ldiskfs/kernel_patches/patches/linux-5.10/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 ---
  11  fs/ext4/dir.c         |   13 ++-
  12  fs/ext4/ext4.h        |  100 +++++++++++++++++++++++-
  13  fs/ext4/fast_commit.c |    2
  14  fs/ext4/inline.c      |    8 -
  15  fs/ext4/namei.c       |  201 +++++++++++++++++++++++++++++++++++++++-----------
  16  fs/ext4/super.c       |    4
  17  6 files changed, 270 insertions(+), 58 deletions(-)
  18
  19 diff -ur a/fs/ext4/dir.c b/fs/ext4/dir.c
  20 --- a/fs/ext4/dir.c     2021-12-14 08:28:35.514373035 -0700
  21 +++ b/fs/ext4/dir.c     2021-12-14 08:29:41.318242202 -0700
  22 @@ -78,7 +78,7 @@
  23                 error_msg = "rec_len is smaller than minimal";
  24         else if (unlikely(rlen % 4 != 0))
  25                 error_msg = "rec_len % 4 != 0";
  26 -       else if (unlikely(rlen < EXT4_DIR_REC_LEN(de->name_len)))
  27 +       else if (unlikely(rlen < EXT4_DIR_ENTRY_LEN(de)))
  28                 error_msg = "rec_len is too small for name_len";
  29         else if (unlikely(next_offset > size))
  30                 error_msg = "directory entry overrun";
  31 @@ -226,7 +226,7 @@
  32                                  * failure will be detected in the
  33                                  * dirent test below. */
  34                                 if (ext4_rec_len_from_disk(de->rec_len,
  35 -                                       sb->s_blocksize) < EXT4_DIR_REC_LEN(1))
  36 +                                   sb->s_blocksize) < EXT4_DIR_REC_LEN(1))
  37                                         break;
  38                                 i += ext4_rec_len_from_disk(de->rec_len,
  39                                                             sb->s_blocksize);
  40 @@ -449,12 +449,17 @@
  41         struct fname *fname, *new_fn;
  42         struct dir_private_info *info;
  43         int len;
  44 +       int extra_data = 0;
  45
  46         info = dir_file->private_data;
  47         p = &info->root.rb_node;
  48
  49         /* Create and allocate the fname structure */
  50 -       len = sizeof(struct fname) + ent_name->len + 1;
  51 +       if (dirent->file_type & EXT4_DIRENT_LUFID)
  52 +               extra_data = ext4_get_dirent_data_len(dirent);
  53 +
  54 +       len = sizeof(struct fname) + ent_name->len + extra_data + 1;
  55 +
  56         new_fn = kzalloc(len, GFP_KERNEL);
  57         if (!new_fn)
  58                 return -ENOMEM;
  59 @@ -463,7 +468,7 @@
  60         new_fn->inode = le32_to_cpu(dirent->inode);
  61         new_fn->name_len = ent_name->len;
  62         new_fn->file_type = dirent->file_type;
  63 -       memcpy(new_fn->name, ent_name->name, ent_name->len);
  64 +       memcpy(new_fn->name, ent_name->name, ent_name->len + extra_data);
  65
  66         while (*p) {
  67                 parent = *p;
  68 diff -ur a/fs/ext4/ext4.h b/fs/ext4/ext4.h
  69 --- a/fs/ext4/ext4.h    2021-12-14 08:28:35.622372837 -0700
  70 +++ b/fs/ext4/ext4.h    2021-12-14 08:30:28.938135681 -0700
  71 @@ -1157,6 +1157,7 @@
  72         __u32 i_csum_seed;
  73
  74         kprojid_t i_projid;
  75 +       void *i_dirdata;
  76  };
  77
  78  /*
  79 @@ -1178,6 +1179,7 @@
  80   * Mount flags set via mount options or defaults
  81   */
  82  #define EXT4_MOUNT_NO_MBCACHE          0x00001 /* Do not use mbcache */
  83 +#define EXT4_MOUNT_DIRDATA             0x00002 /* Data in directory entries */
  84  #define EXT4_MOUNT_GRPID               0x00004 /* Create files with directory's group */
  85  #define EXT4_MOUNT_DEBUG               0x00008 /* Some debugging messages */
  86  #define EXT4_MOUNT_ERRORS_CONT         0x00010 /* Continue on errors */
  87 @@ -2054,6 +2056,7 @@
  88                                          EXT4_FEATURE_INCOMPAT_FLEX_BG| \
  89                                          EXT4_FEATURE_INCOMPAT_EA_INODE| \
  90                                          EXT4_FEATURE_INCOMPAT_MMP | \
  91 +                                        EXT4_FEATURE_INCOMPAT_DIRDATA| \
  92                                          EXT4_FEATURE_INCOMPAT_INLINE_DATA | \
  93                                          EXT4_FEATURE_INCOMPAT_ENCRYPT | \
  94                                          EXT4_FEATURE_INCOMPAT_CASEFOLD | \
  95 @@ -2232,6 +2235,43 @@
  96  #define EXT4_FT_SYMLINK                7
  97
  98  #define EXT4_FT_MAX            8
  99 +#define EXT4_FT_MASK           0xf
 100 +
 101 +#if EXT4_FT_MAX > EXT4_FT_MASK
 102 +#error "conflicting EXT4_FT_MAX and EXT4_FT_MASK"
 103 +#endif
 104 +
 105 +/*
 106 + * d_type has 4 unused bits, so it can hold four types data. these different
 107 + * type of data (e.g. lustre data, high 32 bits of 64-bit inode number) can be
 108 + * stored, in flag order, after file-name in ext4 dirent.
 109 +*/
 110 +/*
 111 + * this flag is added to d_type if ext4 dirent has extra data after
 112 + * filename. this data length is variable and length is stored in first byte
 113 + * of data. data start after filename NUL byte.
 114 + * This is used by Lustre FS.
 115 +  */
 116 +#define EXT4_DIRENT_LUFID              0x10
 117 +
 118 +#define EXT4_LUFID_MAGIC    0xAD200907UL
 119 +struct ext4_dentry_param {
 120 +       __u32  edp_magic;       /* EXT4_LUFID_MAGIC */
 121 +       char   edp_len;         /* size of edp_data in bytes */
 122 +       char   edp_data[0];     /* packed array of data */
 123 +} __packed;
 124 +
 125 +static inline unsigned char *ext4_dentry_get_data(struct super_block *sb,
 126 +                                                 struct ext4_dentry_param *p)
 127 +
 128 +{
 129 +       if (!ext4_has_feature_dirdata(sb))
 130 +               return NULL;
 131 +       if (p && p->edp_magic == EXT4_LUFID_MAGIC)
 132 +               return &p->edp_len;
 133 +       else
 134 +               return NULL;
 135 +}
 136
 137  #define EXT4_FT_DIR_CSUM       0xDE
 138
 139 @@ -2242,8 +2282,16 @@
 140   */
 141  #define EXT4_DIR_PAD                   4
 142  #define EXT4_DIR_ROUND                 (EXT4_DIR_PAD - 1)
 143 -#define EXT4_DIR_REC_LEN(name_len)     (((name_len) + 8 + EXT4_DIR_ROUND) & \
 144 +#define EXT4_DIR_REC_LEN_(name_len)    (((name_len) + 8 + EXT4_DIR_ROUND) & \
 145                                          ~EXT4_DIR_ROUND)
 146 +#define EXT4_DIR_ENTRY_LEN_(de)                (EXT4_DIR_REC_LEN_((de)->name_len +\
 147 +                                       ext4_get_dirent_data_len(de)))
 148 +/* ldiskfs */
 149 +#define EXT4_DIR_REC_LEN(name_len)     EXT4_DIR_REC_LEN_((name_len))
 150 +#define EXT4_DIR_ENTRY_LEN(de)         EXT4_DIR_ENTRY_LEN_((de))
 151 +/* lustre osd_handler compat */
 152 +#define __EXT4_DIR_REC_LEN(name_len)   EXT4_DIR_REC_LEN_((name_len))
 153 +
 154  #define EXT4_MAX_REC_LEN               ((1<<16)-1)
 155
 156  /*
 157 @@ -2710,11 +2758,11 @@
 158                              struct buffer_head *bh,
 159                              void *buf, int buf_size,
 160                              struct ext4_filename *fname,
 161 -                            struct ext4_dir_entry_2 **dest_de);
 162 +                            struct ext4_dir_entry_2 **dest_de, int *dlen);
 163  void ext4_insert_dentry(struct inode *inode,
 164                         struct ext4_dir_entry_2 *de,
 165                         int buf_size,
 166 -                       struct ext4_filename *fname);
 167 +                       struct ext4_filename *fname, void *data);
 168  static inline void ext4_update_dx_flag(struct inode *inode)
 169  {
 170         if (!ext4_has_feature_dir_index(inode->i_sb) &&
 171 @@ -2730,10 +2778,17 @@
 172
 173  static inline  unsigned char get_dtype(struct super_block *sb, int filetype)
 174  {
 175 -       if (!ext4_has_feature_filetype(sb) || filetype >= EXT4_FT_MAX)
 176 +       int fl_index = filetype & EXT4_FT_MASK;
 177 +
 178 +       if (!ext4_has_feature_filetype(sb) || fl_index >= EXT4_FT_MAX)
 179                 return DT_UNKNOWN;
 180
 181 -       return ext4_filetype_table[filetype];
 182 +       if (!test_opt(sb, DIRDATA))
 183 +               return ext4_filetype_table[fl_index];
 184 +
 185 +       return (ext4_filetype_table[fl_index]) |
 186 +               (filetype & EXT4_DIRENT_LUFID);
 187 +
 188  }
 189  extern int ext4_check_all_de(struct inode *dir, struct buffer_head *bh,
 190                              void *buf, int buf_size);
 191 @@ -2922,7 +2977,8 @@
 192
 193  /* namei.c */
 194  extern int ext4_init_new_dir(handle_t *handle, struct inode *dir,
 195 -                            struct inode *inode);
 196 +                            struct inode *inode,
 197 +                            const void *data1, const void *data2);
 198  extern int ext4_dirblock_csum_verify(struct inode *inode,
 199                                      struct buffer_head *bh);
 200  extern int ext4_orphan_add(handle_t *, struct inode *);
 201 @@ -2933,6 +2989,8 @@
 202  extern int ext4_delete_entry(handle_t *handle, struct inode * dir,
 203                              struct ext4_dir_entry_2 *de_del,
 204                              struct buffer_head *bh);
 205 +extern int ext4_add_dot_dotdot(handle_t *handle, struct inode *dir,
 206 +                              struct inode *inode, const void *, const void *);
 207  extern int ext4_htree_fill_tree(struct file *dir_file, __u32 start_hash,
 208                                 __u32 start_minor_hash, __u32 *next_hash);
 209  extern int ext4_search_dir(struct buffer_head *bh,
 210 @@ -3725,6 +3783,36 @@
 211         return buffer_uptodate(bh);
 212  }
 213
 214 +/*
 215 + * Compute the total directory entry data length.
 216 + * This includes the filename and an implicit NUL terminator (always present),
 217 + * and optional extensions.  Each extension has a bit set in the high 4 bits of
 218 + * de->file_type, and the extension length is the first byte in each entry.
 219 + */
 220 +static inline int ext4_get_dirent_data_len(struct ext4_dir_entry_2 *de)
 221 +{
 222 +       char *len = de->name + de->name_len + 1 /* NUL terminator */;
 223 +       int dlen = 0;
 224 +       __u8 extra_data_flags = (de->file_type & ~EXT4_FT_MASK) >> 4;
 225 +       struct ext4_dir_entry_tail *t = (struct ext4_dir_entry_tail *)de;
 226 +
 227 +       if (!t->det_reserved_zero1 &&
 228 +           le16_to_cpu(t->det_rec_len) ==
 229 +               sizeof(struct ext4_dir_entry_tail) &&
 230 +           !t->det_reserved_zero2 &&
 231 +           t->det_reserved_ft == EXT4_FT_DIR_CSUM)
 232 +               return 0;
 233 +
 234 +       while (extra_data_flags) {
 235 +               if (extra_data_flags & 1) {
 236 +                       dlen += *len + (dlen == 0);
 237 +                       len += *len;
 238 +               }
 239 +               extra_data_flags >>= 1;
 240 +       }
 241 +       return dlen;
 242 +}
 243 +
 244  #endif /* __KERNEL__ */
 245
 246  #define EFSBADCRC      EBADMSG         /* Bad CRC detected */
 247 diff -ur a/fs/ext4/fast_commit.c b/fs/ext4/fast_commit.c
 248 --- a/fs/ext4/fast_commit.c     2021-12-14 08:28:35.514373035 -0700
 249 +++ b/fs/ext4/fast_commit.c     2021-12-14 08:29:41.326242185 -0700
 250 @@ -1532,7 +1532,7 @@
 251                         jbd_debug(1, "Dir %d not found.", darg.ino);
 252                         goto out;
 253                 }
 254 -               ret = ext4_init_new_dir(NULL, dir, inode);
 255 +               ret = ext4_init_new_dir(NULL, dir, inode, NULL, NULL);
 256                 iput(dir);
 257                 if (ret) {
 258                         ret = 0;
 259 diff -ur a/fs/ext4/inline.c b/fs/ext4/inline.c
 260 --- a/fs/ext4/inline.c  2021-12-14 08:28:35.518373027 -0700
 261 +++ b/fs/ext4/inline.c  2021-12-14 08:29:41.326242185 -0700
 262 @@ -1023,7 +1023,7 @@
 263         struct ext4_dir_entry_2 *de;
 264
 265         err = ext4_find_dest_de(dir, inode, iloc->bh, inline_start,
 266 -                               inline_size, fname, &de);
 267 +                               inline_size, fname, &de, NULL);
 268         if (err)
 269                 return err;
 270
 271 @@ -1031,7 +1031,7 @@
 272         err = ext4_journal_get_write_access(handle, iloc->bh);
 273         if (err)
 274                 return err;
 275 -       ext4_insert_dentry(inode, de, inline_size, fname);
 276 +       ext4_insert_dentry(inode, de, inline_size, fname, NULL);
 277
 278         ext4_show_inline_dir(dir, iloc->bh, inline_start, inline_size);
 279
 280 @@ -1380,7 +1380,7 @@
 281                         fake.name_len = 1;
 282                         strcpy(fake.name, ".");
 283                         fake.rec_len = ext4_rec_len_to_disk(
 284 -                                               EXT4_DIR_REC_LEN(fake.name_len),
 285 +                                               EXT4_DIR_ENTRY_LEN(&fake),
 286                                                 inline_size);
 287                         ext4_set_de_type(inode->i_sb, &fake, S_IFDIR);
 288                         de = &fake;
 289 @@ -1390,7 +1390,7 @@
 290                         fake.name_len = 2;
 291                         strcpy(fake.name, "..");
 292                         fake.rec_len = ext4_rec_len_to_disk(
 293 -                                               EXT4_DIR_REC_LEN(fake.name_len),
 294 +                                               EXT4_DIR_ENTRY_LEN(&fake),
 295                                                 inline_size);
 296                         ext4_set_de_type(inode->i_sb, &fake, S_IFDIR);
 297                         de = &fake;
 298 diff -ur a/fs/ext4/namei.c b/fs/ext4/namei.c
 299 --- a/fs/ext4/namei.c   2021-12-14 08:28:35.638372808 -0700
 300 +++ b/fs/ext4/namei.c   2021-12-14 08:29:41.330242176 -0700
 301 @@ -265,7 +265,8 @@
 302  static unsigned dx_get_limit(struct dx_entry *entries);
 303  static void dx_set_count(struct dx_entry *entries, unsigned value);
 304  static void dx_set_limit(struct dx_entry *entries, unsigned value);
 305 -static unsigned dx_root_limit(struct inode *dir, unsigned infosize);
 306 +static inline unsigned dx_root_limit(struct inode *dir,
 307 +               struct ext4_dir_entry_2 *dot_de, unsigned infosize);
 308  static unsigned dx_node_limit(struct inode *dir);
 309  static struct dx_frame *dx_probe(struct ext4_filename *fname,
 310                                  struct inode *dir,
 311 @@ -409,22 +410,23 @@
 312  {
 313         struct ext4_dir_entry *dp;
 314         struct dx_root_info *root;
 315 -       int count_offset;
 316 +       int count_offset, dot_rec_len, dotdot_rec_len;
 317
 318         if (le16_to_cpu(dirent->rec_len) == EXT4_BLOCK_SIZE(inode->i_sb))
 319                 count_offset = 8;
 320 -       else if (le16_to_cpu(dirent->rec_len) == 12) {
 321 -               dp = (struct ext4_dir_entry *)(((void *)dirent) + 12);
 322 +       else {
 323 +               dot_rec_len = le16_to_cpu(dirent->rec_len);
 324 +               dp = (struct ext4_dir_entry *)(((void *)dirent) + dot_rec_len);
 325                 if (le16_to_cpu(dp->rec_len) !=
 326 -                   EXT4_BLOCK_SIZE(inode->i_sb) - 12)
 327 +                   EXT4_BLOCK_SIZE(inode->i_sb) - dot_rec_len)
 328                         return NULL;
 329 -               root = (struct dx_root_info *)(((void *)dp + 12));
 330 +               dotdot_rec_len = EXT4_DIR_ENTRY_LEN((struct ext4_dir_entry_2 *)dp);
 331 +               root = (struct dx_root_info *)(((void *)dp + dotdot_rec_len));
 332                 if (root->reserved_zero ||
 333                     root->info_length != sizeof(struct dx_root_info))
 334                         return NULL;
 335 -               count_offset = 32;
 336 -       } else
 337 -               return NULL;
 338 +               count_offset = 8 + dot_rec_len + dotdot_rec_len;
 339 +       }
 340
 341         if (offset)
 342                 *offset = count_offset;
 343 @@ -529,11 +531,12 @@
 344   */
 345  struct dx_root_info *dx_get_dx_info(struct ext4_dir_entry_2 *de)
 346  {
 347 +       BUG_ON(de->name_len != 1);
 348         /* get dotdot first */
 349 -       de = (struct ext4_dir_entry_2 *)((char *)de + EXT4_DIR_REC_LEN(1));
 350 +       de = (struct ext4_dir_entry_2 *)((char *)de + EXT4_DIR_ENTRY_LEN(de));
 351
 352         /* dx root info is after dotdot entry */
 353 -       de = (struct ext4_dir_entry_2 *)((char *)de + EXT4_DIR_REC_LEN(2));
 354 +       de = (struct ext4_dir_entry_2 *)((char *)de + EXT4_DIR_ENTRY_LEN(de));
 355
 356         return (struct dx_root_info *)de;
 357  }
 358 @@ -578,10 +581,16 @@
 359         ((struct dx_countlimit *) entries)->limit = cpu_to_le16(value);
 360  }
 361
 362 -static inline unsigned dx_root_limit(struct inode *dir, unsigned infosize)
 363 +static inline unsigned dx_root_limit(struct inode *dir,
 364 +               struct ext4_dir_entry_2 *dot_de, unsigned infosize)
 365  {
 366 -       unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(1) -
 367 -               EXT4_DIR_REC_LEN(2) - infosize;
 368 +       struct ext4_dir_entry_2 *dotdot_de;
 369 +       unsigned entry_space;
 370 +
 371 +       BUG_ON(dot_de->name_len != 1);
 372 +       dotdot_de = ext4_next_entry(dot_de, dir->i_sb->s_blocksize);
 373 +       entry_space = dir->i_sb->s_blocksize - EXT4_DIR_ENTRY_LEN(dot_de) -
 374 +                        EXT4_DIR_ENTRY_LEN(dotdot_de) - infosize;
 375
 376         if (ext4_has_metadata_csum(dir->i_sb))
 377                 entry_space -= sizeof(struct dx_tail);
 378 @@ -701,7 +710,7 @@
 379                                        (unsigned) ((char *) de - base));
 380  #endif
 381                         }
 382 -                       space += EXT4_DIR_REC_LEN(de->name_len);
 383 +                       space += EXT4_DIR_ENTRY_LEN(de);
 384                         names++;
 385                 }
 386                 de = ext4_next_entry(de, size);
 387 @@ -808,11 +817,14 @@
 388
 389         entries = (struct dx_entry *)(((char *)info) + info->info_length);
 390
 391 -       if (dx_get_limit(entries) != dx_root_limit(dir,
 392 -                                                  info->info_length)) {
 393 +       if (dx_get_limit(entries) !=
 394 +           dx_root_limit(dir, (struct ext4_dir_entry_2 *)frame->bh->b_data,
 395 +                         info->info_length)) {
 396                 ext4_warning_inode(dir, "dx entry: limit %u != root limit %u",
 397                                    dx_get_limit(entries),
 398 -                                  dx_root_limit(dir, info->info_length));
 399 +                                  dx_root_limit(dir,
 400 +                                         (struct ext4_dir_entry_2 *)frame->bh->b_data,
 401 +                                         info->info_length));
 402                 goto fail;
 403         }
 404
 405 @@ -1798,7 +1810,7 @@
 406         while (count--) {
 407                 struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *)
 408                                                 (from + (map->offs<<2));
 409 -               rec_len = EXT4_DIR_REC_LEN(de->name_len);
 410 +               rec_len = EXT4_DIR_ENTRY_LEN(de);
 411                 memcpy (to, de, rec_len);
 412                 ((struct ext4_dir_entry_2 *) to)->rec_len =
 413                                 ext4_rec_len_to_disk(rec_len, blocksize);
 414 @@ -1822,7 +1834,7 @@
 415         while ((char*)de < base + blocksize) {
 416                 next = ext4_next_entry(de, blocksize);
 417                 if (de->inode && de->name_len) {
 418 -                       rec_len = EXT4_DIR_REC_LEN(de->name_len);
 419 +                       rec_len = EXT4_DIR_ENTRY_LEN(de);
 420                         if (de > to)
 421                                 memmove(to, de, rec_len);
 422                         to->rec_len = ext4_rec_len_to_disk(rec_len, blocksize);
 423 @@ -1959,14 +1971,16 @@
 424                       struct buffer_head *bh,
 425                       void *buf, int buf_size,
 426                       struct ext4_filename *fname,
 427 -                     struct ext4_dir_entry_2 **dest_de)
 428 +                     struct ext4_dir_entry_2 **dest_de, int *dlen)
 429  {
 430         struct ext4_dir_entry_2 *de;
 431 -       unsigned short reclen = EXT4_DIR_REC_LEN(fname_len(fname));
 432 +       unsigned short reclen = EXT4_DIR_REC_LEN(fname_len(fname)) +
 433 +                                                 (dlen ? *dlen : 0);
 434         int nlen, rlen;
 435         unsigned int offset = 0;
 436         char *top;
 437
 438 +       dlen ? *dlen = 0 : 0; /* default set to 0 */
 439         de = (struct ext4_dir_entry_2 *)buf;
 440         top = buf + buf_size - reclen;
 441         while ((char *) de <= top) {
 442 @@ -1975,10 +1989,26 @@
 443                         return -EFSCORRUPTED;
 444                 if (ext4_match(dir, fname, de))
 445                         return -EEXIST;
 446 -               nlen = EXT4_DIR_REC_LEN(de->name_len);
 447 +               nlen = EXT4_DIR_ENTRY_LEN(de);
 448                 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
 449                 if ((de->inode ? rlen - nlen : rlen) >= reclen)
 450                         break;
 451 +               /* Then for dotdot entries, check for the smaller space
 452 +                * required for just the entry, no FID */
 453 +               if (fname_len(fname) == 2 && memcmp(fname_name(fname), "..", 2) == 0) {
 454 +                       if ((de->inode ? rlen - nlen : rlen) >=
 455 +                           EXT4_DIR_REC_LEN(fname_len(fname))) {
 456 +                               /* set dlen=1 to indicate not
 457 +                                * enough space store fid */
 458 +                               dlen ? *dlen = 1 : 0;
 459 +                               break;
 460 +                       }
 461 +                       /* The new ".." entry must be written over the
 462 +                        * previous ".." entry, which is the first
 463 +                        * entry traversed by this scan. If it doesn't
 464 +                        * fit, something is badly wrong, so -EIO. */
 465 +                       return -EIO;
 466 +               }
 467                 de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
 468                 offset += rlen;
 469         }
 470 @@ -1992,12 +2022,12 @@
 471  void ext4_insert_dentry(struct inode *inode,
 472                         struct ext4_dir_entry_2 *de,
 473                         int buf_size,
 474 -                       struct ext4_filename *fname)
 475 +                       struct ext4_filename *fname, void *data)
 476  {
 477
 478         int nlen, rlen;
 479
 480 -       nlen = EXT4_DIR_REC_LEN(de->name_len);
 481 +       nlen = EXT4_DIR_ENTRY_LEN(de);
 482         rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
 483         if (de->inode) {
 484                 struct ext4_dir_entry_2 *de1 =
 485 @@ -2011,6 +2041,11 @@
 486         ext4_set_de_type(inode->i_sb, de, inode->i_mode);
 487         de->name_len = fname_len(fname);
 488         memcpy(de->name, fname_name(fname), fname_len(fname));
 489 +       if (data) {
 490 +               de->name[fname_len(fname)] = 0;
 491 +               memcpy(&de->name[fname_len(fname) + 1], data, *(char *)data);
 492 +               de->file_type |= EXT4_DIRENT_LUFID;
 493 +       }
 494  }
 495
 496  /*
 497 @@ -2028,14 +2063,19 @@
 498  {
 499         unsigned int    blocksize = dir->i_sb->s_blocksize;
 500         int             csum_size = 0;
 501 -       int             err, err2;
 502 +       int             err, err2, dlen = 0;
 503 +       unsigned char   *data;
 504
 505 +       data = ext4_dentry_get_data(inode->i_sb, (struct ext4_dentry_param *)
 506 +                                               EXT4_I(inode)->i_dirdata);
 507         if (ext4_has_metadata_csum(inode->i_sb))
 508                 csum_size = sizeof(struct ext4_dir_entry_tail);
 509
 510         if (!de) {
 511 +               if (data)
 512 +                       dlen = (*data) + 1;
 513                 err = ext4_find_dest_de(dir, inode, bh, bh->b_data,
 514 -                                       blocksize - csum_size, fname, &de);
 515 +                                       blocksize - csum_size, fname, &de, &dlen);
 516                 if (err)
 517                         return err;
 518         }
 519 @@ -2047,7 +2087,10 @@
 520         }
 521
 522         /* By now the buffer is marked for journaling */
 523 -       ext4_insert_dentry(inode, de, blocksize, fname);
 524 +       /* If writing the short form of "dotdot", don't add the data section */
 525 +       if (dlen == 1)
 526 +               data = NULL;
 527 +       ext4_insert_dentry(inode, de, blocksize, fname, data);
 528
 529         /*
 530          * XXX shouldn't update any times until successful
 531 @@ -2152,7 +2195,8 @@
 532
 533         dx_set_block(entries, 1);
 534         dx_set_count(entries, 1);
 535 -       dx_set_limit(entries, dx_root_limit(dir, sizeof(*dx_info)));
 536 +       dx_set_limit(entries, dx_root_limit(dir,
 537 +                                        dot_de, sizeof(*dx_info)));
 538
 539         /* Initialize as for dx_probe */
 540         fname->hinfo.hash_version = dx_info->hash_version;
 541 @@ -2202,6 +2246,8 @@
 542         struct buffer_head *dir_block;
 543         struct ext4_dir_entry_2 *de;
 544         int len, journal = 0, err = 0;
 545 +       int dlen = 0;
 546 +       char *data;
 547
 548         if (IS_ERR(handle))
 549                 return PTR_ERR(handle);
 550 @@ -2227,11 +2273,16 @@
 551                         goto out_journal;
 552
 553                 journal = 1;
 554 -               de->rec_len = cpu_to_le16(EXT4_DIR_REC_LEN(1));
 555 +               de->rec_len = cpu_to_le16(EXT4_DIR_ENTRY_LEN(de));
 556         }
 557
 558 -       len -= EXT4_DIR_REC_LEN(1);
 559 -       assert(len == 0 || len >= EXT4_DIR_REC_LEN(2));
 560 +       len -= EXT4_DIR_ENTRY_LEN(de);
 561 +       data = ext4_dentry_get_data(dir->i_sb,
 562 +                       (struct ext4_dentry_param *)dentry->d_fsdata);
 563 +       if (data)
 564 +               dlen = *data + 1;
 565 +       assert(len == 0 || len >= EXT4_DIR_REC_LEN(2 + dlen));
 566 +
 567         de = (struct ext4_dir_entry_2 *)
 568                         ((char *) de + le16_to_cpu(de->rec_len));
 569         if (!journal) {
 570 @@ -2248,7 +2299,12 @@
 571                 assert(le16_to_cpu(de->rec_len) >= EXT4_DIR_REC_LEN(2));
 572         de->name_len = 2;
 573         strcpy(de->name, "..");
 574 -       ext4_set_de_type(dir->i_sb, de, S_IFDIR);
 575 +       if (data != NULL && ext4_get_dirent_data_len(de) >= dlen) {
 576 +               de->name[2] = 0;
 577 +               memcpy(&de->name[2 + 1], data, *data);
 578 +               ext4_set_de_type(dir->i_sb, de, S_IFDIR);
 579 +               de->file_type |= EXT4_DIRENT_LUFID;
 580 +       }
 581
 582  out_journal:
 583         if (journal) {
 584 @@ -2286,6 +2342,7 @@
 585         ext4_lblk_t block, blocks;
 586         int     csum_size = 0;
 587
 588 +       EXT4_I(inode)->i_dirdata = dentry->d_fsdata;
 589         if (ext4_has_metadata_csum(inode->i_sb))
 590                 csum_size = sizeof(struct ext4_dir_entry_tail);
 591
 592 @@ -2839,37 +2896,70 @@
 593         return err;
 594  }
 595
 596 +struct tp_block {
 597 +       struct inode *inode;
 598 +       void *data1;
 599 +       void *data2;
 600 +};
 601 +
 602  struct ext4_dir_entry_2 *ext4_init_dot_dotdot(struct inode *inode,
 603                           struct ext4_dir_entry_2 *de,
 604                           int blocksize, int csum_size,
 605                           unsigned int parent_ino, int dotdot_real_len)
 606  {
 607 +       void *data1 = NULL, *data2 = NULL;
 608 +       int dot_reclen = 0;
 609 +
 610 +       if (dotdot_real_len == 10) {
 611 +               struct tp_block *tpb = (struct tp_block *)inode;
 612 +               data1 = tpb->data1;
 613 +               data2 = tpb->data2;
 614 +               inode = tpb->inode;
 615 +               dotdot_real_len = 0;
 616 +       }
 617         de->inode = cpu_to_le32(inode->i_ino);
 618         de->name_len = 1;
 619 -       de->rec_len = ext4_rec_len_to_disk(EXT4_DIR_REC_LEN(de->name_len),
 620 -                                          blocksize);
 621         strcpy(de->name, ".");
 622         ext4_set_de_type(inode->i_sb, de, S_IFDIR);
 623
 624 +       /* get packed fid data*/
 625 +       data1 = ext4_dentry_get_data(inode->i_sb,
 626 +                               (struct ext4_dentry_param *) data1);
 627 +       if (data1) {
 628 +               de->name[1] = 0;
 629 +               memcpy(&de->name[2], data1, *(char *) data1);
 630 +               de->file_type |= EXT4_DIRENT_LUFID;
 631 +       }
 632 +       de->rec_len = cpu_to_le16(EXT4_DIR_ENTRY_LEN(de));
 633 +       dot_reclen = cpu_to_le16(de->rec_len);
 634         de = ext4_next_entry(de, blocksize);
 635         de->inode = cpu_to_le32(parent_ino);
 636         de->name_len = 2;
 637 +       strcpy(de->name, "..");
 638 +       ext4_set_de_type(inode->i_sb, de, S_IFDIR);
 639 +       data2 = ext4_dentry_get_data(inode->i_sb,
 640 +                       (struct ext4_dentry_param *) data2);
 641 +       if (data2) {
 642 +               de->name[2] = 0;
 643 +               memcpy(&de->name[3], data2, *(char *) data2);
 644 +               de->file_type |= EXT4_DIRENT_LUFID;
 645 +       }
 646         if (!dotdot_real_len)
 647                 de->rec_len = ext4_rec_len_to_disk(blocksize -
 648 -                                       (csum_size + EXT4_DIR_REC_LEN(1)),
 649 +                                       (csum_size + dot_reclen),
 650                                         blocksize);
 651         else
 652                 de->rec_len = ext4_rec_len_to_disk(
 653 -                               EXT4_DIR_REC_LEN(de->name_len), blocksize);
 654 -       strcpy(de->name, "..");
 655 -       ext4_set_de_type(inode->i_sb, de, S_IFDIR);
 656 +                               EXT4_DIR_ENTRY_LEN(de), blocksize);
 657
 658         return ext4_next_entry(de, blocksize);
 659  }
 660
 661  int ext4_init_new_dir(handle_t *handle, struct inode *dir,
 662 -                            struct inode *inode)
 663 +                            struct inode *inode,
 664 +                            const void *data1, const void *data2)
 665  {
 666 +       struct tp_block param;
 667         struct buffer_head *dir_block = NULL;
 668         struct ext4_dir_entry_2 *de;
 669         ext4_lblk_t block = 0;
 670 @@ -2893,7 +2983,11 @@
 671         if (IS_ERR(dir_block))
 672                 return PTR_ERR(dir_block);
 673         de = (struct ext4_dir_entry_2 *)dir_block->b_data;
 674 -       ext4_init_dot_dotdot(inode, de, blocksize, csum_size, dir->i_ino, 0);
 675 +       param.inode = inode;
 676 +       param.data1 = (void *)data1;
 677 +       param.data2 = (void *)data2;
 678 +       ext4_init_dot_dotdot((struct inode *)(&param), de, blocksize,
 679 +                            csum_size, dir->i_ino, 10);
 680         set_nlink(inode, 2);
 681         if (csum_size)
 682                 ext4_initialize_dirent_tail(dir_block, blocksize);
 683 @@ -2908,6 +3002,29 @@
 684         return err;
 685  }
 686
 687 +/* Initialize @inode as a subdirectory of @dir, and add the
 688 + * "." and ".." entries into the first directory block. */
 689 +int ext4_add_dot_dotdot(handle_t *handle, struct inode *dir,
 690 +                       struct inode *inode,
 691 +                       const void *data1, const void *data2)
 692 +{
 693 +       int rc;
 694 +
 695 +       if (IS_ERR(handle))
 696 +               return PTR_ERR(handle);
 697 +
 698 +       if (IS_DIRSYNC(dir))
 699 +               ext4_handle_sync(handle);
 700 +
 701 +       inode->i_op = &ext4_dir_inode_operations;
 702 +       inode->i_fop = &ext4_dir_operations;
 703 +       rc = ext4_init_new_dir(handle, dir, inode, data1, data2);
 704 +       if (!rc)
 705 +               rc = ext4_mark_inode_dirty(handle, inode);
 706 +       return rc;
 707 +}
 708 +EXPORT_SYMBOL(ext4_add_dot_dotdot);
 709 +
 710  static int ext4_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
 711  {
 712         handle_t *handle;
 713 @@ -2934,7 +3051,7 @@
 714
 715         inode->i_op = &ext4_dir_inode_operations;
 716         inode->i_fop = &ext4_dir_operations;
 717 -       err = ext4_init_new_dir(handle, dir, inode);
 718 +       err = ext4_init_new_dir(handle, dir, inode, NULL, NULL);
 719         if (err)
 720                 goto out_clear_inode;
 721         err = ext4_mark_inode_dirty(handle, inode);
 722 diff -ur a/fs/ext4/super.c b/fs/ext4/super.c
 723 --- a/fs/ext4/super.c   2021-12-14 08:28:35.614372852 -0700
 724 +++ b/fs/ext4/super.c   2021-12-14 08:29:41.334242168 -0700
 725 @@ -1694,7 +1694,7 @@
 726         Opt_inlinecrypt,
 727         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
 728         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_jqfmt_vfsv1, Opt_quota,
 729 -       Opt_noquota, Opt_barrier, Opt_nobarrier, Opt_err,
 730 +       Opt_noquota, Opt_barrier, Opt_nobarrier, Opt_err, Opt_dirdata,
 731         Opt_usrquota, Opt_grpquota, Opt_prjquota, Opt_i_version,
 732         Opt_dax, Opt_dax_always, Opt_dax_inode, Opt_dax_never,
 733         Opt_stripe, Opt_delalloc, Opt_nodelalloc, Opt_warn_on_error,
 734 @@ -1778,6 +1778,7 @@
 735         {Opt_nolazytime, "nolazytime"},
 736         {Opt_debug_want_extra_isize, "debug_want_extra_isize=%u"},
 737         {Opt_nodelalloc, "nodelalloc"},
 738 +       {Opt_dirdata, "dirdata"},
 739         {Opt_removed, "mblk_io_submit"},
 740         {Opt_removed, "nomblk_io_submit"},
 741         {Opt_block_validity, "block_validity"},
 742 @@ -2017,6 +2018,7 @@
 743         {Opt_usrjquota, 0, MOPT_Q | MOPT_STRING},
 744         {Opt_grpjquota, 0, MOPT_Q | MOPT_STRING},
 745         {Opt_offusrjquota, 0, MOPT_Q},
 746 +       {Opt_dirdata, EXT4_MOUNT_DIRDATA, MOPT_SET},
 747         {Opt_offgrpjquota, 0, MOPT_Q},
 748         {Opt_jqfmt_vfsold, QFMT_VFS_OLD, MOPT_QFMT},
 749         {Opt_jqfmt_vfsv0, QFMT_VFS_V0, MOPT_QFMT},