blob: b5dff1fff1e5b33af1c583f026b883fae74eca82 [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>
Mingming Cao8f6e39a2008-04-29 22:01:31 -040023#include "ext4_jbd2.h"
24#include "ext4.h"
25#include "group.h"
26
27/*
28 * with AGGRESSIVE_CHECK allocator runs consistency checks over
29 * structures. these checks slow things down a lot
30 */
31#define AGGRESSIVE_CHECK__
32
33/*
34 * with DOUBLE_CHECK defined mballoc creates persistent in-core
35 * bitmaps, maintains and uses them to check for double allocations
36 */
37#define DOUBLE_CHECK__
38
39/*
40 */
41#define MB_DEBUG__
42#ifdef MB_DEBUG
43#define mb_debug(fmt, a...) printk(fmt, ##a)
44#else
45#define mb_debug(fmt, a...)
46#endif
47
48/*
49 * with EXT4_MB_HISTORY mballoc stores last N allocations in memory
50 * and you can monitor it in /proc/fs/ext4/<dev>/mb_history
51 */
52#define EXT4_MB_HISTORY
53#define EXT4_MB_HISTORY_ALLOC 1 /* allocation */
54#define EXT4_MB_HISTORY_PREALLOC 2 /* preallocated blocks used */
55#define EXT4_MB_HISTORY_DISCARD 4 /* preallocation discarded */
56#define EXT4_MB_HISTORY_FREE 8 /* free */
57
58#define EXT4_MB_HISTORY_DEFAULT (EXT4_MB_HISTORY_ALLOC | \
59 EXT4_MB_HISTORY_PREALLOC)
60
61/*
62 * How long mballoc can look for a best extent (in found extents)
63 */
64#define MB_DEFAULT_MAX_TO_SCAN 200
65
66/*
67 * How long mballoc must look for a best extent
68 */
69#define MB_DEFAULT_MIN_TO_SCAN 10
70
71/*
72 * How many groups mballoc will scan looking for the best chunk
73 */
74#define MB_DEFAULT_MAX_GROUPS_TO_SCAN 5
75
76/*
77 * with 'ext4_mb_stats' allocator will collect stats that will be
78 * shown at umount. The collecting costs though!
79 */
80#define MB_DEFAULT_STATS 1
81
82/*
83 * files smaller than MB_DEFAULT_STREAM_THRESHOLD are served
84 * by the stream allocator, which purpose is to pack requests
85 * as close each to other as possible to produce smooth I/O traffic
86 * We use locality group prealloc space for stream request.
87 * We can tune the same via /proc/fs/ext4/<parition>/stream_req
88 */
89#define MB_DEFAULT_STREAM_THRESHOLD 16 /* 64K */
90
91/*
92 * for which requests use 2^N search using buddies
93 */
94#define MB_DEFAULT_ORDER2_REQS 2
95
96/*
97 * default group prealloc size 512 blocks
98 */
99#define MB_DEFAULT_GROUP_PREALLOC 512
100
101static struct kmem_cache *ext4_pspace_cachep;
102static struct kmem_cache *ext4_ac_cachep;
Aneesh Kumar K.Vc8940582008-10-16 10:14:27 -0400103static struct kmem_cache *ext4_free_ext_cachep;
Mingming Cao8f6e39a2008-04-29 22:01:31 -0400104
Aneesh Kumar K.Vc8940582008-10-16 10:14:27 -0400105struct ext4_free_data {
106 /* this links the free block information from group_info */
107 struct rb_node node;
Mingming Cao8f6e39a2008-04-29 22:01:31 -0400108
Aneesh Kumar K.Vc8940582008-10-16 10:14:27 -0400109 /* this links the free block information from ext4_sb_info */
Mingming Cao8f6e39a2008-04-29 22:01:31 -0400110 struct list_head list;
Aneesh Kumar K.Vc8940582008-10-16 10:14:27 -0400111
112 /* group which free block extent belongs */
113 ext4_group_t group;
114
115 /* free block extent */
116 ext4_grpblk_t start_blk;
117 ext4_grpblk_t count;
118
119 /* transaction which freed this extent */
120 tid_t t_tid;
Mingming Cao8f6e39a2008-04-29 22:01:31 -0400121};
122
123struct ext4_group_info {
124 unsigned long bb_state;
Aneesh Kumar K.Vc8940582008-10-16 10:14:27 -0400125 struct rb_root bb_free_root;
Mingming Cao8f6e39a2008-04-29 22:01:31 -0400126 unsigned short bb_first_free;
127 unsigned short bb_free;
128 unsigned short bb_fragments;
129 struct list_head bb_prealloc_list;
130#ifdef DOUBLE_CHECK
131 void *bb_bitmap;
132#endif
133 unsigned short bb_counters[];
134};
135
136#define EXT4_GROUP_INFO_NEED_INIT_BIT 0
137#define EXT4_GROUP_INFO_LOCKED_BIT 1
138
139#define EXT4_MB_GRP_NEED_INIT(grp) \
140 (test_bit(EXT4_GROUP_INFO_NEED_INIT_BIT, &((grp)->bb_state)))
141
142
143struct ext4_prealloc_space {
144 struct list_head pa_inode_list;
145 struct list_head pa_group_list;
146 union {
147 struct list_head pa_tmp_list;
148 struct rcu_head pa_rcu;
149 } u;
150 spinlock_t pa_lock;
151 atomic_t pa_count;
152 unsigned pa_deleted;
153 ext4_fsblk_t pa_pstart; /* phys. block */
154 ext4_lblk_t pa_lstart; /* log. block */
155 unsigned short pa_len; /* len of preallocated chunk */
156 unsigned short pa_free; /* how many blocks are free */
157 unsigned short pa_linear; /* consumed in one direction
158 * strictly, for grp prealloc */
159 spinlock_t *pa_obj_lock;
160 struct inode *pa_inode; /* hack, for history only */
161};
162
163
164struct ext4_free_extent {
165 ext4_lblk_t fe_logical;
166 ext4_grpblk_t fe_start;
167 ext4_group_t fe_group;
168 int fe_len;
169};
170
171/*
172 * Locality group:
173 * we try to group all related changes together
174 * so that writeback can flush/allocate them together as well
Aneesh Kumar K.V6be2ded2008-07-23 14:14:05 -0400175 * Size of lg_prealloc_list hash is determined by MB_DEFAULT_GROUP_PREALLOC
176 * (512). We store prealloc space into the hash based on the pa_free blocks
177 * order value.ie, fls(pa_free)-1;
Mingming Cao8f6e39a2008-04-29 22:01:31 -0400178 */
Aneesh Kumar K.V6be2ded2008-07-23 14:14:05 -0400179#define PREALLOC_TB_SIZE 10
Mingming Cao8f6e39a2008-04-29 22:01:31 -0400180struct ext4_locality_group {
181 /* for allocator */
Aneesh Kumar K.V6be2ded2008-07-23 14:14:05 -0400182 /* to serialize allocates */
183 struct mutex lg_mutex;
184 /* list of preallocations */
185 struct list_head lg_prealloc_list[PREALLOC_TB_SIZE];
Mingming Cao8f6e39a2008-04-29 22:01:31 -0400186 spinlock_t lg_prealloc_lock;
187};
188
189struct ext4_allocation_context {
190 struct inode *ac_inode;
191 struct super_block *ac_sb;
192
193 /* original request */
194 struct ext4_free_extent ac_o_ex;
195
196 /* goal request (after normalization) */
197 struct ext4_free_extent ac_g_ex;
198
199 /* the best found extent */
200 struct ext4_free_extent ac_b_ex;
201
202 /* copy of the bext found extent taken before preallocation efforts */
203 struct ext4_free_extent ac_f_ex;
204
205 /* number of iterations done. we have to track to limit searching */
206 unsigned long ac_ex_scanned;
207 __u16 ac_groups_scanned;
208 __u16 ac_found;
209 __u16 ac_tail;
210 __u16 ac_buddy;
211 __u16 ac_flags; /* allocation hints */
212 __u8 ac_status;
213 __u8 ac_criteria;
214 __u8 ac_repeats;
215 __u8 ac_2order; /* if request is to allocate 2^N blocks and
216 * N > 0, the field stores N, otherwise 0 */
217 __u8 ac_op; /* operation, for history only */
218 struct page *ac_bitmap_page;
219 struct page *ac_buddy_page;
220 struct ext4_prealloc_space *ac_pa;
221 struct ext4_locality_group *ac_lg;
222};
223
224#define AC_STATUS_CONTINUE 1
225#define AC_STATUS_FOUND 2
226#define AC_STATUS_BREAK 3
227
228struct ext4_mb_history {
229 struct ext4_free_extent orig; /* orig allocation */
230 struct ext4_free_extent goal; /* goal allocation */
231 struct ext4_free_extent result; /* result allocation */
232 unsigned pid;
233 unsigned ino;
234 __u16 found; /* how many extents have been found */
235 __u16 groups; /* how many groups have been scanned */
236 __u16 tail; /* what tail broke some buddy */
237 __u16 buddy; /* buddy the tail ^^^ broke */
238 __u16 flags;
239 __u8 cr:3; /* which phase the result extent was found at */
240 __u8 op:4;
241 __u8 merged:1;
242};
243
244struct ext4_buddy {
245 struct page *bd_buddy_page;
246 void *bd_buddy;
247 struct page *bd_bitmap_page;
248 void *bd_bitmap;
249 struct ext4_group_info *bd_info;
250 struct super_block *bd_sb;
251 __u16 bd_blkbits;
252 ext4_group_t bd_group;
253};
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}
262#else
263static void ext4_mb_store_history(struct ext4_allocation_context *ac);
264#endif
265
266#define in_range(b, first, len) ((b) >= (first) && (b) <= (first) + (len) - 1)
267
Mingming Cao8f6e39a2008-04-29 22:01:31 -0400268struct buffer_head *read_block_bitmap(struct super_block *, ext4_group_t);
269
270static void ext4_mb_generate_from_pa(struct super_block *sb, void *bitmap,
271 ext4_group_t group);
Mingming Cao8f6e39a2008-04-29 22:01:31 -0400272static void ext4_mb_return_to_preallocation(struct inode *inode,
273 struct ext4_buddy *e4b, sector_t block,
274 int count);
275static void ext4_mb_put_pa(struct ext4_allocation_context *,
276 struct super_block *, struct ext4_prealloc_space *pa);
277static int ext4_mb_init_per_dev_proc(struct super_block *sb);
278static int ext4_mb_destroy_per_dev_proc(struct super_block *sb);
Theodore Ts'o3e624fc2008-10-16 20:00:24 -0400279static void release_blocks_on_commit(journal_t *journal, transaction_t *txn);
Mingming Cao8f6e39a2008-04-29 22:01:31 -0400280
281
282static inline void ext4_lock_group(struct super_block *sb, ext4_group_t group)
283{
284 struct ext4_group_info *grinfo = ext4_get_group_info(sb, group);
285
286 bit_spin_lock(EXT4_GROUP_INFO_LOCKED_BIT, &(grinfo->bb_state));
287}
288
289static inline void ext4_unlock_group(struct super_block *sb,
290 ext4_group_t group)
291{
292 struct ext4_group_info *grinfo = ext4_get_group_info(sb, group);
293
294 bit_spin_unlock(EXT4_GROUP_INFO_LOCKED_BIT, &(grinfo->bb_state));
295}
296
297static inline int ext4_is_group_locked(struct super_block *sb,
298 ext4_group_t group)
299{
300 struct ext4_group_info *grinfo = ext4_get_group_info(sb, group);
301
302 return bit_spin_is_locked(EXT4_GROUP_INFO_LOCKED_BIT,
303 &(grinfo->bb_state));
304}
305
306static ext4_fsblk_t ext4_grp_offs_to_block(struct super_block *sb,
307 struct ext4_free_extent *fex)
308{
309 ext4_fsblk_t block;
310
311 block = (ext4_fsblk_t) fex->fe_group * EXT4_BLOCKS_PER_GROUP(sb)
312 + fex->fe_start
313 + le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block);
314 return block;
315}
316#endif