blob: 61a183b89df6d15c358e7afc2b418411abe5e728 [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>
Paul Gortmakerb95f1b312011-10-16 02:01:52 -040015#include <linux/export.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070016#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 Magenheimer31677602011-09-21 11:56:28 -040055 cleancache_invalidate_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/**
Liu Bo73c1e202012-02-21 10:57:20 +0800187 * truncate_inode_pages_range - 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 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700202 * We pass down the cache-hot hint to the page freeing code. Even if the
203 * mapping is large, it is probably the case that the final pages are the most
204 * recently touched, and freeing happens in ascending file offset order.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700205 */
Hans Reiserd7339072006-01-06 00:10:36 -0800206void truncate_inode_pages_range(struct address_space *mapping,
207 loff_t lstart, loff_t lend)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700208{
209 const pgoff_t start = (lstart + PAGE_CACHE_SIZE-1) >> PAGE_CACHE_SHIFT;
210 const unsigned partial = lstart & (PAGE_CACHE_SIZE - 1);
211 struct pagevec pvec;
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700212 pgoff_t index;
213 pgoff_t end;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700214 int i;
215
Dan Magenheimer31677602011-09-21 11:56:28 -0400216 cleancache_invalidate_inode(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700217 if (mapping->nrpages == 0)
218 return;
219
Hans Reiserd7339072006-01-06 00:10:36 -0800220 BUG_ON((lend & (PAGE_CACHE_SIZE - 1)) != (PAGE_CACHE_SIZE - 1));
221 end = (lend >> PAGE_CACHE_SHIFT);
222
Linus Torvalds1da177e2005-04-16 15:20:36 -0700223 pagevec_init(&pvec, 0);
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700224 index = start;
225 while (index <= end && pagevec_lookup(&pvec, mapping, index,
226 min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1)) {
Hugh Dickinse5598f82011-02-25 14:44:29 -0800227 mem_cgroup_uncharge_start();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700228 for (i = 0; i < pagevec_count(&pvec); i++) {
229 struct page *page = pvec.pages[i];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700230
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700231 /* We rely upon deletion not changing page->index */
232 index = page->index;
233 if (index > end)
Hans Reiserd7339072006-01-06 00:10:36 -0800234 break;
Hans Reiserd7339072006-01-06 00:10:36 -0800235
Nick Piggin529ae9a2008-08-02 12:01:03 +0200236 if (!trylock_page(page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700237 continue;
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700238 WARN_ON(page->index != index);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700239 if (PageWriteback(page)) {
240 unlock_page(page);
241 continue;
242 }
Nick Piggin750b4982009-09-16 11:50:12 +0200243 truncate_inode_page(mapping, page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700244 unlock_page(page);
245 }
246 pagevec_release(&pvec);
Hugh Dickinse5598f82011-02-25 14:44:29 -0800247 mem_cgroup_uncharge_end();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700248 cond_resched();
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700249 index++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700250 }
251
252 if (partial) {
253 struct page *page = find_lock_page(mapping, start - 1);
254 if (page) {
255 wait_on_page_writeback(page);
256 truncate_partial_page(page, partial);
257 unlock_page(page);
258 page_cache_release(page);
259 }
260 }
261
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700262 index = start;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700263 for ( ; ; ) {
264 cond_resched();
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700265 if (!pagevec_lookup(&pvec, mapping, index,
266 min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1)) {
267 if (index == start)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700268 break;
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700269 index = start;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700270 continue;
271 }
Hugh Dickinsd0823572011-07-25 17:12:25 -0700272 if (index == start && pvec.pages[0]->index > end) {
Hans Reiserd7339072006-01-06 00:10:36 -0800273 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
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700280 /* We rely upon deletion not changing page->index */
281 index = page->index;
282 if (index > end)
Hans Reiserd7339072006-01-06 00:10:36 -0800283 break;
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700284
Linus Torvalds1da177e2005-04-16 15:20:36 -0700285 lock_page(page);
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700286 WARN_ON(page->index != index);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700287 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 unlock_page(page);
290 }
291 pagevec_release(&pvec);
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -0800292 mem_cgroup_uncharge_end();
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700293 index++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700294 }
Dan Magenheimer31677602011-09-21 11:56:28 -0400295 cleancache_invalidate_inode(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700296}
Hans Reiserd7339072006-01-06 00:10:36 -0800297EXPORT_SYMBOL(truncate_inode_pages_range);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700298
Hans Reiserd7339072006-01-06 00:10:36 -0800299/**
300 * truncate_inode_pages - truncate *all* the pages from an offset
301 * @mapping: mapping to truncate
302 * @lstart: offset from which to truncate
303 *
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800304 * Called under (and serialised by) inode->i_mutex.
Jan Kara08142572011-06-27 16:18:10 -0700305 *
306 * Note: When this function returns, there can be a page in the process of
307 * deletion (inside __delete_from_page_cache()) in the specified range. Thus
308 * mapping->nrpages can be non-zero when this function returns even after
309 * truncation of the whole mapping.
Hans Reiserd7339072006-01-06 00:10:36 -0800310 */
311void truncate_inode_pages(struct address_space *mapping, loff_t lstart)
312{
313 truncate_inode_pages_range(mapping, lstart, (loff_t)-1);
314}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700315EXPORT_SYMBOL(truncate_inode_pages);
316
Mike Waychison28697352009-06-16 15:32:59 -0700317/**
318 * invalidate_mapping_pages - Invalidate all the unlocked pages of one inode
319 * @mapping: the address_space which holds the pages to invalidate
320 * @start: the offset 'from' which to invalidate
321 * @end: the offset 'to' which to invalidate (inclusive)
322 *
323 * This function only removes the unlocked pages, if you want to
324 * remove all the pages of one inode, you must call truncate_inode_pages.
325 *
326 * invalidate_mapping_pages() will not block on IO activity. It will not
327 * invalidate pages which are dirty, locked, under writeback or mapped into
328 * pagetables.
329 */
330unsigned long invalidate_mapping_pages(struct address_space *mapping,
Minchan Kim31560182011-03-22 16:32:52 -0700331 pgoff_t start, pgoff_t end)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700332{
333 struct pagevec pvec;
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700334 pgoff_t index = start;
Minchan Kim31560182011-03-22 16:32:52 -0700335 unsigned long ret;
336 unsigned long count = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700337 int i;
338
Hugh Dickins31475dd2011-08-03 16:21:27 -0700339 /*
340 * Note: this function may get called on a shmem/tmpfs mapping:
341 * pagevec_lookup() might then return 0 prematurely (because it
342 * got a gangful of swap entries); but it's hardly worth worrying
343 * about - it can rarely have anything to free from such a mapping
344 * (most pages are dirty), and already skips over any difficulties.
345 */
346
Linus Torvalds1da177e2005-04-16 15:20:36 -0700347 pagevec_init(&pvec, 0);
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700348 while (index <= end && pagevec_lookup(&pvec, mapping, index,
349 min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1)) {
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -0800350 mem_cgroup_uncharge_start();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700351 for (i = 0; i < pagevec_count(&pvec); i++) {
352 struct page *page = pvec.pages[i];
353
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700354 /* We rely upon deletion not changing page->index */
NeilBrowne0f23602006-06-23 02:05:48 -0700355 index = page->index;
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700356 if (index > end)
357 break;
NeilBrowne0f23602006-06-23 02:05:48 -0700358
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700359 if (!trylock_page(page))
360 continue;
361 WARN_ON(page->index != index);
Minchan Kim31560182011-03-22 16:32:52 -0700362 ret = invalidate_inode_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700363 unlock_page(page);
Minchan Kim31560182011-03-22 16:32:52 -0700364 /*
365 * Invalidation is a hint that the page is no longer
366 * of interest and try to speed up its reclaim.
367 */
368 if (!ret)
369 deactivate_page(page);
370 count += ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700371 }
372 pagevec_release(&pvec);
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -0800373 mem_cgroup_uncharge_end();
Mike Waychison28697352009-06-16 15:32:59 -0700374 cond_resched();
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700375 index++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700376 }
Minchan Kim31560182011-03-22 16:32:52 -0700377 return count;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700378}
Anton Altaparmakov54bc4852007-02-10 01:45:38 -0800379EXPORT_SYMBOL(invalidate_mapping_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700380
Andrew Mortonbd4c8ce2006-09-30 23:29:29 -0700381/*
382 * This is like invalidate_complete_page(), except it ignores the page's
383 * refcount. We do this because invalidate_inode_pages2() needs stronger
384 * invalidation guarantees, and cannot afford to leave pages behind because
Anderson Briglia2706a1b2007-07-15 23:38:09 -0700385 * shrink_page_list() has a temp ref on them, or because they're transiently
386 * sitting in the lru_cache_add() pagevecs.
Andrew Mortonbd4c8ce2006-09-30 23:29:29 -0700387 */
388static int
389invalidate_complete_page2(struct address_space *mapping, struct page *page)
390{
391 if (page->mapping != mapping)
392 return 0;
393
David Howells266cf652009-04-03 16:42:36 +0100394 if (page_has_private(page) && !try_to_release_page(page, GFP_KERNEL))
Andrew Mortonbd4c8ce2006-09-30 23:29:29 -0700395 return 0;
396
Nick Piggin19fd6232008-07-25 19:45:32 -0700397 spin_lock_irq(&mapping->tree_lock);
Andrew Mortonbd4c8ce2006-09-30 23:29:29 -0700398 if (PageDirty(page))
399 goto failed;
400
Rik van Rielba470de2008-10-18 20:26:50 -0700401 clear_page_mlock(page);
David Howells266cf652009-04-03 16:42:36 +0100402 BUG_ON(page_has_private(page));
Minchan Kime64a7822011-03-22 16:32:44 -0700403 __delete_from_page_cache(page);
Nick Piggin19fd6232008-07-25 19:45:32 -0700404 spin_unlock_irq(&mapping->tree_lock);
Daisuke Nishimurae767e052009-05-28 14:34:28 -0700405 mem_cgroup_uncharge_cache_page(page);
Linus Torvalds6072d132010-12-01 13:35:19 -0500406
407 if (mapping->a_ops->freepage)
408 mapping->a_ops->freepage(page);
409
Andrew Mortonbd4c8ce2006-09-30 23:29:29 -0700410 page_cache_release(page); /* pagecache ref */
411 return 1;
412failed:
Nick Piggin19fd6232008-07-25 19:45:32 -0700413 spin_unlock_irq(&mapping->tree_lock);
Andrew Mortonbd4c8ce2006-09-30 23:29:29 -0700414 return 0;
415}
416
Trond Myklebuste3db7692007-01-10 23:15:39 -0800417static int do_launder_page(struct address_space *mapping, struct page *page)
418{
419 if (!PageDirty(page))
420 return 0;
421 if (page->mapping != mapping || mapping->a_ops->launder_page == NULL)
422 return 0;
423 return mapping->a_ops->launder_page(page);
424}
425
Linus Torvalds1da177e2005-04-16 15:20:36 -0700426/**
427 * invalidate_inode_pages2_range - remove range of pages from an address_space
Martin Waitz67be2dd2005-05-01 08:59:26 -0700428 * @mapping: the address_space
Linus Torvalds1da177e2005-04-16 15:20:36 -0700429 * @start: the page offset 'from' which to invalidate
430 * @end: the page offset 'to' which to invalidate (inclusive)
431 *
432 * Any pages which are found to be mapped into pagetables are unmapped prior to
433 * invalidation.
434 *
Hisashi Hifumi6ccfa802008-09-02 14:35:40 -0700435 * Returns -EBUSY if any pages could not be invalidated.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700436 */
437int invalidate_inode_pages2_range(struct address_space *mapping,
438 pgoff_t start, pgoff_t end)
439{
440 struct pagevec pvec;
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700441 pgoff_t index;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700442 int i;
443 int ret = 0;
Hisashi Hifumi0dd13342008-04-28 02:12:08 -0700444 int ret2 = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700445 int did_range_unmap = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700446
Dan Magenheimer31677602011-09-21 11:56:28 -0400447 cleancache_invalidate_inode(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700448 pagevec_init(&pvec, 0);
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700449 index = start;
450 while (index <= end && pagevec_lookup(&pvec, mapping, index,
451 min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1)) {
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -0800452 mem_cgroup_uncharge_start();
Trond Myklebust7b965e02007-02-28 20:13:55 -0800453 for (i = 0; i < pagevec_count(&pvec); i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700454 struct page *page = pvec.pages[i];
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700455
456 /* We rely upon deletion not changing page->index */
457 index = page->index;
458 if (index > end)
459 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700460
461 lock_page(page);
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700462 WARN_ON(page->index != index);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700463 if (page->mapping != mapping) {
464 unlock_page(page);
465 continue;
466 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700467 wait_on_page_writeback(page);
Nick Piggind00806b2007-07-19 01:46:57 -0700468 if (page_mapped(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700469 if (!did_range_unmap) {
470 /*
471 * Zap the rest of the file in one hit.
472 */
473 unmap_mapping_range(mapping,
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700474 (loff_t)index << PAGE_CACHE_SHIFT,
475 (loff_t)(1 + end - index)
476 << PAGE_CACHE_SHIFT,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700477 0);
478 did_range_unmap = 1;
479 } else {
480 /*
481 * Just zap this page
482 */
483 unmap_mapping_range(mapping,
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700484 (loff_t)index << PAGE_CACHE_SHIFT,
485 PAGE_CACHE_SIZE, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700486 }
487 }
Nick Piggind00806b2007-07-19 01:46:57 -0700488 BUG_ON(page_mapped(page));
Hisashi Hifumi0dd13342008-04-28 02:12:08 -0700489 ret2 = do_launder_page(mapping, page);
490 if (ret2 == 0) {
491 if (!invalidate_complete_page2(mapping, page))
Hisashi Hifumi6ccfa802008-09-02 14:35:40 -0700492 ret2 = -EBUSY;
Hisashi Hifumi0dd13342008-04-28 02:12:08 -0700493 }
494 if (ret2 < 0)
495 ret = ret2;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700496 unlock_page(page);
497 }
498 pagevec_release(&pvec);
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -0800499 mem_cgroup_uncharge_end();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700500 cond_resched();
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700501 index++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700502 }
Dan Magenheimer31677602011-09-21 11:56:28 -0400503 cleancache_invalidate_inode(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700504 return ret;
505}
506EXPORT_SYMBOL_GPL(invalidate_inode_pages2_range);
507
508/**
509 * invalidate_inode_pages2 - remove all pages from an address_space
Martin Waitz67be2dd2005-05-01 08:59:26 -0700510 * @mapping: the address_space
Linus Torvalds1da177e2005-04-16 15:20:36 -0700511 *
512 * Any pages which are found to be mapped into pagetables are unmapped prior to
513 * invalidation.
514 *
Peng Taoe9de25d2009-10-19 14:48:13 +0800515 * Returns -EBUSY if any pages could not be invalidated.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700516 */
517int invalidate_inode_pages2(struct address_space *mapping)
518{
519 return invalidate_inode_pages2_range(mapping, 0, -1);
520}
521EXPORT_SYMBOL_GPL(invalidate_inode_pages2);
npiggin@suse.de25d9e2d2009-08-21 02:35:05 +1000522
523/**
524 * truncate_pagecache - unmap and remove pagecache that has been truncated
525 * @inode: inode
Hugh Dickins8a549be2011-07-25 17:12:24 -0700526 * @oldsize: old file size
527 * @newsize: new file size
npiggin@suse.de25d9e2d2009-08-21 02:35:05 +1000528 *
529 * inode's new i_size must already be written before truncate_pagecache
530 * is called.
531 *
532 * This function should typically be called before the filesystem
533 * releases resources associated with the freed range (eg. deallocates
534 * blocks). This way, pagecache will always stay logically coherent
535 * with on-disk format, and the filesystem would not have to deal with
536 * situations such as writepage being called for a page that has already
537 * had its underlying blocks deallocated.
538 */
Hugh Dickins8a549be2011-07-25 17:12:24 -0700539void truncate_pagecache(struct inode *inode, loff_t oldsize, loff_t newsize)
npiggin@suse.de25d9e2d2009-08-21 02:35:05 +1000540{
OGAWA Hirofumicedabed2010-01-13 21:14:09 +0900541 struct address_space *mapping = inode->i_mapping;
Hugh Dickins8a549be2011-07-25 17:12:24 -0700542 loff_t holebegin = round_up(newsize, PAGE_SIZE);
npiggin@suse.de25d9e2d2009-08-21 02:35:05 +1000543
OGAWA Hirofumicedabed2010-01-13 21:14:09 +0900544 /*
545 * unmap_mapping_range is called twice, first simply for
546 * efficiency so that truncate_inode_pages does fewer
547 * single-page unmaps. However after this first call, and
548 * before truncate_inode_pages finishes, it is possible for
549 * private pages to be COWed, which remain after
550 * truncate_inode_pages finishes, hence the second
551 * unmap_mapping_range call must be made for correctness.
552 */
Hugh Dickins8a549be2011-07-25 17:12:24 -0700553 unmap_mapping_range(mapping, holebegin, 0, 1);
554 truncate_inode_pages(mapping, newsize);
555 unmap_mapping_range(mapping, holebegin, 0, 1);
npiggin@suse.de25d9e2d2009-08-21 02:35:05 +1000556}
557EXPORT_SYMBOL(truncate_pagecache);
558
559/**
Christoph Hellwig2c27c652010-06-04 11:30:04 +0200560 * truncate_setsize - update inode and pagecache for a new file size
561 * @inode: inode
562 * @newsize: new file size
563 *
Jan Kara382e27d2011-01-20 14:44:26 -0800564 * truncate_setsize updates i_size and performs pagecache truncation (if
565 * necessary) to @newsize. It will be typically be called from the filesystem's
566 * setattr function when ATTR_SIZE is passed in.
Christoph Hellwig2c27c652010-06-04 11:30:04 +0200567 *
Jan Kara382e27d2011-01-20 14:44:26 -0800568 * Must be called with inode_mutex held and before all filesystem specific
569 * block truncation has been performed.
Christoph Hellwig2c27c652010-06-04 11:30:04 +0200570 */
571void truncate_setsize(struct inode *inode, loff_t newsize)
572{
573 loff_t oldsize;
574
575 oldsize = inode->i_size;
576 i_size_write(inode, newsize);
577
578 truncate_pagecache(inode, oldsize, newsize);
579}
580EXPORT_SYMBOL(truncate_setsize);
581
582/**
npiggin@suse.de25d9e2d2009-08-21 02:35:05 +1000583 * vmtruncate - unmap mappings "freed" by truncate() syscall
584 * @inode: inode of the file used
Hugh Dickins8a549be2011-07-25 17:12:24 -0700585 * @newsize: file offset to start truncating
npiggin@suse.de25d9e2d2009-08-21 02:35:05 +1000586 *
Christoph Hellwig2c27c652010-06-04 11:30:04 +0200587 * This function is deprecated and truncate_setsize or truncate_pagecache
588 * should be used instead, together with filesystem specific block truncation.
npiggin@suse.de25d9e2d2009-08-21 02:35:05 +1000589 */
Hugh Dickins8a549be2011-07-25 17:12:24 -0700590int vmtruncate(struct inode *inode, loff_t newsize)
npiggin@suse.de25d9e2d2009-08-21 02:35:05 +1000591{
npiggin@suse.de25d9e2d2009-08-21 02:35:05 +1000592 int error;
593
Hugh Dickins8a549be2011-07-25 17:12:24 -0700594 error = inode_newsize_ok(inode, newsize);
npiggin@suse.de25d9e2d2009-08-21 02:35:05 +1000595 if (error)
596 return error;
npiggin@suse.de7bb46a62010-05-27 01:05:33 +1000597
Hugh Dickins8a549be2011-07-25 17:12:24 -0700598 truncate_setsize(inode, newsize);
npiggin@suse.de25d9e2d2009-08-21 02:35:05 +1000599 if (inode->i_op->truncate)
600 inode->i_op->truncate(inode);
Christoph Hellwig2c27c652010-06-04 11:30:04 +0200601 return 0;
npiggin@suse.de25d9e2d2009-08-21 02:35:05 +1000602}
603EXPORT_SYMBOL(vmtruncate);
Hugh Dickins5b8ba102011-06-27 16:18:01 -0700604
Hugh Dickins8a549be2011-07-25 17:12:24 -0700605int vmtruncate_range(struct inode *inode, loff_t lstart, loff_t lend)
Hugh Dickins5b8ba102011-06-27 16:18:01 -0700606{
607 struct address_space *mapping = inode->i_mapping;
Hugh Dickins8a549be2011-07-25 17:12:24 -0700608 loff_t holebegin = round_up(lstart, PAGE_SIZE);
609 loff_t holelen = 1 + lend - holebegin;
Hugh Dickins5b8ba102011-06-27 16:18:01 -0700610
611 /*
612 * If the underlying filesystem is not going to provide
613 * a way to truncate a range of blocks (punch a hole) -
614 * we should return failure right now.
615 */
616 if (!inode->i_op->truncate_range)
617 return -ENOSYS;
618
619 mutex_lock(&inode->i_mutex);
Christoph Hellwigbd5fe6c2011-06-24 14:29:43 -0400620 inode_dio_wait(inode);
Hugh Dickins8a549be2011-07-25 17:12:24 -0700621 unmap_mapping_range(mapping, holebegin, holelen, 1);
622 inode->i_op->truncate_range(inode, lstart, lend);
Hugh Dickins94c1e622011-06-27 16:18:03 -0700623 /* unmap again to remove racily COWed private pages */
Hugh Dickins8a549be2011-07-25 17:12:24 -0700624 unmap_mapping_range(mapping, holebegin, holelen, 1);
Hugh Dickins5b8ba102011-06-27 16:18:01 -0700625 mutex_unlock(&inode->i_mutex);
626
627 return 0;
628}
Hugh Dickins623e3db2012-03-28 14:42:40 -0700629
630/**
631 * truncate_pagecache_range - unmap and remove pagecache that is hole-punched
632 * @inode: inode
633 * @lstart: offset of beginning of hole
634 * @lend: offset of last byte of hole
635 *
636 * This function should typically be called before the filesystem
637 * releases resources associated with the freed range (eg. deallocates
638 * blocks). This way, pagecache will always stay logically coherent
639 * with on-disk format, and the filesystem would not have to deal with
640 * situations such as writepage being called for a page that has already
641 * had its underlying blocks deallocated.
642 */
643void truncate_pagecache_range(struct inode *inode, loff_t lstart, loff_t lend)
644{
645 struct address_space *mapping = inode->i_mapping;
646 loff_t unmap_start = round_up(lstart, PAGE_SIZE);
647 loff_t unmap_end = round_down(1 + lend, PAGE_SIZE) - 1;
648 /*
649 * This rounding is currently just for example: unmap_mapping_range
650 * expands its hole outwards, whereas we want it to contract the hole
651 * inwards. However, existing callers of truncate_pagecache_range are
652 * doing their own page rounding first; and truncate_inode_pages_range
653 * currently BUGs if lend is not pagealigned-1 (it handles partial
654 * page at start of hole, but not partial page at end of hole). Note
655 * unmap_mapping_range allows holelen 0 for all, and we allow lend -1.
656 */
657
658 /*
659 * Unlike in truncate_pagecache, unmap_mapping_range is called only
660 * once (before truncating pagecache), and without "even_cows" flag:
661 * hole-punching should not remove private COWed pages from the hole.
662 */
663 if ((u64)unmap_end > (u64)unmap_start)
664 unmap_mapping_range(mapping, unmap_start,
665 1 + unmap_end - unmap_start, 0);
666 truncate_inode_pages_range(mapping, lstart, lend);
667}
668EXPORT_SYMBOL(truncate_pagecache_range);