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/<parition>/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 valid buddy orders 83 */ 84 #define MB_NUM_ORDERS(sb) ((sb)->s_blocksize_bits + 2) 85 86 struct ext4_free_data { 87 /* this links the free block information from sb_info */ 88 struct list_head efd_list; 89 90 /* this links the free block information from group_info */ 91 struct rb_node efd_node; 92 93 /* group which free block extent belongs */ 94 ext4_group_t efd_group; 95 96 /* free block extent */ 97 ext4_grpblk_t efd_start_cluster; 98 ext4_grpblk_t efd_count; 99 100 /* transaction which freed this extent */ 101 tid_t efd_tid; 102 }; 103 104 struct ext4_prealloc_space { 105 struct list_head pa_inode_list; 106 struct list_head pa_group_list; 107 union { 108 struct list_head pa_tmp_list; 109 struct rcu_head pa_rcu; 110 } u; 111 spinlock_t pa_lock; 112 atomic_t pa_count; 113 unsigned pa_deleted; 114 ext4_fsblk_t pa_pstart; /* phys. block */ 115 ext4_lblk_t pa_lstart; /* log. block */ 116 ext4_grpblk_t pa_len; /* len of preallocated chunk */ 117 ext4_grpblk_t pa_free; /* how many blocks are free */ 118 unsigned short pa_type; /* pa type. inode or group */ 119 spinlock_t *pa_obj_lock; 120 struct inode *pa_inode; /* hack, for history only */ 121 }; 122 123 enum { 124 MB_INODE_PA = 0, 125 MB_GROUP_PA = 1 126 }; 127 128 struct ext4_free_extent { 129 ext4_lblk_t fe_logical; 130 ext4_grpblk_t fe_start; /* In cluster units */ 131 ext4_group_t fe_group; 132 ext4_grpblk_t fe_len; /* In cluster units */ 133 }; 134 135 /* 136 * Locality group: 137 * we try to group all related changes together 138 * so that writeback can flush/allocate them together as well 139 * Size of lg_prealloc_list hash is determined by MB_DEFAULT_GROUP_PREALLOC 140 * (512). We store prealloc space into the hash based on the pa_free blocks 141 * order value.ie, fls(pa_free)-1; 142 */ 143 #define PREALLOC_TB_SIZE 10 144 struct ext4_locality_group { 145 /* for allocator */ 146 /* to serialize allocates */ 147 struct mutex lg_mutex; 148 /* list of preallocations */ 149 struct list_head lg_prealloc_list[PREALLOC_TB_SIZE]; 150 spinlock_t lg_prealloc_lock; 151 }; 152 153 struct ext4_allocation_context { 154 struct inode *ac_inode; 155 struct super_block *ac_sb; 156 157 /* original request */ 158 struct ext4_free_extent ac_o_ex; 159 160 /* goal request (normalized ac_o_ex) */ 161 struct ext4_free_extent ac_g_ex; 162 163 /* the best found extent */ 164 struct ext4_free_extent ac_b_ex; 165 166 /* copy of the best found extent taken before preallocation efforts */ 167 struct ext4_free_extent ac_f_ex; 168 169 __u16 ac_groups_scanned; 170 __u16 ac_found; 171 __u16 ac_tail; 172 __u16 ac_buddy; 173 __u16 ac_flags; /* allocation hints */ 174 __u8 ac_status; 175 __u8 ac_criteria; 176 __u8 ac_2order; /* if request is to allocate 2^N blocks and 177 * N > 0, the field stores N, otherwise 0 */ 178 __u8 ac_op; /* operation, for history only */ 179 struct page *ac_bitmap_page; 180 struct page *ac_buddy_page; 181 struct ext4_prealloc_space *ac_pa; 182 struct ext4_locality_group *ac_lg; 183 }; 184 185 #define AC_STATUS_CONTINUE 1 186 #define AC_STATUS_FOUND 2 187 #define AC_STATUS_BREAK 3 188 189 struct ext4_buddy { 190 struct page *bd_buddy_page; 191 void *bd_buddy; 192 struct page *bd_bitmap_page; 193 void *bd_bitmap; 194 struct ext4_group_info *bd_info; 195 struct super_block *bd_sb; 196 __u16 bd_blkbits; 197 ext4_group_t bd_group; 198 }; 199 200 static inline ext4_fsblk_t ext4_grp_offs_to_block(struct super_block *sb, 201 struct ext4_free_extent *fex) 202 { 203 return ext4_group_first_block_no(sb, fex->fe_group) + 204 (fex->fe_start << EXT4_SB(sb)->s_cluster_bits); 205 } 206 207 typedef int (*ext4_mballoc_query_range_fn)( 208 struct super_block *sb, 209 ext4_group_t agno, 210 ext4_grpblk_t start, 211 ext4_grpblk_t len, 212 void *priv); 213 214 int 215 ext4_mballoc_query_range( 216 struct super_block *sb, 217 ext4_group_t agno, 218 ext4_grpblk_t start, 219 ext4_grpblk_t end, 220 ext4_mballoc_query_range_fn formatter, 221 void *priv); 222 223 #endif 224