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_page(struct inode *inode, struct page *page)
 471 {
 472         void *kaddr;
 473         int ret = 0;
 474         size_t len;
 475         struct ext4_iloc iloc;
 476
 477         BUG_ON(!PageLocked(page));
 478         BUG_ON(!ext4_has_inline_data(inode));
 479         BUG_ON(page->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         kaddr = kmap_atomic(page);
 493         ret = ext4_read_inline_data(inode, kaddr, len, &iloc);
 494         flush_dcache_page(page);
 495         kunmap_atomic(kaddr);
 496         zero_user_segment(page, len, PAGE_SIZE);
 497         SetPageUptodate(page);
 498         brelse(iloc.bh);
 499
 500 out:
 501         return ret;
 502 }
 503
 504 int ext4_readpage_inline(struct inode *inode, struct page *page)
 505 {
 506         int ret = 0;
 507
 508         down_read(&EXT4_I(inode)->xattr_sem);
 509         if (!ext4_has_inline_data(inode)) {
 510                 up_read(&EXT4_I(inode)->xattr_sem);
 511                 return -EAGAIN;
 512         }
 513
 514         /*
 515          * Current inline data can only exist in the 1st page,
 516          * So for all the other pages, just set them uptodate.
 517          */
 518         if (!page->index)
 519                 ret = ext4_read_inline_page(inode, page);
 520         else if (!PageUptodate(page)) {
 521                 zero_user_segment(page, 0, PAGE_SIZE);
 522                 SetPageUptodate(page);
 523         }
 524
 525         up_read(&EXT4_I(inode)->xattr_sem);
 526
 527         unlock_page(page);
 528         return ret >= 0 ? 0 : ret;
 529 }
 530
 531 static int ext4_convert_inline_data_to_extent(struct address_space *mapping,
 532                                               struct inode *inode)
 533 {
 534         int ret, needed_blocks, no_expand;
 535         handle_t *handle = NULL;
 536         int retries = 0, sem_held = 0;
 537         struct page *page = NULL;
 538         unsigned int flags;
 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         flags = memalloc_nofs_save();
 568         page = grab_cache_page_write_begin(mapping, 0);
 569         memalloc_nofs_restore(flags);
 570         if (!page) {
 571                 ret = -ENOMEM;
 572                 goto out;
 573         }
 574
 575         ext4_write_lock_xattr(inode, &no_expand);
 576         sem_held = 1;
 577         /* If some one has already done this for us, just exit. */
 578         if (!ext4_has_inline_data(inode)) {
 579                 ret = 0;
 580                 goto out;
 581         }
 582
 583         from = 0;
 584         to = ext4_get_inline_size(inode);
 585         if (!PageUptodate(page)) {
 586                 ret = ext4_read_inline_page(inode, page);
 587                 if (ret < 0)
 588                         goto out;
 589         }
 590
 591         ret = ext4_destroy_inline_data_nolock(handle, inode);
 592         if (ret)
 593                 goto out;
 594
 595         if (ext4_should_dioread_nolock(inode)) {
 596                 ret = __block_write_begin(page, from, to,
 597                                           ext4_get_block_unwritten);
 598         } else
 599                 ret = __block_write_begin(page, from, to, ext4_get_block);
 600
 601         if (!ret && ext4_should_journal_data(inode)) {
 602                 ret = ext4_walk_page_buffers(handle, inode, page_buffers(page),
 603                                              from, to, NULL,
 604                                              do_journal_get_write_access);
 605         }
 606
 607         if (ret) {
 608                 unlock_page(page);
 609                 put_page(page);
 610                 page = NULL;
 611                 ext4_orphan_add(handle, inode);
 612                 ext4_write_unlock_xattr(inode, &no_expand);
 613                 sem_held = 0;
 614                 ext4_journal_stop(handle);
 615                 handle = NULL;
 616                 ext4_truncate_failed_write(inode);
 617                 /*
 618                  * If truncate failed early the inode might
 619                  * still be on the orphan list; we need to
 620                  * make sure the inode is removed from the
 621                  * orphan list in that case.
 622                  */
 623                 if (inode->i_nlink)
 624                         ext4_orphan_del(NULL, inode);
 625         }
 626
 627         if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
 628                 goto retry;
 629
 630         if (page)
 631                 block_commit_write(page, from, to);
 632 out:
 633         if (page) {
 634                 unlock_page(page);
 635                 put_page(page);
 636         }
 637         if (sem_held)
 638                 ext4_write_unlock_xattr(inode, &no_expand);
 639         if (handle)
 640                 ext4_journal_stop(handle);
 641         brelse(iloc.bh);
 642         return ret;
 643 }
 644
 645 /*
 646  * Try to write data in the inode.
 647  * If the inode has inline data, check whether the new write can be
 648  * in the inode also. If not, create the page the handle, move the data
 649  * to the page make it update and let the later codes create extent for it.
 650  */
 651 int ext4_try_to_write_inline_data(struct address_space *mapping,
 652                                   struct inode *inode,
 653                                   loff_t pos, unsigned len,
 654                                   struct page **pagep)
 655 {
 656         int ret;
 657         handle_t *handle;
 658         unsigned int flags;
 659         struct page *page;
 660         struct ext4_iloc iloc;
 661
 662         if (pos + len > ext4_get_max_inline_size(inode))
 663                 goto convert;
 664
 665         ret = ext4_get_inode_loc(inode, &iloc);
 666         if (ret)
 667                 return ret;
 668
 669         /*
 670          * The possible write could happen in the inode,
 671          * so try to reserve the space in inode first.
 672          */
 673         handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
 674         if (IS_ERR(handle)) {
 675                 ret = PTR_ERR(handle);
 676                 handle = NULL;
 677                 goto out;
 678         }
 679
 680         ret = ext4_prepare_inline_data(handle, inode, pos + len);
 681         if (ret && ret != -ENOSPC)
 682                 goto out;
 683
 684         /* We don't have space in inline inode, so convert it to extent. */
 685         if (ret == -ENOSPC) {
 686                 ext4_journal_stop(handle);
 687                 brelse(iloc.bh);
 688                 goto convert;
 689         }
 690
 691         ret = ext4_journal_get_write_access(handle, inode->i_sb, iloc.bh,
 692                                             EXT4_JTR_NONE);
 693         if (ret)
 694                 goto out;
 695
 696         flags = memalloc_nofs_save();
 697         page = grab_cache_page_write_begin(mapping, 0);
 698         memalloc_nofs_restore(flags);
 699         if (!page) {
 700                 ret = -ENOMEM;
 701                 goto out;
 702         }
 703
 704         *pagep = page;
 705         down_read(&EXT4_I(inode)->xattr_sem);
 706         if (!ext4_has_inline_data(inode)) {
 707                 ret = 0;
 708                 unlock_page(page);
 709                 put_page(page);
 710                 goto out_up_read;
 711         }
 712
 713         if (!PageUptodate(page)) {
 714                 ret = ext4_read_inline_page(inode, page);
 715                 if (ret < 0) {
 716                         unlock_page(page);
 717                         put_page(page);
 718                         goto out_up_read;
 719                 }
 720         }
 721
 722         ret = 1;
 723         handle = NULL;
 724 out_up_read:
 725         up_read(&EXT4_I(inode)->xattr_sem);
 726 out:
 727         if (handle && (ret != 1))
 728                 ext4_journal_stop(handle);
 729         brelse(iloc.bh);
 730         return ret;
 731 convert:
 732         return ext4_convert_inline_data_to_extent(mapping, inode);
 733 }
 734
 735 int ext4_write_inline_data_end(struct inode *inode, loff_t pos, unsigned len,
 736                                unsigned copied, struct page *page)
 737 {
 738         handle_t *handle = ext4_journal_current_handle();
 739         int no_expand;
 740         void *kaddr;
 741         struct ext4_iloc iloc;
 742         int ret = 0, ret2;
 743
 744         if (unlikely(copied < len) && !PageUptodate(page))
 745                 copied = 0;
 746
 747         if (likely(copied)) {
 748                 ret = ext4_get_inode_loc(inode, &iloc);
 749                 if (ret) {
 750                         unlock_page(page);
 751                         put_page(page);
 752                         ext4_std_error(inode->i_sb, ret);
 753                         goto out;
 754                 }
 755                 ext4_write_lock_xattr(inode, &no_expand);
 756                 BUG_ON(!ext4_has_inline_data(inode));
 757
 758                 /*
 759                  * ei->i_inline_off may have changed since
 760                  * ext4_write_begin() called
 761                  * ext4_try_to_write_inline_data()
 762                  */
 763                 (void) ext4_find_inline_data_nolock(inode);
 764
 765                 kaddr = kmap_atomic(page);
 766                 ext4_write_inline_data(inode, &iloc, kaddr, pos, copied);
 767                 kunmap_atomic(kaddr);
 768                 SetPageUptodate(page);
 769                 /* clear page dirty so that writepages wouldn't work for us. */
 770                 ClearPageDirty(page);
 771
 772                 ext4_write_unlock_xattr(inode, &no_expand);
 773                 brelse(iloc.bh);
 774
 775                 /*
 776                  * It's important to update i_size while still holding page
 777                  * lock: page writeout could otherwise come in and zero
 778                  * beyond i_size.
 779                  */
 780                 ext4_update_inode_size(inode, pos + copied);
 781         }
 782         unlock_page(page);
 783         put_page(page);
 784
 785         /*
 786          * Don't mark the inode dirty under page lock. First, it unnecessarily
 787          * makes the holding time of page lock longer. Second, it forces lock
 788          * ordering of page lock and transaction start for journaling
 789          * filesystems.
 790          */
 791         if (likely(copied))
 792                 mark_inode_dirty(inode);
 793 out:
 794         /*
 795          * If we didn't copy as much data as expected, we need to trim back
 796          * size of xattr containing inline data.
 797          */
 798         if (pos + len > inode->i_size && ext4_can_truncate(inode))
 799                 ext4_orphan_add(handle, inode);
 800
 801         ret2 = ext4_journal_stop(handle);
 802         if (!ret)
 803                 ret = ret2;
 804         if (pos + len > inode->i_size) {
 805                 ext4_truncate_failed_write(inode);
 806                 /*
 807                  * If truncate failed early the inode might still be
 808                  * on the orphan list; we need to make sure the inode
 809                  * is removed from the orphan list in that case.
 810                  */
 811                 if (inode->i_nlink)
 812                         ext4_orphan_del(NULL, inode);
 813         }
 814         return ret ? ret : copied;
 815 }
 816
 817 struct buffer_head *
 818 ext4_journalled_write_inline_data(struct inode *inode,
 819                                   unsigned len,
 820                                   struct page *page)
 821 {
 822         int ret, no_expand;
 823         void *kaddr;
 824         struct ext4_iloc iloc;
 825
 826         ret = ext4_get_inode_loc(inode, &iloc);
 827         if (ret) {
 828                 ext4_std_error(inode->i_sb, ret);
 829                 return NULL;
 830         }
 831
 832         ext4_write_lock_xattr(inode, &no_expand);
 833         kaddr = kmap_atomic(page);
 834         ext4_write_inline_data(inode, &iloc, kaddr, 0, len);
 835         kunmap_atomic(kaddr);
 836         ext4_write_unlock_xattr(inode, &no_expand);
 837
 838         return iloc.bh;
 839 }
 840
 841 /*
 842  * Try to make the page cache and handle ready for the inline data case.
 843  * We can call this function in 2 cases:
 844  * 1. The inode is created and the first write exceeds inline size. We can
 845  *    clear the inode state safely.
 846  * 2. The inode has inline data, then we need to read the data, make it
 847  *    update and dirty so that ext4_da_writepages can handle it. We don't
 848  *    need to start the journal since the file's metadata isn't changed now.
 849  */
 850 static int ext4_da_convert_inline_data_to_extent(struct address_space *mapping,
 851                                                  struct inode *inode,
 852                                                  void **fsdata)
 853 {
 854         int ret = 0, inline_size;
 855         struct page *page;
 856
 857         page = grab_cache_page_write_begin(mapping, 0);
 858         if (!page)
 859                 return -ENOMEM;
 860
 861         down_read(&EXT4_I(inode)->xattr_sem);
 862         if (!ext4_has_inline_data(inode)) {
 863                 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
 864                 goto out;
 865         }
 866
 867         inline_size = ext4_get_inline_size(inode);
 868
 869         if (!PageUptodate(page)) {
 870                 ret = ext4_read_inline_page(inode, page);
 871                 if (ret < 0)
 872                         goto out;
 873         }
 874
 875         ret = __block_write_begin(page, 0, inline_size,
 876                                   ext4_da_get_block_prep);
 877         if (ret) {
 878                 up_read(&EXT4_I(inode)->xattr_sem);
 879                 unlock_page(page);
 880                 put_page(page);
 881                 ext4_truncate_failed_write(inode);
 882                 return ret;
 883         }
 884
 885         SetPageDirty(page);
 886         SetPageUptodate(page);
 887         ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
 888         *fsdata = (void *)CONVERT_INLINE_DATA;
 889
 890 out:
 891         up_read(&EXT4_I(inode)->xattr_sem);
 892         if (page) {
 893                 unlock_page(page);
 894                 put_page(page);
 895         }
 896         return ret;
 897 }
 898
 899 /*
 900  * Prepare the write for the inline data.
 901  * If the data can be written into the inode, we just read
 902  * the page and make it uptodate, and start the journal.
 903  * Otherwise read the page, makes it dirty so that it can be
 904  * handle in writepages(the i_disksize update is left to the
 905  * normal ext4_da_write_end).
 906  */
 907 int ext4_da_write_inline_data_begin(struct address_space *mapping,
 908                                     struct inode *inode,
 909                                     loff_t pos, unsigned len,
 910                                     struct page **pagep,
 911                                     void **fsdata)
 912 {
 913         int ret;
 914         handle_t *handle;
 915         struct page *page;
 916         struct ext4_iloc iloc;
 917         int retries = 0;
 918         unsigned int flags;
 919
 920         ret = ext4_get_inode_loc(inode, &iloc);
 921         if (ret)
 922                 return ret;
 923
 924 retry_journal:
 925         handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
 926         if (IS_ERR(handle)) {
 927                 ret = PTR_ERR(handle);
 928                 goto out;
 929         }
 930
 931         ret = ext4_prepare_inline_data(handle, inode, pos + len);
 932         if (ret && ret != -ENOSPC)
 933                 goto out_journal;
 934
 935         if (ret == -ENOSPC) {
 936                 ext4_journal_stop(handle);
 937                 ret = ext4_da_convert_inline_data_to_extent(mapping,
 938                                                             inode,
 939                                                             fsdata);
 940                 if (ret == -ENOSPC &&
 941                     ext4_should_retry_alloc(inode->i_sb, &retries))
 942                         goto retry_journal;
 943                 goto out;
 944         }
 945
 946         /*
 947          * We cannot recurse into the filesystem as the transaction
 948          * is already started.
 949          */
 950         flags = memalloc_nofs_save();
 951         page = grab_cache_page_write_begin(mapping, 0);
 952         memalloc_nofs_restore(flags);
 953         if (!page) {
 954                 ret = -ENOMEM;
 955                 goto out_journal;
 956         }
 957
 958         down_read(&EXT4_I(inode)->xattr_sem);
 959         if (!ext4_has_inline_data(inode)) {
 960                 ret = 0;
 961                 goto out_release_page;
 962         }
 963
 964         if (!PageUptodate(page)) {
 965                 ret = ext4_read_inline_page(inode, page);
 966                 if (ret < 0)
 967                         goto out_release_page;
 968         }
 969         ret = ext4_journal_get_write_access(handle, inode->i_sb, iloc.bh,
 970                                             EXT4_JTR_NONE);
 971         if (ret)
 972                 goto out_release_page;
 973
 974         up_read(&EXT4_I(inode)->xattr_sem);
 975         *pagep = page;
 976         brelse(iloc.bh);
 977         return 1;
 978 out_release_page:
 979         up_read(&EXT4_I(inode)->xattr_sem);
 980         unlock_page(page);
 981         put_page(page);
 982 out_journal:
 983         ext4_journal_stop(handle);
 984 out:
 985         brelse(iloc.bh);
 986         return ret;
 987 }
 988
 989 #ifdef INLINE_DIR_DEBUG
 990 void ext4_show_inline_dir(struct inode *dir, struct buffer_head *bh,
 991                           void *inline_start, int inline_size)
 992 {
 993         int offset;
 994         unsigned short de_len;
 995         struct ext4_dir_entry_2 *de = inline_start;
 996         void *dlimit = inline_start + inline_size;
 997
 998         trace_printk("inode %lu\n", dir->i_ino);
 999         offset = 0;
1000         while ((void *)de < dlimit) {
1001                 de_len = ext4_rec_len_from_disk(de->rec_len, inline_size);
1002                 trace_printk("de: off %u rlen %u name %.*s nlen %u ino %u\n",
1003                              offset, de_len, de->name_len, de->name,
1004                              de->name_len, le32_to_cpu(de->inode));
1005                 if (ext4_check_dir_entry(dir, NULL, de, bh,
1006                                          inline_start, inline_size, offset))
1007                         BUG();
1008
1009                 offset += de_len;
1010                 de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
1011         }
1012 }
1013 #else
1014 #define ext4_show_inline_dir(dir, bh, inline_start, inline_size)
1015 #endif
1016
1017 /*
1018  * Add a new entry into a inline dir.
1019  * It will return -ENOSPC if no space is available, and -EIO
1020  * and -EEXIST if directory entry already exists.
1021  */
1022 static int ext4_add_dirent_to_inline(handle_t *handle,
1023                                      struct ext4_filename *fname,
1024                                      struct inode *dir,
1025                                      struct inode *inode,
1026                                      struct ext4_iloc *iloc,
1027                                      void *inline_start, int inline_size)
1028 {
1029         int             err;
1030         struct ext4_dir_entry_2 *de;
1031
1032         err = ext4_find_dest_de(dir, inode, iloc->bh, inline_start,
1033                                 inline_size, fname, &de);
1034         if (err)
1035                 return err;
1036
1037         BUFFER_TRACE(iloc->bh, "get_write_access");
1038         err = ext4_journal_get_write_access(handle, dir->i_sb, iloc->bh,
1039                                             EXT4_JTR_NONE);
1040         if (err)
1041                 return err;
1042         ext4_insert_dentry(dir, inode, de, inline_size, fname);
1043
1044         ext4_show_inline_dir(dir, iloc->bh, inline_start, inline_size);
1045
1046         /*
1047          * XXX shouldn't update any times until successful
1048          * completion of syscall, but too many callers depend
1049          * on this.
1050          *
1051          * XXX similarly, too many callers depend on
1052          * ext4_new_inode() setting the times, but error
1053          * recovery deletes the inode, so the worst that can
1054          * happen is that the times are slightly out of date
1055          * and/or different from the directory change time.
1056          */
1057         dir->i_mtime = dir->i_ctime = current_time(dir);
1058         ext4_update_dx_flag(dir);
1059         inode_inc_iversion(dir);
1060         return 1;
1061 }
1062
1063 static void *ext4_get_inline_xattr_pos(struct inode *inode,
1064                                        struct ext4_iloc *iloc)
1065 {
1066         struct ext4_xattr_entry *entry;
1067         struct ext4_xattr_ibody_header *header;
1068
1069         BUG_ON(!EXT4_I(inode)->i_inline_off);
1070
1071         header = IHDR(inode, ext4_raw_inode(iloc));
1072         entry = (struct ext4_xattr_entry *)((void *)ext4_raw_inode(iloc) +
1073                                             EXT4_I(inode)->i_inline_off);
1074
1075         return (void *)IFIRST(header) + le16_to_cpu(entry->e_value_offs);
1076 }
1077
1078 /* Set the final de to cover the whole block. */
1079 static void ext4_update_final_de(void *de_buf, int old_size, int new_size)
1080 {
1081         struct ext4_dir_entry_2 *de, *prev_de;
1082         void *limit;
1083         int de_len;
1084
1085         de = de_buf;
1086         if (old_size) {
1087                 limit = de_buf + old_size;
1088                 do {
1089                         prev_de = de;
1090                         de_len = ext4_rec_len_from_disk(de->rec_len, old_size);
1091                         de_buf += de_len;
1092                         de = de_buf;
1093                 } while (de_buf < limit);
1094
1095                 prev_de->rec_len = ext4_rec_len_to_disk(de_len + new_size -
1096                                                         old_size, new_size);
1097         } else {
1098                 /* this is just created, so create an empty entry. */
1099                 de->inode = 0;
1100                 de->rec_len = ext4_rec_len_to_disk(new_size, new_size);
1101         }
1102 }
1103
1104 static int ext4_update_inline_dir(handle_t *handle, struct inode *dir,
1105                                   struct ext4_iloc *iloc)
1106 {
1107         int ret;
1108         int old_size = EXT4_I(dir)->i_inline_size - EXT4_MIN_INLINE_DATA_SIZE;
1109         int new_size = get_max_inline_xattr_value_size(dir, iloc);
1110
1111         if (new_size - old_size <= ext4_dir_rec_len(1, NULL))
1112                 return -ENOSPC;
1113
1114         ret = ext4_update_inline_data(handle, dir,
1115                                       new_size + EXT4_MIN_INLINE_DATA_SIZE);
1116         if (ret)
1117                 return ret;
1118
1119         ext4_update_final_de(ext4_get_inline_xattr_pos(dir, iloc), old_size,
1120                              EXT4_I(dir)->i_inline_size -
1121                                                 EXT4_MIN_INLINE_DATA_SIZE);
1122         dir->i_size = EXT4_I(dir)->i_disksize = EXT4_I(dir)->i_inline_size;
1123         return 0;
1124 }
1125
1126 static void ext4_restore_inline_data(handle_t *handle, struct inode *inode,
1127                                      struct ext4_iloc *iloc,
1128                                      void *buf, int inline_size)
1129 {
1130         int ret;
1131
1132         ret = ext4_create_inline_data(handle, inode, inline_size);
1133         if (ret) {
1134                 ext4_msg(inode->i_sb, KERN_EMERG,
1135                         "error restoring inline_data for inode -- potential data loss! (inode %lu, error %d)",
1136                         inode->i_ino, ret);
1137                 return;
1138         }
1139         ext4_write_inline_data(inode, iloc, buf, 0, inline_size);
1140         ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
1141 }
1142
1143 static int ext4_finish_convert_inline_dir(handle_t *handle,
1144                                           struct inode *inode,
1145                                           struct buffer_head *dir_block,
1146                                           void *buf,
1147                                           int inline_size)
1148 {
1149         int err, csum_size = 0, header_size = 0;
1150         struct ext4_dir_entry_2 *de;
1151         void *target = dir_block->b_data;
1152
1153         /*
1154          * First create "." and ".." and then copy the dir information
1155          * back to the block.
1156          */
1157         de = target;
1158         de = ext4_init_dot_dotdot(inode, de,
1159                 inode->i_sb->s_blocksize, csum_size,
1160                 le32_to_cpu(((struct ext4_dir_entry_2 *)buf)->inode), 1);
1161         header_size = (void *)de - target;
1162
1163         memcpy((void *)de, buf + EXT4_INLINE_DOTDOT_SIZE,
1164                 inline_size - EXT4_INLINE_DOTDOT_SIZE);
1165
1166         if (ext4_has_metadata_csum(inode->i_sb))
1167                 csum_size = sizeof(struct ext4_dir_entry_tail);
1168
1169         inode->i_size = inode->i_sb->s_blocksize;
1170         i_size_write(inode, inode->i_sb->s_blocksize);
1171         EXT4_I(inode)->i_disksize = inode->i_sb->s_blocksize;
1172         ext4_update_final_de(dir_block->b_data,
1173                         inline_size - EXT4_INLINE_DOTDOT_SIZE + header_size,
1174                         inode->i_sb->s_blocksize - csum_size);
1175
1176         if (csum_size)
1177                 ext4_initialize_dirent_tail(dir_block,
1178                                             inode->i_sb->s_blocksize);
1179         set_buffer_uptodate(dir_block);
1180         err = ext4_handle_dirty_dirblock(handle, inode, dir_block);
1181         if (err)
1182                 return err;
1183         set_buffer_verified(dir_block);
1184         return ext4_mark_inode_dirty(handle, inode);
1185 }
1186
1187 static int ext4_convert_inline_data_nolock(handle_t *handle,
1188                                            struct inode *inode,
1189                                            struct ext4_iloc *iloc)
1190 {
1191         int error;
1192         void *buf = NULL;
1193         struct buffer_head *data_bh = NULL;
1194         struct ext4_map_blocks map;
1195         int inline_size;
1196
1197         inline_size = ext4_get_inline_size(inode);
1198         buf = kmalloc(inline_size, GFP_NOFS);
1199         if (!buf) {
1200                 error = -ENOMEM;
1201                 goto out;
1202         }
1203
1204         error = ext4_read_inline_data(inode, buf, inline_size, iloc);
1205         if (error < 0)
1206                 goto out;
1207
1208         /*
1209          * Make sure the inline directory entries pass checks before we try to
1210          * convert them, so that we avoid touching stuff that needs fsck.
1211          */
1212         if (S_ISDIR(inode->i_mode)) {
1213                 error = ext4_check_all_de(inode, iloc->bh,
1214                                         buf + EXT4_INLINE_DOTDOT_SIZE,
1215                                         inline_size - EXT4_INLINE_DOTDOT_SIZE);
1216                 if (error)
1217                         goto out;
1218         }
1219
1220         error = ext4_destroy_inline_data_nolock(handle, inode);
1221         if (error)
1222                 goto out;
1223
1224         map.m_lblk = 0;
1225         map.m_len = 1;
1226         map.m_flags = 0;
1227         error = ext4_map_blocks(handle, inode, &map, EXT4_GET_BLOCKS_CREATE);
1228         if (error < 0)
1229                 goto out_restore;
1230         if (!(map.m_flags & EXT4_MAP_MAPPED)) {
1231                 error = -EIO;
1232                 goto out_restore;
1233         }
1234
1235         data_bh = sb_getblk(inode->i_sb, map.m_pblk);
1236         if (!data_bh) {
1237                 error = -ENOMEM;
1238                 goto out_restore;
1239         }
1240
1241         lock_buffer(data_bh);
1242         error = ext4_journal_get_create_access(handle, inode->i_sb, data_bh,
1243                                                EXT4_JTR_NONE);
1244         if (error) {
1245                 unlock_buffer(data_bh);
1246                 error = -EIO;
1247                 goto out_restore;
1248         }
1249         memset(data_bh->b_data, 0, inode->i_sb->s_blocksize);
1250
1251         if (!S_ISDIR(inode->i_mode)) {
1252                 memcpy(data_bh->b_data, buf, inline_size);
1253                 set_buffer_uptodate(data_bh);
1254                 error = ext4_handle_dirty_metadata(handle,
1255                                                    inode, data_bh);
1256         } else {
1257                 error = ext4_finish_convert_inline_dir(handle, inode, data_bh,
1258                                                        buf, inline_size);
1259         }
1260
1261         unlock_buffer(data_bh);
1262 out_restore:
1263         if (error)
1264                 ext4_restore_inline_data(handle, inode, iloc, buf, inline_size);
1265
1266 out:
1267         brelse(data_bh);
1268         kfree(buf);
1269         return error;
1270 }
1271
1272 /*
1273  * Try to add the new entry to the inline data.
1274  * If succeeds, return 0. If not, extended the inline dir and copied data to
1275  * the new created block.
1276  */
1277 int ext4_try_add_inline_entry(handle_t *handle, struct ext4_filename *fname,
1278                               struct inode *dir, struct inode *inode)
1279 {
1280         int ret, ret2, inline_size, no_expand;
1281         void *inline_start;
1282         struct ext4_iloc iloc;
1283
1284         ret = ext4_get_inode_loc(dir, &iloc);
1285         if (ret)
1286                 return ret;
1287
1288         ext4_write_lock_xattr(dir, &no_expand);
1289         if (!ext4_has_inline_data(dir))
1290                 goto out;
1291
1292         inline_start = (void *)ext4_raw_inode(&iloc)->i_block +
1293                                                  EXT4_INLINE_DOTDOT_SIZE;
1294         inline_size = EXT4_MIN_INLINE_DATA_SIZE - EXT4_INLINE_DOTDOT_SIZE;
1295
1296         ret = ext4_add_dirent_to_inline(handle, fname, dir, inode, &iloc,
1297                                         inline_start, inline_size);
1298         if (ret != -ENOSPC)
1299                 goto out;
1300
1301         /* check whether it can be inserted to inline xattr space. */
1302         inline_size = EXT4_I(dir)->i_inline_size -
1303                         EXT4_MIN_INLINE_DATA_SIZE;
1304         if (!inline_size) {
1305                 /* Try to use the xattr space.*/
1306                 ret = ext4_update_inline_dir(handle, dir, &iloc);
1307                 if (ret && ret != -ENOSPC)
1308                         goto out;
1309
1310                 inline_size = EXT4_I(dir)->i_inline_size -
1311                                 EXT4_MIN_INLINE_DATA_SIZE;
1312         }
1313
1314         if (inline_size) {
1315                 inline_start = ext4_get_inline_xattr_pos(dir, &iloc);
1316
1317                 ret = ext4_add_dirent_to_inline(handle, fname, dir,
1318                                                 inode, &iloc, inline_start,
1319                                                 inline_size);
1320
1321                 if (ret != -ENOSPC)
1322                         goto out;
1323         }
1324
1325         /*
1326          * The inline space is filled up, so create a new block for it.
1327          * As the extent tree will be created, we have to save the inline
1328          * dir first.
1329          */
1330         ret = ext4_convert_inline_data_nolock(handle, dir, &iloc);
1331
1332 out:
1333         ext4_write_unlock_xattr(dir, &no_expand);
1334         ret2 = ext4_mark_inode_dirty(handle, dir);
1335         if (unlikely(ret2 && !ret))
1336                 ret = ret2;
1337         brelse(iloc.bh);
1338         return ret;
1339 }
1340
1341 /*
1342  * This function fills a red-black tree with information from an
1343  * inlined dir.  It returns the number directory entries loaded
1344  * into the tree.  If there is an error it is returned in err.
1345  */
1346 int ext4_inlinedir_to_tree(struct file *dir_file,
1347                            struct inode *dir, ext4_lblk_t block,
1348                            struct dx_hash_info *hinfo,
1349                            __u32 start_hash, __u32 start_minor_hash,
1350                            int *has_inline_data)
1351 {
1352         int err = 0, count = 0;
1353         unsigned int parent_ino;
1354         int pos;
1355         struct ext4_dir_entry_2 *de;
1356         struct inode *inode = file_inode(dir_file);
1357         int ret, inline_size = 0;
1358         struct ext4_iloc iloc;
1359         void *dir_buf = NULL;
1360         struct ext4_dir_entry_2 fake;
1361         struct fscrypt_str tmp_str;
1362
1363         ret = ext4_get_inode_loc(inode, &iloc);
1364         if (ret)
1365                 return ret;
1366
1367         down_read(&EXT4_I(inode)->xattr_sem);
1368         if (!ext4_has_inline_data(inode)) {
1369                 up_read(&EXT4_I(inode)->xattr_sem);
1370                 *has_inline_data = 0;
1371                 goto out;
1372         }
1373
1374         inline_size = ext4_get_inline_size(inode);
1375         dir_buf = kmalloc(inline_size, GFP_NOFS);
1376         if (!dir_buf) {
1377                 ret = -ENOMEM;
1378                 up_read(&EXT4_I(inode)->xattr_sem);
1379                 goto out;
1380         }
1381
1382         ret = ext4_read_inline_data(inode, dir_buf, inline_size, &iloc);
1383         up_read(&EXT4_I(inode)->xattr_sem);
1384         if (ret < 0)
1385                 goto out;
1386
1387         pos = 0;
1388         parent_ino = le32_to_cpu(((struct ext4_dir_entry_2 *)dir_buf)->inode);
1389         while (pos < inline_size) {
1390                 /*
1391                  * As inlined dir doesn't store any information about '.' and
1392                  * only the inode number of '..' is stored, we have to handle
1393                  * them differently.
1394                  */
1395                 if (pos == 0) {
1396                         fake.inode = cpu_to_le32(inode->i_ino);
1397                         fake.name_len = 1;
1398                         strcpy(fake.name, ".");
1399                         fake.rec_len = ext4_rec_len_to_disk(
1400                                           ext4_dir_rec_len(fake.name_len, NULL),
1401                                           inline_size);
1402                         ext4_set_de_type(inode->i_sb, &fake, S_IFDIR);
1403                         de = &fake;
1404                         pos = EXT4_INLINE_DOTDOT_OFFSET;
1405                 } else if (pos == EXT4_INLINE_DOTDOT_OFFSET) {
1406                         fake.inode = cpu_to_le32(parent_ino);
1407                         fake.name_len = 2;
1408                         strcpy(fake.name, "..");
1409                         fake.rec_len = ext4_rec_len_to_disk(
1410                                           ext4_dir_rec_len(fake.name_len, NULL),
1411                                           inline_size);
1412                         ext4_set_de_type(inode->i_sb, &fake, S_IFDIR);
1413                         de = &fake;
1414                         pos = EXT4_INLINE_DOTDOT_SIZE;
1415                 } else {
1416                         de = (struct ext4_dir_entry_2 *)(dir_buf + pos);
1417                         pos += ext4_rec_len_from_disk(de->rec_len, inline_size);
1418                         if (ext4_check_dir_entry(inode, dir_file, de,
1419                                          iloc.bh, dir_buf,
1420                                          inline_size, pos)) {
1421                                 ret = count;
1422                                 goto out;
1423                         }
1424                 }
1425
1426                 if (ext4_hash_in_dirent(dir)) {
1427                         hinfo->hash = EXT4_DIRENT_HASH(de);
1428                         hinfo->minor_hash = EXT4_DIRENT_MINOR_HASH(de);
1429                 } else {
1430                         ext4fs_dirhash(dir, de->name, de->name_len, hinfo);
1431                 }
1432                 if ((hinfo->hash < start_hash) ||
1433                     ((hinfo->hash == start_hash) &&
1434                      (hinfo->minor_hash < start_minor_hash)))
1435                         continue;
1436                 if (de->inode == 0)
1437                         continue;
1438                 tmp_str.name = de->name;
1439                 tmp_str.len = de->name_len;
1440                 err = ext4_htree_store_dirent(dir_file, hinfo->hash,
1441                                               hinfo->minor_hash, de, &tmp_str);
1442                 if (err) {
1443                         ret = err;
1444                         goto out;
1445                 }
1446                 count++;
1447         }
1448         ret = count;
1449 out:
1450         kfree(dir_buf);
1451         brelse(iloc.bh);
1452         return ret;
1453 }
1454
1455 /*
1456  * So this function is called when the volume is mkfsed with
1457  * dir_index disabled. In order to keep f_pos persistent
1458  * after we convert from an inlined dir to a blocked based,
1459  * we just pretend that we are a normal dir and return the
1460  * offset as if '.' and '..' really take place.
1461  *
1462  */
1463 int ext4_read_inline_dir(struct file *file,
1464                          struct dir_context *ctx,
1465                          int *has_inline_data)
1466 {
1467         unsigned int offset, parent_ino;
1468         int i;
1469         struct ext4_dir_entry_2 *de;
1470         struct super_block *sb;
1471         struct inode *inode = file_inode(file);
1472         int ret, inline_size = 0;
1473         struct ext4_iloc iloc;
1474         void *dir_buf = NULL;
1475         int dotdot_offset, dotdot_size, extra_offset, extra_size;
1476
1477         ret = ext4_get_inode_loc(inode, &iloc);
1478         if (ret)
1479                 return ret;
1480
1481         down_read(&EXT4_I(inode)->xattr_sem);
1482         if (!ext4_has_inline_data(inode)) {
1483                 up_read(&EXT4_I(inode)->xattr_sem);
1484                 *has_inline_data = 0;
1485                 goto out;
1486         }
1487
1488         inline_size = ext4_get_inline_size(inode);
1489         dir_buf = kmalloc(inline_size, GFP_NOFS);
1490         if (!dir_buf) {
1491                 ret = -ENOMEM;
1492                 up_read(&EXT4_I(inode)->xattr_sem);
1493                 goto out;
1494         }
1495
1496         ret = ext4_read_inline_data(inode, dir_buf, inline_size, &iloc);
1497         up_read(&EXT4_I(inode)->xattr_sem);
1498         if (ret < 0)
1499                 goto out;
1500
1501         ret = 0;
1502         sb = inode->i_sb;
1503         parent_ino = le32_to_cpu(((struct ext4_dir_entry_2 *)dir_buf)->inode);
1504         offset = ctx->pos;
1505
1506         /*
1507          * dotdot_offset and dotdot_size is the real offset and
1508          * size for ".." and "." if the dir is block based while
1509          * the real size for them are only EXT4_INLINE_DOTDOT_SIZE.
1510          * So we will use extra_offset and extra_size to indicate them
1511          * during the inline dir iteration.
1512          */
1513         dotdot_offset = ext4_dir_rec_len(1, NULL);
1514         dotdot_size = dotdot_offset + ext4_dir_rec_len(2, NULL);
1515         extra_offset = dotdot_size - EXT4_INLINE_DOTDOT_SIZE;
1516         extra_size = extra_offset + inline_size;
1517
1518         /*
1519          * If the version has changed since the last call to
1520          * readdir(2), then we might be pointing to an invalid
1521          * dirent right now.  Scan from the start of the inline
1522          * dir to make sure.
1523          */
1524         if (!inode_eq_iversion(inode, file->f_version)) {
1525                 for (i = 0; i < extra_size && i < offset;) {
1526                         /*
1527                          * "." is with offset 0 and
1528                          * ".." is dotdot_offset.
1529                          */
1530                         if (!i) {
1531                                 i = dotdot_offset;
1532                                 continue;
1533                         } else if (i == dotdot_offset) {
1534                                 i = dotdot_size;
1535                                 continue;
1536                         }
1537                         /* for other entry, the real offset in
1538                          * the buf has to be tuned accordingly.
1539                          */
1540                         de = (struct ext4_dir_entry_2 *)
1541                                 (dir_buf + i - extra_offset);
1542                         /* It's too expensive to do a full
1543                          * dirent test each time round this
1544                          * loop, but we do have to test at
1545                          * least that it is non-zero.  A
1546                          * failure will be detected in the
1547                          * dirent test below. */
1548                         if (ext4_rec_len_from_disk(de->rec_len, extra_size)
1549                                 < ext4_dir_rec_len(1, NULL))
1550                                 break;
1551                         i += ext4_rec_len_from_disk(de->rec_len,
1552                                                     extra_size);
1553                 }
1554                 offset = i;
1555                 ctx->pos = offset;
1556                 file->f_version = inode_query_iversion(inode);
1557         }
1558
1559         while (ctx->pos < extra_size) {
1560                 if (ctx->pos == 0) {
1561                         if (!dir_emit(ctx, ".", 1, inode->i_ino, DT_DIR))
1562                                 goto out;
1563                         ctx->pos = dotdot_offset;
1564                         continue;
1565                 }
1566
1567                 if (ctx->pos == dotdot_offset) {
1568                         if (!dir_emit(ctx, "..", 2, parent_ino, DT_DIR))
1569                                 goto out;
1570                         ctx->pos = dotdot_size;
1571                         continue;
1572                 }
1573
1574                 de = (struct ext4_dir_entry_2 *)
1575                         (dir_buf + ctx->pos - extra_offset);
1576                 if (ext4_check_dir_entry(inode, file, de, iloc.bh, dir_buf,
1577                                          extra_size, ctx->pos))
1578                         goto out;
1579                 if (le32_to_cpu(de->inode)) {
1580                         if (!dir_emit(ctx, de->name, de->name_len,
1581                                       le32_to_cpu(de->inode),
1582                                       get_dtype(sb, de->file_type)))
1583                                 goto out;
1584                 }
1585                 ctx->pos += ext4_rec_len_from_disk(de->rec_len, extra_size);
1586         }
1587 out:
1588         kfree(dir_buf);
1589         brelse(iloc.bh);
1590         return ret;
1591 }
1592
1593 void *ext4_read_inline_link(struct inode *inode)
1594 {
1595         struct ext4_iloc iloc;
1596         int ret, inline_size;
1597         void *link;
1598
1599         ret = ext4_get_inode_loc(inode, &iloc);
1600         if (ret)
1601                 return ERR_PTR(ret);
1602
1603         ret = -ENOMEM;
1604         inline_size = ext4_get_inline_size(inode);
1605         link = kmalloc(inline_size + 1, GFP_NOFS);
1606         if (!link)
1607                 goto out;
1608
1609         ret = ext4_read_inline_data(inode, link, inline_size, &iloc);
1610         if (ret < 0) {
1611                 kfree(link);
1612                 goto out;
1613         }
1614         nd_terminate_link(link, inode->i_size, ret);
1615 out:
1616         if (ret < 0)
1617                 link = ERR_PTR(ret);
1618         brelse(iloc.bh);
1619         return link;
1620 }
1621
1622 struct buffer_head *ext4_get_first_inline_block(struct inode *inode,
1623                                         struct ext4_dir_entry_2 **parent_de,
1624                                         int *retval)
1625 {
1626         struct ext4_iloc iloc;
1627
1628         *retval = ext4_get_inode_loc(inode, &iloc);
1629         if (*retval)
1630                 return NULL;
1631
1632         *parent_de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block;
1633
1634         return iloc.bh;
1635 }
1636
1637 /*
1638  * Try to create the inline data for the new dir.
1639  * If it succeeds, return 0, otherwise return the error.
1640  * In case of ENOSPC, the caller should create the normal disk layout dir.
1641  */
1642 int ext4_try_create_inline_dir(handle_t *handle, struct inode *parent,
1643                                struct inode *inode)
1644 {
1645         int ret, inline_size = EXT4_MIN_INLINE_DATA_SIZE;
1646         struct ext4_iloc iloc;
1647         struct ext4_dir_entry_2 *de;
1648
1649         ret = ext4_get_inode_loc(inode, &iloc);
1650         if (ret)
1651                 return ret;
1652
1653         ret = ext4_prepare_inline_data(handle, inode, inline_size);
1654         if (ret)
1655                 goto out;
1656
1657         /*
1658          * For inline dir, we only save the inode information for the ".."
1659          * and create a fake dentry to cover the left space.
1660          */
1661         de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block;
1662         de->inode = cpu_to_le32(parent->i_ino);
1663         de = (struct ext4_dir_entry_2 *)((void *)de + EXT4_INLINE_DOTDOT_SIZE);
1664         de->inode = 0;
1665         de->rec_len = ext4_rec_len_to_disk(
1666                                 inline_size - EXT4_INLINE_DOTDOT_SIZE,
1667                                 inline_size);
1668         set_nlink(inode, 2);
1669         inode->i_size = EXT4_I(inode)->i_disksize = inline_size;
1670 out:
1671         brelse(iloc.bh);
1672         return ret;
1673 }
1674
1675 struct buffer_head *ext4_find_inline_entry(struct inode *dir,
1676                                         struct ext4_filename *fname,
1677                                         struct ext4_dir_entry_2 **res_dir,
1678                                         int *has_inline_data)
1679 {
1680         int ret;
1681         struct ext4_iloc iloc;
1682         void *inline_start;
1683         int inline_size;
1684
1685         if (ext4_get_inode_loc(dir, &iloc))
1686                 return NULL;
1687
1688         down_read(&EXT4_I(dir)->xattr_sem);
1689         if (!ext4_has_inline_data(dir)) {
1690                 *has_inline_data = 0;
1691                 goto out;
1692         }
1693
1694         inline_start = (void *)ext4_raw_inode(&iloc)->i_block +
1695                                                 EXT4_INLINE_DOTDOT_SIZE;
1696         inline_size = EXT4_MIN_INLINE_DATA_SIZE - EXT4_INLINE_DOTDOT_SIZE;
1697         ret = ext4_search_dir(iloc.bh, inline_start, inline_size,
1698                               dir, fname, 0, res_dir);
1699         if (ret == 1)
1700                 goto out_find;
1701         if (ret < 0)
1702                 goto out;
1703
1704         if (ext4_get_inline_size(dir) == EXT4_MIN_INLINE_DATA_SIZE)
1705                 goto out;
1706
1707         inline_start = ext4_get_inline_xattr_pos(dir, &iloc);
1708         inline_size = ext4_get_inline_size(dir) - EXT4_MIN_INLINE_DATA_SIZE;
1709
1710         ret = ext4_search_dir(iloc.bh, inline_start, inline_size,
1711                               dir, fname, 0, res_dir);
1712         if (ret == 1)
1713                 goto out_find;
1714
1715 out:
1716         brelse(iloc.bh);
1717         iloc.bh = NULL;
1718 out_find:
1719         up_read(&EXT4_I(dir)->xattr_sem);
1720         return iloc.bh;
1721 }
1722
1723 int ext4_delete_inline_entry(handle_t *handle,
1724                              struct inode *dir,
1725                              struct ext4_dir_entry_2 *de_del,
1726                              struct buffer_head *bh,
1727                              int *has_inline_data)
1728 {
1729         int err, inline_size, no_expand;
1730         struct ext4_iloc iloc;
1731         void *inline_start;
1732
1733         err = ext4_get_inode_loc(dir, &iloc);
1734         if (err)
1735                 return err;
1736
1737         ext4_write_lock_xattr(dir, &no_expand);
1738         if (!ext4_has_inline_data(dir)) {
1739                 *has_inline_data = 0;
1740                 goto out;
1741         }
1742
1743         if ((void *)de_del - ((void *)ext4_raw_inode(&iloc)->i_block) <
1744                 EXT4_MIN_INLINE_DATA_SIZE) {
1745                 inline_start = (void *)ext4_raw_inode(&iloc)->i_block +
1746                                         EXT4_INLINE_DOTDOT_SIZE;
1747                 inline_size = EXT4_MIN_INLINE_DATA_SIZE -
1748                                 EXT4_INLINE_DOTDOT_SIZE;
1749         } else {
1750                 inline_start = ext4_get_inline_xattr_pos(dir, &iloc);
1751                 inline_size = ext4_get_inline_size(dir) -
1752                                 EXT4_MIN_INLINE_DATA_SIZE;
1753         }
1754
1755         BUFFER_TRACE(bh, "get_write_access");
1756         err = ext4_journal_get_write_access(handle, dir->i_sb, bh,
1757                                             EXT4_JTR_NONE);
1758         if (err)
1759                 goto out;
1760
1761         err = ext4_generic_delete_entry(dir, de_del, bh,
1762                                         inline_start, inline_size, 0);
1763         if (err)
1764                 goto out;
1765
1766         ext4_show_inline_dir(dir, iloc.bh, inline_start, inline_size);
1767 out:
1768         ext4_write_unlock_xattr(dir, &no_expand);
1769         if (likely(err == 0))
1770                 err = ext4_mark_inode_dirty(handle, dir);
1771         brelse(iloc.bh);
1772         if (err != -ENOENT)
1773                 ext4_std_error(dir->i_sb, err);
1774         return err;
1775 }
1776
1777 /*
1778  * Get the inline dentry at offset.
1779  */
1780 static inline struct ext4_dir_entry_2 *
1781 ext4_get_inline_entry(struct inode *inode,
1782                       struct ext4_iloc *iloc,
1783                       unsigned int offset,
1784                       void **inline_start,
1785                       int *inline_size)
1786 {
1787         void *inline_pos;
1788
1789         BUG_ON(offset > ext4_get_inline_size(inode));
1790
1791         if (offset < EXT4_MIN_INLINE_DATA_SIZE) {
1792                 inline_pos = (void *)ext4_raw_inode(iloc)->i_block;
1793                 *inline_size = EXT4_MIN_INLINE_DATA_SIZE;
1794         } else {
1795                 inline_pos = ext4_get_inline_xattr_pos(inode, iloc);
1796                 offset -= EXT4_MIN_INLINE_DATA_SIZE;
1797                 *inline_size = ext4_get_inline_size(inode) -
1798                                 EXT4_MIN_INLINE_DATA_SIZE;
1799         }
1800
1801         if (inline_start)
1802                 *inline_start = inline_pos;
1803         return (struct ext4_dir_entry_2 *)(inline_pos + offset);
1804 }
1805
1806 bool empty_inline_dir(struct inode *dir, int *has_inline_data)
1807 {
1808         int err, inline_size;
1809         struct ext4_iloc iloc;
1810         size_t inline_len;
1811         void *inline_pos;
1812         unsigned int offset;
1813         struct ext4_dir_entry_2 *de;
1814         bool ret = false;
1815
1816         err = ext4_get_inode_loc(dir, &iloc);
1817         if (err) {
1818                 EXT4_ERROR_INODE_ERR(dir, -err,
1819                                      "error %d getting inode %lu block",
1820                                      err, dir->i_ino);
1821                 return false;
1822         }
1823
1824         down_read(&EXT4_I(dir)->xattr_sem);
1825         if (!ext4_has_inline_data(dir)) {
1826                 *has_inline_data = 0;
1827                 ret = true;
1828                 goto out;
1829         }
1830
1831         de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block;
1832         if (!le32_to_cpu(de->inode)) {
1833                 ext4_warning(dir->i_sb,
1834                              "bad inline directory (dir #%lu) - no `..'",
1835                              dir->i_ino);
1836                 goto out;
1837         }
1838
1839         inline_len = ext4_get_inline_size(dir);
1840         offset = EXT4_INLINE_DOTDOT_SIZE;
1841         while (offset < inline_len) {
1842                 de = ext4_get_inline_entry(dir, &iloc, offset,
1843                                            &inline_pos, &inline_size);
1844                 if (ext4_check_dir_entry(dir, NULL, de,
1845                                          iloc.bh, inline_pos,
1846                                          inline_size, offset)) {
1847                         ext4_warning(dir->i_sb,
1848                                      "bad inline directory (dir #%lu) - "
1849                                      "inode %u, rec_len %u, name_len %d"
1850                                      "inline size %d",
1851                                      dir->i_ino, le32_to_cpu(de->inode),
1852                                      le16_to_cpu(de->rec_len), de->name_len,
1853                                      inline_size);
1854                         goto out;
1855                 }
1856                 if (le32_to_cpu(de->inode)) {
1857                         goto out;
1858                 }
1859                 offset += ext4_rec_len_from_disk(de->rec_len, inline_size);
1860         }
1861
1862         ret = true;
1863 out:
1864         up_read(&EXT4_I(dir)->xattr_sem);
1865         brelse(iloc.bh);
1866         return ret;
1867 }
1868
1869 int ext4_destroy_inline_data(handle_t *handle, struct inode *inode)
1870 {
1871         int ret, no_expand;
1872
1873         ext4_write_lock_xattr(inode, &no_expand);
1874         ret = ext4_destroy_inline_data_nolock(handle, inode);
1875         ext4_write_unlock_xattr(inode, &no_expand);
1876
1877         return ret;
1878 }
1879
1880 int ext4_inline_data_iomap(struct inode *inode, struct iomap *iomap)
1881 {
1882         __u64 addr;
1883         int error = -EAGAIN;
1884         struct ext4_iloc iloc;
1885
1886         down_read(&EXT4_I(inode)->xattr_sem);
1887         if (!ext4_has_inline_data(inode))
1888                 goto out;
1889
1890         error = ext4_get_inode_loc(inode, &iloc);
1891         if (error)
1892                 goto out;
1893
1894         addr = (__u64)iloc.bh->b_blocknr << inode->i_sb->s_blocksize_bits;
1895         addr += (char *)ext4_raw_inode(&iloc) - iloc.bh->b_data;
1896         addr += offsetof(struct ext4_inode, i_block);
1897
1898         brelse(iloc.bh);
1899
1900         iomap->addr = addr;
1901         iomap->offset = 0;
1902         iomap->length = min_t(loff_t, ext4_get_inline_size(inode),
1903                               i_size_read(inode));
1904         iomap->type = IOMAP_INLINE;
1905         iomap->flags = 0;
1906
1907 out:
1908         up_read(&EXT4_I(inode)->xattr_sem);
1909         return error;
1910 }
1911
1912 int ext4_inline_data_truncate(struct inode *inode, int *has_inline)
1913 {
1914         handle_t *handle;
1915         int inline_size, value_len, needed_blocks, no_expand, err = 0;
1916         size_t i_size;
1917         void *value = NULL;
1918         struct ext4_xattr_ibody_find is = {
1919                 .s = { .not_found = -ENODATA, },
1920         };
1921         struct ext4_xattr_info i = {
1922                 .name_index = EXT4_XATTR_INDEX_SYSTEM,
1923                 .name = EXT4_XATTR_SYSTEM_DATA,
1924         };
1925
1926
1927         needed_blocks = ext4_writepage_trans_blocks(inode);
1928         handle = ext4_journal_start(inode, EXT4_HT_INODE, needed_blocks);
1929         if (IS_ERR(handle))
1930                 return PTR_ERR(handle);
1931
1932         ext4_write_lock_xattr(inode, &no_expand);
1933         if (!ext4_has_inline_data(inode)) {
1934                 ext4_write_unlock_xattr(inode, &no_expand);
1935                 *has_inline = 0;
1936                 ext4_journal_stop(handle);
1937                 return 0;
1938         }
1939
1940         if ((err = ext4_orphan_add(handle, inode)) != 0)
1941                 goto out;
1942
1943         if ((err = ext4_get_inode_loc(inode, &is.iloc)) != 0)
1944                 goto out;
1945
1946         down_write(&EXT4_I(inode)->i_data_sem);
1947         i_size = inode->i_size;
1948         inline_size = ext4_get_inline_size(inode);
1949         EXT4_I(inode)->i_disksize = i_size;
1950
1951         if (i_size < inline_size) {
1952                 /*
1953                  * if there's inline data to truncate and this file was
1954                  * converted to extents after that inline data was written,
1955                  * the extent status cache must be cleared to avoid leaving
1956                  * behind stale delayed allocated extent entries
1957                  */
1958                 if (!ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
1959 retry:
1960                         err = ext4_es_remove_extent(inode, 0, EXT_MAX_BLOCKS);
1961                         if (err == -ENOMEM) {
1962                                 memalloc_retry_wait(GFP_ATOMIC);
1963                                 goto retry;
1964                         }
1965                         if (err)
1966                                 goto out_error;
1967                 }
1968
1969                 /* Clear the content in the xattr space. */
1970                 if (inline_size > EXT4_MIN_INLINE_DATA_SIZE) {
1971                         if ((err = ext4_xattr_ibody_find(inode, &i, &is)) != 0)
1972                                 goto out_error;
1973
1974                         BUG_ON(is.s.not_found);
1975
1976                         value_len = le32_to_cpu(is.s.here->e_value_size);
1977                         value = kmalloc(value_len, GFP_NOFS);
1978                         if (!value) {
1979                                 err = -ENOMEM;
1980                                 goto out_error;
1981                         }
1982
1983                         err = ext4_xattr_ibody_get(inode, i.name_index,
1984                                                    i.name, value, value_len);
1985                         if (err <= 0)
1986                                 goto out_error;
1987
1988                         i.value = value;
1989                         i.value_len = i_size > EXT4_MIN_INLINE_DATA_SIZE ?
1990                                         i_size - EXT4_MIN_INLINE_DATA_SIZE : 0;
1991                         err = ext4_xattr_ibody_set(handle, inode, &i, &is);
1992                         if (err)
1993                                 goto out_error;
1994                 }
1995
1996                 /* Clear the content within i_blocks. */
1997                 if (i_size < EXT4_MIN_INLINE_DATA_SIZE) {
1998                         void *p = (void *) ext4_raw_inode(&is.iloc)->i_block;
1999                         memset(p + i_size, 0,
2000                                EXT4_MIN_INLINE_DATA_SIZE - i_size);
2001                 }
2002
2003                 EXT4_I(inode)->i_inline_size = i_size <
2004                                         EXT4_MIN_INLINE_DATA_SIZE ?
2005                                         EXT4_MIN_INLINE_DATA_SIZE : i_size;
2006         }
2007
2008 out_error:
2009         up_write(&EXT4_I(inode)->i_data_sem);
2010 out:
2011         brelse(is.iloc.bh);
2012         ext4_write_unlock_xattr(inode, &no_expand);
2013         kfree(value);
2014         if (inode->i_nlink)
2015                 ext4_orphan_del(handle, inode);
2016
2017         if (err == 0) {
2018                 inode->i_mtime = inode->i_ctime = current_time(inode);
2019                 err = ext4_mark_inode_dirty(handle, inode);
2020                 if (IS_SYNC(inode))
2021                         ext4_handle_sync(handle);
2022         }
2023         ext4_journal_stop(handle);
2024         return err;
2025 }
2026
2027 int ext4_convert_inline_data(struct inode *inode)
2028 {
2029         int error, needed_blocks, no_expand;
2030         handle_t *handle;
2031         struct ext4_iloc iloc;
2032
2033         if (!ext4_has_inline_data(inode)) {
2034                 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
2035                 return 0;
2036         } else if (!ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2037                 /*
2038                  * Inode has inline data but EXT4_STATE_MAY_INLINE_DATA is
2039                  * cleared. This means we are in the middle of moving of
2040                  * inline data to delay allocated block. Just force writeout
2041                  * here to finish conversion.
2042                  */
2043                 error = filemap_flush(inode->i_mapping);
2044                 if (error)
2045                         return error;
2046                 if (!ext4_has_inline_data(inode))
2047                         return 0;
2048         }
2049
2050         needed_blocks = ext4_writepage_trans_blocks(inode);
2051
2052         iloc.bh = NULL;
2053         error = ext4_get_inode_loc(inode, &iloc);
2054         if (error)
2055                 return error;
2056
2057         handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
2058         if (IS_ERR(handle)) {
2059                 error = PTR_ERR(handle);
2060                 goto out_free;
2061         }
2062
2063         ext4_write_lock_xattr(inode, &no_expand);
2064         if (ext4_has_inline_data(inode))
2065                 error = ext4_convert_inline_data_nolock(handle, inode, &iloc);
2066         ext4_write_unlock_xattr(inode, &no_expand);
2067         ext4_journal_stop(handle);
2068 out_free:
2069         brelse(iloc.bh);
2070         return error;
2071 }