1 /* 2 * Copyright (c) 2008,2009 NEC Software Tohoku, Ltd. 3 * Written by Takashi Sato <t-sato@yk.jp.nec.com> 4 * Akira Fujita <a-fujita@rs.jp.nec.com> 5 * 6 * This program is free software; you can redistribute it and/or modify it 7 * under the terms of version 2.1 of the GNU Lesser General Public License 8 * as published by the Free Software Foundation. 9 * 10 * This program is distributed in the hope that it will be useful, 11 * but WITHOUT ANY WARRANTY; without even the implied warranty of 12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the 13 * GNU General Public License for more details. 14 */ 15 16 #include <linux/fs.h> 17 #include <linux/quotaops.h> 18 #include <linux/slab.h> 19 #include "ext4_jbd2.h" 20 #include "ext4.h" 21 #include "ext4_extents.h" 22 23 /** 24 * get_ext_path - Find an extent path for designated logical block number. 25 * 26 * @inode: an inode which is searched 27 * @lblock: logical block number to find an extent path 28 * @path: pointer to an extent path pointer (for output) 29 * 30 * ext4_find_extent wrapper. Return 0 on success, or a negative error value 31 * on failure. 32 */ 33 static inline int 34 get_ext_path(struct inode *inode, ext4_lblk_t lblock, 35 struct ext4_ext_path **ppath) 36 { 37 struct ext4_ext_path *path; 38 39 path = ext4_find_extent(inode, lblock, ppath, EXT4_EX_NOCACHE); 40 if (IS_ERR(path)) 41 return PTR_ERR(path); 42 if (path[ext_depth(inode)].p_ext == NULL) { 43 ext4_ext_drop_refs(path); 44 kfree(path); 45 *ppath = NULL; 46 return -ENODATA; 47 } 48 *ppath = path; 49 return 0; 50 } 51 52 /** 53 * ext4_double_down_write_data_sem - Acquire two inodes' write lock 54 * of i_data_sem 55 * 56 * Acquire write lock of i_data_sem of the two inodes 57 */ 58 void 59 ext4_double_down_write_data_sem(struct inode *first, struct inode *second) 60 { 61 if (first < second) { 62 down_write(&EXT4_I(first)->i_data_sem); 63 down_write_nested(&EXT4_I(second)->i_data_sem, SINGLE_DEPTH_NESTING); 64 } else { 65 down_write(&EXT4_I(second)->i_data_sem); 66 down_write_nested(&EXT4_I(first)->i_data_sem, SINGLE_DEPTH_NESTING); 67 68 } 69 } 70 71 /** 72 * ext4_double_up_write_data_sem - Release two inodes' write lock of i_data_sem 73 * 74 * @orig_inode: original inode structure to be released its lock first 75 * @donor_inode: donor inode structure to be released its lock second 76 * Release write lock of i_data_sem of two inodes (orig and donor). 77 */ 78 void 79 ext4_double_up_write_data_sem(struct inode *orig_inode, 80 struct inode *donor_inode) 81 { 82 up_write(&EXT4_I(orig_inode)->i_data_sem); 83 up_write(&EXT4_I(donor_inode)->i_data_sem); 84 } 85 86 /** 87 * mext_check_coverage - Check that all extents in range has the same type 88 * 89 * @inode: inode in question 90 * @from: block offset of inode 91 * @count: block count to be checked 92 * @unwritten: extents expected to be unwritten 93 * @err: pointer to save error value 94 * 95 * Return 1 if all extents in range has expected type, and zero otherwise. 96 */ 97 static int 98 mext_check_coverage(struct inode *inode, ext4_lblk_t from, ext4_lblk_t count, 99 int unwritten, int *err) 100 { 101 struct ext4_ext_path *path = NULL; 102 struct ext4_extent *ext; 103 int ret = 0; 104 ext4_lblk_t last = from + count; 105 while (from < last) { 106 *err = get_ext_path(inode, from, &path); 107 if (*err) 108 goto out; 109 ext = path[ext_depth(inode)].p_ext; 110 if (unwritten != ext4_ext_is_unwritten(ext)) 111 goto out; 112 from += ext4_ext_get_actual_len(ext); 113 ext4_ext_drop_refs(path); 114 } 115 ret = 1; 116 out: 117 ext4_ext_drop_refs(path); 118 kfree(path); 119 return ret; 120 } 121 122 /** 123 * mext_page_double_lock - Grab and lock pages on both @inode1 and @inode2 124 * 125 * @inode1: the inode structure 126 * @inode2: the inode structure 127 * @index1: page index 128 * @index2: page index 129 * @page: result page vector 130 * 131 * Grab two locked pages for inode's by inode order 132 */ 133 static int 134 mext_page_double_lock(struct inode *inode1, struct inode *inode2, 135 pgoff_t index1, pgoff_t index2, struct page *page[2]) 136 { 137 struct address_space *mapping[2]; 138 unsigned fl = AOP_FLAG_NOFS; 139 140 BUG_ON(!inode1 || !inode2); 141 if (inode1 < inode2) { 142 mapping[0] = inode1->i_mapping; 143 mapping[1] = inode2->i_mapping; 144 } else { 145 pgoff_t tmp = index1; 146 index1 = index2; 147 index2 = tmp; 148 mapping[0] = inode2->i_mapping; 149 mapping[1] = inode1->i_mapping; 150 } 151 152 page[0] = grab_cache_page_write_begin(mapping[0], index1, fl); 153 if (!page[0]) 154 return -ENOMEM; 155 156 page[1] = grab_cache_page_write_begin(mapping[1], index2, fl); 157 if (!page[1]) { 158 unlock_page(page[0]); 159 page_cache_release(page[0]); 160 return -ENOMEM; 161 } 162 /* 163 * grab_cache_page_write_begin() may not wait on page's writeback if 164 * BDI not demand that. But it is reasonable to be very conservative 165 * here and explicitly wait on page's writeback 166 */ 167 wait_on_page_writeback(page[0]); 168 wait_on_page_writeback(page[1]); 169 if (inode1 > inode2) { 170 struct page *tmp; 171 tmp = page[0]; 172 page[0] = page[1]; 173 page[1] = tmp; 174 } 175 return 0; 176 } 177 178 /* Force page buffers uptodate w/o dropping page's lock */ 179 static int 180 mext_page_mkuptodate(struct page *page, unsigned from, unsigned to) 181 { 182 struct inode *inode = page->mapping->host; 183 sector_t block; 184 struct buffer_head *bh, *head, *arr[MAX_BUF_PER_PAGE]; 185 unsigned int blocksize, block_start, block_end; 186 int i, err, nr = 0, partial = 0; 187 BUG_ON(!PageLocked(page)); 188 BUG_ON(PageWriteback(page)); 189 190 if (PageUptodate(page)) 191 return 0; 192 193 blocksize = 1 << inode->i_blkbits; 194 if (!page_has_buffers(page)) 195 create_empty_buffers(page, blocksize, 0); 196 197 head = page_buffers(page); 198 block = (sector_t)page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits); 199 for (bh = head, block_start = 0; bh != head || !block_start; 200 block++, block_start = block_end, bh = bh->b_this_page) { 201 block_end = block_start + blocksize; 202 if (block_end <= from || block_start >= to) { 203 if (!buffer_uptodate(bh)) 204 partial = 1; 205 continue; 206 } 207 if (buffer_uptodate(bh)) 208 continue; 209 if (!buffer_mapped(bh)) { 210 err = ext4_get_block(inode, block, bh, 0); 211 if (err) { 212 SetPageError(page); 213 return err; 214 } 215 if (!buffer_mapped(bh)) { 216 zero_user(page, block_start, blocksize); 217 set_buffer_uptodate(bh); 218 continue; 219 } 220 } 221 BUG_ON(nr >= MAX_BUF_PER_PAGE); 222 arr[nr++] = bh; 223 } 224 /* No io required */ 225 if (!nr) 226 goto out; 227 228 for (i = 0; i < nr; i++) { 229 bh = arr[i]; 230 if (!bh_uptodate_or_lock(bh)) { 231 err = bh_submit_read(bh); 232 if (err) 233 return err; 234 } 235 } 236 out: 237 if (!partial) 238 SetPageUptodate(page); 239 return 0; 240 } 241 242 /** 243 * move_extent_per_page - Move extent data per page 244 * 245 * @o_filp: file structure of original file 246 * @donor_inode: donor inode 247 * @orig_page_offset: page index on original file 248 * @donor_page_offset: page index on donor file 249 * @data_offset_in_page: block index where data swapping starts 250 * @block_len_in_page: the number of blocks to be swapped 251 * @unwritten: orig extent is unwritten or not 252 * @err: pointer to save return value 253 * 254 * Save the data in original inode blocks and replace original inode extents 255 * with donor inode extents by calling ext4_swap_extents(). 256 * Finally, write out the saved data in new original inode blocks. Return 257 * replaced block count. 258 */ 259 static int 260 move_extent_per_page(struct file *o_filp, struct inode *donor_inode, 261 pgoff_t orig_page_offset, pgoff_t donor_page_offset, 262 int data_offset_in_page, 263 int block_len_in_page, int unwritten, int *err) 264 { 265 struct inode *orig_inode = file_inode(o_filp); 266 struct page *pagep[2] = {NULL, NULL}; 267 handle_t *handle; 268 ext4_lblk_t orig_blk_offset, donor_blk_offset; 269 unsigned long blocksize = orig_inode->i_sb->s_blocksize; 270 unsigned int w_flags = 0; 271 unsigned int tmp_data_size, data_size, replaced_size; 272 int err2, jblocks, retries = 0; 273 int replaced_count = 0; 274 int from = data_offset_in_page << orig_inode->i_blkbits; 275 int blocks_per_page = PAGE_CACHE_SIZE >> orig_inode->i_blkbits; 276 struct super_block *sb = orig_inode->i_sb; 277 278 /* 279 * It needs twice the amount of ordinary journal buffers because 280 * inode and donor_inode may change each different metadata blocks. 281 */ 282 again: 283 *err = 0; 284 jblocks = ext4_writepage_trans_blocks(orig_inode) * 2; 285 handle = ext4_journal_start(orig_inode, EXT4_HT_MOVE_EXTENTS, jblocks); 286 if (IS_ERR(handle)) { 287 *err = PTR_ERR(handle); 288 return 0; 289 } 290 291 if (segment_eq(get_fs(), KERNEL_DS)) 292 w_flags |= AOP_FLAG_UNINTERRUPTIBLE; 293 294 orig_blk_offset = orig_page_offset * blocks_per_page + 295 data_offset_in_page; 296 297 donor_blk_offset = donor_page_offset * blocks_per_page + 298 data_offset_in_page; 299 300 /* Calculate data_size */ 301 if ((orig_blk_offset + block_len_in_page - 1) == 302 ((orig_inode->i_size - 1) >> orig_inode->i_blkbits)) { 303 /* Replace the last block */ 304 tmp_data_size = orig_inode->i_size & (blocksize - 1); 305 /* 306 * If data_size equal zero, it shows data_size is multiples of 307 * blocksize. So we set appropriate value. 308 */ 309 if (tmp_data_size == 0) 310 tmp_data_size = blocksize; 311 312 data_size = tmp_data_size + 313 ((block_len_in_page - 1) << orig_inode->i_blkbits); 314 } else 315 data_size = block_len_in_page << orig_inode->i_blkbits; 316 317 replaced_size = data_size; 318 319 *err = mext_page_double_lock(orig_inode, donor_inode, orig_page_offset, 320 donor_page_offset, pagep); 321 if (unlikely(*err < 0)) 322 goto stop_journal; 323 /* 324 * If orig extent was unwritten it can become initialized 325 * at any time after i_data_sem was dropped, in order to 326 * serialize with delalloc we have recheck extent while we 327 * hold page's lock, if it is still the case data copy is not 328 * necessary, just swap data blocks between orig and donor. 329 */ 330 if (unwritten) { 331 ext4_double_down_write_data_sem(orig_inode, donor_inode); 332 /* If any of extents in range became initialized we have to 333 * fallback to data copying */ 334 unwritten = mext_check_coverage(orig_inode, orig_blk_offset, 335 block_len_in_page, 1, err); 336 if (*err) 337 goto drop_data_sem; 338 339 unwritten &= mext_check_coverage(donor_inode, donor_blk_offset, 340 block_len_in_page, 1, err); 341 if (*err) 342 goto drop_data_sem; 343 344 if (!unwritten) { 345 ext4_double_up_write_data_sem(orig_inode, donor_inode); 346 goto data_copy; 347 } 348 if ((page_has_private(pagep[0]) && 349 !try_to_release_page(pagep[0], 0)) || 350 (page_has_private(pagep[1]) && 351 !try_to_release_page(pagep[1], 0))) { 352 *err = -EBUSY; 353 goto drop_data_sem; 354 } 355 replaced_count = ext4_swap_extents(handle, orig_inode, 356 donor_inode, orig_blk_offset, 357 donor_blk_offset, 358 block_len_in_page, 1, err); 359 drop_data_sem: 360 ext4_double_up_write_data_sem(orig_inode, donor_inode); 361 goto unlock_pages; 362 } 363 data_copy: 364 *err = mext_page_mkuptodate(pagep[0], from, from + replaced_size); 365 if (*err) 366 goto unlock_pages; 367 368 /* At this point all buffers in range are uptodate, old mapping layout 369 * is no longer required, try to drop it now. */ 370 if ((page_has_private(pagep[0]) && !try_to_release_page(pagep[0], 0)) || 371 (page_has_private(pagep[1]) && !try_to_release_page(pagep[1], 0))) { 372 *err = -EBUSY; 373 goto unlock_pages; 374 } 375 ext4_double_down_write_data_sem(orig_inode, donor_inode); 376 replaced_count = ext4_swap_extents(handle, orig_inode, donor_inode, 377 orig_blk_offset, donor_blk_offset, 378 block_len_in_page, 1, err); 379 ext4_double_up_write_data_sem(orig_inode, donor_inode); 380 if (*err) { 381 if (replaced_count) { 382 block_len_in_page = replaced_count; 383 replaced_size = 384 block_len_in_page << orig_inode->i_blkbits; 385 } else 386 goto unlock_pages; 387 } 388 /* Perform all necessary steps similar write_begin()/write_end() 389 * but keeping in mind that i_size will not change */ 390 *err = __block_write_begin(pagep[0], from, replaced_size, 391 ext4_get_block); 392 if (!*err) 393 *err = block_commit_write(pagep[0], from, from + replaced_size); 394 395 if (unlikely(*err < 0)) 396 goto repair_branches; 397 398 /* Even in case of data=writeback it is reasonable to pin 399 * inode to transaction, to prevent unexpected data loss */ 400 *err = ext4_jbd2_file_inode(handle, orig_inode); 401 402 unlock_pages: 403 unlock_page(pagep[0]); 404 page_cache_release(pagep[0]); 405 unlock_page(pagep[1]); 406 page_cache_release(pagep[1]); 407 stop_journal: 408 ext4_journal_stop(handle); 409 if (*err == -ENOSPC && 410 ext4_should_retry_alloc(sb, &retries)) 411 goto again; 412 /* Buffer was busy because probably is pinned to journal transaction, 413 * force transaction commit may help to free it. */ 414 if (*err == -EBUSY && retries++ < 4 && EXT4_SB(sb)->s_journal && 415 jbd2_journal_force_commit_nested(EXT4_SB(sb)->s_journal)) 416 goto again; 417 return replaced_count; 418 419 repair_branches: 420 /* 421 * This should never ever happen! 422 * Extents are swapped already, but we are not able to copy data. 423 * Try to swap extents to it's original places 424 */ 425 ext4_double_down_write_data_sem(orig_inode, donor_inode); 426 replaced_count = ext4_swap_extents(handle, donor_inode, orig_inode, 427 orig_blk_offset, donor_blk_offset, 428 block_len_in_page, 0, &err2); 429 ext4_double_up_write_data_sem(orig_inode, donor_inode); 430 if (replaced_count != block_len_in_page) { 431 EXT4_ERROR_INODE_BLOCK(orig_inode, (sector_t)(orig_blk_offset), 432 "Unable to copy data block," 433 " data will be lost."); 434 *err = -EIO; 435 } 436 replaced_count = 0; 437 goto unlock_pages; 438 } 439 440 /** 441 * mext_check_arguments - Check whether move extent can be done 442 * 443 * @orig_inode: original inode 444 * @donor_inode: donor inode 445 * @orig_start: logical start offset in block for orig 446 * @donor_start: logical start offset in block for donor 447 * @len: the number of blocks to be moved 448 * 449 * Check the arguments of ext4_move_extents() whether the files can be 450 * exchanged with each other. 451 * Return 0 on success, or a negative error value on failure. 452 */ 453 static int 454 mext_check_arguments(struct inode *orig_inode, 455 struct inode *donor_inode, __u64 orig_start, 456 __u64 donor_start, __u64 *len) 457 { 458 __u64 orig_eof, donor_eof; 459 unsigned int blkbits = orig_inode->i_blkbits; 460 unsigned int blocksize = 1 << blkbits; 461 462 orig_eof = (i_size_read(orig_inode) + blocksize - 1) >> blkbits; 463 donor_eof = (i_size_read(donor_inode) + blocksize - 1) >> blkbits; 464 465 466 if (donor_inode->i_mode & (S_ISUID|S_ISGID)) { 467 ext4_debug("ext4 move extent: suid or sgid is set" 468 " to donor file [ino:orig %lu, donor %lu]\n", 469 orig_inode->i_ino, donor_inode->i_ino); 470 return -EINVAL; 471 } 472 473 if (IS_IMMUTABLE(donor_inode) || IS_APPEND(donor_inode)) 474 return -EPERM; 475 476 /* Ext4 move extent does not support swapfile */ 477 if (IS_SWAPFILE(orig_inode) || IS_SWAPFILE(donor_inode)) { 478 ext4_debug("ext4 move extent: The argument files should " 479 "not be swapfile [ino:orig %lu, donor %lu]\n", 480 orig_inode->i_ino, donor_inode->i_ino); 481 return -EBUSY; 482 } 483 484 /* Ext4 move extent supports only extent based file */ 485 if (!(ext4_test_inode_flag(orig_inode, EXT4_INODE_EXTENTS))) { 486 ext4_debug("ext4 move extent: orig file is not extents " 487 "based file [ino:orig %lu]\n", orig_inode->i_ino); 488 return -EOPNOTSUPP; 489 } else if (!(ext4_test_inode_flag(donor_inode, EXT4_INODE_EXTENTS))) { 490 ext4_debug("ext4 move extent: donor file is not extents " 491 "based file [ino:donor %lu]\n", donor_inode->i_ino); 492 return -EOPNOTSUPP; 493 } 494 495 if ((!orig_inode->i_size) || (!donor_inode->i_size)) { 496 ext4_debug("ext4 move extent: File size is 0 byte\n"); 497 return -EINVAL; 498 } 499 500 /* Start offset should be same */ 501 if ((orig_start & ~(PAGE_MASK >> orig_inode->i_blkbits)) != 502 (donor_start & ~(PAGE_MASK >> orig_inode->i_blkbits))) { 503 ext4_debug("ext4 move extent: orig and donor's start " 504 "offset are not alligned [ino:orig %lu, donor %lu]\n", 505 orig_inode->i_ino, donor_inode->i_ino); 506 return -EINVAL; 507 } 508 509 if ((orig_start >= EXT_MAX_BLOCKS) || 510 (donor_start >= EXT_MAX_BLOCKS) || 511 (*len > EXT_MAX_BLOCKS) || 512 (donor_start + *len >= EXT_MAX_BLOCKS) || 513 (orig_start + *len >= EXT_MAX_BLOCKS)) { 514 ext4_debug("ext4 move extent: Can't handle over [%u] blocks " 515 "[ino:orig %lu, donor %lu]\n", EXT_MAX_BLOCKS, 516 orig_inode->i_ino, donor_inode->i_ino); 517 return -EINVAL; 518 } 519 if (orig_eof < orig_start + *len - 1) 520 *len = orig_eof - orig_start; 521 if (donor_eof < donor_start + *len - 1) 522 *len = donor_eof - donor_start; 523 if (!*len) { 524 ext4_debug("ext4 move extent: len should not be 0 " 525 "[ino:orig %lu, donor %lu]\n", orig_inode->i_ino, 526 donor_inode->i_ino); 527 return -EINVAL; 528 } 529 530 return 0; 531 } 532 533 /** 534 * ext4_move_extents - Exchange the specified range of a file 535 * 536 * @o_filp: file structure of the original file 537 * @d_filp: file structure of the donor file 538 * @orig_blk: start offset in block for orig 539 * @donor_blk: start offset in block for donor 540 * @len: the number of blocks to be moved 541 * @moved_len: moved block length 542 * 543 * This function returns 0 and moved block length is set in moved_len 544 * if succeed, otherwise returns error value. 545 * 546 */ 547 int 548 ext4_move_extents(struct file *o_filp, struct file *d_filp, __u64 orig_blk, 549 __u64 donor_blk, __u64 len, __u64 *moved_len) 550 { 551 struct inode *orig_inode = file_inode(o_filp); 552 struct inode *donor_inode = file_inode(d_filp); 553 struct ext4_ext_path *path = NULL; 554 int blocks_per_page = PAGE_CACHE_SIZE >> orig_inode->i_blkbits; 555 ext4_lblk_t o_end, o_start = orig_blk; 556 ext4_lblk_t d_start = donor_blk; 557 int ret; 558 559 if (orig_inode->i_sb != donor_inode->i_sb) { 560 ext4_debug("ext4 move extent: The argument files " 561 "should be in same FS [ino:orig %lu, donor %lu]\n", 562 orig_inode->i_ino, donor_inode->i_ino); 563 return -EINVAL; 564 } 565 566 /* orig and donor should be different inodes */ 567 if (orig_inode == donor_inode) { 568 ext4_debug("ext4 move extent: The argument files should not " 569 "be same inode [ino:orig %lu, donor %lu]\n", 570 orig_inode->i_ino, donor_inode->i_ino); 571 return -EINVAL; 572 } 573 574 /* Regular file check */ 575 if (!S_ISREG(orig_inode->i_mode) || !S_ISREG(donor_inode->i_mode)) { 576 ext4_debug("ext4 move extent: The argument files should be " 577 "regular file [ino:orig %lu, donor %lu]\n", 578 orig_inode->i_ino, donor_inode->i_ino); 579 return -EINVAL; 580 } 581 /* TODO: This is non obvious task to swap blocks for inodes with full 582 jornaling enabled */ 583 if (ext4_should_journal_data(orig_inode) || 584 ext4_should_journal_data(donor_inode)) { 585 return -EINVAL; 586 } 587 /* Protect orig and donor inodes against a truncate */ 588 lock_two_nondirectories(orig_inode, donor_inode); 589 590 /* Wait for all existing dio workers */ 591 ext4_inode_block_unlocked_dio(orig_inode); 592 ext4_inode_block_unlocked_dio(donor_inode); 593 inode_dio_wait(orig_inode); 594 inode_dio_wait(donor_inode); 595 596 /* Protect extent tree against block allocations via delalloc */ 597 ext4_double_down_write_data_sem(orig_inode, donor_inode); 598 /* Check the filesystem environment whether move_extent can be done */ 599 ret = mext_check_arguments(orig_inode, donor_inode, orig_blk, 600 donor_blk, &len); 601 if (ret) 602 goto out; 603 o_end = o_start + len; 604 605 while (o_start < o_end) { 606 struct ext4_extent *ex; 607 ext4_lblk_t cur_blk, next_blk; 608 pgoff_t orig_page_index, donor_page_index; 609 int offset_in_page; 610 int unwritten, cur_len; 611 612 ret = get_ext_path(orig_inode, o_start, &path); 613 if (ret) 614 goto out; 615 ex = path[path->p_depth].p_ext; 616 next_blk = ext4_ext_next_allocated_block(path); 617 cur_blk = le32_to_cpu(ex->ee_block); 618 cur_len = ext4_ext_get_actual_len(ex); 619 /* Check hole before the start pos */ 620 if (cur_blk + cur_len - 1 < o_start) { 621 if (next_blk == EXT_MAX_BLOCKS) { 622 o_start = o_end; 623 ret = -ENODATA; 624 goto out; 625 } 626 d_start += next_blk - o_start; 627 o_start = next_blk; 628 continue; 629 /* Check hole after the start pos */ 630 } else if (cur_blk > o_start) { 631 /* Skip hole */ 632 d_start += cur_blk - o_start; 633 o_start = cur_blk; 634 /* Extent inside requested range ?*/ 635 if (cur_blk >= o_end) 636 goto out; 637 } else { /* in_range(o_start, o_blk, o_len) */ 638 cur_len += cur_blk - o_start; 639 } 640 unwritten = ext4_ext_is_unwritten(ex); 641 if (o_end - o_start < cur_len) 642 cur_len = o_end - o_start; 643 644 orig_page_index = o_start >> (PAGE_CACHE_SHIFT - 645 orig_inode->i_blkbits); 646 donor_page_index = d_start >> (PAGE_CACHE_SHIFT - 647 donor_inode->i_blkbits); 648 offset_in_page = o_start % blocks_per_page; 649 if (cur_len > blocks_per_page- offset_in_page) 650 cur_len = blocks_per_page - offset_in_page; 651 /* 652 * Up semaphore to avoid following problems: 653 * a. transaction deadlock among ext4_journal_start, 654 * ->write_begin via pagefault, and jbd2_journal_commit 655 * b. racing with ->readpage, ->write_begin, and ext4_get_block 656 * in move_extent_per_page 657 */ 658 ext4_double_up_write_data_sem(orig_inode, donor_inode); 659 /* Swap original branches with new branches */ 660 move_extent_per_page(o_filp, donor_inode, 661 orig_page_index, donor_page_index, 662 offset_in_page, cur_len, 663 unwritten, &ret); 664 ext4_double_down_write_data_sem(orig_inode, donor_inode); 665 if (ret < 0) 666 break; 667 o_start += cur_len; 668 d_start += cur_len; 669 } 670 *moved_len = o_start - orig_blk; 671 if (*moved_len > len) 672 *moved_len = len; 673 674 out: 675 if (*moved_len) { 676 ext4_discard_preallocations(orig_inode); 677 ext4_discard_preallocations(donor_inode); 678 } 679 680 ext4_ext_drop_refs(path); 681 kfree(path); 682 ext4_double_up_write_data_sem(orig_inode, donor_inode); 683 ext4_inode_resume_unlocked_dio(orig_inode); 684 ext4_inode_resume_unlocked_dio(donor_inode); 685 unlock_two_nondirectories(orig_inode, donor_inode); 686 687 return ret; 688 } 689