ldiskfs/kernel_patches/patches/linux-5.8/ext4-data-in-dirent.patch

   1 commit 2db3b2b33ee796f4ea61316773452d936303ad27
   2 Author:     Pravin Shelar <Pravin.Shelar@sun.com>
   3 AuthorDate: Sun Oct 4 18:13:14 2009 +0000
   4 Subject: ext4: add ext4-data-in-dirent patch
   5
   6 Allows ext4 to store extra data records inside the ext4_dirent
   7 along with the regular directory entry (type, length, filename).
   8 Data is stored in ext4 dirent after filename, with a bit flag in
   9 de->file_type to indicate if any record after de->name is used.
  10 Each in-use record is variable length and must store a 1-byte
  11 length (including the length byte itself) at the start so that it
  12 can be skipped if the record type is unknown/uneeded.  The record
  13 starts after a NUL byte terminator for the filename.  This extra
  14 space is accounted in de->rec_len but not de->name_len.
  15
  16 Flag EXT4_DIRENT_LUFID is used for a 128-bit file identifier.
  17 Make use of dentry->d_fsdata to pass LUFID to ext4, so no changes
  18 in ext4_add_entry() interface are required.
  19
  20 Bugzilla-ID: b=17670
  21 Signed-off-by: Pravin Shelar <Pravin.Shelar@sun.com>
  22 Reviewed-by: Huang Hua <h.huang@sun.com>
  23 Signed-off-by: Andreas Dilger <andreas.dilger@sun.com>
  24 ---
  25  fs/ext4/dir.c    |   13 ++-
  26  fs/ext4/ext4.h   |   97 +++++++++++++++++++++++++-
  27  fs/ext4/inline.c |    8 +-
  28  fs/ext4/namei.c  |  201 +++++++++++++++++++++++++++++++++++++++++++------------
  29  fs/ext4/super.c  |    4 -
  30  5 files changed, 267 insertions(+), 56 deletions(-)
  31
  32 --- a/fs/ext4/dir.c
  33 +++ b/fs/ext4/dir.c
  34 @@ -78,7 +78,7 @@ int __ext4_check_dir_entry(const char *f
  35                 error_msg = "rec_len is smaller than minimal";
  36         else if (unlikely(rlen % 4 != 0))
  37                 error_msg = "rec_len % 4 != 0";
  38 -       else if (unlikely(rlen < EXT4_DIR_REC_LEN(de->name_len)))
  39 +       else if (unlikely(rlen < EXT4_DIR_ENTRY_LEN(de)))
  40                 error_msg = "rec_len is too small for name_len";
  41         else if (unlikely(next_offset > size))
  42                 error_msg = "directory entry overrun";
  43 @@ -226,7 +226,7 @@ static int ext4_readdir(struct file *fil
  44                                  * failure will be detected in the
  45                                  * dirent test below. */
  46                                 if (ext4_rec_len_from_disk(de->rec_len,
  47 -                                       sb->s_blocksize) < EXT4_DIR_REC_LEN(1))
  48 +                                   sb->s_blocksize) < EXT4_DIR_REC_LEN(1))
  49                                         break;
  50                                 i += ext4_rec_len_from_disk(de->rec_len,
  51                                                             sb->s_blocksize);
  52 @@ -449,12 +449,17 @@ int ext4_htree_store_dirent(struct file
  53         struct fname *fname, *new_fn;
  54         struct dir_private_info *info;
  55         int len;
  56 +       int extra_data = 0;
  57
  58         info = dir_file->private_data;
  59         p = &info->root.rb_node;
  60
  61         /* Create and allocate the fname structure */
  62 -       len = sizeof(struct fname) + ent_name->len + 1;
  63 +       if (dirent->file_type & EXT4_DIRENT_LUFID)
  64 +               extra_data = ext4_get_dirent_data_len(dirent);
  65 +
  66 +       len = sizeof(struct fname) + ent_name->len + extra_data + 1;
  67 +
  68         new_fn = kzalloc(len, GFP_KERNEL);
  69         if (!new_fn)
  70                 return -ENOMEM;
  71 @@ -463,7 +468,7 @@ int ext4_htree_store_dirent(struct file
  72         new_fn->inode = le32_to_cpu(dirent->inode);
  73         new_fn->name_len = ent_name->len;
  74         new_fn->file_type = dirent->file_type;
  75 -       memcpy(new_fn->name, ent_name->name, ent_name->len);
  76 +       memcpy(new_fn->name, ent_name->name, ent_name->len + extra_data);
  77
  78         while (*p) {
  79                 parent = *p;
  80 --- a/fs/ext4/ext4.h
  81 +++ b/fs/ext4/ext4.h
  82 @@ -1118,6 +1118,7 @@ struct ext4_inode_info {
  83         __u32 i_csum_seed;
  84
  85         kprojid_t i_projid;
  86 +       void *i_dirdata;
  87  };
  88
  89  /*
  90 @@ -1139,6 +1140,7 @@
  91   * Mount flags set via mount options or defaults
  92   */
  93  #define EXT4_MOUNT_NO_MBCACHE          0x00001 /* Do not use mbcache */
  94 +#define EXT4_MOUNT_DIRDATA             0x00002 /* Data in directory entries */
  95  #define EXT4_MOUNT_GRPID               0x00004 /* Create files with directory's group */
  96  #define EXT4_MOUNT_DEBUG               0x00008 /* Some debugging messages */
  97  #define EXT4_MOUNT_ERRORS_CONT         0x00010 /* Continue on errors */
  98 @@ -1961,6 +1963,7 @@
  99                                          EXT4_FEATURE_INCOMPAT_FLEX_BG| \
 100                                          EXT4_FEATURE_INCOMPAT_EA_INODE| \
 101                                          EXT4_FEATURE_INCOMPAT_MMP | \
 102 +                                        EXT4_FEATURE_INCOMPAT_DIRDATA| \
 103                                          EXT4_FEATURE_INCOMPAT_INLINE_DATA | \
 104                                          EXT4_FEATURE_INCOMPAT_ENCRYPT | \
 105                                          EXT4_FEATURE_INCOMPAT_CASEFOLD | \
 106 @@ -2133,6 +2136,43 @@ struct ext4_dir_entry_tail {
 107  #define EXT4_FT_SYMLINK                7
 108
 109  #define EXT4_FT_MAX            8
 110 +#define EXT4_FT_MASK           0xf
 111 +
 112 +#if EXT4_FT_MAX > EXT4_FT_MASK
 113 +#error "conflicting EXT4_FT_MAX and EXT4_FT_MASK"
 114 +#endif
 115 +
 116 +/*
 117 + * d_type has 4 unused bits, so it can hold four types data. these different
 118 + * type of data (e.g. lustre data, high 32 bits of 64-bit inode number) can be
 119 + * stored, in flag order, after file-name in ext4 dirent.
 120 +*/
 121 +/*
 122 + * this flag is added to d_type if ext4 dirent has extra data after
 123 + * filename. this data length is variable and length is stored in first byte
 124 + * of data. data start after filename NUL byte.
 125 + * This is used by Lustre FS.
 126 +  */
 127 +#define EXT4_DIRENT_LUFID              0x10
 128 +
 129 +#define EXT4_LUFID_MAGIC    0xAD200907UL
 130 +struct ext4_dentry_param {
 131 +       __u32  edp_magic;       /* EXT4_LUFID_MAGIC */
 132 +       char   edp_len;         /* size of edp_data in bytes */
 133 +       char   edp_data[0];     /* packed array of data */
 134 +} __packed;
 135 +
 136 +static inline unsigned char *ext4_dentry_get_data(struct super_block *sb,
 137 +                                                 struct ext4_dentry_param *p)
 138 +
 139 +{
 140 +       if (!ext4_has_feature_dirdata(sb))
 141 +               return NULL;
 142 +       if (p && p->edp_magic == EXT4_LUFID_MAGIC)
 143 +               return &p->edp_len;
 144 +       else
 145 +               return NULL;
 146 +}
 147
 148  #define EXT4_FT_DIR_CSUM       0xDE
 149
 150 @@ -2143,8 +2183,16 @@ struct ext4_dir_entry_tail {
 151   */
 152  #define EXT4_DIR_PAD                   4
 153  #define EXT4_DIR_ROUND                 (EXT4_DIR_PAD - 1)
 154 -#define EXT4_DIR_REC_LEN(name_len)     (((name_len) + 8 + EXT4_DIR_ROUND) & \
 155 +#define EXT4_DIR_REC_LEN_(name_len)    (((name_len) + 8 + EXT4_DIR_ROUND) & \
 156                                          ~EXT4_DIR_ROUND)
 157 +#define EXT4_DIR_ENTRY_LEN_(de)                (EXT4_DIR_REC_LEN_((de)->name_len +\
 158 +                                       ext4_get_dirent_data_len(de)))
 159 +/* ldiskfs */
 160 +#define EXT4_DIR_REC_LEN(name_len)     EXT4_DIR_REC_LEN_((name_len))
 161 +#define EXT4_DIR_ENTRY_LEN(de)         EXT4_DIR_ENTRY_LEN_((de))
 162 +/* lustre osd_handler compat */
 163 +#define __EXT4_DIR_REC_LEN(name_len)   EXT4_DIR_REC_LEN_((name_len))
 164 +
 165  #define EXT4_MAX_REC_LEN               ((1<<16)-1)
 166
 167  /*
 168 @@ -2604,11 +2652,11 @@ extern int ext4_find_dest_de(struct inod
 169                              struct buffer_head *bh,
 170                              void *buf, int buf_size,
 171                              struct ext4_filename *fname,
 172 -                            struct ext4_dir_entry_2 **dest_de);
 173 +                            struct ext4_dir_entry_2 **dest_de, int *dlen);
 174  void ext4_insert_dentry(struct inode *inode,
 175                         struct ext4_dir_entry_2 *de,
 176                         int buf_size,
 177 -                       struct ext4_filename *fname);
 178 +                       struct ext4_filename *fname, void *data);
 179  static inline void ext4_update_dx_flag(struct inode *inode)
 180  {
 181         if (!ext4_has_feature_dir_index(inode->i_sb)) {
 182 @@ -2623,10 +2671,17 @@ static const unsigned char ext4_filetype
 183
 184  static inline  unsigned char get_dtype(struct super_block *sb, int filetype)
 185  {
 186 -       if (!ext4_has_feature_filetype(sb) || filetype >= EXT4_FT_MAX)
 187 +       int fl_index = filetype & EXT4_FT_MASK;
 188 +
 189 +       if (!ext4_has_feature_filetype(sb) || fl_index >= EXT4_FT_MAX)
 190                 return DT_UNKNOWN;
 191
 192 -       return ext4_filetype_table[filetype];
 193 +       if (!test_opt(sb, DIRDATA))
 194 +               return ext4_filetype_table[fl_index];
 195 +
 196 +       return (ext4_filetype_table[fl_index]) |
 197 +               (filetype & EXT4_DIRENT_LUFID);
 198 +
 199  }
 200  extern int ext4_check_all_de(struct inode *dir, struct buffer_head *bh,
 201                              void *buf, int buf_size);
 202 @@ -2785,6 +2840,8 @@ extern struct inode *ext4_create_inode(h
 203  extern int ext4_delete_entry(handle_t *handle, struct inode * dir,
 204                              struct ext4_dir_entry_2 *de_del,
 205                              struct buffer_head *bh);
 206 +extern int ext4_add_dot_dotdot(handle_t *handle, struct inode *dir,
 207 +                              struct inode *inode, const void *, const void *);
 208  extern int ext4_htree_fill_tree(struct file *dir_file, __u32 start_hash,
 209                                 __u32 start_minor_hash, __u32 *next_hash);
 210  extern int ext4_search_dir(struct buffer_head *bh,
 211 @@ -3559,6 +3616,36 @@ static inline int ext4_buffer_uptodate(s
 212         return buffer_uptodate(bh);
 213  }
 214
 215 +/*
 216 + * Compute the total directory entry data length.
 217 + * This includes the filename and an implicit NUL terminator (always present),
 218 + * and optional extensions.  Each extension has a bit set in the high 4 bits of
 219 + * de->file_type, and the extension length is the first byte in each entry.
 220 + */
 221 +static inline int ext4_get_dirent_data_len(struct ext4_dir_entry_2 *de)
 222 +{
 223 +       char *len = de->name + de->name_len + 1 /* NUL terminator */;
 224 +       int dlen = 0;
 225 +       __u8 extra_data_flags = (de->file_type & ~EXT4_FT_MASK) >> 4;
 226 +       struct ext4_dir_entry_tail *t = (struct ext4_dir_entry_tail *)de;
 227 +
 228 +       if (!t->det_reserved_zero1 &&
 229 +           le16_to_cpu(t->det_rec_len) ==
 230 +               sizeof(struct ext4_dir_entry_tail) &&
 231 +           !t->det_reserved_zero2 &&
 232 +           t->det_reserved_ft == EXT4_FT_DIR_CSUM)
 233 +               return 0;
 234 +
 235 +       while (extra_data_flags) {
 236 +               if (extra_data_flags & 1) {
 237 +                       dlen += *len + (dlen == 0);
 238 +                       len += *len;
 239 +               }
 240 +               extra_data_flags >>= 1;
 241 +       }
 242 +       return dlen;
 243 +}
 244 +
 245  #endif /* __KERNEL__ */
 246
 247  #define EFSBADCRC      EBADMSG         /* Bad CRC detected */
 248 --- a/fs/ext4/inline.c
 249 +++ b/fs/ext4/inline.c
 250 @@ -1023,7 +1023,7 @@ static int ext4_add_dirent_to_inline(han
 251         struct ext4_dir_entry_2 *de;
 252
 253         err = ext4_find_dest_de(dir, inode, iloc->bh, inline_start,
 254 -                               inline_size, fname, &de);
 255 +                               inline_size, fname, &de, NULL);
 256         if (err)
 257                 return err;
 258
 259 @@ -1031,7 +1031,7 @@ static int ext4_add_dirent_to_inline(han
 260         err = ext4_journal_get_write_access(handle, iloc->bh);
 261         if (err)
 262                 return err;
 263 -       ext4_insert_dentry(inode, de, inline_size, fname);
 264 +       ext4_insert_dentry(inode, de, inline_size, fname, NULL);
 265
 266         ext4_show_inline_dir(dir, iloc->bh, inline_start, inline_size);
 267
 268 @@ -1380,7 +1380,7 @@ int ext4_inlinedir_to_tree(struct file *
 269                         fake.name_len = 1;
 270                         strcpy(fake.name, ".");
 271                         fake.rec_len = ext4_rec_len_to_disk(
 272 -                                               EXT4_DIR_REC_LEN(fake.name_len),
 273 +                                               EXT4_DIR_ENTRY_LEN(&fake),
 274                                                 inline_size);
 275                         ext4_set_de_type(inode->i_sb, &fake, S_IFDIR);
 276                         de = &fake;
 277 @@ -1390,7 +1390,7 @@ int ext4_inlinedir_to_tree(struct file *
 278                         fake.name_len = 2;
 279                         strcpy(fake.name, "..");
 280                         fake.rec_len = ext4_rec_len_to_disk(
 281 -                                               EXT4_DIR_REC_LEN(fake.name_len),
 282 +                                               EXT4_DIR_ENTRY_LEN(&fake),
 283                                                 inline_size);
 284                         ext4_set_de_type(inode->i_sb, &fake, S_IFDIR);
 285                         de = &fake;
 286 --- a/fs/ext4/namei.c
 287 +++ b/fs/ext4/namei.c
 288 @@ -265,7 +265,8 @@ static unsigned dx_get_count(struct dx_e
 289  static unsigned dx_get_limit(struct dx_entry *entries);
 290  static void dx_set_count(struct dx_entry *entries, unsigned value);
 291  static void dx_set_limit(struct dx_entry *entries, unsigned value);
 292 -static unsigned dx_root_limit(struct inode *dir, unsigned infosize);
 293 +static inline unsigned dx_root_limit(struct inode *dir,
 294 +               struct ext4_dir_entry_2 *dot_de, unsigned infosize);
 295  static unsigned dx_node_limit(struct inode *dir);
 296  static struct dx_frame *dx_probe(struct ext4_filename *fname,
 297                                  struct inode *dir,
 298 @@ -409,22 +410,23 @@ static struct dx_countlimit *get_dx_coun
 299  {
 300         struct ext4_dir_entry *dp;
 301         struct dx_root_info *root;
 302 -       int count_offset;
 303 +       int count_offset, dot_rec_len, dotdot_rec_len;
 304
 305         if (le16_to_cpu(dirent->rec_len) == EXT4_BLOCK_SIZE(inode->i_sb))
 306                 count_offset = 8;
 307 -       else if (le16_to_cpu(dirent->rec_len) == 12) {
 308 -               dp = (struct ext4_dir_entry *)(((void *)dirent) + 12);
 309 +       else {
 310 +               dot_rec_len = le16_to_cpu(dirent->rec_len);
 311 +               dp = (struct ext4_dir_entry *)(((void *)dirent) + dot_rec_len);
 312                 if (le16_to_cpu(dp->rec_len) !=
 313 -                   EXT4_BLOCK_SIZE(inode->i_sb) - 12)
 314 +                   EXT4_BLOCK_SIZE(inode->i_sb) - dot_rec_len)
 315                         return NULL;
 316 -               root = (struct dx_root_info *)(((void *)dp + 12));
 317 +               dotdot_rec_len = EXT4_DIR_ENTRY_LEN((struct ext4_dir_entry_2 *)dp);
 318 +               root = (struct dx_root_info *)(((void *)dp + dotdot_rec_len));
 319                 if (root->reserved_zero ||
 320                     root->info_length != sizeof(struct dx_root_info))
 321                         return NULL;
 322 -               count_offset = 32;
 323 -       } else
 324 -               return NULL;
 325 +               count_offset = 8 + dot_rec_len + dotdot_rec_len;
 326 +       }
 327
 328         if (offset)
 329                 *offset = count_offset;
 330 @@ -529,11 +531,12 @@ ext4_next_entry(struct ext4_dir_entry_2
 331   */
 332  struct dx_root_info *dx_get_dx_info(struct ext4_dir_entry_2 *de)
 333  {
 334 +       BUG_ON(de->name_len != 1);
 335         /* get dotdot first */
 336 -       de = (struct ext4_dir_entry_2 *)((char *)de + EXT4_DIR_REC_LEN(1));
 337 +       de = (struct ext4_dir_entry_2 *)((char *)de + EXT4_DIR_ENTRY_LEN(de));
 338
 339         /* dx root info is after dotdot entry */
 340 -       de = (struct ext4_dir_entry_2 *)((char *)de + EXT4_DIR_REC_LEN(2));
 341 +       de = (struct ext4_dir_entry_2 *)((char *)de + EXT4_DIR_ENTRY_LEN(de));
 342
 343         return (struct dx_root_info *)de;
 344  }
 345 @@ -578,10 +581,16 @@ static inline void dx_set_limit(struct d
 346         ((struct dx_countlimit *) entries)->limit = cpu_to_le16(value);
 347  }
 348
 349 -static inline unsigned dx_root_limit(struct inode *dir, unsigned infosize)
 350 +static inline unsigned dx_root_limit(struct inode *dir,
 351 +               struct ext4_dir_entry_2 *dot_de, unsigned infosize)
 352  {
 353 -       unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(1) -
 354 -               EXT4_DIR_REC_LEN(2) - infosize;
 355 +       struct ext4_dir_entry_2 *dotdot_de;
 356 +       unsigned entry_space;
 357 +
 358 +       BUG_ON(dot_de->name_len != 1);
 359 +       dotdot_de = ext4_next_entry(dot_de, dir->i_sb->s_blocksize);
 360 +       entry_space = dir->i_sb->s_blocksize - EXT4_DIR_ENTRY_LEN(dot_de) -
 361 +                        EXT4_DIR_ENTRY_LEN(dotdot_de) - infosize;
 362
 363         if (ext4_has_metadata_csum(dir->i_sb))
 364                 entry_space -= sizeof(struct dx_tail);
 365 @@ -702,7 +711,7 @@ static struct stats dx_show_leaf(struct
 366                                        (unsigned) ((char *) de - base));
 367  #endif
 368                         }
 369 -                       space += EXT4_DIR_REC_LEN(de->name_len);
 370 +                       space += EXT4_DIR_ENTRY_LEN(de);
 371                         names++;
 372                 }
 373                 de = ext4_next_entry(de, size);
 374 @@ -809,11 +818,14 @@ dx_probe(struct ext4_filename *fname, st
 375
 376         entries = (struct dx_entry *)(((char *)info) + info->info_length);
 377
 378 -       if (dx_get_limit(entries) != dx_root_limit(dir,
 379 -                                                  info->info_length)) {
 380 +       if (dx_get_limit(entries) !=
 381 +           dx_root_limit(dir, (struct ext4_dir_entry_2 *)frame->bh->b_data,
 382 +                         info->info_length)) {
 383                 ext4_warning_inode(dir, "dx entry: limit %u != root limit %u",
 384                                    dx_get_limit(entries),
 385 -                                  dx_root_limit(dir, info->info_length));
 386 +                                  dx_root_limit(dir,
 387 +                                         (struct ext4_dir_entry_2 *)frame->bh->b_data,
 388 +                                         info->info_length));
 389                 goto fail;
 390         }
 391
 392 @@ -1799,7 +1811,7 @@ dx_move_dirents(char *from, char *to, st
 393         while (count--) {
 394                 struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *)
 395                                                 (from + (map->offs<<2));
 396 -               rec_len = EXT4_DIR_REC_LEN(de->name_len);
 397 +               rec_len = EXT4_DIR_ENTRY_LEN(de);
 398                 memcpy (to, de, rec_len);
 399                 ((struct ext4_dir_entry_2 *) to)->rec_len =
 400                                 ext4_rec_len_to_disk(rec_len, blocksize);
 401 @@ -1823,7 +1835,7 @@ static struct ext4_dir_entry_2* dx_pack_
 402         while ((char*)de < base + blocksize) {
 403                 next = ext4_next_entry(de, blocksize);
 404                 if (de->inode && de->name_len) {
 405 -                       rec_len = EXT4_DIR_REC_LEN(de->name_len);
 406 +                       rec_len = EXT4_DIR_ENTRY_LEN(de);
 407                         if (de > to)
 408                                 memmove(to, de, rec_len);
 409                         to->rec_len = ext4_rec_len_to_disk(rec_len, blocksize);
 410 @@ -1950,14 +1962,16 @@ int ext4_find_dest_de(struct inode *dir,
 411                       struct buffer_head *bh,
 412                       void *buf, int buf_size,
 413                       struct ext4_filename *fname,
 414 -                     struct ext4_dir_entry_2 **dest_de)
 415 +                     struct ext4_dir_entry_2 **dest_de, int *dlen)
 416  {
 417         struct ext4_dir_entry_2 *de;
 418 -       unsigned short reclen = EXT4_DIR_REC_LEN(fname_len(fname));
 419 +       unsigned short reclen = EXT4_DIR_REC_LEN(fname_len(fname)) +
 420 +                                                 (dlen ? *dlen : 0);
 421         int nlen, rlen;
 422         unsigned int offset = 0;
 423         char *top;
 424
 425 +       dlen ? *dlen = 0 : 0; /* default set to 0 */
 426         de = (struct ext4_dir_entry_2 *)buf;
 427         top = buf + buf_size - reclen;
 428         while ((char *) de <= top) {
 429 @@ -1966,10 +1980,26 @@ int ext4_find_dest_de(struct inode *dir,
 430                         return -EFSCORRUPTED;
 431                 if (ext4_match(dir, fname, de))
 432                         return -EEXIST;
 433 -               nlen = EXT4_DIR_REC_LEN(de->name_len);
 434 +               nlen = EXT4_DIR_ENTRY_LEN(de);
 435                 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
 436                 if ((de->inode ? rlen - nlen : rlen) >= reclen)
 437                         break;
 438 +               /* Then for dotdot entries, check for the smaller space
 439 +                * required for just the entry, no FID */
 440 +               if (fname_len(fname) == 2 && memcmp(fname_name(fname), "..", 2) == 0) {
 441 +                       if ((de->inode ? rlen - nlen : rlen) >=
 442 +                           EXT4_DIR_REC_LEN(fname_len(fname))) {
 443 +                               /* set dlen=1 to indicate not
 444 +                                * enough space store fid */
 445 +                               dlen ? *dlen = 1 : 0;
 446 +                               break;
 447 +                       }
 448 +                       /* The new ".." entry must be written over the
 449 +                        * previous ".." entry, which is the first
 450 +                        * entry traversed by this scan. If it doesn't
 451 +                        * fit, something is badly wrong, so -EIO. */
 452 +                       return -EIO;
 453 +               }
 454                 de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
 455                 offset += rlen;
 456         }
 457 @@ -1983,12 +2013,12 @@ int ext4_find_dest_de(struct inode *dir,
 458  void ext4_insert_dentry(struct inode *inode,
 459                         struct ext4_dir_entry_2 *de,
 460                         int buf_size,
 461 -                       struct ext4_filename *fname)
 462 +                       struct ext4_filename *fname, void *data)
 463  {
 464
 465         int nlen, rlen;
 466
 467 -       nlen = EXT4_DIR_REC_LEN(de->name_len);
 468 +       nlen = EXT4_DIR_ENTRY_LEN(de);
 469         rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
 470         if (de->inode) {
 471                 struct ext4_dir_entry_2 *de1 =
 472 @@ -2002,6 +2032,11 @@ void ext4_insert_dentry(struct inode *in
 473         ext4_set_de_type(inode->i_sb, de, inode->i_mode);
 474         de->name_len = fname_len(fname);
 475         memcpy(de->name, fname_name(fname), fname_len(fname));
 476 +       if (data) {
 477 +               de->name[fname_len(fname)] = 0;
 478 +               memcpy(&de->name[fname_len(fname) + 1], data, *(char *)data);
 479 +               de->file_type |= EXT4_DIRENT_LUFID;
 480 +       }
 481  }
 482
 483  /*
 484 @@ -2019,14 +2054,19 @@ static int add_dirent_to_buf(handle_t *h
 485  {
 486         unsigned int    blocksize = dir->i_sb->s_blocksize;
 487         int             csum_size = 0;
 488 -       int             err, err2;
 489 +       int             err, err2, dlen = 0;
 490 +       unsigned char   *data;
 491
 492 +       data = ext4_dentry_get_data(inode->i_sb, (struct ext4_dentry_param *)
 493 +                                               EXT4_I(inode)->i_dirdata);
 494         if (ext4_has_metadata_csum(inode->i_sb))
 495                 csum_size = sizeof(struct ext4_dir_entry_tail);
 496
 497         if (!de) {
 498 +               if (data)
 499 +                       dlen = (*data) + 1;
 500                 err = ext4_find_dest_de(dir, inode, bh, bh->b_data,
 501 -                                       blocksize - csum_size, fname, &de);
 502 +                                       blocksize - csum_size, fname, &de, &dlen);
 503                 if (err)
 504                         return err;
 505         }
 506 @@ -2038,7 +2078,10 @@ static int add_dirent_to_buf(handle_t *h
 507         }
 508
 509         /* By now the buffer is marked for journaling */
 510 -       ext4_insert_dentry(inode, de, blocksize, fname);
 511 +       /* If writing the short form of "dotdot", don't add the data section */
 512 +       if (dlen == 1)
 513 +               data = NULL;
 514 +       ext4_insert_dentry(inode, de, blocksize, fname, data);
 515
 516         /*
 517          * XXX shouldn't update any times until successful
 518 @@ -2143,7 +2186,8 @@ static int make_indexed_dir(handle_t *ha
 519
 520         dx_set_block(entries, 1);
 521         dx_set_count(entries, 1);
 522 -       dx_set_limit(entries, dx_root_limit(dir, sizeof(*dx_info)));
 523 +       dx_set_limit(entries, dx_root_limit(dir,
 524 +                                        dot_de, sizeof(*dx_info)));
 525
 526         /* Initialize as for dx_probe */
 527         fname->hinfo.hash_version = dx_info->hash_version;
 528 @@ -2193,6 +2237,8 @@ static int ext4_update_dotdot(handle_t *
 529         struct buffer_head *dir_block;
 530         struct ext4_dir_entry_2 *de;
 531         int len, journal = 0, err = 0;
 532 +       int dlen = 0;
 533 +       char *data;
 534
 535         if (IS_ERR(handle))
 536                 return PTR_ERR(handle);
 537 @@ -2218,11 +2264,16 @@ static int ext4_update_dotdot(handle_t *
 538                         goto out_journal;
 539
 540                 journal = 1;
 541 -               de->rec_len = cpu_to_le16(EXT4_DIR_REC_LEN(1));
 542 +               de->rec_len = cpu_to_le16(EXT4_DIR_ENTRY_LEN(de));
 543         }
 544
 545 -       len -= EXT4_DIR_REC_LEN(1);
 546 -       assert(len == 0 || len >= EXT4_DIR_REC_LEN(2));
 547 +       len -= EXT4_DIR_ENTRY_LEN(de);
 548 +       data = ext4_dentry_get_data(dir->i_sb,
 549 +                       (struct ext4_dentry_param *)dentry->d_fsdata);
 550 +       if (data)
 551 +               dlen = *data + 1;
 552 +       assert(len == 0 || len >= EXT4_DIR_REC_LEN(2 + dlen));
 553 +
 554         de = (struct ext4_dir_entry_2 *)
 555                         ((char *) de + le16_to_cpu(de->rec_len));
 556         if (!journal) {
 557 @@ -2239,7 +2290,12 @@ static int ext4_update_dotdot(handle_t *
 558                 assert(le16_to_cpu(de->rec_len) >= EXT4_DIR_REC_LEN(2));
 559         de->name_len = 2;
 560         strcpy(de->name, "..");
 561 -       ext4_set_de_type(dir->i_sb, de, S_IFDIR);
 562 +       if (data != NULL && ext4_get_dirent_data_len(de) >= dlen) {
 563 +               de->name[2] = 0;
 564 +               memcpy(&de->name[2 + 1], data, *data);
 565 +               ext4_set_de_type(dir->i_sb, de, S_IFDIR);
 566 +               de->file_type |= EXT4_DIRENT_LUFID;
 567 +       }
 568
 569  out_journal:
 570         if (journal) {
 571 @@ -2280,6 +2336,7 @@ static int ext4_add_entry(handle_t *hand
 572         ext4_lblk_t block, blocks;
 573         int     csum_size = 0;
 574
 575 +       EXT4_I(inode)->i_dirdata = dentry->d_fsdata;
 576         if (ext4_has_metadata_csum(inode->i_sb))
 577                 csum_size = sizeof(struct ext4_dir_entry_tail);
 578
 579 @@ -2832,37 +2889,70 @@ err_unlock_inode:
 580         return err;
 581  }
 582
 583 +struct tp_block {
 584 +       struct inode *inode;
 585 +       void *data1;
 586 +       void *data2;
 587 +};
 588 +
 589  struct ext4_dir_entry_2 *ext4_init_dot_dotdot(struct inode *inode,
 590                           struct ext4_dir_entry_2 *de,
 591                           int blocksize, int csum_size,
 592                           unsigned int parent_ino, int dotdot_real_len)
 593  {
 594 +       void *data1 = NULL, *data2 = NULL;
 595 +       int dot_reclen = 0;
 596 +
 597 +       if (dotdot_real_len == 10) {
 598 +               struct tp_block *tpb = (struct tp_block *)inode;
 599 +               data1 = tpb->data1;
 600 +               data2 = tpb->data2;
 601 +               inode = tpb->inode;
 602 +               dotdot_real_len = 0;
 603 +       }
 604         de->inode = cpu_to_le32(inode->i_ino);
 605         de->name_len = 1;
 606 -       de->rec_len = ext4_rec_len_to_disk(EXT4_DIR_REC_LEN(de->name_len),
 607 -                                          blocksize);
 608         strcpy(de->name, ".");
 609         ext4_set_de_type(inode->i_sb, de, S_IFDIR);
 610
 611 +       /* get packed fid data*/
 612 +       data1 = ext4_dentry_get_data(inode->i_sb,
 613 +                               (struct ext4_dentry_param *) data1);
 614 +       if (data1) {
 615 +               de->name[1] = 0;
 616 +               memcpy(&de->name[2], data1, *(char *) data1);
 617 +               de->file_type |= EXT4_DIRENT_LUFID;
 618 +       }
 619 +       de->rec_len = cpu_to_le16(EXT4_DIR_ENTRY_LEN(de));
 620 +       dot_reclen = cpu_to_le16(de->rec_len);
 621         de = ext4_next_entry(de, blocksize);
 622         de->inode = cpu_to_le32(parent_ino);
 623         de->name_len = 2;
 624 +       strcpy(de->name, "..");
 625 +       ext4_set_de_type(inode->i_sb, de, S_IFDIR);
 626 +       data2 = ext4_dentry_get_data(inode->i_sb,
 627 +                       (struct ext4_dentry_param *) data2);
 628 +       if (data2) {
 629 +               de->name[2] = 0;
 630 +               memcpy(&de->name[3], data2, *(char *) data2);
 631 +               de->file_type |= EXT4_DIRENT_LUFID;
 632 +       }
 633         if (!dotdot_real_len)
 634                 de->rec_len = ext4_rec_len_to_disk(blocksize -
 635 -                                       (csum_size + EXT4_DIR_REC_LEN(1)),
 636 +                                       (csum_size + dot_reclen),
 637                                         blocksize);
 638         else
 639                 de->rec_len = ext4_rec_len_to_disk(
 640 -                               EXT4_DIR_REC_LEN(de->name_len), blocksize);
 641 -       strcpy(de->name, "..");
 642 -       ext4_set_de_type(inode->i_sb, de, S_IFDIR);
 643 +                               EXT4_DIR_ENTRY_LEN(de), blocksize);
 644
 645         return ext4_next_entry(de, blocksize);
 646  }
 647
 648  static int ext4_init_new_dir(handle_t *handle, struct inode *dir,
 649 -                            struct inode *inode)
 650 +                            struct inode *inode,
 651 +                            const void *data1, const void *data2)
 652  {
 653 +       struct tp_block param;
 654         struct buffer_head *dir_block = NULL;
 655         struct ext4_dir_entry_2 *de;
 656         ext4_lblk_t block = 0;
 657 @@ -2886,7 +2976,11 @@ static int ext4_init_new_dir(handle_t *h
 658         if (IS_ERR(dir_block))
 659                 return PTR_ERR(dir_block);
 660         de = (struct ext4_dir_entry_2 *)dir_block->b_data;
 661 -       ext4_init_dot_dotdot(inode, de, blocksize, csum_size, dir->i_ino, 0);
 662 +       param.inode = inode;
 663 +       param.data1 = (void *)data1;
 664 +       param.data2 = (void *)data2;
 665 +       ext4_init_dot_dotdot((struct inode *)(&param), de, blocksize,
 666 +                            csum_size, dir->i_ino, 10);
 667         set_nlink(inode, 2);
 668         if (csum_size)
 669                 ext4_initialize_dirent_tail(dir_block, blocksize);
 670 @@ -2901,6 +2995,29 @@ out:
 671         return err;
 672  }
 673
 674 +/* Initialize @inode as a subdirectory of @dir, and add the
 675 + * "." and ".." entries into the first directory block. */
 676 +int ext4_add_dot_dotdot(handle_t *handle, struct inode *dir,
 677 +                       struct inode *inode,
 678 +                       const void *data1, const void *data2)
 679 +{
 680 +       int rc;
 681 +
 682 +       if (IS_ERR(handle))
 683 +               return PTR_ERR(handle);
 684 +
 685 +       if (IS_DIRSYNC(dir))
 686 +               ext4_handle_sync(handle);
 687 +
 688 +       inode->i_op = &ext4_dir_inode_operations;
 689 +       inode->i_fop = &ext4_dir_operations;
 690 +       rc = ext4_init_new_dir(handle, dir, inode, data1, data2);
 691 +       if (!rc)
 692 +               rc = ext4_mark_inode_dirty(handle, inode);
 693 +       return rc;
 694 +}
 695 +EXPORT_SYMBOL(ext4_add_dot_dotdot);
 696 +
 697  static int ext4_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
 698  {
 699         handle_t *handle;
 700 @@ -2927,7 +3044,7 @@ retry:
 701
 702         inode->i_op = &ext4_dir_inode_operations;
 703         inode->i_fop = &ext4_dir_operations;
 704 -       err = ext4_init_new_dir(handle, dir, inode);
 705 +       err = ext4_init_new_dir(handle, dir, inode, NULL, NULL);
 706         if (err)
 707                 goto out_clear_inode;
 708         err = ext4_mark_inode_dirty(handle, inode);
 709 --- a/fs/ext4/super.c
 710 +++ b/fs/ext4/super.c
 711 @@ -1510,7 +1510,7 @@ enum {
 712         Opt_data_err_abort, Opt_data_err_ignore, Opt_test_dummy_encryption,
 713         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
 714         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_jqfmt_vfsv1, Opt_quota,
 715 -       Opt_noquota, Opt_barrier, Opt_nobarrier, Opt_err,
 716 +       Opt_noquota, Opt_barrier, Opt_nobarrier, Opt_err, Opt_dirdata,
 717         Opt_usrquota, Opt_grpquota, Opt_prjquota, Opt_i_version,
 718         Opt_dax, Opt_dax_always, Opt_dax_inode, Opt_dax_never,
 719         Opt_stripe, Opt_delalloc, Opt_nodelalloc, Opt_warn_on_error,
 720 @@ -1590,6 +1590,7 @@ static const match_table_t tokens = {
 721         {Opt_nolazytime, "nolazytime"},
 722         {Opt_debug_want_extra_isize, "debug_want_extra_isize=%u"},
 723         {Opt_nodelalloc, "nodelalloc"},
 724 +       {Opt_dirdata, "dirdata"},
 725         {Opt_removed, "mblk_io_submit"},
 726         {Opt_removed, "nomblk_io_submit"},
 727         {Opt_block_validity, "block_validity"},
 728 @@ -1822,6 +1823,7 @@ static const struct mount_opts {
 729         {Opt_usrjquota, 0, MOPT_Q},
 730         {Opt_grpjquota, 0, MOPT_Q},
 731         {Opt_offusrjquota, 0, MOPT_Q},
 732 +       {Opt_dirdata, EXT4_MOUNT_DIRDATA, MOPT_SET},
 733         {Opt_offgrpjquota, 0, MOPT_Q},
 734         {Opt_jqfmt_vfsold, QFMT_VFS_OLD, MOPT_QFMT},
 735         {Opt_jqfmt_vfsv0, QFMT_VFS_V0, MOPT_QFMT},