blob: 2d6151fc8f083629a8c4e95ddd157afbb76b8772 [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 Mortone08748ce2006-12-10 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>
Jan Kara6cbdf112014-10-01 21:49:18 -040023#include <linux/rmap.h>
Rik van Rielba470de2008-10-18 20:26:50 -070024#include "internal.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070025
26
David Howellscf9a2ae2006-08-29 19:05:54 +010027/**
Fengguang Wu28bc44d2008-02-03 18:04:10 +020028 * do_invalidatepage - invalidate part or all of a page
David Howellscf9a2ae2006-08-29 19:05:54 +010029 * @page: the page which is affected
30 * @offset: the index of the truncation point
31 *
32 * do_invalidatepage() is called when all or part of the page has become
33 * invalidated by a truncate operation.
34 *
35 * do_invalidatepage() does not have to release all buffers, but it must
36 * ensure that no dirty buffer is left outside @offset and that no I/O
37 * is underway against any of the blocks which are outside the truncation
38 * point. Because the caller is about to free (and possibly reuse) those
39 * blocks on-disk.
40 */
41void do_invalidatepage(struct page *page, unsigned long offset)
42{
43 void (*invalidatepage)(struct page *, unsigned long);
44 invalidatepage = page->mapping->a_ops->invalidatepage;
David Howells93614012006-09-30 20:45:40 +020045#ifdef CONFIG_BLOCK
David Howellscf9a2ae2006-08-29 19:05:54 +010046 if (!invalidatepage)
47 invalidatepage = block_invalidatepage;
David Howells93614012006-09-30 20:45:40 +020048#endif
David Howellscf9a2ae2006-08-29 19:05:54 +010049 if (invalidatepage)
50 (*invalidatepage)(page, offset);
51}
52
Linus Torvalds1da177e2005-04-16 15:20:36 -070053static inline void truncate_partial_page(struct page *page, unsigned partial)
54{
Christoph Lametereebd2aa2008-02-04 22:28:29 -080055 zero_user_segment(page, partial, PAGE_CACHE_SIZE);
Dan Magenheimer31677602011-09-21 11:56:28 -040056 cleancache_invalidate_page(page->mapping, page);
David Howells266cf652009-04-03 16:42:36 +010057 if (page_has_private(page))
Linus Torvalds1da177e2005-04-16 15:20:36 -070058 do_invalidatepage(page, partial);
59}
60
Linus Torvaldsecdfc972007-01-26 12:47:06 -080061/*
62 * This cancels just the dirty bit on the kernel page itself, it
63 * does NOT actually remove dirty bits on any mmap's that may be
64 * around. It also leaves the page tagged dirty, so any sync
65 * activity will still find it on the dirty lists, and in particular,
66 * clear_page_dirty_for_io() will still look at the dirty bits in
67 * the VM.
68 *
69 * Doing this should *normally* only ever be done when a page
70 * is truncated, and is not actually mapped anywhere at all. However,
71 * fs/buffer.c does this when it notices that somebody has cleaned
72 * out all the buffers on a page without actually doing it through
73 * the VM. Can you say "ext3 is horribly ugly"? Tought you could.
74 */
Linus Torvaldsfba25912006-12-20 13:46:42 -080075void cancel_dirty_page(struct page *page, unsigned int account_size)
76{
Linus Torvalds8368e322006-12-23 09:25:04 -080077 if (TestClearPageDirty(page)) {
78 struct address_space *mapping = page->mapping;
79 if (mapping && mapping_cap_account_dirty(mapping)) {
80 dec_zone_page_state(page, NR_FILE_DIRTY);
Peter Zijlstrac9e51e42007-10-16 23:25:47 -070081 dec_bdi_stat(mapping->backing_dev_info,
82 BDI_RECLAIMABLE);
Linus Torvalds8368e322006-12-23 09:25:04 -080083 if (account_size)
84 task_io_account_cancelled_write(account_size);
85 }
Andrew Morton3e67c092006-12-21 11:00:33 -080086 }
Linus Torvaldsfba25912006-12-20 13:46:42 -080087}
Linus Torvalds8368e322006-12-23 09:25:04 -080088EXPORT_SYMBOL(cancel_dirty_page);
Linus Torvaldsfba25912006-12-20 13:46:42 -080089
Linus Torvalds1da177e2005-04-16 15:20:36 -070090/*
91 * If truncate cannot remove the fs-private metadata from the page, the page
Shaohua Li62e1c552008-02-04 22:29:33 -080092 * becomes orphaned. It will be left on the LRU and may even be mapped into
Nick Piggin54cb8822007-07-19 01:46:59 -070093 * user pagetables if we're racing with filemap_fault().
Linus Torvalds1da177e2005-04-16 15:20:36 -070094 *
95 * We need to bale out if page->mapping is no longer equal to the original
96 * mapping. This happens a) when the VM reclaimed the page while we waited on
Andrew Mortonfc0ecff2007-02-10 01:45:39 -080097 * its lock, b) when a concurrent invalidate_mapping_pages got there first and
Linus Torvalds1da177e2005-04-16 15:20:36 -070098 * c) when tmpfs swizzles a page between a tmpfs inode and swapper_space.
99 */
Nick Piggin750b4982009-09-16 11:50:12 +0200100static int
Linus Torvalds1da177e2005-04-16 15:20:36 -0700101truncate_complete_page(struct address_space *mapping, struct page *page)
102{
103 if (page->mapping != mapping)
Nick Piggin750b4982009-09-16 11:50:12 +0200104 return -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700105
David Howells266cf652009-04-03 16:42:36 +0100106 if (page_has_private(page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700107 do_invalidatepage(page, 0);
108
Bjorn Steinbrinka2b34562008-02-04 22:29:28 -0800109 cancel_dirty_page(page, PAGE_CACHE_SIZE);
110
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
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/**
Liu Bo73c1e202012-02-21 10:57:20 +0800186 * truncate_inode_pages_range - 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 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700201 * We pass down the cache-hot hint to the page freeing code. Even if the
202 * mapping is large, it is probably the case that the final pages are the most
203 * recently touched, and freeing happens in ascending file offset order.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700204 */
Hans Reiserd7339072006-01-06 00:10:36 -0800205void truncate_inode_pages_range(struct address_space *mapping,
206 loff_t lstart, loff_t lend)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700207{
208 const pgoff_t start = (lstart + PAGE_CACHE_SIZE-1) >> PAGE_CACHE_SHIFT;
209 const unsigned partial = lstart & (PAGE_CACHE_SIZE - 1);
210 struct pagevec pvec;
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700211 pgoff_t index;
212 pgoff_t end;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700213 int i;
214
Dan Magenheimer31677602011-09-21 11:56:28 -0400215 cleancache_invalidate_inode(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700216 if (mapping->nrpages == 0)
217 return;
218
Hans Reiserd7339072006-01-06 00:10:36 -0800219 BUG_ON((lend & (PAGE_CACHE_SIZE - 1)) != (PAGE_CACHE_SIZE - 1));
220 end = (lend >> PAGE_CACHE_SHIFT);
221
Linus Torvalds1da177e2005-04-16 15:20:36 -0700222 pagevec_init(&pvec, 0);
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700223 index = start;
224 while (index <= end && pagevec_lookup(&pvec, mapping, index,
225 min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1)) {
Hugh Dickinse5598f82011-02-25 14:44:29 -0800226 mem_cgroup_uncharge_start();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700227 for (i = 0; i < pagevec_count(&pvec); i++) {
228 struct page *page = pvec.pages[i];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700229
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700230 /* We rely upon deletion not changing page->index */
231 index = page->index;
232 if (index > end)
Hans Reiserd7339072006-01-06 00:10:36 -0800233 break;
Hans Reiserd7339072006-01-06 00:10:36 -0800234
Nick Piggin529ae9a2008-08-02 12:01:03 +0200235 if (!trylock_page(page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700236 continue;
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700237 WARN_ON(page->index != index);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700238 if (PageWriteback(page)) {
239 unlock_page(page);
240 continue;
241 }
Nick Piggin750b4982009-09-16 11:50:12 +0200242 truncate_inode_page(mapping, page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700243 unlock_page(page);
244 }
245 pagevec_release(&pvec);
Hugh Dickinse5598f82011-02-25 14:44:29 -0800246 mem_cgroup_uncharge_end();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700247 cond_resched();
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700248 index++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700249 }
250
251 if (partial) {
252 struct page *page = find_lock_page(mapping, start - 1);
253 if (page) {
254 wait_on_page_writeback(page);
255 truncate_partial_page(page, partial);
256 unlock_page(page);
257 page_cache_release(page);
258 }
259 }
260
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700261 index = start;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700262 for ( ; ; ) {
263 cond_resched();
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700264 if (!pagevec_lookup(&pvec, mapping, index,
265 min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1)) {
266 if (index == start)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700267 break;
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700268 index = start;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700269 continue;
270 }
Hugh Dickinsd0823572011-07-25 17:12:25 -0700271 if (index == start && pvec.pages[0]->index > end) {
Hans Reiserd7339072006-01-06 00:10:36 -0800272 pagevec_release(&pvec);
273 break;
274 }
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -0800275 mem_cgroup_uncharge_start();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700276 for (i = 0; i < pagevec_count(&pvec); i++) {
277 struct page *page = pvec.pages[i];
278
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700279 /* We rely upon deletion not changing page->index */
280 index = page->index;
281 if (index > end)
Hans Reiserd7339072006-01-06 00:10:36 -0800282 break;
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700283
Linus Torvalds1da177e2005-04-16 15:20:36 -0700284 lock_page(page);
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700285 WARN_ON(page->index != index);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700286 wait_on_page_writeback(page);
Nick Piggin750b4982009-09-16 11:50:12 +0200287 truncate_inode_page(mapping, page);
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();
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700292 index++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700293 }
Dan Magenheimer31677602011-09-21 11:56:28 -0400294 cleancache_invalidate_inode(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700295}
Hans Reiserd7339072006-01-06 00:10:36 -0800296EXPORT_SYMBOL(truncate_inode_pages_range);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700297
Hans Reiserd7339072006-01-06 00:10:36 -0800298/**
299 * truncate_inode_pages - truncate *all* the pages from an offset
300 * @mapping: mapping to truncate
301 * @lstart: offset from which to truncate
302 *
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800303 * Called under (and serialised by) inode->i_mutex.
Jan Kara08142572011-06-27 16:18:10 -0700304 *
305 * Note: When this function returns, there can be a page in the process of
306 * deletion (inside __delete_from_page_cache()) in the specified range. Thus
307 * mapping->nrpages can be non-zero when this function returns even after
308 * truncation of the whole mapping.
Hans Reiserd7339072006-01-06 00:10:36 -0800309 */
310void truncate_inode_pages(struct address_space *mapping, loff_t lstart)
311{
312 truncate_inode_pages_range(mapping, lstart, (loff_t)-1);
313}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700314EXPORT_SYMBOL(truncate_inode_pages);
315
Mike Waychison28697352009-06-16 15:32:59 -0700316/**
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,
Minchan Kim31560182011-03-22 16:32:52 -0700330 pgoff_t start, pgoff_t end)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700331{
332 struct pagevec pvec;
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700333 pgoff_t index = start;
Minchan Kim31560182011-03-22 16:32:52 -0700334 unsigned long ret;
335 unsigned long count = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700336 int i;
337
Hugh Dickins31475dd2011-08-03 16:21:27 -0700338 /*
339 * Note: this function may get called on a shmem/tmpfs mapping:
340 * pagevec_lookup() might then return 0 prematurely (because it
341 * got a gangful of swap entries); but it's hardly worth worrying
342 * about - it can rarely have anything to free from such a mapping
343 * (most pages are dirty), and already skips over any difficulties.
344 */
345
Linus Torvalds1da177e2005-04-16 15:20:36 -0700346 pagevec_init(&pvec, 0);
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700347 while (index <= end && pagevec_lookup(&pvec, mapping, index,
348 min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1)) {
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -0800349 mem_cgroup_uncharge_start();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700350 for (i = 0; i < pagevec_count(&pvec); i++) {
351 struct page *page = pvec.pages[i];
352
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700353 /* We rely upon deletion not changing page->index */
NeilBrowne0f23602006-06-23 02:05:48 -0700354 index = page->index;
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700355 if (index > end)
356 break;
NeilBrowne0f23602006-06-23 02:05:48 -0700357
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700358 if (!trylock_page(page))
359 continue;
360 WARN_ON(page->index != index);
Minchan Kim31560182011-03-22 16:32:52 -0700361 ret = invalidate_inode_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700362 unlock_page(page);
Minchan Kim31560182011-03-22 16:32:52 -0700363 /*
364 * Invalidation is a hint that the page is no longer
365 * of interest and try to speed up its reclaim.
366 */
367 if (!ret)
368 deactivate_page(page);
369 count += ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700370 }
371 pagevec_release(&pvec);
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -0800372 mem_cgroup_uncharge_end();
Mike Waychison28697352009-06-16 15:32:59 -0700373 cond_resched();
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700374 index++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700375 }
Minchan Kim31560182011-03-22 16:32:52 -0700376 return count;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700377}
Anton Altaparmakov54bc4852007-02-10 01:45:38 -0800378EXPORT_SYMBOL(invalidate_mapping_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700379
Andrew Mortonbd4c8ce2006-09-30 23:29:29 -0700380/*
381 * This is like invalidate_complete_page(), except it ignores the page's
382 * refcount. We do this because invalidate_inode_pages2() needs stronger
383 * invalidation guarantees, and cannot afford to leave pages behind because
Anderson Briglia2706a1b2007-07-15 23:38:09 -0700384 * shrink_page_list() has a temp ref on them, or because they're transiently
385 * sitting in the lru_cache_add() pagevecs.
Andrew Mortonbd4c8ce2006-09-30 23:29:29 -0700386 */
387static int
388invalidate_complete_page2(struct address_space *mapping, struct page *page)
389{
390 if (page->mapping != mapping)
391 return 0;
392
David Howells266cf652009-04-03 16:42:36 +0100393 if (page_has_private(page) && !try_to_release_page(page, GFP_KERNEL))
Andrew Mortonbd4c8ce2006-09-30 23:29:29 -0700394 return 0;
395
Nick Piggin19fd6232008-07-25 19:45:32 -0700396 spin_lock_irq(&mapping->tree_lock);
Andrew Mortonbd4c8ce2006-09-30 23:29:29 -0700397 if (PageDirty(page))
398 goto failed;
399
David Howells266cf652009-04-03 16:42:36 +0100400 BUG_ON(page_has_private(page));
Minchan Kime64a7822011-03-22 16:32:44 -0700401 __delete_from_page_cache(page);
Nick Piggin19fd6232008-07-25 19:45:32 -0700402 spin_unlock_irq(&mapping->tree_lock);
Daisuke Nishimurae767e052009-05-28 14:34:28 -0700403 mem_cgroup_uncharge_cache_page(page);
Linus Torvalds6072d132010-12-01 13:35:19 -0500404
405 if (mapping->a_ops->freepage)
406 mapping->a_ops->freepage(page);
407
Andrew Mortonbd4c8ce2006-09-30 23:29:29 -0700408 page_cache_release(page); /* pagecache ref */
409 return 1;
410failed:
Nick Piggin19fd6232008-07-25 19:45:32 -0700411 spin_unlock_irq(&mapping->tree_lock);
Andrew Mortonbd4c8ce2006-09-30 23:29:29 -0700412 return 0;
413}
414
Trond Myklebuste3db7692007-01-10 23:15:39 -0800415static int do_launder_page(struct address_space *mapping, struct page *page)
416{
417 if (!PageDirty(page))
418 return 0;
419 if (page->mapping != mapping || mapping->a_ops->launder_page == NULL)
420 return 0;
421 return mapping->a_ops->launder_page(page);
422}
423
Linus Torvalds1da177e2005-04-16 15:20:36 -0700424/**
425 * invalidate_inode_pages2_range - remove range of pages from an address_space
Martin Waitz67be2dd2005-05-01 08:59:26 -0700426 * @mapping: the address_space
Linus Torvalds1da177e2005-04-16 15:20:36 -0700427 * @start: the page offset 'from' which to invalidate
428 * @end: the page offset 'to' which to invalidate (inclusive)
429 *
430 * Any pages which are found to be mapped into pagetables are unmapped prior to
431 * invalidation.
432 *
Hisashi Hifumi6ccfa802008-09-02 14:35:40 -0700433 * Returns -EBUSY if any pages could not be invalidated.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700434 */
435int invalidate_inode_pages2_range(struct address_space *mapping,
436 pgoff_t start, pgoff_t end)
437{
438 struct pagevec pvec;
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700439 pgoff_t index;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700440 int i;
441 int ret = 0;
Hisashi Hifumi0dd13342008-04-28 02:12:08 -0700442 int ret2 = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700443 int did_range_unmap = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700444
Dan Magenheimer31677602011-09-21 11:56:28 -0400445 cleancache_invalidate_inode(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700446 pagevec_init(&pvec, 0);
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700447 index = start;
448 while (index <= end && pagevec_lookup(&pvec, mapping, index,
449 min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1)) {
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -0800450 mem_cgroup_uncharge_start();
Trond Myklebust7b965e02007-02-28 20:13:55 -0800451 for (i = 0; i < pagevec_count(&pvec); i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700452 struct page *page = pvec.pages[i];
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700453
454 /* We rely upon deletion not changing page->index */
455 index = page->index;
456 if (index > end)
457 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700458
459 lock_page(page);
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700460 WARN_ON(page->index != index);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700461 if (page->mapping != mapping) {
462 unlock_page(page);
463 continue;
464 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700465 wait_on_page_writeback(page);
Nick Piggind00806b2007-07-19 01:46:57 -0700466 if (page_mapped(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700467 if (!did_range_unmap) {
468 /*
469 * Zap the rest of the file in one hit.
470 */
471 unmap_mapping_range(mapping,
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700472 (loff_t)index << PAGE_CACHE_SHIFT,
473 (loff_t)(1 + end - index)
474 << PAGE_CACHE_SHIFT,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700475 0);
476 did_range_unmap = 1;
477 } else {
478 /*
479 * Just zap this page
480 */
481 unmap_mapping_range(mapping,
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700482 (loff_t)index << PAGE_CACHE_SHIFT,
483 PAGE_CACHE_SIZE, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700484 }
485 }
Nick Piggind00806b2007-07-19 01:46:57 -0700486 BUG_ON(page_mapped(page));
Hisashi Hifumi0dd13342008-04-28 02:12:08 -0700487 ret2 = do_launder_page(mapping, page);
488 if (ret2 == 0) {
489 if (!invalidate_complete_page2(mapping, page))
Hisashi Hifumi6ccfa802008-09-02 14:35:40 -0700490 ret2 = -EBUSY;
Hisashi Hifumi0dd13342008-04-28 02:12:08 -0700491 }
492 if (ret2 < 0)
493 ret = ret2;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700494 unlock_page(page);
495 }
496 pagevec_release(&pvec);
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -0800497 mem_cgroup_uncharge_end();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700498 cond_resched();
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700499 index++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700500 }
Dan Magenheimer31677602011-09-21 11:56:28 -0400501 cleancache_invalidate_inode(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700502 return ret;
503}
504EXPORT_SYMBOL_GPL(invalidate_inode_pages2_range);
505
506/**
507 * invalidate_inode_pages2 - remove all pages from an address_space
Martin Waitz67be2dd2005-05-01 08:59:26 -0700508 * @mapping: the address_space
Linus Torvalds1da177e2005-04-16 15:20:36 -0700509 *
510 * Any pages which are found to be mapped into pagetables are unmapped prior to
511 * invalidation.
512 *
Peng Taoe9de25d2009-10-19 14:48:13 +0800513 * Returns -EBUSY if any pages could not be invalidated.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700514 */
515int invalidate_inode_pages2(struct address_space *mapping)
516{
517 return invalidate_inode_pages2_range(mapping, 0, -1);
518}
519EXPORT_SYMBOL_GPL(invalidate_inode_pages2);
npiggin@suse.de25d9e2d2009-08-21 02:35:05 +1000520
521/**
522 * truncate_pagecache - unmap and remove pagecache that has been truncated
523 * @inode: inode
Hugh Dickins8a549be2011-07-25 17:12:24 -0700524 * @oldsize: old file size
525 * @newsize: new file size
npiggin@suse.de25d9e2d2009-08-21 02:35:05 +1000526 *
527 * inode's new i_size must already be written before truncate_pagecache
528 * is called.
529 *
530 * This function should typically be called before the filesystem
531 * releases resources associated with the freed range (eg. deallocates
532 * blocks). This way, pagecache will always stay logically coherent
533 * with on-disk format, and the filesystem would not have to deal with
534 * situations such as writepage being called for a page that has already
535 * had its underlying blocks deallocated.
536 */
Hugh Dickins8a549be2011-07-25 17:12:24 -0700537void truncate_pagecache(struct inode *inode, loff_t oldsize, loff_t newsize)
npiggin@suse.de25d9e2d2009-08-21 02:35:05 +1000538{
OGAWA Hirofumicedabed2010-01-13 21:14:09 +0900539 struct address_space *mapping = inode->i_mapping;
Hugh Dickins8a549be2011-07-25 17:12:24 -0700540 loff_t holebegin = round_up(newsize, PAGE_SIZE);
npiggin@suse.de25d9e2d2009-08-21 02:35:05 +1000541
OGAWA Hirofumicedabed2010-01-13 21:14:09 +0900542 /*
543 * unmap_mapping_range is called twice, first simply for
544 * efficiency so that truncate_inode_pages does fewer
545 * single-page unmaps. However after this first call, and
546 * before truncate_inode_pages finishes, it is possible for
547 * private pages to be COWed, which remain after
548 * truncate_inode_pages finishes, hence the second
549 * unmap_mapping_range call must be made for correctness.
550 */
Hugh Dickins8a549be2011-07-25 17:12:24 -0700551 unmap_mapping_range(mapping, holebegin, 0, 1);
552 truncate_inode_pages(mapping, newsize);
553 unmap_mapping_range(mapping, holebegin, 0, 1);
npiggin@suse.de25d9e2d2009-08-21 02:35:05 +1000554}
555EXPORT_SYMBOL(truncate_pagecache);
556
557/**
Christoph Hellwig2c27c652010-06-04 11:30:04 +0200558 * truncate_setsize - update inode and pagecache for a new file size
559 * @inode: inode
560 * @newsize: new file size
561 *
Jan Kara382e27d2011-01-20 14:44:26 -0800562 * truncate_setsize updates i_size and performs pagecache truncation (if
563 * necessary) to @newsize. It will be typically be called from the filesystem's
564 * setattr function when ATTR_SIZE is passed in.
Christoph Hellwig2c27c652010-06-04 11:30:04 +0200565 *
Jan Kara382e27d2011-01-20 14:44:26 -0800566 * Must be called with inode_mutex held and before all filesystem specific
567 * block truncation has been performed.
Christoph Hellwig2c27c652010-06-04 11:30:04 +0200568 */
569void truncate_setsize(struct inode *inode, loff_t newsize)
570{
Jan Kara6cbdf112014-10-01 21:49:18 -0400571 loff_t oldsize = inode->i_size;
Christoph Hellwig2c27c652010-06-04 11:30:04 +0200572
Christoph Hellwig2c27c652010-06-04 11:30:04 +0200573 i_size_write(inode, newsize);
Jan Kara6cbdf112014-10-01 21:49:18 -0400574 if (newsize > oldsize)
575 pagecache_isize_extended(inode, oldsize, newsize);
Christoph Hellwig2c27c652010-06-04 11:30:04 +0200576 truncate_pagecache(inode, oldsize, newsize);
577}
578EXPORT_SYMBOL(truncate_setsize);
579
580/**
Jan Kara6cbdf112014-10-01 21:49:18 -0400581 * pagecache_isize_extended - update pagecache after extension of i_size
582 * @inode: inode for which i_size was extended
583 * @from: original inode size
584 * @to: new inode size
585 *
586 * Handle extension of inode size either caused by extending truncate or by
587 * write starting after current i_size. We mark the page straddling current
588 * i_size RO so that page_mkwrite() is called on the nearest write access to
589 * the page. This way filesystem can be sure that page_mkwrite() is called on
590 * the page before user writes to the page via mmap after the i_size has been
591 * changed.
592 *
593 * The function must be called after i_size is updated so that page fault
594 * coming after we unlock the page will already see the new i_size.
595 * The function must be called while we still hold i_mutex - this not only
596 * makes sure i_size is stable but also that userspace cannot observe new
597 * i_size value before we are prepared to store mmap writes at new inode size.
598 */
599void pagecache_isize_extended(struct inode *inode, loff_t from, loff_t to)
600{
601 int bsize = 1 << inode->i_blkbits;
602 loff_t rounded_from;
603 struct page *page;
604 pgoff_t index;
605
Jan Kara6cbdf112014-10-01 21:49:18 -0400606 WARN_ON(to > inode->i_size);
607
608 if (from >= to || bsize == PAGE_CACHE_SIZE)
609 return;
610 /* Page straddling @from will not have any hole block created? */
611 rounded_from = round_up(from, bsize);
612 if (to <= rounded_from || !(rounded_from & (PAGE_CACHE_SIZE - 1)))
613 return;
614
615 index = from >> PAGE_CACHE_SHIFT;
616 page = find_lock_page(inode->i_mapping, index);
617 /* Page not cached? Nothing to do */
618 if (!page)
619 return;
620 /*
621 * See clear_page_dirty_for_io() for details why set_page_dirty()
622 * is needed.
623 */
624 if (page_mkclean(page))
625 set_page_dirty(page);
626 unlock_page(page);
627 page_cache_release(page);
628}
629EXPORT_SYMBOL(pagecache_isize_extended);
630
631/**
Hugh Dickins623e3db2012-03-28 14:42:40 -0700632 * truncate_pagecache_range - unmap and remove pagecache that is hole-punched
633 * @inode: inode
634 * @lstart: offset of beginning of hole
635 * @lend: offset of last byte of hole
636 *
637 * This function should typically be called before the filesystem
638 * releases resources associated with the freed range (eg. deallocates
639 * blocks). This way, pagecache will always stay logically coherent
640 * with on-disk format, and the filesystem would not have to deal with
641 * situations such as writepage being called for a page that has already
642 * had its underlying blocks deallocated.
643 */
644void truncate_pagecache_range(struct inode *inode, loff_t lstart, loff_t lend)
645{
646 struct address_space *mapping = inode->i_mapping;
647 loff_t unmap_start = round_up(lstart, PAGE_SIZE);
648 loff_t unmap_end = round_down(1 + lend, PAGE_SIZE) - 1;
649 /*
650 * This rounding is currently just for example: unmap_mapping_range
651 * expands its hole outwards, whereas we want it to contract the hole
652 * inwards. However, existing callers of truncate_pagecache_range are
653 * doing their own page rounding first; and truncate_inode_pages_range
654 * currently BUGs if lend is not pagealigned-1 (it handles partial
655 * page at start of hole, but not partial page at end of hole). Note
656 * unmap_mapping_range allows holelen 0 for all, and we allow lend -1.
657 */
658
659 /*
660 * Unlike in truncate_pagecache, unmap_mapping_range is called only
661 * once (before truncating pagecache), and without "even_cows" flag:
662 * hole-punching should not remove private COWed pages from the hole.
663 */
664 if ((u64)unmap_end > (u64)unmap_start)
665 unmap_mapping_range(mapping, unmap_start,
666 1 + unmap_end - unmap_start, 0);
667 truncate_inode_pages_range(mapping, lstart, lend);
668}
669EXPORT_SYMBOL(truncate_pagecache_range);