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 page buffers uptodate w/o dropping page's lock */ 170 static int 171 mext_page_mkuptodate(struct folio *folio, unsigned from, unsigned to) 172 { 173 struct inode *inode = folio->mapping->host; 174 sector_t block; 175 struct buffer_head *bh, *head, *arr[MAX_BUF_PER_PAGE]; 176 unsigned int blocksize, block_start, block_end; 177 int i, err, nr = 0, partial = 0; 178 BUG_ON(!folio_test_locked(folio)); 179 BUG_ON(folio_test_writeback(folio)); 180 181 if (folio_test_uptodate(folio)) 182 return 0; 183 184 blocksize = i_blocksize(inode); 185 head = folio_buffers(folio); 186 if (!head) { 187 create_empty_buffers(&folio->page, blocksize, 0); 188 head = folio_buffers(folio); 189 } 190 191 block = (sector_t)folio->index << (PAGE_SHIFT - inode->i_blkbits); 192 for (bh = head, block_start = 0; bh != head || !block_start; 193 block++, block_start = block_end, bh = bh->b_this_page) { 194 block_end = block_start + blocksize; 195 if (block_end <= from || block_start >= to) { 196 if (!buffer_uptodate(bh)) 197 partial = 1; 198 continue; 199 } 200 if (buffer_uptodate(bh)) 201 continue; 202 if (!buffer_mapped(bh)) { 203 err = ext4_get_block(inode, block, bh, 0); 204 if (err) { 205 folio_set_error(folio); 206 return err; 207 } 208 if (!buffer_mapped(bh)) { 209 folio_zero_range(folio, block_start, blocksize); 210 set_buffer_uptodate(bh); 211 continue; 212 } 213 } 214 BUG_ON(nr >= MAX_BUF_PER_PAGE); 215 arr[nr++] = bh; 216 } 217 /* No io required */ 218 if (!nr) 219 goto out; 220 221 for (i = 0; i < nr; i++) { 222 bh = arr[i]; 223 if (!bh_uptodate_or_lock(bh)) { 224 err = ext4_read_bh(bh, 0, NULL); 225 if (err) 226 return err; 227 } 228 } 229 out: 230 if (!partial) 231 folio_mark_uptodate(folio); 232 return 0; 233 } 234 235 /** 236 * move_extent_per_page - Move extent data per page 237 * 238 * @o_filp: file structure of original file 239 * @donor_inode: donor inode 240 * @orig_page_offset: page index on original file 241 * @donor_page_offset: page index on donor file 242 * @data_offset_in_page: block index where data swapping starts 243 * @block_len_in_page: the number of blocks to be swapped 244 * @unwritten: orig extent is unwritten or not 245 * @err: pointer to save return value 246 * 247 * Save the data in original inode blocks and replace original inode extents 248 * with donor inode extents by calling ext4_swap_extents(). 249 * Finally, write out the saved data in new original inode blocks. Return 250 * replaced block count. 251 */ 252 static int 253 move_extent_per_page(struct file *o_filp, struct inode *donor_inode, 254 pgoff_t orig_page_offset, pgoff_t donor_page_offset, 255 int data_offset_in_page, 256 int block_len_in_page, int unwritten, int *err) 257 { 258 struct inode *orig_inode = file_inode(o_filp); 259 struct folio *folio[2] = {NULL, NULL}; 260 handle_t *handle; 261 ext4_lblk_t orig_blk_offset, donor_blk_offset; 262 unsigned long blocksize = orig_inode->i_sb->s_blocksize; 263 unsigned int tmp_data_size, data_size, replaced_size; 264 int i, err2, jblocks, retries = 0; 265 int replaced_count = 0; 266 int from = data_offset_in_page << orig_inode->i_blkbits; 267 int blocks_per_page = PAGE_SIZE >> orig_inode->i_blkbits; 268 struct super_block *sb = orig_inode->i_sb; 269 struct buffer_head *bh = NULL; 270 271 /* 272 * It needs twice the amount of ordinary journal buffers because 273 * inode and donor_inode may change each different metadata blocks. 274 */ 275 again: 276 *err = 0; 277 jblocks = ext4_writepage_trans_blocks(orig_inode) * 2; 278 handle = ext4_journal_start(orig_inode, EXT4_HT_MOVE_EXTENTS, jblocks); 279 if (IS_ERR(handle)) { 280 *err = PTR_ERR(handle); 281 return 0; 282 } 283 284 orig_blk_offset = orig_page_offset * blocks_per_page + 285 data_offset_in_page; 286 287 donor_blk_offset = donor_page_offset * blocks_per_page + 288 data_offset_in_page; 289 290 /* Calculate data_size */ 291 if ((orig_blk_offset + block_len_in_page - 1) == 292 ((orig_inode->i_size - 1) >> orig_inode->i_blkbits)) { 293 /* Replace the last block */ 294 tmp_data_size = orig_inode->i_size & (blocksize - 1); 295 /* 296 * If data_size equal zero, it shows data_size is multiples of 297 * blocksize. So we set appropriate value. 298 */ 299 if (tmp_data_size == 0) 300 tmp_data_size = blocksize; 301 302 data_size = tmp_data_size + 303 ((block_len_in_page - 1) << orig_inode->i_blkbits); 304 } else 305 data_size = block_len_in_page << orig_inode->i_blkbits; 306 307 replaced_size = data_size; 308 309 *err = mext_folio_double_lock(orig_inode, donor_inode, orig_page_offset, 310 donor_page_offset, folio); 311 if (unlikely(*err < 0)) 312 goto stop_journal; 313 /* 314 * If orig extent was unwritten it can become initialized 315 * at any time after i_data_sem was dropped, in order to 316 * serialize with delalloc we have recheck extent while we 317 * hold page's lock, if it is still the case data copy is not 318 * necessary, just swap data blocks between orig and donor. 319 */ 320 321 VM_BUG_ON_FOLIO(folio_test_large(folio[0]), folio[0]); 322 VM_BUG_ON_FOLIO(folio_test_large(folio[1]), folio[1]); 323 VM_BUG_ON_FOLIO(folio_nr_pages(folio[0]) != folio_nr_pages(folio[1]), folio[1]); 324 325 if (unwritten) { 326 ext4_double_down_write_data_sem(orig_inode, donor_inode); 327 /* If any of extents in range became initialized we have to 328 * fallback to data copying */ 329 unwritten = mext_check_coverage(orig_inode, orig_blk_offset, 330 block_len_in_page, 1, err); 331 if (*err) 332 goto drop_data_sem; 333 334 unwritten &= mext_check_coverage(donor_inode, donor_blk_offset, 335 block_len_in_page, 1, err); 336 if (*err) 337 goto drop_data_sem; 338 339 if (!unwritten) { 340 ext4_double_up_write_data_sem(orig_inode, donor_inode); 341 goto data_copy; 342 } 343 if ((folio_has_private(folio[0]) && 344 !filemap_release_folio(folio[0], 0)) || 345 (folio_has_private(folio[1]) && 346 !filemap_release_folio(folio[1], 0))) { 347 *err = -EBUSY; 348 goto drop_data_sem; 349 } 350 replaced_count = ext4_swap_extents(handle, orig_inode, 351 donor_inode, orig_blk_offset, 352 donor_blk_offset, 353 block_len_in_page, 1, err); 354 drop_data_sem: 355 ext4_double_up_write_data_sem(orig_inode, donor_inode); 356 goto unlock_folios; 357 } 358 data_copy: 359 *err = mext_page_mkuptodate(folio[0], from, from + replaced_size); 360 if (*err) 361 goto unlock_folios; 362 363 /* At this point all buffers in range are uptodate, old mapping layout 364 * is no longer required, try to drop it now. */ 365 if ((folio_has_private(folio[0]) && 366 !filemap_release_folio(folio[0], 0)) || 367 (folio_has_private(folio[1]) && 368 !filemap_release_folio(folio[1], 0))) { 369 *err = -EBUSY; 370 goto unlock_folios; 371 } 372 ext4_double_down_write_data_sem(orig_inode, donor_inode); 373 replaced_count = ext4_swap_extents(handle, orig_inode, donor_inode, 374 orig_blk_offset, donor_blk_offset, 375 block_len_in_page, 1, err); 376 ext4_double_up_write_data_sem(orig_inode, donor_inode); 377 if (*err) { 378 if (replaced_count) { 379 block_len_in_page = replaced_count; 380 replaced_size = 381 block_len_in_page << orig_inode->i_blkbits; 382 } else 383 goto unlock_folios; 384 } 385 /* Perform all necessary steps similar write_begin()/write_end() 386 * but keeping in mind that i_size will not change */ 387 if (!folio_buffers(folio[0])) 388 create_empty_buffers(&folio[0]->page, 1 << orig_inode->i_blkbits, 0); 389 bh = folio_buffers(folio[0]); 390 for (i = 0; i < data_offset_in_page; i++) 391 bh = bh->b_this_page; 392 for (i = 0; i < block_len_in_page; i++) { 393 *err = ext4_get_block(orig_inode, orig_blk_offset + i, bh, 0); 394 if (*err < 0) 395 break; 396 bh = bh->b_this_page; 397 } 398 if (!*err) 399 *err = block_commit_write(&folio[0]->page, from, from + replaced_size); 400 401 if (unlikely(*err < 0)) 402 goto repair_branches; 403 404 /* Even in case of data=writeback it is reasonable to pin 405 * inode to transaction, to prevent unexpected data loss */ 406 *err = ext4_jbd2_inode_add_write(handle, orig_inode, 407 (loff_t)orig_page_offset << PAGE_SHIFT, replaced_size); 408 409 unlock_folios: 410 folio_unlock(folio[0]); 411 folio_put(folio[0]); 412 folio_unlock(folio[1]); 413 folio_put(folio[1]); 414 stop_journal: 415 ext4_journal_stop(handle); 416 if (*err == -ENOSPC && 417 ext4_should_retry_alloc(sb, &retries)) 418 goto again; 419 /* Buffer was busy because probably is pinned to journal transaction, 420 * force transaction commit may help to free it. */ 421 if (*err == -EBUSY && retries++ < 4 && EXT4_SB(sb)->s_journal && 422 jbd2_journal_force_commit_nested(EXT4_SB(sb)->s_journal)) 423 goto again; 424 return replaced_count; 425 426 repair_branches: 427 /* 428 * This should never ever happen! 429 * Extents are swapped already, but we are not able to copy data. 430 * Try to swap extents to it's original places 431 */ 432 ext4_double_down_write_data_sem(orig_inode, donor_inode); 433 replaced_count = ext4_swap_extents(handle, donor_inode, orig_inode, 434 orig_blk_offset, donor_blk_offset, 435 block_len_in_page, 0, &err2); 436 ext4_double_up_write_data_sem(orig_inode, donor_inode); 437 if (replaced_count != block_len_in_page) { 438 ext4_error_inode_block(orig_inode, (sector_t)(orig_blk_offset), 439 EIO, "Unable to copy data block," 440 " data will be lost."); 441 *err = -EIO; 442 } 443 replaced_count = 0; 444 goto unlock_folios; 445 } 446 447 /** 448 * mext_check_arguments - Check whether move extent can be done 449 * 450 * @orig_inode: original inode 451 * @donor_inode: donor inode 452 * @orig_start: logical start offset in block for orig 453 * @donor_start: logical start offset in block for donor 454 * @len: the number of blocks to be moved 455 * 456 * Check the arguments of ext4_move_extents() whether the files can be 457 * exchanged with each other. 458 * Return 0 on success, or a negative error value on failure. 459 */ 460 static int 461 mext_check_arguments(struct inode *orig_inode, 462 struct inode *donor_inode, __u64 orig_start, 463 __u64 donor_start, __u64 *len) 464 { 465 __u64 orig_eof, donor_eof; 466 unsigned int blkbits = orig_inode->i_blkbits; 467 unsigned int blocksize = 1 << blkbits; 468 469 orig_eof = (i_size_read(orig_inode) + blocksize - 1) >> blkbits; 470 donor_eof = (i_size_read(donor_inode) + blocksize - 1) >> blkbits; 471 472 473 if (donor_inode->i_mode & (S_ISUID|S_ISGID)) { 474 ext4_debug("ext4 move extent: suid or sgid is set" 475 " to donor file [ino:orig %lu, donor %lu]\n", 476 orig_inode->i_ino, donor_inode->i_ino); 477 return -EINVAL; 478 } 479 480 if (IS_IMMUTABLE(donor_inode) || IS_APPEND(donor_inode)) 481 return -EPERM; 482 483 /* Ext4 move extent does not support swap files */ 484 if (IS_SWAPFILE(orig_inode) || IS_SWAPFILE(donor_inode)) { 485 ext4_debug("ext4 move extent: The argument files should not be swap files [ino:orig %lu, donor %lu]\n", 486 orig_inode->i_ino, donor_inode->i_ino); 487 return -ETXTBSY; 488 } 489 490 if (ext4_is_quota_file(orig_inode) && ext4_is_quota_file(donor_inode)) { 491 ext4_debug("ext4 move extent: The argument files should not be quota files [ino:orig %lu, donor %lu]\n", 492 orig_inode->i_ino, donor_inode->i_ino); 493 return -EOPNOTSUPP; 494 } 495 496 /* Ext4 move extent supports only extent based file */ 497 if (!(ext4_test_inode_flag(orig_inode, EXT4_INODE_EXTENTS))) { 498 ext4_debug("ext4 move extent: orig file is not extents " 499 "based file [ino:orig %lu]\n", orig_inode->i_ino); 500 return -EOPNOTSUPP; 501 } else if (!(ext4_test_inode_flag(donor_inode, EXT4_INODE_EXTENTS))) { 502 ext4_debug("ext4 move extent: donor file is not extents " 503 "based file [ino:donor %lu]\n", donor_inode->i_ino); 504 return -EOPNOTSUPP; 505 } 506 507 if ((!orig_inode->i_size) || (!donor_inode->i_size)) { 508 ext4_debug("ext4 move extent: File size is 0 byte\n"); 509 return -EINVAL; 510 } 511 512 /* Start offset should be same */ 513 if ((orig_start & ~(PAGE_MASK >> orig_inode->i_blkbits)) != 514 (donor_start & ~(PAGE_MASK >> orig_inode->i_blkbits))) { 515 ext4_debug("ext4 move extent: orig and donor's start " 516 "offsets are not aligned [ino:orig %lu, donor %lu]\n", 517 orig_inode->i_ino, donor_inode->i_ino); 518 return -EINVAL; 519 } 520 521 if ((orig_start >= EXT_MAX_BLOCKS) || 522 (donor_start >= EXT_MAX_BLOCKS) || 523 (*len > EXT_MAX_BLOCKS) || 524 (donor_start + *len >= EXT_MAX_BLOCKS) || 525 (orig_start + *len >= EXT_MAX_BLOCKS)) { 526 ext4_debug("ext4 move extent: Can't handle over [%u] blocks " 527 "[ino:orig %lu, donor %lu]\n", EXT_MAX_BLOCKS, 528 orig_inode->i_ino, donor_inode->i_ino); 529 return -EINVAL; 530 } 531 if (orig_eof <= orig_start) 532 *len = 0; 533 else if (orig_eof < orig_start + *len - 1) 534 *len = orig_eof - orig_start; 535 if (donor_eof <= donor_start) 536 *len = 0; 537 else if (donor_eof < donor_start + *len - 1) 538 *len = donor_eof - donor_start; 539 if (!*len) { 540 ext4_debug("ext4 move extent: len should not be 0 " 541 "[ino:orig %lu, donor %lu]\n", orig_inode->i_ino, 542 donor_inode->i_ino); 543 return -EINVAL; 544 } 545 546 return 0; 547 } 548 549 /** 550 * ext4_move_extents - Exchange the specified range of a file 551 * 552 * @o_filp: file structure of the original file 553 * @d_filp: file structure of the donor file 554 * @orig_blk: start offset in block for orig 555 * @donor_blk: start offset in block for donor 556 * @len: the number of blocks to be moved 557 * @moved_len: moved block length 558 * 559 * This function returns 0 and moved block length is set in moved_len 560 * if succeed, otherwise returns error value. 561 * 562 */ 563 int 564 ext4_move_extents(struct file *o_filp, struct file *d_filp, __u64 orig_blk, 565 __u64 donor_blk, __u64 len, __u64 *moved_len) 566 { 567 struct inode *orig_inode = file_inode(o_filp); 568 struct inode *donor_inode = file_inode(d_filp); 569 struct ext4_ext_path *path = NULL; 570 int blocks_per_page = PAGE_SIZE >> orig_inode->i_blkbits; 571 ext4_lblk_t o_end, o_start = orig_blk; 572 ext4_lblk_t d_start = donor_blk; 573 int ret; 574 575 if (orig_inode->i_sb != donor_inode->i_sb) { 576 ext4_debug("ext4 move extent: The argument files " 577 "should be in same FS [ino:orig %lu, donor %lu]\n", 578 orig_inode->i_ino, donor_inode->i_ino); 579 return -EINVAL; 580 } 581 582 /* orig and donor should be different inodes */ 583 if (orig_inode == donor_inode) { 584 ext4_debug("ext4 move extent: The argument files should not " 585 "be same inode [ino:orig %lu, donor %lu]\n", 586 orig_inode->i_ino, donor_inode->i_ino); 587 return -EINVAL; 588 } 589 590 /* Regular file check */ 591 if (!S_ISREG(orig_inode->i_mode) || !S_ISREG(donor_inode->i_mode)) { 592 ext4_debug("ext4 move extent: The argument files should be " 593 "regular file [ino:orig %lu, donor %lu]\n", 594 orig_inode->i_ino, donor_inode->i_ino); 595 return -EINVAL; 596 } 597 598 /* TODO: it's not obvious how to swap blocks for inodes with full 599 journaling enabled */ 600 if (ext4_should_journal_data(orig_inode) || 601 ext4_should_journal_data(donor_inode)) { 602 ext4_msg(orig_inode->i_sb, KERN_ERR, 603 "Online defrag not supported with data journaling"); 604 return -EOPNOTSUPP; 605 } 606 607 if (IS_ENCRYPTED(orig_inode) || IS_ENCRYPTED(donor_inode)) { 608 ext4_msg(orig_inode->i_sb, KERN_ERR, 609 "Online defrag not supported for encrypted files"); 610 return -EOPNOTSUPP; 611 } 612 613 /* Protect orig and donor inodes against a truncate */ 614 lock_two_nondirectories(orig_inode, donor_inode); 615 616 /* Wait for all existing dio workers */ 617 inode_dio_wait(orig_inode); 618 inode_dio_wait(donor_inode); 619 620 /* Protect extent tree against block allocations via delalloc */ 621 ext4_double_down_write_data_sem(orig_inode, donor_inode); 622 /* Check the filesystem environment whether move_extent can be done */ 623 ret = mext_check_arguments(orig_inode, donor_inode, orig_blk, 624 donor_blk, &len); 625 if (ret) 626 goto out; 627 o_end = o_start + len; 628 629 while (o_start < o_end) { 630 struct ext4_extent *ex; 631 ext4_lblk_t cur_blk, next_blk; 632 pgoff_t orig_page_index, donor_page_index; 633 int offset_in_page; 634 int unwritten, cur_len; 635 636 ret = get_ext_path(orig_inode, o_start, &path); 637 if (ret) 638 goto out; 639 ex = path[path->p_depth].p_ext; 640 cur_blk = le32_to_cpu(ex->ee_block); 641 cur_len = ext4_ext_get_actual_len(ex); 642 /* Check hole before the start pos */ 643 if (cur_blk + cur_len - 1 < o_start) { 644 next_blk = ext4_ext_next_allocated_block(path); 645 if (next_blk == EXT_MAX_BLOCKS) { 646 ret = -ENODATA; 647 goto out; 648 } 649 d_start += next_blk - o_start; 650 o_start = next_blk; 651 continue; 652 /* Check hole after the start pos */ 653 } else if (cur_blk > o_start) { 654 /* Skip hole */ 655 d_start += cur_blk - o_start; 656 o_start = cur_blk; 657 /* Extent inside requested range ?*/ 658 if (cur_blk >= o_end) 659 goto out; 660 } else { /* in_range(o_start, o_blk, o_len) */ 661 cur_len += cur_blk - o_start; 662 } 663 unwritten = ext4_ext_is_unwritten(ex); 664 if (o_end - o_start < cur_len) 665 cur_len = o_end - o_start; 666 667 orig_page_index = o_start >> (PAGE_SHIFT - 668 orig_inode->i_blkbits); 669 donor_page_index = d_start >> (PAGE_SHIFT - 670 donor_inode->i_blkbits); 671 offset_in_page = o_start % blocks_per_page; 672 if (cur_len > blocks_per_page - offset_in_page) 673 cur_len = blocks_per_page - offset_in_page; 674 /* 675 * Up semaphore to avoid following problems: 676 * a. transaction deadlock among ext4_journal_start, 677 * ->write_begin via pagefault, and jbd2_journal_commit 678 * b. racing with ->read_folio, ->write_begin, and 679 * ext4_get_block in move_extent_per_page 680 */ 681 ext4_double_up_write_data_sem(orig_inode, donor_inode); 682 /* Swap original branches with new branches */ 683 move_extent_per_page(o_filp, donor_inode, 684 orig_page_index, donor_page_index, 685 offset_in_page, cur_len, 686 unwritten, &ret); 687 ext4_double_down_write_data_sem(orig_inode, donor_inode); 688 if (ret < 0) 689 break; 690 o_start += cur_len; 691 d_start += cur_len; 692 } 693 *moved_len = o_start - orig_blk; 694 if (*moved_len > len) 695 *moved_len = len; 696 697 out: 698 if (*moved_len) { 699 ext4_discard_preallocations(orig_inode, 0); 700 ext4_discard_preallocations(donor_inode, 0); 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