fs/ext4/inline.c

   1 // SPDX-License-Identifier: LGPL-2.1
   2 /*
   3  * Copyright (c) 2012 Taobao.
   4  * Written by Tao Ma <[email protected]>
   5  */
   6
   7 #include <linux/iomap.h>
   8 #include <linux/fiemap.h>
   9 #include <linux/namei.h>
  10 #include <linux/iversion.h>
  11 #include <linux/sched/mm.h>
  12
  13 #include "ext4_jbd2.h"
  14 #include "ext4.h"
  15 #include "xattr.h"
  16 #include "truncate.h"
  17
  18 #define EXT4_XATTR_SYSTEM_DATA  "data"
  19 #define EXT4_MIN_INLINE_DATA_SIZE       ((sizeof(__le32) * EXT4_N_BLOCKS))
  20 #define EXT4_INLINE_DOTDOT_OFFSET       2
  21 #define EXT4_INLINE_DOTDOT_SIZE         4
  22
  23 static int ext4_get_inline_size(struct inode *inode)
  24 {
  25         if (EXT4_I(inode)->i_inline_off)
  26                 return EXT4_I(inode)->i_inline_size;
  27
  28         return 0;
  29 }
  30
  31 static int get_max_inline_xattr_value_size(struct inode *inode,
  32                                            struct ext4_iloc *iloc)
  33 {
  34         struct ext4_xattr_ibody_header *header;
  35         struct ext4_xattr_entry *entry;
  36         struct ext4_inode *raw_inode;
  37         int free, min_offs;
  38
  39         if (!EXT4_INODE_HAS_XATTR_SPACE(inode))
  40                 return 0;
  41
  42         min_offs = EXT4_SB(inode->i_sb)->s_inode_size -
  43                         EXT4_GOOD_OLD_INODE_SIZE -
  44                         EXT4_I(inode)->i_extra_isize -
  45                         sizeof(struct ext4_xattr_ibody_header);
  46
  47         /*
  48          * We need to subtract another sizeof(__u32) since an in-inode xattr
  49          * needs an empty 4 bytes to indicate the gap between the xattr entry
  50          * and the name/value pair.
  51          */
  52         if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR))
  53                 return EXT4_XATTR_SIZE(min_offs -
  54                         EXT4_XATTR_LEN(strlen(EXT4_XATTR_SYSTEM_DATA)) -
  55                         EXT4_XATTR_ROUND - sizeof(__u32));
  56
  57         raw_inode = ext4_raw_inode(iloc);
  58         header = IHDR(inode, raw_inode);
  59         entry = IFIRST(header);
  60
  61         /* Compute min_offs. */
  62         for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry)) {
  63                 if (!entry->e_value_inum && entry->e_value_size) {
  64                         size_t offs = le16_to_cpu(entry->e_value_offs);
  65                         if (offs < min_offs)
  66                                 min_offs = offs;
  67                 }
  68         }
  69         free = min_offs -
  70                 ((void *)entry - (void *)IFIRST(header)) - sizeof(__u32);
  71
  72         if (EXT4_I(inode)->i_inline_off) {
  73                 entry = (struct ext4_xattr_entry *)
  74                         ((void *)raw_inode + EXT4_I(inode)->i_inline_off);
  75
  76                 free += EXT4_XATTR_SIZE(le32_to_cpu(entry->e_value_size));
  77                 goto out;
  78         }
  79
  80         free -= EXT4_XATTR_LEN(strlen(EXT4_XATTR_SYSTEM_DATA));
  81
  82         if (free > EXT4_XATTR_ROUND)
  83                 free = EXT4_XATTR_SIZE(free - EXT4_XATTR_ROUND);
  84         else
  85                 free = 0;
  86
  87 out:
  88         return free;
  89 }
  90
  91 /*
  92  * Get the maximum size we now can store in an inode.
  93  * If we can't find the space for a xattr entry, don't use the space
  94  * of the extents since we have no space to indicate the inline data.
  95  */
  96 int ext4_get_max_inline_size(struct inode *inode)
  97 {
  98         int error, max_inline_size;
  99         struct ext4_iloc iloc;
 100
 101         if (EXT4_I(inode)->i_extra_isize == 0)
 102                 return 0;
 103
 104         error = ext4_get_inode_loc(inode, &iloc);
 105         if (error) {
 106                 ext4_error_inode_err(inode, __func__, __LINE__, 0, -error,
 107                                      "can't get inode location %lu",
 108                                      inode->i_ino);
 109                 return 0;
 110         }
 111
 112         down_read(&EXT4_I(inode)->xattr_sem);
 113         max_inline_size = get_max_inline_xattr_value_size(inode, &iloc);
 114         up_read(&EXT4_I(inode)->xattr_sem);
 115
 116         brelse(iloc.bh);
 117
 118         if (!max_inline_size)
 119                 return 0;
 120
 121         return max_inline_size + EXT4_MIN_INLINE_DATA_SIZE;
 122 }
 123
 124 /*
 125  * this function does not take xattr_sem, which is OK because it is
 126  * currently only used in a code path coming form ext4_iget, before
 127  * the new inode has been unlocked
 128  */
 129 int ext4_find_inline_data_nolock(struct inode *inode)
 130 {
 131         struct ext4_xattr_ibody_find is = {
 132                 .s = { .not_found = -ENODATA, },
 133         };
 134         struct ext4_xattr_info i = {
 135                 .name_index = EXT4_XATTR_INDEX_SYSTEM,
 136                 .name = EXT4_XATTR_SYSTEM_DATA,
 137         };
 138         int error;
 139
 140         if (EXT4_I(inode)->i_extra_isize == 0)
 141                 return 0;
 142
 143         error = ext4_get_inode_loc(inode, &is.iloc);
 144         if (error)
 145                 return error;
 146
 147         error = ext4_xattr_ibody_find(inode, &i, &is);
 148         if (error)
 149                 goto out;
 150
 151         if (!is.s.not_found) {
 152                 if (is.s.here->e_value_inum) {
 153                         EXT4_ERROR_INODE(inode, "inline data xattr refers "
 154                                          "to an external xattr inode");
 155                         error = -EFSCORRUPTED;
 156                         goto out;
 157                 }
 158                 EXT4_I(inode)->i_inline_off = (u16)((void *)is.s.here -
 159                                         (void *)ext4_raw_inode(&is.iloc));
 160                 EXT4_I(inode)->i_inline_size = EXT4_MIN_INLINE_DATA_SIZE +
 161                                 le32_to_cpu(is.s.here->e_value_size);
 162                 ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
 163         }
 164 out:
 165         brelse(is.iloc.bh);
 166         return error;
 167 }
 168
 169 static int ext4_read_inline_data(struct inode *inode, void *buffer,
 170                                  unsigned int len,
 171                                  struct ext4_iloc *iloc)
 172 {
 173         struct ext4_xattr_entry *entry;
 174         struct ext4_xattr_ibody_header *header;
 175         int cp_len = 0;
 176         struct ext4_inode *raw_inode;
 177
 178         if (!len)
 179                 return 0;
 180
 181         BUG_ON(len > EXT4_I(inode)->i_inline_size);
 182
 183         cp_len = len < EXT4_MIN_INLINE_DATA_SIZE ?
 184                         len : EXT4_MIN_INLINE_DATA_SIZE;
 185
 186         raw_inode = ext4_raw_inode(iloc);
 187         memcpy(buffer, (void *)(raw_inode->i_block), cp_len);
 188
 189         len -= cp_len;
 190         buffer += cp_len;
 191
 192         if (!len)
 193                 goto out;
 194
 195         header = IHDR(inode, raw_inode);
 196         entry = (struct ext4_xattr_entry *)((void *)raw_inode +
 197                                             EXT4_I(inode)->i_inline_off);
 198         len = min_t(unsigned int, len,
 199                     (unsigned int)le32_to_cpu(entry->e_value_size));
 200
 201         memcpy(buffer,
 202                (void *)IFIRST(header) + le16_to_cpu(entry->e_value_offs), len);
 203         cp_len += len;
 204
 205 out:
 206         return cp_len;
 207 }
 208
 209 /*
 210  * write the buffer to the inline inode.
 211  * If 'create' is set, we don't need to do the extra copy in the xattr
 212  * value since it is already handled by ext4_xattr_ibody_set.
 213  * That saves us one memcpy.
 214  */
 215 static void ext4_write_inline_data(struct inode *inode, struct ext4_iloc *iloc,
 216                                    void *buffer, loff_t pos, unsigned int len)
 217 {
 218         struct ext4_xattr_entry *entry;
 219         struct ext4_xattr_ibody_header *header;
 220         struct ext4_inode *raw_inode;
 221         int cp_len = 0;
 222
 223         if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
 224                 return;
 225
 226         BUG_ON(!EXT4_I(inode)->i_inline_off);
 227         BUG_ON(pos + len > EXT4_I(inode)->i_inline_size);
 228
 229         raw_inode = ext4_raw_inode(iloc);
 230         buffer += pos;
 231
 232         if (pos < EXT4_MIN_INLINE_DATA_SIZE) {
 233                 cp_len = pos + len > EXT4_MIN_INLINE_DATA_SIZE ?
 234                          EXT4_MIN_INLINE_DATA_SIZE - pos : len;
 235                 memcpy((void *)raw_inode->i_block + pos, buffer, cp_len);
 236
 237                 len -= cp_len;
 238                 buffer += cp_len;
 239                 pos += cp_len;
 240         }
 241
 242         if (!len)
 243                 return;
 244
 245         pos -= EXT4_MIN_INLINE_DATA_SIZE;
 246         header = IHDR(inode, raw_inode);
 247         entry = (struct ext4_xattr_entry *)((void *)raw_inode +
 248                                             EXT4_I(inode)->i_inline_off);
 249
 250         memcpy((void *)IFIRST(header) + le16_to_cpu(entry->e_value_offs) + pos,
 251                buffer, len);
 252 }
 253
 254 static int ext4_create_inline_data(handle_t *handle,
 255                                    struct inode *inode, unsigned len)
 256 {
 257         int error;
 258         void *value = NULL;
 259         struct ext4_xattr_ibody_find is = {
 260                 .s = { .not_found = -ENODATA, },
 261         };
 262         struct ext4_xattr_info i = {
 263                 .name_index = EXT4_XATTR_INDEX_SYSTEM,
 264                 .name = EXT4_XATTR_SYSTEM_DATA,
 265         };
 266
 267         error = ext4_get_inode_loc(inode, &is.iloc);
 268         if (error)
 269                 return error;
 270
 271         BUFFER_TRACE(is.iloc.bh, "get_write_access");
 272         error = ext4_journal_get_write_access(handle, inode->i_sb, is.iloc.bh,
 273                                               EXT4_JTR_NONE);
 274         if (error)
 275                 goto out;
 276
 277         if (len > EXT4_MIN_INLINE_DATA_SIZE) {
 278                 value = EXT4_ZERO_XATTR_VALUE;
 279                 len -= EXT4_MIN_INLINE_DATA_SIZE;
 280         } else {
 281                 value = "";
 282                 len = 0;
 283         }
 284
 285         /* Insert the xttr entry. */
 286         i.value = value;
 287         i.value_len = len;
 288
 289         error = ext4_xattr_ibody_find(inode, &i, &is);
 290         if (error)
 291                 goto out;
 292
 293         BUG_ON(!is.s.not_found);
 294
 295         error = ext4_xattr_ibody_set(handle, inode, &i, &is);
 296         if (error) {
 297                 if (error == -ENOSPC)
 298                         ext4_clear_inode_state(inode,
 299                                                EXT4_STATE_MAY_INLINE_DATA);
 300                 goto out;
 301         }
 302
 303         memset((void *)ext4_raw_inode(&is.iloc)->i_block,
 304                 0, EXT4_MIN_INLINE_DATA_SIZE);
 305
 306         EXT4_I(inode)->i_inline_off = (u16)((void *)is.s.here -
 307                                       (void *)ext4_raw_inode(&is.iloc));
 308         EXT4_I(inode)->i_inline_size = len + EXT4_MIN_INLINE_DATA_SIZE;
 309         ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS);
 310         ext4_set_inode_flag(inode, EXT4_INODE_INLINE_DATA);
 311         get_bh(is.iloc.bh);
 312         error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
 313
 314 out:
 315         brelse(is.iloc.bh);
 316         return error;
 317 }
 318
 319 static int ext4_update_inline_data(handle_t *handle, struct inode *inode,
 320                                    unsigned int len)
 321 {
 322         int error;
 323         void *value = NULL;
 324         struct ext4_xattr_ibody_find is = {
 325                 .s = { .not_found = -ENODATA, },
 326         };
 327         struct ext4_xattr_info i = {
 328                 .name_index = EXT4_XATTR_INDEX_SYSTEM,
 329                 .name = EXT4_XATTR_SYSTEM_DATA,
 330         };
 331
 332         /* If the old space is ok, write the data directly. */
 333         if (len <= EXT4_I(inode)->i_inline_size)
 334                 return 0;
 335
 336         error = ext4_get_inode_loc(inode, &is.iloc);
 337         if (error)
 338                 return error;
 339
 340         error = ext4_xattr_ibody_find(inode, &i, &is);
 341         if (error)
 342                 goto out;
 343
 344         BUG_ON(is.s.not_found);
 345
 346         len -= EXT4_MIN_INLINE_DATA_SIZE;
 347         value = kzalloc(len, GFP_NOFS);
 348         if (!value) {
 349                 error = -ENOMEM;
 350                 goto out;
 351         }
 352
 353         error = ext4_xattr_ibody_get(inode, i.name_index, i.name,
 354                                      value, len);
 355         if (error == -ENODATA)
 356                 goto out;
 357
 358         BUFFER_TRACE(is.iloc.bh, "get_write_access");
 359         error = ext4_journal_get_write_access(handle, inode->i_sb, is.iloc.bh,
 360                                               EXT4_JTR_NONE);
 361         if (error)
 362                 goto out;
 363
 364         /* Update the xattr entry. */
 365         i.value = value;
 366         i.value_len = len;
 367
 368         error = ext4_xattr_ibody_set(handle, inode, &i, &is);
 369         if (error)
 370                 goto out;
 371
 372         EXT4_I(inode)->i_inline_off = (u16)((void *)is.s.here -
 373                                       (void *)ext4_raw_inode(&is.iloc));
 374         EXT4_I(inode)->i_inline_size = EXT4_MIN_INLINE_DATA_SIZE +
 375                                 le32_to_cpu(is.s.here->e_value_size);
 376         ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
 377         get_bh(is.iloc.bh);
 378         error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
 379
 380 out:
 381         kfree(value);
 382         brelse(is.iloc.bh);
 383         return error;
 384 }
 385
 386 static int ext4_prepare_inline_data(handle_t *handle, struct inode *inode,
 387                                     unsigned int len)
 388 {
 389         int ret, size, no_expand;
 390         struct ext4_inode_info *ei = EXT4_I(inode);
 391
 392         if (!ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA))
 393                 return -ENOSPC;
 394
 395         size = ext4_get_max_inline_size(inode);
 396         if (size < len)
 397                 return -ENOSPC;
 398
 399         ext4_write_lock_xattr(inode, &no_expand);
 400
 401         if (ei->i_inline_off)
 402                 ret = ext4_update_inline_data(handle, inode, len);
 403         else
 404                 ret = ext4_create_inline_data(handle, inode, len);
 405
 406         ext4_write_unlock_xattr(inode, &no_expand);
 407         return ret;
 408 }
 409
 410 static int ext4_destroy_inline_data_nolock(handle_t *handle,
 411                                            struct inode *inode)
 412 {
 413         struct ext4_inode_info *ei = EXT4_I(inode);
 414         struct ext4_xattr_ibody_find is = {
 415                 .s = { .not_found = 0, },
 416         };
 417         struct ext4_xattr_info i = {
 418                 .name_index = EXT4_XATTR_INDEX_SYSTEM,
 419                 .name = EXT4_XATTR_SYSTEM_DATA,
 420                 .value = NULL,
 421                 .value_len = 0,
 422         };
 423         int error;
 424
 425         if (!ei->i_inline_off)
 426                 return 0;
 427
 428         error = ext4_get_inode_loc(inode, &is.iloc);
 429         if (error)
 430                 return error;
 431
 432         error = ext4_xattr_ibody_find(inode, &i, &is);
 433         if (error)
 434                 goto out;
 435
 436         BUFFER_TRACE(is.iloc.bh, "get_write_access");
 437         error = ext4_journal_get_write_access(handle, inode->i_sb, is.iloc.bh,
 438                                               EXT4_JTR_NONE);
 439         if (error)
 440                 goto out;
 441
 442         error = ext4_xattr_ibody_set(handle, inode, &i, &is);
 443         if (error)
 444                 goto out;
 445
 446         memset((void *)ext4_raw_inode(&is.iloc)->i_block,
 447                 0, EXT4_MIN_INLINE_DATA_SIZE);
 448         memset(ei->i_data, 0, EXT4_MIN_INLINE_DATA_SIZE);
 449
 450         if (ext4_has_feature_extents(inode->i_sb)) {
 451                 if (S_ISDIR(inode->i_mode) ||
 452                     S_ISREG(inode->i_mode) || S_ISLNK(inode->i_mode)) {
 453                         ext4_set_inode_flag(inode, EXT4_INODE_EXTENTS);
 454                         ext4_ext_tree_init(handle, inode);
 455                 }
 456         }
 457         ext4_clear_inode_flag(inode, EXT4_INODE_INLINE_DATA);
 458
 459         get_bh(is.iloc.bh);
 460         error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
 461
 462         EXT4_I(inode)->i_inline_off = 0;
 463         EXT4_I(inode)->i_inline_size = 0;
 464         ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
 465 out:
 466         brelse(is.iloc.bh);
 467         if (error == -ENODATA)
 468                 error = 0;
 469         return error;
 470 }
 471
 472 static int ext4_read_inline_page(struct inode *inode, struct page *page)
 473 {
 474         void *kaddr;
 475         int ret = 0;
 476         size_t len;
 477         struct ext4_iloc iloc;
 478
 479         BUG_ON(!PageLocked(page));
 480         BUG_ON(!ext4_has_inline_data(inode));
 481         BUG_ON(page->index);
 482
 483         if (!EXT4_I(inode)->i_inline_off) {
 484                 ext4_warning(inode->i_sb, "inode %lu doesn't have inline data.",
 485                              inode->i_ino);
 486                 goto out;
 487         }
 488
 489         ret = ext4_get_inode_loc(inode, &iloc);
 490         if (ret)
 491                 goto out;
 492
 493         len = min_t(size_t, ext4_get_inline_size(inode), i_size_read(inode));
 494         kaddr = kmap_atomic(page);
 495         ret = ext4_read_inline_data(inode, kaddr, len, &iloc);
 496         flush_dcache_page(page);
 497         kunmap_atomic(kaddr);
 498         zero_user_segment(page, len, PAGE_SIZE);
 499         SetPageUptodate(page);
 500         brelse(iloc.bh);
 501
 502 out:
 503         return ret;
 504 }
 505
 506 int ext4_readpage_inline(struct inode *inode, struct page *page)
 507 {
 508         int ret = 0;
 509
 510         down_read(&EXT4_I(inode)->xattr_sem);
 511         if (!ext4_has_inline_data(inode)) {
 512                 up_read(&EXT4_I(inode)->xattr_sem);
 513                 return -EAGAIN;
 514         }
 515
 516         /*
 517          * Current inline data can only exist in the 1st page,
 518          * So for all the other pages, just set them uptodate.
 519          */
 520         if (!page->index)
 521                 ret = ext4_read_inline_page(inode, page);
 522         else if (!PageUptodate(page)) {
 523                 zero_user_segment(page, 0, PAGE_SIZE);
 524                 SetPageUptodate(page);
 525         }
 526
 527         up_read(&EXT4_I(inode)->xattr_sem);
 528
 529         unlock_page(page);
 530         return ret >= 0 ? 0 : ret;
 531 }
 532
 533 static int ext4_convert_inline_data_to_extent(struct address_space *mapping,
 534                                               struct inode *inode)
 535 {
 536         int ret, needed_blocks, no_expand;
 537         handle_t *handle = NULL;
 538         int retries = 0, sem_held = 0;
 539         struct page *page = NULL;
 540         unsigned int flags;
 541         unsigned from, to;
 542         struct ext4_iloc iloc;
 543
 544         if (!ext4_has_inline_data(inode)) {
 545                 /*
 546                  * clear the flag so that no new write
 547                  * will trap here again.
 548                  */
 549                 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
 550                 return 0;
 551         }
 552
 553         needed_blocks = ext4_writepage_trans_blocks(inode);
 554
 555         ret = ext4_get_inode_loc(inode, &iloc);
 556         if (ret)
 557                 return ret;
 558
 559 retry:
 560         handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
 561         if (IS_ERR(handle)) {
 562                 ret = PTR_ERR(handle);
 563                 handle = NULL;
 564                 goto out;
 565         }
 566
 567         /* We cannot recurse into the filesystem as the transaction is already
 568          * started */
 569         flags = memalloc_nofs_save();
 570         page = grab_cache_page_write_begin(mapping, 0);
 571         memalloc_nofs_restore(flags);
 572         if (!page) {
 573                 ret = -ENOMEM;
 574                 goto out;
 575         }
 576
 577         ext4_write_lock_xattr(inode, &no_expand);
 578         sem_held = 1;
 579         /* If some one has already done this for us, just exit. */
 580         if (!ext4_has_inline_data(inode)) {
 581                 ret = 0;
 582                 goto out;
 583         }
 584
 585         from = 0;
 586         to = ext4_get_inline_size(inode);
 587         if (!PageUptodate(page)) {
 588                 ret = ext4_read_inline_page(inode, page);
 589                 if (ret < 0)
 590                         goto out;
 591         }
 592
 593         ret = ext4_destroy_inline_data_nolock(handle, inode);
 594         if (ret)
 595                 goto out;
 596
 597         if (ext4_should_dioread_nolock(inode)) {
 598                 ret = __block_write_begin(page, from, to,
 599                                           ext4_get_block_unwritten);
 600         } else
 601                 ret = __block_write_begin(page, from, to, ext4_get_block);
 602
 603         if (!ret && ext4_should_journal_data(inode)) {
 604                 ret = ext4_walk_page_buffers(handle, inode, page_buffers(page),
 605                                              from, to, NULL,
 606                                              do_journal_get_write_access);
 607         }
 608
 609         if (ret) {
 610                 unlock_page(page);
 611                 put_page(page);
 612                 page = NULL;
 613                 ext4_orphan_add(handle, inode);
 614                 ext4_write_unlock_xattr(inode, &no_expand);
 615                 sem_held = 0;
 616                 ext4_journal_stop(handle);
 617                 handle = NULL;
 618                 ext4_truncate_failed_write(inode);
 619                 /*
 620                  * If truncate failed early the inode might
 621                  * still be on the orphan list; we need to
 622                  * make sure the inode is removed from the
 623                  * orphan list in that case.
 624                  */
 625                 if (inode->i_nlink)
 626                         ext4_orphan_del(NULL, inode);
 627         }
 628
 629         if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
 630                 goto retry;
 631
 632         if (page)
 633                 block_commit_write(page, from, to);
 634 out:
 635         if (page) {
 636                 unlock_page(page);
 637                 put_page(page);
 638         }
 639         if (sem_held)
 640                 ext4_write_unlock_xattr(inode, &no_expand);
 641         if (handle)
 642                 ext4_journal_stop(handle);
 643         brelse(iloc.bh);
 644         return ret;
 645 }
 646
 647 /*
 648  * Try to write data in the inode.
 649  * If the inode has inline data, check whether the new write can be
 650  * in the inode also. If not, create the page the handle, move the data
 651  * to the page make it update and let the later codes create extent for it.
 652  */
 653 int ext4_try_to_write_inline_data(struct address_space *mapping,
 654                                   struct inode *inode,
 655                                   loff_t pos, unsigned len,
 656                                   struct page **pagep)
 657 {
 658         int ret;
 659         handle_t *handle;
 660         unsigned int flags;
 661         struct page *page;
 662         struct ext4_iloc iloc;
 663
 664         if (pos + len > ext4_get_max_inline_size(inode))
 665                 goto convert;
 666
 667         ret = ext4_get_inode_loc(inode, &iloc);
 668         if (ret)
 669                 return ret;
 670
 671         /*
 672          * The possible write could happen in the inode,
 673          * so try to reserve the space in inode first.
 674          */
 675         handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
 676         if (IS_ERR(handle)) {
 677                 ret = PTR_ERR(handle);
 678                 handle = NULL;
 679                 goto out;
 680         }
 681
 682         ret = ext4_prepare_inline_data(handle, inode, pos + len);
 683         if (ret && ret != -ENOSPC)
 684                 goto out;
 685
 686         /* We don't have space in inline inode, so convert it to extent. */
 687         if (ret == -ENOSPC) {
 688                 ext4_journal_stop(handle);
 689                 brelse(iloc.bh);
 690                 goto convert;
 691         }
 692
 693         ret = ext4_journal_get_write_access(handle, inode->i_sb, iloc.bh,
 694                                             EXT4_JTR_NONE);
 695         if (ret)
 696                 goto out;
 697
 698         flags = memalloc_nofs_save();
 699         page = grab_cache_page_write_begin(mapping, 0);
 700         memalloc_nofs_restore(flags);
 701         if (!page) {
 702                 ret = -ENOMEM;
 703                 goto out;
 704         }
 705
 706         *pagep = page;
 707         down_read(&EXT4_I(inode)->xattr_sem);
 708         if (!ext4_has_inline_data(inode)) {
 709                 ret = 0;
 710                 unlock_page(page);
 711                 put_page(page);
 712                 goto out_up_read;
 713         }
 714
 715         if (!PageUptodate(page)) {
 716                 ret = ext4_read_inline_page(inode, page);
 717                 if (ret < 0) {
 718                         unlock_page(page);
 719                         put_page(page);
 720                         goto out_up_read;
 721                 }
 722         }
 723
 724         ret = 1;
 725         handle = NULL;
 726 out_up_read:
 727         up_read(&EXT4_I(inode)->xattr_sem);
 728 out:
 729         if (handle && (ret != 1))
 730                 ext4_journal_stop(handle);
 731         brelse(iloc.bh);
 732         return ret;
 733 convert:
 734         return ext4_convert_inline_data_to_extent(mapping, inode);
 735 }
 736
 737 int ext4_write_inline_data_end(struct inode *inode, loff_t pos, unsigned len,
 738                                unsigned copied, struct page *page)
 739 {
 740         handle_t *handle = ext4_journal_current_handle();
 741         int no_expand;
 742         void *kaddr;
 743         struct ext4_iloc iloc;
 744         int ret = 0, ret2;
 745
 746         if (unlikely(copied < len) && !PageUptodate(page))
 747                 copied = 0;
 748
 749         if (likely(copied)) {
 750                 ret = ext4_get_inode_loc(inode, &iloc);
 751                 if (ret) {
 752                         unlock_page(page);
 753                         put_page(page);
 754                         ext4_std_error(inode->i_sb, ret);
 755                         goto out;
 756                 }
 757                 ext4_write_lock_xattr(inode, &no_expand);
 758                 BUG_ON(!ext4_has_inline_data(inode));
 759
 760                 /*
 761                  * ei->i_inline_off may have changed since
 762                  * ext4_write_begin() called
 763                  * ext4_try_to_write_inline_data()
 764                  */
 765                 (void) ext4_find_inline_data_nolock(inode);
 766
 767                 kaddr = kmap_atomic(page);
 768                 ext4_write_inline_data(inode, &iloc, kaddr, pos, copied);
 769                 kunmap_atomic(kaddr);
 770                 SetPageUptodate(page);
 771                 /* clear page dirty so that writepages wouldn't work for us. */
 772                 ClearPageDirty(page);
 773
 774                 ext4_write_unlock_xattr(inode, &no_expand);
 775                 brelse(iloc.bh);
 776
 777                 /*
 778                  * It's important to update i_size while still holding page
 779                  * lock: page writeout could otherwise come in and zero
 780                  * beyond i_size.
 781                  */
 782                 ext4_update_inode_size(inode, pos + copied);
 783         }
 784         unlock_page(page);
 785         put_page(page);
 786
 787         /*
 788          * Don't mark the inode dirty under page lock. First, it unnecessarily
 789          * makes the holding time of page lock longer. Second, it forces lock
 790          * ordering of page lock and transaction start for journaling
 791          * filesystems.
 792          */
 793         if (likely(copied))
 794                 mark_inode_dirty(inode);
 795 out:
 796         /*
 797          * If we didn't copy as much data as expected, we need to trim back
 798          * size of xattr containing inline data.
 799          */
 800         if (pos + len > inode->i_size && ext4_can_truncate(inode))
 801                 ext4_orphan_add(handle, inode);
 802
 803         ret2 = ext4_journal_stop(handle);
 804         if (!ret)
 805                 ret = ret2;
 806         if (pos + len > inode->i_size) {
 807                 ext4_truncate_failed_write(inode);
 808                 /*
 809                  * If truncate failed early the inode might still be
 810                  * on the orphan list; we need to make sure the inode
 811                  * is removed from the orphan list in that case.
 812                  */
 813                 if (inode->i_nlink)
 814                         ext4_orphan_del(NULL, inode);
 815         }
 816         return ret ? ret : copied;
 817 }
 818
 819 struct buffer_head *
 820 ext4_journalled_write_inline_data(struct inode *inode,
 821                                   unsigned len,
 822                                   struct page *page)
 823 {
 824         int ret, no_expand;
 825         void *kaddr;
 826         struct ext4_iloc iloc;
 827
 828         ret = ext4_get_inode_loc(inode, &iloc);
 829         if (ret) {
 830                 ext4_std_error(inode->i_sb, ret);
 831                 return NULL;
 832         }
 833
 834         ext4_write_lock_xattr(inode, &no_expand);
 835         kaddr = kmap_atomic(page);
 836         ext4_write_inline_data(inode, &iloc, kaddr, 0, len);
 837         kunmap_atomic(kaddr);
 838         ext4_write_unlock_xattr(inode, &no_expand);
 839
 840         return iloc.bh;
 841 }
 842
 843 /*
 844  * Try to make the page cache and handle ready for the inline data case.
 845  * We can call this function in 2 cases:
 846  * 1. The inode is created and the first write exceeds inline size. We can
 847  *    clear the inode state safely.
 848  * 2. The inode has inline data, then we need to read the data, make it
 849  *    update and dirty so that ext4_da_writepages can handle it. We don't
 850  *    need to start the journal since the file's metadata isn't changed now.
 851  */
 852 static int ext4_da_convert_inline_data_to_extent(struct address_space *mapping,
 853                                                  struct inode *inode,
 854                                                  void **fsdata)
 855 {
 856         int ret = 0, inline_size;
 857         struct page *page;
 858
 859         page = grab_cache_page_write_begin(mapping, 0);
 860         if (!page)
 861                 return -ENOMEM;
 862
 863         down_read(&EXT4_I(inode)->xattr_sem);
 864         if (!ext4_has_inline_data(inode)) {
 865                 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
 866                 goto out;
 867         }
 868
 869         inline_size = ext4_get_inline_size(inode);
 870
 871         if (!PageUptodate(page)) {
 872                 ret = ext4_read_inline_page(inode, page);
 873                 if (ret < 0)
 874                         goto out;
 875         }
 876
 877         ret = __block_write_begin(page, 0, inline_size,
 878                                   ext4_da_get_block_prep);
 879         if (ret) {
 880                 up_read(&EXT4_I(inode)->xattr_sem);
 881                 unlock_page(page);
 882                 put_page(page);
 883                 ext4_truncate_failed_write(inode);
 884                 return ret;
 885         }
 886
 887         SetPageDirty(page);
 888         SetPageUptodate(page);
 889         ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
 890         *fsdata = (void *)CONVERT_INLINE_DATA;
 891
 892 out:
 893         up_read(&EXT4_I(inode)->xattr_sem);
 894         if (page) {
 895                 unlock_page(page);
 896                 put_page(page);
 897         }
 898         return ret;
 899 }
 900
 901 /*
 902  * Prepare the write for the inline data.
 903  * If the data can be written into the inode, we just read
 904  * the page and make it uptodate, and start the journal.
 905  * Otherwise read the page, makes it dirty so that it can be
 906  * handle in writepages(the i_disksize update is left to the
 907  * normal ext4_da_write_end).
 908  */
 909 int ext4_da_write_inline_data_begin(struct address_space *mapping,
 910                                     struct inode *inode,
 911                                     loff_t pos, unsigned len,
 912                                     struct page **pagep,
 913                                     void **fsdata)
 914 {
 915         int ret;
 916         handle_t *handle;
 917         struct page *page;
 918         struct ext4_iloc iloc;
 919         int retries = 0;
 920         unsigned int flags;
 921
 922         ret = ext4_get_inode_loc(inode, &iloc);
 923         if (ret)
 924                 return ret;
 925
 926 retry_journal:
 927         handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
 928         if (IS_ERR(handle)) {
 929                 ret = PTR_ERR(handle);
 930                 goto out;
 931         }
 932
 933         ret = ext4_prepare_inline_data(handle, inode, pos + len);
 934         if (ret && ret != -ENOSPC)
 935                 goto out_journal;
 936
 937         if (ret == -ENOSPC) {
 938                 ext4_journal_stop(handle);
 939                 ret = ext4_da_convert_inline_data_to_extent(mapping,
 940                                                             inode,
 941                                                             fsdata);
 942                 if (ret == -ENOSPC &&
 943                     ext4_should_retry_alloc(inode->i_sb, &retries))
 944                         goto retry_journal;
 945                 goto out;
 946         }
 947
 948         /*
 949          * We cannot recurse into the filesystem as the transaction
 950          * is already started.
 951          */
 952         flags = memalloc_nofs_save();
 953         page = grab_cache_page_write_begin(mapping, 0);
 954         memalloc_nofs_restore(flags);
 955         if (!page) {
 956                 ret = -ENOMEM;
 957                 goto out_journal;
 958         }
 959
 960         down_read(&EXT4_I(inode)->xattr_sem);
 961         if (!ext4_has_inline_data(inode)) {
 962                 ret = 0;
 963                 goto out_release_page;
 964         }
 965
 966         if (!PageUptodate(page)) {
 967                 ret = ext4_read_inline_page(inode, page);
 968                 if (ret < 0)
 969                         goto out_release_page;
 970         }
 971         ret = ext4_journal_get_write_access(handle, inode->i_sb, iloc.bh,
 972                                             EXT4_JTR_NONE);
 973         if (ret)
 974                 goto out_release_page;
 975
 976         up_read(&EXT4_I(inode)->xattr_sem);
 977         *pagep = page;
 978         brelse(iloc.bh);
 979         return 1;
 980 out_release_page:
 981         up_read(&EXT4_I(inode)->xattr_sem);
 982         unlock_page(page);
 983         put_page(page);
 984 out_journal:
 985         ext4_journal_stop(handle);
 986 out:
 987         brelse(iloc.bh);
 988         return ret;
 989 }
 990
 991 #ifdef INLINE_DIR_DEBUG
 992 void ext4_show_inline_dir(struct inode *dir, struct buffer_head *bh,
 993                           void *inline_start, int inline_size)
 994 {
 995         int offset;
 996         unsigned short de_len;
 997         struct ext4_dir_entry_2 *de = inline_start;
 998         void *dlimit = inline_start + inline_size;
 999
1000         trace_printk("inode %lu\n", dir->i_ino);
1001         offset = 0;
1002         while ((void *)de < dlimit) {
1003                 de_len = ext4_rec_len_from_disk(de->rec_len, inline_size);
1004                 trace_printk("de: off %u rlen %u name %.*s nlen %u ino %u\n",
1005                              offset, de_len, de->name_len, de->name,
1006                              de->name_len, le32_to_cpu(de->inode));
1007                 if (ext4_check_dir_entry(dir, NULL, de, bh,
1008                                          inline_start, inline_size, offset))
1009                         BUG();
1010
1011                 offset += de_len;
1012                 de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
1013         }
1014 }
1015 #else
1016 #define ext4_show_inline_dir(dir, bh, inline_start, inline_size)
1017 #endif
1018
1019 /*
1020  * Add a new entry into a inline dir.
1021  * It will return -ENOSPC if no space is available, and -EIO
1022  * and -EEXIST if directory entry already exists.
1023  */
1024 static int ext4_add_dirent_to_inline(handle_t *handle,
1025                                      struct ext4_filename *fname,
1026                                      struct inode *dir,
1027                                      struct inode *inode,
1028                                      struct ext4_iloc *iloc,
1029                                      void *inline_start, int inline_size)
1030 {
1031         int             err;
1032         struct ext4_dir_entry_2 *de;
1033
1034         err = ext4_find_dest_de(dir, inode, iloc->bh, inline_start,
1035                                 inline_size, fname, &de);
1036         if (err)
1037                 return err;
1038
1039         BUFFER_TRACE(iloc->bh, "get_write_access");
1040         err = ext4_journal_get_write_access(handle, dir->i_sb, iloc->bh,
1041                                             EXT4_JTR_NONE);
1042         if (err)
1043                 return err;
1044         ext4_insert_dentry(dir, inode, de, inline_size, fname);
1045
1046         ext4_show_inline_dir(dir, iloc->bh, inline_start, inline_size);
1047
1048         /*
1049          * XXX shouldn't update any times until successful
1050          * completion of syscall, but too many callers depend
1051          * on this.
1052          *
1053          * XXX similarly, too many callers depend on
1054          * ext4_new_inode() setting the times, but error
1055          * recovery deletes the inode, so the worst that can
1056          * happen is that the times are slightly out of date
1057          * and/or different from the directory change time.
1058          */
1059         dir->i_mtime = dir->i_ctime = current_time(dir);
1060         ext4_update_dx_flag(dir);
1061         inode_inc_iversion(dir);
1062         return 1;
1063 }
1064
1065 static void *ext4_get_inline_xattr_pos(struct inode *inode,
1066                                        struct ext4_iloc *iloc)
1067 {
1068         struct ext4_xattr_entry *entry;
1069         struct ext4_xattr_ibody_header *header;
1070
1071         BUG_ON(!EXT4_I(inode)->i_inline_off);
1072
1073         header = IHDR(inode, ext4_raw_inode(iloc));
1074         entry = (struct ext4_xattr_entry *)((void *)ext4_raw_inode(iloc) +
1075                                             EXT4_I(inode)->i_inline_off);
1076
1077         return (void *)IFIRST(header) + le16_to_cpu(entry->e_value_offs);
1078 }
1079
1080 /* Set the final de to cover the whole block. */
1081 static void ext4_update_final_de(void *de_buf, int old_size, int new_size)
1082 {
1083         struct ext4_dir_entry_2 *de, *prev_de;
1084         void *limit;
1085         int de_len;
1086
1087         de = de_buf;
1088         if (old_size) {
1089                 limit = de_buf + old_size;
1090                 do {
1091                         prev_de = de;
1092                         de_len = ext4_rec_len_from_disk(de->rec_len, old_size);
1093                         de_buf += de_len;
1094                         de = de_buf;
1095                 } while (de_buf < limit);
1096
1097                 prev_de->rec_len = ext4_rec_len_to_disk(de_len + new_size -
1098                                                         old_size, new_size);
1099         } else {
1100                 /* this is just created, so create an empty entry. */
1101                 de->inode = 0;
1102                 de->rec_len = ext4_rec_len_to_disk(new_size, new_size);
1103         }
1104 }
1105
1106 static int ext4_update_inline_dir(handle_t *handle, struct inode *dir,
1107                                   struct ext4_iloc *iloc)
1108 {
1109         int ret;
1110         int old_size = EXT4_I(dir)->i_inline_size - EXT4_MIN_INLINE_DATA_SIZE;
1111         int new_size = get_max_inline_xattr_value_size(dir, iloc);
1112
1113         if (new_size - old_size <= ext4_dir_rec_len(1, NULL))
1114                 return -ENOSPC;
1115
1116         ret = ext4_update_inline_data(handle, dir,
1117                                       new_size + EXT4_MIN_INLINE_DATA_SIZE);
1118         if (ret)
1119                 return ret;
1120
1121         ext4_update_final_de(ext4_get_inline_xattr_pos(dir, iloc), old_size,
1122                              EXT4_I(dir)->i_inline_size -
1123                                                 EXT4_MIN_INLINE_DATA_SIZE);
1124         dir->i_size = EXT4_I(dir)->i_disksize = EXT4_I(dir)->i_inline_size;
1125         return 0;
1126 }
1127
1128 static void ext4_restore_inline_data(handle_t *handle, struct inode *inode,
1129                                      struct ext4_iloc *iloc,
1130                                      void *buf, int inline_size)
1131 {
1132         int ret;
1133
1134         ret = ext4_create_inline_data(handle, inode, inline_size);
1135         if (ret) {
1136                 ext4_msg(inode->i_sb, KERN_EMERG,
1137                         "error restoring inline_data for inode -- potential data loss! (inode %lu, error %d)",
1138                         inode->i_ino, ret);
1139                 return;
1140         }
1141         ext4_write_inline_data(inode, iloc, buf, 0, inline_size);
1142         ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
1143 }
1144
1145 static int ext4_finish_convert_inline_dir(handle_t *handle,
1146                                           struct inode *inode,
1147                                           struct buffer_head *dir_block,
1148                                           void *buf,
1149                                           int inline_size)
1150 {
1151         int err, csum_size = 0, header_size = 0;
1152         struct ext4_dir_entry_2 *de;
1153         void *target = dir_block->b_data;
1154
1155         /*
1156          * First create "." and ".." and then copy the dir information
1157          * back to the block.
1158          */
1159         de = target;
1160         de = ext4_init_dot_dotdot(inode, de,
1161                 inode->i_sb->s_blocksize, csum_size,
1162                 le32_to_cpu(((struct ext4_dir_entry_2 *)buf)->inode), 1);
1163         header_size = (void *)de - target;
1164
1165         memcpy((void *)de, buf + EXT4_INLINE_DOTDOT_SIZE,
1166                 inline_size - EXT4_INLINE_DOTDOT_SIZE);
1167
1168         if (ext4_has_metadata_csum(inode->i_sb))
1169                 csum_size = sizeof(struct ext4_dir_entry_tail);
1170
1171         inode->i_size = inode->i_sb->s_blocksize;
1172         i_size_write(inode, inode->i_sb->s_blocksize);
1173         EXT4_I(inode)->i_disksize = inode->i_sb->s_blocksize;
1174         ext4_update_final_de(dir_block->b_data,
1175                         inline_size - EXT4_INLINE_DOTDOT_SIZE + header_size,
1176                         inode->i_sb->s_blocksize - csum_size);
1177
1178         if (csum_size)
1179                 ext4_initialize_dirent_tail(dir_block,
1180                                             inode->i_sb->s_blocksize);
1181         set_buffer_uptodate(dir_block);
1182         err = ext4_handle_dirty_dirblock(handle, inode, dir_block);
1183         if (err)
1184                 return err;
1185         set_buffer_verified(dir_block);
1186         return ext4_mark_inode_dirty(handle, inode);
1187 }
1188
1189 static int ext4_convert_inline_data_nolock(handle_t *handle,
1190                                            struct inode *inode,
1191                                            struct ext4_iloc *iloc)
1192 {
1193         int error;
1194         void *buf = NULL;
1195         struct buffer_head *data_bh = NULL;
1196         struct ext4_map_blocks map;
1197         int inline_size;
1198
1199         inline_size = ext4_get_inline_size(inode);
1200         buf = kmalloc(inline_size, GFP_NOFS);
1201         if (!buf) {
1202                 error = -ENOMEM;
1203                 goto out;
1204         }
1205
1206         error = ext4_read_inline_data(inode, buf, inline_size, iloc);
1207         if (error < 0)
1208                 goto out;
1209
1210         /*
1211          * Make sure the inline directory entries pass checks before we try to
1212          * convert them, so that we avoid touching stuff that needs fsck.
1213          */
1214         if (S_ISDIR(inode->i_mode)) {
1215                 error = ext4_check_all_de(inode, iloc->bh,
1216                                         buf + EXT4_INLINE_DOTDOT_SIZE,
1217                                         inline_size - EXT4_INLINE_DOTDOT_SIZE);
1218                 if (error)
1219                         goto out;
1220         }
1221
1222         error = ext4_destroy_inline_data_nolock(handle, inode);
1223         if (error)
1224                 goto out;
1225
1226         map.m_lblk = 0;
1227         map.m_len = 1;
1228         map.m_flags = 0;
1229         error = ext4_map_blocks(handle, inode, &map, EXT4_GET_BLOCKS_CREATE);
1230         if (error < 0)
1231                 goto out_restore;
1232         if (!(map.m_flags & EXT4_MAP_MAPPED)) {
1233                 error = -EIO;
1234                 goto out_restore;
1235         }
1236
1237         data_bh = sb_getblk(inode->i_sb, map.m_pblk);
1238         if (!data_bh) {
1239                 error = -ENOMEM;
1240                 goto out_restore;
1241         }
1242
1243         lock_buffer(data_bh);
1244         error = ext4_journal_get_create_access(handle, inode->i_sb, data_bh,
1245                                                EXT4_JTR_NONE);
1246         if (error) {
1247                 unlock_buffer(data_bh);
1248                 error = -EIO;
1249                 goto out_restore;
1250         }
1251         memset(data_bh->b_data, 0, inode->i_sb->s_blocksize);
1252
1253         if (!S_ISDIR(inode->i_mode)) {
1254                 memcpy(data_bh->b_data, buf, inline_size);
1255                 set_buffer_uptodate(data_bh);
1256                 error = ext4_handle_dirty_metadata(handle,
1257                                                    inode, data_bh);
1258         } else {
1259                 error = ext4_finish_convert_inline_dir(handle, inode, data_bh,
1260                                                        buf, inline_size);
1261         }
1262
1263         unlock_buffer(data_bh);
1264 out_restore:
1265         if (error)
1266                 ext4_restore_inline_data(handle, inode, iloc, buf, inline_size);
1267
1268 out:
1269         brelse(data_bh);
1270         kfree(buf);
1271         return error;
1272 }
1273
1274 /*
1275  * Try to add the new entry to the inline data.
1276  * If succeeds, return 0. If not, extended the inline dir and copied data to
1277  * the new created block.
1278  */
1279 int ext4_try_add_inline_entry(handle_t *handle, struct ext4_filename *fname,
1280                               struct inode *dir, struct inode *inode)
1281 {
1282         int ret, ret2, inline_size, no_expand;
1283         void *inline_start;
1284         struct ext4_iloc iloc;
1285
1286         ret = ext4_get_inode_loc(dir, &iloc);
1287         if (ret)
1288                 return ret;
1289
1290         ext4_write_lock_xattr(dir, &no_expand);
1291         if (!ext4_has_inline_data(dir))
1292                 goto out;
1293
1294         inline_start = (void *)ext4_raw_inode(&iloc)->i_block +
1295                                                  EXT4_INLINE_DOTDOT_SIZE;
1296         inline_size = EXT4_MIN_INLINE_DATA_SIZE - EXT4_INLINE_DOTDOT_SIZE;
1297
1298         ret = ext4_add_dirent_to_inline(handle, fname, dir, inode, &iloc,
1299                                         inline_start, inline_size);
1300         if (ret != -ENOSPC)
1301                 goto out;
1302
1303         /* check whether it can be inserted to inline xattr space. */
1304         inline_size = EXT4_I(dir)->i_inline_size -
1305                         EXT4_MIN_INLINE_DATA_SIZE;
1306         if (!inline_size) {
1307                 /* Try to use the xattr space.*/
1308                 ret = ext4_update_inline_dir(handle, dir, &iloc);
1309                 if (ret && ret != -ENOSPC)
1310                         goto out;
1311
1312                 inline_size = EXT4_I(dir)->i_inline_size -
1313                                 EXT4_MIN_INLINE_DATA_SIZE;
1314         }
1315
1316         if (inline_size) {
1317                 inline_start = ext4_get_inline_xattr_pos(dir, &iloc);
1318
1319                 ret = ext4_add_dirent_to_inline(handle, fname, dir,
1320                                                 inode, &iloc, inline_start,
1321                                                 inline_size);
1322
1323                 if (ret != -ENOSPC)
1324                         goto out;
1325         }
1326
1327         /*
1328          * The inline space is filled up, so create a new block for it.
1329          * As the extent tree will be created, we have to save the inline
1330          * dir first.
1331          */
1332         ret = ext4_convert_inline_data_nolock(handle, dir, &iloc);
1333
1334 out:
1335         ext4_write_unlock_xattr(dir, &no_expand);
1336         ret2 = ext4_mark_inode_dirty(handle, dir);
1337         if (unlikely(ret2 && !ret))
1338                 ret = ret2;
1339         brelse(iloc.bh);
1340         return ret;
1341 }
1342
1343 /*
1344  * This function fills a red-black tree with information from an
1345  * inlined dir.  It returns the number directory entries loaded
1346  * into the tree.  If there is an error it is returned in err.
1347  */
1348 int ext4_inlinedir_to_tree(struct file *dir_file,
1349                            struct inode *dir, ext4_lblk_t block,
1350                            struct dx_hash_info *hinfo,
1351                            __u32 start_hash, __u32 start_minor_hash,
1352                            int *has_inline_data)
1353 {
1354         int err = 0, count = 0;
1355         unsigned int parent_ino;
1356         int pos;
1357         struct ext4_dir_entry_2 *de;
1358         struct inode *inode = file_inode(dir_file);
1359         int ret, inline_size = 0;
1360         struct ext4_iloc iloc;
1361         void *dir_buf = NULL;
1362         struct ext4_dir_entry_2 fake;
1363         struct fscrypt_str tmp_str;
1364
1365         ret = ext4_get_inode_loc(inode, &iloc);
1366         if (ret)
1367                 return ret;
1368
1369         down_read(&EXT4_I(inode)->xattr_sem);
1370         if (!ext4_has_inline_data(inode)) {
1371                 up_read(&EXT4_I(inode)->xattr_sem);
1372                 *has_inline_data = 0;
1373                 goto out;
1374         }
1375
1376         inline_size = ext4_get_inline_size(inode);
1377         dir_buf = kmalloc(inline_size, GFP_NOFS);
1378         if (!dir_buf) {
1379                 ret = -ENOMEM;
1380                 up_read(&EXT4_I(inode)->xattr_sem);
1381                 goto out;
1382         }
1383
1384         ret = ext4_read_inline_data(inode, dir_buf, inline_size, &iloc);
1385         up_read(&EXT4_I(inode)->xattr_sem);
1386         if (ret < 0)
1387                 goto out;
1388
1389         pos = 0;
1390         parent_ino = le32_to_cpu(((struct ext4_dir_entry_2 *)dir_buf)->inode);
1391         while (pos < inline_size) {
1392                 /*
1393                  * As inlined dir doesn't store any information about '.' and
1394                  * only the inode number of '..' is stored, we have to handle
1395                  * them differently.
1396                  */
1397                 if (pos == 0) {
1398                         fake.inode = cpu_to_le32(inode->i_ino);
1399                         fake.name_len = 1;
1400                         strcpy(fake.name, ".");
1401                         fake.rec_len = ext4_rec_len_to_disk(
1402                                           ext4_dir_rec_len(fake.name_len, NULL),
1403                                           inline_size);
1404                         ext4_set_de_type(inode->i_sb, &fake, S_IFDIR);
1405                         de = &fake;
1406                         pos = EXT4_INLINE_DOTDOT_OFFSET;
1407                 } else if (pos == EXT4_INLINE_DOTDOT_OFFSET) {
1408                         fake.inode = cpu_to_le32(parent_ino);
1409                         fake.name_len = 2;
1410                         strcpy(fake.name, "..");
1411                         fake.rec_len = ext4_rec_len_to_disk(
1412                                           ext4_dir_rec_len(fake.name_len, NULL),
1413                                           inline_size);
1414                         ext4_set_de_type(inode->i_sb, &fake, S_IFDIR);
1415                         de = &fake;
1416                         pos = EXT4_INLINE_DOTDOT_SIZE;
1417                 } else {
1418                         de = (struct ext4_dir_entry_2 *)(dir_buf + pos);
1419                         pos += ext4_rec_len_from_disk(de->rec_len, inline_size);
1420                         if (ext4_check_dir_entry(inode, dir_file, de,
1421                                          iloc.bh, dir_buf,
1422                                          inline_size, pos)) {
1423                                 ret = count;
1424                                 goto out;
1425                         }
1426                 }
1427
1428                 if (ext4_hash_in_dirent(dir)) {
1429                         hinfo->hash = EXT4_DIRENT_HASH(de);
1430                         hinfo->minor_hash = EXT4_DIRENT_MINOR_HASH(de);
1431                 } else {
1432                         ext4fs_dirhash(dir, de->name, de->name_len, hinfo);
1433                 }
1434                 if ((hinfo->hash < start_hash) ||
1435                     ((hinfo->hash == start_hash) &&
1436                      (hinfo->minor_hash < start_minor_hash)))
1437                         continue;
1438                 if (de->inode == 0)
1439                         continue;
1440                 tmp_str.name = de->name;
1441                 tmp_str.len = de->name_len;
1442                 err = ext4_htree_store_dirent(dir_file, hinfo->hash,
1443                                               hinfo->minor_hash, de, &tmp_str);
1444                 if (err) {
1445                         ret = err;
1446                         goto out;
1447                 }
1448                 count++;
1449         }
1450         ret = count;
1451 out:
1452         kfree(dir_buf);
1453         brelse(iloc.bh);
1454         return ret;
1455 }
1456
1457 /*
1458  * So this function is called when the volume is mkfsed with
1459  * dir_index disabled. In order to keep f_pos persistent
1460  * after we convert from an inlined dir to a blocked based,
1461  * we just pretend that we are a normal dir and return the
1462  * offset as if '.' and '..' really take place.
1463  *
1464  */
1465 int ext4_read_inline_dir(struct file *file,
1466                          struct dir_context *ctx,
1467                          int *has_inline_data)
1468 {
1469         unsigned int offset, parent_ino;
1470         int i;
1471         struct ext4_dir_entry_2 *de;
1472         struct super_block *sb;
1473         struct inode *inode = file_inode(file);
1474         int ret, inline_size = 0;
1475         struct ext4_iloc iloc;
1476         void *dir_buf = NULL;
1477         int dotdot_offset, dotdot_size, extra_offset, extra_size;
1478
1479         ret = ext4_get_inode_loc(inode, &iloc);
1480         if (ret)
1481                 return ret;
1482
1483         down_read(&EXT4_I(inode)->xattr_sem);
1484         if (!ext4_has_inline_data(inode)) {
1485                 up_read(&EXT4_I(inode)->xattr_sem);
1486                 *has_inline_data = 0;
1487                 goto out;
1488         }
1489
1490         inline_size = ext4_get_inline_size(inode);
1491         dir_buf = kmalloc(inline_size, GFP_NOFS);
1492         if (!dir_buf) {
1493                 ret = -ENOMEM;
1494                 up_read(&EXT4_I(inode)->xattr_sem);
1495                 goto out;
1496         }
1497
1498         ret = ext4_read_inline_data(inode, dir_buf, inline_size, &iloc);
1499         up_read(&EXT4_I(inode)->xattr_sem);
1500         if (ret < 0)
1501                 goto out;
1502
1503         ret = 0;
1504         sb = inode->i_sb;
1505         parent_ino = le32_to_cpu(((struct ext4_dir_entry_2 *)dir_buf)->inode);
1506         offset = ctx->pos;
1507
1508         /*
1509          * dotdot_offset and dotdot_size is the real offset and
1510          * size for ".." and "." if the dir is block based while
1511          * the real size for them are only EXT4_INLINE_DOTDOT_SIZE.
1512          * So we will use extra_offset and extra_size to indicate them
1513          * during the inline dir iteration.
1514          */
1515         dotdot_offset = ext4_dir_rec_len(1, NULL);
1516         dotdot_size = dotdot_offset + ext4_dir_rec_len(2, NULL);
1517         extra_offset = dotdot_size - EXT4_INLINE_DOTDOT_SIZE;
1518         extra_size = extra_offset + inline_size;
1519
1520         /*
1521          * If the version has changed since the last call to
1522          * readdir(2), then we might be pointing to an invalid
1523          * dirent right now.  Scan from the start of the inline
1524          * dir to make sure.
1525          */
1526         if (!inode_eq_iversion(inode, file->f_version)) {
1527                 for (i = 0; i < extra_size && i < offset;) {
1528                         /*
1529                          * "." is with offset 0 and
1530                          * ".." is dotdot_offset.
1531                          */
1532                         if (!i) {
1533                                 i = dotdot_offset;
1534                                 continue;
1535                         } else if (i == dotdot_offset) {
1536                                 i = dotdot_size;
1537                                 continue;
1538                         }
1539                         /* for other entry, the real offset in
1540                          * the buf has to be tuned accordingly.
1541                          */
1542                         de = (struct ext4_dir_entry_2 *)
1543                                 (dir_buf + i - extra_offset);
1544                         /* It's too expensive to do a full
1545                          * dirent test each time round this
1546                          * loop, but we do have to test at
1547                          * least that it is non-zero.  A
1548                          * failure will be detected in the
1549                          * dirent test below. */
1550                         if (ext4_rec_len_from_disk(de->rec_len, extra_size)
1551                                 < ext4_dir_rec_len(1, NULL))
1552                                 break;
1553                         i += ext4_rec_len_from_disk(de->rec_len,
1554                                                     extra_size);
1555                 }
1556                 offset = i;
1557                 ctx->pos = offset;
1558                 file->f_version = inode_query_iversion(inode);
1559         }
1560
1561         while (ctx->pos < extra_size) {
1562                 if (ctx->pos == 0) {
1563                         if (!dir_emit(ctx, ".", 1, inode->i_ino, DT_DIR))
1564                                 goto out;
1565                         ctx->pos = dotdot_offset;
1566                         continue;
1567                 }
1568
1569                 if (ctx->pos == dotdot_offset) {
1570                         if (!dir_emit(ctx, "..", 2, parent_ino, DT_DIR))
1571                                 goto out;
1572                         ctx->pos = dotdot_size;
1573                         continue;
1574                 }
1575
1576                 de = (struct ext4_dir_entry_2 *)
1577                         (dir_buf + ctx->pos - extra_offset);
1578                 if (ext4_check_dir_entry(inode, file, de, iloc.bh, dir_buf,
1579                                          extra_size, ctx->pos))
1580                         goto out;
1581                 if (le32_to_cpu(de->inode)) {
1582                         if (!dir_emit(ctx, de->name, de->name_len,
1583                                       le32_to_cpu(de->inode),
1584                                       get_dtype(sb, de->file_type)))
1585                                 goto out;
1586                 }
1587                 ctx->pos += ext4_rec_len_from_disk(de->rec_len, extra_size);
1588         }
1589 out:
1590         kfree(dir_buf);
1591         brelse(iloc.bh);
1592         return ret;
1593 }
1594
1595 void *ext4_read_inline_link(struct inode *inode)
1596 {
1597         struct ext4_iloc iloc;
1598         int ret, inline_size;
1599         void *link;
1600
1601         ret = ext4_get_inode_loc(inode, &iloc);
1602         if (ret)
1603                 return ERR_PTR(ret);
1604
1605         ret = -ENOMEM;
1606         inline_size = ext4_get_inline_size(inode);
1607         link = kmalloc(inline_size + 1, GFP_NOFS);
1608         if (!link)
1609                 goto out;
1610
1611         ret = ext4_read_inline_data(inode, link, inline_size, &iloc);
1612         if (ret < 0) {
1613                 kfree(link);
1614                 goto out;
1615         }
1616         nd_terminate_link(link, inode->i_size, ret);
1617 out:
1618         if (ret < 0)
1619                 link = ERR_PTR(ret);
1620         brelse(iloc.bh);
1621         return link;
1622 }
1623
1624 struct buffer_head *ext4_get_first_inline_block(struct inode *inode,
1625                                         struct ext4_dir_entry_2 **parent_de,
1626                                         int *retval)
1627 {
1628         struct ext4_iloc iloc;
1629
1630         *retval = ext4_get_inode_loc(inode, &iloc);
1631         if (*retval)
1632                 return NULL;
1633
1634         *parent_de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block;
1635
1636         return iloc.bh;
1637 }
1638
1639 /*
1640  * Try to create the inline data for the new dir.
1641  * If it succeeds, return 0, otherwise return the error.
1642  * In case of ENOSPC, the caller should create the normal disk layout dir.
1643  */
1644 int ext4_try_create_inline_dir(handle_t *handle, struct inode *parent,
1645                                struct inode *inode)
1646 {
1647         int ret, inline_size = EXT4_MIN_INLINE_DATA_SIZE;
1648         struct ext4_iloc iloc;
1649         struct ext4_dir_entry_2 *de;
1650
1651         ret = ext4_get_inode_loc(inode, &iloc);
1652         if (ret)
1653                 return ret;
1654
1655         ret = ext4_prepare_inline_data(handle, inode, inline_size);
1656         if (ret)
1657                 goto out;
1658
1659         /*
1660          * For inline dir, we only save the inode information for the ".."
1661          * and create a fake dentry to cover the left space.
1662          */
1663         de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block;
1664         de->inode = cpu_to_le32(parent->i_ino);
1665         de = (struct ext4_dir_entry_2 *)((void *)de + EXT4_INLINE_DOTDOT_SIZE);
1666         de->inode = 0;
1667         de->rec_len = ext4_rec_len_to_disk(
1668                                 inline_size - EXT4_INLINE_DOTDOT_SIZE,
1669                                 inline_size);
1670         set_nlink(inode, 2);
1671         inode->i_size = EXT4_I(inode)->i_disksize = inline_size;
1672 out:
1673         brelse(iloc.bh);
1674         return ret;
1675 }
1676
1677 struct buffer_head *ext4_find_inline_entry(struct inode *dir,
1678                                         struct ext4_filename *fname,
1679                                         struct ext4_dir_entry_2 **res_dir,
1680                                         int *has_inline_data)
1681 {
1682         int ret;
1683         struct ext4_iloc iloc;
1684         void *inline_start;
1685         int inline_size;
1686
1687         if (ext4_get_inode_loc(dir, &iloc))
1688                 return NULL;
1689
1690         down_read(&EXT4_I(dir)->xattr_sem);
1691         if (!ext4_has_inline_data(dir)) {
1692                 *has_inline_data = 0;
1693                 goto out;
1694         }
1695
1696         inline_start = (void *)ext4_raw_inode(&iloc)->i_block +
1697                                                 EXT4_INLINE_DOTDOT_SIZE;
1698         inline_size = EXT4_MIN_INLINE_DATA_SIZE - EXT4_INLINE_DOTDOT_SIZE;
1699         ret = ext4_search_dir(iloc.bh, inline_start, inline_size,
1700                               dir, fname, 0, res_dir);
1701         if (ret == 1)
1702                 goto out_find;
1703         if (ret < 0)
1704                 goto out;
1705
1706         if (ext4_get_inline_size(dir) == EXT4_MIN_INLINE_DATA_SIZE)
1707                 goto out;
1708
1709         inline_start = ext4_get_inline_xattr_pos(dir, &iloc);
1710         inline_size = ext4_get_inline_size(dir) - EXT4_MIN_INLINE_DATA_SIZE;
1711
1712         ret = ext4_search_dir(iloc.bh, inline_start, inline_size,
1713                               dir, fname, 0, res_dir);
1714         if (ret == 1)
1715                 goto out_find;
1716
1717 out:
1718         brelse(iloc.bh);
1719         iloc.bh = NULL;
1720 out_find:
1721         up_read(&EXT4_I(dir)->xattr_sem);
1722         return iloc.bh;
1723 }
1724
1725 int ext4_delete_inline_entry(handle_t *handle,
1726                              struct inode *dir,
1727                              struct ext4_dir_entry_2 *de_del,
1728                              struct buffer_head *bh,
1729                              int *has_inline_data)
1730 {
1731         int err, inline_size, no_expand;
1732         struct ext4_iloc iloc;
1733         void *inline_start;
1734
1735         err = ext4_get_inode_loc(dir, &iloc);
1736         if (err)
1737                 return err;
1738
1739         ext4_write_lock_xattr(dir, &no_expand);
1740         if (!ext4_has_inline_data(dir)) {
1741                 *has_inline_data = 0;
1742                 goto out;
1743         }
1744
1745         if ((void *)de_del - ((void *)ext4_raw_inode(&iloc)->i_block) <
1746                 EXT4_MIN_INLINE_DATA_SIZE) {
1747                 inline_start = (void *)ext4_raw_inode(&iloc)->i_block +
1748                                         EXT4_INLINE_DOTDOT_SIZE;
1749                 inline_size = EXT4_MIN_INLINE_DATA_SIZE -
1750                                 EXT4_INLINE_DOTDOT_SIZE;
1751         } else {
1752                 inline_start = ext4_get_inline_xattr_pos(dir, &iloc);
1753                 inline_size = ext4_get_inline_size(dir) -
1754                                 EXT4_MIN_INLINE_DATA_SIZE;
1755         }
1756
1757         BUFFER_TRACE(bh, "get_write_access");
1758         err = ext4_journal_get_write_access(handle, dir->i_sb, bh,
1759                                             EXT4_JTR_NONE);
1760         if (err)
1761                 goto out;
1762
1763         err = ext4_generic_delete_entry(dir, de_del, bh,
1764                                         inline_start, inline_size, 0);
1765         if (err)
1766                 goto out;
1767
1768         ext4_show_inline_dir(dir, iloc.bh, inline_start, inline_size);
1769 out:
1770         ext4_write_unlock_xattr(dir, &no_expand);
1771         if (likely(err == 0))
1772                 err = ext4_mark_inode_dirty(handle, dir);
1773         brelse(iloc.bh);
1774         if (err != -ENOENT)
1775                 ext4_std_error(dir->i_sb, err);
1776         return err;
1777 }
1778
1779 /*
1780  * Get the inline dentry at offset.
1781  */
1782 static inline struct ext4_dir_entry_2 *
1783 ext4_get_inline_entry(struct inode *inode,
1784                       struct ext4_iloc *iloc,
1785                       unsigned int offset,
1786                       void **inline_start,
1787                       int *inline_size)
1788 {
1789         void *inline_pos;
1790
1791         BUG_ON(offset > ext4_get_inline_size(inode));
1792
1793         if (offset < EXT4_MIN_INLINE_DATA_SIZE) {
1794                 inline_pos = (void *)ext4_raw_inode(iloc)->i_block;
1795                 *inline_size = EXT4_MIN_INLINE_DATA_SIZE;
1796         } else {
1797                 inline_pos = ext4_get_inline_xattr_pos(inode, iloc);
1798                 offset -= EXT4_MIN_INLINE_DATA_SIZE;
1799                 *inline_size = ext4_get_inline_size(inode) -
1800                                 EXT4_MIN_INLINE_DATA_SIZE;
1801         }
1802
1803         if (inline_start)
1804                 *inline_start = inline_pos;
1805         return (struct ext4_dir_entry_2 *)(inline_pos + offset);
1806 }
1807
1808 bool empty_inline_dir(struct inode *dir, int *has_inline_data)
1809 {
1810         int err, inline_size;
1811         struct ext4_iloc iloc;
1812         size_t inline_len;
1813         void *inline_pos;
1814         unsigned int offset;
1815         struct ext4_dir_entry_2 *de;
1816         bool ret = false;
1817
1818         err = ext4_get_inode_loc(dir, &iloc);
1819         if (err) {
1820                 EXT4_ERROR_INODE_ERR(dir, -err,
1821                                      "error %d getting inode %lu block",
1822                                      err, dir->i_ino);
1823                 return false;
1824         }
1825
1826         down_read(&EXT4_I(dir)->xattr_sem);
1827         if (!ext4_has_inline_data(dir)) {
1828                 *has_inline_data = 0;
1829                 ret = true;
1830                 goto out;
1831         }
1832
1833         de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block;
1834         if (!le32_to_cpu(de->inode)) {
1835                 ext4_warning(dir->i_sb,
1836                              "bad inline directory (dir #%lu) - no `..'",
1837                              dir->i_ino);
1838                 goto out;
1839         }
1840
1841         inline_len = ext4_get_inline_size(dir);
1842         offset = EXT4_INLINE_DOTDOT_SIZE;
1843         while (offset < inline_len) {
1844                 de = ext4_get_inline_entry(dir, &iloc, offset,
1845                                            &inline_pos, &inline_size);
1846                 if (ext4_check_dir_entry(dir, NULL, de,
1847                                          iloc.bh, inline_pos,
1848                                          inline_size, offset)) {
1849                         ext4_warning(dir->i_sb,
1850                                      "bad inline directory (dir #%lu) - "
1851                                      "inode %u, rec_len %u, name_len %d"
1852                                      "inline size %d",
1853                                      dir->i_ino, le32_to_cpu(de->inode),
1854                                      le16_to_cpu(de->rec_len), de->name_len,
1855                                      inline_size);
1856                         goto out;
1857                 }
1858                 if (le32_to_cpu(de->inode)) {
1859                         goto out;
1860                 }
1861                 offset += ext4_rec_len_from_disk(de->rec_len, inline_size);
1862         }
1863
1864         ret = true;
1865 out:
1866         up_read(&EXT4_I(dir)->xattr_sem);
1867         brelse(iloc.bh);
1868         return ret;
1869 }
1870
1871 int ext4_destroy_inline_data(handle_t *handle, struct inode *inode)
1872 {
1873         int ret, no_expand;
1874
1875         ext4_write_lock_xattr(inode, &no_expand);
1876         ret = ext4_destroy_inline_data_nolock(handle, inode);
1877         ext4_write_unlock_xattr(inode, &no_expand);
1878
1879         return ret;
1880 }
1881
1882 int ext4_inline_data_iomap(struct inode *inode, struct iomap *iomap)
1883 {
1884         __u64 addr;
1885         int error = -EAGAIN;
1886         struct ext4_iloc iloc;
1887
1888         down_read(&EXT4_I(inode)->xattr_sem);
1889         if (!ext4_has_inline_data(inode))
1890                 goto out;
1891
1892         error = ext4_get_inode_loc(inode, &iloc);
1893         if (error)
1894                 goto out;
1895
1896         addr = (__u64)iloc.bh->b_blocknr << inode->i_sb->s_blocksize_bits;
1897         addr += (char *)ext4_raw_inode(&iloc) - iloc.bh->b_data;
1898         addr += offsetof(struct ext4_inode, i_block);
1899
1900         brelse(iloc.bh);
1901
1902         iomap->addr = addr;
1903         iomap->offset = 0;
1904         iomap->length = min_t(loff_t, ext4_get_inline_size(inode),
1905                               i_size_read(inode));
1906         iomap->type = IOMAP_INLINE;
1907         iomap->flags = 0;
1908
1909 out:
1910         up_read(&EXT4_I(inode)->xattr_sem);
1911         return error;
1912 }
1913
1914 int ext4_inline_data_truncate(struct inode *inode, int *has_inline)
1915 {
1916         handle_t *handle;
1917         int inline_size, value_len, needed_blocks, no_expand, err = 0;
1918         size_t i_size;
1919         void *value = NULL;
1920         struct ext4_xattr_ibody_find is = {
1921                 .s = { .not_found = -ENODATA, },
1922         };
1923         struct ext4_xattr_info i = {
1924                 .name_index = EXT4_XATTR_INDEX_SYSTEM,
1925                 .name = EXT4_XATTR_SYSTEM_DATA,
1926         };
1927
1928
1929         needed_blocks = ext4_writepage_trans_blocks(inode);
1930         handle = ext4_journal_start(inode, EXT4_HT_INODE, needed_blocks);
1931         if (IS_ERR(handle))
1932                 return PTR_ERR(handle);
1933
1934         ext4_write_lock_xattr(inode, &no_expand);
1935         if (!ext4_has_inline_data(inode)) {
1936                 ext4_write_unlock_xattr(inode, &no_expand);
1937                 *has_inline = 0;
1938                 ext4_journal_stop(handle);
1939                 return 0;
1940         }
1941
1942         if ((err = ext4_orphan_add(handle, inode)) != 0)
1943                 goto out;
1944
1945         if ((err = ext4_get_inode_loc(inode, &is.iloc)) != 0)
1946                 goto out;
1947
1948         down_write(&EXT4_I(inode)->i_data_sem);
1949         i_size = inode->i_size;
1950         inline_size = ext4_get_inline_size(inode);
1951         EXT4_I(inode)->i_disksize = i_size;
1952
1953         if (i_size < inline_size) {
1954                 /*
1955                  * if there's inline data to truncate and this file was
1956                  * converted to extents after that inline data was written,
1957                  * the extent status cache must be cleared to avoid leaving
1958                  * behind stale delayed allocated extent entries
1959                  */
1960                 if (!ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
1961 retry:
1962                         err = ext4_es_remove_extent(inode, 0, EXT_MAX_BLOCKS);
1963                         if (err == -ENOMEM) {
1964                                 memalloc_retry_wait(GFP_ATOMIC);
1965                                 goto retry;
1966                         }
1967                         if (err)
1968                                 goto out_error;
1969                 }
1970
1971                 /* Clear the content in the xattr space. */
1972                 if (inline_size > EXT4_MIN_INLINE_DATA_SIZE) {
1973                         if ((err = ext4_xattr_ibody_find(inode, &i, &is)) != 0)
1974                                 goto out_error;
1975
1976                         BUG_ON(is.s.not_found);
1977
1978                         value_len = le32_to_cpu(is.s.here->e_value_size);
1979                         value = kmalloc(value_len, GFP_NOFS);
1980                         if (!value) {
1981                                 err = -ENOMEM;
1982                                 goto out_error;
1983                         }
1984
1985                         err = ext4_xattr_ibody_get(inode, i.name_index,
1986                                                    i.name, value, value_len);
1987                         if (err <= 0)
1988                                 goto out_error;
1989
1990                         i.value = value;
1991                         i.value_len = i_size > EXT4_MIN_INLINE_DATA_SIZE ?
1992                                         i_size - EXT4_MIN_INLINE_DATA_SIZE : 0;
1993                         err = ext4_xattr_ibody_set(handle, inode, &i, &is);
1994                         if (err)
1995                                 goto out_error;
1996                 }
1997
1998                 /* Clear the content within i_blocks. */
1999                 if (i_size < EXT4_MIN_INLINE_DATA_SIZE) {
2000                         void *p = (void *) ext4_raw_inode(&is.iloc)->i_block;
2001                         memset(p + i_size, 0,
2002                                EXT4_MIN_INLINE_DATA_SIZE - i_size);
2003                 }
2004
2005                 EXT4_I(inode)->i_inline_size = i_size <
2006                                         EXT4_MIN_INLINE_DATA_SIZE ?
2007                                         EXT4_MIN_INLINE_DATA_SIZE : i_size;
2008         }
2009
2010 out_error:
2011         up_write(&EXT4_I(inode)->i_data_sem);
2012 out:
2013         brelse(is.iloc.bh);
2014         ext4_write_unlock_xattr(inode, &no_expand);
2015         kfree(value);
2016         if (inode->i_nlink)
2017                 ext4_orphan_del(handle, inode);
2018
2019         if (err == 0) {
2020                 inode->i_mtime = inode->i_ctime = current_time(inode);
2021                 err = ext4_mark_inode_dirty(handle, inode);
2022                 if (IS_SYNC(inode))
2023                         ext4_handle_sync(handle);
2024         }
2025         ext4_journal_stop(handle);
2026         return err;
2027 }
2028
2029 int ext4_convert_inline_data(struct inode *inode)
2030 {
2031         int error, needed_blocks, no_expand;
2032         handle_t *handle;
2033         struct ext4_iloc iloc;
2034
2035         if (!ext4_has_inline_data(inode)) {
2036                 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
2037                 return 0;
2038         } else if (!ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2039                 /*
2040                  * Inode has inline data but EXT4_STATE_MAY_INLINE_DATA is
2041                  * cleared. This means we are in the middle of moving of
2042                  * inline data to delay allocated block. Just force writeout
2043                  * here to finish conversion.
2044                  */
2045                 error = filemap_flush(inode->i_mapping);
2046                 if (error)
2047                         return error;
2048                 if (!ext4_has_inline_data(inode))
2049                         return 0;
2050         }
2051
2052         needed_blocks = ext4_writepage_trans_blocks(inode);
2053
2054         iloc.bh = NULL;
2055         error = ext4_get_inode_loc(inode, &iloc);
2056         if (error)
2057                 return error;
2058
2059         handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
2060         if (IS_ERR(handle)) {
2061                 error = PTR_ERR(handle);
2062                 goto out_free;
2063         }
2064
2065         ext4_write_lock_xattr(inode, &no_expand);
2066         if (ext4_has_inline_data(inode))
2067                 error = ext4_convert_inline_data_nolock(handle, inode, &iloc);
2068         ext4_write_unlock_xattr(inode, &no_expand);
2069         ext4_journal_stop(handle);
2070 out_free:
2071         brelse(iloc.bh);
2072         return error;
2073 }