blob: 4064f8f53daa73aaca001747cbded8d9b451d7f7 [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 */
47 if (!__radix_tree_lookup(&mapping->page_tree, index, &node,
48 &slot))
49 goto unlock;
50 if (*slot != entry)
51 goto unlock;
52 radix_tree_replace_slot(slot, NULL);
53 mapping->nrexceptional--;
54 if (!node)
55 goto unlock;
56 workingset_node_shadows_dec(node);
57 /*
58 * Don't track node without shadow entries.
59 *
60 * Avoid acquiring the list_lru lock if already untracked.
61 * The list_empty() test is safe as node->private_list is
62 * protected by mapping->tree_lock.
63 */
64 if (!workingset_node_shadows(node) &&
65 !list_empty(&node->private_list))
66 list_lru_del(&workingset_shadow_nodes,
67 &node->private_list);
68 __radix_tree_delete_node(&mapping->page_tree, node);
Johannes Weiner449dd692014-04-03 14:47:56 -070069unlock:
Johannes Weiner0cd61442014-04-03 14:47:46 -070070 spin_unlock_irq(&mapping->tree_lock);
71}
Linus Torvalds1da177e2005-04-16 15:20:36 -070072
David Howellscf9a2ae2006-08-29 19:05:54 +010073/**
Fengguang Wu28bc44d2008-02-03 18:04:10 +020074 * do_invalidatepage - invalidate part or all of a page
David Howellscf9a2ae2006-08-29 19:05:54 +010075 * @page: the page which is affected
Lukas Czernerd47992f2013-05-21 23:17:23 -040076 * @offset: start of the range to invalidate
77 * @length: length of the range to invalidate
David Howellscf9a2ae2006-08-29 19:05:54 +010078 *
79 * do_invalidatepage() is called when all or part of the page has become
80 * invalidated by a truncate operation.
81 *
82 * do_invalidatepage() does not have to release all buffers, but it must
83 * ensure that no dirty buffer is left outside @offset and that no I/O
84 * is underway against any of the blocks which are outside the truncation
85 * point. Because the caller is about to free (and possibly reuse) those
86 * blocks on-disk.
87 */
Lukas Czernerd47992f2013-05-21 23:17:23 -040088void do_invalidatepage(struct page *page, unsigned int offset,
89 unsigned int length)
David Howellscf9a2ae2006-08-29 19:05:54 +010090{
Lukas Czernerd47992f2013-05-21 23:17:23 -040091 void (*invalidatepage)(struct page *, unsigned int, unsigned int);
92
David Howellscf9a2ae2006-08-29 19:05:54 +010093 invalidatepage = page->mapping->a_ops->invalidatepage;
David Howells93614012006-09-30 20:45:40 +020094#ifdef CONFIG_BLOCK
David Howellscf9a2ae2006-08-29 19:05:54 +010095 if (!invalidatepage)
96 invalidatepage = block_invalidatepage;
David Howells93614012006-09-30 20:45:40 +020097#endif
David Howellscf9a2ae2006-08-29 19:05:54 +010098 if (invalidatepage)
Lukas Czernerd47992f2013-05-21 23:17:23 -040099 (*invalidatepage)(page, offset, length);
David Howellscf9a2ae2006-08-29 19:05:54 +0100100}
101
Linus Torvaldsecdfc972007-01-26 12:47:06 -0800102/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700103 * If truncate cannot remove the fs-private metadata from the page, the page
Shaohua Li62e1c552008-02-04 22:29:33 -0800104 * becomes orphaned. It will be left on the LRU and may even be mapped into
Nick Piggin54cb8822007-07-19 01:46:59 -0700105 * user pagetables if we're racing with filemap_fault().
Linus Torvalds1da177e2005-04-16 15:20:36 -0700106 *
107 * We need to bale out if page->mapping is no longer equal to the original
108 * mapping. This happens a) when the VM reclaimed the page while we waited on
Andrew Mortonfc0ecff2007-02-10 01:45:39 -0800109 * its lock, b) when a concurrent invalidate_mapping_pages got there first and
Linus Torvalds1da177e2005-04-16 15:20:36 -0700110 * c) when tmpfs swizzles a page between a tmpfs inode and swapper_space.
111 */
Nick Piggin750b4982009-09-16 11:50:12 +0200112static int
Linus Torvalds1da177e2005-04-16 15:20:36 -0700113truncate_complete_page(struct address_space *mapping, struct page *page)
114{
115 if (page->mapping != mapping)
Nick Piggin750b4982009-09-16 11:50:12 +0200116 return -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700117
David Howells266cf652009-04-03 16:42:36 +0100118 if (page_has_private(page))
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300119 do_invalidatepage(page, 0, PAGE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700120
Konstantin Khlebnikovb9ea2512015-04-14 15:45:27 -0700121 /*
122 * Some filesystems seem to re-dirty the page even after
123 * the VM has canceled the dirty bit (eg ext3 journaling).
124 * Hence dirty accounting check is placed after invalidation.
125 */
Tejun Heo11f81be2015-05-22 17:13:15 -0400126 cancel_dirty_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700127 ClearPageMappedToDisk(page);
Minchan Kim5adc7b52011-03-22 16:32:41 -0700128 delete_from_page_cache(page);
Nick Piggin750b4982009-09-16 11:50:12 +0200129 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700130}
131
132/*
Andrew Mortonfc0ecff2007-02-10 01:45:39 -0800133 * This is for invalidate_mapping_pages(). That function can be called at
Linus Torvalds1da177e2005-04-16 15:20:36 -0700134 * any time, and is not supposed to throw away dirty pages. But pages can
Nick Piggin0fd0e6b2006-09-27 01:50:02 -0700135 * be marked dirty at any time too, so use remove_mapping which safely
136 * discards clean, unused pages.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700137 *
138 * Returns non-zero if the page was successfully invalidated.
139 */
140static int
141invalidate_complete_page(struct address_space *mapping, struct page *page)
142{
Nick Piggin0fd0e6b2006-09-27 01:50:02 -0700143 int ret;
144
Linus Torvalds1da177e2005-04-16 15:20:36 -0700145 if (page->mapping != mapping)
146 return 0;
147
David Howells266cf652009-04-03 16:42:36 +0100148 if (page_has_private(page) && !try_to_release_page(page, 0))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700149 return 0;
150
Nick Piggin0fd0e6b2006-09-27 01:50:02 -0700151 ret = remove_mapping(mapping, page);
Nick Piggin0fd0e6b2006-09-27 01:50:02 -0700152
153 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700154}
155
Nick Piggin750b4982009-09-16 11:50:12 +0200156int truncate_inode_page(struct address_space *mapping, struct page *page)
157{
158 if (page_mapped(page)) {
159 unmap_mapping_range(mapping,
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300160 (loff_t)page->index << PAGE_SHIFT,
161 PAGE_SIZE, 0);
Nick Piggin750b4982009-09-16 11:50:12 +0200162 }
163 return truncate_complete_page(mapping, page);
164}
165
Wu Fengguang83f78662009-09-16 11:50:13 +0200166/*
Andi Kleen25718732009-09-16 11:50:13 +0200167 * Used to get rid of pages on hardware memory corruption.
168 */
169int generic_error_remove_page(struct address_space *mapping, struct page *page)
170{
171 if (!mapping)
172 return -EINVAL;
173 /*
174 * Only punch for normal data pages for now.
175 * Handling other types like directories would need more auditing.
176 */
177 if (!S_ISREG(mapping->host->i_mode))
178 return -EIO;
179 return truncate_inode_page(mapping, page);
180}
181EXPORT_SYMBOL(generic_error_remove_page);
182
183/*
Wu Fengguang83f78662009-09-16 11:50:13 +0200184 * Safely invalidate one page from its pagecache mapping.
185 * It only drops clean, unused pages. The page must be locked.
186 *
187 * Returns 1 if the page is successfully invalidated, otherwise 0.
188 */
189int invalidate_inode_page(struct page *page)
190{
191 struct address_space *mapping = page_mapping(page);
192 if (!mapping)
193 return 0;
194 if (PageDirty(page) || PageWriteback(page))
195 return 0;
196 if (page_mapped(page))
197 return 0;
198 return invalidate_complete_page(mapping, page);
199}
200
Linus Torvalds1da177e2005-04-16 15:20:36 -0700201/**
Liu Bo73c1e202012-02-21 10:57:20 +0800202 * truncate_inode_pages_range - truncate range of pages specified by start & end byte offsets
Linus Torvalds1da177e2005-04-16 15:20:36 -0700203 * @mapping: mapping to truncate
204 * @lstart: offset from which to truncate
Lukas Czerner5a720392013-05-27 23:32:35 -0400205 * @lend: offset to which to truncate (inclusive)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700206 *
Hans Reiserd7339072006-01-06 00:10:36 -0800207 * Truncate the page cache, removing the pages that are between
Lukas Czerner5a720392013-05-27 23:32:35 -0400208 * specified offsets (and zeroing out partial pages
209 * if lstart or lend + 1 is not page aligned).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700210 *
211 * Truncate takes two passes - the first pass is nonblocking. It will not
212 * block on page locks and it will not block on writeback. The second pass
213 * will wait. This is to prevent as much IO as possible in the affected region.
214 * The first pass will remove most pages, so the search cost of the second pass
215 * is low.
216 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700217 * We pass down the cache-hot hint to the page freeing code. Even if the
218 * mapping is large, it is probably the case that the final pages are the most
219 * recently touched, and freeing happens in ascending file offset order.
Lukas Czerner5a720392013-05-27 23:32:35 -0400220 *
221 * Note that since ->invalidatepage() accepts range to invalidate
222 * truncate_inode_pages_range is able to handle cases where lend + 1 is not
223 * page aligned properly.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700224 */
Hans Reiserd7339072006-01-06 00:10:36 -0800225void truncate_inode_pages_range(struct address_space *mapping,
226 loff_t lstart, loff_t lend)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700227{
Lukas Czerner5a720392013-05-27 23:32:35 -0400228 pgoff_t start; /* inclusive */
229 pgoff_t end; /* exclusive */
230 unsigned int partial_start; /* inclusive */
231 unsigned int partial_end; /* exclusive */
232 struct pagevec pvec;
Johannes Weiner0cd61442014-04-03 14:47:46 -0700233 pgoff_t indices[PAGEVEC_SIZE];
Lukas Czerner5a720392013-05-27 23:32:35 -0400234 pgoff_t index;
235 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700236
Dan Magenheimer31677602011-09-21 11:56:28 -0400237 cleancache_invalidate_inode(mapping);
Ross Zwislerf9fe48b2016-01-22 15:10:40 -0800238 if (mapping->nrpages == 0 && mapping->nrexceptional == 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700239 return;
240
Lukas Czerner5a720392013-05-27 23:32:35 -0400241 /* Offsets within partial pages */
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300242 partial_start = lstart & (PAGE_SIZE - 1);
243 partial_end = (lend + 1) & (PAGE_SIZE - 1);
Lukas Czerner5a720392013-05-27 23:32:35 -0400244
245 /*
246 * 'start' and 'end' always covers the range of pages to be fully
247 * truncated. Partial pages are covered with 'partial_start' at the
248 * start of the range and 'partial_end' at the end of the range.
249 * Note that 'end' is exclusive while 'lend' is inclusive.
250 */
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300251 start = (lstart + PAGE_SIZE - 1) >> PAGE_SHIFT;
Lukas Czerner5a720392013-05-27 23:32:35 -0400252 if (lend == -1)
253 /*
254 * lend == -1 indicates end-of-file so we have to set 'end'
255 * to the highest possible pgoff_t and since the type is
256 * unsigned we're using -1.
257 */
258 end = -1;
259 else
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300260 end = (lend + 1) >> PAGE_SHIFT;
Hans Reiserd7339072006-01-06 00:10:36 -0800261
Linus Torvalds1da177e2005-04-16 15:20:36 -0700262 pagevec_init(&pvec, 0);
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700263 index = start;
Johannes Weiner0cd61442014-04-03 14:47:46 -0700264 while (index < end && pagevec_lookup_entries(&pvec, mapping, index,
265 min(end - index, (pgoff_t)PAGEVEC_SIZE),
266 indices)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700267 for (i = 0; i < pagevec_count(&pvec); i++) {
268 struct page *page = pvec.pages[i];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700269
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700270 /* We rely upon deletion not changing page->index */
Johannes Weiner0cd61442014-04-03 14:47:46 -0700271 index = indices[i];
Lukas Czerner5a720392013-05-27 23:32:35 -0400272 if (index >= end)
Hans Reiserd7339072006-01-06 00:10:36 -0800273 break;
Hans Reiserd7339072006-01-06 00:10:36 -0800274
Johannes Weiner0cd61442014-04-03 14:47:46 -0700275 if (radix_tree_exceptional_entry(page)) {
276 clear_exceptional_entry(mapping, index, page);
277 continue;
278 }
279
Nick Piggin529ae9a2008-08-02 12:01:03 +0200280 if (!trylock_page(page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700281 continue;
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700282 WARN_ON(page->index != index);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700283 if (PageWriteback(page)) {
284 unlock_page(page);
285 continue;
286 }
Nick Piggin750b4982009-09-16 11:50:12 +0200287 truncate_inode_page(mapping, page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700288 unlock_page(page);
289 }
Johannes Weiner0cd61442014-04-03 14:47:46 -0700290 pagevec_remove_exceptionals(&pvec);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700291 pagevec_release(&pvec);
292 cond_resched();
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700293 index++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700294 }
295
Lukas Czerner5a720392013-05-27 23:32:35 -0400296 if (partial_start) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700297 struct page *page = find_lock_page(mapping, start - 1);
298 if (page) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300299 unsigned int top = PAGE_SIZE;
Lukas Czerner5a720392013-05-27 23:32:35 -0400300 if (start > end) {
301 /* Truncation within a single page */
302 top = partial_end;
303 partial_end = 0;
304 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700305 wait_on_page_writeback(page);
Lukas Czerner5a720392013-05-27 23:32:35 -0400306 zero_user_segment(page, partial_start, top);
307 cleancache_invalidate_page(mapping, page);
308 if (page_has_private(page))
309 do_invalidatepage(page, partial_start,
310 top - partial_start);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700311 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300312 put_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700313 }
314 }
Lukas Czerner5a720392013-05-27 23:32:35 -0400315 if (partial_end) {
316 struct page *page = find_lock_page(mapping, end);
317 if (page) {
318 wait_on_page_writeback(page);
319 zero_user_segment(page, 0, partial_end);
320 cleancache_invalidate_page(mapping, page);
321 if (page_has_private(page))
322 do_invalidatepage(page, 0,
323 partial_end);
324 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300325 put_page(page);
Lukas Czerner5a720392013-05-27 23:32:35 -0400326 }
327 }
328 /*
329 * If the truncation happened within a single page no pages
330 * will be released, just zeroed, so we can bail out now.
331 */
332 if (start >= end)
333 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700334
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700335 index = start;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700336 for ( ; ; ) {
337 cond_resched();
Johannes Weiner0cd61442014-04-03 14:47:46 -0700338 if (!pagevec_lookup_entries(&pvec, mapping, index,
Hugh Dickins792ceae2014-07-23 14:00:15 -0700339 min(end - index, (pgoff_t)PAGEVEC_SIZE), indices)) {
340 /* If all gone from start onwards, we're done */
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700341 if (index == start)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700342 break;
Hugh Dickins792ceae2014-07-23 14:00:15 -0700343 /* Otherwise restart to make sure all gone */
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700344 index = start;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700345 continue;
346 }
Johannes Weiner0cd61442014-04-03 14:47:46 -0700347 if (index == start && indices[0] >= end) {
Hugh Dickins792ceae2014-07-23 14:00:15 -0700348 /* All gone out of hole to be punched, we're done */
Johannes Weiner0cd61442014-04-03 14:47:46 -0700349 pagevec_remove_exceptionals(&pvec);
Hans Reiserd7339072006-01-06 00:10:36 -0800350 pagevec_release(&pvec);
351 break;
352 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700353 for (i = 0; i < pagevec_count(&pvec); i++) {
354 struct page *page = pvec.pages[i];
355
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700356 /* We rely upon deletion not changing page->index */
Johannes Weiner0cd61442014-04-03 14:47:46 -0700357 index = indices[i];
Hugh Dickins792ceae2014-07-23 14:00:15 -0700358 if (index >= end) {
359 /* Restart punch to make sure all gone */
360 index = start - 1;
Hans Reiserd7339072006-01-06 00:10:36 -0800361 break;
Hugh Dickins792ceae2014-07-23 14:00:15 -0700362 }
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700363
Johannes Weiner0cd61442014-04-03 14:47:46 -0700364 if (radix_tree_exceptional_entry(page)) {
365 clear_exceptional_entry(mapping, index, page);
366 continue;
367 }
368
Linus Torvalds1da177e2005-04-16 15:20:36 -0700369 lock_page(page);
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700370 WARN_ON(page->index != index);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700371 wait_on_page_writeback(page);
Nick Piggin750b4982009-09-16 11:50:12 +0200372 truncate_inode_page(mapping, page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700373 unlock_page(page);
374 }
Johannes Weiner0cd61442014-04-03 14:47:46 -0700375 pagevec_remove_exceptionals(&pvec);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700376 pagevec_release(&pvec);
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700377 index++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700378 }
Dan Magenheimer31677602011-09-21 11:56:28 -0400379 cleancache_invalidate_inode(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700380}
Hans Reiserd7339072006-01-06 00:10:36 -0800381EXPORT_SYMBOL(truncate_inode_pages_range);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700382
Hans Reiserd7339072006-01-06 00:10:36 -0800383/**
384 * truncate_inode_pages - truncate *all* the pages from an offset
385 * @mapping: mapping to truncate
386 * @lstart: offset from which to truncate
387 *
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800388 * Called under (and serialised by) inode->i_mutex.
Jan Kara08142572011-06-27 16:18:10 -0700389 *
390 * Note: When this function returns, there can be a page in the process of
391 * deletion (inside __delete_from_page_cache()) in the specified range. Thus
392 * mapping->nrpages can be non-zero when this function returns even after
393 * truncation of the whole mapping.
Hans Reiserd7339072006-01-06 00:10:36 -0800394 */
395void truncate_inode_pages(struct address_space *mapping, loff_t lstart)
396{
397 truncate_inode_pages_range(mapping, lstart, (loff_t)-1);
398}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700399EXPORT_SYMBOL(truncate_inode_pages);
400
Mike Waychison28697352009-06-16 15:32:59 -0700401/**
Johannes Weiner91b0abe2014-04-03 14:47:49 -0700402 * truncate_inode_pages_final - truncate *all* pages before inode dies
403 * @mapping: mapping to truncate
404 *
405 * Called under (and serialized by) inode->i_mutex.
406 *
407 * Filesystems have to use this in the .evict_inode path to inform the
408 * VM that this is the final truncate and the inode is going away.
409 */
410void truncate_inode_pages_final(struct address_space *mapping)
411{
Ross Zwislerf9fe48b2016-01-22 15:10:40 -0800412 unsigned long nrexceptional;
Johannes Weiner91b0abe2014-04-03 14:47:49 -0700413 unsigned long nrpages;
414
415 /*
416 * Page reclaim can not participate in regular inode lifetime
417 * management (can't call iput()) and thus can race with the
418 * inode teardown. Tell it when the address space is exiting,
419 * so that it does not install eviction information after the
420 * final truncate has begun.
421 */
422 mapping_set_exiting(mapping);
423
424 /*
425 * When reclaim installs eviction entries, it increases
Ross Zwislerf9fe48b2016-01-22 15:10:40 -0800426 * nrexceptional first, then decreases nrpages. Make sure we see
Johannes Weiner91b0abe2014-04-03 14:47:49 -0700427 * this in the right order or we might miss an entry.
428 */
429 nrpages = mapping->nrpages;
430 smp_rmb();
Ross Zwislerf9fe48b2016-01-22 15:10:40 -0800431 nrexceptional = mapping->nrexceptional;
Johannes Weiner91b0abe2014-04-03 14:47:49 -0700432
Ross Zwislerf9fe48b2016-01-22 15:10:40 -0800433 if (nrpages || nrexceptional) {
Johannes Weiner91b0abe2014-04-03 14:47:49 -0700434 /*
435 * As truncation uses a lockless tree lookup, cycle
436 * the tree lock to make sure any ongoing tree
437 * modification that does not see AS_EXITING is
438 * completed before starting the final truncate.
439 */
440 spin_lock_irq(&mapping->tree_lock);
441 spin_unlock_irq(&mapping->tree_lock);
442
443 truncate_inode_pages(mapping, 0);
444 }
445}
446EXPORT_SYMBOL(truncate_inode_pages_final);
447
448/**
Mike Waychison28697352009-06-16 15:32:59 -0700449 * invalidate_mapping_pages - Invalidate all the unlocked pages of one inode
450 * @mapping: the address_space which holds the pages to invalidate
451 * @start: the offset 'from' which to invalidate
452 * @end: the offset 'to' which to invalidate (inclusive)
453 *
454 * This function only removes the unlocked pages, if you want to
455 * remove all the pages of one inode, you must call truncate_inode_pages.
456 *
457 * invalidate_mapping_pages() will not block on IO activity. It will not
458 * invalidate pages which are dirty, locked, under writeback or mapped into
459 * pagetables.
460 */
461unsigned long invalidate_mapping_pages(struct address_space *mapping,
Minchan Kim31560182011-03-22 16:32:52 -0700462 pgoff_t start, pgoff_t end)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700463{
Johannes Weiner0cd61442014-04-03 14:47:46 -0700464 pgoff_t indices[PAGEVEC_SIZE];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700465 struct pagevec pvec;
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700466 pgoff_t index = start;
Minchan Kim31560182011-03-22 16:32:52 -0700467 unsigned long ret;
468 unsigned long count = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700469 int i;
470
471 pagevec_init(&pvec, 0);
Johannes Weiner0cd61442014-04-03 14:47:46 -0700472 while (index <= end && pagevec_lookup_entries(&pvec, mapping, index,
473 min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1,
474 indices)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700475 for (i = 0; i < pagevec_count(&pvec); i++) {
476 struct page *page = pvec.pages[i];
477
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700478 /* We rely upon deletion not changing page->index */
Johannes Weiner0cd61442014-04-03 14:47:46 -0700479 index = indices[i];
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700480 if (index > end)
481 break;
NeilBrowne0f23602006-06-23 02:05:48 -0700482
Johannes Weiner0cd61442014-04-03 14:47:46 -0700483 if (radix_tree_exceptional_entry(page)) {
484 clear_exceptional_entry(mapping, index, page);
485 continue;
486 }
487
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700488 if (!trylock_page(page))
489 continue;
490 WARN_ON(page->index != index);
Minchan Kim31560182011-03-22 16:32:52 -0700491 ret = invalidate_inode_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700492 unlock_page(page);
Minchan Kim31560182011-03-22 16:32:52 -0700493 /*
494 * Invalidation is a hint that the page is no longer
495 * of interest and try to speed up its reclaim.
496 */
497 if (!ret)
Minchan Kimcc5993b2015-04-15 16:13:26 -0700498 deactivate_file_page(page);
Minchan Kim31560182011-03-22 16:32:52 -0700499 count += ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700500 }
Johannes Weiner0cd61442014-04-03 14:47:46 -0700501 pagevec_remove_exceptionals(&pvec);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700502 pagevec_release(&pvec);
Mike Waychison28697352009-06-16 15:32:59 -0700503 cond_resched();
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700504 index++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700505 }
Minchan Kim31560182011-03-22 16:32:52 -0700506 return count;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700507}
Anton Altaparmakov54bc4852007-02-10 01:45:38 -0800508EXPORT_SYMBOL(invalidate_mapping_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700509
Andrew Mortonbd4c8ce2006-09-30 23:29:29 -0700510/*
511 * This is like invalidate_complete_page(), except it ignores the page's
512 * refcount. We do this because invalidate_inode_pages2() needs stronger
513 * invalidation guarantees, and cannot afford to leave pages behind because
Anderson Briglia2706a1b2007-07-15 23:38:09 -0700514 * shrink_page_list() has a temp ref on them, or because they're transiently
515 * sitting in the lru_cache_add() pagevecs.
Andrew Mortonbd4c8ce2006-09-30 23:29:29 -0700516 */
517static int
518invalidate_complete_page2(struct address_space *mapping, struct page *page)
519{
Greg Thelenc4843a72015-05-22 17:13:16 -0400520 unsigned long flags;
521
Andrew Mortonbd4c8ce2006-09-30 23:29:29 -0700522 if (page->mapping != mapping)
523 return 0;
524
David Howells266cf652009-04-03 16:42:36 +0100525 if (page_has_private(page) && !try_to_release_page(page, GFP_KERNEL))
Andrew Mortonbd4c8ce2006-09-30 23:29:29 -0700526 return 0;
527
Greg Thelenc4843a72015-05-22 17:13:16 -0400528 spin_lock_irqsave(&mapping->tree_lock, flags);
Andrew Mortonbd4c8ce2006-09-30 23:29:29 -0700529 if (PageDirty(page))
530 goto failed;
531
David Howells266cf652009-04-03 16:42:36 +0100532 BUG_ON(page_has_private(page));
Johannes Weiner62cccb82016-03-15 14:57:22 -0700533 __delete_from_page_cache(page, NULL);
Greg Thelenc4843a72015-05-22 17:13:16 -0400534 spin_unlock_irqrestore(&mapping->tree_lock, flags);
Linus Torvalds6072d132010-12-01 13:35:19 -0500535
536 if (mapping->a_ops->freepage)
537 mapping->a_ops->freepage(page);
538
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300539 put_page(page); /* pagecache ref */
Andrew Mortonbd4c8ce2006-09-30 23:29:29 -0700540 return 1;
541failed:
Greg Thelenc4843a72015-05-22 17:13:16 -0400542 spin_unlock_irqrestore(&mapping->tree_lock, flags);
Andrew Mortonbd4c8ce2006-09-30 23:29:29 -0700543 return 0;
544}
545
Trond Myklebuste3db7692007-01-10 23:15:39 -0800546static int do_launder_page(struct address_space *mapping, struct page *page)
547{
548 if (!PageDirty(page))
549 return 0;
550 if (page->mapping != mapping || mapping->a_ops->launder_page == NULL)
551 return 0;
552 return mapping->a_ops->launder_page(page);
553}
554
Linus Torvalds1da177e2005-04-16 15:20:36 -0700555/**
556 * invalidate_inode_pages2_range - remove range of pages from an address_space
Martin Waitz67be2dd2005-05-01 08:59:26 -0700557 * @mapping: the address_space
Linus Torvalds1da177e2005-04-16 15:20:36 -0700558 * @start: the page offset 'from' which to invalidate
559 * @end: the page offset 'to' which to invalidate (inclusive)
560 *
561 * Any pages which are found to be mapped into pagetables are unmapped prior to
562 * invalidation.
563 *
Hisashi Hifumi6ccfa802008-09-02 14:35:40 -0700564 * Returns -EBUSY if any pages could not be invalidated.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700565 */
566int invalidate_inode_pages2_range(struct address_space *mapping,
567 pgoff_t start, pgoff_t end)
568{
Johannes Weiner0cd61442014-04-03 14:47:46 -0700569 pgoff_t indices[PAGEVEC_SIZE];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700570 struct pagevec pvec;
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700571 pgoff_t index;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700572 int i;
573 int ret = 0;
Hisashi Hifumi0dd13342008-04-28 02:12:08 -0700574 int ret2 = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700575 int did_range_unmap = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700576
Dan Magenheimer31677602011-09-21 11:56:28 -0400577 cleancache_invalidate_inode(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700578 pagevec_init(&pvec, 0);
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700579 index = start;
Johannes Weiner0cd61442014-04-03 14:47:46 -0700580 while (index <= end && pagevec_lookup_entries(&pvec, mapping, index,
581 min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1,
582 indices)) {
Trond Myklebust7b965e02007-02-28 20:13:55 -0800583 for (i = 0; i < pagevec_count(&pvec); i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700584 struct page *page = pvec.pages[i];
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700585
586 /* We rely upon deletion not changing page->index */
Johannes Weiner0cd61442014-04-03 14:47:46 -0700587 index = indices[i];
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700588 if (index > end)
589 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700590
Johannes Weiner0cd61442014-04-03 14:47:46 -0700591 if (radix_tree_exceptional_entry(page)) {
592 clear_exceptional_entry(mapping, index, page);
593 continue;
594 }
595
Linus Torvalds1da177e2005-04-16 15:20:36 -0700596 lock_page(page);
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700597 WARN_ON(page->index != index);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700598 if (page->mapping != mapping) {
599 unlock_page(page);
600 continue;
601 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700602 wait_on_page_writeback(page);
Nick Piggind00806b2007-07-19 01:46:57 -0700603 if (page_mapped(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700604 if (!did_range_unmap) {
605 /*
606 * Zap the rest of the file in one hit.
607 */
608 unmap_mapping_range(mapping,
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300609 (loff_t)index << PAGE_SHIFT,
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700610 (loff_t)(1 + end - index)
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300611 << PAGE_SHIFT,
612 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700613 did_range_unmap = 1;
614 } else {
615 /*
616 * Just zap this page
617 */
618 unmap_mapping_range(mapping,
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300619 (loff_t)index << PAGE_SHIFT,
620 PAGE_SIZE, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700621 }
622 }
Nick Piggind00806b2007-07-19 01:46:57 -0700623 BUG_ON(page_mapped(page));
Hisashi Hifumi0dd13342008-04-28 02:12:08 -0700624 ret2 = do_launder_page(mapping, page);
625 if (ret2 == 0) {
626 if (!invalidate_complete_page2(mapping, page))
Hisashi Hifumi6ccfa802008-09-02 14:35:40 -0700627 ret2 = -EBUSY;
Hisashi Hifumi0dd13342008-04-28 02:12:08 -0700628 }
629 if (ret2 < 0)
630 ret = ret2;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700631 unlock_page(page);
632 }
Johannes Weiner0cd61442014-04-03 14:47:46 -0700633 pagevec_remove_exceptionals(&pvec);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700634 pagevec_release(&pvec);
635 cond_resched();
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700636 index++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700637 }
Dan Magenheimer31677602011-09-21 11:56:28 -0400638 cleancache_invalidate_inode(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700639 return ret;
640}
641EXPORT_SYMBOL_GPL(invalidate_inode_pages2_range);
642
643/**
644 * invalidate_inode_pages2 - remove all pages from an address_space
Martin Waitz67be2dd2005-05-01 08:59:26 -0700645 * @mapping: the address_space
Linus Torvalds1da177e2005-04-16 15:20:36 -0700646 *
647 * Any pages which are found to be mapped into pagetables are unmapped prior to
648 * invalidation.
649 *
Peng Taoe9de25d2009-10-19 14:48:13 +0800650 * Returns -EBUSY if any pages could not be invalidated.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700651 */
652int invalidate_inode_pages2(struct address_space *mapping)
653{
654 return invalidate_inode_pages2_range(mapping, 0, -1);
655}
656EXPORT_SYMBOL_GPL(invalidate_inode_pages2);
npiggin@suse.de25d9e2d2009-08-21 02:35:05 +1000657
658/**
659 * truncate_pagecache - unmap and remove pagecache that has been truncated
660 * @inode: inode
Hugh Dickins8a549be2011-07-25 17:12:24 -0700661 * @newsize: new file size
npiggin@suse.de25d9e2d2009-08-21 02:35:05 +1000662 *
663 * inode's new i_size must already be written before truncate_pagecache
664 * is called.
665 *
666 * This function should typically be called before the filesystem
667 * releases resources associated with the freed range (eg. deallocates
668 * blocks). This way, pagecache will always stay logically coherent
669 * with on-disk format, and the filesystem would not have to deal with
670 * situations such as writepage being called for a page that has already
671 * had its underlying blocks deallocated.
672 */
Kirill A. Shutemov7caef262013-09-12 15:13:56 -0700673void truncate_pagecache(struct inode *inode, loff_t newsize)
npiggin@suse.de25d9e2d2009-08-21 02:35:05 +1000674{
OGAWA Hirofumicedabed2010-01-13 21:14:09 +0900675 struct address_space *mapping = inode->i_mapping;
Hugh Dickins8a549be2011-07-25 17:12:24 -0700676 loff_t holebegin = round_up(newsize, PAGE_SIZE);
npiggin@suse.de25d9e2d2009-08-21 02:35:05 +1000677
OGAWA Hirofumicedabed2010-01-13 21:14:09 +0900678 /*
679 * unmap_mapping_range is called twice, first simply for
680 * efficiency so that truncate_inode_pages does fewer
681 * single-page unmaps. However after this first call, and
682 * before truncate_inode_pages finishes, it is possible for
683 * private pages to be COWed, which remain after
684 * truncate_inode_pages finishes, hence the second
685 * unmap_mapping_range call must be made for correctness.
686 */
Hugh Dickins8a549be2011-07-25 17:12:24 -0700687 unmap_mapping_range(mapping, holebegin, 0, 1);
688 truncate_inode_pages(mapping, newsize);
689 unmap_mapping_range(mapping, holebegin, 0, 1);
npiggin@suse.de25d9e2d2009-08-21 02:35:05 +1000690}
691EXPORT_SYMBOL(truncate_pagecache);
692
693/**
Christoph Hellwig2c27c652010-06-04 11:30:04 +0200694 * truncate_setsize - update inode and pagecache for a new file size
695 * @inode: inode
696 * @newsize: new file size
697 *
Jan Kara382e27d2011-01-20 14:44:26 -0800698 * truncate_setsize updates i_size and performs pagecache truncation (if
699 * necessary) to @newsize. It will be typically be called from the filesystem's
700 * setattr function when ATTR_SIZE is passed in.
Christoph Hellwig2c27c652010-06-04 11:30:04 +0200701 *
Jan Kara77783d02014-11-07 08:29:25 +1100702 * Must be called with a lock serializing truncates and writes (generally
703 * i_mutex but e.g. xfs uses a different lock) and before all filesystem
704 * specific block truncation has been performed.
Christoph Hellwig2c27c652010-06-04 11:30:04 +0200705 */
706void truncate_setsize(struct inode *inode, loff_t newsize)
707{
Jan Kara90a80202014-10-01 21:49:18 -0400708 loff_t oldsize = inode->i_size;
709
Christoph Hellwig2c27c652010-06-04 11:30:04 +0200710 i_size_write(inode, newsize);
Jan Kara90a80202014-10-01 21:49:18 -0400711 if (newsize > oldsize)
712 pagecache_isize_extended(inode, oldsize, newsize);
Kirill A. Shutemov7caef262013-09-12 15:13:56 -0700713 truncate_pagecache(inode, newsize);
Christoph Hellwig2c27c652010-06-04 11:30:04 +0200714}
715EXPORT_SYMBOL(truncate_setsize);
716
717/**
Jan Kara90a80202014-10-01 21:49:18 -0400718 * pagecache_isize_extended - update pagecache after extension of i_size
719 * @inode: inode for which i_size was extended
720 * @from: original inode size
721 * @to: new inode size
722 *
723 * Handle extension of inode size either caused by extending truncate or by
724 * write starting after current i_size. We mark the page straddling current
725 * i_size RO so that page_mkwrite() is called on the nearest write access to
726 * the page. This way filesystem can be sure that page_mkwrite() is called on
727 * the page before user writes to the page via mmap after the i_size has been
728 * changed.
729 *
730 * The function must be called after i_size is updated so that page fault
731 * coming after we unlock the page will already see the new i_size.
732 * The function must be called while we still hold i_mutex - this not only
733 * makes sure i_size is stable but also that userspace cannot observe new
734 * i_size value before we are prepared to store mmap writes at new inode size.
735 */
736void pagecache_isize_extended(struct inode *inode, loff_t from, loff_t to)
737{
738 int bsize = 1 << inode->i_blkbits;
739 loff_t rounded_from;
740 struct page *page;
741 pgoff_t index;
742
Jan Kara90a80202014-10-01 21:49:18 -0400743 WARN_ON(to > inode->i_size);
744
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300745 if (from >= to || bsize == PAGE_SIZE)
Jan Kara90a80202014-10-01 21:49:18 -0400746 return;
747 /* Page straddling @from will not have any hole block created? */
748 rounded_from = round_up(from, bsize);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300749 if (to <= rounded_from || !(rounded_from & (PAGE_SIZE - 1)))
Jan Kara90a80202014-10-01 21:49:18 -0400750 return;
751
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300752 index = from >> PAGE_SHIFT;
Jan Kara90a80202014-10-01 21:49:18 -0400753 page = find_lock_page(inode->i_mapping, index);
754 /* Page not cached? Nothing to do */
755 if (!page)
756 return;
757 /*
758 * See clear_page_dirty_for_io() for details why set_page_dirty()
759 * is needed.
760 */
761 if (page_mkclean(page))
762 set_page_dirty(page);
763 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300764 put_page(page);
Jan Kara90a80202014-10-01 21:49:18 -0400765}
766EXPORT_SYMBOL(pagecache_isize_extended);
767
768/**
Hugh Dickins623e3db2012-03-28 14:42:40 -0700769 * truncate_pagecache_range - unmap and remove pagecache that is hole-punched
770 * @inode: inode
771 * @lstart: offset of beginning of hole
772 * @lend: offset of last byte of hole
773 *
774 * This function should typically be called before the filesystem
775 * releases resources associated with the freed range (eg. deallocates
776 * blocks). This way, pagecache will always stay logically coherent
777 * with on-disk format, and the filesystem would not have to deal with
778 * situations such as writepage being called for a page that has already
779 * had its underlying blocks deallocated.
780 */
781void truncate_pagecache_range(struct inode *inode, loff_t lstart, loff_t lend)
782{
783 struct address_space *mapping = inode->i_mapping;
784 loff_t unmap_start = round_up(lstart, PAGE_SIZE);
785 loff_t unmap_end = round_down(1 + lend, PAGE_SIZE) - 1;
786 /*
787 * This rounding is currently just for example: unmap_mapping_range
788 * expands its hole outwards, whereas we want it to contract the hole
789 * inwards. However, existing callers of truncate_pagecache_range are
Lukas Czerner5a720392013-05-27 23:32:35 -0400790 * doing their own page rounding first. Note that unmap_mapping_range
791 * allows holelen 0 for all, and we allow lend -1 for end of file.
Hugh Dickins623e3db2012-03-28 14:42:40 -0700792 */
793
794 /*
795 * Unlike in truncate_pagecache, unmap_mapping_range is called only
796 * once (before truncating pagecache), and without "even_cows" flag:
797 * hole-punching should not remove private COWed pages from the hole.
798 */
799 if ((u64)unmap_end > (u64)unmap_start)
800 unmap_mapping_range(mapping, unmap_start,
801 1 + unmap_end - unmap_start, 0);
802 truncate_inode_pages_range(mapping, lstart, lend);
803}
804EXPORT_SYMBOL(truncate_pagecache_range);