ldiskfs/kernel_patches/patches/oe2203sp1/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        | 100 +++++++++++++++++++--
  27  fs/ext4/fast_commit.c |   2 +-
  28  fs/ext4/inline.c      |   8 +-
  29  fs/ext4/namei.c       | 201 +++++++++++++++++++++++++++++++++---------
  30  fs/ext4/super.c       |   4 +-
  31  6 files changed, 270 insertions(+), 58 deletions(-)
  32
  33 diff --git a/fs/ext4/dir.c b/fs/ext4/dir.c
  34 index 70a0f5e..ff7d8c3 100644
  35 --- a/fs/ext4/dir.c
  36 +++ b/fs/ext4/dir.c
  37 @@ -78,7 +78,7 @@ int __ext4_check_dir_entry(const char *function, unsigned int line,
  38                 error_msg = "rec_len is smaller than minimal";
  39         else if (unlikely(rlen % 4 != 0))
  40                 error_msg = "rec_len % 4 != 0";
  41 -       else if (unlikely(rlen < EXT4_DIR_REC_LEN(de->name_len)))
  42 +       else if (unlikely(rlen < EXT4_DIR_ENTRY_LEN(de)))
  43                 error_msg = "rec_len is too small for name_len";
  44         else if (unlikely(next_offset > size))
  45                 error_msg = "directory entry overrun";
  46 @@ -226,7 +226,7 @@ static int ext4_readdir(struct file *file, struct dir_context *ctx)
  47                                  * failure will be detected in the
  48                                  * dirent test below. */
  49                                 if (ext4_rec_len_from_disk(de->rec_len,
  50 -                                       sb->s_blocksize) < EXT4_DIR_REC_LEN(1))
  51 +                                   sb->s_blocksize) < EXT4_DIR_REC_LEN(1))
  52                                         break;
  53                                 i += ext4_rec_len_from_disk(de->rec_len,
  54                                                             sb->s_blocksize);
  55 @@ -449,12 +449,17 @@ int ext4_htree_store_dirent(struct file *dir_file, __u32 hash,
  56         struct fname *fname, *new_fn;
  57         struct dir_private_info *info;
  58         int len;
  59 +       int extra_data = 0;
  60
  61         info = dir_file->private_data;
  62         p = &info->root.rb_node;
  63
  64         /* Create and allocate the fname structure */
  65 -       len = sizeof(struct fname) + ent_name->len + 1;
  66 +       if (dirent->file_type & EXT4_DIRENT_LUFID)
  67 +               extra_data = ext4_get_dirent_data_len(dirent);
  68 +
  69 +       len = sizeof(struct fname) + ent_name->len + extra_data + 1;
  70 +
  71         new_fn = kzalloc(len, GFP_KERNEL);
  72         if (!new_fn)
  73                 return -ENOMEM;
  74 @@ -463,7 +468,7 @@ int ext4_htree_store_dirent(struct file *dir_file, __u32 hash,
  75         new_fn->inode = le32_to_cpu(dirent->inode);
  76         new_fn->name_len = ent_name->len;
  77         new_fn->file_type = dirent->file_type;
  78 -       memcpy(new_fn->name, ent_name->name, ent_name->len);
  79 +       memcpy(new_fn->name, ent_name->name, ent_name->len + extra_data);
  80
  81         while (*p) {
  82                 parent = *p;
  83 diff --git a/fs/ext4/ext4.h b/fs/ext4/ext4.h
  84 index 143ce00..98786d8 100644
  85 --- a/fs/ext4/ext4.h
  86 +++ b/fs/ext4/ext4.h
  87 @@ -1167,6 +1167,7 @@ struct ext4_inode_info {
  88         __u32 i_csum_seed;
  89
  90         kprojid_t i_projid;
  91 +       void *i_dirdata;
  92
  93         /* Protect concurrent add cluster delayed block and remove block */
  94         struct mutex i_clu_lock;
  95 @@ -1191,6 +1192,7 @@ struct ext4_inode_info {
  96   * Mount flags set via mount options or defaults
  97   */
  98  #define EXT4_MOUNT_NO_MBCACHE          0x00001 /* Do not use mbcache */
  99 +#define EXT4_MOUNT_DIRDATA             0x00002 /* Data in directory entries */
 100  #define EXT4_MOUNT_GRPID               0x00004 /* Create files with directory's group */
 101  #define EXT4_MOUNT_DEBUG               0x00008 /* Some debugging messages */
 102  #define EXT4_MOUNT_ERRORS_CONT         0x00010 /* Continue on errors */
 103 @@ -2086,6 +2088,7 @@ EXT4_FEATURE_INCOMPAT_FUNCS(casefold,             CASEFOLD)
 104                                          EXT4_FEATURE_INCOMPAT_FLEX_BG| \
 105                                          EXT4_FEATURE_INCOMPAT_EA_INODE| \
 106                                          EXT4_FEATURE_INCOMPAT_MMP | \
 107 +                                        EXT4_FEATURE_INCOMPAT_DIRDATA| \
 108                                          EXT4_FEATURE_INCOMPAT_INLINE_DATA | \
 109                                          EXT4_FEATURE_INCOMPAT_ENCRYPT | \
 110                                          EXT4_FEATURE_INCOMPAT_CASEFOLD | \
 111 @@ -2268,6 +2271,43 @@ struct ext4_dir_entry_tail {
 112  #define EXT4_FT_SYMLINK                7
 113
 114  #define EXT4_FT_MAX            8
 115 +#define EXT4_FT_MASK           0xf
 116 +
 117 +#if EXT4_FT_MAX > EXT4_FT_MASK
 118 +#error "conflicting EXT4_FT_MAX and EXT4_FT_MASK"
 119 +#endif
 120 +
 121 +/*
 122 + * d_type has 4 unused bits, so it can hold four types data. these different
 123 + * type of data (e.g. lustre data, high 32 bits of 64-bit inode number) can be
 124 + * stored, in flag order, after file-name in ext4 dirent.
 125 +*/
 126 +/*
 127 + * this flag is added to d_type if ext4 dirent has extra data after
 128 + * filename. this data length is variable and length is stored in first byte
 129 + * of data. data start after filename NUL byte.
 130 + * This is used by Lustre FS.
 131 +  */
 132 +#define EXT4_DIRENT_LUFID              0x10
 133 +
 134 +#define EXT4_LUFID_MAGIC    0xAD200907UL
 135 +struct ext4_dentry_param {
 136 +       __u32  edp_magic;       /* EXT4_LUFID_MAGIC */
 137 +       char   edp_len;         /* size of edp_data in bytes */
 138 +       char   edp_data[0];     /* packed array of data */
 139 +} __packed;
 140 +
 141 +static inline unsigned char *ext4_dentry_get_data(struct super_block *sb,
 142 +                                                 struct ext4_dentry_param *p)
 143 +
 144 +{
 145 +       if (!ext4_has_feature_dirdata(sb))
 146 +               return NULL;
 147 +       if (p && p->edp_magic == EXT4_LUFID_MAGIC)
 148 +               return &p->edp_len;
 149 +       else
 150 +               return NULL;
 151 +}
 152
 153  #define EXT4_FT_DIR_CSUM       0xDE
 154
 155 @@ -2278,8 +2318,16 @@ struct ext4_dir_entry_tail {
 156   */
 157  #define EXT4_DIR_PAD                   4
 158  #define EXT4_DIR_ROUND                 (EXT4_DIR_PAD - 1)
 159 -#define EXT4_DIR_REC_LEN(name_len)     (((name_len) + 8 + EXT4_DIR_ROUND) & \
 160 +#define EXT4_DIR_REC_LEN_(name_len)    (((name_len) + 8 + EXT4_DIR_ROUND) & \
 161                                          ~EXT4_DIR_ROUND)
 162 +#define EXT4_DIR_ENTRY_LEN_(de)                (EXT4_DIR_REC_LEN_((de)->name_len +\
 163 +                                       ext4_get_dirent_data_len(de)))
 164 +/* ldiskfs */
 165 +#define EXT4_DIR_REC_LEN(name_len)     EXT4_DIR_REC_LEN_((name_len))
 166 +#define EXT4_DIR_ENTRY_LEN(de)         EXT4_DIR_ENTRY_LEN_((de))
 167 +/* lustre osd_handler compat */
 168 +#define __EXT4_DIR_REC_LEN(name_len)   EXT4_DIR_REC_LEN_((name_len))
 169 +
 170  #define EXT4_MAX_REC_LEN               ((1<<16)-1)
 171
 172  /*
 173 @@ -2746,11 +2794,11 @@ extern int ext4_find_dest_de(struct inode *dir, struct inode *inode,
 174                              struct buffer_head *bh,
 175                              void *buf, int buf_size,
 176                              struct ext4_filename *fname,
 177 -                            struct ext4_dir_entry_2 **dest_de);
 178 +                            struct ext4_dir_entry_2 **dest_de, int *dlen);
 179  void ext4_insert_dentry(struct inode *inode,
 180                         struct ext4_dir_entry_2 *de,
 181                         int buf_size,
 182 -                       struct ext4_filename *fname);
 183 +                       struct ext4_filename *fname, void *data);
 184  static inline void ext4_update_dx_flag(struct inode *inode)
 185  {
 186         if (!ext4_has_feature_dir_index(inode->i_sb) &&
 187 @@ -2766,10 +2814,17 @@ static const unsigned char ext4_filetype_table[] = {
 188
 189  static inline  unsigned char get_dtype(struct super_block *sb, int filetype)
 190  {
 191 -       if (!ext4_has_feature_filetype(sb) || filetype >= EXT4_FT_MAX)
 192 +       int fl_index = filetype & EXT4_FT_MASK;
 193 +
 194 +       if (!ext4_has_feature_filetype(sb) || fl_index >= EXT4_FT_MAX)
 195                 return DT_UNKNOWN;
 196
 197 -       return ext4_filetype_table[filetype];
 198 +       if (!test_opt(sb, DIRDATA))
 199 +               return ext4_filetype_table[fl_index];
 200 +
 201 +       return (ext4_filetype_table[fl_index]) |
 202 +               (filetype & EXT4_DIRENT_LUFID);
 203 +
 204  }
 205  extern int ext4_check_all_de(struct inode *dir, struct buffer_head *bh,
 206                              void *buf, int buf_size);
 207 @@ -2965,7 +3020,8 @@ extern int ext4_ind_migrate(struct inode *inode);
 208
 209  /* namei.c */
 210  extern int ext4_init_new_dir(handle_t *handle, struct inode *dir,
 211 -                            struct inode *inode);
 212 +                            struct inode *inode,
 213 +                            const void *data1, const void *data2);
 214  extern int ext4_dirblock_csum_verify(struct inode *inode,
 215                                      struct buffer_head *bh);
 216  extern int ext4_orphan_add(handle_t *, struct inode *);
 217 @@ -2976,6 +3032,8 @@ extern struct inode *ext4_create_inode(handle_t *handle,
 218  extern int ext4_delete_entry(handle_t *handle, struct inode * dir,
 219                              struct ext4_dir_entry_2 *de_del,
 220                              struct buffer_head *bh);
 221 +extern int ext4_add_dot_dotdot(handle_t *handle, struct inode *dir,
 222 +                              struct inode *inode, const void *, const void *);
 223  extern int ext4_htree_fill_tree(struct file *dir_file, __u32 start_hash,
 224                                 __u32 start_minor_hash, __u32 *next_hash);
 225  extern int ext4_search_dir(struct buffer_head *bh,
 226 @@ -3765,6 +3823,36 @@ static inline int ext4_buffer_uptodate(struct buffer_head *bh)
 227         return buffer_uptodate(bh);
 228  }
 229
 230 +/*
 231 + * Compute the total directory entry data length.
 232 + * This includes the filename and an implicit NUL terminator (always present),
 233 + * and optional extensions.  Each extension has a bit set in the high 4 bits of
 234 + * de->file_type, and the extension length is the first byte in each entry.
 235 + */
 236 +static inline int ext4_get_dirent_data_len(struct ext4_dir_entry_2 *de)
 237 +{
 238 +       char *len = de->name + de->name_len + 1 /* NUL terminator */;
 239 +       int dlen = 0;
 240 +       __u8 extra_data_flags = (de->file_type & ~EXT4_FT_MASK) >> 4;
 241 +       struct ext4_dir_entry_tail *t = (struct ext4_dir_entry_tail *)de;
 242 +
 243 +       if (!t->det_reserved_zero1 &&
 244 +           le16_to_cpu(t->det_rec_len) ==
 245 +               sizeof(struct ext4_dir_entry_tail) &&
 246 +           !t->det_reserved_zero2 &&
 247 +           t->det_reserved_ft == EXT4_FT_DIR_CSUM)
 248 +               return 0;
 249 +
 250 +       while (extra_data_flags) {
 251 +               if (extra_data_flags & 1) {
 252 +                       dlen += *len + (dlen == 0);
 253 +                       len += *len;
 254 +               }
 255 +               extra_data_flags >>= 1;
 256 +       }
 257 +       return dlen;
 258 +}
 259 +
 260  #endif /* __KERNEL__ */
 261
 262  #define EFSBADCRC      EBADMSG         /* Bad CRC detected */
 263 diff --git a/fs/ext4/fast_commit.c b/fs/ext4/fast_commit.c
 264 index 41dcf21..1023ace 100644
 265 --- a/fs/ext4/fast_commit.c
 266 +++ b/fs/ext4/fast_commit.c
 267 @@ -1547,7 +1547,7 @@ static int ext4_fc_replay_create(struct super_block *sb, struct ext4_fc_tl *tl,
 268                         jbd_debug(1, "Dir %d not found.", darg.ino);
 269                         goto out;
 270                 }
 271 -               ret = ext4_init_new_dir(NULL, dir, inode);
 272 +               ret = ext4_init_new_dir(NULL, dir, inode, NULL, NULL);
 273                 iput(dir);
 274                 if (ret) {
 275                         ret = 0;
 276 diff --git a/fs/ext4/inline.c b/fs/ext4/inline.c
 277 index c2c688c..686d14a 100644
 278 --- a/fs/ext4/inline.c
 279 +++ b/fs/ext4/inline.c
 280 @@ -1033,7 +1033,7 @@ static int ext4_add_dirent_to_inline(handle_t *handle,
 281         struct ext4_dir_entry_2 *de;
 282
 283         err = ext4_find_dest_de(dir, inode, iloc->bh, inline_start,
 284 -                               inline_size, fname, &de);
 285 +                               inline_size, fname, &de, NULL);
 286         if (err)
 287                 return err;
 288
 289 @@ -1041,7 +1041,7 @@ static int ext4_add_dirent_to_inline(handle_t *handle,
 290         err = ext4_journal_get_write_access(handle, iloc->bh);
 291         if (err)
 292                 return err;
 293 -       ext4_insert_dentry(inode, de, inline_size, fname);
 294 +       ext4_insert_dentry(inode, de, inline_size, fname, NULL);
 295
 296         ext4_show_inline_dir(dir, iloc->bh, inline_start, inline_size);
 297
 298 @@ -1398,7 +1398,7 @@ int ext4_inlinedir_to_tree(struct file *dir_file,
 299                         fake.name_len = 1;
 300                         strcpy(fake.name, ".");
 301                         fake.rec_len = ext4_rec_len_to_disk(
 302 -                                               EXT4_DIR_REC_LEN(fake.name_len),
 303 +                                               EXT4_DIR_ENTRY_LEN(&fake),
 304                                                 inline_size);
 305                         ext4_set_de_type(inode->i_sb, &fake, S_IFDIR);
 306                         de = &fake;
 307 @@ -1408,7 +1408,7 @@ int ext4_inlinedir_to_tree(struct file *dir_file,
 308                         fake.name_len = 2;
 309                         strcpy(fake.name, "..");
 310                         fake.rec_len = ext4_rec_len_to_disk(
 311 -                                               EXT4_DIR_REC_LEN(fake.name_len),
 312 +                                               EXT4_DIR_ENTRY_LEN(&fake),
 313                                                 inline_size);
 314                         ext4_set_de_type(inode->i_sb, &fake, S_IFDIR);
 315                         de = &fake;
 316 diff --git a/fs/ext4/namei.c b/fs/ext4/namei.c
 317 index 1537a76..24e1276 100644
 318 --- a/fs/ext4/namei.c
 319 +++ b/fs/ext4/namei.c
 320 @@ -295,7 +295,8 @@ static unsigned dx_get_count(struct dx_entry *entries);
 321  static unsigned dx_get_limit(struct dx_entry *entries);
 322  static void dx_set_count(struct dx_entry *entries, unsigned value);
 323  static void dx_set_limit(struct dx_entry *entries, unsigned value);
 324 -static unsigned dx_root_limit(struct inode *dir, unsigned infosize);
 325 +static inline unsigned dx_root_limit(struct inode *dir,
 326 +               struct ext4_dir_entry_2 *dot_de, unsigned infosize);
 327  static unsigned dx_node_limit(struct inode *dir);
 328  static struct dx_frame *dx_probe(struct ext4_filename *fname,
 329                                  struct inode *dir,
 330 @@ -439,22 +440,23 @@ static struct dx_countlimit *get_dx_countlimit(struct inode *inode,
 331  {
 332         struct ext4_dir_entry *dp;
 333         struct dx_root_info *root;
 334 -       int count_offset;
 335 +       int count_offset, dot_rec_len, dotdot_rec_len;
 336
 337         if (le16_to_cpu(dirent->rec_len) == EXT4_BLOCK_SIZE(inode->i_sb))
 338                 count_offset = 8;
 339 -       else if (le16_to_cpu(dirent->rec_len) == 12) {
 340 -               dp = (struct ext4_dir_entry *)(((void *)dirent) + 12);
 341 +       else {
 342 +               dot_rec_len = le16_to_cpu(dirent->rec_len);
 343 +               dp = (struct ext4_dir_entry *)(((void *)dirent) + dot_rec_len);
 344                 if (le16_to_cpu(dp->rec_len) !=
 345 -                   EXT4_BLOCK_SIZE(inode->i_sb) - 12)
 346 +                   EXT4_BLOCK_SIZE(inode->i_sb) - dot_rec_len)
 347                         return NULL;
 348 -               root = (struct dx_root_info *)(((void *)dp + 12));
 349 +               dotdot_rec_len = EXT4_DIR_ENTRY_LEN((struct ext4_dir_entry_2 *)dp);
 350 +               root = (struct dx_root_info *)(((void *)dp + dotdot_rec_len));
 351                 if (root->reserved_zero ||
 352                     root->info_length != sizeof(struct dx_root_info))
 353                         return NULL;
 354 -               count_offset = 32;
 355 -       } else
 356 -               return NULL;
 357 +               count_offset = 8 + dot_rec_len + dotdot_rec_len;
 358 +       }
 359
 360         if (offset)
 361                 *offset = count_offset;
 362 @@ -559,11 +561,12 @@ ext4_next_entry(struct ext4_dir_entry_2 *p, unsigned long blocksize)
 363   */
 364  struct dx_root_info *dx_get_dx_info(struct ext4_dir_entry_2 *de)
 365  {
 366 +       BUG_ON(de->name_len != 1);
 367         /* get dotdot first */
 368 -       de = (struct ext4_dir_entry_2 *)((char *)de + EXT4_DIR_REC_LEN(1));
 369 +       de = (struct ext4_dir_entry_2 *)((char *)de + EXT4_DIR_ENTRY_LEN(de));
 370
 371         /* dx root info is after dotdot entry */
 372 -       de = (struct ext4_dir_entry_2 *)((char *)de + EXT4_DIR_REC_LEN(2));
 373 +       de = (struct ext4_dir_entry_2 *)((char *)de + EXT4_DIR_ENTRY_LEN(de));
 374
 375         return (struct dx_root_info *)de;
 376  }
 377 @@ -608,10 +611,16 @@ static inline void dx_set_limit(struct dx_entry *entries, unsigned value)
 378         ((struct dx_countlimit *) entries)->limit = cpu_to_le16(value);
 379  }
 380
 381 -static inline unsigned dx_root_limit(struct inode *dir, unsigned infosize)
 382 +static inline unsigned dx_root_limit(struct inode *dir,
 383 +               struct ext4_dir_entry_2 *dot_de, unsigned infosize)
 384  {
 385 -       unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(1) -
 386 -               EXT4_DIR_REC_LEN(2) - infosize;
 387 +       struct ext4_dir_entry_2 *dotdot_de;
 388 +       unsigned entry_space;
 389 +
 390 +       BUG_ON(dot_de->name_len != 1);
 391 +       dotdot_de = ext4_next_entry(dot_de, dir->i_sb->s_blocksize);
 392 +       entry_space = dir->i_sb->s_blocksize - EXT4_DIR_ENTRY_LEN(dot_de) -
 393 +                        EXT4_DIR_ENTRY_LEN(dotdot_de) - infosize;
 394
 395         if (ext4_has_metadata_csum(dir->i_sb))
 396                 entry_space -= sizeof(struct dx_tail);
 397 @@ -731,7 +740,7 @@ static struct stats dx_show_leaf(struct inode *dir,
 398                                        (unsigned) ((char *) de - base));
 399  #endif
 400                         }
 401 -                       space += EXT4_DIR_REC_LEN(de->name_len);
 402 +                       space += EXT4_DIR_ENTRY_LEN(de);
 403                         names++;
 404                 }
 405                 de = ext4_next_entry(de, size);
 406 @@ -840,11 +849,14 @@ dx_probe(struct ext4_filename *fname, struct inode *dir,
 407
 408         entries = (struct dx_entry *)(((char *)info) + info->info_length);
 409
 410 -       if (dx_get_limit(entries) != dx_root_limit(dir,
 411 -                                                  info->info_length)) {
 412 +       if (dx_get_limit(entries) !=
 413 +           dx_root_limit(dir, (struct ext4_dir_entry_2 *)frame->bh->b_data,
 414 +                         info->info_length)) {
 415                 ext4_warning_inode(dir, "dx entry: limit %u != root limit %u",
 416                                    dx_get_limit(entries),
 417 -                                  dx_root_limit(dir, info->info_length));
 418 +                                  dx_root_limit(dir,
 419 +                                         (struct ext4_dir_entry_2 *)frame->bh->b_data,
 420 +                                         info->info_length));
 421                 goto fail;
 422         }
 423
 424 @@ -1851,7 +1863,7 @@ dx_move_dirents(char *from, char *to, struct dx_map_entry *map, int count,
 425         while (count--) {
 426                 struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *)
 427                                                 (from + (map->offs<<2));
 428 -               rec_len = EXT4_DIR_REC_LEN(de->name_len);
 429 +               rec_len = EXT4_DIR_ENTRY_LEN(de);
 430                 memcpy (to, de, rec_len);
 431                 ((struct ext4_dir_entry_2 *) to)->rec_len =
 432                                 ext4_rec_len_to_disk(rec_len, blocksize);
 433 @@ -1882,7 +1894,7 @@ static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize)
 434         while ((char*)de < base + blocksize) {
 435                 next = ext4_next_entry(de, blocksize);
 436                 if (de->inode && de->name_len) {
 437 -                       rec_len = EXT4_DIR_REC_LEN(de->name_len);
 438 +                       rec_len = EXT4_DIR_ENTRY_LEN(de);
 439                         if (de > to)
 440                                 memmove(to, de, rec_len);
 441                         to->rec_len = ext4_rec_len_to_disk(rec_len, blocksize);
 442 @@ -2023,14 +2035,16 @@ int ext4_find_dest_de(struct inode *dir, struct inode *inode,
 443                       struct buffer_head *bh,
 444                       void *buf, int buf_size,
 445                       struct ext4_filename *fname,
 446 -                     struct ext4_dir_entry_2 **dest_de)
 447 +                     struct ext4_dir_entry_2 **dest_de, int *dlen)
 448  {
 449         struct ext4_dir_entry_2 *de;
 450 -       unsigned short reclen = EXT4_DIR_REC_LEN(fname_len(fname));
 451 +       unsigned short reclen = EXT4_DIR_REC_LEN(fname_len(fname)) +
 452 +                                                 (dlen ? *dlen : 0);
 453         int nlen, rlen;
 454         unsigned int offset = 0;
 455         char *top;
 456
 457 +       dlen ? *dlen = 0 : 0; /* default set to 0 */
 458         de = (struct ext4_dir_entry_2 *)buf;
 459         top = buf + buf_size - reclen;
 460         while ((char *) de <= top) {
 461 @@ -2039,10 +2053,26 @@ int ext4_find_dest_de(struct inode *dir, struct inode *inode,
 462                         return -EFSCORRUPTED;
 463                 if (ext4_match(dir, fname, de))
 464                         return -EEXIST;
 465 -               nlen = EXT4_DIR_REC_LEN(de->name_len);
 466 +               nlen = EXT4_DIR_ENTRY_LEN(de);
 467                 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
 468                 if ((de->inode ? rlen - nlen : rlen) >= reclen)
 469                         break;
 470 +               /* Then for dotdot entries, check for the smaller space
 471 +                * required for just the entry, no FID */
 472 +               if (fname_len(fname) == 2 && memcmp(fname_name(fname), "..", 2) == 0) {
 473 +                       if ((de->inode ? rlen - nlen : rlen) >=
 474 +                           EXT4_DIR_REC_LEN(fname_len(fname))) {
 475 +                               /* set dlen=1 to indicate not
 476 +                                * enough space store fid */
 477 +                               dlen ? *dlen = 1 : 0;
 478 +                               break;
 479 +                       }
 480 +                       /* The new ".." entry must be written over the
 481 +                        * previous ".." entry, which is the first
 482 +                        * entry traversed by this scan. If it doesn't
 483 +                        * fit, something is badly wrong, so -EIO. */
 484 +                       return -EIO;
 485 +               }
 486                 de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
 487                 offset += rlen;
 488         }
 489 @@ -2056,12 +2086,12 @@ int ext4_find_dest_de(struct inode *dir, struct inode *inode,
 490  void ext4_insert_dentry(struct inode *inode,
 491                         struct ext4_dir_entry_2 *de,
 492                         int buf_size,
 493 -                       struct ext4_filename *fname)
 494 +                       struct ext4_filename *fname, void *data)
 495  {
 496
 497         int nlen, rlen;
 498
 499 -       nlen = EXT4_DIR_REC_LEN(de->name_len);
 500 +       nlen = EXT4_DIR_ENTRY_LEN(de);
 501         rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
 502         if (de->inode) {
 503                 struct ext4_dir_entry_2 *de1 =
 504 @@ -2075,6 +2105,11 @@ void ext4_insert_dentry(struct inode *inode,
 505         ext4_set_de_type(inode->i_sb, de, inode->i_mode);
 506         de->name_len = fname_len(fname);
 507         memcpy(de->name, fname_name(fname), fname_len(fname));
 508 +       if (data) {
 509 +               de->name[fname_len(fname)] = 0;
 510 +               memcpy(&de->name[fname_len(fname) + 1], data, *(char *)data);
 511 +               de->file_type |= EXT4_DIRENT_LUFID;
 512 +       }
 513  }
 514
 515  /*
 516 @@ -2092,14 +2127,19 @@ static int add_dirent_to_buf(handle_t *handle, struct ext4_filename *fname,
 517  {
 518         unsigned int    blocksize = dir->i_sb->s_blocksize;
 519         int             csum_size = 0;
 520 -       int             err, err2;
 521 +       int             err, err2, dlen = 0;
 522 +       unsigned char   *data;
 523
 524 +       data = ext4_dentry_get_data(inode->i_sb, (struct ext4_dentry_param *)
 525 +                                               EXT4_I(inode)->i_dirdata);
 526         if (ext4_has_metadata_csum(inode->i_sb))
 527                 csum_size = sizeof(struct ext4_dir_entry_tail);
 528
 529         if (!de) {
 530 +               if (data)
 531 +                       dlen = (*data) + 1;
 532                 err = ext4_find_dest_de(dir, inode, bh, bh->b_data,
 533 -                                       blocksize - csum_size, fname, &de);
 534 +                                       blocksize - csum_size, fname, &de, &dlen);
 535                 if (err)
 536                         return err;
 537         }
 538 @@ -2111,7 +2151,10 @@ static int add_dirent_to_buf(handle_t *handle, struct ext4_filename *fname,
 539         }
 540
 541         /* By now the buffer is marked for journaling */
 542 -       ext4_insert_dentry(inode, de, blocksize, fname);
 543 +       /* If writing the short form of "dotdot", don't add the data section */
 544 +       if (dlen == 1)
 545 +               data = NULL;
 546 +       ext4_insert_dentry(inode, de, blocksize, fname, data);
 547
 548         /*
 549          * XXX shouldn't update any times until successful
 550 @@ -2217,7 +2260,8 @@ static int make_indexed_dir(handle_t *handle, struct ext4_filename *fname,
 551
 552         dx_set_block(entries, 1);
 553         dx_set_count(entries, 1);
 554 -       dx_set_limit(entries, dx_root_limit(dir, sizeof(*dx_info)));
 555 +       dx_set_limit(entries, dx_root_limit(dir,
 556 +                                        dot_de, sizeof(*dx_info)));
 557
 558         /* Initialize as for dx_probe */
 559         fname->hinfo.hash_version = dx_info->hash_version;
 560 @@ -2267,6 +2311,8 @@ static int ext4_update_dotdot(handle_t *handle, struct dentry *dentry,
 561         struct buffer_head *dir_block;
 562         struct ext4_dir_entry_2 *de;
 563         int len, journal = 0, err = 0;
 564 +       int dlen = 0;
 565 +       char *data;
 566
 567         if (IS_ERR(handle))
 568                 return PTR_ERR(handle);
 569 @@ -2292,11 +2338,16 @@ static int ext4_update_dotdot(handle_t *handle, struct dentry *dentry,
 570                         goto out_journal;
 571
 572                 journal = 1;
 573 -               de->rec_len = cpu_to_le16(EXT4_DIR_REC_LEN(1));
 574 +               de->rec_len = cpu_to_le16(EXT4_DIR_ENTRY_LEN(de));
 575         }
 576
 577 -       len -= EXT4_DIR_REC_LEN(1);
 578 -       assert(len == 0 || len >= EXT4_DIR_REC_LEN(2));
 579 +       len -= EXT4_DIR_ENTRY_LEN(de);
 580 +       data = ext4_dentry_get_data(dir->i_sb,
 581 +                       (struct ext4_dentry_param *)dentry->d_fsdata);
 582 +       if (data)
 583 +               dlen = *data + 1;
 584 +       assert(len == 0 || len >= EXT4_DIR_REC_LEN(2 + dlen));
 585 +
 586         de = (struct ext4_dir_entry_2 *)
 587                         ((char *) de + le16_to_cpu(de->rec_len));
 588         if (!journal) {
 589 @@ -2313,7 +2364,12 @@ static int ext4_update_dotdot(handle_t *handle, struct dentry *dentry,
 590                 assert(le16_to_cpu(de->rec_len) >= EXT4_DIR_REC_LEN(2));
 591         de->name_len = 2;
 592         strcpy(de->name, "..");
 593 -       ext4_set_de_type(dir->i_sb, de, S_IFDIR);
 594 +       if (data != NULL && ext4_get_dirent_data_len(de) >= dlen) {
 595 +               de->name[2] = 0;
 596 +               memcpy(&de->name[2 + 1], data, *data);
 597 +               ext4_set_de_type(dir->i_sb, de, S_IFDIR);
 598 +               de->file_type |= EXT4_DIRENT_LUFID;
 599 +       }
 600
 601  out_journal:
 602         if (journal) {
 603 @@ -2351,6 +2407,7 @@ static int ext4_add_entry(handle_t *handle, struct dentry *dentry,
 604         ext4_lblk_t block, blocks;
 605         int     csum_size = 0;
 606
 607 +       EXT4_I(inode)->i_dirdata = dentry->d_fsdata;
 608         if (ext4_has_metadata_csum(inode->i_sb))
 609                 csum_size = sizeof(struct ext4_dir_entry_tail);
 610
 611 @@ -2918,37 +2975,70 @@ err_unlock_inode:
 612         return err;
 613  }
 614
 615 +struct tp_block {
 616 +       struct inode *inode;
 617 +       void *data1;
 618 +       void *data2;
 619 +};
 620 +
 621  struct ext4_dir_entry_2 *ext4_init_dot_dotdot(struct inode *inode,
 622                           struct ext4_dir_entry_2 *de,
 623                           int blocksize, int csum_size,
 624                           unsigned int parent_ino, int dotdot_real_len)
 625  {
 626 +       void *data1 = NULL, *data2 = NULL;
 627 +       int dot_reclen = 0;
 628 +
 629 +       if (dotdot_real_len == 10) {
 630 +               struct tp_block *tpb = (struct tp_block *)inode;
 631 +               data1 = tpb->data1;
 632 +               data2 = tpb->data2;
 633 +               inode = tpb->inode;
 634 +               dotdot_real_len = 0;
 635 +       }
 636         de->inode = cpu_to_le32(inode->i_ino);
 637         de->name_len = 1;
 638 -       de->rec_len = ext4_rec_len_to_disk(EXT4_DIR_REC_LEN(de->name_len),
 639 -                                          blocksize);
 640         strcpy(de->name, ".");
 641         ext4_set_de_type(inode->i_sb, de, S_IFDIR);
 642
 643 +       /* get packed fid data*/
 644 +       data1 = ext4_dentry_get_data(inode->i_sb,
 645 +                               (struct ext4_dentry_param *) data1);
 646 +       if (data1) {
 647 +               de->name[1] = 0;
 648 +               memcpy(&de->name[2], data1, *(char *) data1);
 649 +               de->file_type |= EXT4_DIRENT_LUFID;
 650 +       }
 651 +       de->rec_len = cpu_to_le16(EXT4_DIR_ENTRY_LEN(de));
 652 +       dot_reclen = cpu_to_le16(de->rec_len);
 653         de = ext4_next_entry(de, blocksize);
 654         de->inode = cpu_to_le32(parent_ino);
 655         de->name_len = 2;
 656 +       strcpy(de->name, "..");
 657 +       ext4_set_de_type(inode->i_sb, de, S_IFDIR);
 658 +       data2 = ext4_dentry_get_data(inode->i_sb,
 659 +                       (struct ext4_dentry_param *) data2);
 660 +       if (data2) {
 661 +               de->name[2] = 0;
 662 +               memcpy(&de->name[3], data2, *(char *) data2);
 663 +               de->file_type |= EXT4_DIRENT_LUFID;
 664 +       }
 665         if (!dotdot_real_len)
 666                 de->rec_len = ext4_rec_len_to_disk(blocksize -
 667 -                                       (csum_size + EXT4_DIR_REC_LEN(1)),
 668 +                                       (csum_size + dot_reclen),
 669                                         blocksize);
 670         else
 671                 de->rec_len = ext4_rec_len_to_disk(
 672 -                               EXT4_DIR_REC_LEN(de->name_len), blocksize);
 673 -       strcpy(de->name, "..");
 674 -       ext4_set_de_type(inode->i_sb, de, S_IFDIR);
 675 +                               EXT4_DIR_ENTRY_LEN(de), blocksize);
 676
 677         return ext4_next_entry(de, blocksize);
 678  }
 679
 680  int ext4_init_new_dir(handle_t *handle, struct inode *dir,
 681 -                            struct inode *inode)
 682 +                            struct inode *inode,
 683 +                            const void *data1, const void *data2)
 684  {
 685 +       struct tp_block param;
 686         struct buffer_head *dir_block = NULL;
 687         struct ext4_dir_entry_2 *de;
 688         ext4_lblk_t block = 0;
 689 @@ -2972,7 +3062,11 @@ int ext4_init_new_dir(handle_t *handle, struct inode *dir,
 690         if (IS_ERR(dir_block))
 691                 return PTR_ERR(dir_block);
 692         de = (struct ext4_dir_entry_2 *)dir_block->b_data;
 693 -       ext4_init_dot_dotdot(inode, de, blocksize, csum_size, dir->i_ino, 0);
 694 +       param.inode = inode;
 695 +       param.data1 = (void *)data1;
 696 +       param.data2 = (void *)data2;
 697 +       ext4_init_dot_dotdot((struct inode *)(&param), de, blocksize,
 698 +                            csum_size, dir->i_ino, 10);
 699         set_nlink(inode, 2);
 700         if (csum_size)
 701                 ext4_initialize_dirent_tail(dir_block, blocksize);
 702 @@ -2987,6 +3081,29 @@ out:
 703         return err;
 704  }
 705
 706 +/* Initialize @inode as a subdirectory of @dir, and add the
 707 + * "." and ".." entries into the first directory block. */
 708 +int ext4_add_dot_dotdot(handle_t *handle, struct inode *dir,
 709 +                       struct inode *inode,
 710 +                       const void *data1, const void *data2)
 711 +{
 712 +       int rc;
 713 +
 714 +       if (IS_ERR(handle))
 715 +               return PTR_ERR(handle);
 716 +
 717 +       if (IS_DIRSYNC(dir))
 718 +               ext4_handle_sync(handle);
 719 +
 720 +       inode->i_op = &ext4_dir_inode_operations;
 721 +       inode->i_fop = &ext4_dir_operations;
 722 +       rc = ext4_init_new_dir(handle, dir, inode, data1, data2);
 723 +       if (!rc)
 724 +               rc = ext4_mark_inode_dirty(handle, inode);
 725 +       return rc;
 726 +}
 727 +EXPORT_SYMBOL(ext4_add_dot_dotdot);
 728 +
 729  static int ext4_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
 730  {
 731         handle_t *handle;
 732 @@ -3013,7 +3130,7 @@ retry:
 733
 734         inode->i_op = &ext4_dir_inode_operations;
 735         inode->i_fop = &ext4_dir_operations;
 736 -       err = ext4_init_new_dir(handle, dir, inode);
 737 +       err = ext4_init_new_dir(handle, dir, inode, NULL, NULL);
 738         if (err)
 739                 goto out_clear_inode;
 740         err = ext4_mark_inode_dirty(handle, inode);
 741 diff --git a/fs/ext4/super.c b/fs/ext4/super.c
 742 index 6c33a10..59b87b4 100644
 743 --- a/fs/ext4/super.c
 744 +++ b/fs/ext4/super.c
 745 @@ -1719,7 +1719,7 @@ enum {
 746         Opt_inlinecrypt,
 747         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
 748         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_jqfmt_vfsv1, Opt_quota,
 749 -       Opt_noquota, Opt_barrier, Opt_nobarrier, Opt_err,
 750 +       Opt_noquota, Opt_barrier, Opt_nobarrier, Opt_err, Opt_dirdata,
 751         Opt_usrquota, Opt_grpquota, Opt_prjquota, Opt_i_version,
 752         Opt_dax, Opt_dax_always, Opt_dax_inode, Opt_dax_never,
 753         Opt_stripe, Opt_delalloc, Opt_nodelalloc, Opt_warn_on_error,
 754 @@ -1803,6 +1803,7 @@ static const match_table_t tokens = {
 755         {Opt_nolazytime, "nolazytime"},
 756         {Opt_debug_want_extra_isize, "debug_want_extra_isize=%u"},
 757         {Opt_nodelalloc, "nodelalloc"},
 758 +       {Opt_dirdata, "dirdata"},
 759         {Opt_removed, "mblk_io_submit"},
 760         {Opt_removed, "nomblk_io_submit"},
 761         {Opt_block_validity, "block_validity"},
 762 @@ -2043,6 +2044,7 @@ static const struct mount_opts {
 763         {Opt_usrjquota, 0, MOPT_Q | MOPT_STRING},
 764         {Opt_grpjquota, 0, MOPT_Q | MOPT_STRING},
 765         {Opt_offusrjquota, 0, MOPT_Q},
 766 +       {Opt_dirdata, EXT4_MOUNT_DIRDATA, MOPT_SET},
 767         {Opt_offgrpjquota, 0, MOPT_Q},
 768         {Opt_jqfmt_vfsold, QFMT_VFS_OLD, MOPT_QFMT},
 769         {Opt_jqfmt_vfsv0, QFMT_VFS_V0, MOPT_QFMT},
 770 --
 771 2.33.0
 772