blob: e13f22efaad741bfb0268ba577c3d34807769201 [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 *
Francois Camie1f8e872008-10-15 22:01:59 -07006 * 10Sep2002 Andrew Morton
Linus Torvalds1da177e2005-04-16 15:20:36 -07007 * Initial version.
8 */
9
10#include <linux/kernel.h>
Alexey Dobriyan4af3c9c2007-10-16 23:29:23 -070011#include <linux/backing-dev.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090012#include <linux/gfp.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070013#include <linux/mm.h>
Nick Piggin0fd0e6b2006-09-27 01:50:02 -070014#include <linux/swap.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070015#include <linux/module.h>
16#include <linux/pagemap.h>
Nate Diller01f27052007-05-09 02:35:07 -070017#include <linux/highmem.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070018#include <linux/pagevec.h>
Andrew Mortone08748c2006-12-10 02:19:31 -080019#include <linux/task_io_accounting_ops.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070020#include <linux/buffer_head.h> /* grr. try_to_release_page,
Jan Karaaaa40592005-10-30 15:00:16 -080021 do_invalidatepage */
Dan Magenheimerc515e1f2011-05-26 10:01:43 -060022#include <linux/cleancache.h>
Rik van Rielba470de2008-10-18 20:26:50 -070023#include "internal.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070024
25
David Howellscf9a2ae2006-08-29 19:05:54 +010026/**
Fengguang Wu28bc44d2008-02-03 18:04:10 +020027 * do_invalidatepage - invalidate part or all of a page
David Howellscf9a2ae2006-08-29 19:05:54 +010028 * @page: the page which is affected
29 * @offset: the index of the truncation point
30 *
31 * do_invalidatepage() is called when all or part of the page has become
32 * invalidated by a truncate operation.
33 *
34 * do_invalidatepage() does not have to release all buffers, but it must
35 * ensure that no dirty buffer is left outside @offset and that no I/O
36 * is underway against any of the blocks which are outside the truncation
37 * point. Because the caller is about to free (and possibly reuse) those
38 * blocks on-disk.
39 */
40void do_invalidatepage(struct page *page, unsigned long offset)
41{
42 void (*invalidatepage)(struct page *, unsigned long);
43 invalidatepage = page->mapping->a_ops->invalidatepage;
David Howells93614012006-09-30 20:45:40 +020044#ifdef CONFIG_BLOCK
David Howellscf9a2ae2006-08-29 19:05:54 +010045 if (!invalidatepage)
46 invalidatepage = block_invalidatepage;
David Howells93614012006-09-30 20:45:40 +020047#endif
David Howellscf9a2ae2006-08-29 19:05:54 +010048 if (invalidatepage)
49 (*invalidatepage)(page, offset);
50}
51
Linus Torvalds1da177e2005-04-16 15:20:36 -070052static inline void truncate_partial_page(struct page *page, unsigned partial)
53{
Christoph Lametereebd2aa2008-02-04 22:28:29 -080054 zero_user_segment(page, partial, PAGE_CACHE_SIZE);
Dan Magenheimerc515e1f2011-05-26 10:01:43 -060055 cleancache_flush_page(page->mapping, page);
David Howells266cf652009-04-03 16:42:36 +010056 if (page_has_private(page))
Linus Torvalds1da177e2005-04-16 15:20:36 -070057 do_invalidatepage(page, partial);
58}
59
Linus Torvaldsecdfc972007-01-26 12:47:06 -080060/*
61 * This cancels just the dirty bit on the kernel page itself, it
62 * does NOT actually remove dirty bits on any mmap's that may be
63 * around. It also leaves the page tagged dirty, so any sync
64 * activity will still find it on the dirty lists, and in particular,
65 * clear_page_dirty_for_io() will still look at the dirty bits in
66 * the VM.
67 *
68 * Doing this should *normally* only ever be done when a page
69 * is truncated, and is not actually mapped anywhere at all. However,
70 * fs/buffer.c does this when it notices that somebody has cleaned
71 * out all the buffers on a page without actually doing it through
72 * the VM. Can you say "ext3 is horribly ugly"? Tought you could.
73 */
Linus Torvaldsfba25912006-12-20 13:46:42 -080074void cancel_dirty_page(struct page *page, unsigned int account_size)
75{
Linus Torvalds8368e322006-12-23 09:25:04 -080076 if (TestClearPageDirty(page)) {
77 struct address_space *mapping = page->mapping;
78 if (mapping && mapping_cap_account_dirty(mapping)) {
79 dec_zone_page_state(page, NR_FILE_DIRTY);
Peter Zijlstrac9e51e42007-10-16 23:25:47 -070080 dec_bdi_stat(mapping->backing_dev_info,
81 BDI_RECLAIMABLE);
Linus Torvalds8368e322006-12-23 09:25:04 -080082 if (account_size)
83 task_io_account_cancelled_write(account_size);
84 }
Andrew Morton3e67c092006-12-21 11:00:33 -080085 }
Linus Torvaldsfba25912006-12-20 13:46:42 -080086}
Linus Torvalds8368e322006-12-23 09:25:04 -080087EXPORT_SYMBOL(cancel_dirty_page);
Linus Torvaldsfba25912006-12-20 13:46:42 -080088
Linus Torvalds1da177e2005-04-16 15:20:36 -070089/*
90 * If truncate cannot remove the fs-private metadata from the page, the page
Shaohua Li62e1c552008-02-04 22:29:33 -080091 * becomes orphaned. It will be left on the LRU and may even be mapped into
Nick Piggin54cb8822007-07-19 01:46:59 -070092 * user pagetables if we're racing with filemap_fault().
Linus Torvalds1da177e2005-04-16 15:20:36 -070093 *
94 * We need to bale out if page->mapping is no longer equal to the original
95 * mapping. This happens a) when the VM reclaimed the page while we waited on
Andrew Mortonfc0ecff2007-02-10 01:45:39 -080096 * its lock, b) when a concurrent invalidate_mapping_pages got there first and
Linus Torvalds1da177e2005-04-16 15:20:36 -070097 * c) when tmpfs swizzles a page between a tmpfs inode and swapper_space.
98 */
Nick Piggin750b4982009-09-16 11:50:12 +020099static int
Linus Torvalds1da177e2005-04-16 15:20:36 -0700100truncate_complete_page(struct address_space *mapping, struct page *page)
101{
102 if (page->mapping != mapping)
Nick Piggin750b4982009-09-16 11:50:12 +0200103 return -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700104
David Howells266cf652009-04-03 16:42:36 +0100105 if (page_has_private(page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700106 do_invalidatepage(page, 0);
107
Bjorn Steinbrinka2b34562008-02-04 22:29:28 -0800108 cancel_dirty_page(page, PAGE_CACHE_SIZE);
109
Rik van Rielba470de2008-10-18 20:26:50 -0700110 clear_page_mlock(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700111 ClearPageMappedToDisk(page);
Minchan Kim5adc7b52011-03-22 16:32:41 -0700112 delete_from_page_cache(page);
Nick Piggin750b4982009-09-16 11:50:12 +0200113 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700114}
115
116/*
Andrew Mortonfc0ecff2007-02-10 01:45:39 -0800117 * This is for invalidate_mapping_pages(). That function can be called at
Linus Torvalds1da177e2005-04-16 15:20:36 -0700118 * any time, and is not supposed to throw away dirty pages. But pages can
Nick Piggin0fd0e6b2006-09-27 01:50:02 -0700119 * be marked dirty at any time too, so use remove_mapping which safely
120 * discards clean, unused pages.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700121 *
122 * Returns non-zero if the page was successfully invalidated.
123 */
124static int
125invalidate_complete_page(struct address_space *mapping, struct page *page)
126{
Nick Piggin0fd0e6b2006-09-27 01:50:02 -0700127 int ret;
128
Linus Torvalds1da177e2005-04-16 15:20:36 -0700129 if (page->mapping != mapping)
130 return 0;
131
David Howells266cf652009-04-03 16:42:36 +0100132 if (page_has_private(page) && !try_to_release_page(page, 0))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700133 return 0;
134
Rik van Rielba470de2008-10-18 20:26:50 -0700135 clear_page_mlock(page);
Nick Piggin0fd0e6b2006-09-27 01:50:02 -0700136 ret = remove_mapping(mapping, page);
Nick Piggin0fd0e6b2006-09-27 01:50:02 -0700137
138 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700139}
140
Nick Piggin750b4982009-09-16 11:50:12 +0200141int truncate_inode_page(struct address_space *mapping, struct page *page)
142{
143 if (page_mapped(page)) {
144 unmap_mapping_range(mapping,
145 (loff_t)page->index << PAGE_CACHE_SHIFT,
146 PAGE_CACHE_SIZE, 0);
147 }
148 return truncate_complete_page(mapping, page);
149}
150
Wu Fengguang83f78662009-09-16 11:50:13 +0200151/*
Andi Kleen25718732009-09-16 11:50:13 +0200152 * Used to get rid of pages on hardware memory corruption.
153 */
154int generic_error_remove_page(struct address_space *mapping, struct page *page)
155{
156 if (!mapping)
157 return -EINVAL;
158 /*
159 * Only punch for normal data pages for now.
160 * Handling other types like directories would need more auditing.
161 */
162 if (!S_ISREG(mapping->host->i_mode))
163 return -EIO;
164 return truncate_inode_page(mapping, page);
165}
166EXPORT_SYMBOL(generic_error_remove_page);
167
168/*
Wu Fengguang83f78662009-09-16 11:50:13 +0200169 * Safely invalidate one page from its pagecache mapping.
170 * It only drops clean, unused pages. The page must be locked.
171 *
172 * Returns 1 if the page is successfully invalidated, otherwise 0.
173 */
174int invalidate_inode_page(struct page *page)
175{
176 struct address_space *mapping = page_mapping(page);
177 if (!mapping)
178 return 0;
179 if (PageDirty(page) || PageWriteback(page))
180 return 0;
181 if (page_mapped(page))
182 return 0;
183 return invalidate_complete_page(mapping, page);
184}
185
Linus Torvalds1da177e2005-04-16 15:20:36 -0700186/**
Randy Dunlap06432452008-02-29 22:03:15 -0800187 * truncate_inode_pages - truncate range of pages specified by start & end byte offsets
Linus Torvalds1da177e2005-04-16 15:20:36 -0700188 * @mapping: mapping to truncate
189 * @lstart: offset from which to truncate
Hans Reiserd7339072006-01-06 00:10:36 -0800190 * @lend: offset to which to truncate
Linus Torvalds1da177e2005-04-16 15:20:36 -0700191 *
Hans Reiserd7339072006-01-06 00:10:36 -0800192 * Truncate the page cache, removing the pages that are between
193 * specified offsets (and zeroing out partial page
194 * (if lstart is not page aligned)).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700195 *
196 * Truncate takes two passes - the first pass is nonblocking. It will not
197 * block on page locks and it will not block on writeback. The second pass
198 * will wait. This is to prevent as much IO as possible in the affected region.
199 * The first pass will remove most pages, so the search cost of the second pass
200 * is low.
201 *
202 * When looking at page->index outside the page lock we need to be careful to
203 * copy it into a local to avoid races (it could change at any time).
204 *
205 * We pass down the cache-hot hint to the page freeing code. Even if the
206 * mapping is large, it is probably the case that the final pages are the most
207 * recently touched, and freeing happens in ascending file offset order.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700208 */
Hans Reiserd7339072006-01-06 00:10:36 -0800209void truncate_inode_pages_range(struct address_space *mapping,
210 loff_t lstart, loff_t lend)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700211{
212 const pgoff_t start = (lstart + PAGE_CACHE_SIZE-1) >> PAGE_CACHE_SHIFT;
Hans Reiserd7339072006-01-06 00:10:36 -0800213 pgoff_t end;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700214 const unsigned partial = lstart & (PAGE_CACHE_SIZE - 1);
215 struct pagevec pvec;
216 pgoff_t next;
217 int i;
218
Dan Magenheimerc515e1f2011-05-26 10:01:43 -0600219 cleancache_flush_inode(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700220 if (mapping->nrpages == 0)
221 return;
222
Hans Reiserd7339072006-01-06 00:10:36 -0800223 BUG_ON((lend & (PAGE_CACHE_SIZE - 1)) != (PAGE_CACHE_SIZE - 1));
224 end = (lend >> PAGE_CACHE_SHIFT);
225
Linus Torvalds1da177e2005-04-16 15:20:36 -0700226 pagevec_init(&pvec, 0);
227 next = start;
Hans Reiserd7339072006-01-06 00:10:36 -0800228 while (next <= end &&
229 pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) {
Hugh Dickinse5598f82011-02-25 14:44:29 -0800230 mem_cgroup_uncharge_start();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700231 for (i = 0; i < pagevec_count(&pvec); i++) {
232 struct page *page = pvec.pages[i];
233 pgoff_t page_index = page->index;
234
Hans Reiserd7339072006-01-06 00:10:36 -0800235 if (page_index > end) {
236 next = page_index;
237 break;
238 }
239
Linus Torvalds1da177e2005-04-16 15:20:36 -0700240 if (page_index > next)
241 next = page_index;
242 next++;
Nick Piggin529ae9a2008-08-02 12:01:03 +0200243 if (!trylock_page(page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700244 continue;
245 if (PageWriteback(page)) {
246 unlock_page(page);
247 continue;
248 }
Nick Piggin750b4982009-09-16 11:50:12 +0200249 truncate_inode_page(mapping, page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700250 unlock_page(page);
251 }
252 pagevec_release(&pvec);
Hugh Dickinse5598f82011-02-25 14:44:29 -0800253 mem_cgroup_uncharge_end();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700254 cond_resched();
255 }
256
257 if (partial) {
258 struct page *page = find_lock_page(mapping, start - 1);
259 if (page) {
260 wait_on_page_writeback(page);
261 truncate_partial_page(page, partial);
262 unlock_page(page);
263 page_cache_release(page);
264 }
265 }
266
267 next = start;
268 for ( ; ; ) {
269 cond_resched();
270 if (!pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) {
271 if (next == start)
272 break;
273 next = start;
274 continue;
275 }
Hans Reiserd7339072006-01-06 00:10:36 -0800276 if (pvec.pages[0]->index > end) {
277 pagevec_release(&pvec);
278 break;
279 }
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -0800280 mem_cgroup_uncharge_start();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700281 for (i = 0; i < pagevec_count(&pvec); i++) {
282 struct page *page = pvec.pages[i];
283
Hans Reiserd7339072006-01-06 00:10:36 -0800284 if (page->index > end)
285 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700286 lock_page(page);
287 wait_on_page_writeback(page);
Nick Piggin750b4982009-09-16 11:50:12 +0200288 truncate_inode_page(mapping, page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700289 if (page->index > next)
290 next = page->index;
291 next++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700292 unlock_page(page);
293 }
294 pagevec_release(&pvec);
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -0800295 mem_cgroup_uncharge_end();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700296 }
Dan Magenheimerc515e1f2011-05-26 10:01:43 -0600297 cleancache_flush_inode(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700298}
Hans Reiserd7339072006-01-06 00:10:36 -0800299EXPORT_SYMBOL(truncate_inode_pages_range);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700300
Hans Reiserd7339072006-01-06 00:10:36 -0800301/**
302 * truncate_inode_pages - truncate *all* the pages from an offset
303 * @mapping: mapping to truncate
304 * @lstart: offset from which to truncate
305 *
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800306 * Called under (and serialised by) inode->i_mutex.
Jan Kara08142572011-06-27 16:18:10 -0700307 *
308 * Note: When this function returns, there can be a page in the process of
309 * deletion (inside __delete_from_page_cache()) in the specified range. Thus
310 * mapping->nrpages can be non-zero when this function returns even after
311 * truncation of the whole mapping.
Hans Reiserd7339072006-01-06 00:10:36 -0800312 */
313void truncate_inode_pages(struct address_space *mapping, loff_t lstart)
314{
315 truncate_inode_pages_range(mapping, lstart, (loff_t)-1);
316}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700317EXPORT_SYMBOL(truncate_inode_pages);
318
Mike Waychison28697352009-06-16 15:32:59 -0700319/**
320 * invalidate_mapping_pages - Invalidate all the unlocked pages of one inode
321 * @mapping: the address_space which holds the pages to invalidate
322 * @start: the offset 'from' which to invalidate
323 * @end: the offset 'to' which to invalidate (inclusive)
324 *
325 * This function only removes the unlocked pages, if you want to
326 * remove all the pages of one inode, you must call truncate_inode_pages.
327 *
328 * invalidate_mapping_pages() will not block on IO activity. It will not
329 * invalidate pages which are dirty, locked, under writeback or mapped into
330 * pagetables.
331 */
332unsigned long invalidate_mapping_pages(struct address_space *mapping,
Minchan Kim31560182011-03-22 16:32:52 -0700333 pgoff_t start, pgoff_t end)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700334{
335 struct pagevec pvec;
336 pgoff_t next = start;
Minchan Kim31560182011-03-22 16:32:52 -0700337 unsigned long ret;
338 unsigned long count = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700339 int i;
340
341 pagevec_init(&pvec, 0);
342 while (next <= end &&
343 pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) {
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -0800344 mem_cgroup_uncharge_start();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700345 for (i = 0; i < pagevec_count(&pvec); i++) {
346 struct page *page = pvec.pages[i];
NeilBrowne0f23602006-06-23 02:05:48 -0700347 pgoff_t index;
348 int lock_failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700349
Nick Piggin529ae9a2008-08-02 12:01:03 +0200350 lock_failed = !trylock_page(page);
NeilBrowne0f23602006-06-23 02:05:48 -0700351
352 /*
353 * We really shouldn't be looking at the ->index of an
354 * unlocked page. But we're not allowed to lock these
355 * pages. So we rely upon nobody altering the ->index
356 * of this (pinned-by-us) page.
357 */
358 index = page->index;
359 if (index > next)
360 next = index;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700361 next++;
NeilBrowne0f23602006-06-23 02:05:48 -0700362 if (lock_failed)
363 continue;
364
Minchan Kim31560182011-03-22 16:32:52 -0700365 ret = invalidate_inode_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700366 unlock_page(page);
Minchan Kim31560182011-03-22 16:32:52 -0700367 /*
368 * Invalidation is a hint that the page is no longer
369 * of interest and try to speed up its reclaim.
370 */
371 if (!ret)
372 deactivate_page(page);
373 count += ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700374 if (next > end)
375 break;
376 }
377 pagevec_release(&pvec);
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -0800378 mem_cgroup_uncharge_end();
Mike Waychison28697352009-06-16 15:32:59 -0700379 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700380 }
Minchan Kim31560182011-03-22 16:32:52 -0700381 return count;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700382}
Anton Altaparmakov54bc4852007-02-10 01:45:38 -0800383EXPORT_SYMBOL(invalidate_mapping_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700384
Andrew Mortonbd4c8ce2006-09-30 23:29:29 -0700385/*
386 * This is like invalidate_complete_page(), except it ignores the page's
387 * refcount. We do this because invalidate_inode_pages2() needs stronger
388 * invalidation guarantees, and cannot afford to leave pages behind because
Anderson Briglia2706a1b2007-07-15 23:38:09 -0700389 * shrink_page_list() has a temp ref on them, or because they're transiently
390 * sitting in the lru_cache_add() pagevecs.
Andrew Mortonbd4c8ce2006-09-30 23:29:29 -0700391 */
392static int
393invalidate_complete_page2(struct address_space *mapping, struct page *page)
394{
395 if (page->mapping != mapping)
396 return 0;
397
David Howells266cf652009-04-03 16:42:36 +0100398 if (page_has_private(page) && !try_to_release_page(page, GFP_KERNEL))
Andrew Mortonbd4c8ce2006-09-30 23:29:29 -0700399 return 0;
400
Nick Piggin19fd6232008-07-25 19:45:32 -0700401 spin_lock_irq(&mapping->tree_lock);
Andrew Mortonbd4c8ce2006-09-30 23:29:29 -0700402 if (PageDirty(page))
403 goto failed;
404
Rik van Rielba470de2008-10-18 20:26:50 -0700405 clear_page_mlock(page);
David Howells266cf652009-04-03 16:42:36 +0100406 BUG_ON(page_has_private(page));
Minchan Kime64a7822011-03-22 16:32:44 -0700407 __delete_from_page_cache(page);
Nick Piggin19fd6232008-07-25 19:45:32 -0700408 spin_unlock_irq(&mapping->tree_lock);
Daisuke Nishimurae767e052009-05-28 14:34:28 -0700409 mem_cgroup_uncharge_cache_page(page);
Linus Torvalds6072d132010-12-01 13:35:19 -0500410
411 if (mapping->a_ops->freepage)
412 mapping->a_ops->freepage(page);
413
Andrew Mortonbd4c8ce2006-09-30 23:29:29 -0700414 page_cache_release(page); /* pagecache ref */
415 return 1;
416failed:
Nick Piggin19fd6232008-07-25 19:45:32 -0700417 spin_unlock_irq(&mapping->tree_lock);
Andrew Mortonbd4c8ce2006-09-30 23:29:29 -0700418 return 0;
419}
420
Trond Myklebuste3db7692007-01-10 23:15:39 -0800421static int do_launder_page(struct address_space *mapping, struct page *page)
422{
423 if (!PageDirty(page))
424 return 0;
425 if (page->mapping != mapping || mapping->a_ops->launder_page == NULL)
426 return 0;
427 return mapping->a_ops->launder_page(page);
428}
429
Linus Torvalds1da177e2005-04-16 15:20:36 -0700430/**
431 * invalidate_inode_pages2_range - remove range of pages from an address_space
Martin Waitz67be2dd2005-05-01 08:59:26 -0700432 * @mapping: the address_space
Linus Torvalds1da177e2005-04-16 15:20:36 -0700433 * @start: the page offset 'from' which to invalidate
434 * @end: the page offset 'to' which to invalidate (inclusive)
435 *
436 * Any pages which are found to be mapped into pagetables are unmapped prior to
437 * invalidation.
438 *
Hisashi Hifumi6ccfa802008-09-02 14:35:40 -0700439 * Returns -EBUSY if any pages could not be invalidated.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700440 */
441int invalidate_inode_pages2_range(struct address_space *mapping,
442 pgoff_t start, pgoff_t end)
443{
444 struct pagevec pvec;
445 pgoff_t next;
446 int i;
447 int ret = 0;
Hisashi Hifumi0dd13342008-04-28 02:12:08 -0700448 int ret2 = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700449 int did_range_unmap = 0;
450 int wrapped = 0;
451
Dan Magenheimerc515e1f2011-05-26 10:01:43 -0600452 cleancache_flush_inode(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700453 pagevec_init(&pvec, 0);
454 next = start;
Trond Myklebust7b965e02007-02-28 20:13:55 -0800455 while (next <= end && !wrapped &&
Linus Torvalds1da177e2005-04-16 15:20:36 -0700456 pagevec_lookup(&pvec, mapping, next,
457 min(end - next, (pgoff_t)PAGEVEC_SIZE - 1) + 1)) {
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -0800458 mem_cgroup_uncharge_start();
Trond Myklebust7b965e02007-02-28 20:13:55 -0800459 for (i = 0; i < pagevec_count(&pvec); i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700460 struct page *page = pvec.pages[i];
461 pgoff_t page_index;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700462
463 lock_page(page);
464 if (page->mapping != mapping) {
465 unlock_page(page);
466 continue;
467 }
468 page_index = page->index;
469 next = page_index + 1;
470 if (next == 0)
471 wrapped = 1;
472 if (page_index > end) {
473 unlock_page(page);
474 break;
475 }
476 wait_on_page_writeback(page);
Nick Piggind00806b2007-07-19 01:46:57 -0700477 if (page_mapped(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700478 if (!did_range_unmap) {
479 /*
480 * Zap the rest of the file in one hit.
481 */
482 unmap_mapping_range(mapping,
Oleg Drokin479ef592005-11-23 13:37:47 -0800483 (loff_t)page_index<<PAGE_CACHE_SHIFT,
484 (loff_t)(end - page_index + 1)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700485 << PAGE_CACHE_SHIFT,
486 0);
487 did_range_unmap = 1;
488 } else {
489 /*
490 * Just zap this page
491 */
492 unmap_mapping_range(mapping,
Oleg Drokin479ef592005-11-23 13:37:47 -0800493 (loff_t)page_index<<PAGE_CACHE_SHIFT,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700494 PAGE_CACHE_SIZE, 0);
495 }
496 }
Nick Piggind00806b2007-07-19 01:46:57 -0700497 BUG_ON(page_mapped(page));
Hisashi Hifumi0dd13342008-04-28 02:12:08 -0700498 ret2 = do_launder_page(mapping, page);
499 if (ret2 == 0) {
500 if (!invalidate_complete_page2(mapping, page))
Hisashi Hifumi6ccfa802008-09-02 14:35:40 -0700501 ret2 = -EBUSY;
Hisashi Hifumi0dd13342008-04-28 02:12:08 -0700502 }
503 if (ret2 < 0)
504 ret = ret2;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700505 unlock_page(page);
506 }
507 pagevec_release(&pvec);
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -0800508 mem_cgroup_uncharge_end();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700509 cond_resched();
510 }
Dan Magenheimerc515e1f2011-05-26 10:01:43 -0600511 cleancache_flush_inode(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700512 return ret;
513}
514EXPORT_SYMBOL_GPL(invalidate_inode_pages2_range);
515
516/**
517 * invalidate_inode_pages2 - remove all pages from an address_space
Martin Waitz67be2dd2005-05-01 08:59:26 -0700518 * @mapping: the address_space
Linus Torvalds1da177e2005-04-16 15:20:36 -0700519 *
520 * Any pages which are found to be mapped into pagetables are unmapped prior to
521 * invalidation.
522 *
Peng Taoe9de25d2009-10-19 14:48:13 +0800523 * Returns -EBUSY if any pages could not be invalidated.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700524 */
525int invalidate_inode_pages2(struct address_space *mapping)
526{
527 return invalidate_inode_pages2_range(mapping, 0, -1);
528}
529EXPORT_SYMBOL_GPL(invalidate_inode_pages2);
npiggin@suse.de25d9e2d2009-08-21 02:35:05 +1000530
531/**
532 * truncate_pagecache - unmap and remove pagecache that has been truncated
533 * @inode: inode
534 * @old: old file offset
535 * @new: new file offset
536 *
537 * inode's new i_size must already be written before truncate_pagecache
538 * is called.
539 *
540 * This function should typically be called before the filesystem
541 * releases resources associated with the freed range (eg. deallocates
542 * blocks). This way, pagecache will always stay logically coherent
543 * with on-disk format, and the filesystem would not have to deal with
544 * situations such as writepage being called for a page that has already
545 * had its underlying blocks deallocated.
546 */
547void truncate_pagecache(struct inode *inode, loff_t old, loff_t new)
548{
OGAWA Hirofumicedabed2010-01-13 21:14:09 +0900549 struct address_space *mapping = inode->i_mapping;
npiggin@suse.de25d9e2d2009-08-21 02:35:05 +1000550
OGAWA Hirofumicedabed2010-01-13 21:14:09 +0900551 /*
552 * unmap_mapping_range is called twice, first simply for
553 * efficiency so that truncate_inode_pages does fewer
554 * single-page unmaps. However after this first call, and
555 * before truncate_inode_pages finishes, it is possible for
556 * private pages to be COWed, which remain after
557 * truncate_inode_pages finishes, hence the second
558 * unmap_mapping_range call must be made for correctness.
559 */
560 unmap_mapping_range(mapping, new + PAGE_SIZE - 1, 0, 1);
561 truncate_inode_pages(mapping, new);
562 unmap_mapping_range(mapping, new + PAGE_SIZE - 1, 0, 1);
npiggin@suse.de25d9e2d2009-08-21 02:35:05 +1000563}
564EXPORT_SYMBOL(truncate_pagecache);
565
566/**
Christoph Hellwig2c27c652010-06-04 11:30:04 +0200567 * truncate_setsize - update inode and pagecache for a new file size
568 * @inode: inode
569 * @newsize: new file size
570 *
Jan Kara382e27d2011-01-20 14:44:26 -0800571 * truncate_setsize updates i_size and performs pagecache truncation (if
572 * necessary) to @newsize. It will be typically be called from the filesystem's
573 * setattr function when ATTR_SIZE is passed in.
Christoph Hellwig2c27c652010-06-04 11:30:04 +0200574 *
Jan Kara382e27d2011-01-20 14:44:26 -0800575 * Must be called with inode_mutex held and before all filesystem specific
576 * block truncation has been performed.
Christoph Hellwig2c27c652010-06-04 11:30:04 +0200577 */
578void truncate_setsize(struct inode *inode, loff_t newsize)
579{
580 loff_t oldsize;
581
582 oldsize = inode->i_size;
583 i_size_write(inode, newsize);
584
585 truncate_pagecache(inode, oldsize, newsize);
586}
587EXPORT_SYMBOL(truncate_setsize);
588
589/**
npiggin@suse.de25d9e2d2009-08-21 02:35:05 +1000590 * vmtruncate - unmap mappings "freed" by truncate() syscall
591 * @inode: inode of the file used
592 * @offset: file offset to start truncating
593 *
Christoph Hellwig2c27c652010-06-04 11:30:04 +0200594 * This function is deprecated and truncate_setsize or truncate_pagecache
595 * should be used instead, together with filesystem specific block truncation.
npiggin@suse.de25d9e2d2009-08-21 02:35:05 +1000596 */
597int vmtruncate(struct inode *inode, loff_t offset)
598{
npiggin@suse.de25d9e2d2009-08-21 02:35:05 +1000599 int error;
600
Christoph Hellwig2c27c652010-06-04 11:30:04 +0200601 error = inode_newsize_ok(inode, offset);
npiggin@suse.de25d9e2d2009-08-21 02:35:05 +1000602 if (error)
603 return error;
npiggin@suse.de7bb46a62010-05-27 01:05:33 +1000604
Christoph Hellwig2c27c652010-06-04 11:30:04 +0200605 truncate_setsize(inode, offset);
npiggin@suse.de25d9e2d2009-08-21 02:35:05 +1000606 if (inode->i_op->truncate)
607 inode->i_op->truncate(inode);
Christoph Hellwig2c27c652010-06-04 11:30:04 +0200608 return 0;
npiggin@suse.de25d9e2d2009-08-21 02:35:05 +1000609}
610EXPORT_SYMBOL(vmtruncate);
Hugh Dickins5b8ba102011-06-27 16:18:01 -0700611
612int vmtruncate_range(struct inode *inode, loff_t offset, loff_t end)
613{
614 struct address_space *mapping = inode->i_mapping;
615
616 /*
617 * If the underlying filesystem is not going to provide
618 * a way to truncate a range of blocks (punch a hole) -
619 * we should return failure right now.
620 */
621 if (!inode->i_op->truncate_range)
622 return -ENOSYS;
623
624 mutex_lock(&inode->i_mutex);
625 down_write(&inode->i_alloc_sem);
626 unmap_mapping_range(mapping, offset, (end - offset), 1);
Hugh Dickins5b8ba102011-06-27 16:18:01 -0700627 inode->i_op->truncate_range(inode, offset, end);
Hugh Dickins94c1e622011-06-27 16:18:03 -0700628 /* unmap again to remove racily COWed private pages */
629 unmap_mapping_range(mapping, offset, (end - offset), 1);
Hugh Dickins5b8ba102011-06-27 16:18:01 -0700630 up_write(&inode->i_alloc_sem);
631 mutex_unlock(&inode->i_mutex);
632
633 return 0;
634}