blob: d3a2737cc1885bdb7a1da636ed294ce6dc137201 [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 */
Hugh Dickins3a4f8a02017-02-24 14:59:36 -080023#include <linux/shmem_fs.h>
Dan Magenheimerc515e1f2011-05-26 10:01:43 -060024#include <linux/cleancache.h>
Jan Kara90a80202014-10-01 21:49:18 -040025#include <linux/rmap.h>
Rik van Rielba470de2008-10-18 20:26:50 -070026#include "internal.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070027
Jan Karac6dcf522016-08-10 17:22:44 +020028static void clear_shadow_entry(struct address_space *mapping, pgoff_t index,
29 void *entry)
Johannes Weiner0cd61442014-04-03 14:47:46 -070030{
Johannes Weiner449dd692014-04-03 14:47:56 -070031 struct radix_tree_node *node;
32 void **slot;
33
Jan Karaac401cc2016-05-12 18:29:18 +020034 spin_lock_irq(&mapping->tree_lock);
35 /*
36 * Regular page slots are stabilized by the page lock even
37 * without the tree itself locked. These unlocked entries
38 * need verification under the tree lock.
39 */
Johannes Weiner14b46872016-12-12 16:43:52 -080040 if (!__radix_tree_lookup(&mapping->page_tree, index, &node, &slot))
Jan Karaac401cc2016-05-12 18:29:18 +020041 goto unlock;
42 if (*slot != entry)
43 goto unlock;
Johannes Weiner14b46872016-12-12 16:43:52 -080044 __radix_tree_replace(&mapping->page_tree, node, slot, NULL,
45 workingset_update_node, mapping);
Jan Karaac401cc2016-05-12 18:29:18 +020046 mapping->nrexceptional--;
Johannes Weiner449dd692014-04-03 14:47:56 -070047unlock:
Johannes Weiner0cd61442014-04-03 14:47:46 -070048 spin_unlock_irq(&mapping->tree_lock);
49}
Linus Torvalds1da177e2005-04-16 15:20:36 -070050
Jan Karac6dcf522016-08-10 17:22:44 +020051/*
52 * Unconditionally remove exceptional entry. Usually called from truncate path.
53 */
54static void truncate_exceptional_entry(struct address_space *mapping,
55 pgoff_t index, void *entry)
56{
57 /* Handled by shmem itself */
58 if (shmem_mapping(mapping))
59 return;
60
61 if (dax_mapping(mapping)) {
62 dax_delete_mapping_entry(mapping, index);
63 return;
64 }
65 clear_shadow_entry(mapping, index, entry);
66}
67
68/*
69 * Invalidate exceptional entry if easily possible. This handles exceptional
Ross Zwisler4636e702017-05-12 15:46:47 -070070 * entries for invalidate_inode_pages().
Jan Karac6dcf522016-08-10 17:22:44 +020071 */
72static int invalidate_exceptional_entry(struct address_space *mapping,
73 pgoff_t index, void *entry)
74{
Ross Zwisler4636e702017-05-12 15:46:47 -070075 /* Handled by shmem itself, or for DAX we do nothing. */
76 if (shmem_mapping(mapping) || dax_mapping(mapping))
Jan Karac6dcf522016-08-10 17:22:44 +020077 return 1;
Jan Karac6dcf522016-08-10 17:22:44 +020078 clear_shadow_entry(mapping, index, entry);
79 return 1;
80}
81
82/*
83 * Invalidate exceptional entry if clean. This handles exceptional entries for
84 * invalidate_inode_pages2() so for DAX it evicts only clean entries.
85 */
86static int invalidate_exceptional_entry2(struct address_space *mapping,
87 pgoff_t index, void *entry)
88{
89 /* Handled by shmem itself */
90 if (shmem_mapping(mapping))
91 return 1;
92 if (dax_mapping(mapping))
93 return dax_invalidate_mapping_entry_sync(mapping, index);
94 clear_shadow_entry(mapping, index, entry);
95 return 1;
96}
97
David Howellscf9a2ae2006-08-29 19:05:54 +010098/**
Fengguang Wu28bc44d2008-02-03 18:04:10 +020099 * do_invalidatepage - invalidate part or all of a page
David Howellscf9a2ae2006-08-29 19:05:54 +0100100 * @page: the page which is affected
Lukas Czernerd47992f2013-05-21 23:17:23 -0400101 * @offset: start of the range to invalidate
102 * @length: length of the range to invalidate
David Howellscf9a2ae2006-08-29 19:05:54 +0100103 *
104 * do_invalidatepage() is called when all or part of the page has become
105 * invalidated by a truncate operation.
106 *
107 * do_invalidatepage() does not have to release all buffers, but it must
108 * ensure that no dirty buffer is left outside @offset and that no I/O
109 * is underway against any of the blocks which are outside the truncation
110 * point. Because the caller is about to free (and possibly reuse) those
111 * blocks on-disk.
112 */
Lukas Czernerd47992f2013-05-21 23:17:23 -0400113void do_invalidatepage(struct page *page, unsigned int offset,
114 unsigned int length)
David Howellscf9a2ae2006-08-29 19:05:54 +0100115{
Lukas Czernerd47992f2013-05-21 23:17:23 -0400116 void (*invalidatepage)(struct page *, unsigned int, unsigned int);
117
David Howellscf9a2ae2006-08-29 19:05:54 +0100118 invalidatepage = page->mapping->a_ops->invalidatepage;
David Howells93614012006-09-30 20:45:40 +0200119#ifdef CONFIG_BLOCK
David Howellscf9a2ae2006-08-29 19:05:54 +0100120 if (!invalidatepage)
121 invalidatepage = block_invalidatepage;
David Howells93614012006-09-30 20:45:40 +0200122#endif
David Howellscf9a2ae2006-08-29 19:05:54 +0100123 if (invalidatepage)
Lukas Czernerd47992f2013-05-21 23:17:23 -0400124 (*invalidatepage)(page, offset, length);
David Howellscf9a2ae2006-08-29 19:05:54 +0100125}
126
Linus Torvaldsecdfc972007-01-26 12:47:06 -0800127/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700128 * If truncate cannot remove the fs-private metadata from the page, the page
Shaohua Li62e1c552008-02-04 22:29:33 -0800129 * becomes orphaned. It will be left on the LRU and may even be mapped into
Nick Piggin54cb8822007-07-19 01:46:59 -0700130 * user pagetables if we're racing with filemap_fault().
Linus Torvalds1da177e2005-04-16 15:20:36 -0700131 *
132 * We need to bale out if page->mapping is no longer equal to the original
133 * mapping. This happens a) when the VM reclaimed the page while we waited on
Andrew Mortonfc0ecff2007-02-10 01:45:39 -0800134 * its lock, b) when a concurrent invalidate_mapping_pages got there first and
Linus Torvalds1da177e2005-04-16 15:20:36 -0700135 * c) when tmpfs swizzles a page between a tmpfs inode and swapper_space.
136 */
Nick Piggin750b4982009-09-16 11:50:12 +0200137static int
Linus Torvalds1da177e2005-04-16 15:20:36 -0700138truncate_complete_page(struct address_space *mapping, struct page *page)
139{
140 if (page->mapping != mapping)
Nick Piggin750b4982009-09-16 11:50:12 +0200141 return -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700142
David Howells266cf652009-04-03 16:42:36 +0100143 if (page_has_private(page))
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300144 do_invalidatepage(page, 0, PAGE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700145
Konstantin Khlebnikovb9ea2512015-04-14 15:45:27 -0700146 /*
147 * Some filesystems seem to re-dirty the page even after
148 * the VM has canceled the dirty bit (eg ext3 journaling).
149 * Hence dirty accounting check is placed after invalidation.
150 */
Tejun Heo11f81be2015-05-22 17:13:15 -0400151 cancel_dirty_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700152 ClearPageMappedToDisk(page);
Minchan Kim5adc7b52011-03-22 16:32:41 -0700153 delete_from_page_cache(page);
Nick Piggin750b4982009-09-16 11:50:12 +0200154 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700155}
156
157/*
Andrew Mortonfc0ecff2007-02-10 01:45:39 -0800158 * This is for invalidate_mapping_pages(). That function can be called at
Linus Torvalds1da177e2005-04-16 15:20:36 -0700159 * any time, and is not supposed to throw away dirty pages. But pages can
Nick Piggin0fd0e6b2006-09-27 01:50:02 -0700160 * be marked dirty at any time too, so use remove_mapping which safely
161 * discards clean, unused pages.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700162 *
163 * Returns non-zero if the page was successfully invalidated.
164 */
165static int
166invalidate_complete_page(struct address_space *mapping, struct page *page)
167{
Nick Piggin0fd0e6b2006-09-27 01:50:02 -0700168 int ret;
169
Linus Torvalds1da177e2005-04-16 15:20:36 -0700170 if (page->mapping != mapping)
171 return 0;
172
David Howells266cf652009-04-03 16:42:36 +0100173 if (page_has_private(page) && !try_to_release_page(page, 0))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700174 return 0;
175
Nick Piggin0fd0e6b2006-09-27 01:50:02 -0700176 ret = remove_mapping(mapping, page);
Nick Piggin0fd0e6b2006-09-27 01:50:02 -0700177
178 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700179}
180
Nick Piggin750b4982009-09-16 11:50:12 +0200181int truncate_inode_page(struct address_space *mapping, struct page *page)
182{
Kirill A. Shutemovfc127da2016-07-26 15:26:07 -0700183 loff_t holelen;
184 VM_BUG_ON_PAGE(PageTail(page), page);
185
186 holelen = PageTransHuge(page) ? HPAGE_PMD_SIZE : PAGE_SIZE;
Nick Piggin750b4982009-09-16 11:50:12 +0200187 if (page_mapped(page)) {
188 unmap_mapping_range(mapping,
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300189 (loff_t)page->index << PAGE_SHIFT,
Kirill A. Shutemovfc127da2016-07-26 15:26:07 -0700190 holelen, 0);
Nick Piggin750b4982009-09-16 11:50:12 +0200191 }
192 return truncate_complete_page(mapping, page);
193}
194
Wu Fengguang83f78662009-09-16 11:50:13 +0200195/*
Andi Kleen25718732009-09-16 11:50:13 +0200196 * Used to get rid of pages on hardware memory corruption.
197 */
198int generic_error_remove_page(struct address_space *mapping, struct page *page)
199{
200 if (!mapping)
201 return -EINVAL;
202 /*
203 * Only punch for normal data pages for now.
204 * Handling other types like directories would need more auditing.
205 */
206 if (!S_ISREG(mapping->host->i_mode))
207 return -EIO;
208 return truncate_inode_page(mapping, page);
209}
210EXPORT_SYMBOL(generic_error_remove_page);
211
212/*
Wu Fengguang83f78662009-09-16 11:50:13 +0200213 * Safely invalidate one page from its pagecache mapping.
214 * It only drops clean, unused pages. The page must be locked.
215 *
216 * Returns 1 if the page is successfully invalidated, otherwise 0.
217 */
218int invalidate_inode_page(struct page *page)
219{
220 struct address_space *mapping = page_mapping(page);
221 if (!mapping)
222 return 0;
223 if (PageDirty(page) || PageWriteback(page))
224 return 0;
225 if (page_mapped(page))
226 return 0;
227 return invalidate_complete_page(mapping, page);
228}
229
Linus Torvalds1da177e2005-04-16 15:20:36 -0700230/**
Liu Bo73c1e202012-02-21 10:57:20 +0800231 * truncate_inode_pages_range - truncate range of pages specified by start & end byte offsets
Linus Torvalds1da177e2005-04-16 15:20:36 -0700232 * @mapping: mapping to truncate
233 * @lstart: offset from which to truncate
Lukas Czerner5a720392013-05-27 23:32:35 -0400234 * @lend: offset to which to truncate (inclusive)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700235 *
Hans Reiserd7339072006-01-06 00:10:36 -0800236 * Truncate the page cache, removing the pages that are between
Lukas Czerner5a720392013-05-27 23:32:35 -0400237 * specified offsets (and zeroing out partial pages
238 * if lstart or lend + 1 is not page aligned).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700239 *
240 * Truncate takes two passes - the first pass is nonblocking. It will not
241 * block on page locks and it will not block on writeback. The second pass
242 * will wait. This is to prevent as much IO as possible in the affected region.
243 * The first pass will remove most pages, so the search cost of the second pass
244 * is low.
245 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700246 * We pass down the cache-hot hint to the page freeing code. Even if the
247 * mapping is large, it is probably the case that the final pages are the most
248 * recently touched, and freeing happens in ascending file offset order.
Lukas Czerner5a720392013-05-27 23:32:35 -0400249 *
250 * Note that since ->invalidatepage() accepts range to invalidate
251 * truncate_inode_pages_range is able to handle cases where lend + 1 is not
252 * page aligned properly.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700253 */
Hans Reiserd7339072006-01-06 00:10:36 -0800254void truncate_inode_pages_range(struct address_space *mapping,
255 loff_t lstart, loff_t lend)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700256{
Lukas Czerner5a720392013-05-27 23:32:35 -0400257 pgoff_t start; /* inclusive */
258 pgoff_t end; /* exclusive */
259 unsigned int partial_start; /* inclusive */
260 unsigned int partial_end; /* exclusive */
261 struct pagevec pvec;
Johannes Weiner0cd61442014-04-03 14:47:46 -0700262 pgoff_t indices[PAGEVEC_SIZE];
Lukas Czerner5a720392013-05-27 23:32:35 -0400263 pgoff_t index;
264 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700265
Ross Zwislerf9fe48b2016-01-22 15:10:40 -0800266 if (mapping->nrpages == 0 && mapping->nrexceptional == 0)
Andrey Ryabinin34ccb692017-05-03 14:56:09 -0700267 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700268
Lukas Czerner5a720392013-05-27 23:32:35 -0400269 /* Offsets within partial pages */
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300270 partial_start = lstart & (PAGE_SIZE - 1);
271 partial_end = (lend + 1) & (PAGE_SIZE - 1);
Lukas Czerner5a720392013-05-27 23:32:35 -0400272
273 /*
274 * 'start' and 'end' always covers the range of pages to be fully
275 * truncated. Partial pages are covered with 'partial_start' at the
276 * start of the range and 'partial_end' at the end of the range.
277 * Note that 'end' is exclusive while 'lend' is inclusive.
278 */
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300279 start = (lstart + PAGE_SIZE - 1) >> PAGE_SHIFT;
Lukas Czerner5a720392013-05-27 23:32:35 -0400280 if (lend == -1)
281 /*
282 * lend == -1 indicates end-of-file so we have to set 'end'
283 * to the highest possible pgoff_t and since the type is
284 * unsigned we're using -1.
285 */
286 end = -1;
287 else
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300288 end = (lend + 1) >> PAGE_SHIFT;
Hans Reiserd7339072006-01-06 00:10:36 -0800289
Linus Torvalds1da177e2005-04-16 15:20:36 -0700290 pagevec_init(&pvec, 0);
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700291 index = start;
Johannes Weiner0cd61442014-04-03 14:47:46 -0700292 while (index < end && pagevec_lookup_entries(&pvec, mapping, index,
293 min(end - index, (pgoff_t)PAGEVEC_SIZE),
294 indices)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700295 for (i = 0; i < pagevec_count(&pvec); i++) {
296 struct page *page = pvec.pages[i];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700297
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700298 /* We rely upon deletion not changing page->index */
Johannes Weiner0cd61442014-04-03 14:47:46 -0700299 index = indices[i];
Lukas Czerner5a720392013-05-27 23:32:35 -0400300 if (index >= end)
Hans Reiserd7339072006-01-06 00:10:36 -0800301 break;
Hans Reiserd7339072006-01-06 00:10:36 -0800302
Johannes Weiner0cd61442014-04-03 14:47:46 -0700303 if (radix_tree_exceptional_entry(page)) {
Jan Karac6dcf522016-08-10 17:22:44 +0200304 truncate_exceptional_entry(mapping, index,
305 page);
Johannes Weiner0cd61442014-04-03 14:47:46 -0700306 continue;
307 }
308
Nick Piggin529ae9a2008-08-02 12:01:03 +0200309 if (!trylock_page(page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700310 continue;
Kirill A. Shutemov5cbc1982016-11-30 15:54:19 -0800311 WARN_ON(page_to_index(page) != index);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700312 if (PageWriteback(page)) {
313 unlock_page(page);
314 continue;
315 }
Nick Piggin750b4982009-09-16 11:50:12 +0200316 truncate_inode_page(mapping, page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700317 unlock_page(page);
318 }
Johannes Weiner0cd61442014-04-03 14:47:46 -0700319 pagevec_remove_exceptionals(&pvec);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700320 pagevec_release(&pvec);
321 cond_resched();
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700322 index++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700323 }
324
Lukas Czerner5a720392013-05-27 23:32:35 -0400325 if (partial_start) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700326 struct page *page = find_lock_page(mapping, start - 1);
327 if (page) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300328 unsigned int top = PAGE_SIZE;
Lukas Czerner5a720392013-05-27 23:32:35 -0400329 if (start > end) {
330 /* Truncation within a single page */
331 top = partial_end;
332 partial_end = 0;
333 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700334 wait_on_page_writeback(page);
Lukas Czerner5a720392013-05-27 23:32:35 -0400335 zero_user_segment(page, partial_start, top);
336 cleancache_invalidate_page(mapping, page);
337 if (page_has_private(page))
338 do_invalidatepage(page, partial_start,
339 top - partial_start);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700340 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300341 put_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700342 }
343 }
Lukas Czerner5a720392013-05-27 23:32:35 -0400344 if (partial_end) {
345 struct page *page = find_lock_page(mapping, end);
346 if (page) {
347 wait_on_page_writeback(page);
348 zero_user_segment(page, 0, partial_end);
349 cleancache_invalidate_page(mapping, page);
350 if (page_has_private(page))
351 do_invalidatepage(page, 0,
352 partial_end);
353 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300354 put_page(page);
Lukas Czerner5a720392013-05-27 23:32:35 -0400355 }
356 }
357 /*
358 * If the truncation happened within a single page no pages
359 * will be released, just zeroed, so we can bail out now.
360 */
361 if (start >= end)
Andrey Ryabinin34ccb692017-05-03 14:56:09 -0700362 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700363
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700364 index = start;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700365 for ( ; ; ) {
366 cond_resched();
Johannes Weiner0cd61442014-04-03 14:47:46 -0700367 if (!pagevec_lookup_entries(&pvec, mapping, index,
Hugh Dickins792ceae2014-07-23 14:00:15 -0700368 min(end - index, (pgoff_t)PAGEVEC_SIZE), indices)) {
369 /* If all gone from start onwards, we're done */
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700370 if (index == start)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700371 break;
Hugh Dickins792ceae2014-07-23 14:00:15 -0700372 /* Otherwise restart to make sure all gone */
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700373 index = start;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700374 continue;
375 }
Johannes Weiner0cd61442014-04-03 14:47:46 -0700376 if (index == start && indices[0] >= end) {
Hugh Dickins792ceae2014-07-23 14:00:15 -0700377 /* All gone out of hole to be punched, we're done */
Johannes Weiner0cd61442014-04-03 14:47:46 -0700378 pagevec_remove_exceptionals(&pvec);
Hans Reiserd7339072006-01-06 00:10:36 -0800379 pagevec_release(&pvec);
380 break;
381 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700382 for (i = 0; i < pagevec_count(&pvec); i++) {
383 struct page *page = pvec.pages[i];
384
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700385 /* We rely upon deletion not changing page->index */
Johannes Weiner0cd61442014-04-03 14:47:46 -0700386 index = indices[i];
Hugh Dickins792ceae2014-07-23 14:00:15 -0700387 if (index >= end) {
388 /* Restart punch to make sure all gone */
389 index = start - 1;
Hans Reiserd7339072006-01-06 00:10:36 -0800390 break;
Hugh Dickins792ceae2014-07-23 14:00:15 -0700391 }
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700392
Johannes Weiner0cd61442014-04-03 14:47:46 -0700393 if (radix_tree_exceptional_entry(page)) {
Jan Karac6dcf522016-08-10 17:22:44 +0200394 truncate_exceptional_entry(mapping, index,
395 page);
Johannes Weiner0cd61442014-04-03 14:47:46 -0700396 continue;
397 }
398
Linus Torvalds1da177e2005-04-16 15:20:36 -0700399 lock_page(page);
Kirill A. Shutemov5cbc1982016-11-30 15:54:19 -0800400 WARN_ON(page_to_index(page) != index);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700401 wait_on_page_writeback(page);
Nick Piggin750b4982009-09-16 11:50:12 +0200402 truncate_inode_page(mapping, page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700403 unlock_page(page);
404 }
Johannes Weiner0cd61442014-04-03 14:47:46 -0700405 pagevec_remove_exceptionals(&pvec);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700406 pagevec_release(&pvec);
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700407 index++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700408 }
Andrey Ryabinin34ccb692017-05-03 14:56:09 -0700409
410out:
Dan Magenheimer31677602011-09-21 11:56:28 -0400411 cleancache_invalidate_inode(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700412}
Hans Reiserd7339072006-01-06 00:10:36 -0800413EXPORT_SYMBOL(truncate_inode_pages_range);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700414
Hans Reiserd7339072006-01-06 00:10:36 -0800415/**
416 * truncate_inode_pages - truncate *all* the pages from an offset
417 * @mapping: mapping to truncate
418 * @lstart: offset from which to truncate
419 *
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800420 * Called under (and serialised by) inode->i_mutex.
Jan Kara08142572011-06-27 16:18:10 -0700421 *
422 * Note: When this function returns, there can be a page in the process of
423 * deletion (inside __delete_from_page_cache()) in the specified range. Thus
424 * mapping->nrpages can be non-zero when this function returns even after
425 * truncation of the whole mapping.
Hans Reiserd7339072006-01-06 00:10:36 -0800426 */
427void truncate_inode_pages(struct address_space *mapping, loff_t lstart)
428{
429 truncate_inode_pages_range(mapping, lstart, (loff_t)-1);
430}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700431EXPORT_SYMBOL(truncate_inode_pages);
432
Mike Waychison28697352009-06-16 15:32:59 -0700433/**
Johannes Weiner91b0abe2014-04-03 14:47:49 -0700434 * truncate_inode_pages_final - truncate *all* pages before inode dies
435 * @mapping: mapping to truncate
436 *
437 * Called under (and serialized by) inode->i_mutex.
438 *
439 * Filesystems have to use this in the .evict_inode path to inform the
440 * VM that this is the final truncate and the inode is going away.
441 */
442void truncate_inode_pages_final(struct address_space *mapping)
443{
Ross Zwislerf9fe48b2016-01-22 15:10:40 -0800444 unsigned long nrexceptional;
Johannes Weiner91b0abe2014-04-03 14:47:49 -0700445 unsigned long nrpages;
446
447 /*
448 * Page reclaim can not participate in regular inode lifetime
449 * management (can't call iput()) and thus can race with the
450 * inode teardown. Tell it when the address space is exiting,
451 * so that it does not install eviction information after the
452 * final truncate has begun.
453 */
454 mapping_set_exiting(mapping);
455
456 /*
457 * When reclaim installs eviction entries, it increases
Ross Zwislerf9fe48b2016-01-22 15:10:40 -0800458 * nrexceptional first, then decreases nrpages. Make sure we see
Johannes Weiner91b0abe2014-04-03 14:47:49 -0700459 * this in the right order or we might miss an entry.
460 */
461 nrpages = mapping->nrpages;
462 smp_rmb();
Ross Zwislerf9fe48b2016-01-22 15:10:40 -0800463 nrexceptional = mapping->nrexceptional;
Johannes Weiner91b0abe2014-04-03 14:47:49 -0700464
Ross Zwislerf9fe48b2016-01-22 15:10:40 -0800465 if (nrpages || nrexceptional) {
Johannes Weiner91b0abe2014-04-03 14:47:49 -0700466 /*
467 * As truncation uses a lockless tree lookup, cycle
468 * the tree lock to make sure any ongoing tree
469 * modification that does not see AS_EXITING is
470 * completed before starting the final truncate.
471 */
472 spin_lock_irq(&mapping->tree_lock);
473 spin_unlock_irq(&mapping->tree_lock);
Johannes Weiner91b0abe2014-04-03 14:47:49 -0700474 }
Pavel Tikhomirov60720df2018-11-30 14:09:00 -0800475
476 /*
477 * Cleancache needs notification even if there are no pages or shadow
478 * entries.
479 */
480 truncate_inode_pages(mapping, 0);
Johannes Weiner91b0abe2014-04-03 14:47:49 -0700481}
482EXPORT_SYMBOL(truncate_inode_pages_final);
483
484/**
Mike Waychison28697352009-06-16 15:32:59 -0700485 * invalidate_mapping_pages - Invalidate all the unlocked pages of one inode
486 * @mapping: the address_space which holds the pages to invalidate
487 * @start: the offset 'from' which to invalidate
488 * @end: the offset 'to' which to invalidate (inclusive)
489 *
490 * This function only removes the unlocked pages, if you want to
491 * remove all the pages of one inode, you must call truncate_inode_pages.
492 *
493 * invalidate_mapping_pages() will not block on IO activity. It will not
494 * invalidate pages which are dirty, locked, under writeback or mapped into
495 * pagetables.
496 */
497unsigned long invalidate_mapping_pages(struct address_space *mapping,
Minchan Kim31560182011-03-22 16:32:52 -0700498 pgoff_t start, pgoff_t end)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700499{
Johannes Weiner0cd61442014-04-03 14:47:46 -0700500 pgoff_t indices[PAGEVEC_SIZE];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700501 struct pagevec pvec;
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700502 pgoff_t index = start;
Minchan Kim31560182011-03-22 16:32:52 -0700503 unsigned long ret;
504 unsigned long count = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700505 int i;
506
507 pagevec_init(&pvec, 0);
Johannes Weiner0cd61442014-04-03 14:47:46 -0700508 while (index <= end && pagevec_lookup_entries(&pvec, mapping, index,
509 min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1,
510 indices)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700511 for (i = 0; i < pagevec_count(&pvec); i++) {
512 struct page *page = pvec.pages[i];
513
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700514 /* We rely upon deletion not changing page->index */
Johannes Weiner0cd61442014-04-03 14:47:46 -0700515 index = indices[i];
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700516 if (index > end)
517 break;
NeilBrowne0f23602006-06-23 02:05:48 -0700518
Johannes Weiner0cd61442014-04-03 14:47:46 -0700519 if (radix_tree_exceptional_entry(page)) {
Jan Karac6dcf522016-08-10 17:22:44 +0200520 invalidate_exceptional_entry(mapping, index,
521 page);
Johannes Weiner0cd61442014-04-03 14:47:46 -0700522 continue;
523 }
524
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700525 if (!trylock_page(page))
526 continue;
Kirill A. Shutemovfc127da2016-07-26 15:26:07 -0700527
Kirill A. Shutemov5cbc1982016-11-30 15:54:19 -0800528 WARN_ON(page_to_index(page) != index);
Kirill A. Shutemovfc127da2016-07-26 15:26:07 -0700529
530 /* Middle of THP: skip */
531 if (PageTransTail(page)) {
532 unlock_page(page);
533 continue;
534 } else if (PageTransHuge(page)) {
535 index += HPAGE_PMD_NR - 1;
536 i += HPAGE_PMD_NR - 1;
Jan Kara76b6f9b2017-07-10 15:48:59 -0700537 /*
538 * 'end' is in the middle of THP. Don't
539 * invalidate the page as the part outside of
540 * 'end' could be still useful.
541 */
542 if (index > end) {
543 unlock_page(page);
Kirill A. Shutemovfc127da2016-07-26 15:26:07 -0700544 continue;
Jan Kara76b6f9b2017-07-10 15:48:59 -0700545 }
Kirill A. Shutemovfc127da2016-07-26 15:26:07 -0700546 }
547
Minchan Kim31560182011-03-22 16:32:52 -0700548 ret = invalidate_inode_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700549 unlock_page(page);
Minchan Kim31560182011-03-22 16:32:52 -0700550 /*
551 * Invalidation is a hint that the page is no longer
552 * of interest and try to speed up its reclaim.
553 */
554 if (!ret)
Minchan Kimcc5993b2015-04-15 16:13:26 -0700555 deactivate_file_page(page);
Minchan Kim31560182011-03-22 16:32:52 -0700556 count += ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700557 }
Johannes Weiner0cd61442014-04-03 14:47:46 -0700558 pagevec_remove_exceptionals(&pvec);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700559 pagevec_release(&pvec);
Mike Waychison28697352009-06-16 15:32:59 -0700560 cond_resched();
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700561 index++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700562 }
Minchan Kim31560182011-03-22 16:32:52 -0700563 return count;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700564}
Anton Altaparmakov54bc4852007-02-10 01:45:38 -0800565EXPORT_SYMBOL(invalidate_mapping_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700566
Andrew Mortonbd4c8ce2006-09-30 23:29:29 -0700567/*
568 * This is like invalidate_complete_page(), except it ignores the page's
569 * refcount. We do this because invalidate_inode_pages2() needs stronger
570 * invalidation guarantees, and cannot afford to leave pages behind because
Anderson Briglia2706a1b2007-07-15 23:38:09 -0700571 * shrink_page_list() has a temp ref on them, or because they're transiently
572 * sitting in the lru_cache_add() pagevecs.
Andrew Mortonbd4c8ce2006-09-30 23:29:29 -0700573 */
574static int
575invalidate_complete_page2(struct address_space *mapping, struct page *page)
576{
Greg Thelenc4843a72015-05-22 17:13:16 -0400577 unsigned long flags;
578
Andrew Mortonbd4c8ce2006-09-30 23:29:29 -0700579 if (page->mapping != mapping)
580 return 0;
581
David Howells266cf652009-04-03 16:42:36 +0100582 if (page_has_private(page) && !try_to_release_page(page, GFP_KERNEL))
Andrew Mortonbd4c8ce2006-09-30 23:29:29 -0700583 return 0;
584
Greg Thelenc4843a72015-05-22 17:13:16 -0400585 spin_lock_irqsave(&mapping->tree_lock, flags);
Andrew Mortonbd4c8ce2006-09-30 23:29:29 -0700586 if (PageDirty(page))
587 goto failed;
588
David Howells266cf652009-04-03 16:42:36 +0100589 BUG_ON(page_has_private(page));
Johannes Weiner62cccb82016-03-15 14:57:22 -0700590 __delete_from_page_cache(page, NULL);
Greg Thelenc4843a72015-05-22 17:13:16 -0400591 spin_unlock_irqrestore(&mapping->tree_lock, flags);
Linus Torvalds6072d132010-12-01 13:35:19 -0500592
593 if (mapping->a_ops->freepage)
594 mapping->a_ops->freepage(page);
595
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300596 put_page(page); /* pagecache ref */
Andrew Mortonbd4c8ce2006-09-30 23:29:29 -0700597 return 1;
598failed:
Greg Thelenc4843a72015-05-22 17:13:16 -0400599 spin_unlock_irqrestore(&mapping->tree_lock, flags);
Andrew Mortonbd4c8ce2006-09-30 23:29:29 -0700600 return 0;
601}
602
Trond Myklebuste3db7692007-01-10 23:15:39 -0800603static int do_launder_page(struct address_space *mapping, struct page *page)
604{
605 if (!PageDirty(page))
606 return 0;
607 if (page->mapping != mapping || mapping->a_ops->launder_page == NULL)
608 return 0;
609 return mapping->a_ops->launder_page(page);
610}
611
Linus Torvalds1da177e2005-04-16 15:20:36 -0700612/**
613 * invalidate_inode_pages2_range - remove range of pages from an address_space
Martin Waitz67be2dd2005-05-01 08:59:26 -0700614 * @mapping: the address_space
Linus Torvalds1da177e2005-04-16 15:20:36 -0700615 * @start: the page offset 'from' which to invalidate
616 * @end: the page offset 'to' which to invalidate (inclusive)
617 *
618 * Any pages which are found to be mapped into pagetables are unmapped prior to
619 * invalidation.
620 *
Hisashi Hifumi6ccfa802008-09-02 14:35:40 -0700621 * Returns -EBUSY if any pages could not be invalidated.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700622 */
623int invalidate_inode_pages2_range(struct address_space *mapping,
624 pgoff_t start, pgoff_t end)
625{
Johannes Weiner0cd61442014-04-03 14:47:46 -0700626 pgoff_t indices[PAGEVEC_SIZE];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700627 struct pagevec pvec;
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700628 pgoff_t index;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700629 int i;
630 int ret = 0;
Hisashi Hifumi0dd13342008-04-28 02:12:08 -0700631 int ret2 = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700632 int did_range_unmap = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700633
Andrey Ryabinin32691f02017-05-03 14:56:06 -0700634 if (mapping->nrpages == 0 && mapping->nrexceptional == 0)
Andrey Ryabinin34ccb692017-05-03 14:56:09 -0700635 goto out;
Andrey Ryabinin32691f02017-05-03 14:56:06 -0700636
Linus Torvalds1da177e2005-04-16 15:20:36 -0700637 pagevec_init(&pvec, 0);
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700638 index = start;
Johannes Weiner0cd61442014-04-03 14:47:46 -0700639 while (index <= end && pagevec_lookup_entries(&pvec, mapping, index,
640 min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1,
641 indices)) {
Trond Myklebust7b965e02007-02-28 20:13:55 -0800642 for (i = 0; i < pagevec_count(&pvec); i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700643 struct page *page = pvec.pages[i];
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700644
645 /* We rely upon deletion not changing page->index */
Johannes Weiner0cd61442014-04-03 14:47:46 -0700646 index = indices[i];
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700647 if (index > end)
648 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700649
Johannes Weiner0cd61442014-04-03 14:47:46 -0700650 if (radix_tree_exceptional_entry(page)) {
Jan Karac6dcf522016-08-10 17:22:44 +0200651 if (!invalidate_exceptional_entry2(mapping,
652 index, page))
653 ret = -EBUSY;
Johannes Weiner0cd61442014-04-03 14:47:46 -0700654 continue;
655 }
656
Linus Torvalds1da177e2005-04-16 15:20:36 -0700657 lock_page(page);
Kirill A. Shutemov5cbc1982016-11-30 15:54:19 -0800658 WARN_ON(page_to_index(page) != index);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700659 if (page->mapping != mapping) {
660 unlock_page(page);
661 continue;
662 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700663 wait_on_page_writeback(page);
Nick Piggind00806b2007-07-19 01:46:57 -0700664 if (page_mapped(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700665 if (!did_range_unmap) {
666 /*
667 * Zap the rest of the file in one hit.
668 */
669 unmap_mapping_range(mapping,
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300670 (loff_t)index << PAGE_SHIFT,
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700671 (loff_t)(1 + end - index)
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300672 << PAGE_SHIFT,
673 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700674 did_range_unmap = 1;
675 } else {
676 /*
677 * Just zap this page
678 */
679 unmap_mapping_range(mapping,
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300680 (loff_t)index << PAGE_SHIFT,
681 PAGE_SIZE, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700682 }
683 }
Nick Piggind00806b2007-07-19 01:46:57 -0700684 BUG_ON(page_mapped(page));
Hisashi Hifumi0dd13342008-04-28 02:12:08 -0700685 ret2 = do_launder_page(mapping, page);
686 if (ret2 == 0) {
687 if (!invalidate_complete_page2(mapping, page))
Hisashi Hifumi6ccfa802008-09-02 14:35:40 -0700688 ret2 = -EBUSY;
Hisashi Hifumi0dd13342008-04-28 02:12:08 -0700689 }
690 if (ret2 < 0)
691 ret = ret2;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700692 unlock_page(page);
693 }
Johannes Weiner0cd61442014-04-03 14:47:46 -0700694 pagevec_remove_exceptionals(&pvec);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700695 pagevec_release(&pvec);
696 cond_resched();
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700697 index++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700698 }
Jan Karacd656372017-05-12 15:46:50 -0700699 /*
700 * For DAX we invalidate page tables after invalidating radix tree. We
701 * could invalidate page tables while invalidating each entry however
702 * that would be expensive. And doing range unmapping before doesn't
703 * work as we have no cheap way to find whether radix tree entry didn't
704 * get remapped later.
705 */
706 if (dax_mapping(mapping)) {
707 unmap_mapping_range(mapping, (loff_t)start << PAGE_SHIFT,
708 (loff_t)(end - start + 1) << PAGE_SHIFT, 0);
709 }
Andrey Ryabinin34ccb692017-05-03 14:56:09 -0700710out:
Dan Magenheimer31677602011-09-21 11:56:28 -0400711 cleancache_invalidate_inode(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700712 return ret;
713}
714EXPORT_SYMBOL_GPL(invalidate_inode_pages2_range);
715
716/**
717 * invalidate_inode_pages2 - remove all pages from an address_space
Martin Waitz67be2dd2005-05-01 08:59:26 -0700718 * @mapping: the address_space
Linus Torvalds1da177e2005-04-16 15:20:36 -0700719 *
720 * Any pages which are found to be mapped into pagetables are unmapped prior to
721 * invalidation.
722 *
Peng Taoe9de25d2009-10-19 14:48:13 +0800723 * Returns -EBUSY if any pages could not be invalidated.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700724 */
725int invalidate_inode_pages2(struct address_space *mapping)
726{
727 return invalidate_inode_pages2_range(mapping, 0, -1);
728}
729EXPORT_SYMBOL_GPL(invalidate_inode_pages2);
npiggin@suse.de25d9e2d2009-08-21 02:35:05 +1000730
731/**
732 * truncate_pagecache - unmap and remove pagecache that has been truncated
733 * @inode: inode
Hugh Dickins8a549be2011-07-25 17:12:24 -0700734 * @newsize: new file size
npiggin@suse.de25d9e2d2009-08-21 02:35:05 +1000735 *
736 * inode's new i_size must already be written before truncate_pagecache
737 * is called.
738 *
739 * This function should typically be called before the filesystem
740 * releases resources associated with the freed range (eg. deallocates
741 * blocks). This way, pagecache will always stay logically coherent
742 * with on-disk format, and the filesystem would not have to deal with
743 * situations such as writepage being called for a page that has already
744 * had its underlying blocks deallocated.
745 */
Kirill A. Shutemov7caef262013-09-12 15:13:56 -0700746void truncate_pagecache(struct inode *inode, loff_t newsize)
npiggin@suse.de25d9e2d2009-08-21 02:35:05 +1000747{
OGAWA Hirofumicedabed2010-01-13 21:14:09 +0900748 struct address_space *mapping = inode->i_mapping;
Hugh Dickins8a549be2011-07-25 17:12:24 -0700749 loff_t holebegin = round_up(newsize, PAGE_SIZE);
npiggin@suse.de25d9e2d2009-08-21 02:35:05 +1000750
OGAWA Hirofumicedabed2010-01-13 21:14:09 +0900751 /*
752 * unmap_mapping_range is called twice, first simply for
753 * efficiency so that truncate_inode_pages does fewer
754 * single-page unmaps. However after this first call, and
755 * before truncate_inode_pages finishes, it is possible for
756 * private pages to be COWed, which remain after
757 * truncate_inode_pages finishes, hence the second
758 * unmap_mapping_range call must be made for correctness.
759 */
Hugh Dickins8a549be2011-07-25 17:12:24 -0700760 unmap_mapping_range(mapping, holebegin, 0, 1);
761 truncate_inode_pages(mapping, newsize);
762 unmap_mapping_range(mapping, holebegin, 0, 1);
npiggin@suse.de25d9e2d2009-08-21 02:35:05 +1000763}
764EXPORT_SYMBOL(truncate_pagecache);
765
766/**
Christoph Hellwig2c27c652010-06-04 11:30:04 +0200767 * truncate_setsize - update inode and pagecache for a new file size
768 * @inode: inode
769 * @newsize: new file size
770 *
Jan Kara382e27d2011-01-20 14:44:26 -0800771 * truncate_setsize updates i_size and performs pagecache truncation (if
772 * necessary) to @newsize. It will be typically be called from the filesystem's
773 * setattr function when ATTR_SIZE is passed in.
Christoph Hellwig2c27c652010-06-04 11:30:04 +0200774 *
Jan Kara77783d02014-11-07 08:29:25 +1100775 * Must be called with a lock serializing truncates and writes (generally
776 * i_mutex but e.g. xfs uses a different lock) and before all filesystem
777 * specific block truncation has been performed.
Christoph Hellwig2c27c652010-06-04 11:30:04 +0200778 */
779void truncate_setsize(struct inode *inode, loff_t newsize)
780{
Jan Kara90a80202014-10-01 21:49:18 -0400781 loff_t oldsize = inode->i_size;
782
Christoph Hellwig2c27c652010-06-04 11:30:04 +0200783 i_size_write(inode, newsize);
Jan Kara90a80202014-10-01 21:49:18 -0400784 if (newsize > oldsize)
785 pagecache_isize_extended(inode, oldsize, newsize);
Kirill A. Shutemov7caef262013-09-12 15:13:56 -0700786 truncate_pagecache(inode, newsize);
Christoph Hellwig2c27c652010-06-04 11:30:04 +0200787}
788EXPORT_SYMBOL(truncate_setsize);
789
790/**
Jan Kara90a80202014-10-01 21:49:18 -0400791 * pagecache_isize_extended - update pagecache after extension of i_size
792 * @inode: inode for which i_size was extended
793 * @from: original inode size
794 * @to: new inode size
795 *
796 * Handle extension of inode size either caused by extending truncate or by
797 * write starting after current i_size. We mark the page straddling current
798 * i_size RO so that page_mkwrite() is called on the nearest write access to
799 * the page. This way filesystem can be sure that page_mkwrite() is called on
800 * the page before user writes to the page via mmap after the i_size has been
801 * changed.
802 *
803 * The function must be called after i_size is updated so that page fault
804 * coming after we unlock the page will already see the new i_size.
805 * The function must be called while we still hold i_mutex - this not only
806 * makes sure i_size is stable but also that userspace cannot observe new
807 * i_size value before we are prepared to store mmap writes at new inode size.
808 */
809void pagecache_isize_extended(struct inode *inode, loff_t from, loff_t to)
810{
Fabian Frederick93407472017-02-27 14:28:32 -0800811 int bsize = i_blocksize(inode);
Jan Kara90a80202014-10-01 21:49:18 -0400812 loff_t rounded_from;
813 struct page *page;
814 pgoff_t index;
815
Jan Kara90a80202014-10-01 21:49:18 -0400816 WARN_ON(to > inode->i_size);
817
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300818 if (from >= to || bsize == PAGE_SIZE)
Jan Kara90a80202014-10-01 21:49:18 -0400819 return;
820 /* Page straddling @from will not have any hole block created? */
821 rounded_from = round_up(from, bsize);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300822 if (to <= rounded_from || !(rounded_from & (PAGE_SIZE - 1)))
Jan Kara90a80202014-10-01 21:49:18 -0400823 return;
824
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300825 index = from >> PAGE_SHIFT;
Jan Kara90a80202014-10-01 21:49:18 -0400826 page = find_lock_page(inode->i_mapping, index);
827 /* Page not cached? Nothing to do */
828 if (!page)
829 return;
830 /*
831 * See clear_page_dirty_for_io() for details why set_page_dirty()
832 * is needed.
833 */
834 if (page_mkclean(page))
835 set_page_dirty(page);
836 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300837 put_page(page);
Jan Kara90a80202014-10-01 21:49:18 -0400838}
839EXPORT_SYMBOL(pagecache_isize_extended);
840
841/**
Hugh Dickins623e3db2012-03-28 14:42:40 -0700842 * truncate_pagecache_range - unmap and remove pagecache that is hole-punched
843 * @inode: inode
844 * @lstart: offset of beginning of hole
845 * @lend: offset of last byte of hole
846 *
847 * This function should typically be called before the filesystem
848 * releases resources associated with the freed range (eg. deallocates
849 * blocks). This way, pagecache will always stay logically coherent
850 * with on-disk format, and the filesystem would not have to deal with
851 * situations such as writepage being called for a page that has already
852 * had its underlying blocks deallocated.
853 */
854void truncate_pagecache_range(struct inode *inode, loff_t lstart, loff_t lend)
855{
856 struct address_space *mapping = inode->i_mapping;
857 loff_t unmap_start = round_up(lstart, PAGE_SIZE);
858 loff_t unmap_end = round_down(1 + lend, PAGE_SIZE) - 1;
859 /*
860 * This rounding is currently just for example: unmap_mapping_range
861 * expands its hole outwards, whereas we want it to contract the hole
862 * inwards. However, existing callers of truncate_pagecache_range are
Lukas Czerner5a720392013-05-27 23:32:35 -0400863 * doing their own page rounding first. Note that unmap_mapping_range
864 * allows holelen 0 for all, and we allow lend -1 for end of file.
Hugh Dickins623e3db2012-03-28 14:42:40 -0700865 */
866
867 /*
868 * Unlike in truncate_pagecache, unmap_mapping_range is called only
869 * once (before truncating pagecache), and without "even_cows" flag:
870 * hole-punching should not remove private COWed pages from the hole.
871 */
872 if ((u64)unmap_end > (u64)unmap_start)
873 unmap_mapping_range(mapping, unmap_start,
874 1 + unmap_end - unmap_start, 0);
875 truncate_inode_pages_range(mapping, lstart, lend);
876}
877EXPORT_SYMBOL(truncate_pagecache_range);