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 __dec_lruvec_page_state(page, NR_SHMEM_THPS);
210 } else if (PageTransHuge(page)) {
211 __dec_lruvec_page_state(page, NR_FILE_THPS);
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_write_and_wait_range - write out & wait on a file range
640 * @mapping: the address_space for the pages
641 * @lstart: offset in bytes where the range starts
642 * @lend: offset in bytes where the range ends (inclusive)
644 * Write out and wait upon file offsets lstart->lend, inclusive.
646 * Note that @lend is inclusive (describes the last byte to be written) so
647 * that this function can be used to write to the very end-of-file (end = -1).
649 * Return: error status of the address space.
651 int filemap_write_and_wait_range(struct address_space *mapping,
652 loff_t lstart, loff_t lend)
656 if (mapping_needs_writeback(mapping)) {
657 err = __filemap_fdatawrite_range(mapping, lstart, lend,
660 * Even if the above returned error, the pages may be
661 * written partially (e.g. -ENOSPC), so we wait for it.
662 * But the -EIO is special case, it may indicate the worst
663 * thing (e.g. bug) happened, so we avoid waiting for it.
666 int err2 = filemap_fdatawait_range(mapping,
671 /* Clear any previously stored errors */
672 filemap_check_errors(mapping);
675 err = filemap_check_errors(mapping);
679 EXPORT_SYMBOL(filemap_write_and_wait_range);
681 void __filemap_set_wb_err(struct address_space *mapping, int err)
683 errseq_t eseq = errseq_set(&mapping->wb_err, err);
685 trace_filemap_set_wb_err(mapping, eseq);
687 EXPORT_SYMBOL(__filemap_set_wb_err);
690 * file_check_and_advance_wb_err - report wb error (if any) that was previously
691 * and advance wb_err to current one
692 * @file: struct file on which the error is being reported
694 * When userland calls fsync (or something like nfsd does the equivalent), we
695 * want to report any writeback errors that occurred since the last fsync (or
696 * since the file was opened if there haven't been any).
698 * Grab the wb_err from the mapping. If it matches what we have in the file,
699 * then just quickly return 0. The file is all caught up.
701 * If it doesn't match, then take the mapping value, set the "seen" flag in
702 * it and try to swap it into place. If it works, or another task beat us
703 * to it with the new value, then update the f_wb_err and return the error
704 * portion. The error at this point must be reported via proper channels
705 * (a'la fsync, or NFS COMMIT operation, etc.).
707 * While we handle mapping->wb_err with atomic operations, the f_wb_err
708 * value is protected by the f_lock since we must ensure that it reflects
709 * the latest value swapped in for this file descriptor.
711 * Return: %0 on success, negative error code otherwise.
713 int file_check_and_advance_wb_err(struct file *file)
716 errseq_t old = READ_ONCE(file->f_wb_err);
717 struct address_space *mapping = file->f_mapping;
719 /* Locklessly handle the common case where nothing has changed */
720 if (errseq_check(&mapping->wb_err, old)) {
721 /* Something changed, must use slow path */
722 spin_lock(&file->f_lock);
723 old = file->f_wb_err;
724 err = errseq_check_and_advance(&mapping->wb_err,
726 trace_file_check_and_advance_wb_err(file, old);
727 spin_unlock(&file->f_lock);
731 * We're mostly using this function as a drop in replacement for
732 * filemap_check_errors. Clear AS_EIO/AS_ENOSPC to emulate the effect
733 * that the legacy code would have had on these flags.
735 clear_bit(AS_EIO, &mapping->flags);
736 clear_bit(AS_ENOSPC, &mapping->flags);
739 EXPORT_SYMBOL(file_check_and_advance_wb_err);
742 * file_write_and_wait_range - write out & wait on a file range
743 * @file: file pointing to address_space with pages
744 * @lstart: offset in bytes where the range starts
745 * @lend: offset in bytes where the range ends (inclusive)
747 * Write out and wait upon file offsets lstart->lend, inclusive.
749 * Note that @lend is inclusive (describes the last byte to be written) so
750 * that this function can be used to write to the very end-of-file (end = -1).
752 * After writing out and waiting on the data, we check and advance the
753 * f_wb_err cursor to the latest value, and return any errors detected there.
755 * Return: %0 on success, negative error code otherwise.
757 int file_write_and_wait_range(struct file *file, loff_t lstart, loff_t lend)
760 struct address_space *mapping = file->f_mapping;
762 if (mapping_needs_writeback(mapping)) {
763 err = __filemap_fdatawrite_range(mapping, lstart, lend,
765 /* See comment of filemap_write_and_wait() */
767 __filemap_fdatawait_range(mapping, lstart, lend);
769 err2 = file_check_and_advance_wb_err(file);
774 EXPORT_SYMBOL(file_write_and_wait_range);
777 * replace_page_cache_page - replace a pagecache page with a new one
778 * @old: page to be replaced
779 * @new: page to replace with
781 * This function replaces a page in the pagecache with a new one. On
782 * success it acquires the pagecache reference for the new page and
783 * drops it for the old page. Both the old and new pages must be
784 * locked. This function does not add the new page to the LRU, the
785 * caller must do that.
787 * The remove + add is atomic. This function cannot fail.
789 void replace_page_cache_page(struct page *old, struct page *new)
791 struct address_space *mapping = old->mapping;
792 void (*freepage)(struct page *) = mapping->a_ops->freepage;
793 pgoff_t offset = old->index;
794 XA_STATE(xas, &mapping->i_pages, offset);
797 VM_BUG_ON_PAGE(!PageLocked(old), old);
798 VM_BUG_ON_PAGE(!PageLocked(new), new);
799 VM_BUG_ON_PAGE(new->mapping, new);
802 new->mapping = mapping;
805 mem_cgroup_migrate(old, new);
807 xas_lock_irqsave(&xas, flags);
808 xas_store(&xas, new);
811 /* hugetlb pages do not participate in page cache accounting. */
813 __dec_lruvec_page_state(old, NR_FILE_PAGES);
815 __inc_lruvec_page_state(new, NR_FILE_PAGES);
816 if (PageSwapBacked(old))
817 __dec_lruvec_page_state(old, NR_SHMEM);
818 if (PageSwapBacked(new))
819 __inc_lruvec_page_state(new, NR_SHMEM);
820 xas_unlock_irqrestore(&xas, flags);
825 EXPORT_SYMBOL_GPL(replace_page_cache_page);
827 noinline int __add_to_page_cache_locked(struct page *page,
828 struct address_space *mapping,
829 pgoff_t offset, gfp_t gfp,
832 XA_STATE(xas, &mapping->i_pages, offset);
833 int huge = PageHuge(page);
835 bool charged = false;
837 VM_BUG_ON_PAGE(!PageLocked(page), page);
838 VM_BUG_ON_PAGE(PageSwapBacked(page), page);
839 mapping_set_update(&xas, mapping);
842 page->mapping = mapping;
843 page->index = offset;
846 error = mem_cgroup_charge(page, current->mm, gfp);
852 gfp &= GFP_RECLAIM_MASK;
855 unsigned int order = xa_get_order(xas.xa, xas.xa_index);
856 void *entry, *old = NULL;
858 if (order > thp_order(page))
859 xas_split_alloc(&xas, xa_load(xas.xa, xas.xa_index),
862 xas_for_each_conflict(&xas, entry) {
864 if (!xa_is_value(entry)) {
865 xas_set_err(&xas, -EEXIST);
873 /* entry may have been split before we acquired lock */
874 order = xa_get_order(xas.xa, xas.xa_index);
875 if (order > thp_order(page)) {
876 xas_split(&xas, old, order);
881 xas_store(&xas, page);
886 mapping->nrexceptional--;
889 /* hugetlb pages do not participate in page cache accounting */
891 __inc_lruvec_page_state(page, NR_FILE_PAGES);
893 xas_unlock_irq(&xas);
894 } while (xas_nomem(&xas, gfp));
896 if (xas_error(&xas)) {
897 error = xas_error(&xas);
899 mem_cgroup_uncharge(page);
903 trace_mm_filemap_add_to_page_cache(page);
906 page->mapping = NULL;
907 /* Leave page->index set: truncation relies upon it */
911 ALLOW_ERROR_INJECTION(__add_to_page_cache_locked, ERRNO);
914 * add_to_page_cache_locked - add a locked page to the pagecache
916 * @mapping: the page's address_space
917 * @offset: page index
918 * @gfp_mask: page allocation mode
920 * This function is used to add a page to the pagecache. It must be locked.
921 * This function does not add the page to the LRU. The caller must do that.
923 * Return: %0 on success, negative error code otherwise.
925 int add_to_page_cache_locked(struct page *page, struct address_space *mapping,
926 pgoff_t offset, gfp_t gfp_mask)
928 return __add_to_page_cache_locked(page, mapping, offset,
931 EXPORT_SYMBOL(add_to_page_cache_locked);
933 int add_to_page_cache_lru(struct page *page, struct address_space *mapping,
934 pgoff_t offset, gfp_t gfp_mask)
939 __SetPageLocked(page);
940 ret = __add_to_page_cache_locked(page, mapping, offset,
943 __ClearPageLocked(page);
946 * The page might have been evicted from cache only
947 * recently, in which case it should be activated like
948 * any other repeatedly accessed page.
949 * The exception is pages getting rewritten; evicting other
950 * data from the working set, only to cache data that will
951 * get overwritten with something else, is a waste of memory.
953 WARN_ON_ONCE(PageActive(page));
954 if (!(gfp_mask & __GFP_WRITE) && shadow)
955 workingset_refault(page, shadow);
960 EXPORT_SYMBOL_GPL(add_to_page_cache_lru);
963 struct page *__page_cache_alloc(gfp_t gfp)
968 if (cpuset_do_page_mem_spread()) {
969 unsigned int cpuset_mems_cookie;
971 cpuset_mems_cookie = read_mems_allowed_begin();
972 n = cpuset_mem_spread_node();
973 page = __alloc_pages_node(n, gfp, 0);
974 } while (!page && read_mems_allowed_retry(cpuset_mems_cookie));
978 return alloc_pages(gfp, 0);
980 EXPORT_SYMBOL(__page_cache_alloc);
984 * In order to wait for pages to become available there must be
985 * waitqueues associated with pages. By using a hash table of
986 * waitqueues where the bucket discipline is to maintain all
987 * waiters on the same queue and wake all when any of the pages
988 * become available, and for the woken contexts to check to be
989 * sure the appropriate page became available, this saves space
990 * at a cost of "thundering herd" phenomena during rare hash
993 #define PAGE_WAIT_TABLE_BITS 8
994 #define PAGE_WAIT_TABLE_SIZE (1 << PAGE_WAIT_TABLE_BITS)
995 static wait_queue_head_t page_wait_table[PAGE_WAIT_TABLE_SIZE] __cacheline_aligned;
997 static wait_queue_head_t *page_waitqueue(struct page *page)
999 return &page_wait_table[hash_ptr(page, PAGE_WAIT_TABLE_BITS)];
1002 void __init pagecache_init(void)
1006 for (i = 0; i < PAGE_WAIT_TABLE_SIZE; i++)
1007 init_waitqueue_head(&page_wait_table[i]);
1009 page_writeback_init();
1013 * The page wait code treats the "wait->flags" somewhat unusually, because
1014 * we have multiple different kinds of waits, not just the usual "exclusive"
1019 * (a) no special bits set:
1021 * We're just waiting for the bit to be released, and when a waker
1022 * calls the wakeup function, we set WQ_FLAG_WOKEN and wake it up,
1023 * and remove it from the wait queue.
1025 * Simple and straightforward.
1027 * (b) WQ_FLAG_EXCLUSIVE:
1029 * The waiter is waiting to get the lock, and only one waiter should
1030 * be woken up to avoid any thundering herd behavior. We'll set the
1031 * WQ_FLAG_WOKEN bit, wake it up, and remove it from the wait queue.
1033 * This is the traditional exclusive wait.
1035 * (c) WQ_FLAG_EXCLUSIVE | WQ_FLAG_CUSTOM:
1037 * The waiter is waiting to get the bit, and additionally wants the
1038 * lock to be transferred to it for fair lock behavior. If the lock
1039 * cannot be taken, we stop walking the wait queue without waking
1042 * This is the "fair lock handoff" case, and in addition to setting
1043 * WQ_FLAG_WOKEN, we set WQ_FLAG_DONE to let the waiter easily see
1044 * that it now has the lock.
1046 static int wake_page_function(wait_queue_entry_t *wait, unsigned mode, int sync, void *arg)
1049 struct wait_page_key *key = arg;
1050 struct wait_page_queue *wait_page
1051 = container_of(wait, struct wait_page_queue, wait);
1053 if (!wake_page_match(wait_page, key))
1057 * If it's a lock handoff wait, we get the bit for it, and
1058 * stop walking (and do not wake it up) if we can't.
1060 flags = wait->flags;
1061 if (flags & WQ_FLAG_EXCLUSIVE) {
1062 if (test_bit(key->bit_nr, &key->page->flags))
1064 if (flags & WQ_FLAG_CUSTOM) {
1065 if (test_and_set_bit(key->bit_nr, &key->page->flags))
1067 flags |= WQ_FLAG_DONE;
1072 * We are holding the wait-queue lock, but the waiter that
1073 * is waiting for this will be checking the flags without
1076 * So update the flags atomically, and wake up the waiter
1077 * afterwards to avoid any races. This store-release pairs
1078 * with the load-acquire in wait_on_page_bit_common().
1080 smp_store_release(&wait->flags, flags | WQ_FLAG_WOKEN);
1081 wake_up_state(wait->private, mode);
1084 * Ok, we have successfully done what we're waiting for,
1085 * and we can unconditionally remove the wait entry.
1087 * Note that this pairs with the "finish_wait()" in the
1088 * waiter, and has to be the absolute last thing we do.
1089 * After this list_del_init(&wait->entry) the wait entry
1090 * might be de-allocated and the process might even have
1093 list_del_init_careful(&wait->entry);
1094 return (flags & WQ_FLAG_EXCLUSIVE) != 0;
1097 static void wake_up_page_bit(struct page *page, int bit_nr)
1099 wait_queue_head_t *q = page_waitqueue(page);
1100 struct wait_page_key key;
1101 unsigned long flags;
1102 wait_queue_entry_t bookmark;
1105 key.bit_nr = bit_nr;
1109 bookmark.private = NULL;
1110 bookmark.func = NULL;
1111 INIT_LIST_HEAD(&bookmark.entry);
1113 spin_lock_irqsave(&q->lock, flags);
1114 __wake_up_locked_key_bookmark(q, TASK_NORMAL, &key, &bookmark);
1116 while (bookmark.flags & WQ_FLAG_BOOKMARK) {
1118 * Take a breather from holding the lock,
1119 * allow pages that finish wake up asynchronously
1120 * to acquire the lock and remove themselves
1123 spin_unlock_irqrestore(&q->lock, flags);
1125 spin_lock_irqsave(&q->lock, flags);
1126 __wake_up_locked_key_bookmark(q, TASK_NORMAL, &key, &bookmark);
1130 * It is possible for other pages to have collided on the waitqueue
1131 * hash, so in that case check for a page match. That prevents a long-
1134 * It is still possible to miss a case here, when we woke page waiters
1135 * and removed them from the waitqueue, but there are still other
1138 if (!waitqueue_active(q) || !key.page_match) {
1139 ClearPageWaiters(page);
1141 * It's possible to miss clearing Waiters here, when we woke
1142 * our page waiters, but the hashed waitqueue has waiters for
1143 * other pages on it.
1145 * That's okay, it's a rare case. The next waker will clear it.
1148 spin_unlock_irqrestore(&q->lock, flags);
1151 static void wake_up_page(struct page *page, int bit)
1153 if (!PageWaiters(page))
1155 wake_up_page_bit(page, bit);
1159 * A choice of three behaviors for wait_on_page_bit_common():
1162 EXCLUSIVE, /* Hold ref to page and take the bit when woken, like
1163 * __lock_page() waiting on then setting PG_locked.
1165 SHARED, /* Hold ref to page and check the bit when woken, like
1166 * wait_on_page_writeback() waiting on PG_writeback.
1168 DROP, /* Drop ref to page before wait, no check when woken,
1169 * like put_and_wait_on_page_locked() on PG_locked.
1174 * Attempt to check (or get) the page bit, and mark us done
1177 static inline bool trylock_page_bit_common(struct page *page, int bit_nr,
1178 struct wait_queue_entry *wait)
1180 if (wait->flags & WQ_FLAG_EXCLUSIVE) {
1181 if (test_and_set_bit(bit_nr, &page->flags))
1183 } else if (test_bit(bit_nr, &page->flags))
1186 wait->flags |= WQ_FLAG_WOKEN | WQ_FLAG_DONE;
1190 /* How many times do we accept lock stealing from under a waiter? */
1191 int sysctl_page_lock_unfairness = 5;
1193 static inline int wait_on_page_bit_common(wait_queue_head_t *q,
1194 struct page *page, int bit_nr, int state, enum behavior behavior)
1196 int unfairness = sysctl_page_lock_unfairness;
1197 struct wait_page_queue wait_page;
1198 wait_queue_entry_t *wait = &wait_page.wait;
1199 bool thrashing = false;
1200 bool delayacct = false;
1201 unsigned long pflags;
1203 if (bit_nr == PG_locked &&
1204 !PageUptodate(page) && PageWorkingset(page)) {
1205 if (!PageSwapBacked(page)) {
1206 delayacct_thrashing_start();
1209 psi_memstall_enter(&pflags);
1214 wait->func = wake_page_function;
1215 wait_page.page = page;
1216 wait_page.bit_nr = bit_nr;
1220 if (behavior == EXCLUSIVE) {
1221 wait->flags = WQ_FLAG_EXCLUSIVE;
1222 if (--unfairness < 0)
1223 wait->flags |= WQ_FLAG_CUSTOM;
1227 * Do one last check whether we can get the
1228 * page bit synchronously.
1230 * Do the SetPageWaiters() marking before that
1231 * to let any waker we _just_ missed know they
1232 * need to wake us up (otherwise they'll never
1233 * even go to the slow case that looks at the
1234 * page queue), and add ourselves to the wait
1235 * queue if we need to sleep.
1237 * This part needs to be done under the queue
1238 * lock to avoid races.
1240 spin_lock_irq(&q->lock);
1241 SetPageWaiters(page);
1242 if (!trylock_page_bit_common(page, bit_nr, wait))
1243 __add_wait_queue_entry_tail(q, wait);
1244 spin_unlock_irq(&q->lock);
1247 * From now on, all the logic will be based on
1248 * the WQ_FLAG_WOKEN and WQ_FLAG_DONE flag, to
1249 * see whether the page bit testing has already
1250 * been done by the wake function.
1252 * We can drop our reference to the page.
1254 if (behavior == DROP)
1258 * Note that until the "finish_wait()", or until
1259 * we see the WQ_FLAG_WOKEN flag, we need to
1260 * be very careful with the 'wait->flags', because
1261 * we may race with a waker that sets them.
1266 set_current_state(state);
1268 /* Loop until we've been woken or interrupted */
1269 flags = smp_load_acquire(&wait->flags);
1270 if (!(flags & WQ_FLAG_WOKEN)) {
1271 if (signal_pending_state(state, current))
1278 /* If we were non-exclusive, we're done */
1279 if (behavior != EXCLUSIVE)
1282 /* If the waker got the lock for us, we're done */
1283 if (flags & WQ_FLAG_DONE)
1287 * Otherwise, if we're getting the lock, we need to
1288 * try to get it ourselves.
1290 * And if that fails, we'll have to retry this all.
1292 if (unlikely(test_and_set_bit(bit_nr, &page->flags)))
1295 wait->flags |= WQ_FLAG_DONE;
1300 * If a signal happened, this 'finish_wait()' may remove the last
1301 * waiter from the wait-queues, but the PageWaiters bit will remain
1302 * set. That's ok. The next wakeup will take care of it, and trying
1303 * to do it here would be difficult and prone to races.
1305 finish_wait(q, wait);
1309 delayacct_thrashing_end();
1310 psi_memstall_leave(&pflags);
1314 * NOTE! The wait->flags weren't stable until we've done the
1315 * 'finish_wait()', and we could have exited the loop above due
1316 * to a signal, and had a wakeup event happen after the signal
1317 * test but before the 'finish_wait()'.
1319 * So only after the finish_wait() can we reliably determine
1320 * if we got woken up or not, so we can now figure out the final
1321 * return value based on that state without races.
1323 * Also note that WQ_FLAG_WOKEN is sufficient for a non-exclusive
1324 * waiter, but an exclusive one requires WQ_FLAG_DONE.
1326 if (behavior == EXCLUSIVE)
1327 return wait->flags & WQ_FLAG_DONE ? 0 : -EINTR;
1329 return wait->flags & WQ_FLAG_WOKEN ? 0 : -EINTR;
1332 void wait_on_page_bit(struct page *page, int bit_nr)
1334 wait_queue_head_t *q = page_waitqueue(page);
1335 wait_on_page_bit_common(q, page, bit_nr, TASK_UNINTERRUPTIBLE, SHARED);
1337 EXPORT_SYMBOL(wait_on_page_bit);
1339 int wait_on_page_bit_killable(struct page *page, int bit_nr)
1341 wait_queue_head_t *q = page_waitqueue(page);
1342 return wait_on_page_bit_common(q, page, bit_nr, TASK_KILLABLE, SHARED);
1344 EXPORT_SYMBOL(wait_on_page_bit_killable);
1346 static int __wait_on_page_locked_async(struct page *page,
1347 struct wait_page_queue *wait, bool set)
1349 struct wait_queue_head *q = page_waitqueue(page);
1353 wait->bit_nr = PG_locked;
1355 spin_lock_irq(&q->lock);
1356 __add_wait_queue_entry_tail(q, &wait->wait);
1357 SetPageWaiters(page);
1359 ret = !trylock_page(page);
1361 ret = PageLocked(page);
1363 * If we were successful now, we know we're still on the
1364 * waitqueue as we're still under the lock. This means it's
1365 * safe to remove and return success, we know the callback
1366 * isn't going to trigger.
1369 __remove_wait_queue(q, &wait->wait);
1372 spin_unlock_irq(&q->lock);
1376 static int wait_on_page_locked_async(struct page *page,
1377 struct wait_page_queue *wait)
1379 if (!PageLocked(page))
1381 return __wait_on_page_locked_async(compound_head(page), wait, false);
1385 * put_and_wait_on_page_locked - Drop a reference and wait for it to be unlocked
1386 * @page: The page to wait for.
1388 * The caller should hold a reference on @page. They expect the page to
1389 * become unlocked relatively soon, but do not wish to hold up migration
1390 * (for example) by holding the reference while waiting for the page to
1391 * come unlocked. After this function returns, the caller should not
1392 * dereference @page.
1394 void put_and_wait_on_page_locked(struct page *page)
1396 wait_queue_head_t *q;
1398 page = compound_head(page);
1399 q = page_waitqueue(page);
1400 wait_on_page_bit_common(q, page, PG_locked, TASK_UNINTERRUPTIBLE, DROP);
1404 * add_page_wait_queue - Add an arbitrary waiter to a page's wait queue
1405 * @page: Page defining the wait queue of interest
1406 * @waiter: Waiter to add to the queue
1408 * Add an arbitrary @waiter to the wait queue for the nominated @page.
1410 void add_page_wait_queue(struct page *page, wait_queue_entry_t *waiter)
1412 wait_queue_head_t *q = page_waitqueue(page);
1413 unsigned long flags;
1415 spin_lock_irqsave(&q->lock, flags);
1416 __add_wait_queue_entry_tail(q, waiter);
1417 SetPageWaiters(page);
1418 spin_unlock_irqrestore(&q->lock, flags);
1420 EXPORT_SYMBOL_GPL(add_page_wait_queue);
1422 #ifndef clear_bit_unlock_is_negative_byte
1425 * PG_waiters is the high bit in the same byte as PG_lock.
1427 * On x86 (and on many other architectures), we can clear PG_lock and
1428 * test the sign bit at the same time. But if the architecture does
1429 * not support that special operation, we just do this all by hand
1432 * The read of PG_waiters has to be after (or concurrently with) PG_locked
1433 * being cleared, but a memory barrier should be unnecessary since it is
1434 * in the same byte as PG_locked.
1436 static inline bool clear_bit_unlock_is_negative_byte(long nr, volatile void *mem)
1438 clear_bit_unlock(nr, mem);
1439 /* smp_mb__after_atomic(); */
1440 return test_bit(PG_waiters, mem);
1446 * unlock_page - unlock a locked page
1449 * Unlocks the page and wakes up sleepers in wait_on_page_locked().
1450 * Also wakes sleepers in wait_on_page_writeback() because the wakeup
1451 * mechanism between PageLocked pages and PageWriteback pages is shared.
1452 * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep.
1454 * Note that this depends on PG_waiters being the sign bit in the byte
1455 * that contains PG_locked - thus the BUILD_BUG_ON(). That allows us to
1456 * clear the PG_locked bit and test PG_waiters at the same time fairly
1457 * portably (architectures that do LL/SC can test any bit, while x86 can
1458 * test the sign bit).
1460 void unlock_page(struct page *page)
1462 BUILD_BUG_ON(PG_waiters != 7);
1463 page = compound_head(page);
1464 VM_BUG_ON_PAGE(!PageLocked(page), page);
1465 if (clear_bit_unlock_is_negative_byte(PG_locked, &page->flags))
1466 wake_up_page_bit(page, PG_locked);
1468 EXPORT_SYMBOL(unlock_page);
1471 * end_page_writeback - end writeback against a page
1474 void end_page_writeback(struct page *page)
1477 * TestClearPageReclaim could be used here but it is an atomic
1478 * operation and overkill in this particular case. Failing to
1479 * shuffle a page marked for immediate reclaim is too mild to
1480 * justify taking an atomic operation penalty at the end of
1481 * ever page writeback.
1483 if (PageReclaim(page)) {
1484 ClearPageReclaim(page);
1485 rotate_reclaimable_page(page);
1489 * Writeback does not hold a page reference of its own, relying
1490 * on truncation to wait for the clearing of PG_writeback.
1491 * But here we must make sure that the page is not freed and
1492 * reused before the wake_up_page().
1495 if (!test_clear_page_writeback(page))
1498 smp_mb__after_atomic();
1499 wake_up_page(page, PG_writeback);
1502 EXPORT_SYMBOL(end_page_writeback);
1505 * After completing I/O on a page, call this routine to update the page
1506 * flags appropriately
1508 void page_endio(struct page *page, bool is_write, int err)
1512 SetPageUptodate(page);
1514 ClearPageUptodate(page);
1520 struct address_space *mapping;
1523 mapping = page_mapping(page);
1525 mapping_set_error(mapping, err);
1527 end_page_writeback(page);
1530 EXPORT_SYMBOL_GPL(page_endio);
1533 * __lock_page - get a lock on the page, assuming we need to sleep to get it
1534 * @__page: the page to lock
1536 void __lock_page(struct page *__page)
1538 struct page *page = compound_head(__page);
1539 wait_queue_head_t *q = page_waitqueue(page);
1540 wait_on_page_bit_common(q, page, PG_locked, TASK_UNINTERRUPTIBLE,
1543 EXPORT_SYMBOL(__lock_page);
1545 int __lock_page_killable(struct page *__page)
1547 struct page *page = compound_head(__page);
1548 wait_queue_head_t *q = page_waitqueue(page);
1549 return wait_on_page_bit_common(q, page, PG_locked, TASK_KILLABLE,
1552 EXPORT_SYMBOL_GPL(__lock_page_killable);
1554 int __lock_page_async(struct page *page, struct wait_page_queue *wait)
1556 return __wait_on_page_locked_async(page, wait, true);
1561 * 1 - page is locked; mmap_lock is still held.
1562 * 0 - page is not locked.
1563 * mmap_lock has been released (mmap_read_unlock(), unless flags had both
1564 * FAULT_FLAG_ALLOW_RETRY and FAULT_FLAG_RETRY_NOWAIT set, in
1565 * which case mmap_lock is still held.
1567 * If neither ALLOW_RETRY nor KILLABLE are set, will always return 1
1568 * with the page locked and the mmap_lock unperturbed.
1570 int __lock_page_or_retry(struct page *page, struct mm_struct *mm,
1573 if (fault_flag_allow_retry_first(flags)) {
1575 * CAUTION! In this case, mmap_lock is not released
1576 * even though return 0.
1578 if (flags & FAULT_FLAG_RETRY_NOWAIT)
1581 mmap_read_unlock(mm);
1582 if (flags & FAULT_FLAG_KILLABLE)
1583 wait_on_page_locked_killable(page);
1585 wait_on_page_locked(page);
1588 if (flags & FAULT_FLAG_KILLABLE) {
1591 ret = __lock_page_killable(page);
1593 mmap_read_unlock(mm);
1604 * page_cache_next_miss() - Find the next gap in the page cache.
1605 * @mapping: Mapping.
1607 * @max_scan: Maximum range to search.
1609 * Search the range [index, min(index + max_scan - 1, ULONG_MAX)] for the
1610 * gap with the lowest index.
1612 * This function may be called under the rcu_read_lock. However, this will
1613 * not atomically search a snapshot of the cache at a single point in time.
1614 * For example, if a gap is created at index 5, then subsequently a gap is
1615 * created at index 10, page_cache_next_miss covering both indices may
1616 * return 10 if called under the rcu_read_lock.
1618 * Return: The index of the gap if found, otherwise an index outside the
1619 * range specified (in which case 'return - index >= max_scan' will be true).
1620 * In the rare case of index wrap-around, 0 will be returned.
1622 pgoff_t page_cache_next_miss(struct address_space *mapping,
1623 pgoff_t index, unsigned long max_scan)
1625 XA_STATE(xas, &mapping->i_pages, index);
1627 while (max_scan--) {
1628 void *entry = xas_next(&xas);
1629 if (!entry || xa_is_value(entry))
1631 if (xas.xa_index == 0)
1635 return xas.xa_index;
1637 EXPORT_SYMBOL(page_cache_next_miss);
1640 * page_cache_prev_miss() - Find the previous gap in the page cache.
1641 * @mapping: Mapping.
1643 * @max_scan: Maximum range to search.
1645 * Search the range [max(index - max_scan + 1, 0), index] for the
1646 * gap with the highest index.
1648 * This function may be called under the rcu_read_lock. However, this will
1649 * not atomically search a snapshot of the cache at a single point in time.
1650 * For example, if a gap is created at index 10, then subsequently a gap is
1651 * created at index 5, page_cache_prev_miss() covering both indices may
1652 * return 5 if called under the rcu_read_lock.
1654 * Return: The index of the gap if found, otherwise an index outside the
1655 * range specified (in which case 'index - return >= max_scan' will be true).
1656 * In the rare case of wrap-around, ULONG_MAX will be returned.
1658 pgoff_t page_cache_prev_miss(struct address_space *mapping,
1659 pgoff_t index, unsigned long max_scan)
1661 XA_STATE(xas, &mapping->i_pages, index);
1663 while (max_scan--) {
1664 void *entry = xas_prev(&xas);
1665 if (!entry || xa_is_value(entry))
1667 if (xas.xa_index == ULONG_MAX)
1671 return xas.xa_index;
1673 EXPORT_SYMBOL(page_cache_prev_miss);
1676 * find_get_entry - find and get a page cache entry
1677 * @mapping: the address_space to search
1678 * @index: The page cache index.
1680 * Looks up the page cache slot at @mapping & @offset. If there is a
1681 * page cache page, the head page is returned with an increased refcount.
1683 * If the slot holds a shadow entry of a previously evicted page, or a
1684 * swap entry from shmem/tmpfs, it is returned.
1686 * Return: The head page or shadow entry, %NULL if nothing is found.
1688 struct page *find_get_entry(struct address_space *mapping, pgoff_t index)
1690 XA_STATE(xas, &mapping->i_pages, index);
1696 page = xas_load(&xas);
1697 if (xas_retry(&xas, page))
1700 * A shadow entry of a recently evicted page, or a swap entry from
1701 * shmem/tmpfs. Return it without attempting to raise page count.
1703 if (!page || xa_is_value(page))
1706 if (!page_cache_get_speculative(page))
1710 * Has the page moved or been split?
1711 * This is part of the lockless pagecache protocol. See
1712 * include/linux/pagemap.h for details.
1714 if (unlikely(page != xas_reload(&xas))) {
1725 * find_lock_entry - Locate and lock a page cache entry.
1726 * @mapping: The address_space to search.
1727 * @index: The page cache index.
1729 * Looks up the page at @mapping & @index. If there is a page in the
1730 * cache, the head page is returned locked and with an increased refcount.
1732 * If the slot holds a shadow entry of a previously evicted page, or a
1733 * swap entry from shmem/tmpfs, it is returned.
1735 * Context: May sleep.
1736 * Return: The head page or shadow entry, %NULL if nothing is found.
1738 struct page *find_lock_entry(struct address_space *mapping, pgoff_t index)
1743 page = find_get_entry(mapping, index);
1744 if (page && !xa_is_value(page)) {
1746 /* Has the page been truncated? */
1747 if (unlikely(page->mapping != mapping)) {
1752 VM_BUG_ON_PAGE(!thp_contains(page, index), page);
1758 * pagecache_get_page - Find and get a reference to a page.
1759 * @mapping: The address_space to search.
1760 * @index: The page index.
1761 * @fgp_flags: %FGP flags modify how the page is returned.
1762 * @gfp_mask: Memory allocation flags to use if %FGP_CREAT is specified.
1764 * Looks up the page cache entry at @mapping & @index.
1766 * @fgp_flags can be zero or more of these flags:
1768 * * %FGP_ACCESSED - The page will be marked accessed.
1769 * * %FGP_LOCK - The page is returned locked.
1770 * * %FGP_HEAD - If the page is present and a THP, return the head page
1771 * rather than the exact page specified by the index.
1772 * * %FGP_CREAT - If no page is present then a new page is allocated using
1773 * @gfp_mask and added to the page cache and the VM's LRU list.
1774 * The page is returned locked and with an increased refcount.
1775 * * %FGP_FOR_MMAP - The caller wants to do its own locking dance if the
1776 * page is already in cache. If the page was allocated, unlock it before
1777 * returning so the caller can do the same dance.
1778 * * %FGP_WRITE - The page will be written
1779 * * %FGP_NOFS - __GFP_FS will get cleared in gfp mask
1780 * * %FGP_NOWAIT - Don't get blocked by page lock
1782 * If %FGP_LOCK or %FGP_CREAT are specified then the function may sleep even
1783 * if the %GFP flags specified for %FGP_CREAT are atomic.
1785 * If there is a page cache page, it is returned with an increased refcount.
1787 * Return: The found page or %NULL otherwise.
1789 struct page *pagecache_get_page(struct address_space *mapping, pgoff_t index,
1790 int fgp_flags, gfp_t gfp_mask)
1795 page = find_get_entry(mapping, index);
1796 if (xa_is_value(page))
1801 if (fgp_flags & FGP_LOCK) {
1802 if (fgp_flags & FGP_NOWAIT) {
1803 if (!trylock_page(page)) {
1811 /* Has the page been truncated? */
1812 if (unlikely(page->mapping != mapping)) {
1817 VM_BUG_ON_PAGE(!thp_contains(page, index), page);
1820 if (fgp_flags & FGP_ACCESSED)
1821 mark_page_accessed(page);
1822 else if (fgp_flags & FGP_WRITE) {
1823 /* Clear idle flag for buffer write */
1824 if (page_is_idle(page))
1825 clear_page_idle(page);
1827 if (!(fgp_flags & FGP_HEAD))
1828 page = find_subpage(page, index);
1831 if (!page && (fgp_flags & FGP_CREAT)) {
1833 if ((fgp_flags & FGP_WRITE) && mapping_can_writeback(mapping))
1834 gfp_mask |= __GFP_WRITE;
1835 if (fgp_flags & FGP_NOFS)
1836 gfp_mask &= ~__GFP_FS;
1838 page = __page_cache_alloc(gfp_mask);
1842 if (WARN_ON_ONCE(!(fgp_flags & (FGP_LOCK | FGP_FOR_MMAP))))
1843 fgp_flags |= FGP_LOCK;
1845 /* Init accessed so avoid atomic mark_page_accessed later */
1846 if (fgp_flags & FGP_ACCESSED)
1847 __SetPageReferenced(page);
1849 err = add_to_page_cache_lru(page, mapping, index, gfp_mask);
1850 if (unlikely(err)) {
1858 * add_to_page_cache_lru locks the page, and for mmap we expect
1861 if (page && (fgp_flags & FGP_FOR_MMAP))
1867 EXPORT_SYMBOL(pagecache_get_page);
1870 * find_get_entries - gang pagecache lookup
1871 * @mapping: The address_space to search
1872 * @start: The starting page cache index
1873 * @nr_entries: The maximum number of entries
1874 * @entries: Where the resulting entries are placed
1875 * @indices: The cache indices corresponding to the entries in @entries
1877 * find_get_entries() will search for and return a group of up to
1878 * @nr_entries entries in the mapping. The entries are placed at
1879 * @entries. find_get_entries() takes a reference against any actual
1882 * The search returns a group of mapping-contiguous page cache entries
1883 * with ascending indexes. There may be holes in the indices due to
1884 * not-present pages.
1886 * Any shadow entries of evicted pages, or swap entries from
1887 * shmem/tmpfs, are included in the returned array.
1889 * If it finds a Transparent Huge Page, head or tail, find_get_entries()
1890 * stops at that page: the caller is likely to have a better way to handle
1891 * the compound page as a whole, and then skip its extent, than repeatedly
1892 * calling find_get_entries() to return all its tails.
1894 * Return: the number of pages and shadow entries which were found.
1896 unsigned find_get_entries(struct address_space *mapping,
1897 pgoff_t start, unsigned int nr_entries,
1898 struct page **entries, pgoff_t *indices)
1900 XA_STATE(xas, &mapping->i_pages, start);
1902 unsigned int ret = 0;
1908 xas_for_each(&xas, page, ULONG_MAX) {
1909 if (xas_retry(&xas, page))
1912 * A shadow entry of a recently evicted page, a swap
1913 * entry from shmem/tmpfs or a DAX entry. Return it
1914 * without attempting to raise page count.
1916 if (xa_is_value(page))
1919 if (!page_cache_get_speculative(page))
1922 /* Has the page moved or been split? */
1923 if (unlikely(page != xas_reload(&xas)))
1927 * Terminate early on finding a THP, to allow the caller to
1928 * handle it all at once; but continue if this is hugetlbfs.
1930 if (PageTransHuge(page) && !PageHuge(page)) {
1931 page = find_subpage(page, xas.xa_index);
1932 nr_entries = ret + 1;
1935 indices[ret] = xas.xa_index;
1936 entries[ret] = page;
1937 if (++ret == nr_entries)
1950 * find_get_pages_range - gang pagecache lookup
1951 * @mapping: The address_space to search
1952 * @start: The starting page index
1953 * @end: The final page index (inclusive)
1954 * @nr_pages: The maximum number of pages
1955 * @pages: Where the resulting pages are placed
1957 * find_get_pages_range() will search for and return a group of up to @nr_pages
1958 * pages in the mapping starting at index @start and up to index @end
1959 * (inclusive). The pages are placed at @pages. find_get_pages_range() takes
1960 * a reference against the returned pages.
1962 * The search returns a group of mapping-contiguous pages with ascending
1963 * indexes. There may be holes in the indices due to not-present pages.
1964 * We also update @start to index the next page for the traversal.
1966 * Return: the number of pages which were found. If this number is
1967 * smaller than @nr_pages, the end of specified range has been
1970 unsigned find_get_pages_range(struct address_space *mapping, pgoff_t *start,
1971 pgoff_t end, unsigned int nr_pages,
1972 struct page **pages)
1974 XA_STATE(xas, &mapping->i_pages, *start);
1978 if (unlikely(!nr_pages))
1982 xas_for_each(&xas, page, end) {
1983 if (xas_retry(&xas, page))
1985 /* Skip over shadow, swap and DAX entries */
1986 if (xa_is_value(page))
1989 if (!page_cache_get_speculative(page))
1992 /* Has the page moved or been split? */
1993 if (unlikely(page != xas_reload(&xas)))
1996 pages[ret] = find_subpage(page, xas.xa_index);
1997 if (++ret == nr_pages) {
1998 *start = xas.xa_index + 1;
2009 * We come here when there is no page beyond @end. We take care to not
2010 * overflow the index @start as it confuses some of the callers. This
2011 * breaks the iteration when there is a page at index -1 but that is
2012 * already broken anyway.
2014 if (end == (pgoff_t)-1)
2015 *start = (pgoff_t)-1;
2025 * find_get_pages_contig - gang contiguous pagecache lookup
2026 * @mapping: The address_space to search
2027 * @index: The starting page index
2028 * @nr_pages: The maximum number of pages
2029 * @pages: Where the resulting pages are placed
2031 * find_get_pages_contig() works exactly like find_get_pages(), except
2032 * that the returned number of pages are guaranteed to be contiguous.
2034 * Return: the number of pages which were found.
2036 unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index,
2037 unsigned int nr_pages, struct page **pages)
2039 XA_STATE(xas, &mapping->i_pages, index);
2041 unsigned int ret = 0;
2043 if (unlikely(!nr_pages))
2047 for (page = xas_load(&xas); page; page = xas_next(&xas)) {
2048 if (xas_retry(&xas, page))
2051 * If the entry has been swapped out, we can stop looking.
2052 * No current caller is looking for DAX entries.
2054 if (xa_is_value(page))
2057 if (!page_cache_get_speculative(page))
2060 /* Has the page moved or been split? */
2061 if (unlikely(page != xas_reload(&xas)))
2064 pages[ret] = find_subpage(page, xas.xa_index);
2065 if (++ret == nr_pages)
2076 EXPORT_SYMBOL(find_get_pages_contig);
2079 * find_get_pages_range_tag - find and return pages in given range matching @tag
2080 * @mapping: the address_space to search
2081 * @index: the starting page index
2082 * @end: The final page index (inclusive)
2083 * @tag: the tag index
2084 * @nr_pages: the maximum number of pages
2085 * @pages: where the resulting pages are placed
2087 * Like find_get_pages, except we only return pages which are tagged with
2088 * @tag. We update @index to index the next page for the traversal.
2090 * Return: the number of pages which were found.
2092 unsigned find_get_pages_range_tag(struct address_space *mapping, pgoff_t *index,
2093 pgoff_t end, xa_mark_t tag, unsigned int nr_pages,
2094 struct page **pages)
2096 XA_STATE(xas, &mapping->i_pages, *index);
2100 if (unlikely(!nr_pages))
2104 xas_for_each_marked(&xas, page, end, tag) {
2105 if (xas_retry(&xas, page))
2108 * Shadow entries should never be tagged, but this iteration
2109 * is lockless so there is a window for page reclaim to evict
2110 * a page we saw tagged. Skip over it.
2112 if (xa_is_value(page))
2115 if (!page_cache_get_speculative(page))
2118 /* Has the page moved or been split? */
2119 if (unlikely(page != xas_reload(&xas)))
2122 pages[ret] = find_subpage(page, xas.xa_index);
2123 if (++ret == nr_pages) {
2124 *index = xas.xa_index + 1;
2135 * We come here when we got to @end. We take care to not overflow the
2136 * index @index as it confuses some of the callers. This breaks the
2137 * iteration when there is a page at index -1 but that is already
2140 if (end == (pgoff_t)-1)
2141 *index = (pgoff_t)-1;
2149 EXPORT_SYMBOL(find_get_pages_range_tag);
2152 * CD/DVDs are error prone. When a medium error occurs, the driver may fail
2153 * a _large_ part of the i/o request. Imagine the worst scenario:
2155 * ---R__________________________________________B__________
2156 * ^ reading here ^ bad block(assume 4k)
2158 * read(R) => miss => readahead(R...B) => media error => frustrating retries
2159 * => failing the whole request => read(R) => read(R+1) =>
2160 * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
2161 * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
2162 * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
2164 * It is going insane. Fix it by quickly scaling down the readahead size.
2166 static void shrink_readahead_size_eio(struct file_ra_state *ra)
2171 static int lock_page_for_iocb(struct kiocb *iocb, struct page *page)
2173 if (iocb->ki_flags & IOCB_WAITQ)
2174 return lock_page_async(page, iocb->ki_waitq);
2175 else if (iocb->ki_flags & IOCB_NOWAIT)
2176 return trylock_page(page) ? 0 : -EAGAIN;
2178 return lock_page_killable(page);
2181 static struct page *filemap_read_page(struct kiocb *iocb, struct file *filp,
2182 struct address_space *mapping, struct page *page)
2184 struct file_ra_state *ra = &filp->f_ra;
2187 if (iocb->ki_flags & (IOCB_NOIO | IOCB_NOWAIT)) {
2190 return ERR_PTR(-EAGAIN);
2194 * A previous I/O error may have been due to temporary
2195 * failures, eg. multipath errors.
2196 * PG_error will be set again if readpage fails.
2198 ClearPageError(page);
2199 /* Start the actual read. The read will unlock the page. */
2200 error = mapping->a_ops->readpage(filp, page);
2202 if (unlikely(error)) {
2204 return error != AOP_TRUNCATED_PAGE ? ERR_PTR(error) : NULL;
2207 if (!PageUptodate(page)) {
2208 error = lock_page_for_iocb(iocb, page);
2209 if (unlikely(error)) {
2211 return ERR_PTR(error);
2213 if (!PageUptodate(page)) {
2214 if (page->mapping == NULL) {
2216 * invalidate_mapping_pages got it
2223 shrink_readahead_size_eio(ra);
2225 return ERR_PTR(-EIO);
2233 static struct page *filemap_update_page(struct kiocb *iocb, struct file *filp,
2234 struct iov_iter *iter, struct page *page, loff_t pos,
2237 struct address_space *mapping = filp->f_mapping;
2238 struct inode *inode = mapping->host;
2242 * See comment in do_read_cache_page on why
2243 * wait_on_page_locked is used to avoid unnecessarily
2244 * serialisations and why it's safe.
2246 if (iocb->ki_flags & IOCB_WAITQ) {
2247 error = wait_on_page_locked_async(page,
2250 error = wait_on_page_locked_killable(page);
2252 if (unlikely(error)) {
2254 return ERR_PTR(error);
2256 if (PageUptodate(page))
2259 if (inode->i_blkbits == PAGE_SHIFT ||
2260 !mapping->a_ops->is_partially_uptodate)
2261 goto page_not_up_to_date;
2262 /* pipes can't handle partially uptodate pages */
2263 if (unlikely(iov_iter_is_pipe(iter)))
2264 goto page_not_up_to_date;
2265 if (!trylock_page(page))
2266 goto page_not_up_to_date;
2267 /* Did it get truncated before we got the lock? */
2269 goto page_not_up_to_date_locked;
2270 if (!mapping->a_ops->is_partially_uptodate(page,
2271 pos & ~PAGE_MASK, count))
2272 goto page_not_up_to_date_locked;
2276 page_not_up_to_date:
2277 /* Get exclusive access to the page ... */
2278 error = lock_page_for_iocb(iocb, page);
2279 if (unlikely(error)) {
2281 return ERR_PTR(error);
2284 page_not_up_to_date_locked:
2285 /* Did it get truncated before we got the lock? */
2286 if (!page->mapping) {
2292 /* Did somebody else fill it already? */
2293 if (PageUptodate(page)) {
2298 return filemap_read_page(iocb, filp, mapping, page);
2301 static struct page *filemap_create_page(struct kiocb *iocb,
2302 struct iov_iter *iter)
2304 struct file *filp = iocb->ki_filp;
2305 struct address_space *mapping = filp->f_mapping;
2306 pgoff_t index = iocb->ki_pos >> PAGE_SHIFT;
2310 if (iocb->ki_flags & IOCB_NOIO)
2311 return ERR_PTR(-EAGAIN);
2313 page = page_cache_alloc(mapping);
2315 return ERR_PTR(-ENOMEM);
2317 error = add_to_page_cache_lru(page, mapping, index,
2318 mapping_gfp_constraint(mapping, GFP_KERNEL));
2321 return error != -EEXIST ? ERR_PTR(error) : NULL;
2324 return filemap_read_page(iocb, filp, mapping, page);
2327 static int filemap_get_pages(struct kiocb *iocb, struct iov_iter *iter,
2328 struct pagevec *pvec)
2330 struct file *filp = iocb->ki_filp;
2331 struct address_space *mapping = filp->f_mapping;
2332 struct file_ra_state *ra = &filp->f_ra;
2333 pgoff_t index = iocb->ki_pos >> PAGE_SHIFT;
2334 pgoff_t last_index = (iocb->ki_pos + iter->count + PAGE_SIZE-1) >> PAGE_SHIFT;
2335 unsigned int nr = min_t(unsigned long, last_index - index, PAGEVEC_SIZE);
2339 if (fatal_signal_pending(current))
2342 pvec->nr = find_get_pages_contig(mapping, index, nr, pvec->pages);
2346 if (iocb->ki_flags & IOCB_NOIO)
2349 page_cache_sync_readahead(mapping, ra, filp, index, last_index - index);
2351 pvec->nr = find_get_pages_contig(mapping, index, nr, pvec->pages);
2355 pvec->pages[0] = filemap_create_page(iocb, iter);
2356 err = PTR_ERR_OR_ZERO(pvec->pages[0]);
2357 if (!IS_ERR_OR_NULL(pvec->pages[0]))
2360 for (i = 0; i < pvec->nr; i++) {
2361 struct page *page = pvec->pages[i];
2362 pgoff_t pg_index = index + i;
2363 loff_t pg_pos = max(iocb->ki_pos,
2364 (loff_t) pg_index << PAGE_SHIFT);
2365 loff_t pg_count = iocb->ki_pos + iter->count - pg_pos;
2367 if (PageReadahead(page)) {
2368 if (iocb->ki_flags & IOCB_NOIO) {
2369 for (j = i; j < pvec->nr; j++)
2370 put_page(pvec->pages[j]);
2375 page_cache_async_readahead(mapping, ra, filp, page,
2376 pg_index, last_index - pg_index);
2379 if (!PageUptodate(page)) {
2380 if ((iocb->ki_flags & IOCB_NOWAIT) ||
2381 ((iocb->ki_flags & IOCB_WAITQ) && i)) {
2382 for (j = i; j < pvec->nr; j++)
2383 put_page(pvec->pages[j]);
2389 page = filemap_update_page(iocb, filp, iter, page,
2391 if (IS_ERR_OR_NULL(page)) {
2392 for (j = i + 1; j < pvec->nr; j++)
2393 put_page(pvec->pages[j]);
2395 err = PTR_ERR_OR_ZERO(page);
2401 if (likely(pvec->nr))
2406 * No pages and no error means we raced and should retry:
2412 * generic_file_buffered_read - generic file read routine
2413 * @iocb: the iocb to read
2414 * @iter: data destination
2415 * @written: already copied
2417 * This is a generic file read routine, and uses the
2418 * mapping->a_ops->readpage() function for the actual low-level stuff.
2420 * This is really ugly. But the goto's actually try to clarify some
2421 * of the logic when it comes to error handling etc.
2424 * * total number of bytes copied, including those the were already @written
2425 * * negative error code if nothing was copied
2427 ssize_t generic_file_buffered_read(struct kiocb *iocb,
2428 struct iov_iter *iter, ssize_t written)
2430 struct file *filp = iocb->ki_filp;
2431 struct file_ra_state *ra = &filp->f_ra;
2432 struct address_space *mapping = filp->f_mapping;
2433 struct inode *inode = mapping->host;
2434 struct pagevec pvec;
2436 bool writably_mapped;
2437 loff_t isize, end_offset;
2439 if (unlikely(iocb->ki_pos >= inode->i_sb->s_maxbytes))
2441 if (unlikely(!iov_iter_count(iter)))
2444 iov_iter_truncate(iter, inode->i_sb->s_maxbytes);
2450 * If we've already successfully copied some data, then we
2451 * can no longer safely return -EIOCBQUEUED. Hence mark
2452 * an async read NOWAIT at that point.
2454 if ((iocb->ki_flags & IOCB_WAITQ) && written)
2455 iocb->ki_flags |= IOCB_NOWAIT;
2457 error = filemap_get_pages(iocb, iter, &pvec);
2462 * i_size must be checked after we know the pages are Uptodate.
2464 * Checking i_size after the check allows us to calculate
2465 * the correct value for "nr", which means the zero-filled
2466 * part of the page is not copied back to userspace (unless
2467 * another truncate extends the file - this is desired though).
2469 isize = i_size_read(inode);
2470 if (unlikely(iocb->ki_pos >= isize))
2473 end_offset = min_t(loff_t, isize, iocb->ki_pos + iter->count);
2475 while ((iocb->ki_pos >> PAGE_SHIFT) + pvec.nr >
2476 (end_offset + PAGE_SIZE - 1) >> PAGE_SHIFT)
2477 put_page(pvec.pages[--pvec.nr]);
2480 * Once we start copying data, we don't want to be touching any
2481 * cachelines that might be contended:
2483 writably_mapped = mapping_writably_mapped(mapping);
2486 * When a sequential read accesses a page several times, only
2487 * mark it as accessed the first time.
2489 if (iocb->ki_pos >> PAGE_SHIFT !=
2490 ra->prev_pos >> PAGE_SHIFT)
2491 mark_page_accessed(pvec.pages[0]);
2492 for (i = 1; i < pagevec_count(&pvec); i++)
2493 mark_page_accessed(pvec.pages[i]);
2495 for (i = 0; i < pagevec_count(&pvec); i++) {
2496 unsigned int offset = iocb->ki_pos & ~PAGE_MASK;
2497 unsigned int bytes = min_t(loff_t, end_offset - iocb->ki_pos,
2498 PAGE_SIZE - offset);
2499 unsigned int copied;
2502 * If users can be writing to this page using arbitrary
2503 * virtual addresses, take care about potential aliasing
2504 * before reading the page on the kernel side.
2506 if (writably_mapped)
2507 flush_dcache_page(pvec.pages[i]);
2509 copied = copy_page_to_iter(pvec.pages[i], offset, bytes, iter);
2512 iocb->ki_pos += copied;
2513 ra->prev_pos = iocb->ki_pos;
2515 if (copied < bytes) {
2521 for (i = 0; i < pagevec_count(&pvec); i++)
2522 put_page(pvec.pages[i]);
2523 } while (iov_iter_count(iter) && iocb->ki_pos < isize && !error);
2525 file_accessed(filp);
2527 return written ? written : error;
2529 EXPORT_SYMBOL_GPL(generic_file_buffered_read);
2532 * generic_file_read_iter - generic filesystem read routine
2533 * @iocb: kernel I/O control block
2534 * @iter: destination for the data read
2536 * This is the "read_iter()" routine for all filesystems
2537 * that can use the page cache directly.
2539 * The IOCB_NOWAIT flag in iocb->ki_flags indicates that -EAGAIN shall
2540 * be returned when no data can be read without waiting for I/O requests
2541 * to complete; it doesn't prevent readahead.
2543 * The IOCB_NOIO flag in iocb->ki_flags indicates that no new I/O
2544 * requests shall be made for the read or for readahead. When no data
2545 * can be read, -EAGAIN shall be returned. When readahead would be
2546 * triggered, a partial, possibly empty read shall be returned.
2549 * * number of bytes copied, even for partial reads
2550 * * negative error code (or 0 if IOCB_NOIO) if nothing was read
2553 generic_file_read_iter(struct kiocb *iocb, struct iov_iter *iter)
2555 size_t count = iov_iter_count(iter);
2559 goto out; /* skip atime */
2561 if (iocb->ki_flags & IOCB_DIRECT) {
2562 struct file *file = iocb->ki_filp;
2563 struct address_space *mapping = file->f_mapping;
2564 struct inode *inode = mapping->host;
2567 size = i_size_read(inode);
2568 if (iocb->ki_flags & IOCB_NOWAIT) {
2569 if (filemap_range_has_page(mapping, iocb->ki_pos,
2570 iocb->ki_pos + count - 1))
2573 retval = filemap_write_and_wait_range(mapping,
2575 iocb->ki_pos + count - 1);
2580 file_accessed(file);
2582 retval = mapping->a_ops->direct_IO(iocb, iter);
2584 iocb->ki_pos += retval;
2587 if (retval != -EIOCBQUEUED)
2588 iov_iter_revert(iter, count - iov_iter_count(iter));
2591 * Btrfs can have a short DIO read if we encounter
2592 * compressed extents, so if there was an error, or if
2593 * we've already read everything we wanted to, or if
2594 * there was a short read because we hit EOF, go ahead
2595 * and return. Otherwise fallthrough to buffered io for
2596 * the rest of the read. Buffered reads will not work for
2597 * DAX files, so don't bother trying.
2599 if (retval < 0 || !count || iocb->ki_pos >= size ||
2604 retval = generic_file_buffered_read(iocb, iter, retval);
2608 EXPORT_SYMBOL(generic_file_read_iter);
2611 #define MMAP_LOTSAMISS (100)
2613 * lock_page_maybe_drop_mmap - lock the page, possibly dropping the mmap_lock
2614 * @vmf - the vm_fault for this fault.
2615 * @page - the page to lock.
2616 * @fpin - the pointer to the file we may pin (or is already pinned).
2618 * This works similar to lock_page_or_retry in that it can drop the mmap_lock.
2619 * It differs in that it actually returns the page locked if it returns 1 and 0
2620 * if it couldn't lock the page. If we did have to drop the mmap_lock then fpin
2621 * will point to the pinned file and needs to be fput()'ed at a later point.
2623 static int lock_page_maybe_drop_mmap(struct vm_fault *vmf, struct page *page,
2626 if (trylock_page(page))
2630 * NOTE! This will make us return with VM_FAULT_RETRY, but with
2631 * the mmap_lock still held. That's how FAULT_FLAG_RETRY_NOWAIT
2632 * is supposed to work. We have way too many special cases..
2634 if (vmf->flags & FAULT_FLAG_RETRY_NOWAIT)
2637 *fpin = maybe_unlock_mmap_for_io(vmf, *fpin);
2638 if (vmf->flags & FAULT_FLAG_KILLABLE) {
2639 if (__lock_page_killable(page)) {
2641 * We didn't have the right flags to drop the mmap_lock,
2642 * but all fault_handlers only check for fatal signals
2643 * if we return VM_FAULT_RETRY, so we need to drop the
2644 * mmap_lock here and return 0 if we don't have a fpin.
2647 mmap_read_unlock(vmf->vma->vm_mm);
2657 * Synchronous readahead happens when we don't even find a page in the page
2658 * cache at all. We don't want to perform IO under the mmap sem, so if we have
2659 * to drop the mmap sem we return the file that was pinned in order for us to do
2660 * that. If we didn't pin a file then we return NULL. The file that is
2661 * returned needs to be fput()'ed when we're done with it.
2663 static struct file *do_sync_mmap_readahead(struct vm_fault *vmf)
2665 struct file *file = vmf->vma->vm_file;
2666 struct file_ra_state *ra = &file->f_ra;
2667 struct address_space *mapping = file->f_mapping;
2668 DEFINE_READAHEAD(ractl, file, mapping, vmf->pgoff);
2669 struct file *fpin = NULL;
2670 unsigned int mmap_miss;
2672 /* If we don't want any read-ahead, don't bother */
2673 if (vmf->vma->vm_flags & VM_RAND_READ)
2678 if (vmf->vma->vm_flags & VM_SEQ_READ) {
2679 fpin = maybe_unlock_mmap_for_io(vmf, fpin);
2680 page_cache_sync_ra(&ractl, ra, ra->ra_pages);
2684 /* Avoid banging the cache line if not needed */
2685 mmap_miss = READ_ONCE(ra->mmap_miss);
2686 if (mmap_miss < MMAP_LOTSAMISS * 10)
2687 WRITE_ONCE(ra->mmap_miss, ++mmap_miss);
2690 * Do we miss much more than hit in this file? If so,
2691 * stop bothering with read-ahead. It will only hurt.
2693 if (mmap_miss > MMAP_LOTSAMISS)
2699 fpin = maybe_unlock_mmap_for_io(vmf, fpin);
2700 ra->start = max_t(long, 0, vmf->pgoff - ra->ra_pages / 2);
2701 ra->size = ra->ra_pages;
2702 ra->async_size = ra->ra_pages / 4;
2703 ractl._index = ra->start;
2704 do_page_cache_ra(&ractl, ra->size, ra->async_size);
2709 * Asynchronous readahead happens when we find the page and PG_readahead,
2710 * so we want to possibly extend the readahead further. We return the file that
2711 * was pinned if we have to drop the mmap_lock in order to do IO.
2713 static struct file *do_async_mmap_readahead(struct vm_fault *vmf,
2716 struct file *file = vmf->vma->vm_file;
2717 struct file_ra_state *ra = &file->f_ra;
2718 struct address_space *mapping = file->f_mapping;
2719 struct file *fpin = NULL;
2720 unsigned int mmap_miss;
2721 pgoff_t offset = vmf->pgoff;
2723 /* If we don't want any read-ahead, don't bother */
2724 if (vmf->vma->vm_flags & VM_RAND_READ || !ra->ra_pages)
2726 mmap_miss = READ_ONCE(ra->mmap_miss);
2728 WRITE_ONCE(ra->mmap_miss, --mmap_miss);
2729 if (PageReadahead(page)) {
2730 fpin = maybe_unlock_mmap_for_io(vmf, fpin);
2731 page_cache_async_readahead(mapping, ra, file,
2732 page, offset, ra->ra_pages);
2738 * filemap_fault - read in file data for page fault handling
2739 * @vmf: struct vm_fault containing details of the fault
2741 * filemap_fault() is invoked via the vma operations vector for a
2742 * mapped memory region to read in file data during a page fault.
2744 * The goto's are kind of ugly, but this streamlines the normal case of having
2745 * it in the page cache, and handles the special cases reasonably without
2746 * having a lot of duplicated code.
2748 * vma->vm_mm->mmap_lock must be held on entry.
2750 * If our return value has VM_FAULT_RETRY set, it's because the mmap_lock
2751 * may be dropped before doing I/O or by lock_page_maybe_drop_mmap().
2753 * If our return value does not have VM_FAULT_RETRY set, the mmap_lock
2754 * has not been released.
2756 * We never return with VM_FAULT_RETRY and a bit from VM_FAULT_ERROR set.
2758 * Return: bitwise-OR of %VM_FAULT_ codes.
2760 vm_fault_t filemap_fault(struct vm_fault *vmf)
2763 struct file *file = vmf->vma->vm_file;
2764 struct file *fpin = NULL;
2765 struct address_space *mapping = file->f_mapping;
2766 struct file_ra_state *ra = &file->f_ra;
2767 struct inode *inode = mapping->host;
2768 pgoff_t offset = vmf->pgoff;
2773 max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
2774 if (unlikely(offset >= max_off))
2775 return VM_FAULT_SIGBUS;
2778 * Do we have something in the page cache already?
2780 page = find_get_page(mapping, offset);
2781 if (likely(page) && !(vmf->flags & FAULT_FLAG_TRIED)) {
2783 * We found the page, so try async readahead before
2784 * waiting for the lock.
2786 fpin = do_async_mmap_readahead(vmf, page);
2788 /* No page in the page cache at all */
2789 count_vm_event(PGMAJFAULT);
2790 count_memcg_event_mm(vmf->vma->vm_mm, PGMAJFAULT);
2791 ret = VM_FAULT_MAJOR;
2792 fpin = do_sync_mmap_readahead(vmf);
2794 page = pagecache_get_page(mapping, offset,
2795 FGP_CREAT|FGP_FOR_MMAP,
2800 return VM_FAULT_OOM;
2804 if (!lock_page_maybe_drop_mmap(vmf, page, &fpin))
2807 /* Did it get truncated? */
2808 if (unlikely(compound_head(page)->mapping != mapping)) {
2813 VM_BUG_ON_PAGE(page_to_pgoff(page) != offset, page);
2816 * We have a locked page in the page cache, now we need to check
2817 * that it's up-to-date. If not, it is going to be due to an error.
2819 if (unlikely(!PageUptodate(page)))
2820 goto page_not_uptodate;
2823 * We've made it this far and we had to drop our mmap_lock, now is the
2824 * time to return to the upper layer and have it re-find the vma and
2833 * Found the page and have a reference on it.
2834 * We must recheck i_size under page lock.
2836 max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
2837 if (unlikely(offset >= max_off)) {
2840 return VM_FAULT_SIGBUS;
2844 return ret | VM_FAULT_LOCKED;
2848 * Umm, take care of errors if the page isn't up-to-date.
2849 * Try to re-read it _once_. We do this synchronously,
2850 * because there really aren't any performance issues here
2851 * and we need to check for errors.
2853 ClearPageError(page);
2854 fpin = maybe_unlock_mmap_for_io(vmf, fpin);
2855 error = mapping->a_ops->readpage(file, page);
2857 wait_on_page_locked(page);
2858 if (!PageUptodate(page))
2865 if (!error || error == AOP_TRUNCATED_PAGE)
2868 shrink_readahead_size_eio(ra);
2869 return VM_FAULT_SIGBUS;
2873 * We dropped the mmap_lock, we need to return to the fault handler to
2874 * re-find the vma and come back and find our hopefully still populated
2881 return ret | VM_FAULT_RETRY;
2883 EXPORT_SYMBOL(filemap_fault);
2885 static bool filemap_map_pmd(struct vm_fault *vmf, struct page *page)
2887 struct mm_struct *mm = vmf->vma->vm_mm;
2889 /* Huge page is mapped? No need to proceed. */
2890 if (pmd_trans_huge(*vmf->pmd)) {
2896 if (pmd_none(*vmf->pmd) && PageTransHuge(page)) {
2897 vm_fault_t ret = do_set_pmd(vmf, page);
2899 /* The page is mapped successfully, reference consumed. */
2905 if (pmd_none(*vmf->pmd)) {
2906 vmf->ptl = pmd_lock(mm, vmf->pmd);
2907 if (likely(pmd_none(*vmf->pmd))) {
2909 pmd_populate(mm, vmf->pmd, vmf->prealloc_pte);
2910 vmf->prealloc_pte = NULL;
2912 spin_unlock(vmf->ptl);
2915 /* See comment in handle_pte_fault() */
2916 if (pmd_devmap_trans_unstable(vmf->pmd)) {
2925 static struct page *next_uptodate_page(struct page *page,
2926 struct address_space *mapping,
2927 struct xa_state *xas, pgoff_t end_pgoff)
2929 unsigned long max_idx;
2934 if (xas_retry(xas, page))
2936 if (xa_is_value(page))
2938 if (PageLocked(page))
2940 if (!page_cache_get_speculative(page))
2942 /* Has the page moved or been split? */
2943 if (unlikely(page != xas_reload(xas)))
2945 if (!PageUptodate(page) || PageReadahead(page))
2947 if (PageHWPoison(page))
2949 if (!trylock_page(page))
2951 if (page->mapping != mapping)
2953 if (!PageUptodate(page))
2955 max_idx = DIV_ROUND_UP(i_size_read(mapping->host), PAGE_SIZE);
2956 if (xas->xa_index >= max_idx)
2963 } while ((page = xas_next_entry(xas, end_pgoff)) != NULL);
2968 static inline struct page *first_map_page(struct address_space *mapping,
2969 struct xa_state *xas,
2972 return next_uptodate_page(xas_find(xas, end_pgoff),
2973 mapping, xas, end_pgoff);
2976 static inline struct page *next_map_page(struct address_space *mapping,
2977 struct xa_state *xas,
2980 return next_uptodate_page(xas_next_entry(xas, end_pgoff),
2981 mapping, xas, end_pgoff);
2984 vm_fault_t filemap_map_pages(struct vm_fault *vmf,
2985 pgoff_t start_pgoff, pgoff_t end_pgoff)
2987 struct vm_area_struct *vma = vmf->vma;
2988 struct file *file = vma->vm_file;
2989 struct address_space *mapping = file->f_mapping;
2990 pgoff_t last_pgoff = start_pgoff;
2992 XA_STATE(xas, &mapping->i_pages, start_pgoff);
2993 struct page *head, *page;
2994 unsigned int mmap_miss = READ_ONCE(file->f_ra.mmap_miss);
2998 head = first_map_page(mapping, &xas, end_pgoff);
3002 if (filemap_map_pmd(vmf, head)) {
3003 ret = VM_FAULT_NOPAGE;
3007 addr = vma->vm_start + ((start_pgoff - vma->vm_pgoff) << PAGE_SHIFT);
3008 vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, addr, &vmf->ptl);
3010 page = find_subpage(head, xas.xa_index);
3011 if (PageHWPoison(page))
3017 addr += (xas.xa_index - last_pgoff) << PAGE_SHIFT;
3018 vmf->pte += xas.xa_index - last_pgoff;
3019 last_pgoff = xas.xa_index;
3021 if (!pte_none(*vmf->pte))
3024 /* We're about to handle the fault */
3025 if (vmf->address == addr)
3026 ret = VM_FAULT_NOPAGE;
3028 do_set_pte(vmf, page, addr);
3029 /* no need to invalidate: a not-present page won't be cached */
3030 update_mmu_cache(vma, addr, vmf->pte);
3036 } while ((head = next_map_page(mapping, &xas, end_pgoff)) != NULL);
3037 pte_unmap_unlock(vmf->pte, vmf->ptl);
3040 WRITE_ONCE(file->f_ra.mmap_miss, mmap_miss);
3043 EXPORT_SYMBOL(filemap_map_pages);
3045 vm_fault_t filemap_page_mkwrite(struct vm_fault *vmf)
3047 struct address_space *mapping = vmf->vma->vm_file->f_mapping;
3048 struct page *page = vmf->page;
3049 vm_fault_t ret = VM_FAULT_LOCKED;
3051 sb_start_pagefault(mapping->host->i_sb);
3052 file_update_time(vmf->vma->vm_file);
3054 if (page->mapping != mapping) {
3056 ret = VM_FAULT_NOPAGE;
3060 * We mark the page dirty already here so that when freeze is in
3061 * progress, we are guaranteed that writeback during freezing will
3062 * see the dirty page and writeprotect it again.
3064 set_page_dirty(page);
3065 wait_for_stable_page(page);
3067 sb_end_pagefault(mapping->host->i_sb);
3071 const struct vm_operations_struct generic_file_vm_ops = {
3072 .fault = filemap_fault,
3073 .map_pages = filemap_map_pages,
3074 .page_mkwrite = filemap_page_mkwrite,
3077 /* This is used for a general mmap of a disk file */
3079 int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
3081 struct address_space *mapping = file->f_mapping;
3083 if (!mapping->a_ops->readpage)
3085 file_accessed(file);
3086 vma->vm_ops = &generic_file_vm_ops;
3091 * This is for filesystems which do not implement ->writepage.
3093 int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
3095 if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE))
3097 return generic_file_mmap(file, vma);
3100 vm_fault_t filemap_page_mkwrite(struct vm_fault *vmf)
3102 return VM_FAULT_SIGBUS;
3104 int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
3108 int generic_file_readonly_mmap(struct file * file, struct vm_area_struct * vma)
3112 #endif /* CONFIG_MMU */
3114 EXPORT_SYMBOL(filemap_page_mkwrite);
3115 EXPORT_SYMBOL(generic_file_mmap);
3116 EXPORT_SYMBOL(generic_file_readonly_mmap);
3118 static struct page *wait_on_page_read(struct page *page)
3120 if (!IS_ERR(page)) {
3121 wait_on_page_locked(page);
3122 if (!PageUptodate(page)) {
3124 page = ERR_PTR(-EIO);
3130 static struct page *do_read_cache_page(struct address_space *mapping,
3132 int (*filler)(void *, struct page *),
3139 page = find_get_page(mapping, index);
3141 page = __page_cache_alloc(gfp);
3143 return ERR_PTR(-ENOMEM);
3144 err = add_to_page_cache_lru(page, mapping, index, gfp);
3145 if (unlikely(err)) {
3149 /* Presumably ENOMEM for xarray node */
3150 return ERR_PTR(err);
3155 err = filler(data, page);
3157 err = mapping->a_ops->readpage(data, page);
3161 return ERR_PTR(err);
3164 page = wait_on_page_read(page);
3169 if (PageUptodate(page))
3173 * Page is not up to date and may be locked due to one of the following
3174 * case a: Page is being filled and the page lock is held
3175 * case b: Read/write error clearing the page uptodate status
3176 * case c: Truncation in progress (page locked)
3177 * case d: Reclaim in progress
3179 * Case a, the page will be up to date when the page is unlocked.
3180 * There is no need to serialise on the page lock here as the page
3181 * is pinned so the lock gives no additional protection. Even if the
3182 * page is truncated, the data is still valid if PageUptodate as
3183 * it's a race vs truncate race.
3184 * Case b, the page will not be up to date
3185 * Case c, the page may be truncated but in itself, the data may still
3186 * be valid after IO completes as it's a read vs truncate race. The
3187 * operation must restart if the page is not uptodate on unlock but
3188 * otherwise serialising on page lock to stabilise the mapping gives
3189 * no additional guarantees to the caller as the page lock is
3190 * released before return.
3191 * Case d, similar to truncation. If reclaim holds the page lock, it
3192 * will be a race with remove_mapping that determines if the mapping
3193 * is valid on unlock but otherwise the data is valid and there is
3194 * no need to serialise with page lock.
3196 * As the page lock gives no additional guarantee, we optimistically
3197 * wait on the page to be unlocked and check if it's up to date and
3198 * use the page if it is. Otherwise, the page lock is required to
3199 * distinguish between the different cases. The motivation is that we
3200 * avoid spurious serialisations and wakeups when multiple processes
3201 * wait on the same page for IO to complete.
3203 wait_on_page_locked(page);
3204 if (PageUptodate(page))
3207 /* Distinguish between all the cases under the safety of the lock */
3210 /* Case c or d, restart the operation */
3211 if (!page->mapping) {
3217 /* Someone else locked and filled the page in a very small window */
3218 if (PageUptodate(page)) {
3224 * A previous I/O error may have been due to temporary
3226 * Clear page error before actual read, PG_error will be
3227 * set again if read page fails.
3229 ClearPageError(page);
3233 mark_page_accessed(page);
3238 * read_cache_page - read into page cache, fill it if needed
3239 * @mapping: the page's address_space
3240 * @index: the page index
3241 * @filler: function to perform the read
3242 * @data: first arg to filler(data, page) function, often left as NULL
3244 * Read into the page cache. If a page already exists, and PageUptodate() is
3245 * not set, try to fill the page and wait for it to become unlocked.
3247 * If the page does not get brought uptodate, return -EIO.
3249 * Return: up to date page on success, ERR_PTR() on failure.
3251 struct page *read_cache_page(struct address_space *mapping,
3253 int (*filler)(void *, struct page *),
3256 return do_read_cache_page(mapping, index, filler, data,
3257 mapping_gfp_mask(mapping));
3259 EXPORT_SYMBOL(read_cache_page);
3262 * read_cache_page_gfp - read into page cache, using specified page allocation flags.
3263 * @mapping: the page's address_space
3264 * @index: the page index
3265 * @gfp: the page allocator flags to use if allocating
3267 * This is the same as "read_mapping_page(mapping, index, NULL)", but with
3268 * any new page allocations done using the specified allocation flags.
3270 * If the page does not get brought uptodate, return -EIO.
3272 * Return: up to date page on success, ERR_PTR() on failure.
3274 struct page *read_cache_page_gfp(struct address_space *mapping,
3278 return do_read_cache_page(mapping, index, NULL, NULL, gfp);
3280 EXPORT_SYMBOL(read_cache_page_gfp);
3282 int pagecache_write_begin(struct file *file, struct address_space *mapping,
3283 loff_t pos, unsigned len, unsigned flags,
3284 struct page **pagep, void **fsdata)
3286 const struct address_space_operations *aops = mapping->a_ops;
3288 return aops->write_begin(file, mapping, pos, len, flags,
3291 EXPORT_SYMBOL(pagecache_write_begin);
3293 int pagecache_write_end(struct file *file, struct address_space *mapping,
3294 loff_t pos, unsigned len, unsigned copied,
3295 struct page *page, void *fsdata)
3297 const struct address_space_operations *aops = mapping->a_ops;
3299 return aops->write_end(file, mapping, pos, len, copied, page, fsdata);
3301 EXPORT_SYMBOL(pagecache_write_end);
3304 * Warn about a page cache invalidation failure during a direct I/O write.
3306 void dio_warn_stale_pagecache(struct file *filp)
3308 static DEFINE_RATELIMIT_STATE(_rs, 86400 * HZ, DEFAULT_RATELIMIT_BURST);
3312 errseq_set(&filp->f_mapping->wb_err, -EIO);
3313 if (__ratelimit(&_rs)) {
3314 path = file_path(filp, pathname, sizeof(pathname));
3317 pr_crit("Page cache invalidation failure on direct I/O. Possible data corruption due to collision with buffered I/O!\n");
3318 pr_crit("File: %s PID: %d Comm: %.20s\n", path, current->pid,
3324 generic_file_direct_write(struct kiocb *iocb, struct iov_iter *from)
3326 struct file *file = iocb->ki_filp;
3327 struct address_space *mapping = file->f_mapping;
3328 struct inode *inode = mapping->host;
3329 loff_t pos = iocb->ki_pos;
3334 write_len = iov_iter_count(from);
3335 end = (pos + write_len - 1) >> PAGE_SHIFT;
3337 if (iocb->ki_flags & IOCB_NOWAIT) {
3338 /* If there are pages to writeback, return */
3339 if (filemap_range_has_page(file->f_mapping, pos,
3340 pos + write_len - 1))
3343 written = filemap_write_and_wait_range(mapping, pos,
3344 pos + write_len - 1);
3350 * After a write we want buffered reads to be sure to go to disk to get
3351 * the new data. We invalidate clean cached page from the region we're
3352 * about to write. We do this *before* the write so that we can return
3353 * without clobbering -EIOCBQUEUED from ->direct_IO().
3355 written = invalidate_inode_pages2_range(mapping,
3356 pos >> PAGE_SHIFT, end);
3358 * If a page can not be invalidated, return 0 to fall back
3359 * to buffered write.
3362 if (written == -EBUSY)
3367 written = mapping->a_ops->direct_IO(iocb, from);
3370 * Finally, try again to invalidate clean pages which might have been
3371 * cached by non-direct readahead, or faulted in by get_user_pages()
3372 * if the source of the write was an mmap'ed region of the file
3373 * we're writing. Either one is a pretty crazy thing to do,
3374 * so we don't support it 100%. If this invalidation
3375 * fails, tough, the write still worked...
3377 * Most of the time we do not need this since dio_complete() will do
3378 * the invalidation for us. However there are some file systems that
3379 * do not end up with dio_complete() being called, so let's not break
3380 * them by removing it completely.
3382 * Noticeable example is a blkdev_direct_IO().
3384 * Skip invalidation for async writes or if mapping has no pages.
3386 if (written > 0 && mapping->nrpages &&
3387 invalidate_inode_pages2_range(mapping, pos >> PAGE_SHIFT, end))
3388 dio_warn_stale_pagecache(file);
3392 write_len -= written;
3393 if (pos > i_size_read(inode) && !S_ISBLK(inode->i_mode)) {
3394 i_size_write(inode, pos);
3395 mark_inode_dirty(inode);
3399 if (written != -EIOCBQUEUED)
3400 iov_iter_revert(from, write_len - iov_iter_count(from));
3404 EXPORT_SYMBOL(generic_file_direct_write);
3407 * Find or create a page at the given pagecache position. Return the locked
3408 * page. This function is specifically for buffered writes.
3410 struct page *grab_cache_page_write_begin(struct address_space *mapping,
3411 pgoff_t index, unsigned flags)
3414 int fgp_flags = FGP_LOCK|FGP_WRITE|FGP_CREAT;
3416 if (flags & AOP_FLAG_NOFS)
3417 fgp_flags |= FGP_NOFS;
3419 page = pagecache_get_page(mapping, index, fgp_flags,
3420 mapping_gfp_mask(mapping));
3422 wait_for_stable_page(page);
3426 EXPORT_SYMBOL(grab_cache_page_write_begin);
3428 ssize_t generic_perform_write(struct file *file,
3429 struct iov_iter *i, loff_t pos)
3431 struct address_space *mapping = file->f_mapping;
3432 const struct address_space_operations *a_ops = mapping->a_ops;
3434 ssize_t written = 0;
3435 unsigned int flags = 0;
3439 unsigned long offset; /* Offset into pagecache page */
3440 unsigned long bytes; /* Bytes to write to page */
3441 size_t copied; /* Bytes copied from user */
3444 offset = (pos & (PAGE_SIZE - 1));
3445 bytes = min_t(unsigned long, PAGE_SIZE - offset,
3450 * Bring in the user page that we will copy from _first_.
3451 * Otherwise there's a nasty deadlock on copying from the
3452 * same page as we're writing to, without it being marked
3455 * Not only is this an optimisation, but it is also required
3456 * to check that the address is actually valid, when atomic
3457 * usercopies are used, below.
3459 if (unlikely(iov_iter_fault_in_readable(i, bytes))) {
3464 if (fatal_signal_pending(current)) {
3469 status = a_ops->write_begin(file, mapping, pos, bytes, flags,
3471 if (unlikely(status < 0))
3474 if (mapping_writably_mapped(mapping))
3475 flush_dcache_page(page);
3477 copied = iov_iter_copy_from_user_atomic(page, i, offset, bytes);
3478 flush_dcache_page(page);
3480 status = a_ops->write_end(file, mapping, pos, bytes, copied,
3482 if (unlikely(status < 0))
3488 iov_iter_advance(i, copied);
3489 if (unlikely(copied == 0)) {
3491 * If we were unable to copy any data at all, we must
3492 * fall back to a single segment length write.
3494 * If we didn't fallback here, we could livelock
3495 * because not all segments in the iov can be copied at
3496 * once without a pagefault.
3498 bytes = min_t(unsigned long, PAGE_SIZE - offset,
3499 iov_iter_single_seg_count(i));
3505 balance_dirty_pages_ratelimited(mapping);
3506 } while (iov_iter_count(i));
3508 return written ? written : status;
3510 EXPORT_SYMBOL(generic_perform_write);
3513 * __generic_file_write_iter - write data to a file
3514 * @iocb: IO state structure (file, offset, etc.)
3515 * @from: iov_iter with data to write
3517 * This function does all the work needed for actually writing data to a
3518 * file. It does all basic checks, removes SUID from the file, updates
3519 * modification times and calls proper subroutines depending on whether we
3520 * do direct IO or a standard buffered write.
3522 * It expects i_mutex to be grabbed unless we work on a block device or similar
3523 * object which does not need locking at all.
3525 * This function does *not* take care of syncing data in case of O_SYNC write.
3526 * A caller has to handle it. This is mainly due to the fact that we want to
3527 * avoid syncing under i_mutex.
3530 * * number of bytes written, even for truncated writes
3531 * * negative error code if no data has been written at all
3533 ssize_t __generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
3535 struct file *file = iocb->ki_filp;
3536 struct address_space * mapping = file->f_mapping;
3537 struct inode *inode = mapping->host;
3538 ssize_t written = 0;
3542 /* We can write back this queue in page reclaim */
3543 current->backing_dev_info = inode_to_bdi(inode);
3544 err = file_remove_privs(file);
3548 err = file_update_time(file);
3552 if (iocb->ki_flags & IOCB_DIRECT) {
3553 loff_t pos, endbyte;
3555 written = generic_file_direct_write(iocb, from);
3557 * If the write stopped short of completing, fall back to
3558 * buffered writes. Some filesystems do this for writes to
3559 * holes, for example. For DAX files, a buffered write will
3560 * not succeed (even if it did, DAX does not handle dirty
3561 * page-cache pages correctly).
3563 if (written < 0 || !iov_iter_count(from) || IS_DAX(inode))
3566 status = generic_perform_write(file, from, pos = iocb->ki_pos);
3568 * If generic_perform_write() returned a synchronous error
3569 * then we want to return the number of bytes which were
3570 * direct-written, or the error code if that was zero. Note
3571 * that this differs from normal direct-io semantics, which
3572 * will return -EFOO even if some bytes were written.
3574 if (unlikely(status < 0)) {
3579 * We need to ensure that the page cache pages are written to
3580 * disk and invalidated to preserve the expected O_DIRECT
3583 endbyte = pos + status - 1;
3584 err = filemap_write_and_wait_range(mapping, pos, endbyte);
3586 iocb->ki_pos = endbyte + 1;
3588 invalidate_mapping_pages(mapping,
3590 endbyte >> PAGE_SHIFT);
3593 * We don't know how much we wrote, so just return
3594 * the number of bytes which were direct-written
3598 written = generic_perform_write(file, from, iocb->ki_pos);
3599 if (likely(written > 0))
3600 iocb->ki_pos += written;
3603 current->backing_dev_info = NULL;
3604 return written ? written : err;
3606 EXPORT_SYMBOL(__generic_file_write_iter);
3609 * generic_file_write_iter - write data to a file
3610 * @iocb: IO state structure
3611 * @from: iov_iter with data to write
3613 * This is a wrapper around __generic_file_write_iter() to be used by most
3614 * filesystems. It takes care of syncing the file in case of O_SYNC file
3615 * and acquires i_mutex as needed.
3617 * * negative error code if no data has been written at all of
3618 * vfs_fsync_range() failed for a synchronous write
3619 * * number of bytes written, even for truncated writes
3621 ssize_t generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
3623 struct file *file = iocb->ki_filp;
3624 struct inode *inode = file->f_mapping->host;
3628 ret = generic_write_checks(iocb, from);
3630 ret = __generic_file_write_iter(iocb, from);
3631 inode_unlock(inode);
3634 ret = generic_write_sync(iocb, ret);
3637 EXPORT_SYMBOL(generic_file_write_iter);
3640 * try_to_release_page() - release old fs-specific metadata on a page
3642 * @page: the page which the kernel is trying to free
3643 * @gfp_mask: memory allocation flags (and I/O mode)
3645 * The address_space is to try to release any data against the page
3646 * (presumably at page->private).
3648 * This may also be called if PG_fscache is set on a page, indicating that the
3649 * page is known to the local caching routines.
3651 * The @gfp_mask argument specifies whether I/O may be performed to release
3652 * this page (__GFP_IO), and whether the call may block (__GFP_RECLAIM & __GFP_FS).
3654 * Return: %1 if the release was successful, otherwise return zero.
3656 int try_to_release_page(struct page *page, gfp_t gfp_mask)
3658 struct address_space * const mapping = page->mapping;
3660 BUG_ON(!PageLocked(page));
3661 if (PageWriteback(page))
3664 if (mapping && mapping->a_ops->releasepage)
3665 return mapping->a_ops->releasepage(page, gfp_mask);
3666 return try_to_free_buffers(page);
3669 EXPORT_SYMBOL(try_to_release_page);