xref: /openbmc/linux/mm/truncate.c (revision 9d56dd3b083a3bec56e9da35ce07baca81030b03)
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/mm.h>
13 #include <linux/swap.h>
14 #include <linux/module.h>
15 #include <linux/pagemap.h>
16 #include <linux/highmem.h>
17 #include <linux/pagevec.h>
18 #include <linux/task_io_accounting_ops.h>
19 #include <linux/buffer_head.h>	/* grr. try_to_release_page,
20 				   do_invalidatepage */
21 #include "internal.h"
22 
23 
24 /**
25  * do_invalidatepage - invalidate part or all of a page
26  * @page: the page which is affected
27  * @offset: the index of the truncation point
28  *
29  * do_invalidatepage() is called when all or part of the page has become
30  * invalidated by a truncate operation.
31  *
32  * do_invalidatepage() does not have to release all buffers, but it must
33  * ensure that no dirty buffer is left outside @offset and that no I/O
34  * is underway against any of the blocks which are outside the truncation
35  * point.  Because the caller is about to free (and possibly reuse) those
36  * blocks on-disk.
37  */
38 void do_invalidatepage(struct page *page, unsigned long offset)
39 {
40 	void (*invalidatepage)(struct page *, unsigned long);
41 	invalidatepage = page->mapping->a_ops->invalidatepage;
42 #ifdef CONFIG_BLOCK
43 	if (!invalidatepage)
44 		invalidatepage = block_invalidatepage;
45 #endif
46 	if (invalidatepage)
47 		(*invalidatepage)(page, offset);
48 }
49 
50 static inline void truncate_partial_page(struct page *page, unsigned partial)
51 {
52 	zero_user_segment(page, partial, PAGE_CACHE_SIZE);
53 	if (page_has_private(page))
54 		do_invalidatepage(page, partial);
55 }
56 
57 /*
58  * This cancels just the dirty bit on the kernel page itself, it
59  * does NOT actually remove dirty bits on any mmap's that may be
60  * around. It also leaves the page tagged dirty, so any sync
61  * activity will still find it on the dirty lists, and in particular,
62  * clear_page_dirty_for_io() will still look at the dirty bits in
63  * the VM.
64  *
65  * Doing this should *normally* only ever be done when a page
66  * is truncated, and is not actually mapped anywhere at all. However,
67  * fs/buffer.c does this when it notices that somebody has cleaned
68  * out all the buffers on a page without actually doing it through
69  * the VM. Can you say "ext3 is horribly ugly"? Tought you could.
70  */
71 void cancel_dirty_page(struct page *page, unsigned int account_size)
72 {
73 	if (TestClearPageDirty(page)) {
74 		struct address_space *mapping = page->mapping;
75 		if (mapping && mapping_cap_account_dirty(mapping)) {
76 			dec_zone_page_state(page, NR_FILE_DIRTY);
77 			dec_bdi_stat(mapping->backing_dev_info,
78 					BDI_RECLAIMABLE);
79 			if (account_size)
80 				task_io_account_cancelled_write(account_size);
81 		}
82 	}
83 }
84 EXPORT_SYMBOL(cancel_dirty_page);
85 
86 /*
87  * If truncate cannot remove the fs-private metadata from the page, the page
88  * becomes orphaned.  It will be left on the LRU and may even be mapped into
89  * user pagetables if we're racing with filemap_fault().
90  *
91  * We need to bale out if page->mapping is no longer equal to the original
92  * mapping.  This happens a) when the VM reclaimed the page while we waited on
93  * its lock, b) when a concurrent invalidate_mapping_pages got there first and
94  * c) when tmpfs swizzles a page between a tmpfs inode and swapper_space.
95  */
96 static int
97 truncate_complete_page(struct address_space *mapping, struct page *page)
98 {
99 	if (page->mapping != mapping)
100 		return -EIO;
101 
102 	if (page_has_private(page))
103 		do_invalidatepage(page, 0);
104 
105 	cancel_dirty_page(page, PAGE_CACHE_SIZE);
106 
107 	clear_page_mlock(page);
108 	remove_from_page_cache(page);
109 	ClearPageMappedToDisk(page);
110 	page_cache_release(page);	/* pagecache ref */
111 	return 0;
112 }
113 
114 /*
115  * This is for invalidate_mapping_pages().  That function can be called at
116  * any time, and is not supposed to throw away dirty pages.  But pages can
117  * be marked dirty at any time too, so use remove_mapping which safely
118  * discards clean, unused pages.
119  *
120  * Returns non-zero if the page was successfully invalidated.
121  */
122 static int
123 invalidate_complete_page(struct address_space *mapping, struct page *page)
124 {
125 	int ret;
126 
127 	if (page->mapping != mapping)
128 		return 0;
129 
130 	if (page_has_private(page) && !try_to_release_page(page, 0))
131 		return 0;
132 
133 	clear_page_mlock(page);
134 	ret = remove_mapping(mapping, page);
135 
136 	return ret;
137 }
138 
139 int truncate_inode_page(struct address_space *mapping, struct page *page)
140 {
141 	if (page_mapped(page)) {
142 		unmap_mapping_range(mapping,
143 				   (loff_t)page->index << PAGE_CACHE_SHIFT,
144 				   PAGE_CACHE_SIZE, 0);
145 	}
146 	return truncate_complete_page(mapping, page);
147 }
148 
149 /*
150  * Used to get rid of pages on hardware memory corruption.
151  */
152 int generic_error_remove_page(struct address_space *mapping, struct page *page)
153 {
154 	if (!mapping)
155 		return -EINVAL;
156 	/*
157 	 * Only punch for normal data pages for now.
158 	 * Handling other types like directories would need more auditing.
159 	 */
160 	if (!S_ISREG(mapping->host->i_mode))
161 		return -EIO;
162 	return truncate_inode_page(mapping, page);
163 }
164 EXPORT_SYMBOL(generic_error_remove_page);
165 
166 /*
167  * Safely invalidate one page from its pagecache mapping.
168  * It only drops clean, unused pages. The page must be locked.
169  *
170  * Returns 1 if the page is successfully invalidated, otherwise 0.
171  */
172 int invalidate_inode_page(struct page *page)
173 {
174 	struct address_space *mapping = page_mapping(page);
175 	if (!mapping)
176 		return 0;
177 	if (PageDirty(page) || PageWriteback(page))
178 		return 0;
179 	if (page_mapped(page))
180 		return 0;
181 	return invalidate_complete_page(mapping, page);
182 }
183 
184 /**
185  * truncate_inode_pages - truncate range of pages specified by start & end byte offsets
186  * @mapping: mapping to truncate
187  * @lstart: offset from which to truncate
188  * @lend: offset to which to truncate
189  *
190  * Truncate the page cache, removing the pages that are between
191  * specified offsets (and zeroing out partial page
192  * (if lstart is not page aligned)).
193  *
194  * Truncate takes two passes - the first pass is nonblocking.  It will not
195  * block on page locks and it will not block on writeback.  The second pass
196  * will wait.  This is to prevent as much IO as possible in the affected region.
197  * The first pass will remove most pages, so the search cost of the second pass
198  * is low.
199  *
200  * When looking at page->index outside the page lock we need to be careful to
201  * copy it into a local to avoid races (it could change at any time).
202  *
203  * We pass down the cache-hot hint to the page freeing code.  Even if the
204  * mapping is large, it is probably the case that the final pages are the most
205  * recently touched, and freeing happens in ascending file offset order.
206  */
207 void truncate_inode_pages_range(struct address_space *mapping,
208 				loff_t lstart, loff_t lend)
209 {
210 	const pgoff_t start = (lstart + PAGE_CACHE_SIZE-1) >> PAGE_CACHE_SHIFT;
211 	pgoff_t end;
212 	const unsigned partial = lstart & (PAGE_CACHE_SIZE - 1);
213 	struct pagevec pvec;
214 	pgoff_t next;
215 	int i;
216 
217 	if (mapping->nrpages == 0)
218 		return;
219 
220 	BUG_ON((lend & (PAGE_CACHE_SIZE - 1)) != (PAGE_CACHE_SIZE - 1));
221 	end = (lend >> PAGE_CACHE_SHIFT);
222 
223 	pagevec_init(&pvec, 0);
224 	next = start;
225 	while (next <= end &&
226 	       pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) {
227 		for (i = 0; i < pagevec_count(&pvec); i++) {
228 			struct page *page = pvec.pages[i];
229 			pgoff_t page_index = page->index;
230 
231 			if (page_index > end) {
232 				next = page_index;
233 				break;
234 			}
235 
236 			if (page_index > next)
237 				next = page_index;
238 			next++;
239 			if (!trylock_page(page))
240 				continue;
241 			if (PageWriteback(page)) {
242 				unlock_page(page);
243 				continue;
244 			}
245 			truncate_inode_page(mapping, page);
246 			unlock_page(page);
247 		}
248 		pagevec_release(&pvec);
249 		cond_resched();
250 	}
251 
252 	if (partial) {
253 		struct page *page = find_lock_page(mapping, start - 1);
254 		if (page) {
255 			wait_on_page_writeback(page);
256 			truncate_partial_page(page, partial);
257 			unlock_page(page);
258 			page_cache_release(page);
259 		}
260 	}
261 
262 	next = start;
263 	for ( ; ; ) {
264 		cond_resched();
265 		if (!pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) {
266 			if (next == start)
267 				break;
268 			next = start;
269 			continue;
270 		}
271 		if (pvec.pages[0]->index > end) {
272 			pagevec_release(&pvec);
273 			break;
274 		}
275 		mem_cgroup_uncharge_start();
276 		for (i = 0; i < pagevec_count(&pvec); i++) {
277 			struct page *page = pvec.pages[i];
278 
279 			if (page->index > end)
280 				break;
281 			lock_page(page);
282 			wait_on_page_writeback(page);
283 			truncate_inode_page(mapping, page);
284 			if (page->index > next)
285 				next = page->index;
286 			next++;
287 			unlock_page(page);
288 		}
289 		pagevec_release(&pvec);
290 		mem_cgroup_uncharge_end();
291 	}
292 }
293 EXPORT_SYMBOL(truncate_inode_pages_range);
294 
295 /**
296  * truncate_inode_pages - truncate *all* the pages from an offset
297  * @mapping: mapping to truncate
298  * @lstart: offset from which to truncate
299  *
300  * Called under (and serialised by) inode->i_mutex.
301  */
302 void truncate_inode_pages(struct address_space *mapping, loff_t lstart)
303 {
304 	truncate_inode_pages_range(mapping, lstart, (loff_t)-1);
305 }
306 EXPORT_SYMBOL(truncate_inode_pages);
307 
308 /**
309  * invalidate_mapping_pages - Invalidate all the unlocked pages of one inode
310  * @mapping: the address_space which holds the pages to invalidate
311  * @start: the offset 'from' which to invalidate
312  * @end: the offset 'to' which to invalidate (inclusive)
313  *
314  * This function only removes the unlocked pages, if you want to
315  * remove all the pages of one inode, you must call truncate_inode_pages.
316  *
317  * invalidate_mapping_pages() will not block on IO activity. It will not
318  * invalidate pages which are dirty, locked, under writeback or mapped into
319  * pagetables.
320  */
321 unsigned long invalidate_mapping_pages(struct address_space *mapping,
322 				       pgoff_t start, pgoff_t end)
323 {
324 	struct pagevec pvec;
325 	pgoff_t next = start;
326 	unsigned long ret = 0;
327 	int i;
328 
329 	pagevec_init(&pvec, 0);
330 	while (next <= end &&
331 			pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) {
332 		mem_cgroup_uncharge_start();
333 		for (i = 0; i < pagevec_count(&pvec); i++) {
334 			struct page *page = pvec.pages[i];
335 			pgoff_t index;
336 			int lock_failed;
337 
338 			lock_failed = !trylock_page(page);
339 
340 			/*
341 			 * We really shouldn't be looking at the ->index of an
342 			 * unlocked page.  But we're not allowed to lock these
343 			 * pages.  So we rely upon nobody altering the ->index
344 			 * of this (pinned-by-us) page.
345 			 */
346 			index = page->index;
347 			if (index > next)
348 				next = index;
349 			next++;
350 			if (lock_failed)
351 				continue;
352 
353 			ret += invalidate_inode_page(page);
354 
355 			unlock_page(page);
356 			if (next > end)
357 				break;
358 		}
359 		pagevec_release(&pvec);
360 		mem_cgroup_uncharge_end();
361 		cond_resched();
362 	}
363 	return ret;
364 }
365 EXPORT_SYMBOL(invalidate_mapping_pages);
366 
367 /*
368  * This is like invalidate_complete_page(), except it ignores the page's
369  * refcount.  We do this because invalidate_inode_pages2() needs stronger
370  * invalidation guarantees, and cannot afford to leave pages behind because
371  * shrink_page_list() has a temp ref on them, or because they're transiently
372  * sitting in the lru_cache_add() pagevecs.
373  */
374 static int
375 invalidate_complete_page2(struct address_space *mapping, struct page *page)
376 {
377 	if (page->mapping != mapping)
378 		return 0;
379 
380 	if (page_has_private(page) && !try_to_release_page(page, GFP_KERNEL))
381 		return 0;
382 
383 	spin_lock_irq(&mapping->tree_lock);
384 	if (PageDirty(page))
385 		goto failed;
386 
387 	clear_page_mlock(page);
388 	BUG_ON(page_has_private(page));
389 	__remove_from_page_cache(page);
390 	spin_unlock_irq(&mapping->tree_lock);
391 	mem_cgroup_uncharge_cache_page(page);
392 	page_cache_release(page);	/* pagecache ref */
393 	return 1;
394 failed:
395 	spin_unlock_irq(&mapping->tree_lock);
396 	return 0;
397 }
398 
399 static int do_launder_page(struct address_space *mapping, struct page *page)
400 {
401 	if (!PageDirty(page))
402 		return 0;
403 	if (page->mapping != mapping || mapping->a_ops->launder_page == NULL)
404 		return 0;
405 	return mapping->a_ops->launder_page(page);
406 }
407 
408 /**
409  * invalidate_inode_pages2_range - remove range of pages from an address_space
410  * @mapping: the address_space
411  * @start: the page offset 'from' which to invalidate
412  * @end: the page offset 'to' which to invalidate (inclusive)
413  *
414  * Any pages which are found to be mapped into pagetables are unmapped prior to
415  * invalidation.
416  *
417  * Returns -EBUSY if any pages could not be invalidated.
418  */
419 int invalidate_inode_pages2_range(struct address_space *mapping,
420 				  pgoff_t start, pgoff_t end)
421 {
422 	struct pagevec pvec;
423 	pgoff_t next;
424 	int i;
425 	int ret = 0;
426 	int ret2 = 0;
427 	int did_range_unmap = 0;
428 	int wrapped = 0;
429 
430 	pagevec_init(&pvec, 0);
431 	next = start;
432 	while (next <= end && !wrapped &&
433 		pagevec_lookup(&pvec, mapping, next,
434 			min(end - next, (pgoff_t)PAGEVEC_SIZE - 1) + 1)) {
435 		mem_cgroup_uncharge_start();
436 		for (i = 0; i < pagevec_count(&pvec); i++) {
437 			struct page *page = pvec.pages[i];
438 			pgoff_t page_index;
439 
440 			lock_page(page);
441 			if (page->mapping != mapping) {
442 				unlock_page(page);
443 				continue;
444 			}
445 			page_index = page->index;
446 			next = page_index + 1;
447 			if (next == 0)
448 				wrapped = 1;
449 			if (page_index > end) {
450 				unlock_page(page);
451 				break;
452 			}
453 			wait_on_page_writeback(page);
454 			if (page_mapped(page)) {
455 				if (!did_range_unmap) {
456 					/*
457 					 * Zap the rest of the file in one hit.
458 					 */
459 					unmap_mapping_range(mapping,
460 					   (loff_t)page_index<<PAGE_CACHE_SHIFT,
461 					   (loff_t)(end - page_index + 1)
462 							<< PAGE_CACHE_SHIFT,
463 					    0);
464 					did_range_unmap = 1;
465 				} else {
466 					/*
467 					 * Just zap this page
468 					 */
469 					unmap_mapping_range(mapping,
470 					  (loff_t)page_index<<PAGE_CACHE_SHIFT,
471 					  PAGE_CACHE_SIZE, 0);
472 				}
473 			}
474 			BUG_ON(page_mapped(page));
475 			ret2 = do_launder_page(mapping, page);
476 			if (ret2 == 0) {
477 				if (!invalidate_complete_page2(mapping, page))
478 					ret2 = -EBUSY;
479 			}
480 			if (ret2 < 0)
481 				ret = ret2;
482 			unlock_page(page);
483 		}
484 		pagevec_release(&pvec);
485 		mem_cgroup_uncharge_end();
486 		cond_resched();
487 	}
488 	return ret;
489 }
490 EXPORT_SYMBOL_GPL(invalidate_inode_pages2_range);
491 
492 /**
493  * invalidate_inode_pages2 - remove all pages from an address_space
494  * @mapping: the address_space
495  *
496  * Any pages which are found to be mapped into pagetables are unmapped prior to
497  * invalidation.
498  *
499  * Returns -EBUSY if any pages could not be invalidated.
500  */
501 int invalidate_inode_pages2(struct address_space *mapping)
502 {
503 	return invalidate_inode_pages2_range(mapping, 0, -1);
504 }
505 EXPORT_SYMBOL_GPL(invalidate_inode_pages2);
506 
507 /**
508  * truncate_pagecache - unmap and remove pagecache that has been truncated
509  * @inode: inode
510  * @old: old file offset
511  * @new: new file offset
512  *
513  * inode's new i_size must already be written before truncate_pagecache
514  * is called.
515  *
516  * This function should typically be called before the filesystem
517  * releases resources associated with the freed range (eg. deallocates
518  * blocks). This way, pagecache will always stay logically coherent
519  * with on-disk format, and the filesystem would not have to deal with
520  * situations such as writepage being called for a page that has already
521  * had its underlying blocks deallocated.
522  */
523 void truncate_pagecache(struct inode *inode, loff_t old, loff_t new)
524 {
525 	struct address_space *mapping = inode->i_mapping;
526 
527 	/*
528 	 * unmap_mapping_range is called twice, first simply for
529 	 * efficiency so that truncate_inode_pages does fewer
530 	 * single-page unmaps.  However after this first call, and
531 	 * before truncate_inode_pages finishes, it is possible for
532 	 * private pages to be COWed, which remain after
533 	 * truncate_inode_pages finishes, hence the second
534 	 * unmap_mapping_range call must be made for correctness.
535 	 */
536 	unmap_mapping_range(mapping, new + PAGE_SIZE - 1, 0, 1);
537 	truncate_inode_pages(mapping, new);
538 	unmap_mapping_range(mapping, new + PAGE_SIZE - 1, 0, 1);
539 }
540 EXPORT_SYMBOL(truncate_pagecache);
541 
542 /**
543  * vmtruncate - unmap mappings "freed" by truncate() syscall
544  * @inode: inode of the file used
545  * @offset: file offset to start truncating
546  *
547  * NOTE! We have to be ready to update the memory sharing
548  * between the file and the memory map for a potential last
549  * incomplete page.  Ugly, but necessary.
550  */
551 int vmtruncate(struct inode *inode, loff_t offset)
552 {
553 	loff_t oldsize;
554 	int error;
555 
556 	error = inode_newsize_ok(inode, offset);
557 	if (error)
558 		return error;
559 	oldsize = inode->i_size;
560 	i_size_write(inode, offset);
561 	truncate_pagecache(inode, oldsize, offset);
562 	if (inode->i_op->truncate)
563 		inode->i_op->truncate(inode);
564 
565 	return error;
566 }
567 EXPORT_SYMBOL(vmtruncate);
568