ldiskfs/kernel_patches/patches/sles12/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 @@ -70,11 +70,11 @@ 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 @@ -202,7 +202,7 @@ static int ext4_readdir(struct file *fil
  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 @@ -421,12 +421,17 @@ 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 @@ -435,7 +440,7 @@ 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 Index: linux-stage/fs/ext4/ext4.h
  66 ===================================================================
  67 --- linux-stage.orig/fs/ext4/ext4.h
  68 +++ linux-stage/fs/ext4/ext4.h
  69 @@ -960,6 +960,7 @@ struct ext4_inode_info {
  70  #define EXT4_MOUNT_ERRORS_MASK         0x00070
  71  #define EXT4_MOUNT_MINIX_DF            0x00080 /* Mimics the Minix statfs */
  72  #define EXT4_MOUNT_NOLOAD              0x00100 /* Don't use existing journal*/
  73 +#define EXT4_MOUNT_DIRDATA             0x00200 /* Data in directory entries*/
  74  #define EXT4_MOUNT_DATA_FLAGS          0x00C00 /* Mode for data writes: */
  75  #define EXT4_MOUNT_JOURNAL_DATA                0x00400 /* Write data to journal */
  76  #define EXT4_MOUNT_ORDERED_DATA                0x00800 /* Flush data before commit */
  77 @@ -1538,6 +1539,7 @@ static inline void ext4_clear_state_flag
  78                                          EXT4_FEATURE_INCOMPAT_64BIT| \
  79                                          EXT4_FEATURE_INCOMPAT_FLEX_BG| \
  80                                          EXT4_FEATURE_INCOMPAT_MMP |    \
  81 +                                        EXT4_FEATURE_INCOMPAT_DIRDATA| \
  82                                          EXT4_FEATURE_INCOMPAT_INLINE_DATA)
  83  #define EXT4_FEATURE_RO_COMPAT_SUPP    (EXT4_FEATURE_RO_COMPAT_SPARSE_SUPER| \
  84                                          EXT4_FEATURE_RO_COMPAT_LARGE_FILE| \
  85 @@ -1644,6 +1646,43 @@ struct ext4_dir_entry_tail {
  86  #define EXT4_FT_SYMLINK                7
  87
  88  #define EXT4_FT_MAX            8
  89 +#define EXT4_FT_MASK           0xf
  90 +
  91 +#if EXT4_FT_MAX > EXT4_FT_MASK
  92 +#error "conflicting EXT4_FT_MAX and EXT4_FT_MASK"
  93 +#endif
  94 +
  95 +/*
  96 + * d_type has 4 unused bits, so it can hold four types data. these different
  97 + * type of data (e.g. lustre data, high 32 bits of 64-bit inode number) can be
  98 + * stored, in flag order, after file-name in ext4 dirent.
  99 +*/
 100 +/*
 101 + * this flag is added to d_type if ext4 dirent has extra data after
 102 + * filename. this data length is variable and length is stored in first byte
 103 + * of data. data start after filename NUL byte.
 104 + * This is used by Lustre FS.
 105 +  */
 106 +#define EXT4_DIRENT_LUFID              0x10
 107 +
 108 +#define EXT4_LUFID_MAGIC    0xAD200907UL
 109 +struct ext4_dentry_param {
 110 +       __u32  edp_magic;       /* EXT4_LUFID_MAGIC */
 111 +       char   edp_len;         /* size of edp_data in bytes */
 112 +       char   edp_data[0];     /* packed array of data */
 113 +} __packed;
 114 +
 115 +static inline unsigned char *ext4_dentry_get_data(struct super_block *sb,
 116 +                                                 struct ext4_dentry_param *p)
 117 +
 118 +{
 119 +       if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_DIRDATA))
 120 +               return NULL;
 121 +       if (p && p->edp_magic == EXT4_LUFID_MAGIC)
 122 +               return &p->edp_len;
 123 +       else
 124 +               return NULL;
 125 +}
 126
 127  #define EXT4_FT_DIR_CSUM       0xDE
 128
 129 @@ -1654,8 +1693,11 @@ struct ext4_dir_entry_tail {
 130   */
 131  #define EXT4_DIR_PAD                   4
 132  #define EXT4_DIR_ROUND                 (EXT4_DIR_PAD - 1)
 133 -#define EXT4_DIR_REC_LEN(name_len)     (((name_len) + 8 + EXT4_DIR_ROUND) & \
 134 +#define __EXT4_DIR_REC_LEN(name_len)   (((name_len) + 8 + EXT4_DIR_ROUND) & \
 135                                          ~EXT4_DIR_ROUND)
 136 +#define EXT4_DIR_REC_LEN(de)           (__EXT4_DIR_REC_LEN((de)->name_len +\
 137 +                                       ext4_get_dirent_data_len(de)))
 138 +
 139  #define EXT4_MAX_REC_LEN               ((1<<16)-1)
 140
 141  /*
 142 @@ -1991,11 +2033,11 @@ extern int ext4_find_dest_de(struct inod
 143                              struct buffer_head *bh,
 144                              void *buf, int buf_size,
 145                              const char *name, int namelen,
 146 -                            struct ext4_dir_entry_2 **dest_de);
 147 +                            struct ext4_dir_entry_2 **dest_de, int *dlen);
 148  void ext4_insert_dentry(struct inode *inode,
 149                         struct ext4_dir_entry_2 *de,
 150                         int buf_size,
 151 -                       const char *name, int namelen);
 152 +                       const char *name, int namelen, void *data);
 153  static inline void ext4_update_dx_flag(struct inode *inode)
 154  {
 155         if (!EXT4_HAS_COMPAT_FEATURE(inode->i_sb,
 156 @@ -2008,11 +2050,18 @@ static unsigned char ext4_filetype_table
 157
 158  static inline  unsigned char get_dtype(struct super_block *sb, int filetype)
 159  {
 160 +       int fl_index = filetype & EXT4_FT_MASK;
 161 +
 162         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FILETYPE) ||
 163 -           (filetype >= EXT4_FT_MAX))
 164 +           (fl_index >= EXT4_FT_MAX))
 165                 return DT_UNKNOWN;
 166
 167 -       return ext4_filetype_table[filetype];
 168 +       if (!test_opt(sb, DIRDATA))
 169 +               return ext4_filetype_table[fl_index];
 170 +
 171 +       return (ext4_filetype_table[fl_index]) |
 172 +               (filetype & EXT4_DIRENT_LUFID);
 173 +
 174  }
 175
 176  /* fsync.c */
 177 @@ -2158,6 +2207,8 @@ extern struct inode *ext4_create_inode(h
 178  extern int ext4_delete_entry(handle_t *handle, struct inode * dir,
 179                              struct ext4_dir_entry_2 *de_del,
 180                              struct buffer_head *bh);
 181 +extern int ext4_add_dot_dotdot(handle_t *handle, struct inode *dir,
 182 +                              struct inode *inode, const void *, const void *);
 183  extern int search_dir(struct buffer_head *bh,
 184                       char *search_buf,
 185                       int buf_size,
 186 @@ -2834,6 +2885,36 @@ extern struct mutex ext4__aio_mutex[EXT4
 187  extern int ext4_resize_begin(struct super_block *sb);
 188  extern void ext4_resize_end(struct super_block *sb);
 189
 190 +/*
 191 + * Compute the total directory entry data length.
 192 + * This includes the filename and an implicit NUL terminator (always present),
 193 + * and optional extensions.  Each extension has a bit set in the high 4 bits of
 194 + * de->file_type, and the extension length is the first byte in each entry.
 195 + */
 196 +static inline int ext4_get_dirent_data_len(struct ext4_dir_entry_2 *de)
 197 +{
 198 +       char *len = de->name + de->name_len + 1 /* NUL terminator */;
 199 +       int dlen = 0;
 200 +       __u8 extra_data_flags = (de->file_type & ~EXT4_FT_MASK) >> 4;
 201 +       struct ext4_dir_entry_tail *t = (struct ext4_dir_entry_tail *)de;
 202 +
 203 +       if (!t->det_reserved_zero1 &&
 204 +           le16_to_cpu(t->det_rec_len) ==
 205 +               sizeof(struct ext4_dir_entry_tail) &&
 206 +           !t->det_reserved_zero2 &&
 207 +           t->det_reserved_ft == EXT4_FT_DIR_CSUM)
 208 +               return 0;
 209 +
 210 +       while (extra_data_flags) {
 211 +               if (extra_data_flags & 1) {
 212 +                       dlen += *len + (dlen == 0);
 213 +                       len += *len;
 214 +               }
 215 +               extra_data_flags >>= 1;
 216 +       }
 217 +       return dlen;
 218 +}
 219 +
 220  #endif /* __KERNEL__ */
 221
 222  #endif /* _EXT4_H */
 223 Index: linux-stage/fs/ext4/namei.c
 224 ===================================================================
 225 --- linux-stage.orig/fs/ext4/namei.c
 226 +++ linux-stage/fs/ext4/namei.c
 227 @@ -239,7 +239,8 @@ static unsigned dx_get_count(struct dx_e
 228  static unsigned dx_get_limit(struct dx_entry *entries);
 229  static void dx_set_count(struct dx_entry *entries, unsigned value);
 230  static void dx_set_limit(struct dx_entry *entries, unsigned value);
 231 -static unsigned dx_root_limit(struct inode *dir, unsigned infosize);
 232 +static inline unsigned dx_root_limit(struct inode *dir,
 233 +               struct ext4_dir_entry_2 *dot_de, unsigned infosize);
 234  static unsigned dx_node_limit(struct inode *dir);
 235  static struct dx_frame *dx_probe(const struct qstr *d_name,
 236                                  struct inode *dir,
 237 @@ -379,22 +380,23 @@ static struct dx_countlimit *get_dx_coun
 238  {
 239         struct ext4_dir_entry *dp;
 240         struct dx_root_info *root;
 241 -       int count_offset;
 242 +       int count_offset, dot_rec_len, dotdot_rec_len;
 243
 244         if (le16_to_cpu(dirent->rec_len) == EXT4_BLOCK_SIZE(inode->i_sb))
 245                 count_offset = 8;
 246 -       else if (le16_to_cpu(dirent->rec_len) == 12) {
 247 -               dp = (struct ext4_dir_entry *)(((void *)dirent) + 12);
 248 +       else {
 249 +               dot_rec_len = le16_to_cpu(dirent->rec_len);
 250 +               dp = (struct ext4_dir_entry *)(((void *)dirent) + dot_rec_len);
 251                 if (le16_to_cpu(dp->rec_len) !=
 252 -                   EXT4_BLOCK_SIZE(inode->i_sb) - 12)
 253 +                   EXT4_BLOCK_SIZE(inode->i_sb) - dot_rec_len)
 254                         return NULL;
 255 -               root = (struct dx_root_info *)(((void *)dp + 12));
 256 +               dotdot_rec_len = EXT4_DIR_REC_LEN((struct ext4_dir_entry_2 *)dp);
 257 +               root = (struct dx_root_info *)(((void *)dp + dotdot_rec_len));
 258                 if (root->reserved_zero ||
 259                     root->info_length != sizeof(struct dx_root_info))
 260                         return NULL;
 261 -               count_offset = 32;
 262 -       } else
 263 -               return NULL;
 264 +               count_offset = 8 + dot_rec_len + dotdot_rec_len;
 265 +       }
 266
 267         if (offset)
 268                 *offset = count_offset;
 269 @@ -500,11 +501,12 @@ ext4_next_entry(struct ext4_dir_entry_2
 270   */
 271  struct dx_root_info *dx_get_dx_info(struct ext4_dir_entry_2 *de)
 272  {
 273 +       BUG_ON(de->name_len != 1);
 274         /* get dotdot first */
 275 -       de = (struct ext4_dir_entry_2 *)((char *)de + EXT4_DIR_REC_LEN(1));
 276 +       de = (struct ext4_dir_entry_2 *)((char *)de + EXT4_DIR_REC_LEN(de));
 277
 278         /* dx root info is after dotdot entry */
 279 -       de = (struct ext4_dir_entry_2 *)((char *)de + EXT4_DIR_REC_LEN(2));
 280 +       de = (struct ext4_dir_entry_2 *)((char *)de + EXT4_DIR_REC_LEN(de));
 281
 282         return (struct dx_root_info *)de;
 283  }
 284 @@ -549,10 +551,16 @@ static inline void dx_set_limit(struct d
 285         ((struct dx_countlimit *) entries)->limit = cpu_to_le16(value);
 286  }
 287
 288 -static inline unsigned dx_root_limit(struct inode *dir, unsigned infosize)
 289 +static inline unsigned dx_root_limit(struct inode *dir,
 290 +               struct ext4_dir_entry_2 *dot_de, unsigned infosize)
 291  {
 292 -       unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(1) -
 293 -               EXT4_DIR_REC_LEN(2) - infosize;
 294 +       struct ext4_dir_entry_2 *dotdot_de;
 295 +       unsigned entry_space;
 296 +
 297 +       BUG_ON(dot_de->name_len != 1);
 298 +       dotdot_de = ext4_next_entry(dot_de, dir->i_sb->s_blocksize);
 299 +       entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(dot_de) -
 300 +                        EXT4_DIR_REC_LEN(dotdot_de) - infosize;
 301
 302         if (ext4_has_metadata_csum(dir->i_sb))
 303                 entry_space -= sizeof(struct dx_tail);
 304 @@ -561,7 +569,7 @@ static inline unsigned dx_root_limit(str
 305
 306  static inline unsigned dx_node_limit(struct inode *dir)
 307  {
 308 -       unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(0);
 309 +       unsigned entry_space = dir->i_sb->s_blocksize - __EXT4_DIR_REC_LEN(0);
 310
 311         if (ext4_has_metadata_csum(dir->i_sb))
 312                 entry_space -= sizeof(struct dx_tail);
 313 @@ -611,7 +619,7 @@ static struct stats dx_show_leaf(struct
 314                                 printk(":%x.%u ", h.hash,
 315                                        (unsigned) ((char *) de - base));
 316                         }
 317 -                       space += EXT4_DIR_REC_LEN(de->name_len);
 318 +                       space += EXT4_DIR_REC_LEN(de);
 319                         names++;
 320                 }
 321                 de = ext4_next_entry(de, size);
 322 @@ -719,12 +727,15 @@ dx_probe(const struct qstr *d_name, stru
 323
 324         entries = (struct dx_entry *)(((char *)info) + info->info_length);
 325
 326 -       if (dx_get_limit(entries) != dx_root_limit(dir,
 327 -                                                  info->info_length)) {
 328 +       if (dx_get_limit(entries) !=
 329 +           dx_root_limit(dir, (struct ext4_dir_entry_2 *)bh->b_data,
 330 +                         info->info_length)) {
 331                 ext4_warning(dir->i_sb, "dx entry: limit != root limit "
 332                              "inode #%lu: dx entry: limit %u != root limit %u",
 333                              dir->i_ino, dx_get_limit(entries),
 334 -                            dx_root_limit(dir, info->info_length));
 335 +                            dx_root_limit(dir,
 336 +                                         (struct ext4_dir_entry_2 *)bh->b_data,
 337 +                                         info->info_length));
 338                 brelse(bh);
 339                 *err = ERR_BAD_DX_DIR;
 340                 goto fail;
 341 @@ -916,7 +927,7 @@ static int htree_dirblock_to_tree(struct
 342         de = (struct ext4_dir_entry_2 *) bh->b_data;
 343         top = (struct ext4_dir_entry_2 *) ((char *) de +
 344                                            dir->i_sb->s_blocksize -
 345 -                                          EXT4_DIR_REC_LEN(0));
 346 +                                          __EXT4_DIR_REC_LEN(0));
 347         for (; de < top; de = ext4_next_entry(de, dir->i_sb->s_blocksize)) {
 348                 if (ext4_check_dir_entry(dir, NULL, de, bh,
 349                                 bh->b_data, bh->b_size,
 350 @@ -1508,7 +1519,7 @@ dx_move_dirents(char *from, char *to, st
 351         while (count--) {
 352                 struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *)
 353                                                 (from + (map->offs<<2));
 354 -               rec_len = EXT4_DIR_REC_LEN(de->name_len);
 355 +               rec_len = EXT4_DIR_REC_LEN(de);
 356                 memcpy (to, de, rec_len);
 357                 ((struct ext4_dir_entry_2 *) to)->rec_len =
 358                                 ext4_rec_len_to_disk(rec_len, blocksize);
 359 @@ -1532,7 +1543,7 @@ static struct ext4_dir_entry_2* dx_pack_
 360         while ((char*)de < base + blocksize) {
 361                 next = ext4_next_entry(de, blocksize);
 362                 if (de->inode && de->name_len) {
 363 -                       rec_len = EXT4_DIR_REC_LEN(de->name_len);
 364 +                       rec_len = EXT4_DIR_REC_LEN(de);
 365                         if (de > to)
 366                                 memmove(to, de, rec_len);
 367                         to->rec_len = ext4_rec_len_to_disk(rec_len, blocksize);
 368 @@ -1663,14 +1674,16 @@ int ext4_find_dest_de(struct inode *dir,
 369                       struct buffer_head *bh,
 370                       void *buf, int buf_size,
 371                       const char *name, int namelen,
 372 -                     struct ext4_dir_entry_2 **dest_de)
 373 +                     struct ext4_dir_entry_2 **dest_de, int *dlen)
 374  {
 375         struct ext4_dir_entry_2 *de;
 376 -       unsigned short reclen = EXT4_DIR_REC_LEN(namelen);
 377 +       unsigned short reclen = __EXT4_DIR_REC_LEN(namelen) +
 378 +                                                       (dlen ? *dlen : 0);
 379         int nlen, rlen;
 380         unsigned int offset = 0;
 381         char *top;
 382
 383 +       dlen ? *dlen = 0 : 0; /* default set to 0 */
 384         de = (struct ext4_dir_entry_2 *)buf;
 385         top = buf + buf_size - reclen;
 386         while ((char *) de <= top) {
 387 @@ -1679,10 +1692,26 @@ int ext4_find_dest_de(struct inode *dir,
 388                         return -EIO;
 389                 if (ext4_match(namelen, name, de))
 390                         return -EEXIST;
 391 -               nlen = EXT4_DIR_REC_LEN(de->name_len);
 392 +               nlen = EXT4_DIR_REC_LEN(de);
 393                 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
 394                 if ((de->inode ? rlen - nlen : rlen) >= reclen)
 395                         break;
 396 +               /* Then for dotdot entries, check for the smaller space
 397 +                * required for just the entry, no FID */
 398 +               if (namelen == 2 && memcmp(name, "..", 2) == 0) {
 399 +                       if ((de->inode ? rlen - nlen : rlen) >=
 400 +                           __EXT4_DIR_REC_LEN(namelen)) {
 401 +                               /* set dlen=1 to indicate not
 402 +                                * enough space store fid */
 403 +                               dlen ? *dlen = 1 : 0;
 404 +                               break;
 405 +                       }
 406 +                       /* The new ".." entry must be written over the
 407 +                        * previous ".." entry, which is the first
 408 +                        * entry traversed by this scan. If it doesn't
 409 +                        * fit, something is badly wrong, so -EIO. */
 410 +                       return -EIO;
 411 +               }
 412                 de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
 413                 offset += rlen;
 414         }
 415 @@ -1696,12 +1725,12 @@ int ext4_find_dest_de(struct inode *dir,
 416  void ext4_insert_dentry(struct inode *inode,
 417                         struct ext4_dir_entry_2 *de,
 418                         int buf_size,
 419 -                       const char *name, int namelen)
 420 +                       const char *name, int namelen, void *data)
 421  {
 422
 423         int nlen, rlen;
 424
 425 -       nlen = EXT4_DIR_REC_LEN(de->name_len);
 426 +       nlen = EXT4_DIR_REC_LEN(de);
 427         rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
 428         if (de->inode) {
 429                 struct ext4_dir_entry_2 *de1 =
 430 @@ -1715,6 +1744,11 @@ void ext4_insert_dentry(struct inode *in
 431         ext4_set_de_type(inode->i_sb, de, inode->i_mode);
 432         de->name_len = namelen;
 433         memcpy(de->name, name, namelen);
 434 +       if (data) {
 435 +               de->name[namelen] = 0;
 436 +               memcpy(&de->name[namelen + 1], data, *(char *)data);
 437 +               de->file_type |= EXT4_DIRENT_LUFID;
 438 +       }
 439  }
 440  /*
 441   * Add a new entry into a directory (leaf) block.  If de is non-NULL,
 442 @@ -1733,15 +1767,20 @@ static int add_dirent_to_buf(handle_t *h
 443         int             namelen = dentry->d_name.len;
 444         unsigned int    blocksize = dir->i_sb->s_blocksize;
 445         int             csum_size = 0;
 446 -       int             err;
 447 +       int             err, dlen = 0;
 448 +       unsigned char   *data;
 449
 450 +       data = ext4_dentry_get_data(inode->i_sb, (struct ext4_dentry_param *)
 451 +                                               dentry->d_fsdata);
 452         if (ext4_has_metadata_csum(inode->i_sb))
 453                 csum_size = sizeof(struct ext4_dir_entry_tail);
 454
 455         if (!de) {
 456 +               if (data)
 457 +                       dlen = (*data) + 1;
 458                 err = ext4_find_dest_de(dir, inode,
 459                                         bh, bh->b_data, blocksize - csum_size,
 460 -                                       name, namelen, &de);
 461 +                                       name, namelen, &de, &dlen);
 462                 if (err)
 463                         return err;
 464         }
 465 @@ -1753,7 +1792,10 @@ static int add_dirent_to_buf(handle_t *h
 466         }
 467
 468         /* By now the buffer is marked for journaling */
 469 -       ext4_insert_dentry(inode, de, blocksize, name, namelen);
 470 +       /* If writing the short form of "dotdot", don't add the data section */
 471 +       if (dlen == 1)
 472 +               data = NULL;
 473 +       ext4_insert_dentry(inode, de, blocksize, name, namelen, data);
 474
 475         /*
 476          * XXX shouldn't update any times until successful
 477 @@ -1864,7 +1906,8 @@ static int make_indexed_dir(handle_t *ha
 478
 479         dx_set_block(entries, 1);
 480         dx_set_count(entries, 1);
 481 -       dx_set_limit(entries, dx_root_limit(dir, sizeof(*dx_info)));
 482 +       dx_set_limit(entries, dx_root_limit(dir,
 483 +                                        dot_de, sizeof(*dx_info)));
 484
 485         /* Initialize as for dx_probe */
 486         hinfo.hash_version = dx_info->hash_version;
 487 @@ -1907,6 +1950,8 @@ static int ext4_update_dotdot(handle_t *
 488         struct buffer_head *dir_block;
 489         struct ext4_dir_entry_2 *de;
 490         int len, journal = 0, err = 0;
 491 +       int dlen = 0;
 492 +       char *data;
 493
 494         if (IS_ERR(handle))
 495                 return PTR_ERR(handle);
 496 @@ -1922,19 +1967,24 @@ static int ext4_update_dotdot(handle_t *
 497         /* the first item must be "." */
 498         assert(de->name_len == 1 && de->name[0] == '.');
 499         len = le16_to_cpu(de->rec_len);
 500 -       assert(len >= EXT4_DIR_REC_LEN(1));
 501 -       if (len > EXT4_DIR_REC_LEN(1)) {
 502 +       assert(len >= __EXT4_DIR_REC_LEN(1));
 503 +       if (len > __EXT4_DIR_REC_LEN(1)) {
 504                 BUFFER_TRACE(dir_block, "get_write_access");
 505                 err = ext4_journal_get_write_access(handle, dir_block);
 506                 if (err)
 507                         goto out_journal;
 508
 509                 journal = 1;
 510 -               de->rec_len = cpu_to_le16(EXT4_DIR_REC_LEN(1));
 511 +               de->rec_len = cpu_to_le16(EXT4_DIR_REC_LEN(de));
 512         }
 513
 514 -       len -= EXT4_DIR_REC_LEN(1);
 515 -       assert(len == 0 || len >= EXT4_DIR_REC_LEN(2));
 516 +       len -= EXT4_DIR_REC_LEN(de);
 517 +       data = ext4_dentry_get_data(dir->i_sb,
 518 +                       (struct ext4_dentry_param *)dentry->d_fsdata);
 519 +       if (data)
 520 +               dlen = *data + 1;
 521 +       assert(len == 0 || len >= __EXT4_DIR_REC_LEN(2 + dlen));
 522 +
 523         de = (struct ext4_dir_entry_2 *)
 524                         ((char *) de + le16_to_cpu(de->rec_len));
 525         if (!journal) {
 526 @@ -1948,10 +1998,15 @@ static int ext4_update_dotdot(handle_t *
 527         if (len > 0)
 528                 de->rec_len = cpu_to_le16(len);
 529         else
 530 -               assert(le16_to_cpu(de->rec_len) >= EXT4_DIR_REC_LEN(2));
 531 +               assert(le16_to_cpu(de->rec_len) >= __EXT4_DIR_REC_LEN(2));
 532         de->name_len = 2;
 533         strcpy(de->name, "..");
 534 -       ext4_set_de_type(dir->i_sb, de, S_IFDIR);
 535 +       if (data != NULL && ext4_get_dirent_data_len(de) >= dlen) {
 536 +               de->name[2] = 0;
 537 +               memcpy(&de->name[2 + 1], data, *data);
 538 +               ext4_set_de_type(dir->i_sb, de, S_IFDIR);
 539 +               de->file_type |= EXT4_DIRENT_LUFID;
 540 +       }
 541
 542  out_journal:
 543         if (journal) {
 544 @@ -2463,37 +2518,70 @@ err_unlock_inode:
 545         return err;
 546  }
 547
 548 +struct tp_block {
 549 +       struct inode *inode;
 550 +       void *data1;
 551 +       void *data2;
 552 +};
 553 +
 554  struct ext4_dir_entry_2 *ext4_init_dot_dotdot(struct inode *inode,
 555                           struct ext4_dir_entry_2 *de,
 556                           int blocksize, int csum_size,
 557                           unsigned int parent_ino, int dotdot_real_len)
 558  {
 559 +       void *data1 = NULL, *data2 = NULL;
 560 +       int dot_reclen = 0;
 561 +
 562 +       if (dotdot_real_len == 10) {
 563 +               struct tp_block *tpb = (struct tp_block *)inode;
 564 +               data1 = tpb->data1;
 565 +               data2 = tpb->data2;
 566 +               inode = tpb->inode;
 567 +               dotdot_real_len = 0;
 568 +       }
 569         de->inode = cpu_to_le32(inode->i_ino);
 570         de->name_len = 1;
 571 -       de->rec_len = ext4_rec_len_to_disk(EXT4_DIR_REC_LEN(de->name_len),
 572 -                                          blocksize);
 573         strcpy(de->name, ".");
 574         ext4_set_de_type(inode->i_sb, de, S_IFDIR);
 575
 576 +       /* get packed fid data*/
 577 +       data1 = ext4_dentry_get_data(inode->i_sb,
 578 +                               (struct ext4_dentry_param *) data1);
 579 +       if (data1) {
 580 +               de->name[1] = 0;
 581 +               memcpy(&de->name[2], data1, *(char *) data1);
 582 +               de->file_type |= EXT4_DIRENT_LUFID;
 583 +       }
 584 +       de->rec_len = cpu_to_le16(EXT4_DIR_REC_LEN(de));
 585 +       dot_reclen = cpu_to_le16(de->rec_len);
 586         de = ext4_next_entry(de, blocksize);
 587         de->inode = cpu_to_le32(parent_ino);
 588         de->name_len = 2;
 589 +       strcpy(de->name, "..");
 590 +       ext4_set_de_type(inode->i_sb, de, S_IFDIR);
 591 +       data2 = ext4_dentry_get_data(inode->i_sb,
 592 +                       (struct ext4_dentry_param *) data2);
 593 +       if (data2) {
 594 +               de->name[2] = 0;
 595 +               memcpy(&de->name[3], data2, *(char *) data2);
 596 +               de->file_type |= EXT4_DIRENT_LUFID;
 597 +       }
 598         if (!dotdot_real_len)
 599                 de->rec_len = ext4_rec_len_to_disk(blocksize -
 600 -                                       (csum_size + EXT4_DIR_REC_LEN(1)),
 601 +                                       (csum_size + dot_reclen),
 602                                         blocksize);
 603         else
 604                 de->rec_len = ext4_rec_len_to_disk(
 605 -                               EXT4_DIR_REC_LEN(de->name_len), blocksize);
 606 -       strcpy(de->name, "..");
 607 -       ext4_set_de_type(inode->i_sb, de, S_IFDIR);
 608 +                               EXT4_DIR_REC_LEN(de), blocksize);
 609
 610         return ext4_next_entry(de, blocksize);
 611  }
 612
 613  static int ext4_init_new_dir(handle_t *handle, struct inode *dir,
 614 -                            struct inode *inode)
 615 +                            struct inode *inode,
 616 +                            const void *data1, const void *data2)
 617  {
 618 +       struct tp_block param;
 619         struct buffer_head *dir_block = NULL;
 620         struct ext4_dir_entry_2 *de;
 621         struct ext4_dir_entry_tail *t;
 622 @@ -2522,7 +2610,11 @@ static int ext4_init_new_dir(handle_t *h
 623         if (err)
 624                 goto out;
 625         de = (struct ext4_dir_entry_2 *)dir_block->b_data;
 626 -       ext4_init_dot_dotdot(inode, de, blocksize, csum_size, dir->i_ino, 0);
 627 +       param.inode = inode;
 628 +       param.data1 = (void *)data1;
 629 +       param.data2 = (void *)data2;
 630 +       ext4_init_dot_dotdot((struct inode *)(&param), de, blocksize,
 631 +                            csum_size, dir->i_ino, 10);
 632         set_nlink(inode, 2);
 633         if (csum_size) {
 634                 t = EXT4_DIRENT_TAIL(dir_block->b_data, blocksize);
 635 @@ -2539,6 +2631,29 @@ out:
 636         return err;
 637  }
 638
 639 +/* Initialize @inode as a subdirectory of @dir, and add the
 640 + * "." and ".." entries into the first directory block. */
 641 +int ext4_add_dot_dotdot(handle_t *handle, struct inode *dir,
 642 +                       struct inode *inode,
 643 +                       const void *data1, const void *data2)
 644 +{
 645 +       int rc;
 646 +
 647 +       if (IS_ERR(handle))
 648 +               return PTR_ERR(handle);
 649 +
 650 +       if (IS_DIRSYNC(dir))
 651 +               ext4_handle_sync(handle);
 652 +
 653 +       inode->i_op = &ext4_dir_inode_operations;
 654 +       inode->i_fop = &ext4_dir_operations;
 655 +       rc = ext4_init_new_dir(handle, dir, inode, data1, data2);
 656 +       if (!rc)
 657 +               rc = ext4_mark_inode_dirty(handle, inode);
 658 +       return rc;
 659 +}
 660 +EXPORT_SYMBOL(ext4_add_dot_dotdot);
 661 +
 662  static int ext4_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
 663  {
 664         handle_t *handle;
 665 @@ -2563,7 +2678,7 @@ retry:
 666
 667         inode->i_op = &ext4_dir_inode_operations;
 668         inode->i_fop = &ext4_dir_operations;
 669 -       err = ext4_init_new_dir(handle, dir, inode);
 670 +       err = ext4_init_new_dir(handle, dir, inode, NULL, NULL);
 671         if (err)
 672                 goto out_clear_inode;
 673         err = ext4_mark_inode_dirty(handle, inode);
 674 @@ -2615,7 +2730,7 @@ static int empty_dir(struct inode *inode
 675         }
 676
 677         sb = inode->i_sb;
 678 -       if (inode->i_size < EXT4_DIR_REC_LEN(1) + EXT4_DIR_REC_LEN(2)) {
 679 +       if (inode->i_size < __EXT4_DIR_REC_LEN(1) + __EXT4_DIR_REC_LEN(2)) {
 680                 EXT4_ERROR_INODE(inode, "invalid size");
 681                 return 1;
 682         }
 683 Index: linux-stage/fs/ext4/inline.c
 684 ===================================================================
 685 --- linux-stage.orig/fs/ext4/inline.c
 686 +++ linux-stage/fs/ext4/inline.c
 687 @@ -998,17 +998,17 @@ static int ext4_add_dirent_to_inline(han
 688         int             err;
 689         struct ext4_dir_entry_2 *de;
 690
 691 -       reclen = EXT4_DIR_REC_LEN(namelen);
 692 +       reclen = __EXT4_DIR_REC_LEN(namelen);
 693         err = ext4_find_dest_de(dir, inode, iloc->bh,
 694                                 inline_start, inline_size,
 695 -                               name, namelen, &de);
 696 +                               name, namelen, &de, NULL);
 697         if (err)
 698                 return err;
 699
 700         err = ext4_journal_get_write_access(handle, iloc->bh);
 701         if (err)
 702                 return err;
 703 -       ext4_insert_dentry(inode, de, inline_size, name, namelen);
 704 +       ext4_insert_dentry(inode, de, inline_size, name, namelen, NULL);
 705
 706         ext4_show_inline_dir(dir, iloc->bh, inline_start, inline_size);
 707
 708 @@ -1078,7 +1078,7 @@ static int ext4_update_inline_dir(handle
 709         int old_size = EXT4_I(dir)->i_inline_size - EXT4_MIN_INLINE_DATA_SIZE;
 710         int new_size = get_max_inline_xattr_value_size(dir, iloc);
 711
 712 -       if (new_size - old_size <= EXT4_DIR_REC_LEN(1))
 713 +       if (new_size - old_size <= __EXT4_DIR_REC_LEN(1))
 714                 return -ENOSPC;
 715
 716         ret = ext4_update_inline_data(handle, dir,
 717 @@ -1347,7 +1347,7 @@ int htree_inlinedir_to_tree(struct file
 718                         fake.name_len = 1;
 719                         strcpy(fake.name, ".");
 720                         fake.rec_len = ext4_rec_len_to_disk(
 721 -                                               EXT4_DIR_REC_LEN(fake.name_len),
 722 +                                               EXT4_DIR_REC_LEN(&fake),
 723                                                 inline_size);
 724                         ext4_set_de_type(inode->i_sb, &fake, S_IFDIR);
 725                         de = &fake;
 726 @@ -1357,7 +1357,7 @@ int htree_inlinedir_to_tree(struct file
 727                         fake.name_len = 2;
 728                         strcpy(fake.name, "..");
 729                         fake.rec_len = ext4_rec_len_to_disk(
 730 -                                               EXT4_DIR_REC_LEN(fake.name_len),
 731 +                                               EXT4_DIR_REC_LEN(&fake),
 732                                                 inline_size);
 733                         ext4_set_de_type(inode->i_sb, &fake, S_IFDIR);
 734                         de = &fake;
 735 @@ -1452,8 +1452,8 @@ int ext4_read_inline_dir(struct file *fi
 736          * So we will use extra_offset and extra_size to indicate them
 737          * during the inline dir iteration.
 738          */
 739 -       dotdot_offset = EXT4_DIR_REC_LEN(1);
 740 -       dotdot_size = dotdot_offset + EXT4_DIR_REC_LEN(2);
 741 +       dotdot_offset = __EXT4_DIR_REC_LEN(1);
 742 +       dotdot_size = dotdot_offset + __EXT4_DIR_REC_LEN(2);
 743         extra_offset = dotdot_size - EXT4_INLINE_DOTDOT_SIZE;
 744         extra_size = extra_offset + inline_size;
 745
 746 @@ -1488,7 +1488,7 @@ int ext4_read_inline_dir(struct file *fi
 747                          * failure will be detected in the
 748                          * dirent test below. */
 749                         if (ext4_rec_len_from_disk(de->rec_len, extra_size)
 750 -                               < EXT4_DIR_REC_LEN(1))
 751 +                               < __EXT4_DIR_REC_LEN(1))
 752                                 break;
 753                         i += ext4_rec_len_from_disk(de->rec_len,
 754                                                     extra_size);
 755 Index: linux-stage/fs/ext4/super.c
 756 ===================================================================
 757 --- linux-stage.orig/fs/ext4/super.c
 758 +++ linux-stage/fs/ext4/super.c
 759 @@ -1133,7 +1133,7 @@ enum {
 760         Opt_data_err_abort, Opt_data_err_ignore,
 761         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
 762         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_jqfmt_vfsv1, Opt_quota,
 763 -       Opt_noquota, Opt_barrier, Opt_nobarrier, Opt_err,
 764 +       Opt_noquota, Opt_barrier, Opt_nobarrier, Opt_err, Opt_dirdata,
 765         Opt_usrquota, Opt_grpquota, Opt_i_version,
 766         Opt_stripe, Opt_delalloc, Opt_nodelalloc, Opt_mblk_io_submit,
 767         Opt_nomblk_io_submit, Opt_block_validity, Opt_noblock_validity,
 768 @@ -1199,6 +1199,7 @@ static const match_table_t tokens = {
 769         {Opt_stripe, "stripe=%u"},
 770         {Opt_delalloc, "delalloc"},
 771         {Opt_nodelalloc, "nodelalloc"},
 772 +       {Opt_dirdata, "dirdata"},
 773         {Opt_removed, "mblk_io_submit"},
 774         {Opt_removed, "nomblk_io_submit"},
 775         {Opt_block_validity, "block_validity"},
 776 @@ -1409,6 +1410,7 @@ static const struct mount_opts {
 777         {Opt_usrjquota, 0, MOPT_Q},
 778         {Opt_grpjquota, 0, MOPT_Q},
 779         {Opt_offusrjquota, 0, MOPT_Q},
 780 +       {Opt_dirdata, EXT4_MOUNT_DIRDATA, MOPT_SET},
 781         {Opt_offgrpjquota, 0, MOPT_Q},
 782         {Opt_jqfmt_vfsold, QFMT_VFS_OLD, MOPT_QFMT},
 783         {Opt_jqfmt_vfsv0, QFMT_VFS_V0, MOPT_QFMT},