1 // SPDX-License-Identifier: GPL-2.0-only 2 /* 3 * mm/truncate.c - code for taking down pages from address_spaces 4 * 5 * Copyright (C) 2002, Linus Torvalds 6 * 7 * 10Sep2002 Andrew Morton 8 * Initial version. 9 */ 10 11 #include <linux/kernel.h> 12 #include <linux/backing-dev.h> 13 #include <linux/dax.h> 14 #include <linux/gfp.h> 15 #include <linux/mm.h> 16 #include <linux/swap.h> 17 #include <linux/export.h> 18 #include <linux/pagemap.h> 19 #include <linux/highmem.h> 20 #include <linux/pagevec.h> 21 #include <linux/task_io_accounting_ops.h> 22 #include <linux/buffer_head.h> /* grr. try_to_release_page, 23 do_invalidatepage */ 24 #include <linux/shmem_fs.h> 25 #include <linux/cleancache.h> 26 #include <linux/rmap.h> 27 #include "internal.h" 28 29 /* 30 * Regular page slots are stabilized by the page lock even without the tree 31 * itself locked. These unlocked entries need verification under the tree 32 * lock. 33 */ 34 static inline void __clear_shadow_entry(struct address_space *mapping, 35 pgoff_t index, void *entry) 36 { 37 XA_STATE(xas, &mapping->i_pages, index); 38 39 xas_set_update(&xas, workingset_update_node); 40 if (xas_load(&xas) != entry) 41 return; 42 xas_store(&xas, NULL); 43 } 44 45 static void clear_shadow_entry(struct address_space *mapping, pgoff_t index, 46 void *entry) 47 { 48 spin_lock(&mapping->host->i_lock); 49 xa_lock_irq(&mapping->i_pages); 50 __clear_shadow_entry(mapping, index, entry); 51 xa_unlock_irq(&mapping->i_pages); 52 if (mapping_shrinkable(mapping)) 53 inode_add_lru(mapping->host); 54 spin_unlock(&mapping->host->i_lock); 55 } 56 57 /* 58 * Unconditionally remove exceptional entries. Usually called from truncate 59 * path. Note that the folio_batch may be altered by this function by removing 60 * exceptional entries similar to what folio_batch_remove_exceptionals() does. 61 */ 62 static void truncate_folio_batch_exceptionals(struct address_space *mapping, 63 struct folio_batch *fbatch, pgoff_t *indices) 64 { 65 int i, j; 66 bool dax; 67 68 /* Handled by shmem itself */ 69 if (shmem_mapping(mapping)) 70 return; 71 72 for (j = 0; j < folio_batch_count(fbatch); j++) 73 if (xa_is_value(fbatch->folios[j])) 74 break; 75 76 if (j == folio_batch_count(fbatch)) 77 return; 78 79 dax = dax_mapping(mapping); 80 if (!dax) { 81 spin_lock(&mapping->host->i_lock); 82 xa_lock_irq(&mapping->i_pages); 83 } 84 85 for (i = j; i < folio_batch_count(fbatch); i++) { 86 struct folio *folio = fbatch->folios[i]; 87 pgoff_t index = indices[i]; 88 89 if (!xa_is_value(folio)) { 90 fbatch->folios[j++] = folio; 91 continue; 92 } 93 94 if (unlikely(dax)) { 95 dax_delete_mapping_entry(mapping, index); 96 continue; 97 } 98 99 __clear_shadow_entry(mapping, index, folio); 100 } 101 102 if (!dax) { 103 xa_unlock_irq(&mapping->i_pages); 104 if (mapping_shrinkable(mapping)) 105 inode_add_lru(mapping->host); 106 spin_unlock(&mapping->host->i_lock); 107 } 108 fbatch->nr = j; 109 } 110 111 /* 112 * Invalidate exceptional entry if easily possible. This handles exceptional 113 * entries for invalidate_inode_pages(). 114 */ 115 static int invalidate_exceptional_entry(struct address_space *mapping, 116 pgoff_t index, void *entry) 117 { 118 /* Handled by shmem itself, or for DAX we do nothing. */ 119 if (shmem_mapping(mapping) || dax_mapping(mapping)) 120 return 1; 121 clear_shadow_entry(mapping, index, entry); 122 return 1; 123 } 124 125 /* 126 * Invalidate exceptional entry if clean. This handles exceptional entries for 127 * invalidate_inode_pages2() so for DAX it evicts only clean entries. 128 */ 129 static int invalidate_exceptional_entry2(struct address_space *mapping, 130 pgoff_t index, void *entry) 131 { 132 /* Handled by shmem itself */ 133 if (shmem_mapping(mapping)) 134 return 1; 135 if (dax_mapping(mapping)) 136 return dax_invalidate_mapping_entry_sync(mapping, index); 137 clear_shadow_entry(mapping, index, entry); 138 return 1; 139 } 140 141 /** 142 * do_invalidatepage - invalidate part or all of a page 143 * @page: the page which is affected 144 * @offset: start of the range to invalidate 145 * @length: length of the range to invalidate 146 * 147 * do_invalidatepage() is called when all or part of the page has become 148 * invalidated by a truncate operation. 149 * 150 * do_invalidatepage() does not have to release all buffers, but it must 151 * ensure that no dirty buffer is left outside @offset and that no I/O 152 * is underway against any of the blocks which are outside the truncation 153 * point. Because the caller is about to free (and possibly reuse) those 154 * blocks on-disk. 155 */ 156 void do_invalidatepage(struct page *page, unsigned int offset, 157 unsigned int length) 158 { 159 void (*invalidatepage)(struct page *, unsigned int, unsigned int); 160 161 invalidatepage = page->mapping->a_ops->invalidatepage; 162 #ifdef CONFIG_BLOCK 163 if (!invalidatepage) 164 invalidatepage = block_invalidatepage; 165 #endif 166 if (invalidatepage) 167 (*invalidatepage)(page, offset, length); 168 } 169 170 /* 171 * If truncate cannot remove the fs-private metadata from the page, the page 172 * becomes orphaned. It will be left on the LRU and may even be mapped into 173 * user pagetables if we're racing with filemap_fault(). 174 * 175 * We need to bail out if page->mapping is no longer equal to the original 176 * mapping. This happens a) when the VM reclaimed the page while we waited on 177 * its lock, b) when a concurrent invalidate_mapping_pages got there first and 178 * c) when tmpfs swizzles a page between a tmpfs inode and swapper_space. 179 */ 180 static void truncate_cleanup_folio(struct folio *folio) 181 { 182 if (folio_mapped(folio)) 183 unmap_mapping_folio(folio); 184 185 if (folio_has_private(folio)) 186 do_invalidatepage(&folio->page, 0, folio_size(folio)); 187 188 /* 189 * Some filesystems seem to re-dirty the page even after 190 * the VM has canceled the dirty bit (eg ext3 journaling). 191 * Hence dirty accounting check is placed after invalidation. 192 */ 193 folio_cancel_dirty(folio); 194 folio_clear_mappedtodisk(folio); 195 } 196 197 /* 198 * This is for invalidate_mapping_pages(). That function can be called at 199 * any time, and is not supposed to throw away dirty pages. But pages can 200 * be marked dirty at any time too, so use remove_mapping which safely 201 * discards clean, unused pages. 202 * 203 * Returns non-zero if the page was successfully invalidated. 204 */ 205 static int 206 invalidate_complete_page(struct address_space *mapping, struct page *page) 207 { 208 209 if (page->mapping != mapping) 210 return 0; 211 212 if (page_has_private(page) && !try_to_release_page(page, 0)) 213 return 0; 214 215 return remove_mapping(mapping, page); 216 } 217 218 int truncate_inode_folio(struct address_space *mapping, struct folio *folio) 219 { 220 if (folio->mapping != mapping) 221 return -EIO; 222 223 truncate_cleanup_folio(folio); 224 filemap_remove_folio(folio); 225 return 0; 226 } 227 228 /* 229 * Handle partial folios. The folio may be entirely within the 230 * range if a split has raced with us. If not, we zero the part of the 231 * folio that's within the [start, end] range, and then split the folio if 232 * it's large. split_page_range() will discard pages which now lie beyond 233 * i_size, and we rely on the caller to discard pages which lie within a 234 * newly created hole. 235 * 236 * Returns false if splitting failed so the caller can avoid 237 * discarding the entire folio which is stubbornly unsplit. 238 */ 239 bool truncate_inode_partial_folio(struct folio *folio, loff_t start, loff_t end) 240 { 241 loff_t pos = folio_pos(folio); 242 unsigned int offset, length; 243 244 if (pos < start) 245 offset = start - pos; 246 else 247 offset = 0; 248 length = folio_size(folio); 249 if (pos + length <= (u64)end) 250 length = length - offset; 251 else 252 length = end + 1 - pos - offset; 253 254 folio_wait_writeback(folio); 255 if (length == folio_size(folio)) { 256 truncate_inode_folio(folio->mapping, folio); 257 return true; 258 } 259 260 /* 261 * We may be zeroing pages we're about to discard, but it avoids 262 * doing a complex calculation here, and then doing the zeroing 263 * anyway if the page split fails. 264 */ 265 folio_zero_range(folio, offset, length); 266 267 cleancache_invalidate_page(folio->mapping, &folio->page); 268 if (folio_has_private(folio)) 269 do_invalidatepage(&folio->page, offset, length); 270 if (!folio_test_large(folio)) 271 return true; 272 if (split_huge_page(&folio->page) == 0) 273 return true; 274 if (folio_test_dirty(folio)) 275 return false; 276 truncate_inode_folio(folio->mapping, folio); 277 return true; 278 } 279 280 /* 281 * Used to get rid of pages on hardware memory corruption. 282 */ 283 int generic_error_remove_page(struct address_space *mapping, struct page *page) 284 { 285 VM_BUG_ON_PAGE(PageTail(page), page); 286 287 if (!mapping) 288 return -EINVAL; 289 /* 290 * Only punch for normal data pages for now. 291 * Handling other types like directories would need more auditing. 292 */ 293 if (!S_ISREG(mapping->host->i_mode)) 294 return -EIO; 295 return truncate_inode_folio(mapping, page_folio(page)); 296 } 297 EXPORT_SYMBOL(generic_error_remove_page); 298 299 /* 300 * Safely invalidate one page from its pagecache mapping. 301 * It only drops clean, unused pages. The page must be locked. 302 * 303 * Returns 1 if the page is successfully invalidated, otherwise 0. 304 */ 305 int invalidate_inode_page(struct page *page) 306 { 307 struct address_space *mapping = page_mapping(page); 308 if (!mapping) 309 return 0; 310 if (PageDirty(page) || PageWriteback(page)) 311 return 0; 312 if (page_mapped(page)) 313 return 0; 314 return invalidate_complete_page(mapping, page); 315 } 316 317 /** 318 * truncate_inode_pages_range - truncate range of pages specified by start & end byte offsets 319 * @mapping: mapping to truncate 320 * @lstart: offset from which to truncate 321 * @lend: offset to which to truncate (inclusive) 322 * 323 * Truncate the page cache, removing the pages that are between 324 * specified offsets (and zeroing out partial pages 325 * if lstart or lend + 1 is not page aligned). 326 * 327 * Truncate takes two passes - the first pass is nonblocking. It will not 328 * block on page locks and it will not block on writeback. The second pass 329 * will wait. This is to prevent as much IO as possible in the affected region. 330 * The first pass will remove most pages, so the search cost of the second pass 331 * is low. 332 * 333 * We pass down the cache-hot hint to the page freeing code. Even if the 334 * mapping is large, it is probably the case that the final pages are the most 335 * recently touched, and freeing happens in ascending file offset order. 336 * 337 * Note that since ->invalidatepage() accepts range to invalidate 338 * truncate_inode_pages_range is able to handle cases where lend + 1 is not 339 * page aligned properly. 340 */ 341 void truncate_inode_pages_range(struct address_space *mapping, 342 loff_t lstart, loff_t lend) 343 { 344 pgoff_t start; /* inclusive */ 345 pgoff_t end; /* exclusive */ 346 struct folio_batch fbatch; 347 pgoff_t indices[PAGEVEC_SIZE]; 348 pgoff_t index; 349 int i; 350 struct folio *folio; 351 bool same_folio; 352 353 if (mapping_empty(mapping)) 354 goto out; 355 356 /* 357 * 'start' and 'end' always covers the range of pages to be fully 358 * truncated. Partial pages are covered with 'partial_start' at the 359 * start of the range and 'partial_end' at the end of the range. 360 * Note that 'end' is exclusive while 'lend' is inclusive. 361 */ 362 start = (lstart + PAGE_SIZE - 1) >> PAGE_SHIFT; 363 if (lend == -1) 364 /* 365 * lend == -1 indicates end-of-file so we have to set 'end' 366 * to the highest possible pgoff_t and since the type is 367 * unsigned we're using -1. 368 */ 369 end = -1; 370 else 371 end = (lend + 1) >> PAGE_SHIFT; 372 373 folio_batch_init(&fbatch); 374 index = start; 375 while (index < end && find_lock_entries(mapping, index, end - 1, 376 &fbatch, indices)) { 377 index = indices[folio_batch_count(&fbatch) - 1] + 1; 378 truncate_folio_batch_exceptionals(mapping, &fbatch, indices); 379 for (i = 0; i < folio_batch_count(&fbatch); i++) 380 truncate_cleanup_folio(fbatch.folios[i]); 381 delete_from_page_cache_batch(mapping, &fbatch); 382 for (i = 0; i < folio_batch_count(&fbatch); i++) 383 folio_unlock(fbatch.folios[i]); 384 folio_batch_release(&fbatch); 385 cond_resched(); 386 } 387 388 same_folio = (lstart >> PAGE_SHIFT) == (lend >> PAGE_SHIFT); 389 folio = __filemap_get_folio(mapping, lstart >> PAGE_SHIFT, FGP_LOCK, 0); 390 if (folio) { 391 same_folio = lend < folio_pos(folio) + folio_size(folio); 392 if (!truncate_inode_partial_folio(folio, lstart, lend)) { 393 start = folio->index + folio_nr_pages(folio); 394 if (same_folio) 395 end = folio->index; 396 } 397 folio_unlock(folio); 398 folio_put(folio); 399 folio = NULL; 400 } 401 402 if (!same_folio) 403 folio = __filemap_get_folio(mapping, lend >> PAGE_SHIFT, 404 FGP_LOCK, 0); 405 if (folio) { 406 if (!truncate_inode_partial_folio(folio, lstart, lend)) 407 end = folio->index; 408 folio_unlock(folio); 409 folio_put(folio); 410 } 411 412 index = start; 413 while (index < end) { 414 cond_resched(); 415 if (!find_get_entries(mapping, index, end - 1, &fbatch, 416 indices)) { 417 /* If all gone from start onwards, we're done */ 418 if (index == start) 419 break; 420 /* Otherwise restart to make sure all gone */ 421 index = start; 422 continue; 423 } 424 425 for (i = 0; i < folio_batch_count(&fbatch); i++) { 426 struct folio *folio = fbatch.folios[i]; 427 428 /* We rely upon deletion not changing page->index */ 429 index = indices[i]; 430 431 if (xa_is_value(folio)) 432 continue; 433 434 folio_lock(folio); 435 VM_BUG_ON_FOLIO(!folio_contains(folio, index), folio); 436 folio_wait_writeback(folio); 437 truncate_inode_folio(mapping, folio); 438 folio_unlock(folio); 439 index = folio_index(folio) + folio_nr_pages(folio) - 1; 440 } 441 truncate_folio_batch_exceptionals(mapping, &fbatch, indices); 442 folio_batch_release(&fbatch); 443 index++; 444 } 445 446 out: 447 cleancache_invalidate_inode(mapping); 448 } 449 EXPORT_SYMBOL(truncate_inode_pages_range); 450 451 /** 452 * truncate_inode_pages - truncate *all* the pages from an offset 453 * @mapping: mapping to truncate 454 * @lstart: offset from which to truncate 455 * 456 * Called under (and serialised by) inode->i_rwsem and 457 * mapping->invalidate_lock. 458 * 459 * Note: When this function returns, there can be a page in the process of 460 * deletion (inside __delete_from_page_cache()) in the specified range. Thus 461 * mapping->nrpages can be non-zero when this function returns even after 462 * truncation of the whole mapping. 463 */ 464 void truncate_inode_pages(struct address_space *mapping, loff_t lstart) 465 { 466 truncate_inode_pages_range(mapping, lstart, (loff_t)-1); 467 } 468 EXPORT_SYMBOL(truncate_inode_pages); 469 470 /** 471 * truncate_inode_pages_final - truncate *all* pages before inode dies 472 * @mapping: mapping to truncate 473 * 474 * Called under (and serialized by) inode->i_rwsem. 475 * 476 * Filesystems have to use this in the .evict_inode path to inform the 477 * VM that this is the final truncate and the inode is going away. 478 */ 479 void truncate_inode_pages_final(struct address_space *mapping) 480 { 481 /* 482 * Page reclaim can not participate in regular inode lifetime 483 * management (can't call iput()) and thus can race with the 484 * inode teardown. Tell it when the address space is exiting, 485 * so that it does not install eviction information after the 486 * final truncate has begun. 487 */ 488 mapping_set_exiting(mapping); 489 490 if (!mapping_empty(mapping)) { 491 /* 492 * As truncation uses a lockless tree lookup, cycle 493 * the tree lock to make sure any ongoing tree 494 * modification that does not see AS_EXITING is 495 * completed before starting the final truncate. 496 */ 497 xa_lock_irq(&mapping->i_pages); 498 xa_unlock_irq(&mapping->i_pages); 499 } 500 501 /* 502 * Cleancache needs notification even if there are no pages or shadow 503 * entries. 504 */ 505 truncate_inode_pages(mapping, 0); 506 } 507 EXPORT_SYMBOL(truncate_inode_pages_final); 508 509 static unsigned long __invalidate_mapping_pages(struct address_space *mapping, 510 pgoff_t start, pgoff_t end, unsigned long *nr_pagevec) 511 { 512 pgoff_t indices[PAGEVEC_SIZE]; 513 struct folio_batch fbatch; 514 pgoff_t index = start; 515 unsigned long ret; 516 unsigned long count = 0; 517 int i; 518 519 folio_batch_init(&fbatch); 520 while (find_lock_entries(mapping, index, end, &fbatch, indices)) { 521 for (i = 0; i < folio_batch_count(&fbatch); i++) { 522 struct page *page = &fbatch.folios[i]->page; 523 524 /* We rely upon deletion not changing page->index */ 525 index = indices[i]; 526 527 if (xa_is_value(page)) { 528 count += invalidate_exceptional_entry(mapping, 529 index, 530 page); 531 continue; 532 } 533 index += thp_nr_pages(page) - 1; 534 535 ret = invalidate_inode_page(page); 536 unlock_page(page); 537 /* 538 * Invalidation is a hint that the page is no longer 539 * of interest and try to speed up its reclaim. 540 */ 541 if (!ret) { 542 deactivate_file_page(page); 543 /* It is likely on the pagevec of a remote CPU */ 544 if (nr_pagevec) 545 (*nr_pagevec)++; 546 } 547 count += ret; 548 } 549 folio_batch_remove_exceptionals(&fbatch); 550 folio_batch_release(&fbatch); 551 cond_resched(); 552 index++; 553 } 554 return count; 555 } 556 557 /** 558 * invalidate_mapping_pages - Invalidate all clean, unlocked cache of one inode 559 * @mapping: the address_space which holds the cache to invalidate 560 * @start: the offset 'from' which to invalidate 561 * @end: the offset 'to' which to invalidate (inclusive) 562 * 563 * This function removes pages that are clean, unmapped and unlocked, 564 * as well as shadow entries. It will not block on IO activity. 565 * 566 * If you want to remove all the pages of one inode, regardless of 567 * their use and writeback state, use truncate_inode_pages(). 568 * 569 * Return: the number of the cache entries that were invalidated 570 */ 571 unsigned long invalidate_mapping_pages(struct address_space *mapping, 572 pgoff_t start, pgoff_t end) 573 { 574 return __invalidate_mapping_pages(mapping, start, end, NULL); 575 } 576 EXPORT_SYMBOL(invalidate_mapping_pages); 577 578 /** 579 * invalidate_mapping_pagevec - Invalidate all the unlocked pages of one inode 580 * @mapping: the address_space which holds the pages to invalidate 581 * @start: the offset 'from' which to invalidate 582 * @end: the offset 'to' which to invalidate (inclusive) 583 * @nr_pagevec: invalidate failed page number for caller 584 * 585 * This helper is similar to invalidate_mapping_pages(), except that it accounts 586 * for pages that are likely on a pagevec and counts them in @nr_pagevec, which 587 * will be used by the caller. 588 */ 589 void invalidate_mapping_pagevec(struct address_space *mapping, 590 pgoff_t start, pgoff_t end, unsigned long *nr_pagevec) 591 { 592 __invalidate_mapping_pages(mapping, start, end, nr_pagevec); 593 } 594 595 /* 596 * This is like invalidate_complete_page(), except it ignores the page's 597 * refcount. We do this because invalidate_inode_pages2() needs stronger 598 * invalidation guarantees, and cannot afford to leave pages behind because 599 * shrink_page_list() has a temp ref on them, or because they're transiently 600 * sitting in the lru_cache_add() pagevecs. 601 */ 602 static int invalidate_complete_folio2(struct address_space *mapping, 603 struct folio *folio) 604 { 605 if (folio->mapping != mapping) 606 return 0; 607 608 if (folio_has_private(folio) && 609 !filemap_release_folio(folio, GFP_KERNEL)) 610 return 0; 611 612 spin_lock(&mapping->host->i_lock); 613 xa_lock_irq(&mapping->i_pages); 614 if (folio_test_dirty(folio)) 615 goto failed; 616 617 BUG_ON(folio_has_private(folio)); 618 __filemap_remove_folio(folio, NULL); 619 xa_unlock_irq(&mapping->i_pages); 620 if (mapping_shrinkable(mapping)) 621 inode_add_lru(mapping->host); 622 spin_unlock(&mapping->host->i_lock); 623 624 filemap_free_folio(mapping, folio); 625 return 1; 626 failed: 627 xa_unlock_irq(&mapping->i_pages); 628 spin_unlock(&mapping->host->i_lock); 629 return 0; 630 } 631 632 static int do_launder_folio(struct address_space *mapping, struct folio *folio) 633 { 634 if (!folio_test_dirty(folio)) 635 return 0; 636 if (folio->mapping != mapping || mapping->a_ops->launder_page == NULL) 637 return 0; 638 return mapping->a_ops->launder_page(&folio->page); 639 } 640 641 /** 642 * invalidate_inode_pages2_range - remove range of pages from an address_space 643 * @mapping: the address_space 644 * @start: the page offset 'from' which to invalidate 645 * @end: the page offset 'to' which to invalidate (inclusive) 646 * 647 * Any pages which are found to be mapped into pagetables are unmapped prior to 648 * invalidation. 649 * 650 * Return: -EBUSY if any pages could not be invalidated. 651 */ 652 int invalidate_inode_pages2_range(struct address_space *mapping, 653 pgoff_t start, pgoff_t end) 654 { 655 pgoff_t indices[PAGEVEC_SIZE]; 656 struct folio_batch fbatch; 657 pgoff_t index; 658 int i; 659 int ret = 0; 660 int ret2 = 0; 661 int did_range_unmap = 0; 662 663 if (mapping_empty(mapping)) 664 goto out; 665 666 folio_batch_init(&fbatch); 667 index = start; 668 while (find_get_entries(mapping, index, end, &fbatch, indices)) { 669 for (i = 0; i < folio_batch_count(&fbatch); i++) { 670 struct folio *folio = fbatch.folios[i]; 671 672 /* We rely upon deletion not changing folio->index */ 673 index = indices[i]; 674 675 if (xa_is_value(folio)) { 676 if (!invalidate_exceptional_entry2(mapping, 677 index, folio)) 678 ret = -EBUSY; 679 continue; 680 } 681 682 if (!did_range_unmap && folio_mapped(folio)) { 683 /* 684 * If folio is mapped, before taking its lock, 685 * zap the rest of the file in one hit. 686 */ 687 unmap_mapping_pages(mapping, index, 688 (1 + end - index), false); 689 did_range_unmap = 1; 690 } 691 692 folio_lock(folio); 693 VM_BUG_ON_FOLIO(!folio_contains(folio, index), folio); 694 if (folio->mapping != mapping) { 695 folio_unlock(folio); 696 continue; 697 } 698 folio_wait_writeback(folio); 699 700 if (folio_mapped(folio)) 701 unmap_mapping_folio(folio); 702 BUG_ON(folio_mapped(folio)); 703 704 ret2 = do_launder_folio(mapping, folio); 705 if (ret2 == 0) { 706 if (!invalidate_complete_folio2(mapping, folio)) 707 ret2 = -EBUSY; 708 } 709 if (ret2 < 0) 710 ret = ret2; 711 folio_unlock(folio); 712 } 713 folio_batch_remove_exceptionals(&fbatch); 714 folio_batch_release(&fbatch); 715 cond_resched(); 716 index++; 717 } 718 /* 719 * For DAX we invalidate page tables after invalidating page cache. We 720 * could invalidate page tables while invalidating each entry however 721 * that would be expensive. And doing range unmapping before doesn't 722 * work as we have no cheap way to find whether page cache entry didn't 723 * get remapped later. 724 */ 725 if (dax_mapping(mapping)) { 726 unmap_mapping_pages(mapping, start, end - start + 1, false); 727 } 728 out: 729 cleancache_invalidate_inode(mapping); 730 return ret; 731 } 732 EXPORT_SYMBOL_GPL(invalidate_inode_pages2_range); 733 734 /** 735 * invalidate_inode_pages2 - remove all pages from an address_space 736 * @mapping: the address_space 737 * 738 * Any pages which are found to be mapped into pagetables are unmapped prior to 739 * invalidation. 740 * 741 * Return: -EBUSY if any pages could not be invalidated. 742 */ 743 int invalidate_inode_pages2(struct address_space *mapping) 744 { 745 return invalidate_inode_pages2_range(mapping, 0, -1); 746 } 747 EXPORT_SYMBOL_GPL(invalidate_inode_pages2); 748 749 /** 750 * truncate_pagecache - unmap and remove pagecache that has been truncated 751 * @inode: inode 752 * @newsize: new file size 753 * 754 * inode's new i_size must already be written before truncate_pagecache 755 * is called. 756 * 757 * This function should typically be called before the filesystem 758 * releases resources associated with the freed range (eg. deallocates 759 * blocks). This way, pagecache will always stay logically coherent 760 * with on-disk format, and the filesystem would not have to deal with 761 * situations such as writepage being called for a page that has already 762 * had its underlying blocks deallocated. 763 */ 764 void truncate_pagecache(struct inode *inode, loff_t newsize) 765 { 766 struct address_space *mapping = inode->i_mapping; 767 loff_t holebegin = round_up(newsize, PAGE_SIZE); 768 769 /* 770 * unmap_mapping_range is called twice, first simply for 771 * efficiency so that truncate_inode_pages does fewer 772 * single-page unmaps. However after this first call, and 773 * before truncate_inode_pages finishes, it is possible for 774 * private pages to be COWed, which remain after 775 * truncate_inode_pages finishes, hence the second 776 * unmap_mapping_range call must be made for correctness. 777 */ 778 unmap_mapping_range(mapping, holebegin, 0, 1); 779 truncate_inode_pages(mapping, newsize); 780 unmap_mapping_range(mapping, holebegin, 0, 1); 781 } 782 EXPORT_SYMBOL(truncate_pagecache); 783 784 /** 785 * truncate_setsize - update inode and pagecache for a new file size 786 * @inode: inode 787 * @newsize: new file size 788 * 789 * truncate_setsize updates i_size and performs pagecache truncation (if 790 * necessary) to @newsize. It will be typically be called from the filesystem's 791 * setattr function when ATTR_SIZE is passed in. 792 * 793 * Must be called with a lock serializing truncates and writes (generally 794 * i_rwsem but e.g. xfs uses a different lock) and before all filesystem 795 * specific block truncation has been performed. 796 */ 797 void truncate_setsize(struct inode *inode, loff_t newsize) 798 { 799 loff_t oldsize = inode->i_size; 800 801 i_size_write(inode, newsize); 802 if (newsize > oldsize) 803 pagecache_isize_extended(inode, oldsize, newsize); 804 truncate_pagecache(inode, newsize); 805 } 806 EXPORT_SYMBOL(truncate_setsize); 807 808 /** 809 * pagecache_isize_extended - update pagecache after extension of i_size 810 * @inode: inode for which i_size was extended 811 * @from: original inode size 812 * @to: new inode size 813 * 814 * Handle extension of inode size either caused by extending truncate or by 815 * write starting after current i_size. We mark the page straddling current 816 * i_size RO so that page_mkwrite() is called on the nearest write access to 817 * the page. This way filesystem can be sure that page_mkwrite() is called on 818 * the page before user writes to the page via mmap after the i_size has been 819 * changed. 820 * 821 * The function must be called after i_size is updated so that page fault 822 * coming after we unlock the page will already see the new i_size. 823 * The function must be called while we still hold i_rwsem - this not only 824 * makes sure i_size is stable but also that userspace cannot observe new 825 * i_size value before we are prepared to store mmap writes at new inode size. 826 */ 827 void pagecache_isize_extended(struct inode *inode, loff_t from, loff_t to) 828 { 829 int bsize = i_blocksize(inode); 830 loff_t rounded_from; 831 struct page *page; 832 pgoff_t index; 833 834 WARN_ON(to > inode->i_size); 835 836 if (from >= to || bsize == PAGE_SIZE) 837 return; 838 /* Page straddling @from will not have any hole block created? */ 839 rounded_from = round_up(from, bsize); 840 if (to <= rounded_from || !(rounded_from & (PAGE_SIZE - 1))) 841 return; 842 843 index = from >> PAGE_SHIFT; 844 page = find_lock_page(inode->i_mapping, index); 845 /* Page not cached? Nothing to do */ 846 if (!page) 847 return; 848 /* 849 * See clear_page_dirty_for_io() for details why set_page_dirty() 850 * is needed. 851 */ 852 if (page_mkclean(page)) 853 set_page_dirty(page); 854 unlock_page(page); 855 put_page(page); 856 } 857 EXPORT_SYMBOL(pagecache_isize_extended); 858 859 /** 860 * truncate_pagecache_range - unmap and remove pagecache that is hole-punched 861 * @inode: inode 862 * @lstart: offset of beginning of hole 863 * @lend: offset of last byte of hole 864 * 865 * This function should typically be called before the filesystem 866 * releases resources associated with the freed range (eg. deallocates 867 * blocks). This way, pagecache will always stay logically coherent 868 * with on-disk format, and the filesystem would not have to deal with 869 * situations such as writepage being called for a page that has already 870 * had its underlying blocks deallocated. 871 */ 872 void truncate_pagecache_range(struct inode *inode, loff_t lstart, loff_t lend) 873 { 874 struct address_space *mapping = inode->i_mapping; 875 loff_t unmap_start = round_up(lstart, PAGE_SIZE); 876 loff_t unmap_end = round_down(1 + lend, PAGE_SIZE) - 1; 877 /* 878 * This rounding is currently just for example: unmap_mapping_range 879 * expands its hole outwards, whereas we want it to contract the hole 880 * inwards. However, existing callers of truncate_pagecache_range are 881 * doing their own page rounding first. Note that unmap_mapping_range 882 * allows holelen 0 for all, and we allow lend -1 for end of file. 883 */ 884 885 /* 886 * Unlike in truncate_pagecache, unmap_mapping_range is called only 887 * once (before truncating pagecache), and without "even_cows" flag: 888 * hole-punching should not remove private COWed pages from the hole. 889 */ 890 if ((u64)unmap_end > (u64)unmap_start) 891 unmap_mapping_range(mapping, unmap_start, 892 1 + unmap_end - unmap_start, 0); 893 truncate_inode_pages_range(mapping, lstart, lend); 894 } 895 EXPORT_SYMBOL(truncate_pagecache_range); 896