]>
Commit | Line | Data |
---|---|---|
1 | /* | |
2 | * mm/truncate.c - code for taking down pages from address_spaces | |
3 | * | |
4 | * Copyright (C) 2002, Linus Torvalds | |
5 | * | |
6 | * 10Sep2002 Andrew Morton | |
7 | * Initial version. | |
8 | */ | |
9 | ||
10 | #include <linux/kernel.h> | |
11 | #include <linux/backing-dev.h> | |
12 | #include <linux/dax.h> | |
13 | #include <linux/gfp.h> | |
14 | #include <linux/mm.h> | |
15 | #include <linux/swap.h> | |
16 | #include <linux/export.h> | |
17 | #include <linux/pagemap.h> | |
18 | #include <linux/highmem.h> | |
19 | #include <linux/pagevec.h> | |
20 | #include <linux/task_io_accounting_ops.h> | |
21 | #include <linux/buffer_head.h> /* grr. try_to_release_page, | |
22 | do_invalidatepage */ | |
23 | #include <linux/shmem_fs.h> | |
24 | #include <linux/cleancache.h> | |
25 | #include <linux/rmap.h> | |
26 | #include "internal.h" | |
27 | ||
28 | static void clear_shadow_entry(struct address_space *mapping, pgoff_t index, | |
29 | void *entry) | |
30 | { | |
31 | struct radix_tree_node *node; | |
32 | void **slot; | |
33 | ||
34 | spin_lock_irq(&mapping->tree_lock); | |
35 | /* | |
36 | * Regular page slots are stabilized by the page lock even | |
37 | * without the tree itself locked. These unlocked entries | |
38 | * need verification under the tree lock. | |
39 | */ | |
40 | if (!__radix_tree_lookup(&mapping->page_tree, index, &node, &slot)) | |
41 | goto unlock; | |
42 | if (*slot != entry) | |
43 | goto unlock; | |
44 | __radix_tree_replace(&mapping->page_tree, node, slot, NULL, | |
45 | workingset_update_node, mapping); | |
46 | mapping->nrexceptional--; | |
47 | unlock: | |
48 | spin_unlock_irq(&mapping->tree_lock); | |
49 | } | |
50 | ||
51 | /* | |
52 | * Unconditionally remove exceptional entry. Usually called from truncate path. | |
53 | */ | |
54 | static void truncate_exceptional_entry(struct address_space *mapping, | |
55 | pgoff_t index, void *entry) | |
56 | { | |
57 | /* Handled by shmem itself */ | |
58 | if (shmem_mapping(mapping)) | |
59 | return; | |
60 | ||
61 | if (dax_mapping(mapping)) { | |
62 | dax_delete_mapping_entry(mapping, index); | |
63 | return; | |
64 | } | |
65 | clear_shadow_entry(mapping, index, entry); | |
66 | } | |
67 | ||
68 | /* | |
69 | * Invalidate exceptional entry if easily possible. This handles exceptional | |
70 | * entries for invalidate_inode_pages(). | |
71 | */ | |
72 | static int invalidate_exceptional_entry(struct address_space *mapping, | |
73 | pgoff_t index, void *entry) | |
74 | { | |
75 | /* Handled by shmem itself, or for DAX we do nothing. */ | |
76 | if (shmem_mapping(mapping) || dax_mapping(mapping)) | |
77 | return 1; | |
78 | clear_shadow_entry(mapping, index, entry); | |
79 | return 1; | |
80 | } | |
81 | ||
82 | /* | |
83 | * Invalidate exceptional entry if clean. This handles exceptional entries for | |
84 | * invalidate_inode_pages2() so for DAX it evicts only clean entries. | |
85 | */ | |
86 | static int invalidate_exceptional_entry2(struct address_space *mapping, | |
87 | pgoff_t index, void *entry) | |
88 | { | |
89 | /* Handled by shmem itself */ | |
90 | if (shmem_mapping(mapping)) | |
91 | return 1; | |
92 | if (dax_mapping(mapping)) | |
93 | return dax_invalidate_mapping_entry_sync(mapping, index); | |
94 | clear_shadow_entry(mapping, index, entry); | |
95 | return 1; | |
96 | } | |
97 | ||
98 | /** | |
99 | * do_invalidatepage - invalidate part or all of a page | |
100 | * @page: the page which is affected | |
101 | * @offset: start of the range to invalidate | |
102 | * @length: length of the range to invalidate | |
103 | * | |
104 | * do_invalidatepage() is called when all or part of the page has become | |
105 | * invalidated by a truncate operation. | |
106 | * | |
107 | * do_invalidatepage() does not have to release all buffers, but it must | |
108 | * ensure that no dirty buffer is left outside @offset and that no I/O | |
109 | * is underway against any of the blocks which are outside the truncation | |
110 | * point. Because the caller is about to free (and possibly reuse) those | |
111 | * blocks on-disk. | |
112 | */ | |
113 | void do_invalidatepage(struct page *page, unsigned int offset, | |
114 | unsigned int length) | |
115 | { | |
116 | void (*invalidatepage)(struct page *, unsigned int, unsigned int); | |
117 | ||
118 | invalidatepage = page->mapping->a_ops->invalidatepage; | |
119 | #ifdef CONFIG_BLOCK | |
120 | if (!invalidatepage) | |
121 | invalidatepage = block_invalidatepage; | |
122 | #endif | |
123 | if (invalidatepage) | |
124 | (*invalidatepage)(page, offset, length); | |
125 | } | |
126 | ||
127 | /* | |
128 | * If truncate cannot remove the fs-private metadata from the page, the page | |
129 | * becomes orphaned. It will be left on the LRU and may even be mapped into | |
130 | * user pagetables if we're racing with filemap_fault(). | |
131 | * | |
132 | * We need to bale out if page->mapping is no longer equal to the original | |
133 | * mapping. This happens a) when the VM reclaimed the page while we waited on | |
134 | * its lock, b) when a concurrent invalidate_mapping_pages got there first and | |
135 | * c) when tmpfs swizzles a page between a tmpfs inode and swapper_space. | |
136 | */ | |
137 | static int | |
138 | truncate_complete_page(struct address_space *mapping, struct page *page) | |
139 | { | |
140 | if (page->mapping != mapping) | |
141 | return -EIO; | |
142 | ||
143 | if (page_has_private(page)) | |
144 | do_invalidatepage(page, 0, PAGE_SIZE); | |
145 | ||
146 | /* | |
147 | * Some filesystems seem to re-dirty the page even after | |
148 | * the VM has canceled the dirty bit (eg ext3 journaling). | |
149 | * Hence dirty accounting check is placed after invalidation. | |
150 | */ | |
151 | cancel_dirty_page(page); | |
152 | ClearPageMappedToDisk(page); | |
153 | delete_from_page_cache(page); | |
154 | return 0; | |
155 | } | |
156 | ||
157 | /* | |
158 | * This is for invalidate_mapping_pages(). That function can be called at | |
159 | * any time, and is not supposed to throw away dirty pages. But pages can | |
160 | * be marked dirty at any time too, so use remove_mapping which safely | |
161 | * discards clean, unused pages. | |
162 | * | |
163 | * Returns non-zero if the page was successfully invalidated. | |
164 | */ | |
165 | static int | |
166 | invalidate_complete_page(struct address_space *mapping, struct page *page) | |
167 | { | |
168 | int ret; | |
169 | ||
170 | if (page->mapping != mapping) | |
171 | return 0; | |
172 | ||
173 | if (page_has_private(page) && !try_to_release_page(page, 0)) | |
174 | return 0; | |
175 | ||
176 | ret = remove_mapping(mapping, page); | |
177 | ||
178 | return ret; | |
179 | } | |
180 | ||
181 | int truncate_inode_page(struct address_space *mapping, struct page *page) | |
182 | { | |
183 | loff_t holelen; | |
184 | VM_BUG_ON_PAGE(PageTail(page), page); | |
185 | ||
186 | holelen = PageTransHuge(page) ? HPAGE_PMD_SIZE : PAGE_SIZE; | |
187 | if (page_mapped(page)) { | |
188 | unmap_mapping_range(mapping, | |
189 | (loff_t)page->index << PAGE_SHIFT, | |
190 | holelen, 0); | |
191 | } | |
192 | return truncate_complete_page(mapping, page); | |
193 | } | |
194 | ||
195 | /* | |
196 | * Used to get rid of pages on hardware memory corruption. | |
197 | */ | |
198 | int generic_error_remove_page(struct address_space *mapping, struct page *page) | |
199 | { | |
200 | if (!mapping) | |
201 | return -EINVAL; | |
202 | /* | |
203 | * Only punch for normal data pages for now. | |
204 | * Handling other types like directories would need more auditing. | |
205 | */ | |
206 | if (!S_ISREG(mapping->host->i_mode)) | |
207 | return -EIO; | |
208 | return truncate_inode_page(mapping, page); | |
209 | } | |
210 | EXPORT_SYMBOL(generic_error_remove_page); | |
211 | ||
212 | /* | |
213 | * Safely invalidate one page from its pagecache mapping. | |
214 | * It only drops clean, unused pages. The page must be locked. | |
215 | * | |
216 | * Returns 1 if the page is successfully invalidated, otherwise 0. | |
217 | */ | |
218 | int invalidate_inode_page(struct page *page) | |
219 | { | |
220 | struct address_space *mapping = page_mapping(page); | |
221 | if (!mapping) | |
222 | return 0; | |
223 | if (PageDirty(page) || PageWriteback(page)) | |
224 | return 0; | |
225 | if (page_mapped(page)) | |
226 | return 0; | |
227 | return invalidate_complete_page(mapping, page); | |
228 | } | |
229 | ||
230 | /** | |
231 | * truncate_inode_pages_range - truncate range of pages specified by start & end byte offsets | |
232 | * @mapping: mapping to truncate | |
233 | * @lstart: offset from which to truncate | |
234 | * @lend: offset to which to truncate (inclusive) | |
235 | * | |
236 | * Truncate the page cache, removing the pages that are between | |
237 | * specified offsets (and zeroing out partial pages | |
238 | * if lstart or lend + 1 is not page aligned). | |
239 | * | |
240 | * Truncate takes two passes - the first pass is nonblocking. It will not | |
241 | * block on page locks and it will not block on writeback. The second pass | |
242 | * will wait. This is to prevent as much IO as possible in the affected region. | |
243 | * The first pass will remove most pages, so the search cost of the second pass | |
244 | * is low. | |
245 | * | |
246 | * We pass down the cache-hot hint to the page freeing code. Even if the | |
247 | * mapping is large, it is probably the case that the final pages are the most | |
248 | * recently touched, and freeing happens in ascending file offset order. | |
249 | * | |
250 | * Note that since ->invalidatepage() accepts range to invalidate | |
251 | * truncate_inode_pages_range is able to handle cases where lend + 1 is not | |
252 | * page aligned properly. | |
253 | */ | |
254 | void truncate_inode_pages_range(struct address_space *mapping, | |
255 | loff_t lstart, loff_t lend) | |
256 | { | |
257 | pgoff_t start; /* inclusive */ | |
258 | pgoff_t end; /* exclusive */ | |
259 | unsigned int partial_start; /* inclusive */ | |
260 | unsigned int partial_end; /* exclusive */ | |
261 | struct pagevec pvec; | |
262 | pgoff_t indices[PAGEVEC_SIZE]; | |
263 | pgoff_t index; | |
264 | int i; | |
265 | ||
266 | if (mapping->nrpages == 0 && mapping->nrexceptional == 0) | |
267 | goto out; | |
268 | ||
269 | /* Offsets within partial pages */ | |
270 | partial_start = lstart & (PAGE_SIZE - 1); | |
271 | partial_end = (lend + 1) & (PAGE_SIZE - 1); | |
272 | ||
273 | /* | |
274 | * 'start' and 'end' always covers the range of pages to be fully | |
275 | * truncated. Partial pages are covered with 'partial_start' at the | |
276 | * start of the range and 'partial_end' at the end of the range. | |
277 | * Note that 'end' is exclusive while 'lend' is inclusive. | |
278 | */ | |
279 | start = (lstart + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
280 | if (lend == -1) | |
281 | /* | |
282 | * lend == -1 indicates end-of-file so we have to set 'end' | |
283 | * to the highest possible pgoff_t and since the type is | |
284 | * unsigned we're using -1. | |
285 | */ | |
286 | end = -1; | |
287 | else | |
288 | end = (lend + 1) >> PAGE_SHIFT; | |
289 | ||
290 | pagevec_init(&pvec, 0); | |
291 | index = start; | |
292 | while (index < end && pagevec_lookup_entries(&pvec, mapping, index, | |
293 | min(end - index, (pgoff_t)PAGEVEC_SIZE), | |
294 | indices)) { | |
295 | for (i = 0; i < pagevec_count(&pvec); i++) { | |
296 | struct page *page = pvec.pages[i]; | |
297 | ||
298 | /* We rely upon deletion not changing page->index */ | |
299 | index = indices[i]; | |
300 | if (index >= end) | |
301 | break; | |
302 | ||
303 | if (radix_tree_exceptional_entry(page)) { | |
304 | truncate_exceptional_entry(mapping, index, | |
305 | page); | |
306 | continue; | |
307 | } | |
308 | ||
309 | if (!trylock_page(page)) | |
310 | continue; | |
311 | WARN_ON(page_to_index(page) != index); | |
312 | if (PageWriteback(page)) { | |
313 | unlock_page(page); | |
314 | continue; | |
315 | } | |
316 | truncate_inode_page(mapping, page); | |
317 | unlock_page(page); | |
318 | } | |
319 | pagevec_remove_exceptionals(&pvec); | |
320 | pagevec_release(&pvec); | |
321 | cond_resched(); | |
322 | index++; | |
323 | } | |
324 | ||
325 | if (partial_start) { | |
326 | struct page *page = find_lock_page(mapping, start - 1); | |
327 | if (page) { | |
328 | unsigned int top = PAGE_SIZE; | |
329 | if (start > end) { | |
330 | /* Truncation within a single page */ | |
331 | top = partial_end; | |
332 | partial_end = 0; | |
333 | } | |
334 | wait_on_page_writeback(page); | |
335 | zero_user_segment(page, partial_start, top); | |
336 | cleancache_invalidate_page(mapping, page); | |
337 | if (page_has_private(page)) | |
338 | do_invalidatepage(page, partial_start, | |
339 | top - partial_start); | |
340 | unlock_page(page); | |
341 | put_page(page); | |
342 | } | |
343 | } | |
344 | if (partial_end) { | |
345 | struct page *page = find_lock_page(mapping, end); | |
346 | if (page) { | |
347 | wait_on_page_writeback(page); | |
348 | zero_user_segment(page, 0, partial_end); | |
349 | cleancache_invalidate_page(mapping, page); | |
350 | if (page_has_private(page)) | |
351 | do_invalidatepage(page, 0, | |
352 | partial_end); | |
353 | unlock_page(page); | |
354 | put_page(page); | |
355 | } | |
356 | } | |
357 | /* | |
358 | * If the truncation happened within a single page no pages | |
359 | * will be released, just zeroed, so we can bail out now. | |
360 | */ | |
361 | if (start >= end) | |
362 | goto out; | |
363 | ||
364 | index = start; | |
365 | for ( ; ; ) { | |
366 | cond_resched(); | |
367 | if (!pagevec_lookup_entries(&pvec, mapping, index, | |
368 | min(end - index, (pgoff_t)PAGEVEC_SIZE), indices)) { | |
369 | /* If all gone from start onwards, we're done */ | |
370 | if (index == start) | |
371 | break; | |
372 | /* Otherwise restart to make sure all gone */ | |
373 | index = start; | |
374 | continue; | |
375 | } | |
376 | if (index == start && indices[0] >= end) { | |
377 | /* All gone out of hole to be punched, we're done */ | |
378 | pagevec_remove_exceptionals(&pvec); | |
379 | pagevec_release(&pvec); | |
380 | break; | |
381 | } | |
382 | for (i = 0; i < pagevec_count(&pvec); i++) { | |
383 | struct page *page = pvec.pages[i]; | |
384 | ||
385 | /* We rely upon deletion not changing page->index */ | |
386 | index = indices[i]; | |
387 | if (index >= end) { | |
388 | /* Restart punch to make sure all gone */ | |
389 | index = start - 1; | |
390 | break; | |
391 | } | |
392 | ||
393 | if (radix_tree_exceptional_entry(page)) { | |
394 | truncate_exceptional_entry(mapping, index, | |
395 | page); | |
396 | continue; | |
397 | } | |
398 | ||
399 | lock_page(page); | |
400 | WARN_ON(page_to_index(page) != index); | |
401 | wait_on_page_writeback(page); | |
402 | truncate_inode_page(mapping, page); | |
403 | unlock_page(page); | |
404 | } | |
405 | pagevec_remove_exceptionals(&pvec); | |
406 | pagevec_release(&pvec); | |
407 | index++; | |
408 | } | |
409 | ||
410 | out: | |
411 | cleancache_invalidate_inode(mapping); | |
412 | } | |
413 | EXPORT_SYMBOL(truncate_inode_pages_range); | |
414 | ||
415 | /** | |
416 | * truncate_inode_pages - truncate *all* the pages from an offset | |
417 | * @mapping: mapping to truncate | |
418 | * @lstart: offset from which to truncate | |
419 | * | |
420 | * Called under (and serialised by) inode->i_mutex. | |
421 | * | |
422 | * Note: When this function returns, there can be a page in the process of | |
423 | * deletion (inside __delete_from_page_cache()) in the specified range. Thus | |
424 | * mapping->nrpages can be non-zero when this function returns even after | |
425 | * truncation of the whole mapping. | |
426 | */ | |
427 | void truncate_inode_pages(struct address_space *mapping, loff_t lstart) | |
428 | { | |
429 | truncate_inode_pages_range(mapping, lstart, (loff_t)-1); | |
430 | } | |
431 | EXPORT_SYMBOL(truncate_inode_pages); | |
432 | ||
433 | /** | |
434 | * truncate_inode_pages_final - truncate *all* pages before inode dies | |
435 | * @mapping: mapping to truncate | |
436 | * | |
437 | * Called under (and serialized by) inode->i_mutex. | |
438 | * | |
439 | * Filesystems have to use this in the .evict_inode path to inform the | |
440 | * VM that this is the final truncate and the inode is going away. | |
441 | */ | |
442 | void truncate_inode_pages_final(struct address_space *mapping) | |
443 | { | |
444 | unsigned long nrexceptional; | |
445 | unsigned long nrpages; | |
446 | ||
447 | /* | |
448 | * Page reclaim can not participate in regular inode lifetime | |
449 | * management (can't call iput()) and thus can race with the | |
450 | * inode teardown. Tell it when the address space is exiting, | |
451 | * so that it does not install eviction information after the | |
452 | * final truncate has begun. | |
453 | */ | |
454 | mapping_set_exiting(mapping); | |
455 | ||
456 | /* | |
457 | * When reclaim installs eviction entries, it increases | |
458 | * nrexceptional first, then decreases nrpages. Make sure we see | |
459 | * this in the right order or we might miss an entry. | |
460 | */ | |
461 | nrpages = mapping->nrpages; | |
462 | smp_rmb(); | |
463 | nrexceptional = mapping->nrexceptional; | |
464 | ||
465 | if (nrpages || nrexceptional) { | |
466 | /* | |
467 | * As truncation uses a lockless tree lookup, cycle | |
468 | * the tree lock to make sure any ongoing tree | |
469 | * modification that does not see AS_EXITING is | |
470 | * completed before starting the final truncate. | |
471 | */ | |
472 | spin_lock_irq(&mapping->tree_lock); | |
473 | spin_unlock_irq(&mapping->tree_lock); | |
474 | ||
475 | truncate_inode_pages(mapping, 0); | |
476 | } | |
477 | } | |
478 | EXPORT_SYMBOL(truncate_inode_pages_final); | |
479 | ||
480 | /** | |
481 | * invalidate_mapping_pages - Invalidate all the unlocked pages of one inode | |
482 | * @mapping: the address_space which holds the pages to invalidate | |
483 | * @start: the offset 'from' which to invalidate | |
484 | * @end: the offset 'to' which to invalidate (inclusive) | |
485 | * | |
486 | * This function only removes the unlocked pages, if you want to | |
487 | * remove all the pages of one inode, you must call truncate_inode_pages. | |
488 | * | |
489 | * invalidate_mapping_pages() will not block on IO activity. It will not | |
490 | * invalidate pages which are dirty, locked, under writeback or mapped into | |
491 | * pagetables. | |
492 | */ | |
493 | unsigned long invalidate_mapping_pages(struct address_space *mapping, | |
494 | pgoff_t start, pgoff_t end) | |
495 | { | |
496 | pgoff_t indices[PAGEVEC_SIZE]; | |
497 | struct pagevec pvec; | |
498 | pgoff_t index = start; | |
499 | unsigned long ret; | |
500 | unsigned long count = 0; | |
501 | int i; | |
502 | ||
503 | pagevec_init(&pvec, 0); | |
504 | while (index <= end && pagevec_lookup_entries(&pvec, mapping, index, | |
505 | min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1, | |
506 | indices)) { | |
507 | for (i = 0; i < pagevec_count(&pvec); i++) { | |
508 | struct page *page = pvec.pages[i]; | |
509 | ||
510 | /* We rely upon deletion not changing page->index */ | |
511 | index = indices[i]; | |
512 | if (index > end) | |
513 | break; | |
514 | ||
515 | if (radix_tree_exceptional_entry(page)) { | |
516 | invalidate_exceptional_entry(mapping, index, | |
517 | page); | |
518 | continue; | |
519 | } | |
520 | ||
521 | if (!trylock_page(page)) | |
522 | continue; | |
523 | ||
524 | WARN_ON(page_to_index(page) != index); | |
525 | ||
526 | /* Middle of THP: skip */ | |
527 | if (PageTransTail(page)) { | |
528 | unlock_page(page); | |
529 | continue; | |
530 | } else if (PageTransHuge(page)) { | |
531 | index += HPAGE_PMD_NR - 1; | |
532 | i += HPAGE_PMD_NR - 1; | |
533 | /* | |
534 | * 'end' is in the middle of THP. Don't | |
535 | * invalidate the page as the part outside of | |
536 | * 'end' could be still useful. | |
537 | */ | |
538 | if (index > end) { | |
539 | unlock_page(page); | |
540 | continue; | |
541 | } | |
542 | } | |
543 | ||
544 | ret = invalidate_inode_page(page); | |
545 | unlock_page(page); | |
546 | /* | |
547 | * Invalidation is a hint that the page is no longer | |
548 | * of interest and try to speed up its reclaim. | |
549 | */ | |
550 | if (!ret) | |
551 | deactivate_file_page(page); | |
552 | count += ret; | |
553 | } | |
554 | pagevec_remove_exceptionals(&pvec); | |
555 | pagevec_release(&pvec); | |
556 | cond_resched(); | |
557 | index++; | |
558 | } | |
559 | return count; | |
560 | } | |
561 | EXPORT_SYMBOL(invalidate_mapping_pages); | |
562 | ||
563 | /* | |
564 | * This is like invalidate_complete_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 lru_cache_add() pagevecs. | |
569 | */ | |
570 | static int | |
571 | invalidate_complete_page2(struct address_space *mapping, struct page *page) | |
572 | { | |
573 | unsigned long flags; | |
574 | ||
575 | if (page->mapping != mapping) | |
576 | return 0; | |
577 | ||
578 | if (page_has_private(page) && !try_to_release_page(page, GFP_KERNEL)) | |
579 | return 0; | |
580 | ||
581 | spin_lock_irqsave(&mapping->tree_lock, flags); | |
582 | if (PageDirty(page)) | |
583 | goto failed; | |
584 | ||
585 | BUG_ON(page_has_private(page)); | |
586 | __delete_from_page_cache(page, NULL); | |
587 | spin_unlock_irqrestore(&mapping->tree_lock, flags); | |
588 | ||
589 | if (mapping->a_ops->freepage) | |
590 | mapping->a_ops->freepage(page); | |
591 | ||
592 | put_page(page); /* pagecache ref */ | |
593 | return 1; | |
594 | failed: | |
595 | spin_unlock_irqrestore(&mapping->tree_lock, flags); | |
596 | return 0; | |
597 | } | |
598 | ||
599 | static int do_launder_page(struct address_space *mapping, struct page *page) | |
600 | { | |
601 | if (!PageDirty(page)) | |
602 | return 0; | |
603 | if (page->mapping != mapping || mapping->a_ops->launder_page == NULL) | |
604 | return 0; | |
605 | return mapping->a_ops->launder_page(page); | |
606 | } | |
607 | ||
608 | /** | |
609 | * invalidate_inode_pages2_range - remove range of pages from an address_space | |
610 | * @mapping: the address_space | |
611 | * @start: the page offset 'from' which to invalidate | |
612 | * @end: the page offset 'to' which to invalidate (inclusive) | |
613 | * | |
614 | * Any pages which are found to be mapped into pagetables are unmapped prior to | |
615 | * invalidation. | |
616 | * | |
617 | * Returns -EBUSY if any pages could not be invalidated. | |
618 | */ | |
619 | int invalidate_inode_pages2_range(struct address_space *mapping, | |
620 | pgoff_t start, pgoff_t end) | |
621 | { | |
622 | pgoff_t indices[PAGEVEC_SIZE]; | |
623 | struct pagevec pvec; | |
624 | pgoff_t index; | |
625 | int i; | |
626 | int ret = 0; | |
627 | int ret2 = 0; | |
628 | int did_range_unmap = 0; | |
629 | ||
630 | if (mapping->nrpages == 0 && mapping->nrexceptional == 0) | |
631 | goto out; | |
632 | ||
633 | pagevec_init(&pvec, 0); | |
634 | index = start; | |
635 | while (index <= end && pagevec_lookup_entries(&pvec, mapping, index, | |
636 | min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1, | |
637 | indices)) { | |
638 | for (i = 0; i < pagevec_count(&pvec); i++) { | |
639 | struct page *page = pvec.pages[i]; | |
640 | ||
641 | /* We rely upon deletion not changing page->index */ | |
642 | index = indices[i]; | |
643 | if (index > end) | |
644 | break; | |
645 | ||
646 | if (radix_tree_exceptional_entry(page)) { | |
647 | if (!invalidate_exceptional_entry2(mapping, | |
648 | index, page)) | |
649 | ret = -EBUSY; | |
650 | continue; | |
651 | } | |
652 | ||
653 | lock_page(page); | |
654 | WARN_ON(page_to_index(page) != index); | |
655 | if (page->mapping != mapping) { | |
656 | unlock_page(page); | |
657 | continue; | |
658 | } | |
659 | wait_on_page_writeback(page); | |
660 | if (page_mapped(page)) { | |
661 | if (!did_range_unmap) { | |
662 | /* | |
663 | * Zap the rest of the file in one hit. | |
664 | */ | |
665 | unmap_mapping_range(mapping, | |
666 | (loff_t)index << PAGE_SHIFT, | |
667 | (loff_t)(1 + end - index) | |
668 | << PAGE_SHIFT, | |
669 | 0); | |
670 | did_range_unmap = 1; | |
671 | } else { | |
672 | /* | |
673 | * Just zap this page | |
674 | */ | |
675 | unmap_mapping_range(mapping, | |
676 | (loff_t)index << PAGE_SHIFT, | |
677 | PAGE_SIZE, 0); | |
678 | } | |
679 | } | |
680 | BUG_ON(page_mapped(page)); | |
681 | ret2 = do_launder_page(mapping, page); | |
682 | if (ret2 == 0) { | |
683 | if (!invalidate_complete_page2(mapping, page)) | |
684 | ret2 = -EBUSY; | |
685 | } | |
686 | if (ret2 < 0) | |
687 | ret = ret2; | |
688 | unlock_page(page); | |
689 | } | |
690 | pagevec_remove_exceptionals(&pvec); | |
691 | pagevec_release(&pvec); | |
692 | cond_resched(); | |
693 | index++; | |
694 | } | |
695 | /* | |
696 | * For DAX we invalidate page tables after invalidating radix tree. We | |
697 | * could invalidate page tables while invalidating each entry however | |
698 | * that would be expensive. And doing range unmapping before doesn't | |
699 | * work as we have no cheap way to find whether radix tree entry didn't | |
700 | * get remapped later. | |
701 | */ | |
702 | if (dax_mapping(mapping)) { | |
703 | unmap_mapping_range(mapping, (loff_t)start << PAGE_SHIFT, | |
704 | (loff_t)(end - start + 1) << PAGE_SHIFT, 0); | |
705 | } | |
706 | out: | |
707 | cleancache_invalidate_inode(mapping); | |
708 | return ret; | |
709 | } | |
710 | EXPORT_SYMBOL_GPL(invalidate_inode_pages2_range); | |
711 | ||
712 | /** | |
713 | * invalidate_inode_pages2 - remove all pages from an address_space | |
714 | * @mapping: the address_space | |
715 | * | |
716 | * Any pages which are found to be mapped into pagetables are unmapped prior to | |
717 | * invalidation. | |
718 | * | |
719 | * Returns -EBUSY if any pages could not be invalidated. | |
720 | */ | |
721 | int invalidate_inode_pages2(struct address_space *mapping) | |
722 | { | |
723 | return invalidate_inode_pages2_range(mapping, 0, -1); | |
724 | } | |
725 | EXPORT_SYMBOL_GPL(invalidate_inode_pages2); | |
726 | ||
727 | /** | |
728 | * truncate_pagecache - unmap and remove pagecache that has been truncated | |
729 | * @inode: inode | |
730 | * @newsize: new file size | |
731 | * | |
732 | * inode's new i_size must already be written before truncate_pagecache | |
733 | * is called. | |
734 | * | |
735 | * This function should typically be called before the filesystem | |
736 | * releases resources associated with the freed range (eg. deallocates | |
737 | * blocks). This way, pagecache will always stay logically coherent | |
738 | * with on-disk format, and the filesystem would not have to deal with | |
739 | * situations such as writepage being called for a page that has already | |
740 | * had its underlying blocks deallocated. | |
741 | */ | |
742 | void truncate_pagecache(struct inode *inode, loff_t newsize) | |
743 | { | |
744 | struct address_space *mapping = inode->i_mapping; | |
745 | loff_t holebegin = round_up(newsize, PAGE_SIZE); | |
746 | ||
747 | /* | |
748 | * unmap_mapping_range is called twice, first simply for | |
749 | * efficiency so that truncate_inode_pages does fewer | |
750 | * single-page unmaps. However after this first call, and | |
751 | * before truncate_inode_pages finishes, it is possible for | |
752 | * private pages to be COWed, which remain after | |
753 | * truncate_inode_pages finishes, hence the second | |
754 | * unmap_mapping_range call must be made for correctness. | |
755 | */ | |
756 | unmap_mapping_range(mapping, holebegin, 0, 1); | |
757 | truncate_inode_pages(mapping, newsize); | |
758 | unmap_mapping_range(mapping, holebegin, 0, 1); | |
759 | } | |
760 | EXPORT_SYMBOL(truncate_pagecache); | |
761 | ||
762 | /** | |
763 | * truncate_setsize - update inode and pagecache for a new file size | |
764 | * @inode: inode | |
765 | * @newsize: new file size | |
766 | * | |
767 | * truncate_setsize updates i_size and performs pagecache truncation (if | |
768 | * necessary) to @newsize. It will be typically be called from the filesystem's | |
769 | * setattr function when ATTR_SIZE is passed in. | |
770 | * | |
771 | * Must be called with a lock serializing truncates and writes (generally | |
772 | * i_mutex but e.g. xfs uses a different lock) and before all filesystem | |
773 | * specific block truncation has been performed. | |
774 | */ | |
775 | void truncate_setsize(struct inode *inode, loff_t newsize) | |
776 | { | |
777 | loff_t oldsize = inode->i_size; | |
778 | ||
779 | i_size_write(inode, newsize); | |
780 | if (newsize > oldsize) | |
781 | pagecache_isize_extended(inode, oldsize, newsize); | |
782 | truncate_pagecache(inode, newsize); | |
783 | } | |
784 | EXPORT_SYMBOL(truncate_setsize); | |
785 | ||
786 | /** | |
787 | * pagecache_isize_extended - update pagecache after extension of i_size | |
788 | * @inode: inode for which i_size was extended | |
789 | * @from: original inode size | |
790 | * @to: new inode size | |
791 | * | |
792 | * Handle extension of inode size either caused by extending truncate or by | |
793 | * write starting after current i_size. We mark the page straddling current | |
794 | * i_size RO so that page_mkwrite() is called on the nearest write access to | |
795 | * the page. This way filesystem can be sure that page_mkwrite() is called on | |
796 | * the page before user writes to the page via mmap after the i_size has been | |
797 | * changed. | |
798 | * | |
799 | * The function must be called after i_size is updated so that page fault | |
800 | * coming after we unlock the page will already see the new i_size. | |
801 | * The function must be called while we still hold i_mutex - this not only | |
802 | * makes sure i_size is stable but also that userspace cannot observe new | |
803 | * i_size value before we are prepared to store mmap writes at new inode size. | |
804 | */ | |
805 | void pagecache_isize_extended(struct inode *inode, loff_t from, loff_t to) | |
806 | { | |
807 | int bsize = i_blocksize(inode); | |
808 | loff_t rounded_from; | |
809 | struct page *page; | |
810 | pgoff_t index; | |
811 | ||
812 | WARN_ON(to > inode->i_size); | |
813 | ||
814 | if (from >= to || bsize == PAGE_SIZE) | |
815 | return; | |
816 | /* Page straddling @from will not have any hole block created? */ | |
817 | rounded_from = round_up(from, bsize); | |
818 | if (to <= rounded_from || !(rounded_from & (PAGE_SIZE - 1))) | |
819 | return; | |
820 | ||
821 | index = from >> PAGE_SHIFT; | |
822 | page = find_lock_page(inode->i_mapping, index); | |
823 | /* Page not cached? Nothing to do */ | |
824 | if (!page) | |
825 | return; | |
826 | /* | |
827 | * See clear_page_dirty_for_io() for details why set_page_dirty() | |
828 | * is needed. | |
829 | */ | |
830 | if (page_mkclean(page)) | |
831 | set_page_dirty(page); | |
832 | unlock_page(page); | |
833 | put_page(page); | |
834 | } | |
835 | EXPORT_SYMBOL(pagecache_isize_extended); | |
836 | ||
837 | /** | |
838 | * truncate_pagecache_range - unmap and remove pagecache that is hole-punched | |
839 | * @inode: inode | |
840 | * @lstart: offset of beginning of hole | |
841 | * @lend: offset of last byte of hole | |
842 | * | |
843 | * This function should typically be called before the filesystem | |
844 | * releases resources associated with the freed range (eg. deallocates | |
845 | * blocks). This way, pagecache will always stay logically coherent | |
846 | * with on-disk format, and the filesystem would not have to deal with | |
847 | * situations such as writepage being called for a page that has already | |
848 | * had its underlying blocks deallocated. | |
849 | */ | |
850 | void truncate_pagecache_range(struct inode *inode, loff_t lstart, loff_t lend) | |
851 | { | |
852 | struct address_space *mapping = inode->i_mapping; | |
853 | loff_t unmap_start = round_up(lstart, PAGE_SIZE); | |
854 | loff_t unmap_end = round_down(1 + lend, PAGE_SIZE) - 1; | |
855 | /* | |
856 | * This rounding is currently just for example: unmap_mapping_range | |
857 | * expands its hole outwards, whereas we want it to contract the hole | |
858 | * inwards. However, existing callers of truncate_pagecache_range are | |
859 | * doing their own page rounding first. Note that unmap_mapping_range | |
860 | * allows holelen 0 for all, and we allow lend -1 for end of file. | |
861 | */ | |
862 | ||
863 | /* | |
864 | * Unlike in truncate_pagecache, unmap_mapping_range is called only | |
865 | * once (before truncating pagecache), and without "even_cows" flag: | |
866 | * hole-punching should not remove private COWed pages from the hole. | |
867 | */ | |
868 | if ((u64)unmap_end > (u64)unmap_start) | |
869 | unmap_mapping_range(mapping, unmap_start, | |
870 | 1 + unmap_end - unmap_start, 0); | |
871 | truncate_inode_pages_range(mapping, lstart, lend); | |
872 | } | |
873 | EXPORT_SYMBOL(truncate_pagecache_range); |