1 // SPDX-License-Identifier: LGPL-2.1 2 /* 3 * Copyright (c) 2008,2009 NEC Software Tohoku, Ltd. 4 * Written by Takashi Sato <t-sato@yk.jp.nec.com> 5 * Akira Fujita <a-fujita@rs.jp.nec.com> 6 */ 7 8 #include <linux/fs.h> 9 #include <linux/quotaops.h> 10 #include <linux/slab.h> 11 #include <linux/sched/mm.h> 12 #include "ext4_jbd2.h" 13 #include "ext4.h" 14 #include "ext4_extents.h" 15 16 /** 17 * get_ext_path() - Find an extent path for designated logical block number. 18 * @inode: inode to be searched 19 * @lblock: logical block number to find an extent path 20 * @ppath: pointer to an extent path pointer (for output) 21 * 22 * ext4_find_extent wrapper. Return 0 on success, or a negative error value 23 * on failure. 24 */ 25 static inline int 26 get_ext_path(struct inode *inode, ext4_lblk_t lblock, 27 struct ext4_ext_path **ppath) 28 { 29 struct ext4_ext_path *path; 30 31 path = ext4_find_extent(inode, lblock, ppath, EXT4_EX_NOCACHE); 32 if (IS_ERR(path)) 33 return PTR_ERR(path); 34 if (path[ext_depth(inode)].p_ext == NULL) { 35 ext4_free_ext_path(path); 36 *ppath = NULL; 37 return -ENODATA; 38 } 39 *ppath = path; 40 return 0; 41 } 42 43 /** 44 * ext4_double_down_write_data_sem() - write lock two inodes's i_data_sem 45 * @first: inode to be locked 46 * @second: inode to be locked 47 * 48 * Acquire write lock of i_data_sem of the two inodes 49 */ 50 void 51 ext4_double_down_write_data_sem(struct inode *first, struct inode *second) 52 { 53 if (first < second) { 54 down_write(&EXT4_I(first)->i_data_sem); 55 down_write_nested(&EXT4_I(second)->i_data_sem, I_DATA_SEM_OTHER); 56 } else { 57 down_write(&EXT4_I(second)->i_data_sem); 58 down_write_nested(&EXT4_I(first)->i_data_sem, I_DATA_SEM_OTHER); 59 60 } 61 } 62 63 /** 64 * ext4_double_up_write_data_sem - Release two inodes' write lock of i_data_sem 65 * 66 * @orig_inode: original inode structure to be released its lock first 67 * @donor_inode: donor inode structure to be released its lock second 68 * Release write lock of i_data_sem of two inodes (orig and donor). 69 */ 70 void 71 ext4_double_up_write_data_sem(struct inode *orig_inode, 72 struct inode *donor_inode) 73 { 74 up_write(&EXT4_I(orig_inode)->i_data_sem); 75 up_write(&EXT4_I(donor_inode)->i_data_sem); 76 } 77 78 /** 79 * mext_check_coverage - Check that all extents in range has the same type 80 * 81 * @inode: inode in question 82 * @from: block offset of inode 83 * @count: block count to be checked 84 * @unwritten: extents expected to be unwritten 85 * @err: pointer to save error value 86 * 87 * Return 1 if all extents in range has expected type, and zero otherwise. 88 */ 89 static int 90 mext_check_coverage(struct inode *inode, ext4_lblk_t from, ext4_lblk_t count, 91 int unwritten, int *err) 92 { 93 struct ext4_ext_path *path = NULL; 94 struct ext4_extent *ext; 95 int ret = 0; 96 ext4_lblk_t last = from + count; 97 while (from < last) { 98 *err = get_ext_path(inode, from, &path); 99 if (*err) 100 goto out; 101 ext = path[ext_depth(inode)].p_ext; 102 if (unwritten != ext4_ext_is_unwritten(ext)) 103 goto out; 104 from += ext4_ext_get_actual_len(ext); 105 } 106 ret = 1; 107 out: 108 ext4_free_ext_path(path); 109 return ret; 110 } 111 112 /** 113 * mext_folio_double_lock - Grab and lock folio on both @inode1 and @inode2 114 * 115 * @inode1: the inode structure 116 * @inode2: the inode structure 117 * @index1: folio index 118 * @index2: folio index 119 * @folio: result folio vector 120 * 121 * Grab two locked folio for inode's by inode order 122 */ 123 static int 124 mext_folio_double_lock(struct inode *inode1, struct inode *inode2, 125 pgoff_t index1, pgoff_t index2, struct folio *folio[2]) 126 { 127 struct address_space *mapping[2]; 128 unsigned int flags; 129 130 BUG_ON(!inode1 || !inode2); 131 if (inode1 < inode2) { 132 mapping[0] = inode1->i_mapping; 133 mapping[1] = inode2->i_mapping; 134 } else { 135 swap(index1, index2); 136 mapping[0] = inode2->i_mapping; 137 mapping[1] = inode1->i_mapping; 138 } 139 140 flags = memalloc_nofs_save(); 141 folio[0] = __filemap_get_folio(mapping[0], index1, FGP_WRITEBEGIN, 142 mapping_gfp_mask(mapping[0])); 143 if (IS_ERR(folio[0])) { 144 memalloc_nofs_restore(flags); 145 return PTR_ERR(folio[0]); 146 } 147 148 folio[1] = __filemap_get_folio(mapping[1], index2, FGP_WRITEBEGIN, 149 mapping_gfp_mask(mapping[1])); 150 memalloc_nofs_restore(flags); 151 if (IS_ERR(folio[1])) { 152 folio_unlock(folio[0]); 153 folio_put(folio[0]); 154 return PTR_ERR(folio[1]); 155 } 156 /* 157 * __filemap_get_folio() may not wait on folio's writeback if 158 * BDI not demand that. But it is reasonable to be very conservative 159 * here and explicitly wait on folio's writeback 160 */ 161 folio_wait_writeback(folio[0]); 162 folio_wait_writeback(folio[1]); 163 if (inode1 > inode2) 164 swap(folio[0], folio[1]); 165 166 return 0; 167 } 168 169 /* Force folio buffers uptodate w/o dropping folio's lock */ 170 static int mext_page_mkuptodate(struct folio *folio, size_t from, size_t to) 171 { 172 struct inode *inode = folio->mapping->host; 173 sector_t block; 174 struct buffer_head *bh, *head; 175 unsigned int blocksize, block_start, block_end; 176 int nr = 0; 177 bool partial = false; 178 179 BUG_ON(!folio_test_locked(folio)); 180 BUG_ON(folio_test_writeback(folio)); 181 182 if (folio_test_uptodate(folio)) 183 return 0; 184 185 blocksize = i_blocksize(inode); 186 head = folio_buffers(folio); 187 if (!head) 188 head = create_empty_buffers(folio, blocksize, 0); 189 190 block = folio_pos(folio) >> inode->i_blkbits; 191 block_end = 0; 192 bh = head; 193 do { 194 block_start = block_end; 195 block_end = block_start + blocksize; 196 if (block_end <= from || block_start >= to) { 197 if (!buffer_uptodate(bh)) 198 partial = true; 199 continue; 200 } 201 if (buffer_uptodate(bh)) 202 continue; 203 if (!buffer_mapped(bh)) { 204 int err = ext4_get_block(inode, block, bh, 0); 205 if (err) 206 return err; 207 if (!buffer_mapped(bh)) { 208 folio_zero_range(folio, block_start, blocksize); 209 set_buffer_uptodate(bh); 210 continue; 211 } 212 } 213 lock_buffer(bh); 214 if (buffer_uptodate(bh)) { 215 unlock_buffer(bh); 216 continue; 217 } 218 ext4_read_bh_nowait(bh, 0, NULL); 219 nr++; 220 } while (block++, (bh = bh->b_this_page) != head); 221 222 /* No io required */ 223 if (!nr) 224 goto out; 225 226 bh = head; 227 do { 228 if (bh_offset(bh) + blocksize <= from) 229 continue; 230 if (bh_offset(bh) > to) 231 break; 232 wait_on_buffer(bh); 233 if (buffer_uptodate(bh)) 234 continue; 235 return -EIO; 236 } while ((bh = bh->b_this_page) != head); 237 out: 238 if (!partial) 239 folio_mark_uptodate(folio); 240 return 0; 241 } 242 243 /** 244 * move_extent_per_page - Move extent data per page 245 * 246 * @o_filp: file structure of original file 247 * @donor_inode: donor inode 248 * @orig_page_offset: page index on original file 249 * @donor_page_offset: page index on donor file 250 * @data_offset_in_page: block index where data swapping starts 251 * @block_len_in_page: the number of blocks to be swapped 252 * @unwritten: orig extent is unwritten or not 253 * @err: pointer to save return value 254 * 255 * Save the data in original inode blocks and replace original inode extents 256 * with donor inode extents by calling ext4_swap_extents(). 257 * Finally, write out the saved data in new original inode blocks. Return 258 * replaced block count. 259 */ 260 static int 261 move_extent_per_page(struct file *o_filp, struct inode *donor_inode, 262 pgoff_t orig_page_offset, pgoff_t donor_page_offset, 263 int data_offset_in_page, 264 int block_len_in_page, int unwritten, int *err) 265 { 266 struct inode *orig_inode = file_inode(o_filp); 267 struct folio *folio[2] = {NULL, NULL}; 268 handle_t *handle; 269 ext4_lblk_t orig_blk_offset, donor_blk_offset; 270 unsigned long blocksize = orig_inode->i_sb->s_blocksize; 271 unsigned int tmp_data_size, data_size, replaced_size; 272 int i, 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_SIZE >> orig_inode->i_blkbits; 276 struct super_block *sb = orig_inode->i_sb; 277 struct buffer_head *bh = NULL; 278 279 /* 280 * It needs twice the amount of ordinary journal buffers because 281 * inode and donor_inode may change each different metadata blocks. 282 */ 283 again: 284 *err = 0; 285 jblocks = ext4_writepage_trans_blocks(orig_inode) * 2; 286 handle = ext4_journal_start(orig_inode, EXT4_HT_MOVE_EXTENTS, jblocks); 287 if (IS_ERR(handle)) { 288 *err = PTR_ERR(handle); 289 return 0; 290 } 291 292 orig_blk_offset = orig_page_offset * blocks_per_page + 293 data_offset_in_page; 294 295 donor_blk_offset = donor_page_offset * blocks_per_page + 296 data_offset_in_page; 297 298 /* Calculate data_size */ 299 if ((orig_blk_offset + block_len_in_page - 1) == 300 ((orig_inode->i_size - 1) >> orig_inode->i_blkbits)) { 301 /* Replace the last block */ 302 tmp_data_size = orig_inode->i_size & (blocksize - 1); 303 /* 304 * If data_size equal zero, it shows data_size is multiples of 305 * blocksize. So we set appropriate value. 306 */ 307 if (tmp_data_size == 0) 308 tmp_data_size = blocksize; 309 310 data_size = tmp_data_size + 311 ((block_len_in_page - 1) << orig_inode->i_blkbits); 312 } else 313 data_size = block_len_in_page << orig_inode->i_blkbits; 314 315 replaced_size = data_size; 316 317 *err = mext_folio_double_lock(orig_inode, donor_inode, orig_page_offset, 318 donor_page_offset, folio); 319 if (unlikely(*err < 0)) 320 goto stop_journal; 321 /* 322 * If orig extent was unwritten it can become initialized 323 * at any time after i_data_sem was dropped, in order to 324 * serialize with delalloc we have recheck extent while we 325 * hold page's lock, if it is still the case data copy is not 326 * necessary, just swap data blocks between orig and donor. 327 */ 328 329 VM_BUG_ON_FOLIO(folio_test_large(folio[0]), folio[0]); 330 VM_BUG_ON_FOLIO(folio_test_large(folio[1]), folio[1]); 331 VM_BUG_ON_FOLIO(folio_nr_pages(folio[0]) != folio_nr_pages(folio[1]), folio[1]); 332 333 if (unwritten) { 334 ext4_double_down_write_data_sem(orig_inode, donor_inode); 335 /* If any of extents in range became initialized we have to 336 * fallback to data copying */ 337 unwritten = mext_check_coverage(orig_inode, orig_blk_offset, 338 block_len_in_page, 1, err); 339 if (*err) 340 goto drop_data_sem; 341 342 unwritten &= mext_check_coverage(donor_inode, donor_blk_offset, 343 block_len_in_page, 1, err); 344 if (*err) 345 goto drop_data_sem; 346 347 if (!unwritten) { 348 ext4_double_up_write_data_sem(orig_inode, donor_inode); 349 goto data_copy; 350 } 351 if (!filemap_release_folio(folio[0], 0) || 352 !filemap_release_folio(folio[1], 0)) { 353 *err = -EBUSY; 354 goto drop_data_sem; 355 } 356 replaced_count = ext4_swap_extents(handle, orig_inode, 357 donor_inode, orig_blk_offset, 358 donor_blk_offset, 359 block_len_in_page, 1, err); 360 drop_data_sem: 361 ext4_double_up_write_data_sem(orig_inode, donor_inode); 362 goto unlock_folios; 363 } 364 data_copy: 365 *err = mext_page_mkuptodate(folio[0], from, from + replaced_size); 366 if (*err) 367 goto unlock_folios; 368 369 /* At this point all buffers in range are uptodate, old mapping layout 370 * is no longer required, try to drop it now. */ 371 if (!filemap_release_folio(folio[0], 0) || 372 !filemap_release_folio(folio[1], 0)) { 373 *err = -EBUSY; 374 goto unlock_folios; 375 } 376 ext4_double_down_write_data_sem(orig_inode, donor_inode); 377 replaced_count = ext4_swap_extents(handle, orig_inode, donor_inode, 378 orig_blk_offset, donor_blk_offset, 379 block_len_in_page, 1, err); 380 ext4_double_up_write_data_sem(orig_inode, donor_inode); 381 if (*err) { 382 if (replaced_count) { 383 block_len_in_page = replaced_count; 384 replaced_size = 385 block_len_in_page << orig_inode->i_blkbits; 386 } else 387 goto unlock_folios; 388 } 389 /* Perform all necessary steps similar write_begin()/write_end() 390 * but keeping in mind that i_size will not change */ 391 bh = folio_buffers(folio[0]); 392 if (!bh) 393 bh = create_empty_buffers(folio[0], 394 1 << orig_inode->i_blkbits, 0); 395 for (i = 0; i < data_offset_in_page; i++) 396 bh = bh->b_this_page; 397 for (i = 0; i < block_len_in_page; i++) { 398 *err = ext4_get_block(orig_inode, orig_blk_offset + i, bh, 0); 399 if (*err < 0) 400 goto repair_branches; 401 bh = bh->b_this_page; 402 } 403 404 block_commit_write(&folio[0]->page, from, from + replaced_size); 405 406 /* Even in case of data=writeback it is reasonable to pin 407 * inode to transaction, to prevent unexpected data loss */ 408 *err = ext4_jbd2_inode_add_write(handle, orig_inode, 409 (loff_t)orig_page_offset << PAGE_SHIFT, replaced_size); 410 411 unlock_folios: 412 folio_unlock(folio[0]); 413 folio_put(folio[0]); 414 folio_unlock(folio[1]); 415 folio_put(folio[1]); 416 stop_journal: 417 ext4_journal_stop(handle); 418 if (*err == -ENOSPC && 419 ext4_should_retry_alloc(sb, &retries)) 420 goto again; 421 /* Buffer was busy because probably is pinned to journal transaction, 422 * force transaction commit may help to free it. */ 423 if (*err == -EBUSY && retries++ < 4 && EXT4_SB(sb)->s_journal && 424 jbd2_journal_force_commit_nested(EXT4_SB(sb)->s_journal)) 425 goto again; 426 return replaced_count; 427 428 repair_branches: 429 /* 430 * This should never ever happen! 431 * Extents are swapped already, but we are not able to copy data. 432 * Try to swap extents to it's original places 433 */ 434 ext4_double_down_write_data_sem(orig_inode, donor_inode); 435 replaced_count = ext4_swap_extents(handle, donor_inode, orig_inode, 436 orig_blk_offset, donor_blk_offset, 437 block_len_in_page, 0, &err2); 438 ext4_double_up_write_data_sem(orig_inode, donor_inode); 439 if (replaced_count != block_len_in_page) { 440 ext4_error_inode_block(orig_inode, (sector_t)(orig_blk_offset), 441 EIO, "Unable to copy data block," 442 " data will be lost."); 443 *err = -EIO; 444 } 445 replaced_count = 0; 446 goto unlock_folios; 447 } 448 449 /** 450 * mext_check_arguments - Check whether move extent can be done 451 * 452 * @orig_inode: original inode 453 * @donor_inode: donor inode 454 * @orig_start: logical start offset in block for orig 455 * @donor_start: logical start offset in block for donor 456 * @len: the number of blocks to be moved 457 * 458 * Check the arguments of ext4_move_extents() whether the files can be 459 * exchanged with each other. 460 * Return 0 on success, or a negative error value on failure. 461 */ 462 static int 463 mext_check_arguments(struct inode *orig_inode, 464 struct inode *donor_inode, __u64 orig_start, 465 __u64 donor_start, __u64 *len) 466 { 467 __u64 orig_eof, donor_eof; 468 unsigned int blkbits = orig_inode->i_blkbits; 469 unsigned int blocksize = 1 << blkbits; 470 471 orig_eof = (i_size_read(orig_inode) + blocksize - 1) >> blkbits; 472 donor_eof = (i_size_read(donor_inode) + blocksize - 1) >> blkbits; 473 474 475 if (donor_inode->i_mode & (S_ISUID|S_ISGID)) { 476 ext4_debug("ext4 move extent: suid or sgid is set" 477 " to donor file [ino:orig %lu, donor %lu]\n", 478 orig_inode->i_ino, donor_inode->i_ino); 479 return -EINVAL; 480 } 481 482 if (IS_IMMUTABLE(donor_inode) || IS_APPEND(donor_inode)) 483 return -EPERM; 484 485 /* Ext4 move extent does not support swap files */ 486 if (IS_SWAPFILE(orig_inode) || IS_SWAPFILE(donor_inode)) { 487 ext4_debug("ext4 move extent: The argument files should not be swap files [ino:orig %lu, donor %lu]\n", 488 orig_inode->i_ino, donor_inode->i_ino); 489 return -ETXTBSY; 490 } 491 492 if (ext4_is_quota_file(orig_inode) && ext4_is_quota_file(donor_inode)) { 493 ext4_debug("ext4 move extent: The argument files should not be quota files [ino:orig %lu, donor %lu]\n", 494 orig_inode->i_ino, donor_inode->i_ino); 495 return -EOPNOTSUPP; 496 } 497 498 /* Ext4 move extent supports only extent based file */ 499 if (!(ext4_test_inode_flag(orig_inode, EXT4_INODE_EXTENTS))) { 500 ext4_debug("ext4 move extent: orig file is not extents " 501 "based file [ino:orig %lu]\n", orig_inode->i_ino); 502 return -EOPNOTSUPP; 503 } else if (!(ext4_test_inode_flag(donor_inode, EXT4_INODE_EXTENTS))) { 504 ext4_debug("ext4 move extent: donor file is not extents " 505 "based file [ino:donor %lu]\n", donor_inode->i_ino); 506 return -EOPNOTSUPP; 507 } 508 509 if ((!orig_inode->i_size) || (!donor_inode->i_size)) { 510 ext4_debug("ext4 move extent: File size is 0 byte\n"); 511 return -EINVAL; 512 } 513 514 /* Start offset should be same */ 515 if ((orig_start & ~(PAGE_MASK >> orig_inode->i_blkbits)) != 516 (donor_start & ~(PAGE_MASK >> orig_inode->i_blkbits))) { 517 ext4_debug("ext4 move extent: orig and donor's start " 518 "offsets are not aligned [ino:orig %lu, donor %lu]\n", 519 orig_inode->i_ino, donor_inode->i_ino); 520 return -EINVAL; 521 } 522 523 if ((orig_start >= EXT_MAX_BLOCKS) || 524 (donor_start >= EXT_MAX_BLOCKS) || 525 (*len > EXT_MAX_BLOCKS) || 526 (donor_start + *len >= EXT_MAX_BLOCKS) || 527 (orig_start + *len >= EXT_MAX_BLOCKS)) { 528 ext4_debug("ext4 move extent: Can't handle over [%u] blocks " 529 "[ino:orig %lu, donor %lu]\n", EXT_MAX_BLOCKS, 530 orig_inode->i_ino, donor_inode->i_ino); 531 return -EINVAL; 532 } 533 if (orig_eof <= orig_start) 534 *len = 0; 535 else if (orig_eof < orig_start + *len - 1) 536 *len = orig_eof - orig_start; 537 if (donor_eof <= donor_start) 538 *len = 0; 539 else if (donor_eof < donor_start + *len - 1) 540 *len = donor_eof - donor_start; 541 if (!*len) { 542 ext4_debug("ext4 move extent: len should not be 0 " 543 "[ino:orig %lu, donor %lu]\n", orig_inode->i_ino, 544 donor_inode->i_ino); 545 return -EINVAL; 546 } 547 548 return 0; 549 } 550 551 /** 552 * ext4_move_extents - Exchange the specified range of a file 553 * 554 * @o_filp: file structure of the original file 555 * @d_filp: file structure of the donor file 556 * @orig_blk: start offset in block for orig 557 * @donor_blk: start offset in block for donor 558 * @len: the number of blocks to be moved 559 * @moved_len: moved block length 560 * 561 * This function returns 0 and moved block length is set in moved_len 562 * if succeed, otherwise returns error value. 563 * 564 */ 565 int 566 ext4_move_extents(struct file *o_filp, struct file *d_filp, __u64 orig_blk, 567 __u64 donor_blk, __u64 len, __u64 *moved_len) 568 { 569 struct inode *orig_inode = file_inode(o_filp); 570 struct inode *donor_inode = file_inode(d_filp); 571 struct ext4_ext_path *path = NULL; 572 int blocks_per_page = PAGE_SIZE >> orig_inode->i_blkbits; 573 ext4_lblk_t o_end, o_start = orig_blk; 574 ext4_lblk_t d_start = donor_blk; 575 int ret; 576 577 if (orig_inode->i_sb != donor_inode->i_sb) { 578 ext4_debug("ext4 move extent: The argument files " 579 "should be in same FS [ino:orig %lu, donor %lu]\n", 580 orig_inode->i_ino, donor_inode->i_ino); 581 return -EINVAL; 582 } 583 584 /* orig and donor should be different inodes */ 585 if (orig_inode == donor_inode) { 586 ext4_debug("ext4 move extent: The argument files should not " 587 "be same inode [ino:orig %lu, donor %lu]\n", 588 orig_inode->i_ino, donor_inode->i_ino); 589 return -EINVAL; 590 } 591 592 /* Regular file check */ 593 if (!S_ISREG(orig_inode->i_mode) || !S_ISREG(donor_inode->i_mode)) { 594 ext4_debug("ext4 move extent: The argument files should be " 595 "regular file [ino:orig %lu, donor %lu]\n", 596 orig_inode->i_ino, donor_inode->i_ino); 597 return -EINVAL; 598 } 599 600 /* TODO: it's not obvious how to swap blocks for inodes with full 601 journaling enabled */ 602 if (ext4_should_journal_data(orig_inode) || 603 ext4_should_journal_data(donor_inode)) { 604 ext4_msg(orig_inode->i_sb, KERN_ERR, 605 "Online defrag not supported with data journaling"); 606 return -EOPNOTSUPP; 607 } 608 609 if (IS_ENCRYPTED(orig_inode) || IS_ENCRYPTED(donor_inode)) { 610 ext4_msg(orig_inode->i_sb, KERN_ERR, 611 "Online defrag not supported for encrypted files"); 612 return -EOPNOTSUPP; 613 } 614 615 /* Protect orig and donor inodes against a truncate */ 616 lock_two_nondirectories(orig_inode, donor_inode); 617 618 /* Wait for all existing dio workers */ 619 inode_dio_wait(orig_inode); 620 inode_dio_wait(donor_inode); 621 622 /* Protect extent tree against block allocations via delalloc */ 623 ext4_double_down_write_data_sem(orig_inode, donor_inode); 624 /* Check the filesystem environment whether move_extent can be done */ 625 ret = mext_check_arguments(orig_inode, donor_inode, orig_blk, 626 donor_blk, &len); 627 if (ret) 628 goto out; 629 o_end = o_start + len; 630 631 *moved_len = 0; 632 while (o_start < o_end) { 633 struct ext4_extent *ex; 634 ext4_lblk_t cur_blk, next_blk; 635 pgoff_t orig_page_index, donor_page_index; 636 int offset_in_page; 637 int unwritten, cur_len; 638 639 ret = get_ext_path(orig_inode, o_start, &path); 640 if (ret) 641 goto out; 642 ex = path[path->p_depth].p_ext; 643 cur_blk = le32_to_cpu(ex->ee_block); 644 cur_len = ext4_ext_get_actual_len(ex); 645 /* Check hole before the start pos */ 646 if (cur_blk + cur_len - 1 < o_start) { 647 next_blk = ext4_ext_next_allocated_block(path); 648 if (next_blk == EXT_MAX_BLOCKS) { 649 ret = -ENODATA; 650 goto out; 651 } 652 d_start += next_blk - o_start; 653 o_start = next_blk; 654 continue; 655 /* Check hole after the start pos */ 656 } else if (cur_blk > o_start) { 657 /* Skip hole */ 658 d_start += cur_blk - o_start; 659 o_start = cur_blk; 660 /* Extent inside requested range ?*/ 661 if (cur_blk >= o_end) 662 goto out; 663 } else { /* in_range(o_start, o_blk, o_len) */ 664 cur_len += cur_blk - o_start; 665 } 666 unwritten = ext4_ext_is_unwritten(ex); 667 if (o_end - o_start < cur_len) 668 cur_len = o_end - o_start; 669 670 orig_page_index = o_start >> (PAGE_SHIFT - 671 orig_inode->i_blkbits); 672 donor_page_index = d_start >> (PAGE_SHIFT - 673 donor_inode->i_blkbits); 674 offset_in_page = o_start % blocks_per_page; 675 if (cur_len > blocks_per_page - offset_in_page) 676 cur_len = blocks_per_page - offset_in_page; 677 /* 678 * Up semaphore to avoid following problems: 679 * a. transaction deadlock among ext4_journal_start, 680 * ->write_begin via pagefault, and jbd2_journal_commit 681 * b. racing with ->read_folio, ->write_begin, and 682 * ext4_get_block in move_extent_per_page 683 */ 684 ext4_double_up_write_data_sem(orig_inode, donor_inode); 685 /* Swap original branches with new branches */ 686 *moved_len += move_extent_per_page(o_filp, donor_inode, 687 orig_page_index, donor_page_index, 688 offset_in_page, cur_len, 689 unwritten, &ret); 690 ext4_double_down_write_data_sem(orig_inode, donor_inode); 691 if (ret < 0) 692 break; 693 o_start += cur_len; 694 d_start += cur_len; 695 } 696 697 out: 698 if (*moved_len) { 699 ext4_discard_preallocations(orig_inode); 700 ext4_discard_preallocations(donor_inode); 701 } 702 703 ext4_free_ext_path(path); 704 ext4_double_up_write_data_sem(orig_inode, donor_inode); 705 unlock_two_nondirectories(orig_inode, donor_inode); 706 707 return ret; 708 } 709