ldiskfs/kernel_patches/patches/rhel7/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-3.10.0-123.13.2.el7.x86_64/fs/ext4/ext4.h
   8 ===================================================================
   9 --- linux-3.10.0-123.13.2.el7.x86_64.orig/fs/ext4/ext4.h
  10 +++ linux-3.10.0-123.13.2.el7.x86_64/fs/ext4/ext4.h
  11 @@ -1534,6 +1534,7 @@ 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 @@ -1934,6 +1935,12 @@ 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 @@ -1945,6 +1952,10 @@ 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 @@ -2157,6 +2168,7 @@ 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_discard_partial_page_buffers(handle_t *handle,
  51 Index: linux-3.10.0-123.13.2.el7.x86_64/fs/ext4/inode.c
  52 ===================================================================
  53 --- linux-3.10.0-123.13.2.el7.x86_64.orig/fs/ext4/inode.c
  54 +++ linux-3.10.0-123.13.2.el7.x86_64/fs/ext4/inode.c
  55 @@ -133,8 +183,6 @@ void ext4_evict_inode(struct inode *inod
  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 @@ -183,6 +183,8 @@ 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 +237,8 @@ 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 +250,33 @@ 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 +331,12 @@ 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 @@ -4841,7 +4871,7 @@ 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         int gdpblocks;
 142 Index: linux-3.10.0-123.13.2.el7.x86_64/fs/ext4/xattr.c
 143 ===================================================================
 144 --- linux-3.10.0-123.13.2.el7.x86_64.orig/fs/ext4/xattr.c
 145 +++ linux-3.10.0-123.13.2.el7.x86_64/fs/ext4/xattr.c
 146 @@ -220,19 +220,26 @@ ext4_xattr_check_block(struct inode *ino
 147  }
 148
 149  static inline int
 150 -ext4_xattr_check_entry(struct ext4_xattr_entry *entry, size_t size)
 151 +ext4_xattr_check_entry(struct ext4_xattr_entry *entry, size_t size,
 152 +                      struct inode *inode)
 153  {
 154         size_t value_size = le32_to_cpu(entry->e_value_size);
 155
 156 -       if (entry->e_value_block != 0 || value_size > size ||
 157 -           le16_to_cpu(entry->e_value_offs) + value_size > size)
 158 +       if ((entry->e_value_inum == 0) &&
 159 +          (le16_to_cpu(entry->e_value_offs) + value_size > size))
 160 +               return -EIO;
 161 +       if (entry->e_value_inum != 0 &&
 162 +           (le32_to_cpu(entry->e_value_inum) < EXT4_FIRST_INO(inode->i_sb) ||
 163 +            le32_to_cpu(entry->e_value_inum) >
 164 +            le32_to_cpu(EXT4_SB(inode->i_sb)->s_es->s_inodes_count)))
 165                 return -EIO;
 166         return 0;
 167  }
 168
 169  static int
 170  ext4_xattr_find_entry(struct ext4_xattr_entry **pentry, int name_index,
 171 -                     const char *name, size_t size, int sorted)
 172 +                     const char *name, size_t size, int sorted,
 173 +                     struct inode *inode)
 174  {
 175         struct ext4_xattr_entry *entry;
 176         size_t name_len;
 177 @@ -252,11 +259,103 @@ ext4_xattr_find_entry(struct ext4_xattr_
 178                         break;
 179         }
 180         *pentry = entry;
 181 -       if (!cmp && ext4_xattr_check_entry(entry, size))
 182 +       if (!cmp && ext4_xattr_check_entry(entry, size, inode))
 183                         return -EIO;
 184         return cmp ? -ENODATA : 0;
 185  }
 186
 187 +/*
 188 + * Read the EA value from an inode.
 189 + */
 190 +static int
 191 +ext4_xattr_inode_read(struct inode *ea_inode, void *buf, size_t *size)
 192 +{
 193 +       unsigned long block = 0;
 194 +       struct buffer_head *bh = NULL;
 195 +       int err, blocksize;
 196 +       size_t csize, ret_size = 0;
 197 +
 198 +       if (*size == 0)
 199 +               return 0;
 200 +
 201 +       blocksize = ea_inode->i_sb->s_blocksize;
 202 +
 203 +       while (ret_size < *size) {
 204 +               csize = (*size - ret_size) > blocksize ? blocksize :
 205 +                                                       *size - ret_size;
 206 +               bh = ext4_bread(NULL, ea_inode, block, 0, &err);
 207 +               if (!bh) {
 208 +                       *size = ret_size;
 209 +                       return err;
 210 +               }
 211 +               memcpy(buf, bh->b_data, csize);
 212 +               brelse(bh);
 213 +
 214 +               buf += csize;
 215 +               block += 1;
 216 +               ret_size += csize;
 217 +       }
 218 +
 219 +       *size = ret_size;
 220 +
 221 +       return err;
 222 +}
 223 +
 224 +struct inode *ext4_xattr_inode_iget(struct inode *parent, int ea_ino, int *err)
 225 +{
 226 +       struct inode *ea_inode = NULL;
 227 +
 228 +       ea_inode = ext4_iget(parent->i_sb, ea_ino);
 229 +       if (IS_ERR(ea_inode) || is_bad_inode(ea_inode)) {
 230 +               ext4_error(parent->i_sb, "error while reading EA inode %d",
 231 +                          ea_ino);
 232 +               *err = -EIO;
 233 +               return NULL;
 234 +       }
 235 +
 236 +       if (ea_inode->i_xattr_inode_parent != parent->i_ino ||
 237 +           ea_inode->i_generation != parent->i_generation) {
 238 +               ext4_error(parent->i_sb, "Backpointer from EA inode %d "
 239 +                          "to parent invalid.", ea_ino);
 240 +               *err = -EINVAL;
 241 +               goto error;
 242 +       }
 243 +
 244 +       if (!(EXT4_I(ea_inode)->i_flags & EXT4_EA_INODE_FL)) {
 245 +               ext4_error(parent->i_sb, "EA inode %d does not have "
 246 +                          "EXT4_EA_INODE_FL flag set.\n", ea_ino);
 247 +               *err = -EINVAL;
 248 +               goto error;
 249 +       }
 250 +
 251 +       *err = 0;
 252 +       return ea_inode;
 253 +
 254 +error:
 255 +       iput(ea_inode);
 256 +       return NULL;
 257 +}
 258 +
 259 +/*
 260 + * Read the value from the EA inode.
 261 + */
 262 +static int
 263 +ext4_xattr_inode_get(struct inode *inode, int ea_ino, void *buffer,
 264 +                    size_t *size)
 265 +{
 266 +       struct inode *ea_inode = NULL;
 267 +       int err;
 268 +
 269 +       ea_inode = ext4_xattr_inode_iget(inode, ea_ino, &err);
 270 +       if (err)
 271 +               return err;
 272 +
 273 +       err = ext4_xattr_inode_read(ea_inode, buffer, size);
 274 +       iput(ea_inode);
 275 +
 276 +       return err;
 277 +}
 278 +
 279  static int
 280  ext4_xattr_block_get(struct inode *inode, int name_index, const char *name,
 281                      void *buffer, size_t buffer_size)
 282 @@ -288,7 +387,8 @@ bad_block:
 283         }
 284         ext4_xattr_cache_insert(bh);
 285         entry = BFIRST(bh);
 286 -       error = ext4_xattr_find_entry(&entry, name_index, name, bh->b_size, 1);
 287 +       error = ext4_xattr_find_entry(&entry, name_index, name, bh->b_size, 1,
 288 +                                     inode);
 289         if (error == -EIO)
 290                 goto bad_block;
 291         if (error)
 292 @@ -298,8 +398,16 @@ bad_block:
 293                 error = -ERANGE;
 294                 if (size > buffer_size)
 295                         goto cleanup;
 296 -               memcpy(buffer, bh->b_data + le16_to_cpu(entry->e_value_offs),
 297 -                      size);
 298 +               if (entry->e_value_inum != 0) {
 299 +                       error = ext4_xattr_inode_get(inode,
 300 +                                            le32_to_cpu(entry->e_value_inum),
 301 +                                            buffer, &size);
 302 +                       if (error)
 303 +                               goto cleanup;
 304 +               } else {
 305 +                       memcpy(buffer, bh->b_data +
 306 +                              le16_to_cpu(entry->e_value_offs), size);
 307 +               }
 308         }
 309         error = size;
 310
 311 @@ -333,7 +441,7 @@ ext4_xattr_ibody_get(struct inode *inode
 312         if (error)
 313                 goto cleanup;
 314         error = ext4_xattr_find_entry(&entry, name_index, name,
 315 -                                     end - (void *)entry, 0);
 316 +                                     end - (void *)entry, 0, inode);
 317         if (error)
 318                 goto cleanup;
 319         size = le32_to_cpu(entry->e_value_size);
 320 @@ -341,8 +449,16 @@ ext4_xattr_ibody_get(struct inode *inode
 321                 error = -ERANGE;
 322                 if (size > buffer_size)
 323                         goto cleanup;
 324 -               memcpy(buffer, (void *)IFIRST(header) +
 325 -                      le16_to_cpu(entry->e_value_offs), size);
 326 +               if (entry->e_value_inum != 0) {
 327 +                       error = ext4_xattr_inode_get(inode,
 328 +                                            le32_to_cpu(entry->e_value_inum),
 329 +                                            buffer, &size);
 330 +                       if (error)
 331 +                               goto cleanup;
 332 +               } else {
 333 +                       memcpy(buffer, (void *)IFIRST(header) +
 334 +                              le16_to_cpu(entry->e_value_offs), size);
 335 +               }
 336         }
 337         error = size;
 338
 339 @@ -568,7 +684,7 @@ static size_t ext4_xattr_free_space(stru
 340                 *total += EXT4_XATTR_LEN(last->e_name_len);
 341  {
 342         for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
 343 -               if (!last->e_value_block && last->e_value_size) {
 344 +               if (last->e_value_inum == 0 && last->e_value_size > 0) {
 345                         size_t offs = le16_to_cpu(last->e_value_offs);
 346                         if (offs < *min_offs)
 347                                 *min_offs = offs;
 348 @@ -577,16 +693,172 @@ static size_t ext4_xattr_free_space(stru
 349         return (*min_offs - ((void *)last - base) - sizeof(__u32));
 350  }
 351
 352 +/*
 353 + * Write the value of the EA in an inode.
 354 + */
 355 +static int
 356 +ext4_xattr_inode_write(handle_t *handle, struct inode *ea_inode,
 357 +                      const void *buf, int bufsize)
 358 +{
 359 +       struct buffer_head *bh = NULL;
 360 +       unsigned long block = 0;
 361 +       unsigned blocksize = ea_inode->i_sb->s_blocksize;
 362 +       unsigned max_blocks = (bufsize + blocksize - 1) >> ea_inode->i_blkbits;
 363 +       int csize, wsize = 0;
 364 +       int ret = 0;
 365 +       int retries = 0;
 366 +
 367 +retry:
 368 +       while (ret >= 0 && ret < max_blocks) {
 369 +               struct ext4_map_blocks map;
 370 +               map.m_lblk = block += ret;
 371 +               map.m_len = max_blocks -= ret;
 372 +
 373 +               ret = ext4_map_blocks(handle, ea_inode, &map,
 374 +                                     EXT4_GET_BLOCKS_CREATE);
 375 +               if (ret <= 0) {
 376 +                       ext4_mark_inode_dirty(handle, ea_inode);
 377 +                       if (ret == -ENOSPC &&
 378 +                           ext4_should_retry_alloc(ea_inode->i_sb, &retries)) {
 379 +                               ret = 0;
 380 +                               goto retry;
 381 +                       }
 382 +                       break;
 383 +               }
 384 +       }
 385 +
 386 +       if (ret < 0)
 387 +               return ret;
 388 +
 389 +       block = 0;
 390 +       while (wsize < bufsize) {
 391 +               if (bh != NULL)
 392 +                       brelse(bh);
 393 +               csize = (bufsize - wsize) > blocksize ? blocksize :
 394 +                                                               bufsize - wsize;
 395 +               bh = ext4_getblk(handle, ea_inode, block, 0, &ret);
 396 +               if (!bh)
 397 +                       goto out;
 398 +               ret = ext4_journal_get_write_access(handle, bh);
 399 +               if (ret)
 400 +                       goto out;
 401 +
 402 +               memcpy(bh->b_data, buf, csize);
 403 +               set_buffer_uptodate(bh);
 404 +               ext4_handle_dirty_metadata(handle, ea_inode, bh);
 405 +
 406 +               buf += csize;
 407 +               wsize += csize;
 408 +               block += 1;
 409 +       }
 410 +
 411 +       i_size_write(ea_inode, wsize);
 412 +       ext4_update_i_disksize(ea_inode, wsize);
 413 +
 414 +       ext4_mark_inode_dirty(handle, ea_inode);
 415 +
 416 +out:
 417 +       brelse(bh);
 418 +
 419 +       return ret;
 420 +}
 421 +
 422 +/*
 423 + * Create an inode to store the value of a large EA.
 424 + */
 425 +static struct inode *
 426 +ext4_xattr_inode_create(handle_t *handle, struct inode *inode)
 427 +{
 428 +       struct inode *ea_inode = NULL;
 429 +
 430 +       /*
 431 +        * Let the next inode be the goal, so we try and allocate the EA inode
 432 +        * in the same group, or nearby one.
 433 +        */
 434 +       ea_inode = ext4_new_inode(handle, inode->i_sb->s_root->d_inode,
 435 +                                 S_IFREG|0600, NULL, inode->i_ino + 1, NULL);
 436 +
 437 +       if (!IS_ERR(ea_inode)) {
 438 +               ea_inode->i_op = &ext4_file_inode_operations;
 439 +               ea_inode->i_fop = &ext4_file_operations;
 440 +               ext4_set_aops(ea_inode);
 441 +               ea_inode->i_generation = inode->i_generation;
 442 +               EXT4_I(ea_inode)->i_flags |= EXT4_EA_INODE_FL;
 443 +
 444 +               /*
 445 +                * A back-pointer from EA inode to parent inode will be useful
 446 +                * for e2fsck.
 447 +                */
 448 +               ea_inode->i_xattr_inode_parent = inode->i_ino;
 449 +               unlock_new_inode(ea_inode);
 450 +       }
 451 +
 452 +       return ea_inode;
 453 +}
 454 +
 455 +/*
 456 + * Unlink the inode storing the value of the EA.
 457 + */
 458 +int
 459 +ext4_xattr_inode_unlink(struct inode *inode, int ea_ino)
 460 +{
 461 +       struct inode *ea_inode = NULL;
 462 +       int err;
 463 +
 464 +       ea_inode = ext4_xattr_inode_iget(inode, ea_ino, &err);
 465 +       if (err)
 466 +               return err;
 467 +
 468 +       clear_nlink(ea_inode);
 469 +       iput(ea_inode);
 470 +
 471 +       return 0;
 472 +}
 473 +
 474 +/*
 475 + * Add value of the EA in an inode.
 476 + */
 477  static int
 478 -ext4_xattr_set_entry(struct ext4_xattr_info *i, struct ext4_xattr_search *s)
 479 +ext4_xattr_inode_set(handle_t *handle, struct inode *inode, int *ea_ino,
 480 +                    const void *value, size_t value_len)
 481 +{
 482 +       struct inode *ea_inode = NULL;
 483 +       int err;
 484 +
 485 +       /* Create an inode for the EA value */
 486 +       ea_inode = ext4_xattr_inode_create(handle, inode);
 487 +       if (IS_ERR(ea_inode))
 488 +               return -1;
 489 +
 490 +       err = ext4_xattr_inode_write(handle, ea_inode, value, value_len);
 491 +       if (err)
 492 +               clear_nlink(ea_inode);
 493 +       else
 494 +               *ea_ino = ea_inode->i_ino;
 495 +
 496 +       iput(ea_inode);
 497 +
 498 +       return err;
 499 +}
 500 +
 501 +static int
 502 +ext4_xattr_set_entry(struct ext4_xattr_info *i, struct ext4_xattr_search *s,
 503 +                    handle_t *handle, struct inode *inode)
 504  {
 505         struct ext4_xattr_entry *last;
 506         size_t free, min_offs = s->end - s->base, name_len = strlen(i->name);
 507 +       int in_inode = i->in_inode;
 508 +
 509 +       if (EXT4_HAS_INCOMPAT_FEATURE(inode->i_sb,
 510 +                EXT4_FEATURE_INCOMPAT_EA_INODE) &&
 511 +           (EXT4_XATTR_SIZE(i->value_len) >
 512 +            EXT4_XATTR_MIN_LARGE_EA_SIZE(inode->i_sb->s_blocksize)))
 513 +               in_inode = 1;
 514
 515         /* Compute min_offs and last. */
 516         last = s->first;
 517         for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
 518 -               if (!last->e_value_block && last->e_value_size) {
 519 +               if (last->e_value_inum == 0 && last->e_value_size > 0) {
 520                         size_t offs = le16_to_cpu(last->e_value_offs);
 521                         if (offs < min_offs)
 522                                 min_offs = offs;
 523 @@ -594,16 +865,21 @@ ext4_xattr_set_entry(struct ext4_xattr_i
 524         }
 525         free = min_offs - ((void *)last - s->base) - sizeof(__u32);
 526         if (!s->not_found) {
 527 -               if (!s->here->e_value_block && s->here->e_value_size) {
 528 +               if (!in_inode && s->here->e_value_inum == 0 &&
 529 +                   s->here->e_value_size > 0) {
 530                         size_t size = le32_to_cpu(s->here->e_value_size);
 531                         free += EXT4_XATTR_SIZE(size);
 532                 }
 533                 free += EXT4_XATTR_LEN(name_len);
 534         }
 535         if (i->value) {
 536 -               if (free < EXT4_XATTR_SIZE(i->value_len) ||
 537 -                   free < EXT4_XATTR_LEN(name_len) +
 538 -                          EXT4_XATTR_SIZE(i->value_len))
 539 +               size_t value_len = EXT4_XATTR_SIZE(i->value_len);
 540 +
 541 +               if (in_inode)
 542 +                       value_len = 0;
 543 +
 544 +               if (free < value_len ||
 545 +                   free < EXT4_XATTR_LEN(name_len) + value_len)
 546                         return -ENOSPC;
 547         }
 548
 549 @@ -617,7 +893,8 @@ ext4_xattr_set_entry(struct ext4_xattr_i
 550                 s->here->e_name_len = name_len;
 551                 memcpy(s->here->e_name, i->name, name_len);
 552         } else {
 553 -               if (!s->here->e_value_block && s->here->e_value_size) {
 554 +               if (s->here->e_value_offs > 0 && s->here->e_value_inum == 0 &&
 555 +                   s->here->e_value_size > 0) {
 556                         void *first_val = s->base + min_offs;
 557                         size_t offs = le16_to_cpu(s->here->e_value_offs);
 558                         void *val = s->base + offs;
 559 @@ -651,13 +928,17 @@ ext4_xattr_set_entry(struct ext4_xattr_i
 560                         last = s->first;
 561                         while (!IS_LAST_ENTRY(last)) {
 562                                 size_t o = le16_to_cpu(last->e_value_offs);
 563 -                               if (!last->e_value_block &&
 564 -                                   last->e_value_size && o < offs)
 565 +                               if (last->e_value_size > 0 && o < offs)
 566                                         last->e_value_offs =
 567                                                 cpu_to_le16(o + size);
 568                                 last = EXT4_XATTR_NEXT(last);
 569                         }
 570                 }
 571 +               if (s->here->e_value_inum != 0) {
 572 +                       ext4_xattr_inode_unlink(inode,
 573 +                                       le32_to_cpu(s->here->e_value_inum));
 574 +                       s->here->e_value_inum = 0;
 575 +               }
 576                 if (!i->value) {
 577                         /* Remove the old name. */
 578                         size_t size = EXT4_XATTR_LEN(name_len);
 579 @@ -671,10 +952,17 @@ ext4_xattr_set_entry(struct ext4_xattr_i
 580         if (i->value) {
 581                 /* Insert the new value. */
 582                 s->here->e_value_size = cpu_to_le32(i->value_len);
 583 -               if (i->value_len) {
 584 +               if (in_inode) {
 585 +                       int ea_ino = le32_to_cpu(s->here->e_value_inum);
 586 +                       ext4_xattr_inode_set(handle, inode, &ea_ino, i->value,
 587 +                                            i->value_len);
 588 +                       s->here->e_value_inum = cpu_to_le32(ea_ino);
 589 +                       s->here->e_value_offs = 0;
 590 +               } else if (i->value_len) {
 591                         size_t size = EXT4_XATTR_SIZE(i->value_len);
 592                         void *val = s->base + min_offs - size;
 593                         s->here->e_value_offs = cpu_to_le16(min_offs - size);
 594 +                       s->here->e_value_inum = 0;
 595                         if (i->value == EXT4_ZERO_XATTR_VALUE) {
 596                                 memset(val, 0, size);
 597                         } else {
 598 @@ -724,7 +1012,7 @@ ext4_xattr_block_find(struct inode *inod
 599                 bs->s.end = bs->bh->b_data + bs->bh->b_size;
 600                 bs->s.here = bs->s.first;
 601                 error = ext4_xattr_find_entry(&bs->s.here, i->name_index,
 602 -                                             i->name, bs->bh->b_size, 1);
 603 +                                            i->name, bs->bh->b_size, 1, inode);
 604                 if (error && error != -ENODATA)
 605                         goto cleanup;
 606                 bs->s.not_found = error;
 607 @@ -748,8 +1036,6 @@ ext4_xattr_block_set(handle_t *handle, s
 608
 609  #define header(x) ((struct ext4_xattr_header *)(x))
 610
 611 -       if (i->value && i->value_len > sb->s_blocksize)
 612 -               return -ENOSPC;
 613         if (s->base) {
 614                 ce = mb_cache_entry_get(ext4_xattr_cache, bs->bh->b_bdev,
 615                                         bs->bh->b_blocknr);
 616 @@ -764,7 +1050,7 @@ ext4_xattr_block_set(handle_t *handle, s
 617                                 ce = NULL;
 618                         }
 619                         ea_bdebug(bs->bh, "modifying in-place");
 620 -                       error = ext4_xattr_set_entry(i, s);
 621 +                       error = ext4_xattr_set_entry(i, s, handle, inode);
 622                         if (!error) {
 623                                 if (!IS_LAST_ENTRY(s->first))
 624                                         ext4_xattr_rehash(header(s->base),
 625 @@ -815,7 +1101,7 @@ ext4_xattr_block_set(handle_t *handle, s
 626                 s->end = s->base + sb->s_blocksize;
 627         }
 628
 629 -       error = ext4_xattr_set_entry(i, s);
 630 +       error = ext4_xattr_set_entry(i, s, handle, inode);
 631         if (error == -EIO)
 632                 goto bad_block;
 633         if (error)
 634 @@ -963,7 +1249,7 @@ int ext4_xattr_ibody_find(struct inode *
 635                 /* Find the named attribute. */
 636                 error = ext4_xattr_find_entry(&is->s.here, i->name_index,
 637                                               i->name, is->s.end -
 638 -                                             (void *)is->s.base, 0);
 639 +                                             (void *)is->s.base, 0, inode);
 640                 if (error && error != -ENODATA)
 641                         return error;
 642                 is->s.not_found = error;
 643 @@ -981,7 +1267,7 @@ int ext4_xattr_ibody_inline_set(handle_t
 644
 645         if (EXT4_I(inode)->i_extra_isize == 0)
 646                 return -ENOSPC;
 647 -       error = ext4_xattr_set_entry(i, s);
 648 +       error = ext4_xattr_set_entry(i, s, handle, inode);
 649         if (error) {
 650                 if (error == -ENOSPC &&
 651                     ext4_has_inline_data(inode)) {
 652 @@ -993,7 +1279,7 @@ int ext4_xattr_ibody_inline_set(handle_t
 653                         error = ext4_xattr_ibody_find(inode, i, is);
 654                         if (error)
 655                                 return error;
 656 -                       error = ext4_xattr_set_entry(i, s);
 657 +                       error = ext4_xattr_set_entry(i, s, handle, inode);
 658                 }
 659                 if (error)
 660                         return error;
 661 @@ -1019,7 +1305,7 @@ static int ext4_xattr_ibody_set(handle_t
 662
 663         if (EXT4_I(inode)->i_extra_isize == 0)
 664                 return -ENOSPC;
 665 -       error = ext4_xattr_set_entry(i, s);
 666 +       error = ext4_xattr_set_entry(i, s, handle, inode);
 667         if (error)
 668                 return error;
 669         header = IHDR(inode, ext4_raw_inode(&is->iloc));
 670 @@ -1055,7 +1341,7 @@ ext4_xattr_set_handle(handle_t *handle,
 671                 .name = name,
 672                 .value = value,
 673                 .value_len = value_len,
 674 -
 675 +               .in_inode = 0,
 676         };
 677         struct ext4_xattr_ibody_find is = {
 678                 .s = { .not_found = -ENODATA, },
 679 @@ -1120,6 +1406,15 @@ ext4_xattr_set_handle(handle_t *handle,
 680                                         goto cleanup;
 681                         }
 682                         error = ext4_xattr_block_set(handle, inode, &i, &bs);
 683 +                       if (EXT4_HAS_INCOMPAT_FEATURE(inode->i_sb,
 684 +                                       EXT4_FEATURE_INCOMPAT_EA_INODE) &&
 685 +                           error == -ENOSPC) {
 686 +                               /* xattr not fit to block, store at external
 687 +                                * inode */
 688 +                               i.in_inode = 1;
 689 +                               error = ext4_xattr_ibody_set(handle, inode,
 690 +                                                            &i, &is);
 691 +                       }
 692                         if (error)
 693                                 goto cleanup;
 694                         if (!is.s.not_found) {
 695 @@ -1166,9 +1461,22 @@ ext4_xattr_set(struct inode *inode, int
 696                const void *value, size_t value_len, int flags)
 697  {
 698         handle_t *handle;
 699 +       struct super_block *sb = inode->i_sb;
 700         int error, retries = 0;
 701         int credits = ext4_jbd2_credits_xattr(inode);
 702
 703 +       if ((value_len >= EXT4_XATTR_MIN_LARGE_EA_SIZE(sb->s_blocksize)) &&
 704 +           EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EA_INODE)) {
 705 +               int nrblocks = (value_len + sb->s_blocksize - 1) >>
 706 +                                       sb->s_blocksize_bits;
 707 +
 708 +               /* For new inode */
 709 +               credits += EXT4_SINGLEDATA_TRANS_BLOCKS(sb) + 3;
 710 +
 711 +               /* For data blocks of EA inode */
 712 +               credits += ext4_meta_trans_blocks(inode, nrblocks, 0);
 713 +       }
 714 +
 715  retry:
 716         handle = ext4_journal_start(inode, EXT4_HT_XATTR, credits);
 717         if (IS_ERR(handle)) {
 718 @@ -1180,7 +1488,7 @@ retry:
 719                                               value, value_len, flags);
 720                 error2 = ext4_journal_stop(handle);
 721                 if (error == -ENOSPC &&
 722 -                   ext4_should_retry_alloc(inode->i_sb, &retries))
 723 +                   ext4_should_retry_alloc(sb, &retries))
 724                         goto retry;
 725                 if (error == 0)
 726                         error = error2;
 727 @@ -1202,7 +1510,7 @@ static void ext4_xattr_shift_entries(str
 728
 729         /* Adjust the value offsets of the entries */
 730         for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
 731 -               if (!last->e_value_block && last->e_value_size) {
 732 +               if (last->e_value_inum == 0 && last->e_value_size > 0) {
 733                         new_offs = le16_to_cpu(last->e_value_offs) +
 734                                                         value_offs_shift;
 735                         BUG_ON(new_offs + le32_to_cpu(last->e_value_size)
 736 @@ -1441,19 +1749,91 @@ cleanup:
 737  }
 738
 739
 740 -
 741 +#define EIA_INCR 16 /* must be 2^n */
 742 +#define EIA_MASK (EIA_INCR - 1)
 743 +/* Add the large xattr @ino into @lea_ino_array for later deletion.
 744 + * If @lea_ino_array is new or full it will be grown and the old
 745 + * contents copied over.
 746 + */
 747 +static int
 748 +ext4_expand_ino_array(struct ext4_xattr_ino_array **lea_ino_array, __u32 ino)
 749 +{
 750 +       if (*lea_ino_array == NULL) {
 751 +               /*
 752 +                * Start with 15 inodes, so it fits into a power-of-two size.
 753 +                * If *lea_ino_array is NULL, this is essentially offsetof()
 754 +                */
 755 +               (*lea_ino_array) =
 756 +                       kmalloc(offsetof(struct ext4_xattr_ino_array,
 757 +                                        xia_inodes[EIA_MASK]),
 758 +                               GFP_NOFS);
 759 +               if (*lea_ino_array == NULL)
 760 +                       return -ENOMEM;
 761 +               (*lea_ino_array)->xia_count = 0;
 762 +       } else if (((*lea_ino_array)->xia_count & EIA_MASK) == EIA_MASK) {
 763 +               /* expand the array once all 15 + n * 16 slots are full */
 764 +               struct ext4_xattr_ino_array *new_array = NULL;
 765 +               int count = (*lea_ino_array)->xia_count;
 766 +
 767 +               /* if new_array is NULL, this is essentially offsetof() */
 768 +               new_array = kmalloc(
 769 +                               offsetof(struct ext4_xattr_ino_array,
 770 +                                        xia_inodes[count + EIA_INCR]),
 771 +                               GFP_NOFS);
 772 +               if (new_array == NULL)
 773 +                       return -ENOMEM;
 774 +               memcpy(new_array, *lea_ino_array,
 775 +                      offsetof(struct ext4_xattr_ino_array,
 776 +                               xia_inodes[count]));
 777 +               kfree(*lea_ino_array);
 778 +               *lea_ino_array = new_array;
 779 +       }
 780 +       (*lea_ino_array)->xia_inodes[(*lea_ino_array)->xia_count++] = ino;
 781 +       return 0;
 782 +}
 783  /*
 784   * ext4_xattr_delete_inode()
 785   *
 786 - * Free extended attribute resources associated with this inode. This
 787 + * Free extended attribute resources associated with this inode. Traverse
 788 + * all entries and unlink any xattr inodes associated with this inode. This
 789   * is called immediately before an inode is freed. We have exclusive
 790 - * access to the inode.
 791 + * access to the inode. If an orphan inode is deleted it will also delete any
 792 + * xattr block and all xattr inodes. They are checked by ext4_xattr_inode_iget()
 793 + * to ensure they belong to the parent inode and were not deleted already.
 794   */
 795 -void
 796 -ext4_xattr_delete_inode(handle_t *handle, struct inode *inode)
 797 +int
 798 +ext4_xattr_delete_inode(handle_t *handle, struct inode *inode,
 799 +                       struct ext4_xattr_ino_array **lea_ino_array)
 800  {
 801         struct buffer_head *bh = NULL;
 802 +       struct ext4_xattr_ibody_header *header;
 803 +       struct ext4_inode *raw_inode;
 804 +       struct ext4_iloc iloc;
 805 +       struct ext4_xattr_entry *entry;
 806 +       int error = 0;
 807 +
 808 +       if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR))
 809 +               goto delete_external_ea;
 810 +
 811 +       error = ext4_get_inode_loc(inode, &iloc);
 812 +       if (error)
 813 +               goto cleanup;
 814 +       raw_inode = ext4_raw_inode(&iloc);
 815 +       header = IHDR(inode, raw_inode);
 816 +       entry = IFIRST(header);
 817 +       for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry)) {
 818 +               if (entry->e_value_inum == 0)
 819 +                       continue;
 820 +               if (ext4_expand_ino_array(lea_ino_array,
 821 +                                         entry->e_value_inum) != 0) {
 822 +                       brelse(iloc.bh);
 823 +                       goto cleanup;
 824 +               }
 825 +               entry->e_value_inum = 0;
 826 +       }
 827 +       brelse(iloc.bh);
 828
 829 +delete_external_ea:
 830         if (!EXT4_I(inode)->i_file_acl)
 831                 goto cleanup;
 832         bh = sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl);
 833 @@ -1468,11 +1848,74 @@ ext4_xattr_delete_inode(handle_t *handle
 834                                  EXT4_I(inode)->i_file_acl);
 835                 goto cleanup;
 836         }
 837 +
 838 +       entry = BFIRST(bh);
 839 +       for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry)) {
 840 +               if (entry->e_value_inum == 0)
 841 +                       continue;
 842 +               if (ext4_expand_ino_array(lea_ino_array,
 843 +                                         entry->e_value_inum) != 0)
 844 +                       goto cleanup;
 845 +               entry->e_value_inum = 0;
 846 +       }
 847 +
 848 +       /* add xattr inode to orphan list */
 849 +       if (*lea_ino_array != NULL) {
 850 +               struct inode *ea_inode = NULL;
 851 +               int idx = 0;
 852 +
 853 +               for (; idx < (*lea_ino_array)->xia_count; ++idx) {
 854 +                       if (!ext4_handle_has_enough_credits(handle, 3)) {
 855 +                               error = ext4_journal_extend(handle, 3);
 856 +                               if (error > 0)
 857 +                                       error = ext4_journal_restart(handle, 3);
 858 +                               if (error != 0) {
 859 +                                       ext4_warning(inode->i_sb,
 860 +                                               "couldn't extend journal "
 861 +                                               "(err %d)", error);
 862 +                                       goto cleanup;
 863 +                               }
 864 +                       }
 865 +                       ea_inode = ext4_xattr_inode_iget(inode,
 866 +                               (*lea_ino_array)->xia_inodes[idx], &error);
 867 +                       if (error)
 868 +                               continue;
 869 +                       ext4_orphan_add(handle, ea_inode);
 870 +                       /* the inode's i_count will be released by caller */
 871 +               }
 872 +       }
 873 +
 874         ext4_xattr_release_block(handle, inode, bh);
 875         EXT4_I(inode)->i_file_acl = 0;
 876
 877  cleanup:
 878         brelse(bh);
 879 +
 880 +       return error;
 881 +}
 882 +
 883 +void
 884 +ext4_xattr_inode_array_free(struct inode *inode,
 885 +                           struct ext4_xattr_ino_array *lea_ino_array)
 886 +{
 887 +       struct inode    *ea_inode = NULL;
 888 +       int             idx = 0;
 889 +       int             err;
 890 +
 891 +       if (lea_ino_array == NULL)
 892 +               return;
 893 +
 894 +       for (; idx < lea_ino_array->xia_count; ++idx) {
 895 +               ea_inode = ext4_xattr_inode_iget(inode,
 896 +                               lea_ino_array->xia_inodes[idx], &err);
 897 +               if (err)
 898 +                       continue;
 899 +               clear_nlink(ea_inode);
 900 +               iput(ea_inode);
 901 +               /* for inode's i_count get from ext4_xattr_delete_inode */
 902 +               iput(ea_inode);
 903 +       }
 904 +       kfree(lea_ino_array);
 905  }
 906
 907  /*
 908 @@ -1542,10 +1985,9 @@ ext4_xattr_cmp(struct ext4_xattr_header
 909                     entry1->e_name_index != entry2->e_name_index ||
 910                     entry1->e_name_len != entry2->e_name_len ||
 911                     entry1->e_value_size != entry2->e_value_size ||
 912 +                   entry1->e_value_inum != entry2->e_value_inum ||
 913                     memcmp(entry1->e_name, entry2->e_name, entry1->e_name_len))
 914                         return 1;
 915 -               if (entry1->e_value_block != 0 || entry2->e_value_block != 0)
 916 -                       return -EIO;
 917                 if (memcmp((char *)header1 + le16_to_cpu(entry1->e_value_offs),
 918                            (char *)header2 + le16_to_cpu(entry2->e_value_offs),
 919                            le32_to_cpu(entry1->e_value_size)))
 920 @@ -1629,7 +2071,7 @@ static inline void ext4_xattr_hash_entry
 921                        *name++;
 922         }
 923
 924 -       if (entry->e_value_block == 0 && entry->e_value_size != 0) {
 925 +       if (entry->e_value_inum == 0 && entry->e_value_size != 0) {
 926                 __le32 *value = (__le32 *)((char *)header +
 927                         le16_to_cpu(entry->e_value_offs));
 928                 for (n = (le32_to_cpu(entry->e_value_size) +
 929 Index: linux-3.10.0-123.13.2.el7.x86_64/fs/ext4/xattr.h
 930 ===================================================================
 931 --- linux-3.10.0-123.13.2.el7.x86_64.orig/fs/ext4/xattr.h
 932 +++ linux-3.10.0-123.13.2.el7.x86_64/fs/ext4/xattr.h
 933 @@ -42,7 +42,7 @@ struct ext4_xattr_entry {
 934         __u8    e_name_len;     /* length of name */
 935         __u8    e_name_index;   /* attribute name index */
 936         __le16  e_value_offs;   /* offset in disk block of value */
 937 -       __le32  e_value_block;  /* disk block attribute is stored on (n/i) */
 938 +       __le32  e_value_inum;   /* inode in which the value is stored */
 939         __le32  e_value_size;   /* size of attribute value */
 940         __le32  e_hash;         /* hash value of name and value */
 941         char    e_name[0];      /* attribute name */
 942 @@ -67,6 +67,15 @@ struct ext4_xattr_entry {
 943                 EXT4_I(inode)->i_extra_isize))
 944  #define IFIRST(hdr) ((struct ext4_xattr_entry *)((hdr)+1))
 945
 946 +#define i_xattr_inode_parent i_mtime.tv_sec
 947 +
 948 +/*
 949 + * The minimum size of EA value when you start storing it in an external inode
 950 + * size of block - size of header - size of 1 entry - 4 null bytes
 951 +*/
 952 +#define EXT4_XATTR_MIN_LARGE_EA_SIZE(b)                                        \
 953 +       ((b) - EXT4_XATTR_LEN(3) - sizeof(struct ext4_xattr_header) - 4)
 954 +
 955  #define BHDR(bh) ((struct ext4_xattr_header *)((bh)->b_data))
 956  #define ENTRY(ptr) ((struct ext4_xattr_entry *)(ptr))
 957  #define BFIRST(bh) ENTRY(BHDR(bh)+1)
 958 @@ -75,10 +84,11 @@ struct ext4_xattr_entry {
 959  #define EXT4_ZERO_XATTR_VALUE ((void *)-1)
 960
 961  struct ext4_xattr_info {
 962 -       int name_index;
 963         const char *name;
 964         const void *value;
 965         size_t value_len;
 966 +       int name_index;
 967 +       int in_inode;
 968  };
 969
 970  struct ext4_xattr_search {
 971 @@ -106,7 +116,13 @@ extern int ext4_xattr_get(struct inode *
 972  extern int ext4_xattr_set(struct inode *, int, const char *, const void *, size_t, int);
 973  extern int ext4_xattr_set_handle(handle_t *, struct inode *, int, const char *, const void *, size_t, int);
 974
 975 -extern void ext4_xattr_delete_inode(handle_t *, struct inode *);
 976 +extern struct inode *ext4_xattr_inode_iget(struct inode *parent, int ea_ino,
 977 +                                          int *err);
 978 +extern int ext4_xattr_inode_unlink(struct inode *inode, int ea_ino);
 979 +extern int ext4_xattr_delete_inode(handle_t *handle, struct inode *inode,
 980 +                                  struct ext4_xattr_ino_array **array);
 981 +extern void ext4_xattr_inode_array_free(struct inode *inode,
 982 +                                       struct ext4_xattr_ino_array *array);
 983  extern void ext4_xattr_put_super(struct super_block *);
 984
 985  extern int ext4_expand_extra_isize_ea(struct inode *inode, int new_extra_isize,
 986 Index: linux-3.10.0-123.13.2.el7.x86_64/fs/ext4/ialloc.c
 987 ===================================================================
 988 --- linux-3.10.0-123.13.2.el7.x86_64.orig/fs/ext4/ialloc.c
 989 +++ linux-3.10.0-123.13.2.el7.x86_64/fs/ext4/ialloc.c
 990 @@ -250,7 +250,6 @@ void ext4_free_inode(handle_t *handle, s
 991          * as writing the quota to disk may need the lock as well.
 992          */
 993         dquot_initialize(inode);
 994 -       ext4_xattr_delete_inode(handle, inode);
 995         dquot_free_inode(inode);
 996         dquot_drop(inode);
 997
 998 Index: linux-3.10.0-123.13.2.el7.x86_64/fs/ext4/inline.c
 999 ===================================================================
1000 --- linux-3.10.0-123.13.2.el7.x86_64.orig/fs/ext4/inline.c
1001 +++ linux-3.10.0-123.13.2.el7.x86_64/fs/ext4/inline.c
1002 @@ -59,7 +59,7 @@ static int get_max_inline_xattr_value_si
1003
1004         /* Compute min_offs. */
1005         for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry)) {
1006 -               if (!entry->e_value_block && entry->e_value_size) {
1007 +               if (!entry->e_value_inum && entry->e_value_size) {
1008                         size_t offs = le16_to_cpu(entry->e_value_offs);
1009                         if (offs < min_offs)
1010                                 min_offs = offs;