1 /* 2 * Copyright (c) 2012 Taobao. 3 * Written by Tao Ma <boyu.mt@taobao.com> 4 * 5 * This program is free software; you can redistribute it and/or modify it 6 * under the terms of version 2.1 of the GNU Lesser General Public License 7 * as published by the Free Software Foundation. 8 * 9 * This program is distributed in the hope that it will be useful, 10 * but WITHOUT ANY WARRANTY; without even the implied warranty of 11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the 12 * GNU General Public License for more details. 13 */ 14 15 #include <linux/fiemap.h> 16 17 #include "ext4_jbd2.h" 18 #include "ext4.h" 19 #include "xattr.h" 20 #include "truncate.h" 21 22 #define EXT4_XATTR_SYSTEM_DATA "data" 23 #define EXT4_MIN_INLINE_DATA_SIZE ((sizeof(__le32) * EXT4_N_BLOCKS)) 24 #define EXT4_INLINE_DOTDOT_OFFSET 2 25 #define EXT4_INLINE_DOTDOT_SIZE 4 26 27 static int ext4_get_inline_size(struct inode *inode) 28 { 29 if (EXT4_I(inode)->i_inline_off) 30 return EXT4_I(inode)->i_inline_size; 31 32 return 0; 33 } 34 35 static int get_max_inline_xattr_value_size(struct inode *inode, 36 struct ext4_iloc *iloc) 37 { 38 struct ext4_xattr_ibody_header *header; 39 struct ext4_xattr_entry *entry; 40 struct ext4_inode *raw_inode; 41 int free, min_offs; 42 43 min_offs = EXT4_SB(inode->i_sb)->s_inode_size - 44 EXT4_GOOD_OLD_INODE_SIZE - 45 EXT4_I(inode)->i_extra_isize - 46 sizeof(struct ext4_xattr_ibody_header); 47 48 /* 49 * We need to subtract another sizeof(__u32) since an in-inode xattr 50 * needs an empty 4 bytes to indicate the gap between the xattr entry 51 * and the name/value pair. 52 */ 53 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR)) 54 return EXT4_XATTR_SIZE(min_offs - 55 EXT4_XATTR_LEN(strlen(EXT4_XATTR_SYSTEM_DATA)) - 56 EXT4_XATTR_ROUND - sizeof(__u32)); 57 58 raw_inode = ext4_raw_inode(iloc); 59 header = IHDR(inode, raw_inode); 60 entry = IFIRST(header); 61 62 /* Compute min_offs. */ 63 for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry)) { 64 if (!entry->e_value_block && entry->e_value_size) { 65 size_t offs = le16_to_cpu(entry->e_value_offs); 66 if (offs < min_offs) 67 min_offs = offs; 68 } 69 } 70 free = min_offs - 71 ((void *)entry - (void *)IFIRST(header)) - sizeof(__u32); 72 73 if (EXT4_I(inode)->i_inline_off) { 74 entry = (struct ext4_xattr_entry *) 75 ((void *)raw_inode + EXT4_I(inode)->i_inline_off); 76 77 free += EXT4_XATTR_SIZE(le32_to_cpu(entry->e_value_size)); 78 goto out; 79 } 80 81 free -= EXT4_XATTR_LEN(strlen(EXT4_XATTR_SYSTEM_DATA)); 82 83 if (free > EXT4_XATTR_ROUND) 84 free = EXT4_XATTR_SIZE(free - EXT4_XATTR_ROUND); 85 else 86 free = 0; 87 88 out: 89 return free; 90 } 91 92 /* 93 * Get the maximum size we now can store in an inode. 94 * If we can't find the space for a xattr entry, don't use the space 95 * of the extents since we have no space to indicate the inline data. 96 */ 97 int ext4_get_max_inline_size(struct inode *inode) 98 { 99 int error, max_inline_size; 100 struct ext4_iloc iloc; 101 102 if (EXT4_I(inode)->i_extra_isize == 0) 103 return 0; 104 105 error = ext4_get_inode_loc(inode, &iloc); 106 if (error) { 107 ext4_error_inode(inode, __func__, __LINE__, 0, 108 "can't get inode location %lu", 109 inode->i_ino); 110 return 0; 111 } 112 113 down_read(&EXT4_I(inode)->xattr_sem); 114 max_inline_size = get_max_inline_xattr_value_size(inode, &iloc); 115 up_read(&EXT4_I(inode)->xattr_sem); 116 117 brelse(iloc.bh); 118 119 if (!max_inline_size) 120 return 0; 121 122 return max_inline_size + EXT4_MIN_INLINE_DATA_SIZE; 123 } 124 125 /* 126 * this function does not take xattr_sem, which is OK because it is 127 * currently only used in a code path coming form ext4_iget, before 128 * the new inode has been unlocked 129 */ 130 int ext4_find_inline_data_nolock(struct inode *inode) 131 { 132 struct ext4_xattr_ibody_find is = { 133 .s = { .not_found = -ENODATA, }, 134 }; 135 struct ext4_xattr_info i = { 136 .name_index = EXT4_XATTR_INDEX_SYSTEM, 137 .name = EXT4_XATTR_SYSTEM_DATA, 138 }; 139 int error; 140 141 if (EXT4_I(inode)->i_extra_isize == 0) 142 return 0; 143 144 error = ext4_get_inode_loc(inode, &is.iloc); 145 if (error) 146 return error; 147 148 error = ext4_xattr_ibody_find(inode, &i, &is); 149 if (error) 150 goto out; 151 152 if (!is.s.not_found) { 153 EXT4_I(inode)->i_inline_off = (u16)((void *)is.s.here - 154 (void *)ext4_raw_inode(&is.iloc)); 155 EXT4_I(inode)->i_inline_size = EXT4_MIN_INLINE_DATA_SIZE + 156 le32_to_cpu(is.s.here->e_value_size); 157 ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA); 158 } 159 out: 160 brelse(is.iloc.bh); 161 return error; 162 } 163 164 static int ext4_read_inline_data(struct inode *inode, void *buffer, 165 unsigned int len, 166 struct ext4_iloc *iloc) 167 { 168 struct ext4_xattr_entry *entry; 169 struct ext4_xattr_ibody_header *header; 170 int cp_len = 0; 171 struct ext4_inode *raw_inode; 172 173 if (!len) 174 return 0; 175 176 BUG_ON(len > EXT4_I(inode)->i_inline_size); 177 178 cp_len = len < EXT4_MIN_INLINE_DATA_SIZE ? 179 len : EXT4_MIN_INLINE_DATA_SIZE; 180 181 raw_inode = ext4_raw_inode(iloc); 182 memcpy(buffer, (void *)(raw_inode->i_block), cp_len); 183 184 len -= cp_len; 185 buffer += cp_len; 186 187 if (!len) 188 goto out; 189 190 header = IHDR(inode, raw_inode); 191 entry = (struct ext4_xattr_entry *)((void *)raw_inode + 192 EXT4_I(inode)->i_inline_off); 193 len = min_t(unsigned int, len, 194 (unsigned int)le32_to_cpu(entry->e_value_size)); 195 196 memcpy(buffer, 197 (void *)IFIRST(header) + le16_to_cpu(entry->e_value_offs), len); 198 cp_len += len; 199 200 out: 201 return cp_len; 202 } 203 204 /* 205 * write the buffer to the inline inode. 206 * If 'create' is set, we don't need to do the extra copy in the xattr 207 * value since it is already handled by ext4_xattr_ibody_inline_set. 208 * That saves us one memcpy. 209 */ 210 static void ext4_write_inline_data(struct inode *inode, struct ext4_iloc *iloc, 211 void *buffer, loff_t pos, unsigned int len) 212 { 213 struct ext4_xattr_entry *entry; 214 struct ext4_xattr_ibody_header *header; 215 struct ext4_inode *raw_inode; 216 int cp_len = 0; 217 218 BUG_ON(!EXT4_I(inode)->i_inline_off); 219 BUG_ON(pos + len > EXT4_I(inode)->i_inline_size); 220 221 raw_inode = ext4_raw_inode(iloc); 222 buffer += pos; 223 224 if (pos < EXT4_MIN_INLINE_DATA_SIZE) { 225 cp_len = pos + len > EXT4_MIN_INLINE_DATA_SIZE ? 226 EXT4_MIN_INLINE_DATA_SIZE - pos : len; 227 memcpy((void *)raw_inode->i_block + pos, buffer, cp_len); 228 229 len -= cp_len; 230 buffer += cp_len; 231 pos += cp_len; 232 } 233 234 if (!len) 235 return; 236 237 pos -= EXT4_MIN_INLINE_DATA_SIZE; 238 header = IHDR(inode, raw_inode); 239 entry = (struct ext4_xattr_entry *)((void *)raw_inode + 240 EXT4_I(inode)->i_inline_off); 241 242 memcpy((void *)IFIRST(header) + le16_to_cpu(entry->e_value_offs) + pos, 243 buffer, len); 244 } 245 246 static int ext4_create_inline_data(handle_t *handle, 247 struct inode *inode, unsigned len) 248 { 249 int error; 250 void *value = NULL; 251 struct ext4_xattr_ibody_find is = { 252 .s = { .not_found = -ENODATA, }, 253 }; 254 struct ext4_xattr_info i = { 255 .name_index = EXT4_XATTR_INDEX_SYSTEM, 256 .name = EXT4_XATTR_SYSTEM_DATA, 257 }; 258 259 error = ext4_get_inode_loc(inode, &is.iloc); 260 if (error) 261 return error; 262 263 BUFFER_TRACE(is.iloc.bh, "get_write_access"); 264 error = ext4_journal_get_write_access(handle, is.iloc.bh); 265 if (error) 266 goto out; 267 268 if (len > EXT4_MIN_INLINE_DATA_SIZE) { 269 value = EXT4_ZERO_XATTR_VALUE; 270 len -= EXT4_MIN_INLINE_DATA_SIZE; 271 } else { 272 value = ""; 273 len = 0; 274 } 275 276 /* Insert the the xttr entry. */ 277 i.value = value; 278 i.value_len = len; 279 280 error = ext4_xattr_ibody_find(inode, &i, &is); 281 if (error) 282 goto out; 283 284 BUG_ON(!is.s.not_found); 285 286 error = ext4_xattr_ibody_inline_set(handle, inode, &i, &is); 287 if (error) { 288 if (error == -ENOSPC) 289 ext4_clear_inode_state(inode, 290 EXT4_STATE_MAY_INLINE_DATA); 291 goto out; 292 } 293 294 memset((void *)ext4_raw_inode(&is.iloc)->i_block, 295 0, EXT4_MIN_INLINE_DATA_SIZE); 296 297 EXT4_I(inode)->i_inline_off = (u16)((void *)is.s.here - 298 (void *)ext4_raw_inode(&is.iloc)); 299 EXT4_I(inode)->i_inline_size = len + EXT4_MIN_INLINE_DATA_SIZE; 300 ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS); 301 ext4_set_inode_flag(inode, EXT4_INODE_INLINE_DATA); 302 get_bh(is.iloc.bh); 303 error = ext4_mark_iloc_dirty(handle, inode, &is.iloc); 304 305 out: 306 brelse(is.iloc.bh); 307 return error; 308 } 309 310 static int ext4_update_inline_data(handle_t *handle, struct inode *inode, 311 unsigned int len) 312 { 313 int error; 314 void *value = NULL; 315 struct ext4_xattr_ibody_find is = { 316 .s = { .not_found = -ENODATA, }, 317 }; 318 struct ext4_xattr_info i = { 319 .name_index = EXT4_XATTR_INDEX_SYSTEM, 320 .name = EXT4_XATTR_SYSTEM_DATA, 321 }; 322 323 /* If the old space is ok, write the data directly. */ 324 if (len <= EXT4_I(inode)->i_inline_size) 325 return 0; 326 327 error = ext4_get_inode_loc(inode, &is.iloc); 328 if (error) 329 return error; 330 331 error = ext4_xattr_ibody_find(inode, &i, &is); 332 if (error) 333 goto out; 334 335 BUG_ON(is.s.not_found); 336 337 len -= EXT4_MIN_INLINE_DATA_SIZE; 338 value = kzalloc(len, GFP_NOFS); 339 if (!value) 340 goto out; 341 342 error = ext4_xattr_ibody_get(inode, i.name_index, i.name, 343 value, len); 344 if (error == -ENODATA) 345 goto out; 346 347 BUFFER_TRACE(is.iloc.bh, "get_write_access"); 348 error = ext4_journal_get_write_access(handle, is.iloc.bh); 349 if (error) 350 goto out; 351 352 /* Update the xttr entry. */ 353 i.value = value; 354 i.value_len = len; 355 356 error = ext4_xattr_ibody_inline_set(handle, inode, &i, &is); 357 if (error) 358 goto out; 359 360 EXT4_I(inode)->i_inline_off = (u16)((void *)is.s.here - 361 (void *)ext4_raw_inode(&is.iloc)); 362 EXT4_I(inode)->i_inline_size = EXT4_MIN_INLINE_DATA_SIZE + 363 le32_to_cpu(is.s.here->e_value_size); 364 ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA); 365 get_bh(is.iloc.bh); 366 error = ext4_mark_iloc_dirty(handle, inode, &is.iloc); 367 368 out: 369 kfree(value); 370 brelse(is.iloc.bh); 371 return error; 372 } 373 374 static int ext4_prepare_inline_data(handle_t *handle, struct inode *inode, 375 unsigned int len) 376 { 377 int ret, size; 378 struct ext4_inode_info *ei = EXT4_I(inode); 379 380 if (!ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) 381 return -ENOSPC; 382 383 size = ext4_get_max_inline_size(inode); 384 if (size < len) 385 return -ENOSPC; 386 387 down_write(&EXT4_I(inode)->xattr_sem); 388 389 if (ei->i_inline_off) 390 ret = ext4_update_inline_data(handle, inode, len); 391 else 392 ret = ext4_create_inline_data(handle, inode, len); 393 394 up_write(&EXT4_I(inode)->xattr_sem); 395 396 return ret; 397 } 398 399 static int ext4_destroy_inline_data_nolock(handle_t *handle, 400 struct inode *inode) 401 { 402 struct ext4_inode_info *ei = EXT4_I(inode); 403 struct ext4_xattr_ibody_find is = { 404 .s = { .not_found = 0, }, 405 }; 406 struct ext4_xattr_info i = { 407 .name_index = EXT4_XATTR_INDEX_SYSTEM, 408 .name = EXT4_XATTR_SYSTEM_DATA, 409 .value = NULL, 410 .value_len = 0, 411 }; 412 int error; 413 414 if (!ei->i_inline_off) 415 return 0; 416 417 error = ext4_get_inode_loc(inode, &is.iloc); 418 if (error) 419 return error; 420 421 error = ext4_xattr_ibody_find(inode, &i, &is); 422 if (error) 423 goto out; 424 425 BUFFER_TRACE(is.iloc.bh, "get_write_access"); 426 error = ext4_journal_get_write_access(handle, is.iloc.bh); 427 if (error) 428 goto out; 429 430 error = ext4_xattr_ibody_inline_set(handle, inode, &i, &is); 431 if (error) 432 goto out; 433 434 memset((void *)ext4_raw_inode(&is.iloc)->i_block, 435 0, EXT4_MIN_INLINE_DATA_SIZE); 436 437 if (EXT4_HAS_INCOMPAT_FEATURE(inode->i_sb, 438 EXT4_FEATURE_INCOMPAT_EXTENTS)) { 439 if (S_ISDIR(inode->i_mode) || 440 S_ISREG(inode->i_mode) || S_ISLNK(inode->i_mode)) { 441 ext4_set_inode_flag(inode, EXT4_INODE_EXTENTS); 442 ext4_ext_tree_init(handle, inode); 443 } 444 } 445 ext4_clear_inode_flag(inode, EXT4_INODE_INLINE_DATA); 446 447 get_bh(is.iloc.bh); 448 error = ext4_mark_iloc_dirty(handle, inode, &is.iloc); 449 450 EXT4_I(inode)->i_inline_off = 0; 451 EXT4_I(inode)->i_inline_size = 0; 452 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA); 453 out: 454 brelse(is.iloc.bh); 455 if (error == -ENODATA) 456 error = 0; 457 return error; 458 } 459 460 static int ext4_read_inline_page(struct inode *inode, struct page *page) 461 { 462 void *kaddr; 463 int ret = 0; 464 size_t len; 465 struct ext4_iloc iloc; 466 467 BUG_ON(!PageLocked(page)); 468 BUG_ON(!ext4_has_inline_data(inode)); 469 BUG_ON(page->index); 470 471 if (!EXT4_I(inode)->i_inline_off) { 472 ext4_warning(inode->i_sb, "inode %lu doesn't have inline data.", 473 inode->i_ino); 474 goto out; 475 } 476 477 ret = ext4_get_inode_loc(inode, &iloc); 478 if (ret) 479 goto out; 480 481 len = min_t(size_t, ext4_get_inline_size(inode), i_size_read(inode)); 482 kaddr = kmap_atomic(page); 483 ret = ext4_read_inline_data(inode, kaddr, len, &iloc); 484 flush_dcache_page(page); 485 kunmap_atomic(kaddr); 486 zero_user_segment(page, len, PAGE_CACHE_SIZE); 487 SetPageUptodate(page); 488 brelse(iloc.bh); 489 490 out: 491 return ret; 492 } 493 494 int ext4_readpage_inline(struct inode *inode, struct page *page) 495 { 496 int ret = 0; 497 498 down_read(&EXT4_I(inode)->xattr_sem); 499 if (!ext4_has_inline_data(inode)) { 500 up_read(&EXT4_I(inode)->xattr_sem); 501 return -EAGAIN; 502 } 503 504 /* 505 * Current inline data can only exist in the 1st page, 506 * So for all the other pages, just set them uptodate. 507 */ 508 if (!page->index) 509 ret = ext4_read_inline_page(inode, page); 510 else if (!PageUptodate(page)) { 511 zero_user_segment(page, 0, PAGE_CACHE_SIZE); 512 SetPageUptodate(page); 513 } 514 515 up_read(&EXT4_I(inode)->xattr_sem); 516 517 unlock_page(page); 518 return ret >= 0 ? 0 : ret; 519 } 520 521 static int ext4_convert_inline_data_to_extent(struct address_space *mapping, 522 struct inode *inode, 523 unsigned flags) 524 { 525 int ret, needed_blocks; 526 handle_t *handle = NULL; 527 int retries = 0, sem_held = 0; 528 struct page *page = NULL; 529 unsigned from, to; 530 struct ext4_iloc iloc; 531 532 if (!ext4_has_inline_data(inode)) { 533 /* 534 * clear the flag so that no new write 535 * will trap here again. 536 */ 537 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA); 538 return 0; 539 } 540 541 needed_blocks = ext4_writepage_trans_blocks(inode); 542 543 ret = ext4_get_inode_loc(inode, &iloc); 544 if (ret) 545 return ret; 546 547 retry: 548 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks); 549 if (IS_ERR(handle)) { 550 ret = PTR_ERR(handle); 551 handle = NULL; 552 goto out; 553 } 554 555 /* We cannot recurse into the filesystem as the transaction is already 556 * started */ 557 flags |= AOP_FLAG_NOFS; 558 559 page = grab_cache_page_write_begin(mapping, 0, flags); 560 if (!page) { 561 ret = -ENOMEM; 562 goto out; 563 } 564 565 down_write(&EXT4_I(inode)->xattr_sem); 566 sem_held = 1; 567 /* If some one has already done this for us, just exit. */ 568 if (!ext4_has_inline_data(inode)) { 569 ret = 0; 570 goto out; 571 } 572 573 from = 0; 574 to = ext4_get_inline_size(inode); 575 if (!PageUptodate(page)) { 576 ret = ext4_read_inline_page(inode, page); 577 if (ret < 0) 578 goto out; 579 } 580 581 ret = ext4_destroy_inline_data_nolock(handle, inode); 582 if (ret) 583 goto out; 584 585 if (ext4_should_dioread_nolock(inode)) 586 ret = __block_write_begin(page, from, to, ext4_get_block_write); 587 else 588 ret = __block_write_begin(page, from, to, ext4_get_block); 589 590 if (!ret && ext4_should_journal_data(inode)) { 591 ret = ext4_walk_page_buffers(handle, page_buffers(page), 592 from, to, NULL, 593 do_journal_get_write_access); 594 } 595 596 if (ret) { 597 unlock_page(page); 598 page_cache_release(page); 599 page = NULL; 600 ext4_orphan_add(handle, inode); 601 up_write(&EXT4_I(inode)->xattr_sem); 602 sem_held = 0; 603 ext4_journal_stop(handle); 604 handle = NULL; 605 ext4_truncate_failed_write(inode); 606 /* 607 * If truncate failed early the inode might 608 * still be on the orphan list; we need to 609 * make sure the inode is removed from the 610 * orphan list in that case. 611 */ 612 if (inode->i_nlink) 613 ext4_orphan_del(NULL, inode); 614 } 615 616 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries)) 617 goto retry; 618 619 if (page) 620 block_commit_write(page, from, to); 621 out: 622 if (page) { 623 unlock_page(page); 624 page_cache_release(page); 625 } 626 if (sem_held) 627 up_write(&EXT4_I(inode)->xattr_sem); 628 if (handle) 629 ext4_journal_stop(handle); 630 brelse(iloc.bh); 631 return ret; 632 } 633 634 /* 635 * Try to write data in the inode. 636 * If the inode has inline data, check whether the new write can be 637 * in the inode also. If not, create the page the handle, move the data 638 * to the page make it update and let the later codes create extent for it. 639 */ 640 int ext4_try_to_write_inline_data(struct address_space *mapping, 641 struct inode *inode, 642 loff_t pos, unsigned len, 643 unsigned flags, 644 struct page **pagep) 645 { 646 int ret; 647 handle_t *handle; 648 struct page *page; 649 struct ext4_iloc iloc; 650 651 if (pos + len > ext4_get_max_inline_size(inode)) 652 goto convert; 653 654 ret = ext4_get_inode_loc(inode, &iloc); 655 if (ret) 656 return ret; 657 658 /* 659 * The possible write could happen in the inode, 660 * so try to reserve the space in inode first. 661 */ 662 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1); 663 if (IS_ERR(handle)) { 664 ret = PTR_ERR(handle); 665 handle = NULL; 666 goto out; 667 } 668 669 ret = ext4_prepare_inline_data(handle, inode, pos + len); 670 if (ret && ret != -ENOSPC) 671 goto out; 672 673 /* We don't have space in inline inode, so convert it to extent. */ 674 if (ret == -ENOSPC) { 675 ext4_journal_stop(handle); 676 brelse(iloc.bh); 677 goto convert; 678 } 679 680 flags |= AOP_FLAG_NOFS; 681 682 page = grab_cache_page_write_begin(mapping, 0, flags); 683 if (!page) { 684 ret = -ENOMEM; 685 goto out; 686 } 687 688 *pagep = page; 689 down_read(&EXT4_I(inode)->xattr_sem); 690 if (!ext4_has_inline_data(inode)) { 691 ret = 0; 692 unlock_page(page); 693 page_cache_release(page); 694 goto out_up_read; 695 } 696 697 if (!PageUptodate(page)) { 698 ret = ext4_read_inline_page(inode, page); 699 if (ret < 0) 700 goto out_up_read; 701 } 702 703 ret = 1; 704 handle = NULL; 705 out_up_read: 706 up_read(&EXT4_I(inode)->xattr_sem); 707 out: 708 if (handle) 709 ext4_journal_stop(handle); 710 brelse(iloc.bh); 711 return ret; 712 convert: 713 return ext4_convert_inline_data_to_extent(mapping, 714 inode, flags); 715 } 716 717 int ext4_write_inline_data_end(struct inode *inode, loff_t pos, unsigned len, 718 unsigned copied, struct page *page) 719 { 720 int ret; 721 void *kaddr; 722 struct ext4_iloc iloc; 723 724 if (unlikely(copied < len)) { 725 if (!PageUptodate(page)) { 726 copied = 0; 727 goto out; 728 } 729 } 730 731 ret = ext4_get_inode_loc(inode, &iloc); 732 if (ret) { 733 ext4_std_error(inode->i_sb, ret); 734 copied = 0; 735 goto out; 736 } 737 738 down_write(&EXT4_I(inode)->xattr_sem); 739 BUG_ON(!ext4_has_inline_data(inode)); 740 741 kaddr = kmap_atomic(page); 742 ext4_write_inline_data(inode, &iloc, kaddr, pos, len); 743 kunmap_atomic(kaddr); 744 SetPageUptodate(page); 745 /* clear page dirty so that writepages wouldn't work for us. */ 746 ClearPageDirty(page); 747 748 up_write(&EXT4_I(inode)->xattr_sem); 749 brelse(iloc.bh); 750 out: 751 return copied; 752 } 753 754 struct buffer_head * 755 ext4_journalled_write_inline_data(struct inode *inode, 756 unsigned len, 757 struct page *page) 758 { 759 int ret; 760 void *kaddr; 761 struct ext4_iloc iloc; 762 763 ret = ext4_get_inode_loc(inode, &iloc); 764 if (ret) { 765 ext4_std_error(inode->i_sb, ret); 766 return NULL; 767 } 768 769 down_write(&EXT4_I(inode)->xattr_sem); 770 kaddr = kmap_atomic(page); 771 ext4_write_inline_data(inode, &iloc, kaddr, 0, len); 772 kunmap_atomic(kaddr); 773 up_write(&EXT4_I(inode)->xattr_sem); 774 775 return iloc.bh; 776 } 777 778 /* 779 * Try to make the page cache and handle ready for the inline data case. 780 * We can call this function in 2 cases: 781 * 1. The inode is created and the first write exceeds inline size. We can 782 * clear the inode state safely. 783 * 2. The inode has inline data, then we need to read the data, make it 784 * update and dirty so that ext4_da_writepages can handle it. We don't 785 * need to start the journal since the file's metatdata isn't changed now. 786 */ 787 static int ext4_da_convert_inline_data_to_extent(struct address_space *mapping, 788 struct inode *inode, 789 unsigned flags, 790 void **fsdata) 791 { 792 int ret = 0, inline_size; 793 struct page *page; 794 795 page = grab_cache_page_write_begin(mapping, 0, flags); 796 if (!page) 797 return -ENOMEM; 798 799 down_read(&EXT4_I(inode)->xattr_sem); 800 if (!ext4_has_inline_data(inode)) { 801 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA); 802 goto out; 803 } 804 805 inline_size = ext4_get_inline_size(inode); 806 807 if (!PageUptodate(page)) { 808 ret = ext4_read_inline_page(inode, page); 809 if (ret < 0) 810 goto out; 811 } 812 813 ret = __block_write_begin(page, 0, inline_size, 814 ext4_da_get_block_prep); 815 if (ret) { 816 up_read(&EXT4_I(inode)->xattr_sem); 817 unlock_page(page); 818 page_cache_release(page); 819 ext4_truncate_failed_write(inode); 820 return ret; 821 } 822 823 SetPageDirty(page); 824 SetPageUptodate(page); 825 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA); 826 *fsdata = (void *)CONVERT_INLINE_DATA; 827 828 out: 829 up_read(&EXT4_I(inode)->xattr_sem); 830 if (page) { 831 unlock_page(page); 832 page_cache_release(page); 833 } 834 return ret; 835 } 836 837 /* 838 * Prepare the write for the inline data. 839 * If the the data can be written into the inode, we just read 840 * the page and make it uptodate, and start the journal. 841 * Otherwise read the page, makes it dirty so that it can be 842 * handle in writepages(the i_disksize update is left to the 843 * normal ext4_da_write_end). 844 */ 845 int ext4_da_write_inline_data_begin(struct address_space *mapping, 846 struct inode *inode, 847 loff_t pos, unsigned len, 848 unsigned flags, 849 struct page **pagep, 850 void **fsdata) 851 { 852 int ret, inline_size; 853 handle_t *handle; 854 struct page *page; 855 struct ext4_iloc iloc; 856 int retries; 857 858 ret = ext4_get_inode_loc(inode, &iloc); 859 if (ret) 860 return ret; 861 862 retry_journal: 863 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1); 864 if (IS_ERR(handle)) { 865 ret = PTR_ERR(handle); 866 goto out; 867 } 868 869 inline_size = ext4_get_max_inline_size(inode); 870 871 ret = -ENOSPC; 872 if (inline_size >= pos + len) { 873 ret = ext4_prepare_inline_data(handle, inode, pos + len); 874 if (ret && ret != -ENOSPC) 875 goto out_journal; 876 } 877 878 /* 879 * We cannot recurse into the filesystem as the transaction 880 * is already started. 881 */ 882 flags |= AOP_FLAG_NOFS; 883 884 if (ret == -ENOSPC) { 885 ret = ext4_da_convert_inline_data_to_extent(mapping, 886 inode, 887 flags, 888 fsdata); 889 ext4_journal_stop(handle); 890 if (ret == -ENOSPC && 891 ext4_should_retry_alloc(inode->i_sb, &retries)) 892 goto retry_journal; 893 goto out; 894 } 895 896 897 page = grab_cache_page_write_begin(mapping, 0, flags); 898 if (!page) { 899 ret = -ENOMEM; 900 goto out_journal; 901 } 902 903 down_read(&EXT4_I(inode)->xattr_sem); 904 if (!ext4_has_inline_data(inode)) { 905 ret = 0; 906 goto out_release_page; 907 } 908 909 if (!PageUptodate(page)) { 910 ret = ext4_read_inline_page(inode, page); 911 if (ret < 0) 912 goto out_release_page; 913 } 914 915 up_read(&EXT4_I(inode)->xattr_sem); 916 *pagep = page; 917 brelse(iloc.bh); 918 return 1; 919 out_release_page: 920 up_read(&EXT4_I(inode)->xattr_sem); 921 unlock_page(page); 922 page_cache_release(page); 923 out_journal: 924 ext4_journal_stop(handle); 925 out: 926 brelse(iloc.bh); 927 return ret; 928 } 929 930 int ext4_da_write_inline_data_end(struct inode *inode, loff_t pos, 931 unsigned len, unsigned copied, 932 struct page *page) 933 { 934 int i_size_changed = 0; 935 936 copied = ext4_write_inline_data_end(inode, pos, len, copied, page); 937 938 /* 939 * No need to use i_size_read() here, the i_size 940 * cannot change under us because we hold i_mutex. 941 * 942 * But it's important to update i_size while still holding page lock: 943 * page writeout could otherwise come in and zero beyond i_size. 944 */ 945 if (pos+copied > inode->i_size) { 946 i_size_write(inode, pos+copied); 947 i_size_changed = 1; 948 } 949 unlock_page(page); 950 page_cache_release(page); 951 952 /* 953 * Don't mark the inode dirty under page lock. First, it unnecessarily 954 * makes the holding time of page lock longer. Second, it forces lock 955 * ordering of page lock and transaction start for journaling 956 * filesystems. 957 */ 958 if (i_size_changed) 959 mark_inode_dirty(inode); 960 961 return copied; 962 } 963 964 #ifdef INLINE_DIR_DEBUG 965 void ext4_show_inline_dir(struct inode *dir, struct buffer_head *bh, 966 void *inline_start, int inline_size) 967 { 968 int offset; 969 unsigned short de_len; 970 struct ext4_dir_entry_2 *de = inline_start; 971 void *dlimit = inline_start + inline_size; 972 973 trace_printk("inode %lu\n", dir->i_ino); 974 offset = 0; 975 while ((void *)de < dlimit) { 976 de_len = ext4_rec_len_from_disk(de->rec_len, inline_size); 977 trace_printk("de: off %u rlen %u name %.*s nlen %u ino %u\n", 978 offset, de_len, de->name_len, de->name, 979 de->name_len, le32_to_cpu(de->inode)); 980 if (ext4_check_dir_entry(dir, NULL, de, bh, 981 inline_start, inline_size, offset)) 982 BUG(); 983 984 offset += de_len; 985 de = (struct ext4_dir_entry_2 *) ((char *) de + de_len); 986 } 987 } 988 #else 989 #define ext4_show_inline_dir(dir, bh, inline_start, inline_size) 990 #endif 991 992 /* 993 * Add a new entry into a inline dir. 994 * It will return -ENOSPC if no space is available, and -EIO 995 * and -EEXIST if directory entry already exists. 996 */ 997 static int ext4_add_dirent_to_inline(handle_t *handle, 998 struct ext4_filename *fname, 999 struct dentry *dentry, 1000 struct inode *inode, 1001 struct ext4_iloc *iloc, 1002 void *inline_start, int inline_size) 1003 { 1004 struct inode *dir = d_inode(dentry->d_parent); 1005 int err; 1006 struct ext4_dir_entry_2 *de; 1007 1008 err = ext4_find_dest_de(dir, inode, iloc->bh, inline_start, 1009 inline_size, fname, &de); 1010 if (err) 1011 return err; 1012 1013 BUFFER_TRACE(iloc->bh, "get_write_access"); 1014 err = ext4_journal_get_write_access(handle, iloc->bh); 1015 if (err) 1016 return err; 1017 ext4_insert_dentry(dir, inode, de, inline_size, fname); 1018 1019 ext4_show_inline_dir(dir, iloc->bh, inline_start, inline_size); 1020 1021 /* 1022 * XXX shouldn't update any times until successful 1023 * completion of syscall, but too many callers depend 1024 * on this. 1025 * 1026 * XXX similarly, too many callers depend on 1027 * ext4_new_inode() setting the times, but error 1028 * recovery deletes the inode, so the worst that can 1029 * happen is that the times are slightly out of date 1030 * and/or different from the directory change time. 1031 */ 1032 dir->i_mtime = dir->i_ctime = ext4_current_time(dir); 1033 ext4_update_dx_flag(dir); 1034 dir->i_version++; 1035 ext4_mark_inode_dirty(handle, dir); 1036 return 1; 1037 } 1038 1039 static void *ext4_get_inline_xattr_pos(struct inode *inode, 1040 struct ext4_iloc *iloc) 1041 { 1042 struct ext4_xattr_entry *entry; 1043 struct ext4_xattr_ibody_header *header; 1044 1045 BUG_ON(!EXT4_I(inode)->i_inline_off); 1046 1047 header = IHDR(inode, ext4_raw_inode(iloc)); 1048 entry = (struct ext4_xattr_entry *)((void *)ext4_raw_inode(iloc) + 1049 EXT4_I(inode)->i_inline_off); 1050 1051 return (void *)IFIRST(header) + le16_to_cpu(entry->e_value_offs); 1052 } 1053 1054 /* Set the final de to cover the whole block. */ 1055 static void ext4_update_final_de(void *de_buf, int old_size, int new_size) 1056 { 1057 struct ext4_dir_entry_2 *de, *prev_de; 1058 void *limit; 1059 int de_len; 1060 1061 de = (struct ext4_dir_entry_2 *)de_buf; 1062 if (old_size) { 1063 limit = de_buf + old_size; 1064 do { 1065 prev_de = de; 1066 de_len = ext4_rec_len_from_disk(de->rec_len, old_size); 1067 de_buf += de_len; 1068 de = (struct ext4_dir_entry_2 *)de_buf; 1069 } while (de_buf < limit); 1070 1071 prev_de->rec_len = ext4_rec_len_to_disk(de_len + new_size - 1072 old_size, new_size); 1073 } else { 1074 /* this is just created, so create an empty entry. */ 1075 de->inode = 0; 1076 de->rec_len = ext4_rec_len_to_disk(new_size, new_size); 1077 } 1078 } 1079 1080 static int ext4_update_inline_dir(handle_t *handle, struct inode *dir, 1081 struct ext4_iloc *iloc) 1082 { 1083 int ret; 1084 int old_size = EXT4_I(dir)->i_inline_size - EXT4_MIN_INLINE_DATA_SIZE; 1085 int new_size = get_max_inline_xattr_value_size(dir, iloc); 1086 1087 if (new_size - old_size <= EXT4_DIR_REC_LEN(1)) 1088 return -ENOSPC; 1089 1090 ret = ext4_update_inline_data(handle, dir, 1091 new_size + EXT4_MIN_INLINE_DATA_SIZE); 1092 if (ret) 1093 return ret; 1094 1095 ext4_update_final_de(ext4_get_inline_xattr_pos(dir, iloc), old_size, 1096 EXT4_I(dir)->i_inline_size - 1097 EXT4_MIN_INLINE_DATA_SIZE); 1098 dir->i_size = EXT4_I(dir)->i_disksize = EXT4_I(dir)->i_inline_size; 1099 return 0; 1100 } 1101 1102 static void ext4_restore_inline_data(handle_t *handle, struct inode *inode, 1103 struct ext4_iloc *iloc, 1104 void *buf, int inline_size) 1105 { 1106 ext4_create_inline_data(handle, inode, inline_size); 1107 ext4_write_inline_data(inode, iloc, buf, 0, inline_size); 1108 ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA); 1109 } 1110 1111 static int ext4_finish_convert_inline_dir(handle_t *handle, 1112 struct inode *inode, 1113 struct buffer_head *dir_block, 1114 void *buf, 1115 int inline_size) 1116 { 1117 int err, csum_size = 0, header_size = 0; 1118 struct ext4_dir_entry_2 *de; 1119 struct ext4_dir_entry_tail *t; 1120 void *target = dir_block->b_data; 1121 1122 /* 1123 * First create "." and ".." and then copy the dir information 1124 * back to the block. 1125 */ 1126 de = (struct ext4_dir_entry_2 *)target; 1127 de = ext4_init_dot_dotdot(inode, de, 1128 inode->i_sb->s_blocksize, csum_size, 1129 le32_to_cpu(((struct ext4_dir_entry_2 *)buf)->inode), 1); 1130 header_size = (void *)de - target; 1131 1132 memcpy((void *)de, buf + EXT4_INLINE_DOTDOT_SIZE, 1133 inline_size - EXT4_INLINE_DOTDOT_SIZE); 1134 1135 if (ext4_has_metadata_csum(inode->i_sb)) 1136 csum_size = sizeof(struct ext4_dir_entry_tail); 1137 1138 inode->i_size = inode->i_sb->s_blocksize; 1139 i_size_write(inode, inode->i_sb->s_blocksize); 1140 EXT4_I(inode)->i_disksize = inode->i_sb->s_blocksize; 1141 ext4_update_final_de(dir_block->b_data, 1142 inline_size - EXT4_INLINE_DOTDOT_SIZE + header_size, 1143 inode->i_sb->s_blocksize - csum_size); 1144 1145 if (csum_size) { 1146 t = EXT4_DIRENT_TAIL(dir_block->b_data, 1147 inode->i_sb->s_blocksize); 1148 initialize_dirent_tail(t, inode->i_sb->s_blocksize); 1149 } 1150 set_buffer_uptodate(dir_block); 1151 err = ext4_handle_dirty_dirent_node(handle, inode, dir_block); 1152 if (err) 1153 goto out; 1154 set_buffer_verified(dir_block); 1155 out: 1156 return err; 1157 } 1158 1159 static int ext4_convert_inline_data_nolock(handle_t *handle, 1160 struct inode *inode, 1161 struct ext4_iloc *iloc) 1162 { 1163 int error; 1164 void *buf = NULL; 1165 struct buffer_head *data_bh = NULL; 1166 struct ext4_map_blocks map; 1167 int inline_size; 1168 1169 inline_size = ext4_get_inline_size(inode); 1170 buf = kmalloc(inline_size, GFP_NOFS); 1171 if (!buf) { 1172 error = -ENOMEM; 1173 goto out; 1174 } 1175 1176 error = ext4_read_inline_data(inode, buf, inline_size, iloc); 1177 if (error < 0) 1178 goto out; 1179 1180 /* 1181 * Make sure the inline directory entries pass checks before we try to 1182 * convert them, so that we avoid touching stuff that needs fsck. 1183 */ 1184 if (S_ISDIR(inode->i_mode)) { 1185 error = ext4_check_all_de(inode, iloc->bh, 1186 buf + EXT4_INLINE_DOTDOT_SIZE, 1187 inline_size - EXT4_INLINE_DOTDOT_SIZE); 1188 if (error) 1189 goto out; 1190 } 1191 1192 error = ext4_destroy_inline_data_nolock(handle, inode); 1193 if (error) 1194 goto out; 1195 1196 map.m_lblk = 0; 1197 map.m_len = 1; 1198 map.m_flags = 0; 1199 error = ext4_map_blocks(handle, inode, &map, EXT4_GET_BLOCKS_CREATE); 1200 if (error < 0) 1201 goto out_restore; 1202 if (!(map.m_flags & EXT4_MAP_MAPPED)) { 1203 error = -EIO; 1204 goto out_restore; 1205 } 1206 1207 data_bh = sb_getblk(inode->i_sb, map.m_pblk); 1208 if (!data_bh) { 1209 error = -ENOMEM; 1210 goto out_restore; 1211 } 1212 1213 lock_buffer(data_bh); 1214 error = ext4_journal_get_create_access(handle, data_bh); 1215 if (error) { 1216 unlock_buffer(data_bh); 1217 error = -EIO; 1218 goto out_restore; 1219 } 1220 memset(data_bh->b_data, 0, inode->i_sb->s_blocksize); 1221 1222 if (!S_ISDIR(inode->i_mode)) { 1223 memcpy(data_bh->b_data, buf, inline_size); 1224 set_buffer_uptodate(data_bh); 1225 error = ext4_handle_dirty_metadata(handle, 1226 inode, data_bh); 1227 } else { 1228 error = ext4_finish_convert_inline_dir(handle, inode, data_bh, 1229 buf, inline_size); 1230 } 1231 1232 unlock_buffer(data_bh); 1233 out_restore: 1234 if (error) 1235 ext4_restore_inline_data(handle, inode, iloc, buf, inline_size); 1236 1237 out: 1238 brelse(data_bh); 1239 kfree(buf); 1240 return error; 1241 } 1242 1243 /* 1244 * Try to add the new entry to the inline data. 1245 * If succeeds, return 0. If not, extended the inline dir and copied data to 1246 * the new created block. 1247 */ 1248 int ext4_try_add_inline_entry(handle_t *handle, struct ext4_filename *fname, 1249 struct dentry *dentry, struct inode *inode) 1250 { 1251 int ret, inline_size; 1252 void *inline_start; 1253 struct ext4_iloc iloc; 1254 struct inode *dir = d_inode(dentry->d_parent); 1255 1256 ret = ext4_get_inode_loc(dir, &iloc); 1257 if (ret) 1258 return ret; 1259 1260 down_write(&EXT4_I(dir)->xattr_sem); 1261 if (!ext4_has_inline_data(dir)) 1262 goto out; 1263 1264 inline_start = (void *)ext4_raw_inode(&iloc)->i_block + 1265 EXT4_INLINE_DOTDOT_SIZE; 1266 inline_size = EXT4_MIN_INLINE_DATA_SIZE - EXT4_INLINE_DOTDOT_SIZE; 1267 1268 ret = ext4_add_dirent_to_inline(handle, fname, dentry, inode, &iloc, 1269 inline_start, inline_size); 1270 if (ret != -ENOSPC) 1271 goto out; 1272 1273 /* check whether it can be inserted to inline xattr space. */ 1274 inline_size = EXT4_I(dir)->i_inline_size - 1275 EXT4_MIN_INLINE_DATA_SIZE; 1276 if (!inline_size) { 1277 /* Try to use the xattr space.*/ 1278 ret = ext4_update_inline_dir(handle, dir, &iloc); 1279 if (ret && ret != -ENOSPC) 1280 goto out; 1281 1282 inline_size = EXT4_I(dir)->i_inline_size - 1283 EXT4_MIN_INLINE_DATA_SIZE; 1284 } 1285 1286 if (inline_size) { 1287 inline_start = ext4_get_inline_xattr_pos(dir, &iloc); 1288 1289 ret = ext4_add_dirent_to_inline(handle, fname, dentry, 1290 inode, &iloc, inline_start, 1291 inline_size); 1292 1293 if (ret != -ENOSPC) 1294 goto out; 1295 } 1296 1297 /* 1298 * The inline space is filled up, so create a new block for it. 1299 * As the extent tree will be created, we have to save the inline 1300 * dir first. 1301 */ 1302 ret = ext4_convert_inline_data_nolock(handle, dir, &iloc); 1303 1304 out: 1305 ext4_mark_inode_dirty(handle, dir); 1306 up_write(&EXT4_I(dir)->xattr_sem); 1307 brelse(iloc.bh); 1308 return ret; 1309 } 1310 1311 /* 1312 * This function fills a red-black tree with information from an 1313 * inlined dir. It returns the number directory entries loaded 1314 * into the tree. If there is an error it is returned in err. 1315 */ 1316 int htree_inlinedir_to_tree(struct file *dir_file, 1317 struct inode *dir, ext4_lblk_t block, 1318 struct dx_hash_info *hinfo, 1319 __u32 start_hash, __u32 start_minor_hash, 1320 int *has_inline_data) 1321 { 1322 int err = 0, count = 0; 1323 unsigned int parent_ino; 1324 int pos; 1325 struct ext4_dir_entry_2 *de; 1326 struct inode *inode = file_inode(dir_file); 1327 int ret, inline_size = 0; 1328 struct ext4_iloc iloc; 1329 void *dir_buf = NULL; 1330 struct ext4_dir_entry_2 fake; 1331 struct ext4_str tmp_str; 1332 1333 ret = ext4_get_inode_loc(inode, &iloc); 1334 if (ret) 1335 return ret; 1336 1337 down_read(&EXT4_I(inode)->xattr_sem); 1338 if (!ext4_has_inline_data(inode)) { 1339 up_read(&EXT4_I(inode)->xattr_sem); 1340 *has_inline_data = 0; 1341 goto out; 1342 } 1343 1344 inline_size = ext4_get_inline_size(inode); 1345 dir_buf = kmalloc(inline_size, GFP_NOFS); 1346 if (!dir_buf) { 1347 ret = -ENOMEM; 1348 up_read(&EXT4_I(inode)->xattr_sem); 1349 goto out; 1350 } 1351 1352 ret = ext4_read_inline_data(inode, dir_buf, inline_size, &iloc); 1353 up_read(&EXT4_I(inode)->xattr_sem); 1354 if (ret < 0) 1355 goto out; 1356 1357 pos = 0; 1358 parent_ino = le32_to_cpu(((struct ext4_dir_entry_2 *)dir_buf)->inode); 1359 while (pos < inline_size) { 1360 /* 1361 * As inlined dir doesn't store any information about '.' and 1362 * only the inode number of '..' is stored, we have to handle 1363 * them differently. 1364 */ 1365 if (pos == 0) { 1366 fake.inode = cpu_to_le32(inode->i_ino); 1367 fake.name_len = 1; 1368 strcpy(fake.name, "."); 1369 fake.rec_len = ext4_rec_len_to_disk( 1370 EXT4_DIR_REC_LEN(fake.name_len), 1371 inline_size); 1372 ext4_set_de_type(inode->i_sb, &fake, S_IFDIR); 1373 de = &fake; 1374 pos = EXT4_INLINE_DOTDOT_OFFSET; 1375 } else if (pos == EXT4_INLINE_DOTDOT_OFFSET) { 1376 fake.inode = cpu_to_le32(parent_ino); 1377 fake.name_len = 2; 1378 strcpy(fake.name, ".."); 1379 fake.rec_len = ext4_rec_len_to_disk( 1380 EXT4_DIR_REC_LEN(fake.name_len), 1381 inline_size); 1382 ext4_set_de_type(inode->i_sb, &fake, S_IFDIR); 1383 de = &fake; 1384 pos = EXT4_INLINE_DOTDOT_SIZE; 1385 } else { 1386 de = (struct ext4_dir_entry_2 *)(dir_buf + pos); 1387 pos += ext4_rec_len_from_disk(de->rec_len, inline_size); 1388 if (ext4_check_dir_entry(inode, dir_file, de, 1389 iloc.bh, dir_buf, 1390 inline_size, pos)) { 1391 ret = count; 1392 goto out; 1393 } 1394 } 1395 1396 ext4fs_dirhash(de->name, de->name_len, hinfo); 1397 if ((hinfo->hash < start_hash) || 1398 ((hinfo->hash == start_hash) && 1399 (hinfo->minor_hash < start_minor_hash))) 1400 continue; 1401 if (de->inode == 0) 1402 continue; 1403 tmp_str.name = de->name; 1404 tmp_str.len = de->name_len; 1405 err = ext4_htree_store_dirent(dir_file, hinfo->hash, 1406 hinfo->minor_hash, de, &tmp_str); 1407 if (err) { 1408 count = err; 1409 goto out; 1410 } 1411 count++; 1412 } 1413 ret = count; 1414 out: 1415 kfree(dir_buf); 1416 brelse(iloc.bh); 1417 return ret; 1418 } 1419 1420 /* 1421 * So this function is called when the volume is mkfsed with 1422 * dir_index disabled. In order to keep f_pos persistent 1423 * after we convert from an inlined dir to a blocked based, 1424 * we just pretend that we are a normal dir and return the 1425 * offset as if '.' and '..' really take place. 1426 * 1427 */ 1428 int ext4_read_inline_dir(struct file *file, 1429 struct dir_context *ctx, 1430 int *has_inline_data) 1431 { 1432 unsigned int offset, parent_ino; 1433 int i; 1434 struct ext4_dir_entry_2 *de; 1435 struct super_block *sb; 1436 struct inode *inode = file_inode(file); 1437 int ret, inline_size = 0; 1438 struct ext4_iloc iloc; 1439 void *dir_buf = NULL; 1440 int dotdot_offset, dotdot_size, extra_offset, extra_size; 1441 1442 ret = ext4_get_inode_loc(inode, &iloc); 1443 if (ret) 1444 return ret; 1445 1446 down_read(&EXT4_I(inode)->xattr_sem); 1447 if (!ext4_has_inline_data(inode)) { 1448 up_read(&EXT4_I(inode)->xattr_sem); 1449 *has_inline_data = 0; 1450 goto out; 1451 } 1452 1453 inline_size = ext4_get_inline_size(inode); 1454 dir_buf = kmalloc(inline_size, GFP_NOFS); 1455 if (!dir_buf) { 1456 ret = -ENOMEM; 1457 up_read(&EXT4_I(inode)->xattr_sem); 1458 goto out; 1459 } 1460 1461 ret = ext4_read_inline_data(inode, dir_buf, inline_size, &iloc); 1462 up_read(&EXT4_I(inode)->xattr_sem); 1463 if (ret < 0) 1464 goto out; 1465 1466 ret = 0; 1467 sb = inode->i_sb; 1468 parent_ino = le32_to_cpu(((struct ext4_dir_entry_2 *)dir_buf)->inode); 1469 offset = ctx->pos; 1470 1471 /* 1472 * dotdot_offset and dotdot_size is the real offset and 1473 * size for ".." and "." if the dir is block based while 1474 * the real size for them are only EXT4_INLINE_DOTDOT_SIZE. 1475 * So we will use extra_offset and extra_size to indicate them 1476 * during the inline dir iteration. 1477 */ 1478 dotdot_offset = EXT4_DIR_REC_LEN(1); 1479 dotdot_size = dotdot_offset + EXT4_DIR_REC_LEN(2); 1480 extra_offset = dotdot_size - EXT4_INLINE_DOTDOT_SIZE; 1481 extra_size = extra_offset + inline_size; 1482 1483 /* 1484 * If the version has changed since the last call to 1485 * readdir(2), then we might be pointing to an invalid 1486 * dirent right now. Scan from the start of the inline 1487 * dir to make sure. 1488 */ 1489 if (file->f_version != inode->i_version) { 1490 for (i = 0; i < extra_size && i < offset;) { 1491 /* 1492 * "." is with offset 0 and 1493 * ".." is dotdot_offset. 1494 */ 1495 if (!i) { 1496 i = dotdot_offset; 1497 continue; 1498 } else if (i == dotdot_offset) { 1499 i = dotdot_size; 1500 continue; 1501 } 1502 /* for other entry, the real offset in 1503 * the buf has to be tuned accordingly. 1504 */ 1505 de = (struct ext4_dir_entry_2 *) 1506 (dir_buf + i - extra_offset); 1507 /* It's too expensive to do a full 1508 * dirent test each time round this 1509 * loop, but we do have to test at 1510 * least that it is non-zero. A 1511 * failure will be detected in the 1512 * dirent test below. */ 1513 if (ext4_rec_len_from_disk(de->rec_len, extra_size) 1514 < EXT4_DIR_REC_LEN(1)) 1515 break; 1516 i += ext4_rec_len_from_disk(de->rec_len, 1517 extra_size); 1518 } 1519 offset = i; 1520 ctx->pos = offset; 1521 file->f_version = inode->i_version; 1522 } 1523 1524 while (ctx->pos < extra_size) { 1525 if (ctx->pos == 0) { 1526 if (!dir_emit(ctx, ".", 1, inode->i_ino, DT_DIR)) 1527 goto out; 1528 ctx->pos = dotdot_offset; 1529 continue; 1530 } 1531 1532 if (ctx->pos == dotdot_offset) { 1533 if (!dir_emit(ctx, "..", 2, parent_ino, DT_DIR)) 1534 goto out; 1535 ctx->pos = dotdot_size; 1536 continue; 1537 } 1538 1539 de = (struct ext4_dir_entry_2 *) 1540 (dir_buf + ctx->pos - extra_offset); 1541 if (ext4_check_dir_entry(inode, file, de, iloc.bh, dir_buf, 1542 extra_size, ctx->pos)) 1543 goto out; 1544 if (le32_to_cpu(de->inode)) { 1545 if (!dir_emit(ctx, de->name, de->name_len, 1546 le32_to_cpu(de->inode), 1547 get_dtype(sb, de->file_type))) 1548 goto out; 1549 } 1550 ctx->pos += ext4_rec_len_from_disk(de->rec_len, extra_size); 1551 } 1552 out: 1553 kfree(dir_buf); 1554 brelse(iloc.bh); 1555 return ret; 1556 } 1557 1558 struct buffer_head *ext4_get_first_inline_block(struct inode *inode, 1559 struct ext4_dir_entry_2 **parent_de, 1560 int *retval) 1561 { 1562 struct ext4_iloc iloc; 1563 1564 *retval = ext4_get_inode_loc(inode, &iloc); 1565 if (*retval) 1566 return NULL; 1567 1568 *parent_de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block; 1569 1570 return iloc.bh; 1571 } 1572 1573 /* 1574 * Try to create the inline data for the new dir. 1575 * If it succeeds, return 0, otherwise return the error. 1576 * In case of ENOSPC, the caller should create the normal disk layout dir. 1577 */ 1578 int ext4_try_create_inline_dir(handle_t *handle, struct inode *parent, 1579 struct inode *inode) 1580 { 1581 int ret, inline_size = EXT4_MIN_INLINE_DATA_SIZE; 1582 struct ext4_iloc iloc; 1583 struct ext4_dir_entry_2 *de; 1584 1585 ret = ext4_get_inode_loc(inode, &iloc); 1586 if (ret) 1587 return ret; 1588 1589 ret = ext4_prepare_inline_data(handle, inode, inline_size); 1590 if (ret) 1591 goto out; 1592 1593 /* 1594 * For inline dir, we only save the inode information for the ".." 1595 * and create a fake dentry to cover the left space. 1596 */ 1597 de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block; 1598 de->inode = cpu_to_le32(parent->i_ino); 1599 de = (struct ext4_dir_entry_2 *)((void *)de + EXT4_INLINE_DOTDOT_SIZE); 1600 de->inode = 0; 1601 de->rec_len = ext4_rec_len_to_disk( 1602 inline_size - EXT4_INLINE_DOTDOT_SIZE, 1603 inline_size); 1604 set_nlink(inode, 2); 1605 inode->i_size = EXT4_I(inode)->i_disksize = inline_size; 1606 out: 1607 brelse(iloc.bh); 1608 return ret; 1609 } 1610 1611 struct buffer_head *ext4_find_inline_entry(struct inode *dir, 1612 struct ext4_filename *fname, 1613 const struct qstr *d_name, 1614 struct ext4_dir_entry_2 **res_dir, 1615 int *has_inline_data) 1616 { 1617 int ret; 1618 struct ext4_iloc iloc; 1619 void *inline_start; 1620 int inline_size; 1621 1622 if (ext4_get_inode_loc(dir, &iloc)) 1623 return NULL; 1624 1625 down_read(&EXT4_I(dir)->xattr_sem); 1626 if (!ext4_has_inline_data(dir)) { 1627 *has_inline_data = 0; 1628 goto out; 1629 } 1630 1631 inline_start = (void *)ext4_raw_inode(&iloc)->i_block + 1632 EXT4_INLINE_DOTDOT_SIZE; 1633 inline_size = EXT4_MIN_INLINE_DATA_SIZE - EXT4_INLINE_DOTDOT_SIZE; 1634 ret = ext4_search_dir(iloc.bh, inline_start, inline_size, 1635 dir, fname, d_name, 0, res_dir); 1636 if (ret == 1) 1637 goto out_find; 1638 if (ret < 0) 1639 goto out; 1640 1641 if (ext4_get_inline_size(dir) == EXT4_MIN_INLINE_DATA_SIZE) 1642 goto out; 1643 1644 inline_start = ext4_get_inline_xattr_pos(dir, &iloc); 1645 inline_size = ext4_get_inline_size(dir) - EXT4_MIN_INLINE_DATA_SIZE; 1646 1647 ret = ext4_search_dir(iloc.bh, inline_start, inline_size, 1648 dir, fname, d_name, 0, res_dir); 1649 if (ret == 1) 1650 goto out_find; 1651 1652 out: 1653 brelse(iloc.bh); 1654 iloc.bh = NULL; 1655 out_find: 1656 up_read(&EXT4_I(dir)->xattr_sem); 1657 return iloc.bh; 1658 } 1659 1660 int ext4_delete_inline_entry(handle_t *handle, 1661 struct inode *dir, 1662 struct ext4_dir_entry_2 *de_del, 1663 struct buffer_head *bh, 1664 int *has_inline_data) 1665 { 1666 int err, inline_size; 1667 struct ext4_iloc iloc; 1668 void *inline_start; 1669 1670 err = ext4_get_inode_loc(dir, &iloc); 1671 if (err) 1672 return err; 1673 1674 down_write(&EXT4_I(dir)->xattr_sem); 1675 if (!ext4_has_inline_data(dir)) { 1676 *has_inline_data = 0; 1677 goto out; 1678 } 1679 1680 if ((void *)de_del - ((void *)ext4_raw_inode(&iloc)->i_block) < 1681 EXT4_MIN_INLINE_DATA_SIZE) { 1682 inline_start = (void *)ext4_raw_inode(&iloc)->i_block + 1683 EXT4_INLINE_DOTDOT_SIZE; 1684 inline_size = EXT4_MIN_INLINE_DATA_SIZE - 1685 EXT4_INLINE_DOTDOT_SIZE; 1686 } else { 1687 inline_start = ext4_get_inline_xattr_pos(dir, &iloc); 1688 inline_size = ext4_get_inline_size(dir) - 1689 EXT4_MIN_INLINE_DATA_SIZE; 1690 } 1691 1692 BUFFER_TRACE(bh, "get_write_access"); 1693 err = ext4_journal_get_write_access(handle, bh); 1694 if (err) 1695 goto out; 1696 1697 err = ext4_generic_delete_entry(handle, dir, de_del, bh, 1698 inline_start, inline_size, 0); 1699 if (err) 1700 goto out; 1701 1702 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata"); 1703 err = ext4_mark_inode_dirty(handle, dir); 1704 if (unlikely(err)) 1705 goto out; 1706 1707 ext4_show_inline_dir(dir, iloc.bh, inline_start, inline_size); 1708 out: 1709 up_write(&EXT4_I(dir)->xattr_sem); 1710 brelse(iloc.bh); 1711 if (err != -ENOENT) 1712 ext4_std_error(dir->i_sb, err); 1713 return err; 1714 } 1715 1716 /* 1717 * Get the inline dentry at offset. 1718 */ 1719 static inline struct ext4_dir_entry_2 * 1720 ext4_get_inline_entry(struct inode *inode, 1721 struct ext4_iloc *iloc, 1722 unsigned int offset, 1723 void **inline_start, 1724 int *inline_size) 1725 { 1726 void *inline_pos; 1727 1728 BUG_ON(offset > ext4_get_inline_size(inode)); 1729 1730 if (offset < EXT4_MIN_INLINE_DATA_SIZE) { 1731 inline_pos = (void *)ext4_raw_inode(iloc)->i_block; 1732 *inline_size = EXT4_MIN_INLINE_DATA_SIZE; 1733 } else { 1734 inline_pos = ext4_get_inline_xattr_pos(inode, iloc); 1735 offset -= EXT4_MIN_INLINE_DATA_SIZE; 1736 *inline_size = ext4_get_inline_size(inode) - 1737 EXT4_MIN_INLINE_DATA_SIZE; 1738 } 1739 1740 if (inline_start) 1741 *inline_start = inline_pos; 1742 return (struct ext4_dir_entry_2 *)(inline_pos + offset); 1743 } 1744 1745 int empty_inline_dir(struct inode *dir, int *has_inline_data) 1746 { 1747 int err, inline_size; 1748 struct ext4_iloc iloc; 1749 void *inline_pos; 1750 unsigned int offset; 1751 struct ext4_dir_entry_2 *de; 1752 int ret = 1; 1753 1754 err = ext4_get_inode_loc(dir, &iloc); 1755 if (err) { 1756 EXT4_ERROR_INODE(dir, "error %d getting inode %lu block", 1757 err, dir->i_ino); 1758 return 1; 1759 } 1760 1761 down_read(&EXT4_I(dir)->xattr_sem); 1762 if (!ext4_has_inline_data(dir)) { 1763 *has_inline_data = 0; 1764 goto out; 1765 } 1766 1767 de = (struct ext4_dir_entry_2 *)ext4_raw_inode(&iloc)->i_block; 1768 if (!le32_to_cpu(de->inode)) { 1769 ext4_warning(dir->i_sb, 1770 "bad inline directory (dir #%lu) - no `..'", 1771 dir->i_ino); 1772 ret = 1; 1773 goto out; 1774 } 1775 1776 offset = EXT4_INLINE_DOTDOT_SIZE; 1777 while (offset < dir->i_size) { 1778 de = ext4_get_inline_entry(dir, &iloc, offset, 1779 &inline_pos, &inline_size); 1780 if (ext4_check_dir_entry(dir, NULL, de, 1781 iloc.bh, inline_pos, 1782 inline_size, offset)) { 1783 ext4_warning(dir->i_sb, 1784 "bad inline directory (dir #%lu) - " 1785 "inode %u, rec_len %u, name_len %d" 1786 "inline size %d\n", 1787 dir->i_ino, le32_to_cpu(de->inode), 1788 le16_to_cpu(de->rec_len), de->name_len, 1789 inline_size); 1790 ret = 1; 1791 goto out; 1792 } 1793 if (le32_to_cpu(de->inode)) { 1794 ret = 0; 1795 goto out; 1796 } 1797 offset += ext4_rec_len_from_disk(de->rec_len, inline_size); 1798 } 1799 1800 out: 1801 up_read(&EXT4_I(dir)->xattr_sem); 1802 brelse(iloc.bh); 1803 return ret; 1804 } 1805 1806 int ext4_destroy_inline_data(handle_t *handle, struct inode *inode) 1807 { 1808 int ret; 1809 1810 down_write(&EXT4_I(inode)->xattr_sem); 1811 ret = ext4_destroy_inline_data_nolock(handle, inode); 1812 up_write(&EXT4_I(inode)->xattr_sem); 1813 1814 return ret; 1815 } 1816 1817 int ext4_inline_data_fiemap(struct inode *inode, 1818 struct fiemap_extent_info *fieinfo, 1819 int *has_inline, __u64 start, __u64 len) 1820 { 1821 __u64 physical = 0; 1822 __u64 inline_len; 1823 __u32 flags = FIEMAP_EXTENT_DATA_INLINE | FIEMAP_EXTENT_NOT_ALIGNED | 1824 FIEMAP_EXTENT_LAST; 1825 int error = 0; 1826 struct ext4_iloc iloc; 1827 1828 down_read(&EXT4_I(inode)->xattr_sem); 1829 if (!ext4_has_inline_data(inode)) { 1830 *has_inline = 0; 1831 goto out; 1832 } 1833 inline_len = min_t(size_t, ext4_get_inline_size(inode), 1834 i_size_read(inode)); 1835 if (start >= inline_len) 1836 goto out; 1837 if (start + len < inline_len) 1838 inline_len = start + len; 1839 inline_len -= start; 1840 1841 error = ext4_get_inode_loc(inode, &iloc); 1842 if (error) 1843 goto out; 1844 1845 physical = (__u64)iloc.bh->b_blocknr << inode->i_sb->s_blocksize_bits; 1846 physical += (char *)ext4_raw_inode(&iloc) - iloc.bh->b_data; 1847 physical += offsetof(struct ext4_inode, i_block); 1848 1849 if (physical) 1850 error = fiemap_fill_next_extent(fieinfo, start, physical, 1851 inline_len, flags); 1852 brelse(iloc.bh); 1853 out: 1854 up_read(&EXT4_I(inode)->xattr_sem); 1855 return (error < 0 ? error : 0); 1856 } 1857 1858 /* 1859 * Called during xattr set, and if we can sparse space 'needed', 1860 * just create the extent tree evict the data to the outer block. 1861 * 1862 * We use jbd2 instead of page cache to move data to the 1st block 1863 * so that the whole transaction can be committed as a whole and 1864 * the data isn't lost because of the delayed page cache write. 1865 */ 1866 int ext4_try_to_evict_inline_data(handle_t *handle, 1867 struct inode *inode, 1868 int needed) 1869 { 1870 int error; 1871 struct ext4_xattr_entry *entry; 1872 struct ext4_inode *raw_inode; 1873 struct ext4_iloc iloc; 1874 1875 error = ext4_get_inode_loc(inode, &iloc); 1876 if (error) 1877 return error; 1878 1879 raw_inode = ext4_raw_inode(&iloc); 1880 entry = (struct ext4_xattr_entry *)((void *)raw_inode + 1881 EXT4_I(inode)->i_inline_off); 1882 if (EXT4_XATTR_LEN(entry->e_name_len) + 1883 EXT4_XATTR_SIZE(le32_to_cpu(entry->e_value_size)) < needed) { 1884 error = -ENOSPC; 1885 goto out; 1886 } 1887 1888 error = ext4_convert_inline_data_nolock(handle, inode, &iloc); 1889 out: 1890 brelse(iloc.bh); 1891 return error; 1892 } 1893 1894 void ext4_inline_data_truncate(struct inode *inode, int *has_inline) 1895 { 1896 handle_t *handle; 1897 int inline_size, value_len, needed_blocks; 1898 size_t i_size; 1899 void *value = NULL; 1900 struct ext4_xattr_ibody_find is = { 1901 .s = { .not_found = -ENODATA, }, 1902 }; 1903 struct ext4_xattr_info i = { 1904 .name_index = EXT4_XATTR_INDEX_SYSTEM, 1905 .name = EXT4_XATTR_SYSTEM_DATA, 1906 }; 1907 1908 1909 needed_blocks = ext4_writepage_trans_blocks(inode); 1910 handle = ext4_journal_start(inode, EXT4_HT_INODE, needed_blocks); 1911 if (IS_ERR(handle)) 1912 return; 1913 1914 down_write(&EXT4_I(inode)->xattr_sem); 1915 if (!ext4_has_inline_data(inode)) { 1916 *has_inline = 0; 1917 ext4_journal_stop(handle); 1918 return; 1919 } 1920 1921 if (ext4_orphan_add(handle, inode)) 1922 goto out; 1923 1924 if (ext4_get_inode_loc(inode, &is.iloc)) 1925 goto out; 1926 1927 down_write(&EXT4_I(inode)->i_data_sem); 1928 i_size = inode->i_size; 1929 inline_size = ext4_get_inline_size(inode); 1930 EXT4_I(inode)->i_disksize = i_size; 1931 1932 if (i_size < inline_size) { 1933 /* Clear the content in the xattr space. */ 1934 if (inline_size > EXT4_MIN_INLINE_DATA_SIZE) { 1935 if (ext4_xattr_ibody_find(inode, &i, &is)) 1936 goto out_error; 1937 1938 BUG_ON(is.s.not_found); 1939 1940 value_len = le32_to_cpu(is.s.here->e_value_size); 1941 value = kmalloc(value_len, GFP_NOFS); 1942 if (!value) 1943 goto out_error; 1944 1945 if (ext4_xattr_ibody_get(inode, i.name_index, i.name, 1946 value, value_len)) 1947 goto out_error; 1948 1949 i.value = value; 1950 i.value_len = i_size > EXT4_MIN_INLINE_DATA_SIZE ? 1951 i_size - EXT4_MIN_INLINE_DATA_SIZE : 0; 1952 if (ext4_xattr_ibody_inline_set(handle, inode, &i, &is)) 1953 goto out_error; 1954 } 1955 1956 /* Clear the content within i_blocks. */ 1957 if (i_size < EXT4_MIN_INLINE_DATA_SIZE) { 1958 void *p = (void *) ext4_raw_inode(&is.iloc)->i_block; 1959 memset(p + i_size, 0, 1960 EXT4_MIN_INLINE_DATA_SIZE - i_size); 1961 } 1962 1963 EXT4_I(inode)->i_inline_size = i_size < 1964 EXT4_MIN_INLINE_DATA_SIZE ? 1965 EXT4_MIN_INLINE_DATA_SIZE : i_size; 1966 } 1967 1968 out_error: 1969 up_write(&EXT4_I(inode)->i_data_sem); 1970 out: 1971 brelse(is.iloc.bh); 1972 up_write(&EXT4_I(inode)->xattr_sem); 1973 kfree(value); 1974 if (inode->i_nlink) 1975 ext4_orphan_del(handle, inode); 1976 1977 inode->i_mtime = inode->i_ctime = ext4_current_time(inode); 1978 ext4_mark_inode_dirty(handle, inode); 1979 if (IS_SYNC(inode)) 1980 ext4_handle_sync(handle); 1981 1982 ext4_journal_stop(handle); 1983 return; 1984 } 1985 1986 int ext4_convert_inline_data(struct inode *inode) 1987 { 1988 int error, needed_blocks; 1989 handle_t *handle; 1990 struct ext4_iloc iloc; 1991 1992 if (!ext4_has_inline_data(inode)) { 1993 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA); 1994 return 0; 1995 } 1996 1997 needed_blocks = ext4_writepage_trans_blocks(inode); 1998 1999 iloc.bh = NULL; 2000 error = ext4_get_inode_loc(inode, &iloc); 2001 if (error) 2002 return error; 2003 2004 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks); 2005 if (IS_ERR(handle)) { 2006 error = PTR_ERR(handle); 2007 goto out_free; 2008 } 2009 2010 down_write(&EXT4_I(inode)->xattr_sem); 2011 if (!ext4_has_inline_data(inode)) { 2012 up_write(&EXT4_I(inode)->xattr_sem); 2013 goto out; 2014 } 2015 2016 error = ext4_convert_inline_data_nolock(handle, inode, &iloc); 2017 up_write(&EXT4_I(inode)->xattr_sem); 2018 out: 2019 ext4_journal_stop(handle); 2020 out_free: 2021 brelse(iloc.bh); 2022 return error; 2023 } 2024