ldiskfs/kernel_patches/patches/sles11sp2/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-stage/fs/ext4/ext4.h
   8 ===================================================================
   9 --- linux-stage.orig/fs/ext4/ext4.h
  10 +++ linux-stage/fs/ext4/ext4.h
  11 @@ -1408,6 +1408,7 @@ 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
  19 @@ -1764,6 +1765,12 @@ 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 @@ -1775,6 +1782,10 @@ 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
  43 @@ -1893,6 +1904,7 @@ 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-stage/fs/ext4/inode.c
  52 ===================================================================
  53 --- linux-stage.orig/fs/ext4/inode.c
  54 +++ linux-stage/fs/ext4/inode.c
  55 @@ -187,6 +187,8 @@ void ext4_evict_inode(struct inode *inod
  56  {
  57         handle_t *handle;
  58         int err;
  59 +       int extra_credits = 3;
  60 +       struct ext4_xattr_ino_array *lea_ino_array = NULL;
  61
  62         trace_ext4_evict_inode(inode);
  63
  64 @@ -207,7 +209,10 @@ void ext4_evict_inode(struct inode *inod
  65         if (is_bad_inode(inode))
  66                 goto no_delete;
  67
  68 -       handle = ext4_journal_start(inode, blocks_for_truncate(inode)+3);
  69 +       /*
  70 +        * Delete xattr inode before deleting the main inode.
  71 +        */
  72 +       handle = ext4_journal_start(inode, extra_credits);
  73         if (IS_ERR(handle)) {
  74                 ext4_std_error(inode->i_sb, PTR_ERR(handle));
  75                 /*
  76 @@ -218,9 +223,33 @@ void ext4_evict_inode(struct inode *inod
  77                 ext4_orphan_del(NULL, inode);
  78                 goto no_delete;
  79         }
  80 -
  81         if (IS_SYNC(inode))
  82                 ext4_handle_sync(handle);
  83 +
  84 +       err = ext4_xattr_delete_inode(handle, inode, &lea_ino_array);
  85 +       if (err) {
  86 +               ext4_warning(inode->i_sb,
  87 +                            "couldn't delete inode's xattr (err %d)", err);
  88 +               goto stop_handle;
  89 +       }
  90 +
  91 +       if (!IS_NOQUOTA(inode))
  92 +               extra_credits += 2 * EXT4_QUOTA_DEL_BLOCKS(inode->i_sb);
  93 +
  94 +       if (!ext4_handle_has_enough_credits(handle,
  95 +                               blocks_for_truncate(inode) + extra_credits)) {
  96 +               err = ext4_journal_extend(handle,
  97 +                               blocks_for_truncate(inode) + extra_credits);
  98 +               if (err > 0)
  99 +                       err = ext4_journal_restart(handle,
 100 +                               blocks_for_truncate(inode) + extra_credits);
 101 +               if (err != 0) {
 102 +                       ext4_warning(inode->i_sb,
 103 +                                    "couldn't extend journal (err %d)", err);
 104 +                       goto stop_handle;
 105 +               }
 106 +       }
 107 +
 108         inode->i_size = 0;
 109         err = ext4_mark_inode_dirty(handle, inode);
 110         if (err) {
 111 @@ -237,10 +266,10 @@ void ext4_evict_inode(struct inode *inod
 112          * enough credits left in the handle to remove the inode from
 113          * the orphan list and set the dtime field.
 114          */
 115 -       if (!ext4_handle_has_enough_credits(handle, 3)) {
 116 -               err = ext4_journal_extend(handle, 3);
 117 +       if (!ext4_handle_has_enough_credits(handle, extra_credits)) {
 118 +               err = ext4_journal_extend(handle, extra_credits);
 119                 if (err > 0)
 120 -                       err = ext4_journal_restart(handle, 3);
 121 +                       err = ext4_journal_restart(handle, extra_credits);
 122                 if (err != 0) {
 123                         ext4_warning(inode->i_sb,
 124                                      "couldn't extend journal (err %d)", err);
 125 @@ -274,7 +303,11 @@ void ext4_evict_inode(struct inode *inod
 126                 ext4_clear_inode(inode);
 127         else
 128                 ext4_free_inode(handle, inode);
 129 +
 130         ext4_journal_stop(handle);
 131 +
 132 +       if (lea_ino_array != NULL)
 133 +               ext4_xattr_inode_array_free(inode, lea_ino_array);
 134         return;
 135  no_delete:
 136         ext4_clear_inode(inode);        /* We must guarantee clearing of inode... */
 137 @@ -5552,7 +5585,7 @@ static int ext4_index_trans_blocks(struc
 138   *
 139   * Also account for superblock, inode, quota and xattr blocks
 140   */
 141 -static int ext4_meta_trans_blocks(struct inode *inode, int nrblocks, int chunk)
 142 +int ext4_meta_trans_blocks(struct inode *inode, int nrblocks, int chunk)
 143  {
 144         ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb);
 145         int gdpblocks;
 146 Index: linux-stage/fs/ext4/xattr.c
 147 ===================================================================
 148 --- linux-stage.orig/fs/ext4/xattr.c
 149 +++ linux-stage/fs/ext4/xattr.c
 150 @@ -168,19 +168,26 @@ ext4_xattr_check_block(struct buffer_hea
 151  }
 152
 153  static inline int
 154 -ext4_xattr_check_entry(struct ext4_xattr_entry *entry, size_t size)
 155 +ext4_xattr_check_entry(struct ext4_xattr_entry *entry, size_t size,
 156 +                      struct inode *inode)
 157  {
 158         size_t value_size = le32_to_cpu(entry->e_value_size);
 159
 160 -       if (entry->e_value_block != 0 || value_size > size ||
 161 -           le16_to_cpu(entry->e_value_offs) + value_size > size)
 162 +       if ((entry->e_value_inum == 0) &&
 163 +          (le16_to_cpu(entry->e_value_offs) + value_size > size))
 164 +               return -EIO;
 165 +       if (entry->e_value_inum != 0 &&
 166 +           (le32_to_cpu(entry->e_value_inum) < EXT4_FIRST_INO(inode->i_sb) ||
 167 +            le32_to_cpu(entry->e_value_inum) >
 168 +            le32_to_cpu(EXT4_SB(inode->i_sb)->s_es->s_inodes_count)))
 169                 return -EIO;
 170         return 0;
 171  }
 172
 173  static int
 174  ext4_xattr_find_entry(struct ext4_xattr_entry **pentry, int name_index,
 175 -                     const char *name, size_t size, int sorted)
 176 +                     const char *name, size_t size, int sorted,
 177 +                     struct inode *inode)
 178  {
 179         struct ext4_xattr_entry *entry;
 180         size_t name_len;
 181 @@ -200,11 +207,104 @@ ext4_xattr_find_entry(struct ext4_xattr_
 182                         break;
 183         }
 184         *pentry = entry;
 185 -       if (!cmp && ext4_xattr_check_entry(entry, size))
 186 +       if (!cmp && ext4_xattr_check_entry(entry, size, inode))
 187                         return -EIO;
 188         return cmp ? -ENODATA : 0;
 189  }
 190
 191 +/*
 192 + * Read the EA value from an inode.
 193 + */
 194 +static int
 195 +ext4_xattr_inode_read(struct inode *ea_inode, void *buf, size_t *size)
 196 +{
 197 +       unsigned long block = 0;
 198 +       struct buffer_head *bh = NULL;
 199 +       int err, blocksize;
 200 +       size_t csize, ret_size = 0;
 201 +
 202 +       if (*size == 0)
 203 +               return 0;
 204 +
 205 +       blocksize = ea_inode->i_sb->s_blocksize;
 206 +
 207 +       while (ret_size < *size) {
 208 +               csize = (*size - ret_size) > blocksize ? blocksize :
 209 +                                                       *size - ret_size;
 210 +               bh = ext4_bread(NULL, ea_inode, block, 0, &err);
 211 +               if (!bh) {
 212 +                       *size = ret_size;
 213 +                       return err;
 214 +               }
 215 +               memcpy(buf, bh->b_data, csize);
 216 +               brelse(bh);
 217 +
 218 +               buf += csize;
 219 +               block += 1;
 220 +               ret_size += csize;
 221 +       }
 222 +
 223 +       *size = ret_size;
 224 +
 225 +       return err;
 226 +}
 227 +
 228 +struct inode *ext4_xattr_inode_iget(struct inode *parent, unsigned long ea_ino, int *err)
 229 +{
 230 +       struct inode *ea_inode = NULL;
 231 +
 232 +       ea_inode = ext4_iget(parent->i_sb, ea_ino);
 233 +       if (IS_ERR(ea_inode) || is_bad_inode(ea_inode)) {
 234 +               int rc = IS_ERR(ea_inode) ? PTR_ERR(ea_inode) : 0;
 235 +               ext4_error(parent->i_sb, "error while reading EA inode %lu "
 236 +                          "/ %d %d", ea_ino, rc, is_bad_inode(ea_inode));
 237 +               *err = rc != 0 ? rc : -EIO;
 238 +               return NULL;
 239 +       }
 240 +
 241 +       if (ea_inode->i_xattr_inode_parent != parent->i_ino ||
 242 +           ea_inode->i_generation != parent->i_generation) {
 243 +               ext4_error(parent->i_sb, "Backpointer from EA inode %lu "
 244 +                          "to parent invalid.", ea_ino);
 245 +               *err = -EINVAL;
 246 +               goto error;
 247 +       }
 248 +
 249 +       if (!(EXT4_I(ea_inode)->i_flags & EXT4_EA_INODE_FL)) {
 250 +               ext4_error(parent->i_sb, "EA inode %lu does not have "
 251 +                          "EXT4_EA_INODE_FL flag set.\n", ea_ino);
 252 +               *err = -EINVAL;
 253 +               goto error;
 254 +       }
 255 +
 256 +       *err = 0;
 257 +       return ea_inode;
 258 +
 259 +error:
 260 +       iput(ea_inode);
 261 +       return NULL;
 262 +}
 263 +
 264 +/*
 265 + * Read the value from the EA inode.
 266 + */
 267 +static int
 268 +ext4_xattr_inode_get(struct inode *inode, unsigned long ea_ino, void *buffer,
 269 +                    size_t *size)
 270 +{
 271 +       struct inode *ea_inode = NULL;
 272 +       int err;
 273 +
 274 +       ea_inode = ext4_xattr_inode_iget(inode, ea_ino, &err);
 275 +       if (err)
 276 +               return err;
 277 +
 278 +       err = ext4_xattr_inode_read(ea_inode, buffer, size);
 279 +       iput(ea_inode);
 280 +
 281 +       return err;
 282 +}
 283 +
 284  static int
 285  ext4_xattr_block_get(struct inode *inode, int name_index, const char *name,
 286                      void *buffer, size_t buffer_size)
 287 @@ -235,7 +334,8 @@ bad_block:
 288         }
 289         ext4_xattr_cache_insert(bh);
 290         entry = BFIRST(bh);
 291 -       error = ext4_xattr_find_entry(&entry, name_index, name, bh->b_size, 1);
 292 +       error = ext4_xattr_find_entry(&entry, name_index, name, bh->b_size, 1,
 293 +                                     inode);
 294         if (error == -EIO)
 295                 goto bad_block;
 296         if (error)
 297 @@ -245,8 +345,16 @@ bad_block:
 298                 error = -ERANGE;
 299                 if (size > buffer_size)
 300                         goto cleanup;
 301 -               memcpy(buffer, bh->b_data + le16_to_cpu(entry->e_value_offs),
 302 -                      size);
 303 +               if (entry->e_value_inum != 0) {
 304 +                       error = ext4_xattr_inode_get(inode,
 305 +                                            le32_to_cpu(entry->e_value_inum),
 306 +                                            buffer, &size);
 307 +                       if (error)
 308 +                               goto cleanup;
 309 +               } else {
 310 +                       memcpy(buffer, bh->b_data +
 311 +                              le16_to_cpu(entry->e_value_offs), size);
 312 +               }
 313         }
 314         error = size;
 315
 316 @@ -280,7 +388,7 @@ ext4_xattr_ibody_get(struct inode *inode
 317         if (error)
 318                 goto cleanup;
 319         error = ext4_xattr_find_entry(&entry, name_index, name,
 320 -                                     end - (void *)entry, 0);
 321 +                                     end - (void *)entry, 0, inode);
 322         if (error)
 323                 goto cleanup;
 324         size = le32_to_cpu(entry->e_value_size);
 325 @@ -288,8 +396,16 @@ ext4_xattr_ibody_get(struct inode *inode
 326                 error = -ERANGE;
 327                 if (size > buffer_size)
 328                         goto cleanup;
 329 -               memcpy(buffer, (void *)IFIRST(header) +
 330 -                      le16_to_cpu(entry->e_value_offs), size);
 331 +               if (entry->e_value_inum != 0) {
 332 +                       error = ext4_xattr_inode_get(inode,
 333 +                                            le32_to_cpu(entry->e_value_inum),
 334 +                                            buffer, &size);
 335 +                       if (error)
 336 +                               goto cleanup;
 337 +               } else {
 338 +                       memcpy(buffer, (void *)IFIRST(header) +
 339 +                              le16_to_cpu(entry->e_value_offs), size);
 340 +               }
 341         }
 342         error = size;
 343
 344 @@ -514,7 +630,7 @@ static size_t ext4_xattr_free_space(stru
 345  {
 346         for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
 347                 *total += EXT4_XATTR_LEN(last->e_name_len);
 348 -               if (!last->e_value_block && last->e_value_size) {
 349 +               if (last->e_value_inum == 0 && last->e_value_size > 0) {
 350                         size_t offs = le16_to_cpu(last->e_value_offs);
 351                         if (offs < *min_offs)
 352                                 *min_offs = offs;
 353 @@ -523,11 +639,162 @@ static size_t ext4_xattr_free_space(stru
 354         return (*min_offs - ((void *)last - base) - sizeof(__u32));
 355  }
 356
 357 +/*
 358 + * Write the value of the EA in an inode.
 359 + */
 360 +static int
 361 +ext4_xattr_inode_write(handle_t *handle, struct inode *ea_inode,
 362 +                      const void *buf, int bufsize)
 363 +{
 364 +       struct buffer_head *bh = NULL;
 365 +       struct ext4_map_blocks map;
 366 +       unsigned long block = 0;
 367 +       unsigned blocksize = ea_inode->i_sb->s_blocksize;
 368 +       unsigned max_blocks = (bufsize + blocksize - 1) >> ea_inode->i_blkbits;
 369 +       int csize, wsize = 0;
 370 +       int ret = 0;
 371 +       int retries = 0;
 372 +
 373 +retry:
 374 +       while (ret >= 0 && ret < max_blocks) {
 375 +               block += ret;
 376 +               max_blocks -= ret;
 377 +
 378 +               map.m_lblk = block;
 379 +               map.m_len = max_blocks;
 380 +               ret = ext4_map_blocks(handle, ea_inode, &map,
 381 +                                     EXT4_GET_BLOCKS_CREATE);
 382 +               if (ret <= 0) {
 383 +                       ext4_mark_inode_dirty(handle, ea_inode);
 384 +                       if (ret == -ENOSPC &&
 385 +                           ext4_should_retry_alloc(ea_inode->i_sb, &retries)) {
 386 +                               ret = 0;
 387 +                               goto retry;
 388 +                       }
 389 +                       break;
 390 +               }
 391 +       }
 392 +
 393 +       if (ret < 0)
 394 +               return ret;
 395 +
 396 +       block = 0;
 397 +       while (wsize < bufsize) {
 398 +               if (bh != NULL)
 399 +                       brelse(bh);
 400 +               csize = (bufsize - wsize) > blocksize ? blocksize :
 401 +                                                               bufsize - wsize;
 402 +               bh = ext4_getblk(handle, ea_inode, block, 0, &ret);
 403 +               if (!bh)
 404 +                       goto out;
 405 +               ret = ext4_journal_get_write_access(handle, bh);
 406 +               if (ret)
 407 +                       goto out;
 408 +
 409 +               memcpy(bh->b_data, buf, csize);
 410 +               set_buffer_uptodate(bh);
 411 +               ext4_handle_dirty_metadata(handle, ea_inode, bh);
 412 +
 413 +               buf += csize;
 414 +               wsize += csize;
 415 +               block += 1;
 416 +       }
 417 +
 418 +       i_size_write(ea_inode, wsize);
 419 +       ext4_update_i_disksize(ea_inode, wsize);
 420 +
 421 +       ext4_mark_inode_dirty(handle, ea_inode);
 422 +
 423 +out:
 424 +       brelse(bh);
 425 +
 426 +       return ret;
 427 +}
 428 +
 429 +/*
 430 + * Create an inode to store the value of a large EA.
 431 + */
 432 +static struct inode *
 433 +ext4_xattr_inode_create(handle_t *handle, struct inode *inode)
 434 +{
 435 +       struct inode *ea_inode = NULL;
 436 +
 437 +       /*
 438 +        * Let the next inode be the goal, so we try and allocate the EA inode
 439 +        * in the same group, or nearby one.
 440 +        */
 441 +       ea_inode = ext4_new_inode(handle, inode->i_sb->s_root->d_inode,
 442 +                                 S_IFREG|0600, NULL, inode->i_ino + 1);
 443 +
 444 +       if (!IS_ERR(ea_inode)) {
 445 +               ea_inode->i_op = &ext4_file_inode_operations;
 446 +               ea_inode->i_fop = &ext4_file_operations;
 447 +               ext4_set_aops(ea_inode);
 448 +               ea_inode->i_generation = inode->i_generation;
 449 +               EXT4_I(ea_inode)->i_flags |= EXT4_EA_INODE_FL;
 450 +
 451 +               /*
 452 +                * A back-pointer from EA inode to parent inode will be useful
 453 +                * for e2fsck.
 454 +                */
 455 +               ea_inode->i_xattr_inode_parent = inode->i_ino;
 456 +               unlock_new_inode(ea_inode);
 457 +       }
 458 +
 459 +       return ea_inode;
 460 +}
 461 +
 462 +/*
 463 + * Unlink the inode storing the value of the EA.
 464 + */
 465 +int
 466 +ext4_xattr_inode_unlink(struct inode *inode, unsigned long ea_ino)
 467 +{
 468 +       struct inode *ea_inode = NULL;
 469 +       int err;
 470 +
 471 +       ea_inode = ext4_xattr_inode_iget(inode, ea_ino, &err);
 472 +       if (err)
 473 +               return err;
 474 +
 475 +       ea_inode->i_nlink = 0;
 476 +       iput(ea_inode);
 477 +
 478 +       return 0;
 479 +}
 480 +
 481 +/*
 482 + * Add value of the EA in an inode.
 483 + */
 484 +static int
 485 +ext4_xattr_inode_set(handle_t *handle, struct inode *inode, unsigned long *ea_ino,
 486 +                    const void *value, size_t value_len)
 487 +{
 488 +       struct inode *ea_inode = NULL;
 489 +       int err;
 490 +
 491 +       /* Create an inode for the EA value */
 492 +       ea_inode = ext4_xattr_inode_create(handle, inode);
 493 +       if (IS_ERR(ea_inode))
 494 +               return -1;
 495 +
 496 +       err = ext4_xattr_inode_write(handle, ea_inode, value, value_len);
 497 +       if (err)
 498 +               ea_inode->i_nlink = 0;
 499 +       else
 500 +               *ea_ino = ea_inode->i_ino;
 501 +
 502 +       iput(ea_inode);
 503 +
 504 +       return err;
 505 +}
 506 +
 507  struct ext4_xattr_info {
 508 -       int name_index;
 509         const char *name;
 510         const void *value;
 511         size_t value_len;
 512 +       int name_index;
 513 +       int in_inode;
 514  };
 515
 516  struct ext4_xattr_search {
 517 @@ -539,15 +806,23 @@ struct ext4_xattr_search {
 518  };
 519
 520  static int
 521 -ext4_xattr_set_entry(struct ext4_xattr_info *i, struct ext4_xattr_search *s)
 522 +ext4_xattr_set_entry(struct ext4_xattr_info *i, struct ext4_xattr_search *s,
 523 +                    handle_t *handle, struct inode *inode)
 524  {
 525         struct ext4_xattr_entry *last;
 526         size_t free, min_offs = s->end - s->base, name_len = strlen(i->name);
 527 +       int in_inode = i->in_inode;
 528 +
 529 +       if (EXT4_HAS_INCOMPAT_FEATURE(inode->i_sb,
 530 +                EXT4_FEATURE_INCOMPAT_EA_INODE) &&
 531 +           (EXT4_XATTR_SIZE(i->value_len) >
 532 +            EXT4_XATTR_MIN_LARGE_EA_SIZE(inode->i_sb->s_blocksize)))
 533 +               in_inode = 1;
 534
 535         /* Compute min_offs and last. */
 536         last = s->first;
 537         for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
 538 -               if (!last->e_value_block && last->e_value_size) {
 539 +               if (last->e_value_inum == 0 && last->e_value_size > 0) {
 540                         size_t offs = le16_to_cpu(last->e_value_offs);
 541                         if (offs < min_offs)
 542                                 min_offs = offs;
 543 @@ -555,16 +830,21 @@ ext4_xattr_set_entry(struct ext4_xattr_i
 544         }
 545         free = min_offs - ((void *)last - s->base) - sizeof(__u32);
 546         if (!s->not_found) {
 547 -               if (!s->here->e_value_block && s->here->e_value_size) {
 548 +               if (!in_inode && s->here->e_value_inum == 0 &&
 549 +                   s->here->e_value_size > 0) {
 550                         size_t size = le32_to_cpu(s->here->e_value_size);
 551                         free += EXT4_XATTR_SIZE(size);
 552                 }
 553                 free += EXT4_XATTR_LEN(name_len);
 554         }
 555         if (i->value) {
 556 -               if (free < EXT4_XATTR_SIZE(i->value_len) ||
 557 -                   free < EXT4_XATTR_LEN(name_len) +
 558 -                          EXT4_XATTR_SIZE(i->value_len))
 559 +               size_t value_len = EXT4_XATTR_SIZE(i->value_len);
 560 +
 561 +               if (in_inode)
 562 +                       value_len = 0;
 563 +
 564 +               if (free < value_len ||
 565 +                   free < EXT4_XATTR_LEN(name_len) + value_len)
 566                         return -ENOSPC;
 567         }
 568
 569 @@ -578,7 +858,8 @@ ext4_xattr_set_entry(struct ext4_xattr_i
 570                 s->here->e_name_len = name_len;
 571                 memcpy(s->here->e_name, i->name, name_len);
 572         } else {
 573 -               if (!s->here->e_value_block && s->here->e_value_size) {
 574 +               if (s->here->e_value_offs > 0 && s->here->e_value_inum == 0 &&
 575 +                   s->here->e_value_size > 0) {
 576                         void *first_val = s->base + min_offs;
 577                         size_t offs = le16_to_cpu(s->here->e_value_offs);
 578                         void *val = s->base + offs;
 579 @@ -607,13 +888,18 @@ ext4_xattr_set_entry(struct ext4_xattr_i
 580                         last = s->first;
 581                         while (!IS_LAST_ENTRY(last)) {
 582                                 size_t o = le16_to_cpu(last->e_value_offs);
 583 -                               if (!last->e_value_block &&
 584 +                               if (!last->e_value_inum &&
 585                                     last->e_value_size && o < offs)
 586                                         last->e_value_offs =
 587                                                 cpu_to_le16(o + size);
 588                                 last = EXT4_XATTR_NEXT(last);
 589                         }
 590                 }
 591 +               if (s->here->e_value_inum != 0) {
 592 +                       ext4_xattr_inode_unlink(inode,
 593 +                                       le32_to_cpu(s->here->e_value_inum));
 594 +                       s->here->e_value_inum = 0;
 595 +               }
 596                 if (!i->value) {
 597                         /* Remove the old name. */
 598                         size_t size = EXT4_XATTR_LEN(name_len);
 599 @@ -627,10 +912,17 @@ ext4_xattr_set_entry(struct ext4_xattr_i
 600         if (i->value) {
 601                 /* Insert the new value. */
 602                 s->here->e_value_size = cpu_to_le32(i->value_len);
 603 -               if (i->value_len) {
 604 +               if (in_inode) {
 605 +                       unsigned long ea_ino = le32_to_cpu(s->here->e_value_inum);
 606 +                       ext4_xattr_inode_set(handle, inode, &ea_ino, i->value,
 607 +                                            i->value_len);
 608 +                       s->here->e_value_inum = cpu_to_le32(ea_ino);
 609 +                       s->here->e_value_offs = 0;
 610 +               } else if (i->value_len) {
 611                         size_t size = EXT4_XATTR_SIZE(i->value_len);
 612                         void *val = s->base + min_offs - size;
 613                         s->here->e_value_offs = cpu_to_le16(min_offs - size);
 614 +                       s->here->e_value_inum = 0;
 615                         memset(val + size - EXT4_XATTR_PAD, 0,
 616                                EXT4_XATTR_PAD); /* Clear the pad bytes. */
 617                         memcpy(val, i->value, i->value_len);
 618 @@ -675,7 +967,7 @@ ext4_xattr_block_find(struct inode *inod
 619                 bs->s.end = bs->bh->b_data + bs->bh->b_size;
 620                 bs->s.here = bs->s.first;
 621                 error = ext4_xattr_find_entry(&bs->s.here, i->name_index,
 622 -                                             i->name, bs->bh->b_size, 1);
 623 +                                            i->name, bs->bh->b_size, 1, inode);
 624                 if (error && error != -ENODATA)
 625                         goto cleanup;
 626                 bs->s.not_found = error;
 627 @@ -699,8 +991,6 @@ ext4_xattr_block_set(handle_t *handle, s
 628
 629  #define header(x) ((struct ext4_xattr_header *)(x))
 630
 631 -       if (i->value && i->value_len > sb->s_blocksize)
 632 -               return -ENOSPC;
 633         if (s->base) {
 634                 ce = mb_cache_entry_get(ext4_xattr_cache, bs->bh->b_bdev,
 635                                         bs->bh->b_blocknr);
 636 @@ -715,7 +1005,7 @@ ext4_xattr_block_set(handle_t *handle, s
 637                                 ce = NULL;
 638                         }
 639                         ea_bdebug(bs->bh, "modifying in-place");
 640 -                       error = ext4_xattr_set_entry(i, s);
 641 +                       error = ext4_xattr_set_entry(i, s, handle, inode);
 642                         if (!error) {
 643                                 if (!IS_LAST_ENTRY(s->first))
 644                                         ext4_xattr_rehash(header(s->base),
 645 @@ -767,7 +1057,7 @@ ext4_xattr_block_set(handle_t *handle, s
 646                 s->end = s->base + sb->s_blocksize;
 647         }
 648
 649 -       error = ext4_xattr_set_entry(i, s);
 650 +       error = ext4_xattr_set_entry(i, s, handle, inode);
 651         if (error == -EIO)
 652                 goto bad_block;
 653         if (error)
 654 @@ -918,7 +1208,7 @@ ext4_xattr_ibody_find(struct inode *inod
 655                 /* Find the named attribute. */
 656                 error = ext4_xattr_find_entry(&is->s.here, i->name_index,
 657                                               i->name, is->s.end -
 658 -                                             (void *)is->s.base, 0);
 659 +                                             (void *)is->s.base, 0, inode);
 660                 if (error && error != -ENODATA)
 661                         return error;
 662                 is->s.not_found = error;
 663 @@ -937,7 +1227,7 @@ ext4_xattr_ibody_set(handle_t *handle, s
 664
 665         if (EXT4_I(inode)->i_extra_isize == 0)
 666                 return -ENOSPC;
 667 -       error = ext4_xattr_set_entry(i, s);
 668 +       error = ext4_xattr_set_entry(i, s, handle, inode);
 669         if (error)
 670                 return error;
 671         header = IHDR(inode, ext4_raw_inode(&is->iloc));
 672 @@ -973,7 +1263,7 @@ ext4_xattr_set_handle(handle_t *handle,
 673                 .name = name,
 674                 .value = value,
 675                 .value_len = value_len,
 676 -
 677 +               .in_inode = 0,
 678         };
 679         struct ext4_xattr_ibody_find is = {
 680                 .s = { .not_found = -ENODATA, },
 681 @@ -1042,6 +1332,15 @@ ext4_xattr_set_handle(handle_t *handle,
 682                                         goto cleanup;
 683                         }
 684                         error = ext4_xattr_block_set(handle, inode, &i, &bs);
 685 +                       if (EXT4_HAS_INCOMPAT_FEATURE(inode->i_sb,
 686 +                                       EXT4_FEATURE_INCOMPAT_EA_INODE) &&
 687 +                           error == -ENOSPC) {
 688 +                               /* xattr not fit to block, store at external
 689 +                                * inode */
 690 +                               i.in_inode = 1;
 691 +                               error = ext4_xattr_ibody_set(handle, inode,
 692 +                                                            &i, &is);
 693 +                       }
 694                         if (error)
 695                                 goto cleanup;
 696                         if (!is.s.not_found) {
 697 @@ -1088,10 +1387,25 @@ ext4_xattr_set(struct inode *inode, int
 698                const void *value, size_t value_len, int flags)
 699  {
 700         handle_t *handle;
 701 +       struct super_block *sb = inode->i_sb;
 702 +       int buffer_credits;
 703         int error, retries = 0;
 704
 705 +       buffer_credits = EXT4_DATA_TRANS_BLOCKS(sb);
 706 +       if ((value_len >= EXT4_XATTR_MIN_LARGE_EA_SIZE(sb->s_blocksize)) &&
 707 +           EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EA_INODE)) {
 708 +               int nrblocks = (value_len + sb->s_blocksize - 1) >>
 709 +                                       sb->s_blocksize_bits;
 710 +
 711 +               /* For new inode */
 712 +               buffer_credits += EXT4_SINGLEDATA_TRANS_BLOCKS(sb) + 3;
 713 +
 714 +               /* For data blocks of EA inode */
 715 +               buffer_credits += ext4_meta_trans_blocks(inode, nrblocks, 0);
 716 +       }
 717 +
 718  retry:
 719 -       handle = ext4_journal_start(inode, EXT4_DATA_TRANS_BLOCKS(inode->i_sb));
 720 +       handle = ext4_journal_start(inode, buffer_credits);
 721         if (IS_ERR(handle)) {
 722                 error = PTR_ERR(handle);
 723         } else {
 724 @@ -1101,7 +1415,7 @@ retry:
 725                                               value, value_len, flags);
 726                 error2 = ext4_journal_stop(handle);
 727                 if (error == -ENOSPC &&
 728 -                   ext4_should_retry_alloc(inode->i_sb, &retries))
 729 +                   ext4_should_retry_alloc(sb, &retries))
 730                         goto retry;
 731                 if (error == 0)
 732                         error = error2;
 733 @@ -1123,7 +1437,7 @@ static void ext4_xattr_shift_entries(str
 734
 735         /* Adjust the value offsets of the entries */
 736         for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
 737 -               if (!last->e_value_block && last->e_value_size) {
 738 +               if (last->e_value_inum == 0 && last->e_value_size > 0) {
 739                         new_offs = le16_to_cpu(last->e_value_offs) +
 740                                                         value_offs_shift;
 741                         BUG_ON(new_offs + le32_to_cpu(last->e_value_size)
 742 @@ -1358,22 +1672,135 @@ cleanup:
 743         return error;
 744  }
 745
 746 +#define EIA_INCR 16 /* must be 2^n */
 747 +#define EIA_MASK (EIA_INCR - 1)
 748 +/* Add the large xattr @ino into @lea_ino_array for later deletion.
 749 + * If @lea_ino_array is new or full it will be grown and the old
 750 + * contents copied over.
 751 + */
 752 +static int
 753 +ext4_expand_ino_array(struct ext4_xattr_ino_array **lea_ino_array, __u32 ino)
 754 +{
 755 +       if (*lea_ino_array == NULL) {
 756 +               /*
 757 +                * Start with 15 inodes, so it fits into a powr-of-two size.
 758 +                * If *lea_ino_array is NULL, this is essentially offsetof()
 759 +                */
 760 +               (*lea_ino_array) =
 761 +                       kmalloc(offsetof(struct ext4_xattr_ino_array,
 762 +                                        xia_inodes[EIA_MASK]),
 763 +                               GFP_NOFS);
 764 +               if (*lea_ino_array == NULL)
 765 +                       return -ENOMEM;
 766 +               (*lea_ino_array)->xia_count = 0;
 767 +       } else if (((*lea_ino_array)->xia_count & EIA_MASK) == EIA_MASK) {
 768 +               /* expand the array once all 15 + n * 16 slots are full */
 769 +               struct ext4_xattr_ino_array *new_array = NULL;
 770 +               int count = (*lea_ino_array)->xia_count;
 771 +
 772 +               /* if new_array is NULL, this is essentially offsetof() */
 773 +               new_array = kmalloc(
 774 +                               offsetof(struct ext4_xattr_ino_array,
 775 +                                        xia_inodes[count + EIA_INCR]),
 776 +                               GFP_NOFS);
 777 +               if (new_array == NULL)
 778 +                       return -ENOMEM;
 779 +               memcpy(new_array, *lea_ino_array,
 780 +                      offsetof(struct ext4_xattr_ino_array,
 781 +                               xia_inodes[count]));
 782 +               kfree(*lea_ino_array);
 783 +               *lea_ino_array = new_array;
 784 +       }
 785 +       (*lea_ino_array)->xia_inodes[(*lea_ino_array)->xia_count++] = ino;
 786 +       return 0;
 787 +}
 788 +
 789 +/**
 790 + * Add xattr inode to orphan list
 791 + */
 792 +static int
 793 +ext4_xattr_inode_orphan_add(handle_t *handle, struct inode *inode,
 794 +                       int credits, struct ext4_xattr_ino_array *lea_ino_array)
 795 +{
 796 +       struct inode *ea_inode = NULL;
 797 +       int idx = 0, error = 0;
 798 +
 799 +       if (lea_ino_array == NULL)
 800 +               return 0;
 801
 802 +       for (; idx < lea_ino_array->xia_count; ++idx) {
 803 +               if (!ext4_handle_has_enough_credits(handle, credits)) {
 804 +                       error = ext4_journal_extend(handle, credits);
 805 +                       if (error > 0)
 806 +                               error = ext4_journal_restart(handle, credits);
 807 +
 808 +                       if (error != 0) {
 809 +                               ext4_warning(inode->i_sb,
 810 +                                       "couldn't extend journal "
 811 +                                       "(err %d)", error);
 812 +                               return error;
 813 +                       }
 814 +               }
 815 +               ea_inode = ext4_xattr_inode_iget(inode,
 816 +                               lea_ino_array->xia_inodes[idx], &error);
 817 +               if (error)
 818 +                       continue;
 819 +               ext4_orphan_add(handle, ea_inode);
 820 +               /* the inode's i_count will be released by caller */
 821 +       }
 822 +
 823 +       return 0;
 824 +}
 825
 826  /*
 827   * ext4_xattr_delete_inode()
 828   *
 829 - * Free extended attribute resources associated with this inode. This
 830 + * Free extended attribute resources associated with this inode. Traverse
 831 + * all entries and unlink any xattr inodes associated with this inode. This
 832   * is called immediately before an inode is freed. We have exclusive
 833 - * access to the inode.
 834 + * access to the inode. If an orphan inode is deleted it will also delete any
 835 + * xattr block and all xattr inodes. They are checked by ext4_xattr_inode_iget()
 836 + * to ensure they belong to the parent inode and were not deleted already.
 837   */
 838 -void
 839 -ext4_xattr_delete_inode(handle_t *handle, struct inode *inode)
 840 +int
 841 +ext4_xattr_delete_inode(handle_t *handle, struct inode *inode,
 842 +                       struct ext4_xattr_ino_array **lea_ino_array)
 843  {
 844         struct buffer_head *bh = NULL;
 845 +       struct ext4_xattr_ibody_header *header;
 846 +       struct ext4_inode *raw_inode;
 847 +       struct ext4_iloc iloc;
 848 +       struct ext4_xattr_entry *entry;
 849 +       int credits = 3, error = 0;
 850
 851 -       if (!EXT4_I(inode)->i_file_acl)
 852 +       if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR))
 853 +               goto delete_external_ea;
 854 +
 855 +       error = ext4_get_inode_loc(inode, &iloc);
 856 +       if (error)
 857                 goto cleanup;
 858 +       raw_inode = ext4_raw_inode(&iloc);
 859 +       header = IHDR(inode, raw_inode);
 860 +       entry = IFIRST(header);
 861 +       for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry)) {
 862 +               if (entry->e_value_inum == 0)
 863 +                       continue;
 864 +               if (ext4_expand_ino_array(lea_ino_array,
 865 +                                         entry->e_value_inum) != 0) {
 866 +                       brelse(iloc.bh);
 867 +                       goto cleanup;
 868 +               }
 869 +               entry->e_value_inum = 0;
 870 +       }
 871 +       brelse(iloc.bh);
 872 +
 873 +delete_external_ea:
 874 +       if (!EXT4_I(inode)->i_file_acl) {
 875 +               /* add xattr inode to orphan list */
 876 +               ext4_xattr_inode_orphan_add(handle, inode, credits,
 877 +                                               *lea_ino_array);
 878 +               goto cleanup;
 879 +       }
 880         bh = sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl);
 881         if (!bh) {
 882                 EXT4_ERROR_INODE(inode, "block %llu read error",
 883 @@ -1386,11 +1813,69 @@ ext4_xattr_delete_inode(handle_t *handle
 884                                  EXT4_I(inode)->i_file_acl);
 885                 goto cleanup;
 886         }
 887 +
 888 +       entry = BFIRST(bh);
 889 +       for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry)) {
 890 +               if (entry->e_value_inum == 0)
 891 +                       continue;
 892 +               if (ext4_expand_ino_array(lea_ino_array,
 893 +                                         entry->e_value_inum) != 0)
 894 +                       goto cleanup;
 895 +               entry->e_value_inum = 0;
 896 +       }
 897 +
 898 +       /* adding xattr inode to orphan list */
 899 +       error = ext4_xattr_inode_orphan_add(handle, inode, credits,
 900 +                                       *lea_ino_array);
 901 +       if (error != 0)
 902 +               goto cleanup;
 903 +
 904 +       if (!IS_NOQUOTA(inode))
 905 +               credits += 2 * EXT4_QUOTA_DEL_BLOCKS(inode->i_sb);
 906 +
 907 +       if (!ext4_handle_has_enough_credits(handle, credits)) {
 908 +               error = ext4_journal_extend(handle, credits);
 909 +               if (error > 0)
 910 +                       error = ext4_journal_restart(handle, credits);
 911 +               if (error != 0) {
 912 +                       ext4_warning(inode->i_sb,
 913 +                               "couldn't extend journal (err %d)", error);
 914 +                       goto cleanup;
 915 +               }
 916 +       }
 917 +
 918         ext4_xattr_release_block(handle, inode, bh);
 919         EXT4_I(inode)->i_file_acl = 0;
 920
 921  cleanup:
 922         brelse(bh);
 923 +
 924 +       return error;
 925 +}
 926 +
 927 +void
 928 +ext4_xattr_inode_array_free(struct inode *inode,
 929 +                           struct ext4_xattr_ino_array *lea_ino_array)
 930 +{
 931 +       struct inode    *ea_inode = NULL;
 932 +       int             idx = 0;
 933 +       int             err;
 934 +
 935 +       if (lea_ino_array == NULL)
 936 +               return;
 937 +
 938 +       for (; idx < lea_ino_array->xia_count; ++idx) {
 939 +               ea_inode = ext4_xattr_inode_iget(inode,
 940 +                               lea_ino_array->xia_inodes[idx], &err);
 941 +               if (err)
 942 +                       continue;
 943 +               /* for inode's i_count get from ext4_xattr_delete_inode */
 944 +               if (!list_empty(&EXT4_I(ea_inode)->i_orphan))
 945 +                       iput(ea_inode);
 946 +               ea_inode->i_nlink = 0;
 947 +               iput(ea_inode);
 948 +       }
 949 +       kfree(lea_ino_array);
 950  }
 951
 952  /*
 953 @@ -1460,10 +1945,9 @@ ext4_xattr_cmp(struct ext4_xattr_header
 954                     entry1->e_name_index != entry2->e_name_index ||
 955                     entry1->e_name_len != entry2->e_name_len ||
 956                     entry1->e_value_size != entry2->e_value_size ||
 957 +                   entry1->e_value_inum != entry2->e_value_inum ||
 958                     memcmp(entry1->e_name, entry2->e_name, entry1->e_name_len))
 959                         return 1;
 960 -               if (entry1->e_value_block != 0 || entry2->e_value_block != 0)
 961 -                       return -EIO;
 962                 if (memcmp((char *)header1 + le16_to_cpu(entry1->e_value_offs),
 963                            (char *)header2 + le16_to_cpu(entry2->e_value_offs),
 964                            le32_to_cpu(entry1->e_value_size)))
 965 @@ -1547,7 +2031,7 @@ static inline void ext4_xattr_hash_entry
 966                        *name++;
 967         }
 968
 969 -       if (entry->e_value_block == 0 && entry->e_value_size != 0) {
 970 +       if (entry->e_value_inum == 0 && entry->e_value_size != 0) {
 971                 __le32 *value = (__le32 *)((char *)header +
 972                         le16_to_cpu(entry->e_value_offs));
 973                 for (n = (le32_to_cpu(entry->e_value_size) +
 974 Index: linux-stage/fs/ext4/xattr.h
 975 ===================================================================
 976 --- linux-stage.orig/fs/ext4/xattr.h
 977 +++ linux-stage/fs/ext4/xattr.h
 978 @@ -38,7 +38,7 @@ struct ext4_xattr_entry {
 979         __u8    e_name_len;     /* length of name */
 980         __u8    e_name_index;   /* attribute name index */
 981         __le16  e_value_offs;   /* offset in disk block of value */
 982 -       __le32  e_value_block;  /* disk block attribute is stored on (n/i) */
 983 +       __le32  e_value_inum;   /* inode in which the value is stored */
 984         __le32  e_value_size;   /* size of attribute value */
 985         __le32  e_hash;         /* hash value of name and value */
 986         char    e_name[0];      /* attribute name */
 987 @@ -63,6 +63,15 @@ struct ext4_xattr_entry {
 988                 EXT4_I(inode)->i_extra_isize))
 989  #define IFIRST(hdr) ((struct ext4_xattr_entry *)((hdr)+1))
 990
 991 +#define i_xattr_inode_parent i_mtime.tv_sec
 992 +
 993 +/*
 994 + * The minimum size of EA value when you start storing it in an external inode
 995 + * size of block - size of header - size of 1 entry - 4 null bytes
 996 +*/
 997 +#define EXT4_XATTR_MIN_LARGE_EA_SIZE(b)                                        \
 998 +       ((b) - EXT4_XATTR_LEN(3) - sizeof(struct ext4_xattr_header) - 4)
 999 +
1000  # ifdef CONFIG_EXT4_FS_XATTR
1001
1002  extern const struct xattr_handler ext4_xattr_user_handler;
1003 @@ -77,7 +86,13 @@ extern int ext4_xattr_get(struct inode *
1004  extern int ext4_xattr_set(struct inode *, int, const char *, const void *, size_t, int);
1005  extern int ext4_xattr_set_handle(handle_t *, struct inode *, int, const char *, const void *, size_t, int);
1006
1007 -extern void ext4_xattr_delete_inode(handle_t *, struct inode *);
1008 +extern struct inode *ext4_xattr_inode_iget(struct inode *parent, unsigned long ea_ino,
1009 +                                          int *err);
1010 +extern int ext4_xattr_inode_unlink(struct inode *inode, unsigned long ea_ino);
1011 +extern int ext4_xattr_delete_inode(handle_t *handle, struct inode *inode,
1012 +                                  struct ext4_xattr_ino_array **array);
1013 +extern void ext4_xattr_inode_array_free(struct inode *inode,
1014 +                                       struct ext4_xattr_ino_array *array);
1015  extern void ext4_xattr_put_super(struct super_block *);
1016
1017  extern int ext4_expand_extra_isize_ea(struct inode *inode, int new_extra_isize,
1018 @@ -111,9 +126,11 @@ ext4_xattr_set_handle(handle_t *handle,
1019         return -EOPNOTSUPP;
1020  }
1021
1022 -static inline void
1023 -ext4_xattr_delete_inode(handle_t *handle, struct inode *inode)
1024 +inline int
1025 +ext4_xattr_delete_inode(handle_t *handle, struct inode *inode,
1026 +                       struct ext4_xattr_ino_array *array)
1027  {
1028 +       return -EOPNOTSUPP;
1029  }
1030
1031  static inline void
1032 Index: linux-stage/fs/ext4/ialloc.c
1033 ===================================================================
1034 --- linux-stage.orig/fs/ext4/ialloc.c
1035 +++ linux-stage/fs/ext4/ialloc.c
1036 @@ -221,7 +221,6 @@ void ext4_free_inode(handle_t *handle, s
1037          * as writing the quota to disk may need the lock as well.
1038          */
1039         dquot_initialize(inode);
1040 -       ext4_xattr_delete_inode(handle, inode);
1041         dquot_free_inode(inode);
1042         dquot_drop(inode);
1043