blob: fd97f1dbce290f39e1d0d0367006df954f20e8f1 [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>
Ross Zwislerf9fe48b2016-01-22 15:10:40 -080012#include <linux/dax.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090013#include <linux/gfp.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070014#include <linux/mm.h>
Nick Piggin0fd0e6b2006-09-27 01:50:02 -070015#include <linux/swap.h>
Paul Gortmakerb95f1b312011-10-16 02:01:52 -040016#include <linux/export.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070017#include <linux/pagemap.h>
Nate Diller01f27052007-05-09 02:35:07 -070018#include <linux/highmem.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070019#include <linux/pagevec.h>
Andrew Mortone08748ce2006-12-10 02:19:31 -080020#include <linux/task_io_accounting_ops.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070021#include <linux/buffer_head.h> /* grr. try_to_release_page,
Jan Karaaaa40592005-10-30 15:00:16 -080022 do_invalidatepage */
Dan Magenheimerc515e1f2011-05-26 10:01:43 -060023#include <linux/cleancache.h>
Jan Kara90a80202014-10-01 21:49:18 -040024#include <linux/rmap.h>
Rik van Rielba470de2008-10-18 20:26:50 -070025#include "internal.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070026
Johannes Weiner0cd61442014-04-03 14:47:46 -070027static void clear_exceptional_entry(struct address_space *mapping,
28 pgoff_t index, void *entry)
29{
Johannes Weiner449dd692014-04-03 14:47:56 -070030 struct radix_tree_node *node;
31 void **slot;
32
Johannes Weiner0cd61442014-04-03 14:47:46 -070033 /* Handled by shmem itself */
34 if (shmem_mapping(mapping))
35 return;
36
Ross Zwislerf9fe48b2016-01-22 15:10:40 -080037 if (dax_mapping(mapping)) {
Jan Karaac401cc2016-05-12 18:29:18 +020038 dax_delete_mapping_entry(mapping, index);
39 return;
Ross Zwislerf9fe48b2016-01-22 15:10:40 -080040 }
Jan Karaac401cc2016-05-12 18:29:18 +020041 spin_lock_irq(&mapping->tree_lock);
42 /*
43 * Regular page slots are stabilized by the page lock even
44 * without the tree itself locked. These unlocked entries
45 * need verification under the tree lock.
46 */
Johannes Weiner14b46872016-12-12 16:43:52 -080047 if (!__radix_tree_lookup(&mapping->page_tree, index, &node, &slot))
Jan Karaac401cc2016-05-12 18:29:18 +020048 goto unlock;
49 if (*slot != entry)
50 goto unlock;
Johannes Weiner14b46872016-12-12 16:43:52 -080051 __radix_tree_replace(&mapping->page_tree, node, slot, NULL,
52 workingset_update_node, mapping);
Jan Karaac401cc2016-05-12 18:29:18 +020053 mapping->nrexceptional--;
Johannes Weiner449dd692014-04-03 14:47:56 -070054unlock:
Johannes Weiner0cd61442014-04-03 14:47:46 -070055 spin_unlock_irq(&mapping->tree_lock);
56}
Linus Torvalds1da177e2005-04-16 15:20:36 -070057
David Howellscf9a2ae2006-08-29 19:05:54 +010058/**
Fengguang Wu28bc44d2008-02-03 18:04:10 +020059 * do_invalidatepage - invalidate part or all of a page
David Howellscf9a2ae2006-08-29 19:05:54 +010060 * @page: the page which is affected
Lukas Czernerd47992f2013-05-21 23:17:23 -040061 * @offset: start of the range to invalidate
62 * @length: length of the range to invalidate
David Howellscf9a2ae2006-08-29 19:05:54 +010063 *
64 * do_invalidatepage() is called when all or part of the page has become
65 * invalidated by a truncate operation.
66 *
67 * do_invalidatepage() does not have to release all buffers, but it must
68 * ensure that no dirty buffer is left outside @offset and that no I/O
69 * is underway against any of the blocks which are outside the truncation
70 * point. Because the caller is about to free (and possibly reuse) those
71 * blocks on-disk.
72 */
Lukas Czernerd47992f2013-05-21 23:17:23 -040073void do_invalidatepage(struct page *page, unsigned int offset,
74 unsigned int length)
David Howellscf9a2ae2006-08-29 19:05:54 +010075{
Lukas Czernerd47992f2013-05-21 23:17:23 -040076 void (*invalidatepage)(struct page *, unsigned int, unsigned int);
77
David Howellscf9a2ae2006-08-29 19:05:54 +010078 invalidatepage = page->mapping->a_ops->invalidatepage;
David Howells93614012006-09-30 20:45:40 +020079#ifdef CONFIG_BLOCK
David Howellscf9a2ae2006-08-29 19:05:54 +010080 if (!invalidatepage)
81 invalidatepage = block_invalidatepage;
David Howells93614012006-09-30 20:45:40 +020082#endif
David Howellscf9a2ae2006-08-29 19:05:54 +010083 if (invalidatepage)
Lukas Czernerd47992f2013-05-21 23:17:23 -040084 (*invalidatepage)(page, offset, length);
David Howellscf9a2ae2006-08-29 19:05:54 +010085}
86
Linus Torvaldsecdfc972007-01-26 12:47:06 -080087/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070088 * If truncate cannot remove the fs-private metadata from the page, the page
Shaohua Li62e1c552008-02-04 22:29:33 -080089 * becomes orphaned. It will be left on the LRU and may even be mapped into
Nick Piggin54cb8822007-07-19 01:46:59 -070090 * user pagetables if we're racing with filemap_fault().
Linus Torvalds1da177e2005-04-16 15:20:36 -070091 *
92 * We need to bale out if page->mapping is no longer equal to the original
93 * mapping. This happens a) when the VM reclaimed the page while we waited on
Andrew Mortonfc0ecff2007-02-10 01:45:39 -080094 * its lock, b) when a concurrent invalidate_mapping_pages got there first and
Linus Torvalds1da177e2005-04-16 15:20:36 -070095 * c) when tmpfs swizzles a page between a tmpfs inode and swapper_space.
96 */
Nick Piggin750b4982009-09-16 11:50:12 +020097static int
Linus Torvalds1da177e2005-04-16 15:20:36 -070098truncate_complete_page(struct address_space *mapping, struct page *page)
99{
100 if (page->mapping != mapping)
Nick Piggin750b4982009-09-16 11:50:12 +0200101 return -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700102
David Howells266cf652009-04-03 16:42:36 +0100103 if (page_has_private(page))
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300104 do_invalidatepage(page, 0, PAGE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700105
Konstantin Khlebnikovb9ea2512015-04-14 15:45:27 -0700106 /*
107 * Some filesystems seem to re-dirty the page even after
108 * the VM has canceled the dirty bit (eg ext3 journaling).
109 * Hence dirty accounting check is placed after invalidation.
110 */
Tejun Heo11f81be2015-05-22 17:13:15 -0400111 cancel_dirty_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700112 ClearPageMappedToDisk(page);
Minchan Kim5adc7b52011-03-22 16:32:41 -0700113 delete_from_page_cache(page);
Nick Piggin750b4982009-09-16 11:50:12 +0200114 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700115}
116
117/*
Andrew Mortonfc0ecff2007-02-10 01:45:39 -0800118 * This is for invalidate_mapping_pages(). That function can be called at
Linus Torvalds1da177e2005-04-16 15:20:36 -0700119 * any time, and is not supposed to throw away dirty pages. But pages can
Nick Piggin0fd0e6b2006-09-27 01:50:02 -0700120 * be marked dirty at any time too, so use remove_mapping which safely
121 * discards clean, unused pages.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700122 *
123 * Returns non-zero if the page was successfully invalidated.
124 */
125static int
126invalidate_complete_page(struct address_space *mapping, struct page *page)
127{
Nick Piggin0fd0e6b2006-09-27 01:50:02 -0700128 int ret;
129
Linus Torvalds1da177e2005-04-16 15:20:36 -0700130 if (page->mapping != mapping)
131 return 0;
132
David Howells266cf652009-04-03 16:42:36 +0100133 if (page_has_private(page) && !try_to_release_page(page, 0))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700134 return 0;
135
Nick Piggin0fd0e6b2006-09-27 01:50:02 -0700136 ret = remove_mapping(mapping, page);
Nick Piggin0fd0e6b2006-09-27 01:50:02 -0700137
138 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700139}
140
Nick Piggin750b4982009-09-16 11:50:12 +0200141int truncate_inode_page(struct address_space *mapping, struct page *page)
142{
Kirill A. Shutemovfc127da2016-07-26 15:26:07 -0700143 loff_t holelen;
144 VM_BUG_ON_PAGE(PageTail(page), page);
145
146 holelen = PageTransHuge(page) ? HPAGE_PMD_SIZE : PAGE_SIZE;
Nick Piggin750b4982009-09-16 11:50:12 +0200147 if (page_mapped(page)) {
148 unmap_mapping_range(mapping,
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300149 (loff_t)page->index << PAGE_SHIFT,
Kirill A. Shutemovfc127da2016-07-26 15:26:07 -0700150 holelen, 0);
Nick Piggin750b4982009-09-16 11:50:12 +0200151 }
152 return truncate_complete_page(mapping, page);
153}
154
Wu Fengguang83f78662009-09-16 11:50:13 +0200155/*
Andi Kleen25718732009-09-16 11:50:13 +0200156 * Used to get rid of pages on hardware memory corruption.
157 */
158int generic_error_remove_page(struct address_space *mapping, struct page *page)
159{
160 if (!mapping)
161 return -EINVAL;
162 /*
163 * Only punch for normal data pages for now.
164 * Handling other types like directories would need more auditing.
165 */
166 if (!S_ISREG(mapping->host->i_mode))
167 return -EIO;
168 return truncate_inode_page(mapping, page);
169}
170EXPORT_SYMBOL(generic_error_remove_page);
171
172/*
Wu Fengguang83f78662009-09-16 11:50:13 +0200173 * Safely invalidate one page from its pagecache mapping.
174 * It only drops clean, unused pages. The page must be locked.
175 *
176 * Returns 1 if the page is successfully invalidated, otherwise 0.
177 */
178int invalidate_inode_page(struct page *page)
179{
180 struct address_space *mapping = page_mapping(page);
181 if (!mapping)
182 return 0;
183 if (PageDirty(page) || PageWriteback(page))
184 return 0;
185 if (page_mapped(page))
186 return 0;
187 return invalidate_complete_page(mapping, page);
188}
189
Linus Torvalds1da177e2005-04-16 15:20:36 -0700190/**
Liu Bo73c1e202012-02-21 10:57:20 +0800191 * truncate_inode_pages_range - truncate range of pages specified by start & end byte offsets
Linus Torvalds1da177e2005-04-16 15:20:36 -0700192 * @mapping: mapping to truncate
193 * @lstart: offset from which to truncate
Lukas Czerner5a720392013-05-27 23:32:35 -0400194 * @lend: offset to which to truncate (inclusive)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700195 *
Hans Reiserd7339072006-01-06 00:10:36 -0800196 * Truncate the page cache, removing the pages that are between
Lukas Czerner5a720392013-05-27 23:32:35 -0400197 * specified offsets (and zeroing out partial pages
198 * if lstart or lend + 1 is not page aligned).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700199 *
200 * Truncate takes two passes - the first pass is nonblocking. It will not
201 * block on page locks and it will not block on writeback. The second pass
202 * will wait. This is to prevent as much IO as possible in the affected region.
203 * The first pass will remove most pages, so the search cost of the second pass
204 * is low.
205 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700206 * We pass down the cache-hot hint to the page freeing code. Even if the
207 * mapping is large, it is probably the case that the final pages are the most
208 * recently touched, and freeing happens in ascending file offset order.
Lukas Czerner5a720392013-05-27 23:32:35 -0400209 *
210 * Note that since ->invalidatepage() accepts range to invalidate
211 * truncate_inode_pages_range is able to handle cases where lend + 1 is not
212 * page aligned properly.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700213 */
Hans Reiserd7339072006-01-06 00:10:36 -0800214void truncate_inode_pages_range(struct address_space *mapping,
215 loff_t lstart, loff_t lend)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700216{
Lukas Czerner5a720392013-05-27 23:32:35 -0400217 pgoff_t start; /* inclusive */
218 pgoff_t end; /* exclusive */
219 unsigned int partial_start; /* inclusive */
220 unsigned int partial_end; /* exclusive */
221 struct pagevec pvec;
Johannes Weiner0cd61442014-04-03 14:47:46 -0700222 pgoff_t indices[PAGEVEC_SIZE];
Lukas Czerner5a720392013-05-27 23:32:35 -0400223 pgoff_t index;
224 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700225
Dan Magenheimer31677602011-09-21 11:56:28 -0400226 cleancache_invalidate_inode(mapping);
Ross Zwislerf9fe48b2016-01-22 15:10:40 -0800227 if (mapping->nrpages == 0 && mapping->nrexceptional == 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700228 return;
229
Lukas Czerner5a720392013-05-27 23:32:35 -0400230 /* Offsets within partial pages */
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300231 partial_start = lstart & (PAGE_SIZE - 1);
232 partial_end = (lend + 1) & (PAGE_SIZE - 1);
Lukas Czerner5a720392013-05-27 23:32:35 -0400233
234 /*
235 * 'start' and 'end' always covers the range of pages to be fully
236 * truncated. Partial pages are covered with 'partial_start' at the
237 * start of the range and 'partial_end' at the end of the range.
238 * Note that 'end' is exclusive while 'lend' is inclusive.
239 */
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300240 start = (lstart + PAGE_SIZE - 1) >> PAGE_SHIFT;
Lukas Czerner5a720392013-05-27 23:32:35 -0400241 if (lend == -1)
242 /*
243 * lend == -1 indicates end-of-file so we have to set 'end'
244 * to the highest possible pgoff_t and since the type is
245 * unsigned we're using -1.
246 */
247 end = -1;
248 else
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300249 end = (lend + 1) >> PAGE_SHIFT;
Hans Reiserd7339072006-01-06 00:10:36 -0800250
Linus Torvalds1da177e2005-04-16 15:20:36 -0700251 pagevec_init(&pvec, 0);
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700252 index = start;
Johannes Weiner0cd61442014-04-03 14:47:46 -0700253 while (index < end && pagevec_lookup_entries(&pvec, mapping, index,
254 min(end - index, (pgoff_t)PAGEVEC_SIZE),
255 indices)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700256 for (i = 0; i < pagevec_count(&pvec); i++) {
257 struct page *page = pvec.pages[i];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700258
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700259 /* We rely upon deletion not changing page->index */
Johannes Weiner0cd61442014-04-03 14:47:46 -0700260 index = indices[i];
Lukas Czerner5a720392013-05-27 23:32:35 -0400261 if (index >= end)
Hans Reiserd7339072006-01-06 00:10:36 -0800262 break;
Hans Reiserd7339072006-01-06 00:10:36 -0800263
Johannes Weiner0cd61442014-04-03 14:47:46 -0700264 if (radix_tree_exceptional_entry(page)) {
265 clear_exceptional_entry(mapping, index, page);
266 continue;
267 }
268
Nick Piggin529ae9a2008-08-02 12:01:03 +0200269 if (!trylock_page(page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700270 continue;
Kirill A. Shutemov5cbc1982016-11-30 15:54:19 -0800271 WARN_ON(page_to_index(page) != index);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700272 if (PageWriteback(page)) {
273 unlock_page(page);
274 continue;
275 }
Nick Piggin750b4982009-09-16 11:50:12 +0200276 truncate_inode_page(mapping, page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700277 unlock_page(page);
278 }
Johannes Weiner0cd61442014-04-03 14:47:46 -0700279 pagevec_remove_exceptionals(&pvec);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700280 pagevec_release(&pvec);
281 cond_resched();
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700282 index++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700283 }
284
Lukas Czerner5a720392013-05-27 23:32:35 -0400285 if (partial_start) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700286 struct page *page = find_lock_page(mapping, start - 1);
287 if (page) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300288 unsigned int top = PAGE_SIZE;
Lukas Czerner5a720392013-05-27 23:32:35 -0400289 if (start > end) {
290 /* Truncation within a single page */
291 top = partial_end;
292 partial_end = 0;
293 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700294 wait_on_page_writeback(page);
Lukas Czerner5a720392013-05-27 23:32:35 -0400295 zero_user_segment(page, partial_start, top);
296 cleancache_invalidate_page(mapping, page);
297 if (page_has_private(page))
298 do_invalidatepage(page, partial_start,
299 top - partial_start);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700300 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300301 put_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700302 }
303 }
Lukas Czerner5a720392013-05-27 23:32:35 -0400304 if (partial_end) {
305 struct page *page = find_lock_page(mapping, end);
306 if (page) {
307 wait_on_page_writeback(page);
308 zero_user_segment(page, 0, partial_end);
309 cleancache_invalidate_page(mapping, page);
310 if (page_has_private(page))
311 do_invalidatepage(page, 0,
312 partial_end);
313 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300314 put_page(page);
Lukas Czerner5a720392013-05-27 23:32:35 -0400315 }
316 }
317 /*
318 * If the truncation happened within a single page no pages
319 * will be released, just zeroed, so we can bail out now.
320 */
321 if (start >= end)
322 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700323
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700324 index = start;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700325 for ( ; ; ) {
326 cond_resched();
Johannes Weiner0cd61442014-04-03 14:47:46 -0700327 if (!pagevec_lookup_entries(&pvec, mapping, index,
Hugh Dickins792ceae2014-07-23 14:00:15 -0700328 min(end - index, (pgoff_t)PAGEVEC_SIZE), indices)) {
329 /* If all gone from start onwards, we're done */
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700330 if (index == start)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700331 break;
Hugh Dickins792ceae2014-07-23 14:00:15 -0700332 /* Otherwise restart to make sure all gone */
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700333 index = start;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700334 continue;
335 }
Johannes Weiner0cd61442014-04-03 14:47:46 -0700336 if (index == start && indices[0] >= end) {
Hugh Dickins792ceae2014-07-23 14:00:15 -0700337 /* All gone out of hole to be punched, we're done */
Johannes Weiner0cd61442014-04-03 14:47:46 -0700338 pagevec_remove_exceptionals(&pvec);
Hans Reiserd7339072006-01-06 00:10:36 -0800339 pagevec_release(&pvec);
340 break;
341 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700342 for (i = 0; i < pagevec_count(&pvec); i++) {
343 struct page *page = pvec.pages[i];
344
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700345 /* We rely upon deletion not changing page->index */
Johannes Weiner0cd61442014-04-03 14:47:46 -0700346 index = indices[i];
Hugh Dickins792ceae2014-07-23 14:00:15 -0700347 if (index >= end) {
348 /* Restart punch to make sure all gone */
349 index = start - 1;
Hans Reiserd7339072006-01-06 00:10:36 -0800350 break;
Hugh Dickins792ceae2014-07-23 14:00:15 -0700351 }
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700352
Johannes Weiner0cd61442014-04-03 14:47:46 -0700353 if (radix_tree_exceptional_entry(page)) {
354 clear_exceptional_entry(mapping, index, page);
355 continue;
356 }
357
Linus Torvalds1da177e2005-04-16 15:20:36 -0700358 lock_page(page);
Kirill A. Shutemov5cbc1982016-11-30 15:54:19 -0800359 WARN_ON(page_to_index(page) != index);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700360 wait_on_page_writeback(page);
Nick Piggin750b4982009-09-16 11:50:12 +0200361 truncate_inode_page(mapping, page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700362 unlock_page(page);
363 }
Johannes Weiner0cd61442014-04-03 14:47:46 -0700364 pagevec_remove_exceptionals(&pvec);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700365 pagevec_release(&pvec);
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700366 index++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700367 }
Dan Magenheimer31677602011-09-21 11:56:28 -0400368 cleancache_invalidate_inode(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700369}
Hans Reiserd7339072006-01-06 00:10:36 -0800370EXPORT_SYMBOL(truncate_inode_pages_range);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700371
Hans Reiserd7339072006-01-06 00:10:36 -0800372/**
373 * truncate_inode_pages - truncate *all* the pages from an offset
374 * @mapping: mapping to truncate
375 * @lstart: offset from which to truncate
376 *
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800377 * Called under (and serialised by) inode->i_mutex.
Jan Kara08142572011-06-27 16:18:10 -0700378 *
379 * Note: When this function returns, there can be a page in the process of
380 * deletion (inside __delete_from_page_cache()) in the specified range. Thus
381 * mapping->nrpages can be non-zero when this function returns even after
382 * truncation of the whole mapping.
Hans Reiserd7339072006-01-06 00:10:36 -0800383 */
384void truncate_inode_pages(struct address_space *mapping, loff_t lstart)
385{
386 truncate_inode_pages_range(mapping, lstart, (loff_t)-1);
387}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700388EXPORT_SYMBOL(truncate_inode_pages);
389
Mike Waychison28697352009-06-16 15:32:59 -0700390/**
Johannes Weiner91b0abe2014-04-03 14:47:49 -0700391 * truncate_inode_pages_final - truncate *all* pages before inode dies
392 * @mapping: mapping to truncate
393 *
394 * Called under (and serialized by) inode->i_mutex.
395 *
396 * Filesystems have to use this in the .evict_inode path to inform the
397 * VM that this is the final truncate and the inode is going away.
398 */
399void truncate_inode_pages_final(struct address_space *mapping)
400{
Ross Zwislerf9fe48b2016-01-22 15:10:40 -0800401 unsigned long nrexceptional;
Johannes Weiner91b0abe2014-04-03 14:47:49 -0700402 unsigned long nrpages;
403
404 /*
405 * Page reclaim can not participate in regular inode lifetime
406 * management (can't call iput()) and thus can race with the
407 * inode teardown. Tell it when the address space is exiting,
408 * so that it does not install eviction information after the
409 * final truncate has begun.
410 */
411 mapping_set_exiting(mapping);
412
413 /*
414 * When reclaim installs eviction entries, it increases
Ross Zwislerf9fe48b2016-01-22 15:10:40 -0800415 * nrexceptional first, then decreases nrpages. Make sure we see
Johannes Weiner91b0abe2014-04-03 14:47:49 -0700416 * this in the right order or we might miss an entry.
417 */
418 nrpages = mapping->nrpages;
419 smp_rmb();
Ross Zwislerf9fe48b2016-01-22 15:10:40 -0800420 nrexceptional = mapping->nrexceptional;
Johannes Weiner91b0abe2014-04-03 14:47:49 -0700421
Ross Zwislerf9fe48b2016-01-22 15:10:40 -0800422 if (nrpages || nrexceptional) {
Johannes Weiner91b0abe2014-04-03 14:47:49 -0700423 /*
424 * As truncation uses a lockless tree lookup, cycle
425 * the tree lock to make sure any ongoing tree
426 * modification that does not see AS_EXITING is
427 * completed before starting the final truncate.
428 */
429 spin_lock_irq(&mapping->tree_lock);
430 spin_unlock_irq(&mapping->tree_lock);
431
432 truncate_inode_pages(mapping, 0);
433 }
434}
435EXPORT_SYMBOL(truncate_inode_pages_final);
436
437/**
Mike Waychison28697352009-06-16 15:32:59 -0700438 * invalidate_mapping_pages - Invalidate all the unlocked pages of one inode
439 * @mapping: the address_space which holds the pages to invalidate
440 * @start: the offset 'from' which to invalidate
441 * @end: the offset 'to' which to invalidate (inclusive)
442 *
443 * This function only removes the unlocked pages, if you want to
444 * remove all the pages of one inode, you must call truncate_inode_pages.
445 *
446 * invalidate_mapping_pages() will not block on IO activity. It will not
447 * invalidate pages which are dirty, locked, under writeback or mapped into
448 * pagetables.
449 */
450unsigned long invalidate_mapping_pages(struct address_space *mapping,
Minchan Kim31560182011-03-22 16:32:52 -0700451 pgoff_t start, pgoff_t end)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700452{
Johannes Weiner0cd61442014-04-03 14:47:46 -0700453 pgoff_t indices[PAGEVEC_SIZE];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700454 struct pagevec pvec;
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700455 pgoff_t index = start;
Minchan Kim31560182011-03-22 16:32:52 -0700456 unsigned long ret;
457 unsigned long count = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700458 int i;
459
460 pagevec_init(&pvec, 0);
Johannes Weiner0cd61442014-04-03 14:47:46 -0700461 while (index <= end && pagevec_lookup_entries(&pvec, mapping, index,
462 min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1,
463 indices)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700464 for (i = 0; i < pagevec_count(&pvec); i++) {
465 struct page *page = pvec.pages[i];
466
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700467 /* We rely upon deletion not changing page->index */
Johannes Weiner0cd61442014-04-03 14:47:46 -0700468 index = indices[i];
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700469 if (index > end)
470 break;
NeilBrowne0f23602006-06-23 02:05:48 -0700471
Johannes Weiner0cd61442014-04-03 14:47:46 -0700472 if (radix_tree_exceptional_entry(page)) {
473 clear_exceptional_entry(mapping, index, page);
474 continue;
475 }
476
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700477 if (!trylock_page(page))
478 continue;
Kirill A. Shutemovfc127da2016-07-26 15:26:07 -0700479
Kirill A. Shutemov5cbc1982016-11-30 15:54:19 -0800480 WARN_ON(page_to_index(page) != index);
Kirill A. Shutemovfc127da2016-07-26 15:26:07 -0700481
482 /* Middle of THP: skip */
483 if (PageTransTail(page)) {
484 unlock_page(page);
485 continue;
486 } else if (PageTransHuge(page)) {
487 index += HPAGE_PMD_NR - 1;
488 i += HPAGE_PMD_NR - 1;
489 /* 'end' is in the middle of THP */
490 if (index == round_down(end, HPAGE_PMD_NR))
491 continue;
492 }
493
Minchan Kim31560182011-03-22 16:32:52 -0700494 ret = invalidate_inode_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700495 unlock_page(page);
Minchan Kim31560182011-03-22 16:32:52 -0700496 /*
497 * Invalidation is a hint that the page is no longer
498 * of interest and try to speed up its reclaim.
499 */
500 if (!ret)
Minchan Kimcc5993b2015-04-15 16:13:26 -0700501 deactivate_file_page(page);
Minchan Kim31560182011-03-22 16:32:52 -0700502 count += ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700503 }
Johannes Weiner0cd61442014-04-03 14:47:46 -0700504 pagevec_remove_exceptionals(&pvec);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700505 pagevec_release(&pvec);
Mike Waychison28697352009-06-16 15:32:59 -0700506 cond_resched();
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700507 index++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700508 }
Minchan Kim31560182011-03-22 16:32:52 -0700509 return count;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700510}
Anton Altaparmakov54bc4852007-02-10 01:45:38 -0800511EXPORT_SYMBOL(invalidate_mapping_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700512
Andrew Mortonbd4c8ce2006-09-30 23:29:29 -0700513/*
514 * This is like invalidate_complete_page(), except it ignores the page's
515 * refcount. We do this because invalidate_inode_pages2() needs stronger
516 * invalidation guarantees, and cannot afford to leave pages behind because
Anderson Briglia2706a1b2007-07-15 23:38:09 -0700517 * shrink_page_list() has a temp ref on them, or because they're transiently
518 * sitting in the lru_cache_add() pagevecs.
Andrew Mortonbd4c8ce2006-09-30 23:29:29 -0700519 */
520static int
521invalidate_complete_page2(struct address_space *mapping, struct page *page)
522{
Greg Thelenc4843a72015-05-22 17:13:16 -0400523 unsigned long flags;
524
Andrew Mortonbd4c8ce2006-09-30 23:29:29 -0700525 if (page->mapping != mapping)
526 return 0;
527
David Howells266cf652009-04-03 16:42:36 +0100528 if (page_has_private(page) && !try_to_release_page(page, GFP_KERNEL))
Andrew Mortonbd4c8ce2006-09-30 23:29:29 -0700529 return 0;
530
Greg Thelenc4843a72015-05-22 17:13:16 -0400531 spin_lock_irqsave(&mapping->tree_lock, flags);
Andrew Mortonbd4c8ce2006-09-30 23:29:29 -0700532 if (PageDirty(page))
533 goto failed;
534
David Howells266cf652009-04-03 16:42:36 +0100535 BUG_ON(page_has_private(page));
Johannes Weiner62cccb82016-03-15 14:57:22 -0700536 __delete_from_page_cache(page, NULL);
Greg Thelenc4843a72015-05-22 17:13:16 -0400537 spin_unlock_irqrestore(&mapping->tree_lock, flags);
Linus Torvalds6072d132010-12-01 13:35:19 -0500538
539 if (mapping->a_ops->freepage)
540 mapping->a_ops->freepage(page);
541
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300542 put_page(page); /* pagecache ref */
Andrew Mortonbd4c8ce2006-09-30 23:29:29 -0700543 return 1;
544failed:
Greg Thelenc4843a72015-05-22 17:13:16 -0400545 spin_unlock_irqrestore(&mapping->tree_lock, flags);
Andrew Mortonbd4c8ce2006-09-30 23:29:29 -0700546 return 0;
547}
548
Trond Myklebuste3db7692007-01-10 23:15:39 -0800549static int do_launder_page(struct address_space *mapping, struct page *page)
550{
551 if (!PageDirty(page))
552 return 0;
553 if (page->mapping != mapping || mapping->a_ops->launder_page == NULL)
554 return 0;
555 return mapping->a_ops->launder_page(page);
556}
557
Linus Torvalds1da177e2005-04-16 15:20:36 -0700558/**
559 * invalidate_inode_pages2_range - remove range of pages from an address_space
Martin Waitz67be2dd2005-05-01 08:59:26 -0700560 * @mapping: the address_space
Linus Torvalds1da177e2005-04-16 15:20:36 -0700561 * @start: the page offset 'from' which to invalidate
562 * @end: the page offset 'to' which to invalidate (inclusive)
563 *
564 * Any pages which are found to be mapped into pagetables are unmapped prior to
565 * invalidation.
566 *
Hisashi Hifumi6ccfa802008-09-02 14:35:40 -0700567 * Returns -EBUSY if any pages could not be invalidated.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700568 */
569int invalidate_inode_pages2_range(struct address_space *mapping,
570 pgoff_t start, pgoff_t end)
571{
Johannes Weiner0cd61442014-04-03 14:47:46 -0700572 pgoff_t indices[PAGEVEC_SIZE];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700573 struct pagevec pvec;
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700574 pgoff_t index;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700575 int i;
576 int ret = 0;
Hisashi Hifumi0dd13342008-04-28 02:12:08 -0700577 int ret2 = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700578 int did_range_unmap = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700579
Dan Magenheimer31677602011-09-21 11:56:28 -0400580 cleancache_invalidate_inode(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700581 pagevec_init(&pvec, 0);
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700582 index = start;
Johannes Weiner0cd61442014-04-03 14:47:46 -0700583 while (index <= end && pagevec_lookup_entries(&pvec, mapping, index,
584 min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1,
585 indices)) {
Trond Myklebust7b965e02007-02-28 20:13:55 -0800586 for (i = 0; i < pagevec_count(&pvec); i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700587 struct page *page = pvec.pages[i];
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700588
589 /* We rely upon deletion not changing page->index */
Johannes Weiner0cd61442014-04-03 14:47:46 -0700590 index = indices[i];
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700591 if (index > end)
592 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700593
Johannes Weiner0cd61442014-04-03 14:47:46 -0700594 if (radix_tree_exceptional_entry(page)) {
595 clear_exceptional_entry(mapping, index, page);
596 continue;
597 }
598
Linus Torvalds1da177e2005-04-16 15:20:36 -0700599 lock_page(page);
Kirill A. Shutemov5cbc1982016-11-30 15:54:19 -0800600 WARN_ON(page_to_index(page) != index);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700601 if (page->mapping != mapping) {
602 unlock_page(page);
603 continue;
604 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700605 wait_on_page_writeback(page);
Nick Piggind00806b2007-07-19 01:46:57 -0700606 if (page_mapped(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700607 if (!did_range_unmap) {
608 /*
609 * Zap the rest of the file in one hit.
610 */
611 unmap_mapping_range(mapping,
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300612 (loff_t)index << PAGE_SHIFT,
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700613 (loff_t)(1 + end - index)
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300614 << PAGE_SHIFT,
615 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700616 did_range_unmap = 1;
617 } else {
618 /*
619 * Just zap this page
620 */
621 unmap_mapping_range(mapping,
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300622 (loff_t)index << PAGE_SHIFT,
623 PAGE_SIZE, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700624 }
625 }
Nick Piggind00806b2007-07-19 01:46:57 -0700626 BUG_ON(page_mapped(page));
Hisashi Hifumi0dd13342008-04-28 02:12:08 -0700627 ret2 = do_launder_page(mapping, page);
628 if (ret2 == 0) {
629 if (!invalidate_complete_page2(mapping, page))
Hisashi Hifumi6ccfa802008-09-02 14:35:40 -0700630 ret2 = -EBUSY;
Hisashi Hifumi0dd13342008-04-28 02:12:08 -0700631 }
632 if (ret2 < 0)
633 ret = ret2;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700634 unlock_page(page);
635 }
Johannes Weiner0cd61442014-04-03 14:47:46 -0700636 pagevec_remove_exceptionals(&pvec);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700637 pagevec_release(&pvec);
638 cond_resched();
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700639 index++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700640 }
Dan Magenheimer31677602011-09-21 11:56:28 -0400641 cleancache_invalidate_inode(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700642 return ret;
643}
644EXPORT_SYMBOL_GPL(invalidate_inode_pages2_range);
645
646/**
647 * invalidate_inode_pages2 - remove all pages from an address_space
Martin Waitz67be2dd2005-05-01 08:59:26 -0700648 * @mapping: the address_space
Linus Torvalds1da177e2005-04-16 15:20:36 -0700649 *
650 * Any pages which are found to be mapped into pagetables are unmapped prior to
651 * invalidation.
652 *
Peng Taoe9de25d2009-10-19 14:48:13 +0800653 * Returns -EBUSY if any pages could not be invalidated.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700654 */
655int invalidate_inode_pages2(struct address_space *mapping)
656{
657 return invalidate_inode_pages2_range(mapping, 0, -1);
658}
659EXPORT_SYMBOL_GPL(invalidate_inode_pages2);
npiggin@suse.de25d9e2d2009-08-21 02:35:05 +1000660
661/**
662 * truncate_pagecache - unmap and remove pagecache that has been truncated
663 * @inode: inode
Hugh Dickins8a549be2011-07-25 17:12:24 -0700664 * @newsize: new file size
npiggin@suse.de25d9e2d2009-08-21 02:35:05 +1000665 *
666 * inode's new i_size must already be written before truncate_pagecache
667 * is called.
668 *
669 * This function should typically be called before the filesystem
670 * releases resources associated with the freed range (eg. deallocates
671 * blocks). This way, pagecache will always stay logically coherent
672 * with on-disk format, and the filesystem would not have to deal with
673 * situations such as writepage being called for a page that has already
674 * had its underlying blocks deallocated.
675 */
Kirill A. Shutemov7caef262013-09-12 15:13:56 -0700676void truncate_pagecache(struct inode *inode, loff_t newsize)
npiggin@suse.de25d9e2d2009-08-21 02:35:05 +1000677{
OGAWA Hirofumicedabed2010-01-13 21:14:09 +0900678 struct address_space *mapping = inode->i_mapping;
Hugh Dickins8a549be2011-07-25 17:12:24 -0700679 loff_t holebegin = round_up(newsize, PAGE_SIZE);
npiggin@suse.de25d9e2d2009-08-21 02:35:05 +1000680
OGAWA Hirofumicedabed2010-01-13 21:14:09 +0900681 /*
682 * unmap_mapping_range is called twice, first simply for
683 * efficiency so that truncate_inode_pages does fewer
684 * single-page unmaps. However after this first call, and
685 * before truncate_inode_pages finishes, it is possible for
686 * private pages to be COWed, which remain after
687 * truncate_inode_pages finishes, hence the second
688 * unmap_mapping_range call must be made for correctness.
689 */
Hugh Dickins8a549be2011-07-25 17:12:24 -0700690 unmap_mapping_range(mapping, holebegin, 0, 1);
691 truncate_inode_pages(mapping, newsize);
692 unmap_mapping_range(mapping, holebegin, 0, 1);
npiggin@suse.de25d9e2d2009-08-21 02:35:05 +1000693}
694EXPORT_SYMBOL(truncate_pagecache);
695
696/**
Christoph Hellwig2c27c652010-06-04 11:30:04 +0200697 * truncate_setsize - update inode and pagecache for a new file size
698 * @inode: inode
699 * @newsize: new file size
700 *
Jan Kara382e27d2011-01-20 14:44:26 -0800701 * truncate_setsize updates i_size and performs pagecache truncation (if
702 * necessary) to @newsize. It will be typically be called from the filesystem's
703 * setattr function when ATTR_SIZE is passed in.
Christoph Hellwig2c27c652010-06-04 11:30:04 +0200704 *
Jan Kara77783d02014-11-07 08:29:25 +1100705 * Must be called with a lock serializing truncates and writes (generally
706 * i_mutex but e.g. xfs uses a different lock) and before all filesystem
707 * specific block truncation has been performed.
Christoph Hellwig2c27c652010-06-04 11:30:04 +0200708 */
709void truncate_setsize(struct inode *inode, loff_t newsize)
710{
Jan Kara90a80202014-10-01 21:49:18 -0400711 loff_t oldsize = inode->i_size;
712
Christoph Hellwig2c27c652010-06-04 11:30:04 +0200713 i_size_write(inode, newsize);
Jan Kara90a80202014-10-01 21:49:18 -0400714 if (newsize > oldsize)
715 pagecache_isize_extended(inode, oldsize, newsize);
Kirill A. Shutemov7caef262013-09-12 15:13:56 -0700716 truncate_pagecache(inode, newsize);
Christoph Hellwig2c27c652010-06-04 11:30:04 +0200717}
718EXPORT_SYMBOL(truncate_setsize);
719
720/**
Jan Kara90a80202014-10-01 21:49:18 -0400721 * pagecache_isize_extended - update pagecache after extension of i_size
722 * @inode: inode for which i_size was extended
723 * @from: original inode size
724 * @to: new inode size
725 *
726 * Handle extension of inode size either caused by extending truncate or by
727 * write starting after current i_size. We mark the page straddling current
728 * i_size RO so that page_mkwrite() is called on the nearest write access to
729 * the page. This way filesystem can be sure that page_mkwrite() is called on
730 * the page before user writes to the page via mmap after the i_size has been
731 * changed.
732 *
733 * The function must be called after i_size is updated so that page fault
734 * coming after we unlock the page will already see the new i_size.
735 * The function must be called while we still hold i_mutex - this not only
736 * makes sure i_size is stable but also that userspace cannot observe new
737 * i_size value before we are prepared to store mmap writes at new inode size.
738 */
739void pagecache_isize_extended(struct inode *inode, loff_t from, loff_t to)
740{
741 int bsize = 1 << inode->i_blkbits;
742 loff_t rounded_from;
743 struct page *page;
744 pgoff_t index;
745
Jan Kara90a80202014-10-01 21:49:18 -0400746 WARN_ON(to > inode->i_size);
747
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300748 if (from >= to || bsize == PAGE_SIZE)
Jan Kara90a80202014-10-01 21:49:18 -0400749 return;
750 /* Page straddling @from will not have any hole block created? */
751 rounded_from = round_up(from, bsize);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300752 if (to <= rounded_from || !(rounded_from & (PAGE_SIZE - 1)))
Jan Kara90a80202014-10-01 21:49:18 -0400753 return;
754
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300755 index = from >> PAGE_SHIFT;
Jan Kara90a80202014-10-01 21:49:18 -0400756 page = find_lock_page(inode->i_mapping, index);
757 /* Page not cached? Nothing to do */
758 if (!page)
759 return;
760 /*
761 * See clear_page_dirty_for_io() for details why set_page_dirty()
762 * is needed.
763 */
764 if (page_mkclean(page))
765 set_page_dirty(page);
766 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300767 put_page(page);
Jan Kara90a80202014-10-01 21:49:18 -0400768}
769EXPORT_SYMBOL(pagecache_isize_extended);
770
771/**
Hugh Dickins623e3db2012-03-28 14:42:40 -0700772 * truncate_pagecache_range - unmap and remove pagecache that is hole-punched
773 * @inode: inode
774 * @lstart: offset of beginning of hole
775 * @lend: offset of last byte of hole
776 *
777 * This function should typically be called before the filesystem
778 * releases resources associated with the freed range (eg. deallocates
779 * blocks). This way, pagecache will always stay logically coherent
780 * with on-disk format, and the filesystem would not have to deal with
781 * situations such as writepage being called for a page that has already
782 * had its underlying blocks deallocated.
783 */
784void truncate_pagecache_range(struct inode *inode, loff_t lstart, loff_t lend)
785{
786 struct address_space *mapping = inode->i_mapping;
787 loff_t unmap_start = round_up(lstart, PAGE_SIZE);
788 loff_t unmap_end = round_down(1 + lend, PAGE_SIZE) - 1;
789 /*
790 * This rounding is currently just for example: unmap_mapping_range
791 * expands its hole outwards, whereas we want it to contract the hole
792 * inwards. However, existing callers of truncate_pagecache_range are
Lukas Czerner5a720392013-05-27 23:32:35 -0400793 * doing their own page rounding first. Note that unmap_mapping_range
794 * allows holelen 0 for all, and we allow lend -1 for end of file.
Hugh Dickins623e3db2012-03-28 14:42:40 -0700795 */
796
797 /*
798 * Unlike in truncate_pagecache, unmap_mapping_range is called only
799 * once (before truncating pagecache), and without "even_cows" flag:
800 * hole-punching should not remove private COWed pages from the hole.
801 */
802 if ((u64)unmap_end > (u64)unmap_start)
803 unmap_mapping_range(mapping, unmap_start,
804 1 + unmap_end - unmap_start, 0);
805 truncate_inode_pages_range(mapping, lstart, lend);
806}
807EXPORT_SYMBOL(truncate_pagecache_range);