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>
45 #include <asm/pgalloc.h>
46 #include <asm/tlbflush.h>
49 #define CREATE_TRACE_POINTS
50 #include <trace/events/filemap.h>
53 * FIXME: remove all knowledge of the buffer layer from the core VM
55 #include <linux/buffer_head.h> /* for try_to_free_buffers */
60 * Shared mappings implemented 30.11.1994. It's not fully working yet,
63 * Shared mappings now work. 15.8.1995 Bruno.
65 * finished 'unifying' the page and buffer cache and SMP-threaded the
66 * page-cache, 21.05.1999, Ingo Molnar <mingo@redhat.com>
68 * SMP-threaded pagemap-LRU 1999, Andrea Arcangeli <andrea@suse.de>
74 * ->i_mmap_rwsem (truncate_pagecache)
75 * ->private_lock (__free_pte->__set_page_dirty_buffers)
76 * ->swap_lock (exclusive_swap_page, others)
80 * ->i_mmap_rwsem (truncate->unmap_mapping_range)
84 * ->page_table_lock or pte_lock (various, mainly in memory.c)
85 * ->i_pages lock (arch-dependent flush_dcache_mmap_lock)
88 * ->lock_page (access_process_vm)
90 * ->i_mutex (generic_perform_write)
91 * ->mmap_lock (fault_in_pages_readable->do_page_fault)
94 * sb_lock (fs/fs-writeback.c)
95 * ->i_pages lock (__sync_single_inode)
98 * ->anon_vma.lock (vma_adjust)
101 * ->page_table_lock or pte_lock (anon_vma_prepare and various)
103 * ->page_table_lock or pte_lock
104 * ->swap_lock (try_to_unmap_one)
105 * ->private_lock (try_to_unmap_one)
106 * ->i_pages lock (try_to_unmap_one)
107 * ->lruvec->lru_lock (follow_page->mark_page_accessed)
108 * ->lruvec->lru_lock (check_pte_range->isolate_lru_page)
109 * ->private_lock (page_remove_rmap->set_page_dirty)
110 * ->i_pages lock (page_remove_rmap->set_page_dirty)
111 * bdi.wb->list_lock (page_remove_rmap->set_page_dirty)
112 * ->inode->i_lock (page_remove_rmap->set_page_dirty)
113 * ->memcg->move_lock (page_remove_rmap->lock_page_memcg)
114 * bdi.wb->list_lock (zap_pte_range->set_page_dirty)
115 * ->inode->i_lock (zap_pte_range->set_page_dirty)
116 * ->private_lock (zap_pte_range->__set_page_dirty_buffers)
119 * ->tasklist_lock (memory_failure, collect_procs_ao)
122 static void page_cache_delete(struct address_space *mapping,
123 struct page *page, void *shadow)
125 XA_STATE(xas, &mapping->i_pages, page->index);
128 mapping_set_update(&xas, mapping);
130 /* hugetlb pages are represented by a single entry in the xarray */
131 if (!PageHuge(page)) {
132 xas_set_order(&xas, page->index, compound_order(page));
133 nr = compound_nr(page);
136 VM_BUG_ON_PAGE(!PageLocked(page), page);
137 VM_BUG_ON_PAGE(PageTail(page), page);
138 VM_BUG_ON_PAGE(nr != 1 && shadow, page);
140 xas_store(&xas, shadow);
141 xas_init_marks(&xas);
143 page->mapping = NULL;
144 /* Leave page->index set: truncation lookup relies upon it */
147 mapping->nrexceptional += nr;
149 * Make sure the nrexceptional update is committed before
150 * the nrpages update so that final truncate racing
151 * with reclaim does not see both counters 0 at the
152 * same time and miss a shadow entry.
156 mapping->nrpages -= nr;
159 static void unaccount_page_cache_page(struct address_space *mapping,
165 * if we're uptodate, flush out into the cleancache, otherwise
166 * invalidate any existing cleancache entries. We can't leave
167 * stale data around in the cleancache once our page is gone
169 if (PageUptodate(page) && PageMappedToDisk(page))
170 cleancache_put_page(page);
172 cleancache_invalidate_page(mapping, page);
174 VM_BUG_ON_PAGE(PageTail(page), page);
175 VM_BUG_ON_PAGE(page_mapped(page), page);
176 if (!IS_ENABLED(CONFIG_DEBUG_VM) && unlikely(page_mapped(page))) {
179 pr_alert("BUG: Bad page cache in process %s pfn:%05lx\n",
180 current->comm, page_to_pfn(page));
181 dump_page(page, "still mapped when deleted");
183 add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE);
185 mapcount = page_mapcount(page);
186 if (mapping_exiting(mapping) &&
187 page_count(page) >= mapcount + 2) {
189 * All vmas have already been torn down, so it's
190 * a good bet that actually the page is unmapped,
191 * and we'd prefer not to leak it: if we're wrong,
192 * some other bad page check should catch it later.
194 page_mapcount_reset(page);
195 page_ref_sub(page, mapcount);
199 /* hugetlb pages do not participate in page cache accounting. */
203 nr = thp_nr_pages(page);
205 __mod_lruvec_page_state(page, NR_FILE_PAGES, -nr);
206 if (PageSwapBacked(page)) {
207 __mod_lruvec_page_state(page, NR_SHMEM, -nr);
208 if (PageTransHuge(page))
209 __mod_lruvec_page_state(page, NR_SHMEM_THPS, -nr);
210 } else if (PageTransHuge(page)) {
211 __mod_lruvec_page_state(page, NR_FILE_THPS, -nr);
212 filemap_nr_thps_dec(mapping);
216 * At this point page must be either written or cleaned by
217 * truncate. Dirty page here signals a bug and loss of
220 * This fixes dirty accounting after removing the page entirely
221 * but leaves PageDirty set: it has no effect for truncated
222 * page and anyway will be cleared before returning page into
225 if (WARN_ON_ONCE(PageDirty(page)))
226 account_page_cleaned(page, mapping, inode_to_wb(mapping->host));
230 * Delete a page from the page cache and free it. Caller has to make
231 * sure the page is locked and that nobody else uses it - or that usage
232 * is safe. The caller must hold the i_pages lock.
234 void __delete_from_page_cache(struct page *page, void *shadow)
236 struct address_space *mapping = page->mapping;
238 trace_mm_filemap_delete_from_page_cache(page);
240 unaccount_page_cache_page(mapping, page);
241 page_cache_delete(mapping, page, shadow);
244 static void page_cache_free_page(struct address_space *mapping,
247 void (*freepage)(struct page *);
249 freepage = mapping->a_ops->freepage;
253 if (PageTransHuge(page) && !PageHuge(page)) {
254 page_ref_sub(page, thp_nr_pages(page));
255 VM_BUG_ON_PAGE(page_count(page) <= 0, page);
262 * delete_from_page_cache - delete page from page cache
263 * @page: the page which the kernel is trying to remove from page cache
265 * This must be called only on pages that have been verified to be in the page
266 * cache and locked. It will never put the page into the free list, the caller
267 * has a reference on the page.
269 void delete_from_page_cache(struct page *page)
271 struct address_space *mapping = page_mapping(page);
274 BUG_ON(!PageLocked(page));
275 xa_lock_irqsave(&mapping->i_pages, flags);
276 __delete_from_page_cache(page, NULL);
277 xa_unlock_irqrestore(&mapping->i_pages, flags);
279 page_cache_free_page(mapping, page);
281 EXPORT_SYMBOL(delete_from_page_cache);
284 * page_cache_delete_batch - delete several pages from page cache
285 * @mapping: the mapping to which pages belong
286 * @pvec: pagevec with pages to delete
288 * The function walks over mapping->i_pages and removes pages passed in @pvec
289 * from the mapping. The function expects @pvec to be sorted by page index
290 * and is optimised for it to be dense.
291 * It tolerates holes in @pvec (mapping entries at those indices are not
292 * modified). The function expects only THP head pages to be present in the
295 * The function expects the i_pages lock to be held.
297 static void page_cache_delete_batch(struct address_space *mapping,
298 struct pagevec *pvec)
300 XA_STATE(xas, &mapping->i_pages, pvec->pages[0]->index);
305 mapping_set_update(&xas, mapping);
306 xas_for_each(&xas, page, ULONG_MAX) {
307 if (i >= pagevec_count(pvec))
310 /* A swap/dax/shadow entry got inserted? Skip it. */
311 if (xa_is_value(page))
314 * A page got inserted in our range? Skip it. We have our
315 * pages locked so they are protected from being removed.
316 * If we see a page whose index is higher than ours, it
317 * means our page has been removed, which shouldn't be
318 * possible because we're holding the PageLock.
320 if (page != pvec->pages[i]) {
321 VM_BUG_ON_PAGE(page->index > pvec->pages[i]->index,
326 WARN_ON_ONCE(!PageLocked(page));
328 if (page->index == xas.xa_index)
329 page->mapping = NULL;
330 /* Leave page->index set: truncation lookup relies on it */
333 * Move to the next page in the vector if this is a regular
334 * page or the index is of the last sub-page of this compound
337 if (page->index + compound_nr(page) - 1 == xas.xa_index)
339 xas_store(&xas, NULL);
342 mapping->nrpages -= total_pages;
345 void delete_from_page_cache_batch(struct address_space *mapping,
346 struct pagevec *pvec)
351 if (!pagevec_count(pvec))
354 xa_lock_irqsave(&mapping->i_pages, flags);
355 for (i = 0; i < pagevec_count(pvec); i++) {
356 trace_mm_filemap_delete_from_page_cache(pvec->pages[i]);
358 unaccount_page_cache_page(mapping, pvec->pages[i]);
360 page_cache_delete_batch(mapping, pvec);
361 xa_unlock_irqrestore(&mapping->i_pages, flags);
363 for (i = 0; i < pagevec_count(pvec); i++)
364 page_cache_free_page(mapping, pvec->pages[i]);
367 int filemap_check_errors(struct address_space *mapping)
370 /* Check for outstanding write errors */
371 if (test_bit(AS_ENOSPC, &mapping->flags) &&
372 test_and_clear_bit(AS_ENOSPC, &mapping->flags))
374 if (test_bit(AS_EIO, &mapping->flags) &&
375 test_and_clear_bit(AS_EIO, &mapping->flags))
379 EXPORT_SYMBOL(filemap_check_errors);
381 static int filemap_check_and_keep_errors(struct address_space *mapping)
383 /* Check for outstanding write errors */
384 if (test_bit(AS_EIO, &mapping->flags))
386 if (test_bit(AS_ENOSPC, &mapping->flags))
392 * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range
393 * @mapping: address space structure to write
394 * @start: offset in bytes where the range starts
395 * @end: offset in bytes where the range ends (inclusive)
396 * @sync_mode: enable synchronous operation
398 * Start writeback against all of a mapping's dirty pages that lie
399 * within the byte offsets <start, end> inclusive.
401 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
402 * opposed to a regular memory cleansing writeback. The difference between
403 * these two operations is that if a dirty page/buffer is encountered, it must
404 * be waited upon, and not just skipped over.
406 * Return: %0 on success, negative error code otherwise.
408 int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
409 loff_t end, int sync_mode)
412 struct writeback_control wbc = {
413 .sync_mode = sync_mode,
414 .nr_to_write = LONG_MAX,
415 .range_start = start,
419 if (!mapping_can_writeback(mapping) ||
420 !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
423 wbc_attach_fdatawrite_inode(&wbc, mapping->host);
424 ret = do_writepages(mapping, &wbc);
425 wbc_detach_inode(&wbc);
429 static inline int __filemap_fdatawrite(struct address_space *mapping,
432 return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode);
435 int filemap_fdatawrite(struct address_space *mapping)
437 return __filemap_fdatawrite(mapping, WB_SYNC_ALL);
439 EXPORT_SYMBOL(filemap_fdatawrite);
441 int filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
444 return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL);
446 EXPORT_SYMBOL(filemap_fdatawrite_range);
449 * filemap_flush - mostly a non-blocking flush
450 * @mapping: target address_space
452 * This is a mostly non-blocking flush. Not suitable for data-integrity
453 * purposes - I/O may not be started against all dirty pages.
455 * Return: %0 on success, negative error code otherwise.
457 int filemap_flush(struct address_space *mapping)
459 return __filemap_fdatawrite(mapping, WB_SYNC_NONE);
461 EXPORT_SYMBOL(filemap_flush);
464 * filemap_range_has_page - check if a page exists in range.
465 * @mapping: address space within which to check
466 * @start_byte: offset in bytes where the range starts
467 * @end_byte: offset in bytes where the range ends (inclusive)
469 * Find at least one page in the range supplied, usually used to check if
470 * direct writing in this range will trigger a writeback.
472 * Return: %true if at least one page exists in the specified range,
475 bool filemap_range_has_page(struct address_space *mapping,
476 loff_t start_byte, loff_t end_byte)
479 XA_STATE(xas, &mapping->i_pages, start_byte >> PAGE_SHIFT);
480 pgoff_t max = end_byte >> PAGE_SHIFT;
482 if (end_byte < start_byte)
487 page = xas_find(&xas, max);
488 if (xas_retry(&xas, page))
490 /* Shadow entries don't count */
491 if (xa_is_value(page))
494 * We don't need to try to pin this page; we're about to
495 * release the RCU lock anyway. It is enough to know that
496 * there was a page here recently.
504 EXPORT_SYMBOL(filemap_range_has_page);
506 static void __filemap_fdatawait_range(struct address_space *mapping,
507 loff_t start_byte, loff_t end_byte)
509 pgoff_t index = start_byte >> PAGE_SHIFT;
510 pgoff_t end = end_byte >> PAGE_SHIFT;
514 if (end_byte < start_byte)
518 while (index <= end) {
521 nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index,
522 end, PAGECACHE_TAG_WRITEBACK);
526 for (i = 0; i < nr_pages; i++) {
527 struct page *page = pvec.pages[i];
529 wait_on_page_writeback(page);
530 ClearPageError(page);
532 pagevec_release(&pvec);
538 * filemap_fdatawait_range - wait for writeback to complete
539 * @mapping: address space structure to wait for
540 * @start_byte: offset in bytes where the range starts
541 * @end_byte: offset in bytes where the range ends (inclusive)
543 * Walk the list of under-writeback pages of the given address space
544 * in the given range and wait for all of them. Check error status of
545 * the address space and return it.
547 * Since the error status of the address space is cleared by this function,
548 * callers are responsible for checking the return value and handling and/or
549 * reporting the error.
551 * Return: error status of the address space.
553 int filemap_fdatawait_range(struct address_space *mapping, loff_t start_byte,
556 __filemap_fdatawait_range(mapping, start_byte, end_byte);
557 return filemap_check_errors(mapping);
559 EXPORT_SYMBOL(filemap_fdatawait_range);
562 * filemap_fdatawait_range_keep_errors - wait for writeback to complete
563 * @mapping: address space structure to wait for
564 * @start_byte: offset in bytes where the range starts
565 * @end_byte: offset in bytes where the range ends (inclusive)
567 * Walk the list of under-writeback pages of the given address space in the
568 * given range and wait for all of them. Unlike filemap_fdatawait_range(),
569 * this function does not clear error status of the address space.
571 * Use this function if callers don't handle errors themselves. Expected
572 * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2),
575 int filemap_fdatawait_range_keep_errors(struct address_space *mapping,
576 loff_t start_byte, loff_t end_byte)
578 __filemap_fdatawait_range(mapping, start_byte, end_byte);
579 return filemap_check_and_keep_errors(mapping);
581 EXPORT_SYMBOL(filemap_fdatawait_range_keep_errors);
584 * file_fdatawait_range - wait for writeback to complete
585 * @file: file pointing to address space structure to wait for
586 * @start_byte: offset in bytes where the range starts
587 * @end_byte: offset in bytes where the range ends (inclusive)
589 * Walk the list of under-writeback pages of the address space that file
590 * refers to, in the given range and wait for all of them. Check error
591 * status of the address space vs. the file->f_wb_err cursor and return it.
593 * Since the error status of the file is advanced by this function,
594 * callers are responsible for checking the return value and handling and/or
595 * reporting the error.
597 * Return: error status of the address space vs. the file->f_wb_err cursor.
599 int file_fdatawait_range(struct file *file, loff_t start_byte, loff_t end_byte)
601 struct address_space *mapping = file->f_mapping;
603 __filemap_fdatawait_range(mapping, start_byte, end_byte);
604 return file_check_and_advance_wb_err(file);
606 EXPORT_SYMBOL(file_fdatawait_range);
609 * filemap_fdatawait_keep_errors - wait for writeback without clearing errors
610 * @mapping: address space structure to wait for
612 * Walk the list of under-writeback pages of the given address space
613 * and wait for all of them. Unlike filemap_fdatawait(), this function
614 * does not clear error status of the address space.
616 * Use this function if callers don't handle errors themselves. Expected
617 * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2),
620 * Return: error status of the address space.
622 int filemap_fdatawait_keep_errors(struct address_space *mapping)
624 __filemap_fdatawait_range(mapping, 0, LLONG_MAX);
625 return filemap_check_and_keep_errors(mapping);
627 EXPORT_SYMBOL(filemap_fdatawait_keep_errors);
629 /* Returns true if writeback might be needed or already in progress. */
630 static bool mapping_needs_writeback(struct address_space *mapping)
632 if (dax_mapping(mapping))
633 return mapping->nrexceptional;
635 return mapping->nrpages;
639 * filemap_range_needs_writeback - check if range potentially needs writeback
640 * @mapping: address space within which to check
641 * @start_byte: offset in bytes where the range starts
642 * @end_byte: offset in bytes where the range ends (inclusive)
644 * Find at least one page in the range supplied, usually used to check if
645 * direct writing in this range will trigger a writeback. Used by O_DIRECT
646 * read/write with IOCB_NOWAIT, to see if the caller needs to do
647 * filemap_write_and_wait_range() before proceeding.
649 * Return: %true if the caller should do filemap_write_and_wait_range() before
650 * doing O_DIRECT to a page in this range, %false otherwise.
652 bool filemap_range_needs_writeback(struct address_space *mapping,
653 loff_t start_byte, loff_t end_byte)
655 XA_STATE(xas, &mapping->i_pages, start_byte >> PAGE_SHIFT);
656 pgoff_t max = end_byte >> PAGE_SHIFT;
659 if (!mapping_needs_writeback(mapping))
661 if (!mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) &&
662 !mapping_tagged(mapping, PAGECACHE_TAG_WRITEBACK))
664 if (end_byte < start_byte)
668 xas_for_each(&xas, page, max) {
669 if (xas_retry(&xas, page))
671 if (xa_is_value(page))
673 if (PageDirty(page) || PageLocked(page) || PageWriteback(page))
679 EXPORT_SYMBOL_GPL(filemap_range_needs_writeback);
682 * filemap_write_and_wait_range - write out & wait on a file range
683 * @mapping: the address_space for the pages
684 * @lstart: offset in bytes where the range starts
685 * @lend: offset in bytes where the range ends (inclusive)
687 * Write out and wait upon file offsets lstart->lend, inclusive.
689 * Note that @lend is inclusive (describes the last byte to be written) so
690 * that this function can be used to write to the very end-of-file (end = -1).
692 * Return: error status of the address space.
694 int filemap_write_and_wait_range(struct address_space *mapping,
695 loff_t lstart, loff_t lend)
699 if (mapping_needs_writeback(mapping)) {
700 err = __filemap_fdatawrite_range(mapping, lstart, lend,
703 * Even if the above returned error, the pages may be
704 * written partially (e.g. -ENOSPC), so we wait for it.
705 * But the -EIO is special case, it may indicate the worst
706 * thing (e.g. bug) happened, so we avoid waiting for it.
709 int err2 = filemap_fdatawait_range(mapping,
714 /* Clear any previously stored errors */
715 filemap_check_errors(mapping);
718 err = filemap_check_errors(mapping);
722 EXPORT_SYMBOL(filemap_write_and_wait_range);
724 void __filemap_set_wb_err(struct address_space *mapping, int err)
726 errseq_t eseq = errseq_set(&mapping->wb_err, err);
728 trace_filemap_set_wb_err(mapping, eseq);
730 EXPORT_SYMBOL(__filemap_set_wb_err);
733 * file_check_and_advance_wb_err - report wb error (if any) that was previously
734 * and advance wb_err to current one
735 * @file: struct file on which the error is being reported
737 * When userland calls fsync (or something like nfsd does the equivalent), we
738 * want to report any writeback errors that occurred since the last fsync (or
739 * since the file was opened if there haven't been any).
741 * Grab the wb_err from the mapping. If it matches what we have in the file,
742 * then just quickly return 0. The file is all caught up.
744 * If it doesn't match, then take the mapping value, set the "seen" flag in
745 * it and try to swap it into place. If it works, or another task beat us
746 * to it with the new value, then update the f_wb_err and return the error
747 * portion. The error at this point must be reported via proper channels
748 * (a'la fsync, or NFS COMMIT operation, etc.).
750 * While we handle mapping->wb_err with atomic operations, the f_wb_err
751 * value is protected by the f_lock since we must ensure that it reflects
752 * the latest value swapped in for this file descriptor.
754 * Return: %0 on success, negative error code otherwise.
756 int file_check_and_advance_wb_err(struct file *file)
759 errseq_t old = READ_ONCE(file->f_wb_err);
760 struct address_space *mapping = file->f_mapping;
762 /* Locklessly handle the common case where nothing has changed */
763 if (errseq_check(&mapping->wb_err, old)) {
764 /* Something changed, must use slow path */
765 spin_lock(&file->f_lock);
766 old = file->f_wb_err;
767 err = errseq_check_and_advance(&mapping->wb_err,
769 trace_file_check_and_advance_wb_err(file, old);
770 spin_unlock(&file->f_lock);
774 * We're mostly using this function as a drop in replacement for
775 * filemap_check_errors. Clear AS_EIO/AS_ENOSPC to emulate the effect
776 * that the legacy code would have had on these flags.
778 clear_bit(AS_EIO, &mapping->flags);
779 clear_bit(AS_ENOSPC, &mapping->flags);
782 EXPORT_SYMBOL(file_check_and_advance_wb_err);
785 * file_write_and_wait_range - write out & wait on a file range
786 * @file: file pointing to address_space with pages
787 * @lstart: offset in bytes where the range starts
788 * @lend: offset in bytes where the range ends (inclusive)
790 * Write out and wait upon file offsets lstart->lend, inclusive.
792 * Note that @lend is inclusive (describes the last byte to be written) so
793 * that this function can be used to write to the very end-of-file (end = -1).
795 * After writing out and waiting on the data, we check and advance the
796 * f_wb_err cursor to the latest value, and return any errors detected there.
798 * Return: %0 on success, negative error code otherwise.
800 int file_write_and_wait_range(struct file *file, loff_t lstart, loff_t lend)
803 struct address_space *mapping = file->f_mapping;
805 if (mapping_needs_writeback(mapping)) {
806 err = __filemap_fdatawrite_range(mapping, lstart, lend,
808 /* See comment of filemap_write_and_wait() */
810 __filemap_fdatawait_range(mapping, lstart, lend);
812 err2 = file_check_and_advance_wb_err(file);
817 EXPORT_SYMBOL(file_write_and_wait_range);
820 * replace_page_cache_page - replace a pagecache page with a new one
821 * @old: page to be replaced
822 * @new: page to replace with
824 * This function replaces a page in the pagecache with a new one. On
825 * success it acquires the pagecache reference for the new page and
826 * drops it for the old page. Both the old and new pages must be
827 * locked. This function does not add the new page to the LRU, the
828 * caller must do that.
830 * The remove + add is atomic. This function cannot fail.
832 void replace_page_cache_page(struct page *old, struct page *new)
834 struct address_space *mapping = old->mapping;
835 void (*freepage)(struct page *) = mapping->a_ops->freepage;
836 pgoff_t offset = old->index;
837 XA_STATE(xas, &mapping->i_pages, offset);
840 VM_BUG_ON_PAGE(!PageLocked(old), old);
841 VM_BUG_ON_PAGE(!PageLocked(new), new);
842 VM_BUG_ON_PAGE(new->mapping, new);
845 new->mapping = mapping;
848 mem_cgroup_migrate(old, new);
850 xas_lock_irqsave(&xas, flags);
851 xas_store(&xas, new);
854 /* hugetlb pages do not participate in page cache accounting. */
856 __dec_lruvec_page_state(old, NR_FILE_PAGES);
858 __inc_lruvec_page_state(new, NR_FILE_PAGES);
859 if (PageSwapBacked(old))
860 __dec_lruvec_page_state(old, NR_SHMEM);
861 if (PageSwapBacked(new))
862 __inc_lruvec_page_state(new, NR_SHMEM);
863 xas_unlock_irqrestore(&xas, flags);
868 EXPORT_SYMBOL_GPL(replace_page_cache_page);
870 noinline int __add_to_page_cache_locked(struct page *page,
871 struct address_space *mapping,
872 pgoff_t offset, gfp_t gfp,
875 XA_STATE(xas, &mapping->i_pages, offset);
876 int huge = PageHuge(page);
878 bool charged = false;
880 VM_BUG_ON_PAGE(!PageLocked(page), page);
881 VM_BUG_ON_PAGE(PageSwapBacked(page), page);
882 mapping_set_update(&xas, mapping);
885 page->mapping = mapping;
886 page->index = offset;
889 error = mem_cgroup_charge(page, current->mm, gfp);
895 gfp &= GFP_RECLAIM_MASK;
898 unsigned int order = xa_get_order(xas.xa, xas.xa_index);
899 void *entry, *old = NULL;
901 if (order > thp_order(page))
902 xas_split_alloc(&xas, xa_load(xas.xa, xas.xa_index),
905 xas_for_each_conflict(&xas, entry) {
907 if (!xa_is_value(entry)) {
908 xas_set_err(&xas, -EEXIST);
916 /* entry may have been split before we acquired lock */
917 order = xa_get_order(xas.xa, xas.xa_index);
918 if (order > thp_order(page)) {
919 xas_split(&xas, old, order);
924 xas_store(&xas, page);
929 mapping->nrexceptional--;
932 /* hugetlb pages do not participate in page cache accounting */
934 __inc_lruvec_page_state(page, NR_FILE_PAGES);
936 xas_unlock_irq(&xas);
937 } while (xas_nomem(&xas, gfp));
939 if (xas_error(&xas)) {
940 error = xas_error(&xas);
942 mem_cgroup_uncharge(page);
946 trace_mm_filemap_add_to_page_cache(page);
949 page->mapping = NULL;
950 /* Leave page->index set: truncation relies upon it */
954 ALLOW_ERROR_INJECTION(__add_to_page_cache_locked, ERRNO);
957 * add_to_page_cache_locked - add a locked page to the pagecache
959 * @mapping: the page's address_space
960 * @offset: page index
961 * @gfp_mask: page allocation mode
963 * This function is used to add a page to the pagecache. It must be locked.
964 * This function does not add the page to the LRU. The caller must do that.
966 * Return: %0 on success, negative error code otherwise.
968 int add_to_page_cache_locked(struct page *page, struct address_space *mapping,
969 pgoff_t offset, gfp_t gfp_mask)
971 return __add_to_page_cache_locked(page, mapping, offset,
974 EXPORT_SYMBOL(add_to_page_cache_locked);
976 int add_to_page_cache_lru(struct page *page, struct address_space *mapping,
977 pgoff_t offset, gfp_t gfp_mask)
982 __SetPageLocked(page);
983 ret = __add_to_page_cache_locked(page, mapping, offset,
986 __ClearPageLocked(page);
989 * The page might have been evicted from cache only
990 * recently, in which case it should be activated like
991 * any other repeatedly accessed page.
992 * The exception is pages getting rewritten; evicting other
993 * data from the working set, only to cache data that will
994 * get overwritten with something else, is a waste of memory.
996 WARN_ON_ONCE(PageActive(page));
997 if (!(gfp_mask & __GFP_WRITE) && shadow)
998 workingset_refault(page, shadow);
1003 EXPORT_SYMBOL_GPL(add_to_page_cache_lru);
1006 struct page *__page_cache_alloc(gfp_t gfp)
1011 if (cpuset_do_page_mem_spread()) {
1012 unsigned int cpuset_mems_cookie;
1014 cpuset_mems_cookie = read_mems_allowed_begin();
1015 n = cpuset_mem_spread_node();
1016 page = __alloc_pages_node(n, gfp, 0);
1017 } while (!page && read_mems_allowed_retry(cpuset_mems_cookie));
1021 return alloc_pages(gfp, 0);
1023 EXPORT_SYMBOL(__page_cache_alloc);
1027 * In order to wait for pages to become available there must be
1028 * waitqueues associated with pages. By using a hash table of
1029 * waitqueues where the bucket discipline is to maintain all
1030 * waiters on the same queue and wake all when any of the pages
1031 * become available, and for the woken contexts to check to be
1032 * sure the appropriate page became available, this saves space
1033 * at a cost of "thundering herd" phenomena during rare hash
1036 #define PAGE_WAIT_TABLE_BITS 8
1037 #define PAGE_WAIT_TABLE_SIZE (1 << PAGE_WAIT_TABLE_BITS)
1038 static wait_queue_head_t page_wait_table[PAGE_WAIT_TABLE_SIZE] __cacheline_aligned;
1040 static wait_queue_head_t *page_waitqueue(struct page *page)
1042 return &page_wait_table[hash_ptr(page, PAGE_WAIT_TABLE_BITS)];
1045 void __init pagecache_init(void)
1049 for (i = 0; i < PAGE_WAIT_TABLE_SIZE; i++)
1050 init_waitqueue_head(&page_wait_table[i]);
1052 page_writeback_init();
1056 * The page wait code treats the "wait->flags" somewhat unusually, because
1057 * we have multiple different kinds of waits, not just the usual "exclusive"
1062 * (a) no special bits set:
1064 * We're just waiting for the bit to be released, and when a waker
1065 * calls the wakeup function, we set WQ_FLAG_WOKEN and wake it up,
1066 * and remove it from the wait queue.
1068 * Simple and straightforward.
1070 * (b) WQ_FLAG_EXCLUSIVE:
1072 * The waiter is waiting to get the lock, and only one waiter should
1073 * be woken up to avoid any thundering herd behavior. We'll set the
1074 * WQ_FLAG_WOKEN bit, wake it up, and remove it from the wait queue.
1076 * This is the traditional exclusive wait.
1078 * (c) WQ_FLAG_EXCLUSIVE | WQ_FLAG_CUSTOM:
1080 * The waiter is waiting to get the bit, and additionally wants the
1081 * lock to be transferred to it for fair lock behavior. If the lock
1082 * cannot be taken, we stop walking the wait queue without waking
1085 * This is the "fair lock handoff" case, and in addition to setting
1086 * WQ_FLAG_WOKEN, we set WQ_FLAG_DONE to let the waiter easily see
1087 * that it now has the lock.
1089 static int wake_page_function(wait_queue_entry_t *wait, unsigned mode, int sync, void *arg)
1092 struct wait_page_key *key = arg;
1093 struct wait_page_queue *wait_page
1094 = container_of(wait, struct wait_page_queue, wait);
1096 if (!wake_page_match(wait_page, key))
1100 * If it's a lock handoff wait, we get the bit for it, and
1101 * stop walking (and do not wake it up) if we can't.
1103 flags = wait->flags;
1104 if (flags & WQ_FLAG_EXCLUSIVE) {
1105 if (test_bit(key->bit_nr, &key->page->flags))
1107 if (flags & WQ_FLAG_CUSTOM) {
1108 if (test_and_set_bit(key->bit_nr, &key->page->flags))
1110 flags |= WQ_FLAG_DONE;
1115 * We are holding the wait-queue lock, but the waiter that
1116 * is waiting for this will be checking the flags without
1119 * So update the flags atomically, and wake up the waiter
1120 * afterwards to avoid any races. This store-release pairs
1121 * with the load-acquire in wait_on_page_bit_common().
1123 smp_store_release(&wait->flags, flags | WQ_FLAG_WOKEN);
1124 wake_up_state(wait->private, mode);
1127 * Ok, we have successfully done what we're waiting for,
1128 * and we can unconditionally remove the wait entry.
1130 * Note that this pairs with the "finish_wait()" in the
1131 * waiter, and has to be the absolute last thing we do.
1132 * After this list_del_init(&wait->entry) the wait entry
1133 * might be de-allocated and the process might even have
1136 list_del_init_careful(&wait->entry);
1137 return (flags & WQ_FLAG_EXCLUSIVE) != 0;
1140 static void wake_up_page_bit(struct page *page, int bit_nr)
1142 wait_queue_head_t *q = page_waitqueue(page);
1143 struct wait_page_key key;
1144 unsigned long flags;
1145 wait_queue_entry_t bookmark;
1148 key.bit_nr = bit_nr;
1152 bookmark.private = NULL;
1153 bookmark.func = NULL;
1154 INIT_LIST_HEAD(&bookmark.entry);
1156 spin_lock_irqsave(&q->lock, flags);
1157 __wake_up_locked_key_bookmark(q, TASK_NORMAL, &key, &bookmark);
1159 while (bookmark.flags & WQ_FLAG_BOOKMARK) {
1161 * Take a breather from holding the lock,
1162 * allow pages that finish wake up asynchronously
1163 * to acquire the lock and remove themselves
1166 spin_unlock_irqrestore(&q->lock, flags);
1168 spin_lock_irqsave(&q->lock, flags);
1169 __wake_up_locked_key_bookmark(q, TASK_NORMAL, &key, &bookmark);
1173 * It is possible for other pages to have collided on the waitqueue
1174 * hash, so in that case check for a page match. That prevents a long-
1177 * It is still possible to miss a case here, when we woke page waiters
1178 * and removed them from the waitqueue, but there are still other
1181 if (!waitqueue_active(q) || !key.page_match) {
1182 ClearPageWaiters(page);
1184 * It's possible to miss clearing Waiters here, when we woke
1185 * our page waiters, but the hashed waitqueue has waiters for
1186 * other pages on it.
1188 * That's okay, it's a rare case. The next waker will clear it.
1191 spin_unlock_irqrestore(&q->lock, flags);
1194 static void wake_up_page(struct page *page, int bit)
1196 if (!PageWaiters(page))
1198 wake_up_page_bit(page, bit);
1202 * A choice of three behaviors for wait_on_page_bit_common():
1205 EXCLUSIVE, /* Hold ref to page and take the bit when woken, like
1206 * __lock_page() waiting on then setting PG_locked.
1208 SHARED, /* Hold ref to page and check the bit when woken, like
1209 * wait_on_page_writeback() waiting on PG_writeback.
1211 DROP, /* Drop ref to page before wait, no check when woken,
1212 * like put_and_wait_on_page_locked() on PG_locked.
1217 * Attempt to check (or get) the page bit, and mark us done
1220 static inline bool trylock_page_bit_common(struct page *page, int bit_nr,
1221 struct wait_queue_entry *wait)
1223 if (wait->flags & WQ_FLAG_EXCLUSIVE) {
1224 if (test_and_set_bit(bit_nr, &page->flags))
1226 } else if (test_bit(bit_nr, &page->flags))
1229 wait->flags |= WQ_FLAG_WOKEN | WQ_FLAG_DONE;
1233 /* How many times do we accept lock stealing from under a waiter? */
1234 int sysctl_page_lock_unfairness = 5;
1236 static inline int wait_on_page_bit_common(wait_queue_head_t *q,
1237 struct page *page, int bit_nr, int state, enum behavior behavior)
1239 int unfairness = sysctl_page_lock_unfairness;
1240 struct wait_page_queue wait_page;
1241 wait_queue_entry_t *wait = &wait_page.wait;
1242 bool thrashing = false;
1243 bool delayacct = false;
1244 unsigned long pflags;
1246 if (bit_nr == PG_locked &&
1247 !PageUptodate(page) && PageWorkingset(page)) {
1248 if (!PageSwapBacked(page)) {
1249 delayacct_thrashing_start();
1252 psi_memstall_enter(&pflags);
1257 wait->func = wake_page_function;
1258 wait_page.page = page;
1259 wait_page.bit_nr = bit_nr;
1263 if (behavior == EXCLUSIVE) {
1264 wait->flags = WQ_FLAG_EXCLUSIVE;
1265 if (--unfairness < 0)
1266 wait->flags |= WQ_FLAG_CUSTOM;
1270 * Do one last check whether we can get the
1271 * page bit synchronously.
1273 * Do the SetPageWaiters() marking before that
1274 * to let any waker we _just_ missed know they
1275 * need to wake us up (otherwise they'll never
1276 * even go to the slow case that looks at the
1277 * page queue), and add ourselves to the wait
1278 * queue if we need to sleep.
1280 * This part needs to be done under the queue
1281 * lock to avoid races.
1283 spin_lock_irq(&q->lock);
1284 SetPageWaiters(page);
1285 if (!trylock_page_bit_common(page, bit_nr, wait))
1286 __add_wait_queue_entry_tail(q, wait);
1287 spin_unlock_irq(&q->lock);
1290 * From now on, all the logic will be based on
1291 * the WQ_FLAG_WOKEN and WQ_FLAG_DONE flag, to
1292 * see whether the page bit testing has already
1293 * been done by the wake function.
1295 * We can drop our reference to the page.
1297 if (behavior == DROP)
1301 * Note that until the "finish_wait()", or until
1302 * we see the WQ_FLAG_WOKEN flag, we need to
1303 * be very careful with the 'wait->flags', because
1304 * we may race with a waker that sets them.
1309 set_current_state(state);
1311 /* Loop until we've been woken or interrupted */
1312 flags = smp_load_acquire(&wait->flags);
1313 if (!(flags & WQ_FLAG_WOKEN)) {
1314 if (signal_pending_state(state, current))
1321 /* If we were non-exclusive, we're done */
1322 if (behavior != EXCLUSIVE)
1325 /* If the waker got the lock for us, we're done */
1326 if (flags & WQ_FLAG_DONE)
1330 * Otherwise, if we're getting the lock, we need to
1331 * try to get it ourselves.
1333 * And if that fails, we'll have to retry this all.
1335 if (unlikely(test_and_set_bit(bit_nr, &page->flags)))
1338 wait->flags |= WQ_FLAG_DONE;
1343 * If a signal happened, this 'finish_wait()' may remove the last
1344 * waiter from the wait-queues, but the PageWaiters bit will remain
1345 * set. That's ok. The next wakeup will take care of it, and trying
1346 * to do it here would be difficult and prone to races.
1348 finish_wait(q, wait);
1352 delayacct_thrashing_end();
1353 psi_memstall_leave(&pflags);
1357 * NOTE! The wait->flags weren't stable until we've done the
1358 * 'finish_wait()', and we could have exited the loop above due
1359 * to a signal, and had a wakeup event happen after the signal
1360 * test but before the 'finish_wait()'.
1362 * So only after the finish_wait() can we reliably determine
1363 * if we got woken up or not, so we can now figure out the final
1364 * return value based on that state without races.
1366 * Also note that WQ_FLAG_WOKEN is sufficient for a non-exclusive
1367 * waiter, but an exclusive one requires WQ_FLAG_DONE.
1369 if (behavior == EXCLUSIVE)
1370 return wait->flags & WQ_FLAG_DONE ? 0 : -EINTR;
1372 return wait->flags & WQ_FLAG_WOKEN ? 0 : -EINTR;
1375 void wait_on_page_bit(struct page *page, int bit_nr)
1377 wait_queue_head_t *q = page_waitqueue(page);
1378 wait_on_page_bit_common(q, page, bit_nr, TASK_UNINTERRUPTIBLE, SHARED);
1380 EXPORT_SYMBOL(wait_on_page_bit);
1382 int wait_on_page_bit_killable(struct page *page, int bit_nr)
1384 wait_queue_head_t *q = page_waitqueue(page);
1385 return wait_on_page_bit_common(q, page, bit_nr, TASK_KILLABLE, SHARED);
1387 EXPORT_SYMBOL(wait_on_page_bit_killable);
1390 * put_and_wait_on_page_locked - Drop a reference and wait for it to be unlocked
1391 * @page: The page to wait for.
1392 * @state: The sleep state (TASK_KILLABLE, TASK_UNINTERRUPTIBLE, etc).
1394 * The caller should hold a reference on @page. They expect the page to
1395 * become unlocked relatively soon, but do not wish to hold up migration
1396 * (for example) by holding the reference while waiting for the page to
1397 * come unlocked. After this function returns, the caller should not
1398 * dereference @page.
1400 * Return: 0 if the page was unlocked or -EINTR if interrupted by a signal.
1402 int put_and_wait_on_page_locked(struct page *page, int state)
1404 wait_queue_head_t *q;
1406 page = compound_head(page);
1407 q = page_waitqueue(page);
1408 return wait_on_page_bit_common(q, page, PG_locked, state, DROP);
1412 * add_page_wait_queue - Add an arbitrary waiter to a page's wait queue
1413 * @page: Page defining the wait queue of interest
1414 * @waiter: Waiter to add to the queue
1416 * Add an arbitrary @waiter to the wait queue for the nominated @page.
1418 void add_page_wait_queue(struct page *page, wait_queue_entry_t *waiter)
1420 wait_queue_head_t *q = page_waitqueue(page);
1421 unsigned long flags;
1423 spin_lock_irqsave(&q->lock, flags);
1424 __add_wait_queue_entry_tail(q, waiter);
1425 SetPageWaiters(page);
1426 spin_unlock_irqrestore(&q->lock, flags);
1428 EXPORT_SYMBOL_GPL(add_page_wait_queue);
1430 #ifndef clear_bit_unlock_is_negative_byte
1433 * PG_waiters is the high bit in the same byte as PG_lock.
1435 * On x86 (and on many other architectures), we can clear PG_lock and
1436 * test the sign bit at the same time. But if the architecture does
1437 * not support that special operation, we just do this all by hand
1440 * The read of PG_waiters has to be after (or concurrently with) PG_locked
1441 * being cleared, but a memory barrier should be unnecessary since it is
1442 * in the same byte as PG_locked.
1444 static inline bool clear_bit_unlock_is_negative_byte(long nr, volatile void *mem)
1446 clear_bit_unlock(nr, mem);
1447 /* smp_mb__after_atomic(); */
1448 return test_bit(PG_waiters, mem);
1454 * unlock_page - unlock a locked page
1457 * Unlocks the page and wakes up sleepers in wait_on_page_locked().
1458 * Also wakes sleepers in wait_on_page_writeback() because the wakeup
1459 * mechanism between PageLocked pages and PageWriteback pages is shared.
1460 * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep.
1462 * Note that this depends on PG_waiters being the sign bit in the byte
1463 * that contains PG_locked - thus the BUILD_BUG_ON(). That allows us to
1464 * clear the PG_locked bit and test PG_waiters at the same time fairly
1465 * portably (architectures that do LL/SC can test any bit, while x86 can
1466 * test the sign bit).
1468 void unlock_page(struct page *page)
1470 BUILD_BUG_ON(PG_waiters != 7);
1471 page = compound_head(page);
1472 VM_BUG_ON_PAGE(!PageLocked(page), page);
1473 if (clear_bit_unlock_is_negative_byte(PG_locked, &page->flags))
1474 wake_up_page_bit(page, PG_locked);
1476 EXPORT_SYMBOL(unlock_page);
1479 * end_page_private_2 - Clear PG_private_2 and release any waiters
1482 * Clear the PG_private_2 bit on a page and wake up any sleepers waiting for
1483 * this. The page ref held for PG_private_2 being set is released.
1485 * This is, for example, used when a netfs page is being written to a local
1486 * disk cache, thereby allowing writes to the cache for the same page to be
1489 void end_page_private_2(struct page *page)
1491 page = compound_head(page);
1492 VM_BUG_ON_PAGE(!PagePrivate2(page), page);
1493 clear_bit_unlock(PG_private_2, &page->flags);
1494 wake_up_page_bit(page, PG_private_2);
1497 EXPORT_SYMBOL(end_page_private_2);
1500 * wait_on_page_private_2 - Wait for PG_private_2 to be cleared on a page
1501 * @page: The page to wait on
1503 * Wait for PG_private_2 (aka PG_fscache) to be cleared on a page.
1505 void wait_on_page_private_2(struct page *page)
1507 page = compound_head(page);
1508 while (PagePrivate2(page))
1509 wait_on_page_bit(page, PG_private_2);
1511 EXPORT_SYMBOL(wait_on_page_private_2);
1514 * wait_on_page_private_2_killable - Wait for PG_private_2 to be cleared on a page
1515 * @page: The page to wait on
1517 * Wait for PG_private_2 (aka PG_fscache) to be cleared on a page or until a
1518 * fatal signal is received by the calling task.
1521 * - 0 if successful.
1522 * - -EINTR if a fatal signal was encountered.
1524 int wait_on_page_private_2_killable(struct page *page)
1528 page = compound_head(page);
1529 while (PagePrivate2(page)) {
1530 ret = wait_on_page_bit_killable(page, PG_private_2);
1537 EXPORT_SYMBOL(wait_on_page_private_2_killable);
1540 * end_page_writeback - end writeback against a page
1543 void end_page_writeback(struct page *page)
1546 * TestClearPageReclaim could be used here but it is an atomic
1547 * operation and overkill in this particular case. Failing to
1548 * shuffle a page marked for immediate reclaim is too mild to
1549 * justify taking an atomic operation penalty at the end of
1550 * ever page writeback.
1552 if (PageReclaim(page)) {
1553 ClearPageReclaim(page);
1554 rotate_reclaimable_page(page);
1558 * Writeback does not hold a page reference of its own, relying
1559 * on truncation to wait for the clearing of PG_writeback.
1560 * But here we must make sure that the page is not freed and
1561 * reused before the wake_up_page().
1564 if (!test_clear_page_writeback(page))
1567 smp_mb__after_atomic();
1568 wake_up_page(page, PG_writeback);
1571 EXPORT_SYMBOL(end_page_writeback);
1574 * After completing I/O on a page, call this routine to update the page
1575 * flags appropriately
1577 void page_endio(struct page *page, bool is_write, int err)
1581 SetPageUptodate(page);
1583 ClearPageUptodate(page);
1589 struct address_space *mapping;
1592 mapping = page_mapping(page);
1594 mapping_set_error(mapping, err);
1596 end_page_writeback(page);
1599 EXPORT_SYMBOL_GPL(page_endio);
1602 * __lock_page - get a lock on the page, assuming we need to sleep to get it
1603 * @__page: the page to lock
1605 void __lock_page(struct page *__page)
1607 struct page *page = compound_head(__page);
1608 wait_queue_head_t *q = page_waitqueue(page);
1609 wait_on_page_bit_common(q, page, PG_locked, TASK_UNINTERRUPTIBLE,
1612 EXPORT_SYMBOL(__lock_page);
1614 int __lock_page_killable(struct page *__page)
1616 struct page *page = compound_head(__page);
1617 wait_queue_head_t *q = page_waitqueue(page);
1618 return wait_on_page_bit_common(q, page, PG_locked, TASK_KILLABLE,
1621 EXPORT_SYMBOL_GPL(__lock_page_killable);
1623 int __lock_page_async(struct page *page, struct wait_page_queue *wait)
1625 struct wait_queue_head *q = page_waitqueue(page);
1629 wait->bit_nr = PG_locked;
1631 spin_lock_irq(&q->lock);
1632 __add_wait_queue_entry_tail(q, &wait->wait);
1633 SetPageWaiters(page);
1634 ret = !trylock_page(page);
1636 * If we were successful now, we know we're still on the
1637 * waitqueue as we're still under the lock. This means it's
1638 * safe to remove and return success, we know the callback
1639 * isn't going to trigger.
1642 __remove_wait_queue(q, &wait->wait);
1645 spin_unlock_irq(&q->lock);
1651 * 1 - page is locked; mmap_lock is still held.
1652 * 0 - page is not locked.
1653 * mmap_lock has been released (mmap_read_unlock(), unless flags had both
1654 * FAULT_FLAG_ALLOW_RETRY and FAULT_FLAG_RETRY_NOWAIT set, in
1655 * which case mmap_lock is still held.
1657 * If neither ALLOW_RETRY nor KILLABLE are set, will always return 1
1658 * with the page locked and the mmap_lock unperturbed.
1660 int __lock_page_or_retry(struct page *page, struct mm_struct *mm,
1663 if (fault_flag_allow_retry_first(flags)) {
1665 * CAUTION! In this case, mmap_lock is not released
1666 * even though return 0.
1668 if (flags & FAULT_FLAG_RETRY_NOWAIT)
1671 mmap_read_unlock(mm);
1672 if (flags & FAULT_FLAG_KILLABLE)
1673 wait_on_page_locked_killable(page);
1675 wait_on_page_locked(page);
1678 if (flags & FAULT_FLAG_KILLABLE) {
1681 ret = __lock_page_killable(page);
1683 mmap_read_unlock(mm);
1694 * page_cache_next_miss() - Find the next gap in the page cache.
1695 * @mapping: Mapping.
1697 * @max_scan: Maximum range to search.
1699 * Search the range [index, min(index + max_scan - 1, ULONG_MAX)] for the
1700 * gap with the lowest index.
1702 * This function may be called under the rcu_read_lock. However, this will
1703 * not atomically search a snapshot of the cache at a single point in time.
1704 * For example, if a gap is created at index 5, then subsequently a gap is
1705 * created at index 10, page_cache_next_miss covering both indices may
1706 * return 10 if called under the rcu_read_lock.
1708 * Return: The index of the gap if found, otherwise an index outside the
1709 * range specified (in which case 'return - index >= max_scan' will be true).
1710 * In the rare case of index wrap-around, 0 will be returned.
1712 pgoff_t page_cache_next_miss(struct address_space *mapping,
1713 pgoff_t index, unsigned long max_scan)
1715 XA_STATE(xas, &mapping->i_pages, index);
1717 while (max_scan--) {
1718 void *entry = xas_next(&xas);
1719 if (!entry || xa_is_value(entry))
1721 if (xas.xa_index == 0)
1725 return xas.xa_index;
1727 EXPORT_SYMBOL(page_cache_next_miss);
1730 * page_cache_prev_miss() - Find the previous gap in the page cache.
1731 * @mapping: Mapping.
1733 * @max_scan: Maximum range to search.
1735 * Search the range [max(index - max_scan + 1, 0), index] for the
1736 * gap with the highest index.
1738 * This function may be called under the rcu_read_lock. However, this will
1739 * not atomically search a snapshot of the cache at a single point in time.
1740 * For example, if a gap is created at index 10, then subsequently a gap is
1741 * created at index 5, page_cache_prev_miss() covering both indices may
1742 * return 5 if called under the rcu_read_lock.
1744 * Return: The index of the gap if found, otherwise an index outside the
1745 * range specified (in which case 'index - return >= max_scan' will be true).
1746 * In the rare case of wrap-around, ULONG_MAX will be returned.
1748 pgoff_t page_cache_prev_miss(struct address_space *mapping,
1749 pgoff_t index, unsigned long max_scan)
1751 XA_STATE(xas, &mapping->i_pages, index);
1753 while (max_scan--) {
1754 void *entry = xas_prev(&xas);
1755 if (!entry || xa_is_value(entry))
1757 if (xas.xa_index == ULONG_MAX)
1761 return xas.xa_index;
1763 EXPORT_SYMBOL(page_cache_prev_miss);
1766 * mapping_get_entry - Get a page cache entry.
1767 * @mapping: the address_space to search
1768 * @index: The page cache index.
1770 * Looks up the page cache slot at @mapping & @index. If there is a
1771 * page cache page, the head page is returned with an increased refcount.
1773 * If the slot holds a shadow entry of a previously evicted page, or a
1774 * swap entry from shmem/tmpfs, it is returned.
1776 * Return: The head page or shadow entry, %NULL if nothing is found.
1778 static struct page *mapping_get_entry(struct address_space *mapping,
1781 XA_STATE(xas, &mapping->i_pages, index);
1787 page = xas_load(&xas);
1788 if (xas_retry(&xas, page))
1791 * A shadow entry of a recently evicted page, or a swap entry from
1792 * shmem/tmpfs. Return it without attempting to raise page count.
1794 if (!page || xa_is_value(page))
1797 if (!page_cache_get_speculative(page))
1801 * Has the page moved or been split?
1802 * This is part of the lockless pagecache protocol. See
1803 * include/linux/pagemap.h for details.
1805 if (unlikely(page != xas_reload(&xas))) {
1816 * pagecache_get_page - Find and get a reference to a page.
1817 * @mapping: The address_space to search.
1818 * @index: The page index.
1819 * @fgp_flags: %FGP flags modify how the page is returned.
1820 * @gfp_mask: Memory allocation flags to use if %FGP_CREAT is specified.
1822 * Looks up the page cache entry at @mapping & @index.
1824 * @fgp_flags can be zero or more of these flags:
1826 * * %FGP_ACCESSED - The page will be marked accessed.
1827 * * %FGP_LOCK - The page is returned locked.
1828 * * %FGP_HEAD - If the page is present and a THP, return the head page
1829 * rather than the exact page specified by the index.
1830 * * %FGP_ENTRY - If there is a shadow / swap / DAX entry, return it
1831 * instead of allocating a new page to replace it.
1832 * * %FGP_CREAT - If no page is present then a new page is allocated using
1833 * @gfp_mask and added to the page cache and the VM's LRU list.
1834 * The page is returned locked and with an increased refcount.
1835 * * %FGP_FOR_MMAP - The caller wants to do its own locking dance if the
1836 * page is already in cache. If the page was allocated, unlock it before
1837 * returning so the caller can do the same dance.
1838 * * %FGP_WRITE - The page will be written
1839 * * %FGP_NOFS - __GFP_FS will get cleared in gfp mask
1840 * * %FGP_NOWAIT - Don't get blocked by page lock
1842 * If %FGP_LOCK or %FGP_CREAT are specified then the function may sleep even
1843 * if the %GFP flags specified for %FGP_CREAT are atomic.
1845 * If there is a page cache page, it is returned with an increased refcount.
1847 * Return: The found page or %NULL otherwise.
1849 struct page *pagecache_get_page(struct address_space *mapping, pgoff_t index,
1850 int fgp_flags, gfp_t gfp_mask)
1855 page = mapping_get_entry(mapping, index);
1856 if (xa_is_value(page)) {
1857 if (fgp_flags & FGP_ENTRY)
1864 if (fgp_flags & FGP_LOCK) {
1865 if (fgp_flags & FGP_NOWAIT) {
1866 if (!trylock_page(page)) {
1874 /* Has the page been truncated? */
1875 if (unlikely(page->mapping != mapping)) {
1880 VM_BUG_ON_PAGE(!thp_contains(page, index), page);
1883 if (fgp_flags & FGP_ACCESSED)
1884 mark_page_accessed(page);
1885 else if (fgp_flags & FGP_WRITE) {
1886 /* Clear idle flag for buffer write */
1887 if (page_is_idle(page))
1888 clear_page_idle(page);
1890 if (!(fgp_flags & FGP_HEAD))
1891 page = find_subpage(page, index);
1894 if (!page && (fgp_flags & FGP_CREAT)) {
1896 if ((fgp_flags & FGP_WRITE) && mapping_can_writeback(mapping))
1897 gfp_mask |= __GFP_WRITE;
1898 if (fgp_flags & FGP_NOFS)
1899 gfp_mask &= ~__GFP_FS;
1901 page = __page_cache_alloc(gfp_mask);
1905 if (WARN_ON_ONCE(!(fgp_flags & (FGP_LOCK | FGP_FOR_MMAP))))
1906 fgp_flags |= FGP_LOCK;
1908 /* Init accessed so avoid atomic mark_page_accessed later */
1909 if (fgp_flags & FGP_ACCESSED)
1910 __SetPageReferenced(page);
1912 err = add_to_page_cache_lru(page, mapping, index, gfp_mask);
1913 if (unlikely(err)) {
1921 * add_to_page_cache_lru locks the page, and for mmap we expect
1924 if (page && (fgp_flags & FGP_FOR_MMAP))
1930 EXPORT_SYMBOL(pagecache_get_page);
1932 static inline struct page *find_get_entry(struct xa_state *xas, pgoff_t max,
1938 if (mark == XA_PRESENT)
1939 page = xas_find(xas, max);
1941 page = xas_find_marked(xas, max, mark);
1943 if (xas_retry(xas, page))
1946 * A shadow entry of a recently evicted page, a swap
1947 * entry from shmem/tmpfs or a DAX entry. Return it
1948 * without attempting to raise page count.
1950 if (!page || xa_is_value(page))
1953 if (!page_cache_get_speculative(page))
1956 /* Has the page moved or been split? */
1957 if (unlikely(page != xas_reload(xas))) {
1969 * find_get_entries - gang pagecache lookup
1970 * @mapping: The address_space to search
1971 * @start: The starting page cache index
1972 * @end: The final page index (inclusive).
1973 * @pvec: Where the resulting entries are placed.
1974 * @indices: The cache indices corresponding to the entries in @entries
1976 * find_get_entries() will search for and return a batch of entries in
1977 * the mapping. The entries are placed in @pvec. find_get_entries()
1978 * takes a reference on any actual pages it returns.
1980 * The search returns a group of mapping-contiguous page cache entries
1981 * with ascending indexes. There may be holes in the indices due to
1982 * not-present pages.
1984 * Any shadow entries of evicted pages, or swap entries from
1985 * shmem/tmpfs, are included in the returned array.
1987 * If it finds a Transparent Huge Page, head or tail, find_get_entries()
1988 * stops at that page: the caller is likely to have a better way to handle
1989 * the compound page as a whole, and then skip its extent, than repeatedly
1990 * calling find_get_entries() to return all its tails.
1992 * Return: the number of pages and shadow entries which were found.
1994 unsigned find_get_entries(struct address_space *mapping, pgoff_t start,
1995 pgoff_t end, struct pagevec *pvec, pgoff_t *indices)
1997 XA_STATE(xas, &mapping->i_pages, start);
1999 unsigned int ret = 0;
2000 unsigned nr_entries = PAGEVEC_SIZE;
2003 while ((page = find_get_entry(&xas, end, XA_PRESENT))) {
2005 * Terminate early on finding a THP, to allow the caller to
2006 * handle it all at once; but continue if this is hugetlbfs.
2008 if (!xa_is_value(page) && PageTransHuge(page) &&
2010 page = find_subpage(page, xas.xa_index);
2011 nr_entries = ret + 1;
2014 indices[ret] = xas.xa_index;
2015 pvec->pages[ret] = page;
2016 if (++ret == nr_entries)
2026 * find_lock_entries - Find a batch of pagecache entries.
2027 * @mapping: The address_space to search.
2028 * @start: The starting page cache index.
2029 * @end: The final page index (inclusive).
2030 * @pvec: Where the resulting entries are placed.
2031 * @indices: The cache indices of the entries in @pvec.
2033 * find_lock_entries() will return a batch of entries from @mapping.
2034 * Swap, shadow and DAX entries are included. Pages are returned
2035 * locked and with an incremented refcount. Pages which are locked by
2036 * somebody else or under writeback are skipped. Only the head page of
2037 * a THP is returned. Pages which are partially outside the range are
2040 * The entries have ascending indexes. The indices may not be consecutive
2041 * due to not-present entries, THP pages, pages which could not be locked
2042 * or pages under writeback.
2044 * Return: The number of entries which were found.
2046 unsigned find_lock_entries(struct address_space *mapping, pgoff_t start,
2047 pgoff_t end, struct pagevec *pvec, pgoff_t *indices)
2049 XA_STATE(xas, &mapping->i_pages, start);
2053 while ((page = find_get_entry(&xas, end, XA_PRESENT))) {
2054 if (!xa_is_value(page)) {
2055 if (page->index < start)
2057 VM_BUG_ON_PAGE(page->index != xas.xa_index, page);
2058 if (page->index + thp_nr_pages(page) - 1 > end)
2060 if (!trylock_page(page))
2062 if (page->mapping != mapping || PageWriteback(page))
2064 VM_BUG_ON_PAGE(!thp_contains(page, xas.xa_index),
2067 indices[pvec->nr] = xas.xa_index;
2068 if (!pagevec_add(pvec, page))
2076 if (!xa_is_value(page) && PageTransHuge(page)) {
2077 unsigned int nr_pages = thp_nr_pages(page);
2079 /* Final THP may cross MAX_LFS_FILESIZE on 32-bit */
2080 xas_set(&xas, page->index + nr_pages);
2081 if (xas.xa_index < nr_pages)
2087 return pagevec_count(pvec);
2091 * find_get_pages_range - gang pagecache lookup
2092 * @mapping: The address_space to search
2093 * @start: The starting page index
2094 * @end: The final page index (inclusive)
2095 * @nr_pages: The maximum number of pages
2096 * @pages: Where the resulting pages are placed
2098 * find_get_pages_range() will search for and return a group of up to @nr_pages
2099 * pages in the mapping starting at index @start and up to index @end
2100 * (inclusive). The pages are placed at @pages. find_get_pages_range() takes
2101 * a reference against the returned pages.
2103 * The search returns a group of mapping-contiguous pages with ascending
2104 * indexes. There may be holes in the indices due to not-present pages.
2105 * We also update @start to index the next page for the traversal.
2107 * Return: the number of pages which were found. If this number is
2108 * smaller than @nr_pages, the end of specified range has been
2111 unsigned find_get_pages_range(struct address_space *mapping, pgoff_t *start,
2112 pgoff_t end, unsigned int nr_pages,
2113 struct page **pages)
2115 XA_STATE(xas, &mapping->i_pages, *start);
2119 if (unlikely(!nr_pages))
2123 while ((page = find_get_entry(&xas, end, XA_PRESENT))) {
2124 /* Skip over shadow, swap and DAX entries */
2125 if (xa_is_value(page))
2128 pages[ret] = find_subpage(page, xas.xa_index);
2129 if (++ret == nr_pages) {
2130 *start = xas.xa_index + 1;
2136 * We come here when there is no page beyond @end. We take care to not
2137 * overflow the index @start as it confuses some of the callers. This
2138 * breaks the iteration when there is a page at index -1 but that is
2139 * already broken anyway.
2141 if (end == (pgoff_t)-1)
2142 *start = (pgoff_t)-1;
2152 * find_get_pages_contig - gang contiguous pagecache lookup
2153 * @mapping: The address_space to search
2154 * @index: The starting page index
2155 * @nr_pages: The maximum number of pages
2156 * @pages: Where the resulting pages are placed
2158 * find_get_pages_contig() works exactly like find_get_pages(), except
2159 * that the returned number of pages are guaranteed to be contiguous.
2161 * Return: the number of pages which were found.
2163 unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index,
2164 unsigned int nr_pages, struct page **pages)
2166 XA_STATE(xas, &mapping->i_pages, index);
2168 unsigned int ret = 0;
2170 if (unlikely(!nr_pages))
2174 for (page = xas_load(&xas); page; page = xas_next(&xas)) {
2175 if (xas_retry(&xas, page))
2178 * If the entry has been swapped out, we can stop looking.
2179 * No current caller is looking for DAX entries.
2181 if (xa_is_value(page))
2184 if (!page_cache_get_speculative(page))
2187 /* Has the page moved or been split? */
2188 if (unlikely(page != xas_reload(&xas)))
2191 pages[ret] = find_subpage(page, xas.xa_index);
2192 if (++ret == nr_pages)
2203 EXPORT_SYMBOL(find_get_pages_contig);
2206 * find_get_pages_range_tag - Find and return head pages matching @tag.
2207 * @mapping: the address_space to search
2208 * @index: the starting page index
2209 * @end: The final page index (inclusive)
2210 * @tag: the tag index
2211 * @nr_pages: the maximum number of pages
2212 * @pages: where the resulting pages are placed
2214 * Like find_get_pages(), except we only return head pages which are tagged
2215 * with @tag. @index is updated to the index immediately after the last
2216 * page we return, ready for the next iteration.
2218 * Return: the number of pages which were found.
2220 unsigned find_get_pages_range_tag(struct address_space *mapping, pgoff_t *index,
2221 pgoff_t end, xa_mark_t tag, unsigned int nr_pages,
2222 struct page **pages)
2224 XA_STATE(xas, &mapping->i_pages, *index);
2228 if (unlikely(!nr_pages))
2232 while ((page = find_get_entry(&xas, end, tag))) {
2234 * Shadow entries should never be tagged, but this iteration
2235 * is lockless so there is a window for page reclaim to evict
2236 * a page we saw tagged. Skip over it.
2238 if (xa_is_value(page))
2242 if (++ret == nr_pages) {
2243 *index = page->index + thp_nr_pages(page);
2249 * We come here when we got to @end. We take care to not overflow the
2250 * index @index as it confuses some of the callers. This breaks the
2251 * iteration when there is a page at index -1 but that is already
2254 if (end == (pgoff_t)-1)
2255 *index = (pgoff_t)-1;
2263 EXPORT_SYMBOL(find_get_pages_range_tag);
2266 * CD/DVDs are error prone. When a medium error occurs, the driver may fail
2267 * a _large_ part of the i/o request. Imagine the worst scenario:
2269 * ---R__________________________________________B__________
2270 * ^ reading here ^ bad block(assume 4k)
2272 * read(R) => miss => readahead(R...B) => media error => frustrating retries
2273 * => failing the whole request => read(R) => read(R+1) =>
2274 * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
2275 * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
2276 * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
2278 * It is going insane. Fix it by quickly scaling down the readahead size.
2280 static void shrink_readahead_size_eio(struct file_ra_state *ra)
2286 * filemap_get_read_batch - Get a batch of pages for read
2288 * Get a batch of pages which represent a contiguous range of bytes
2289 * in the file. No tail pages will be returned. If @index is in the
2290 * middle of a THP, the entire THP will be returned. The last page in
2291 * the batch may have Readahead set or be not Uptodate so that the
2292 * caller can take the appropriate action.
2294 static void filemap_get_read_batch(struct address_space *mapping,
2295 pgoff_t index, pgoff_t max, struct pagevec *pvec)
2297 XA_STATE(xas, &mapping->i_pages, index);
2301 for (head = xas_load(&xas); head; head = xas_next(&xas)) {
2302 if (xas_retry(&xas, head))
2304 if (xas.xa_index > max || xa_is_value(head))
2306 if (!page_cache_get_speculative(head))
2309 /* Has the page moved or been split? */
2310 if (unlikely(head != xas_reload(&xas)))
2313 if (!pagevec_add(pvec, head))
2315 if (!PageUptodate(head))
2317 if (PageReadahead(head))
2319 xas.xa_index = head->index + thp_nr_pages(head) - 1;
2320 xas.xa_offset = (xas.xa_index >> xas.xa_shift) & XA_CHUNK_MASK;
2330 static int filemap_read_page(struct file *file, struct address_space *mapping,
2336 * A previous I/O error may have been due to temporary failures,
2337 * eg. multipath errors. PG_error will be set again if readpage
2340 ClearPageError(page);
2341 /* Start the actual read. The read will unlock the page. */
2342 error = mapping->a_ops->readpage(file, page);
2346 error = wait_on_page_locked_killable(page);
2349 if (PageUptodate(page))
2351 shrink_readahead_size_eio(&file->f_ra);
2355 static bool filemap_range_uptodate(struct address_space *mapping,
2356 loff_t pos, struct iov_iter *iter, struct page *page)
2360 if (PageUptodate(page))
2362 /* pipes can't handle partially uptodate pages */
2363 if (iov_iter_is_pipe(iter))
2365 if (!mapping->a_ops->is_partially_uptodate)
2367 if (mapping->host->i_blkbits >= (PAGE_SHIFT + thp_order(page)))
2370 count = iter->count;
2371 if (page_offset(page) > pos) {
2372 count -= page_offset(page) - pos;
2375 pos -= page_offset(page);
2378 return mapping->a_ops->is_partially_uptodate(page, pos, count);
2381 static int filemap_update_page(struct kiocb *iocb,
2382 struct address_space *mapping, struct iov_iter *iter,
2387 if (!trylock_page(page)) {
2388 if (iocb->ki_flags & (IOCB_NOWAIT | IOCB_NOIO))
2390 if (!(iocb->ki_flags & IOCB_WAITQ)) {
2391 put_and_wait_on_page_locked(page, TASK_KILLABLE);
2392 return AOP_TRUNCATED_PAGE;
2394 error = __lock_page_async(page, iocb->ki_waitq);
2403 if (filemap_range_uptodate(mapping, iocb->ki_pos, iter, page))
2407 if (iocb->ki_flags & (IOCB_NOIO | IOCB_NOWAIT | IOCB_WAITQ))
2410 error = filemap_read_page(iocb->ki_filp, mapping, page);
2411 if (error == AOP_TRUNCATED_PAGE)
2417 return AOP_TRUNCATED_PAGE;
2423 static int filemap_create_page(struct file *file,
2424 struct address_space *mapping, pgoff_t index,
2425 struct pagevec *pvec)
2430 page = page_cache_alloc(mapping);
2434 error = add_to_page_cache_lru(page, mapping, index,
2435 mapping_gfp_constraint(mapping, GFP_KERNEL));
2436 if (error == -EEXIST)
2437 error = AOP_TRUNCATED_PAGE;
2441 error = filemap_read_page(file, mapping, page);
2445 pagevec_add(pvec, page);
2452 static int filemap_readahead(struct kiocb *iocb, struct file *file,
2453 struct address_space *mapping, struct page *page,
2456 if (iocb->ki_flags & IOCB_NOIO)
2458 page_cache_async_readahead(mapping, &file->f_ra, file, page,
2459 page->index, last_index - page->index);
2463 static int filemap_get_pages(struct kiocb *iocb, struct iov_iter *iter,
2464 struct pagevec *pvec)
2466 struct file *filp = iocb->ki_filp;
2467 struct address_space *mapping = filp->f_mapping;
2468 struct file_ra_state *ra = &filp->f_ra;
2469 pgoff_t index = iocb->ki_pos >> PAGE_SHIFT;
2474 last_index = DIV_ROUND_UP(iocb->ki_pos + iter->count, PAGE_SIZE);
2476 if (fatal_signal_pending(current))
2479 filemap_get_read_batch(mapping, index, last_index, pvec);
2480 if (!pagevec_count(pvec)) {
2481 if (iocb->ki_flags & IOCB_NOIO)
2483 page_cache_sync_readahead(mapping, ra, filp, index,
2484 last_index - index);
2485 filemap_get_read_batch(mapping, index, last_index, pvec);
2487 if (!pagevec_count(pvec)) {
2488 if (iocb->ki_flags & (IOCB_NOWAIT | IOCB_WAITQ))
2490 err = filemap_create_page(filp, mapping,
2491 iocb->ki_pos >> PAGE_SHIFT, pvec);
2492 if (err == AOP_TRUNCATED_PAGE)
2497 page = pvec->pages[pagevec_count(pvec) - 1];
2498 if (PageReadahead(page)) {
2499 err = filemap_readahead(iocb, filp, mapping, page, last_index);
2503 if (!PageUptodate(page)) {
2504 if ((iocb->ki_flags & IOCB_WAITQ) && pagevec_count(pvec) > 1)
2505 iocb->ki_flags |= IOCB_NOWAIT;
2506 err = filemap_update_page(iocb, mapping, iter, page);
2515 if (likely(--pvec->nr))
2517 if (err == AOP_TRUNCATED_PAGE)
2523 * filemap_read - Read data from the page cache.
2524 * @iocb: The iocb to read.
2525 * @iter: Destination for the data.
2526 * @already_read: Number of bytes already read by the caller.
2528 * Copies data from the page cache. If the data is not currently present,
2529 * uses the readahead and readpage address_space operations to fetch it.
2531 * Return: Total number of bytes copied, including those already read by
2532 * the caller. If an error happens before any bytes are copied, returns
2533 * a negative error number.
2535 ssize_t filemap_read(struct kiocb *iocb, struct iov_iter *iter,
2536 ssize_t already_read)
2538 struct file *filp = iocb->ki_filp;
2539 struct file_ra_state *ra = &filp->f_ra;
2540 struct address_space *mapping = filp->f_mapping;
2541 struct inode *inode = mapping->host;
2542 struct pagevec pvec;
2544 bool writably_mapped;
2545 loff_t isize, end_offset;
2547 if (unlikely(iocb->ki_pos >= inode->i_sb->s_maxbytes))
2549 if (unlikely(!iov_iter_count(iter)))
2552 iov_iter_truncate(iter, inode->i_sb->s_maxbytes);
2553 pagevec_init(&pvec);
2559 * If we've already successfully copied some data, then we
2560 * can no longer safely return -EIOCBQUEUED. Hence mark
2561 * an async read NOWAIT at that point.
2563 if ((iocb->ki_flags & IOCB_WAITQ) && already_read)
2564 iocb->ki_flags |= IOCB_NOWAIT;
2566 error = filemap_get_pages(iocb, iter, &pvec);
2571 * i_size must be checked after we know the pages are Uptodate.
2573 * Checking i_size after the check allows us to calculate
2574 * the correct value for "nr", which means the zero-filled
2575 * part of the page is not copied back to userspace (unless
2576 * another truncate extends the file - this is desired though).
2578 isize = i_size_read(inode);
2579 if (unlikely(iocb->ki_pos >= isize))
2581 end_offset = min_t(loff_t, isize, iocb->ki_pos + iter->count);
2584 * Once we start copying data, we don't want to be touching any
2585 * cachelines that might be contended:
2587 writably_mapped = mapping_writably_mapped(mapping);
2590 * When a sequential read accesses a page several times, only
2591 * mark it as accessed the first time.
2593 if (iocb->ki_pos >> PAGE_SHIFT !=
2594 ra->prev_pos >> PAGE_SHIFT)
2595 mark_page_accessed(pvec.pages[0]);
2597 for (i = 0; i < pagevec_count(&pvec); i++) {
2598 struct page *page = pvec.pages[i];
2599 size_t page_size = thp_size(page);
2600 size_t offset = iocb->ki_pos & (page_size - 1);
2601 size_t bytes = min_t(loff_t, end_offset - iocb->ki_pos,
2602 page_size - offset);
2605 if (end_offset < page_offset(page))
2608 mark_page_accessed(page);
2610 * If users can be writing to this page using arbitrary
2611 * virtual addresses, take care about potential aliasing
2612 * before reading the page on the kernel side.
2614 if (writably_mapped) {
2617 for (j = 0; j < thp_nr_pages(page); j++)
2618 flush_dcache_page(page + j);
2621 copied = copy_page_to_iter(page, offset, bytes, iter);
2623 already_read += copied;
2624 iocb->ki_pos += copied;
2625 ra->prev_pos = iocb->ki_pos;
2627 if (copied < bytes) {
2633 for (i = 0; i < pagevec_count(&pvec); i++)
2634 put_page(pvec.pages[i]);
2635 pagevec_reinit(&pvec);
2636 } while (iov_iter_count(iter) && iocb->ki_pos < isize && !error);
2638 file_accessed(filp);
2640 return already_read ? already_read : error;
2642 EXPORT_SYMBOL_GPL(filemap_read);
2645 * generic_file_read_iter - generic filesystem read routine
2646 * @iocb: kernel I/O control block
2647 * @iter: destination for the data read
2649 * This is the "read_iter()" routine for all filesystems
2650 * that can use the page cache directly.
2652 * The IOCB_NOWAIT flag in iocb->ki_flags indicates that -EAGAIN shall
2653 * be returned when no data can be read without waiting for I/O requests
2654 * to complete; it doesn't prevent readahead.
2656 * The IOCB_NOIO flag in iocb->ki_flags indicates that no new I/O
2657 * requests shall be made for the read or for readahead. When no data
2658 * can be read, -EAGAIN shall be returned. When readahead would be
2659 * triggered, a partial, possibly empty read shall be returned.
2662 * * number of bytes copied, even for partial reads
2663 * * negative error code (or 0 if IOCB_NOIO) if nothing was read
2666 generic_file_read_iter(struct kiocb *iocb, struct iov_iter *iter)
2668 size_t count = iov_iter_count(iter);
2672 return 0; /* skip atime */
2674 if (iocb->ki_flags & IOCB_DIRECT) {
2675 struct file *file = iocb->ki_filp;
2676 struct address_space *mapping = file->f_mapping;
2677 struct inode *inode = mapping->host;
2680 size = i_size_read(inode);
2681 if (iocb->ki_flags & IOCB_NOWAIT) {
2682 if (filemap_range_needs_writeback(mapping, iocb->ki_pos,
2683 iocb->ki_pos + count - 1))
2686 retval = filemap_write_and_wait_range(mapping,
2688 iocb->ki_pos + count - 1);
2693 file_accessed(file);
2695 retval = mapping->a_ops->direct_IO(iocb, iter);
2697 iocb->ki_pos += retval;
2700 if (retval != -EIOCBQUEUED)
2701 iov_iter_revert(iter, count - iov_iter_count(iter));
2704 * Btrfs can have a short DIO read if we encounter
2705 * compressed extents, so if there was an error, or if
2706 * we've already read everything we wanted to, or if
2707 * there was a short read because we hit EOF, go ahead
2708 * and return. Otherwise fallthrough to buffered io for
2709 * the rest of the read. Buffered reads will not work for
2710 * DAX files, so don't bother trying.
2712 if (retval < 0 || !count || iocb->ki_pos >= size ||
2717 return filemap_read(iocb, iter, retval);
2719 EXPORT_SYMBOL(generic_file_read_iter);
2721 static inline loff_t page_seek_hole_data(struct xa_state *xas,
2722 struct address_space *mapping, struct page *page,
2723 loff_t start, loff_t end, bool seek_data)
2725 const struct address_space_operations *ops = mapping->a_ops;
2726 size_t offset, bsz = i_blocksize(mapping->host);
2728 if (xa_is_value(page) || PageUptodate(page))
2729 return seek_data ? start : end;
2730 if (!ops->is_partially_uptodate)
2731 return seek_data ? end : start;
2736 if (unlikely(page->mapping != mapping))
2739 offset = offset_in_thp(page, start) & ~(bsz - 1);
2742 if (ops->is_partially_uptodate(page, offset, bsz) == seek_data)
2744 start = (start + bsz) & ~(bsz - 1);
2746 } while (offset < thp_size(page));
2754 unsigned int seek_page_size(struct xa_state *xas, struct page *page)
2756 if (xa_is_value(page))
2757 return PAGE_SIZE << xa_get_order(xas->xa, xas->xa_index);
2758 return thp_size(page);
2762 * mapping_seek_hole_data - Seek for SEEK_DATA / SEEK_HOLE in the page cache.
2763 * @mapping: Address space to search.
2764 * @start: First byte to consider.
2765 * @end: Limit of search (exclusive).
2766 * @whence: Either SEEK_HOLE or SEEK_DATA.
2768 * If the page cache knows which blocks contain holes and which blocks
2769 * contain data, your filesystem can use this function to implement
2770 * SEEK_HOLE and SEEK_DATA. This is useful for filesystems which are
2771 * entirely memory-based such as tmpfs, and filesystems which support
2772 * unwritten extents.
2774 * Return: The requested offset on successs, or -ENXIO if @whence specifies
2775 * SEEK_DATA and there is no data after @start. There is an implicit hole
2776 * after @end - 1, so SEEK_HOLE returns @end if all the bytes between @start
2777 * and @end contain data.
2779 loff_t mapping_seek_hole_data(struct address_space *mapping, loff_t start,
2780 loff_t end, int whence)
2782 XA_STATE(xas, &mapping->i_pages, start >> PAGE_SHIFT);
2783 pgoff_t max = (end - 1) >> PAGE_SHIFT;
2784 bool seek_data = (whence == SEEK_DATA);
2791 while ((page = find_get_entry(&xas, max, XA_PRESENT))) {
2792 loff_t pos = (u64)xas.xa_index << PAGE_SHIFT;
2793 unsigned int seek_size;
2801 seek_size = seek_page_size(&xas, page);
2802 pos = round_up(pos + 1, seek_size);
2803 start = page_seek_hole_data(&xas, mapping, page, start, pos,
2809 if (seek_size > PAGE_SIZE)
2810 xas_set(&xas, pos >> PAGE_SHIFT);
2811 if (!xa_is_value(page))
2818 if (page && !xa_is_value(page))
2826 #define MMAP_LOTSAMISS (100)
2828 * lock_page_maybe_drop_mmap - lock the page, possibly dropping the mmap_lock
2829 * @vmf - the vm_fault for this fault.
2830 * @page - the page to lock.
2831 * @fpin - the pointer to the file we may pin (or is already pinned).
2833 * This works similar to lock_page_or_retry in that it can drop the mmap_lock.
2834 * It differs in that it actually returns the page locked if it returns 1 and 0
2835 * if it couldn't lock the page. If we did have to drop the mmap_lock then fpin
2836 * will point to the pinned file and needs to be fput()'ed at a later point.
2838 static int lock_page_maybe_drop_mmap(struct vm_fault *vmf, struct page *page,
2841 if (trylock_page(page))
2845 * NOTE! This will make us return with VM_FAULT_RETRY, but with
2846 * the mmap_lock still held. That's how FAULT_FLAG_RETRY_NOWAIT
2847 * is supposed to work. We have way too many special cases..
2849 if (vmf->flags & FAULT_FLAG_RETRY_NOWAIT)
2852 *fpin = maybe_unlock_mmap_for_io(vmf, *fpin);
2853 if (vmf->flags & FAULT_FLAG_KILLABLE) {
2854 if (__lock_page_killable(page)) {
2856 * We didn't have the right flags to drop the mmap_lock,
2857 * but all fault_handlers only check for fatal signals
2858 * if we return VM_FAULT_RETRY, so we need to drop the
2859 * mmap_lock here and return 0 if we don't have a fpin.
2862 mmap_read_unlock(vmf->vma->vm_mm);
2872 * Synchronous readahead happens when we don't even find a page in the page
2873 * cache at all. We don't want to perform IO under the mmap sem, so if we have
2874 * to drop the mmap sem we return the file that was pinned in order for us to do
2875 * that. If we didn't pin a file then we return NULL. The file that is
2876 * returned needs to be fput()'ed when we're done with it.
2878 static struct file *do_sync_mmap_readahead(struct vm_fault *vmf)
2880 struct file *file = vmf->vma->vm_file;
2881 struct file_ra_state *ra = &file->f_ra;
2882 struct address_space *mapping = file->f_mapping;
2883 DEFINE_READAHEAD(ractl, file, ra, mapping, vmf->pgoff);
2884 struct file *fpin = NULL;
2885 unsigned int mmap_miss;
2887 /* If we don't want any read-ahead, don't bother */
2888 if (vmf->vma->vm_flags & VM_RAND_READ)
2893 if (vmf->vma->vm_flags & VM_SEQ_READ) {
2894 fpin = maybe_unlock_mmap_for_io(vmf, fpin);
2895 page_cache_sync_ra(&ractl, ra->ra_pages);
2899 /* Avoid banging the cache line if not needed */
2900 mmap_miss = READ_ONCE(ra->mmap_miss);
2901 if (mmap_miss < MMAP_LOTSAMISS * 10)
2902 WRITE_ONCE(ra->mmap_miss, ++mmap_miss);
2905 * Do we miss much more than hit in this file? If so,
2906 * stop bothering with read-ahead. It will only hurt.
2908 if (mmap_miss > MMAP_LOTSAMISS)
2914 fpin = maybe_unlock_mmap_for_io(vmf, fpin);
2915 ra->start = max_t(long, 0, vmf->pgoff - ra->ra_pages / 2);
2916 ra->size = ra->ra_pages;
2917 ra->async_size = ra->ra_pages / 4;
2918 ractl._index = ra->start;
2919 do_page_cache_ra(&ractl, ra->size, ra->async_size);
2924 * Asynchronous readahead happens when we find the page and PG_readahead,
2925 * so we want to possibly extend the readahead further. We return the file that
2926 * was pinned if we have to drop the mmap_lock in order to do IO.
2928 static struct file *do_async_mmap_readahead(struct vm_fault *vmf,
2931 struct file *file = vmf->vma->vm_file;
2932 struct file_ra_state *ra = &file->f_ra;
2933 struct address_space *mapping = file->f_mapping;
2934 struct file *fpin = NULL;
2935 unsigned int mmap_miss;
2936 pgoff_t offset = vmf->pgoff;
2938 /* If we don't want any read-ahead, don't bother */
2939 if (vmf->vma->vm_flags & VM_RAND_READ || !ra->ra_pages)
2941 mmap_miss = READ_ONCE(ra->mmap_miss);
2943 WRITE_ONCE(ra->mmap_miss, --mmap_miss);
2944 if (PageReadahead(page)) {
2945 fpin = maybe_unlock_mmap_for_io(vmf, fpin);
2946 page_cache_async_readahead(mapping, ra, file,
2947 page, offset, ra->ra_pages);
2953 * filemap_fault - read in file data for page fault handling
2954 * @vmf: struct vm_fault containing details of the fault
2956 * filemap_fault() is invoked via the vma operations vector for a
2957 * mapped memory region to read in file data during a page fault.
2959 * The goto's are kind of ugly, but this streamlines the normal case of having
2960 * it in the page cache, and handles the special cases reasonably without
2961 * having a lot of duplicated code.
2963 * vma->vm_mm->mmap_lock must be held on entry.
2965 * If our return value has VM_FAULT_RETRY set, it's because the mmap_lock
2966 * may be dropped before doing I/O or by lock_page_maybe_drop_mmap().
2968 * If our return value does not have VM_FAULT_RETRY set, the mmap_lock
2969 * has not been released.
2971 * We never return with VM_FAULT_RETRY and a bit from VM_FAULT_ERROR set.
2973 * Return: bitwise-OR of %VM_FAULT_ codes.
2975 vm_fault_t filemap_fault(struct vm_fault *vmf)
2978 struct file *file = vmf->vma->vm_file;
2979 struct file *fpin = NULL;
2980 struct address_space *mapping = file->f_mapping;
2981 struct inode *inode = mapping->host;
2982 pgoff_t offset = vmf->pgoff;
2987 max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
2988 if (unlikely(offset >= max_off))
2989 return VM_FAULT_SIGBUS;
2992 * Do we have something in the page cache already?
2994 page = find_get_page(mapping, offset);
2995 if (likely(page) && !(vmf->flags & FAULT_FLAG_TRIED)) {
2997 * We found the page, so try async readahead before
2998 * waiting for the lock.
3000 fpin = do_async_mmap_readahead(vmf, page);
3002 /* No page in the page cache at all */
3003 count_vm_event(PGMAJFAULT);
3004 count_memcg_event_mm(vmf->vma->vm_mm, PGMAJFAULT);
3005 ret = VM_FAULT_MAJOR;
3006 fpin = do_sync_mmap_readahead(vmf);
3008 page = pagecache_get_page(mapping, offset,
3009 FGP_CREAT|FGP_FOR_MMAP,
3014 return VM_FAULT_OOM;
3018 if (!lock_page_maybe_drop_mmap(vmf, page, &fpin))
3021 /* Did it get truncated? */
3022 if (unlikely(compound_head(page)->mapping != mapping)) {
3027 VM_BUG_ON_PAGE(page_to_pgoff(page) != offset, page);
3030 * We have a locked page in the page cache, now we need to check
3031 * that it's up-to-date. If not, it is going to be due to an error.
3033 if (unlikely(!PageUptodate(page)))
3034 goto page_not_uptodate;
3037 * We've made it this far and we had to drop our mmap_lock, now is the
3038 * time to return to the upper layer and have it re-find the vma and
3047 * Found the page and have a reference on it.
3048 * We must recheck i_size under page lock.
3050 max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
3051 if (unlikely(offset >= max_off)) {
3054 return VM_FAULT_SIGBUS;
3058 return ret | VM_FAULT_LOCKED;
3062 * Umm, take care of errors if the page isn't up-to-date.
3063 * Try to re-read it _once_. We do this synchronously,
3064 * because there really aren't any performance issues here
3065 * and we need to check for errors.
3067 fpin = maybe_unlock_mmap_for_io(vmf, fpin);
3068 error = filemap_read_page(file, mapping, page);
3073 if (!error || error == AOP_TRUNCATED_PAGE)
3076 return VM_FAULT_SIGBUS;
3080 * We dropped the mmap_lock, we need to return to the fault handler to
3081 * re-find the vma and come back and find our hopefully still populated
3088 return ret | VM_FAULT_RETRY;
3090 EXPORT_SYMBOL(filemap_fault);
3092 static bool filemap_map_pmd(struct vm_fault *vmf, struct page *page)
3094 struct mm_struct *mm = vmf->vma->vm_mm;
3096 /* Huge page is mapped? No need to proceed. */
3097 if (pmd_trans_huge(*vmf->pmd)) {
3103 if (pmd_none(*vmf->pmd) && PageTransHuge(page)) {
3104 vm_fault_t ret = do_set_pmd(vmf, page);
3106 /* The page is mapped successfully, reference consumed. */
3112 if (pmd_none(*vmf->pmd)) {
3113 vmf->ptl = pmd_lock(mm, vmf->pmd);
3114 if (likely(pmd_none(*vmf->pmd))) {
3116 pmd_populate(mm, vmf->pmd, vmf->prealloc_pte);
3117 vmf->prealloc_pte = NULL;
3119 spin_unlock(vmf->ptl);
3122 /* See comment in handle_pte_fault() */
3123 if (pmd_devmap_trans_unstable(vmf->pmd)) {
3132 static struct page *next_uptodate_page(struct page *page,
3133 struct address_space *mapping,
3134 struct xa_state *xas, pgoff_t end_pgoff)
3136 unsigned long max_idx;
3141 if (xas_retry(xas, page))
3143 if (xa_is_value(page))
3145 if (PageLocked(page))
3147 if (!page_cache_get_speculative(page))
3149 /* Has the page moved or been split? */
3150 if (unlikely(page != xas_reload(xas)))
3152 if (!PageUptodate(page) || PageReadahead(page))
3154 if (PageHWPoison(page))
3156 if (!trylock_page(page))
3158 if (page->mapping != mapping)
3160 if (!PageUptodate(page))
3162 max_idx = DIV_ROUND_UP(i_size_read(mapping->host), PAGE_SIZE);
3163 if (xas->xa_index >= max_idx)
3170 } while ((page = xas_next_entry(xas, end_pgoff)) != NULL);
3175 static inline struct page *first_map_page(struct address_space *mapping,
3176 struct xa_state *xas,
3179 return next_uptodate_page(xas_find(xas, end_pgoff),
3180 mapping, xas, end_pgoff);
3183 static inline struct page *next_map_page(struct address_space *mapping,
3184 struct xa_state *xas,
3187 return next_uptodate_page(xas_next_entry(xas, end_pgoff),
3188 mapping, xas, end_pgoff);
3191 vm_fault_t filemap_map_pages(struct vm_fault *vmf,
3192 pgoff_t start_pgoff, pgoff_t end_pgoff)
3194 struct vm_area_struct *vma = vmf->vma;
3195 struct file *file = vma->vm_file;
3196 struct address_space *mapping = file->f_mapping;
3197 pgoff_t last_pgoff = start_pgoff;
3199 XA_STATE(xas, &mapping->i_pages, start_pgoff);
3200 struct page *head, *page;
3201 unsigned int mmap_miss = READ_ONCE(file->f_ra.mmap_miss);
3205 head = first_map_page(mapping, &xas, end_pgoff);
3209 if (filemap_map_pmd(vmf, head)) {
3210 ret = VM_FAULT_NOPAGE;
3214 addr = vma->vm_start + ((start_pgoff - vma->vm_pgoff) << PAGE_SHIFT);
3215 vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, addr, &vmf->ptl);
3217 page = find_subpage(head, xas.xa_index);
3218 if (PageHWPoison(page))
3224 addr += (xas.xa_index - last_pgoff) << PAGE_SHIFT;
3225 vmf->pte += xas.xa_index - last_pgoff;
3226 last_pgoff = xas.xa_index;
3228 if (!pte_none(*vmf->pte))
3231 /* We're about to handle the fault */
3232 if (vmf->address == addr)
3233 ret = VM_FAULT_NOPAGE;
3235 do_set_pte(vmf, page, addr);
3236 /* no need to invalidate: a not-present page won't be cached */
3237 update_mmu_cache(vma, addr, vmf->pte);
3243 } while ((head = next_map_page(mapping, &xas, end_pgoff)) != NULL);
3244 pte_unmap_unlock(vmf->pte, vmf->ptl);
3247 WRITE_ONCE(file->f_ra.mmap_miss, mmap_miss);
3250 EXPORT_SYMBOL(filemap_map_pages);
3252 vm_fault_t filemap_page_mkwrite(struct vm_fault *vmf)
3254 struct address_space *mapping = vmf->vma->vm_file->f_mapping;
3255 struct page *page = vmf->page;
3256 vm_fault_t ret = VM_FAULT_LOCKED;
3258 sb_start_pagefault(mapping->host->i_sb);
3259 file_update_time(vmf->vma->vm_file);
3261 if (page->mapping != mapping) {
3263 ret = VM_FAULT_NOPAGE;
3267 * We mark the page dirty already here so that when freeze is in
3268 * progress, we are guaranteed that writeback during freezing will
3269 * see the dirty page and writeprotect it again.
3271 set_page_dirty(page);
3272 wait_for_stable_page(page);
3274 sb_end_pagefault(mapping->host->i_sb);
3278 const struct vm_operations_struct generic_file_vm_ops = {
3279 .fault = filemap_fault,
3280 .map_pages = filemap_map_pages,
3281 .page_mkwrite = filemap_page_mkwrite,
3284 /* This is used for a general mmap of a disk file */
3286 int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
3288 struct address_space *mapping = file->f_mapping;
3290 if (!mapping->a_ops->readpage)
3292 file_accessed(file);
3293 vma->vm_ops = &generic_file_vm_ops;
3298 * This is for filesystems which do not implement ->writepage.
3300 int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
3302 if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE))
3304 return generic_file_mmap(file, vma);
3307 vm_fault_t filemap_page_mkwrite(struct vm_fault *vmf)
3309 return VM_FAULT_SIGBUS;
3311 int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
3315 int generic_file_readonly_mmap(struct file * file, struct vm_area_struct * vma)
3319 #endif /* CONFIG_MMU */
3321 EXPORT_SYMBOL(filemap_page_mkwrite);
3322 EXPORT_SYMBOL(generic_file_mmap);
3323 EXPORT_SYMBOL(generic_file_readonly_mmap);
3325 static struct page *wait_on_page_read(struct page *page)
3327 if (!IS_ERR(page)) {
3328 wait_on_page_locked(page);
3329 if (!PageUptodate(page)) {
3331 page = ERR_PTR(-EIO);
3337 static struct page *do_read_cache_page(struct address_space *mapping,
3339 int (*filler)(void *, struct page *),
3346 page = find_get_page(mapping, index);
3348 page = __page_cache_alloc(gfp);
3350 return ERR_PTR(-ENOMEM);
3351 err = add_to_page_cache_lru(page, mapping, index, gfp);
3352 if (unlikely(err)) {
3356 /* Presumably ENOMEM for xarray node */
3357 return ERR_PTR(err);
3362 err = filler(data, page);
3364 err = mapping->a_ops->readpage(data, page);
3368 return ERR_PTR(err);
3371 page = wait_on_page_read(page);
3376 if (PageUptodate(page))
3380 * Page is not up to date and may be locked due to one of the following
3381 * case a: Page is being filled and the page lock is held
3382 * case b: Read/write error clearing the page uptodate status
3383 * case c: Truncation in progress (page locked)
3384 * case d: Reclaim in progress
3386 * Case a, the page will be up to date when the page is unlocked.
3387 * There is no need to serialise on the page lock here as the page
3388 * is pinned so the lock gives no additional protection. Even if the
3389 * page is truncated, the data is still valid if PageUptodate as
3390 * it's a race vs truncate race.
3391 * Case b, the page will not be up to date
3392 * Case c, the page may be truncated but in itself, the data may still
3393 * be valid after IO completes as it's a read vs truncate race. The
3394 * operation must restart if the page is not uptodate on unlock but
3395 * otherwise serialising on page lock to stabilise the mapping gives
3396 * no additional guarantees to the caller as the page lock is
3397 * released before return.
3398 * Case d, similar to truncation. If reclaim holds the page lock, it
3399 * will be a race with remove_mapping that determines if the mapping
3400 * is valid on unlock but otherwise the data is valid and there is
3401 * no need to serialise with page lock.
3403 * As the page lock gives no additional guarantee, we optimistically
3404 * wait on the page to be unlocked and check if it's up to date and
3405 * use the page if it is. Otherwise, the page lock is required to
3406 * distinguish between the different cases. The motivation is that we
3407 * avoid spurious serialisations and wakeups when multiple processes
3408 * wait on the same page for IO to complete.
3410 wait_on_page_locked(page);
3411 if (PageUptodate(page))
3414 /* Distinguish between all the cases under the safety of the lock */
3417 /* Case c or d, restart the operation */
3418 if (!page->mapping) {
3424 /* Someone else locked and filled the page in a very small window */
3425 if (PageUptodate(page)) {
3431 * A previous I/O error may have been due to temporary
3433 * Clear page error before actual read, PG_error will be
3434 * set again if read page fails.
3436 ClearPageError(page);
3440 mark_page_accessed(page);
3445 * read_cache_page - read into page cache, fill it if needed
3446 * @mapping: the page's address_space
3447 * @index: the page index
3448 * @filler: function to perform the read
3449 * @data: first arg to filler(data, page) function, often left as NULL
3451 * Read into the page cache. If a page already exists, and PageUptodate() is
3452 * not set, try to fill the page and wait for it to become unlocked.
3454 * If the page does not get brought uptodate, return -EIO.
3456 * Return: up to date page on success, ERR_PTR() on failure.
3458 struct page *read_cache_page(struct address_space *mapping,
3460 int (*filler)(void *, struct page *),
3463 return do_read_cache_page(mapping, index, filler, data,
3464 mapping_gfp_mask(mapping));
3466 EXPORT_SYMBOL(read_cache_page);
3469 * read_cache_page_gfp - read into page cache, using specified page allocation flags.
3470 * @mapping: the page's address_space
3471 * @index: the page index
3472 * @gfp: the page allocator flags to use if allocating
3474 * This is the same as "read_mapping_page(mapping, index, NULL)", but with
3475 * any new page allocations done using the specified allocation flags.
3477 * If the page does not get brought uptodate, return -EIO.
3479 * Return: up to date page on success, ERR_PTR() on failure.
3481 struct page *read_cache_page_gfp(struct address_space *mapping,
3485 return do_read_cache_page(mapping, index, NULL, NULL, gfp);
3487 EXPORT_SYMBOL(read_cache_page_gfp);
3489 int pagecache_write_begin(struct file *file, struct address_space *mapping,
3490 loff_t pos, unsigned len, unsigned flags,
3491 struct page **pagep, void **fsdata)
3493 const struct address_space_operations *aops = mapping->a_ops;
3495 return aops->write_begin(file, mapping, pos, len, flags,
3498 EXPORT_SYMBOL(pagecache_write_begin);
3500 int pagecache_write_end(struct file *file, struct address_space *mapping,
3501 loff_t pos, unsigned len, unsigned copied,
3502 struct page *page, void *fsdata)
3504 const struct address_space_operations *aops = mapping->a_ops;
3506 return aops->write_end(file, mapping, pos, len, copied, page, fsdata);
3508 EXPORT_SYMBOL(pagecache_write_end);
3511 * Warn about a page cache invalidation failure during a direct I/O write.
3513 void dio_warn_stale_pagecache(struct file *filp)
3515 static DEFINE_RATELIMIT_STATE(_rs, 86400 * HZ, DEFAULT_RATELIMIT_BURST);
3519 errseq_set(&filp->f_mapping->wb_err, -EIO);
3520 if (__ratelimit(&_rs)) {
3521 path = file_path(filp, pathname, sizeof(pathname));
3524 pr_crit("Page cache invalidation failure on direct I/O. Possible data corruption due to collision with buffered I/O!\n");
3525 pr_crit("File: %s PID: %d Comm: %.20s\n", path, current->pid,
3531 generic_file_direct_write(struct kiocb *iocb, struct iov_iter *from)
3533 struct file *file = iocb->ki_filp;
3534 struct address_space *mapping = file->f_mapping;
3535 struct inode *inode = mapping->host;
3536 loff_t pos = iocb->ki_pos;
3541 write_len = iov_iter_count(from);
3542 end = (pos + write_len - 1) >> PAGE_SHIFT;
3544 if (iocb->ki_flags & IOCB_NOWAIT) {
3545 /* If there are pages to writeback, return */
3546 if (filemap_range_has_page(file->f_mapping, pos,
3547 pos + write_len - 1))
3550 written = filemap_write_and_wait_range(mapping, pos,
3551 pos + write_len - 1);
3557 * After a write we want buffered reads to be sure to go to disk to get
3558 * the new data. We invalidate clean cached page from the region we're
3559 * about to write. We do this *before* the write so that we can return
3560 * without clobbering -EIOCBQUEUED from ->direct_IO().
3562 written = invalidate_inode_pages2_range(mapping,
3563 pos >> PAGE_SHIFT, end);
3565 * If a page can not be invalidated, return 0 to fall back
3566 * to buffered write.
3569 if (written == -EBUSY)
3574 written = mapping->a_ops->direct_IO(iocb, from);
3577 * Finally, try again to invalidate clean pages which might have been
3578 * cached by non-direct readahead, or faulted in by get_user_pages()
3579 * if the source of the write was an mmap'ed region of the file
3580 * we're writing. Either one is a pretty crazy thing to do,
3581 * so we don't support it 100%. If this invalidation
3582 * fails, tough, the write still worked...
3584 * Most of the time we do not need this since dio_complete() will do
3585 * the invalidation for us. However there are some file systems that
3586 * do not end up with dio_complete() being called, so let's not break
3587 * them by removing it completely.
3589 * Noticeable example is a blkdev_direct_IO().
3591 * Skip invalidation for async writes or if mapping has no pages.
3593 if (written > 0 && mapping->nrpages &&
3594 invalidate_inode_pages2_range(mapping, pos >> PAGE_SHIFT, end))
3595 dio_warn_stale_pagecache(file);
3599 write_len -= written;
3600 if (pos > i_size_read(inode) && !S_ISBLK(inode->i_mode)) {
3601 i_size_write(inode, pos);
3602 mark_inode_dirty(inode);
3606 if (written != -EIOCBQUEUED)
3607 iov_iter_revert(from, write_len - iov_iter_count(from));
3611 EXPORT_SYMBOL(generic_file_direct_write);
3614 * Find or create a page at the given pagecache position. Return the locked
3615 * page. This function is specifically for buffered writes.
3617 struct page *grab_cache_page_write_begin(struct address_space *mapping,
3618 pgoff_t index, unsigned flags)
3621 int fgp_flags = FGP_LOCK|FGP_WRITE|FGP_CREAT;
3623 if (flags & AOP_FLAG_NOFS)
3624 fgp_flags |= FGP_NOFS;
3626 page = pagecache_get_page(mapping, index, fgp_flags,
3627 mapping_gfp_mask(mapping));
3629 wait_for_stable_page(page);
3633 EXPORT_SYMBOL(grab_cache_page_write_begin);
3635 ssize_t generic_perform_write(struct file *file,
3636 struct iov_iter *i, loff_t pos)
3638 struct address_space *mapping = file->f_mapping;
3639 const struct address_space_operations *a_ops = mapping->a_ops;
3641 ssize_t written = 0;
3642 unsigned int flags = 0;
3646 unsigned long offset; /* Offset into pagecache page */
3647 unsigned long bytes; /* Bytes to write to page */
3648 size_t copied; /* Bytes copied from user */
3651 offset = (pos & (PAGE_SIZE - 1));
3652 bytes = min_t(unsigned long, PAGE_SIZE - offset,
3657 * Bring in the user page that we will copy from _first_.
3658 * Otherwise there's a nasty deadlock on copying from the
3659 * same page as we're writing to, without it being marked
3662 * Not only is this an optimisation, but it is also required
3663 * to check that the address is actually valid, when atomic
3664 * usercopies are used, below.
3666 if (unlikely(iov_iter_fault_in_readable(i, bytes))) {
3671 if (fatal_signal_pending(current)) {
3676 status = a_ops->write_begin(file, mapping, pos, bytes, flags,
3678 if (unlikely(status < 0))
3681 if (mapping_writably_mapped(mapping))
3682 flush_dcache_page(page);
3684 copied = iov_iter_copy_from_user_atomic(page, i, offset, bytes);
3685 flush_dcache_page(page);
3687 status = a_ops->write_end(file, mapping, pos, bytes, copied,
3689 if (unlikely(status < 0))
3695 iov_iter_advance(i, copied);
3696 if (unlikely(copied == 0)) {
3698 * If we were unable to copy any data at all, we must
3699 * fall back to a single segment length write.
3701 * If we didn't fallback here, we could livelock
3702 * because not all segments in the iov can be copied at
3703 * once without a pagefault.
3705 bytes = min_t(unsigned long, PAGE_SIZE - offset,
3706 iov_iter_single_seg_count(i));
3712 balance_dirty_pages_ratelimited(mapping);
3713 } while (iov_iter_count(i));
3715 return written ? written : status;
3717 EXPORT_SYMBOL(generic_perform_write);
3720 * __generic_file_write_iter - write data to a file
3721 * @iocb: IO state structure (file, offset, etc.)
3722 * @from: iov_iter with data to write
3724 * This function does all the work needed for actually writing data to a
3725 * file. It does all basic checks, removes SUID from the file, updates
3726 * modification times and calls proper subroutines depending on whether we
3727 * do direct IO or a standard buffered write.
3729 * It expects i_mutex to be grabbed unless we work on a block device or similar
3730 * object which does not need locking at all.
3732 * This function does *not* take care of syncing data in case of O_SYNC write.
3733 * A caller has to handle it. This is mainly due to the fact that we want to
3734 * avoid syncing under i_mutex.
3737 * * number of bytes written, even for truncated writes
3738 * * negative error code if no data has been written at all
3740 ssize_t __generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
3742 struct file *file = iocb->ki_filp;
3743 struct address_space * mapping = file->f_mapping;
3744 struct inode *inode = mapping->host;
3745 ssize_t written = 0;
3749 /* We can write back this queue in page reclaim */
3750 current->backing_dev_info = inode_to_bdi(inode);
3751 err = file_remove_privs(file);
3755 err = file_update_time(file);
3759 if (iocb->ki_flags & IOCB_DIRECT) {
3760 loff_t pos, endbyte;
3762 written = generic_file_direct_write(iocb, from);
3764 * If the write stopped short of completing, fall back to
3765 * buffered writes. Some filesystems do this for writes to
3766 * holes, for example. For DAX files, a buffered write will
3767 * not succeed (even if it did, DAX does not handle dirty
3768 * page-cache pages correctly).
3770 if (written < 0 || !iov_iter_count(from) || IS_DAX(inode))
3773 status = generic_perform_write(file, from, pos = iocb->ki_pos);
3775 * If generic_perform_write() returned a synchronous error
3776 * then we want to return the number of bytes which were
3777 * direct-written, or the error code if that was zero. Note
3778 * that this differs from normal direct-io semantics, which
3779 * will return -EFOO even if some bytes were written.
3781 if (unlikely(status < 0)) {
3786 * We need to ensure that the page cache pages are written to
3787 * disk and invalidated to preserve the expected O_DIRECT
3790 endbyte = pos + status - 1;
3791 err = filemap_write_and_wait_range(mapping, pos, endbyte);
3793 iocb->ki_pos = endbyte + 1;
3795 invalidate_mapping_pages(mapping,
3797 endbyte >> PAGE_SHIFT);
3800 * We don't know how much we wrote, so just return
3801 * the number of bytes which were direct-written
3805 written = generic_perform_write(file, from, iocb->ki_pos);
3806 if (likely(written > 0))
3807 iocb->ki_pos += written;
3810 current->backing_dev_info = NULL;
3811 return written ? written : err;
3813 EXPORT_SYMBOL(__generic_file_write_iter);
3816 * generic_file_write_iter - write data to a file
3817 * @iocb: IO state structure
3818 * @from: iov_iter with data to write
3820 * This is a wrapper around __generic_file_write_iter() to be used by most
3821 * filesystems. It takes care of syncing the file in case of O_SYNC file
3822 * and acquires i_mutex as needed.
3824 * * negative error code if no data has been written at all of
3825 * vfs_fsync_range() failed for a synchronous write
3826 * * number of bytes written, even for truncated writes
3828 ssize_t generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
3830 struct file *file = iocb->ki_filp;
3831 struct inode *inode = file->f_mapping->host;
3835 ret = generic_write_checks(iocb, from);
3837 ret = __generic_file_write_iter(iocb, from);
3838 inode_unlock(inode);
3841 ret = generic_write_sync(iocb, ret);
3844 EXPORT_SYMBOL(generic_file_write_iter);
3847 * try_to_release_page() - release old fs-specific metadata on a page
3849 * @page: the page which the kernel is trying to free
3850 * @gfp_mask: memory allocation flags (and I/O mode)
3852 * The address_space is to try to release any data against the page
3853 * (presumably at page->private).
3855 * This may also be called if PG_fscache is set on a page, indicating that the
3856 * page is known to the local caching routines.
3858 * The @gfp_mask argument specifies whether I/O may be performed to release
3859 * this page (__GFP_IO), and whether the call may block (__GFP_RECLAIM & __GFP_FS).
3861 * Return: %1 if the release was successful, otherwise return zero.
3863 int try_to_release_page(struct page *page, gfp_t gfp_mask)
3865 struct address_space * const mapping = page->mapping;
3867 BUG_ON(!PageLocked(page));
3868 if (PageWriteback(page))
3871 if (mapping && mapping->a_ops->releasepage)
3872 return mapping->a_ops->releasepage(page, gfp_mask);
3873 return try_to_free_buffers(page);
3876 EXPORT_SYMBOL(try_to_release_page);