ldiskfs/kernel_patches/patches/sles12sp2/ext4-large-dir.patch

   1 This INCOMPAT_LARGEDIR feature allows larger directories
   2 to be created in ldiskfs, both with directory sizes over
   3 2GB and and a maximum htree depth of 3 instead of the
   4 current limit of 2. These features are needed in order
   5 to exceed the current limit of approximately 10M entries
   6 in a single directory.
   7
   8 Index: linux-3.10.0-229.1.2.fc21.x86_64/fs/ext4/ext4.h
   9 ===================================================================
  10 --- linux-3.10.0-229.1.2.fc21.x86_64.orig/fs/ext4/ext4.h
  11 +++ linux-3.10.0-229.1.2.fc21.x86_64/fs/ext4/ext4.h
  12 @@ -1585,6 +1585,7 @@ static inline void ext4_clear_state_flag
  13                                          EXT4_FEATURE_INCOMPAT_MMP |    \
  14                                          EXT4_FEATURE_INCOMPAT_DIRDATA| \
  15                                          EXT4_FEATURE_INCOMPAT_INLINE_DATA | \
  16 +                                        EXT4_FEATURE_INCOMPAT_LARGEDIR | \
  17                                          EXT4_FEATURE_INCOMPAT_ENCRYPT | \
  18                                          EXT4_FEATURE_INCOMPAT_CSUM_SEED)
  19  #define EXT4_FEATURE_RO_COMPAT_SUPP    (EXT4_FEATURE_RO_COMPAT_SPARSE_SUPER| \
  20 @@ -1999,6 +2000,9 @@ struct mmpd_data {
  21  # define NORET_TYPE    /**/
  22  # define ATTRIB_NORET  __attribute__((noreturn))
  23  # define NORET_AND     noreturn,
  24 +/* htree levels for ext4 */
  25 +#define EXT4_HTREE_LEVEL_COMPAT 2
  26 +#define EXT4_HTREE_LEVEL       3
  27
  28  struct ext4_xattr_ino_array {
  29         unsigned int xia_count;         /* # of used item in the array */
  30 @@ -2472,13 +2476,16 @@ static inline void ext4_r_blocks_count_s
  31         es->s_r_blocks_count_hi = cpu_to_le32(blk >> 32);
  32  }
  33
  34 -static inline loff_t ext4_isize(struct ext4_inode *raw_inode)
  35 +static inline loff_t ext4_isize(struct super_block *sb,
  36 +                               struct ext4_inode *raw_inode)
  37  {
  38 -       if (S_ISREG(le16_to_cpu(raw_inode->i_mode)))
  39 +       if (S_ISREG(le16_to_cpu(raw_inode->i_mode)) ||
  40 +           (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_LARGEDIR) &&
  41 +           S_ISDIR(le16_to_cpu(raw_inode->i_mode))))
  42                 return ((loff_t)le32_to_cpu(raw_inode->i_size_high) << 32) |
  43                         le32_to_cpu(raw_inode->i_size_lo);
  44 -       else
  45 -               return (loff_t) le32_to_cpu(raw_inode->i_size_lo);
  46 +
  47 +       return (loff_t)le32_to_cpu(raw_inode->i_size_lo);
  48  }
  49
  50  static inline void ext4_isize_set(struct ext4_inode *raw_inode, loff_t i_size)
  51 Index: linux-3.10.0-229.1.2.fc21.x86_64/fs/ext4/namei.c
  52 ===================================================================
  53 --- linux-3.10.0-229.1.2.fc21.x86_64.orig/fs/ext4/namei.c
  54 +++ linux-3.10.0-229.1.2.fc21.x86_64/fs/ext4/namei.c
  55 @@ -513,7 +513,14 @@ struct dx_root_info * dx_get_dx_info(str
  56
  57  static inline ext4_lblk_t dx_get_block(struct dx_entry *entry)
  58  {
  59 -       return le32_to_cpu(entry->block) & 0x00ffffff;
  60 +       return le32_to_cpu(entry->block) & 0x0fffffff;
  61 +}
  62 +
  63 +static inline int
  64 +ext4_dir_htree_level(struct super_block *sb)
  65 +{
  66 +       return EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_LARGEDIR) ?
  67 +               EXT4_HTREE_LEVEL : EXT4_HTREE_LEVEL_COMPAT;
  68  }
  69
  70  static inline void dx_set_block(struct dx_entry *entry, ext4_lblk_t value)
  71 @@ -681,6 +688,7 @@ dx_probe(const struct qstr *d_name, stru
  72         struct dx_frame *frame = frame_in;
  73         u32 hash;
  74
  75 +       memset(frame_in, 0, EXT4_HTREE_LEVEL * sizeof(frame_in[0]));
  76         frame->bh = ext4_read_dirblock(dir, 0, INDEX);
  77         if (IS_ERR(frame->bh))
  78                 return (struct dx_frame *) frame->bh;
  79 @@ -714,9 +721,13 @@ dx_probe(const struct qstr *d_name, stru
  80         }
  81
  82         indirect = info->indirect_levels;
  83 -       if (indirect > 1) {
  84 -               ext4_warning_inode(dir, "Unimplemented hash depth: %#06x",
  85 -                                  info->indirect_levels);
  86 +       if (indirect >= ext4_dir_htree_level(dir->i_sb)) {
  87 +               ext4_warning_inode(dir, "htree depth: %#06x exceed max depth %u",
  88 +                                  indirect, ext4_dir_htree_level(dir->i_sb));
  89 +               if (ext4_dir_htree_level(dir->i_sb) < EXT4_HTREE_LEVEL) {
  90 +                       ext4_warning(dir->i_sb, "Enable large directory "
  91 +                                               "feature to access it");
  92 +               }
  93                 goto fail;
  94         }
  95
  96 @@ -812,12 +826,20 @@ fail:
  97
  98  static void dx_release (struct dx_frame *frames)
  99  {
 100 +       int i;
 101 +       struct dx_root_info *info;
 102 +
 103         if (frames[0].bh == NULL)
 104                 return;
 105
 106 -       if (((struct dx_root *)frames[0].bh->b_data)->info.indirect_levels)
 107 -               brelse(frames[1].bh);
 108 -       brelse(frames[0].bh);
 109 +       for (i = 0, info = dx_get_dx_info((struct ext4_dir_entry_2 *)frames[0].bh->b_data);
 110 +            i <= info->indirect_levels;
 111 +            i++) {
 112 +               if (frames[i].bh == NULL)
 113 +                       break;
 114 +               brelse(frames[i].bh);
 115 +               frames[i].bh = NULL;
 116 +       }
 117  }
 118
 119  /*
 120 @@ -960,7 +979,7 @@ int ext4_htree_fill_tree(struct file *di
 121  {
 122         struct dx_hash_info hinfo;
 123         struct ext4_dir_entry_2 *de;
 124 -       struct dx_frame frames[2], *frame;
 125 +       struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
 126         struct inode *dir;
 127         ext4_lblk_t block;
 128         int count = 0;
 129 @@ -1376,7 +1395,7 @@ static struct buffer_head * ext4_dx_find
 130         struct dx_hash_info     hinfo;
 131  {
 132         struct super_block * sb = dir->i_sb;
 133 -       struct dx_frame frames[2], *frame;
 134 +       struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
 135         const struct qstr *d_name = fname->usr_fname;
 136         struct buffer_head *bh;
 137         ext4_lblk_t block;
 138 @@ -1832,7 +1851,7 @@ static int make_indexed_dir(handle_t *ha
 139         const char      *name = dentry->d_name.name;
 140         int             namelen = dentry->d_name.len;
 141         struct buffer_head *bh2;
 142 -       struct dx_frame frames[2], *frame;
 143 +       struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
 144         struct dx_entry *entries;
 145         struct ext4_dir_entry_2 *de, *de2, *dot_de, *dotdot_de;
 146         struct ext4_dir_entry_tail *t;
 147 @@ -2117,14 +2136,17 @@ static int ext4_add_entry(handle_t *hand
 148  static int ext4_dx_add_entry(handle_t *handle, struct dentry *dentry,
 149                              struct inode *inode)
 150  {
 151 -       struct dx_frame frames[2], *frame;
 152 +       struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
 153         struct dx_entry *entries, *at;
 154         struct buffer_head *bh;
 155         struct inode *dir = d_inode(dentry->d_parent);
 156         struct super_block *sb = dir->i_sb;
 157         struct ext4_dir_entry_2 *de;
 158 +       int restart;
 159         int err;
 160
 161 +again:
 162 +       restart = 0;
 163         frame = dx_probe(fname, dir, NULL, frames);
 164         if (IS_ERR(frame))
 165                 return PTR_ERR(frame);
 166 @@ -2138,33 +2160,48 @@ static int ext4_dx_add_entry(handle_t *h
 167                 goto cleanup;
 168         }
 169
 170 -       BUFFER_TRACE(bh, "get_write_access");
 171 -       err = ext4_journal_get_write_access(handle, bh);
 172 -       if (err)
 173 -               goto journal_error;
 174 -
 175         err = add_dirent_to_buf(handle, fname, dir, inode, NULL, bh, dentry);
 176         if (err != -ENOSPC)
 177                 goto cleanup;
 178
 179 +       err = 0;
 180         /* Block full, should compress but for now just split */
 181         dxtrace(printk(KERN_DEBUG "using %u of %u node entries\n",
 182                        dx_get_count(entries), dx_get_limit(entries)));
 183         /* Need to split index? */
 184         if (dx_get_count(entries) == dx_get_limit(entries)) {
 185                 ext4_lblk_t newblock;
 186 -               unsigned icount = dx_get_count(entries);
 187 -               int levels = frame - frames;
 188 +               int levels = frame - frames + 1;
 189 +               unsigned icount;
 190 +               int add_level = 1;
 191                 struct dx_entry *entries2;
 192                 struct dx_node *node2;
 193                 struct buffer_head *bh2;
 194
 195 -               if (levels && (dx_get_count(frames->entries) ==
 196 -                              dx_get_limit(frames->entries))) {
 197 -                       ext4_warning_inode(dir, "Directory index full!");
 198 +               while (frame > frames) {
 199 +                       if (dx_get_count((frame - 1)->entries) <
 200 +                           dx_get_limit((frame - 1)->entries)) {
 201 +                               add_level = 0;
 202 +                               break;
 203 +                       }
 204 +                       frame--; /* split higher index block */
 205 +                       at = frame->at;
 206 +                       entries = frame->entries;
 207 +                       restart = 1;
 208 +               }
 209 +               if (add_level && levels == ext4_dir_htree_level(sb)) {
 210 +                       ext4_warning(sb, "inode %lu: comm %s: index %u: reach max htree level %u",
 211 +                                        dir->i_ino, current->comm, levels,
 212 +                                        ext4_dir_htree_level(sb));
 213 +                       if (ext4_dir_htree_level(sb) < EXT4_HTREE_LEVEL) {
 214 +                               ext4_warning(sb, "Large directory feature is"
 215 +                                                "not enabled on this "
 216 +                                                "filesystem");
 217 +                       }
 218                         err = -ENOSPC;
 219                         goto cleanup;
 220                 }
 221 +               icount = dx_get_count(entries);
 222                 bh2 = ext4_append(handle, dir, &newblock);
 223                 if (IS_ERR(bh2)) {
 224                         err = PTR_ERR(bh2);
 225 @@ -2179,7 +2216,7 @@ static int ext4_dx_add_entry(handle_t *h
 226                 err = ext4_journal_get_write_access(handle, frame->bh);
 227                 if (err)
 228                         goto journal_error;
 229 -               if (levels) {
 230 +               if (!add_level) {
 231                         unsigned icount1 = icount/2, icount2 = icount - icount1;
 232                         unsigned hash2 = dx_get_hash(entries + icount1);
 233                         dxtrace(printk(KERN_DEBUG "Split index %i/%i\n",
 234 @@ -2187,7 +2224,7 @@ static int ext4_dx_add_entry(handle_t *h
 235
 236                         BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
 237                         err = ext4_journal_get_write_access(handle,
 238 -                                                            frames[0].bh);
 239 +                                                           (frame - 1)->bh);
 240                         if (err)
 241                                 goto journal_error;
 242
 243 @@ -2203,19 +2240,27 @@ static int ext4_dx_add_entry(handle_t *h
 244                                 frame->entries = entries = entries2;
 245                                 swap(frame->bh, bh2);
 246                         }
 247 -                       dx_insert_block(frames + 0, hash2, newblock);
 248 -                       dxtrace(dx_show_index("node", frames[1].entries));
 249 +                       dx_insert_block(frame - 1, hash2, newblock);
 250 +                       dxtrace(dx_show_index("node", frame->entries));
 251                         dxtrace(dx_show_index("node",
 252 -                              ((struct dx_node *) bh2->b_data)->entries));
 253 +                              ((struct dx_node *)bh2->b_data)->entries));
 254                         err = ext4_handle_dirty_dx_node(handle, dir, bh2);
 255                         if (err)
 256                                 goto journal_error;
 257                         brelse (bh2);
 258 +                       err = ext4_handle_dirty_dx_node(handle, dir,
 259 +                                                  (frame - 1)->bh);
 260 +                       if (err)
 261 +                               goto journal_error;
 262 +                       if (restart) {
 263 +                               err = ext4_handle_dirty_dx_node(handle, dir,
 264 +                                                          frame->bh);
 265 +                               goto journal_error;
 266 +                       }
 267                 } else {
 268                         struct dx_root_info *info;
 269 -                       dxtrace(printk(KERN_DEBUG
 270 -                                      "Creating second level index...\n"));
 271 -                       memcpy((char *) entries2, (char *) entries,
 272 +
 273 +                       memcpy((char *)entries2, (char *)entries,
 274                                icount * sizeof(struct dx_entry));
 275                         dx_set_limit(entries2, dx_node_limit(dir));
 276
 277 @@ -2224,22 +2267,17 @@ static int ext4_dx_add_entry(handle_t *h
 278                         dx_set_block(entries + 0, newblock);
 279                         info = dx_get_dx_info((struct ext4_dir_entry_2*)
 280                                               frames[0].bh->b_data);
 281 -                       info->indirect_levels = 1;
 282 -
 283 -                       /* Add new access path frame */
 284 -                       frame = frames + 1;
 285 -                       frame->at = at = at - entries + entries2;
 286 -                       frame->entries = entries = entries2;
 287 -                       frame->bh = bh2;
 288 -                       err = ext4_journal_get_write_access(handle,
 289 -                                                            frame->bh);
 290 +                       info->indirect_levels += 1;
 291 +                       dxtrace(printk(KERN_DEBUG
 292 +                                      "Creating %d level index...\n",
 293 +                                      info->indirect_levels));
 294 +                       err = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
 295                         if (err)
 296                                 goto journal_error;
 297 -               }
 298 -               err = ext4_handle_dirty_dx_node(handle, dir, frames[0].bh);
 299 -               if (err) {
 300 -                       ext4_std_error(inode->i_sb, err);
 301 -                       goto cleanup;
 302 +                       err = ext4_handle_dirty_dx_node(handle, dir, bh2);
 303 +                       brelse(bh2);
 304 +                       restart = 1;
 305 +                       goto journal_error;
 306                 }
 307         }
 308         de = do_split(handle, dir, &bh, frame, &fname->hinfo);
 309 @@ -2249,10 +2285,14 @@ static int ext4_dx_add_entry(handle_t *h
 310         goto cleanup;
 311
 312  journal_error:
 313 -       ext4_std_error(dir->i_sb, err);
 314 +       ext4_std_error(dir->i_sb, err); /* this is a no-op if err == 0 */
 315  cleanup:
 316         brelse(bh);
 317         dx_release(frames);
 318 +       /* @restart is true means htree-path has been changed, we need to
 319 +        * repeat dx_probe() to find out valid htree-path */
 320 +       if (restart && err == 0)
 321 +               goto again;
 322         return err;
 323  }
 324
 325 Index: linux-3.10.0-229.1.2.fc21.x86_64/fs/ext4/inode.c
 326 ===================================================================
 327 --- linux-3.10.0-229.1.2.fc21.x86_64.orig/fs/ext4/inode.c
 328 +++ linux-3.10.0-229.1.2.fc21.x86_64/fs/ext4/inode.c
 329 @@ -4056,12 +4056,12 @@ struct inode *ext4_iget(struct super_blo
 330         if (ext4_has_feature_64bit(sb))
 331                 ei->i_file_acl |=
 332                         ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32;
 333 -       inode->i_size = ext4_isize(raw_inode);
 334 +       inode->i_size = ext4_isize(sb, raw_inode);
 335         if ((size = i_size_read(inode)) < 0) {
 336                 EXT4_ERROR_INODE(inode, "bad i_size value: %lld", size);
 337                 ret = -EFSCORRUPTED;
 338                 goto bad_inode;
 339         }
 340         ei->i_disksize = inode->i_size;
 341  #ifdef CONFIG_QUOTA
 342         ei->i_reserved_quota = 0;
 343 @@ -4306,7 +4306,7 @@ static int ext4_do_update_inode(handle_t
 344                 raw_inode->i_file_acl_high =
 345                         cpu_to_le16(ei->i_file_acl >> 32);
 346         raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl);
 347 -       if (ei->i_disksize != ext4_isize(raw_inode)) {
 348 +       if (ei->i_disksize != ext4_isize(inode->i_sb, raw_inode)) {
 349                 ext4_isize_set(raw_inode, ei->i_disksize);
 350                 need_datasync = 1;
 351         }