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

   1 Subject: [PATCH] This patch implements feature which allows ext4 fs users
   2  (e.g. Lustre) to store data in ext4 dirent.
   3
   4 Data is stored in ext4 dirent after file-name, this space is accounted
   5 in de->rec_len. Flag EXT4_DIRENT_LUFID is added to d_type when extra
   6 data is present.
   7
   8 This uses dentry->d_fsdata to pass fid to ext4. so no
   9 changes in ext4_add_entry() interface required.
  10 ---
  11  fs/ext4/dir.c         |   9 +-
  12  fs/ext4/ext4.h        | 106 ++++++++++++++++--
  13  fs/ext4/fast_commit.c |   2 +-
  14  fs/ext4/inline.c      |   8 +-
  15  fs/ext4/namei.c       | 249 ++++++++++++++++++++++++++++++++----------
  16  fs/ext4/super.c       |   4 +-
  17  6 files changed, 303 insertions(+), 75 deletions(-)
  18
  19 diff --git a/fs/ext4/dir.c b/fs/ext4/dir.c
  20 index 74b172a..c6afabc 100644
  21 --- a/fs/ext4/dir.c
  22 +++ b/fs/ext4/dir.c
  23 @@ -466,12 +466,17 @@ 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 @@ -480,7 +485,7 @@ 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 0791a8b..f1bc21d 100644
  53 --- a/fs/ext4/ext4.h
  54 +++ b/fs/ext4/ext4.h
  55 @@ -1165,6 +1165,7 @@ struct ext4_inode_info {
  56         __u32 i_csum_seed;
  57
  58         kprojid_t i_projid;
  59 +       void *i_dirdata;
  60  };
  61
  62  /*
  63 @@ -1186,6 +1187,7 @@ 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 @@ -2117,6 +2119,7 @@ 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 @@ -2326,6 +2329,42 @@ 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 @@ -2337,6 +2376,17 @@ 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 @@ -2344,10 +2394,10 @@ 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 @@ -2821,11 +2871,13 @@ 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 @@ -2841,10 +2893,17 @@ 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 @@ -3048,7 +3107,8 @@ 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_orphan_add(handle_t *, struct inode *);
 199 @@ -3059,6 +3119,8 @@ extern struct inode *ext4_create_inode(handle_t *handle,
 200  extern int ext4_delete_entry(handle_t *handle, struct inode * dir,
 201                              struct ext4_dir_entry_2 *de_del,
 202                              struct buffer_head *bh);
 203 +extern int ext4_add_dot_dotdot(handle_t *handle, struct inode *dir,
 204 +                              struct inode *inode, const void *, const void *);
 205  extern int ext4_htree_fill_tree(struct file *dir_file, __u32 start_hash,
 206                                 __u32 start_minor_hash, __u32 *next_hash);
 207  extern int ext4_search_dir(struct buffer_head *bh,
 208 @@ -3862,6 +3924,36 @@ static inline int ext4_buffer_uptodate(struct buffer_head *bh)
 209         return buffer_uptodate(bh);
 210  }
 211
 212 +/*
 213 + * Compute the total directory entry data length.
 214 + * This includes the filename and an implicit NUL terminator (always present),
 215 + * and optional extensions.  Each extension has a bit set in the high 4 bits of
 216 + * de->file_type, and the extension length is the first byte in each entry.
 217 + */
 218 +static inline int ext4_get_dirent_data_len(struct ext4_dir_entry_2 *de)
 219 +{
 220 +       char *len = de->name + de->name_len + 1 /* NUL terminator */;
 221 +       int dlen = 0;
 222 +       __u8 extra_data_flags = (de->file_type & ~EXT4_FT_MASK) >> 4;
 223 +       struct ext4_dir_entry_tail *t = (struct ext4_dir_entry_tail *)de;
 224 +
 225 +       if (!t->det_reserved_zero1 &&
 226 +           le16_to_cpu(t->det_rec_len) ==
 227 +               sizeof(struct ext4_dir_entry_tail) &&
 228 +           !t->det_reserved_zero2 &&
 229 +           t->det_reserved_ft == EXT4_FT_DIR_CSUM)
 230 +               return 0;
 231 +
 232 +       while (extra_data_flags) {
 233 +               if (extra_data_flags & 1) {
 234 +                       dlen += *len + (dlen == 0);
 235 +                       len += *len;
 236 +               }
 237 +               extra_data_flags >>= 1;
 238 +       }
 239 +       return dlen;
 240 +}
 241 +
 242  #endif /* __KERNEL__ */
 243
 244  #define EFSBADCRC      EBADMSG         /* Bad CRC detected */
 245 diff --git a/fs/ext4/fast_commit.c b/fs/ext4/fast_commit.c
 246 index 276d9e6..3be0f08 100644
 247 --- a/fs/ext4/fast_commit.c
 248 +++ b/fs/ext4/fast_commit.c
 249 @@ -1596,7 +1596,7 @@ static int ext4_fc_replay_create(struct super_block *sb, struct ext4_fc_tl *tl,
 250                         jbd_debug(1, "Dir %d not found.", darg.ino);
 251                         goto out;
 252                 }
 253 -               ret = ext4_init_new_dir(NULL, dir, inode);
 254 +               ret = ext4_init_new_dir(NULL, dir, inode, NULL, NULL);
 255                 iput(dir);
 256                 if (ret) {
 257                         ret = 0;
 258 diff --git a/fs/ext4/inline.c b/fs/ext4/inline.c
 259 index 9626c31..ed31b5c 100644
 260 --- a/fs/ext4/inline.c
 261 +++ b/fs/ext4/inline.c
 262 @@ -1029,7 +1029,7 @@ static int ext4_add_dirent_to_inline(handle_t *handle,
 263         struct ext4_dir_entry_2 *de;
 264
 265         err = ext4_find_dest_de(dir, inode, iloc->bh, inline_start,
 266 -                               inline_size, fname, &de);
 267 +                               inline_size, fname, &de, NULL);
 268         if (err)
 269                 return err;
 270
 271 @@ -1038,7 +1038,7 @@ static int ext4_add_dirent_to_inline(handle_t *handle,
 272                                             EXT4_JTR_NONE);
 273         if (err)
 274                 return err;
 275 -       ext4_insert_dentry(dir, inode, de, inline_size, fname);
 276 +       ext4_insert_dentry(dir, inode, de, inline_size, fname, NULL);
 277
 278         ext4_show_inline_dir(dir, iloc->bh, inline_start, inline_size);
 279
 280 @@ -1396,7 +1396,7 @@ int ext4_inlinedir_to_tree(struct file *dir_file,
 281                         fake.name_len = 1;
 282                         strcpy(fake.name, ".");
 283                         fake.rec_len = ext4_rec_len_to_disk(
 284 -                                         ext4_dir_rec_len(fake.name_len, NULL),
 285 +                                         EXT4_DIR_ENTRY_LEN(&fake, NULL),
 286                                           inline_size);
 287                         ext4_set_de_type(inode->i_sb, &fake, S_IFDIR);
 288                         de = &fake;
 289 @@ -1406,7 +1406,7 @@ int ext4_inlinedir_to_tree(struct file *dir_file,
 290                         fake.name_len = 2;
 291                         strcpy(fake.name, "..");
 292                         fake.rec_len = ext4_rec_len_to_disk(
 293 -                                         ext4_dir_rec_len(fake.name_len, NULL),
 294 +                                         EXT4_DIR_ENTRY_LEN(&fake, NULL),
 295                                           inline_size);
 296                         ext4_set_de_type(inode->i_sb, &fake, S_IFDIR);
 297                         de = &fake;
 298 diff --git a/fs/ext4/namei.c b/fs/ext4/namei.c
 299 index 7f00dc3..51c950b 100644
 300 --- a/fs/ext4/namei.c
 301 +++ b/fs/ext4/namei.c
 302 @@ -285,13 +285,14 @@ static unsigned dx_get_count(struct dx_entry *entries);
 303  static unsigned dx_get_limit(struct dx_entry *entries);
 304  static void dx_set_count(struct dx_entry *entries, unsigned value);
 305  static void dx_set_limit(struct dx_entry *entries, unsigned value);
 306 -static unsigned dx_root_limit(struct inode *dir, unsigned infosize);
 307 +static inline unsigned dx_root_limit(struct inode *dir,
 308 +               struct ext4_dir_entry_2 *dot_de, unsigned infosize);
 309  static unsigned dx_node_limit(struct inode *dir);
 310  static struct dx_frame *dx_probe(struct ext4_filename *fname,
 311                                  struct inode *dir,
 312                                  struct dx_hash_info *hinfo,
 313                                  struct dx_frame *frame);
 314 -static void dx_release(struct dx_frame *frames);
 315 +static void dx_release(struct dx_frame *frames, struct inode *dir);
 316  static int dx_make_map(struct inode *dir, struct buffer_head *bh,
 317                        struct dx_hash_info *hinfo,
 318                        struct dx_map_entry *map_tail);
 319 @@ -431,22 +432,23 @@ static struct dx_countlimit *get_dx_countlimit(struct inode *inode,
 320  {
 321         struct ext4_dir_entry *dp;
 322         struct dx_root_info *root;
 323 -       int count_offset;
 324 +       int count_offset, dot_rec_len, dotdot_rec_len;
 325
 326         if (le16_to_cpu(dirent->rec_len) == EXT4_BLOCK_SIZE(inode->i_sb))
 327                 count_offset = 8;
 328 -       else if (le16_to_cpu(dirent->rec_len) == 12) {
 329 -               dp = (struct ext4_dir_entry *)(((void *)dirent) + 12);
 330 +       else {
 331 +               dot_rec_len = le16_to_cpu(dirent->rec_len);
 332 +               dp = (struct ext4_dir_entry *)(((void *)dirent) + dot_rec_len);
 333                 if (le16_to_cpu(dp->rec_len) !=
 334 -                   EXT4_BLOCK_SIZE(inode->i_sb) - 12)
 335 +                   EXT4_BLOCK_SIZE(inode->i_sb) - dot_rec_len)
 336                         return NULL;
 337 -               root = (struct dx_root_info *)(((void *)dp + 12));
 338 +               dotdot_rec_len = EXT4_DIR_ENTRY_LEN((struct ext4_dir_entry_2 *)dp, NULL);
 339 +               root = (struct dx_root_info *)(((void *)dp + dotdot_rec_len));
 340                 if (root->reserved_zero ||
 341                     root->info_length != sizeof(struct dx_root_info))
 342                         return NULL;
 343 -               count_offset = 32;
 344 -       } else
 345 -               return NULL;
 346 +               count_offset = 8 + dot_rec_len + dotdot_rec_len;
 347 +       }
 348
 349         if (offset)
 350                 *offset = count_offset;
 351 @@ -549,13 +551,14 @@ ext4_next_entry(struct ext4_dir_entry_2 *p, unsigned long blocksize)
 352   * Future: use high four bits of block for coalesce-on-delete flags
 353   * Mask them off for now.
 354   */
 355 -struct dx_root_info *dx_get_dx_info(struct ext4_dir_entry_2 *de)
 356 +struct dx_root_info *dx_get_dx_info(struct ext4_dir_entry_2 *de, struct inode *i_dir)
 357  {
 358 +       BUG_ON(de->name_len != 1);
 359         /* get dotdot first */
 360 -       de = (struct ext4_dir_entry_2 *)((char *)de + EXT4_DIR_REC_LEN(1));
 361 +       de = (struct ext4_dir_entry_2 *)((char *)de + EXT4_DIR_ENTRY_LEN(de, i_dir));
 362
 363         /* dx root info is after dotdot entry */
 364 -       de = (struct ext4_dir_entry_2 *)((char *)de + EXT4_DIR_REC_LEN(2));
 365 +       de = (struct ext4_dir_entry_2 *)((char *)de + EXT4_DIR_ENTRY_LEN(de, i_dir));
 366
 367         return (struct dx_root_info *)de;
 368  }
 369 @@ -600,11 +603,16 @@ static inline void dx_set_limit(struct dx_entry *entries, unsigned value)
 370         ((struct dx_countlimit *) entries)->limit = cpu_to_le16(value);
 371  }
 372
 373 -static inline unsigned dx_root_limit(struct inode *dir, unsigned infosize)
 374 +static inline unsigned dx_root_limit(struct inode *dir,
 375 +               struct ext4_dir_entry_2 *dot_de, unsigned infosize)
 376  {
 377 -       unsigned int entry_space = dir->i_sb->s_blocksize -
 378 -                       ext4_dir_rec_len(1, NULL) -
 379 -                       ext4_dir_rec_len(2, NULL) - infosize;
 380 +       struct ext4_dir_entry_2 *dotdot_de;
 381 +       unsigned entry_space;
 382 +
 383 +       BUG_ON(dot_de->name_len != 1);
 384 +       dotdot_de = ext4_next_entry(dot_de, dir->i_sb->s_blocksize);
 385 +       entry_space = dir->i_sb->s_blocksize - EXT4_DIR_ENTRY_LEN(dot_de, NULL) -
 386 +                        EXT4_DIR_ENTRY_LEN(dotdot_de, NULL) - infosize;
 387
 388         if (ext4_has_metadata_csum(dir->i_sb))
 389                 entry_space -= sizeof(struct dx_tail);
 390 @@ -722,7 +730,7 @@ static struct stats dx_show_leaf(struct inode *dir,
 391                                        (unsigned) ((char *) de - base));
 392  #endif
 393                         }
 394 -                       space += ext4_dir_rec_len(de->name_len, dir);
 395 +                       space += EXT4_DIR_ENTRY_LEN(de, dir);
 396                         names++;
 397                 }
 398                 de = ext4_next_entry(de, size);
 399 @@ -816,7 +824,7 @@ dx_probe(struct ext4_filename *fname, struct inode *dir,
 400         if (IS_ERR(frame->bh))
 401                 return (struct dx_frame *) frame->bh;
 402
 403 -       info = dx_get_dx_info((struct ext4_dir_entry_2 *)frame->bh->b_data);
 404 +       info = dx_get_dx_info((struct ext4_dir_entry_2 *)frame->bh->b_data, dir);
 405         if (info->hash_version != DX_HASH_TEA &&
 406             info->hash_version != DX_HASH_HALF_MD4 &&
 407             info->hash_version != DX_HASH_LEGACY &&
 408 @@ -872,11 +880,14 @@ dx_probe(struct ext4_filename *fname, struct inode *dir,
 409
 410         entries = (struct dx_entry *)(((char *)info) + info->info_length);
 411
 412 -       if (dx_get_limit(entries) != dx_root_limit(dir,
 413 -                                                  info->info_length)) {
 414 +       if (dx_get_limit(entries) !=
 415 +           dx_root_limit(dir, (struct ext4_dir_entry_2 *)frame->bh->b_data,
 416 +                         info->info_length)) {
 417                 ext4_warning_inode(dir, "dx entry: limit %u != root limit %u",
 418                                    dx_get_limit(entries),
 419 -                                  dx_root_limit(dir, info->info_length));
 420 +                                  dx_root_limit(dir,
 421 +                                         (struct ext4_dir_entry_2 *)frame->bh->b_data,
 422 +                                         info->info_length));
 423                 goto fail;
 424         }
 425
 426 @@ -953,7 +964,7 @@ fail:
 427         return ret_err;
 428  }
 429
 430 -static void dx_release(struct dx_frame *frames)
 431 +static void dx_release(struct dx_frame *frames, struct inode *dir)
 432  {
 433         struct dx_root_info *info;
 434         int i;
 435 @@ -962,7 +973,7 @@ static void dx_release(struct dx_frame *frames)
 436         if (frames[0].bh == NULL)
 437                 return;
 438
 439 -       info = dx_get_dx_info((struct ext4_dir_entry_2 *)frames[0].bh->b_data);
 440 +       info = dx_get_dx_info((struct ext4_dir_entry_2 *)frames[0].bh->b_data, dir);
 441         /* save local copy, "info" may be freed after brelse() */
 442         indirect_levels = info->indirect_levels;
 443         for (i = 0; i <= indirect_levels; i++) {
 444 @@ -1263,12 +1274,12 @@ int ext4_htree_fill_tree(struct file *dir_file, __u32 start_hash,
 445                     (count && ((hashval & 1) == 0)))
 446                         break;
 447         }
 448 -       dx_release(frames);
 449 +       dx_release(frames, dir);
 450         dxtrace(printk(KERN_DEBUG "Fill tree: returned %d entries, "
 451                        "next hash: %x\n", count, *next_hash));
 452         return count;
 453  errout:
 454 -       dx_release(frames);
 455 +       dx_release(frames, dir);
 456         return (err);
 457  }
 458
 459 @@ -1801,7 +1812,7 @@ static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
 460  errout:
 461         dxtrace(printk(KERN_DEBUG "%s not found\n", fname->usr_fname->name));
 462  success:
 463 -       dx_release(frames);
 464 +       dx_release(frames, dir);
 465         return bh;
 466  }
 467
 468 @@ -1925,7 +1936,7 @@ dx_move_dirents(struct inode *dir, char *from, char *to,
 469         while (count--) {
 470                 struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *)
 471                                                 (from + (map->offs<<2));
 472 -               rec_len = ext4_dir_rec_len(de->name_len, dir);
 473 +               rec_len = EXT4_DIR_ENTRY_LEN(de, dir);
 474
 475                 memcpy (to, de, rec_len);
 476                 ((struct ext4_dir_entry_2 *) to)->rec_len =
 477 @@ -1958,7 +1969,7 @@ static struct ext4_dir_entry_2 *dx_pack_dirents(struct inode *dir, char *base,
 478         while ((char*)de < base + blocksize) {
 479                 next = ext4_next_entry(de, blocksize);
 480                 if (de->inode && de->name_len) {
 481 -                       rec_len = ext4_dir_rec_len(de->name_len, dir);
 482 +                       rec_len = EXT4_DIR_ENTRY_LEN(de, dir);
 483                         if (de > to)
 484                                 memmove(to, de, rec_len);
 485                         to->rec_len = ext4_rec_len_to_disk(rec_len, blocksize);
 486 @@ -2101,14 +2112,21 @@ int ext4_find_dest_de(struct inode *dir, struct inode *inode,
 487                       struct buffer_head *bh,
 488                       void *buf, int buf_size,
 489                       struct ext4_filename *fname,
 490 -                     struct ext4_dir_entry_2 **dest_de)
 491 +                     struct ext4_dir_entry_2 **dest_de,
 492 +                     int *dlen)
 493  {
 494         struct ext4_dir_entry_2 *de;
 495 -       unsigned short reclen = ext4_dir_rec_len(fname_len(fname), dir);
 496 +       unsigned short reclen;
 497         int nlen, rlen;
 498         unsigned int offset = 0;
 499         char *top;
 500
 501 +       if (dlen) {
 502 +               reclen = ext4_dir_rec_len(fname_len(fname) + *dlen, dir);
 503 +               *dlen = 0;
 504 +       } else {
 505 +               reclen = ext4_dir_rec_len(fname_len(fname), dir);
 506 +       }
 507         de = (struct ext4_dir_entry_2 *)buf;
 508         top = buf + buf_size - reclen;
 509         while ((char *) de <= top) {
 510 @@ -2117,10 +2135,31 @@ int ext4_find_dest_de(struct inode *dir, struct inode *inode,
 511                         return -EFSCORRUPTED;
 512                 if (ext4_match(dir, fname, de))
 513                         return -EEXIST;
 514 -               nlen = ext4_dir_rec_len(de->name_len, dir);
 515 +               nlen = EXT4_DIR_ENTRY_LEN(de, dir);
 516                 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
 517                 if ((de->inode ? rlen - nlen : rlen) >= reclen)
 518                         break;
 519 +
 520 +               /* Then for dotdot entries, check for the smaller space
 521 +                * required for just the entry, no FID
 522 +                */
 523 +               if (fname_len(fname) == 2 && memcmp(fname_name(fname), "..", 2) == 0) {
 524 +                       if ((de->inode ? rlen - nlen : rlen) >=
 525 +                           ext4_dir_rec_len(fname_len(fname), dir)) {
 526 +                               /* set dlen = 1 to indicate not
 527 +                                * enough space store fid
 528 +                                */
 529 +                               if (dlen)
 530 +                                       *dlen = 1;
 531 +                               break;
 532 +                       }
 533 +                       /* The new ".." entry must be written over the
 534 +                        * previous ".." entry, which is the first
 535 +                        * entry traversed by this scan. If it doesn't
 536 +                        * fit, something is badly wrong, so -EIO.
 537 +                        */
 538 +                       return -EIO;
 539 +               }
 540                 de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
 541                 offset += rlen;
 542         }
 543 @@ -2135,12 +2174,13 @@ void ext4_insert_dentry(struct inode *dir,
 544                         struct inode *inode,
 545                         struct ext4_dir_entry_2 *de,
 546                         int buf_size,
 547 -                       struct ext4_filename *fname)
 548 +                       struct ext4_filename *fname,
 549 +                       void *data)
 550  {
 551
 552         int nlen, rlen;
 553
 554 -       nlen = ext4_dir_rec_len(de->name_len, dir);
 555 +       nlen = EXT4_DIR_ENTRY_LEN(de, dir);
 556         rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
 557         if (de->inode) {
 558                 struct ext4_dir_entry_2 *de1 =
 559 @@ -2161,6 +2201,12 @@ void ext4_insert_dentry(struct inode *dir,
 560                 EXT4_DIRENT_HASHES(de)->minor_hash =
 561                                                 cpu_to_le32(hinfo->minor_hash);
 562         }
 563 +       if (data) {
 564 +               de->name[fname_len(fname)] = 0;
 565 +               memcpy(&de->name[fname_len(fname) + 1], data, *(char *)data);
 566 +               de->file_type |= EXT4_DIRENT_LUFID;
 567 +       }
 568 +
 569  }
 570
 571  /*
 572 @@ -2178,14 +2224,19 @@ static int add_dirent_to_buf(handle_t *handle, struct ext4_filename *fname,
 573  {
 574         unsigned int    blocksize = dir->i_sb->s_blocksize;
 575         int             csum_size = 0;
 576 -       int             err, err2;
 577 +       int             err, err2, dlen = 0;
 578 +       unsigned char   *data;
 579
 580 +       data = ext4_dentry_get_data(inode->i_sb, (struct ext4_dentry_param *)
 581 +                                               EXT4_I(inode)->i_dirdata);
 582         if (ext4_has_metadata_csum(inode->i_sb))
 583                 csum_size = sizeof(struct ext4_dir_entry_tail);
 584
 585         if (!de) {
 586 +               if (data)
 587 +                       dlen = (*data) + 1;
 588                 err = ext4_find_dest_de(dir, inode, bh, bh->b_data,
 589 -                                       blocksize - csum_size, fname, &de);
 590 +                                       blocksize - csum_size, fname, &de, &dlen);
 591                 if (err)
 592                         return err;
 593         }
 594 @@ -2198,7 +2249,10 @@ static int add_dirent_to_buf(handle_t *handle, struct ext4_filename *fname,
 595         }
 596
 597         /* By now the buffer is marked for journaling */
 598 -       ext4_insert_dentry(dir, inode, de, blocksize, fname);
 599 +       /* If writing the short form of "dotdot", don't add the data section */
 600 +       if (dlen == 1)
 601 +               data = NULL;
 602 +       ext4_insert_dentry(dir, inode, de, blocksize, fname, data);
 603
 604         /*
 605          * XXX shouldn't update any times until successful
 606 @@ -2296,7 +2350,7 @@ static int make_indexed_dir(handle_t *handle, struct ext4_filename *fname,
 607                                      blocksize);
 608
 609         /* initialize hashing info */
 610 -       dx_info = dx_get_dx_info(dot_de);
 611 +       dx_info = dx_get_dx_info(dot_de, dir);
 612         memset(dx_info, 0, sizeof(*dx_info));
 613         dx_info->info_length = sizeof(*dx_info);
 614         if (ext4_hash_in_dirent(dir))
 615 @@ -2307,7 +2361,8 @@ static int make_indexed_dir(handle_t *handle, struct ext4_filename *fname,
 616         entries = (void *)dx_info + sizeof(*dx_info);
 617         dx_set_block(entries, 1);
 618         dx_set_count(entries, 1);
 619 -       dx_set_limit(entries, dx_root_limit(dir, sizeof(*dx_info)));
 620 +       dx_set_limit(entries, dx_root_limit(dir,
 621 +                                        dot_de, sizeof(*dx_info)));
 622
 623         /* Initialize as for dx_probe */
 624         fname->hinfo.hash_version = dx_info->hash_version;
 625 @@ -2348,7 +2403,7 @@ out_frames:
 626          */
 627         if (retval)
 628                 ext4_mark_inode_dirty(handle, dir);
 629 -       dx_release(frames);
 630 +       dx_release(frames, dir);
 631         brelse(bh2);
 632         return retval;
 633  }
 634 @@ -2361,6 +2416,8 @@ static int ext4_update_dotdot(handle_t *handle, struct dentry *dentry,
 635         struct buffer_head *dir_block;
 636         struct ext4_dir_entry_2 *de;
 637         int len, journal = 0, err = 0;
 638 +       int dlen = 0;
 639 +       char *data;
 640
 641         if (IS_ERR(handle))
 642                 return PTR_ERR(handle);
 643 @@ -2376,21 +2433,25 @@ static int ext4_update_dotdot(handle_t *handle, struct dentry *dentry,
 644
 645         de = (struct ext4_dir_entry_2 *)dir_block->b_data;
 646         /* the first item must be "." */
 647 -       assert(de->name_len == 1 && de->name[0] == '.');
 648 +       ASSERT(de->name_len == 1 && de->name[0] == '.');
 649         len = le16_to_cpu(de->rec_len);
 650 -       assert(len >= EXT4_DIR_REC_LEN(1));
 651 -       if (len > EXT4_DIR_REC_LEN(1)) {
 652 +       ASSERT(len >= EXT4_DIR_REC_LEN(1, dir));
 653 +       if (len > EXT4_DIR_REC_LEN(1, dir)) {
 654                 BUFFER_TRACE(dir_block, "get_write_access");
 655                 err = ext4_journal_get_write_access(handle, dir->i_sb, dir_block, EXT4_JTR_NONE);
 656                 if (err)
 657                         goto out_journal;
 658
 659                 journal = 1;
 660 -               de->rec_len = cpu_to_le16(EXT4_DIR_REC_LEN(1));
 661 +               de->rec_len = cpu_to_le16(EXT4_DIR_REC_LEN(1, dir));
 662         }
 663
 664 -       len -= EXT4_DIR_REC_LEN(1);
 665 -       assert(len == 0 || len >= EXT4_DIR_REC_LEN(2));
 666 +       len -= EXT4_DIR_REC_LEN(1, dir);
 667 +       data = ext4_dentry_get_data(dir->i_sb,
 668 +                       (struct ext4_dentry_param *)dentry->d_fsdata);
 669 +       if (data)
 670 +               dlen = *data + 1;
 671 +       ASSERT(len == 0 || len >= EXT4_DIR_REC_LEN(2 + dlen, dir));
 672         de = (struct ext4_dir_entry_2 *)
 673                         ((char *) de + le16_to_cpu(de->rec_len));
 674         if (!journal) {
 675 @@ -2404,10 +2465,15 @@ static int ext4_update_dotdot(handle_t *handle, struct dentry *dentry,
 676         if (len > 0)
 677                 de->rec_len = cpu_to_le16(len);
 678         else
 679 -               assert(le16_to_cpu(de->rec_len) >= EXT4_DIR_REC_LEN(2));
 680 +               ASSERT(le16_to_cpu(de->rec_len) >= EXT4_DIR_REC_LEN(2, dir));
 681         de->name_len = 2;
 682         strcpy(de->name, "..");
 683 -       ext4_set_de_type(dir->i_sb, de, S_IFDIR);
 684 +       if (data != NULL && ext4_get_dirent_data_len(de) >= dlen) {
 685 +               de->name[2] = 0;
 686 +               memcpy(&de->name[2 + 1], data, *data);
 687 +               ext4_set_de_type(dir->i_sb, de, S_IFDIR);
 688 +               de->file_type |= EXT4_DIRENT_LUFID;
 689 +       }
 690
 691  out_journal:
 692         if (journal) {
 693 @@ -2445,6 +2511,7 @@ static int ext4_add_entry(handle_t *handle, struct dentry *dentry,
 694         ext4_lblk_t block, blocks;
 695         int     csum_size = 0;
 696
 697 +       EXT4_I(inode)->i_dirdata = dentry->d_fsdata;
 698         if (ext4_has_metadata_csum(inode->i_sb))
 699                 csum_size = sizeof(struct ext4_dir_entry_tail);
 700
 701 @@ -2687,7 +2754,7 @@ again:
 702                         dx_set_count(entries, 1);
 703                         dx_set_block(entries + 0, newblock);
 704                         info = dx_get_dx_info((struct ext4_dir_entry_2 *)
 705 -                                             frames[0].bh->b_data);
 706 +                                             frames[0].bh->b_data, dir);
 707                         info->indirect_levels = 1;
 708                         dxtrace(printk(KERN_DEBUG
 709                                        "Creating %d level index...\n",
 710 @@ -2713,7 +2780,7 @@ journal_error:
 711         ext4_std_error(dir->i_sb, err); /* this is a no-op if err == 0 */
 712  cleanup:
 713         brelse(bh);
 714 -       dx_release(frames);
 715 +       dx_release(frames, dir);
 716         /* @restart is true means htree-path has been changed, we need to
 717          * repeat dx_probe() to find out valid htree-path
 718          */
 719 @@ -3016,38 +3083,73 @@ err_unlock_inode:
 720         return err;
 721  }
 722
 723 +struct tp_block {
 724 +       struct inode *inode;
 725 +       void *data1;
 726 +       void *data2;
 727 +};
 728 +
 729  struct ext4_dir_entry_2 *ext4_init_dot_dotdot(struct inode *inode,
 730                           struct ext4_dir_entry_2 *de,
 731                           int blocksize, int csum_size,
 732                           unsigned int parent_ino, int dotdot_real_len)
 733  {
 734 +       void *data1 = NULL, *data2 = NULL;
 735 +       int dot_reclen = 0;
 736 +
 737 +       if (dotdot_real_len == 10) {
 738 +               struct tp_block *tpb = (struct tp_block *)inode;
 739 +               data1 = tpb->data1;
 740 +               data2 = tpb->data2;
 741 +               inode = tpb->inode;
 742 +               dotdot_real_len = 0;
 743 +       }
 744         de->inode = cpu_to_le32(inode->i_ino);
 745         de->name_len = 1;
 746 -       de->rec_len = ext4_rec_len_to_disk(ext4_dir_rec_len(de->name_len, NULL),
 747 -                                          blocksize);
 748         strcpy(de->name, ".");
 749         ext4_set_de_type(inode->i_sb, de, S_IFDIR);
 750
 751 +       /* get packed fid data*/
 752 +       data1 = ext4_dentry_get_data(inode->i_sb,
 753 +                               (struct ext4_dentry_param *) data1);
 754 +       if (data1) {
 755 +               de->name[1] = 0;
 756 +               memcpy(&de->name[2], data1, *(char *) data1);
 757 +               de->file_type |= EXT4_DIRENT_LUFID;
 758 +       }
 759 +       de->rec_len = cpu_to_le16(EXT4_DIR_ENTRY_LEN(de, NULL));
 760 +
 761 +       dot_reclen = cpu_to_le16(de->rec_len);
 762         de = ext4_next_entry(de, blocksize);
 763         de->inode = cpu_to_le32(parent_ino);
 764         de->name_len = 2;
 765 +
 766 +       strcpy(de->name, "..");
 767 +       ext4_set_de_type(inode->i_sb, de, S_IFDIR);
 768 +       data2 = ext4_dentry_get_data(inode->i_sb,
 769 +                       (struct ext4_dentry_param *) data2);
 770 +       if (data2) {
 771 +               de->name[2] = 0;
 772 +               memcpy(&de->name[3], data2, *(char *) data2);
 773 +               de->file_type |= EXT4_DIRENT_LUFID;
 774 +       }
 775 +
 776         if (!dotdot_real_len)
 777                 de->rec_len = ext4_rec_len_to_disk(blocksize -
 778 -                                       (csum_size + ext4_dir_rec_len(1, NULL)),
 779 -                                       blocksize);
 780 +                                       (csum_size + dot_reclen), blocksize);
 781         else
 782                 de->rec_len = ext4_rec_len_to_disk(
 783 -                                       ext4_dir_rec_len(de->name_len, NULL),
 784 +                                       EXT4_DIR_ENTRY_LEN(de, NULL),
 785                                         blocksize);
 786 -       strcpy(de->name, "..");
 787 -       ext4_set_de_type(inode->i_sb, de, S_IFDIR);
 788
 789         return ext4_next_entry(de, blocksize);
 790  }
 791
 792  int ext4_init_new_dir(handle_t *handle, struct inode *dir,
 793 -                            struct inode *inode)
 794 +                            struct inode *inode,
 795 +                            const void *data1, const void *data2)
 796  {
 797 +       struct tp_block param;
 798         struct buffer_head *dir_block = NULL;
 799         struct ext4_dir_entry_2 *de;
 800         ext4_lblk_t block = 0;
 801 @@ -3071,7 +3173,11 @@ int ext4_init_new_dir(handle_t *handle, struct inode *dir,
 802         if (IS_ERR(dir_block))
 803                 return PTR_ERR(dir_block);
 804         de = (struct ext4_dir_entry_2 *)dir_block->b_data;
 805 -       ext4_init_dot_dotdot(inode, de, blocksize, csum_size, dir->i_ino, 0);
 806 +       param.inode = inode;
 807 +       param.data1 = (void *)data1;
 808 +       param.data2 = (void *)data2;
 809 +       ext4_init_dot_dotdot((struct inode *)(&param), de, blocksize,
 810 +                            csum_size, dir->i_ino, 10);
 811         set_nlink(inode, 2);
 812         if (csum_size)
 813                 ext4_initialize_dirent_tail(dir_block, blocksize);
 814 @@ -3086,6 +3192,29 @@ out:
 815         return err;
 816  }
 817
 818 +/* Initialize @inode as a subdirectory of @dir, and add the
 819 + * "." and ".." entries into the first directory block. */
 820 +int ext4_add_dot_dotdot(handle_t *handle, struct inode *dir,
 821 +                       struct inode *inode,
 822 +                       const void *data1, const void *data2)
 823 +{
 824 +       int rc;
 825 +
 826 +       if (IS_ERR(handle))
 827 +               return PTR_ERR(handle);
 828 +
 829 +       if (IS_DIRSYNC(dir))
 830 +               ext4_handle_sync(handle);
 831 +
 832 +       inode->i_op = &ext4_dir_inode_operations;
 833 +       inode->i_fop = &ext4_dir_operations;
 834 +       rc = ext4_init_new_dir(handle, dir, inode, data1, data2);
 835 +       if (!rc)
 836 +               rc = ext4_mark_inode_dirty(handle, inode);
 837 +       return rc;
 838 +}
 839 +EXPORT_SYMBOL(ext4_add_dot_dotdot);
 840 +
 841  static int ext4_mkdir(struct user_namespace *mnt_userns, struct inode *dir,
 842                       struct dentry *dentry, umode_t mode)
 843  {
 844 @@ -3113,7 +3242,7 @@ retry:
 845
 846         inode->i_op = &ext4_dir_inode_operations;
 847         inode->i_fop = &ext4_dir_operations;
 848 -       err = ext4_init_new_dir(handle, dir, inode);
 849 +       err = ext4_init_new_dir(handle, dir, inode, NULL, NULL);
 850         if (err)
 851                 goto out_clear_inode;
 852         err = ext4_mark_inode_dirty(handle, inode);
 853 diff --git a/fs/ext4/super.c b/fs/ext4/super.c
 854 index 4be1994..a2fcbf8 100644
 855 --- a/fs/ext4/super.c
 856 +++ b/fs/ext4/super.c
 857 @@ -1672,7 +1672,7 @@ enum {
 858         Opt_inlinecrypt,
 859         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
 860         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_jqfmt_vfsv1, Opt_quota,
 861 -       Opt_noquota, Opt_barrier, Opt_nobarrier, Opt_err,
 862 +       Opt_noquota, Opt_barrier, Opt_nobarrier, Opt_err, Opt_dirdata,
 863         Opt_usrquota, Opt_grpquota, Opt_prjquota, Opt_i_version,
 864         Opt_dax, Opt_dax_always, Opt_dax_inode, Opt_dax_never,
 865         Opt_stripe, Opt_delalloc, Opt_nodelalloc, Opt_warn_on_error,
 866 @@ -1756,6 +1756,7 @@ static const match_table_t tokens = {
 867         {Opt_nolazytime, "nolazytime"},
 868         {Opt_debug_want_extra_isize, "debug_want_extra_isize=%u"},
 869         {Opt_nodelalloc, "nodelalloc"},
 870 +       {Opt_dirdata, "dirdata"},
 871         {Opt_removed, "mblk_io_submit"},
 872         {Opt_removed, "nomblk_io_submit"},
 873         {Opt_block_validity, "block_validity"},
 874 @@ -2000,6 +2001,7 @@ static const struct mount_opts {
 875         {Opt_usrjquota, 0, MOPT_Q | MOPT_STRING},
 876         {Opt_grpjquota, 0, MOPT_Q | MOPT_STRING},
 877         {Opt_offusrjquota, 0, MOPT_Q},
 878 +       {Opt_dirdata, EXT4_MOUNT_DIRDATA, MOPT_SET},
 879         {Opt_offgrpjquota, 0, MOPT_Q},
 880         {Opt_jqfmt_vfsold, QFMT_VFS_OLD, MOPT_QFMT},
 881         {Opt_jqfmt_vfsv0, QFMT_VFS_V0, MOPT_QFMT},
 882 --
 883 2.34.1
 884