blob: 997f78fff1297ea1f20cee77ee0befb28ef0dfd7 [file] [log] [blame]
Mingming Cao8f6e39a2008-04-29 22:01:31 -04001/*
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>
Theodore Ts'o8a0aba72008-10-16 10:06:27 -040021#include <linux/blkdev.h>
22#include <linux/marker.h>
Aneesh Kumar K.V920313a2009-01-05 21:36:19 -050023#include <linux/mutex.h>
Mingming Cao8f6e39a2008-04-29 22:01:31 -040024#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
Mingming Cao8f6e39a2008-04-29 22:01:31 -0400102
Aneesh Kumar K.Vc8940582008-10-16 10:14:27 -0400103struct ext4_free_data {
104 /* this links the free block information from group_info */
105 struct rb_node node;
Mingming Cao8f6e39a2008-04-29 22:01:31 -0400106
Aneesh Kumar K.Vc8940582008-10-16 10:14:27 -0400107 /* this links the free block information from ext4_sb_info */
Mingming Cao8f6e39a2008-04-29 22:01:31 -0400108 struct list_head list;
Aneesh Kumar K.Vc8940582008-10-16 10:14:27 -0400109
110 /* group which free block extent belongs */
111 ext4_group_t group;
112
113 /* free block extent */
114 ext4_grpblk_t start_blk;
115 ext4_grpblk_t count;
116
117 /* transaction which freed this extent */
118 tid_t t_tid;
Mingming Cao8f6e39a2008-04-29 22:01:31 -0400119};
120
121struct ext4_group_info {
122 unsigned long bb_state;
Aneesh Kumar K.Vc8940582008-10-16 10:14:27 -0400123 struct rb_root bb_free_root;
Mingming Cao8f6e39a2008-04-29 22:01:31 -0400124 unsigned short bb_first_free;
125 unsigned short bb_free;
126 unsigned short bb_fragments;
127 struct list_head bb_prealloc_list;
128#ifdef DOUBLE_CHECK
129 void *bb_bitmap;
130#endif
Aneesh Kumar K.V920313a2009-01-05 21:36:19 -0500131 struct rw_semaphore alloc_sem;
Mingming Cao8f6e39a2008-04-29 22:01:31 -0400132 unsigned short bb_counters[];
133};
134
135#define EXT4_GROUP_INFO_NEED_INIT_BIT 0
136#define EXT4_GROUP_INFO_LOCKED_BIT 1
137
138#define EXT4_MB_GRP_NEED_INIT(grp) \
139 (test_bit(EXT4_GROUP_INFO_NEED_INIT_BIT, &((grp)->bb_state)))
140
141
142struct ext4_prealloc_space {
143 struct list_head pa_inode_list;
144 struct list_head pa_group_list;
145 union {
146 struct list_head pa_tmp_list;
147 struct rcu_head pa_rcu;
148 } u;
149 spinlock_t pa_lock;
150 atomic_t pa_count;
151 unsigned pa_deleted;
152 ext4_fsblk_t pa_pstart; /* phys. block */
153 ext4_lblk_t pa_lstart; /* log. block */
154 unsigned short pa_len; /* len of preallocated chunk */
155 unsigned short pa_free; /* how many blocks are free */
156 unsigned short pa_linear; /* consumed in one direction
157 * strictly, for grp prealloc */
158 spinlock_t *pa_obj_lock;
159 struct inode *pa_inode; /* hack, for history only */
160};
161
162
163struct ext4_free_extent {
164 ext4_lblk_t fe_logical;
165 ext4_grpblk_t fe_start;
166 ext4_group_t fe_group;
167 int fe_len;
168};
169
170/*
171 * Locality group:
172 * we try to group all related changes together
173 * so that writeback can flush/allocate them together as well
Aneesh Kumar K.V6be2ded2008-07-23 14:14:05 -0400174 * Size of lg_prealloc_list hash is determined by MB_DEFAULT_GROUP_PREALLOC
175 * (512). We store prealloc space into the hash based on the pa_free blocks
176 * order value.ie, fls(pa_free)-1;
Mingming Cao8f6e39a2008-04-29 22:01:31 -0400177 */
Aneesh Kumar K.V6be2ded2008-07-23 14:14:05 -0400178#define PREALLOC_TB_SIZE 10
Mingming Cao8f6e39a2008-04-29 22:01:31 -0400179struct ext4_locality_group {
180 /* for allocator */
Aneesh Kumar K.V6be2ded2008-07-23 14:14:05 -0400181 /* to serialize allocates */
182 struct mutex lg_mutex;
183 /* list of preallocations */
184 struct list_head lg_prealloc_list[PREALLOC_TB_SIZE];
Mingming Cao8f6e39a2008-04-29 22:01:31 -0400185 spinlock_t lg_prealloc_lock;
186};
187
188struct ext4_allocation_context {
189 struct inode *ac_inode;
190 struct super_block *ac_sb;
191
192 /* original request */
193 struct ext4_free_extent ac_o_ex;
194
195 /* goal request (after normalization) */
196 struct ext4_free_extent ac_g_ex;
197
198 /* the best found extent */
199 struct ext4_free_extent ac_b_ex;
200
201 /* copy of the bext found extent taken before preallocation efforts */
202 struct ext4_free_extent ac_f_ex;
203
204 /* number of iterations done. we have to track to limit searching */
205 unsigned long ac_ex_scanned;
206 __u16 ac_groups_scanned;
207 __u16 ac_found;
208 __u16 ac_tail;
209 __u16 ac_buddy;
210 __u16 ac_flags; /* allocation hints */
211 __u8 ac_status;
212 __u8 ac_criteria;
213 __u8 ac_repeats;
214 __u8 ac_2order; /* if request is to allocate 2^N blocks and
215 * N > 0, the field stores N, otherwise 0 */
216 __u8 ac_op; /* operation, for history only */
217 struct page *ac_bitmap_page;
218 struct page *ac_buddy_page;
219 struct ext4_prealloc_space *ac_pa;
220 struct ext4_locality_group *ac_lg;
221};
222
223#define AC_STATUS_CONTINUE 1
224#define AC_STATUS_FOUND 2
225#define AC_STATUS_BREAK 3
226
227struct ext4_mb_history {
228 struct ext4_free_extent orig; /* orig allocation */
229 struct ext4_free_extent goal; /* goal allocation */
230 struct ext4_free_extent result; /* result allocation */
231 unsigned pid;
232 unsigned ino;
233 __u16 found; /* how many extents have been found */
234 __u16 groups; /* how many groups have been scanned */
235 __u16 tail; /* what tail broke some buddy */
236 __u16 buddy; /* buddy the tail ^^^ broke */
237 __u16 flags;
238 __u8 cr:3; /* which phase the result extent was found at */
239 __u8 op:4;
240 __u8 merged:1;
241};
242
243struct ext4_buddy {
244 struct page *bd_buddy_page;
245 void *bd_buddy;
246 struct page *bd_bitmap_page;
247 void *bd_bitmap;
248 struct ext4_group_info *bd_info;
249 struct super_block *bd_sb;
250 __u16 bd_blkbits;
251 ext4_group_t bd_group;
Aneesh Kumar K.V920313a2009-01-05 21:36:19 -0500252 struct rw_semaphore *alloc_semp;
Mingming Cao8f6e39a2008-04-29 22:01:31 -0400253};
254#define EXT4_MB_BITMAP(e4b) ((e4b)->bd_bitmap)
255#define EXT4_MB_BUDDY(e4b) ((e4b)->bd_buddy)
256
257#ifndef EXT4_MB_HISTORY
258static inline void ext4_mb_store_history(struct ext4_allocation_context *ac)
259{
260 return;
261}
Mingming Cao8f6e39a2008-04-29 22:01:31 -0400262#endif
263
264#define in_range(b, first, len) ((b) >= (first) && (b) <= (first) + (len) - 1)
265
Mingming Cao8f6e39a2008-04-29 22:01:31 -0400266struct buffer_head *read_block_bitmap(struct super_block *, ext4_group_t);
267
Mingming Cao8f6e39a2008-04-29 22:01:31 -0400268
269static inline void ext4_lock_group(struct super_block *sb, ext4_group_t group)
270{
271 struct ext4_group_info *grinfo = ext4_get_group_info(sb, group);
272
273 bit_spin_lock(EXT4_GROUP_INFO_LOCKED_BIT, &(grinfo->bb_state));
274}
275
276static inline void ext4_unlock_group(struct super_block *sb,
277 ext4_group_t group)
278{
279 struct ext4_group_info *grinfo = ext4_get_group_info(sb, group);
280
281 bit_spin_unlock(EXT4_GROUP_INFO_LOCKED_BIT, &(grinfo->bb_state));
282}
283
284static inline int ext4_is_group_locked(struct super_block *sb,
285 ext4_group_t group)
286{
287 struct ext4_group_info *grinfo = ext4_get_group_info(sb, group);
288
289 return bit_spin_is_locked(EXT4_GROUP_INFO_LOCKED_BIT,
290 &(grinfo->bb_state));
291}
292
Aneesh Kumar K.Vc3a326a2008-11-25 15:11:52 -0500293static inline ext4_fsblk_t ext4_grp_offs_to_block(struct super_block *sb,
Mingming Cao8f6e39a2008-04-29 22:01:31 -0400294 struct ext4_free_extent *fex)
295{
296 ext4_fsblk_t block;
297
298 block = (ext4_fsblk_t) fex->fe_group * EXT4_BLOCKS_PER_GROUP(sb)
299 + fex->fe_start
300 + le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block);
301 return block;
302}
303#endif