ldiskfs/kernel_patches/patches/rhel9.1/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         |   9 +-
  26  fs/ext4/ext4.h        | 107 ++++++++++++++++--
  27  fs/ext4/fast_commit.c |   2 +-
  28  fs/ext4/inline.c      |   8 +-
  29  fs/ext4/namei.c       | 250 ++++++++++++++++++++++++++++++++----------
  30  fs/ext4/super.c       |   4 +-
  31  6 files changed, 305 insertions(+), 75 deletions(-)
  32
  33 diff --git a/fs/ext4/dir.c b/fs/ext4/dir.c
  34 index a6bb86f..edd0540 100644
  35 --- a/fs/ext4/dir.c
  36 +++ b/fs/ext4/dir.c
  37 @@ -465,12 +465,17 @@ int ext4_htree_store_dirent(struct file *dir_file, __u32 hash,
  38         struct fname *fname, *new_fn;
  39         struct dir_private_info *info;
  40         int len;
  41 +       int extra_data = 0;
  42
  43         info = dir_file->private_data;
  44         p = &info->root.rb_node;
  45
  46         /* Create and allocate the fname structure */
  47 -       len = sizeof(struct fname) + ent_name->len + 1;
  48 +       if (dirent->file_type & EXT4_DIRENT_LUFID)
  49 +               extra_data = ext4_get_dirent_data_len(dirent);
  50 +
  51 +       len = sizeof(struct fname) + ent_name->len + extra_data + 1;
  52 +
  53         new_fn = kzalloc(len, GFP_KERNEL);
  54         if (!new_fn)
  55                 return -ENOMEM;
  56 @@ -479,7 +484,7 @@ int ext4_htree_store_dirent(struct file *dir_file, __u32 hash,
  57         new_fn->inode = le32_to_cpu(dirent->inode);
  58         new_fn->name_len = ent_name->len;
  59         new_fn->file_type = dirent->file_type;
  60 -       memcpy(new_fn->name, ent_name->name, ent_name->len);
  61 +       memcpy(new_fn->name, ent_name->name, ent_name->len + extra_data);
  62
  63         while (*p) {
  64                 parent = *p;
  65 diff --git a/fs/ext4/ext4.h b/fs/ext4/ext4.h
  66 index f4a1557..85556ce 100644
  67 --- a/fs/ext4/ext4.h
  68 +++ b/fs/ext4/ext4.h
  69 @@ -1173,6 +1173,7 @@ struct ext4_inode_info {
  70         __u32 i_csum_seed;
  71
  72         kprojid_t i_projid;
  73 +       void *i_dirdata;
  74  };
  75
  76  /*
  77 @@ -1194,6 +1195,7 @@ struct ext4_inode_info {
  78   * Mount flags set via mount options or defaults
  79   */
  80  #define EXT4_MOUNT_NO_MBCACHE          0x00001 /* Do not use mbcache */
  81 +#define EXT4_MOUNT_DIRDATA             0x00002 /* Data in directory entries */
  82  #define EXT4_MOUNT_GRPID               0x00004 /* Create files with directory's group */
  83  #define EXT4_MOUNT_DEBUG               0x00008 /* Some debugging messages */
  84  #define EXT4_MOUNT_ERRORS_CONT         0x00010 /* Continue on errors */
  85 @@ -2162,6 +2164,7 @@ EXT4_FEATURE_INCOMPAT_FUNCS(casefold,             CASEFOLD)
  86                                          EXT4_FEATURE_INCOMPAT_FLEX_BG| \
  87                                          EXT4_FEATURE_INCOMPAT_EA_INODE| \
  88                                          EXT4_FEATURE_INCOMPAT_MMP | \
  89 +                                        EXT4_FEATURE_INCOMPAT_DIRDATA| \
  90                                          EXT4_FEATURE_INCOMPAT_INLINE_DATA | \
  91                                          EXT4_FEATURE_INCOMPAT_ENCRYPT | \
  92                                          EXT4_FEATURE_INCOMPAT_CASEFOLD | \
  93 @@ -2373,6 +2376,42 @@ struct ext4_dir_entry_tail {
  94  #define EXT4_FT_SYMLINK                7
  95
  96  #define EXT4_FT_MAX            8
  97 +#define EXT4_FT_MASK           0xf
  98 +
  99 +#if EXT4_FT_MAX > EXT4_FT_MASK
 100 +#error "conflicting EXT4_FT_MAX and EXT4_FT_MASK"
 101 +#endif
 102 +
 103 +/*
 104 + * d_type has 4 unused bits, so it can hold four types data. these different
 105 + * type of data (e.g. lustre data, high 32 bits of 64-bit inode number) can be
 106 + * stored, in flag order, after file-name in ext4 dirent.
 107 +*/
 108 +/*
 109 + * this flag is added to d_type if ext4 dirent has extra data after
 110 + * filename. this data length is variable and length is stored in first byte
 111 + * of data. data start after filename NUL byte.
 112 + * This is used by Lustre FS.
 113 +  */
 114 +#define EXT4_DIRENT_LUFID              0x10
 115 +
 116 +#define EXT4_LUFID_MAGIC    0xAD200907UL
 117 +struct ext4_dentry_param {
 118 +       __u32  edp_magic;       /* EXT4_LUFID_MAGIC */
 119 +       char   edp_len;         /* size of edp_data in bytes */
 120 +       char   edp_data[0];     /* packed array of data */
 121 +} __packed;
 122 +
 123 +static inline unsigned char *ext4_dentry_get_data(struct super_block *sb,
 124 +                                                 struct ext4_dentry_param *p)
 125 +{
 126 +       if (!ext4_has_feature_dirdata(sb))
 127 +               return NULL;
 128 +       if (p && p->edp_magic == EXT4_LUFID_MAGIC)
 129 +               return &p->edp_len;
 130 +       else
 131 +               return NULL;
 132 +}
 133
 134  #define EXT4_FT_DIR_CSUM       0xDE
 135
 136 @@ -2384,6 +2423,17 @@ struct ext4_dir_entry_tail {
 137  #define EXT4_DIR_PAD                   4
 138  #define EXT4_DIR_ROUND                 (EXT4_DIR_PAD - 1)
 139  #define EXT4_MAX_REC_LEN               ((1<<16)-1)
 140 +#define EXT4_DIR_REC_LEN_(name_len, i_dir) \
 141 +       ext4_dir_rec_len((name_len), (i_dir))
 142 +#define EXT4_DIR_ENTRY_LEN_(de, i_dir) \
 143 +       (EXT4_DIR_REC_LEN_((de)->name_len + ext4_get_dirent_data_len(de), \
 144 +               (i_dir)))
 145 +/* ldiskfs */
 146 +#define EXT4_DIR_REC_LEN(name_len, i_dir)      EXT4_DIR_REC_LEN_((name_len), (i_dir))
 147 +#define EXT4_DIR_ENTRY_LEN(de, i_dir)          EXT4_DIR_ENTRY_LEN_((de), (i_dir))
 148 +/* lustre osd_handler compat -- ifdef LDISKFS_DIR_REC_LEN_WITH_DIR */
 149 +#define EXT4_DIR_REC_LEN_WITH_DIR              1
 150 +#define __EXT4_DIR_REC_LEN(name_len)           EXT4_DIR_REC_LEN_((name_len), NULL)
 151
 152  /*
 153   * The rec_len is dependent on the type of directory. Directories that are
 154 @@ -2391,10 +2441,10 @@ struct ext4_dir_entry_tail {
 155   * ext4_extended_dir_entry_2. For all entries related to '.' or '..' you should
 156   * pass NULL for dir, as those entries do not use the extra fields.
 157   */
 158 -static inline unsigned int ext4_dir_rec_len(__u8 name_len,
 159 +static inline unsigned int ext4_dir_rec_len(__u32 name_len,
 160                                                 const struct inode *dir)
 161  {
 162 -       int rec_len = (name_len + 8 + EXT4_DIR_ROUND);
 163 +       __u32 rec_len = (name_len + 8 + EXT4_DIR_ROUND);
 164
 165         if (dir && ext4_hash_in_dirent(dir))
 166                 rec_len += sizeof(struct ext4_dir_entry_hash);
 167 @@ -2867,11 +2917,13 @@ extern int ext4_find_dest_de(struct inode *dir, struct inode *inode,
 168                              struct buffer_head *bh,
 169                              void *buf, int buf_size,
 170                              struct ext4_filename *fname,
 171 -                            struct ext4_dir_entry_2 **dest_de);
 172 +                            struct ext4_dir_entry_2 **dest_de,
 173 +                            int dlen);
 174  void ext4_insert_dentry(struct inode *dir, struct inode *inode,
 175                         struct ext4_dir_entry_2 *de,
 176                         int buf_size,
 177 -                       struct ext4_filename *fname);
 178 +                       struct ext4_filename *fname,
 179 +                       void *data);
 180  static inline void ext4_update_dx_flag(struct inode *inode)
 181  {
 182         if (!ext4_has_feature_dir_index(inode->i_sb) &&
 183 @@ -2887,10 +2939,17 @@ static const unsigned char ext4_filetype_table[] = {
 184
 185  static inline  unsigned char get_dtype(struct super_block *sb, int filetype)
 186  {
 187 -       if (!ext4_has_feature_filetype(sb) || filetype >= EXT4_FT_MAX)
 188 +       int fl_index = filetype & EXT4_FT_MASK;
 189 +
 190 +       if (!ext4_has_feature_filetype(sb) || fl_index >= EXT4_FT_MAX)
 191                 return DT_UNKNOWN;
 192
 193 -       return ext4_filetype_table[filetype];
 194 +       if (!test_opt(sb, DIRDATA))
 195 +               return ext4_filetype_table[fl_index];
 196 +
 197 +       return (ext4_filetype_table[fl_index]) |
 198 +               (filetype & EXT4_DIRENT_LUFID);
 199 +
 200  }
 201  extern int ext4_check_all_de(struct inode *dir, struct buffer_head *bh,
 202                              void *buf, int buf_size);
 203 @@ -3093,9 +3152,13 @@ extern int ext4_ind_migrate(struct inode *inode);
 204
 205  /* namei.c */
 206  extern int ext4_init_new_dir(handle_t *handle, struct inode *dir,
 207 -                            struct inode *inode);
 208 +                            struct inode *inode,
 209 +                            const void *data1, const void *data2);
 210  extern int ext4_dirblock_csum_verify(struct inode *inode,
 211                                      struct buffer_head *bh);
 212 +extern int ext4_add_dot_dotdot(handle_t *handle, struct inode *dir,
 213 +                              struct inode *inode,
 214 +                              const void *data1, const void *data2);
 215  extern int ext4_htree_fill_tree(struct file *dir_file, __u32 start_hash,
 216                                 __u32 start_minor_hash, __u32 *next_hash);
 217  extern struct inode *ext4_create_inode(handle_t *handle,
 218 @@ -3924,6 +3987,36 @@ static inline int ext4_buffer_uptodate(struct buffer_head *bh)
 219         return buffer_uptodate(bh);
 220  }
 221
 222 +/*
 223 + * Compute the total directory entry data length.
 224 + * This includes the filename and an implicit NUL terminator (always present),
 225 + * and optional extensions.  Each extension has a bit set in the high 4 bits of
 226 + * de->file_type, and the extension length is the first byte in each entry.
 227 + */
 228 +static inline int ext4_get_dirent_data_len(struct ext4_dir_entry_2 *de)
 229 +{
 230 +       char *len = de->name + de->name_len + 1 /* NUL terminator */;
 231 +       int dlen = 0;
 232 +       __u8 extra_data_flags = (de->file_type & ~EXT4_FT_MASK) >> 4;
 233 +       struct ext4_dir_entry_tail *t = (struct ext4_dir_entry_tail *)de;
 234 +
 235 +       if (!t->det_reserved_zero1 &&
 236 +           le16_to_cpu(t->det_rec_len) ==
 237 +               sizeof(struct ext4_dir_entry_tail) &&
 238 +           !t->det_reserved_zero2 &&
 239 +           t->det_reserved_ft == EXT4_FT_DIR_CSUM)
 240 +               return 0;
 241 +
 242 +       while (extra_data_flags) {
 243 +               if (extra_data_flags & 1) {
 244 +                       dlen += *len + (dlen == 0);
 245 +                       len += *len;
 246 +               }
 247 +               extra_data_flags >>= 1;
 248 +       }
 249 +       return dlen;
 250 +}
 251 +
 252  #endif /* __KERNEL__ */
 253
 254  #define EFSBADCRC      EBADMSG         /* Bad CRC detected */
 255 diff --git a/fs/ext4/fast_commit.c b/fs/ext4/fast_commit.c
 256 index 3d72565..2331515 100644
 257 --- a/fs/ext4/fast_commit.c
 258 +++ b/fs/ext4/fast_commit.c
 259 @@ -1646,7 +1646,7 @@ static int ext4_fc_replay_create(struct super_block *sb, struct ext4_fc_tl *tl,
 260                         jbd_debug(1, "Dir %d not found.", darg.ino);
 261                         goto out;
 262                 }
 263 -               ret = ext4_init_new_dir(NULL, dir, inode);
 264 +               ret = ext4_init_new_dir(NULL, dir, inode, NULL, NULL);
 265                 iput(dir);
 266                 if (ret) {
 267                         ret = 0;
 268 diff --git a/fs/ext4/inline.c b/fs/ext4/inline.c
 269 index e9ef5cf..a23fd25 100644
 270 --- a/fs/ext4/inline.c
 271 +++ b/fs/ext4/inline.c
 272 @@ -1031,7 +1031,7 @@ static int ext4_add_dirent_to_inline(handle_t *handle,
 273         struct ext4_dir_entry_2 *de;
 274
 275         err = ext4_find_dest_de(dir, inode, iloc->bh, inline_start,
 276 -                               inline_size, fname, &de);
 277 +                               inline_size, fname, &de, 0);
 278         if (err)
 279                 return err;
 280
 281 @@ -1040,7 +1040,7 @@ static int ext4_add_dirent_to_inline(handle_t *handle,
 282                                             EXT4_JTR_NONE);
 283         if (err)
 284                 return err;
 285 -       ext4_insert_dentry(dir, inode, de, inline_size, fname);
 286 +       ext4_insert_dentry(dir, inode, de, inline_size, fname, NULL);
 287
 288         ext4_show_inline_dir(dir, iloc->bh, inline_start, inline_size);
 289
 290 @@ -1398,7 +1398,7 @@ int ext4_inlinedir_to_tree(struct file *dir_file,
 291                         fake.name_len = 1;
 292                         strcpy(fake.name, ".");
 293                         fake.rec_len = ext4_rec_len_to_disk(
 294 -                                         ext4_dir_rec_len(fake.name_len, NULL),
 295 +                                         EXT4_DIR_ENTRY_LEN(&fake, NULL),
 296                                           inline_size);
 297                         ext4_set_de_type(inode->i_sb, &fake, S_IFDIR);
 298                         de = &fake;
 299 @@ -1408,7 +1408,7 @@ int ext4_inlinedir_to_tree(struct file *dir_file,
 300                         fake.name_len = 2;
 301                         strcpy(fake.name, "..");
 302                         fake.rec_len = ext4_rec_len_to_disk(
 303 -                                         ext4_dir_rec_len(fake.name_len, NULL),
 304 +                                         EXT4_DIR_ENTRY_LEN(&fake, NULL),
 305                                           inline_size);
 306                         ext4_set_de_type(inode->i_sb, &fake, S_IFDIR);
 307                         de = &fake;
 308 diff --git a/fs/ext4/namei.c b/fs/ext4/namei.c
 309 index 4d932a2..ee1a058 100644
 310 --- a/fs/ext4/namei.c
 311 +++ b/fs/ext4/namei.c
 312 @@ -262,13 +262,14 @@ static unsigned dx_get_count(struct dx_entry *entries);
 313  static unsigned dx_get_limit(struct dx_entry *entries);
 314  static void dx_set_count(struct dx_entry *entries, unsigned value);
 315  static void dx_set_limit(struct dx_entry *entries, unsigned value);
 316 -static unsigned dx_root_limit(struct inode *dir, unsigned infosize);
 317 +static inline unsigned dx_root_limit(struct inode *dir,
 318 +               struct ext4_dir_entry_2 *dot_de, unsigned infosize);
 319  static unsigned dx_node_limit(struct inode *dir);
 320  static struct dx_frame *dx_probe(struct ext4_filename *fname,
 321                                  struct inode *dir,
 322                                  struct dx_hash_info *hinfo,
 323                                  struct dx_frame *frame);
 324 -static void dx_release(struct dx_frame *frames);
 325 +static void dx_release(struct dx_frame *frames, struct inode *dir);
 326  static int dx_make_map(struct inode *dir, struct buffer_head *bh,
 327                        struct dx_hash_info *hinfo,
 328                        struct dx_map_entry *map_tail);
 329 @@ -408,22 +409,23 @@ static struct dx_countlimit *get_dx_countlimit(struct inode *inode,
 330  {
 331         struct ext4_dir_entry *dp;
 332         struct dx_root_info *root;
 333 -       int count_offset;
 334 +       int count_offset, dot_rec_len, dotdot_rec_len;
 335
 336         if (le16_to_cpu(dirent->rec_len) == EXT4_BLOCK_SIZE(inode->i_sb))
 337                 count_offset = 8;
 338 -       else if (le16_to_cpu(dirent->rec_len) == 12) {
 339 -               dp = (struct ext4_dir_entry *)(((void *)dirent) + 12);
 340 +       else {
 341 +               dot_rec_len = le16_to_cpu(dirent->rec_len);
 342 +               dp = (struct ext4_dir_entry *)(((void *)dirent) + dot_rec_len);
 343                 if (le16_to_cpu(dp->rec_len) !=
 344 -                   EXT4_BLOCK_SIZE(inode->i_sb) - 12)
 345 +                   EXT4_BLOCK_SIZE(inode->i_sb) - dot_rec_len)
 346                         return NULL;
 347 -               root = (struct dx_root_info *)(((void *)dp + 12));
 348 +               dotdot_rec_len = EXT4_DIR_ENTRY_LEN((struct ext4_dir_entry_2 *)dp, NULL);
 349 +               root = (struct dx_root_info *)(((void *)dp + dotdot_rec_len));
 350                 if (root->reserved_zero ||
 351                     root->info_length != sizeof(struct dx_root_info))
 352                         return NULL;
 353 -               count_offset = 32;
 354 -       } else
 355 -               return NULL;
 356 +               count_offset = 8 + dot_rec_len + dotdot_rec_len;
 357 +       }
 358
 359         if (offset)
 360                 *offset = count_offset;
 361 @@ -526,13 +528,14 @@ ext4_next_entry(struct ext4_dir_entry_2 *p, unsigned long blocksize)
 362   * Future: use high four bits of block for coalesce-on-delete flags
 363   * Mask them off for now.
 364   */
 365 -struct dx_root_info *dx_get_dx_info(struct ext4_dir_entry_2 *de)
 366 +struct dx_root_info *dx_get_dx_info(struct ext4_dir_entry_2 *de, struct inode *i_dir)
 367  {
 368 +       BUG_ON(de->name_len != 1);
 369         /* get dotdot first */
 370 -       de = (struct ext4_dir_entry_2 *)((char *)de + EXT4_DIR_REC_LEN(1));
 371 +       de = (struct ext4_dir_entry_2 *)((char *)de + EXT4_DIR_ENTRY_LEN(de, i_dir));
 372
 373         /* dx root info is after dotdot entry */
 374 -       de = (struct ext4_dir_entry_2 *)((char *)de + EXT4_DIR_REC_LEN(2));
 375 +       de = (struct ext4_dir_entry_2 *)((char *)de + EXT4_DIR_ENTRY_LEN(de, i_dir));
 376
 377         return (struct dx_root_info *)de;
 378  }
 379 @@ -577,11 +580,16 @@ static inline void dx_set_limit(struct dx_entry *entries, unsigned value)
 380         ((struct dx_countlimit *) entries)->limit = cpu_to_le16(value);
 381  }
 382
 383 -static inline unsigned dx_root_limit(struct inode *dir, unsigned infosize)
 384 +static inline unsigned dx_root_limit(struct inode *dir,
 385 +               struct ext4_dir_entry_2 *dot_de, unsigned infosize)
 386  {
 387 -       unsigned int entry_space = dir->i_sb->s_blocksize -
 388 -                       ext4_dir_rec_len(1, NULL) -
 389 -                       ext4_dir_rec_len(2, NULL) - infosize;
 390 +       struct ext4_dir_entry_2 *dotdot_de;
 391 +       unsigned entry_space;
 392 +
 393 +       BUG_ON(dot_de->name_len != 1);
 394 +       dotdot_de = ext4_next_entry(dot_de, dir->i_sb->s_blocksize);
 395 +       entry_space = dir->i_sb->s_blocksize - EXT4_DIR_ENTRY_LEN(dot_de, NULL) -
 396 +                        EXT4_DIR_ENTRY_LEN(dotdot_de, NULL) - infosize;
 397
 398         if (ext4_has_metadata_csum(dir->i_sb))
 399                 entry_space -= sizeof(struct dx_tail);
 400 @@ -699,7 +707,7 @@ static struct stats dx_show_leaf(struct inode *dir,
 401                                        (unsigned) ((char *) de - base));
 402  #endif
 403                         }
 404 -                       space += ext4_dir_rec_len(de->name_len, dir);
 405 +                       space += EXT4_DIR_ENTRY_LEN(de, dir);
 406                         names++;
 407                 }
 408                 de = ext4_next_entry(de, size);
 409 @@ -793,7 +801,7 @@ dx_probe(struct ext4_filename *fname, struct inode *dir,
 410         if (IS_ERR(frame->bh))
 411                 return (struct dx_frame *) frame->bh;
 412
 413 -       info = dx_get_dx_info((struct ext4_dir_entry_2 *)frame->bh->b_data);
 414 +       info = dx_get_dx_info((struct ext4_dir_entry_2 *)frame->bh->b_data, dir);
 415         if (info->hash_version != DX_HASH_TEA &&
 416             info->hash_version != DX_HASH_HALF_MD4 &&
 417             info->hash_version != DX_HASH_LEGACY &&
 418 @@ -849,11 +857,14 @@ dx_probe(struct ext4_filename *fname, struct inode *dir,
 419
 420         entries = (struct dx_entry *)(((char *)info) + info->info_length);
 421
 422 -       if (dx_get_limit(entries) != dx_root_limit(dir,
 423 -                                                  info->info_length)) {
 424 +       if (dx_get_limit(entries) !=
 425 +           dx_root_limit(dir, (struct ext4_dir_entry_2 *)frame->bh->b_data,
 426 +                         info->info_length)) {
 427                 ext4_warning_inode(dir, "dx entry: limit %u != root limit %u",
 428                                    dx_get_limit(entries),
 429 -                                  dx_root_limit(dir, info->info_length));
 430 +                                  dx_root_limit(dir,
 431 +                                         (struct ext4_dir_entry_2 *)frame->bh->b_data,
 432 +                                         info->info_length));
 433                 goto fail;
 434         }
 435
 436 @@ -930,7 +941,7 @@ fail:
 437         return ret_err;
 438  }
 439
 440 -static void dx_release(struct dx_frame *frames)
 441 +static void dx_release(struct dx_frame *frames, struct inode *dir)
 442  {
 443         struct dx_root_info *info;
 444         int i;
 445 @@ -939,7 +950,7 @@ static void dx_release(struct dx_frame *frames)
 446         if (frames[0].bh == NULL)
 447                 return;
 448
 449 -       info = dx_get_dx_info((struct ext4_dir_entry_2 *)frames[0].bh->b_data);
 450 +       info = dx_get_dx_info((struct ext4_dir_entry_2 *)frames[0].bh->b_data, dir);
 451         /* save local copy, "info" may be freed after brelse() */
 452         indirect_levels = info->indirect_levels;
 453         for (i = 0; i <= indirect_levels; i++) {
 454 @@ -1240,12 +1251,12 @@ int ext4_htree_fill_tree(struct file *dir_file, __u32 start_hash,
 455                     (count && ((hashval & 1) == 0)))
 456                         break;
 457         }
 458 -       dx_release(frames);
 459 +       dx_release(frames, dir);
 460         dxtrace(printk(KERN_DEBUG "Fill tree: returned %d entries, "
 461                        "next hash: %x\n", count, *next_hash));
 462         return count;
 463  errout:
 464 -       dx_release(frames);
 465 +       dx_release(frames, dir);
 466         return (err);
 467  }
 468
 469 @@ -1778,7 +1789,7 @@ static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
 470  errout:
 471         dxtrace(printk(KERN_DEBUG "%s not found\n", fname->usr_fname->name));
 472  success:
 473 -       dx_release(frames);
 474 +       dx_release(frames, dir);
 475         return bh;
 476  }
 477
 478 @@ -1902,7 +1913,7 @@ dx_move_dirents(struct inode *dir, char *from, char *to,
 479         while (count--) {
 480                 struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *)
 481                                                 (from + (map->offs<<2));
 482 -               rec_len = ext4_dir_rec_len(de->name_len, dir);
 483 +               rec_len = EXT4_DIR_ENTRY_LEN(de, dir);
 484
 485                 memcpy (to, de, rec_len);
 486                 ((struct ext4_dir_entry_2 *) to)->rec_len =
 487 @@ -1935,7 +1946,7 @@ static struct ext4_dir_entry_2 *dx_pack_dirents(struct inode *dir, char *base,
 488         while ((char*)de < base + blocksize) {
 489                 next = ext4_next_entry(de, blocksize);
 490                 if (de->inode && de->name_len) {
 491 -                       rec_len = ext4_dir_rec_len(de->name_len, dir);
 492 +                       rec_len = EXT4_DIR_ENTRY_LEN(de, dir);
 493                         if (de > to)
 494                                 memmove(to, de, rec_len);
 495                         to->rec_len = ext4_rec_len_to_disk(rec_len, blocksize);
 496 @@ -2078,10 +2089,11 @@ int ext4_find_dest_de(struct inode *dir, struct inode *inode,
 497                       struct buffer_head *bh,
 498                       void *buf, int buf_size,
 499                       struct ext4_filename *fname,
 500 -                     struct ext4_dir_entry_2 **dest_de)
 501 +                     struct ext4_dir_entry_2 **dest_de,
 502 +                     int dlen)
 503  {
 504         struct ext4_dir_entry_2 *de;
 505 -       unsigned short reclen = ext4_dir_rec_len(fname_len(fname), dir);
 506 +       unsigned short reclen = ext4_dir_rec_len(fname_len(fname) + dlen, dir);
 507         int nlen, rlen;
 508         unsigned int offset = 0;
 509         char *top;
 510 @@ -2094,7 +2112,7 @@ int ext4_find_dest_de(struct inode *dir, struct inode *inode,
 511                         return -EFSCORRUPTED;
 512                 if (ext4_match(dir, fname, de))
 513                         return -EEXIST;
 514 -               nlen = ext4_dir_rec_len(de->name_len, dir);
 515 +               nlen = EXT4_DIR_ENTRY_LEN(de, dir);
 516                 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
 517                 if ((de->inode ? rlen - nlen : rlen) >= reclen)
 518                         break;
 519 @@ -2112,12 +2151,13 @@ void ext4_insert_dentry(struct inode *dir,
 520                         struct inode *inode,
 521                         struct ext4_dir_entry_2 *de,
 522                         int buf_size,
 523 -                       struct ext4_filename *fname)
 524 +                       struct ext4_filename *fname,
 525 +                       void *data)
 526  {
 527
 528         int nlen, rlen;
 529
 530 -       nlen = ext4_dir_rec_len(de->name_len, dir);
 531 +       nlen = EXT4_DIR_ENTRY_LEN(de, dir);
 532         rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
 533         if (de->inode) {
 534                 struct ext4_dir_entry_2 *de1 =
 535 @@ -2138,6 +2178,12 @@ void ext4_insert_dentry(struct inode *dir,
 536                 EXT4_DIRENT_HASHES(de)->minor_hash =
 537                                                 cpu_to_le32(hinfo->minor_hash);
 538         }
 539 +       if (data) {
 540 +               de->name[fname_len(fname)] = 0;
 541 +               memcpy(&de->name[fname_len(fname) + 1], data, *(char *)data);
 542 +               de->file_type |= EXT4_DIRENT_LUFID;
 543 +       }
 544 +
 545  }
 546
 547  /*
 548 @@ -2155,14 +2201,19 @@ static int add_dirent_to_buf(handle_t *handle, struct ext4_filename *fname,
 549  {
 550         unsigned int    blocksize = dir->i_sb->s_blocksize;
 551         int             csum_size = 0;
 552 -       int             err, err2;
 553 +       int             err, err2, dlen = 0;
 554 +       unsigned char   *data;
 555
 556 +       data = ext4_dentry_get_data(inode->i_sb, (struct ext4_dentry_param *)
 557 +                                               EXT4_I(inode)->i_dirdata);
 558         if (ext4_has_metadata_csum(inode->i_sb))
 559                 csum_size = sizeof(struct ext4_dir_entry_tail);
 560
 561         if (!de) {
 562 +               if (data)
 563 +                       dlen = (*data) + 1;
 564                 err = ext4_find_dest_de(dir, inode, bh, bh->b_data,
 565 -                                       blocksize - csum_size, fname, &de);
 566 +                                       blocksize - csum_size, fname, &de, dlen);
 567                 if (err)
 568                         return err;
 569         }
 570 @@ -2175,7 +2226,7 @@ static int add_dirent_to_buf(handle_t *handle, struct ext4_filename *fname,
 571         }
 572
 573         /* By now the buffer is marked for journaling */
 574 -       ext4_insert_dentry(dir, inode, de, blocksize, fname);
 575 +       ext4_insert_dentry(dir, inode, de, blocksize, fname, data);
 576
 577         /*
 578          * XXX shouldn't update any times until successful
 579 @@ -2273,7 +2327,7 @@ static int make_indexed_dir(handle_t *handle, struct ext4_filename *fname,
 580                                      blocksize);
 581
 582         /* initialize hashing info */
 583 -       dx_info = dx_get_dx_info(dot_de);
 584 +       dx_info = dx_get_dx_info(dot_de, dir);
 585         memset(dx_info, 0, sizeof(*dx_info));
 586         dx_info->info_length = sizeof(*dx_info);
 587         if (ext4_hash_in_dirent(dir))
 588 @@ -2284,7 +2338,8 @@ static int make_indexed_dir(handle_t *handle, struct ext4_filename *fname,
 589         entries = (void *)dx_info + sizeof(*dx_info);
 590         dx_set_block(entries, 1);
 591         dx_set_count(entries, 1);
 592 -       dx_set_limit(entries, dx_root_limit(dir, sizeof(*dx_info)));
 593 +       dx_set_limit(entries, dx_root_limit(dir,
 594 +                                        dot_de, sizeof(*dx_info)));
 595
 596         /* Initialize as for dx_probe */
 597         fname->hinfo.hash_version = dx_info->hash_version;
 598 @@ -2348,12 +2373,111 @@ out_frames:
 599          */
 600         if (retval)
 601                 ext4_mark_inode_dirty(handle, dir);
 602 -       dx_release(frames);
 603 +       dx_release(frames, dir);
 604         brelse(bh2);
 605         return retval;
 606  }
 607
 608 -/* update ".." entry */
 609 +static int ext4_expand_dotdot(struct inode *dir,
 610 +                             struct buffer_head *bh,
 611 +                             int dlen)
 612 +{
 613 +       struct ext4_dir_entry_2 *dot_de;
 614 +       struct ext4_dir_entry_2 *dotdot_de;
 615 +       int len;
 616 +       unsigned blocksize = dir->i_sb->s_blocksize;
 617 +
 618 +       dot_de = (struct ext4_dir_entry_2 *)bh->b_data;
 619 +       dotdot_de = ext4_next_entry(dot_de, blocksize);
 620 +
 621 +       if (is_dx(dir)) {
 622 +               struct dx_entry *entries;
 623 +               struct dx_root_info *dx_info;
 624 +               int limit, count;
 625 +               int entry_space;
 626 +
 627 +               len = EXT4_DIR_REC_LEN(2 + dlen, NULL) -
 628 +                       EXT4_DIR_ENTRY_LEN(dotdot_de, NULL);
 629 +
 630 +               dx_info = dx_get_dx_info(dot_de, NULL);
 631 +               entries = (struct dx_entry *)((char *)dx_info +
 632 +                                                       sizeof(*dx_info));
 633 +               count = dx_get_count(entries);
 634 +
 635 +               /*
 636 +                * figure out new limit with dlen,
 637 +                * check if we have enough space
 638 +                */
 639 +               entry_space = blocksize;
 640 +               entry_space -= (char *)dotdot_de - (char *)dot_de +
 641 +                              EXT4_DIR_REC_LEN(2 + dlen, NULL) +
 642 +                              sizeof(*dx_info);
 643 +               if (ext4_has_metadata_csum(dir->i_sb))
 644 +                       entry_space -= sizeof(struct dx_tail);
 645 +               limit = entry_space / sizeof(struct dx_entry);
 646 +               if (count > limit)
 647 +                       return -ENOSPC;
 648 +
 649 +               /* set the new limit, move dx_info and the entries */
 650 +               dx_set_limit(entries, limit);
 651 +               memmove((char *)dx_info + len, dx_info,
 652 +                       sizeof(*dx_info) + count * sizeof(struct dx_entry));
 653 +       } else {
 654 +               struct ext4_dir_entry_2 *next, *to, *prev, *de;
 655 +               char *top = (char *)bh->b_data + blocksize;
 656 +               int space = 0;
 657 +               unsigned rec_len = 0;
 658 +
 659 +               len = EXT4_DIR_REC_LEN(2 + dlen, NULL) -
 660 +                       ext4_rec_len_from_disk(dotdot_de->rec_len, blocksize);
 661 +
 662 +               if (ext4_has_metadata_csum(dir->i_sb))
 663 +                       top -= sizeof(struct ext4_dir_entry_tail);
 664 +
 665 +               de = ext4_next_entry(dotdot_de, blocksize);
 666 +               while ((char *)de < top) {
 667 +                       space += ext4_rec_len_from_disk(de->rec_len, blocksize) -
 668 +                                       EXT4_DIR_ENTRY_LEN(de, dir);
 669 +                       de = ext4_next_entry(de, blocksize);
 670 +               }
 671 +
 672 +               if (space < len)
 673 +                       return -ENOSPC;
 674 +
 675 +               /* pack all the entries after dotdot */
 676 +               de = ext4_next_entry(dotdot_de, blocksize);
 677 +               prev = to = de;
 678 +               while ((char *)de < top) {
 679 +                       next = ext4_next_entry(de, blocksize);
 680 +                       if (de->inode && de->name_len) {
 681 +                               rec_len = EXT4_DIR_ENTRY_LEN(de, dir);
 682 +                               if (de > to)
 683 +                                       memmove(to, de, rec_len);
 684 +                               to->rec_len = ext4_rec_len_to_disk(rec_len,
 685 +                                                                  blocksize);
 686 +                               prev = to;
 687 +                               to = (struct ext4_dir_entry_2 *)
 688 +                                               (((char *)to) + rec_len);
 689 +                       }
 690 +                       de = next;
 691 +               }
 692 +               /* fix up rec_len for the last entry */
 693 +               prev->rec_len = ext4_rec_len_to_disk(top - (char *)prev - len,
 694 +                                                    blocksize);
 695 +               /* move all the entries after dotdot to make space */
 696 +               de = ext4_next_entry(dotdot_de, blocksize);
 697 +               memmove((char *)de + len, de, (char *)prev - (char *)de +
 698 +                       EXT4_DIR_ENTRY_LEN(prev, dir));
 699 +               /* fix the rec_len for dotdot */
 700 +               dotdot_de->rec_len = ext4_rec_len_to_disk(
 701 +                                       EXT4_DIR_REC_LEN(2 + dlen, NULL),
 702 +                                       blocksize);
 703 +       }
 704 +
 705 +       return 0;
 706 +}
 707 +
 708 +/* update ".." entry, try to expand the entry if necessary */
 709  static int ext4_update_dotdot(handle_t *handle, struct dentry *dentry,
 710                               struct inode *inode)
 711  {
 712 @@ -2362,6 +2486,8 @@ static int ext4_update_dotdot(handle_t *
 713         struct ext4_dir_entry_2 *dot_de, *dotdot_de;
 714         unsigned int offset;
 715         int retval = 0;
 716 +       int dlen = 0;
 717 +       char *data;
 718
 719         if (IS_ERR(handle))
 720                 return PTR_ERR(handle);
 721 @@ -2402,6 +2528,30 @@ static int ext4_update_dotdot(handle_t *
 722
 723         dotdot_de->inode = cpu_to_le32(inode->i_ino);
 724
 725 +       data = ext4_dentry_get_data(dir->i_sb,
 726 +                       (struct ext4_dentry_param *)dentry->d_fsdata);
 727 +       if (data != NULL) {
 728 +               dlen = *data + 1;
 729 +               if (is_dx(dir)) {
 730 +                       if (ext4_get_dirent_data_len(dotdot_de) < dlen) {
 731 +                               if (ext4_expand_dotdot(dir, bh, dlen) < 0)
 732 +                                       dlen = 0;
 733 +                       }
 734 +               } else {
 735 +                       if (ext4_rec_len_from_disk(dotdot_de->rec_len,
 736 +                                                  dir->i_sb->s_blocksize) <
 737 +                           EXT4_DIR_REC_LEN(2 + dlen, NULL)) {
 738 +                               if (ext4_expand_dotdot(dir, bh, dlen) < 0)
 739 +                                       dlen = 0;
 740 +                       }
 741 +               }
 742 +       }
 743 +       if (dlen) {
 744 +               dotdot_de->name[2] = 0;
 745 +               memcpy(&dotdot_de->name[2 + 1], data, *data);
 746 +               dotdot_de->file_type |= LDISKFS_DIRENT_LUFID;
 747 +       }
 748 +
 749         ext4_mark_inode_dirty(handle, dir);
 750         BUFFER_TRACE(dir_block, "call ext4_handle_dirty_metadata");
 751         if (is_dx(dir)) {
 752 @@ -2422,6 +2489,7 @@ static int ext4_add_entry(handle_t *handle, struct dentry *dentry,
 753         ext4_lblk_t block, blocks;
 754         int     csum_size = 0;
 755
 756 +       EXT4_I(inode)->i_dirdata = dentry->d_fsdata;
 757         if (ext4_has_metadata_csum(inode->i_sb))
 758                 csum_size = sizeof(struct ext4_dir_entry_tail);
 759
 760 @@ -2664,7 +2732,7 @@ again:
 761                         dx_set_count(entries, 1);
 762                         dx_set_block(entries + 0, newblock);
 763                         info = dx_get_dx_info((struct ext4_dir_entry_2 *)
 764 -                                             frames[0].bh->b_data);
 765 +                                             frames[0].bh->b_data, dir);
 766                         info->indirect_levels = 1;
 767                         dxtrace(printk(KERN_DEBUG
 768                                        "Creating %d level index...\n",
 769 @@ -2690,7 +2758,7 @@ journal_error:
 770         ext4_std_error(dir->i_sb, err); /* this is a no-op if err == 0 */
 771  cleanup:
 772         brelse(bh);
 773 -       dx_release(frames);
 774 +       dx_release(frames, dir);
 775         /* @restart is true means htree-path has been changed, we need to
 776          * repeat dx_probe() to find out valid htree-path
 777          */
 778 @@ -2993,38 +3061,73 @@ err_unlock_inode:
 779         return err;
 780  }
 781
 782 +struct tp_block {
 783 +       struct inode *inode;
 784 +       void *data1;
 785 +       void *data2;
 786 +};
 787 +
 788  struct ext4_dir_entry_2 *ext4_init_dot_dotdot(struct inode *inode,
 789                           struct ext4_dir_entry_2 *de,
 790                           int blocksize, int csum_size,
 791                           unsigned int parent_ino, int dotdot_real_len)
 792  {
 793 +       void *data1 = NULL, *data2 = NULL;
 794 +       int dot_reclen = 0;
 795 +
 796 +       if (dotdot_real_len == 10) {
 797 +               struct tp_block *tpb = (struct tp_block *)inode;
 798 +               data1 = tpb->data1;
 799 +               data2 = tpb->data2;
 800 +               inode = tpb->inode;
 801 +               dotdot_real_len = 0;
 802 +       }
 803         de->inode = cpu_to_le32(inode->i_ino);
 804         de->name_len = 1;
 805 -       de->rec_len = ext4_rec_len_to_disk(ext4_dir_rec_len(de->name_len, NULL),
 806 -                                          blocksize);
 807         strcpy(de->name, ".");
 808         ext4_set_de_type(inode->i_sb, de, S_IFDIR);
 809
 810 +       /* get packed fid data*/
 811 +       data1 = ext4_dentry_get_data(inode->i_sb,
 812 +                               (struct ext4_dentry_param *) data1);
 813 +       if (data1) {
 814 +               de->name[1] = 0;
 815 +               memcpy(&de->name[2], data1, *(char *) data1);
 816 +               de->file_type |= EXT4_DIRENT_LUFID;
 817 +       }
 818 +       de->rec_len = cpu_to_le16(EXT4_DIR_ENTRY_LEN(de, NULL));
 819 +
 820 +       dot_reclen = cpu_to_le16(de->rec_len);
 821         de = ext4_next_entry(de, blocksize);
 822         de->inode = cpu_to_le32(parent_ino);
 823         de->name_len = 2;
 824 +
 825 +       strcpy(de->name, "..");
 826 +       ext4_set_de_type(inode->i_sb, de, S_IFDIR);
 827 +       data2 = ext4_dentry_get_data(inode->i_sb,
 828 +                       (struct ext4_dentry_param *) data2);
 829 +       if (data2) {
 830 +               de->name[2] = 0;
 831 +               memcpy(&de->name[3], data2, *(char *) data2);
 832 +               de->file_type |= EXT4_DIRENT_LUFID;
 833 +       }
 834 +
 835         if (!dotdot_real_len)
 836                 de->rec_len = ext4_rec_len_to_disk(blocksize -
 837 -                                       (csum_size + ext4_dir_rec_len(1, NULL)),
 838 -                                       blocksize);
 839 +                                       (csum_size + dot_reclen), blocksize);
 840         else
 841                 de->rec_len = ext4_rec_len_to_disk(
 842 -                                       ext4_dir_rec_len(de->name_len, NULL),
 843 +                                       EXT4_DIR_ENTRY_LEN(de, NULL),
 844                                         blocksize);
 845 -       strcpy(de->name, "..");
 846 -       ext4_set_de_type(inode->i_sb, de, S_IFDIR);
 847
 848         return ext4_next_entry(de, blocksize);
 849  }
 850
 851  int ext4_init_new_dir(handle_t *handle, struct inode *dir,
 852 -                            struct inode *inode)
 853 +                            struct inode *inode,
 854 +                            const void *data1, const void *data2)
 855  {
 856 +       struct tp_block param;
 857         struct buffer_head *dir_block = NULL;
 858         struct ext4_dir_entry_2 *de;
 859         ext4_lblk_t block = 0;
 860 @@ -3048,7 +3151,11 @@ int ext4_init_new_dir(handle_t *handle, struct inode *dir,
 861         if (IS_ERR(dir_block))
 862                 return PTR_ERR(dir_block);
 863         de = (struct ext4_dir_entry_2 *)dir_block->b_data;
 864 -       ext4_init_dot_dotdot(inode, de, blocksize, csum_size, dir->i_ino, 0);
 865 +       param.inode = inode;
 866 +       param.data1 = (void *)data1;
 867 +       param.data2 = (void *)data2;
 868 +       ext4_init_dot_dotdot((struct inode *)(&param), de, blocksize,
 869 +                            csum_size, dir->i_ino, 10);
 870         set_nlink(inode, 2);
 871         if (csum_size)
 872                 ext4_initialize_dirent_tail(dir_block, blocksize);
 873 @@ -3063,6 +3170,29 @@ out:
 874         return err;
 875  }
 876
 877 +/* Initialize @inode as a subdirectory of @dir, and add the
 878 + * "." and ".." entries into the first directory block. */
 879 +int ext4_add_dot_dotdot(handle_t *handle, struct inode *dir,
 880 +                       struct inode *inode,
 881 +                       const void *data1, const void *data2)
 882 +{
 883 +       int rc;
 884 +
 885 +       if (IS_ERR(handle))
 886 +               return PTR_ERR(handle);
 887 +
 888 +       if (IS_DIRSYNC(dir))
 889 +               ext4_handle_sync(handle);
 890 +
 891 +       inode->i_op = &ext4_dir_inode_operations;
 892 +       inode->i_fop = &ext4_dir_operations;
 893 +       rc = ext4_init_new_dir(handle, dir, inode, data1, data2);
 894 +       if (!rc)
 895 +               rc = ext4_mark_inode_dirty(handle, inode);
 896 +       return rc;
 897 +}
 898 +EXPORT_SYMBOL(ext4_add_dot_dotdot);
 899 +
 900  static int ext4_mkdir(struct user_namespace *mnt_userns, struct inode *dir,
 901                       struct dentry *dentry, umode_t mode)
 902  {
 903 @@ -3090,7 +3220,7 @@ retry:
 904
 905         inode->i_op = &ext4_dir_inode_operations;
 906         inode->i_fop = &ext4_dir_operations;
 907 -       err = ext4_init_new_dir(handle, dir, inode);
 908 +       err = ext4_init_new_dir(handle, dir, inode, NULL, NULL);
 909         if (err)
 910                 goto out_clear_inode;
 911         err = ext4_mark_inode_dirty(handle, inode);
 912 diff --git a/fs/ext4/super.c b/fs/ext4/super.c
 913 index 9a0af5a..917cca7 100644
 914 --- a/fs/ext4/super.c
 915 +++ b/fs/ext4/super.c
 916 @@ -1703,7 +1703,7 @@ enum {
 917         Opt_data_err_abort, Opt_data_err_ignore, Opt_test_dummy_encryption,
 918         Opt_inlinecrypt,
 919         Opt_usrjquota, Opt_grpjquota, Opt_quota,
 920 -       Opt_noquota, Opt_barrier, Opt_nobarrier, Opt_err,
 921 +       Opt_noquota, Opt_barrier, Opt_nobarrier, Opt_err, Opt_dirdata,
 922         Opt_usrquota, Opt_grpquota, Opt_prjquota, Opt_i_version,
 923         Opt_dax, Opt_dax_always, Opt_dax_inode, Opt_dax_never,
 924         Opt_stripe, Opt_delalloc, Opt_nodelalloc, Opt_warn_on_error,
 925 @@ -1819,6 +1819,7 @@ static const struct fs_parameter_spec ext4_param_specs[] = {
 926         fsparam_u32     ("stripe",              Opt_stripe),
 927         fsparam_flag    ("delalloc",            Opt_delalloc),
 928         fsparam_flag    ("nodelalloc",          Opt_nodelalloc),
 929 +       fsparam_flag    ("dirdata",             Opt_dirdata),
 930         fsparam_flag    ("warn_on_error",       Opt_warn_on_error),
 931         fsparam_flag    ("nowarn_on_error",     Opt_nowarn_on_error),
 932         fsparam_u32     ("debug_want_extra_isize",
 933 @@ -1956,6 +1957,7 @@ static const struct mount_opts {
 934                                                         MOPT_CLEAR | MOPT_Q},
 935         {Opt_usrjquota, 0, MOPT_Q},
 936         {Opt_grpjquota, 0, MOPT_Q},
 937 +       {Opt_dirdata, EXT4_MOUNT_DIRDATA, MOPT_SET},
 938         {Opt_jqfmt, 0, MOPT_QFMT},
 939         {Opt_nombcache, EXT4_MOUNT_NO_MBCACHE, MOPT_SET},
 940         {Opt_no_prefetch_block_bitmaps, EXT4_MOUNT_NO_PREFETCH_BLOCK_BITMAPS,
 941 --
 942 2.34.1
 943