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