• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
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  */
__clear_shadow_entry(struct address_space * mapping,pgoff_t index,void * entry)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 
clear_shadow_entry(struct address_space * mapping,pgoff_t index,void * entry)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  */
truncate_exceptional_pvec_entries(struct address_space * mapping,struct pagevec * pvec,pgoff_t * indices)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  */
invalidate_exceptional_entry(struct address_space * mapping,pgoff_t index,void * entry)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  */
invalidate_exceptional_entry2(struct address_space * mapping,pgoff_t index,void * entry)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  */
do_invalidatepage(struct page * page,unsigned int offset,unsigned int length)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  */
truncate_cleanup_page(struct page * page)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
invalidate_complete_page(struct address_space * mapping,struct page * page)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 
truncate_inode_page(struct address_space * mapping,struct page * page)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  */
generic_error_remove_page(struct address_space * mapping,struct page * page)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  */
invalidate_inode_page(struct page * page)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  */
truncate_inode_pages_range(struct address_space * mapping,loff_t lstart,loff_t lend)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_rwsem and
416  * mapping->invalidate_lock.
417  *
418  * Note: When this function returns, there can be a page in the process of
419  * deletion (inside __delete_from_page_cache()) in the specified range.  Thus
420  * mapping->nrpages can be non-zero when this function returns even after
421  * truncation of the whole mapping.
422  */
truncate_inode_pages(struct address_space * mapping,loff_t lstart)423 void truncate_inode_pages(struct address_space *mapping, loff_t lstart)
424 {
425 	truncate_inode_pages_range(mapping, lstart, (loff_t)-1);
426 }
427 EXPORT_SYMBOL(truncate_inode_pages);
428 
429 /**
430  * truncate_inode_pages_final - truncate *all* pages before inode dies
431  * @mapping: mapping to truncate
432  *
433  * Called under (and serialized by) inode->i_rwsem.
434  *
435  * Filesystems have to use this in the .evict_inode path to inform the
436  * VM that this is the final truncate and the inode is going away.
437  */
truncate_inode_pages_final(struct address_space * mapping)438 void truncate_inode_pages_final(struct address_space *mapping)
439 {
440 	/*
441 	 * Page reclaim can not participate in regular inode lifetime
442 	 * management (can't call iput()) and thus can race with the
443 	 * inode teardown.  Tell it when the address space is exiting,
444 	 * so that it does not install eviction information after the
445 	 * final truncate has begun.
446 	 */
447 	mapping_set_exiting(mapping);
448 
449 	if (!mapping_empty(mapping)) {
450 		/*
451 		 * As truncation uses a lockless tree lookup, cycle
452 		 * the tree lock to make sure any ongoing tree
453 		 * modification that does not see AS_EXITING is
454 		 * completed before starting the final truncate.
455 		 */
456 		xa_lock_irq(&mapping->i_pages);
457 		xa_unlock_irq(&mapping->i_pages);
458 	}
459 
460 	/*
461 	 * Cleancache needs notification even if there are no pages or shadow
462 	 * entries.
463 	 */
464 	truncate_inode_pages(mapping, 0);
465 }
466 EXPORT_SYMBOL(truncate_inode_pages_final);
467 
__invalidate_mapping_pages(struct address_space * mapping,pgoff_t start,pgoff_t end,unsigned long * nr_pagevec)468 static unsigned long __invalidate_mapping_pages(struct address_space *mapping,
469 		pgoff_t start, pgoff_t end, unsigned long *nr_pagevec)
470 {
471 	pgoff_t indices[PAGEVEC_SIZE];
472 	struct pagevec pvec;
473 	pgoff_t index = start;
474 	unsigned long ret;
475 	unsigned long count = 0;
476 	int i;
477 
478 	pagevec_init(&pvec);
479 	while (find_lock_entries(mapping, index, end, &pvec, indices)) {
480 		for (i = 0; i < pagevec_count(&pvec); i++) {
481 			struct page *page = pvec.pages[i];
482 
483 			/* We rely upon deletion not changing page->index */
484 			index = indices[i];
485 
486 			if (xa_is_value(page)) {
487 				count += invalidate_exceptional_entry(mapping,
488 								      index,
489 								      page);
490 				continue;
491 			}
492 			index += thp_nr_pages(page) - 1;
493 
494 			ret = invalidate_inode_page(page);
495 			unlock_page(page);
496 			/*
497 			 * Invalidation is a hint that the page is no longer
498 			 * of interest and try to speed up its reclaim.
499 			 */
500 			if (!ret) {
501 				deactivate_file_page(page);
502 				/* It is likely on the pagevec of a remote CPU */
503 				if (nr_pagevec)
504 					(*nr_pagevec)++;
505 			}
506 			count += ret;
507 		}
508 		pagevec_remove_exceptionals(&pvec);
509 		pagevec_release(&pvec);
510 		cond_resched();
511 		index++;
512 	}
513 	return count;
514 }
515 
516 /**
517  * invalidate_mapping_pages - Invalidate all clean, unlocked cache of one inode
518  * @mapping: the address_space which holds the cache to invalidate
519  * @start: the offset 'from' which to invalidate
520  * @end: the offset 'to' which to invalidate (inclusive)
521  *
522  * This function removes pages that are clean, unmapped and unlocked,
523  * as well as shadow entries. It will not block on IO activity.
524  *
525  * If you want to remove all the pages of one inode, regardless of
526  * their use and writeback state, use truncate_inode_pages().
527  *
528  * Return: the number of the cache entries that were invalidated
529  */
invalidate_mapping_pages(struct address_space * mapping,pgoff_t start,pgoff_t end)530 unsigned long invalidate_mapping_pages(struct address_space *mapping,
531 		pgoff_t start, pgoff_t end)
532 {
533 	return __invalidate_mapping_pages(mapping, start, end, NULL);
534 }
535 EXPORT_SYMBOL(invalidate_mapping_pages);
536 
537 /**
538  * invalidate_mapping_pagevec - Invalidate all the unlocked pages of one inode
539  * @mapping: the address_space which holds the pages to invalidate
540  * @start: the offset 'from' which to invalidate
541  * @end: the offset 'to' which to invalidate (inclusive)
542  * @nr_pagevec: invalidate failed page number for caller
543  *
544  * This helper is similar to invalidate_mapping_pages(), except that it accounts
545  * for pages that are likely on a pagevec and counts them in @nr_pagevec, which
546  * will be used by the caller.
547  */
invalidate_mapping_pagevec(struct address_space * mapping,pgoff_t start,pgoff_t end,unsigned long * nr_pagevec)548 void invalidate_mapping_pagevec(struct address_space *mapping,
549 		pgoff_t start, pgoff_t end, unsigned long *nr_pagevec)
550 {
551 	__invalidate_mapping_pages(mapping, start, end, nr_pagevec);
552 }
553 
554 /*
555  * This is like invalidate_complete_page(), except it ignores the page's
556  * refcount.  We do this because invalidate_inode_pages2() needs stronger
557  * invalidation guarantees, and cannot afford to leave pages behind because
558  * shrink_page_list() has a temp ref on them, or because they're transiently
559  * sitting in the lru_cache_add() pagevecs.
560  */
561 static int
invalidate_complete_page2(struct address_space * mapping,struct page * page)562 invalidate_complete_page2(struct address_space *mapping, struct page *page)
563 {
564 	if (page->mapping != mapping)
565 		return 0;
566 
567 	if (page_has_private(page) && !try_to_release_page(page, GFP_KERNEL))
568 		return 0;
569 
570 	xa_lock_irq(&mapping->i_pages);
571 	if (PageDirty(page))
572 		goto failed;
573 
574 	BUG_ON(page_has_private(page));
575 	__delete_from_page_cache(page, NULL);
576 	xa_unlock_irq(&mapping->i_pages);
577 
578 	if (mapping->a_ops->freepage)
579 		mapping->a_ops->freepage(page);
580 
581 	put_page(page);	/* pagecache ref */
582 	return 1;
583 failed:
584 	xa_unlock_irq(&mapping->i_pages);
585 	return 0;
586 }
587 
do_launder_page(struct address_space * mapping,struct page * page)588 static int do_launder_page(struct address_space *mapping, struct page *page)
589 {
590 	if (!PageDirty(page))
591 		return 0;
592 	if (page->mapping != mapping || mapping->a_ops->launder_page == NULL)
593 		return 0;
594 	return mapping->a_ops->launder_page(page);
595 }
596 
597 /**
598  * invalidate_inode_pages2_range - remove range of pages from an address_space
599  * @mapping: the address_space
600  * @start: the page offset 'from' which to invalidate
601  * @end: the page offset 'to' which to invalidate (inclusive)
602  *
603  * Any pages which are found to be mapped into pagetables are unmapped prior to
604  * invalidation.
605  *
606  * Return: -EBUSY if any pages could not be invalidated.
607  */
invalidate_inode_pages2_range(struct address_space * mapping,pgoff_t start,pgoff_t end)608 int invalidate_inode_pages2_range(struct address_space *mapping,
609 				  pgoff_t start, pgoff_t end)
610 {
611 	pgoff_t indices[PAGEVEC_SIZE];
612 	struct pagevec pvec;
613 	pgoff_t index;
614 	int i;
615 	int ret = 0;
616 	int ret2 = 0;
617 	int did_range_unmap = 0;
618 
619 	if (mapping_empty(mapping))
620 		goto out;
621 
622 	pagevec_init(&pvec);
623 	index = start;
624 	while (find_get_entries(mapping, index, end, &pvec, indices)) {
625 		for (i = 0; i < pagevec_count(&pvec); i++) {
626 			struct page *page = pvec.pages[i];
627 
628 			/* We rely upon deletion not changing page->index */
629 			index = indices[i];
630 
631 			if (xa_is_value(page)) {
632 				if (!invalidate_exceptional_entry2(mapping,
633 								   index, page))
634 					ret = -EBUSY;
635 				continue;
636 			}
637 
638 			if (!did_range_unmap && page_mapped(page)) {
639 				/*
640 				 * If page is mapped, before taking its lock,
641 				 * zap the rest of the file in one hit.
642 				 */
643 				unmap_mapping_pages(mapping, index,
644 						(1 + end - index), false);
645 				did_range_unmap = 1;
646 			}
647 
648 			lock_page(page);
649 			WARN_ON(page_to_index(page) != index);
650 			if (page->mapping != mapping) {
651 				unlock_page(page);
652 				continue;
653 			}
654 			wait_on_page_writeback(page);
655 
656 			if (page_mapped(page))
657 				unmap_mapping_page(page);
658 			BUG_ON(page_mapped(page));
659 
660 			ret2 = do_launder_page(mapping, page);
661 			if (ret2 == 0) {
662 				if (!invalidate_complete_page2(mapping, page))
663 					ret2 = -EBUSY;
664 			}
665 			if (ret2 < 0)
666 				ret = ret2;
667 			unlock_page(page);
668 		}
669 		pagevec_remove_exceptionals(&pvec);
670 		pagevec_release(&pvec);
671 		cond_resched();
672 		index++;
673 	}
674 	/*
675 	 * For DAX we invalidate page tables after invalidating page cache.  We
676 	 * could invalidate page tables while invalidating each entry however
677 	 * that would be expensive. And doing range unmapping before doesn't
678 	 * work as we have no cheap way to find whether page cache entry didn't
679 	 * get remapped later.
680 	 */
681 	if (dax_mapping(mapping)) {
682 		unmap_mapping_pages(mapping, start, end - start + 1, false);
683 	}
684 out:
685 	cleancache_invalidate_inode(mapping);
686 	return ret;
687 }
688 EXPORT_SYMBOL_GPL(invalidate_inode_pages2_range);
689 
690 /**
691  * invalidate_inode_pages2 - remove all pages from an address_space
692  * @mapping: the address_space
693  *
694  * Any pages which are found to be mapped into pagetables are unmapped prior to
695  * invalidation.
696  *
697  * Return: -EBUSY if any pages could not be invalidated.
698  */
invalidate_inode_pages2(struct address_space * mapping)699 int invalidate_inode_pages2(struct address_space *mapping)
700 {
701 	return invalidate_inode_pages2_range(mapping, 0, -1);
702 }
703 EXPORT_SYMBOL_GPL(invalidate_inode_pages2);
704 
705 /**
706  * truncate_pagecache - unmap and remove pagecache that has been truncated
707  * @inode: inode
708  * @newsize: new file size
709  *
710  * inode's new i_size must already be written before truncate_pagecache
711  * is called.
712  *
713  * This function should typically be called before the filesystem
714  * releases resources associated with the freed range (eg. deallocates
715  * blocks). This way, pagecache will always stay logically coherent
716  * with on-disk format, and the filesystem would not have to deal with
717  * situations such as writepage being called for a page that has already
718  * had its underlying blocks deallocated.
719  */
truncate_pagecache(struct inode * inode,loff_t newsize)720 void truncate_pagecache(struct inode *inode, loff_t newsize)
721 {
722 	struct address_space *mapping = inode->i_mapping;
723 	loff_t holebegin = round_up(newsize, PAGE_SIZE);
724 
725 	/*
726 	 * unmap_mapping_range is called twice, first simply for
727 	 * efficiency so that truncate_inode_pages does fewer
728 	 * single-page unmaps.  However after this first call, and
729 	 * before truncate_inode_pages finishes, it is possible for
730 	 * private pages to be COWed, which remain after
731 	 * truncate_inode_pages finishes, hence the second
732 	 * unmap_mapping_range call must be made for correctness.
733 	 */
734 	unmap_mapping_range(mapping, holebegin, 0, 1);
735 	truncate_inode_pages(mapping, newsize);
736 	unmap_mapping_range(mapping, holebegin, 0, 1);
737 }
738 EXPORT_SYMBOL(truncate_pagecache);
739 
740 /**
741  * truncate_setsize - update inode and pagecache for a new file size
742  * @inode: inode
743  * @newsize: new file size
744  *
745  * truncate_setsize updates i_size and performs pagecache truncation (if
746  * necessary) to @newsize. It will be typically be called from the filesystem's
747  * setattr function when ATTR_SIZE is passed in.
748  *
749  * Must be called with a lock serializing truncates and writes (generally
750  * i_rwsem but e.g. xfs uses a different lock) and before all filesystem
751  * specific block truncation has been performed.
752  */
truncate_setsize(struct inode * inode,loff_t newsize)753 void truncate_setsize(struct inode *inode, loff_t newsize)
754 {
755 	loff_t oldsize = inode->i_size;
756 
757 	i_size_write(inode, newsize);
758 	if (newsize > oldsize)
759 		pagecache_isize_extended(inode, oldsize, newsize);
760 	truncate_pagecache(inode, newsize);
761 }
762 EXPORT_SYMBOL(truncate_setsize);
763 
764 /**
765  * pagecache_isize_extended - update pagecache after extension of i_size
766  * @inode:	inode for which i_size was extended
767  * @from:	original inode size
768  * @to:		new inode size
769  *
770  * Handle extension of inode size either caused by extending truncate or by
771  * write starting after current i_size. We mark the page straddling current
772  * i_size RO so that page_mkwrite() is called on the nearest write access to
773  * the page.  This way filesystem can be sure that page_mkwrite() is called on
774  * the page before user writes to the page via mmap after the i_size has been
775  * changed.
776  *
777  * The function must be called after i_size is updated so that page fault
778  * coming after we unlock the page will already see the new i_size.
779  * The function must be called while we still hold i_rwsem - this not only
780  * makes sure i_size is stable but also that userspace cannot observe new
781  * i_size value before we are prepared to store mmap writes at new inode size.
782  */
pagecache_isize_extended(struct inode * inode,loff_t from,loff_t to)783 void pagecache_isize_extended(struct inode *inode, loff_t from, loff_t to)
784 {
785 	int bsize = i_blocksize(inode);
786 	loff_t rounded_from;
787 	struct page *page;
788 	pgoff_t index;
789 
790 	WARN_ON(to > inode->i_size);
791 
792 	if (from >= to || bsize == PAGE_SIZE)
793 		return;
794 	/* Page straddling @from will not have any hole block created? */
795 	rounded_from = round_up(from, bsize);
796 	if (to <= rounded_from || !(rounded_from & (PAGE_SIZE - 1)))
797 		return;
798 
799 	index = from >> PAGE_SHIFT;
800 	page = find_lock_page(inode->i_mapping, index);
801 	/* Page not cached? Nothing to do */
802 	if (!page)
803 		return;
804 	/*
805 	 * See clear_page_dirty_for_io() for details why set_page_dirty()
806 	 * is needed.
807 	 */
808 	if (page_mkclean(page))
809 		set_page_dirty(page);
810 	unlock_page(page);
811 	put_page(page);
812 }
813 EXPORT_SYMBOL(pagecache_isize_extended);
814 
815 /**
816  * truncate_pagecache_range - unmap and remove pagecache that is hole-punched
817  * @inode: inode
818  * @lstart: offset of beginning of hole
819  * @lend: offset of last byte of hole
820  *
821  * This function should typically be called before the filesystem
822  * releases resources associated with the freed range (eg. deallocates
823  * blocks). This way, pagecache will always stay logically coherent
824  * with on-disk format, and the filesystem would not have to deal with
825  * situations such as writepage being called for a page that has already
826  * had its underlying blocks deallocated.
827  */
truncate_pagecache_range(struct inode * inode,loff_t lstart,loff_t lend)828 void truncate_pagecache_range(struct inode *inode, loff_t lstart, loff_t lend)
829 {
830 	struct address_space *mapping = inode->i_mapping;
831 	loff_t unmap_start = round_up(lstart, PAGE_SIZE);
832 	loff_t unmap_end = round_down(1 + lend, PAGE_SIZE) - 1;
833 	/*
834 	 * This rounding is currently just for example: unmap_mapping_range
835 	 * expands its hole outwards, whereas we want it to contract the hole
836 	 * inwards.  However, existing callers of truncate_pagecache_range are
837 	 * doing their own page rounding first.  Note that unmap_mapping_range
838 	 * allows holelen 0 for all, and we allow lend -1 for end of file.
839 	 */
840 
841 	/*
842 	 * Unlike in truncate_pagecache, unmap_mapping_range is called only
843 	 * once (before truncating pagecache), and without "even_cows" flag:
844 	 * hole-punching should not remove private COWed pages from the hole.
845 	 */
846 	if ((u64)unmap_end > (u64)unmap_start)
847 		unmap_mapping_range(mapping, unmap_start,
848 				    1 + unmap_end - unmap_start, 0);
849 	truncate_inode_pages_range(mapping, lstart, lend);
850 }
851 EXPORT_SYMBOL(truncate_pagecache_range);
852