blob: f1e4d60523694eb758b93fcd6487866c203956dd [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/*
96 * This cancels just the dirty bit on the kernel page itself, it
97 * does NOT actually remove dirty bits on any mmap's that may be
98 * around. It also leaves the page tagged dirty, so any sync
99 * activity will still find it on the dirty lists, and in particular,
100 * clear_page_dirty_for_io() will still look at the dirty bits in
101 * the VM.
102 *
103 * Doing this should *normally* only ever be done when a page
104 * is truncated, and is not actually mapped anywhere at all. However,
105 * fs/buffer.c does this when it notices that somebody has cleaned
106 * out all the buffers on a page without actually doing it through
107 * the VM. Can you say "ext3 is horribly ugly"? Tought you could.
108 */
Linus Torvaldsfba25912006-12-20 13:46:42 -0800109void cancel_dirty_page(struct page *page, unsigned int account_size)
110{
Linus Torvalds8368e322006-12-23 09:25:04 -0800111 if (TestClearPageDirty(page)) {
112 struct address_space *mapping = page->mapping;
113 if (mapping && mapping_cap_account_dirty(mapping)) {
114 dec_zone_page_state(page, NR_FILE_DIRTY);
Peter Zijlstrac9e51e42007-10-16 23:25:47 -0700115 dec_bdi_stat(mapping->backing_dev_info,
116 BDI_RECLAIMABLE);
Linus Torvalds8368e322006-12-23 09:25:04 -0800117 if (account_size)
118 task_io_account_cancelled_write(account_size);
119 }
Andrew Morton3e67c092006-12-21 11:00:33 -0800120 }
Linus Torvaldsfba25912006-12-20 13:46:42 -0800121}
Linus Torvalds8368e322006-12-23 09:25:04 -0800122EXPORT_SYMBOL(cancel_dirty_page);
Linus Torvaldsfba25912006-12-20 13:46:42 -0800123
Linus Torvalds1da177e2005-04-16 15:20:36 -0700124/*
125 * If truncate cannot remove the fs-private metadata from the page, the page
Shaohua Li62e1c552008-02-04 22:29:33 -0800126 * becomes orphaned. It will be left on the LRU and may even be mapped into
Nick Piggin54cb8822007-07-19 01:46:59 -0700127 * user pagetables if we're racing with filemap_fault().
Linus Torvalds1da177e2005-04-16 15:20:36 -0700128 *
129 * We need to bale out if page->mapping is no longer equal to the original
130 * mapping. This happens a) when the VM reclaimed the page while we waited on
Andrew Mortonfc0ecff2007-02-10 01:45:39 -0800131 * its lock, b) when a concurrent invalidate_mapping_pages got there first and
Linus Torvalds1da177e2005-04-16 15:20:36 -0700132 * c) when tmpfs swizzles a page between a tmpfs inode and swapper_space.
133 */
Nick Piggin750b4982009-09-16 11:50:12 +0200134static int
Linus Torvalds1da177e2005-04-16 15:20:36 -0700135truncate_complete_page(struct address_space *mapping, struct page *page)
136{
137 if (page->mapping != mapping)
Nick Piggin750b4982009-09-16 11:50:12 +0200138 return -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700139
David Howells266cf652009-04-03 16:42:36 +0100140 if (page_has_private(page))
Lukas Czernerd47992f2013-05-21 23:17:23 -0400141 do_invalidatepage(page, 0, PAGE_CACHE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700142
Bjorn Steinbrinka2b34562008-02-04 22:29:28 -0800143 cancel_dirty_page(page, PAGE_CACHE_SIZE);
144
Linus Torvalds1da177e2005-04-16 15:20:36 -0700145 ClearPageMappedToDisk(page);
Minchan Kim5adc7b52011-03-22 16:32:41 -0700146 delete_from_page_cache(page);
Nick Piggin750b4982009-09-16 11:50:12 +0200147 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700148}
149
150/*
Andrew Mortonfc0ecff2007-02-10 01:45:39 -0800151 * This is for invalidate_mapping_pages(). That function can be called at
Linus Torvalds1da177e2005-04-16 15:20:36 -0700152 * any time, and is not supposed to throw away dirty pages. But pages can
Nick Piggin0fd0e6b2006-09-27 01:50:02 -0700153 * be marked dirty at any time too, so use remove_mapping which safely
154 * discards clean, unused pages.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700155 *
156 * Returns non-zero if the page was successfully invalidated.
157 */
158static int
159invalidate_complete_page(struct address_space *mapping, struct page *page)
160{
Nick Piggin0fd0e6b2006-09-27 01:50:02 -0700161 int ret;
162
Linus Torvalds1da177e2005-04-16 15:20:36 -0700163 if (page->mapping != mapping)
164 return 0;
165
David Howells266cf652009-04-03 16:42:36 +0100166 if (page_has_private(page) && !try_to_release_page(page, 0))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700167 return 0;
168
Nick Piggin0fd0e6b2006-09-27 01:50:02 -0700169 ret = remove_mapping(mapping, page);
Nick Piggin0fd0e6b2006-09-27 01:50:02 -0700170
171 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700172}
173
Nick Piggin750b4982009-09-16 11:50:12 +0200174int truncate_inode_page(struct address_space *mapping, struct page *page)
175{
176 if (page_mapped(page)) {
177 unmap_mapping_range(mapping,
178 (loff_t)page->index << PAGE_CACHE_SHIFT,
179 PAGE_CACHE_SIZE, 0);
180 }
181 return truncate_complete_page(mapping, page);
182}
183
Wu Fengguang83f78662009-09-16 11:50:13 +0200184/*
Andi Kleen25718732009-09-16 11:50:13 +0200185 * Used to get rid of pages on hardware memory corruption.
186 */
187int generic_error_remove_page(struct address_space *mapping, struct page *page)
188{
189 if (!mapping)
190 return -EINVAL;
191 /*
192 * Only punch for normal data pages for now.
193 * Handling other types like directories would need more auditing.
194 */
195 if (!S_ISREG(mapping->host->i_mode))
196 return -EIO;
197 return truncate_inode_page(mapping, page);
198}
199EXPORT_SYMBOL(generic_error_remove_page);
200
201/*
Wu Fengguang83f78662009-09-16 11:50:13 +0200202 * Safely invalidate one page from its pagecache mapping.
203 * It only drops clean, unused pages. The page must be locked.
204 *
205 * Returns 1 if the page is successfully invalidated, otherwise 0.
206 */
207int invalidate_inode_page(struct page *page)
208{
209 struct address_space *mapping = page_mapping(page);
210 if (!mapping)
211 return 0;
212 if (PageDirty(page) || PageWriteback(page))
213 return 0;
214 if (page_mapped(page))
215 return 0;
216 return invalidate_complete_page(mapping, page);
217}
218
Linus Torvalds1da177e2005-04-16 15:20:36 -0700219/**
Liu Bo73c1e202012-02-21 10:57:20 +0800220 * truncate_inode_pages_range - truncate range of pages specified by start & end byte offsets
Linus Torvalds1da177e2005-04-16 15:20:36 -0700221 * @mapping: mapping to truncate
222 * @lstart: offset from which to truncate
Lukas Czerner5a720392013-05-27 23:32:35 -0400223 * @lend: offset to which to truncate (inclusive)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700224 *
Hans Reiserd7339072006-01-06 00:10:36 -0800225 * Truncate the page cache, removing the pages that are between
Lukas Czerner5a720392013-05-27 23:32:35 -0400226 * specified offsets (and zeroing out partial pages
227 * if lstart or lend + 1 is not page aligned).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700228 *
229 * Truncate takes two passes - the first pass is nonblocking. It will not
230 * block on page locks and it will not block on writeback. The second pass
231 * will wait. This is to prevent as much IO as possible in the affected region.
232 * The first pass will remove most pages, so the search cost of the second pass
233 * is low.
234 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700235 * We pass down the cache-hot hint to the page freeing code. Even if the
236 * mapping is large, it is probably the case that the final pages are the most
237 * recently touched, and freeing happens in ascending file offset order.
Lukas Czerner5a720392013-05-27 23:32:35 -0400238 *
239 * Note that since ->invalidatepage() accepts range to invalidate
240 * truncate_inode_pages_range is able to handle cases where lend + 1 is not
241 * page aligned properly.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700242 */
Hans Reiserd7339072006-01-06 00:10:36 -0800243void truncate_inode_pages_range(struct address_space *mapping,
244 loff_t lstart, loff_t lend)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700245{
Lukas Czerner5a720392013-05-27 23:32:35 -0400246 pgoff_t start; /* inclusive */
247 pgoff_t end; /* exclusive */
248 unsigned int partial_start; /* inclusive */
249 unsigned int partial_end; /* exclusive */
250 struct pagevec pvec;
Johannes Weiner0cd61442014-04-03 14:47:46 -0700251 pgoff_t indices[PAGEVEC_SIZE];
Lukas Czerner5a720392013-05-27 23:32:35 -0400252 pgoff_t index;
253 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700254
Dan Magenheimer31677602011-09-21 11:56:28 -0400255 cleancache_invalidate_inode(mapping);
Johannes Weiner91b0abe2014-04-03 14:47:49 -0700256 if (mapping->nrpages == 0 && mapping->nrshadows == 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700257 return;
258
Lukas Czerner5a720392013-05-27 23:32:35 -0400259 /* Offsets within partial pages */
260 partial_start = lstart & (PAGE_CACHE_SIZE - 1);
261 partial_end = (lend + 1) & (PAGE_CACHE_SIZE - 1);
262
263 /*
264 * 'start' and 'end' always covers the range of pages to be fully
265 * truncated. Partial pages are covered with 'partial_start' at the
266 * start of the range and 'partial_end' at the end of the range.
267 * Note that 'end' is exclusive while 'lend' is inclusive.
268 */
269 start = (lstart + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
270 if (lend == -1)
271 /*
272 * lend == -1 indicates end-of-file so we have to set 'end'
273 * to the highest possible pgoff_t and since the type is
274 * unsigned we're using -1.
275 */
276 end = -1;
277 else
278 end = (lend + 1) >> PAGE_CACHE_SHIFT;
Hans Reiserd7339072006-01-06 00:10:36 -0800279
Linus Torvalds1da177e2005-04-16 15:20:36 -0700280 pagevec_init(&pvec, 0);
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700281 index = start;
Johannes Weiner0cd61442014-04-03 14:47:46 -0700282 while (index < end && pagevec_lookup_entries(&pvec, mapping, index,
283 min(end - index, (pgoff_t)PAGEVEC_SIZE),
284 indices)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700285 for (i = 0; i < pagevec_count(&pvec); i++) {
286 struct page *page = pvec.pages[i];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700287
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700288 /* We rely upon deletion not changing page->index */
Johannes Weiner0cd61442014-04-03 14:47:46 -0700289 index = indices[i];
Lukas Czerner5a720392013-05-27 23:32:35 -0400290 if (index >= end)
Hans Reiserd7339072006-01-06 00:10:36 -0800291 break;
Hans Reiserd7339072006-01-06 00:10:36 -0800292
Johannes Weiner0cd61442014-04-03 14:47:46 -0700293 if (radix_tree_exceptional_entry(page)) {
294 clear_exceptional_entry(mapping, index, page);
295 continue;
296 }
297
Nick Piggin529ae9a2008-08-02 12:01:03 +0200298 if (!trylock_page(page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700299 continue;
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700300 WARN_ON(page->index != index);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700301 if (PageWriteback(page)) {
302 unlock_page(page);
303 continue;
304 }
Nick Piggin750b4982009-09-16 11:50:12 +0200305 truncate_inode_page(mapping, page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700306 unlock_page(page);
307 }
Johannes Weiner0cd61442014-04-03 14:47:46 -0700308 pagevec_remove_exceptionals(&pvec);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700309 pagevec_release(&pvec);
310 cond_resched();
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700311 index++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700312 }
313
Lukas Czerner5a720392013-05-27 23:32:35 -0400314 if (partial_start) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700315 struct page *page = find_lock_page(mapping, start - 1);
316 if (page) {
Lukas Czerner5a720392013-05-27 23:32:35 -0400317 unsigned int top = PAGE_CACHE_SIZE;
318 if (start > end) {
319 /* Truncation within a single page */
320 top = partial_end;
321 partial_end = 0;
322 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700323 wait_on_page_writeback(page);
Lukas Czerner5a720392013-05-27 23:32:35 -0400324 zero_user_segment(page, partial_start, top);
325 cleancache_invalidate_page(mapping, page);
326 if (page_has_private(page))
327 do_invalidatepage(page, partial_start,
328 top - partial_start);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700329 unlock_page(page);
330 page_cache_release(page);
331 }
332 }
Lukas Czerner5a720392013-05-27 23:32:35 -0400333 if (partial_end) {
334 struct page *page = find_lock_page(mapping, end);
335 if (page) {
336 wait_on_page_writeback(page);
337 zero_user_segment(page, 0, partial_end);
338 cleancache_invalidate_page(mapping, page);
339 if (page_has_private(page))
340 do_invalidatepage(page, 0,
341 partial_end);
342 unlock_page(page);
343 page_cache_release(page);
344 }
345 }
346 /*
347 * If the truncation happened within a single page no pages
348 * will be released, just zeroed, so we can bail out now.
349 */
350 if (start >= end)
351 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700352
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700353 index = start;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700354 for ( ; ; ) {
355 cond_resched();
Johannes Weiner0cd61442014-04-03 14:47:46 -0700356 if (!pagevec_lookup_entries(&pvec, mapping, index,
Hugh Dickins792ceae2014-07-23 14:00:15 -0700357 min(end - index, (pgoff_t)PAGEVEC_SIZE), indices)) {
358 /* If all gone from start onwards, we're done */
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700359 if (index == start)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700360 break;
Hugh Dickins792ceae2014-07-23 14:00:15 -0700361 /* Otherwise restart to make sure all gone */
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700362 index = start;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700363 continue;
364 }
Johannes Weiner0cd61442014-04-03 14:47:46 -0700365 if (index == start && indices[0] >= end) {
Hugh Dickins792ceae2014-07-23 14:00:15 -0700366 /* All gone out of hole to be punched, we're done */
Johannes Weiner0cd61442014-04-03 14:47:46 -0700367 pagevec_remove_exceptionals(&pvec);
Hans Reiserd7339072006-01-06 00:10:36 -0800368 pagevec_release(&pvec);
369 break;
370 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700371 for (i = 0; i < pagevec_count(&pvec); i++) {
372 struct page *page = pvec.pages[i];
373
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700374 /* We rely upon deletion not changing page->index */
Johannes Weiner0cd61442014-04-03 14:47:46 -0700375 index = indices[i];
Hugh Dickins792ceae2014-07-23 14:00:15 -0700376 if (index >= end) {
377 /* Restart punch to make sure all gone */
378 index = start - 1;
Hans Reiserd7339072006-01-06 00:10:36 -0800379 break;
Hugh Dickins792ceae2014-07-23 14:00:15 -0700380 }
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700381
Johannes Weiner0cd61442014-04-03 14:47:46 -0700382 if (radix_tree_exceptional_entry(page)) {
383 clear_exceptional_entry(mapping, index, page);
384 continue;
385 }
386
Linus Torvalds1da177e2005-04-16 15:20:36 -0700387 lock_page(page);
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700388 WARN_ON(page->index != index);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700389 wait_on_page_writeback(page);
Nick Piggin750b4982009-09-16 11:50:12 +0200390 truncate_inode_page(mapping, page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700391 unlock_page(page);
392 }
Johannes Weiner0cd61442014-04-03 14:47:46 -0700393 pagevec_remove_exceptionals(&pvec);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700394 pagevec_release(&pvec);
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700395 index++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700396 }
Dan Magenheimer31677602011-09-21 11:56:28 -0400397 cleancache_invalidate_inode(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700398}
Hans Reiserd7339072006-01-06 00:10:36 -0800399EXPORT_SYMBOL(truncate_inode_pages_range);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700400
Hans Reiserd7339072006-01-06 00:10:36 -0800401/**
402 * truncate_inode_pages - truncate *all* the pages from an offset
403 * @mapping: mapping to truncate
404 * @lstart: offset from which to truncate
405 *
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800406 * Called under (and serialised by) inode->i_mutex.
Jan Kara08142572011-06-27 16:18:10 -0700407 *
408 * Note: When this function returns, there can be a page in the process of
409 * deletion (inside __delete_from_page_cache()) in the specified range. Thus
410 * mapping->nrpages can be non-zero when this function returns even after
411 * truncation of the whole mapping.
Hans Reiserd7339072006-01-06 00:10:36 -0800412 */
413void truncate_inode_pages(struct address_space *mapping, loff_t lstart)
414{
415 truncate_inode_pages_range(mapping, lstart, (loff_t)-1);
416}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700417EXPORT_SYMBOL(truncate_inode_pages);
418
Mike Waychison28697352009-06-16 15:32:59 -0700419/**
Johannes Weiner91b0abe2014-04-03 14:47:49 -0700420 * truncate_inode_pages_final - truncate *all* pages before inode dies
421 * @mapping: mapping to truncate
422 *
423 * Called under (and serialized by) inode->i_mutex.
424 *
425 * Filesystems have to use this in the .evict_inode path to inform the
426 * VM that this is the final truncate and the inode is going away.
427 */
428void truncate_inode_pages_final(struct address_space *mapping)
429{
430 unsigned long nrshadows;
431 unsigned long nrpages;
432
433 /*
434 * Page reclaim can not participate in regular inode lifetime
435 * management (can't call iput()) and thus can race with the
436 * inode teardown. Tell it when the address space is exiting,
437 * so that it does not install eviction information after the
438 * final truncate has begun.
439 */
440 mapping_set_exiting(mapping);
441
442 /*
443 * When reclaim installs eviction entries, it increases
444 * nrshadows first, then decreases nrpages. Make sure we see
445 * this in the right order or we might miss an entry.
446 */
447 nrpages = mapping->nrpages;
448 smp_rmb();
449 nrshadows = mapping->nrshadows;
450
451 if (nrpages || nrshadows) {
452 /*
453 * As truncation uses a lockless tree lookup, cycle
454 * the tree lock to make sure any ongoing tree
455 * modification that does not see AS_EXITING is
456 * completed before starting the final truncate.
457 */
458 spin_lock_irq(&mapping->tree_lock);
459 spin_unlock_irq(&mapping->tree_lock);
460
461 truncate_inode_pages(mapping, 0);
462 }
463}
464EXPORT_SYMBOL(truncate_inode_pages_final);
465
466/**
Mike Waychison28697352009-06-16 15:32:59 -0700467 * invalidate_mapping_pages - Invalidate all the unlocked pages of one inode
468 * @mapping: the address_space which holds the pages to invalidate
469 * @start: the offset 'from' which to invalidate
470 * @end: the offset 'to' which to invalidate (inclusive)
471 *
472 * This function only removes the unlocked pages, if you want to
473 * remove all the pages of one inode, you must call truncate_inode_pages.
474 *
475 * invalidate_mapping_pages() will not block on IO activity. It will not
476 * invalidate pages which are dirty, locked, under writeback or mapped into
477 * pagetables.
478 */
479unsigned long invalidate_mapping_pages(struct address_space *mapping,
Minchan Kim31560182011-03-22 16:32:52 -0700480 pgoff_t start, pgoff_t end)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700481{
Johannes Weiner0cd61442014-04-03 14:47:46 -0700482 pgoff_t indices[PAGEVEC_SIZE];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700483 struct pagevec pvec;
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700484 pgoff_t index = start;
Minchan Kim31560182011-03-22 16:32:52 -0700485 unsigned long ret;
486 unsigned long count = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700487 int i;
488
489 pagevec_init(&pvec, 0);
Johannes Weiner0cd61442014-04-03 14:47:46 -0700490 while (index <= end && pagevec_lookup_entries(&pvec, mapping, index,
491 min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1,
492 indices)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700493 for (i = 0; i < pagevec_count(&pvec); i++) {
494 struct page *page = pvec.pages[i];
495
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700496 /* We rely upon deletion not changing page->index */
Johannes Weiner0cd61442014-04-03 14:47:46 -0700497 index = indices[i];
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700498 if (index > end)
499 break;
NeilBrowne0f23602006-06-23 02:05:48 -0700500
Johannes Weiner0cd61442014-04-03 14:47:46 -0700501 if (radix_tree_exceptional_entry(page)) {
502 clear_exceptional_entry(mapping, index, page);
503 continue;
504 }
505
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700506 if (!trylock_page(page))
507 continue;
508 WARN_ON(page->index != index);
Minchan Kim31560182011-03-22 16:32:52 -0700509 ret = invalidate_inode_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700510 unlock_page(page);
Minchan Kim31560182011-03-22 16:32:52 -0700511 /*
512 * Invalidation is a hint that the page is no longer
513 * of interest and try to speed up its reclaim.
514 */
515 if (!ret)
516 deactivate_page(page);
517 count += ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700518 }
Johannes Weiner0cd61442014-04-03 14:47:46 -0700519 pagevec_remove_exceptionals(&pvec);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700520 pagevec_release(&pvec);
Mike Waychison28697352009-06-16 15:32:59 -0700521 cond_resched();
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700522 index++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700523 }
Minchan Kim31560182011-03-22 16:32:52 -0700524 return count;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700525}
Anton Altaparmakov54bc4852007-02-10 01:45:38 -0800526EXPORT_SYMBOL(invalidate_mapping_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700527
Andrew Mortonbd4c8ce2006-09-30 23:29:29 -0700528/*
529 * This is like invalidate_complete_page(), except it ignores the page's
530 * refcount. We do this because invalidate_inode_pages2() needs stronger
531 * invalidation guarantees, and cannot afford to leave pages behind because
Anderson Briglia2706a1b2007-07-15 23:38:09 -0700532 * shrink_page_list() has a temp ref on them, or because they're transiently
533 * sitting in the lru_cache_add() pagevecs.
Andrew Mortonbd4c8ce2006-09-30 23:29:29 -0700534 */
535static int
536invalidate_complete_page2(struct address_space *mapping, struct page *page)
537{
538 if (page->mapping != mapping)
539 return 0;
540
David Howells266cf652009-04-03 16:42:36 +0100541 if (page_has_private(page) && !try_to_release_page(page, GFP_KERNEL))
Andrew Mortonbd4c8ce2006-09-30 23:29:29 -0700542 return 0;
543
Nick Piggin19fd6232008-07-25 19:45:32 -0700544 spin_lock_irq(&mapping->tree_lock);
Andrew Mortonbd4c8ce2006-09-30 23:29:29 -0700545 if (PageDirty(page))
546 goto failed;
547
David Howells266cf652009-04-03 16:42:36 +0100548 BUG_ON(page_has_private(page));
Johannes Weiner91b0abe2014-04-03 14:47:49 -0700549 __delete_from_page_cache(page, NULL);
Nick Piggin19fd6232008-07-25 19:45:32 -0700550 spin_unlock_irq(&mapping->tree_lock);
Linus Torvalds6072d132010-12-01 13:35:19 -0500551
552 if (mapping->a_ops->freepage)
553 mapping->a_ops->freepage(page);
554
Andrew Mortonbd4c8ce2006-09-30 23:29:29 -0700555 page_cache_release(page); /* pagecache ref */
556 return 1;
557failed:
Nick Piggin19fd6232008-07-25 19:45:32 -0700558 spin_unlock_irq(&mapping->tree_lock);
Andrew Mortonbd4c8ce2006-09-30 23:29:29 -0700559 return 0;
560}
561
Trond Myklebuste3db7692007-01-10 23:15:39 -0800562static int do_launder_page(struct address_space *mapping, struct page *page)
563{
564 if (!PageDirty(page))
565 return 0;
566 if (page->mapping != mapping || mapping->a_ops->launder_page == NULL)
567 return 0;
568 return mapping->a_ops->launder_page(page);
569}
570
Linus Torvalds1da177e2005-04-16 15:20:36 -0700571/**
572 * invalidate_inode_pages2_range - remove range of pages from an address_space
Martin Waitz67be2dd2005-05-01 08:59:26 -0700573 * @mapping: the address_space
Linus Torvalds1da177e2005-04-16 15:20:36 -0700574 * @start: the page offset 'from' which to invalidate
575 * @end: the page offset 'to' which to invalidate (inclusive)
576 *
577 * Any pages which are found to be mapped into pagetables are unmapped prior to
578 * invalidation.
579 *
Hisashi Hifumi6ccfa802008-09-02 14:35:40 -0700580 * Returns -EBUSY if any pages could not be invalidated.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700581 */
582int invalidate_inode_pages2_range(struct address_space *mapping,
583 pgoff_t start, pgoff_t end)
584{
Johannes Weiner0cd61442014-04-03 14:47:46 -0700585 pgoff_t indices[PAGEVEC_SIZE];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700586 struct pagevec pvec;
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700587 pgoff_t index;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700588 int i;
589 int ret = 0;
Hisashi Hifumi0dd13342008-04-28 02:12:08 -0700590 int ret2 = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700591 int did_range_unmap = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700592
Dan Magenheimer31677602011-09-21 11:56:28 -0400593 cleancache_invalidate_inode(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700594 pagevec_init(&pvec, 0);
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700595 index = start;
Johannes Weiner0cd61442014-04-03 14:47:46 -0700596 while (index <= end && pagevec_lookup_entries(&pvec, mapping, index,
597 min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1,
598 indices)) {
Trond Myklebust7b965e02007-02-28 20:13:55 -0800599 for (i = 0; i < pagevec_count(&pvec); i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700600 struct page *page = pvec.pages[i];
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700601
602 /* We rely upon deletion not changing page->index */
Johannes Weiner0cd61442014-04-03 14:47:46 -0700603 index = indices[i];
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700604 if (index > end)
605 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700606
Johannes Weiner0cd61442014-04-03 14:47:46 -0700607 if (radix_tree_exceptional_entry(page)) {
608 clear_exceptional_entry(mapping, index, page);
609 continue;
610 }
611
Linus Torvalds1da177e2005-04-16 15:20:36 -0700612 lock_page(page);
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700613 WARN_ON(page->index != index);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700614 if (page->mapping != mapping) {
615 unlock_page(page);
616 continue;
617 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700618 wait_on_page_writeback(page);
Nick Piggind00806b2007-07-19 01:46:57 -0700619 if (page_mapped(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700620 if (!did_range_unmap) {
621 /*
622 * Zap the rest of the file in one hit.
623 */
624 unmap_mapping_range(mapping,
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700625 (loff_t)index << PAGE_CACHE_SHIFT,
626 (loff_t)(1 + end - index)
627 << PAGE_CACHE_SHIFT,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700628 0);
629 did_range_unmap = 1;
630 } else {
631 /*
632 * Just zap this page
633 */
634 unmap_mapping_range(mapping,
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700635 (loff_t)index << PAGE_CACHE_SHIFT,
636 PAGE_CACHE_SIZE, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700637 }
638 }
Nick Piggind00806b2007-07-19 01:46:57 -0700639 BUG_ON(page_mapped(page));
Hisashi Hifumi0dd13342008-04-28 02:12:08 -0700640 ret2 = do_launder_page(mapping, page);
641 if (ret2 == 0) {
642 if (!invalidate_complete_page2(mapping, page))
Hisashi Hifumi6ccfa802008-09-02 14:35:40 -0700643 ret2 = -EBUSY;
Hisashi Hifumi0dd13342008-04-28 02:12:08 -0700644 }
645 if (ret2 < 0)
646 ret = ret2;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700647 unlock_page(page);
648 }
Johannes Weiner0cd61442014-04-03 14:47:46 -0700649 pagevec_remove_exceptionals(&pvec);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700650 pagevec_release(&pvec);
651 cond_resched();
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700652 index++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700653 }
Dan Magenheimer31677602011-09-21 11:56:28 -0400654 cleancache_invalidate_inode(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700655 return ret;
656}
657EXPORT_SYMBOL_GPL(invalidate_inode_pages2_range);
658
659/**
660 * invalidate_inode_pages2 - remove all pages from an address_space
Martin Waitz67be2dd2005-05-01 08:59:26 -0700661 * @mapping: the address_space
Linus Torvalds1da177e2005-04-16 15:20:36 -0700662 *
663 * Any pages which are found to be mapped into pagetables are unmapped prior to
664 * invalidation.
665 *
Peng Taoe9de25d2009-10-19 14:48:13 +0800666 * Returns -EBUSY if any pages could not be invalidated.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700667 */
668int invalidate_inode_pages2(struct address_space *mapping)
669{
670 return invalidate_inode_pages2_range(mapping, 0, -1);
671}
672EXPORT_SYMBOL_GPL(invalidate_inode_pages2);
npiggin@suse.de25d9e2d2009-08-21 02:35:05 +1000673
674/**
675 * truncate_pagecache - unmap and remove pagecache that has been truncated
676 * @inode: inode
Hugh Dickins8a549be2011-07-25 17:12:24 -0700677 * @newsize: new file size
npiggin@suse.de25d9e2d2009-08-21 02:35:05 +1000678 *
679 * inode's new i_size must already be written before truncate_pagecache
680 * is called.
681 *
682 * This function should typically be called before the filesystem
683 * releases resources associated with the freed range (eg. deallocates
684 * blocks). This way, pagecache will always stay logically coherent
685 * with on-disk format, and the filesystem would not have to deal with
686 * situations such as writepage being called for a page that has already
687 * had its underlying blocks deallocated.
688 */
Kirill A. Shutemov7caef262013-09-12 15:13:56 -0700689void truncate_pagecache(struct inode *inode, loff_t newsize)
npiggin@suse.de25d9e2d2009-08-21 02:35:05 +1000690{
OGAWA Hirofumicedabed2010-01-13 21:14:09 +0900691 struct address_space *mapping = inode->i_mapping;
Hugh Dickins8a549be2011-07-25 17:12:24 -0700692 loff_t holebegin = round_up(newsize, PAGE_SIZE);
npiggin@suse.de25d9e2d2009-08-21 02:35:05 +1000693
OGAWA Hirofumicedabed2010-01-13 21:14:09 +0900694 /*
695 * unmap_mapping_range is called twice, first simply for
696 * efficiency so that truncate_inode_pages does fewer
697 * single-page unmaps. However after this first call, and
698 * before truncate_inode_pages finishes, it is possible for
699 * private pages to be COWed, which remain after
700 * truncate_inode_pages finishes, hence the second
701 * unmap_mapping_range call must be made for correctness.
702 */
Hugh Dickins8a549be2011-07-25 17:12:24 -0700703 unmap_mapping_range(mapping, holebegin, 0, 1);
704 truncate_inode_pages(mapping, newsize);
705 unmap_mapping_range(mapping, holebegin, 0, 1);
npiggin@suse.de25d9e2d2009-08-21 02:35:05 +1000706}
707EXPORT_SYMBOL(truncate_pagecache);
708
709/**
Christoph Hellwig2c27c652010-06-04 11:30:04 +0200710 * truncate_setsize - update inode and pagecache for a new file size
711 * @inode: inode
712 * @newsize: new file size
713 *
Jan Kara382e27d2011-01-20 14:44:26 -0800714 * truncate_setsize updates i_size and performs pagecache truncation (if
715 * necessary) to @newsize. It will be typically be called from the filesystem's
716 * setattr function when ATTR_SIZE is passed in.
Christoph Hellwig2c27c652010-06-04 11:30:04 +0200717 *
Jan Kara77783d02014-11-07 08:29:25 +1100718 * Must be called with a lock serializing truncates and writes (generally
719 * i_mutex but e.g. xfs uses a different lock) and before all filesystem
720 * specific block truncation has been performed.
Christoph Hellwig2c27c652010-06-04 11:30:04 +0200721 */
722void truncate_setsize(struct inode *inode, loff_t newsize)
723{
Jan Kara90a80202014-10-01 21:49:18 -0400724 loff_t oldsize = inode->i_size;
725
Christoph Hellwig2c27c652010-06-04 11:30:04 +0200726 i_size_write(inode, newsize);
Jan Kara90a80202014-10-01 21:49:18 -0400727 if (newsize > oldsize)
728 pagecache_isize_extended(inode, oldsize, newsize);
Kirill A. Shutemov7caef262013-09-12 15:13:56 -0700729 truncate_pagecache(inode, newsize);
Christoph Hellwig2c27c652010-06-04 11:30:04 +0200730}
731EXPORT_SYMBOL(truncate_setsize);
732
733/**
Jan Kara90a80202014-10-01 21:49:18 -0400734 * pagecache_isize_extended - update pagecache after extension of i_size
735 * @inode: inode for which i_size was extended
736 * @from: original inode size
737 * @to: new inode size
738 *
739 * Handle extension of inode size either caused by extending truncate or by
740 * write starting after current i_size. We mark the page straddling current
741 * i_size RO so that page_mkwrite() is called on the nearest write access to
742 * the page. This way filesystem can be sure that page_mkwrite() is called on
743 * the page before user writes to the page via mmap after the i_size has been
744 * changed.
745 *
746 * The function must be called after i_size is updated so that page fault
747 * coming after we unlock the page will already see the new i_size.
748 * The function must be called while we still hold i_mutex - this not only
749 * makes sure i_size is stable but also that userspace cannot observe new
750 * i_size value before we are prepared to store mmap writes at new inode size.
751 */
752void pagecache_isize_extended(struct inode *inode, loff_t from, loff_t to)
753{
754 int bsize = 1 << inode->i_blkbits;
755 loff_t rounded_from;
756 struct page *page;
757 pgoff_t index;
758
Jan Kara90a80202014-10-01 21:49:18 -0400759 WARN_ON(to > inode->i_size);
760
761 if (from >= to || bsize == PAGE_CACHE_SIZE)
762 return;
763 /* Page straddling @from will not have any hole block created? */
764 rounded_from = round_up(from, bsize);
765 if (to <= rounded_from || !(rounded_from & (PAGE_CACHE_SIZE - 1)))
766 return;
767
768 index = from >> PAGE_CACHE_SHIFT;
769 page = find_lock_page(inode->i_mapping, index);
770 /* Page not cached? Nothing to do */
771 if (!page)
772 return;
773 /*
774 * See clear_page_dirty_for_io() for details why set_page_dirty()
775 * is needed.
776 */
777 if (page_mkclean(page))
778 set_page_dirty(page);
779 unlock_page(page);
780 page_cache_release(page);
781}
782EXPORT_SYMBOL(pagecache_isize_extended);
783
784/**
Hugh Dickins623e3db2012-03-28 14:42:40 -0700785 * truncate_pagecache_range - unmap and remove pagecache that is hole-punched
786 * @inode: inode
787 * @lstart: offset of beginning of hole
788 * @lend: offset of last byte of hole
789 *
790 * This function should typically be called before the filesystem
791 * releases resources associated with the freed range (eg. deallocates
792 * blocks). This way, pagecache will always stay logically coherent
793 * with on-disk format, and the filesystem would not have to deal with
794 * situations such as writepage being called for a page that has already
795 * had its underlying blocks deallocated.
796 */
797void truncate_pagecache_range(struct inode *inode, loff_t lstart, loff_t lend)
798{
799 struct address_space *mapping = inode->i_mapping;
800 loff_t unmap_start = round_up(lstart, PAGE_SIZE);
801 loff_t unmap_end = round_down(1 + lend, PAGE_SIZE) - 1;
802 /*
803 * This rounding is currently just for example: unmap_mapping_range
804 * expands its hole outwards, whereas we want it to contract the hole
805 * inwards. However, existing callers of truncate_pagecache_range are
Lukas Czerner5a720392013-05-27 23:32:35 -0400806 * doing their own page rounding first. Note that unmap_mapping_range
807 * allows holelen 0 for all, and we allow lend -1 for end of file.
Hugh Dickins623e3db2012-03-28 14:42:40 -0700808 */
809
810 /*
811 * Unlike in truncate_pagecache, unmap_mapping_range is called only
812 * once (before truncating pagecache), and without "even_cows" flag:
813 * hole-punching should not remove private COWed pages from the hole.
814 */
815 if ((u64)unmap_end > (u64)unmap_start)
816 unmap_mapping_range(mapping, unmap_start,
817 1 + unmap_end - unmap_start, 0);
818 truncate_inode_pages_range(mapping, lstart, lend);
819}
820EXPORT_SYMBOL(truncate_pagecache_range);