ldiskfs/kernel_patches/patches/rhel7.6/ext4-simple-blockalloc.patch

   1 Index: linux-stage/fs/ext4/mballoc.c
   2 ===================================================================
   3 --- linux-stage.orig/fs/ext4/mballoc.c
   4 +++ linux-stage/fs/ext4/mballoc.c
   5 @@ -2078,6 +2078,21 @@ static int ext4_mb_good_group(struct ext
   6         return 0;
   7  }
   8
   9 +static u64 available_blocks_count(struct ext4_sb_info *sbi)
  10 +{
  11 +       ext4_fsblk_t resv_blocks;
  12 +       u64 bfree;
  13 +       struct ext4_super_block *es = sbi->s_es;
  14 +
  15 +       resv_blocks = EXT4_C2B(sbi, atomic64_read(&sbi->s_resv_clusters));
  16 +       bfree = percpu_counter_sum_positive(&sbi->s_freeclusters_counter) -
  17 +                percpu_counter_sum_positive(&sbi->s_dirtyclusters_counter);
  18 +
  19 +       bfree = EXT4_C2B(sbi, max_t(s64, bfree, 0));
  20 +       return bfree - (ext4_r_blocks_count(es) + resv_blocks);
  21 +}
  22 +
  23 +
  24  static noinline_for_stack int
  25  ext4_mb_regular_allocator(struct ext4_allocation_context *ac)
  26  {
  27 @@ -2087,6 +2102,7 @@ ext4_mb_regular_allocator(struct ext4_al
  28         struct ext4_sb_info *sbi;
  29         struct super_block *sb;
  30         struct ext4_buddy e4b;
  31 +       ext4_fsblk_t avail_blocks;
  32
  33         sb = ac->ac_sb;
  34         sbi = EXT4_SB(sb);
  35 @@ -2136,6 +2152,21 @@ ext4_mb_regular_allocator(struct ext4_al
  36
  37         /* Let's just scan groups to find more-less suitable blocks */
  38         cr = ac->ac_2order ? 0 : 1;
  39 +
  40 +       /* Choose what loop to pass based on disk fullness */
  41 +       avail_blocks = available_blocks_count(sbi) ;
  42 +
  43 +       if (avail_blocks < sbi->s_mb_c3_blocks) {
  44 +               cr = 3;
  45 +               atomic64_inc(&sbi->s_bal_cX_skipped[2]);
  46 +       } else if(avail_blocks < sbi->s_mb_c2_blocks) {
  47 +               cr = 2;
  48 +               atomic64_inc(&sbi->s_bal_cX_skipped[1]);
  49 +       } else if(avail_blocks < sbi->s_mb_c1_blocks) {
  50 +               cr = 1;
  51 +               atomic64_inc(&sbi->s_bal_cX_skipped[0]);
  52 +       }
  53 +
  54         /*
  55          * cr == 0 try to get exact allocation,
  56          * cr == 3  try to get anything
  57 @@ -2193,6 +2224,9 @@ repeat:
  58                         if (ac->ac_status != AC_STATUS_CONTINUE)
  59                                 break;
  60                 }
  61 +               /* Processed all groups and haven't found blocks */
  62 +               if (i == ngroups)
  63 +                       atomic64_inc(&sbi->s_bal_cX_failed[cr]);
  64         }
  65
  66         if (ac->ac_b_ex.fe_len > 0 && ac->ac_status != AC_STATUS_FOUND &&
  67 @@ -2316,6 +2350,93 @@ static const struct seq_operations ext4_
  68         .show   = ext4_mb_seq_groups_show,
  69  };
  70
  71 +static int mb_seq_alloc_show(struct seq_file *seq, void *v)
  72 +{
  73 +       struct super_block *sb = seq->private;
  74 +       struct ext4_sb_info *sbi = EXT4_SB(sb);
  75 +
  76 +       seq_printf(seq, "mballoc:\n");
  77 +       seq_printf(seq, "\tblocks: %u\n", atomic_read(&sbi->s_bal_allocated));
  78 +       seq_printf(seq, "\treqs: %u\n", atomic_read(&sbi->s_bal_reqs));
  79 +       seq_printf(seq, "\tsuccess: %u\n", atomic_read(&sbi->s_bal_success));
  80 +
  81 +       seq_printf(seq, "\textents_scanned: %u\n",
  82 +                  atomic_read(&sbi->s_bal_ex_scanned));
  83 +       seq_printf(seq, "\t\tgoal_hits: %u\n", atomic_read(&sbi->s_bal_goals));
  84 +       seq_printf(seq, "\t\t2^n_hits: %u\n", atomic_read(&sbi->s_bal_2orders));
  85 +       seq_printf(seq, "\t\tbreaks: %u\n", atomic_read(&sbi->s_bal_breaks));
  86 +       seq_printf(seq, "\t\tlost: %u\n", atomic_read(&sbi->s_mb_lost_chunks));
  87 +
  88 +       seq_printf(seq, "\tuseless_c1_loops: %lu\n",
  89 +                  atomic64_read(&sbi->s_bal_cX_failed[0]));
  90 +       seq_printf(seq, "\tuseless_c2_loops: %lu\n",
  91 +                  atomic64_read(&sbi->s_bal_cX_failed[1]));
  92 +       seq_printf(seq, "\tuseless_c3_loops: %lu\n",
  93 +                  atomic64_read(&sbi->s_bal_cX_failed[2]));
  94 +       seq_printf(seq, "\tskipped_c1_loops: %lu\n",
  95 +                  atomic64_read(&sbi->s_bal_cX_skipped[0]));
  96 +       seq_printf(seq, "\tskipped_c2_loops: %lu\n",
  97 +                  atomic64_read(&sbi->s_bal_cX_skipped[1]));
  98 +       seq_printf(seq, "\tskipped_c3_loops: %lu\n",
  99 +                  atomic64_read(&sbi->s_bal_cX_skipped[2]));
 100 +       seq_printf(seq, "\tbuddies_generated: %lu\n",
 101 +                  sbi->s_mb_buddies_generated);
 102 +       seq_printf(seq, "\tbuddies_time_used: %llu\n", sbi->s_mb_generation_time);
 103 +       seq_printf(seq, "\tpreallocated: %u\n",
 104 +                  atomic_read(&sbi->s_mb_preallocated));
 105 +       seq_printf(seq, "\tdiscarded: %u\n",
 106 +                  atomic_read(&sbi->s_mb_discarded));
 107 +       return 0;
 108 +}
 109 +
 110 +static ssize_t mb_seq_alloc_write(struct file *file,
 111 +                             const char __user *buf,
 112 +                             size_t cnt, loff_t *pos)
 113 +{
 114 +       struct ext4_sb_info *sbi = EXT4_SB(PDE_DATA(file_inode(file)));
 115 +
 116 +       atomic_set(&sbi->s_bal_allocated, 0),
 117 +       atomic_set(&sbi->s_bal_reqs, 0),
 118 +       atomic_set(&sbi->s_bal_success, 0);
 119 +
 120 +       atomic_set(&sbi->s_bal_ex_scanned, 0),
 121 +       atomic_set(&sbi->s_bal_goals, 0),
 122 +       atomic_set(&sbi->s_bal_2orders, 0),
 123 +       atomic_set(&sbi->s_bal_breaks, 0),
 124 +       atomic_set(&sbi->s_mb_lost_chunks, 0);
 125 +
 126 +       atomic64_set(&sbi->s_bal_cX_failed[0], 0),
 127 +       atomic64_set(&sbi->s_bal_cX_failed[1], 0),
 128 +       atomic64_set(&sbi->s_bal_cX_failed[2], 0);
 129 +
 130 +       atomic64_set(&sbi->s_bal_cX_skipped[0], 0),
 131 +       atomic64_set(&sbi->s_bal_cX_skipped[1], 0),
 132 +       atomic64_set(&sbi->s_bal_cX_skipped[2], 0);
 133 +
 134 +
 135 +       sbi->s_mb_buddies_generated = 0;
 136 +       sbi->s_mb_generation_time = 0;
 137 +
 138 +       atomic_set(&sbi->s_mb_preallocated, 0),
 139 +       atomic_set(&sbi->s_mb_discarded, 0);
 140 +
 141 +       return cnt;
 142 +}
 143 +
 144 +static int mb_seq_alloc_open(struct inode *inode, struct file *file)
 145 +{
 146 +       return single_open(file, mb_seq_alloc_show, PDE_DATA(inode));
 147 +}
 148 +
 149 +static const struct file_operations ext4_mb_seq_alloc_fops = {
 150 +       .owner          = THIS_MODULE,
 151 +       .open           = mb_seq_alloc_open,
 152 +       .read           = seq_read,
 153 +       .llseek         = seq_lseek,
 154 +       .release        = single_release,
 155 +       .write          = mb_seq_alloc_write,
 156 +};
 157 +
 158  #define EXT4_MB_PREALLOC_TABLE          "prealloc_table"
 159
 160  static int ext4_mb_check_and_update_prealloc(struct ext4_sb_info *sbi,
 161 @@ -2730,6 +2851,8 @@ static int ext4_groupinfo_create_slab(si
 162         return 0;
 163  }
 164
 165 +#define THRESHOLD_BLOCKS(sbi, percent)                                 \
 166 +       (ext4_blocks_count((sbi)->s_es) / 100 * (percent))
 167  int ext4_mb_init(struct super_block *sb)
 168  {
 169         struct ext4_sb_info *sbi = EXT4_SB(sb);
 170 @@ -2781,6 +2903,15 @@ int ext4_mb_init(struct super_block *sb)
 171         sbi->s_mb_min_to_scan = MB_DEFAULT_MIN_TO_SCAN;
 172         sbi->s_mb_stats = MB_DEFAULT_STATS;
 173         sbi->s_mb_order2_reqs = MB_DEFAULT_ORDER2_REQS;
 174 +       if (!sbi->s_mb_c1_blocks)
 175 +               sbi->s_mb_c1_blocks =
 176 +                       THRESHOLD_BLOCKS(sbi, MB_DEFAULT_C1_THRESHOLD);
 177 +       if (!sbi->s_mb_c2_blocks)
 178 +               sbi->s_mb_c2_blocks =
 179 +                       THRESHOLD_BLOCKS(sbi, MB_DEFAULT_C2_THRESHOLD);
 180 +       if (!sbi->s_mb_c3_blocks)
 181 +               sbi->s_mb_c3_blocks =
 182 +                       THRESHOLD_BLOCKS(sbi, MB_DEFAULT_C3_THRESHOLD);
 183         /*
 184          * The default group preallocation is 512, which for 4k block
 185          * sizes translates to 2 megabytes.  However for bigalloc file
 186 @@ -2853,6 +2978,8 @@ int ext4_mb_init(struct super_block *sb)
 187                 proc_create_data(EXT4_MB_PREALLOC_TABLE, S_IFREG | S_IRUGO |
 188                                  S_IWUSR, sbi->s_proc,
 189                                  &ext4_mb_prealloc_seq_fops, sb);
 190 +               proc_create_data("mb_alloc_stats", S_IFREG | S_IRUGO | S_IWUSR,
 191 +                                sbi->s_proc, &ext4_mb_seq_alloc_fops, sb);
 192                 proc_create_data("mb_last_group", S_IFREG | S_IRUGO |
 193                                  S_IWUSR, sbi->s_proc,
 194                                  &ext4_mb_seq_last_group_fops, sb);
 195 @@ -2906,6 +3033,7 @@ int ext4_mb_release(struct super_block *
 196                 remove_proc_entry("mb_last_group", sbi->s_proc);
 197                 remove_proc_entry("mb_last_start", sbi->s_proc);
 198                 remove_proc_entry(EXT4_MB_PREALLOC_TABLE, sbi->s_proc);
 199 +               remove_proc_entry("mb_alloc_stats", sbi->s_proc);
 200         }
 201
 202         if (sbi->s_group_info) {
 203 @@ -2936,6 +3064,16 @@ int ext4_mb_release(struct super_block *
 204                                 atomic_read(&sbi->s_bal_reqs),
 205                                 atomic_read(&sbi->s_bal_success));
 206                 ext4_msg(sb, KERN_INFO,
 207 +                       "mballoc: (%lu, %lu, %lu) useless c(0,1,2) loops",
 208 +                               atomic64_read(&sbi->s_bal_cX_failed[0]),
 209 +                               atomic64_read(&sbi->s_bal_cX_failed[1]),
 210 +                               atomic64_read(&sbi->s_bal_cX_failed[2]));
 211 +               ext4_msg(sb, KERN_INFO,
 212 +                       "mballoc: (%lu, %lu, %lu) skipped c(0,1,2) loops",
 213 +                               atomic64_read(&sbi->s_bal_cX_skipped[0]),
 214 +                               atomic64_read(&sbi->s_bal_cX_skipped[1]),
 215 +                               atomic64_read(&sbi->s_bal_cX_skipped[2]));
 216 +               ext4_msg(sb, KERN_INFO,
 217                       "mballoc: %u extents scanned, %u goal hits, "
 218                                 "%u 2^N hits, %u breaks, %u lost",
 219                                 atomic_read(&sbi->s_bal_ex_scanned),
 220 Index: linux-stage/fs/ext4/ext4.h
 221 ===================================================================
 222 --- linux-stage.orig/fs/ext4/ext4.h
 223 +++ linux-stage/fs/ext4/ext4.h
 224 @@ -1409,6 +1409,9 @@ struct ext4_sb_info {
 225         unsigned int s_mb_min_to_scan;
 226         unsigned int s_mb_stats;
 227         unsigned int s_mb_order2_reqs;
 228 +       ext4_fsblk_t s_mb_c1_blocks;
 229 +       ext4_fsblk_t s_mb_c2_blocks;
 230 +       ext4_fsblk_t s_mb_c3_blocks;
 231         unsigned long *s_mb_prealloc_table;
 232         unsigned int s_mb_group_prealloc;
 233         unsigned int s_max_dir_size_kb;
 234 @@ -1425,6 +1428,9 @@ struct ext4_sb_info {
 235         atomic_t s_bal_goals;   /* goal hits */
 236         atomic_t s_bal_breaks;  /* too long searches */
 237         atomic_t s_bal_2orders; /* 2^order hits */
 238 +       /* cX loop didn't find blocks */
 239 +       atomic64_t s_bal_cX_failed[3];
 240 +       atomic64_t s_bal_cX_skipped[3];
 241         spinlock_t s_bal_lock;
 242         unsigned long s_mb_buddies_generated;
 243         unsigned long long s_mb_generation_time;
 244 @@ -2115,6 +2121,8 @@ struct ext4_sb_info {
 245  extern void ext4_end_bitmap_read(struct buffer_head *bh, int uptodate);
 246
 247  /* mballoc.c */
 248 +extern int save_threshold_percent(struct ext4_sb_info *sbi, const char *buf,
 249 +                                 ext4_fsblk_t *blocks);
 250  extern long ext4_mb_stats;
 251  extern long ext4_mb_max_to_scan;
 252  extern int ext4_mb_init(struct super_block *);
 253 Index: linux-stage/fs/ext4/super.c
 254 ===================================================================
 255 --- linux-stage.orig/fs/ext4/super.c
 256 +++ linux-stage/fs/ext4/super.c
 257 @@ -1208,6 +1208,7 @@ enum {
 258         Opt_nomblk_io_submit, Opt_block_validity, Opt_noblock_validity,
 259         Opt_inode_readahead_blks, Opt_journal_ioprio,
 260         Opt_dioread_nolock, Opt_dioread_lock,
 261 +       Opt_mb_c1_threshold, Opt_mb_c2_threshold, Opt_mb_c3_threshold,
 262         Opt_no_mbcache,
 263         Opt_discard, Opt_nodiscard, Opt_init_itable, Opt_noinit_itable,
 264         Opt_max_dir_size_kb, Opt_nojournal_checksum,
 265 @@ -1287,6 +1288,9 @@ static const match_table_t tokens = {
 266         {Opt_nodiscard, "nodiscard"},
 267         {Opt_init_itable, "init_itable=%u"},
 268         {Opt_no_mbcache, "no_mbcache"},
 269 +       {Opt_mb_c1_threshold, "mb_c1_threshold=%s"},
 270 +       {Opt_mb_c2_threshold, "mb_c2_threshold=%s"},
 271 +       {Opt_mb_c3_threshold, "mb_c3_threshold=%s"},
 272         {Opt_init_itable, "init_itable"},
 273         {Opt_noinit_itable, "noinit_itable"},
 274         {Opt_max_dir_size_kb, "max_dir_size_kb=%u"},
 275 @@ -1449,6 +1453,9 @@ static const struct mount_opts {
 276         {Opt_auto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_CLEAR},
 277         {Opt_noinit_itable, EXT4_MOUNT_INIT_INODE_TABLE, MOPT_CLEAR},
 278         {Opt_no_mbcache, EXT4_MOUNT_NO_MBCACHE, MOPT_SET},
 279 +       {Opt_mb_c1_threshold, 0, MOPT_STRING},
 280 +       {Opt_mb_c2_threshold, 0, MOPT_STRING},
 281 +       {Opt_mb_c3_threshold, 0, MOPT_STRING},
 282         {Opt_commit, 0, MOPT_GTE0},
 283         {Opt_max_batch_time, 0, MOPT_GTE0},
 284         {Opt_min_batch_time, 0, MOPT_GTE0},
 285 @@ -1571,6 +1578,12 @@ static const struct mount_opts {
 286                 sbi->s_max_dir_size_kb = arg;
 287                 /* reset s_warning_dir_size and make it re-calculated */
 288                 sbi->s_warning_dir_size = 0;
 289 +       } else if (token == Opt_mb_c1_threshold) {
 290 +               save_threshold_percent(sbi, args[0].from, &sbi->s_mb_c1_blocks);
 291 +       } else if (token == Opt_mb_c2_threshold) {
 292 +               save_threshold_percent(sbi, args[0].from, &sbi->s_mb_c2_blocks);
 293 +       } else if (token == Opt_mb_c3_threshold) {
 294 +               save_threshold_percent(sbi, args[0].from, &sbi->s_mb_c3_blocks);
 295         } else if (token == Opt_stripe) {
 296                 sbi->s_stripe = arg;
 297         } else if (token == Opt_resuid) {
 298 @@ -2734,6 +2747,74 @@ static ssize_t sbi_deprecated_show(struc
 299         return snprintf(buf, PAGE_SIZE, "%d\n", a->u.deprecated_val);
 300  }
 301
 302 +int save_threshold_percent(struct ext4_sb_info *sbi, const char *buf,
 303 +                          ext4_fsblk_t *blocks) {
 304 +       unsigned long long val;
 305 +
 306 +       if (!parse_strtoull(buf, 100, &val) && val <= 100) {
 307 +               *blocks = val * ext4_blocks_count(sbi->s_es) / 100;
 308 +               return 0;
 309 +       }
 310 +
 311 +       return -EINVAL;
 312 +}
 313 +
 314 +#define THRESHOLD_PERCENT(sbi, blocks)                                 \
 315 +       (((blocks) - 1) * 100 / ext4_blocks_count((sbi)->s_es) + 1)
 316 +static ssize_t mb_c1_threshold_store(struct ext4_attr *a,
 317 +                                   struct ext4_sb_info *sbi,
 318 +                                   const char *buf, size_t count)
 319 +{
 320 +       int ret;
 321 +
 322 +       ret = save_threshold_percent(sbi, buf, &sbi->s_mb_c1_blocks);
 323 +
 324 +       return ret ? ret : count;
 325 +}
 326 +
 327 +static ssize_t mb_c1_threshold_show(struct ext4_attr *a,
 328 +                                  struct ext4_sb_info *sbi, char *buf)
 329 +{
 330 +       return snprintf(buf, PAGE_SIZE, "%llu\n",
 331 +                       THRESHOLD_PERCENT(sbi, sbi->s_mb_c1_blocks));
 332 +}
 333 +
 334 +static ssize_t mb_c2_threshold_store(struct ext4_attr *a,
 335 +                                   struct ext4_sb_info *sbi,
 336 +                                   const char *buf, size_t count)
 337 +{
 338 +       int ret;
 339 +
 340 +       ret = save_threshold_percent(sbi, buf, &sbi->s_mb_c2_blocks);
 341 +       return ret ? ret : count;
 342 +}
 343 +
 344 +static ssize_t mb_c2_threshold_show(struct ext4_attr *a,
 345 +                                  struct ext4_sb_info *sbi, char *buf)
 346 +{
 347 +               return snprintf(buf, PAGE_SIZE, "%llu\n",
 348 +                               THRESHOLD_PERCENT(sbi, sbi->s_mb_c2_blocks));
 349 +}
 350 +
 351 +static ssize_t mb_c3_threshold_store(struct ext4_attr *a,
 352 +                                   struct ext4_sb_info *sbi,
 353 +                                   const char *buf, size_t count)
 354 +{
 355 +       int ret;
 356 +
 357 +       ret = save_threshold_percent(sbi, buf, &sbi->s_mb_c3_blocks);
 358 +
 359 +       return ret ? ret : count;
 360 +}
 361 +
 362 +static ssize_t mb_c3_threshold_show(struct ext4_attr *a,
 363 +                                  struct ext4_sb_info *sbi, char *buf)
 364 +{
 365 +               return snprintf(buf, PAGE_SIZE, "%llu\n",
 366 +                               THRESHOLD_PERCENT(sbi, sbi->s_mb_c3_blocks));
 367 +}
 368 +
 369 +
 370  #define EXT4_ATTR_OFFSET(_name,_mode,_show,_store,_elname) \
 371  static struct ext4_attr ext4_attr_##_name = {                  \
 372         .attr = {.name = __stringify(_name), .mode = _mode },   \
 373 @@ -2790,6 +2857,9 @@ EXT4_RW_ATTR_SBI_UI(mb_stats, s_mb_stats
 374  EXT4_RW_ATTR_SBI_UI(mb_max_to_scan, s_mb_max_to_scan);
 375  EXT4_RW_ATTR_SBI_UI(mb_min_to_scan, s_mb_min_to_scan);
 376  EXT4_RW_ATTR_SBI_UI(mb_order2_req, s_mb_order2_reqs);
 377 +EXT4_RW_ATTR(mb_c1_threshold);
 378 +EXT4_RW_ATTR(mb_c2_threshold);
 379 +EXT4_RW_ATTR(mb_c3_threshold);
 380  EXT4_RW_ATTR_SBI_UI(mb_small_req, s_mb_small_req);
 381  EXT4_RW_ATTR_SBI_UI(mb_large_req, s_mb_large_req);
 382  EXT4_RW_ATTR_SBI_UI(mb_group_prealloc, s_mb_group_prealloc);
 383 @@ -2820,6 +2890,9 @@ static struct attribute *ext4_attrs[] =
 384         ATTR_LIST(mb_max_to_scan),
 385         ATTR_LIST(mb_min_to_scan),
 386         ATTR_LIST(mb_order2_req),
 387 +       ATTR_LIST(mb_c1_threshold),
 388 +       ATTR_LIST(mb_c2_threshold),
 389 +       ATTR_LIST(mb_c3_threshold),
 390         ATTR_LIST(mb_small_req),
 391         ATTR_LIST(mb_large_req),
 392         ATTR_LIST(mb_group_prealloc),
 393 Index: linux-stage/fs/ext4/mballoc.h
 394 ===================================================================
 395 --- linux-stage.orig/fs/ext4/mballoc.h
 396 +++ linux-stage/fs/ext4/mballoc.h
 397 @@ -84,6 +84,9 @@ extern ushort ext4_mballoc_debug;
 398   * for which requests use 2^N search using buddies
 399   */
 400  #define MB_DEFAULT_ORDER2_REQS         8
 401 +#define MB_DEFAULT_C1_THRESHOLD                25
 402 +#define MB_DEFAULT_C2_THRESHOLD                15
 403 +#define MB_DEFAULT_C3_THRESHOLD                5
 404
 405  /*
 406   * default group prealloc size 512 blocks