ALSA: usb-audio: Generic application of implicit fb to Roland/BOSS devices
[linux-block.git] / mm / truncate.c
CommitLineData
457c8996 1// SPDX-License-Identifier: GPL-2.0-only
1da177e4
LT
2/*
3 * mm/truncate.c - code for taking down pages from address_spaces
4 *
5 * Copyright (C) 2002, Linus Torvalds
6 *
e1f8e874 7 * 10Sep2002 Andrew Morton
1da177e4
LT
8 * Initial version.
9 */
10
11#include <linux/kernel.h>
4af3c9cc 12#include <linux/backing-dev.h>
f9fe48be 13#include <linux/dax.h>
5a0e3ad6 14#include <linux/gfp.h>
1da177e4 15#include <linux/mm.h>
0fd0e6b0 16#include <linux/swap.h>
b95f1b31 17#include <linux/export.h>
1da177e4 18#include <linux/pagemap.h>
01f2705d 19#include <linux/highmem.h>
1da177e4 20#include <linux/pagevec.h>
e08748ce 21#include <linux/task_io_accounting_ops.h>
1da177e4 22#include <linux/buffer_head.h> /* grr. try_to_release_page,
aaa4059b 23 do_invalidatepage */
3a4f8a0b 24#include <linux/shmem_fs.h>
c515e1fd 25#include <linux/cleancache.h>
90a80202 26#include <linux/rmap.h>
ba470de4 27#include "internal.h"
1da177e4 28
f2187599
MG
29/*
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)
0cd6144a 36{
69b6c131 37 XA_STATE(xas, &mapping->i_pages, index);
449dd698 38
69b6c131
MW
39 xas_set_update(&xas, workingset_update_node);
40 if (xas_load(&xas) != entry)
f2187599 41 return;
69b6c131 42 xas_store(&xas, NULL);
ac401cc7 43 mapping->nrexceptional--;
f2187599
MG
44}
45
46static void clear_shadow_entry(struct address_space *mapping, pgoff_t index,
47 void *entry)
48{
b93b0163 49 xa_lock_irq(&mapping->i_pages);
f2187599 50 __clear_shadow_entry(mapping, index, entry);
b93b0163 51 xa_unlock_irq(&mapping->i_pages);
0cd6144a 52}
1da177e4 53
c6dcf52c 54/*
f2187599
MG
55 * 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.
c6dcf52c 58 */
f2187599 59static void truncate_exceptional_pvec_entries(struct address_space *mapping,
31d270fd 60 struct pagevec *pvec, pgoff_t *indices)
c6dcf52c 61{
f2187599 62 int i, j;
31d270fd 63 bool dax;
f2187599 64
c6dcf52c
JK
65 /* Handled by shmem itself */
66 if (shmem_mapping(mapping))
67 return;
68
f2187599 69 for (j = 0; j < pagevec_count(pvec); j++)
3159f943 70 if (xa_is_value(pvec->pages[j]))
f2187599
MG
71 break;
72
73 if (j == pagevec_count(pvec))
c6dcf52c 74 return;
f2187599
MG
75
76 dax = dax_mapping(mapping);
31d270fd 77 if (!dax)
b93b0163 78 xa_lock_irq(&mapping->i_pages);
f2187599
MG
79
80 for (i = j; i < pagevec_count(pvec); i++) {
81 struct page *page = pvec->pages[i];
82 pgoff_t index = indices[i];
83
3159f943 84 if (!xa_is_value(page)) {
f2187599
MG
85 pvec->pages[j++] = page;
86 continue;
87 }
88
f2187599
MG
89 if (unlikely(dax)) {
90 dax_delete_mapping_entry(mapping, index);
91 continue;
92 }
93
94 __clear_shadow_entry(mapping, index, page);
c6dcf52c 95 }
f2187599 96
31d270fd 97 if (!dax)
b93b0163 98 xa_unlock_irq(&mapping->i_pages);
f2187599 99 pvec->nr = j;
c6dcf52c
JK
100}
101
102/*
103 * Invalidate exceptional entry if easily possible. This handles exceptional
4636e70b 104 * entries for invalidate_inode_pages().
c6dcf52c
JK
105 */
106static int invalidate_exceptional_entry(struct address_space *mapping,
107 pgoff_t index, void *entry)
108{
4636e70b
RZ
109 /* Handled by shmem itself, or for DAX we do nothing. */
110 if (shmem_mapping(mapping) || dax_mapping(mapping))
c6dcf52c 111 return 1;
c6dcf52c
JK
112 clear_shadow_entry(mapping, index, entry);
113 return 1;
114}
115
116/*
117 * Invalidate exceptional entry if clean. This handles exceptional entries for
118 * invalidate_inode_pages2() so for DAX it evicts only clean entries.
119 */
120static int invalidate_exceptional_entry2(struct address_space *mapping,
121 pgoff_t index, void *entry)
122{
123 /* Handled by shmem itself */
124 if (shmem_mapping(mapping))
125 return 1;
126 if (dax_mapping(mapping))
127 return dax_invalidate_mapping_entry_sync(mapping, index);
128 clear_shadow_entry(mapping, index, entry);
129 return 1;
130}
131
cf9a2ae8 132/**
28bc44d7 133 * do_invalidatepage - invalidate part or all of a page
cf9a2ae8 134 * @page: the page which is affected
d47992f8
LC
135 * @offset: start of the range to invalidate
136 * @length: length of the range to invalidate
cf9a2ae8
DH
137 *
138 * do_invalidatepage() is called when all or part of the page has become
139 * invalidated by a truncate operation.
140 *
141 * do_invalidatepage() does not have to release all buffers, but it must
142 * ensure that no dirty buffer is left outside @offset and that no I/O
143 * is underway against any of the blocks which are outside the truncation
144 * point. Because the caller is about to free (and possibly reuse) those
145 * blocks on-disk.
146 */
d47992f8
LC
147void do_invalidatepage(struct page *page, unsigned int offset,
148 unsigned int length)
cf9a2ae8 149{
d47992f8
LC
150 void (*invalidatepage)(struct page *, unsigned int, unsigned int);
151
cf9a2ae8 152 invalidatepage = page->mapping->a_ops->invalidatepage;
9361401e 153#ifdef CONFIG_BLOCK
cf9a2ae8
DH
154 if (!invalidatepage)
155 invalidatepage = block_invalidatepage;
9361401e 156#endif
cf9a2ae8 157 if (invalidatepage)
d47992f8 158 (*invalidatepage)(page, offset, length);
cf9a2ae8
DH
159}
160
1da177e4
LT
161/*
162 * If truncate cannot remove the fs-private metadata from the page, the page
62e1c553 163 * becomes orphaned. It will be left on the LRU and may even be mapped into
54cb8821 164 * user pagetables if we're racing with filemap_fault().
1da177e4 165 *
fc3a5ac5 166 * We need to bail out if page->mapping is no longer equal to the original
1da177e4 167 * mapping. This happens a) when the VM reclaimed the page while we waited on
fc0ecff6 168 * its lock, b) when a concurrent invalidate_mapping_pages got there first and
1da177e4
LT
169 * c) when tmpfs swizzles a page between a tmpfs inode and swapper_space.
170 */
9f4e41f4
JK
171static void
172truncate_cleanup_page(struct address_space *mapping, struct page *page)
1da177e4 173{
9f4e41f4 174 if (page_mapped(page)) {
fc3a5ac5 175 unsigned int nr = thp_nr_pages(page);
977fbdcd 176 unmap_mapping_pages(mapping, page->index, nr, false);
9f4e41f4 177 }
1da177e4 178
266cf658 179 if (page_has_private(page))
fc3a5ac5 180 do_invalidatepage(page, 0, thp_size(page));
1da177e4 181
b9ea2515
KK
182 /*
183 * Some filesystems seem to re-dirty the page even after
184 * the VM has canceled the dirty bit (eg ext3 journaling).
185 * Hence dirty accounting check is placed after invalidation.
186 */
11f81bec 187 cancel_dirty_page(page);
1da177e4 188 ClearPageMappedToDisk(page);
1da177e4
LT
189}
190
191/*
fc0ecff6 192 * This is for invalidate_mapping_pages(). That function can be called at
1da177e4 193 * any time, and is not supposed to throw away dirty pages. But pages can
0fd0e6b0
NP
194 * be marked dirty at any time too, so use remove_mapping which safely
195 * discards clean, unused pages.
1da177e4
LT
196 *
197 * Returns non-zero if the page was successfully invalidated.
198 */
199static int
200invalidate_complete_page(struct address_space *mapping, struct page *page)
201{
0fd0e6b0
NP
202 int ret;
203
1da177e4
LT
204 if (page->mapping != mapping)
205 return 0;
206
266cf658 207 if (page_has_private(page) && !try_to_release_page(page, 0))
1da177e4
LT
208 return 0;
209
0fd0e6b0 210 ret = remove_mapping(mapping, page);
0fd0e6b0
NP
211
212 return ret;
1da177e4
LT
213}
214
750b4987
NP
215int truncate_inode_page(struct address_space *mapping, struct page *page)
216{
fc127da0
KS
217 VM_BUG_ON_PAGE(PageTail(page), page);
218
9f4e41f4
JK
219 if (page->mapping != mapping)
220 return -EIO;
221
222 truncate_cleanup_page(mapping, page);
223 delete_from_page_cache(page);
224 return 0;
750b4987
NP
225}
226
25718736
AK
227/*
228 * Used to get rid of pages on hardware memory corruption.
229 */
230int generic_error_remove_page(struct address_space *mapping, struct page *page)
231{
232 if (!mapping)
233 return -EINVAL;
234 /*
235 * Only punch for normal data pages for now.
236 * Handling other types like directories would need more auditing.
237 */
238 if (!S_ISREG(mapping->host->i_mode))
239 return -EIO;
240 return truncate_inode_page(mapping, page);
241}
242EXPORT_SYMBOL(generic_error_remove_page);
243
83f78668
WF
244/*
245 * Safely invalidate one page from its pagecache mapping.
246 * It only drops clean, unused pages. The page must be locked.
247 *
248 * Returns 1 if the page is successfully invalidated, otherwise 0.
249 */
250int invalidate_inode_page(struct page *page)
251{
252 struct address_space *mapping = page_mapping(page);
253 if (!mapping)
254 return 0;
255 if (PageDirty(page) || PageWriteback(page))
256 return 0;
257 if (page_mapped(page))
258 return 0;
259 return invalidate_complete_page(mapping, page);
260}
261
1da177e4 262/**
73c1e204 263 * truncate_inode_pages_range - truncate range of pages specified by start & end byte offsets
1da177e4
LT
264 * @mapping: mapping to truncate
265 * @lstart: offset from which to truncate
5a720394 266 * @lend: offset to which to truncate (inclusive)
1da177e4 267 *
d7339071 268 * Truncate the page cache, removing the pages that are between
5a720394
LC
269 * specified offsets (and zeroing out partial pages
270 * if lstart or lend + 1 is not page aligned).
1da177e4
LT
271 *
272 * Truncate takes two passes - the first pass is nonblocking. It will not
273 * block on page locks and it will not block on writeback. The second pass
274 * will wait. This is to prevent as much IO as possible in the affected region.
275 * The first pass will remove most pages, so the search cost of the second pass
276 * is low.
277 *
1da177e4
LT
278 * We pass down the cache-hot hint to the page freeing code. Even if the
279 * mapping is large, it is probably the case that the final pages are the most
280 * recently touched, and freeing happens in ascending file offset order.
5a720394
LC
281 *
282 * Note that since ->invalidatepage() accepts range to invalidate
283 * truncate_inode_pages_range is able to handle cases where lend + 1 is not
284 * page aligned properly.
1da177e4 285 */
d7339071
HR
286void truncate_inode_pages_range(struct address_space *mapping,
287 loff_t lstart, loff_t lend)
1da177e4 288{
5a720394
LC
289 pgoff_t start; /* inclusive */
290 pgoff_t end; /* exclusive */
291 unsigned int partial_start; /* inclusive */
292 unsigned int partial_end; /* exclusive */
293 struct pagevec pvec;
0cd6144a 294 pgoff_t indices[PAGEVEC_SIZE];
5a720394
LC
295 pgoff_t index;
296 int i;
1da177e4 297
f9fe48be 298 if (mapping->nrpages == 0 && mapping->nrexceptional == 0)
34ccb69e 299 goto out;
1da177e4 300
5a720394 301 /* Offsets within partial pages */
09cbfeaf
KS
302 partial_start = lstart & (PAGE_SIZE - 1);
303 partial_end = (lend + 1) & (PAGE_SIZE - 1);
5a720394
LC
304
305 /*
306 * 'start' and 'end' always covers the range of pages to be fully
307 * truncated. Partial pages are covered with 'partial_start' at the
308 * start of the range and 'partial_end' at the end of the range.
309 * Note that 'end' is exclusive while 'lend' is inclusive.
310 */
09cbfeaf 311 start = (lstart + PAGE_SIZE - 1) >> PAGE_SHIFT;
5a720394
LC
312 if (lend == -1)
313 /*
314 * lend == -1 indicates end-of-file so we have to set 'end'
315 * to the highest possible pgoff_t and since the type is
316 * unsigned we're using -1.
317 */
318 end = -1;
319 else
09cbfeaf 320 end = (lend + 1) >> PAGE_SHIFT;
d7339071 321
86679820 322 pagevec_init(&pvec);
b85e0eff 323 index = start;
5c211ba2
MWO
324 while (index < end && find_lock_entries(mapping, index, end - 1,
325 &pvec, indices)) {
326 index = indices[pagevec_count(&pvec) - 1] + 1;
31d270fd 327 truncate_exceptional_pvec_entries(mapping, &pvec, indices);
5c211ba2
MWO
328 for (i = 0; i < pagevec_count(&pvec); i++)
329 truncate_cleanup_page(mapping, pvec.pages[i]);
330 delete_from_page_cache_batch(mapping, &pvec);
331 for (i = 0; i < pagevec_count(&pvec); i++)
332 unlock_page(pvec.pages[i]);
1da177e4
LT
333 pagevec_release(&pvec);
334 cond_resched();
335 }
5c211ba2 336
5a720394 337 if (partial_start) {
1da177e4
LT
338 struct page *page = find_lock_page(mapping, start - 1);
339 if (page) {
09cbfeaf 340 unsigned int top = PAGE_SIZE;
5a720394
LC
341 if (start > end) {
342 /* Truncation within a single page */
343 top = partial_end;
344 partial_end = 0;
345 }
1da177e4 346 wait_on_page_writeback(page);
5a720394
LC
347 zero_user_segment(page, partial_start, top);
348 cleancache_invalidate_page(mapping, page);
349 if (page_has_private(page))
350 do_invalidatepage(page, partial_start,
351 top - partial_start);
1da177e4 352 unlock_page(page);
09cbfeaf 353 put_page(page);
1da177e4
LT
354 }
355 }
5a720394
LC
356 if (partial_end) {
357 struct page *page = find_lock_page(mapping, end);
358 if (page) {
359 wait_on_page_writeback(page);
360 zero_user_segment(page, 0, partial_end);
361 cleancache_invalidate_page(mapping, page);
362 if (page_has_private(page))
363 do_invalidatepage(page, 0,
364 partial_end);
365 unlock_page(page);
09cbfeaf 366 put_page(page);
5a720394
LC
367 }
368 }
369 /*
370 * If the truncation happened within a single page no pages
371 * will be released, just zeroed, so we can bail out now.
372 */
373 if (start >= end)
34ccb69e 374 goto out;
1da177e4 375
b85e0eff 376 index = start;
1da177e4
LT
377 for ( ; ; ) {
378 cond_resched();
a656a202 379 if (!find_get_entries(mapping, index, end - 1, &pvec,
38cefeb3 380 indices)) {
792ceaef 381 /* If all gone from start onwards, we're done */
b85e0eff 382 if (index == start)
1da177e4 383 break;
792ceaef 384 /* Otherwise restart to make sure all gone */
b85e0eff 385 index = start;
1da177e4
LT
386 continue;
387 }
f2187599 388
1da177e4
LT
389 for (i = 0; i < pagevec_count(&pvec); i++) {
390 struct page *page = pvec.pages[i];
391
b85e0eff 392 /* We rely upon deletion not changing page->index */
0cd6144a 393 index = indices[i];
b85e0eff 394
3159f943 395 if (xa_is_value(page))
0cd6144a 396 continue;
0cd6144a 397
1da177e4 398 lock_page(page);
5cbc198a 399 WARN_ON(page_to_index(page) != index);
1da177e4 400 wait_on_page_writeback(page);
750b4987 401 truncate_inode_page(mapping, page);
1da177e4
LT
402 unlock_page(page);
403 }
31d270fd 404 truncate_exceptional_pvec_entries(mapping, &pvec, indices);
1da177e4 405 pagevec_release(&pvec);
b85e0eff 406 index++;
1da177e4 407 }
34ccb69e
AR
408
409out:
3167760f 410 cleancache_invalidate_inode(mapping);
1da177e4 411}
d7339071 412EXPORT_SYMBOL(truncate_inode_pages_range);
1da177e4 413
d7339071
HR
414/**
415 * truncate_inode_pages - truncate *all* the pages from an offset
416 * @mapping: mapping to truncate
417 * @lstart: offset from which to truncate
418 *
1b1dcc1b 419 * Called under (and serialised by) inode->i_mutex.
08142579
JK
420 *
421 * Note: When this function returns, there can be a page in the process of
422 * deletion (inside __delete_from_page_cache()) in the specified range. Thus
423 * mapping->nrpages can be non-zero when this function returns even after
424 * truncation of the whole mapping.
d7339071
HR
425 */
426void truncate_inode_pages(struct address_space *mapping, loff_t lstart)
427{
428 truncate_inode_pages_range(mapping, lstart, (loff_t)-1);
429}
1da177e4
LT
430EXPORT_SYMBOL(truncate_inode_pages);
431
91b0abe3
JW
432/**
433 * truncate_inode_pages_final - truncate *all* pages before inode dies
434 * @mapping: mapping to truncate
435 *
436 * Called under (and serialized by) inode->i_mutex.
437 *
438 * Filesystems have to use this in the .evict_inode path to inform the
439 * VM that this is the final truncate and the inode is going away.
440 */
441void truncate_inode_pages_final(struct address_space *mapping)
442{
f9fe48be 443 unsigned long nrexceptional;
91b0abe3
JW
444 unsigned long nrpages;
445
446 /*
447 * Page reclaim can not participate in regular inode lifetime
448 * management (can't call iput()) and thus can race with the
449 * inode teardown. Tell it when the address space is exiting,
450 * so that it does not install eviction information after the
451 * final truncate has begun.
452 */
453 mapping_set_exiting(mapping);
454
455 /*
456 * When reclaim installs eviction entries, it increases
f9fe48be 457 * nrexceptional first, then decreases nrpages. Make sure we see
91b0abe3
JW
458 * this in the right order or we might miss an entry.
459 */
460 nrpages = mapping->nrpages;
461 smp_rmb();
f9fe48be 462 nrexceptional = mapping->nrexceptional;
91b0abe3 463
f9fe48be 464 if (nrpages || nrexceptional) {
91b0abe3
JW
465 /*
466 * As truncation uses a lockless tree lookup, cycle
467 * the tree lock to make sure any ongoing tree
468 * modification that does not see AS_EXITING is
469 * completed before starting the final truncate.
470 */
b93b0163
MW
471 xa_lock_irq(&mapping->i_pages);
472 xa_unlock_irq(&mapping->i_pages);
91b0abe3 473 }
6ff38bd4
PT
474
475 /*
476 * Cleancache needs notification even if there are no pages or shadow
477 * entries.
478 */
479 truncate_inode_pages(mapping, 0);
91b0abe3
JW
480}
481EXPORT_SYMBOL(truncate_inode_pages_final);
482
a77eedbc 483static unsigned long __invalidate_mapping_pages(struct address_space *mapping,
eb1d7a65 484 pgoff_t start, pgoff_t end, unsigned long *nr_pagevec)
1da177e4 485{
0cd6144a 486 pgoff_t indices[PAGEVEC_SIZE];
1da177e4 487 struct pagevec pvec;
b85e0eff 488 pgoff_t index = start;
31560180
MK
489 unsigned long ret;
490 unsigned long count = 0;
1da177e4
LT
491 int i;
492
86679820 493 pagevec_init(&pvec);
5c211ba2 494 while (find_lock_entries(mapping, index, end, &pvec, indices)) {
1da177e4
LT
495 for (i = 0; i < pagevec_count(&pvec); i++) {
496 struct page *page = pvec.pages[i];
e0f23603 497
b85e0eff 498 /* We rely upon deletion not changing page->index */
0cd6144a 499 index = indices[i];
e0f23603 500
3159f943 501 if (xa_is_value(page)) {
c6dcf52c
JK
502 invalidate_exceptional_entry(mapping, index,
503 page);
0cd6144a
JW
504 continue;
505 }
5c211ba2 506 index += thp_nr_pages(page) - 1;
fc127da0 507
31560180 508 ret = invalidate_inode_page(page);
1da177e4 509 unlock_page(page);
31560180
MK
510 /*
511 * Invalidation is a hint that the page is no longer
512 * of interest and try to speed up its reclaim.
513 */
eb1d7a65 514 if (!ret) {
cc5993bd 515 deactivate_file_page(page);
eb1d7a65
YS
516 /* It is likely on the pagevec of a remote CPU */
517 if (nr_pagevec)
518 (*nr_pagevec)++;
519 }
31560180 520 count += ret;
1da177e4 521 }
0cd6144a 522 pagevec_remove_exceptionals(&pvec);
1da177e4 523 pagevec_release(&pvec);
28697355 524 cond_resched();
b85e0eff 525 index++;
1da177e4 526 }
31560180 527 return count;
1da177e4 528}
eb1d7a65
YS
529
530/**
531 * invalidate_mapping_pages - Invalidate all the unlocked pages of one inode
532 * @mapping: the address_space which holds the pages to invalidate
533 * @start: the offset 'from' which to invalidate
534 * @end: the offset 'to' which to invalidate (inclusive)
535 *
536 * This function only removes the unlocked pages, if you want to
537 * remove all the pages of one inode, you must call truncate_inode_pages.
538 *
539 * invalidate_mapping_pages() will not block on IO activity. It will not
540 * invalidate pages which are dirty, locked, under writeback or mapped into
541 * pagetables.
542 *
543 * Return: the number of the pages that were invalidated
544 */
545unsigned long invalidate_mapping_pages(struct address_space *mapping,
546 pgoff_t start, pgoff_t end)
547{
548 return __invalidate_mapping_pages(mapping, start, end, NULL);
549}
54bc4855 550EXPORT_SYMBOL(invalidate_mapping_pages);
1da177e4 551
eb1d7a65 552/**
649c6dfe
AS
553 * invalidate_mapping_pagevec - Invalidate all the unlocked pages of one inode
554 * @mapping: the address_space which holds the pages to invalidate
555 * @start: the offset 'from' which to invalidate
556 * @end: the offset 'to' which to invalidate (inclusive)
557 * @nr_pagevec: invalidate failed page number for caller
558 *
a00cda3f
MCC
559 * This helper is similar to invalidate_mapping_pages(), except that it accounts
560 * for pages that are likely on a pagevec and counts them in @nr_pagevec, which
561 * will be used by the caller.
eb1d7a65
YS
562 */
563void invalidate_mapping_pagevec(struct address_space *mapping,
564 pgoff_t start, pgoff_t end, unsigned long *nr_pagevec)
565{
566 __invalidate_mapping_pages(mapping, start, end, nr_pagevec);
567}
568
bd4c8ce4
AM
569/*
570 * This is like invalidate_complete_page(), except it ignores the page's
571 * refcount. We do this because invalidate_inode_pages2() needs stronger
572 * invalidation guarantees, and cannot afford to leave pages behind because
2706a1b8
AB
573 * shrink_page_list() has a temp ref on them, or because they're transiently
574 * sitting in the lru_cache_add() pagevecs.
bd4c8ce4
AM
575 */
576static int
577invalidate_complete_page2(struct address_space *mapping, struct page *page)
578{
c4843a75
GT
579 unsigned long flags;
580
bd4c8ce4
AM
581 if (page->mapping != mapping)
582 return 0;
583
266cf658 584 if (page_has_private(page) && !try_to_release_page(page, GFP_KERNEL))
bd4c8ce4
AM
585 return 0;
586
b93b0163 587 xa_lock_irqsave(&mapping->i_pages, flags);
bd4c8ce4
AM
588 if (PageDirty(page))
589 goto failed;
590
266cf658 591 BUG_ON(page_has_private(page));
62cccb8c 592 __delete_from_page_cache(page, NULL);
b93b0163 593 xa_unlock_irqrestore(&mapping->i_pages, flags);
6072d13c
LT
594
595 if (mapping->a_ops->freepage)
596 mapping->a_ops->freepage(page);
597
09cbfeaf 598 put_page(page); /* pagecache ref */
bd4c8ce4
AM
599 return 1;
600failed:
b93b0163 601 xa_unlock_irqrestore(&mapping->i_pages, flags);
bd4c8ce4
AM
602 return 0;
603}
604
e3db7691
TM
605static int do_launder_page(struct address_space *mapping, struct page *page)
606{
607 if (!PageDirty(page))
608 return 0;
609 if (page->mapping != mapping || mapping->a_ops->launder_page == NULL)
610 return 0;
611 return mapping->a_ops->launder_page(page);
612}
613
1da177e4
LT
614/**
615 * invalidate_inode_pages2_range - remove range of pages from an address_space
67be2dd1 616 * @mapping: the address_space
1da177e4
LT
617 * @start: the page offset 'from' which to invalidate
618 * @end: the page offset 'to' which to invalidate (inclusive)
619 *
620 * Any pages which are found to be mapped into pagetables are unmapped prior to
621 * invalidation.
622 *
a862f68a 623 * Return: -EBUSY if any pages could not be invalidated.
1da177e4
LT
624 */
625int invalidate_inode_pages2_range(struct address_space *mapping,
626 pgoff_t start, pgoff_t end)
627{
0cd6144a 628 pgoff_t indices[PAGEVEC_SIZE];
1da177e4 629 struct pagevec pvec;
b85e0eff 630 pgoff_t index;
1da177e4
LT
631 int i;
632 int ret = 0;
0dd1334f 633 int ret2 = 0;
1da177e4 634 int did_range_unmap = 0;
1da177e4 635
32691f0f 636 if (mapping->nrpages == 0 && mapping->nrexceptional == 0)
34ccb69e 637 goto out;
32691f0f 638
86679820 639 pagevec_init(&pvec);
b85e0eff 640 index = start;
a656a202 641 while (find_get_entries(mapping, index, end, &pvec, indices)) {
7b965e08 642 for (i = 0; i < pagevec_count(&pvec); i++) {
1da177e4 643 struct page *page = pvec.pages[i];
b85e0eff
HD
644
645 /* We rely upon deletion not changing page->index */
0cd6144a 646 index = indices[i];
1da177e4 647
3159f943 648 if (xa_is_value(page)) {
c6dcf52c
JK
649 if (!invalidate_exceptional_entry2(mapping,
650 index, page))
651 ret = -EBUSY;
0cd6144a
JW
652 continue;
653 }
654
1da177e4 655 lock_page(page);
5cbc198a 656 WARN_ON(page_to_index(page) != index);
1da177e4
LT
657 if (page->mapping != mapping) {
658 unlock_page(page);
659 continue;
660 }
1da177e4 661 wait_on_page_writeback(page);
d00806b1 662 if (page_mapped(page)) {
1da177e4
LT
663 if (!did_range_unmap) {
664 /*
665 * Zap the rest of the file in one hit.
666 */
977fbdcd
MW
667 unmap_mapping_pages(mapping, index,
668 (1 + end - index), false);
1da177e4
LT
669 did_range_unmap = 1;
670 } else {
671 /*
672 * Just zap this page
673 */
977fbdcd
MW
674 unmap_mapping_pages(mapping, index,
675 1, false);
1da177e4
LT
676 }
677 }
d00806b1 678 BUG_ON(page_mapped(page));
0dd1334f
HH
679 ret2 = do_launder_page(mapping, page);
680 if (ret2 == 0) {
681 if (!invalidate_complete_page2(mapping, page))
6ccfa806 682 ret2 = -EBUSY;
0dd1334f
HH
683 }
684 if (ret2 < 0)
685 ret = ret2;
1da177e4
LT
686 unlock_page(page);
687 }
0cd6144a 688 pagevec_remove_exceptionals(&pvec);
1da177e4
LT
689 pagevec_release(&pvec);
690 cond_resched();
b85e0eff 691 index++;
1da177e4 692 }
cd656375 693 /*
69b6c131 694 * For DAX we invalidate page tables after invalidating page cache. We
cd656375
JK
695 * could invalidate page tables while invalidating each entry however
696 * that would be expensive. And doing range unmapping before doesn't
69b6c131 697 * work as we have no cheap way to find whether page cache entry didn't
cd656375
JK
698 * get remapped later.
699 */
700 if (dax_mapping(mapping)) {
977fbdcd 701 unmap_mapping_pages(mapping, start, end - start + 1, false);
cd656375 702 }
34ccb69e 703out:
3167760f 704 cleancache_invalidate_inode(mapping);
1da177e4
LT
705 return ret;
706}
707EXPORT_SYMBOL_GPL(invalidate_inode_pages2_range);
708
709/**
710 * invalidate_inode_pages2 - remove all pages from an address_space
67be2dd1 711 * @mapping: the address_space
1da177e4
LT
712 *
713 * Any pages which are found to be mapped into pagetables are unmapped prior to
714 * invalidation.
715 *
a862f68a 716 * Return: -EBUSY if any pages could not be invalidated.
1da177e4
LT
717 */
718int invalidate_inode_pages2(struct address_space *mapping)
719{
720 return invalidate_inode_pages2_range(mapping, 0, -1);
721}
722EXPORT_SYMBOL_GPL(invalidate_inode_pages2);
25d9e2d1 723
724/**
725 * truncate_pagecache - unmap and remove pagecache that has been truncated
726 * @inode: inode
8a549bea 727 * @newsize: new file size
25d9e2d1 728 *
729 * inode's new i_size must already be written before truncate_pagecache
730 * is called.
731 *
732 * This function should typically be called before the filesystem
733 * releases resources associated with the freed range (eg. deallocates
734 * blocks). This way, pagecache will always stay logically coherent
735 * with on-disk format, and the filesystem would not have to deal with
736 * situations such as writepage being called for a page that has already
737 * had its underlying blocks deallocated.
738 */
7caef267 739void truncate_pagecache(struct inode *inode, loff_t newsize)
25d9e2d1 740{
cedabed4 741 struct address_space *mapping = inode->i_mapping;
8a549bea 742 loff_t holebegin = round_up(newsize, PAGE_SIZE);
cedabed4
OH
743
744 /*
745 * unmap_mapping_range is called twice, first simply for
746 * efficiency so that truncate_inode_pages does fewer
747 * single-page unmaps. However after this first call, and
748 * before truncate_inode_pages finishes, it is possible for
749 * private pages to be COWed, which remain after
750 * truncate_inode_pages finishes, hence the second
751 * unmap_mapping_range call must be made for correctness.
752 */
8a549bea
HD
753 unmap_mapping_range(mapping, holebegin, 0, 1);
754 truncate_inode_pages(mapping, newsize);
755 unmap_mapping_range(mapping, holebegin, 0, 1);
25d9e2d1 756}
757EXPORT_SYMBOL(truncate_pagecache);
758
2c27c65e
CH
759/**
760 * truncate_setsize - update inode and pagecache for a new file size
761 * @inode: inode
762 * @newsize: new file size
763 *
382e27da
JK
764 * truncate_setsize updates i_size and performs pagecache truncation (if
765 * necessary) to @newsize. It will be typically be called from the filesystem's
766 * setattr function when ATTR_SIZE is passed in.
2c27c65e 767 *
77783d06
JK
768 * Must be called with a lock serializing truncates and writes (generally
769 * i_mutex but e.g. xfs uses a different lock) and before all filesystem
770 * specific block truncation has been performed.
2c27c65e
CH
771 */
772void truncate_setsize(struct inode *inode, loff_t newsize)
773{
90a80202
JK
774 loff_t oldsize = inode->i_size;
775
2c27c65e 776 i_size_write(inode, newsize);
90a80202
JK
777 if (newsize > oldsize)
778 pagecache_isize_extended(inode, oldsize, newsize);
7caef267 779 truncate_pagecache(inode, newsize);
2c27c65e
CH
780}
781EXPORT_SYMBOL(truncate_setsize);
782
90a80202
JK
783/**
784 * pagecache_isize_extended - update pagecache after extension of i_size
785 * @inode: inode for which i_size was extended
786 * @from: original inode size
787 * @to: new inode size
788 *
789 * Handle extension of inode size either caused by extending truncate or by
790 * write starting after current i_size. We mark the page straddling current
791 * i_size RO so that page_mkwrite() is called on the nearest write access to
792 * the page. This way filesystem can be sure that page_mkwrite() is called on
793 * the page before user writes to the page via mmap after the i_size has been
794 * changed.
795 *
796 * The function must be called after i_size is updated so that page fault
797 * coming after we unlock the page will already see the new i_size.
798 * The function must be called while we still hold i_mutex - this not only
799 * makes sure i_size is stable but also that userspace cannot observe new
800 * i_size value before we are prepared to store mmap writes at new inode size.
801 */
802void pagecache_isize_extended(struct inode *inode, loff_t from, loff_t to)
803{
93407472 804 int bsize = i_blocksize(inode);
90a80202
JK
805 loff_t rounded_from;
806 struct page *page;
807 pgoff_t index;
808
90a80202
JK
809 WARN_ON(to > inode->i_size);
810
09cbfeaf 811 if (from >= to || bsize == PAGE_SIZE)
90a80202
JK
812 return;
813 /* Page straddling @from will not have any hole block created? */
814 rounded_from = round_up(from, bsize);
09cbfeaf 815 if (to <= rounded_from || !(rounded_from & (PAGE_SIZE - 1)))
90a80202
JK
816 return;
817
09cbfeaf 818 index = from >> PAGE_SHIFT;
90a80202
JK
819 page = find_lock_page(inode->i_mapping, index);
820 /* Page not cached? Nothing to do */
821 if (!page)
822 return;
823 /*
824 * See clear_page_dirty_for_io() for details why set_page_dirty()
825 * is needed.
826 */
827 if (page_mkclean(page))
828 set_page_dirty(page);
829 unlock_page(page);
09cbfeaf 830 put_page(page);
90a80202
JK
831}
832EXPORT_SYMBOL(pagecache_isize_extended);
833
623e3db9
HD
834/**
835 * truncate_pagecache_range - unmap and remove pagecache that is hole-punched
836 * @inode: inode
837 * @lstart: offset of beginning of hole
838 * @lend: offset of last byte of hole
839 *
840 * This function should typically be called before the filesystem
841 * releases resources associated with the freed range (eg. deallocates
842 * blocks). This way, pagecache will always stay logically coherent
843 * with on-disk format, and the filesystem would not have to deal with
844 * situations such as writepage being called for a page that has already
845 * had its underlying blocks deallocated.
846 */
847void truncate_pagecache_range(struct inode *inode, loff_t lstart, loff_t lend)
848{
849 struct address_space *mapping = inode->i_mapping;
850 loff_t unmap_start = round_up(lstart, PAGE_SIZE);
851 loff_t unmap_end = round_down(1 + lend, PAGE_SIZE) - 1;
852 /*
853 * This rounding is currently just for example: unmap_mapping_range
854 * expands its hole outwards, whereas we want it to contract the hole
855 * inwards. However, existing callers of truncate_pagecache_range are
5a720394
LC
856 * doing their own page rounding first. Note that unmap_mapping_range
857 * allows holelen 0 for all, and we allow lend -1 for end of file.
623e3db9
HD
858 */
859
860 /*
861 * Unlike in truncate_pagecache, unmap_mapping_range is called only
862 * once (before truncating pagecache), and without "even_cows" flag:
863 * hole-punching should not remove private COWed pages from the hole.
864 */
865 if ((u64)unmap_end > (u64)unmap_start)
866 unmap_mapping_range(mapping, unmap_start,
867 1 + unmap_end - unmap_start, 0);
868 truncate_inode_pages_range(mapping, lstart, lend);
869}
870EXPORT_SYMBOL(truncate_pagecache_range);