2
* Copyright (C) 2007 Oracle. All rights reserved.
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.
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.
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.
19
#include <linux/delay.h>
20
#include <linux/kthread.h>
21
#include <linux/pagemap.h>
25
#include "free-space-cache.h"
26
#include "inode-map.h"
27
#include "transaction.h"
29
static int caching_kthread(void *data)
31
struct btrfs_root *root = data;
32
struct btrfs_fs_info *fs_info = root->fs_info;
33
struct btrfs_free_space_ctl *ctl = root->free_ino_ctl;
35
struct btrfs_path *path;
36
struct extent_buffer *leaf;
41
if (!btrfs_test_opt(root, INODE_MAP_CACHE))
44
path = btrfs_alloc_path();
48
/* Since the commit root is read-only, we can safely skip locking. */
49
path->skip_locking = 1;
50
path->search_commit_root = 1;
53
key.objectid = BTRFS_FIRST_FREE_OBJECTID;
55
key.type = BTRFS_INODE_ITEM_KEY;
57
/* need to make sure the commit_root doesn't disappear */
58
mutex_lock(&root->fs_commit_mutex);
60
ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
65
if (btrfs_fs_closing(fs_info))
68
leaf = path->nodes[0];
69
slot = path->slots[0];
70
if (slot >= btrfs_header_nritems(leaf)) {
71
ret = btrfs_next_leaf(root, path);
78
btrfs_transaction_in_commit(fs_info)) {
79
leaf = path->nodes[0];
81
if (btrfs_header_nritems(leaf) == 0) {
87
* Save the key so we can advances forward
90
btrfs_item_key_to_cpu(leaf, &key, 0);
91
btrfs_release_path(path);
92
root->cache_progress = last;
93
mutex_unlock(&root->fs_commit_mutex);
100
btrfs_item_key_to_cpu(leaf, &key, slot);
102
if (key.type != BTRFS_INODE_ITEM_KEY)
105
if (key.objectid >= root->highest_objectid)
108
if (last != (u64)-1 && last + 1 != key.objectid) {
109
__btrfs_add_free_space(ctl, last + 1,
110
key.objectid - last - 1);
111
wake_up(&root->cache_wait);
119
if (last < root->highest_objectid - 1) {
120
__btrfs_add_free_space(ctl, last + 1,
121
root->highest_objectid - last - 1);
124
spin_lock(&root->cache_lock);
125
root->cached = BTRFS_CACHE_FINISHED;
126
spin_unlock(&root->cache_lock);
128
root->cache_progress = (u64)-1;
129
btrfs_unpin_free_ino(root);
131
wake_up(&root->cache_wait);
132
mutex_unlock(&root->fs_commit_mutex);
134
btrfs_free_path(path);
139
static void start_caching(struct btrfs_root *root)
141
struct btrfs_free_space_ctl *ctl = root->free_ino_ctl;
142
struct task_struct *tsk;
146
if (!btrfs_test_opt(root, INODE_MAP_CACHE))
149
spin_lock(&root->cache_lock);
150
if (root->cached != BTRFS_CACHE_NO) {
151
spin_unlock(&root->cache_lock);
155
root->cached = BTRFS_CACHE_STARTED;
156
spin_unlock(&root->cache_lock);
158
ret = load_free_ino_cache(root->fs_info, root);
160
spin_lock(&root->cache_lock);
161
root->cached = BTRFS_CACHE_FINISHED;
162
spin_unlock(&root->cache_lock);
167
* It can be quite time-consuming to fill the cache by searching
168
* through the extent tree, and this can keep ino allocation path
169
* waiting. Therefore at start we quickly find out the highest
170
* inode number and we know we can use inode numbers which fall in
171
* [highest_ino + 1, BTRFS_LAST_FREE_OBJECTID].
173
ret = btrfs_find_free_objectid(root, &objectid);
174
if (!ret && objectid <= BTRFS_LAST_FREE_OBJECTID) {
175
__btrfs_add_free_space(ctl, objectid,
176
BTRFS_LAST_FREE_OBJECTID - objectid + 1);
179
tsk = kthread_run(caching_kthread, root, "btrfs-ino-cache-%llu\n",
180
root->root_key.objectid);
184
int btrfs_find_free_ino(struct btrfs_root *root, u64 *objectid)
186
if (!btrfs_test_opt(root, INODE_MAP_CACHE))
187
return btrfs_find_free_objectid(root, objectid);
190
*objectid = btrfs_find_ino_for_alloc(root);
197
wait_event(root->cache_wait,
198
root->cached == BTRFS_CACHE_FINISHED ||
199
root->free_ino_ctl->free_space > 0);
201
if (root->cached == BTRFS_CACHE_FINISHED &&
202
root->free_ino_ctl->free_space == 0)
208
void btrfs_return_ino(struct btrfs_root *root, u64 objectid)
210
struct btrfs_free_space_ctl *ctl = root->free_ino_ctl;
211
struct btrfs_free_space_ctl *pinned = root->free_ino_pinned;
213
if (!btrfs_test_opt(root, INODE_MAP_CACHE))
217
if (root->cached == BTRFS_CACHE_FINISHED) {
218
__btrfs_add_free_space(ctl, objectid, 1);
221
* If we are in the process of caching free ino chunks,
222
* to avoid adding the same inode number to the free_ino
223
* tree twice due to cross transaction, we'll leave it
224
* in the pinned tree until a transaction is committed
225
* or the caching work is done.
228
mutex_lock(&root->fs_commit_mutex);
229
spin_lock(&root->cache_lock);
230
if (root->cached == BTRFS_CACHE_FINISHED) {
231
spin_unlock(&root->cache_lock);
232
mutex_unlock(&root->fs_commit_mutex);
235
spin_unlock(&root->cache_lock);
239
if (objectid <= root->cache_progress ||
240
objectid > root->highest_objectid)
241
__btrfs_add_free_space(ctl, objectid, 1);
243
__btrfs_add_free_space(pinned, objectid, 1);
245
mutex_unlock(&root->fs_commit_mutex);
250
* When a transaction is committed, we'll move those inode numbers which
251
* are smaller than root->cache_progress from pinned tree to free_ino tree,
252
* and others will just be dropped, because the commit root we were
253
* searching has changed.
255
* Must be called with root->fs_commit_mutex held
257
void btrfs_unpin_free_ino(struct btrfs_root *root)
259
struct btrfs_free_space_ctl *ctl = root->free_ino_ctl;
260
struct rb_root *rbroot = &root->free_ino_pinned->free_space_offset;
261
struct btrfs_free_space *info;
265
if (!btrfs_test_opt(root, INODE_MAP_CACHE))
269
n = rb_first(rbroot);
273
info = rb_entry(n, struct btrfs_free_space, offset_index);
274
BUG_ON(info->bitmap);
276
if (info->offset > root->cache_progress)
278
else if (info->offset + info->bytes > root->cache_progress)
279
count = root->cache_progress - info->offset + 1;
283
__btrfs_add_free_space(ctl, info->offset, count);
285
rb_erase(&info->offset_index, rbroot);
290
#define INIT_THRESHOLD (((1024 * 32) / 2) / sizeof(struct btrfs_free_space))
291
#define INODES_PER_BITMAP (PAGE_CACHE_SIZE * 8)
294
* The goal is to keep the memory used by the free_ino tree won't
295
* exceed the memory if we use bitmaps only.
297
static void recalculate_thresholds(struct btrfs_free_space_ctl *ctl)
299
struct btrfs_free_space *info;
304
n = rb_last(&ctl->free_space_offset);
306
ctl->extents_thresh = INIT_THRESHOLD;
309
info = rb_entry(n, struct btrfs_free_space, offset_index);
312
* Find the maximum inode number in the filesystem. Note we
313
* ignore the fact that this can be a bitmap, because we are
314
* not doing precise calculation.
316
max_ino = info->bytes - 1;
318
max_bitmaps = ALIGN(max_ino, INODES_PER_BITMAP) / INODES_PER_BITMAP;
319
if (max_bitmaps <= ctl->total_bitmaps) {
320
ctl->extents_thresh = 0;
324
ctl->extents_thresh = (max_bitmaps - ctl->total_bitmaps) *
325
PAGE_CACHE_SIZE / sizeof(*info);
329
* We don't fall back to bitmap, if we are below the extents threshold
330
* or this chunk of inode numbers is a big one.
332
static bool use_bitmap(struct btrfs_free_space_ctl *ctl,
333
struct btrfs_free_space *info)
335
if (ctl->free_extents < ctl->extents_thresh ||
336
info->bytes > INODES_PER_BITMAP / 10)
342
static struct btrfs_free_space_op free_ino_op = {
343
.recalc_thresholds = recalculate_thresholds,
344
.use_bitmap = use_bitmap,
347
static void pinned_recalc_thresholds(struct btrfs_free_space_ctl *ctl)
351
static bool pinned_use_bitmap(struct btrfs_free_space_ctl *ctl,
352
struct btrfs_free_space *info)
355
* We always use extents for two reasons:
357
* - The pinned tree is only used during the process of caching
359
* - Make code simpler. See btrfs_unpin_free_ino().
364
static struct btrfs_free_space_op pinned_free_ino_op = {
365
.recalc_thresholds = pinned_recalc_thresholds,
366
.use_bitmap = pinned_use_bitmap,
369
void btrfs_init_free_ino_ctl(struct btrfs_root *root)
371
struct btrfs_free_space_ctl *ctl = root->free_ino_ctl;
372
struct btrfs_free_space_ctl *pinned = root->free_ino_pinned;
374
spin_lock_init(&ctl->tree_lock);
378
ctl->op = &free_ino_op;
381
* Initially we allow to use 16K of ram to cache chunks of
382
* inode numbers before we resort to bitmaps. This is somewhat
383
* arbitrary, but it will be adjusted in runtime.
385
ctl->extents_thresh = INIT_THRESHOLD;
387
spin_lock_init(&pinned->tree_lock);
390
pinned->private = NULL;
391
pinned->extents_thresh = 0;
392
pinned->op = &pinned_free_ino_op;
395
int btrfs_save_ino_cache(struct btrfs_root *root,
396
struct btrfs_trans_handle *trans)
398
struct btrfs_free_space_ctl *ctl = root->free_ino_ctl;
399
struct btrfs_path *path;
401
struct btrfs_block_rsv *rsv;
408
/* only fs tree and subvol/snap needs ino cache */
409
if (root->root_key.objectid != BTRFS_FS_TREE_OBJECTID &&
410
(root->root_key.objectid < BTRFS_FIRST_FREE_OBJECTID ||
411
root->root_key.objectid > BTRFS_LAST_FREE_OBJECTID))
414
/* Don't save inode cache if we are deleting this root */
415
if (btrfs_root_refs(&root->root_item) == 0 &&
416
root != root->fs_info->tree_root)
419
if (!btrfs_test_opt(root, INODE_MAP_CACHE))
422
path = btrfs_alloc_path();
426
rsv = trans->block_rsv;
427
trans->block_rsv = &root->fs_info->trans_block_rsv;
429
num_bytes = trans->bytes_reserved;
431
* 1 item for inode item insertion if need
432
* 3 items for inode item update (in the worst case)
433
* 1 item for free space object
434
* 3 items for pre-allocation
436
trans->bytes_reserved = btrfs_calc_trans_metadata_size(root, 8);
437
ret = btrfs_block_rsv_add_noflush(root, trans->block_rsv,
438
trans->bytes_reserved);
442
inode = lookup_free_ino_inode(root, path);
443
if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
444
ret = PTR_ERR(inode);
452
ret = create_free_ino_inode(root, trans, path);
458
BTRFS_I(inode)->generation = 0;
459
ret = btrfs_update_inode(trans, root, inode);
462
if (i_size_read(inode) > 0) {
463
ret = btrfs_truncate_free_space_cache(root, trans, path, inode);
468
spin_lock(&root->cache_lock);
469
if (root->cached != BTRFS_CACHE_FINISHED) {
471
spin_unlock(&root->cache_lock);
474
spin_unlock(&root->cache_lock);
476
spin_lock(&ctl->tree_lock);
477
prealloc = sizeof(struct btrfs_free_space) * ctl->free_extents;
478
prealloc = ALIGN(prealloc, PAGE_CACHE_SIZE);
479
prealloc += ctl->total_bitmaps * PAGE_CACHE_SIZE;
480
spin_unlock(&ctl->tree_lock);
482
/* Just to make sure we have enough space */
483
prealloc += 8 * PAGE_CACHE_SIZE;
485
ret = btrfs_delalloc_reserve_space(inode, prealloc);
489
ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, prealloc,
490
prealloc, prealloc, &alloc_hint);
492
btrfs_delalloc_release_space(inode, prealloc);
495
btrfs_free_reserved_data_space(inode, prealloc);
497
ret = btrfs_write_out_ino_cache(root, trans, path);
501
btrfs_block_rsv_release(root, trans->block_rsv, trans->bytes_reserved);
503
trans->block_rsv = rsv;
504
trans->bytes_reserved = num_bytes;
506
btrfs_free_path(path);
510
static int btrfs_find_highest_objectid(struct btrfs_root *root, u64 *objectid)
512
struct btrfs_path *path;
514
struct extent_buffer *l;
515
struct btrfs_key search_key;
516
struct btrfs_key found_key;
519
path = btrfs_alloc_path();
523
search_key.objectid = BTRFS_LAST_FREE_OBJECTID;
524
search_key.type = -1;
525
search_key.offset = (u64)-1;
526
ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
530
if (path->slots[0] > 0) {
531
slot = path->slots[0] - 1;
533
btrfs_item_key_to_cpu(l, &found_key, slot);
534
*objectid = max_t(u64, found_key.objectid,
535
BTRFS_FIRST_FREE_OBJECTID - 1);
537
*objectid = BTRFS_FIRST_FREE_OBJECTID - 1;
541
btrfs_free_path(path);
545
int btrfs_find_free_objectid(struct btrfs_root *root, u64 *objectid)
548
mutex_lock(&root->objectid_mutex);
550
if (unlikely(root->highest_objectid < BTRFS_FIRST_FREE_OBJECTID)) {
551
ret = btrfs_find_highest_objectid(root,
552
&root->highest_objectid);
557
if (unlikely(root->highest_objectid >= BTRFS_LAST_FREE_OBJECTID)) {
562
*objectid = ++root->highest_objectid;
565
mutex_unlock(&root->objectid_mutex);