ldiskfs/kernel_patches/patches/rhel8/ext4-simple-blockalloc.patch

   1 diff --git a/fs/ext4/ext4.h b/fs/ext4/ext4.h
   2 index 3b9ec24..64dc5fd 100644
   3 --- a/fs/ext4/ext4.h
   4 +++ b/fs/ext4/ext4.h
   5 @@ -1450,6 +1450,9 @@ struct ext4_sb_info {
   6         unsigned int s_mb_min_to_scan;
   7         unsigned int s_mb_stats;
   8         unsigned int s_mb_order2_reqs;
   9 +       ext4_fsblk_t s_mb_c1_blocks;
  10 +       ext4_fsblk_t s_mb_c2_blocks;
  11 +       ext4_fsblk_t s_mb_c3_blocks;
  12         unsigned long *s_mb_prealloc_table;
  13         unsigned int s_mb_group_prealloc;
  14         unsigned int s_max_dir_size_kb;
  15 @@ -1466,6 +1469,9 @@ struct ext4_sb_info {
  16         atomic_t s_bal_goals;   /* goal hits */
  17         atomic_t s_bal_breaks;  /* too long searches */
  18         atomic_t s_bal_2orders; /* 2^order hits */
  19 +       /* cX loop didn't find blocks */
  20 +       atomic64_t s_bal_cX_failed[3];
  21 +       atomic64_t s_bal_cX_skipped[3];
  22         spinlock_t s_bal_lock;
  23         unsigned long s_mb_buddies_generated;
  24         unsigned long long s_mb_generation_time;
  25 @@ -2563,6 +2569,9 @@ extern void ext4_end_bitmap_read(struct buffer_head *bh, int uptodate);
  26  /* mballoc.c */
  27  extern const struct file_operations ext4_seq_prealloc_table_fops;
  28  extern const struct seq_operations ext4_mb_seq_groups_ops;
  29 +extern const struct file_operations ext4_mb_seq_alloc_fops;
  30 +extern int save_threshold_percent(struct ext4_sb_info *sbi, const char *buf,
  31 +                                 ext4_fsblk_t *blocks);
  32  extern const struct file_operations ext4_seq_mb_last_group_fops;
  33  extern int ext4_mb_seq_last_start_seq_show(struct seq_file *m, void *v);
  34  extern long ext4_mb_stats;
  35 diff --git a/fs/ext4/mballoc.c b/fs/ext4/mballoc.c
  36 index 15c962f..7870406 100644
  37 --- a/fs/ext4/mballoc.c
  38 +++ b/fs/ext4/mballoc.c
  39 @@ -2104,6 +2104,20 @@ static int ext4_mb_good_group(struct ext4_allocation_context *ac,
  40         return 0;
  41  }
  42
  43 +static u64 available_blocks_count(struct ext4_sb_info *sbi)
  44 +{
  45 +       ext4_fsblk_t resv_blocks;
  46 +       u64 bfree;
  47 +       struct ext4_super_block *es = sbi->s_es;
  48 +
  49 +       resv_blocks = EXT4_C2B(sbi, atomic64_read(&sbi->s_resv_clusters));
  50 +       bfree = percpu_counter_sum_positive(&sbi->s_freeclusters_counter) -
  51 +                percpu_counter_sum_positive(&sbi->s_dirtyclusters_counter);
  52 +
  53 +       bfree = EXT4_C2B(sbi, max_t(s64, bfree, 0));
  54 +       return bfree - (ext4_r_blocks_count(es) + resv_blocks);
  55 +}
  56 +
  57  static noinline_for_stack int
  58  ext4_mb_regular_allocator(struct ext4_allocation_context *ac)
  59  {
  60 @@ -2113,6 +2127,7 @@ ext4_mb_regular_allocator(struct ext4_allocation_context *ac)
  61         struct ext4_sb_info *sbi;
  62         struct super_block *sb;
  63         struct ext4_buddy e4b;
  64 +       ext4_fsblk_t avail_blocks;
  65
  66         sb = ac->ac_sb;
  67         sbi = EXT4_SB(sb);
  68 @@ -2165,6 +2180,21 @@ ext4_mb_regular_allocator(struct ext4_allocation_context *ac)
  69
  70         /* Let's just scan groups to find more-less suitable blocks */
  71         cr = ac->ac_2order ? 0 : 1;
  72 +
  73 +       /* Choose what loop to pass based on disk fullness */
  74 +       avail_blocks = available_blocks_count(sbi) ;
  75 +
  76 +       if (avail_blocks < sbi->s_mb_c3_blocks) {
  77 +               cr = 3;
  78 +               atomic64_inc(&sbi->s_bal_cX_skipped[2]);
  79 +       } else if(avail_blocks < sbi->s_mb_c2_blocks) {
  80 +               cr = 2;
  81 +               atomic64_inc(&sbi->s_bal_cX_skipped[1]);
  82 +       } else if(avail_blocks < sbi->s_mb_c1_blocks) {
  83 +               cr = 1;
  84 +               atomic64_inc(&sbi->s_bal_cX_skipped[0]);
  85 +       }
  86 +
  87         /*
  88          * cr == 0 try to get exact allocation,
  89          * cr == 3  try to get anything
  90 @@ -2230,6 +2260,9 @@ repeat:
  91                         if (ac->ac_status != AC_STATUS_CONTINUE)
  92                                 break;
  93                 }
  94 +               /* Processed all groups and haven't found blocks */
  95 +               if (i == ngroups)
  96 +                       atomic64_inc(&sbi->s_bal_cX_failed[cr]);
  97         }
  98
  99         if (ac->ac_b_ex.fe_len > 0 && ac->ac_status != AC_STATUS_FOUND &&
 100 @@ -2510,6 +2543,93 @@ const struct file_operations ext4_seq_mb_last_group_fops = {
 101         .write         = ext4_mb_last_group_write,
 102  };
 103
 104 +static int mb_seq_alloc_show(struct seq_file *seq, void *v)
 105 +{
 106 +       struct super_block *sb = seq->private;
 107 +       struct ext4_sb_info *sbi = EXT4_SB(sb);
 108 +
 109 +       seq_printf(seq, "mballoc:\n");
 110 +       seq_printf(seq, "\tblocks: %u\n", atomic_read(&sbi->s_bal_allocated));
 111 +       seq_printf(seq, "\treqs: %u\n", atomic_read(&sbi->s_bal_reqs));
 112 +       seq_printf(seq, "\tsuccess: %u\n", atomic_read(&sbi->s_bal_success));
 113 +
 114 +       seq_printf(seq, "\textents_scanned: %u\n",
 115 +                  atomic_read(&sbi->s_bal_ex_scanned));
 116 +       seq_printf(seq, "\t\tgoal_hits: %u\n", atomic_read(&sbi->s_bal_goals));
 117 +       seq_printf(seq, "\t\t2^n_hits: %u\n", atomic_read(&sbi->s_bal_2orders));
 118 +       seq_printf(seq, "\t\tbreaks: %u\n", atomic_read(&sbi->s_bal_breaks));
 119 +       seq_printf(seq, "\t\tlost: %u\n", atomic_read(&sbi->s_mb_lost_chunks));
 120 +
 121 +       seq_printf(seq, "\tuseless_c1_loops: %llu\n",
 122 +                  (unsigned long long)atomic64_read(&sbi->s_bal_cX_failed[0]));
 123 +       seq_printf(seq, "\tuseless_c2_loops: %llu\n",
 124 +                  (unsigned long long)atomic64_read(&sbi->s_bal_cX_failed[1]));
 125 +       seq_printf(seq, "\tuseless_c3_loops: %llu\n",
 126 +                  (unsigned long long)atomic64_read(&sbi->s_bal_cX_failed[2]));
 127 +       seq_printf(seq, "\tskipped_c1_loops: %llu\n",
 128 +                  (unsigned long long)atomic64_read(&sbi->s_bal_cX_skipped[0]));
 129 +       seq_printf(seq, "\tskipped_c2_loops: %llu\n",
 130 +                  (unsigned long long)atomic64_read(&sbi->s_bal_cX_skipped[1]));
 131 +       seq_printf(seq, "\tskipped_c3_loops: %llu\n",
 132 +                  (unsigned long long)atomic64_read(&sbi->s_bal_cX_skipped[2]));
 133 +       seq_printf(seq, "\tbuddies_generated: %lu\n",
 134 +                  sbi->s_mb_buddies_generated);
 135 +       seq_printf(seq, "\tbuddies_time_used: %llu\n", sbi->s_mb_generation_time);
 136 +       seq_printf(seq, "\tpreallocated: %u\n",
 137 +                  atomic_read(&sbi->s_mb_preallocated));
 138 +       seq_printf(seq, "\tdiscarded: %u\n",
 139 +                  atomic_read(&sbi->s_mb_discarded));
 140 +       return 0;
 141 +}
 142 +
 143 +static ssize_t mb_seq_alloc_write(struct file *file,
 144 +                             const char __user *buf,
 145 +                             size_t cnt, loff_t *pos)
 146 +{
 147 +       struct ext4_sb_info *sbi = EXT4_SB(PDE_DATA(file_inode(file)));
 148 +
 149 +       atomic_set(&sbi->s_bal_allocated, 0),
 150 +       atomic_set(&sbi->s_bal_reqs, 0),
 151 +       atomic_set(&sbi->s_bal_success, 0);
 152 +
 153 +       atomic_set(&sbi->s_bal_ex_scanned, 0),
 154 +       atomic_set(&sbi->s_bal_goals, 0),
 155 +       atomic_set(&sbi->s_bal_2orders, 0),
 156 +       atomic_set(&sbi->s_bal_breaks, 0),
 157 +       atomic_set(&sbi->s_mb_lost_chunks, 0);
 158 +
 159 +       atomic64_set(&sbi->s_bal_cX_failed[0], 0),
 160 +       atomic64_set(&sbi->s_bal_cX_failed[1], 0),
 161 +       atomic64_set(&sbi->s_bal_cX_failed[2], 0);
 162 +
 163 +       atomic64_set(&sbi->s_bal_cX_skipped[0], 0),
 164 +       atomic64_set(&sbi->s_bal_cX_skipped[1], 0),
 165 +       atomic64_set(&sbi->s_bal_cX_skipped[2], 0);
 166 +
 167 +
 168 +       sbi->s_mb_buddies_generated = 0;
 169 +       sbi->s_mb_generation_time = 0;
 170 +
 171 +       atomic_set(&sbi->s_mb_preallocated, 0),
 172 +       atomic_set(&sbi->s_mb_discarded, 0);
 173 +
 174 +       return cnt;
 175 +}
 176 +
 177 +static int mb_seq_alloc_open(struct inode *inode, struct file *file)
 178 +{
 179 +       return single_open(file, mb_seq_alloc_show, PDE_DATA(inode));
 180 +}
 181 +
 182 +const struct file_operations ext4_mb_seq_alloc_fops = {
 183 +       .owner          = THIS_MODULE,
 184 +       .open           = mb_seq_alloc_open,
 185 +       .read           = seq_read,
 186 +       .llseek         = seq_lseek,
 187 +       .release        = single_release,
 188 +       .write          = mb_seq_alloc_write,
 189 +};
 190 +
 191  int ext4_mb_seq_last_start_seq_show(struct seq_file *m, void *v)
 192  {
 193         struct ext4_sb_info *sbi = EXT4_SB(m->private);
 194 @@ -2734,6 +2854,8 @@ static int ext4_groupinfo_create_slab(size_t size)
 195         return 0;
 196  }
 197
 198 +#define THRESHOLD_BLOCKS(sbi, percent)                                 \
 199 +       (ext4_blocks_count((sbi)->s_es) / 100 * (percent))
 200  int ext4_mb_init(struct super_block *sb)
 201  {
 202         struct ext4_sb_info *sbi = EXT4_SB(sb);
 203 @@ -2787,6 +2908,15 @@ int ext4_mb_init(struct super_block *sb)
 204         sbi->s_mb_min_to_scan = MB_DEFAULT_MIN_TO_SCAN;
 205         sbi->s_mb_stats = MB_DEFAULT_STATS;
 206         sbi->s_mb_order2_reqs = MB_DEFAULT_ORDER2_REQS;
 207 +       if (!sbi->s_mb_c1_blocks)
 208 +               sbi->s_mb_c1_blocks =
 209 +                       THRESHOLD_BLOCKS(sbi, MB_DEFAULT_C1_THRESHOLD);
 210 +       if (!sbi->s_mb_c2_blocks)
 211 +               sbi->s_mb_c2_blocks =
 212 +                       THRESHOLD_BLOCKS(sbi, MB_DEFAULT_C2_THRESHOLD);
 213 +       if (!sbi->s_mb_c3_blocks)
 214 +               sbi->s_mb_c3_blocks =
 215 +                       THRESHOLD_BLOCKS(sbi, MB_DEFAULT_C3_THRESHOLD);
 216         /*
 217          * The default group preallocation is 512, which for 4k block
 218          * sizes translates to 2 megabytes.  However for bigalloc file
 219 @@ -2922,6 +3046,16 @@ int ext4_mb_release(struct super_block *sb)
 220                                 atomic_read(&sbi->s_bal_allocated),
 221                                 atomic_read(&sbi->s_bal_reqs),
 222                                 atomic_read(&sbi->s_bal_success));
 223 +               ext4_msg(sb, KERN_INFO,
 224 +                       "mballoc: (%llu, %llu, %llu) useless c(0,1,2) loops",
 225 +                               (unsigned long long)atomic64_read(&sbi->s_bal_cX_failed[0]),
 226 +                               (unsigned long long)atomic64_read(&sbi->s_bal_cX_failed[1]),
 227 +                               (unsigned long long)atomic64_read(&sbi->s_bal_cX_failed[2]));
 228 +               ext4_msg(sb, KERN_INFO,
 229 +                       "mballoc: (%llu, %llu, %llu) skipped c(0,1,2) loops",
 230 +                               (unsigned long long)atomic64_read(&sbi->s_bal_cX_skipped[0]),
 231 +                               (unsigned long long)atomic64_read(&sbi->s_bal_cX_skipped[1]),
 232 +                               (unsigned long long)atomic64_read(&sbi->s_bal_cX_skipped[2]));
 233                 ext4_msg(sb, KERN_INFO,
 234                       "mballoc: %u extents scanned, %u goal hits, "
 235                                 "%u 2^N hits, %u breaks, %u lost",
 236 diff --git a/fs/ext4/mballoc.h b/fs/ext4/mballoc.h
 237 index e00c3b7..d02daaf 100644
 238 --- a/fs/ext4/mballoc.h
 239 +++ b/fs/ext4/mballoc.h
 240 @@ -72,6 +72,9 @@ do {                                                                  \
 241   * for which requests use 2^N search using buddies
 242   */
 243  #define MB_DEFAULT_ORDER2_REQS         8
 244 +#define MB_DEFAULT_C1_THRESHOLD                25
 245 +#define MB_DEFAULT_C2_THRESHOLD                15
 246 +#define MB_DEFAULT_C3_THRESHOLD                5
 247
 248  /*
 249   * default group prealloc size 512 blocks
 250 diff --git a/fs/ext4/super.c b/fs/ext4/super.c
 251 ===================================================================
 252 --- linux-stage.orig/fs/ext4/super.c
 253 +++ linux-stage/fs/ext4/super.c
 254 @@ -1450,6 +1450,7 @@ enum {
 255         Opt_nomblk_io_submit, Opt_block_validity, Opt_noblock_validity,
 256         Opt_inode_readahead_blks, Opt_journal_ioprio,
 257         Opt_dioread_nolock, Opt_dioread_lock,
 258 +       Opt_mb_c1_threshold, Opt_mb_c2_threshold, Opt_mb_c3_threshold,
 259         Opt_discard, Opt_nodiscard, Opt_init_itable, Opt_noinit_itable,
 260         Opt_max_dir_size_kb, Opt_nojournal_checksum, Opt_nombcache,
 261  };
 262 @@ -1604,6 +1605,9 @@ static const match_table_t tokens = {
 263         {Opt_init_itable, "init_itable"},
 264         {Opt_noinit_itable, "noinit_itable"},
 265         {Opt_max_dir_size_kb, "max_dir_size_kb=%u"},
 266 +       {Opt_mb_c1_threshold, "mb_c1_threshold=%s"},
 267 +       {Opt_mb_c2_threshold, "mb_c2_threshold=%s"},
 268 +       {Opt_mb_c3_threshold, "mb_c3_threshold=%s"},
 269         {Opt_test_dummy_encryption, "test_dummy_encryption"},
 270         {Opt_nombcache, "nombcache"},
 271         {Opt_nombcache, "no_mbcache"},  /* for backward compatibility */
 272 @@ -1748,6 +1752,9 @@ static const struct mount_opts {
 273         {Opt_jqfmt_vfsv0, QFMT_VFS_V0, MOPT_QFMT},
 274         {Opt_jqfmt_vfsv1, QFMT_VFS_V1, MOPT_QFMT},
 275         {Opt_max_dir_size_kb, 0, MOPT_GTE0},
 276 +       {Opt_mb_c1_threshold, 0, MOPT_STRING},
 277 +       {Opt_mb_c2_threshold, 0, MOPT_STRING},
 278 +       {Opt_mb_c3_threshold, 0, MOPT_STRING},
 279         {Opt_test_dummy_encryption, 0, MOPT_GTE0},
 280         {Opt_nombcache, EXT4_MOUNT_NO_MBCACHE, MOPT_SET},
 281         {Opt_err, 0, 0}
 282 @@ -1874,6 +1881,12 @@ static const struct mount_opts {
 283                 sbi->s_max_dir_size_kb = arg;
 284                 /* reset s_warning_dir_size and make it re-calculated */
 285                 sbi->s_warning_dir_size = 0;
 286 +       } else if (token == Opt_mb_c1_threshold) {
 287 +               save_threshold_percent(sbi, args[0].from, &sbi->s_mb_c1_blocks);
 288 +       } else if (token == Opt_mb_c2_threshold) {
 289 +               save_threshold_percent(sbi, args[0].from, &sbi->s_mb_c2_blocks);
 290 +       } else if (token == Opt_mb_c3_threshold) {
 291 +               save_threshold_percent(sbi, args[0].from, &sbi->s_mb_c3_blocks);
 292         } else if (token == Opt_stripe) {
 293                 sbi->s_stripe = arg;
 294         } else if (token == Opt_resuid) {
 295 diff --git a/fs/ext4/sysfs.c b/fs/ext4/sysfs.c
 296 index 417b33a..f49821e 100644
 297 --- a/fs/ext4/sysfs.c
 298 +++ b/fs/ext4/sysfs.c
 299 @@ -20,6 +20,9 @@
 300  typedef enum {
 301         attr_noop,
 302         attr_delayed_allocation_blocks,
 303 +       attr_mb_c1_threshold,
 304 +       attr_mb_c2_threshold,
 305 +       attr_mb_c3_threshold,
 306         attr_session_write_kbytes,
 307         attr_lifetime_write_kbytes,
 308         attr_reserved_clusters,
 309 @@ -134,6 +137,32 @@ static ssize_t journal_task_show(struct ext4_sb_info *sbi, char *buf)
 310                         task_pid_vnr(sbi->s_journal->j_task));
 311  }
 312
 313 +int save_threshold_percent(struct ext4_sb_info *sbi, const char *buf,
 314 +                          ext4_fsblk_t *blocks)
 315 +{
 316 +       unsigned long long val;
 317 +
 318 +       int ret;
 319 +
 320 +       ret = kstrtoull(skip_spaces(buf), 0, &val);
 321 +       if (ret || val > 100)
 322 +               return -EINVAL;
 323 +
 324 +       *blocks = val * ext4_blocks_count(sbi->s_es) / 100;
 325 +       return 0;
 326 +}
 327 +
 328 +#define THRESHOLD_PERCENT(sbi, blocks)                                 \
 329 +       (((blocks) - 1) * 100 / ext4_blocks_count((sbi)->s_es) + 1)
 330 +static ssize_t mb_threshold_store(struct ext4_sb_info *sbi,
 331 +                                 const char *buf, size_t count,
 332 +                                 ext4_fsblk_t *blocks)
 333 +{
 334 +       int ret = save_threshold_percent(sbi, buf, blocks);
 335 +
 336 +       return ret ?: count;
 337 +}
 338 +
 339  #define EXT4_ATTR(_name,_mode,_id)                                     \
 340  static struct ext4_attr ext4_attr_##_name = {                          \
 341         .attr = {.name = __stringify(_name), .mode = _mode },           \
 342 @@ -176,6 +205,9 @@ EXT4_ATTR_FUNC(delayed_allocation_blocks, 0444);
 343  EXT4_ATTR_FUNC(session_write_kbytes, 0444);
 344  EXT4_ATTR_FUNC(lifetime_write_kbytes, 0444);
 345  EXT4_ATTR_FUNC(reserved_clusters, 0644);
 346 +EXT4_ATTR_FUNC(mb_c1_threshold, 0644);
 347 +EXT4_ATTR_FUNC(mb_c2_threshold, 0644);
 348 +EXT4_ATTR_FUNC(mb_c3_threshold, 0644);
 349
 350  EXT4_ATTR_OFFSET(inode_readahead_blks, 0644, inode_readahead,
 351                  ext4_sb_info, s_inode_readahead_blks);
 352 @@ -211,6 +243,9 @@ static struct attribute *ext4_attrs[] = {
 353         ATTR_LIST(session_write_kbytes),
 354         ATTR_LIST(lifetime_write_kbytes),
 355         ATTR_LIST(reserved_clusters),
 356 +       ATTR_LIST(mb_c1_threshold),
 357 +       ATTR_LIST(mb_c2_threshold),
 358 +       ATTR_LIST(mb_c3_threshold),
 359         ATTR_LIST(inode_readahead_blks),
 360         ATTR_LIST(inode_goal),
 361         ATTR_LIST(max_dir_size),
 362 @@ -294,6 +329,15 @@ static ssize_t ext4_attr_show(struct kobject *kobj,
 363                 return snprintf(buf, PAGE_SIZE, "%llu\n",
 364                                 (s64) EXT4_C2B(sbi,
 365                        percpu_counter_sum(&sbi->s_dirtyclusters_counter)));
 366 +       case attr_mb_c1_threshold:
 367 +               return scnprintf(buf, PAGE_SIZE, "%llu\n",
 368 +                                THRESHOLD_PERCENT(sbi, sbi->s_mb_c1_blocks));
 369 +       case attr_mb_c2_threshold:
 370 +               return scnprintf(buf, PAGE_SIZE, "%llu\n",
 371 +                                THRESHOLD_PERCENT(sbi, sbi->s_mb_c2_blocks));
 372 +       case attr_mb_c3_threshold:
 373 +               return scnprintf(buf, PAGE_SIZE, "%llu\n",
 374 +                                THRESHOLD_PERCENT(sbi, sbi->s_mb_c3_blocks));
 375         case attr_session_write_kbytes:
 376                 return session_write_kbytes_show(sbi, buf);
 377         case attr_lifetime_write_kbytes:
 378 @@ -363,6 +407,12 @@ static ssize_t ext4_attr_store(struct kobject *kobj,
 379                 return inode_readahead_blks_store(sbi, buf, len);
 380         case attr_trigger_test_error:
 381                 return trigger_test_error(sbi, buf, len);
 382 +       case attr_mb_c1_threshold:
 383 +               return mb_threshold_store(sbi, buf, len, &sbi->s_mb_c1_blocks);
 384 +       case attr_mb_c2_threshold:
 385 +               return mb_threshold_store(sbi, buf, len, &sbi->s_mb_c2_blocks);
 386 +       case attr_mb_c3_threshold:
 387 +               return mb_threshold_store(sbi, buf, len, &sbi->s_mb_c3_blocks);
 388         }
 389         return 0;
 390  }
 391 @@ -425,6 +475,8 @@ int ext4_register_sysfs(struct super_block *sb)
 392                                 &ext4_seq_mb_last_group_fops, sb);
 393                 proc_create_single_data("mb_last_start", S_IRUGO, sbi->s_proc,
 394                                 ext4_mb_seq_last_start_seq_show, sb);
 395 +               proc_create_data("mb_alloc_stats", S_IFREG | S_IRUGO | S_IWUSR,
 396 +                                sbi->s_proc, &ext4_mb_seq_alloc_fops, sb);
 397         }
 398         return 0;
 399  }