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>
44 #include <linux/page_idle.h>
47 #define CREATE_TRACE_POINTS
48 #include <trace/events/filemap.h>
51 * FIXME: remove all knowledge of the buffer layer from the core VM
53 #include <linux/buffer_head.h> /* for try_to_free_buffers */
58 * Shared mappings implemented 30.11.1994. It's not fully working yet,
61 * Shared mappings now work. 15.8.1995 Bruno.
63 * finished 'unifying' the page and buffer cache and SMP-threaded the
64 * page-cache, 21.05.1999, Ingo Molnar <mingo@redhat.com>
66 * SMP-threaded pagemap-LRU 1999, Andrea Arcangeli <andrea@suse.de>
72 * ->i_mmap_rwsem (truncate_pagecache)
73 * ->private_lock (__free_pte->__set_page_dirty_buffers)
74 * ->swap_lock (exclusive_swap_page, others)
78 * ->i_mmap_rwsem (truncate->unmap_mapping_range)
82 * ->page_table_lock or pte_lock (various, mainly in memory.c)
83 * ->i_pages lock (arch-dependent flush_dcache_mmap_lock)
86 * ->lock_page (access_process_vm)
88 * ->i_mutex (generic_perform_write)
89 * ->mmap_lock (fault_in_pages_readable->do_page_fault)
92 * sb_lock (fs/fs-writeback.c)
93 * ->i_pages lock (__sync_single_inode)
96 * ->anon_vma.lock (vma_adjust)
99 * ->page_table_lock or pte_lock (anon_vma_prepare and various)
101 * ->page_table_lock or pte_lock
102 * ->swap_lock (try_to_unmap_one)
103 * ->private_lock (try_to_unmap_one)
104 * ->i_pages lock (try_to_unmap_one)
105 * ->pgdat->lru_lock (follow_page->mark_page_accessed)
106 * ->pgdat->lru_lock (check_pte_range->isolate_lru_page)
107 * ->private_lock (page_remove_rmap->set_page_dirty)
108 * ->i_pages lock (page_remove_rmap->set_page_dirty)
109 * bdi.wb->list_lock (page_remove_rmap->set_page_dirty)
110 * ->inode->i_lock (page_remove_rmap->set_page_dirty)
111 * ->memcg->move_lock (page_remove_rmap->lock_page_memcg)
112 * bdi.wb->list_lock (zap_pte_range->set_page_dirty)
113 * ->inode->i_lock (zap_pte_range->set_page_dirty)
114 * ->private_lock (zap_pte_range->__set_page_dirty_buffers)
117 * ->tasklist_lock (memory_failure, collect_procs_ao)
120 static void page_cache_delete(struct address_space
*mapping
,
121 struct page
*page
, void *shadow
)
123 XA_STATE(xas
, &mapping
->i_pages
, page
->index
);
126 mapping_set_update(&xas
, mapping
);
128 /* hugetlb pages are represented by a single entry in the xarray */
129 if (!PageHuge(page
)) {
130 xas_set_order(&xas
, page
->index
, compound_order(page
));
131 nr
= compound_nr(page
);
134 VM_BUG_ON_PAGE(!PageLocked(page
), page
);
135 VM_BUG_ON_PAGE(PageTail(page
), page
);
136 VM_BUG_ON_PAGE(nr
!= 1 && shadow
, page
);
138 xas_store(&xas
, shadow
);
139 xas_init_marks(&xas
);
141 page
->mapping
= NULL
;
142 /* Leave page->index set: truncation lookup relies upon it */
145 mapping
->nrexceptional
+= nr
;
147 * Make sure the nrexceptional update is committed before
148 * the nrpages update so that final truncate racing
149 * with reclaim does not see both counters 0 at the
150 * same time and miss a shadow entry.
154 mapping
->nrpages
-= nr
;
157 static void unaccount_page_cache_page(struct address_space
*mapping
,
163 * if we're uptodate, flush out into the cleancache, otherwise
164 * invalidate any existing cleancache entries. We can't leave
165 * stale data around in the cleancache once our page is gone
167 if (PageUptodate(page
) && PageMappedToDisk(page
))
168 cleancache_put_page(page
);
170 cleancache_invalidate_page(mapping
, page
);
172 VM_BUG_ON_PAGE(PageTail(page
), page
);
173 VM_BUG_ON_PAGE(page_mapped(page
), page
);
174 if (!IS_ENABLED(CONFIG_DEBUG_VM
) && unlikely(page_mapped(page
))) {
177 pr_alert("BUG: Bad page cache in process %s pfn:%05lx\n",
178 current
->comm
, page_to_pfn(page
));
179 dump_page(page
, "still mapped when deleted");
181 add_taint(TAINT_BAD_PAGE
, LOCKDEP_NOW_UNRELIABLE
);
183 mapcount
= page_mapcount(page
);
184 if (mapping_exiting(mapping
) &&
185 page_count(page
) >= mapcount
+ 2) {
187 * All vmas have already been torn down, so it's
188 * a good bet that actually the page is unmapped,
189 * and we'd prefer not to leak it: if we're wrong,
190 * some other bad page check should catch it later.
192 page_mapcount_reset(page
);
193 page_ref_sub(page
, mapcount
);
197 /* hugetlb pages do not participate in page cache accounting. */
201 nr
= thp_nr_pages(page
);
203 __mod_lruvec_page_state(page
, NR_FILE_PAGES
, -nr
);
204 if (PageSwapBacked(page
)) {
205 __mod_lruvec_page_state(page
, NR_SHMEM
, -nr
);
206 if (PageTransHuge(page
))
207 __dec_node_page_state(page
, NR_SHMEM_THPS
);
208 } else if (PageTransHuge(page
)) {
209 __dec_node_page_state(page
, NR_FILE_THPS
);
210 filemap_nr_thps_dec(mapping
);
214 * At this point page must be either written or cleaned by
215 * truncate. Dirty page here signals a bug and loss of
218 * This fixes dirty accounting after removing the page entirely
219 * but leaves PageDirty set: it has no effect for truncated
220 * page and anyway will be cleared before returning page into
223 if (WARN_ON_ONCE(PageDirty(page
)))
224 account_page_cleaned(page
, mapping
, inode_to_wb(mapping
->host
));
228 * Delete a page from the page cache and free it. Caller has to make
229 * sure the page is locked and that nobody else uses it - or that usage
230 * is safe. The caller must hold the i_pages lock.
232 void __delete_from_page_cache(struct page
*page
, void *shadow
)
234 struct address_space
*mapping
= page
->mapping
;
236 trace_mm_filemap_delete_from_page_cache(page
);
238 unaccount_page_cache_page(mapping
, page
);
239 page_cache_delete(mapping
, page
, shadow
);
242 static void page_cache_free_page(struct address_space
*mapping
,
245 void (*freepage
)(struct page
*);
247 freepage
= mapping
->a_ops
->freepage
;
251 if (PageTransHuge(page
) && !PageHuge(page
)) {
252 page_ref_sub(page
, thp_nr_pages(page
));
253 VM_BUG_ON_PAGE(page_count(page
) <= 0, page
);
260 * delete_from_page_cache - delete page from page cache
261 * @page: the page which the kernel is trying to remove from page cache
263 * This must be called only on pages that have been verified to be in the page
264 * cache and locked. It will never put the page into the free list, the caller
265 * has a reference on the page.
267 void delete_from_page_cache(struct page
*page
)
269 struct address_space
*mapping
= page_mapping(page
);
272 BUG_ON(!PageLocked(page
));
273 xa_lock_irqsave(&mapping
->i_pages
, flags
);
274 __delete_from_page_cache(page
, NULL
);
275 xa_unlock_irqrestore(&mapping
->i_pages
, flags
);
277 page_cache_free_page(mapping
, page
);
279 EXPORT_SYMBOL(delete_from_page_cache
);
282 * page_cache_delete_batch - delete several pages from page cache
283 * @mapping: the mapping to which pages belong
284 * @pvec: pagevec with pages to delete
286 * The function walks over mapping->i_pages and removes pages passed in @pvec
287 * from the mapping. The function expects @pvec to be sorted by page index
288 * and is optimised for it to be dense.
289 * It tolerates holes in @pvec (mapping entries at those indices are not
290 * modified). The function expects only THP head pages to be present in the
293 * The function expects the i_pages lock to be held.
295 static void page_cache_delete_batch(struct address_space
*mapping
,
296 struct pagevec
*pvec
)
298 XA_STATE(xas
, &mapping
->i_pages
, pvec
->pages
[0]->index
);
303 mapping_set_update(&xas
, mapping
);
304 xas_for_each(&xas
, page
, ULONG_MAX
) {
305 if (i
>= pagevec_count(pvec
))
308 /* A swap/dax/shadow entry got inserted? Skip it. */
309 if (xa_is_value(page
))
312 * A page got inserted in our range? Skip it. We have our
313 * pages locked so they are protected from being removed.
314 * If we see a page whose index is higher than ours, it
315 * means our page has been removed, which shouldn't be
316 * possible because we're holding the PageLock.
318 if (page
!= pvec
->pages
[i
]) {
319 VM_BUG_ON_PAGE(page
->index
> pvec
->pages
[i
]->index
,
324 WARN_ON_ONCE(!PageLocked(page
));
326 if (page
->index
== xas
.xa_index
)
327 page
->mapping
= NULL
;
328 /* Leave page->index set: truncation lookup relies on it */
331 * Move to the next page in the vector if this is a regular
332 * page or the index is of the last sub-page of this compound
335 if (page
->index
+ compound_nr(page
) - 1 == xas
.xa_index
)
337 xas_store(&xas
, NULL
);
340 mapping
->nrpages
-= total_pages
;
343 void delete_from_page_cache_batch(struct address_space
*mapping
,
344 struct pagevec
*pvec
)
349 if (!pagevec_count(pvec
))
352 xa_lock_irqsave(&mapping
->i_pages
, flags
);
353 for (i
= 0; i
< pagevec_count(pvec
); i
++) {
354 trace_mm_filemap_delete_from_page_cache(pvec
->pages
[i
]);
356 unaccount_page_cache_page(mapping
, pvec
->pages
[i
]);
358 page_cache_delete_batch(mapping
, pvec
);
359 xa_unlock_irqrestore(&mapping
->i_pages
, flags
);
361 for (i
= 0; i
< pagevec_count(pvec
); i
++)
362 page_cache_free_page(mapping
, pvec
->pages
[i
]);
365 int filemap_check_errors(struct address_space
*mapping
)
368 /* Check for outstanding write errors */
369 if (test_bit(AS_ENOSPC
, &mapping
->flags
) &&
370 test_and_clear_bit(AS_ENOSPC
, &mapping
->flags
))
372 if (test_bit(AS_EIO
, &mapping
->flags
) &&
373 test_and_clear_bit(AS_EIO
, &mapping
->flags
))
377 EXPORT_SYMBOL(filemap_check_errors
);
379 static int filemap_check_and_keep_errors(struct address_space
*mapping
)
381 /* Check for outstanding write errors */
382 if (test_bit(AS_EIO
, &mapping
->flags
))
384 if (test_bit(AS_ENOSPC
, &mapping
->flags
))
390 * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range
391 * @mapping: address space structure to write
392 * @start: offset in bytes where the range starts
393 * @end: offset in bytes where the range ends (inclusive)
394 * @sync_mode: enable synchronous operation
396 * Start writeback against all of a mapping's dirty pages that lie
397 * within the byte offsets <start, end> inclusive.
399 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
400 * opposed to a regular memory cleansing writeback. The difference between
401 * these two operations is that if a dirty page/buffer is encountered, it must
402 * be waited upon, and not just skipped over.
404 * Return: %0 on success, negative error code otherwise.
406 int __filemap_fdatawrite_range(struct address_space
*mapping
, loff_t start
,
407 loff_t end
, int sync_mode
)
410 struct writeback_control wbc
= {
411 .sync_mode
= sync_mode
,
412 .nr_to_write
= LONG_MAX
,
413 .range_start
= start
,
417 if (!mapping_can_writeback(mapping
) ||
418 !mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
))
421 wbc_attach_fdatawrite_inode(&wbc
, mapping
->host
);
422 ret
= do_writepages(mapping
, &wbc
);
423 wbc_detach_inode(&wbc
);
427 static inline int __filemap_fdatawrite(struct address_space
*mapping
,
430 return __filemap_fdatawrite_range(mapping
, 0, LLONG_MAX
, sync_mode
);
433 int filemap_fdatawrite(struct address_space
*mapping
)
435 return __filemap_fdatawrite(mapping
, WB_SYNC_ALL
);
437 EXPORT_SYMBOL(filemap_fdatawrite
);
439 int filemap_fdatawrite_range(struct address_space
*mapping
, loff_t start
,
442 return __filemap_fdatawrite_range(mapping
, start
, end
, WB_SYNC_ALL
);
444 EXPORT_SYMBOL(filemap_fdatawrite_range
);
447 * filemap_flush - mostly a non-blocking flush
448 * @mapping: target address_space
450 * This is a mostly non-blocking flush. Not suitable for data-integrity
451 * purposes - I/O may not be started against all dirty pages.
453 * Return: %0 on success, negative error code otherwise.
455 int filemap_flush(struct address_space
*mapping
)
457 return __filemap_fdatawrite(mapping
, WB_SYNC_NONE
);
459 EXPORT_SYMBOL(filemap_flush
);
462 * filemap_range_has_page - check if a page exists in range.
463 * @mapping: address space within which to check
464 * @start_byte: offset in bytes where the range starts
465 * @end_byte: offset in bytes where the range ends (inclusive)
467 * Find at least one page in the range supplied, usually used to check if
468 * direct writing in this range will trigger a writeback.
470 * Return: %true if at least one page exists in the specified range,
473 bool filemap_range_has_page(struct address_space
*mapping
,
474 loff_t start_byte
, loff_t end_byte
)
477 XA_STATE(xas
, &mapping
->i_pages
, start_byte
>> PAGE_SHIFT
);
478 pgoff_t max
= end_byte
>> PAGE_SHIFT
;
480 if (end_byte
< start_byte
)
485 page
= xas_find(&xas
, max
);
486 if (xas_retry(&xas
, page
))
488 /* Shadow entries don't count */
489 if (xa_is_value(page
))
492 * We don't need to try to pin this page; we're about to
493 * release the RCU lock anyway. It is enough to know that
494 * there was a page here recently.
502 EXPORT_SYMBOL(filemap_range_has_page
);
504 static void __filemap_fdatawait_range(struct address_space
*mapping
,
505 loff_t start_byte
, loff_t end_byte
)
507 pgoff_t index
= start_byte
>> PAGE_SHIFT
;
508 pgoff_t end
= end_byte
>> PAGE_SHIFT
;
512 if (end_byte
< start_byte
)
516 while (index
<= end
) {
519 nr_pages
= pagevec_lookup_range_tag(&pvec
, mapping
, &index
,
520 end
, PAGECACHE_TAG_WRITEBACK
);
524 for (i
= 0; i
< nr_pages
; i
++) {
525 struct page
*page
= pvec
.pages
[i
];
527 wait_on_page_writeback(page
);
528 ClearPageError(page
);
530 pagevec_release(&pvec
);
536 * filemap_fdatawait_range - wait for writeback to complete
537 * @mapping: address space structure to wait for
538 * @start_byte: offset in bytes where the range starts
539 * @end_byte: offset in bytes where the range ends (inclusive)
541 * Walk the list of under-writeback pages of the given address space
542 * in the given range and wait for all of them. Check error status of
543 * the address space and return it.
545 * Since the error status of the address space is cleared by this function,
546 * callers are responsible for checking the return value and handling and/or
547 * reporting the error.
549 * Return: error status of the address space.
551 int filemap_fdatawait_range(struct address_space
*mapping
, loff_t start_byte
,
554 __filemap_fdatawait_range(mapping
, start_byte
, end_byte
);
555 return filemap_check_errors(mapping
);
557 EXPORT_SYMBOL(filemap_fdatawait_range
);
560 * filemap_fdatawait_range_keep_errors - wait for writeback to complete
561 * @mapping: address space structure to wait for
562 * @start_byte: offset in bytes where the range starts
563 * @end_byte: offset in bytes where the range ends (inclusive)
565 * Walk the list of under-writeback pages of the given address space in the
566 * given range and wait for all of them. Unlike filemap_fdatawait_range(),
567 * this function does not clear error status of the address space.
569 * Use this function if callers don't handle errors themselves. Expected
570 * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2),
573 int filemap_fdatawait_range_keep_errors(struct address_space
*mapping
,
574 loff_t start_byte
, loff_t end_byte
)
576 __filemap_fdatawait_range(mapping
, start_byte
, end_byte
);
577 return filemap_check_and_keep_errors(mapping
);
579 EXPORT_SYMBOL(filemap_fdatawait_range_keep_errors
);
582 * file_fdatawait_range - wait for writeback to complete
583 * @file: file pointing to address space structure to wait for
584 * @start_byte: offset in bytes where the range starts
585 * @end_byte: offset in bytes where the range ends (inclusive)
587 * Walk the list of under-writeback pages of the address space that file
588 * refers to, in the given range and wait for all of them. Check error
589 * status of the address space vs. the file->f_wb_err cursor and return it.
591 * Since the error status of the file is advanced by this function,
592 * callers are responsible for checking the return value and handling and/or
593 * reporting the error.
595 * Return: error status of the address space vs. the file->f_wb_err cursor.
597 int file_fdatawait_range(struct file
*file
, loff_t start_byte
, loff_t end_byte
)
599 struct address_space
*mapping
= file
->f_mapping
;
601 __filemap_fdatawait_range(mapping
, start_byte
, end_byte
);
602 return file_check_and_advance_wb_err(file
);
604 EXPORT_SYMBOL(file_fdatawait_range
);
607 * filemap_fdatawait_keep_errors - wait for writeback without clearing errors
608 * @mapping: address space structure to wait for
610 * Walk the list of under-writeback pages of the given address space
611 * and wait for all of them. Unlike filemap_fdatawait(), this function
612 * does not clear error status of the address space.
614 * Use this function if callers don't handle errors themselves. Expected
615 * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2),
618 * Return: error status of the address space.
620 int filemap_fdatawait_keep_errors(struct address_space
*mapping
)
622 __filemap_fdatawait_range(mapping
, 0, LLONG_MAX
);
623 return filemap_check_and_keep_errors(mapping
);
625 EXPORT_SYMBOL(filemap_fdatawait_keep_errors
);
627 /* Returns true if writeback might be needed or already in progress. */
628 static bool mapping_needs_writeback(struct address_space
*mapping
)
630 if (dax_mapping(mapping
))
631 return mapping
->nrexceptional
;
633 return mapping
->nrpages
;
637 * filemap_write_and_wait_range - write out & wait on a file range
638 * @mapping: the address_space for the pages
639 * @lstart: offset in bytes where the range starts
640 * @lend: offset in bytes where the range ends (inclusive)
642 * Write out and wait upon file offsets lstart->lend, inclusive.
644 * Note that @lend is inclusive (describes the last byte to be written) so
645 * that this function can be used to write to the very end-of-file (end = -1).
647 * Return: error status of the address space.
649 int filemap_write_and_wait_range(struct address_space
*mapping
,
650 loff_t lstart
, loff_t lend
)
654 if (mapping_needs_writeback(mapping
)) {
655 err
= __filemap_fdatawrite_range(mapping
, lstart
, lend
,
658 * Even if the above returned error, the pages may be
659 * written partially (e.g. -ENOSPC), so we wait for it.
660 * But the -EIO is special case, it may indicate the worst
661 * thing (e.g. bug) happened, so we avoid waiting for it.
664 int err2
= filemap_fdatawait_range(mapping
,
669 /* Clear any previously stored errors */
670 filemap_check_errors(mapping
);
673 err
= filemap_check_errors(mapping
);
677 EXPORT_SYMBOL(filemap_write_and_wait_range
);
679 void __filemap_set_wb_err(struct address_space
*mapping
, int err
)
681 errseq_t eseq
= errseq_set(&mapping
->wb_err
, err
);
683 trace_filemap_set_wb_err(mapping
, eseq
);
685 EXPORT_SYMBOL(__filemap_set_wb_err
);
688 * file_check_and_advance_wb_err - report wb error (if any) that was previously
689 * and advance wb_err to current one
690 * @file: struct file on which the error is being reported
692 * When userland calls fsync (or something like nfsd does the equivalent), we
693 * want to report any writeback errors that occurred since the last fsync (or
694 * since the file was opened if there haven't been any).
696 * Grab the wb_err from the mapping. If it matches what we have in the file,
697 * then just quickly return 0. The file is all caught up.
699 * If it doesn't match, then take the mapping value, set the "seen" flag in
700 * it and try to swap it into place. If it works, or another task beat us
701 * to it with the new value, then update the f_wb_err and return the error
702 * portion. The error at this point must be reported via proper channels
703 * (a'la fsync, or NFS COMMIT operation, etc.).
705 * While we handle mapping->wb_err with atomic operations, the f_wb_err
706 * value is protected by the f_lock since we must ensure that it reflects
707 * the latest value swapped in for this file descriptor.
709 * Return: %0 on success, negative error code otherwise.
711 int file_check_and_advance_wb_err(struct file
*file
)
714 errseq_t old
= READ_ONCE(file
->f_wb_err
);
715 struct address_space
*mapping
= file
->f_mapping
;
717 /* Locklessly handle the common case where nothing has changed */
718 if (errseq_check(&mapping
->wb_err
, old
)) {
719 /* Something changed, must use slow path */
720 spin_lock(&file
->f_lock
);
721 old
= file
->f_wb_err
;
722 err
= errseq_check_and_advance(&mapping
->wb_err
,
724 trace_file_check_and_advance_wb_err(file
, old
);
725 spin_unlock(&file
->f_lock
);
729 * We're mostly using this function as a drop in replacement for
730 * filemap_check_errors. Clear AS_EIO/AS_ENOSPC to emulate the effect
731 * that the legacy code would have had on these flags.
733 clear_bit(AS_EIO
, &mapping
->flags
);
734 clear_bit(AS_ENOSPC
, &mapping
->flags
);
737 EXPORT_SYMBOL(file_check_and_advance_wb_err
);
740 * file_write_and_wait_range - write out & wait on a file range
741 * @file: file pointing to address_space with pages
742 * @lstart: offset in bytes where the range starts
743 * @lend: offset in bytes where the range ends (inclusive)
745 * Write out and wait upon file offsets lstart->lend, inclusive.
747 * Note that @lend is inclusive (describes the last byte to be written) so
748 * that this function can be used to write to the very end-of-file (end = -1).
750 * After writing out and waiting on the data, we check and advance the
751 * f_wb_err cursor to the latest value, and return any errors detected there.
753 * Return: %0 on success, negative error code otherwise.
755 int file_write_and_wait_range(struct file
*file
, loff_t lstart
, loff_t lend
)
758 struct address_space
*mapping
= file
->f_mapping
;
760 if (mapping_needs_writeback(mapping
)) {
761 err
= __filemap_fdatawrite_range(mapping
, lstart
, lend
,
763 /* See comment of filemap_write_and_wait() */
765 __filemap_fdatawait_range(mapping
, lstart
, lend
);
767 err2
= file_check_and_advance_wb_err(file
);
772 EXPORT_SYMBOL(file_write_and_wait_range
);
775 * replace_page_cache_page - replace a pagecache page with a new one
776 * @old: page to be replaced
777 * @new: page to replace with
778 * @gfp_mask: allocation mode
780 * This function replaces a page in the pagecache with a new one. On
781 * success it acquires the pagecache reference for the new page and
782 * drops it for the old page. Both the old and new pages must be
783 * locked. This function does not add the new page to the LRU, the
784 * caller must do that.
786 * The remove + add is atomic. This function cannot fail.
790 int replace_page_cache_page(struct page
*old
, struct page
*new, gfp_t gfp_mask
)
792 struct address_space
*mapping
= old
->mapping
;
793 void (*freepage
)(struct page
*) = mapping
->a_ops
->freepage
;
794 pgoff_t offset
= old
->index
;
795 XA_STATE(xas
, &mapping
->i_pages
, offset
);
798 VM_BUG_ON_PAGE(!PageLocked(old
), old
);
799 VM_BUG_ON_PAGE(!PageLocked(new), new);
800 VM_BUG_ON_PAGE(new->mapping
, new);
803 new->mapping
= mapping
;
806 mem_cgroup_migrate(old
, new);
808 xas_lock_irqsave(&xas
, flags
);
809 xas_store(&xas
, new);
812 /* hugetlb pages do not participate in page cache accounting. */
814 __dec_lruvec_page_state(old
, NR_FILE_PAGES
);
816 __inc_lruvec_page_state(new, NR_FILE_PAGES
);
817 if (PageSwapBacked(old
))
818 __dec_lruvec_page_state(old
, NR_SHMEM
);
819 if (PageSwapBacked(new))
820 __inc_lruvec_page_state(new, NR_SHMEM
);
821 xas_unlock_irqrestore(&xas
, flags
);
828 EXPORT_SYMBOL_GPL(replace_page_cache_page
);
830 noinline
int __add_to_page_cache_locked(struct page
*page
,
831 struct address_space
*mapping
,
832 pgoff_t offset
, gfp_t gfp
,
835 XA_STATE(xas
, &mapping
->i_pages
, offset
);
836 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
);
853 gfp
&= GFP_RECLAIM_MASK
;
856 unsigned int order
= xa_get_order(xas
.xa
, xas
.xa_index
);
857 void *entry
, *old
= NULL
;
859 if (order
> thp_order(page
))
860 xas_split_alloc(&xas
, xa_load(xas
.xa
, xas
.xa_index
),
863 xas_for_each_conflict(&xas
, entry
) {
865 if (!xa_is_value(entry
)) {
866 xas_set_err(&xas
, -EEXIST
);
874 /* entry may have been split before we acquired lock */
875 order
= xa_get_order(xas
.xa
, xas
.xa_index
);
876 if (order
> thp_order(page
)) {
877 xas_split(&xas
, old
, order
);
882 xas_store(&xas
, page
);
887 mapping
->nrexceptional
--;
890 /* hugetlb pages do not participate in page cache accounting */
892 __inc_lruvec_page_state(page
, NR_FILE_PAGES
);
894 xas_unlock_irq(&xas
);
895 } while (xas_nomem(&xas
, gfp
));
897 if (xas_error(&xas
)) {
898 error
= xas_error(&xas
);
902 trace_mm_filemap_add_to_page_cache(page
);
905 page
->mapping
= NULL
;
906 /* Leave page->index set: truncation relies upon it */
910 ALLOW_ERROR_INJECTION(__add_to_page_cache_locked
, ERRNO
);
913 * add_to_page_cache_locked - add a locked page to the pagecache
915 * @mapping: the page's address_space
916 * @offset: page index
917 * @gfp_mask: page allocation mode
919 * This function is used to add a page to the pagecache. It must be locked.
920 * This function does not add the page to the LRU. The caller must do that.
922 * Return: %0 on success, negative error code otherwise.
924 int add_to_page_cache_locked(struct page
*page
, struct address_space
*mapping
,
925 pgoff_t offset
, gfp_t gfp_mask
)
927 return __add_to_page_cache_locked(page
, mapping
, offset
,
930 EXPORT_SYMBOL(add_to_page_cache_locked
);
932 int add_to_page_cache_lru(struct page
*page
, struct address_space
*mapping
,
933 pgoff_t offset
, gfp_t gfp_mask
)
938 __SetPageLocked(page
);
939 ret
= __add_to_page_cache_locked(page
, mapping
, offset
,
942 __ClearPageLocked(page
);
945 * The page might have been evicted from cache only
946 * recently, in which case it should be activated like
947 * any other repeatedly accessed page.
948 * The exception is pages getting rewritten; evicting other
949 * data from the working set, only to cache data that will
950 * get overwritten with something else, is a waste of memory.
952 WARN_ON_ONCE(PageActive(page
));
953 if (!(gfp_mask
& __GFP_WRITE
) && shadow
)
954 workingset_refault(page
, shadow
);
959 EXPORT_SYMBOL_GPL(add_to_page_cache_lru
);
962 struct page
*__page_cache_alloc(gfp_t gfp
)
967 if (cpuset_do_page_mem_spread()) {
968 unsigned int cpuset_mems_cookie
;
970 cpuset_mems_cookie
= read_mems_allowed_begin();
971 n
= cpuset_mem_spread_node();
972 page
= __alloc_pages_node(n
, gfp
, 0);
973 } while (!page
&& read_mems_allowed_retry(cpuset_mems_cookie
));
977 return alloc_pages(gfp
, 0);
979 EXPORT_SYMBOL(__page_cache_alloc
);
983 * In order to wait for pages to become available there must be
984 * waitqueues associated with pages. By using a hash table of
985 * waitqueues where the bucket discipline is to maintain all
986 * waiters on the same queue and wake all when any of the pages
987 * become available, and for the woken contexts to check to be
988 * sure the appropriate page became available, this saves space
989 * at a cost of "thundering herd" phenomena during rare hash
992 #define PAGE_WAIT_TABLE_BITS 8
993 #define PAGE_WAIT_TABLE_SIZE (1 << PAGE_WAIT_TABLE_BITS)
994 static wait_queue_head_t page_wait_table
[PAGE_WAIT_TABLE_SIZE
] __cacheline_aligned
;
996 static wait_queue_head_t
*page_waitqueue(struct page
*page
)
998 return &page_wait_table
[hash_ptr(page
, PAGE_WAIT_TABLE_BITS
)];
1001 void __init
pagecache_init(void)
1005 for (i
= 0; i
< PAGE_WAIT_TABLE_SIZE
; i
++)
1006 init_waitqueue_head(&page_wait_table
[i
]);
1008 page_writeback_init();
1012 * The page wait code treats the "wait->flags" somewhat unusually, because
1013 * we have multiple different kinds of waits, not just the usual "exclusive"
1018 * (a) no special bits set:
1020 * We're just waiting for the bit to be released, and when a waker
1021 * calls the wakeup function, we set WQ_FLAG_WOKEN and wake it up,
1022 * and remove it from the wait queue.
1024 * Simple and straightforward.
1026 * (b) WQ_FLAG_EXCLUSIVE:
1028 * The waiter is waiting to get the lock, and only one waiter should
1029 * be woken up to avoid any thundering herd behavior. We'll set the
1030 * WQ_FLAG_WOKEN bit, wake it up, and remove it from the wait queue.
1032 * This is the traditional exclusive wait.
1034 * (c) WQ_FLAG_EXCLUSIVE | WQ_FLAG_CUSTOM:
1036 * The waiter is waiting to get the bit, and additionally wants the
1037 * lock to be transferred to it for fair lock behavior. If the lock
1038 * cannot be taken, we stop walking the wait queue without waking
1041 * This is the "fair lock handoff" case, and in addition to setting
1042 * WQ_FLAG_WOKEN, we set WQ_FLAG_DONE to let the waiter easily see
1043 * that it now has the lock.
1045 static int wake_page_function(wait_queue_entry_t
*wait
, unsigned mode
, int sync
, void *arg
)
1048 struct wait_page_key
*key
= arg
;
1049 struct wait_page_queue
*wait_page
1050 = container_of(wait
, struct wait_page_queue
, wait
);
1052 if (!wake_page_match(wait_page
, key
))
1056 * If it's a lock handoff wait, we get the bit for it, and
1057 * stop walking (and do not wake it up) if we can't.
1059 flags
= wait
->flags
;
1060 if (flags
& WQ_FLAG_EXCLUSIVE
) {
1061 if (test_bit(key
->bit_nr
, &key
->page
->flags
))
1063 if (flags
& WQ_FLAG_CUSTOM
) {
1064 if (test_and_set_bit(key
->bit_nr
, &key
->page
->flags
))
1066 flags
|= WQ_FLAG_DONE
;
1071 * We are holding the wait-queue lock, but the waiter that
1072 * is waiting for this will be checking the flags without
1075 * So update the flags atomically, and wake up the waiter
1076 * afterwards to avoid any races. This store-release pairs
1077 * with the load-acquire in wait_on_page_bit_common().
1079 smp_store_release(&wait
->flags
, flags
| WQ_FLAG_WOKEN
);
1080 wake_up_state(wait
->private, mode
);
1083 * Ok, we have successfully done what we're waiting for,
1084 * and we can unconditionally remove the wait entry.
1086 * Note that this pairs with the "finish_wait()" in the
1087 * waiter, and has to be the absolute last thing we do.
1088 * After this list_del_init(&wait->entry) the wait entry
1089 * might be de-allocated and the process might even have
1092 list_del_init_careful(&wait
->entry
);
1093 return (flags
& WQ_FLAG_EXCLUSIVE
) != 0;
1096 static void wake_up_page_bit(struct page
*page
, int bit_nr
)
1098 wait_queue_head_t
*q
= page_waitqueue(page
);
1099 struct wait_page_key key
;
1100 unsigned long flags
;
1101 wait_queue_entry_t bookmark
;
1104 key
.bit_nr
= bit_nr
;
1108 bookmark
.private = NULL
;
1109 bookmark
.func
= NULL
;
1110 INIT_LIST_HEAD(&bookmark
.entry
);
1112 spin_lock_irqsave(&q
->lock
, flags
);
1113 __wake_up_locked_key_bookmark(q
, TASK_NORMAL
, &key
, &bookmark
);
1115 while (bookmark
.flags
& WQ_FLAG_BOOKMARK
) {
1117 * Take a breather from holding the lock,
1118 * allow pages that finish wake up asynchronously
1119 * to acquire the lock and remove themselves
1122 spin_unlock_irqrestore(&q
->lock
, flags
);
1124 spin_lock_irqsave(&q
->lock
, flags
);
1125 __wake_up_locked_key_bookmark(q
, TASK_NORMAL
, &key
, &bookmark
);
1129 * It is possible for other pages to have collided on the waitqueue
1130 * hash, so in that case check for a page match. That prevents a long-
1133 * It is still possible to miss a case here, when we woke page waiters
1134 * and removed them from the waitqueue, but there are still other
1137 if (!waitqueue_active(q
) || !key
.page_match
) {
1138 ClearPageWaiters(page
);
1140 * It's possible to miss clearing Waiters here, when we woke
1141 * our page waiters, but the hashed waitqueue has waiters for
1142 * other pages on it.
1144 * That's okay, it's a rare case. The next waker will clear it.
1147 spin_unlock_irqrestore(&q
->lock
, flags
);
1150 static void wake_up_page(struct page
*page
, int bit
)
1152 if (!PageWaiters(page
))
1154 wake_up_page_bit(page
, bit
);
1158 * A choice of three behaviors for wait_on_page_bit_common():
1161 EXCLUSIVE
, /* Hold ref to page and take the bit when woken, like
1162 * __lock_page() waiting on then setting PG_locked.
1164 SHARED
, /* Hold ref to page and check the bit when woken, like
1165 * wait_on_page_writeback() waiting on PG_writeback.
1167 DROP
, /* Drop ref to page before wait, no check when woken,
1168 * like put_and_wait_on_page_locked() on PG_locked.
1173 * Attempt to check (or get) the page bit, and mark us done
1176 static inline bool trylock_page_bit_common(struct page
*page
, int bit_nr
,
1177 struct wait_queue_entry
*wait
)
1179 if (wait
->flags
& WQ_FLAG_EXCLUSIVE
) {
1180 if (test_and_set_bit(bit_nr
, &page
->flags
))
1182 } else if (test_bit(bit_nr
, &page
->flags
))
1185 wait
->flags
|= WQ_FLAG_WOKEN
| WQ_FLAG_DONE
;
1189 /* How many times do we accept lock stealing from under a waiter? */
1190 int sysctl_page_lock_unfairness
= 5;
1192 static inline int wait_on_page_bit_common(wait_queue_head_t
*q
,
1193 struct page
*page
, int bit_nr
, int state
, enum behavior behavior
)
1195 int unfairness
= sysctl_page_lock_unfairness
;
1196 struct wait_page_queue wait_page
;
1197 wait_queue_entry_t
*wait
= &wait_page
.wait
;
1198 bool thrashing
= false;
1199 bool delayacct
= false;
1200 unsigned long pflags
;
1202 if (bit_nr
== PG_locked
&&
1203 !PageUptodate(page
) && PageWorkingset(page
)) {
1204 if (!PageSwapBacked(page
)) {
1205 delayacct_thrashing_start();
1208 psi_memstall_enter(&pflags
);
1213 wait
->func
= wake_page_function
;
1214 wait_page
.page
= page
;
1215 wait_page
.bit_nr
= bit_nr
;
1219 if (behavior
== EXCLUSIVE
) {
1220 wait
->flags
= WQ_FLAG_EXCLUSIVE
;
1221 if (--unfairness
< 0)
1222 wait
->flags
|= WQ_FLAG_CUSTOM
;
1226 * Do one last check whether we can get the
1227 * page bit synchronously.
1229 * Do the SetPageWaiters() marking before that
1230 * to let any waker we _just_ missed know they
1231 * need to wake us up (otherwise they'll never
1232 * even go to the slow case that looks at the
1233 * page queue), and add ourselves to the wait
1234 * queue if we need to sleep.
1236 * This part needs to be done under the queue
1237 * lock to avoid races.
1239 spin_lock_irq(&q
->lock
);
1240 SetPageWaiters(page
);
1241 if (!trylock_page_bit_common(page
, bit_nr
, wait
))
1242 __add_wait_queue_entry_tail(q
, wait
);
1243 spin_unlock_irq(&q
->lock
);
1246 * From now on, all the logic will be based on
1247 * the WQ_FLAG_WOKEN and WQ_FLAG_DONE flag, to
1248 * see whether the page bit testing has already
1249 * been done by the wake function.
1251 * We can drop our reference to the page.
1253 if (behavior
== DROP
)
1257 * Note that until the "finish_wait()", or until
1258 * we see the WQ_FLAG_WOKEN flag, we need to
1259 * be very careful with the 'wait->flags', because
1260 * we may race with a waker that sets them.
1265 set_current_state(state
);
1267 /* Loop until we've been woken or interrupted */
1268 flags
= smp_load_acquire(&wait
->flags
);
1269 if (!(flags
& WQ_FLAG_WOKEN
)) {
1270 if (signal_pending_state(state
, current
))
1277 /* If we were non-exclusive, we're done */
1278 if (behavior
!= EXCLUSIVE
)
1281 /* If the waker got the lock for us, we're done */
1282 if (flags
& WQ_FLAG_DONE
)
1286 * Otherwise, if we're getting the lock, we need to
1287 * try to get it ourselves.
1289 * And if that fails, we'll have to retry this all.
1291 if (unlikely(test_and_set_bit(bit_nr
, &page
->flags
)))
1294 wait
->flags
|= WQ_FLAG_DONE
;
1299 * If a signal happened, this 'finish_wait()' may remove the last
1300 * waiter from the wait-queues, but the PageWaiters bit will remain
1301 * set. That's ok. The next wakeup will take care of it, and trying
1302 * to do it here would be difficult and prone to races.
1304 finish_wait(q
, wait
);
1308 delayacct_thrashing_end();
1309 psi_memstall_leave(&pflags
);
1313 * NOTE! The wait->flags weren't stable until we've done the
1314 * 'finish_wait()', and we could have exited the loop above due
1315 * to a signal, and had a wakeup event happen after the signal
1316 * test but before the 'finish_wait()'.
1318 * So only after the finish_wait() can we reliably determine
1319 * if we got woken up or not, so we can now figure out the final
1320 * return value based on that state without races.
1322 * Also note that WQ_FLAG_WOKEN is sufficient for a non-exclusive
1323 * waiter, but an exclusive one requires WQ_FLAG_DONE.
1325 if (behavior
== EXCLUSIVE
)
1326 return wait
->flags
& WQ_FLAG_DONE
? 0 : -EINTR
;
1328 return wait
->flags
& WQ_FLAG_WOKEN
? 0 : -EINTR
;
1331 void wait_on_page_bit(struct page
*page
, int bit_nr
)
1333 wait_queue_head_t
*q
= page_waitqueue(page
);
1334 wait_on_page_bit_common(q
, page
, bit_nr
, TASK_UNINTERRUPTIBLE
, SHARED
);
1336 EXPORT_SYMBOL(wait_on_page_bit
);
1338 int wait_on_page_bit_killable(struct page
*page
, int bit_nr
)
1340 wait_queue_head_t
*q
= page_waitqueue(page
);
1341 return wait_on_page_bit_common(q
, page
, bit_nr
, TASK_KILLABLE
, SHARED
);
1343 EXPORT_SYMBOL(wait_on_page_bit_killable
);
1345 static int __wait_on_page_locked_async(struct page
*page
,
1346 struct wait_page_queue
*wait
, bool set
)
1348 struct wait_queue_head
*q
= page_waitqueue(page
);
1352 wait
->bit_nr
= PG_locked
;
1354 spin_lock_irq(&q
->lock
);
1355 __add_wait_queue_entry_tail(q
, &wait
->wait
);
1356 SetPageWaiters(page
);
1358 ret
= !trylock_page(page
);
1360 ret
= PageLocked(page
);
1362 * If we were succesful now, we know we're still on the
1363 * waitqueue as we're still under the lock. This means it's
1364 * safe to remove and return success, we know the callback
1365 * isn't going to trigger.
1368 __remove_wait_queue(q
, &wait
->wait
);
1371 spin_unlock_irq(&q
->lock
);
1375 static int wait_on_page_locked_async(struct page
*page
,
1376 struct wait_page_queue
*wait
)
1378 if (!PageLocked(page
))
1380 return __wait_on_page_locked_async(compound_head(page
), wait
, false);
1384 * put_and_wait_on_page_locked - Drop a reference and wait for it to be unlocked
1385 * @page: The page to wait for.
1387 * The caller should hold a reference on @page. They expect the page to
1388 * become unlocked relatively soon, but do not wish to hold up migration
1389 * (for example) by holding the reference while waiting for the page to
1390 * come unlocked. After this function returns, the caller should not
1391 * dereference @page.
1393 void put_and_wait_on_page_locked(struct page
*page
)
1395 wait_queue_head_t
*q
;
1397 page
= compound_head(page
);
1398 q
= page_waitqueue(page
);
1399 wait_on_page_bit_common(q
, page
, PG_locked
, TASK_UNINTERRUPTIBLE
, DROP
);
1403 * add_page_wait_queue - Add an arbitrary waiter to a page's wait queue
1404 * @page: Page defining the wait queue of interest
1405 * @waiter: Waiter to add to the queue
1407 * Add an arbitrary @waiter to the wait queue for the nominated @page.
1409 void add_page_wait_queue(struct page
*page
, wait_queue_entry_t
*waiter
)
1411 wait_queue_head_t
*q
= page_waitqueue(page
);
1412 unsigned long flags
;
1414 spin_lock_irqsave(&q
->lock
, flags
);
1415 __add_wait_queue_entry_tail(q
, waiter
);
1416 SetPageWaiters(page
);
1417 spin_unlock_irqrestore(&q
->lock
, flags
);
1419 EXPORT_SYMBOL_GPL(add_page_wait_queue
);
1421 #ifndef clear_bit_unlock_is_negative_byte
1424 * PG_waiters is the high bit in the same byte as PG_lock.
1426 * On x86 (and on many other architectures), we can clear PG_lock and
1427 * test the sign bit at the same time. But if the architecture does
1428 * not support that special operation, we just do this all by hand
1431 * The read of PG_waiters has to be after (or concurrently with) PG_locked
1432 * being cleared, but a memory barrier should be unnecessary since it is
1433 * in the same byte as PG_locked.
1435 static inline bool clear_bit_unlock_is_negative_byte(long nr
, volatile void *mem
)
1437 clear_bit_unlock(nr
, mem
);
1438 /* smp_mb__after_atomic(); */
1439 return test_bit(PG_waiters
, mem
);
1445 * unlock_page - unlock a locked page
1448 * Unlocks the page and wakes up sleepers in wait_on_page_locked().
1449 * Also wakes sleepers in wait_on_page_writeback() because the wakeup
1450 * mechanism between PageLocked pages and PageWriteback pages is shared.
1451 * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep.
1453 * Note that this depends on PG_waiters being the sign bit in the byte
1454 * that contains PG_locked - thus the BUILD_BUG_ON(). That allows us to
1455 * clear the PG_locked bit and test PG_waiters at the same time fairly
1456 * portably (architectures that do LL/SC can test any bit, while x86 can
1457 * test the sign bit).
1459 void unlock_page(struct page
*page
)
1461 BUILD_BUG_ON(PG_waiters
!= 7);
1462 page
= compound_head(page
);
1463 VM_BUG_ON_PAGE(!PageLocked(page
), page
);
1464 if (clear_bit_unlock_is_negative_byte(PG_locked
, &page
->flags
))
1465 wake_up_page_bit(page
, PG_locked
);
1467 EXPORT_SYMBOL(unlock_page
);
1470 * end_page_writeback - end writeback against a page
1473 void end_page_writeback(struct page
*page
)
1476 * TestClearPageReclaim could be used here but it is an atomic
1477 * operation and overkill in this particular case. Failing to
1478 * shuffle a page marked for immediate reclaim is too mild to
1479 * justify taking an atomic operation penalty at the end of
1480 * ever page writeback.
1482 if (PageReclaim(page
)) {
1483 ClearPageReclaim(page
);
1484 rotate_reclaimable_page(page
);
1488 * Writeback does not hold a page reference of its own, relying
1489 * on truncation to wait for the clearing of PG_writeback.
1490 * But here we must make sure that the page is not freed and
1491 * reused before the wake_up_page().
1494 if (!test_clear_page_writeback(page
))
1497 smp_mb__after_atomic();
1498 wake_up_page(page
, PG_writeback
);
1501 EXPORT_SYMBOL(end_page_writeback
);
1504 * After completing I/O on a page, call this routine to update the page
1505 * flags appropriately
1507 void page_endio(struct page
*page
, bool is_write
, int err
)
1511 SetPageUptodate(page
);
1513 ClearPageUptodate(page
);
1519 struct address_space
*mapping
;
1522 mapping
= page_mapping(page
);
1524 mapping_set_error(mapping
, err
);
1526 end_page_writeback(page
);
1529 EXPORT_SYMBOL_GPL(page_endio
);
1532 * __lock_page - get a lock on the page, assuming we need to sleep to get it
1533 * @__page: the page to lock
1535 void __lock_page(struct page
*__page
)
1537 struct page
*page
= compound_head(__page
);
1538 wait_queue_head_t
*q
= page_waitqueue(page
);
1539 wait_on_page_bit_common(q
, page
, PG_locked
, TASK_UNINTERRUPTIBLE
,
1542 EXPORT_SYMBOL(__lock_page
);
1544 int __lock_page_killable(struct page
*__page
)
1546 struct page
*page
= compound_head(__page
);
1547 wait_queue_head_t
*q
= page_waitqueue(page
);
1548 return wait_on_page_bit_common(q
, page
, PG_locked
, TASK_KILLABLE
,
1551 EXPORT_SYMBOL_GPL(__lock_page_killable
);
1553 int __lock_page_async(struct page
*page
, struct wait_page_queue
*wait
)
1555 return __wait_on_page_locked_async(page
, wait
, true);
1560 * 1 - page is locked; mmap_lock is still held.
1561 * 0 - page is not locked.
1562 * mmap_lock has been released (mmap_read_unlock(), unless flags had both
1563 * FAULT_FLAG_ALLOW_RETRY and FAULT_FLAG_RETRY_NOWAIT set, in
1564 * which case mmap_lock is still held.
1566 * If neither ALLOW_RETRY nor KILLABLE are set, will always return 1
1567 * with the page locked and the mmap_lock unperturbed.
1569 int __lock_page_or_retry(struct page
*page
, struct mm_struct
*mm
,
1572 if (fault_flag_allow_retry_first(flags
)) {
1574 * CAUTION! In this case, mmap_lock is not released
1575 * even though return 0.
1577 if (flags
& FAULT_FLAG_RETRY_NOWAIT
)
1580 mmap_read_unlock(mm
);
1581 if (flags
& FAULT_FLAG_KILLABLE
)
1582 wait_on_page_locked_killable(page
);
1584 wait_on_page_locked(page
);
1587 if (flags
& FAULT_FLAG_KILLABLE
) {
1590 ret
= __lock_page_killable(page
);
1592 mmap_read_unlock(mm
);
1602 * page_cache_next_miss() - Find the next gap in the page cache.
1603 * @mapping: Mapping.
1605 * @max_scan: Maximum range to search.
1607 * Search the range [index, min(index + max_scan - 1, ULONG_MAX)] for the
1608 * gap with the lowest index.
1610 * This function may be called under the rcu_read_lock. However, this will
1611 * not atomically search a snapshot of the cache at a single point in time.
1612 * For example, if a gap is created at index 5, then subsequently a gap is
1613 * created at index 10, page_cache_next_miss covering both indices may
1614 * return 10 if called under the rcu_read_lock.
1616 * Return: The index of the gap if found, otherwise an index outside the
1617 * range specified (in which case 'return - index >= max_scan' will be true).
1618 * In the rare case of index wrap-around, 0 will be returned.
1620 pgoff_t
page_cache_next_miss(struct address_space
*mapping
,
1621 pgoff_t index
, unsigned long max_scan
)
1623 XA_STATE(xas
, &mapping
->i_pages
, index
);
1625 while (max_scan
--) {
1626 void *entry
= xas_next(&xas
);
1627 if (!entry
|| xa_is_value(entry
))
1629 if (xas
.xa_index
== 0)
1633 return xas
.xa_index
;
1635 EXPORT_SYMBOL(page_cache_next_miss
);
1638 * page_cache_prev_miss() - Find the previous gap in the page cache.
1639 * @mapping: Mapping.
1641 * @max_scan: Maximum range to search.
1643 * Search the range [max(index - max_scan + 1, 0), index] for the
1644 * gap with the highest index.
1646 * This function may be called under the rcu_read_lock. However, this will
1647 * not atomically search a snapshot of the cache at a single point in time.
1648 * For example, if a gap is created at index 10, then subsequently a gap is
1649 * created at index 5, page_cache_prev_miss() covering both indices may
1650 * return 5 if called under the rcu_read_lock.
1652 * Return: The index of the gap if found, otherwise an index outside the
1653 * range specified (in which case 'index - return >= max_scan' will be true).
1654 * In the rare case of wrap-around, ULONG_MAX will be returned.
1656 pgoff_t
page_cache_prev_miss(struct address_space
*mapping
,
1657 pgoff_t index
, unsigned long max_scan
)
1659 XA_STATE(xas
, &mapping
->i_pages
, index
);
1661 while (max_scan
--) {
1662 void *entry
= xas_prev(&xas
);
1663 if (!entry
|| xa_is_value(entry
))
1665 if (xas
.xa_index
== ULONG_MAX
)
1669 return xas
.xa_index
;
1671 EXPORT_SYMBOL(page_cache_prev_miss
);
1674 * find_get_entry - find and get a page cache entry
1675 * @mapping: the address_space to search
1676 * @index: The page cache index.
1678 * Looks up the page cache slot at @mapping & @offset. If there is a
1679 * page cache page, the head page is returned with an increased refcount.
1681 * If the slot holds a shadow entry of a previously evicted page, or a
1682 * swap entry from shmem/tmpfs, it is returned.
1684 * Return: The head page or shadow entry, %NULL if nothing is found.
1686 struct page
*find_get_entry(struct address_space
*mapping
, pgoff_t index
)
1688 XA_STATE(xas
, &mapping
->i_pages
, index
);
1694 page
= xas_load(&xas
);
1695 if (xas_retry(&xas
, page
))
1698 * A shadow entry of a recently evicted page, or a swap entry from
1699 * shmem/tmpfs. Return it without attempting to raise page count.
1701 if (!page
|| xa_is_value(page
))
1704 if (!page_cache_get_speculative(page
))
1708 * Has the page moved or been split?
1709 * This is part of the lockless pagecache protocol. See
1710 * include/linux/pagemap.h for details.
1712 if (unlikely(page
!= xas_reload(&xas
))) {
1723 * find_lock_entry - Locate and lock a page cache entry.
1724 * @mapping: The address_space to search.
1725 * @index: The page cache index.
1727 * Looks up the page at @mapping & @index. If there is a page in the
1728 * cache, the head page is returned locked and with an increased refcount.
1730 * If the slot holds a shadow entry of a previously evicted page, or a
1731 * swap entry from shmem/tmpfs, it is returned.
1733 * Context: May sleep.
1734 * Return: The head page or shadow entry, %NULL if nothing is found.
1736 struct page
*find_lock_entry(struct address_space
*mapping
, pgoff_t index
)
1741 page
= find_get_entry(mapping
, index
);
1742 if (page
&& !xa_is_value(page
)) {
1744 /* Has the page been truncated? */
1745 if (unlikely(page
->mapping
!= mapping
)) {
1750 VM_BUG_ON_PAGE(!thp_contains(page
, index
), page
);
1756 * pagecache_get_page - Find and get a reference to a page.
1757 * @mapping: The address_space to search.
1758 * @index: The page index.
1759 * @fgp_flags: %FGP flags modify how the page is returned.
1760 * @gfp_mask: Memory allocation flags to use if %FGP_CREAT is specified.
1762 * Looks up the page cache entry at @mapping & @index.
1764 * @fgp_flags can be zero or more of these flags:
1766 * * %FGP_ACCESSED - The page will be marked accessed.
1767 * * %FGP_LOCK - The page is returned locked.
1768 * * %FGP_HEAD - If the page is present and a THP, return the head page
1769 * rather than the exact page specified by the index.
1770 * * %FGP_CREAT - If no page is present then a new page is allocated using
1771 * @gfp_mask and added to the page cache and the VM's LRU list.
1772 * The page is returned locked and with an increased refcount.
1773 * * %FGP_FOR_MMAP - The caller wants to do its own locking dance if the
1774 * page is already in cache. If the page was allocated, unlock it before
1775 * returning so the caller can do the same dance.
1776 * * %FGP_WRITE - The page will be written
1777 * * %FGP_NOFS - __GFP_FS will get cleared in gfp mask
1778 * * %FGP_NOWAIT - Don't get blocked by page lock
1780 * If %FGP_LOCK or %FGP_CREAT are specified then the function may sleep even
1781 * if the %GFP flags specified for %FGP_CREAT are atomic.
1783 * If there is a page cache page, it is returned with an increased refcount.
1785 * Return: The found page or %NULL otherwise.
1787 struct page
*pagecache_get_page(struct address_space
*mapping
, pgoff_t index
,
1788 int fgp_flags
, gfp_t gfp_mask
)
1793 page
= find_get_entry(mapping
, index
);
1794 if (xa_is_value(page
))
1799 if (fgp_flags
& FGP_LOCK
) {
1800 if (fgp_flags
& FGP_NOWAIT
) {
1801 if (!trylock_page(page
)) {
1809 /* Has the page been truncated? */
1810 if (unlikely(page
->mapping
!= mapping
)) {
1815 VM_BUG_ON_PAGE(!thp_contains(page
, index
), page
);
1818 if (fgp_flags
& FGP_ACCESSED
)
1819 mark_page_accessed(page
);
1820 else if (fgp_flags
& FGP_WRITE
) {
1821 /* Clear idle flag for buffer write */
1822 if (page_is_idle(page
))
1823 clear_page_idle(page
);
1825 if (!(fgp_flags
& FGP_HEAD
))
1826 page
= find_subpage(page
, index
);
1829 if (!page
&& (fgp_flags
& FGP_CREAT
)) {
1831 if ((fgp_flags
& FGP_WRITE
) && mapping_can_writeback(mapping
))
1832 gfp_mask
|= __GFP_WRITE
;
1833 if (fgp_flags
& FGP_NOFS
)
1834 gfp_mask
&= ~__GFP_FS
;
1836 page
= __page_cache_alloc(gfp_mask
);
1840 if (WARN_ON_ONCE(!(fgp_flags
& (FGP_LOCK
| FGP_FOR_MMAP
))))
1841 fgp_flags
|= FGP_LOCK
;
1843 /* Init accessed so avoid atomic mark_page_accessed later */
1844 if (fgp_flags
& FGP_ACCESSED
)
1845 __SetPageReferenced(page
);
1847 err
= add_to_page_cache_lru(page
, mapping
, index
, gfp_mask
);
1848 if (unlikely(err
)) {
1856 * add_to_page_cache_lru locks the page, and for mmap we expect
1859 if (page
&& (fgp_flags
& FGP_FOR_MMAP
))
1865 EXPORT_SYMBOL(pagecache_get_page
);
1868 * find_get_entries - gang pagecache lookup
1869 * @mapping: The address_space to search
1870 * @start: The starting page cache index
1871 * @nr_entries: The maximum number of entries
1872 * @entries: Where the resulting entries are placed
1873 * @indices: The cache indices corresponding to the entries in @entries
1875 * find_get_entries() will search for and return a group of up to
1876 * @nr_entries entries in the mapping. The entries are placed at
1877 * @entries. find_get_entries() takes a reference against any actual
1880 * The search returns a group of mapping-contiguous page cache entries
1881 * with ascending indexes. There may be holes in the indices due to
1882 * not-present pages.
1884 * Any shadow entries of evicted pages, or swap entries from
1885 * shmem/tmpfs, are included in the returned array.
1887 * If it finds a Transparent Huge Page, head or tail, find_get_entries()
1888 * stops at that page: the caller is likely to have a better way to handle
1889 * the compound page as a whole, and then skip its extent, than repeatedly
1890 * calling find_get_entries() to return all its tails.
1892 * Return: the number of pages and shadow entries which were found.
1894 unsigned find_get_entries(struct address_space
*mapping
,
1895 pgoff_t start
, unsigned int nr_entries
,
1896 struct page
**entries
, pgoff_t
*indices
)
1898 XA_STATE(xas
, &mapping
->i_pages
, start
);
1900 unsigned int ret
= 0;
1906 xas_for_each(&xas
, page
, ULONG_MAX
) {
1907 if (xas_retry(&xas
, page
))
1910 * A shadow entry of a recently evicted page, a swap
1911 * entry from shmem/tmpfs or a DAX entry. Return it
1912 * without attempting to raise page count.
1914 if (xa_is_value(page
))
1917 if (!page_cache_get_speculative(page
))
1920 /* Has the page moved or been split? */
1921 if (unlikely(page
!= xas_reload(&xas
)))
1925 * Terminate early on finding a THP, to allow the caller to
1926 * handle it all at once; but continue if this is hugetlbfs.
1928 if (PageTransHuge(page
) && !PageHuge(page
)) {
1929 page
= find_subpage(page
, xas
.xa_index
);
1930 nr_entries
= ret
+ 1;
1933 indices
[ret
] = xas
.xa_index
;
1934 entries
[ret
] = page
;
1935 if (++ret
== nr_entries
)
1948 * find_get_pages_range - gang pagecache lookup
1949 * @mapping: The address_space to search
1950 * @start: The starting page index
1951 * @end: The final page index (inclusive)
1952 * @nr_pages: The maximum number of pages
1953 * @pages: Where the resulting pages are placed
1955 * find_get_pages_range() will search for and return a group of up to @nr_pages
1956 * pages in the mapping starting at index @start and up to index @end
1957 * (inclusive). The pages are placed at @pages. find_get_pages_range() takes
1958 * a reference against the returned pages.
1960 * The search returns a group of mapping-contiguous pages with ascending
1961 * indexes. There may be holes in the indices due to not-present pages.
1962 * We also update @start to index the next page for the traversal.
1964 * Return: the number of pages which were found. If this number is
1965 * smaller than @nr_pages, the end of specified range has been
1968 unsigned find_get_pages_range(struct address_space
*mapping
, pgoff_t
*start
,
1969 pgoff_t end
, unsigned int nr_pages
,
1970 struct page
**pages
)
1972 XA_STATE(xas
, &mapping
->i_pages
, *start
);
1976 if (unlikely(!nr_pages
))
1980 xas_for_each(&xas
, page
, end
) {
1981 if (xas_retry(&xas
, page
))
1983 /* Skip over shadow, swap and DAX entries */
1984 if (xa_is_value(page
))
1987 if (!page_cache_get_speculative(page
))
1990 /* Has the page moved or been split? */
1991 if (unlikely(page
!= xas_reload(&xas
)))
1994 pages
[ret
] = find_subpage(page
, xas
.xa_index
);
1995 if (++ret
== nr_pages
) {
1996 *start
= xas
.xa_index
+ 1;
2007 * We come here when there is no page beyond @end. We take care to not
2008 * overflow the index @start as it confuses some of the callers. This
2009 * breaks the iteration when there is a page at index -1 but that is
2010 * already broken anyway.
2012 if (end
== (pgoff_t
)-1)
2013 *start
= (pgoff_t
)-1;
2023 * find_get_pages_contig - gang contiguous pagecache lookup
2024 * @mapping: The address_space to search
2025 * @index: The starting page index
2026 * @nr_pages: The maximum number of pages
2027 * @pages: Where the resulting pages are placed
2029 * find_get_pages_contig() works exactly like find_get_pages(), except
2030 * that the returned number of pages are guaranteed to be contiguous.
2032 * Return: the number of pages which were found.
2034 unsigned find_get_pages_contig(struct address_space
*mapping
, pgoff_t index
,
2035 unsigned int nr_pages
, struct page
**pages
)
2037 XA_STATE(xas
, &mapping
->i_pages
, index
);
2039 unsigned int ret
= 0;
2041 if (unlikely(!nr_pages
))
2045 for (page
= xas_load(&xas
); page
; page
= xas_next(&xas
)) {
2046 if (xas_retry(&xas
, page
))
2049 * If the entry has been swapped out, we can stop looking.
2050 * No current caller is looking for DAX entries.
2052 if (xa_is_value(page
))
2055 if (!page_cache_get_speculative(page
))
2058 /* Has the page moved or been split? */
2059 if (unlikely(page
!= xas_reload(&xas
)))
2062 pages
[ret
] = find_subpage(page
, xas
.xa_index
);
2063 if (++ret
== nr_pages
)
2074 EXPORT_SYMBOL(find_get_pages_contig
);
2077 * find_get_pages_range_tag - find and return pages in given range matching @tag
2078 * @mapping: the address_space to search
2079 * @index: the starting page index
2080 * @end: The final page index (inclusive)
2081 * @tag: the tag index
2082 * @nr_pages: the maximum number of pages
2083 * @pages: where the resulting pages are placed
2085 * Like find_get_pages, except we only return pages which are tagged with
2086 * @tag. We update @index to index the next page for the traversal.
2088 * Return: the number of pages which were found.
2090 unsigned find_get_pages_range_tag(struct address_space
*mapping
, pgoff_t
*index
,
2091 pgoff_t end
, xa_mark_t tag
, unsigned int nr_pages
,
2092 struct page
**pages
)
2094 XA_STATE(xas
, &mapping
->i_pages
, *index
);
2098 if (unlikely(!nr_pages
))
2102 xas_for_each_marked(&xas
, page
, end
, tag
) {
2103 if (xas_retry(&xas
, page
))
2106 * Shadow entries should never be tagged, but this iteration
2107 * is lockless so there is a window for page reclaim to evict
2108 * a page we saw tagged. Skip over it.
2110 if (xa_is_value(page
))
2113 if (!page_cache_get_speculative(page
))
2116 /* Has the page moved or been split? */
2117 if (unlikely(page
!= xas_reload(&xas
)))
2120 pages
[ret
] = find_subpage(page
, xas
.xa_index
);
2121 if (++ret
== nr_pages
) {
2122 *index
= xas
.xa_index
+ 1;
2133 * We come here when we got to @end. We take care to not overflow the
2134 * index @index as it confuses some of the callers. This breaks the
2135 * iteration when there is a page at index -1 but that is already
2138 if (end
== (pgoff_t
)-1)
2139 *index
= (pgoff_t
)-1;
2147 EXPORT_SYMBOL(find_get_pages_range_tag
);
2150 * CD/DVDs are error prone. When a medium error occurs, the driver may fail
2151 * a _large_ part of the i/o request. Imagine the worst scenario:
2153 * ---R__________________________________________B__________
2154 * ^ reading here ^ bad block(assume 4k)
2156 * read(R) => miss => readahead(R...B) => media error => frustrating retries
2157 * => failing the whole request => read(R) => read(R+1) =>
2158 * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
2159 * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
2160 * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
2162 * It is going insane. Fix it by quickly scaling down the readahead size.
2164 static void shrink_readahead_size_eio(struct file_ra_state
*ra
)
2170 * generic_file_buffered_read - generic file read routine
2171 * @iocb: the iocb to read
2172 * @iter: data destination
2173 * @written: already copied
2175 * This is a generic file read routine, and uses the
2176 * mapping->a_ops->readpage() function for the actual low-level stuff.
2178 * This is really ugly. But the goto's actually try to clarify some
2179 * of the logic when it comes to error handling etc.
2182 * * total number of bytes copied, including those the were already @written
2183 * * negative error code if nothing was copied
2185 ssize_t
generic_file_buffered_read(struct kiocb
*iocb
,
2186 struct iov_iter
*iter
, ssize_t written
)
2188 struct file
*filp
= iocb
->ki_filp
;
2189 struct address_space
*mapping
= filp
->f_mapping
;
2190 struct inode
*inode
= mapping
->host
;
2191 struct file_ra_state
*ra
= &filp
->f_ra
;
2192 loff_t
*ppos
= &iocb
->ki_pos
;
2196 unsigned long offset
; /* offset into pagecache page */
2197 unsigned int prev_offset
;
2200 if (unlikely(*ppos
>= inode
->i_sb
->s_maxbytes
))
2202 iov_iter_truncate(iter
, inode
->i_sb
->s_maxbytes
);
2204 index
= *ppos
>> PAGE_SHIFT
;
2205 prev_index
= ra
->prev_pos
>> PAGE_SHIFT
;
2206 prev_offset
= ra
->prev_pos
& (PAGE_SIZE
-1);
2207 last_index
= (*ppos
+ iter
->count
+ PAGE_SIZE
-1) >> PAGE_SHIFT
;
2208 offset
= *ppos
& ~PAGE_MASK
;
2211 * If we've already successfully copied some data, then we
2212 * can no longer safely return -EIOCBQUEUED. Hence mark
2213 * an async read NOWAIT at that point.
2215 if (written
&& (iocb
->ki_flags
& IOCB_WAITQ
))
2216 iocb
->ki_flags
|= IOCB_NOWAIT
;
2222 unsigned long nr
, ret
;
2226 if (fatal_signal_pending(current
)) {
2231 page
= find_get_page(mapping
, index
);
2233 if (iocb
->ki_flags
& IOCB_NOIO
)
2235 page_cache_sync_readahead(mapping
,
2237 index
, last_index
- index
);
2238 page
= find_get_page(mapping
, index
);
2239 if (unlikely(page
== NULL
))
2240 goto no_cached_page
;
2242 if (PageReadahead(page
)) {
2243 if (iocb
->ki_flags
& IOCB_NOIO
) {
2247 page_cache_async_readahead(mapping
,
2249 index
, last_index
- index
);
2251 if (!PageUptodate(page
)) {
2253 * See comment in do_read_cache_page on why
2254 * wait_on_page_locked is used to avoid unnecessarily
2255 * serialisations and why it's safe.
2257 if (iocb
->ki_flags
& IOCB_WAITQ
) {
2262 error
= wait_on_page_locked_async(page
,
2265 if (iocb
->ki_flags
& IOCB_NOWAIT
) {
2269 error
= wait_on_page_locked_killable(page
);
2271 if (unlikely(error
))
2272 goto readpage_error
;
2273 if (PageUptodate(page
))
2276 if (inode
->i_blkbits
== PAGE_SHIFT
||
2277 !mapping
->a_ops
->is_partially_uptodate
)
2278 goto page_not_up_to_date
;
2279 /* pipes can't handle partially uptodate pages */
2280 if (unlikely(iov_iter_is_pipe(iter
)))
2281 goto page_not_up_to_date
;
2282 if (!trylock_page(page
))
2283 goto page_not_up_to_date
;
2284 /* Did it get truncated before we got the lock? */
2286 goto page_not_up_to_date_locked
;
2287 if (!mapping
->a_ops
->is_partially_uptodate(page
,
2288 offset
, iter
->count
))
2289 goto page_not_up_to_date_locked
;
2294 * i_size must be checked after we know the page is Uptodate.
2296 * Checking i_size after the check allows us to calculate
2297 * the correct value for "nr", which means the zero-filled
2298 * part of the page is not copied back to userspace (unless
2299 * another truncate extends the file - this is desired though).
2302 isize
= i_size_read(inode
);
2303 end_index
= (isize
- 1) >> PAGE_SHIFT
;
2304 if (unlikely(!isize
|| index
> end_index
)) {
2309 /* nr is the maximum number of bytes to copy from this page */
2311 if (index
== end_index
) {
2312 nr
= ((isize
- 1) & ~PAGE_MASK
) + 1;
2320 /* If users can be writing to this page using arbitrary
2321 * virtual addresses, take care about potential aliasing
2322 * before reading the page on the kernel side.
2324 if (mapping_writably_mapped(mapping
))
2325 flush_dcache_page(page
);
2328 * When a sequential read accesses a page several times,
2329 * only mark it as accessed the first time.
2331 if (prev_index
!= index
|| offset
!= prev_offset
)
2332 mark_page_accessed(page
);
2336 * Ok, we have the page, and it's up-to-date, so
2337 * now we can copy it to user space...
2340 ret
= copy_page_to_iter(page
, offset
, nr
, iter
);
2342 index
+= offset
>> PAGE_SHIFT
;
2343 offset
&= ~PAGE_MASK
;
2344 prev_offset
= offset
;
2348 if (!iov_iter_count(iter
))
2356 page_not_up_to_date
:
2357 /* Get exclusive access to the page ... */
2358 if (iocb
->ki_flags
& IOCB_WAITQ
) {
2363 error
= lock_page_async(page
, iocb
->ki_waitq
);
2365 error
= lock_page_killable(page
);
2367 if (unlikely(error
))
2368 goto readpage_error
;
2370 page_not_up_to_date_locked
:
2371 /* Did it get truncated before we got the lock? */
2372 if (!page
->mapping
) {
2378 /* Did somebody else fill it already? */
2379 if (PageUptodate(page
)) {
2385 if (iocb
->ki_flags
& (IOCB_NOIO
| IOCB_NOWAIT
)) {
2391 * A previous I/O error may have been due to temporary
2392 * failures, eg. multipath errors.
2393 * PG_error will be set again if readpage fails.
2395 ClearPageError(page
);
2396 /* Start the actual read. The read will unlock the page. */
2397 error
= mapping
->a_ops
->readpage(filp
, page
);
2399 if (unlikely(error
)) {
2400 if (error
== AOP_TRUNCATED_PAGE
) {
2405 goto readpage_error
;
2408 if (!PageUptodate(page
)) {
2409 if (iocb
->ki_flags
& IOCB_WAITQ
) {
2414 error
= lock_page_async(page
, iocb
->ki_waitq
);
2416 error
= lock_page_killable(page
);
2419 if (unlikely(error
))
2420 goto readpage_error
;
2421 if (!PageUptodate(page
)) {
2422 if (page
->mapping
== NULL
) {
2424 * invalidate_mapping_pages got it
2431 shrink_readahead_size_eio(ra
);
2433 goto readpage_error
;
2441 /* UHHUH! A synchronous read error occurred. Report it */
2447 * Ok, it wasn't cached, so we need to create a new
2450 page
= page_cache_alloc(mapping
);
2455 error
= add_to_page_cache_lru(page
, mapping
, index
,
2456 mapping_gfp_constraint(mapping
, GFP_KERNEL
));
2459 if (error
== -EEXIST
) {
2471 ra
->prev_pos
= prev_index
;
2472 ra
->prev_pos
<<= PAGE_SHIFT
;
2473 ra
->prev_pos
|= prev_offset
;
2475 *ppos
= ((loff_t
)index
<< PAGE_SHIFT
) + offset
;
2476 file_accessed(filp
);
2477 return written
? written
: error
;
2479 EXPORT_SYMBOL_GPL(generic_file_buffered_read
);
2482 * generic_file_read_iter - generic filesystem read routine
2483 * @iocb: kernel I/O control block
2484 * @iter: destination for the data read
2486 * This is the "read_iter()" routine for all filesystems
2487 * that can use the page cache directly.
2489 * The IOCB_NOWAIT flag in iocb->ki_flags indicates that -EAGAIN shall
2490 * be returned when no data can be read without waiting for I/O requests
2491 * to complete; it doesn't prevent readahead.
2493 * The IOCB_NOIO flag in iocb->ki_flags indicates that no new I/O
2494 * requests shall be made for the read or for readahead. When no data
2495 * can be read, -EAGAIN shall be returned. When readahead would be
2496 * triggered, a partial, possibly empty read shall be returned.
2499 * * number of bytes copied, even for partial reads
2500 * * negative error code (or 0 if IOCB_NOIO) if nothing was read
2503 generic_file_read_iter(struct kiocb
*iocb
, struct iov_iter
*iter
)
2505 size_t count
= iov_iter_count(iter
);
2509 goto out
; /* skip atime */
2511 if (iocb
->ki_flags
& IOCB_DIRECT
) {
2512 struct file
*file
= iocb
->ki_filp
;
2513 struct address_space
*mapping
= file
->f_mapping
;
2514 struct inode
*inode
= mapping
->host
;
2517 size
= i_size_read(inode
);
2518 if (iocb
->ki_flags
& IOCB_NOWAIT
) {
2519 if (filemap_range_has_page(mapping
, iocb
->ki_pos
,
2520 iocb
->ki_pos
+ count
- 1))
2523 retval
= filemap_write_and_wait_range(mapping
,
2525 iocb
->ki_pos
+ count
- 1);
2530 file_accessed(file
);
2532 retval
= mapping
->a_ops
->direct_IO(iocb
, iter
);
2534 iocb
->ki_pos
+= retval
;
2537 iov_iter_revert(iter
, count
- iov_iter_count(iter
));
2540 * Btrfs can have a short DIO read if we encounter
2541 * compressed extents, so if there was an error, or if
2542 * we've already read everything we wanted to, or if
2543 * there was a short read because we hit EOF, go ahead
2544 * and return. Otherwise fallthrough to buffered io for
2545 * the rest of the read. Buffered reads will not work for
2546 * DAX files, so don't bother trying.
2548 if (retval
< 0 || !count
|| iocb
->ki_pos
>= size
||
2553 retval
= generic_file_buffered_read(iocb
, iter
, retval
);
2557 EXPORT_SYMBOL(generic_file_read_iter
);
2560 #define MMAP_LOTSAMISS (100)
2562 * lock_page_maybe_drop_mmap - lock the page, possibly dropping the mmap_lock
2563 * @vmf - the vm_fault for this fault.
2564 * @page - the page to lock.
2565 * @fpin - the pointer to the file we may pin (or is already pinned).
2567 * This works similar to lock_page_or_retry in that it can drop the mmap_lock.
2568 * It differs in that it actually returns the page locked if it returns 1 and 0
2569 * if it couldn't lock the page. If we did have to drop the mmap_lock then fpin
2570 * will point to the pinned file and needs to be fput()'ed at a later point.
2572 static int lock_page_maybe_drop_mmap(struct vm_fault
*vmf
, struct page
*page
,
2575 if (trylock_page(page
))
2579 * NOTE! This will make us return with VM_FAULT_RETRY, but with
2580 * the mmap_lock still held. That's how FAULT_FLAG_RETRY_NOWAIT
2581 * is supposed to work. We have way too many special cases..
2583 if (vmf
->flags
& FAULT_FLAG_RETRY_NOWAIT
)
2586 *fpin
= maybe_unlock_mmap_for_io(vmf
, *fpin
);
2587 if (vmf
->flags
& FAULT_FLAG_KILLABLE
) {
2588 if (__lock_page_killable(page
)) {
2590 * We didn't have the right flags to drop the mmap_lock,
2591 * but all fault_handlers only check for fatal signals
2592 * if we return VM_FAULT_RETRY, so we need to drop the
2593 * mmap_lock here and return 0 if we don't have a fpin.
2596 mmap_read_unlock(vmf
->vma
->vm_mm
);
2606 * Synchronous readahead happens when we don't even find a page in the page
2607 * cache at all. We don't want to perform IO under the mmap sem, so if we have
2608 * to drop the mmap sem we return the file that was pinned in order for us to do
2609 * that. If we didn't pin a file then we return NULL. The file that is
2610 * returned needs to be fput()'ed when we're done with it.
2612 static struct file
*do_sync_mmap_readahead(struct vm_fault
*vmf
)
2614 struct file
*file
= vmf
->vma
->vm_file
;
2615 struct file_ra_state
*ra
= &file
->f_ra
;
2616 struct address_space
*mapping
= file
->f_mapping
;
2617 DEFINE_READAHEAD(ractl
, file
, mapping
, vmf
->pgoff
);
2618 struct file
*fpin
= NULL
;
2619 unsigned int mmap_miss
;
2621 /* If we don't want any read-ahead, don't bother */
2622 if (vmf
->vma
->vm_flags
& VM_RAND_READ
)
2627 if (vmf
->vma
->vm_flags
& VM_SEQ_READ
) {
2628 fpin
= maybe_unlock_mmap_for_io(vmf
, fpin
);
2629 page_cache_sync_ra(&ractl
, ra
, ra
->ra_pages
);
2633 /* Avoid banging the cache line if not needed */
2634 mmap_miss
= READ_ONCE(ra
->mmap_miss
);
2635 if (mmap_miss
< MMAP_LOTSAMISS
* 10)
2636 WRITE_ONCE(ra
->mmap_miss
, ++mmap_miss
);
2639 * Do we miss much more than hit in this file? If so,
2640 * stop bothering with read-ahead. It will only hurt.
2642 if (mmap_miss
> MMAP_LOTSAMISS
)
2648 fpin
= maybe_unlock_mmap_for_io(vmf
, fpin
);
2649 ra
->start
= max_t(long, 0, vmf
->pgoff
- ra
->ra_pages
/ 2);
2650 ra
->size
= ra
->ra_pages
;
2651 ra
->async_size
= ra
->ra_pages
/ 4;
2652 ractl
._index
= ra
->start
;
2653 do_page_cache_ra(&ractl
, ra
->size
, ra
->async_size
);
2658 * Asynchronous readahead happens when we find the page and PG_readahead,
2659 * so we want to possibly extend the readahead further. We return the file that
2660 * was pinned if we have to drop the mmap_lock in order to do IO.
2662 static struct file
*do_async_mmap_readahead(struct vm_fault
*vmf
,
2665 struct file
*file
= vmf
->vma
->vm_file
;
2666 struct file_ra_state
*ra
= &file
->f_ra
;
2667 struct address_space
*mapping
= file
->f_mapping
;
2668 struct file
*fpin
= NULL
;
2669 unsigned int mmap_miss
;
2670 pgoff_t offset
= vmf
->pgoff
;
2672 /* If we don't want any read-ahead, don't bother */
2673 if (vmf
->vma
->vm_flags
& VM_RAND_READ
|| !ra
->ra_pages
)
2675 mmap_miss
= READ_ONCE(ra
->mmap_miss
);
2677 WRITE_ONCE(ra
->mmap_miss
, --mmap_miss
);
2678 if (PageReadahead(page
)) {
2679 fpin
= maybe_unlock_mmap_for_io(vmf
, fpin
);
2680 page_cache_async_readahead(mapping
, ra
, file
,
2681 page
, offset
, ra
->ra_pages
);
2687 * filemap_fault - read in file data for page fault handling
2688 * @vmf: struct vm_fault containing details of the fault
2690 * filemap_fault() is invoked via the vma operations vector for a
2691 * mapped memory region to read in file data during a page fault.
2693 * The goto's are kind of ugly, but this streamlines the normal case of having
2694 * it in the page cache, and handles the special cases reasonably without
2695 * having a lot of duplicated code.
2697 * vma->vm_mm->mmap_lock must be held on entry.
2699 * If our return value has VM_FAULT_RETRY set, it's because the mmap_lock
2700 * may be dropped before doing I/O or by lock_page_maybe_drop_mmap().
2702 * If our return value does not have VM_FAULT_RETRY set, the mmap_lock
2703 * has not been released.
2705 * We never return with VM_FAULT_RETRY and a bit from VM_FAULT_ERROR set.
2707 * Return: bitwise-OR of %VM_FAULT_ codes.
2709 vm_fault_t
filemap_fault(struct vm_fault
*vmf
)
2712 struct file
*file
= vmf
->vma
->vm_file
;
2713 struct file
*fpin
= NULL
;
2714 struct address_space
*mapping
= file
->f_mapping
;
2715 struct file_ra_state
*ra
= &file
->f_ra
;
2716 struct inode
*inode
= mapping
->host
;
2717 pgoff_t offset
= vmf
->pgoff
;
2722 max_off
= DIV_ROUND_UP(i_size_read(inode
), PAGE_SIZE
);
2723 if (unlikely(offset
>= max_off
))
2724 return VM_FAULT_SIGBUS
;
2727 * Do we have something in the page cache already?
2729 page
= find_get_page(mapping
, offset
);
2730 if (likely(page
) && !(vmf
->flags
& FAULT_FLAG_TRIED
)) {
2732 * We found the page, so try async readahead before
2733 * waiting for the lock.
2735 fpin
= do_async_mmap_readahead(vmf
, page
);
2737 /* No page in the page cache at all */
2738 count_vm_event(PGMAJFAULT
);
2739 count_memcg_event_mm(vmf
->vma
->vm_mm
, PGMAJFAULT
);
2740 ret
= VM_FAULT_MAJOR
;
2741 fpin
= do_sync_mmap_readahead(vmf
);
2743 page
= pagecache_get_page(mapping
, offset
,
2744 FGP_CREAT
|FGP_FOR_MMAP
,
2749 return VM_FAULT_OOM
;
2753 if (!lock_page_maybe_drop_mmap(vmf
, page
, &fpin
))
2756 /* Did it get truncated? */
2757 if (unlikely(compound_head(page
)->mapping
!= mapping
)) {
2762 VM_BUG_ON_PAGE(page_to_pgoff(page
) != offset
, page
);
2765 * We have a locked page in the page cache, now we need to check
2766 * that it's up-to-date. If not, it is going to be due to an error.
2768 if (unlikely(!PageUptodate(page
)))
2769 goto page_not_uptodate
;
2772 * We've made it this far and we had to drop our mmap_lock, now is the
2773 * time to return to the upper layer and have it re-find the vma and
2782 * Found the page and have a reference on it.
2783 * We must recheck i_size under page lock.
2785 max_off
= DIV_ROUND_UP(i_size_read(inode
), PAGE_SIZE
);
2786 if (unlikely(offset
>= max_off
)) {
2789 return VM_FAULT_SIGBUS
;
2793 return ret
| VM_FAULT_LOCKED
;
2797 * Umm, take care of errors if the page isn't up-to-date.
2798 * Try to re-read it _once_. We do this synchronously,
2799 * because there really aren't any performance issues here
2800 * and we need to check for errors.
2802 ClearPageError(page
);
2803 fpin
= maybe_unlock_mmap_for_io(vmf
, fpin
);
2804 error
= mapping
->a_ops
->readpage(file
, page
);
2806 wait_on_page_locked(page
);
2807 if (!PageUptodate(page
))
2814 if (!error
|| error
== AOP_TRUNCATED_PAGE
)
2817 shrink_readahead_size_eio(ra
);
2818 return VM_FAULT_SIGBUS
;
2822 * We dropped the mmap_lock, we need to return to the fault handler to
2823 * re-find the vma and come back and find our hopefully still populated
2830 return ret
| VM_FAULT_RETRY
;
2832 EXPORT_SYMBOL(filemap_fault
);
2834 void filemap_map_pages(struct vm_fault
*vmf
,
2835 pgoff_t start_pgoff
, pgoff_t end_pgoff
)
2837 struct file
*file
= vmf
->vma
->vm_file
;
2838 struct address_space
*mapping
= file
->f_mapping
;
2839 pgoff_t last_pgoff
= start_pgoff
;
2840 unsigned long max_idx
;
2841 XA_STATE(xas
, &mapping
->i_pages
, start_pgoff
);
2842 struct page
*head
, *page
;
2843 unsigned int mmap_miss
= READ_ONCE(file
->f_ra
.mmap_miss
);
2846 xas_for_each(&xas
, head
, end_pgoff
) {
2847 if (xas_retry(&xas
, head
))
2849 if (xa_is_value(head
))
2853 * Check for a locked page first, as a speculative
2854 * reference may adversely influence page migration.
2856 if (PageLocked(head
))
2858 if (!page_cache_get_speculative(head
))
2861 /* Has the page moved or been split? */
2862 if (unlikely(head
!= xas_reload(&xas
)))
2864 page
= find_subpage(head
, xas
.xa_index
);
2866 if (!PageUptodate(head
) ||
2867 PageReadahead(page
) ||
2870 if (!trylock_page(head
))
2873 if (head
->mapping
!= mapping
|| !PageUptodate(head
))
2876 max_idx
= DIV_ROUND_UP(i_size_read(mapping
->host
), PAGE_SIZE
);
2877 if (xas
.xa_index
>= max_idx
)
2883 vmf
->address
+= (xas
.xa_index
- last_pgoff
) << PAGE_SHIFT
;
2885 vmf
->pte
+= xas
.xa_index
- last_pgoff
;
2886 last_pgoff
= xas
.xa_index
;
2887 if (alloc_set_pte(vmf
, page
))
2896 /* Huge page is mapped? No need to proceed. */
2897 if (pmd_trans_huge(*vmf
->pmd
))
2901 WRITE_ONCE(file
->f_ra
.mmap_miss
, mmap_miss
);
2903 EXPORT_SYMBOL(filemap_map_pages
);
2905 vm_fault_t
filemap_page_mkwrite(struct vm_fault
*vmf
)
2907 struct page
*page
= vmf
->page
;
2908 struct inode
*inode
= file_inode(vmf
->vma
->vm_file
);
2909 vm_fault_t ret
= VM_FAULT_LOCKED
;
2911 sb_start_pagefault(inode
->i_sb
);
2912 file_update_time(vmf
->vma
->vm_file
);
2914 if (page
->mapping
!= inode
->i_mapping
) {
2916 ret
= VM_FAULT_NOPAGE
;
2920 * We mark the page dirty already here so that when freeze is in
2921 * progress, we are guaranteed that writeback during freezing will
2922 * see the dirty page and writeprotect it again.
2924 set_page_dirty(page
);
2925 wait_for_stable_page(page
);
2927 sb_end_pagefault(inode
->i_sb
);
2931 const struct vm_operations_struct generic_file_vm_ops
= {
2932 .fault
= filemap_fault
,
2933 .map_pages
= filemap_map_pages
,
2934 .page_mkwrite
= filemap_page_mkwrite
,
2937 /* This is used for a general mmap of a disk file */
2939 int generic_file_mmap(struct file
* file
, struct vm_area_struct
* vma
)
2941 struct address_space
*mapping
= file
->f_mapping
;
2943 if (!mapping
->a_ops
->readpage
)
2945 file_accessed(file
);
2946 vma
->vm_ops
= &generic_file_vm_ops
;
2951 * This is for filesystems which do not implement ->writepage.
2953 int generic_file_readonly_mmap(struct file
*file
, struct vm_area_struct
*vma
)
2955 if ((vma
->vm_flags
& VM_SHARED
) && (vma
->vm_flags
& VM_MAYWRITE
))
2957 return generic_file_mmap(file
, vma
);
2960 vm_fault_t
filemap_page_mkwrite(struct vm_fault
*vmf
)
2962 return VM_FAULT_SIGBUS
;
2964 int generic_file_mmap(struct file
* file
, struct vm_area_struct
* vma
)
2968 int generic_file_readonly_mmap(struct file
* file
, struct vm_area_struct
* vma
)
2972 #endif /* CONFIG_MMU */
2974 EXPORT_SYMBOL(filemap_page_mkwrite
);
2975 EXPORT_SYMBOL(generic_file_mmap
);
2976 EXPORT_SYMBOL(generic_file_readonly_mmap
);
2978 static struct page
*wait_on_page_read(struct page
*page
)
2980 if (!IS_ERR(page
)) {
2981 wait_on_page_locked(page
);
2982 if (!PageUptodate(page
)) {
2984 page
= ERR_PTR(-EIO
);
2990 static struct page
*do_read_cache_page(struct address_space
*mapping
,
2992 int (*filler
)(void *, struct page
*),
2999 page
= find_get_page(mapping
, index
);
3001 page
= __page_cache_alloc(gfp
);
3003 return ERR_PTR(-ENOMEM
);
3004 err
= add_to_page_cache_lru(page
, mapping
, index
, gfp
);
3005 if (unlikely(err
)) {
3009 /* Presumably ENOMEM for xarray node */
3010 return ERR_PTR(err
);
3015 err
= filler(data
, page
);
3017 err
= mapping
->a_ops
->readpage(data
, page
);
3021 return ERR_PTR(err
);
3024 page
= wait_on_page_read(page
);
3029 if (PageUptodate(page
))
3033 * Page is not up to date and may be locked due to one of the following
3034 * case a: Page is being filled and the page lock is held
3035 * case b: Read/write error clearing the page uptodate status
3036 * case c: Truncation in progress (page locked)
3037 * case d: Reclaim in progress
3039 * Case a, the page will be up to date when the page is unlocked.
3040 * There is no need to serialise on the page lock here as the page
3041 * is pinned so the lock gives no additional protection. Even if the
3042 * page is truncated, the data is still valid if PageUptodate as
3043 * it's a race vs truncate race.
3044 * Case b, the page will not be up to date
3045 * Case c, the page may be truncated but in itself, the data may still
3046 * be valid after IO completes as it's a read vs truncate race. The
3047 * operation must restart if the page is not uptodate on unlock but
3048 * otherwise serialising on page lock to stabilise the mapping gives
3049 * no additional guarantees to the caller as the page lock is
3050 * released before return.
3051 * Case d, similar to truncation. If reclaim holds the page lock, it
3052 * will be a race with remove_mapping that determines if the mapping
3053 * is valid on unlock but otherwise the data is valid and there is
3054 * no need to serialise with page lock.
3056 * As the page lock gives no additional guarantee, we optimistically
3057 * wait on the page to be unlocked and check if it's up to date and
3058 * use the page if it is. Otherwise, the page lock is required to
3059 * distinguish between the different cases. The motivation is that we
3060 * avoid spurious serialisations and wakeups when multiple processes
3061 * wait on the same page for IO to complete.
3063 wait_on_page_locked(page
);
3064 if (PageUptodate(page
))
3067 /* Distinguish between all the cases under the safety of the lock */
3070 /* Case c or d, restart the operation */
3071 if (!page
->mapping
) {
3077 /* Someone else locked and filled the page in a very small window */
3078 if (PageUptodate(page
)) {
3084 * A previous I/O error may have been due to temporary
3086 * Clear page error before actual read, PG_error will be
3087 * set again if read page fails.
3089 ClearPageError(page
);
3093 mark_page_accessed(page
);
3098 * read_cache_page - read into page cache, fill it if needed
3099 * @mapping: the page's address_space
3100 * @index: the page index
3101 * @filler: function to perform the read
3102 * @data: first arg to filler(data, page) function, often left as NULL
3104 * Read into the page cache. If a page already exists, and PageUptodate() is
3105 * not set, try to fill the page and wait for it to become unlocked.
3107 * If the page does not get brought uptodate, return -EIO.
3109 * Return: up to date page on success, ERR_PTR() on failure.
3111 struct page
*read_cache_page(struct address_space
*mapping
,
3113 int (*filler
)(void *, struct page
*),
3116 return do_read_cache_page(mapping
, index
, filler
, data
,
3117 mapping_gfp_mask(mapping
));
3119 EXPORT_SYMBOL(read_cache_page
);
3122 * read_cache_page_gfp - read into page cache, using specified page allocation flags.
3123 * @mapping: the page's address_space
3124 * @index: the page index
3125 * @gfp: the page allocator flags to use if allocating
3127 * This is the same as "read_mapping_page(mapping, index, NULL)", but with
3128 * any new page allocations done using the specified allocation flags.
3130 * If the page does not get brought uptodate, return -EIO.
3132 * Return: up to date page on success, ERR_PTR() on failure.
3134 struct page
*read_cache_page_gfp(struct address_space
*mapping
,
3138 return do_read_cache_page(mapping
, index
, NULL
, NULL
, gfp
);
3140 EXPORT_SYMBOL(read_cache_page_gfp
);
3142 int pagecache_write_begin(struct file
*file
, struct address_space
*mapping
,
3143 loff_t pos
, unsigned len
, unsigned flags
,
3144 struct page
**pagep
, void **fsdata
)
3146 const struct address_space_operations
*aops
= mapping
->a_ops
;
3148 return aops
->write_begin(file
, mapping
, pos
, len
, flags
,
3151 EXPORT_SYMBOL(pagecache_write_begin
);
3153 int pagecache_write_end(struct file
*file
, struct address_space
*mapping
,
3154 loff_t pos
, unsigned len
, unsigned copied
,
3155 struct page
*page
, void *fsdata
)
3157 const struct address_space_operations
*aops
= mapping
->a_ops
;
3159 return aops
->write_end(file
, mapping
, pos
, len
, copied
, page
, fsdata
);
3161 EXPORT_SYMBOL(pagecache_write_end
);
3164 * Warn about a page cache invalidation failure during a direct I/O write.
3166 void dio_warn_stale_pagecache(struct file
*filp
)
3168 static DEFINE_RATELIMIT_STATE(_rs
, 86400 * HZ
, DEFAULT_RATELIMIT_BURST
);
3170 struct inode
*inode
= file_inode(filp
);
3173 errseq_set(&inode
->i_mapping
->wb_err
, -EIO
);
3174 if (__ratelimit(&_rs
)) {
3175 path
= file_path(filp
, pathname
, sizeof(pathname
));
3178 pr_crit("Page cache invalidation failure on direct I/O. Possible data corruption due to collision with buffered I/O!\n");
3179 pr_crit("File: %s PID: %d Comm: %.20s\n", path
, current
->pid
,
3185 generic_file_direct_write(struct kiocb
*iocb
, struct iov_iter
*from
)
3187 struct file
*file
= iocb
->ki_filp
;
3188 struct address_space
*mapping
= file
->f_mapping
;
3189 struct inode
*inode
= mapping
->host
;
3190 loff_t pos
= iocb
->ki_pos
;
3195 write_len
= iov_iter_count(from
);
3196 end
= (pos
+ write_len
- 1) >> PAGE_SHIFT
;
3198 if (iocb
->ki_flags
& IOCB_NOWAIT
) {
3199 /* If there are pages to writeback, return */
3200 if (filemap_range_has_page(inode
->i_mapping
, pos
,
3201 pos
+ write_len
- 1))
3204 written
= filemap_write_and_wait_range(mapping
, pos
,
3205 pos
+ write_len
- 1);
3211 * After a write we want buffered reads to be sure to go to disk to get
3212 * the new data. We invalidate clean cached page from the region we're
3213 * about to write. We do this *before* the write so that we can return
3214 * without clobbering -EIOCBQUEUED from ->direct_IO().
3216 written
= invalidate_inode_pages2_range(mapping
,
3217 pos
>> PAGE_SHIFT
, end
);
3219 * If a page can not be invalidated, return 0 to fall back
3220 * to buffered write.
3223 if (written
== -EBUSY
)
3228 written
= mapping
->a_ops
->direct_IO(iocb
, from
);
3231 * Finally, try again to invalidate clean pages which might have been
3232 * cached by non-direct readahead, or faulted in by get_user_pages()
3233 * if the source of the write was an mmap'ed region of the file
3234 * we're writing. Either one is a pretty crazy thing to do,
3235 * so we don't support it 100%. If this invalidation
3236 * fails, tough, the write still worked...
3238 * Most of the time we do not need this since dio_complete() will do
3239 * the invalidation for us. However there are some file systems that
3240 * do not end up with dio_complete() being called, so let's not break
3241 * them by removing it completely.
3243 * Noticeable example is a blkdev_direct_IO().
3245 * Skip invalidation for async writes or if mapping has no pages.
3247 if (written
> 0 && mapping
->nrpages
&&
3248 invalidate_inode_pages2_range(mapping
, pos
>> PAGE_SHIFT
, end
))
3249 dio_warn_stale_pagecache(file
);
3253 write_len
-= written
;
3254 if (pos
> i_size_read(inode
) && !S_ISBLK(inode
->i_mode
)) {
3255 i_size_write(inode
, pos
);
3256 mark_inode_dirty(inode
);
3260 iov_iter_revert(from
, write_len
- iov_iter_count(from
));
3264 EXPORT_SYMBOL(generic_file_direct_write
);
3267 * Find or create a page at the given pagecache position. Return the locked
3268 * page. This function is specifically for buffered writes.
3270 struct page
*grab_cache_page_write_begin(struct address_space
*mapping
,
3271 pgoff_t index
, unsigned flags
)
3274 int fgp_flags
= FGP_LOCK
|FGP_WRITE
|FGP_CREAT
;
3276 if (flags
& AOP_FLAG_NOFS
)
3277 fgp_flags
|= FGP_NOFS
;
3279 page
= pagecache_get_page(mapping
, index
, fgp_flags
,
3280 mapping_gfp_mask(mapping
));
3282 wait_for_stable_page(page
);
3286 EXPORT_SYMBOL(grab_cache_page_write_begin
);
3288 ssize_t
generic_perform_write(struct file
*file
,
3289 struct iov_iter
*i
, loff_t pos
)
3291 struct address_space
*mapping
= file
->f_mapping
;
3292 const struct address_space_operations
*a_ops
= mapping
->a_ops
;
3294 ssize_t written
= 0;
3295 unsigned int flags
= 0;
3299 unsigned long offset
; /* Offset into pagecache page */
3300 unsigned long bytes
; /* Bytes to write to page */
3301 size_t copied
; /* Bytes copied from user */
3304 offset
= (pos
& (PAGE_SIZE
- 1));
3305 bytes
= min_t(unsigned long, PAGE_SIZE
- offset
,
3310 * Bring in the user page that we will copy from _first_.
3311 * Otherwise there's a nasty deadlock on copying from the
3312 * same page as we're writing to, without it being marked
3315 * Not only is this an optimisation, but it is also required
3316 * to check that the address is actually valid, when atomic
3317 * usercopies are used, below.
3319 if (unlikely(iov_iter_fault_in_readable(i
, bytes
))) {
3324 if (fatal_signal_pending(current
)) {
3329 status
= a_ops
->write_begin(file
, mapping
, pos
, bytes
, flags
,
3331 if (unlikely(status
< 0))
3334 if (mapping_writably_mapped(mapping
))
3335 flush_dcache_page(page
);
3337 copied
= iov_iter_copy_from_user_atomic(page
, i
, offset
, bytes
);
3338 flush_dcache_page(page
);
3340 status
= a_ops
->write_end(file
, mapping
, pos
, bytes
, copied
,
3342 if (unlikely(status
< 0))
3348 iov_iter_advance(i
, copied
);
3349 if (unlikely(copied
== 0)) {
3351 * If we were unable to copy any data at all, we must
3352 * fall back to a single segment length write.
3354 * If we didn't fallback here, we could livelock
3355 * because not all segments in the iov can be copied at
3356 * once without a pagefault.
3358 bytes
= min_t(unsigned long, PAGE_SIZE
- offset
,
3359 iov_iter_single_seg_count(i
));
3365 balance_dirty_pages_ratelimited(mapping
);
3366 } while (iov_iter_count(i
));
3368 return written
? written
: status
;
3370 EXPORT_SYMBOL(generic_perform_write
);
3373 * __generic_file_write_iter - write data to a file
3374 * @iocb: IO state structure (file, offset, etc.)
3375 * @from: iov_iter with data to write
3377 * This function does all the work needed for actually writing data to a
3378 * file. It does all basic checks, removes SUID from the file, updates
3379 * modification times and calls proper subroutines depending on whether we
3380 * do direct IO or a standard buffered write.
3382 * It expects i_mutex to be grabbed unless we work on a block device or similar
3383 * object which does not need locking at all.
3385 * This function does *not* take care of syncing data in case of O_SYNC write.
3386 * A caller has to handle it. This is mainly due to the fact that we want to
3387 * avoid syncing under i_mutex.
3390 * * number of bytes written, even for truncated writes
3391 * * negative error code if no data has been written at all
3393 ssize_t
__generic_file_write_iter(struct kiocb
*iocb
, struct iov_iter
*from
)
3395 struct file
*file
= iocb
->ki_filp
;
3396 struct address_space
* mapping
= file
->f_mapping
;
3397 struct inode
*inode
= mapping
->host
;
3398 ssize_t written
= 0;
3402 /* We can write back this queue in page reclaim */
3403 current
->backing_dev_info
= inode_to_bdi(inode
);
3404 err
= file_remove_privs(file
);
3408 err
= file_update_time(file
);
3412 if (iocb
->ki_flags
& IOCB_DIRECT
) {
3413 loff_t pos
, endbyte
;
3415 written
= generic_file_direct_write(iocb
, from
);
3417 * If the write stopped short of completing, fall back to
3418 * buffered writes. Some filesystems do this for writes to
3419 * holes, for example. For DAX files, a buffered write will
3420 * not succeed (even if it did, DAX does not handle dirty
3421 * page-cache pages correctly).
3423 if (written
< 0 || !iov_iter_count(from
) || IS_DAX(inode
))
3426 status
= generic_perform_write(file
, from
, pos
= iocb
->ki_pos
);
3428 * If generic_perform_write() returned a synchronous error
3429 * then we want to return the number of bytes which were
3430 * direct-written, or the error code if that was zero. Note
3431 * that this differs from normal direct-io semantics, which
3432 * will return -EFOO even if some bytes were written.
3434 if (unlikely(status
< 0)) {
3439 * We need to ensure that the page cache pages are written to
3440 * disk and invalidated to preserve the expected O_DIRECT
3443 endbyte
= pos
+ status
- 1;
3444 err
= filemap_write_and_wait_range(mapping
, pos
, endbyte
);
3446 iocb
->ki_pos
= endbyte
+ 1;
3448 invalidate_mapping_pages(mapping
,
3450 endbyte
>> PAGE_SHIFT
);
3453 * We don't know how much we wrote, so just return
3454 * the number of bytes which were direct-written
3458 written
= generic_perform_write(file
, from
, iocb
->ki_pos
);
3459 if (likely(written
> 0))
3460 iocb
->ki_pos
+= written
;
3463 current
->backing_dev_info
= NULL
;
3464 return written
? written
: err
;
3466 EXPORT_SYMBOL(__generic_file_write_iter
);
3469 * generic_file_write_iter - write data to a file
3470 * @iocb: IO state structure
3471 * @from: iov_iter with data to write
3473 * This is a wrapper around __generic_file_write_iter() to be used by most
3474 * filesystems. It takes care of syncing the file in case of O_SYNC file
3475 * and acquires i_mutex as needed.
3477 * * negative error code if no data has been written at all of
3478 * vfs_fsync_range() failed for a synchronous write
3479 * * number of bytes written, even for truncated writes
3481 ssize_t
generic_file_write_iter(struct kiocb
*iocb
, struct iov_iter
*from
)
3483 struct file
*file
= iocb
->ki_filp
;
3484 struct inode
*inode
= file
->f_mapping
->host
;
3488 ret
= generic_write_checks(iocb
, from
);
3490 ret
= __generic_file_write_iter(iocb
, from
);
3491 inode_unlock(inode
);
3494 ret
= generic_write_sync(iocb
, ret
);
3497 EXPORT_SYMBOL(generic_file_write_iter
);
3500 * try_to_release_page() - release old fs-specific metadata on a page
3502 * @page: the page which the kernel is trying to free
3503 * @gfp_mask: memory allocation flags (and I/O mode)
3505 * The address_space is to try to release any data against the page
3506 * (presumably at page->private).
3508 * This may also be called if PG_fscache is set on a page, indicating that the
3509 * page is known to the local caching routines.
3511 * The @gfp_mask argument specifies whether I/O may be performed to release
3512 * this page (__GFP_IO), and whether the call may block (__GFP_RECLAIM & __GFP_FS).
3514 * Return: %1 if the release was successful, otherwise return zero.
3516 int try_to_release_page(struct page
*page
, gfp_t gfp_mask
)
3518 struct address_space
* const mapping
= page
->mapping
;
3520 BUG_ON(!PageLocked(page
));
3521 if (PageWriteback(page
))
3524 if (mapping
&& mapping
->a_ops
->releasepage
)
3525 return mapping
->a_ops
->releasepage(page
, gfp_mask
);
3526 return try_to_free_buffers(page
);
3529 EXPORT_SYMBOL(try_to_release_page
);