-04-03 09:47:43 -04005228 }
Josef Bacik817d52f2009-07-13 21:29:25 -04005229
Chris Masonfa9c0d72009-04-03 09:47:43 -04005230 /*
5231 * at this point we either didn't find a cluster
5232 * or we weren't able to allocate a block from our
5233 * cluster. Free the cluster we've been trying
5234 * to use, and go to the next block group
5235 */
Josef Bacik0a243252009-09-11 16:11:20 -04005236 btrfs_return_cluster_to_free_space(NULL, last_ptr);
Chris Masonfa9c0d72009-04-03 09:47:43 -04005237 spin_unlock(&last_ptr->refill_lock);
Josef Bacik0a243252009-09-11 16:11:20 -04005238 goto loop;
Chris Masonfa9c0d72009-04-03 09:47:43 -04005239 }
5240
Josef Bacik6226cb02009-04-03 10:14:18 -04005241 offset = btrfs_find_space_for_alloc(block_group, search_start,
5242 num_bytes, empty_size);
Josef Bacik1cdda9b2009-10-06 10:04:28 -04005243 /*
5244 * If we didn't find a chunk, and we haven't failed on this
5245 * block group before, and this block group is in the middle of
5246 * caching and we are ok with waiting, then go ahead and wait
5247 * for progress to be made, and set failed_alloc to true.
5248 *
5249 * If failed_alloc is true then we've already waited on this
5250 * block group once and should move on to the next block group.
5251 */
5252 if (!offset && !failed_alloc && !cached &&
5253 loop > LOOP_CACHING_NOWAIT) {
Josef Bacik817d52f2009-07-13 21:29:25 -04005254 wait_block_group_cache_progress(block_group,
Josef Bacik1cdda9b2009-10-06 10:04:28 -04005255 num_bytes + empty_size);
5256 failed_alloc = true;
Josef Bacik817d52f2009-07-13 21:29:25 -04005257 goto have_block_group;
Josef Bacik1cdda9b2009-10-06 10:04:28 -04005258 } else if (!offset) {
5259 goto loop;
Josef Bacik817d52f2009-07-13 21:29:25 -04005260 }
Chris Masonfa9c0d72009-04-03 09:47:43 -04005261checks:
Josef Bacik6226cb02009-04-03 10:14:18 -04005262 search_start = stripe_align(root, offset);
Josef Bacik2552d172009-04-03 10:14:19 -04005263 /* move on to the next group */
Josef Bacik6226cb02009-04-03 10:14:18 -04005264 if (search_start + num_bytes >= search_end) {
5265 btrfs_add_free_space(block_group, offset, num_bytes);
Josef Bacik2552d172009-04-03 10:14:19 -04005266 goto loop;
Josef Bacik6226cb02009-04-03 10:14:18 -04005267 }
Chris Masone37c9e62007-05-09 20:13:14 -04005268
Josef Bacik2552d172009-04-03 10:14:19 -04005269 /* move on to the next group */
5270 if (search_start + num_bytes >
Josef Bacik6226cb02009-04-03 10:14:18 -04005271 block_group->key.objectid + block_group->key.offset) {
5272 btrfs_add_free_space(block_group, offset, num_bytes);
Josef Bacik2552d172009-04-03 10:14:19 -04005273 goto loop;
Josef Bacik6226cb02009-04-03 10:14:18 -04005274 }
Josef Bacik80eb2342008-10-29 14:49:05 -04005275
Josef Bacik2552d172009-04-03 10:14:19 -04005276 ins->objectid = search_start;
5277 ins->offset = num_bytes;
5278
Josef Bacik6226cb02009-04-03 10:14:18 -04005279 if (offset < search_start)
5280 btrfs_add_free_space(block_group, offset,
5281 search_start - offset);
5282 BUG_ON(offset > search_start);
5283
Li Dongyangb4d00d52011-03-24 10:24:25 +00005284 ret = btrfs_update_reserved_bytes(block_group, num_bytes, 1,
Yan, Zhengf0486c62010-05-16 10:46:25 -04005285 (data & BTRFS_BLOCK_GROUP_DATA));
5286 if (ret == -EAGAIN) {
5287 btrfs_add_free_space(block_group, offset, num_bytes);
5288 goto loop;
5289 }
Yan Zheng11833d62009-09-11 16:11:19 -04005290
Josef Bacik2552d172009-04-03 10:14:19 -04005291 /* we are all good, lets return */
Yan, Zhengf0486c62010-05-16 10:46:25 -04005292 ins->objectid = search_start;
5293 ins->offset = num_bytes;
5294
5295 if (offset < search_start)
5296 btrfs_add_free_space(block_group, offset,
5297 search_start - offset);
5298 BUG_ON(offset > search_start);
Josef Bacik2552d172009-04-03 10:14:19 -04005299 break;
5300loop:
Josef Bacik0a243252009-09-11 16:11:20 -04005301 failed_cluster_refill = false;
Josef Bacik1cdda9b2009-10-06 10:04:28 -04005302 failed_alloc = false;
Yan, Zhengb742bb82010-05-16 10:46:24 -04005303 BUG_ON(index != get_block_group_index(block_group));
Chris Masonfa9c0d72009-04-03 09:47:43 -04005304 btrfs_put_block_group(block_group);
Josef Bacik2552d172009-04-03 10:14:19 -04005305 }
5306 up_read(&space_info->groups_sem);
Chris Masonf5a31e12008-11-10 11:47:09 -05005307
Yan, Zhengb742bb82010-05-16 10:46:24 -04005308 if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
5309 goto search;
5310
Josef Bacikccf0e722009-11-10 21:23:48 -05005311 /* LOOP_FIND_IDEAL, only search caching/cached bg's, and don't wait for
5312 * for them to make caching progress. Also
5313 * determine the best possible bg to cache
5314 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
5315 * caching kthreads as we move along
Josef Bacik817d52f2009-07-13 21:29:25 -04005316 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
5317 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
5318 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
5319 * again
Chris Masonfa9c0d72009-04-03 09:47:43 -04005320 */
Josef Bacik817d52f2009-07-13 21:29:25 -04005321 if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE &&
5322 (found_uncached_bg || empty_size || empty_cluster ||
5323 allowed_chunk_alloc)) {
Yan, Zhengb742bb82010-05-16 10:46:24 -04005324 index = 0;
Josef Bacikccf0e722009-11-10 21:23:48 -05005325 if (loop == LOOP_FIND_IDEAL && found_uncached_bg) {
Josef Bacik817d52f2009-07-13 21:29:25 -04005326 found_uncached_bg = false;
Josef Bacikccf0e722009-11-10 21:23:48 -05005327 loop++;
5328 if (!ideal_cache_percent &&
5329 atomic_read(&space_info->caching_threads))
Josef Bacik817d52f2009-07-13 21:29:25 -04005330 goto search;
Josef Bacikccf0e722009-11-10 21:23:48 -05005331
5332 /*
5333 * 1 of the following 2 things have happened so far
5334 *
5335 * 1) We found an ideal block group for caching that
5336 * is mostly full and will cache quickly, so we might
5337 * as well wait for it.
5338 *
5339 * 2) We searched for cached only and we didn't find
5340 * anything, and we didn't start any caching kthreads
5341 * either, so chances are we will loop through and
5342 * start a couple caching kthreads, and then come back
5343 * around and just wait for them. This will be slower
5344 * because we will have 2 caching kthreads reading at
5345 * the same time when we could have just started one
5346 * and waited for it to get far enough to give us an
5347 * allocation, so go ahead and go to the wait caching
5348 * loop.
5349 */
5350 loop = LOOP_CACHING_WAIT;
5351 search_start = ideal_cache_offset;
5352 ideal_cache_percent = 0;
5353 goto ideal_cache;
5354 } else if (loop == LOOP_FIND_IDEAL) {
5355 /*
5356 * Didn't find a uncached bg, wait on anything we find
5357 * next.
5358 */
5359 loop = LOOP_CACHING_WAIT;
5360 goto search;
5361 }
5362
5363 if (loop < LOOP_CACHING_WAIT) {
5364 loop++;
5365 goto search;
Josef Bacik817d52f2009-07-13 21:29:25 -04005366 }
5367
5368 if (loop == LOOP_ALLOC_CHUNK) {
Chris Masonfa9c0d72009-04-03 09:47:43 -04005369 empty_size = 0;
5370 empty_cluster = 0;
5371 }
Chris Mason42e70e72008-11-07 18:17:11 -05005372
Josef Bacik2552d172009-04-03 10:14:19 -04005373 if (allowed_chunk_alloc) {
5374 ret = do_chunk_alloc(trans, root, num_bytes +
Chris Mason0e4f8f82011-04-15 16:05:44 -04005375 2 * 1024 * 1024, data,
5376 CHUNK_ALLOC_LIMITED);
Josef Bacik2552d172009-04-03 10:14:19 -04005377 allowed_chunk_alloc = 0;
Josef Bacikccf0e722009-11-10 21:23:48 -05005378 done_chunk_alloc = 1;
Chris Mason0e4f8f82011-04-15 16:05:44 -04005379 } else if (!done_chunk_alloc &&
5380 space_info->force_alloc == CHUNK_ALLOC_NO_FORCE) {
5381 space_info->force_alloc = CHUNK_ALLOC_LIMITED;
Josef Bacik0f9dd462008-09-23 13:14:11 -04005382 }
Josef Bacik80eb2342008-10-29 14:49:05 -04005383
Josef Bacik817d52f2009-07-13 21:29:25 -04005384 if (loop < LOOP_NO_EMPTY_SIZE) {
Chris Masonfa9c0d72009-04-03 09:47:43 -04005385 loop++;
Josef Bacik2552d172009-04-03 10:14:19 -04005386 goto search;
Chris Masonfa9c0d72009-04-03 09:47:43 -04005387 }
Josef Bacik2552d172009-04-03 10:14:19 -04005388 ret = -ENOSPC;
5389 } else if (!ins->objectid) {
5390 ret = -ENOSPC;
Chris Masone19caa52007-10-15 16:17:44 -04005391 }
Chris Mason0b86a832008-03-24 15:01:56 -04005392
Josef Bacik80eb2342008-10-29 14:49:05 -04005393 /* we found what we needed */
5394 if (ins->objectid) {
5395 if (!(data & BTRFS_BLOCK_GROUP_DATA))
Yan Zhengd2fb3432008-12-11 16:30:39 -05005396 trans->block_group = block_group->key.objectid;
Josef Bacik80eb2342008-10-29 14:49:05 -04005397
Chris Masonfa9c0d72009-04-03 09:47:43 -04005398 btrfs_put_block_group(block_group);
Josef Bacik2552d172009-04-03 10:14:19 -04005399 ret = 0;
5400 }
Chris Mason0b86a832008-03-24 15:01:56 -04005401
Chris Mason0f70abe2007-02-28 16:46:22 -05005402 return ret;
Chris Masonfec577f2007-02-26 10:40:21 -05005403}
Chris Masonec44a352008-04-28 15:29:52 -04005404
Josef Bacik9ed74f22009-09-11 16:12:44 -04005405static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
5406 int dump_block_groups)
Josef Bacik0f9dd462008-09-23 13:14:11 -04005407{
5408 struct btrfs_block_group_cache *cache;
Yan, Zhengb742bb82010-05-16 10:46:24 -04005409 int index = 0;
Josef Bacik0f9dd462008-09-23 13:14:11 -04005410
Josef Bacik9ed74f22009-09-11 16:12:44 -04005411 spin_lock(&info->lock);
Chris Masond3977122009-01-05 21:25:51 -05005412 printk(KERN_INFO "space_info has %llu free, is %sfull\n",
5413 (unsigned long long)(info->total_bytes - info->bytes_used -
Josef Bacik9ed74f22009-09-11 16:12:44 -04005414 info->bytes_pinned - info->bytes_reserved -
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04005415 info->bytes_readonly),
Chris Masond3977122009-01-05 21:25:51 -05005416 (info->full) ? "" : "not ");
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04005417 printk(KERN_INFO "space_info total=%llu, used=%llu, pinned=%llu, "
5418 "reserved=%llu, may_use=%llu, readonly=%llu\n",
Joel Becker21380932009-04-21 12:38:29 -07005419 (unsigned long long)info->total_bytes,
Josef Bacik9ed74f22009-09-11 16:12:44 -04005420 (unsigned long long)info->bytes_used,
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04005421 (unsigned long long)info->bytes_pinned,
5422 (unsigned long long)info->bytes_reserved,
5423 (unsigned long long)info->bytes_may_use,
5424 (unsigned long long)info->bytes_readonly);
Josef Bacik9ed74f22009-09-11 16:12:44 -04005425 spin_unlock(&info->lock);
5426
5427 if (!dump_block_groups)
5428 return;
Josef Bacik0f9dd462008-09-23 13:14:11 -04005429
Josef Bacik80eb2342008-10-29 14:49:05 -04005430 down_read(&info->groups_sem);
Yan, Zhengb742bb82010-05-16 10:46:24 -04005431again:
5432 list_for_each_entry(cache, &info->block_groups[index], list) {
Josef Bacik0f9dd462008-09-23 13:14:11 -04005433 spin_lock(&cache->lock);
Chris Masond3977122009-01-05 21:25:51 -05005434 printk(KERN_INFO "block group %llu has %llu bytes, %llu used "
5435 "%llu pinned %llu reserved\n",
5436 (unsigned long long)cache->key.objectid,
5437 (unsigned long long)cache->key.offset,
5438 (unsigned long long)btrfs_block_group_used(&cache->item),
5439 (unsigned long long)cache->pinned,
5440 (unsigned long long)cache->reserved);
Josef Bacik0f9dd462008-09-23 13:14:11 -04005441 btrfs_dump_free_space(cache, bytes);
5442 spin_unlock(&cache->lock);
5443 }
Yan, Zhengb742bb82010-05-16 10:46:24 -04005444 if (++index < BTRFS_NR_RAID_TYPES)
5445 goto again;
Josef Bacik80eb2342008-10-29 14:49:05 -04005446 up_read(&info->groups_sem);
Josef Bacik0f9dd462008-09-23 13:14:11 -04005447}
Zheng Yane8569812008-09-26 10:05:48 -04005448
Yan Zheng11833d62009-09-11 16:11:19 -04005449int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
5450 struct btrfs_root *root,
5451 u64 num_bytes, u64 min_alloc_size,
5452 u64 empty_size, u64 hint_byte,
5453 u64 search_end, struct btrfs_key *ins,
5454 u64 data)
Chris Masonfec577f2007-02-26 10:40:21 -05005455{
5456 int ret;
Chris Masonfbdc7622007-05-30 10:22:12 -04005457 u64 search_start = 0;
Chris Mason925baed2008-06-25 16:01:30 -04005458
Josef Bacik6a632092009-02-20 11:00:09 -05005459 data = btrfs_get_alloc_profile(root, data);
Chris Mason98d20f62008-04-14 09:46:10 -04005460again:
Chris Mason0ef3e662008-05-24 14:04:53 -04005461 /*
5462 * the only place that sets empty_size is btrfs_realloc_node, which
5463 * is not called recursively on allocations
5464 */
Josef Bacik83d3c962009-12-07 21:45:59 +00005465 if (empty_size || root->ref_cows)
Chris Mason6324fbf2008-03-24 15:01:59 -04005466 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
Chris Mason0e4f8f82011-04-15 16:05:44 -04005467 num_bytes + 2 * 1024 * 1024, data,
5468 CHUNK_ALLOC_NO_FORCE);
Chris Mason0b86a832008-03-24 15:01:56 -04005469
Chris Masondb945352007-10-15 16:15:53 -04005470 WARN_ON(num_bytes < root->sectorsize);
5471 ret = find_free_extent(trans, root, num_bytes, empty_size,
Yan, Zhengf0486c62010-05-16 10:46:25 -04005472 search_start, search_end, hint_byte,
5473 ins, data);
Chris Mason3b951512008-04-17 11:29:12 -04005474
Chris Mason98d20f62008-04-14 09:46:10 -04005475 if (ret == -ENOSPC && num_bytes > min_alloc_size) {
5476 num_bytes = num_bytes >> 1;
Josef Bacik0f9dd462008-09-23 13:14:11 -04005477 num_bytes = num_bytes & ~(root->sectorsize - 1);
Chris Mason98d20f62008-04-14 09:46:10 -04005478 num_bytes = max(num_bytes, min_alloc_size);
Chris Mason0ef3e662008-05-24 14:04:53 -04005479 do_chunk_alloc(trans, root->fs_info->extent_root,
Chris Mason0e4f8f82011-04-15 16:05:44 -04005480 num_bytes, data, CHUNK_ALLOC_FORCE);
Chris Mason98d20f62008-04-14 09:46:10 -04005481 goto again;
5482 }
Chris Mason91435652011-02-16 13:10:41 -05005483 if (ret == -ENOSPC && btrfs_test_opt(root, ENOSPC_DEBUG)) {
Josef Bacik0f9dd462008-09-23 13:14:11 -04005484 struct btrfs_space_info *sinfo;
5485
5486 sinfo = __find_space_info(root->fs_info, data);
Chris Masond3977122009-01-05 21:25:51 -05005487 printk(KERN_ERR "btrfs allocation failed flags %llu, "
5488 "wanted %llu\n", (unsigned long long)data,
5489 (unsigned long long)num_bytes);
Josef Bacik9ed74f22009-09-11 16:12:44 -04005490 dump_space_info(sinfo, num_bytes, 1);
Chris Mason925baed2008-06-25 16:01:30 -04005491 }
Josef Bacik0f9dd462008-09-23 13:14:11 -04005492
liubo1abe9b82011-03-24 11:18:59 +00005493 trace_btrfs_reserved_extent_alloc(root, ins->objectid, ins->offset);
5494
Josef Bacik0f9dd462008-09-23 13:14:11 -04005495 return ret;
Chris Masone6dcd2d2008-07-17 12:53:50 -04005496}
5497
Chris Mason65b51a02008-08-01 15:11:20 -04005498int btrfs_free_reserved_extent(struct btrfs_root *root, u64 start, u64 len)
5499{
Josef Bacik0f9dd462008-09-23 13:14:11 -04005500 struct btrfs_block_group_cache *cache;
Liu Hui1f3c79a2009-01-05 15:57:51 -05005501 int ret = 0;
Josef Bacik0f9dd462008-09-23 13:14:11 -04005502
Josef Bacik0f9dd462008-09-23 13:14:11 -04005503 cache = btrfs_lookup_block_group(root->fs_info, start);
5504 if (!cache) {
Chris Masond3977122009-01-05 21:25:51 -05005505 printk(KERN_ERR "Unable to find block group for %llu\n",
5506 (unsigned long long)start);
Josef Bacik0f9dd462008-09-23 13:14:11 -04005507 return -ENOSPC;
5508 }
Liu Hui1f3c79a2009-01-05 15:57:51 -05005509
Li Dongyang5378e602011-03-24 10:24:27 +00005510 if (btrfs_test_opt(root, DISCARD))
5511 ret = btrfs_discard_extent(root, start, len, NULL);
Liu Hui1f3c79a2009-01-05 15:57:51 -05005512
Josef Bacik0f9dd462008-09-23 13:14:11 -04005513 btrfs_add_free_space(cache, start, len);
Li Dongyangb4d00d52011-03-24 10:24:25 +00005514 btrfs_update_reserved_bytes(cache, len, 0, 1);
Chris Masonfa9c0d72009-04-03 09:47:43 -04005515 btrfs_put_block_group(cache);
Josef Bacik817d52f2009-07-13 21:29:25 -04005516
liubo1abe9b82011-03-24 11:18:59 +00005517 trace_btrfs_reserved_extent_free(root, start, len);
5518
Chris Masone6dcd2d2008-07-17 12:53:50 -04005519 return ret;
5520}
5521
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005522static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
5523 struct btrfs_root *root,
5524 u64 parent, u64 root_objectid,
5525 u64 flags, u64 owner, u64 offset,
5526 struct btrfs_key *ins, int ref_mod)
Chris Masone6dcd2d2008-07-17 12:53:50 -04005527{
5528 int ret;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005529 struct btrfs_fs_info *fs_info = root->fs_info;
Chris Masone6dcd2d2008-07-17 12:53:50 -04005530 struct btrfs_extent_item *extent_item;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005531 struct btrfs_extent_inline_ref *iref;
Chris Masone6dcd2d2008-07-17 12:53:50 -04005532 struct btrfs_path *path;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005533 struct extent_buffer *leaf;
5534 int type;
5535 u32 size;
Chris Masonf2654de2007-06-26 12:20:46 -04005536
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005537 if (parent > 0)
5538 type = BTRFS_SHARED_DATA_REF_KEY;
5539 else
5540 type = BTRFS_EXTENT_DATA_REF_KEY;
Zheng Yan31840ae2008-09-23 13:14:14 -04005541
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005542 size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
Chris Mason7bb86312007-12-11 09:25:06 -05005543
5544 path = btrfs_alloc_path();
Tsutomu Itohdb5b4932011-03-23 08:14:16 +00005545 if (!path)
5546 return -ENOMEM;
Chris Mason47e4bb92008-02-01 14:51:59 -05005547
Chris Masonb9473432009-03-13 11:00:37 -04005548 path->leave_spinning = 1;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005549 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
5550 ins, size);
Chris Masonccd467d2007-06-28 15:57:36 -04005551 BUG_ON(ret);
Josef Bacik0f9dd462008-09-23 13:14:11 -04005552
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005553 leaf = path->nodes[0];
5554 extent_item = btrfs_item_ptr(leaf, path->slots[0],
Chris Mason47e4bb92008-02-01 14:51:59 -05005555 struct btrfs_extent_item);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005556 btrfs_set_extent_refs(leaf, extent_item, ref_mod);
5557 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
5558 btrfs_set_extent_flags(leaf, extent_item,
5559 flags | BTRFS_EXTENT_FLAG_DATA);
Chris Mason47e4bb92008-02-01 14:51:59 -05005560
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005561 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
5562 btrfs_set_extent_inline_ref_type(leaf, iref, type);
5563 if (parent > 0) {
5564 struct btrfs_shared_data_ref *ref;
5565 ref = (struct btrfs_shared_data_ref *)(iref + 1);
5566 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
5567 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
5568 } else {
5569 struct btrfs_extent_data_ref *ref;
5570 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
5571 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
5572 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
5573 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
5574 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
5575 }
Chris Mason47e4bb92008-02-01 14:51:59 -05005576
5577 btrfs_mark_buffer_dirty(path->nodes[0]);
Chris Mason7bb86312007-12-11 09:25:06 -05005578 btrfs_free_path(path);
Chris Masonf510cfe2007-10-15 16:14:48 -04005579
Yan, Zhengf0486c62010-05-16 10:46:25 -04005580 ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
Chris Masonf5947062008-02-04 10:10:13 -05005581 if (ret) {
Chris Masond3977122009-01-05 21:25:51 -05005582 printk(KERN_ERR "btrfs update block group failed for %llu "
5583 "%llu\n", (unsigned long long)ins->objectid,
5584 (unsigned long long)ins->offset);
Chris Masonf5947062008-02-04 10:10:13 -05005585 BUG();
5586 }
Chris Masone6dcd2d2008-07-17 12:53:50 -04005587 return ret;
5588}
5589
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005590static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
5591 struct btrfs_root *root,
5592 u64 parent, u64 root_objectid,
5593 u64 flags, struct btrfs_disk_key *key,
5594 int level, struct btrfs_key *ins)
5595{
5596 int ret;
5597 struct btrfs_fs_info *fs_info = root->fs_info;
5598 struct btrfs_extent_item *extent_item;
5599 struct btrfs_tree_block_info *block_info;
5600 struct btrfs_extent_inline_ref *iref;
5601 struct btrfs_path *path;
5602 struct extent_buffer *leaf;
5603 u32 size = sizeof(*extent_item) + sizeof(*block_info) + sizeof(*iref);
5604
5605 path = btrfs_alloc_path();
5606 BUG_ON(!path);
5607
5608 path->leave_spinning = 1;
5609 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
5610 ins, size);
5611 BUG_ON(ret);
5612
5613 leaf = path->nodes[0];
5614 extent_item = btrfs_item_ptr(leaf, path->slots[0],
5615 struct btrfs_extent_item);
5616 btrfs_set_extent_refs(leaf, extent_item, 1);
5617 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
5618 btrfs_set_extent_flags(leaf, extent_item,
5619 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
5620 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
5621
5622 btrfs_set_tree_block_key(leaf, block_info, key);
5623 btrfs_set_tree_block_level(leaf, block_info, level);
5624
5625 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
5626 if (parent > 0) {
5627 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
5628 btrfs_set_extent_inline_ref_type(leaf, iref,
5629 BTRFS_SHARED_BLOCK_REF_KEY);
5630 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
5631 } else {
5632 btrfs_set_extent_inline_ref_type(leaf, iref,
5633 BTRFS_TREE_BLOCK_REF_KEY);
5634 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
5635 }
5636
5637 btrfs_mark_buffer_dirty(leaf);
5638 btrfs_free_path(path);
5639
Yan, Zhengf0486c62010-05-16 10:46:25 -04005640 ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005641 if (ret) {
5642 printk(KERN_ERR "btrfs update block group failed for %llu "
5643 "%llu\n", (unsigned long long)ins->objectid,
5644 (unsigned long long)ins->offset);
5645 BUG();
5646 }
5647 return ret;
5648}
5649
5650int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
5651 struct btrfs_root *root,
5652 u64 root_objectid, u64 owner,
5653 u64 offset, struct btrfs_key *ins)
Chris Masone6dcd2d2008-07-17 12:53:50 -04005654{
5655 int ret;
Chris Mason1c2308f2008-09-23 13:14:13 -04005656
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005657 BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
Chris Mason56bec292009-03-13 10:10:06 -04005658
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005659 ret = btrfs_add_delayed_data_ref(trans, ins->objectid, ins->offset,
5660 0, root_objectid, owner, offset,
5661 BTRFS_ADD_DELAYED_EXTENT, NULL);
Chris Masone6dcd2d2008-07-17 12:53:50 -04005662 return ret;
5663}
Chris Masone02119d2008-09-05 16:13:11 -04005664
5665/*
5666 * this is used by the tree logging recovery code. It records that
5667 * an extent has been allocated and makes sure to clear the free
5668 * space cache bits as well
5669 */
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005670int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
5671 struct btrfs_root *root,
5672 u64 root_objectid, u64 owner, u64 offset,
5673 struct btrfs_key *ins)
Chris Masone02119d2008-09-05 16:13:11 -04005674{
5675 int ret;
5676 struct btrfs_block_group_cache *block_group;
Yan Zheng11833d62009-09-11 16:11:19 -04005677 struct btrfs_caching_control *caching_ctl;
5678 u64 start = ins->objectid;
5679 u64 num_bytes = ins->offset;
Chris Masone02119d2008-09-05 16:13:11 -04005680
Chris Masone02119d2008-09-05 16:13:11 -04005681 block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
Josef Bacikb8399de2010-12-08 09:15:11 -05005682 cache_block_group(block_group, trans, NULL, 0);
Yan Zheng11833d62009-09-11 16:11:19 -04005683 caching_ctl = get_caching_control(block_group);
Chris Masone02119d2008-09-05 16:13:11 -04005684
Yan Zheng11833d62009-09-11 16:11:19 -04005685 if (!caching_ctl) {
5686 BUG_ON(!block_group_cache_done(block_group));
5687 ret = btrfs_remove_free_space(block_group, start, num_bytes);
5688 BUG_ON(ret);
5689 } else {
5690 mutex_lock(&caching_ctl->mutex);
5691
5692 if (start >= caching_ctl->progress) {
5693 ret = add_excluded_extent(root, start, num_bytes);
5694 BUG_ON(ret);
5695 } else if (start + num_bytes <= caching_ctl->progress) {
5696 ret = btrfs_remove_free_space(block_group,
5697 start, num_bytes);
5698 BUG_ON(ret);
5699 } else {
5700 num_bytes = caching_ctl->progress - start;
5701 ret = btrfs_remove_free_space(block_group,
5702 start, num_bytes);
5703 BUG_ON(ret);
5704
5705 start = caching_ctl->progress;
5706 num_bytes = ins->objectid + ins->offset -
5707 caching_ctl->progress;
5708 ret = add_excluded_extent(root, start, num_bytes);
5709 BUG_ON(ret);
5710 }
5711
5712 mutex_unlock(&caching_ctl->mutex);
5713 put_caching_control(caching_ctl);
5714 }
5715
Li Dongyangb4d00d52011-03-24 10:24:25 +00005716 ret = btrfs_update_reserved_bytes(block_group, ins->offset, 1, 1);
Yan, Zhengf0486c62010-05-16 10:46:25 -04005717 BUG_ON(ret);
Chris Masonfa9c0d72009-04-03 09:47:43 -04005718 btrfs_put_block_group(block_group);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005719 ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
5720 0, owner, offset, ins, 1);
Chris Masone02119d2008-09-05 16:13:11 -04005721 return ret;
5722}
5723
Chris Mason65b51a02008-08-01 15:11:20 -04005724struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans,
5725 struct btrfs_root *root,
Chris Mason4008c042009-02-12 14:09:45 -05005726 u64 bytenr, u32 blocksize,
5727 int level)
Chris Mason65b51a02008-08-01 15:11:20 -04005728{
5729 struct extent_buffer *buf;
5730
5731 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
5732 if (!buf)
5733 return ERR_PTR(-ENOMEM);
5734 btrfs_set_header_generation(buf, trans->transid);
Chris Mason4008c042009-02-12 14:09:45 -05005735 btrfs_set_buffer_lockdep_class(buf, level);
Chris Mason65b51a02008-08-01 15:11:20 -04005736 btrfs_tree_lock(buf);
5737 clean_tree_block(trans, root, buf);
Chris Masonb4ce94d2009-02-04 09:25:08 -05005738
5739 btrfs_set_lock_blocking(buf);
Chris Mason65b51a02008-08-01 15:11:20 -04005740 btrfs_set_buffer_uptodate(buf);
Chris Masonb4ce94d2009-02-04 09:25:08 -05005741
Chris Masond0c803c2008-09-11 16:17:57 -04005742 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
Yan, Zheng8cef4e12009-11-12 09:33:26 +00005743 /*
5744 * we allow two log transactions at a time, use different
5745 * EXENT bit to differentiate dirty pages.
5746 */
5747 if (root->log_transid % 2 == 0)
5748 set_extent_dirty(&root->dirty_log_pages, buf->start,
5749 buf->start + buf->len - 1, GFP_NOFS);
5750 else
5751 set_extent_new(&root->dirty_log_pages, buf->start,
5752 buf->start + buf->len - 1, GFP_NOFS);
Chris Masond0c803c2008-09-11 16:17:57 -04005753 } else {
5754 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
5755 buf->start + buf->len - 1, GFP_NOFS);
5756 }
Chris Mason65b51a02008-08-01 15:11:20 -04005757 trans->blocks_used++;
Chris Masonb4ce94d2009-02-04 09:25:08 -05005758 /* this returns a buffer locked for blocking */
Chris Mason65b51a02008-08-01 15:11:20 -04005759 return buf;
5760}
5761
Yan, Zhengf0486c62010-05-16 10:46:25 -04005762static struct btrfs_block_rsv *
5763use_block_rsv(struct btrfs_trans_handle *trans,
5764 struct btrfs_root *root, u32 blocksize)
5765{
5766 struct btrfs_block_rsv *block_rsv;
Josef Bacik68a82272011-01-24 21:43:20 +00005767 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
Yan, Zhengf0486c62010-05-16 10:46:25 -04005768 int ret;
5769
5770 block_rsv = get_block_rsv(trans, root);
5771
5772 if (block_rsv->size == 0) {
Josef Bacik8bb8ab22010-10-15 16:52:49 -04005773 ret = reserve_metadata_bytes(trans, root, block_rsv,
5774 blocksize, 0);
Josef Bacik68a82272011-01-24 21:43:20 +00005775 /*
5776 * If we couldn't reserve metadata bytes try and use some from
5777 * the global reserve.
5778 */
5779 if (ret && block_rsv != global_rsv) {
5780 ret = block_rsv_use_bytes(global_rsv, blocksize);
5781 if (!ret)
5782 return global_rsv;
Yan, Zhengf0486c62010-05-16 10:46:25 -04005783 return ERR_PTR(ret);
Josef Bacik68a82272011-01-24 21:43:20 +00005784 } else if (ret) {
5785 return ERR_PTR(ret);
5786 }
Yan, Zhengf0486c62010-05-16 10:46:25 -04005787 return block_rsv;
5788 }
5789
5790 ret = block_rsv_use_bytes(block_rsv, blocksize);
5791 if (!ret)
5792 return block_rsv;
Josef Bacik68a82272011-01-24 21:43:20 +00005793 if (ret) {
5794 WARN_ON(1);
5795 ret = reserve_metadata_bytes(trans, root, block_rsv, blocksize,
5796 0);
5797 if (!ret) {
5798 spin_lock(&block_rsv->lock);
5799 block_rsv->size += blocksize;
5800 spin_unlock(&block_rsv->lock);
5801 return block_rsv;
5802 } else if (ret && block_rsv != global_rsv) {
5803 ret = block_rsv_use_bytes(global_rsv, blocksize);
5804 if (!ret)
5805 return global_rsv;
5806 }
5807 }
Yan, Zhengf0486c62010-05-16 10:46:25 -04005808
Yan, Zhengf0486c62010-05-16 10:46:25 -04005809 return ERR_PTR(-ENOSPC);
5810}
5811
5812static void unuse_block_rsv(struct btrfs_block_rsv *block_rsv, u32 blocksize)
5813{
5814 block_rsv_add_bytes(block_rsv, blocksize, 0);
5815 block_rsv_release_bytes(block_rsv, NULL, 0);
5816}
5817
Chris Masonfec577f2007-02-26 10:40:21 -05005818/*
Yan, Zhengf0486c62010-05-16 10:46:25 -04005819 * finds a free extent and does all the dirty work required for allocation
5820 * returns the key for the extent through ins, and a tree buffer for
5821 * the first block of the extent through buf.
5822 *
Chris Masonfec577f2007-02-26 10:40:21 -05005823 * returns the tree buffer or NULL.
5824 */
Chris Mason5f39d392007-10-15 16:14:19 -04005825struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005826 struct btrfs_root *root, u32 blocksize,
5827 u64 parent, u64 root_objectid,
5828 struct btrfs_disk_key *key, int level,
5829 u64 hint, u64 empty_size)
Chris Masonfec577f2007-02-26 10:40:21 -05005830{
Chris Masone2fa7222007-03-12 16:22:34 -04005831 struct btrfs_key ins;
Yan, Zhengf0486c62010-05-16 10:46:25 -04005832 struct btrfs_block_rsv *block_rsv;
Chris Mason5f39d392007-10-15 16:14:19 -04005833 struct extent_buffer *buf;
Yan, Zhengf0486c62010-05-16 10:46:25 -04005834 u64 flags = 0;
5835 int ret;
Chris Masonfec577f2007-02-26 10:40:21 -05005836
Yan, Zhengf0486c62010-05-16 10:46:25 -04005837
5838 block_rsv = use_block_rsv(trans, root, blocksize);
5839 if (IS_ERR(block_rsv))
5840 return ERR_CAST(block_rsv);
5841
5842 ret = btrfs_reserve_extent(trans, root, blocksize, blocksize,
5843 empty_size, hint, (u64)-1, &ins, 0);
Chris Masonfec577f2007-02-26 10:40:21 -05005844 if (ret) {
Yan, Zhengf0486c62010-05-16 10:46:25 -04005845 unuse_block_rsv(block_rsv, blocksize);
Chris Mason54aa1f42007-06-22 14:16:25 -04005846 return ERR_PTR(ret);
Chris Masonfec577f2007-02-26 10:40:21 -05005847 }
Chris Mason55c69072008-01-09 15:55:33 -05005848
Chris Mason4008c042009-02-12 14:09:45 -05005849 buf = btrfs_init_new_buffer(trans, root, ins.objectid,
5850 blocksize, level);
Yan, Zhengf0486c62010-05-16 10:46:25 -04005851 BUG_ON(IS_ERR(buf));
5852
5853 if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
5854 if (parent == 0)
5855 parent = ins.objectid;
5856 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
5857 } else
5858 BUG_ON(parent > 0);
5859
5860 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
5861 struct btrfs_delayed_extent_op *extent_op;
5862 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
5863 BUG_ON(!extent_op);
5864 if (key)
5865 memcpy(&extent_op->key, key, sizeof(extent_op->key));
5866 else
5867 memset(&extent_op->key, 0, sizeof(extent_op->key));
5868 extent_op->flags_to_set = flags;
5869 extent_op->update_key = 1;
5870 extent_op->update_flags = 1;
5871 extent_op->is_data = 0;
5872
5873 ret = btrfs_add_delayed_tree_ref(trans, ins.objectid,
5874 ins.offset, parent, root_objectid,
5875 level, BTRFS_ADD_DELAYED_EXTENT,
5876 extent_op);
5877 BUG_ON(ret);
5878 }
Chris Masonfec577f2007-02-26 10:40:21 -05005879 return buf;
5880}
Chris Masona28ec192007-03-06 20:08:01 -05005881
Yan Zheng2c47e6052009-06-27 21:07:35 -04005882struct walk_control {
5883 u64 refs[BTRFS_MAX_LEVEL];
5884 u64 flags[BTRFS_MAX_LEVEL];
5885 struct btrfs_key update_progress;
5886 int stage;
5887 int level;
5888 int shared_level;
5889 int update_ref;
5890 int keep_locks;
Yan, Zheng1c4850e2009-09-21 15:55:59 -04005891 int reada_slot;
5892 int reada_count;
Yan Zheng2c47e6052009-06-27 21:07:35 -04005893};
5894
5895#define DROP_REFERENCE 1
5896#define UPDATE_BACKREF 2
5897
Yan, Zheng1c4850e2009-09-21 15:55:59 -04005898static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
5899 struct btrfs_root *root,
5900 struct walk_control *wc,
5901 struct btrfs_path *path)
5902{
5903 u64 bytenr;
5904 u64 generation;
5905 u64 refs;
Yan, Zheng94fcca92009-10-09 09:25:16 -04005906 u64 flags;
Yan, Zheng1c4850e2009-09-21 15:55:59 -04005907 u32 nritems;
5908 u32 blocksize;
5909 struct btrfs_key key;
5910 struct extent_buffer *eb;
5911 int ret;
5912 int slot;
5913 int nread = 0;
5914
5915 if (path->slots[wc->level] < wc->reada_slot) {
5916 wc->reada_count = wc->reada_count * 2 / 3;
5917 wc->reada_count = max(wc->reada_count, 2);
5918 } else {
5919 wc->reada_count = wc->reada_count * 3 / 2;
5920 wc->reada_count = min_t(int, wc->reada_count,
5921 BTRFS_NODEPTRS_PER_BLOCK(root));
5922 }
5923
5924 eb = path->nodes[wc->level];
5925 nritems = btrfs_header_nritems(eb);
5926 blocksize = btrfs_level_size(root, wc->level - 1);
5927
5928 for (slot = path->slots[wc->level]; slot < nritems; slot++) {
5929 if (nread >= wc->reada_count)
5930 break;
5931
5932 cond_resched();
5933 bytenr = btrfs_node_blockptr(eb, slot);
5934 generation = btrfs_node_ptr_generation(eb, slot);
5935
5936 if (slot == path->slots[wc->level])
5937 goto reada;
5938
5939 if (wc->stage == UPDATE_BACKREF &&
5940 generation <= root->root_key.offset)
5941 continue;
5942
Yan, Zheng94fcca92009-10-09 09:25:16 -04005943 /* We don't lock the tree block, it's OK to be racy here */
5944 ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
5945 &refs, &flags);
5946 BUG_ON(ret);
5947 BUG_ON(refs == 0);
5948
Yan, Zheng1c4850e2009-09-21 15:55:59 -04005949 if (wc->stage == DROP_REFERENCE) {
Yan, Zheng1c4850e2009-09-21 15:55:59 -04005950 if (refs == 1)
5951 goto reada;
5952
Yan, Zheng94fcca92009-10-09 09:25:16 -04005953 if (wc->level == 1 &&
5954 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5955 continue;
Yan, Zheng1c4850e2009-09-21 15:55:59 -04005956 if (!wc->update_ref ||
5957 generation <= root->root_key.offset)
5958 continue;
5959 btrfs_node_key_to_cpu(eb, &key, slot);
5960 ret = btrfs_comp_cpu_keys(&key,
5961 &wc->update_progress);
5962 if (ret < 0)
5963 continue;
Yan, Zheng94fcca92009-10-09 09:25:16 -04005964 } else {
5965 if (wc->level == 1 &&
5966 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5967 continue;
Yan, Zheng1c4850e2009-09-21 15:55:59 -04005968 }
5969reada:
5970 ret = readahead_tree_block(root, bytenr, blocksize,
5971 generation);
5972 if (ret)
5973 break;
Yan, Zheng1c4850e2009-09-21 15:55:59 -04005974 nread++;
5975 }
5976 wc->reada_slot = slot;
5977}
5978
Chris Mason9aca1d52007-03-13 11:09:37 -04005979/*
Yan Zheng2c47e6052009-06-27 21:07:35 -04005980 * hepler to process tree block while walking down the tree.
5981 *
Yan Zheng2c47e6052009-06-27 21:07:35 -04005982 * when wc->stage == UPDATE_BACKREF, this function updates
5983 * back refs for pointers in the block.
5984 *
5985 * NOTE: return value 1 means we should stop walking down.
Yan Zhengf82d02d2008-10-29 14:49:05 -04005986 */
Yan Zheng2c47e6052009-06-27 21:07:35 -04005987static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
5988 struct btrfs_root *root,
5989 struct btrfs_path *path,
Yan, Zheng94fcca92009-10-09 09:25:16 -04005990 struct walk_control *wc, int lookup_info)
Yan Zheng2c47e6052009-06-27 21:07:35 -04005991{
5992 int level = wc->level;
5993 struct extent_buffer *eb = path->nodes[level];
Yan Zheng2c47e6052009-06-27 21:07:35 -04005994 u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
5995 int ret;
5996
5997 if (wc->stage == UPDATE_BACKREF &&
5998 btrfs_header_owner(eb) != root->root_key.objectid)
5999 return 1;
6000
6001 /*
6002 * when reference count of tree block is 1, it won't increase
6003 * again. once full backref flag is set, we never clear it.
6004 */
Yan, Zheng94fcca92009-10-09 09:25:16 -04006005 if (lookup_info &&
6006 ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
6007 (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
Yan Zheng2c47e6052009-06-27 21:07:35 -04006008 BUG_ON(!path->locks[level]);
6009 ret = btrfs_lookup_extent_info(trans, root,
6010 eb->start, eb->len,
6011 &wc->refs[level],
6012 &wc->flags[level]);
6013 BUG_ON(ret);
6014 BUG_ON(wc->refs[level] == 0);
6015 }
6016
Yan Zheng2c47e6052009-06-27 21:07:35 -04006017 if (wc->stage == DROP_REFERENCE) {
6018 if (wc->refs[level] > 1)
6019 return 1;
6020
6021 if (path->locks[level] && !wc->keep_locks) {
6022 btrfs_tree_unlock(eb);
6023 path->locks[level] = 0;
6024 }
6025 return 0;
6026 }
6027
6028 /* wc->stage == UPDATE_BACKREF */
6029 if (!(wc->flags[level] & flag)) {
6030 BUG_ON(!path->locks[level]);
6031 ret = btrfs_inc_ref(trans, root, eb, 1);
6032 BUG_ON(ret);
6033 ret = btrfs_dec_ref(trans, root, eb, 0);
6034 BUG_ON(ret);
6035 ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
6036 eb->len, flag, 0);
6037 BUG_ON(ret);
6038 wc->flags[level] |= flag;
6039 }
6040
6041 /*
6042 * the block is shared by multiple trees, so it's not good to
6043 * keep the tree lock
6044 */
6045 if (path->locks[level] && level > 0) {
6046 btrfs_tree_unlock(eb);
6047 path->locks[level] = 0;
6048 }
6049 return 0;
6050}
6051
6052/*
Yan, Zheng1c4850e2009-09-21 15:55:59 -04006053 * hepler to process tree block pointer.
6054 *
6055 * when wc->stage == DROP_REFERENCE, this function checks
6056 * reference count of the block pointed to. if the block
6057 * is shared and we need update back refs for the subtree
6058 * rooted at the block, this function changes wc->stage to
6059 * UPDATE_BACKREF. if the block is shared and there is no
6060 * need to update back, this function drops the reference
6061 * to the block.
6062 *
6063 * NOTE: return value 1 means we should stop walking down.
6064 */
6065static noinline int do_walk_down(struct btrfs_trans_handle *trans,
6066 struct btrfs_root *root,
6067 struct btrfs_path *path,
Yan, Zheng94fcca92009-10-09 09:25:16 -04006068 struct walk_control *wc, int *lookup_info)
Yan, Zheng1c4850e2009-09-21 15:55:59 -04006069{
6070 u64 bytenr;
6071 u64 generation;
6072 u64 parent;
6073 u32 blocksize;
6074 struct btrfs_key key;
6075 struct extent_buffer *next;
6076 int level = wc->level;
6077 int reada = 0;
6078 int ret = 0;
6079
6080 generation = btrfs_node_ptr_generation(path->nodes[level],
6081 path->slots[level]);
6082 /*
6083 * if the lower level block was created before the snapshot
6084 * was created, we know there is no need to update back refs
6085 * for the subtree
6086 */
6087 if (wc->stage == UPDATE_BACKREF &&
Yan, Zheng94fcca92009-10-09 09:25:16 -04006088 generation <= root->root_key.offset) {
6089 *lookup_info = 1;
Yan, Zheng1c4850e2009-09-21 15:55:59 -04006090 return 1;
Yan, Zheng94fcca92009-10-09 09:25:16 -04006091 }
Yan, Zheng1c4850e2009-09-21 15:55:59 -04006092
6093 bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
6094 blocksize = btrfs_level_size(root, level - 1);
6095
6096 next = btrfs_find_tree_block(root, bytenr, blocksize);
6097 if (!next) {
6098 next = btrfs_find_create_tree_block(root, bytenr, blocksize);
Miao Xie90d2c512010-03-25 12:37:12 +00006099 if (!next)
6100 return -ENOMEM;
Yan, Zheng1c4850e2009-09-21 15:55:59 -04006101 reada = 1;
6102 }
6103 btrfs_tree_lock(next);
6104 btrfs_set_lock_blocking(next);
6105
Yan, Zheng94fcca92009-10-09 09:25:16 -04006106 ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
6107 &wc->refs[level - 1],
6108 &wc->flags[level - 1]);
6109 BUG_ON(ret);
6110 BUG_ON(wc->refs[level - 1] == 0);
6111 *lookup_info = 0;
Yan, Zheng1c4850e2009-09-21 15:55:59 -04006112
Yan, Zheng94fcca92009-10-09 09:25:16 -04006113 if (wc->stage == DROP_REFERENCE) {
Yan, Zheng1c4850e2009-09-21 15:55:59 -04006114 if (wc->refs[level - 1] > 1) {
Yan, Zheng94fcca92009-10-09 09:25:16 -04006115 if (level == 1 &&
6116 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6117 goto skip;
6118
Yan, Zheng1c4850e2009-09-21 15:55:59 -04006119 if (!wc->update_ref ||
6120 generation <= root->root_key.offset)
6121 goto skip;
6122
6123 btrfs_node_key_to_cpu(path->nodes[level], &key,
6124 path->slots[level]);
6125 ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
6126 if (ret < 0)
6127 goto skip;
6128
6129 wc->stage = UPDATE_BACKREF;
6130 wc->shared_level = level - 1;
6131 }
Yan, Zheng94fcca92009-10-09 09:25:16 -04006132 } else {
6133 if (level == 1 &&
6134 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6135 goto skip;
Yan, Zheng1c4850e2009-09-21 15:55:59 -04006136 }
6137
6138 if (!btrfs_buffer_uptodate(next, generation)) {
6139 btrfs_tree_unlock(next);
6140 free_extent_buffer(next);
6141 next = NULL;
Yan, Zheng94fcca92009-10-09 09:25:16 -04006142 *lookup_info = 1;
Yan, Zheng1c4850e2009-09-21 15:55:59 -04006143 }
6144
6145 if (!next) {
6146 if (reada && level == 1)
6147 reada_walk_down(trans, root, wc, path);
6148 next = read_tree_block(root, bytenr, blocksize, generation);
Tsutomu Itoh97d9a8a2011-03-24 06:33:21 +00006149 if (!next)
6150 return -EIO;
Yan, Zheng1c4850e2009-09-21 15:55:59 -04006151 btrfs_tree_lock(next);
6152 btrfs_set_lock_blocking(next);
6153 }
6154
6155 level--;
6156 BUG_ON(level != btrfs_header_level(next));
6157 path->nodes[level] = next;
6158 path->slots[level] = 0;
6159 path->locks[level] = 1;
6160 wc->level = level;
6161 if (wc->level == 1)
6162 wc->reada_slot = 0;
6163 return 0;
6164skip:
6165 wc->refs[level - 1] = 0;
6166 wc->flags[level - 1] = 0;
Yan, Zheng94fcca92009-10-09 09:25:16 -04006167 if (wc->stage == DROP_REFERENCE) {
6168 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
6169 parent = path->nodes[level]->start;
6170 } else {
6171 BUG_ON(root->root_key.objectid !=
6172 btrfs_header_owner(path->nodes[level]));
6173 parent = 0;
6174 }
Yan, Zheng1c4850e2009-09-21 15:55:59 -04006175
Yan, Zheng94fcca92009-10-09 09:25:16 -04006176 ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
6177 root->root_key.objectid, level - 1, 0);
6178 BUG_ON(ret);
Yan, Zheng1c4850e2009-09-21 15:55:59 -04006179 }
Yan, Zheng1c4850e2009-09-21 15:55:59 -04006180 btrfs_tree_unlock(next);
6181 free_extent_buffer(next);
Yan, Zheng94fcca92009-10-09 09:25:16 -04006182 *lookup_info = 1;
Yan, Zheng1c4850e2009-09-21 15:55:59 -04006183 return 1;
6184}
6185
6186/*
Yan Zheng2c47e6052009-06-27 21:07:35 -04006187 * hepler to process tree block while walking up the tree.
6188 *
6189 * when wc->stage == DROP_REFERENCE, this function drops
6190 * reference count on the block.
6191 *
6192 * when wc->stage == UPDATE_BACKREF, this function changes
6193 * wc->stage back to DROP_REFERENCE if we changed wc->stage
6194 * to UPDATE_BACKREF previously while processing the block.
6195 *
6196 * NOTE: return value 1 means we should stop walking up.
6197 */
6198static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
6199 struct btrfs_root *root,
6200 struct btrfs_path *path,
6201 struct walk_control *wc)
6202{
Yan, Zhengf0486c62010-05-16 10:46:25 -04006203 int ret;
Yan Zheng2c47e6052009-06-27 21:07:35 -04006204 int level = wc->level;
6205 struct extent_buffer *eb = path->nodes[level];
6206 u64 parent = 0;
6207
6208 if (wc->stage == UPDATE_BACKREF) {
6209 BUG_ON(wc->shared_level < level);
6210 if (level < wc->shared_level)
6211 goto out;
6212
Yan Zheng2c47e6052009-06-27 21:07:35 -04006213 ret = find_next_key(path, level + 1, &wc->update_progress);
6214 if (ret > 0)
6215 wc->update_ref = 0;
6216
6217 wc->stage = DROP_REFERENCE;
6218 wc->shared_level = -1;
6219 path->slots[level] = 0;
6220
6221 /*
6222 * check reference count again if the block isn't locked.
6223 * we should start walking down the tree again if reference
6224 * count is one.
6225 */
6226 if (!path->locks[level]) {
6227 BUG_ON(level == 0);
6228 btrfs_tree_lock(eb);
6229 btrfs_set_lock_blocking(eb);
6230 path->locks[level] = 1;
6231
6232 ret = btrfs_lookup_extent_info(trans, root,
6233 eb->start, eb->len,
6234 &wc->refs[level],
6235 &wc->flags[level]);
6236 BUG_ON(ret);
6237 BUG_ON(wc->refs[level] == 0);
6238 if (wc->refs[level] == 1) {
6239 btrfs_tree_unlock(eb);
6240 path->locks[level] = 0;
6241 return 1;
6242 }
Yan Zheng2c47e6052009-06-27 21:07:35 -04006243 }
6244 }
6245
6246 /* wc->stage == DROP_REFERENCE */
6247 BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
6248
6249 if (wc->refs[level] == 1) {
6250 if (level == 0) {
6251 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6252 ret = btrfs_dec_ref(trans, root, eb, 1);
6253 else
6254 ret = btrfs_dec_ref(trans, root, eb, 0);
6255 BUG_ON(ret);
6256 }
6257 /* make block locked assertion in clean_tree_block happy */
6258 if (!path->locks[level] &&
6259 btrfs_header_generation(eb) == trans->transid) {
6260 btrfs_tree_lock(eb);
6261 btrfs_set_lock_blocking(eb);
6262 path->locks[level] = 1;
6263 }
6264 clean_tree_block(trans, root, eb);
6265 }
6266
6267 if (eb == root->node) {
6268 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6269 parent = eb->start;
6270 else
6271 BUG_ON(root->root_key.objectid !=
6272 btrfs_header_owner(eb));
6273 } else {
6274 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6275 parent = path->nodes[level + 1]->start;
6276 else
6277 BUG_ON(root->root_key.objectid !=
6278 btrfs_header_owner(path->nodes[level + 1]));
6279 }
6280
Yan, Zhengf0486c62010-05-16 10:46:25 -04006281 btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
Yan Zheng2c47e6052009-06-27 21:07:35 -04006282out:
6283 wc->refs[level] = 0;
6284 wc->flags[level] = 0;
Yan, Zhengf0486c62010-05-16 10:46:25 -04006285 return 0;
Yan Zheng2c47e6052009-06-27 21:07:35 -04006286}
6287
Yan Zheng5d4f98a2009-06-10 10:45:14 -04006288static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
6289 struct btrfs_root *root,
Yan Zheng2c47e6052009-06-27 21:07:35 -04006290 struct btrfs_path *path,
6291 struct walk_control *wc)
Yan Zhengf82d02d2008-10-29 14:49:05 -04006292{
Yan Zheng2c47e6052009-06-27 21:07:35 -04006293 int level = wc->level;
Yan, Zheng94fcca92009-10-09 09:25:16 -04006294 int lookup_info = 1;
Yan Zhengf82d02d2008-10-29 14:49:05 -04006295 int ret;
6296
Yan Zheng2c47e6052009-06-27 21:07:35 -04006297 while (level >= 0) {
Yan, Zheng94fcca92009-10-09 09:25:16 -04006298 ret = walk_down_proc(trans, root, path, wc, lookup_info);
Yan Zheng2c47e6052009-06-27 21:07:35 -04006299 if (ret > 0)
Yan Zhengf82d02d2008-10-29 14:49:05 -04006300 break;
Yan Zhengf82d02d2008-10-29 14:49:05 -04006301
Yan Zheng2c47e6052009-06-27 21:07:35 -04006302 if (level == 0)
6303 break;
6304
Yan, Zheng7a7965f2010-02-01 02:41:17 +00006305 if (path->slots[level] >=
6306 btrfs_header_nritems(path->nodes[level]))
6307 break;
6308
Yan, Zheng94fcca92009-10-09 09:25:16 -04006309 ret = do_walk_down(trans, root, path, wc, &lookup_info);
Yan, Zheng1c4850e2009-09-21 15:55:59 -04006310 if (ret > 0) {
6311 path->slots[level]++;
6312 continue;
Miao Xie90d2c512010-03-25 12:37:12 +00006313 } else if (ret < 0)
6314 return ret;
Yan, Zheng1c4850e2009-09-21 15:55:59 -04006315 level = wc->level;
Yan Zhengf82d02d2008-10-29 14:49:05 -04006316 }
Yan Zhengf82d02d2008-10-29 14:49:05 -04006317 return 0;
6318}
6319
Chris Masond3977122009-01-05 21:25:51 -05006320static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
Chris Mason98ed5172008-01-03 10:01:48 -05006321 struct btrfs_root *root,
Yan Zhengf82d02d2008-10-29 14:49:05 -04006322 struct btrfs_path *path,
Yan Zheng2c47e6052009-06-27 21:07:35 -04006323 struct walk_control *wc, int max_level)
Chris Mason20524f02007-03-10 06:35:47 -05006324{
Yan Zheng2c47e6052009-06-27 21:07:35 -04006325 int level = wc->level;
Chris Mason20524f02007-03-10 06:35:47 -05006326 int ret;
Chris Mason9f3a7422007-08-07 15:52:19 -04006327
Yan Zheng2c47e6052009-06-27 21:07:35 -04006328 path->slots[level] = btrfs_header_nritems(path->nodes[level]);
6329 while (level < max_level && path->nodes[level]) {
6330 wc->level = level;
6331 if (path->slots[level] + 1 <
6332 btrfs_header_nritems(path->nodes[level])) {
6333 path->slots[level]++;
Chris Mason20524f02007-03-10 06:35:47 -05006334 return 0;
6335 } else {
Yan Zheng2c47e6052009-06-27 21:07:35 -04006336 ret = walk_up_proc(trans, root, path, wc);
6337 if (ret > 0)
6338 return 0;
Chris Masonbd56b302009-02-04 09:27:02 -05006339
Yan Zheng2c47e6052009-06-27 21:07:35 -04006340 if (path->locks[level]) {
6341 btrfs_tree_unlock(path->nodes[level]);
6342 path->locks[level] = 0;
Yan Zhengf82d02d2008-10-29 14:49:05 -04006343 }
Yan Zheng2c47e6052009-06-27 21:07:35 -04006344 free_extent_buffer(path->nodes[level]);
6345 path->nodes[level] = NULL;
6346 level++;
Chris Mason20524f02007-03-10 06:35:47 -05006347 }
6348 }
6349 return 1;
6350}
6351
Chris Mason9aca1d52007-03-13 11:09:37 -04006352/*
Yan Zheng2c47e6052009-06-27 21:07:35 -04006353 * drop a subvolume tree.
6354 *
6355 * this function traverses the tree freeing any blocks that only
6356 * referenced by the tree.
6357 *
6358 * when a shared tree block is found. this function decreases its
6359 * reference count by one. if update_ref is true, this function
6360 * also make sure backrefs for the shared block and all lower level
6361 * blocks are properly updated.
Chris Mason9aca1d52007-03-13 11:09:37 -04006362 */
Yan, Zheng3fd0a552010-05-16 10:49:59 -04006363int btrfs_drop_snapshot(struct btrfs_root *root,
6364 struct btrfs_block_rsv *block_rsv, int update_ref)
Chris Mason20524f02007-03-10 06:35:47 -05006365{
Chris Mason5caf2a02007-04-02 11:20:42 -04006366 struct btrfs_path *path;
Yan Zheng2c47e6052009-06-27 21:07:35 -04006367 struct btrfs_trans_handle *trans;
6368 struct btrfs_root *tree_root = root->fs_info->tree_root;
Chris Mason9f3a7422007-08-07 15:52:19 -04006369 struct btrfs_root_item *root_item = &root->root_item;
Yan Zheng2c47e6052009-06-27 21:07:35 -04006370 struct walk_control *wc;
6371 struct btrfs_key key;
6372 int err = 0;
6373 int ret;
6374 int level;
Chris Mason20524f02007-03-10 06:35:47 -05006375
Chris Mason5caf2a02007-04-02 11:20:42 -04006376 path = btrfs_alloc_path();
6377 BUG_ON(!path);
Chris Mason20524f02007-03-10 06:35:47 -05006378
Yan Zheng2c47e6052009-06-27 21:07:35 -04006379 wc = kzalloc(sizeof(*wc), GFP_NOFS);
6380 BUG_ON(!wc);
6381
Yan, Zhenga22285a2010-05-16 10:48:46 -04006382 trans = btrfs_start_transaction(tree_root, 0);
Tsutomu Itoh98d5dc12011-01-20 06:19:37 +00006383 BUG_ON(IS_ERR(trans));
6384
Yan, Zheng3fd0a552010-05-16 10:49:59 -04006385 if (block_rsv)
6386 trans->block_rsv = block_rsv;
Yan Zheng2c47e6052009-06-27 21:07:35 -04006387
Chris Mason9f3a7422007-08-07 15:52:19 -04006388 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
Yan Zheng2c47e6052009-06-27 21:07:35 -04006389 level = btrfs_header_level(root->node);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04006390 path->nodes[level] = btrfs_lock_root_node(root);
6391 btrfs_set_lock_blocking(path->nodes[level]);
Chris Mason9f3a7422007-08-07 15:52:19 -04006392 path->slots[level] = 0;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04006393 path->locks[level] = 1;
Yan Zheng2c47e6052009-06-27 21:07:35 -04006394 memset(&wc->update_progress, 0,
6395 sizeof(wc->update_progress));
Chris Mason9f3a7422007-08-07 15:52:19 -04006396 } else {
Chris Mason9f3a7422007-08-07 15:52:19 -04006397 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
Yan Zheng2c47e6052009-06-27 21:07:35 -04006398 memcpy(&wc->update_progress, &key,
6399 sizeof(wc->update_progress));
6400
Chris Mason6702ed42007-08-07 16:15:09 -04006401 level = root_item->drop_level;
Yan Zheng2c47e6052009-06-27 21:07:35 -04006402 BUG_ON(level == 0);
Chris Mason6702ed42007-08-07 16:15:09 -04006403 path->lowest_level = level;
Yan Zheng2c47e6052009-06-27 21:07:35 -04006404 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6405 path->lowest_level = 0;
6406 if (ret < 0) {
6407 err = ret;
Chris Mason9f3a7422007-08-07 15:52:19 -04006408 goto out;
6409 }
Yan, Zheng1c4850e2009-09-21 15:55:59 -04006410 WARN_ON(ret > 0);
Yan Zheng2c47e6052009-06-27 21:07:35 -04006411
Chris Mason7d9eb122008-07-08 14:19:17 -04006412 /*
6413 * unlock our path, this is safe because only this
6414 * function is allowed to delete this snapshot
6415 */
Yan Zheng5d4f98a2009-06-10 10:45:14 -04006416 btrfs_unlock_up_safe(path, 0);
Chris Mason9aca1d52007-03-13 11:09:37 -04006417
Yan Zheng2c47e6052009-06-27 21:07:35 -04006418 level = btrfs_header_level(root->node);
6419 while (1) {
6420 btrfs_tree_lock(path->nodes[level]);
6421 btrfs_set_lock_blocking(path->nodes[level]);
6422
6423 ret = btrfs_lookup_extent_info(trans, root,
6424 path->nodes[level]->start,
6425 path->nodes[level]->len,
6426 &wc->refs[level],
6427 &wc->flags[level]);
6428 BUG_ON(ret);
6429 BUG_ON(wc->refs[level] == 0);
6430
6431 if (level == root_item->drop_level)
6432 break;
6433
6434 btrfs_tree_unlock(path->nodes[level]);
6435 WARN_ON(wc->refs[level] != 1);
6436 level--;
6437 }
6438 }
6439
6440 wc->level = level;
6441 wc->shared_level = -1;
6442 wc->stage = DROP_REFERENCE;
6443 wc->update_ref = update_ref;
6444 wc->keep_locks = 0;
Yan, Zheng1c4850e2009-09-21 15:55:59 -04006445 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
Yan Zheng2c47e6052009-06-27 21:07:35 -04006446
6447 while (1) {
6448 ret = walk_down_tree(trans, root, path, wc);
6449 if (ret < 0) {
6450 err = ret;
Chris Masone7a84562008-06-25 16:01:31 -04006451 break;
6452 }
Yan Zheng2c47e6052009-06-27 21:07:35 -04006453
6454 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
6455 if (ret < 0) {
6456 err = ret;
6457 break;
6458 }
6459
6460 if (ret > 0) {
6461 BUG_ON(wc->stage != DROP_REFERENCE);
6462 break;
6463 }
6464
6465 if (wc->stage == DROP_REFERENCE) {
6466 level = wc->level;
6467 btrfs_node_key(path->nodes[level],
6468 &root_item->drop_progress,
6469 path->slots[level]);
6470 root_item->drop_level = level;
6471 }
6472
6473 BUG_ON(wc->level == 0);
Yan, Zheng3fd0a552010-05-16 10:49:59 -04006474 if (btrfs_should_end_transaction(trans, tree_root)) {
Yan Zheng2c47e6052009-06-27 21:07:35 -04006475 ret = btrfs_update_root(trans, tree_root,
6476 &root->root_key,
6477 root_item);
6478 BUG_ON(ret);
6479
Yan, Zheng3fd0a552010-05-16 10:49:59 -04006480 btrfs_end_transaction_throttle(trans, tree_root);
Yan, Zhenga22285a2010-05-16 10:48:46 -04006481 trans = btrfs_start_transaction(tree_root, 0);
Tsutomu Itoh98d5dc12011-01-20 06:19:37 +00006482 BUG_ON(IS_ERR(trans));
Yan, Zheng3fd0a552010-05-16 10:49:59 -04006483 if (block_rsv)
6484 trans->block_rsv = block_rsv;
Chris Masonc3e69d52009-03-13 10:17:05 -04006485 }
Chris Mason20524f02007-03-10 06:35:47 -05006486 }
Yan Zheng2c47e6052009-06-27 21:07:35 -04006487 btrfs_release_path(root, path);
6488 BUG_ON(err);
6489
6490 ret = btrfs_del_root(trans, tree_root, &root->root_key);
6491 BUG_ON(ret);
6492
Yan, Zheng76dda932009-09-21 16:00:26 -04006493 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
6494 ret = btrfs_find_last_root(tree_root, root->root_key.objectid,
6495 NULL, NULL);
6496 BUG_ON(ret < 0);
6497 if (ret > 0) {
Josef Bacik84cd9482010-12-08 12:24:01 -05006498 /* if we fail to delete the orphan item this time
6499 * around, it'll get picked up the next time.
6500 *
6501 * The most common failure here is just -ENOENT.
6502 */
6503 btrfs_del_orphan_item(trans, tree_root,
6504 root->root_key.objectid);
Yan, Zheng76dda932009-09-21 16:00:26 -04006505 }
6506 }
6507
6508 if (root->in_radix) {
6509 btrfs_free_fs_root(tree_root->fs_info, root);
6510 } else {
6511 free_extent_buffer(root->node);
6512 free_extent_buffer(root->commit_root);
6513 kfree(root);
6514 }
Chris Mason9f3a7422007-08-07 15:52:19 -04006515out:
Yan, Zheng3fd0a552010-05-16 10:49:59 -04006516 btrfs_end_transaction_throttle(trans, tree_root);
Yan Zheng2c47e6052009-06-27 21:07:35 -04006517 kfree(wc);
Chris Mason5caf2a02007-04-02 11:20:42 -04006518 btrfs_free_path(path);
Yan Zheng2c47e6052009-06-27 21:07:35 -04006519 return err;
Chris Mason20524f02007-03-10 06:35:47 -05006520}
Chris Mason9078a3e2007-04-26 16:46:15 -04006521
Yan Zheng2c47e6052009-06-27 21:07:35 -04006522/*
6523 * drop subtree rooted at tree block 'node'.
6524 *
6525 * NOTE: this function will unlock and release tree block 'node'
6526 */
Yan Zhengf82d02d2008-10-29 14:49:05 -04006527int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
6528 struct btrfs_root *root,
6529 struct extent_buffer *node,
6530 struct extent_buffer *parent)
6531{
6532 struct btrfs_path *path;
Yan Zheng2c47e6052009-06-27 21:07:35 -04006533 struct walk_control *wc;
Yan Zhengf82d02d2008-10-29 14:49:05 -04006534 int level;
6535 int parent_level;
6536 int ret = 0;
6537 int wret;
6538
Yan Zheng2c47e6052009-06-27 21:07:35 -04006539 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
6540
Yan Zhengf82d02d2008-10-29 14:49:05 -04006541 path = btrfs_alloc_path();
Tsutomu Itohdb5b4932011-03-23 08:14:16 +00006542 if (!path)
6543 return -ENOMEM;
Yan Zhengf82d02d2008-10-29 14:49:05 -04006544
Yan Zheng2c47e6052009-06-27 21:07:35 -04006545 wc = kzalloc(sizeof(*wc), GFP_NOFS);
Tsutomu Itohdb5b4932011-03-23 08:14:16 +00006546 if (!wc) {
6547 btrfs_free_path(path);
6548 return -ENOMEM;
6549 }
Yan Zheng2c47e6052009-06-27 21:07:35 -04006550
Chris Masonb9447ef2009-03-09 11:45:38 -04006551 btrfs_assert_tree_locked(parent);
Yan Zhengf82d02d2008-10-29 14:49:05 -04006552 parent_level = btrfs_header_level(parent);
6553 extent_buffer_get(parent);
6554 path->nodes[parent_level] = parent;
6555 path->slots[parent_level] = btrfs_header_nritems(parent);
6556
Chris Masonb9447ef2009-03-09 11:45:38 -04006557 btrfs_assert_tree_locked(node);
Yan Zhengf82d02d2008-10-29 14:49:05 -04006558 level = btrfs_header_level(node);
Yan Zhengf82d02d2008-10-29 14:49:05 -04006559 path->nodes[level] = node;
6560 path->slots[level] = 0;
Yan Zheng2c47e6052009-06-27 21:07:35 -04006561 path->locks[level] = 1;
6562
6563 wc->refs[parent_level] = 1;
6564 wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
6565 wc->level = level;
6566 wc->shared_level = -1;
6567 wc->stage = DROP_REFERENCE;
6568 wc->update_ref = 0;
6569 wc->keep_locks = 1;
Yan, Zheng1c4850e2009-09-21 15:55:59 -04006570 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
Yan Zhengf82d02d2008-10-29 14:49:05 -04006571
6572 while (1) {
Yan Zheng2c47e6052009-06-27 21:07:35 -04006573 wret = walk_down_tree(trans, root, path, wc);
6574 if (wret < 0) {
Yan Zhengf82d02d2008-10-29 14:49:05 -04006575 ret = wret;
Yan Zhengf82d02d2008-10-29 14:49:05 -04006576 break;
Yan Zheng2c47e6052009-06-27 21:07:35 -04006577 }
Yan Zhengf82d02d2008-10-29 14:49:05 -04006578
Yan Zheng2c47e6052009-06-27 21:07:35 -04006579 wret = walk_up_tree(trans, root, path, wc, parent_level);
Yan Zhengf82d02d2008-10-29 14:49:05 -04006580 if (wret < 0)
6581 ret = wret;
6582 if (wret != 0)
6583 break;
6584 }
6585
Yan Zheng2c47e6052009-06-27 21:07:35 -04006586 kfree(wc);
Yan Zhengf82d02d2008-10-29 14:49:05 -04006587 btrfs_free_path(path);
6588 return ret;
6589}
6590
Yan Zheng5d4f98a2009-06-10 10:45:14 -04006591#if 0
Chris Mason8e7bf942008-04-28 09:02:36 -04006592static unsigned long calc_ra(unsigned long start, unsigned long last,
6593 unsigned long nr)
6594{
6595 return min(last, start + nr - 1);
6596}
6597
Chris Masond3977122009-01-05 21:25:51 -05006598static noinline int relocate_inode_pages(struct inode *inode, u64 start,
Chris Mason98ed5172008-01-03 10:01:48 -05006599 u64 len)
Chris Masonedbd8d42007-12-21 16:27:24 -05006600{
6601 u64 page_start;
6602 u64 page_end;
Zheng Yan1a40e232008-09-26 10:09:34 -04006603 unsigned long first_index;
Chris Masonedbd8d42007-12-21 16:27:24 -05006604 unsigned long last_index;
Chris Masonedbd8d42007-12-21 16:27:24 -05006605 unsigned long i;
6606 struct page *page;
Chris Masond1310b22008-01-24 16:13:08 -05006607 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
Chris Mason4313b392008-01-03 09:08:48 -05006608 struct file_ra_state *ra;
Chris Mason3eaa2882008-07-24 11:57:52 -04006609 struct btrfs_ordered_extent *ordered;
Zheng Yan1a40e232008-09-26 10:09:34 -04006610 unsigned int total_read = 0;
6611 unsigned int total_dirty = 0;
6612 int ret = 0;
Chris Mason4313b392008-01-03 09:08:48 -05006613
6614 ra = kzalloc(sizeof(*ra), GFP_NOFS);
Tsutomu Itoh5df67082011-02-01 09:17:35 +00006615 if (!ra)
6616 return -ENOMEM;
Chris Masonedbd8d42007-12-21 16:27:24 -05006617
6618 mutex_lock(&inode->i_mutex);
Zheng Yan1a40e232008-09-26 10:09:34 -04006619 first_index = start >> PAGE_CACHE_SHIFT;
Chris Masonedbd8d42007-12-21 16:27:24 -05006620 last_index = (start + len - 1) >> PAGE_CACHE_SHIFT;
6621
Zheng Yan1a40e232008-09-26 10:09:34 -04006622 /* make sure the dirty trick played by the caller work */
6623 ret = invalidate_inode_pages2_range(inode->i_mapping,
6624 first_index, last_index);
6625 if (ret)
6626 goto out_unlock;
Chris Mason8e7bf942008-04-28 09:02:36 -04006627
Chris Mason4313b392008-01-03 09:08:48 -05006628 file_ra_state_init(ra, inode->i_mapping);
Chris Masonedbd8d42007-12-21 16:27:24 -05006629
Zheng Yan1a40e232008-09-26 10:09:34 -04006630 for (i = first_index ; i <= last_index; i++) {
6631 if (total_read % ra->ra_pages == 0) {
Chris Mason8e7bf942008-04-28 09:02:36 -04006632 btrfs_force_ra(inode->i_mapping, ra, NULL, i,
Zheng Yan1a40e232008-09-26 10:09:34 -04006633 calc_ra(i, last_index, ra->ra_pages));
Chris Mason8e7bf942008-04-28 09:02:36 -04006634 }
6635 total_read++;
Chris Mason3eaa2882008-07-24 11:57:52 -04006636again:
6637 if (((u64)i << PAGE_CACHE_SHIFT) > i_size_read(inode))
Zheng Yan1a40e232008-09-26 10:09:34 -04006638 BUG_ON(1);
Chris Masonedbd8d42007-12-21 16:27:24 -05006639 page = grab_cache_page(inode->i_mapping, i);
Chris Masona061fc82008-05-07 11:43:44 -04006640 if (!page) {
Zheng Yan1a40e232008-09-26 10:09:34 -04006641 ret = -ENOMEM;
Chris Masonedbd8d42007-12-21 16:27:24 -05006642 goto out_unlock;
Chris Masona061fc82008-05-07 11:43:44 -04006643 }
Chris Masonedbd8d42007-12-21 16:27:24 -05006644 if (!PageUptodate(page)) {
6645 btrfs_readpage(NULL, page);
6646 lock_page(page);
6647 if (!PageUptodate(page)) {
6648 unlock_page(page);
6649 page_cache_release(page);
Zheng Yan1a40e232008-09-26 10:09:34 -04006650 ret = -EIO;
Chris Masonedbd8d42007-12-21 16:27:24 -05006651 goto out_unlock;
6652 }
6653 }
Chris Masonec44a352008-04-28 15:29:52 -04006654 wait_on_page_writeback(page);
Chris Mason3eaa2882008-07-24 11:57:52 -04006655
Chris Masonedbd8d42007-12-21 16:27:24 -05006656 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
6657 page_end = page_start + PAGE_CACHE_SIZE - 1;
Chris Masond1310b22008-01-24 16:13:08 -05006658 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
Chris Masonedbd8d42007-12-21 16:27:24 -05006659
Chris Mason3eaa2882008-07-24 11:57:52 -04006660 ordered = btrfs_lookup_ordered_extent(inode, page_start);
6661 if (ordered) {
6662 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
6663 unlock_page(page);
6664 page_cache_release(page);
6665 btrfs_start_ordered_extent(inode, ordered, 1);
6666 btrfs_put_ordered_extent(ordered);
6667 goto again;
6668 }
6669 set_page_extent_mapped(page);
6670
Zheng Yan1a40e232008-09-26 10:09:34 -04006671 if (i == first_index)
6672 set_extent_bits(io_tree, page_start, page_end,
6673 EXTENT_BOUNDARY, GFP_NOFS);
Yan Zheng1f80e4d2008-12-19 10:59:04 -05006674 btrfs_set_extent_delalloc(inode, page_start, page_end);
Zheng Yan1a40e232008-09-26 10:09:34 -04006675
Chris Masona061fc82008-05-07 11:43:44 -04006676 set_page_dirty(page);
Zheng Yan1a40e232008-09-26 10:09:34 -04006677 total_dirty++;
Chris Masonedbd8d42007-12-21 16:27:24 -05006678
Chris Masond1310b22008-01-24 16:13:08 -05006679 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
Chris Masonedbd8d42007-12-21 16:27:24 -05006680 unlock_page(page);
6681 page_cache_release(page);
6682 }
6683
6684out_unlock:
Chris Masonec44a352008-04-28 15:29:52 -04006685 kfree(ra);
Chris Masonedbd8d42007-12-21 16:27:24 -05006686 mutex_unlock(&inode->i_mutex);
Zheng Yan1a40e232008-09-26 10:09:34 -04006687 balance_dirty_pages_ratelimited_nr(inode->i_mapping, total_dirty);
Chris Masonbf4ef672008-05-08 13:26:18 -04006688 return ret;
6689}
6690
Chris Masond3977122009-01-05 21:25:51 -05006691static noinline int relocate_data_extent(struct inode *reloc_inode,
Zheng Yan1a40e232008-09-26 10:09:34 -04006692 struct btrfs_key *extent_key,
6693 u64 offset)
Chris Masonedbd8d42007-12-21 16:27:24 -05006694{
Zheng Yan1a40e232008-09-26 10:09:34 -04006695 struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
6696 struct extent_map_tree *em_tree = &BTRFS_I(reloc_inode)->extent_tree;
6697 struct extent_map *em;
Yan Zheng66435582008-10-30 14:19:50 -04006698 u64 start = extent_key->objectid - offset;
6699 u64 end = start + extent_key->offset - 1;
Zheng Yan1a40e232008-09-26 10:09:34 -04006700
6701 em = alloc_extent_map(GFP_NOFS);
Tsutomu Itohc26a9202011-02-14 00:45:29 +00006702 BUG_ON(!em);
Zheng Yan1a40e232008-09-26 10:09:34 -04006703
Yan Zheng66435582008-10-30 14:19:50 -04006704 em->start = start;
Zheng Yan1a40e232008-09-26 10:09:34 -04006705 em->len = extent_key->offset;
Chris Masonc8b97812008-10-29 14:49:59 -04006706 em->block_len = extent_key->offset;
Zheng Yan1a40e232008-09-26 10:09:34 -04006707 em->block_start = extent_key->objectid;
6708 em->bdev = root->fs_info->fs_devices->latest_bdev;
6709 set_bit(EXTENT_FLAG_PINNED, &em->flags);
6710
6711 /* setup extent map to cheat btrfs_readpage */
Yan Zheng66435582008-10-30 14:19:50 -04006712 lock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
Zheng Yan1a40e232008-09-26 10:09:34 -04006713 while (1) {
6714 int ret;
Chris Mason890871b2009-09-02 16:24:52 -04006715 write_lock(&em_tree->lock);
Zheng Yan1a40e232008-09-26 10:09:34 -04006716 ret = add_extent_mapping(em_tree, em);
Chris Mason890871b2009-09-02 16:24:52 -04006717 write_unlock(&em_tree->lock);
Zheng Yan1a40e232008-09-26 10:09:34 -04006718 if (ret != -EEXIST) {
6719 free_extent_map(em);
6720 break;
6721 }
Yan Zheng66435582008-10-30 14:19:50 -04006722 btrfs_drop_extent_cache(reloc_inode, start, end, 0);
Zheng Yan1a40e232008-09-26 10:09:34 -04006723 }
Yan Zheng66435582008-10-30 14:19:50 -04006724 unlock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
Zheng Yan1a40e232008-09-26 10:09:34 -04006725
Yan Zheng66435582008-10-30 14:19:50 -04006726 return relocate_inode_pages(reloc_inode, start, extent_key->offset);
Zheng Yan1a40e232008-09-26 10:09:34 -04006727}
6728
6729struct btrfs_ref_path {
6730 u64 extent_start;
6731 u64 nodes[BTRFS_MAX_LEVEL];
6732 u64 root_objectid;
6733 u64 root_generation;
6734 u64 owner_objectid;
Zheng Yan1a40e232008-09-26 10:09:34 -04006735 u32 num_refs;
6736 int lowest_level;
6737 int current_level;
Yan Zhengf82d02d2008-10-29 14:49:05 -04006738 int shared_level;
6739
6740 struct btrfs_key node_keys[BTRFS_MAX_LEVEL];
6741 u64 new_nodes[BTRFS_MAX_LEVEL];
Zheng Yan1a40e232008-09-26 10:09:34 -04006742};
6743
6744struct disk_extent {
Chris Masonc8b97812008-10-29 14:49:59 -04006745 u64 ram_bytes;
Zheng Yan1a40e232008-09-26 10:09:34 -04006746 u64 disk_bytenr;
6747 u64 disk_num_bytes;
6748 u64 offset;
6749 u64 num_bytes;
Chris Masonc8b97812008-10-29 14:49:59 -04006750 u8 compression;
6751 u8 encryption;
6752 u16 other_encoding;
Zheng Yan1a40e232008-09-26 10:09:34 -04006753};
6754
6755static int is_cowonly_root(u64 root_objectid)
6756{
6757 if (root_objectid == BTRFS_ROOT_TREE_OBJECTID ||
6758 root_objectid == BTRFS_EXTENT_TREE_OBJECTID ||
6759 root_objectid == BTRFS_CHUNK_TREE_OBJECTID ||
6760 root_objectid == BTRFS_DEV_TREE_OBJECTID ||
Yan Zheng0403e472008-12-10 20:32:51 -05006761 root_objectid == BTRFS_TREE_LOG_OBJECTID ||
6762 root_objectid == BTRFS_CSUM_TREE_OBJECTID)
Zheng Yan1a40e232008-09-26 10:09:34 -04006763 return 1;
6764 return 0;
6765}
6766
Chris Masond3977122009-01-05 21:25:51 -05006767static noinline int __next_ref_path(struct btrfs_trans_handle *trans,
Zheng Yan1a40e232008-09-26 10:09:34 -04006768 struct btrfs_root *extent_root,
6769 struct btrfs_ref_path *ref_path,
6770 int first_time)
6771{
6772 struct extent_buffer *leaf;
6773 struct btrfs_path *path;
Chris Mason4313b392008-01-03 09:08:48 -05006774 struct btrfs_extent_ref *ref;
Chris Masonedbd8d42007-12-21 16:27:24 -05006775 struct btrfs_key key;
6776 struct btrfs_key found_key;
Zheng Yan1a40e232008-09-26 10:09:34 -04006777 u64 bytenr;
Chris Masonedbd8d42007-12-21 16:27:24 -05006778 u32 nritems;
Zheng Yan1a40e232008-09-26 10:09:34 -04006779 int level;
6780 int ret = 1;
Chris Masonedbd8d42007-12-21 16:27:24 -05006781
Zheng Yan1a40e232008-09-26 10:09:34 -04006782 path = btrfs_alloc_path();
6783 if (!path)
6784 return -ENOMEM;
6785
Zheng Yan1a40e232008-09-26 10:09:34 -04006786 if (first_time) {
6787 ref_path->lowest_level = -1;
6788 ref_path->current_level = -1;
Yan Zhengf82d02d2008-10-29 14:49:05 -04006789 ref_path->shared_level = -1;
Zheng Yan1a40e232008-09-26 10:09:34 -04006790 goto walk_up;
Chris Masona061fc82008-05-07 11:43:44 -04006791 }
Zheng Yan1a40e232008-09-26 10:09:34 -04006792walk_down:
6793 level = ref_path->current_level - 1;
6794 while (level >= -1) {
6795 u64 parent;
6796 if (level < ref_path->lowest_level)
6797 break;
Chris Masonedbd8d42007-12-21 16:27:24 -05006798
Chris Masond3977122009-01-05 21:25:51 -05006799 if (level >= 0)
Zheng Yan1a40e232008-09-26 10:09:34 -04006800 bytenr = ref_path->nodes[level];
Chris Masond3977122009-01-05 21:25:51 -05006801 else
Zheng Yan1a40e232008-09-26 10:09:34 -04006802 bytenr = ref_path->extent_start;
Zheng Yan1a40e232008-09-26 10:09:34 -04006803 BUG_ON(bytenr == 0);
Chris Masonedbd8d42007-12-21 16:27:24 -05006804
Zheng Yan1a40e232008-09-26 10:09:34 -04006805 parent = ref_path->nodes[level + 1];
6806 ref_path->nodes[level + 1] = 0;
6807 ref_path->current_level = level;
6808 BUG_ON(parent == 0);
6809
6810 key.objectid = bytenr;
6811 key.offset = parent + 1;
6812 key.type = BTRFS_EXTENT_REF_KEY;
6813
6814 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
Chris Masonedbd8d42007-12-21 16:27:24 -05006815 if (ret < 0)
6816 goto out;
Zheng Yan1a40e232008-09-26 10:09:34 -04006817 BUG_ON(ret == 0);
Chris Masonedbd8d42007-12-21 16:27:24 -05006818
Chris Masonedbd8d42007-12-21 16:27:24 -05006819 leaf = path->nodes[0];
6820 nritems = btrfs_header_nritems(leaf);
Zheng Yan1a40e232008-09-26 10:09:34 -04006821 if (path->slots[0] >= nritems) {
Chris Masona061fc82008-05-07 11:43:44 -04006822 ret = btrfs_next_leaf(extent_root, path);
Chris Masona061fc82008-05-07 11:43:44 -04006823 if (ret < 0)
6824 goto out;
Zheng Yan1a40e232008-09-26 10:09:34 -04006825 if (ret > 0)
6826 goto next;
Chris Masonbf4ef672008-05-08 13:26:18 -04006827 leaf = path->nodes[0];
Chris Masona061fc82008-05-07 11:43:44 -04006828 }
Chris Masonedbd8d42007-12-21 16:27:24 -05006829
6830 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
Zheng Yan1a40e232008-09-26 10:09:34 -04006831 if (found_key.objectid == bytenr &&
Yan Zhengf82d02d2008-10-29 14:49:05 -04006832 found_key.type == BTRFS_EXTENT_REF_KEY) {
6833 if (level < ref_path->shared_level)
6834 ref_path->shared_level = level;
Zheng Yan1a40e232008-09-26 10:09:34 -04006835 goto found;
Yan Zhengf82d02d2008-10-29 14:49:05 -04006836 }
Zheng Yan1a40e232008-09-26 10:09:34 -04006837next:
6838 level--;
6839 btrfs_release_path(extent_root, path);
Yan Zhengd899e052008-10-30 14:25:28 -04006840 cond_resched();
Zheng Yan1a40e232008-09-26 10:09:34 -04006841 }
6842 /* reached lowest level */
6843 ret = 1;
6844 goto out;
6845walk_up:
6846 level = ref_path->current_level;
6847 while (level < BTRFS_MAX_LEVEL - 1) {
6848 u64 ref_objectid;
Chris Masond3977122009-01-05 21:25:51 -05006849
6850 if (level >= 0)
Zheng Yan1a40e232008-09-26 10:09:34 -04006851 bytenr = ref_path->nodes[level];
Chris Masond3977122009-01-05 21:25:51 -05006852 else
Zheng Yan1a40e232008-09-26 10:09:34 -04006853 bytenr = ref_path->extent_start;
Chris Masond3977122009-01-05 21:25:51 -05006854
Zheng Yan1a40e232008-09-26 10:09:34 -04006855 BUG_ON(bytenr == 0);
Chris Masonedbd8d42007-12-21 16:27:24 -05006856
Zheng Yan1a40e232008-09-26 10:09:34 -04006857 key.objectid = bytenr;
6858 key.offset = 0;
6859 key.type = BTRFS_EXTENT_REF_KEY;
Chris Masonedbd8d42007-12-21 16:27:24 -05006860
Zheng Yan1a40e232008-09-26 10:09:34 -04006861 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
6862 if (ret < 0)
Chris Masonedbd8d42007-12-21 16:27:24 -05006863 goto out;
Zheng Yan1a40e232008-09-26 10:09:34 -04006864
6865 leaf = path->nodes[0];
6866 nritems = btrfs_header_nritems(leaf);
6867 if (path->slots[0] >= nritems) {
6868 ret = btrfs_next_leaf(extent_root, path);
6869 if (ret < 0)
6870 goto out;
6871 if (ret > 0) {
6872 /* the extent was freed by someone */
6873 if (ref_path->lowest_level == level)
6874 goto out;
6875 btrfs_release_path(extent_root, path);
6876 goto walk_down;
6877 }
6878 leaf = path->nodes[0];
6879 }
6880
6881 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6882 if (found_key.objectid != bytenr ||
6883 found_key.type != BTRFS_EXTENT_REF_KEY) {
6884 /* the extent was freed by someone */
6885 if (ref_path->lowest_level == level) {
6886 ret = 1;
6887 goto out;
6888 }
6889 btrfs_release_path(extent_root, path);
6890 goto walk_down;
6891 }
6892found:
6893 ref = btrfs_item_ptr(leaf, path->slots[0],
6894 struct btrfs_extent_ref);
6895 ref_objectid = btrfs_ref_objectid(leaf, ref);
6896 if (ref_objectid < BTRFS_FIRST_FREE_OBJECTID) {
6897 if (first_time) {
6898 level = (int)ref_objectid;
6899 BUG_ON(level >= BTRFS_MAX_LEVEL);
6900 ref_path->lowest_level = level;
6901 ref_path->current_level = level;
6902 ref_path->nodes[level] = bytenr;
6903 } else {
6904 WARN_ON(ref_objectid != level);
6905 }
6906 } else {
6907 WARN_ON(level != -1);
6908 }
6909 first_time = 0;
6910
6911 if (ref_path->lowest_level == level) {
6912 ref_path->owner_objectid = ref_objectid;
Zheng Yan1a40e232008-09-26 10:09:34 -04006913 ref_path->num_refs = btrfs_ref_num_refs(leaf, ref);
6914 }
6915
6916 /*
6917 * the block is tree root or the block isn't in reference
6918 * counted tree.
6919 */
6920 if (found_key.objectid == found_key.offset ||
6921 is_cowonly_root(btrfs_ref_root(leaf, ref))) {
6922 ref_path->root_objectid = btrfs_ref_root(leaf, ref);
6923 ref_path->root_generation =
6924 btrfs_ref_generation(leaf, ref);
6925 if (level < 0) {
6926 /* special reference from the tree log */
6927 ref_path->nodes[0] = found_key.offset;
6928 ref_path->current_level = 0;
6929 }
6930 ret = 0;
6931 goto out;
6932 }
6933
6934 level++;
6935 BUG_ON(ref_path->nodes[level] != 0);
6936 ref_path->nodes[level] = found_key.offset;
6937 ref_path->current_level = level;
6938
6939 /*
6940 * the reference was created in the running transaction,
6941 * no need to continue walking up.
6942 */
6943 if (btrfs_ref_generation(leaf, ref) == trans->transid) {
6944 ref_path->root_objectid = btrfs_ref_root(leaf, ref);
6945 ref_path->root_generation =
6946 btrfs_ref_generation(leaf, ref);
6947 ret = 0;
6948 goto out;
6949 }
6950
6951 btrfs_release_path(extent_root, path);
Yan Zhengd899e052008-10-30 14:25:28 -04006952 cond_resched();
Zheng Yan1a40e232008-09-26 10:09:34 -04006953 }
6954 /* reached max tree level, but no tree root found. */
6955 BUG();
6956out:
Zheng Yan1a40e232008-09-26 10:09:34 -04006957 btrfs_free_path(path);
6958 return ret;
6959}
6960
6961static int btrfs_first_ref_path(struct btrfs_trans_handle *trans,
6962 struct btrfs_root *extent_root,
6963 struct btrfs_ref_path *ref_path,
6964 u64 extent_start)
6965{
6966 memset(ref_path, 0, sizeof(*ref_path));
6967 ref_path->extent_start = extent_start;
6968
6969 return __next_ref_path(trans, extent_root, ref_path, 1);
6970}
6971
6972static int btrfs_next_ref_path(struct btrfs_trans_handle *trans,
6973 struct btrfs_root *extent_root,
6974 struct btrfs_ref_path *ref_path)
6975{
6976 return __next_ref_path(trans, extent_root, ref_path, 0);
6977}
6978
Chris Masond3977122009-01-05 21:25:51 -05006979static noinline int get_new_locations(struct inode *reloc_inode,
Zheng Yan1a40e232008-09-26 10:09:34 -04006980 struct btrfs_key *extent_key,
6981 u64 offset, int no_fragment,
6982 struct disk_extent **extents,
6983 int *nr_extents)
6984{
6985 struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
6986 struct btrfs_path *path;
6987 struct btrfs_file_extent_item *fi;
6988 struct extent_buffer *leaf;
6989 struct disk_extent *exts = *extents;
6990 struct btrfs_key found_key;
6991 u64 cur_pos;
6992 u64 last_byte;
6993 u32 nritems;
6994 int nr = 0;
6995 int max = *nr_extents;
6996 int ret;
6997
6998 WARN_ON(!no_fragment && *extents);
6999 if (!exts) {
7000 max = 1;
7001 exts = kmalloc(sizeof(*exts) * max, GFP_NOFS);
7002 if (!exts)
7003 return -ENOMEM;
7004 }
7005
7006 path = btrfs_alloc_path();
Tsutomu Itohdb5b4932011-03-23 08:14:16 +00007007 if (!path) {
7008 if (exts != *extents)
7009 kfree(exts);
7010 return -ENOMEM;
7011 }
Zheng Yan1a40e232008-09-26 10:09:34 -04007012
7013 cur_pos = extent_key->objectid - offset;
7014 last_byte = extent_key->objectid + extent_key->offset;
7015 ret = btrfs_lookup_file_extent(NULL, root, path, reloc_inode->i_ino,
7016 cur_pos, 0);
7017 if (ret < 0)
7018 goto out;
7019 if (ret > 0) {
7020 ret = -ENOENT;
7021 goto out;
7022 }
7023
7024 while (1) {
7025 leaf = path->nodes[0];
7026 nritems = btrfs_header_nritems(leaf);
7027 if (path->slots[0] >= nritems) {
7028 ret = btrfs_next_leaf(root, path);
7029 if (ret < 0)
7030 goto out;
7031 if (ret > 0)
7032 break;
7033 leaf = path->nodes[0];
7034 }
7035
7036 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
7037 if (found_key.offset != cur_pos ||
7038 found_key.type != BTRFS_EXTENT_DATA_KEY ||
7039 found_key.objectid != reloc_inode->i_ino)
7040 break;
7041
7042 fi = btrfs_item_ptr(leaf, path->slots[0],
7043 struct btrfs_file_extent_item);
7044 if (btrfs_file_extent_type(leaf, fi) !=
7045 BTRFS_FILE_EXTENT_REG ||
7046 btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
7047 break;
7048
7049 if (nr == max) {
7050 struct disk_extent *old = exts;
7051 max *= 2;
7052 exts = kzalloc(sizeof(*exts) * max, GFP_NOFS);
Yoshinori Sanodac97e52011-02-15 12:01:42 +00007053 if (!exts) {
7054 ret = -ENOMEM;
7055 goto out;
7056 }
Zheng Yan1a40e232008-09-26 10:09:34 -04007057 memcpy(exts, old, sizeof(*exts) * nr);
7058 if (old != *extents)
7059 kfree(old);
7060 }
7061
7062 exts[nr].disk_bytenr =
7063 btrfs_file_extent_disk_bytenr(leaf, fi);
7064 exts[nr].disk_num_bytes =
7065 btrfs_file_extent_disk_num_bytes(leaf, fi);
7066 exts[nr].offset = btrfs_file_extent_offset(leaf, fi);
7067 exts[nr].num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
Chris Masonc8b97812008-10-29 14:49:59 -04007068 exts[nr].ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
7069 exts[nr].compression = btrfs_file_extent_compression(leaf, fi);
7070 exts[nr].encryption = btrfs_file_extent_encryption(leaf, fi);
7071 exts[nr].other_encoding = btrfs_file_extent_other_encoding(leaf,
7072 fi);
Yan Zhengd899e052008-10-30 14:25:28 -04007073 BUG_ON(exts[nr].offset > 0);
7074 BUG_ON(exts[nr].compression || exts[nr].encryption);
7075 BUG_ON(exts[nr].num_bytes != exts[nr].disk_num_bytes);
Zheng Yan1a40e232008-09-26 10:09:34 -04007076
7077 cur_pos += exts[nr].num_bytes;
7078 nr++;
7079
7080 if (cur_pos + offset >= last_byte)
7081 break;
7082
7083 if (no_fragment) {
7084 ret = 1;
7085 goto out;
7086 }
7087 path->slots[0]++;
7088 }
7089
Yan Zheng1f80e4d2008-12-19 10:59:04 -05007090 BUG_ON(cur_pos + offset > last_byte);
Zheng Yan1a40e232008-09-26 10:09:34 -04007091 if (cur_pos + offset < last_byte) {
7092 ret = -ENOENT;
7093 goto out;
Chris Masonedbd8d42007-12-21 16:27:24 -05007094 }
7095 ret = 0;
7096out:
Zheng Yan1a40e232008-09-26 10:09:34 -04007097 btrfs_free_path(path);
7098 if (ret) {
7099 if (exts != *extents)
7100 kfree(exts);
7101 } else {
7102 *extents = exts;
7103 *nr_extents = nr;
7104 }
7105 return ret;
7106}
7107
Chris Masond3977122009-01-05 21:25:51 -05007108static noinline int replace_one_extent(struct btrfs_trans_handle *trans,
Zheng Yan1a40e232008-09-26 10:09:34 -04007109 struct btrfs_root *root,
7110 struct btrfs_path *path,
7111 struct btrfs_key *extent_key,
7112 struct btrfs_key *leaf_key,
7113 struct btrfs_ref_path *ref_path,
7114 struct disk_extent *new_extents,
7115 int nr_extents)
7116{
7117 struct extent_buffer *leaf;
7118 struct btrfs_file_extent_item *fi;
7119 struct inode *inode = NULL;
7120 struct btrfs_key key;
7121 u64 lock_start = 0;
7122 u64 lock_end = 0;
7123 u64 num_bytes;
7124 u64 ext_offset;
Yan Zheng86288a12009-01-21 10:49:16 -05007125 u64 search_end = (u64)-1;
Zheng Yan1a40e232008-09-26 10:09:34 -04007126 u32 nritems;
Yan Zheng3bb1a1b2008-10-09 11:46:24 -04007127 int nr_scaned = 0;
Zheng Yan1a40e232008-09-26 10:09:34 -04007128 int extent_locked = 0;
Yan Zhengd899e052008-10-30 14:25:28 -04007129 int extent_type;
Zheng Yan1a40e232008-09-26 10:09:34 -04007130 int ret;
7131
Yan Zheng3bb1a1b2008-10-09 11:46:24 -04007132 memcpy(&key, leaf_key, sizeof(key));
Zheng Yan1a40e232008-09-26 10:09:34 -04007133 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
Yan Zheng3bb1a1b2008-10-09 11:46:24 -04007134 if (key.objectid < ref_path->owner_objectid ||
7135 (key.objectid == ref_path->owner_objectid &&
7136 key.type < BTRFS_EXTENT_DATA_KEY)) {
7137 key.objectid = ref_path->owner_objectid;
7138 key.type = BTRFS_EXTENT_DATA_KEY;
7139 key.offset = 0;
7140 }
Zheng Yan1a40e232008-09-26 10:09:34 -04007141 }
7142
7143 while (1) {
7144 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
7145 if (ret < 0)
7146 goto out;
7147
7148 leaf = path->nodes[0];
7149 nritems = btrfs_header_nritems(leaf);
7150next:
7151 if (extent_locked && ret > 0) {
7152 /*
7153 * the file extent item was modified by someone
7154 * before the extent got locked.
7155 */
Zheng Yan1a40e232008-09-26 10:09:34 -04007156 unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
7157 lock_end, GFP_NOFS);
7158 extent_locked = 0;
7159 }
7160
7161 if (path->slots[0] >= nritems) {
Yan Zheng3bb1a1b2008-10-09 11:46:24 -04007162 if (++nr_scaned > 2)
Zheng Yan1a40e232008-09-26 10:09:34 -04007163 break;
7164
7165 BUG_ON(extent_locked);
7166 ret = btrfs_next_leaf(root, path);
7167 if (ret < 0)
7168 goto out;
7169 if (ret > 0)
7170 break;
7171 leaf = path->nodes[0];
7172 nritems = btrfs_header_nritems(leaf);
7173 }
7174
7175 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
7176
7177 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
7178 if ((key.objectid > ref_path->owner_objectid) ||
7179 (key.objectid == ref_path->owner_objectid &&
7180 key.type > BTRFS_EXTENT_DATA_KEY) ||
Yan Zheng86288a12009-01-21 10:49:16 -05007181 key.offset >= search_end)
Zheng Yan1a40e232008-09-26 10:09:34 -04007182 break;
7183 }
7184
7185 if (inode && key.objectid != inode->i_ino) {
7186 BUG_ON(extent_locked);
7187 btrfs_release_path(root, path);
7188 mutex_unlock(&inode->i_mutex);
7189 iput(inode);
7190 inode = NULL;
7191 continue;
7192 }
7193
7194 if (key.type != BTRFS_EXTENT_DATA_KEY) {
7195 path->slots[0]++;
7196 ret = 1;
7197 goto next;
7198 }
7199 fi = btrfs_item_ptr(leaf, path->slots[0],
7200 struct btrfs_file_extent_item);
Yan Zhengd899e052008-10-30 14:25:28 -04007201 extent_type = btrfs_file_extent_type(leaf, fi);
7202 if ((extent_type != BTRFS_FILE_EXTENT_REG &&
7203 extent_type != BTRFS_FILE_EXTENT_PREALLOC) ||
Zheng Yan1a40e232008-09-26 10:09:34 -04007204 (btrfs_file_extent_disk_bytenr(leaf, fi) !=
7205 extent_key->objectid)) {
7206 path->slots[0]++;
7207 ret = 1;
7208 goto next;
7209 }
7210
7211 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
7212 ext_offset = btrfs_file_extent_offset(leaf, fi);
7213
Yan Zheng86288a12009-01-21 10:49:16 -05007214 if (search_end == (u64)-1) {
7215 search_end = key.offset - ext_offset +
7216 btrfs_file_extent_ram_bytes(leaf, fi);
7217 }
Zheng Yan1a40e232008-09-26 10:09:34 -04007218
7219 if (!extent_locked) {
7220 lock_start = key.offset;
7221 lock_end = lock_start + num_bytes - 1;
7222 } else {
Yan Zheng66435582008-10-30 14:19:50 -04007223 if (lock_start > key.offset ||
7224 lock_end + 1 < key.offset + num_bytes) {
7225 unlock_extent(&BTRFS_I(inode)->io_tree,
7226 lock_start, lock_end, GFP_NOFS);
7227 extent_locked = 0;
7228 }
Zheng Yan1a40e232008-09-26 10:09:34 -04007229 }
7230
7231 if (!inode) {
7232 btrfs_release_path(root, path);
7233
7234 inode = btrfs_iget_locked(root->fs_info->sb,
7235 key.objectid, root);
7236 if (inode->i_state & I_NEW) {
7237 BTRFS_I(inode)->root = root;
7238 BTRFS_I(inode)->location.objectid =
7239 key.objectid;
7240 BTRFS_I(inode)->location.type =
7241 BTRFS_INODE_ITEM_KEY;
7242 BTRFS_I(inode)->location.offset = 0;
7243 btrfs_read_locked_inode(inode);
7244 unlock_new_inode(inode);
7245 }
7246 /*
7247 * some code call btrfs_commit_transaction while
7248 * holding the i_mutex, so we can't use mutex_lock
7249 * here.
7250 */
7251 if (is_bad_inode(inode) ||
7252 !mutex_trylock(&inode->i_mutex)) {
7253 iput(inode);
7254 inode = NULL;
7255 key.offset = (u64)-1;
7256 goto skip;
7257 }
7258 }
7259
7260 if (!extent_locked) {
7261 struct btrfs_ordered_extent *ordered;
7262
7263 btrfs_release_path(root, path);
7264
7265 lock_extent(&BTRFS_I(inode)->io_tree, lock_start,
7266 lock_end, GFP_NOFS);
7267 ordered = btrfs_lookup_first_ordered_extent(inode,
7268 lock_end);
7269 if (ordered &&
7270 ordered->file_offset <= lock_end &&
7271 ordered->file_offset + ordered->len > lock_start) {
7272 unlock_extent(&BTRFS_I(inode)->io_tree,
7273 lock_start, lock_end, GFP_NOFS);
7274 btrfs_start_ordered_extent(inode, ordered, 1);
7275 btrfs_put_ordered_extent(ordered);
7276 key.offset += num_bytes;
7277 goto skip;
7278 }
7279 if (ordered)
7280 btrfs_put_ordered_extent(ordered);
7281
Zheng Yan1a40e232008-09-26 10:09:34 -04007282 extent_locked = 1;
7283 continue;
7284 }
7285
7286 if (nr_extents == 1) {
7287 /* update extent pointer in place */
Zheng Yan1a40e232008-09-26 10:09:34 -04007288 btrfs_set_file_extent_disk_bytenr(leaf, fi,
7289 new_extents[0].disk_bytenr);
7290 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
7291 new_extents[0].disk_num_bytes);
Zheng Yan1a40e232008-09-26 10:09:34 -04007292 btrfs_mark_buffer_dirty(leaf);
7293
7294 btrfs_drop_extent_cache(inode, key.offset,
7295 key.offset + num_bytes - 1, 0);
7296
7297 ret = btrfs_inc_extent_ref(trans, root,
7298 new_extents[0].disk_bytenr,
7299 new_extents[0].disk_num_bytes,
7300 leaf->start,
7301 root->root_key.objectid,
7302 trans->transid,
Yan Zheng3bb1a1b2008-10-09 11:46:24 -04007303 key.objectid);
Zheng Yan1a40e232008-09-26 10:09:34 -04007304 BUG_ON(ret);
7305
7306 ret = btrfs_free_extent(trans, root,
7307 extent_key->objectid,
7308 extent_key->offset,
7309 leaf->start,
7310 btrfs_header_owner(leaf),
7311 btrfs_header_generation(leaf),
Yan Zheng3bb1a1b2008-10-09 11:46:24 -04007312 key.objectid, 0);
Zheng Yan1a40e232008-09-26 10:09:34 -04007313 BUG_ON(ret);
7314
7315 btrfs_release_path(root, path);
7316 key.offset += num_bytes;
7317 } else {
Yan Zhengd899e052008-10-30 14:25:28 -04007318 BUG_ON(1);
7319#if 0
Zheng Yan1a40e232008-09-26 10:09:34 -04007320 u64 alloc_hint;
7321 u64 extent_len;
7322 int i;
7323 /*
7324 * drop old extent pointer at first, then insert the
7325 * new pointers one bye one
7326 */
7327 btrfs_release_path(root, path);
7328 ret = btrfs_drop_extents(trans, root, inode, key.offset,
7329 key.offset + num_bytes,
7330 key.offset, &alloc_hint);
7331 BUG_ON(ret);
7332
7333 for (i = 0; i < nr_extents; i++) {
7334 if (ext_offset >= new_extents[i].num_bytes) {
7335 ext_offset -= new_extents[i].num_bytes;
7336 continue;
7337 }
7338 extent_len = min(new_extents[i].num_bytes -
7339 ext_offset, num_bytes);
7340
7341 ret = btrfs_insert_empty_item(trans, root,
7342 path, &key,
7343 sizeof(*fi));
7344 BUG_ON(ret);
7345
7346 leaf = path->nodes[0];
7347 fi = btrfs_item_ptr(leaf, path->slots[0],
7348 struct btrfs_file_extent_item);
7349 btrfs_set_file_extent_generation(leaf, fi,
7350 trans->transid);
7351 btrfs_set_file_extent_type(leaf, fi,
7352 BTRFS_FILE_EXTENT_REG);
7353 btrfs_set_file_extent_disk_bytenr(leaf, fi,
7354 new_extents[i].disk_bytenr);
7355 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
7356 new_extents[i].disk_num_bytes);
Chris Masonc8b97812008-10-29 14:49:59 -04007357 btrfs_set_file_extent_ram_bytes(leaf, fi,
7358 new_extents[i].ram_bytes);
7359
7360 btrfs_set_file_extent_compression(leaf, fi,
7361 new_extents[i].compression);
7362 btrfs_set_file_extent_encryption(leaf, fi,
7363 new_extents[i].encryption);
7364 btrfs_set_file_extent_other_encoding(leaf, fi,
7365 new_extents[i].other_encoding);
7366
Zheng Yan1a40e232008-09-26 10:09:34 -04007367 btrfs_set_file_extent_num_bytes(leaf, fi,
7368 extent_len);
7369 ext_offset += new_extents[i].offset;
7370 btrfs_set_file_extent_offset(leaf, fi,
7371 ext_offset);
7372 btrfs_mark_buffer_dirty(leaf);
7373
7374 btrfs_drop_extent_cache(inode, key.offset,
7375 key.offset + extent_len - 1, 0);
7376
7377 ret = btrfs_inc_extent_ref(trans, root,
7378 new_extents[i].disk_bytenr,
7379 new_extents[i].disk_num_bytes,
7380 leaf->start,
7381 root->root_key.objectid,
Yan Zheng3bb1a1b2008-10-09 11:46:24 -04007382 trans->transid, key.objectid);
Zheng Yan1a40e232008-09-26 10:09:34 -04007383 BUG_ON(ret);
7384 btrfs_release_path(root, path);
7385
Yan Zhenga76a3cd2008-10-09 11:46:29 -04007386 inode_add_bytes(inode, extent_len);
Zheng Yan1a40e232008-09-26 10:09:34 -04007387
7388 ext_offset = 0;
7389 num_bytes -= extent_len;
7390 key.offset += extent_len;
7391
7392 if (num_bytes == 0)
7393 break;
7394 }
7395 BUG_ON(i >= nr_extents);
Yan Zhengd899e052008-10-30 14:25:28 -04007396#endif
Zheng Yan1a40e232008-09-26 10:09:34 -04007397 }
7398
7399 if (extent_locked) {
Zheng Yan1a40e232008-09-26 10:09:34 -04007400 unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
7401 lock_end, GFP_NOFS);
7402 extent_locked = 0;
7403 }
7404skip:
7405 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS &&
Yan Zheng86288a12009-01-21 10:49:16 -05007406 key.offset >= search_end)
Zheng Yan1a40e232008-09-26 10:09:34 -04007407 break;
7408
7409 cond_resched();
7410 }
7411 ret = 0;
7412out:
7413 btrfs_release_path(root, path);
7414 if (inode) {
7415 mutex_unlock(&inode->i_mutex);
7416 if (extent_locked) {
Zheng Yan1a40e232008-09-26 10:09:34 -04007417 unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
7418 lock_end, GFP_NOFS);
7419 }
7420 iput(inode);
7421 }
7422 return ret;
7423}
7424
Zheng Yan1a40e232008-09-26 10:09:34 -04007425int btrfs_reloc_tree_cache_ref(struct btrfs_trans_handle *trans,
7426 struct btrfs_root *root,
7427 struct extent_buffer *buf, u64 orig_start)
7428{
7429 int level;
7430 int ret;
7431
7432 BUG_ON(btrfs_header_generation(buf) != trans->transid);
7433 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
7434
7435 level = btrfs_header_level(buf);
7436 if (level == 0) {
7437 struct btrfs_leaf_ref *ref;
7438 struct btrfs_leaf_ref *orig_ref;
7439
7440 orig_ref = btrfs_lookup_leaf_ref(root, orig_start);
7441 if (!orig_ref)
7442 return -ENOENT;
7443
7444 ref = btrfs_alloc_leaf_ref(root, orig_ref->nritems);
7445 if (!ref) {
7446 btrfs_free_leaf_ref(root, orig_ref);
7447 return -ENOMEM;
7448 }
7449
7450 ref->nritems = orig_ref->nritems;
7451 memcpy(ref->extents, orig_ref->extents,
7452 sizeof(ref->extents[0]) * ref->nritems);
7453
7454 btrfs_free_leaf_ref(root, orig_ref);
7455
7456 ref->root_gen = trans->transid;
7457 ref->bytenr = buf->start;
7458 ref->owner = btrfs_header_owner(buf);
7459 ref->generation = btrfs_header_generation(buf);
Chris Masonbd56b302009-02-04 09:27:02 -05007460
Zheng Yan1a40e232008-09-26 10:09:34 -04007461 ret = btrfs_add_leaf_ref(root, ref, 0);
7462 WARN_ON(ret);
7463 btrfs_free_leaf_ref(root, ref);
7464 }
7465 return 0;
7466}
7467
Chris Masond3977122009-01-05 21:25:51 -05007468static noinline int invalidate_extent_cache(struct btrfs_root *root,
Zheng Yan1a40e232008-09-26 10:09:34 -04007469 struct extent_buffer *leaf,
7470 struct btrfs_block_group_cache *group,
7471 struct btrfs_root *target_root)
7472{
7473 struct btrfs_key key;
7474 struct inode *inode = NULL;
7475 struct btrfs_file_extent_item *fi;
Josef Bacik2ac55d42010-02-03 19:33:23 +00007476 struct extent_state *cached_state = NULL;
Zheng Yan1a40e232008-09-26 10:09:34 -04007477 u64 num_bytes;
7478 u64 skip_objectid = 0;
7479 u32 nritems;
7480 u32 i;
7481
7482 nritems = btrfs_header_nritems(leaf);
7483 for (i = 0; i < nritems; i++) {
7484 btrfs_item_key_to_cpu(leaf, &key, i);
7485 if (key.objectid == skip_objectid ||
7486 key.type != BTRFS_EXTENT_DATA_KEY)
7487 continue;
7488 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
7489 if (btrfs_file_extent_type(leaf, fi) ==
7490 BTRFS_FILE_EXTENT_INLINE)
7491 continue;
7492 if (btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
7493 continue;
7494 if (!inode || inode->i_ino != key.objectid) {
7495 iput(inode);
7496 inode = btrfs_ilookup(target_root->fs_info->sb,
7497 key.objectid, target_root, 1);
7498 }
7499 if (!inode) {
7500 skip_objectid = key.objectid;
7501 continue;
7502 }
7503 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
7504
Josef Bacik2ac55d42010-02-03 19:33:23 +00007505 lock_extent_bits(&BTRFS_I(inode)->io_tree, key.offset,
7506 key.offset + num_bytes - 1, 0, &cached_state,
7507 GFP_NOFS);
Zheng Yan1a40e232008-09-26 10:09:34 -04007508 btrfs_drop_extent_cache(inode, key.offset,
7509 key.offset + num_bytes - 1, 1);
Josef Bacik2ac55d42010-02-03 19:33:23 +00007510 unlock_extent_cached(&BTRFS_I(inode)->io_tree, key.offset,
7511 key.offset + num_bytes - 1, &cached_state,
7512 GFP_NOFS);
Zheng Yan1a40e232008-09-26 10:09:34 -04007513 cond_resched();
7514 }
7515 iput(inode);
7516 return 0;
7517}
7518
Chris Masond3977122009-01-05 21:25:51 -05007519static noinline int replace_extents_in_leaf(struct btrfs_trans_handle *trans,
Zheng Yan1a40e232008-09-26 10:09:34 -04007520 struct btrfs_root *root,
7521 struct extent_buffer *leaf,
7522 struct btrfs_block_group_cache *group,
7523 struct inode *reloc_inode)
7524{
7525 struct btrfs_key key;
7526 struct btrfs_key extent_key;
7527 struct btrfs_file_extent_item *fi;
7528 struct btrfs_leaf_ref *ref;
7529 struct disk_extent *new_extent;
7530 u64 bytenr;
7531 u64 num_bytes;
7532 u32 nritems;
7533 u32 i;
7534 int ext_index;
7535 int nr_extent;
7536 int ret;
7537
7538 new_extent = kmalloc(sizeof(*new_extent), GFP_NOFS);
Tsutomu Itohdb5b4932011-03-23 08:14:16 +00007539 if (!new_extent)
7540 return -ENOMEM;
Zheng Yan1a40e232008-09-26 10:09:34 -04007541
7542 ref = btrfs_lookup_leaf_ref(root, leaf->start);
7543 BUG_ON(!ref);
7544
7545 ext_index = -1;
7546 nritems = btrfs_header_nritems(leaf);
7547 for (i = 0; i < nritems; i++) {
7548 btrfs_item_key_to_cpu(leaf, &key, i);
7549 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
7550 continue;
7551 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
7552 if (btrfs_file_extent_type(leaf, fi) ==
7553 BTRFS_FILE_EXTENT_INLINE)
7554 continue;
7555 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
7556 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
7557 if (bytenr == 0)
7558 continue;
7559
7560 ext_index++;
7561 if (bytenr >= group->key.objectid + group->key.offset ||
7562 bytenr + num_bytes <= group->key.objectid)
7563 continue;
7564
7565 extent_key.objectid = bytenr;
7566 extent_key.offset = num_bytes;
7567 extent_key.type = BTRFS_EXTENT_ITEM_KEY;
7568 nr_extent = 1;
7569 ret = get_new_locations(reloc_inode, &extent_key,
7570 group->key.objectid, 1,
7571 &new_extent, &nr_extent);
7572 if (ret > 0)
7573 continue;
7574 BUG_ON(ret < 0);
7575
7576 BUG_ON(ref->extents[ext_index].bytenr != bytenr);
7577 BUG_ON(ref->extents[ext_index].num_bytes != num_bytes);
7578 ref->extents[ext_index].bytenr = new_extent->disk_bytenr;
7579 ref->extents[ext_index].num_bytes = new_extent->disk_num_bytes;
7580
Zheng Yan1a40e232008-09-26 10:09:34 -04007581 btrfs_set_file_extent_disk_bytenr(leaf, fi,
7582 new_extent->disk_bytenr);
7583 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
7584 new_extent->disk_num_bytes);
Zheng Yan1a40e232008-09-26 10:09:34 -04007585 btrfs_mark_buffer_dirty(leaf);
7586
7587 ret = btrfs_inc_extent_ref(trans, root,
7588 new_extent->disk_bytenr,
7589 new_extent->disk_num_bytes,
7590 leaf->start,
7591 root->root_key.objectid,
Yan Zheng3bb1a1b2008-10-09 11:46:24 -04007592 trans->transid, key.objectid);
Zheng Yan1a40e232008-09-26 10:09:34 -04007593 BUG_ON(ret);
Chris Mason56bec292009-03-13 10:10:06 -04007594
Zheng Yan1a40e232008-09-26 10:09:34 -04007595 ret = btrfs_free_extent(trans, root,
7596 bytenr, num_bytes, leaf->start,
7597 btrfs_header_owner(leaf),
7598 btrfs_header_generation(leaf),
Yan Zheng3bb1a1b2008-10-09 11:46:24 -04007599 key.objectid, 0);
Zheng Yan1a40e232008-09-26 10:09:34 -04007600 BUG_ON(ret);
7601 cond_resched();
7602 }
7603 kfree(new_extent);
7604 BUG_ON(ext_index + 1 != ref->nritems);
7605 btrfs_free_leaf_ref(root, ref);
7606 return 0;
7607}
7608
Yan Zhengf82d02d2008-10-29 14:49:05 -04007609int btrfs_free_reloc_root(struct btrfs_trans_handle *trans,
7610 struct btrfs_root *root)
Zheng Yan1a40e232008-09-26 10:09:34 -04007611{
7612 struct btrfs_root *reloc_root;
Yan Zhengf82d02d2008-10-29 14:49:05 -04007613 int ret;
Zheng Yan1a40e232008-09-26 10:09:34 -04007614
7615 if (root->reloc_root) {
7616 reloc_root = root->reloc_root;
7617 root->reloc_root = NULL;
7618 list_add(&reloc_root->dead_list,
7619 &root->fs_info->dead_reloc_roots);
Yan Zhengf82d02d2008-10-29 14:49:05 -04007620
7621 btrfs_set_root_bytenr(&reloc_root->root_item,
7622 reloc_root->node->start);
7623 btrfs_set_root_level(&root->root_item,
7624 btrfs_header_level(reloc_root->node));
7625 memset(&reloc_root->root_item.drop_progress, 0,
7626 sizeof(struct btrfs_disk_key));
7627 reloc_root->root_item.drop_level = 0;
7628
7629 ret = btrfs_update_root(trans, root->fs_info->tree_root,
7630 &reloc_root->root_key,
7631 &reloc_root->root_item);
7632 BUG_ON(ret);
Zheng Yan1a40e232008-09-26 10:09:34 -04007633 }
7634 return 0;
7635}
7636
7637int btrfs_drop_dead_reloc_roots(struct btrfs_root *root)
7638{
7639 struct btrfs_trans_handle *trans;
7640 struct btrfs_root *reloc_root;
7641 struct btrfs_root *prev_root = NULL;
7642 struct list_head dead_roots;
7643 int ret;
7644 unsigned long nr;
7645
7646 INIT_LIST_HEAD(&dead_roots);
7647 list_splice_init(&root->fs_info->dead_reloc_roots, &dead_roots);
7648
7649 while (!list_empty(&dead_roots)) {
7650 reloc_root = list_entry(dead_roots.prev,
7651 struct btrfs_root, dead_list);
7652 list_del_init(&reloc_root->dead_list);
7653
7654 BUG_ON(reloc_root->commit_root != NULL);
7655 while (1) {
7656 trans = btrfs_join_transaction(root, 1);
Tsutomu Itoh3612b492011-01-25 02:51:38 +00007657 BUG_ON(IS_ERR(trans));
Zheng Yan1a40e232008-09-26 10:09:34 -04007658
7659 mutex_lock(&root->fs_info->drop_mutex);
7660 ret = btrfs_drop_snapshot(trans, reloc_root);
7661 if (ret != -EAGAIN)
7662 break;
7663 mutex_unlock(&root->fs_info->drop_mutex);
7664
7665 nr = trans->blocks_used;
7666 ret = btrfs_end_transaction(trans, root);
7667 BUG_ON(ret);
7668 btrfs_btree_balance_dirty(root, nr);
7669 }
7670
7671 free_extent_buffer(reloc_root->node);
7672
7673 ret = btrfs_del_root(trans, root->fs_info->tree_root,
7674 &reloc_root->root_key);
7675 BUG_ON(ret);
7676 mutex_unlock(&root->fs_info->drop_mutex);
7677
7678 nr = trans->blocks_used;
7679 ret = btrfs_end_transaction(trans, root);
7680 BUG_ON(ret);
7681 btrfs_btree_balance_dirty(root, nr);
7682
7683 kfree(prev_root);
7684 prev_root = reloc_root;
7685 }
7686 if (prev_root) {
7687 btrfs_remove_leaf_refs(prev_root, (u64)-1, 0);
7688 kfree(prev_root);
7689 }
7690 return 0;
7691}
7692
7693int btrfs_add_dead_reloc_root(struct btrfs_root *root)
7694{
7695 list_add(&root->dead_list, &root->fs_info->dead_reloc_roots);
7696 return 0;
7697}
7698
7699int btrfs_cleanup_reloc_trees(struct btrfs_root *root)
7700{
7701 struct btrfs_root *reloc_root;
7702 struct btrfs_trans_handle *trans;
7703 struct btrfs_key location;
7704 int found;
7705 int ret;
7706
7707 mutex_lock(&root->fs_info->tree_reloc_mutex);
7708 ret = btrfs_find_dead_roots(root, BTRFS_TREE_RELOC_OBJECTID, NULL);
7709 BUG_ON(ret);
7710 found = !list_empty(&root->fs_info->dead_reloc_roots);
7711 mutex_unlock(&root->fs_info->tree_reloc_mutex);
7712
7713 if (found) {
7714 trans = btrfs_start_transaction(root, 1);
Tsutomu Itoh98d5dc12011-01-20 06:19:37 +00007715 BUG_ON(IS_ERR(trans));
Zheng Yan1a40e232008-09-26 10:09:34 -04007716 ret = btrfs_commit_transaction(trans, root);
7717 BUG_ON(ret);
7718 }
7719
7720 location.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
7721 location.offset = (u64)-1;
7722 location.type = BTRFS_ROOT_ITEM_KEY;
7723
7724 reloc_root = btrfs_read_fs_root_no_name(root->fs_info, &location);
7725 BUG_ON(!reloc_root);
Josef Bacik66b4ffd2011-01-31 16:22:42 -05007726 ret = btrfs_orphan_cleanup(reloc_root);
7727 BUG_ON(ret);
Zheng Yan1a40e232008-09-26 10:09:34 -04007728 return 0;
7729}
7730
Chris Masond3977122009-01-05 21:25:51 -05007731static noinline int init_reloc_tree(struct btrfs_trans_handle *trans,
Zheng Yan1a40e232008-09-26 10:09:34 -04007732 struct btrfs_root *root)
7733{
7734 struct btrfs_root *reloc_root;
7735 struct extent_buffer *eb;
7736 struct btrfs_root_item *root_item;
7737 struct btrfs_key root_key;
7738 int ret;
7739
7740 BUG_ON(!root->ref_cows);
7741 if (root->reloc_root)
7742 return 0;
7743
7744 root_item = kmalloc(sizeof(*root_item), GFP_NOFS);
Tsutomu Itohdb5b4932011-03-23 08:14:16 +00007745 if (!root_item)
7746 return -ENOMEM;
Zheng Yan1a40e232008-09-26 10:09:34 -04007747
7748 ret = btrfs_copy_root(trans, root, root->commit_root,
7749 &eb, BTRFS_TREE_RELOC_OBJECTID);
7750 BUG_ON(ret);
7751
7752 root_key.objectid = BTRFS_TREE_RELOC_OBJECTID;
7753 root_key.offset = root->root_key.objectid;
7754 root_key.type = BTRFS_ROOT_ITEM_KEY;
7755
7756 memcpy(root_item, &root->root_item, sizeof(root_item));
7757 btrfs_set_root_refs(root_item, 0);
7758 btrfs_set_root_bytenr(root_item, eb->start);
7759 btrfs_set_root_level(root_item, btrfs_header_level(eb));
Yan Zheng84234f32008-10-29 14:49:05 -04007760 btrfs_set_root_generation(root_item, trans->transid);
Zheng Yan1a40e232008-09-26 10:09:34 -04007761
7762 btrfs_tree_unlock(eb);
7763 free_extent_buffer(eb);
7764
7765 ret = btrfs_insert_root(trans, root->fs_info->tree_root,
7766 &root_key, root_item);
7767 BUG_ON(ret);
7768 kfree(root_item);
7769
7770 reloc_root = btrfs_read_fs_root_no_radix(root->fs_info->tree_root,
7771 &root_key);
Tsutomu Itohdb5b4932011-03-23 08:14:16 +00007772 BUG_ON(IS_ERR(reloc_root));
Zheng Yan1a40e232008-09-26 10:09:34 -04007773 reloc_root->last_trans = trans->transid;
7774 reloc_root->commit_root = NULL;
7775 reloc_root->ref_tree = &root->fs_info->reloc_ref_tree;
7776
7777 root->reloc_root = reloc_root;
7778 return 0;
7779}
7780
7781/*
7782 * Core function of space balance.
7783 *
7784 * The idea is using reloc trees to relocate tree blocks in reference
Yan Zhengf82d02d2008-10-29 14:49:05 -04007785 * counted roots. There is one reloc tree for each subvol, and all
7786 * reloc trees share same root key objectid. Reloc trees are snapshots
7787 * of the latest committed roots of subvols (root->commit_root).
7788 *
7789 * To relocate a tree block referenced by a subvol, there are two steps.
7790 * COW the block through subvol's reloc tree, then update block pointer
7791 * in the subvol to point to the new block. Since all reloc trees share
7792 * same root key objectid, doing special handing for tree blocks owned
7793 * by them is easy. Once a tree block has been COWed in one reloc tree,
7794 * we can use the resulting new block directly when the same block is
7795 * required to COW again through other reloc trees. By this way, relocated
7796 * tree blocks are shared between reloc trees, so they are also shared
7797 * between subvols.
Zheng Yan1a40e232008-09-26 10:09:34 -04007798 */
Chris Masond3977122009-01-05 21:25:51 -05007799static noinline int relocate_one_path(struct btrfs_trans_handle *trans,
Zheng Yan1a40e232008-09-26 10:09:34 -04007800 struct btrfs_root *root,
7801 struct btrfs_path *path,
7802 struct btrfs_key *first_key,
7803 struct btrfs_ref_path *ref_path,
7804 struct btrfs_block_group_cache *group,
7805 struct inode *reloc_inode)
7806{
7807 struct btrfs_root *reloc_root;
7808 struct extent_buffer *eb = NULL;
7809 struct btrfs_key *keys;
7810 u64 *nodes;
7811 int level;
Yan Zhengf82d02d2008-10-29 14:49:05 -04007812 int shared_level;
Zheng Yan1a40e232008-09-26 10:09:34 -04007813 int lowest_level = 0;
Zheng Yan1a40e232008-09-26 10:09:34 -04007814 int ret;
7815
7816 if (ref_path->owner_objectid < BTRFS_FIRST_FREE_OBJECTID)
7817 lowest_level = ref_path->owner_objectid;
7818
Yan Zhengf82d02d2008-10-29 14:49:05 -04007819 if (!root->ref_cows) {
Zheng Yan1a40e232008-09-26 10:09:34 -04007820 path->lowest_level = lowest_level;
7821 ret = btrfs_search_slot(trans, root, first_key, path, 0, 1);
7822 BUG_ON(ret < 0);
7823 path->lowest_level = 0;
7824 btrfs_release_path(root, path);
7825 return 0;
7826 }
7827
Zheng Yan1a40e232008-09-26 10:09:34 -04007828 mutex_lock(&root->fs_info->tree_reloc_mutex);
7829 ret = init_reloc_tree(trans, root);
7830 BUG_ON(ret);
7831 reloc_root = root->reloc_root;
7832
Yan Zhengf82d02d2008-10-29 14:49:05 -04007833 shared_level = ref_path->shared_level;
7834 ref_path->shared_level = BTRFS_MAX_LEVEL - 1;
Zheng Yan1a40e232008-09-26 10:09:34 -04007835
Yan Zhengf82d02d2008-10-29 14:49:05 -04007836 keys = ref_path->node_keys;
7837 nodes = ref_path->new_nodes;
7838 memset(&keys[shared_level + 1], 0,
7839 sizeof(*keys) * (BTRFS_MAX_LEVEL - shared_level - 1));
7840 memset(&nodes[shared_level + 1], 0,
7841 sizeof(*nodes) * (BTRFS_MAX_LEVEL - shared_level - 1));
Zheng Yan1a40e232008-09-26 10:09:34 -04007842
Yan Zhengf82d02d2008-10-29 14:49:05 -04007843 if (nodes[lowest_level] == 0) {
7844 path->lowest_level = lowest_level;
7845 ret = btrfs_search_slot(trans, reloc_root, first_key, path,
7846 0, 1);
7847 BUG_ON(ret);
7848 for (level = lowest_level; level < BTRFS_MAX_LEVEL; level++) {
7849 eb = path->nodes[level];
7850 if (!eb || eb == reloc_root->node)
7851 break;
7852 nodes[level] = eb->start;
7853 if (level == 0)
7854 btrfs_item_key_to_cpu(eb, &keys[level], 0);
7855 else
7856 btrfs_node_key_to_cpu(eb, &keys[level], 0);
7857 }
Yan Zheng2b820322008-11-17 21:11:30 -05007858 if (nodes[0] &&
7859 ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
Yan Zhengf82d02d2008-10-29 14:49:05 -04007860 eb = path->nodes[0];
7861 ret = replace_extents_in_leaf(trans, reloc_root, eb,
7862 group, reloc_inode);
7863 BUG_ON(ret);
7864 }
7865 btrfs_release_path(reloc_root, path);
7866 } else {
Zheng Yan1a40e232008-09-26 10:09:34 -04007867 ret = btrfs_merge_path(trans, reloc_root, keys, nodes,
Yan Zhengf82d02d2008-10-29 14:49:05 -04007868 lowest_level);
Zheng Yan1a40e232008-09-26 10:09:34 -04007869 BUG_ON(ret);
7870 }
7871
Zheng Yan1a40e232008-09-26 10:09:34 -04007872 /*
7873 * replace tree blocks in the fs tree with tree blocks in
7874 * the reloc tree.
7875 */
7876 ret = btrfs_merge_path(trans, root, keys, nodes, lowest_level);
7877 BUG_ON(ret < 0);
7878
7879 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
Yan Zhengf82d02d2008-10-29 14:49:05 -04007880 ret = btrfs_search_slot(trans, reloc_root, first_key, path,
7881 0, 0);
7882 BUG_ON(ret);
7883 extent_buffer_get(path->nodes[0]);
7884 eb = path->nodes[0];
7885 btrfs_release_path(reloc_root, path);
Zheng Yan1a40e232008-09-26 10:09:34 -04007886 ret = invalidate_extent_cache(reloc_root, eb, group, root);
7887 BUG_ON(ret);
7888 free_extent_buffer(eb);
7889 }
Zheng Yan1a40e232008-09-26 10:09:34 -04007890
Yan Zhengf82d02d2008-10-29 14:49:05 -04007891 mutex_unlock(&root->fs_info->tree_reloc_mutex);
Zheng Yan1a40e232008-09-26 10:09:34 -04007892 path->lowest_level = 0;
Zheng Yan1a40e232008-09-26 10:09:34 -04007893 return 0;
7894}
7895
Chris Masond3977122009-01-05 21:25:51 -05007896static noinline int relocate_tree_block(struct btrfs_trans_handle *trans,
Zheng Yan1a40e232008-09-26 10:09:34 -04007897 struct btrfs_root *root,
7898 struct btrfs_path *path,
7899 struct btrfs_key *first_key,
7900 struct btrfs_ref_path *ref_path)
7901{
7902 int ret;
Zheng Yan1a40e232008-09-26 10:09:34 -04007903
7904 ret = relocate_one_path(trans, root, path, first_key,
7905 ref_path, NULL, NULL);
7906 BUG_ON(ret);
7907
Zheng Yan1a40e232008-09-26 10:09:34 -04007908 return 0;
7909}
7910
Chris Masond3977122009-01-05 21:25:51 -05007911static noinline int del_extent_zero(struct btrfs_trans_handle *trans,
Zheng Yan1a40e232008-09-26 10:09:34 -04007912 struct btrfs_root *extent_root,
7913 struct btrfs_path *path,
7914 struct btrfs_key *extent_key)
7915{
7916 int ret;
7917
Zheng Yan1a40e232008-09-26 10:09:34 -04007918 ret = btrfs_search_slot(trans, extent_root, extent_key, path, -1, 1);
7919 if (ret)
7920 goto out;
7921 ret = btrfs_del_item(trans, extent_root, path);
7922out:
Chris Masonedbd8d42007-12-21 16:27:24 -05007923 btrfs_release_path(extent_root, path);
Zheng Yan1a40e232008-09-26 10:09:34 -04007924 return ret;
7925}
7926
Chris Masond3977122009-01-05 21:25:51 -05007927static noinline struct btrfs_root *read_ref_root(struct btrfs_fs_info *fs_info,
Zheng Yan1a40e232008-09-26 10:09:34 -04007928 struct btrfs_ref_path *ref_path)
7929{
7930 struct btrfs_key root_key;
7931
7932 root_key.objectid = ref_path->root_objectid;
7933 root_key.type = BTRFS_ROOT_ITEM_KEY;
7934 if (is_cowonly_root(ref_path->root_objectid))
7935 root_key.offset = 0;
7936 else
7937 root_key.offset = (u64)-1;
7938
7939 return btrfs_read_fs_root_no_name(fs_info, &root_key);
7940}
7941
Chris Masond3977122009-01-05 21:25:51 -05007942static noinline int relocate_one_extent(struct btrfs_root *extent_root,
Zheng Yan1a40e232008-09-26 10:09:34 -04007943 struct btrfs_path *path,
7944 struct btrfs_key *extent_key,
7945 struct btrfs_block_group_cache *group,
7946 struct inode *reloc_inode, int pass)
7947{
7948 struct btrfs_trans_handle *trans;
7949 struct btrfs_root *found_root;
7950 struct btrfs_ref_path *ref_path = NULL;
7951 struct disk_extent *new_extents = NULL;
7952 int nr_extents = 0;
7953 int loops;
7954 int ret;
7955 int level;
7956 struct btrfs_key first_key;
7957 u64 prev_block = 0;
7958
Zheng Yan1a40e232008-09-26 10:09:34 -04007959
7960 trans = btrfs_start_transaction(extent_root, 1);
Tsutomu Itoh98d5dc12011-01-20 06:19:37 +00007961 BUG_ON(IS_ERR(trans));
Zheng Yan1a40e232008-09-26 10:09:34 -04007962
7963 if (extent_key->objectid == 0) {
7964 ret = del_extent_zero(trans, extent_root, path, extent_key);
7965 goto out;
7966 }
7967
7968 ref_path = kmalloc(sizeof(*ref_path), GFP_NOFS);
7969 if (!ref_path) {
Chris Masond3977122009-01-05 21:25:51 -05007970 ret = -ENOMEM;
7971 goto out;
Zheng Yan1a40e232008-09-26 10:09:34 -04007972 }
7973
7974 for (loops = 0; ; loops++) {
7975 if (loops == 0) {
7976 ret = btrfs_first_ref_path(trans, extent_root, ref_path,
7977 extent_key->objectid);
7978 } else {
7979 ret = btrfs_next_ref_path(trans, extent_root, ref_path);
7980 }
7981 if (ret < 0)
7982 goto out;
7983 if (ret > 0)
7984 break;
7985
7986 if (ref_path->root_objectid == BTRFS_TREE_LOG_OBJECTID ||
7987 ref_path->root_objectid == BTRFS_TREE_RELOC_OBJECTID)
7988 continue;
7989
7990 found_root = read_ref_root(extent_root->fs_info, ref_path);
7991 BUG_ON(!found_root);
7992 /*
7993 * for reference counted tree, only process reference paths
7994 * rooted at the latest committed root.
7995 */
7996 if (found_root->ref_cows &&
7997 ref_path->root_generation != found_root->root_key.offset)
7998 continue;
7999
8000 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
8001 if (pass == 0) {
8002 /*
8003 * copy data extents to new locations
8004 */
8005 u64 group_start = group->key.objectid;
8006 ret = relocate_data_extent(reloc_inode,
8007 extent_key,
8008 group_start);
8009 if (ret < 0)
8010 goto out;
8011 break;
8012 }
8013 level = 0;
8014 } else {
8015 level = ref_path->owner_objectid;
8016 }
8017
8018 if (prev_block != ref_path->nodes[level]) {
8019 struct extent_buffer *eb;
8020 u64 block_start = ref_path->nodes[level];
8021 u64 block_size = btrfs_level_size(found_root, level);
8022
8023 eb = read_tree_block(found_root, block_start,
8024 block_size, 0);
Tsutomu Itoh97d9a8a2011-03-24 06:33:21 +00008025 if (!eb) {
8026 ret = -EIO;
8027 goto out;
8028 }
Zheng Yan1a40e232008-09-26 10:09:34 -04008029 btrfs_tree_lock(eb);
8030 BUG_ON(level != btrfs_header_level(eb));
8031
8032 if (level == 0)
8033 btrfs_item_key_to_cpu(eb, &first_key, 0);
8034 else
8035 btrfs_node_key_to_cpu(eb, &first_key, 0);
8036
8037 btrfs_tree_unlock(eb);
8038 free_extent_buffer(eb);
8039 prev_block = block_start;
8040 }
8041
Yan Zheng24562422009-02-12 14:14:53 -05008042 mutex_lock(&extent_root->fs_info->trans_mutex);
Yan Zhenge4404d62008-12-12 10:03:26 -05008043 btrfs_record_root_in_trans(found_root);
Yan Zheng24562422009-02-12 14:14:53 -05008044 mutex_unlock(&extent_root->fs_info->trans_mutex);
Yan Zhenge4404d62008-12-12 10:03:26 -05008045 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
8046 /*
8047 * try to update data extent references while
8048 * keeping metadata shared between snapshots.
8049 */
8050 if (pass == 1) {
8051 ret = relocate_one_path(trans, found_root,
8052 path, &first_key, ref_path,
8053 group, reloc_inode);
8054 if (ret < 0)
8055 goto out;
8056 continue;
8057 }
Zheng Yan1a40e232008-09-26 10:09:34 -04008058 /*
8059 * use fallback method to process the remaining
8060 * references.
8061 */
8062 if (!new_extents) {
8063 u64 group_start = group->key.objectid;
Yan Zhengd899e052008-10-30 14:25:28 -04008064 new_extents = kmalloc(sizeof(*new_extents),
8065 GFP_NOFS);
Tsutomu Itoh8d413712011-04-25 19:43:52 -04008066 if (!new_extents) {
8067 ret = -ENOMEM;
8068 goto out;
8069 }
Yan Zhengd899e052008-10-30 14:25:28 -04008070 nr_extents = 1;
Zheng Yan1a40e232008-09-26 10:09:34 -04008071 ret = get_new_locations(reloc_inode,
8072 extent_key,
Yan Zhengd899e052008-10-30 14:25:28 -04008073 group_start, 1,
Zheng Yan1a40e232008-09-26 10:09:34 -04008074 &new_extents,
8075 &nr_extents);
Yan Zhengd899e052008-10-30 14:25:28 -04008076 if (ret)
Zheng Yan1a40e232008-09-26 10:09:34 -04008077 goto out;
8078 }
Zheng Yan1a40e232008-09-26 10:09:34 -04008079 ret = replace_one_extent(trans, found_root,
8080 path, extent_key,
8081 &first_key, ref_path,
8082 new_extents, nr_extents);
Yan Zhenge4404d62008-12-12 10:03:26 -05008083 } else {
Zheng Yan1a40e232008-09-26 10:09:34 -04008084 ret = relocate_tree_block(trans, found_root, path,
8085 &first_key, ref_path);
Zheng Yan1a40e232008-09-26 10:09:34 -04008086 }
8087 if (ret < 0)
8088 goto out;
8089 }
8090 ret = 0;
8091out:
8092 btrfs_end_transaction(trans, extent_root);
8093 kfree(new_extents);
8094 kfree(ref_path);
Chris Masonedbd8d42007-12-21 16:27:24 -05008095 return ret;
8096}
Yan Zheng5d4f98a2009-06-10 10:45:14 -04008097#endif
Chris Masonedbd8d42007-12-21 16:27:24 -05008098
Chris Masonec44a352008-04-28 15:29:52 -04008099static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
8100{
8101 u64 num_devices;
8102 u64 stripped = BTRFS_BLOCK_GROUP_RAID0 |
8103 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
8104
Chris Masoncd02dca2010-12-13 14:56:23 -05008105 /*
8106 * we add in the count of missing devices because we want
8107 * to make sure that any RAID levels on a degraded FS
8108 * continue to be honored.
8109 */
8110 num_devices = root->fs_info->fs_devices->rw_devices +
8111 root->fs_info->fs_devices->missing_devices;
8112
Chris Masonec44a352008-04-28 15:29:52 -04008113 if (num_devices == 1) {
8114 stripped |= BTRFS_BLOCK_GROUP_DUP;
8115 stripped = flags & ~stripped;
8116
8117 /* turn raid0 into single device chunks */
8118 if (flags & BTRFS_BLOCK_GROUP_RAID0)
8119 return stripped;
8120
8121 /* turn mirroring into duplication */
8122 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
8123 BTRFS_BLOCK_GROUP_RAID10))
8124 return stripped | BTRFS_BLOCK_GROUP_DUP;
8125 return flags;
8126 } else {
8127 /* they already had raid on here, just return */
Chris Masonec44a352008-04-28 15:29:52 -04008128 if (flags & stripped)
8129 return flags;
8130
8131 stripped |= BTRFS_BLOCK_GROUP_DUP;
8132 stripped = flags & ~stripped;
8133
8134 /* switch duplicated blocks with raid1 */
8135 if (flags & BTRFS_BLOCK_GROUP_DUP)
8136 return stripped | BTRFS_BLOCK_GROUP_RAID1;
8137
8138 /* turn single device chunks into raid0 */
8139 return stripped | BTRFS_BLOCK_GROUP_RAID0;
8140 }
8141 return flags;
8142}
8143
Yan, Zhengf0486c62010-05-16 10:46:25 -04008144static int set_block_group_ro(struct btrfs_block_group_cache *cache)
Chris Mason0ef3e662008-05-24 14:04:53 -04008145{
Yan, Zhengf0486c62010-05-16 10:46:25 -04008146 struct btrfs_space_info *sinfo = cache->space_info;
8147 u64 num_bytes;
8148 int ret = -ENOSPC;
Chris Mason0ef3e662008-05-24 14:04:53 -04008149
Yan, Zhengf0486c62010-05-16 10:46:25 -04008150 if (cache->ro)
8151 return 0;
Chris Masonc286ac42008-07-22 23:06:41 -04008152
Yan, Zhengf0486c62010-05-16 10:46:25 -04008153 spin_lock(&sinfo->lock);
8154 spin_lock(&cache->lock);
8155 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
8156 cache->bytes_super - btrfs_block_group_used(&cache->item);
Chris Mason7d9eb122008-07-08 14:19:17 -04008157
Yan, Zhengf0486c62010-05-16 10:46:25 -04008158 if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned +
8159 sinfo->bytes_may_use + sinfo->bytes_readonly +
Chris Mason65e53412010-12-24 06:41:52 -05008160 cache->reserved_pinned + num_bytes <= sinfo->total_bytes) {
Yan, Zhengf0486c62010-05-16 10:46:25 -04008161 sinfo->bytes_readonly += num_bytes;
8162 sinfo->bytes_reserved += cache->reserved_pinned;
8163 cache->reserved_pinned = 0;
8164 cache->ro = 1;
8165 ret = 0;
8166 }
Chris Mason65e53412010-12-24 06:41:52 -05008167
Yan, Zhengf0486c62010-05-16 10:46:25 -04008168 spin_unlock(&cache->lock);
8169 spin_unlock(&sinfo->lock);
8170 return ret;
Chris Mason0ef3e662008-05-24 14:04:53 -04008171}
8172
Yan, Zhengf0486c62010-05-16 10:46:25 -04008173int btrfs_set_block_group_ro(struct btrfs_root *root,
8174 struct btrfs_block_group_cache *cache)
Yan Zheng5d4f98a2009-06-10 10:45:14 -04008175
8176{
Yan, Zhengf0486c62010-05-16 10:46:25 -04008177 struct btrfs_trans_handle *trans;
8178 u64 alloc_flags;
8179 int ret;
8180
8181 BUG_ON(cache->ro);
8182
8183 trans = btrfs_join_transaction(root, 1);
8184 BUG_ON(IS_ERR(trans));
8185
8186 alloc_flags = update_block_group_flags(root, cache->flags);
8187 if (alloc_flags != cache->flags)
Chris Mason0e4f8f82011-04-15 16:05:44 -04008188 do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags,
8189 CHUNK_ALLOC_FORCE);
Yan, Zhengf0486c62010-05-16 10:46:25 -04008190
8191 ret = set_block_group_ro(cache);
8192 if (!ret)
8193 goto out;
8194 alloc_flags = get_alloc_profile(root, cache->space_info->flags);
Chris Mason0e4f8f82011-04-15 16:05:44 -04008195 ret = do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags,
8196 CHUNK_ALLOC_FORCE);
Yan, Zhengf0486c62010-05-16 10:46:25 -04008197 if (ret < 0)
8198 goto out;
8199 ret = set_block_group_ro(cache);
8200out:
8201 btrfs_end_transaction(trans, root);
8202 return ret;
8203}
8204
Chris Masonc87f08c2011-02-16 13:57:04 -05008205int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans,
8206 struct btrfs_root *root, u64 type)
8207{
8208 u64 alloc_flags = get_alloc_profile(root, type);
Chris Mason0e4f8f82011-04-15 16:05:44 -04008209 return do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags,
8210 CHUNK_ALLOC_FORCE);
Chris Masonc87f08c2011-02-16 13:57:04 -05008211}
8212
Miao Xie6d07bce2011-01-05 10:07:31 +00008213/*
8214 * helper to account the unused space of all the readonly block group in the
8215 * list. takes mirrors into account.
8216 */
8217static u64 __btrfs_get_ro_block_group_free_space(struct list_head *groups_list)
8218{
8219 struct btrfs_block_group_cache *block_group;
8220 u64 free_bytes = 0;
8221 int factor;
8222
8223 list_for_each_entry(block_group, groups_list, list) {
8224 spin_lock(&block_group->lock);
8225
8226 if (!block_group->ro) {
8227 spin_unlock(&block_group->lock);
8228 continue;
8229 }
8230
8231 if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 |
8232 BTRFS_BLOCK_GROUP_RAID10 |
8233 BTRFS_BLOCK_GROUP_DUP))
8234 factor = 2;
8235 else
8236 factor = 1;
8237
8238 free_bytes += (block_group->key.offset -
8239 btrfs_block_group_used(&block_group->item)) *
8240 factor;
8241
8242 spin_unlock(&block_group->lock);
8243 }
8244
8245 return free_bytes;
8246}
8247
8248/*
8249 * helper to account the unused space of all the readonly block group in the
8250 * space_info. takes mirrors into account.
8251 */
8252u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
8253{
8254 int i;
8255 u64 free_bytes = 0;
8256
8257 spin_lock(&sinfo->lock);
8258
8259 for(i = 0; i < BTRFS_NR_RAID_TYPES; i++)
8260 if (!list_empty(&sinfo->block_groups[i]))
8261 free_bytes += __btrfs_get_ro_block_group_free_space(
8262 &sinfo->block_groups[i]);
8263
8264 spin_unlock(&sinfo->lock);
8265
8266 return free_bytes;
8267}
8268
Yan, Zhengf0486c62010-05-16 10:46:25 -04008269int btrfs_set_block_group_rw(struct btrfs_root *root,
8270 struct btrfs_block_group_cache *cache)
8271{
8272 struct btrfs_space_info *sinfo = cache->space_info;
8273 u64 num_bytes;
8274
8275 BUG_ON(!cache->ro);
8276
8277 spin_lock(&sinfo->lock);
8278 spin_lock(&cache->lock);
8279 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
8280 cache->bytes_super - btrfs_block_group_used(&cache->item);
8281 sinfo->bytes_readonly -= num_bytes;
8282 cache->ro = 0;
8283 spin_unlock(&cache->lock);
8284 spin_unlock(&sinfo->lock);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04008285 return 0;
8286}
8287
Josef Bacikba1bf482009-09-11 16:11:19 -04008288/*
8289 * checks to see if its even possible to relocate this block group.
8290 *
8291 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
8292 * ok to go ahead and try.
8293 */
8294int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr)
Zheng Yan1a40e232008-09-26 10:09:34 -04008295{
Zheng Yan1a40e232008-09-26 10:09:34 -04008296 struct btrfs_block_group_cache *block_group;
Josef Bacikba1bf482009-09-11 16:11:19 -04008297 struct btrfs_space_info *space_info;
8298 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
8299 struct btrfs_device *device;
8300 int full = 0;
8301 int ret = 0;
Chris Masonedbd8d42007-12-21 16:27:24 -05008302
Josef Bacikba1bf482009-09-11 16:11:19 -04008303 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
Zheng Yan1a40e232008-09-26 10:09:34 -04008304
Josef Bacikba1bf482009-09-11 16:11:19 -04008305 /* odd, couldn't find the block group, leave it alone */
8306 if (!block_group)
8307 return -1;
Chris Masonedbd8d42007-12-21 16:27:24 -05008308
Josef Bacikba1bf482009-09-11 16:11:19 -04008309 /* no bytes used, we're good */
8310 if (!btrfs_block_group_used(&block_group->item))
8311 goto out;
Chris Mason323da792008-05-09 11:46:48 -04008312
Josef Bacikba1bf482009-09-11 16:11:19 -04008313 space_info = block_group->space_info;
8314 spin_lock(&space_info->lock);
Chris Mason323da792008-05-09 11:46:48 -04008315
Josef Bacikba1bf482009-09-11 16:11:19 -04008316 full = space_info->full;
Zheng Yan1a40e232008-09-26 10:09:34 -04008317
Josef Bacikba1bf482009-09-11 16:11:19 -04008318 /*
8319 * if this is the last block group we have in this space, we can't
Chris Mason7ce618d2009-09-22 14:48:44 -04008320 * relocate it unless we're able to allocate a new chunk below.
8321 *
8322 * Otherwise, we need to make sure we have room in the space to handle
8323 * all of the extents from this block group. If we can, we're good
Josef Bacikba1bf482009-09-11 16:11:19 -04008324 */
Chris Mason7ce618d2009-09-22 14:48:44 -04008325 if ((space_info->total_bytes != block_group->key.offset) &&
8326 (space_info->bytes_used + space_info->bytes_reserved +
Josef Bacikba1bf482009-09-11 16:11:19 -04008327 space_info->bytes_pinned + space_info->bytes_readonly +
8328 btrfs_block_group_used(&block_group->item) <
Chris Mason7ce618d2009-09-22 14:48:44 -04008329 space_info->total_bytes)) {
Josef Bacikba1bf482009-09-11 16:11:19 -04008330 spin_unlock(&space_info->lock);
8331 goto out;
8332 }
8333 spin_unlock(&space_info->lock);
Zheng Yan1a40e232008-09-26 10:09:34 -04008334
Josef Bacikba1bf482009-09-11 16:11:19 -04008335 /*
8336 * ok we don't have enough space, but maybe we have free space on our
8337 * devices to allocate new chunks for relocation, so loop through our
8338 * alloc devices and guess if we have enough space. However, if we
8339 * were marked as full, then we know there aren't enough chunks, and we
8340 * can just return.
8341 */
8342 ret = -1;
8343 if (full)
8344 goto out;
Chris Mason4313b392008-01-03 09:08:48 -05008345
Josef Bacikba1bf482009-09-11 16:11:19 -04008346 mutex_lock(&root->fs_info->chunk_mutex);
8347 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
8348 u64 min_free = btrfs_block_group_used(&block_group->item);
Miao Xie7bfc8372011-01-05 10:07:26 +00008349 u64 dev_offset;
Chris Masonea8c2812008-08-04 23:17:27 -04008350
Josef Bacikba1bf482009-09-11 16:11:19 -04008351 /*
8352 * check to make sure we can actually find a chunk with enough
8353 * space to fit our block group in.
8354 */
8355 if (device->total_bytes > device->bytes_used + min_free) {
8356 ret = find_free_dev_extent(NULL, device, min_free,
Miao Xie7bfc8372011-01-05 10:07:26 +00008357 &dev_offset, NULL);
Josef Bacikba1bf482009-09-11 16:11:19 -04008358 if (!ret)
Yan73e48b22008-01-03 14:14:39 -05008359 break;
Josef Bacikba1bf482009-09-11 16:11:19 -04008360 ret = -1;
Yan73e48b22008-01-03 14:14:39 -05008361 }
Chris Masonedbd8d42007-12-21 16:27:24 -05008362 }
Josef Bacikba1bf482009-09-11 16:11:19 -04008363 mutex_unlock(&root->fs_info->chunk_mutex);
Chris Masonedbd8d42007-12-21 16:27:24 -05008364out:
Josef Bacikba1bf482009-09-11 16:11:19 -04008365 btrfs_put_block_group(block_group);
Chris Masonedbd8d42007-12-21 16:27:24 -05008366 return ret;
8367}
8368
Christoph Hellwigb2950862008-12-02 09:54:17 -05008369static int find_first_block_group(struct btrfs_root *root,
8370 struct btrfs_path *path, struct btrfs_key *key)
Chris Mason0b86a832008-03-24 15:01:56 -04008371{
Chris Mason925baed2008-06-25 16:01:30 -04008372 int ret = 0;
Chris Mason0b86a832008-03-24 15:01:56 -04008373 struct btrfs_key found_key;
8374 struct extent_buffer *leaf;
8375 int slot;
8376
8377 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
8378 if (ret < 0)
Chris Mason925baed2008-06-25 16:01:30 -04008379 goto out;
8380
Chris Masond3977122009-01-05 21:25:51 -05008381 while (1) {
Chris Mason0b86a832008-03-24 15:01:56 -04008382 slot = path->slots[0];
8383 leaf = path->nodes[0];
8384 if (slot >= btrfs_header_nritems(leaf)) {
8385 ret = btrfs_next_leaf(root, path);
8386 if (ret == 0)
8387 continue;
8388 if (ret < 0)
Chris Mason925baed2008-06-25 16:01:30 -04008389 goto out;
Chris Mason0b86a832008-03-24 15:01:56 -04008390 break;
8391 }
8392 btrfs_item_key_to_cpu(leaf, &found_key, slot);
8393
8394 if (found_key.objectid >= key->objectid &&
Chris Mason925baed2008-06-25 16:01:30 -04008395 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
8396 ret = 0;
8397 goto out;
8398 }
Chris Mason0b86a832008-03-24 15:01:56 -04008399 path->slots[0]++;
8400 }
Chris Mason925baed2008-06-25 16:01:30 -04008401out:
Chris Mason0b86a832008-03-24 15:01:56 -04008402 return ret;
8403}
8404
Josef Bacik0af3d002010-06-21 14:48:16 -04008405void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
8406{
8407 struct btrfs_block_group_cache *block_group;
8408 u64 last = 0;
8409
8410 while (1) {
8411 struct inode *inode;
8412
8413 block_group = btrfs_lookup_first_block_group(info, last);
8414 while (block_group) {
8415 spin_lock(&block_group->lock);
8416 if (block_group->iref)
8417 break;
8418 spin_unlock(&block_group->lock);
8419 block_group = next_block_group(info->tree_root,
8420 block_group);
8421 }
8422 if (!block_group) {
8423 if (last == 0)
8424 break;
8425 last = 0;
8426 continue;
8427 }
8428
8429 inode = block_group->inode;
8430 block_group->iref = 0;
8431 block_group->inode = NULL;
8432 spin_unlock(&block_group->lock);
8433 iput(inode);
8434 last = block_group->key.objectid + block_group->key.offset;
8435 btrfs_put_block_group(block_group);
8436 }
8437}
8438
Zheng Yan1a40e232008-09-26 10:09:34 -04008439int btrfs_free_block_groups(struct btrfs_fs_info *info)
8440{
8441 struct btrfs_block_group_cache *block_group;
Chris Mason4184ea72009-03-10 12:39:20 -04008442 struct btrfs_space_info *space_info;
Yan Zheng11833d62009-09-11 16:11:19 -04008443 struct btrfs_caching_control *caching_ctl;
Zheng Yan1a40e232008-09-26 10:09:34 -04008444 struct rb_node *n;
8445
Yan Zheng11833d62009-09-11 16:11:19 -04008446 down_write(&info->extent_commit_sem);
8447 while (!list_empty(&info->caching_block_groups)) {
8448 caching_ctl = list_entry(info->caching_block_groups.next,
8449 struct btrfs_caching_control, list);
8450 list_del(&caching_ctl->list);
8451 put_caching_control(caching_ctl);
8452 }
8453 up_write(&info->extent_commit_sem);
8454
Zheng Yan1a40e232008-09-26 10:09:34 -04008455 spin_lock(&info->block_group_cache_lock);
8456 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
8457 block_group = rb_entry(n, struct btrfs_block_group_cache,
8458 cache_node);
Zheng Yan1a40e232008-09-26 10:09:34 -04008459 rb_erase(&block_group->cache_node,
8460 &info->block_group_cache_tree);
Yan Zhengd899e052008-10-30 14:25:28 -04008461 spin_unlock(&info->block_group_cache_lock);
8462
Josef Bacik80eb2342008-10-29 14:49:05 -04008463 down_write(&block_group->space_info->groups_sem);
Zheng Yan1a40e232008-09-26 10:09:34 -04008464 list_del(&block_group->list);
Josef Bacik80eb2342008-10-29 14:49:05 -04008465 up_write(&block_group->space_info->groups_sem);
Yan Zhengd2fb3432008-12-11 16:30:39 -05008466
Josef Bacik817d52f2009-07-13 21:29:25 -04008467 if (block_group->cached == BTRFS_CACHE_STARTED)
Yan Zheng11833d62009-09-11 16:11:19 -04008468 wait_block_group_cache_done(block_group);
Josef Bacik817d52f2009-07-13 21:29:25 -04008469
Josef Bacik3c148742011-02-02 15:53:47 +00008470 /*
8471 * We haven't cached this block group, which means we could
8472 * possibly have excluded extents on this block group.
8473 */
8474 if (block_group->cached == BTRFS_CACHE_NO)
8475 free_excluded_extents(info->extent_root, block_group);
8476
Josef Bacik817d52f2009-07-13 21:29:25 -04008477 btrfs_remove_free_space_cache(block_group);
Josef Bacik11dfe352009-11-13 20:12:59 +00008478 btrfs_put_block_group(block_group);
Yan Zhengd899e052008-10-30 14:25:28 -04008479
8480 spin_lock(&info->block_group_cache_lock);
Zheng Yan1a40e232008-09-26 10:09:34 -04008481 }
8482 spin_unlock(&info->block_group_cache_lock);
Chris Mason4184ea72009-03-10 12:39:20 -04008483
8484 /* now that all the block groups are freed, go through and
8485 * free all the space_info structs. This is only called during
8486 * the final stages of unmount, and so we know nobody is
8487 * using them. We call synchronize_rcu() once before we start,
8488 * just to be on the safe side.
8489 */
8490 synchronize_rcu();
8491
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04008492 release_global_block_rsv(info);
8493
Chris Mason4184ea72009-03-10 12:39:20 -04008494 while(!list_empty(&info->space_info)) {
8495 space_info = list_entry(info->space_info.next,
8496 struct btrfs_space_info,
8497 list);
Yan, Zhengf0486c62010-05-16 10:46:25 -04008498 if (space_info->bytes_pinned > 0 ||
8499 space_info->bytes_reserved > 0) {
8500 WARN_ON(1);
8501 dump_space_info(space_info, 0, 0);
8502 }
Chris Mason4184ea72009-03-10 12:39:20 -04008503 list_del(&space_info->list);
8504 kfree(space_info);
8505 }
Zheng Yan1a40e232008-09-26 10:09:34 -04008506 return 0;
8507}
8508
Yan, Zhengb742bb82010-05-16 10:46:24 -04008509static void __link_block_group(struct btrfs_space_info *space_info,
8510 struct btrfs_block_group_cache *cache)
8511{
8512 int index = get_block_group_index(cache);
8513
8514 down_write(&space_info->groups_sem);
8515 list_add_tail(&cache->list, &space_info->block_groups[index]);
8516 up_write(&space_info->groups_sem);
8517}
8518
Chris Mason9078a3e2007-04-26 16:46:15 -04008519int btrfs_read_block_groups(struct btrfs_root *root)
8520{
8521 struct btrfs_path *path;
8522 int ret;
Chris Mason9078a3e2007-04-26 16:46:15 -04008523 struct btrfs_block_group_cache *cache;
Chris Masonbe744172007-05-06 10:15:01 -04008524 struct btrfs_fs_info *info = root->fs_info;
Chris Mason6324fbf2008-03-24 15:01:59 -04008525 struct btrfs_space_info *space_info;
Chris Mason9078a3e2007-04-26 16:46:15 -04008526 struct btrfs_key key;
8527 struct btrfs_key found_key;
Chris Mason5f39d392007-10-15 16:14:19 -04008528 struct extent_buffer *leaf;
Josef Bacik0af3d002010-06-21 14:48:16 -04008529 int need_clear = 0;
8530 u64 cache_gen;
Chris Mason96b51792007-10-15 16:15:19 -04008531
Chris Masonbe744172007-05-06 10:15:01 -04008532 root = info->extent_root;
Chris Mason9078a3e2007-04-26 16:46:15 -04008533 key.objectid = 0;
Chris Mason0b86a832008-03-24 15:01:56 -04008534 key.offset = 0;
Chris Mason9078a3e2007-04-26 16:46:15 -04008535 btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
Chris Mason9078a3e2007-04-26 16:46:15 -04008536 path = btrfs_alloc_path();
8537 if (!path)
8538 return -ENOMEM;
8539
Josef Bacik0af3d002010-06-21 14:48:16 -04008540 cache_gen = btrfs_super_cache_generation(&root->fs_info->super_copy);
8541 if (cache_gen != 0 &&
8542 btrfs_super_generation(&root->fs_info->super_copy) != cache_gen)
8543 need_clear = 1;
Josef Bacik88c2ba32010-09-21 14:21:34 -04008544 if (btrfs_test_opt(root, CLEAR_CACHE))
8545 need_clear = 1;
Josef Bacik8216ef82010-10-28 16:55:47 -04008546 if (!btrfs_test_opt(root, SPACE_CACHE) && cache_gen)
8547 printk(KERN_INFO "btrfs: disk space caching is enabled\n");
Josef Bacik0af3d002010-06-21 14:48:16 -04008548
Chris Masond3977122009-01-05 21:25:51 -05008549 while (1) {
Chris Mason0b86a832008-03-24 15:01:56 -04008550 ret = find_first_block_group(root, path, &key);
Yan, Zhengb742bb82010-05-16 10:46:24 -04008551 if (ret > 0)
8552 break;
Chris Mason0b86a832008-03-24 15:01:56 -04008553 if (ret != 0)
8554 goto error;
Chris Mason5f39d392007-10-15 16:14:19 -04008555 leaf = path->nodes[0];
8556 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
Chris Mason8f18cf12008-04-25 16:53:30 -04008557 cache = kzalloc(sizeof(*cache), GFP_NOFS);
Chris Mason9078a3e2007-04-26 16:46:15 -04008558 if (!cache) {
Chris Mason0b86a832008-03-24 15:01:56 -04008559 ret = -ENOMEM;
Yan, Zhengf0486c62010-05-16 10:46:25 -04008560 goto error;
Chris Mason9078a3e2007-04-26 16:46:15 -04008561 }
Chris Mason3e1ad542007-05-07 20:03:49 -04008562
Yan Zhengd2fb3432008-12-11 16:30:39 -05008563 atomic_set(&cache->count, 1);
Chris Masonc286ac42008-07-22 23:06:41 -04008564 spin_lock_init(&cache->lock);
Josef Bacik6226cb02009-04-03 10:14:18 -04008565 spin_lock_init(&cache->tree_lock);
Josef Bacik817d52f2009-07-13 21:29:25 -04008566 cache->fs_info = info;
Josef Bacik0f9dd462008-09-23 13:14:11 -04008567 INIT_LIST_HEAD(&cache->list);
Chris Masonfa9c0d72009-04-03 09:47:43 -04008568 INIT_LIST_HEAD(&cache->cluster_list);
Josef Bacik96303082009-07-13 21:29:25 -04008569
Josef Bacik0af3d002010-06-21 14:48:16 -04008570 if (need_clear)
8571 cache->disk_cache_state = BTRFS_DC_CLEAR;
8572
Josef Bacik96303082009-07-13 21:29:25 -04008573 /*
8574 * we only want to have 32k of ram per block group for keeping
8575 * track of free space, and if we pass 1/2 of that we want to
8576 * start converting things over to using bitmaps
8577 */
8578 cache->extents_thresh = ((1024 * 32) / 2) /
8579 sizeof(struct btrfs_free_space);
8580
Chris Mason5f39d392007-10-15 16:14:19 -04008581 read_extent_buffer(leaf, &cache->item,
8582 btrfs_item_ptr_offset(leaf, path->slots[0]),
8583 sizeof(cache->item));
Chris Mason9078a3e2007-04-26 16:46:15 -04008584 memcpy(&cache->key, &found_key, sizeof(found_key));
Chris Mason0b86a832008-03-24 15:01:56 -04008585
Chris Mason9078a3e2007-04-26 16:46:15 -04008586 key.objectid = found_key.objectid + found_key.offset;
8587 btrfs_release_path(root, path);
Chris Mason0b86a832008-03-24 15:01:56 -04008588 cache->flags = btrfs_block_group_flags(&cache->item);
Josef Bacik817d52f2009-07-13 21:29:25 -04008589 cache->sectorsize = root->sectorsize;
8590
Josef Bacik817d52f2009-07-13 21:29:25 -04008591 /*
Josef Bacik3c148742011-02-02 15:53:47 +00008592 * We need to exclude the super stripes now so that the space
8593 * info has super bytes accounted for, otherwise we'll think
8594 * we have more space than we actually do.
8595 */
8596 exclude_super_stripes(root, cache);
8597
8598 /*
Josef Bacik817d52f2009-07-13 21:29:25 -04008599 * check for two cases, either we are full, and therefore
8600 * don't need to bother with the caching work since we won't
8601 * find any space, or we are empty, and we can just add all
8602 * the space in and be done with it. This saves us _alot_ of
8603 * time, particularly in the full case.
8604 */
8605 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
Yan Zheng11833d62009-09-11 16:11:19 -04008606 cache->last_byte_to_unpin = (u64)-1;
Josef Bacik817d52f2009-07-13 21:29:25 -04008607 cache->cached = BTRFS_CACHE_FINISHED;
Josef Bacik1b2da372009-09-11 16:11:20 -04008608 free_excluded_extents(root, cache);
Josef Bacik817d52f2009-07-13 21:29:25 -04008609 } else if (btrfs_block_group_used(&cache->item) == 0) {
Yan Zheng11833d62009-09-11 16:11:19 -04008610 cache->last_byte_to_unpin = (u64)-1;
Josef Bacik817d52f2009-07-13 21:29:25 -04008611 cache->cached = BTRFS_CACHE_FINISHED;
8612 add_new_free_space(cache, root->fs_info,
8613 found_key.objectid,
8614 found_key.objectid +
8615 found_key.offset);
Yan Zheng11833d62009-09-11 16:11:19 -04008616 free_excluded_extents(root, cache);
Josef Bacik817d52f2009-07-13 21:29:25 -04008617 }
Chris Mason96b51792007-10-15 16:15:19 -04008618
Chris Mason6324fbf2008-03-24 15:01:59 -04008619 ret = update_space_info(info, cache->flags, found_key.offset,
8620 btrfs_block_group_used(&cache->item),
8621 &space_info);
8622 BUG_ON(ret);
8623 cache->space_info = space_info;
Josef Bacik1b2da372009-09-11 16:11:20 -04008624 spin_lock(&cache->space_info->lock);
Yan, Zhengf0486c62010-05-16 10:46:25 -04008625 cache->space_info->bytes_readonly += cache->bytes_super;
Josef Bacik1b2da372009-09-11 16:11:20 -04008626 spin_unlock(&cache->space_info->lock);
8627
Yan, Zhengb742bb82010-05-16 10:46:24 -04008628 __link_block_group(space_info, cache);
Chris Mason6324fbf2008-03-24 15:01:59 -04008629
Josef Bacik0f9dd462008-09-23 13:14:11 -04008630 ret = btrfs_add_block_group_cache(root->fs_info, cache);
8631 BUG_ON(ret);
Chris Mason75ccf472008-09-30 19:24:06 -04008632
8633 set_avail_alloc_bits(root->fs_info, cache->flags);
Yan Zheng2b820322008-11-17 21:11:30 -05008634 if (btrfs_chunk_readonly(root, cache->key.objectid))
Yan, Zhengf0486c62010-05-16 10:46:25 -04008635 set_block_group_ro(cache);
Chris Mason9078a3e2007-04-26 16:46:15 -04008636 }
Yan, Zhengb742bb82010-05-16 10:46:24 -04008637
8638 list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) {
8639 if (!(get_alloc_profile(root, space_info->flags) &
8640 (BTRFS_BLOCK_GROUP_RAID10 |
8641 BTRFS_BLOCK_GROUP_RAID1 |
8642 BTRFS_BLOCK_GROUP_DUP)))
8643 continue;
8644 /*
8645 * avoid allocating from un-mirrored block group if there are
8646 * mirrored block groups.
8647 */
8648 list_for_each_entry(cache, &space_info->block_groups[3], list)
Yan, Zhengf0486c62010-05-16 10:46:25 -04008649 set_block_group_ro(cache);
Yan, Zhengb742bb82010-05-16 10:46:24 -04008650 list_for_each_entry(cache, &space_info->block_groups[4], list)
Yan, Zhengf0486c62010-05-16 10:46:25 -04008651 set_block_group_ro(cache);
Yan, Zhengb742bb82010-05-16 10:46:24 -04008652 }
Yan, Zhengf0486c62010-05-16 10:46:25 -04008653
8654 init_global_block_rsv(info);
Chris Mason0b86a832008-03-24 15:01:56 -04008655 ret = 0;
8656error:
Chris Mason9078a3e2007-04-26 16:46:15 -04008657 btrfs_free_path(path);
Chris Mason0b86a832008-03-24 15:01:56 -04008658 return ret;
Chris Mason9078a3e2007-04-26 16:46:15 -04008659}
Chris Mason6324fbf2008-03-24 15:01:59 -04008660
8661int btrfs_make_block_group(struct btrfs_trans_handle *trans,
8662 struct btrfs_root *root, u64 bytes_used,
Chris Masone17cade2008-04-15 15:41:47 -04008663 u64 type, u64 chunk_objectid, u64 chunk_offset,
Chris Mason6324fbf2008-03-24 15:01:59 -04008664 u64 size)
8665{
8666 int ret;
Chris Mason6324fbf2008-03-24 15:01:59 -04008667 struct btrfs_root *extent_root;
8668 struct btrfs_block_group_cache *cache;
Chris Mason6324fbf2008-03-24 15:01:59 -04008669
8670 extent_root = root->fs_info->extent_root;
Chris Mason6324fbf2008-03-24 15:01:59 -04008671
Chris Mason12fcfd22009-03-24 10:24:20 -04008672 root->fs_info->last_trans_log_full_commit = trans->transid;
Chris Masone02119d2008-09-05 16:13:11 -04008673
Chris Mason8f18cf12008-04-25 16:53:30 -04008674 cache = kzalloc(sizeof(*cache), GFP_NOFS);
Josef Bacik0f9dd462008-09-23 13:14:11 -04008675 if (!cache)
8676 return -ENOMEM;
8677
Chris Masone17cade2008-04-15 15:41:47 -04008678 cache->key.objectid = chunk_offset;
Chris Mason6324fbf2008-03-24 15:01:59 -04008679 cache->key.offset = size;
Yan Zhengd2fb3432008-12-11 16:30:39 -05008680 cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
Josef Bacik96303082009-07-13 21:29:25 -04008681 cache->sectorsize = root->sectorsize;
Josef Bacik0af3d002010-06-21 14:48:16 -04008682 cache->fs_info = root->fs_info;
Josef Bacik96303082009-07-13 21:29:25 -04008683
8684 /*
8685 * we only want to have 32k of ram per block group for keeping track
8686 * of free space, and if we pass 1/2 of that we want to start
8687 * converting things over to using bitmaps
8688 */
8689 cache->extents_thresh = ((1024 * 32) / 2) /
8690 sizeof(struct btrfs_free_space);
Yan Zhengd2fb3432008-12-11 16:30:39 -05008691 atomic_set(&cache->count, 1);
Chris Masonc286ac42008-07-22 23:06:41 -04008692 spin_lock_init(&cache->lock);
Josef Bacik6226cb02009-04-03 10:14:18 -04008693 spin_lock_init(&cache->tree_lock);
Josef Bacik0f9dd462008-09-23 13:14:11 -04008694 INIT_LIST_HEAD(&cache->list);
Chris Masonfa9c0d72009-04-03 09:47:43 -04008695 INIT_LIST_HEAD(&cache->cluster_list);
Chris Mason0ef3e662008-05-24 14:04:53 -04008696
Chris Mason6324fbf2008-03-24 15:01:59 -04008697 btrfs_set_block_group_used(&cache->item, bytes_used);
Chris Mason6324fbf2008-03-24 15:01:59 -04008698 btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
8699 cache->flags = type;
8700 btrfs_set_block_group_flags(&cache->item, type);
8701
Yan Zheng11833d62009-09-11 16:11:19 -04008702 cache->last_byte_to_unpin = (u64)-1;
Josef Bacik817d52f2009-07-13 21:29:25 -04008703 cache->cached = BTRFS_CACHE_FINISHED;
Yan Zheng11833d62009-09-11 16:11:19 -04008704 exclude_super_stripes(root, cache);
Josef Bacik96303082009-07-13 21:29:25 -04008705
Josef Bacik817d52f2009-07-13 21:29:25 -04008706 add_new_free_space(cache, root->fs_info, chunk_offset,
8707 chunk_offset + size);
8708
Yan Zheng11833d62009-09-11 16:11:19 -04008709 free_excluded_extents(root, cache);
8710
Chris Mason6324fbf2008-03-24 15:01:59 -04008711 ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
8712 &cache->space_info);
8713 BUG_ON(ret);
Josef Bacik1b2da372009-09-11 16:11:20 -04008714
8715 spin_lock(&cache->space_info->lock);
Yan, Zhengf0486c62010-05-16 10:46:25 -04008716 cache->space_info->bytes_readonly += cache->bytes_super;
Josef Bacik1b2da372009-09-11 16:11:20 -04008717 spin_unlock(&cache->space_info->lock);
8718
Yan, Zhengb742bb82010-05-16 10:46:24 -04008719 __link_block_group(cache->space_info, cache);
Chris Mason6324fbf2008-03-24 15:01:59 -04008720
Josef Bacik0f9dd462008-09-23 13:14:11 -04008721 ret = btrfs_add_block_group_cache(root->fs_info, cache);
8722 BUG_ON(ret);
Chris Masonc286ac42008-07-22 23:06:41 -04008723
Chris Mason6324fbf2008-03-24 15:01:59 -04008724 ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
8725 sizeof(cache->item));
8726 BUG_ON(ret);
8727
Chris Masond18a2c42008-04-04 15:40:00 -04008728 set_avail_alloc_bits(extent_root->fs_info, type);
Chris Mason925baed2008-06-25 16:01:30 -04008729
Chris Mason6324fbf2008-03-24 15:01:59 -04008730 return 0;
8731}
Zheng Yan1a40e232008-09-26 10:09:34 -04008732
8733int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
8734 struct btrfs_root *root, u64 group_start)
8735{
8736 struct btrfs_path *path;
8737 struct btrfs_block_group_cache *block_group;
Chris Mason44fb5512009-06-04 15:34:51 -04008738 struct btrfs_free_cluster *cluster;
Josef Bacik0af3d002010-06-21 14:48:16 -04008739 struct btrfs_root *tree_root = root->fs_info->tree_root;
Zheng Yan1a40e232008-09-26 10:09:34 -04008740 struct btrfs_key key;
Josef Bacik0af3d002010-06-21 14:48:16 -04008741 struct inode *inode;
Zheng Yan1a40e232008-09-26 10:09:34 -04008742 int ret;
Josef Bacik89a55892010-10-14 14:52:27 -04008743 int factor;
Zheng Yan1a40e232008-09-26 10:09:34 -04008744
Zheng Yan1a40e232008-09-26 10:09:34 -04008745 root = root->fs_info->extent_root;
8746
8747 block_group = btrfs_lookup_block_group(root->fs_info, group_start);
8748 BUG_ON(!block_group);
Yan Zhengc146afa2008-11-12 14:34:12 -05008749 BUG_ON(!block_group->ro);
Zheng Yan1a40e232008-09-26 10:09:34 -04008750
liubo9f7c43c2011-03-07 02:13:33 +00008751 /*
8752 * Free the reserved super bytes from this block group before
8753 * remove it.
8754 */
8755 free_excluded_extents(root, block_group);
8756
Zheng Yan1a40e232008-09-26 10:09:34 -04008757 memcpy(&key, &block_group->key, sizeof(key));
Josef Bacik89a55892010-10-14 14:52:27 -04008758 if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP |
8759 BTRFS_BLOCK_GROUP_RAID1 |
8760 BTRFS_BLOCK_GROUP_RAID10))
8761 factor = 2;
8762 else
8763 factor = 1;
Zheng Yan1a40e232008-09-26 10:09:34 -04008764
Chris Mason44fb5512009-06-04 15:34:51 -04008765 /* make sure this block group isn't part of an allocation cluster */
8766 cluster = &root->fs_info->data_alloc_cluster;
8767 spin_lock(&cluster->refill_lock);
8768 btrfs_return_cluster_to_free_space(block_group, cluster);
8769 spin_unlock(&cluster->refill_lock);
8770
8771 /*
8772 * make sure this block group isn't part of a metadata
8773 * allocation cluster
8774 */
8775 cluster = &root->fs_info->meta_alloc_cluster;
8776 spin_lock(&cluster->refill_lock);
8777 btrfs_return_cluster_to_free_space(block_group, cluster);
8778 spin_unlock(&cluster->refill_lock);
8779
Zheng Yan1a40e232008-09-26 10:09:34 -04008780 path = btrfs_alloc_path();
8781 BUG_ON(!path);
8782
Josef Bacik0af3d002010-06-21 14:48:16 -04008783 inode = lookup_free_space_inode(root, block_group, path);
8784 if (!IS_ERR(inode)) {
8785 btrfs_orphan_add(trans, inode);
8786 clear_nlink(inode);
8787 /* One for the block groups ref */
8788 spin_lock(&block_group->lock);
8789 if (block_group->iref) {
8790 block_group->iref = 0;
8791 block_group->inode = NULL;
8792 spin_unlock(&block_group->lock);
8793 iput(inode);
8794 } else {
8795 spin_unlock(&block_group->lock);
8796 }
8797 /* One for our lookup ref */
8798 iput(inode);
8799 }
8800
8801 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
8802 key.offset = block_group->key.objectid;
8803 key.type = 0;
8804
8805 ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
8806 if (ret < 0)
8807 goto out;
8808 if (ret > 0)
8809 btrfs_release_path(tree_root, path);
8810 if (ret == 0) {
8811 ret = btrfs_del_item(trans, tree_root, path);
8812 if (ret)
8813 goto out;
8814 btrfs_release_path(tree_root, path);
8815 }
8816
Yan Zheng3dfdb932009-01-21 10:49:16 -05008817 spin_lock(&root->fs_info->block_group_cache_lock);
Zheng Yan1a40e232008-09-26 10:09:34 -04008818 rb_erase(&block_group->cache_node,
8819 &root->fs_info->block_group_cache_tree);
Yan Zheng3dfdb932009-01-21 10:49:16 -05008820 spin_unlock(&root->fs_info->block_group_cache_lock);
Josef Bacik817d52f2009-07-13 21:29:25 -04008821
Josef Bacik80eb2342008-10-29 14:49:05 -04008822 down_write(&block_group->space_info->groups_sem);
Chris Mason44fb5512009-06-04 15:34:51 -04008823 /*
8824 * we must use list_del_init so people can check to see if they
8825 * are still on the list after taking the semaphore
8826 */
8827 list_del_init(&block_group->list);
Josef Bacik80eb2342008-10-29 14:49:05 -04008828 up_write(&block_group->space_info->groups_sem);
Zheng Yan1a40e232008-09-26 10:09:34 -04008829
Josef Bacik817d52f2009-07-13 21:29:25 -04008830 if (block_group->cached == BTRFS_CACHE_STARTED)
Yan Zheng11833d62009-09-11 16:11:19 -04008831 wait_block_group_cache_done(block_group);
Josef Bacik817d52f2009-07-13 21:29:25 -04008832
8833 btrfs_remove_free_space_cache(block_group);
8834
Yan Zhengc146afa2008-11-12 14:34:12 -05008835 spin_lock(&block_group->space_info->lock);
8836 block_group->space_info->total_bytes -= block_group->key.offset;
8837 block_group->space_info->bytes_readonly -= block_group->key.offset;
Josef Bacik89a55892010-10-14 14:52:27 -04008838 block_group->space_info->disk_total -= block_group->key.offset * factor;
Yan Zhengc146afa2008-11-12 14:34:12 -05008839 spin_unlock(&block_group->space_info->lock);
Chris Mason283bb192009-07-24 16:30:55 -04008840
Josef Bacik0af3d002010-06-21 14:48:16 -04008841 memcpy(&key, &block_group->key, sizeof(key));
8842
Chris Mason283bb192009-07-24 16:30:55 -04008843 btrfs_clear_space_info_full(root->fs_info);
Yan Zhengc146afa2008-11-12 14:34:12 -05008844
Chris Masonfa9c0d72009-04-03 09:47:43 -04008845 btrfs_put_block_group(block_group);
8846 btrfs_put_block_group(block_group);
Zheng Yan1a40e232008-09-26 10:09:34 -04008847
8848 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
8849 if (ret > 0)
8850 ret = -EIO;
8851 if (ret < 0)
8852 goto out;
8853
8854 ret = btrfs_del_item(trans, root, path);
8855out:
8856 btrfs_free_path(path);
8857 return ret;
8858}
liuboacce9522011-01-06 19:30:25 +08008859
liuboc59021f2011-03-07 02:13:14 +00008860int btrfs_init_space_info(struct btrfs_fs_info *fs_info)
8861{
8862 struct btrfs_space_info *space_info;
liubo1aba86d2011-04-08 08:44:37 +00008863 struct btrfs_super_block *disk_super;
8864 u64 features;
8865 u64 flags;
8866 int mixed = 0;
liuboc59021f2011-03-07 02:13:14 +00008867 int ret;
8868
liubo1aba86d2011-04-08 08:44:37 +00008869 disk_super = &fs_info->super_copy;
8870 if (!btrfs_super_root(disk_super))
8871 return 1;
liuboc59021f2011-03-07 02:13:14 +00008872
liubo1aba86d2011-04-08 08:44:37 +00008873 features = btrfs_super_incompat_flags(disk_super);
8874 if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
8875 mixed = 1;
liuboc59021f2011-03-07 02:13:14 +00008876
liubo1aba86d2011-04-08 08:44:37 +00008877 flags = BTRFS_BLOCK_GROUP_SYSTEM;
8878 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
liuboc59021f2011-03-07 02:13:14 +00008879 if (ret)
liubo1aba86d2011-04-08 08:44:37 +00008880 goto out;
liuboc59021f2011-03-07 02:13:14 +00008881
liubo1aba86d2011-04-08 08:44:37 +00008882 if (mixed) {
8883 flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA;
8884 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8885 } else {
8886 flags = BTRFS_BLOCK_GROUP_METADATA;
8887 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8888 if (ret)
8889 goto out;
8890
8891 flags = BTRFS_BLOCK_GROUP_DATA;
8892 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8893 }
8894out:
liuboc59021f2011-03-07 02:13:14 +00008895 return ret;
8896}
8897
liuboacce9522011-01-06 19:30:25 +08008898int btrfs_error_unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
8899{
8900 return unpin_extent_range(root, start, end);
8901}
8902
8903int btrfs_error_discard_extent(struct btrfs_root *root, u64 bytenr,
Li Dongyang5378e602011-03-24 10:24:27 +00008904 u64 num_bytes, u64 *actual_bytes)
liuboacce9522011-01-06 19:30:25 +08008905{
Li Dongyang5378e602011-03-24 10:24:27 +00008906 return btrfs_discard_extent(root, bytenr, num_bytes, actual_bytes);
liuboacce9522011-01-06 19:30:25 +08008907}
Li Dongyangf7039b12011-03-24 10:24:28 +00008908
8909int btrfs_trim_fs(struct btrfs_root *root, struct fstrim_range *range)
8910{
8911 struct btrfs_fs_info *fs_info = root->fs_info;
8912 struct btrfs_block_group_cache *cache = NULL;
8913 u64 group_trimmed;
8914 u64 start;
8915 u64 end;
8916 u64 trimmed = 0;
8917 int ret = 0;
8918
8919 cache = btrfs_lookup_block_group(fs_info, range->start);
8920
8921 while (cache) {
8922 if (cache->key.objectid >= (range->start + range->len)) {
8923 btrfs_put_block_group(cache);
8924 break;
8925 }
8926
8927 start = max(range->start, cache->key.objectid);
8928 end = min(range->start + range->len,
8929 cache->key.objectid + cache->key.offset);
8930
8931 if (end - start >= range->minlen) {
8932 if (!block_group_cache_done(cache)) {
8933 ret = cache_block_group(cache, NULL, root, 0);
8934 if (!ret)
8935 wait_block_group_cache_done(cache);
8936 }
8937 ret = btrfs_trim_block_group(cache,
8938 &group_trimmed,
8939 start,
8940 end,
8941 range->minlen);
8942
8943 trimmed += group_trimmed;
8944 if (ret) {
8945 btrfs_put_block_group(cache);
8946 break;
8947 }
8948 }
8949
8950 cache = next_block_group(fs_info->tree_root, cache);
8951 }
8952
8953 range->len = trimmed;
8954 return ret;
8955}