blob: dd9ebc1da35664d04dd508539b34fb96dc764f28 [file] [log] [blame]
Thomas Gleixner457c8992019-05-19 13:08:55 +01001// SPDX-License-Identifier: GPL-2.0-only
Linus Torvalds1da177e2005-04-16 15:20:36 -07002/*
3 * mm/truncate.c - code for taking down pages from address_spaces
4 *
5 * Copyright (C) 2002, Linus Torvalds
6 *
Francois Camie1f8e872008-10-15 22:01:59 -07007 * 10Sep2002 Andrew Morton
Linus Torvalds1da177e2005-04-16 15:20:36 -07008 * Initial version.
9 */
10
11#include <linux/kernel.h>
Alexey Dobriyan4af3c9c2007-10-16 23:29:23 -070012#include <linux/backing-dev.h>
Ross Zwislerf9fe48b2016-01-22 15:10:40 -080013#include <linux/dax.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090014#include <linux/gfp.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070015#include <linux/mm.h>
Nick Piggin0fd0e6b2006-09-27 01:50:02 -070016#include <linux/swap.h>
Paul Gortmakerb95f1b312011-10-16 02:01:52 -040017#include <linux/export.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070018#include <linux/pagemap.h>
Nate Diller01f2705d2007-05-09 02:35:07 -070019#include <linux/highmem.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070020#include <linux/pagevec.h>
Andrew Mortone08748ce2006-12-10 02:19:31 -080021#include <linux/task_io_accounting_ops.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070022#include <linux/buffer_head.h> /* grr. try_to_release_page,
Jan Karaaaa40592005-10-30 15:00:16 -080023 do_invalidatepage */
Hugh Dickins3a4f8a02017-02-24 14:59:36 -080024#include <linux/shmem_fs.h>
Dan Magenheimerc515e1f2011-05-26 10:01:43 -060025#include <linux/cleancache.h>
Jan Kara90a80202014-10-01 21:49:18 -040026#include <linux/rmap.h>
Rik van Rielba470de2008-10-18 20:26:50 -070027#include "internal.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070028
Mel Gormanf2187592017-11-15 17:37:44 -080029/*
30 * Regular page slots are stabilized by the page lock even without the tree
31 * itself locked. These unlocked entries need verification under the tree
32 * lock.
33 */
34static inline void __clear_shadow_entry(struct address_space *mapping,
35 pgoff_t index, void *entry)
Johannes Weiner0cd61442014-04-03 14:47:46 -070036{
Matthew Wilcox69b6c132017-11-25 22:52:46 -050037 XA_STATE(xas, &mapping->i_pages, index);
Johannes Weiner449dd692014-04-03 14:47:56 -070038
Matthew Wilcox69b6c132017-11-25 22:52:46 -050039 xas_set_update(&xas, workingset_update_node);
40 if (xas_load(&xas) != entry)
Mel Gormanf2187592017-11-15 17:37:44 -080041 return;
Matthew Wilcox69b6c132017-11-25 22:52:46 -050042 xas_store(&xas, NULL);
Jan Karaac401cc2016-05-12 18:29:18 +020043 mapping->nrexceptional--;
Mel Gormanf2187592017-11-15 17:37:44 -080044}
45
46static void clear_shadow_entry(struct address_space *mapping, pgoff_t index,
47 void *entry)
48{
Matthew Wilcoxb93b0162018-04-10 16:36:56 -070049 xa_lock_irq(&mapping->i_pages);
Mel Gormanf2187592017-11-15 17:37:44 -080050 __clear_shadow_entry(mapping, index, entry);
Matthew Wilcoxb93b0162018-04-10 16:36:56 -070051 xa_unlock_irq(&mapping->i_pages);
Johannes Weiner0cd61442014-04-03 14:47:46 -070052}
Linus Torvalds1da177e2005-04-16 15:20:36 -070053
Jan Karac6dcf522016-08-10 17:22:44 +020054/*
Mel Gormanf2187592017-11-15 17:37:44 -080055 * Unconditionally remove exceptional entries. Usually called from truncate
56 * path. Note that the pagevec may be altered by this function by removing
57 * exceptional entries similar to what pagevec_remove_exceptionals does.
Jan Karac6dcf522016-08-10 17:22:44 +020058 */
Mel Gormanf2187592017-11-15 17:37:44 -080059static void truncate_exceptional_pvec_entries(struct address_space *mapping,
60 struct pagevec *pvec, pgoff_t *indices,
61 pgoff_t end)
Jan Karac6dcf522016-08-10 17:22:44 +020062{
Mel Gormanf2187592017-11-15 17:37:44 -080063 int i, j;
64 bool dax, lock;
65
Jan Karac6dcf522016-08-10 17:22:44 +020066 /* Handled by shmem itself */
67 if (shmem_mapping(mapping))
68 return;
69
Mel Gormanf2187592017-11-15 17:37:44 -080070 for (j = 0; j < pagevec_count(pvec); j++)
Matthew Wilcox3159f942017-11-03 13:30:42 -040071 if (xa_is_value(pvec->pages[j]))
Mel Gormanf2187592017-11-15 17:37:44 -080072 break;
73
74 if (j == pagevec_count(pvec))
Jan Karac6dcf522016-08-10 17:22:44 +020075 return;
Mel Gormanf2187592017-11-15 17:37:44 -080076
77 dax = dax_mapping(mapping);
78 lock = !dax && indices[j] < end;
79 if (lock)
Matthew Wilcoxb93b0162018-04-10 16:36:56 -070080 xa_lock_irq(&mapping->i_pages);
Mel Gormanf2187592017-11-15 17:37:44 -080081
82 for (i = j; i < pagevec_count(pvec); i++) {
83 struct page *page = pvec->pages[i];
84 pgoff_t index = indices[i];
85
Matthew Wilcox3159f942017-11-03 13:30:42 -040086 if (!xa_is_value(page)) {
Mel Gormanf2187592017-11-15 17:37:44 -080087 pvec->pages[j++] = page;
88 continue;
89 }
90
91 if (index >= end)
92 continue;
93
94 if (unlikely(dax)) {
95 dax_delete_mapping_entry(mapping, index);
96 continue;
97 }
98
99 __clear_shadow_entry(mapping, index, page);
Jan Karac6dcf522016-08-10 17:22:44 +0200100 }
Mel Gormanf2187592017-11-15 17:37:44 -0800101
102 if (lock)
Matthew Wilcoxb93b0162018-04-10 16:36:56 -0700103 xa_unlock_irq(&mapping->i_pages);
Mel Gormanf2187592017-11-15 17:37:44 -0800104 pvec->nr = j;
Jan Karac6dcf522016-08-10 17:22:44 +0200105}
106
107/*
108 * Invalidate exceptional entry if easily possible. This handles exceptional
Ross Zwisler4636e702017-05-12 15:46:47 -0700109 * entries for invalidate_inode_pages().
Jan Karac6dcf522016-08-10 17:22:44 +0200110 */
111static int invalidate_exceptional_entry(struct address_space *mapping,
112 pgoff_t index, void *entry)
113{
Ross Zwisler4636e702017-05-12 15:46:47 -0700114 /* Handled by shmem itself, or for DAX we do nothing. */
115 if (shmem_mapping(mapping) || dax_mapping(mapping))
Jan Karac6dcf522016-08-10 17:22:44 +0200116 return 1;
Jan Karac6dcf522016-08-10 17:22:44 +0200117 clear_shadow_entry(mapping, index, entry);
118 return 1;
119}
120
121/*
122 * Invalidate exceptional entry if clean. This handles exceptional entries for
123 * invalidate_inode_pages2() so for DAX it evicts only clean entries.
124 */
125static int invalidate_exceptional_entry2(struct address_space *mapping,
126 pgoff_t index, void *entry)
127{
128 /* Handled by shmem itself */
129 if (shmem_mapping(mapping))
130 return 1;
131 if (dax_mapping(mapping))
132 return dax_invalidate_mapping_entry_sync(mapping, index);
133 clear_shadow_entry(mapping, index, entry);
134 return 1;
135}
136
David Howellscf9a2ae2006-08-29 19:05:54 +0100137/**
Fengguang Wu28bc44d2008-02-03 18:04:10 +0200138 * do_invalidatepage - invalidate part or all of a page
David Howellscf9a2ae2006-08-29 19:05:54 +0100139 * @page: the page which is affected
Lukas Czernerd47992f2013-05-21 23:17:23 -0400140 * @offset: start of the range to invalidate
141 * @length: length of the range to invalidate
David Howellscf9a2ae2006-08-29 19:05:54 +0100142 *
143 * do_invalidatepage() is called when all or part of the page has become
144 * invalidated by a truncate operation.
145 *
146 * do_invalidatepage() does not have to release all buffers, but it must
147 * ensure that no dirty buffer is left outside @offset and that no I/O
148 * is underway against any of the blocks which are outside the truncation
149 * point. Because the caller is about to free (and possibly reuse) those
150 * blocks on-disk.
151 */
Lukas Czernerd47992f2013-05-21 23:17:23 -0400152void do_invalidatepage(struct page *page, unsigned int offset,
153 unsigned int length)
David Howellscf9a2ae2006-08-29 19:05:54 +0100154{
Lukas Czernerd47992f2013-05-21 23:17:23 -0400155 void (*invalidatepage)(struct page *, unsigned int, unsigned int);
156
David Howellscf9a2ae2006-08-29 19:05:54 +0100157 invalidatepage = page->mapping->a_ops->invalidatepage;
David Howells93614012006-09-30 20:45:40 +0200158#ifdef CONFIG_BLOCK
David Howellscf9a2ae2006-08-29 19:05:54 +0100159 if (!invalidatepage)
160 invalidatepage = block_invalidatepage;
David Howells93614012006-09-30 20:45:40 +0200161#endif
David Howellscf9a2ae2006-08-29 19:05:54 +0100162 if (invalidatepage)
Lukas Czernerd47992f2013-05-21 23:17:23 -0400163 (*invalidatepage)(page, offset, length);
David Howellscf9a2ae2006-08-29 19:05:54 +0100164}
165
Linus Torvaldsecdfc972007-01-26 12:47:06 -0800166/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700167 * If truncate cannot remove the fs-private metadata from the page, the page
Shaohua Li62e1c552008-02-04 22:29:33 -0800168 * becomes orphaned. It will be left on the LRU and may even be mapped into
Nick Piggin54cb8822007-07-19 01:46:59 -0700169 * user pagetables if we're racing with filemap_fault().
Linus Torvalds1da177e2005-04-16 15:20:36 -0700170 *
171 * We need to bale out if page->mapping is no longer equal to the original
172 * mapping. This happens a) when the VM reclaimed the page while we waited on
Andrew Mortonfc0ecff2007-02-10 01:45:39 -0800173 * its lock, b) when a concurrent invalidate_mapping_pages got there first and
Linus Torvalds1da177e2005-04-16 15:20:36 -0700174 * c) when tmpfs swizzles a page between a tmpfs inode and swapper_space.
175 */
Jan Kara9f4e41f2017-11-15 17:37:15 -0800176static void
177truncate_cleanup_page(struct address_space *mapping, struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700178{
Jan Kara9f4e41f2017-11-15 17:37:15 -0800179 if (page_mapped(page)) {
Matthew Wilcox977fbdc2018-01-31 16:17:36 -0800180 pgoff_t nr = PageTransHuge(page) ? HPAGE_PMD_NR : 1;
181 unmap_mapping_pages(mapping, page->index, nr, false);
Jan Kara9f4e41f2017-11-15 17:37:15 -0800182 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700183
David Howells266cf652009-04-03 16:42:36 +0100184 if (page_has_private(page))
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300185 do_invalidatepage(page, 0, PAGE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700186
Konstantin Khlebnikovb9ea2512015-04-14 15:45:27 -0700187 /*
188 * Some filesystems seem to re-dirty the page even after
189 * the VM has canceled the dirty bit (eg ext3 journaling).
190 * Hence dirty accounting check is placed after invalidation.
191 */
Tejun Heo11f81be2015-05-22 17:13:15 -0400192 cancel_dirty_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700193 ClearPageMappedToDisk(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700194}
195
196/*
Andrew Mortonfc0ecff2007-02-10 01:45:39 -0800197 * This is for invalidate_mapping_pages(). That function can be called at
Linus Torvalds1da177e2005-04-16 15:20:36 -0700198 * any time, and is not supposed to throw away dirty pages. But pages can
Nick Piggin0fd0e6b2006-09-27 01:50:02 -0700199 * be marked dirty at any time too, so use remove_mapping which safely
200 * discards clean, unused pages.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700201 *
202 * Returns non-zero if the page was successfully invalidated.
203 */
204static int
205invalidate_complete_page(struct address_space *mapping, struct page *page)
206{
Nick Piggin0fd0e6b2006-09-27 01:50:02 -0700207 int ret;
208
Linus Torvalds1da177e2005-04-16 15:20:36 -0700209 if (page->mapping != mapping)
210 return 0;
211
David Howells266cf652009-04-03 16:42:36 +0100212 if (page_has_private(page) && !try_to_release_page(page, 0))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700213 return 0;
214
Nick Piggin0fd0e6b2006-09-27 01:50:02 -0700215 ret = remove_mapping(mapping, page);
Nick Piggin0fd0e6b2006-09-27 01:50:02 -0700216
217 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700218}
219
Nick Piggin750b4982009-09-16 11:50:12 +0200220int truncate_inode_page(struct address_space *mapping, struct page *page)
221{
Kirill A. Shutemovfc127da2016-07-26 15:26:07 -0700222 VM_BUG_ON_PAGE(PageTail(page), page);
223
Jan Kara9f4e41f2017-11-15 17:37:15 -0800224 if (page->mapping != mapping)
225 return -EIO;
226
227 truncate_cleanup_page(mapping, page);
228 delete_from_page_cache(page);
229 return 0;
Nick Piggin750b4982009-09-16 11:50:12 +0200230}
231
Wu Fengguang83f78662009-09-16 11:50:13 +0200232/*
Andi Kleen25718732009-09-16 11:50:13 +0200233 * Used to get rid of pages on hardware memory corruption.
234 */
235int generic_error_remove_page(struct address_space *mapping, struct page *page)
236{
237 if (!mapping)
238 return -EINVAL;
239 /*
240 * Only punch for normal data pages for now.
241 * Handling other types like directories would need more auditing.
242 */
243 if (!S_ISREG(mapping->host->i_mode))
244 return -EIO;
245 return truncate_inode_page(mapping, page);
246}
247EXPORT_SYMBOL(generic_error_remove_page);
248
249/*
Wu Fengguang83f78662009-09-16 11:50:13 +0200250 * Safely invalidate one page from its pagecache mapping.
251 * It only drops clean, unused pages. The page must be locked.
252 *
253 * Returns 1 if the page is successfully invalidated, otherwise 0.
254 */
255int invalidate_inode_page(struct page *page)
256{
257 struct address_space *mapping = page_mapping(page);
258 if (!mapping)
259 return 0;
260 if (PageDirty(page) || PageWriteback(page))
261 return 0;
262 if (page_mapped(page))
263 return 0;
264 return invalidate_complete_page(mapping, page);
265}
266
Linus Torvalds1da177e2005-04-16 15:20:36 -0700267/**
Liu Bo73c1e202012-02-21 10:57:20 +0800268 * truncate_inode_pages_range - truncate range of pages specified by start & end byte offsets
Linus Torvalds1da177e2005-04-16 15:20:36 -0700269 * @mapping: mapping to truncate
270 * @lstart: offset from which to truncate
Lukas Czerner5a720392013-05-27 23:32:35 -0400271 * @lend: offset to which to truncate (inclusive)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700272 *
Hans Reiserd7339072006-01-06 00:10:36 -0800273 * Truncate the page cache, removing the pages that are between
Lukas Czerner5a720392013-05-27 23:32:35 -0400274 * specified offsets (and zeroing out partial pages
275 * if lstart or lend + 1 is not page aligned).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700276 *
277 * Truncate takes two passes - the first pass is nonblocking. It will not
278 * block on page locks and it will not block on writeback. The second pass
279 * will wait. This is to prevent as much IO as possible in the affected region.
280 * The first pass will remove most pages, so the search cost of the second pass
281 * is low.
282 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700283 * We pass down the cache-hot hint to the page freeing code. Even if the
284 * mapping is large, it is probably the case that the final pages are the most
285 * recently touched, and freeing happens in ascending file offset order.
Lukas Czerner5a720392013-05-27 23:32:35 -0400286 *
287 * Note that since ->invalidatepage() accepts range to invalidate
288 * truncate_inode_pages_range is able to handle cases where lend + 1 is not
289 * page aligned properly.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700290 */
Hans Reiserd7339072006-01-06 00:10:36 -0800291void truncate_inode_pages_range(struct address_space *mapping,
292 loff_t lstart, loff_t lend)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700293{
Lukas Czerner5a720392013-05-27 23:32:35 -0400294 pgoff_t start; /* inclusive */
295 pgoff_t end; /* exclusive */
296 unsigned int partial_start; /* inclusive */
297 unsigned int partial_end; /* exclusive */
298 struct pagevec pvec;
Johannes Weiner0cd61442014-04-03 14:47:46 -0700299 pgoff_t indices[PAGEVEC_SIZE];
Lukas Czerner5a720392013-05-27 23:32:35 -0400300 pgoff_t index;
301 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700302
Ross Zwislerf9fe48b2016-01-22 15:10:40 -0800303 if (mapping->nrpages == 0 && mapping->nrexceptional == 0)
Andrey Ryabinin34ccb692017-05-03 14:56:09 -0700304 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700305
Lukas Czerner5a720392013-05-27 23:32:35 -0400306 /* Offsets within partial pages */
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300307 partial_start = lstart & (PAGE_SIZE - 1);
308 partial_end = (lend + 1) & (PAGE_SIZE - 1);
Lukas Czerner5a720392013-05-27 23:32:35 -0400309
310 /*
311 * 'start' and 'end' always covers the range of pages to be fully
312 * truncated. Partial pages are covered with 'partial_start' at the
313 * start of the range and 'partial_end' at the end of the range.
314 * Note that 'end' is exclusive while 'lend' is inclusive.
315 */
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300316 start = (lstart + PAGE_SIZE - 1) >> PAGE_SHIFT;
Lukas Czerner5a720392013-05-27 23:32:35 -0400317 if (lend == -1)
318 /*
319 * lend == -1 indicates end-of-file so we have to set 'end'
320 * to the highest possible pgoff_t and since the type is
321 * unsigned we're using -1.
322 */
323 end = -1;
324 else
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300325 end = (lend + 1) >> PAGE_SHIFT;
Hans Reiserd7339072006-01-06 00:10:36 -0800326
Mel Gorman86679822017-11-15 17:37:52 -0800327 pagevec_init(&pvec);
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700328 index = start;
Johannes Weiner0cd61442014-04-03 14:47:46 -0700329 while (index < end && pagevec_lookup_entries(&pvec, mapping, index,
330 min(end - index, (pgoff_t)PAGEVEC_SIZE),
331 indices)) {
Jan Karaaa65c292017-11-15 17:37:33 -0800332 /*
333 * Pagevec array has exceptional entries and we may also fail
334 * to lock some pages. So we store pages that can be deleted
335 * in a new pagevec.
336 */
337 struct pagevec locked_pvec;
338
Mel Gorman86679822017-11-15 17:37:52 -0800339 pagevec_init(&locked_pvec);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700340 for (i = 0; i < pagevec_count(&pvec); i++) {
341 struct page *page = pvec.pages[i];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700342
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700343 /* We rely upon deletion not changing page->index */
Johannes Weiner0cd61442014-04-03 14:47:46 -0700344 index = indices[i];
Lukas Czerner5a720392013-05-27 23:32:35 -0400345 if (index >= end)
Hans Reiserd7339072006-01-06 00:10:36 -0800346 break;
Hans Reiserd7339072006-01-06 00:10:36 -0800347
Matthew Wilcox3159f942017-11-03 13:30:42 -0400348 if (xa_is_value(page))
Johannes Weiner0cd61442014-04-03 14:47:46 -0700349 continue;
Johannes Weiner0cd61442014-04-03 14:47:46 -0700350
Nick Piggin529ae9a2008-08-02 12:01:03 +0200351 if (!trylock_page(page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700352 continue;
Kirill A. Shutemov5cbc1982016-11-30 15:54:19 -0800353 WARN_ON(page_to_index(page) != index);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700354 if (PageWriteback(page)) {
355 unlock_page(page);
356 continue;
357 }
Jan Karaaa65c292017-11-15 17:37:33 -0800358 if (page->mapping != mapping) {
359 unlock_page(page);
360 continue;
361 }
362 pagevec_add(&locked_pvec, page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700363 }
Jan Karaaa65c292017-11-15 17:37:33 -0800364 for (i = 0; i < pagevec_count(&locked_pvec); i++)
365 truncate_cleanup_page(mapping, locked_pvec.pages[i]);
366 delete_from_page_cache_batch(mapping, &locked_pvec);
367 for (i = 0; i < pagevec_count(&locked_pvec); i++)
368 unlock_page(locked_pvec.pages[i]);
Mel Gormanf2187592017-11-15 17:37:44 -0800369 truncate_exceptional_pvec_entries(mapping, &pvec, indices, end);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700370 pagevec_release(&pvec);
371 cond_resched();
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700372 index++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700373 }
Lukas Czerner5a720392013-05-27 23:32:35 -0400374 if (partial_start) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700375 struct page *page = find_lock_page(mapping, start - 1);
376 if (page) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300377 unsigned int top = PAGE_SIZE;
Lukas Czerner5a720392013-05-27 23:32:35 -0400378 if (start > end) {
379 /* Truncation within a single page */
380 top = partial_end;
381 partial_end = 0;
382 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700383 wait_on_page_writeback(page);
Lukas Czerner5a720392013-05-27 23:32:35 -0400384 zero_user_segment(page, partial_start, top);
385 cleancache_invalidate_page(mapping, page);
386 if (page_has_private(page))
387 do_invalidatepage(page, partial_start,
388 top - partial_start);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700389 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300390 put_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700391 }
392 }
Lukas Czerner5a720392013-05-27 23:32:35 -0400393 if (partial_end) {
394 struct page *page = find_lock_page(mapping, end);
395 if (page) {
396 wait_on_page_writeback(page);
397 zero_user_segment(page, 0, partial_end);
398 cleancache_invalidate_page(mapping, page);
399 if (page_has_private(page))
400 do_invalidatepage(page, 0,
401 partial_end);
402 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300403 put_page(page);
Lukas Czerner5a720392013-05-27 23:32:35 -0400404 }
405 }
406 /*
407 * If the truncation happened within a single page no pages
408 * will be released, just zeroed, so we can bail out now.
409 */
410 if (start >= end)
Andrey Ryabinin34ccb692017-05-03 14:56:09 -0700411 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700412
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700413 index = start;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700414 for ( ; ; ) {
415 cond_resched();
Johannes Weiner0cd61442014-04-03 14:47:46 -0700416 if (!pagevec_lookup_entries(&pvec, mapping, index,
Hugh Dickins792ceae2014-07-23 14:00:15 -0700417 min(end - index, (pgoff_t)PAGEVEC_SIZE), indices)) {
418 /* If all gone from start onwards, we're done */
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700419 if (index == start)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700420 break;
Hugh Dickins792ceae2014-07-23 14:00:15 -0700421 /* Otherwise restart to make sure all gone */
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700422 index = start;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700423 continue;
424 }
Johannes Weiner0cd61442014-04-03 14:47:46 -0700425 if (index == start && indices[0] >= end) {
Hugh Dickins792ceae2014-07-23 14:00:15 -0700426 /* All gone out of hole to be punched, we're done */
Johannes Weiner0cd61442014-04-03 14:47:46 -0700427 pagevec_remove_exceptionals(&pvec);
Hans Reiserd7339072006-01-06 00:10:36 -0800428 pagevec_release(&pvec);
429 break;
430 }
Mel Gormanf2187592017-11-15 17:37:44 -0800431
Linus Torvalds1da177e2005-04-16 15:20:36 -0700432 for (i = 0; i < pagevec_count(&pvec); i++) {
433 struct page *page = pvec.pages[i];
434
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700435 /* We rely upon deletion not changing page->index */
Johannes Weiner0cd61442014-04-03 14:47:46 -0700436 index = indices[i];
Hugh Dickins792ceae2014-07-23 14:00:15 -0700437 if (index >= end) {
438 /* Restart punch to make sure all gone */
439 index = start - 1;
Hans Reiserd7339072006-01-06 00:10:36 -0800440 break;
Hugh Dickins792ceae2014-07-23 14:00:15 -0700441 }
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700442
Matthew Wilcox3159f942017-11-03 13:30:42 -0400443 if (xa_is_value(page))
Johannes Weiner0cd61442014-04-03 14:47:46 -0700444 continue;
Johannes Weiner0cd61442014-04-03 14:47:46 -0700445
Linus Torvalds1da177e2005-04-16 15:20:36 -0700446 lock_page(page);
Kirill A. Shutemov5cbc1982016-11-30 15:54:19 -0800447 WARN_ON(page_to_index(page) != index);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700448 wait_on_page_writeback(page);
Nick Piggin750b4982009-09-16 11:50:12 +0200449 truncate_inode_page(mapping, page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700450 unlock_page(page);
451 }
Mel Gormanf2187592017-11-15 17:37:44 -0800452 truncate_exceptional_pvec_entries(mapping, &pvec, indices, end);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700453 pagevec_release(&pvec);
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700454 index++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700455 }
Andrey Ryabinin34ccb692017-05-03 14:56:09 -0700456
457out:
Dan Magenheimer31677602011-09-21 11:56:28 -0400458 cleancache_invalidate_inode(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700459}
Hans Reiserd7339072006-01-06 00:10:36 -0800460EXPORT_SYMBOL(truncate_inode_pages_range);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700461
Hans Reiserd7339072006-01-06 00:10:36 -0800462/**
463 * truncate_inode_pages - truncate *all* the pages from an offset
464 * @mapping: mapping to truncate
465 * @lstart: offset from which to truncate
466 *
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800467 * Called under (and serialised by) inode->i_mutex.
Jan Kara08142572011-06-27 16:18:10 -0700468 *
469 * Note: When this function returns, there can be a page in the process of
470 * deletion (inside __delete_from_page_cache()) in the specified range. Thus
471 * mapping->nrpages can be non-zero when this function returns even after
472 * truncation of the whole mapping.
Hans Reiserd7339072006-01-06 00:10:36 -0800473 */
474void truncate_inode_pages(struct address_space *mapping, loff_t lstart)
475{
476 truncate_inode_pages_range(mapping, lstart, (loff_t)-1);
477}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700478EXPORT_SYMBOL(truncate_inode_pages);
479
Mike Waychison28697352009-06-16 15:32:59 -0700480/**
Johannes Weiner91b0abe2014-04-03 14:47:49 -0700481 * truncate_inode_pages_final - truncate *all* pages before inode dies
482 * @mapping: mapping to truncate
483 *
484 * Called under (and serialized by) inode->i_mutex.
485 *
486 * Filesystems have to use this in the .evict_inode path to inform the
487 * VM that this is the final truncate and the inode is going away.
488 */
489void truncate_inode_pages_final(struct address_space *mapping)
490{
Ross Zwislerf9fe48b2016-01-22 15:10:40 -0800491 unsigned long nrexceptional;
Johannes Weiner91b0abe2014-04-03 14:47:49 -0700492 unsigned long nrpages;
493
494 /*
495 * Page reclaim can not participate in regular inode lifetime
496 * management (can't call iput()) and thus can race with the
497 * inode teardown. Tell it when the address space is exiting,
498 * so that it does not install eviction information after the
499 * final truncate has begun.
500 */
501 mapping_set_exiting(mapping);
502
503 /*
504 * When reclaim installs eviction entries, it increases
Ross Zwislerf9fe48b2016-01-22 15:10:40 -0800505 * nrexceptional first, then decreases nrpages. Make sure we see
Johannes Weiner91b0abe2014-04-03 14:47:49 -0700506 * this in the right order or we might miss an entry.
507 */
508 nrpages = mapping->nrpages;
509 smp_rmb();
Ross Zwislerf9fe48b2016-01-22 15:10:40 -0800510 nrexceptional = mapping->nrexceptional;
Johannes Weiner91b0abe2014-04-03 14:47:49 -0700511
Ross Zwislerf9fe48b2016-01-22 15:10:40 -0800512 if (nrpages || nrexceptional) {
Johannes Weiner91b0abe2014-04-03 14:47:49 -0700513 /*
514 * As truncation uses a lockless tree lookup, cycle
515 * the tree lock to make sure any ongoing tree
516 * modification that does not see AS_EXITING is
517 * completed before starting the final truncate.
518 */
Matthew Wilcoxb93b0162018-04-10 16:36:56 -0700519 xa_lock_irq(&mapping->i_pages);
520 xa_unlock_irq(&mapping->i_pages);
Johannes Weiner91b0abe2014-04-03 14:47:49 -0700521 }
Pavel Tikhomirov6ff38bd2018-11-30 14:09:00 -0800522
523 /*
524 * Cleancache needs notification even if there are no pages or shadow
525 * entries.
526 */
527 truncate_inode_pages(mapping, 0);
Johannes Weiner91b0abe2014-04-03 14:47:49 -0700528}
529EXPORT_SYMBOL(truncate_inode_pages_final);
530
531/**
Mike Waychison28697352009-06-16 15:32:59 -0700532 * invalidate_mapping_pages - Invalidate all the unlocked pages of one inode
533 * @mapping: the address_space which holds the pages to invalidate
534 * @start: the offset 'from' which to invalidate
535 * @end: the offset 'to' which to invalidate (inclusive)
536 *
537 * This function only removes the unlocked pages, if you want to
538 * remove all the pages of one inode, you must call truncate_inode_pages.
539 *
540 * invalidate_mapping_pages() will not block on IO activity. It will not
541 * invalidate pages which are dirty, locked, under writeback or mapped into
542 * pagetables.
Mike Rapoporta862f682019-03-05 15:48:42 -0800543 *
544 * Return: the number of the pages that were invalidated
Mike Waychison28697352009-06-16 15:32:59 -0700545 */
546unsigned long invalidate_mapping_pages(struct address_space *mapping,
Minchan Kim31560182011-03-22 16:32:52 -0700547 pgoff_t start, pgoff_t end)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700548{
Johannes Weiner0cd61442014-04-03 14:47:46 -0700549 pgoff_t indices[PAGEVEC_SIZE];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700550 struct pagevec pvec;
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700551 pgoff_t index = start;
Minchan Kim31560182011-03-22 16:32:52 -0700552 unsigned long ret;
553 unsigned long count = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700554 int i;
555
Mel Gorman86679822017-11-15 17:37:52 -0800556 pagevec_init(&pvec);
Johannes Weiner0cd61442014-04-03 14:47:46 -0700557 while (index <= end && pagevec_lookup_entries(&pvec, mapping, index,
558 min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1,
559 indices)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700560 for (i = 0; i < pagevec_count(&pvec); i++) {
561 struct page *page = pvec.pages[i];
562
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700563 /* We rely upon deletion not changing page->index */
Johannes Weiner0cd61442014-04-03 14:47:46 -0700564 index = indices[i];
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700565 if (index > end)
566 break;
NeilBrowne0f236032006-06-23 02:05:48 -0700567
Matthew Wilcox3159f942017-11-03 13:30:42 -0400568 if (xa_is_value(page)) {
Jan Karac6dcf522016-08-10 17:22:44 +0200569 invalidate_exceptional_entry(mapping, index,
570 page);
Johannes Weiner0cd61442014-04-03 14:47:46 -0700571 continue;
572 }
573
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700574 if (!trylock_page(page))
575 continue;
Kirill A. Shutemovfc127da2016-07-26 15:26:07 -0700576
Kirill A. Shutemov5cbc1982016-11-30 15:54:19 -0800577 WARN_ON(page_to_index(page) != index);
Kirill A. Shutemovfc127da2016-07-26 15:26:07 -0700578
579 /* Middle of THP: skip */
580 if (PageTransTail(page)) {
581 unlock_page(page);
582 continue;
583 } else if (PageTransHuge(page)) {
584 index += HPAGE_PMD_NR - 1;
585 i += HPAGE_PMD_NR - 1;
Jan Kara76b6f9b2017-07-10 15:48:59 -0700586 /*
587 * 'end' is in the middle of THP. Don't
588 * invalidate the page as the part outside of
589 * 'end' could be still useful.
590 */
591 if (index > end) {
592 unlock_page(page);
Kirill A. Shutemovfc127da2016-07-26 15:26:07 -0700593 continue;
Jan Kara76b6f9b2017-07-10 15:48:59 -0700594 }
Kirill A. Shutemovef18a1c2019-10-18 20:20:36 -0700595
596 /* Take a pin outside pagevec */
597 get_page(page);
598
599 /*
600 * Drop extra pins before trying to invalidate
601 * the huge page.
602 */
603 pagevec_remove_exceptionals(&pvec);
604 pagevec_release(&pvec);
Kirill A. Shutemovfc127da2016-07-26 15:26:07 -0700605 }
606
Minchan Kim31560182011-03-22 16:32:52 -0700607 ret = invalidate_inode_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700608 unlock_page(page);
Minchan Kim31560182011-03-22 16:32:52 -0700609 /*
610 * Invalidation is a hint that the page is no longer
611 * of interest and try to speed up its reclaim.
612 */
613 if (!ret)
Minchan Kimcc5993b2015-04-15 16:13:26 -0700614 deactivate_file_page(page);
Kirill A. Shutemovef18a1c2019-10-18 20:20:36 -0700615 if (PageTransHuge(page))
616 put_page(page);
Minchan Kim31560182011-03-22 16:32:52 -0700617 count += ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700618 }
Johannes Weiner0cd61442014-04-03 14:47:46 -0700619 pagevec_remove_exceptionals(&pvec);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700620 pagevec_release(&pvec);
Mike Waychison28697352009-06-16 15:32:59 -0700621 cond_resched();
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700622 index++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700623 }
Minchan Kim31560182011-03-22 16:32:52 -0700624 return count;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700625}
Anton Altaparmakov54bc4852007-02-10 01:45:38 -0800626EXPORT_SYMBOL(invalidate_mapping_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700627
Andrew Mortonbd4c8ce2006-09-30 23:29:29 -0700628/*
629 * This is like invalidate_complete_page(), except it ignores the page's
630 * refcount. We do this because invalidate_inode_pages2() needs stronger
631 * invalidation guarantees, and cannot afford to leave pages behind because
Anderson Briglia2706a1b2007-07-15 23:38:09 -0700632 * shrink_page_list() has a temp ref on them, or because they're transiently
633 * sitting in the lru_cache_add() pagevecs.
Andrew Mortonbd4c8ce2006-09-30 23:29:29 -0700634 */
635static int
636invalidate_complete_page2(struct address_space *mapping, struct page *page)
637{
Greg Thelenc4843a72015-05-22 17:13:16 -0400638 unsigned long flags;
639
Andrew Mortonbd4c8ce2006-09-30 23:29:29 -0700640 if (page->mapping != mapping)
641 return 0;
642
David Howells266cf652009-04-03 16:42:36 +0100643 if (page_has_private(page) && !try_to_release_page(page, GFP_KERNEL))
Andrew Mortonbd4c8ce2006-09-30 23:29:29 -0700644 return 0;
645
Matthew Wilcoxb93b0162018-04-10 16:36:56 -0700646 xa_lock_irqsave(&mapping->i_pages, flags);
Andrew Mortonbd4c8ce2006-09-30 23:29:29 -0700647 if (PageDirty(page))
648 goto failed;
649
David Howells266cf652009-04-03 16:42:36 +0100650 BUG_ON(page_has_private(page));
Johannes Weiner62cccb82016-03-15 14:57:22 -0700651 __delete_from_page_cache(page, NULL);
Matthew Wilcoxb93b0162018-04-10 16:36:56 -0700652 xa_unlock_irqrestore(&mapping->i_pages, flags);
Linus Torvalds6072d132010-12-01 13:35:19 -0500653
654 if (mapping->a_ops->freepage)
655 mapping->a_ops->freepage(page);
656
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300657 put_page(page); /* pagecache ref */
Andrew Mortonbd4c8ce2006-09-30 23:29:29 -0700658 return 1;
659failed:
Matthew Wilcoxb93b0162018-04-10 16:36:56 -0700660 xa_unlock_irqrestore(&mapping->i_pages, flags);
Andrew Mortonbd4c8ce2006-09-30 23:29:29 -0700661 return 0;
662}
663
Trond Myklebuste3db7692007-01-10 23:15:39 -0800664static int do_launder_page(struct address_space *mapping, struct page *page)
665{
666 if (!PageDirty(page))
667 return 0;
668 if (page->mapping != mapping || mapping->a_ops->launder_page == NULL)
669 return 0;
670 return mapping->a_ops->launder_page(page);
671}
672
Linus Torvalds1da177e2005-04-16 15:20:36 -0700673/**
674 * invalidate_inode_pages2_range - remove range of pages from an address_space
Martin Waitz67be2dd2005-05-01 08:59:26 -0700675 * @mapping: the address_space
Linus Torvalds1da177e2005-04-16 15:20:36 -0700676 * @start: the page offset 'from' which to invalidate
677 * @end: the page offset 'to' which to invalidate (inclusive)
678 *
679 * Any pages which are found to be mapped into pagetables are unmapped prior to
680 * invalidation.
681 *
Mike Rapoporta862f682019-03-05 15:48:42 -0800682 * Return: -EBUSY if any pages could not be invalidated.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700683 */
684int invalidate_inode_pages2_range(struct address_space *mapping,
685 pgoff_t start, pgoff_t end)
686{
Johannes Weiner0cd61442014-04-03 14:47:46 -0700687 pgoff_t indices[PAGEVEC_SIZE];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700688 struct pagevec pvec;
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700689 pgoff_t index;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700690 int i;
691 int ret = 0;
Hisashi Hifumi0dd13342008-04-28 02:12:08 -0700692 int ret2 = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700693 int did_range_unmap = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700694
Andrey Ryabinin32691f02017-05-03 14:56:06 -0700695 if (mapping->nrpages == 0 && mapping->nrexceptional == 0)
Andrey Ryabinin34ccb692017-05-03 14:56:09 -0700696 goto out;
Andrey Ryabinin32691f02017-05-03 14:56:06 -0700697
Mel Gorman86679822017-11-15 17:37:52 -0800698 pagevec_init(&pvec);
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700699 index = start;
Johannes Weiner0cd61442014-04-03 14:47:46 -0700700 while (index <= end && pagevec_lookup_entries(&pvec, mapping, index,
701 min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1,
702 indices)) {
Trond Myklebust7b965e02007-02-28 20:13:55 -0800703 for (i = 0; i < pagevec_count(&pvec); i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700704 struct page *page = pvec.pages[i];
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700705
706 /* We rely upon deletion not changing page->index */
Johannes Weiner0cd61442014-04-03 14:47:46 -0700707 index = indices[i];
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700708 if (index > end)
709 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700710
Matthew Wilcox3159f942017-11-03 13:30:42 -0400711 if (xa_is_value(page)) {
Jan Karac6dcf522016-08-10 17:22:44 +0200712 if (!invalidate_exceptional_entry2(mapping,
713 index, page))
714 ret = -EBUSY;
Johannes Weiner0cd61442014-04-03 14:47:46 -0700715 continue;
716 }
717
Linus Torvalds1da177e2005-04-16 15:20:36 -0700718 lock_page(page);
Kirill A. Shutemov5cbc1982016-11-30 15:54:19 -0800719 WARN_ON(page_to_index(page) != index);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700720 if (page->mapping != mapping) {
721 unlock_page(page);
722 continue;
723 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700724 wait_on_page_writeback(page);
Nick Piggind00806b2007-07-19 01:46:57 -0700725 if (page_mapped(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700726 if (!did_range_unmap) {
727 /*
728 * Zap the rest of the file in one hit.
729 */
Matthew Wilcox977fbdc2018-01-31 16:17:36 -0800730 unmap_mapping_pages(mapping, index,
731 (1 + end - index), false);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700732 did_range_unmap = 1;
733 } else {
734 /*
735 * Just zap this page
736 */
Matthew Wilcox977fbdc2018-01-31 16:17:36 -0800737 unmap_mapping_pages(mapping, index,
738 1, false);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700739 }
740 }
Nick Piggind00806b2007-07-19 01:46:57 -0700741 BUG_ON(page_mapped(page));
Hisashi Hifumi0dd13342008-04-28 02:12:08 -0700742 ret2 = do_launder_page(mapping, page);
743 if (ret2 == 0) {
744 if (!invalidate_complete_page2(mapping, page))
Hisashi Hifumi6ccfa802008-09-02 14:35:40 -0700745 ret2 = -EBUSY;
Hisashi Hifumi0dd13342008-04-28 02:12:08 -0700746 }
747 if (ret2 < 0)
748 ret = ret2;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700749 unlock_page(page);
750 }
Johannes Weiner0cd61442014-04-03 14:47:46 -0700751 pagevec_remove_exceptionals(&pvec);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700752 pagevec_release(&pvec);
753 cond_resched();
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700754 index++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700755 }
Jan Karacd656372017-05-12 15:46:50 -0700756 /*
Matthew Wilcox69b6c132017-11-25 22:52:46 -0500757 * For DAX we invalidate page tables after invalidating page cache. We
Jan Karacd656372017-05-12 15:46:50 -0700758 * could invalidate page tables while invalidating each entry however
759 * that would be expensive. And doing range unmapping before doesn't
Matthew Wilcox69b6c132017-11-25 22:52:46 -0500760 * work as we have no cheap way to find whether page cache entry didn't
Jan Karacd656372017-05-12 15:46:50 -0700761 * get remapped later.
762 */
763 if (dax_mapping(mapping)) {
Matthew Wilcox977fbdc2018-01-31 16:17:36 -0800764 unmap_mapping_pages(mapping, start, end - start + 1, false);
Jan Karacd656372017-05-12 15:46:50 -0700765 }
Andrey Ryabinin34ccb692017-05-03 14:56:09 -0700766out:
Dan Magenheimer31677602011-09-21 11:56:28 -0400767 cleancache_invalidate_inode(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700768 return ret;
769}
770EXPORT_SYMBOL_GPL(invalidate_inode_pages2_range);
771
772/**
773 * invalidate_inode_pages2 - remove all pages from an address_space
Martin Waitz67be2dd2005-05-01 08:59:26 -0700774 * @mapping: the address_space
Linus Torvalds1da177e2005-04-16 15:20:36 -0700775 *
776 * Any pages which are found to be mapped into pagetables are unmapped prior to
777 * invalidation.
778 *
Mike Rapoporta862f682019-03-05 15:48:42 -0800779 * Return: -EBUSY if any pages could not be invalidated.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700780 */
781int invalidate_inode_pages2(struct address_space *mapping)
782{
783 return invalidate_inode_pages2_range(mapping, 0, -1);
784}
785EXPORT_SYMBOL_GPL(invalidate_inode_pages2);
npiggin@suse.de25d9e2d2009-08-21 02:35:05 +1000786
787/**
788 * truncate_pagecache - unmap and remove pagecache that has been truncated
789 * @inode: inode
Hugh Dickins8a549be2011-07-25 17:12:24 -0700790 * @newsize: new file size
npiggin@suse.de25d9e2d2009-08-21 02:35:05 +1000791 *
792 * inode's new i_size must already be written before truncate_pagecache
793 * is called.
794 *
795 * This function should typically be called before the filesystem
796 * releases resources associated with the freed range (eg. deallocates
797 * blocks). This way, pagecache will always stay logically coherent
798 * with on-disk format, and the filesystem would not have to deal with
799 * situations such as writepage being called for a page that has already
800 * had its underlying blocks deallocated.
801 */
Kirill A. Shutemov7caef262013-09-12 15:13:56 -0700802void truncate_pagecache(struct inode *inode, loff_t newsize)
npiggin@suse.de25d9e2d2009-08-21 02:35:05 +1000803{
OGAWA Hirofumicedabed2010-01-13 21:14:09 +0900804 struct address_space *mapping = inode->i_mapping;
Hugh Dickins8a549be2011-07-25 17:12:24 -0700805 loff_t holebegin = round_up(newsize, PAGE_SIZE);
npiggin@suse.de25d9e2d2009-08-21 02:35:05 +1000806
OGAWA Hirofumicedabed2010-01-13 21:14:09 +0900807 /*
808 * unmap_mapping_range is called twice, first simply for
809 * efficiency so that truncate_inode_pages does fewer
810 * single-page unmaps. However after this first call, and
811 * before truncate_inode_pages finishes, it is possible for
812 * private pages to be COWed, which remain after
813 * truncate_inode_pages finishes, hence the second
814 * unmap_mapping_range call must be made for correctness.
815 */
Hugh Dickins8a549be2011-07-25 17:12:24 -0700816 unmap_mapping_range(mapping, holebegin, 0, 1);
817 truncate_inode_pages(mapping, newsize);
818 unmap_mapping_range(mapping, holebegin, 0, 1);
npiggin@suse.de25d9e2d2009-08-21 02:35:05 +1000819}
820EXPORT_SYMBOL(truncate_pagecache);
821
822/**
Christoph Hellwig2c27c652010-06-04 11:30:04 +0200823 * truncate_setsize - update inode and pagecache for a new file size
824 * @inode: inode
825 * @newsize: new file size
826 *
Jan Kara382e27d2011-01-20 14:44:26 -0800827 * truncate_setsize updates i_size and performs pagecache truncation (if
828 * necessary) to @newsize. It will be typically be called from the filesystem's
829 * setattr function when ATTR_SIZE is passed in.
Christoph Hellwig2c27c652010-06-04 11:30:04 +0200830 *
Jan Kara77783d02014-11-07 08:29:25 +1100831 * Must be called with a lock serializing truncates and writes (generally
832 * i_mutex but e.g. xfs uses a different lock) and before all filesystem
833 * specific block truncation has been performed.
Christoph Hellwig2c27c652010-06-04 11:30:04 +0200834 */
835void truncate_setsize(struct inode *inode, loff_t newsize)
836{
Jan Kara90a80202014-10-01 21:49:18 -0400837 loff_t oldsize = inode->i_size;
838
Christoph Hellwig2c27c652010-06-04 11:30:04 +0200839 i_size_write(inode, newsize);
Jan Kara90a80202014-10-01 21:49:18 -0400840 if (newsize > oldsize)
841 pagecache_isize_extended(inode, oldsize, newsize);
Kirill A. Shutemov7caef262013-09-12 15:13:56 -0700842 truncate_pagecache(inode, newsize);
Christoph Hellwig2c27c652010-06-04 11:30:04 +0200843}
844EXPORT_SYMBOL(truncate_setsize);
845
846/**
Jan Kara90a80202014-10-01 21:49:18 -0400847 * pagecache_isize_extended - update pagecache after extension of i_size
848 * @inode: inode for which i_size was extended
849 * @from: original inode size
850 * @to: new inode size
851 *
852 * Handle extension of inode size either caused by extending truncate or by
853 * write starting after current i_size. We mark the page straddling current
854 * i_size RO so that page_mkwrite() is called on the nearest write access to
855 * the page. This way filesystem can be sure that page_mkwrite() is called on
856 * the page before user writes to the page via mmap after the i_size has been
857 * changed.
858 *
859 * The function must be called after i_size is updated so that page fault
860 * coming after we unlock the page will already see the new i_size.
861 * The function must be called while we still hold i_mutex - this not only
862 * makes sure i_size is stable but also that userspace cannot observe new
863 * i_size value before we are prepared to store mmap writes at new inode size.
864 */
865void pagecache_isize_extended(struct inode *inode, loff_t from, loff_t to)
866{
Fabian Frederick93407472017-02-27 14:28:32 -0800867 int bsize = i_blocksize(inode);
Jan Kara90a80202014-10-01 21:49:18 -0400868 loff_t rounded_from;
869 struct page *page;
870 pgoff_t index;
871
Jan Kara90a80202014-10-01 21:49:18 -0400872 WARN_ON(to > inode->i_size);
873
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300874 if (from >= to || bsize == PAGE_SIZE)
Jan Kara90a80202014-10-01 21:49:18 -0400875 return;
876 /* Page straddling @from will not have any hole block created? */
877 rounded_from = round_up(from, bsize);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300878 if (to <= rounded_from || !(rounded_from & (PAGE_SIZE - 1)))
Jan Kara90a80202014-10-01 21:49:18 -0400879 return;
880
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300881 index = from >> PAGE_SHIFT;
Jan Kara90a80202014-10-01 21:49:18 -0400882 page = find_lock_page(inode->i_mapping, index);
883 /* Page not cached? Nothing to do */
884 if (!page)
885 return;
886 /*
887 * See clear_page_dirty_for_io() for details why set_page_dirty()
888 * is needed.
889 */
890 if (page_mkclean(page))
891 set_page_dirty(page);
892 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300893 put_page(page);
Jan Kara90a80202014-10-01 21:49:18 -0400894}
895EXPORT_SYMBOL(pagecache_isize_extended);
896
897/**
Hugh Dickins623e3db2012-03-28 14:42:40 -0700898 * truncate_pagecache_range - unmap and remove pagecache that is hole-punched
899 * @inode: inode
900 * @lstart: offset of beginning of hole
901 * @lend: offset of last byte of hole
902 *
903 * This function should typically be called before the filesystem
904 * releases resources associated with the freed range (eg. deallocates
905 * blocks). This way, pagecache will always stay logically coherent
906 * with on-disk format, and the filesystem would not have to deal with
907 * situations such as writepage being called for a page that has already
908 * had its underlying blocks deallocated.
909 */
910void truncate_pagecache_range(struct inode *inode, loff_t lstart, loff_t lend)
911{
912 struct address_space *mapping = inode->i_mapping;
913 loff_t unmap_start = round_up(lstart, PAGE_SIZE);
914 loff_t unmap_end = round_down(1 + lend, PAGE_SIZE) - 1;
915 /*
916 * This rounding is currently just for example: unmap_mapping_range
917 * expands its hole outwards, whereas we want it to contract the hole
918 * inwards. However, existing callers of truncate_pagecache_range are
Lukas Czerner5a720392013-05-27 23:32:35 -0400919 * doing their own page rounding first. Note that unmap_mapping_range
920 * allows holelen 0 for all, and we allow lend -1 for end of file.
Hugh Dickins623e3db2012-03-28 14:42:40 -0700921 */
922
923 /*
924 * Unlike in truncate_pagecache, unmap_mapping_range is called only
925 * once (before truncating pagecache), and without "even_cows" flag:
926 * hole-punching should not remove private COWed pages from the hole.
927 */
928 if ((u64)unmap_end > (u64)unmap_start)
929 unmap_mapping_range(mapping, unmap_start,
930 1 + unmap_end - unmap_start, 0);
931 truncate_inode_pages_range(mapping, lstart, lend);
932}
933EXPORT_SYMBOL(truncate_pagecache_range);