fs/ext4/inline.c

   1 // SPDX-License-Identifier: LGPL-2.1
   2 /*
   3  * Copyright (c) 2012 Taobao.
   4  * Written by Tao Ma <boyu.mt@taobao.com>
   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         }
 163 out:
 164         brelse(is.iloc.bh);
 165         return error;
 166 }
 167
 168 static int ext4_read_inline_data(struct inode *inode, void *buffer,
 169                                  unsigned int len,
 170                                  struct ext4_iloc *iloc)
 171 {
 172         struct ext4_xattr_entry *entry;
 173         struct ext4_xattr_ibody_header *header;
 174         int cp_len = 0;
 175         struct ext4_inode *raw_inode;
 176
 177         if (!len)
 178                 return 0;
 179
 180         BUG_ON(len > EXT4_I(inode)->i_inline_size);
 181
 182         cp_len = min_t(unsigned int, len, EXT4_MIN_INLINE_DATA_SIZE);
 183
 184         raw_inode = ext4_raw_inode(iloc);
 185         memcpy(buffer, (void *)(raw_inode->i_block), cp_len);
 186
 187         len -= cp_len;
 188         buffer += cp_len;
 189
 190         if (!len)
 191                 goto out;
 192
 193         header = IHDR(inode, raw_inode);
 194         entry = (struct ext4_xattr_entry *)((void *)raw_inode +
 195                                             EXT4_I(inode)->i_inline_off);
 196         len = min_t(unsigned int, len,
 197                     (unsigned int)le32_to_cpu(entry->e_value_size));
 198
 199         memcpy(buffer,
 200                (void *)IFIRST(header) + le16_to_cpu(entry->e_value_offs), len);
 201         cp_len += len;
 202
 203 out:
 204         return cp_len;
 205 }
 206
 207 /*
 208  * write the buffer to the inline inode.
 209  * If 'create' is set, we don't need to do the extra copy in the xattr
 210  * value since it is already handled by ext4_xattr_ibody_set.
 211  * That saves us one memcpy.
 212  */
 213 static void ext4_write_inline_data(struct inode *inode, struct ext4_iloc *iloc,
 214                                    void *buffer, loff_t pos, unsigned int len)
 215 {
 216         struct ext4_xattr_entry *entry;
 217         struct ext4_xattr_ibody_header *header;
 218         struct ext4_inode *raw_inode;
 219         int cp_len = 0;
 220
 221         if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
 222                 return;
 223
 224         BUG_ON(!EXT4_I(inode)->i_inline_off);
 225         BUG_ON(pos + len > EXT4_I(inode)->i_inline_size);
 226
 227         raw_inode = ext4_raw_inode(iloc);
 228         buffer += pos;
 229
 230         if (pos < EXT4_MIN_INLINE_DATA_SIZE) {
 231                 cp_len = pos + len > EXT4_MIN_INLINE_DATA_SIZE ?
 232                          EXT4_MIN_INLINE_DATA_SIZE - pos : len;
 233                 memcpy((void *)raw_inode->i_block + pos, buffer, cp_len);
 234
 235                 len -= cp_len;
 236                 buffer += cp_len;
 237                 pos += cp_len;
 238         }
 239
 240         if (!len)
 241                 return;
 242
 243         pos -= EXT4_MIN_INLINE_DATA_SIZE;
 244         header = IHDR(inode, raw_inode);
 245         entry = (struct ext4_xattr_entry *)((void *)raw_inode +
 246                                             EXT4_I(inode)->i_inline_off);
 247
 248         memcpy((void *)IFIRST(header) + le16_to_cpu(entry->e_value_offs) + pos,
 249                buffer, len);
 250 }
 251
 252 static int ext4_create_inline_data(handle_t *handle,
 253                                    struct inode *inode, unsigned len)
 254 {
 255         int error;
 256         void *value = NULL;
 257         struct ext4_xattr_ibody_find is = {
 258                 .s = { .not_found = -ENODATA, },
 259         };
 260         struct ext4_xattr_info i = {
 261                 .name_index = EXT4_XATTR_INDEX_SYSTEM,
 262                 .name = EXT4_XATTR_SYSTEM_DATA,
 263         };
 264
 265         error = ext4_get_inode_loc(inode, &is.iloc);
 266         if (error)
 267                 return error;
 268
 269         BUFFER_TRACE(is.iloc.bh, "get_write_access");
 270         error = ext4_journal_get_write_access(handle, inode->i_sb, is.iloc.bh,
 271                                               EXT4_JTR_NONE);
 272         if (error)
 273                 goto out;
 274
 275         if (len > EXT4_MIN_INLINE_DATA_SIZE) {
 276                 value = EXT4_ZERO_XATTR_VALUE;
 277                 len -= EXT4_MIN_INLINE_DATA_SIZE;
 278         } else {
 279                 value = "";
 280                 len = 0;
 281         }
 282
 283         /* Insert the xttr entry. */
 284         i.value = value;
 285         i.value_len = len;
 286
 287         error = ext4_xattr_ibody_find(inode, &i, &is);
 288         if (error)
 289                 goto out;
 290
 291         BUG_ON(!is.s.not_found);
 292
 293         error = ext4_xattr_ibody_set(handle, inode, &i, &is);
 294         if (error) {
 295                 if (error == -ENOSPC)
 296                         ext4_clear_inode_state(inode,
 297                                                EXT4_STATE_MAY_INLINE_DATA);
 298                 goto out;
 299         }
 300
 301         memset((void *)ext4_raw_inode(&is.iloc)->i_block,
 302                 0, EXT4_MIN_INLINE_DATA_SIZE);
 303
 304         EXT4_I(inode)->i_inline_off = (u16)((void *)is.s.here -
 305                                       (void *)ext4_raw_inode(&is.iloc));
 306         EXT4_I(inode)->i_inline_size = len + EXT4_MIN_INLINE_DATA_SIZE;
 307         ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS);
 308         ext4_set_inode_flag(inode, EXT4_INODE_INLINE_DATA);
 309         get_bh(is.iloc.bh);
 310         error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
 311
 312 out:
 313         brelse(is.iloc.bh);
 314         return error;
 315 }
 316
 317 static int ext4_update_inline_data(handle_t *handle, struct inode *inode,
 318                                    unsigned int len)
 319 {
 320         int error;
 321         void *value = NULL;
 322         struct ext4_xattr_ibody_find is = {
 323                 .s = { .not_found = -ENODATA, },
 324         };
 325         struct ext4_xattr_info i = {
 326                 .name_index = EXT4_XATTR_INDEX_SYSTEM,
 327                 .name = EXT4_XATTR_SYSTEM_DATA,
 328         };
 329
 330         /* If the old space is ok, write the data directly. */
 331         if (len <= EXT4_I(inode)->i_inline_size)
 332                 return 0;
 333
 334         error = ext4_get_inode_loc(inode, &is.iloc);
 335         if (error)
 336                 return error;
 337
 338         error = ext4_xattr_ibody_find(inode, &i, &is);
 339         if (error)
 340                 goto out;
 341
 342         BUG_ON(is.s.not_found);
 343
 344         len -= EXT4_MIN_INLINE_DATA_SIZE;
 345         value = kzalloc(len, GFP_NOFS);
 346         if (!value) {
 347                 error = -ENOMEM;
 348                 goto out;
 349         }
 350
 351         error = ext4_xattr_ibody_get(inode, i.name_index, i.name,
 352                                      value, len);
 353         if (error == -ENODATA)
 354                 goto out;
 355
 356         BUFFER_TRACE(is.iloc.bh, "get_write_access");
 357         error = ext4_journal_get_write_access(handle, inode->i_sb, is.iloc.bh,
 358                                               EXT4_JTR_NONE);
 359         if (error)
 360                 goto out;
 361
 362         /* Update the xattr entry. */
 363         i.value = value;
 364         i.value_len = len;
 365
 366         error = ext4_xattr_ibody_set(handle, inode, &i, &is);
 367         if (error)
 368                 goto out;
 369
 370         EXT4_I(inode)->i_inline_off = (u16)((void *)is.s.here -
 371                                       (void *)ext4_raw_inode(&is.iloc));
 372         EXT4_I(inode)->i_inline_size = EXT4_MIN_INLINE_DATA_SIZE +
 373                                 le32_to_cpu(is.s.here->e_value_size);
 374         ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
 375         get_bh(is.iloc.bh);
 376         error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
 377
 378 out:
 379         kfree(value);
 380         brelse(is.iloc.bh);
 381         return error;
 382 }
 383
 384 static int ext4_prepare_inline_data(handle_t *handle, struct inode *inode,
 385                                     unsigned int len)
 386 {
 387         int ret, size, no_expand;
 388         struct ext4_inode_info *ei = EXT4_I(inode);
 389
 390         if (!ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA))
 391                 return -ENOSPC;
 392
 393         size = ext4_get_max_inline_size(inode);
 394         if (size < len)
 395                 return -ENOSPC;
 396
 397         ext4_write_lock_xattr(inode, &no_expand);
 398
 399         if (ei->i_inline_off)
 400                 ret = ext4_update_inline_data(handle, inode, len);
 401         else
 402                 ret = ext4_create_inline_data(handle, inode, len);
 403
 404         ext4_write_unlock_xattr(inode, &no_expand);
 405         return ret;
 406 }
 407
 408 static int ext4_destroy_inline_data_nolock(handle_t *handle,
 409                                            struct inode *inode)
 410 {
 411         struct ext4_inode_info *ei = EXT4_I(inode);
 412         struct ext4_xattr_ibody_find is = {
 413                 .s = { .not_found = 0, },
 414         };
 415         struct ext4_xattr_info i = {
 416                 .name_index = EXT4_XATTR_INDEX_SYSTEM,
 417                 .name = EXT4_XATTR_SYSTEM_DATA,
 418                 .value = NULL,
 419                 .value_len = 0,
 420         };
 421         int error;
 422
 423         if (!ei->i_inline_off)
 424                 return 0;
 425
 426         error = ext4_get_inode_loc(inode, &is.iloc);
 427         if (error)
 428                 return error;
 429
 430         error = ext4_xattr_ibody_find(inode, &i, &is);
 431         if (error)
 432                 goto out;
 433
 434         BUFFER_TRACE(is.iloc.bh, "get_write_access");
 435         error = ext4_journal_get_write_access(handle, inode->i_sb, is.iloc.bh,
 436                                               EXT4_JTR_NONE);
 437         if (error)
 438                 goto out;
 439
 440         error = ext4_xattr_ibody_set(handle, inode, &i, &is);
 441         if (error)
 442                 goto out;
 443
 444         memset((void *)ext4_raw_inode(&is.iloc)->i_block,
 445                 0, EXT4_MIN_INLINE_DATA_SIZE);
 446         memset(ei->i_data, 0, EXT4_MIN_INLINE_DATA_SIZE);
 447
 448         if (ext4_has_feature_extents(inode->i_sb)) {
 449                 if (S_ISDIR(inode->i_mode) ||
 450                     S_ISREG(inode->i_mode) || S_ISLNK(inode->i_mode)) {
 451                         ext4_set_inode_flag(inode, EXT4_INODE_EXTENTS);
 452                         ext4_ext_tree_init(handle, inode);
 453                 }
 454         }
 455         ext4_clear_inode_flag(inode, EXT4_INODE_INLINE_DATA);
 456
 457         get_bh(is.iloc.bh);
 458         error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
 459
 460         EXT4_I(inode)->i_inline_off = 0;
 461         EXT4_I(inode)->i_inline_size = 0;
 462         ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
 463 out:
 464         brelse(is.iloc.bh);
 465         if (error == -ENODATA)
 466                 error = 0;
 467         return error;
 468 }
 469
 470 static int ext4_read_inline_folio(struct inode *inode, struct folio *folio)
 471 {
 472         void *kaddr;
 473         int ret = 0;
 474         size_t len;
 475         struct ext4_iloc iloc;
 476
 477         BUG_ON(!folio_test_locked(folio));
 478         BUG_ON(!ext4_has_inline_data(inode));
 479         BUG_ON(folio->index);
 480
 481         if (!EXT4_I(inode)->i_inline_off) {
 482                 ext4_warning(inode->i_sb, "inode %lu doesn't have inline data.",
 483                              inode->i_ino);
 484                 goto out;
 485         }
 486
 487         ret = ext4_get_inode_loc(inode, &iloc);
 488         if (ret)
 489                 goto out;
 490
 491         len = min_t(size_t, ext4_get_inline_size(inode), i_size_read(inode));
 492         BUG_ON(len > PAGE_SIZE);
 493         kaddr = kmap_local_folio(folio, 0);
 494         ret = ext4_read_inline_data(inode, kaddr, len, &iloc);
 495         flush_dcache_folio(folio);
 496         kunmap_local(kaddr);
 497         folio_zero_segment(folio, len, folio_size(folio));
 498         folio_mark_uptodate(folio);
 499         brelse(iloc.bh);
 500
 501 out:
 502         return ret;
 503 }
 504
 505 int ext4_readpage_inline(struct inode *inode, struct folio *folio)
 506 {
 507         int ret = 0;
 508
 509         down_read(&EXT4_I(inode)->xattr_sem);
 510         if (!ext4_has_inline_data(inode)) {
 511                 up_read(&EXT4_I(inode)->xattr_sem);
 512                 return -EAGAIN;
 513         }
 514
 515         /*
 516          * Current inline data can only exist in the 1st page,
 517          * So for all the other pages, just set them uptodate.
 518          */
 519         if (!folio->index)
 520                 ret = ext4_read_inline_folio(inode, folio);
 521         else if (!folio_test_uptodate(folio)) {
 522                 folio_zero_segment(folio, 0, folio_size(folio));
 523                 folio_mark_uptodate(folio);
 524         }
 525
 526         up_read(&EXT4_I(inode)->xattr_sem);
 527
 528         folio_unlock(folio);
 529         return ret >= 0 ? 0 : ret;
 530 }
 531
 532 static int ext4_convert_inline_data_to_extent(struct address_space *mapping,
 533                                               struct inode *inode)
 534 {
 535         int ret, needed_blocks, no_expand;
 536         handle_t *handle = NULL;
 537         int retries = 0, sem_held = 0;
 538         struct folio *folio = NULL;
 539         unsigned from, to;
 540         struct ext4_iloc iloc;
 541
 542         if (!ext4_has_inline_data(inode)) {
 543                 /*
 544                  * clear the flag so that no new write
 545                  * will trap here again.
 546                  */
 547                 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
 548                 return 0;
 549         }
 550
 551         needed_blocks = ext4_writepage_trans_blocks(inode);
 552
 553         ret = ext4_get_inode_loc(inode, &iloc);
 554         if (ret)
 555                 return ret;
 556
 557 retry:
 558         handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
 559         if (IS_ERR(handle)) {
 560                 ret = PTR_ERR(handle);
 561                 handle = NULL;
 562                 goto out;
 563         }
 564
 565         /* We cannot recurse into the filesystem as the transaction is already
 566          * started */
 567         folio = __filemap_get_folio(mapping, 0, FGP_WRITEBEGIN | FGP_NOFS,
 568                         mapping_gfp_mask(mapping));
 569         if (!folio) {
 570                 ret = -ENOMEM;
 571                 goto out;
 572         }
 573
 574         ext4_write_lock_xattr(inode, &no_expand);
 575         sem_held = 1;
 576         /* If some one has already done this for us, just exit. */
 577         if (!ext4_has_inline_data(inode)) {
 578                 ret = 0;
 579                 goto out;
 580         }
 581
 582         from = 0;
 583         to = ext4_get_inline_size(inode);
 584         if (!folio_test_uptodate(folio)) {
 585                 ret = ext4_read_inline_folio(inode, folio);
 586                 if (ret < 0)
 587                         goto out;
 588         }
 589
 590         ret = ext4_destroy_inline_data_nolock(handle, inode);
 591         if (ret)
 592                 goto out;
 593
 594         if (ext4_should_dioread_nolock(inode)) {
 595                 ret = __block_write_begin(&folio->page, from, to,
 596                                           ext4_get_block_unwritten);
 597         } else
 598                 ret = __block_write_begin(&folio->page, from, to, ext4_get_block);
 599
 600         if (!ret && ext4_should_journal_data(inode)) {
 601                 ret = ext4_walk_page_buffers(handle, inode,
 602                                              folio_buffers(folio), from, to,
 603                                              NULL, do_journal_get_write_access);
 604         }
 605
 606         if (ret) {
 607                 folio_unlock(folio);
 608                 folio_put(folio);
 609                 folio = NULL;
 610                 ext4_orphan_add(handle, inode);
 611                 ext4_write_unlock_xattr(inode, &no_expand);
 612                 sem_held = 0;
 613                 ext4_journal_stop(handle);
 614                 handle = NULL;
 615                 ext4_truncate_failed_write(inode);
 616                 /*
 617                  * If truncate failed early the inode might
 618                  * still be on the orphan list; we need to
 619                  * make sure the inode is removed from the
 620                  * orphan list in that case.
 621                  */
 622                 if (inode->i_nlink)
 623                         ext4_orphan_del(NULL, inode);
 624         }
 625
 626         if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
 627                 goto retry;
 628
 629         if (folio)
 630                 block_commit_write(&folio->page, from, to);
 631 out:
 632         if (folio) {
 633                 folio_unlock(folio);
 634                 folio_put(folio);
 635         }
 636         if (sem_held)
 637                 ext4_write_unlock_xattr(inode, &no_expand);
 638         if (handle)
 639                 ext4_journal_stop(handle);
 640         brelse(iloc.bh);
 641         return ret;
 642 }
 643
 644 /*
 645  * Try to write data in the inode.
 646  * If the inode has inline data, check whether the new write can be
 647  * in the inode also. If not, create the page the handle, move the data
 648  * to the page make it update and let the later codes create extent for it.
 649  */
 650 int ext4_try_to_write_inline_data(struct address_space *mapping,
 651                                   struct inode *inode,
 652                                   loff_t pos, unsigned len,
 653                                   struct page **pagep)
 654 {
 655         int ret;
 656         handle_t *handle;
 657         struct folio *folio;
 658         struct ext4_iloc iloc;
 659
 660         if (pos + len > ext4_get_max_inline_size(inode))
 661                 goto convert;
 662
 663         ret = ext4_get_inode_loc(inode, &iloc);
 664         if (ret)
 665                 return ret;
 666
 667         /*
 668          * The possible write could happen in the inode,
 669          * so try to reserve the space in inode first.
 670          */
 671         handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
 672         if (IS_ERR(handle)) {
 673                 ret = PTR_ERR(handle);
 674                 handle = NULL;
 675                 goto out;
 676         }
 677
 678         ret = ext4_prepare_inline_data(handle, inode, pos + len);
 679         if (ret && ret != -ENOSPC)
 680                 goto out;
 681
 682         /* We don't have space in inline inode, so convert it to extent. */
 683         if (ret == -ENOSPC) {
 684                 ext4_journal_stop(handle);
 685                 brelse(iloc.bh);
 686                 goto convert;
 687         }
 688
 689         ret = ext4_journal_get_write_access(handle, inode->i_sb, iloc.bh,
 690                                             EXT4_JTR_NONE);
 691         if (ret)
 692                 goto out;
 693
 694         folio = __filemap_get_folio(mapping, 0, FGP_WRITEBEGIN | FGP_NOFS,
 695                                         mapping_gfp_mask(mapping));
 696         if (!folio) {
 697                 ret = -ENOMEM;
 698                 goto out;
 699         }
 700
 701         *pagep = &folio->page;
 702         down_read(&EXT4_I(inode)->xattr_sem);
 703         if (!ext4_has_inline_data(inode)) {
 704                 ret = 0;
 705                 folio_unlock(folio);
 706                 folio_put(folio);
 707                 goto out_up_read;
 708         }
 709
 710         if (!folio_test_uptodate(folio)) {
 711                 ret = ext4_read_inline_folio(inode, folio);
 712                 if (ret < 0) {
 713                         folio_unlock(folio);
 714                         folio_put(folio);
 715                         goto out_up_read;
 716                 }
 717         }
 718
 719         ret = 1;
 720         handle = NULL;
 721 out_up_read:
 722         up_read(&EXT4_I(inode)->xattr_sem);
 723 out:
 724         if (handle && (ret != 1))
 725                 ext4_journal_stop(handle);
 726         brelse(iloc.bh);
 727         return ret;
 728 convert:
 729         return ext4_convert_inline_data_to_extent(mapping, inode);
 730 }
 731
 732 int ext4_write_inline_data_end(struct inode *inode, loff_t pos, unsigned len,
 733                                unsigned copied, struct page *page)
 734 {
 735         struct folio *folio = page_folio(page);
 736         handle_t *handle = ext4_journal_current_handle();
 737         int no_expand;
 738         void *kaddr;
 739         struct ext4_iloc iloc;
 740         int ret = 0, ret2;
 741
 742         if (unlikely(copied < len) && !folio_test_uptodate(folio))
 743                 copied = 0;
 744
 745         if (likely(copied)) {
 746                 ret = ext4_get_inode_loc(inode, &iloc);
 747                 if (ret) {
 748                         folio_unlock(folio);
 749                         folio_put(folio);
 750                         ext4_std_error(inode->i_sb, ret);
 751                         goto out;
 752                 }
 753                 ext4_write_lock_xattr(inode, &no_expand);
 754                 BUG_ON(!ext4_has_inline_data(inode));
 755
 756                 /*
 757                  * ei->i_inline_off may have changed since
 758                  * ext4_write_begin() called
 759                  * ext4_try_to_write_inline_data()
 760                  */
 761                 (void) ext4_find_inline_data_nolock(inode);
 762
 763                 kaddr = kmap_local_folio(folio, 0);
 764                 ext4_write_inline_data(inode, &iloc, kaddr, pos, copied);
 765                 kunmap_local(kaddr);
 766                 folio_mark_uptodate(folio);
 767                 /* clear dirty flag so that writepages wouldn't work for us. */
 768                 folio_clear_dirty(folio);
 769
 770                 ext4_write_unlock_xattr(inode, &no_expand);
 771                 brelse(iloc.bh);
 772
 773                 /*
 774                  * It's important to update i_size while still holding folio
 775                  * lock: page writeout could otherwise come in and zero
 776                  * beyond i_size.
 777                  */
 778                 ext4_update_inode_size(inode, pos + copied);
 779         }
 780         folio_unlock(folio);
 781         folio_put(folio);
 782
 783         /*
 784          * Don't mark the inode dirty under folio lock. First, it unnecessarily
 785          * makes the holding time of folio lock longer. Second, it forces lock
 786          * ordering of folio lock and transaction start for journaling
 787          * filesystems.
 788          */
 789         if (likely(copied))
 790                 mark_inode_dirty(inode);
 791 out:
 792         /*
 793          * If we didn't copy as much data as expected, we need to trim back
 794          * size of xattr containing inline data.
 795          */
 796         if (pos + len > inode->i_size && ext4_can_truncate(inode))
 797                 ext4_orphan_add(handle, inode);
 798
 799         ret2 = ext4_journal_stop(handle);
 800         if (!ret)
 801                 ret = ret2;
 802         if (pos + len > inode->i_size) {
 803                 ext4_truncate_failed_write(inode);
 804                 /*
 805                  * If truncate failed early the inode might still be
 806                  * on the orphan list; we need to make sure the inode
 807                  * is removed from the orphan list in that case.
 808                  */
 809                 if (inode->i_nlink)
 810                         ext4_orphan_del(NULL, inode);
 811         }
 812         return ret ? ret : copied;
 813 }
 814
 815 struct buffer_head *
 816 ext4_journalled_write_inline_data(struct inode *inode,
 817                                   unsigned len,
 818                                   struct page *page)
 819 {
 820         int ret, no_expand;
 821         void *kaddr;
 822         struct ext4_iloc iloc;
 823
 824         ret = ext4_get_inode_loc(inode, &iloc);
 825         if (ret) {
 826                 ext4_std_error(inode->i_sb, ret);
 827                 return NULL;
 828         }
 829
 830         ext4_write_lock_xattr(inode, &no_expand);
 831         kaddr = kmap_atomic(page);
 832         ext4_write_inline_data(inode, &iloc, kaddr, 0, len);
 833         kunmap_atomic(kaddr);
 834         ext4_write_unlock_xattr(inode, &no_expand);
 835
 836         return iloc.bh;
 837 }
 838
 839 /*
 840  * Try to make the page cache and handle ready for the inline data case.
 841  * We can call this function in 2 cases:
 842  * 1. The inode is created and the first write exceeds inline size. We can
 843  *    clear the inode state safely.
 844  * 2. The inode has inline data, then we need to read the data, make it
 845  *    update and dirty so that ext4_da_writepages can handle it. We don't
 846  *    need to start the journal since the file's metadata isn't changed now.
 847  */
 848 static int ext4_da_convert_inline_data_to_extent(struct address_space *mapping,
 849                                                  struct inode *inode,
 850                                                  void **fsdata)
 851 {
 852         int ret = 0, inline_size;
 853         struct folio *folio;
 854
 855         folio = __filemap_get_folio(mapping, 0, FGP_WRITEBEGIN,
 856                                         mapping_gfp_mask(mapping));
 857         if (!folio)
 858                 return -ENOMEM;
 859
 860         down_read(&EXT4_I(inode)->xattr_sem);
 861         if (!ext4_has_inline_data(inode)) {
 862                 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
 863                 goto out;
 864         }
 865
 866         inline_size = ext4_get_inline_size(inode);
 867
 868         if (!folio_test_uptodate(folio)) {
 869                 ret = ext4_read_inline_folio(inode, folio);
 870                 if (ret < 0)
 871                         goto out;
 872         }
 873
 874         ret = __block_write_begin(&folio->page, 0, inline_size,
 875                                   ext4_da_get_block_prep);
 876         if (ret) {
 877                 up_read(&EXT4_I(inode)->xattr_sem);
 878                 folio_unlock(folio);
 879                 folio_put(folio);
 880                 ext4_truncate_failed_write(inode);
 881                 return ret;
 882         }
 883
 884         folio_mark_dirty(folio);
 885         folio_mark_uptodate(folio);
 886         ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
 887         *fsdata = (void *)CONVERT_INLINE_DATA;
 888
 889 out:
 890         up_read(&EXT4_I(inode)->xattr_sem);
 891         if (folio) {
 892                 folio_unlock(folio);
 893                 folio_put(folio);
 894         }
 895         return ret;
 896 }
 897
 898 /*
 899  * Prepare the write for the inline data.
 900  * If the data can be written into the inode, we just read
 901  * the page and make it uptodate, and start the journal.
 902  * Otherwise read the page, makes it dirty so that it can be
 903  * handle in writepages(the i_disksize update is left to the
 904  * normal ext4_da_write_end).
 905  */
 906 int ext4_da_write_inline_data_begin(struct address_space *mapping,
 907                                     struct inode *inode,
 908                                     loff_t pos, unsigned len,
 909                                     struct page **pagep,
 910                                     void **fsdata)
 911 {
 912         int ret;
 913         handle_t *handle;
 914         struct folio *folio;
 915         struct ext4_iloc iloc;
 916         int retries = 0;
 917
 918         ret = ext4_get_inode_loc(inode, &iloc);
 919         if (ret)
 920                 return ret;
 921
 922 retry_journal:
 923         handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
 924         if (IS_ERR(handle)) {
 925                 ret = PTR_ERR(handle);
 926                 goto out;
 927         }
 928
 929         ret = ext4_prepare_inline_data(handle, inode, pos + len);
 930         if (ret && ret != -ENOSPC)
 931                 goto out_journal;
 932
 933         if (ret == -ENOSPC) {
 934                 ext4_journal_stop(handle);
 935                 ret = ext4_da_convert_inline_data_to_extent(mapping,
 936                                                             inode,
 937                                                             fsdata);
 938                 if (ret == -ENOSPC &&
 939                     ext4_should_retry_alloc(inode->i_sb, &retries))
 940                         goto retry_journal;
 941                 goto out;
 942         }
 943
 944         /*
 945          * We cannot recurse into the filesystem as the transaction
 946          * is already started.
 947          */
 948         folio = __filemap_get_folio(mapping, 0, FGP_WRITEBEGIN | FGP_NOFS,
 949                                         mapping_gfp_mask(mapping));
 950         if (!folio) {
 951                 ret = -ENOMEM;
 952                 goto out_journal;
 953         }
 954
 955         down_read(&EXT4_I(inode)->xattr_sem);
 956         if (!ext4_has_inline_data(inode)) {
 957                 ret = 0;
 958                 goto out_release_page;
 959         }
 960
 961         if (!folio_test_uptodate(folio)) {
 962                 ret = ext4_read_inline_folio(inode, folio);
 963                 if (ret < 0)
 964                         goto out_release_page;
 965         }
 966         ret = ext4_journal_get_write_access(handle, inode->i_sb, iloc.bh,
 967                                             EXT4_JTR_NONE);
 968         if (ret)
 969                 goto out_release_page;
 970
 971         up_read(&EXT4_I(inode)->xattr_sem);
 972         *pagep = &folio->page;
 973         brelse(iloc.bh);
 974         return 1;
 975 out_release_page:
 976         up_read(&EXT4_I(inode)->xattr_sem);
 977         folio_unlock(folio);
 978         folio_put(folio);
 979 out_journal:
 980         ext4_journal_stop(handle);
 981 out:
 982         brelse(iloc.bh);
 983         return ret;
 984 }
 985
 986 #ifdef INLINE_DIR_DEBUG
 987 void ext4_show_inline_dir(struct inode *dir, struct buffer_head *bh,
 988                           void *inline_start, int inline_size)
 989 {
 990         int offset;
 991         unsigned short de_len;
 992         struct ext4_dir_entry_2 *de = inline_start;
 993         void *dlimit = inline_start + inline_size;
 994
 995         trace_printk("inode %lu\n", dir->i_ino);
 996         offset = 0;
 997         while ((void *)de < dlimit) {
 998                 de_len = ext4_rec_len_from_disk(de->rec_len, inline_size);
 999                 trace_printk("de: off %u rlen %u name %.*s nlen %u ino %u\n",
1000                              offset, de_len, de->name_len, de->name,
1001                              de->name_len, le32_to_cpu(de->inode));
1002                 if (ext4_check_dir_entry(dir, NULL, de, bh,
1003                                          inline_start, inline_size, offset))
1004                         BUG();
1005
1006                 offset += de_len;
1007                 de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
1008         }
1009 }
1010 #else
1011 #define ext4_show_inline_dir(dir, bh, inline_start, inline_size)
1012 #endif
1013
1014 /*
1015  * Add a new entry into a inline dir.
1016  * It will return -ENOSPC if no space is available, and -EIO
1017  * and -EEXIST if directory entry already exists.
1018  */
1019 static int ext4_add_dirent_to_inline(handle_t *handle,
1020                                      struct ext4_filename *fname,
1021                                      struct inode *dir,
1022                                      struct inode *inode,
1023                                      struct ext4_iloc *iloc,
1024                                      void *inline_start, int inline_size)
1025 {
1026         int             err;
1027         struct ext4_dir_entry_2 *de;
1028
1029         err = ext4_find_dest_de(dir, inode, iloc->bh, inline_start,
1030                                 inline_size, fname, &de);
1031         if (err)
1032                 return err;
1033
1034         BUFFER_TRACE(iloc->bh, "get_write_access");
1035         err = ext4_journal_get_write_access(handle, dir->i_sb, iloc->bh,
1036                                             EXT4_JTR_NONE);
1037         if (err)
1038                 return err;
1039         ext4_insert_dentry(dir, inode, de, inline_size, fname);
1040
1041         ext4_show_inline_dir(dir, iloc->bh, inline_start, inline_size);
1042
1043         /*
1044          * XXX shouldn't update any times until successful
1045          * completion of syscall, but too many callers depend
1046          * on this.
1047          *
1048          * XXX similarly, too many callers depend on
1049          * ext4_new_inode() setting the times, but error
1050          * recovery deletes the inode, so the worst that can
1051          * happen is that the times are slightly out of date
1052          * and/or different from the directory change time.
1053          */
1054         dir->i_mtime = dir->i_ctime = current_time(dir);
1055         ext4_update_dx_flag(dir);
1056         inode_inc_iversion(dir);
1057         return 1;
1058 }
1059
1060 static void *ext4_get_inline_xattr_pos(struct inode *inode,
1061                                        struct ext4_iloc *iloc)
1062 {
1063         struct ext4_xattr_entry *entry;
1064         struct ext4_xattr_ibody_header *header;
1065
1066         BUG_ON(!EXT4_I(inode)->i_inline_off);
1067
1068         header = IHDR(inode, ext4_raw_inode(iloc));
1069         entry = (struct ext4_xattr_entry *)((void *)ext4_raw_inode(iloc) +
1070                                             EXT4_I(inode)->i_inline_off);
1071
1072         return (void *)IFIRST(header) + le16_to_cpu(entry->e_value_offs);
1073 }
1074
1075 /* Set the final de to cover the whole block. */
1076 static void ext4_update_final_de(void *de_buf, int old_size, int new_size)
1077 {
1078         struct ext4_dir_entry_2 *de, *prev_de;
1079         void *limit;
1080         int de_len;
1081
1082         de = de_buf;
1083         if (old_size) {
1084                 limit = de_buf + old_size;
1085                 do {
1086                         prev_de = de;
1087                         de_len = ext4_rec_len_from_disk(de->rec_len, old_size);
1088                         de_buf += de_len;
1089                         de = de_buf;
1090                 } while (de_buf < limit);
1091
1092                 prev_de->rec_len = ext4_rec_len_to_disk(de_len + new_size -
1093                                                         old_size, new_size);
1094         } else {
1095                 /* this is just created, so create an empty entry. */
1096                 de->inode = 0;
1097                 de->rec_len = ext4_rec_len_to_disk(new_size, new_size);
1098         }
1099 }
1100
1101 static int ext4_update_inline_dir(handle_t *handle, struct inode *dir,
1102                                   struct ext4_iloc *iloc)
1103 {
1104         int ret;
1105         int old_size = EXT4_I(dir)->i_inline_size - EXT4_MIN_INLINE_DATA_SIZE;
1106         int new_size = get_max_inline_xattr_value_size(dir, iloc);
1107
1108         if (new_size - old_size <= ext4_dir_rec_len(1, NULL))
1109                 return -ENOSPC;
1110
1111         ret = ext4_update_inline_data(handle, dir,
1112                                       new_size + EXT4_MIN_INLINE_DATA_SIZE);
1113         if (ret)
1114                 return ret;
1115
1116         ext4_update_final_de(ext4_get_inline_xattr_pos(dir, iloc), old_size,
1117                              EXT4_I(dir)->i_inline_size -
1118                                                 EXT4_MIN_INLINE_DATA_SIZE);
1119         dir->i_size = EXT4_I(dir)->i_disksize = EXT4_I(dir)->i_inline_size;
1120         return 0;
1121 }
1122
1123 static void ext4_restore_inline_data(handle_t *handle, struct inode *inode,
1124                                      struct ext4_iloc *iloc,
1125                                      void *buf, int inline_size)
1126 {
1127         int ret;
1128
1129         ret = ext4_create_inline_data(handle, inode, inline_size);
1130         if (ret) {
1131                 ext4_msg(inode->i_sb, KERN_EMERG,
1132                         "error restoring inline_data for inode -- potential data loss! (inode %lu, error %d)",
1133                         inode->i_ino, ret);
1134                 return;
1135         }
1136         ext4_write_inline_data(inode, iloc, buf, 0, inline_size);
1137         ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
1138 }
1139
1140 static int ext4_finish_convert_inline_dir(handle_t *handle,
1141                                           struct inode *inode,
1142                                           struct buffer_head *dir_block,
1143                                           void *buf,
1144                                           int inline_size)
1145 {
1146         int err, csum_size = 0, header_size = 0;
1147         struct ext4_dir_entry_2 *de;
1148         void *target = dir_block->b_data;
1149
1150         /*
1151          * First create "." and ".." and then copy the dir information
1152          * back to the block.
1153          */
1154         de = target;
1155         de = ext4_init_dot_dotdot(inode, de,
1156                 inode->i_sb->s_blocksize, csum_size,
1157                 le32_to_cpu(((struct ext4_dir_entry_2 *)buf)->inode), 1);
1158         header_size = (void *)de - target;
1159
1160         memcpy((void *)de, buf + EXT4_INLINE_DOTDOT_SIZE,
1161                 inline_size - EXT4_INLINE_DOTDOT_SIZE);
1162
1163         if (ext4_has_metadata_csum(inode->i_sb))
1164                 csum_size = sizeof(struct ext4_dir_entry_tail);
1165
1166         inode->i_size = inode->i_sb->s_blocksize;
1167         i_size_write(inode, inode->i_sb->s_blocksize);
1168         EXT4_I(inode)->i_disksize = inode->i_sb->s_blocksize;
1169         ext4_update_final_de(dir_block->b_data,
1170                         inline_size - EXT4_INLINE_DOTDOT_SIZE + header_size,
1171                         inode->i_sb->s_blocksize - csum_size);
1172
1173         if (csum_size)
1174                 ext4_initialize_dirent_tail(dir_block,
1175                                             inode->i_sb->s_blocksize);
1176         set_buffer_uptodate(dir_block);
1177         err = ext4_handle_dirty_dirblock(handle, inode, dir_block);
1178         if (err)
1179                 return err;
1180         set_buffer_verified(dir_block);
1181         return ext4_mark_inode_dirty(handle, inode);
1182 }
1183
1184 static int ext4_convert_inline_data_nolock(handle_t *handle,
1185                                            struct inode *inode,
1186                                            struct ext4_iloc *iloc)
1187 {
1188         int error;
1189         void *buf = NULL;
1190         struct buffer_head *data_bh = NULL;
1191         struct ext4_map_blocks map;
1192         int inline_size;
1193
1194         inline_size = ext4_get_inline_size(inode);
1195         buf = kmalloc(inline_size, GFP_NOFS);
1196         if (!buf) {
1197                 error = -ENOMEM;
1198                 goto out;
1199         }
1200
1201         error = ext4_read_inline_data(inode, buf, inline_size, iloc);
1202         if (error < 0)
1203                 goto out;
1204
1205         /*
1206          * Make sure the inline directory entries pass checks before we try to
1207          * convert them, so that we avoid touching stuff that needs fsck.
1208          */
1209         if (S_ISDIR(inode->i_mode)) {
1210                 error = ext4_check_all_de(inode, iloc->bh,
1211                                         buf + EXT4_INLINE_DOTDOT_SIZE,
1212                                         inline_size - EXT4_INLINE_DOTDOT_SIZE);
1213                 if (error)
1214                         goto out;
1215         }
1216
1217         error = ext4_destroy_inline_data_nolock(handle, inode);
1218         if (error)
1219                 goto out;
1220
1221         map.m_lblk = 0;
1222         map.m_len = 1;
1223         map.m_flags = 0;
1224         error = ext4_map_blocks(handle, inode, &map, EXT4_GET_BLOCKS_CREATE);
1225         if (error < 0)
1226                 goto out_restore;
1227         if (!(map.m_flags & EXT4_MAP_MAPPED)) {
1228                 error = -EIO;
1229                 goto out_restore;
1230         }
1231
1232         data_bh = sb_getblk(inode->i_sb, map.m_pblk);
1233         if (!data_bh) {
1234                 error = -ENOMEM;
1235                 goto out_restore;
1236         }
1237
1238         lock_buffer(data_bh);
1239         error = ext4_journal_get_create_access(handle, inode->i_sb, data_bh,
1240                                                EXT4_JTR_NONE);
1241         if (error) {
1242                 unlock_buffer(data_bh);
1243                 error = -EIO;
1244                 goto out_restore;
1245         }
1246         memset(data_bh->b_data, 0, inode->i_sb->s_blocksize);
1247
1248         if (!S_ISDIR(inode->i_mode)) {
1249                 memcpy(data_bh->b_data, buf, inline_size);
1250                 set_buffer_uptodate(data_bh);
1251                 error = ext4_handle_dirty_metadata(handle,
1252                                                    inode, data_bh);
1253         } else {
1254                 error = ext4_finish_convert_inline_dir(handle, inode, data_bh,
1255                                                        buf, inline_size);
1256         }
1257
1258         unlock_buffer(data_bh);
1259 out_restore:
1260         if (error)
1261                 ext4_restore_inline_data(handle, inode, iloc, buf, inline_size);
1262
1263 out:
1264         brelse(data_bh);
1265         kfree(buf);
1266         return error;
1267 }
1268
1269 /*
1270  * Try to add the new entry to the inline data.
1271  * If succeeds, return 0. If not, extended the inline dir and copied data to
1272  * the new created block.
1273  */
1274 int ext4_try_add_inline_entry(handle_t *handle, struct ext4_filename *fname,
1275                               struct inode *dir, struct inode *inode)
1276 {
1277         int ret, ret2, inline_size, no_expand;
1278         void *inline_start;
1279         struct ext4_iloc iloc;
1280
1281         ret = ext4_get_inode_loc(dir, &iloc);
1282         if (ret)
1283                 return ret;
1284
1285         ext4_write_lock_xattr(dir, &no_expand);
1286         if (!ext4_has_inline_data(dir))
1287                 goto out;
1288
1289         inline_start = (void *)ext4_raw_inode(&iloc)->i_block +
1290                                                  EXT4_INLINE_DOTDOT_SIZE;
1291         inline_size = EXT4_MIN_INLINE_DATA_SIZE - EXT4_INLINE_DOTDOT_SIZE;
1292
1293         ret = ext4_add_dirent_to_inline(handle, fname, dir, inode, &iloc,
1294                                         inline_start, inline_size);
1295         if (ret != -ENOSPC)
1296                 goto out;
1297
1298         /* check whether it can be inserted to inline xattr space. */
1299         inline_size = EXT4_I(dir)->i_inline_size -
1300                         EXT4_MIN_INLINE_DATA_SIZE;
1301         if (!inline_size) {
1302                 /* Try to use the xattr space.*/
1303                 ret = ext4_update_inline_dir(handle, dir, &iloc);
1304                 if (ret && ret != -ENOSPC)
1305                         goto out;
1306
1307                 inline_size = EXT4_I(dir)->i_inline_size -
1308                                 EXT4_MIN_INLINE_DATA_SIZE;
1309         }
1310
1311         if (inline_size) {
1312                 inline_start = ext4_get_inline_xattr_pos(dir, &iloc);
1313
1314                 ret = ext4_add_dirent_to_inline(handle, fname, dir,
1315                                                 inode, &iloc, inline_start,
1316                                                 inline_size);
1317
1318                 if (ret != -ENOSPC)
1319                         goto out;
1320         }
1321
1322         /*
1323          * The inline space is filled up, so create a new block for it.
1324          * As the extent tree will be created, we have to save the inline
1325          * dir first.
1326          */
1327         ret = ext4_convert_inline_data_nolock(handle, dir, &iloc);
1328
1329 out:
1330         ext4_write_unlock_xattr(dir, &no_expand);
1331         ret2 = ext4_mark_inode_dirty(handle, dir);
1332         if (unlikely(ret2 && !ret))
1333                 ret = ret2;
1334         brelse(iloc.bh);
1335         return ret;
1336 }
1337
1338 /*
1339  * This function fills a red-black tree with information from an
1340  * inlined dir.  It returns the number directory entries loaded
1341  * into the tree.  If there is an error it is returned in err.
1342  */
1343 int ext4_inlinedir_to_tree(struct file *dir_file,
1344                            struct inode *dir, ext4_lblk_t block,
1345                            struct dx_hash_info *hinfo,
1346                            __u32 start_hash, __u32 start_minor_hash,
1347                            int *has_inline_data)
1348 {
1349         int err = 0, count = 0;
1350         unsigned int parent_ino;
1351         int pos;
1352         struct ext4_dir_entry_2 *de;
1353         struct inode *inode = file_inode(dir_file);
1354         int ret, inline_size = 0;
1355         struct ext4_iloc iloc;
1356         void *dir_buf = NULL;
1357         struct ext4_dir_entry_2 fake;
1358         struct fscrypt_str tmp_str;
1359
1360         ret = ext4_get_inode_loc(inode, &iloc);
1361         if (ret)
1362                 return ret;
1363
1364         down_read(&EXT4_I(inode)->xattr_sem);
1365         if (!ext4_has_inline_data(inode)) {
1366                 up_read(&EXT4_I(inode)->xattr_sem);
1367                 *has_inline_data = 0;
1368                 goto out;
1369         }
1370
1371         inline_size = ext4_get_inline_size(inode);
1372         dir_buf = kmalloc(inline_size, GFP_NOFS);
1373         if (!dir_buf) {
1374                 ret = -ENOMEM;
1375                 up_read(&EXT4_I(inode)->xattr_sem);
1376                 goto out;
1377         }
1378
1379         ret = ext4_read_inline_data(inode, dir_buf, inline_size, &iloc);
1380         up_read(&EXT4_I(inode)->xattr_sem);
1381         if (ret < 0)
1382                 goto out;
1383
1384         pos = 0;
1385         parent_ino = le32_to_cpu(((struct ext4_dir_entry_2 *)dir_buf)->inode);
1386         while (pos < inline_size) {
1387                 /*
1388                  * As inlined dir doesn't store any information about '.' and
1389                  * only the inode number of '..' is stored, we have to handle
1390                  * them differently.
1391                  */
1392                 if (pos == 0) {
1393                         fake.inode = cpu_to_le32(inode->i_ino);
1394                         fake.name_len = 1;
1395                         strcpy(fake.name, ".");
1396                         fake.rec_len = ext4_rec_len_to_disk(
1397                                           ext4_dir_rec_len(fake.name_len, NULL),
1398                                           inline_size);
1399                         ext4_set_de_type(inode->i_sb, &fake, S_IFDIR);
1400                         de = &fake;
1401                         pos = EXT4_INLINE_DOTDOT_OFFSET;
1402                 } else if (pos == EXT4_INLINE_DOTDOT_OFFSET) {
1403                         fake.inode = cpu_to_le32(parent_ino);
1404                         fake.name_len = 2;
1405                         strcpy(fake.name, "..");
1406                         fake.rec_len = ext4_rec_len_to_disk(
1407                                           ext4_dir_rec_len(fake.name_len, NULL),
1408                                           inline_size);
1409                         ext4_set_de_type(inode->i_sb, &fake, S_IFDIR);
1410                         de = &fake;
1411                         pos = EXT4_INLINE_DOTDOT_SIZE;
1412                 } else {
1413                         de = (struct ext4_dir_entry_2 *)(dir_buf + pos);
1414                         pos += ext4_rec_len_from_disk(de->rec_len, inline_size);
1415                         if (ext4_check_dir_entry(inode, dir_file, de,
1416                                          iloc.bh, dir_buf,
1417                                          inline_size, pos)) {
1418                                 ret = count;
1419                                 goto out;
1420                         }
1421                 }
1422
1423                 if (ext4_hash_in_dirent(dir)) {
1424                         hinfo->hash = EXT4_DIRENT_HASH(de);
1425                         hinfo->minor_hash = EXT4_DIRENT_MINOR_HASH(de);
1426                 } else {
1427                         ext4fs_dirhash(dir, de->name, de->name_len, hinfo);
1428                 }
1429                 if ((hinfo->hash < start_hash) ||
1430                     ((hinfo->hash == start_hash) &&
1431                      (hinfo->minor_hash < start_minor_hash)))
1432                         continue;
1433                 if (de->inode == 0)
1434                         continue;
1435                 tmp_str.name = de->name;
1436                 tmp_str.len = de->name_len;
1437                 err = ext4_htree_store_dirent(dir_file, hinfo->hash,
1438                                               hinfo->minor_hash, de, &tmp_str);
1439                 if (err) {
1440                         ret = err;
1441                         goto out;
1442                 }
1443                 count++;
1444         }
1445         ret = count;
1446 out:
1447         kfree(dir_buf);
1448         brelse(iloc.bh);
1449         return ret;
1450 }
1451
1452 /*
1453  * So this function is called when the volume is mkfsed with
1454  * dir_index disabled. In order to keep f_pos persistent
1455  * after we convert from an inlined dir to a blocked based,
1456  * we just pretend that we are a normal dir and return the
1457  * offset as if '.' and '..' really take place.
1458  *
1459  */
1460 int ext4_read_inline_dir(struct file *file,
1461                          struct dir_context *ctx,
1462                          int *has_inline_data)
1463 {
1464         unsigned int offset, parent_ino;
1465         int i;
1466         struct ext4_dir_entry_2 *de;
1467         struct super_block *sb;
1468         struct inode *inode = file_inode(file);
1469         int ret, inline_size = 0;
1470         struct ext4_iloc iloc;
1471         void *dir_buf = NULL;
1472         int dotdot_offset, dotdot_size, extra_offset, extra_size;
1473
1474         ret = ext4_get_inode_loc(inode, &iloc);
1475         if (ret)
1476                 return ret;
1477
1478         down_read(&EXT4_I(inode)->xattr_sem);
1479         if (!ext4_has_inline_data(inode)) {
1480                 up_read(&EXT4_I(inode)->xattr_sem);
1481                 *has_inline_data = 0;
1482                 goto out;
1483         }
1484
1485         inline_size = ext4_get_inline_size(inode);
1486         dir_buf = kmalloc(inline_size, GFP_NOFS);
1487         if (!dir_buf) {
1488                 ret = -ENOMEM;
1489                 up_read(&EXT4_I(inode)->xattr_sem);
1490                 goto out;
1491         }
1492
1493         ret = ext4_read_inline_data(inode, dir_buf, inline_size, &iloc);
1494         up_read(&EXT4_I(inode)->xattr_sem);
1495         if (ret < 0)
1496                 goto out;
1497
1498         ret = 0;
1499         sb = inode->i_sb;
1500         parent_ino = le32_to_cpu(((struct ext4_dir_entry_2 *)dir_buf)->inode);
1501         offset = ctx->pos;
1502
1503         /*
1504          * dotdot_offset and dotdot_size is the real offset and
1505          * size for ".." and "." if the dir is block based while
1506          * the real size for them are only EXT4_INLINE_DOTDOT_SIZE.
1507          * So we will use extra_offset and extra_size to indicate them
1508          * during the inline dir iteration.
1509          */
1510         dotdot_offset = ext4_dir_rec_len(1, NULL);
1511         dotdot_size = dotdot_offset + ext4_dir_rec_len(2, NULL);
1512         extra_offset = dotdot_size - EXT4_INLINE_DOTDOT_SIZE;
1513         extra_size = extra_offset + inline_size;
1514
1515         /*
1516          * If the version has changed since the last call to
1517          * readdir(2), then we might be pointing to an invalid
1518          * dirent right now.  Scan from the start of the inline
1519          * dir to make sure.
1520          */
1521         if (!inode_eq_iversion(inode, file->f_version)) {
1522                 for (i = 0; i < extra_size && i < offset;) {
1523                         /*
1524                          * "." is with offset 0 and
1525                          * ".." is dotdot_offset.
1526                          */
1527                         if (!i) {
1528                                 i = dotdot_offset;
1529                                 continue;
1530                         } else if (i == dotdot_offset) {
1531                                 i = dotdot_size;
1532                                 continue;
1533                         }
1534                         /* for other entry, the real offset in
1535                          * the buf has to be tuned accordingly.
1536                          */
1537                         de = (struct ext4_dir_entry_2 *)
1538                                 (dir_buf + i - extra_offset);
1539                         /* It's too expensive to do a full
1540                          * dirent test each time round this
1541                          * loop, but we do have to test at
1542                          * least that it is non-zero.  A
1543                          * failure will be detected in the
1544                          * dirent test below. */
1545                         if (ext4_rec_len_from_disk(de->rec_len, extra_size)
1546                                 < ext4_dir_rec_len(1, NULL))
1547                                 break;
1548                         i += ext4_rec_len_from_disk(de->rec_len,
1549                                                     extra_size);
1550                 }
1551                 offset = i;
1552                 ctx->pos = offset;
1553                 file->f_version = inode_query_iversion(inode);
1554         }
1555
1556         while (ctx->pos < extra_size) {
1557                 if (ctx->pos == 0) {
1558                         if (!dir_emit(ctx, ".", 1, inode->i_ino, DT_DIR))
1559                                 goto out;
1560                         ctx->pos = dotdot_offset;
1561                         continue;
1562                 }
1563
1564                 if (ctx->pos == dotdot_offset) {
1565                         if (!dir_emit(ctx, "..", 2, parent_ino, DT_DIR))
1566                                 goto out;
1567                         ctx->pos = dotdot_size;
1568                         continue;
1569                 }
1570
1571                 de = (struct ext4_dir_entry_2 *)
1572                         (dir_buf + ctx->pos - extra_offset);
1573                 if (ext4_check_dir_entry(inode, file, de, iloc.bh, dir_buf,
1574                                          extra_size, ctx->pos))
1575                         goto out;
1576                 if (le32_to_cpu(de->inode)) {
1577                         if (!dir_emit(ctx, de->name, de->name_len,
1578                                       le32_to_cpu(de->inode),
1579                                       get_dtype(sb, de->file_type)))
1580                                 goto out;
1581                 }
1582                 ctx->pos += ext4_rec_len_from_disk(de->rec_len, extra_size);
1583         }
1584 out:
1585         kfree(dir_buf);
1586         brelse(iloc.bh);
1587         return ret;
1588 }
1589
1590 void *ext4_read_inline_link(struct inode *inode)
1591 {
1592         struct ext4_iloc iloc;
1593         int ret, inline_size;
1594         void *link;
1595
1596         ret = ext4_get_inode_loc(inode, &iloc);
1597         if (ret)
1598                 return ERR_PTR(ret);
1599
1600         ret = -ENOMEM;
1601         inline_size = ext4_get_inline_size(inode);
1602         link = kmalloc(inline_size + 1, GFP_NOFS);
1603         if (!link)
1604                 goto out;
1605
1606         ret = ext4_read_inline_data(inode, link, inline_size, &iloc);
1607         if (ret < 0) {
1608                 kfree(link);
1609                 goto out;
1610         }
1611         nd_terminate_link(link, inode->i_size, ret);
1612 out:
1613         if (ret < 0)
1614                 link = ERR_PTR(ret);
1615         brelse(iloc.bh);
1616         return link;
1617 }
1618
1619 struct buffer_head *ext4_get_first_inline_block(struct inode *inode,
1620                                         struct ext4_dir_entry_2 **parent_de,
1621                                         int *retval)
1622 {
1623         struct ext4_iloc iloc;
1624
1625         *retval = ext4_get_inode_loc(inode, &iloc);
1626         if (*retval)
1627                 return NULL;
1628
1629         *parent_de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block;
1630
1631         return iloc.bh;
1632 }
1633
1634 /*
1635  * Try to create the inline data for the new dir.
1636  * If it succeeds, return 0, otherwise return the error.
1637  * In case of ENOSPC, the caller should create the normal disk layout dir.
1638  */
1639 int ext4_try_create_inline_dir(handle_t *handle, struct inode *parent,
1640                                struct inode *inode)
1641 {
1642         int ret, inline_size = EXT4_MIN_INLINE_DATA_SIZE;
1643         struct ext4_iloc iloc;
1644         struct ext4_dir_entry_2 *de;
1645
1646         ret = ext4_get_inode_loc(inode, &iloc);
1647         if (ret)
1648                 return ret;
1649
1650         ret = ext4_prepare_inline_data(handle, inode, inline_size);
1651         if (ret)
1652                 goto out;
1653
1654         /*
1655          * For inline dir, we only save the inode information for the ".."
1656          * and create a fake dentry to cover the left space.
1657          */
1658         de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block;
1659         de->inode = cpu_to_le32(parent->i_ino);
1660         de = (struct ext4_dir_entry_2 *)((void *)de + EXT4_INLINE_DOTDOT_SIZE);
1661         de->inode = 0;
1662         de->rec_len = ext4_rec_len_to_disk(
1663                                 inline_size - EXT4_INLINE_DOTDOT_SIZE,
1664                                 inline_size);
1665         set_nlink(inode, 2);
1666         inode->i_size = EXT4_I(inode)->i_disksize = inline_size;
1667 out:
1668         brelse(iloc.bh);
1669         return ret;
1670 }
1671
1672 struct buffer_head *ext4_find_inline_entry(struct inode *dir,
1673                                         struct ext4_filename *fname,
1674                                         struct ext4_dir_entry_2 **res_dir,
1675                                         int *has_inline_data)
1676 {
1677         int ret;
1678         struct ext4_iloc iloc;
1679         void *inline_start;
1680         int inline_size;
1681
1682         if (ext4_get_inode_loc(dir, &iloc))
1683                 return NULL;
1684
1685         down_read(&EXT4_I(dir)->xattr_sem);
1686         if (!ext4_has_inline_data(dir)) {
1687                 *has_inline_data = 0;
1688                 goto out;
1689         }
1690
1691         inline_start = (void *)ext4_raw_inode(&iloc)->i_block +
1692                                                 EXT4_INLINE_DOTDOT_SIZE;
1693         inline_size = EXT4_MIN_INLINE_DATA_SIZE - EXT4_INLINE_DOTDOT_SIZE;
1694         ret = ext4_search_dir(iloc.bh, inline_start, inline_size,
1695                               dir, fname, 0, res_dir);
1696         if (ret == 1)
1697                 goto out_find;
1698         if (ret < 0)
1699                 goto out;
1700
1701         if (ext4_get_inline_size(dir) == EXT4_MIN_INLINE_DATA_SIZE)
1702                 goto out;
1703
1704         inline_start = ext4_get_inline_xattr_pos(dir, &iloc);
1705         inline_size = ext4_get_inline_size(dir) - EXT4_MIN_INLINE_DATA_SIZE;
1706
1707         ret = ext4_search_dir(iloc.bh, inline_start, inline_size,
1708                               dir, fname, 0, res_dir);
1709         if (ret == 1)
1710                 goto out_find;
1711
1712 out:
1713         brelse(iloc.bh);
1714         iloc.bh = NULL;
1715 out_find:
1716         up_read(&EXT4_I(dir)->xattr_sem);
1717         return iloc.bh;
1718 }
1719
1720 int ext4_delete_inline_entry(handle_t *handle,
1721                              struct inode *dir,
1722                              struct ext4_dir_entry_2 *de_del,
1723                              struct buffer_head *bh,
1724                              int *has_inline_data)
1725 {
1726         int err, inline_size, no_expand;
1727         struct ext4_iloc iloc;
1728         void *inline_start;
1729
1730         err = ext4_get_inode_loc(dir, &iloc);
1731         if (err)
1732                 return err;
1733
1734         ext4_write_lock_xattr(dir, &no_expand);
1735         if (!ext4_has_inline_data(dir)) {
1736                 *has_inline_data = 0;
1737                 goto out;
1738         }
1739
1740         if ((void *)de_del - ((void *)ext4_raw_inode(&iloc)->i_block) <
1741                 EXT4_MIN_INLINE_DATA_SIZE) {
1742                 inline_start = (void *)ext4_raw_inode(&iloc)->i_block +
1743                                         EXT4_INLINE_DOTDOT_SIZE;
1744                 inline_size = EXT4_MIN_INLINE_DATA_SIZE -
1745                                 EXT4_INLINE_DOTDOT_SIZE;
1746         } else {
1747                 inline_start = ext4_get_inline_xattr_pos(dir, &iloc);
1748                 inline_size = ext4_get_inline_size(dir) -
1749                                 EXT4_MIN_INLINE_DATA_SIZE;
1750         }
1751
1752         BUFFER_TRACE(bh, "get_write_access");
1753         err = ext4_journal_get_write_access(handle, dir->i_sb, bh,
1754                                             EXT4_JTR_NONE);
1755         if (err)
1756                 goto out;
1757
1758         err = ext4_generic_delete_entry(dir, de_del, bh,
1759                                         inline_start, inline_size, 0);
1760         if (err)
1761                 goto out;
1762
1763         ext4_show_inline_dir(dir, iloc.bh, inline_start, inline_size);
1764 out:
1765         ext4_write_unlock_xattr(dir, &no_expand);
1766         if (likely(err == 0))
1767                 err = ext4_mark_inode_dirty(handle, dir);
1768         brelse(iloc.bh);
1769         if (err != -ENOENT)
1770                 ext4_std_error(dir->i_sb, err);
1771         return err;
1772 }
1773
1774 /*
1775  * Get the inline dentry at offset.
1776  */
1777 static inline struct ext4_dir_entry_2 *
1778 ext4_get_inline_entry(struct inode *inode,
1779                       struct ext4_iloc *iloc,
1780                       unsigned int offset,
1781                       void **inline_start,
1782                       int *inline_size)
1783 {
1784         void *inline_pos;
1785
1786         BUG_ON(offset > ext4_get_inline_size(inode));
1787
1788         if (offset < EXT4_MIN_INLINE_DATA_SIZE) {
1789                 inline_pos = (void *)ext4_raw_inode(iloc)->i_block;
1790                 *inline_size = EXT4_MIN_INLINE_DATA_SIZE;
1791         } else {
1792                 inline_pos = ext4_get_inline_xattr_pos(inode, iloc);
1793                 offset -= EXT4_MIN_INLINE_DATA_SIZE;
1794                 *inline_size = ext4_get_inline_size(inode) -
1795                                 EXT4_MIN_INLINE_DATA_SIZE;
1796         }
1797
1798         if (inline_start)
1799                 *inline_start = inline_pos;
1800         return (struct ext4_dir_entry_2 *)(inline_pos + offset);
1801 }
1802
1803 bool empty_inline_dir(struct inode *dir, int *has_inline_data)
1804 {
1805         int err, inline_size;
1806         struct ext4_iloc iloc;
1807         size_t inline_len;
1808         void *inline_pos;
1809         unsigned int offset;
1810         struct ext4_dir_entry_2 *de;
1811         bool ret = false;
1812
1813         err = ext4_get_inode_loc(dir, &iloc);
1814         if (err) {
1815                 EXT4_ERROR_INODE_ERR(dir, -err,
1816                                      "error %d getting inode %lu block",
1817                                      err, dir->i_ino);
1818                 return false;
1819         }
1820
1821         down_read(&EXT4_I(dir)->xattr_sem);
1822         if (!ext4_has_inline_data(dir)) {
1823                 *has_inline_data = 0;
1824                 ret = true;
1825                 goto out;
1826         }
1827
1828         de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block;
1829         if (!le32_to_cpu(de->inode)) {
1830                 ext4_warning(dir->i_sb,
1831                              "bad inline directory (dir #%lu) - no `..'",
1832                              dir->i_ino);
1833                 goto out;
1834         }
1835
1836         inline_len = ext4_get_inline_size(dir);
1837         offset = EXT4_INLINE_DOTDOT_SIZE;
1838         while (offset < inline_len) {
1839                 de = ext4_get_inline_entry(dir, &iloc, offset,
1840                                            &inline_pos, &inline_size);
1841                 if (ext4_check_dir_entry(dir, NULL, de,
1842                                          iloc.bh, inline_pos,
1843                                          inline_size, offset)) {
1844                         ext4_warning(dir->i_sb,
1845                                      "bad inline directory (dir #%lu) - "
1846                                      "inode %u, rec_len %u, name_len %d"
1847                                      "inline size %d",
1848                                      dir->i_ino, le32_to_cpu(de->inode),
1849                                      le16_to_cpu(de->rec_len), de->name_len,
1850                                      inline_size);
1851                         goto out;
1852                 }
1853                 if (le32_to_cpu(de->inode)) {
1854                         goto out;
1855                 }
1856                 offset += ext4_rec_len_from_disk(de->rec_len, inline_size);
1857         }
1858
1859         ret = true;
1860 out:
1861         up_read(&EXT4_I(dir)->xattr_sem);
1862         brelse(iloc.bh);
1863         return ret;
1864 }
1865
1866 int ext4_destroy_inline_data(handle_t *handle, struct inode *inode)
1867 {
1868         int ret, no_expand;
1869
1870         ext4_write_lock_xattr(inode, &no_expand);
1871         ret = ext4_destroy_inline_data_nolock(handle, inode);
1872         ext4_write_unlock_xattr(inode, &no_expand);
1873
1874         return ret;
1875 }
1876
1877 int ext4_inline_data_iomap(struct inode *inode, struct iomap *iomap)
1878 {
1879         __u64 addr;
1880         int error = -EAGAIN;
1881         struct ext4_iloc iloc;
1882
1883         down_read(&EXT4_I(inode)->xattr_sem);
1884         if (!ext4_has_inline_data(inode))
1885                 goto out;
1886
1887         error = ext4_get_inode_loc(inode, &iloc);
1888         if (error)
1889                 goto out;
1890
1891         addr = (__u64)iloc.bh->b_blocknr << inode->i_sb->s_blocksize_bits;
1892         addr += (char *)ext4_raw_inode(&iloc) - iloc.bh->b_data;
1893         addr += offsetof(struct ext4_inode, i_block);
1894
1895         brelse(iloc.bh);
1896
1897         iomap->addr = addr;
1898         iomap->offset = 0;
1899         iomap->length = min_t(loff_t, ext4_get_inline_size(inode),
1900                               i_size_read(inode));
1901         iomap->type = IOMAP_INLINE;
1902         iomap->flags = 0;
1903
1904 out:
1905         up_read(&EXT4_I(inode)->xattr_sem);
1906         return error;
1907 }
1908
1909 int ext4_inline_data_truncate(struct inode *inode, int *has_inline)
1910 {
1911         handle_t *handle;
1912         int inline_size, value_len, needed_blocks, no_expand, err = 0;
1913         size_t i_size;
1914         void *value = NULL;
1915         struct ext4_xattr_ibody_find is = {
1916                 .s = { .not_found = -ENODATA, },
1917         };
1918         struct ext4_xattr_info i = {
1919                 .name_index = EXT4_XATTR_INDEX_SYSTEM,
1920                 .name = EXT4_XATTR_SYSTEM_DATA,
1921         };
1922
1923
1924         needed_blocks = ext4_writepage_trans_blocks(inode);
1925         handle = ext4_journal_start(inode, EXT4_HT_INODE, needed_blocks);
1926         if (IS_ERR(handle))
1927                 return PTR_ERR(handle);
1928
1929         ext4_write_lock_xattr(inode, &no_expand);
1930         if (!ext4_has_inline_data(inode)) {
1931                 ext4_write_unlock_xattr(inode, &no_expand);
1932                 *has_inline = 0;
1933                 ext4_journal_stop(handle);
1934                 return 0;
1935         }
1936
1937         if ((err = ext4_orphan_add(handle, inode)) != 0)
1938                 goto out;
1939
1940         if ((err = ext4_get_inode_loc(inode, &is.iloc)) != 0)
1941                 goto out;
1942
1943         down_write(&EXT4_I(inode)->i_data_sem);
1944         i_size = inode->i_size;
1945         inline_size = ext4_get_inline_size(inode);
1946         EXT4_I(inode)->i_disksize = i_size;
1947
1948         if (i_size < inline_size) {
1949                 /*
1950                  * if there's inline data to truncate and this file was
1951                  * converted to extents after that inline data was written,
1952                  * the extent status cache must be cleared to avoid leaving
1953                  * behind stale delayed allocated extent entries
1954                  */
1955                 if (!ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
1956 retry:
1957                         err = ext4_es_remove_extent(inode, 0, EXT_MAX_BLOCKS);
1958                         if (err == -ENOMEM) {
1959                                 memalloc_retry_wait(GFP_ATOMIC);
1960                                 goto retry;
1961                         }
1962                         if (err)
1963                                 goto out_error;
1964                 }
1965
1966                 /* Clear the content in the xattr space. */
1967                 if (inline_size > EXT4_MIN_INLINE_DATA_SIZE) {
1968                         if ((err = ext4_xattr_ibody_find(inode, &i, &is)) != 0)
1969                                 goto out_error;
1970
1971                         BUG_ON(is.s.not_found);
1972
1973                         value_len = le32_to_cpu(is.s.here->e_value_size);
1974                         value = kmalloc(value_len, GFP_NOFS);
1975                         if (!value) {
1976                                 err = -ENOMEM;
1977                                 goto out_error;
1978                         }
1979
1980                         err = ext4_xattr_ibody_get(inode, i.name_index,
1981                                                    i.name, value, value_len);
1982                         if (err <= 0)
1983                                 goto out_error;
1984
1985                         i.value = value;
1986                         i.value_len = i_size > EXT4_MIN_INLINE_DATA_SIZE ?
1987                                         i_size - EXT4_MIN_INLINE_DATA_SIZE : 0;
1988                         err = ext4_xattr_ibody_set(handle, inode, &i, &is);
1989                         if (err)
1990                                 goto out_error;
1991                 }
1992
1993                 /* Clear the content within i_blocks. */
1994                 if (i_size < EXT4_MIN_INLINE_DATA_SIZE) {
1995                         void *p = (void *) ext4_raw_inode(&is.iloc)->i_block;
1996                         memset(p + i_size, 0,
1997                                EXT4_MIN_INLINE_DATA_SIZE - i_size);
1998                 }
1999
2000                 EXT4_I(inode)->i_inline_size = i_size <
2001                                         EXT4_MIN_INLINE_DATA_SIZE ?
2002                                         EXT4_MIN_INLINE_DATA_SIZE : i_size;
2003         }
2004
2005 out_error:
2006         up_write(&EXT4_I(inode)->i_data_sem);
2007 out:
2008         brelse(is.iloc.bh);
2009         ext4_write_unlock_xattr(inode, &no_expand);
2010         kfree(value);
2011         if (inode->i_nlink)
2012                 ext4_orphan_del(handle, inode);
2013
2014         if (err == 0) {
2015                 inode->i_mtime = inode->i_ctime = current_time(inode);
2016                 err = ext4_mark_inode_dirty(handle, inode);
2017                 if (IS_SYNC(inode))
2018                         ext4_handle_sync(handle);
2019         }
2020         ext4_journal_stop(handle);
2021         return err;
2022 }
2023
2024 int ext4_convert_inline_data(struct inode *inode)
2025 {
2026         int error, needed_blocks, no_expand;
2027         handle_t *handle;
2028         struct ext4_iloc iloc;
2029
2030         if (!ext4_has_inline_data(inode)) {
2031                 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
2032                 return 0;
2033         } else if (!ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2034                 /*
2035                  * Inode has inline data but EXT4_STATE_MAY_INLINE_DATA is
2036                  * cleared. This means we are in the middle of moving of
2037                  * inline data to delay allocated block. Just force writeout
2038                  * here to finish conversion.
2039                  */
2040                 error = filemap_flush(inode->i_mapping);
2041                 if (error)
2042                         return error;
2043                 if (!ext4_has_inline_data(inode))
2044                         return 0;
2045         }
2046
2047         needed_blocks = ext4_writepage_trans_blocks(inode);
2048
2049         iloc.bh = NULL;
2050         error = ext4_get_inode_loc(inode, &iloc);
2051         if (error)
2052                 return error;
2053
2054         handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
2055         if (IS_ERR(handle)) {
2056                 error = PTR_ERR(handle);
2057                 goto out_free;
2058         }
2059
2060         ext4_write_lock_xattr(inode, &no_expand);
2061         if (ext4_has_inline_data(inode))
2062                 error = ext4_convert_inline_data_nolock(handle, inode, &iloc);
2063         ext4_write_unlock_xattr(inode, &no_expand);
2064         ext4_journal_stop(handle);
2065 out_free:
2066         brelse(iloc.bh);
2067         return error;
2068 }