ext4: Use EXT4_GROUP_INFO_NEED_INIT_BIT during resize
[linux-block.git] / fs / ext4 / mballoc.h
CommitLineData
8f6e39a7
MC
1/*
2 * fs/ext4/mballoc.h
3 *
4 * Written by: Alex Tomas <alex@clusterfs.com>
5 *
6 */
7#ifndef _EXT4_MBALLOC_H
8#define _EXT4_MBALLOC_H
9
10#include <linux/time.h>
11#include <linux/fs.h>
12#include <linux/namei.h>
13#include <linux/quotaops.h>
14#include <linux/buffer_head.h>
15#include <linux/module.h>
16#include <linux/swap.h>
17#include <linux/proc_fs.h>
18#include <linux/pagemap.h>
19#include <linux/seq_file.h>
20#include <linux/version.h>
8a0aba73
TT
21#include <linux/blkdev.h>
22#include <linux/marker.h>
920313a7 23#include <linux/mutex.h>
8f6e39a7
MC
24#include "ext4_jbd2.h"
25#include "ext4.h"
26#include "group.h"
27
28/*
29 * with AGGRESSIVE_CHECK allocator runs consistency checks over
30 * structures. these checks slow things down a lot
31 */
32#define AGGRESSIVE_CHECK__
33
34/*
35 * with DOUBLE_CHECK defined mballoc creates persistent in-core
36 * bitmaps, maintains and uses them to check for double allocations
37 */
38#define DOUBLE_CHECK__
39
40/*
41 */
42#define MB_DEBUG__
43#ifdef MB_DEBUG
44#define mb_debug(fmt, a...) printk(fmt, ##a)
45#else
46#define mb_debug(fmt, a...)
47#endif
48
49/*
50 * with EXT4_MB_HISTORY mballoc stores last N allocations in memory
51 * and you can monitor it in /proc/fs/ext4/<dev>/mb_history
52 */
53#define EXT4_MB_HISTORY
54#define EXT4_MB_HISTORY_ALLOC 1 /* allocation */
55#define EXT4_MB_HISTORY_PREALLOC 2 /* preallocated blocks used */
56#define EXT4_MB_HISTORY_DISCARD 4 /* preallocation discarded */
57#define EXT4_MB_HISTORY_FREE 8 /* free */
58
59#define EXT4_MB_HISTORY_DEFAULT (EXT4_MB_HISTORY_ALLOC | \
60 EXT4_MB_HISTORY_PREALLOC)
61
62/*
63 * How long mballoc can look for a best extent (in found extents)
64 */
65#define MB_DEFAULT_MAX_TO_SCAN 200
66
67/*
68 * How long mballoc must look for a best extent
69 */
70#define MB_DEFAULT_MIN_TO_SCAN 10
71
72/*
73 * How many groups mballoc will scan looking for the best chunk
74 */
75#define MB_DEFAULT_MAX_GROUPS_TO_SCAN 5
76
77/*
78 * with 'ext4_mb_stats' allocator will collect stats that will be
79 * shown at umount. The collecting costs though!
80 */
81#define MB_DEFAULT_STATS 1
82
83/*
84 * files smaller than MB_DEFAULT_STREAM_THRESHOLD are served
85 * by the stream allocator, which purpose is to pack requests
86 * as close each to other as possible to produce smooth I/O traffic
87 * We use locality group prealloc space for stream request.
88 * We can tune the same via /proc/fs/ext4/<parition>/stream_req
89 */
90#define MB_DEFAULT_STREAM_THRESHOLD 16 /* 64K */
91
92/*
93 * for which requests use 2^N search using buddies
94 */
95#define MB_DEFAULT_ORDER2_REQS 2
96
97/*
98 * default group prealloc size 512 blocks
99 */
100#define MB_DEFAULT_GROUP_PREALLOC 512
101
102static struct kmem_cache *ext4_pspace_cachep;
103static struct kmem_cache *ext4_ac_cachep;
c894058d 104static struct kmem_cache *ext4_free_ext_cachep;
8f6e39a7 105
c894058d
AK
106struct ext4_free_data {
107 /* this links the free block information from group_info */
108 struct rb_node node;
8f6e39a7 109
c894058d 110 /* this links the free block information from ext4_sb_info */
8f6e39a7 111 struct list_head list;
c894058d
AK
112
113 /* group which free block extent belongs */
114 ext4_group_t group;
115
116 /* free block extent */
117 ext4_grpblk_t start_blk;
118 ext4_grpblk_t count;
119
120 /* transaction which freed this extent */
121 tid_t t_tid;
8f6e39a7
MC
122};
123
124struct ext4_group_info {
125 unsigned long bb_state;
c894058d 126 struct rb_root bb_free_root;
8f6e39a7
MC
127 unsigned short bb_first_free;
128 unsigned short bb_free;
129 unsigned short bb_fragments;
130 struct list_head bb_prealloc_list;
131#ifdef DOUBLE_CHECK
132 void *bb_bitmap;
133#endif
920313a7 134 struct rw_semaphore alloc_sem;
8f6e39a7
MC
135 unsigned short bb_counters[];
136};
137
138#define EXT4_GROUP_INFO_NEED_INIT_BIT 0
139#define EXT4_GROUP_INFO_LOCKED_BIT 1
140
141#define EXT4_MB_GRP_NEED_INIT(grp) \
142 (test_bit(EXT4_GROUP_INFO_NEED_INIT_BIT, &((grp)->bb_state)))
143
144
145struct ext4_prealloc_space {
146 struct list_head pa_inode_list;
147 struct list_head pa_group_list;
148 union {
149 struct list_head pa_tmp_list;
150 struct rcu_head pa_rcu;
151 } u;
152 spinlock_t pa_lock;
153 atomic_t pa_count;
154 unsigned pa_deleted;
155 ext4_fsblk_t pa_pstart; /* phys. block */
156 ext4_lblk_t pa_lstart; /* log. block */
157 unsigned short pa_len; /* len of preallocated chunk */
158 unsigned short pa_free; /* how many blocks are free */
159 unsigned short pa_linear; /* consumed in one direction
160 * strictly, for grp prealloc */
161 spinlock_t *pa_obj_lock;
162 struct inode *pa_inode; /* hack, for history only */
163};
164
165
166struct ext4_free_extent {
167 ext4_lblk_t fe_logical;
168 ext4_grpblk_t fe_start;
169 ext4_group_t fe_group;
170 int fe_len;
171};
172
173/*
174 * Locality group:
175 * we try to group all related changes together
176 * so that writeback can flush/allocate them together as well
6be2ded1
AK
177 * Size of lg_prealloc_list hash is determined by MB_DEFAULT_GROUP_PREALLOC
178 * (512). We store prealloc space into the hash based on the pa_free blocks
179 * order value.ie, fls(pa_free)-1;
8f6e39a7 180 */
6be2ded1 181#define PREALLOC_TB_SIZE 10
8f6e39a7
MC
182struct ext4_locality_group {
183 /* for allocator */
6be2ded1
AK
184 /* to serialize allocates */
185 struct mutex lg_mutex;
186 /* list of preallocations */
187 struct list_head lg_prealloc_list[PREALLOC_TB_SIZE];
8f6e39a7
MC
188 spinlock_t lg_prealloc_lock;
189};
190
191struct ext4_allocation_context {
192 struct inode *ac_inode;
193 struct super_block *ac_sb;
194
195 /* original request */
196 struct ext4_free_extent ac_o_ex;
197
198 /* goal request (after normalization) */
199 struct ext4_free_extent ac_g_ex;
200
201 /* the best found extent */
202 struct ext4_free_extent ac_b_ex;
203
204 /* copy of the bext found extent taken before preallocation efforts */
205 struct ext4_free_extent ac_f_ex;
206
207 /* number of iterations done. we have to track to limit searching */
208 unsigned long ac_ex_scanned;
209 __u16 ac_groups_scanned;
210 __u16 ac_found;
211 __u16 ac_tail;
212 __u16 ac_buddy;
213 __u16 ac_flags; /* allocation hints */
214 __u8 ac_status;
215 __u8 ac_criteria;
216 __u8 ac_repeats;
217 __u8 ac_2order; /* if request is to allocate 2^N blocks and
218 * N > 0, the field stores N, otherwise 0 */
219 __u8 ac_op; /* operation, for history only */
220 struct page *ac_bitmap_page;
221 struct page *ac_buddy_page;
222 struct ext4_prealloc_space *ac_pa;
223 struct ext4_locality_group *ac_lg;
224};
225
226#define AC_STATUS_CONTINUE 1
227#define AC_STATUS_FOUND 2
228#define AC_STATUS_BREAK 3
229
230struct ext4_mb_history {
231 struct ext4_free_extent orig; /* orig allocation */
232 struct ext4_free_extent goal; /* goal allocation */
233 struct ext4_free_extent result; /* result allocation */
234 unsigned pid;
235 unsigned ino;
236 __u16 found; /* how many extents have been found */
237 __u16 groups; /* how many groups have been scanned */
238 __u16 tail; /* what tail broke some buddy */
239 __u16 buddy; /* buddy the tail ^^^ broke */
240 __u16 flags;
241 __u8 cr:3; /* which phase the result extent was found at */
242 __u8 op:4;
243 __u8 merged:1;
244};
245
246struct ext4_buddy {
247 struct page *bd_buddy_page;
248 void *bd_buddy;
249 struct page *bd_bitmap_page;
250 void *bd_bitmap;
251 struct ext4_group_info *bd_info;
252 struct super_block *bd_sb;
253 __u16 bd_blkbits;
254 ext4_group_t bd_group;
920313a7 255 struct rw_semaphore *alloc_semp;
8f6e39a7
MC
256};
257#define EXT4_MB_BITMAP(e4b) ((e4b)->bd_bitmap)
258#define EXT4_MB_BUDDY(e4b) ((e4b)->bd_buddy)
259
260#ifndef EXT4_MB_HISTORY
261static inline void ext4_mb_store_history(struct ext4_allocation_context *ac)
262{
263 return;
264}
265#else
266static void ext4_mb_store_history(struct ext4_allocation_context *ac);
267#endif
268
269#define in_range(b, first, len) ((b) >= (first) && (b) <= (first) + (len) - 1)
270
8f6e39a7
MC
271struct buffer_head *read_block_bitmap(struct super_block *, ext4_group_t);
272
273static void ext4_mb_generate_from_pa(struct super_block *sb, void *bitmap,
274 ext4_group_t group);
8f6e39a7
MC
275static void ext4_mb_return_to_preallocation(struct inode *inode,
276 struct ext4_buddy *e4b, sector_t block,
277 int count);
278static void ext4_mb_put_pa(struct ext4_allocation_context *,
279 struct super_block *, struct ext4_prealloc_space *pa);
280static int ext4_mb_init_per_dev_proc(struct super_block *sb);
281static int ext4_mb_destroy_per_dev_proc(struct super_block *sb);
3e624fc7 282static void release_blocks_on_commit(journal_t *journal, transaction_t *txn);
8f6e39a7
MC
283
284
285static inline void ext4_lock_group(struct super_block *sb, ext4_group_t group)
286{
287 struct ext4_group_info *grinfo = ext4_get_group_info(sb, group);
288
289 bit_spin_lock(EXT4_GROUP_INFO_LOCKED_BIT, &(grinfo->bb_state));
290}
291
292static inline void ext4_unlock_group(struct super_block *sb,
293 ext4_group_t group)
294{
295 struct ext4_group_info *grinfo = ext4_get_group_info(sb, group);
296
297 bit_spin_unlock(EXT4_GROUP_INFO_LOCKED_BIT, &(grinfo->bb_state));
298}
299
300static inline int ext4_is_group_locked(struct super_block *sb,
301 ext4_group_t group)
302{
303 struct ext4_group_info *grinfo = ext4_get_group_info(sb, group);
304
305 return bit_spin_is_locked(EXT4_GROUP_INFO_LOCKED_BIT,
306 &(grinfo->bb_state));
307}
308
309static ext4_fsblk_t ext4_grp_offs_to_block(struct super_block *sb,
310 struct ext4_free_extent *fex)
311{
312 ext4_fsblk_t block;
313
314 block = (ext4_fsblk_t) fex->fe_group * EXT4_BLOCKS_PER_GROUP(sb)
315 + fex->fe_start
316 + le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block);
317 return block;
318}
319#endif