blob: 66af9031fae8071c6d7a8c0b7e2a3da2c90684c7 [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 Kara90a80202014-10-01 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
Johannes Weiner0cd61442014-04-03 14:47:46 -070026static void clear_exceptional_entry(struct address_space *mapping,
27 pgoff_t index, void *entry)
28{
Johannes Weiner449dd692014-04-03 14:47:56 -070029 struct radix_tree_node *node;
30 void **slot;
31
Johannes Weiner0cd61442014-04-03 14:47:46 -070032 /* Handled by shmem itself */
33 if (shmem_mapping(mapping))
34 return;
35
36 spin_lock_irq(&mapping->tree_lock);
37 /*
38 * Regular page slots are stabilized by the page lock even
39 * without the tree itself locked. These unlocked entries
40 * need verification under the tree lock.
41 */
Johannes Weiner449dd692014-04-03 14:47:56 -070042 if (!__radix_tree_lookup(&mapping->page_tree, index, &node, &slot))
43 goto unlock;
44 if (*slot != entry)
45 goto unlock;
46 radix_tree_replace_slot(slot, NULL);
47 mapping->nrshadows--;
48 if (!node)
49 goto unlock;
50 workingset_node_shadows_dec(node);
51 /*
52 * Don't track node without shadow entries.
53 *
54 * Avoid acquiring the list_lru lock if already untracked.
55 * The list_empty() test is safe as node->private_list is
56 * protected by mapping->tree_lock.
57 */
58 if (!workingset_node_shadows(node) &&
59 !list_empty(&node->private_list))
60 list_lru_del(&workingset_shadow_nodes, &node->private_list);
61 __radix_tree_delete_node(&mapping->page_tree, node);
62unlock:
Johannes Weiner0cd61442014-04-03 14:47:46 -070063 spin_unlock_irq(&mapping->tree_lock);
64}
Linus Torvalds1da177e2005-04-16 15:20:36 -070065
David Howellscf9a2ae2006-08-29 19:05:54 +010066/**
Fengguang Wu28bc44d2008-02-03 18:04:10 +020067 * do_invalidatepage - invalidate part or all of a page
David Howellscf9a2ae2006-08-29 19:05:54 +010068 * @page: the page which is affected
Lukas Czernerd47992f2013-05-21 23:17:23 -040069 * @offset: start of the range to invalidate
70 * @length: length of the range to invalidate
David Howellscf9a2ae2006-08-29 19:05:54 +010071 *
72 * do_invalidatepage() is called when all or part of the page has become
73 * invalidated by a truncate operation.
74 *
75 * do_invalidatepage() does not have to release all buffers, but it must
76 * ensure that no dirty buffer is left outside @offset and that no I/O
77 * is underway against any of the blocks which are outside the truncation
78 * point. Because the caller is about to free (and possibly reuse) those
79 * blocks on-disk.
80 */
Lukas Czernerd47992f2013-05-21 23:17:23 -040081void do_invalidatepage(struct page *page, unsigned int offset,
82 unsigned int length)
David Howellscf9a2ae2006-08-29 19:05:54 +010083{
Lukas Czernerd47992f2013-05-21 23:17:23 -040084 void (*invalidatepage)(struct page *, unsigned int, unsigned int);
85
David Howellscf9a2ae2006-08-29 19:05:54 +010086 invalidatepage = page->mapping->a_ops->invalidatepage;
David Howells93614012006-09-30 20:45:40 +020087#ifdef CONFIG_BLOCK
David Howellscf9a2ae2006-08-29 19:05:54 +010088 if (!invalidatepage)
89 invalidatepage = block_invalidatepage;
David Howells93614012006-09-30 20:45:40 +020090#endif
David Howellscf9a2ae2006-08-29 19:05:54 +010091 if (invalidatepage)
Lukas Czernerd47992f2013-05-21 23:17:23 -040092 (*invalidatepage)(page, offset, length);
David Howellscf9a2ae2006-08-29 19:05:54 +010093}
94
Linus Torvaldsecdfc972007-01-26 12:47:06 -080095/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070096 * If truncate cannot remove the fs-private metadata from the page, the page
Shaohua Li62e1c552008-02-04 22:29:33 -080097 * becomes orphaned. It will be left on the LRU and may even be mapped into
Nick Piggin54cb8822007-07-19 01:46:59 -070098 * user pagetables if we're racing with filemap_fault().
Linus Torvalds1da177e2005-04-16 15:20:36 -070099 *
100 * We need to bale out if page->mapping is no longer equal to the original
101 * mapping. This happens a) when the VM reclaimed the page while we waited on
Andrew Mortonfc0ecff2007-02-10 01:45:39 -0800102 * its lock, b) when a concurrent invalidate_mapping_pages got there first and
Linus Torvalds1da177e2005-04-16 15:20:36 -0700103 * c) when tmpfs swizzles a page between a tmpfs inode and swapper_space.
104 */
Nick Piggin750b4982009-09-16 11:50:12 +0200105static int
Linus Torvalds1da177e2005-04-16 15:20:36 -0700106truncate_complete_page(struct address_space *mapping, struct page *page)
107{
108 if (page->mapping != mapping)
Nick Piggin750b4982009-09-16 11:50:12 +0200109 return -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700110
David Howells266cf652009-04-03 16:42:36 +0100111 if (page_has_private(page))
Lukas Czernerd47992f2013-05-21 23:17:23 -0400112 do_invalidatepage(page, 0, PAGE_CACHE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700113
Konstantin Khlebnikovb9ea2512015-04-14 15:45:27 -0700114 /*
115 * Some filesystems seem to re-dirty the page even after
116 * the VM has canceled the dirty bit (eg ext3 journaling).
117 * Hence dirty accounting check is placed after invalidation.
118 */
119 if (TestClearPageDirty(page))
120 account_page_cleaned(page, mapping);
Bjorn Steinbrinka2b34562008-02-04 22:29:28 -0800121
Linus Torvalds1da177e2005-04-16 15:20:36 -0700122 ClearPageMappedToDisk(page);
Minchan Kim5adc7b52011-03-22 16:32:41 -0700123 delete_from_page_cache(page);
Nick Piggin750b4982009-09-16 11:50:12 +0200124 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700125}
126
127/*
Andrew Mortonfc0ecff2007-02-10 01:45:39 -0800128 * This is for invalidate_mapping_pages(). That function can be called at
Linus Torvalds1da177e2005-04-16 15:20:36 -0700129 * any time, and is not supposed to throw away dirty pages. But pages can
Nick Piggin0fd0e6b2006-09-27 01:50:02 -0700130 * be marked dirty at any time too, so use remove_mapping which safely
131 * discards clean, unused pages.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700132 *
133 * Returns non-zero if the page was successfully invalidated.
134 */
135static int
136invalidate_complete_page(struct address_space *mapping, struct page *page)
137{
Nick Piggin0fd0e6b2006-09-27 01:50:02 -0700138 int ret;
139
Linus Torvalds1da177e2005-04-16 15:20:36 -0700140 if (page->mapping != mapping)
141 return 0;
142
David Howells266cf652009-04-03 16:42:36 +0100143 if (page_has_private(page) && !try_to_release_page(page, 0))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700144 return 0;
145
Nick Piggin0fd0e6b2006-09-27 01:50:02 -0700146 ret = remove_mapping(mapping, page);
Nick Piggin0fd0e6b2006-09-27 01:50:02 -0700147
148 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700149}
150
Nick Piggin750b4982009-09-16 11:50:12 +0200151int truncate_inode_page(struct address_space *mapping, struct page *page)
152{
153 if (page_mapped(page)) {
154 unmap_mapping_range(mapping,
155 (loff_t)page->index << PAGE_CACHE_SHIFT,
156 PAGE_CACHE_SIZE, 0);
157 }
158 return truncate_complete_page(mapping, page);
159}
160
Wu Fengguang83f78662009-09-16 11:50:13 +0200161/*
Andi Kleen25718732009-09-16 11:50:13 +0200162 * Used to get rid of pages on hardware memory corruption.
163 */
164int generic_error_remove_page(struct address_space *mapping, struct page *page)
165{
166 if (!mapping)
167 return -EINVAL;
168 /*
169 * Only punch for normal data pages for now.
170 * Handling other types like directories would need more auditing.
171 */
172 if (!S_ISREG(mapping->host->i_mode))
173 return -EIO;
174 return truncate_inode_page(mapping, page);
175}
176EXPORT_SYMBOL(generic_error_remove_page);
177
178/*
Wu Fengguang83f78662009-09-16 11:50:13 +0200179 * Safely invalidate one page from its pagecache mapping.
180 * It only drops clean, unused pages. The page must be locked.
181 *
182 * Returns 1 if the page is successfully invalidated, otherwise 0.
183 */
184int invalidate_inode_page(struct page *page)
185{
186 struct address_space *mapping = page_mapping(page);
187 if (!mapping)
188 return 0;
189 if (PageDirty(page) || PageWriteback(page))
190 return 0;
191 if (page_mapped(page))
192 return 0;
193 return invalidate_complete_page(mapping, page);
194}
195
Linus Torvalds1da177e2005-04-16 15:20:36 -0700196/**
Liu Bo73c1e202012-02-21 10:57:20 +0800197 * truncate_inode_pages_range - truncate range of pages specified by start & end byte offsets
Linus Torvalds1da177e2005-04-16 15:20:36 -0700198 * @mapping: mapping to truncate
199 * @lstart: offset from which to truncate
Lukas Czerner5a720392013-05-27 23:32:35 -0400200 * @lend: offset to which to truncate (inclusive)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700201 *
Hans Reiserd7339072006-01-06 00:10:36 -0800202 * Truncate the page cache, removing the pages that are between
Lukas Czerner5a720392013-05-27 23:32:35 -0400203 * specified offsets (and zeroing out partial pages
204 * if lstart or lend + 1 is not page aligned).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700205 *
206 * Truncate takes two passes - the first pass is nonblocking. It will not
207 * block on page locks and it will not block on writeback. The second pass
208 * will wait. This is to prevent as much IO as possible in the affected region.
209 * The first pass will remove most pages, so the search cost of the second pass
210 * is low.
211 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700212 * We pass down the cache-hot hint to the page freeing code. Even if the
213 * mapping is large, it is probably the case that the final pages are the most
214 * recently touched, and freeing happens in ascending file offset order.
Lukas Czerner5a720392013-05-27 23:32:35 -0400215 *
216 * Note that since ->invalidatepage() accepts range to invalidate
217 * truncate_inode_pages_range is able to handle cases where lend + 1 is not
218 * page aligned properly.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700219 */
Hans Reiserd7339072006-01-06 00:10:36 -0800220void truncate_inode_pages_range(struct address_space *mapping,
221 loff_t lstart, loff_t lend)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700222{
Lukas Czerner5a720392013-05-27 23:32:35 -0400223 pgoff_t start; /* inclusive */
224 pgoff_t end; /* exclusive */
225 unsigned int partial_start; /* inclusive */
226 unsigned int partial_end; /* exclusive */
227 struct pagevec pvec;
Johannes Weiner0cd61442014-04-03 14:47:46 -0700228 pgoff_t indices[PAGEVEC_SIZE];
Lukas Czerner5a720392013-05-27 23:32:35 -0400229 pgoff_t index;
230 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700231
Dan Magenheimer31677602011-09-21 11:56:28 -0400232 cleancache_invalidate_inode(mapping);
Johannes Weiner91b0abe2014-04-03 14:47:49 -0700233 if (mapping->nrpages == 0 && mapping->nrshadows == 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700234 return;
235
Lukas Czerner5a720392013-05-27 23:32:35 -0400236 /* Offsets within partial pages */
237 partial_start = lstart & (PAGE_CACHE_SIZE - 1);
238 partial_end = (lend + 1) & (PAGE_CACHE_SIZE - 1);
239
240 /*
241 * 'start' and 'end' always covers the range of pages to be fully
242 * truncated. Partial pages are covered with 'partial_start' at the
243 * start of the range and 'partial_end' at the end of the range.
244 * Note that 'end' is exclusive while 'lend' is inclusive.
245 */
246 start = (lstart + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
247 if (lend == -1)
248 /*
249 * lend == -1 indicates end-of-file so we have to set 'end'
250 * to the highest possible pgoff_t and since the type is
251 * unsigned we're using -1.
252 */
253 end = -1;
254 else
255 end = (lend + 1) >> PAGE_CACHE_SHIFT;
Hans Reiserd7339072006-01-06 00:10:36 -0800256
Linus Torvalds1da177e2005-04-16 15:20:36 -0700257 pagevec_init(&pvec, 0);
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700258 index = start;
Johannes Weiner0cd61442014-04-03 14:47:46 -0700259 while (index < end && pagevec_lookup_entries(&pvec, mapping, index,
260 min(end - index, (pgoff_t)PAGEVEC_SIZE),
261 indices)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700262 for (i = 0; i < pagevec_count(&pvec); i++) {
263 struct page *page = pvec.pages[i];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700264
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700265 /* We rely upon deletion not changing page->index */
Johannes Weiner0cd61442014-04-03 14:47:46 -0700266 index = indices[i];
Lukas Czerner5a720392013-05-27 23:32:35 -0400267 if (index >= end)
Hans Reiserd7339072006-01-06 00:10:36 -0800268 break;
Hans Reiserd7339072006-01-06 00:10:36 -0800269
Johannes Weiner0cd61442014-04-03 14:47:46 -0700270 if (radix_tree_exceptional_entry(page)) {
271 clear_exceptional_entry(mapping, index, page);
272 continue;
273 }
274
Nick Piggin529ae9a2008-08-02 12:01:03 +0200275 if (!trylock_page(page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700276 continue;
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700277 WARN_ON(page->index != index);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700278 if (PageWriteback(page)) {
279 unlock_page(page);
280 continue;
281 }
Nick Piggin750b4982009-09-16 11:50:12 +0200282 truncate_inode_page(mapping, page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700283 unlock_page(page);
284 }
Johannes Weiner0cd61442014-04-03 14:47:46 -0700285 pagevec_remove_exceptionals(&pvec);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700286 pagevec_release(&pvec);
287 cond_resched();
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700288 index++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700289 }
290
Lukas Czerner5a720392013-05-27 23:32:35 -0400291 if (partial_start) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700292 struct page *page = find_lock_page(mapping, start - 1);
293 if (page) {
Lukas Czerner5a720392013-05-27 23:32:35 -0400294 unsigned int top = PAGE_CACHE_SIZE;
295 if (start > end) {
296 /* Truncation within a single page */
297 top = partial_end;
298 partial_end = 0;
299 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700300 wait_on_page_writeback(page);
Lukas Czerner5a720392013-05-27 23:32:35 -0400301 zero_user_segment(page, partial_start, top);
302 cleancache_invalidate_page(mapping, page);
303 if (page_has_private(page))
304 do_invalidatepage(page, partial_start,
305 top - partial_start);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700306 unlock_page(page);
307 page_cache_release(page);
308 }
309 }
Lukas Czerner5a720392013-05-27 23:32:35 -0400310 if (partial_end) {
311 struct page *page = find_lock_page(mapping, end);
312 if (page) {
313 wait_on_page_writeback(page);
314 zero_user_segment(page, 0, partial_end);
315 cleancache_invalidate_page(mapping, page);
316 if (page_has_private(page))
317 do_invalidatepage(page, 0,
318 partial_end);
319 unlock_page(page);
320 page_cache_release(page);
321 }
322 }
323 /*
324 * If the truncation happened within a single page no pages
325 * will be released, just zeroed, so we can bail out now.
326 */
327 if (start >= end)
328 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700329
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700330 index = start;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700331 for ( ; ; ) {
332 cond_resched();
Johannes Weiner0cd61442014-04-03 14:47:46 -0700333 if (!pagevec_lookup_entries(&pvec, mapping, index,
Hugh Dickins792ceae2014-07-23 14:00:15 -0700334 min(end - index, (pgoff_t)PAGEVEC_SIZE), indices)) {
335 /* If all gone from start onwards, we're done */
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700336 if (index == start)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700337 break;
Hugh Dickins792ceae2014-07-23 14:00:15 -0700338 /* Otherwise restart to make sure all gone */
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700339 index = start;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700340 continue;
341 }
Johannes Weiner0cd61442014-04-03 14:47:46 -0700342 if (index == start && indices[0] >= end) {
Hugh Dickins792ceae2014-07-23 14:00:15 -0700343 /* All gone out of hole to be punched, we're done */
Johannes Weiner0cd61442014-04-03 14:47:46 -0700344 pagevec_remove_exceptionals(&pvec);
Hans Reiserd7339072006-01-06 00:10:36 -0800345 pagevec_release(&pvec);
346 break;
347 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700348 for (i = 0; i < pagevec_count(&pvec); i++) {
349 struct page *page = pvec.pages[i];
350
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700351 /* We rely upon deletion not changing page->index */
Johannes Weiner0cd61442014-04-03 14:47:46 -0700352 index = indices[i];
Hugh Dickins792ceae2014-07-23 14:00:15 -0700353 if (index >= end) {
354 /* Restart punch to make sure all gone */
355 index = start - 1;
Hans Reiserd7339072006-01-06 00:10:36 -0800356 break;
Hugh Dickins792ceae2014-07-23 14:00:15 -0700357 }
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700358
Johannes Weiner0cd61442014-04-03 14:47:46 -0700359 if (radix_tree_exceptional_entry(page)) {
360 clear_exceptional_entry(mapping, index, page);
361 continue;
362 }
363
Linus Torvalds1da177e2005-04-16 15:20:36 -0700364 lock_page(page);
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700365 WARN_ON(page->index != index);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700366 wait_on_page_writeback(page);
Nick Piggin750b4982009-09-16 11:50:12 +0200367 truncate_inode_page(mapping, page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700368 unlock_page(page);
369 }
Johannes Weiner0cd61442014-04-03 14:47:46 -0700370 pagevec_remove_exceptionals(&pvec);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700371 pagevec_release(&pvec);
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700372 index++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700373 }
Dan Magenheimer31677602011-09-21 11:56:28 -0400374 cleancache_invalidate_inode(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700375}
Hans Reiserd7339072006-01-06 00:10:36 -0800376EXPORT_SYMBOL(truncate_inode_pages_range);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700377
Hans Reiserd7339072006-01-06 00:10:36 -0800378/**
379 * truncate_inode_pages - truncate *all* the pages from an offset
380 * @mapping: mapping to truncate
381 * @lstart: offset from which to truncate
382 *
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800383 * Called under (and serialised by) inode->i_mutex.
Jan Kara08142572011-06-27 16:18:10 -0700384 *
385 * Note: When this function returns, there can be a page in the process of
386 * deletion (inside __delete_from_page_cache()) in the specified range. Thus
387 * mapping->nrpages can be non-zero when this function returns even after
388 * truncation of the whole mapping.
Hans Reiserd7339072006-01-06 00:10:36 -0800389 */
390void truncate_inode_pages(struct address_space *mapping, loff_t lstart)
391{
392 truncate_inode_pages_range(mapping, lstart, (loff_t)-1);
393}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700394EXPORT_SYMBOL(truncate_inode_pages);
395
Mike Waychison28697352009-06-16 15:32:59 -0700396/**
Johannes Weiner91b0abe2014-04-03 14:47:49 -0700397 * truncate_inode_pages_final - truncate *all* pages before inode dies
398 * @mapping: mapping to truncate
399 *
400 * Called under (and serialized by) inode->i_mutex.
401 *
402 * Filesystems have to use this in the .evict_inode path to inform the
403 * VM that this is the final truncate and the inode is going away.
404 */
405void truncate_inode_pages_final(struct address_space *mapping)
406{
407 unsigned long nrshadows;
408 unsigned long nrpages;
409
410 /*
411 * Page reclaim can not participate in regular inode lifetime
412 * management (can't call iput()) and thus can race with the
413 * inode teardown. Tell it when the address space is exiting,
414 * so that it does not install eviction information after the
415 * final truncate has begun.
416 */
417 mapping_set_exiting(mapping);
418
419 /*
420 * When reclaim installs eviction entries, it increases
421 * nrshadows first, then decreases nrpages. Make sure we see
422 * this in the right order or we might miss an entry.
423 */
424 nrpages = mapping->nrpages;
425 smp_rmb();
426 nrshadows = mapping->nrshadows;
427
428 if (nrpages || nrshadows) {
429 /*
430 * As truncation uses a lockless tree lookup, cycle
431 * the tree lock to make sure any ongoing tree
432 * modification that does not see AS_EXITING is
433 * completed before starting the final truncate.
434 */
435 spin_lock_irq(&mapping->tree_lock);
436 spin_unlock_irq(&mapping->tree_lock);
437
438 truncate_inode_pages(mapping, 0);
439 }
440}
441EXPORT_SYMBOL(truncate_inode_pages_final);
442
443/**
Mike Waychison28697352009-06-16 15:32:59 -0700444 * invalidate_mapping_pages - Invalidate all the unlocked pages of one inode
445 * @mapping: the address_space which holds the pages to invalidate
446 * @start: the offset 'from' which to invalidate
447 * @end: the offset 'to' which to invalidate (inclusive)
448 *
449 * This function only removes the unlocked pages, if you want to
450 * remove all the pages of one inode, you must call truncate_inode_pages.
451 *
452 * invalidate_mapping_pages() will not block on IO activity. It will not
453 * invalidate pages which are dirty, locked, under writeback or mapped into
454 * pagetables.
455 */
456unsigned long invalidate_mapping_pages(struct address_space *mapping,
Minchan Kim31560182011-03-22 16:32:52 -0700457 pgoff_t start, pgoff_t end)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700458{
Johannes Weiner0cd61442014-04-03 14:47:46 -0700459 pgoff_t indices[PAGEVEC_SIZE];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700460 struct pagevec pvec;
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700461 pgoff_t index = start;
Minchan Kim31560182011-03-22 16:32:52 -0700462 unsigned long ret;
463 unsigned long count = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700464 int i;
465
466 pagevec_init(&pvec, 0);
Johannes Weiner0cd61442014-04-03 14:47:46 -0700467 while (index <= end && pagevec_lookup_entries(&pvec, mapping, index,
468 min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1,
469 indices)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700470 for (i = 0; i < pagevec_count(&pvec); i++) {
471 struct page *page = pvec.pages[i];
472
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700473 /* We rely upon deletion not changing page->index */
Johannes Weiner0cd61442014-04-03 14:47:46 -0700474 index = indices[i];
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700475 if (index > end)
476 break;
NeilBrowne0f23602006-06-23 02:05:48 -0700477
Johannes Weiner0cd61442014-04-03 14:47:46 -0700478 if (radix_tree_exceptional_entry(page)) {
479 clear_exceptional_entry(mapping, index, page);
480 continue;
481 }
482
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700483 if (!trylock_page(page))
484 continue;
485 WARN_ON(page->index != index);
Minchan Kim31560182011-03-22 16:32:52 -0700486 ret = invalidate_inode_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700487 unlock_page(page);
Minchan Kim31560182011-03-22 16:32:52 -0700488 /*
489 * Invalidation is a hint that the page is no longer
490 * of interest and try to speed up its reclaim.
491 */
492 if (!ret)
Minchan Kimcc5993b2015-04-15 16:13:26 -0700493 deactivate_file_page(page);
Minchan Kim31560182011-03-22 16:32:52 -0700494 count += ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700495 }
Johannes Weiner0cd61442014-04-03 14:47:46 -0700496 pagevec_remove_exceptionals(&pvec);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700497 pagevec_release(&pvec);
Mike Waychison28697352009-06-16 15:32:59 -0700498 cond_resched();
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700499 index++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700500 }
Minchan Kim31560182011-03-22 16:32:52 -0700501 return count;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700502}
Anton Altaparmakov54bc4852007-02-10 01:45:38 -0800503EXPORT_SYMBOL(invalidate_mapping_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700504
Andrew Mortonbd4c8ce2006-09-30 23:29:29 -0700505/*
506 * This is like invalidate_complete_page(), except it ignores the page's
507 * refcount. We do this because invalidate_inode_pages2() needs stronger
508 * invalidation guarantees, and cannot afford to leave pages behind because
Anderson Briglia2706a1b2007-07-15 23:38:09 -0700509 * shrink_page_list() has a temp ref on them, or because they're transiently
510 * sitting in the lru_cache_add() pagevecs.
Andrew Mortonbd4c8ce2006-09-30 23:29:29 -0700511 */
512static int
513invalidate_complete_page2(struct address_space *mapping, struct page *page)
514{
515 if (page->mapping != mapping)
516 return 0;
517
David Howells266cf652009-04-03 16:42:36 +0100518 if (page_has_private(page) && !try_to_release_page(page, GFP_KERNEL))
Andrew Mortonbd4c8ce2006-09-30 23:29:29 -0700519 return 0;
520
Nick Piggin19fd6232008-07-25 19:45:32 -0700521 spin_lock_irq(&mapping->tree_lock);
Andrew Mortonbd4c8ce2006-09-30 23:29:29 -0700522 if (PageDirty(page))
523 goto failed;
524
David Howells266cf652009-04-03 16:42:36 +0100525 BUG_ON(page_has_private(page));
Johannes Weiner91b0abe2014-04-03 14:47:49 -0700526 __delete_from_page_cache(page, NULL);
Nick Piggin19fd6232008-07-25 19:45:32 -0700527 spin_unlock_irq(&mapping->tree_lock);
Linus Torvalds6072d132010-12-01 13:35:19 -0500528
529 if (mapping->a_ops->freepage)
530 mapping->a_ops->freepage(page);
531
Andrew Mortonbd4c8ce2006-09-30 23:29:29 -0700532 page_cache_release(page); /* pagecache ref */
533 return 1;
534failed:
Nick Piggin19fd6232008-07-25 19:45:32 -0700535 spin_unlock_irq(&mapping->tree_lock);
Andrew Mortonbd4c8ce2006-09-30 23:29:29 -0700536 return 0;
537}
538
Trond Myklebuste3db7692007-01-10 23:15:39 -0800539static int do_launder_page(struct address_space *mapping, struct page *page)
540{
541 if (!PageDirty(page))
542 return 0;
543 if (page->mapping != mapping || mapping->a_ops->launder_page == NULL)
544 return 0;
545 return mapping->a_ops->launder_page(page);
546}
547
Linus Torvalds1da177e2005-04-16 15:20:36 -0700548/**
549 * invalidate_inode_pages2_range - remove range of pages from an address_space
Martin Waitz67be2dd2005-05-01 08:59:26 -0700550 * @mapping: the address_space
Linus Torvalds1da177e2005-04-16 15:20:36 -0700551 * @start: the page offset 'from' which to invalidate
552 * @end: the page offset 'to' which to invalidate (inclusive)
553 *
554 * Any pages which are found to be mapped into pagetables are unmapped prior to
555 * invalidation.
556 *
Hisashi Hifumi6ccfa802008-09-02 14:35:40 -0700557 * Returns -EBUSY if any pages could not be invalidated.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700558 */
559int invalidate_inode_pages2_range(struct address_space *mapping,
560 pgoff_t start, pgoff_t end)
561{
Johannes Weiner0cd61442014-04-03 14:47:46 -0700562 pgoff_t indices[PAGEVEC_SIZE];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700563 struct pagevec pvec;
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700564 pgoff_t index;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700565 int i;
566 int ret = 0;
Hisashi Hifumi0dd13342008-04-28 02:12:08 -0700567 int ret2 = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700568 int did_range_unmap = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700569
Dan Magenheimer31677602011-09-21 11:56:28 -0400570 cleancache_invalidate_inode(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700571 pagevec_init(&pvec, 0);
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700572 index = start;
Johannes Weiner0cd61442014-04-03 14:47:46 -0700573 while (index <= end && pagevec_lookup_entries(&pvec, mapping, index,
574 min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1,
575 indices)) {
Trond Myklebust7b965e02007-02-28 20:13:55 -0800576 for (i = 0; i < pagevec_count(&pvec); i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700577 struct page *page = pvec.pages[i];
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700578
579 /* We rely upon deletion not changing page->index */
Johannes Weiner0cd61442014-04-03 14:47:46 -0700580 index = indices[i];
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700581 if (index > end)
582 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700583
Johannes Weiner0cd61442014-04-03 14:47:46 -0700584 if (radix_tree_exceptional_entry(page)) {
585 clear_exceptional_entry(mapping, index, page);
586 continue;
587 }
588
Linus Torvalds1da177e2005-04-16 15:20:36 -0700589 lock_page(page);
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700590 WARN_ON(page->index != index);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700591 if (page->mapping != mapping) {
592 unlock_page(page);
593 continue;
594 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700595 wait_on_page_writeback(page);
Nick Piggind00806b2007-07-19 01:46:57 -0700596 if (page_mapped(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700597 if (!did_range_unmap) {
598 /*
599 * Zap the rest of the file in one hit.
600 */
601 unmap_mapping_range(mapping,
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700602 (loff_t)index << PAGE_CACHE_SHIFT,
603 (loff_t)(1 + end - index)
604 << PAGE_CACHE_SHIFT,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700605 0);
606 did_range_unmap = 1;
607 } else {
608 /*
609 * Just zap this page
610 */
611 unmap_mapping_range(mapping,
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700612 (loff_t)index << PAGE_CACHE_SHIFT,
613 PAGE_CACHE_SIZE, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700614 }
615 }
Nick Piggind00806b2007-07-19 01:46:57 -0700616 BUG_ON(page_mapped(page));
Hisashi Hifumi0dd13342008-04-28 02:12:08 -0700617 ret2 = do_launder_page(mapping, page);
618 if (ret2 == 0) {
619 if (!invalidate_complete_page2(mapping, page))
Hisashi Hifumi6ccfa802008-09-02 14:35:40 -0700620 ret2 = -EBUSY;
Hisashi Hifumi0dd13342008-04-28 02:12:08 -0700621 }
622 if (ret2 < 0)
623 ret = ret2;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700624 unlock_page(page);
625 }
Johannes Weiner0cd61442014-04-03 14:47:46 -0700626 pagevec_remove_exceptionals(&pvec);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700627 pagevec_release(&pvec);
628 cond_resched();
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700629 index++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700630 }
Dan Magenheimer31677602011-09-21 11:56:28 -0400631 cleancache_invalidate_inode(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700632 return ret;
633}
634EXPORT_SYMBOL_GPL(invalidate_inode_pages2_range);
635
636/**
637 * invalidate_inode_pages2 - remove all pages from an address_space
Martin Waitz67be2dd2005-05-01 08:59:26 -0700638 * @mapping: the address_space
Linus Torvalds1da177e2005-04-16 15:20:36 -0700639 *
640 * Any pages which are found to be mapped into pagetables are unmapped prior to
641 * invalidation.
642 *
Peng Taoe9de25d2009-10-19 14:48:13 +0800643 * Returns -EBUSY if any pages could not be invalidated.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700644 */
645int invalidate_inode_pages2(struct address_space *mapping)
646{
647 return invalidate_inode_pages2_range(mapping, 0, -1);
648}
649EXPORT_SYMBOL_GPL(invalidate_inode_pages2);
npiggin@suse.de25d9e2d2009-08-21 02:35:05 +1000650
651/**
652 * truncate_pagecache - unmap and remove pagecache that has been truncated
653 * @inode: inode
Hugh Dickins8a549be2011-07-25 17:12:24 -0700654 * @newsize: new file size
npiggin@suse.de25d9e2d2009-08-21 02:35:05 +1000655 *
656 * inode's new i_size must already be written before truncate_pagecache
657 * is called.
658 *
659 * This function should typically be called before the filesystem
660 * releases resources associated with the freed range (eg. deallocates
661 * blocks). This way, pagecache will always stay logically coherent
662 * with on-disk format, and the filesystem would not have to deal with
663 * situations such as writepage being called for a page that has already
664 * had its underlying blocks deallocated.
665 */
Kirill A. Shutemov7caef262013-09-12 15:13:56 -0700666void truncate_pagecache(struct inode *inode, loff_t newsize)
npiggin@suse.de25d9e2d2009-08-21 02:35:05 +1000667{
OGAWA Hirofumicedabed2010-01-13 21:14:09 +0900668 struct address_space *mapping = inode->i_mapping;
Hugh Dickins8a549be2011-07-25 17:12:24 -0700669 loff_t holebegin = round_up(newsize, PAGE_SIZE);
npiggin@suse.de25d9e2d2009-08-21 02:35:05 +1000670
OGAWA Hirofumicedabed2010-01-13 21:14:09 +0900671 /*
672 * unmap_mapping_range is called twice, first simply for
673 * efficiency so that truncate_inode_pages does fewer
674 * single-page unmaps. However after this first call, and
675 * before truncate_inode_pages finishes, it is possible for
676 * private pages to be COWed, which remain after
677 * truncate_inode_pages finishes, hence the second
678 * unmap_mapping_range call must be made for correctness.
679 */
Hugh Dickins8a549be2011-07-25 17:12:24 -0700680 unmap_mapping_range(mapping, holebegin, 0, 1);
681 truncate_inode_pages(mapping, newsize);
682 unmap_mapping_range(mapping, holebegin, 0, 1);
npiggin@suse.de25d9e2d2009-08-21 02:35:05 +1000683}
684EXPORT_SYMBOL(truncate_pagecache);
685
686/**
Christoph Hellwig2c27c652010-06-04 11:30:04 +0200687 * truncate_setsize - update inode and pagecache for a new file size
688 * @inode: inode
689 * @newsize: new file size
690 *
Jan Kara382e27d2011-01-20 14:44:26 -0800691 * truncate_setsize updates i_size and performs pagecache truncation (if
692 * necessary) to @newsize. It will be typically be called from the filesystem's
693 * setattr function when ATTR_SIZE is passed in.
Christoph Hellwig2c27c652010-06-04 11:30:04 +0200694 *
Jan Kara77783d02014-11-07 08:29:25 +1100695 * Must be called with a lock serializing truncates and writes (generally
696 * i_mutex but e.g. xfs uses a different lock) and before all filesystem
697 * specific block truncation has been performed.
Christoph Hellwig2c27c652010-06-04 11:30:04 +0200698 */
699void truncate_setsize(struct inode *inode, loff_t newsize)
700{
Jan Kara90a80202014-10-01 21:49:18 -0400701 loff_t oldsize = inode->i_size;
702
Christoph Hellwig2c27c652010-06-04 11:30:04 +0200703 i_size_write(inode, newsize);
Jan Kara90a80202014-10-01 21:49:18 -0400704 if (newsize > oldsize)
705 pagecache_isize_extended(inode, oldsize, newsize);
Kirill A. Shutemov7caef262013-09-12 15:13:56 -0700706 truncate_pagecache(inode, newsize);
Christoph Hellwig2c27c652010-06-04 11:30:04 +0200707}
708EXPORT_SYMBOL(truncate_setsize);
709
710/**
Jan Kara90a80202014-10-01 21:49:18 -0400711 * pagecache_isize_extended - update pagecache after extension of i_size
712 * @inode: inode for which i_size was extended
713 * @from: original inode size
714 * @to: new inode size
715 *
716 * Handle extension of inode size either caused by extending truncate or by
717 * write starting after current i_size. We mark the page straddling current
718 * i_size RO so that page_mkwrite() is called on the nearest write access to
719 * the page. This way filesystem can be sure that page_mkwrite() is called on
720 * the page before user writes to the page via mmap after the i_size has been
721 * changed.
722 *
723 * The function must be called after i_size is updated so that page fault
724 * coming after we unlock the page will already see the new i_size.
725 * The function must be called while we still hold i_mutex - this not only
726 * makes sure i_size is stable but also that userspace cannot observe new
727 * i_size value before we are prepared to store mmap writes at new inode size.
728 */
729void pagecache_isize_extended(struct inode *inode, loff_t from, loff_t to)
730{
731 int bsize = 1 << inode->i_blkbits;
732 loff_t rounded_from;
733 struct page *page;
734 pgoff_t index;
735
Jan Kara90a80202014-10-01 21:49:18 -0400736 WARN_ON(to > inode->i_size);
737
738 if (from >= to || bsize == PAGE_CACHE_SIZE)
739 return;
740 /* Page straddling @from will not have any hole block created? */
741 rounded_from = round_up(from, bsize);
742 if (to <= rounded_from || !(rounded_from & (PAGE_CACHE_SIZE - 1)))
743 return;
744
745 index = from >> PAGE_CACHE_SHIFT;
746 page = find_lock_page(inode->i_mapping, index);
747 /* Page not cached? Nothing to do */
748 if (!page)
749 return;
750 /*
751 * See clear_page_dirty_for_io() for details why set_page_dirty()
752 * is needed.
753 */
754 if (page_mkclean(page))
755 set_page_dirty(page);
756 unlock_page(page);
757 page_cache_release(page);
758}
759EXPORT_SYMBOL(pagecache_isize_extended);
760
761/**
Hugh Dickins623e3db2012-03-28 14:42:40 -0700762 * truncate_pagecache_range - unmap and remove pagecache that is hole-punched
763 * @inode: inode
764 * @lstart: offset of beginning of hole
765 * @lend: offset of last byte of hole
766 *
767 * This function should typically be called before the filesystem
768 * releases resources associated with the freed range (eg. deallocates
769 * blocks). This way, pagecache will always stay logically coherent
770 * with on-disk format, and the filesystem would not have to deal with
771 * situations such as writepage being called for a page that has already
772 * had its underlying blocks deallocated.
773 */
774void truncate_pagecache_range(struct inode *inode, loff_t lstart, loff_t lend)
775{
776 struct address_space *mapping = inode->i_mapping;
777 loff_t unmap_start = round_up(lstart, PAGE_SIZE);
778 loff_t unmap_end = round_down(1 + lend, PAGE_SIZE) - 1;
779 /*
780 * This rounding is currently just for example: unmap_mapping_range
781 * expands its hole outwards, whereas we want it to contract the hole
782 * inwards. However, existing callers of truncate_pagecache_range are
Lukas Czerner5a720392013-05-27 23:32:35 -0400783 * doing their own page rounding first. Note that unmap_mapping_range
784 * allows holelen 0 for all, and we allow lend -1 for end of file.
Hugh Dickins623e3db2012-03-28 14:42:40 -0700785 */
786
787 /*
788 * Unlike in truncate_pagecache, unmap_mapping_range is called only
789 * once (before truncating pagecache), and without "even_cows" flag:
790 * hole-punching should not remove private COWed pages from the hole.
791 */
792 if ((u64)unmap_end > (u64)unmap_start)
793 unmap_mapping_range(mapping, unmap_start,
794 1 + unmap_end - unmap_start, 0);
795 truncate_inode_pages_range(mapping, lstart, lend);
796}
797EXPORT_SYMBOL(truncate_pagecache_range);