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 		index = indices[folio_batch_count(&fbatch) - 1] + 1;
367 		truncate_folio_batch_exceptionals(mapping, &fbatch, indices);
368 		for (i = 0; i < folio_batch_count(&fbatch); i++)
369 			truncate_cleanup_folio(fbatch.folios[i]);
370 		delete_from_page_cache_batch(mapping, &fbatch);
371 		for (i = 0; i < folio_batch_count(&fbatch); i++)
372 			folio_unlock(fbatch.folios[i]);
373 		folio_batch_release(&fbatch);
374 		cond_resched();
375 	}
376 
377 	same_folio = (lstart >> PAGE_SHIFT) == (lend >> PAGE_SHIFT);
378 	folio = __filemap_get_folio(mapping, lstart >> PAGE_SHIFT, FGP_LOCK, 0);
379 	if (folio) {
380 		same_folio = lend < folio_pos(folio) + folio_size(folio);
381 		if (!truncate_inode_partial_folio(folio, lstart, lend)) {
382 			start = folio->index + folio_nr_pages(folio);
383 			if (same_folio)
384 				end = folio->index;
385 		}
386 		folio_unlock(folio);
387 		folio_put(folio);
388 		folio = NULL;
389 	}
390 
391 	if (!same_folio)
392 		folio = __filemap_get_folio(mapping, lend >> PAGE_SHIFT,
393 						FGP_LOCK, 0);
394 	if (folio) {
395 		if (!truncate_inode_partial_folio(folio, lstart, lend))
396 			end = folio->index;
397 		folio_unlock(folio);
398 		folio_put(folio);
399 	}
400 
401 	index = start;
402 	while (index < end) {
403 		cond_resched();
404 		if (!find_get_entries(mapping, index, end - 1, &fbatch,
405 				indices)) {
406 			/* If all gone from start onwards, we're done */
407 			if (index == start)
408 				break;
409 			/* Otherwise restart to make sure all gone */
410 			index = start;
411 			continue;
412 		}
413 
414 		for (i = 0; i < folio_batch_count(&fbatch); i++) {
415 			struct folio *folio = fbatch.folios[i];
416 
417 			/* We rely upon deletion not changing page->index */
418 			index = indices[i];
419 
420 			if (xa_is_value(folio))
421 				continue;
422 
423 			folio_lock(folio);
424 			VM_BUG_ON_FOLIO(!folio_contains(folio, index), folio);
425 			folio_wait_writeback(folio);
426 			truncate_inode_folio(mapping, folio);
427 			folio_unlock(folio);
428 			index = folio_index(folio) + folio_nr_pages(folio) - 1;
429 		}
430 		truncate_folio_batch_exceptionals(mapping, &fbatch, indices);
431 		folio_batch_release(&fbatch);
432 		index++;
433 	}
434 }
435 EXPORT_SYMBOL(truncate_inode_pages_range);
436 
437 /**
438  * truncate_inode_pages - truncate *all* the pages from an offset
439  * @mapping: mapping to truncate
440  * @lstart: offset from which to truncate
441  *
442  * Called under (and serialised by) inode->i_rwsem and
443  * mapping->invalidate_lock.
444  *
445  * Note: When this function returns, there can be a page in the process of
446  * deletion (inside __filemap_remove_folio()) in the specified range.  Thus
447  * mapping->nrpages can be non-zero when this function returns even after
448  * truncation of the whole mapping.
449  */
truncate_inode_pages(struct address_space * mapping,loff_t lstart)450 void truncate_inode_pages(struct address_space *mapping, loff_t lstart)
451 {
452 	truncate_inode_pages_range(mapping, lstart, (loff_t)-1);
453 }
454 EXPORT_SYMBOL(truncate_inode_pages);
455 
456 /**
457  * truncate_inode_pages_final - truncate *all* pages before inode dies
458  * @mapping: mapping to truncate
459  *
460  * Called under (and serialized by) inode->i_rwsem.
461  *
462  * Filesystems have to use this in the .evict_inode path to inform the
463  * VM that this is the final truncate and the inode is going away.
464  */
truncate_inode_pages_final(struct address_space * mapping)465 void truncate_inode_pages_final(struct address_space *mapping)
466 {
467 	/*
468 	 * Page reclaim can not participate in regular inode lifetime
469 	 * management (can't call iput()) and thus can race with the
470 	 * inode teardown.  Tell it when the address space is exiting,
471 	 * so that it does not install eviction information after the
472 	 * final truncate has begun.
473 	 */
474 	mapping_set_exiting(mapping);
475 
476 	if (!mapping_empty(mapping)) {
477 		/*
478 		 * As truncation uses a lockless tree lookup, cycle
479 		 * the tree lock to make sure any ongoing tree
480 		 * modification that does not see AS_EXITING is
481 		 * completed before starting the final truncate.
482 		 */
483 		xa_lock_irq(&mapping->i_pages);
484 		xa_unlock_irq(&mapping->i_pages);
485 	}
486 
487 	truncate_inode_pages(mapping, 0);
488 }
489 EXPORT_SYMBOL(truncate_inode_pages_final);
490 
491 /**
492  * invalidate_mapping_pagevec - Invalidate all the unlocked pages of one inode
493  * @mapping: the address_space which holds the pages to invalidate
494  * @start: the offset 'from' which to invalidate
495  * @end: the offset 'to' which to invalidate (inclusive)
496  * @nr_pagevec: invalidate failed page number for caller
497  *
498  * This helper is similar to invalidate_mapping_pages(), except that it accounts
499  * for pages that are likely on a pagevec and counts them in @nr_pagevec, which
500  * will be used by the caller.
501  */
invalidate_mapping_pagevec(struct address_space * mapping,pgoff_t start,pgoff_t end,unsigned long * nr_pagevec)502 unsigned long invalidate_mapping_pagevec(struct address_space *mapping,
503 		pgoff_t start, pgoff_t end, unsigned long *nr_pagevec)
504 {
505 	pgoff_t indices[PAGEVEC_SIZE];
506 	struct folio_batch fbatch;
507 	pgoff_t index = start;
508 	unsigned long ret;
509 	unsigned long count = 0;
510 	int i;
511 
512 	folio_batch_init(&fbatch);
513 	while (find_lock_entries(mapping, index, end, &fbatch, indices)) {
514 		for (i = 0; i < folio_batch_count(&fbatch); i++) {
515 			struct folio *folio = fbatch.folios[i];
516 
517 			/* We rely upon deletion not changing folio->index */
518 			index = indices[i];
519 
520 			if (xa_is_value(folio)) {
521 				count += invalidate_exceptional_entry(mapping,
522 								      index,
523 								      folio);
524 				continue;
525 			}
526 			index += folio_nr_pages(folio) - 1;
527 
528 			ret = mapping_evict_folio(mapping, folio);
529 			folio_unlock(folio);
530 			/*
531 			 * Invalidation is a hint that the folio is no longer
532 			 * of interest and try to speed up its reclaim.
533 			 */
534 			if (!ret) {
535 				deactivate_file_folio(folio);
536 				/* It is likely on the pagevec of a remote CPU */
537 				if (nr_pagevec)
538 					(*nr_pagevec)++;
539 			}
540 			count += ret;
541 		}
542 		folio_batch_remove_exceptionals(&fbatch);
543 		folio_batch_release(&fbatch);
544 		cond_resched();
545 		index++;
546 	}
547 	return count;
548 }
549 
550 /**
551  * invalidate_mapping_pages - Invalidate all clean, unlocked cache of one inode
552  * @mapping: the address_space which holds the cache to invalidate
553  * @start: the offset 'from' which to invalidate
554  * @end: the offset 'to' which to invalidate (inclusive)
555  *
556  * This function removes pages that are clean, unmapped and unlocked,
557  * as well as shadow entries. It will not block on IO activity.
558  *
559  * If you want to remove all the pages of one inode, regardless of
560  * their use and writeback state, use truncate_inode_pages().
561  *
562  * Return: the number of the cache entries that were invalidated
563  */
invalidate_mapping_pages(struct address_space * mapping,pgoff_t start,pgoff_t end)564 unsigned long invalidate_mapping_pages(struct address_space *mapping,
565 		pgoff_t start, pgoff_t end)
566 {
567 	return invalidate_mapping_pagevec(mapping, start, end, NULL);
568 }
569 EXPORT_SYMBOL(invalidate_mapping_pages);
570 
571 /*
572  * This is like invalidate_inode_page(), except it ignores the page's
573  * refcount.  We do this because invalidate_inode_pages2() needs stronger
574  * invalidation guarantees, and cannot afford to leave pages behind because
575  * shrink_page_list() has a temp ref on them, or because they're transiently
576  * sitting in the lru_cache_add() pagevecs.
577  */
invalidate_complete_folio2(struct address_space * mapping,struct folio * folio)578 static int invalidate_complete_folio2(struct address_space *mapping,
579 					struct folio *folio)
580 {
581 	if (folio->mapping != mapping)
582 		return 0;
583 
584 	if (folio_has_private(folio) &&
585 	    !filemap_release_folio(folio, GFP_KERNEL))
586 		return 0;
587 
588 	spin_lock(&mapping->host->i_lock);
589 	xa_lock_irq(&mapping->i_pages);
590 	if (folio_test_dirty(folio))
591 		goto failed;
592 
593 	BUG_ON(folio_has_private(folio));
594 	__filemap_remove_folio(folio, NULL);
595 	xa_unlock_irq(&mapping->i_pages);
596 	if (mapping_shrinkable(mapping))
597 		inode_add_lru(mapping->host);
598 	spin_unlock(&mapping->host->i_lock);
599 
600 	filemap_free_folio(mapping, folio);
601 	return 1;
602 failed:
603 	xa_unlock_irq(&mapping->i_pages);
604 	spin_unlock(&mapping->host->i_lock);
605 	return 0;
606 }
607 
folio_launder(struct address_space * mapping,struct folio * folio)608 static int folio_launder(struct address_space *mapping, struct folio *folio)
609 {
610 	if (!folio_test_dirty(folio))
611 		return 0;
612 	if (folio->mapping != mapping || mapping->a_ops->launder_folio == NULL)
613 		return 0;
614 	return mapping->a_ops->launder_folio(folio);
615 }
616 
617 /**
618  * invalidate_inode_pages2_range - remove range of pages from an address_space
619  * @mapping: the address_space
620  * @start: the page offset 'from' which to invalidate
621  * @end: the page offset 'to' which to invalidate (inclusive)
622  *
623  * Any pages which are found to be mapped into pagetables are unmapped prior to
624  * invalidation.
625  *
626  * Return: -EBUSY if any pages could not be invalidated.
627  */
invalidate_inode_pages2_range(struct address_space * mapping,pgoff_t start,pgoff_t end)628 int invalidate_inode_pages2_range(struct address_space *mapping,
629 				  pgoff_t start, pgoff_t end)
630 {
631 	pgoff_t indices[PAGEVEC_SIZE];
632 	struct folio_batch fbatch;
633 	pgoff_t index;
634 	int i;
635 	int ret = 0;
636 	int ret2 = 0;
637 	int did_range_unmap = 0;
638 
639 	if (mapping_empty(mapping))
640 		return 0;
641 
642 	folio_batch_init(&fbatch);
643 	index = start;
644 	while (find_get_entries(mapping, index, end, &fbatch, indices)) {
645 		for (i = 0; i < folio_batch_count(&fbatch); i++) {
646 			struct folio *folio = fbatch.folios[i];
647 
648 			/* We rely upon deletion not changing folio->index */
649 			index = indices[i];
650 
651 			if (xa_is_value(folio)) {
652 				if (!invalidate_exceptional_entry2(mapping,
653 						index, folio))
654 					ret = -EBUSY;
655 				continue;
656 			}
657 
658 			if (!did_range_unmap && folio_mapped(folio)) {
659 				/*
660 				 * If folio is mapped, before taking its lock,
661 				 * zap the rest of the file in one hit.
662 				 */
663 				unmap_mapping_pages(mapping, index,
664 						(1 + end - index), false);
665 				did_range_unmap = 1;
666 			}
667 
668 			folio_lock(folio);
669 			VM_BUG_ON_FOLIO(!folio_contains(folio, index), folio);
670 			if (folio->mapping != mapping) {
671 				folio_unlock(folio);
672 				continue;
673 			}
674 			folio_wait_writeback(folio);
675 
676 			if (folio_mapped(folio))
677 				unmap_mapping_folio(folio);
678 			BUG_ON(folio_mapped(folio));
679 
680 			ret2 = folio_launder(mapping, folio);
681 			if (ret2 == 0) {
682 				if (!invalidate_complete_folio2(mapping, folio))
683 					ret2 = -EBUSY;
684 			}
685 			if (ret2 < 0)
686 				ret = ret2;
687 			folio_unlock(folio);
688 		}
689 		folio_batch_remove_exceptionals(&fbatch);
690 		folio_batch_release(&fbatch);
691 		cond_resched();
692 		index++;
693 	}
694 	/*
695 	 * For DAX we invalidate page tables after invalidating page cache.  We
696 	 * could invalidate page tables while invalidating each entry however
697 	 * that would be expensive. And doing range unmapping before doesn't
698 	 * work as we have no cheap way to find whether page cache entry didn't
699 	 * get remapped later.
700 	 */
701 	if (dax_mapping(mapping)) {
702 		unmap_mapping_pages(mapping, start, end - start + 1, false);
703 	}
704 	return ret;
705 }
706 EXPORT_SYMBOL_GPL(invalidate_inode_pages2_range);
707 
708 /**
709  * invalidate_inode_pages2 - remove all pages from an address_space
710  * @mapping: the address_space
711  *
712  * Any pages which are found to be mapped into pagetables are unmapped prior to
713  * invalidation.
714  *
715  * Return: -EBUSY if any pages could not be invalidated.
716  */
invalidate_inode_pages2(struct address_space * mapping)717 int invalidate_inode_pages2(struct address_space *mapping)
718 {
719 	return invalidate_inode_pages2_range(mapping, 0, -1);
720 }
721 EXPORT_SYMBOL_GPL(invalidate_inode_pages2);
722 
723 /**
724  * truncate_pagecache - unmap and remove pagecache that has been truncated
725  * @inode: inode
726  * @newsize: new file size
727  *
728  * inode's new i_size must already be written before truncate_pagecache
729  * is called.
730  *
731  * This function should typically be called before the filesystem
732  * releases resources associated with the freed range (eg. deallocates
733  * blocks). This way, pagecache will always stay logically coherent
734  * with on-disk format, and the filesystem would not have to deal with
735  * situations such as writepage being called for a page that has already
736  * had its underlying blocks deallocated.
737  */
truncate_pagecache(struct inode * inode,loff_t newsize)738 void truncate_pagecache(struct inode *inode, loff_t newsize)
739 {
740 	struct address_space *mapping = inode->i_mapping;
741 	loff_t holebegin = round_up(newsize, PAGE_SIZE);
742 
743 	/*
744 	 * unmap_mapping_range is called twice, first simply for
745 	 * efficiency so that truncate_inode_pages does fewer
746 	 * single-page unmaps.  However after this first call, and
747 	 * before truncate_inode_pages finishes, it is possible for
748 	 * private pages to be COWed, which remain after
749 	 * truncate_inode_pages finishes, hence the second
750 	 * unmap_mapping_range call must be made for correctness.
751 	 */
752 	unmap_mapping_range(mapping, holebegin, 0, 1);
753 	truncate_inode_pages(mapping, newsize);
754 	unmap_mapping_range(mapping, holebegin, 0, 1);
755 }
756 EXPORT_SYMBOL(truncate_pagecache);
757 
758 /**
759  * truncate_setsize - update inode and pagecache for a new file size
760  * @inode: inode
761  * @newsize: new file size
762  *
763  * truncate_setsize updates i_size and performs pagecache truncation (if
764  * necessary) to @newsize. It will be typically be called from the filesystem's
765  * setattr function when ATTR_SIZE is passed in.
766  *
767  * Must be called with a lock serializing truncates and writes (generally
768  * i_rwsem but e.g. xfs uses a different lock) and before all filesystem
769  * specific block truncation has been performed.
770  */
truncate_setsize(struct inode * inode,loff_t newsize)771 void truncate_setsize(struct inode *inode, loff_t newsize)
772 {
773 	loff_t oldsize = inode->i_size;
774 
775 	i_size_write(inode, newsize);
776 	if (newsize > oldsize)
777 		pagecache_isize_extended(inode, oldsize, newsize);
778 	truncate_pagecache(inode, newsize);
779 }
780 EXPORT_SYMBOL(truncate_setsize);
781 
782 /**
783  * pagecache_isize_extended - update pagecache after extension of i_size
784  * @inode:	inode for which i_size was extended
785  * @from:	original inode size
786  * @to:		new inode size
787  *
788  * Handle extension of inode size either caused by extending truncate or by
789  * write starting after current i_size. We mark the page straddling current
790  * i_size RO so that page_mkwrite() is called on the nearest write access to
791  * the page.  This way filesystem can be sure that page_mkwrite() is called on
792  * the page before user writes to the page via mmap after the i_size has been
793  * changed.
794  *
795  * The function must be called after i_size is updated so that page fault
796  * coming after we unlock the page will already see the new i_size.
797  * The function must be called while we still hold i_rwsem - this not only
798  * makes sure i_size is stable but also that userspace cannot observe new
799  * i_size value before we are prepared to store mmap writes at new inode size.
800  */
pagecache_isize_extended(struct inode * inode,loff_t from,loff_t to)801 void pagecache_isize_extended(struct inode *inode, loff_t from, loff_t to)
802 {
803 	int bsize = i_blocksize(inode);
804 	loff_t rounded_from;
805 	struct page *page;
806 	pgoff_t index;
807 
808 	WARN_ON(to > inode->i_size);
809 
810 	if (from >= to || bsize == PAGE_SIZE)
811 		return;
812 	/* Page straddling @from will not have any hole block created? */
813 	rounded_from = round_up(from, bsize);
814 	if (to <= rounded_from || !(rounded_from & (PAGE_SIZE - 1)))
815 		return;
816 
817 	index = from >> PAGE_SHIFT;
818 	page = find_lock_page(inode->i_mapping, index);
819 	/* Page not cached? Nothing to do */
820 	if (!page)
821 		return;
822 	/*
823 	 * See clear_page_dirty_for_io() for details why set_page_dirty()
824 	 * is needed.
825 	 */
826 	if (page_mkclean(page))
827 		set_page_dirty(page);
828 	unlock_page(page);
829 	put_page(page);
830 }
831 EXPORT_SYMBOL(pagecache_isize_extended);
832 
833 /**
834  * truncate_pagecache_range - unmap and remove pagecache that is hole-punched
835  * @inode: inode
836  * @lstart: offset of beginning of hole
837  * @lend: offset of last byte of hole
838  *
839  * This function should typically be called before the filesystem
840  * releases resources associated with the freed range (eg. deallocates
841  * blocks). This way, pagecache will always stay logically coherent
842  * with on-disk format, and the filesystem would not have to deal with
843  * situations such as writepage being called for a page that has already
844  * had its underlying blocks deallocated.
845  */
truncate_pagecache_range(struct inode * inode,loff_t lstart,loff_t lend)846 void truncate_pagecache_range(struct inode *inode, loff_t lstart, loff_t lend)
847 {
848 	struct address_space *mapping = inode->i_mapping;
849 	loff_t unmap_start = round_up(lstart, PAGE_SIZE);
850 	loff_t unmap_end = round_down(1 + lend, PAGE_SIZE) - 1;
851 	/*
852 	 * This rounding is currently just for example: unmap_mapping_range
853 	 * expands its hole outwards, whereas we want it to contract the hole
854 	 * inwards.  However, existing callers of truncate_pagecache_range are
855 	 * doing their own page rounding first.  Note that unmap_mapping_range
856 	 * allows holelen 0 for all, and we allow lend -1 for end of file.
857 	 */
858 
859 	/*
860 	 * Unlike in truncate_pagecache, unmap_mapping_range is called only
861 	 * once (before truncating pagecache), and without "even_cows" flag:
862 	 * hole-punching should not remove private COWed pages from the hole.
863 	 */
864 	if ((u64)unmap_end > (u64)unmap_start)
865 		unmap_mapping_range(mapping, unmap_start,
866 				    1 + unmap_end - unmap_start, 0);
867 	truncate_inode_pages_range(mapping, lstart, lend);
868 }
869 EXPORT_SYMBOL(truncate_pagecache_range);
870