1 // SPDX-License-Identifier: GPL-2.0-only
5 * Copyright (C) 1994-1999 Linus Torvalds
9 * This file handles the generic file mmap semantics used by
10 * most "normal" filesystems (but you don't /have/ to use this:
11 * the NFS filesystem used to do this differently, for example)
13 #include <linux/export.h>
14 #include <linux/compiler.h>
15 #include <linux/dax.h>
17 #include <linux/sched/signal.h>
18 #include <linux/uaccess.h>
19 #include <linux/capability.h>
20 #include <linux/kernel_stat.h>
21 #include <linux/gfp.h>
23 #include <linux/swap.h>
24 #include <linux/mman.h>
25 #include <linux/pagemap.h>
26 #include <linux/file.h>
27 #include <linux/uio.h>
28 #include <linux/error-injection.h>
29 #include <linux/hash.h>
30 #include <linux/writeback.h>
31 #include <linux/backing-dev.h>
32 #include <linux/pagevec.h>
33 #include <linux/blkdev.h>
34 #include <linux/security.h>
35 #include <linux/cpuset.h>
36 #include <linux/hugetlb.h>
37 #include <linux/memcontrol.h>
38 #include <linux/cleancache.h>
39 #include <linux/shmem_fs.h>
40 #include <linux/rmap.h>
41 #include <linux/delayacct.h>
42 #include <linux/psi.h>
43 #include <linux/ramfs.h>
46 #define CREATE_TRACE_POINTS
47 #include <trace/events/filemap.h>
50 * FIXME: remove all knowledge of the buffer layer from the core VM
52 #include <linux/buffer_head.h> /* for try_to_free_buffers */
57 * Shared mappings implemented 30.11.1994. It's not fully working yet,
60 * Shared mappings now work. 15.8.1995 Bruno.
62 * finished 'unifying' the page and buffer cache and SMP-threaded the
63 * page-cache, 21.05.1999, Ingo Molnar <mingo@redhat.com>
65 * SMP-threaded pagemap-LRU 1999, Andrea Arcangeli <andrea@suse.de>
71 * ->i_mmap_rwsem (truncate_pagecache)
72 * ->private_lock (__free_pte->__set_page_dirty_buffers)
73 * ->swap_lock (exclusive_swap_page, others)
77 * ->i_mmap_rwsem (truncate->unmap_mapping_range)
81 * ->page_table_lock or pte_lock (various, mainly in memory.c)
82 * ->i_pages lock (arch-dependent flush_dcache_mmap_lock)
85 * ->lock_page (access_process_vm)
87 * ->i_mutex (generic_perform_write)
88 * ->mmap_lock (fault_in_pages_readable->do_page_fault)
91 * sb_lock (fs/fs-writeback.c)
92 * ->i_pages lock (__sync_single_inode)
95 * ->anon_vma.lock (vma_adjust)
98 * ->page_table_lock or pte_lock (anon_vma_prepare and various)
100 * ->page_table_lock or pte_lock
101 * ->swap_lock (try_to_unmap_one)
102 * ->private_lock (try_to_unmap_one)
103 * ->i_pages lock (try_to_unmap_one)
104 * ->pgdat->lru_lock (follow_page->mark_page_accessed)
105 * ->pgdat->lru_lock (check_pte_range->isolate_lru_page)
106 * ->private_lock (page_remove_rmap->set_page_dirty)
107 * ->i_pages lock (page_remove_rmap->set_page_dirty)
108 * bdi.wb->list_lock (page_remove_rmap->set_page_dirty)
109 * ->inode->i_lock (page_remove_rmap->set_page_dirty)
110 * ->memcg->move_lock (page_remove_rmap->lock_page_memcg)
111 * bdi.wb->list_lock (zap_pte_range->set_page_dirty)
112 * ->inode->i_lock (zap_pte_range->set_page_dirty)
113 * ->private_lock (zap_pte_range->__set_page_dirty_buffers)
116 * ->tasklist_lock (memory_failure, collect_procs_ao)
119 static void page_cache_delete(struct address_space
*mapping
,
120 struct page
*page
, void *shadow
)
122 XA_STATE(xas
, &mapping
->i_pages
, page
->index
);
125 mapping_set_update(&xas
, mapping
);
127 /* hugetlb pages are represented by a single entry in the xarray */
128 if (!PageHuge(page
)) {
129 xas_set_order(&xas
, page
->index
, compound_order(page
));
130 nr
= compound_nr(page
);
133 VM_BUG_ON_PAGE(!PageLocked(page
), page
);
134 VM_BUG_ON_PAGE(PageTail(page
), page
);
135 VM_BUG_ON_PAGE(nr
!= 1 && shadow
, page
);
137 xas_store(&xas
, shadow
);
138 xas_init_marks(&xas
);
140 page
->mapping
= NULL
;
141 /* Leave page->index set: truncation lookup relies upon it */
144 mapping
->nrexceptional
+= nr
;
146 * Make sure the nrexceptional update is committed before
147 * the nrpages update so that final truncate racing
148 * with reclaim does not see both counters 0 at the
149 * same time and miss a shadow entry.
153 mapping
->nrpages
-= nr
;
156 static void unaccount_page_cache_page(struct address_space
*mapping
,
162 * if we're uptodate, flush out into the cleancache, otherwise
163 * invalidate any existing cleancache entries. We can't leave
164 * stale data around in the cleancache once our page is gone
166 if (PageUptodate(page
) && PageMappedToDisk(page
))
167 cleancache_put_page(page
);
169 cleancache_invalidate_page(mapping
, page
);
171 VM_BUG_ON_PAGE(PageTail(page
), page
);
172 VM_BUG_ON_PAGE(page_mapped(page
), page
);
173 if (!IS_ENABLED(CONFIG_DEBUG_VM
) && unlikely(page_mapped(page
))) {
176 pr_alert("BUG: Bad page cache in process %s pfn:%05lx\n",
177 current
->comm
, page_to_pfn(page
));
178 dump_page(page
, "still mapped when deleted");
180 add_taint(TAINT_BAD_PAGE
, LOCKDEP_NOW_UNRELIABLE
);
182 mapcount
= page_mapcount(page
);
183 if (mapping_exiting(mapping
) &&
184 page_count(page
) >= mapcount
+ 2) {
186 * All vmas have already been torn down, so it's
187 * a good bet that actually the page is unmapped,
188 * and we'd prefer not to leak it: if we're wrong,
189 * some other bad page check should catch it later.
191 page_mapcount_reset(page
);
192 page_ref_sub(page
, mapcount
);
196 /* hugetlb pages do not participate in page cache accounting. */
200 nr
= hpage_nr_pages(page
);
202 __mod_lruvec_page_state(page
, NR_FILE_PAGES
, -nr
);
203 if (PageSwapBacked(page
)) {
204 __mod_lruvec_page_state(page
, NR_SHMEM
, -nr
);
205 if (PageTransHuge(page
))
206 __dec_node_page_state(page
, NR_SHMEM_THPS
);
207 } else if (PageTransHuge(page
)) {
208 __dec_node_page_state(page
, NR_FILE_THPS
);
209 filemap_nr_thps_dec(mapping
);
213 * At this point page must be either written or cleaned by
214 * truncate. Dirty page here signals a bug and loss of
217 * This fixes dirty accounting after removing the page entirely
218 * but leaves PageDirty set: it has no effect for truncated
219 * page and anyway will be cleared before returning page into
222 if (WARN_ON_ONCE(PageDirty(page
)))
223 account_page_cleaned(page
, mapping
, inode_to_wb(mapping
->host
));
227 * Delete a page from the page cache and free it. Caller has to make
228 * sure the page is locked and that nobody else uses it - or that usage
229 * is safe. The caller must hold the i_pages lock.
231 void __delete_from_page_cache(struct page
*page
, void *shadow
)
233 struct address_space
*mapping
= page
->mapping
;
235 trace_mm_filemap_delete_from_page_cache(page
);
237 unaccount_page_cache_page(mapping
, page
);
238 page_cache_delete(mapping
, page
, shadow
);
241 static void page_cache_free_page(struct address_space
*mapping
,
244 void (*freepage
)(struct page
*);
246 freepage
= mapping
->a_ops
->freepage
;
250 if (PageTransHuge(page
) && !PageHuge(page
)) {
251 page_ref_sub(page
, HPAGE_PMD_NR
);
252 VM_BUG_ON_PAGE(page_count(page
) <= 0, page
);
259 * delete_from_page_cache - delete page from page cache
260 * @page: the page which the kernel is trying to remove from page cache
262 * This must be called only on pages that have been verified to be in the page
263 * cache and locked. It will never put the page into the free list, the caller
264 * has a reference on the page.
266 void delete_from_page_cache(struct page
*page
)
268 struct address_space
*mapping
= page_mapping(page
);
271 BUG_ON(!PageLocked(page
));
272 xa_lock_irqsave(&mapping
->i_pages
, flags
);
273 __delete_from_page_cache(page
, NULL
);
274 xa_unlock_irqrestore(&mapping
->i_pages
, flags
);
276 page_cache_free_page(mapping
, page
);
278 EXPORT_SYMBOL(delete_from_page_cache
);
281 * page_cache_delete_batch - delete several pages from page cache
282 * @mapping: the mapping to which pages belong
283 * @pvec: pagevec with pages to delete
285 * The function walks over mapping->i_pages and removes pages passed in @pvec
286 * from the mapping. The function expects @pvec to be sorted by page index
287 * and is optimised for it to be dense.
288 * It tolerates holes in @pvec (mapping entries at those indices are not
289 * modified). The function expects only THP head pages to be present in the
292 * The function expects the i_pages lock to be held.
294 static void page_cache_delete_batch(struct address_space
*mapping
,
295 struct pagevec
*pvec
)
297 XA_STATE(xas
, &mapping
->i_pages
, pvec
->pages
[0]->index
);
302 mapping_set_update(&xas
, mapping
);
303 xas_for_each(&xas
, page
, ULONG_MAX
) {
304 if (i
>= pagevec_count(pvec
))
307 /* A swap/dax/shadow entry got inserted? Skip it. */
308 if (xa_is_value(page
))
311 * A page got inserted in our range? Skip it. We have our
312 * pages locked so they are protected from being removed.
313 * If we see a page whose index is higher than ours, it
314 * means our page has been removed, which shouldn't be
315 * possible because we're holding the PageLock.
317 if (page
!= pvec
->pages
[i
]) {
318 VM_BUG_ON_PAGE(page
->index
> pvec
->pages
[i
]->index
,
323 WARN_ON_ONCE(!PageLocked(page
));
325 if (page
->index
== xas
.xa_index
)
326 page
->mapping
= NULL
;
327 /* Leave page->index set: truncation lookup relies on it */
330 * Move to the next page in the vector if this is a regular
331 * page or the index is of the last sub-page of this compound
334 if (page
->index
+ compound_nr(page
) - 1 == xas
.xa_index
)
336 xas_store(&xas
, NULL
);
339 mapping
->nrpages
-= total_pages
;
342 void delete_from_page_cache_batch(struct address_space
*mapping
,
343 struct pagevec
*pvec
)
348 if (!pagevec_count(pvec
))
351 xa_lock_irqsave(&mapping
->i_pages
, flags
);
352 for (i
= 0; i
< pagevec_count(pvec
); i
++) {
353 trace_mm_filemap_delete_from_page_cache(pvec
->pages
[i
]);
355 unaccount_page_cache_page(mapping
, pvec
->pages
[i
]);
357 page_cache_delete_batch(mapping
, pvec
);
358 xa_unlock_irqrestore(&mapping
->i_pages
, flags
);
360 for (i
= 0; i
< pagevec_count(pvec
); i
++)
361 page_cache_free_page(mapping
, pvec
->pages
[i
]);
364 int filemap_check_errors(struct address_space
*mapping
)
367 /* Check for outstanding write errors */
368 if (test_bit(AS_ENOSPC
, &mapping
->flags
) &&
369 test_and_clear_bit(AS_ENOSPC
, &mapping
->flags
))
371 if (test_bit(AS_EIO
, &mapping
->flags
) &&
372 test_and_clear_bit(AS_EIO
, &mapping
->flags
))
376 EXPORT_SYMBOL(filemap_check_errors
);
378 static int filemap_check_and_keep_errors(struct address_space
*mapping
)
380 /* Check for outstanding write errors */
381 if (test_bit(AS_EIO
, &mapping
->flags
))
383 if (test_bit(AS_ENOSPC
, &mapping
->flags
))
389 * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range
390 * @mapping: address space structure to write
391 * @start: offset in bytes where the range starts
392 * @end: offset in bytes where the range ends (inclusive)
393 * @sync_mode: enable synchronous operation
395 * Start writeback against all of a mapping's dirty pages that lie
396 * within the byte offsets <start, end> inclusive.
398 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
399 * opposed to a regular memory cleansing writeback. The difference between
400 * these two operations is that if a dirty page/buffer is encountered, it must
401 * be waited upon, and not just skipped over.
403 * Return: %0 on success, negative error code otherwise.
405 int __filemap_fdatawrite_range(struct address_space
*mapping
, loff_t start
,
406 loff_t end
, int sync_mode
)
409 struct writeback_control wbc
= {
410 .sync_mode
= sync_mode
,
411 .nr_to_write
= LONG_MAX
,
412 .range_start
= start
,
416 if (!mapping_cap_writeback_dirty(mapping
) ||
417 !mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
))
420 wbc_attach_fdatawrite_inode(&wbc
, mapping
->host
);
421 ret
= do_writepages(mapping
, &wbc
);
422 wbc_detach_inode(&wbc
);
426 static inline int __filemap_fdatawrite(struct address_space
*mapping
,
429 return __filemap_fdatawrite_range(mapping
, 0, LLONG_MAX
, sync_mode
);
432 int filemap_fdatawrite(struct address_space
*mapping
)
434 return __filemap_fdatawrite(mapping
, WB_SYNC_ALL
);
436 EXPORT_SYMBOL(filemap_fdatawrite
);
438 int filemap_fdatawrite_range(struct address_space
*mapping
, loff_t start
,
441 return __filemap_fdatawrite_range(mapping
, start
, end
, WB_SYNC_ALL
);
443 EXPORT_SYMBOL(filemap_fdatawrite_range
);
446 * filemap_flush - mostly a non-blocking flush
447 * @mapping: target address_space
449 * This is a mostly non-blocking flush. Not suitable for data-integrity
450 * purposes - I/O may not be started against all dirty pages.
452 * Return: %0 on success, negative error code otherwise.
454 int filemap_flush(struct address_space
*mapping
)
456 return __filemap_fdatawrite(mapping
, WB_SYNC_NONE
);
458 EXPORT_SYMBOL(filemap_flush
);
461 * filemap_range_has_page - check if a page exists in range.
462 * @mapping: address space within which to check
463 * @start_byte: offset in bytes where the range starts
464 * @end_byte: offset in bytes where the range ends (inclusive)
466 * Find at least one page in the range supplied, usually used to check if
467 * direct writing in this range will trigger a writeback.
469 * Return: %true if at least one page exists in the specified range,
472 bool filemap_range_has_page(struct address_space
*mapping
,
473 loff_t start_byte
, loff_t end_byte
)
476 XA_STATE(xas
, &mapping
->i_pages
, start_byte
>> PAGE_SHIFT
);
477 pgoff_t max
= end_byte
>> PAGE_SHIFT
;
479 if (end_byte
< start_byte
)
484 page
= xas_find(&xas
, max
);
485 if (xas_retry(&xas
, page
))
487 /* Shadow entries don't count */
488 if (xa_is_value(page
))
491 * We don't need to try to pin this page; we're about to
492 * release the RCU lock anyway. It is enough to know that
493 * there was a page here recently.
501 EXPORT_SYMBOL(filemap_range_has_page
);
503 static void __filemap_fdatawait_range(struct address_space
*mapping
,
504 loff_t start_byte
, loff_t end_byte
)
506 pgoff_t index
= start_byte
>> PAGE_SHIFT
;
507 pgoff_t end
= end_byte
>> PAGE_SHIFT
;
511 if (end_byte
< start_byte
)
515 while (index
<= end
) {
518 nr_pages
= pagevec_lookup_range_tag(&pvec
, mapping
, &index
,
519 end
, PAGECACHE_TAG_WRITEBACK
);
523 for (i
= 0; i
< nr_pages
; i
++) {
524 struct page
*page
= pvec
.pages
[i
];
526 wait_on_page_writeback(page
);
527 ClearPageError(page
);
529 pagevec_release(&pvec
);
535 * filemap_fdatawait_range - wait for writeback to complete
536 * @mapping: address space structure to wait for
537 * @start_byte: offset in bytes where the range starts
538 * @end_byte: offset in bytes where the range ends (inclusive)
540 * Walk the list of under-writeback pages of the given address space
541 * in the given range and wait for all of them. Check error status of
542 * the address space and return it.
544 * Since the error status of the address space is cleared by this function,
545 * callers are responsible for checking the return value and handling and/or
546 * reporting the error.
548 * Return: error status of the address space.
550 int filemap_fdatawait_range(struct address_space
*mapping
, loff_t start_byte
,
553 __filemap_fdatawait_range(mapping
, start_byte
, end_byte
);
554 return filemap_check_errors(mapping
);
556 EXPORT_SYMBOL(filemap_fdatawait_range
);
559 * filemap_fdatawait_range_keep_errors - wait for writeback to complete
560 * @mapping: address space structure to wait for
561 * @start_byte: offset in bytes where the range starts
562 * @end_byte: offset in bytes where the range ends (inclusive)
564 * Walk the list of under-writeback pages of the given address space in the
565 * given range and wait for all of them. Unlike filemap_fdatawait_range(),
566 * this function does not clear error status of the address space.
568 * Use this function if callers don't handle errors themselves. Expected
569 * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2),
572 int filemap_fdatawait_range_keep_errors(struct address_space
*mapping
,
573 loff_t start_byte
, loff_t end_byte
)
575 __filemap_fdatawait_range(mapping
, start_byte
, end_byte
);
576 return filemap_check_and_keep_errors(mapping
);
578 EXPORT_SYMBOL(filemap_fdatawait_range_keep_errors
);
581 * file_fdatawait_range - wait for writeback to complete
582 * @file: file pointing to address space structure to wait for
583 * @start_byte: offset in bytes where the range starts
584 * @end_byte: offset in bytes where the range ends (inclusive)
586 * Walk the list of under-writeback pages of the address space that file
587 * refers to, in the given range and wait for all of them. Check error
588 * status of the address space vs. the file->f_wb_err cursor and return it.
590 * Since the error status of the file is advanced by this function,
591 * callers are responsible for checking the return value and handling and/or
592 * reporting the error.
594 * Return: error status of the address space vs. the file->f_wb_err cursor.
596 int file_fdatawait_range(struct file
*file
, loff_t start_byte
, loff_t end_byte
)
598 struct address_space
*mapping
= file
->f_mapping
;
600 __filemap_fdatawait_range(mapping
, start_byte
, end_byte
);
601 return file_check_and_advance_wb_err(file
);
603 EXPORT_SYMBOL(file_fdatawait_range
);
606 * filemap_fdatawait_keep_errors - wait for writeback without clearing errors
607 * @mapping: address space structure to wait for
609 * Walk the list of under-writeback pages of the given address space
610 * and wait for all of them. Unlike filemap_fdatawait(), this function
611 * does not clear error status of the address space.
613 * Use this function if callers don't handle errors themselves. Expected
614 * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2),
617 * Return: error status of the address space.
619 int filemap_fdatawait_keep_errors(struct address_space
*mapping
)
621 __filemap_fdatawait_range(mapping
, 0, LLONG_MAX
);
622 return filemap_check_and_keep_errors(mapping
);
624 EXPORT_SYMBOL(filemap_fdatawait_keep_errors
);
626 /* Returns true if writeback might be needed or already in progress. */
627 static bool mapping_needs_writeback(struct address_space
*mapping
)
629 if (dax_mapping(mapping
))
630 return mapping
->nrexceptional
;
632 return mapping
->nrpages
;
636 * filemap_write_and_wait_range - write out & wait on a file range
637 * @mapping: the address_space for the pages
638 * @lstart: offset in bytes where the range starts
639 * @lend: offset in bytes where the range ends (inclusive)
641 * Write out and wait upon file offsets lstart->lend, inclusive.
643 * Note that @lend is inclusive (describes the last byte to be written) so
644 * that this function can be used to write to the very end-of-file (end = -1).
646 * Return: error status of the address space.
648 int filemap_write_and_wait_range(struct address_space
*mapping
,
649 loff_t lstart
, loff_t lend
)
653 if (mapping_needs_writeback(mapping
)) {
654 err
= __filemap_fdatawrite_range(mapping
, lstart
, lend
,
657 * Even if the above returned error, the pages may be
658 * written partially (e.g. -ENOSPC), so we wait for it.
659 * But the -EIO is special case, it may indicate the worst
660 * thing (e.g. bug) happened, so we avoid waiting for it.
663 int err2
= filemap_fdatawait_range(mapping
,
668 /* Clear any previously stored errors */
669 filemap_check_errors(mapping
);
672 err
= filemap_check_errors(mapping
);
676 EXPORT_SYMBOL(filemap_write_and_wait_range
);
678 void __filemap_set_wb_err(struct address_space
*mapping
, int err
)
680 errseq_t eseq
= errseq_set(&mapping
->wb_err
, err
);
682 trace_filemap_set_wb_err(mapping
, eseq
);
684 EXPORT_SYMBOL(__filemap_set_wb_err
);
687 * file_check_and_advance_wb_err - report wb error (if any) that was previously
688 * and advance wb_err to current one
689 * @file: struct file on which the error is being reported
691 * When userland calls fsync (or something like nfsd does the equivalent), we
692 * want to report any writeback errors that occurred since the last fsync (or
693 * since the file was opened if there haven't been any).
695 * Grab the wb_err from the mapping. If it matches what we have in the file,
696 * then just quickly return 0. The file is all caught up.
698 * If it doesn't match, then take the mapping value, set the "seen" flag in
699 * it and try to swap it into place. If it works, or another task beat us
700 * to it with the new value, then update the f_wb_err and return the error
701 * portion. The error at this point must be reported via proper channels
702 * (a'la fsync, or NFS COMMIT operation, etc.).
704 * While we handle mapping->wb_err with atomic operations, the f_wb_err
705 * value is protected by the f_lock since we must ensure that it reflects
706 * the latest value swapped in for this file descriptor.
708 * Return: %0 on success, negative error code otherwise.
710 int file_check_and_advance_wb_err(struct file
*file
)
713 errseq_t old
= READ_ONCE(file
->f_wb_err
);
714 struct address_space
*mapping
= file
->f_mapping
;
716 /* Locklessly handle the common case where nothing has changed */
717 if (errseq_check(&mapping
->wb_err
, old
)) {
718 /* Something changed, must use slow path */
719 spin_lock(&file
->f_lock
);
720 old
= file
->f_wb_err
;
721 err
= errseq_check_and_advance(&mapping
->wb_err
,
723 trace_file_check_and_advance_wb_err(file
, old
);
724 spin_unlock(&file
->f_lock
);
728 * We're mostly using this function as a drop in replacement for
729 * filemap_check_errors. Clear AS_EIO/AS_ENOSPC to emulate the effect
730 * that the legacy code would have had on these flags.
732 clear_bit(AS_EIO
, &mapping
->flags
);
733 clear_bit(AS_ENOSPC
, &mapping
->flags
);
736 EXPORT_SYMBOL(file_check_and_advance_wb_err
);
739 * file_write_and_wait_range - write out & wait on a file range
740 * @file: file pointing to address_space with pages
741 * @lstart: offset in bytes where the range starts
742 * @lend: offset in bytes where the range ends (inclusive)
744 * Write out and wait upon file offsets lstart->lend, inclusive.
746 * Note that @lend is inclusive (describes the last byte to be written) so
747 * that this function can be used to write to the very end-of-file (end = -1).
749 * After writing out and waiting on the data, we check and advance the
750 * f_wb_err cursor to the latest value, and return any errors detected there.
752 * Return: %0 on success, negative error code otherwise.
754 int file_write_and_wait_range(struct file
*file
, loff_t lstart
, loff_t lend
)
757 struct address_space
*mapping
= file
->f_mapping
;
759 if (mapping_needs_writeback(mapping
)) {
760 err
= __filemap_fdatawrite_range(mapping
, lstart
, lend
,
762 /* See comment of filemap_write_and_wait() */
764 __filemap_fdatawait_range(mapping
, lstart
, lend
);
766 err2
= file_check_and_advance_wb_err(file
);
771 EXPORT_SYMBOL(file_write_and_wait_range
);
774 * replace_page_cache_page - replace a pagecache page with a new one
775 * @old: page to be replaced
776 * @new: page to replace with
777 * @gfp_mask: allocation mode
779 * This function replaces a page in the pagecache with a new one. On
780 * success it acquires the pagecache reference for the new page and
781 * drops it for the old page. Both the old and new pages must be
782 * locked. This function does not add the new page to the LRU, the
783 * caller must do that.
785 * The remove + add is atomic. This function cannot fail.
789 int replace_page_cache_page(struct page
*old
, struct page
*new, gfp_t gfp_mask
)
791 struct address_space
*mapping
= old
->mapping
;
792 void (*freepage
)(struct page
*) = mapping
->a_ops
->freepage
;
793 pgoff_t offset
= old
->index
;
794 XA_STATE(xas
, &mapping
->i_pages
, offset
);
797 VM_BUG_ON_PAGE(!PageLocked(old
), old
);
798 VM_BUG_ON_PAGE(!PageLocked(new), new);
799 VM_BUG_ON_PAGE(new->mapping
, new);
802 new->mapping
= mapping
;
805 mem_cgroup_migrate(old
, new);
807 xas_lock_irqsave(&xas
, flags
);
808 xas_store(&xas
, new);
811 /* hugetlb pages do not participate in page cache accounting. */
813 __dec_lruvec_page_state(old
, NR_FILE_PAGES
);
815 __inc_lruvec_page_state(new, NR_FILE_PAGES
);
816 if (PageSwapBacked(old
))
817 __dec_lruvec_page_state(old
, NR_SHMEM
);
818 if (PageSwapBacked(new))
819 __inc_lruvec_page_state(new, NR_SHMEM
);
820 xas_unlock_irqrestore(&xas
, flags
);
827 EXPORT_SYMBOL_GPL(replace_page_cache_page
);
829 static int __add_to_page_cache_locked(struct page
*page
,
830 struct address_space
*mapping
,
831 pgoff_t offset
, gfp_t gfp_mask
,
834 XA_STATE(xas
, &mapping
->i_pages
, offset
);
835 int huge
= PageHuge(page
);
839 VM_BUG_ON_PAGE(!PageLocked(page
), page
);
840 VM_BUG_ON_PAGE(PageSwapBacked(page
), page
);
841 mapping_set_update(&xas
, mapping
);
844 page
->mapping
= mapping
;
845 page
->index
= offset
;
848 error
= mem_cgroup_charge(page
, current
->mm
, gfp_mask
);
855 old
= xas_load(&xas
);
856 if (old
&& !xa_is_value(old
))
857 xas_set_err(&xas
, -EEXIST
);
858 xas_store(&xas
, page
);
862 if (xa_is_value(old
)) {
863 mapping
->nrexceptional
--;
869 /* hugetlb pages do not participate in page cache accounting */
871 __inc_lruvec_page_state(page
, NR_FILE_PAGES
);
873 xas_unlock_irq(&xas
);
874 } while (xas_nomem(&xas
, gfp_mask
& GFP_RECLAIM_MASK
));
876 if (xas_error(&xas
)) {
877 error
= xas_error(&xas
);
881 trace_mm_filemap_add_to_page_cache(page
);
884 page
->mapping
= NULL
;
885 /* Leave page->index set: truncation relies upon it */
889 ALLOW_ERROR_INJECTION(__add_to_page_cache_locked
, ERRNO
);
892 * add_to_page_cache_locked - add a locked page to the pagecache
894 * @mapping: the page's address_space
895 * @offset: page index
896 * @gfp_mask: page allocation mode
898 * This function is used to add a page to the pagecache. It must be locked.
899 * This function does not add the page to the LRU. The caller must do that.
901 * Return: %0 on success, negative error code otherwise.
903 int add_to_page_cache_locked(struct page
*page
, struct address_space
*mapping
,
904 pgoff_t offset
, gfp_t gfp_mask
)
906 return __add_to_page_cache_locked(page
, mapping
, offset
,
909 EXPORT_SYMBOL(add_to_page_cache_locked
);
911 int add_to_page_cache_lru(struct page
*page
, struct address_space
*mapping
,
912 pgoff_t offset
, gfp_t gfp_mask
)
917 __SetPageLocked(page
);
918 ret
= __add_to_page_cache_locked(page
, mapping
, offset
,
921 __ClearPageLocked(page
);
924 * The page might have been evicted from cache only
925 * recently, in which case it should be activated like
926 * any other repeatedly accessed page.
927 * The exception is pages getting rewritten; evicting other
928 * data from the working set, only to cache data that will
929 * get overwritten with something else, is a waste of memory.
931 WARN_ON_ONCE(PageActive(page
));
932 if (!(gfp_mask
& __GFP_WRITE
) && shadow
)
933 workingset_refault(page
, shadow
);
938 EXPORT_SYMBOL_GPL(add_to_page_cache_lru
);
941 struct page
*__page_cache_alloc(gfp_t gfp
)
946 if (cpuset_do_page_mem_spread()) {
947 unsigned int cpuset_mems_cookie
;
949 cpuset_mems_cookie
= read_mems_allowed_begin();
950 n
= cpuset_mem_spread_node();
951 page
= __alloc_pages_node(n
, gfp
, 0);
952 } while (!page
&& read_mems_allowed_retry(cpuset_mems_cookie
));
956 return alloc_pages(gfp
, 0);
958 EXPORT_SYMBOL(__page_cache_alloc
);
962 * In order to wait for pages to become available there must be
963 * waitqueues associated with pages. By using a hash table of
964 * waitqueues where the bucket discipline is to maintain all
965 * waiters on the same queue and wake all when any of the pages
966 * become available, and for the woken contexts to check to be
967 * sure the appropriate page became available, this saves space
968 * at a cost of "thundering herd" phenomena during rare hash
971 #define PAGE_WAIT_TABLE_BITS 8
972 #define PAGE_WAIT_TABLE_SIZE (1 << PAGE_WAIT_TABLE_BITS)
973 static wait_queue_head_t page_wait_table
[PAGE_WAIT_TABLE_SIZE
] __cacheline_aligned
;
975 static wait_queue_head_t
*page_waitqueue(struct page
*page
)
977 return &page_wait_table
[hash_ptr(page
, PAGE_WAIT_TABLE_BITS
)];
980 void __init
pagecache_init(void)
984 for (i
= 0; i
< PAGE_WAIT_TABLE_SIZE
; i
++)
985 init_waitqueue_head(&page_wait_table
[i
]);
987 page_writeback_init();
990 /* This has the same layout as wait_bit_key - see fs/cachefiles/rdwr.c */
991 struct wait_page_key
{
997 struct wait_page_queue
{
1000 wait_queue_entry_t wait
;
1003 static int wake_page_function(wait_queue_entry_t
*wait
, unsigned mode
, int sync
, void *arg
)
1005 struct wait_page_key
*key
= arg
;
1006 struct wait_page_queue
*wait_page
1007 = container_of(wait
, struct wait_page_queue
, wait
);
1009 if (wait_page
->page
!= key
->page
)
1011 key
->page_match
= 1;
1013 if (wait_page
->bit_nr
!= key
->bit_nr
)
1017 * Stop walking if it's locked.
1018 * Is this safe if put_and_wait_on_page_locked() is in use?
1019 * Yes: the waker must hold a reference to this page, and if PG_locked
1020 * has now already been set by another task, that task must also hold
1021 * a reference to the *same usage* of this page; so there is no need
1022 * to walk on to wake even the put_and_wait_on_page_locked() callers.
1024 if (test_bit(key
->bit_nr
, &key
->page
->flags
))
1027 return autoremove_wake_function(wait
, mode
, sync
, key
);
1030 static void wake_up_page_bit(struct page
*page
, int bit_nr
)
1032 wait_queue_head_t
*q
= page_waitqueue(page
);
1033 struct wait_page_key key
;
1034 unsigned long flags
;
1035 wait_queue_entry_t bookmark
;
1038 key
.bit_nr
= bit_nr
;
1042 bookmark
.private = NULL
;
1043 bookmark
.func
= NULL
;
1044 INIT_LIST_HEAD(&bookmark
.entry
);
1046 spin_lock_irqsave(&q
->lock
, flags
);
1047 __wake_up_locked_key_bookmark(q
, TASK_NORMAL
, &key
, &bookmark
);
1049 while (bookmark
.flags
& WQ_FLAG_BOOKMARK
) {
1051 * Take a breather from holding the lock,
1052 * allow pages that finish wake up asynchronously
1053 * to acquire the lock and remove themselves
1056 spin_unlock_irqrestore(&q
->lock
, flags
);
1058 spin_lock_irqsave(&q
->lock
, flags
);
1059 __wake_up_locked_key_bookmark(q
, TASK_NORMAL
, &key
, &bookmark
);
1063 * It is possible for other pages to have collided on the waitqueue
1064 * hash, so in that case check for a page match. That prevents a long-
1067 * It is still possible to miss a case here, when we woke page waiters
1068 * and removed them from the waitqueue, but there are still other
1071 if (!waitqueue_active(q
) || !key
.page_match
) {
1072 ClearPageWaiters(page
);
1074 * It's possible to miss clearing Waiters here, when we woke
1075 * our page waiters, but the hashed waitqueue has waiters for
1076 * other pages on it.
1078 * That's okay, it's a rare case. The next waker will clear it.
1081 spin_unlock_irqrestore(&q
->lock
, flags
);
1084 static void wake_up_page(struct page
*page
, int bit
)
1086 if (!PageWaiters(page
))
1088 wake_up_page_bit(page
, bit
);
1092 * A choice of three behaviors for wait_on_page_bit_common():
1095 EXCLUSIVE
, /* Hold ref to page and take the bit when woken, like
1096 * __lock_page() waiting on then setting PG_locked.
1098 SHARED
, /* Hold ref to page and check the bit when woken, like
1099 * wait_on_page_writeback() waiting on PG_writeback.
1101 DROP
, /* Drop ref to page before wait, no check when woken,
1102 * like put_and_wait_on_page_locked() on PG_locked.
1106 static inline int wait_on_page_bit_common(wait_queue_head_t
*q
,
1107 struct page
*page
, int bit_nr
, int state
, enum behavior behavior
)
1109 struct wait_page_queue wait_page
;
1110 wait_queue_entry_t
*wait
= &wait_page
.wait
;
1112 bool thrashing
= false;
1113 bool delayacct
= false;
1114 unsigned long pflags
;
1117 if (bit_nr
== PG_locked
&&
1118 !PageUptodate(page
) && PageWorkingset(page
)) {
1119 if (!PageSwapBacked(page
)) {
1120 delayacct_thrashing_start();
1123 psi_memstall_enter(&pflags
);
1128 wait
->flags
= behavior
== EXCLUSIVE
? WQ_FLAG_EXCLUSIVE
: 0;
1129 wait
->func
= wake_page_function
;
1130 wait_page
.page
= page
;
1131 wait_page
.bit_nr
= bit_nr
;
1134 spin_lock_irq(&q
->lock
);
1136 if (likely(list_empty(&wait
->entry
))) {
1137 __add_wait_queue_entry_tail(q
, wait
);
1138 SetPageWaiters(page
);
1141 set_current_state(state
);
1143 spin_unlock_irq(&q
->lock
);
1145 bit_is_set
= test_bit(bit_nr
, &page
->flags
);
1146 if (behavior
== DROP
)
1149 if (likely(bit_is_set
))
1152 if (behavior
== EXCLUSIVE
) {
1153 if (!test_and_set_bit_lock(bit_nr
, &page
->flags
))
1155 } else if (behavior
== SHARED
) {
1156 if (!test_bit(bit_nr
, &page
->flags
))
1160 if (signal_pending_state(state
, current
)) {
1165 if (behavior
== DROP
) {
1167 * We can no longer safely access page->flags:
1168 * even if CONFIG_MEMORY_HOTREMOVE is not enabled,
1169 * there is a risk of waiting forever on a page reused
1170 * for something that keeps it locked indefinitely.
1171 * But best check for -EINTR above before breaking.
1177 finish_wait(q
, wait
);
1181 delayacct_thrashing_end();
1182 psi_memstall_leave(&pflags
);
1186 * A signal could leave PageWaiters set. Clearing it here if
1187 * !waitqueue_active would be possible (by open-coding finish_wait),
1188 * but still fail to catch it in the case of wait hash collision. We
1189 * already can fail to clear wait hash collision cases, so don't
1190 * bother with signals either.
1196 void wait_on_page_bit(struct page
*page
, int bit_nr
)
1198 wait_queue_head_t
*q
= page_waitqueue(page
);
1199 wait_on_page_bit_common(q
, page
, bit_nr
, TASK_UNINTERRUPTIBLE
, SHARED
);
1201 EXPORT_SYMBOL(wait_on_page_bit
);
1203 int wait_on_page_bit_killable(struct page
*page
, int bit_nr
)
1205 wait_queue_head_t
*q
= page_waitqueue(page
);
1206 return wait_on_page_bit_common(q
, page
, bit_nr
, TASK_KILLABLE
, SHARED
);
1208 EXPORT_SYMBOL(wait_on_page_bit_killable
);
1211 * put_and_wait_on_page_locked - Drop a reference and wait for it to be unlocked
1212 * @page: The page to wait for.
1214 * The caller should hold a reference on @page. They expect the page to
1215 * become unlocked relatively soon, but do not wish to hold up migration
1216 * (for example) by holding the reference while waiting for the page to
1217 * come unlocked. After this function returns, the caller should not
1218 * dereference @page.
1220 void put_and_wait_on_page_locked(struct page
*page
)
1222 wait_queue_head_t
*q
;
1224 page
= compound_head(page
);
1225 q
= page_waitqueue(page
);
1226 wait_on_page_bit_common(q
, page
, PG_locked
, TASK_UNINTERRUPTIBLE
, DROP
);
1230 * add_page_wait_queue - Add an arbitrary waiter to a page's wait queue
1231 * @page: Page defining the wait queue of interest
1232 * @waiter: Waiter to add to the queue
1234 * Add an arbitrary @waiter to the wait queue for the nominated @page.
1236 void add_page_wait_queue(struct page
*page
, wait_queue_entry_t
*waiter
)
1238 wait_queue_head_t
*q
= page_waitqueue(page
);
1239 unsigned long flags
;
1241 spin_lock_irqsave(&q
->lock
, flags
);
1242 __add_wait_queue_entry_tail(q
, waiter
);
1243 SetPageWaiters(page
);
1244 spin_unlock_irqrestore(&q
->lock
, flags
);
1246 EXPORT_SYMBOL_GPL(add_page_wait_queue
);
1248 #ifndef clear_bit_unlock_is_negative_byte
1251 * PG_waiters is the high bit in the same byte as PG_lock.
1253 * On x86 (and on many other architectures), we can clear PG_lock and
1254 * test the sign bit at the same time. But if the architecture does
1255 * not support that special operation, we just do this all by hand
1258 * The read of PG_waiters has to be after (or concurrently with) PG_locked
1259 * being cleared, but a memory barrier should be unnecessary since it is
1260 * in the same byte as PG_locked.
1262 static inline bool clear_bit_unlock_is_negative_byte(long nr
, volatile void *mem
)
1264 clear_bit_unlock(nr
, mem
);
1265 /* smp_mb__after_atomic(); */
1266 return test_bit(PG_waiters
, mem
);
1272 * unlock_page - unlock a locked page
1275 * Unlocks the page and wakes up sleepers in ___wait_on_page_locked().
1276 * Also wakes sleepers in wait_on_page_writeback() because the wakeup
1277 * mechanism between PageLocked pages and PageWriteback pages is shared.
1278 * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep.
1280 * Note that this depends on PG_waiters being the sign bit in the byte
1281 * that contains PG_locked - thus the BUILD_BUG_ON(). That allows us to
1282 * clear the PG_locked bit and test PG_waiters at the same time fairly
1283 * portably (architectures that do LL/SC can test any bit, while x86 can
1284 * test the sign bit).
1286 void unlock_page(struct page
*page
)
1288 BUILD_BUG_ON(PG_waiters
!= 7);
1289 page
= compound_head(page
);
1290 VM_BUG_ON_PAGE(!PageLocked(page
), page
);
1291 if (clear_bit_unlock_is_negative_byte(PG_locked
, &page
->flags
))
1292 wake_up_page_bit(page
, PG_locked
);
1294 EXPORT_SYMBOL(unlock_page
);
1297 * end_page_writeback - end writeback against a page
1300 void end_page_writeback(struct page
*page
)
1303 * TestClearPageReclaim could be used here but it is an atomic
1304 * operation and overkill in this particular case. Failing to
1305 * shuffle a page marked for immediate reclaim is too mild to
1306 * justify taking an atomic operation penalty at the end of
1307 * ever page writeback.
1309 if (PageReclaim(page
)) {
1310 ClearPageReclaim(page
);
1311 rotate_reclaimable_page(page
);
1314 if (!test_clear_page_writeback(page
))
1317 smp_mb__after_atomic();
1318 wake_up_page(page
, PG_writeback
);
1320 EXPORT_SYMBOL(end_page_writeback
);
1323 * After completing I/O on a page, call this routine to update the page
1324 * flags appropriately
1326 void page_endio(struct page
*page
, bool is_write
, int err
)
1330 SetPageUptodate(page
);
1332 ClearPageUptodate(page
);
1338 struct address_space
*mapping
;
1341 mapping
= page_mapping(page
);
1343 mapping_set_error(mapping
, err
);
1345 end_page_writeback(page
);
1348 EXPORT_SYMBOL_GPL(page_endio
);
1351 * __lock_page - get a lock on the page, assuming we need to sleep to get it
1352 * @__page: the page to lock
1354 void __lock_page(struct page
*__page
)
1356 struct page
*page
= compound_head(__page
);
1357 wait_queue_head_t
*q
= page_waitqueue(page
);
1358 wait_on_page_bit_common(q
, page
, PG_locked
, TASK_UNINTERRUPTIBLE
,
1361 EXPORT_SYMBOL(__lock_page
);
1363 int __lock_page_killable(struct page
*__page
)
1365 struct page
*page
= compound_head(__page
);
1366 wait_queue_head_t
*q
= page_waitqueue(page
);
1367 return wait_on_page_bit_common(q
, page
, PG_locked
, TASK_KILLABLE
,
1370 EXPORT_SYMBOL_GPL(__lock_page_killable
);
1374 * 1 - page is locked; mmap_lock is still held.
1375 * 0 - page is not locked.
1376 * mmap_lock has been released (mmap_read_unlock(), unless flags had both
1377 * FAULT_FLAG_ALLOW_RETRY and FAULT_FLAG_RETRY_NOWAIT set, in
1378 * which case mmap_lock is still held.
1380 * If neither ALLOW_RETRY nor KILLABLE are set, will always return 1
1381 * with the page locked and the mmap_lock unperturbed.
1383 int __lock_page_or_retry(struct page
*page
, struct mm_struct
*mm
,
1386 if (fault_flag_allow_retry_first(flags
)) {
1388 * CAUTION! In this case, mmap_lock is not released
1389 * even though return 0.
1391 if (flags
& FAULT_FLAG_RETRY_NOWAIT
)
1394 mmap_read_unlock(mm
);
1395 if (flags
& FAULT_FLAG_KILLABLE
)
1396 wait_on_page_locked_killable(page
);
1398 wait_on_page_locked(page
);
1401 if (flags
& FAULT_FLAG_KILLABLE
) {
1404 ret
= __lock_page_killable(page
);
1406 mmap_read_unlock(mm
);
1416 * page_cache_next_miss() - Find the next gap in the page cache.
1417 * @mapping: Mapping.
1419 * @max_scan: Maximum range to search.
1421 * Search the range [index, min(index + max_scan - 1, ULONG_MAX)] for the
1422 * gap with the lowest index.
1424 * This function may be called under the rcu_read_lock. However, this will
1425 * not atomically search a snapshot of the cache at a single point in time.
1426 * For example, if a gap is created at index 5, then subsequently a gap is
1427 * created at index 10, page_cache_next_miss covering both indices may
1428 * return 10 if called under the rcu_read_lock.
1430 * Return: The index of the gap if found, otherwise an index outside the
1431 * range specified (in which case 'return - index >= max_scan' will be true).
1432 * In the rare case of index wrap-around, 0 will be returned.
1434 pgoff_t
page_cache_next_miss(struct address_space
*mapping
,
1435 pgoff_t index
, unsigned long max_scan
)
1437 XA_STATE(xas
, &mapping
->i_pages
, index
);
1439 while (max_scan
--) {
1440 void *entry
= xas_next(&xas
);
1441 if (!entry
|| xa_is_value(entry
))
1443 if (xas
.xa_index
== 0)
1447 return xas
.xa_index
;
1449 EXPORT_SYMBOL(page_cache_next_miss
);
1452 * page_cache_prev_miss() - Find the previous gap in the page cache.
1453 * @mapping: Mapping.
1455 * @max_scan: Maximum range to search.
1457 * Search the range [max(index - max_scan + 1, 0), index] for the
1458 * gap with the highest index.
1460 * This function may be called under the rcu_read_lock. However, this will
1461 * not atomically search a snapshot of the cache at a single point in time.
1462 * For example, if a gap is created at index 10, then subsequently a gap is
1463 * created at index 5, page_cache_prev_miss() covering both indices may
1464 * return 5 if called under the rcu_read_lock.
1466 * Return: The index of the gap if found, otherwise an index outside the
1467 * range specified (in which case 'index - return >= max_scan' will be true).
1468 * In the rare case of wrap-around, ULONG_MAX will be returned.
1470 pgoff_t
page_cache_prev_miss(struct address_space
*mapping
,
1471 pgoff_t index
, unsigned long max_scan
)
1473 XA_STATE(xas
, &mapping
->i_pages
, index
);
1475 while (max_scan
--) {
1476 void *entry
= xas_prev(&xas
);
1477 if (!entry
|| xa_is_value(entry
))
1479 if (xas
.xa_index
== ULONG_MAX
)
1483 return xas
.xa_index
;
1485 EXPORT_SYMBOL(page_cache_prev_miss
);
1488 * find_get_entry - find and get a page cache entry
1489 * @mapping: the address_space to search
1490 * @offset: the page cache index
1492 * Looks up the page cache slot at @mapping & @offset. If there is a
1493 * page cache page, it is returned with an increased refcount.
1495 * If the slot holds a shadow entry of a previously evicted page, or a
1496 * swap entry from shmem/tmpfs, it is returned.
1498 * Return: the found page or shadow entry, %NULL if nothing is found.
1500 struct page
*find_get_entry(struct address_space
*mapping
, pgoff_t offset
)
1502 XA_STATE(xas
, &mapping
->i_pages
, offset
);
1508 page
= xas_load(&xas
);
1509 if (xas_retry(&xas
, page
))
1512 * A shadow entry of a recently evicted page, or a swap entry from
1513 * shmem/tmpfs. Return it without attempting to raise page count.
1515 if (!page
|| xa_is_value(page
))
1518 if (!page_cache_get_speculative(page
))
1522 * Has the page moved or been split?
1523 * This is part of the lockless pagecache protocol. See
1524 * include/linux/pagemap.h for details.
1526 if (unlikely(page
!= xas_reload(&xas
))) {
1530 page
= find_subpage(page
, offset
);
1538 * find_lock_entry - locate, pin and lock a page cache entry
1539 * @mapping: the address_space to search
1540 * @offset: the page cache index
1542 * Looks up the page cache slot at @mapping & @offset. If there is a
1543 * page cache page, it is returned locked and with an increased
1546 * If the slot holds a shadow entry of a previously evicted page, or a
1547 * swap entry from shmem/tmpfs, it is returned.
1549 * find_lock_entry() may sleep.
1551 * Return: the found page or shadow entry, %NULL if nothing is found.
1553 struct page
*find_lock_entry(struct address_space
*mapping
, pgoff_t offset
)
1558 page
= find_get_entry(mapping
, offset
);
1559 if (page
&& !xa_is_value(page
)) {
1561 /* Has the page been truncated? */
1562 if (unlikely(page_mapping(page
) != mapping
)) {
1567 VM_BUG_ON_PAGE(page_to_pgoff(page
) != offset
, page
);
1571 EXPORT_SYMBOL(find_lock_entry
);
1574 * pagecache_get_page - Find and get a reference to a page.
1575 * @mapping: The address_space to search.
1576 * @index: The page index.
1577 * @fgp_flags: %FGP flags modify how the page is returned.
1578 * @gfp_mask: Memory allocation flags to use if %FGP_CREAT is specified.
1580 * Looks up the page cache entry at @mapping & @index.
1582 * @fgp_flags can be zero or more of these flags:
1584 * * %FGP_ACCESSED - The page will be marked accessed.
1585 * * %FGP_LOCK - The page is returned locked.
1586 * * %FGP_CREAT - If no page is present then a new page is allocated using
1587 * @gfp_mask and added to the page cache and the VM's LRU list.
1588 * The page is returned locked and with an increased refcount.
1589 * * %FGP_FOR_MMAP - The caller wants to do its own locking dance if the
1590 * page is already in cache. If the page was allocated, unlock it before
1591 * returning so the caller can do the same dance.
1593 * If %FGP_LOCK or %FGP_CREAT are specified then the function may sleep even
1594 * if the %GFP flags specified for %FGP_CREAT are atomic.
1596 * If there is a page cache page, it is returned with an increased refcount.
1598 * Return: The found page or %NULL otherwise.
1600 struct page
*pagecache_get_page(struct address_space
*mapping
, pgoff_t index
,
1601 int fgp_flags
, gfp_t gfp_mask
)
1606 page
= find_get_entry(mapping
, index
);
1607 if (xa_is_value(page
))
1612 if (fgp_flags
& FGP_LOCK
) {
1613 if (fgp_flags
& FGP_NOWAIT
) {
1614 if (!trylock_page(page
)) {
1622 /* Has the page been truncated? */
1623 if (unlikely(compound_head(page
)->mapping
!= mapping
)) {
1628 VM_BUG_ON_PAGE(page
->index
!= index
, page
);
1631 if (fgp_flags
& FGP_ACCESSED
)
1632 mark_page_accessed(page
);
1635 if (!page
&& (fgp_flags
& FGP_CREAT
)) {
1637 if ((fgp_flags
& FGP_WRITE
) && mapping_cap_account_dirty(mapping
))
1638 gfp_mask
|= __GFP_WRITE
;
1639 if (fgp_flags
& FGP_NOFS
)
1640 gfp_mask
&= ~__GFP_FS
;
1642 page
= __page_cache_alloc(gfp_mask
);
1646 if (WARN_ON_ONCE(!(fgp_flags
& (FGP_LOCK
| FGP_FOR_MMAP
))))
1647 fgp_flags
|= FGP_LOCK
;
1649 /* Init accessed so avoid atomic mark_page_accessed later */
1650 if (fgp_flags
& FGP_ACCESSED
)
1651 __SetPageReferenced(page
);
1653 err
= add_to_page_cache_lru(page
, mapping
, index
, gfp_mask
);
1654 if (unlikely(err
)) {
1662 * add_to_page_cache_lru locks the page, and for mmap we expect
1665 if (page
&& (fgp_flags
& FGP_FOR_MMAP
))
1671 EXPORT_SYMBOL(pagecache_get_page
);
1674 * find_get_entries - gang pagecache lookup
1675 * @mapping: The address_space to search
1676 * @start: The starting page cache index
1677 * @nr_entries: The maximum number of entries
1678 * @entries: Where the resulting entries are placed
1679 * @indices: The cache indices corresponding to the entries in @entries
1681 * find_get_entries() will search for and return a group of up to
1682 * @nr_entries entries in the mapping. The entries are placed at
1683 * @entries. find_get_entries() takes a reference against any actual
1686 * The search returns a group of mapping-contiguous page cache entries
1687 * with ascending indexes. There may be holes in the indices due to
1688 * not-present pages.
1690 * Any shadow entries of evicted pages, or swap entries from
1691 * shmem/tmpfs, are included in the returned array.
1693 * If it finds a Transparent Huge Page, head or tail, find_get_entries()
1694 * stops at that page: the caller is likely to have a better way to handle
1695 * the compound page as a whole, and then skip its extent, than repeatedly
1696 * calling find_get_entries() to return all its tails.
1698 * Return: the number of pages and shadow entries which were found.
1700 unsigned find_get_entries(struct address_space
*mapping
,
1701 pgoff_t start
, unsigned int nr_entries
,
1702 struct page
**entries
, pgoff_t
*indices
)
1704 XA_STATE(xas
, &mapping
->i_pages
, start
);
1706 unsigned int ret
= 0;
1712 xas_for_each(&xas
, page
, ULONG_MAX
) {
1713 if (xas_retry(&xas
, page
))
1716 * A shadow entry of a recently evicted page, a swap
1717 * entry from shmem/tmpfs or a DAX entry. Return it
1718 * without attempting to raise page count.
1720 if (xa_is_value(page
))
1723 if (!page_cache_get_speculative(page
))
1726 /* Has the page moved or been split? */
1727 if (unlikely(page
!= xas_reload(&xas
)))
1731 * Terminate early on finding a THP, to allow the caller to
1732 * handle it all at once; but continue if this is hugetlbfs.
1734 if (PageTransHuge(page
) && !PageHuge(page
)) {
1735 page
= find_subpage(page
, xas
.xa_index
);
1736 nr_entries
= ret
+ 1;
1739 indices
[ret
] = xas
.xa_index
;
1740 entries
[ret
] = page
;
1741 if (++ret
== nr_entries
)
1754 * find_get_pages_range - gang pagecache lookup
1755 * @mapping: The address_space to search
1756 * @start: The starting page index
1757 * @end: The final page index (inclusive)
1758 * @nr_pages: The maximum number of pages
1759 * @pages: Where the resulting pages are placed
1761 * find_get_pages_range() will search for and return a group of up to @nr_pages
1762 * pages in the mapping starting at index @start and up to index @end
1763 * (inclusive). The pages are placed at @pages. find_get_pages_range() takes
1764 * a reference against the returned pages.
1766 * The search returns a group of mapping-contiguous pages with ascending
1767 * indexes. There may be holes in the indices due to not-present pages.
1768 * We also update @start to index the next page for the traversal.
1770 * Return: the number of pages which were found. If this number is
1771 * smaller than @nr_pages, the end of specified range has been
1774 unsigned find_get_pages_range(struct address_space
*mapping
, pgoff_t
*start
,
1775 pgoff_t end
, unsigned int nr_pages
,
1776 struct page
**pages
)
1778 XA_STATE(xas
, &mapping
->i_pages
, *start
);
1782 if (unlikely(!nr_pages
))
1786 xas_for_each(&xas
, page
, end
) {
1787 if (xas_retry(&xas
, page
))
1789 /* Skip over shadow, swap and DAX entries */
1790 if (xa_is_value(page
))
1793 if (!page_cache_get_speculative(page
))
1796 /* Has the page moved or been split? */
1797 if (unlikely(page
!= xas_reload(&xas
)))
1800 pages
[ret
] = find_subpage(page
, xas
.xa_index
);
1801 if (++ret
== nr_pages
) {
1802 *start
= xas
.xa_index
+ 1;
1813 * We come here when there is no page beyond @end. We take care to not
1814 * overflow the index @start as it confuses some of the callers. This
1815 * breaks the iteration when there is a page at index -1 but that is
1816 * already broken anyway.
1818 if (end
== (pgoff_t
)-1)
1819 *start
= (pgoff_t
)-1;
1829 * find_get_pages_contig - gang contiguous pagecache lookup
1830 * @mapping: The address_space to search
1831 * @index: The starting page index
1832 * @nr_pages: The maximum number of pages
1833 * @pages: Where the resulting pages are placed
1835 * find_get_pages_contig() works exactly like find_get_pages(), except
1836 * that the returned number of pages are guaranteed to be contiguous.
1838 * Return: the number of pages which were found.
1840 unsigned find_get_pages_contig(struct address_space
*mapping
, pgoff_t index
,
1841 unsigned int nr_pages
, struct page
**pages
)
1843 XA_STATE(xas
, &mapping
->i_pages
, index
);
1845 unsigned int ret
= 0;
1847 if (unlikely(!nr_pages
))
1851 for (page
= xas_load(&xas
); page
; page
= xas_next(&xas
)) {
1852 if (xas_retry(&xas
, page
))
1855 * If the entry has been swapped out, we can stop looking.
1856 * No current caller is looking for DAX entries.
1858 if (xa_is_value(page
))
1861 if (!page_cache_get_speculative(page
))
1864 /* Has the page moved or been split? */
1865 if (unlikely(page
!= xas_reload(&xas
)))
1868 pages
[ret
] = find_subpage(page
, xas
.xa_index
);
1869 if (++ret
== nr_pages
)
1880 EXPORT_SYMBOL(find_get_pages_contig
);
1883 * find_get_pages_range_tag - find and return pages in given range matching @tag
1884 * @mapping: the address_space to search
1885 * @index: the starting page index
1886 * @end: The final page index (inclusive)
1887 * @tag: the tag index
1888 * @nr_pages: the maximum number of pages
1889 * @pages: where the resulting pages are placed
1891 * Like find_get_pages, except we only return pages which are tagged with
1892 * @tag. We update @index to index the next page for the traversal.
1894 * Return: the number of pages which were found.
1896 unsigned find_get_pages_range_tag(struct address_space
*mapping
, pgoff_t
*index
,
1897 pgoff_t end
, xa_mark_t tag
, unsigned int nr_pages
,
1898 struct page
**pages
)
1900 XA_STATE(xas
, &mapping
->i_pages
, *index
);
1904 if (unlikely(!nr_pages
))
1908 xas_for_each_marked(&xas
, page
, end
, tag
) {
1909 if (xas_retry(&xas
, page
))
1912 * Shadow entries should never be tagged, but this iteration
1913 * is lockless so there is a window for page reclaim to evict
1914 * a page we saw tagged. Skip over it.
1916 if (xa_is_value(page
))
1919 if (!page_cache_get_speculative(page
))
1922 /* Has the page moved or been split? */
1923 if (unlikely(page
!= xas_reload(&xas
)))
1926 pages
[ret
] = find_subpage(page
, xas
.xa_index
);
1927 if (++ret
== nr_pages
) {
1928 *index
= xas
.xa_index
+ 1;
1939 * We come here when we got to @end. We take care to not overflow the
1940 * index @index as it confuses some of the callers. This breaks the
1941 * iteration when there is a page at index -1 but that is already
1944 if (end
== (pgoff_t
)-1)
1945 *index
= (pgoff_t
)-1;
1953 EXPORT_SYMBOL(find_get_pages_range_tag
);
1956 * CD/DVDs are error prone. When a medium error occurs, the driver may fail
1957 * a _large_ part of the i/o request. Imagine the worst scenario:
1959 * ---R__________________________________________B__________
1960 * ^ reading here ^ bad block(assume 4k)
1962 * read(R) => miss => readahead(R...B) => media error => frustrating retries
1963 * => failing the whole request => read(R) => read(R+1) =>
1964 * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
1965 * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
1966 * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
1968 * It is going insane. Fix it by quickly scaling down the readahead size.
1970 static void shrink_readahead_size_eio(struct file_ra_state
*ra
)
1976 * generic_file_buffered_read - generic file read routine
1977 * @iocb: the iocb to read
1978 * @iter: data destination
1979 * @written: already copied
1981 * This is a generic file read routine, and uses the
1982 * mapping->a_ops->readpage() function for the actual low-level stuff.
1984 * This is really ugly. But the goto's actually try to clarify some
1985 * of the logic when it comes to error handling etc.
1988 * * total number of bytes copied, including those the were already @written
1989 * * negative error code if nothing was copied
1991 ssize_t
generic_file_buffered_read(struct kiocb
*iocb
,
1992 struct iov_iter
*iter
, ssize_t written
)
1994 struct file
*filp
= iocb
->ki_filp
;
1995 struct address_space
*mapping
= filp
->f_mapping
;
1996 struct inode
*inode
= mapping
->host
;
1997 struct file_ra_state
*ra
= &filp
->f_ra
;
1998 loff_t
*ppos
= &iocb
->ki_pos
;
2002 unsigned long offset
; /* offset into pagecache page */
2003 unsigned int prev_offset
;
2006 if (unlikely(*ppos
>= inode
->i_sb
->s_maxbytes
))
2008 iov_iter_truncate(iter
, inode
->i_sb
->s_maxbytes
);
2010 index
= *ppos
>> PAGE_SHIFT
;
2011 prev_index
= ra
->prev_pos
>> PAGE_SHIFT
;
2012 prev_offset
= ra
->prev_pos
& (PAGE_SIZE
-1);
2013 last_index
= (*ppos
+ iter
->count
+ PAGE_SIZE
-1) >> PAGE_SHIFT
;
2014 offset
= *ppos
& ~PAGE_MASK
;
2020 unsigned long nr
, ret
;
2024 if (fatal_signal_pending(current
)) {
2029 page
= find_get_page(mapping
, index
);
2031 if (iocb
->ki_flags
& IOCB_NOWAIT
)
2033 page_cache_sync_readahead(mapping
,
2035 index
, last_index
- index
);
2036 page
= find_get_page(mapping
, index
);
2037 if (unlikely(page
== NULL
))
2038 goto no_cached_page
;
2040 if (PageReadahead(page
)) {
2041 page_cache_async_readahead(mapping
,
2043 index
, last_index
- index
);
2045 if (!PageUptodate(page
)) {
2046 if (iocb
->ki_flags
& IOCB_NOWAIT
) {
2052 * See comment in do_read_cache_page on why
2053 * wait_on_page_locked is used to avoid unnecessarily
2054 * serialisations and why it's safe.
2056 error
= wait_on_page_locked_killable(page
);
2057 if (unlikely(error
))
2058 goto readpage_error
;
2059 if (PageUptodate(page
))
2062 if (inode
->i_blkbits
== PAGE_SHIFT
||
2063 !mapping
->a_ops
->is_partially_uptodate
)
2064 goto page_not_up_to_date
;
2065 /* pipes can't handle partially uptodate pages */
2066 if (unlikely(iov_iter_is_pipe(iter
)))
2067 goto page_not_up_to_date
;
2068 if (!trylock_page(page
))
2069 goto page_not_up_to_date
;
2070 /* Did it get truncated before we got the lock? */
2072 goto page_not_up_to_date_locked
;
2073 if (!mapping
->a_ops
->is_partially_uptodate(page
,
2074 offset
, iter
->count
))
2075 goto page_not_up_to_date_locked
;
2080 * i_size must be checked after we know the page is Uptodate.
2082 * Checking i_size after the check allows us to calculate
2083 * the correct value for "nr", which means the zero-filled
2084 * part of the page is not copied back to userspace (unless
2085 * another truncate extends the file - this is desired though).
2088 isize
= i_size_read(inode
);
2089 end_index
= (isize
- 1) >> PAGE_SHIFT
;
2090 if (unlikely(!isize
|| index
> end_index
)) {
2095 /* nr is the maximum number of bytes to copy from this page */
2097 if (index
== end_index
) {
2098 nr
= ((isize
- 1) & ~PAGE_MASK
) + 1;
2106 /* If users can be writing to this page using arbitrary
2107 * virtual addresses, take care about potential aliasing
2108 * before reading the page on the kernel side.
2110 if (mapping_writably_mapped(mapping
))
2111 flush_dcache_page(page
);
2114 * When a sequential read accesses a page several times,
2115 * only mark it as accessed the first time.
2117 if (prev_index
!= index
|| offset
!= prev_offset
)
2118 mark_page_accessed(page
);
2122 * Ok, we have the page, and it's up-to-date, so
2123 * now we can copy it to user space...
2126 ret
= copy_page_to_iter(page
, offset
, nr
, iter
);
2128 index
+= offset
>> PAGE_SHIFT
;
2129 offset
&= ~PAGE_MASK
;
2130 prev_offset
= offset
;
2134 if (!iov_iter_count(iter
))
2142 page_not_up_to_date
:
2143 /* Get exclusive access to the page ... */
2144 error
= lock_page_killable(page
);
2145 if (unlikely(error
))
2146 goto readpage_error
;
2148 page_not_up_to_date_locked
:
2149 /* Did it get truncated before we got the lock? */
2150 if (!page
->mapping
) {
2156 /* Did somebody else fill it already? */
2157 if (PageUptodate(page
)) {
2164 * A previous I/O error may have been due to temporary
2165 * failures, eg. multipath errors.
2166 * PG_error will be set again if readpage fails.
2168 ClearPageError(page
);
2169 /* Start the actual read. The read will unlock the page. */
2170 error
= mapping
->a_ops
->readpage(filp
, page
);
2172 if (unlikely(error
)) {
2173 if (error
== AOP_TRUNCATED_PAGE
) {
2178 goto readpage_error
;
2181 if (!PageUptodate(page
)) {
2182 error
= lock_page_killable(page
);
2183 if (unlikely(error
))
2184 goto readpage_error
;
2185 if (!PageUptodate(page
)) {
2186 if (page
->mapping
== NULL
) {
2188 * invalidate_mapping_pages got it
2195 shrink_readahead_size_eio(ra
);
2197 goto readpage_error
;
2205 /* UHHUH! A synchronous read error occurred. Report it */
2211 * Ok, it wasn't cached, so we need to create a new
2214 page
= page_cache_alloc(mapping
);
2219 error
= add_to_page_cache_lru(page
, mapping
, index
,
2220 mapping_gfp_constraint(mapping
, GFP_KERNEL
));
2223 if (error
== -EEXIST
) {
2235 ra
->prev_pos
= prev_index
;
2236 ra
->prev_pos
<<= PAGE_SHIFT
;
2237 ra
->prev_pos
|= prev_offset
;
2239 *ppos
= ((loff_t
)index
<< PAGE_SHIFT
) + offset
;
2240 file_accessed(filp
);
2241 return written
? written
: error
;
2243 EXPORT_SYMBOL_GPL(generic_file_buffered_read
);
2246 * generic_file_read_iter - generic filesystem read routine
2247 * @iocb: kernel I/O control block
2248 * @iter: destination for the data read
2250 * This is the "read_iter()" routine for all filesystems
2251 * that can use the page cache directly.
2253 * * number of bytes copied, even for partial reads
2254 * * negative error code if nothing was read
2257 generic_file_read_iter(struct kiocb
*iocb
, struct iov_iter
*iter
)
2259 size_t count
= iov_iter_count(iter
);
2263 goto out
; /* skip atime */
2265 if (iocb
->ki_flags
& IOCB_DIRECT
) {
2266 struct file
*file
= iocb
->ki_filp
;
2267 struct address_space
*mapping
= file
->f_mapping
;
2268 struct inode
*inode
= mapping
->host
;
2271 size
= i_size_read(inode
);
2272 if (iocb
->ki_flags
& IOCB_NOWAIT
) {
2273 if (filemap_range_has_page(mapping
, iocb
->ki_pos
,
2274 iocb
->ki_pos
+ count
- 1))
2277 retval
= filemap_write_and_wait_range(mapping
,
2279 iocb
->ki_pos
+ count
- 1);
2284 file_accessed(file
);
2286 retval
= mapping
->a_ops
->direct_IO(iocb
, iter
);
2288 iocb
->ki_pos
+= retval
;
2291 iov_iter_revert(iter
, count
- iov_iter_count(iter
));
2294 * Btrfs can have a short DIO read if we encounter
2295 * compressed extents, so if there was an error, or if
2296 * we've already read everything we wanted to, or if
2297 * there was a short read because we hit EOF, go ahead
2298 * and return. Otherwise fallthrough to buffered io for
2299 * the rest of the read. Buffered reads will not work for
2300 * DAX files, so don't bother trying.
2302 if (retval
< 0 || !count
|| iocb
->ki_pos
>= size
||
2307 retval
= generic_file_buffered_read(iocb
, iter
, retval
);
2311 EXPORT_SYMBOL(generic_file_read_iter
);
2314 #define MMAP_LOTSAMISS (100)
2316 * lock_page_maybe_drop_mmap - lock the page, possibly dropping the mmap_lock
2317 * @vmf - the vm_fault for this fault.
2318 * @page - the page to lock.
2319 * @fpin - the pointer to the file we may pin (or is already pinned).
2321 * This works similar to lock_page_or_retry in that it can drop the mmap_lock.
2322 * It differs in that it actually returns the page locked if it returns 1 and 0
2323 * if it couldn't lock the page. If we did have to drop the mmap_lock then fpin
2324 * will point to the pinned file and needs to be fput()'ed at a later point.
2326 static int lock_page_maybe_drop_mmap(struct vm_fault
*vmf
, struct page
*page
,
2329 if (trylock_page(page
))
2333 * NOTE! This will make us return with VM_FAULT_RETRY, but with
2334 * the mmap_lock still held. That's how FAULT_FLAG_RETRY_NOWAIT
2335 * is supposed to work. We have way too many special cases..
2337 if (vmf
->flags
& FAULT_FLAG_RETRY_NOWAIT
)
2340 *fpin
= maybe_unlock_mmap_for_io(vmf
, *fpin
);
2341 if (vmf
->flags
& FAULT_FLAG_KILLABLE
) {
2342 if (__lock_page_killable(page
)) {
2344 * We didn't have the right flags to drop the mmap_lock,
2345 * but all fault_handlers only check for fatal signals
2346 * if we return VM_FAULT_RETRY, so we need to drop the
2347 * mmap_lock here and return 0 if we don't have a fpin.
2350 mmap_read_unlock(vmf
->vma
->vm_mm
);
2360 * Synchronous readahead happens when we don't even find a page in the page
2361 * cache at all. We don't want to perform IO under the mmap sem, so if we have
2362 * to drop the mmap sem we return the file that was pinned in order for us to do
2363 * that. If we didn't pin a file then we return NULL. The file that is
2364 * returned needs to be fput()'ed when we're done with it.
2366 static struct file
*do_sync_mmap_readahead(struct vm_fault
*vmf
)
2368 struct file
*file
= vmf
->vma
->vm_file
;
2369 struct file_ra_state
*ra
= &file
->f_ra
;
2370 struct address_space
*mapping
= file
->f_mapping
;
2371 struct file
*fpin
= NULL
;
2372 pgoff_t offset
= vmf
->pgoff
;
2374 /* If we don't want any read-ahead, don't bother */
2375 if (vmf
->vma
->vm_flags
& VM_RAND_READ
)
2380 if (vmf
->vma
->vm_flags
& VM_SEQ_READ
) {
2381 fpin
= maybe_unlock_mmap_for_io(vmf
, fpin
);
2382 page_cache_sync_readahead(mapping
, ra
, file
, offset
,
2387 /* Avoid banging the cache line if not needed */
2388 if (ra
->mmap_miss
< MMAP_LOTSAMISS
* 10)
2392 * Do we miss much more than hit in this file? If so,
2393 * stop bothering with read-ahead. It will only hurt.
2395 if (ra
->mmap_miss
> MMAP_LOTSAMISS
)
2401 fpin
= maybe_unlock_mmap_for_io(vmf
, fpin
);
2402 ra
->start
= max_t(long, 0, offset
- ra
->ra_pages
/ 2);
2403 ra
->size
= ra
->ra_pages
;
2404 ra
->async_size
= ra
->ra_pages
/ 4;
2405 ra_submit(ra
, mapping
, file
);
2410 * Asynchronous readahead happens when we find the page and PG_readahead,
2411 * so we want to possibly extend the readahead further. We return the file that
2412 * was pinned if we have to drop the mmap_lock in order to do IO.
2414 static struct file
*do_async_mmap_readahead(struct vm_fault
*vmf
,
2417 struct file
*file
= vmf
->vma
->vm_file
;
2418 struct file_ra_state
*ra
= &file
->f_ra
;
2419 struct address_space
*mapping
= file
->f_mapping
;
2420 struct file
*fpin
= NULL
;
2421 pgoff_t offset
= vmf
->pgoff
;
2423 /* If we don't want any read-ahead, don't bother */
2424 if (vmf
->vma
->vm_flags
& VM_RAND_READ
|| !ra
->ra_pages
)
2426 if (ra
->mmap_miss
> 0)
2428 if (PageReadahead(page
)) {
2429 fpin
= maybe_unlock_mmap_for_io(vmf
, fpin
);
2430 page_cache_async_readahead(mapping
, ra
, file
,
2431 page
, offset
, ra
->ra_pages
);
2437 * filemap_fault - read in file data for page fault handling
2438 * @vmf: struct vm_fault containing details of the fault
2440 * filemap_fault() is invoked via the vma operations vector for a
2441 * mapped memory region to read in file data during a page fault.
2443 * The goto's are kind of ugly, but this streamlines the normal case of having
2444 * it in the page cache, and handles the special cases reasonably without
2445 * having a lot of duplicated code.
2447 * vma->vm_mm->mmap_lock must be held on entry.
2449 * If our return value has VM_FAULT_RETRY set, it's because the mmap_lock
2450 * may be dropped before doing I/O or by lock_page_maybe_drop_mmap().
2452 * If our return value does not have VM_FAULT_RETRY set, the mmap_lock
2453 * has not been released.
2455 * We never return with VM_FAULT_RETRY and a bit from VM_FAULT_ERROR set.
2457 * Return: bitwise-OR of %VM_FAULT_ codes.
2459 vm_fault_t
filemap_fault(struct vm_fault
*vmf
)
2462 struct file
*file
= vmf
->vma
->vm_file
;
2463 struct file
*fpin
= NULL
;
2464 struct address_space
*mapping
= file
->f_mapping
;
2465 struct file_ra_state
*ra
= &file
->f_ra
;
2466 struct inode
*inode
= mapping
->host
;
2467 pgoff_t offset
= vmf
->pgoff
;
2472 max_off
= DIV_ROUND_UP(i_size_read(inode
), PAGE_SIZE
);
2473 if (unlikely(offset
>= max_off
))
2474 return VM_FAULT_SIGBUS
;
2477 * Do we have something in the page cache already?
2479 page
= find_get_page(mapping
, offset
);
2480 if (likely(page
) && !(vmf
->flags
& FAULT_FLAG_TRIED
)) {
2482 * We found the page, so try async readahead before
2483 * waiting for the lock.
2485 fpin
= do_async_mmap_readahead(vmf
, page
);
2487 /* No page in the page cache at all */
2488 count_vm_event(PGMAJFAULT
);
2489 count_memcg_event_mm(vmf
->vma
->vm_mm
, PGMAJFAULT
);
2490 ret
= VM_FAULT_MAJOR
;
2491 fpin
= do_sync_mmap_readahead(vmf
);
2493 page
= pagecache_get_page(mapping
, offset
,
2494 FGP_CREAT
|FGP_FOR_MMAP
,
2499 return VM_FAULT_OOM
;
2503 if (!lock_page_maybe_drop_mmap(vmf
, page
, &fpin
))
2506 /* Did it get truncated? */
2507 if (unlikely(compound_head(page
)->mapping
!= mapping
)) {
2512 VM_BUG_ON_PAGE(page_to_pgoff(page
) != offset
, page
);
2515 * We have a locked page in the page cache, now we need to check
2516 * that it's up-to-date. If not, it is going to be due to an error.
2518 if (unlikely(!PageUptodate(page
)))
2519 goto page_not_uptodate
;
2522 * We've made it this far and we had to drop our mmap_lock, now is the
2523 * time to return to the upper layer and have it re-find the vma and
2532 * Found the page and have a reference on it.
2533 * We must recheck i_size under page lock.
2535 max_off
= DIV_ROUND_UP(i_size_read(inode
), PAGE_SIZE
);
2536 if (unlikely(offset
>= max_off
)) {
2539 return VM_FAULT_SIGBUS
;
2543 return ret
| VM_FAULT_LOCKED
;
2547 * Umm, take care of errors if the page isn't up-to-date.
2548 * Try to re-read it _once_. We do this synchronously,
2549 * because there really aren't any performance issues here
2550 * and we need to check for errors.
2552 ClearPageError(page
);
2553 fpin
= maybe_unlock_mmap_for_io(vmf
, fpin
);
2554 error
= mapping
->a_ops
->readpage(file
, page
);
2556 wait_on_page_locked(page
);
2557 if (!PageUptodate(page
))
2564 if (!error
|| error
== AOP_TRUNCATED_PAGE
)
2567 shrink_readahead_size_eio(ra
);
2568 return VM_FAULT_SIGBUS
;
2572 * We dropped the mmap_lock, we need to return to the fault handler to
2573 * re-find the vma and come back and find our hopefully still populated
2580 return ret
| VM_FAULT_RETRY
;
2582 EXPORT_SYMBOL(filemap_fault
);
2584 void filemap_map_pages(struct vm_fault
*vmf
,
2585 pgoff_t start_pgoff
, pgoff_t end_pgoff
)
2587 struct file
*file
= vmf
->vma
->vm_file
;
2588 struct address_space
*mapping
= file
->f_mapping
;
2589 pgoff_t last_pgoff
= start_pgoff
;
2590 unsigned long max_idx
;
2591 XA_STATE(xas
, &mapping
->i_pages
, start_pgoff
);
2595 xas_for_each(&xas
, page
, end_pgoff
) {
2596 if (xas_retry(&xas
, page
))
2598 if (xa_is_value(page
))
2602 * Check for a locked page first, as a speculative
2603 * reference may adversely influence page migration.
2605 if (PageLocked(page
))
2607 if (!page_cache_get_speculative(page
))
2610 /* Has the page moved or been split? */
2611 if (unlikely(page
!= xas_reload(&xas
)))
2613 page
= find_subpage(page
, xas
.xa_index
);
2615 if (!PageUptodate(page
) ||
2616 PageReadahead(page
) ||
2619 if (!trylock_page(page
))
2622 if (page
->mapping
!= mapping
|| !PageUptodate(page
))
2625 max_idx
= DIV_ROUND_UP(i_size_read(mapping
->host
), PAGE_SIZE
);
2626 if (page
->index
>= max_idx
)
2629 if (file
->f_ra
.mmap_miss
> 0)
2630 file
->f_ra
.mmap_miss
--;
2632 vmf
->address
+= (xas
.xa_index
- last_pgoff
) << PAGE_SHIFT
;
2634 vmf
->pte
+= xas
.xa_index
- last_pgoff
;
2635 last_pgoff
= xas
.xa_index
;
2636 if (alloc_set_pte(vmf
, page
))
2645 /* Huge page is mapped? No need to proceed. */
2646 if (pmd_trans_huge(*vmf
->pmd
))
2651 EXPORT_SYMBOL(filemap_map_pages
);
2653 vm_fault_t
filemap_page_mkwrite(struct vm_fault
*vmf
)
2655 struct page
*page
= vmf
->page
;
2656 struct inode
*inode
= file_inode(vmf
->vma
->vm_file
);
2657 vm_fault_t ret
= VM_FAULT_LOCKED
;
2659 sb_start_pagefault(inode
->i_sb
);
2660 file_update_time(vmf
->vma
->vm_file
);
2662 if (page
->mapping
!= inode
->i_mapping
) {
2664 ret
= VM_FAULT_NOPAGE
;
2668 * We mark the page dirty already here so that when freeze is in
2669 * progress, we are guaranteed that writeback during freezing will
2670 * see the dirty page and writeprotect it again.
2672 set_page_dirty(page
);
2673 wait_for_stable_page(page
);
2675 sb_end_pagefault(inode
->i_sb
);
2679 const struct vm_operations_struct generic_file_vm_ops
= {
2680 .fault
= filemap_fault
,
2681 .map_pages
= filemap_map_pages
,
2682 .page_mkwrite
= filemap_page_mkwrite
,
2685 /* This is used for a general mmap of a disk file */
2687 int generic_file_mmap(struct file
* file
, struct vm_area_struct
* vma
)
2689 struct address_space
*mapping
= file
->f_mapping
;
2691 if (!mapping
->a_ops
->readpage
)
2693 file_accessed(file
);
2694 vma
->vm_ops
= &generic_file_vm_ops
;
2699 * This is for filesystems which do not implement ->writepage.
2701 int generic_file_readonly_mmap(struct file
*file
, struct vm_area_struct
*vma
)
2703 if ((vma
->vm_flags
& VM_SHARED
) && (vma
->vm_flags
& VM_MAYWRITE
))
2705 return generic_file_mmap(file
, vma
);
2708 vm_fault_t
filemap_page_mkwrite(struct vm_fault
*vmf
)
2710 return VM_FAULT_SIGBUS
;
2712 int generic_file_mmap(struct file
* file
, struct vm_area_struct
* vma
)
2716 int generic_file_readonly_mmap(struct file
* file
, struct vm_area_struct
* vma
)
2720 #endif /* CONFIG_MMU */
2722 EXPORT_SYMBOL(filemap_page_mkwrite
);
2723 EXPORT_SYMBOL(generic_file_mmap
);
2724 EXPORT_SYMBOL(generic_file_readonly_mmap
);
2726 static struct page
*wait_on_page_read(struct page
*page
)
2728 if (!IS_ERR(page
)) {
2729 wait_on_page_locked(page
);
2730 if (!PageUptodate(page
)) {
2732 page
= ERR_PTR(-EIO
);
2738 static struct page
*do_read_cache_page(struct address_space
*mapping
,
2740 int (*filler
)(void *, struct page
*),
2747 page
= find_get_page(mapping
, index
);
2749 page
= __page_cache_alloc(gfp
);
2751 return ERR_PTR(-ENOMEM
);
2752 err
= add_to_page_cache_lru(page
, mapping
, index
, gfp
);
2753 if (unlikely(err
)) {
2757 /* Presumably ENOMEM for xarray node */
2758 return ERR_PTR(err
);
2763 err
= filler(data
, page
);
2765 err
= mapping
->a_ops
->readpage(data
, page
);
2769 return ERR_PTR(err
);
2772 page
= wait_on_page_read(page
);
2777 if (PageUptodate(page
))
2781 * Page is not up to date and may be locked due one of the following
2782 * case a: Page is being filled and the page lock is held
2783 * case b: Read/write error clearing the page uptodate status
2784 * case c: Truncation in progress (page locked)
2785 * case d: Reclaim in progress
2787 * Case a, the page will be up to date when the page is unlocked.
2788 * There is no need to serialise on the page lock here as the page
2789 * is pinned so the lock gives no additional protection. Even if the
2790 * the page is truncated, the data is still valid if PageUptodate as
2791 * it's a race vs truncate race.
2792 * Case b, the page will not be up to date
2793 * Case c, the page may be truncated but in itself, the data may still
2794 * be valid after IO completes as it's a read vs truncate race. The
2795 * operation must restart if the page is not uptodate on unlock but
2796 * otherwise serialising on page lock to stabilise the mapping gives
2797 * no additional guarantees to the caller as the page lock is
2798 * released before return.
2799 * Case d, similar to truncation. If reclaim holds the page lock, it
2800 * will be a race with remove_mapping that determines if the mapping
2801 * is valid on unlock but otherwise the data is valid and there is
2802 * no need to serialise with page lock.
2804 * As the page lock gives no additional guarantee, we optimistically
2805 * wait on the page to be unlocked and check if it's up to date and
2806 * use the page if it is. Otherwise, the page lock is required to
2807 * distinguish between the different cases. The motivation is that we
2808 * avoid spurious serialisations and wakeups when multiple processes
2809 * wait on the same page for IO to complete.
2811 wait_on_page_locked(page
);
2812 if (PageUptodate(page
))
2815 /* Distinguish between all the cases under the safety of the lock */
2818 /* Case c or d, restart the operation */
2819 if (!page
->mapping
) {
2825 /* Someone else locked and filled the page in a very small window */
2826 if (PageUptodate(page
)) {
2832 * A previous I/O error may have been due to temporary
2834 * Clear page error before actual read, PG_error will be
2835 * set again if read page fails.
2837 ClearPageError(page
);
2841 mark_page_accessed(page
);
2846 * read_cache_page - read into page cache, fill it if needed
2847 * @mapping: the page's address_space
2848 * @index: the page index
2849 * @filler: function to perform the read
2850 * @data: first arg to filler(data, page) function, often left as NULL
2852 * Read into the page cache. If a page already exists, and PageUptodate() is
2853 * not set, try to fill the page and wait for it to become unlocked.
2855 * If the page does not get brought uptodate, return -EIO.
2857 * Return: up to date page on success, ERR_PTR() on failure.
2859 struct page
*read_cache_page(struct address_space
*mapping
,
2861 int (*filler
)(void *, struct page
*),
2864 return do_read_cache_page(mapping
, index
, filler
, data
,
2865 mapping_gfp_mask(mapping
));
2867 EXPORT_SYMBOL(read_cache_page
);
2870 * read_cache_page_gfp - read into page cache, using specified page allocation flags.
2871 * @mapping: the page's address_space
2872 * @index: the page index
2873 * @gfp: the page allocator flags to use if allocating
2875 * This is the same as "read_mapping_page(mapping, index, NULL)", but with
2876 * any new page allocations done using the specified allocation flags.
2878 * If the page does not get brought uptodate, return -EIO.
2880 * Return: up to date page on success, ERR_PTR() on failure.
2882 struct page
*read_cache_page_gfp(struct address_space
*mapping
,
2886 return do_read_cache_page(mapping
, index
, NULL
, NULL
, gfp
);
2888 EXPORT_SYMBOL(read_cache_page_gfp
);
2891 * Don't operate on ranges the page cache doesn't support, and don't exceed the
2892 * LFS limits. If pos is under the limit it becomes a short access. If it
2893 * exceeds the limit we return -EFBIG.
2895 static int generic_write_check_limits(struct file
*file
, loff_t pos
,
2898 struct inode
*inode
= file
->f_mapping
->host
;
2899 loff_t max_size
= inode
->i_sb
->s_maxbytes
;
2900 loff_t limit
= rlimit(RLIMIT_FSIZE
);
2902 if (limit
!= RLIM_INFINITY
) {
2904 send_sig(SIGXFSZ
, current
, 0);
2907 *count
= min(*count
, limit
- pos
);
2910 if (!(file
->f_flags
& O_LARGEFILE
))
2911 max_size
= MAX_NON_LFS
;
2913 if (unlikely(pos
>= max_size
))
2916 *count
= min(*count
, max_size
- pos
);
2922 * Performs necessary checks before doing a write
2924 * Can adjust writing position or amount of bytes to write.
2925 * Returns appropriate error code that caller should return or
2926 * zero in case that write should be allowed.
2928 inline ssize_t
generic_write_checks(struct kiocb
*iocb
, struct iov_iter
*from
)
2930 struct file
*file
= iocb
->ki_filp
;
2931 struct inode
*inode
= file
->f_mapping
->host
;
2935 if (IS_SWAPFILE(inode
))
2938 if (!iov_iter_count(from
))
2941 /* FIXME: this is for backwards compatibility with 2.4 */
2942 if (iocb
->ki_flags
& IOCB_APPEND
)
2943 iocb
->ki_pos
= i_size_read(inode
);
2945 if ((iocb
->ki_flags
& IOCB_NOWAIT
) && !(iocb
->ki_flags
& IOCB_DIRECT
))
2948 count
= iov_iter_count(from
);
2949 ret
= generic_write_check_limits(file
, iocb
->ki_pos
, &count
);
2953 iov_iter_truncate(from
, count
);
2954 return iov_iter_count(from
);
2956 EXPORT_SYMBOL(generic_write_checks
);
2959 * Performs necessary checks before doing a clone.
2961 * Can adjust amount of bytes to clone via @req_count argument.
2962 * Returns appropriate error code that caller should return or
2963 * zero in case the clone should be allowed.
2965 int generic_remap_checks(struct file
*file_in
, loff_t pos_in
,
2966 struct file
*file_out
, loff_t pos_out
,
2967 loff_t
*req_count
, unsigned int remap_flags
)
2969 struct inode
*inode_in
= file_in
->f_mapping
->host
;
2970 struct inode
*inode_out
= file_out
->f_mapping
->host
;
2971 uint64_t count
= *req_count
;
2973 loff_t size_in
, size_out
;
2974 loff_t bs
= inode_out
->i_sb
->s_blocksize
;
2977 /* The start of both ranges must be aligned to an fs block. */
2978 if (!IS_ALIGNED(pos_in
, bs
) || !IS_ALIGNED(pos_out
, bs
))
2981 /* Ensure offsets don't wrap. */
2982 if (pos_in
+ count
< pos_in
|| pos_out
+ count
< pos_out
)
2985 size_in
= i_size_read(inode_in
);
2986 size_out
= i_size_read(inode_out
);
2988 /* Dedupe requires both ranges to be within EOF. */
2989 if ((remap_flags
& REMAP_FILE_DEDUP
) &&
2990 (pos_in
>= size_in
|| pos_in
+ count
> size_in
||
2991 pos_out
>= size_out
|| pos_out
+ count
> size_out
))
2994 /* Ensure the infile range is within the infile. */
2995 if (pos_in
>= size_in
)
2997 count
= min(count
, size_in
- (uint64_t)pos_in
);
2999 ret
= generic_write_check_limits(file_out
, pos_out
, &count
);
3004 * If the user wanted us to link to the infile's EOF, round up to the
3005 * next block boundary for this check.
3007 * Otherwise, make sure the count is also block-aligned, having
3008 * already confirmed the starting offsets' block alignment.
3010 if (pos_in
+ count
== size_in
) {
3011 bcount
= ALIGN(size_in
, bs
) - pos_in
;
3013 if (!IS_ALIGNED(count
, bs
))
3014 count
= ALIGN_DOWN(count
, bs
);
3018 /* Don't allow overlapped cloning within the same file. */
3019 if (inode_in
== inode_out
&&
3020 pos_out
+ bcount
> pos_in
&&
3021 pos_out
< pos_in
+ bcount
)
3025 * We shortened the request but the caller can't deal with that, so
3026 * bounce the request back to userspace.
3028 if (*req_count
!= count
&& !(remap_flags
& REMAP_FILE_CAN_SHORTEN
))
3037 * Performs common checks before doing a file copy/clone
3038 * from @file_in to @file_out.
3040 int generic_file_rw_checks(struct file
*file_in
, struct file
*file_out
)
3042 struct inode
*inode_in
= file_inode(file_in
);
3043 struct inode
*inode_out
= file_inode(file_out
);
3045 /* Don't copy dirs, pipes, sockets... */
3046 if (S_ISDIR(inode_in
->i_mode
) || S_ISDIR(inode_out
->i_mode
))
3048 if (!S_ISREG(inode_in
->i_mode
) || !S_ISREG(inode_out
->i_mode
))
3051 if (!(file_in
->f_mode
& FMODE_READ
) ||
3052 !(file_out
->f_mode
& FMODE_WRITE
) ||
3053 (file_out
->f_flags
& O_APPEND
))
3060 * Performs necessary checks before doing a file copy
3062 * Can adjust amount of bytes to copy via @req_count argument.
3063 * Returns appropriate error code that caller should return or
3064 * zero in case the copy should be allowed.
3066 int generic_copy_file_checks(struct file
*file_in
, loff_t pos_in
,
3067 struct file
*file_out
, loff_t pos_out
,
3068 size_t *req_count
, unsigned int flags
)
3070 struct inode
*inode_in
= file_inode(file_in
);
3071 struct inode
*inode_out
= file_inode(file_out
);
3072 uint64_t count
= *req_count
;
3076 ret
= generic_file_rw_checks(file_in
, file_out
);
3080 /* Don't touch certain kinds of inodes */
3081 if (IS_IMMUTABLE(inode_out
))
3084 if (IS_SWAPFILE(inode_in
) || IS_SWAPFILE(inode_out
))
3087 /* Ensure offsets don't wrap. */
3088 if (pos_in
+ count
< pos_in
|| pos_out
+ count
< pos_out
)
3091 /* Shorten the copy to EOF */
3092 size_in
= i_size_read(inode_in
);
3093 if (pos_in
>= size_in
)
3096 count
= min(count
, size_in
- (uint64_t)pos_in
);
3098 ret
= generic_write_check_limits(file_out
, pos_out
, &count
);
3102 /* Don't allow overlapped copying within the same file. */
3103 if (inode_in
== inode_out
&&
3104 pos_out
+ count
> pos_in
&&
3105 pos_out
< pos_in
+ count
)
3112 int pagecache_write_begin(struct file
*file
, struct address_space
*mapping
,
3113 loff_t pos
, unsigned len
, unsigned flags
,
3114 struct page
**pagep
, void **fsdata
)
3116 const struct address_space_operations
*aops
= mapping
->a_ops
;
3118 return aops
->write_begin(file
, mapping
, pos
, len
, flags
,
3121 EXPORT_SYMBOL(pagecache_write_begin
);
3123 int pagecache_write_end(struct file
*file
, struct address_space
*mapping
,
3124 loff_t pos
, unsigned len
, unsigned copied
,
3125 struct page
*page
, void *fsdata
)
3127 const struct address_space_operations
*aops
= mapping
->a_ops
;
3129 return aops
->write_end(file
, mapping
, pos
, len
, copied
, page
, fsdata
);
3131 EXPORT_SYMBOL(pagecache_write_end
);
3134 * Warn about a page cache invalidation failure during a direct I/O write.
3136 void dio_warn_stale_pagecache(struct file
*filp
)
3138 static DEFINE_RATELIMIT_STATE(_rs
, 86400 * HZ
, DEFAULT_RATELIMIT_BURST
);
3140 struct inode
*inode
= file_inode(filp
);
3143 errseq_set(&inode
->i_mapping
->wb_err
, -EIO
);
3144 if (__ratelimit(&_rs
)) {
3145 path
= file_path(filp
, pathname
, sizeof(pathname
));
3148 pr_crit("Page cache invalidation failure on direct I/O. Possible data corruption due to collision with buffered I/O!\n");
3149 pr_crit("File: %s PID: %d Comm: %.20s\n", path
, current
->pid
,
3155 generic_file_direct_write(struct kiocb
*iocb
, struct iov_iter
*from
)
3157 struct file
*file
= iocb
->ki_filp
;
3158 struct address_space
*mapping
= file
->f_mapping
;
3159 struct inode
*inode
= mapping
->host
;
3160 loff_t pos
= iocb
->ki_pos
;
3165 write_len
= iov_iter_count(from
);
3166 end
= (pos
+ write_len
- 1) >> PAGE_SHIFT
;
3168 if (iocb
->ki_flags
& IOCB_NOWAIT
) {
3169 /* If there are pages to writeback, return */
3170 if (filemap_range_has_page(inode
->i_mapping
, pos
,
3171 pos
+ write_len
- 1))
3174 written
= filemap_write_and_wait_range(mapping
, pos
,
3175 pos
+ write_len
- 1);
3181 * After a write we want buffered reads to be sure to go to disk to get
3182 * the new data. We invalidate clean cached page from the region we're
3183 * about to write. We do this *before* the write so that we can return
3184 * without clobbering -EIOCBQUEUED from ->direct_IO().
3186 written
= invalidate_inode_pages2_range(mapping
,
3187 pos
>> PAGE_SHIFT
, end
);
3189 * If a page can not be invalidated, return 0 to fall back
3190 * to buffered write.
3193 if (written
== -EBUSY
)
3198 written
= mapping
->a_ops
->direct_IO(iocb
, from
);
3201 * Finally, try again to invalidate clean pages which might have been
3202 * cached by non-direct readahead, or faulted in by get_user_pages()
3203 * if the source of the write was an mmap'ed region of the file
3204 * we're writing. Either one is a pretty crazy thing to do,
3205 * so we don't support it 100%. If this invalidation
3206 * fails, tough, the write still worked...
3208 * Most of the time we do not need this since dio_complete() will do
3209 * the invalidation for us. However there are some file systems that
3210 * do not end up with dio_complete() being called, so let's not break
3211 * them by removing it completely.
3213 * Noticeable example is a blkdev_direct_IO().
3215 * Skip invalidation for async writes or if mapping has no pages.
3217 if (written
> 0 && mapping
->nrpages
&&
3218 invalidate_inode_pages2_range(mapping
, pos
>> PAGE_SHIFT
, end
))
3219 dio_warn_stale_pagecache(file
);
3223 write_len
-= written
;
3224 if (pos
> i_size_read(inode
) && !S_ISBLK(inode
->i_mode
)) {
3225 i_size_write(inode
, pos
);
3226 mark_inode_dirty(inode
);
3230 iov_iter_revert(from
, write_len
- iov_iter_count(from
));
3234 EXPORT_SYMBOL(generic_file_direct_write
);
3237 * Find or create a page at the given pagecache position. Return the locked
3238 * page. This function is specifically for buffered writes.
3240 struct page
*grab_cache_page_write_begin(struct address_space
*mapping
,
3241 pgoff_t index
, unsigned flags
)
3244 int fgp_flags
= FGP_LOCK
|FGP_WRITE
|FGP_CREAT
;
3246 if (flags
& AOP_FLAG_NOFS
)
3247 fgp_flags
|= FGP_NOFS
;
3249 page
= pagecache_get_page(mapping
, index
, fgp_flags
,
3250 mapping_gfp_mask(mapping
));
3252 wait_for_stable_page(page
);
3256 EXPORT_SYMBOL(grab_cache_page_write_begin
);
3258 ssize_t
generic_perform_write(struct file
*file
,
3259 struct iov_iter
*i
, loff_t pos
)
3261 struct address_space
*mapping
= file
->f_mapping
;
3262 const struct address_space_operations
*a_ops
= mapping
->a_ops
;
3264 ssize_t written
= 0;
3265 unsigned int flags
= 0;
3269 unsigned long offset
; /* Offset into pagecache page */
3270 unsigned long bytes
; /* Bytes to write to page */
3271 size_t copied
; /* Bytes copied from user */
3274 offset
= (pos
& (PAGE_SIZE
- 1));
3275 bytes
= min_t(unsigned long, PAGE_SIZE
- offset
,
3280 * Bring in the user page that we will copy from _first_.
3281 * Otherwise there's a nasty deadlock on copying from the
3282 * same page as we're writing to, without it being marked
3285 * Not only is this an optimisation, but it is also required
3286 * to check that the address is actually valid, when atomic
3287 * usercopies are used, below.
3289 if (unlikely(iov_iter_fault_in_readable(i
, bytes
))) {
3294 if (fatal_signal_pending(current
)) {
3299 status
= a_ops
->write_begin(file
, mapping
, pos
, bytes
, flags
,
3301 if (unlikely(status
< 0))
3304 if (mapping_writably_mapped(mapping
))
3305 flush_dcache_page(page
);
3307 copied
= iov_iter_copy_from_user_atomic(page
, i
, offset
, bytes
);
3308 flush_dcache_page(page
);
3310 status
= a_ops
->write_end(file
, mapping
, pos
, bytes
, copied
,
3312 if (unlikely(status
< 0))
3318 iov_iter_advance(i
, copied
);
3319 if (unlikely(copied
== 0)) {
3321 * If we were unable to copy any data at all, we must
3322 * fall back to a single segment length write.
3324 * If we didn't fallback here, we could livelock
3325 * because not all segments in the iov can be copied at
3326 * once without a pagefault.
3328 bytes
= min_t(unsigned long, PAGE_SIZE
- offset
,
3329 iov_iter_single_seg_count(i
));
3335 balance_dirty_pages_ratelimited(mapping
);
3336 } while (iov_iter_count(i
));
3338 return written
? written
: status
;
3340 EXPORT_SYMBOL(generic_perform_write
);
3343 * __generic_file_write_iter - write data to a file
3344 * @iocb: IO state structure (file, offset, etc.)
3345 * @from: iov_iter with data to write
3347 * This function does all the work needed for actually writing data to a
3348 * file. It does all basic checks, removes SUID from the file, updates
3349 * modification times and calls proper subroutines depending on whether we
3350 * do direct IO or a standard buffered write.
3352 * It expects i_mutex to be grabbed unless we work on a block device or similar
3353 * object which does not need locking at all.
3355 * This function does *not* take care of syncing data in case of O_SYNC write.
3356 * A caller has to handle it. This is mainly due to the fact that we want to
3357 * avoid syncing under i_mutex.
3360 * * number of bytes written, even for truncated writes
3361 * * negative error code if no data has been written at all
3363 ssize_t
__generic_file_write_iter(struct kiocb
*iocb
, struct iov_iter
*from
)
3365 struct file
*file
= iocb
->ki_filp
;
3366 struct address_space
* mapping
= file
->f_mapping
;
3367 struct inode
*inode
= mapping
->host
;
3368 ssize_t written
= 0;
3372 /* We can write back this queue in page reclaim */
3373 current
->backing_dev_info
= inode_to_bdi(inode
);
3374 err
= file_remove_privs(file
);
3378 err
= file_update_time(file
);
3382 if (iocb
->ki_flags
& IOCB_DIRECT
) {
3383 loff_t pos
, endbyte
;
3385 written
= generic_file_direct_write(iocb
, from
);
3387 * If the write stopped short of completing, fall back to
3388 * buffered writes. Some filesystems do this for writes to
3389 * holes, for example. For DAX files, a buffered write will
3390 * not succeed (even if it did, DAX does not handle dirty
3391 * page-cache pages correctly).
3393 if (written
< 0 || !iov_iter_count(from
) || IS_DAX(inode
))
3396 status
= generic_perform_write(file
, from
, pos
= iocb
->ki_pos
);
3398 * If generic_perform_write() returned a synchronous error
3399 * then we want to return the number of bytes which were
3400 * direct-written, or the error code if that was zero. Note
3401 * that this differs from normal direct-io semantics, which
3402 * will return -EFOO even if some bytes were written.
3404 if (unlikely(status
< 0)) {
3409 * We need to ensure that the page cache pages are written to
3410 * disk and invalidated to preserve the expected O_DIRECT
3413 endbyte
= pos
+ status
- 1;
3414 err
= filemap_write_and_wait_range(mapping
, pos
, endbyte
);
3416 iocb
->ki_pos
= endbyte
+ 1;
3418 invalidate_mapping_pages(mapping
,
3420 endbyte
>> PAGE_SHIFT
);
3423 * We don't know how much we wrote, so just return
3424 * the number of bytes which were direct-written
3428 written
= generic_perform_write(file
, from
, iocb
->ki_pos
);
3429 if (likely(written
> 0))
3430 iocb
->ki_pos
+= written
;
3433 current
->backing_dev_info
= NULL
;
3434 return written
? written
: err
;
3436 EXPORT_SYMBOL(__generic_file_write_iter
);
3439 * generic_file_write_iter - write data to a file
3440 * @iocb: IO state structure
3441 * @from: iov_iter with data to write
3443 * This is a wrapper around __generic_file_write_iter() to be used by most
3444 * filesystems. It takes care of syncing the file in case of O_SYNC file
3445 * and acquires i_mutex as needed.
3447 * * negative error code if no data has been written at all of
3448 * vfs_fsync_range() failed for a synchronous write
3449 * * number of bytes written, even for truncated writes
3451 ssize_t
generic_file_write_iter(struct kiocb
*iocb
, struct iov_iter
*from
)
3453 struct file
*file
= iocb
->ki_filp
;
3454 struct inode
*inode
= file
->f_mapping
->host
;
3458 ret
= generic_write_checks(iocb
, from
);
3460 ret
= __generic_file_write_iter(iocb
, from
);
3461 inode_unlock(inode
);
3464 ret
= generic_write_sync(iocb
, ret
);
3467 EXPORT_SYMBOL(generic_file_write_iter
);
3470 * try_to_release_page() - release old fs-specific metadata on a page
3472 * @page: the page which the kernel is trying to free
3473 * @gfp_mask: memory allocation flags (and I/O mode)
3475 * The address_space is to try to release any data against the page
3476 * (presumably at page->private).
3478 * This may also be called if PG_fscache is set on a page, indicating that the
3479 * page is known to the local caching routines.
3481 * The @gfp_mask argument specifies whether I/O may be performed to release
3482 * this page (__GFP_IO), and whether the call may block (__GFP_RECLAIM & __GFP_FS).
3484 * Return: %1 if the release was successful, otherwise return zero.
3486 int try_to_release_page(struct page
*page
, gfp_t gfp_mask
)
3488 struct address_space
* const mapping
= page
->mapping
;
3490 BUG_ON(!PageLocked(page
));
3491 if (PageWriteback(page
))
3494 if (mapping
&& mapping
->a_ops
->releasepage
)
3495 return mapping
->a_ops
->releasepage(page
, gfp_mask
);
3496 return try_to_free_buffers(page
);
3499 EXPORT_SYMBOL(try_to_release_page
);