ldiskfs/kernel_patches/patches/rhel8.4/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 diff --git a/fs/ext4/dir.c b/fs/ext4/dir.c
  11 index 9fdd2b2..2a73a69 100644
  12 --- a/fs/ext4/dir.c
  13 +++ b/fs/ext4/dir.c
  14 @@ -77,7 +77,7 @@ int __ext4_check_dir_entry(const char *function, unsigned int line,
  15                 error_msg = "rec_len is smaller than minimal";
  16         else if (unlikely(rlen % 4 != 0))
  17                 error_msg = "rec_len % 4 != 0";
  18 -       else if (unlikely(rlen < EXT4_DIR_REC_LEN(de->name_len)))
  19 +       else if (unlikely(rlen < EXT4_DIR_ENTRY_LEN(de)))
  20                 error_msg = "rec_len is too small for name_len";
  21         else if (unlikely(((char *) de - buf) + rlen > size))
  22                 error_msg = "directory entry overrun";
  23 @@ -220,7 +220,7 @@ static int ext4_readdir(struct file *file, struct dir_context *ctx)
  24                                  * failure will be detected in the
  25                                  * dirent test below. */
  26                                 if (ext4_rec_len_from_disk(de->rec_len,
  27 -                                       sb->s_blocksize) < EXT4_DIR_REC_LEN(1))
  28 +                                   sb->s_blocksize) < EXT4_DIR_REC_LEN(1))
  29                                         break;
  30                                 i += ext4_rec_len_from_disk(de->rec_len,
  31                                                             sb->s_blocksize);
  32 @@ -443,12 +443,17 @@ int ext4_htree_store_dirent(struct file *dir_file, __u32 hash,
  33         struct fname *fname, *new_fn;
  34         struct dir_private_info *info;
  35         int len;
  36 +       int extra_data = 0;
  37
  38         info = dir_file->private_data;
  39         p = &info->root.rb_node;
  40
  41         /* Create and allocate the fname structure */
  42 -       len = sizeof(struct fname) + ent_name->len + 1;
  43 +       if (dirent->file_type & EXT4_DIRENT_LUFID)
  44 +               extra_data = ext4_get_dirent_data_len(dirent);
  45 +
  46 +       len = sizeof(struct fname) + ent_name->len + extra_data + 1;
  47 +
  48         new_fn = kzalloc(len, GFP_KERNEL);
  49         if (!new_fn)
  50                 return -ENOMEM;
  51 @@ -457,7 +462,7 @@ int ext4_htree_store_dirent(struct file *dir_file, __u32 hash,
  52         new_fn->inode = le32_to_cpu(dirent->inode);
  53         new_fn->name_len = ent_name->len;
  54         new_fn->file_type = dirent->file_type;
  55 -       memcpy(new_fn->name, ent_name->name, ent_name->len);
  56 +       memcpy(new_fn->name, ent_name->name, ent_name->len + extra_data);
  57         new_fn->name[ent_name->len] = 0;
  58
  59         while (*p) {
  60 diff --git a/fs/ext4/ext4.h b/fs/ext4/ext4.h
  61 index cb649f0..fe35251 100644
  62 --- a/fs/ext4/ext4.h
  63 +++ b/fs/ext4/ext4.h
  64 @@ -1094,6 +1094,7 @@ struct ext4_inode_info {
  65         __u32 i_csum_seed;
  66
  67         kprojid_t i_projid;
  68 +       void *i_dirdata;
  69  };
  70
  71  /*
  72 @@ -1137,6 +1138,7 @@ struct ext4_inode_info {
  73  #define EXT4_MOUNT_POSIX_ACL           0x08000 /* POSIX Access Control Lists */
  74  #define EXT4_MOUNT_NO_AUTO_DA_ALLOC    0x10000 /* No auto delalloc mapping */
  75  #define EXT4_MOUNT_BARRIER             0x20000 /* Use block barriers */
  76 +#define EXT4_MOUNT_DIRDATA             0x60000 /* Data in directory entries*/
  77  #define EXT4_MOUNT_QUOTA               0x40000 /* Some quota option set */
  78  #define EXT4_MOUNT_USRQUOTA            0x80000 /* "old" user quota,
  79                                                  * enable enforcement for hidden
  80 @@ -1843,6 +1845,7 @@ EXT4_FEATURE_INCOMPAT_FUNCS(casefold,             CASEFOLD)
  81                                          EXT4_FEATURE_INCOMPAT_FLEX_BG| \
  82                                          EXT4_FEATURE_INCOMPAT_EA_INODE| \
  83                                          EXT4_FEATURE_INCOMPAT_MMP | \
  84 +                                        EXT4_FEATURE_INCOMPAT_DIRDATA| \
  85                                          EXT4_FEATURE_INCOMPAT_INLINE_DATA | \
  86                                          EXT4_FEATURE_INCOMPAT_ENCRYPT | \
  87                                          EXT4_FEATURE_INCOMPAT_CASEFOLD | \
  88 @@ -2020,6 +2023,43 @@ struct ext4_dir_entry_tail {
  89  #define EXT4_FT_SYMLINK                7
  90
  91  #define EXT4_FT_MAX            8
  92 +#define EXT4_FT_MASK           0xf
  93 +
  94 +#if EXT4_FT_MAX > EXT4_FT_MASK
  95 +#error "conflicting EXT4_FT_MAX and EXT4_FT_MASK"
  96 +#endif
  97 +
  98 +/*
  99 + * d_type has 4 unused bits, so it can hold four types data. these different
 100 + * type of data (e.g. lustre data, high 32 bits of 64-bit inode number) can be
 101 + * stored, in flag order, after file-name in ext4 dirent.
 102 +*/
 103 +/*
 104 + * this flag is added to d_type if ext4 dirent has extra data after
 105 + * filename. this data length is variable and length is stored in first byte
 106 + * of data. data start after filename NUL byte.
 107 + * This is used by Lustre FS.
 108 +  */
 109 +#define EXT4_DIRENT_LUFID              0x10
 110 +
 111 +#define EXT4_LUFID_MAGIC    0xAD200907UL
 112 +struct ext4_dentry_param {
 113 +       __u32  edp_magic;       /* EXT4_LUFID_MAGIC */
 114 +       char   edp_len;         /* size of edp_data in bytes */
 115 +       char   edp_data[0];     /* packed array of data */
 116 +} __packed;
 117 +
 118 +static inline unsigned char *ext4_dentry_get_data(struct super_block *sb,
 119 +                                                 struct ext4_dentry_param *p)
 120 +
 121 +{
 122 +       if (!ext4_has_feature_dirdata(sb))
 123 +               return NULL;
 124 +       if (p && p->edp_magic == EXT4_LUFID_MAGIC)
 125 +               return &p->edp_len;
 126 +       else
 127 +               return NULL;
 128 +}
 129
 130  #define EXT4_FT_DIR_CSUM       0xDE
 131
 132 @@ -2030,8 +2070,16 @@ struct ext4_dir_entry_tail {
 133   */
 134  #define EXT4_DIR_PAD                   4
 135  #define EXT4_DIR_ROUND                 (EXT4_DIR_PAD - 1)
 136 -#define EXT4_DIR_REC_LEN(name_len)     (((name_len) + 8 + EXT4_DIR_ROUND) & \
 137 +#define EXT4_DIR_REC_LEN_(name_len)    (((name_len) + 8 + EXT4_DIR_ROUND) & \
 138                                          ~EXT4_DIR_ROUND)
 139 +#define EXT4_DIR_ENTRY_LEN_(de)                (EXT4_DIR_REC_LEN_((de)->name_len +\
 140 +                                       ext4_get_dirent_data_len(de)))
 141 +/* ldiskfs */
 142 +#define EXT4_DIR_REC_LEN(name_len)     EXT4_DIR_REC_LEN_((name_len))
 143 +#define EXT4_DIR_ENTRY_LEN(de)         EXT4_DIR_ENTRY_LEN_((de))
 144 +/* lustre osd_handler compat */
 145 +#define __EXT4_DIR_REC_LEN(name_len)   EXT4_DIR_REC_LEN_((name_len))
 146 +
 147  #define EXT4_MAX_REC_LEN               ((1<<16)-1)
 148
 149  /*
 150 @@ -2491,11 +2539,11 @@ extern int ext4_find_dest_de(struct inode *dir, struct inode *inode,
 151                              struct buffer_head *bh,
 152                              void *buf, int buf_size,
 153                              struct ext4_filename *fname,
 154 -                            struct ext4_dir_entry_2 **dest_de);
 155 +                            struct ext4_dir_entry_2 **dest_de, int *dlen);
 156  void ext4_insert_dentry(struct inode *inode,
 157                         struct ext4_dir_entry_2 *de,
 158                         int buf_size,
 159 -                       struct ext4_filename *fname);
 160 +                       struct ext4_filename *fname, void *data);
 161  static inline void ext4_update_dx_flag(struct inode *inode)
 162  {
 163         if (!ext4_has_feature_dir_index(inode->i_sb))
 164 @@ -2507,10 +2555,17 @@ static const unsigned char ext4_filetype_table[] = {
 165
 166  static inline  unsigned char get_dtype(struct super_block *sb, int filetype)
 167  {
 168 -       if (!ext4_has_feature_filetype(sb) || filetype >= EXT4_FT_MAX)
 169 +       int fl_index = filetype & EXT4_FT_MASK;
 170 +
 171 +       if (!ext4_has_feature_filetype(sb) || fl_index >= EXT4_FT_MAX)
 172                 return DT_UNKNOWN;
 173
 174 -       return ext4_filetype_table[filetype];
 175 +       if (!test_opt(sb, DIRDATA))
 176 +               return ext4_filetype_table[fl_index];
 177 +
 178 +       return (ext4_filetype_table[fl_index]) |
 179 +               (filetype & EXT4_DIRENT_LUFID);
 180 +
 181  }
 182  extern int ext4_check_all_de(struct inode *dir, struct buffer_head *bh,
 183                              void *buf, int buf_size);
 184 @@ -2673,6 +2728,8 @@ extern struct inode *ext4_create_inode(handle_t *handle,
 185  extern int ext4_delete_entry(handle_t *handle, struct inode * dir,
 186                              struct ext4_dir_entry_2 *de_del,
 187                              struct buffer_head *bh);
 188 +extern int ext4_add_dot_dotdot(handle_t *handle, struct inode *dir,
 189 +                              struct inode *inode, const void *, const void *);
 190  extern int ext4_htree_fill_tree(struct file *dir_file, __u32 start_hash,
 191                                 __u32 start_minor_hash, __u32 *next_hash);
 192  extern int ext4_search_dir(struct buffer_head *bh,
 193 @@ -3428,6 +3485,36 @@ static inline int ext4_buffer_uptodate(struct buffer_head *bh)
 194         return buffer_uptodate(bh);
 195  }
 196
 197 +/*
 198 + * Compute the total directory entry data length.
 199 + * This includes the filename and an implicit NUL terminator (always present),
 200 + * and optional extensions.  Each extension has a bit set in the high 4 bits of
 201 + * de->file_type, and the extension length is the first byte in each entry.
 202 + */
 203 +static inline int ext4_get_dirent_data_len(struct ext4_dir_entry_2 *de)
 204 +{
 205 +       char *len = de->name + de->name_len + 1 /* NUL terminator */;
 206 +       int dlen = 0;
 207 +       __u8 extra_data_flags = (de->file_type & ~EXT4_FT_MASK) >> 4;
 208 +       struct ext4_dir_entry_tail *t = (struct ext4_dir_entry_tail *)de;
 209 +
 210 +       if (!t->det_reserved_zero1 &&
 211 +           le16_to_cpu(t->det_rec_len) ==
 212 +               sizeof(struct ext4_dir_entry_tail) &&
 213 +           !t->det_reserved_zero2 &&
 214 +           t->det_reserved_ft == EXT4_FT_DIR_CSUM)
 215 +               return 0;
 216 +
 217 +       while (extra_data_flags) {
 218 +               if (extra_data_flags & 1) {
 219 +                       dlen += *len + (dlen == 0);
 220 +                       len += *len;
 221 +               }
 222 +               extra_data_flags >>= 1;
 223 +       }
 224 +       return dlen;
 225 +}
 226 +
 227  #endif /* __KERNEL__ */
 228
 229  #define EFSBADCRC      EBADMSG         /* Bad CRC detected */
 230 diff --git a/fs/ext4/inline.c b/fs/ext4/inline.c
 231 index 2fec62d..3f35821 100644
 232 --- a/fs/ext4/inline.c
 233 +++ b/fs/ext4/inline.c
 234 @@ -1023,7 +1023,7 @@ static int ext4_add_dirent_to_inline(handle_t *handle,
 235         struct ext4_dir_entry_2 *de;
 236
 237         err = ext4_find_dest_de(dir, inode, iloc->bh, inline_start,
 238 -                               inline_size, fname, &de);
 239 +                               inline_size, fname, &de, NULL);
 240         if (err)
 241                 return err;
 242
 243 @@ -1031,7 +1031,7 @@ static int ext4_add_dirent_to_inline(handle_t *handle,
 244         err = ext4_journal_get_write_access(handle, iloc->bh);
 245         if (err)
 246                 return err;
 247 -       ext4_insert_dentry(inode, de, inline_size, fname);
 248 +       ext4_insert_dentry(inode, de, inline_size, fname, NULL);
 249
 250         ext4_show_inline_dir(dir, iloc->bh, inline_start, inline_size);
 251
 252 @@ -1378,7 +1378,7 @@ int ext4_inlinedir_to_tree(struct file *dir_file,
 253                         fake.name_len = 1;
 254                         strcpy(fake.name, ".");
 255                         fake.rec_len = ext4_rec_len_to_disk(
 256 -                                               EXT4_DIR_REC_LEN(fake.name_len),
 257 +                                               EXT4_DIR_ENTRY_LEN(&fake),
 258                                                 inline_size);
 259                         ext4_set_de_type(inode->i_sb, &fake, S_IFDIR);
 260                         de = &fake;
 261 @@ -1388,7 +1388,7 @@ int ext4_inlinedir_to_tree(struct file *dir_file,
 262                         fake.name_len = 2;
 263                         strcpy(fake.name, "..");
 264                         fake.rec_len = ext4_rec_len_to_disk(
 265 -                                               EXT4_DIR_REC_LEN(fake.name_len),
 266 +                                               EXT4_DIR_ENTRY_LEN(&fake),
 267                                                 inline_size);
 268                         ext4_set_de_type(inode->i_sb, &fake, S_IFDIR);
 269                         de = &fake;
 270 diff --git a/fs/ext4/namei.c b/fs/ext4/namei.c
 271 index 8713671..23bd871 100644
 272 --- a/fs/ext4/namei.c
 273 +++ b/fs/ext4/namei.c
 274 @@ -258,7 +258,8 @@ static unsigned dx_get_count(struct dx_entry *entries);
 275  static unsigned dx_get_limit(struct dx_entry *entries);
 276  static void dx_set_count(struct dx_entry *entries, unsigned value);
 277  static void dx_set_limit(struct dx_entry *entries, unsigned value);
 278 -static unsigned dx_root_limit(struct inode *dir, unsigned infosize);
 279 +static inline unsigned dx_root_limit(struct inode *dir,
 280 +               struct ext4_dir_entry_2 *dot_de, unsigned infosize);
 281  static unsigned dx_node_limit(struct inode *dir);
 282  static struct dx_frame *dx_probe(struct ext4_filename *fname,
 283                                  struct inode *dir,
 284 @@ -402,22 +403,23 @@ static struct dx_countlimit *get_dx_countlimit(struct inode *inode,
 285  {
 286         struct ext4_dir_entry *dp;
 287         struct dx_root_info *root;
 288 -       int count_offset;
 289 +       int count_offset, dot_rec_len, dotdot_rec_len;
 290
 291         if (le16_to_cpu(dirent->rec_len) == EXT4_BLOCK_SIZE(inode->i_sb))
 292                 count_offset = 8;
 293 -       else if (le16_to_cpu(dirent->rec_len) == 12) {
 294 -               dp = (struct ext4_dir_entry *)(((void *)dirent) + 12);
 295 +       else {
 296 +               dot_rec_len = le16_to_cpu(dirent->rec_len);
 297 +               dp = (struct ext4_dir_entry *)(((void *)dirent) + dot_rec_len);
 298                 if (le16_to_cpu(dp->rec_len) !=
 299 -                   EXT4_BLOCK_SIZE(inode->i_sb) - 12)
 300 +                   EXT4_BLOCK_SIZE(inode->i_sb) - dot_rec_len)
 301                         return NULL;
 302 -               root = (struct dx_root_info *)(((void *)dp + 12));
 303 +               dotdot_rec_len = EXT4_DIR_ENTRY_LEN((struct ext4_dir_entry_2 *)dp);
 304 +               root = (struct dx_root_info *)(((void *)dp + dotdot_rec_len));
 305                 if (root->reserved_zero ||
 306                     root->info_length != sizeof(struct dx_root_info))
 307                         return NULL;
 308 -               count_offset = 32;
 309 -       } else
 310 -               return NULL;
 311 +               count_offset = 8 + dot_rec_len + dotdot_rec_len;
 312 +       }
 313
 314         if (offset)
 315                 *offset = count_offset;
 316 @@ -522,11 +524,12 @@ ext4_next_entry(struct ext4_dir_entry_2 *p, unsigned long blocksize)
 317   */
 318  struct dx_root_info *dx_get_dx_info(struct ext4_dir_entry_2 *de)
 319  {
 320 +       BUG_ON(de->name_len != 1);
 321         /* get dotdot first */
 322 -       de = (struct ext4_dir_entry_2 *)((char *)de + EXT4_DIR_REC_LEN(1));
 323 +       de = (struct ext4_dir_entry_2 *)((char *)de + EXT4_DIR_ENTRY_LEN(de));
 324
 325         /* dx root info is after dotdot entry */
 326 -       de = (struct ext4_dir_entry_2 *)((char *)de + EXT4_DIR_REC_LEN(2));
 327 +       de = (struct ext4_dir_entry_2 *)((char *)de + EXT4_DIR_ENTRY_LEN(de));
 328
 329         return (struct dx_root_info *)de;
 330  }
 331 @@ -571,10 +574,16 @@ static inline void dx_set_limit(struct dx_entry *entries, unsigned value)
 332         ((struct dx_countlimit *) entries)->limit = cpu_to_le16(value);
 333  }
 334
 335 -static inline unsigned dx_root_limit(struct inode *dir, unsigned infosize)
 336 +static inline unsigned dx_root_limit(struct inode *dir,
 337 +               struct ext4_dir_entry_2 *dot_de, unsigned infosize)
 338  {
 339 -       unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(1) -
 340 -               EXT4_DIR_REC_LEN(2) - infosize;
 341 +       struct ext4_dir_entry_2 *dotdot_de;
 342 +       unsigned entry_space;
 343 +
 344 +       BUG_ON(dot_de->name_len != 1);
 345 +       dotdot_de = ext4_next_entry(dot_de, dir->i_sb->s_blocksize);
 346 +       entry_space = dir->i_sb->s_blocksize - EXT4_DIR_ENTRY_LEN(dot_de) -
 347 +                        EXT4_DIR_ENTRY_LEN(dotdot_de) - infosize;
 348
 349         if (ext4_has_metadata_csum(dir->i_sb))
 350                 entry_space -= sizeof(struct dx_tail);
 351 @@ -695,7 +704,7 @@ static struct stats dx_show_leaf(struct inode *dir,
 352                                        (unsigned) ((char *) de - base));
 353  #endif
 354                         }
 355 -                       space += EXT4_DIR_REC_LEN(de->name_len);
 356 +                       space += EXT4_DIR_ENTRY_LEN(de);
 357                         names++;
 358                 }
 359                 de = ext4_next_entry(de, size);
 360 @@ -802,11 +811,14 @@ dx_probe(struct ext4_filename *fname, struct inode *dir,
 361
 362         entries = (struct dx_entry *)(((char *)info) + info->info_length);
 363
 364 -       if (dx_get_limit(entries) != dx_root_limit(dir,
 365 -                                                  info->info_length)) {
 366 +       if (dx_get_limit(entries) !=
 367 +           dx_root_limit(dir, (struct ext4_dir_entry_2 *)frame->bh->b_data,
 368 +                         info->info_length)) {
 369                 ext4_warning_inode(dir, "dx entry: limit %u != root limit %u",
 370                                    dx_get_limit(entries),
 371 -                                  dx_root_limit(dir, info->info_length));
 372 +                                  dx_root_limit(dir,
 373 +                                         (struct ext4_dir_entry_2 *)frame->bh->b_data,
 374 +                                         info->info_length));
 375                 goto fail;
 376         }
 377
 378 @@ -1792,7 +1804,7 @@ dx_move_dirents(char *from, char *to, struct dx_map_entry *map, int count,
 379         while (count--) {
 380                 struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *)
 381                                                 (from + (map->offs<<2));
 382 -               rec_len = EXT4_DIR_REC_LEN(de->name_len);
 383 +               rec_len = EXT4_DIR_ENTRY_LEN(de);
 384                 memcpy (to, de, rec_len);
 385                 ((struct ext4_dir_entry_2 *) to)->rec_len =
 386                                 ext4_rec_len_to_disk(rec_len, blocksize);
 387 @@ -1816,7 +1828,7 @@ static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize)
 388         while ((char*)de < base + blocksize) {
 389                 next = ext4_next_entry(de, blocksize);
 390                 if (de->inode && de->name_len) {
 391 -                       rec_len = EXT4_DIR_REC_LEN(de->name_len);
 392 +                       rec_len = EXT4_DIR_ENTRY_LEN(de);
 393                         if (de > to)
 394                                 memmove(to, de, rec_len);
 395                         to->rec_len = ext4_rec_len_to_disk(rec_len, blocksize);
 396 @@ -1943,14 +1955,16 @@ int ext4_find_dest_de(struct inode *dir, struct inode *inode,
 397                       struct buffer_head *bh,
 398                       void *buf, int buf_size,
 399                       struct ext4_filename *fname,
 400 -                     struct ext4_dir_entry_2 **dest_de)
 401 +                     struct ext4_dir_entry_2 **dest_de, int *dlen)
 402  {
 403         struct ext4_dir_entry_2 *de;
 404 -       unsigned short reclen = EXT4_DIR_REC_LEN(fname_len(fname));
 405 +       unsigned short reclen = EXT4_DIR_REC_LEN(fname_len(fname)) +
 406 +                                                 (dlen ? *dlen : 0);
 407         int nlen, rlen;
 408         unsigned int offset = 0;
 409         char *top;
 410
 411 +       dlen ? *dlen = 0 : 0; /* default set to 0 */
 412         de = (struct ext4_dir_entry_2 *)buf;
 413         top = buf + buf_size - reclen;
 414         while ((char *) de <= top) {
 415 @@ -1959,10 +1973,26 @@ int ext4_find_dest_de(struct inode *dir, struct inode *inode,
 416                         return -EFSCORRUPTED;
 417                 if (ext4_match(dir, fname, de))
 418                         return -EEXIST;
 419 -               nlen = EXT4_DIR_REC_LEN(de->name_len);
 420 +               nlen = EXT4_DIR_ENTRY_LEN(de);
 421                 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
 422                 if ((de->inode ? rlen - nlen : rlen) >= reclen)
 423                         break;
 424 +               /* Then for dotdot entries, check for the smaller space
 425 +                * required for just the entry, no FID */
 426 +               if (fname_len(fname) == 2 && memcmp(fname_name(fname), "..", 2) == 0) {
 427 +                       if ((de->inode ? rlen - nlen : rlen) >=
 428 +                           EXT4_DIR_REC_LEN(fname_len(fname))) {
 429 +                               /* set dlen=1 to indicate not
 430 +                                * enough space store fid */
 431 +                               dlen ? *dlen = 1 : 0;
 432 +                               break;
 433 +                       }
 434 +                       /* The new ".." entry must be written over the
 435 +                        * previous ".." entry, which is the first
 436 +                        * entry traversed by this scan. If it doesn't
 437 +                        * fit, something is badly wrong, so -EIO. */
 438 +                       return -EIO;
 439 +               }
 440                 de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
 441                 offset += rlen;
 442         }
 443 @@ -1976,12 +2006,12 @@ int ext4_find_dest_de(struct inode *dir, struct inode *inode,
 444  void ext4_insert_dentry(struct inode *inode,
 445                         struct ext4_dir_entry_2 *de,
 446                         int buf_size,
 447 -                       struct ext4_filename *fname)
 448 +                       struct ext4_filename *fname, void *data)
 449  {
 450
 451         int nlen, rlen;
 452
 453 -       nlen = EXT4_DIR_REC_LEN(de->name_len);
 454 +       nlen = EXT4_DIR_ENTRY_LEN(de);
 455         rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
 456         if (de->inode) {
 457                 struct ext4_dir_entry_2 *de1 =
 458 @@ -1995,6 +2025,11 @@ void ext4_insert_dentry(struct inode *inode,
 459         ext4_set_de_type(inode->i_sb, de, inode->i_mode);
 460         de->name_len = fname_len(fname);
 461         memcpy(de->name, fname_name(fname), fname_len(fname));
 462 +       if (data) {
 463 +               de->name[fname_len(fname)] = 0;
 464 +               memcpy(&de->name[fname_len(fname) + 1], data, *(char *)data);
 465 +               de->file_type |= EXT4_DIRENT_LUFID;
 466 +       }
 467  }
 468
 469  /*
 470 @@ -2012,14 +2047,19 @@ static int add_dirent_to_buf(handle_t *handle, struct ext4_filename *fname,
 471  {
 472         unsigned int    blocksize = dir->i_sb->s_blocksize;
 473         int             csum_size = 0;
 474 -       int             err;
 475 +       int             err, dlen = 0;
 476 +       unsigned char   *data;
 477
 478 +       data = ext4_dentry_get_data(inode->i_sb, (struct ext4_dentry_param *)
 479 +                                               EXT4_I(inode)->i_dirdata);
 480         if (ext4_has_metadata_csum(inode->i_sb))
 481                 csum_size = sizeof(struct ext4_dir_entry_tail);
 482
 483         if (!de) {
 484 +               if (data)
 485 +                       dlen = (*data) + 1;
 486                 err = ext4_find_dest_de(dir, inode, bh, bh->b_data,
 487 -                                       blocksize - csum_size, fname, &de);
 488 +                                       blocksize - csum_size, fname, &de, &dlen);
 489                 if (err)
 490                         return err;
 491         }
 492 @@ -2031,7 +2071,10 @@ static int add_dirent_to_buf(handle_t *handle, struct ext4_filename *fname,
 493         }
 494
 495         /* By now the buffer is marked for journaling */
 496 -       ext4_insert_dentry(inode, de, blocksize, fname);
 497 +       /* If writing the short form of "dotdot", don't add the data section */
 498 +       if (dlen == 1)
 499 +               data = NULL;
 500 +       ext4_insert_dentry(inode, de, blocksize, fname, data);
 501
 502         /*
 503          * XXX shouldn't update any times until successful
 504 @@ -2136,7 +2179,8 @@ static int make_indexed_dir(handle_t *handle, struct ext4_filename *fname,
 505
 506         dx_set_block(entries, 1);
 507         dx_set_count(entries, 1);
 508 -       dx_set_limit(entries, dx_root_limit(dir, sizeof(*dx_info)));
 509 +       dx_set_limit(entries, dx_root_limit(dir,
 510 +                                        dot_de, sizeof(*dx_info)));
 511
 512         /* Initialize as for dx_probe */
 513         fname->hinfo.hash_version = dx_info->hash_version;
 514 @@ -2186,6 +2230,8 @@ static int ext4_update_dotdot(handle_t *handle, struct dentry *dentry,
 515         struct buffer_head *dir_block;
 516         struct ext4_dir_entry_2 *de;
 517         int len, journal = 0, err = 0;
 518 +       int dlen = 0;
 519 +       char *data;
 520
 521         if (IS_ERR(handle))
 522                 return PTR_ERR(handle);
 523 @@ -2211,11 +2257,16 @@ static int ext4_update_dotdot(handle_t *handle, struct dentry *dentry,
 524                         goto out_journal;
 525
 526                 journal = 1;
 527 -               de->rec_len = cpu_to_le16(EXT4_DIR_REC_LEN(1));
 528 +               de->rec_len = cpu_to_le16(EXT4_DIR_ENTRY_LEN(de));
 529         }
 530
 531 -       len -= EXT4_DIR_REC_LEN(1);
 532 -       assert(len == 0 || len >= EXT4_DIR_REC_LEN(2));
 533 +       len -= EXT4_DIR_ENTRY_LEN(de);
 534 +       data = ext4_dentry_get_data(dir->i_sb,
 535 +                       (struct ext4_dentry_param *)dentry->d_fsdata);
 536 +       if (data)
 537 +               dlen = *data + 1;
 538 +       assert(len == 0 || len >= EXT4_DIR_REC_LEN(2 + dlen));
 539 +
 540         de = (struct ext4_dir_entry_2 *)
 541                         ((char *) de + le16_to_cpu(de->rec_len));
 542         if (!journal) {
 543 @@ -2232,7 +2283,12 @@ static int ext4_update_dotdot(handle_t *handle, struct dentry *dentry,
 544                 assert(le16_to_cpu(de->rec_len) >= EXT4_DIR_REC_LEN(2));
 545         de->name_len = 2;
 546         strcpy(de->name, "..");
 547 -       ext4_set_de_type(dir->i_sb, de, S_IFDIR);
 548 +       if (data != NULL && ext4_get_dirent_data_len(de) >= dlen) {
 549 +               de->name[2] = 0;
 550 +               memcpy(&de->name[2 + 1], data, *data);
 551 +               ext4_set_de_type(dir->i_sb, de, S_IFDIR);
 552 +               de->file_type |= EXT4_DIRENT_LUFID;
 553 +       }
 554
 555  out_journal:
 556         if (journal) {
 557 @@ -2271,6 +2327,7 @@ static int ext4_add_entry(handle_t *handle, struct dentry *dentry,
 558         ext4_lblk_t block, blocks;
 559         int     csum_size = 0;
 560
 561 +       EXT4_I(inode)->i_dirdata = dentry->d_fsdata;
 562         if (ext4_has_metadata_csum(inode->i_sb))
 563                 csum_size = sizeof(struct ext4_dir_entry_tail);
 564
 565 @@ -2803,37 +2860,70 @@ err_unlock_inode:
 566         return err;
 567  }
 568
 569 +struct tp_block {
 570 +       struct inode *inode;
 571 +       void *data1;
 572 +       void *data2;
 573 +};
 574 +
 575  struct ext4_dir_entry_2 *ext4_init_dot_dotdot(struct inode *inode,
 576                           struct ext4_dir_entry_2 *de,
 577                           int blocksize, int csum_size,
 578                           unsigned int parent_ino, int dotdot_real_len)
 579  {
 580 +       void *data1 = NULL, *data2 = NULL;
 581 +       int dot_reclen = 0;
 582 +
 583 +       if (dotdot_real_len == 10) {
 584 +               struct tp_block *tpb = (struct tp_block *)inode;
 585 +               data1 = tpb->data1;
 586 +               data2 = tpb->data2;
 587 +               inode = tpb->inode;
 588 +               dotdot_real_len = 0;
 589 +       }
 590         de->inode = cpu_to_le32(inode->i_ino);
 591         de->name_len = 1;
 592 -       de->rec_len = ext4_rec_len_to_disk(EXT4_DIR_REC_LEN(de->name_len),
 593 -                                          blocksize);
 594         strcpy(de->name, ".");
 595         ext4_set_de_type(inode->i_sb, de, S_IFDIR);
 596
 597 +       /* get packed fid data*/
 598 +       data1 = ext4_dentry_get_data(inode->i_sb,
 599 +                               (struct ext4_dentry_param *) data1);
 600 +       if (data1) {
 601 +               de->name[1] = 0;
 602 +               memcpy(&de->name[2], data1, *(char *) data1);
 603 +               de->file_type |= EXT4_DIRENT_LUFID;
 604 +       }
 605 +       de->rec_len = cpu_to_le16(EXT4_DIR_ENTRY_LEN(de));
 606 +       dot_reclen = cpu_to_le16(de->rec_len);
 607         de = ext4_next_entry(de, blocksize);
 608         de->inode = cpu_to_le32(parent_ino);
 609         de->name_len = 2;
 610 +       strcpy(de->name, "..");
 611 +       ext4_set_de_type(inode->i_sb, de, S_IFDIR);
 612 +       data2 = ext4_dentry_get_data(inode->i_sb,
 613 +                       (struct ext4_dentry_param *) data2);
 614 +       if (data2) {
 615 +               de->name[2] = 0;
 616 +               memcpy(&de->name[3], data2, *(char *) data2);
 617 +               de->file_type |= EXT4_DIRENT_LUFID;
 618 +       }
 619         if (!dotdot_real_len)
 620                 de->rec_len = ext4_rec_len_to_disk(blocksize -
 621 -                                       (csum_size + EXT4_DIR_REC_LEN(1)),
 622 +                                       (csum_size + dot_reclen),
 623                                         blocksize);
 624         else
 625                 de->rec_len = ext4_rec_len_to_disk(
 626 -                               EXT4_DIR_REC_LEN(de->name_len), blocksize);
 627 -       strcpy(de->name, "..");
 628 -       ext4_set_de_type(inode->i_sb, de, S_IFDIR);
 629 +                               EXT4_DIR_ENTRY_LEN(de), blocksize);
 630
 631         return ext4_next_entry(de, blocksize);
 632  }
 633
 634  static int ext4_init_new_dir(handle_t *handle, struct inode *dir,
 635 -                            struct inode *inode)
 636 +                            struct inode *inode,
 637 +                            const void *data1, const void *data2)
 638  {
 639 +       struct tp_block param;
 640         struct buffer_head *dir_block = NULL;
 641         struct ext4_dir_entry_2 *de;
 642         ext4_lblk_t block = 0;
 643 @@ -2857,7 +2947,11 @@ static int ext4_init_new_dir(handle_t *handle, struct inode *dir,
 644         if (IS_ERR(dir_block))
 645                 return PTR_ERR(dir_block);
 646         de = (struct ext4_dir_entry_2 *)dir_block->b_data;
 647 -       ext4_init_dot_dotdot(inode, de, blocksize, csum_size, dir->i_ino, 0);
 648 +       param.inode = inode;
 649 +       param.data1 = (void *)data1;
 650 +       param.data2 = (void *)data2;
 651 +       ext4_init_dot_dotdot((struct inode *)(&param), de, blocksize,
 652 +                            csum_size, dir->i_ino, 10);
 653         set_nlink(inode, 2);
 654         if (csum_size)
 655                 ext4_initialize_dirent_tail(dir_block, blocksize);
 656 @@ -2872,6 +2966,29 @@ out:
 657         return err;
 658  }
 659
 660 +/* Initialize @inode as a subdirectory of @dir, and add the
 661 + * "." and ".." entries into the first directory block. */
 662 +int ext4_add_dot_dotdot(handle_t *handle, struct inode *dir,
 663 +                       struct inode *inode,
 664 +                       const void *data1, const void *data2)
 665 +{
 666 +       int rc;
 667 +
 668 +       if (IS_ERR(handle))
 669 +               return PTR_ERR(handle);
 670 +
 671 +       if (IS_DIRSYNC(dir))
 672 +               ext4_handle_sync(handle);
 673 +
 674 +       inode->i_op = &ext4_dir_inode_operations;
 675 +       inode->i_fop = &ext4_dir_operations;
 676 +       rc = ext4_init_new_dir(handle, dir, inode, data1, data2);
 677 +       if (!rc)
 678 +               rc = ext4_mark_inode_dirty(handle, inode);
 679 +       return rc;
 680 +}
 681 +EXPORT_SYMBOL(ext4_add_dot_dotdot);
 682 +
 683  static int ext4_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
 684  {
 685         handle_t *handle;
 686 @@ -2898,7 +3015,7 @@ retry:
 687
 688         inode->i_op = &ext4_dir_inode_operations;
 689         inode->i_fop = &ext4_dir_operations;
 690 -       err = ext4_init_new_dir(handle, dir, inode);
 691 +       err = ext4_init_new_dir(handle, dir, inode, NULL, NULL);
 692         if (err)
 693                 goto out_clear_inode;
 694         err = ext4_mark_inode_dirty(handle, inode);
 695 diff --git a/fs/ext4/super.c b/fs/ext4/super.c
 696 index d202d6d..0fcc33b 100644
 697 --- a/fs/ext4/super.c
 698 +++ b/fs/ext4/super.c
 699 @@ -1510,7 +1510,7 @@ enum {
 700         Opt_data_err_abort, Opt_data_err_ignore, Opt_test_dummy_encryption,
 701         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
 702         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_jqfmt_vfsv1, Opt_quota,
 703 -       Opt_noquota, Opt_barrier, Opt_nobarrier, Opt_err,
 704 +       Opt_noquota, Opt_barrier, Opt_nobarrier, Opt_err, Opt_dirdata,
 705         Opt_usrquota, Opt_grpquota, Opt_prjquota, Opt_i_version,
 706         Opt_dax, Opt_dax_always, Opt_dax_inode, Opt_dax_never,
 707         Opt_stripe, Opt_delalloc, Opt_nodelalloc, Opt_warn_on_error,
 708 @@ -1590,6 +1590,7 @@ static const match_table_t tokens = {
 709         {Opt_nolazytime, "nolazytime"},
 710         {Opt_debug_want_extra_isize, "debug_want_extra_isize=%u"},
 711         {Opt_nodelalloc, "nodelalloc"},
 712 +       {Opt_dirdata, "dirdata"},
 713         {Opt_removed, "mblk_io_submit"},
 714         {Opt_removed, "nomblk_io_submit"},
 715         {Opt_block_validity, "block_validity"},
 716 @@ -1820,6 +1821,7 @@ static const struct mount_opts {
 717         {Opt_usrjquota, 0, MOPT_Q},
 718         {Opt_grpjquota, 0, MOPT_Q},
 719         {Opt_offusrjquota, 0, MOPT_Q},
 720 +       {Opt_dirdata, EXT4_MOUNT_DIRDATA, MOPT_SET},
 721         {Opt_offgrpjquota, 0, MOPT_Q},
 722         {Opt_jqfmt_vfsold, QFMT_VFS_OLD, MOPT_QFMT},
 723         {Opt_jqfmt_vfsv0, QFMT_VFS_V0, MOPT_QFMT},