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> 21 #include <linux/blkdev.h> 22 #include <linux/marker.h> 23 #include <linux/mutex.h> 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 512 101 102 103 struct ext4_free_data { 104 /* this links the free block information from group_info */ 105 struct rb_node node; 106 107 /* this links the free block information from ext4_sb_info */ 108 struct list_head list; 109 110 /* 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; 119 }; 120 121 struct ext4_prealloc_space { 122 struct list_head pa_inode_list; 123 struct list_head pa_group_list; 124 union { 125 struct list_head pa_tmp_list; 126 struct rcu_head pa_rcu; 127 } u; 128 spinlock_t pa_lock; 129 atomic_t pa_count; 130 unsigned pa_deleted; 131 ext4_fsblk_t pa_pstart; /* phys. block */ 132 ext4_lblk_t pa_lstart; /* log. block */ 133 unsigned short pa_len; /* len of preallocated chunk */ 134 unsigned short pa_free; /* how many blocks are free */ 135 unsigned short pa_type; /* pa type. inode or group */ 136 spinlock_t *pa_obj_lock; 137 struct inode *pa_inode; /* hack, for history only */ 138 }; 139 140 enum { 141 MB_INODE_PA = 0, 142 MB_GROUP_PA = 1 143 }; 144 145 struct ext4_free_extent { 146 ext4_lblk_t fe_logical; 147 ext4_grpblk_t fe_start; 148 ext4_group_t fe_group; 149 int fe_len; 150 }; 151 152 /* 153 * Locality group: 154 * we try to group all related changes together 155 * so that writeback can flush/allocate them together as well 156 * Size of lg_prealloc_list hash is determined by MB_DEFAULT_GROUP_PREALLOC 157 * (512). We store prealloc space into the hash based on the pa_free blocks 158 * order value.ie, fls(pa_free)-1; 159 */ 160 #define PREALLOC_TB_SIZE 10 161 struct ext4_locality_group { 162 /* for allocator */ 163 /* to serialize allocates */ 164 struct mutex lg_mutex; 165 /* list of preallocations */ 166 struct list_head lg_prealloc_list[PREALLOC_TB_SIZE]; 167 spinlock_t lg_prealloc_lock; 168 }; 169 170 struct ext4_allocation_context { 171 struct inode *ac_inode; 172 struct super_block *ac_sb; 173 174 /* original request */ 175 struct ext4_free_extent ac_o_ex; 176 177 /* goal request (after normalization) */ 178 struct ext4_free_extent ac_g_ex; 179 180 /* the best found extent */ 181 struct ext4_free_extent ac_b_ex; 182 183 /* copy of the bext found extent taken before preallocation efforts */ 184 struct ext4_free_extent ac_f_ex; 185 186 /* number of iterations done. we have to track to limit searching */ 187 unsigned long ac_ex_scanned; 188 __u16 ac_groups_scanned; 189 __u16 ac_found; 190 __u16 ac_tail; 191 __u16 ac_buddy; 192 __u16 ac_flags; /* allocation hints */ 193 __u8 ac_status; 194 __u8 ac_criteria; 195 __u8 ac_repeats; 196 __u8 ac_2order; /* if request is to allocate 2^N blocks and 197 * N > 0, the field stores N, otherwise 0 */ 198 __u8 ac_op; /* operation, for history only */ 199 struct page *ac_bitmap_page; 200 struct page *ac_buddy_page; 201 /* 202 * pointer to the held semaphore upon successful 203 * block allocation 204 */ 205 struct rw_semaphore *alloc_semp; 206 struct ext4_prealloc_space *ac_pa; 207 struct ext4_locality_group *ac_lg; 208 }; 209 210 #define AC_STATUS_CONTINUE 1 211 #define AC_STATUS_FOUND 2 212 #define AC_STATUS_BREAK 3 213 214 struct ext4_mb_history { 215 struct ext4_free_extent orig; /* orig allocation */ 216 struct ext4_free_extent goal; /* goal allocation */ 217 struct ext4_free_extent result; /* result allocation */ 218 unsigned pid; 219 unsigned ino; 220 __u16 found; /* how many extents have been found */ 221 __u16 groups; /* how many groups have been scanned */ 222 __u16 tail; /* what tail broke some buddy */ 223 __u16 buddy; /* buddy the tail ^^^ broke */ 224 __u16 flags; 225 __u8 cr:3; /* which phase the result extent was found at */ 226 __u8 op:4; 227 __u8 merged:1; 228 }; 229 230 struct ext4_buddy { 231 struct page *bd_buddy_page; 232 void *bd_buddy; 233 struct page *bd_bitmap_page; 234 void *bd_bitmap; 235 struct ext4_group_info *bd_info; 236 struct super_block *bd_sb; 237 __u16 bd_blkbits; 238 ext4_group_t bd_group; 239 struct rw_semaphore *alloc_semp; 240 }; 241 #define EXT4_MB_BITMAP(e4b) ((e4b)->bd_bitmap) 242 #define EXT4_MB_BUDDY(e4b) ((e4b)->bd_buddy) 243 244 #ifndef EXT4_MB_HISTORY 245 static inline void ext4_mb_store_history(struct ext4_allocation_context *ac) 246 { 247 return; 248 } 249 #endif 250 251 #define in_range(b, first, len) ((b) >= (first) && (b) <= (first) + (len) - 1) 252 253 static inline ext4_fsblk_t ext4_grp_offs_to_block(struct super_block *sb, 254 struct ext4_free_extent *fex) 255 { 256 ext4_fsblk_t block; 257 258 block = (ext4_fsblk_t) fex->fe_group * EXT4_BLOCKS_PER_GROUP(sb) 259 + fex->fe_start 260 + le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block); 261 return block; 262 } 263 #endif 264