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