ldiskfs/kernel_patches/patches/rhel9.3/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       | 251 ++++++++++++++++++++++++++++++++----------
  30  fs/ext4/super.c       |  12 +-
  31  6 files changed, 314 insertions(+), 75 deletions(-)
  32
  33 diff --git a/fs/ext4/dir.c b/fs/ext4/dir.c
  34 index 3985f8c3..b8e4df14 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 a4af3ec8..930ca3a5 100644
  67 --- a/fs/ext4/ext4.h
  68 +++ b/fs/ext4/ext4.h
  69 @@ -1184,6 +1184,7 @@ struct ext4_inode_info {
  70         __u32 i_csum_seed;
  71
  72         kprojid_t i_projid;
  73 +       void *i_dirdata;
  74  };
  75
  76  /*
  77 @@ -1205,6 +1206,7 @@ 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 @@ -2174,6 +2176,7 @@ 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 @@ -2385,6 +2388,42 @@ 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 @@ -2396,6 +2435,17 @@ 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 @@ -2403,10 +2453,10 @@ 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 @@ -2831,11 +2881,13 @@ 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 @@ -2851,10 +2903,17 @@ 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 @@ -3058,9 +3117,13 @@ 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 @@ -3887,6 +3950,36 @@ 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 8de354dc..00c4769b 100644
 257 --- a/fs/ext4/fast_commit.c
 258 +++ b/fs/ext4/fast_commit.c
 259 @@ -1674,7 +1674,7 @@ static int ext4_fc_replay_create(struct super_block *sb,
 260                         ext4_debug("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 c4475a74..3fc75d80 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, NULL);
 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 d0afa8f2..839d51ba 100644
 310 --- a/fs/ext4/namei.c
 311 +++ b/fs/ext4/namei.c
 312 @@ -291,13 +291,14 @@ 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 @@ -437,22 +438,23 @@ 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 @@ -555,13 +557,14 @@ 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 @@ -606,11 +609,16 @@ 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 @@ -728,7 +736,7 @@ 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 @@ -822,7 +830,7 @@ 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 @@ -878,11 +886,14 @@ 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 @@ -959,7 +970,7 @@ 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 @@ -968,7 +979,7 @@ 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 @@ -1269,12 +1280,12 @@ 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 @@ -1806,7 +1817,7 @@ 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 @@ -1930,7 +1941,7 @@ 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 @@ -1963,7 +1974,7 @@ 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 @@ -2106,14 +2117,22 @@ 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;
 507         int nlen, rlen;
 508         unsigned int offset = 0;
 509         char *top;
 510
 511 +       if (dlen) {
 512 +               reclen = ext4_dir_rec_len(fname_len(fname) + *dlen, dir);
 513 +               *dlen = 0;
 514 +       } else {
 515 +               reclen = ext4_dir_rec_len(fname_len(fname), dir);
 516 +       }
 517 +
 518         de = buf;
 519         top = buf + buf_size - reclen;
 520         while ((char *) de <= top) {
 521 @@ -2122,10 +2141,31 @@ int ext4_find_dest_de(struct inode *dir, struct inode *inode,
 522                         return -EFSCORRUPTED;
 523                 if (ext4_match(dir, fname, de))
 524                         return -EEXIST;
 525 -               nlen = ext4_dir_rec_len(de->name_len, dir);
 526 +               nlen = EXT4_DIR_ENTRY_LEN(de, dir);
 527                 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
 528                 if ((de->inode ? rlen - nlen : rlen) >= reclen)
 529                         break;
 530 +
 531 +               /* Then for dotdot entries, check for the smaller space
 532 +                * required for just the entry, no FID
 533 +                */
 534 +               if (fname_len(fname) == 2 && memcmp(fname_name(fname), "..", 2) == 0) {
 535 +                       if ((de->inode ? rlen - nlen : rlen) >=
 536 +                           ext4_dir_rec_len(fname_len(fname), dir)) {
 537 +                               /* set dlen = 1 to indicate not
 538 +                                * enough space store fid
 539 +                                */
 540 +                               if (dlen)
 541 +                                       *dlen = 1;
 542 +                               break;
 543 +                       }
 544 +                       /* The new ".." entry must be written over the
 545 +                        * previous ".." entry, which is the first
 546 +                        * entry traversed by this scan. If it doesn't
 547 +                        * fit, something is badly wrong, so -EIO.
 548 +                        */
 549 +                       return -EIO;
 550 +               }
 551                 de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
 552                 offset += rlen;
 553         }
 554 @@ -2140,12 +2180,13 @@ void ext4_insert_dentry(struct inode *dir,
 555                         struct inode *inode,
 556                         struct ext4_dir_entry_2 *de,
 557                         int buf_size,
 558 -                       struct ext4_filename *fname)
 559 +                       struct ext4_filename *fname,
 560 +                       void *data)
 561  {
 562
 563         int nlen, rlen;
 564
 565 -       nlen = ext4_dir_rec_len(de->name_len, dir);
 566 +       nlen = EXT4_DIR_ENTRY_LEN(de, dir);
 567         rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
 568         if (de->inode) {
 569                 struct ext4_dir_entry_2 *de1 =
 570 @@ -2166,6 +2207,12 @@ void ext4_insert_dentry(struct inode *dir,
 571                 EXT4_DIRENT_HASHES(de)->minor_hash =
 572                                                 cpu_to_le32(hinfo->minor_hash);
 573         }
 574 +       if (data) {
 575 +               de->name[fname_len(fname)] = 0;
 576 +               memcpy(&de->name[fname_len(fname) + 1], data, *(char *)data);
 577 +               de->file_type |= EXT4_DIRENT_LUFID;
 578 +       }
 579 +
 580  }
 581
 582  /*
 583 @@ -2183,14 +2230,19 @@ static int add_dirent_to_buf(handle_t *handle, struct ext4_filename *fname,
 584  {
 585         unsigned int    blocksize = dir->i_sb->s_blocksize;
 586         int             csum_size = 0;
 587 -       int             err, err2;
 588 +       int             err, err2, dlen = 0;
 589 +       unsigned char   *data;
 590
 591 +       data = ext4_dentry_get_data(inode->i_sb, (struct ext4_dentry_param *)
 592 +                                               EXT4_I(inode)->i_dirdata);
 593         if (ext4_has_metadata_csum(inode->i_sb))
 594                 csum_size = sizeof(struct ext4_dir_entry_tail);
 595
 596         if (!de) {
 597 +               if (data)
 598 +                       dlen = (*data) + 1;
 599                 err = ext4_find_dest_de(dir, inode, bh, bh->b_data,
 600 -                                       blocksize - csum_size, fname, &de);
 601 +                                       blocksize - csum_size, fname, &de, &dlen);
 602                 if (err)
 603                         return err;
 604         }
 605 @@ -2203,7 +2255,10 @@ static int add_dirent_to_buf(handle_t *handle, struct ext4_filename *fname,
 606         }
 607
 608         /* By now the buffer is marked for journaling */
 609 -       ext4_insert_dentry(dir, inode, de, blocksize, fname);
 610 +       /* If writing the short form of "dotdot", don't add the data section */
 611 +       if (dlen == 1)
 612 +               data = NULL;
 613 +       ext4_insert_dentry(dir, inode, de, blocksize, fname, data);
 614
 615         /*
 616          * XXX shouldn't update any times until successful
 617 @@ -2309,7 +2364,7 @@ static int make_indexed_dir(handle_t *handle, struct ext4_filename *fname,
 618                                      blocksize);
 619
 620         /* initialize hashing info */
 621 -       dx_info = dx_get_dx_info(dot_de);
 622 +       dx_info = dx_get_dx_info(dot_de, dir);
 623         memset(dx_info, 0, sizeof(*dx_info));
 624         dx_info->info_length = sizeof(*dx_info);
 625         if (ext4_hash_in_dirent(dir))
 626 @@ -2320,7 +2375,8 @@ static int make_indexed_dir(handle_t *handle, struct ext4_filename *fname,
 627         entries = (void *)dx_info + sizeof(*dx_info);
 628         dx_set_block(entries, 1);
 629         dx_set_count(entries, 1);
 630 -       dx_set_limit(entries, dx_root_limit(dir, sizeof(*dx_info)));
 631 +       dx_set_limit(entries, dx_root_limit(dir,
 632 +                                        dot_de, sizeof(*dx_info)));
 633
 634         /* Initialize as for dx_probe */
 635         fname->hinfo.hash_version = dx_info->hash_version;
 636 @@ -2361,7 +2417,7 @@ out_frames:
 637          */
 638         if (retval)
 639                 ext4_mark_inode_dirty(handle, dir);
 640 -       dx_release(frames);
 641 +       dx_release(frames, dir);
 642         brelse(bh2);
 643         return retval;
 644  }
 645 @@ -2374,6 +2430,8 @@ static int ext4_update_dotdot(handle_t *handle, struct dentry *dentry,
 646         struct buffer_head *dir_block;
 647         struct ext4_dir_entry_2 *de;
 648         int len, journal = 0, err = 0;
 649 +       int dlen = 0;
 650 +       char *data;
 651
 652         if (IS_ERR(handle))
 653                 return PTR_ERR(handle);
 654 @@ -2389,21 +2447,26 @@ static int ext4_update_dotdot(handle_t *handle, struct dentry *dentry,
 655
 656         de = (struct ext4_dir_entry_2 *)dir_block->b_data;
 657         /* the first item must be "." */
 658 -       assert(de->name_len == 1 && de->name[0] == '.');
 659 +       ASSERT(de->name_len == 1 && de->name[0] == '.');
 660         len = le16_to_cpu(de->rec_len);
 661 -       assert(len >= EXT4_DIR_REC_LEN(1));
 662 -       if (len > EXT4_DIR_REC_LEN(1)) {
 663 +       ASSERT(len >= EXT4_DIR_REC_LEN(1, dir));
 664 +       if (len > EXT4_DIR_REC_LEN(1, dir)) {
 665                 BUFFER_TRACE(dir_block, "get_write_access");
 666                 err = ext4_journal_get_write_access(handle, dir->i_sb, dir_block, EXT4_JTR_NONE);
 667                 if (err)
 668                         goto out_journal;
 669
 670                 journal = 1;
 671 -               de->rec_len = cpu_to_le16(EXT4_DIR_REC_LEN(1));
 672 +               de->rec_len = cpu_to_le16(EXT4_DIR_ENTRY_LEN(de, dir));
 673         }
 674
 675 -       len -= EXT4_DIR_REC_LEN(1);
 676 -       assert(len == 0 || len >= EXT4_DIR_REC_LEN(2));
 677 +       len -= EXT4_DIR_ENTRY_LEN(de, NULL);
 678 +       data = ext4_dentry_get_data(dir->i_sb,
 679 +                       (struct ext4_dentry_param *)dentry->d_fsdata);
 680 +       if (data)
 681 +               dlen = *data + 1;
 682 +       ASSERT(len == 0 || len >= EXT4_DIR_REC_LEN(2 + dlen, dir));
 683 +
 684         de = (struct ext4_dir_entry_2 *)
 685                         ((char *) de + le16_to_cpu(de->rec_len));
 686         if (!journal) {
 687 @@ -2417,10 +2480,15 @@ static int ext4_update_dotdot(handle_t *handle, struct dentry *dentry,
 688         if (len > 0)
 689                 de->rec_len = cpu_to_le16(len);
 690         else
 691 -               assert(le16_to_cpu(de->rec_len) >= EXT4_DIR_REC_LEN(2));
 692 +               ASSERT(le16_to_cpu(de->rec_len) >= EXT4_DIR_REC_LEN(2, dir));
 693         de->name_len = 2;
 694         strcpy(de->name, "..");
 695 -       ext4_set_de_type(dir->i_sb, de, S_IFDIR);
 696 +       if (data != NULL && ext4_get_dirent_data_len(de) >= dlen) {
 697 +               de->name[2] = 0;
 698 +               memcpy(&de->name[2 + 1], data, *data);
 699 +               ext4_set_de_type(dir->i_sb, de, S_IFDIR);
 700 +               de->file_type |= EXT4_DIRENT_LUFID;
 701 +       }
 702
 703  out_journal:
 704         if (journal) {
 705 @@ -2458,6 +2526,7 @@ static int ext4_add_entry(handle_t *handle, struct dentry *dentry,
 706         ext4_lblk_t block, blocks;
 707         int     csum_size = 0;
 708
 709 +       EXT4_I(inode)->i_dirdata = dentry->d_fsdata;
 710         if (ext4_has_metadata_csum(inode->i_sb))
 711                 csum_size = sizeof(struct ext4_dir_entry_tail);
 712
 713 @@ -2700,7 +2769,7 @@ again:
 714                         dx_set_count(entries, 1);
 715                         dx_set_block(entries + 0, newblock);
 716                         info = dx_get_dx_info((struct ext4_dir_entry_2 *)
 717 -                                             frames[0].bh->b_data);
 718 +                                             frames[0].bh->b_data, dir);
 719                         info->indirect_levels = 1;
 720                         dxtrace(printk(KERN_DEBUG
 721                                        "Creating %d level index...\n",
 722 @@ -2726,7 +2795,7 @@ journal_error:
 723         ext4_std_error(dir->i_sb, err); /* this is a no-op if err == 0 */
 724  cleanup:
 725         brelse(bh);
 726 -       dx_release(frames);
 727 +       dx_release(frames, dir);
 728         /* @restart is true means htree-path has been changed, we need to
 729          * repeat dx_probe() to find out valid htree-path
 730          */
 731 @@ -3029,38 +3098,73 @@ err_unlock_inode:
 732         return err;
 733  }
 734
 735 +struct tp_block {
 736 +       struct inode *inode;
 737 +       void *data1;
 738 +       void *data2;
 739 +};
 740 +
 741  struct ext4_dir_entry_2 *ext4_init_dot_dotdot(struct inode *inode,
 742                           struct ext4_dir_entry_2 *de,
 743                           int blocksize, int csum_size,
 744                           unsigned int parent_ino, int dotdot_real_len)
 745  {
 746 +       void *data1 = NULL, *data2 = NULL;
 747 +       int dot_reclen = 0;
 748 +
 749 +       if (dotdot_real_len == 10) {
 750 +               struct tp_block *tpb = (struct tp_block *)inode;
 751 +               data1 = tpb->data1;
 752 +               data2 = tpb->data2;
 753 +               inode = tpb->inode;
 754 +               dotdot_real_len = 0;
 755 +       }
 756         de->inode = cpu_to_le32(inode->i_ino);
 757         de->name_len = 1;
 758 -       de->rec_len = ext4_rec_len_to_disk(ext4_dir_rec_len(de->name_len, NULL),
 759 -                                          blocksize);
 760         strcpy(de->name, ".");
 761         ext4_set_de_type(inode->i_sb, de, S_IFDIR);
 762
 763 +       /* get packed fid data*/
 764 +       data1 = ext4_dentry_get_data(inode->i_sb,
 765 +                               (struct ext4_dentry_param *) data1);
 766 +       if (data1) {
 767 +               de->name[1] = 0;
 768 +               memcpy(&de->name[2], data1, *(char *) data1);
 769 +               de->file_type |= EXT4_DIRENT_LUFID;
 770 +       }
 771 +       de->rec_len = cpu_to_le16(EXT4_DIR_ENTRY_LEN(de, NULL));
 772 +
 773 +       dot_reclen = cpu_to_le16(de->rec_len);
 774         de = ext4_next_entry(de, blocksize);
 775         de->inode = cpu_to_le32(parent_ino);
 776         de->name_len = 2;
 777 +
 778 +       strcpy(de->name, "..");
 779 +       ext4_set_de_type(inode->i_sb, de, S_IFDIR);
 780 +       data2 = ext4_dentry_get_data(inode->i_sb,
 781 +                       (struct ext4_dentry_param *) data2);
 782 +       if (data2) {
 783 +               de->name[2] = 0;
 784 +               memcpy(&de->name[3], data2, *(char *) data2);
 785 +               de->file_type |= EXT4_DIRENT_LUFID;
 786 +       }
 787 +
 788         if (!dotdot_real_len)
 789                 de->rec_len = ext4_rec_len_to_disk(blocksize -
 790 -                                       (csum_size + ext4_dir_rec_len(1, NULL)),
 791 -                                       blocksize);
 792 +                                       (csum_size + dot_reclen), blocksize);
 793         else
 794                 de->rec_len = ext4_rec_len_to_disk(
 795 -                                       ext4_dir_rec_len(de->name_len, NULL),
 796 +                                       EXT4_DIR_ENTRY_LEN(de, NULL),
 797                                         blocksize);
 798 -       strcpy(de->name, "..");
 799 -       ext4_set_de_type(inode->i_sb, de, S_IFDIR);
 800
 801         return ext4_next_entry(de, blocksize);
 802  }
 803
 804  int ext4_init_new_dir(handle_t *handle, struct inode *dir,
 805 -                            struct inode *inode)
 806 +                            struct inode *inode,
 807 +                            const void *data1, const void *data2)
 808  {
 809 +       struct tp_block param;
 810         struct buffer_head *dir_block = NULL;
 811         struct ext4_dir_entry_2 *de;
 812         ext4_lblk_t block = 0;
 813 @@ -3084,7 +3188,11 @@ int ext4_init_new_dir(handle_t *handle, struct inode *dir,
 814         if (IS_ERR(dir_block))
 815                 return PTR_ERR(dir_block);
 816         de = (struct ext4_dir_entry_2 *)dir_block->b_data;
 817 -       ext4_init_dot_dotdot(inode, de, blocksize, csum_size, dir->i_ino, 0);
 818 +       param.inode = inode;
 819 +       param.data1 = (void *)data1;
 820 +       param.data2 = (void *)data2;
 821 +       ext4_init_dot_dotdot((struct inode *)(&param), de, blocksize,
 822 +                            csum_size, dir->i_ino, 10);
 823         set_nlink(inode, 2);
 824         if (csum_size)
 825                 ext4_initialize_dirent_tail(dir_block, blocksize);
 826 @@ -3099,6 +3207,29 @@ out:
 827         return err;
 828  }
 829
 830 +/* Initialize @inode as a subdirectory of @dir, and add the
 831 + * "." and ".." entries into the first directory block. */
 832 +int ext4_add_dot_dotdot(handle_t *handle, struct inode *dir,
 833 +                       struct inode *inode,
 834 +                       const void *data1, const void *data2)
 835 +{
 836 +       int rc;
 837 +
 838 +       if (IS_ERR(handle))
 839 +               return PTR_ERR(handle);
 840 +
 841 +       if (IS_DIRSYNC(dir))
 842 +               ext4_handle_sync(handle);
 843 +
 844 +       inode->i_op = &ext4_dir_inode_operations;
 845 +       inode->i_fop = &ext4_dir_operations;
 846 +       rc = ext4_init_new_dir(handle, dir, inode, data1, data2);
 847 +       if (!rc)
 848 +               rc = ext4_mark_inode_dirty(handle, inode);
 849 +       return rc;
 850 +}
 851 +EXPORT_SYMBOL(ext4_add_dot_dotdot);
 852 +
 853  static int ext4_mkdir(struct user_namespace *mnt_userns, struct inode *dir,
 854                       struct dentry *dentry, umode_t mode)
 855  {
 856 @@ -3126,7 +3257,7 @@ retry:
 857
 858         inode->i_op = &ext4_dir_inode_operations;
 859         inode->i_fop = &ext4_dir_operations;
 860 -       err = ext4_init_new_dir(handle, dir, inode);
 861 +       err = ext4_init_new_dir(handle, dir, inode, NULL, NULL);
 862         if (err)
 863                 goto out_clear_inode;
 864         err = ext4_mark_inode_dirty(handle, inode);
 865 diff --git a/fs/ext4/super.c b/fs/ext4/super.c
 866 index c013357d..1838ee86 100644
 867 --- a/fs/ext4/super.c
 868 +++ b/fs/ext4/super.c
 869 @@ -1571,7 +1571,7 @@ enum {
 870         Opt_data_err_abort, Opt_data_err_ignore, Opt_test_dummy_encryption,
 871         Opt_inlinecrypt,
 872         Opt_usrjquota, Opt_grpjquota, Opt_quota,
 873 -       Opt_noquota, Opt_barrier, Opt_nobarrier, Opt_err,
 874 +       Opt_noquota, Opt_barrier, Opt_nobarrier, Opt_err, Opt_dirdata,
 875         Opt_usrquota, Opt_grpquota, Opt_prjquota,
 876         Opt_dax, Opt_dax_always, Opt_dax_inode, Opt_dax_never,
 877         Opt_stripe, Opt_delalloc, Opt_nodelalloc, Opt_warn_on_error,
 878 @@ -1687,6 +1687,7 @@ static const struct fs_parameter_spec ext4_param_specs[] = {
 879         fsparam_u32     ("stripe",              Opt_stripe),
 880         fsparam_flag    ("delalloc",            Opt_delalloc),
 881         fsparam_flag    ("nodelalloc",          Opt_nodelalloc),
 882 +       fsparam_flag    ("dirdata",             Opt_dirdata),
 883         fsparam_flag    ("warn_on_error",       Opt_warn_on_error),
 884         fsparam_flag    ("nowarn_on_error",     Opt_nowarn_on_error),
 885         fsparam_u32     ("debug_want_extra_isize",
 886 @@ -1823,6 +1824,7 @@ static const struct mount_opts {
 887                                                         MOPT_CLEAR | MOPT_Q},
 888         {Opt_usrjquota, 0, MOPT_Q},
 889         {Opt_grpjquota, 0, MOPT_Q},
 890 +       {Opt_dirdata, EXT4_MOUNT_DIRDATA, MOPT_SET},
 891         {Opt_jqfmt, 0, MOPT_QFMT},
 892         {Opt_nombcache, EXT4_MOUNT_NO_MBCACHE, MOPT_SET},
 893         {Opt_no_prefetch_block_bitmaps, EXT4_MOUNT_NO_PREFETCH_BLOCK_BITMAPS,
 894 @@ -3563,6 +3565,14 @@ int ext4_feature_set_ok(struct super_block *sb, int readonly)
 895         }
 896  #endif
 897
 898 +       if ((ext4_has_feature_casefold(sb) || ext4_has_feature_encrypt(sb)) &&
 899 +            ext4_has_feature_dirdata(sb)) {
 900 +               ext4_msg(sb, KERN_ERR,
 901 +                        "Filesystem with casefold and encrypt are incompatible "
 902 +                        "with dirdata");
 903 +               return 0;
 904 +       }
 905 +
 906         if (readonly)
 907                 return 1;
 908
 909 --
 910 2.25.1
 911