ldiskfs/kernel_patches/patches/sles15sp4/ext4-data-in-dirent.patch

   1
   2 commit 2db3b2b33ee796f4ea61316773452d936303ad27
   3 Author:     Pravin Shelar <Pravin.Shelar@sun.com>
   4 AuthorDate: Sun Oct 4 18:13:14 2009 +0000
   5 Subject: ext4: add ext4-data-in-dirent patch
   6
   7 Allows ext4 to store extra data records inside the ext4_dirent
   8 along with the regular directory entry (type, length, filename).
   9 Data is stored in ext4 dirent after filename, with a bit flag in
  10 de->file_type to indicate if any record after de->name is used.
  11 Each in-use record is variable length and must store a 1-byte
  12 length (including the length byte itself) at the start so that it
  13 can be skipped if the record type is unknown/uneeded.  The record
  14 starts after a NUL byte terminator for the filename.  This extra
  15 space is accounted in de->rec_len but not de->name_len.
  16
  17 Flag EXT4_DIRENT_LUFID is used for a 128-bit file identifier.
  18 Make use of dentry->d_fsdata to pass LUFID to ext4, so no changes
  19 in ext4_add_entry() interface are required.
  20
  21 Bugzilla-ID: b=17670
  22 Signed-off-by: Pravin Shelar <Pravin.Shelar@sun.com>
  23 Reviewed-by: Huang Hua <h.huang@sun.com>
  24 Signed-off-by: Andreas Dilger <andreas.dilger@sun.com>
  25 ---
  26  fs/ext4/dir.c         |   9 +-
  27  fs/ext4/ext4.h        | 106 ++++++++++++++++--
  28  fs/ext4/fast_commit.c |   2 +-
  29  fs/ext4/inline.c      |   8 +-
  30  fs/ext4/namei.c       | 249 ++++++++++++++++++++++++++++++++----------
  31  fs/ext4/super.c       |   4 +-
  32  6 files changed, 303 insertions(+), 75 deletions(-)
  33
  34 diff --git a/fs/ext4/dir.c b/fs/ext4/dir.c
  35 index 74b172a..c6afabc 100644
  36 --- a/fs/ext4/dir.c
  37 +++ b/fs/ext4/dir.c
  38 @@ -466,12 +466,17 @@ int ext4_htree_store_dirent(struct file *dir_file, __u32 hash,
  39         struct fname *fname, *new_fn;
  40         struct dir_private_info *info;
  41         int len;
  42 +       int extra_data = 0;
  43
  44         info = dir_file->private_data;
  45         p = &info->root.rb_node;
  46
  47         /* Create and allocate the fname structure */
  48 -       len = sizeof(struct fname) + ent_name->len + 1;
  49 +       if (dirent->file_type & EXT4_DIRENT_LUFID)
  50 +               extra_data = ext4_get_dirent_data_len(dirent);
  51 +
  52 +       len = sizeof(struct fname) + ent_name->len + extra_data + 1;
  53 +
  54         new_fn = kzalloc(len, GFP_KERNEL);
  55         if (!new_fn)
  56                 return -ENOMEM;
  57 @@ -480,7 +485,7 @@ int ext4_htree_store_dirent(struct file *dir_file, __u32 hash,
  58         new_fn->inode = le32_to_cpu(dirent->inode);
  59         new_fn->name_len = ent_name->len;
  60         new_fn->file_type = dirent->file_type;
  61 -       memcpy(new_fn->name, ent_name->name, ent_name->len);
  62 +       memcpy(new_fn->name, ent_name->name, ent_name->len + extra_data);
  63
  64         while (*p) {
  65                 parent = *p;
  66 diff --git a/fs/ext4/ext4.h b/fs/ext4/ext4.h
  67 index 0791a8b..f1bc21d 100644
  68 --- a/fs/ext4/ext4.h
  69 +++ b/fs/ext4/ext4.h
  70 @@ -1165,6 +1165,7 @@ struct ext4_inode_info {
  71         __u32 i_csum_seed;
  72
  73         kprojid_t i_projid;
  74 +       void *i_dirdata;
  75  };
  76
  77  /*
  78 @@ -1186,6 +1187,7 @@ struct ext4_inode_info {
  79   * Mount flags set via mount options or defaults
  80   */
  81  #define EXT4_MOUNT_NO_MBCACHE          0x00001 /* Do not use mbcache */
  82 +#define EXT4_MOUNT_DIRDATA             0x00002 /* Data in directory entries */
  83  #define EXT4_MOUNT_GRPID               0x00004 /* Create files with directory's group */
  84  #define EXT4_MOUNT_DEBUG               0x00008 /* Some debugging messages */
  85  #define EXT4_MOUNT_ERRORS_CONT         0x00010 /* Continue on errors */
  86 @@ -2117,6 +2119,7 @@ EXT4_FEATURE_INCOMPAT_FUNCS(casefold,             CASEFOLD)
  87                                          EXT4_FEATURE_INCOMPAT_FLEX_BG| \
  88                                          EXT4_FEATURE_INCOMPAT_EA_INODE| \
  89                                          EXT4_FEATURE_INCOMPAT_MMP | \
  90 +                                        EXT4_FEATURE_INCOMPAT_DIRDATA| \
  91                                          EXT4_FEATURE_INCOMPAT_INLINE_DATA | \
  92                                          EXT4_FEATURE_INCOMPAT_ENCRYPT | \
  93                                          EXT4_FEATURE_INCOMPAT_CASEFOLD | \
  94 @@ -2326,6 +2329,42 @@ struct ext4_dir_entry_tail {
  95  #define EXT4_FT_SYMLINK                7
  96
  97  #define EXT4_FT_MAX            8
  98 +#define EXT4_FT_MASK           0xf
  99 +
 100 +#if EXT4_FT_MAX > EXT4_FT_MASK
 101 +#error "conflicting EXT4_FT_MAX and EXT4_FT_MASK"
 102 +#endif
 103 +
 104 +/*
 105 + * d_type has 4 unused bits, so it can hold four types data. these different
 106 + * type of data (e.g. lustre data, high 32 bits of 64-bit inode number) can be
 107 + * stored, in flag order, after file-name in ext4 dirent.
 108 +*/
 109 +/*
 110 + * this flag is added to d_type if ext4 dirent has extra data after
 111 + * filename. this data length is variable and length is stored in first byte
 112 + * of data. data start after filename NUL byte.
 113 + * This is used by Lustre FS.
 114 +  */
 115 +#define EXT4_DIRENT_LUFID              0x10
 116 +
 117 +#define EXT4_LUFID_MAGIC    0xAD200907UL
 118 +struct ext4_dentry_param {
 119 +       __u32  edp_magic;       /* EXT4_LUFID_MAGIC */
 120 +       char   edp_len;         /* size of edp_data in bytes */
 121 +       char   edp_data[0];     /* packed array of data */
 122 +} __packed;
 123 +
 124 +static inline unsigned char *ext4_dentry_get_data(struct super_block *sb,
 125 +                                                 struct ext4_dentry_param *p)
 126 +{
 127 +       if (!ext4_has_feature_dirdata(sb))
 128 +               return NULL;
 129 +       if (p && p->edp_magic == EXT4_LUFID_MAGIC)
 130 +               return &p->edp_len;
 131 +       else
 132 +               return NULL;
 133 +}
 134
 135  #define EXT4_FT_DIR_CSUM       0xDE
 136
 137 @@ -2337,6 +2376,17 @@ struct ext4_dir_entry_tail {
 138  #define EXT4_DIR_PAD                   4
 139  #define EXT4_DIR_ROUND                 (EXT4_DIR_PAD - 1)
 140  #define EXT4_MAX_REC_LEN               ((1<<16)-1)
 141 +#define EXT4_DIR_REC_LEN_(name_len, i_dir) \
 142 +       ext4_dir_rec_len((name_len), (i_dir))
 143 +#define EXT4_DIR_ENTRY_LEN_(de, i_dir) \
 144 +       (EXT4_DIR_REC_LEN_((de)->name_len + ext4_get_dirent_data_len(de), \
 145 +               (i_dir)))
 146 +/* ldiskfs */
 147 +#define EXT4_DIR_REC_LEN(name_len, i_dir)      EXT4_DIR_REC_LEN_((name_len), (i_dir))
 148 +#define EXT4_DIR_ENTRY_LEN(de, i_dir)          EXT4_DIR_ENTRY_LEN_((de), (i_dir))
 149 +/* lustre osd_handler compat -- ifdef LDISKFS_DIR_REC_LEN_WITH_DIR */
 150 +#define EXT4_DIR_REC_LEN_WITH_DIR              1
 151 +#define __EXT4_DIR_REC_LEN(name_len)           EXT4_DIR_REC_LEN_((name_len), NULL)
 152
 153  /*
 154   * The rec_len is dependent on the type of directory. Directories that are
 155 @@ -2344,10 +2394,10 @@ struct ext4_dir_entry_tail {
 156   * ext4_extended_dir_entry_2. For all entries related to '.' or '..' you should
 157   * pass NULL for dir, as those entries do not use the extra fields.
 158   */
 159 -static inline unsigned int ext4_dir_rec_len(__u8 name_len,
 160 +static inline unsigned int ext4_dir_rec_len(__u32 name_len,
 161                                                 const struct inode *dir)
 162  {
 163 -       int rec_len = (name_len + 8 + EXT4_DIR_ROUND);
 164 +       __u32 rec_len = (name_len + 8 + EXT4_DIR_ROUND);
 165
 166         if (dir && ext4_hash_in_dirent(dir))
 167                 rec_len += sizeof(struct ext4_dir_entry_hash);
 168 @@ -2821,11 +2871,13 @@ extern int ext4_find_dest_de(struct inode *dir, struct inode *inode,
 169                              struct buffer_head *bh,
 170                              void *buf, int buf_size,
 171                              struct ext4_filename *fname,
 172 -                            struct ext4_dir_entry_2 **dest_de);
 173 +                            struct ext4_dir_entry_2 **dest_de,
 174 +                            int *dlen);
 175  void ext4_insert_dentry(struct inode *dir, struct inode *inode,
 176                         struct ext4_dir_entry_2 *de,
 177                         int buf_size,
 178 -                       struct ext4_filename *fname);
 179 +                       struct ext4_filename *fname,
 180 +                       void *data);
 181  static inline void ext4_update_dx_flag(struct inode *inode)
 182  {
 183         if (!ext4_has_feature_dir_index(inode->i_sb) &&
 184 @@ -2841,10 +2893,17 @@ static const unsigned char ext4_filetype_table[] = {
 185
 186  static inline  unsigned char get_dtype(struct super_block *sb, int filetype)
 187  {
 188 -       if (!ext4_has_feature_filetype(sb) || filetype >= EXT4_FT_MAX)
 189 +       int fl_index = filetype & EXT4_FT_MASK;
 190 +
 191 +       if (!ext4_has_feature_filetype(sb) || fl_index >= EXT4_FT_MAX)
 192                 return DT_UNKNOWN;
 193
 194 -       return ext4_filetype_table[filetype];
 195 +       if (!test_opt(sb, DIRDATA))
 196 +               return ext4_filetype_table[fl_index];
 197 +
 198 +       return (ext4_filetype_table[fl_index]) |
 199 +               (filetype & EXT4_DIRENT_LUFID);
 200 +
 201  }
 202  extern int ext4_check_all_de(struct inode *dir, struct buffer_head *bh,
 203                              void *buf, int buf_size);
 204 @@ -3048,7 +3107,8 @@ extern int ext4_ind_migrate(struct inode *inode);
 205
 206  /* namei.c */
 207  extern int ext4_init_new_dir(handle_t *handle, struct inode *dir,
 208 -                            struct inode *inode);
 209 +                            struct inode *inode,
 210 +                            const void *data1, const void *data2);
 211  extern int ext4_dirblock_csum_verify(struct inode *inode,
 212                                      struct buffer_head *bh);
 213  extern int ext4_orphan_add(handle_t *, struct inode *);
 214 @@ -3059,6 +3119,8 @@ extern struct inode *ext4_create_inode(handle_t *handle,
 215  extern int ext4_delete_entry(handle_t *handle, struct inode * dir,
 216                              struct ext4_dir_entry_2 *de_del,
 217                              struct buffer_head *bh);
 218 +extern int ext4_add_dot_dotdot(handle_t *handle, struct inode *dir,
 219 +                              struct inode *inode, const void *, const void *);
 220  extern int ext4_htree_fill_tree(struct file *dir_file, __u32 start_hash,
 221                                 __u32 start_minor_hash, __u32 *next_hash);
 222  extern int ext4_search_dir(struct buffer_head *bh,
 223 @@ -3862,6 +3924,36 @@ static inline int ext4_buffer_uptodate(struct buffer_head *bh)
 224         return buffer_uptodate(bh);
 225  }
 226
 227 +/*
 228 + * Compute the total directory entry data length.
 229 + * This includes the filename and an implicit NUL terminator (always present),
 230 + * and optional extensions.  Each extension has a bit set in the high 4 bits of
 231 + * de->file_type, and the extension length is the first byte in each entry.
 232 + */
 233 +static inline int ext4_get_dirent_data_len(struct ext4_dir_entry_2 *de)
 234 +{
 235 +       char *len = de->name + de->name_len + 1 /* NUL terminator */;
 236 +       int dlen = 0;
 237 +       __u8 extra_data_flags = (de->file_type & ~EXT4_FT_MASK) >> 4;
 238 +       struct ext4_dir_entry_tail *t = (struct ext4_dir_entry_tail *)de;
 239 +
 240 +       if (!t->det_reserved_zero1 &&
 241 +           le16_to_cpu(t->det_rec_len) ==
 242 +               sizeof(struct ext4_dir_entry_tail) &&
 243 +           !t->det_reserved_zero2 &&
 244 +           t->det_reserved_ft == EXT4_FT_DIR_CSUM)
 245 +               return 0;
 246 +
 247 +       while (extra_data_flags) {
 248 +               if (extra_data_flags & 1) {
 249 +                       dlen += *len + (dlen == 0);
 250 +                       len += *len;
 251 +               }
 252 +               extra_data_flags >>= 1;
 253 +       }
 254 +       return dlen;
 255 +}
 256 +
 257  #endif /* __KERNEL__ */
 258
 259  #define EFSBADCRC      EBADMSG         /* Bad CRC detected */
 260 diff --git a/fs/ext4/fast_commit.c b/fs/ext4/fast_commit.c
 261 index 276d9e6..3be0f08 100644
 262 --- a/fs/ext4/fast_commit.c
 263 +++ b/fs/ext4/fast_commit.c
 264 @@ -1596,7 +1596,7 @@ static int ext4_fc_replay_create(struct super_block *sb, struct ext4_fc_tl *tl,
 265                         jbd_debug(1, "Dir %d not found.", darg.ino);
 266                         goto out;
 267                 }
 268 -               ret = ext4_init_new_dir(NULL, dir, inode);
 269 +               ret = ext4_init_new_dir(NULL, dir, inode, NULL, NULL);
 270                 iput(dir);
 271                 if (ret) {
 272                         ret = 0;
 273 diff --git a/fs/ext4/inline.c b/fs/ext4/inline.c
 274 index 9626c31..ed31b5c 100644
 275 --- a/fs/ext4/inline.c
 276 +++ b/fs/ext4/inline.c
 277 @@ -1029,7 +1029,7 @@ static int ext4_add_dirent_to_inline(handle_t *handle,
 278         struct ext4_dir_entry_2 *de;
 279
 280         err = ext4_find_dest_de(dir, inode, iloc->bh, inline_start,
 281 -                               inline_size, fname, &de);
 282 +                               inline_size, fname, &de, NULL);
 283         if (err)
 284                 return err;
 285
 286 @@ -1038,7 +1038,7 @@ static int ext4_add_dirent_to_inline(handle_t *handle,
 287                                             EXT4_JTR_NONE);
 288         if (err)
 289                 return err;
 290 -       ext4_insert_dentry(dir, inode, de, inline_size, fname);
 291 +       ext4_insert_dentry(dir, inode, de, inline_size, fname, NULL);
 292
 293         ext4_show_inline_dir(dir, iloc->bh, inline_start, inline_size);
 294
 295 @@ -1396,7 +1396,7 @@ int ext4_inlinedir_to_tree(struct file *dir_file,
 296                         fake.name_len = 1;
 297                         strcpy(fake.name, ".");
 298                         fake.rec_len = ext4_rec_len_to_disk(
 299 -                                         ext4_dir_rec_len(fake.name_len, NULL),
 300 +                                         EXT4_DIR_ENTRY_LEN(&fake, NULL),
 301                                           inline_size);
 302                         ext4_set_de_type(inode->i_sb, &fake, S_IFDIR);
 303                         de = &fake;
 304 @@ -1406,7 +1406,7 @@ int ext4_inlinedir_to_tree(struct file *dir_file,
 305                         fake.name_len = 2;
 306                         strcpy(fake.name, "..");
 307                         fake.rec_len = ext4_rec_len_to_disk(
 308 -                                         ext4_dir_rec_len(fake.name_len, NULL),
 309 +                                         EXT4_DIR_ENTRY_LEN(&fake, NULL),
 310                                           inline_size);
 311                         ext4_set_de_type(inode->i_sb, &fake, S_IFDIR);
 312                         de = &fake;
 313 diff --git a/fs/ext4/namei.c b/fs/ext4/namei.c
 314 index 7f00dc3..51c950b 100644
 315 --- a/fs/ext4/namei.c
 316 +++ b/fs/ext4/namei.c
 317 @@ -285,13 +285,14 @@ static unsigned dx_get_count(struct dx_entry *entries);
 318  static unsigned dx_get_limit(struct dx_entry *entries);
 319  static void dx_set_count(struct dx_entry *entries, unsigned value);
 320  static void dx_set_limit(struct dx_entry *entries, unsigned value);
 321 -static unsigned dx_root_limit(struct inode *dir, unsigned infosize);
 322 +static inline unsigned dx_root_limit(struct inode *dir,
 323 +               struct ext4_dir_entry_2 *dot_de, unsigned infosize);
 324  static unsigned dx_node_limit(struct inode *dir);
 325  static struct dx_frame *dx_probe(struct ext4_filename *fname,
 326                                  struct inode *dir,
 327                                  struct dx_hash_info *hinfo,
 328                                  struct dx_frame *frame);
 329 -static void dx_release(struct dx_frame *frames);
 330 +static void dx_release(struct dx_frame *frames, struct inode *dir);
 331  static int dx_make_map(struct inode *dir, struct buffer_head *bh,
 332                        struct dx_hash_info *hinfo,
 333                        struct dx_map_entry *map_tail);
 334 @@ -431,22 +432,23 @@ static struct dx_countlimit *get_dx_countlimit(struct inode *inode,
 335  {
 336         struct ext4_dir_entry *dp;
 337         struct dx_root_info *root;
 338 -       int count_offset;
 339 +       int count_offset, dot_rec_len, dotdot_rec_len;
 340
 341         if (le16_to_cpu(dirent->rec_len) == EXT4_BLOCK_SIZE(inode->i_sb))
 342                 count_offset = 8;
 343 -       else if (le16_to_cpu(dirent->rec_len) == 12) {
 344 -               dp = (struct ext4_dir_entry *)(((void *)dirent) + 12);
 345 +       else {
 346 +               dot_rec_len = le16_to_cpu(dirent->rec_len);
 347 +               dp = (struct ext4_dir_entry *)(((void *)dirent) + dot_rec_len);
 348                 if (le16_to_cpu(dp->rec_len) !=
 349 -                   EXT4_BLOCK_SIZE(inode->i_sb) - 12)
 350 +                   EXT4_BLOCK_SIZE(inode->i_sb) - dot_rec_len)
 351                         return NULL;
 352 -               root = (struct dx_root_info *)(((void *)dp + 12));
 353 +               dotdot_rec_len = EXT4_DIR_ENTRY_LEN((struct ext4_dir_entry_2 *)dp, NULL);
 354 +               root = (struct dx_root_info *)(((void *)dp + dotdot_rec_len));
 355                 if (root->reserved_zero ||
 356                     root->info_length != sizeof(struct dx_root_info))
 357                         return NULL;
 358 -               count_offset = 32;
 359 -       } else
 360 -               return NULL;
 361 +               count_offset = 8 + dot_rec_len + dotdot_rec_len;
 362 +       }
 363
 364         if (offset)
 365                 *offset = count_offset;
 366 @@ -549,13 +551,14 @@ ext4_next_entry(struct ext4_dir_entry_2 *p, unsigned long blocksize)
 367   * Future: use high four bits of block for coalesce-on-delete flags
 368   * Mask them off for now.
 369   */
 370 -struct dx_root_info *dx_get_dx_info(struct ext4_dir_entry_2 *de)
 371 +struct dx_root_info *dx_get_dx_info(struct ext4_dir_entry_2 *de, struct inode *i_dir)
 372  {
 373 +       BUG_ON(de->name_len != 1);
 374         /* get dotdot first */
 375 -       de = (struct ext4_dir_entry_2 *)((char *)de + EXT4_DIR_REC_LEN(1));
 376 +       de = (struct ext4_dir_entry_2 *)((char *)de + EXT4_DIR_ENTRY_LEN(de, i_dir));
 377
 378         /* dx root info is after dotdot entry */
 379 -       de = (struct ext4_dir_entry_2 *)((char *)de + EXT4_DIR_REC_LEN(2));
 380 +       de = (struct ext4_dir_entry_2 *)((char *)de + EXT4_DIR_ENTRY_LEN(de, i_dir));
 381
 382         return (struct dx_root_info *)de;
 383  }
 384 @@ -600,11 +603,16 @@ static inline void dx_set_limit(struct dx_entry *entries, unsigned value)
 385         ((struct dx_countlimit *) entries)->limit = cpu_to_le16(value);
 386  }
 387
 388 -static inline unsigned dx_root_limit(struct inode *dir, unsigned infosize)
 389 +static inline unsigned dx_root_limit(struct inode *dir,
 390 +               struct ext4_dir_entry_2 *dot_de, unsigned infosize)
 391  {
 392 -       unsigned int entry_space = dir->i_sb->s_blocksize -
 393 -                       ext4_dir_rec_len(1, NULL) -
 394 -                       ext4_dir_rec_len(2, NULL) - infosize;
 395 +       struct ext4_dir_entry_2 *dotdot_de;
 396 +       unsigned entry_space;
 397 +
 398 +       BUG_ON(dot_de->name_len != 1);
 399 +       dotdot_de = ext4_next_entry(dot_de, dir->i_sb->s_blocksize);
 400 +       entry_space = dir->i_sb->s_blocksize - EXT4_DIR_ENTRY_LEN(dot_de, NULL) -
 401 +                        EXT4_DIR_ENTRY_LEN(dotdot_de, NULL) - infosize;
 402
 403         if (ext4_has_metadata_csum(dir->i_sb))
 404                 entry_space -= sizeof(struct dx_tail);
 405 @@ -722,7 +730,7 @@ static struct stats dx_show_leaf(struct inode *dir,
 406                                        (unsigned) ((char *) de - base));
 407  #endif
 408                         }
 409 -                       space += ext4_dir_rec_len(de->name_len, dir);
 410 +                       space += EXT4_DIR_ENTRY_LEN(de, dir);
 411                         names++;
 412                 }
 413                 de = ext4_next_entry(de, size);
 414 @@ -816,7 +824,7 @@ dx_probe(struct ext4_filename *fname, struct inode *dir,
 415         if (IS_ERR(frame->bh))
 416                 return (struct dx_frame *) frame->bh;
 417
 418 -       info = dx_get_dx_info((struct ext4_dir_entry_2 *)frame->bh->b_data);
 419 +       info = dx_get_dx_info((struct ext4_dir_entry_2 *)frame->bh->b_data, dir);
 420         if (info->hash_version != DX_HASH_TEA &&
 421             info->hash_version != DX_HASH_HALF_MD4 &&
 422             info->hash_version != DX_HASH_LEGACY &&
 423 @@ -872,11 +880,14 @@ dx_probe(struct ext4_filename *fname, struct inode *dir,
 424
 425         entries = (struct dx_entry *)(((char *)info) + info->info_length);
 426
 427 -       if (dx_get_limit(entries) != dx_root_limit(dir,
 428 -                                                  info->info_length)) {
 429 +       if (dx_get_limit(entries) !=
 430 +           dx_root_limit(dir, (struct ext4_dir_entry_2 *)frame->bh->b_data,
 431 +                         info->info_length)) {
 432                 ext4_warning_inode(dir, "dx entry: limit %u != root limit %u",
 433                                    dx_get_limit(entries),
 434 -                                  dx_root_limit(dir, info->info_length));
 435 +                                  dx_root_limit(dir,
 436 +                                         (struct ext4_dir_entry_2 *)frame->bh->b_data,
 437 +                                         info->info_length));
 438                 goto fail;
 439         }
 440
 441 @@ -953,7 +964,7 @@ fail:
 442         return ret_err;
 443  }
 444
 445 -static void dx_release(struct dx_frame *frames)
 446 +static void dx_release(struct dx_frame *frames, struct inode *dir)
 447  {
 448         struct dx_root_info *info;
 449         int i;
 450 @@ -962,7 +973,7 @@ static void dx_release(struct dx_frame *frames)
 451         if (frames[0].bh == NULL)
 452                 return;
 453
 454 -       info = dx_get_dx_info((struct ext4_dir_entry_2 *)frames[0].bh->b_data);
 455 +       info = dx_get_dx_info((struct ext4_dir_entry_2 *)frames[0].bh->b_data, dir);
 456         /* save local copy, "info" may be freed after brelse() */
 457         indirect_levels = info->indirect_levels;
 458         for (i = 0; i <= indirect_levels; i++) {
 459 @@ -1263,12 +1274,12 @@ int ext4_htree_fill_tree(struct file *dir_file, __u32 start_hash,
 460                     (count && ((hashval & 1) == 0)))
 461                         break;
 462         }
 463 -       dx_release(frames);
 464 +       dx_release(frames, dir);
 465         dxtrace(printk(KERN_DEBUG "Fill tree: returned %d entries, "
 466                        "next hash: %x\n", count, *next_hash));
 467         return count;
 468  errout:
 469 -       dx_release(frames);
 470 +       dx_release(frames, dir);
 471         return (err);
 472  }
 473
 474 @@ -1801,7 +1812,7 @@ static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
 475  errout:
 476         dxtrace(printk(KERN_DEBUG "%s not found\n", fname->usr_fname->name));
 477  success:
 478 -       dx_release(frames);
 479 +       dx_release(frames, dir);
 480         return bh;
 481  }
 482
 483 @@ -1925,7 +1936,7 @@ dx_move_dirents(struct inode *dir, char *from, char *to,
 484         while (count--) {
 485                 struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *)
 486                                                 (from + (map->offs<<2));
 487 -               rec_len = ext4_dir_rec_len(de->name_len, dir);
 488 +               rec_len = EXT4_DIR_ENTRY_LEN(de, dir);
 489
 490                 memcpy (to, de, rec_len);
 491                 ((struct ext4_dir_entry_2 *) to)->rec_len =
 492 @@ -1958,7 +1969,7 @@ static struct ext4_dir_entry_2 *dx_pack_dirents(struct inode *dir, char *base,
 493         while ((char*)de < base + blocksize) {
 494                 next = ext4_next_entry(de, blocksize);
 495                 if (de->inode && de->name_len) {
 496 -                       rec_len = ext4_dir_rec_len(de->name_len, dir);
 497 +                       rec_len = EXT4_DIR_ENTRY_LEN(de, dir);
 498                         if (de > to)
 499                                 memmove(to, de, rec_len);
 500                         to->rec_len = ext4_rec_len_to_disk(rec_len, blocksize);
 501 @@ -2101,14 +2112,21 @@ int ext4_find_dest_de(struct inode *dir, struct inode *inode,
 502                       struct buffer_head *bh,
 503                       void *buf, int buf_size,
 504                       struct ext4_filename *fname,
 505 -                     struct ext4_dir_entry_2 **dest_de)
 506 +                     struct ext4_dir_entry_2 **dest_de,
 507 +                     int *dlen)
 508  {
 509         struct ext4_dir_entry_2 *de;
 510 -       unsigned short reclen = ext4_dir_rec_len(fname_len(fname), dir);
 511 +       unsigned short reclen;
 512         int nlen, rlen;
 513         unsigned int offset = 0;
 514         char *top;
 515
 516 +       if (dlen) {
 517 +               reclen = ext4_dir_rec_len(fname_len(fname) + *dlen, dir);
 518 +               *dlen = 0;
 519 +       } else {
 520 +               reclen = ext4_dir_rec_len(fname_len(fname), dir);
 521 +       }
 522         de = (struct ext4_dir_entry_2 *)buf;
 523         top = buf + buf_size - reclen;
 524         while ((char *) de <= top) {
 525 @@ -2117,10 +2135,31 @@ int ext4_find_dest_de(struct inode *dir, struct inode *inode,
 526                         return -EFSCORRUPTED;
 527                 if (ext4_match(dir, fname, de))
 528                         return -EEXIST;
 529 -               nlen = ext4_dir_rec_len(de->name_len, dir);
 530 +               nlen = EXT4_DIR_ENTRY_LEN(de, dir);
 531                 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
 532                 if ((de->inode ? rlen - nlen : rlen) >= reclen)
 533                         break;
 534 +
 535 +               /* Then for dotdot entries, check for the smaller space
 536 +                * required for just the entry, no FID
 537 +                */
 538 +               if (fname_len(fname) == 2 && memcmp(fname_name(fname), "..", 2) == 0) {
 539 +                       if ((de->inode ? rlen - nlen : rlen) >=
 540 +                           ext4_dir_rec_len(fname_len(fname), dir)) {
 541 +                               /* set dlen = 1 to indicate not
 542 +                                * enough space store fid
 543 +                                */
 544 +                               if (dlen)
 545 +                                       *dlen = 1;
 546 +                               break;
 547 +                       }
 548 +                       /* The new ".." entry must be written over the
 549 +                        * previous ".." entry, which is the first
 550 +                        * entry traversed by this scan. If it doesn't
 551 +                        * fit, something is badly wrong, so -EIO.
 552 +                        */
 553 +                       return -EIO;
 554 +               }
 555                 de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
 556                 offset += rlen;
 557         }
 558 @@ -2135,12 +2174,13 @@ void ext4_insert_dentry(struct inode *dir,
 559                         struct inode *inode,
 560                         struct ext4_dir_entry_2 *de,
 561                         int buf_size,
 562 -                       struct ext4_filename *fname)
 563 +                       struct ext4_filename *fname,
 564 +                       void *data)
 565  {
 566
 567         int nlen, rlen;
 568
 569 -       nlen = ext4_dir_rec_len(de->name_len, dir);
 570 +       nlen = EXT4_DIR_ENTRY_LEN(de, dir);
 571         rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
 572         if (de->inode) {
 573                 struct ext4_dir_entry_2 *de1 =
 574 @@ -2161,6 +2201,12 @@ void ext4_insert_dentry(struct inode *dir,
 575                 EXT4_DIRENT_HASHES(de)->minor_hash =
 576                                                 cpu_to_le32(hinfo->minor_hash);
 577         }
 578 +       if (data) {
 579 +               de->name[fname_len(fname)] = 0;
 580 +               memcpy(&de->name[fname_len(fname) + 1], data, *(char *)data);
 581 +               de->file_type |= EXT4_DIRENT_LUFID;
 582 +       }
 583 +
 584  }
 585
 586  /*
 587 @@ -2178,14 +2224,19 @@ static int add_dirent_to_buf(handle_t *handle, struct ext4_filename *fname,
 588  {
 589         unsigned int    blocksize = dir->i_sb->s_blocksize;
 590         int             csum_size = 0;
 591 -       int             err, err2;
 592 +       int             err, err2, dlen = 0;
 593 +       unsigned char   *data;
 594
 595 +       data = ext4_dentry_get_data(inode->i_sb, (struct ext4_dentry_param *)
 596 +                                               EXT4_I(inode)->i_dirdata);
 597         if (ext4_has_metadata_csum(inode->i_sb))
 598                 csum_size = sizeof(struct ext4_dir_entry_tail);
 599
 600         if (!de) {
 601 +               if (data)
 602 +                       dlen = (*data) + 1;
 603                 err = ext4_find_dest_de(dir, inode, bh, bh->b_data,
 604 -                                       blocksize - csum_size, fname, &de);
 605 +                                       blocksize - csum_size, fname, &de, &dlen);
 606                 if (err)
 607                         return err;
 608         }
 609 @@ -2198,7 +2249,10 @@ static int add_dirent_to_buf(handle_t *handle, struct ext4_filename *fname,
 610         }
 611
 612         /* By now the buffer is marked for journaling */
 613 -       ext4_insert_dentry(dir, inode, de, blocksize, fname);
 614 +       /* If writing the short form of "dotdot", don't add the data section */
 615 +       if (dlen == 1)
 616 +               data = NULL;
 617 +       ext4_insert_dentry(dir, inode, de, blocksize, fname, data);
 618
 619         /*
 620          * XXX shouldn't update any times until successful
 621 @@ -2296,7 +2350,7 @@ static int make_indexed_dir(handle_t *handle, struct ext4_filename *fname,
 622                                      blocksize);
 623
 624         /* initialize hashing info */
 625 -       dx_info = dx_get_dx_info(dot_de);
 626 +       dx_info = dx_get_dx_info(dot_de, dir);
 627         memset(dx_info, 0, sizeof(*dx_info));
 628         dx_info->info_length = sizeof(*dx_info);
 629         if (ext4_hash_in_dirent(dir))
 630 @@ -2307,7 +2361,8 @@ static int make_indexed_dir(handle_t *handle, struct ext4_filename *fname,
 631         entries = (void *)dx_info + sizeof(*dx_info);
 632         dx_set_block(entries, 1);
 633         dx_set_count(entries, 1);
 634 -       dx_set_limit(entries, dx_root_limit(dir, sizeof(*dx_info)));
 635 +       dx_set_limit(entries, dx_root_limit(dir,
 636 +                                        dot_de, sizeof(*dx_info)));
 637
 638         /* Initialize as for dx_probe */
 639         fname->hinfo.hash_version = dx_info->hash_version;
 640 @@ -2348,7 +2403,7 @@ out_frames:
 641          */
 642         if (retval)
 643                 ext4_mark_inode_dirty(handle, dir);
 644 -       dx_release(frames);
 645 +       dx_release(frames, dir);
 646         brelse(bh2);
 647         return retval;
 648  }
 649 @@ -2361,6 +2416,8 @@ static int ext4_update_dotdot(handle_t *handle, struct dentry *dentry,
 650         struct buffer_head *dir_block;
 651         struct ext4_dir_entry_2 *de;
 652         int len, journal = 0, err = 0;
 653 +       int dlen = 0;
 654 +       char *data;
 655
 656         if (IS_ERR(handle))
 657                 return PTR_ERR(handle);
 658 @@ -2376,21 +2433,25 @@ static int ext4_update_dotdot(handle_t *handle, struct dentry *dentry,
 659
 660         de = (struct ext4_dir_entry_2 *)dir_block->b_data;
 661         /* the first item must be "." */
 662 -       assert(de->name_len == 1 && de->name[0] == '.');
 663 +       ASSERT(de->name_len == 1 && de->name[0] == '.');
 664         len = le16_to_cpu(de->rec_len);
 665 -       assert(len >= EXT4_DIR_REC_LEN(1));
 666 -       if (len > EXT4_DIR_REC_LEN(1)) {
 667 +       ASSERT(len >= EXT4_DIR_REC_LEN(1, dir));
 668 +       if (len > EXT4_DIR_REC_LEN(1, dir)) {
 669                 BUFFER_TRACE(dir_block, "get_write_access");
 670                 err = ext4_journal_get_write_access(handle, dir->i_sb, dir_block, EXT4_JTR_NONE);
 671                 if (err)
 672                         goto out_journal;
 673
 674                 journal = 1;
 675 -               de->rec_len = cpu_to_le16(EXT4_DIR_REC_LEN(1));
 676 +               de->rec_len = cpu_to_le16(EXT4_DIR_REC_LEN(1, dir));
 677         }
 678
 679 -       len -= EXT4_DIR_REC_LEN(1);
 680 -       assert(len == 0 || len >= EXT4_DIR_REC_LEN(2));
 681 +       len -= EXT4_DIR_REC_LEN(1, dir);
 682 +       data = ext4_dentry_get_data(dir->i_sb,
 683 +                       (struct ext4_dentry_param *)dentry->d_fsdata);
 684 +       if (data)
 685 +               dlen = *data + 1;
 686 +       ASSERT(len == 0 || len >= EXT4_DIR_REC_LEN(2 + dlen, dir));
 687         de = (struct ext4_dir_entry_2 *)
 688                         ((char *) de + le16_to_cpu(de->rec_len));
 689         if (!journal) {
 690 @@ -2404,10 +2465,15 @@ static int ext4_update_dotdot(handle_t *handle, struct dentry *dentry,
 691         if (len > 0)
 692                 de->rec_len = cpu_to_le16(len);
 693         else
 694 -               assert(le16_to_cpu(de->rec_len) >= EXT4_DIR_REC_LEN(2));
 695 +               ASSERT(le16_to_cpu(de->rec_len) >= EXT4_DIR_REC_LEN(2, dir));
 696         de->name_len = 2;
 697         strcpy(de->name, "..");
 698 -       ext4_set_de_type(dir->i_sb, de, S_IFDIR);
 699 +       if (data != NULL && ext4_get_dirent_data_len(de) >= dlen) {
 700 +               de->name[2] = 0;
 701 +               memcpy(&de->name[2 + 1], data, *data);
 702 +               ext4_set_de_type(dir->i_sb, de, S_IFDIR);
 703 +               de->file_type |= EXT4_DIRENT_LUFID;
 704 +       }
 705
 706  out_journal:
 707         if (journal) {
 708 @@ -2445,6 +2511,7 @@ static int ext4_add_entry(handle_t *handle, struct dentry *dentry,
 709         ext4_lblk_t block, blocks;
 710         int     csum_size = 0;
 711
 712 +       EXT4_I(inode)->i_dirdata = dentry->d_fsdata;
 713         if (ext4_has_metadata_csum(inode->i_sb))
 714                 csum_size = sizeof(struct ext4_dir_entry_tail);
 715
 716 @@ -2687,7 +2754,7 @@ again:
 717                         dx_set_count(entries, 1);
 718                         dx_set_block(entries + 0, newblock);
 719                         info = dx_get_dx_info((struct ext4_dir_entry_2 *)
 720 -                                             frames[0].bh->b_data);
 721 +                                             frames[0].bh->b_data, dir);
 722                         info->indirect_levels = 1;
 723                         dxtrace(printk(KERN_DEBUG
 724                                        "Creating %d level index...\n",
 725 @@ -2713,7 +2780,7 @@ journal_error:
 726         ext4_std_error(dir->i_sb, err); /* this is a no-op if err == 0 */
 727  cleanup:
 728         brelse(bh);
 729 -       dx_release(frames);
 730 +       dx_release(frames, dir);
 731         /* @restart is true means htree-path has been changed, we need to
 732          * repeat dx_probe() to find out valid htree-path
 733          */
 734 @@ -3016,38 +3083,73 @@ err_unlock_inode:
 735         return err;
 736  }
 737
 738 +struct tp_block {
 739 +       struct inode *inode;
 740 +       void *data1;
 741 +       void *data2;
 742 +};
 743 +
 744  struct ext4_dir_entry_2 *ext4_init_dot_dotdot(struct inode *inode,
 745                           struct ext4_dir_entry_2 *de,
 746                           int blocksize, int csum_size,
 747                           unsigned int parent_ino, int dotdot_real_len)
 748  {
 749 +       void *data1 = NULL, *data2 = NULL;
 750 +       int dot_reclen = 0;
 751 +
 752 +       if (dotdot_real_len == 10) {
 753 +               struct tp_block *tpb = (struct tp_block *)inode;
 754 +               data1 = tpb->data1;
 755 +               data2 = tpb->data2;
 756 +               inode = tpb->inode;
 757 +               dotdot_real_len = 0;
 758 +       }
 759         de->inode = cpu_to_le32(inode->i_ino);
 760         de->name_len = 1;
 761 -       de->rec_len = ext4_rec_len_to_disk(ext4_dir_rec_len(de->name_len, NULL),
 762 -                                          blocksize);
 763         strcpy(de->name, ".");
 764         ext4_set_de_type(inode->i_sb, de, S_IFDIR);
 765
 766 +       /* get packed fid data*/
 767 +       data1 = ext4_dentry_get_data(inode->i_sb,
 768 +                               (struct ext4_dentry_param *) data1);
 769 +       if (data1) {
 770 +               de->name[1] = 0;
 771 +               memcpy(&de->name[2], data1, *(char *) data1);
 772 +               de->file_type |= EXT4_DIRENT_LUFID;
 773 +       }
 774 +       de->rec_len = cpu_to_le16(EXT4_DIR_ENTRY_LEN(de, NULL));
 775 +
 776 +       dot_reclen = cpu_to_le16(de->rec_len);
 777         de = ext4_next_entry(de, blocksize);
 778         de->inode = cpu_to_le32(parent_ino);
 779         de->name_len = 2;
 780 +
 781 +       strcpy(de->name, "..");
 782 +       ext4_set_de_type(inode->i_sb, de, S_IFDIR);
 783 +       data2 = ext4_dentry_get_data(inode->i_sb,
 784 +                       (struct ext4_dentry_param *) data2);
 785 +       if (data2) {
 786 +               de->name[2] = 0;
 787 +               memcpy(&de->name[3], data2, *(char *) data2);
 788 +               de->file_type |= EXT4_DIRENT_LUFID;
 789 +       }
 790 +
 791         if (!dotdot_real_len)
 792                 de->rec_len = ext4_rec_len_to_disk(blocksize -
 793 -                                       (csum_size + ext4_dir_rec_len(1, NULL)),
 794 -                                       blocksize);
 795 +                                       (csum_size + dot_reclen), blocksize);
 796         else
 797                 de->rec_len = ext4_rec_len_to_disk(
 798 -                                       ext4_dir_rec_len(de->name_len, NULL),
 799 +                                       EXT4_DIR_ENTRY_LEN(de, NULL),
 800                                         blocksize);
 801 -       strcpy(de->name, "..");
 802 -       ext4_set_de_type(inode->i_sb, de, S_IFDIR);
 803
 804         return ext4_next_entry(de, blocksize);
 805  }
 806
 807  int ext4_init_new_dir(handle_t *handle, struct inode *dir,
 808 -                            struct inode *inode)
 809 +                            struct inode *inode,
 810 +                            const void *data1, const void *data2)
 811  {
 812 +       struct tp_block param;
 813         struct buffer_head *dir_block = NULL;
 814         struct ext4_dir_entry_2 *de;
 815         ext4_lblk_t block = 0;
 816 @@ -3071,7 +3173,11 @@ int ext4_init_new_dir(handle_t *handle, struct inode *dir,
 817         if (IS_ERR(dir_block))
 818                 return PTR_ERR(dir_block);
 819         de = (struct ext4_dir_entry_2 *)dir_block->b_data;
 820 -       ext4_init_dot_dotdot(inode, de, blocksize, csum_size, dir->i_ino, 0);
 821 +       param.inode = inode;
 822 +       param.data1 = (void *)data1;
 823 +       param.data2 = (void *)data2;
 824 +       ext4_init_dot_dotdot((struct inode *)(&param), de, blocksize,
 825 +                            csum_size, dir->i_ino, 10);
 826         set_nlink(inode, 2);
 827         if (csum_size)
 828                 ext4_initialize_dirent_tail(dir_block, blocksize);
 829 @@ -3086,6 +3192,29 @@ out:
 830         return err;
 831  }
 832
 833 +/* Initialize @inode as a subdirectory of @dir, and add the
 834 + * "." and ".." entries into the first directory block. */
 835 +int ext4_add_dot_dotdot(handle_t *handle, struct inode *dir,
 836 +                       struct inode *inode,
 837 +                       const void *data1, const void *data2)
 838 +{
 839 +       int rc;
 840 +
 841 +       if (IS_ERR(handle))
 842 +               return PTR_ERR(handle);
 843 +
 844 +       if (IS_DIRSYNC(dir))
 845 +               ext4_handle_sync(handle);
 846 +
 847 +       inode->i_op = &ext4_dir_inode_operations;
 848 +       inode->i_fop = &ext4_dir_operations;
 849 +       rc = ext4_init_new_dir(handle, dir, inode, data1, data2);
 850 +       if (!rc)
 851 +               rc = ext4_mark_inode_dirty(handle, inode);
 852 +       return rc;
 853 +}
 854 +EXPORT_SYMBOL(ext4_add_dot_dotdot);
 855 +
 856  static int ext4_mkdir(struct user_namespace *mnt_userns, struct inode *dir,
 857                       struct dentry *dentry, umode_t mode)
 858  {
 859 @@ -3113,7 +3242,7 @@ retry:
 860
 861         inode->i_op = &ext4_dir_inode_operations;
 862         inode->i_fop = &ext4_dir_operations;
 863 -       err = ext4_init_new_dir(handle, dir, inode);
 864 +       err = ext4_init_new_dir(handle, dir, inode, NULL, NULL);
 865         if (err)
 866                 goto out_clear_inode;
 867         err = ext4_mark_inode_dirty(handle, inode);
 868 diff --git a/fs/ext4/super.c b/fs/ext4/super.c
 869 index 4be1994..a2fcbf8 100644
 870 --- a/fs/ext4/super.c
 871 +++ b/fs/ext4/super.c
 872 @@ -1672,7 +1672,7 @@ enum {
 873         Opt_inlinecrypt,
 874         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
 875         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_jqfmt_vfsv1, Opt_quota,
 876 -       Opt_noquota, Opt_barrier, Opt_nobarrier, Opt_err,
 877 +       Opt_noquota, Opt_barrier, Opt_nobarrier, Opt_err, Opt_dirdata,
 878         Opt_usrquota, Opt_grpquota, Opt_prjquota, Opt_i_version,
 879         Opt_dax, Opt_dax_always, Opt_dax_inode, Opt_dax_never,
 880         Opt_stripe, Opt_delalloc, Opt_nodelalloc, Opt_warn_on_error,
 881 @@ -1756,6 +1756,7 @@ static const match_table_t tokens = {
 882         {Opt_nolazytime, "nolazytime"},
 883         {Opt_debug_want_extra_isize, "debug_want_extra_isize=%u"},
 884         {Opt_nodelalloc, "nodelalloc"},
 885 +       {Opt_dirdata, "dirdata"},
 886         {Opt_removed, "mblk_io_submit"},
 887         {Opt_removed, "nomblk_io_submit"},
 888         {Opt_block_validity, "block_validity"},
 889 @@ -2000,6 +2001,7 @@ static const struct mount_opts {
 890         {Opt_usrjquota, 0, MOPT_Q | MOPT_STRING},
 891         {Opt_grpjquota, 0, MOPT_Q | MOPT_STRING},
 892         {Opt_offusrjquota, 0, MOPT_Q},
 893 +       {Opt_dirdata, EXT4_MOUNT_DIRDATA, MOPT_SET},
 894         {Opt_offgrpjquota, 0, MOPT_Q},
 895         {Opt_jqfmt_vfsold, QFMT_VFS_OLD, MOPT_QFMT},
 896         {Opt_jqfmt_vfsv0, QFMT_VFS_V0, MOPT_QFMT},
 897 --
 898 2.34.1
 899