LXR linux/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 */
  28#ifdef CONFIG_EXT4_DEBUG
  29extern ushort ext4_mballoc_debug;
  30
  31#define mb_debug(n, fmt, ...)                                           \
  32do {                                                                    \
  33        if ((n) <= ext4_mballoc_debug) {                                \
  34                printk(KERN_DEBUG "(%s, %d): %s: " fmt,                 \
  35                       __FILE__, __LINE__, __func__, ##__VA_ARGS__);    \
  36        }                                                               \
  37} while (0)
  38#else
  39#define mb_debug(n, fmt, ...)   no_printk(fmt, ##__VA_ARGS__)
  40#endif
  41
  42#define EXT4_MB_HISTORY_ALLOC           1       /* allocation */
  43#define EXT4_MB_HISTORY_PREALLOC        2       /* preallocated blocks used */
  44
  45/*
  46 * How long mballoc can look for a best extent (in found extents)
  47 */
  48#define MB_DEFAULT_MAX_TO_SCAN          200
  49
  50/*
  51 * How long mballoc must look for a best extent
  52 */
  53#define MB_DEFAULT_MIN_TO_SCAN          10
  54
  55/*
  56 * with 'ext4_mb_stats' allocator will collect stats that will be
  57 * shown at umount. The collecting costs though!
  58 */
  59#define MB_DEFAULT_STATS                0
  60
  61/*
  62 * files smaller than MB_DEFAULT_STREAM_THRESHOLD are served
  63 * by the stream allocator, which purpose is to pack requests
  64 * as close each to other as possible to produce smooth I/O traffic
  65 * We use locality group prealloc space for stream request.
  66 * We can tune the same via /proc/fs/ext4/<parition>/stream_req
  67 */
  68#define MB_DEFAULT_STREAM_THRESHOLD     16      /* 64K */
  69
  70/*
  71 * for which requests use 2^N search using buddies
  72 */
  73#define MB_DEFAULT_ORDER2_REQS          2
  74
  75/*
  76 * default group prealloc size 512 blocks
  77 */
  78#define MB_DEFAULT_GROUP_PREALLOC       512
  79
  80
  81struct ext4_free_data {
  82        /* this links the free block information from sb_info */
  83        struct list_head                efd_list;
  84
  85        /* this links the free block information from group_info */
  86        struct rb_node                  efd_node;
  87
  88        /* group which free block extent belongs */
  89        ext4_group_t                    efd_group;
  90
  91        /* free block extent */
  92        ext4_grpblk_t                   efd_start_cluster;
  93        ext4_grpblk_t                   efd_count;
  94
  95        /* transaction which freed this extent */
  96        tid_t                           efd_tid;
  97};
  98
  99struct ext4_prealloc_space {
 100        struct list_head        pa_inode_list;
 101        struct list_head        pa_group_list;
 102        union {
 103                struct list_head pa_tmp_list;
 104                struct rcu_head pa_rcu;
 105        } u;
 106        spinlock_t              pa_lock;
 107        atomic_t                pa_count;
 108        unsigned                pa_deleted;
 109        ext4_fsblk_t            pa_pstart;      /* phys. block */
 110        ext4_lblk_t             pa_lstart;      /* log. block */
 111        ext4_grpblk_t           pa_len;         /* len of preallocated chunk */
 112        ext4_grpblk_t           pa_free;        /* how many blocks are free */
 113        unsigned short          pa_type;        /* pa type. inode or group */
 114        spinlock_t              *pa_obj_lock;
 115        struct inode            *pa_inode;      /* hack, for history only */
 116};
 117
 118enum {
 119        MB_INODE_PA = 0,
 120        MB_GROUP_PA = 1
 121};
 122
 123struct ext4_free_extent {
 124        ext4_lblk_t fe_logical;
 125        ext4_grpblk_t fe_start; /* In cluster units */
 126        ext4_group_t fe_group;
 127        ext4_grpblk_t fe_len;   /* In cluster units */
 128};
 129
 130/*
 131 * Locality group:
 132 *   we try to group all related changes together
 133 *   so that writeback can flush/allocate them together as well
 134 *   Size of lg_prealloc_list hash is determined by MB_DEFAULT_GROUP_PREALLOC
 135 *   (512). We store prealloc space into the hash based on the pa_free blocks
 136 *   order value.ie, fls(pa_free)-1;
 137 */
 138#define PREALLOC_TB_SIZE 10
 139struct ext4_locality_group {
 140        /* for allocator */
 141        /* to serialize allocates */
 142        struct mutex            lg_mutex;
 143        /* list of preallocations */
 144        struct list_head        lg_prealloc_list[PREALLOC_TB_SIZE];
 145        spinlock_t              lg_prealloc_lock;
 146};
 147
 148struct ext4_allocation_context {
 149        struct inode *ac_inode;
 150        struct super_block *ac_sb;
 151
 152        /* original request */
 153        struct ext4_free_extent ac_o_ex;
 154
 155        /* goal request (normalized ac_o_ex) */
 156        struct ext4_free_extent ac_g_ex;
 157
 158        /* the best found extent */
 159        struct ext4_free_extent ac_b_ex;
 160
 161        /* copy of the best found extent taken before preallocation efforts */
 162        struct ext4_free_extent ac_f_ex;
 163
 164        __u16 ac_groups_scanned;
 165        __u16 ac_found;
 166        __u16 ac_tail;
 167        __u16 ac_buddy;
 168        __u16 ac_flags;         /* allocation hints */
 169        __u8 ac_status;
 170        __u8 ac_criteria;
 171        __u8 ac_2order;         /* if request is to allocate 2^N blocks and
 172                                 * N > 0, the field stores N, otherwise 0 */
 173        __u8 ac_op;             /* operation, for history only */
 174        struct page *ac_bitmap_page;
 175        struct page *ac_buddy_page;
 176        struct ext4_prealloc_space *ac_pa;
 177        struct ext4_locality_group *ac_lg;
 178};
 179
 180#define AC_STATUS_CONTINUE      1
 181#define AC_STATUS_FOUND         2
 182#define AC_STATUS_BREAK         3
 183
 184struct ext4_buddy {
 185        struct page *bd_buddy_page;
 186        void *bd_buddy;
 187        struct page *bd_bitmap_page;
 188        void *bd_bitmap;
 189        struct ext4_group_info *bd_info;
 190        struct super_block *bd_sb;
 191        __u16 bd_blkbits;
 192        ext4_group_t bd_group;
 193};
 194
 195static inline ext4_fsblk_t ext4_grp_offs_to_block(struct super_block *sb,
 196                                        struct ext4_free_extent *fex)
 197{
 198        return ext4_group_first_block_no(sb, fex->fe_group) +
 199                (fex->fe_start << EXT4_SB(sb)->s_cluster_bits);
 200}
 201
 202typedef int (*ext4_mballoc_query_range_fn)(
 203        struct super_block              *sb,
 204        ext4_group_t                    agno,
 205        ext4_grpblk_t                   start,
 206        ext4_grpblk_t                   len,
 207        void                            *priv);
 208
 209int
 210ext4_mballoc_query_range(
 211        struct super_block              *sb,
 212        ext4_group_t                    agno,
 213        ext4_grpblk_t                   start,
 214        ext4_grpblk_t                   end,
 215        ext4_mballoc_query_range_fn     formatter,
 216        void                            *priv);
 217
 218#endif
 219