blob: b8961cb6341401e27f9debffc5832154525fadcb [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 Mortone08748c2006-12-10 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 ClearPageUptodate(page);
108 ClearPageMappedToDisk(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700109 page_cache_release(page); /* pagecache ref */
110}
111
112/*
Andrew Mortonfc0ecff2007-02-10 01:45:39 -0800113 * This is for invalidate_mapping_pages(). That function can be called at
Linus Torvalds1da177e2005-04-16 15:20:36 -0700114 * any time, and is not supposed to throw away dirty pages. But pages can
Nick Piggin0fd0e6b2006-09-27 01:50:02 -0700115 * be marked dirty at any time too, so use remove_mapping which safely
116 * discards clean, unused pages.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700117 *
118 * Returns non-zero if the page was successfully invalidated.
119 */
120static int
121invalidate_complete_page(struct address_space *mapping, struct page *page)
122{
Nick Piggin0fd0e6b2006-09-27 01:50:02 -0700123 int ret;
124
Linus Torvalds1da177e2005-04-16 15:20:36 -0700125 if (page->mapping != mapping)
126 return 0;
127
128 if (PagePrivate(page) && !try_to_release_page(page, 0))
129 return 0;
130
Nick Piggin0fd0e6b2006-09-27 01:50:02 -0700131 ret = remove_mapping(mapping, page);
Nick Piggin0fd0e6b2006-09-27 01:50:02 -0700132
133 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700134}
135
136/**
Randy Dunlap06432452008-02-29 22:03:15 -0800137 * truncate_inode_pages - truncate range of pages specified by start & end byte offsets
Linus Torvalds1da177e2005-04-16 15:20:36 -0700138 * @mapping: mapping to truncate
139 * @lstart: offset from which to truncate
Hans Reiserd7339072006-01-06 00:10:36 -0800140 * @lend: offset to which to truncate
Linus Torvalds1da177e2005-04-16 15:20:36 -0700141 *
Hans Reiserd7339072006-01-06 00:10:36 -0800142 * Truncate the page cache, removing the pages that are between
143 * specified offsets (and zeroing out partial page
144 * (if lstart is not page aligned)).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700145 *
146 * Truncate takes two passes - the first pass is nonblocking. It will not
147 * block on page locks and it will not block on writeback. The second pass
148 * will wait. This is to prevent as much IO as possible in the affected region.
149 * The first pass will remove most pages, so the search cost of the second pass
150 * is low.
151 *
152 * When looking at page->index outside the page lock we need to be careful to
153 * copy it into a local to avoid races (it could change at any time).
154 *
155 * We pass down the cache-hot hint to the page freeing code. Even if the
156 * mapping is large, it is probably the case that the final pages are the most
157 * recently touched, and freeing happens in ascending file offset order.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700158 */
Hans Reiserd7339072006-01-06 00:10:36 -0800159void truncate_inode_pages_range(struct address_space *mapping,
160 loff_t lstart, loff_t lend)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700161{
162 const pgoff_t start = (lstart + PAGE_CACHE_SIZE-1) >> PAGE_CACHE_SHIFT;
Hans Reiserd7339072006-01-06 00:10:36 -0800163 pgoff_t end;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700164 const unsigned partial = lstart & (PAGE_CACHE_SIZE - 1);
165 struct pagevec pvec;
166 pgoff_t next;
167 int i;
168
169 if (mapping->nrpages == 0)
170 return;
171
Hans Reiserd7339072006-01-06 00:10:36 -0800172 BUG_ON((lend & (PAGE_CACHE_SIZE - 1)) != (PAGE_CACHE_SIZE - 1));
173 end = (lend >> PAGE_CACHE_SHIFT);
174
Linus Torvalds1da177e2005-04-16 15:20:36 -0700175 pagevec_init(&pvec, 0);
176 next = start;
Hans Reiserd7339072006-01-06 00:10:36 -0800177 while (next <= end &&
178 pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700179 for (i = 0; i < pagevec_count(&pvec); i++) {
180 struct page *page = pvec.pages[i];
181 pgoff_t page_index = page->index;
182
Hans Reiserd7339072006-01-06 00:10:36 -0800183 if (page_index > end) {
184 next = page_index;
185 break;
186 }
187
Linus Torvalds1da177e2005-04-16 15:20:36 -0700188 if (page_index > next)
189 next = page_index;
190 next++;
191 if (TestSetPageLocked(page))
192 continue;
193 if (PageWriteback(page)) {
194 unlock_page(page);
195 continue;
196 }
Nick Piggind00806b2007-07-19 01:46:57 -0700197 if (page_mapped(page)) {
198 unmap_mapping_range(mapping,
199 (loff_t)page_index<<PAGE_CACHE_SHIFT,
200 PAGE_CACHE_SIZE, 0);
201 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700202 truncate_complete_page(mapping, page);
203 unlock_page(page);
204 }
205 pagevec_release(&pvec);
206 cond_resched();
207 }
208
209 if (partial) {
210 struct page *page = find_lock_page(mapping, start - 1);
211 if (page) {
212 wait_on_page_writeback(page);
213 truncate_partial_page(page, partial);
214 unlock_page(page);
215 page_cache_release(page);
216 }
217 }
218
219 next = start;
220 for ( ; ; ) {
221 cond_resched();
222 if (!pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) {
223 if (next == start)
224 break;
225 next = start;
226 continue;
227 }
Hans Reiserd7339072006-01-06 00:10:36 -0800228 if (pvec.pages[0]->index > end) {
229 pagevec_release(&pvec);
230 break;
231 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700232 for (i = 0; i < pagevec_count(&pvec); i++) {
233 struct page *page = pvec.pages[i];
234
Hans Reiserd7339072006-01-06 00:10:36 -0800235 if (page->index > end)
236 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700237 lock_page(page);
238 wait_on_page_writeback(page);
Nick Piggind00806b2007-07-19 01:46:57 -0700239 if (page_mapped(page)) {
240 unmap_mapping_range(mapping,
241 (loff_t)page->index<<PAGE_CACHE_SHIFT,
242 PAGE_CACHE_SIZE, 0);
243 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700244 if (page->index > next)
245 next = page->index;
246 next++;
247 truncate_complete_page(mapping, page);
248 unlock_page(page);
249 }
250 pagevec_release(&pvec);
251 }
252}
Hans Reiserd7339072006-01-06 00:10:36 -0800253EXPORT_SYMBOL(truncate_inode_pages_range);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700254
Hans Reiserd7339072006-01-06 00:10:36 -0800255/**
256 * truncate_inode_pages - truncate *all* the pages from an offset
257 * @mapping: mapping to truncate
258 * @lstart: offset from which to truncate
259 *
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800260 * Called under (and serialised by) inode->i_mutex.
Hans Reiserd7339072006-01-06 00:10:36 -0800261 */
262void truncate_inode_pages(struct address_space *mapping, loff_t lstart)
263{
264 truncate_inode_pages_range(mapping, lstart, (loff_t)-1);
265}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700266EXPORT_SYMBOL(truncate_inode_pages);
267
Andrew Mortonfc9a07e2007-07-15 23:38:14 -0700268unsigned long __invalidate_mapping_pages(struct address_space *mapping,
269 pgoff_t start, pgoff_t end, bool be_atomic)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700270{
271 struct pagevec pvec;
272 pgoff_t next = start;
273 unsigned long ret = 0;
274 int i;
275
276 pagevec_init(&pvec, 0);
277 while (next <= end &&
278 pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) {
279 for (i = 0; i < pagevec_count(&pvec); i++) {
280 struct page *page = pvec.pages[i];
NeilBrowne0f23602006-06-23 02:05:48 -0700281 pgoff_t index;
282 int lock_failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700283
NeilBrowne0f23602006-06-23 02:05:48 -0700284 lock_failed = TestSetPageLocked(page);
285
286 /*
287 * We really shouldn't be looking at the ->index of an
288 * unlocked page. But we're not allowed to lock these
289 * pages. So we rely upon nobody altering the ->index
290 * of this (pinned-by-us) page.
291 */
292 index = page->index;
293 if (index > next)
294 next = index;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700295 next++;
NeilBrowne0f23602006-06-23 02:05:48 -0700296 if (lock_failed)
297 continue;
298
Linus Torvalds1da177e2005-04-16 15:20:36 -0700299 if (PageDirty(page) || PageWriteback(page))
300 goto unlock;
301 if (page_mapped(page))
302 goto unlock;
303 ret += invalidate_complete_page(mapping, page);
304unlock:
305 unlock_page(page);
306 if (next > end)
307 break;
308 }
309 pagevec_release(&pvec);
Andrew Mortonfc9a07e2007-07-15 23:38:14 -0700310 if (likely(!be_atomic))
311 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700312 }
313 return ret;
314}
Andrew Mortonfc9a07e2007-07-15 23:38:14 -0700315
316/**
317 * invalidate_mapping_pages - Invalidate all the unlocked pages of one inode
318 * @mapping: the address_space which holds the pages to invalidate
319 * @start: the offset 'from' which to invalidate
320 * @end: the offset 'to' which to invalidate (inclusive)
321 *
322 * This function only removes the unlocked pages, if you want to
323 * remove all the pages of one inode, you must call truncate_inode_pages.
324 *
325 * invalidate_mapping_pages() will not block on IO activity. It will not
326 * invalidate pages which are dirty, locked, under writeback or mapped into
327 * pagetables.
328 */
329unsigned long invalidate_mapping_pages(struct address_space *mapping,
330 pgoff_t start, pgoff_t end)
331{
332 return __invalidate_mapping_pages(mapping, start, end, false);
333}
Anton Altaparmakov54bc4852007-02-10 01:45:38 -0800334EXPORT_SYMBOL(invalidate_mapping_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700335
Andrew Mortonbd4c8ce2006-09-30 23:29:29 -0700336/*
337 * This is like invalidate_complete_page(), except it ignores the page's
338 * refcount. We do this because invalidate_inode_pages2() needs stronger
339 * invalidation guarantees, and cannot afford to leave pages behind because
Anderson Briglia2706a1b2007-07-15 23:38:09 -0700340 * shrink_page_list() has a temp ref on them, or because they're transiently
341 * sitting in the lru_cache_add() pagevecs.
Andrew Mortonbd4c8ce2006-09-30 23:29:29 -0700342 */
343static int
344invalidate_complete_page2(struct address_space *mapping, struct page *page)
345{
346 if (page->mapping != mapping)
347 return 0;
348
Trond Myklebust887ed2f2006-10-11 01:21:58 -0700349 if (PagePrivate(page) && !try_to_release_page(page, GFP_KERNEL))
Andrew Mortonbd4c8ce2006-09-30 23:29:29 -0700350 return 0;
351
352 write_lock_irq(&mapping->tree_lock);
353 if (PageDirty(page))
354 goto failed;
355
356 BUG_ON(PagePrivate(page));
357 __remove_from_page_cache(page);
358 write_unlock_irq(&mapping->tree_lock);
359 ClearPageUptodate(page);
360 page_cache_release(page); /* pagecache ref */
361 return 1;
362failed:
363 write_unlock_irq(&mapping->tree_lock);
364 return 0;
365}
366
Trond Myklebuste3db7692007-01-10 23:15:39 -0800367static int do_launder_page(struct address_space *mapping, struct page *page)
368{
369 if (!PageDirty(page))
370 return 0;
371 if (page->mapping != mapping || mapping->a_ops->launder_page == NULL)
372 return 0;
373 return mapping->a_ops->launder_page(page);
374}
375
Linus Torvalds1da177e2005-04-16 15:20:36 -0700376/**
377 * invalidate_inode_pages2_range - remove range of pages from an address_space
Martin Waitz67be2dd2005-05-01 08:59:26 -0700378 * @mapping: the address_space
Linus Torvalds1da177e2005-04-16 15:20:36 -0700379 * @start: the page offset 'from' which to invalidate
380 * @end: the page offset 'to' which to invalidate (inclusive)
381 *
382 * Any pages which are found to be mapped into pagetables are unmapped prior to
383 * invalidation.
384 *
385 * Returns -EIO if any pages could not be invalidated.
386 */
387int invalidate_inode_pages2_range(struct address_space *mapping,
388 pgoff_t start, pgoff_t end)
389{
390 struct pagevec pvec;
391 pgoff_t next;
392 int i;
393 int ret = 0;
Hisashi Hifumi0dd13342008-04-28 02:12:08 -0700394 int ret2 = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700395 int did_range_unmap = 0;
396 int wrapped = 0;
397
398 pagevec_init(&pvec, 0);
399 next = start;
Trond Myklebust7b965e02007-02-28 20:13:55 -0800400 while (next <= end && !wrapped &&
Linus Torvalds1da177e2005-04-16 15:20:36 -0700401 pagevec_lookup(&pvec, mapping, next,
402 min(end - next, (pgoff_t)PAGEVEC_SIZE - 1) + 1)) {
Trond Myklebust7b965e02007-02-28 20:13:55 -0800403 for (i = 0; i < pagevec_count(&pvec); i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700404 struct page *page = pvec.pages[i];
405 pgoff_t page_index;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700406
407 lock_page(page);
408 if (page->mapping != mapping) {
409 unlock_page(page);
410 continue;
411 }
412 page_index = page->index;
413 next = page_index + 1;
414 if (next == 0)
415 wrapped = 1;
416 if (page_index > end) {
417 unlock_page(page);
418 break;
419 }
420 wait_on_page_writeback(page);
Nick Piggind00806b2007-07-19 01:46:57 -0700421 if (page_mapped(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700422 if (!did_range_unmap) {
423 /*
424 * Zap the rest of the file in one hit.
425 */
426 unmap_mapping_range(mapping,
Oleg Drokin479ef592005-11-23 13:37:47 -0800427 (loff_t)page_index<<PAGE_CACHE_SHIFT,
428 (loff_t)(end - page_index + 1)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700429 << PAGE_CACHE_SHIFT,
430 0);
431 did_range_unmap = 1;
432 } else {
433 /*
434 * Just zap this page
435 */
436 unmap_mapping_range(mapping,
Oleg Drokin479ef592005-11-23 13:37:47 -0800437 (loff_t)page_index<<PAGE_CACHE_SHIFT,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700438 PAGE_CACHE_SIZE, 0);
439 }
440 }
Nick Piggind00806b2007-07-19 01:46:57 -0700441 BUG_ON(page_mapped(page));
Hisashi Hifumi0dd13342008-04-28 02:12:08 -0700442 ret2 = do_launder_page(mapping, page);
443 if (ret2 == 0) {
444 if (!invalidate_complete_page2(mapping, page))
445 ret2 = -EIO;
446 }
447 if (ret2 < 0)
448 ret = ret2;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700449 unlock_page(page);
450 }
451 pagevec_release(&pvec);
452 cond_resched();
453 }
454 return ret;
455}
456EXPORT_SYMBOL_GPL(invalidate_inode_pages2_range);
457
458/**
459 * invalidate_inode_pages2 - remove all pages from an address_space
Martin Waitz67be2dd2005-05-01 08:59:26 -0700460 * @mapping: the address_space
Linus Torvalds1da177e2005-04-16 15:20:36 -0700461 *
462 * Any pages which are found to be mapped into pagetables are unmapped prior to
463 * invalidation.
464 *
465 * Returns -EIO if any pages could not be invalidated.
466 */
467int invalidate_inode_pages2(struct address_space *mapping)
468{
469 return invalidate_inode_pages2_range(mapping, 0, -1);
470}
471EXPORT_SYMBOL_GPL(invalidate_inode_pages2);