1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * fs/ext4/mballoc.h 4 * 5 * Written by: Alex Tomas <alex@clusterfs.com> 6 * 7 */ 8 #ifndef _EXT4_MBALLOC_H 9 #define _EXT4_MBALLOC_H 10 11 #include <linux/time.h> 12 #include <linux/fs.h> 13 #include <linux/namei.h> 14 #include <linux/quotaops.h> 15 #include <linux/buffer_head.h> 16 #include <linux/module.h> 17 #include <linux/swap.h> 18 #include <linux/proc_fs.h> 19 #include <linux/pagemap.h> 20 #include <linux/seq_file.h> 21 #include <linux/blkdev.h> 22 #include <linux/mutex.h> 23 #include "ext4_jbd2.h" 24 #include "ext4.h" 25 26 /* 27 * mb_debug() dynamic printk msgs could be used to debug mballoc code. 28 */ 29 #ifdef CONFIG_EXT4_DEBUG 30 #define mb_debug(sb, fmt, ...) \ 31 pr_debug("[%s/%d] EXT4-fs (%s): (%s, %d): %s: " fmt, \ 32 current->comm, task_pid_nr(current), sb->s_id, \ 33 __FILE__, __LINE__, __func__, ##__VA_ARGS__) 34 #else 35 #define mb_debug(sb, fmt, ...) no_printk(fmt, ##__VA_ARGS__) 36 #endif 37 38 #define EXT4_MB_HISTORY_ALLOC 1 /* allocation */ 39 #define EXT4_MB_HISTORY_PREALLOC 2 /* preallocated blocks used */ 40 41 /* 42 * How long mballoc can look for a best extent (in found extents) 43 */ 44 #define MB_DEFAULT_MAX_TO_SCAN 200 45 46 /* 47 * How long mballoc must look for a best extent 48 */ 49 #define MB_DEFAULT_MIN_TO_SCAN 10 50 51 /* 52 * with 'ext4_mb_stats' allocator will collect stats that will be 53 * shown at umount. The collecting costs though! 54 */ 55 #define MB_DEFAULT_STATS 0 56 57 /* 58 * files smaller than MB_DEFAULT_STREAM_THRESHOLD are served 59 * by the stream allocator, which purpose is to pack requests 60 * as close each to other as possible to produce smooth I/O traffic 61 * We use locality group prealloc space for stream request. 62 * We can tune the same via /proc/fs/ext4/<partition>/stream_req 63 */ 64 #define MB_DEFAULT_STREAM_THRESHOLD 16 /* 64K */ 65 66 /* 67 * for which requests use 2^N search using buddies 68 */ 69 #define MB_DEFAULT_ORDER2_REQS 2 70 71 /* 72 * default group prealloc size 512 blocks 73 */ 74 #define MB_DEFAULT_GROUP_PREALLOC 512 75 76 /* 77 * maximum length of inode prealloc list 78 */ 79 #define MB_DEFAULT_MAX_INODE_PREALLOC 512 80 81 /* 82 * Number of groups to search linearly before performing group scanning 83 * optimization. 84 */ 85 #define MB_DEFAULT_LINEAR_LIMIT 4 86 87 /* 88 * Minimum number of groups that should be present in the file system to perform 89 * group scanning optimizations. 90 */ 91 #define MB_DEFAULT_LINEAR_SCAN_THRESHOLD 16 92 93 /* 94 * Number of valid buddy orders 95 */ 96 #define MB_NUM_ORDERS(sb) ((sb)->s_blocksize_bits + 2) 97 98 struct ext4_free_data { 99 /* this links the free block information from sb_info */ 100 struct list_head efd_list; 101 102 /* this links the free block information from group_info */ 103 struct rb_node efd_node; 104 105 /* group which free block extent belongs */ 106 ext4_group_t efd_group; 107 108 /* free block extent */ 109 ext4_grpblk_t efd_start_cluster; 110 ext4_grpblk_t efd_count; 111 112 /* transaction which freed this extent */ 113 tid_t efd_tid; 114 }; 115 116 struct ext4_prealloc_space { 117 union { 118 struct list_head inode_list; /* for inode PAs */ 119 struct list_head lg_list; /* for lg PAs */ 120 } pa_node; 121 struct list_head pa_group_list; 122 union { 123 struct list_head pa_tmp_list; 124 struct rcu_head pa_rcu; 125 } u; 126 spinlock_t pa_lock; 127 atomic_t pa_count; 128 unsigned pa_deleted; 129 ext4_fsblk_t pa_pstart; /* phys. block */ 130 ext4_lblk_t pa_lstart; /* log. block */ 131 ext4_grpblk_t pa_len; /* len of preallocated chunk */ 132 ext4_grpblk_t pa_free; /* how many blocks are free */ 133 unsigned short pa_type; /* pa type. inode or group */ 134 union { 135 spinlock_t *inode_lock; /* locks the inode list holding this PA */ 136 spinlock_t *lg_lock; /* locks the lg list holding this PA */ 137 } pa_node_lock; 138 struct inode *pa_inode; /* hack, for history only */ 139 }; 140 141 enum { 142 MB_INODE_PA = 0, 143 MB_GROUP_PA = 1 144 }; 145 146 struct ext4_free_extent { 147 ext4_lblk_t fe_logical; 148 ext4_grpblk_t fe_start; /* In cluster units */ 149 ext4_group_t fe_group; 150 ext4_grpblk_t fe_len; /* In cluster units */ 151 }; 152 153 /* 154 * Locality group: 155 * we try to group all related changes together 156 * so that writeback can flush/allocate them together as well 157 * Size of lg_prealloc_list hash is determined by MB_DEFAULT_GROUP_PREALLOC 158 * (512). We store prealloc space into the hash based on the pa_free blocks 159 * order value.ie, fls(pa_free)-1; 160 */ 161 #define PREALLOC_TB_SIZE 10 162 struct ext4_locality_group { 163 /* for allocator */ 164 /* to serialize allocates */ 165 struct mutex lg_mutex; 166 /* list of preallocations */ 167 struct list_head lg_prealloc_list[PREALLOC_TB_SIZE]; 168 spinlock_t lg_prealloc_lock; 169 }; 170 171 struct ext4_allocation_context { 172 struct inode *ac_inode; 173 struct super_block *ac_sb; 174 175 /* original request */ 176 struct ext4_free_extent ac_o_ex; 177 178 /* goal request (normalized ac_o_ex) */ 179 struct ext4_free_extent ac_g_ex; 180 181 /* the best found extent */ 182 struct ext4_free_extent ac_b_ex; 183 184 /* copy of the best found extent taken before preallocation efforts */ 185 struct ext4_free_extent ac_f_ex; 186 187 __u32 ac_groups_considered; 188 __u32 ac_flags; /* allocation hints */ 189 __u16 ac_groups_scanned; 190 __u16 ac_groups_linear_remaining; 191 __u16 ac_found; 192 __u16 ac_tail; 193 __u16 ac_buddy; 194 __u8 ac_status; 195 __u8 ac_criteria; 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 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 3 208 209 struct 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; 218 }; 219 220 static inline ext4_fsblk_t ext4_grp_offs_to_block(struct super_block *sb, 221 struct ext4_free_extent *fex) 222 { 223 return ext4_group_first_block_no(sb, fex->fe_group) + 224 (fex->fe_start << EXT4_SB(sb)->s_cluster_bits); 225 } 226 227 typedef int (*ext4_mballoc_query_range_fn)( 228 struct super_block *sb, 229 ext4_group_t agno, 230 ext4_grpblk_t start, 231 ext4_grpblk_t len, 232 void *priv); 233 234 int 235 ext4_mballoc_query_range( 236 struct super_block *sb, 237 ext4_group_t agno, 238 ext4_grpblk_t start, 239 ext4_grpblk_t end, 240 ext4_mballoc_query_range_fn formatter, 241 void *priv); 242 243 #endif 244