Line data Source code
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 : */
32 0 : static inline void __clear_shadow_entry(struct address_space *mapping,
33 : pgoff_t index, void *entry)
34 : {
35 0 : XA_STATE(xas, &mapping->i_pages, index);
36 :
37 0 : xas_set_update(&xas, workingset_update_node);
38 0 : if (xas_load(&xas) != entry)
39 0 : return;
40 0 : xas_store(&xas, NULL);
41 : }
42 :
43 0 : static void clear_shadow_entry(struct address_space *mapping, pgoff_t index,
44 : void *entry)
45 : {
46 0 : spin_lock(&mapping->host->i_lock);
47 0 : xa_lock_irq(&mapping->i_pages);
48 0 : __clear_shadow_entry(mapping, index, entry);
49 0 : xa_unlock_irq(&mapping->i_pages);
50 0 : if (mapping_shrinkable(mapping))
51 0 : inode_add_lru(mapping->host);
52 0 : spin_unlock(&mapping->host->i_lock);
53 0 : }
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 : */
60 0 : 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 0 : if (shmem_mapping(mapping))
68 : return;
69 :
70 0 : for (j = 0; j < folio_batch_count(fbatch); j++)
71 0 : if (xa_is_value(fbatch->folios[j]))
72 : break;
73 :
74 0 : if (j == folio_batch_count(fbatch))
75 : return;
76 :
77 0 : dax = dax_mapping(mapping);
78 : if (!dax) {
79 0 : spin_lock(&mapping->host->i_lock);
80 0 : xa_lock_irq(&mapping->i_pages);
81 : }
82 :
83 0 : for (i = j; i < folio_batch_count(fbatch); i++) {
84 0 : struct folio *folio = fbatch->folios[i];
85 0 : pgoff_t index = indices[i];
86 :
87 0 : if (!xa_is_value(folio)) {
88 0 : fbatch->folios[j++] = folio;
89 0 : continue;
90 : }
91 :
92 : if (unlikely(dax)) {
93 : dax_delete_mapping_entry(mapping, index);
94 : continue;
95 : }
96 :
97 0 : __clear_shadow_entry(mapping, index, folio);
98 : }
99 :
100 : if (!dax) {
101 0 : xa_unlock_irq(&mapping->i_pages);
102 0 : if (mapping_shrinkable(mapping))
103 0 : inode_add_lru(mapping->host);
104 0 : spin_unlock(&mapping->host->i_lock);
105 : }
106 0 : fbatch->nr = j;
107 : }
108 :
109 : /*
110 : * Invalidate exceptional entry if easily possible. This handles exceptional
111 : * entries for invalidate_inode_pages().
112 : */
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 0 : if (shmem_mapping(mapping) || dax_mapping(mapping))
118 : return 1;
119 0 : 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 : */
127 : static int invalidate_exceptional_entry2(struct address_space *mapping,
128 : pgoff_t index, void *entry)
129 : {
130 : /* Handled by shmem itself */
131 0 : if (shmem_mapping(mapping))
132 : return 1;
133 0 : if (dax_mapping(mapping))
134 : return dax_invalidate_mapping_entry_sync(mapping, index);
135 0 : 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 : */
154 0 : void folio_invalidate(struct folio *folio, size_t offset, size_t length)
155 : {
156 0 : const struct address_space_operations *aops = folio->mapping->a_ops;
157 :
158 0 : if (aops->invalidate_folio)
159 0 : aops->invalidate_folio(folio, offset, length);
160 0 : }
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 : */
173 0 : static void truncate_cleanup_folio(struct folio *folio)
174 : {
175 0 : if (folio_mapped(folio))
176 0 : unmap_mapping_folio(folio);
177 :
178 0 : if (folio_has_private(folio))
179 0 : 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 0 : folio_cancel_dirty(folio);
187 0 : folio_clear_mappedtodisk(folio);
188 0 : }
189 :
190 0 : int truncate_inode_folio(struct address_space *mapping, struct folio *folio)
191 : {
192 0 : if (folio->mapping != mapping)
193 : return -EIO;
194 :
195 0 : truncate_cleanup_folio(folio);
196 0 : filemap_remove_folio(folio);
197 0 : 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 : */
211 0 : bool truncate_inode_partial_folio(struct folio *folio, loff_t start, loff_t end)
212 : {
213 0 : loff_t pos = folio_pos(folio);
214 : unsigned int offset, length;
215 :
216 0 : if (pos < start)
217 0 : offset = start - pos;
218 : else
219 : offset = 0;
220 0 : length = folio_size(folio);
221 0 : if (pos + length <= (u64)end)
222 0 : length = length - offset;
223 : else
224 0 : length = end + 1 - pos - offset;
225 :
226 0 : folio_wait_writeback(folio);
227 0 : if (length == folio_size(folio)) {
228 0 : truncate_inode_folio(folio->mapping, folio);
229 0 : 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 0 : folio_zero_range(folio, offset, length);
238 :
239 0 : if (folio_has_private(folio))
240 0 : folio_invalidate(folio, offset, length);
241 0 : 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 : */
254 0 : int generic_error_remove_page(struct address_space *mapping, struct page *page)
255 : {
256 : VM_BUG_ON_PAGE(PageTail(page), page);
257 :
258 0 : 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 0 : if (!S_ISREG(mapping->host->i_mode))
265 : return -EIO;
266 0 : return truncate_inode_folio(mapping, page_folio(page));
267 : }
268 : EXPORT_SYMBOL(generic_error_remove_page);
269 :
270 0 : static long mapping_evict_folio(struct address_space *mapping,
271 : struct folio *folio)
272 : {
273 0 : 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 0 : if (folio_ref_count(folio) >
277 0 : folio_nr_pages(folio) + folio_has_private(folio) + 1)
278 : return 0;
279 0 : if (folio_has_private(folio) && !filemap_release_folio(folio, 0))
280 : return 0;
281 :
282 0 : 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 : */
295 0 : long invalidate_inode_page(struct page *page)
296 : {
297 0 : struct folio *folio = page_folio(page);
298 0 : struct address_space *mapping = folio_mapping(folio);
299 :
300 : /* The page may have been truncated before it was locked */
301 0 : if (!mapping)
302 : return 0;
303 0 : 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 : */
330 10 : 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 10 : if (mapping_empty(mapping))
343 10 : 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 0 : start = (lstart + PAGE_SIZE - 1) >> PAGE_SHIFT;
352 0 : 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 0 : end = (lend + 1) >> PAGE_SHIFT;
361 :
362 0 : folio_batch_init(&fbatch);
363 0 : index = start;
364 0 : while (index < end && find_lock_entries(mapping, &index, end - 1,
365 : &fbatch, indices)) {
366 0 : truncate_folio_batch_exceptionals(mapping, &fbatch, indices);
367 0 : for (i = 0; i < folio_batch_count(&fbatch); i++)
368 0 : truncate_cleanup_folio(fbatch.folios[i]);
369 0 : delete_from_page_cache_batch(mapping, &fbatch);
370 0 : for (i = 0; i < folio_batch_count(&fbatch); i++)
371 0 : folio_unlock(fbatch.folios[i]);
372 0 : folio_batch_release(&fbatch);
373 0 : cond_resched();
374 : }
375 :
376 0 : same_folio = (lstart >> PAGE_SHIFT) == (lend >> PAGE_SHIFT);
377 0 : folio = __filemap_get_folio(mapping, lstart >> PAGE_SHIFT, FGP_LOCK, 0);
378 0 : if (!IS_ERR(folio)) {
379 0 : same_folio = lend < folio_pos(folio) + folio_size(folio);
380 0 : if (!truncate_inode_partial_folio(folio, lstart, lend)) {
381 0 : start = folio->index + folio_nr_pages(folio);
382 0 : if (same_folio)
383 0 : end = folio->index;
384 : }
385 0 : folio_unlock(folio);
386 : folio_put(folio);
387 : folio = NULL;
388 : }
389 :
390 0 : if (!same_folio) {
391 0 : folio = __filemap_get_folio(mapping, lend >> PAGE_SHIFT,
392 : FGP_LOCK, 0);
393 0 : if (!IS_ERR(folio)) {
394 0 : if (!truncate_inode_partial_folio(folio, lstart, lend))
395 0 : end = folio->index;
396 0 : folio_unlock(folio);
397 : folio_put(folio);
398 : }
399 : }
400 :
401 0 : index = start;
402 0 : while (index < end) {
403 0 : cond_resched();
404 0 : if (!find_get_entries(mapping, &index, end - 1, &fbatch,
405 : indices)) {
406 : /* If all gone from start onwards, we're done */
407 0 : if (index == start)
408 : break;
409 : /* Otherwise restart to make sure all gone */
410 0 : index = start;
411 0 : continue;
412 : }
413 :
414 0 : for (i = 0; i < folio_batch_count(&fbatch); i++) {
415 0 : struct folio *folio = fbatch.folios[i];
416 :
417 : /* We rely upon deletion not changing page->index */
418 :
419 0 : if (xa_is_value(folio))
420 0 : continue;
421 :
422 0 : folio_lock(folio);
423 : VM_BUG_ON_FOLIO(!folio_contains(folio, indices[i]), folio);
424 0 : folio_wait_writeback(folio);
425 0 : truncate_inode_folio(mapping, folio);
426 0 : folio_unlock(folio);
427 : }
428 0 : truncate_folio_batch_exceptionals(mapping, &fbatch, indices);
429 : folio_batch_release(&fbatch);
430 : }
431 : }
432 : EXPORT_SYMBOL(truncate_inode_pages_range);
433 :
434 : /**
435 : * truncate_inode_pages - truncate *all* the pages from an offset
436 : * @mapping: mapping to truncate
437 : * @lstart: offset from which to truncate
438 : *
439 : * Called under (and serialised by) inode->i_rwsem and
440 : * mapping->invalidate_lock.
441 : *
442 : * Note: When this function returns, there can be a page in the process of
443 : * deletion (inside __filemap_remove_folio()) in the specified range. Thus
444 : * mapping->nrpages can be non-zero when this function returns even after
445 : * truncation of the whole mapping.
446 : */
447 0 : void truncate_inode_pages(struct address_space *mapping, loff_t lstart)
448 : {
449 10 : truncate_inode_pages_range(mapping, lstart, (loff_t)-1);
450 0 : }
451 : EXPORT_SYMBOL(truncate_inode_pages);
452 :
453 : /**
454 : * truncate_inode_pages_final - truncate *all* pages before inode dies
455 : * @mapping: mapping to truncate
456 : *
457 : * Called under (and serialized by) inode->i_rwsem.
458 : *
459 : * Filesystems have to use this in the .evict_inode path to inform the
460 : * VM that this is the final truncate and the inode is going away.
461 : */
462 10 : void truncate_inode_pages_final(struct address_space *mapping)
463 : {
464 : /*
465 : * Page reclaim can not participate in regular inode lifetime
466 : * management (can't call iput()) and thus can race with the
467 : * inode teardown. Tell it when the address space is exiting,
468 : * so that it does not install eviction information after the
469 : * final truncate has begun.
470 : */
471 10 : mapping_set_exiting(mapping);
472 :
473 10 : if (!mapping_empty(mapping)) {
474 : /*
475 : * As truncation uses a lockless tree lookup, cycle
476 : * the tree lock to make sure any ongoing tree
477 : * modification that does not see AS_EXITING is
478 : * completed before starting the final truncate.
479 : */
480 0 : xa_lock_irq(&mapping->i_pages);
481 0 : xa_unlock_irq(&mapping->i_pages);
482 : }
483 :
484 10 : truncate_inode_pages(mapping, 0);
485 10 : }
486 : EXPORT_SYMBOL(truncate_inode_pages_final);
487 :
488 : /**
489 : * mapping_try_invalidate - Invalidate all the evictable folios of one inode
490 : * @mapping: the address_space which holds the folios to invalidate
491 : * @start: the offset 'from' which to invalidate
492 : * @end: the offset 'to' which to invalidate (inclusive)
493 : * @nr_failed: How many folio invalidations failed
494 : *
495 : * This function is similar to invalidate_mapping_pages(), except that it
496 : * returns the number of folios which could not be evicted in @nr_failed.
497 : */
498 0 : unsigned long mapping_try_invalidate(struct address_space *mapping,
499 : pgoff_t start, pgoff_t end, unsigned long *nr_failed)
500 : {
501 : pgoff_t indices[PAGEVEC_SIZE];
502 : struct folio_batch fbatch;
503 0 : pgoff_t index = start;
504 : unsigned long ret;
505 0 : unsigned long count = 0;
506 : int i;
507 :
508 0 : folio_batch_init(&fbatch);
509 0 : while (find_lock_entries(mapping, &index, end, &fbatch, indices)) {
510 0 : for (i = 0; i < folio_batch_count(&fbatch); i++) {
511 0 : struct folio *folio = fbatch.folios[i];
512 :
513 : /* We rely upon deletion not changing folio->index */
514 :
515 0 : if (xa_is_value(folio)) {
516 0 : count += invalidate_exceptional_entry(mapping,
517 : indices[i], folio);
518 0 : continue;
519 : }
520 :
521 0 : ret = mapping_evict_folio(mapping, folio);
522 0 : 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 0 : if (!ret) {
528 0 : deactivate_file_folio(folio);
529 : /* Likely in the lru cache of a remote CPU */
530 0 : if (nr_failed)
531 0 : (*nr_failed)++;
532 : }
533 0 : count += ret;
534 : }
535 0 : folio_batch_remove_exceptionals(&fbatch);
536 0 : folio_batch_release(&fbatch);
537 0 : cond_resched();
538 : }
539 0 : 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 indices that had their contents invalidated
555 : */
556 0 : unsigned long invalidate_mapping_pages(struct address_space *mapping,
557 : pgoff_t start, pgoff_t end)
558 : {
559 0 : return mapping_try_invalidate(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() caches.
569 : */
570 0 : static int invalidate_complete_folio2(struct address_space *mapping,
571 : struct folio *folio)
572 : {
573 0 : if (folio->mapping != mapping)
574 : return 0;
575 :
576 0 : if (folio_has_private(folio) &&
577 0 : !filemap_release_folio(folio, GFP_KERNEL))
578 : return 0;
579 :
580 0 : spin_lock(&mapping->host->i_lock);
581 0 : xa_lock_irq(&mapping->i_pages);
582 0 : if (folio_test_dirty(folio))
583 : goto failed;
584 :
585 0 : BUG_ON(folio_has_private(folio));
586 0 : __filemap_remove_folio(folio, NULL);
587 0 : xa_unlock_irq(&mapping->i_pages);
588 0 : if (mapping_shrinkable(mapping))
589 0 : inode_add_lru(mapping->host);
590 0 : spin_unlock(&mapping->host->i_lock);
591 :
592 0 : filemap_free_folio(mapping, folio);
593 0 : return 1;
594 : failed:
595 0 : xa_unlock_irq(&mapping->i_pages);
596 0 : spin_unlock(&mapping->host->i_lock);
597 0 : return 0;
598 : }
599 :
600 : static int folio_launder(struct address_space *mapping, struct folio *folio)
601 : {
602 0 : if (!folio_test_dirty(folio))
603 : return 0;
604 0 : if (folio->mapping != mapping || mapping->a_ops->launder_folio == NULL)
605 : return 0;
606 0 : 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 : */
620 0 : 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 0 : int ret = 0;
628 0 : int ret2 = 0;
629 0 : int did_range_unmap = 0;
630 :
631 0 : if (mapping_empty(mapping))
632 : return 0;
633 :
634 0 : folio_batch_init(&fbatch);
635 0 : index = start;
636 0 : while (find_get_entries(mapping, &index, end, &fbatch, indices)) {
637 0 : for (i = 0; i < folio_batch_count(&fbatch); i++) {
638 0 : struct folio *folio = fbatch.folios[i];
639 :
640 : /* We rely upon deletion not changing folio->index */
641 :
642 0 : if (xa_is_value(folio)) {
643 0 : if (!invalidate_exceptional_entry2(mapping,
644 : indices[i], folio))
645 : ret = -EBUSY;
646 0 : continue;
647 : }
648 :
649 0 : 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 0 : unmap_mapping_pages(mapping, indices[i],
655 0 : (1 + end - indices[i]), false);
656 0 : did_range_unmap = 1;
657 : }
658 :
659 0 : folio_lock(folio);
660 : VM_BUG_ON_FOLIO(!folio_contains(folio, indices[i]), folio);
661 0 : if (folio->mapping != mapping) {
662 0 : folio_unlock(folio);
663 0 : continue;
664 : }
665 0 : folio_wait_writeback(folio);
666 :
667 0 : if (folio_mapped(folio))
668 0 : unmap_mapping_folio(folio);
669 0 : BUG_ON(folio_mapped(folio));
670 :
671 0 : ret2 = folio_launder(mapping, folio);
672 0 : if (ret2 == 0) {
673 0 : if (!invalidate_complete_folio2(mapping, folio))
674 0 : ret2 = -EBUSY;
675 : }
676 0 : if (ret2 < 0)
677 0 : ret = ret2;
678 0 : folio_unlock(folio);
679 : }
680 0 : folio_batch_remove_exceptionals(&fbatch);
681 0 : folio_batch_release(&fbatch);
682 0 : 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 : */
707 0 : int invalidate_inode_pages2(struct address_space *mapping)
708 : {
709 0 : 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 : */
728 0 : void truncate_pagecache(struct inode *inode, loff_t newsize)
729 : {
730 0 : struct address_space *mapping = inode->i_mapping;
731 0 : 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 0 : unmap_mapping_range(mapping, holebegin, 0, 1);
743 0 : truncate_inode_pages(mapping, newsize);
744 0 : unmap_mapping_range(mapping, holebegin, 0, 1);
745 0 : }
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 : */
761 0 : void truncate_setsize(struct inode *inode, loff_t newsize)
762 : {
763 0 : loff_t oldsize = inode->i_size;
764 :
765 0 : i_size_write(inode, newsize);
766 0 : if (newsize > oldsize)
767 0 : pagecache_isize_extended(inode, oldsize, newsize);
768 0 : truncate_pagecache(inode, newsize);
769 0 : }
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 : */
791 0 : void pagecache_isize_extended(struct inode *inode, loff_t from, loff_t to)
792 : {
793 0 : int bsize = i_blocksize(inode);
794 : loff_t rounded_from;
795 : struct page *page;
796 : pgoff_t index;
797 :
798 0 : WARN_ON(to > inode->i_size);
799 :
800 0 : if (from >= to || bsize == PAGE_SIZE)
801 : return;
802 : /* Page straddling @from will not have any hole block created? */
803 0 : rounded_from = round_up(from, bsize);
804 0 : if (to <= rounded_from || !(rounded_from & (PAGE_SIZE - 1)))
805 : return;
806 :
807 0 : index = from >> PAGE_SHIFT;
808 0 : page = find_lock_page(inode->i_mapping, index);
809 : /* Page not cached? Nothing to do */
810 0 : if (!page)
811 : return;
812 : /*
813 : * See clear_page_dirty_for_io() for details why set_page_dirty()
814 : * is needed.
815 : */
816 0 : if (page_mkclean(page))
817 0 : set_page_dirty(page);
818 0 : unlock_page(page);
819 0 : 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 : */
836 0 : void truncate_pagecache_range(struct inode *inode, loff_t lstart, loff_t lend)
837 : {
838 0 : struct address_space *mapping = inode->i_mapping;
839 0 : loff_t unmap_start = round_up(lstart, PAGE_SIZE);
840 0 : 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 0 : if ((u64)unmap_end > (u64)unmap_start)
855 0 : unmap_mapping_range(mapping, unmap_start,
856 : 1 + unmap_end - unmap_start, 0);
857 0 : truncate_inode_pages_range(mapping, lstart, lend);
858 0 : }
859 : EXPORT_SYMBOL(truncate_pagecache_range);
|
