blob: 436521cae45662629d02871d5ff8429aee92065b [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>
Theodore Ts'o8a0aba72008-10-16 10:06:27 -040020#include <linux/blkdev.h>
Aneesh Kumar K.V920313a2009-01-05 21:36:19 -050021#include <linux/mutex.h>
Mingming Cao8f6e39a2008-04-29 22:01:31 -040022#include "ext4_jbd2.h"
23#include "ext4.h"
Mingming Cao8f6e39a2008-04-29 22:01:31 -040024
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 */
Theodore Ts'o6ba495e2009-09-18 13:38:55 -040039#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)
Mingming Cao8f6e39a2008-04-29 22:01:31 -040050#else
Theodore Ts'o6ba495e2009-09-18 13:38:55 -040051#define mb_debug(n, fmt, a...)
Mingming Cao8f6e39a2008-04-29 22:01:31 -040052#endif
53
Mingming Cao8f6e39a2008-04-29 22:01:31 -040054#define EXT4_MB_HISTORY_ALLOC 1 /* allocation */
55#define EXT4_MB_HISTORY_PREALLOC 2 /* preallocated blocks used */
Mingming Cao8f6e39a2008-04-29 22:01:31 -040056
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 */
Theodore Ts'o90576c02009-09-29 15:51:30 -040076#define MB_DEFAULT_STATS 0
Mingming Cao8f6e39a2008-04-29 22:01:31 -040077
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
Mingming Cao8f6e39a2008-04-29 22:01:31 -040097
Aneesh Kumar K.Vc8940582008-10-16 10:14:27 -040098struct ext4_free_data {
99 /* this links the free block information from group_info */
100 struct rb_node node;
Mingming Cao8f6e39a2008-04-29 22:01:31 -0400101
Aneesh Kumar K.Vc8940582008-10-16 10:14:27 -0400102 /* this links the free block information from ext4_sb_info */
Mingming Cao8f6e39a2008-04-29 22:01:31 -0400103 struct list_head list;
Aneesh Kumar K.Vc8940582008-10-16 10:14:27 -0400104
105 /* group which free block extent belongs */
106 ext4_group_t group;
107
108 /* free block extent */
109 ext4_grpblk_t start_blk;
110 ext4_grpblk_t count;
111
112 /* transaction which freed this extent */
113 tid_t t_tid;
Mingming Cao8f6e39a2008-04-29 22:01:31 -0400114};
115
Mingming Cao8f6e39a2008-04-29 22:01:31 -0400116struct ext4_prealloc_space {
117 struct list_head pa_inode_list;
118 struct list_head pa_group_list;
119 union {
120 struct list_head pa_tmp_list;
121 struct rcu_head pa_rcu;
122 } u;
123 spinlock_t pa_lock;
124 atomic_t pa_count;
125 unsigned pa_deleted;
126 ext4_fsblk_t pa_pstart; /* phys. block */
127 ext4_lblk_t pa_lstart; /* log. block */
Eric Sandeena36b4492009-08-25 22:36:45 -0400128 ext4_grpblk_t pa_len; /* len of preallocated chunk */
129 ext4_grpblk_t pa_free; /* how many blocks are free */
Aneesh Kumar K.Vcc0fb9a2009-03-27 17:16:58 -0400130 unsigned short pa_type; /* pa type. inode or group */
Mingming Cao8f6e39a2008-04-29 22:01:31 -0400131 spinlock_t *pa_obj_lock;
132 struct inode *pa_inode; /* hack, for history only */
133};
134
Aneesh Kumar K.Vcc0fb9a2009-03-27 17:16:58 -0400135enum {
136 MB_INODE_PA = 0,
137 MB_GROUP_PA = 1
138};
Mingming Cao8f6e39a2008-04-29 22:01:31 -0400139
140struct ext4_free_extent {
141 ext4_lblk_t fe_logical;
142 ext4_grpblk_t fe_start;
143 ext4_group_t fe_group;
Eric Sandeena36b4492009-08-25 22:36:45 -0400144 ext4_grpblk_t fe_len;
Mingming Cao8f6e39a2008-04-29 22:01:31 -0400145};
146
147/*
148 * Locality group:
149 * we try to group all related changes together
150 * so that writeback can flush/allocate them together as well
Aneesh Kumar K.V6be2ded2008-07-23 14:14:05 -0400151 * Size of lg_prealloc_list hash is determined by MB_DEFAULT_GROUP_PREALLOC
152 * (512). We store prealloc space into the hash based on the pa_free blocks
153 * order value.ie, fls(pa_free)-1;
Mingming Cao8f6e39a2008-04-29 22:01:31 -0400154 */
Aneesh Kumar K.V6be2ded2008-07-23 14:14:05 -0400155#define PREALLOC_TB_SIZE 10
Mingming Cao8f6e39a2008-04-29 22:01:31 -0400156struct ext4_locality_group {
157 /* for allocator */
Aneesh Kumar K.V6be2ded2008-07-23 14:14:05 -0400158 /* to serialize allocates */
159 struct mutex lg_mutex;
160 /* list of preallocations */
161 struct list_head lg_prealloc_list[PREALLOC_TB_SIZE];
Mingming Cao8f6e39a2008-04-29 22:01:31 -0400162 spinlock_t lg_prealloc_lock;
163};
164
165struct ext4_allocation_context {
166 struct inode *ac_inode;
167 struct super_block *ac_sb;
168
169 /* original request */
170 struct ext4_free_extent ac_o_ex;
171
172 /* goal request (after normalization) */
173 struct ext4_free_extent ac_g_ex;
174
175 /* the best found extent */
176 struct ext4_free_extent ac_b_ex;
177
178 /* copy of the bext found extent taken before preallocation efforts */
179 struct ext4_free_extent ac_f_ex;
180
181 /* number of iterations done. we have to track to limit searching */
182 unsigned long ac_ex_scanned;
183 __u16 ac_groups_scanned;
184 __u16 ac_found;
185 __u16 ac_tail;
186 __u16 ac_buddy;
187 __u16 ac_flags; /* allocation hints */
188 __u8 ac_status;
189 __u8 ac_criteria;
190 __u8 ac_repeats;
191 __u8 ac_2order; /* if request is to allocate 2^N blocks and
192 * N > 0, the field stores N, otherwise 0 */
193 __u8 ac_op; /* operation, for history only */
194 struct page *ac_bitmap_page;
195 struct page *ac_buddy_page;
Aneesh Kumar K.V8556e8f2009-01-05 21:46:55 -0500196 /*
197 * pointer to the held semaphore upon successful
198 * block allocation
199 */
200 struct rw_semaphore *alloc_semp;
Mingming Cao8f6e39a2008-04-29 22:01:31 -0400201 struct ext4_prealloc_space *ac_pa;
202 struct ext4_locality_group *ac_lg;
203};
204
205#define AC_STATUS_CONTINUE 1
206#define AC_STATUS_FOUND 2
207#define AC_STATUS_BREAK 3
208
Mingming Cao8f6e39a2008-04-29 22:01:31 -0400209struct ext4_buddy {
210 struct page *bd_buddy_page;
211 void *bd_buddy;
212 struct page *bd_bitmap_page;
213 void *bd_bitmap;
214 struct ext4_group_info *bd_info;
215 struct super_block *bd_sb;
216 __u16 bd_blkbits;
217 ext4_group_t bd_group;
Aneesh Kumar K.V920313a2009-01-05 21:36:19 -0500218 struct rw_semaphore *alloc_semp;
Mingming Cao8f6e39a2008-04-29 22:01:31 -0400219};
220#define EXT4_MB_BITMAP(e4b) ((e4b)->bd_bitmap)
221#define EXT4_MB_BUDDY(e4b) ((e4b)->bd_buddy)
222
Mingming Cao8f6e39a2008-04-29 22:01:31 -0400223#define in_range(b, first, len) ((b) >= (first) && (b) <= (first) + (len) - 1)
224
Aneesh Kumar K.Vc3a326a2008-11-25 15:11:52 -0500225static inline ext4_fsblk_t ext4_grp_offs_to_block(struct super_block *sb,
Mingming Cao8f6e39a2008-04-29 22:01:31 -0400226 struct ext4_free_extent *fex)
227{
228 ext4_fsblk_t block;
229
230 block = (ext4_fsblk_t) fex->fe_group * EXT4_BLOCKS_PER_GROUP(sb)
231 + fex->fe_start
232 + le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block);
233 return block;
234}
235#endif