blob: e5cc39ab0751f08b30691c3e7e8c98e32ded409c [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * mm/truncate.c - code for taking down pages from address_spaces
3 *
4 * Copyright (C) 2002, Linus Torvalds
5 *
Francois Camie1f8e872008-10-15 22:01:59 -07006 * 10Sep2002 Andrew Morton
Linus Torvalds1da177e2005-04-16 15:20:36 -07007 * Initial version.
8 */
9
10#include <linux/kernel.h>
Alexey Dobriyan4af3c9c2007-10-16 23:29:23 -070011#include <linux/backing-dev.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090012#include <linux/gfp.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070013#include <linux/mm.h>
Nick Piggin0fd0e6b2006-09-27 01:50:02 -070014#include <linux/swap.h>
Paul Gortmakerb95f1b312011-10-16 02:01:52 -040015#include <linux/export.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070016#include <linux/pagemap.h>
Nate Diller01f27052007-05-09 02:35:07 -070017#include <linux/highmem.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070018#include <linux/pagevec.h>
Andrew Mortone08748ce2006-12-10 02:19:31 -080019#include <linux/task_io_accounting_ops.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070020#include <linux/buffer_head.h> /* grr. try_to_release_page,
Jan Karaaaa40592005-10-30 15:00:16 -080021 do_invalidatepage */
Dan Magenheimerc515e1f2011-05-26 10:01:43 -060022#include <linux/cleancache.h>
Rik van Rielba470de2008-10-18 20:26:50 -070023#include "internal.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070024
Johannes Weiner0cd61442014-04-03 14:47:46 -070025static void clear_exceptional_entry(struct address_space *mapping,
26 pgoff_t index, void *entry)
27{
Johannes Weiner449dd692014-04-03 14:47:56 -070028 struct radix_tree_node *node;
29 void **slot;
30
Johannes Weiner0cd61442014-04-03 14:47:46 -070031 /* Handled by shmem itself */
32 if (shmem_mapping(mapping))
33 return;
34
35 spin_lock_irq(&mapping->tree_lock);
36 /*
37 * Regular page slots are stabilized by the page lock even
38 * without the tree itself locked. These unlocked entries
39 * need verification under the tree lock.
40 */
Johannes Weiner449dd692014-04-03 14:47:56 -070041 if (!__radix_tree_lookup(&mapping->page_tree, index, &node, &slot))
42 goto unlock;
43 if (*slot != entry)
44 goto unlock;
45 radix_tree_replace_slot(slot, NULL);
46 mapping->nrshadows--;
47 if (!node)
48 goto unlock;
49 workingset_node_shadows_dec(node);
50 /*
51 * Don't track node without shadow entries.
52 *
53 * Avoid acquiring the list_lru lock if already untracked.
54 * The list_empty() test is safe as node->private_list is
55 * protected by mapping->tree_lock.
56 */
57 if (!workingset_node_shadows(node) &&
58 !list_empty(&node->private_list))
59 list_lru_del(&workingset_shadow_nodes, &node->private_list);
60 __radix_tree_delete_node(&mapping->page_tree, node);
61unlock:
Johannes Weiner0cd61442014-04-03 14:47:46 -070062 spin_unlock_irq(&mapping->tree_lock);
63}
Linus Torvalds1da177e2005-04-16 15:20:36 -070064
David Howellscf9a2ae2006-08-29 19:05:54 +010065/**
Fengguang Wu28bc44d2008-02-03 18:04:10 +020066 * do_invalidatepage - invalidate part or all of a page
David Howellscf9a2ae2006-08-29 19:05:54 +010067 * @page: the page which is affected
Lukas Czernerd47992f2013-05-21 23:17:23 -040068 * @offset: start of the range to invalidate
69 * @length: length of the range to invalidate
David Howellscf9a2ae2006-08-29 19:05:54 +010070 *
71 * do_invalidatepage() is called when all or part of the page has become
72 * invalidated by a truncate operation.
73 *
74 * do_invalidatepage() does not have to release all buffers, but it must
75 * ensure that no dirty buffer is left outside @offset and that no I/O
76 * is underway against any of the blocks which are outside the truncation
77 * point. Because the caller is about to free (and possibly reuse) those
78 * blocks on-disk.
79 */
Lukas Czernerd47992f2013-05-21 23:17:23 -040080void do_invalidatepage(struct page *page, unsigned int offset,
81 unsigned int length)
David Howellscf9a2ae2006-08-29 19:05:54 +010082{
Lukas Czernerd47992f2013-05-21 23:17:23 -040083 void (*invalidatepage)(struct page *, unsigned int, unsigned int);
84
David Howellscf9a2ae2006-08-29 19:05:54 +010085 invalidatepage = page->mapping->a_ops->invalidatepage;
David Howells93614012006-09-30 20:45:40 +020086#ifdef CONFIG_BLOCK
David Howellscf9a2ae2006-08-29 19:05:54 +010087 if (!invalidatepage)
88 invalidatepage = block_invalidatepage;
David Howells93614012006-09-30 20:45:40 +020089#endif
David Howellscf9a2ae2006-08-29 19:05:54 +010090 if (invalidatepage)
Lukas Czernerd47992f2013-05-21 23:17:23 -040091 (*invalidatepage)(page, offset, length);
David Howellscf9a2ae2006-08-29 19:05:54 +010092}
93
Linus Torvaldsecdfc972007-01-26 12:47:06 -080094/*
95 * This cancels just the dirty bit on the kernel page itself, it
96 * does NOT actually remove dirty bits on any mmap's that may be
97 * around. It also leaves the page tagged dirty, so any sync
98 * activity will still find it on the dirty lists, and in particular,
99 * clear_page_dirty_for_io() will still look at the dirty bits in
100 * the VM.
101 *
102 * Doing this should *normally* only ever be done when a page
103 * is truncated, and is not actually mapped anywhere at all. However,
104 * fs/buffer.c does this when it notices that somebody has cleaned
105 * out all the buffers on a page without actually doing it through
106 * the VM. Can you say "ext3 is horribly ugly"? Tought you could.
107 */
Linus Torvaldsfba25912006-12-20 13:46:42 -0800108void cancel_dirty_page(struct page *page, unsigned int account_size)
109{
Linus Torvalds8368e322006-12-23 09:25:04 -0800110 if (TestClearPageDirty(page)) {
111 struct address_space *mapping = page->mapping;
112 if (mapping && mapping_cap_account_dirty(mapping)) {
113 dec_zone_page_state(page, NR_FILE_DIRTY);
Peter Zijlstrac9e51e42007-10-16 23:25:47 -0700114 dec_bdi_stat(mapping->backing_dev_info,
115 BDI_RECLAIMABLE);
Linus Torvalds8368e322006-12-23 09:25:04 -0800116 if (account_size)
117 task_io_account_cancelled_write(account_size);
118 }
Andrew Morton3e67c092006-12-21 11:00:33 -0800119 }
Linus Torvaldsfba25912006-12-20 13:46:42 -0800120}
Linus Torvalds8368e322006-12-23 09:25:04 -0800121EXPORT_SYMBOL(cancel_dirty_page);
Linus Torvaldsfba25912006-12-20 13:46:42 -0800122
Linus Torvalds1da177e2005-04-16 15:20:36 -0700123/*
124 * If truncate cannot remove the fs-private metadata from the page, the page
Shaohua Li62e1c552008-02-04 22:29:33 -0800125 * becomes orphaned. It will be left on the LRU and may even be mapped into
Nick Piggin54cb8822007-07-19 01:46:59 -0700126 * user pagetables if we're racing with filemap_fault().
Linus Torvalds1da177e2005-04-16 15:20:36 -0700127 *
128 * We need to bale out if page->mapping is no longer equal to the original
129 * mapping. This happens a) when the VM reclaimed the page while we waited on
Andrew Mortonfc0ecff2007-02-10 01:45:39 -0800130 * its lock, b) when a concurrent invalidate_mapping_pages got there first and
Linus Torvalds1da177e2005-04-16 15:20:36 -0700131 * c) when tmpfs swizzles a page between a tmpfs inode and swapper_space.
132 */
Nick Piggin750b4982009-09-16 11:50:12 +0200133static int
Linus Torvalds1da177e2005-04-16 15:20:36 -0700134truncate_complete_page(struct address_space *mapping, struct page *page)
135{
136 if (page->mapping != mapping)
Nick Piggin750b4982009-09-16 11:50:12 +0200137 return -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700138
David Howells266cf652009-04-03 16:42:36 +0100139 if (page_has_private(page))
Lukas Czernerd47992f2013-05-21 23:17:23 -0400140 do_invalidatepage(page, 0, PAGE_CACHE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700141
Bjorn Steinbrinka2b34562008-02-04 22:29:28 -0800142 cancel_dirty_page(page, PAGE_CACHE_SIZE);
143
Linus Torvalds1da177e2005-04-16 15:20:36 -0700144 ClearPageMappedToDisk(page);
Minchan Kim5adc7b52011-03-22 16:32:41 -0700145 delete_from_page_cache(page);
Nick Piggin750b4982009-09-16 11:50:12 +0200146 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700147}
148
149/*
Andrew Mortonfc0ecff2007-02-10 01:45:39 -0800150 * This is for invalidate_mapping_pages(). That function can be called at
Linus Torvalds1da177e2005-04-16 15:20:36 -0700151 * any time, and is not supposed to throw away dirty pages. But pages can
Nick Piggin0fd0e6b2006-09-27 01:50:02 -0700152 * be marked dirty at any time too, so use remove_mapping which safely
153 * discards clean, unused pages.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700154 *
155 * Returns non-zero if the page was successfully invalidated.
156 */
157static int
158invalidate_complete_page(struct address_space *mapping, struct page *page)
159{
Nick Piggin0fd0e6b2006-09-27 01:50:02 -0700160 int ret;
161
Linus Torvalds1da177e2005-04-16 15:20:36 -0700162 if (page->mapping != mapping)
163 return 0;
164
David Howells266cf652009-04-03 16:42:36 +0100165 if (page_has_private(page) && !try_to_release_page(page, 0))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700166 return 0;
167
Nick Piggin0fd0e6b2006-09-27 01:50:02 -0700168 ret = remove_mapping(mapping, page);
Nick Piggin0fd0e6b2006-09-27 01:50:02 -0700169
170 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700171}
172
Nick Piggin750b4982009-09-16 11:50:12 +0200173int truncate_inode_page(struct address_space *mapping, struct page *page)
174{
175 if (page_mapped(page)) {
176 unmap_mapping_range(mapping,
177 (loff_t)page->index << PAGE_CACHE_SHIFT,
178 PAGE_CACHE_SIZE, 0);
179 }
180 return truncate_complete_page(mapping, page);
181}
182
Wu Fengguang83f78662009-09-16 11:50:13 +0200183/*
Andi Kleen25718732009-09-16 11:50:13 +0200184 * Used to get rid of pages on hardware memory corruption.
185 */
186int generic_error_remove_page(struct address_space *mapping, struct page *page)
187{
188 if (!mapping)
189 return -EINVAL;
190 /*
191 * Only punch for normal data pages for now.
192 * Handling other types like directories would need more auditing.
193 */
194 if (!S_ISREG(mapping->host->i_mode))
195 return -EIO;
196 return truncate_inode_page(mapping, page);
197}
198EXPORT_SYMBOL(generic_error_remove_page);
199
200/*
Wu Fengguang83f78662009-09-16 11:50:13 +0200201 * Safely invalidate one page from its pagecache mapping.
202 * It only drops clean, unused pages. The page must be locked.
203 *
204 * Returns 1 if the page is successfully invalidated, otherwise 0.
205 */
206int invalidate_inode_page(struct page *page)
207{
208 struct address_space *mapping = page_mapping(page);
209 if (!mapping)
210 return 0;
211 if (PageDirty(page) || PageWriteback(page))
212 return 0;
213 if (page_mapped(page))
214 return 0;
215 return invalidate_complete_page(mapping, page);
216}
217
Linus Torvalds1da177e2005-04-16 15:20:36 -0700218/**
Liu Bo73c1e202012-02-21 10:57:20 +0800219 * truncate_inode_pages_range - truncate range of pages specified by start & end byte offsets
Linus Torvalds1da177e2005-04-16 15:20:36 -0700220 * @mapping: mapping to truncate
221 * @lstart: offset from which to truncate
Lukas Czerner5a720392013-05-27 23:32:35 -0400222 * @lend: offset to which to truncate (inclusive)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700223 *
Hans Reiserd7339072006-01-06 00:10:36 -0800224 * Truncate the page cache, removing the pages that are between
Lukas Czerner5a720392013-05-27 23:32:35 -0400225 * specified offsets (and zeroing out partial pages
226 * if lstart or lend + 1 is not page aligned).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700227 *
228 * Truncate takes two passes - the first pass is nonblocking. It will not
229 * block on page locks and it will not block on writeback. The second pass
230 * will wait. This is to prevent as much IO as possible in the affected region.
231 * The first pass will remove most pages, so the search cost of the second pass
232 * is low.
233 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700234 * We pass down the cache-hot hint to the page freeing code. Even if the
235 * mapping is large, it is probably the case that the final pages are the most
236 * recently touched, and freeing happens in ascending file offset order.
Lukas Czerner5a720392013-05-27 23:32:35 -0400237 *
238 * Note that since ->invalidatepage() accepts range to invalidate
239 * truncate_inode_pages_range is able to handle cases where lend + 1 is not
240 * page aligned properly.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700241 */
Hans Reiserd7339072006-01-06 00:10:36 -0800242void truncate_inode_pages_range(struct address_space *mapping,
243 loff_t lstart, loff_t lend)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700244{
Lukas Czerner5a720392013-05-27 23:32:35 -0400245 pgoff_t start; /* inclusive */
246 pgoff_t end; /* exclusive */
247 unsigned int partial_start; /* inclusive */
248 unsigned int partial_end; /* exclusive */
249 struct pagevec pvec;
Johannes Weiner0cd61442014-04-03 14:47:46 -0700250 pgoff_t indices[PAGEVEC_SIZE];
Lukas Czerner5a720392013-05-27 23:32:35 -0400251 pgoff_t index;
252 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700253
Dan Magenheimer31677602011-09-21 11:56:28 -0400254 cleancache_invalidate_inode(mapping);
Johannes Weiner91b0abe2014-04-03 14:47:49 -0700255 if (mapping->nrpages == 0 && mapping->nrshadows == 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700256 return;
257
Lukas Czerner5a720392013-05-27 23:32:35 -0400258 /* Offsets within partial pages */
259 partial_start = lstart & (PAGE_CACHE_SIZE - 1);
260 partial_end = (lend + 1) & (PAGE_CACHE_SIZE - 1);
261
262 /*
263 * 'start' and 'end' always covers the range of pages to be fully
264 * truncated. Partial pages are covered with 'partial_start' at the
265 * start of the range and 'partial_end' at the end of the range.
266 * Note that 'end' is exclusive while 'lend' is inclusive.
267 */
268 start = (lstart + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
269 if (lend == -1)
270 /*
271 * lend == -1 indicates end-of-file so we have to set 'end'
272 * to the highest possible pgoff_t and since the type is
273 * unsigned we're using -1.
274 */
275 end = -1;
276 else
277 end = (lend + 1) >> PAGE_CACHE_SHIFT;
Hans Reiserd7339072006-01-06 00:10:36 -0800278
Linus Torvalds1da177e2005-04-16 15:20:36 -0700279 pagevec_init(&pvec, 0);
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700280 index = start;
Johannes Weiner0cd61442014-04-03 14:47:46 -0700281 while (index < end && pagevec_lookup_entries(&pvec, mapping, index,
282 min(end - index, (pgoff_t)PAGEVEC_SIZE),
283 indices)) {
Hugh Dickinse5598f82011-02-25 14:44:29 -0800284 mem_cgroup_uncharge_start();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700285 for (i = 0; i < pagevec_count(&pvec); i++) {
286 struct page *page = pvec.pages[i];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700287
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700288 /* We rely upon deletion not changing page->index */
Johannes Weiner0cd61442014-04-03 14:47:46 -0700289 index = indices[i];
Lukas Czerner5a720392013-05-27 23:32:35 -0400290 if (index >= end)
Hans Reiserd7339072006-01-06 00:10:36 -0800291 break;
Hans Reiserd7339072006-01-06 00:10:36 -0800292
Johannes Weiner0cd61442014-04-03 14:47:46 -0700293 if (radix_tree_exceptional_entry(page)) {
294 clear_exceptional_entry(mapping, index, page);
295 continue;
296 }
297
Nick Piggin529ae9a2008-08-02 12:01:03 +0200298 if (!trylock_page(page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700299 continue;
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700300 WARN_ON(page->index != index);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700301 if (PageWriteback(page)) {
302 unlock_page(page);
303 continue;
304 }
Nick Piggin750b4982009-09-16 11:50:12 +0200305 truncate_inode_page(mapping, page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700306 unlock_page(page);
307 }
Johannes Weiner0cd61442014-04-03 14:47:46 -0700308 pagevec_remove_exceptionals(&pvec);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700309 pagevec_release(&pvec);
Hugh Dickinse5598f82011-02-25 14:44:29 -0800310 mem_cgroup_uncharge_end();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700311 cond_resched();
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700312 index++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700313 }
314
Lukas Czerner5a720392013-05-27 23:32:35 -0400315 if (partial_start) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700316 struct page *page = find_lock_page(mapping, start - 1);
317 if (page) {
Lukas Czerner5a720392013-05-27 23:32:35 -0400318 unsigned int top = PAGE_CACHE_SIZE;
319 if (start > end) {
320 /* Truncation within a single page */
321 top = partial_end;
322 partial_end = 0;
323 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700324 wait_on_page_writeback(page);
Lukas Czerner5a720392013-05-27 23:32:35 -0400325 zero_user_segment(page, partial_start, top);
326 cleancache_invalidate_page(mapping, page);
327 if (page_has_private(page))
328 do_invalidatepage(page, partial_start,
329 top - partial_start);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700330 unlock_page(page);
331 page_cache_release(page);
332 }
333 }
Lukas Czerner5a720392013-05-27 23:32:35 -0400334 if (partial_end) {
335 struct page *page = find_lock_page(mapping, end);
336 if (page) {
337 wait_on_page_writeback(page);
338 zero_user_segment(page, 0, partial_end);
339 cleancache_invalidate_page(mapping, page);
340 if (page_has_private(page))
341 do_invalidatepage(page, 0,
342 partial_end);
343 unlock_page(page);
344 page_cache_release(page);
345 }
346 }
347 /*
348 * If the truncation happened within a single page no pages
349 * will be released, just zeroed, so we can bail out now.
350 */
351 if (start >= end)
352 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700353
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700354 index = start;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700355 for ( ; ; ) {
356 cond_resched();
Johannes Weiner0cd61442014-04-03 14:47:46 -0700357 if (!pagevec_lookup_entries(&pvec, mapping, index,
358 min(end - index, (pgoff_t)PAGEVEC_SIZE),
359 indices)) {
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700360 if (index == start)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700361 break;
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700362 index = start;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700363 continue;
364 }
Johannes Weiner0cd61442014-04-03 14:47:46 -0700365 if (index == start && indices[0] >= end) {
366 pagevec_remove_exceptionals(&pvec);
Hans Reiserd7339072006-01-06 00:10:36 -0800367 pagevec_release(&pvec);
368 break;
369 }
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -0800370 mem_cgroup_uncharge_start();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700371 for (i = 0; i < pagevec_count(&pvec); i++) {
372 struct page *page = pvec.pages[i];
373
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700374 /* We rely upon deletion not changing page->index */
Johannes Weiner0cd61442014-04-03 14:47:46 -0700375 index = indices[i];
Lukas Czerner5a720392013-05-27 23:32:35 -0400376 if (index >= end)
Hans Reiserd7339072006-01-06 00:10:36 -0800377 break;
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700378
Johannes Weiner0cd61442014-04-03 14:47:46 -0700379 if (radix_tree_exceptional_entry(page)) {
380 clear_exceptional_entry(mapping, index, page);
381 continue;
382 }
383
Linus Torvalds1da177e2005-04-16 15:20:36 -0700384 lock_page(page);
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700385 WARN_ON(page->index != index);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700386 wait_on_page_writeback(page);
Nick Piggin750b4982009-09-16 11:50:12 +0200387 truncate_inode_page(mapping, page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700388 unlock_page(page);
389 }
Johannes Weiner0cd61442014-04-03 14:47:46 -0700390 pagevec_remove_exceptionals(&pvec);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700391 pagevec_release(&pvec);
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -0800392 mem_cgroup_uncharge_end();
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700393 index++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700394 }
Dan Magenheimer31677602011-09-21 11:56:28 -0400395 cleancache_invalidate_inode(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700396}
Hans Reiserd7339072006-01-06 00:10:36 -0800397EXPORT_SYMBOL(truncate_inode_pages_range);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700398
Hans Reiserd7339072006-01-06 00:10:36 -0800399/**
400 * truncate_inode_pages - truncate *all* the pages from an offset
401 * @mapping: mapping to truncate
402 * @lstart: offset from which to truncate
403 *
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800404 * Called under (and serialised by) inode->i_mutex.
Jan Kara08142572011-06-27 16:18:10 -0700405 *
406 * Note: When this function returns, there can be a page in the process of
407 * deletion (inside __delete_from_page_cache()) in the specified range. Thus
408 * mapping->nrpages can be non-zero when this function returns even after
409 * truncation of the whole mapping.
Hans Reiserd7339072006-01-06 00:10:36 -0800410 */
411void truncate_inode_pages(struct address_space *mapping, loff_t lstart)
412{
413 truncate_inode_pages_range(mapping, lstart, (loff_t)-1);
414}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700415EXPORT_SYMBOL(truncate_inode_pages);
416
Mike Waychison28697352009-06-16 15:32:59 -0700417/**
Johannes Weiner91b0abe2014-04-03 14:47:49 -0700418 * truncate_inode_pages_final - truncate *all* pages before inode dies
419 * @mapping: mapping to truncate
420 *
421 * Called under (and serialized by) inode->i_mutex.
422 *
423 * Filesystems have to use this in the .evict_inode path to inform the
424 * VM that this is the final truncate and the inode is going away.
425 */
426void truncate_inode_pages_final(struct address_space *mapping)
427{
428 unsigned long nrshadows;
429 unsigned long nrpages;
430
431 /*
432 * Page reclaim can not participate in regular inode lifetime
433 * management (can't call iput()) and thus can race with the
434 * inode teardown. Tell it when the address space is exiting,
435 * so that it does not install eviction information after the
436 * final truncate has begun.
437 */
438 mapping_set_exiting(mapping);
439
440 /*
441 * When reclaim installs eviction entries, it increases
442 * nrshadows first, then decreases nrpages. Make sure we see
443 * this in the right order or we might miss an entry.
444 */
445 nrpages = mapping->nrpages;
446 smp_rmb();
447 nrshadows = mapping->nrshadows;
448
449 if (nrpages || nrshadows) {
450 /*
451 * As truncation uses a lockless tree lookup, cycle
452 * the tree lock to make sure any ongoing tree
453 * modification that does not see AS_EXITING is
454 * completed before starting the final truncate.
455 */
456 spin_lock_irq(&mapping->tree_lock);
457 spin_unlock_irq(&mapping->tree_lock);
458
459 truncate_inode_pages(mapping, 0);
460 }
461}
462EXPORT_SYMBOL(truncate_inode_pages_final);
463
464/**
Mike Waychison28697352009-06-16 15:32:59 -0700465 * invalidate_mapping_pages - Invalidate all the unlocked pages of one inode
466 * @mapping: the address_space which holds the pages to invalidate
467 * @start: the offset 'from' which to invalidate
468 * @end: the offset 'to' which to invalidate (inclusive)
469 *
470 * This function only removes the unlocked pages, if you want to
471 * remove all the pages of one inode, you must call truncate_inode_pages.
472 *
473 * invalidate_mapping_pages() will not block on IO activity. It will not
474 * invalidate pages which are dirty, locked, under writeback or mapped into
475 * pagetables.
476 */
477unsigned long invalidate_mapping_pages(struct address_space *mapping,
Minchan Kim31560182011-03-22 16:32:52 -0700478 pgoff_t start, pgoff_t end)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700479{
Johannes Weiner0cd61442014-04-03 14:47:46 -0700480 pgoff_t indices[PAGEVEC_SIZE];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700481 struct pagevec pvec;
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700482 pgoff_t index = start;
Minchan Kim31560182011-03-22 16:32:52 -0700483 unsigned long ret;
484 unsigned long count = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700485 int i;
486
Hugh Dickins31475dd2011-08-03 16:21:27 -0700487 /*
488 * Note: this function may get called on a shmem/tmpfs mapping:
489 * pagevec_lookup() might then return 0 prematurely (because it
490 * got a gangful of swap entries); but it's hardly worth worrying
491 * about - it can rarely have anything to free from such a mapping
492 * (most pages are dirty), and already skips over any difficulties.
493 */
494
Linus Torvalds1da177e2005-04-16 15:20:36 -0700495 pagevec_init(&pvec, 0);
Johannes Weiner0cd61442014-04-03 14:47:46 -0700496 while (index <= end && pagevec_lookup_entries(&pvec, mapping, index,
497 min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1,
498 indices)) {
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -0800499 mem_cgroup_uncharge_start();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700500 for (i = 0; i < pagevec_count(&pvec); i++) {
501 struct page *page = pvec.pages[i];
502
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700503 /* We rely upon deletion not changing page->index */
Johannes Weiner0cd61442014-04-03 14:47:46 -0700504 index = indices[i];
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700505 if (index > end)
506 break;
NeilBrowne0f236032006-06-23 02:05:48 -0700507
Johannes Weiner0cd61442014-04-03 14:47:46 -0700508 if (radix_tree_exceptional_entry(page)) {
509 clear_exceptional_entry(mapping, index, page);
510 continue;
511 }
512
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700513 if (!trylock_page(page))
514 continue;
515 WARN_ON(page->index != index);
Minchan Kim31560182011-03-22 16:32:52 -0700516 ret = invalidate_inode_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700517 unlock_page(page);
Minchan Kim31560182011-03-22 16:32:52 -0700518 /*
519 * Invalidation is a hint that the page is no longer
520 * of interest and try to speed up its reclaim.
521 */
522 if (!ret)
523 deactivate_page(page);
524 count += ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700525 }
Johannes Weiner0cd61442014-04-03 14:47:46 -0700526 pagevec_remove_exceptionals(&pvec);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700527 pagevec_release(&pvec);
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -0800528 mem_cgroup_uncharge_end();
Mike Waychison28697352009-06-16 15:32:59 -0700529 cond_resched();
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700530 index++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700531 }
Minchan Kim31560182011-03-22 16:32:52 -0700532 return count;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700533}
Anton Altaparmakov54bc4852007-02-10 01:45:38 -0800534EXPORT_SYMBOL(invalidate_mapping_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700535
Andrew Mortonbd4c8ce2006-09-30 23:29:29 -0700536/*
537 * This is like invalidate_complete_page(), except it ignores the page's
538 * refcount. We do this because invalidate_inode_pages2() needs stronger
539 * invalidation guarantees, and cannot afford to leave pages behind because
Anderson Briglia2706a1b2007-07-15 23:38:09 -0700540 * shrink_page_list() has a temp ref on them, or because they're transiently
541 * sitting in the lru_cache_add() pagevecs.
Andrew Mortonbd4c8ce2006-09-30 23:29:29 -0700542 */
543static int
544invalidate_complete_page2(struct address_space *mapping, struct page *page)
545{
546 if (page->mapping != mapping)
547 return 0;
548
David Howells266cf652009-04-03 16:42:36 +0100549 if (page_has_private(page) && !try_to_release_page(page, GFP_KERNEL))
Andrew Mortonbd4c8ce2006-09-30 23:29:29 -0700550 return 0;
551
Nick Piggin19fd6232008-07-25 19:45:32 -0700552 spin_lock_irq(&mapping->tree_lock);
Andrew Mortonbd4c8ce2006-09-30 23:29:29 -0700553 if (PageDirty(page))
554 goto failed;
555
David Howells266cf652009-04-03 16:42:36 +0100556 BUG_ON(page_has_private(page));
Johannes Weiner91b0abe2014-04-03 14:47:49 -0700557 __delete_from_page_cache(page, NULL);
Nick Piggin19fd6232008-07-25 19:45:32 -0700558 spin_unlock_irq(&mapping->tree_lock);
Daisuke Nishimurae767e052009-05-28 14:34:28 -0700559 mem_cgroup_uncharge_cache_page(page);
Linus Torvalds6072d132010-12-01 13:35:19 -0500560
561 if (mapping->a_ops->freepage)
562 mapping->a_ops->freepage(page);
563
Andrew Mortonbd4c8ce2006-09-30 23:29:29 -0700564 page_cache_release(page); /* pagecache ref */
565 return 1;
566failed:
Nick Piggin19fd6232008-07-25 19:45:32 -0700567 spin_unlock_irq(&mapping->tree_lock);
Andrew Mortonbd4c8ce2006-09-30 23:29:29 -0700568 return 0;
569}
570
Trond Myklebuste3db7692007-01-10 23:15:39 -0800571static int do_launder_page(struct address_space *mapping, struct page *page)
572{
573 if (!PageDirty(page))
574 return 0;
575 if (page->mapping != mapping || mapping->a_ops->launder_page == NULL)
576 return 0;
577 return mapping->a_ops->launder_page(page);
578}
579
Linus Torvalds1da177e2005-04-16 15:20:36 -0700580/**
581 * invalidate_inode_pages2_range - remove range of pages from an address_space
Martin Waitz67be2dd2005-05-01 08:59:26 -0700582 * @mapping: the address_space
Linus Torvalds1da177e2005-04-16 15:20:36 -0700583 * @start: the page offset 'from' which to invalidate
584 * @end: the page offset 'to' which to invalidate (inclusive)
585 *
586 * Any pages which are found to be mapped into pagetables are unmapped prior to
587 * invalidation.
588 *
Hisashi Hifumi6ccfa802008-09-02 14:35:40 -0700589 * Returns -EBUSY if any pages could not be invalidated.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700590 */
591int invalidate_inode_pages2_range(struct address_space *mapping,
592 pgoff_t start, pgoff_t end)
593{
Johannes Weiner0cd61442014-04-03 14:47:46 -0700594 pgoff_t indices[PAGEVEC_SIZE];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700595 struct pagevec pvec;
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700596 pgoff_t index;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700597 int i;
598 int ret = 0;
Hisashi Hifumi0dd13342008-04-28 02:12:08 -0700599 int ret2 = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700600 int did_range_unmap = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700601
Dan Magenheimer31677602011-09-21 11:56:28 -0400602 cleancache_invalidate_inode(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700603 pagevec_init(&pvec, 0);
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700604 index = start;
Johannes Weiner0cd61442014-04-03 14:47:46 -0700605 while (index <= end && pagevec_lookup_entries(&pvec, mapping, index,
606 min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1,
607 indices)) {
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -0800608 mem_cgroup_uncharge_start();
Trond Myklebust7b965e02007-02-28 20:13:55 -0800609 for (i = 0; i < pagevec_count(&pvec); i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700610 struct page *page = pvec.pages[i];
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700611
612 /* We rely upon deletion not changing page->index */
Johannes Weiner0cd61442014-04-03 14:47:46 -0700613 index = indices[i];
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700614 if (index > end)
615 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700616
Johannes Weiner0cd61442014-04-03 14:47:46 -0700617 if (radix_tree_exceptional_entry(page)) {
618 clear_exceptional_entry(mapping, index, page);
619 continue;
620 }
621
Linus Torvalds1da177e2005-04-16 15:20:36 -0700622 lock_page(page);
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700623 WARN_ON(page->index != index);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700624 if (page->mapping != mapping) {
625 unlock_page(page);
626 continue;
627 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700628 wait_on_page_writeback(page);
Nick Piggind00806b2007-07-19 01:46:57 -0700629 if (page_mapped(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700630 if (!did_range_unmap) {
631 /*
632 * Zap the rest of the file in one hit.
633 */
634 unmap_mapping_range(mapping,
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700635 (loff_t)index << PAGE_CACHE_SHIFT,
636 (loff_t)(1 + end - index)
637 << PAGE_CACHE_SHIFT,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700638 0);
639 did_range_unmap = 1;
640 } else {
641 /*
642 * Just zap this page
643 */
644 unmap_mapping_range(mapping,
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700645 (loff_t)index << PAGE_CACHE_SHIFT,
646 PAGE_CACHE_SIZE, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700647 }
648 }
Nick Piggind00806b2007-07-19 01:46:57 -0700649 BUG_ON(page_mapped(page));
Hisashi Hifumi0dd13342008-04-28 02:12:08 -0700650 ret2 = do_launder_page(mapping, page);
651 if (ret2 == 0) {
652 if (!invalidate_complete_page2(mapping, page))
Hisashi Hifumi6ccfa802008-09-02 14:35:40 -0700653 ret2 = -EBUSY;
Hisashi Hifumi0dd13342008-04-28 02:12:08 -0700654 }
655 if (ret2 < 0)
656 ret = ret2;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700657 unlock_page(page);
658 }
Johannes Weiner0cd61442014-04-03 14:47:46 -0700659 pagevec_remove_exceptionals(&pvec);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700660 pagevec_release(&pvec);
KAMEZAWA Hiroyuki569b8462009-12-15 16:47:03 -0800661 mem_cgroup_uncharge_end();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700662 cond_resched();
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700663 index++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700664 }
Dan Magenheimer31677602011-09-21 11:56:28 -0400665 cleancache_invalidate_inode(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700666 return ret;
667}
668EXPORT_SYMBOL_GPL(invalidate_inode_pages2_range);
669
670/**
671 * invalidate_inode_pages2 - remove all pages from an address_space
Martin Waitz67be2dd2005-05-01 08:59:26 -0700672 * @mapping: the address_space
Linus Torvalds1da177e2005-04-16 15:20:36 -0700673 *
674 * Any pages which are found to be mapped into pagetables are unmapped prior to
675 * invalidation.
676 *
Peng Taoe9de25d2009-10-19 14:48:13 +0800677 * Returns -EBUSY if any pages could not be invalidated.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700678 */
679int invalidate_inode_pages2(struct address_space *mapping)
680{
681 return invalidate_inode_pages2_range(mapping, 0, -1);
682}
683EXPORT_SYMBOL_GPL(invalidate_inode_pages2);
npiggin@suse.de25d9e2d2009-08-21 02:35:05 +1000684
685/**
686 * truncate_pagecache - unmap and remove pagecache that has been truncated
687 * @inode: inode
Hugh Dickins8a549be2011-07-25 17:12:24 -0700688 * @newsize: new file size
npiggin@suse.de25d9e2d2009-08-21 02:35:05 +1000689 *
690 * inode's new i_size must already be written before truncate_pagecache
691 * is called.
692 *
693 * This function should typically be called before the filesystem
694 * releases resources associated with the freed range (eg. deallocates
695 * blocks). This way, pagecache will always stay logically coherent
696 * with on-disk format, and the filesystem would not have to deal with
697 * situations such as writepage being called for a page that has already
698 * had its underlying blocks deallocated.
699 */
Kirill A. Shutemov7caef262013-09-12 15:13:56 -0700700void truncate_pagecache(struct inode *inode, loff_t newsize)
npiggin@suse.de25d9e2d2009-08-21 02:35:05 +1000701{
OGAWA Hirofumicedabed2010-01-13 21:14:09 +0900702 struct address_space *mapping = inode->i_mapping;
Hugh Dickins8a549be2011-07-25 17:12:24 -0700703 loff_t holebegin = round_up(newsize, PAGE_SIZE);
npiggin@suse.de25d9e2d2009-08-21 02:35:05 +1000704
OGAWA Hirofumicedabed2010-01-13 21:14:09 +0900705 /*
706 * unmap_mapping_range is called twice, first simply for
707 * efficiency so that truncate_inode_pages does fewer
708 * single-page unmaps. However after this first call, and
709 * before truncate_inode_pages finishes, it is possible for
710 * private pages to be COWed, which remain after
711 * truncate_inode_pages finishes, hence the second
712 * unmap_mapping_range call must be made for correctness.
713 */
Hugh Dickins8a549be2011-07-25 17:12:24 -0700714 unmap_mapping_range(mapping, holebegin, 0, 1);
715 truncate_inode_pages(mapping, newsize);
716 unmap_mapping_range(mapping, holebegin, 0, 1);
npiggin@suse.de25d9e2d2009-08-21 02:35:05 +1000717}
718EXPORT_SYMBOL(truncate_pagecache);
719
720/**
Christoph Hellwig2c27c652010-06-04 11:30:04 +0200721 * truncate_setsize - update inode and pagecache for a new file size
722 * @inode: inode
723 * @newsize: new file size
724 *
Jan Kara382e27d2011-01-20 14:44:26 -0800725 * truncate_setsize updates i_size and performs pagecache truncation (if
726 * necessary) to @newsize. It will be typically be called from the filesystem's
727 * setattr function when ATTR_SIZE is passed in.
Christoph Hellwig2c27c652010-06-04 11:30:04 +0200728 *
Jan Kara382e27d2011-01-20 14:44:26 -0800729 * Must be called with inode_mutex held and before all filesystem specific
730 * block truncation has been performed.
Christoph Hellwig2c27c652010-06-04 11:30:04 +0200731 */
732void truncate_setsize(struct inode *inode, loff_t newsize)
733{
Christoph Hellwig2c27c652010-06-04 11:30:04 +0200734 i_size_write(inode, newsize);
Kirill A. Shutemov7caef262013-09-12 15:13:56 -0700735 truncate_pagecache(inode, newsize);
Christoph Hellwig2c27c652010-06-04 11:30:04 +0200736}
737EXPORT_SYMBOL(truncate_setsize);
738
739/**
Hugh Dickins623e3db2012-03-28 14:42:40 -0700740 * truncate_pagecache_range - unmap and remove pagecache that is hole-punched
741 * @inode: inode
742 * @lstart: offset of beginning of hole
743 * @lend: offset of last byte of hole
744 *
745 * This function should typically be called before the filesystem
746 * releases resources associated with the freed range (eg. deallocates
747 * blocks). This way, pagecache will always stay logically coherent
748 * with on-disk format, and the filesystem would not have to deal with
749 * situations such as writepage being called for a page that has already
750 * had its underlying blocks deallocated.
751 */
752void truncate_pagecache_range(struct inode *inode, loff_t lstart, loff_t lend)
753{
754 struct address_space *mapping = inode->i_mapping;
755 loff_t unmap_start = round_up(lstart, PAGE_SIZE);
756 loff_t unmap_end = round_down(1 + lend, PAGE_SIZE) - 1;
757 /*
758 * This rounding is currently just for example: unmap_mapping_range
759 * expands its hole outwards, whereas we want it to contract the hole
760 * inwards. However, existing callers of truncate_pagecache_range are
Lukas Czerner5a720392013-05-27 23:32:35 -0400761 * doing their own page rounding first. Note that unmap_mapping_range
762 * allows holelen 0 for all, and we allow lend -1 for end of file.
Hugh Dickins623e3db2012-03-28 14:42:40 -0700763 */
764
765 /*
766 * Unlike in truncate_pagecache, unmap_mapping_range is called only
767 * once (before truncating pagecache), and without "even_cows" flag:
768 * hole-punching should not remove private COWed pages from the hole.
769 */
770 if ((u64)unmap_end > (u64)unmap_start)
771 unmap_mapping_range(mapping, unmap_start,
772 1 + unmap_end - unmap_start, 0);
773 truncate_inode_pages_range(mapping, lstart, lend);
774}
775EXPORT_SYMBOL(truncate_pagecache_range);