blob: e4b4cf0f407068caf2175624731f3d8f9b361098 [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
Mel Gormanf2187592017-11-15 17:37:44 -080028/*
29 * Regular page slots are stabilized by the page lock even without the tree
30 * itself locked. These unlocked entries need verification under the tree
31 * lock.
32 */
33static inline void __clear_shadow_entry(struct address_space *mapping,
34 pgoff_t index, void *entry)
Johannes Weiner0cd61442014-04-03 14:47:46 -070035{
Johannes Weiner449dd692014-04-03 14:47:56 -070036 struct radix_tree_node *node;
37 void **slot;
38
Johannes Weiner14b46872016-12-12 16:43:52 -080039 if (!__radix_tree_lookup(&mapping->page_tree, index, &node, &slot))
Mel Gormanf2187592017-11-15 17:37:44 -080040 return;
Jan Karaac401cc2016-05-12 18:29:18 +020041 if (*slot != entry)
Mel Gormanf2187592017-11-15 17:37:44 -080042 return;
Johannes Weiner14b46872016-12-12 16:43:52 -080043 __radix_tree_replace(&mapping->page_tree, node, slot, NULL,
Mel Gormanc7df8ad2017-11-15 17:37:41 -080044 workingset_update_node);
Jan Karaac401cc2016-05-12 18:29:18 +020045 mapping->nrexceptional--;
Mel Gormanf2187592017-11-15 17:37:44 -080046}
47
48static void clear_shadow_entry(struct address_space *mapping, pgoff_t index,
49 void *entry)
50{
51 spin_lock_irq(&mapping->tree_lock);
52 __clear_shadow_entry(mapping, index, entry);
Johannes Weiner0cd61442014-04-03 14:47:46 -070053 spin_unlock_irq(&mapping->tree_lock);
54}
Linus Torvalds1da177e2005-04-16 15:20:36 -070055
Jan Karac6dcf522016-08-10 17:22:44 +020056/*
Mel Gormanf2187592017-11-15 17:37:44 -080057 * Unconditionally remove exceptional entries. Usually called from truncate
58 * path. Note that the pagevec may be altered by this function by removing
59 * exceptional entries similar to what pagevec_remove_exceptionals does.
Jan Karac6dcf522016-08-10 17:22:44 +020060 */
Mel Gormanf2187592017-11-15 17:37:44 -080061static void truncate_exceptional_pvec_entries(struct address_space *mapping,
62 struct pagevec *pvec, pgoff_t *indices,
63 pgoff_t end)
Jan Karac6dcf522016-08-10 17:22:44 +020064{
Mel Gormanf2187592017-11-15 17:37:44 -080065 int i, j;
66 bool dax, lock;
67
Jan Karac6dcf522016-08-10 17:22:44 +020068 /* Handled by shmem itself */
69 if (shmem_mapping(mapping))
70 return;
71
Mel Gormanf2187592017-11-15 17:37:44 -080072 for (j = 0; j < pagevec_count(pvec); j++)
73 if (radix_tree_exceptional_entry(pvec->pages[j]))
74 break;
75
76 if (j == pagevec_count(pvec))
Jan Karac6dcf522016-08-10 17:22:44 +020077 return;
Mel Gormanf2187592017-11-15 17:37:44 -080078
79 dax = dax_mapping(mapping);
80 lock = !dax && indices[j] < end;
81 if (lock)
82 spin_lock_irq(&mapping->tree_lock);
83
84 for (i = j; i < pagevec_count(pvec); i++) {
85 struct page *page = pvec->pages[i];
86 pgoff_t index = indices[i];
87
88 if (!radix_tree_exceptional_entry(page)) {
89 pvec->pages[j++] = page;
90 continue;
91 }
92
93 if (index >= end)
94 continue;
95
96 if (unlikely(dax)) {
97 dax_delete_mapping_entry(mapping, index);
98 continue;
99 }
100
101 __clear_shadow_entry(mapping, index, page);
Jan Karac6dcf522016-08-10 17:22:44 +0200102 }
Mel Gormanf2187592017-11-15 17:37:44 -0800103
104 if (lock)
105 spin_unlock_irq(&mapping->tree_lock);
106 pvec->nr = j;
Jan Karac6dcf522016-08-10 17:22:44 +0200107}
108
109/*
110 * Invalidate exceptional entry if easily possible. This handles exceptional
Ross Zwisler4636e702017-05-12 15:46:47 -0700111 * entries for invalidate_inode_pages().
Jan Karac6dcf522016-08-10 17:22:44 +0200112 */
113static int invalidate_exceptional_entry(struct address_space *mapping,
114 pgoff_t index, void *entry)
115{
Ross Zwisler4636e702017-05-12 15:46:47 -0700116 /* Handled by shmem itself, or for DAX we do nothing. */
117 if (shmem_mapping(mapping) || dax_mapping(mapping))
Jan Karac6dcf522016-08-10 17:22:44 +0200118 return 1;
Jan Karac6dcf522016-08-10 17:22:44 +0200119 clear_shadow_entry(mapping, index, entry);
120 return 1;
121}
122
123/*
124 * Invalidate exceptional entry if clean. This handles exceptional entries for
125 * invalidate_inode_pages2() so for DAX it evicts only clean entries.
126 */
127static int invalidate_exceptional_entry2(struct address_space *mapping,
128 pgoff_t index, void *entry)
129{
130 /* Handled by shmem itself */
131 if (shmem_mapping(mapping))
132 return 1;
133 if (dax_mapping(mapping))
134 return dax_invalidate_mapping_entry_sync(mapping, index);
135 clear_shadow_entry(mapping, index, entry);
136 return 1;
137}
138
David Howellscf9a2ae2006-08-29 19:05:54 +0100139/**
Fengguang Wu28bc44d2008-02-03 18:04:10 +0200140 * do_invalidatepage - invalidate part or all of a page
David Howellscf9a2ae2006-08-29 19:05:54 +0100141 * @page: the page which is affected
Lukas Czernerd47992f2013-05-21 23:17:23 -0400142 * @offset: start of the range to invalidate
143 * @length: length of the range to invalidate
David Howellscf9a2ae2006-08-29 19:05:54 +0100144 *
145 * do_invalidatepage() is called when all or part of the page has become
146 * invalidated by a truncate operation.
147 *
148 * do_invalidatepage() does not have to release all buffers, but it must
149 * ensure that no dirty buffer is left outside @offset and that no I/O
150 * is underway against any of the blocks which are outside the truncation
151 * point. Because the caller is about to free (and possibly reuse) those
152 * blocks on-disk.
153 */
Lukas Czernerd47992f2013-05-21 23:17:23 -0400154void do_invalidatepage(struct page *page, unsigned int offset,
155 unsigned int length)
David Howellscf9a2ae2006-08-29 19:05:54 +0100156{
Lukas Czernerd47992f2013-05-21 23:17:23 -0400157 void (*invalidatepage)(struct page *, unsigned int, unsigned int);
158
David Howellscf9a2ae2006-08-29 19:05:54 +0100159 invalidatepage = page->mapping->a_ops->invalidatepage;
David Howells93614012006-09-30 20:45:40 +0200160#ifdef CONFIG_BLOCK
David Howellscf9a2ae2006-08-29 19:05:54 +0100161 if (!invalidatepage)
162 invalidatepage = block_invalidatepage;
David Howells93614012006-09-30 20:45:40 +0200163#endif
David Howellscf9a2ae2006-08-29 19:05:54 +0100164 if (invalidatepage)
Lukas Czernerd47992f2013-05-21 23:17:23 -0400165 (*invalidatepage)(page, offset, length);
David Howellscf9a2ae2006-08-29 19:05:54 +0100166}
167
Linus Torvaldsecdfc972007-01-26 12:47:06 -0800168/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700169 * If truncate cannot remove the fs-private metadata from the page, the page
Shaohua Li62e1c552008-02-04 22:29:33 -0800170 * becomes orphaned. It will be left on the LRU and may even be mapped into
Nick Piggin54cb8822007-07-19 01:46:59 -0700171 * user pagetables if we're racing with filemap_fault().
Linus Torvalds1da177e2005-04-16 15:20:36 -0700172 *
173 * We need to bale out if page->mapping is no longer equal to the original
174 * mapping. This happens a) when the VM reclaimed the page while we waited on
Andrew Mortonfc0ecff2007-02-10 01:45:39 -0800175 * its lock, b) when a concurrent invalidate_mapping_pages got there first and
Linus Torvalds1da177e2005-04-16 15:20:36 -0700176 * c) when tmpfs swizzles a page between a tmpfs inode and swapper_space.
177 */
Jan Kara9f4e41f2017-11-15 17:37:15 -0800178static void
179truncate_cleanup_page(struct address_space *mapping, struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700180{
Jan Kara9f4e41f2017-11-15 17:37:15 -0800181 if (page_mapped(page)) {
182 loff_t holelen;
183
184 holelen = PageTransHuge(page) ? HPAGE_PMD_SIZE : PAGE_SIZE;
185 unmap_mapping_range(mapping,
186 (loff_t)page->index << PAGE_SHIFT,
187 holelen, 0);
188 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700189
David Howells266cf652009-04-03 16:42:36 +0100190 if (page_has_private(page))
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300191 do_invalidatepage(page, 0, PAGE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700192
Konstantin Khlebnikovb9ea2512015-04-14 15:45:27 -0700193 /*
194 * Some filesystems seem to re-dirty the page even after
195 * the VM has canceled the dirty bit (eg ext3 journaling).
196 * Hence dirty accounting check is placed after invalidation.
197 */
Tejun Heo11f81be2015-05-22 17:13:15 -0400198 cancel_dirty_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700199 ClearPageMappedToDisk(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700200}
201
202/*
Andrew Mortonfc0ecff2007-02-10 01:45:39 -0800203 * This is for invalidate_mapping_pages(). That function can be called at
Linus Torvalds1da177e2005-04-16 15:20:36 -0700204 * any time, and is not supposed to throw away dirty pages. But pages can
Nick Piggin0fd0e6b2006-09-27 01:50:02 -0700205 * be marked dirty at any time too, so use remove_mapping which safely
206 * discards clean, unused pages.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700207 *
208 * Returns non-zero if the page was successfully invalidated.
209 */
210static int
211invalidate_complete_page(struct address_space *mapping, struct page *page)
212{
Nick Piggin0fd0e6b2006-09-27 01:50:02 -0700213 int ret;
214
Linus Torvalds1da177e2005-04-16 15:20:36 -0700215 if (page->mapping != mapping)
216 return 0;
217
David Howells266cf652009-04-03 16:42:36 +0100218 if (page_has_private(page) && !try_to_release_page(page, 0))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700219 return 0;
220
Nick Piggin0fd0e6b2006-09-27 01:50:02 -0700221 ret = remove_mapping(mapping, page);
Nick Piggin0fd0e6b2006-09-27 01:50:02 -0700222
223 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700224}
225
Nick Piggin750b4982009-09-16 11:50:12 +0200226int truncate_inode_page(struct address_space *mapping, struct page *page)
227{
Kirill A. Shutemovfc127da2016-07-26 15:26:07 -0700228 VM_BUG_ON_PAGE(PageTail(page), page);
229
Jan Kara9f4e41f2017-11-15 17:37:15 -0800230 if (page->mapping != mapping)
231 return -EIO;
232
233 truncate_cleanup_page(mapping, page);
234 delete_from_page_cache(page);
235 return 0;
Nick Piggin750b4982009-09-16 11:50:12 +0200236}
237
Wu Fengguang83f78662009-09-16 11:50:13 +0200238/*
Andi Kleen25718732009-09-16 11:50:13 +0200239 * Used to get rid of pages on hardware memory corruption.
240 */
241int generic_error_remove_page(struct address_space *mapping, struct page *page)
242{
243 if (!mapping)
244 return -EINVAL;
245 /*
246 * Only punch for normal data pages for now.
247 * Handling other types like directories would need more auditing.
248 */
249 if (!S_ISREG(mapping->host->i_mode))
250 return -EIO;
251 return truncate_inode_page(mapping, page);
252}
253EXPORT_SYMBOL(generic_error_remove_page);
254
255/*
Wu Fengguang83f78662009-09-16 11:50:13 +0200256 * Safely invalidate one page from its pagecache mapping.
257 * It only drops clean, unused pages. The page must be locked.
258 *
259 * Returns 1 if the page is successfully invalidated, otherwise 0.
260 */
261int invalidate_inode_page(struct page *page)
262{
263 struct address_space *mapping = page_mapping(page);
264 if (!mapping)
265 return 0;
266 if (PageDirty(page) || PageWriteback(page))
267 return 0;
268 if (page_mapped(page))
269 return 0;
270 return invalidate_complete_page(mapping, page);
271}
272
Linus Torvalds1da177e2005-04-16 15:20:36 -0700273/**
Liu Bo73c1e202012-02-21 10:57:20 +0800274 * truncate_inode_pages_range - truncate range of pages specified by start & end byte offsets
Linus Torvalds1da177e2005-04-16 15:20:36 -0700275 * @mapping: mapping to truncate
276 * @lstart: offset from which to truncate
Lukas Czerner5a720392013-05-27 23:32:35 -0400277 * @lend: offset to which to truncate (inclusive)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700278 *
Hans Reiserd7339072006-01-06 00:10:36 -0800279 * Truncate the page cache, removing the pages that are between
Lukas Czerner5a720392013-05-27 23:32:35 -0400280 * specified offsets (and zeroing out partial pages
281 * if lstart or lend + 1 is not page aligned).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700282 *
283 * Truncate takes two passes - the first pass is nonblocking. It will not
284 * block on page locks and it will not block on writeback. The second pass
285 * will wait. This is to prevent as much IO as possible in the affected region.
286 * The first pass will remove most pages, so the search cost of the second pass
287 * is low.
288 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700289 * We pass down the cache-hot hint to the page freeing code. Even if the
290 * mapping is large, it is probably the case that the final pages are the most
291 * recently touched, and freeing happens in ascending file offset order.
Lukas Czerner5a720392013-05-27 23:32:35 -0400292 *
293 * Note that since ->invalidatepage() accepts range to invalidate
294 * truncate_inode_pages_range is able to handle cases where lend + 1 is not
295 * page aligned properly.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700296 */
Hans Reiserd7339072006-01-06 00:10:36 -0800297void truncate_inode_pages_range(struct address_space *mapping,
298 loff_t lstart, loff_t lend)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700299{
Lukas Czerner5a720392013-05-27 23:32:35 -0400300 pgoff_t start; /* inclusive */
301 pgoff_t end; /* exclusive */
302 unsigned int partial_start; /* inclusive */
303 unsigned int partial_end; /* exclusive */
304 struct pagevec pvec;
Johannes Weiner0cd61442014-04-03 14:47:46 -0700305 pgoff_t indices[PAGEVEC_SIZE];
Lukas Czerner5a720392013-05-27 23:32:35 -0400306 pgoff_t index;
307 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700308
Ross Zwislerf9fe48b2016-01-22 15:10:40 -0800309 if (mapping->nrpages == 0 && mapping->nrexceptional == 0)
Andrey Ryabinin34ccb692017-05-03 14:56:09 -0700310 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700311
Lukas Czerner5a720392013-05-27 23:32:35 -0400312 /* Offsets within partial pages */
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300313 partial_start = lstart & (PAGE_SIZE - 1);
314 partial_end = (lend + 1) & (PAGE_SIZE - 1);
Lukas Czerner5a720392013-05-27 23:32:35 -0400315
316 /*
317 * 'start' and 'end' always covers the range of pages to be fully
318 * truncated. Partial pages are covered with 'partial_start' at the
319 * start of the range and 'partial_end' at the end of the range.
320 * Note that 'end' is exclusive while 'lend' is inclusive.
321 */
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300322 start = (lstart + PAGE_SIZE - 1) >> PAGE_SHIFT;
Lukas Czerner5a720392013-05-27 23:32:35 -0400323 if (lend == -1)
324 /*
325 * lend == -1 indicates end-of-file so we have to set 'end'
326 * to the highest possible pgoff_t and since the type is
327 * unsigned we're using -1.
328 */
329 end = -1;
330 else
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300331 end = (lend + 1) >> PAGE_SHIFT;
Hans Reiserd7339072006-01-06 00:10:36 -0800332
Mel Gorman86679822017-11-15 17:37:52 -0800333 pagevec_init(&pvec);
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700334 index = start;
Johannes Weiner0cd61442014-04-03 14:47:46 -0700335 while (index < end && pagevec_lookup_entries(&pvec, mapping, index,
336 min(end - index, (pgoff_t)PAGEVEC_SIZE),
337 indices)) {
Jan Karaaa65c292017-11-15 17:37:33 -0800338 /*
339 * Pagevec array has exceptional entries and we may also fail
340 * to lock some pages. So we store pages that can be deleted
341 * in a new pagevec.
342 */
343 struct pagevec locked_pvec;
344
Mel Gorman86679822017-11-15 17:37:52 -0800345 pagevec_init(&locked_pvec);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700346 for (i = 0; i < pagevec_count(&pvec); i++) {
347 struct page *page = pvec.pages[i];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700348
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700349 /* We rely upon deletion not changing page->index */
Johannes Weiner0cd61442014-04-03 14:47:46 -0700350 index = indices[i];
Lukas Czerner5a720392013-05-27 23:32:35 -0400351 if (index >= end)
Hans Reiserd7339072006-01-06 00:10:36 -0800352 break;
Hans Reiserd7339072006-01-06 00:10:36 -0800353
Mel Gormanf2187592017-11-15 17:37:44 -0800354 if (radix_tree_exceptional_entry(page))
Johannes Weiner0cd61442014-04-03 14:47:46 -0700355 continue;
Johannes Weiner0cd61442014-04-03 14:47:46 -0700356
Nick Piggin529ae9a2008-08-02 12:01:03 +0200357 if (!trylock_page(page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700358 continue;
Kirill A. Shutemov5cbc1982016-11-30 15:54:19 -0800359 WARN_ON(page_to_index(page) != index);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700360 if (PageWriteback(page)) {
361 unlock_page(page);
362 continue;
363 }
Jan Karaaa65c292017-11-15 17:37:33 -0800364 if (page->mapping != mapping) {
365 unlock_page(page);
366 continue;
367 }
368 pagevec_add(&locked_pvec, page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700369 }
Jan Karaaa65c292017-11-15 17:37:33 -0800370 for (i = 0; i < pagevec_count(&locked_pvec); i++)
371 truncate_cleanup_page(mapping, locked_pvec.pages[i]);
372 delete_from_page_cache_batch(mapping, &locked_pvec);
373 for (i = 0; i < pagevec_count(&locked_pvec); i++)
374 unlock_page(locked_pvec.pages[i]);
Mel Gormanf2187592017-11-15 17:37:44 -0800375 truncate_exceptional_pvec_entries(mapping, &pvec, indices, end);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700376 pagevec_release(&pvec);
377 cond_resched();
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700378 index++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700379 }
Lukas Czerner5a720392013-05-27 23:32:35 -0400380 if (partial_start) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700381 struct page *page = find_lock_page(mapping, start - 1);
382 if (page) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300383 unsigned int top = PAGE_SIZE;
Lukas Czerner5a720392013-05-27 23:32:35 -0400384 if (start > end) {
385 /* Truncation within a single page */
386 top = partial_end;
387 partial_end = 0;
388 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700389 wait_on_page_writeback(page);
Lukas Czerner5a720392013-05-27 23:32:35 -0400390 zero_user_segment(page, partial_start, top);
391 cleancache_invalidate_page(mapping, page);
392 if (page_has_private(page))
393 do_invalidatepage(page, partial_start,
394 top - partial_start);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700395 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300396 put_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700397 }
398 }
Lukas Czerner5a720392013-05-27 23:32:35 -0400399 if (partial_end) {
400 struct page *page = find_lock_page(mapping, end);
401 if (page) {
402 wait_on_page_writeback(page);
403 zero_user_segment(page, 0, partial_end);
404 cleancache_invalidate_page(mapping, page);
405 if (page_has_private(page))
406 do_invalidatepage(page, 0,
407 partial_end);
408 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300409 put_page(page);
Lukas Czerner5a720392013-05-27 23:32:35 -0400410 }
411 }
412 /*
413 * If the truncation happened within a single page no pages
414 * will be released, just zeroed, so we can bail out now.
415 */
416 if (start >= end)
Andrey Ryabinin34ccb692017-05-03 14:56:09 -0700417 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700418
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700419 index = start;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700420 for ( ; ; ) {
421 cond_resched();
Johannes Weiner0cd61442014-04-03 14:47:46 -0700422 if (!pagevec_lookup_entries(&pvec, mapping, index,
Hugh Dickins792ceae2014-07-23 14:00:15 -0700423 min(end - index, (pgoff_t)PAGEVEC_SIZE), indices)) {
424 /* If all gone from start onwards, we're done */
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700425 if (index == start)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700426 break;
Hugh Dickins792ceae2014-07-23 14:00:15 -0700427 /* Otherwise restart to make sure all gone */
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700428 index = start;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700429 continue;
430 }
Johannes Weiner0cd61442014-04-03 14:47:46 -0700431 if (index == start && indices[0] >= end) {
Hugh Dickins792ceae2014-07-23 14:00:15 -0700432 /* All gone out of hole to be punched, we're done */
Johannes Weiner0cd61442014-04-03 14:47:46 -0700433 pagevec_remove_exceptionals(&pvec);
Hans Reiserd7339072006-01-06 00:10:36 -0800434 pagevec_release(&pvec);
435 break;
436 }
Mel Gormanf2187592017-11-15 17:37:44 -0800437
Linus Torvalds1da177e2005-04-16 15:20:36 -0700438 for (i = 0; i < pagevec_count(&pvec); i++) {
439 struct page *page = pvec.pages[i];
440
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700441 /* We rely upon deletion not changing page->index */
Johannes Weiner0cd61442014-04-03 14:47:46 -0700442 index = indices[i];
Hugh Dickins792ceae2014-07-23 14:00:15 -0700443 if (index >= end) {
444 /* Restart punch to make sure all gone */
445 index = start - 1;
Hans Reiserd7339072006-01-06 00:10:36 -0800446 break;
Hugh Dickins792ceae2014-07-23 14:00:15 -0700447 }
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700448
Mel Gormanf2187592017-11-15 17:37:44 -0800449 if (radix_tree_exceptional_entry(page))
Johannes Weiner0cd61442014-04-03 14:47:46 -0700450 continue;
Johannes Weiner0cd61442014-04-03 14:47:46 -0700451
Linus Torvalds1da177e2005-04-16 15:20:36 -0700452 lock_page(page);
Kirill A. Shutemov5cbc1982016-11-30 15:54:19 -0800453 WARN_ON(page_to_index(page) != index);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700454 wait_on_page_writeback(page);
Nick Piggin750b4982009-09-16 11:50:12 +0200455 truncate_inode_page(mapping, page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700456 unlock_page(page);
457 }
Mel Gormanf2187592017-11-15 17:37:44 -0800458 truncate_exceptional_pvec_entries(mapping, &pvec, indices, end);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700459 pagevec_release(&pvec);
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700460 index++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700461 }
Andrey Ryabinin34ccb692017-05-03 14:56:09 -0700462
463out:
Dan Magenheimer31677602011-09-21 11:56:28 -0400464 cleancache_invalidate_inode(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700465}
Hans Reiserd7339072006-01-06 00:10:36 -0800466EXPORT_SYMBOL(truncate_inode_pages_range);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700467
Hans Reiserd7339072006-01-06 00:10:36 -0800468/**
469 * truncate_inode_pages - truncate *all* the pages from an offset
470 * @mapping: mapping to truncate
471 * @lstart: offset from which to truncate
472 *
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800473 * Called under (and serialised by) inode->i_mutex.
Jan Kara08142572011-06-27 16:18:10 -0700474 *
475 * Note: When this function returns, there can be a page in the process of
476 * deletion (inside __delete_from_page_cache()) in the specified range. Thus
477 * mapping->nrpages can be non-zero when this function returns even after
478 * truncation of the whole mapping.
Hans Reiserd7339072006-01-06 00:10:36 -0800479 */
480void truncate_inode_pages(struct address_space *mapping, loff_t lstart)
481{
482 truncate_inode_pages_range(mapping, lstart, (loff_t)-1);
483}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700484EXPORT_SYMBOL(truncate_inode_pages);
485
Mike Waychison28697352009-06-16 15:32:59 -0700486/**
Johannes Weiner91b0abe2014-04-03 14:47:49 -0700487 * truncate_inode_pages_final - truncate *all* pages before inode dies
488 * @mapping: mapping to truncate
489 *
490 * Called under (and serialized by) inode->i_mutex.
491 *
492 * Filesystems have to use this in the .evict_inode path to inform the
493 * VM that this is the final truncate and the inode is going away.
494 */
495void truncate_inode_pages_final(struct address_space *mapping)
496{
Ross Zwislerf9fe48b2016-01-22 15:10:40 -0800497 unsigned long nrexceptional;
Johannes Weiner91b0abe2014-04-03 14:47:49 -0700498 unsigned long nrpages;
499
500 /*
501 * Page reclaim can not participate in regular inode lifetime
502 * management (can't call iput()) and thus can race with the
503 * inode teardown. Tell it when the address space is exiting,
504 * so that it does not install eviction information after the
505 * final truncate has begun.
506 */
507 mapping_set_exiting(mapping);
508
509 /*
510 * When reclaim installs eviction entries, it increases
Ross Zwislerf9fe48b2016-01-22 15:10:40 -0800511 * nrexceptional first, then decreases nrpages. Make sure we see
Johannes Weiner91b0abe2014-04-03 14:47:49 -0700512 * this in the right order or we might miss an entry.
513 */
514 nrpages = mapping->nrpages;
515 smp_rmb();
Ross Zwislerf9fe48b2016-01-22 15:10:40 -0800516 nrexceptional = mapping->nrexceptional;
Johannes Weiner91b0abe2014-04-03 14:47:49 -0700517
Ross Zwislerf9fe48b2016-01-22 15:10:40 -0800518 if (nrpages || nrexceptional) {
Johannes Weiner91b0abe2014-04-03 14:47:49 -0700519 /*
520 * As truncation uses a lockless tree lookup, cycle
521 * the tree lock to make sure any ongoing tree
522 * modification that does not see AS_EXITING is
523 * completed before starting the final truncate.
524 */
525 spin_lock_irq(&mapping->tree_lock);
526 spin_unlock_irq(&mapping->tree_lock);
527
528 truncate_inode_pages(mapping, 0);
529 }
530}
531EXPORT_SYMBOL(truncate_inode_pages_final);
532
533/**
Mike Waychison28697352009-06-16 15:32:59 -0700534 * invalidate_mapping_pages - Invalidate all the unlocked pages of one inode
535 * @mapping: the address_space which holds the pages to invalidate
536 * @start: the offset 'from' which to invalidate
537 * @end: the offset 'to' which to invalidate (inclusive)
538 *
539 * This function only removes the unlocked pages, if you want to
540 * remove all the pages of one inode, you must call truncate_inode_pages.
541 *
542 * invalidate_mapping_pages() will not block on IO activity. It will not
543 * invalidate pages which are dirty, locked, under writeback or mapped into
544 * pagetables.
545 */
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;
NeilBrowne0f23602006-06-23 02:05:48 -0700567
Johannes Weiner0cd61442014-04-03 14:47:46 -0700568 if (radix_tree_exceptional_entry(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. Shutemovfc127da2016-07-26 15:26:07 -0700595 }
596
Minchan Kim31560182011-03-22 16:32:52 -0700597 ret = invalidate_inode_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700598 unlock_page(page);
Minchan Kim31560182011-03-22 16:32:52 -0700599 /*
600 * Invalidation is a hint that the page is no longer
601 * of interest and try to speed up its reclaim.
602 */
603 if (!ret)
Minchan Kimcc5993b2015-04-15 16:13:26 -0700604 deactivate_file_page(page);
Minchan Kim31560182011-03-22 16:32:52 -0700605 count += ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700606 }
Johannes Weiner0cd61442014-04-03 14:47:46 -0700607 pagevec_remove_exceptionals(&pvec);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700608 pagevec_release(&pvec);
Mike Waychison28697352009-06-16 15:32:59 -0700609 cond_resched();
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700610 index++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700611 }
Minchan Kim31560182011-03-22 16:32:52 -0700612 return count;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700613}
Anton Altaparmakov54bc4852007-02-10 01:45:38 -0800614EXPORT_SYMBOL(invalidate_mapping_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700615
Andrew Mortonbd4c8ce2006-09-30 23:29:29 -0700616/*
617 * This is like invalidate_complete_page(), except it ignores the page's
618 * refcount. We do this because invalidate_inode_pages2() needs stronger
619 * invalidation guarantees, and cannot afford to leave pages behind because
Anderson Briglia2706a1b2007-07-15 23:38:09 -0700620 * shrink_page_list() has a temp ref on them, or because they're transiently
621 * sitting in the lru_cache_add() pagevecs.
Andrew Mortonbd4c8ce2006-09-30 23:29:29 -0700622 */
623static int
624invalidate_complete_page2(struct address_space *mapping, struct page *page)
625{
Greg Thelenc4843a72015-05-22 17:13:16 -0400626 unsigned long flags;
627
Andrew Mortonbd4c8ce2006-09-30 23:29:29 -0700628 if (page->mapping != mapping)
629 return 0;
630
David Howells266cf652009-04-03 16:42:36 +0100631 if (page_has_private(page) && !try_to_release_page(page, GFP_KERNEL))
Andrew Mortonbd4c8ce2006-09-30 23:29:29 -0700632 return 0;
633
Greg Thelenc4843a72015-05-22 17:13:16 -0400634 spin_lock_irqsave(&mapping->tree_lock, flags);
Andrew Mortonbd4c8ce2006-09-30 23:29:29 -0700635 if (PageDirty(page))
636 goto failed;
637
David Howells266cf652009-04-03 16:42:36 +0100638 BUG_ON(page_has_private(page));
Johannes Weiner62cccb82016-03-15 14:57:22 -0700639 __delete_from_page_cache(page, NULL);
Greg Thelenc4843a72015-05-22 17:13:16 -0400640 spin_unlock_irqrestore(&mapping->tree_lock, flags);
Linus Torvalds6072d132010-12-01 13:35:19 -0500641
642 if (mapping->a_ops->freepage)
643 mapping->a_ops->freepage(page);
644
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300645 put_page(page); /* pagecache ref */
Andrew Mortonbd4c8ce2006-09-30 23:29:29 -0700646 return 1;
647failed:
Greg Thelenc4843a72015-05-22 17:13:16 -0400648 spin_unlock_irqrestore(&mapping->tree_lock, flags);
Andrew Mortonbd4c8ce2006-09-30 23:29:29 -0700649 return 0;
650}
651
Trond Myklebuste3db7692007-01-10 23:15:39 -0800652static int do_launder_page(struct address_space *mapping, struct page *page)
653{
654 if (!PageDirty(page))
655 return 0;
656 if (page->mapping != mapping || mapping->a_ops->launder_page == NULL)
657 return 0;
658 return mapping->a_ops->launder_page(page);
659}
660
Linus Torvalds1da177e2005-04-16 15:20:36 -0700661/**
662 * invalidate_inode_pages2_range - remove range of pages from an address_space
Martin Waitz67be2dd2005-05-01 08:59:26 -0700663 * @mapping: the address_space
Linus Torvalds1da177e2005-04-16 15:20:36 -0700664 * @start: the page offset 'from' which to invalidate
665 * @end: the page offset 'to' which to invalidate (inclusive)
666 *
667 * Any pages which are found to be mapped into pagetables are unmapped prior to
668 * invalidation.
669 *
Hisashi Hifumi6ccfa802008-09-02 14:35:40 -0700670 * Returns -EBUSY if any pages could not be invalidated.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700671 */
672int invalidate_inode_pages2_range(struct address_space *mapping,
673 pgoff_t start, pgoff_t end)
674{
Johannes Weiner0cd61442014-04-03 14:47:46 -0700675 pgoff_t indices[PAGEVEC_SIZE];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700676 struct pagevec pvec;
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700677 pgoff_t index;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700678 int i;
679 int ret = 0;
Hisashi Hifumi0dd13342008-04-28 02:12:08 -0700680 int ret2 = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700681 int did_range_unmap = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700682
Andrey Ryabinin32691f02017-05-03 14:56:06 -0700683 if (mapping->nrpages == 0 && mapping->nrexceptional == 0)
Andrey Ryabinin34ccb692017-05-03 14:56:09 -0700684 goto out;
Andrey Ryabinin32691f02017-05-03 14:56:06 -0700685
Mel Gorman86679822017-11-15 17:37:52 -0800686 pagevec_init(&pvec);
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700687 index = start;
Johannes Weiner0cd61442014-04-03 14:47:46 -0700688 while (index <= end && pagevec_lookup_entries(&pvec, mapping, index,
689 min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1,
690 indices)) {
Trond Myklebust7b965e02007-02-28 20:13:55 -0800691 for (i = 0; i < pagevec_count(&pvec); i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700692 struct page *page = pvec.pages[i];
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700693
694 /* We rely upon deletion not changing page->index */
Johannes Weiner0cd61442014-04-03 14:47:46 -0700695 index = indices[i];
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700696 if (index > end)
697 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700698
Johannes Weiner0cd61442014-04-03 14:47:46 -0700699 if (radix_tree_exceptional_entry(page)) {
Jan Karac6dcf522016-08-10 17:22:44 +0200700 if (!invalidate_exceptional_entry2(mapping,
701 index, page))
702 ret = -EBUSY;
Johannes Weiner0cd61442014-04-03 14:47:46 -0700703 continue;
704 }
705
Linus Torvalds1da177e2005-04-16 15:20:36 -0700706 lock_page(page);
Kirill A. Shutemov5cbc1982016-11-30 15:54:19 -0800707 WARN_ON(page_to_index(page) != index);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700708 if (page->mapping != mapping) {
709 unlock_page(page);
710 continue;
711 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700712 wait_on_page_writeback(page);
Nick Piggind00806b2007-07-19 01:46:57 -0700713 if (page_mapped(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700714 if (!did_range_unmap) {
715 /*
716 * Zap the rest of the file in one hit.
717 */
718 unmap_mapping_range(mapping,
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300719 (loff_t)index << PAGE_SHIFT,
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700720 (loff_t)(1 + end - index)
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300721 << PAGE_SHIFT,
722 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700723 did_range_unmap = 1;
724 } else {
725 /*
726 * Just zap this page
727 */
728 unmap_mapping_range(mapping,
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300729 (loff_t)index << PAGE_SHIFT,
730 PAGE_SIZE, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700731 }
732 }
Nick Piggind00806b2007-07-19 01:46:57 -0700733 BUG_ON(page_mapped(page));
Hisashi Hifumi0dd13342008-04-28 02:12:08 -0700734 ret2 = do_launder_page(mapping, page);
735 if (ret2 == 0) {
736 if (!invalidate_complete_page2(mapping, page))
Hisashi Hifumi6ccfa802008-09-02 14:35:40 -0700737 ret2 = -EBUSY;
Hisashi Hifumi0dd13342008-04-28 02:12:08 -0700738 }
739 if (ret2 < 0)
740 ret = ret2;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700741 unlock_page(page);
742 }
Johannes Weiner0cd61442014-04-03 14:47:46 -0700743 pagevec_remove_exceptionals(&pvec);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700744 pagevec_release(&pvec);
745 cond_resched();
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700746 index++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700747 }
Jan Karacd656372017-05-12 15:46:50 -0700748 /*
749 * For DAX we invalidate page tables after invalidating radix tree. We
750 * could invalidate page tables while invalidating each entry however
751 * that would be expensive. And doing range unmapping before doesn't
752 * work as we have no cheap way to find whether radix tree entry didn't
753 * get remapped later.
754 */
755 if (dax_mapping(mapping)) {
756 unmap_mapping_range(mapping, (loff_t)start << PAGE_SHIFT,
757 (loff_t)(end - start + 1) << PAGE_SHIFT, 0);
758 }
Andrey Ryabinin34ccb692017-05-03 14:56:09 -0700759out:
Dan Magenheimer31677602011-09-21 11:56:28 -0400760 cleancache_invalidate_inode(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700761 return ret;
762}
763EXPORT_SYMBOL_GPL(invalidate_inode_pages2_range);
764
765/**
766 * invalidate_inode_pages2 - remove all pages from an address_space
Martin Waitz67be2dd2005-05-01 08:59:26 -0700767 * @mapping: the address_space
Linus Torvalds1da177e2005-04-16 15:20:36 -0700768 *
769 * Any pages which are found to be mapped into pagetables are unmapped prior to
770 * invalidation.
771 *
Peng Taoe9de25d2009-10-19 14:48:13 +0800772 * Returns -EBUSY if any pages could not be invalidated.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700773 */
774int invalidate_inode_pages2(struct address_space *mapping)
775{
776 return invalidate_inode_pages2_range(mapping, 0, -1);
777}
778EXPORT_SYMBOL_GPL(invalidate_inode_pages2);
npiggin@suse.de25d9e2d2009-08-21 02:35:05 +1000779
780/**
781 * truncate_pagecache - unmap and remove pagecache that has been truncated
782 * @inode: inode
Hugh Dickins8a549be2011-07-25 17:12:24 -0700783 * @newsize: new file size
npiggin@suse.de25d9e2d2009-08-21 02:35:05 +1000784 *
785 * inode's new i_size must already be written before truncate_pagecache
786 * is called.
787 *
788 * This function should typically be called before the filesystem
789 * releases resources associated with the freed range (eg. deallocates
790 * blocks). This way, pagecache will always stay logically coherent
791 * with on-disk format, and the filesystem would not have to deal with
792 * situations such as writepage being called for a page that has already
793 * had its underlying blocks deallocated.
794 */
Kirill A. Shutemov7caef262013-09-12 15:13:56 -0700795void truncate_pagecache(struct inode *inode, loff_t newsize)
npiggin@suse.de25d9e2d2009-08-21 02:35:05 +1000796{
OGAWA Hirofumicedabed2010-01-13 21:14:09 +0900797 struct address_space *mapping = inode->i_mapping;
Hugh Dickins8a549be2011-07-25 17:12:24 -0700798 loff_t holebegin = round_up(newsize, PAGE_SIZE);
npiggin@suse.de25d9e2d2009-08-21 02:35:05 +1000799
OGAWA Hirofumicedabed2010-01-13 21:14:09 +0900800 /*
801 * unmap_mapping_range is called twice, first simply for
802 * efficiency so that truncate_inode_pages does fewer
803 * single-page unmaps. However after this first call, and
804 * before truncate_inode_pages finishes, it is possible for
805 * private pages to be COWed, which remain after
806 * truncate_inode_pages finishes, hence the second
807 * unmap_mapping_range call must be made for correctness.
808 */
Hugh Dickins8a549be2011-07-25 17:12:24 -0700809 unmap_mapping_range(mapping, holebegin, 0, 1);
810 truncate_inode_pages(mapping, newsize);
811 unmap_mapping_range(mapping, holebegin, 0, 1);
npiggin@suse.de25d9e2d2009-08-21 02:35:05 +1000812}
813EXPORT_SYMBOL(truncate_pagecache);
814
815/**
Christoph Hellwig2c27c652010-06-04 11:30:04 +0200816 * truncate_setsize - update inode and pagecache for a new file size
817 * @inode: inode
818 * @newsize: new file size
819 *
Jan Kara382e27d2011-01-20 14:44:26 -0800820 * truncate_setsize updates i_size and performs pagecache truncation (if
821 * necessary) to @newsize. It will be typically be called from the filesystem's
822 * setattr function when ATTR_SIZE is passed in.
Christoph Hellwig2c27c652010-06-04 11:30:04 +0200823 *
Jan Kara77783d02014-11-07 08:29:25 +1100824 * Must be called with a lock serializing truncates and writes (generally
825 * i_mutex but e.g. xfs uses a different lock) and before all filesystem
826 * specific block truncation has been performed.
Christoph Hellwig2c27c652010-06-04 11:30:04 +0200827 */
828void truncate_setsize(struct inode *inode, loff_t newsize)
829{
Jan Kara90a80202014-10-01 21:49:18 -0400830 loff_t oldsize = inode->i_size;
831
Christoph Hellwig2c27c652010-06-04 11:30:04 +0200832 i_size_write(inode, newsize);
Jan Kara90a80202014-10-01 21:49:18 -0400833 if (newsize > oldsize)
834 pagecache_isize_extended(inode, oldsize, newsize);
Kirill A. Shutemov7caef262013-09-12 15:13:56 -0700835 truncate_pagecache(inode, newsize);
Christoph Hellwig2c27c652010-06-04 11:30:04 +0200836}
837EXPORT_SYMBOL(truncate_setsize);
838
839/**
Jan Kara90a80202014-10-01 21:49:18 -0400840 * pagecache_isize_extended - update pagecache after extension of i_size
841 * @inode: inode for which i_size was extended
842 * @from: original inode size
843 * @to: new inode size
844 *
845 * Handle extension of inode size either caused by extending truncate or by
846 * write starting after current i_size. We mark the page straddling current
847 * i_size RO so that page_mkwrite() is called on the nearest write access to
848 * the page. This way filesystem can be sure that page_mkwrite() is called on
849 * the page before user writes to the page via mmap after the i_size has been
850 * changed.
851 *
852 * The function must be called after i_size is updated so that page fault
853 * coming after we unlock the page will already see the new i_size.
854 * The function must be called while we still hold i_mutex - this not only
855 * makes sure i_size is stable but also that userspace cannot observe new
856 * i_size value before we are prepared to store mmap writes at new inode size.
857 */
858void pagecache_isize_extended(struct inode *inode, loff_t from, loff_t to)
859{
Fabian Frederick93407472017-02-27 14:28:32 -0800860 int bsize = i_blocksize(inode);
Jan Kara90a80202014-10-01 21:49:18 -0400861 loff_t rounded_from;
862 struct page *page;
863 pgoff_t index;
864
Jan Kara90a80202014-10-01 21:49:18 -0400865 WARN_ON(to > inode->i_size);
866
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300867 if (from >= to || bsize == PAGE_SIZE)
Jan Kara90a80202014-10-01 21:49:18 -0400868 return;
869 /* Page straddling @from will not have any hole block created? */
870 rounded_from = round_up(from, bsize);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300871 if (to <= rounded_from || !(rounded_from & (PAGE_SIZE - 1)))
Jan Kara90a80202014-10-01 21:49:18 -0400872 return;
873
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300874 index = from >> PAGE_SHIFT;
Jan Kara90a80202014-10-01 21:49:18 -0400875 page = find_lock_page(inode->i_mapping, index);
876 /* Page not cached? Nothing to do */
877 if (!page)
878 return;
879 /*
880 * See clear_page_dirty_for_io() for details why set_page_dirty()
881 * is needed.
882 */
883 if (page_mkclean(page))
884 set_page_dirty(page);
885 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300886 put_page(page);
Jan Kara90a80202014-10-01 21:49:18 -0400887}
888EXPORT_SYMBOL(pagecache_isize_extended);
889
890/**
Hugh Dickins623e3db2012-03-28 14:42:40 -0700891 * truncate_pagecache_range - unmap and remove pagecache that is hole-punched
892 * @inode: inode
893 * @lstart: offset of beginning of hole
894 * @lend: offset of last byte of hole
895 *
896 * This function should typically be called before the filesystem
897 * releases resources associated with the freed range (eg. deallocates
898 * blocks). This way, pagecache will always stay logically coherent
899 * with on-disk format, and the filesystem would not have to deal with
900 * situations such as writepage being called for a page that has already
901 * had its underlying blocks deallocated.
902 */
903void truncate_pagecache_range(struct inode *inode, loff_t lstart, loff_t lend)
904{
905 struct address_space *mapping = inode->i_mapping;
906 loff_t unmap_start = round_up(lstart, PAGE_SIZE);
907 loff_t unmap_end = round_down(1 + lend, PAGE_SIZE) - 1;
908 /*
909 * This rounding is currently just for example: unmap_mapping_range
910 * expands its hole outwards, whereas we want it to contract the hole
911 * inwards. However, existing callers of truncate_pagecache_range are
Lukas Czerner5a720392013-05-27 23:32:35 -0400912 * doing their own page rounding first. Note that unmap_mapping_range
913 * allows holelen 0 for all, and we allow lend -1 for end of file.
Hugh Dickins623e3db2012-03-28 14:42:40 -0700914 */
915
916 /*
917 * Unlike in truncate_pagecache, unmap_mapping_range is called only
918 * once (before truncating pagecache), and without "even_cows" flag:
919 * hole-punching should not remove private COWed pages from the hole.
920 */
921 if ((u64)unmap_end > (u64)unmap_start)
922 unmap_mapping_range(mapping, unmap_start,
923 1 + unmap_end - unmap_start, 0);
924 truncate_inode_pages_range(mapping, lstart, lend);
925}
926EXPORT_SYMBOL(truncate_pagecache_range);