ldiskfs/kernel_patches/patches/rhel9/ext4-data-in-dirent.patch

   1 this patch implements feature which allows ext4 fs users (e.g. Lustre)
   2 to store data in ext4 dirent.
   3 data is stored in ext4 dirent after file-name, this space is accounted
   4 in de->rec_len. flag EXT4_DIRENT_LUFID added to d_type if extra data
   5 is present.
   6
   7 make use of dentry->d_fsdata to pass fid to ext4. so no
   8 changes in ext4_add_entry() interface required.
   9
  10 ---
  11  fs/ext4/dir.c         |   9 +-
  12  fs/ext4/ext4.h        | 107 ++++++++++++++++--
  13  fs/ext4/fast_commit.c |   2 +-
  14  fs/ext4/inline.c      |   8 +-
  15  fs/ext4/namei.c       | 250 ++++++++++++++++++++++++++++++++----------
  16  fs/ext4/super.c       |   4 +-
  17  6 files changed, 305 insertions(+), 75 deletions(-)
  18
  19 diff --git a/fs/ext4/dir.c b/fs/ext4/dir.c
  20 index a6bb86f..edd0540 100644
  21 --- a/fs/ext4/dir.c
  22 +++ b/fs/ext4/dir.c
  23 @@ -465,12 +465,17 @@ int ext4_htree_store_dirent(struct file *dir_file, __u32 hash,
  24         struct fname *fname, *new_fn;
  25         struct dir_private_info *info;
  26         int len;
  27 +       int extra_data = 0;
  28
  29         info = dir_file->private_data;
  30         p = &info->root.rb_node;
  31
  32         /* Create and allocate the fname structure */
  33 -       len = sizeof(struct fname) + ent_name->len + 1;
  34 +       if (dirent->file_type & EXT4_DIRENT_LUFID)
  35 +               extra_data = ext4_get_dirent_data_len(dirent);
  36 +
  37 +       len = sizeof(struct fname) + ent_name->len + extra_data + 1;
  38 +
  39         new_fn = kzalloc(len, GFP_KERNEL);
  40         if (!new_fn)
  41                 return -ENOMEM;
  42 @@ -479,7 +484,7 @@ int ext4_htree_store_dirent(struct file *dir_file, __u32 hash,
  43         new_fn->inode = le32_to_cpu(dirent->inode);
  44         new_fn->name_len = ent_name->len;
  45         new_fn->file_type = dirent->file_type;
  46 -       memcpy(new_fn->name, ent_name->name, ent_name->len);
  47 +       memcpy(new_fn->name, ent_name->name, ent_name->len + extra_data);
  48
  49         while (*p) {
  50                 parent = *p;
  51 diff --git a/fs/ext4/ext4.h b/fs/ext4/ext4.h
  52 index 58645be..8c6864c 100644
  53 --- a/fs/ext4/ext4.h
  54 +++ b/fs/ext4/ext4.h
  55 @@ -1171,6 +1171,7 @@ struct ext4_inode_info {
  56         __u32 i_csum_seed;
  57
  58         kprojid_t i_projid;
  59 +       void *i_dirdata;
  60  };
  61
  62  /*
  63 @@ -1192,6 +1193,7 @@ struct ext4_inode_info {
  64   * Mount flags set via mount options or defaults
  65   */
  66  #define EXT4_MOUNT_NO_MBCACHE          0x00001 /* Do not use mbcache */
  67 +#define EXT4_MOUNT_DIRDATA             0x00002 /* Data in directory entries */
  68  #define EXT4_MOUNT_GRPID               0x00004 /* Create files with directory's group */
  69  #define EXT4_MOUNT_DEBUG               0x00008 /* Some debugging messages */
  70  #define EXT4_MOUNT_ERRORS_CONT         0x00010 /* Continue on errors */
  71 @@ -2164,6 +2166,7 @@ EXT4_FEATURE_INCOMPAT_FUNCS(casefold,             CASEFOLD)
  72                                          EXT4_FEATURE_INCOMPAT_FLEX_BG| \
  73                                          EXT4_FEATURE_INCOMPAT_EA_INODE| \
  74                                          EXT4_FEATURE_INCOMPAT_MMP | \
  75 +                                        EXT4_FEATURE_INCOMPAT_DIRDATA| \
  76                                          EXT4_FEATURE_INCOMPAT_INLINE_DATA | \
  77                                          EXT4_FEATURE_INCOMPAT_ENCRYPT | \
  78                                          EXT4_FEATURE_INCOMPAT_CASEFOLD | \
  79 @@ -2371,6 +2374,42 @@ struct ext4_dir_entry_tail {
  80  #define EXT4_FT_SYMLINK                7
  81
  82  #define EXT4_FT_MAX            8
  83 +#define EXT4_FT_MASK           0xf
  84 +
  85 +#if EXT4_FT_MAX > EXT4_FT_MASK
  86 +#error "conflicting EXT4_FT_MAX and EXT4_FT_MASK"
  87 +#endif
  88 +
  89 +/*
  90 + * d_type has 4 unused bits, so it can hold four types data. these different
  91 + * type of data (e.g. lustre data, high 32 bits of 64-bit inode number) can be
  92 + * stored, in flag order, after file-name in ext4 dirent.
  93 +*/
  94 +/*
  95 + * this flag is added to d_type if ext4 dirent has extra data after
  96 + * filename. this data length is variable and length is stored in first byte
  97 + * of data. data start after filename NUL byte.
  98 + * This is used by Lustre FS.
  99 +  */
 100 +#define EXT4_DIRENT_LUFID              0x10
 101 +
 102 +#define EXT4_LUFID_MAGIC    0xAD200907UL
 103 +struct ext4_dentry_param {
 104 +       __u32  edp_magic;       /* EXT4_LUFID_MAGIC */
 105 +       char   edp_len;         /* size of edp_data in bytes */
 106 +       char   edp_data[0];     /* packed array of data */
 107 +} __packed;
 108 +
 109 +static inline unsigned char *ext4_dentry_get_data(struct super_block *sb,
 110 +                                                 struct ext4_dentry_param *p)
 111 +{
 112 +       if (!ext4_has_feature_dirdata(sb))
 113 +               return NULL;
 114 +       if (p && p->edp_magic == EXT4_LUFID_MAGIC)
 115 +               return &p->edp_len;
 116 +       else
 117 +               return NULL;
 118 +}
 119
 120  #define EXT4_FT_DIR_CSUM       0xDE
 121
 122 @@ -2382,6 +2421,17 @@ struct ext4_dir_entry_tail {
 123  #define EXT4_DIR_PAD                   4
 124  #define EXT4_DIR_ROUND                 (EXT4_DIR_PAD - 1)
 125  #define EXT4_MAX_REC_LEN               ((1<<16)-1)
 126 +#define EXT4_DIR_REC_LEN_(name_len, i_dir) \
 127 +       ext4_dir_rec_len((name_len), (i_dir))
 128 +#define EXT4_DIR_ENTRY_LEN_(de, i_dir) \
 129 +       (EXT4_DIR_REC_LEN_((de)->name_len + ext4_get_dirent_data_len(de), \
 130 +               (i_dir)))
 131 +/* ldiskfs */
 132 +#define EXT4_DIR_REC_LEN(name_len, i_dir)      EXT4_DIR_REC_LEN_((name_len), (i_dir))
 133 +#define EXT4_DIR_ENTRY_LEN(de, i_dir)          EXT4_DIR_ENTRY_LEN_((de), (i_dir))
 134 +/* lustre osd_handler compat -- ifdef LDISKFS_DIR_REC_LEN_WITH_DIR */
 135 +#define EXT4_DIR_REC_LEN_WITH_DIR              1
 136 +#define __EXT4_DIR_REC_LEN(name_len)           EXT4_DIR_REC_LEN_((name_len), NULL)
 137
 138  /*
 139   * The rec_len is dependent on the type of directory. Directories that are
 140 @@ -2389,10 +2439,10 @@ struct ext4_dir_entry_tail {
 141   * ext4_extended_dir_entry_2. For all entries related to '.' or '..' you should
 142   * pass NULL for dir, as those entries do not use the extra fields.
 143   */
 144 -static inline unsigned int ext4_dir_rec_len(__u8 name_len,
 145 +static inline unsigned int ext4_dir_rec_len(__u32 name_len,
 146                                                 const struct inode *dir)
 147  {
 148 -       int rec_len = (name_len + 8 + EXT4_DIR_ROUND);
 149 +       __u32 rec_len = (name_len + 8 + EXT4_DIR_ROUND);
 150
 151         if (dir && ext4_hash_in_dirent(dir))
 152                 rec_len += sizeof(struct ext4_dir_entry_hash);
 153 @@ -2865,11 +2915,13 @@ extern int ext4_find_dest_de(struct inode *dir, struct inode *inode,
 154                              struct buffer_head *bh,
 155                              void *buf, int buf_size,
 156                              struct ext4_filename *fname,
 157 -                            struct ext4_dir_entry_2 **dest_de);
 158 +                            struct ext4_dir_entry_2 **dest_de,
 159 +                            int *dlen);
 160  void ext4_insert_dentry(struct inode *dir, struct inode *inode,
 161                         struct ext4_dir_entry_2 *de,
 162                         int buf_size,
 163 -                       struct ext4_filename *fname);
 164 +                       struct ext4_filename *fname,
 165 +                       void *data);
 166  static inline void ext4_update_dx_flag(struct inode *inode)
 167  {
 168         if (!ext4_has_feature_dir_index(inode->i_sb) &&
 169 @@ -2885,10 +2937,17 @@ static const unsigned char ext4_filetype_table[] = {
 170
 171  static inline  unsigned char get_dtype(struct super_block *sb, int filetype)
 172  {
 173 -       if (!ext4_has_feature_filetype(sb) || filetype >= EXT4_FT_MAX)
 174 +       int fl_index = filetype & EXT4_FT_MASK;
 175 +
 176 +       if (!ext4_has_feature_filetype(sb) || fl_index >= EXT4_FT_MAX)
 177                 return DT_UNKNOWN;
 178
 179 -       return ext4_filetype_table[filetype];
 180 +       if (!test_opt(sb, DIRDATA))
 181 +               return ext4_filetype_table[fl_index];
 182 +
 183 +       return (ext4_filetype_table[fl_index]) |
 184 +               (filetype & EXT4_DIRENT_LUFID);
 185 +
 186  }
 187  extern int ext4_check_all_de(struct inode *dir, struct buffer_head *bh,
 188                              void *buf, int buf_size);
 189 @@ -3087,9 +3146,13 @@ extern int ext4_ind_migrate(struct inode *inode);
 190
 191  /* namei.c */
 192  extern int ext4_init_new_dir(handle_t *handle, struct inode *dir,
 193 -                            struct inode *inode);
 194 +                            struct inode *inode,
 195 +                            const void *data1, const void *data2);
 196  extern int ext4_dirblock_csum_verify(struct inode *inode,
 197                                      struct buffer_head *bh);
 198 +extern int ext4_add_dot_dotdot(handle_t *handle, struct inode *dir,
 199 +                              struct inode *inode,
 200 +                              const void *data1, const void *data2);
 201  extern int ext4_htree_fill_tree(struct file *dir_file, __u32 start_hash,
 202                                 __u32 start_minor_hash, __u32 *next_hash);
 203  extern struct inode *ext4_create_inode(handle_t *handle,
 204 @@ -3915,6 +3978,36 @@ static inline int ext4_buffer_uptodate(struct buffer_head *bh)
 205         return buffer_uptodate(bh);
 206  }
 207
 208 +/*
 209 + * Compute the total directory entry data length.
 210 + * This includes the filename and an implicit NUL terminator (always present),
 211 + * and optional extensions.  Each extension has a bit set in the high 4 bits of
 212 + * de->file_type, and the extension length is the first byte in each entry.
 213 + */
 214 +static inline int ext4_get_dirent_data_len(struct ext4_dir_entry_2 *de)
 215 +{
 216 +       char *len = de->name + de->name_len + 1 /* NUL terminator */;
 217 +       int dlen = 0;
 218 +       __u8 extra_data_flags = (de->file_type & ~EXT4_FT_MASK) >> 4;
 219 +       struct ext4_dir_entry_tail *t = (struct ext4_dir_entry_tail *)de;
 220 +
 221 +       if (!t->det_reserved_zero1 &&
 222 +           le16_to_cpu(t->det_rec_len) ==
 223 +               sizeof(struct ext4_dir_entry_tail) &&
 224 +           !t->det_reserved_zero2 &&
 225 +           t->det_reserved_ft == EXT4_FT_DIR_CSUM)
 226 +               return 0;
 227 +
 228 +       while (extra_data_flags) {
 229 +               if (extra_data_flags & 1) {
 230 +                       dlen += *len + (dlen == 0);
 231 +                       len += *len;
 232 +               }
 233 +               extra_data_flags >>= 1;
 234 +       }
 235 +       return dlen;
 236 +}
 237 +
 238  #endif /* __KERNEL__ */
 239
 240  #define EFSBADCRC      EBADMSG         /* Bad CRC detected */
 241 diff --git a/fs/ext4/fast_commit.c b/fs/ext4/fast_commit.c
 242 index 5ae8026..ecfa9b8 100644
 243 --- a/fs/ext4/fast_commit.c
 244 +++ b/fs/ext4/fast_commit.c
 245 @@ -1544,7 +1544,7 @@ static int ext4_fc_replay_create(struct super_block *sb, struct ext4_fc_tl *tl,
 246                         jbd_debug(1, "Dir %d not found.", darg.ino);
 247                         goto out;
 248                 }
 249 -               ret = ext4_init_new_dir(NULL, dir, inode);
 250 +               ret = ext4_init_new_dir(NULL, dir, inode, NULL, NULL);
 251                 iput(dir);
 252                 if (ret) {
 253                         ret = 0;
 254 diff --git a/fs/ext4/inline.c b/fs/ext4/inline.c
 255 index 39a1ab1..46f2e1e 100644
 256 --- a/fs/ext4/inline.c
 257 +++ b/fs/ext4/inline.c
 258 @@ -1036,7 +1036,7 @@ static int ext4_add_dirent_to_inline(handle_t *handle,
 259         struct ext4_dir_entry_2 *de;
 260
 261         err = ext4_find_dest_de(dir, inode, iloc->bh, inline_start,
 262 -                               inline_size, fname, &de);
 263 +                               inline_size, fname, &de, NULL);
 264         if (err)
 265                 return err;
 266
 267 @@ -1045,7 +1045,7 @@ static int ext4_add_dirent_to_inline(handle_t *handle,
 268                                             EXT4_JTR_NONE);
 269         if (err)
 270                 return err;
 271 -       ext4_insert_dentry(dir, inode, de, inline_size, fname);
 272 +       ext4_insert_dentry(dir, inode, de, inline_size, fname, NULL);
 273
 274         ext4_show_inline_dir(dir, iloc->bh, inline_start, inline_size);
 275
 276 @@ -1395,7 +1395,7 @@ int ext4_inlinedir_to_tree(struct file *dir_file,
 277                         fake.name_len = 1;
 278                         strcpy(fake.name, ".");
 279                         fake.rec_len = ext4_rec_len_to_disk(
 280 -                                         ext4_dir_rec_len(fake.name_len, NULL),
 281 +                                         EXT4_DIR_ENTRY_LEN(&fake, NULL),
 282                                           inline_size);
 283                         ext4_set_de_type(inode->i_sb, &fake, S_IFDIR);
 284                         de = &fake;
 285 @@ -1405,7 +1405,7 @@ int ext4_inlinedir_to_tree(struct file *dir_file,
 286                         fake.name_len = 2;
 287                         strcpy(fake.name, "..");
 288                         fake.rec_len = ext4_rec_len_to_disk(
 289 -                                         ext4_dir_rec_len(fake.name_len, NULL),
 290 +                                         EXT4_DIR_ENTRY_LEN(&fake, NULL),
 291                                           inline_size);
 292                         ext4_set_de_type(inode->i_sb, &fake, S_IFDIR);
 293                         de = &fake;
 294 diff --git a/fs/ext4/namei.c b/fs/ext4/namei.c
 295 index 1f95773..9edb487 100644
 296 --- a/fs/ext4/namei.c
 297 +++ b/fs/ext4/namei.c
 298 @@ -262,13 +262,14 @@ static unsigned dx_get_count(struct dx_entry *entries);
 299  static unsigned dx_get_limit(struct dx_entry *entries);
 300  static void dx_set_count(struct dx_entry *entries, unsigned value);
 301  static void dx_set_limit(struct dx_entry *entries, unsigned value);
 302 -static unsigned dx_root_limit(struct inode *dir, unsigned infosize);
 303 +static inline unsigned dx_root_limit(struct inode *dir,
 304 +               struct ext4_dir_entry_2 *dot_de, unsigned infosize);
 305  static unsigned dx_node_limit(struct inode *dir);
 306  static struct dx_frame *dx_probe(struct ext4_filename *fname,
 307                                  struct inode *dir,
 308                                  struct dx_hash_info *hinfo,
 309                                  struct dx_frame *frame);
 310 -static void dx_release(struct dx_frame *frames);
 311 +static void dx_release(struct dx_frame *frames, struct inode *dir);
 312  static int dx_make_map(struct inode *dir, struct ext4_dir_entry_2 *de,
 313                        unsigned blocksize, struct dx_hash_info *hinfo,
 314                        struct dx_map_entry map[]);
 315 @@ -408,22 +409,23 @@ static struct dx_countlimit *get_dx_countlimit(struct inode *inode,
 316  {
 317         struct ext4_dir_entry *dp;
 318         struct dx_root_info *root;
 319 -       int count_offset;
 320 +       int count_offset, dot_rec_len, dotdot_rec_len;
 321
 322         if (le16_to_cpu(dirent->rec_len) == EXT4_BLOCK_SIZE(inode->i_sb))
 323                 count_offset = 8;
 324 -       else if (le16_to_cpu(dirent->rec_len) == 12) {
 325 -               dp = (struct ext4_dir_entry *)(((void *)dirent) + 12);
 326 +       else {
 327 +               dot_rec_len = le16_to_cpu(dirent->rec_len);
 328 +               dp = (struct ext4_dir_entry *)(((void *)dirent) + dot_rec_len);
 329                 if (le16_to_cpu(dp->rec_len) !=
 330 -                   EXT4_BLOCK_SIZE(inode->i_sb) - 12)
 331 +                   EXT4_BLOCK_SIZE(inode->i_sb) - dot_rec_len)
 332                         return NULL;
 333 -               root = (struct dx_root_info *)(((void *)dp + 12));
 334 +               dotdot_rec_len = EXT4_DIR_ENTRY_LEN((struct ext4_dir_entry_2 *)dp, NULL);
 335 +               root = (struct dx_root_info *)(((void *)dp + dotdot_rec_len));
 336                 if (root->reserved_zero ||
 337                     root->info_length != sizeof(struct dx_root_info))
 338                         return NULL;
 339 -               count_offset = 32;
 340 -       } else
 341 -               return NULL;
 342 +               count_offset = 8 + dot_rec_len + dotdot_rec_len;
 343 +       }
 344
 345         if (offset)
 346                 *offset = count_offset;
 347 @@ -526,13 +528,14 @@ ext4_next_entry(struct ext4_dir_entry_2 *p, unsigned long blocksize)
 348   * Future: use high four bits of block for coalesce-on-delete flags
 349   * Mask them off for now.
 350   */
 351 -struct dx_root_info *dx_get_dx_info(struct ext4_dir_entry_2 *de)
 352 +struct dx_root_info *dx_get_dx_info(struct ext4_dir_entry_2 *de, struct inode *i_dir)
 353  {
 354 +       BUG_ON(de->name_len != 1);
 355         /* get dotdot first */
 356 -       de = (struct ext4_dir_entry_2 *)((char *)de + EXT4_DIR_REC_LEN(1));
 357 +       de = (struct ext4_dir_entry_2 *)((char *)de + EXT4_DIR_ENTRY_LEN(de, i_dir));
 358
 359         /* dx root info is after dotdot entry */
 360 -       de = (struct ext4_dir_entry_2 *)((char *)de + EXT4_DIR_REC_LEN(2));
 361 +       de = (struct ext4_dir_entry_2 *)((char *)de + EXT4_DIR_ENTRY_LEN(de, i_dir));
 362
 363         return (struct dx_root_info *)de;
 364  }
 365 @@ -577,11 +580,16 @@ static inline void dx_set_limit(struct dx_entry *entries, unsigned value)
 366         ((struct dx_countlimit *) entries)->limit = cpu_to_le16(value);
 367  }
 368
 369 -static inline unsigned dx_root_limit(struct inode *dir, unsigned infosize)
 370 +static inline unsigned dx_root_limit(struct inode *dir,
 371 +               struct ext4_dir_entry_2 *dot_de, unsigned infosize)
 372  {
 373 -       unsigned int entry_space = dir->i_sb->s_blocksize -
 374 -                       ext4_dir_rec_len(1, NULL) -
 375 -                       ext4_dir_rec_len(2, NULL) - infosize;
 376 +       struct ext4_dir_entry_2 *dotdot_de;
 377 +       unsigned entry_space;
 378 +
 379 +       BUG_ON(dot_de->name_len != 1);
 380 +       dotdot_de = ext4_next_entry(dot_de, dir->i_sb->s_blocksize);
 381 +       entry_space = dir->i_sb->s_blocksize - EXT4_DIR_ENTRY_LEN(dot_de, NULL) -
 382 +                        EXT4_DIR_ENTRY_LEN(dotdot_de, NULL) - infosize;
 383
 384         if (ext4_has_metadata_csum(dir->i_sb))
 385                 entry_space -= sizeof(struct dx_tail);
 386 @@ -699,7 +707,7 @@ static struct stats dx_show_leaf(struct inode *dir,
 387                                        (unsigned) ((char *) de - base));
 388  #endif
 389                         }
 390 -                       space += ext4_dir_rec_len(de->name_len, dir);
 391 +                       space += EXT4_DIR_ENTRY_LEN(de, dir);
 392                         names++;
 393                 }
 394                 de = ext4_next_entry(de, size);
 395 @@ -791,7 +799,7 @@ dx_probe(struct ext4_filename *fname, struct inode *dir,
 396         if (IS_ERR(frame->bh))
 397                 return (struct dx_frame *) frame->bh;
 398
 399 -       info = dx_get_dx_info((struct ext4_dir_entry_2 *)frame->bh->b_data);
 400 +       info = dx_get_dx_info((struct ext4_dir_entry_2 *)frame->bh->b_data, dir);
 401         if (info->hash_version != DX_HASH_TEA &&
 402             info->hash_version != DX_HASH_HALF_MD4 &&
 403             info->hash_version != DX_HASH_LEGACY &&
 404 @@ -847,11 +855,14 @@ dx_probe(struct ext4_filename *fname, struct inode *dir,
 405
 406         entries = (struct dx_entry *)(((char *)info) + info->info_length);
 407
 408 -       if (dx_get_limit(entries) != dx_root_limit(dir,
 409 -                                                  info->info_length)) {
 410 +       if (dx_get_limit(entries) !=
 411 +           dx_root_limit(dir, (struct ext4_dir_entry_2 *)frame->bh->b_data,
 412 +                         info->info_length)) {
 413                 ext4_warning_inode(dir, "dx entry: limit %u != root limit %u",
 414                                    dx_get_limit(entries),
 415 -                                  dx_root_limit(dir, info->info_length));
 416 +                                  dx_root_limit(dir,
 417 +                                         (struct ext4_dir_entry_2 *)frame->bh->b_data,
 418 +                                         info->info_length));
 419                 goto fail;
 420         }
 421
 422 @@ -914,7 +925,7 @@ fail:
 423         return ret_err;
 424  }
 425
 426 -static void dx_release(struct dx_frame *frames)
 427 +static void dx_release(struct dx_frame *frames, struct inode *dir)
 428  {
 429         struct dx_root_info *info;
 430         int i;
 431 @@ -923,7 +934,7 @@ static void dx_release(struct dx_frame *frames)
 432         if (frames[0].bh == NULL)
 433                 return;
 434
 435 -       info = dx_get_dx_info((struct ext4_dir_entry_2 *)frames[0].bh->b_data);
 436 +       info = dx_get_dx_info((struct ext4_dir_entry_2 *)frames[0].bh->b_data, dir);
 437         /* save local copy, "info" may be freed after brelse() */
 438         indirect_levels = info->indirect_levels;
 439         for (i = 0; i <= indirect_levels; i++) {
 440 @@ -1224,12 +1235,12 @@ int ext4_htree_fill_tree(struct file *dir_file, __u32 start_hash,
 441                     (count && ((hashval & 1) == 0)))
 442                         break;
 443         }
 444 -       dx_release(frames);
 445 +       dx_release(frames, dir);
 446         dxtrace(printk(KERN_DEBUG "Fill tree: returned %d entries, "
 447                        "next hash: %x\n", count, *next_hash));
 448         return count;
 449  errout:
 450 -       dx_release(frames);
 451 +       dx_release(frames, dir);
 452         return (err);
 453  }
 454
 455 @@ -1755,7 +1766,7 @@ static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
 456  errout:
 457         dxtrace(printk(KERN_DEBUG "%s not found\n", fname->usr_fname->name));
 458  success:
 459 -       dx_release(frames);
 460 +       dx_release(frames, dir);
 461         return bh;
 462  }
 463
 464 @@ -1879,7 +1890,7 @@ dx_move_dirents(struct inode *dir, char *from, char *to,
 465         while (count--) {
 466                 struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *)
 467                                                 (from + (map->offs<<2));
 468 -               rec_len = ext4_dir_rec_len(de->name_len, dir);
 469 +               rec_len = EXT4_DIR_ENTRY_LEN(de, dir);
 470
 471                 memcpy (to, de, rec_len);
 472                 ((struct ext4_dir_entry_2 *) to)->rec_len =
 473 @@ -1912,7 +1923,7 @@ static struct ext4_dir_entry_2 *dx_pack_dirents(struct inode *dir, char *base,
 474         while ((char*)de < base + blocksize) {
 475                 next = ext4_next_entry(de, blocksize);
 476                 if (de->inode && de->name_len) {
 477 -                       rec_len = ext4_dir_rec_len(de->name_len, dir);
 478 +                       rec_len = EXT4_DIR_ENTRY_LEN(de, dir);
 479                         if (de > to)
 480                                 memmove(to, de, rec_len);
 481                         to->rec_len = ext4_rec_len_to_disk(rec_len, blocksize);
 482 @@ -2051,14 +2062,21 @@ int ext4_find_dest_de(struct inode *dir, struct inode *inode,
 483                       struct buffer_head *bh,
 484                       void *buf, int buf_size,
 485                       struct ext4_filename *fname,
 486 -                     struct ext4_dir_entry_2 **dest_de)
 487 +                     struct ext4_dir_entry_2 **dest_de,
 488 +                     int *dlen)
 489  {
 490         struct ext4_dir_entry_2 *de;
 491 -       unsigned short reclen = ext4_dir_rec_len(fname_len(fname), dir);
 492 +       unsigned short reclen;
 493         int nlen, rlen;
 494         unsigned int offset = 0;
 495         char *top;
 496
 497 +       if (dlen) {
 498 +               reclen = ext4_dir_rec_len(fname_len(fname) + *dlen, dir);
 499 +               *dlen = 0;
 500 +       } else {
 501 +               reclen = ext4_dir_rec_len(fname_len(fname), dir);
 502 +       }
 503         de = (struct ext4_dir_entry_2 *)buf;
 504         top = buf + buf_size - reclen;
 505         while ((char *) de <= top) {
 506 @@ -2067,10 +2085,31 @@ int ext4_find_dest_de(struct inode *dir, struct inode *inode,
 507                         return -EFSCORRUPTED;
 508                 if (ext4_match(dir, fname, de))
 509                         return -EEXIST;
 510 -               nlen = ext4_dir_rec_len(de->name_len, dir);
 511 +               nlen = EXT4_DIR_ENTRY_LEN(de, dir);
 512                 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
 513                 if ((de->inode ? rlen - nlen : rlen) >= reclen)
 514                         break;
 515 +
 516 +               /* Then for dotdot entries, check for the smaller space
 517 +                * required for just the entry, no FID
 518 +                */
 519 +               if (fname_len(fname) == 2 && memcmp(fname_name(fname), "..", 2) == 0) {
 520 +                       if ((de->inode ? rlen - nlen : rlen) >=
 521 +                           ext4_dir_rec_len(fname_len(fname), dir)) {
 522 +                               /* set dlen = 1 to indicate not
 523 +                                * enough space store fid
 524 +                                */
 525 +                               if (dlen)
 526 +                                       *dlen = 1;
 527 +                               break;
 528 +                       }
 529 +                       /* The new ".." entry must be written over the
 530 +                        * previous ".." entry, which is the first
 531 +                        * entry traversed by this scan. If it doesn't
 532 +                        * fit, something is badly wrong, so -EIO.
 533 +                        */
 534 +                       return -EIO;
 535 +               }
 536                 de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
 537                 offset += rlen;
 538         }
 539 @@ -2085,12 +2124,13 @@ void ext4_insert_dentry(struct inode *dir,
 540                         struct inode *inode,
 541                         struct ext4_dir_entry_2 *de,
 542                         int buf_size,
 543 -                       struct ext4_filename *fname)
 544 +                       struct ext4_filename *fname,
 545 +                       void *data)
 546  {
 547
 548         int nlen, rlen;
 549
 550 -       nlen = ext4_dir_rec_len(de->name_len, dir);
 551 +       nlen = EXT4_DIR_ENTRY_LEN(de, dir);
 552         rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
 553         if (de->inode) {
 554                 struct ext4_dir_entry_2 *de1 =
 555 @@ -2111,6 +2151,12 @@ void ext4_insert_dentry(struct inode *dir,
 556                 EXT4_DIRENT_HASHES(de)->minor_hash =
 557                                                 cpu_to_le32(hinfo->minor_hash);
 558         }
 559 +       if (data) {
 560 +               de->name[fname_len(fname)] = 0;
 561 +               memcpy(&de->name[fname_len(fname) + 1], data, *(char *)data);
 562 +               de->file_type |= EXT4_DIRENT_LUFID;
 563 +       }
 564 +
 565  }
 566
 567  /*
 568 @@ -2128,14 +2174,19 @@ static int add_dirent_to_buf(handle_t *handle, struct ext4_filename *fname,
 569  {
 570         unsigned int    blocksize = dir->i_sb->s_blocksize;
 571         int             csum_size = 0;
 572 -       int             err, err2;
 573 +       int             err, err2, dlen = 0;
 574 +       unsigned char   *data;
 575
 576 +       data = ext4_dentry_get_data(inode->i_sb, (struct ext4_dentry_param *)
 577 +                                               EXT4_I(inode)->i_dirdata);
 578         if (ext4_has_metadata_csum(inode->i_sb))
 579                 csum_size = sizeof(struct ext4_dir_entry_tail);
 580
 581         if (!de) {
 582 +               if (data)
 583 +                       dlen = (*data) + 1;
 584                 err = ext4_find_dest_de(dir, inode, bh, bh->b_data,
 585 -                                       blocksize - csum_size, fname, &de);
 586 +                                       blocksize - csum_size, fname, &de, &dlen);
 587                 if (err)
 588                         return err;
 589         }
 590 @@ -2148,7 +2199,10 @@ static int add_dirent_to_buf(handle_t *handle, struct ext4_filename *fname,
 591         }
 592
 593         /* By now the buffer is marked for journaling */
 594 -       ext4_insert_dentry(dir, inode, de, blocksize, fname);
 595 +       /* If writing the short form of "dotdot", don't add the data section */
 596 +       if (dlen == 1)
 597 +               data = NULL;
 598 +       ext4_insert_dentry(dir, inode, de, blocksize, fname, data);
 599
 600         /*
 601          * XXX shouldn't update any times until successful
 602 @@ -2246,7 +2300,7 @@ static int make_indexed_dir(handle_t *handle, struct ext4_filename *fname,
 603                                      blocksize);
 604
 605         /* initialize hashing info */
 606 -       dx_info = dx_get_dx_info(dot_de);
 607 +       dx_info = dx_get_dx_info(dot_de, dir);
 608         memset(dx_info, 0, sizeof(*dx_info));
 609         dx_info->info_length = sizeof(*dx_info);
 610         if (ext4_hash_in_dirent(dir))
 611 @@ -2257,7 +2311,8 @@ static int make_indexed_dir(handle_t *handle, struct ext4_filename *fname,
 612         entries = (void *)dx_info + sizeof(*dx_info);
 613         dx_set_block(entries, 1);
 614         dx_set_count(entries, 1);
 615 -       dx_set_limit(entries, dx_root_limit(dir, sizeof(*dx_info)));
 616 +       dx_set_limit(entries, dx_root_limit(dir,
 617 +                                        dot_de, sizeof(*dx_info)));
 618
 619         /* Initialize as for dx_probe */
 620         fname->hinfo.hash_version = dx_info->hash_version;
 621 @@ -2298,7 +2353,7 @@ out_frames:
 622          */
 623         if (retval)
 624                 ext4_mark_inode_dirty(handle, dir);
 625 -       dx_release(frames);
 626 +       dx_release(frames, dir);
 627         brelse(bh2);
 628         return retval;
 629  }
 630 @@ -2311,6 +2366,8 @@ static int ext4_update_dotdot(handle_t *handle, struct dentry *dentry,
 631         struct buffer_head *dir_block;
 632         struct ext4_dir_entry_2 *de;
 633         int len, journal = 0, err = 0;
 634 +       int dlen = 0;
 635 +       char *data;
 636
 637         if (IS_ERR(handle))
 638                 return PTR_ERR(handle);
 639 @@ -2326,21 +2383,26 @@ static int ext4_update_dotdot(handle_t *handle, struct dentry *dentry,
 640
 641         de = (struct ext4_dir_entry_2 *)dir_block->b_data;
 642         /* the first item must be "." */
 643 -       assert(de->name_len == 1 && de->name[0] == '.');
 644 +       ASSERT(de->name_len == 1 && de->name[0] == '.');
 645         len = le16_to_cpu(de->rec_len);
 646 -       assert(len >= EXT4_DIR_REC_LEN(1));
 647 -       if (len > EXT4_DIR_REC_LEN(1)) {
 648 +       ASSERT(len >= EXT4_DIR_REC_LEN(1, dir));
 649 +       if (len > EXT4_DIR_REC_LEN(1, dir)) {
 650                 BUFFER_TRACE(dir_block, "get_write_access");
 651                 err = ext4_journal_get_write_access(handle, dir->i_sb, dir_block, EXT4_JTR_NONE);
 652                 if (err)
 653                         goto out_journal;
 654
 655                 journal = 1;
 656 -               de->rec_len = cpu_to_le16(EXT4_DIR_REC_LEN(1));
 657 +               de->rec_len = cpu_to_le16(EXT4_DIR_ENTRY_LEN(de, dir));
 658         }
 659
 660 -       len -= EXT4_DIR_REC_LEN(1);
 661 -       assert(len == 0 || len >= EXT4_DIR_REC_LEN(2));
 662 +       len -= EXT4_DIR_ENTRY_LEN(de, NULL);
 663 +       data = ext4_dentry_get_data(dir->i_sb,
 664 +                       (struct ext4_dentry_param *)dentry->d_fsdata);
 665 +       if (data)
 666 +               dlen = *data + 1;
 667 +       ASSERT(len == 0 || len >= EXT4_DIR_REC_LEN(2 + dlen, dir));
 668 +
 669         de = (struct ext4_dir_entry_2 *)
 670                         ((char *) de + le16_to_cpu(de->rec_len));
 671         if (!journal) {
 672 @@ -2354,10 +2416,15 @@ static int ext4_update_dotdot(handle_t *handle, struct dentry *dentry,
 673         if (len > 0)
 674                 de->rec_len = cpu_to_le16(len);
 675         else
 676 -               assert(le16_to_cpu(de->rec_len) >= EXT4_DIR_REC_LEN(2));
 677 +               ASSERT(le16_to_cpu(de->rec_len) >= EXT4_DIR_REC_LEN(2, dir));
 678         de->name_len = 2;
 679         strcpy(de->name, "..");
 680 -       ext4_set_de_type(dir->i_sb, de, S_IFDIR);
 681 +       if (data != NULL && ext4_get_dirent_data_len(de) >= dlen) {
 682 +               de->name[2] = 0;
 683 +               memcpy(&de->name[2 + 1], data, *data);
 684 +               ext4_set_de_type(dir->i_sb, de, S_IFDIR);
 685 +               de->file_type |= EXT4_DIRENT_LUFID;
 686 +       }
 687
 688  out_journal:
 689         if (journal) {
 690 @@ -2395,6 +2462,7 @@ static int ext4_add_entry(handle_t *handle, struct dentry *dentry,
 691         ext4_lblk_t block, blocks;
 692         int     csum_size = 0;
 693
 694 +       EXT4_I(inode)->i_dirdata = dentry->d_fsdata;
 695         if (ext4_has_metadata_csum(inode->i_sb))
 696                 csum_size = sizeof(struct ext4_dir_entry_tail);
 697
 698 @@ -2637,7 +2705,7 @@ again:
 699                         dx_set_count(entries, 1);
 700                         dx_set_block(entries + 0, newblock);
 701                         info = dx_get_dx_info((struct ext4_dir_entry_2 *)
 702 -                                             frames[0].bh->b_data);
 703 +                                             frames[0].bh->b_data, dir);
 704                         info->indirect_levels = 1;
 705                         dxtrace(printk(KERN_DEBUG
 706                                        "Creating %d level index...\n",
 707 @@ -2663,7 +2731,7 @@ journal_error:
 708         ext4_std_error(dir->i_sb, err); /* this is a no-op if err == 0 */
 709  cleanup:
 710         brelse(bh);
 711 -       dx_release(frames);
 712 +       dx_release(frames, dir);
 713         /* @restart is true means htree-path has been changed, we need to
 714          * repeat dx_probe() to find out valid htree-path
 715          */
 716 @@ -2966,38 +3034,73 @@ err_unlock_inode:
 717         return err;
 718  }
 719
 720 +struct tp_block {
 721 +       struct inode *inode;
 722 +       void *data1;
 723 +       void *data2;
 724 +};
 725 +
 726  struct ext4_dir_entry_2 *ext4_init_dot_dotdot(struct inode *inode,
 727                           struct ext4_dir_entry_2 *de,
 728                           int blocksize, int csum_size,
 729                           unsigned int parent_ino, int dotdot_real_len)
 730  {
 731 +       void *data1 = NULL, *data2 = NULL;
 732 +       int dot_reclen = 0;
 733 +
 734 +       if (dotdot_real_len == 10) {
 735 +               struct tp_block *tpb = (struct tp_block *)inode;
 736 +               data1 = tpb->data1;
 737 +               data2 = tpb->data2;
 738 +               inode = tpb->inode;
 739 +               dotdot_real_len = 0;
 740 +       }
 741         de->inode = cpu_to_le32(inode->i_ino);
 742         de->name_len = 1;
 743 -       de->rec_len = ext4_rec_len_to_disk(ext4_dir_rec_len(de->name_len, NULL),
 744 -                                          blocksize);
 745         strcpy(de->name, ".");
 746         ext4_set_de_type(inode->i_sb, de, S_IFDIR);
 747
 748 +       /* get packed fid data*/
 749 +       data1 = ext4_dentry_get_data(inode->i_sb,
 750 +                               (struct ext4_dentry_param *) data1);
 751 +       if (data1) {
 752 +               de->name[1] = 0;
 753 +               memcpy(&de->name[2], data1, *(char *) data1);
 754 +               de->file_type |= EXT4_DIRENT_LUFID;
 755 +       }
 756 +       de->rec_len = cpu_to_le16(EXT4_DIR_ENTRY_LEN(de, NULL));
 757 +
 758 +       dot_reclen = cpu_to_le16(de->rec_len);
 759         de = ext4_next_entry(de, blocksize);
 760         de->inode = cpu_to_le32(parent_ino);
 761         de->name_len = 2;
 762 +
 763 +       strcpy(de->name, "..");
 764 +       ext4_set_de_type(inode->i_sb, de, S_IFDIR);
 765 +       data2 = ext4_dentry_get_data(inode->i_sb,
 766 +                       (struct ext4_dentry_param *) data2);
 767 +       if (data2) {
 768 +               de->name[2] = 0;
 769 +               memcpy(&de->name[3], data2, *(char *) data2);
 770 +               de->file_type |= EXT4_DIRENT_LUFID;
 771 +       }
 772 +
 773         if (!dotdot_real_len)
 774                 de->rec_len = ext4_rec_len_to_disk(blocksize -
 775 -                                       (csum_size + ext4_dir_rec_len(1, NULL)),
 776 -                                       blocksize);
 777 +                                       (csum_size + dot_reclen), blocksize);
 778         else
 779                 de->rec_len = ext4_rec_len_to_disk(
 780 -                                       ext4_dir_rec_len(de->name_len, NULL),
 781 +                                       EXT4_DIR_ENTRY_LEN(de, NULL),
 782                                         blocksize);
 783 -       strcpy(de->name, "..");
 784 -       ext4_set_de_type(inode->i_sb, de, S_IFDIR);
 785
 786         return ext4_next_entry(de, blocksize);
 787  }
 788
 789  int ext4_init_new_dir(handle_t *handle, struct inode *dir,
 790 -                            struct inode *inode)
 791 +                            struct inode *inode,
 792 +                            const void *data1, const void *data2)
 793  {
 794 +       struct tp_block param;
 795         struct buffer_head *dir_block = NULL;
 796         struct ext4_dir_entry_2 *de;
 797         ext4_lblk_t block = 0;
 798 @@ -3021,7 +3124,11 @@ int ext4_init_new_dir(handle_t *handle, struct inode *dir,
 799         if (IS_ERR(dir_block))
 800                 return PTR_ERR(dir_block);
 801         de = (struct ext4_dir_entry_2 *)dir_block->b_data;
 802 -       ext4_init_dot_dotdot(inode, de, blocksize, csum_size, dir->i_ino, 0);
 803 +       param.inode = inode;
 804 +       param.data1 = (void *)data1;
 805 +       param.data2 = (void *)data2;
 806 +       ext4_init_dot_dotdot((struct inode *)(&param), de, blocksize,
 807 +                            csum_size, dir->i_ino, 10);
 808         set_nlink(inode, 2);
 809         if (csum_size)
 810                 ext4_initialize_dirent_tail(dir_block, blocksize);
 811 @@ -3036,6 +3143,29 @@ out:
 812         return err;
 813  }
 814
 815 +/* Initialize @inode as a subdirectory of @dir, and add the
 816 + * "." and ".." entries into the first directory block. */
 817 +int ext4_add_dot_dotdot(handle_t *handle, struct inode *dir,
 818 +                       struct inode *inode,
 819 +                       const void *data1, const void *data2)
 820 +{
 821 +       int rc;
 822 +
 823 +       if (IS_ERR(handle))
 824 +               return PTR_ERR(handle);
 825 +
 826 +       if (IS_DIRSYNC(dir))
 827 +               ext4_handle_sync(handle);
 828 +
 829 +       inode->i_op = &ext4_dir_inode_operations;
 830 +       inode->i_fop = &ext4_dir_operations;
 831 +       rc = ext4_init_new_dir(handle, dir, inode, data1, data2);
 832 +       if (!rc)
 833 +               rc = ext4_mark_inode_dirty(handle, inode);
 834 +       return rc;
 835 +}
 836 +EXPORT_SYMBOL(ext4_add_dot_dotdot);
 837 +
 838  static int ext4_mkdir(struct user_namespace *mnt_userns, struct inode *dir,
 839                       struct dentry *dentry, umode_t mode)
 840  {
 841 @@ -3063,7 +3193,7 @@ retry:
 842
 843         inode->i_op = &ext4_dir_inode_operations;
 844         inode->i_fop = &ext4_dir_operations;
 845 -       err = ext4_init_new_dir(handle, dir, inode);
 846 +       err = ext4_init_new_dir(handle, dir, inode, NULL, NULL);
 847         if (err)
 848                 goto out_clear_inode;
 849         err = ext4_mark_inode_dirty(handle, inode);
 850 diff --git a/fs/ext4/super.c b/fs/ext4/super.c
 851 index 8a95bd1..488162e 100644
 852 --- a/fs/ext4/super.c
 853 +++ b/fs/ext4/super.c
 854 @@ -1692,7 +1692,7 @@ enum {
 855         Opt_data_err_abort, Opt_data_err_ignore, Opt_test_dummy_encryption,
 856         Opt_inlinecrypt,
 857         Opt_usrjquota, Opt_grpjquota, Opt_quota,
 858 -       Opt_noquota, Opt_barrier, Opt_nobarrier, Opt_err,
 859 +       Opt_noquota, Opt_barrier, Opt_nobarrier, Opt_err, Opt_dirdata,
 860         Opt_usrquota, Opt_grpquota, Opt_prjquota, Opt_i_version,
 861         Opt_dax, Opt_dax_always, Opt_dax_inode, Opt_dax_never,
 862         Opt_stripe, Opt_delalloc, Opt_nodelalloc, Opt_warn_on_error,
 863 @@ -1808,6 +1808,7 @@ static const struct fs_parameter_spec ext4_param_specs[] = {
 864         fsparam_u32     ("stripe",              Opt_stripe),
 865         fsparam_flag    ("delalloc",            Opt_delalloc),
 866         fsparam_flag    ("nodelalloc",          Opt_nodelalloc),
 867 +       fsparam_flag    ("dirdata",             Opt_dirdata),
 868         fsparam_flag    ("warn_on_error",       Opt_warn_on_error),
 869         fsparam_flag    ("nowarn_on_error",     Opt_nowarn_on_error),
 870         fsparam_u32     ("debug_want_extra_isize",
 871 @@ -1944,6 +1945,7 @@ static const struct mount_opts {
 872                                                         MOPT_CLEAR | MOPT_Q},
 873         {Opt_usrjquota, 0, MOPT_Q},
 874         {Opt_grpjquota, 0, MOPT_Q},
 875 +       {Opt_dirdata, EXT4_MOUNT_DIRDATA, MOPT_SET},
 876         {Opt_jqfmt, 0, MOPT_QFMT},
 877         {Opt_nombcache, EXT4_MOUNT_NO_MBCACHE, MOPT_SET},
 878         {Opt_no_prefetch_block_bitmaps, EXT4_MOUNT_NO_PREFETCH_BLOCK_BITMAPS,
 879 --
 880 2.34.1
 881