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