blob: 6650c1d878b463a114e0e024688ecdc1d0fa6b74 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
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>
Alexey Dobriyan4af3c9c2007-10-16 23:29:23 -070011#include <linux/backing-dev.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070012#include <linux/mm.h>
Nick Piggin0fd0e6b2006-09-27 01:50:02 -070013#include <linux/swap.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070014#include <linux/module.h>
15#include <linux/pagemap.h>
Nate Diller01f27052007-05-09 02:35:07 -070016#include <linux/highmem.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070017#include <linux/pagevec.h>
Andrew Mortone08748ce2006-12-10 02:19:31 -080018#include <linux/task_io_accounting_ops.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070019#include <linux/buffer_head.h> /* grr. try_to_release_page,
Jan Karaaaa40592005-10-30 15:00:16 -080020 do_invalidatepage */
Linus Torvalds1da177e2005-04-16 15:20:36 -070021
22
David Howellscf9a2ae2006-08-29 19:05:54 +010023/**
Fengguang Wu28bc44d2008-02-03 18:04:10 +020024 * do_invalidatepage - invalidate part or all of a page
David Howellscf9a2ae2006-08-29 19:05:54 +010025 * @page: the page which is affected
26 * @offset: the index of the truncation point
27 *
28 * do_invalidatepage() is called when all or part of the page has become
29 * invalidated by a truncate operation.
30 *
31 * do_invalidatepage() does not have to release all buffers, but it must
32 * ensure that no dirty buffer is left outside @offset and that no I/O
33 * is underway against any of the blocks which are outside the truncation
34 * point. Because the caller is about to free (and possibly reuse) those
35 * blocks on-disk.
36 */
37void do_invalidatepage(struct page *page, unsigned long offset)
38{
39 void (*invalidatepage)(struct page *, unsigned long);
40 invalidatepage = page->mapping->a_ops->invalidatepage;
David Howells93614012006-09-30 20:45:40 +020041#ifdef CONFIG_BLOCK
David Howellscf9a2ae2006-08-29 19:05:54 +010042 if (!invalidatepage)
43 invalidatepage = block_invalidatepage;
David Howells93614012006-09-30 20:45:40 +020044#endif
David Howellscf9a2ae2006-08-29 19:05:54 +010045 if (invalidatepage)
46 (*invalidatepage)(page, offset);
47}
48
Linus Torvalds1da177e2005-04-16 15:20:36 -070049static inline void truncate_partial_page(struct page *page, unsigned partial)
50{
Christoph Lametereebd2aa2008-02-04 22:28:29 -080051 zero_user_segment(page, partial, PAGE_CACHE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -070052 if (PagePrivate(page))
53 do_invalidatepage(page, partial);
54}
55
Linus Torvaldsecdfc972007-01-26 12:47:06 -080056/*
57 * This cancels just the dirty bit on the kernel page itself, it
58 * does NOT actually remove dirty bits on any mmap's that may be
59 * around. It also leaves the page tagged dirty, so any sync
60 * activity will still find it on the dirty lists, and in particular,
61 * clear_page_dirty_for_io() will still look at the dirty bits in
62 * the VM.
63 *
64 * Doing this should *normally* only ever be done when a page
65 * is truncated, and is not actually mapped anywhere at all. However,
66 * fs/buffer.c does this when it notices that somebody has cleaned
67 * out all the buffers on a page without actually doing it through
68 * the VM. Can you say "ext3 is horribly ugly"? Tought you could.
69 */
Linus Torvaldsfba25912006-12-20 13:46:42 -080070void cancel_dirty_page(struct page *page, unsigned int account_size)
71{
Linus Torvalds8368e322006-12-23 09:25:04 -080072 if (TestClearPageDirty(page)) {
73 struct address_space *mapping = page->mapping;
74 if (mapping && mapping_cap_account_dirty(mapping)) {
75 dec_zone_page_state(page, NR_FILE_DIRTY);
Peter Zijlstrac9e51e42007-10-16 23:25:47 -070076 dec_bdi_stat(mapping->backing_dev_info,
77 BDI_RECLAIMABLE);
Linus Torvalds8368e322006-12-23 09:25:04 -080078 if (account_size)
79 task_io_account_cancelled_write(account_size);
80 }
Andrew Morton3e67c092006-12-21 11:00:33 -080081 }
Linus Torvaldsfba25912006-12-20 13:46:42 -080082}
Linus Torvalds8368e322006-12-23 09:25:04 -080083EXPORT_SYMBOL(cancel_dirty_page);
Linus Torvaldsfba25912006-12-20 13:46:42 -080084
Linus Torvalds1da177e2005-04-16 15:20:36 -070085/*
86 * If truncate cannot remove the fs-private metadata from the page, the page
Shaohua Li62e1c552008-02-04 22:29:33 -080087 * becomes orphaned. It will be left on the LRU and may even be mapped into
Nick Piggin54cb8822007-07-19 01:46:59 -070088 * user pagetables if we're racing with filemap_fault().
Linus Torvalds1da177e2005-04-16 15:20:36 -070089 *
90 * We need to bale out if page->mapping is no longer equal to the original
91 * mapping. This happens a) when the VM reclaimed the page while we waited on
Andrew Mortonfc0ecff2007-02-10 01:45:39 -080092 * its lock, b) when a concurrent invalidate_mapping_pages got there first and
Linus Torvalds1da177e2005-04-16 15:20:36 -070093 * c) when tmpfs swizzles a page between a tmpfs inode and swapper_space.
94 */
95static void
96truncate_complete_page(struct address_space *mapping, struct page *page)
97{
98 if (page->mapping != mapping)
99 return;
100
101 if (PagePrivate(page))
102 do_invalidatepage(page, 0);
103
Bjorn Steinbrinka2b34562008-02-04 22:29:28 -0800104 cancel_dirty_page(page, PAGE_CACHE_SIZE);
105
Nick Piggin787d2212007-07-17 04:03:34 -0700106 remove_from_page_cache(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700107 ClearPageMappedToDisk(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700108 page_cache_release(page); /* pagecache ref */
109}
110
111/*
Andrew Mortonfc0ecff2007-02-10 01:45:39 -0800112 * This is for invalidate_mapping_pages(). That function can be called at
Linus Torvalds1da177e2005-04-16 15:20:36 -0700113 * any time, and is not supposed to throw away dirty pages. But pages can
Nick Piggin0fd0e6b2006-09-27 01:50:02 -0700114 * be marked dirty at any time too, so use remove_mapping which safely
115 * discards clean, unused pages.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700116 *
117 * Returns non-zero if the page was successfully invalidated.
118 */
119static int
120invalidate_complete_page(struct address_space *mapping, struct page *page)
121{
Nick Piggin0fd0e6b2006-09-27 01:50:02 -0700122 int ret;
123
Linus Torvalds1da177e2005-04-16 15:20:36 -0700124 if (page->mapping != mapping)
125 return 0;
126
127 if (PagePrivate(page) && !try_to_release_page(page, 0))
128 return 0;
129
Nick Piggin0fd0e6b2006-09-27 01:50:02 -0700130 ret = remove_mapping(mapping, page);
Nick Piggin0fd0e6b2006-09-27 01:50:02 -0700131
132 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700133}
134
135/**
Randy Dunlap06432452008-02-29 22:03:15 -0800136 * truncate_inode_pages - truncate range of pages specified by start & end byte offsets
Linus Torvalds1da177e2005-04-16 15:20:36 -0700137 * @mapping: mapping to truncate
138 * @lstart: offset from which to truncate
Hans Reiserd7339072006-01-06 00:10:36 -0800139 * @lend: offset to which to truncate
Linus Torvalds1da177e2005-04-16 15:20:36 -0700140 *
Hans Reiserd7339072006-01-06 00:10:36 -0800141 * Truncate the page cache, removing the pages that are between
142 * specified offsets (and zeroing out partial page
143 * (if lstart is not page aligned)).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700144 *
145 * Truncate takes two passes - the first pass is nonblocking. It will not
146 * block on page locks and it will not block on writeback. The second pass
147 * will wait. This is to prevent as much IO as possible in the affected region.
148 * The first pass will remove most pages, so the search cost of the second pass
149 * is low.
150 *
151 * When looking at page->index outside the page lock we need to be careful to
152 * copy it into a local to avoid races (it could change at any time).
153 *
154 * We pass down the cache-hot hint to the page freeing code. Even if the
155 * mapping is large, it is probably the case that the final pages are the most
156 * recently touched, and freeing happens in ascending file offset order.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700157 */
Hans Reiserd7339072006-01-06 00:10:36 -0800158void truncate_inode_pages_range(struct address_space *mapping,
159 loff_t lstart, loff_t lend)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700160{
161 const pgoff_t start = (lstart + PAGE_CACHE_SIZE-1) >> PAGE_CACHE_SHIFT;
Hans Reiserd7339072006-01-06 00:10:36 -0800162 pgoff_t end;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700163 const unsigned partial = lstart & (PAGE_CACHE_SIZE - 1);
164 struct pagevec pvec;
165 pgoff_t next;
166 int i;
167
168 if (mapping->nrpages == 0)
169 return;
170
Hans Reiserd7339072006-01-06 00:10:36 -0800171 BUG_ON((lend & (PAGE_CACHE_SIZE - 1)) != (PAGE_CACHE_SIZE - 1));
172 end = (lend >> PAGE_CACHE_SHIFT);
173
Linus Torvalds1da177e2005-04-16 15:20:36 -0700174 pagevec_init(&pvec, 0);
175 next = start;
Hans Reiserd7339072006-01-06 00:10:36 -0800176 while (next <= end &&
177 pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700178 for (i = 0; i < pagevec_count(&pvec); i++) {
179 struct page *page = pvec.pages[i];
180 pgoff_t page_index = page->index;
181
Hans Reiserd7339072006-01-06 00:10:36 -0800182 if (page_index > end) {
183 next = page_index;
184 break;
185 }
186
Linus Torvalds1da177e2005-04-16 15:20:36 -0700187 if (page_index > next)
188 next = page_index;
189 next++;
Nick Piggin529ae9a2008-08-02 12:01:03 +0200190 if (!trylock_page(page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700191 continue;
192 if (PageWriteback(page)) {
193 unlock_page(page);
194 continue;
195 }
Nick Piggind00806b2007-07-19 01:46:57 -0700196 if (page_mapped(page)) {
197 unmap_mapping_range(mapping,
198 (loff_t)page_index<<PAGE_CACHE_SHIFT,
199 PAGE_CACHE_SIZE, 0);
200 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700201 truncate_complete_page(mapping, page);
202 unlock_page(page);
203 }
204 pagevec_release(&pvec);
205 cond_resched();
206 }
207
208 if (partial) {
209 struct page *page = find_lock_page(mapping, start - 1);
210 if (page) {
211 wait_on_page_writeback(page);
212 truncate_partial_page(page, partial);
213 unlock_page(page);
214 page_cache_release(page);
215 }
216 }
217
218 next = start;
219 for ( ; ; ) {
220 cond_resched();
221 if (!pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) {
222 if (next == start)
223 break;
224 next = start;
225 continue;
226 }
Hans Reiserd7339072006-01-06 00:10:36 -0800227 if (pvec.pages[0]->index > end) {
228 pagevec_release(&pvec);
229 break;
230 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700231 for (i = 0; i < pagevec_count(&pvec); i++) {
232 struct page *page = pvec.pages[i];
233
Hans Reiserd7339072006-01-06 00:10:36 -0800234 if (page->index > end)
235 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700236 lock_page(page);
237 wait_on_page_writeback(page);
Nick Piggind00806b2007-07-19 01:46:57 -0700238 if (page_mapped(page)) {
239 unmap_mapping_range(mapping,
240 (loff_t)page->index<<PAGE_CACHE_SHIFT,
241 PAGE_CACHE_SIZE, 0);
242 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700243 if (page->index > next)
244 next = page->index;
245 next++;
246 truncate_complete_page(mapping, page);
247 unlock_page(page);
248 }
249 pagevec_release(&pvec);
250 }
251}
Hans Reiserd7339072006-01-06 00:10:36 -0800252EXPORT_SYMBOL(truncate_inode_pages_range);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700253
Hans Reiserd7339072006-01-06 00:10:36 -0800254/**
255 * truncate_inode_pages - truncate *all* the pages from an offset
256 * @mapping: mapping to truncate
257 * @lstart: offset from which to truncate
258 *
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800259 * Called under (and serialised by) inode->i_mutex.
Hans Reiserd7339072006-01-06 00:10:36 -0800260 */
261void truncate_inode_pages(struct address_space *mapping, loff_t lstart)
262{
263 truncate_inode_pages_range(mapping, lstart, (loff_t)-1);
264}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700265EXPORT_SYMBOL(truncate_inode_pages);
266
Andrew Mortonfc9a07e2007-07-15 23:38:14 -0700267unsigned long __invalidate_mapping_pages(struct address_space *mapping,
268 pgoff_t start, pgoff_t end, bool be_atomic)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700269{
270 struct pagevec pvec;
271 pgoff_t next = start;
272 unsigned long ret = 0;
273 int i;
274
275 pagevec_init(&pvec, 0);
276 while (next <= end &&
277 pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) {
278 for (i = 0; i < pagevec_count(&pvec); i++) {
279 struct page *page = pvec.pages[i];
NeilBrowne0f23602006-06-23 02:05:48 -0700280 pgoff_t index;
281 int lock_failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700282
Nick Piggin529ae9a2008-08-02 12:01:03 +0200283 lock_failed = !trylock_page(page);
NeilBrowne0f23602006-06-23 02:05:48 -0700284
285 /*
286 * We really shouldn't be looking at the ->index of an
287 * unlocked page. But we're not allowed to lock these
288 * pages. So we rely upon nobody altering the ->index
289 * of this (pinned-by-us) page.
290 */
291 index = page->index;
292 if (index > next)
293 next = index;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700294 next++;
NeilBrowne0f23602006-06-23 02:05:48 -0700295 if (lock_failed)
296 continue;
297
Linus Torvalds1da177e2005-04-16 15:20:36 -0700298 if (PageDirty(page) || PageWriteback(page))
299 goto unlock;
300 if (page_mapped(page))
301 goto unlock;
302 ret += invalidate_complete_page(mapping, page);
303unlock:
304 unlock_page(page);
305 if (next > end)
306 break;
307 }
308 pagevec_release(&pvec);
Andrew Mortonfc9a07e2007-07-15 23:38:14 -0700309 if (likely(!be_atomic))
310 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700311 }
312 return ret;
313}
Andrew Mortonfc9a07e2007-07-15 23:38:14 -0700314
315/**
316 * invalidate_mapping_pages - Invalidate all the unlocked pages of one inode
317 * @mapping: the address_space which holds the pages to invalidate
318 * @start: the offset 'from' which to invalidate
319 * @end: the offset 'to' which to invalidate (inclusive)
320 *
321 * This function only removes the unlocked pages, if you want to
322 * remove all the pages of one inode, you must call truncate_inode_pages.
323 *
324 * invalidate_mapping_pages() will not block on IO activity. It will not
325 * invalidate pages which are dirty, locked, under writeback or mapped into
326 * pagetables.
327 */
328unsigned long invalidate_mapping_pages(struct address_space *mapping,
329 pgoff_t start, pgoff_t end)
330{
331 return __invalidate_mapping_pages(mapping, start, end, false);
332}
Anton Altaparmakov54bc4852007-02-10 01:45:38 -0800333EXPORT_SYMBOL(invalidate_mapping_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700334
Andrew Mortonbd4c8ce2006-09-30 23:29:29 -0700335/*
336 * This is like invalidate_complete_page(), except it ignores the page's
337 * refcount. We do this because invalidate_inode_pages2() needs stronger
338 * invalidation guarantees, and cannot afford to leave pages behind because
Anderson Briglia2706a1b2007-07-15 23:38:09 -0700339 * shrink_page_list() has a temp ref on them, or because they're transiently
340 * sitting in the lru_cache_add() pagevecs.
Andrew Mortonbd4c8ce2006-09-30 23:29:29 -0700341 */
342static int
343invalidate_complete_page2(struct address_space *mapping, struct page *page)
344{
345 if (page->mapping != mapping)
346 return 0;
347
Trond Myklebust887ed2f2006-10-11 01:21:58 -0700348 if (PagePrivate(page) && !try_to_release_page(page, GFP_KERNEL))
Andrew Mortonbd4c8ce2006-09-30 23:29:29 -0700349 return 0;
350
Nick Piggin19fd6232008-07-25 19:45:32 -0700351 spin_lock_irq(&mapping->tree_lock);
Andrew Mortonbd4c8ce2006-09-30 23:29:29 -0700352 if (PageDirty(page))
353 goto failed;
354
355 BUG_ON(PagePrivate(page));
356 __remove_from_page_cache(page);
Nick Piggin19fd6232008-07-25 19:45:32 -0700357 spin_unlock_irq(&mapping->tree_lock);
Andrew Mortonbd4c8ce2006-09-30 23:29:29 -0700358 page_cache_release(page); /* pagecache ref */
359 return 1;
360failed:
Nick Piggin19fd6232008-07-25 19:45:32 -0700361 spin_unlock_irq(&mapping->tree_lock);
Andrew Mortonbd4c8ce2006-09-30 23:29:29 -0700362 return 0;
363}
364
Trond Myklebuste3db7692007-01-10 23:15:39 -0800365static int do_launder_page(struct address_space *mapping, struct page *page)
366{
367 if (!PageDirty(page))
368 return 0;
369 if (page->mapping != mapping || mapping->a_ops->launder_page == NULL)
370 return 0;
371 return mapping->a_ops->launder_page(page);
372}
373
Linus Torvalds1da177e2005-04-16 15:20:36 -0700374/**
375 * invalidate_inode_pages2_range - remove range of pages from an address_space
Martin Waitz67be2dd2005-05-01 08:59:26 -0700376 * @mapping: the address_space
Linus Torvalds1da177e2005-04-16 15:20:36 -0700377 * @start: the page offset 'from' which to invalidate
378 * @end: the page offset 'to' which to invalidate (inclusive)
379 *
380 * Any pages which are found to be mapped into pagetables are unmapped prior to
381 * invalidation.
382 *
Hisashi Hifumi6ccfa802008-09-02 14:35:40 -0700383 * Returns -EBUSY if any pages could not be invalidated.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700384 */
385int invalidate_inode_pages2_range(struct address_space *mapping,
386 pgoff_t start, pgoff_t end)
387{
388 struct pagevec pvec;
389 pgoff_t next;
390 int i;
391 int ret = 0;
Hisashi Hifumi0dd13342008-04-28 02:12:08 -0700392 int ret2 = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700393 int did_range_unmap = 0;
394 int wrapped = 0;
395
396 pagevec_init(&pvec, 0);
397 next = start;
Trond Myklebust7b965e02007-02-28 20:13:55 -0800398 while (next <= end && !wrapped &&
Linus Torvalds1da177e2005-04-16 15:20:36 -0700399 pagevec_lookup(&pvec, mapping, next,
400 min(end - next, (pgoff_t)PAGEVEC_SIZE - 1) + 1)) {
Trond Myklebust7b965e02007-02-28 20:13:55 -0800401 for (i = 0; i < pagevec_count(&pvec); i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700402 struct page *page = pvec.pages[i];
403 pgoff_t page_index;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700404
405 lock_page(page);
406 if (page->mapping != mapping) {
407 unlock_page(page);
408 continue;
409 }
410 page_index = page->index;
411 next = page_index + 1;
412 if (next == 0)
413 wrapped = 1;
414 if (page_index > end) {
415 unlock_page(page);
416 break;
417 }
418 wait_on_page_writeback(page);
Nick Piggind00806b2007-07-19 01:46:57 -0700419 if (page_mapped(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700420 if (!did_range_unmap) {
421 /*
422 * Zap the rest of the file in one hit.
423 */
424 unmap_mapping_range(mapping,
Oleg Drokin479ef592005-11-23 13:37:47 -0800425 (loff_t)page_index<<PAGE_CACHE_SHIFT,
426 (loff_t)(end - page_index + 1)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700427 << PAGE_CACHE_SHIFT,
428 0);
429 did_range_unmap = 1;
430 } else {
431 /*
432 * Just zap this page
433 */
434 unmap_mapping_range(mapping,
Oleg Drokin479ef592005-11-23 13:37:47 -0800435 (loff_t)page_index<<PAGE_CACHE_SHIFT,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700436 PAGE_CACHE_SIZE, 0);
437 }
438 }
Nick Piggind00806b2007-07-19 01:46:57 -0700439 BUG_ON(page_mapped(page));
Hisashi Hifumi0dd13342008-04-28 02:12:08 -0700440 ret2 = do_launder_page(mapping, page);
441 if (ret2 == 0) {
442 if (!invalidate_complete_page2(mapping, page))
Hisashi Hifumi6ccfa802008-09-02 14:35:40 -0700443 ret2 = -EBUSY;
Hisashi Hifumi0dd13342008-04-28 02:12:08 -0700444 }
445 if (ret2 < 0)
446 ret = ret2;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700447 unlock_page(page);
448 }
449 pagevec_release(&pvec);
450 cond_resched();
451 }
452 return ret;
453}
454EXPORT_SYMBOL_GPL(invalidate_inode_pages2_range);
455
456/**
457 * invalidate_inode_pages2 - remove all pages from an address_space
Martin Waitz67be2dd2005-05-01 08:59:26 -0700458 * @mapping: the address_space
Linus Torvalds1da177e2005-04-16 15:20:36 -0700459 *
460 * Any pages which are found to be mapped into pagetables are unmapped prior to
461 * invalidation.
462 *
463 * Returns -EIO if any pages could not be invalidated.
464 */
465int invalidate_inode_pages2(struct address_space *mapping)
466{
467 return invalidate_inode_pages2_range(mapping, 0, -1);
468}
469EXPORT_SYMBOL_GPL(invalidate_inode_pages2);