blob: c75b736e54b793f338fce25f35afa60eb0faf5ef [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>
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
Linus Torvalds1da177e2005-04-16 15:20:36 -0700110 ClearPageMappedToDisk(page);
Minchan Kim5adc7b52011-03-22 16:32:41 -0700111 delete_from_page_cache(page);
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
Nick Piggin0fd0e6b2006-09-27 01:50:02 -0700134 ret = remove_mapping(mapping, page);
Nick Piggin0fd0e6b2006-09-27 01:50:02 -0700135
136 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700137}
138
Nick Piggin750b4982009-09-16 11:50:12 +0200139int truncate_inode_page(struct address_space *mapping, struct page *page)
140{
141 if (page_mapped(page)) {
142 unmap_mapping_range(mapping,
143 (loff_t)page->index << PAGE_CACHE_SHIFT,
144 PAGE_CACHE_SIZE, 0);
145 }
146 return truncate_complete_page(mapping, page);
147}
148
Wu Fengguang83f78662009-09-16 11:50:13 +0200149/*
Andi Kleen25718732009-09-16 11:50:13 +0200150 * Used to get rid of pages on hardware memory corruption.
151 */
152int generic_error_remove_page(struct address_space *mapping, struct page *page)
153{
154 if (!mapping)
155 return -EINVAL;
156 /*
157 * Only punch for normal data pages for now.
158 * Handling other types like directories would need more auditing.
159 */
160 if (!S_ISREG(mapping->host->i_mode))
161 return -EIO;
162 return truncate_inode_page(mapping, page);
163}
164EXPORT_SYMBOL(generic_error_remove_page);
165
166/*
Wu Fengguang83f78662009-09-16 11:50:13 +0200167 * Safely invalidate one page from its pagecache mapping.
168 * It only drops clean, unused pages. The page must be locked.
169 *
170 * Returns 1 if the page is successfully invalidated, otherwise 0.
171 */
172int invalidate_inode_page(struct page *page)
173{
174 struct address_space *mapping = page_mapping(page);
175 if (!mapping)
176 return 0;
177 if (PageDirty(page) || PageWriteback(page))
178 return 0;
179 if (page_mapped(page))
180 return 0;
181 return invalidate_complete_page(mapping, page);
182}
183
Linus Torvalds1da177e2005-04-16 15:20:36 -0700184/**
Liu Bo73c1e202012-02-21 10:57:20 +0800185 * truncate_inode_pages_range - truncate range of pages specified by start & end byte offsets
Linus Torvalds1da177e2005-04-16 15:20:36 -0700186 * @mapping: mapping to truncate
187 * @lstart: offset from which to truncate
Hans Reiserd7339072006-01-06 00:10:36 -0800188 * @lend: offset to which to truncate
Linus Torvalds1da177e2005-04-16 15:20:36 -0700189 *
Hans Reiserd7339072006-01-06 00:10:36 -0800190 * Truncate the page cache, removing the pages that are between
191 * specified offsets (and zeroing out partial page
192 * (if lstart is not page aligned)).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700193 *
194 * Truncate takes two passes - the first pass is nonblocking. It will not
195 * block on page locks and it will not block on writeback. The second pass
196 * will wait. This is to prevent as much IO as possible in the affected region.
197 * The first pass will remove most pages, so the search cost of the second pass
198 * is low.
199 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700200 * We pass down the cache-hot hint to the page freeing code. Even if the
201 * mapping is large, it is probably the case that the final pages are the most
202 * recently touched, and freeing happens in ascending file offset order.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700203 */
Hans Reiserd7339072006-01-06 00:10:36 -0800204void truncate_inode_pages_range(struct address_space *mapping,
205 loff_t lstart, loff_t lend)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700206{
207 const pgoff_t start = (lstart + PAGE_CACHE_SIZE-1) >> PAGE_CACHE_SHIFT;
208 const unsigned partial = lstart & (PAGE_CACHE_SIZE - 1);
209 struct pagevec pvec;
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700210 pgoff_t index;
211 pgoff_t end;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700212 int i;
213
Dan Magenheimer31677602011-09-21 11:56:28 -0400214 cleancache_invalidate_inode(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700215 if (mapping->nrpages == 0)
216 return;
217
Hans Reiserd7339072006-01-06 00:10:36 -0800218 BUG_ON((lend & (PAGE_CACHE_SIZE - 1)) != (PAGE_CACHE_SIZE - 1));
219 end = (lend >> PAGE_CACHE_SHIFT);
220
Linus Torvalds1da177e2005-04-16 15:20:36 -0700221 pagevec_init(&pvec, 0);
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700222 index = start;
223 while (index <= end && pagevec_lookup(&pvec, mapping, index,
224 min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1)) {
Hugh Dickinse5598f82011-02-25 14:44:29 -0800225 mem_cgroup_uncharge_start();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700226 for (i = 0; i < pagevec_count(&pvec); i++) {
227 struct page *page = pvec.pages[i];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700228
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700229 /* We rely upon deletion not changing page->index */
230 index = page->index;
231 if (index > end)
Hans Reiserd7339072006-01-06 00:10:36 -0800232 break;
Hans Reiserd7339072006-01-06 00:10:36 -0800233
Nick Piggin529ae9a2008-08-02 12:01:03 +0200234 if (!trylock_page(page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700235 continue;
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700236 WARN_ON(page->index != index);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700237 if (PageWriteback(page)) {
238 unlock_page(page);
239 continue;
240 }
Nick Piggin750b4982009-09-16 11:50:12 +0200241 truncate_inode_page(mapping, page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700242 unlock_page(page);
243 }
244 pagevec_release(&pvec);
Hugh Dickinse5598f82011-02-25 14:44:29 -0800245 mem_cgroup_uncharge_end();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700246 cond_resched();
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700247 index++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700248 }
249
250 if (partial) {
251 struct page *page = find_lock_page(mapping, start - 1);
252 if (page) {
253 wait_on_page_writeback(page);
254 truncate_partial_page(page, partial);
255 unlock_page(page);
256 page_cache_release(page);
257 }
258 }
259
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700260 index = start;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700261 for ( ; ; ) {
262 cond_resched();
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700263 if (!pagevec_lookup(&pvec, mapping, index,
264 min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1)) {
265 if (index == start)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700266 break;
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700267 index = start;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700268 continue;
269 }
Hugh Dickinsd0823572011-07-25 17:12:25 -0700270 if (index == start && pvec.pages[0]->index > end) {
Hans Reiserd7339072006-01-06 00:10:36 -0800271 pagevec_release(&pvec);
272 break;
273 }
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -0800274 mem_cgroup_uncharge_start();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700275 for (i = 0; i < pagevec_count(&pvec); i++) {
276 struct page *page = pvec.pages[i];
277
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700278 /* We rely upon deletion not changing page->index */
279 index = page->index;
280 if (index > end)
Hans Reiserd7339072006-01-06 00:10:36 -0800281 break;
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700282
Linus Torvalds1da177e2005-04-16 15:20:36 -0700283 lock_page(page);
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700284 WARN_ON(page->index != index);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700285 wait_on_page_writeback(page);
Nick Piggin750b4982009-09-16 11:50:12 +0200286 truncate_inode_page(mapping, page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700287 unlock_page(page);
288 }
289 pagevec_release(&pvec);
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -0800290 mem_cgroup_uncharge_end();
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700291 index++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700292 }
Dan Magenheimer31677602011-09-21 11:56:28 -0400293 cleancache_invalidate_inode(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700294}
Hans Reiserd7339072006-01-06 00:10:36 -0800295EXPORT_SYMBOL(truncate_inode_pages_range);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700296
Hans Reiserd7339072006-01-06 00:10:36 -0800297/**
298 * truncate_inode_pages - truncate *all* the pages from an offset
299 * @mapping: mapping to truncate
300 * @lstart: offset from which to truncate
301 *
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800302 * Called under (and serialised by) inode->i_mutex.
Jan Kara08142572011-06-27 16:18:10 -0700303 *
304 * Note: When this function returns, there can be a page in the process of
305 * deletion (inside __delete_from_page_cache()) in the specified range. Thus
306 * mapping->nrpages can be non-zero when this function returns even after
307 * truncation of the whole mapping.
Hans Reiserd7339072006-01-06 00:10:36 -0800308 */
309void truncate_inode_pages(struct address_space *mapping, loff_t lstart)
310{
311 truncate_inode_pages_range(mapping, lstart, (loff_t)-1);
312}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700313EXPORT_SYMBOL(truncate_inode_pages);
314
Mike Waychison28697352009-06-16 15:32:59 -0700315/**
316 * invalidate_mapping_pages - Invalidate all the unlocked pages of one inode
317 * @mapping: the address_space which holds the pages to invalidate
318 * @start: the offset 'from' which to invalidate
319 * @end: the offset 'to' which to invalidate (inclusive)
320 *
321 * This function only removes the unlocked pages, if you want to
322 * remove all the pages of one inode, you must call truncate_inode_pages.
323 *
324 * invalidate_mapping_pages() will not block on IO activity. It will not
325 * invalidate pages which are dirty, locked, under writeback or mapped into
326 * pagetables.
327 */
328unsigned long invalidate_mapping_pages(struct address_space *mapping,
Minchan Kim31560182011-03-22 16:32:52 -0700329 pgoff_t start, pgoff_t end)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700330{
331 struct pagevec pvec;
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700332 pgoff_t index = start;
Minchan Kim31560182011-03-22 16:32:52 -0700333 unsigned long ret;
334 unsigned long count = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700335 int i;
336
Hugh Dickins31475dd2011-08-03 16:21:27 -0700337 /*
338 * Note: this function may get called on a shmem/tmpfs mapping:
339 * pagevec_lookup() might then return 0 prematurely (because it
340 * got a gangful of swap entries); but it's hardly worth worrying
341 * about - it can rarely have anything to free from such a mapping
342 * (most pages are dirty), and already skips over any difficulties.
343 */
344
Linus Torvalds1da177e2005-04-16 15:20:36 -0700345 pagevec_init(&pvec, 0);
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700346 while (index <= end && pagevec_lookup(&pvec, mapping, index,
347 min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1)) {
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -0800348 mem_cgroup_uncharge_start();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700349 for (i = 0; i < pagevec_count(&pvec); i++) {
350 struct page *page = pvec.pages[i];
351
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700352 /* We rely upon deletion not changing page->index */
NeilBrowne0f23602006-06-23 02:05:48 -0700353 index = page->index;
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700354 if (index > end)
355 break;
NeilBrowne0f23602006-06-23 02:05:48 -0700356
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700357 if (!trylock_page(page))
358 continue;
359 WARN_ON(page->index != index);
Minchan Kim31560182011-03-22 16:32:52 -0700360 ret = invalidate_inode_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700361 unlock_page(page);
Minchan Kim31560182011-03-22 16:32:52 -0700362 /*
363 * Invalidation is a hint that the page is no longer
364 * of interest and try to speed up its reclaim.
365 */
366 if (!ret)
367 deactivate_page(page);
368 count += ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700369 }
370 pagevec_release(&pvec);
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -0800371 mem_cgroup_uncharge_end();
Mike Waychison28697352009-06-16 15:32:59 -0700372 cond_resched();
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700373 index++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700374 }
Minchan Kim31560182011-03-22 16:32:52 -0700375 return count;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700376}
Anton Altaparmakov54bc4852007-02-10 01:45:38 -0800377EXPORT_SYMBOL(invalidate_mapping_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700378
Andrew Mortonbd4c8ce2006-09-30 23:29:29 -0700379/*
380 * This is like invalidate_complete_page(), except it ignores the page's
381 * refcount. We do this because invalidate_inode_pages2() needs stronger
382 * invalidation guarantees, and cannot afford to leave pages behind because
Anderson Briglia2706a1b2007-07-15 23:38:09 -0700383 * shrink_page_list() has a temp ref on them, or because they're transiently
384 * sitting in the lru_cache_add() pagevecs.
Andrew Mortonbd4c8ce2006-09-30 23:29:29 -0700385 */
386static int
387invalidate_complete_page2(struct address_space *mapping, struct page *page)
388{
389 if (page->mapping != mapping)
390 return 0;
391
David Howells266cf652009-04-03 16:42:36 +0100392 if (page_has_private(page) && !try_to_release_page(page, GFP_KERNEL))
Andrew Mortonbd4c8ce2006-09-30 23:29:29 -0700393 return 0;
394
Nick Piggin19fd6232008-07-25 19:45:32 -0700395 spin_lock_irq(&mapping->tree_lock);
Andrew Mortonbd4c8ce2006-09-30 23:29:29 -0700396 if (PageDirty(page))
397 goto failed;
398
David Howells266cf652009-04-03 16:42:36 +0100399 BUG_ON(page_has_private(page));
Minchan Kime64a7822011-03-22 16:32:44 -0700400 __delete_from_page_cache(page);
Nick Piggin19fd6232008-07-25 19:45:32 -0700401 spin_unlock_irq(&mapping->tree_lock);
Daisuke Nishimurae767e052009-05-28 14:34:28 -0700402 mem_cgroup_uncharge_cache_page(page);
Linus Torvalds6072d132010-12-01 13:35:19 -0500403
404 if (mapping->a_ops->freepage)
405 mapping->a_ops->freepage(page);
406
Andrew Mortonbd4c8ce2006-09-30 23:29:29 -0700407 page_cache_release(page); /* pagecache ref */
408 return 1;
409failed:
Nick Piggin19fd6232008-07-25 19:45:32 -0700410 spin_unlock_irq(&mapping->tree_lock);
Andrew Mortonbd4c8ce2006-09-30 23:29:29 -0700411 return 0;
412}
413
Trond Myklebuste3db7692007-01-10 23:15:39 -0800414static int do_launder_page(struct address_space *mapping, struct page *page)
415{
416 if (!PageDirty(page))
417 return 0;
418 if (page->mapping != mapping || mapping->a_ops->launder_page == NULL)
419 return 0;
420 return mapping->a_ops->launder_page(page);
421}
422
Linus Torvalds1da177e2005-04-16 15:20:36 -0700423/**
424 * invalidate_inode_pages2_range - remove range of pages from an address_space
Martin Waitz67be2dd2005-05-01 08:59:26 -0700425 * @mapping: the address_space
Linus Torvalds1da177e2005-04-16 15:20:36 -0700426 * @start: the page offset 'from' which to invalidate
427 * @end: the page offset 'to' which to invalidate (inclusive)
428 *
429 * Any pages which are found to be mapped into pagetables are unmapped prior to
430 * invalidation.
431 *
Hisashi Hifumi6ccfa802008-09-02 14:35:40 -0700432 * Returns -EBUSY if any pages could not be invalidated.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700433 */
434int invalidate_inode_pages2_range(struct address_space *mapping,
435 pgoff_t start, pgoff_t end)
436{
437 struct pagevec pvec;
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700438 pgoff_t index;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700439 int i;
440 int ret = 0;
Hisashi Hifumi0dd13342008-04-28 02:12:08 -0700441 int ret2 = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700442 int did_range_unmap = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700443
Dan Magenheimer31677602011-09-21 11:56:28 -0400444 cleancache_invalidate_inode(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700445 pagevec_init(&pvec, 0);
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700446 index = start;
447 while (index <= end && pagevec_lookup(&pvec, mapping, index,
448 min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1)) {
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -0800449 mem_cgroup_uncharge_start();
Trond Myklebust7b965e02007-02-28 20:13:55 -0800450 for (i = 0; i < pagevec_count(&pvec); i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700451 struct page *page = pvec.pages[i];
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700452
453 /* We rely upon deletion not changing page->index */
454 index = page->index;
455 if (index > end)
456 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700457
458 lock_page(page);
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700459 WARN_ON(page->index != index);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700460 if (page->mapping != mapping) {
461 unlock_page(page);
462 continue;
463 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700464 wait_on_page_writeback(page);
Nick Piggind00806b2007-07-19 01:46:57 -0700465 if (page_mapped(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700466 if (!did_range_unmap) {
467 /*
468 * Zap the rest of the file in one hit.
469 */
470 unmap_mapping_range(mapping,
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700471 (loff_t)index << PAGE_CACHE_SHIFT,
472 (loff_t)(1 + end - index)
473 << PAGE_CACHE_SHIFT,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700474 0);
475 did_range_unmap = 1;
476 } else {
477 /*
478 * Just zap this page
479 */
480 unmap_mapping_range(mapping,
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700481 (loff_t)index << PAGE_CACHE_SHIFT,
482 PAGE_CACHE_SIZE, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700483 }
484 }
Nick Piggind00806b2007-07-19 01:46:57 -0700485 BUG_ON(page_mapped(page));
Hisashi Hifumi0dd13342008-04-28 02:12:08 -0700486 ret2 = do_launder_page(mapping, page);
487 if (ret2 == 0) {
488 if (!invalidate_complete_page2(mapping, page))
Hisashi Hifumi6ccfa802008-09-02 14:35:40 -0700489 ret2 = -EBUSY;
Hisashi Hifumi0dd13342008-04-28 02:12:08 -0700490 }
491 if (ret2 < 0)
492 ret = ret2;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700493 unlock_page(page);
494 }
495 pagevec_release(&pvec);
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -0800496 mem_cgroup_uncharge_end();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700497 cond_resched();
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700498 index++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700499 }
Dan Magenheimer31677602011-09-21 11:56:28 -0400500 cleancache_invalidate_inode(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700501 return ret;
502}
503EXPORT_SYMBOL_GPL(invalidate_inode_pages2_range);
504
505/**
506 * invalidate_inode_pages2 - remove all pages from an address_space
Martin Waitz67be2dd2005-05-01 08:59:26 -0700507 * @mapping: the address_space
Linus Torvalds1da177e2005-04-16 15:20:36 -0700508 *
509 * Any pages which are found to be mapped into pagetables are unmapped prior to
510 * invalidation.
511 *
Peng Taoe9de25d2009-10-19 14:48:13 +0800512 * Returns -EBUSY if any pages could not be invalidated.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700513 */
514int invalidate_inode_pages2(struct address_space *mapping)
515{
516 return invalidate_inode_pages2_range(mapping, 0, -1);
517}
518EXPORT_SYMBOL_GPL(invalidate_inode_pages2);
npiggin@suse.de25d9e2d2009-08-21 02:35:05 +1000519
520/**
521 * truncate_pagecache - unmap and remove pagecache that has been truncated
522 * @inode: inode
Hugh Dickins8a549be2011-07-25 17:12:24 -0700523 * @oldsize: old file size
524 * @newsize: new file size
npiggin@suse.de25d9e2d2009-08-21 02:35:05 +1000525 *
526 * inode's new i_size must already be written before truncate_pagecache
527 * is called.
528 *
529 * This function should typically be called before the filesystem
530 * releases resources associated with the freed range (eg. deallocates
531 * blocks). This way, pagecache will always stay logically coherent
532 * with on-disk format, and the filesystem would not have to deal with
533 * situations such as writepage being called for a page that has already
534 * had its underlying blocks deallocated.
535 */
Hugh Dickins8a549be2011-07-25 17:12:24 -0700536void truncate_pagecache(struct inode *inode, loff_t oldsize, loff_t newsize)
npiggin@suse.de25d9e2d2009-08-21 02:35:05 +1000537{
OGAWA Hirofumicedabed2010-01-13 21:14:09 +0900538 struct address_space *mapping = inode->i_mapping;
Hugh Dickins8a549be2011-07-25 17:12:24 -0700539 loff_t holebegin = round_up(newsize, PAGE_SIZE);
npiggin@suse.de25d9e2d2009-08-21 02:35:05 +1000540
OGAWA Hirofumicedabed2010-01-13 21:14:09 +0900541 /*
542 * unmap_mapping_range is called twice, first simply for
543 * efficiency so that truncate_inode_pages does fewer
544 * single-page unmaps. However after this first call, and
545 * before truncate_inode_pages finishes, it is possible for
546 * private pages to be COWed, which remain after
547 * truncate_inode_pages finishes, hence the second
548 * unmap_mapping_range call must be made for correctness.
549 */
Hugh Dickins8a549be2011-07-25 17:12:24 -0700550 unmap_mapping_range(mapping, holebegin, 0, 1);
551 truncate_inode_pages(mapping, newsize);
552 unmap_mapping_range(mapping, holebegin, 0, 1);
npiggin@suse.de25d9e2d2009-08-21 02:35:05 +1000553}
554EXPORT_SYMBOL(truncate_pagecache);
555
556/**
Christoph Hellwig2c27c652010-06-04 11:30:04 +0200557 * truncate_setsize - update inode and pagecache for a new file size
558 * @inode: inode
559 * @newsize: new file size
560 *
Jan Kara382e27d2011-01-20 14:44:26 -0800561 * truncate_setsize updates i_size and performs pagecache truncation (if
562 * necessary) to @newsize. It will be typically be called from the filesystem's
563 * setattr function when ATTR_SIZE is passed in.
Christoph Hellwig2c27c652010-06-04 11:30:04 +0200564 *
Jan Kara382e27d2011-01-20 14:44:26 -0800565 * Must be called with inode_mutex held and before all filesystem specific
566 * block truncation has been performed.
Christoph Hellwig2c27c652010-06-04 11:30:04 +0200567 */
568void truncate_setsize(struct inode *inode, loff_t newsize)
569{
570 loff_t oldsize;
571
572 oldsize = inode->i_size;
573 i_size_write(inode, newsize);
574
575 truncate_pagecache(inode, oldsize, newsize);
576}
577EXPORT_SYMBOL(truncate_setsize);
578
579/**
Hugh Dickins623e3db2012-03-28 14:42:40 -0700580 * truncate_pagecache_range - unmap and remove pagecache that is hole-punched
581 * @inode: inode
582 * @lstart: offset of beginning of hole
583 * @lend: offset of last byte of hole
584 *
585 * This function should typically be called before the filesystem
586 * releases resources associated with the freed range (eg. deallocates
587 * blocks). This way, pagecache will always stay logically coherent
588 * with on-disk format, and the filesystem would not have to deal with
589 * situations such as writepage being called for a page that has already
590 * had its underlying blocks deallocated.
591 */
592void truncate_pagecache_range(struct inode *inode, loff_t lstart, loff_t lend)
593{
594 struct address_space *mapping = inode->i_mapping;
595 loff_t unmap_start = round_up(lstart, PAGE_SIZE);
596 loff_t unmap_end = round_down(1 + lend, PAGE_SIZE) - 1;
597 /*
598 * This rounding is currently just for example: unmap_mapping_range
599 * expands its hole outwards, whereas we want it to contract the hole
600 * inwards. However, existing callers of truncate_pagecache_range are
601 * doing their own page rounding first; and truncate_inode_pages_range
602 * currently BUGs if lend is not pagealigned-1 (it handles partial
603 * page at start of hole, but not partial page at end of hole). Note
604 * unmap_mapping_range allows holelen 0 for all, and we allow lend -1.
605 */
606
607 /*
608 * Unlike in truncate_pagecache, unmap_mapping_range is called only
609 * once (before truncating pagecache), and without "even_cows" flag:
610 * hole-punching should not remove private COWed pages from the hole.
611 */
612 if ((u64)unmap_end > (u64)unmap_start)
613 unmap_mapping_range(mapping, unmap_start,
614 1 + unmap_end - unmap_start, 0);
615 truncate_inode_pages_range(mapping, lstart, lend);
616}
617EXPORT_SYMBOL(truncate_pagecache_range);