blob: 1204c8ef6f32751bd859d2ef1a2ab3444061f708 [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>
David Sterbadff51cd2011-06-14 12:52:17 +020026#include <linux/ratelimit.h>
Josef Bacikb150a4f2013-06-19 15:00:04 -040027#include <linux/percpu_counter.h>
Chris Mason4b4e25f2008-11-20 10:22:27 -050028#include "compat.h"
Chris Mason74493f72007-12-11 09:25:06 -050029#include "hash.h"
Chris Masonfec577f2007-02-26 10:40:21 -050030#include "ctree.h"
31#include "disk-io.h"
32#include "print-tree.h"
Chris Masone089f052007-03-16 16:20:31 -040033#include "transaction.h"
Chris Mason0b86a832008-03-24 15:01:56 -040034#include "volumes.h"
David Woodhouse53b381b2013-01-29 18:40:14 -050035#include "raid56.h"
Chris Mason925baed2008-06-25 16:01:30 -040036#include "locking.h"
Chris Masonfa9c0d792009-04-03 09:47:43 -040037#include "free-space-cache.h"
Miao Xie3fed40c2012-09-13 04:51:36 -060038#include "math.h"
Chris Masonfec577f2007-02-26 10:40:21 -050039
Arne Jansen709c0482011-09-12 12:22:57 +020040#undef SCRAMBLE_DELAYED_REFS
41
Miao Xie9e622d62012-01-26 15:01:12 -050042/*
43 * control flags for do_chunk_alloc's force field
Chris Mason0e4f8f82011-04-15 16:05:44 -040044 * CHUNK_ALLOC_NO_FORCE means to only allocate a chunk
45 * if we really need one.
46 *
Chris Mason0e4f8f82011-04-15 16:05:44 -040047 * CHUNK_ALLOC_LIMITED means to only try and allocate one
48 * if we have very few chunks already allocated. This is
49 * used as part of the clustering code to help make sure
50 * we have a good pool of storage to cluster in, without
51 * filling the FS with empty chunks
52 *
Miao Xie9e622d62012-01-26 15:01:12 -050053 * CHUNK_ALLOC_FORCE means it must try to allocate one
54 *
Chris Mason0e4f8f82011-04-15 16:05:44 -040055 */
56enum {
57 CHUNK_ALLOC_NO_FORCE = 0,
Miao Xie9e622d62012-01-26 15:01:12 -050058 CHUNK_ALLOC_LIMITED = 1,
59 CHUNK_ALLOC_FORCE = 2,
Chris Mason0e4f8f82011-04-15 16:05:44 -040060};
61
Josef Bacikfb25e912011-07-26 17:00:46 -040062/*
63 * Control how reservations are dealt with.
64 *
65 * RESERVE_FREE - freeing a reservation.
66 * RESERVE_ALLOC - allocating space and we need to update bytes_may_use for
67 * ENOSPC accounting
68 * RESERVE_ALLOC_NO_ACCOUNT - allocating space and we should not update
69 * bytes_may_use as the ENOSPC accounting is done elsewhere
70 */
71enum {
72 RESERVE_FREE = 0,
73 RESERVE_ALLOC = 1,
74 RESERVE_ALLOC_NO_ACCOUNT = 2,
75};
76
Liu Boc53d6132012-12-27 09:01:19 +000077static int update_block_group(struct btrfs_root *root,
Yan, Zhengf0486c62010-05-16 10:46:25 -040078 u64 bytenr, u64 num_bytes, int alloc);
Yan Zheng5d4f98a2009-06-10 10:45:14 -040079static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
80 struct btrfs_root *root,
81 u64 bytenr, u64 num_bytes, u64 parent,
82 u64 root_objectid, u64 owner_objectid,
83 u64 owner_offset, int refs_to_drop,
84 struct btrfs_delayed_extent_op *extra_op);
85static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
86 struct extent_buffer *leaf,
87 struct btrfs_extent_item *ei);
88static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
89 struct btrfs_root *root,
90 u64 parent, u64 root_objectid,
91 u64 flags, u64 owner, u64 offset,
92 struct btrfs_key *ins, int ref_mod);
93static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
94 struct btrfs_root *root,
95 u64 parent, u64 root_objectid,
96 u64 flags, struct btrfs_disk_key *key,
97 int level, struct btrfs_key *ins);
Josef Bacik6a632092009-02-20 11:00:09 -050098static int do_chunk_alloc(struct btrfs_trans_handle *trans,
Josef Bacik698d0082012-09-12 14:08:47 -040099 struct btrfs_root *extent_root, u64 flags,
100 int force);
Yan Zheng11833d62009-09-11 16:11:19 -0400101static int find_next_key(struct btrfs_path *path, int level,
102 struct btrfs_key *key);
Josef Bacik9ed74f22009-09-11 16:12:44 -0400103static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
104 int dump_block_groups);
Josef Bacikfb25e912011-07-26 17:00:46 -0400105static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
106 u64 num_bytes, int reserve);
Josef Bacik5d803662013-02-07 16:06:02 -0500107static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
108 u64 num_bytes);
Eric Sandeen48a3b632013-04-25 20:41:01 +0000109int btrfs_pin_extent(struct btrfs_root *root,
110 u64 bytenr, u64 num_bytes, int reserved);
Josef Bacik6a632092009-02-20 11:00:09 -0500111
Josef Bacik817d52f2009-07-13 21:29:25 -0400112static noinline int
113block_group_cache_done(struct btrfs_block_group_cache *cache)
114{
115 smp_mb();
116 return cache->cached == BTRFS_CACHE_FINISHED;
117}
118
Josef Bacik0f9dd462008-09-23 13:14:11 -0400119static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
120{
121 return (cache->flags & bits) == bits;
122}
123
David Sterba62a45b62011-04-20 15:52:26 +0200124static void btrfs_get_block_group(struct btrfs_block_group_cache *cache)
Josef Bacik11dfe352009-11-13 20:12:59 +0000125{
126 atomic_inc(&cache->count);
127}
128
129void btrfs_put_block_group(struct btrfs_block_group_cache *cache)
130{
Yan, Zhengf0486c62010-05-16 10:46:25 -0400131 if (atomic_dec_and_test(&cache->count)) {
132 WARN_ON(cache->pinned > 0);
133 WARN_ON(cache->reserved > 0);
Li Zefan34d52cb2011-03-29 13:46:06 +0800134 kfree(cache->free_space_ctl);
Josef Bacik11dfe352009-11-13 20:12:59 +0000135 kfree(cache);
Yan, Zhengf0486c62010-05-16 10:46:25 -0400136 }
Josef Bacik11dfe352009-11-13 20:12:59 +0000137}
138
Josef Bacik0f9dd462008-09-23 13:14:11 -0400139/*
140 * this adds the block group to the fs_info rb tree for the block group
141 * cache
142 */
Christoph Hellwigb2950862008-12-02 09:54:17 -0500143static int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
Josef Bacik0f9dd462008-09-23 13:14:11 -0400144 struct btrfs_block_group_cache *block_group)
145{
146 struct rb_node **p;
147 struct rb_node *parent = NULL;
148 struct btrfs_block_group_cache *cache;
149
150 spin_lock(&info->block_group_cache_lock);
151 p = &info->block_group_cache_tree.rb_node;
152
153 while (*p) {
154 parent = *p;
155 cache = rb_entry(parent, struct btrfs_block_group_cache,
156 cache_node);
157 if (block_group->key.objectid < cache->key.objectid) {
158 p = &(*p)->rb_left;
159 } else if (block_group->key.objectid > cache->key.objectid) {
160 p = &(*p)->rb_right;
161 } else {
162 spin_unlock(&info->block_group_cache_lock);
163 return -EEXIST;
164 }
165 }
166
167 rb_link_node(&block_group->cache_node, parent, p);
168 rb_insert_color(&block_group->cache_node,
169 &info->block_group_cache_tree);
Liu Boa1897fd2012-12-27 09:01:23 +0000170
171 if (info->first_logical_byte > block_group->key.objectid)
172 info->first_logical_byte = block_group->key.objectid;
173
Josef Bacik0f9dd462008-09-23 13:14:11 -0400174 spin_unlock(&info->block_group_cache_lock);
175
176 return 0;
177}
178
179/*
180 * This will return the block group at or after bytenr if contains is 0, else
181 * it will return the block group that contains the bytenr
182 */
183static struct btrfs_block_group_cache *
184block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
185 int contains)
186{
187 struct btrfs_block_group_cache *cache, *ret = NULL;
188 struct rb_node *n;
189 u64 end, start;
190
191 spin_lock(&info->block_group_cache_lock);
192 n = info->block_group_cache_tree.rb_node;
193
194 while (n) {
195 cache = rb_entry(n, struct btrfs_block_group_cache,
196 cache_node);
197 end = cache->key.objectid + cache->key.offset - 1;
198 start = cache->key.objectid;
199
200 if (bytenr < start) {
201 if (!contains && (!ret || start < ret->key.objectid))
202 ret = cache;
203 n = n->rb_left;
204 } else if (bytenr > start) {
205 if (contains && bytenr <= end) {
206 ret = cache;
207 break;
208 }
209 n = n->rb_right;
210 } else {
211 ret = cache;
212 break;
213 }
214 }
Liu Boa1897fd2012-12-27 09:01:23 +0000215 if (ret) {
Josef Bacik11dfe352009-11-13 20:12:59 +0000216 btrfs_get_block_group(ret);
Liu Boa1897fd2012-12-27 09:01:23 +0000217 if (bytenr == 0 && info->first_logical_byte > ret->key.objectid)
218 info->first_logical_byte = ret->key.objectid;
219 }
Josef Bacik0f9dd462008-09-23 13:14:11 -0400220 spin_unlock(&info->block_group_cache_lock);
221
222 return ret;
223}
224
Yan Zheng11833d62009-09-11 16:11:19 -0400225static int add_excluded_extent(struct btrfs_root *root,
226 u64 start, u64 num_bytes)
Josef Bacik817d52f2009-07-13 21:29:25 -0400227{
Yan Zheng11833d62009-09-11 16:11:19 -0400228 u64 end = start + num_bytes - 1;
229 set_extent_bits(&root->fs_info->freed_extents[0],
230 start, end, EXTENT_UPTODATE, GFP_NOFS);
231 set_extent_bits(&root->fs_info->freed_extents[1],
232 start, end, EXTENT_UPTODATE, GFP_NOFS);
233 return 0;
Josef Bacik817d52f2009-07-13 21:29:25 -0400234}
235
Yan Zheng11833d62009-09-11 16:11:19 -0400236static void free_excluded_extents(struct btrfs_root *root,
237 struct btrfs_block_group_cache *cache)
Josef Bacik817d52f2009-07-13 21:29:25 -0400238{
Yan Zheng11833d62009-09-11 16:11:19 -0400239 u64 start, end;
240
241 start = cache->key.objectid;
242 end = start + cache->key.offset - 1;
243
244 clear_extent_bits(&root->fs_info->freed_extents[0],
245 start, end, EXTENT_UPTODATE, GFP_NOFS);
246 clear_extent_bits(&root->fs_info->freed_extents[1],
247 start, end, EXTENT_UPTODATE, GFP_NOFS);
248}
249
250static int exclude_super_stripes(struct btrfs_root *root,
251 struct btrfs_block_group_cache *cache)
252{
Josef Bacik817d52f2009-07-13 21:29:25 -0400253 u64 bytenr;
254 u64 *logical;
255 int stripe_len;
256 int i, nr, ret;
257
Yan, Zheng06b23312009-11-26 09:31:11 +0000258 if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) {
259 stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid;
260 cache->bytes_super += stripe_len;
261 ret = add_excluded_extent(root, cache->key.objectid,
262 stripe_len);
Josef Bacik835d9742013-03-19 12:13:25 -0400263 if (ret)
264 return ret;
Yan, Zheng06b23312009-11-26 09:31:11 +0000265 }
266
Josef Bacik817d52f2009-07-13 21:29:25 -0400267 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
268 bytenr = btrfs_sb_offset(i);
269 ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
270 cache->key.objectid, bytenr,
271 0, &logical, &nr, &stripe_len);
Josef Bacik835d9742013-03-19 12:13:25 -0400272 if (ret)
273 return ret;
Yan Zheng11833d62009-09-11 16:11:19 -0400274
Josef Bacik817d52f2009-07-13 21:29:25 -0400275 while (nr--) {
Josef Bacik51bf5f02013-04-23 12:55:21 -0400276 u64 start, len;
277
278 if (logical[nr] > cache->key.objectid +
279 cache->key.offset)
280 continue;
281
282 if (logical[nr] + stripe_len <= cache->key.objectid)
283 continue;
284
285 start = logical[nr];
286 if (start < cache->key.objectid) {
287 start = cache->key.objectid;
288 len = (logical[nr] + stripe_len) - start;
289 } else {
290 len = min_t(u64, stripe_len,
291 cache->key.objectid +
292 cache->key.offset - start);
293 }
294
295 cache->bytes_super += len;
296 ret = add_excluded_extent(root, start, len);
Josef Bacik835d9742013-03-19 12:13:25 -0400297 if (ret) {
298 kfree(logical);
299 return ret;
300 }
Josef Bacik817d52f2009-07-13 21:29:25 -0400301 }
Yan Zheng11833d62009-09-11 16:11:19 -0400302
Josef Bacik817d52f2009-07-13 21:29:25 -0400303 kfree(logical);
304 }
Josef Bacik817d52f2009-07-13 21:29:25 -0400305 return 0;
306}
307
Yan Zheng11833d62009-09-11 16:11:19 -0400308static struct btrfs_caching_control *
309get_caching_control(struct btrfs_block_group_cache *cache)
310{
311 struct btrfs_caching_control *ctl;
312
313 spin_lock(&cache->lock);
314 if (cache->cached != BTRFS_CACHE_STARTED) {
315 spin_unlock(&cache->lock);
316 return NULL;
317 }
318
Josef Bacikdde5abe2010-09-16 16:17:03 -0400319 /* We're loading it the fast way, so we don't have a caching_ctl. */
320 if (!cache->caching_ctl) {
321 spin_unlock(&cache->lock);
322 return NULL;
323 }
324
Yan Zheng11833d62009-09-11 16:11:19 -0400325 ctl = cache->caching_ctl;
326 atomic_inc(&ctl->count);
327 spin_unlock(&cache->lock);
328 return ctl;
329}
330
331static void put_caching_control(struct btrfs_caching_control *ctl)
332{
333 if (atomic_dec_and_test(&ctl->count))
334 kfree(ctl);
335}
336
Josef Bacik0f9dd462008-09-23 13:14:11 -0400337/*
338 * this is only called by cache_block_group, since we could have freed extents
339 * we need to check the pinned_extents for any extents that can't be used yet
340 * since their free space will be released as soon as the transaction commits.
341 */
Josef Bacik817d52f2009-07-13 21:29:25 -0400342static u64 add_new_free_space(struct btrfs_block_group_cache *block_group,
Josef Bacik0f9dd462008-09-23 13:14:11 -0400343 struct btrfs_fs_info *info, u64 start, u64 end)
344{
Josef Bacik817d52f2009-07-13 21:29:25 -0400345 u64 extent_start, extent_end, size, total_added = 0;
Josef Bacik0f9dd462008-09-23 13:14:11 -0400346 int ret;
347
348 while (start < end) {
Yan Zheng11833d62009-09-11 16:11:19 -0400349 ret = find_first_extent_bit(info->pinned_extents, start,
Josef Bacik0f9dd462008-09-23 13:14:11 -0400350 &extent_start, &extent_end,
Josef Bacike6138872012-09-27 17:07:30 -0400351 EXTENT_DIRTY | EXTENT_UPTODATE,
352 NULL);
Josef Bacik0f9dd462008-09-23 13:14:11 -0400353 if (ret)
354 break;
355
Yan, Zheng06b23312009-11-26 09:31:11 +0000356 if (extent_start <= start) {
Josef Bacik0f9dd462008-09-23 13:14:11 -0400357 start = extent_end + 1;
358 } else if (extent_start > start && extent_start < end) {
359 size = extent_start - start;
Josef Bacik817d52f2009-07-13 21:29:25 -0400360 total_added += size;
Josef Bacikea6a4782008-11-20 12:16:16 -0500361 ret = btrfs_add_free_space(block_group, start,
362 size);
Jeff Mahoney79787ea2012-03-12 16:03:00 +0100363 BUG_ON(ret); /* -ENOMEM or logic error */
Josef Bacik0f9dd462008-09-23 13:14:11 -0400364 start = extent_end + 1;
365 } else {
366 break;
367 }
368 }
369
370 if (start < end) {
371 size = end - start;
Josef Bacik817d52f2009-07-13 21:29:25 -0400372 total_added += size;
Josef Bacikea6a4782008-11-20 12:16:16 -0500373 ret = btrfs_add_free_space(block_group, start, size);
Jeff Mahoney79787ea2012-03-12 16:03:00 +0100374 BUG_ON(ret); /* -ENOMEM or logic error */
Josef Bacik0f9dd462008-09-23 13:14:11 -0400375 }
376
Josef Bacik817d52f2009-07-13 21:29:25 -0400377 return total_added;
Josef Bacik0f9dd462008-09-23 13:14:11 -0400378}
379
Josef Bacikbab39bf2011-06-30 14:42:28 -0400380static noinline void caching_thread(struct btrfs_work *work)
Chris Masone37c9e62007-05-09 20:13:14 -0400381{
Josef Bacikbab39bf2011-06-30 14:42:28 -0400382 struct btrfs_block_group_cache *block_group;
383 struct btrfs_fs_info *fs_info;
384 struct btrfs_caching_control *caching_ctl;
385 struct btrfs_root *extent_root;
Chris Masone37c9e62007-05-09 20:13:14 -0400386 struct btrfs_path *path;
Chris Mason5f39d392007-10-15 16:14:19 -0400387 struct extent_buffer *leaf;
Yan Zheng11833d62009-09-11 16:11:19 -0400388 struct btrfs_key key;
Josef Bacik817d52f2009-07-13 21:29:25 -0400389 u64 total_found = 0;
Yan Zheng11833d62009-09-11 16:11:19 -0400390 u64 last = 0;
391 u32 nritems;
392 int ret = 0;
Chris Masonf510cfe2007-10-15 16:14:48 -0400393
Josef Bacikbab39bf2011-06-30 14:42:28 -0400394 caching_ctl = container_of(work, struct btrfs_caching_control, work);
395 block_group = caching_ctl->block_group;
396 fs_info = block_group->fs_info;
397 extent_root = fs_info->extent_root;
398
Chris Masone37c9e62007-05-09 20:13:14 -0400399 path = btrfs_alloc_path();
400 if (!path)
Josef Bacikbab39bf2011-06-30 14:42:28 -0400401 goto out;
Yan7d7d6062007-09-14 16:15:28 -0400402
Josef Bacik817d52f2009-07-13 21:29:25 -0400403 last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
Yan Zheng11833d62009-09-11 16:11:19 -0400404
Chris Mason5cd57b22008-06-25 16:01:30 -0400405 /*
Josef Bacik817d52f2009-07-13 21:29:25 -0400406 * We don't want to deadlock with somebody trying to allocate a new
407 * extent for the extent root while also trying to search the extent
408 * root to add free space. So we skip locking and search the commit
409 * root, since its read-only
Chris Mason5cd57b22008-06-25 16:01:30 -0400410 */
411 path->skip_locking = 1;
Josef Bacik817d52f2009-07-13 21:29:25 -0400412 path->search_commit_root = 1;
Josef Bacik026fd312011-05-13 10:32:11 -0400413 path->reada = 1;
Josef Bacik817d52f2009-07-13 21:29:25 -0400414
Yan Zhenge4404d62008-12-12 10:03:26 -0500415 key.objectid = last;
Chris Masone37c9e62007-05-09 20:13:14 -0400416 key.offset = 0;
Yan Zheng11833d62009-09-11 16:11:19 -0400417 key.type = BTRFS_EXTENT_ITEM_KEY;
Chris Mason013f1b12009-07-31 14:57:55 -0400418again:
Yan Zheng11833d62009-09-11 16:11:19 -0400419 mutex_lock(&caching_ctl->mutex);
Chris Mason013f1b12009-07-31 14:57:55 -0400420 /* need to make sure the commit_root doesn't disappear */
421 down_read(&fs_info->extent_commit_sem);
422
Yan Zheng11833d62009-09-11 16:11:19 -0400423 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
Chris Masone37c9e62007-05-09 20:13:14 -0400424 if (ret < 0)
Josef Bacikef8bbdf2008-09-23 13:14:11 -0400425 goto err;
Yan Zhenga512bbf2008-12-08 16:46:26 -0500426
Yan Zheng11833d62009-09-11 16:11:19 -0400427 leaf = path->nodes[0];
428 nritems = btrfs_header_nritems(leaf);
429
Chris Masond3977122009-01-05 21:25:51 -0500430 while (1) {
David Sterba7841cb22011-05-31 18:07:27 +0200431 if (btrfs_fs_closing(fs_info) > 1) {
Yan Zhengf25784b2009-07-28 08:41:57 -0400432 last = (u64)-1;
Josef Bacik817d52f2009-07-13 21:29:25 -0400433 break;
Yan Zhengf25784b2009-07-28 08:41:57 -0400434 }
Josef Bacik817d52f2009-07-13 21:29:25 -0400435
Yan Zheng11833d62009-09-11 16:11:19 -0400436 if (path->slots[0] < nritems) {
437 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
438 } else {
439 ret = find_next_key(path, 0, &key);
440 if (ret)
Chris Masone37c9e62007-05-09 20:13:14 -0400441 break;
Josef Bacik817d52f2009-07-13 21:29:25 -0400442
Josef Bacik0a3896d2013-04-19 14:37:26 -0400443 if (need_resched()) {
Josef Bacik589d8ad2011-05-11 17:30:53 -0400444 caching_ctl->progress = last;
Chris Masonff5714c2011-05-28 07:00:39 -0400445 btrfs_release_path(path);
Josef Bacik589d8ad2011-05-11 17:30:53 -0400446 up_read(&fs_info->extent_commit_sem);
447 mutex_unlock(&caching_ctl->mutex);
Yan Zheng11833d62009-09-11 16:11:19 -0400448 cond_resched();
Josef Bacik589d8ad2011-05-11 17:30:53 -0400449 goto again;
450 }
Josef Bacik0a3896d2013-04-19 14:37:26 -0400451
452 ret = btrfs_next_leaf(extent_root, path);
453 if (ret < 0)
454 goto err;
455 if (ret)
456 break;
Josef Bacik589d8ad2011-05-11 17:30:53 -0400457 leaf = path->nodes[0];
458 nritems = btrfs_header_nritems(leaf);
459 continue;
Yan Zheng11833d62009-09-11 16:11:19 -0400460 }
Josef Bacik817d52f2009-07-13 21:29:25 -0400461
Yan Zheng11833d62009-09-11 16:11:19 -0400462 if (key.objectid < block_group->key.objectid) {
463 path->slots[0]++;
Josef Bacik817d52f2009-07-13 21:29:25 -0400464 continue;
Chris Masone37c9e62007-05-09 20:13:14 -0400465 }
Josef Bacik0f9dd462008-09-23 13:14:11 -0400466
Chris Masone37c9e62007-05-09 20:13:14 -0400467 if (key.objectid >= block_group->key.objectid +
Josef Bacik0f9dd462008-09-23 13:14:11 -0400468 block_group->key.offset)
Yan7d7d6062007-09-14 16:15:28 -0400469 break;
Yan7d7d6062007-09-14 16:15:28 -0400470
Josef Bacik3173a182013-03-07 14:22:04 -0500471 if (key.type == BTRFS_EXTENT_ITEM_KEY ||
472 key.type == BTRFS_METADATA_ITEM_KEY) {
Josef Bacik817d52f2009-07-13 21:29:25 -0400473 total_found += add_new_free_space(block_group,
474 fs_info, last,
475 key.objectid);
Josef Bacik3173a182013-03-07 14:22:04 -0500476 if (key.type == BTRFS_METADATA_ITEM_KEY)
477 last = key.objectid +
478 fs_info->tree_root->leafsize;
479 else
480 last = key.objectid + key.offset;
Josef Bacik817d52f2009-07-13 21:29:25 -0400481
Yan Zheng11833d62009-09-11 16:11:19 -0400482 if (total_found > (1024 * 1024 * 2)) {
483 total_found = 0;
484 wake_up(&caching_ctl->wait);
485 }
Josef Bacik817d52f2009-07-13 21:29:25 -0400486 }
Chris Masone37c9e62007-05-09 20:13:14 -0400487 path->slots[0]++;
488 }
Josef Bacikef8bbdf2008-09-23 13:14:11 -0400489 ret = 0;
Josef Bacik817d52f2009-07-13 21:29:25 -0400490
491 total_found += add_new_free_space(block_group, fs_info, last,
492 block_group->key.objectid +
493 block_group->key.offset);
Yan Zheng11833d62009-09-11 16:11:19 -0400494 caching_ctl->progress = (u64)-1;
Josef Bacik817d52f2009-07-13 21:29:25 -0400495
496 spin_lock(&block_group->lock);
Yan Zheng11833d62009-09-11 16:11:19 -0400497 block_group->caching_ctl = NULL;
Josef Bacik817d52f2009-07-13 21:29:25 -0400498 block_group->cached = BTRFS_CACHE_FINISHED;
499 spin_unlock(&block_group->lock);
500
Chris Mason54aa1f42007-06-22 14:16:25 -0400501err:
Chris Masone37c9e62007-05-09 20:13:14 -0400502 btrfs_free_path(path);
Yan Zheng276e6802009-07-30 09:40:40 -0400503 up_read(&fs_info->extent_commit_sem);
Josef Bacik817d52f2009-07-13 21:29:25 -0400504
Yan Zheng11833d62009-09-11 16:11:19 -0400505 free_excluded_extents(extent_root, block_group);
506
507 mutex_unlock(&caching_ctl->mutex);
Josef Bacikbab39bf2011-06-30 14:42:28 -0400508out:
Yan Zheng11833d62009-09-11 16:11:19 -0400509 wake_up(&caching_ctl->wait);
510
511 put_caching_control(caching_ctl);
Josef Bacik11dfe352009-11-13 20:12:59 +0000512 btrfs_put_block_group(block_group);
Josef Bacik817d52f2009-07-13 21:29:25 -0400513}
514
Josef Bacik9d66e232010-08-25 16:54:15 -0400515static int cache_block_group(struct btrfs_block_group_cache *cache,
Josef Bacik9d66e232010-08-25 16:54:15 -0400516 int load_cache_only)
Josef Bacik817d52f2009-07-13 21:29:25 -0400517{
Josef Bacik291c7d22011-11-14 13:52:14 -0500518 DEFINE_WAIT(wait);
Yan Zheng11833d62009-09-11 16:11:19 -0400519 struct btrfs_fs_info *fs_info = cache->fs_info;
520 struct btrfs_caching_control *caching_ctl;
Josef Bacik817d52f2009-07-13 21:29:25 -0400521 int ret = 0;
522
Josef Bacik291c7d22011-11-14 13:52:14 -0500523 caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_NOFS);
Jeff Mahoney79787ea2012-03-12 16:03:00 +0100524 if (!caching_ctl)
525 return -ENOMEM;
Josef Bacik291c7d22011-11-14 13:52:14 -0500526
527 INIT_LIST_HEAD(&caching_ctl->list);
528 mutex_init(&caching_ctl->mutex);
529 init_waitqueue_head(&caching_ctl->wait);
530 caching_ctl->block_group = cache;
531 caching_ctl->progress = cache->key.objectid;
532 atomic_set(&caching_ctl->count, 1);
533 caching_ctl->work.func = caching_thread;
534
535 spin_lock(&cache->lock);
536 /*
537 * This should be a rare occasion, but this could happen I think in the
538 * case where one thread starts to load the space cache info, and then
539 * some other thread starts a transaction commit which tries to do an
540 * allocation while the other thread is still loading the space cache
541 * info. The previous loop should have kept us from choosing this block
542 * group, but if we've moved to the state where we will wait on caching
543 * block groups we need to first check if we're doing a fast load here,
544 * so we can wait for it to finish, otherwise we could end up allocating
545 * from a block group who's cache gets evicted for one reason or
546 * another.
547 */
548 while (cache->cached == BTRFS_CACHE_FAST) {
549 struct btrfs_caching_control *ctl;
550
551 ctl = cache->caching_ctl;
552 atomic_inc(&ctl->count);
553 prepare_to_wait(&ctl->wait, &wait, TASK_UNINTERRUPTIBLE);
554 spin_unlock(&cache->lock);
555
556 schedule();
557
558 finish_wait(&ctl->wait, &wait);
559 put_caching_control(ctl);
560 spin_lock(&cache->lock);
561 }
562
563 if (cache->cached != BTRFS_CACHE_NO) {
564 spin_unlock(&cache->lock);
565 kfree(caching_ctl);
Yan Zheng11833d62009-09-11 16:11:19 -0400566 return 0;
Josef Bacik291c7d22011-11-14 13:52:14 -0500567 }
568 WARN_ON(cache->caching_ctl);
569 cache->caching_ctl = caching_ctl;
570 cache->cached = BTRFS_CACHE_FAST;
571 spin_unlock(&cache->lock);
Yan Zheng11833d62009-09-11 16:11:19 -0400572
Josef Bacikd53ba472012-04-12 16:03:57 -0400573 if (fs_info->mount_opt & BTRFS_MOUNT_SPACE_CACHE) {
Josef Bacik9d66e232010-08-25 16:54:15 -0400574 ret = load_free_space_cache(fs_info, cache);
575
576 spin_lock(&cache->lock);
577 if (ret == 1) {
Josef Bacik291c7d22011-11-14 13:52:14 -0500578 cache->caching_ctl = NULL;
Josef Bacik9d66e232010-08-25 16:54:15 -0400579 cache->cached = BTRFS_CACHE_FINISHED;
580 cache->last_byte_to_unpin = (u64)-1;
581 } else {
Josef Bacik291c7d22011-11-14 13:52:14 -0500582 if (load_cache_only) {
583 cache->caching_ctl = NULL;
584 cache->cached = BTRFS_CACHE_NO;
585 } else {
586 cache->cached = BTRFS_CACHE_STARTED;
587 }
Josef Bacik9d66e232010-08-25 16:54:15 -0400588 }
589 spin_unlock(&cache->lock);
Josef Bacik291c7d22011-11-14 13:52:14 -0500590 wake_up(&caching_ctl->wait);
Josef Bacik3c148742011-02-02 15:53:47 +0000591 if (ret == 1) {
Josef Bacik291c7d22011-11-14 13:52:14 -0500592 put_caching_control(caching_ctl);
Josef Bacik3c148742011-02-02 15:53:47 +0000593 free_excluded_extents(fs_info->extent_root, cache);
Josef Bacik9d66e232010-08-25 16:54:15 -0400594 return 0;
Josef Bacik3c148742011-02-02 15:53:47 +0000595 }
Josef Bacik291c7d22011-11-14 13:52:14 -0500596 } else {
597 /*
598 * We are not going to do the fast caching, set cached to the
599 * appropriate value and wakeup any waiters.
600 */
601 spin_lock(&cache->lock);
602 if (load_cache_only) {
603 cache->caching_ctl = NULL;
604 cache->cached = BTRFS_CACHE_NO;
605 } else {
606 cache->cached = BTRFS_CACHE_STARTED;
607 }
Josef Bacik817d52f2009-07-13 21:29:25 -0400608 spin_unlock(&cache->lock);
Josef Bacik291c7d22011-11-14 13:52:14 -0500609 wake_up(&caching_ctl->wait);
610 }
611
612 if (load_cache_only) {
613 put_caching_control(caching_ctl);
Yan Zheng11833d62009-09-11 16:11:19 -0400614 return 0;
Josef Bacik817d52f2009-07-13 21:29:25 -0400615 }
Josef Bacik817d52f2009-07-13 21:29:25 -0400616
Yan Zheng11833d62009-09-11 16:11:19 -0400617 down_write(&fs_info->extent_commit_sem);
Josef Bacik291c7d22011-11-14 13:52:14 -0500618 atomic_inc(&caching_ctl->count);
Yan Zheng11833d62009-09-11 16:11:19 -0400619 list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups);
620 up_write(&fs_info->extent_commit_sem);
621
Josef Bacik11dfe352009-11-13 20:12:59 +0000622 btrfs_get_block_group(cache);
Yan Zheng11833d62009-09-11 16:11:19 -0400623
Josef Bacikbab39bf2011-06-30 14:42:28 -0400624 btrfs_queue_worker(&fs_info->caching_workers, &caching_ctl->work);
Josef Bacik817d52f2009-07-13 21:29:25 -0400625
Josef Bacikef8bbdf2008-09-23 13:14:11 -0400626 return ret;
Chris Masone37c9e62007-05-09 20:13:14 -0400627}
628
Josef Bacik0f9dd462008-09-23 13:14:11 -0400629/*
630 * return the block group that starts at or after bytenr
631 */
Chris Masond3977122009-01-05 21:25:51 -0500632static struct btrfs_block_group_cache *
633btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr)
Chris Mason0ef3e662008-05-24 14:04:53 -0400634{
Josef Bacik0f9dd462008-09-23 13:14:11 -0400635 struct btrfs_block_group_cache *cache;
Chris Mason0ef3e662008-05-24 14:04:53 -0400636
Josef Bacik0f9dd462008-09-23 13:14:11 -0400637 cache = block_group_cache_tree_search(info, bytenr, 0);
Chris Mason0ef3e662008-05-24 14:04:53 -0400638
Josef Bacik0f9dd462008-09-23 13:14:11 -0400639 return cache;
Chris Mason0ef3e662008-05-24 14:04:53 -0400640}
641
Josef Bacik0f9dd462008-09-23 13:14:11 -0400642/*
Sankar P9f556842009-05-14 13:52:22 -0400643 * return the block group that contains the given bytenr
Josef Bacik0f9dd462008-09-23 13:14:11 -0400644 */
Chris Masond3977122009-01-05 21:25:51 -0500645struct btrfs_block_group_cache *btrfs_lookup_block_group(
646 struct btrfs_fs_info *info,
647 u64 bytenr)
Chris Masonbe744172007-05-06 10:15:01 -0400648{
Josef Bacik0f9dd462008-09-23 13:14:11 -0400649 struct btrfs_block_group_cache *cache;
Chris Masonbe744172007-05-06 10:15:01 -0400650
Josef Bacik0f9dd462008-09-23 13:14:11 -0400651 cache = block_group_cache_tree_search(info, bytenr, 1);
Chris Mason96b51792007-10-15 16:15:19 -0400652
Josef Bacik0f9dd462008-09-23 13:14:11 -0400653 return cache;
Chris Masonbe744172007-05-06 10:15:01 -0400654}
Chris Mason0b86a832008-03-24 15:01:56 -0400655
Josef Bacik0f9dd462008-09-23 13:14:11 -0400656static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
657 u64 flags)
Chris Mason6324fbf2008-03-24 15:01:59 -0400658{
Josef Bacik0f9dd462008-09-23 13:14:11 -0400659 struct list_head *head = &info->space_info;
Josef Bacik0f9dd462008-09-23 13:14:11 -0400660 struct btrfs_space_info *found;
Chris Mason4184ea72009-03-10 12:39:20 -0400661
Ilya Dryomov52ba6922012-01-16 22:04:47 +0200662 flags &= BTRFS_BLOCK_GROUP_TYPE_MASK;
Yan, Zhengb742bb82010-05-16 10:46:24 -0400663
Chris Mason4184ea72009-03-10 12:39:20 -0400664 rcu_read_lock();
665 list_for_each_entry_rcu(found, head, list) {
Josef Bacik67377732010-09-16 16:19:09 -0400666 if (found->flags & flags) {
Chris Mason4184ea72009-03-10 12:39:20 -0400667 rcu_read_unlock();
Josef Bacik0f9dd462008-09-23 13:14:11 -0400668 return found;
Chris Mason4184ea72009-03-10 12:39:20 -0400669 }
Josef Bacik0f9dd462008-09-23 13:14:11 -0400670 }
Chris Mason4184ea72009-03-10 12:39:20 -0400671 rcu_read_unlock();
Josef Bacik0f9dd462008-09-23 13:14:11 -0400672 return NULL;
Chris Mason6324fbf2008-03-24 15:01:59 -0400673}
674
Chris Mason4184ea72009-03-10 12:39:20 -0400675/*
676 * after adding space to the filesystem, we need to clear the full flags
677 * on all the space infos.
678 */
679void btrfs_clear_space_info_full(struct btrfs_fs_info *info)
680{
681 struct list_head *head = &info->space_info;
682 struct btrfs_space_info *found;
683
684 rcu_read_lock();
685 list_for_each_entry_rcu(found, head, list)
686 found->full = 0;
687 rcu_read_unlock();
688}
689
Chris Masone02119d2008-09-05 16:13:11 -0400690/* simple helper to search for an existing extent at a given offset */
Zheng Yan31840ae2008-09-23 13:14:14 -0400691int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len)
Chris Masone02119d2008-09-05 16:13:11 -0400692{
693 int ret;
694 struct btrfs_key key;
Zheng Yan31840ae2008-09-23 13:14:14 -0400695 struct btrfs_path *path;
Chris Masone02119d2008-09-05 16:13:11 -0400696
Zheng Yan31840ae2008-09-23 13:14:14 -0400697 path = btrfs_alloc_path();
Mark Fashehd8926bb2011-07-13 10:38:47 -0700698 if (!path)
699 return -ENOMEM;
700
Chris Masone02119d2008-09-05 16:13:11 -0400701 key.objectid = start;
702 key.offset = len;
Josef Bacik3173a182013-03-07 14:22:04 -0500703 key.type = BTRFS_EXTENT_ITEM_KEY;
Chris Masone02119d2008-09-05 16:13:11 -0400704 ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
705 0, 0);
Josef Bacik3173a182013-03-07 14:22:04 -0500706 if (ret > 0) {
707 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
708 if (key.objectid == start &&
709 key.type == BTRFS_METADATA_ITEM_KEY)
710 ret = 0;
711 }
Zheng Yan31840ae2008-09-23 13:14:14 -0400712 btrfs_free_path(path);
Chris Mason7bb86312007-12-11 09:25:06 -0500713 return ret;
714}
715
Chris Masond8d5f3e2007-12-11 12:42:00 -0500716/*
Josef Bacik3173a182013-03-07 14:22:04 -0500717 * helper function to lookup reference count and flags of a tree block.
Yan, Zhenga22285a2010-05-16 10:48:46 -0400718 *
719 * the head node for delayed ref is used to store the sum of all the
720 * reference count modifications queued up in the rbtree. the head
721 * node may also store the extent flags to set. This way you can check
722 * to see what the reference count and extent flags would be if all of
723 * the delayed refs are not processed.
724 */
725int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans,
726 struct btrfs_root *root, u64 bytenr,
Josef Bacik3173a182013-03-07 14:22:04 -0500727 u64 offset, int metadata, u64 *refs, u64 *flags)
Yan, Zhenga22285a2010-05-16 10:48:46 -0400728{
729 struct btrfs_delayed_ref_head *head;
730 struct btrfs_delayed_ref_root *delayed_refs;
731 struct btrfs_path *path;
732 struct btrfs_extent_item *ei;
733 struct extent_buffer *leaf;
734 struct btrfs_key key;
735 u32 item_size;
736 u64 num_refs;
737 u64 extent_flags;
738 int ret;
739
Josef Bacik3173a182013-03-07 14:22:04 -0500740 /*
741 * If we don't have skinny metadata, don't bother doing anything
742 * different
743 */
744 if (metadata && !btrfs_fs_incompat(root->fs_info, SKINNY_METADATA)) {
745 offset = root->leafsize;
746 metadata = 0;
747 }
748
Yan, Zhenga22285a2010-05-16 10:48:46 -0400749 path = btrfs_alloc_path();
750 if (!path)
751 return -ENOMEM;
752
Josef Bacik3173a182013-03-07 14:22:04 -0500753 if (metadata) {
754 key.objectid = bytenr;
755 key.type = BTRFS_METADATA_ITEM_KEY;
756 key.offset = offset;
757 } else {
758 key.objectid = bytenr;
759 key.type = BTRFS_EXTENT_ITEM_KEY;
760 key.offset = offset;
761 }
762
Yan, Zhenga22285a2010-05-16 10:48:46 -0400763 if (!trans) {
764 path->skip_locking = 1;
765 path->search_commit_root = 1;
766 }
767again:
768 ret = btrfs_search_slot(trans, root->fs_info->extent_root,
769 &key, path, 0, 0);
770 if (ret < 0)
771 goto out_free;
772
Josef Bacik3173a182013-03-07 14:22:04 -0500773 if (ret > 0 && metadata && key.type == BTRFS_METADATA_ITEM_KEY) {
774 key.type = BTRFS_EXTENT_ITEM_KEY;
775 key.offset = root->leafsize;
776 btrfs_release_path(path);
777 goto again;
778 }
779
Yan, Zhenga22285a2010-05-16 10:48:46 -0400780 if (ret == 0) {
781 leaf = path->nodes[0];
782 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
783 if (item_size >= sizeof(*ei)) {
784 ei = btrfs_item_ptr(leaf, path->slots[0],
785 struct btrfs_extent_item);
786 num_refs = btrfs_extent_refs(leaf, ei);
787 extent_flags = btrfs_extent_flags(leaf, ei);
788 } else {
789#ifdef BTRFS_COMPAT_EXTENT_TREE_V0
790 struct btrfs_extent_item_v0 *ei0;
791 BUG_ON(item_size != sizeof(*ei0));
792 ei0 = btrfs_item_ptr(leaf, path->slots[0],
793 struct btrfs_extent_item_v0);
794 num_refs = btrfs_extent_refs_v0(leaf, ei0);
795 /* FIXME: this isn't correct for data */
796 extent_flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
797#else
798 BUG();
799#endif
800 }
801 BUG_ON(num_refs == 0);
802 } else {
803 num_refs = 0;
804 extent_flags = 0;
805 ret = 0;
806 }
807
808 if (!trans)
809 goto out;
810
811 delayed_refs = &trans->transaction->delayed_refs;
812 spin_lock(&delayed_refs->lock);
813 head = btrfs_find_delayed_ref_head(trans, bytenr);
814 if (head) {
815 if (!mutex_trylock(&head->mutex)) {
816 atomic_inc(&head->node.refs);
817 spin_unlock(&delayed_refs->lock);
818
David Sterbab3b4aa72011-04-21 01:20:15 +0200819 btrfs_release_path(path);
Yan, Zhenga22285a2010-05-16 10:48:46 -0400820
David Sterba8cc33e52011-05-02 15:29:25 +0200821 /*
822 * Mutex was contended, block until it's released and try
823 * again
824 */
Yan, Zhenga22285a2010-05-16 10:48:46 -0400825 mutex_lock(&head->mutex);
826 mutex_unlock(&head->mutex);
827 btrfs_put_delayed_ref(&head->node);
828 goto again;
829 }
830 if (head->extent_op && head->extent_op->update_flags)
831 extent_flags |= head->extent_op->flags_to_set;
832 else
833 BUG_ON(num_refs == 0);
834
835 num_refs += head->node.ref_mod;
836 mutex_unlock(&head->mutex);
837 }
838 spin_unlock(&delayed_refs->lock);
839out:
840 WARN_ON(num_refs == 0);
841 if (refs)
842 *refs = num_refs;
843 if (flags)
844 *flags = extent_flags;
845out_free:
846 btrfs_free_path(path);
847 return ret;
848}
849
850/*
Chris Masond8d5f3e2007-12-11 12:42:00 -0500851 * Back reference rules. Back refs have three main goals:
852 *
853 * 1) differentiate between all holders of references to an extent so that
854 * when a reference is dropped we can make sure it was a valid reference
855 * before freeing the extent.
856 *
857 * 2) Provide enough information to quickly find the holders of an extent
858 * if we notice a given block is corrupted or bad.
859 *
860 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
861 * maintenance. This is actually the same as #2, but with a slightly
862 * different use case.
863 *
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400864 * There are two kinds of back refs. The implicit back refs is optimized
865 * for pointers in non-shared tree blocks. For a given pointer in a block,
866 * back refs of this kind provide information about the block's owner tree
867 * and the pointer's key. These information allow us to find the block by
868 * b-tree searching. The full back refs is for pointers in tree blocks not
869 * referenced by their owner trees. The location of tree block is recorded
870 * in the back refs. Actually the full back refs is generic, and can be
871 * used in all cases the implicit back refs is used. The major shortcoming
872 * of the full back refs is its overhead. Every time a tree block gets
873 * COWed, we have to update back refs entry for all pointers in it.
874 *
875 * For a newly allocated tree block, we use implicit back refs for
876 * pointers in it. This means most tree related operations only involve
877 * implicit back refs. For a tree block created in old transaction, the
878 * only way to drop a reference to it is COW it. So we can detect the
879 * event that tree block loses its owner tree's reference and do the
880 * back refs conversion.
881 *
882 * When a tree block is COW'd through a tree, there are four cases:
883 *
884 * The reference count of the block is one and the tree is the block's
885 * owner tree. Nothing to do in this case.
886 *
887 * The reference count of the block is one and the tree is not the
888 * block's owner tree. In this case, full back refs is used for pointers
889 * in the block. Remove these full back refs, add implicit back refs for
890 * every pointers in the new block.
891 *
892 * The reference count of the block is greater than one and the tree is
893 * the block's owner tree. In this case, implicit back refs is used for
894 * pointers in the block. Add full back refs for every pointers in the
895 * block, increase lower level extents' reference counts. The original
896 * implicit back refs are entailed to the new block.
897 *
898 * The reference count of the block is greater than one and the tree is
899 * not the block's owner tree. Add implicit back refs for every pointer in
900 * the new block, increase lower level extents' reference count.
901 *
902 * Back Reference Key composing:
903 *
904 * The key objectid corresponds to the first byte in the extent,
905 * The key type is used to differentiate between types of back refs.
906 * There are different meanings of the key offset for different types
907 * of back refs.
908 *
Chris Masond8d5f3e2007-12-11 12:42:00 -0500909 * File extents can be referenced by:
910 *
911 * - multiple snapshots, subvolumes, or different generations in one subvol
Zheng Yan31840ae2008-09-23 13:14:14 -0400912 * - different files inside a single subvolume
Chris Masond8d5f3e2007-12-11 12:42:00 -0500913 * - different offsets inside a file (bookend extents in file.c)
914 *
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400915 * The extent ref structure for the implicit back refs has fields for:
Chris Masond8d5f3e2007-12-11 12:42:00 -0500916 *
917 * - Objectid of the subvolume root
Chris Masond8d5f3e2007-12-11 12:42:00 -0500918 * - objectid of the file holding the reference
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400919 * - original offset in the file
920 * - how many bookend extents
Zheng Yan31840ae2008-09-23 13:14:14 -0400921 *
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400922 * The key offset for the implicit back refs is hash of the first
923 * three fields.
Chris Masond8d5f3e2007-12-11 12:42:00 -0500924 *
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400925 * The extent ref structure for the full back refs has field for:
Chris Masond8d5f3e2007-12-11 12:42:00 -0500926 *
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400927 * - number of pointers in the tree leaf
Chris Masond8d5f3e2007-12-11 12:42:00 -0500928 *
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400929 * The key offset for the implicit back refs is the first byte of
930 * the tree leaf
Chris Masond8d5f3e2007-12-11 12:42:00 -0500931 *
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400932 * When a file extent is allocated, The implicit back refs is used.
933 * the fields are filled in:
Chris Masond8d5f3e2007-12-11 12:42:00 -0500934 *
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400935 * (root_key.objectid, inode objectid, offset in file, 1)
936 *
937 * When a file extent is removed file truncation, we find the
938 * corresponding implicit back refs and check the following fields:
939 *
940 * (btrfs_header_owner(leaf), inode objectid, offset in file)
Chris Masond8d5f3e2007-12-11 12:42:00 -0500941 *
942 * Btree extents can be referenced by:
943 *
944 * - Different subvolumes
Chris Masond8d5f3e2007-12-11 12:42:00 -0500945 *
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400946 * Both the implicit back refs and the full back refs for tree blocks
947 * only consist of key. The key offset for the implicit back refs is
948 * objectid of block's owner tree. The key offset for the full back refs
949 * is the first byte of parent block.
Chris Masond8d5f3e2007-12-11 12:42:00 -0500950 *
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400951 * When implicit back refs is used, information about the lowest key and
952 * level of the tree block are required. These information are stored in
953 * tree block info structure.
Chris Masond8d5f3e2007-12-11 12:42:00 -0500954 */
Zheng Yan31840ae2008-09-23 13:14:14 -0400955
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400956#ifdef BTRFS_COMPAT_EXTENT_TREE_V0
957static int convert_extent_item_v0(struct btrfs_trans_handle *trans,
958 struct btrfs_root *root,
959 struct btrfs_path *path,
960 u64 owner, u32 extra_size)
Chris Mason74493f72007-12-11 09:25:06 -0500961{
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400962 struct btrfs_extent_item *item;
963 struct btrfs_extent_item_v0 *ei0;
964 struct btrfs_extent_ref_v0 *ref0;
965 struct btrfs_tree_block_info *bi;
Zheng Yan31840ae2008-09-23 13:14:14 -0400966 struct extent_buffer *leaf;
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400967 struct btrfs_key key;
968 struct btrfs_key found_key;
969 u32 new_size = sizeof(*item);
970 u64 refs;
Chris Mason74493f72007-12-11 09:25:06 -0500971 int ret;
972
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400973 leaf = path->nodes[0];
974 BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0));
Chris Mason74493f72007-12-11 09:25:06 -0500975
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400976 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
977 ei0 = btrfs_item_ptr(leaf, path->slots[0],
978 struct btrfs_extent_item_v0);
979 refs = btrfs_extent_refs_v0(leaf, ei0);
980
981 if (owner == (u64)-1) {
982 while (1) {
983 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
984 ret = btrfs_next_leaf(root, path);
985 if (ret < 0)
986 return ret;
Jeff Mahoney79787ea2012-03-12 16:03:00 +0100987 BUG_ON(ret > 0); /* Corruption */
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400988 leaf = path->nodes[0];
989 }
990 btrfs_item_key_to_cpu(leaf, &found_key,
991 path->slots[0]);
992 BUG_ON(key.objectid != found_key.objectid);
993 if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) {
994 path->slots[0]++;
995 continue;
996 }
997 ref0 = btrfs_item_ptr(leaf, path->slots[0],
998 struct btrfs_extent_ref_v0);
999 owner = btrfs_ref_objectid_v0(leaf, ref0);
1000 break;
1001 }
1002 }
David Sterbab3b4aa72011-04-21 01:20:15 +02001003 btrfs_release_path(path);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001004
1005 if (owner < BTRFS_FIRST_FREE_OBJECTID)
1006 new_size += sizeof(*bi);
1007
1008 new_size -= sizeof(*ei0);
1009 ret = btrfs_search_slot(trans, root, &key, path,
1010 new_size + extra_size, 1);
Zheng Yan31840ae2008-09-23 13:14:14 -04001011 if (ret < 0)
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001012 return ret;
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001013 BUG_ON(ret); /* Corruption */
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001014
Tsutomu Itoh4b90c682013-04-16 05:18:49 +00001015 btrfs_extend_item(root, path, new_size);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001016
1017 leaf = path->nodes[0];
1018 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1019 btrfs_set_extent_refs(leaf, item, refs);
1020 /* FIXME: get real generation */
1021 btrfs_set_extent_generation(leaf, item, 0);
1022 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1023 btrfs_set_extent_flags(leaf, item,
1024 BTRFS_EXTENT_FLAG_TREE_BLOCK |
1025 BTRFS_BLOCK_FLAG_FULL_BACKREF);
1026 bi = (struct btrfs_tree_block_info *)(item + 1);
1027 /* FIXME: get first key of the block */
1028 memset_extent_buffer(leaf, 0, (unsigned long)bi, sizeof(*bi));
1029 btrfs_set_tree_block_level(leaf, bi, (int)owner);
1030 } else {
1031 btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA);
1032 }
1033 btrfs_mark_buffer_dirty(leaf);
1034 return 0;
1035}
1036#endif
1037
1038static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset)
1039{
1040 u32 high_crc = ~(u32)0;
1041 u32 low_crc = ~(u32)0;
1042 __le64 lenum;
1043
1044 lenum = cpu_to_le64(root_objectid);
David Woodhouse163e7832009-04-19 13:02:41 +01001045 high_crc = crc32c(high_crc, &lenum, sizeof(lenum));
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001046 lenum = cpu_to_le64(owner);
David Woodhouse163e7832009-04-19 13:02:41 +01001047 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001048 lenum = cpu_to_le64(offset);
David Woodhouse163e7832009-04-19 13:02:41 +01001049 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001050
1051 return ((u64)high_crc << 31) ^ (u64)low_crc;
1052}
1053
1054static u64 hash_extent_data_ref_item(struct extent_buffer *leaf,
1055 struct btrfs_extent_data_ref *ref)
1056{
1057 return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref),
1058 btrfs_extent_data_ref_objectid(leaf, ref),
1059 btrfs_extent_data_ref_offset(leaf, ref));
1060}
1061
1062static int match_extent_data_ref(struct extent_buffer *leaf,
1063 struct btrfs_extent_data_ref *ref,
1064 u64 root_objectid, u64 owner, u64 offset)
1065{
1066 if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid ||
1067 btrfs_extent_data_ref_objectid(leaf, ref) != owner ||
1068 btrfs_extent_data_ref_offset(leaf, ref) != offset)
1069 return 0;
1070 return 1;
1071}
1072
1073static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans,
1074 struct btrfs_root *root,
1075 struct btrfs_path *path,
1076 u64 bytenr, u64 parent,
1077 u64 root_objectid,
1078 u64 owner, u64 offset)
1079{
1080 struct btrfs_key key;
1081 struct btrfs_extent_data_ref *ref;
1082 struct extent_buffer *leaf;
1083 u32 nritems;
1084 int ret;
1085 int recow;
1086 int err = -ENOENT;
1087
1088 key.objectid = bytenr;
1089 if (parent) {
1090 key.type = BTRFS_SHARED_DATA_REF_KEY;
1091 key.offset = parent;
1092 } else {
1093 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1094 key.offset = hash_extent_data_ref(root_objectid,
1095 owner, offset);
1096 }
1097again:
1098 recow = 0;
1099 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1100 if (ret < 0) {
1101 err = ret;
1102 goto fail;
1103 }
1104
1105 if (parent) {
1106 if (!ret)
1107 return 0;
1108#ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1109 key.type = BTRFS_EXTENT_REF_V0_KEY;
David Sterbab3b4aa72011-04-21 01:20:15 +02001110 btrfs_release_path(path);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001111 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1112 if (ret < 0) {
1113 err = ret;
1114 goto fail;
1115 }
1116 if (!ret)
1117 return 0;
1118#endif
1119 goto fail;
Zheng Yan31840ae2008-09-23 13:14:14 -04001120 }
1121
1122 leaf = path->nodes[0];
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001123 nritems = btrfs_header_nritems(leaf);
1124 while (1) {
1125 if (path->slots[0] >= nritems) {
1126 ret = btrfs_next_leaf(root, path);
1127 if (ret < 0)
1128 err = ret;
1129 if (ret)
1130 goto fail;
1131
1132 leaf = path->nodes[0];
1133 nritems = btrfs_header_nritems(leaf);
1134 recow = 1;
1135 }
1136
1137 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1138 if (key.objectid != bytenr ||
1139 key.type != BTRFS_EXTENT_DATA_REF_KEY)
1140 goto fail;
1141
1142 ref = btrfs_item_ptr(leaf, path->slots[0],
1143 struct btrfs_extent_data_ref);
1144
1145 if (match_extent_data_ref(leaf, ref, root_objectid,
1146 owner, offset)) {
1147 if (recow) {
David Sterbab3b4aa72011-04-21 01:20:15 +02001148 btrfs_release_path(path);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001149 goto again;
1150 }
1151 err = 0;
1152 break;
1153 }
1154 path->slots[0]++;
Zheng Yan31840ae2008-09-23 13:14:14 -04001155 }
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001156fail:
1157 return err;
Zheng Yan31840ae2008-09-23 13:14:14 -04001158}
1159
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001160static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans,
1161 struct btrfs_root *root,
1162 struct btrfs_path *path,
1163 u64 bytenr, u64 parent,
1164 u64 root_objectid, u64 owner,
1165 u64 offset, int refs_to_add)
Zheng Yan31840ae2008-09-23 13:14:14 -04001166{
1167 struct btrfs_key key;
1168 struct extent_buffer *leaf;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001169 u32 size;
Zheng Yan31840ae2008-09-23 13:14:14 -04001170 u32 num_refs;
1171 int ret;
1172
1173 key.objectid = bytenr;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001174 if (parent) {
1175 key.type = BTRFS_SHARED_DATA_REF_KEY;
1176 key.offset = parent;
1177 size = sizeof(struct btrfs_shared_data_ref);
Zheng Yan31840ae2008-09-23 13:14:14 -04001178 } else {
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001179 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1180 key.offset = hash_extent_data_ref(root_objectid,
1181 owner, offset);
1182 size = sizeof(struct btrfs_extent_data_ref);
Zheng Yan31840ae2008-09-23 13:14:14 -04001183 }
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001184
1185 ret = btrfs_insert_empty_item(trans, root, path, &key, size);
1186 if (ret && ret != -EEXIST)
1187 goto fail;
1188
1189 leaf = path->nodes[0];
1190 if (parent) {
1191 struct btrfs_shared_data_ref *ref;
1192 ref = btrfs_item_ptr(leaf, path->slots[0],
1193 struct btrfs_shared_data_ref);
1194 if (ret == 0) {
1195 btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add);
1196 } else {
1197 num_refs = btrfs_shared_data_ref_count(leaf, ref);
1198 num_refs += refs_to_add;
1199 btrfs_set_shared_data_ref_count(leaf, ref, num_refs);
1200 }
1201 } else {
1202 struct btrfs_extent_data_ref *ref;
1203 while (ret == -EEXIST) {
1204 ref = btrfs_item_ptr(leaf, path->slots[0],
1205 struct btrfs_extent_data_ref);
1206 if (match_extent_data_ref(leaf, ref, root_objectid,
1207 owner, offset))
1208 break;
David Sterbab3b4aa72011-04-21 01:20:15 +02001209 btrfs_release_path(path);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001210 key.offset++;
1211 ret = btrfs_insert_empty_item(trans, root, path, &key,
1212 size);
1213 if (ret && ret != -EEXIST)
1214 goto fail;
1215
1216 leaf = path->nodes[0];
1217 }
1218 ref = btrfs_item_ptr(leaf, path->slots[0],
1219 struct btrfs_extent_data_ref);
1220 if (ret == 0) {
1221 btrfs_set_extent_data_ref_root(leaf, ref,
1222 root_objectid);
1223 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
1224 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
1225 btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add);
1226 } else {
1227 num_refs = btrfs_extent_data_ref_count(leaf, ref);
1228 num_refs += refs_to_add;
1229 btrfs_set_extent_data_ref_count(leaf, ref, num_refs);
1230 }
1231 }
1232 btrfs_mark_buffer_dirty(leaf);
1233 ret = 0;
1234fail:
David Sterbab3b4aa72011-04-21 01:20:15 +02001235 btrfs_release_path(path);
Chris Mason7bb86312007-12-11 09:25:06 -05001236 return ret;
Chris Mason74493f72007-12-11 09:25:06 -05001237}
1238
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001239static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans,
1240 struct btrfs_root *root,
1241 struct btrfs_path *path,
1242 int refs_to_drop)
Zheng Yan31840ae2008-09-23 13:14:14 -04001243{
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001244 struct btrfs_key key;
1245 struct btrfs_extent_data_ref *ref1 = NULL;
1246 struct btrfs_shared_data_ref *ref2 = NULL;
Zheng Yan31840ae2008-09-23 13:14:14 -04001247 struct extent_buffer *leaf;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001248 u32 num_refs = 0;
Zheng Yan31840ae2008-09-23 13:14:14 -04001249 int ret = 0;
1250
1251 leaf = path->nodes[0];
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001252 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1253
1254 if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1255 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1256 struct btrfs_extent_data_ref);
1257 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1258 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1259 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1260 struct btrfs_shared_data_ref);
1261 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1262#ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1263 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1264 struct btrfs_extent_ref_v0 *ref0;
1265 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1266 struct btrfs_extent_ref_v0);
1267 num_refs = btrfs_ref_count_v0(leaf, ref0);
1268#endif
1269 } else {
1270 BUG();
1271 }
1272
Chris Mason56bec292009-03-13 10:10:06 -04001273 BUG_ON(num_refs < refs_to_drop);
1274 num_refs -= refs_to_drop;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001275
Zheng Yan31840ae2008-09-23 13:14:14 -04001276 if (num_refs == 0) {
1277 ret = btrfs_del_item(trans, root, path);
1278 } else {
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001279 if (key.type == BTRFS_EXTENT_DATA_REF_KEY)
1280 btrfs_set_extent_data_ref_count(leaf, ref1, num_refs);
1281 else if (key.type == BTRFS_SHARED_DATA_REF_KEY)
1282 btrfs_set_shared_data_ref_count(leaf, ref2, num_refs);
1283#ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1284 else {
1285 struct btrfs_extent_ref_v0 *ref0;
1286 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1287 struct btrfs_extent_ref_v0);
1288 btrfs_set_ref_count_v0(leaf, ref0, num_refs);
1289 }
1290#endif
Zheng Yan31840ae2008-09-23 13:14:14 -04001291 btrfs_mark_buffer_dirty(leaf);
1292 }
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001293 return ret;
1294}
1295
1296static noinline u32 extent_data_ref_count(struct btrfs_root *root,
1297 struct btrfs_path *path,
1298 struct btrfs_extent_inline_ref *iref)
1299{
1300 struct btrfs_key key;
1301 struct extent_buffer *leaf;
1302 struct btrfs_extent_data_ref *ref1;
1303 struct btrfs_shared_data_ref *ref2;
1304 u32 num_refs = 0;
1305
1306 leaf = path->nodes[0];
1307 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1308 if (iref) {
1309 if (btrfs_extent_inline_ref_type(leaf, iref) ==
1310 BTRFS_EXTENT_DATA_REF_KEY) {
1311 ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
1312 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1313 } else {
1314 ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
1315 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1316 }
1317 } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1318 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1319 struct btrfs_extent_data_ref);
1320 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1321 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1322 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1323 struct btrfs_shared_data_ref);
1324 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1325#ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1326 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1327 struct btrfs_extent_ref_v0 *ref0;
1328 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1329 struct btrfs_extent_ref_v0);
1330 num_refs = btrfs_ref_count_v0(leaf, ref0);
1331#endif
1332 } else {
1333 WARN_ON(1);
1334 }
1335 return num_refs;
1336}
1337
1338static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
1339 struct btrfs_root *root,
1340 struct btrfs_path *path,
1341 u64 bytenr, u64 parent,
1342 u64 root_objectid)
1343{
1344 struct btrfs_key key;
1345 int ret;
1346
1347 key.objectid = bytenr;
1348 if (parent) {
1349 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1350 key.offset = parent;
1351 } else {
1352 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1353 key.offset = root_objectid;
1354 }
1355
1356 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1357 if (ret > 0)
1358 ret = -ENOENT;
1359#ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1360 if (ret == -ENOENT && parent) {
David Sterbab3b4aa72011-04-21 01:20:15 +02001361 btrfs_release_path(path);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001362 key.type = BTRFS_EXTENT_REF_V0_KEY;
1363 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1364 if (ret > 0)
1365 ret = -ENOENT;
1366 }
1367#endif
1368 return ret;
1369}
1370
1371static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
1372 struct btrfs_root *root,
1373 struct btrfs_path *path,
1374 u64 bytenr, u64 parent,
1375 u64 root_objectid)
1376{
1377 struct btrfs_key key;
1378 int ret;
1379
1380 key.objectid = bytenr;
1381 if (parent) {
1382 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1383 key.offset = parent;
1384 } else {
1385 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1386 key.offset = root_objectid;
1387 }
1388
1389 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
David Sterbab3b4aa72011-04-21 01:20:15 +02001390 btrfs_release_path(path);
Zheng Yan31840ae2008-09-23 13:14:14 -04001391 return ret;
1392}
1393
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001394static inline int extent_ref_type(u64 parent, u64 owner)
1395{
1396 int type;
1397 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1398 if (parent > 0)
1399 type = BTRFS_SHARED_BLOCK_REF_KEY;
1400 else
1401 type = BTRFS_TREE_BLOCK_REF_KEY;
1402 } else {
1403 if (parent > 0)
1404 type = BTRFS_SHARED_DATA_REF_KEY;
1405 else
1406 type = BTRFS_EXTENT_DATA_REF_KEY;
1407 }
1408 return type;
1409}
1410
Yan Zheng2c47e6052009-06-27 21:07:35 -04001411static int find_next_key(struct btrfs_path *path, int level,
1412 struct btrfs_key *key)
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001413
1414{
Yan Zheng2c47e6052009-06-27 21:07:35 -04001415 for (; level < BTRFS_MAX_LEVEL; level++) {
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001416 if (!path->nodes[level])
1417 break;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001418 if (path->slots[level] + 1 >=
1419 btrfs_header_nritems(path->nodes[level]))
1420 continue;
1421 if (level == 0)
1422 btrfs_item_key_to_cpu(path->nodes[level], key,
1423 path->slots[level] + 1);
1424 else
1425 btrfs_node_key_to_cpu(path->nodes[level], key,
1426 path->slots[level] + 1);
1427 return 0;
1428 }
1429 return 1;
1430}
1431
1432/*
1433 * look for inline back ref. if back ref is found, *ref_ret is set
1434 * to the address of inline back ref, and 0 is returned.
1435 *
1436 * if back ref isn't found, *ref_ret is set to the address where it
1437 * should be inserted, and -ENOENT is returned.
1438 *
1439 * if insert is true and there are too many inline back refs, the path
1440 * points to the extent item, and -EAGAIN is returned.
1441 *
1442 * NOTE: inline back refs are ordered in the same way that back ref
1443 * items in the tree are ordered.
1444 */
1445static noinline_for_stack
1446int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
1447 struct btrfs_root *root,
1448 struct btrfs_path *path,
1449 struct btrfs_extent_inline_ref **ref_ret,
1450 u64 bytenr, u64 num_bytes,
1451 u64 parent, u64 root_objectid,
1452 u64 owner, u64 offset, int insert)
1453{
1454 struct btrfs_key key;
1455 struct extent_buffer *leaf;
1456 struct btrfs_extent_item *ei;
1457 struct btrfs_extent_inline_ref *iref;
1458 u64 flags;
1459 u64 item_size;
1460 unsigned long ptr;
1461 unsigned long end;
1462 int extra_size;
1463 int type;
1464 int want;
1465 int ret;
1466 int err = 0;
Josef Bacik3173a182013-03-07 14:22:04 -05001467 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
1468 SKINNY_METADATA);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001469
1470 key.objectid = bytenr;
1471 key.type = BTRFS_EXTENT_ITEM_KEY;
1472 key.offset = num_bytes;
1473
1474 want = extent_ref_type(parent, owner);
1475 if (insert) {
1476 extra_size = btrfs_extent_inline_ref_size(want);
Yan Zheng85d41982009-06-11 08:51:10 -04001477 path->keep_locks = 1;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001478 } else
1479 extra_size = -1;
Josef Bacik3173a182013-03-07 14:22:04 -05001480
1481 /*
1482 * Owner is our parent level, so we can just add one to get the level
1483 * for the block we are interested in.
1484 */
1485 if (skinny_metadata && owner < BTRFS_FIRST_FREE_OBJECTID) {
1486 key.type = BTRFS_METADATA_ITEM_KEY;
1487 key.offset = owner;
1488 }
1489
1490again:
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001491 ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
1492 if (ret < 0) {
1493 err = ret;
1494 goto out;
1495 }
Josef Bacik3173a182013-03-07 14:22:04 -05001496
1497 /*
1498 * We may be a newly converted file system which still has the old fat
1499 * extent entries for metadata, so try and see if we have one of those.
1500 */
1501 if (ret > 0 && skinny_metadata) {
1502 skinny_metadata = false;
1503 if (path->slots[0]) {
1504 path->slots[0]--;
1505 btrfs_item_key_to_cpu(path->nodes[0], &key,
1506 path->slots[0]);
1507 if (key.objectid == bytenr &&
1508 key.type == BTRFS_EXTENT_ITEM_KEY &&
1509 key.offset == num_bytes)
1510 ret = 0;
1511 }
1512 if (ret) {
1513 key.type = BTRFS_EXTENT_ITEM_KEY;
1514 key.offset = num_bytes;
1515 btrfs_release_path(path);
1516 goto again;
1517 }
1518 }
1519
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001520 if (ret && !insert) {
1521 err = -ENOENT;
1522 goto out;
Josef Bacik492104c2013-03-08 15:41:02 -05001523 } else if (ret) {
1524 err = -EIO;
1525 WARN_ON(1);
1526 goto out;
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001527 }
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001528
1529 leaf = path->nodes[0];
1530 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1531#ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1532 if (item_size < sizeof(*ei)) {
1533 if (!insert) {
1534 err = -ENOENT;
1535 goto out;
1536 }
1537 ret = convert_extent_item_v0(trans, root, path, owner,
1538 extra_size);
1539 if (ret < 0) {
1540 err = ret;
1541 goto out;
1542 }
1543 leaf = path->nodes[0];
1544 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1545 }
1546#endif
1547 BUG_ON(item_size < sizeof(*ei));
1548
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001549 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1550 flags = btrfs_extent_flags(leaf, ei);
1551
1552 ptr = (unsigned long)(ei + 1);
1553 end = (unsigned long)ei + item_size;
1554
Josef Bacik3173a182013-03-07 14:22:04 -05001555 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK && !skinny_metadata) {
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001556 ptr += sizeof(struct btrfs_tree_block_info);
1557 BUG_ON(ptr > end);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001558 }
1559
1560 err = -ENOENT;
1561 while (1) {
1562 if (ptr >= end) {
1563 WARN_ON(ptr > end);
1564 break;
1565 }
1566 iref = (struct btrfs_extent_inline_ref *)ptr;
1567 type = btrfs_extent_inline_ref_type(leaf, iref);
1568 if (want < type)
1569 break;
1570 if (want > type) {
1571 ptr += btrfs_extent_inline_ref_size(type);
1572 continue;
1573 }
1574
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 if (match_extent_data_ref(leaf, dref, root_objectid,
1579 owner, offset)) {
1580 err = 0;
1581 break;
1582 }
1583 if (hash_extent_data_ref_item(leaf, dref) <
1584 hash_extent_data_ref(root_objectid, owner, offset))
1585 break;
1586 } else {
1587 u64 ref_offset;
1588 ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1589 if (parent > 0) {
1590 if (parent == ref_offset) {
1591 err = 0;
1592 break;
1593 }
1594 if (ref_offset < parent)
1595 break;
1596 } else {
1597 if (root_objectid == ref_offset) {
1598 err = 0;
1599 break;
1600 }
1601 if (ref_offset < root_objectid)
1602 break;
1603 }
1604 }
1605 ptr += btrfs_extent_inline_ref_size(type);
1606 }
1607 if (err == -ENOENT && insert) {
1608 if (item_size + extra_size >=
1609 BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
1610 err = -EAGAIN;
1611 goto out;
1612 }
1613 /*
1614 * To add new inline back ref, we have to make sure
1615 * there is no corresponding back ref item.
1616 * For simplicity, we just do not add new inline back
1617 * ref if there is any kind of item for this block
1618 */
Yan Zheng2c47e6052009-06-27 21:07:35 -04001619 if (find_next_key(path, 0, &key) == 0 &&
1620 key.objectid == bytenr &&
Yan Zheng85d41982009-06-11 08:51:10 -04001621 key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001622 err = -EAGAIN;
1623 goto out;
1624 }
1625 }
1626 *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
1627out:
Yan Zheng85d41982009-06-11 08:51:10 -04001628 if (insert) {
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001629 path->keep_locks = 0;
1630 btrfs_unlock_up_safe(path, 1);
1631 }
1632 return err;
1633}
1634
1635/*
1636 * helper to add new inline back ref
1637 */
1638static noinline_for_stack
Tsutomu Itohfd279fa2013-04-16 05:19:11 +00001639void setup_inline_extent_backref(struct btrfs_root *root,
Jeff Mahoney143bede2012-03-01 14:56:26 +01001640 struct btrfs_path *path,
1641 struct btrfs_extent_inline_ref *iref,
1642 u64 parent, u64 root_objectid,
1643 u64 owner, u64 offset, int refs_to_add,
1644 struct btrfs_delayed_extent_op *extent_op)
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001645{
1646 struct extent_buffer *leaf;
1647 struct btrfs_extent_item *ei;
1648 unsigned long ptr;
1649 unsigned long end;
1650 unsigned long item_offset;
1651 u64 refs;
1652 int size;
1653 int type;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001654
1655 leaf = path->nodes[0];
1656 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1657 item_offset = (unsigned long)iref - (unsigned long)ei;
1658
1659 type = extent_ref_type(parent, owner);
1660 size = btrfs_extent_inline_ref_size(type);
1661
Tsutomu Itoh4b90c682013-04-16 05:18:49 +00001662 btrfs_extend_item(root, path, size);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001663
1664 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1665 refs = btrfs_extent_refs(leaf, ei);
1666 refs += refs_to_add;
1667 btrfs_set_extent_refs(leaf, ei, refs);
1668 if (extent_op)
1669 __run_delayed_extent_op(extent_op, leaf, ei);
1670
1671 ptr = (unsigned long)ei + item_offset;
1672 end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
1673 if (ptr < end - size)
1674 memmove_extent_buffer(leaf, ptr + size, ptr,
1675 end - size - ptr);
1676
1677 iref = (struct btrfs_extent_inline_ref *)ptr;
1678 btrfs_set_extent_inline_ref_type(leaf, iref, type);
1679 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1680 struct btrfs_extent_data_ref *dref;
1681 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1682 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1683 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1684 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1685 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1686 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1687 struct btrfs_shared_data_ref *sref;
1688 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1689 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1690 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1691 } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1692 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1693 } else {
1694 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1695 }
1696 btrfs_mark_buffer_dirty(leaf);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001697}
1698
1699static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1700 struct btrfs_root *root,
1701 struct btrfs_path *path,
1702 struct btrfs_extent_inline_ref **ref_ret,
1703 u64 bytenr, u64 num_bytes, u64 parent,
1704 u64 root_objectid, u64 owner, u64 offset)
1705{
1706 int ret;
1707
1708 ret = lookup_inline_extent_backref(trans, root, path, ref_ret,
1709 bytenr, num_bytes, parent,
1710 root_objectid, owner, offset, 0);
1711 if (ret != -ENOENT)
1712 return ret;
1713
David Sterbab3b4aa72011-04-21 01:20:15 +02001714 btrfs_release_path(path);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001715 *ref_ret = NULL;
1716
1717 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1718 ret = lookup_tree_block_ref(trans, root, path, bytenr, parent,
1719 root_objectid);
1720 } else {
1721 ret = lookup_extent_data_ref(trans, root, path, bytenr, parent,
1722 root_objectid, owner, offset);
1723 }
1724 return ret;
1725}
1726
1727/*
1728 * helper to update/remove inline back ref
1729 */
1730static noinline_for_stack
Tsutomu Itohafe5fea2013-04-16 05:18:22 +00001731void update_inline_extent_backref(struct btrfs_root *root,
Jeff Mahoney143bede2012-03-01 14:56:26 +01001732 struct btrfs_path *path,
1733 struct btrfs_extent_inline_ref *iref,
1734 int refs_to_mod,
1735 struct btrfs_delayed_extent_op *extent_op)
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001736{
1737 struct extent_buffer *leaf;
1738 struct btrfs_extent_item *ei;
1739 struct btrfs_extent_data_ref *dref = NULL;
1740 struct btrfs_shared_data_ref *sref = NULL;
1741 unsigned long ptr;
1742 unsigned long end;
1743 u32 item_size;
1744 int size;
1745 int type;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001746 u64 refs;
1747
1748 leaf = path->nodes[0];
1749 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1750 refs = btrfs_extent_refs(leaf, ei);
1751 WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1752 refs += refs_to_mod;
1753 btrfs_set_extent_refs(leaf, ei, refs);
1754 if (extent_op)
1755 __run_delayed_extent_op(extent_op, leaf, ei);
1756
1757 type = btrfs_extent_inline_ref_type(leaf, iref);
1758
1759 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1760 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1761 refs = btrfs_extent_data_ref_count(leaf, dref);
1762 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1763 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1764 refs = btrfs_shared_data_ref_count(leaf, sref);
1765 } else {
1766 refs = 1;
1767 BUG_ON(refs_to_mod != -1);
1768 }
1769
1770 BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1771 refs += refs_to_mod;
1772
1773 if (refs > 0) {
1774 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1775 btrfs_set_extent_data_ref_count(leaf, dref, refs);
1776 else
1777 btrfs_set_shared_data_ref_count(leaf, sref, refs);
1778 } else {
1779 size = btrfs_extent_inline_ref_size(type);
1780 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1781 ptr = (unsigned long)iref;
1782 end = (unsigned long)ei + item_size;
1783 if (ptr + size < end)
1784 memmove_extent_buffer(leaf, ptr, ptr + size,
1785 end - ptr - size);
1786 item_size -= size;
Tsutomu Itohafe5fea2013-04-16 05:18:22 +00001787 btrfs_truncate_item(root, path, item_size, 1);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001788 }
1789 btrfs_mark_buffer_dirty(leaf);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001790}
1791
1792static noinline_for_stack
1793int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1794 struct btrfs_root *root,
1795 struct btrfs_path *path,
1796 u64 bytenr, u64 num_bytes, u64 parent,
1797 u64 root_objectid, u64 owner,
1798 u64 offset, int refs_to_add,
1799 struct btrfs_delayed_extent_op *extent_op)
1800{
1801 struct btrfs_extent_inline_ref *iref;
1802 int ret;
1803
1804 ret = lookup_inline_extent_backref(trans, root, path, &iref,
1805 bytenr, num_bytes, parent,
1806 root_objectid, owner, offset, 1);
1807 if (ret == 0) {
1808 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID);
Tsutomu Itohafe5fea2013-04-16 05:18:22 +00001809 update_inline_extent_backref(root, path, iref,
Jeff Mahoney143bede2012-03-01 14:56:26 +01001810 refs_to_add, extent_op);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001811 } else if (ret == -ENOENT) {
Tsutomu Itohfd279fa2013-04-16 05:19:11 +00001812 setup_inline_extent_backref(root, path, iref, parent,
Jeff Mahoney143bede2012-03-01 14:56:26 +01001813 root_objectid, owner, offset,
1814 refs_to_add, extent_op);
1815 ret = 0;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001816 }
1817 return ret;
1818}
1819
1820static int insert_extent_backref(struct btrfs_trans_handle *trans,
1821 struct btrfs_root *root,
1822 struct btrfs_path *path,
1823 u64 bytenr, u64 parent, u64 root_objectid,
1824 u64 owner, u64 offset, int refs_to_add)
1825{
1826 int ret;
1827 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1828 BUG_ON(refs_to_add != 1);
1829 ret = insert_tree_block_ref(trans, root, path, bytenr,
1830 parent, root_objectid);
1831 } else {
1832 ret = insert_extent_data_ref(trans, root, path, bytenr,
1833 parent, root_objectid,
1834 owner, offset, refs_to_add);
1835 }
1836 return ret;
1837}
1838
1839static int remove_extent_backref(struct btrfs_trans_handle *trans,
1840 struct btrfs_root *root,
1841 struct btrfs_path *path,
1842 struct btrfs_extent_inline_ref *iref,
1843 int refs_to_drop, int is_data)
1844{
Jeff Mahoney143bede2012-03-01 14:56:26 +01001845 int ret = 0;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001846
1847 BUG_ON(!is_data && refs_to_drop != 1);
1848 if (iref) {
Tsutomu Itohafe5fea2013-04-16 05:18:22 +00001849 update_inline_extent_backref(root, path, iref,
Jeff Mahoney143bede2012-03-01 14:56:26 +01001850 -refs_to_drop, NULL);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001851 } else if (is_data) {
1852 ret = remove_extent_data_ref(trans, root, path, refs_to_drop);
1853 } else {
1854 ret = btrfs_del_item(trans, root, path);
1855 }
1856 return ret;
1857}
1858
Li Dongyang5378e602011-03-24 10:24:27 +00001859static int btrfs_issue_discard(struct block_device *bdev,
Chris Mason15916de2008-11-19 21:17:22 -05001860 u64 start, u64 len)
1861{
Li Dongyang5378e602011-03-24 10:24:27 +00001862 return blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_NOFS, 0);
Chris Mason15916de2008-11-19 21:17:22 -05001863}
Chris Mason15916de2008-11-19 21:17:22 -05001864
Liu Hui1f3c79a2009-01-05 15:57:51 -05001865static int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr,
Li Dongyang5378e602011-03-24 10:24:27 +00001866 u64 num_bytes, u64 *actual_bytes)
Liu Hui1f3c79a2009-01-05 15:57:51 -05001867{
Liu Hui1f3c79a2009-01-05 15:57:51 -05001868 int ret;
Li Dongyang5378e602011-03-24 10:24:27 +00001869 u64 discarded_bytes = 0;
Jan Schmidta1d3c472011-08-04 17:15:33 +02001870 struct btrfs_bio *bbio = NULL;
Liu Hui1f3c79a2009-01-05 15:57:51 -05001871
Christoph Hellwige244a0a2009-10-14 09:24:59 -04001872
Liu Hui1f3c79a2009-01-05 15:57:51 -05001873 /* Tell the block device(s) that the sectors can be discarded */
Stefan Behrens3ec706c2012-11-05 15:46:42 +01001874 ret = btrfs_map_block(root->fs_info, REQ_DISCARD,
Jan Schmidta1d3c472011-08-04 17:15:33 +02001875 bytenr, &num_bytes, &bbio, 0);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001876 /* Error condition is -ENOMEM */
Liu Hui1f3c79a2009-01-05 15:57:51 -05001877 if (!ret) {
Jan Schmidta1d3c472011-08-04 17:15:33 +02001878 struct btrfs_bio_stripe *stripe = bbio->stripes;
Liu Hui1f3c79a2009-01-05 15:57:51 -05001879 int i;
1880
Liu Hui1f3c79a2009-01-05 15:57:51 -05001881
Jan Schmidta1d3c472011-08-04 17:15:33 +02001882 for (i = 0; i < bbio->num_stripes; i++, stripe++) {
Josef Bacikd5e20032011-08-04 14:52:27 +00001883 if (!stripe->dev->can_discard)
1884 continue;
1885
Li Dongyang5378e602011-03-24 10:24:27 +00001886 ret = btrfs_issue_discard(stripe->dev->bdev,
1887 stripe->physical,
1888 stripe->length);
1889 if (!ret)
1890 discarded_bytes += stripe->length;
1891 else if (ret != -EOPNOTSUPP)
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001892 break; /* Logic errors or -ENOMEM, or -EIO but I don't know how that could happen JDM */
Josef Bacikd5e20032011-08-04 14:52:27 +00001893
1894 /*
1895 * Just in case we get back EOPNOTSUPP for some reason,
1896 * just ignore the return value so we don't screw up
1897 * people calling discard_extent.
1898 */
1899 ret = 0;
Liu Hui1f3c79a2009-01-05 15:57:51 -05001900 }
Jan Schmidta1d3c472011-08-04 17:15:33 +02001901 kfree(bbio);
Liu Hui1f3c79a2009-01-05 15:57:51 -05001902 }
Li Dongyang5378e602011-03-24 10:24:27 +00001903
1904 if (actual_bytes)
1905 *actual_bytes = discarded_bytes;
1906
Liu Hui1f3c79a2009-01-05 15:57:51 -05001907
David Woodhouse53b381b2013-01-29 18:40:14 -05001908 if (ret == -EOPNOTSUPP)
1909 ret = 0;
Liu Hui1f3c79a2009-01-05 15:57:51 -05001910 return ret;
Liu Hui1f3c79a2009-01-05 15:57:51 -05001911}
1912
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001913/* Can return -ENOMEM */
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001914int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1915 struct btrfs_root *root,
1916 u64 bytenr, u64 num_bytes, u64 parent,
Arne Jansen66d7e7f2011-09-12 15:26:38 +02001917 u64 root_objectid, u64 owner, u64 offset, int for_cow)
Zheng Yan31840ae2008-09-23 13:14:14 -04001918{
1919 int ret;
Arne Jansen66d7e7f2011-09-12 15:26:38 +02001920 struct btrfs_fs_info *fs_info = root->fs_info;
1921
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001922 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID &&
1923 root_objectid == BTRFS_TREE_LOG_OBJECTID);
Zheng Yan31840ae2008-09-23 13:14:14 -04001924
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001925 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
Arne Jansen66d7e7f2011-09-12 15:26:38 +02001926 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
1927 num_bytes,
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001928 parent, root_objectid, (int)owner,
Arne Jansen66d7e7f2011-09-12 15:26:38 +02001929 BTRFS_ADD_DELAYED_REF, NULL, for_cow);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001930 } else {
Arne Jansen66d7e7f2011-09-12 15:26:38 +02001931 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
1932 num_bytes,
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001933 parent, root_objectid, owner, offset,
Arne Jansen66d7e7f2011-09-12 15:26:38 +02001934 BTRFS_ADD_DELAYED_REF, NULL, for_cow);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001935 }
Zheng Yan31840ae2008-09-23 13:14:14 -04001936 return ret;
1937}
1938
Chris Mason925baed2008-06-25 16:01:30 -04001939static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001940 struct btrfs_root *root,
1941 u64 bytenr, u64 num_bytes,
1942 u64 parent, u64 root_objectid,
1943 u64 owner, u64 offset, int refs_to_add,
1944 struct btrfs_delayed_extent_op *extent_op)
Chris Mason56bec292009-03-13 10:10:06 -04001945{
Chris Mason5caf2a02007-04-02 11:20:42 -04001946 struct btrfs_path *path;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001947 struct extent_buffer *leaf;
Chris Mason234b63a2007-03-13 10:46:10 -04001948 struct btrfs_extent_item *item;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001949 u64 refs;
1950 int ret;
1951 int err = 0;
Chris Mason037e6392007-03-07 11:50:24 -05001952
Chris Mason5caf2a02007-04-02 11:20:42 -04001953 path = btrfs_alloc_path();
Chris Mason54aa1f42007-06-22 14:16:25 -04001954 if (!path)
1955 return -ENOMEM;
Chris Mason26b80032007-08-08 20:17:12 -04001956
Chris Mason3c12ac72008-04-21 12:01:38 -04001957 path->reada = 1;
Chris Masonb9473432009-03-13 11:00:37 -04001958 path->leave_spinning = 1;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001959 /* this will setup the path even if it fails to insert the back ref */
1960 ret = insert_inline_extent_backref(trans, root->fs_info->extent_root,
1961 path, bytenr, num_bytes, parent,
1962 root_objectid, owner, offset,
1963 refs_to_add, extent_op);
1964 if (ret == 0)
1965 goto out;
Zheng Yan31840ae2008-09-23 13:14:14 -04001966
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001967 if (ret != -EAGAIN) {
1968 err = ret;
1969 goto out;
Chris Masonb9473432009-03-13 11:00:37 -04001970 }
Zheng Yan31840ae2008-09-23 13:14:14 -04001971
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001972 leaf = path->nodes[0];
1973 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1974 refs = btrfs_extent_refs(leaf, item);
1975 btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
1976 if (extent_op)
1977 __run_delayed_extent_op(extent_op, leaf, item);
Zheng Yan31840ae2008-09-23 13:14:14 -04001978
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001979 btrfs_mark_buffer_dirty(leaf);
David Sterbab3b4aa72011-04-21 01:20:15 +02001980 btrfs_release_path(path);
Chris Mason7bb86312007-12-11 09:25:06 -05001981
Chris Mason3c12ac72008-04-21 12:01:38 -04001982 path->reada = 1;
Chris Masonb9473432009-03-13 11:00:37 -04001983 path->leave_spinning = 1;
1984
Chris Mason56bec292009-03-13 10:10:06 -04001985 /* now insert the actual backref */
Zheng Yan31840ae2008-09-23 13:14:14 -04001986 ret = insert_extent_backref(trans, root->fs_info->extent_root,
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001987 path, bytenr, parent, root_objectid,
1988 owner, offset, refs_to_add);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001989 if (ret)
1990 btrfs_abort_transaction(trans, root, ret);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001991out:
Chris Mason74493f72007-12-11 09:25:06 -05001992 btrfs_free_path(path);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001993 return err;
Chris Mason02217ed2007-03-02 16:08:05 -05001994}
1995
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001996static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
1997 struct btrfs_root *root,
1998 struct btrfs_delayed_ref_node *node,
1999 struct btrfs_delayed_extent_op *extent_op,
2000 int insert_reserved)
Chris Masone9d0b132007-08-10 14:06:19 -04002001{
Chris Mason56bec292009-03-13 10:10:06 -04002002 int ret = 0;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002003 struct btrfs_delayed_data_ref *ref;
2004 struct btrfs_key ins;
2005 u64 parent = 0;
2006 u64 ref_root = 0;
2007 u64 flags = 0;
Chris Mason56bec292009-03-13 10:10:06 -04002008
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002009 ins.objectid = node->bytenr;
2010 ins.offset = node->num_bytes;
2011 ins.type = BTRFS_EXTENT_ITEM_KEY;
Chris Mason56bec292009-03-13 10:10:06 -04002012
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002013 ref = btrfs_delayed_node_to_data_ref(node);
2014 if (node->type == BTRFS_SHARED_DATA_REF_KEY)
2015 parent = ref->parent;
2016 else
2017 ref_root = ref->root;
2018
2019 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
Josef Bacik3173a182013-03-07 14:22:04 -05002020 if (extent_op)
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002021 flags |= extent_op->flags_to_set;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002022 ret = alloc_reserved_file_extent(trans, root,
2023 parent, ref_root, flags,
2024 ref->objectid, ref->offset,
2025 &ins, node->ref_mod);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002026 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2027 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
2028 node->num_bytes, parent,
2029 ref_root, ref->objectid,
2030 ref->offset, node->ref_mod,
2031 extent_op);
2032 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2033 ret = __btrfs_free_extent(trans, root, node->bytenr,
2034 node->num_bytes, parent,
2035 ref_root, ref->objectid,
2036 ref->offset, node->ref_mod,
2037 extent_op);
2038 } else {
2039 BUG();
2040 }
Chris Mason56bec292009-03-13 10:10:06 -04002041 return ret;
2042}
2043
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002044static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
2045 struct extent_buffer *leaf,
2046 struct btrfs_extent_item *ei)
2047{
2048 u64 flags = btrfs_extent_flags(leaf, ei);
2049 if (extent_op->update_flags) {
2050 flags |= extent_op->flags_to_set;
2051 btrfs_set_extent_flags(leaf, ei, flags);
2052 }
2053
2054 if (extent_op->update_key) {
2055 struct btrfs_tree_block_info *bi;
2056 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
2057 bi = (struct btrfs_tree_block_info *)(ei + 1);
2058 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
2059 }
2060}
2061
2062static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
2063 struct btrfs_root *root,
2064 struct btrfs_delayed_ref_node *node,
2065 struct btrfs_delayed_extent_op *extent_op)
2066{
2067 struct btrfs_key key;
2068 struct btrfs_path *path;
2069 struct btrfs_extent_item *ei;
2070 struct extent_buffer *leaf;
2071 u32 item_size;
2072 int ret;
2073 int err = 0;
Josef Bacikb1c79e02013-05-09 13:49:30 -04002074 int metadata = !extent_op->is_data;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002075
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002076 if (trans->aborted)
2077 return 0;
2078
Josef Bacik3173a182013-03-07 14:22:04 -05002079 if (metadata && !btrfs_fs_incompat(root->fs_info, SKINNY_METADATA))
2080 metadata = 0;
2081
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002082 path = btrfs_alloc_path();
2083 if (!path)
2084 return -ENOMEM;
2085
2086 key.objectid = node->bytenr;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002087
Josef Bacik3173a182013-03-07 14:22:04 -05002088 if (metadata) {
Josef Bacik3173a182013-03-07 14:22:04 -05002089 key.type = BTRFS_METADATA_ITEM_KEY;
Josef Bacikb1c79e02013-05-09 13:49:30 -04002090 key.offset = extent_op->level;
Josef Bacik3173a182013-03-07 14:22:04 -05002091 } else {
2092 key.type = BTRFS_EXTENT_ITEM_KEY;
2093 key.offset = node->num_bytes;
2094 }
2095
2096again:
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002097 path->reada = 1;
2098 path->leave_spinning = 1;
2099 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key,
2100 path, 0, 1);
2101 if (ret < 0) {
2102 err = ret;
2103 goto out;
2104 }
2105 if (ret > 0) {
Josef Bacik3173a182013-03-07 14:22:04 -05002106 if (metadata) {
2107 btrfs_release_path(path);
2108 metadata = 0;
2109
2110 key.offset = node->num_bytes;
2111 key.type = BTRFS_EXTENT_ITEM_KEY;
2112 goto again;
2113 }
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002114 err = -EIO;
2115 goto out;
2116 }
2117
2118 leaf = path->nodes[0];
2119 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2120#ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2121 if (item_size < sizeof(*ei)) {
2122 ret = convert_extent_item_v0(trans, root->fs_info->extent_root,
2123 path, (u64)-1, 0);
2124 if (ret < 0) {
2125 err = ret;
2126 goto out;
2127 }
2128 leaf = path->nodes[0];
2129 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2130 }
2131#endif
2132 BUG_ON(item_size < sizeof(*ei));
2133 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2134 __run_delayed_extent_op(extent_op, leaf, ei);
2135
2136 btrfs_mark_buffer_dirty(leaf);
2137out:
2138 btrfs_free_path(path);
2139 return err;
2140}
2141
2142static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
2143 struct btrfs_root *root,
2144 struct btrfs_delayed_ref_node *node,
2145 struct btrfs_delayed_extent_op *extent_op,
2146 int insert_reserved)
2147{
2148 int ret = 0;
2149 struct btrfs_delayed_tree_ref *ref;
2150 struct btrfs_key ins;
2151 u64 parent = 0;
2152 u64 ref_root = 0;
Josef Bacik3173a182013-03-07 14:22:04 -05002153 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
2154 SKINNY_METADATA);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002155
2156 ref = btrfs_delayed_node_to_tree_ref(node);
2157 if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2158 parent = ref->parent;
2159 else
2160 ref_root = ref->root;
2161
Josef Bacik3173a182013-03-07 14:22:04 -05002162 ins.objectid = node->bytenr;
2163 if (skinny_metadata) {
2164 ins.offset = ref->level;
2165 ins.type = BTRFS_METADATA_ITEM_KEY;
2166 } else {
2167 ins.offset = node->num_bytes;
2168 ins.type = BTRFS_EXTENT_ITEM_KEY;
2169 }
2170
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002171 BUG_ON(node->ref_mod != 1);
2172 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
Josef Bacik3173a182013-03-07 14:22:04 -05002173 BUG_ON(!extent_op || !extent_op->update_flags);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002174 ret = alloc_reserved_tree_block(trans, root,
2175 parent, ref_root,
2176 extent_op->flags_to_set,
2177 &extent_op->key,
2178 ref->level, &ins);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002179 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2180 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
2181 node->num_bytes, parent, ref_root,
2182 ref->level, 0, 1, extent_op);
2183 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2184 ret = __btrfs_free_extent(trans, root, node->bytenr,
2185 node->num_bytes, parent, ref_root,
2186 ref->level, 0, 1, extent_op);
2187 } else {
2188 BUG();
2189 }
2190 return ret;
2191}
2192
Chris Mason56bec292009-03-13 10:10:06 -04002193/* helper function to actually process a single delayed ref entry */
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002194static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
2195 struct btrfs_root *root,
2196 struct btrfs_delayed_ref_node *node,
2197 struct btrfs_delayed_extent_op *extent_op,
2198 int insert_reserved)
Chris Mason56bec292009-03-13 10:10:06 -04002199{
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002200 int ret = 0;
2201
2202 if (trans->aborted)
2203 return 0;
2204
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002205 if (btrfs_delayed_ref_is_head(node)) {
Chris Mason56bec292009-03-13 10:10:06 -04002206 struct btrfs_delayed_ref_head *head;
2207 /*
2208 * we've hit the end of the chain and we were supposed
2209 * to insert this extent into the tree. But, it got
2210 * deleted before we ever needed to insert it, so all
2211 * we have to do is clean up the accounting
2212 */
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002213 BUG_ON(extent_op);
2214 head = btrfs_delayed_node_to_head(node);
Chris Mason56bec292009-03-13 10:10:06 -04002215 if (insert_reserved) {
Yan, Zhengf0486c62010-05-16 10:46:25 -04002216 btrfs_pin_extent(root, node->bytenr,
2217 node->num_bytes, 1);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002218 if (head->is_data) {
2219 ret = btrfs_del_csums(trans, root,
2220 node->bytenr,
2221 node->num_bytes);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002222 }
Chris Mason56bec292009-03-13 10:10:06 -04002223 }
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002224 return ret;
Chris Mason56bec292009-03-13 10:10:06 -04002225 }
Josef Bacikeb099672009-02-12 09:27:38 -05002226
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002227 if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
2228 node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2229 ret = run_delayed_tree_ref(trans, root, node, extent_op,
2230 insert_reserved);
2231 else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
2232 node->type == BTRFS_SHARED_DATA_REF_KEY)
2233 ret = run_delayed_data_ref(trans, root, node, extent_op,
2234 insert_reserved);
2235 else
2236 BUG();
2237 return ret;
Chris Masone9d0b132007-08-10 14:06:19 -04002238}
2239
Chris Mason56bec292009-03-13 10:10:06 -04002240static noinline struct btrfs_delayed_ref_node *
2241select_delayed_ref(struct btrfs_delayed_ref_head *head)
Chris Masona28ec192007-03-06 20:08:01 -05002242{
Chris Mason56bec292009-03-13 10:10:06 -04002243 struct rb_node *node;
2244 struct btrfs_delayed_ref_node *ref;
2245 int action = BTRFS_ADD_DELAYED_REF;
2246again:
2247 /*
2248 * select delayed ref of type BTRFS_ADD_DELAYED_REF first.
2249 * this prevents ref count from going down to zero when
2250 * there still are pending delayed ref.
2251 */
2252 node = rb_prev(&head->node.rb_node);
2253 while (1) {
2254 if (!node)
2255 break;
2256 ref = rb_entry(node, struct btrfs_delayed_ref_node,
2257 rb_node);
2258 if (ref->bytenr != head->node.bytenr)
2259 break;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002260 if (ref->action == action)
Chris Mason56bec292009-03-13 10:10:06 -04002261 return ref;
2262 node = rb_prev(node);
Chris Mason5f39d392007-10-15 16:14:19 -04002263 }
Chris Mason56bec292009-03-13 10:10:06 -04002264 if (action == BTRFS_ADD_DELAYED_REF) {
2265 action = BTRFS_DROP_DELAYED_REF;
2266 goto again;
2267 }
2268 return NULL;
2269}
2270
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002271/*
2272 * Returns 0 on success or if called with an already aborted transaction.
2273 * Returns -ENOMEM or -EIO on failure and will abort the transaction.
2274 */
Chris Masonc3e69d52009-03-13 10:17:05 -04002275static noinline int run_clustered_refs(struct btrfs_trans_handle *trans,
2276 struct btrfs_root *root,
2277 struct list_head *cluster)
Chris Mason56bec292009-03-13 10:10:06 -04002278{
Chris Mason56bec292009-03-13 10:10:06 -04002279 struct btrfs_delayed_ref_root *delayed_refs;
2280 struct btrfs_delayed_ref_node *ref;
2281 struct btrfs_delayed_ref_head *locked_ref = NULL;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002282 struct btrfs_delayed_extent_op *extent_op;
Jan Schmidt097b8a72012-06-21 11:08:04 +02002283 struct btrfs_fs_info *fs_info = root->fs_info;
Chris Mason56bec292009-03-13 10:10:06 -04002284 int ret;
Chris Masonc3e69d52009-03-13 10:17:05 -04002285 int count = 0;
Chris Mason56bec292009-03-13 10:10:06 -04002286 int must_insert_reserved = 0;
Chris Mason56bec292009-03-13 10:10:06 -04002287
2288 delayed_refs = &trans->transaction->delayed_refs;
Chris Mason56bec292009-03-13 10:10:06 -04002289 while (1) {
2290 if (!locked_ref) {
Chris Masonc3e69d52009-03-13 10:17:05 -04002291 /* pick a new head ref from the cluster list */
2292 if (list_empty(cluster))
Chris Mason56bec292009-03-13 10:10:06 -04002293 break;
Chris Mason56bec292009-03-13 10:10:06 -04002294
Chris Masonc3e69d52009-03-13 10:17:05 -04002295 locked_ref = list_entry(cluster->next,
2296 struct btrfs_delayed_ref_head, cluster);
2297
2298 /* grab the lock that says we are going to process
2299 * all the refs for this head */
2300 ret = btrfs_delayed_ref_lock(trans, locked_ref);
2301
2302 /*
2303 * we may have dropped the spin lock to get the head
2304 * mutex lock, and that might have given someone else
2305 * time to free the head. If that's true, it has been
2306 * removed from our list and we can move on.
2307 */
2308 if (ret == -EAGAIN) {
2309 locked_ref = NULL;
2310 count++;
2311 continue;
Chris Mason56bec292009-03-13 10:10:06 -04002312 }
2313 }
2314
2315 /*
Josef Bacikae1e2062012-08-07 16:00:32 -04002316 * We need to try and merge add/drops of the same ref since we
2317 * can run into issues with relocate dropping the implicit ref
2318 * and then it being added back again before the drop can
2319 * finish. If we merged anything we need to re-loop so we can
2320 * get a good ref.
2321 */
2322 btrfs_merge_delayed_refs(trans, fs_info, delayed_refs,
2323 locked_ref);
2324
2325 /*
Arne Jansend1270cd2011-09-13 15:16:43 +02002326 * locked_ref is the head node, so we have to go one
2327 * node back for any delayed ref updates
2328 */
2329 ref = select_delayed_ref(locked_ref);
2330
2331 if (ref && ref->seq &&
Jan Schmidt097b8a72012-06-21 11:08:04 +02002332 btrfs_check_delayed_seq(fs_info, delayed_refs, ref->seq)) {
Arne Jansend1270cd2011-09-13 15:16:43 +02002333 /*
2334 * there are still refs with lower seq numbers in the
2335 * process of being added. Don't run this ref yet.
2336 */
2337 list_del_init(&locked_ref->cluster);
Miao Xie093486c2012-12-19 08:10:10 +00002338 btrfs_delayed_ref_unlock(locked_ref);
Arne Jansend1270cd2011-09-13 15:16:43 +02002339 locked_ref = NULL;
2340 delayed_refs->num_heads_ready++;
2341 spin_unlock(&delayed_refs->lock);
2342 cond_resched();
2343 spin_lock(&delayed_refs->lock);
2344 continue;
2345 }
2346
2347 /*
Chris Mason56bec292009-03-13 10:10:06 -04002348 * record the must insert reserved flag before we
2349 * drop the spin lock.
2350 */
2351 must_insert_reserved = locked_ref->must_insert_reserved;
2352 locked_ref->must_insert_reserved = 0;
2353
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002354 extent_op = locked_ref->extent_op;
2355 locked_ref->extent_op = NULL;
2356
Chris Mason56bec292009-03-13 10:10:06 -04002357 if (!ref) {
2358 /* All delayed refs have been processed, Go ahead
2359 * and send the head node to run_one_delayed_ref,
2360 * so that any accounting fixes can happen
2361 */
2362 ref = &locked_ref->node;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002363
2364 if (extent_op && must_insert_reserved) {
Miao Xie78a61842012-11-21 02:21:28 +00002365 btrfs_free_delayed_extent_op(extent_op);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002366 extent_op = NULL;
2367 }
2368
2369 if (extent_op) {
2370 spin_unlock(&delayed_refs->lock);
2371
2372 ret = run_delayed_extent_op(trans, root,
2373 ref, extent_op);
Miao Xie78a61842012-11-21 02:21:28 +00002374 btrfs_free_delayed_extent_op(extent_op);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002375
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002376 if (ret) {
Simon Kirbyc2cf52e2013-03-19 22:41:23 +00002377 btrfs_debug(fs_info, "run_delayed_extent_op returned %d", ret);
Dan Carpenter253beeb2012-04-18 09:59:03 +03002378 spin_lock(&delayed_refs->lock);
Miao Xie093486c2012-12-19 08:10:10 +00002379 btrfs_delayed_ref_unlock(locked_ref);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002380 return ret;
2381 }
2382
Chris Mason203bf282012-01-06 15:23:57 -05002383 goto next;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002384 }
Chris Mason56bec292009-03-13 10:10:06 -04002385 }
2386
2387 ref->in_tree = 0;
2388 rb_erase(&ref->rb_node, &delayed_refs->root);
2389 delayed_refs->num_entries--;
Miao Xie093486c2012-12-19 08:10:10 +00002390 if (!btrfs_delayed_ref_is_head(ref)) {
Arne Jansen22cd2e72012-08-09 00:16:53 -06002391 /*
2392 * when we play the delayed ref, also correct the
2393 * ref_mod on head
2394 */
2395 switch (ref->action) {
2396 case BTRFS_ADD_DELAYED_REF:
2397 case BTRFS_ADD_DELAYED_EXTENT:
2398 locked_ref->node.ref_mod -= ref->ref_mod;
2399 break;
2400 case BTRFS_DROP_DELAYED_REF:
2401 locked_ref->node.ref_mod += ref->ref_mod;
2402 break;
2403 default:
2404 WARN_ON(1);
2405 }
2406 }
Chris Mason56bec292009-03-13 10:10:06 -04002407 spin_unlock(&delayed_refs->lock);
2408
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002409 ret = run_one_delayed_ref(trans, root, ref, extent_op,
Chris Mason56bec292009-03-13 10:10:06 -04002410 must_insert_reserved);
Chris Mason56bec292009-03-13 10:10:06 -04002411
Miao Xie78a61842012-11-21 02:21:28 +00002412 btrfs_free_delayed_extent_op(extent_op);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002413 if (ret) {
Miao Xie093486c2012-12-19 08:10:10 +00002414 btrfs_delayed_ref_unlock(locked_ref);
2415 btrfs_put_delayed_ref(ref);
Simon Kirbyc2cf52e2013-03-19 22:41:23 +00002416 btrfs_debug(fs_info, "run_one_delayed_ref returned %d", ret);
Dan Carpenter253beeb2012-04-18 09:59:03 +03002417 spin_lock(&delayed_refs->lock);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002418 return ret;
2419 }
2420
Miao Xie093486c2012-12-19 08:10:10 +00002421 /*
2422 * If this node is a head, that means all the refs in this head
2423 * have been dealt with, and we will pick the next head to deal
2424 * with, so we must unlock the head and drop it from the cluster
2425 * list before we release it.
2426 */
2427 if (btrfs_delayed_ref_is_head(ref)) {
2428 list_del_init(&locked_ref->cluster);
2429 btrfs_delayed_ref_unlock(locked_ref);
2430 locked_ref = NULL;
2431 }
2432 btrfs_put_delayed_ref(ref);
2433 count++;
Chris Mason203bf282012-01-06 15:23:57 -05002434next:
Chris Mason1887be62009-03-13 10:11:24 -04002435 cond_resched();
Chris Mason56bec292009-03-13 10:10:06 -04002436 spin_lock(&delayed_refs->lock);
2437 }
Chris Masonc3e69d52009-03-13 10:17:05 -04002438 return count;
2439}
2440
Arne Jansen709c0482011-09-12 12:22:57 +02002441#ifdef SCRAMBLE_DELAYED_REFS
2442/*
2443 * Normally delayed refs get processed in ascending bytenr order. This
2444 * correlates in most cases to the order added. To expose dependencies on this
2445 * order, we start to process the tree in the middle instead of the beginning
2446 */
2447static u64 find_middle(struct rb_root *root)
2448{
2449 struct rb_node *n = root->rb_node;
2450 struct btrfs_delayed_ref_node *entry;
2451 int alt = 1;
2452 u64 middle;
2453 u64 first = 0, last = 0;
2454
2455 n = rb_first(root);
2456 if (n) {
2457 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2458 first = entry->bytenr;
2459 }
2460 n = rb_last(root);
2461 if (n) {
2462 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2463 last = entry->bytenr;
2464 }
2465 n = root->rb_node;
2466
2467 while (n) {
2468 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2469 WARN_ON(!entry->in_tree);
2470
2471 middle = entry->bytenr;
2472
2473 if (alt)
2474 n = n->rb_left;
2475 else
2476 n = n->rb_right;
2477
2478 alt = 1 - alt;
2479 }
2480 return middle;
2481}
2482#endif
2483
Arne Jansenbed92ea2012-06-28 18:03:02 +02002484int btrfs_delayed_refs_qgroup_accounting(struct btrfs_trans_handle *trans,
2485 struct btrfs_fs_info *fs_info)
2486{
2487 struct qgroup_update *qgroup_update;
2488 int ret = 0;
2489
2490 if (list_empty(&trans->qgroup_ref_list) !=
2491 !trans->delayed_ref_elem.seq) {
2492 /* list without seq or seq without list */
Simon Kirbyc2cf52e2013-03-19 22:41:23 +00002493 btrfs_err(fs_info,
Jan Schmidtfc36ed7e2013-04-24 16:57:33 +00002494 "qgroup accounting update error, list is%s empty, seq is %#x.%x",
Arne Jansenbed92ea2012-06-28 18:03:02 +02002495 list_empty(&trans->qgroup_ref_list) ? "" : " not",
Jan Schmidtfc36ed7e2013-04-24 16:57:33 +00002496 (u32)(trans->delayed_ref_elem.seq >> 32),
2497 (u32)trans->delayed_ref_elem.seq);
Arne Jansenbed92ea2012-06-28 18:03:02 +02002498 BUG();
2499 }
2500
2501 if (!trans->delayed_ref_elem.seq)
2502 return 0;
2503
2504 while (!list_empty(&trans->qgroup_ref_list)) {
2505 qgroup_update = list_first_entry(&trans->qgroup_ref_list,
2506 struct qgroup_update, list);
2507 list_del(&qgroup_update->list);
2508 if (!ret)
2509 ret = btrfs_qgroup_account_ref(
2510 trans, fs_info, qgroup_update->node,
2511 qgroup_update->extent_op);
2512 kfree(qgroup_update);
2513 }
2514
2515 btrfs_put_tree_mod_seq(fs_info, &trans->delayed_ref_elem);
2516
2517 return ret;
2518}
2519
Chris Masonbb721702013-01-29 18:44:12 -05002520static int refs_newer(struct btrfs_delayed_ref_root *delayed_refs, int seq,
2521 int count)
2522{
2523 int val = atomic_read(&delayed_refs->ref_seq);
2524
2525 if (val < seq || val >= seq + count)
2526 return 1;
2527 return 0;
2528}
2529
Josef Bacik1be41b72013-06-12 13:56:06 -04002530static inline u64 heads_to_leaves(struct btrfs_root *root, u64 heads)
2531{
2532 u64 num_bytes;
2533
2534 num_bytes = heads * (sizeof(struct btrfs_extent_item) +
2535 sizeof(struct btrfs_extent_inline_ref));
2536 if (!btrfs_fs_incompat(root->fs_info, SKINNY_METADATA))
2537 num_bytes += heads * sizeof(struct btrfs_tree_block_info);
2538
2539 /*
2540 * We don't ever fill up leaves all the way so multiply by 2 just to be
2541 * closer to what we're really going to want to ouse.
2542 */
2543 return div64_u64(num_bytes, BTRFS_LEAF_DATA_SIZE(root));
2544}
2545
2546int btrfs_should_throttle_delayed_refs(struct btrfs_trans_handle *trans,
2547 struct btrfs_root *root)
2548{
2549 struct btrfs_block_rsv *global_rsv;
2550 u64 num_heads = trans->transaction->delayed_refs.num_heads_ready;
2551 u64 num_bytes;
2552 int ret = 0;
2553
2554 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
2555 num_heads = heads_to_leaves(root, num_heads);
2556 if (num_heads > 1)
2557 num_bytes += (num_heads - 1) * root->leafsize;
2558 num_bytes <<= 1;
2559 global_rsv = &root->fs_info->global_block_rsv;
2560
2561 /*
2562 * If we can't allocate any more chunks lets make sure we have _lots_ of
2563 * wiggle room since running delayed refs can create more delayed refs.
2564 */
2565 if (global_rsv->space_info->full)
2566 num_bytes <<= 1;
2567
2568 spin_lock(&global_rsv->lock);
2569 if (global_rsv->reserved <= num_bytes)
2570 ret = 1;
2571 spin_unlock(&global_rsv->lock);
2572 return ret;
2573}
2574
Chris Masonc3e69d52009-03-13 10:17:05 -04002575/*
2576 * this starts processing the delayed reference count updates and
2577 * extent insertions we have queued up so far. count can be
2578 * 0, which means to process everything in the tree at the start
2579 * of the run (but not newly added entries), or it can be some target
2580 * number you'd like to process.
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002581 *
2582 * Returns 0 on success or if called with an aborted transaction
2583 * Returns <0 on error and aborts the transaction
Chris Masonc3e69d52009-03-13 10:17:05 -04002584 */
2585int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2586 struct btrfs_root *root, unsigned long count)
2587{
2588 struct rb_node *node;
2589 struct btrfs_delayed_ref_root *delayed_refs;
2590 struct btrfs_delayed_ref_node *ref;
2591 struct list_head cluster;
2592 int ret;
Jan Schmidta1686502011-12-12 16:10:07 +01002593 u64 delayed_start;
Chris Masonc3e69d52009-03-13 10:17:05 -04002594 int run_all = count == (unsigned long)-1;
2595 int run_most = 0;
Arne Jansen1fa11e22012-08-06 14:18:51 -06002596 int loops;
Chris Masonc3e69d52009-03-13 10:17:05 -04002597
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002598 /* We'll clean this up in btrfs_cleanup_transaction */
2599 if (trans->aborted)
2600 return 0;
2601
Chris Masonc3e69d52009-03-13 10:17:05 -04002602 if (root == root->fs_info->extent_root)
2603 root = root->fs_info->tree_root;
2604
Jan Schmidtedf39272012-06-28 18:04:55 +02002605 btrfs_delayed_refs_qgroup_accounting(trans, root->fs_info);
2606
Chris Masonc3e69d52009-03-13 10:17:05 -04002607 delayed_refs = &trans->transaction->delayed_refs;
2608 INIT_LIST_HEAD(&cluster);
Chris Masonbb721702013-01-29 18:44:12 -05002609 if (count == 0) {
2610 count = delayed_refs->num_entries * 2;
2611 run_most = 1;
2612 }
2613
2614 if (!run_all && !run_most) {
2615 int old;
2616 int seq = atomic_read(&delayed_refs->ref_seq);
2617
2618progress:
2619 old = atomic_cmpxchg(&delayed_refs->procs_running_refs, 0, 1);
2620 if (old) {
2621 DEFINE_WAIT(__wait);
Josef Bacik1be41b72013-06-12 13:56:06 -04002622 if (delayed_refs->flushing ||
2623 !btrfs_should_throttle_delayed_refs(trans, root))
Chris Masonbb721702013-01-29 18:44:12 -05002624 return 0;
2625
2626 prepare_to_wait(&delayed_refs->wait, &__wait,
2627 TASK_UNINTERRUPTIBLE);
2628
2629 old = atomic_cmpxchg(&delayed_refs->procs_running_refs, 0, 1);
2630 if (old) {
2631 schedule();
2632 finish_wait(&delayed_refs->wait, &__wait);
2633
2634 if (!refs_newer(delayed_refs, seq, 256))
2635 goto progress;
2636 else
2637 return 0;
2638 } else {
2639 finish_wait(&delayed_refs->wait, &__wait);
2640 goto again;
2641 }
2642 }
2643
2644 } else {
2645 atomic_inc(&delayed_refs->procs_running_refs);
2646 }
2647
Chris Masonc3e69d52009-03-13 10:17:05 -04002648again:
Arne Jansen1fa11e22012-08-06 14:18:51 -06002649 loops = 0;
Chris Masonc3e69d52009-03-13 10:17:05 -04002650 spin_lock(&delayed_refs->lock);
Jan Schmidt097b8a72012-06-21 11:08:04 +02002651
Arne Jansen709c0482011-09-12 12:22:57 +02002652#ifdef SCRAMBLE_DELAYED_REFS
2653 delayed_refs->run_delayed_start = find_middle(&delayed_refs->root);
2654#endif
2655
Chris Masonc3e69d52009-03-13 10:17:05 -04002656 while (1) {
2657 if (!(run_all || run_most) &&
Josef Bacik1be41b72013-06-12 13:56:06 -04002658 !btrfs_should_throttle_delayed_refs(trans, root))
Chris Masonc3e69d52009-03-13 10:17:05 -04002659 break;
2660
2661 /*
2662 * go find something we can process in the rbtree. We start at
2663 * the beginning of the tree, and then build a cluster
2664 * of refs to process starting at the first one we are able to
2665 * lock
2666 */
Jan Schmidta1686502011-12-12 16:10:07 +01002667 delayed_start = delayed_refs->run_delayed_start;
Chris Masonc3e69d52009-03-13 10:17:05 -04002668 ret = btrfs_find_ref_cluster(trans, &cluster,
2669 delayed_refs->run_delayed_start);
2670 if (ret)
2671 break;
2672
2673 ret = run_clustered_refs(trans, root, &cluster);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002674 if (ret < 0) {
Miao Xie093486c2012-12-19 08:10:10 +00002675 btrfs_release_ref_cluster(&cluster);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002676 spin_unlock(&delayed_refs->lock);
2677 btrfs_abort_transaction(trans, root, ret);
Chris Masonbb721702013-01-29 18:44:12 -05002678 atomic_dec(&delayed_refs->procs_running_refs);
Josef Bacikf971fe22013-06-10 11:52:32 -04002679 wake_up(&delayed_refs->wait);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002680 return ret;
2681 }
Chris Masonc3e69d52009-03-13 10:17:05 -04002682
Chris Masonbb721702013-01-29 18:44:12 -05002683 atomic_add(ret, &delayed_refs->ref_seq);
2684
Chris Masonc3e69d52009-03-13 10:17:05 -04002685 count -= min_t(unsigned long, ret, count);
2686
2687 if (count == 0)
2688 break;
Jan Schmidta1686502011-12-12 16:10:07 +01002689
Arne Jansen1fa11e22012-08-06 14:18:51 -06002690 if (delayed_start >= delayed_refs->run_delayed_start) {
2691 if (loops == 0) {
2692 /*
2693 * btrfs_find_ref_cluster looped. let's do one
2694 * more cycle. if we don't run any delayed ref
2695 * during that cycle (because we can't because
2696 * all of them are blocked), bail out.
2697 */
2698 loops = 1;
2699 } else {
2700 /*
2701 * no runnable refs left, stop trying
2702 */
2703 BUG_ON(run_all);
2704 break;
2705 }
2706 }
2707 if (ret) {
Jan Schmidta1686502011-12-12 16:10:07 +01002708 /* refs were run, let's reset staleness detection */
Arne Jansen1fa11e22012-08-06 14:18:51 -06002709 loops = 0;
Jan Schmidta1686502011-12-12 16:10:07 +01002710 }
Chris Masonc3e69d52009-03-13 10:17:05 -04002711 }
2712
Chris Mason56bec292009-03-13 10:10:06 -04002713 if (run_all) {
Josef Bacikea658ba2012-09-11 16:57:25 -04002714 if (!list_empty(&trans->new_bgs)) {
2715 spin_unlock(&delayed_refs->lock);
2716 btrfs_create_pending_block_groups(trans, root);
2717 spin_lock(&delayed_refs->lock);
2718 }
2719
Chris Mason56bec292009-03-13 10:10:06 -04002720 node = rb_first(&delayed_refs->root);
Chris Masonc3e69d52009-03-13 10:17:05 -04002721 if (!node)
Chris Mason56bec292009-03-13 10:10:06 -04002722 goto out;
Chris Masonc3e69d52009-03-13 10:17:05 -04002723 count = (unsigned long)-1;
Chris Mason56bec292009-03-13 10:10:06 -04002724
2725 while (node) {
2726 ref = rb_entry(node, struct btrfs_delayed_ref_node,
2727 rb_node);
2728 if (btrfs_delayed_ref_is_head(ref)) {
2729 struct btrfs_delayed_ref_head *head;
2730
2731 head = btrfs_delayed_node_to_head(ref);
2732 atomic_inc(&ref->refs);
2733
2734 spin_unlock(&delayed_refs->lock);
David Sterba8cc33e52011-05-02 15:29:25 +02002735 /*
2736 * Mutex was contended, block until it's
2737 * released and try again
2738 */
Chris Mason56bec292009-03-13 10:10:06 -04002739 mutex_lock(&head->mutex);
2740 mutex_unlock(&head->mutex);
2741
2742 btrfs_put_delayed_ref(ref);
Chris Mason1887be62009-03-13 10:11:24 -04002743 cond_resched();
Chris Mason56bec292009-03-13 10:10:06 -04002744 goto again;
2745 }
2746 node = rb_next(node);
2747 }
2748 spin_unlock(&delayed_refs->lock);
Chris Mason56bec292009-03-13 10:10:06 -04002749 schedule_timeout(1);
2750 goto again;
2751 }
Chris Mason54aa1f42007-06-22 14:16:25 -04002752out:
Chris Masonbb721702013-01-29 18:44:12 -05002753 atomic_dec(&delayed_refs->procs_running_refs);
2754 smp_mb();
2755 if (waitqueue_active(&delayed_refs->wait))
2756 wake_up(&delayed_refs->wait);
2757
Chris Masonc3e69d52009-03-13 10:17:05 -04002758 spin_unlock(&delayed_refs->lock);
Jan Schmidtedf39272012-06-28 18:04:55 +02002759 assert_qgroups_uptodate(trans);
Chris Masona28ec192007-03-06 20:08:01 -05002760 return 0;
2761}
2762
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002763int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
2764 struct btrfs_root *root,
2765 u64 bytenr, u64 num_bytes, u64 flags,
Josef Bacikb1c79e02013-05-09 13:49:30 -04002766 int level, int is_data)
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002767{
2768 struct btrfs_delayed_extent_op *extent_op;
2769 int ret;
2770
Miao Xie78a61842012-11-21 02:21:28 +00002771 extent_op = btrfs_alloc_delayed_extent_op();
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002772 if (!extent_op)
2773 return -ENOMEM;
2774
2775 extent_op->flags_to_set = flags;
2776 extent_op->update_flags = 1;
2777 extent_op->update_key = 0;
2778 extent_op->is_data = is_data ? 1 : 0;
Josef Bacikb1c79e02013-05-09 13:49:30 -04002779 extent_op->level = level;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002780
Arne Jansen66d7e7f2011-09-12 15:26:38 +02002781 ret = btrfs_add_delayed_extent_op(root->fs_info, trans, bytenr,
2782 num_bytes, extent_op);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002783 if (ret)
Miao Xie78a61842012-11-21 02:21:28 +00002784 btrfs_free_delayed_extent_op(extent_op);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002785 return ret;
2786}
2787
2788static noinline int check_delayed_ref(struct btrfs_trans_handle *trans,
2789 struct btrfs_root *root,
2790 struct btrfs_path *path,
2791 u64 objectid, u64 offset, u64 bytenr)
2792{
2793 struct btrfs_delayed_ref_head *head;
2794 struct btrfs_delayed_ref_node *ref;
2795 struct btrfs_delayed_data_ref *data_ref;
2796 struct btrfs_delayed_ref_root *delayed_refs;
2797 struct rb_node *node;
2798 int ret = 0;
2799
2800 ret = -ENOENT;
2801 delayed_refs = &trans->transaction->delayed_refs;
2802 spin_lock(&delayed_refs->lock);
2803 head = btrfs_find_delayed_ref_head(trans, bytenr);
2804 if (!head)
2805 goto out;
2806
2807 if (!mutex_trylock(&head->mutex)) {
2808 atomic_inc(&head->node.refs);
2809 spin_unlock(&delayed_refs->lock);
2810
David Sterbab3b4aa72011-04-21 01:20:15 +02002811 btrfs_release_path(path);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002812
David Sterba8cc33e52011-05-02 15:29:25 +02002813 /*
2814 * Mutex was contended, block until it's released and let
2815 * caller try again
2816 */
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002817 mutex_lock(&head->mutex);
2818 mutex_unlock(&head->mutex);
2819 btrfs_put_delayed_ref(&head->node);
2820 return -EAGAIN;
2821 }
2822
2823 node = rb_prev(&head->node.rb_node);
2824 if (!node)
2825 goto out_unlock;
2826
2827 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2828
2829 if (ref->bytenr != bytenr)
2830 goto out_unlock;
2831
2832 ret = 1;
2833 if (ref->type != BTRFS_EXTENT_DATA_REF_KEY)
2834 goto out_unlock;
2835
2836 data_ref = btrfs_delayed_node_to_data_ref(ref);
2837
2838 node = rb_prev(node);
2839 if (node) {
Liu Bodf57dbe2012-07-23 05:50:03 -06002840 int seq = ref->seq;
2841
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002842 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
Liu Bodf57dbe2012-07-23 05:50:03 -06002843 if (ref->bytenr == bytenr && ref->seq == seq)
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002844 goto out_unlock;
2845 }
2846
2847 if (data_ref->root != root->root_key.objectid ||
2848 data_ref->objectid != objectid || data_ref->offset != offset)
2849 goto out_unlock;
2850
2851 ret = 0;
2852out_unlock:
2853 mutex_unlock(&head->mutex);
2854out:
2855 spin_unlock(&delayed_refs->lock);
2856 return ret;
2857}
2858
2859static noinline int check_committed_ref(struct btrfs_trans_handle *trans,
2860 struct btrfs_root *root,
2861 struct btrfs_path *path,
2862 u64 objectid, u64 offset, u64 bytenr)
Chris Masonbe20aa92007-12-17 20:14:01 -05002863{
2864 struct btrfs_root *extent_root = root->fs_info->extent_root;
Yan Zhengf321e492008-07-30 09:26:11 -04002865 struct extent_buffer *leaf;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002866 struct btrfs_extent_data_ref *ref;
2867 struct btrfs_extent_inline_ref *iref;
2868 struct btrfs_extent_item *ei;
Chris Masonbe20aa92007-12-17 20:14:01 -05002869 struct btrfs_key key;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002870 u32 item_size;
Yan Zhengf321e492008-07-30 09:26:11 -04002871 int ret;
Chris Masonbe20aa92007-12-17 20:14:01 -05002872
Chris Masonbe20aa92007-12-17 20:14:01 -05002873 key.objectid = bytenr;
Zheng Yan31840ae2008-09-23 13:14:14 -04002874 key.offset = (u64)-1;
Yan Zhengf321e492008-07-30 09:26:11 -04002875 key.type = BTRFS_EXTENT_ITEM_KEY;
Chris Masonbe20aa92007-12-17 20:14:01 -05002876
Chris Masonbe20aa92007-12-17 20:14:01 -05002877 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2878 if (ret < 0)
2879 goto out;
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002880 BUG_ON(ret == 0); /* Corruption */
Yan Zheng80ff3852008-10-30 14:20:02 -04002881
2882 ret = -ENOENT;
2883 if (path->slots[0] == 0)
Zheng Yan31840ae2008-09-23 13:14:14 -04002884 goto out;
Chris Masonbe20aa92007-12-17 20:14:01 -05002885
Zheng Yan31840ae2008-09-23 13:14:14 -04002886 path->slots[0]--;
Yan Zhengf321e492008-07-30 09:26:11 -04002887 leaf = path->nodes[0];
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002888 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
Chris Masonbe20aa92007-12-17 20:14:01 -05002889
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002890 if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
Chris Masonbe20aa92007-12-17 20:14:01 -05002891 goto out;
Chris Masonbe20aa92007-12-17 20:14:01 -05002892
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002893 ret = 1;
2894 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2895#ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2896 if (item_size < sizeof(*ei)) {
2897 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
2898 goto out;
Chris Masonbe20aa92007-12-17 20:14:01 -05002899 }
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002900#endif
2901 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2902
2903 if (item_size != sizeof(*ei) +
2904 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
2905 goto out;
2906
2907 if (btrfs_extent_generation(leaf, ei) <=
2908 btrfs_root_last_snapshot(&root->root_item))
2909 goto out;
2910
2911 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
2912 if (btrfs_extent_inline_ref_type(leaf, iref) !=
2913 BTRFS_EXTENT_DATA_REF_KEY)
2914 goto out;
2915
2916 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
2917 if (btrfs_extent_refs(leaf, ei) !=
2918 btrfs_extent_data_ref_count(leaf, ref) ||
2919 btrfs_extent_data_ref_root(leaf, ref) !=
2920 root->root_key.objectid ||
2921 btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
2922 btrfs_extent_data_ref_offset(leaf, ref) != offset)
2923 goto out;
2924
Yan Zhengf321e492008-07-30 09:26:11 -04002925 ret = 0;
Chris Masonbe20aa92007-12-17 20:14:01 -05002926out:
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002927 return ret;
2928}
2929
2930int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
2931 struct btrfs_root *root,
2932 u64 objectid, u64 offset, u64 bytenr)
2933{
2934 struct btrfs_path *path;
2935 int ret;
2936 int ret2;
2937
2938 path = btrfs_alloc_path();
2939 if (!path)
2940 return -ENOENT;
2941
2942 do {
2943 ret = check_committed_ref(trans, root, path, objectid,
2944 offset, bytenr);
2945 if (ret && ret != -ENOENT)
2946 goto out;
2947
2948 ret2 = check_delayed_ref(trans, root, path, objectid,
2949 offset, bytenr);
2950 } while (ret2 == -EAGAIN);
2951
2952 if (ret2 && ret2 != -ENOENT) {
2953 ret = ret2;
2954 goto out;
2955 }
2956
2957 if (ret != -ENOENT || ret2 != -ENOENT)
2958 ret = 0;
2959out:
Yan Zhengf321e492008-07-30 09:26:11 -04002960 btrfs_free_path(path);
Yan, Zhengf0486c62010-05-16 10:46:25 -04002961 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
2962 WARN_ON(ret > 0);
Yan Zhengf321e492008-07-30 09:26:11 -04002963 return ret;
2964}
2965
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002966static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
Chris Masonb7a9f292009-02-04 09:23:45 -05002967 struct btrfs_root *root,
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002968 struct extent_buffer *buf,
Arne Jansen66d7e7f2011-09-12 15:26:38 +02002969 int full_backref, int inc, int for_cow)
Zheng Yan31840ae2008-09-23 13:14:14 -04002970{
2971 u64 bytenr;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002972 u64 num_bytes;
2973 u64 parent;
Zheng Yan31840ae2008-09-23 13:14:14 -04002974 u64 ref_root;
Zheng Yan31840ae2008-09-23 13:14:14 -04002975 u32 nritems;
Zheng Yan31840ae2008-09-23 13:14:14 -04002976 struct btrfs_key key;
2977 struct btrfs_file_extent_item *fi;
2978 int i;
2979 int level;
2980 int ret = 0;
Zheng Yan31840ae2008-09-23 13:14:14 -04002981 int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
Arne Jansen66d7e7f2011-09-12 15:26:38 +02002982 u64, u64, u64, u64, u64, u64, int);
Zheng Yan31840ae2008-09-23 13:14:14 -04002983
2984 ref_root = btrfs_header_owner(buf);
Zheng Yan31840ae2008-09-23 13:14:14 -04002985 nritems = btrfs_header_nritems(buf);
2986 level = btrfs_header_level(buf);
2987
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002988 if (!root->ref_cows && level == 0)
2989 return 0;
Chris Masonb7a9f292009-02-04 09:23:45 -05002990
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002991 if (inc)
2992 process_func = btrfs_inc_extent_ref;
2993 else
2994 process_func = btrfs_free_extent;
Zheng Yan31840ae2008-09-23 13:14:14 -04002995
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002996 if (full_backref)
2997 parent = buf->start;
2998 else
2999 parent = 0;
3000
Zheng Yan31840ae2008-09-23 13:14:14 -04003001 for (i = 0; i < nritems; i++) {
Chris Masondb945352007-10-15 16:15:53 -04003002 if (level == 0) {
Chris Mason5f39d392007-10-15 16:14:19 -04003003 btrfs_item_key_to_cpu(buf, &key, i);
3004 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
Chris Mason54aa1f42007-06-22 14:16:25 -04003005 continue;
Chris Mason5f39d392007-10-15 16:14:19 -04003006 fi = btrfs_item_ptr(buf, i,
Chris Mason54aa1f42007-06-22 14:16:25 -04003007 struct btrfs_file_extent_item);
Chris Mason5f39d392007-10-15 16:14:19 -04003008 if (btrfs_file_extent_type(buf, fi) ==
Chris Mason54aa1f42007-06-22 14:16:25 -04003009 BTRFS_FILE_EXTENT_INLINE)
3010 continue;
Zheng Yan31840ae2008-09-23 13:14:14 -04003011 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
3012 if (bytenr == 0)
Chris Mason54aa1f42007-06-22 14:16:25 -04003013 continue;
Zheng Yan31840ae2008-09-23 13:14:14 -04003014
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003015 num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
3016 key.offset -= btrfs_file_extent_offset(buf, fi);
3017 ret = process_func(trans, root, bytenr, num_bytes,
3018 parent, ref_root, key.objectid,
Arne Jansen66d7e7f2011-09-12 15:26:38 +02003019 key.offset, for_cow);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003020 if (ret)
3021 goto fail;
Chris Masonb7a9f292009-02-04 09:23:45 -05003022 } else {
3023 bytenr = btrfs_node_blockptr(buf, i);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003024 num_bytes = btrfs_level_size(root, level - 1);
3025 ret = process_func(trans, root, bytenr, num_bytes,
Arne Jansen66d7e7f2011-09-12 15:26:38 +02003026 parent, ref_root, level - 1, 0,
3027 for_cow);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003028 if (ret)
Zheng Yan31840ae2008-09-23 13:14:14 -04003029 goto fail;
Chris Mason54aa1f42007-06-22 14:16:25 -04003030 }
3031 }
Zheng Yan31840ae2008-09-23 13:14:14 -04003032 return 0;
3033fail:
Chris Mason54aa1f42007-06-22 14:16:25 -04003034 return ret;
Chris Mason02217ed2007-03-02 16:08:05 -05003035}
3036
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003037int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
Arne Jansen66d7e7f2011-09-12 15:26:38 +02003038 struct extent_buffer *buf, int full_backref, int for_cow)
Zheng Yan31840ae2008-09-23 13:14:14 -04003039{
Arne Jansen66d7e7f2011-09-12 15:26:38 +02003040 return __btrfs_mod_ref(trans, root, buf, full_backref, 1, for_cow);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003041}
Zheng Yan31840ae2008-09-23 13:14:14 -04003042
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003043int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
Arne Jansen66d7e7f2011-09-12 15:26:38 +02003044 struct extent_buffer *buf, int full_backref, int for_cow)
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003045{
Arne Jansen66d7e7f2011-09-12 15:26:38 +02003046 return __btrfs_mod_ref(trans, root, buf, full_backref, 0, for_cow);
Zheng Yan31840ae2008-09-23 13:14:14 -04003047}
3048
Chris Mason9078a3e2007-04-26 16:46:15 -04003049static int write_one_cache_group(struct btrfs_trans_handle *trans,
3050 struct btrfs_root *root,
3051 struct btrfs_path *path,
3052 struct btrfs_block_group_cache *cache)
3053{
3054 int ret;
Chris Mason9078a3e2007-04-26 16:46:15 -04003055 struct btrfs_root *extent_root = root->fs_info->extent_root;
Chris Mason5f39d392007-10-15 16:14:19 -04003056 unsigned long bi;
3057 struct extent_buffer *leaf;
Chris Mason9078a3e2007-04-26 16:46:15 -04003058
Chris Mason9078a3e2007-04-26 16:46:15 -04003059 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
Chris Mason54aa1f42007-06-22 14:16:25 -04003060 if (ret < 0)
3061 goto fail;
Jeff Mahoney79787ea2012-03-12 16:03:00 +01003062 BUG_ON(ret); /* Corruption */
Chris Mason5f39d392007-10-15 16:14:19 -04003063
3064 leaf = path->nodes[0];
3065 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
3066 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
3067 btrfs_mark_buffer_dirty(leaf);
David Sterbab3b4aa72011-04-21 01:20:15 +02003068 btrfs_release_path(path);
Chris Mason54aa1f42007-06-22 14:16:25 -04003069fail:
Jeff Mahoney79787ea2012-03-12 16:03:00 +01003070 if (ret) {
3071 btrfs_abort_transaction(trans, root, ret);
Chris Mason9078a3e2007-04-26 16:46:15 -04003072 return ret;
Jeff Mahoney79787ea2012-03-12 16:03:00 +01003073 }
Chris Mason9078a3e2007-04-26 16:46:15 -04003074 return 0;
3075
3076}
3077
Yan Zheng4a8c9a62009-07-22 10:07:05 -04003078static struct btrfs_block_group_cache *
3079next_block_group(struct btrfs_root *root,
3080 struct btrfs_block_group_cache *cache)
3081{
3082 struct rb_node *node;
3083 spin_lock(&root->fs_info->block_group_cache_lock);
3084 node = rb_next(&cache->cache_node);
3085 btrfs_put_block_group(cache);
3086 if (node) {
3087 cache = rb_entry(node, struct btrfs_block_group_cache,
3088 cache_node);
Josef Bacik11dfe352009-11-13 20:12:59 +00003089 btrfs_get_block_group(cache);
Yan Zheng4a8c9a62009-07-22 10:07:05 -04003090 } else
3091 cache = NULL;
3092 spin_unlock(&root->fs_info->block_group_cache_lock);
3093 return cache;
3094}
3095
Josef Bacik0af3d002010-06-21 14:48:16 -04003096static int cache_save_setup(struct btrfs_block_group_cache *block_group,
3097 struct btrfs_trans_handle *trans,
3098 struct btrfs_path *path)
3099{
3100 struct btrfs_root *root = block_group->fs_info->tree_root;
3101 struct inode *inode = NULL;
3102 u64 alloc_hint = 0;
Josef Bacik2b209822010-12-03 13:17:53 -05003103 int dcs = BTRFS_DC_ERROR;
Josef Bacik0af3d002010-06-21 14:48:16 -04003104 int num_pages = 0;
3105 int retries = 0;
3106 int ret = 0;
3107
3108 /*
3109 * If this block group is smaller than 100 megs don't bother caching the
3110 * block group.
3111 */
3112 if (block_group->key.offset < (100 * 1024 * 1024)) {
3113 spin_lock(&block_group->lock);
3114 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
3115 spin_unlock(&block_group->lock);
3116 return 0;
3117 }
3118
3119again:
3120 inode = lookup_free_space_inode(root, block_group, path);
3121 if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
3122 ret = PTR_ERR(inode);
David Sterbab3b4aa72011-04-21 01:20:15 +02003123 btrfs_release_path(path);
Josef Bacik0af3d002010-06-21 14:48:16 -04003124 goto out;
3125 }
3126
3127 if (IS_ERR(inode)) {
3128 BUG_ON(retries);
3129 retries++;
3130
3131 if (block_group->ro)
3132 goto out_free;
3133
3134 ret = create_free_space_inode(root, trans, block_group, path);
3135 if (ret)
3136 goto out_free;
3137 goto again;
3138 }
3139
Josef Bacik5b0e95b2011-10-06 08:58:24 -04003140 /* We've already setup this transaction, go ahead and exit */
3141 if (block_group->cache_generation == trans->transid &&
3142 i_size_read(inode)) {
3143 dcs = BTRFS_DC_SETUP;
3144 goto out_put;
3145 }
3146
Josef Bacik0af3d002010-06-21 14:48:16 -04003147 /*
3148 * We want to set the generation to 0, that way if anything goes wrong
3149 * from here on out we know not to trust this cache when we load up next
3150 * time.
3151 */
3152 BTRFS_I(inode)->generation = 0;
3153 ret = btrfs_update_inode(trans, root, inode);
3154 WARN_ON(ret);
3155
3156 if (i_size_read(inode) > 0) {
Miao Xie7b61cd92013-05-13 13:55:09 +00003157 ret = btrfs_check_trunc_cache_free_space(root,
3158 &root->fs_info->global_block_rsv);
3159 if (ret)
3160 goto out_put;
3161
Josef Bacik0af3d002010-06-21 14:48:16 -04003162 ret = btrfs_truncate_free_space_cache(root, trans, path,
3163 inode);
3164 if (ret)
3165 goto out_put;
3166 }
3167
3168 spin_lock(&block_group->lock);
Liu Bocf7c1ef2012-07-06 03:31:34 -06003169 if (block_group->cached != BTRFS_CACHE_FINISHED ||
3170 !btrfs_test_opt(root, SPACE_CACHE)) {
3171 /*
3172 * don't bother trying to write stuff out _if_
3173 * a) we're not cached,
3174 * b) we're with nospace_cache mount option.
3175 */
Josef Bacik2b209822010-12-03 13:17:53 -05003176 dcs = BTRFS_DC_WRITTEN;
Josef Bacik0af3d002010-06-21 14:48:16 -04003177 spin_unlock(&block_group->lock);
3178 goto out_put;
3179 }
3180 spin_unlock(&block_group->lock);
3181
Josef Bacik6fc823b2012-08-06 13:46:38 -06003182 /*
3183 * Try to preallocate enough space based on how big the block group is.
3184 * Keep in mind this has to include any pinned space which could end up
3185 * taking up quite a bit since it's not folded into the other space
3186 * cache.
3187 */
3188 num_pages = (int)div64_u64(block_group->key.offset, 256 * 1024 * 1024);
Josef Bacik0af3d002010-06-21 14:48:16 -04003189 if (!num_pages)
3190 num_pages = 1;
3191
Josef Bacik0af3d002010-06-21 14:48:16 -04003192 num_pages *= 16;
3193 num_pages *= PAGE_CACHE_SIZE;
3194
3195 ret = btrfs_check_data_free_space(inode, num_pages);
3196 if (ret)
3197 goto out_put;
3198
3199 ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages,
3200 num_pages, num_pages,
3201 &alloc_hint);
Josef Bacik2b209822010-12-03 13:17:53 -05003202 if (!ret)
3203 dcs = BTRFS_DC_SETUP;
Josef Bacik0af3d002010-06-21 14:48:16 -04003204 btrfs_free_reserved_data_space(inode, num_pages);
Josef Bacikc09544e2011-08-30 10:19:10 -04003205
Josef Bacik0af3d002010-06-21 14:48:16 -04003206out_put:
3207 iput(inode);
3208out_free:
David Sterbab3b4aa72011-04-21 01:20:15 +02003209 btrfs_release_path(path);
Josef Bacik0af3d002010-06-21 14:48:16 -04003210out:
3211 spin_lock(&block_group->lock);
Josef Bacike65cbb92011-12-13 16:04:54 -05003212 if (!ret && dcs == BTRFS_DC_SETUP)
Josef Bacik5b0e95b2011-10-06 08:58:24 -04003213 block_group->cache_generation = trans->transid;
Josef Bacik2b209822010-12-03 13:17:53 -05003214 block_group->disk_cache_state = dcs;
Josef Bacik0af3d002010-06-21 14:48:16 -04003215 spin_unlock(&block_group->lock);
3216
3217 return ret;
3218}
3219
Chris Mason96b51792007-10-15 16:15:19 -04003220int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
3221 struct btrfs_root *root)
Chris Mason9078a3e2007-04-26 16:46:15 -04003222{
Yan Zheng4a8c9a62009-07-22 10:07:05 -04003223 struct btrfs_block_group_cache *cache;
Chris Mason9078a3e2007-04-26 16:46:15 -04003224 int err = 0;
Chris Mason9078a3e2007-04-26 16:46:15 -04003225 struct btrfs_path *path;
Chris Mason96b51792007-10-15 16:15:19 -04003226 u64 last = 0;
Chris Mason9078a3e2007-04-26 16:46:15 -04003227
3228 path = btrfs_alloc_path();
3229 if (!path)
3230 return -ENOMEM;
3231
Josef Bacik0af3d002010-06-21 14:48:16 -04003232again:
3233 while (1) {
3234 cache = btrfs_lookup_first_block_group(root->fs_info, last);
3235 while (cache) {
3236 if (cache->disk_cache_state == BTRFS_DC_CLEAR)
3237 break;
3238 cache = next_block_group(root, cache);
3239 }
3240 if (!cache) {
3241 if (last == 0)
3242 break;
3243 last = 0;
3244 continue;
3245 }
3246 err = cache_save_setup(cache, trans, path);
3247 last = cache->key.objectid + cache->key.offset;
3248 btrfs_put_block_group(cache);
3249 }
3250
Chris Masond3977122009-01-05 21:25:51 -05003251 while (1) {
Yan Zheng4a8c9a62009-07-22 10:07:05 -04003252 if (last == 0) {
3253 err = btrfs_run_delayed_refs(trans, root,
3254 (unsigned long)-1);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01003255 if (err) /* File system offline */
3256 goto out;
Josef Bacik0f9dd462008-09-23 13:14:11 -04003257 }
Josef Bacik0f9dd462008-09-23 13:14:11 -04003258
Yan Zheng4a8c9a62009-07-22 10:07:05 -04003259 cache = btrfs_lookup_first_block_group(root->fs_info, last);
3260 while (cache) {
Josef Bacik0af3d002010-06-21 14:48:16 -04003261 if (cache->disk_cache_state == BTRFS_DC_CLEAR) {
3262 btrfs_put_block_group(cache);
3263 goto again;
3264 }
3265
Yan Zheng4a8c9a62009-07-22 10:07:05 -04003266 if (cache->dirty)
3267 break;
3268 cache = next_block_group(root, cache);
3269 }
3270 if (!cache) {
3271 if (last == 0)
3272 break;
3273 last = 0;
Chris Mason96b51792007-10-15 16:15:19 -04003274 continue;
Chris Mason9078a3e2007-04-26 16:46:15 -04003275 }
Yan Zheng4a8c9a62009-07-22 10:07:05 -04003276
Josef Bacik0cb59c92010-07-02 12:14:14 -04003277 if (cache->disk_cache_state == BTRFS_DC_SETUP)
3278 cache->disk_cache_state = BTRFS_DC_NEED_WRITE;
Yan Zheng4a8c9a62009-07-22 10:07:05 -04003279 cache->dirty = 0;
3280 last = cache->key.objectid + cache->key.offset;
3281
3282 err = write_one_cache_group(trans, root, path, cache);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01003283 if (err) /* File system offline */
3284 goto out;
3285
Yan Zheng4a8c9a62009-07-22 10:07:05 -04003286 btrfs_put_block_group(cache);
Chris Mason9078a3e2007-04-26 16:46:15 -04003287 }
Yan Zheng4a8c9a62009-07-22 10:07:05 -04003288
Josef Bacik0cb59c92010-07-02 12:14:14 -04003289 while (1) {
3290 /*
3291 * I don't think this is needed since we're just marking our
3292 * preallocated extent as written, but just in case it can't
3293 * hurt.
3294 */
3295 if (last == 0) {
3296 err = btrfs_run_delayed_refs(trans, root,
3297 (unsigned long)-1);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01003298 if (err) /* File system offline */
3299 goto out;
Josef Bacik0cb59c92010-07-02 12:14:14 -04003300 }
3301
3302 cache = btrfs_lookup_first_block_group(root->fs_info, last);
3303 while (cache) {
3304 /*
3305 * Really this shouldn't happen, but it could if we
3306 * couldn't write the entire preallocated extent and
3307 * splitting the extent resulted in a new block.
3308 */
3309 if (cache->dirty) {
3310 btrfs_put_block_group(cache);
3311 goto again;
3312 }
3313 if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
3314 break;
3315 cache = next_block_group(root, cache);
3316 }
3317 if (!cache) {
3318 if (last == 0)
3319 break;
3320 last = 0;
3321 continue;
3322 }
3323
Jeff Mahoney79787ea2012-03-12 16:03:00 +01003324 err = btrfs_write_out_cache(root, trans, cache, path);
Josef Bacik0cb59c92010-07-02 12:14:14 -04003325
3326 /*
3327 * If we didn't have an error then the cache state is still
3328 * NEED_WRITE, so we can set it to WRITTEN.
3329 */
Jeff Mahoney79787ea2012-03-12 16:03:00 +01003330 if (!err && cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
Josef Bacik0cb59c92010-07-02 12:14:14 -04003331 cache->disk_cache_state = BTRFS_DC_WRITTEN;
3332 last = cache->key.objectid + cache->key.offset;
3333 btrfs_put_block_group(cache);
3334 }
Jeff Mahoney79787ea2012-03-12 16:03:00 +01003335out:
Josef Bacik0cb59c92010-07-02 12:14:14 -04003336
Chris Mason9078a3e2007-04-26 16:46:15 -04003337 btrfs_free_path(path);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01003338 return err;
Chris Mason9078a3e2007-04-26 16:46:15 -04003339}
3340
Yan Zhengd2fb3432008-12-11 16:30:39 -05003341int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
3342{
3343 struct btrfs_block_group_cache *block_group;
3344 int readonly = 0;
3345
3346 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
3347 if (!block_group || block_group->ro)
3348 readonly = 1;
3349 if (block_group)
Chris Masonfa9c0d792009-04-03 09:47:43 -04003350 btrfs_put_block_group(block_group);
Yan Zhengd2fb3432008-12-11 16:30:39 -05003351 return readonly;
3352}
3353
Chris Mason593060d2008-03-25 16:50:33 -04003354static int update_space_info(struct btrfs_fs_info *info, u64 flags,
3355 u64 total_bytes, u64 bytes_used,
3356 struct btrfs_space_info **space_info)
3357{
3358 struct btrfs_space_info *found;
Yan, Zhengb742bb82010-05-16 10:46:24 -04003359 int i;
3360 int factor;
Josef Bacikb150a4f2013-06-19 15:00:04 -04003361 int ret;
Yan, Zhengb742bb82010-05-16 10:46:24 -04003362
3363 if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3364 BTRFS_BLOCK_GROUP_RAID10))
3365 factor = 2;
3366 else
3367 factor = 1;
Chris Mason593060d2008-03-25 16:50:33 -04003368
3369 found = __find_space_info(info, flags);
3370 if (found) {
Josef Bacik25179202008-10-29 14:49:05 -04003371 spin_lock(&found->lock);
Chris Mason593060d2008-03-25 16:50:33 -04003372 found->total_bytes += total_bytes;
Josef Bacik89a55892010-10-14 14:52:27 -04003373 found->disk_total += total_bytes * factor;
Chris Mason593060d2008-03-25 16:50:33 -04003374 found->bytes_used += bytes_used;
Yan, Zhengb742bb82010-05-16 10:46:24 -04003375 found->disk_used += bytes_used * factor;
Chris Mason8f18cf12008-04-25 16:53:30 -04003376 found->full = 0;
Josef Bacik25179202008-10-29 14:49:05 -04003377 spin_unlock(&found->lock);
Chris Mason593060d2008-03-25 16:50:33 -04003378 *space_info = found;
3379 return 0;
3380 }
Yan Zhengc146afa2008-11-12 14:34:12 -05003381 found = kzalloc(sizeof(*found), GFP_NOFS);
Chris Mason593060d2008-03-25 16:50:33 -04003382 if (!found)
3383 return -ENOMEM;
3384
Josef Bacikb150a4f2013-06-19 15:00:04 -04003385 ret = percpu_counter_init(&found->total_bytes_pinned, 0);
3386 if (ret) {
3387 kfree(found);
3388 return ret;
3389 }
3390
Yan, Zhengb742bb82010-05-16 10:46:24 -04003391 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
3392 INIT_LIST_HEAD(&found->block_groups[i]);
Josef Bacik80eb2342008-10-29 14:49:05 -04003393 init_rwsem(&found->groups_sem);
Josef Bacik0f9dd462008-09-23 13:14:11 -04003394 spin_lock_init(&found->lock);
Ilya Dryomov52ba6922012-01-16 22:04:47 +02003395 found->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK;
Chris Mason593060d2008-03-25 16:50:33 -04003396 found->total_bytes = total_bytes;
Josef Bacik89a55892010-10-14 14:52:27 -04003397 found->disk_total = total_bytes * factor;
Chris Mason593060d2008-03-25 16:50:33 -04003398 found->bytes_used = bytes_used;
Yan, Zhengb742bb82010-05-16 10:46:24 -04003399 found->disk_used = bytes_used * factor;
Chris Mason593060d2008-03-25 16:50:33 -04003400 found->bytes_pinned = 0;
Zheng Yane8569812008-09-26 10:05:48 -04003401 found->bytes_reserved = 0;
Yan Zhengc146afa2008-11-12 14:34:12 -05003402 found->bytes_readonly = 0;
Yan, Zhengf0486c62010-05-16 10:46:25 -04003403 found->bytes_may_use = 0;
Chris Mason593060d2008-03-25 16:50:33 -04003404 found->full = 0;
Chris Mason0e4f8f82011-04-15 16:05:44 -04003405 found->force_alloc = CHUNK_ALLOC_NO_FORCE;
Josef Bacik6d741192011-04-11 20:20:11 -04003406 found->chunk_alloc = 0;
Josef Bacikfdb5eff2011-06-07 16:07:44 -04003407 found->flush = 0;
3408 init_waitqueue_head(&found->wait);
Chris Mason593060d2008-03-25 16:50:33 -04003409 *space_info = found;
Chris Mason4184ea72009-03-10 12:39:20 -04003410 list_add_rcu(&found->list, &info->space_info);
Li Zefanb4d7c3c2012-07-09 20:21:07 -06003411 if (flags & BTRFS_BLOCK_GROUP_DATA)
3412 info->data_sinfo = found;
Chris Mason593060d2008-03-25 16:50:33 -04003413 return 0;
3414}
3415
Chris Mason8790d502008-04-03 16:29:03 -04003416static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
3417{
Ilya Dryomov899c81e2012-03-27 17:09:16 +03003418 u64 extra_flags = chunk_to_extended(flags) &
3419 BTRFS_EXTENDED_PROFILE_MASK;
Ilya Dryomova46d11a2012-01-16 22:04:47 +02003420
Miao Xiede98ced2013-01-29 10:13:12 +00003421 write_seqlock(&fs_info->profiles_lock);
Ilya Dryomova46d11a2012-01-16 22:04:47 +02003422 if (flags & BTRFS_BLOCK_GROUP_DATA)
3423 fs_info->avail_data_alloc_bits |= extra_flags;
3424 if (flags & BTRFS_BLOCK_GROUP_METADATA)
3425 fs_info->avail_metadata_alloc_bits |= extra_flags;
3426 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3427 fs_info->avail_system_alloc_bits |= extra_flags;
Miao Xiede98ced2013-01-29 10:13:12 +00003428 write_sequnlock(&fs_info->profiles_lock);
Chris Mason8790d502008-04-03 16:29:03 -04003429}
Chris Mason593060d2008-03-25 16:50:33 -04003430
Ilya Dryomova46d11a2012-01-16 22:04:47 +02003431/*
Ilya Dryomovfc67c452012-03-27 17:09:17 +03003432 * returns target flags in extended format or 0 if restripe for this
3433 * chunk_type is not in progress
Ilya Dryomovc6664b42012-04-12 16:03:56 -04003434 *
3435 * should be called with either volume_mutex or balance_lock held
Ilya Dryomovfc67c452012-03-27 17:09:17 +03003436 */
3437static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags)
3438{
3439 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3440 u64 target = 0;
3441
Ilya Dryomovfc67c452012-03-27 17:09:17 +03003442 if (!bctl)
3443 return 0;
3444
3445 if (flags & BTRFS_BLOCK_GROUP_DATA &&
3446 bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3447 target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target;
3448 } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM &&
3449 bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3450 target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target;
3451 } else if (flags & BTRFS_BLOCK_GROUP_METADATA &&
3452 bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3453 target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target;
3454 }
3455
3456 return target;
3457}
3458
3459/*
Ilya Dryomova46d11a2012-01-16 22:04:47 +02003460 * @flags: available profiles in extended format (see ctree.h)
3461 *
Ilya Dryomove4d8ec02012-01-16 22:04:48 +02003462 * Returns reduced profile in chunk format. If profile changing is in
3463 * progress (either running or paused) picks the target profile (if it's
3464 * already available), otherwise falls back to plain reducing.
Ilya Dryomova46d11a2012-01-16 22:04:47 +02003465 */
Eric Sandeen48a3b632013-04-25 20:41:01 +00003466static u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
Chris Masonec44a352008-04-28 15:29:52 -04003467{
Chris Masoncd02dca2010-12-13 14:56:23 -05003468 /*
3469 * we add in the count of missing devices because we want
3470 * to make sure that any RAID levels on a degraded FS
3471 * continue to be honored.
3472 */
3473 u64 num_devices = root->fs_info->fs_devices->rw_devices +
3474 root->fs_info->fs_devices->missing_devices;
Ilya Dryomovfc67c452012-03-27 17:09:17 +03003475 u64 target;
David Woodhouse53b381b2013-01-29 18:40:14 -05003476 u64 tmp;
Chris Masona061fc82008-05-07 11:43:44 -04003477
Ilya Dryomovfc67c452012-03-27 17:09:17 +03003478 /*
3479 * see if restripe for this chunk_type is in progress, if so
3480 * try to reduce to the target profile
3481 */
Ilya Dryomove4d8ec02012-01-16 22:04:48 +02003482 spin_lock(&root->fs_info->balance_lock);
Ilya Dryomovfc67c452012-03-27 17:09:17 +03003483 target = get_restripe_target(root->fs_info, flags);
3484 if (target) {
3485 /* pick target profile only if it's already available */
3486 if ((flags & target) & BTRFS_EXTENDED_PROFILE_MASK) {
Ilya Dryomove4d8ec02012-01-16 22:04:48 +02003487 spin_unlock(&root->fs_info->balance_lock);
Ilya Dryomovfc67c452012-03-27 17:09:17 +03003488 return extended_to_chunk(target);
Ilya Dryomove4d8ec02012-01-16 22:04:48 +02003489 }
3490 }
3491 spin_unlock(&root->fs_info->balance_lock);
3492
David Woodhouse53b381b2013-01-29 18:40:14 -05003493 /* First, mask out the RAID levels which aren't possible */
Chris Masona061fc82008-05-07 11:43:44 -04003494 if (num_devices == 1)
David Woodhouse53b381b2013-01-29 18:40:14 -05003495 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0 |
3496 BTRFS_BLOCK_GROUP_RAID5);
3497 if (num_devices < 3)
3498 flags &= ~BTRFS_BLOCK_GROUP_RAID6;
Chris Masona061fc82008-05-07 11:43:44 -04003499 if (num_devices < 4)
3500 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
3501
David Woodhouse53b381b2013-01-29 18:40:14 -05003502 tmp = flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID0 |
3503 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID5 |
3504 BTRFS_BLOCK_GROUP_RAID6 | BTRFS_BLOCK_GROUP_RAID10);
3505 flags &= ~tmp;
Chris Masonec44a352008-04-28 15:29:52 -04003506
David Woodhouse53b381b2013-01-29 18:40:14 -05003507 if (tmp & BTRFS_BLOCK_GROUP_RAID6)
3508 tmp = BTRFS_BLOCK_GROUP_RAID6;
3509 else if (tmp & BTRFS_BLOCK_GROUP_RAID5)
3510 tmp = BTRFS_BLOCK_GROUP_RAID5;
3511 else if (tmp & BTRFS_BLOCK_GROUP_RAID10)
3512 tmp = BTRFS_BLOCK_GROUP_RAID10;
3513 else if (tmp & BTRFS_BLOCK_GROUP_RAID1)
3514 tmp = BTRFS_BLOCK_GROUP_RAID1;
3515 else if (tmp & BTRFS_BLOCK_GROUP_RAID0)
3516 tmp = BTRFS_BLOCK_GROUP_RAID0;
Chris Masonec44a352008-04-28 15:29:52 -04003517
David Woodhouse53b381b2013-01-29 18:40:14 -05003518 return extended_to_chunk(flags | tmp);
Chris Masonec44a352008-04-28 15:29:52 -04003519}
3520
Yan, Zhengb742bb82010-05-16 10:46:24 -04003521static u64 get_alloc_profile(struct btrfs_root *root, u64 flags)
Josef Bacik6a632092009-02-20 11:00:09 -05003522{
Miao Xiede98ced2013-01-29 10:13:12 +00003523 unsigned seq;
3524
3525 do {
3526 seq = read_seqbegin(&root->fs_info->profiles_lock);
3527
3528 if (flags & BTRFS_BLOCK_GROUP_DATA)
3529 flags |= root->fs_info->avail_data_alloc_bits;
3530 else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3531 flags |= root->fs_info->avail_system_alloc_bits;
3532 else if (flags & BTRFS_BLOCK_GROUP_METADATA)
3533 flags |= root->fs_info->avail_metadata_alloc_bits;
3534 } while (read_seqretry(&root->fs_info->profiles_lock, seq));
Ilya Dryomov6fef8df2012-01-16 22:04:47 +02003535
Yan, Zhengb742bb82010-05-16 10:46:24 -04003536 return btrfs_reduce_alloc_profile(root, flags);
3537}
Josef Bacik6a632092009-02-20 11:00:09 -05003538
Miao Xie6d07bce2011-01-05 10:07:31 +00003539u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data)
Yan, Zhengb742bb82010-05-16 10:46:24 -04003540{
3541 u64 flags;
David Woodhouse53b381b2013-01-29 18:40:14 -05003542 u64 ret;
Josef Bacik6a632092009-02-20 11:00:09 -05003543
Yan, Zhengb742bb82010-05-16 10:46:24 -04003544 if (data)
3545 flags = BTRFS_BLOCK_GROUP_DATA;
3546 else if (root == root->fs_info->chunk_root)
3547 flags = BTRFS_BLOCK_GROUP_SYSTEM;
3548 else
3549 flags = BTRFS_BLOCK_GROUP_METADATA;
3550
David Woodhouse53b381b2013-01-29 18:40:14 -05003551 ret = get_alloc_profile(root, flags);
3552 return ret;
Josef Bacik6a632092009-02-20 11:00:09 -05003553}
3554
Josef Bacik6a632092009-02-20 11:00:09 -05003555/*
3556 * This will check the space that the inode allocates from to make sure we have
3557 * enough space for bytes.
3558 */
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04003559int btrfs_check_data_free_space(struct inode *inode, u64 bytes)
Josef Bacik6a632092009-02-20 11:00:09 -05003560{
3561 struct btrfs_space_info *data_sinfo;
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04003562 struct btrfs_root *root = BTRFS_I(inode)->root;
Li Zefanb4d7c3c2012-07-09 20:21:07 -06003563 struct btrfs_fs_info *fs_info = root->fs_info;
Josef Bacikab6e24102010-03-19 14:38:13 +00003564 u64 used;
Josef Bacik0af3d002010-06-21 14:48:16 -04003565 int ret = 0, committed = 0, alloc_chunk = 1;
Josef Bacik6a632092009-02-20 11:00:09 -05003566
3567 /* make sure bytes are sectorsize aligned */
Qu Wenruofda28322013-02-26 08:10:22 +00003568 bytes = ALIGN(bytes, root->sectorsize);
Josef Bacik6a632092009-02-20 11:00:09 -05003569
Li Zefan82d59022011-04-20 10:33:24 +08003570 if (root == root->fs_info->tree_root ||
3571 BTRFS_I(inode)->location.objectid == BTRFS_FREE_INO_OBJECTID) {
Josef Bacik0af3d002010-06-21 14:48:16 -04003572 alloc_chunk = 0;
3573 committed = 1;
3574 }
3575
Li Zefanb4d7c3c2012-07-09 20:21:07 -06003576 data_sinfo = fs_info->data_sinfo;
Chris Mason33b4d472009-09-22 14:45:50 -04003577 if (!data_sinfo)
3578 goto alloc;
3579
Josef Bacik6a632092009-02-20 11:00:09 -05003580again:
3581 /* make sure we have enough space to handle the data first */
3582 spin_lock(&data_sinfo->lock);
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04003583 used = data_sinfo->bytes_used + data_sinfo->bytes_reserved +
3584 data_sinfo->bytes_pinned + data_sinfo->bytes_readonly +
3585 data_sinfo->bytes_may_use;
Josef Bacikab6e24102010-03-19 14:38:13 +00003586
3587 if (used + bytes > data_sinfo->total_bytes) {
Josef Bacik4e06bdd2009-02-20 10:59:53 -05003588 struct btrfs_trans_handle *trans;
3589
Josef Bacik6a632092009-02-20 11:00:09 -05003590 /*
3591 * if we don't have enough free bytes in this space then we need
3592 * to alloc a new chunk.
3593 */
Josef Bacik0af3d002010-06-21 14:48:16 -04003594 if (!data_sinfo->full && alloc_chunk) {
Josef Bacik6a632092009-02-20 11:00:09 -05003595 u64 alloc_target;
Josef Bacik6a632092009-02-20 11:00:09 -05003596
Chris Mason0e4f8f82011-04-15 16:05:44 -04003597 data_sinfo->force_alloc = CHUNK_ALLOC_FORCE;
Josef Bacik6a632092009-02-20 11:00:09 -05003598 spin_unlock(&data_sinfo->lock);
Chris Mason33b4d472009-09-22 14:45:50 -04003599alloc:
Josef Bacik6a632092009-02-20 11:00:09 -05003600 alloc_target = btrfs_get_alloc_profile(root, 1);
Josef Bacik7a7eaa42011-04-13 12:54:33 -04003601 trans = btrfs_join_transaction(root);
Yan, Zhenga22285a2010-05-16 10:48:46 -04003602 if (IS_ERR(trans))
3603 return PTR_ERR(trans);
Josef Bacik6a632092009-02-20 11:00:09 -05003604
3605 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
Chris Mason0e4f8f82011-04-15 16:05:44 -04003606 alloc_target,
3607 CHUNK_ALLOC_NO_FORCE);
Josef Bacik6a632092009-02-20 11:00:09 -05003608 btrfs_end_transaction(trans, root);
Miao Xied52a5b52011-01-05 10:07:18 +00003609 if (ret < 0) {
3610 if (ret != -ENOSPC)
3611 return ret;
3612 else
3613 goto commit_trans;
3614 }
Chris Mason33b4d472009-09-22 14:45:50 -04003615
Li Zefanb4d7c3c2012-07-09 20:21:07 -06003616 if (!data_sinfo)
3617 data_sinfo = fs_info->data_sinfo;
3618
Josef Bacik6a632092009-02-20 11:00:09 -05003619 goto again;
3620 }
Josef Bacikf2bb8f52011-05-25 13:10:16 -04003621
3622 /*
Josef Bacikb150a4f2013-06-19 15:00:04 -04003623 * If we don't have enough pinned space to deal with this
3624 * allocation don't bother committing the transaction.
Josef Bacikf2bb8f52011-05-25 13:10:16 -04003625 */
Josef Bacikb150a4f2013-06-19 15:00:04 -04003626 if (percpu_counter_compare(&data_sinfo->total_bytes_pinned,
3627 bytes) < 0)
Josef Bacikf2bb8f52011-05-25 13:10:16 -04003628 committed = 1;
Josef Bacik6a632092009-02-20 11:00:09 -05003629 spin_unlock(&data_sinfo->lock);
Josef Bacik4e06bdd2009-02-20 10:59:53 -05003630
3631 /* commit the current transaction and try again */
Miao Xied52a5b52011-01-05 10:07:18 +00003632commit_trans:
Josef Bacika4abeea2011-04-11 17:25:13 -04003633 if (!committed &&
3634 !atomic_read(&root->fs_info->open_ioctl_trans)) {
Josef Bacik4e06bdd2009-02-20 10:59:53 -05003635 committed = 1;
Josef Bacikb150a4f2013-06-19 15:00:04 -04003636
Josef Bacik7a7eaa42011-04-13 12:54:33 -04003637 trans = btrfs_join_transaction(root);
Yan, Zhenga22285a2010-05-16 10:48:46 -04003638 if (IS_ERR(trans))
3639 return PTR_ERR(trans);
Josef Bacik4e06bdd2009-02-20 10:59:53 -05003640 ret = btrfs_commit_transaction(trans, root);
3641 if (ret)
3642 return ret;
3643 goto again;
3644 }
3645
Josef Bacik6a632092009-02-20 11:00:09 -05003646 return -ENOSPC;
3647 }
3648 data_sinfo->bytes_may_use += bytes;
Josef Bacik8c2a3ca2012-01-10 10:31:31 -05003649 trace_btrfs_space_reservation(root->fs_info, "space_info",
Liu Bo2bcc0322012-03-29 09:57:44 -04003650 data_sinfo->flags, bytes, 1);
Josef Bacik6a632092009-02-20 11:00:09 -05003651 spin_unlock(&data_sinfo->lock);
3652
Josef Bacik9ed74f22009-09-11 16:12:44 -04003653 return 0;
Josef Bacik6a632092009-02-20 11:00:09 -05003654}
3655
3656/*
Josef Bacikfb25e912011-07-26 17:00:46 -04003657 * Called if we need to clear a data reservation for this inode.
Josef Bacik6a632092009-02-20 11:00:09 -05003658 */
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04003659void btrfs_free_reserved_data_space(struct inode *inode, u64 bytes)
Josef Bacik6a632092009-02-20 11:00:09 -05003660{
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04003661 struct btrfs_root *root = BTRFS_I(inode)->root;
Josef Bacik6a632092009-02-20 11:00:09 -05003662 struct btrfs_space_info *data_sinfo;
3663
3664 /* make sure bytes are sectorsize aligned */
Qu Wenruofda28322013-02-26 08:10:22 +00003665 bytes = ALIGN(bytes, root->sectorsize);
Josef Bacik6a632092009-02-20 11:00:09 -05003666
Li Zefanb4d7c3c2012-07-09 20:21:07 -06003667 data_sinfo = root->fs_info->data_sinfo;
Josef Bacik6a632092009-02-20 11:00:09 -05003668 spin_lock(&data_sinfo->lock);
Josef Bacik7ee9e442013-06-21 16:37:03 -04003669 WARN_ON(data_sinfo->bytes_may_use < bytes);
Josef Bacik6a632092009-02-20 11:00:09 -05003670 data_sinfo->bytes_may_use -= bytes;
Josef Bacik8c2a3ca2012-01-10 10:31:31 -05003671 trace_btrfs_space_reservation(root->fs_info, "space_info",
Liu Bo2bcc0322012-03-29 09:57:44 -04003672 data_sinfo->flags, bytes, 0);
Josef Bacik6a632092009-02-20 11:00:09 -05003673 spin_unlock(&data_sinfo->lock);
3674}
3675
Josef Bacik97e728d2009-04-21 17:40:57 -04003676static void force_metadata_allocation(struct btrfs_fs_info *info)
3677{
3678 struct list_head *head = &info->space_info;
3679 struct btrfs_space_info *found;
3680
3681 rcu_read_lock();
3682 list_for_each_entry_rcu(found, head, list) {
3683 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
Chris Mason0e4f8f82011-04-15 16:05:44 -04003684 found->force_alloc = CHUNK_ALLOC_FORCE;
Josef Bacik97e728d2009-04-21 17:40:57 -04003685 }
3686 rcu_read_unlock();
3687}
3688
Miao Xie3c76cd82013-04-25 10:12:38 +00003689static inline u64 calc_global_rsv_need_space(struct btrfs_block_rsv *global)
3690{
3691 return (global->size << 1);
3692}
3693
Chris Masone5bc2452010-10-26 13:37:56 -04003694static int should_alloc_chunk(struct btrfs_root *root,
Josef Bacik698d0082012-09-12 14:08:47 -04003695 struct btrfs_space_info *sinfo, int force)
Yan, Zheng424499d2010-05-16 10:46:25 -04003696{
Josef Bacikfb25e912011-07-26 17:00:46 -04003697 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
Yan, Zheng424499d2010-05-16 10:46:25 -04003698 u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly;
Chris Mason0e4f8f82011-04-15 16:05:44 -04003699 u64 num_allocated = sinfo->bytes_used + sinfo->bytes_reserved;
Chris Masone5bc2452010-10-26 13:37:56 -04003700 u64 thresh;
Yan, Zheng424499d2010-05-16 10:46:25 -04003701
Chris Mason0e4f8f82011-04-15 16:05:44 -04003702 if (force == CHUNK_ALLOC_FORCE)
3703 return 1;
3704
3705 /*
Josef Bacikfb25e912011-07-26 17:00:46 -04003706 * We need to take into account the global rsv because for all intents
3707 * and purposes it's used space. Don't worry about locking the
3708 * global_rsv, it doesn't change except when the transaction commits.
3709 */
Josef Bacik54338b52012-08-14 16:20:52 -04003710 if (sinfo->flags & BTRFS_BLOCK_GROUP_METADATA)
Miao Xie3c76cd82013-04-25 10:12:38 +00003711 num_allocated += calc_global_rsv_need_space(global_rsv);
Josef Bacikfb25e912011-07-26 17:00:46 -04003712
3713 /*
Chris Mason0e4f8f82011-04-15 16:05:44 -04003714 * in limited mode, we want to have some free space up to
3715 * about 1% of the FS size.
3716 */
3717 if (force == CHUNK_ALLOC_LIMITED) {
David Sterba6c417612011-04-13 15:41:04 +02003718 thresh = btrfs_super_total_bytes(root->fs_info->super_copy);
Chris Mason0e4f8f82011-04-15 16:05:44 -04003719 thresh = max_t(u64, 64 * 1024 * 1024,
3720 div_factor_fine(thresh, 1));
3721
3722 if (num_bytes - num_allocated < thresh)
3723 return 1;
3724 }
Chris Mason0e4f8f82011-04-15 16:05:44 -04003725
Josef Bacik698d0082012-09-12 14:08:47 -04003726 if (num_allocated + 2 * 1024 * 1024 < div_factor(num_bytes, 8))
Josef Bacik14ed0ca2010-10-15 15:23:48 -04003727 return 0;
Yan, Zheng424499d2010-05-16 10:46:25 -04003728 return 1;
3729}
3730
Liu Bo15d1ff82012-03-29 09:57:44 -04003731static u64 get_system_chunk_thresh(struct btrfs_root *root, u64 type)
3732{
3733 u64 num_dev;
3734
David Woodhouse53b381b2013-01-29 18:40:14 -05003735 if (type & (BTRFS_BLOCK_GROUP_RAID10 |
3736 BTRFS_BLOCK_GROUP_RAID0 |
3737 BTRFS_BLOCK_GROUP_RAID5 |
3738 BTRFS_BLOCK_GROUP_RAID6))
Liu Bo15d1ff82012-03-29 09:57:44 -04003739 num_dev = root->fs_info->fs_devices->rw_devices;
3740 else if (type & BTRFS_BLOCK_GROUP_RAID1)
3741 num_dev = 2;
3742 else
3743 num_dev = 1; /* DUP or single */
3744
3745 /* metadata for updaing devices and chunk tree */
3746 return btrfs_calc_trans_metadata_size(root, num_dev + 1);
3747}
3748
3749static void check_system_chunk(struct btrfs_trans_handle *trans,
3750 struct btrfs_root *root, u64 type)
3751{
3752 struct btrfs_space_info *info;
3753 u64 left;
3754 u64 thresh;
3755
3756 info = __find_space_info(root->fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
3757 spin_lock(&info->lock);
3758 left = info->total_bytes - info->bytes_used - info->bytes_pinned -
3759 info->bytes_reserved - info->bytes_readonly;
3760 spin_unlock(&info->lock);
3761
3762 thresh = get_system_chunk_thresh(root, type);
3763 if (left < thresh && btrfs_test_opt(root, ENOSPC_DEBUG)) {
Simon Kirbyc2cf52e2013-03-19 22:41:23 +00003764 btrfs_info(root->fs_info, "left=%llu, need=%llu, flags=%llu",
3765 left, thresh, type);
Liu Bo15d1ff82012-03-29 09:57:44 -04003766 dump_space_info(info, 0, 0);
3767 }
3768
3769 if (left < thresh) {
3770 u64 flags;
3771
3772 flags = btrfs_get_alloc_profile(root->fs_info->chunk_root, 0);
3773 btrfs_alloc_chunk(trans, root, flags);
3774 }
3775}
3776
Chris Mason6324fbf2008-03-24 15:01:59 -04003777static int do_chunk_alloc(struct btrfs_trans_handle *trans,
Josef Bacik698d0082012-09-12 14:08:47 -04003778 struct btrfs_root *extent_root, u64 flags, int force)
Chris Mason6324fbf2008-03-24 15:01:59 -04003779{
3780 struct btrfs_space_info *space_info;
Josef Bacik97e728d2009-04-21 17:40:57 -04003781 struct btrfs_fs_info *fs_info = extent_root->fs_info;
Josef Bacik6d741192011-04-11 20:20:11 -04003782 int wait_for_alloc = 0;
Yan Zhengc146afa2008-11-12 14:34:12 -05003783 int ret = 0;
3784
Josef Bacikc6b305a2012-12-18 09:16:16 -05003785 /* Don't re-enter if we're already allocating a chunk */
3786 if (trans->allocating_chunk)
3787 return -ENOSPC;
3788
Chris Mason6324fbf2008-03-24 15:01:59 -04003789 space_info = __find_space_info(extent_root->fs_info, flags);
Chris Mason593060d2008-03-25 16:50:33 -04003790 if (!space_info) {
3791 ret = update_space_info(extent_root->fs_info, flags,
3792 0, 0, &space_info);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01003793 BUG_ON(ret); /* -ENOMEM */
Chris Mason593060d2008-03-25 16:50:33 -04003794 }
Jeff Mahoney79787ea2012-03-12 16:03:00 +01003795 BUG_ON(!space_info); /* Logic error */
Chris Mason6324fbf2008-03-24 15:01:59 -04003796
Josef Bacik6d741192011-04-11 20:20:11 -04003797again:
Josef Bacik25179202008-10-29 14:49:05 -04003798 spin_lock(&space_info->lock);
Miao Xie9e622d62012-01-26 15:01:12 -05003799 if (force < space_info->force_alloc)
Chris Mason0e4f8f82011-04-15 16:05:44 -04003800 force = space_info->force_alloc;
Josef Bacik25179202008-10-29 14:49:05 -04003801 if (space_info->full) {
3802 spin_unlock(&space_info->lock);
Josef Bacik6d741192011-04-11 20:20:11 -04003803 return 0;
Josef Bacik25179202008-10-29 14:49:05 -04003804 }
Chris Mason6324fbf2008-03-24 15:01:59 -04003805
Josef Bacik698d0082012-09-12 14:08:47 -04003806 if (!should_alloc_chunk(extent_root, space_info, force)) {
Josef Bacik25179202008-10-29 14:49:05 -04003807 spin_unlock(&space_info->lock);
Josef Bacik6d741192011-04-11 20:20:11 -04003808 return 0;
3809 } else if (space_info->chunk_alloc) {
3810 wait_for_alloc = 1;
3811 } else {
3812 space_info->chunk_alloc = 1;
Josef Bacik25179202008-10-29 14:49:05 -04003813 }
Chris Mason0e4f8f82011-04-15 16:05:44 -04003814
Josef Bacik25179202008-10-29 14:49:05 -04003815 spin_unlock(&space_info->lock);
3816
Josef Bacik6d741192011-04-11 20:20:11 -04003817 mutex_lock(&fs_info->chunk_mutex);
3818
3819 /*
3820 * The chunk_mutex is held throughout the entirety of a chunk
3821 * allocation, so once we've acquired the chunk_mutex we know that the
3822 * other guy is done and we need to recheck and see if we should
3823 * allocate.
3824 */
3825 if (wait_for_alloc) {
3826 mutex_unlock(&fs_info->chunk_mutex);
3827 wait_for_alloc = 0;
3828 goto again;
3829 }
3830
Josef Bacikc6b305a2012-12-18 09:16:16 -05003831 trans->allocating_chunk = true;
3832
Josef Bacik97e728d2009-04-21 17:40:57 -04003833 /*
Josef Bacik67377732010-09-16 16:19:09 -04003834 * If we have mixed data/metadata chunks we want to make sure we keep
3835 * allocating mixed chunks instead of individual chunks.
3836 */
3837 if (btrfs_mixed_space_info(space_info))
3838 flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
3839
3840 /*
Josef Bacik97e728d2009-04-21 17:40:57 -04003841 * if we're doing a data chunk, go ahead and make sure that
3842 * we keep a reasonable number of metadata chunks allocated in the
3843 * FS as well.
3844 */
Josef Bacik9ed74f22009-09-11 16:12:44 -04003845 if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
Josef Bacik97e728d2009-04-21 17:40:57 -04003846 fs_info->data_chunk_allocations++;
3847 if (!(fs_info->data_chunk_allocations %
3848 fs_info->metadata_ratio))
3849 force_metadata_allocation(fs_info);
3850 }
3851
Liu Bo15d1ff82012-03-29 09:57:44 -04003852 /*
3853 * Check if we have enough space in SYSTEM chunk because we may need
3854 * to update devices.
3855 */
3856 check_system_chunk(trans, extent_root, flags);
3857
Yan Zheng2b820322008-11-17 21:11:30 -05003858 ret = btrfs_alloc_chunk(trans, extent_root, flags);
Josef Bacikc6b305a2012-12-18 09:16:16 -05003859 trans->allocating_chunk = false;
Mark Fasheh92b8e8972011-07-12 10:57:59 -07003860
Josef Bacik9ed74f22009-09-11 16:12:44 -04003861 spin_lock(&space_info->lock);
Alexandre Olivaa81cb9a2013-02-21 21:15:14 +00003862 if (ret < 0 && ret != -ENOSPC)
3863 goto out;
Chris Masond3977122009-01-05 21:25:51 -05003864 if (ret)
Chris Mason6324fbf2008-03-24 15:01:59 -04003865 space_info->full = 1;
Yan, Zheng424499d2010-05-16 10:46:25 -04003866 else
3867 ret = 1;
Josef Bacik6d741192011-04-11 20:20:11 -04003868
Chris Mason0e4f8f82011-04-15 16:05:44 -04003869 space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
Alexandre Olivaa81cb9a2013-02-21 21:15:14 +00003870out:
Josef Bacik6d741192011-04-11 20:20:11 -04003871 space_info->chunk_alloc = 0;
Josef Bacik9ed74f22009-09-11 16:12:44 -04003872 spin_unlock(&space_info->lock);
Dan Carpentera25c75d2012-04-18 09:59:29 +03003873 mutex_unlock(&fs_info->chunk_mutex);
Josef Bacik0f9dd462008-09-23 13:14:11 -04003874 return ret;
Chris Mason6324fbf2008-03-24 15:01:59 -04003875}
3876
Josef Bacika80c8dc2012-09-06 16:59:33 -04003877static int can_overcommit(struct btrfs_root *root,
3878 struct btrfs_space_info *space_info, u64 bytes,
Miao Xie08e007d2012-10-16 11:33:38 +00003879 enum btrfs_reserve_flush_enum flush)
Josef Bacika80c8dc2012-09-06 16:59:33 -04003880{
Josef Bacik96f1bb52013-01-30 17:02:51 -05003881 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
Josef Bacika80c8dc2012-09-06 16:59:33 -04003882 u64 profile = btrfs_get_alloc_profile(root, 0);
Miao Xie3c76cd82013-04-25 10:12:38 +00003883 u64 space_size;
Josef Bacika80c8dc2012-09-06 16:59:33 -04003884 u64 avail;
3885 u64 used;
Josef Bacik70afa392013-02-06 13:53:19 -05003886 u64 to_add;
Josef Bacika80c8dc2012-09-06 16:59:33 -04003887
3888 used = space_info->bytes_used + space_info->bytes_reserved +
Josef Bacik96f1bb52013-01-30 17:02:51 -05003889 space_info->bytes_pinned + space_info->bytes_readonly;
3890
Josef Bacik96f1bb52013-01-30 17:02:51 -05003891 /*
3892 * We only want to allow over committing if we have lots of actual space
3893 * free, but if we don't have enough space to handle the global reserve
3894 * space then we could end up having a real enospc problem when trying
3895 * to allocate a chunk or some other such important allocation.
3896 */
Miao Xie3c76cd82013-04-25 10:12:38 +00003897 spin_lock(&global_rsv->lock);
3898 space_size = calc_global_rsv_need_space(global_rsv);
3899 spin_unlock(&global_rsv->lock);
3900 if (used + space_size >= space_info->total_bytes)
Josef Bacik96f1bb52013-01-30 17:02:51 -05003901 return 0;
3902
3903 used += space_info->bytes_may_use;
Josef Bacika80c8dc2012-09-06 16:59:33 -04003904
3905 spin_lock(&root->fs_info->free_chunk_lock);
3906 avail = root->fs_info->free_chunk_space;
3907 spin_unlock(&root->fs_info->free_chunk_lock);
3908
3909 /*
3910 * If we have dup, raid1 or raid10 then only half of the free
David Woodhouse53b381b2013-01-29 18:40:14 -05003911 * space is actually useable. For raid56, the space info used
3912 * doesn't include the parity drive, so we don't have to
3913 * change the math
Josef Bacika80c8dc2012-09-06 16:59:33 -04003914 */
3915 if (profile & (BTRFS_BLOCK_GROUP_DUP |
3916 BTRFS_BLOCK_GROUP_RAID1 |
3917 BTRFS_BLOCK_GROUP_RAID10))
3918 avail >>= 1;
3919
Josef Bacik70afa392013-02-06 13:53:19 -05003920 to_add = space_info->total_bytes;
3921
Josef Bacika80c8dc2012-09-06 16:59:33 -04003922 /*
Miao Xie561c2942012-10-16 11:32:18 +00003923 * If we aren't flushing all things, let us overcommit up to
3924 * 1/2th of the space. If we can flush, don't let us overcommit
3925 * too much, let it overcommit up to 1/8 of the space.
Josef Bacika80c8dc2012-09-06 16:59:33 -04003926 */
Miao Xie08e007d2012-10-16 11:33:38 +00003927 if (flush == BTRFS_RESERVE_FLUSH_ALL)
Josef Bacik70afa392013-02-06 13:53:19 -05003928 to_add >>= 3;
Josef Bacika80c8dc2012-09-06 16:59:33 -04003929 else
Josef Bacik70afa392013-02-06 13:53:19 -05003930 to_add >>= 1;
Josef Bacika80c8dc2012-09-06 16:59:33 -04003931
Josef Bacik70afa392013-02-06 13:53:19 -05003932 /*
3933 * Limit the overcommit to the amount of free space we could possibly
3934 * allocate for chunks.
3935 */
3936 to_add = min(avail, to_add);
3937
3938 if (used + bytes < space_info->total_bytes + to_add)
Josef Bacika80c8dc2012-09-06 16:59:33 -04003939 return 1;
3940 return 0;
3941}
3942
Eric Sandeen48a3b632013-04-25 20:41:01 +00003943static void btrfs_writeback_inodes_sb_nr(struct btrfs_root *root,
3944 unsigned long nr_pages)
Miao Xieda633a42012-12-20 11:19:09 +00003945{
3946 struct super_block *sb = root->fs_info->sb;
Miao Xieda633a42012-12-20 11:19:09 +00003947
Josef Bacik925a6ef2013-06-20 12:31:27 -04003948 if (down_read_trylock(&sb->s_umount)) {
3949 writeback_inodes_sb_nr(sb, nr_pages, WB_REASON_FS_FREE_SPACE);
3950 up_read(&sb->s_umount);
3951 } else {
Miao Xieda633a42012-12-20 11:19:09 +00003952 /*
3953 * We needn't worry the filesystem going from r/w to r/o though
3954 * we don't acquire ->s_umount mutex, because the filesystem
3955 * should guarantee the delalloc inodes list be empty after
3956 * the filesystem is readonly(all dirty pages are written to
3957 * the disk).
3958 */
Miao Xieeb73c1b2013-05-15 07:48:22 +00003959 btrfs_start_all_delalloc_inodes(root->fs_info, 0);
Josef Bacik98ad69c2013-04-04 11:55:49 -04003960 if (!current->journal_info)
Miao Xie199c2a92013-05-15 07:48:23 +00003961 btrfs_wait_all_ordered_extents(root->fs_info, 0);
Miao Xieda633a42012-12-20 11:19:09 +00003962 }
3963}
3964
Yan, Zheng5da9d012010-05-16 10:46:25 -04003965/*
3966 * shrink metadata reservation for delalloc
3967 */
Josef Bacikf4c738c2012-07-02 17:10:51 -04003968static void shrink_delalloc(struct btrfs_root *root, u64 to_reclaim, u64 orig,
3969 bool wait_ordered)
Yan, Zheng5da9d012010-05-16 10:46:25 -04003970{
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04003971 struct btrfs_block_rsv *block_rsv;
Josef Bacik0019f102010-10-15 15:18:40 -04003972 struct btrfs_space_info *space_info;
Josef Bacik663350a2011-11-03 22:54:25 -04003973 struct btrfs_trans_handle *trans;
Josef Bacikf4c738c2012-07-02 17:10:51 -04003974 u64 delalloc_bytes;
Yan, Zheng5da9d012010-05-16 10:46:25 -04003975 u64 max_reclaim;
Josef Bacikb1953bc2011-01-21 21:10:01 +00003976 long time_left;
Josef Bacik877da172011-10-14 14:02:10 -04003977 unsigned long nr_pages = (2 * 1024 * 1024) >> PAGE_CACHE_SHIFT;
Josef Bacikb1953bc2011-01-21 21:10:01 +00003978 int loops = 0;
Miao Xie08e007d2012-10-16 11:33:38 +00003979 enum btrfs_reserve_flush_enum flush;
Yan, Zheng5da9d012010-05-16 10:46:25 -04003980
Josef Bacik663350a2011-11-03 22:54:25 -04003981 trans = (struct btrfs_trans_handle *)current->journal_info;
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04003982 block_rsv = &root->fs_info->delalloc_block_rsv;
Josef Bacik0019f102010-10-15 15:18:40 -04003983 space_info = block_rsv->space_info;
Chris Masonbf9022e2010-10-26 13:40:45 -04003984
3985 smp_mb();
Miao Xie963d6782013-01-29 10:10:51 +00003986 delalloc_bytes = percpu_counter_sum_positive(
3987 &root->fs_info->delalloc_bytes);
Josef Bacikf4c738c2012-07-02 17:10:51 -04003988 if (delalloc_bytes == 0) {
Josef Bacikfdb5eff2011-06-07 16:07:44 -04003989 if (trans)
Josef Bacikf4c738c2012-07-02 17:10:51 -04003990 return;
Miao Xie199c2a92013-05-15 07:48:23 +00003991 btrfs_wait_all_ordered_extents(root->fs_info, 0);
Josef Bacikf4c738c2012-07-02 17:10:51 -04003992 return;
Josef Bacikfdb5eff2011-06-07 16:07:44 -04003993 }
3994
Josef Bacikf4c738c2012-07-02 17:10:51 -04003995 while (delalloc_bytes && loops < 3) {
3996 max_reclaim = min(delalloc_bytes, to_reclaim);
3997 nr_pages = max_reclaim >> PAGE_CACHE_SHIFT;
Miao Xieda633a42012-12-20 11:19:09 +00003998 btrfs_writeback_inodes_sb_nr(root, nr_pages);
Josef Bacikdea31f52012-09-06 16:47:00 -04003999 /*
4000 * We need to wait for the async pages to actually start before
4001 * we do anything.
4002 */
4003 wait_event(root->fs_info->async_submit_wait,
4004 !atomic_read(&root->fs_info->async_delalloc_pages));
4005
Miao Xie08e007d2012-10-16 11:33:38 +00004006 if (!trans)
4007 flush = BTRFS_RESERVE_FLUSH_ALL;
4008 else
4009 flush = BTRFS_RESERVE_NO_FLUSH;
Josef Bacik0019f102010-10-15 15:18:40 -04004010 spin_lock(&space_info->lock);
Miao Xie08e007d2012-10-16 11:33:38 +00004011 if (can_overcommit(root, space_info, orig, flush)) {
Josef Bacikf4c738c2012-07-02 17:10:51 -04004012 spin_unlock(&space_info->lock);
4013 break;
4014 }
Josef Bacik0019f102010-10-15 15:18:40 -04004015 spin_unlock(&space_info->lock);
Yan, Zheng5da9d012010-05-16 10:46:25 -04004016
Chris Mason36e39c42011-03-12 07:08:42 -05004017 loops++;
Josef Bacikf104d042011-10-14 13:56:58 -04004018 if (wait_ordered && !trans) {
Miao Xie199c2a92013-05-15 07:48:23 +00004019 btrfs_wait_all_ordered_extents(root->fs_info, 0);
Josef Bacikf104d042011-10-14 13:56:58 -04004020 } else {
Josef Bacikf4c738c2012-07-02 17:10:51 -04004021 time_left = schedule_timeout_killable(1);
Josef Bacikf104d042011-10-14 13:56:58 -04004022 if (time_left)
4023 break;
4024 }
Josef Bacikf4c738c2012-07-02 17:10:51 -04004025 smp_mb();
Miao Xie963d6782013-01-29 10:10:51 +00004026 delalloc_bytes = percpu_counter_sum_positive(
4027 &root->fs_info->delalloc_bytes);
Yan, Zheng5da9d012010-05-16 10:46:25 -04004028 }
Yan, Zheng5da9d012010-05-16 10:46:25 -04004029}
4030
Josef Bacik4a92b1b2011-08-30 12:34:28 -04004031/**
Josef Bacik663350a2011-11-03 22:54:25 -04004032 * maybe_commit_transaction - possibly commit the transaction if its ok to
4033 * @root - the root we're allocating for
4034 * @bytes - the number of bytes we want to reserve
4035 * @force - force the commit
Josef Bacik8bb8ab22010-10-15 16:52:49 -04004036 *
Josef Bacik663350a2011-11-03 22:54:25 -04004037 * This will check to make sure that committing the transaction will actually
4038 * get us somewhere and then commit the transaction if it does. Otherwise it
4039 * will return -ENOSPC.
Josef Bacik8bb8ab22010-10-15 16:52:49 -04004040 */
Josef Bacik663350a2011-11-03 22:54:25 -04004041static int may_commit_transaction(struct btrfs_root *root,
4042 struct btrfs_space_info *space_info,
4043 u64 bytes, int force)
4044{
4045 struct btrfs_block_rsv *delayed_rsv = &root->fs_info->delayed_block_rsv;
4046 struct btrfs_trans_handle *trans;
4047
4048 trans = (struct btrfs_trans_handle *)current->journal_info;
4049 if (trans)
4050 return -EAGAIN;
4051
4052 if (force)
4053 goto commit;
4054
4055 /* See if there is enough pinned space to make this reservation */
4056 spin_lock(&space_info->lock);
Josef Bacikb150a4f2013-06-19 15:00:04 -04004057 if (percpu_counter_compare(&space_info->total_bytes_pinned,
4058 bytes) >= 0) {
Josef Bacik663350a2011-11-03 22:54:25 -04004059 spin_unlock(&space_info->lock);
4060 goto commit;
4061 }
4062 spin_unlock(&space_info->lock);
4063
4064 /*
4065 * See if there is some space in the delayed insertion reservation for
4066 * this reservation.
4067 */
4068 if (space_info != delayed_rsv->space_info)
4069 return -ENOSPC;
4070
Liu Bod9b02182012-02-16 18:34:39 +08004071 spin_lock(&space_info->lock);
Josef Bacik663350a2011-11-03 22:54:25 -04004072 spin_lock(&delayed_rsv->lock);
Josef Bacikb150a4f2013-06-19 15:00:04 -04004073 if (percpu_counter_compare(&space_info->total_bytes_pinned,
4074 bytes - delayed_rsv->size) >= 0) {
Josef Bacik663350a2011-11-03 22:54:25 -04004075 spin_unlock(&delayed_rsv->lock);
Liu Bod9b02182012-02-16 18:34:39 +08004076 spin_unlock(&space_info->lock);
Josef Bacik663350a2011-11-03 22:54:25 -04004077 return -ENOSPC;
4078 }
4079 spin_unlock(&delayed_rsv->lock);
Liu Bod9b02182012-02-16 18:34:39 +08004080 spin_unlock(&space_info->lock);
Josef Bacik663350a2011-11-03 22:54:25 -04004081
4082commit:
4083 trans = btrfs_join_transaction(root);
4084 if (IS_ERR(trans))
4085 return -ENOSPC;
4086
4087 return btrfs_commit_transaction(trans, root);
4088}
4089
Josef Bacik96c3f432012-06-21 14:05:49 -04004090enum flush_state {
Josef Bacik67b0fd62012-09-24 13:42:00 -04004091 FLUSH_DELAYED_ITEMS_NR = 1,
4092 FLUSH_DELAYED_ITEMS = 2,
4093 FLUSH_DELALLOC = 3,
4094 FLUSH_DELALLOC_WAIT = 4,
Josef Bacikea658ba2012-09-11 16:57:25 -04004095 ALLOC_CHUNK = 5,
4096 COMMIT_TRANS = 6,
Josef Bacik96c3f432012-06-21 14:05:49 -04004097};
4098
4099static int flush_space(struct btrfs_root *root,
4100 struct btrfs_space_info *space_info, u64 num_bytes,
4101 u64 orig_bytes, int state)
4102{
4103 struct btrfs_trans_handle *trans;
4104 int nr;
Josef Bacikf4c738c2012-07-02 17:10:51 -04004105 int ret = 0;
Josef Bacik96c3f432012-06-21 14:05:49 -04004106
4107 switch (state) {
Josef Bacik96c3f432012-06-21 14:05:49 -04004108 case FLUSH_DELAYED_ITEMS_NR:
4109 case FLUSH_DELAYED_ITEMS:
4110 if (state == FLUSH_DELAYED_ITEMS_NR) {
4111 u64 bytes = btrfs_calc_trans_metadata_size(root, 1);
4112
4113 nr = (int)div64_u64(num_bytes, bytes);
4114 if (!nr)
4115 nr = 1;
4116 nr *= 2;
4117 } else {
4118 nr = -1;
4119 }
4120 trans = btrfs_join_transaction(root);
4121 if (IS_ERR(trans)) {
4122 ret = PTR_ERR(trans);
4123 break;
4124 }
4125 ret = btrfs_run_delayed_items_nr(trans, root, nr);
4126 btrfs_end_transaction(trans, root);
4127 break;
Josef Bacik67b0fd62012-09-24 13:42:00 -04004128 case FLUSH_DELALLOC:
4129 case FLUSH_DELALLOC_WAIT:
4130 shrink_delalloc(root, num_bytes, orig_bytes,
4131 state == FLUSH_DELALLOC_WAIT);
4132 break;
Josef Bacikea658ba2012-09-11 16:57:25 -04004133 case ALLOC_CHUNK:
4134 trans = btrfs_join_transaction(root);
4135 if (IS_ERR(trans)) {
4136 ret = PTR_ERR(trans);
4137 break;
4138 }
4139 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
Josef Bacikea658ba2012-09-11 16:57:25 -04004140 btrfs_get_alloc_profile(root, 0),
4141 CHUNK_ALLOC_NO_FORCE);
4142 btrfs_end_transaction(trans, root);
4143 if (ret == -ENOSPC)
4144 ret = 0;
4145 break;
Josef Bacik96c3f432012-06-21 14:05:49 -04004146 case COMMIT_TRANS:
4147 ret = may_commit_transaction(root, space_info, orig_bytes, 0);
4148 break;
4149 default:
4150 ret = -ENOSPC;
4151 break;
4152 }
4153
4154 return ret;
4155}
Josef Bacik663350a2011-11-03 22:54:25 -04004156/**
Josef Bacik4a92b1b2011-08-30 12:34:28 -04004157 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
4158 * @root - the root we're allocating for
4159 * @block_rsv - the block_rsv we're allocating for
4160 * @orig_bytes - the number of bytes we want
Adam Buchbinder48fc7f72012-09-19 21:48:00 -04004161 * @flush - whether or not we can flush to make our reservation
Josef Bacik8bb8ab22010-10-15 16:52:49 -04004162 *
Josef Bacik4a92b1b2011-08-30 12:34:28 -04004163 * This will reserve orgi_bytes number of bytes from the space info associated
4164 * with the block_rsv. If there is not enough space it will make an attempt to
4165 * flush out space to make room. It will do this by flushing delalloc if
4166 * possible or committing the transaction. If flush is 0 then no attempts to
4167 * regain reservations will be made and this will fail if there is not enough
4168 * space already.
Josef Bacik8bb8ab22010-10-15 16:52:49 -04004169 */
Josef Bacik4a92b1b2011-08-30 12:34:28 -04004170static int reserve_metadata_bytes(struct btrfs_root *root,
Josef Bacik8bb8ab22010-10-15 16:52:49 -04004171 struct btrfs_block_rsv *block_rsv,
Miao Xie08e007d2012-10-16 11:33:38 +00004172 u64 orig_bytes,
4173 enum btrfs_reserve_flush_enum flush)
Yan, Zhengf0486c62010-05-16 10:46:25 -04004174{
4175 struct btrfs_space_info *space_info = block_rsv->space_info;
Josef Bacik2bf64752011-09-26 17:12:22 -04004176 u64 used;
Josef Bacik8bb8ab22010-10-15 16:52:49 -04004177 u64 num_bytes = orig_bytes;
Josef Bacik67b0fd62012-09-24 13:42:00 -04004178 int flush_state = FLUSH_DELAYED_ITEMS_NR;
Josef Bacik8bb8ab22010-10-15 16:52:49 -04004179 int ret = 0;
Josef Bacikfdb5eff2011-06-07 16:07:44 -04004180 bool flushing = false;
Josef Bacik8bb8ab22010-10-15 16:52:49 -04004181
4182again:
Josef Bacikfdb5eff2011-06-07 16:07:44 -04004183 ret = 0;
Yan, Zhengf0486c62010-05-16 10:46:25 -04004184 spin_lock(&space_info->lock);
Josef Bacikfdb5eff2011-06-07 16:07:44 -04004185 /*
Miao Xie08e007d2012-10-16 11:33:38 +00004186 * We only want to wait if somebody other than us is flushing and we
4187 * are actually allowed to flush all things.
Josef Bacikfdb5eff2011-06-07 16:07:44 -04004188 */
Miao Xie08e007d2012-10-16 11:33:38 +00004189 while (flush == BTRFS_RESERVE_FLUSH_ALL && !flushing &&
4190 space_info->flush) {
Josef Bacikfdb5eff2011-06-07 16:07:44 -04004191 spin_unlock(&space_info->lock);
4192 /*
4193 * If we have a trans handle we can't wait because the flusher
4194 * may have to commit the transaction, which would mean we would
4195 * deadlock since we are waiting for the flusher to finish, but
4196 * hold the current transaction open.
4197 */
Josef Bacik663350a2011-11-03 22:54:25 -04004198 if (current->journal_info)
Josef Bacikfdb5eff2011-06-07 16:07:44 -04004199 return -EAGAIN;
Arne Jansenb9688bb2012-04-18 10:27:16 +02004200 ret = wait_event_killable(space_info->wait, !space_info->flush);
4201 /* Must have been killed, return */
4202 if (ret)
Josef Bacikfdb5eff2011-06-07 16:07:44 -04004203 return -EINTR;
4204
4205 spin_lock(&space_info->lock);
4206 }
4207
4208 ret = -ENOSPC;
Josef Bacik2bf64752011-09-26 17:12:22 -04004209 used = space_info->bytes_used + space_info->bytes_reserved +
4210 space_info->bytes_pinned + space_info->bytes_readonly +
4211 space_info->bytes_may_use;
Yan, Zhengf0486c62010-05-16 10:46:25 -04004212
Josef Bacik8bb8ab22010-10-15 16:52:49 -04004213 /*
4214 * The idea here is that we've not already over-reserved the block group
4215 * then we can go ahead and save our reservation first and then start
4216 * flushing if we need to. Otherwise if we've already overcommitted
4217 * lets start flushing stuff first and then come back and try to make
4218 * our reservation.
4219 */
Josef Bacik2bf64752011-09-26 17:12:22 -04004220 if (used <= space_info->total_bytes) {
4221 if (used + orig_bytes <= space_info->total_bytes) {
Josef Bacikfb25e912011-07-26 17:00:46 -04004222 space_info->bytes_may_use += orig_bytes;
Josef Bacik8c2a3ca2012-01-10 10:31:31 -05004223 trace_btrfs_space_reservation(root->fs_info,
Liu Bo2bcc0322012-03-29 09:57:44 -04004224 "space_info", space_info->flags, orig_bytes, 1);
Yan, Zhengf0486c62010-05-16 10:46:25 -04004225 ret = 0;
4226 } else {
Josef Bacik8bb8ab22010-10-15 16:52:49 -04004227 /*
4228 * Ok set num_bytes to orig_bytes since we aren't
4229 * overocmmitted, this way we only try and reclaim what
4230 * we need.
4231 */
4232 num_bytes = orig_bytes;
Yan, Zhengf0486c62010-05-16 10:46:25 -04004233 }
Josef Bacik8bb8ab22010-10-15 16:52:49 -04004234 } else {
4235 /*
4236 * Ok we're over committed, set num_bytes to the overcommitted
4237 * amount plus the amount of bytes that we need for this
4238 * reservation.
4239 */
Josef Bacik2bf64752011-09-26 17:12:22 -04004240 num_bytes = used - space_info->total_bytes +
Josef Bacik96c3f432012-06-21 14:05:49 -04004241 (orig_bytes * 2);
Yan, Zhengf0486c62010-05-16 10:46:25 -04004242 }
Josef Bacik8bb8ab22010-10-15 16:52:49 -04004243
Josef Bacik44734ed2012-09-28 16:04:19 -04004244 if (ret && can_overcommit(root, space_info, orig_bytes, flush)) {
4245 space_info->bytes_may_use += orig_bytes;
4246 trace_btrfs_space_reservation(root->fs_info, "space_info",
4247 space_info->flags, orig_bytes,
4248 1);
4249 ret = 0;
Josef Bacik2bf64752011-09-26 17:12:22 -04004250 }
4251
Josef Bacik8bb8ab22010-10-15 16:52:49 -04004252 /*
4253 * Couldn't make our reservation, save our place so while we're trying
4254 * to reclaim space we can actually use it instead of somebody else
4255 * stealing it from us.
Miao Xie08e007d2012-10-16 11:33:38 +00004256 *
4257 * We make the other tasks wait for the flush only when we can flush
4258 * all things.
Josef Bacik8bb8ab22010-10-15 16:52:49 -04004259 */
Josef Bacik72bcd992012-12-18 15:16:34 -05004260 if (ret && flush != BTRFS_RESERVE_NO_FLUSH) {
Josef Bacikfdb5eff2011-06-07 16:07:44 -04004261 flushing = true;
4262 space_info->flush = 1;
Josef Bacik8bb8ab22010-10-15 16:52:49 -04004263 }
4264
Yan, Zhengf0486c62010-05-16 10:46:25 -04004265 spin_unlock(&space_info->lock);
4266
Miao Xie08e007d2012-10-16 11:33:38 +00004267 if (!ret || flush == BTRFS_RESERVE_NO_FLUSH)
Josef Bacik8bb8ab22010-10-15 16:52:49 -04004268 goto out;
4269
Josef Bacik96c3f432012-06-21 14:05:49 -04004270 ret = flush_space(root, space_info, num_bytes, orig_bytes,
4271 flush_state);
4272 flush_state++;
Miao Xie08e007d2012-10-16 11:33:38 +00004273
4274 /*
4275 * If we are FLUSH_LIMIT, we can not flush delalloc, or the deadlock
4276 * would happen. So skip delalloc flush.
4277 */
4278 if (flush == BTRFS_RESERVE_FLUSH_LIMIT &&
4279 (flush_state == FLUSH_DELALLOC ||
4280 flush_state == FLUSH_DELALLOC_WAIT))
4281 flush_state = ALLOC_CHUNK;
4282
Josef Bacik96c3f432012-06-21 14:05:49 -04004283 if (!ret)
Josef Bacik8bb8ab22010-10-15 16:52:49 -04004284 goto again;
Miao Xie08e007d2012-10-16 11:33:38 +00004285 else if (flush == BTRFS_RESERVE_FLUSH_LIMIT &&
4286 flush_state < COMMIT_TRANS)
4287 goto again;
4288 else if (flush == BTRFS_RESERVE_FLUSH_ALL &&
4289 flush_state <= COMMIT_TRANS)
Josef Bacik8bb8ab22010-10-15 16:52:49 -04004290 goto again;
4291
4292out:
Josef Bacik5d803662013-02-07 16:06:02 -05004293 if (ret == -ENOSPC &&
4294 unlikely(root->orphan_cleanup_state == ORPHAN_CLEANUP_STARTED)) {
4295 struct btrfs_block_rsv *global_rsv =
4296 &root->fs_info->global_block_rsv;
4297
4298 if (block_rsv != global_rsv &&
4299 !block_rsv_use_bytes(global_rsv, orig_bytes))
4300 ret = 0;
4301 }
Josef Bacikfdb5eff2011-06-07 16:07:44 -04004302 if (flushing) {
Josef Bacik8bb8ab22010-10-15 16:52:49 -04004303 spin_lock(&space_info->lock);
Josef Bacikfdb5eff2011-06-07 16:07:44 -04004304 space_info->flush = 0;
4305 wake_up_all(&space_info->wait);
Josef Bacik8bb8ab22010-10-15 16:52:49 -04004306 spin_unlock(&space_info->lock);
4307 }
Yan, Zhengf0486c62010-05-16 10:46:25 -04004308 return ret;
4309}
4310
Jeff Mahoney79787ea2012-03-12 16:03:00 +01004311static struct btrfs_block_rsv *get_block_rsv(
4312 const struct btrfs_trans_handle *trans,
4313 const struct btrfs_root *root)
Yan, Zhengf0486c62010-05-16 10:46:25 -04004314{
Josef Bacik4c13d752011-08-30 11:31:29 -04004315 struct btrfs_block_rsv *block_rsv = NULL;
4316
Josef Bacik0e721102012-06-26 16:13:18 -04004317 if (root->ref_cows)
4318 block_rsv = trans->block_rsv;
4319
4320 if (root == root->fs_info->csum_root && trans->adding_csums)
Yan, Zhengf0486c62010-05-16 10:46:25 -04004321 block_rsv = trans->block_rsv;
Josef Bacik4c13d752011-08-30 11:31:29 -04004322
4323 if (!block_rsv)
Yan, Zhengf0486c62010-05-16 10:46:25 -04004324 block_rsv = root->block_rsv;
4325
4326 if (!block_rsv)
4327 block_rsv = &root->fs_info->empty_block_rsv;
4328
4329 return block_rsv;
4330}
4331
4332static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
4333 u64 num_bytes)
4334{
4335 int ret = -ENOSPC;
4336 spin_lock(&block_rsv->lock);
4337 if (block_rsv->reserved >= num_bytes) {
4338 block_rsv->reserved -= num_bytes;
4339 if (block_rsv->reserved < block_rsv->size)
4340 block_rsv->full = 0;
4341 ret = 0;
4342 }
4343 spin_unlock(&block_rsv->lock);
4344 return ret;
4345}
4346
4347static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
4348 u64 num_bytes, int update_size)
4349{
4350 spin_lock(&block_rsv->lock);
4351 block_rsv->reserved += num_bytes;
4352 if (update_size)
4353 block_rsv->size += num_bytes;
4354 else if (block_rsv->reserved >= block_rsv->size)
4355 block_rsv->full = 1;
4356 spin_unlock(&block_rsv->lock);
4357}
4358
Josef Bacikd52be812013-05-29 14:54:47 -04004359int btrfs_cond_migrate_bytes(struct btrfs_fs_info *fs_info,
4360 struct btrfs_block_rsv *dest, u64 num_bytes,
4361 int min_factor)
4362{
4363 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
4364 u64 min_bytes;
4365
4366 if (global_rsv->space_info != dest->space_info)
4367 return -ENOSPC;
4368
4369 spin_lock(&global_rsv->lock);
4370 min_bytes = div_factor(global_rsv->size, min_factor);
4371 if (global_rsv->reserved < min_bytes + num_bytes) {
4372 spin_unlock(&global_rsv->lock);
4373 return -ENOSPC;
4374 }
4375 global_rsv->reserved -= num_bytes;
4376 if (global_rsv->reserved < global_rsv->size)
4377 global_rsv->full = 0;
4378 spin_unlock(&global_rsv->lock);
4379
4380 block_rsv_add_bytes(dest, num_bytes, 1);
4381 return 0;
4382}
4383
Josef Bacik8c2a3ca2012-01-10 10:31:31 -05004384static void block_rsv_release_bytes(struct btrfs_fs_info *fs_info,
4385 struct btrfs_block_rsv *block_rsv,
David Sterba62a45b62011-04-20 15:52:26 +02004386 struct btrfs_block_rsv *dest, u64 num_bytes)
Yan, Zhengf0486c62010-05-16 10:46:25 -04004387{
4388 struct btrfs_space_info *space_info = block_rsv->space_info;
4389
4390 spin_lock(&block_rsv->lock);
4391 if (num_bytes == (u64)-1)
4392 num_bytes = block_rsv->size;
4393 block_rsv->size -= num_bytes;
4394 if (block_rsv->reserved >= block_rsv->size) {
4395 num_bytes = block_rsv->reserved - block_rsv->size;
4396 block_rsv->reserved = block_rsv->size;
4397 block_rsv->full = 1;
4398 } else {
4399 num_bytes = 0;
4400 }
4401 spin_unlock(&block_rsv->lock);
4402
4403 if (num_bytes > 0) {
4404 if (dest) {
Josef Bacike9e22892011-01-24 21:43:19 +00004405 spin_lock(&dest->lock);
4406 if (!dest->full) {
4407 u64 bytes_to_add;
4408
4409 bytes_to_add = dest->size - dest->reserved;
4410 bytes_to_add = min(num_bytes, bytes_to_add);
4411 dest->reserved += bytes_to_add;
4412 if (dest->reserved >= dest->size)
4413 dest->full = 1;
4414 num_bytes -= bytes_to_add;
4415 }
4416 spin_unlock(&dest->lock);
4417 }
4418 if (num_bytes) {
Yan, Zhengf0486c62010-05-16 10:46:25 -04004419 spin_lock(&space_info->lock);
Josef Bacikfb25e912011-07-26 17:00:46 -04004420 space_info->bytes_may_use -= num_bytes;
Josef Bacik8c2a3ca2012-01-10 10:31:31 -05004421 trace_btrfs_space_reservation(fs_info, "space_info",
Liu Bo2bcc0322012-03-29 09:57:44 -04004422 space_info->flags, num_bytes, 0);
Chris Mason36e39c42011-03-12 07:08:42 -05004423 space_info->reservation_progress++;
Yan, Zhengf0486c62010-05-16 10:46:25 -04004424 spin_unlock(&space_info->lock);
4425 }
4426 }
4427}
4428
4429static int block_rsv_migrate_bytes(struct btrfs_block_rsv *src,
4430 struct btrfs_block_rsv *dst, u64 num_bytes)
4431{
4432 int ret;
4433
4434 ret = block_rsv_use_bytes(src, num_bytes);
4435 if (ret)
4436 return ret;
4437
4438 block_rsv_add_bytes(dst, num_bytes, 1);
4439 return 0;
4440}
4441
Miao Xie66d8f3d2012-09-06 04:02:28 -06004442void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv, unsigned short type)
Yan, Zhengf0486c62010-05-16 10:46:25 -04004443{
4444 memset(rsv, 0, sizeof(*rsv));
4445 spin_lock_init(&rsv->lock);
Miao Xie66d8f3d2012-09-06 04:02:28 -06004446 rsv->type = type;
Yan, Zhengf0486c62010-05-16 10:46:25 -04004447}
4448
Miao Xie66d8f3d2012-09-06 04:02:28 -06004449struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root,
4450 unsigned short type)
Yan, Zhengf0486c62010-05-16 10:46:25 -04004451{
4452 struct btrfs_block_rsv *block_rsv;
4453 struct btrfs_fs_info *fs_info = root->fs_info;
Yan, Zhengf0486c62010-05-16 10:46:25 -04004454
4455 block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
4456 if (!block_rsv)
4457 return NULL;
4458
Miao Xie66d8f3d2012-09-06 04:02:28 -06004459 btrfs_init_block_rsv(block_rsv, type);
Yan, Zhengf0486c62010-05-16 10:46:25 -04004460 block_rsv->space_info = __find_space_info(fs_info,
4461 BTRFS_BLOCK_GROUP_METADATA);
Yan, Zhengf0486c62010-05-16 10:46:25 -04004462 return block_rsv;
4463}
4464
4465void btrfs_free_block_rsv(struct btrfs_root *root,
4466 struct btrfs_block_rsv *rsv)
4467{
Josef Bacik2aaa6652012-08-29 14:27:18 -04004468 if (!rsv)
4469 return;
Josef Bacikdabdb642011-08-08 12:50:18 -04004470 btrfs_block_rsv_release(root, rsv, (u64)-1);
4471 kfree(rsv);
Yan, Zhengf0486c62010-05-16 10:46:25 -04004472}
4473
Miao Xie08e007d2012-10-16 11:33:38 +00004474int btrfs_block_rsv_add(struct btrfs_root *root,
4475 struct btrfs_block_rsv *block_rsv, u64 num_bytes,
4476 enum btrfs_reserve_flush_enum flush)
Yan, Zhengf0486c62010-05-16 10:46:25 -04004477{
4478 int ret;
4479
4480 if (num_bytes == 0)
4481 return 0;
Josef Bacik8bb8ab22010-10-15 16:52:49 -04004482
Miao Xie61b520a2011-11-10 20:45:05 -05004483 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
Yan, Zhengf0486c62010-05-16 10:46:25 -04004484 if (!ret) {
4485 block_rsv_add_bytes(block_rsv, num_bytes, 1);
4486 return 0;
4487 }
4488
Yan, Zhengf0486c62010-05-16 10:46:25 -04004489 return ret;
4490}
4491
Josef Bacik4a92b1b2011-08-30 12:34:28 -04004492int btrfs_block_rsv_check(struct btrfs_root *root,
Josef Bacik36ba0222011-10-18 12:15:48 -04004493 struct btrfs_block_rsv *block_rsv, int min_factor)
Yan, Zhengf0486c62010-05-16 10:46:25 -04004494{
4495 u64 num_bytes = 0;
Yan, Zhengf0486c62010-05-16 10:46:25 -04004496 int ret = -ENOSPC;
4497
4498 if (!block_rsv)
4499 return 0;
4500
4501 spin_lock(&block_rsv->lock);
Josef Bacik36ba0222011-10-18 12:15:48 -04004502 num_bytes = div_factor(block_rsv->size, min_factor);
4503 if (block_rsv->reserved >= num_bytes)
Yan, Zhengf0486c62010-05-16 10:46:25 -04004504 ret = 0;
Yan, Zhengf0486c62010-05-16 10:46:25 -04004505 spin_unlock(&block_rsv->lock);
Yan, Zhengf0486c62010-05-16 10:46:25 -04004506
Josef Bacik36ba0222011-10-18 12:15:48 -04004507 return ret;
4508}
4509
Miao Xie08e007d2012-10-16 11:33:38 +00004510int btrfs_block_rsv_refill(struct btrfs_root *root,
4511 struct btrfs_block_rsv *block_rsv, u64 min_reserved,
4512 enum btrfs_reserve_flush_enum flush)
Josef Bacik36ba0222011-10-18 12:15:48 -04004513{
4514 u64 num_bytes = 0;
4515 int ret = -ENOSPC;
4516
4517 if (!block_rsv)
4518 return 0;
4519
4520 spin_lock(&block_rsv->lock);
4521 num_bytes = min_reserved;
Josef Bacik13553e52011-08-08 13:33:21 -04004522 if (block_rsv->reserved >= num_bytes)
Yan, Zhengf0486c62010-05-16 10:46:25 -04004523 ret = 0;
Josef Bacik13553e52011-08-08 13:33:21 -04004524 else
Yan, Zhengf0486c62010-05-16 10:46:25 -04004525 num_bytes -= block_rsv->reserved;
Yan, Zhengf0486c62010-05-16 10:46:25 -04004526 spin_unlock(&block_rsv->lock);
Josef Bacik13553e52011-08-08 13:33:21 -04004527
Yan, Zhengf0486c62010-05-16 10:46:25 -04004528 if (!ret)
4529 return 0;
4530
Miao Xieaa38a712011-11-18 17:43:00 +08004531 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
Josef Bacikdabdb642011-08-08 12:50:18 -04004532 if (!ret) {
4533 block_rsv_add_bytes(block_rsv, num_bytes, 0);
Yan, Zhengf0486c62010-05-16 10:46:25 -04004534 return 0;
4535 }
4536
Josef Bacik13553e52011-08-08 13:33:21 -04004537 return ret;
Yan, Zhengf0486c62010-05-16 10:46:25 -04004538}
4539
4540int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv,
4541 struct btrfs_block_rsv *dst_rsv,
4542 u64 num_bytes)
4543{
4544 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4545}
4546
4547void btrfs_block_rsv_release(struct btrfs_root *root,
4548 struct btrfs_block_rsv *block_rsv,
4549 u64 num_bytes)
4550{
4551 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
4552 if (global_rsv->full || global_rsv == block_rsv ||
4553 block_rsv->space_info != global_rsv->space_info)
4554 global_rsv = NULL;
Josef Bacik8c2a3ca2012-01-10 10:31:31 -05004555 block_rsv_release_bytes(root->fs_info, block_rsv, global_rsv,
4556 num_bytes);
Yan, Zhengf0486c62010-05-16 10:46:25 -04004557}
4558
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04004559/*
4560 * helper to calculate size of global block reservation.
4561 * the desired value is sum of space used by extent tree,
4562 * checksum tree and root tree
4563 */
4564static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info)
4565{
4566 struct btrfs_space_info *sinfo;
4567 u64 num_bytes;
4568 u64 meta_used;
4569 u64 data_used;
David Sterba6c417612011-04-13 15:41:04 +02004570 int csum_size = btrfs_super_csum_size(fs_info->super_copy);
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04004571
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04004572 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA);
4573 spin_lock(&sinfo->lock);
4574 data_used = sinfo->bytes_used;
4575 spin_unlock(&sinfo->lock);
4576
4577 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4578 spin_lock(&sinfo->lock);
Josef Bacik6d487552010-10-15 15:13:32 -04004579 if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA)
4580 data_used = 0;
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04004581 meta_used = sinfo->bytes_used;
4582 spin_unlock(&sinfo->lock);
4583
4584 num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) *
4585 csum_size * 2;
4586 num_bytes += div64_u64(data_used + meta_used, 50);
4587
4588 if (num_bytes * 3 > meta_used)
Chris Mason8e62c2d2012-04-12 13:46:48 -04004589 num_bytes = div64_u64(meta_used, 3);
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04004590
4591 return ALIGN(num_bytes, fs_info->extent_root->leafsize << 10);
4592}
4593
4594static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
4595{
4596 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
4597 struct btrfs_space_info *sinfo = block_rsv->space_info;
4598 u64 num_bytes;
4599
4600 num_bytes = calc_global_metadata_size(fs_info);
4601
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04004602 spin_lock(&sinfo->lock);
Stefan Behrens1f699d32012-04-27 12:41:46 -04004603 spin_lock(&block_rsv->lock);
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04004604
Josef Bacikfdf30d12013-03-26 15:31:45 -04004605 block_rsv->size = min_t(u64, num_bytes, 512 * 1024 * 1024);
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04004606
4607 num_bytes = sinfo->bytes_used + sinfo->bytes_pinned +
Josef Bacik6d487552010-10-15 15:13:32 -04004608 sinfo->bytes_reserved + sinfo->bytes_readonly +
4609 sinfo->bytes_may_use;
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04004610
4611 if (sinfo->total_bytes > num_bytes) {
4612 num_bytes = sinfo->total_bytes - num_bytes;
4613 block_rsv->reserved += num_bytes;
Josef Bacikfb25e912011-07-26 17:00:46 -04004614 sinfo->bytes_may_use += num_bytes;
Josef Bacik8c2a3ca2012-01-10 10:31:31 -05004615 trace_btrfs_space_reservation(fs_info, "space_info",
Liu Bo2bcc0322012-03-29 09:57:44 -04004616 sinfo->flags, num_bytes, 1);
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04004617 }
4618
4619 if (block_rsv->reserved >= block_rsv->size) {
4620 num_bytes = block_rsv->reserved - block_rsv->size;
Josef Bacikfb25e912011-07-26 17:00:46 -04004621 sinfo->bytes_may_use -= num_bytes;
Josef Bacik8c2a3ca2012-01-10 10:31:31 -05004622 trace_btrfs_space_reservation(fs_info, "space_info",
Liu Bo2bcc0322012-03-29 09:57:44 -04004623 sinfo->flags, num_bytes, 0);
Chris Mason36e39c42011-03-12 07:08:42 -05004624 sinfo->reservation_progress++;
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04004625 block_rsv->reserved = block_rsv->size;
4626 block_rsv->full = 1;
4627 }
David Sterba182608c2011-05-05 13:13:16 +02004628
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04004629 spin_unlock(&block_rsv->lock);
Stefan Behrens1f699d32012-04-27 12:41:46 -04004630 spin_unlock(&sinfo->lock);
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04004631}
4632
Yan, Zhengf0486c62010-05-16 10:46:25 -04004633static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
4634{
4635 struct btrfs_space_info *space_info;
4636
4637 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
4638 fs_info->chunk_block_rsv.space_info = space_info;
Yan, Zhengf0486c62010-05-16 10:46:25 -04004639
4640 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04004641 fs_info->global_block_rsv.space_info = space_info;
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04004642 fs_info->delalloc_block_rsv.space_info = space_info;
Yan, Zhengf0486c62010-05-16 10:46:25 -04004643 fs_info->trans_block_rsv.space_info = space_info;
4644 fs_info->empty_block_rsv.space_info = space_info;
Josef Bacik6d668dd2011-11-03 22:54:25 -04004645 fs_info->delayed_block_rsv.space_info = space_info;
Yan, Zhengf0486c62010-05-16 10:46:25 -04004646
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04004647 fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
4648 fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
4649 fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
4650 fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
Stefan Behrens3a6cad92013-05-16 14:48:19 +00004651 if (fs_info->quota_root)
4652 fs_info->quota_root->block_rsv = &fs_info->global_block_rsv;
Yan, Zhengf0486c62010-05-16 10:46:25 -04004653 fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04004654
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04004655 update_global_block_rsv(fs_info);
4656}
4657
4658static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
4659{
Josef Bacik8c2a3ca2012-01-10 10:31:31 -05004660 block_rsv_release_bytes(fs_info, &fs_info->global_block_rsv, NULL,
4661 (u64)-1);
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04004662 WARN_ON(fs_info->delalloc_block_rsv.size > 0);
4663 WARN_ON(fs_info->delalloc_block_rsv.reserved > 0);
4664 WARN_ON(fs_info->trans_block_rsv.size > 0);
4665 WARN_ON(fs_info->trans_block_rsv.reserved > 0);
4666 WARN_ON(fs_info->chunk_block_rsv.size > 0);
4667 WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
Josef Bacik6d668dd2011-11-03 22:54:25 -04004668 WARN_ON(fs_info->delayed_block_rsv.size > 0);
4669 WARN_ON(fs_info->delayed_block_rsv.reserved > 0);
Josef Bacikfcb80c22011-05-03 10:40:22 -04004670}
4671
Yan, Zhenga22285a2010-05-16 10:48:46 -04004672void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans,
4673 struct btrfs_root *root)
4674{
Josef Bacik0e721102012-06-26 16:13:18 -04004675 if (!trans->block_rsv)
4676 return;
4677
Yan, Zhenga22285a2010-05-16 10:48:46 -04004678 if (!trans->bytes_reserved)
4679 return;
4680
Chris Masone77266e2012-02-24 10:39:05 -05004681 trace_btrfs_space_reservation(root->fs_info, "transaction",
Liu Bo2bcc0322012-03-29 09:57:44 -04004682 trans->transid, trans->bytes_reserved, 0);
Josef Bacikb24e03d2011-10-14 14:40:17 -04004683 btrfs_block_rsv_release(root, trans->block_rsv, trans->bytes_reserved);
Yan, Zhenga22285a2010-05-16 10:48:46 -04004684 trans->bytes_reserved = 0;
4685}
4686
Jeff Mahoney79787ea2012-03-12 16:03:00 +01004687/* Can only return 0 or -ENOSPC */
Yan, Zhengd68fc572010-05-16 10:49:58 -04004688int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
4689 struct inode *inode)
4690{
4691 struct btrfs_root *root = BTRFS_I(inode)->root;
4692 struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
4693 struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv;
4694
4695 /*
Josef Bacikfcb80c22011-05-03 10:40:22 -04004696 * We need to hold space in order to delete our orphan item once we've
4697 * added it, so this takes the reservation so we can release it later
4698 * when we are truly done with the orphan item.
Yan, Zhengd68fc572010-05-16 10:49:58 -04004699 */
Chris Masonff5714c2011-05-28 07:00:39 -04004700 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
Josef Bacik8c2a3ca2012-01-10 10:31:31 -05004701 trace_btrfs_space_reservation(root->fs_info, "orphan",
4702 btrfs_ino(inode), num_bytes, 1);
Yan, Zhengd68fc572010-05-16 10:49:58 -04004703 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4704}
4705
4706void btrfs_orphan_release_metadata(struct inode *inode)
4707{
4708 struct btrfs_root *root = BTRFS_I(inode)->root;
Chris Masonff5714c2011-05-28 07:00:39 -04004709 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
Josef Bacik8c2a3ca2012-01-10 10:31:31 -05004710 trace_btrfs_space_reservation(root->fs_info, "orphan",
4711 btrfs_ino(inode), num_bytes, 0);
Yan, Zhengd68fc572010-05-16 10:49:58 -04004712 btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes);
4713}
4714
Miao Xied5c12072013-02-28 10:04:33 +00004715/*
4716 * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation
4717 * root: the root of the parent directory
4718 * rsv: block reservation
4719 * items: the number of items that we need do reservation
4720 * qgroup_reserved: used to return the reserved size in qgroup
4721 *
4722 * This function is used to reserve the space for snapshot/subvolume
4723 * creation and deletion. Those operations are different with the
4724 * common file/directory operations, they change two fs/file trees
4725 * and root tree, the number of items that the qgroup reserves is
4726 * different with the free space reservation. So we can not use
4727 * the space reseravtion mechanism in start_transaction().
4728 */
4729int btrfs_subvolume_reserve_metadata(struct btrfs_root *root,
4730 struct btrfs_block_rsv *rsv,
4731 int items,
4732 u64 *qgroup_reserved)
Yan, Zhenga22285a2010-05-16 10:48:46 -04004733{
Miao Xied5c12072013-02-28 10:04:33 +00004734 u64 num_bytes;
4735 int ret;
4736
4737 if (root->fs_info->quota_enabled) {
4738 /* One for parent inode, two for dir entries */
4739 num_bytes = 3 * root->leafsize;
4740 ret = btrfs_qgroup_reserve(root, num_bytes);
4741 if (ret)
4742 return ret;
4743 } else {
4744 num_bytes = 0;
4745 }
4746
4747 *qgroup_reserved = num_bytes;
4748
4749 num_bytes = btrfs_calc_trans_metadata_size(root, items);
4750 rsv->space_info = __find_space_info(root->fs_info,
4751 BTRFS_BLOCK_GROUP_METADATA);
4752 ret = btrfs_block_rsv_add(root, rsv, num_bytes,
4753 BTRFS_RESERVE_FLUSH_ALL);
4754 if (ret) {
4755 if (*qgroup_reserved)
4756 btrfs_qgroup_free(root, *qgroup_reserved);
4757 }
4758
4759 return ret;
4760}
4761
4762void btrfs_subvolume_release_metadata(struct btrfs_root *root,
4763 struct btrfs_block_rsv *rsv,
4764 u64 qgroup_reserved)
4765{
4766 btrfs_block_rsv_release(root, rsv, (u64)-1);
4767 if (qgroup_reserved)
4768 btrfs_qgroup_free(root, qgroup_reserved);
Yan, Zhenga22285a2010-05-16 10:48:46 -04004769}
4770
Josef Bacik7709cde2011-08-04 10:25:02 -04004771/**
4772 * drop_outstanding_extent - drop an outstanding extent
4773 * @inode: the inode we're dropping the extent for
4774 *
4775 * This is called when we are freeing up an outstanding extent, either called
4776 * after an error or after an extent is written. This will return the number of
4777 * reserved extents that need to be freed. This must be called with
4778 * BTRFS_I(inode)->lock held.
4779 */
Josef Bacik9e0baf62011-07-15 15:16:44 +00004780static unsigned drop_outstanding_extent(struct inode *inode)
4781{
Josef Bacik7fd2ae22011-11-08 15:47:34 -05004782 unsigned drop_inode_space = 0;
Josef Bacik9e0baf62011-07-15 15:16:44 +00004783 unsigned dropped_extents = 0;
4784
Josef Bacik9e0baf62011-07-15 15:16:44 +00004785 BUG_ON(!BTRFS_I(inode)->outstanding_extents);
4786 BTRFS_I(inode)->outstanding_extents--;
4787
Josef Bacik7fd2ae22011-11-08 15:47:34 -05004788 if (BTRFS_I(inode)->outstanding_extents == 0 &&
Josef Bacik72ac3c02012-05-23 14:13:11 -04004789 test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
4790 &BTRFS_I(inode)->runtime_flags))
Josef Bacik7fd2ae22011-11-08 15:47:34 -05004791 drop_inode_space = 1;
Josef Bacik7fd2ae22011-11-08 15:47:34 -05004792
Josef Bacik9e0baf62011-07-15 15:16:44 +00004793 /*
4794 * If we have more or the same amount of outsanding extents than we have
4795 * reserved then we need to leave the reserved extents count alone.
4796 */
4797 if (BTRFS_I(inode)->outstanding_extents >=
4798 BTRFS_I(inode)->reserved_extents)
Josef Bacik7fd2ae22011-11-08 15:47:34 -05004799 return drop_inode_space;
Josef Bacik9e0baf62011-07-15 15:16:44 +00004800
4801 dropped_extents = BTRFS_I(inode)->reserved_extents -
4802 BTRFS_I(inode)->outstanding_extents;
4803 BTRFS_I(inode)->reserved_extents -= dropped_extents;
Josef Bacik7fd2ae22011-11-08 15:47:34 -05004804 return dropped_extents + drop_inode_space;
Josef Bacik9e0baf62011-07-15 15:16:44 +00004805}
4806
Josef Bacik7709cde2011-08-04 10:25:02 -04004807/**
4808 * calc_csum_metadata_size - return the amount of metada space that must be
4809 * reserved/free'd for the given bytes.
4810 * @inode: the inode we're manipulating
4811 * @num_bytes: the number of bytes in question
4812 * @reserve: 1 if we are reserving space, 0 if we are freeing space
4813 *
4814 * This adjusts the number of csum_bytes in the inode and then returns the
4815 * correct amount of metadata that must either be reserved or freed. We
4816 * calculate how many checksums we can fit into one leaf and then divide the
4817 * number of bytes that will need to be checksumed by this value to figure out
4818 * how many checksums will be required. If we are adding bytes then the number
4819 * may go up and we will return the number of additional bytes that must be
4820 * reserved. If it is going down we will return the number of bytes that must
4821 * be freed.
4822 *
4823 * This must be called with BTRFS_I(inode)->lock held.
4824 */
4825static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes,
4826 int reserve)
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04004827{
Josef Bacik7709cde2011-08-04 10:25:02 -04004828 struct btrfs_root *root = BTRFS_I(inode)->root;
4829 u64 csum_size;
4830 int num_csums_per_leaf;
4831 int num_csums;
4832 int old_csums;
4833
4834 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM &&
4835 BTRFS_I(inode)->csum_bytes == 0)
4836 return 0;
4837
4838 old_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize);
4839 if (reserve)
4840 BTRFS_I(inode)->csum_bytes += num_bytes;
4841 else
4842 BTRFS_I(inode)->csum_bytes -= num_bytes;
4843 csum_size = BTRFS_LEAF_DATA_SIZE(root) - sizeof(struct btrfs_item);
4844 num_csums_per_leaf = (int)div64_u64(csum_size,
4845 sizeof(struct btrfs_csum_item) +
4846 sizeof(struct btrfs_disk_key));
4847 num_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize);
4848 num_csums = num_csums + num_csums_per_leaf - 1;
4849 num_csums = num_csums / num_csums_per_leaf;
4850
4851 old_csums = old_csums + num_csums_per_leaf - 1;
4852 old_csums = old_csums / num_csums_per_leaf;
4853
4854 /* No change, no need to reserve more */
4855 if (old_csums == num_csums)
4856 return 0;
4857
4858 if (reserve)
4859 return btrfs_calc_trans_metadata_size(root,
4860 num_csums - old_csums);
4861
4862 return btrfs_calc_trans_metadata_size(root, old_csums - num_csums);
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04004863}
4864
4865int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes)
4866{
4867 struct btrfs_root *root = BTRFS_I(inode)->root;
4868 struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv;
Josef Bacik9e0baf62011-07-15 15:16:44 +00004869 u64 to_reserve = 0;
Josef Bacik660d3f62011-12-09 11:18:51 -05004870 u64 csum_bytes;
Josef Bacik9e0baf62011-07-15 15:16:44 +00004871 unsigned nr_extents = 0;
Josef Bacik660d3f62011-12-09 11:18:51 -05004872 int extra_reserve = 0;
Miao Xie08e007d2012-10-16 11:33:38 +00004873 enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_FLUSH_ALL;
Jan Schmidteb6b88d2013-01-27 23:26:00 -07004874 int ret = 0;
Josef Bacikc64c2bd2012-12-14 13:48:14 -05004875 bool delalloc_lock = true;
Wang Shilong88e081bf2013-03-01 11:36:01 +00004876 u64 to_free = 0;
4877 unsigned dropped;
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04004878
Josef Bacikc64c2bd2012-12-14 13:48:14 -05004879 /* If we are a free space inode we need to not flush since we will be in
4880 * the middle of a transaction commit. We also don't need the delalloc
4881 * mutex since we won't race with anybody. We need this mostly to make
4882 * lockdep shut its filthy mouth.
4883 */
4884 if (btrfs_is_free_space_inode(inode)) {
Miao Xie08e007d2012-10-16 11:33:38 +00004885 flush = BTRFS_RESERVE_NO_FLUSH;
Josef Bacikc64c2bd2012-12-14 13:48:14 -05004886 delalloc_lock = false;
4887 }
Josef Bacikc09544e2011-08-30 10:19:10 -04004888
Miao Xie08e007d2012-10-16 11:33:38 +00004889 if (flush != BTRFS_RESERVE_NO_FLUSH &&
4890 btrfs_transaction_in_commit(root->fs_info))
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04004891 schedule_timeout(1);
4892
Josef Bacikc64c2bd2012-12-14 13:48:14 -05004893 if (delalloc_lock)
4894 mutex_lock(&BTRFS_I(inode)->delalloc_mutex);
4895
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04004896 num_bytes = ALIGN(num_bytes, root->sectorsize);
Josef Bacik8bb8ab22010-10-15 16:52:49 -04004897
Josef Bacik9e0baf62011-07-15 15:16:44 +00004898 spin_lock(&BTRFS_I(inode)->lock);
4899 BTRFS_I(inode)->outstanding_extents++;
Josef Bacik57a45ced2011-01-25 16:30:38 -05004900
Josef Bacik9e0baf62011-07-15 15:16:44 +00004901 if (BTRFS_I(inode)->outstanding_extents >
Josef Bacik660d3f62011-12-09 11:18:51 -05004902 BTRFS_I(inode)->reserved_extents)
Josef Bacik9e0baf62011-07-15 15:16:44 +00004903 nr_extents = BTRFS_I(inode)->outstanding_extents -
4904 BTRFS_I(inode)->reserved_extents;
Josef Bacik7fd2ae22011-11-08 15:47:34 -05004905
4906 /*
4907 * Add an item to reserve for updating the inode when we complete the
4908 * delalloc io.
4909 */
Josef Bacik72ac3c02012-05-23 14:13:11 -04004910 if (!test_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
4911 &BTRFS_I(inode)->runtime_flags)) {
Josef Bacik7fd2ae22011-11-08 15:47:34 -05004912 nr_extents++;
Josef Bacik660d3f62011-12-09 11:18:51 -05004913 extra_reserve = 1;
Josef Bacik7fd2ae22011-11-08 15:47:34 -05004914 }
4915
4916 to_reserve = btrfs_calc_trans_metadata_size(root, nr_extents);
Josef Bacik7709cde2011-08-04 10:25:02 -04004917 to_reserve += calc_csum_metadata_size(inode, num_bytes, 1);
Josef Bacik660d3f62011-12-09 11:18:51 -05004918 csum_bytes = BTRFS_I(inode)->csum_bytes;
Josef Bacik9e0baf62011-07-15 15:16:44 +00004919 spin_unlock(&BTRFS_I(inode)->lock);
Josef Bacik57a45ced2011-01-25 16:30:38 -05004920
Wang Shilong88e081bf2013-03-01 11:36:01 +00004921 if (root->fs_info->quota_enabled) {
Arne Jansenc5567232011-09-14 15:44:05 +02004922 ret = btrfs_qgroup_reserve(root, num_bytes +
4923 nr_extents * root->leafsize);
Wang Shilong88e081bf2013-03-01 11:36:01 +00004924 if (ret)
4925 goto out_fail;
Wang Shilonga9870c02013-03-01 11:33:01 +00004926 }
Arne Jansenc5567232011-09-14 15:44:05 +02004927
Wang Shilong88e081bf2013-03-01 11:36:01 +00004928 ret = reserve_metadata_bytes(root, block_rsv, to_reserve, flush);
4929 if (unlikely(ret)) {
4930 if (root->fs_info->quota_enabled)
Miao Xie4b5829a2012-12-05 10:53:25 +00004931 btrfs_qgroup_free(root, num_bytes +
4932 nr_extents * root->leafsize);
Wang Shilong88e081bf2013-03-01 11:36:01 +00004933 goto out_fail;
Josef Bacik9e0baf62011-07-15 15:16:44 +00004934 }
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04004935
Josef Bacik660d3f62011-12-09 11:18:51 -05004936 spin_lock(&BTRFS_I(inode)->lock);
4937 if (extra_reserve) {
Josef Bacik72ac3c02012-05-23 14:13:11 -04004938 set_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
4939 &BTRFS_I(inode)->runtime_flags);
Josef Bacik660d3f62011-12-09 11:18:51 -05004940 nr_extents--;
4941 }
4942 BTRFS_I(inode)->reserved_extents += nr_extents;
4943 spin_unlock(&BTRFS_I(inode)->lock);
Josef Bacikc64c2bd2012-12-14 13:48:14 -05004944
4945 if (delalloc_lock)
4946 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
Josef Bacik660d3f62011-12-09 11:18:51 -05004947
Josef Bacik8c2a3ca2012-01-10 10:31:31 -05004948 if (to_reserve)
4949 trace_btrfs_space_reservation(root->fs_info,"delalloc",
4950 btrfs_ino(inode), to_reserve, 1);
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04004951 block_rsv_add_bytes(block_rsv, to_reserve, 1);
4952
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04004953 return 0;
Wang Shilong88e081bf2013-03-01 11:36:01 +00004954
4955out_fail:
4956 spin_lock(&BTRFS_I(inode)->lock);
4957 dropped = drop_outstanding_extent(inode);
4958 /*
4959 * If the inodes csum_bytes is the same as the original
4960 * csum_bytes then we know we haven't raced with any free()ers
4961 * so we can just reduce our inodes csum bytes and carry on.
Wang Shilong88e081bf2013-03-01 11:36:01 +00004962 */
Josef Bacikf4881bc2013-03-25 16:03:35 -04004963 if (BTRFS_I(inode)->csum_bytes == csum_bytes) {
Wang Shilong88e081bf2013-03-01 11:36:01 +00004964 calc_csum_metadata_size(inode, num_bytes, 0);
Josef Bacikf4881bc2013-03-25 16:03:35 -04004965 } else {
4966 u64 orig_csum_bytes = BTRFS_I(inode)->csum_bytes;
4967 u64 bytes;
4968
4969 /*
4970 * This is tricky, but first we need to figure out how much we
4971 * free'd from any free-ers that occured during this
4972 * reservation, so we reset ->csum_bytes to the csum_bytes
4973 * before we dropped our lock, and then call the free for the
4974 * number of bytes that were freed while we were trying our
4975 * reservation.
4976 */
4977 bytes = csum_bytes - BTRFS_I(inode)->csum_bytes;
4978 BTRFS_I(inode)->csum_bytes = csum_bytes;
4979 to_free = calc_csum_metadata_size(inode, bytes, 0);
4980
4981
4982 /*
4983 * Now we need to see how much we would have freed had we not
4984 * been making this reservation and our ->csum_bytes were not
4985 * artificially inflated.
4986 */
4987 BTRFS_I(inode)->csum_bytes = csum_bytes - num_bytes;
4988 bytes = csum_bytes - orig_csum_bytes;
4989 bytes = calc_csum_metadata_size(inode, bytes, 0);
4990
4991 /*
4992 * Now reset ->csum_bytes to what it should be. If bytes is
4993 * more than to_free then we would have free'd more space had we
4994 * not had an artificially high ->csum_bytes, so we need to free
4995 * the remainder. If bytes is the same or less then we don't
4996 * need to do anything, the other free-ers did the correct
4997 * thing.
4998 */
4999 BTRFS_I(inode)->csum_bytes = orig_csum_bytes - num_bytes;
5000 if (bytes > to_free)
5001 to_free = bytes - to_free;
5002 else
5003 to_free = 0;
5004 }
Wang Shilong88e081bf2013-03-01 11:36:01 +00005005 spin_unlock(&BTRFS_I(inode)->lock);
5006 if (dropped)
5007 to_free += btrfs_calc_trans_metadata_size(root, dropped);
5008
5009 if (to_free) {
5010 btrfs_block_rsv_release(root, block_rsv, to_free);
5011 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5012 btrfs_ino(inode), to_free, 0);
5013 }
5014 if (delalloc_lock)
5015 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
5016 return ret;
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04005017}
5018
Josef Bacik7709cde2011-08-04 10:25:02 -04005019/**
5020 * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
5021 * @inode: the inode to release the reservation for
5022 * @num_bytes: the number of bytes we're releasing
5023 *
5024 * This will release the metadata reservation for an inode. This can be called
5025 * once we complete IO for a given set of bytes to release their metadata
5026 * reservations.
5027 */
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04005028void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes)
5029{
5030 struct btrfs_root *root = BTRFS_I(inode)->root;
Josef Bacik9e0baf62011-07-15 15:16:44 +00005031 u64 to_free = 0;
5032 unsigned dropped;
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04005033
5034 num_bytes = ALIGN(num_bytes, root->sectorsize);
Josef Bacik7709cde2011-08-04 10:25:02 -04005035 spin_lock(&BTRFS_I(inode)->lock);
Josef Bacik9e0baf62011-07-15 15:16:44 +00005036 dropped = drop_outstanding_extent(inode);
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04005037
Miao Xie09348562013-02-07 10:12:07 +00005038 if (num_bytes)
5039 to_free = calc_csum_metadata_size(inode, num_bytes, 0);
Josef Bacik7709cde2011-08-04 10:25:02 -04005040 spin_unlock(&BTRFS_I(inode)->lock);
Josef Bacik9e0baf62011-07-15 15:16:44 +00005041 if (dropped > 0)
5042 to_free += btrfs_calc_trans_metadata_size(root, dropped);
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04005043
Josef Bacik8c2a3ca2012-01-10 10:31:31 -05005044 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5045 btrfs_ino(inode), to_free, 0);
Arne Jansenc5567232011-09-14 15:44:05 +02005046 if (root->fs_info->quota_enabled) {
5047 btrfs_qgroup_free(root, num_bytes +
5048 dropped * root->leafsize);
5049 }
5050
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04005051 btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv,
5052 to_free);
5053}
5054
Josef Bacik7709cde2011-08-04 10:25:02 -04005055/**
5056 * btrfs_delalloc_reserve_space - reserve data and metadata space for delalloc
5057 * @inode: inode we're writing to
5058 * @num_bytes: the number of bytes we want to allocate
5059 *
5060 * This will do the following things
5061 *
5062 * o reserve space in the data space info for num_bytes
5063 * o reserve space in the metadata space info based on number of outstanding
5064 * extents and how much csums will be needed
5065 * o add to the inodes ->delalloc_bytes
5066 * o add it to the fs_info's delalloc inodes list.
5067 *
5068 * This will return 0 for success and -ENOSPC if there is no space left.
5069 */
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04005070int btrfs_delalloc_reserve_space(struct inode *inode, u64 num_bytes)
5071{
5072 int ret;
5073
5074 ret = btrfs_check_data_free_space(inode, num_bytes);
5075 if (ret)
5076 return ret;
5077
5078 ret = btrfs_delalloc_reserve_metadata(inode, num_bytes);
5079 if (ret) {
5080 btrfs_free_reserved_data_space(inode, num_bytes);
5081 return ret;
5082 }
5083
5084 return 0;
5085}
5086
Josef Bacik7709cde2011-08-04 10:25:02 -04005087/**
5088 * btrfs_delalloc_release_space - release data and metadata space for delalloc
5089 * @inode: inode we're releasing space for
5090 * @num_bytes: the number of bytes we want to free up
5091 *
5092 * This must be matched with a call to btrfs_delalloc_reserve_space. This is
5093 * called in the case that we don't need the metadata AND data reservations
5094 * anymore. So if there is an error or we insert an inline extent.
5095 *
5096 * This function will release the metadata space that was not used and will
5097 * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
5098 * list if there are no delalloc bytes left.
5099 */
Yan, Zheng0ca1f7c2010-05-16 10:48:47 -04005100void btrfs_delalloc_release_space(struct inode *inode, u64 num_bytes)
5101{
5102 btrfs_delalloc_release_metadata(inode, num_bytes);
5103 btrfs_free_reserved_data_space(inode, num_bytes);
5104}
5105
Liu Boc53d6132012-12-27 09:01:19 +00005106static int update_block_group(struct btrfs_root *root,
Yan, Zhengf0486c62010-05-16 10:46:25 -04005107 u64 bytenr, u64 num_bytes, int alloc)
Chris Mason9078a3e2007-04-26 16:46:15 -04005108{
Josef Bacik0af3d002010-06-21 14:48:16 -04005109 struct btrfs_block_group_cache *cache = NULL;
Chris Mason9078a3e2007-04-26 16:46:15 -04005110 struct btrfs_fs_info *info = root->fs_info;
Chris Masondb945352007-10-15 16:15:53 -04005111 u64 total = num_bytes;
Chris Mason9078a3e2007-04-26 16:46:15 -04005112 u64 old_val;
Chris Masondb945352007-10-15 16:15:53 -04005113 u64 byte_in_group;
Josef Bacik0af3d002010-06-21 14:48:16 -04005114 int factor;
Chris Mason3e1ad542007-05-07 20:03:49 -04005115
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005116 /* block accounting for super block */
Miao Xieeb73c1b2013-05-15 07:48:22 +00005117 spin_lock(&info->delalloc_root_lock);
David Sterba6c417612011-04-13 15:41:04 +02005118 old_val = btrfs_super_bytes_used(info->super_copy);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005119 if (alloc)
5120 old_val += num_bytes;
5121 else
5122 old_val -= num_bytes;
David Sterba6c417612011-04-13 15:41:04 +02005123 btrfs_set_super_bytes_used(info->super_copy, old_val);
Miao Xieeb73c1b2013-05-15 07:48:22 +00005124 spin_unlock(&info->delalloc_root_lock);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005125
Chris Masond3977122009-01-05 21:25:51 -05005126 while (total) {
Chris Masondb945352007-10-15 16:15:53 -04005127 cache = btrfs_lookup_block_group(info, bytenr);
Josef Bacikf3465ca2008-11-12 14:19:50 -05005128 if (!cache)
Jeff Mahoney79787ea2012-03-12 16:03:00 +01005129 return -ENOENT;
Yan, Zhengb742bb82010-05-16 10:46:24 -04005130 if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
5131 BTRFS_BLOCK_GROUP_RAID1 |
5132 BTRFS_BLOCK_GROUP_RAID10))
5133 factor = 2;
5134 else
5135 factor = 1;
Josef Bacik9d66e232010-08-25 16:54:15 -04005136 /*
5137 * If this block group has free space cache written out, we
5138 * need to make sure to load it if we are removing space. This
5139 * is because we need the unpinning stage to actually add the
5140 * space back to the block group, otherwise we will leak space.
5141 */
5142 if (!alloc && cache->cached == BTRFS_CACHE_NO)
Liu Bof6373bf2012-12-27 09:01:18 +00005143 cache_block_group(cache, 1);
Josef Bacik0af3d002010-06-21 14:48:16 -04005144
Chris Masondb945352007-10-15 16:15:53 -04005145 byte_in_group = bytenr - cache->key.objectid;
5146 WARN_ON(byte_in_group > cache->key.offset);
Chris Mason9078a3e2007-04-26 16:46:15 -04005147
Josef Bacik25179202008-10-29 14:49:05 -04005148 spin_lock(&cache->space_info->lock);
Chris Masonc286ac42008-07-22 23:06:41 -04005149 spin_lock(&cache->lock);
Josef Bacik0af3d002010-06-21 14:48:16 -04005150
Josef Bacik73bc1872011-10-03 14:07:49 -04005151 if (btrfs_test_opt(root, SPACE_CACHE) &&
Josef Bacik0af3d002010-06-21 14:48:16 -04005152 cache->disk_cache_state < BTRFS_DC_CLEAR)
5153 cache->disk_cache_state = BTRFS_DC_CLEAR;
5154
Josef Bacik0f9dd462008-09-23 13:14:11 -04005155 cache->dirty = 1;
Chris Mason9078a3e2007-04-26 16:46:15 -04005156 old_val = btrfs_block_group_used(&cache->item);
Chris Masondb945352007-10-15 16:15:53 -04005157 num_bytes = min(total, cache->key.offset - byte_in_group);
Chris Masoncd1bc462007-04-27 10:08:34 -04005158 if (alloc) {
Chris Masondb945352007-10-15 16:15:53 -04005159 old_val += num_bytes;
Yan Zheng11833d62009-09-11 16:11:19 -04005160 btrfs_set_block_group_used(&cache->item, old_val);
5161 cache->reserved -= num_bytes;
Yan Zheng11833d62009-09-11 16:11:19 -04005162 cache->space_info->bytes_reserved -= num_bytes;
Yan, Zhengb742bb82010-05-16 10:46:24 -04005163 cache->space_info->bytes_used += num_bytes;
5164 cache->space_info->disk_used += num_bytes * factor;
Chris Masonc286ac42008-07-22 23:06:41 -04005165 spin_unlock(&cache->lock);
Josef Bacik25179202008-10-29 14:49:05 -04005166 spin_unlock(&cache->space_info->lock);
Chris Masoncd1bc462007-04-27 10:08:34 -04005167 } else {
Chris Masondb945352007-10-15 16:15:53 -04005168 old_val -= num_bytes;
Chris Masonc286ac42008-07-22 23:06:41 -04005169 btrfs_set_block_group_used(&cache->item, old_val);
Yan, Zhengf0486c62010-05-16 10:46:25 -04005170 cache->pinned += num_bytes;
5171 cache->space_info->bytes_pinned += num_bytes;
Chris Mason9078a3e2007-04-26 16:46:15 -04005172 cache->space_info->bytes_used -= num_bytes;
Yan, Zhengb742bb82010-05-16 10:46:24 -04005173 cache->space_info->disk_used -= num_bytes * factor;
Chris Masonc286ac42008-07-22 23:06:41 -04005174 spin_unlock(&cache->lock);
Josef Bacik25179202008-10-29 14:49:05 -04005175 spin_unlock(&cache->space_info->lock);
Liu Hui1f3c79a2009-01-05 15:57:51 -05005176
Yan, Zhengf0486c62010-05-16 10:46:25 -04005177 set_extent_dirty(info->pinned_extents,
5178 bytenr, bytenr + num_bytes - 1,
5179 GFP_NOFS | __GFP_NOFAIL);
Chris Masoncd1bc462007-04-27 10:08:34 -04005180 }
Chris Masonfa9c0d792009-04-03 09:47:43 -04005181 btrfs_put_block_group(cache);
Chris Masondb945352007-10-15 16:15:53 -04005182 total -= num_bytes;
5183 bytenr += num_bytes;
Chris Mason9078a3e2007-04-26 16:46:15 -04005184 }
5185 return 0;
5186}
Chris Mason6324fbf2008-03-24 15:01:59 -04005187
Chris Masona061fc82008-05-07 11:43:44 -04005188static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
5189{
Josef Bacik0f9dd462008-09-23 13:14:11 -04005190 struct btrfs_block_group_cache *cache;
Yan Zhengd2fb3432008-12-11 16:30:39 -05005191 u64 bytenr;
Josef Bacik0f9dd462008-09-23 13:14:11 -04005192
Liu Boa1897fd2012-12-27 09:01:23 +00005193 spin_lock(&root->fs_info->block_group_cache_lock);
5194 bytenr = root->fs_info->first_logical_byte;
5195 spin_unlock(&root->fs_info->block_group_cache_lock);
5196
5197 if (bytenr < (u64)-1)
5198 return bytenr;
5199
Josef Bacik0f9dd462008-09-23 13:14:11 -04005200 cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
5201 if (!cache)
Chris Masona061fc82008-05-07 11:43:44 -04005202 return 0;
Josef Bacik0f9dd462008-09-23 13:14:11 -04005203
Yan Zhengd2fb3432008-12-11 16:30:39 -05005204 bytenr = cache->key.objectid;
Chris Masonfa9c0d792009-04-03 09:47:43 -04005205 btrfs_put_block_group(cache);
Yan Zhengd2fb3432008-12-11 16:30:39 -05005206
5207 return bytenr;
Chris Masona061fc82008-05-07 11:43:44 -04005208}
5209
Yan, Zhengf0486c62010-05-16 10:46:25 -04005210static int pin_down_extent(struct btrfs_root *root,
5211 struct btrfs_block_group_cache *cache,
5212 u64 bytenr, u64 num_bytes, int reserved)
Yan324ae4d2007-11-16 14:57:08 -05005213{
Yan Zheng11833d62009-09-11 16:11:19 -04005214 spin_lock(&cache->space_info->lock);
5215 spin_lock(&cache->lock);
5216 cache->pinned += num_bytes;
5217 cache->space_info->bytes_pinned += num_bytes;
5218 if (reserved) {
5219 cache->reserved -= num_bytes;
5220 cache->space_info->bytes_reserved -= num_bytes;
Yan324ae4d2007-11-16 14:57:08 -05005221 }
Yan Zheng11833d62009-09-11 16:11:19 -04005222 spin_unlock(&cache->lock);
5223 spin_unlock(&cache->space_info->lock);
5224
Yan, Zhengf0486c62010-05-16 10:46:25 -04005225 set_extent_dirty(root->fs_info->pinned_extents, bytenr,
5226 bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
Yan324ae4d2007-11-16 14:57:08 -05005227 return 0;
5228}
Chris Mason9078a3e2007-04-26 16:46:15 -04005229
Yan, Zhengf0486c62010-05-16 10:46:25 -04005230/*
5231 * this function must be called within transaction
5232 */
5233int btrfs_pin_extent(struct btrfs_root *root,
5234 u64 bytenr, u64 num_bytes, int reserved)
Zheng Yane8569812008-09-26 10:05:48 -04005235{
Yan, Zhengf0486c62010-05-16 10:46:25 -04005236 struct btrfs_block_group_cache *cache;
5237
5238 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01005239 BUG_ON(!cache); /* Logic error */
Yan, Zhengf0486c62010-05-16 10:46:25 -04005240
5241 pin_down_extent(root, cache, bytenr, num_bytes, reserved);
5242
5243 btrfs_put_block_group(cache);
Yan Zheng11833d62009-09-11 16:11:19 -04005244 return 0;
5245}
Zheng Yane8569812008-09-26 10:05:48 -04005246
Yan, Zhengf0486c62010-05-16 10:46:25 -04005247/*
Chris Masone688b7252011-10-31 20:52:39 -04005248 * this function must be called within transaction
Yan, Zhengf0486c62010-05-16 10:46:25 -04005249 */
Liu Bodcfac412012-12-27 09:01:20 +00005250int btrfs_pin_extent_for_log_replay(struct btrfs_root *root,
Chris Masone688b7252011-10-31 20:52:39 -04005251 u64 bytenr, u64 num_bytes)
Yan, Zhengf0486c62010-05-16 10:46:25 -04005252{
Chris Masone688b7252011-10-31 20:52:39 -04005253 struct btrfs_block_group_cache *cache;
Josef Bacikb50c6e22013-04-25 15:55:30 -04005254 int ret;
Chris Masone688b7252011-10-31 20:52:39 -04005255
5256 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
Josef Bacikb50c6e22013-04-25 15:55:30 -04005257 if (!cache)
5258 return -EINVAL;
Chris Masone688b7252011-10-31 20:52:39 -04005259
5260 /*
5261 * pull in the free space cache (if any) so that our pin
5262 * removes the free space from the cache. We have load_only set
5263 * to one because the slow code to read in the free extents does check
5264 * the pinned extents.
5265 */
Liu Bof6373bf2012-12-27 09:01:18 +00005266 cache_block_group(cache, 1);
Chris Masone688b7252011-10-31 20:52:39 -04005267
5268 pin_down_extent(root, cache, bytenr, num_bytes, 0);
5269
5270 /* remove us from the free space cache (if we're there at all) */
Josef Bacikb50c6e22013-04-25 15:55:30 -04005271 ret = btrfs_remove_free_space(cache, bytenr, num_bytes);
Chris Masone688b7252011-10-31 20:52:39 -04005272 btrfs_put_block_group(cache);
Josef Bacikb50c6e22013-04-25 15:55:30 -04005273 return ret;
Chris Masone688b7252011-10-31 20:52:39 -04005274}
5275
Josef Bacik8c2a1a32013-06-06 13:19:32 -04005276static int __exclude_logged_extent(struct btrfs_root *root, u64 start, u64 num_bytes)
5277{
5278 int ret;
5279 struct btrfs_block_group_cache *block_group;
5280 struct btrfs_caching_control *caching_ctl;
5281
5282 block_group = btrfs_lookup_block_group(root->fs_info, start);
5283 if (!block_group)
5284 return -EINVAL;
5285
5286 cache_block_group(block_group, 0);
5287 caching_ctl = get_caching_control(block_group);
5288
5289 if (!caching_ctl) {
5290 /* Logic error */
5291 BUG_ON(!block_group_cache_done(block_group));
5292 ret = btrfs_remove_free_space(block_group, start, num_bytes);
5293 } else {
5294 mutex_lock(&caching_ctl->mutex);
5295
5296 if (start >= caching_ctl->progress) {
5297 ret = add_excluded_extent(root, start, num_bytes);
5298 } else if (start + num_bytes <= caching_ctl->progress) {
5299 ret = btrfs_remove_free_space(block_group,
5300 start, num_bytes);
5301 } else {
5302 num_bytes = caching_ctl->progress - start;
5303 ret = btrfs_remove_free_space(block_group,
5304 start, num_bytes);
5305 if (ret)
5306 goto out_lock;
5307
5308 num_bytes = (start + num_bytes) -
5309 caching_ctl->progress;
5310 start = caching_ctl->progress;
5311 ret = add_excluded_extent(root, start, num_bytes);
5312 }
5313out_lock:
5314 mutex_unlock(&caching_ctl->mutex);
5315 put_caching_control(caching_ctl);
5316 }
5317 btrfs_put_block_group(block_group);
5318 return ret;
5319}
5320
5321int btrfs_exclude_logged_extents(struct btrfs_root *log,
5322 struct extent_buffer *eb)
5323{
5324 struct btrfs_file_extent_item *item;
5325 struct btrfs_key key;
5326 int found_type;
5327 int i;
5328
5329 if (!btrfs_fs_incompat(log->fs_info, MIXED_GROUPS))
5330 return 0;
5331
5332 for (i = 0; i < btrfs_header_nritems(eb); i++) {
5333 btrfs_item_key_to_cpu(eb, &key, i);
5334 if (key.type != BTRFS_EXTENT_DATA_KEY)
5335 continue;
5336 item = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
5337 found_type = btrfs_file_extent_type(eb, item);
5338 if (found_type == BTRFS_FILE_EXTENT_INLINE)
5339 continue;
5340 if (btrfs_file_extent_disk_bytenr(eb, item) == 0)
5341 continue;
5342 key.objectid = btrfs_file_extent_disk_bytenr(eb, item);
5343 key.offset = btrfs_file_extent_disk_num_bytes(eb, item);
5344 __exclude_logged_extent(log, key.objectid, key.offset);
5345 }
5346
5347 return 0;
5348}
5349
Josef Bacikfb25e912011-07-26 17:00:46 -04005350/**
5351 * btrfs_update_reserved_bytes - update the block_group and space info counters
5352 * @cache: The cache we are manipulating
5353 * @num_bytes: The number of bytes in question
5354 * @reserve: One of the reservation enums
5355 *
5356 * This is called by the allocator when it reserves space, or by somebody who is
5357 * freeing space that was never actually used on disk. For example if you
5358 * reserve some space for a new leaf in transaction A and before transaction A
5359 * commits you free that leaf, you call this with reserve set to 0 in order to
5360 * clear the reservation.
5361 *
5362 * Metadata reservations should be called with RESERVE_ALLOC so we do the proper
5363 * ENOSPC accounting. For data we handle the reservation through clearing the
5364 * delalloc bits in the io_tree. We have to do this since we could end up
5365 * allocating less disk space for the amount of data we have reserved in the
5366 * case of compression.
5367 *
5368 * If this is a reservation and the block group has become read only we cannot
5369 * make the reservation and return -EAGAIN, otherwise this function always
5370 * succeeds.
Yan, Zhengf0486c62010-05-16 10:46:25 -04005371 */
Josef Bacikfb25e912011-07-26 17:00:46 -04005372static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
5373 u64 num_bytes, int reserve)
Yan, Zhengf0486c62010-05-16 10:46:25 -04005374{
Josef Bacikfb25e912011-07-26 17:00:46 -04005375 struct btrfs_space_info *space_info = cache->space_info;
Yan, Zhengf0486c62010-05-16 10:46:25 -04005376 int ret = 0;
Jeff Mahoney79787ea2012-03-12 16:03:00 +01005377
Josef Bacikfb25e912011-07-26 17:00:46 -04005378 spin_lock(&space_info->lock);
5379 spin_lock(&cache->lock);
5380 if (reserve != RESERVE_FREE) {
Yan, Zhengf0486c62010-05-16 10:46:25 -04005381 if (cache->ro) {
5382 ret = -EAGAIN;
5383 } else {
Josef Bacikfb25e912011-07-26 17:00:46 -04005384 cache->reserved += num_bytes;
5385 space_info->bytes_reserved += num_bytes;
5386 if (reserve == RESERVE_ALLOC) {
Josef Bacik8c2a3ca2012-01-10 10:31:31 -05005387 trace_btrfs_space_reservation(cache->fs_info,
Liu Bo2bcc0322012-03-29 09:57:44 -04005388 "space_info", space_info->flags,
5389 num_bytes, 0);
Josef Bacikfb25e912011-07-26 17:00:46 -04005390 space_info->bytes_may_use -= num_bytes;
5391 }
Yan, Zhengf0486c62010-05-16 10:46:25 -04005392 }
Josef Bacikfb25e912011-07-26 17:00:46 -04005393 } else {
5394 if (cache->ro)
5395 space_info->bytes_readonly += num_bytes;
5396 cache->reserved -= num_bytes;
5397 space_info->bytes_reserved -= num_bytes;
5398 space_info->reservation_progress++;
Yan, Zhengf0486c62010-05-16 10:46:25 -04005399 }
Josef Bacikfb25e912011-07-26 17:00:46 -04005400 spin_unlock(&cache->lock);
5401 spin_unlock(&space_info->lock);
Yan, Zhengf0486c62010-05-16 10:46:25 -04005402 return ret;
5403}
5404
Jeff Mahoney143bede2012-03-01 14:56:26 +01005405void btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
Yan Zheng11833d62009-09-11 16:11:19 -04005406 struct btrfs_root *root)
5407{
5408 struct btrfs_fs_info *fs_info = root->fs_info;
5409 struct btrfs_caching_control *next;
5410 struct btrfs_caching_control *caching_ctl;
5411 struct btrfs_block_group_cache *cache;
Josef Bacikb150a4f2013-06-19 15:00:04 -04005412 struct btrfs_space_info *space_info;
Yan Zheng11833d62009-09-11 16:11:19 -04005413
5414 down_write(&fs_info->extent_commit_sem);
5415
5416 list_for_each_entry_safe(caching_ctl, next,
5417 &fs_info->caching_block_groups, list) {
5418 cache = caching_ctl->block_group;
5419 if (block_group_cache_done(cache)) {
5420 cache->last_byte_to_unpin = (u64)-1;
5421 list_del_init(&caching_ctl->list);
5422 put_caching_control(caching_ctl);
5423 } else {
5424 cache->last_byte_to_unpin = caching_ctl->progress;
5425 }
5426 }
5427
5428 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
5429 fs_info->pinned_extents = &fs_info->freed_extents[1];
5430 else
5431 fs_info->pinned_extents = &fs_info->freed_extents[0];
5432
5433 up_write(&fs_info->extent_commit_sem);
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04005434
Josef Bacikb150a4f2013-06-19 15:00:04 -04005435 list_for_each_entry_rcu(space_info, &fs_info->space_info, list)
5436 percpu_counter_set(&space_info->total_bytes_pinned, 0);
5437
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04005438 update_global_block_rsv(fs_info);
Yan Zheng11833d62009-09-11 16:11:19 -04005439}
5440
5441static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
5442{
5443 struct btrfs_fs_info *fs_info = root->fs_info;
5444 struct btrfs_block_group_cache *cache = NULL;
Josef Bacik7b398f82012-10-22 15:52:28 -04005445 struct btrfs_space_info *space_info;
5446 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
Yan Zheng11833d62009-09-11 16:11:19 -04005447 u64 len;
Josef Bacik7b398f82012-10-22 15:52:28 -04005448 bool readonly;
Yan Zheng11833d62009-09-11 16:11:19 -04005449
5450 while (start <= end) {
Josef Bacik7b398f82012-10-22 15:52:28 -04005451 readonly = false;
Yan Zheng11833d62009-09-11 16:11:19 -04005452 if (!cache ||
5453 start >= cache->key.objectid + cache->key.offset) {
5454 if (cache)
5455 btrfs_put_block_group(cache);
5456 cache = btrfs_lookup_block_group(fs_info, start);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01005457 BUG_ON(!cache); /* Logic error */
Yan Zheng11833d62009-09-11 16:11:19 -04005458 }
5459
5460 len = cache->key.objectid + cache->key.offset - start;
5461 len = min(len, end + 1 - start);
5462
5463 if (start < cache->last_byte_to_unpin) {
5464 len = min(len, cache->last_byte_to_unpin - start);
5465 btrfs_add_free_space(cache, start, len);
5466 }
Josef Bacik25179202008-10-29 14:49:05 -04005467
Yan, Zhengf0486c62010-05-16 10:46:25 -04005468 start += len;
Josef Bacik7b398f82012-10-22 15:52:28 -04005469 space_info = cache->space_info;
Yan, Zhengf0486c62010-05-16 10:46:25 -04005470
Josef Bacik7b398f82012-10-22 15:52:28 -04005471 spin_lock(&space_info->lock);
Josef Bacik25179202008-10-29 14:49:05 -04005472 spin_lock(&cache->lock);
Yan Zheng11833d62009-09-11 16:11:19 -04005473 cache->pinned -= len;
Josef Bacik7b398f82012-10-22 15:52:28 -04005474 space_info->bytes_pinned -= len;
5475 if (cache->ro) {
5476 space_info->bytes_readonly += len;
5477 readonly = true;
5478 }
Josef Bacik25179202008-10-29 14:49:05 -04005479 spin_unlock(&cache->lock);
Josef Bacik7b398f82012-10-22 15:52:28 -04005480 if (!readonly && global_rsv->space_info == space_info) {
5481 spin_lock(&global_rsv->lock);
5482 if (!global_rsv->full) {
5483 len = min(len, global_rsv->size -
5484 global_rsv->reserved);
5485 global_rsv->reserved += len;
5486 space_info->bytes_may_use += len;
5487 if (global_rsv->reserved >= global_rsv->size)
5488 global_rsv->full = 1;
5489 }
5490 spin_unlock(&global_rsv->lock);
5491 }
5492 spin_unlock(&space_info->lock);
Yan Zheng11833d62009-09-11 16:11:19 -04005493 }
5494
5495 if (cache)
Chris Masonfa9c0d792009-04-03 09:47:43 -04005496 btrfs_put_block_group(cache);
Chris Masonccd467d2007-06-28 15:57:36 -04005497 return 0;
5498}
5499
5500int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
Yan Zheng11833d62009-09-11 16:11:19 -04005501 struct btrfs_root *root)
Chris Masona28ec192007-03-06 20:08:01 -05005502{
Yan Zheng11833d62009-09-11 16:11:19 -04005503 struct btrfs_fs_info *fs_info = root->fs_info;
5504 struct extent_io_tree *unpin;
Chris Mason1a5bc162007-10-15 16:15:26 -04005505 u64 start;
5506 u64 end;
Chris Masona28ec192007-03-06 20:08:01 -05005507 int ret;
Chris Masona28ec192007-03-06 20:08:01 -05005508
Jeff Mahoney79787ea2012-03-12 16:03:00 +01005509 if (trans->aborted)
5510 return 0;
5511
Yan Zheng11833d62009-09-11 16:11:19 -04005512 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
5513 unpin = &fs_info->freed_extents[1];
5514 else
5515 unpin = &fs_info->freed_extents[0];
5516
Chris Masond3977122009-01-05 21:25:51 -05005517 while (1) {
Chris Mason1a5bc162007-10-15 16:15:26 -04005518 ret = find_first_extent_bit(unpin, 0, &start, &end,
Josef Bacike6138872012-09-27 17:07:30 -04005519 EXTENT_DIRTY, NULL);
Chris Mason1a5bc162007-10-15 16:15:26 -04005520 if (ret)
Chris Masona28ec192007-03-06 20:08:01 -05005521 break;
Liu Hui1f3c79a2009-01-05 15:57:51 -05005522
Li Dongyang5378e602011-03-24 10:24:27 +00005523 if (btrfs_test_opt(root, DISCARD))
5524 ret = btrfs_discard_extent(root, start,
5525 end + 1 - start, NULL);
Liu Hui1f3c79a2009-01-05 15:57:51 -05005526
Chris Mason1a5bc162007-10-15 16:15:26 -04005527 clear_extent_dirty(unpin, start, end, GFP_NOFS);
Yan Zheng11833d62009-09-11 16:11:19 -04005528 unpin_extent_range(root, start, end);
Chris Masonb9473432009-03-13 11:00:37 -04005529 cond_resched();
Chris Masona28ec192007-03-06 20:08:01 -05005530 }
Josef Bacik817d52f2009-07-13 21:29:25 -04005531
Chris Masone20d96d2007-03-22 12:13:20 -04005532 return 0;
5533}
5534
Josef Bacikb150a4f2013-06-19 15:00:04 -04005535static void add_pinned_bytes(struct btrfs_fs_info *fs_info, u64 num_bytes,
5536 u64 owner, u64 root_objectid)
5537{
5538 struct btrfs_space_info *space_info;
5539 u64 flags;
5540
5541 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
5542 if (root_objectid == BTRFS_CHUNK_TREE_OBJECTID)
5543 flags = BTRFS_BLOCK_GROUP_SYSTEM;
5544 else
5545 flags = BTRFS_BLOCK_GROUP_METADATA;
5546 } else {
5547 flags = BTRFS_BLOCK_GROUP_DATA;
5548 }
5549
5550 space_info = __find_space_info(fs_info, flags);
5551 BUG_ON(!space_info); /* Logic bug */
5552 percpu_counter_add(&space_info->total_bytes_pinned, num_bytes);
5553}
5554
5555
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005556static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
5557 struct btrfs_root *root,
5558 u64 bytenr, u64 num_bytes, u64 parent,
5559 u64 root_objectid, u64 owner_objectid,
5560 u64 owner_offset, int refs_to_drop,
5561 struct btrfs_delayed_extent_op *extent_op)
Chris Masona28ec192007-03-06 20:08:01 -05005562{
Chris Masone2fa7222007-03-12 16:22:34 -04005563 struct btrfs_key key;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005564 struct btrfs_path *path;
Chris Mason1261ec42007-03-20 20:35:03 -04005565 struct btrfs_fs_info *info = root->fs_info;
5566 struct btrfs_root *extent_root = info->extent_root;
Chris Mason5f39d392007-10-15 16:14:19 -04005567 struct extent_buffer *leaf;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005568 struct btrfs_extent_item *ei;
5569 struct btrfs_extent_inline_ref *iref;
Chris Masona28ec192007-03-06 20:08:01 -05005570 int ret;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005571 int is_data;
Chris Mason952fcca2008-02-18 16:33:44 -05005572 int extent_slot = 0;
5573 int found_extent = 0;
5574 int num_to_del = 1;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005575 u32 item_size;
5576 u64 refs;
Josef Bacik3173a182013-03-07 14:22:04 -05005577 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
5578 SKINNY_METADATA);
Chris Mason037e6392007-03-07 11:50:24 -05005579
Chris Mason5caf2a02007-04-02 11:20:42 -04005580 path = btrfs_alloc_path();
Chris Mason54aa1f42007-06-22 14:16:25 -04005581 if (!path)
5582 return -ENOMEM;
5583
Chris Mason3c12ac72008-04-21 12:01:38 -04005584 path->reada = 1;
Chris Masonb9473432009-03-13 11:00:37 -04005585 path->leave_spinning = 1;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005586
5587 is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
5588 BUG_ON(!is_data && refs_to_drop != 1);
5589
Josef Bacik3173a182013-03-07 14:22:04 -05005590 if (is_data)
5591 skinny_metadata = 0;
5592
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005593 ret = lookup_extent_backref(trans, extent_root, path, &iref,
5594 bytenr, num_bytes, parent,
5595 root_objectid, owner_objectid,
5596 owner_offset);
Chris Mason7bb86312007-12-11 09:25:06 -05005597 if (ret == 0) {
Chris Mason952fcca2008-02-18 16:33:44 -05005598 extent_slot = path->slots[0];
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005599 while (extent_slot >= 0) {
5600 btrfs_item_key_to_cpu(path->nodes[0], &key,
Chris Mason952fcca2008-02-18 16:33:44 -05005601 extent_slot);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005602 if (key.objectid != bytenr)
Chris Mason952fcca2008-02-18 16:33:44 -05005603 break;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005604 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
5605 key.offset == num_bytes) {
Chris Mason952fcca2008-02-18 16:33:44 -05005606 found_extent = 1;
5607 break;
5608 }
Josef Bacik3173a182013-03-07 14:22:04 -05005609 if (key.type == BTRFS_METADATA_ITEM_KEY &&
5610 key.offset == owner_objectid) {
5611 found_extent = 1;
5612 break;
5613 }
Chris Mason952fcca2008-02-18 16:33:44 -05005614 if (path->slots[0] - extent_slot > 5)
5615 break;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005616 extent_slot--;
Chris Mason952fcca2008-02-18 16:33:44 -05005617 }
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005618#ifdef BTRFS_COMPAT_EXTENT_TREE_V0
5619 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
5620 if (found_extent && item_size < sizeof(*ei))
5621 found_extent = 0;
5622#endif
Zheng Yan31840ae2008-09-23 13:14:14 -04005623 if (!found_extent) {
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005624 BUG_ON(iref);
Chris Mason56bec292009-03-13 10:10:06 -04005625 ret = remove_extent_backref(trans, extent_root, path,
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005626 NULL, refs_to_drop,
5627 is_data);
David Sterba005d6422012-09-18 07:52:32 -06005628 if (ret) {
5629 btrfs_abort_transaction(trans, extent_root, ret);
5630 goto out;
5631 }
David Sterbab3b4aa72011-04-21 01:20:15 +02005632 btrfs_release_path(path);
Chris Masonb9473432009-03-13 11:00:37 -04005633 path->leave_spinning = 1;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005634
5635 key.objectid = bytenr;
5636 key.type = BTRFS_EXTENT_ITEM_KEY;
5637 key.offset = num_bytes;
5638
Josef Bacik3173a182013-03-07 14:22:04 -05005639 if (!is_data && skinny_metadata) {
5640 key.type = BTRFS_METADATA_ITEM_KEY;
5641 key.offset = owner_objectid;
5642 }
5643
Zheng Yan31840ae2008-09-23 13:14:14 -04005644 ret = btrfs_search_slot(trans, extent_root,
5645 &key, path, -1, 1);
Josef Bacik3173a182013-03-07 14:22:04 -05005646 if (ret > 0 && skinny_metadata && path->slots[0]) {
5647 /*
5648 * Couldn't find our skinny metadata item,
5649 * see if we have ye olde extent item.
5650 */
5651 path->slots[0]--;
5652 btrfs_item_key_to_cpu(path->nodes[0], &key,
5653 path->slots[0]);
5654 if (key.objectid == bytenr &&
5655 key.type == BTRFS_EXTENT_ITEM_KEY &&
5656 key.offset == num_bytes)
5657 ret = 0;
5658 }
5659
5660 if (ret > 0 && skinny_metadata) {
5661 skinny_metadata = false;
5662 key.type = BTRFS_EXTENT_ITEM_KEY;
5663 key.offset = num_bytes;
5664 btrfs_release_path(path);
5665 ret = btrfs_search_slot(trans, extent_root,
5666 &key, path, -1, 1);
5667 }
5668
Josef Bacikf3465ca2008-11-12 14:19:50 -05005669 if (ret) {
Simon Kirbyc2cf52e2013-03-19 22:41:23 +00005670 btrfs_err(info, "umm, got %d back from search, was looking for %llu",
5671 ret, (unsigned long long)bytenr);
Josef Bacikb783e622011-07-13 15:03:50 +00005672 if (ret > 0)
5673 btrfs_print_leaf(extent_root,
5674 path->nodes[0]);
Josef Bacikf3465ca2008-11-12 14:19:50 -05005675 }
David Sterba005d6422012-09-18 07:52:32 -06005676 if (ret < 0) {
5677 btrfs_abort_transaction(trans, extent_root, ret);
5678 goto out;
5679 }
Zheng Yan31840ae2008-09-23 13:14:14 -04005680 extent_slot = path->slots[0];
5681 }
Jeff Mahoney79787ea2012-03-12 16:03:00 +01005682 } else if (ret == -ENOENT) {
Chris Mason7bb86312007-12-11 09:25:06 -05005683 btrfs_print_leaf(extent_root, path->nodes[0]);
5684 WARN_ON(1);
Simon Kirbyc2cf52e2013-03-19 22:41:23 +00005685 btrfs_err(info,
5686 "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu",
5687 (unsigned long long)bytenr,
5688 (unsigned long long)parent,
5689 (unsigned long long)root_objectid,
5690 (unsigned long long)owner_objectid,
5691 (unsigned long long)owner_offset);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01005692 } else {
David Sterba005d6422012-09-18 07:52:32 -06005693 btrfs_abort_transaction(trans, extent_root, ret);
5694 goto out;
Chris Mason7bb86312007-12-11 09:25:06 -05005695 }
Chris Mason5f39d392007-10-15 16:14:19 -04005696
5697 leaf = path->nodes[0];
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005698 item_size = btrfs_item_size_nr(leaf, extent_slot);
5699#ifdef BTRFS_COMPAT_EXTENT_TREE_V0
5700 if (item_size < sizeof(*ei)) {
5701 BUG_ON(found_extent || extent_slot != path->slots[0]);
5702 ret = convert_extent_item_v0(trans, extent_root, path,
5703 owner_objectid, 0);
David Sterba005d6422012-09-18 07:52:32 -06005704 if (ret < 0) {
5705 btrfs_abort_transaction(trans, extent_root, ret);
5706 goto out;
5707 }
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005708
David Sterbab3b4aa72011-04-21 01:20:15 +02005709 btrfs_release_path(path);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005710 path->leave_spinning = 1;
5711
5712 key.objectid = bytenr;
5713 key.type = BTRFS_EXTENT_ITEM_KEY;
5714 key.offset = num_bytes;
5715
5716 ret = btrfs_search_slot(trans, extent_root, &key, path,
5717 -1, 1);
5718 if (ret) {
Simon Kirbyc2cf52e2013-03-19 22:41:23 +00005719 btrfs_err(info, "umm, got %d back from search, was looking for %llu",
5720 ret, (unsigned long long)bytenr);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005721 btrfs_print_leaf(extent_root, path->nodes[0]);
5722 }
David Sterba005d6422012-09-18 07:52:32 -06005723 if (ret < 0) {
5724 btrfs_abort_transaction(trans, extent_root, ret);
5725 goto out;
5726 }
5727
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005728 extent_slot = path->slots[0];
5729 leaf = path->nodes[0];
5730 item_size = btrfs_item_size_nr(leaf, extent_slot);
5731 }
5732#endif
5733 BUG_ON(item_size < sizeof(*ei));
Chris Mason952fcca2008-02-18 16:33:44 -05005734 ei = btrfs_item_ptr(leaf, extent_slot,
Chris Mason123abc82007-03-14 14:14:43 -04005735 struct btrfs_extent_item);
Josef Bacik3173a182013-03-07 14:22:04 -05005736 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID &&
5737 key.type == BTRFS_EXTENT_ITEM_KEY) {
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005738 struct btrfs_tree_block_info *bi;
5739 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
5740 bi = (struct btrfs_tree_block_info *)(ei + 1);
5741 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
Chris Mason952fcca2008-02-18 16:33:44 -05005742 }
5743
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005744 refs = btrfs_extent_refs(leaf, ei);
Josef Bacik32b02532013-04-24 16:38:50 -04005745 if (refs < refs_to_drop) {
5746 btrfs_err(info, "trying to drop %d refs but we only have %Lu "
5747 "for bytenr %Lu\n", refs_to_drop, refs, bytenr);
5748 ret = -EINVAL;
5749 btrfs_abort_transaction(trans, extent_root, ret);
5750 goto out;
5751 }
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005752 refs -= refs_to_drop;
5753
5754 if (refs > 0) {
5755 if (extent_op)
5756 __run_delayed_extent_op(extent_op, leaf, ei);
5757 /*
5758 * In the case of inline back ref, reference count will
5759 * be updated by remove_extent_backref
5760 */
5761 if (iref) {
5762 BUG_ON(!found_extent);
5763 } else {
5764 btrfs_set_extent_refs(leaf, ei, refs);
5765 btrfs_mark_buffer_dirty(leaf);
5766 }
5767 if (found_extent) {
5768 ret = remove_extent_backref(trans, extent_root, path,
5769 iref, refs_to_drop,
5770 is_data);
David Sterba005d6422012-09-18 07:52:32 -06005771 if (ret) {
5772 btrfs_abort_transaction(trans, extent_root, ret);
5773 goto out;
5774 }
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005775 }
Josef Bacikb150a4f2013-06-19 15:00:04 -04005776 add_pinned_bytes(root->fs_info, -num_bytes, owner_objectid,
5777 root_objectid);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005778 } else {
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005779 if (found_extent) {
5780 BUG_ON(is_data && refs_to_drop !=
5781 extent_data_ref_count(root, path, iref));
5782 if (iref) {
5783 BUG_ON(path->slots[0] != extent_slot);
5784 } else {
5785 BUG_ON(path->slots[0] != extent_slot + 1);
5786 path->slots[0] = extent_slot;
5787 num_to_del = 2;
5788 }
Chris Mason78fae272007-03-25 11:35:08 -04005789 }
Chris Masonb9473432009-03-13 11:00:37 -04005790
Chris Mason952fcca2008-02-18 16:33:44 -05005791 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
5792 num_to_del);
David Sterba005d6422012-09-18 07:52:32 -06005793 if (ret) {
5794 btrfs_abort_transaction(trans, extent_root, ret);
5795 goto out;
5796 }
David Sterbab3b4aa72011-04-21 01:20:15 +02005797 btrfs_release_path(path);
David Woodhouse21af8042008-08-12 14:13:26 +01005798
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005799 if (is_data) {
Chris Mason459931e2008-12-10 09:10:46 -05005800 ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
David Sterba005d6422012-09-18 07:52:32 -06005801 if (ret) {
5802 btrfs_abort_transaction(trans, extent_root, ret);
5803 goto out;
5804 }
Chris Mason459931e2008-12-10 09:10:46 -05005805 }
5806
Liu Boc53d6132012-12-27 09:01:19 +00005807 ret = update_block_group(root, bytenr, num_bytes, 0);
David Sterba005d6422012-09-18 07:52:32 -06005808 if (ret) {
5809 btrfs_abort_transaction(trans, extent_root, ret);
5810 goto out;
5811 }
Chris Masona28ec192007-03-06 20:08:01 -05005812 }
Jeff Mahoney79787ea2012-03-12 16:03:00 +01005813out:
Chris Mason5caf2a02007-04-02 11:20:42 -04005814 btrfs_free_path(path);
Chris Masona28ec192007-03-06 20:08:01 -05005815 return ret;
5816}
5817
5818/*
Yan, Zhengf0486c62010-05-16 10:46:25 -04005819 * when we free an block, it is possible (and likely) that we free the last
Chris Mason1887be62009-03-13 10:11:24 -04005820 * delayed ref for that extent as well. This searches the delayed ref tree for
5821 * a given extent, and if there are no other delayed refs to be processed, it
5822 * removes it from the tree.
5823 */
5824static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
5825 struct btrfs_root *root, u64 bytenr)
5826{
5827 struct btrfs_delayed_ref_head *head;
5828 struct btrfs_delayed_ref_root *delayed_refs;
5829 struct btrfs_delayed_ref_node *ref;
5830 struct rb_node *node;
Yan, Zhengf0486c62010-05-16 10:46:25 -04005831 int ret = 0;
Chris Mason1887be62009-03-13 10:11:24 -04005832
5833 delayed_refs = &trans->transaction->delayed_refs;
5834 spin_lock(&delayed_refs->lock);
5835 head = btrfs_find_delayed_ref_head(trans, bytenr);
5836 if (!head)
5837 goto out;
5838
5839 node = rb_prev(&head->node.rb_node);
5840 if (!node)
5841 goto out;
5842
5843 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
5844
5845 /* there are still entries for this ref, we can't drop it */
5846 if (ref->bytenr == bytenr)
5847 goto out;
5848
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005849 if (head->extent_op) {
5850 if (!head->must_insert_reserved)
5851 goto out;
Miao Xie78a61842012-11-21 02:21:28 +00005852 btrfs_free_delayed_extent_op(head->extent_op);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005853 head->extent_op = NULL;
5854 }
5855
Chris Mason1887be62009-03-13 10:11:24 -04005856 /*
5857 * waiting for the lock here would deadlock. If someone else has it
5858 * locked they are already in the process of dropping it anyway
5859 */
5860 if (!mutex_trylock(&head->mutex))
5861 goto out;
5862
5863 /*
5864 * at this point we have a head with no other entries. Go
5865 * ahead and process it.
5866 */
5867 head->node.in_tree = 0;
5868 rb_erase(&head->node.rb_node, &delayed_refs->root);
Chris Masonc3e69d52009-03-13 10:17:05 -04005869
Chris Mason1887be62009-03-13 10:11:24 -04005870 delayed_refs->num_entries--;
5871
5872 /*
5873 * we don't take a ref on the node because we're removing it from the
5874 * tree, so we just steal the ref the tree was holding.
5875 */
Chris Masonc3e69d52009-03-13 10:17:05 -04005876 delayed_refs->num_heads--;
5877 if (list_empty(&head->cluster))
5878 delayed_refs->num_heads_ready--;
5879
5880 list_del_init(&head->cluster);
Chris Mason1887be62009-03-13 10:11:24 -04005881 spin_unlock(&delayed_refs->lock);
5882
Yan, Zhengf0486c62010-05-16 10:46:25 -04005883 BUG_ON(head->extent_op);
5884 if (head->must_insert_reserved)
5885 ret = 1;
5886
5887 mutex_unlock(&head->mutex);
Chris Mason1887be62009-03-13 10:11:24 -04005888 btrfs_put_delayed_ref(&head->node);
Yan, Zhengf0486c62010-05-16 10:46:25 -04005889 return ret;
Chris Mason1887be62009-03-13 10:11:24 -04005890out:
5891 spin_unlock(&delayed_refs->lock);
5892 return 0;
5893}
5894
Yan, Zhengf0486c62010-05-16 10:46:25 -04005895void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
5896 struct btrfs_root *root,
5897 struct extent_buffer *buf,
Jan Schmidt5581a512012-05-16 17:04:52 +02005898 u64 parent, int last_ref)
Yan, Zhengf0486c62010-05-16 10:46:25 -04005899{
Yan, Zhengf0486c62010-05-16 10:46:25 -04005900 struct btrfs_block_group_cache *cache = NULL;
Josef Bacikb150a4f2013-06-19 15:00:04 -04005901 int pin = 1;
Yan, Zhengf0486c62010-05-16 10:46:25 -04005902 int ret;
5903
5904 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
Arne Jansen66d7e7f2011-09-12 15:26:38 +02005905 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
5906 buf->start, buf->len,
5907 parent, root->root_key.objectid,
5908 btrfs_header_level(buf),
Jan Schmidt5581a512012-05-16 17:04:52 +02005909 BTRFS_DROP_DELAYED_REF, NULL, 0);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01005910 BUG_ON(ret); /* -ENOMEM */
Yan, Zhengf0486c62010-05-16 10:46:25 -04005911 }
5912
5913 if (!last_ref)
5914 return;
5915
Yan, Zhengf0486c62010-05-16 10:46:25 -04005916 cache = btrfs_lookup_block_group(root->fs_info, buf->start);
Yan, Zhengf0486c62010-05-16 10:46:25 -04005917
5918 if (btrfs_header_generation(buf) == trans->transid) {
5919 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
5920 ret = check_ref_cleanup(trans, root, buf->start);
5921 if (!ret)
Josef Bacik37be25b2011-08-05 10:25:38 -04005922 goto out;
Yan, Zhengf0486c62010-05-16 10:46:25 -04005923 }
5924
5925 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
5926 pin_down_extent(root, cache, buf->start, buf->len, 1);
Josef Bacik37be25b2011-08-05 10:25:38 -04005927 goto out;
Yan, Zhengf0486c62010-05-16 10:46:25 -04005928 }
5929
5930 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
5931
5932 btrfs_add_free_space(cache, buf->start, buf->len);
Josef Bacikfb25e912011-07-26 17:00:46 -04005933 btrfs_update_reserved_bytes(cache, buf->len, RESERVE_FREE);
Josef Bacikb150a4f2013-06-19 15:00:04 -04005934 pin = 0;
Yan, Zhengf0486c62010-05-16 10:46:25 -04005935 }
5936out:
Josef Bacikb150a4f2013-06-19 15:00:04 -04005937 if (pin)
5938 add_pinned_bytes(root->fs_info, buf->len,
5939 btrfs_header_level(buf),
5940 root->root_key.objectid);
5941
Josef Bacika826d6d2011-03-16 13:42:43 -04005942 /*
5943 * Deleting the buffer, clear the corrupt flag since it doesn't matter
5944 * anymore.
5945 */
5946 clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
Yan, Zhengf0486c62010-05-16 10:46:25 -04005947 btrfs_put_block_group(cache);
5948}
5949
Jeff Mahoney79787ea2012-03-12 16:03:00 +01005950/* Can return -ENOMEM */
Arne Jansen66d7e7f2011-09-12 15:26:38 +02005951int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root *root,
5952 u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid,
5953 u64 owner, u64 offset, int for_cow)
Chris Mason925baed2008-06-25 16:01:30 -04005954{
5955 int ret;
Arne Jansen66d7e7f2011-09-12 15:26:38 +02005956 struct btrfs_fs_info *fs_info = root->fs_info;
Chris Mason925baed2008-06-25 16:01:30 -04005957
Josef Bacikb150a4f2013-06-19 15:00:04 -04005958 add_pinned_bytes(root->fs_info, num_bytes, owner, root_objectid);
5959
Chris Mason56bec292009-03-13 10:10:06 -04005960 /*
5961 * tree log blocks never actually go into the extent allocation
5962 * tree, just update pinning info and exit early.
Chris Mason56bec292009-03-13 10:10:06 -04005963 */
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005964 if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
5965 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
Chris Masonb9473432009-03-13 11:00:37 -04005966 /* unlocks the pinned mutex */
Yan Zheng11833d62009-09-11 16:11:19 -04005967 btrfs_pin_extent(root, bytenr, num_bytes, 1);
Chris Mason56bec292009-03-13 10:10:06 -04005968 ret = 0;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005969 } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
Arne Jansen66d7e7f2011-09-12 15:26:38 +02005970 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
5971 num_bytes,
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005972 parent, root_objectid, (int)owner,
Arne Jansen66d7e7f2011-09-12 15:26:38 +02005973 BTRFS_DROP_DELAYED_REF, NULL, for_cow);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005974 } else {
Arne Jansen66d7e7f2011-09-12 15:26:38 +02005975 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
5976 num_bytes,
5977 parent, root_objectid, owner,
5978 offset, BTRFS_DROP_DELAYED_REF,
5979 NULL, for_cow);
Chris Mason56bec292009-03-13 10:10:06 -04005980 }
Chris Mason925baed2008-06-25 16:01:30 -04005981 return ret;
5982}
5983
David Woodhouse53b381b2013-01-29 18:40:14 -05005984static u64 stripe_align(struct btrfs_root *root,
5985 struct btrfs_block_group_cache *cache,
5986 u64 val, u64 num_bytes)
Chris Mason87ee04e2007-11-30 11:30:34 -05005987{
Qu Wenruofda28322013-02-26 08:10:22 +00005988 u64 ret = ALIGN(val, root->stripesize);
Chris Mason87ee04e2007-11-30 11:30:34 -05005989 return ret;
5990}
5991
Chris Masonfec577f2007-02-26 10:40:21 -05005992/*
Josef Bacik817d52f2009-07-13 21:29:25 -04005993 * when we wait for progress in the block group caching, its because
5994 * our allocation attempt failed at least once. So, we must sleep
5995 * and let some progress happen before we try again.
5996 *
5997 * This function will sleep at least once waiting for new free space to
5998 * show up, and then it will check the block group free space numbers
5999 * for our min num_bytes. Another option is to have it go ahead
6000 * and look in the rbtree for a free extent of a given size, but this
6001 * is a good start.
6002 */
6003static noinline int
6004wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
6005 u64 num_bytes)
6006{
Yan Zheng11833d62009-09-11 16:11:19 -04006007 struct btrfs_caching_control *caching_ctl;
Josef Bacik817d52f2009-07-13 21:29:25 -04006008
Yan Zheng11833d62009-09-11 16:11:19 -04006009 caching_ctl = get_caching_control(cache);
6010 if (!caching_ctl)
Josef Bacik817d52f2009-07-13 21:29:25 -04006011 return 0;
Josef Bacik817d52f2009-07-13 21:29:25 -04006012
Yan Zheng11833d62009-09-11 16:11:19 -04006013 wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
Li Zefan34d52cb2011-03-29 13:46:06 +08006014 (cache->free_space_ctl->free_space >= num_bytes));
Yan Zheng11833d62009-09-11 16:11:19 -04006015
6016 put_caching_control(caching_ctl);
6017 return 0;
6018}
6019
6020static noinline int
6021wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
6022{
6023 struct btrfs_caching_control *caching_ctl;
Yan Zheng11833d62009-09-11 16:11:19 -04006024
6025 caching_ctl = get_caching_control(cache);
6026 if (!caching_ctl)
6027 return 0;
6028
6029 wait_event(caching_ctl->wait, block_group_cache_done(cache));
6030
6031 put_caching_control(caching_ctl);
Josef Bacik817d52f2009-07-13 21:29:25 -04006032 return 0;
6033}
6034
Liu Bo31e50222012-11-21 14:18:10 +00006035int __get_raid_index(u64 flags)
Yan, Zhengb742bb82010-05-16 10:46:24 -04006036{
Ilya Dryomov7738a532012-03-27 17:09:17 +03006037 if (flags & BTRFS_BLOCK_GROUP_RAID10)
Miao Xiee6ec7162013-01-17 05:38:51 +00006038 return BTRFS_RAID_RAID10;
Ilya Dryomov7738a532012-03-27 17:09:17 +03006039 else if (flags & BTRFS_BLOCK_GROUP_RAID1)
Miao Xiee6ec7162013-01-17 05:38:51 +00006040 return BTRFS_RAID_RAID1;
Ilya Dryomov7738a532012-03-27 17:09:17 +03006041 else if (flags & BTRFS_BLOCK_GROUP_DUP)
Miao Xiee6ec7162013-01-17 05:38:51 +00006042 return BTRFS_RAID_DUP;
Ilya Dryomov7738a532012-03-27 17:09:17 +03006043 else if (flags & BTRFS_BLOCK_GROUP_RAID0)
Miao Xiee6ec7162013-01-17 05:38:51 +00006044 return BTRFS_RAID_RAID0;
David Woodhouse53b381b2013-01-29 18:40:14 -05006045 else if (flags & BTRFS_BLOCK_GROUP_RAID5)
Chris Masone942f882013-02-20 14:06:05 -05006046 return BTRFS_RAID_RAID5;
David Woodhouse53b381b2013-01-29 18:40:14 -05006047 else if (flags & BTRFS_BLOCK_GROUP_RAID6)
Chris Masone942f882013-02-20 14:06:05 -05006048 return BTRFS_RAID_RAID6;
Ilya Dryomov7738a532012-03-27 17:09:17 +03006049
Chris Masone942f882013-02-20 14:06:05 -05006050 return BTRFS_RAID_SINGLE; /* BTRFS_BLOCK_GROUP_SINGLE */
Yan, Zhengb742bb82010-05-16 10:46:24 -04006051}
6052
Ilya Dryomov7738a532012-03-27 17:09:17 +03006053static int get_block_group_index(struct btrfs_block_group_cache *cache)
6054{
Liu Bo31e50222012-11-21 14:18:10 +00006055 return __get_raid_index(cache->flags);
Ilya Dryomov7738a532012-03-27 17:09:17 +03006056}
6057
Josef Bacik817d52f2009-07-13 21:29:25 -04006058enum btrfs_loop_type {
Josef Bacik285ff5a2012-01-13 15:27:45 -05006059 LOOP_CACHING_NOWAIT = 0,
6060 LOOP_CACHING_WAIT = 1,
6061 LOOP_ALLOC_CHUNK = 2,
6062 LOOP_NO_EMPTY_SIZE = 3,
Josef Bacik817d52f2009-07-13 21:29:25 -04006063};
6064
6065/*
Chris Masonfec577f2007-02-26 10:40:21 -05006066 * walks the btree of allocated extents and find a hole of a given size.
6067 * The key ins is changed to record the hole:
6068 * ins->objectid == block start
Chris Mason62e27492007-03-15 12:56:47 -04006069 * ins->flags = BTRFS_EXTENT_ITEM_KEY
Chris Masonfec577f2007-02-26 10:40:21 -05006070 * ins->offset == number of blocks
6071 * Any available blocks before search_start are skipped.
6072 */
Chris Masond3977122009-01-05 21:25:51 -05006073static noinline int find_free_extent(struct btrfs_trans_handle *trans,
Chris Mason98ed5172008-01-03 10:01:48 -05006074 struct btrfs_root *orig_root,
6075 u64 num_bytes, u64 empty_size,
Chris Mason98ed5172008-01-03 10:01:48 -05006076 u64 hint_byte, struct btrfs_key *ins,
David Sterbab6919a52013-04-29 13:39:40 +00006077 u64 flags)
Chris Masonfec577f2007-02-26 10:40:21 -05006078{
Josef Bacik80eb2342008-10-29 14:49:05 -04006079 int ret = 0;
Chris Masond3977122009-01-05 21:25:51 -05006080 struct btrfs_root *root = orig_root->fs_info->extent_root;
Chris Masonfa9c0d792009-04-03 09:47:43 -04006081 struct btrfs_free_cluster *last_ptr = NULL;
Josef Bacik80eb2342008-10-29 14:49:05 -04006082 struct btrfs_block_group_cache *block_group = NULL;
Alexandre Oliva274bd4f2011-12-07 20:08:40 -05006083 struct btrfs_block_group_cache *used_block_group;
Josef Bacik81c9ad22012-01-18 10:56:06 -05006084 u64 search_start = 0;
Chris Mason239b14b2008-03-24 15:02:07 -04006085 int empty_cluster = 2 * 1024 * 1024;
Josef Bacik80eb2342008-10-29 14:49:05 -04006086 struct btrfs_space_info *space_info;
Chris Masonfa9c0d792009-04-03 09:47:43 -04006087 int loop = 0;
David Sterbab6919a52013-04-29 13:39:40 +00006088 int index = __get_raid_index(flags);
6089 int alloc_type = (flags & BTRFS_BLOCK_GROUP_DATA) ?
Josef Bacikfb25e912011-07-26 17:00:46 -04006090 RESERVE_ALLOC_NO_ACCOUNT : RESERVE_ALLOC;
Josef Bacik817d52f2009-07-13 21:29:25 -04006091 bool found_uncached_bg = false;
Josef Bacik0a243252009-09-11 16:11:20 -04006092 bool failed_cluster_refill = false;
Josef Bacik1cdda9b2009-10-06 10:04:28 -04006093 bool failed_alloc = false;
Josef Bacik67377732010-09-16 16:19:09 -04006094 bool use_cluster = true;
Miao Xie60d2adb2011-09-09 17:34:35 +08006095 bool have_caching_bg = false;
Chris Masonfec577f2007-02-26 10:40:21 -05006096
Chris Masondb945352007-10-15 16:15:53 -04006097 WARN_ON(num_bytes < root->sectorsize);
Chris Masonb1a4d962007-04-04 15:27:52 -04006098 btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
Josef Bacik80eb2342008-10-29 14:49:05 -04006099 ins->objectid = 0;
6100 ins->offset = 0;
Chris Masonb1a4d962007-04-04 15:27:52 -04006101
David Sterbab6919a52013-04-29 13:39:40 +00006102 trace_find_free_extent(orig_root, num_bytes, empty_size, flags);
Josef Bacik3f7de032011-11-10 08:29:20 -05006103
David Sterbab6919a52013-04-29 13:39:40 +00006104 space_info = __find_space_info(root->fs_info, flags);
Josef Bacik1b1d1f62010-03-19 20:49:55 +00006105 if (!space_info) {
David Sterbab6919a52013-04-29 13:39:40 +00006106 btrfs_err(root->fs_info, "No space info for %llu", flags);
Josef Bacik1b1d1f62010-03-19 20:49:55 +00006107 return -ENOSPC;
6108 }
Josef Bacik2552d172009-04-03 10:14:19 -04006109
Josef Bacik67377732010-09-16 16:19:09 -04006110 /*
6111 * If the space info is for both data and metadata it means we have a
6112 * small filesystem and we can't use the clustering stuff.
6113 */
6114 if (btrfs_mixed_space_info(space_info))
6115 use_cluster = false;
6116
David Sterbab6919a52013-04-29 13:39:40 +00006117 if (flags & BTRFS_BLOCK_GROUP_METADATA && use_cluster) {
Chris Masonfa9c0d792009-04-03 09:47:43 -04006118 last_ptr = &root->fs_info->meta_alloc_cluster;
Chris Mason536ac8a2009-02-12 09:41:38 -05006119 if (!btrfs_test_opt(root, SSD))
6120 empty_cluster = 64 * 1024;
Chris Mason239b14b2008-03-24 15:02:07 -04006121 }
6122
David Sterbab6919a52013-04-29 13:39:40 +00006123 if ((flags & BTRFS_BLOCK_GROUP_DATA) && use_cluster &&
Josef Bacik67377732010-09-16 16:19:09 -04006124 btrfs_test_opt(root, SSD)) {
Chris Masonfa9c0d792009-04-03 09:47:43 -04006125 last_ptr = &root->fs_info->data_alloc_cluster;
6126 }
Josef Bacik0f9dd462008-09-23 13:14:11 -04006127
Chris Mason239b14b2008-03-24 15:02:07 -04006128 if (last_ptr) {
Chris Masonfa9c0d792009-04-03 09:47:43 -04006129 spin_lock(&last_ptr->lock);
6130 if (last_ptr->block_group)
6131 hint_byte = last_ptr->window_start;
6132 spin_unlock(&last_ptr->lock);
Chris Mason239b14b2008-03-24 15:02:07 -04006133 }
Chris Masonfa9c0d792009-04-03 09:47:43 -04006134
Chris Masona061fc82008-05-07 11:43:44 -04006135 search_start = max(search_start, first_logical_byte(root, 0));
Chris Mason239b14b2008-03-24 15:02:07 -04006136 search_start = max(search_start, hint_byte);
Chris Mason0b86a832008-03-24 15:01:56 -04006137
Josef Bacik817d52f2009-07-13 21:29:25 -04006138 if (!last_ptr)
Chris Masonfa9c0d792009-04-03 09:47:43 -04006139 empty_cluster = 0;
Chris Masonfa9c0d792009-04-03 09:47:43 -04006140
Josef Bacik2552d172009-04-03 10:14:19 -04006141 if (search_start == hint_byte) {
Josef Bacik2552d172009-04-03 10:14:19 -04006142 block_group = btrfs_lookup_block_group(root->fs_info,
6143 search_start);
Alexandre Oliva274bd4f2011-12-07 20:08:40 -05006144 used_block_group = block_group;
Josef Bacik817d52f2009-07-13 21:29:25 -04006145 /*
6146 * we don't want to use the block group if it doesn't match our
6147 * allocation bits, or if its not cached.
Josef Bacikccf0e722009-11-10 21:23:48 -05006148 *
6149 * However if we are re-searching with an ideal block group
6150 * picked out then we don't care that the block group is cached.
Josef Bacik817d52f2009-07-13 21:29:25 -04006151 */
David Sterbab6919a52013-04-29 13:39:40 +00006152 if (block_group && block_group_bits(block_group, flags) &&
Josef Bacik285ff5a2012-01-13 15:27:45 -05006153 block_group->cached != BTRFS_CACHE_NO) {
Josef Bacik2552d172009-04-03 10:14:19 -04006154 down_read(&space_info->groups_sem);
Chris Mason44fb5512009-06-04 15:34:51 -04006155 if (list_empty(&block_group->list) ||
6156 block_group->ro) {
6157 /*
6158 * someone is removing this block group,
6159 * we can't jump into the have_block_group
6160 * target because our list pointers are not
6161 * valid
6162 */
6163 btrfs_put_block_group(block_group);
6164 up_read(&space_info->groups_sem);
Josef Bacikccf0e722009-11-10 21:23:48 -05006165 } else {
Yan, Zhengb742bb82010-05-16 10:46:24 -04006166 index = get_block_group_index(block_group);
Chris Mason44fb5512009-06-04 15:34:51 -04006167 goto have_block_group;
Josef Bacikccf0e722009-11-10 21:23:48 -05006168 }
Josef Bacik2552d172009-04-03 10:14:19 -04006169 } else if (block_group) {
Chris Masonfa9c0d792009-04-03 09:47:43 -04006170 btrfs_put_block_group(block_group);
Josef Bacik2552d172009-04-03 10:14:19 -04006171 }
Chris Mason42e70e72008-11-07 18:17:11 -05006172 }
Josef Bacik2552d172009-04-03 10:14:19 -04006173search:
Miao Xie60d2adb2011-09-09 17:34:35 +08006174 have_caching_bg = false;
Josef Bacik80eb2342008-10-29 14:49:05 -04006175 down_read(&space_info->groups_sem);
Yan, Zhengb742bb82010-05-16 10:46:24 -04006176 list_for_each_entry(block_group, &space_info->block_groups[index],
6177 list) {
Josef Bacik6226cb0a2009-04-03 10:14:18 -04006178 u64 offset;
Josef Bacik817d52f2009-07-13 21:29:25 -04006179 int cached;
Chris Mason8a1413a2008-11-10 16:13:54 -05006180
Alexandre Oliva274bd4f2011-12-07 20:08:40 -05006181 used_block_group = block_group;
Josef Bacik11dfe352009-11-13 20:12:59 +00006182 btrfs_get_block_group(block_group);
Josef Bacik2552d172009-04-03 10:14:19 -04006183 search_start = block_group->key.objectid;
Chris Mason42e70e72008-11-07 18:17:11 -05006184
Chris Mason83a50de2010-12-13 15:06:46 -05006185 /*
6186 * this can happen if we end up cycling through all the
6187 * raid types, but we want to make sure we only allocate
6188 * for the proper type.
6189 */
David Sterbab6919a52013-04-29 13:39:40 +00006190 if (!block_group_bits(block_group, flags)) {
Chris Mason83a50de2010-12-13 15:06:46 -05006191 u64 extra = BTRFS_BLOCK_GROUP_DUP |
6192 BTRFS_BLOCK_GROUP_RAID1 |
David Woodhouse53b381b2013-01-29 18:40:14 -05006193 BTRFS_BLOCK_GROUP_RAID5 |
6194 BTRFS_BLOCK_GROUP_RAID6 |
Chris Mason83a50de2010-12-13 15:06:46 -05006195 BTRFS_BLOCK_GROUP_RAID10;
6196
6197 /*
6198 * if they asked for extra copies and this block group
6199 * doesn't provide them, bail. This does allow us to
6200 * fill raid0 from raid1.
6201 */
David Sterbab6919a52013-04-29 13:39:40 +00006202 if ((flags & extra) && !(block_group->flags & extra))
Chris Mason83a50de2010-12-13 15:06:46 -05006203 goto loop;
6204 }
6205
Josef Bacik2552d172009-04-03 10:14:19 -04006206have_block_group:
Josef Bacik291c7d22011-11-14 13:52:14 -05006207 cached = block_group_cache_done(block_group);
6208 if (unlikely(!cached)) {
Josef Bacik291c7d22011-11-14 13:52:14 -05006209 found_uncached_bg = true;
Liu Bof6373bf2012-12-27 09:01:18 +00006210 ret = cache_block_group(block_group, 0);
Chris Mason1d4284b2012-03-28 20:31:37 -04006211 BUG_ON(ret < 0);
6212 ret = 0;
Josef Bacikea6a4782008-11-20 12:16:16 -05006213 }
6214
Josef Bacikea6a4782008-11-20 12:16:16 -05006215 if (unlikely(block_group->ro))
Josef Bacik2552d172009-04-03 10:14:19 -04006216 goto loop;
Josef Bacik0f9dd462008-09-23 13:14:11 -04006217
Josef Bacik0a243252009-09-11 16:11:20 -04006218 /*
Alexandre Oliva062c05c2011-12-07 19:50:42 -05006219 * Ok we want to try and use the cluster allocator, so
6220 * lets look there
Josef Bacik0a243252009-09-11 16:11:20 -04006221 */
Alexandre Oliva062c05c2011-12-07 19:50:42 -05006222 if (last_ptr) {
Chris Mason8de972b2013-01-04 15:39:43 -05006223 unsigned long aligned_cluster;
Chris Masonfa9c0d792009-04-03 09:47:43 -04006224 /*
6225 * the refill lock keeps out other
6226 * people trying to start a new cluster
6227 */
6228 spin_lock(&last_ptr->refill_lock);
Alexandre Oliva274bd4f2011-12-07 20:08:40 -05006229 used_block_group = last_ptr->block_group;
6230 if (used_block_group != block_group &&
6231 (!used_block_group ||
6232 used_block_group->ro ||
David Sterbab6919a52013-04-29 13:39:40 +00006233 !block_group_bits(used_block_group, flags))) {
Alexandre Oliva274bd4f2011-12-07 20:08:40 -05006234 used_block_group = block_group;
Chris Mason44fb5512009-06-04 15:34:51 -04006235 goto refill_cluster;
Alexandre Oliva274bd4f2011-12-07 20:08:40 -05006236 }
Chris Mason44fb5512009-06-04 15:34:51 -04006237
Alexandre Oliva274bd4f2011-12-07 20:08:40 -05006238 if (used_block_group != block_group)
6239 btrfs_get_block_group(used_block_group);
6240
6241 offset = btrfs_alloc_from_cluster(used_block_group,
6242 last_ptr, num_bytes, used_block_group->key.objectid);
Chris Masonfa9c0d792009-04-03 09:47:43 -04006243 if (offset) {
6244 /* we have a block, we're done */
6245 spin_unlock(&last_ptr->refill_lock);
Josef Bacik3f7de032011-11-10 08:29:20 -05006246 trace_btrfs_reserve_extent_cluster(root,
6247 block_group, search_start, num_bytes);
Chris Masonfa9c0d792009-04-03 09:47:43 -04006248 goto checks;
6249 }
6250
Alexandre Oliva274bd4f2011-12-07 20:08:40 -05006251 WARN_ON(last_ptr->block_group != used_block_group);
6252 if (used_block_group != block_group) {
6253 btrfs_put_block_group(used_block_group);
6254 used_block_group = block_group;
Chris Masonfa9c0d792009-04-03 09:47:43 -04006255 }
Chris Mason44fb5512009-06-04 15:34:51 -04006256refill_cluster:
Alexandre Oliva274bd4f2011-12-07 20:08:40 -05006257 BUG_ON(used_block_group != block_group);
Alexandre Oliva062c05c2011-12-07 19:50:42 -05006258 /* If we are on LOOP_NO_EMPTY_SIZE, we can't
6259 * set up a new clusters, so lets just skip it
6260 * and let the allocator find whatever block
6261 * it can find. If we reach this point, we
6262 * will have tried the cluster allocator
6263 * plenty of times and not have found
6264 * anything, so we are likely way too
6265 * fragmented for the clustering stuff to find
Alexandre Olivaa5f6f712011-12-12 04:48:19 -02006266 * anything.
6267 *
6268 * However, if the cluster is taken from the
6269 * current block group, release the cluster
6270 * first, so that we stand a better chance of
6271 * succeeding in the unclustered
6272 * allocation. */
6273 if (loop >= LOOP_NO_EMPTY_SIZE &&
6274 last_ptr->block_group != block_group) {
Alexandre Oliva062c05c2011-12-07 19:50:42 -05006275 spin_unlock(&last_ptr->refill_lock);
6276 goto unclustered_alloc;
6277 }
6278
Chris Masonfa9c0d792009-04-03 09:47:43 -04006279 /*
6280 * this cluster didn't work out, free it and
6281 * start over
6282 */
6283 btrfs_return_cluster_to_free_space(NULL, last_ptr);
6284
Alexandre Olivaa5f6f712011-12-12 04:48:19 -02006285 if (loop >= LOOP_NO_EMPTY_SIZE) {
6286 spin_unlock(&last_ptr->refill_lock);
6287 goto unclustered_alloc;
6288 }
6289
Chris Mason8de972b2013-01-04 15:39:43 -05006290 aligned_cluster = max_t(unsigned long,
6291 empty_cluster + empty_size,
6292 block_group->full_stripe_len);
6293
Chris Masonfa9c0d792009-04-03 09:47:43 -04006294 /* allocate a cluster in this block group */
Chris Mason451d7582009-06-09 20:28:34 -04006295 ret = btrfs_find_space_cluster(trans, root,
Chris Masonfa9c0d792009-04-03 09:47:43 -04006296 block_group, last_ptr,
Alexandre Oliva1b22bad2011-11-30 13:43:00 -05006297 search_start, num_bytes,
Chris Mason8de972b2013-01-04 15:39:43 -05006298 aligned_cluster);
Chris Masonfa9c0d792009-04-03 09:47:43 -04006299 if (ret == 0) {
6300 /*
6301 * now pull our allocation out of this
6302 * cluster
6303 */
6304 offset = btrfs_alloc_from_cluster(block_group,
6305 last_ptr, num_bytes,
6306 search_start);
6307 if (offset) {
6308 /* we found one, proceed */
6309 spin_unlock(&last_ptr->refill_lock);
Josef Bacik3f7de032011-11-10 08:29:20 -05006310 trace_btrfs_reserve_extent_cluster(root,
6311 block_group, search_start,
6312 num_bytes);
Chris Masonfa9c0d792009-04-03 09:47:43 -04006313 goto checks;
6314 }
Josef Bacik0a243252009-09-11 16:11:20 -04006315 } else if (!cached && loop > LOOP_CACHING_NOWAIT
6316 && !failed_cluster_refill) {
Josef Bacik817d52f2009-07-13 21:29:25 -04006317 spin_unlock(&last_ptr->refill_lock);
6318
Josef Bacik0a243252009-09-11 16:11:20 -04006319 failed_cluster_refill = true;
Josef Bacik817d52f2009-07-13 21:29:25 -04006320 wait_block_group_cache_progress(block_group,
6321 num_bytes + empty_cluster + empty_size);
6322 goto have_block_group;
Chris Masonfa9c0d792009-04-03 09:47:43 -04006323 }
Josef Bacik817d52f2009-07-13 21:29:25 -04006324
Chris Masonfa9c0d792009-04-03 09:47:43 -04006325 /*
6326 * at this point we either didn't find a cluster
6327 * or we weren't able to allocate a block from our
6328 * cluster. Free the cluster we've been trying
6329 * to use, and go to the next block group
6330 */
Josef Bacik0a243252009-09-11 16:11:20 -04006331 btrfs_return_cluster_to_free_space(NULL, last_ptr);
Chris Masonfa9c0d792009-04-03 09:47:43 -04006332 spin_unlock(&last_ptr->refill_lock);
Josef Bacik0a243252009-09-11 16:11:20 -04006333 goto loop;
Chris Masonfa9c0d792009-04-03 09:47:43 -04006334 }
6335
Alexandre Oliva062c05c2011-12-07 19:50:42 -05006336unclustered_alloc:
Alexandre Olivaa5f6f712011-12-12 04:48:19 -02006337 spin_lock(&block_group->free_space_ctl->tree_lock);
6338 if (cached &&
6339 block_group->free_space_ctl->free_space <
6340 num_bytes + empty_cluster + empty_size) {
6341 spin_unlock(&block_group->free_space_ctl->tree_lock);
6342 goto loop;
6343 }
6344 spin_unlock(&block_group->free_space_ctl->tree_lock);
6345
Josef Bacik6226cb0a2009-04-03 10:14:18 -04006346 offset = btrfs_find_space_for_alloc(block_group, search_start,
6347 num_bytes, empty_size);
Josef Bacik1cdda9b2009-10-06 10:04:28 -04006348 /*
6349 * If we didn't find a chunk, and we haven't failed on this
6350 * block group before, and this block group is in the middle of
6351 * caching and we are ok with waiting, then go ahead and wait
6352 * for progress to be made, and set failed_alloc to true.
6353 *
6354 * If failed_alloc is true then we've already waited on this
6355 * block group once and should move on to the next block group.
6356 */
6357 if (!offset && !failed_alloc && !cached &&
6358 loop > LOOP_CACHING_NOWAIT) {
Josef Bacik817d52f2009-07-13 21:29:25 -04006359 wait_block_group_cache_progress(block_group,
Josef Bacik1cdda9b2009-10-06 10:04:28 -04006360 num_bytes + empty_size);
6361 failed_alloc = true;
Josef Bacik817d52f2009-07-13 21:29:25 -04006362 goto have_block_group;
Josef Bacik1cdda9b2009-10-06 10:04:28 -04006363 } else if (!offset) {
Miao Xie60d2adb2011-09-09 17:34:35 +08006364 if (!cached)
6365 have_caching_bg = true;
Josef Bacik1cdda9b2009-10-06 10:04:28 -04006366 goto loop;
Josef Bacik817d52f2009-07-13 21:29:25 -04006367 }
Chris Masonfa9c0d792009-04-03 09:47:43 -04006368checks:
David Woodhouse53b381b2013-01-29 18:40:14 -05006369 search_start = stripe_align(root, used_block_group,
6370 offset, num_bytes);
Chris Masone37c9e62007-05-09 20:13:14 -04006371
Josef Bacik2552d172009-04-03 10:14:19 -04006372 /* move on to the next group */
6373 if (search_start + num_bytes >
Alexandre Oliva274bd4f2011-12-07 20:08:40 -05006374 used_block_group->key.objectid + used_block_group->key.offset) {
6375 btrfs_add_free_space(used_block_group, offset, num_bytes);
Josef Bacik2552d172009-04-03 10:14:19 -04006376 goto loop;
Josef Bacik6226cb0a2009-04-03 10:14:18 -04006377 }
Josef Bacik80eb2342008-10-29 14:49:05 -04006378
Josef Bacik6226cb0a2009-04-03 10:14:18 -04006379 if (offset < search_start)
Alexandre Oliva274bd4f2011-12-07 20:08:40 -05006380 btrfs_add_free_space(used_block_group, offset,
Josef Bacik6226cb0a2009-04-03 10:14:18 -04006381 search_start - offset);
6382 BUG_ON(offset > search_start);
6383
Alexandre Oliva274bd4f2011-12-07 20:08:40 -05006384 ret = btrfs_update_reserved_bytes(used_block_group, num_bytes,
Josef Bacikfb25e912011-07-26 17:00:46 -04006385 alloc_type);
Yan, Zhengf0486c62010-05-16 10:46:25 -04006386 if (ret == -EAGAIN) {
Alexandre Oliva274bd4f2011-12-07 20:08:40 -05006387 btrfs_add_free_space(used_block_group, offset, num_bytes);
Yan, Zhengf0486c62010-05-16 10:46:25 -04006388 goto loop;
6389 }
Yan Zheng11833d62009-09-11 16:11:19 -04006390
Josef Bacik2552d172009-04-03 10:14:19 -04006391 /* we are all good, lets return */
Yan, Zhengf0486c62010-05-16 10:46:25 -04006392 ins->objectid = search_start;
6393 ins->offset = num_bytes;
6394
Josef Bacik3f7de032011-11-10 08:29:20 -05006395 trace_btrfs_reserve_extent(orig_root, block_group,
6396 search_start, num_bytes);
Alexandre Oliva274bd4f2011-12-07 20:08:40 -05006397 if (used_block_group != block_group)
6398 btrfs_put_block_group(used_block_group);
Josef Bacikd82a6f12011-05-11 15:26:06 -04006399 btrfs_put_block_group(block_group);
Josef Bacik2552d172009-04-03 10:14:19 -04006400 break;
6401loop:
Josef Bacik0a243252009-09-11 16:11:20 -04006402 failed_cluster_refill = false;
Josef Bacik1cdda9b2009-10-06 10:04:28 -04006403 failed_alloc = false;
Yan, Zhengb742bb82010-05-16 10:46:24 -04006404 BUG_ON(index != get_block_group_index(block_group));
Alexandre Oliva274bd4f2011-12-07 20:08:40 -05006405 if (used_block_group != block_group)
6406 btrfs_put_block_group(used_block_group);
Chris Masonfa9c0d792009-04-03 09:47:43 -04006407 btrfs_put_block_group(block_group);
Josef Bacik2552d172009-04-03 10:14:19 -04006408 }
6409 up_read(&space_info->groups_sem);
Chris Masonf5a31e12008-11-10 11:47:09 -05006410
Miao Xie60d2adb2011-09-09 17:34:35 +08006411 if (!ins->objectid && loop >= LOOP_CACHING_WAIT && have_caching_bg)
6412 goto search;
6413
Yan, Zhengb742bb82010-05-16 10:46:24 -04006414 if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
6415 goto search;
6416
Josef Bacik285ff5a2012-01-13 15:27:45 -05006417 /*
Josef Bacikccf0e722009-11-10 21:23:48 -05006418 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
6419 * caching kthreads as we move along
Josef Bacik817d52f2009-07-13 21:29:25 -04006420 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
6421 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
6422 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
6423 * again
Chris Masonfa9c0d792009-04-03 09:47:43 -04006424 */
Josef Bacik723bda22011-05-27 16:11:38 -04006425 if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) {
Yan, Zhengb742bb82010-05-16 10:46:24 -04006426 index = 0;
Josef Bacik723bda22011-05-27 16:11:38 -04006427 loop++;
Josef Bacik817d52f2009-07-13 21:29:25 -04006428 if (loop == LOOP_ALLOC_CHUNK) {
David Sterbab6919a52013-04-29 13:39:40 +00006429 ret = do_chunk_alloc(trans, root, flags,
Josef Bacikea658ba2012-09-11 16:57:25 -04006430 CHUNK_ALLOC_FORCE);
6431 /*
6432 * Do not bail out on ENOSPC since we
6433 * can do more things.
6434 */
6435 if (ret < 0 && ret != -ENOSPC) {
6436 btrfs_abort_transaction(trans,
6437 root, ret);
6438 goto out;
Josef Bacik723bda22011-05-27 16:11:38 -04006439 }
Josef Bacik723bda22011-05-27 16:11:38 -04006440 }
6441
6442 if (loop == LOOP_NO_EMPTY_SIZE) {
Chris Masonfa9c0d792009-04-03 09:47:43 -04006443 empty_size = 0;
6444 empty_cluster = 0;
6445 }
Chris Mason42e70e72008-11-07 18:17:11 -05006446
Josef Bacik723bda22011-05-27 16:11:38 -04006447 goto search;
Josef Bacik2552d172009-04-03 10:14:19 -04006448 } else if (!ins->objectid) {
6449 ret = -ENOSPC;
Josef Bacikd82a6f12011-05-11 15:26:06 -04006450 } else if (ins->objectid) {
Josef Bacik2552d172009-04-03 10:14:19 -04006451 ret = 0;
6452 }
Jeff Mahoney79787ea2012-03-12 16:03:00 +01006453out:
Chris Mason0b86a832008-03-24 15:01:56 -04006454
Chris Mason0f70abe2007-02-28 16:46:22 -05006455 return ret;
Chris Masonfec577f2007-02-26 10:40:21 -05006456}
Chris Masonec44a352008-04-28 15:29:52 -04006457
Josef Bacik9ed74f22009-09-11 16:12:44 -04006458static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
6459 int dump_block_groups)
Josef Bacik0f9dd462008-09-23 13:14:11 -04006460{
6461 struct btrfs_block_group_cache *cache;
Yan, Zhengb742bb82010-05-16 10:46:24 -04006462 int index = 0;
Josef Bacik0f9dd462008-09-23 13:14:11 -04006463
Josef Bacik9ed74f22009-09-11 16:12:44 -04006464 spin_lock(&info->lock);
Josef Bacikfb25e912011-07-26 17:00:46 -04006465 printk(KERN_INFO "space_info %llu has %llu free, is %sfull\n",
6466 (unsigned long long)info->flags,
Chris Masond3977122009-01-05 21:25:51 -05006467 (unsigned long long)(info->total_bytes - info->bytes_used -
Josef Bacik9ed74f22009-09-11 16:12:44 -04006468 info->bytes_pinned - info->bytes_reserved -
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04006469 info->bytes_readonly),
Chris Masond3977122009-01-05 21:25:51 -05006470 (info->full) ? "" : "not ");
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04006471 printk(KERN_INFO "space_info total=%llu, used=%llu, pinned=%llu, "
6472 "reserved=%llu, may_use=%llu, readonly=%llu\n",
Joel Becker21380932009-04-21 12:38:29 -07006473 (unsigned long long)info->total_bytes,
Josef Bacik9ed74f22009-09-11 16:12:44 -04006474 (unsigned long long)info->bytes_used,
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04006475 (unsigned long long)info->bytes_pinned,
6476 (unsigned long long)info->bytes_reserved,
6477 (unsigned long long)info->bytes_may_use,
6478 (unsigned long long)info->bytes_readonly);
Josef Bacik9ed74f22009-09-11 16:12:44 -04006479 spin_unlock(&info->lock);
6480
6481 if (!dump_block_groups)
6482 return;
Josef Bacik0f9dd462008-09-23 13:14:11 -04006483
Josef Bacik80eb2342008-10-29 14:49:05 -04006484 down_read(&info->groups_sem);
Yan, Zhengb742bb82010-05-16 10:46:24 -04006485again:
6486 list_for_each_entry(cache, &info->block_groups[index], list) {
Josef Bacik0f9dd462008-09-23 13:14:11 -04006487 spin_lock(&cache->lock);
Liu Bo799ffc32012-07-06 03:31:35 -06006488 printk(KERN_INFO "block group %llu has %llu bytes, %llu used %llu pinned %llu reserved %s\n",
Chris Masond3977122009-01-05 21:25:51 -05006489 (unsigned long long)cache->key.objectid,
6490 (unsigned long long)cache->key.offset,
6491 (unsigned long long)btrfs_block_group_used(&cache->item),
6492 (unsigned long long)cache->pinned,
Liu Bo799ffc32012-07-06 03:31:35 -06006493 (unsigned long long)cache->reserved,
6494 cache->ro ? "[readonly]" : "");
Josef Bacik0f9dd462008-09-23 13:14:11 -04006495 btrfs_dump_free_space(cache, bytes);
6496 spin_unlock(&cache->lock);
6497 }
Yan, Zhengb742bb82010-05-16 10:46:24 -04006498 if (++index < BTRFS_NR_RAID_TYPES)
6499 goto again;
Josef Bacik80eb2342008-10-29 14:49:05 -04006500 up_read(&info->groups_sem);
Josef Bacik0f9dd462008-09-23 13:14:11 -04006501}
Zheng Yane8569812008-09-26 10:05:48 -04006502
Yan Zheng11833d62009-09-11 16:11:19 -04006503int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
6504 struct btrfs_root *root,
6505 u64 num_bytes, u64 min_alloc_size,
6506 u64 empty_size, u64 hint_byte,
David Sterbab6919a52013-04-29 13:39:40 +00006507 struct btrfs_key *ins, int is_data)
Chris Masonfec577f2007-02-26 10:40:21 -05006508{
Miao Xie9e622d62012-01-26 15:01:12 -05006509 bool final_tried = false;
David Sterbab6919a52013-04-29 13:39:40 +00006510 u64 flags;
Chris Masonfec577f2007-02-26 10:40:21 -05006511 int ret;
Chris Mason925baed2008-06-25 16:01:30 -04006512
David Sterbab6919a52013-04-29 13:39:40 +00006513 flags = btrfs_get_alloc_profile(root, is_data);
Chris Mason98d20f62008-04-14 09:46:10 -04006514again:
Chris Masondb945352007-10-15 16:15:53 -04006515 WARN_ON(num_bytes < root->sectorsize);
6516 ret = find_free_extent(trans, root, num_bytes, empty_size,
David Sterbab6919a52013-04-29 13:39:40 +00006517 hint_byte, ins, flags);
Chris Mason3b951512008-04-17 11:29:12 -04006518
Miao Xie9e622d62012-01-26 15:01:12 -05006519 if (ret == -ENOSPC) {
6520 if (!final_tried) {
6521 num_bytes = num_bytes >> 1;
Zach Brown24542bf2012-11-16 00:04:43 +00006522 num_bytes = round_down(num_bytes, root->sectorsize);
Miao Xie9e622d62012-01-26 15:01:12 -05006523 num_bytes = max(num_bytes, min_alloc_size);
Miao Xie9e622d62012-01-26 15:01:12 -05006524 if (num_bytes == min_alloc_size)
6525 final_tried = true;
6526 goto again;
6527 } else if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
6528 struct btrfs_space_info *sinfo;
Josef Bacik0f9dd462008-09-23 13:14:11 -04006529
David Sterbab6919a52013-04-29 13:39:40 +00006530 sinfo = __find_space_info(root->fs_info, flags);
Simon Kirbyc2cf52e2013-03-19 22:41:23 +00006531 btrfs_err(root->fs_info, "allocation failed flags %llu, wanted %llu",
David Sterbab6919a52013-04-29 13:39:40 +00006532 (unsigned long long)flags,
Simon Kirbyc2cf52e2013-03-19 22:41:23 +00006533 (unsigned long long)num_bytes);
Jeff Mahoney53804282012-03-01 14:56:28 +01006534 if (sinfo)
6535 dump_space_info(sinfo, num_bytes, 1);
Miao Xie9e622d62012-01-26 15:01:12 -05006536 }
Chris Mason925baed2008-06-25 16:01:30 -04006537 }
Josef Bacik0f9dd462008-09-23 13:14:11 -04006538
liubo1abe9b82011-03-24 11:18:59 +00006539 trace_btrfs_reserved_extent_alloc(root, ins->objectid, ins->offset);
6540
Josef Bacik0f9dd462008-09-23 13:14:11 -04006541 return ret;
Chris Masone6dcd2d2008-07-17 12:53:50 -04006542}
6543
Chris Masone688b7252011-10-31 20:52:39 -04006544static int __btrfs_free_reserved_extent(struct btrfs_root *root,
6545 u64 start, u64 len, int pin)
Chris Mason65b51a02008-08-01 15:11:20 -04006546{
Josef Bacik0f9dd462008-09-23 13:14:11 -04006547 struct btrfs_block_group_cache *cache;
Liu Hui1f3c79a2009-01-05 15:57:51 -05006548 int ret = 0;
Josef Bacik0f9dd462008-09-23 13:14:11 -04006549
Josef Bacik0f9dd462008-09-23 13:14:11 -04006550 cache = btrfs_lookup_block_group(root->fs_info, start);
6551 if (!cache) {
Simon Kirbyc2cf52e2013-03-19 22:41:23 +00006552 btrfs_err(root->fs_info, "Unable to find block group for %llu",
6553 (unsigned long long)start);
Josef Bacik0f9dd462008-09-23 13:14:11 -04006554 return -ENOSPC;
6555 }
Liu Hui1f3c79a2009-01-05 15:57:51 -05006556
Li Dongyang5378e602011-03-24 10:24:27 +00006557 if (btrfs_test_opt(root, DISCARD))
6558 ret = btrfs_discard_extent(root, start, len, NULL);
Liu Hui1f3c79a2009-01-05 15:57:51 -05006559
Chris Masone688b7252011-10-31 20:52:39 -04006560 if (pin)
6561 pin_down_extent(root, cache, start, len, 1);
6562 else {
6563 btrfs_add_free_space(cache, start, len);
6564 btrfs_update_reserved_bytes(cache, len, RESERVE_FREE);
6565 }
Chris Masonfa9c0d792009-04-03 09:47:43 -04006566 btrfs_put_block_group(cache);
Josef Bacik817d52f2009-07-13 21:29:25 -04006567
liubo1abe9b82011-03-24 11:18:59 +00006568 trace_btrfs_reserved_extent_free(root, start, len);
6569
Chris Masone6dcd2d2008-07-17 12:53:50 -04006570 return ret;
6571}
6572
Chris Masone688b7252011-10-31 20:52:39 -04006573int btrfs_free_reserved_extent(struct btrfs_root *root,
6574 u64 start, u64 len)
6575{
6576 return __btrfs_free_reserved_extent(root, start, len, 0);
6577}
6578
6579int btrfs_free_and_pin_reserved_extent(struct btrfs_root *root,
6580 u64 start, u64 len)
6581{
6582 return __btrfs_free_reserved_extent(root, start, len, 1);
6583}
6584
Yan Zheng5d4f98a2009-06-10 10:45:14 -04006585static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
6586 struct btrfs_root *root,
6587 u64 parent, u64 root_objectid,
6588 u64 flags, u64 owner, u64 offset,
6589 struct btrfs_key *ins, int ref_mod)
Chris Masone6dcd2d2008-07-17 12:53:50 -04006590{
6591 int ret;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04006592 struct btrfs_fs_info *fs_info = root->fs_info;
Chris Masone6dcd2d2008-07-17 12:53:50 -04006593 struct btrfs_extent_item *extent_item;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04006594 struct btrfs_extent_inline_ref *iref;
Chris Masone6dcd2d2008-07-17 12:53:50 -04006595 struct btrfs_path *path;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04006596 struct extent_buffer *leaf;
6597 int type;
6598 u32 size;
Chris Masonf2654de2007-06-26 12:20:46 -04006599
Yan Zheng5d4f98a2009-06-10 10:45:14 -04006600 if (parent > 0)
6601 type = BTRFS_SHARED_DATA_REF_KEY;
6602 else
6603 type = BTRFS_EXTENT_DATA_REF_KEY;
Zheng Yan31840ae2008-09-23 13:14:14 -04006604
Yan Zheng5d4f98a2009-06-10 10:45:14 -04006605 size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
Chris Mason7bb86312007-12-11 09:25:06 -05006606
6607 path = btrfs_alloc_path();
Tsutomu Itohdb5b4932011-03-23 08:14:16 +00006608 if (!path)
6609 return -ENOMEM;
Chris Mason47e4bb92008-02-01 14:51:59 -05006610
Chris Masonb9473432009-03-13 11:00:37 -04006611 path->leave_spinning = 1;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04006612 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
6613 ins, size);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01006614 if (ret) {
6615 btrfs_free_path(path);
6616 return ret;
6617 }
Josef Bacik0f9dd462008-09-23 13:14:11 -04006618
Yan Zheng5d4f98a2009-06-10 10:45:14 -04006619 leaf = path->nodes[0];
6620 extent_item = btrfs_item_ptr(leaf, path->slots[0],
Chris Mason47e4bb92008-02-01 14:51:59 -05006621 struct btrfs_extent_item);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04006622 btrfs_set_extent_refs(leaf, extent_item, ref_mod);
6623 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
6624 btrfs_set_extent_flags(leaf, extent_item,
6625 flags | BTRFS_EXTENT_FLAG_DATA);
Chris Mason47e4bb92008-02-01 14:51:59 -05006626
Yan Zheng5d4f98a2009-06-10 10:45:14 -04006627 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
6628 btrfs_set_extent_inline_ref_type(leaf, iref, type);
6629 if (parent > 0) {
6630 struct btrfs_shared_data_ref *ref;
6631 ref = (struct btrfs_shared_data_ref *)(iref + 1);
6632 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
6633 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
6634 } else {
6635 struct btrfs_extent_data_ref *ref;
6636 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
6637 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
6638 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
6639 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
6640 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
6641 }
Chris Mason47e4bb92008-02-01 14:51:59 -05006642
6643 btrfs_mark_buffer_dirty(path->nodes[0]);
Chris Mason7bb86312007-12-11 09:25:06 -05006644 btrfs_free_path(path);
Chris Masonf510cfe2007-10-15 16:14:48 -04006645
Liu Boc53d6132012-12-27 09:01:19 +00006646 ret = update_block_group(root, ins->objectid, ins->offset, 1);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01006647 if (ret) { /* -ENOENT, logic error */
Simon Kirbyc2cf52e2013-03-19 22:41:23 +00006648 btrfs_err(fs_info, "update block group failed for %llu %llu",
6649 (unsigned long long)ins->objectid,
6650 (unsigned long long)ins->offset);
Chris Masonf5947062008-02-04 10:10:13 -05006651 BUG();
6652 }
Chris Masone6dcd2d2008-07-17 12:53:50 -04006653 return ret;
6654}
6655
Yan Zheng5d4f98a2009-06-10 10:45:14 -04006656static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
6657 struct btrfs_root *root,
6658 u64 parent, u64 root_objectid,
6659 u64 flags, struct btrfs_disk_key *key,
6660 int level, struct btrfs_key *ins)
6661{
6662 int ret;
6663 struct btrfs_fs_info *fs_info = root->fs_info;
6664 struct btrfs_extent_item *extent_item;
6665 struct btrfs_tree_block_info *block_info;
6666 struct btrfs_extent_inline_ref *iref;
6667 struct btrfs_path *path;
6668 struct extent_buffer *leaf;
Josef Bacik3173a182013-03-07 14:22:04 -05006669 u32 size = sizeof(*extent_item) + sizeof(*iref);
6670 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
6671 SKINNY_METADATA);
6672
6673 if (!skinny_metadata)
6674 size += sizeof(*block_info);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04006675
6676 path = btrfs_alloc_path();
Mark Fashehd8926bb2011-07-13 10:38:47 -07006677 if (!path)
6678 return -ENOMEM;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04006679
6680 path->leave_spinning = 1;
6681 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
6682 ins, size);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01006683 if (ret) {
6684 btrfs_free_path(path);
6685 return ret;
6686 }
Yan Zheng5d4f98a2009-06-10 10:45:14 -04006687
6688 leaf = path->nodes[0];
6689 extent_item = btrfs_item_ptr(leaf, path->slots[0],
6690 struct btrfs_extent_item);
6691 btrfs_set_extent_refs(leaf, extent_item, 1);
6692 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
6693 btrfs_set_extent_flags(leaf, extent_item,
6694 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04006695
Josef Bacik3173a182013-03-07 14:22:04 -05006696 if (skinny_metadata) {
6697 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
6698 } else {
6699 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
6700 btrfs_set_tree_block_key(leaf, block_info, key);
6701 btrfs_set_tree_block_level(leaf, block_info, level);
6702 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
6703 }
Yan Zheng5d4f98a2009-06-10 10:45:14 -04006704
Yan Zheng5d4f98a2009-06-10 10:45:14 -04006705 if (parent > 0) {
6706 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
6707 btrfs_set_extent_inline_ref_type(leaf, iref,
6708 BTRFS_SHARED_BLOCK_REF_KEY);
6709 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
6710 } else {
6711 btrfs_set_extent_inline_ref_type(leaf, iref,
6712 BTRFS_TREE_BLOCK_REF_KEY);
6713 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
6714 }
6715
6716 btrfs_mark_buffer_dirty(leaf);
6717 btrfs_free_path(path);
6718
Josef Bacik3173a182013-03-07 14:22:04 -05006719 ret = update_block_group(root, ins->objectid, root->leafsize, 1);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01006720 if (ret) { /* -ENOENT, logic error */
Simon Kirbyc2cf52e2013-03-19 22:41:23 +00006721 btrfs_err(fs_info, "update block group failed for %llu %llu",
6722 (unsigned long long)ins->objectid,
6723 (unsigned long long)ins->offset);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04006724 BUG();
6725 }
6726 return ret;
6727}
6728
6729int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
6730 struct btrfs_root *root,
6731 u64 root_objectid, u64 owner,
6732 u64 offset, struct btrfs_key *ins)
Chris Masone6dcd2d2008-07-17 12:53:50 -04006733{
6734 int ret;
Chris Mason1c2308f2008-09-23 13:14:13 -04006735
Yan Zheng5d4f98a2009-06-10 10:45:14 -04006736 BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
Chris Mason56bec292009-03-13 10:10:06 -04006737
Arne Jansen66d7e7f2011-09-12 15:26:38 +02006738 ret = btrfs_add_delayed_data_ref(root->fs_info, trans, ins->objectid,
6739 ins->offset, 0,
6740 root_objectid, owner, offset,
6741 BTRFS_ADD_DELAYED_EXTENT, NULL, 0);
Chris Masone6dcd2d2008-07-17 12:53:50 -04006742 return ret;
6743}
Chris Masone02119d2008-09-05 16:13:11 -04006744
6745/*
6746 * this is used by the tree logging recovery code. It records that
6747 * an extent has been allocated and makes sure to clear the free
6748 * space cache bits as well
6749 */
Yan Zheng5d4f98a2009-06-10 10:45:14 -04006750int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
6751 struct btrfs_root *root,
6752 u64 root_objectid, u64 owner, u64 offset,
6753 struct btrfs_key *ins)
Chris Masone02119d2008-09-05 16:13:11 -04006754{
6755 int ret;
6756 struct btrfs_block_group_cache *block_group;
Josef Bacik8c2a1a32013-06-06 13:19:32 -04006757
6758 /*
6759 * Mixed block groups will exclude before processing the log so we only
6760 * need to do the exlude dance if this fs isn't mixed.
6761 */
6762 if (!btrfs_fs_incompat(root->fs_info, MIXED_GROUPS)) {
6763 ret = __exclude_logged_extent(root, ins->objectid, ins->offset);
6764 if (ret)
6765 return ret;
6766 }
Chris Masone02119d2008-09-05 16:13:11 -04006767
Chris Masone02119d2008-09-05 16:13:11 -04006768 block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
Josef Bacik8c2a1a32013-06-06 13:19:32 -04006769 if (!block_group)
6770 return -EINVAL;
Yan Zheng11833d62009-09-11 16:11:19 -04006771
Josef Bacikfb25e912011-07-26 17:00:46 -04006772 ret = btrfs_update_reserved_bytes(block_group, ins->offset,
6773 RESERVE_ALLOC_NO_ACCOUNT);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01006774 BUG_ON(ret); /* logic error */
Yan Zheng5d4f98a2009-06-10 10:45:14 -04006775 ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
6776 0, owner, offset, ins, 1);
Josef Bacikb50c6e22013-04-25 15:55:30 -04006777 btrfs_put_block_group(block_group);
Chris Masone02119d2008-09-05 16:13:11 -04006778 return ret;
6779}
6780
Eric Sandeen48a3b632013-04-25 20:41:01 +00006781static struct extent_buffer *
6782btrfs_init_new_buffer(struct btrfs_trans_handle *trans, struct btrfs_root *root,
6783 u64 bytenr, u32 blocksize, int level)
Chris Mason65b51a02008-08-01 15:11:20 -04006784{
6785 struct extent_buffer *buf;
6786
6787 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
6788 if (!buf)
6789 return ERR_PTR(-ENOMEM);
6790 btrfs_set_header_generation(buf, trans->transid);
Chris Mason85d4e462011-07-26 16:11:19 -04006791 btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level);
Chris Mason65b51a02008-08-01 15:11:20 -04006792 btrfs_tree_lock(buf);
6793 clean_tree_block(trans, root, buf);
Josef Bacik3083ee22012-03-09 16:01:49 -05006794 clear_bit(EXTENT_BUFFER_STALE, &buf->bflags);
Chris Masonb4ce94d2009-02-04 09:25:08 -05006795
6796 btrfs_set_lock_blocking(buf);
Chris Mason65b51a02008-08-01 15:11:20 -04006797 btrfs_set_buffer_uptodate(buf);
Chris Masonb4ce94d2009-02-04 09:25:08 -05006798
Chris Masond0c803c2008-09-11 16:17:57 -04006799 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
Yan, Zheng8cef4e12009-11-12 09:33:26 +00006800 /*
6801 * we allow two log transactions at a time, use different
6802 * EXENT bit to differentiate dirty pages.
6803 */
6804 if (root->log_transid % 2 == 0)
6805 set_extent_dirty(&root->dirty_log_pages, buf->start,
6806 buf->start + buf->len - 1, GFP_NOFS);
6807 else
6808 set_extent_new(&root->dirty_log_pages, buf->start,
6809 buf->start + buf->len - 1, GFP_NOFS);
Chris Masond0c803c2008-09-11 16:17:57 -04006810 } else {
6811 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
6812 buf->start + buf->len - 1, GFP_NOFS);
6813 }
Chris Mason65b51a02008-08-01 15:11:20 -04006814 trans->blocks_used++;
Chris Masonb4ce94d2009-02-04 09:25:08 -05006815 /* this returns a buffer locked for blocking */
Chris Mason65b51a02008-08-01 15:11:20 -04006816 return buf;
6817}
6818
Yan, Zhengf0486c62010-05-16 10:46:25 -04006819static struct btrfs_block_rsv *
6820use_block_rsv(struct btrfs_trans_handle *trans,
6821 struct btrfs_root *root, u32 blocksize)
6822{
6823 struct btrfs_block_rsv *block_rsv;
Josef Bacik68a82272011-01-24 21:43:20 +00006824 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
Yan, Zhengf0486c62010-05-16 10:46:25 -04006825 int ret;
Miao Xied88033d2013-05-13 13:55:12 +00006826 bool global_updated = false;
Yan, Zhengf0486c62010-05-16 10:46:25 -04006827
6828 block_rsv = get_block_rsv(trans, root);
6829
Miao Xieb586b322013-05-13 13:55:10 +00006830 if (unlikely(block_rsv->size == 0))
6831 goto try_reserve;
Miao Xied88033d2013-05-13 13:55:12 +00006832again:
Yan, Zhengf0486c62010-05-16 10:46:25 -04006833 ret = block_rsv_use_bytes(block_rsv, blocksize);
6834 if (!ret)
6835 return block_rsv;
6836
Miao Xieb586b322013-05-13 13:55:10 +00006837 if (block_rsv->failfast)
6838 return ERR_PTR(ret);
6839
Miao Xied88033d2013-05-13 13:55:12 +00006840 if (block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL && !global_updated) {
6841 global_updated = true;
6842 update_global_block_rsv(root->fs_info);
6843 goto again;
6844 }
6845
Miao Xieb586b322013-05-13 13:55:10 +00006846 if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
6847 static DEFINE_RATELIMIT_STATE(_rs,
6848 DEFAULT_RATELIMIT_INTERVAL * 10,
6849 /*DEFAULT_RATELIMIT_BURST*/ 1);
6850 if (__ratelimit(&_rs))
6851 WARN(1, KERN_DEBUG
6852 "btrfs: block rsv returned %d\n", ret);
6853 }
6854try_reserve:
6855 ret = reserve_metadata_bytes(root, block_rsv, blocksize,
6856 BTRFS_RESERVE_NO_FLUSH);
6857 if (!ret)
6858 return block_rsv;
6859 /*
6860 * If we couldn't reserve metadata bytes try and use some from
Miao Xie5881cfc2013-05-13 13:55:11 +00006861 * the global reserve if its space type is the same as the global
6862 * reservation.
Miao Xieb586b322013-05-13 13:55:10 +00006863 */
Miao Xie5881cfc2013-05-13 13:55:11 +00006864 if (block_rsv->type != BTRFS_BLOCK_RSV_GLOBAL &&
6865 block_rsv->space_info == global_rsv->space_info) {
Miao Xieb586b322013-05-13 13:55:10 +00006866 ret = block_rsv_use_bytes(global_rsv, blocksize);
6867 if (!ret)
6868 return global_rsv;
6869 }
6870 return ERR_PTR(ret);
Yan, Zhengf0486c62010-05-16 10:46:25 -04006871}
6872
Josef Bacik8c2a3ca2012-01-10 10:31:31 -05006873static void unuse_block_rsv(struct btrfs_fs_info *fs_info,
6874 struct btrfs_block_rsv *block_rsv, u32 blocksize)
Yan, Zhengf0486c62010-05-16 10:46:25 -04006875{
6876 block_rsv_add_bytes(block_rsv, blocksize, 0);
Josef Bacik8c2a3ca2012-01-10 10:31:31 -05006877 block_rsv_release_bytes(fs_info, block_rsv, NULL, 0);
Yan, Zhengf0486c62010-05-16 10:46:25 -04006878}
6879
Chris Masonfec577f2007-02-26 10:40:21 -05006880/*
Yan, Zhengf0486c62010-05-16 10:46:25 -04006881 * finds a free extent and does all the dirty work required for allocation
6882 * returns the key for the extent through ins, and a tree buffer for
6883 * the first block of the extent through buf.
6884 *
Chris Masonfec577f2007-02-26 10:40:21 -05006885 * returns the tree buffer or NULL.
6886 */
Chris Mason5f39d392007-10-15 16:14:19 -04006887struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
Yan Zheng5d4f98a2009-06-10 10:45:14 -04006888 struct btrfs_root *root, u32 blocksize,
6889 u64 parent, u64 root_objectid,
6890 struct btrfs_disk_key *key, int level,
Jan Schmidt5581a512012-05-16 17:04:52 +02006891 u64 hint, u64 empty_size)
Chris Masonfec577f2007-02-26 10:40:21 -05006892{
Chris Masone2fa7222007-03-12 16:22:34 -04006893 struct btrfs_key ins;
Yan, Zhengf0486c62010-05-16 10:46:25 -04006894 struct btrfs_block_rsv *block_rsv;
Chris Mason5f39d392007-10-15 16:14:19 -04006895 struct extent_buffer *buf;
Yan, Zhengf0486c62010-05-16 10:46:25 -04006896 u64 flags = 0;
6897 int ret;
Josef Bacik3173a182013-03-07 14:22:04 -05006898 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
6899 SKINNY_METADATA);
Yan, Zhengf0486c62010-05-16 10:46:25 -04006900
6901 block_rsv = use_block_rsv(trans, root, blocksize);
6902 if (IS_ERR(block_rsv))
6903 return ERR_CAST(block_rsv);
6904
6905 ret = btrfs_reserve_extent(trans, root, blocksize, blocksize,
Josef Bacik81c9ad22012-01-18 10:56:06 -05006906 empty_size, hint, &ins, 0);
Chris Masonfec577f2007-02-26 10:40:21 -05006907 if (ret) {
Josef Bacik8c2a3ca2012-01-10 10:31:31 -05006908 unuse_block_rsv(root->fs_info, block_rsv, blocksize);
Chris Mason54aa1f42007-06-22 14:16:25 -04006909 return ERR_PTR(ret);
Chris Masonfec577f2007-02-26 10:40:21 -05006910 }
Chris Mason55c69072008-01-09 15:55:33 -05006911
Chris Mason4008c042009-02-12 14:09:45 -05006912 buf = btrfs_init_new_buffer(trans, root, ins.objectid,
6913 blocksize, level);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01006914 BUG_ON(IS_ERR(buf)); /* -ENOMEM */
Yan, Zhengf0486c62010-05-16 10:46:25 -04006915
6916 if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
6917 if (parent == 0)
6918 parent = ins.objectid;
6919 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
6920 } else
6921 BUG_ON(parent > 0);
6922
6923 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
6924 struct btrfs_delayed_extent_op *extent_op;
Miao Xie78a61842012-11-21 02:21:28 +00006925 extent_op = btrfs_alloc_delayed_extent_op();
Jeff Mahoney79787ea2012-03-12 16:03:00 +01006926 BUG_ON(!extent_op); /* -ENOMEM */
Yan, Zhengf0486c62010-05-16 10:46:25 -04006927 if (key)
6928 memcpy(&extent_op->key, key, sizeof(extent_op->key));
6929 else
6930 memset(&extent_op->key, 0, sizeof(extent_op->key));
6931 extent_op->flags_to_set = flags;
Josef Bacik3173a182013-03-07 14:22:04 -05006932 if (skinny_metadata)
6933 extent_op->update_key = 0;
6934 else
6935 extent_op->update_key = 1;
Yan, Zhengf0486c62010-05-16 10:46:25 -04006936 extent_op->update_flags = 1;
6937 extent_op->is_data = 0;
Josef Bacikb1c79e02013-05-09 13:49:30 -04006938 extent_op->level = level;
Yan, Zhengf0486c62010-05-16 10:46:25 -04006939
Arne Jansen66d7e7f2011-09-12 15:26:38 +02006940 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
6941 ins.objectid,
Yan, Zhengf0486c62010-05-16 10:46:25 -04006942 ins.offset, parent, root_objectid,
6943 level, BTRFS_ADD_DELAYED_EXTENT,
Jan Schmidt5581a512012-05-16 17:04:52 +02006944 extent_op, 0);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01006945 BUG_ON(ret); /* -ENOMEM */
Yan, Zhengf0486c62010-05-16 10:46:25 -04006946 }
Chris Masonfec577f2007-02-26 10:40:21 -05006947 return buf;
6948}
Chris Masona28ec192007-03-06 20:08:01 -05006949
Yan Zheng2c47e6052009-06-27 21:07:35 -04006950struct walk_control {
6951 u64 refs[BTRFS_MAX_LEVEL];
6952 u64 flags[BTRFS_MAX_LEVEL];
6953 struct btrfs_key update_progress;
6954 int stage;
6955 int level;
6956 int shared_level;
6957 int update_ref;
6958 int keep_locks;
Yan, Zheng1c4850e2009-09-21 15:55:59 -04006959 int reada_slot;
6960 int reada_count;
Arne Jansen66d7e7f2011-09-12 15:26:38 +02006961 int for_reloc;
Yan Zheng2c47e6052009-06-27 21:07:35 -04006962};
6963
6964#define DROP_REFERENCE 1
6965#define UPDATE_BACKREF 2
6966
Yan, Zheng1c4850e2009-09-21 15:55:59 -04006967static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
6968 struct btrfs_root *root,
6969 struct walk_control *wc,
6970 struct btrfs_path *path)
6971{
6972 u64 bytenr;
6973 u64 generation;
6974 u64 refs;
Yan, Zheng94fcca92009-10-09 09:25:16 -04006975 u64 flags;
Yan, Zheng1c4850e2009-09-21 15:55:59 -04006976 u32 nritems;
6977 u32 blocksize;
6978 struct btrfs_key key;
6979 struct extent_buffer *eb;
6980 int ret;
6981 int slot;
6982 int nread = 0;
6983
6984 if (path->slots[wc->level] < wc->reada_slot) {
6985 wc->reada_count = wc->reada_count * 2 / 3;
6986 wc->reada_count = max(wc->reada_count, 2);
6987 } else {
6988 wc->reada_count = wc->reada_count * 3 / 2;
6989 wc->reada_count = min_t(int, wc->reada_count,
6990 BTRFS_NODEPTRS_PER_BLOCK(root));
6991 }
6992
6993 eb = path->nodes[wc->level];
6994 nritems = btrfs_header_nritems(eb);
6995 blocksize = btrfs_level_size(root, wc->level - 1);
6996
6997 for (slot = path->slots[wc->level]; slot < nritems; slot++) {
6998 if (nread >= wc->reada_count)
6999 break;
7000
7001 cond_resched();
7002 bytenr = btrfs_node_blockptr(eb, slot);
7003 generation = btrfs_node_ptr_generation(eb, slot);
7004
7005 if (slot == path->slots[wc->level])
7006 goto reada;
7007
7008 if (wc->stage == UPDATE_BACKREF &&
7009 generation <= root->root_key.offset)
7010 continue;
7011
Yan, Zheng94fcca92009-10-09 09:25:16 -04007012 /* We don't lock the tree block, it's OK to be racy here */
Josef Bacik3173a182013-03-07 14:22:04 -05007013 ret = btrfs_lookup_extent_info(trans, root, bytenr,
7014 wc->level - 1, 1, &refs,
7015 &flags);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01007016 /* We don't care about errors in readahead. */
7017 if (ret < 0)
7018 continue;
Yan, Zheng94fcca92009-10-09 09:25:16 -04007019 BUG_ON(refs == 0);
7020
Yan, Zheng1c4850e2009-09-21 15:55:59 -04007021 if (wc->stage == DROP_REFERENCE) {
Yan, Zheng1c4850e2009-09-21 15:55:59 -04007022 if (refs == 1)
7023 goto reada;
7024
Yan, Zheng94fcca92009-10-09 09:25:16 -04007025 if (wc->level == 1 &&
7026 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7027 continue;
Yan, Zheng1c4850e2009-09-21 15:55:59 -04007028 if (!wc->update_ref ||
7029 generation <= root->root_key.offset)
7030 continue;
7031 btrfs_node_key_to_cpu(eb, &key, slot);
7032 ret = btrfs_comp_cpu_keys(&key,
7033 &wc->update_progress);
7034 if (ret < 0)
7035 continue;
Yan, Zheng94fcca92009-10-09 09:25:16 -04007036 } else {
7037 if (wc->level == 1 &&
7038 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7039 continue;
Yan, Zheng1c4850e2009-09-21 15:55:59 -04007040 }
7041reada:
7042 ret = readahead_tree_block(root, bytenr, blocksize,
7043 generation);
7044 if (ret)
7045 break;
Yan, Zheng1c4850e2009-09-21 15:55:59 -04007046 nread++;
7047 }
7048 wc->reada_slot = slot;
7049}
7050
Chris Mason9aca1d52007-03-13 11:09:37 -04007051/*
Liu Bo2c016dc2012-12-26 15:32:17 +08007052 * helper to process tree block while walking down the tree.
Yan Zheng2c47e6052009-06-27 21:07:35 -04007053 *
Yan Zheng2c47e6052009-06-27 21:07:35 -04007054 * when wc->stage == UPDATE_BACKREF, this function updates
7055 * back refs for pointers in the block.
7056 *
7057 * NOTE: return value 1 means we should stop walking down.
Yan Zhengf82d02d2008-10-29 14:49:05 -04007058 */
Yan Zheng2c47e6052009-06-27 21:07:35 -04007059static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
7060 struct btrfs_root *root,
7061 struct btrfs_path *path,
Yan, Zheng94fcca92009-10-09 09:25:16 -04007062 struct walk_control *wc, int lookup_info)
Yan Zheng2c47e6052009-06-27 21:07:35 -04007063{
7064 int level = wc->level;
7065 struct extent_buffer *eb = path->nodes[level];
Yan Zheng2c47e6052009-06-27 21:07:35 -04007066 u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
7067 int ret;
7068
7069 if (wc->stage == UPDATE_BACKREF &&
7070 btrfs_header_owner(eb) != root->root_key.objectid)
7071 return 1;
7072
7073 /*
7074 * when reference count of tree block is 1, it won't increase
7075 * again. once full backref flag is set, we never clear it.
7076 */
Yan, Zheng94fcca92009-10-09 09:25:16 -04007077 if (lookup_info &&
7078 ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
7079 (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
Yan Zheng2c47e6052009-06-27 21:07:35 -04007080 BUG_ON(!path->locks[level]);
7081 ret = btrfs_lookup_extent_info(trans, root,
Josef Bacik3173a182013-03-07 14:22:04 -05007082 eb->start, level, 1,
Yan Zheng2c47e6052009-06-27 21:07:35 -04007083 &wc->refs[level],
7084 &wc->flags[level]);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01007085 BUG_ON(ret == -ENOMEM);
7086 if (ret)
7087 return ret;
Yan Zheng2c47e6052009-06-27 21:07:35 -04007088 BUG_ON(wc->refs[level] == 0);
7089 }
7090
Yan Zheng2c47e6052009-06-27 21:07:35 -04007091 if (wc->stage == DROP_REFERENCE) {
7092 if (wc->refs[level] > 1)
7093 return 1;
7094
7095 if (path->locks[level] && !wc->keep_locks) {
Chris Masonbd681512011-07-16 15:23:14 -04007096 btrfs_tree_unlock_rw(eb, path->locks[level]);
Yan Zheng2c47e6052009-06-27 21:07:35 -04007097 path->locks[level] = 0;
7098 }
7099 return 0;
7100 }
7101
7102 /* wc->stage == UPDATE_BACKREF */
7103 if (!(wc->flags[level] & flag)) {
7104 BUG_ON(!path->locks[level]);
Arne Jansen66d7e7f2011-09-12 15:26:38 +02007105 ret = btrfs_inc_ref(trans, root, eb, 1, wc->for_reloc);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01007106 BUG_ON(ret); /* -ENOMEM */
Arne Jansen66d7e7f2011-09-12 15:26:38 +02007107 ret = btrfs_dec_ref(trans, root, eb, 0, wc->for_reloc);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01007108 BUG_ON(ret); /* -ENOMEM */
Yan Zheng2c47e6052009-06-27 21:07:35 -04007109 ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
Josef Bacikb1c79e02013-05-09 13:49:30 -04007110 eb->len, flag,
7111 btrfs_header_level(eb), 0);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01007112 BUG_ON(ret); /* -ENOMEM */
Yan Zheng2c47e6052009-06-27 21:07:35 -04007113 wc->flags[level] |= flag;
7114 }
7115
7116 /*
7117 * the block is shared by multiple trees, so it's not good to
7118 * keep the tree lock
7119 */
7120 if (path->locks[level] && level > 0) {
Chris Masonbd681512011-07-16 15:23:14 -04007121 btrfs_tree_unlock_rw(eb, path->locks[level]);
Yan Zheng2c47e6052009-06-27 21:07:35 -04007122 path->locks[level] = 0;
7123 }
7124 return 0;
7125}
7126
7127/*
Liu Bo2c016dc2012-12-26 15:32:17 +08007128 * helper to process tree block pointer.
Yan, Zheng1c4850e2009-09-21 15:55:59 -04007129 *
7130 * when wc->stage == DROP_REFERENCE, this function checks
7131 * reference count of the block pointed to. if the block
7132 * is shared and we need update back refs for the subtree
7133 * rooted at the block, this function changes wc->stage to
7134 * UPDATE_BACKREF. if the block is shared and there is no
7135 * need to update back, this function drops the reference
7136 * to the block.
7137 *
7138 * NOTE: return value 1 means we should stop walking down.
7139 */
7140static noinline int do_walk_down(struct btrfs_trans_handle *trans,
7141 struct btrfs_root *root,
7142 struct btrfs_path *path,
Yan, Zheng94fcca92009-10-09 09:25:16 -04007143 struct walk_control *wc, int *lookup_info)
Yan, Zheng1c4850e2009-09-21 15:55:59 -04007144{
7145 u64 bytenr;
7146 u64 generation;
7147 u64 parent;
7148 u32 blocksize;
7149 struct btrfs_key key;
7150 struct extent_buffer *next;
7151 int level = wc->level;
7152 int reada = 0;
7153 int ret = 0;
7154
7155 generation = btrfs_node_ptr_generation(path->nodes[level],
7156 path->slots[level]);
7157 /*
7158 * if the lower level block was created before the snapshot
7159 * was created, we know there is no need to update back refs
7160 * for the subtree
7161 */
7162 if (wc->stage == UPDATE_BACKREF &&
Yan, Zheng94fcca92009-10-09 09:25:16 -04007163 generation <= root->root_key.offset) {
7164 *lookup_info = 1;
Yan, Zheng1c4850e2009-09-21 15:55:59 -04007165 return 1;
Yan, Zheng94fcca92009-10-09 09:25:16 -04007166 }
Yan, Zheng1c4850e2009-09-21 15:55:59 -04007167
7168 bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
7169 blocksize = btrfs_level_size(root, level - 1);
7170
7171 next = btrfs_find_tree_block(root, bytenr, blocksize);
7172 if (!next) {
7173 next = btrfs_find_create_tree_block(root, bytenr, blocksize);
Miao Xie90d2c51d2010-03-25 12:37:12 +00007174 if (!next)
7175 return -ENOMEM;
Yan, Zheng1c4850e2009-09-21 15:55:59 -04007176 reada = 1;
7177 }
7178 btrfs_tree_lock(next);
7179 btrfs_set_lock_blocking(next);
7180
Josef Bacik3173a182013-03-07 14:22:04 -05007181 ret = btrfs_lookup_extent_info(trans, root, bytenr, level - 1, 1,
Yan, Zheng94fcca92009-10-09 09:25:16 -04007182 &wc->refs[level - 1],
7183 &wc->flags[level - 1]);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01007184 if (ret < 0) {
7185 btrfs_tree_unlock(next);
7186 return ret;
7187 }
7188
Simon Kirbyc2cf52e2013-03-19 22:41:23 +00007189 if (unlikely(wc->refs[level - 1] == 0)) {
7190 btrfs_err(root->fs_info, "Missing references.");
7191 BUG();
7192 }
Yan, Zheng94fcca92009-10-09 09:25:16 -04007193 *lookup_info = 0;
Yan, Zheng1c4850e2009-09-21 15:55:59 -04007194
Yan, Zheng94fcca92009-10-09 09:25:16 -04007195 if (wc->stage == DROP_REFERENCE) {
Yan, Zheng1c4850e2009-09-21 15:55:59 -04007196 if (wc->refs[level - 1] > 1) {
Yan, Zheng94fcca92009-10-09 09:25:16 -04007197 if (level == 1 &&
7198 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7199 goto skip;
7200
Yan, Zheng1c4850e2009-09-21 15:55:59 -04007201 if (!wc->update_ref ||
7202 generation <= root->root_key.offset)
7203 goto skip;
7204
7205 btrfs_node_key_to_cpu(path->nodes[level], &key,
7206 path->slots[level]);
7207 ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
7208 if (ret < 0)
7209 goto skip;
7210
7211 wc->stage = UPDATE_BACKREF;
7212 wc->shared_level = level - 1;
7213 }
Yan, Zheng94fcca92009-10-09 09:25:16 -04007214 } else {
7215 if (level == 1 &&
7216 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7217 goto skip;
Yan, Zheng1c4850e2009-09-21 15:55:59 -04007218 }
7219
Chris Masonb9fab912012-05-06 07:23:47 -04007220 if (!btrfs_buffer_uptodate(next, generation, 0)) {
Yan, Zheng1c4850e2009-09-21 15:55:59 -04007221 btrfs_tree_unlock(next);
7222 free_extent_buffer(next);
7223 next = NULL;
Yan, Zheng94fcca92009-10-09 09:25:16 -04007224 *lookup_info = 1;
Yan, Zheng1c4850e2009-09-21 15:55:59 -04007225 }
7226
7227 if (!next) {
7228 if (reada && level == 1)
7229 reada_walk_down(trans, root, wc, path);
7230 next = read_tree_block(root, bytenr, blocksize, generation);
Josef Bacik416bc652013-04-23 14:17:42 -04007231 if (!next || !extent_buffer_uptodate(next)) {
7232 free_extent_buffer(next);
Tsutomu Itoh97d9a8a2011-03-24 06:33:21 +00007233 return -EIO;
Josef Bacik416bc652013-04-23 14:17:42 -04007234 }
Yan, Zheng1c4850e2009-09-21 15:55:59 -04007235 btrfs_tree_lock(next);
7236 btrfs_set_lock_blocking(next);
7237 }
7238
7239 level--;
7240 BUG_ON(level != btrfs_header_level(next));
7241 path->nodes[level] = next;
7242 path->slots[level] = 0;
Chris Masonbd681512011-07-16 15:23:14 -04007243 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
Yan, Zheng1c4850e2009-09-21 15:55:59 -04007244 wc->level = level;
7245 if (wc->level == 1)
7246 wc->reada_slot = 0;
7247 return 0;
7248skip:
7249 wc->refs[level - 1] = 0;
7250 wc->flags[level - 1] = 0;
Yan, Zheng94fcca92009-10-09 09:25:16 -04007251 if (wc->stage == DROP_REFERENCE) {
7252 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
7253 parent = path->nodes[level]->start;
7254 } else {
7255 BUG_ON(root->root_key.objectid !=
7256 btrfs_header_owner(path->nodes[level]));
7257 parent = 0;
7258 }
Yan, Zheng1c4850e2009-09-21 15:55:59 -04007259
Yan, Zheng94fcca92009-10-09 09:25:16 -04007260 ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
Arne Jansen66d7e7f2011-09-12 15:26:38 +02007261 root->root_key.objectid, level - 1, 0, 0);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01007262 BUG_ON(ret); /* -ENOMEM */
Yan, Zheng1c4850e2009-09-21 15:55:59 -04007263 }
Yan, Zheng1c4850e2009-09-21 15:55:59 -04007264 btrfs_tree_unlock(next);
7265 free_extent_buffer(next);
Yan, Zheng94fcca92009-10-09 09:25:16 -04007266 *lookup_info = 1;
Yan, Zheng1c4850e2009-09-21 15:55:59 -04007267 return 1;
7268}
7269
7270/*
Liu Bo2c016dc2012-12-26 15:32:17 +08007271 * helper to process tree block while walking up the tree.
Yan Zheng2c47e6052009-06-27 21:07:35 -04007272 *
7273 * when wc->stage == DROP_REFERENCE, this function drops
7274 * reference count on the block.
7275 *
7276 * when wc->stage == UPDATE_BACKREF, this function changes
7277 * wc->stage back to DROP_REFERENCE if we changed wc->stage
7278 * to UPDATE_BACKREF previously while processing the block.
7279 *
7280 * NOTE: return value 1 means we should stop walking up.
7281 */
7282static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
7283 struct btrfs_root *root,
7284 struct btrfs_path *path,
7285 struct walk_control *wc)
7286{
Yan, Zhengf0486c62010-05-16 10:46:25 -04007287 int ret;
Yan Zheng2c47e6052009-06-27 21:07:35 -04007288 int level = wc->level;
7289 struct extent_buffer *eb = path->nodes[level];
7290 u64 parent = 0;
7291
7292 if (wc->stage == UPDATE_BACKREF) {
7293 BUG_ON(wc->shared_level < level);
7294 if (level < wc->shared_level)
7295 goto out;
7296
Yan Zheng2c47e6052009-06-27 21:07:35 -04007297 ret = find_next_key(path, level + 1, &wc->update_progress);
7298 if (ret > 0)
7299 wc->update_ref = 0;
7300
7301 wc->stage = DROP_REFERENCE;
7302 wc->shared_level = -1;
7303 path->slots[level] = 0;
7304
7305 /*
7306 * check reference count again if the block isn't locked.
7307 * we should start walking down the tree again if reference
7308 * count is one.
7309 */
7310 if (!path->locks[level]) {
7311 BUG_ON(level == 0);
7312 btrfs_tree_lock(eb);
7313 btrfs_set_lock_blocking(eb);
Chris Masonbd681512011-07-16 15:23:14 -04007314 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
Yan Zheng2c47e6052009-06-27 21:07:35 -04007315
7316 ret = btrfs_lookup_extent_info(trans, root,
Josef Bacik3173a182013-03-07 14:22:04 -05007317 eb->start, level, 1,
Yan Zheng2c47e6052009-06-27 21:07:35 -04007318 &wc->refs[level],
7319 &wc->flags[level]);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01007320 if (ret < 0) {
7321 btrfs_tree_unlock_rw(eb, path->locks[level]);
Liu Bo3268a242012-12-28 09:33:19 +00007322 path->locks[level] = 0;
Jeff Mahoney79787ea2012-03-12 16:03:00 +01007323 return ret;
7324 }
Yan Zheng2c47e6052009-06-27 21:07:35 -04007325 BUG_ON(wc->refs[level] == 0);
7326 if (wc->refs[level] == 1) {
Chris Masonbd681512011-07-16 15:23:14 -04007327 btrfs_tree_unlock_rw(eb, path->locks[level]);
Liu Bo3268a242012-12-28 09:33:19 +00007328 path->locks[level] = 0;
Yan Zheng2c47e6052009-06-27 21:07:35 -04007329 return 1;
7330 }
Yan Zheng2c47e6052009-06-27 21:07:35 -04007331 }
7332 }
7333
7334 /* wc->stage == DROP_REFERENCE */
7335 BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
7336
7337 if (wc->refs[level] == 1) {
7338 if (level == 0) {
7339 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
Arne Jansen66d7e7f2011-09-12 15:26:38 +02007340 ret = btrfs_dec_ref(trans, root, eb, 1,
7341 wc->for_reloc);
Yan Zheng2c47e6052009-06-27 21:07:35 -04007342 else
Arne Jansen66d7e7f2011-09-12 15:26:38 +02007343 ret = btrfs_dec_ref(trans, root, eb, 0,
7344 wc->for_reloc);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01007345 BUG_ON(ret); /* -ENOMEM */
Yan Zheng2c47e6052009-06-27 21:07:35 -04007346 }
7347 /* make block locked assertion in clean_tree_block happy */
7348 if (!path->locks[level] &&
7349 btrfs_header_generation(eb) == trans->transid) {
7350 btrfs_tree_lock(eb);
7351 btrfs_set_lock_blocking(eb);
Chris Masonbd681512011-07-16 15:23:14 -04007352 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
Yan Zheng2c47e6052009-06-27 21:07:35 -04007353 }
7354 clean_tree_block(trans, root, eb);
7355 }
7356
7357 if (eb == root->node) {
7358 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
7359 parent = eb->start;
7360 else
7361 BUG_ON(root->root_key.objectid !=
7362 btrfs_header_owner(eb));
7363 } else {
7364 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
7365 parent = path->nodes[level + 1]->start;
7366 else
7367 BUG_ON(root->root_key.objectid !=
7368 btrfs_header_owner(path->nodes[level + 1]));
7369 }
7370
Jan Schmidt5581a512012-05-16 17:04:52 +02007371 btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
Yan Zheng2c47e6052009-06-27 21:07:35 -04007372out:
7373 wc->refs[level] = 0;
7374 wc->flags[level] = 0;
Yan, Zhengf0486c62010-05-16 10:46:25 -04007375 return 0;
Yan Zheng2c47e6052009-06-27 21:07:35 -04007376}
7377
Yan Zheng5d4f98a2009-06-10 10:45:14 -04007378static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
7379 struct btrfs_root *root,
Yan Zheng2c47e6052009-06-27 21:07:35 -04007380 struct btrfs_path *path,
7381 struct walk_control *wc)
Yan Zhengf82d02d2008-10-29 14:49:05 -04007382{
Yan Zheng2c47e6052009-06-27 21:07:35 -04007383 int level = wc->level;
Yan, Zheng94fcca92009-10-09 09:25:16 -04007384 int lookup_info = 1;
Yan Zhengf82d02d2008-10-29 14:49:05 -04007385 int ret;
7386
Yan Zheng2c47e6052009-06-27 21:07:35 -04007387 while (level >= 0) {
Yan, Zheng94fcca92009-10-09 09:25:16 -04007388 ret = walk_down_proc(trans, root, path, wc, lookup_info);
Yan Zheng2c47e6052009-06-27 21:07:35 -04007389 if (ret > 0)
Yan Zhengf82d02d2008-10-29 14:49:05 -04007390 break;
Yan Zhengf82d02d2008-10-29 14:49:05 -04007391
Yan Zheng2c47e6052009-06-27 21:07:35 -04007392 if (level == 0)
7393 break;
7394
Yan, Zheng7a7965f2010-02-01 02:41:17 +00007395 if (path->slots[level] >=
7396 btrfs_header_nritems(path->nodes[level]))
7397 break;
7398
Yan, Zheng94fcca92009-10-09 09:25:16 -04007399 ret = do_walk_down(trans, root, path, wc, &lookup_info);
Yan, Zheng1c4850e2009-09-21 15:55:59 -04007400 if (ret > 0) {
7401 path->slots[level]++;
7402 continue;
Miao Xie90d2c51d2010-03-25 12:37:12 +00007403 } else if (ret < 0)
7404 return ret;
Yan, Zheng1c4850e2009-09-21 15:55:59 -04007405 level = wc->level;
Yan Zhengf82d02d2008-10-29 14:49:05 -04007406 }
Yan Zhengf82d02d2008-10-29 14:49:05 -04007407 return 0;
7408}
7409
Chris Masond3977122009-01-05 21:25:51 -05007410static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
Chris Mason98ed5172008-01-03 10:01:48 -05007411 struct btrfs_root *root,
Yan Zhengf82d02d2008-10-29 14:49:05 -04007412 struct btrfs_path *path,
Yan Zheng2c47e6052009-06-27 21:07:35 -04007413 struct walk_control *wc, int max_level)
Chris Mason20524f02007-03-10 06:35:47 -05007414{
Yan Zheng2c47e6052009-06-27 21:07:35 -04007415 int level = wc->level;
Chris Mason20524f02007-03-10 06:35:47 -05007416 int ret;
Chris Mason9f3a7422007-08-07 15:52:19 -04007417
Yan Zheng2c47e6052009-06-27 21:07:35 -04007418 path->slots[level] = btrfs_header_nritems(path->nodes[level]);
7419 while (level < max_level && path->nodes[level]) {
7420 wc->level = level;
7421 if (path->slots[level] + 1 <
7422 btrfs_header_nritems(path->nodes[level])) {
7423 path->slots[level]++;
Chris Mason20524f02007-03-10 06:35:47 -05007424 return 0;
7425 } else {
Yan Zheng2c47e6052009-06-27 21:07:35 -04007426 ret = walk_up_proc(trans, root, path, wc);
7427 if (ret > 0)
7428 return 0;
Chris Masonbd56b302009-02-04 09:27:02 -05007429
Yan Zheng2c47e6052009-06-27 21:07:35 -04007430 if (path->locks[level]) {
Chris Masonbd681512011-07-16 15:23:14 -04007431 btrfs_tree_unlock_rw(path->nodes[level],
7432 path->locks[level]);
Yan Zheng2c47e6052009-06-27 21:07:35 -04007433 path->locks[level] = 0;
Yan Zhengf82d02d2008-10-29 14:49:05 -04007434 }
Yan Zheng2c47e6052009-06-27 21:07:35 -04007435 free_extent_buffer(path->nodes[level]);
7436 path->nodes[level] = NULL;
7437 level++;
Chris Mason20524f02007-03-10 06:35:47 -05007438 }
7439 }
7440 return 1;
7441}
7442
Chris Mason9aca1d52007-03-13 11:09:37 -04007443/*
Yan Zheng2c47e6052009-06-27 21:07:35 -04007444 * drop a subvolume tree.
7445 *
7446 * this function traverses the tree freeing any blocks that only
7447 * referenced by the tree.
7448 *
7449 * when a shared tree block is found. this function decreases its
7450 * reference count by one. if update_ref is true, this function
7451 * also make sure backrefs for the shared block and all lower level
7452 * blocks are properly updated.
David Sterba9d1a2a32013-03-12 15:13:28 +00007453 *
7454 * If called with for_reloc == 0, may exit early with -EAGAIN
Chris Mason9aca1d52007-03-13 11:09:37 -04007455 */
Jeff Mahoney2c536792011-10-03 23:22:41 -04007456int btrfs_drop_snapshot(struct btrfs_root *root,
Arne Jansen66d7e7f2011-09-12 15:26:38 +02007457 struct btrfs_block_rsv *block_rsv, int update_ref,
7458 int for_reloc)
Chris Mason20524f02007-03-10 06:35:47 -05007459{
Chris Mason5caf2a02007-04-02 11:20:42 -04007460 struct btrfs_path *path;
Yan Zheng2c47e6052009-06-27 21:07:35 -04007461 struct btrfs_trans_handle *trans;
7462 struct btrfs_root *tree_root = root->fs_info->tree_root;
Chris Mason9f3a7422007-08-07 15:52:19 -04007463 struct btrfs_root_item *root_item = &root->root_item;
Yan Zheng2c47e6052009-06-27 21:07:35 -04007464 struct walk_control *wc;
7465 struct btrfs_key key;
7466 int err = 0;
7467 int ret;
7468 int level;
Josef Bacikd29a9f62013-07-17 19:30:20 -04007469 bool root_dropped = false;
Chris Mason20524f02007-03-10 06:35:47 -05007470
Chris Mason5caf2a02007-04-02 11:20:42 -04007471 path = btrfs_alloc_path();
Tsutomu Itohcb1b69f2011-08-09 07:11:13 +00007472 if (!path) {
7473 err = -ENOMEM;
7474 goto out;
7475 }
Chris Mason20524f02007-03-10 06:35:47 -05007476
Yan Zheng2c47e6052009-06-27 21:07:35 -04007477 wc = kzalloc(sizeof(*wc), GFP_NOFS);
Mark Fasheh38a1a912011-07-13 10:59:59 -07007478 if (!wc) {
7479 btrfs_free_path(path);
Tsutomu Itohcb1b69f2011-08-09 07:11:13 +00007480 err = -ENOMEM;
7481 goto out;
Mark Fasheh38a1a912011-07-13 10:59:59 -07007482 }
Yan Zheng2c47e6052009-06-27 21:07:35 -04007483
Yan, Zhenga22285a2010-05-16 10:48:46 -04007484 trans = btrfs_start_transaction(tree_root, 0);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01007485 if (IS_ERR(trans)) {
7486 err = PTR_ERR(trans);
7487 goto out_free;
7488 }
Tsutomu Itoh98d5dc12011-01-20 06:19:37 +00007489
Yan, Zheng3fd0a552010-05-16 10:49:59 -04007490 if (block_rsv)
7491 trans->block_rsv = block_rsv;
Yan Zheng2c47e6052009-06-27 21:07:35 -04007492
Chris Mason9f3a7422007-08-07 15:52:19 -04007493 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
Yan Zheng2c47e6052009-06-27 21:07:35 -04007494 level = btrfs_header_level(root->node);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04007495 path->nodes[level] = btrfs_lock_root_node(root);
7496 btrfs_set_lock_blocking(path->nodes[level]);
Chris Mason9f3a7422007-08-07 15:52:19 -04007497 path->slots[level] = 0;
Chris Masonbd681512011-07-16 15:23:14 -04007498 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
Yan Zheng2c47e6052009-06-27 21:07:35 -04007499 memset(&wc->update_progress, 0,
7500 sizeof(wc->update_progress));
Chris Mason9f3a7422007-08-07 15:52:19 -04007501 } else {
Chris Mason9f3a7422007-08-07 15:52:19 -04007502 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
Yan Zheng2c47e6052009-06-27 21:07:35 -04007503 memcpy(&wc->update_progress, &key,
7504 sizeof(wc->update_progress));
7505
Chris Mason6702ed42007-08-07 16:15:09 -04007506 level = root_item->drop_level;
Yan Zheng2c47e6052009-06-27 21:07:35 -04007507 BUG_ON(level == 0);
Chris Mason6702ed42007-08-07 16:15:09 -04007508 path->lowest_level = level;
Yan Zheng2c47e6052009-06-27 21:07:35 -04007509 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
7510 path->lowest_level = 0;
7511 if (ret < 0) {
7512 err = ret;
Jeff Mahoney79787ea2012-03-12 16:03:00 +01007513 goto out_end_trans;
Chris Mason9f3a7422007-08-07 15:52:19 -04007514 }
Yan, Zheng1c4850e2009-09-21 15:55:59 -04007515 WARN_ON(ret > 0);
Yan Zheng2c47e6052009-06-27 21:07:35 -04007516
Chris Mason7d9eb122008-07-08 14:19:17 -04007517 /*
7518 * unlock our path, this is safe because only this
7519 * function is allowed to delete this snapshot
7520 */
Yan Zheng5d4f98a2009-06-10 10:45:14 -04007521 btrfs_unlock_up_safe(path, 0);
Chris Mason9aca1d52007-03-13 11:09:37 -04007522
Yan Zheng2c47e6052009-06-27 21:07:35 -04007523 level = btrfs_header_level(root->node);
7524 while (1) {
7525 btrfs_tree_lock(path->nodes[level]);
7526 btrfs_set_lock_blocking(path->nodes[level]);
Josef Bacikfec386a2013-07-15 12:41:42 -04007527 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
Yan Zheng2c47e6052009-06-27 21:07:35 -04007528
7529 ret = btrfs_lookup_extent_info(trans, root,
7530 path->nodes[level]->start,
Josef Bacik3173a182013-03-07 14:22:04 -05007531 level, 1, &wc->refs[level],
Yan Zheng2c47e6052009-06-27 21:07:35 -04007532 &wc->flags[level]);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01007533 if (ret < 0) {
7534 err = ret;
7535 goto out_end_trans;
7536 }
Yan Zheng2c47e6052009-06-27 21:07:35 -04007537 BUG_ON(wc->refs[level] == 0);
7538
7539 if (level == root_item->drop_level)
7540 break;
7541
7542 btrfs_tree_unlock(path->nodes[level]);
Josef Bacikfec386a2013-07-15 12:41:42 -04007543 path->locks[level] = 0;
Yan Zheng2c47e6052009-06-27 21:07:35 -04007544 WARN_ON(wc->refs[level] != 1);
7545 level--;
7546 }
7547 }
7548
7549 wc->level = level;
7550 wc->shared_level = -1;
7551 wc->stage = DROP_REFERENCE;
7552 wc->update_ref = update_ref;
7553 wc->keep_locks = 0;
Arne Jansen66d7e7f2011-09-12 15:26:38 +02007554 wc->for_reloc = for_reloc;
Yan, Zheng1c4850e2009-09-21 15:55:59 -04007555 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
Yan Zheng2c47e6052009-06-27 21:07:35 -04007556
7557 while (1) {
David Sterba9d1a2a32013-03-12 15:13:28 +00007558
Yan Zheng2c47e6052009-06-27 21:07:35 -04007559 ret = walk_down_tree(trans, root, path, wc);
7560 if (ret < 0) {
7561 err = ret;
Chris Masone7a84562008-06-25 16:01:31 -04007562 break;
7563 }
Yan Zheng2c47e6052009-06-27 21:07:35 -04007564
7565 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
7566 if (ret < 0) {
7567 err = ret;
7568 break;
7569 }
7570
7571 if (ret > 0) {
7572 BUG_ON(wc->stage != DROP_REFERENCE);
7573 break;
7574 }
7575
7576 if (wc->stage == DROP_REFERENCE) {
7577 level = wc->level;
7578 btrfs_node_key(path->nodes[level],
7579 &root_item->drop_progress,
7580 path->slots[level]);
7581 root_item->drop_level = level;
7582 }
7583
7584 BUG_ON(wc->level == 0);
Josef Bacik3c8f2422013-07-15 11:57:06 -04007585 if (btrfs_should_end_transaction(trans, tree_root) ||
7586 (!for_reloc && btrfs_need_cleaner_sleep(root))) {
Yan Zheng2c47e6052009-06-27 21:07:35 -04007587 ret = btrfs_update_root(trans, tree_root,
7588 &root->root_key,
7589 root_item);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01007590 if (ret) {
7591 btrfs_abort_transaction(trans, tree_root, ret);
7592 err = ret;
7593 goto out_end_trans;
7594 }
Yan Zheng2c47e6052009-06-27 21:07:35 -04007595
Yan, Zheng3fd0a552010-05-16 10:49:59 -04007596 btrfs_end_transaction_throttle(trans, tree_root);
Josef Bacik3c8f2422013-07-15 11:57:06 -04007597 if (!for_reloc && btrfs_need_cleaner_sleep(root)) {
7598 pr_debug("btrfs: drop snapshot early exit\n");
7599 err = -EAGAIN;
7600 goto out_free;
7601 }
7602
Yan, Zhenga22285a2010-05-16 10:48:46 -04007603 trans = btrfs_start_transaction(tree_root, 0);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01007604 if (IS_ERR(trans)) {
7605 err = PTR_ERR(trans);
7606 goto out_free;
7607 }
Yan, Zheng3fd0a552010-05-16 10:49:59 -04007608 if (block_rsv)
7609 trans->block_rsv = block_rsv;
Chris Masonc3e69d52009-03-13 10:17:05 -04007610 }
Chris Mason20524f02007-03-10 06:35:47 -05007611 }
David Sterbab3b4aa72011-04-21 01:20:15 +02007612 btrfs_release_path(path);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01007613 if (err)
7614 goto out_end_trans;
Yan Zheng2c47e6052009-06-27 21:07:35 -04007615
7616 ret = btrfs_del_root(trans, tree_root, &root->root_key);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01007617 if (ret) {
7618 btrfs_abort_transaction(trans, tree_root, ret);
7619 goto out_end_trans;
7620 }
Yan Zheng2c47e6052009-06-27 21:07:35 -04007621
Yan, Zheng76dda932009-09-21 16:00:26 -04007622 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
Miao Xiecb517ea2013-05-15 07:48:19 +00007623 ret = btrfs_find_root(tree_root, &root->root_key, path,
7624 NULL, NULL);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01007625 if (ret < 0) {
7626 btrfs_abort_transaction(trans, tree_root, ret);
7627 err = ret;
7628 goto out_end_trans;
7629 } else if (ret > 0) {
Josef Bacik84cd9482010-12-08 12:24:01 -05007630 /* if we fail to delete the orphan item this time
7631 * around, it'll get picked up the next time.
7632 *
7633 * The most common failure here is just -ENOENT.
7634 */
7635 btrfs_del_orphan_item(trans, tree_root,
7636 root->root_key.objectid);
Yan, Zheng76dda932009-09-21 16:00:26 -04007637 }
7638 }
7639
7640 if (root->in_radix) {
Miao Xiecb517ea2013-05-15 07:48:19 +00007641 btrfs_drop_and_free_fs_root(tree_root->fs_info, root);
Yan, Zheng76dda932009-09-21 16:00:26 -04007642 } else {
7643 free_extent_buffer(root->node);
7644 free_extent_buffer(root->commit_root);
Miao Xieb0feb9d2013-05-15 07:48:20 +00007645 btrfs_put_fs_root(root);
Yan, Zheng76dda932009-09-21 16:00:26 -04007646 }
Josef Bacikd29a9f62013-07-17 19:30:20 -04007647 root_dropped = true;
Jeff Mahoney79787ea2012-03-12 16:03:00 +01007648out_end_trans:
Yan, Zheng3fd0a552010-05-16 10:49:59 -04007649 btrfs_end_transaction_throttle(trans, tree_root);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01007650out_free:
Yan Zheng2c47e6052009-06-27 21:07:35 -04007651 kfree(wc);
Chris Mason5caf2a02007-04-02 11:20:42 -04007652 btrfs_free_path(path);
Tsutomu Itohcb1b69f2011-08-09 07:11:13 +00007653out:
Josef Bacikd29a9f62013-07-17 19:30:20 -04007654 /*
7655 * So if we need to stop dropping the snapshot for whatever reason we
7656 * need to make sure to add it back to the dead root list so that we
7657 * keep trying to do the work later. This also cleans up roots if we
7658 * don't have it in the radix (like when we recover after a power fail
7659 * or unmount) so we don't leak memory.
7660 */
7661 if (root_dropped == false)
7662 btrfs_add_dead_root(root);
Tsutomu Itohcb1b69f2011-08-09 07:11:13 +00007663 if (err)
7664 btrfs_std_error(root->fs_info, err);
Jeff Mahoney2c536792011-10-03 23:22:41 -04007665 return err;
Chris Mason20524f02007-03-10 06:35:47 -05007666}
Chris Mason9078a3e2007-04-26 16:46:15 -04007667
Yan Zheng2c47e6052009-06-27 21:07:35 -04007668/*
7669 * drop subtree rooted at tree block 'node'.
7670 *
7671 * NOTE: this function will unlock and release tree block 'node'
Arne Jansen66d7e7f2011-09-12 15:26:38 +02007672 * only used by relocation code
Yan Zheng2c47e6052009-06-27 21:07:35 -04007673 */
Yan Zhengf82d02d2008-10-29 14:49:05 -04007674int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
7675 struct btrfs_root *root,
7676 struct extent_buffer *node,
7677 struct extent_buffer *parent)
7678{
7679 struct btrfs_path *path;
Yan Zheng2c47e6052009-06-27 21:07:35 -04007680 struct walk_control *wc;
Yan Zhengf82d02d2008-10-29 14:49:05 -04007681 int level;
7682 int parent_level;
7683 int ret = 0;
7684 int wret;
7685
Yan Zheng2c47e6052009-06-27 21:07:35 -04007686 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
7687
Yan Zhengf82d02d2008-10-29 14:49:05 -04007688 path = btrfs_alloc_path();
Tsutomu Itohdb5b4932011-03-23 08:14:16 +00007689 if (!path)
7690 return -ENOMEM;
Yan Zhengf82d02d2008-10-29 14:49:05 -04007691
Yan Zheng2c47e6052009-06-27 21:07:35 -04007692 wc = kzalloc(sizeof(*wc), GFP_NOFS);
Tsutomu Itohdb5b4932011-03-23 08:14:16 +00007693 if (!wc) {
7694 btrfs_free_path(path);
7695 return -ENOMEM;
7696 }
Yan Zheng2c47e6052009-06-27 21:07:35 -04007697
Chris Masonb9447ef2009-03-09 11:45:38 -04007698 btrfs_assert_tree_locked(parent);
Yan Zhengf82d02d2008-10-29 14:49:05 -04007699 parent_level = btrfs_header_level(parent);
7700 extent_buffer_get(parent);
7701 path->nodes[parent_level] = parent;
7702 path->slots[parent_level] = btrfs_header_nritems(parent);
7703
Chris Masonb9447ef2009-03-09 11:45:38 -04007704 btrfs_assert_tree_locked(node);
Yan Zhengf82d02d2008-10-29 14:49:05 -04007705 level = btrfs_header_level(node);
Yan Zhengf82d02d2008-10-29 14:49:05 -04007706 path->nodes[level] = node;
7707 path->slots[level] = 0;
Chris Masonbd681512011-07-16 15:23:14 -04007708 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
Yan Zheng2c47e6052009-06-27 21:07:35 -04007709
7710 wc->refs[parent_level] = 1;
7711 wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
7712 wc->level = level;
7713 wc->shared_level = -1;
7714 wc->stage = DROP_REFERENCE;
7715 wc->update_ref = 0;
7716 wc->keep_locks = 1;
Arne Jansen66d7e7f2011-09-12 15:26:38 +02007717 wc->for_reloc = 1;
Yan, Zheng1c4850e2009-09-21 15:55:59 -04007718 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
Yan Zhengf82d02d2008-10-29 14:49:05 -04007719
7720 while (1) {
Yan Zheng2c47e6052009-06-27 21:07:35 -04007721 wret = walk_down_tree(trans, root, path, wc);
7722 if (wret < 0) {
Yan Zhengf82d02d2008-10-29 14:49:05 -04007723 ret = wret;
Yan Zhengf82d02d2008-10-29 14:49:05 -04007724 break;
Yan Zheng2c47e6052009-06-27 21:07:35 -04007725 }
Yan Zhengf82d02d2008-10-29 14:49:05 -04007726
Yan Zheng2c47e6052009-06-27 21:07:35 -04007727 wret = walk_up_tree(trans, root, path, wc, parent_level);
Yan Zhengf82d02d2008-10-29 14:49:05 -04007728 if (wret < 0)
7729 ret = wret;
7730 if (wret != 0)
7731 break;
7732 }
7733
Yan Zheng2c47e6052009-06-27 21:07:35 -04007734 kfree(wc);
Yan Zhengf82d02d2008-10-29 14:49:05 -04007735 btrfs_free_path(path);
7736 return ret;
7737}
7738
Chris Masonec44a352008-04-28 15:29:52 -04007739static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
7740{
7741 u64 num_devices;
Ilya Dryomovfc67c452012-03-27 17:09:17 +03007742 u64 stripped;
Chris Masonec44a352008-04-28 15:29:52 -04007743
Ilya Dryomovfc67c452012-03-27 17:09:17 +03007744 /*
7745 * if restripe for this chunk_type is on pick target profile and
7746 * return, otherwise do the usual balance
7747 */
7748 stripped = get_restripe_target(root->fs_info, flags);
7749 if (stripped)
7750 return extended_to_chunk(stripped);
Ilya Dryomove4d8ec02012-01-16 22:04:48 +02007751
Chris Masoncd02dca2010-12-13 14:56:23 -05007752 /*
7753 * we add in the count of missing devices because we want
7754 * to make sure that any RAID levels on a degraded FS
7755 * continue to be honored.
7756 */
7757 num_devices = root->fs_info->fs_devices->rw_devices +
7758 root->fs_info->fs_devices->missing_devices;
7759
Ilya Dryomovfc67c452012-03-27 17:09:17 +03007760 stripped = BTRFS_BLOCK_GROUP_RAID0 |
David Woodhouse53b381b2013-01-29 18:40:14 -05007761 BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 |
Ilya Dryomovfc67c452012-03-27 17:09:17 +03007762 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
7763
Chris Masonec44a352008-04-28 15:29:52 -04007764 if (num_devices == 1) {
7765 stripped |= BTRFS_BLOCK_GROUP_DUP;
7766 stripped = flags & ~stripped;
7767
7768 /* turn raid0 into single device chunks */
7769 if (flags & BTRFS_BLOCK_GROUP_RAID0)
7770 return stripped;
7771
7772 /* turn mirroring into duplication */
7773 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
7774 BTRFS_BLOCK_GROUP_RAID10))
7775 return stripped | BTRFS_BLOCK_GROUP_DUP;
Chris Masonec44a352008-04-28 15:29:52 -04007776 } else {
7777 /* they already had raid on here, just return */
Chris Masonec44a352008-04-28 15:29:52 -04007778 if (flags & stripped)
7779 return flags;
7780
7781 stripped |= BTRFS_BLOCK_GROUP_DUP;
7782 stripped = flags & ~stripped;
7783
7784 /* switch duplicated blocks with raid1 */
7785 if (flags & BTRFS_BLOCK_GROUP_DUP)
7786 return stripped | BTRFS_BLOCK_GROUP_RAID1;
7787
Ilya Dryomove3176ca2012-03-27 17:09:16 +03007788 /* this is drive concat, leave it alone */
Chris Masonec44a352008-04-28 15:29:52 -04007789 }
Ilya Dryomove3176ca2012-03-27 17:09:16 +03007790
Chris Masonec44a352008-04-28 15:29:52 -04007791 return flags;
7792}
7793
Miao Xie199c36e2011-07-15 10:34:36 +00007794static int set_block_group_ro(struct btrfs_block_group_cache *cache, int force)
Chris Mason0ef3e662008-05-24 14:04:53 -04007795{
Yan, Zhengf0486c62010-05-16 10:46:25 -04007796 struct btrfs_space_info *sinfo = cache->space_info;
7797 u64 num_bytes;
Miao Xie199c36e2011-07-15 10:34:36 +00007798 u64 min_allocable_bytes;
Yan, Zhengf0486c62010-05-16 10:46:25 -04007799 int ret = -ENOSPC;
Chris Mason0ef3e662008-05-24 14:04:53 -04007800
Chris Masonc286ac42008-07-22 23:06:41 -04007801
Miao Xie199c36e2011-07-15 10:34:36 +00007802 /*
7803 * We need some metadata space and system metadata space for
7804 * allocating chunks in some corner cases until we force to set
7805 * it to be readonly.
7806 */
7807 if ((sinfo->flags &
7808 (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) &&
7809 !force)
7810 min_allocable_bytes = 1 * 1024 * 1024;
7811 else
7812 min_allocable_bytes = 0;
7813
Yan, Zhengf0486c62010-05-16 10:46:25 -04007814 spin_lock(&sinfo->lock);
7815 spin_lock(&cache->lock);
WuBo61cfea92011-07-26 03:30:11 +00007816
7817 if (cache->ro) {
7818 ret = 0;
7819 goto out;
7820 }
7821
Yan, Zhengf0486c62010-05-16 10:46:25 -04007822 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
7823 cache->bytes_super - btrfs_block_group_used(&cache->item);
Chris Mason7d9eb122008-07-08 14:19:17 -04007824
Yan, Zhengf0486c62010-05-16 10:46:25 -04007825 if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned +
Josef Bacik37be25b2011-08-05 10:25:38 -04007826 sinfo->bytes_may_use + sinfo->bytes_readonly + num_bytes +
7827 min_allocable_bytes <= sinfo->total_bytes) {
Yan, Zhengf0486c62010-05-16 10:46:25 -04007828 sinfo->bytes_readonly += num_bytes;
Yan, Zhengf0486c62010-05-16 10:46:25 -04007829 cache->ro = 1;
7830 ret = 0;
7831 }
WuBo61cfea92011-07-26 03:30:11 +00007832out:
Yan, Zhengf0486c62010-05-16 10:46:25 -04007833 spin_unlock(&cache->lock);
7834 spin_unlock(&sinfo->lock);
7835 return ret;
Chris Mason0ef3e662008-05-24 14:04:53 -04007836}
7837
Yan, Zhengf0486c62010-05-16 10:46:25 -04007838int btrfs_set_block_group_ro(struct btrfs_root *root,
7839 struct btrfs_block_group_cache *cache)
Yan Zheng5d4f98a2009-06-10 10:45:14 -04007840
7841{
Yan, Zhengf0486c62010-05-16 10:46:25 -04007842 struct btrfs_trans_handle *trans;
7843 u64 alloc_flags;
7844 int ret;
7845
7846 BUG_ON(cache->ro);
7847
Josef Bacik7a7eaa42011-04-13 12:54:33 -04007848 trans = btrfs_join_transaction(root);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01007849 if (IS_ERR(trans))
7850 return PTR_ERR(trans);
Yan, Zhengf0486c62010-05-16 10:46:25 -04007851
7852 alloc_flags = update_block_group_flags(root, cache->flags);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01007853 if (alloc_flags != cache->flags) {
Josef Bacik698d0082012-09-12 14:08:47 -04007854 ret = do_chunk_alloc(trans, root, alloc_flags,
Jeff Mahoney79787ea2012-03-12 16:03:00 +01007855 CHUNK_ALLOC_FORCE);
7856 if (ret < 0)
7857 goto out;
7858 }
Yan, Zhengf0486c62010-05-16 10:46:25 -04007859
Miao Xie199c36e2011-07-15 10:34:36 +00007860 ret = set_block_group_ro(cache, 0);
Yan, Zhengf0486c62010-05-16 10:46:25 -04007861 if (!ret)
7862 goto out;
7863 alloc_flags = get_alloc_profile(root, cache->space_info->flags);
Josef Bacik698d0082012-09-12 14:08:47 -04007864 ret = do_chunk_alloc(trans, root, alloc_flags,
Chris Mason0e4f8f82011-04-15 16:05:44 -04007865 CHUNK_ALLOC_FORCE);
Yan, Zhengf0486c62010-05-16 10:46:25 -04007866 if (ret < 0)
7867 goto out;
Miao Xie199c36e2011-07-15 10:34:36 +00007868 ret = set_block_group_ro(cache, 0);
Yan, Zhengf0486c62010-05-16 10:46:25 -04007869out:
7870 btrfs_end_transaction(trans, root);
7871 return ret;
7872}
7873
Chris Masonc87f08c2011-02-16 13:57:04 -05007874int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans,
7875 struct btrfs_root *root, u64 type)
7876{
7877 u64 alloc_flags = get_alloc_profile(root, type);
Josef Bacik698d0082012-09-12 14:08:47 -04007878 return do_chunk_alloc(trans, root, alloc_flags,
Chris Mason0e4f8f82011-04-15 16:05:44 -04007879 CHUNK_ALLOC_FORCE);
Chris Masonc87f08c2011-02-16 13:57:04 -05007880}
7881
Miao Xie6d07bce2011-01-05 10:07:31 +00007882/*
7883 * helper to account the unused space of all the readonly block group in the
7884 * list. takes mirrors into account.
7885 */
7886static u64 __btrfs_get_ro_block_group_free_space(struct list_head *groups_list)
7887{
7888 struct btrfs_block_group_cache *block_group;
7889 u64 free_bytes = 0;
7890 int factor;
7891
7892 list_for_each_entry(block_group, groups_list, list) {
7893 spin_lock(&block_group->lock);
7894
7895 if (!block_group->ro) {
7896 spin_unlock(&block_group->lock);
7897 continue;
7898 }
7899
7900 if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 |
7901 BTRFS_BLOCK_GROUP_RAID10 |
7902 BTRFS_BLOCK_GROUP_DUP))
7903 factor = 2;
7904 else
7905 factor = 1;
7906
7907 free_bytes += (block_group->key.offset -
7908 btrfs_block_group_used(&block_group->item)) *
7909 factor;
7910
7911 spin_unlock(&block_group->lock);
7912 }
7913
7914 return free_bytes;
7915}
7916
7917/*
7918 * helper to account the unused space of all the readonly block group in the
7919 * space_info. takes mirrors into account.
7920 */
7921u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
7922{
7923 int i;
7924 u64 free_bytes = 0;
7925
7926 spin_lock(&sinfo->lock);
7927
7928 for(i = 0; i < BTRFS_NR_RAID_TYPES; i++)
7929 if (!list_empty(&sinfo->block_groups[i]))
7930 free_bytes += __btrfs_get_ro_block_group_free_space(
7931 &sinfo->block_groups[i]);
7932
7933 spin_unlock(&sinfo->lock);
7934
7935 return free_bytes;
7936}
7937
Jeff Mahoney143bede2012-03-01 14:56:26 +01007938void btrfs_set_block_group_rw(struct btrfs_root *root,
Yan, Zhengf0486c62010-05-16 10:46:25 -04007939 struct btrfs_block_group_cache *cache)
7940{
7941 struct btrfs_space_info *sinfo = cache->space_info;
7942 u64 num_bytes;
7943
7944 BUG_ON(!cache->ro);
7945
7946 spin_lock(&sinfo->lock);
7947 spin_lock(&cache->lock);
7948 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
7949 cache->bytes_super - btrfs_block_group_used(&cache->item);
7950 sinfo->bytes_readonly -= num_bytes;
7951 cache->ro = 0;
7952 spin_unlock(&cache->lock);
7953 spin_unlock(&sinfo->lock);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04007954}
7955
Josef Bacikba1bf482009-09-11 16:11:19 -04007956/*
7957 * checks to see if its even possible to relocate this block group.
7958 *
7959 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
7960 * ok to go ahead and try.
7961 */
7962int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr)
Zheng Yan1a40e232008-09-26 10:09:34 -04007963{
Zheng Yan1a40e232008-09-26 10:09:34 -04007964 struct btrfs_block_group_cache *block_group;
Josef Bacikba1bf482009-09-11 16:11:19 -04007965 struct btrfs_space_info *space_info;
7966 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
7967 struct btrfs_device *device;
Josef Bacik6df9a952013-06-27 13:22:46 -04007968 struct btrfs_trans_handle *trans;
liubocdcb7252011-08-03 10:15:25 +00007969 u64 min_free;
Josef Bacik6719db62011-08-20 08:29:51 -04007970 u64 dev_min = 1;
7971 u64 dev_nr = 0;
Ilya Dryomov4a5e98f2012-03-27 17:09:17 +03007972 u64 target;
liubocdcb7252011-08-03 10:15:25 +00007973 int index;
Josef Bacikba1bf482009-09-11 16:11:19 -04007974 int full = 0;
7975 int ret = 0;
Chris Masonedbd8d42007-12-21 16:27:24 -05007976
Josef Bacikba1bf482009-09-11 16:11:19 -04007977 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
Zheng Yan1a40e232008-09-26 10:09:34 -04007978
Josef Bacikba1bf482009-09-11 16:11:19 -04007979 /* odd, couldn't find the block group, leave it alone */
7980 if (!block_group)
7981 return -1;
Chris Masonedbd8d42007-12-21 16:27:24 -05007982
liubocdcb7252011-08-03 10:15:25 +00007983 min_free = btrfs_block_group_used(&block_group->item);
7984
Josef Bacikba1bf482009-09-11 16:11:19 -04007985 /* no bytes used, we're good */
liubocdcb7252011-08-03 10:15:25 +00007986 if (!min_free)
Josef Bacikba1bf482009-09-11 16:11:19 -04007987 goto out;
Chris Mason323da792008-05-09 11:46:48 -04007988
Josef Bacikba1bf482009-09-11 16:11:19 -04007989 space_info = block_group->space_info;
7990 spin_lock(&space_info->lock);
Chris Mason323da792008-05-09 11:46:48 -04007991
Josef Bacikba1bf482009-09-11 16:11:19 -04007992 full = space_info->full;
Zheng Yan1a40e232008-09-26 10:09:34 -04007993
Josef Bacikba1bf482009-09-11 16:11:19 -04007994 /*
7995 * if this is the last block group we have in this space, we can't
Chris Mason7ce618d2009-09-22 14:48:44 -04007996 * relocate it unless we're able to allocate a new chunk below.
7997 *
7998 * Otherwise, we need to make sure we have room in the space to handle
7999 * all of the extents from this block group. If we can, we're good
Josef Bacikba1bf482009-09-11 16:11:19 -04008000 */
Chris Mason7ce618d2009-09-22 14:48:44 -04008001 if ((space_info->total_bytes != block_group->key.offset) &&
liubocdcb7252011-08-03 10:15:25 +00008002 (space_info->bytes_used + space_info->bytes_reserved +
8003 space_info->bytes_pinned + space_info->bytes_readonly +
8004 min_free < space_info->total_bytes)) {
Josef Bacikba1bf482009-09-11 16:11:19 -04008005 spin_unlock(&space_info->lock);
8006 goto out;
8007 }
8008 spin_unlock(&space_info->lock);
Zheng Yan1a40e232008-09-26 10:09:34 -04008009
Josef Bacikba1bf482009-09-11 16:11:19 -04008010 /*
8011 * ok we don't have enough space, but maybe we have free space on our
8012 * devices to allocate new chunks for relocation, so loop through our
Ilya Dryomov4a5e98f2012-03-27 17:09:17 +03008013 * alloc devices and guess if we have enough space. if this block
8014 * group is going to be restriped, run checks against the target
8015 * profile instead of the current one.
Josef Bacikba1bf482009-09-11 16:11:19 -04008016 */
8017 ret = -1;
Chris Mason4313b392008-01-03 09:08:48 -05008018
liubocdcb7252011-08-03 10:15:25 +00008019 /*
8020 * index:
8021 * 0: raid10
8022 * 1: raid1
8023 * 2: dup
8024 * 3: raid0
8025 * 4: single
8026 */
Ilya Dryomov4a5e98f2012-03-27 17:09:17 +03008027 target = get_restripe_target(root->fs_info, block_group->flags);
8028 if (target) {
Liu Bo31e50222012-11-21 14:18:10 +00008029 index = __get_raid_index(extended_to_chunk(target));
Ilya Dryomov4a5e98f2012-03-27 17:09:17 +03008030 } else {
8031 /*
8032 * this is just a balance, so if we were marked as full
8033 * we know there is no space for a new chunk
8034 */
8035 if (full)
8036 goto out;
8037
8038 index = get_block_group_index(block_group);
8039 }
8040
Miao Xiee6ec7162013-01-17 05:38:51 +00008041 if (index == BTRFS_RAID_RAID10) {
liubocdcb7252011-08-03 10:15:25 +00008042 dev_min = 4;
Josef Bacik6719db62011-08-20 08:29:51 -04008043 /* Divide by 2 */
8044 min_free >>= 1;
Miao Xiee6ec7162013-01-17 05:38:51 +00008045 } else if (index == BTRFS_RAID_RAID1) {
liubocdcb7252011-08-03 10:15:25 +00008046 dev_min = 2;
Miao Xiee6ec7162013-01-17 05:38:51 +00008047 } else if (index == BTRFS_RAID_DUP) {
Josef Bacik6719db62011-08-20 08:29:51 -04008048 /* Multiply by 2 */
8049 min_free <<= 1;
Miao Xiee6ec7162013-01-17 05:38:51 +00008050 } else if (index == BTRFS_RAID_RAID0) {
liubocdcb7252011-08-03 10:15:25 +00008051 dev_min = fs_devices->rw_devices;
Josef Bacik6719db62011-08-20 08:29:51 -04008052 do_div(min_free, dev_min);
liubocdcb7252011-08-03 10:15:25 +00008053 }
8054
Josef Bacik6df9a952013-06-27 13:22:46 -04008055 /* We need to do this so that we can look at pending chunks */
8056 trans = btrfs_join_transaction(root);
8057 if (IS_ERR(trans)) {
8058 ret = PTR_ERR(trans);
8059 goto out;
8060 }
8061
Josef Bacikba1bf482009-09-11 16:11:19 -04008062 mutex_lock(&root->fs_info->chunk_mutex);
8063 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
Miao Xie7bfc8372011-01-05 10:07:26 +00008064 u64 dev_offset;
Chris Masonea8c2812008-08-04 23:17:27 -04008065
Josef Bacikba1bf482009-09-11 16:11:19 -04008066 /*
8067 * check to make sure we can actually find a chunk with enough
8068 * space to fit our block group in.
8069 */
Stefan Behrens63a212a2012-11-05 18:29:28 +01008070 if (device->total_bytes > device->bytes_used + min_free &&
8071 !device->is_tgtdev_for_dev_replace) {
Josef Bacik6df9a952013-06-27 13:22:46 -04008072 ret = find_free_dev_extent(trans, device, min_free,
Miao Xie7bfc8372011-01-05 10:07:26 +00008073 &dev_offset, NULL);
Josef Bacikba1bf482009-09-11 16:11:19 -04008074 if (!ret)
liubocdcb7252011-08-03 10:15:25 +00008075 dev_nr++;
8076
8077 if (dev_nr >= dev_min)
Yan73e48b22008-01-03 14:14:39 -05008078 break;
liubocdcb7252011-08-03 10:15:25 +00008079
Josef Bacikba1bf482009-09-11 16:11:19 -04008080 ret = -1;
Yan73e48b22008-01-03 14:14:39 -05008081 }
Chris Masonedbd8d42007-12-21 16:27:24 -05008082 }
Josef Bacikba1bf482009-09-11 16:11:19 -04008083 mutex_unlock(&root->fs_info->chunk_mutex);
Josef Bacik6df9a952013-06-27 13:22:46 -04008084 btrfs_end_transaction(trans, root);
Chris Masonedbd8d42007-12-21 16:27:24 -05008085out:
Josef Bacikba1bf482009-09-11 16:11:19 -04008086 btrfs_put_block_group(block_group);
Chris Masonedbd8d42007-12-21 16:27:24 -05008087 return ret;
8088}
8089
Christoph Hellwigb2950862008-12-02 09:54:17 -05008090static int find_first_block_group(struct btrfs_root *root,
8091 struct btrfs_path *path, struct btrfs_key *key)
Chris Mason0b86a832008-03-24 15:01:56 -04008092{
Chris Mason925baed2008-06-25 16:01:30 -04008093 int ret = 0;
Chris Mason0b86a832008-03-24 15:01:56 -04008094 struct btrfs_key found_key;
8095 struct extent_buffer *leaf;
8096 int slot;
8097
8098 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
8099 if (ret < 0)
Chris Mason925baed2008-06-25 16:01:30 -04008100 goto out;
8101
Chris Masond3977122009-01-05 21:25:51 -05008102 while (1) {
Chris Mason0b86a832008-03-24 15:01:56 -04008103 slot = path->slots[0];
8104 leaf = path->nodes[0];
8105 if (slot >= btrfs_header_nritems(leaf)) {
8106 ret = btrfs_next_leaf(root, path);
8107 if (ret == 0)
8108 continue;
8109 if (ret < 0)
Chris Mason925baed2008-06-25 16:01:30 -04008110 goto out;
Chris Mason0b86a832008-03-24 15:01:56 -04008111 break;
8112 }
8113 btrfs_item_key_to_cpu(leaf, &found_key, slot);
8114
8115 if (found_key.objectid >= key->objectid &&
Chris Mason925baed2008-06-25 16:01:30 -04008116 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
8117 ret = 0;
8118 goto out;
8119 }
Chris Mason0b86a832008-03-24 15:01:56 -04008120 path->slots[0]++;
8121 }
Chris Mason925baed2008-06-25 16:01:30 -04008122out:
Chris Mason0b86a832008-03-24 15:01:56 -04008123 return ret;
8124}
8125
Josef Bacik0af3d002010-06-21 14:48:16 -04008126void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
8127{
8128 struct btrfs_block_group_cache *block_group;
8129 u64 last = 0;
8130
8131 while (1) {
8132 struct inode *inode;
8133
8134 block_group = btrfs_lookup_first_block_group(info, last);
8135 while (block_group) {
8136 spin_lock(&block_group->lock);
8137 if (block_group->iref)
8138 break;
8139 spin_unlock(&block_group->lock);
8140 block_group = next_block_group(info->tree_root,
8141 block_group);
8142 }
8143 if (!block_group) {
8144 if (last == 0)
8145 break;
8146 last = 0;
8147 continue;
8148 }
8149
8150 inode = block_group->inode;
8151 block_group->iref = 0;
8152 block_group->inode = NULL;
8153 spin_unlock(&block_group->lock);
8154 iput(inode);
8155 last = block_group->key.objectid + block_group->key.offset;
8156 btrfs_put_block_group(block_group);
8157 }
8158}
8159
Zheng Yan1a40e232008-09-26 10:09:34 -04008160int btrfs_free_block_groups(struct btrfs_fs_info *info)
8161{
8162 struct btrfs_block_group_cache *block_group;
Chris Mason4184ea72009-03-10 12:39:20 -04008163 struct btrfs_space_info *space_info;
Yan Zheng11833d62009-09-11 16:11:19 -04008164 struct btrfs_caching_control *caching_ctl;
Zheng Yan1a40e232008-09-26 10:09:34 -04008165 struct rb_node *n;
8166
Yan Zheng11833d62009-09-11 16:11:19 -04008167 down_write(&info->extent_commit_sem);
8168 while (!list_empty(&info->caching_block_groups)) {
8169 caching_ctl = list_entry(info->caching_block_groups.next,
8170 struct btrfs_caching_control, list);
8171 list_del(&caching_ctl->list);
8172 put_caching_control(caching_ctl);
8173 }
8174 up_write(&info->extent_commit_sem);
8175
Zheng Yan1a40e232008-09-26 10:09:34 -04008176 spin_lock(&info->block_group_cache_lock);
8177 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
8178 block_group = rb_entry(n, struct btrfs_block_group_cache,
8179 cache_node);
Zheng Yan1a40e232008-09-26 10:09:34 -04008180 rb_erase(&block_group->cache_node,
8181 &info->block_group_cache_tree);
Yan Zhengd899e052008-10-30 14:25:28 -04008182 spin_unlock(&info->block_group_cache_lock);
8183
Josef Bacik80eb2342008-10-29 14:49:05 -04008184 down_write(&block_group->space_info->groups_sem);
Zheng Yan1a40e232008-09-26 10:09:34 -04008185 list_del(&block_group->list);
Josef Bacik80eb2342008-10-29 14:49:05 -04008186 up_write(&block_group->space_info->groups_sem);
Yan Zhengd2fb3432008-12-11 16:30:39 -05008187
Josef Bacik817d52f2009-07-13 21:29:25 -04008188 if (block_group->cached == BTRFS_CACHE_STARTED)
Yan Zheng11833d62009-09-11 16:11:19 -04008189 wait_block_group_cache_done(block_group);
Josef Bacik817d52f2009-07-13 21:29:25 -04008190
Josef Bacik3c148742011-02-02 15:53:47 +00008191 /*
8192 * We haven't cached this block group, which means we could
8193 * possibly have excluded extents on this block group.
8194 */
8195 if (block_group->cached == BTRFS_CACHE_NO)
8196 free_excluded_extents(info->extent_root, block_group);
8197
Josef Bacik817d52f2009-07-13 21:29:25 -04008198 btrfs_remove_free_space_cache(block_group);
Josef Bacik11dfe352009-11-13 20:12:59 +00008199 btrfs_put_block_group(block_group);
Yan Zhengd899e052008-10-30 14:25:28 -04008200
8201 spin_lock(&info->block_group_cache_lock);
Zheng Yan1a40e232008-09-26 10:09:34 -04008202 }
8203 spin_unlock(&info->block_group_cache_lock);
Chris Mason4184ea72009-03-10 12:39:20 -04008204
8205 /* now that all the block groups are freed, go through and
8206 * free all the space_info structs. This is only called during
8207 * the final stages of unmount, and so we know nobody is
8208 * using them. We call synchronize_rcu() once before we start,
8209 * just to be on the safe side.
8210 */
8211 synchronize_rcu();
8212
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04008213 release_global_block_rsv(info);
8214
Chris Mason4184ea72009-03-10 12:39:20 -04008215 while(!list_empty(&info->space_info)) {
8216 space_info = list_entry(info->space_info.next,
8217 struct btrfs_space_info,
8218 list);
David Sterbab069e0c2013-02-08 21:28:17 +00008219 if (btrfs_test_opt(info->tree_root, ENOSPC_DEBUG)) {
8220 if (space_info->bytes_pinned > 0 ||
8221 space_info->bytes_reserved > 0 ||
8222 space_info->bytes_may_use > 0) {
8223 WARN_ON(1);
8224 dump_space_info(space_info, 0, 0);
8225 }
Yan, Zhengf0486c62010-05-16 10:46:25 -04008226 }
Josef Bacikb150a4f2013-06-19 15:00:04 -04008227 percpu_counter_destroy(&space_info->total_bytes_pinned);
Chris Mason4184ea72009-03-10 12:39:20 -04008228 list_del(&space_info->list);
8229 kfree(space_info);
8230 }
Zheng Yan1a40e232008-09-26 10:09:34 -04008231 return 0;
8232}
8233
Yan, Zhengb742bb82010-05-16 10:46:24 -04008234static void __link_block_group(struct btrfs_space_info *space_info,
8235 struct btrfs_block_group_cache *cache)
8236{
8237 int index = get_block_group_index(cache);
8238
8239 down_write(&space_info->groups_sem);
8240 list_add_tail(&cache->list, &space_info->block_groups[index]);
8241 up_write(&space_info->groups_sem);
8242}
8243
Chris Mason9078a3e2007-04-26 16:46:15 -04008244int btrfs_read_block_groups(struct btrfs_root *root)
8245{
8246 struct btrfs_path *path;
8247 int ret;
Chris Mason9078a3e2007-04-26 16:46:15 -04008248 struct btrfs_block_group_cache *cache;
Chris Masonbe744172007-05-06 10:15:01 -04008249 struct btrfs_fs_info *info = root->fs_info;
Chris Mason6324fbf2008-03-24 15:01:59 -04008250 struct btrfs_space_info *space_info;
Chris Mason9078a3e2007-04-26 16:46:15 -04008251 struct btrfs_key key;
8252 struct btrfs_key found_key;
Chris Mason5f39d392007-10-15 16:14:19 -04008253 struct extent_buffer *leaf;
Josef Bacik0af3d002010-06-21 14:48:16 -04008254 int need_clear = 0;
8255 u64 cache_gen;
Chris Mason96b51792007-10-15 16:15:19 -04008256
Chris Masonbe744172007-05-06 10:15:01 -04008257 root = info->extent_root;
Chris Mason9078a3e2007-04-26 16:46:15 -04008258 key.objectid = 0;
Chris Mason0b86a832008-03-24 15:01:56 -04008259 key.offset = 0;
Chris Mason9078a3e2007-04-26 16:46:15 -04008260 btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
Chris Mason9078a3e2007-04-26 16:46:15 -04008261 path = btrfs_alloc_path();
8262 if (!path)
8263 return -ENOMEM;
Josef Bacik026fd312011-05-13 10:32:11 -04008264 path->reada = 1;
Chris Mason9078a3e2007-04-26 16:46:15 -04008265
David Sterba6c417612011-04-13 15:41:04 +02008266 cache_gen = btrfs_super_cache_generation(root->fs_info->super_copy);
Josef Bacik73bc1872011-10-03 14:07:49 -04008267 if (btrfs_test_opt(root, SPACE_CACHE) &&
David Sterba6c417612011-04-13 15:41:04 +02008268 btrfs_super_generation(root->fs_info->super_copy) != cache_gen)
Josef Bacik0af3d002010-06-21 14:48:16 -04008269 need_clear = 1;
Josef Bacik88c2ba32010-09-21 14:21:34 -04008270 if (btrfs_test_opt(root, CLEAR_CACHE))
8271 need_clear = 1;
Josef Bacik0af3d002010-06-21 14:48:16 -04008272
Chris Masond3977122009-01-05 21:25:51 -05008273 while (1) {
Chris Mason0b86a832008-03-24 15:01:56 -04008274 ret = find_first_block_group(root, path, &key);
Yan, Zhengb742bb82010-05-16 10:46:24 -04008275 if (ret > 0)
8276 break;
Chris Mason0b86a832008-03-24 15:01:56 -04008277 if (ret != 0)
8278 goto error;
Chris Mason5f39d392007-10-15 16:14:19 -04008279 leaf = path->nodes[0];
8280 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
Chris Mason8f18cf12008-04-25 16:53:30 -04008281 cache = kzalloc(sizeof(*cache), GFP_NOFS);
Chris Mason9078a3e2007-04-26 16:46:15 -04008282 if (!cache) {
Chris Mason0b86a832008-03-24 15:01:56 -04008283 ret = -ENOMEM;
Yan, Zhengf0486c62010-05-16 10:46:25 -04008284 goto error;
Chris Mason9078a3e2007-04-26 16:46:15 -04008285 }
Li Zefan34d52cb2011-03-29 13:46:06 +08008286 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
8287 GFP_NOFS);
8288 if (!cache->free_space_ctl) {
8289 kfree(cache);
8290 ret = -ENOMEM;
8291 goto error;
8292 }
Chris Mason3e1ad542007-05-07 20:03:49 -04008293
Yan Zhengd2fb3432008-12-11 16:30:39 -05008294 atomic_set(&cache->count, 1);
Chris Masonc286ac42008-07-22 23:06:41 -04008295 spin_lock_init(&cache->lock);
Josef Bacik817d52f2009-07-13 21:29:25 -04008296 cache->fs_info = info;
Josef Bacik0f9dd462008-09-23 13:14:11 -04008297 INIT_LIST_HEAD(&cache->list);
Chris Masonfa9c0d792009-04-03 09:47:43 -04008298 INIT_LIST_HEAD(&cache->cluster_list);
Josef Bacik96303082009-07-13 21:29:25 -04008299
Liu Bocf7c1ef2012-07-06 03:31:34 -06008300 if (need_clear) {
8301 /*
8302 * When we mount with old space cache, we need to
8303 * set BTRFS_DC_CLEAR and set dirty flag.
8304 *
8305 * a) Setting 'BTRFS_DC_CLEAR' makes sure that we
8306 * truncate the old free space cache inode and
8307 * setup a new one.
8308 * b) Setting 'dirty flag' makes sure that we flush
8309 * the new space cache info onto disk.
8310 */
Josef Bacik0af3d002010-06-21 14:48:16 -04008311 cache->disk_cache_state = BTRFS_DC_CLEAR;
Liu Bocf7c1ef2012-07-06 03:31:34 -06008312 if (btrfs_test_opt(root, SPACE_CACHE))
8313 cache->dirty = 1;
8314 }
Josef Bacik0af3d002010-06-21 14:48:16 -04008315
Chris Mason5f39d392007-10-15 16:14:19 -04008316 read_extent_buffer(leaf, &cache->item,
8317 btrfs_item_ptr_offset(leaf, path->slots[0]),
8318 sizeof(cache->item));
Chris Mason9078a3e2007-04-26 16:46:15 -04008319 memcpy(&cache->key, &found_key, sizeof(found_key));
Chris Mason0b86a832008-03-24 15:01:56 -04008320
Chris Mason9078a3e2007-04-26 16:46:15 -04008321 key.objectid = found_key.objectid + found_key.offset;
David Sterbab3b4aa72011-04-21 01:20:15 +02008322 btrfs_release_path(path);
Chris Mason0b86a832008-03-24 15:01:56 -04008323 cache->flags = btrfs_block_group_flags(&cache->item);
Josef Bacik817d52f2009-07-13 21:29:25 -04008324 cache->sectorsize = root->sectorsize;
David Woodhouse53b381b2013-01-29 18:40:14 -05008325 cache->full_stripe_len = btrfs_full_stripe_len(root,
8326 &root->fs_info->mapping_tree,
8327 found_key.objectid);
Li Zefan34d52cb2011-03-29 13:46:06 +08008328 btrfs_init_free_space_ctl(cache);
8329
Josef Bacik817d52f2009-07-13 21:29:25 -04008330 /*
Josef Bacik3c148742011-02-02 15:53:47 +00008331 * We need to exclude the super stripes now so that the space
8332 * info has super bytes accounted for, otherwise we'll think
8333 * we have more space than we actually do.
8334 */
Josef Bacik835d9742013-03-19 12:13:25 -04008335 ret = exclude_super_stripes(root, cache);
8336 if (ret) {
8337 /*
8338 * We may have excluded something, so call this just in
8339 * case.
8340 */
8341 free_excluded_extents(root, cache);
8342 kfree(cache->free_space_ctl);
8343 kfree(cache);
8344 goto error;
8345 }
Josef Bacik3c148742011-02-02 15:53:47 +00008346
8347 /*
Josef Bacik817d52f2009-07-13 21:29:25 -04008348 * check for two cases, either we are full, and therefore
8349 * don't need to bother with the caching work since we won't
8350 * find any space, or we are empty, and we can just add all
8351 * the space in and be done with it. This saves us _alot_ of
8352 * time, particularly in the full case.
8353 */
8354 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
Yan Zheng11833d62009-09-11 16:11:19 -04008355 cache->last_byte_to_unpin = (u64)-1;
Josef Bacik817d52f2009-07-13 21:29:25 -04008356 cache->cached = BTRFS_CACHE_FINISHED;
Josef Bacik1b2da372009-09-11 16:11:20 -04008357 free_excluded_extents(root, cache);
Josef Bacik817d52f2009-07-13 21:29:25 -04008358 } else if (btrfs_block_group_used(&cache->item) == 0) {
Yan Zheng11833d62009-09-11 16:11:19 -04008359 cache->last_byte_to_unpin = (u64)-1;
Josef Bacik817d52f2009-07-13 21:29:25 -04008360 cache->cached = BTRFS_CACHE_FINISHED;
8361 add_new_free_space(cache, root->fs_info,
8362 found_key.objectid,
8363 found_key.objectid +
8364 found_key.offset);
Yan Zheng11833d62009-09-11 16:11:19 -04008365 free_excluded_extents(root, cache);
Josef Bacik817d52f2009-07-13 21:29:25 -04008366 }
Chris Mason96b51792007-10-15 16:15:19 -04008367
Josef Bacik8c579fe2013-04-02 12:40:42 -04008368 ret = btrfs_add_block_group_cache(root->fs_info, cache);
8369 if (ret) {
8370 btrfs_remove_free_space_cache(cache);
8371 btrfs_put_block_group(cache);
8372 goto error;
8373 }
8374
Chris Mason6324fbf2008-03-24 15:01:59 -04008375 ret = update_space_info(info, cache->flags, found_key.offset,
8376 btrfs_block_group_used(&cache->item),
8377 &space_info);
Josef Bacik8c579fe2013-04-02 12:40:42 -04008378 if (ret) {
8379 btrfs_remove_free_space_cache(cache);
8380 spin_lock(&info->block_group_cache_lock);
8381 rb_erase(&cache->cache_node,
8382 &info->block_group_cache_tree);
8383 spin_unlock(&info->block_group_cache_lock);
8384 btrfs_put_block_group(cache);
8385 goto error;
8386 }
8387
Chris Mason6324fbf2008-03-24 15:01:59 -04008388 cache->space_info = space_info;
Josef Bacik1b2da372009-09-11 16:11:20 -04008389 spin_lock(&cache->space_info->lock);
Yan, Zhengf0486c62010-05-16 10:46:25 -04008390 cache->space_info->bytes_readonly += cache->bytes_super;
Josef Bacik1b2da372009-09-11 16:11:20 -04008391 spin_unlock(&cache->space_info->lock);
8392
Yan, Zhengb742bb82010-05-16 10:46:24 -04008393 __link_block_group(space_info, cache);
Chris Mason6324fbf2008-03-24 15:01:59 -04008394
Chris Mason75ccf472008-09-30 19:24:06 -04008395 set_avail_alloc_bits(root->fs_info, cache->flags);
Yan Zheng2b820322008-11-17 21:11:30 -05008396 if (btrfs_chunk_readonly(root, cache->key.objectid))
Miao Xie199c36e2011-07-15 10:34:36 +00008397 set_block_group_ro(cache, 1);
Chris Mason9078a3e2007-04-26 16:46:15 -04008398 }
Yan, Zhengb742bb82010-05-16 10:46:24 -04008399
8400 list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) {
8401 if (!(get_alloc_profile(root, space_info->flags) &
8402 (BTRFS_BLOCK_GROUP_RAID10 |
8403 BTRFS_BLOCK_GROUP_RAID1 |
David Woodhouse53b381b2013-01-29 18:40:14 -05008404 BTRFS_BLOCK_GROUP_RAID5 |
8405 BTRFS_BLOCK_GROUP_RAID6 |
Yan, Zhengb742bb82010-05-16 10:46:24 -04008406 BTRFS_BLOCK_GROUP_DUP)))
8407 continue;
8408 /*
8409 * avoid allocating from un-mirrored block group if there are
8410 * mirrored block groups.
8411 */
8412 list_for_each_entry(cache, &space_info->block_groups[3], list)
Miao Xie199c36e2011-07-15 10:34:36 +00008413 set_block_group_ro(cache, 1);
Yan, Zhengb742bb82010-05-16 10:46:24 -04008414 list_for_each_entry(cache, &space_info->block_groups[4], list)
Miao Xie199c36e2011-07-15 10:34:36 +00008415 set_block_group_ro(cache, 1);
Yan, Zhengb742bb82010-05-16 10:46:24 -04008416 }
Yan, Zhengf0486c62010-05-16 10:46:25 -04008417
8418 init_global_block_rsv(info);
Chris Mason0b86a832008-03-24 15:01:56 -04008419 ret = 0;
8420error:
Chris Mason9078a3e2007-04-26 16:46:15 -04008421 btrfs_free_path(path);
Chris Mason0b86a832008-03-24 15:01:56 -04008422 return ret;
Chris Mason9078a3e2007-04-26 16:46:15 -04008423}
Chris Mason6324fbf2008-03-24 15:01:59 -04008424
Josef Bacikea658ba2012-09-11 16:57:25 -04008425void btrfs_create_pending_block_groups(struct btrfs_trans_handle *trans,
8426 struct btrfs_root *root)
8427{
8428 struct btrfs_block_group_cache *block_group, *tmp;
8429 struct btrfs_root *extent_root = root->fs_info->extent_root;
8430 struct btrfs_block_group_item item;
8431 struct btrfs_key key;
8432 int ret = 0;
8433
8434 list_for_each_entry_safe(block_group, tmp, &trans->new_bgs,
8435 new_bg_list) {
8436 list_del_init(&block_group->new_bg_list);
8437
8438 if (ret)
8439 continue;
8440
8441 spin_lock(&block_group->lock);
8442 memcpy(&item, &block_group->item, sizeof(item));
8443 memcpy(&key, &block_group->key, sizeof(key));
8444 spin_unlock(&block_group->lock);
8445
8446 ret = btrfs_insert_item(trans, extent_root, &key, &item,
8447 sizeof(item));
8448 if (ret)
8449 btrfs_abort_transaction(trans, extent_root, ret);
Josef Bacik6df9a952013-06-27 13:22:46 -04008450 ret = btrfs_finish_chunk_alloc(trans, extent_root,
8451 key.objectid, key.offset);
8452 if (ret)
8453 btrfs_abort_transaction(trans, extent_root, ret);
Josef Bacikea658ba2012-09-11 16:57:25 -04008454 }
8455}
8456
Chris Mason6324fbf2008-03-24 15:01:59 -04008457int btrfs_make_block_group(struct btrfs_trans_handle *trans,
8458 struct btrfs_root *root, u64 bytes_used,
Chris Masone17cade2008-04-15 15:41:47 -04008459 u64 type, u64 chunk_objectid, u64 chunk_offset,
Chris Mason6324fbf2008-03-24 15:01:59 -04008460 u64 size)
8461{
8462 int ret;
Chris Mason6324fbf2008-03-24 15:01:59 -04008463 struct btrfs_root *extent_root;
8464 struct btrfs_block_group_cache *cache;
Chris Mason6324fbf2008-03-24 15:01:59 -04008465
8466 extent_root = root->fs_info->extent_root;
Chris Mason6324fbf2008-03-24 15:01:59 -04008467
Chris Mason12fcfd22009-03-24 10:24:20 -04008468 root->fs_info->last_trans_log_full_commit = trans->transid;
Chris Masone02119d2008-09-05 16:13:11 -04008469
Chris Mason8f18cf12008-04-25 16:53:30 -04008470 cache = kzalloc(sizeof(*cache), GFP_NOFS);
Josef Bacik0f9dd462008-09-23 13:14:11 -04008471 if (!cache)
8472 return -ENOMEM;
Li Zefan34d52cb2011-03-29 13:46:06 +08008473 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
8474 GFP_NOFS);
8475 if (!cache->free_space_ctl) {
8476 kfree(cache);
8477 return -ENOMEM;
8478 }
Josef Bacik0f9dd462008-09-23 13:14:11 -04008479
Chris Masone17cade2008-04-15 15:41:47 -04008480 cache->key.objectid = chunk_offset;
Chris Mason6324fbf2008-03-24 15:01:59 -04008481 cache->key.offset = size;
Yan Zhengd2fb3432008-12-11 16:30:39 -05008482 cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
Josef Bacik96303082009-07-13 21:29:25 -04008483 cache->sectorsize = root->sectorsize;
Josef Bacik0af3d002010-06-21 14:48:16 -04008484 cache->fs_info = root->fs_info;
David Woodhouse53b381b2013-01-29 18:40:14 -05008485 cache->full_stripe_len = btrfs_full_stripe_len(root,
8486 &root->fs_info->mapping_tree,
8487 chunk_offset);
Josef Bacik96303082009-07-13 21:29:25 -04008488
Yan Zhengd2fb3432008-12-11 16:30:39 -05008489 atomic_set(&cache->count, 1);
Chris Masonc286ac42008-07-22 23:06:41 -04008490 spin_lock_init(&cache->lock);
Josef Bacik0f9dd462008-09-23 13:14:11 -04008491 INIT_LIST_HEAD(&cache->list);
Chris Masonfa9c0d792009-04-03 09:47:43 -04008492 INIT_LIST_HEAD(&cache->cluster_list);
Josef Bacikea658ba2012-09-11 16:57:25 -04008493 INIT_LIST_HEAD(&cache->new_bg_list);
Chris Mason0ef3e662008-05-24 14:04:53 -04008494
Li Zefan34d52cb2011-03-29 13:46:06 +08008495 btrfs_init_free_space_ctl(cache);
8496
Chris Mason6324fbf2008-03-24 15:01:59 -04008497 btrfs_set_block_group_used(&cache->item, bytes_used);
Chris Mason6324fbf2008-03-24 15:01:59 -04008498 btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
8499 cache->flags = type;
8500 btrfs_set_block_group_flags(&cache->item, type);
8501
Yan Zheng11833d62009-09-11 16:11:19 -04008502 cache->last_byte_to_unpin = (u64)-1;
Josef Bacik817d52f2009-07-13 21:29:25 -04008503 cache->cached = BTRFS_CACHE_FINISHED;
Josef Bacik835d9742013-03-19 12:13:25 -04008504 ret = exclude_super_stripes(root, cache);
8505 if (ret) {
8506 /*
8507 * We may have excluded something, so call this just in
8508 * case.
8509 */
8510 free_excluded_extents(root, cache);
8511 kfree(cache->free_space_ctl);
8512 kfree(cache);
8513 return ret;
8514 }
Josef Bacik96303082009-07-13 21:29:25 -04008515
Josef Bacik817d52f2009-07-13 21:29:25 -04008516 add_new_free_space(cache, root->fs_info, chunk_offset,
8517 chunk_offset + size);
8518
Yan Zheng11833d62009-09-11 16:11:19 -04008519 free_excluded_extents(root, cache);
8520
Josef Bacik8c579fe2013-04-02 12:40:42 -04008521 ret = btrfs_add_block_group_cache(root->fs_info, cache);
8522 if (ret) {
8523 btrfs_remove_free_space_cache(cache);
8524 btrfs_put_block_group(cache);
8525 return ret;
8526 }
8527
Chris Mason6324fbf2008-03-24 15:01:59 -04008528 ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
8529 &cache->space_info);
Josef Bacik8c579fe2013-04-02 12:40:42 -04008530 if (ret) {
8531 btrfs_remove_free_space_cache(cache);
8532 spin_lock(&root->fs_info->block_group_cache_lock);
8533 rb_erase(&cache->cache_node,
8534 &root->fs_info->block_group_cache_tree);
8535 spin_unlock(&root->fs_info->block_group_cache_lock);
8536 btrfs_put_block_group(cache);
8537 return ret;
8538 }
Li Zefanc7c144d2011-12-07 10:39:22 +08008539 update_global_block_rsv(root->fs_info);
Josef Bacik1b2da372009-09-11 16:11:20 -04008540
8541 spin_lock(&cache->space_info->lock);
Yan, Zhengf0486c62010-05-16 10:46:25 -04008542 cache->space_info->bytes_readonly += cache->bytes_super;
Josef Bacik1b2da372009-09-11 16:11:20 -04008543 spin_unlock(&cache->space_info->lock);
8544
Yan, Zhengb742bb82010-05-16 10:46:24 -04008545 __link_block_group(cache->space_info, cache);
Chris Mason6324fbf2008-03-24 15:01:59 -04008546
Josef Bacikea658ba2012-09-11 16:57:25 -04008547 list_add_tail(&cache->new_bg_list, &trans->new_bgs);
Chris Mason6324fbf2008-03-24 15:01:59 -04008548
Chris Masond18a2c42008-04-04 15:40:00 -04008549 set_avail_alloc_bits(extent_root->fs_info, type);
Chris Mason925baed2008-06-25 16:01:30 -04008550
Chris Mason6324fbf2008-03-24 15:01:59 -04008551 return 0;
8552}
Zheng Yan1a40e232008-09-26 10:09:34 -04008553
Ilya Dryomov10ea00f2012-01-16 22:04:47 +02008554static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
8555{
Ilya Dryomov899c81e2012-03-27 17:09:16 +03008556 u64 extra_flags = chunk_to_extended(flags) &
8557 BTRFS_EXTENDED_PROFILE_MASK;
Ilya Dryomov10ea00f2012-01-16 22:04:47 +02008558
Miao Xiede98ced2013-01-29 10:13:12 +00008559 write_seqlock(&fs_info->profiles_lock);
Ilya Dryomov10ea00f2012-01-16 22:04:47 +02008560 if (flags & BTRFS_BLOCK_GROUP_DATA)
8561 fs_info->avail_data_alloc_bits &= ~extra_flags;
8562 if (flags & BTRFS_BLOCK_GROUP_METADATA)
8563 fs_info->avail_metadata_alloc_bits &= ~extra_flags;
8564 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
8565 fs_info->avail_system_alloc_bits &= ~extra_flags;
Miao Xiede98ced2013-01-29 10:13:12 +00008566 write_sequnlock(&fs_info->profiles_lock);
Ilya Dryomov10ea00f2012-01-16 22:04:47 +02008567}
8568
Zheng Yan1a40e232008-09-26 10:09:34 -04008569int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
8570 struct btrfs_root *root, u64 group_start)
8571{
8572 struct btrfs_path *path;
8573 struct btrfs_block_group_cache *block_group;
Chris Mason44fb5512009-06-04 15:34:51 -04008574 struct btrfs_free_cluster *cluster;
Josef Bacik0af3d002010-06-21 14:48:16 -04008575 struct btrfs_root *tree_root = root->fs_info->tree_root;
Zheng Yan1a40e232008-09-26 10:09:34 -04008576 struct btrfs_key key;
Josef Bacik0af3d002010-06-21 14:48:16 -04008577 struct inode *inode;
Zheng Yan1a40e232008-09-26 10:09:34 -04008578 int ret;
Ilya Dryomov10ea00f2012-01-16 22:04:47 +02008579 int index;
Josef Bacik89a55892010-10-14 14:52:27 -04008580 int factor;
Zheng Yan1a40e232008-09-26 10:09:34 -04008581
Zheng Yan1a40e232008-09-26 10:09:34 -04008582 root = root->fs_info->extent_root;
8583
8584 block_group = btrfs_lookup_block_group(root->fs_info, group_start);
8585 BUG_ON(!block_group);
Yan Zhengc146afa2008-11-12 14:34:12 -05008586 BUG_ON(!block_group->ro);
Zheng Yan1a40e232008-09-26 10:09:34 -04008587
liubo9f7c43c2011-03-07 02:13:33 +00008588 /*
8589 * Free the reserved super bytes from this block group before
8590 * remove it.
8591 */
8592 free_excluded_extents(root, block_group);
8593
Zheng Yan1a40e232008-09-26 10:09:34 -04008594 memcpy(&key, &block_group->key, sizeof(key));
Ilya Dryomov10ea00f2012-01-16 22:04:47 +02008595 index = get_block_group_index(block_group);
Josef Bacik89a55892010-10-14 14:52:27 -04008596 if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP |
8597 BTRFS_BLOCK_GROUP_RAID1 |
8598 BTRFS_BLOCK_GROUP_RAID10))
8599 factor = 2;
8600 else
8601 factor = 1;
Zheng Yan1a40e232008-09-26 10:09:34 -04008602
Chris Mason44fb5512009-06-04 15:34:51 -04008603 /* make sure this block group isn't part of an allocation cluster */
8604 cluster = &root->fs_info->data_alloc_cluster;
8605 spin_lock(&cluster->refill_lock);
8606 btrfs_return_cluster_to_free_space(block_group, cluster);
8607 spin_unlock(&cluster->refill_lock);
8608
8609 /*
8610 * make sure this block group isn't part of a metadata
8611 * allocation cluster
8612 */
8613 cluster = &root->fs_info->meta_alloc_cluster;
8614 spin_lock(&cluster->refill_lock);
8615 btrfs_return_cluster_to_free_space(block_group, cluster);
8616 spin_unlock(&cluster->refill_lock);
8617
Zheng Yan1a40e232008-09-26 10:09:34 -04008618 path = btrfs_alloc_path();
Mark Fashehd8926bb2011-07-13 10:38:47 -07008619 if (!path) {
8620 ret = -ENOMEM;
8621 goto out;
8622 }
Zheng Yan1a40e232008-09-26 10:09:34 -04008623
Ilya Dryomov10b2f342011-10-02 13:56:53 +03008624 inode = lookup_free_space_inode(tree_root, block_group, path);
Josef Bacik0af3d002010-06-21 14:48:16 -04008625 if (!IS_ERR(inode)) {
Tsutomu Itohb5324022011-07-19 07:27:20 +00008626 ret = btrfs_orphan_add(trans, inode);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01008627 if (ret) {
8628 btrfs_add_delayed_iput(inode);
8629 goto out;
8630 }
Josef Bacik0af3d002010-06-21 14:48:16 -04008631 clear_nlink(inode);
8632 /* One for the block groups ref */
8633 spin_lock(&block_group->lock);
8634 if (block_group->iref) {
8635 block_group->iref = 0;
8636 block_group->inode = NULL;
8637 spin_unlock(&block_group->lock);
8638 iput(inode);
8639 } else {
8640 spin_unlock(&block_group->lock);
8641 }
8642 /* One for our lookup ref */
Josef Bacik455757c2011-09-19 12:26:24 -04008643 btrfs_add_delayed_iput(inode);
Josef Bacik0af3d002010-06-21 14:48:16 -04008644 }
8645
8646 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
8647 key.offset = block_group->key.objectid;
8648 key.type = 0;
8649
8650 ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
8651 if (ret < 0)
8652 goto out;
8653 if (ret > 0)
David Sterbab3b4aa72011-04-21 01:20:15 +02008654 btrfs_release_path(path);
Josef Bacik0af3d002010-06-21 14:48:16 -04008655 if (ret == 0) {
8656 ret = btrfs_del_item(trans, tree_root, path);
8657 if (ret)
8658 goto out;
David Sterbab3b4aa72011-04-21 01:20:15 +02008659 btrfs_release_path(path);
Josef Bacik0af3d002010-06-21 14:48:16 -04008660 }
8661
Yan Zheng3dfdb932009-01-21 10:49:16 -05008662 spin_lock(&root->fs_info->block_group_cache_lock);
Zheng Yan1a40e232008-09-26 10:09:34 -04008663 rb_erase(&block_group->cache_node,
8664 &root->fs_info->block_group_cache_tree);
Liu Boa1897fd2012-12-27 09:01:23 +00008665
8666 if (root->fs_info->first_logical_byte == block_group->key.objectid)
8667 root->fs_info->first_logical_byte = (u64)-1;
Yan Zheng3dfdb932009-01-21 10:49:16 -05008668 spin_unlock(&root->fs_info->block_group_cache_lock);
Josef Bacik817d52f2009-07-13 21:29:25 -04008669
Josef Bacik80eb2342008-10-29 14:49:05 -04008670 down_write(&block_group->space_info->groups_sem);
Chris Mason44fb5512009-06-04 15:34:51 -04008671 /*
8672 * we must use list_del_init so people can check to see if they
8673 * are still on the list after taking the semaphore
8674 */
8675 list_del_init(&block_group->list);
Ilya Dryomov10ea00f2012-01-16 22:04:47 +02008676 if (list_empty(&block_group->space_info->block_groups[index]))
8677 clear_avail_alloc_bits(root->fs_info, block_group->flags);
Josef Bacik80eb2342008-10-29 14:49:05 -04008678 up_write(&block_group->space_info->groups_sem);
Zheng Yan1a40e232008-09-26 10:09:34 -04008679
Josef Bacik817d52f2009-07-13 21:29:25 -04008680 if (block_group->cached == BTRFS_CACHE_STARTED)
Yan Zheng11833d62009-09-11 16:11:19 -04008681 wait_block_group_cache_done(block_group);
Josef Bacik817d52f2009-07-13 21:29:25 -04008682
8683 btrfs_remove_free_space_cache(block_group);
8684
Yan Zhengc146afa2008-11-12 14:34:12 -05008685 spin_lock(&block_group->space_info->lock);
8686 block_group->space_info->total_bytes -= block_group->key.offset;
8687 block_group->space_info->bytes_readonly -= block_group->key.offset;
Josef Bacik89a55892010-10-14 14:52:27 -04008688 block_group->space_info->disk_total -= block_group->key.offset * factor;
Yan Zhengc146afa2008-11-12 14:34:12 -05008689 spin_unlock(&block_group->space_info->lock);
Chris Mason283bb192009-07-24 16:30:55 -04008690
Josef Bacik0af3d002010-06-21 14:48:16 -04008691 memcpy(&key, &block_group->key, sizeof(key));
8692
Chris Mason283bb192009-07-24 16:30:55 -04008693 btrfs_clear_space_info_full(root->fs_info);
Yan Zhengc146afa2008-11-12 14:34:12 -05008694
Chris Masonfa9c0d792009-04-03 09:47:43 -04008695 btrfs_put_block_group(block_group);
8696 btrfs_put_block_group(block_group);
Zheng Yan1a40e232008-09-26 10:09:34 -04008697
8698 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
8699 if (ret > 0)
8700 ret = -EIO;
8701 if (ret < 0)
8702 goto out;
8703
8704 ret = btrfs_del_item(trans, root, path);
8705out:
8706 btrfs_free_path(path);
8707 return ret;
8708}
liuboacce9522011-01-06 19:30:25 +08008709
liuboc59021f2011-03-07 02:13:14 +00008710int btrfs_init_space_info(struct btrfs_fs_info *fs_info)
8711{
8712 struct btrfs_space_info *space_info;
liubo1aba86d2011-04-08 08:44:37 +00008713 struct btrfs_super_block *disk_super;
8714 u64 features;
8715 u64 flags;
8716 int mixed = 0;
liuboc59021f2011-03-07 02:13:14 +00008717 int ret;
8718
David Sterba6c417612011-04-13 15:41:04 +02008719 disk_super = fs_info->super_copy;
liubo1aba86d2011-04-08 08:44:37 +00008720 if (!btrfs_super_root(disk_super))
8721 return 1;
liuboc59021f2011-03-07 02:13:14 +00008722
liubo1aba86d2011-04-08 08:44:37 +00008723 features = btrfs_super_incompat_flags(disk_super);
8724 if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
8725 mixed = 1;
liuboc59021f2011-03-07 02:13:14 +00008726
liubo1aba86d2011-04-08 08:44:37 +00008727 flags = BTRFS_BLOCK_GROUP_SYSTEM;
8728 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
liuboc59021f2011-03-07 02:13:14 +00008729 if (ret)
liubo1aba86d2011-04-08 08:44:37 +00008730 goto out;
liuboc59021f2011-03-07 02:13:14 +00008731
liubo1aba86d2011-04-08 08:44:37 +00008732 if (mixed) {
8733 flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA;
8734 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8735 } else {
8736 flags = BTRFS_BLOCK_GROUP_METADATA;
8737 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8738 if (ret)
8739 goto out;
8740
8741 flags = BTRFS_BLOCK_GROUP_DATA;
8742 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8743 }
8744out:
liuboc59021f2011-03-07 02:13:14 +00008745 return ret;
8746}
8747
liuboacce9522011-01-06 19:30:25 +08008748int btrfs_error_unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
8749{
8750 return unpin_extent_range(root, start, end);
8751}
8752
8753int btrfs_error_discard_extent(struct btrfs_root *root, u64 bytenr,
Li Dongyang5378e602011-03-24 10:24:27 +00008754 u64 num_bytes, u64 *actual_bytes)
liuboacce9522011-01-06 19:30:25 +08008755{
Li Dongyang5378e602011-03-24 10:24:27 +00008756 return btrfs_discard_extent(root, bytenr, num_bytes, actual_bytes);
liuboacce9522011-01-06 19:30:25 +08008757}
Li Dongyangf7039b12011-03-24 10:24:28 +00008758
8759int btrfs_trim_fs(struct btrfs_root *root, struct fstrim_range *range)
8760{
8761 struct btrfs_fs_info *fs_info = root->fs_info;
8762 struct btrfs_block_group_cache *cache = NULL;
8763 u64 group_trimmed;
8764 u64 start;
8765 u64 end;
8766 u64 trimmed = 0;
Liu Bo2cac13e2012-02-09 18:17:41 +08008767 u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
Li Dongyangf7039b12011-03-24 10:24:28 +00008768 int ret = 0;
8769
Liu Bo2cac13e2012-02-09 18:17:41 +08008770 /*
8771 * try to trim all FS space, our block group may start from non-zero.
8772 */
8773 if (range->len == total_bytes)
8774 cache = btrfs_lookup_first_block_group(fs_info, range->start);
8775 else
8776 cache = btrfs_lookup_block_group(fs_info, range->start);
Li Dongyangf7039b12011-03-24 10:24:28 +00008777
8778 while (cache) {
8779 if (cache->key.objectid >= (range->start + range->len)) {
8780 btrfs_put_block_group(cache);
8781 break;
8782 }
8783
8784 start = max(range->start, cache->key.objectid);
8785 end = min(range->start + range->len,
8786 cache->key.objectid + cache->key.offset);
8787
8788 if (end - start >= range->minlen) {
8789 if (!block_group_cache_done(cache)) {
Liu Bof6373bf2012-12-27 09:01:18 +00008790 ret = cache_block_group(cache, 0);
Josef Bacik1be41b72013-06-12 13:56:06 -04008791 if (ret) {
8792 btrfs_put_block_group(cache);
8793 break;
8794 }
8795 ret = wait_block_group_cache_done(cache);
8796 if (ret) {
8797 btrfs_put_block_group(cache);
8798 break;
8799 }
Li Dongyangf7039b12011-03-24 10:24:28 +00008800 }
8801 ret = btrfs_trim_block_group(cache,
8802 &group_trimmed,
8803 start,
8804 end,
8805 range->minlen);
8806
8807 trimmed += group_trimmed;
8808 if (ret) {
8809 btrfs_put_block_group(cache);
8810 break;
8811 }
8812 }
8813
8814 cache = next_block_group(fs_info->tree_root, cache);
8815 }
8816
8817 range->len = trimmed;
8818 return ret;
8819}