ldiskfs/kernel_patches/patches/sles12/ext4-large-eas.patch

   1 This patch implements the large EA support in ext4. If the size of
   2 an EA value is larger than the blocksize, then the EA value would
   3 not be saved in the external EA block, instead it would be saved
   4 in an external EA inode. So, the patch also helps support a larger
   5 number of EAs.
   6
   7 Index: linux-3.12.39-47.1/fs/ext4/ext4.h
   8 ===================================================================
   9 --- linux-3.12.39-47.1.orig/fs/ext4/ext4.h
  10 +++ linux-3.12.39-47.1/fs/ext4/ext4.h
  11 @@ -1538,6 +1538,7 @@ static inline void ext4_clear_state_flag
  12                                          EXT4_FEATURE_INCOMPAT_EXTENTS| \
  13                                          EXT4_FEATURE_INCOMPAT_64BIT| \
  14                                          EXT4_FEATURE_INCOMPAT_FLEX_BG| \
  15 +                                        EXT4_FEATURE_INCOMPAT_EA_INODE| \
  16                                          EXT4_FEATURE_INCOMPAT_MMP |    \
  17                                          EXT4_FEATURE_INCOMPAT_DIRDATA| \
  18                                          EXT4_FEATURE_INCOMPAT_INLINE_DATA)
  19 @@ -1938,6 +1939,12 @@ struct mmpd_data {
  20  #define EXT4_MMP_MAX_CHECK_INTERVAL    300UL
  21
  22  /*
  23 + * Maximum size of xattr attributes for FEATURE_INCOMPAT_EA_INODE 1Mb
  24 + * This limit is arbitrary, but is reasonable for the xattr API.
  25 + */
  26 +#define EXT4_XATTR_MAX_LARGE_EA_SIZE    (1024 * 1024)
  27 +
  28 +/*
  29   * Function prototypes
  30   */
  31
  32 @@ -1949,6 +1956,10 @@ struct mmpd_data {
  33  # define ATTRIB_NORET  __attribute__((noreturn))
  34  # define NORET_AND     noreturn,
  35
  36 +struct ext4_xattr_ino_array {
  37 +       unsigned int xia_count;         /* # of used item in the array */
  38 +       unsigned int xia_inodes[0];
  39 +};
  40  /* bitmap.c */
  41  extern unsigned int ext4_count_free(char *bitmap, unsigned numchars);
  42  void ext4_inode_bitmap_csum_set(struct super_block *sb, ext4_group_t group,
  43 @@ -2162,6 +2173,7 @@ extern void ext4_set_inode_flags(struct
  44  extern void ext4_get_inode_flags(struct ext4_inode_info *);
  45  extern int ext4_alloc_da_blocks(struct inode *inode);
  46  extern void ext4_set_aops(struct inode *inode);
  47 +extern int ext4_meta_trans_blocks(struct inode *, int nrblocks, int chunk);
  48  extern int ext4_writepage_trans_blocks(struct inode *);
  49  extern int ext4_chunk_trans_blocks(struct inode *, int nrblocks);
  50  extern int ext4_block_truncate_page(handle_t *handle,
  51 Index: linux-3.12.39-47.1/fs/ext4/inode.c
  52 ===================================================================
  53 --- linux-3.12.39-47.1.orig/fs/ext4/inode.c
  54 +++ linux-3.12.39-47.1/fs/ext4/inode.c
  55 @@ -135,8 +135,6 @@ static void ext4_invalidatepage(struct p
  56                                 unsigned int length);
  57  static int __ext4_journalled_writepage(struct page *page, unsigned int len);
  58  static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh);
  59 -static int ext4_meta_trans_blocks(struct inode *inode, int lblocks,
  60 -                                 int pextents);
  61
  62  /*
  63   * Test whether an inode is a fast symlink.
  64 @@ -182,6 +180,8 @@ void ext4_evict_inode(struct inode *inod
  65  {
  66         handle_t *handle;
  67         int err;
  68 +       int extra_credits = 3;
  69 +       struct ext4_xattr_ino_array *lea_ino_array = NULL;
  70
  71         trace_ext4_evict_inode(inode);
  72
  73 @@ -235,8 +235,8 @@ void ext4_evict_inode(struct inode *inod
  74          * protection against it
  75          */
  76         sb_start_intwrite(inode->i_sb);
  77 -       handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE,
  78 -                                   ext4_blocks_for_truncate(inode)+3);
  79 +
  80 +       handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, extra_credits);
  81         if (IS_ERR(handle)) {
  82                 ext4_std_error(inode->i_sb, PTR_ERR(handle));
  83                 /*
  84 @@ -248,9 +248,33 @@ void ext4_evict_inode(struct inode *inod
  85                 sb_end_intwrite(inode->i_sb);
  86                 goto no_delete;
  87         }
  88 -
  89         if (IS_SYNC(inode))
  90                 ext4_handle_sync(handle);
  91 +
  92 +       /*
  93 +        * Delete xattr inode before deleting the main inode.
  94 +        */
  95 +       err = ext4_xattr_delete_inode(handle, inode, &lea_ino_array);
  96 +       if (err) {
  97 +               ext4_warning(inode->i_sb,
  98 +                            "couldn't delete inode's xattr (err %d)", err);
  99 +               goto stop_handle;
 100 +       }
 101 +
 102 +       if (!ext4_handle_has_enough_credits(handle,
 103 +                       ext4_blocks_for_truncate(inode) + extra_credits)) {
 104 +               err = ext4_journal_extend(handle,
 105 +                       ext4_blocks_for_truncate(inode) + extra_credits);
 106 +               if (err > 0)
 107 +                       err = ext4_journal_restart(handle,
 108 +                       ext4_blocks_for_truncate(inode) + extra_credits);
 109 +               if (err != 0) {
 110 +                       ext4_warning(inode->i_sb,
 111 +                                    "couldn't extend journal (err %d)", err);
 112 +                       goto stop_handle;
 113 +               }
 114 +       }
 115 +
 116         inode->i_size = 0;
 117         err = ext4_mark_inode_dirty(handle, inode);
 118         if (err) {
 119 @@ -305,8 +329,12 @@ void ext4_evict_inode(struct inode *inod
 120                 ext4_clear_inode(inode);
 121         else
 122                 ext4_free_inode(handle, inode);
 123 +
 124         ext4_journal_stop(handle);
 125         sb_end_intwrite(inode->i_sb);
 126 +
 127 +       if (lea_ino_array != NULL)
 128 +               ext4_xattr_inode_array_free(inode, lea_ino_array);
 129         return;
 130  no_delete:
 131         ext4_clear_inode(inode);        /* We must guarantee clearing of inode... */
 132 @@ -4754,7 +4782,7 @@ static int ext4_index_trans_blocks(struc
 133   *
 134   * Also account for superblock, inode, quota and xattr blocks
 135   */
 136 -static int ext4_meta_trans_blocks(struct inode *inode, int lblocks,
 137 +int ext4_meta_trans_blocks(struct inode *inode, int lblocks,
 138                                   int pextents)
 139  {
 140         ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb);
 141 Index: linux-3.12.39-47.1/fs/ext4/xattr.c
 142 ===================================================================
 143 --- linux-3.12.39-47.1.orig/fs/ext4/xattr.c
 144 +++ linux-3.12.39-47.1/fs/ext4/xattr.c
 145 @@ -233,19 +233,26 @@ ext4_xattr_check_block(struct inode *ino
 146  }
 147
 148  static inline int
 149 -ext4_xattr_check_entry(struct ext4_xattr_entry *entry, size_t size)
 150 +ext4_xattr_check_entry(struct ext4_xattr_entry *entry, size_t size,
 151 +                      struct inode *inode)
 152  {
 153         size_t value_size = le32_to_cpu(entry->e_value_size);
 154
 155 -       if (entry->e_value_block != 0 || value_size > size ||
 156 -           le16_to_cpu(entry->e_value_offs) + value_size > size)
 157 +       if ((entry->e_value_inum == 0) &&
 158 +          (le16_to_cpu(entry->e_value_offs) + value_size > size))
 159 +               return -EIO;
 160 +       if (entry->e_value_inum != 0 &&
 161 +           (le32_to_cpu(entry->e_value_inum) < EXT4_FIRST_INO(inode->i_sb) ||
 162 +            le32_to_cpu(entry->e_value_inum) >
 163 +            le32_to_cpu(EXT4_SB(inode->i_sb)->s_es->s_inodes_count)))
 164                 return -EIO;
 165         return 0;
 166  }
 167
 168  static int
 169  ext4_xattr_find_entry(struct ext4_xattr_entry **pentry, int name_index,
 170 -                     const char *name, size_t size, int sorted)
 171 +                     const char *name, size_t size, int sorted,
 172 +                     struct inode *inode)
 173  {
 174         struct ext4_xattr_entry *entry;
 175         size_t name_len;
 176 @@ -265,11 +272,103 @@ ext4_xattr_find_entry(struct ext4_xattr_
 177                         break;
 178         }
 179         *pentry = entry;
 180 -       if (!cmp && ext4_xattr_check_entry(entry, size))
 181 +       if (!cmp && ext4_xattr_check_entry(entry, size, inode))
 182                         return -EIO;
 183         return cmp ? -ENODATA : 0;
 184  }
 185
 186 +/*
 187 + * Read the EA value from an inode.
 188 + */
 189 +static int
 190 +ext4_xattr_inode_read(struct inode *ea_inode, void *buf, size_t *size)
 191 +{
 192 +       unsigned long block = 0;
 193 +       struct buffer_head *bh = NULL;
 194 +       int err, blocksize;
 195 +       size_t csize, ret_size = 0;
 196 +
 197 +       if (*size == 0)
 198 +               return 0;
 199 +
 200 +       blocksize = ea_inode->i_sb->s_blocksize;
 201 +
 202 +       while (ret_size < *size) {
 203 +               csize = (*size - ret_size) > blocksize ? blocksize :
 204 +                                                       *size - ret_size;
 205 +               bh = ext4_bread(NULL, ea_inode, block, 0, &err);
 206 +               if (!bh) {
 207 +                       *size = ret_size;
 208 +                       return err;
 209 +               }
 210 +               memcpy(buf, bh->b_data, csize);
 211 +               brelse(bh);
 212 +
 213 +               buf += csize;
 214 +               block += 1;
 215 +               ret_size += csize;
 216 +       }
 217 +
 218 +       *size = ret_size;
 219 +
 220 +       return err;
 221 +}
 222 +
 223 +struct inode *ext4_xattr_inode_iget(struct inode *parent, int ea_ino, int *err)
 224 +{
 225 +       struct inode *ea_inode = NULL;
 226 +
 227 +       ea_inode = ext4_iget(parent->i_sb, ea_ino);
 228 +       if (IS_ERR(ea_inode) || is_bad_inode(ea_inode)) {
 229 +               ext4_error(parent->i_sb, "error while reading EA inode %d",
 230 +                          ea_ino);
 231 +               *err = -EIO;
 232 +               return NULL;
 233 +       }
 234 +
 235 +       if (ea_inode->i_xattr_inode_parent != parent->i_ino ||
 236 +           ea_inode->i_generation != parent->i_generation) {
 237 +               ext4_error(parent->i_sb, "Backpointer from EA inode %d "
 238 +                          "to parent invalid.", ea_ino);
 239 +               *err = -EINVAL;
 240 +               goto error;
 241 +       }
 242 +
 243 +       if (!(EXT4_I(ea_inode)->i_flags & EXT4_EA_INODE_FL)) {
 244 +               ext4_error(parent->i_sb, "EA inode %d does not have "
 245 +                          "EXT4_EA_INODE_FL flag set.\n", ea_ino);
 246 +               *err = -EINVAL;
 247 +               goto error;
 248 +       }
 249 +
 250 +       *err = 0;
 251 +       return ea_inode;
 252 +
 253 +error:
 254 +       iput(ea_inode);
 255 +       return NULL;
 256 +}
 257 +
 258 +/*
 259 + * Read the value from the EA inode.
 260 + */
 261 +static int
 262 +ext4_xattr_inode_get(struct inode *inode, int ea_ino, void *buffer,
 263 +                    size_t *size)
 264 +{
 265 +       struct inode *ea_inode = NULL;
 266 +       int err;
 267 +
 268 +       ea_inode = ext4_xattr_inode_iget(inode, ea_ino, &err);
 269 +       if (err)
 270 +               return err;
 271 +
 272 +       err = ext4_xattr_inode_read(ea_inode, buffer, size);
 273 +       iput(ea_inode);
 274 +
 275 +       return err;
 276 +}
 277 +
 278  static int
 279  ext4_xattr_block_get(struct inode *inode, int name_index, const char *name,
 280                      void *buffer, size_t buffer_size)
 281 @@ -301,7 +400,8 @@ bad_block:
 282         }
 283         ext4_xattr_cache_insert(bh);
 284         entry = BFIRST(bh);
 285 -       error = ext4_xattr_find_entry(&entry, name_index, name, bh->b_size, 1);
 286 +       error = ext4_xattr_find_entry(&entry, name_index, name, bh->b_size, 1,
 287 +                                     inode);
 288         if (error == -EIO)
 289                 goto bad_block;
 290         if (error)
 291 @@ -311,8 +411,16 @@ bad_block:
 292                 error = -ERANGE;
 293                 if (size > buffer_size)
 294                         goto cleanup;
 295 -               memcpy(buffer, bh->b_data + le16_to_cpu(entry->e_value_offs),
 296 -                      size);
 297 +               if (entry->e_value_inum != 0) {
 298 +                       error = ext4_xattr_inode_get(inode,
 299 +                                            le32_to_cpu(entry->e_value_inum),
 300 +                                            buffer, &size);
 301 +                       if (error)
 302 +                               goto cleanup;
 303 +               } else {
 304 +                       memcpy(buffer, bh->b_data +
 305 +                              le16_to_cpu(entry->e_value_offs), size);
 306 +               }
 307         }
 308         error = size;
 309
 310 @@ -346,7 +454,7 @@ ext4_xattr_ibody_get(struct inode *inode
 311         if (error)
 312                 goto cleanup;
 313         error = ext4_xattr_find_entry(&entry, name_index, name,
 314 -                                     end - (void *)entry, 0);
 315 +                                     end - (void *)entry, 0, inode);
 316         if (error)
 317                 goto cleanup;
 318         size = le32_to_cpu(entry->e_value_size);
 319 @@ -354,8 +462,16 @@ ext4_xattr_ibody_get(struct inode *inode
 320                 error = -ERANGE;
 321                 if (size > buffer_size)
 322                         goto cleanup;
 323 -               memcpy(buffer, (void *)IFIRST(header) +
 324 -                      le16_to_cpu(entry->e_value_offs), size);
 325 +               if (entry->e_value_inum != 0) {
 326 +                       error = ext4_xattr_inode_get(inode,
 327 +                                            le32_to_cpu(entry->e_value_inum),
 328 +                                            buffer, &size);
 329 +                       if (error)
 330 +                               goto cleanup;
 331 +               } else {
 332 +                       memcpy(buffer, (void *)IFIRST(header) +
 333 +                              le16_to_cpu(entry->e_value_offs), size);
 334 +               }
 335         }
 336         error = size;
 337
 338 @@ -597,7 +713,7 @@ static size_t ext4_xattr_free_space(stru
 339  {
 340         for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
 341                 *total += EXT4_XATTR_LEN(last->e_name_len);
 342 -               if (!last->e_value_block && last->e_value_size) {
 343 +               if (last->e_value_inum == 0 && last->e_value_size > 0) {
 344                         size_t offs = le16_to_cpu(last->e_value_offs);
 345                         if (offs < *min_offs)
 346                                 *min_offs = offs;
 347 @@ -606,16 +722,172 @@ static size_t ext4_xattr_free_space(stru
 348         return (*min_offs - ((void *)last - base) - sizeof(__u32));
 349  }
 350
 351 +/*
 352 + * Write the value of the EA in an inode.
 353 + */
 354 +static int
 355 +ext4_xattr_inode_write(handle_t *handle, struct inode *ea_inode,
 356 +                      const void *buf, int bufsize)
 357 +{
 358 +       struct buffer_head *bh = NULL;
 359 +       unsigned long block = 0;
 360 +       unsigned blocksize = ea_inode->i_sb->s_blocksize;
 361 +       unsigned max_blocks = (bufsize + blocksize - 1) >> ea_inode->i_blkbits;
 362 +       int csize, wsize = 0;
 363 +       int ret = 0;
 364 +       int retries = 0;
 365 +
 366 +retry:
 367 +       while (ret >= 0 && ret < max_blocks) {
 368 +               struct ext4_map_blocks map;
 369 +               map.m_lblk = block += ret;
 370 +               map.m_len = max_blocks -= ret;
 371 +
 372 +               ret = ext4_map_blocks(handle, ea_inode, &map,
 373 +                                     EXT4_GET_BLOCKS_CREATE);
 374 +               if (ret <= 0) {
 375 +                       ext4_mark_inode_dirty(handle, ea_inode);
 376 +                       if (ret == -ENOSPC &&
 377 +                           ext4_should_retry_alloc(ea_inode->i_sb, &retries)) {
 378 +                               ret = 0;
 379 +                               goto retry;
 380 +                       }
 381 +                       break;
 382 +               }
 383 +       }
 384 +
 385 +       if (ret < 0)
 386 +               return ret;
 387 +
 388 +       block = 0;
 389 +       while (wsize < bufsize) {
 390 +               if (bh != NULL)
 391 +                       brelse(bh);
 392 +               csize = (bufsize - wsize) > blocksize ? blocksize :
 393 +                                                               bufsize - wsize;
 394 +               bh = ext4_getblk(handle, ea_inode, block, 0, &ret);
 395 +               if (!bh)
 396 +                       goto out;
 397 +               ret = ext4_journal_get_write_access(handle, bh);
 398 +               if (ret)
 399 +                       goto out;
 400 +
 401 +               memcpy(bh->b_data, buf, csize);
 402 +               set_buffer_uptodate(bh);
 403 +               ext4_handle_dirty_metadata(handle, ea_inode, bh);
 404 +
 405 +               buf += csize;
 406 +               wsize += csize;
 407 +               block += 1;
 408 +       }
 409 +
 410 +       i_size_write(ea_inode, wsize);
 411 +       ext4_update_i_disksize(ea_inode, wsize);
 412 +
 413 +       ext4_mark_inode_dirty(handle, ea_inode);
 414 +
 415 +out:
 416 +       brelse(bh);
 417 +
 418 +       return ret;
 419 +}
 420 +
 421 +/*
 422 + * Create an inode to store the value of a large EA.
 423 + */
 424 +static struct inode *
 425 +ext4_xattr_inode_create(handle_t *handle, struct inode *inode)
 426 +{
 427 +       struct inode *ea_inode = NULL;
 428 +
 429 +       /*
 430 +        * Let the next inode be the goal, so we try and allocate the EA inode
 431 +        * in the same group, or nearby one.
 432 +        */
 433 +       ea_inode = ext4_new_inode(handle, inode->i_sb->s_root->d_inode,
 434 +                                 S_IFREG|0600, NULL, inode->i_ino + 1, NULL);
 435 +
 436 +       if (!IS_ERR(ea_inode)) {
 437 +               ea_inode->i_op = &ext4_file_inode_operations;
 438 +               ea_inode->i_fop = &ext4_file_operations;
 439 +               ext4_set_aops(ea_inode);
 440 +               ea_inode->i_generation = inode->i_generation;
 441 +               EXT4_I(ea_inode)->i_flags |= EXT4_EA_INODE_FL;
 442 +
 443 +               /*
 444 +                * A back-pointer from EA inode to parent inode will be useful
 445 +                * for e2fsck.
 446 +                */
 447 +               ea_inode->i_xattr_inode_parent = inode->i_ino;
 448 +               unlock_new_inode(ea_inode);
 449 +       }
 450 +
 451 +       return ea_inode;
 452 +}
 453 +
 454 +/*
 455 + * Unlink the inode storing the value of the EA.
 456 + */
 457 +int
 458 +ext4_xattr_inode_unlink(struct inode *inode, int ea_ino)
 459 +{
 460 +       struct inode *ea_inode = NULL;
 461 +       int err;
 462 +
 463 +       ea_inode = ext4_xattr_inode_iget(inode, ea_ino, &err);
 464 +       if (err)
 465 +               return err;
 466 +
 467 +       clear_nlink(ea_inode);
 468 +       iput(ea_inode);
 469 +
 470 +       return 0;
 471 +}
 472 +
 473 +/*
 474 + * Add value of the EA in an inode.
 475 + */
 476  static int
 477 -ext4_xattr_set_entry(struct ext4_xattr_info *i, struct ext4_xattr_search *s)
 478 +ext4_xattr_inode_set(handle_t *handle, struct inode *inode, int *ea_ino,
 479 +                    const void *value, size_t value_len)
 480 +{
 481 +       struct inode *ea_inode = NULL;
 482 +       int err;
 483 +
 484 +       /* Create an inode for the EA value */
 485 +       ea_inode = ext4_xattr_inode_create(handle, inode);
 486 +       if (IS_ERR(ea_inode))
 487 +               return -1;
 488 +
 489 +       err = ext4_xattr_inode_write(handle, ea_inode, value, value_len);
 490 +       if (err)
 491 +               clear_nlink(ea_inode);
 492 +       else
 493 +               *ea_ino = ea_inode->i_ino;
 494 +
 495 +       iput(ea_inode);
 496 +
 497 +       return err;
 498 +}
 499 +
 500 +static int
 501 +ext4_xattr_set_entry(struct ext4_xattr_info *i, struct ext4_xattr_search *s,
 502 +                    handle_t *handle, struct inode *inode)
 503  {
 504         struct ext4_xattr_entry *last;
 505         size_t free, min_offs = s->end - s->base, name_len = strlen(i->name);
 506 +       int in_inode = i->in_inode;
 507 +
 508 +       if (EXT4_HAS_INCOMPAT_FEATURE(inode->i_sb,
 509 +                EXT4_FEATURE_INCOMPAT_EA_INODE) &&
 510 +           (EXT4_XATTR_SIZE(i->value_len) >
 511 +            EXT4_XATTR_MIN_LARGE_EA_SIZE(inode->i_sb->s_blocksize)))
 512 +               in_inode = 1;
 513
 514         /* Compute min_offs and last. */
 515         last = s->first;
 516         for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
 517 -               if (!last->e_value_block && last->e_value_size) {
 518 +               if (last->e_value_inum == 0 && last->e_value_size > 0) {
 519                         size_t offs = le16_to_cpu(last->e_value_offs);
 520                         if (offs < min_offs)
 521                                 min_offs = offs;
 522 @@ -623,16 +895,21 @@ ext4_xattr_set_entry(struct ext4_xattr_i
 523         }
 524         free = min_offs - ((void *)last - s->base) - sizeof(__u32);
 525         if (!s->not_found) {
 526 -               if (!s->here->e_value_block && s->here->e_value_size) {
 527 +               if (!in_inode && s->here->e_value_inum == 0 &&
 528 +                   s->here->e_value_size > 0) {
 529                         size_t size = le32_to_cpu(s->here->e_value_size);
 530                         free += EXT4_XATTR_SIZE(size);
 531                 }
 532                 free += EXT4_XATTR_LEN(name_len);
 533         }
 534         if (i->value) {
 535 -               if (free < EXT4_XATTR_SIZE(i->value_len) ||
 536 -                   free < EXT4_XATTR_LEN(name_len) +
 537 -                          EXT4_XATTR_SIZE(i->value_len))
 538 +               size_t value_len = EXT4_XATTR_SIZE(i->value_len);
 539 +
 540 +               if (in_inode)
 541 +                       value_len = 0;
 542 +
 543 +               if (free < value_len ||
 544 +                   free < EXT4_XATTR_LEN(name_len) + value_len)
 545                         return -ENOSPC;
 546         }
 547
 548 @@ -646,7 +923,8 @@ ext4_xattr_set_entry(struct ext4_xattr_i
 549                 s->here->e_name_len = name_len;
 550                 memcpy(s->here->e_name, i->name, name_len);
 551         } else {
 552 -               if (!s->here->e_value_block && s->here->e_value_size) {
 553 +               if (s->here->e_value_offs > 0 && s->here->e_value_inum == 0 &&
 554 +                   s->here->e_value_size > 0) {
 555                         void *first_val = s->base + min_offs;
 556                         size_t offs = le16_to_cpu(s->here->e_value_offs);
 557                         void *val = s->base + offs;
 558 @@ -680,13 +958,17 @@ ext4_xattr_set_entry(struct ext4_xattr_i
 559                         last = s->first;
 560                         while (!IS_LAST_ENTRY(last)) {
 561                                 size_t o = le16_to_cpu(last->e_value_offs);
 562 -                               if (!last->e_value_block &&
 563 -                                   last->e_value_size && o < offs)
 564 +                               if (last->e_value_size > 0 && o < offs)
 565                                         last->e_value_offs =
 566                                                 cpu_to_le16(o + size);
 567                                 last = EXT4_XATTR_NEXT(last);
 568                         }
 569                 }
 570 +               if (s->here->e_value_inum != 0) {
 571 +                       ext4_xattr_inode_unlink(inode,
 572 +                                       le32_to_cpu(s->here->e_value_inum));
 573 +                       s->here->e_value_inum = 0;
 574 +               }
 575                 if (!i->value) {
 576                         /* Remove the old name. */
 577                         size_t size = EXT4_XATTR_LEN(name_len);
 578 @@ -700,10 +982,17 @@ ext4_xattr_set_entry(struct ext4_xattr_i
 579         if (i->value) {
 580                 /* Insert the new value. */
 581                 s->here->e_value_size = cpu_to_le32(i->value_len);
 582 -               if (i->value_len) {
 583 +               if (in_inode) {
 584 +                       int ea_ino = le32_to_cpu(s->here->e_value_inum);
 585 +                       ext4_xattr_inode_set(handle, inode, &ea_ino, i->value,
 586 +                                            i->value_len);
 587 +                       s->here->e_value_inum = cpu_to_le32(ea_ino);
 588 +                       s->here->e_value_offs = 0;
 589 +               } else if (i->value_len) {
 590                         size_t size = EXT4_XATTR_SIZE(i->value_len);
 591                         void *val = s->base + min_offs - size;
 592                         s->here->e_value_offs = cpu_to_le16(min_offs - size);
 593 +                       s->here->e_value_inum = 0;
 594                         if (i->value == EXT4_ZERO_XATTR_VALUE) {
 595                                 memset(val, 0, size);
 596                         } else {
 597 @@ -753,7 +1042,7 @@ ext4_xattr_block_find(struct inode *inod
 598                 bs->s.end = bs->bh->b_data + bs->bh->b_size;
 599                 bs->s.here = bs->s.first;
 600                 error = ext4_xattr_find_entry(&bs->s.here, i->name_index,
 601 -                                             i->name, bs->bh->b_size, 1);
 602 +                                            i->name, bs->bh->b_size, 1, inode);
 603                 if (error && error != -ENODATA)
 604                         goto cleanup;
 605                 bs->s.not_found = error;
 606 @@ -777,8 +1066,6 @@ ext4_xattr_block_set(handle_t *handle, s
 607
 608  #define header(x) ((struct ext4_xattr_header *)(x))
 609
 610 -       if (i->value && i->value_len > sb->s_blocksize)
 611 -               return -ENOSPC;
 612         if (s->base) {
 613                 ce = mb_cache_entry_get(ext4_xattr_cache, bs->bh->b_bdev,
 614                                         bs->bh->b_blocknr);
 615 @@ -794,7 +1081,7 @@ ext4_xattr_block_set(handle_t *handle, s
 616                                 ce = NULL;
 617                         }
 618                         ea_bdebug(bs->bh, "modifying in-place");
 619 -                       error = ext4_xattr_set_entry(i, s);
 620 +                       error = ext4_xattr_set_entry(i, s, handle, inode);
 621                         if (!error) {
 622                                 if (!IS_LAST_ENTRY(s->first))
 623                                         ext4_xattr_rehash(header(s->base),
 624 @@ -845,7 +1132,7 @@ ext4_xattr_block_set(handle_t *handle, s
 625                 s->end = s->base + sb->s_blocksize;
 626         }
 627
 628 -       error = ext4_xattr_set_entry(i, s);
 629 +       error = ext4_xattr_set_entry(i, s, handle, inode);
 630         if (error == -EIO)
 631                 goto bad_block;
 632         if (error)
 633 @@ -994,7 +1281,7 @@ int ext4_xattr_ibody_find(struct inode *
 634                 /* Find the named attribute. */
 635                 error = ext4_xattr_find_entry(&is->s.here, i->name_index,
 636                                               i->name, is->s.end -
 637 -                                             (void *)is->s.base, 0);
 638 +                                             (void *)is->s.base, 0, inode);
 639                 if (error && error != -ENODATA)
 640                         return error;
 641                 is->s.not_found = error;
 642 @@ -1012,7 +1299,7 @@ int ext4_xattr_ibody_inline_set(handle_t
 643
 644         if (EXT4_I(inode)->i_extra_isize == 0)
 645                 return -ENOSPC;
 646 -       error = ext4_xattr_set_entry(i, s);
 647 +       error = ext4_xattr_set_entry(i, s, handle, inode);
 648         if (error) {
 649                 if (error == -ENOSPC &&
 650                     ext4_has_inline_data(inode)) {
 651 @@ -1024,7 +1311,7 @@ int ext4_xattr_ibody_inline_set(handle_t
 652                         error = ext4_xattr_ibody_find(inode, i, is);
 653                         if (error)
 654                                 return error;
 655 -                       error = ext4_xattr_set_entry(i, s);
 656 +                       error = ext4_xattr_set_entry(i, s, handle, inode);
 657                 }
 658                 if (error)
 659                         return error;
 660 @@ -1050,7 +1337,7 @@ static int ext4_xattr_ibody_set(handle_t
 661
 662         if (EXT4_I(inode)->i_extra_isize == 0)
 663                 return -ENOSPC;
 664 -       error = ext4_xattr_set_entry(i, s);
 665 +       error = ext4_xattr_set_entry(i, s, handle, inode);
 666         if (error)
 667                 return error;
 668         header = IHDR(inode, ext4_raw_inode(&is->iloc));
 669 @@ -1086,7 +1373,7 @@ ext4_xattr_set_handle(handle_t *handle,
 670                 .name = name,
 671                 .value = value,
 672                 .value_len = value_len,
 673 -
 674 +               .in_inode = 0,
 675         };
 676         struct ext4_xattr_ibody_find is = {
 677                 .s = { .not_found = -ENODATA, },
 678 @@ -1151,6 +1438,15 @@ ext4_xattr_set_handle(handle_t *handle,
 679                                         goto cleanup;
 680                         }
 681                         error = ext4_xattr_block_set(handle, inode, &i, &bs);
 682 +                       if (EXT4_HAS_INCOMPAT_FEATURE(inode->i_sb,
 683 +                                       EXT4_FEATURE_INCOMPAT_EA_INODE) &&
 684 +                           error == -ENOSPC) {
 685 +                               /* xattr not fit to block, store at external
 686 +                                * inode */
 687 +                               i.in_inode = 1;
 688 +                               error = ext4_xattr_ibody_set(handle, inode,
 689 +                                                            &i, &is);
 690 +                       }
 691                         if (error)
 692                                 goto cleanup;
 693                         if (!is.s.not_found) {
 694 @@ -1197,9 +1493,22 @@ ext4_xattr_set(struct inode *inode, int
 695                const void *value, size_t value_len, int flags)
 696  {
 697         handle_t *handle;
 698 +       struct super_block *sb = inode->i_sb;
 699         int error, retries = 0;
 700         int credits = ext4_jbd2_credits_xattr(inode);
 701
 702 +       if ((value_len >= EXT4_XATTR_MIN_LARGE_EA_SIZE(sb->s_blocksize)) &&
 703 +           EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EA_INODE)) {
 704 +               int nrblocks = (value_len + sb->s_blocksize - 1) >>
 705 +                                       sb->s_blocksize_bits;
 706 +
 707 +               /* For new inode */
 708 +               credits += EXT4_SINGLEDATA_TRANS_BLOCKS(sb) + 3;
 709 +
 710 +               /* For data blocks of EA inode */
 711 +               credits += ext4_meta_trans_blocks(inode, nrblocks, 0);
 712 +       }
 713 +
 714  retry:
 715         handle = ext4_journal_start(inode, EXT4_HT_XATTR, credits);
 716         if (IS_ERR(handle)) {
 717 @@ -1211,7 +1520,7 @@ retry:
 718                                               value, value_len, flags);
 719                 error2 = ext4_journal_stop(handle);
 720                 if (error == -ENOSPC &&
 721 -                   ext4_should_retry_alloc(inode->i_sb, &retries))
 722 +                   ext4_should_retry_alloc(sb, &retries))
 723                         goto retry;
 724                 if (error == 0)
 725                         error = error2;
 726 @@ -1233,7 +1542,7 @@ static void ext4_xattr_shift_entries(str
 727
 728         /* Adjust the value offsets of the entries */
 729         for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
 730 -               if (!last->e_value_block && last->e_value_size) {
 731 +               if (last->e_value_inum == 0 && last->e_value_size > 0) {
 732                         new_offs = le16_to_cpu(last->e_value_offs) +
 733                                                         value_offs_shift;
 734                         BUG_ON(new_offs + le32_to_cpu(last->e_value_size)
 735 @@ -1472,19 +1781,91 @@ cleanup:
 736  }
 737
 738
 739 -
 740 +#define EIA_INCR 16 /* must be 2^n */
 741 +#define EIA_MASK (EIA_INCR - 1)
 742 +/* Add the large xattr @ino into @lea_ino_array for later deletion.
 743 + * If @lea_ino_array is new or full it will be grown and the old
 744 + * contents copied over.
 745 + */
 746 +static int
 747 +ext4_expand_ino_array(struct ext4_xattr_ino_array **lea_ino_array, __u32 ino)
 748 +{
 749 +       if (*lea_ino_array == NULL) {
 750 +               /*
 751 +                * Start with 15 inodes, so it fits into a power-of-two size.
 752 +                * If *lea_ino_array is NULL, this is essentially offsetof()
 753 +                */
 754 +               (*lea_ino_array) =
 755 +                       kmalloc(offsetof(struct ext4_xattr_ino_array,
 756 +                                        xia_inodes[EIA_MASK]),
 757 +                               GFP_NOFS);
 758 +               if (*lea_ino_array == NULL)
 759 +                       return -ENOMEM;
 760 +               (*lea_ino_array)->xia_count = 0;
 761 +       } else if (((*lea_ino_array)->xia_count & EIA_MASK) == EIA_MASK) {
 762 +               /* expand the array once all 15 + n * 16 slots are full */
 763 +               struct ext4_xattr_ino_array *new_array = NULL;
 764 +               int count = (*lea_ino_array)->xia_count;
 765 +
 766 +               /* if new_array is NULL, this is essentially offsetof() */
 767 +               new_array = kmalloc(
 768 +                               offsetof(struct ext4_xattr_ino_array,
 769 +                                        xia_inodes[count + EIA_INCR]),
 770 +                               GFP_NOFS);
 771 +               if (new_array == NULL)
 772 +                       return -ENOMEM;
 773 +               memcpy(new_array, *lea_ino_array,
 774 +                      offsetof(struct ext4_xattr_ino_array,
 775 +                               xia_inodes[count]));
 776 +               kfree(*lea_ino_array);
 777 +               *lea_ino_array = new_array;
 778 +       }
 779 +       (*lea_ino_array)->xia_inodes[(*lea_ino_array)->xia_count++] = ino;
 780 +       return 0;
 781 +}
 782  /*
 783   * ext4_xattr_delete_inode()
 784   *
 785 - * Free extended attribute resources associated with this inode. This
 786 + * Free extended attribute resources associated with this inode. Traverse
 787 + * all entries and unlink any xattr inodes associated with this inode. This
 788   * is called immediately before an inode is freed. We have exclusive
 789 - * access to the inode.
 790 + * access to the inode. If an orphan inode is deleted it will also delete any
 791 + * xattr block and all xattr inodes. They are checked by ext4_xattr_inode_iget()
 792 + * to ensure they belong to the parent inode and were not deleted already.
 793   */
 794 -void
 795 -ext4_xattr_delete_inode(handle_t *handle, struct inode *inode)
 796 +int
 797 +ext4_xattr_delete_inode(handle_t *handle, struct inode *inode,
 798 +                       struct ext4_xattr_ino_array **lea_ino_array)
 799  {
 800         struct buffer_head *bh = NULL;
 801 +       struct ext4_xattr_ibody_header *header;
 802 +       struct ext4_inode *raw_inode;
 803 +       struct ext4_iloc iloc;
 804 +       struct ext4_xattr_entry *entry;
 805 +       int error = 0;
 806 +
 807 +       if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR))
 808 +               goto delete_external_ea;
 809 +
 810 +       error = ext4_get_inode_loc(inode, &iloc);
 811 +       if (error)
 812 +               goto cleanup;
 813 +       raw_inode = ext4_raw_inode(&iloc);
 814 +       header = IHDR(inode, raw_inode);
 815 +       entry = IFIRST(header);
 816 +       for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry)) {
 817 +               if (entry->e_value_inum == 0)
 818 +                       continue;
 819 +               if (ext4_expand_ino_array(lea_ino_array,
 820 +                                         entry->e_value_inum) != 0) {
 821 +                       brelse(iloc.bh);
 822 +                       goto cleanup;
 823 +               }
 824 +               entry->e_value_inum = 0;
 825 +       }
 826 +       brelse(iloc.bh);
 827
 828 +delete_external_ea:
 829         if (!EXT4_I(inode)->i_file_acl)
 830                 goto cleanup;
 831         bh = sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl);
 832 @@ -1499,11 +1880,74 @@ ext4_xattr_delete_inode(handle_t *handle
 833                                  EXT4_I(inode)->i_file_acl);
 834                 goto cleanup;
 835         }
 836 +
 837 +       entry = BFIRST(bh);
 838 +       for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry)) {
 839 +               if (entry->e_value_inum == 0)
 840 +                       continue;
 841 +               if (ext4_expand_ino_array(lea_ino_array,
 842 +                                         entry->e_value_inum) != 0)
 843 +                       goto cleanup;
 844 +               entry->e_value_inum = 0;
 845 +       }
 846 +
 847 +       /* add xattr inode to orphan list */
 848 +       if (*lea_ino_array != NULL) {
 849 +               struct inode *ea_inode = NULL;
 850 +               int idx = 0;
 851 +
 852 +               for (; idx < (*lea_ino_array)->xia_count; ++idx) {
 853 +                       if (!ext4_handle_has_enough_credits(handle, 3)) {
 854 +                               error = ext4_journal_extend(handle, 3);
 855 +                               if (error > 0)
 856 +                                       error = ext4_journal_restart(handle, 3);
 857 +                               if (error != 0) {
 858 +                                       ext4_warning(inode->i_sb,
 859 +                                               "couldn't extend journal "
 860 +                                               "(err %d)", error);
 861 +                                       goto cleanup;
 862 +                               }
 863 +                       }
 864 +                       ea_inode = ext4_xattr_inode_iget(inode,
 865 +                               (*lea_ino_array)->xia_inodes[idx], &error);
 866 +                       if (error)
 867 +                               continue;
 868 +                       ext4_orphan_add(handle, ea_inode);
 869 +                       /* the inode's i_count will be released by caller */
 870 +               }
 871 +       }
 872 +
 873         ext4_xattr_release_block(handle, inode, bh);
 874         EXT4_I(inode)->i_file_acl = 0;
 875
 876  cleanup:
 877         brelse(bh);
 878 +
 879 +       return error;
 880 +}
 881 +
 882 +void
 883 +ext4_xattr_inode_array_free(struct inode *inode,
 884 +                           struct ext4_xattr_ino_array *lea_ino_array)
 885 +{
 886 +       struct inode    *ea_inode = NULL;
 887 +       int             idx = 0;
 888 +       int             err;
 889 +
 890 +       if (lea_ino_array == NULL)
 891 +               return;
 892 +
 893 +       for (; idx < lea_ino_array->xia_count; ++idx) {
 894 +               ea_inode = ext4_xattr_inode_iget(inode,
 895 +                               lea_ino_array->xia_inodes[idx], &err);
 896 +               if (err)
 897 +                       continue;
 898 +               clear_nlink(ea_inode);
 899 +               iput(ea_inode);
 900 +               /* for inode's i_count get from ext4_xattr_delete_inode */
 901 +               iput(ea_inode);
 902 +       }
 903 +       kfree(lea_ino_array);
 904  }
 905
 906  /*
 907 @@ -1573,10 +2017,9 @@ ext4_xattr_cmp(struct ext4_xattr_header
 908                     entry1->e_name_index != entry2->e_name_index ||
 909                     entry1->e_name_len != entry2->e_name_len ||
 910                     entry1->e_value_size != entry2->e_value_size ||
 911 +                   entry1->e_value_inum != entry2->e_value_inum ||
 912                     memcmp(entry1->e_name, entry2->e_name, entry1->e_name_len))
 913                         return 1;
 914 -               if (entry1->e_value_block != 0 || entry2->e_value_block != 0)
 915 -                       return -EIO;
 916                 if (memcmp((char *)header1 + le16_to_cpu(entry1->e_value_offs),
 917                            (char *)header2 + le16_to_cpu(entry2->e_value_offs),
 918                            le32_to_cpu(entry1->e_value_size)))
 919 @@ -1660,7 +2103,7 @@ static inline void ext4_xattr_hash_entry
 920                        *name++;
 921         }
 922
 923 -       if (entry->e_value_block == 0 && entry->e_value_size != 0) {
 924 +       if (entry->e_value_inum == 0 && entry->e_value_size != 0) {
 925                 __le32 *value = (__le32 *)((char *)header +
 926                         le16_to_cpu(entry->e_value_offs));
 927                 for (n = (le32_to_cpu(entry->e_value_size) +
 928 Index: linux-3.12.39-47.1/fs/ext4/xattr.h
 929 ===================================================================
 930 --- linux-3.12.39-47.1.orig/fs/ext4/xattr.h
 931 +++ linux-3.12.39-47.1/fs/ext4/xattr.h
 932 @@ -42,7 +42,7 @@ struct ext4_xattr_entry {
 933         __u8    e_name_len;     /* length of name */
 934         __u8    e_name_index;   /* attribute name index */
 935         __le16  e_value_offs;   /* offset in disk block of value */
 936 -       __le32  e_value_block;  /* disk block attribute is stored on (n/i) */
 937 +       __le32  e_value_inum;   /* inode in which the value is stored */
 938         __le32  e_value_size;   /* size of attribute value */
 939         __le32  e_hash;         /* hash value of name and value */
 940         char    e_name[0];      /* attribute name */
 941 @@ -67,6 +67,15 @@ struct ext4_xattr_entry {
 942                 EXT4_I(inode)->i_extra_isize))
 943  #define IFIRST(hdr) ((struct ext4_xattr_entry *)((hdr)+1))
 944
 945 +#define i_xattr_inode_parent i_mtime.tv_sec
 946 +
 947 +/*
 948 + * The minimum size of EA value when you start storing it in an external inode
 949 + * size of block - size of header - size of 1 entry - 4 null bytes
 950 +*/
 951 +#define EXT4_XATTR_MIN_LARGE_EA_SIZE(b)                                        \
 952 +       ((b) - EXT4_XATTR_LEN(3) - sizeof(struct ext4_xattr_header) - 4)
 953 +
 954  #define BHDR(bh) ((struct ext4_xattr_header *)((bh)->b_data))
 955  #define ENTRY(ptr) ((struct ext4_xattr_entry *)(ptr))
 956  #define BFIRST(bh) ENTRY(BHDR(bh)+1)
 957 @@ -75,10 +84,11 @@ struct ext4_xattr_entry {
 958  #define EXT4_ZERO_XATTR_VALUE ((void *)-1)
 959
 960  struct ext4_xattr_info {
 961 -       int name_index;
 962         const char *name;
 963         const void *value;
 964         size_t value_len;
 965 +       int name_index;
 966 +       int in_inode;
 967  };
 968
 969  struct ext4_xattr_search {
 970 @@ -106,7 +116,13 @@ extern int ext4_xattr_get(struct inode *
 971  extern int ext4_xattr_set(struct inode *, int, const char *, const void *, size_t, int);
 972  extern int ext4_xattr_set_handle(handle_t *, struct inode *, int, const char *, const void *, size_t, int);
 973
 974 -extern void ext4_xattr_delete_inode(handle_t *, struct inode *);
 975 +extern struct inode *ext4_xattr_inode_iget(struct inode *parent, int ea_ino,
 976 +                                          int *err);
 977 +extern int ext4_xattr_inode_unlink(struct inode *inode, int ea_ino);
 978 +extern int ext4_xattr_delete_inode(handle_t *handle, struct inode *inode,
 979 +                                  struct ext4_xattr_ino_array **array);
 980 +extern void ext4_xattr_inode_array_free(struct inode *inode,
 981 +                                       struct ext4_xattr_ino_array *array);
 982  extern void ext4_xattr_put_super(struct super_block *);
 983
 984  extern int ext4_expand_extra_isize_ea(struct inode *inode, int new_extra_isize,
 985 Index: linux-3.12.39-47.1/fs/ext4/ialloc.c
 986 ===================================================================
 987 --- linux-3.12.39-47.1.orig/fs/ext4/ialloc.c
 988 +++ linux-3.12.39-47.1/fs/ext4/ialloc.c
 989 @@ -252,7 +252,6 @@ void ext4_free_inode(handle_t *handle, s
 990          * as writing the quota to disk may need the lock as well.
 991          */
 992         dquot_initialize(inode);
 993 -       ext4_xattr_delete_inode(handle, inode);
 994         dquot_free_inode(inode);
 995         dquot_drop(inode);
 996
 997 Index: linux-3.12.39-47.1/fs/ext4/inline.c
 998 ===================================================================
 999 --- linux-3.12.39-47.1.orig/fs/ext4/inline.c
1000 +++ linux-3.12.39-47.1/fs/ext4/inline.c
1001 @@ -59,7 +59,7 @@ static int get_max_inline_xattr_value_si
1002
1003         /* Compute min_offs. */
1004         for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry)) {
1005 -               if (!entry->e_value_block && entry->e_value_size) {
1006 +               if (!entry->e_value_inum && entry->e_value_size) {
1007                         size_t offs = le16_to_cpu(entry->e_value_offs);
1008                         if (offs < min_offs)
1009                                 min_offs = offs;