blob: 49feb46e77b8802803d20009f758ec65d2db0835 [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 */
Rik van Rielba470de2008-10-18 20:26:50 -070022#include "internal.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070023
24
David Howellscf9a2ae2006-08-29 19:05:54 +010025/**
Fengguang Wu28bc44d2008-02-03 18:04:10 +020026 * do_invalidatepage - invalidate part or all of a page
David Howellscf9a2ae2006-08-29 19:05:54 +010027 * @page: the page which is affected
28 * @offset: the index of the truncation point
29 *
30 * do_invalidatepage() is called when all or part of the page has become
31 * invalidated by a truncate operation.
32 *
33 * do_invalidatepage() does not have to release all buffers, but it must
34 * ensure that no dirty buffer is left outside @offset and that no I/O
35 * is underway against any of the blocks which are outside the truncation
36 * point. Because the caller is about to free (and possibly reuse) those
37 * blocks on-disk.
38 */
39void do_invalidatepage(struct page *page, unsigned long offset)
40{
41 void (*invalidatepage)(struct page *, unsigned long);
42 invalidatepage = page->mapping->a_ops->invalidatepage;
David Howells93614012006-09-30 20:45:40 +020043#ifdef CONFIG_BLOCK
David Howellscf9a2ae2006-08-29 19:05:54 +010044 if (!invalidatepage)
45 invalidatepage = block_invalidatepage;
David Howells93614012006-09-30 20:45:40 +020046#endif
David Howellscf9a2ae2006-08-29 19:05:54 +010047 if (invalidatepage)
48 (*invalidatepage)(page, offset);
49}
50
Linus Torvalds1da177e2005-04-16 15:20:36 -070051static inline void truncate_partial_page(struct page *page, unsigned partial)
52{
Christoph Lametereebd2aa2008-02-04 22:28:29 -080053 zero_user_segment(page, partial, PAGE_CACHE_SIZE);
David Howells266cf652009-04-03 16:42:36 +010054 if (page_has_private(page))
Linus Torvalds1da177e2005-04-16 15:20:36 -070055 do_invalidatepage(page, partial);
56}
57
Linus Torvaldsecdfc972007-01-26 12:47:06 -080058/*
59 * This cancels just the dirty bit on the kernel page itself, it
60 * does NOT actually remove dirty bits on any mmap's that may be
61 * around. It also leaves the page tagged dirty, so any sync
62 * activity will still find it on the dirty lists, and in particular,
63 * clear_page_dirty_for_io() will still look at the dirty bits in
64 * the VM.
65 *
66 * Doing this should *normally* only ever be done when a page
67 * is truncated, and is not actually mapped anywhere at all. However,
68 * fs/buffer.c does this when it notices that somebody has cleaned
69 * out all the buffers on a page without actually doing it through
70 * the VM. Can you say "ext3 is horribly ugly"? Tought you could.
71 */
Linus Torvaldsfba25912006-12-20 13:46:42 -080072void cancel_dirty_page(struct page *page, unsigned int account_size)
73{
Linus Torvalds8368e322006-12-23 09:25:04 -080074 if (TestClearPageDirty(page)) {
75 struct address_space *mapping = page->mapping;
76 if (mapping && mapping_cap_account_dirty(mapping)) {
77 dec_zone_page_state(page, NR_FILE_DIRTY);
Peter Zijlstrac9e51e42007-10-16 23:25:47 -070078 dec_bdi_stat(mapping->backing_dev_info,
79 BDI_RECLAIMABLE);
Linus Torvalds8368e322006-12-23 09:25:04 -080080 if (account_size)
81 task_io_account_cancelled_write(account_size);
82 }
Andrew Morton3e67c092006-12-21 11:00:33 -080083 }
Linus Torvaldsfba25912006-12-20 13:46:42 -080084}
Linus Torvalds8368e322006-12-23 09:25:04 -080085EXPORT_SYMBOL(cancel_dirty_page);
Linus Torvaldsfba25912006-12-20 13:46:42 -080086
Linus Torvalds1da177e2005-04-16 15:20:36 -070087/*
88 * If truncate cannot remove the fs-private metadata from the page, the page
Shaohua Li62e1c552008-02-04 22:29:33 -080089 * becomes orphaned. It will be left on the LRU and may even be mapped into
Nick Piggin54cb8822007-07-19 01:46:59 -070090 * user pagetables if we're racing with filemap_fault().
Linus Torvalds1da177e2005-04-16 15:20:36 -070091 *
92 * We need to bale out if page->mapping is no longer equal to the original
93 * mapping. This happens a) when the VM reclaimed the page while we waited on
Andrew Mortonfc0ecff2007-02-10 01:45:39 -080094 * its lock, b) when a concurrent invalidate_mapping_pages got there first and
Linus Torvalds1da177e2005-04-16 15:20:36 -070095 * c) when tmpfs swizzles a page between a tmpfs inode and swapper_space.
96 */
Nick Piggin750b4982009-09-16 11:50:12 +020097static int
Linus Torvalds1da177e2005-04-16 15:20:36 -070098truncate_complete_page(struct address_space *mapping, struct page *page)
99{
100 if (page->mapping != mapping)
Nick Piggin750b4982009-09-16 11:50:12 +0200101 return -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700102
David Howells266cf652009-04-03 16:42:36 +0100103 if (page_has_private(page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700104 do_invalidatepage(page, 0);
105
Bjorn Steinbrinka2b34562008-02-04 22:29:28 -0800106 cancel_dirty_page(page, PAGE_CACHE_SIZE);
107
Rik van Rielba470de2008-10-18 20:26:50 -0700108 clear_page_mlock(page);
Nick Piggin787d2212007-07-17 04:03:34 -0700109 remove_from_page_cache(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700110 ClearPageMappedToDisk(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700111 page_cache_release(page); /* pagecache ref */
Nick Piggin750b4982009-09-16 11:50:12 +0200112 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700113}
114
115/*
Andrew Mortonfc0ecff2007-02-10 01:45:39 -0800116 * This is for invalidate_mapping_pages(). That function can be called at
Linus Torvalds1da177e2005-04-16 15:20:36 -0700117 * any time, and is not supposed to throw away dirty pages. But pages can
Nick Piggin0fd0e6b2006-09-27 01:50:02 -0700118 * be marked dirty at any time too, so use remove_mapping which safely
119 * discards clean, unused pages.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700120 *
121 * Returns non-zero if the page was successfully invalidated.
122 */
123static int
124invalidate_complete_page(struct address_space *mapping, struct page *page)
125{
Nick Piggin0fd0e6b2006-09-27 01:50:02 -0700126 int ret;
127
Linus Torvalds1da177e2005-04-16 15:20:36 -0700128 if (page->mapping != mapping)
129 return 0;
130
David Howells266cf652009-04-03 16:42:36 +0100131 if (page_has_private(page) && !try_to_release_page(page, 0))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700132 return 0;
133
Rik van Rielba470de2008-10-18 20:26:50 -0700134 clear_page_mlock(page);
Nick Piggin0fd0e6b2006-09-27 01:50:02 -0700135 ret = remove_mapping(mapping, page);
Nick Piggin0fd0e6b2006-09-27 01:50:02 -0700136
137 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700138}
139
Nick Piggin750b4982009-09-16 11:50:12 +0200140int truncate_inode_page(struct address_space *mapping, struct page *page)
141{
142 if (page_mapped(page)) {
143 unmap_mapping_range(mapping,
144 (loff_t)page->index << PAGE_CACHE_SHIFT,
145 PAGE_CACHE_SIZE, 0);
146 }
147 return truncate_complete_page(mapping, page);
148}
149
Wu Fengguang83f78662009-09-16 11:50:13 +0200150/*
Andi Kleen25718732009-09-16 11:50:13 +0200151 * Used to get rid of pages on hardware memory corruption.
152 */
153int generic_error_remove_page(struct address_space *mapping, struct page *page)
154{
155 if (!mapping)
156 return -EINVAL;
157 /*
158 * Only punch for normal data pages for now.
159 * Handling other types like directories would need more auditing.
160 */
161 if (!S_ISREG(mapping->host->i_mode))
162 return -EIO;
163 return truncate_inode_page(mapping, page);
164}
165EXPORT_SYMBOL(generic_error_remove_page);
166
167/*
Wu Fengguang83f78662009-09-16 11:50:13 +0200168 * Safely invalidate one page from its pagecache mapping.
169 * It only drops clean, unused pages. The page must be locked.
170 *
171 * Returns 1 if the page is successfully invalidated, otherwise 0.
172 */
173int invalidate_inode_page(struct page *page)
174{
175 struct address_space *mapping = page_mapping(page);
176 if (!mapping)
177 return 0;
178 if (PageDirty(page) || PageWriteback(page))
179 return 0;
180 if (page_mapped(page))
181 return 0;
182 return invalidate_complete_page(mapping, page);
183}
184
Linus Torvalds1da177e2005-04-16 15:20:36 -0700185/**
Randy Dunlap06432452008-02-29 22:03:15 -0800186 * truncate_inode_pages - truncate range of pages specified by start & end byte offsets
Linus Torvalds1da177e2005-04-16 15:20:36 -0700187 * @mapping: mapping to truncate
188 * @lstart: offset from which to truncate
Hans Reiserd7339072006-01-06 00:10:36 -0800189 * @lend: offset to which to truncate
Linus Torvalds1da177e2005-04-16 15:20:36 -0700190 *
Hans Reiserd7339072006-01-06 00:10:36 -0800191 * Truncate the page cache, removing the pages that are between
192 * specified offsets (and zeroing out partial page
193 * (if lstart is not page aligned)).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700194 *
195 * Truncate takes two passes - the first pass is nonblocking. It will not
196 * block on page locks and it will not block on writeback. The second pass
197 * will wait. This is to prevent as much IO as possible in the affected region.
198 * The first pass will remove most pages, so the search cost of the second pass
199 * is low.
200 *
201 * When looking at page->index outside the page lock we need to be careful to
202 * copy it into a local to avoid races (it could change at any time).
203 *
204 * We pass down the cache-hot hint to the page freeing code. Even if the
205 * mapping is large, it is probably the case that the final pages are the most
206 * recently touched, and freeing happens in ascending file offset order.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700207 */
Hans Reiserd7339072006-01-06 00:10:36 -0800208void truncate_inode_pages_range(struct address_space *mapping,
209 loff_t lstart, loff_t lend)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700210{
211 const pgoff_t start = (lstart + PAGE_CACHE_SIZE-1) >> PAGE_CACHE_SHIFT;
Hans Reiserd7339072006-01-06 00:10:36 -0800212 pgoff_t end;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700213 const unsigned partial = lstart & (PAGE_CACHE_SIZE - 1);
214 struct pagevec pvec;
215 pgoff_t next;
216 int i;
217
218 if (mapping->nrpages == 0)
219 return;
220
Hans Reiserd7339072006-01-06 00:10:36 -0800221 BUG_ON((lend & (PAGE_CACHE_SIZE - 1)) != (PAGE_CACHE_SIZE - 1));
222 end = (lend >> PAGE_CACHE_SHIFT);
223
Linus Torvalds1da177e2005-04-16 15:20:36 -0700224 pagevec_init(&pvec, 0);
225 next = start;
Hans Reiserd7339072006-01-06 00:10:36 -0800226 while (next <= end &&
227 pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700228 for (i = 0; i < pagevec_count(&pvec); i++) {
229 struct page *page = pvec.pages[i];
230 pgoff_t page_index = page->index;
231
Hans Reiserd7339072006-01-06 00:10:36 -0800232 if (page_index > end) {
233 next = page_index;
234 break;
235 }
236
Linus Torvalds1da177e2005-04-16 15:20:36 -0700237 if (page_index > next)
238 next = page_index;
239 next++;
Nick Piggin529ae9a2008-08-02 12:01:03 +0200240 if (!trylock_page(page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700241 continue;
242 if (PageWriteback(page)) {
243 unlock_page(page);
244 continue;
245 }
Nick Piggin750b4982009-09-16 11:50:12 +0200246 truncate_inode_page(mapping, page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700247 unlock_page(page);
248 }
249 pagevec_release(&pvec);
250 cond_resched();
251 }
252
253 if (partial) {
254 struct page *page = find_lock_page(mapping, start - 1);
255 if (page) {
256 wait_on_page_writeback(page);
257 truncate_partial_page(page, partial);
258 unlock_page(page);
259 page_cache_release(page);
260 }
261 }
262
263 next = start;
264 for ( ; ; ) {
265 cond_resched();
266 if (!pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) {
267 if (next == start)
268 break;
269 next = start;
270 continue;
271 }
Hans Reiserd7339072006-01-06 00:10:36 -0800272 if (pvec.pages[0]->index > end) {
273 pagevec_release(&pvec);
274 break;
275 }
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -0800276 mem_cgroup_uncharge_start();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700277 for (i = 0; i < pagevec_count(&pvec); i++) {
278 struct page *page = pvec.pages[i];
279
Hans Reiserd7339072006-01-06 00:10:36 -0800280 if (page->index > end)
281 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700282 lock_page(page);
283 wait_on_page_writeback(page);
Nick Piggin750b4982009-09-16 11:50:12 +0200284 truncate_inode_page(mapping, page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700285 if (page->index > next)
286 next = page->index;
287 next++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700288 unlock_page(page);
289 }
290 pagevec_release(&pvec);
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -0800291 mem_cgroup_uncharge_end();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700292 }
293}
Hans Reiserd7339072006-01-06 00:10:36 -0800294EXPORT_SYMBOL(truncate_inode_pages_range);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700295
Hans Reiserd7339072006-01-06 00:10:36 -0800296/**
297 * truncate_inode_pages - truncate *all* the pages from an offset
298 * @mapping: mapping to truncate
299 * @lstart: offset from which to truncate
300 *
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800301 * Called under (and serialised by) inode->i_mutex.
Hans Reiserd7339072006-01-06 00:10:36 -0800302 */
303void truncate_inode_pages(struct address_space *mapping, loff_t lstart)
304{
305 truncate_inode_pages_range(mapping, lstart, (loff_t)-1);
306}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700307EXPORT_SYMBOL(truncate_inode_pages);
308
Mike Waychison28697352009-06-16 15:32:59 -0700309/**
310 * invalidate_mapping_pages - Invalidate all the unlocked pages of one inode
311 * @mapping: the address_space which holds the pages to invalidate
312 * @start: the offset 'from' which to invalidate
313 * @end: the offset 'to' which to invalidate (inclusive)
314 *
315 * This function only removes the unlocked pages, if you want to
316 * remove all the pages of one inode, you must call truncate_inode_pages.
317 *
318 * invalidate_mapping_pages() will not block on IO activity. It will not
319 * invalidate pages which are dirty, locked, under writeback or mapped into
320 * pagetables.
321 */
322unsigned long invalidate_mapping_pages(struct address_space *mapping,
323 pgoff_t start, pgoff_t end)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700324{
325 struct pagevec pvec;
326 pgoff_t next = start;
327 unsigned long ret = 0;
328 int i;
329
330 pagevec_init(&pvec, 0);
331 while (next <= end &&
332 pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) {
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -0800333 mem_cgroup_uncharge_start();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700334 for (i = 0; i < pagevec_count(&pvec); i++) {
335 struct page *page = pvec.pages[i];
NeilBrowne0f23602006-06-23 02:05:48 -0700336 pgoff_t index;
337 int lock_failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700338
Nick Piggin529ae9a2008-08-02 12:01:03 +0200339 lock_failed = !trylock_page(page);
NeilBrowne0f23602006-06-23 02:05:48 -0700340
341 /*
342 * We really shouldn't be looking at the ->index of an
343 * unlocked page. But we're not allowed to lock these
344 * pages. So we rely upon nobody altering the ->index
345 * of this (pinned-by-us) page.
346 */
347 index = page->index;
348 if (index > next)
349 next = index;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700350 next++;
NeilBrowne0f23602006-06-23 02:05:48 -0700351 if (lock_failed)
352 continue;
353
Wu Fengguang83f78662009-09-16 11:50:13 +0200354 ret += invalidate_inode_page(page);
355
Linus Torvalds1da177e2005-04-16 15:20:36 -0700356 unlock_page(page);
357 if (next > end)
358 break;
359 }
360 pagevec_release(&pvec);
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -0800361 mem_cgroup_uncharge_end();
Mike Waychison28697352009-06-16 15:32:59 -0700362 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700363 }
364 return ret;
365}
Anton Altaparmakov54bc4852007-02-10 01:45:38 -0800366EXPORT_SYMBOL(invalidate_mapping_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700367
Andrew Mortonbd4c8ce2006-09-30 23:29:29 -0700368/*
369 * This is like invalidate_complete_page(), except it ignores the page's
370 * refcount. We do this because invalidate_inode_pages2() needs stronger
371 * invalidation guarantees, and cannot afford to leave pages behind because
Anderson Briglia2706a1b2007-07-15 23:38:09 -0700372 * shrink_page_list() has a temp ref on them, or because they're transiently
373 * sitting in the lru_cache_add() pagevecs.
Andrew Mortonbd4c8ce2006-09-30 23:29:29 -0700374 */
375static int
376invalidate_complete_page2(struct address_space *mapping, struct page *page)
377{
378 if (page->mapping != mapping)
379 return 0;
380
David Howells266cf652009-04-03 16:42:36 +0100381 if (page_has_private(page) && !try_to_release_page(page, GFP_KERNEL))
Andrew Mortonbd4c8ce2006-09-30 23:29:29 -0700382 return 0;
383
Nick Piggin19fd6232008-07-25 19:45:32 -0700384 spin_lock_irq(&mapping->tree_lock);
Andrew Mortonbd4c8ce2006-09-30 23:29:29 -0700385 if (PageDirty(page))
386 goto failed;
387
Rik van Rielba470de2008-10-18 20:26:50 -0700388 clear_page_mlock(page);
David Howells266cf652009-04-03 16:42:36 +0100389 BUG_ON(page_has_private(page));
Andrew Mortonbd4c8ce2006-09-30 23:29:29 -0700390 __remove_from_page_cache(page);
Nick Piggin19fd6232008-07-25 19:45:32 -0700391 spin_unlock_irq(&mapping->tree_lock);
Daisuke Nishimurae767e052009-05-28 14:34:28 -0700392 mem_cgroup_uncharge_cache_page(page);
Linus Torvalds6072d132010-12-01 13:35:19 -0500393
394 if (mapping->a_ops->freepage)
395 mapping->a_ops->freepage(page);
396
Andrew Mortonbd4c8ce2006-09-30 23:29:29 -0700397 page_cache_release(page); /* pagecache ref */
398 return 1;
399failed:
Nick Piggin19fd6232008-07-25 19:45:32 -0700400 spin_unlock_irq(&mapping->tree_lock);
Andrew Mortonbd4c8ce2006-09-30 23:29:29 -0700401 return 0;
402}
403
Trond Myklebuste3db7692007-01-10 23:15:39 -0800404static int do_launder_page(struct address_space *mapping, struct page *page)
405{
406 if (!PageDirty(page))
407 return 0;
408 if (page->mapping != mapping || mapping->a_ops->launder_page == NULL)
409 return 0;
410 return mapping->a_ops->launder_page(page);
411}
412
Linus Torvalds1da177e2005-04-16 15:20:36 -0700413/**
414 * invalidate_inode_pages2_range - remove range of pages from an address_space
Martin Waitz67be2dd2005-05-01 08:59:26 -0700415 * @mapping: the address_space
Linus Torvalds1da177e2005-04-16 15:20:36 -0700416 * @start: the page offset 'from' which to invalidate
417 * @end: the page offset 'to' which to invalidate (inclusive)
418 *
419 * Any pages which are found to be mapped into pagetables are unmapped prior to
420 * invalidation.
421 *
Hisashi Hifumi6ccfa802008-09-02 14:35:40 -0700422 * Returns -EBUSY if any pages could not be invalidated.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700423 */
424int invalidate_inode_pages2_range(struct address_space *mapping,
425 pgoff_t start, pgoff_t end)
426{
427 struct pagevec pvec;
428 pgoff_t next;
429 int i;
430 int ret = 0;
Hisashi Hifumi0dd13342008-04-28 02:12:08 -0700431 int ret2 = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700432 int did_range_unmap = 0;
433 int wrapped = 0;
434
435 pagevec_init(&pvec, 0);
436 next = start;
Trond Myklebust7b965e02007-02-28 20:13:55 -0800437 while (next <= end && !wrapped &&
Linus Torvalds1da177e2005-04-16 15:20:36 -0700438 pagevec_lookup(&pvec, mapping, next,
439 min(end - next, (pgoff_t)PAGEVEC_SIZE - 1) + 1)) {
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -0800440 mem_cgroup_uncharge_start();
Trond Myklebust7b965e02007-02-28 20:13:55 -0800441 for (i = 0; i < pagevec_count(&pvec); i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700442 struct page *page = pvec.pages[i];
443 pgoff_t page_index;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700444
445 lock_page(page);
446 if (page->mapping != mapping) {
447 unlock_page(page);
448 continue;
449 }
450 page_index = page->index;
451 next = page_index + 1;
452 if (next == 0)
453 wrapped = 1;
454 if (page_index > end) {
455 unlock_page(page);
456 break;
457 }
458 wait_on_page_writeback(page);
Nick Piggind00806b2007-07-19 01:46:57 -0700459 if (page_mapped(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700460 if (!did_range_unmap) {
461 /*
462 * Zap the rest of the file in one hit.
463 */
464 unmap_mapping_range(mapping,
Oleg Drokin479ef592005-11-23 13:37:47 -0800465 (loff_t)page_index<<PAGE_CACHE_SHIFT,
466 (loff_t)(end - page_index + 1)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700467 << PAGE_CACHE_SHIFT,
468 0);
469 did_range_unmap = 1;
470 } else {
471 /*
472 * Just zap this page
473 */
474 unmap_mapping_range(mapping,
Oleg Drokin479ef592005-11-23 13:37:47 -0800475 (loff_t)page_index<<PAGE_CACHE_SHIFT,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700476 PAGE_CACHE_SIZE, 0);
477 }
478 }
Nick Piggind00806b2007-07-19 01:46:57 -0700479 BUG_ON(page_mapped(page));
Hisashi Hifumi0dd13342008-04-28 02:12:08 -0700480 ret2 = do_launder_page(mapping, page);
481 if (ret2 == 0) {
482 if (!invalidate_complete_page2(mapping, page))
Hisashi Hifumi6ccfa802008-09-02 14:35:40 -0700483 ret2 = -EBUSY;
Hisashi Hifumi0dd13342008-04-28 02:12:08 -0700484 }
485 if (ret2 < 0)
486 ret = ret2;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700487 unlock_page(page);
488 }
489 pagevec_release(&pvec);
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -0800490 mem_cgroup_uncharge_end();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700491 cond_resched();
492 }
493 return ret;
494}
495EXPORT_SYMBOL_GPL(invalidate_inode_pages2_range);
496
497/**
498 * invalidate_inode_pages2 - remove all pages from an address_space
Martin Waitz67be2dd2005-05-01 08:59:26 -0700499 * @mapping: the address_space
Linus Torvalds1da177e2005-04-16 15:20:36 -0700500 *
501 * Any pages which are found to be mapped into pagetables are unmapped prior to
502 * invalidation.
503 *
Peng Taoe9de25d2009-10-19 14:48:13 +0800504 * Returns -EBUSY if any pages could not be invalidated.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700505 */
506int invalidate_inode_pages2(struct address_space *mapping)
507{
508 return invalidate_inode_pages2_range(mapping, 0, -1);
509}
510EXPORT_SYMBOL_GPL(invalidate_inode_pages2);
npiggin@suse.de25d9e2d2009-08-21 02:35:05 +1000511
512/**
513 * truncate_pagecache - unmap and remove pagecache that has been truncated
514 * @inode: inode
515 * @old: old file offset
516 * @new: new file offset
517 *
518 * inode's new i_size must already be written before truncate_pagecache
519 * is called.
520 *
521 * This function should typically be called before the filesystem
522 * releases resources associated with the freed range (eg. deallocates
523 * blocks). This way, pagecache will always stay logically coherent
524 * with on-disk format, and the filesystem would not have to deal with
525 * situations such as writepage being called for a page that has already
526 * had its underlying blocks deallocated.
527 */
528void truncate_pagecache(struct inode *inode, loff_t old, loff_t new)
529{
OGAWA Hirofumicedabed2010-01-13 21:14:09 +0900530 struct address_space *mapping = inode->i_mapping;
npiggin@suse.de25d9e2d2009-08-21 02:35:05 +1000531
OGAWA Hirofumicedabed2010-01-13 21:14:09 +0900532 /*
533 * unmap_mapping_range is called twice, first simply for
534 * efficiency so that truncate_inode_pages does fewer
535 * single-page unmaps. However after this first call, and
536 * before truncate_inode_pages finishes, it is possible for
537 * private pages to be COWed, which remain after
538 * truncate_inode_pages finishes, hence the second
539 * unmap_mapping_range call must be made for correctness.
540 */
541 unmap_mapping_range(mapping, new + PAGE_SIZE - 1, 0, 1);
542 truncate_inode_pages(mapping, new);
543 unmap_mapping_range(mapping, new + PAGE_SIZE - 1, 0, 1);
npiggin@suse.de25d9e2d2009-08-21 02:35:05 +1000544}
545EXPORT_SYMBOL(truncate_pagecache);
546
547/**
Christoph Hellwig2c27c652010-06-04 11:30:04 +0200548 * truncate_setsize - update inode and pagecache for a new file size
549 * @inode: inode
550 * @newsize: new file size
551 *
Jan Kara382e27d2011-01-20 14:44:26 -0800552 * truncate_setsize updates i_size and performs pagecache truncation (if
553 * necessary) to @newsize. It will be typically be called from the filesystem's
554 * setattr function when ATTR_SIZE is passed in.
Christoph Hellwig2c27c652010-06-04 11:30:04 +0200555 *
Jan Kara382e27d2011-01-20 14:44:26 -0800556 * Must be called with inode_mutex held and before all filesystem specific
557 * block truncation has been performed.
Christoph Hellwig2c27c652010-06-04 11:30:04 +0200558 */
559void truncate_setsize(struct inode *inode, loff_t newsize)
560{
561 loff_t oldsize;
562
563 oldsize = inode->i_size;
564 i_size_write(inode, newsize);
565
566 truncate_pagecache(inode, oldsize, newsize);
567}
568EXPORT_SYMBOL(truncate_setsize);
569
570/**
npiggin@suse.de25d9e2d2009-08-21 02:35:05 +1000571 * vmtruncate - unmap mappings "freed" by truncate() syscall
572 * @inode: inode of the file used
573 * @offset: file offset to start truncating
574 *
Christoph Hellwig2c27c652010-06-04 11:30:04 +0200575 * This function is deprecated and truncate_setsize or truncate_pagecache
576 * should be used instead, together with filesystem specific block truncation.
npiggin@suse.de25d9e2d2009-08-21 02:35:05 +1000577 */
578int vmtruncate(struct inode *inode, loff_t offset)
579{
npiggin@suse.de25d9e2d2009-08-21 02:35:05 +1000580 int error;
581
Christoph Hellwig2c27c652010-06-04 11:30:04 +0200582 error = inode_newsize_ok(inode, offset);
npiggin@suse.de25d9e2d2009-08-21 02:35:05 +1000583 if (error)
584 return error;
npiggin@suse.de7bb46a62010-05-27 01:05:33 +1000585
Christoph Hellwig2c27c652010-06-04 11:30:04 +0200586 truncate_setsize(inode, offset);
npiggin@suse.de25d9e2d2009-08-21 02:35:05 +1000587 if (inode->i_op->truncate)
588 inode->i_op->truncate(inode);
Christoph Hellwig2c27c652010-06-04 11:30:04 +0200589 return 0;
npiggin@suse.de25d9e2d2009-08-21 02:35:05 +1000590}
591EXPORT_SYMBOL(vmtruncate);