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 mapping->nrexceptional--; 44 } 45 46 static void clear_shadow_entry(struct address_space *mapping, pgoff_t index, 47 void *entry) 48 { 49 xa_lock_irq(&mapping->i_pages); 50 __clear_shadow_entry(mapping, index, entry); 51 xa_unlock_irq(&mapping->i_pages); 52 } 53 54 /* 55 * Unconditionally remove exceptional entries. Usually called from truncate 56 * path. Note that the pagevec may be altered by this function by removing 57 * exceptional entries similar to what pagevec_remove_exceptionals does. 58 */ 59 static void truncate_exceptional_pvec_entries(struct address_space *mapping, 60 struct pagevec *pvec, pgoff_t *indices) 61 { 62 int i, j; 63 bool dax; 64 65 /* Handled by shmem itself */ 66 if (shmem_mapping(mapping)) 67 return; 68 69 for (j = 0; j < pagevec_count(pvec); j++) 70 if (xa_is_value(pvec->pages[j])) 71 break; 72 73 if (j == pagevec_count(pvec)) 74 return; 75 76 dax = dax_mapping(mapping); 77 if (!dax) 78 xa_lock_irq(&mapping->i_pages); 79 80 for (i = j; i < pagevec_count(pvec); i++) { 81 struct page *page = pvec->pages[i]; 82 pgoff_t index = indices[i]; 83 84 if (!xa_is_value(page)) { 85 pvec->pages[j++] = page; 86 continue; 87 } 88 89 if (unlikely(dax)) { 90 dax_delete_mapping_entry(mapping, index); 91 continue; 92 } 93 94 __clear_shadow_entry(mapping, index, page); 95 } 96 97 if (!dax) 98 xa_unlock_irq(&mapping->i_pages); 99 pvec->nr = j; 100 } 101 102 /* 103 * Invalidate exceptional entry if easily possible. This handles exceptional 104 * entries for invalidate_inode_pages(). 105 */ 106 static int invalidate_exceptional_entry(struct address_space *mapping, 107 pgoff_t index, void *entry) 108 { 109 /* Handled by shmem itself, or for DAX we do nothing. */ 110 if (shmem_mapping(mapping) || dax_mapping(mapping)) 111 return 1; 112 clear_shadow_entry(mapping, index, entry); 113 return 1; 114 } 115 116 /* 117 * Invalidate exceptional entry if clean. This handles exceptional entries for 118 * invalidate_inode_pages2() so for DAX it evicts only clean entries. 119 */ 120 static int invalidate_exceptional_entry2(struct address_space *mapping, 121 pgoff_t index, void *entry) 122 { 123 /* Handled by shmem itself */ 124 if (shmem_mapping(mapping)) 125 return 1; 126 if (dax_mapping(mapping)) 127 return dax_invalidate_mapping_entry_sync(mapping, index); 128 clear_shadow_entry(mapping, index, entry); 129 return 1; 130 } 131 132 /** 133 * do_invalidatepage - invalidate part or all of a page 134 * @page: the page which is affected 135 * @offset: start of the range to invalidate 136 * @length: length of the range to invalidate 137 * 138 * do_invalidatepage() is called when all or part of the page has become 139 * invalidated by a truncate operation. 140 * 141 * do_invalidatepage() does not have to release all buffers, but it must 142 * ensure that no dirty buffer is left outside @offset and that no I/O 143 * is underway against any of the blocks which are outside the truncation 144 * point. Because the caller is about to free (and possibly reuse) those 145 * blocks on-disk. 146 */ 147 void do_invalidatepage(struct page *page, unsigned int offset, 148 unsigned int length) 149 { 150 void (*invalidatepage)(struct page *, unsigned int, unsigned int); 151 152 invalidatepage = page->mapping->a_ops->invalidatepage; 153 #ifdef CONFIG_BLOCK 154 if (!invalidatepage) 155 invalidatepage = block_invalidatepage; 156 #endif 157 if (invalidatepage) 158 (*invalidatepage)(page, offset, length); 159 } 160 161 /* 162 * If truncate cannot remove the fs-private metadata from the page, the page 163 * becomes orphaned. It will be left on the LRU and may even be mapped into 164 * user pagetables if we're racing with filemap_fault(). 165 * 166 * We need to bail out if page->mapping is no longer equal to the original 167 * mapping. This happens a) when the VM reclaimed the page while we waited on 168 * its lock, b) when a concurrent invalidate_mapping_pages got there first and 169 * c) when tmpfs swizzles a page between a tmpfs inode and swapper_space. 170 */ 171 static void 172 truncate_cleanup_page(struct address_space *mapping, struct page *page) 173 { 174 if (page_mapped(page)) { 175 unsigned int nr = thp_nr_pages(page); 176 unmap_mapping_pages(mapping, page->index, nr, false); 177 } 178 179 if (page_has_private(page)) 180 do_invalidatepage(page, 0, thp_size(page)); 181 182 /* 183 * Some filesystems seem to re-dirty the page even after 184 * the VM has canceled the dirty bit (eg ext3 journaling). 185 * Hence dirty accounting check is placed after invalidation. 186 */ 187 cancel_dirty_page(page); 188 ClearPageMappedToDisk(page); 189 } 190 191 /* 192 * This is for invalidate_mapping_pages(). That function can be called at 193 * any time, and is not supposed to throw away dirty pages. But pages can 194 * be marked dirty at any time too, so use remove_mapping which safely 195 * discards clean, unused pages. 196 * 197 * Returns non-zero if the page was successfully invalidated. 198 */ 199 static int 200 invalidate_complete_page(struct address_space *mapping, struct page *page) 201 { 202 int ret; 203 204 if (page->mapping != mapping) 205 return 0; 206 207 if (page_has_private(page) && !try_to_release_page(page, 0)) 208 return 0; 209 210 ret = remove_mapping(mapping, page); 211 212 return ret; 213 } 214 215 int truncate_inode_page(struct address_space *mapping, struct page *page) 216 { 217 VM_BUG_ON_PAGE(PageTail(page), page); 218 219 if (page->mapping != mapping) 220 return -EIO; 221 222 truncate_cleanup_page(mapping, page); 223 delete_from_page_cache(page); 224 return 0; 225 } 226 227 /* 228 * Used to get rid of pages on hardware memory corruption. 229 */ 230 int generic_error_remove_page(struct address_space *mapping, struct page *page) 231 { 232 if (!mapping) 233 return -EINVAL; 234 /* 235 * Only punch for normal data pages for now. 236 * Handling other types like directories would need more auditing. 237 */ 238 if (!S_ISREG(mapping->host->i_mode)) 239 return -EIO; 240 return truncate_inode_page(mapping, page); 241 } 242 EXPORT_SYMBOL(generic_error_remove_page); 243 244 /* 245 * Safely invalidate one page from its pagecache mapping. 246 * It only drops clean, unused pages. The page must be locked. 247 * 248 * Returns 1 if the page is successfully invalidated, otherwise 0. 249 */ 250 int invalidate_inode_page(struct page *page) 251 { 252 struct address_space *mapping = page_mapping(page); 253 if (!mapping) 254 return 0; 255 if (PageDirty(page) || PageWriteback(page)) 256 return 0; 257 if (page_mapped(page)) 258 return 0; 259 return invalidate_complete_page(mapping, page); 260 } 261 262 /** 263 * truncate_inode_pages_range - truncate range of pages specified by start & end byte offsets 264 * @mapping: mapping to truncate 265 * @lstart: offset from which to truncate 266 * @lend: offset to which to truncate (inclusive) 267 * 268 * Truncate the page cache, removing the pages that are between 269 * specified offsets (and zeroing out partial pages 270 * if lstart or lend + 1 is not page aligned). 271 * 272 * Truncate takes two passes - the first pass is nonblocking. It will not 273 * block on page locks and it will not block on writeback. The second pass 274 * will wait. This is to prevent as much IO as possible in the affected region. 275 * The first pass will remove most pages, so the search cost of the second pass 276 * is low. 277 * 278 * We pass down the cache-hot hint to the page freeing code. Even if the 279 * mapping is large, it is probably the case that the final pages are the most 280 * recently touched, and freeing happens in ascending file offset order. 281 * 282 * Note that since ->invalidatepage() accepts range to invalidate 283 * truncate_inode_pages_range is able to handle cases where lend + 1 is not 284 * page aligned properly. 285 */ 286 void truncate_inode_pages_range(struct address_space *mapping, 287 loff_t lstart, loff_t lend) 288 { 289 pgoff_t start; /* inclusive */ 290 pgoff_t end; /* exclusive */ 291 unsigned int partial_start; /* inclusive */ 292 unsigned int partial_end; /* exclusive */ 293 struct pagevec pvec; 294 pgoff_t indices[PAGEVEC_SIZE]; 295 pgoff_t index; 296 int i; 297 298 if (mapping->nrpages == 0 && mapping->nrexceptional == 0) 299 goto out; 300 301 /* Offsets within partial pages */ 302 partial_start = lstart & (PAGE_SIZE - 1); 303 partial_end = (lend + 1) & (PAGE_SIZE - 1); 304 305 /* 306 * 'start' and 'end' always covers the range of pages to be fully 307 * truncated. Partial pages are covered with 'partial_start' at the 308 * start of the range and 'partial_end' at the end of the range. 309 * Note that 'end' is exclusive while 'lend' is inclusive. 310 */ 311 start = (lstart + PAGE_SIZE - 1) >> PAGE_SHIFT; 312 if (lend == -1) 313 /* 314 * lend == -1 indicates end-of-file so we have to set 'end' 315 * to the highest possible pgoff_t and since the type is 316 * unsigned we're using -1. 317 */ 318 end = -1; 319 else 320 end = (lend + 1) >> PAGE_SHIFT; 321 322 pagevec_init(&pvec); 323 index = start; 324 while (index < end && find_lock_entries(mapping, index, end - 1, 325 &pvec, indices)) { 326 index = indices[pagevec_count(&pvec) - 1] + 1; 327 truncate_exceptional_pvec_entries(mapping, &pvec, indices); 328 for (i = 0; i < pagevec_count(&pvec); i++) 329 truncate_cleanup_page(mapping, pvec.pages[i]); 330 delete_from_page_cache_batch(mapping, &pvec); 331 for (i = 0; i < pagevec_count(&pvec); i++) 332 unlock_page(pvec.pages[i]); 333 pagevec_release(&pvec); 334 cond_resched(); 335 } 336 337 if (partial_start) { 338 struct page *page = find_lock_page(mapping, start - 1); 339 if (page) { 340 unsigned int top = PAGE_SIZE; 341 if (start > end) { 342 /* Truncation within a single page */ 343 top = partial_end; 344 partial_end = 0; 345 } 346 wait_on_page_writeback(page); 347 zero_user_segment(page, partial_start, top); 348 cleancache_invalidate_page(mapping, page); 349 if (page_has_private(page)) 350 do_invalidatepage(page, partial_start, 351 top - partial_start); 352 unlock_page(page); 353 put_page(page); 354 } 355 } 356 if (partial_end) { 357 struct page *page = find_lock_page(mapping, end); 358 if (page) { 359 wait_on_page_writeback(page); 360 zero_user_segment(page, 0, partial_end); 361 cleancache_invalidate_page(mapping, page); 362 if (page_has_private(page)) 363 do_invalidatepage(page, 0, 364 partial_end); 365 unlock_page(page); 366 put_page(page); 367 } 368 } 369 /* 370 * If the truncation happened within a single page no pages 371 * will be released, just zeroed, so we can bail out now. 372 */ 373 if (start >= end) 374 goto out; 375 376 index = start; 377 for ( ; ; ) { 378 cond_resched(); 379 if (!find_get_entries(mapping, index, end - 1, &pvec, 380 indices)) { 381 /* If all gone from start onwards, we're done */ 382 if (index == start) 383 break; 384 /* Otherwise restart to make sure all gone */ 385 index = start; 386 continue; 387 } 388 389 for (i = 0; i < pagevec_count(&pvec); i++) { 390 struct page *page = pvec.pages[i]; 391 392 /* We rely upon deletion not changing page->index */ 393 index = indices[i]; 394 395 if (xa_is_value(page)) 396 continue; 397 398 lock_page(page); 399 WARN_ON(page_to_index(page) != index); 400 wait_on_page_writeback(page); 401 truncate_inode_page(mapping, page); 402 unlock_page(page); 403 } 404 truncate_exceptional_pvec_entries(mapping, &pvec, indices); 405 pagevec_release(&pvec); 406 index++; 407 } 408 409 out: 410 cleancache_invalidate_inode(mapping); 411 } 412 EXPORT_SYMBOL(truncate_inode_pages_range); 413 414 /** 415 * truncate_inode_pages - truncate *all* the pages from an offset 416 * @mapping: mapping to truncate 417 * @lstart: offset from which to truncate 418 * 419 * Called under (and serialised by) inode->i_mutex. 420 * 421 * Note: When this function returns, there can be a page in the process of 422 * deletion (inside __delete_from_page_cache()) in the specified range. Thus 423 * mapping->nrpages can be non-zero when this function returns even after 424 * truncation of the whole mapping. 425 */ 426 void truncate_inode_pages(struct address_space *mapping, loff_t lstart) 427 { 428 truncate_inode_pages_range(mapping, lstart, (loff_t)-1); 429 } 430 EXPORT_SYMBOL(truncate_inode_pages); 431 432 /** 433 * truncate_inode_pages_final - truncate *all* pages before inode dies 434 * @mapping: mapping to truncate 435 * 436 * Called under (and serialized by) inode->i_mutex. 437 * 438 * Filesystems have to use this in the .evict_inode path to inform the 439 * VM that this is the final truncate and the inode is going away. 440 */ 441 void truncate_inode_pages_final(struct address_space *mapping) 442 { 443 unsigned long nrexceptional; 444 unsigned long nrpages; 445 446 /* 447 * Page reclaim can not participate in regular inode lifetime 448 * management (can't call iput()) and thus can race with the 449 * inode teardown. Tell it when the address space is exiting, 450 * so that it does not install eviction information after the 451 * final truncate has begun. 452 */ 453 mapping_set_exiting(mapping); 454 455 /* 456 * When reclaim installs eviction entries, it increases 457 * nrexceptional first, then decreases nrpages. Make sure we see 458 * this in the right order or we might miss an entry. 459 */ 460 nrpages = mapping->nrpages; 461 smp_rmb(); 462 nrexceptional = mapping->nrexceptional; 463 464 if (nrpages || nrexceptional) { 465 /* 466 * As truncation uses a lockless tree lookup, cycle 467 * the tree lock to make sure any ongoing tree 468 * modification that does not see AS_EXITING is 469 * completed before starting the final truncate. 470 */ 471 xa_lock_irq(&mapping->i_pages); 472 xa_unlock_irq(&mapping->i_pages); 473 } 474 475 /* 476 * Cleancache needs notification even if there are no pages or shadow 477 * entries. 478 */ 479 truncate_inode_pages(mapping, 0); 480 } 481 EXPORT_SYMBOL(truncate_inode_pages_final); 482 483 static unsigned long __invalidate_mapping_pages(struct address_space *mapping, 484 pgoff_t start, pgoff_t end, unsigned long *nr_pagevec) 485 { 486 pgoff_t indices[PAGEVEC_SIZE]; 487 struct pagevec pvec; 488 pgoff_t index = start; 489 unsigned long ret; 490 unsigned long count = 0; 491 int i; 492 493 pagevec_init(&pvec); 494 while (find_lock_entries(mapping, index, end, &pvec, indices)) { 495 for (i = 0; i < pagevec_count(&pvec); i++) { 496 struct page *page = pvec.pages[i]; 497 498 /* We rely upon deletion not changing page->index */ 499 index = indices[i]; 500 501 if (xa_is_value(page)) { 502 invalidate_exceptional_entry(mapping, index, 503 page); 504 continue; 505 } 506 index += thp_nr_pages(page) - 1; 507 508 ret = invalidate_inode_page(page); 509 unlock_page(page); 510 /* 511 * Invalidation is a hint that the page is no longer 512 * of interest and try to speed up its reclaim. 513 */ 514 if (!ret) { 515 deactivate_file_page(page); 516 /* It is likely on the pagevec of a remote CPU */ 517 if (nr_pagevec) 518 (*nr_pagevec)++; 519 } 520 count += ret; 521 } 522 pagevec_remove_exceptionals(&pvec); 523 pagevec_release(&pvec); 524 cond_resched(); 525 index++; 526 } 527 return count; 528 } 529 530 /** 531 * invalidate_mapping_pages - Invalidate all the unlocked pages of one inode 532 * @mapping: the address_space which holds the pages to invalidate 533 * @start: the offset 'from' which to invalidate 534 * @end: the offset 'to' which to invalidate (inclusive) 535 * 536 * This function only removes the unlocked pages, if you want to 537 * remove all the pages of one inode, you must call truncate_inode_pages. 538 * 539 * invalidate_mapping_pages() will not block on IO activity. It will not 540 * invalidate pages which are dirty, locked, under writeback or mapped into 541 * pagetables. 542 * 543 * Return: the number of the pages that were invalidated 544 */ 545 unsigned long invalidate_mapping_pages(struct address_space *mapping, 546 pgoff_t start, pgoff_t end) 547 { 548 return __invalidate_mapping_pages(mapping, start, end, NULL); 549 } 550 EXPORT_SYMBOL(invalidate_mapping_pages); 551 552 /** 553 * invalidate_mapping_pagevec - Invalidate all the unlocked pages of one inode 554 * @mapping: the address_space which holds the pages to invalidate 555 * @start: the offset 'from' which to invalidate 556 * @end: the offset 'to' which to invalidate (inclusive) 557 * @nr_pagevec: invalidate failed page number for caller 558 * 559 * This helper is similar to invalidate_mapping_pages(), except that it accounts 560 * for pages that are likely on a pagevec and counts them in @nr_pagevec, which 561 * will be used by the caller. 562 */ 563 void invalidate_mapping_pagevec(struct address_space *mapping, 564 pgoff_t start, pgoff_t end, unsigned long *nr_pagevec) 565 { 566 __invalidate_mapping_pages(mapping, start, end, nr_pagevec); 567 } 568 569 /* 570 * This is like invalidate_complete_page(), except it ignores the page's 571 * refcount. We do this because invalidate_inode_pages2() needs stronger 572 * invalidation guarantees, and cannot afford to leave pages behind because 573 * shrink_page_list() has a temp ref on them, or because they're transiently 574 * sitting in the lru_cache_add() pagevecs. 575 */ 576 static int 577 invalidate_complete_page2(struct address_space *mapping, struct page *page) 578 { 579 unsigned long flags; 580 581 if (page->mapping != mapping) 582 return 0; 583 584 if (page_has_private(page) && !try_to_release_page(page, GFP_KERNEL)) 585 return 0; 586 587 xa_lock_irqsave(&mapping->i_pages, flags); 588 if (PageDirty(page)) 589 goto failed; 590 591 BUG_ON(page_has_private(page)); 592 __delete_from_page_cache(page, NULL); 593 xa_unlock_irqrestore(&mapping->i_pages, flags); 594 595 if (mapping->a_ops->freepage) 596 mapping->a_ops->freepage(page); 597 598 put_page(page); /* pagecache ref */ 599 return 1; 600 failed: 601 xa_unlock_irqrestore(&mapping->i_pages, flags); 602 return 0; 603 } 604 605 static int do_launder_page(struct address_space *mapping, struct page *page) 606 { 607 if (!PageDirty(page)) 608 return 0; 609 if (page->mapping != mapping || mapping->a_ops->launder_page == NULL) 610 return 0; 611 return mapping->a_ops->launder_page(page); 612 } 613 614 /** 615 * invalidate_inode_pages2_range - remove range of pages from an address_space 616 * @mapping: the address_space 617 * @start: the page offset 'from' which to invalidate 618 * @end: the page offset 'to' which to invalidate (inclusive) 619 * 620 * Any pages which are found to be mapped into pagetables are unmapped prior to 621 * invalidation. 622 * 623 * Return: -EBUSY if any pages could not be invalidated. 624 */ 625 int invalidate_inode_pages2_range(struct address_space *mapping, 626 pgoff_t start, pgoff_t end) 627 { 628 pgoff_t indices[PAGEVEC_SIZE]; 629 struct pagevec pvec; 630 pgoff_t index; 631 int i; 632 int ret = 0; 633 int ret2 = 0; 634 int did_range_unmap = 0; 635 636 if (mapping->nrpages == 0 && mapping->nrexceptional == 0) 637 goto out; 638 639 pagevec_init(&pvec); 640 index = start; 641 while (find_get_entries(mapping, index, end, &pvec, indices)) { 642 for (i = 0; i < pagevec_count(&pvec); i++) { 643 struct page *page = pvec.pages[i]; 644 645 /* We rely upon deletion not changing page->index */ 646 index = indices[i]; 647 648 if (xa_is_value(page)) { 649 if (!invalidate_exceptional_entry2(mapping, 650 index, page)) 651 ret = -EBUSY; 652 continue; 653 } 654 655 lock_page(page); 656 WARN_ON(page_to_index(page) != index); 657 if (page->mapping != mapping) { 658 unlock_page(page); 659 continue; 660 } 661 wait_on_page_writeback(page); 662 if (page_mapped(page)) { 663 if (!did_range_unmap) { 664 /* 665 * Zap the rest of the file in one hit. 666 */ 667 unmap_mapping_pages(mapping, index, 668 (1 + end - index), false); 669 did_range_unmap = 1; 670 } else { 671 /* 672 * Just zap this page 673 */ 674 unmap_mapping_pages(mapping, index, 675 1, false); 676 } 677 } 678 BUG_ON(page_mapped(page)); 679 ret2 = do_launder_page(mapping, page); 680 if (ret2 == 0) { 681 if (!invalidate_complete_page2(mapping, page)) 682 ret2 = -EBUSY; 683 } 684 if (ret2 < 0) 685 ret = ret2; 686 unlock_page(page); 687 } 688 pagevec_remove_exceptionals(&pvec); 689 pagevec_release(&pvec); 690 cond_resched(); 691 index++; 692 } 693 /* 694 * For DAX we invalidate page tables after invalidating page cache. We 695 * could invalidate page tables while invalidating each entry however 696 * that would be expensive. And doing range unmapping before doesn't 697 * work as we have no cheap way to find whether page cache entry didn't 698 * get remapped later. 699 */ 700 if (dax_mapping(mapping)) { 701 unmap_mapping_pages(mapping, start, end - start + 1, false); 702 } 703 out: 704 cleancache_invalidate_inode(mapping); 705 return ret; 706 } 707 EXPORT_SYMBOL_GPL(invalidate_inode_pages2_range); 708 709 /** 710 * invalidate_inode_pages2 - remove all pages from an address_space 711 * @mapping: the address_space 712 * 713 * Any pages which are found to be mapped into pagetables are unmapped prior to 714 * invalidation. 715 * 716 * Return: -EBUSY if any pages could not be invalidated. 717 */ 718 int invalidate_inode_pages2(struct address_space *mapping) 719 { 720 return invalidate_inode_pages2_range(mapping, 0, -1); 721 } 722 EXPORT_SYMBOL_GPL(invalidate_inode_pages2); 723 724 /** 725 * truncate_pagecache - unmap and remove pagecache that has been truncated 726 * @inode: inode 727 * @newsize: new file size 728 * 729 * inode's new i_size must already be written before truncate_pagecache 730 * is called. 731 * 732 * This function should typically be called before the filesystem 733 * releases resources associated with the freed range (eg. deallocates 734 * blocks). This way, pagecache will always stay logically coherent 735 * with on-disk format, and the filesystem would not have to deal with 736 * situations such as writepage being called for a page that has already 737 * had its underlying blocks deallocated. 738 */ 739 void truncate_pagecache(struct inode *inode, loff_t newsize) 740 { 741 struct address_space *mapping = inode->i_mapping; 742 loff_t holebegin = round_up(newsize, PAGE_SIZE); 743 744 /* 745 * unmap_mapping_range is called twice, first simply for 746 * efficiency so that truncate_inode_pages does fewer 747 * single-page unmaps. However after this first call, and 748 * before truncate_inode_pages finishes, it is possible for 749 * private pages to be COWed, which remain after 750 * truncate_inode_pages finishes, hence the second 751 * unmap_mapping_range call must be made for correctness. 752 */ 753 unmap_mapping_range(mapping, holebegin, 0, 1); 754 truncate_inode_pages(mapping, newsize); 755 unmap_mapping_range(mapping, holebegin, 0, 1); 756 } 757 EXPORT_SYMBOL(truncate_pagecache); 758 759 /** 760 * truncate_setsize - update inode and pagecache for a new file size 761 * @inode: inode 762 * @newsize: new file size 763 * 764 * truncate_setsize updates i_size and performs pagecache truncation (if 765 * necessary) to @newsize. It will be typically be called from the filesystem's 766 * setattr function when ATTR_SIZE is passed in. 767 * 768 * Must be called with a lock serializing truncates and writes (generally 769 * i_mutex but e.g. xfs uses a different lock) and before all filesystem 770 * specific block truncation has been performed. 771 */ 772 void truncate_setsize(struct inode *inode, loff_t newsize) 773 { 774 loff_t oldsize = inode->i_size; 775 776 i_size_write(inode, newsize); 777 if (newsize > oldsize) 778 pagecache_isize_extended(inode, oldsize, newsize); 779 truncate_pagecache(inode, newsize); 780 } 781 EXPORT_SYMBOL(truncate_setsize); 782 783 /** 784 * pagecache_isize_extended - update pagecache after extension of i_size 785 * @inode: inode for which i_size was extended 786 * @from: original inode size 787 * @to: new inode size 788 * 789 * Handle extension of inode size either caused by extending truncate or by 790 * write starting after current i_size. We mark the page straddling current 791 * i_size RO so that page_mkwrite() is called on the nearest write access to 792 * the page. This way filesystem can be sure that page_mkwrite() is called on 793 * the page before user writes to the page via mmap after the i_size has been 794 * changed. 795 * 796 * The function must be called after i_size is updated so that page fault 797 * coming after we unlock the page will already see the new i_size. 798 * The function must be called while we still hold i_mutex - this not only 799 * makes sure i_size is stable but also that userspace cannot observe new 800 * i_size value before we are prepared to store mmap writes at new inode size. 801 */ 802 void pagecache_isize_extended(struct inode *inode, loff_t from, loff_t to) 803 { 804 int bsize = i_blocksize(inode); 805 loff_t rounded_from; 806 struct page *page; 807 pgoff_t index; 808 809 WARN_ON(to > inode->i_size); 810 811 if (from >= to || bsize == PAGE_SIZE) 812 return; 813 /* Page straddling @from will not have any hole block created? */ 814 rounded_from = round_up(from, bsize); 815 if (to <= rounded_from || !(rounded_from & (PAGE_SIZE - 1))) 816 return; 817 818 index = from >> PAGE_SHIFT; 819 page = find_lock_page(inode->i_mapping, index); 820 /* Page not cached? Nothing to do */ 821 if (!page) 822 return; 823 /* 824 * See clear_page_dirty_for_io() for details why set_page_dirty() 825 * is needed. 826 */ 827 if (page_mkclean(page)) 828 set_page_dirty(page); 829 unlock_page(page); 830 put_page(page); 831 } 832 EXPORT_SYMBOL(pagecache_isize_extended); 833 834 /** 835 * truncate_pagecache_range - unmap and remove pagecache that is hole-punched 836 * @inode: inode 837 * @lstart: offset of beginning of hole 838 * @lend: offset of last byte of hole 839 * 840 * This function should typically be called before the filesystem 841 * releases resources associated with the freed range (eg. deallocates 842 * blocks). This way, pagecache will always stay logically coherent 843 * with on-disk format, and the filesystem would not have to deal with 844 * situations such as writepage being called for a page that has already 845 * had its underlying blocks deallocated. 846 */ 847 void truncate_pagecache_range(struct inode *inode, loff_t lstart, loff_t lend) 848 { 849 struct address_space *mapping = inode->i_mapping; 850 loff_t unmap_start = round_up(lstart, PAGE_SIZE); 851 loff_t unmap_end = round_down(1 + lend, PAGE_SIZE) - 1; 852 /* 853 * This rounding is currently just for example: unmap_mapping_range 854 * expands its hole outwards, whereas we want it to contract the hole 855 * inwards. However, existing callers of truncate_pagecache_range are 856 * doing their own page rounding first. Note that unmap_mapping_range 857 * allows holelen 0 for all, and we allow lend -1 for end of file. 858 */ 859 860 /* 861 * Unlike in truncate_pagecache, unmap_mapping_range is called only 862 * once (before truncating pagecache), and without "even_cows" flag: 863 * hole-punching should not remove private COWed pages from the hole. 864 */ 865 if ((u64)unmap_end > (u64)unmap_start) 866 unmap_mapping_range(mapping, unmap_start, 867 1 + unmap_end - unmap_start, 0); 868 truncate_inode_pages_range(mapping, lstart, lend); 869 } 870 EXPORT_SYMBOL(truncate_pagecache_range); 871