fs/ext4/mballoc.h

   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 struct ext4_free_data {
  78         /* this links the free block information from sb_info */
  79         struct list_head                efd_list;
  80
  81         /* this links the free block information from group_info */
  82         struct rb_node                  efd_node;
  83
  84         /* group which free block extent belongs */
  85         ext4_group_t                    efd_group;
  86
  87         /* free block extent */
  88         ext4_grpblk_t                   efd_start_cluster;
  89         ext4_grpblk_t                   efd_count;
  90
  91         /* transaction which freed this extent */
  92         tid_t                           efd_tid;
  93 };
  94
  95 struct ext4_prealloc_space {
  96         struct list_head        pa_inode_list;
  97         struct list_head        pa_group_list;
  98         union {
  99                 struct list_head pa_tmp_list;
 100                 struct rcu_head pa_rcu;
 101         } u;
 102         spinlock_t              pa_lock;
 103         atomic_t                pa_count;
 104         unsigned                pa_deleted;
 105         ext4_fsblk_t            pa_pstart;      /* phys. block */
 106         ext4_lblk_t             pa_lstart;      /* log. block */
 107         ext4_grpblk_t           pa_len;         /* len of preallocated chunk */
 108         ext4_grpblk_t           pa_free;        /* how many blocks are free */
 109         unsigned short          pa_type;        /* pa type. inode or group */
 110         spinlock_t              *pa_obj_lock;
 111         struct inode            *pa_inode;      /* hack, for history only */
 112 };
 113
 114 enum {
 115         MB_INODE_PA = 0,
 116         MB_GROUP_PA = 1
 117 };
 118
 119 struct ext4_free_extent {
 120         ext4_lblk_t fe_logical;
 121         ext4_grpblk_t fe_start; /* In cluster units */
 122         ext4_group_t fe_group;
 123         ext4_grpblk_t fe_len;   /* In cluster units */
 124 };
 125
 126 /*
 127  * Locality group:
 128  *   we try to group all related changes together
 129  *   so that writeback can flush/allocate them together as well
 130  *   Size of lg_prealloc_list hash is determined by MB_DEFAULT_GROUP_PREALLOC
 131  *   (512). We store prealloc space into the hash based on the pa_free blocks
 132  *   order value.ie, fls(pa_free)-1;
 133  */
 134 #define PREALLOC_TB_SIZE 10
 135 struct ext4_locality_group {
 136         /* for allocator */
 137         /* to serialize allocates */
 138         struct mutex            lg_mutex;
 139         /* list of preallocations */
 140         struct list_head        lg_prealloc_list[PREALLOC_TB_SIZE];
 141         spinlock_t              lg_prealloc_lock;
 142 };
 143
 144 struct ext4_allocation_context {
 145         struct inode *ac_inode;
 146         struct super_block *ac_sb;
 147
 148         /* original request */
 149         struct ext4_free_extent ac_o_ex;
 150
 151         /* goal request (normalized ac_o_ex) */
 152         struct ext4_free_extent ac_g_ex;
 153
 154         /* the best found extent */
 155         struct ext4_free_extent ac_b_ex;
 156
 157         /* copy of the best found extent taken before preallocation efforts */
 158         struct ext4_free_extent ac_f_ex;
 159
 160         __u16 ac_groups_scanned;
 161         __u16 ac_found;
 162         __u16 ac_tail;
 163         __u16 ac_buddy;
 164         __u16 ac_flags;         /* allocation hints */
 165         __u8 ac_status;
 166         __u8 ac_criteria;
 167         __u8 ac_2order;         /* if request is to allocate 2^N blocks and
 168                                  * N > 0, the field stores N, otherwise 0 */
 169         __u8 ac_op;             /* operation, for history only */
 170         struct page *ac_bitmap_page;
 171         struct page *ac_buddy_page;
 172         struct ext4_prealloc_space *ac_pa;
 173         struct ext4_locality_group *ac_lg;
 174 };
 175
 176 #define AC_STATUS_CONTINUE      1
 177 #define AC_STATUS_FOUND         2
 178 #define AC_STATUS_BREAK         3
 179
 180 struct ext4_buddy {
 181         struct page *bd_buddy_page;
 182         void *bd_buddy;
 183         struct page *bd_bitmap_page;
 184         void *bd_bitmap;
 185         struct ext4_group_info *bd_info;
 186         struct super_block *bd_sb;
 187         __u16 bd_blkbits;
 188         ext4_group_t bd_group;
 189 };
 190
 191 static inline ext4_fsblk_t ext4_grp_offs_to_block(struct super_block *sb,
 192                                         struct ext4_free_extent *fex)
 193 {
 194         return ext4_group_first_block_no(sb, fex->fe_group) +
 195                 (fex->fe_start << EXT4_SB(sb)->s_cluster_bits);
 196 }
 197
 198 typedef int (*ext4_mballoc_query_range_fn)(
 199         struct super_block              *sb,
 200         ext4_group_t                    agno,
 201         ext4_grpblk_t                   start,
 202         ext4_grpblk_t                   len,
 203         void                            *priv);
 204
 205 int
 206 ext4_mballoc_query_range(
 207         struct super_block              *sb,
 208         ext4_group_t                    agno,
 209         ext4_grpblk_t                   start,
 210         ext4_grpblk_t                   end,
 211         ext4_mballoc_query_range_fn     formatter,
 212         void                            *priv);
 213
 214 #endif