~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

TOMOYO Linux Cross Reference
Linux/mm/truncate.c

Version: ~ [ linux-5.19-rc3 ] ~ [ linux-5.18.5 ] ~ [ linux-5.17.15 ] ~ [ linux-5.16.20 ] ~ [ linux-5.15.48 ] ~ [ linux-5.14.21 ] ~ [ linux-5.13.19 ] ~ [ linux-5.12.19 ] ~ [ linux-5.11.22 ] ~ [ linux-5.10.123 ] ~ [ linux-5.9.16 ] ~ [ linux-5.8.18 ] ~ [ linux-5.7.19 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.199 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.248 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.284 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.319 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.302 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.10.108 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.9 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

  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/cleancache.h>
 24 #include <linux/rmap.h>
 25 #include "internal.h"
 26 
 27 static void clear_exceptional_entry(struct address_space *mapping,
 28                                     pgoff_t index, void *entry)
 29 {
 30         struct radix_tree_node *node;
 31         void **slot;
 32 
 33         /* Handled by shmem itself */
 34         if (shmem_mapping(mapping))
 35                 return;
 36 
 37         spin_lock_irq(&mapping->tree_lock);
 38 
 39         if (dax_mapping(mapping)) {
 40                 if (radix_tree_delete_item(&mapping->page_tree, index, entry))
 41                         mapping->nrexceptional--;
 42         } else {
 43                 /*
 44                  * Regular page slots are stabilized by the page lock even
 45                  * without the tree itself locked.  These unlocked entries
 46                  * need verification under the tree lock.
 47                  */
 48                 if (!__radix_tree_lookup(&mapping->page_tree, index, &node,
 49                                         &slot))
 50                         goto unlock;
 51                 if (*slot != entry)
 52                         goto unlock;
 53                 radix_tree_replace_slot(slot, NULL);
 54                 mapping->nrexceptional--;
 55                 if (!node)
 56                         goto unlock;
 57                 workingset_node_shadows_dec(node);
 58                 /*
 59                  * Don't track node without shadow entries.
 60                  *
 61                  * Avoid acquiring the list_lru lock if already untracked.
 62                  * The list_empty() test is safe as node->private_list is
 63                  * protected by mapping->tree_lock.
 64                  */
 65                 if (!workingset_node_shadows(node) &&
 66                     !list_empty(&node->private_list))
 67                         list_lru_del(&workingset_shadow_nodes,
 68                                         &node->private_list);
 69                 __radix_tree_delete_node(&mapping->page_tree, node);
 70         }
 71 unlock:
 72         spin_unlock_irq(&mapping->tree_lock);
 73 }
 74 
 75 /**
 76  * do_invalidatepage - invalidate part or all of a page
 77  * @page: the page which is affected
 78  * @offset: start of the range to invalidate
 79  * @length: length of the range to invalidate
 80  *
 81  * do_invalidatepage() is called when all or part of the page has become
 82  * invalidated by a truncate operation.
 83  *
 84  * do_invalidatepage() does not have to release all buffers, but it must
 85  * ensure that no dirty buffer is left outside @offset and that no I/O
 86  * is underway against any of the blocks which are outside the truncation
 87  * point.  Because the caller is about to free (and possibly reuse) those
 88  * blocks on-disk.
 89  */
 90 void do_invalidatepage(struct page *page, unsigned int offset,
 91                        unsigned int length)
 92 {
 93         void (*invalidatepage)(struct page *, unsigned int, unsigned int);
 94 
 95         invalidatepage = page->mapping->a_ops->invalidatepage;
 96 #ifdef CONFIG_BLOCK
 97         if (!invalidatepage)
 98                 invalidatepage = block_invalidatepage;
 99 #endif
100         if (invalidatepage)
101                 (*invalidatepage)(page, offset, length);
102 }
103 
104 /*
105  * If truncate cannot remove the fs-private metadata from the page, the page
106  * becomes orphaned.  It will be left on the LRU and may even be mapped into
107  * user pagetables if we're racing with filemap_fault().
108  *
109  * We need to bale out if page->mapping is no longer equal to the original
110  * mapping.  This happens a) when the VM reclaimed the page while we waited on
111  * its lock, b) when a concurrent invalidate_mapping_pages got there first and
112  * c) when tmpfs swizzles a page between a tmpfs inode and swapper_space.
113  */
114 static int
115 truncate_complete_page(struct address_space *mapping, struct page *page)
116 {
117         if (page->mapping != mapping)
118                 return -EIO;
119 
120         if (page_has_private(page))
121                 do_invalidatepage(page, 0, PAGE_CACHE_SIZE);
122 
123         /*
124          * Some filesystems seem to re-dirty the page even after
125          * the VM has canceled the dirty bit (eg ext3 journaling).
126          * Hence dirty accounting check is placed after invalidation.
127          */
128         cancel_dirty_page(page);
129         ClearPageMappedToDisk(page);
130         delete_from_page_cache(page);
131         return 0;
132 }
133 
134 /*
135  * This is for invalidate_mapping_pages().  That function can be called at
136  * any time, and is not supposed to throw away dirty pages.  But pages can
137  * be marked dirty at any time too, so use remove_mapping which safely
138  * discards clean, unused pages.
139  *
140  * Returns non-zero if the page was successfully invalidated.
141  */
142 static int
143 invalidate_complete_page(struct address_space *mapping, struct page *page)
144 {
145         int ret;
146 
147         if (page->mapping != mapping)
148                 return 0;
149 
150         if (page_has_private(page) && !try_to_release_page(page, 0))
151                 return 0;
152 
153         ret = remove_mapping(mapping, page);
154 
155         return ret;
156 }
157 
158 int truncate_inode_page(struct address_space *mapping, struct page *page)
159 {
160         if (page_mapped(page)) {
161                 unmap_mapping_range(mapping,
162                                    (loff_t)page->index << PAGE_CACHE_SHIFT,
163                                    PAGE_CACHE_SIZE, 0);
164         }
165         return truncate_complete_page(mapping, page);
166 }
167 
168 /*
169  * Used to get rid of pages on hardware memory corruption.
170  */
171 int generic_error_remove_page(struct address_space *mapping, struct page *page)
172 {
173         if (!mapping)
174                 return -EINVAL;
175         /*
176          * Only punch for normal data pages for now.
177          * Handling other types like directories would need more auditing.
178          */
179         if (!S_ISREG(mapping->host->i_mode))
180                 return -EIO;
181         return truncate_inode_page(mapping, page);
182 }
183 EXPORT_SYMBOL(generic_error_remove_page);
184 
185 /*
186  * Safely invalidate one page from its pagecache mapping.
187  * It only drops clean, unused pages. The page must be locked.
188  *
189  * Returns 1 if the page is successfully invalidated, otherwise 0.
190  */
191 int invalidate_inode_page(struct page *page)
192 {
193         struct address_space *mapping = page_mapping(page);
194         if (!mapping)
195                 return 0;
196         if (PageDirty(page) || PageWriteback(page))
197                 return 0;
198         if (page_mapped(page))
199                 return 0;
200         return invalidate_complete_page(mapping, page);
201 }
202 
203 /**
204  * truncate_inode_pages_range - truncate range of pages specified by start & end byte offsets
205  * @mapping: mapping to truncate
206  * @lstart: offset from which to truncate
207  * @lend: offset to which to truncate (inclusive)
208  *
209  * Truncate the page cache, removing the pages that are between
210  * specified offsets (and zeroing out partial pages
211  * if lstart or lend + 1 is not page aligned).
212  *
213  * Truncate takes two passes - the first pass is nonblocking.  It will not
214  * block on page locks and it will not block on writeback.  The second pass
215  * will wait.  This is to prevent as much IO as possible in the affected region.
216  * The first pass will remove most pages, so the search cost of the second pass
217  * is low.
218  *
219  * We pass down the cache-hot hint to the page freeing code.  Even if the
220  * mapping is large, it is probably the case that the final pages are the most
221  * recently touched, and freeing happens in ascending file offset order.
222  *
223  * Note that since ->invalidatepage() accepts range to invalidate
224  * truncate_inode_pages_range is able to handle cases where lend + 1 is not
225  * page aligned properly.
226  */
227 void truncate_inode_pages_range(struct address_space *mapping,
228                                 loff_t lstart, loff_t lend)
229 {
230         pgoff_t         start;          /* inclusive */
231         pgoff_t         end;            /* exclusive */
232         unsigned int    partial_start;  /* inclusive */
233         unsigned int    partial_end;    /* exclusive */
234         struct pagevec  pvec;
235         pgoff_t         indices[PAGEVEC_SIZE];
236         pgoff_t         index;
237         int             i;
238 
239         cleancache_invalidate_inode(mapping);
240         if (mapping->nrpages == 0 && mapping->nrexceptional == 0)
241                 return;
242 
243         /* Offsets within partial pages */
244         partial_start = lstart & (PAGE_CACHE_SIZE - 1);
245         partial_end = (lend + 1) & (PAGE_CACHE_SIZE - 1);
246 
247         /*
248          * 'start' and 'end' always covers the range of pages to be fully
249          * truncated. Partial pages are covered with 'partial_start' at the
250          * start of the range and 'partial_end' at the end of the range.
251          * Note that 'end' is exclusive while 'lend' is inclusive.
252          */
253         start = (lstart + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
254         if (lend == -1)
255                 /*
256                  * lend == -1 indicates end-of-file so we have to set 'end'
257                  * to the highest possible pgoff_t and since the type is
258                  * unsigned we're using -1.
259                  */
260                 end = -1;
261         else
262                 end = (lend + 1) >> PAGE_CACHE_SHIFT;
263 
264         pagevec_init(&pvec, 0);
265         index = start;
266         while (index < end && pagevec_lookup_entries(&pvec, mapping, index,
267                         min(end - index, (pgoff_t)PAGEVEC_SIZE),
268                         indices)) {
269                 for (i = 0; i < pagevec_count(&pvec); i++) {
270                         struct page *page = pvec.pages[i];
271 
272                         /* We rely upon deletion not changing page->index */
273                         index = indices[i];
274                         if (index >= end)
275                                 break;
276 
277                         if (radix_tree_exceptional_entry(page)) {
278                                 clear_exceptional_entry(mapping, index, page);
279                                 continue;
280                         }
281 
282                         if (!trylock_page(page))
283                                 continue;
284                         WARN_ON(page->index != index);
285                         if (PageWriteback(page)) {
286                                 unlock_page(page);
287                                 continue;
288                         }
289                         truncate_inode_page(mapping, page);
290                         unlock_page(page);
291                 }
292                 pagevec_remove_exceptionals(&pvec);
293                 pagevec_release(&pvec);
294                 cond_resched();
295                 index++;
296         }
297 
298         if (partial_start) {
299                 struct page *page = find_lock_page(mapping, start - 1);
300                 if (page) {
301                         unsigned int top = PAGE_CACHE_SIZE;
302                         if (start > end) {
303                                 /* Truncation within a single page */
304                                 top = partial_end;
305                                 partial_end = 0;
306                         }
307                         wait_on_page_writeback(page);
308                         zero_user_segment(page, partial_start, top);
309                         cleancache_invalidate_page(mapping, page);
310                         if (page_has_private(page))
311                                 do_invalidatepage(page, partial_start,
312                                                   top - partial_start);
313                         unlock_page(page);
314                         page_cache_release(page);
315                 }
316         }
317         if (partial_end) {
318                 struct page *page = find_lock_page(mapping, end);
319                 if (page) {
320                         wait_on_page_writeback(page);
321                         zero_user_segment(page, 0, partial_end);
322                         cleancache_invalidate_page(mapping, page);
323                         if (page_has_private(page))
324                                 do_invalidatepage(page, 0,
325                                                   partial_end);
326                         unlock_page(page);
327                         page_cache_release(page);
328                 }
329         }
330         /*
331          * If the truncation happened within a single page no pages
332          * will be released, just zeroed, so we can bail out now.
333          */
334         if (start >= end)
335                 return;
336 
337         index = start;
338         for ( ; ; ) {
339                 cond_resched();
340                 if (!pagevec_lookup_entries(&pvec, mapping, index,
341                         min(end - index, (pgoff_t)PAGEVEC_SIZE), indices)) {
342                         /* If all gone from start onwards, we're done */
343                         if (index == start)
344                                 break;
345                         /* Otherwise restart to make sure all gone */
346                         index = start;
347                         continue;
348                 }
349                 if (index == start && indices[0] >= end) {
350                         /* All gone out of hole to be punched, we're done */
351                         pagevec_remove_exceptionals(&pvec);
352                         pagevec_release(&pvec);
353                         break;
354                 }
355                 for (i = 0; i < pagevec_count(&pvec); i++) {
356                         struct page *page = pvec.pages[i];
357 
358                         /* We rely upon deletion not changing page->index */
359                         index = indices[i];
360                         if (index >= end) {
361                                 /* Restart punch to make sure all gone */
362                                 index = start - 1;
363                                 break;
364                         }
365 
366                         if (radix_tree_exceptional_entry(page)) {
367                                 clear_exceptional_entry(mapping, index, page);
368                                 continue;
369                         }
370 
371                         lock_page(page);
372                         WARN_ON(page->index != index);
373                         wait_on_page_writeback(page);
374                         truncate_inode_page(mapping, page);
375                         unlock_page(page);
376                 }
377                 pagevec_remove_exceptionals(&pvec);
378                 pagevec_release(&pvec);
379                 index++;
380         }
381         cleancache_invalidate_inode(mapping);
382 }
383 EXPORT_SYMBOL(truncate_inode_pages_range);
384 
385 /**
386  * truncate_inode_pages - truncate *all* the pages from an offset
387  * @mapping: mapping to truncate
388  * @lstart: offset from which to truncate
389  *
390  * Called under (and serialised by) inode->i_mutex.
391  *
392  * Note: When this function returns, there can be a page in the process of
393  * deletion (inside __delete_from_page_cache()) in the specified range.  Thus
394  * mapping->nrpages can be non-zero when this function returns even after
395  * truncation of the whole mapping.
396  */
397 void truncate_inode_pages(struct address_space *mapping, loff_t lstart)
398 {
399         truncate_inode_pages_range(mapping, lstart, (loff_t)-1);
400 }
401 EXPORT_SYMBOL(truncate_inode_pages);
402 
403 /**
404  * truncate_inode_pages_final - truncate *all* pages before inode dies
405  * @mapping: mapping to truncate
406  *
407  * Called under (and serialized by) inode->i_mutex.
408  *
409  * Filesystems have to use this in the .evict_inode path to inform the
410  * VM that this is the final truncate and the inode is going away.
411  */
412 void truncate_inode_pages_final(struct address_space *mapping)
413 {
414         unsigned long nrexceptional;
415         unsigned long nrpages;
416 
417         /*
418          * Page reclaim can not participate in regular inode lifetime
419          * management (can't call iput()) and thus can race with the
420          * inode teardown.  Tell it when the address space is exiting,
421          * so that it does not install eviction information after the
422          * final truncate has begun.
423          */
424         mapping_set_exiting(mapping);
425 
426         /*
427          * When reclaim installs eviction entries, it increases
428          * nrexceptional first, then decreases nrpages.  Make sure we see
429          * this in the right order or we might miss an entry.
430          */
431         nrpages = mapping->nrpages;
432         smp_rmb();
433         nrexceptional = mapping->nrexceptional;
434 
435         if (nrpages || nrexceptional) {
436                 /*
437                  * As truncation uses a lockless tree lookup, cycle
438                  * the tree lock to make sure any ongoing tree
439                  * modification that does not see AS_EXITING is
440                  * completed before starting the final truncate.
441                  */
442                 spin_lock_irq(&mapping->tree_lock);
443                 spin_unlock_irq(&mapping->tree_lock);
444 
445                 truncate_inode_pages(mapping, 0);
446         }
447 }
448 EXPORT_SYMBOL(truncate_inode_pages_final);
449 
450 /**
451  * invalidate_mapping_pages - Invalidate all the unlocked pages of one inode
452  * @mapping: the address_space which holds the pages to invalidate
453  * @start: the offset 'from' which to invalidate
454  * @end: the offset 'to' which to invalidate (inclusive)
455  *
456  * This function only removes the unlocked pages, if you want to
457  * remove all the pages of one inode, you must call truncate_inode_pages.
458  *
459  * invalidate_mapping_pages() will not block on IO activity. It will not
460  * invalidate pages which are dirty, locked, under writeback or mapped into
461  * pagetables.
462  */
463 unsigned long invalidate_mapping_pages(struct address_space *mapping,
464                 pgoff_t start, pgoff_t end)
465 {
466         pgoff_t indices[PAGEVEC_SIZE];
467         struct pagevec pvec;
468         pgoff_t index = start;
469         unsigned long ret;
470         unsigned long count = 0;
471         int i;
472 
473         pagevec_init(&pvec, 0);
474         while (index <= end && pagevec_lookup_entries(&pvec, mapping, index,
475                         min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1,
476                         indices)) {
477                 for (i = 0; i < pagevec_count(&pvec); i++) {
478                         struct page *page = pvec.pages[i];
479 
480                         /* We rely upon deletion not changing page->index */
481                         index = indices[i];
482                         if (index > end)
483                                 break;
484 
485                         if (radix_tree_exceptional_entry(page)) {
486                                 clear_exceptional_entry(mapping, index, page);
487                                 continue;
488                         }
489 
490                         if (!trylock_page(page))
491                                 continue;
492                         WARN_ON(page->index != index);
493                         ret = invalidate_inode_page(page);
494                         unlock_page(page);
495                         /*
496                          * Invalidation is a hint that the page is no longer
497                          * of interest and try to speed up its reclaim.
498                          */
499                         if (!ret)
500                                 deactivate_file_page(page);
501                         count += ret;
502                 }
503                 pagevec_remove_exceptionals(&pvec);
504                 pagevec_release(&pvec);
505                 cond_resched();
506                 index++;
507         }
508         return count;
509 }
510 EXPORT_SYMBOL(invalidate_mapping_pages);
511 
512 /*
513  * This is like invalidate_complete_page(), except it ignores the page's
514  * refcount.  We do this because invalidate_inode_pages2() needs stronger
515  * invalidation guarantees, and cannot afford to leave pages behind because
516  * shrink_page_list() has a temp ref on them, or because they're transiently
517  * sitting in the lru_cache_add() pagevecs.
518  */
519 static int
520 invalidate_complete_page2(struct address_space *mapping, struct page *page)
521 {
522         struct mem_cgroup *memcg;
523         unsigned long flags;
524 
525         if (page->mapping != mapping)
526                 return 0;
527 
528         if (page_has_private(page) && !try_to_release_page(page, GFP_KERNEL))
529                 return 0;
530 
531         memcg = mem_cgroup_begin_page_stat(page);
532         spin_lock_irqsave(&mapping->tree_lock, flags);
533         if (PageDirty(page))
534                 goto failed;
535 
536         BUG_ON(page_has_private(page));
537         __delete_from_page_cache(page, NULL, memcg);
538         spin_unlock_irqrestore(&mapping->tree_lock, flags);
539         mem_cgroup_end_page_stat(memcg);
540 
541         if (mapping->a_ops->freepage)
542                 mapping->a_ops->freepage(page);
543 
544         page_cache_release(page);       /* pagecache ref */
545         return 1;
546 failed:
547         spin_unlock_irqrestore(&mapping->tree_lock, flags);
548         mem_cgroup_end_page_stat(memcg);
549         return 0;
550 }
551 
552 static int do_launder_page(struct address_space *mapping, struct page *page)
553 {
554         if (!PageDirty(page))
555                 return 0;
556         if (page->mapping != mapping || mapping->a_ops->launder_page == NULL)
557                 return 0;
558         return mapping->a_ops->launder_page(page);
559 }
560 
561 /**
562  * invalidate_inode_pages2_range - remove range of pages from an address_space
563  * @mapping: the address_space
564  * @start: the page offset 'from' which to invalidate
565  * @end: the page offset 'to' which to invalidate (inclusive)
566  *
567  * Any pages which are found to be mapped into pagetables are unmapped prior to
568  * invalidation.
569  *
570  * Returns -EBUSY if any pages could not be invalidated.
571  */
572 int invalidate_inode_pages2_range(struct address_space *mapping,
573                                   pgoff_t start, pgoff_t end)
574 {
575         pgoff_t indices[PAGEVEC_SIZE];
576         struct pagevec pvec;
577         pgoff_t index;
578         int i;
579         int ret = 0;
580         int ret2 = 0;
581         int did_range_unmap = 0;
582 
583         cleancache_invalidate_inode(mapping);
584         pagevec_init(&pvec, 0);
585         index = start;
586         while (index <= end && pagevec_lookup_entries(&pvec, mapping, index,
587                         min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1,
588                         indices)) {
589                 for (i = 0; i < pagevec_count(&pvec); i++) {
590                         struct page *page = pvec.pages[i];
591 
592                         /* We rely upon deletion not changing page->index */
593                         index = indices[i];
594                         if (index > end)
595                                 break;
596 
597                         if (radix_tree_exceptional_entry(page)) {
598                                 clear_exceptional_entry(mapping, index, page);
599                                 continue;
600                         }
601 
602                         lock_page(page);
603                         WARN_ON(page->index != index);
604                         if (page->mapping != mapping) {
605                                 unlock_page(page);
606                                 continue;
607                         }
608                         wait_on_page_writeback(page);
609                         if (page_mapped(page)) {
610                                 if (!did_range_unmap) {
611                                         /*
612                                          * Zap the rest of the file in one hit.
613                                          */
614                                         unmap_mapping_range(mapping,
615                                            (loff_t)index << PAGE_CACHE_SHIFT,
616                                            (loff_t)(1 + end - index)
617                                                          << PAGE_CACHE_SHIFT,
618                                             0);
619                                         did_range_unmap = 1;
620                                 } else {
621                                         /*
622                                          * Just zap this page
623                                          */
624                                         unmap_mapping_range(mapping,
625                                            (loff_t)index << PAGE_CACHE_SHIFT,
626                                            PAGE_CACHE_SIZE, 0);
627                                 }
628                         }
629                         BUG_ON(page_mapped(page));
630                         ret2 = do_launder_page(mapping, page);
631                         if (ret2 == 0) {
632                                 if (!invalidate_complete_page2(mapping, page))
633                                         ret2 = -EBUSY;
634                         }
635                         if (ret2 < 0)
636                                 ret = ret2;
637                         unlock_page(page);
638                 }
639                 pagevec_remove_exceptionals(&pvec);
640                 pagevec_release(&pvec);
641                 cond_resched();
642                 index++;
643         }
644         cleancache_invalidate_inode(mapping);
645         return ret;
646 }
647 EXPORT_SYMBOL_GPL(invalidate_inode_pages2_range);
648 
649 /**
650  * invalidate_inode_pages2 - remove all pages from an address_space
651  * @mapping: the address_space
652  *
653  * Any pages which are found to be mapped into pagetables are unmapped prior to
654  * invalidation.
655  *
656  * Returns -EBUSY if any pages could not be invalidated.
657  */
658 int invalidate_inode_pages2(struct address_space *mapping)
659 {
660         return invalidate_inode_pages2_range(mapping, 0, -1);
661 }
662 EXPORT_SYMBOL_GPL(invalidate_inode_pages2);
663 
664 /**
665  * truncate_pagecache - unmap and remove pagecache that has been truncated
666  * @inode: inode
667  * @newsize: new file size
668  *
669  * inode's new i_size must already be written before truncate_pagecache
670  * is called.
671  *
672  * This function should typically be called before the filesystem
673  * releases resources associated with the freed range (eg. deallocates
674  * blocks). This way, pagecache will always stay logically coherent
675  * with on-disk format, and the filesystem would not have to deal with
676  * situations such as writepage being called for a page that has already
677  * had its underlying blocks deallocated.
678  */
679 void truncate_pagecache(struct inode *inode, loff_t newsize)
680 {
681         struct address_space *mapping = inode->i_mapping;
682         loff_t holebegin = round_up(newsize, PAGE_SIZE);
683 
684         /*
685          * unmap_mapping_range is called twice, first simply for
686          * efficiency so that truncate_inode_pages does fewer
687          * single-page unmaps.  However after this first call, and
688          * before truncate_inode_pages finishes, it is possible for
689          * private pages to be COWed, which remain after
690          * truncate_inode_pages finishes, hence the second
691          * unmap_mapping_range call must be made for correctness.
692          */
693         unmap_mapping_range(mapping, holebegin, 0, 1);
694         truncate_inode_pages(mapping, newsize);
695         unmap_mapping_range(mapping, holebegin, 0, 1);
696 }
697 EXPORT_SYMBOL(truncate_pagecache);
698 
699 /**
700  * truncate_setsize - update inode and pagecache for a new file size
701  * @inode: inode
702  * @newsize: new file size
703  *
704  * truncate_setsize updates i_size and performs pagecache truncation (if
705  * necessary) to @newsize. It will be typically be called from the filesystem's
706  * setattr function when ATTR_SIZE is passed in.
707  *
708  * Must be called with a lock serializing truncates and writes (generally
709  * i_mutex but e.g. xfs uses a different lock) and before all filesystem
710  * specific block truncation has been performed.
711  */
712 void truncate_setsize(struct inode *inode, loff_t newsize)
713 {
714         loff_t oldsize = inode->i_size;
715 
716         i_size_write(inode, newsize);
717         if (newsize > oldsize)
718                 pagecache_isize_extended(inode, oldsize, newsize);
719         truncate_pagecache(inode, newsize);
720 }
721 EXPORT_SYMBOL(truncate_setsize);
722 
723 /**
724  * pagecache_isize_extended - update pagecache after extension of i_size
725  * @inode:      inode for which i_size was extended
726  * @from:       original inode size
727  * @to:         new inode size
728  *
729  * Handle extension of inode size either caused by extending truncate or by
730  * write starting after current i_size. We mark the page straddling current
731  * i_size RO so that page_mkwrite() is called on the nearest write access to
732  * the page.  This way filesystem can be sure that page_mkwrite() is called on
733  * the page before user writes to the page via mmap after the i_size has been
734  * changed.
735  *
736  * The function must be called after i_size is updated so that page fault
737  * coming after we unlock the page will already see the new i_size.
738  * The function must be called while we still hold i_mutex - this not only
739  * makes sure i_size is stable but also that userspace cannot observe new
740  * i_size value before we are prepared to store mmap writes at new inode size.
741  */
742 void pagecache_isize_extended(struct inode *inode, loff_t from, loff_t to)
743 {
744         int bsize = 1 << inode->i_blkbits;
745         loff_t rounded_from;
746         struct page *page;
747         pgoff_t index;
748 
749         WARN_ON(to > inode->i_size);
750 
751         if (from >= to || bsize == PAGE_CACHE_SIZE)
752                 return;
753         /* Page straddling @from will not have any hole block created? */
754         rounded_from = round_up(from, bsize);
755         if (to <= rounded_from || !(rounded_from & (PAGE_CACHE_SIZE - 1)))
756                 return;
757 
758         index = from >> PAGE_CACHE_SHIFT;
759         page = find_lock_page(inode->i_mapping, index);
760         /* Page not cached? Nothing to do */
761         if (!page)
762                 return;
763         /*
764          * See clear_page_dirty_for_io() for details why set_page_dirty()
765          * is needed.
766          */
767         if (page_mkclean(page))
768                 set_page_dirty(page);
769         unlock_page(page);
770         page_cache_release(page);
771 }
772 EXPORT_SYMBOL(pagecache_isize_extended);
773 
774 /**
775  * truncate_pagecache_range - unmap and remove pagecache that is hole-punched
776  * @inode: inode
777  * @lstart: offset of beginning of hole
778  * @lend: offset of last byte of hole
779  *
780  * This function should typically be called before the filesystem
781  * releases resources associated with the freed range (eg. deallocates
782  * blocks). This way, pagecache will always stay logically coherent
783  * with on-disk format, and the filesystem would not have to deal with
784  * situations such as writepage being called for a page that has already
785  * had its underlying blocks deallocated.
786  */
787 void truncate_pagecache_range(struct inode *inode, loff_t lstart, loff_t lend)
788 {
789         struct address_space *mapping = inode->i_mapping;
790         loff_t unmap_start = round_up(lstart, PAGE_SIZE);
791         loff_t unmap_end = round_down(1 + lend, PAGE_SIZE) - 1;
792         /*
793          * This rounding is currently just for example: unmap_mapping_range
794          * expands its hole outwards, whereas we want it to contract the hole
795          * inwards.  However, existing callers of truncate_pagecache_range are
796          * doing their own page rounding first.  Note that unmap_mapping_range
797          * allows holelen 0 for all, and we allow lend -1 for end of file.
798          */
799 
800         /*
801          * Unlike in truncate_pagecache, unmap_mapping_range is called only
802          * once (before truncating pagecache), and without "even_cows" flag:
803          * hole-punching should not remove private COWed pages from the hole.
804          */
805         if ((u64)unmap_end > (u64)unmap_start)
806                 unmap_mapping_range(mapping, unmap_start,
807                                     1 + unmap_end - unmap_start, 0);
808         truncate_inode_pages_range(mapping, lstart, lend);
809 }
810 EXPORT_SYMBOL(truncate_pagecache_range);
811 

~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

kernel.org | git.kernel.org | LWN.net | Project Home | Wiki (Japanese) | Wiki (English) | SVN repository | Mail admin

Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.

osdn.jp