xref: /openbmc/linux/mm/truncate.c (revision e868d61272caa648214046a096e5a6bfc068dc8c)
1 /*
2  * mm/truncate.c - code for taking down pages from address_spaces
3  *
4  * Copyright (C) 2002, Linus Torvalds
5  *
6  * 10Sep2002	akpm@zip.com.au
7  *		Initial version.
8  */
9 
10 #include <linux/kernel.h>
11 #include <linux/mm.h>
12 #include <linux/swap.h>
13 #include <linux/module.h>
14 #include <linux/pagemap.h>
15 #include <linux/highmem.h>
16 #include <linux/pagevec.h>
17 #include <linux/task_io_accounting_ops.h>
18 #include <linux/buffer_head.h>	/* grr. try_to_release_page,
19 				   do_invalidatepage */
20 
21 
22 /**
23  * do_invalidatepage - invalidate part of all of a page
24  * @page: the page which is affected
25  * @offset: the index of the truncation point
26  *
27  * do_invalidatepage() is called when all or part of the page has become
28  * invalidated by a truncate operation.
29  *
30  * do_invalidatepage() does not have to release all buffers, but it must
31  * ensure that no dirty buffer is left outside @offset and that no I/O
32  * is underway against any of the blocks which are outside the truncation
33  * point.  Because the caller is about to free (and possibly reuse) those
34  * blocks on-disk.
35  */
36 void do_invalidatepage(struct page *page, unsigned long offset)
37 {
38 	void (*invalidatepage)(struct page *, unsigned long);
39 	invalidatepage = page->mapping->a_ops->invalidatepage;
40 #ifdef CONFIG_BLOCK
41 	if (!invalidatepage)
42 		invalidatepage = block_invalidatepage;
43 #endif
44 	if (invalidatepage)
45 		(*invalidatepage)(page, offset);
46 }
47 
48 static inline void truncate_partial_page(struct page *page, unsigned partial)
49 {
50 	zero_user_page(page, partial, PAGE_CACHE_SIZE - partial, KM_USER0);
51 	if (PagePrivate(page))
52 		do_invalidatepage(page, partial);
53 }
54 
55 /*
56  * This cancels just the dirty bit on the kernel page itself, it
57  * does NOT actually remove dirty bits on any mmap's that may be
58  * around. It also leaves the page tagged dirty, so any sync
59  * activity will still find it on the dirty lists, and in particular,
60  * clear_page_dirty_for_io() will still look at the dirty bits in
61  * the VM.
62  *
63  * Doing this should *normally* only ever be done when a page
64  * is truncated, and is not actually mapped anywhere at all. However,
65  * fs/buffer.c does this when it notices that somebody has cleaned
66  * out all the buffers on a page without actually doing it through
67  * the VM. Can you say "ext3 is horribly ugly"? Tought you could.
68  */
69 void cancel_dirty_page(struct page *page, unsigned int account_size)
70 {
71 	if (TestClearPageDirty(page)) {
72 		struct address_space *mapping = page->mapping;
73 		if (mapping && mapping_cap_account_dirty(mapping)) {
74 			dec_zone_page_state(page, NR_FILE_DIRTY);
75 			if (account_size)
76 				task_io_account_cancelled_write(account_size);
77 		}
78 	}
79 }
80 EXPORT_SYMBOL(cancel_dirty_page);
81 
82 /*
83  * If truncate cannot remove the fs-private metadata from the page, the page
84  * becomes anonymous.  It will be left on the LRU and may even be mapped into
85  * user pagetables if we're racing with filemap_nopage().
86  *
87  * We need to bale out if page->mapping is no longer equal to the original
88  * mapping.  This happens a) when the VM reclaimed the page while we waited on
89  * its lock, b) when a concurrent invalidate_mapping_pages got there first and
90  * c) when tmpfs swizzles a page between a tmpfs inode and swapper_space.
91  */
92 static void
93 truncate_complete_page(struct address_space *mapping, struct page *page)
94 {
95 	if (page->mapping != mapping)
96 		return;
97 
98 	cancel_dirty_page(page, PAGE_CACHE_SIZE);
99 
100 	if (PagePrivate(page))
101 		do_invalidatepage(page, 0);
102 
103 	ClearPageUptodate(page);
104 	ClearPageMappedToDisk(page);
105 	remove_from_page_cache(page);
106 	page_cache_release(page);	/* pagecache ref */
107 }
108 
109 /*
110  * This is for invalidate_mapping_pages().  That function can be called at
111  * any time, and is not supposed to throw away dirty pages.  But pages can
112  * be marked dirty at any time too, so use remove_mapping which safely
113  * discards clean, unused pages.
114  *
115  * Returns non-zero if the page was successfully invalidated.
116  */
117 static int
118 invalidate_complete_page(struct address_space *mapping, struct page *page)
119 {
120 	int ret;
121 
122 	if (page->mapping != mapping)
123 		return 0;
124 
125 	if (PagePrivate(page) && !try_to_release_page(page, 0))
126 		return 0;
127 
128 	ret = remove_mapping(mapping, page);
129 
130 	return ret;
131 }
132 
133 /**
134  * truncate_inode_pages - truncate range of pages specified by start and
135  * end byte offsets
136  * @mapping: mapping to truncate
137  * @lstart: offset from which to truncate
138  * @lend: offset to which to truncate
139  *
140  * Truncate the page cache, removing the pages that are between
141  * specified offsets (and zeroing out partial page
142  * (if lstart is not page aligned)).
143  *
144  * Truncate takes two passes - the first pass is nonblocking.  It will not
145  * block on page locks and it will not block on writeback.  The second pass
146  * will wait.  This is to prevent as much IO as possible in the affected region.
147  * The first pass will remove most pages, so the search cost of the second pass
148  * is low.
149  *
150  * When looking at page->index outside the page lock we need to be careful to
151  * copy it into a local to avoid races (it could change at any time).
152  *
153  * We pass down the cache-hot hint to the page freeing code.  Even if the
154  * mapping is large, it is probably the case that the final pages are the most
155  * recently touched, and freeing happens in ascending file offset order.
156  */
157 void truncate_inode_pages_range(struct address_space *mapping,
158 				loff_t lstart, loff_t lend)
159 {
160 	const pgoff_t start = (lstart + PAGE_CACHE_SIZE-1) >> PAGE_CACHE_SHIFT;
161 	pgoff_t end;
162 	const unsigned partial = lstart & (PAGE_CACHE_SIZE - 1);
163 	struct pagevec pvec;
164 	pgoff_t next;
165 	int i;
166 
167 	if (mapping->nrpages == 0)
168 		return;
169 
170 	BUG_ON((lend & (PAGE_CACHE_SIZE - 1)) != (PAGE_CACHE_SIZE - 1));
171 	end = (lend >> PAGE_CACHE_SHIFT);
172 
173 	pagevec_init(&pvec, 0);
174 	next = start;
175 	while (next <= end &&
176 	       pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) {
177 		for (i = 0; i < pagevec_count(&pvec); i++) {
178 			struct page *page = pvec.pages[i];
179 			pgoff_t page_index = page->index;
180 
181 			if (page_index > end) {
182 				next = page_index;
183 				break;
184 			}
185 
186 			if (page_index > next)
187 				next = page_index;
188 			next++;
189 			if (TestSetPageLocked(page))
190 				continue;
191 			if (PageWriteback(page)) {
192 				unlock_page(page);
193 				continue;
194 			}
195 			truncate_complete_page(mapping, page);
196 			unlock_page(page);
197 		}
198 		pagevec_release(&pvec);
199 		cond_resched();
200 	}
201 
202 	if (partial) {
203 		struct page *page = find_lock_page(mapping, start - 1);
204 		if (page) {
205 			wait_on_page_writeback(page);
206 			truncate_partial_page(page, partial);
207 			unlock_page(page);
208 			page_cache_release(page);
209 		}
210 	}
211 
212 	next = start;
213 	for ( ; ; ) {
214 		cond_resched();
215 		if (!pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) {
216 			if (next == start)
217 				break;
218 			next = start;
219 			continue;
220 		}
221 		if (pvec.pages[0]->index > end) {
222 			pagevec_release(&pvec);
223 			break;
224 		}
225 		for (i = 0; i < pagevec_count(&pvec); i++) {
226 			struct page *page = pvec.pages[i];
227 
228 			if (page->index > end)
229 				break;
230 			lock_page(page);
231 			wait_on_page_writeback(page);
232 			if (page->index > next)
233 				next = page->index;
234 			next++;
235 			truncate_complete_page(mapping, page);
236 			unlock_page(page);
237 		}
238 		pagevec_release(&pvec);
239 	}
240 }
241 EXPORT_SYMBOL(truncate_inode_pages_range);
242 
243 /**
244  * truncate_inode_pages - truncate *all* the pages from an offset
245  * @mapping: mapping to truncate
246  * @lstart: offset from which to truncate
247  *
248  * Called under (and serialised by) inode->i_mutex.
249  */
250 void truncate_inode_pages(struct address_space *mapping, loff_t lstart)
251 {
252 	truncate_inode_pages_range(mapping, lstart, (loff_t)-1);
253 }
254 EXPORT_SYMBOL(truncate_inode_pages);
255 
256 /**
257  * invalidate_mapping_pages - Invalidate all the unlocked pages of one inode
258  * @mapping: the address_space which holds the pages to invalidate
259  * @start: the offset 'from' which to invalidate
260  * @end: the offset 'to' which to invalidate (inclusive)
261  *
262  * This function only removes the unlocked pages, if you want to
263  * remove all the pages of one inode, you must call truncate_inode_pages.
264  *
265  * invalidate_mapping_pages() will not block on IO activity. It will not
266  * invalidate pages which are dirty, locked, under writeback or mapped into
267  * pagetables.
268  */
269 unsigned long invalidate_mapping_pages(struct address_space *mapping,
270 				pgoff_t start, pgoff_t end)
271 {
272 	struct pagevec pvec;
273 	pgoff_t next = start;
274 	unsigned long ret = 0;
275 	int i;
276 
277 	pagevec_init(&pvec, 0);
278 	while (next <= end &&
279 			pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) {
280 		for (i = 0; i < pagevec_count(&pvec); i++) {
281 			struct page *page = pvec.pages[i];
282 			pgoff_t index;
283 			int lock_failed;
284 
285 			lock_failed = TestSetPageLocked(page);
286 
287 			/*
288 			 * We really shouldn't be looking at the ->index of an
289 			 * unlocked page.  But we're not allowed to lock these
290 			 * pages.  So we rely upon nobody altering the ->index
291 			 * of this (pinned-by-us) page.
292 			 */
293 			index = page->index;
294 			if (index > next)
295 				next = index;
296 			next++;
297 			if (lock_failed)
298 				continue;
299 
300 			if (PageDirty(page) || PageWriteback(page))
301 				goto unlock;
302 			if (page_mapped(page))
303 				goto unlock;
304 			ret += invalidate_complete_page(mapping, page);
305 unlock:
306 			unlock_page(page);
307 			if (next > end)
308 				break;
309 		}
310 		pagevec_release(&pvec);
311 	}
312 	return ret;
313 }
314 EXPORT_SYMBOL(invalidate_mapping_pages);
315 
316 /*
317  * This is like invalidate_complete_page(), except it ignores the page's
318  * refcount.  We do this because invalidate_inode_pages2() needs stronger
319  * invalidation guarantees, and cannot afford to leave pages behind because
320  * shrink_list() has a temp ref on them, or because they're transiently sitting
321  * in the lru_cache_add() pagevecs.
322  */
323 static int
324 invalidate_complete_page2(struct address_space *mapping, struct page *page)
325 {
326 	if (page->mapping != mapping)
327 		return 0;
328 
329 	if (PagePrivate(page) && !try_to_release_page(page, GFP_KERNEL))
330 		return 0;
331 
332 	write_lock_irq(&mapping->tree_lock);
333 	if (PageDirty(page))
334 		goto failed;
335 
336 	BUG_ON(PagePrivate(page));
337 	__remove_from_page_cache(page);
338 	write_unlock_irq(&mapping->tree_lock);
339 	ClearPageUptodate(page);
340 	page_cache_release(page);	/* pagecache ref */
341 	return 1;
342 failed:
343 	write_unlock_irq(&mapping->tree_lock);
344 	return 0;
345 }
346 
347 static int do_launder_page(struct address_space *mapping, struct page *page)
348 {
349 	if (!PageDirty(page))
350 		return 0;
351 	if (page->mapping != mapping || mapping->a_ops->launder_page == NULL)
352 		return 0;
353 	return mapping->a_ops->launder_page(page);
354 }
355 
356 /**
357  * invalidate_inode_pages2_range - remove range of pages from an address_space
358  * @mapping: the address_space
359  * @start: the page offset 'from' which to invalidate
360  * @end: the page offset 'to' which to invalidate (inclusive)
361  *
362  * Any pages which are found to be mapped into pagetables are unmapped prior to
363  * invalidation.
364  *
365  * Returns -EIO if any pages could not be invalidated.
366  */
367 int invalidate_inode_pages2_range(struct address_space *mapping,
368 				  pgoff_t start, pgoff_t end)
369 {
370 	struct pagevec pvec;
371 	pgoff_t next;
372 	int i;
373 	int ret = 0;
374 	int did_range_unmap = 0;
375 	int wrapped = 0;
376 
377 	pagevec_init(&pvec, 0);
378 	next = start;
379 	while (next <= end && !wrapped &&
380 		pagevec_lookup(&pvec, mapping, next,
381 			min(end - next, (pgoff_t)PAGEVEC_SIZE - 1) + 1)) {
382 		for (i = 0; i < pagevec_count(&pvec); i++) {
383 			struct page *page = pvec.pages[i];
384 			pgoff_t page_index;
385 
386 			lock_page(page);
387 			if (page->mapping != mapping) {
388 				unlock_page(page);
389 				continue;
390 			}
391 			page_index = page->index;
392 			next = page_index + 1;
393 			if (next == 0)
394 				wrapped = 1;
395 			if (page_index > end) {
396 				unlock_page(page);
397 				break;
398 			}
399 			wait_on_page_writeback(page);
400 			while (page_mapped(page)) {
401 				if (!did_range_unmap) {
402 					/*
403 					 * Zap the rest of the file in one hit.
404 					 */
405 					unmap_mapping_range(mapping,
406 					   (loff_t)page_index<<PAGE_CACHE_SHIFT,
407 					   (loff_t)(end - page_index + 1)
408 							<< PAGE_CACHE_SHIFT,
409 					    0);
410 					did_range_unmap = 1;
411 				} else {
412 					/*
413 					 * Just zap this page
414 					 */
415 					unmap_mapping_range(mapping,
416 					  (loff_t)page_index<<PAGE_CACHE_SHIFT,
417 					  PAGE_CACHE_SIZE, 0);
418 				}
419 			}
420 			ret = do_launder_page(mapping, page);
421 			if (ret == 0 && !invalidate_complete_page2(mapping, page))
422 				ret = -EIO;
423 			unlock_page(page);
424 		}
425 		pagevec_release(&pvec);
426 		cond_resched();
427 	}
428 	return ret;
429 }
430 EXPORT_SYMBOL_GPL(invalidate_inode_pages2_range);
431 
432 /**
433  * invalidate_inode_pages2 - remove all pages from an address_space
434  * @mapping: the address_space
435  *
436  * Any pages which are found to be mapped into pagetables are unmapped prior to
437  * invalidation.
438  *
439  * Returns -EIO if any pages could not be invalidated.
440  */
441 int invalidate_inode_pages2(struct address_space *mapping)
442 {
443 	return invalidate_inode_pages2_range(mapping, 0, -1);
444 }
445 EXPORT_SYMBOL_GPL(invalidate_inode_pages2);
446