Loading...
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/blkdev.h>
21#include <linux/mutex.h>
22#include "ext4_jbd2.h"
23#include "ext4.h"
24
25/*
26 */
27#ifdef CONFIG_EXT4_DEBUG
28extern ushort ext4_mballoc_debug;
29
30#define mb_debug(n, fmt, a...) \
31 do { \
32 if ((n) <= ext4_mballoc_debug) { \
33 printk(KERN_DEBUG "(%s, %d): %s: ", \
34 __FILE__, __LINE__, __func__); \
35 printk(fmt, ## a); \
36 } \
37 } while (0)
38#else
39#define mb_debug(n, fmt, a...) no_printk(fmt, ## a)
40#endif
41
42#define EXT4_MB_HISTORY_ALLOC 1 /* allocation */
43#define EXT4_MB_HISTORY_PREALLOC 2 /* preallocated blocks used */
44
45/*
46 * How long mballoc can look for a best extent (in found extents)
47 */
48#define MB_DEFAULT_MAX_TO_SCAN 200
49
50/*
51 * How long mballoc must look for a best extent
52 */
53#define MB_DEFAULT_MIN_TO_SCAN 10
54
55/*
56 * with 'ext4_mb_stats' allocator will collect stats that will be
57 * shown at umount. The collecting costs though!
58 */
59#define MB_DEFAULT_STATS 0
60
61/*
62 * files smaller than MB_DEFAULT_STREAM_THRESHOLD are served
63 * by the stream allocator, which purpose is to pack requests
64 * as close each to other as possible to produce smooth I/O traffic
65 * We use locality group prealloc space for stream request.
66 * We can tune the same via /proc/fs/ext4/<parition>/stream_req
67 */
68#define MB_DEFAULT_STREAM_THRESHOLD 16 /* 64K */
69
70/*
71 * for which requests use 2^N search using buddies
72 */
73#define MB_DEFAULT_ORDER2_REQS 2
74
75/*
76 * default group prealloc size 512 blocks
77 */
78#define MB_DEFAULT_GROUP_PREALLOC 512
79
80
81struct ext4_free_data {
82 /* MUST be the first member */
83 struct ext4_journal_cb_entry efd_jce;
84
85 /* ext4_free_data private data starts from here */
86
87 /* this links the free block information from group_info */
88 struct rb_node efd_node;
89
90 /* group which free block extent belongs */
91 ext4_group_t efd_group;
92
93 /* free block extent */
94 ext4_grpblk_t efd_start_cluster;
95 ext4_grpblk_t efd_count;
96
97 /* transaction which freed this extent */
98 tid_t efd_tid;
99};
100
101struct ext4_prealloc_space {
102 struct list_head pa_inode_list;
103 struct list_head pa_group_list;
104 union {
105 struct list_head pa_tmp_list;
106 struct rcu_head pa_rcu;
107 } u;
108 spinlock_t pa_lock;
109 atomic_t pa_count;
110 unsigned pa_deleted;
111 ext4_fsblk_t pa_pstart; /* phys. block */
112 ext4_lblk_t pa_lstart; /* log. block */
113 ext4_grpblk_t pa_len; /* len of preallocated chunk */
114 ext4_grpblk_t pa_free; /* how many blocks are free */
115 unsigned short pa_type; /* pa type. inode or group */
116 spinlock_t *pa_obj_lock;
117 struct inode *pa_inode; /* hack, for history only */
118};
119
120enum {
121 MB_INODE_PA = 0,
122 MB_GROUP_PA = 1
123};
124
125struct ext4_free_extent {
126 ext4_lblk_t fe_logical;
127 ext4_grpblk_t fe_start; /* In cluster units */
128 ext4_group_t fe_group;
129 ext4_grpblk_t fe_len; /* In cluster units */
130};
131
132/*
133 * Locality group:
134 * we try to group all related changes together
135 * so that writeback can flush/allocate them together as well
136 * Size of lg_prealloc_list hash is determined by MB_DEFAULT_GROUP_PREALLOC
137 * (512). We store prealloc space into the hash based on the pa_free blocks
138 * order value.ie, fls(pa_free)-1;
139 */
140#define PREALLOC_TB_SIZE 10
141struct ext4_locality_group {
142 /* for allocator */
143 /* to serialize allocates */
144 struct mutex lg_mutex;
145 /* list of preallocations */
146 struct list_head lg_prealloc_list[PREALLOC_TB_SIZE];
147 spinlock_t lg_prealloc_lock;
148};
149
150struct ext4_allocation_context {
151 struct inode *ac_inode;
152 struct super_block *ac_sb;
153
154 /* original request */
155 struct ext4_free_extent ac_o_ex;
156
157 /* goal request (normalized ac_o_ex) */
158 struct ext4_free_extent ac_g_ex;
159
160 /* the best found extent */
161 struct ext4_free_extent ac_b_ex;
162
163 /* copy of the best found extent taken before preallocation efforts */
164 struct ext4_free_extent ac_f_ex;
165
166 __u16 ac_groups_scanned;
167 __u16 ac_found;
168 __u16 ac_tail;
169 __u16 ac_buddy;
170 __u16 ac_flags; /* allocation hints */
171 __u8 ac_status;
172 __u8 ac_criteria;
173 __u8 ac_2order; /* if request is to allocate 2^N blocks and
174 * N > 0, the field stores N, otherwise 0 */
175 __u8 ac_op; /* operation, for history only */
176 struct page *ac_bitmap_page;
177 struct page *ac_buddy_page;
178 struct ext4_prealloc_space *ac_pa;
179 struct ext4_locality_group *ac_lg;
180};
181
182#define AC_STATUS_CONTINUE 1
183#define AC_STATUS_FOUND 2
184#define AC_STATUS_BREAK 3
185
186struct ext4_buddy {
187 struct page *bd_buddy_page;
188 void *bd_buddy;
189 struct page *bd_bitmap_page;
190 void *bd_bitmap;
191 struct ext4_group_info *bd_info;
192 struct super_block *bd_sb;
193 __u16 bd_blkbits;
194 ext4_group_t bd_group;
195};
196
197static inline ext4_fsblk_t ext4_grp_offs_to_block(struct super_block *sb,
198 struct ext4_free_extent *fex)
199{
200 return ext4_group_first_block_no(sb, fex->fe_group) +
201 (fex->fe_start << EXT4_SB(sb)->s_cluster_bits);
202}
203#endif
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/blkdev.h>
21#include <linux/mutex.h>
22#include "ext4_jbd2.h"
23#include "ext4.h"
24
25/*
26 * with AGGRESSIVE_CHECK allocator runs consistency checks over
27 * structures. these checks slow things down a lot
28 */
29#define AGGRESSIVE_CHECK__
30
31/*
32 * with DOUBLE_CHECK defined mballoc creates persistent in-core
33 * bitmaps, maintains and uses them to check for double allocations
34 */
35#define DOUBLE_CHECK__
36
37/*
38 */
39#ifdef CONFIG_EXT4_DEBUG
40extern u8 mb_enable_debug;
41
42#define mb_debug(n, fmt, a...) \
43 do { \
44 if ((n) <= mb_enable_debug) { \
45 printk(KERN_DEBUG "(%s, %d): %s: ", \
46 __FILE__, __LINE__, __func__); \
47 printk(fmt, ## a); \
48 } \
49 } while (0)
50#else
51#define mb_debug(n, fmt, a...)
52#endif
53
54#define EXT4_MB_HISTORY_ALLOC 1 /* allocation */
55#define EXT4_MB_HISTORY_PREALLOC 2 /* preallocated blocks used */
56
57/*
58 * How long mballoc can look for a best extent (in found extents)
59 */
60#define MB_DEFAULT_MAX_TO_SCAN 200
61
62/*
63 * How long mballoc must look for a best extent
64 */
65#define MB_DEFAULT_MIN_TO_SCAN 10
66
67/*
68 * How many groups mballoc will scan looking for the best chunk
69 */
70#define MB_DEFAULT_MAX_GROUPS_TO_SCAN 5
71
72/*
73 * with 'ext4_mb_stats' allocator will collect stats that will be
74 * shown at umount. The collecting costs though!
75 */
76#define MB_DEFAULT_STATS 0
77
78/*
79 * files smaller than MB_DEFAULT_STREAM_THRESHOLD are served
80 * by the stream allocator, which purpose is to pack requests
81 * as close each to other as possible to produce smooth I/O traffic
82 * We use locality group prealloc space for stream request.
83 * We can tune the same via /proc/fs/ext4/<parition>/stream_req
84 */
85#define MB_DEFAULT_STREAM_THRESHOLD 16 /* 64K */
86
87/*
88 * for which requests use 2^N search using buddies
89 */
90#define MB_DEFAULT_ORDER2_REQS 2
91
92/*
93 * default group prealloc size 512 blocks
94 */
95#define MB_DEFAULT_GROUP_PREALLOC 512
96
97
98struct ext4_free_data {
99 /* MUST be the first member */
100 struct ext4_journal_cb_entry efd_jce;
101
102 /* ext4_free_data private data starts from here */
103
104 /* this links the free block information from group_info */
105 struct rb_node efd_node;
106
107 /* group which free block extent belongs */
108 ext4_group_t efd_group;
109
110 /* free block extent */
111 ext4_grpblk_t efd_start_cluster;
112 ext4_grpblk_t efd_count;
113
114 /* transaction which freed this extent */
115 tid_t efd_tid;
116};
117
118struct ext4_prealloc_space {
119 struct list_head pa_inode_list;
120 struct list_head pa_group_list;
121 union {
122 struct list_head pa_tmp_list;
123 struct rcu_head pa_rcu;
124 } u;
125 spinlock_t pa_lock;
126 atomic_t pa_count;
127 unsigned pa_deleted;
128 ext4_fsblk_t pa_pstart; /* phys. block */
129 ext4_lblk_t pa_lstart; /* log. block */
130 ext4_grpblk_t pa_len; /* len of preallocated chunk */
131 ext4_grpblk_t pa_free; /* how many blocks are free */
132 unsigned short pa_type; /* pa type. inode or group */
133 spinlock_t *pa_obj_lock;
134 struct inode *pa_inode; /* hack, for history only */
135};
136
137enum {
138 MB_INODE_PA = 0,
139 MB_GROUP_PA = 1
140};
141
142struct ext4_free_extent {
143 ext4_lblk_t fe_logical;
144 ext4_grpblk_t fe_start; /* In cluster units */
145 ext4_group_t fe_group;
146 ext4_grpblk_t fe_len; /* In cluster units */
147};
148
149/*
150 * Locality group:
151 * we try to group all related changes together
152 * so that writeback can flush/allocate them together as well
153 * Size of lg_prealloc_list hash is determined by MB_DEFAULT_GROUP_PREALLOC
154 * (512). We store prealloc space into the hash based on the pa_free blocks
155 * order value.ie, fls(pa_free)-1;
156 */
157#define PREALLOC_TB_SIZE 10
158struct ext4_locality_group {
159 /* for allocator */
160 /* to serialize allocates */
161 struct mutex lg_mutex;
162 /* list of preallocations */
163 struct list_head lg_prealloc_list[PREALLOC_TB_SIZE];
164 spinlock_t lg_prealloc_lock;
165};
166
167struct ext4_allocation_context {
168 struct inode *ac_inode;
169 struct super_block *ac_sb;
170
171 /* original request */
172 struct ext4_free_extent ac_o_ex;
173
174 /* goal request (normalized ac_o_ex) */
175 struct ext4_free_extent ac_g_ex;
176
177 /* the best found extent */
178 struct ext4_free_extent ac_b_ex;
179
180 /* copy of the best found extent taken before preallocation efforts */
181 struct ext4_free_extent ac_f_ex;
182
183 /* number of iterations done. we have to track to limit searching */
184 unsigned long ac_ex_scanned;
185 __u16 ac_groups_scanned;
186 __u16 ac_found;
187 __u16 ac_tail;
188 __u16 ac_buddy;
189 __u16 ac_flags; /* allocation hints */
190 __u8 ac_status;
191 __u8 ac_criteria;
192 __u8 ac_2order; /* if request is to allocate 2^N blocks and
193 * N > 0, the field stores N, otherwise 0 */
194 __u8 ac_op; /* operation, for history only */
195 struct page *ac_bitmap_page;
196 struct page *ac_buddy_page;
197 struct ext4_prealloc_space *ac_pa;
198 struct ext4_locality_group *ac_lg;
199};
200
201#define AC_STATUS_CONTINUE 1
202#define AC_STATUS_FOUND 2
203#define AC_STATUS_BREAK 3
204
205struct ext4_buddy {
206 struct page *bd_buddy_page;
207 void *bd_buddy;
208 struct page *bd_bitmap_page;
209 void *bd_bitmap;
210 struct ext4_group_info *bd_info;
211 struct super_block *bd_sb;
212 __u16 bd_blkbits;
213 ext4_group_t bd_group;
214};
215
216static inline ext4_fsblk_t ext4_grp_offs_to_block(struct super_block *sb,
217 struct ext4_free_extent *fex)
218{
219 return ext4_group_first_block_no(sb, fex->fe_group) +
220 (fex->fe_start << EXT4_SB(sb)->s_cluster_bits);
221}
222#endif