ldiskfs/kernel_patches/patches/rhel7.6/ext4-data-in-dirent.patch

   1 this patch implements feature which allows ext4 fs users (e.g. Lustre)
   2 to store data in ext4 dirent.
   3 data is stored in ext4 dirent after file-name, this space is accounted
   4 in de->rec_len. flag EXT4_DIRENT_LUFID added to d_type if extra data
   5 is present.
   6
   7 make use of dentry->d_fsdata to pass fid to ext4. so no
   8 changes in ext4_add_entry() interface required.
   9
  10 Index: linux-stage/fs/ext4/dir.c
  11 ===================================================================
  12 --- linux-stage.orig/fs/ext4/dir.c
  13 +++ linux-stage/fs/ext4/dir.c
  14 @@ -71,11 +71,11 @@ int __ext4_check_dir_entry(const char *f
  15         const int rlen = ext4_rec_len_from_disk(de->rec_len,
  16                                                 dir->i_sb->s_blocksize);
  17
  18 -       if (unlikely(rlen < EXT4_DIR_REC_LEN(1)))
  19 +       if (unlikely(rlen < __EXT4_DIR_REC_LEN(1)))
  20                 error_msg = "rec_len is smaller than minimal";
  21         else if (unlikely(rlen % 4 != 0))
  22                 error_msg = "rec_len % 4 != 0";
  23 -       else if (unlikely(rlen < EXT4_DIR_REC_LEN(de->name_len)))
  24 +       else if (unlikely(rlen < EXT4_DIR_REC_LEN(de)))
  25                 error_msg = "rec_len is too small for name_len";
  26         else if (unlikely(((char *) de - buf) + rlen > size))
  27                 error_msg = "directory entry across range";
  28 @@ -208,7 +208,7 @@ revalidate:
  29                                  * failure will be detected in the
  30                                  * dirent test below. */
  31                                 if (ext4_rec_len_from_disk(de->rec_len,
  32 -                                       sb->s_blocksize) < EXT4_DIR_REC_LEN(1))
  33 +                                   sb->s_blocksize) < __EXT4_DIR_REC_LEN(1))
  34                                         break;
  35                                 i += ext4_rec_len_from_disk(de->rec_len,
  36                                                             sb->s_blocksize);
  37 @@ -438,12 +438,17 @@ int ext4_htree_store_dirent(struct file
  38         struct fname *fname, *new_fn;
  39         struct dir_private_info *info;
  40         int len;
  41 +       int extra_data = 0;
  42
  43         info = dir_file->private_data;
  44         p = &info->root.rb_node;
  45
  46         /* Create and allocate the fname structure */
  47 -       len = sizeof(struct fname) + dirent->name_len + 1;
  48 +       if (dirent->file_type & EXT4_DIRENT_LUFID)
  49 +               extra_data = ext4_get_dirent_data_len(dirent);
  50 +
  51 +       len = sizeof(struct fname) + dirent->name_len + extra_data + 1;
  52 +
  53         new_fn = kzalloc(len, GFP_KERNEL);
  54         if (!new_fn)
  55                 return -ENOMEM;
  56 @@ -452,7 +457,7 @@ int ext4_htree_store_dirent(struct file
  57         new_fn->inode = le32_to_cpu(dirent->inode);
  58         new_fn->name_len = dirent->name_len;
  59         new_fn->file_type = dirent->file_type;
  60 -       memcpy(new_fn->name, dirent->name, dirent->name_len);
  61 +       memcpy(new_fn->name, dirent->name, dirent->name_len + extra_data);
  62         new_fn->name[dirent->name_len] = 0;
  63
  64         while (*p) {
  65 @@ -635,7 +640,7 @@ int ext4_check_all_de(struct inode *dir,
  66                 if (ext4_check_dir_entry(dir, NULL, de, bh,
  67                                          buf, buf_size, offset))
  68                         return -EIO;
  69 -               nlen = EXT4_DIR_REC_LEN(de->name_len);
  70 +               nlen = EXT4_DIR_REC_LEN(de);
  71                 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
  72                 de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
  73                 offset += rlen;
  74 Index: linux-stage/fs/ext4/ext4.h
  75 ===================================================================
  76 --- linux-stage.orig/fs/ext4/ext4.h
  77 +++ linux-stage/fs/ext4/ext4.h
  78 @@ -963,6 +963,7 @@ struct ext4_inode_info {
  79  /*
  80   * Mount flags set via mount options or defaults
  81   */
  82 +#define EXT4_MOUNT_DIRDATA             0x00002 /* Data in directory entries*/
  83  #define EXT4_MOUNT_GRPID               0x00004 /* Create files with directory's group */
  84  #define EXT4_MOUNT_DEBUG               0x00008 /* Some debugging messages */
  85  #define EXT4_MOUNT_ERRORS_CONT         0x00010 /* Continue on errors */
  86 @@ -1574,6 +1575,7 @@ static inline void ext4_clear_state_flag
  87                                          EXT4_FEATURE_INCOMPAT_64BIT| \
  88                                          EXT4_FEATURE_INCOMPAT_FLEX_BG| \
  89                                          EXT4_FEATURE_INCOMPAT_MMP |    \
  90 +                                        EXT4_FEATURE_INCOMPAT_DIRDATA| \
  91                                          EXT4_FEATURE_INCOMPAT_INLINE_DATA)
  92  #define EXT4_FEATURE_RO_COMPAT_SUPP    (EXT4_FEATURE_RO_COMPAT_SPARSE_SUPER| \
  93                                          EXT4_FEATURE_RO_COMPAT_LARGE_FILE| \
  94 @@ -1680,6 +1682,43 @@ struct ext4_dir_entry_tail {
  95  #define EXT4_FT_SYMLINK                7
  96
  97  #define EXT4_FT_MAX            8
  98 +#define EXT4_FT_MASK           0xf
  99 +
 100 +#if EXT4_FT_MAX > EXT4_FT_MASK
 101 +#error "conflicting EXT4_FT_MAX and EXT4_FT_MASK"
 102 +#endif
 103 +
 104 +/*
 105 + * d_type has 4 unused bits, so it can hold four types data. these different
 106 + * type of data (e.g. lustre data, high 32 bits of 64-bit inode number) can be
 107 + * stored, in flag order, after file-name in ext4 dirent.
 108 +*/
 109 +/*
 110 + * this flag is added to d_type if ext4 dirent has extra data after
 111 + * filename. this data length is variable and length is stored in first byte
 112 + * of data. data start after filename NUL byte.
 113 + * This is used by Lustre FS.
 114 +  */
 115 +#define EXT4_DIRENT_LUFID              0x10
 116 +
 117 +#define EXT4_LUFID_MAGIC    0xAD200907UL
 118 +struct ext4_dentry_param {
 119 +       __u32  edp_magic;       /* EXT4_LUFID_MAGIC */
 120 +       char   edp_len;         /* size of edp_data in bytes */
 121 +       char   edp_data[0];     /* packed array of data */
 122 +} __packed;
 123 +
 124 +static inline unsigned char *ext4_dentry_get_data(struct super_block *sb,
 125 +                                                 struct ext4_dentry_param *p)
 126 +
 127 +{
 128 +       if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_DIRDATA))
 129 +               return NULL;
 130 +       if (p && p->edp_magic == EXT4_LUFID_MAGIC)
 131 +               return &p->edp_len;
 132 +       else
 133 +               return NULL;
 134 +}
 135
 136  #define EXT4_FT_DIR_CSUM       0xDE
 137
 138 @@ -1690,8 +1729,11 @@ struct ext4_dir_entry_tail {
 139   */
 140  #define EXT4_DIR_PAD                   4
 141  #define EXT4_DIR_ROUND                 (EXT4_DIR_PAD - 1)
 142 -#define EXT4_DIR_REC_LEN(name_len)     (((name_len) + 8 + EXT4_DIR_ROUND) & \
 143 +#define __EXT4_DIR_REC_LEN(name_len)   (((name_len) + 8 + EXT4_DIR_ROUND) & \
 144                                          ~EXT4_DIR_ROUND)
 145 +#define EXT4_DIR_REC_LEN(de)           (__EXT4_DIR_REC_LEN((de)->name_len +\
 146 +                                       ext4_get_dirent_data_len(de)))
 147 +
 148  #define EXT4_MAX_REC_LEN               ((1<<16)-1)
 149
 150  /*
 151 @@ -2016,11 +2058,11 @@ extern int ext4_find_dest_de(struct inod
 152                              struct buffer_head *bh,
 153                              void *buf, int buf_size,
 154                              const char *name, int namelen,
 155 -                            struct ext4_dir_entry_2 **dest_de);
 156 +                            struct ext4_dir_entry_2 **dest_de, int *dlen);
 157  void ext4_insert_dentry(struct inode *inode,
 158                         struct ext4_dir_entry_2 *de,
 159                         int buf_size,
 160 -                       const char *name, int namelen);
 161 +                       const char *name, int namelen, void *data);
 162  static inline void ext4_update_dx_flag(struct inode *inode)
 163  {
 164         if (!EXT4_HAS_COMPAT_FEATURE(inode->i_sb,
 165 @@ -2033,11 +2075,18 @@ static unsigned char ext4_filetype_table
 166
 167  static inline  unsigned char get_dtype(struct super_block *sb, int filetype)
 168  {
 169 +       int fl_index = filetype & EXT4_FT_MASK;
 170 +
 171         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FILETYPE) ||
 172 -           (filetype >= EXT4_FT_MAX))
 173 +           (fl_index >= EXT4_FT_MAX))
 174                 return DT_UNKNOWN;
 175
 176 -       return ext4_filetype_table[filetype];
 177 +       if (!test_opt(sb, DIRDATA))
 178 +               return ext4_filetype_table[fl_index];
 179 +
 180 +       return (ext4_filetype_table[fl_index]) |
 181 +               (filetype & EXT4_DIRENT_LUFID);
 182 +
 183  }
 184  extern int ext4_check_all_de(struct inode *dir, struct buffer_head *bh,
 185                              void *buf, int buf_size);
 186 @@ -2186,6 +2235,8 @@ extern struct inode *ext4_create_inode(h
 187  extern int ext4_delete_entry(handle_t *handle, struct inode * dir,
 188                              struct ext4_dir_entry_2 *de_del,
 189                              struct buffer_head *bh);
 190 +extern int ext4_add_dot_dotdot(handle_t *handle, struct inode *dir,
 191 +                              struct inode *inode, const void *, const void *);
 192  extern int search_dir(struct buffer_head *bh,
 193                       char *search_buf,
 194                       int buf_size,
 195 @@ -2865,6 +2916,36 @@ extern struct mutex ext4__aio_mutex[EXT4
 196  extern int ext4_resize_begin(struct super_block *sb);
 197  extern void ext4_resize_end(struct super_block *sb);
 198
 199 +/*
 200 + * Compute the total directory entry data length.
 201 + * This includes the filename and an implicit NUL terminator (always present),
 202 + * and optional extensions.  Each extension has a bit set in the high 4 bits of
 203 + * de->file_type, and the extension length is the first byte in each entry.
 204 + */
 205 +static inline int ext4_get_dirent_data_len(struct ext4_dir_entry_2 *de)
 206 +{
 207 +       char *len = de->name + de->name_len + 1 /* NUL terminator */;
 208 +       int dlen = 0;
 209 +       __u8 extra_data_flags = (de->file_type & ~EXT4_FT_MASK) >> 4;
 210 +       struct ext4_dir_entry_tail *t = (struct ext4_dir_entry_tail *)de;
 211 +
 212 +       if (!t->det_reserved_zero1 &&
 213 +           le16_to_cpu(t->det_rec_len) ==
 214 +               sizeof(struct ext4_dir_entry_tail) &&
 215 +           !t->det_reserved_zero2 &&
 216 +           t->det_reserved_ft == EXT4_FT_DIR_CSUM)
 217 +               return 0;
 218 +
 219 +       while (extra_data_flags) {
 220 +               if (extra_data_flags & 1) {
 221 +                       dlen += *len + (dlen == 0);
 222 +                       len += *len;
 223 +               }
 224 +               extra_data_flags >>= 1;
 225 +       }
 226 +       return dlen;
 227 +}
 228 +
 229  #endif /* __KERNEL__ */
 230
 231  #endif /* _EXT4_H */
 232 Index: linux-stage/fs/ext4/namei.c
 233 ===================================================================
 234 --- linux-stage.orig/fs/ext4/namei.c
 235 +++ linux-stage/fs/ext4/namei.c
 236 @@ -239,7 +239,8 @@ static unsigned dx_get_count(struct dx_e
 237  static unsigned dx_get_limit(struct dx_entry *entries);
 238  static void dx_set_count(struct dx_entry *entries, unsigned value);
 239  static void dx_set_limit(struct dx_entry *entries, unsigned value);
 240 -static unsigned dx_root_limit(struct inode *dir, unsigned infosize);
 241 +static inline unsigned dx_root_limit(struct inode *dir,
 242 +               struct ext4_dir_entry_2 *dot_de, unsigned infosize);
 243  static unsigned dx_node_limit(struct inode *dir);
 244  static struct dx_frame *dx_probe(const struct qstr *d_name,
 245                                  struct inode *dir,
 246 @@ -379,22 +380,23 @@ static struct dx_countlimit *get_dx_coun
 247  {
 248         struct ext4_dir_entry *dp;
 249         struct dx_root_info *root;
 250 -       int count_offset;
 251 +       int count_offset, dot_rec_len, dotdot_rec_len;
 252
 253         if (le16_to_cpu(dirent->rec_len) == EXT4_BLOCK_SIZE(inode->i_sb))
 254                 count_offset = 8;
 255 -       else if (le16_to_cpu(dirent->rec_len) == 12) {
 256 -               dp = (struct ext4_dir_entry *)(((void *)dirent) + 12);
 257 +       else {
 258 +               dot_rec_len = le16_to_cpu(dirent->rec_len);
 259 +               dp = (struct ext4_dir_entry *)(((void *)dirent) + dot_rec_len);
 260                 if (le16_to_cpu(dp->rec_len) !=
 261 -                   EXT4_BLOCK_SIZE(inode->i_sb) - 12)
 262 +                   EXT4_BLOCK_SIZE(inode->i_sb) - dot_rec_len)
 263                         return NULL;
 264 -               root = (struct dx_root_info *)(((void *)dp + 12));
 265 +               dotdot_rec_len = EXT4_DIR_REC_LEN((struct ext4_dir_entry_2 *)dp);
 266 +               root = (struct dx_root_info *)(((void *)dp + dotdot_rec_len));
 267                 if (root->reserved_zero ||
 268                     root->info_length != sizeof(struct dx_root_info))
 269                         return NULL;
 270 -               count_offset = 32;
 271 -       } else
 272 -               return NULL;
 273 +               count_offset = 8 + dot_rec_len + dotdot_rec_len;
 274 +       }
 275
 276         if (offset)
 277                 *offset = count_offset;
 278 @@ -500,11 +501,12 @@ ext4_next_entry(struct ext4_dir_entry_2
 279   */
 280  struct dx_root_info *dx_get_dx_info(struct ext4_dir_entry_2 *de)
 281  {
 282 +       BUG_ON(de->name_len != 1);
 283         /* get dotdot first */
 284 -       de = (struct ext4_dir_entry_2 *)((char *)de + EXT4_DIR_REC_LEN(1));
 285 +       de = (struct ext4_dir_entry_2 *)((char *)de + EXT4_DIR_REC_LEN(de));
 286
 287         /* dx root info is after dotdot entry */
 288 -       de = (struct ext4_dir_entry_2 *)((char *)de + EXT4_DIR_REC_LEN(2));
 289 +       de = (struct ext4_dir_entry_2 *)((char *)de + EXT4_DIR_REC_LEN(de));
 290
 291         return (struct dx_root_info *)de;
 292  }
 293 @@ -549,10 +551,16 @@ static inline void dx_set_limit(struct d
 294         ((struct dx_countlimit *) entries)->limit = cpu_to_le16(value);
 295  }
 296
 297 -static inline unsigned dx_root_limit(struct inode *dir, unsigned infosize)
 298 +static inline unsigned dx_root_limit(struct inode *dir,
 299 +               struct ext4_dir_entry_2 *dot_de, unsigned infosize)
 300  {
 301 -       unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(1) -
 302 -               EXT4_DIR_REC_LEN(2) - infosize;
 303 +       struct ext4_dir_entry_2 *dotdot_de;
 304 +       unsigned entry_space;
 305 +
 306 +       BUG_ON(dot_de->name_len != 1);
 307 +       dotdot_de = ext4_next_entry(dot_de, dir->i_sb->s_blocksize);
 308 +       entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(dot_de) -
 309 +                        EXT4_DIR_REC_LEN(dotdot_de) - infosize;
 310
 311         if (ext4_has_metadata_csum(dir->i_sb))
 312                 entry_space -= sizeof(struct dx_tail);
 313 @@ -561,7 +569,7 @@ static inline unsigned dx_root_limit(str
 314
 315  static inline unsigned dx_node_limit(struct inode *dir)
 316  {
 317 -       unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(0);
 318 +       unsigned entry_space = dir->i_sb->s_blocksize - __EXT4_DIR_REC_LEN(0);
 319
 320         if (ext4_has_metadata_csum(dir->i_sb))
 321                 entry_space -= sizeof(struct dx_tail);
 322 @@ -611,7 +619,7 @@ static struct stats dx_show_leaf(struct
 323                                 printk(":%x.%u ", h.hash,
 324                                        (unsigned) ((char *) de - base));
 325                         }
 326 -                       space += EXT4_DIR_REC_LEN(de->name_len);
 327 +                       space += EXT4_DIR_REC_LEN(de);
 328                         names++;
 329                 }
 330                 de = ext4_next_entry(de, size);
 331 @@ -719,12 +727,15 @@ dx_probe(const struct qstr *d_name, stru
 332
 333         entries = (struct dx_entry *)(((char *)info) + info->info_length);
 334
 335 -       if (dx_get_limit(entries) != dx_root_limit(dir,
 336 -                                                  info->info_length)) {
 337 +       if (dx_get_limit(entries) !=
 338 +           dx_root_limit(dir, (struct ext4_dir_entry_2 *)bh->b_data,
 339 +                         info->info_length)) {
 340                 ext4_warning(dir->i_sb, "dx entry: limit != root limit "
 341                              "inode #%lu: dx entry: limit %u != root limit %u",
 342                              dir->i_ino, dx_get_limit(entries),
 343 -                            dx_root_limit(dir, info->info_length));
 344 +                            dx_root_limit(dir,
 345 +                                         (struct ext4_dir_entry_2 *)bh->b_data,
 346 +                                         info->info_length));
 347                 brelse(bh);
 348                 *err = ERR_BAD_DX_DIR;
 349                 goto fail;
 350 @@ -916,7 +927,7 @@ static int htree_dirblock_to_tree(struct
 351         de = (struct ext4_dir_entry_2 *) bh->b_data;
 352         top = (struct ext4_dir_entry_2 *) ((char *) de +
 353                                            dir->i_sb->s_blocksize -
 354 -                                          EXT4_DIR_REC_LEN(0));
 355 +                                          __EXT4_DIR_REC_LEN(0));
 356         for (; de < top; de = ext4_next_entry(de, dir->i_sb->s_blocksize)) {
 357                 if (ext4_check_dir_entry(dir, NULL, de, bh,
 358                                 bh->b_data, bh->b_size,
 359 @@ -1520,7 +1531,7 @@ dx_move_dirents(char *from, char *to, st
 360         while (count--) {
 361                 struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *)
 362                                                 (from + (map->offs<<2));
 363 -               rec_len = EXT4_DIR_REC_LEN(de->name_len);
 364 +               rec_len = EXT4_DIR_REC_LEN(de);
 365                 memcpy (to, de, rec_len);
 366                 ((struct ext4_dir_entry_2 *) to)->rec_len =
 367                                 ext4_rec_len_to_disk(rec_len, blocksize);
 368 @@ -1544,7 +1555,7 @@ static struct ext4_dir_entry_2* dx_pack_
 369         while ((char*)de < base + blocksize) {
 370                 next = ext4_next_entry(de, blocksize);
 371                 if (de->inode && de->name_len) {
 372 -                       rec_len = EXT4_DIR_REC_LEN(de->name_len);
 373 +                       rec_len = EXT4_DIR_REC_LEN(de);
 374                         if (de > to)
 375                                 memmove(to, de, rec_len);
 376                         to->rec_len = ext4_rec_len_to_disk(rec_len, blocksize);
 377 @@ -1675,14 +1686,16 @@ int ext4_find_dest_de(struct inode *dir,
 378                       struct buffer_head *bh,
 379                       void *buf, int buf_size,
 380                       const char *name, int namelen,
 381 -                     struct ext4_dir_entry_2 **dest_de)
 382 +                     struct ext4_dir_entry_2 **dest_de, int *dlen)
 383  {
 384         struct ext4_dir_entry_2 *de;
 385 -       unsigned short reclen = EXT4_DIR_REC_LEN(namelen);
 386 +       unsigned short reclen = __EXT4_DIR_REC_LEN(namelen) +
 387 +                                                       (dlen ? *dlen : 0);
 388         int nlen, rlen;
 389         unsigned int offset = 0;
 390         char *top;
 391
 392 +       dlen ? *dlen = 0 : 0; /* default set to 0 */
 393         de = (struct ext4_dir_entry_2 *)buf;
 394         top = buf + buf_size - reclen;
 395         while ((char *) de <= top) {
 396 @@ -1691,10 +1704,26 @@ int ext4_find_dest_de(struct inode *dir,
 397                         return -EIO;
 398                 if (ext4_match(namelen, name, de))
 399                         return -EEXIST;
 400 -               nlen = EXT4_DIR_REC_LEN(de->name_len);
 401 +               nlen = EXT4_DIR_REC_LEN(de);
 402                 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
 403                 if ((de->inode ? rlen - nlen : rlen) >= reclen)
 404                         break;
 405 +               /* Then for dotdot entries, check for the smaller space
 406 +                * required for just the entry, no FID */
 407 +               if (namelen == 2 && memcmp(name, "..", 2) == 0) {
 408 +                       if ((de->inode ? rlen - nlen : rlen) >=
 409 +                           __EXT4_DIR_REC_LEN(namelen)) {
 410 +                               /* set dlen=1 to indicate not
 411 +                                * enough space store fid */
 412 +                               dlen ? *dlen = 1 : 0;
 413 +                               break;
 414 +                       }
 415 +                       /* The new ".." entry must be written over the
 416 +                        * previous ".." entry, which is the first
 417 +                        * entry traversed by this scan. If it doesn't
 418 +                        * fit, something is badly wrong, so -EIO. */
 419 +                       return -EIO;
 420 +               }
 421                 de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
 422                 offset += rlen;
 423         }
 424 @@ -1708,12 +1737,12 @@ int ext4_find_dest_de(struct inode *dir,
 425  void ext4_insert_dentry(struct inode *inode,
 426                         struct ext4_dir_entry_2 *de,
 427                         int buf_size,
 428 -                       const char *name, int namelen)
 429 +                       const char *name, int namelen, void *data)
 430  {
 431
 432         int nlen, rlen;
 433
 434 -       nlen = EXT4_DIR_REC_LEN(de->name_len);
 435 +       nlen = EXT4_DIR_REC_LEN(de);
 436         rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
 437         if (de->inode) {
 438                 struct ext4_dir_entry_2 *de1 =
 439 @@ -1727,6 +1756,11 @@ void ext4_insert_dentry(struct inode *in
 440         ext4_set_de_type(inode->i_sb, de, inode->i_mode);
 441         de->name_len = namelen;
 442         memcpy(de->name, name, namelen);
 443 +       if (data) {
 444 +               de->name[namelen] = 0;
 445 +               memcpy(&de->name[namelen + 1], data, *(char *)data);
 446 +               de->file_type |= EXT4_DIRENT_LUFID;
 447 +       }
 448  }
 449  /*
 450   * Add a new entry into a directory (leaf) block.  If de is non-NULL,
 451 @@ -1745,15 +1779,20 @@ static int add_dirent_to_buf(handle_t *h
 452         int             namelen = dentry->d_name.len;
 453         unsigned int    blocksize = dir->i_sb->s_blocksize;
 454         int             csum_size = 0;
 455 -       int             err;
 456 +       int             err, dlen = 0;
 457 +       unsigned char   *data;
 458
 459 +       data = ext4_dentry_get_data(inode->i_sb, (struct ext4_dentry_param *)
 460 +                                               dentry->d_fsdata);
 461         if (ext4_has_metadata_csum(inode->i_sb))
 462                 csum_size = sizeof(struct ext4_dir_entry_tail);
 463
 464         if (!de) {
 465 +               if (data)
 466 +                       dlen = (*data) + 1;
 467                 err = ext4_find_dest_de(dir, inode,
 468                                         bh, bh->b_data, blocksize - csum_size,
 469 -                                       name, namelen, &de);
 470 +                                       name, namelen, &de, &dlen);
 471                 if (err)
 472                         return err;
 473         }
 474 @@ -1765,7 +1804,10 @@ static int add_dirent_to_buf(handle_t *h
 475         }
 476
 477         /* By now the buffer is marked for journaling */
 478 -       ext4_insert_dentry(inode, de, blocksize, name, namelen);
 479 +       /* If writing the short form of "dotdot", don't add the data section */
 480 +       if (dlen == 1)
 481 +               data = NULL;
 482 +       ext4_insert_dentry(inode, de, blocksize, name, namelen, data);
 483
 484         /*
 485          * XXX shouldn't update any times until successful
 486 @@ -1877,7 +1919,8 @@ static int make_indexed_dir(handle_t *ha
 487
 488         dx_set_block(entries, 1);
 489         dx_set_count(entries, 1);
 490 -       dx_set_limit(entries, dx_root_limit(dir, sizeof(*dx_info)));
 491 +       dx_set_limit(entries, dx_root_limit(dir,
 492 +                                        dot_de, sizeof(*dx_info)));
 493
 494         /* Initialize as for dx_probe */
 495         hinfo.hash_version = dx_info->hash_version;
 496 @@ -1927,6 +1970,8 @@ static int ext4_update_dotdot(handle_t *
 497         struct buffer_head *dir_block;
 498         struct ext4_dir_entry_2 *de;
 499         int len, journal = 0, err = 0;
 500 +       int dlen = 0;
 501 +       char *data;
 502
 503         if (IS_ERR(handle))
 504                 return PTR_ERR(handle);
 505 @@ -1942,19 +1987,24 @@ static int ext4_update_dotdot(handle_t *
 506         /* the first item must be "." */
 507         assert(de->name_len == 1 && de->name[0] == '.');
 508         len = le16_to_cpu(de->rec_len);
 509 -       assert(len >= EXT4_DIR_REC_LEN(1));
 510 -       if (len > EXT4_DIR_REC_LEN(1)) {
 511 +       assert(len >= __EXT4_DIR_REC_LEN(1));
 512 +       if (len > __EXT4_DIR_REC_LEN(1)) {
 513                 BUFFER_TRACE(dir_block, "get_write_access");
 514                 err = ext4_journal_get_write_access(handle, dir_block);
 515                 if (err)
 516                         goto out_journal;
 517
 518                 journal = 1;
 519 -               de->rec_len = cpu_to_le16(EXT4_DIR_REC_LEN(1));
 520 +               de->rec_len = cpu_to_le16(EXT4_DIR_REC_LEN(de));
 521         }
 522
 523 -       len -= EXT4_DIR_REC_LEN(1);
 524 -       assert(len == 0 || len >= EXT4_DIR_REC_LEN(2));
 525 +       len -= EXT4_DIR_REC_LEN(de);
 526 +       data = ext4_dentry_get_data(dir->i_sb,
 527 +                       (struct ext4_dentry_param *)dentry->d_fsdata);
 528 +       if (data)
 529 +               dlen = *data + 1;
 530 +       assert(len == 0 || len >= __EXT4_DIR_REC_LEN(2 + dlen));
 531 +
 532         de = (struct ext4_dir_entry_2 *)
 533                         ((char *) de + le16_to_cpu(de->rec_len));
 534         if (!journal) {
 535 @@ -1968,10 +2018,15 @@ static int ext4_update_dotdot(handle_t *
 536         if (len > 0)
 537                 de->rec_len = cpu_to_le16(len);
 538         else
 539 -               assert(le16_to_cpu(de->rec_len) >= EXT4_DIR_REC_LEN(2));
 540 +               assert(le16_to_cpu(de->rec_len) >= __EXT4_DIR_REC_LEN(2));
 541         de->name_len = 2;
 542         strcpy(de->name, "..");
 543 -       ext4_set_de_type(dir->i_sb, de, S_IFDIR);
 544 +       if (data != NULL && ext4_get_dirent_data_len(de) >= dlen) {
 545 +               de->name[2] = 0;
 546 +               memcpy(&de->name[2 + 1], data, *data);
 547 +               ext4_set_de_type(dir->i_sb, de, S_IFDIR);
 548 +               de->file_type |= EXT4_DIRENT_LUFID;
 549 +       }
 550
 551  out_journal:
 552         if (journal) {
 553 @@ -2445,37 +2500,70 @@ retry:
 554         return err;
 555  }
 556
 557 +struct tp_block {
 558 +       struct inode *inode;
 559 +       void *data1;
 560 +       void *data2;
 561 +};
 562 +
 563  struct ext4_dir_entry_2 *ext4_init_dot_dotdot(struct inode *inode,
 564                           struct ext4_dir_entry_2 *de,
 565                           int blocksize, int csum_size,
 566                           unsigned int parent_ino, int dotdot_real_len)
 567  {
 568 +       void *data1 = NULL, *data2 = NULL;
 569 +       int dot_reclen = 0;
 570 +
 571 +       if (dotdot_real_len == 10) {
 572 +               struct tp_block *tpb = (struct tp_block *)inode;
 573 +               data1 = tpb->data1;
 574 +               data2 = tpb->data2;
 575 +               inode = tpb->inode;
 576 +               dotdot_real_len = 0;
 577 +       }
 578         de->inode = cpu_to_le32(inode->i_ino);
 579         de->name_len = 1;
 580 -       de->rec_len = ext4_rec_len_to_disk(EXT4_DIR_REC_LEN(de->name_len),
 581 -                                          blocksize);
 582         strcpy(de->name, ".");
 583         ext4_set_de_type(inode->i_sb, de, S_IFDIR);
 584
 585 +       /* get packed fid data*/
 586 +       data1 = ext4_dentry_get_data(inode->i_sb,
 587 +                               (struct ext4_dentry_param *) data1);
 588 +       if (data1) {
 589 +               de->name[1] = 0;
 590 +               memcpy(&de->name[2], data1, *(char *) data1);
 591 +               de->file_type |= EXT4_DIRENT_LUFID;
 592 +       }
 593 +       de->rec_len = cpu_to_le16(EXT4_DIR_REC_LEN(de));
 594 +       dot_reclen = cpu_to_le16(de->rec_len);
 595         de = ext4_next_entry(de, blocksize);
 596         de->inode = cpu_to_le32(parent_ino);
 597         de->name_len = 2;
 598 +       strcpy(de->name, "..");
 599 +       ext4_set_de_type(inode->i_sb, de, S_IFDIR);
 600 +       data2 = ext4_dentry_get_data(inode->i_sb,
 601 +                       (struct ext4_dentry_param *) data2);
 602 +       if (data2) {
 603 +               de->name[2] = 0;
 604 +               memcpy(&de->name[3], data2, *(char *) data2);
 605 +               de->file_type |= EXT4_DIRENT_LUFID;
 606 +       }
 607         if (!dotdot_real_len)
 608                 de->rec_len = ext4_rec_len_to_disk(blocksize -
 609 -                                       (csum_size + EXT4_DIR_REC_LEN(1)),
 610 +                                       (csum_size + dot_reclen),
 611                                         blocksize);
 612         else
 613                 de->rec_len = ext4_rec_len_to_disk(
 614 -                               EXT4_DIR_REC_LEN(de->name_len), blocksize);
 615 -       strcpy(de->name, "..");
 616 -       ext4_set_de_type(inode->i_sb, de, S_IFDIR);
 617 +                               EXT4_DIR_REC_LEN(de), blocksize);
 618
 619         return ext4_next_entry(de, blocksize);
 620  }
 621
 622  static int ext4_init_new_dir(handle_t *handle, struct inode *dir,
 623 -                            struct inode *inode)
 624 +                            struct inode *inode,
 625 +                            const void *data1, const void *data2)
 626  {
 627 +       struct tp_block param;
 628         struct buffer_head *dir_block = NULL;
 629         struct ext4_dir_entry_2 *de;
 630         struct ext4_dir_entry_tail *t;
 631 @@ -2500,7 +2588,11 @@ static int ext4_init_new_dir(handle_t *h
 632         if (IS_ERR(dir_block))
 633                 return PTR_ERR(dir_block);
 634         de = (struct ext4_dir_entry_2 *)dir_block->b_data;
 635 -       ext4_init_dot_dotdot(inode, de, blocksize, csum_size, dir->i_ino, 0);
 636 +       param.inode = inode;
 637 +       param.data1 = (void *)data1;
 638 +       param.data2 = (void *)data2;
 639 +       ext4_init_dot_dotdot((struct inode *)(&param), de, blocksize,
 640 +                            csum_size, dir->i_ino, 10);
 641         set_nlink(inode, 2);
 642         if (csum_size) {
 643                 t = EXT4_DIRENT_TAIL(dir_block->b_data, blocksize);
 644 @@ -2517,6 +2609,29 @@ out:
 645         return err;
 646  }
 647
 648 +/* Initialize @inode as a subdirectory of @dir, and add the
 649 + * "." and ".." entries into the first directory block. */
 650 +int ext4_add_dot_dotdot(handle_t *handle, struct inode *dir,
 651 +                       struct inode *inode,
 652 +                       const void *data1, const void *data2)
 653 +{
 654 +       int rc;
 655 +
 656 +       if (IS_ERR(handle))
 657 +               return PTR_ERR(handle);
 658 +
 659 +       if (IS_DIRSYNC(dir))
 660 +               ext4_handle_sync(handle);
 661 +
 662 +       inode->i_op = &ext4_dir_inode_operations.ops;
 663 +       inode->i_fop = &ext4_dir_operations;
 664 +       rc = ext4_init_new_dir(handle, dir, inode, data1, data2);
 665 +       if (!rc)
 666 +               rc = ext4_mark_inode_dirty(handle, inode);
 667 +       return rc;
 668 +}
 669 +EXPORT_SYMBOL(ext4_add_dot_dotdot);
 670 +
 671  static int ext4_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
 672  {
 673         handle_t *handle;
 674 @@ -2542,7 +2657,7 @@ retry:
 675         inode->i_op = &ext4_dir_inode_operations.ops;
 676         inode->i_fop = &ext4_dir_operations;
 677         inode->i_flags |= S_IOPS_WRAPPER;
 678 -       err = ext4_init_new_dir(handle, dir, inode);
 679 +       err = ext4_init_new_dir(handle, dir, inode, NULL, NULL);
 680         if (err)
 681                 goto out_clear_inode;
 682         err = ext4_mark_inode_dirty(handle, inode);
 683 @@ -2594,7 +2709,7 @@ static int empty_dir(struct inode *inode
 684         }
 685
 686         sb = inode->i_sb;
 687 -       if (inode->i_size < EXT4_DIR_REC_LEN(1) + EXT4_DIR_REC_LEN(2)) {
 688 +       if (inode->i_size < __EXT4_DIR_REC_LEN(1) + __EXT4_DIR_REC_LEN(2)) {
 689                 EXT4_ERROR_INODE(inode, "invalid size");
 690                 return 1;
 691         }
 692 Index: linux-stage/fs/ext4/inline.c
 693 ===================================================================
 694 --- linux-stage.orig/fs/ext4/inline.c
 695 +++ linux-stage/fs/ext4/inline.c
 696 @@ -1006,7 +1006,7 @@ static int ext4_add_dirent_to_inline(han
 697
 698         err = ext4_find_dest_de(dir, inode, iloc->bh,
 699                                 inline_start, inline_size,
 700 -                               name, namelen, &de);
 701 +                               name, namelen, &de, NULL);
 702         if (err)
 703                 return err;
 704
 705 @@ -1014,7 +1014,7 @@ static int ext4_add_dirent_to_inline(han
 706         err = ext4_journal_get_write_access(handle, iloc->bh);
 707         if (err)
 708                 return err;
 709 -       ext4_insert_dentry(inode, de, inline_size, name, namelen);
 710 +       ext4_insert_dentry(inode, de, inline_size, name, namelen, NULL);
 711
 712         ext4_show_inline_dir(dir, iloc->bh, inline_start, inline_size);
 713
 714 @@ -1084,7 +1084,7 @@ static int ext4_update_inline_dir(handle
 715         int old_size = EXT4_I(dir)->i_inline_size - EXT4_MIN_INLINE_DATA_SIZE;
 716         int new_size = get_max_inline_xattr_value_size(dir, iloc);
 717
 718 -       if (new_size - old_size <= EXT4_DIR_REC_LEN(1))
 719 +       if (new_size - old_size <= __EXT4_DIR_REC_LEN(1))
 720                 return -ENOSPC;
 721
 722         ret = ext4_update_inline_data(handle, dir,
 723 @@ -1365,7 +1365,7 @@ int htree_inlinedir_to_tree(struct file
 724                         fake.name_len = 1;
 725                         strcpy(fake.name, ".");
 726                         fake.rec_len = ext4_rec_len_to_disk(
 727 -                                               EXT4_DIR_REC_LEN(fake.name_len),
 728 +                                               EXT4_DIR_REC_LEN(&fake),
 729                                                 inline_size);
 730                         ext4_set_de_type(inode->i_sb, &fake, S_IFDIR);
 731                         de = &fake;
 732 @@ -1375,7 +1375,7 @@ int htree_inlinedir_to_tree(struct file
 733                         fake.name_len = 2;
 734                         strcpy(fake.name, "..");
 735                         fake.rec_len = ext4_rec_len_to_disk(
 736 -                                               EXT4_DIR_REC_LEN(fake.name_len),
 737 +                                               EXT4_DIR_REC_LEN(&fake),
 738                                                 inline_size);
 739                         ext4_set_de_type(inode->i_sb, &fake, S_IFDIR);
 740                         de = &fake;
 741 @@ -1473,8 +1473,8 @@ int ext4_read_inline_dir(struct file *fi
 742          * So we will use extra_offset and extra_size to indicate them
 743          * during the inline dir iteration.
 744          */
 745 -       dotdot_offset = EXT4_DIR_REC_LEN(1);
 746 -       dotdot_size = dotdot_offset + EXT4_DIR_REC_LEN(2);
 747 +       dotdot_offset = __EXT4_DIR_REC_LEN(1);
 748 +       dotdot_size = dotdot_offset + __EXT4_DIR_REC_LEN(2);
 749         extra_offset = dotdot_size - EXT4_INLINE_DOTDOT_SIZE;
 750         extra_size = extra_offset + inline_size;
 751
 752 @@ -1511,7 +1511,7 @@ revalidate:
 753                                  * failure will be detected in the
 754                                  * dirent test below. */
 755                                 if (ext4_rec_len_from_disk(de->rec_len,
 756 -                                       extra_size) < EXT4_DIR_REC_LEN(1))
 757 +                                       extra_size) < __EXT4_DIR_REC_LEN(1))
 758                                         break;
 759                                 i += ext4_rec_len_from_disk(de->rec_len,
 760                                                             extra_size);
 761 Index: linux-stage/fs/ext4/super.c
 762 ===================================================================
 763 --- linux-stage.orig/fs/ext4/super.c
 764 +++ linux-stage/fs/ext4/super.c
 765 @@ -1155,7 +1155,7 @@ enum {
 766         Opt_data_err_abort, Opt_data_err_ignore,
 767         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
 768         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_jqfmt_vfsv1, Opt_quota,
 769 -       Opt_noquota, Opt_barrier, Opt_nobarrier, Opt_err,
 770 +       Opt_noquota, Opt_barrier, Opt_nobarrier, Opt_err, Opt_dirdata,
 771         Opt_usrquota, Opt_grpquota, Opt_i_version, Opt_dax,
 772         Opt_stripe, Opt_delalloc, Opt_nodelalloc, Opt_mblk_io_submit,
 773         Opt_nomblk_io_submit, Opt_block_validity, Opt_noblock_validity,
 774 @@ -1223,6 +1223,7 @@ static const match_table_t tokens = {
 775         {Opt_stripe, "stripe=%u"},
 776         {Opt_delalloc, "delalloc"},
 777         {Opt_nodelalloc, "nodelalloc"},
 778 +       {Opt_dirdata, "dirdata"},
 779         {Opt_removed, "mblk_io_submit"},
 780         {Opt_removed, "nomblk_io_submit"},
 781         {Opt_block_validity, "block_validity"},
 782 @@ -1436,6 +1437,7 @@ static const struct mount_opts {
 783         {Opt_usrjquota, 0, MOPT_Q},
 784         {Opt_grpjquota, 0, MOPT_Q},
 785         {Opt_offusrjquota, 0, MOPT_Q},
 786 +       {Opt_dirdata, EXT4_MOUNT_DIRDATA, MOPT_SET},
 787         {Opt_offgrpjquota, 0, MOPT_Q},
 788         {Opt_jqfmt_vfsold, QFMT_VFS_OLD, MOPT_QFMT},
 789         {Opt_jqfmt_vfsv0, QFMT_VFS_V0, MOPT_QFMT},