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>
45 #define CREATE_TRACE_POINTS
46 #include <trace/events/filemap.h>
49 * FIXME: remove all knowledge of the buffer layer from the core VM
51 #include <linux/buffer_head.h> /* for try_to_free_buffers */
56 * Shared mappings implemented 30.11.1994. It's not fully working yet,
59 * Shared mappings now work. 15.8.1995 Bruno.
61 * finished 'unifying' the page and buffer cache and SMP-threaded the
62 * page-cache, 21.05.1999, Ingo Molnar <mingo@redhat.com>
64 * SMP-threaded pagemap-LRU 1999, Andrea Arcangeli <andrea@suse.de>
70 * ->i_mmap_rwsem (truncate_pagecache)
71 * ->private_lock (__free_pte->__set_page_dirty_buffers)
72 * ->swap_lock (exclusive_swap_page, others)
76 * ->i_mmap_rwsem (truncate->unmap_mapping_range)
80 * ->page_table_lock or pte_lock (various, mainly in memory.c)
81 * ->i_pages lock (arch-dependent flush_dcache_mmap_lock)
84 * ->lock_page (access_process_vm)
86 * ->i_mutex (generic_perform_write)
87 * ->mmap_sem (fault_in_pages_readable->do_page_fault)
90 * sb_lock (fs/fs-writeback.c)
91 * ->i_pages lock (__sync_single_inode)
94 * ->anon_vma.lock (vma_adjust)
97 * ->page_table_lock or pte_lock (anon_vma_prepare and various)
99 * ->page_table_lock or pte_lock
100 * ->swap_lock (try_to_unmap_one)
101 * ->private_lock (try_to_unmap_one)
102 * ->i_pages lock (try_to_unmap_one)
103 * ->pgdat->lru_lock (follow_page->mark_page_accessed)
104 * ->pgdat->lru_lock (check_pte_range->isolate_lru_page)
105 * ->private_lock (page_remove_rmap->set_page_dirty)
106 * ->i_pages lock (page_remove_rmap->set_page_dirty)
107 * bdi.wb->list_lock (page_remove_rmap->set_page_dirty)
108 * ->inode->i_lock (page_remove_rmap->set_page_dirty)
109 * ->memcg->move_lock (page_remove_rmap->lock_page_memcg)
110 * bdi.wb->list_lock (zap_pte_range->set_page_dirty)
111 * ->inode->i_lock (zap_pte_range->set_page_dirty)
112 * ->private_lock (zap_pte_range->__set_page_dirty_buffers)
115 * ->tasklist_lock (memory_failure, collect_procs_ao)
118 static void page_cache_delete(struct address_space *mapping,
119 struct page *page, void *shadow)
121 XA_STATE(xas, &mapping->i_pages, page->index);
124 mapping_set_update(&xas, mapping);
126 /* hugetlb pages are represented by a single entry in the xarray */
127 if (!PageHuge(page)) {
128 xas_set_order(&xas, page->index, compound_order(page));
129 nr = compound_nr(page);
132 VM_BUG_ON_PAGE(!PageLocked(page), page);
133 VM_BUG_ON_PAGE(PageTail(page), page);
134 VM_BUG_ON_PAGE(nr != 1 && shadow, page);
136 xas_store(&xas, shadow);
137 xas_init_marks(&xas);
139 page->mapping = NULL;
140 /* Leave page->index set: truncation lookup relies upon it */
143 mapping->nrexceptional += nr;
145 * Make sure the nrexceptional update is committed before
146 * the nrpages update so that final truncate racing
147 * with reclaim does not see both counters 0 at the
148 * same time and miss a shadow entry.
152 mapping->nrpages -= nr;
155 static void unaccount_page_cache_page(struct address_space *mapping,
161 * if we're uptodate, flush out into the cleancache, otherwise
162 * invalidate any existing cleancache entries. We can't leave
163 * stale data around in the cleancache once our page is gone
165 if (PageUptodate(page) && PageMappedToDisk(page))
166 cleancache_put_page(page);
168 cleancache_invalidate_page(mapping, page);
170 VM_BUG_ON_PAGE(PageTail(page), page);
171 VM_BUG_ON_PAGE(page_mapped(page), page);
172 if (!IS_ENABLED(CONFIG_DEBUG_VM) && unlikely(page_mapped(page))) {
175 pr_alert("BUG: Bad page cache in process %s pfn:%05lx\n",
176 current->comm, page_to_pfn(page));
177 dump_page(page, "still mapped when deleted");
179 add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE);
181 mapcount = page_mapcount(page);
182 if (mapping_exiting(mapping) &&
183 page_count(page) >= mapcount + 2) {
185 * All vmas have already been torn down, so it's
186 * a good bet that actually the page is unmapped,
187 * and we'd prefer not to leak it: if we're wrong,
188 * some other bad page check should catch it later.
190 page_mapcount_reset(page);
191 page_ref_sub(page, mapcount);
195 /* hugetlb pages do not participate in page cache accounting. */
199 nr = hpage_nr_pages(page);
201 __mod_node_page_state(page_pgdat(page), NR_FILE_PAGES, -nr);
202 if (PageSwapBacked(page)) {
203 __mod_node_page_state(page_pgdat(page), NR_SHMEM, -nr);
204 if (PageTransHuge(page))
205 __dec_node_page_state(page, NR_SHMEM_THPS);
207 VM_BUG_ON_PAGE(PageTransHuge(page), page);
211 * At this point page must be either written or cleaned by
212 * truncate. Dirty page here signals a bug and loss of
215 * This fixes dirty accounting after removing the page entirely
216 * but leaves PageDirty set: it has no effect for truncated
217 * page and anyway will be cleared before returning page into
220 if (WARN_ON_ONCE(PageDirty(page)))
221 account_page_cleaned(page, mapping, inode_to_wb(mapping->host));
225 * Delete a page from the page cache and free it. Caller has to make
226 * sure the page is locked and that nobody else uses it - or that usage
227 * is safe. The caller must hold the i_pages lock.
229 void __delete_from_page_cache(struct page *page, void *shadow)
231 struct address_space *mapping = page->mapping;
233 trace_mm_filemap_delete_from_page_cache(page);
235 unaccount_page_cache_page(mapping, page);
236 page_cache_delete(mapping, page, shadow);
239 static void page_cache_free_page(struct address_space *mapping,
242 void (*freepage)(struct page *);
244 freepage = mapping->a_ops->freepage;
248 if (PageTransHuge(page) && !PageHuge(page)) {
249 page_ref_sub(page, HPAGE_PMD_NR);
250 VM_BUG_ON_PAGE(page_count(page) <= 0, page);
257 * delete_from_page_cache - delete page from page cache
258 * @page: the page which the kernel is trying to remove from page cache
260 * This must be called only on pages that have been verified to be in the page
261 * cache and locked. It will never put the page into the free list, the caller
262 * has a reference on the page.
264 void delete_from_page_cache(struct page *page)
266 struct address_space *mapping = page_mapping(page);
269 BUG_ON(!PageLocked(page));
270 xa_lock_irqsave(&mapping->i_pages, flags);
271 __delete_from_page_cache(page, NULL);
272 xa_unlock_irqrestore(&mapping->i_pages, flags);
274 page_cache_free_page(mapping, page);
276 EXPORT_SYMBOL(delete_from_page_cache);
279 * page_cache_delete_batch - delete several pages from page cache
280 * @mapping: the mapping to which pages belong
281 * @pvec: pagevec with pages to delete
283 * The function walks over mapping->i_pages and removes pages passed in @pvec
284 * from the mapping. The function expects @pvec to be sorted by page index.
285 * It tolerates holes in @pvec (mapping entries at those indices are not
286 * modified). The function expects only THP head pages to be present in the
287 * @pvec and takes care to delete all corresponding tail pages from the
290 * The function expects the i_pages lock to be held.
292 static void page_cache_delete_batch(struct address_space *mapping,
293 struct pagevec *pvec)
295 XA_STATE(xas, &mapping->i_pages, pvec->pages[0]->index);
297 int i = 0, tail_pages = 0;
300 mapping_set_update(&xas, mapping);
301 xas_for_each(&xas, page, ULONG_MAX) {
302 if (i >= pagevec_count(pvec) && !tail_pages)
304 if (xa_is_value(page))
308 * Some page got inserted in our range? Skip it. We
309 * have our pages locked so they are protected from
312 if (page != pvec->pages[i]) {
313 VM_BUG_ON_PAGE(page->index >
314 pvec->pages[i]->index, page);
317 WARN_ON_ONCE(!PageLocked(page));
318 if (PageTransHuge(page) && !PageHuge(page))
319 tail_pages = HPAGE_PMD_NR - 1;
320 page->mapping = NULL;
322 * Leave page->index set: truncation lookup relies
327 VM_BUG_ON_PAGE(page->index + HPAGE_PMD_NR - tail_pages
328 != pvec->pages[i]->index, page);
331 xas_store(&xas, NULL);
334 mapping->nrpages -= total_pages;
337 void delete_from_page_cache_batch(struct address_space *mapping,
338 struct pagevec *pvec)
343 if (!pagevec_count(pvec))
346 xa_lock_irqsave(&mapping->i_pages, flags);
347 for (i = 0; i < pagevec_count(pvec); i++) {
348 trace_mm_filemap_delete_from_page_cache(pvec->pages[i]);
350 unaccount_page_cache_page(mapping, pvec->pages[i]);
352 page_cache_delete_batch(mapping, pvec);
353 xa_unlock_irqrestore(&mapping->i_pages, flags);
355 for (i = 0; i < pagevec_count(pvec); i++)
356 page_cache_free_page(mapping, pvec->pages[i]);
359 int filemap_check_errors(struct address_space *mapping)
362 /* Check for outstanding write errors */
363 if (test_bit(AS_ENOSPC, &mapping->flags) &&
364 test_and_clear_bit(AS_ENOSPC, &mapping->flags))
366 if (test_bit(AS_EIO, &mapping->flags) &&
367 test_and_clear_bit(AS_EIO, &mapping->flags))
371 EXPORT_SYMBOL(filemap_check_errors);
373 static int filemap_check_and_keep_errors(struct address_space *mapping)
375 /* Check for outstanding write errors */
376 if (test_bit(AS_EIO, &mapping->flags))
378 if (test_bit(AS_ENOSPC, &mapping->flags))
384 * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range
385 * @mapping: address space structure to write
386 * @start: offset in bytes where the range starts
387 * @end: offset in bytes where the range ends (inclusive)
388 * @sync_mode: enable synchronous operation
390 * Start writeback against all of a mapping's dirty pages that lie
391 * within the byte offsets <start, end> inclusive.
393 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
394 * opposed to a regular memory cleansing writeback. The difference between
395 * these two operations is that if a dirty page/buffer is encountered, it must
396 * be waited upon, and not just skipped over.
398 * Return: %0 on success, negative error code otherwise.
400 int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
401 loff_t end, int sync_mode)
404 struct writeback_control wbc = {
405 .sync_mode = sync_mode,
406 .nr_to_write = LONG_MAX,
407 .range_start = start,
411 if (!mapping_cap_writeback_dirty(mapping) ||
412 !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
415 wbc_attach_fdatawrite_inode(&wbc, mapping->host);
416 ret = do_writepages(mapping, &wbc);
417 wbc_detach_inode(&wbc);
421 static inline int __filemap_fdatawrite(struct address_space *mapping,
424 return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode);
427 int filemap_fdatawrite(struct address_space *mapping)
429 return __filemap_fdatawrite(mapping, WB_SYNC_ALL);
431 EXPORT_SYMBOL(filemap_fdatawrite);
433 int filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
436 return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL);
438 EXPORT_SYMBOL(filemap_fdatawrite_range);
441 * filemap_flush - mostly a non-blocking flush
442 * @mapping: target address_space
444 * This is a mostly non-blocking flush. Not suitable for data-integrity
445 * purposes - I/O may not be started against all dirty pages.
447 * Return: %0 on success, negative error code otherwise.
449 int filemap_flush(struct address_space *mapping)
451 return __filemap_fdatawrite(mapping, WB_SYNC_NONE);
453 EXPORT_SYMBOL(filemap_flush);
456 * filemap_range_has_page - check if a page exists in range.
457 * @mapping: address space within which to check
458 * @start_byte: offset in bytes where the range starts
459 * @end_byte: offset in bytes where the range ends (inclusive)
461 * Find at least one page in the range supplied, usually used to check if
462 * direct writing in this range will trigger a writeback.
464 * Return: %true if at least one page exists in the specified range,
467 bool filemap_range_has_page(struct address_space *mapping,
468 loff_t start_byte, loff_t end_byte)
471 XA_STATE(xas, &mapping->i_pages, start_byte >> PAGE_SHIFT);
472 pgoff_t max = end_byte >> PAGE_SHIFT;
474 if (end_byte < start_byte)
479 page = xas_find(&xas, max);
480 if (xas_retry(&xas, page))
482 /* Shadow entries don't count */
483 if (xa_is_value(page))
486 * We don't need to try to pin this page; we're about to
487 * release the RCU lock anyway. It is enough to know that
488 * there was a page here recently.
496 EXPORT_SYMBOL(filemap_range_has_page);
498 static void __filemap_fdatawait_range(struct address_space *mapping,
499 loff_t start_byte, loff_t end_byte)
501 pgoff_t index = start_byte >> PAGE_SHIFT;
502 pgoff_t end = end_byte >> PAGE_SHIFT;
506 if (end_byte < start_byte)
510 while (index <= end) {
513 nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index,
514 end, PAGECACHE_TAG_WRITEBACK);
518 for (i = 0; i < nr_pages; i++) {
519 struct page *page = pvec.pages[i];
521 wait_on_page_writeback(page);
522 ClearPageError(page);
524 pagevec_release(&pvec);
530 * filemap_fdatawait_range - wait for writeback to complete
531 * @mapping: address space structure to wait for
532 * @start_byte: offset in bytes where the range starts
533 * @end_byte: offset in bytes where the range ends (inclusive)
535 * Walk the list of under-writeback pages of the given address space
536 * in the given range and wait for all of them. Check error status of
537 * the address space and return it.
539 * Since the error status of the address space is cleared by this function,
540 * callers are responsible for checking the return value and handling and/or
541 * reporting the error.
543 * Return: error status of the address space.
545 int filemap_fdatawait_range(struct address_space *mapping, loff_t start_byte,
548 __filemap_fdatawait_range(mapping, start_byte, end_byte);
549 return filemap_check_errors(mapping);
551 EXPORT_SYMBOL(filemap_fdatawait_range);
554 * filemap_fdatawait_range_keep_errors - wait for writeback to complete
555 * @mapping: address space structure to wait for
556 * @start_byte: offset in bytes where the range starts
557 * @end_byte: offset in bytes where the range ends (inclusive)
559 * Walk the list of under-writeback pages of the given address space in the
560 * given range and wait for all of them. Unlike filemap_fdatawait_range(),
561 * this function does not clear error status of the address space.
563 * Use this function if callers don't handle errors themselves. Expected
564 * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2),
567 int filemap_fdatawait_range_keep_errors(struct address_space *mapping,
568 loff_t start_byte, loff_t end_byte)
570 __filemap_fdatawait_range(mapping, start_byte, end_byte);
571 return filemap_check_and_keep_errors(mapping);
573 EXPORT_SYMBOL(filemap_fdatawait_range_keep_errors);
576 * file_fdatawait_range - wait for writeback to complete
577 * @file: file pointing to address space structure to wait for
578 * @start_byte: offset in bytes where the range starts
579 * @end_byte: offset in bytes where the range ends (inclusive)
581 * Walk the list of under-writeback pages of the address space that file
582 * refers to, in the given range and wait for all of them. Check error
583 * status of the address space vs. the file->f_wb_err cursor and return it.
585 * Since the error status of the file is advanced by this function,
586 * callers are responsible for checking the return value and handling and/or
587 * reporting the error.
589 * Return: error status of the address space vs. the file->f_wb_err cursor.
591 int file_fdatawait_range(struct file *file, loff_t start_byte, loff_t end_byte)
593 struct address_space *mapping = file->f_mapping;
595 __filemap_fdatawait_range(mapping, start_byte, end_byte);
596 return file_check_and_advance_wb_err(file);
598 EXPORT_SYMBOL(file_fdatawait_range);
601 * filemap_fdatawait_keep_errors - wait for writeback without clearing errors
602 * @mapping: address space structure to wait for
604 * Walk the list of under-writeback pages of the given address space
605 * and wait for all of them. Unlike filemap_fdatawait(), this function
606 * does not clear error status of the address space.
608 * Use this function if callers don't handle errors themselves. Expected
609 * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2),
612 * Return: error status of the address space.
614 int filemap_fdatawait_keep_errors(struct address_space *mapping)
616 __filemap_fdatawait_range(mapping, 0, LLONG_MAX);
617 return filemap_check_and_keep_errors(mapping);
619 EXPORT_SYMBOL(filemap_fdatawait_keep_errors);
621 /* Returns true if writeback might be needed or already in progress. */
622 static bool mapping_needs_writeback(struct address_space *mapping)
624 if (dax_mapping(mapping))
625 return mapping->nrexceptional;
627 return mapping->nrpages;
630 int filemap_write_and_wait(struct address_space *mapping)
634 if (mapping_needs_writeback(mapping)) {
635 err = filemap_fdatawrite(mapping);
637 * Even if the above returned error, the pages may be
638 * written partially (e.g. -ENOSPC), so we wait for it.
639 * But the -EIO is special case, it may indicate the worst
640 * thing (e.g. bug) happened, so we avoid waiting for it.
643 int err2 = filemap_fdatawait(mapping);
647 /* Clear any previously stored errors */
648 filemap_check_errors(mapping);
651 err = filemap_check_errors(mapping);
655 EXPORT_SYMBOL(filemap_write_and_wait);
658 * filemap_write_and_wait_range - write out & wait on a file range
659 * @mapping: the address_space for the pages
660 * @lstart: offset in bytes where the range starts
661 * @lend: offset in bytes where the range ends (inclusive)
663 * Write out and wait upon file offsets lstart->lend, inclusive.
665 * Note that @lend is inclusive (describes the last byte to be written) so
666 * that this function can be used to write to the very end-of-file (end = -1).
668 * Return: error status of the address space.
670 int filemap_write_and_wait_range(struct address_space *mapping,
671 loff_t lstart, loff_t lend)
675 if (mapping_needs_writeback(mapping)) {
676 err = __filemap_fdatawrite_range(mapping, lstart, lend,
678 /* See comment of filemap_write_and_wait() */
680 int err2 = filemap_fdatawait_range(mapping,
685 /* Clear any previously stored errors */
686 filemap_check_errors(mapping);
689 err = filemap_check_errors(mapping);
693 EXPORT_SYMBOL(filemap_write_and_wait_range);
695 void __filemap_set_wb_err(struct address_space *mapping, int err)
697 errseq_t eseq = errseq_set(&mapping->wb_err, err);
699 trace_filemap_set_wb_err(mapping, eseq);
701 EXPORT_SYMBOL(__filemap_set_wb_err);
704 * file_check_and_advance_wb_err - report wb error (if any) that was previously
705 * and advance wb_err to current one
706 * @file: struct file on which the error is being reported
708 * When userland calls fsync (or something like nfsd does the equivalent), we
709 * want to report any writeback errors that occurred since the last fsync (or
710 * since the file was opened if there haven't been any).
712 * Grab the wb_err from the mapping. If it matches what we have in the file,
713 * then just quickly return 0. The file is all caught up.
715 * If it doesn't match, then take the mapping value, set the "seen" flag in
716 * it and try to swap it into place. If it works, or another task beat us
717 * to it with the new value, then update the f_wb_err and return the error
718 * portion. The error at this point must be reported via proper channels
719 * (a'la fsync, or NFS COMMIT operation, etc.).
721 * While we handle mapping->wb_err with atomic operations, the f_wb_err
722 * value is protected by the f_lock since we must ensure that it reflects
723 * the latest value swapped in for this file descriptor.
725 * Return: %0 on success, negative error code otherwise.
727 int file_check_and_advance_wb_err(struct file *file)
730 errseq_t old = READ_ONCE(file->f_wb_err);
731 struct address_space *mapping = file->f_mapping;
733 /* Locklessly handle the common case where nothing has changed */
734 if (errseq_check(&mapping->wb_err, old)) {
735 /* Something changed, must use slow path */
736 spin_lock(&file->f_lock);
737 old = file->f_wb_err;
738 err = errseq_check_and_advance(&mapping->wb_err,
740 trace_file_check_and_advance_wb_err(file, old);
741 spin_unlock(&file->f_lock);
745 * We're mostly using this function as a drop in replacement for
746 * filemap_check_errors. Clear AS_EIO/AS_ENOSPC to emulate the effect
747 * that the legacy code would have had on these flags.
749 clear_bit(AS_EIO, &mapping->flags);
750 clear_bit(AS_ENOSPC, &mapping->flags);
753 EXPORT_SYMBOL(file_check_and_advance_wb_err);
756 * file_write_and_wait_range - write out & wait on a file range
757 * @file: file pointing to address_space with pages
758 * @lstart: offset in bytes where the range starts
759 * @lend: offset in bytes where the range ends (inclusive)
761 * Write out and wait upon file offsets lstart->lend, inclusive.
763 * Note that @lend is inclusive (describes the last byte to be written) so
764 * that this function can be used to write to the very end-of-file (end = -1).
766 * After writing out and waiting on the data, we check and advance the
767 * f_wb_err cursor to the latest value, and return any errors detected there.
769 * Return: %0 on success, negative error code otherwise.
771 int file_write_and_wait_range(struct file *file, loff_t lstart, loff_t lend)
774 struct address_space *mapping = file->f_mapping;
776 if (mapping_needs_writeback(mapping)) {
777 err = __filemap_fdatawrite_range(mapping, lstart, lend,
779 /* See comment of filemap_write_and_wait() */
781 __filemap_fdatawait_range(mapping, lstart, lend);
783 err2 = file_check_and_advance_wb_err(file);
788 EXPORT_SYMBOL(file_write_and_wait_range);
791 * replace_page_cache_page - replace a pagecache page with a new one
792 * @old: page to be replaced
793 * @new: page to replace with
794 * @gfp_mask: allocation mode
796 * This function replaces a page in the pagecache with a new one. On
797 * success it acquires the pagecache reference for the new page and
798 * drops it for the old page. Both the old and new pages must be
799 * locked. This function does not add the new page to the LRU, the
800 * caller must do that.
802 * The remove + add is atomic. This function cannot fail.
806 int replace_page_cache_page(struct page *old, struct page *new, gfp_t gfp_mask)
808 struct address_space *mapping = old->mapping;
809 void (*freepage)(struct page *) = mapping->a_ops->freepage;
810 pgoff_t offset = old->index;
811 XA_STATE(xas, &mapping->i_pages, offset);
814 VM_BUG_ON_PAGE(!PageLocked(old), old);
815 VM_BUG_ON_PAGE(!PageLocked(new), new);
816 VM_BUG_ON_PAGE(new->mapping, new);
819 new->mapping = mapping;
822 xas_lock_irqsave(&xas, flags);
823 xas_store(&xas, new);
826 /* hugetlb pages do not participate in page cache accounting. */
828 __dec_node_page_state(new, NR_FILE_PAGES);
830 __inc_node_page_state(new, NR_FILE_PAGES);
831 if (PageSwapBacked(old))
832 __dec_node_page_state(new, NR_SHMEM);
833 if (PageSwapBacked(new))
834 __inc_node_page_state(new, NR_SHMEM);
835 xas_unlock_irqrestore(&xas, flags);
836 mem_cgroup_migrate(old, new);
843 EXPORT_SYMBOL_GPL(replace_page_cache_page);
845 static int __add_to_page_cache_locked(struct page *page,
846 struct address_space *mapping,
847 pgoff_t offset, gfp_t gfp_mask,
850 XA_STATE(xas, &mapping->i_pages, offset);
851 int huge = PageHuge(page);
852 struct mem_cgroup *memcg;
856 VM_BUG_ON_PAGE(!PageLocked(page), page);
857 VM_BUG_ON_PAGE(PageSwapBacked(page), page);
858 mapping_set_update(&xas, mapping);
861 error = mem_cgroup_try_charge(page, current->mm,
862 gfp_mask, &memcg, false);
868 page->mapping = mapping;
869 page->index = offset;
873 old = xas_load(&xas);
874 if (old && !xa_is_value(old))
875 xas_set_err(&xas, -EEXIST);
876 xas_store(&xas, page);
880 if (xa_is_value(old)) {
881 mapping->nrexceptional--;
887 /* hugetlb pages do not participate in page cache accounting */
889 __inc_node_page_state(page, NR_FILE_PAGES);
891 xas_unlock_irq(&xas);
892 } while (xas_nomem(&xas, gfp_mask & GFP_RECLAIM_MASK));
898 mem_cgroup_commit_charge(page, memcg, false, false);
899 trace_mm_filemap_add_to_page_cache(page);
902 page->mapping = NULL;
903 /* Leave page->index set: truncation relies upon it */
905 mem_cgroup_cancel_charge(page, memcg, false);
907 return xas_error(&xas);
909 ALLOW_ERROR_INJECTION(__add_to_page_cache_locked, ERRNO);
912 * add_to_page_cache_locked - add a locked page to the pagecache
914 * @mapping: the page's address_space
915 * @offset: page index
916 * @gfp_mask: page allocation mode
918 * This function is used to add a page to the pagecache. It must be locked.
919 * This function does not add the page to the LRU. The caller must do that.
921 * Return: %0 on success, negative error code otherwise.
923 int add_to_page_cache_locked(struct page *page, struct address_space *mapping,
924 pgoff_t offset, gfp_t gfp_mask)
926 return __add_to_page_cache_locked(page, mapping, offset,
929 EXPORT_SYMBOL(add_to_page_cache_locked);
931 int add_to_page_cache_lru(struct page *page, struct address_space *mapping,
932 pgoff_t offset, gfp_t gfp_mask)
937 __SetPageLocked(page);
938 ret = __add_to_page_cache_locked(page, mapping, offset,
941 __ClearPageLocked(page);
944 * The page might have been evicted from cache only
945 * recently, in which case it should be activated like
946 * any other repeatedly accessed page.
947 * The exception is pages getting rewritten; evicting other
948 * data from the working set, only to cache data that will
949 * get overwritten with something else, is a waste of memory.
951 WARN_ON_ONCE(PageActive(page));
952 if (!(gfp_mask & __GFP_WRITE) && shadow)
953 workingset_refault(page, shadow);
958 EXPORT_SYMBOL_GPL(add_to_page_cache_lru);
961 struct page *__page_cache_alloc(gfp_t gfp)
966 if (cpuset_do_page_mem_spread()) {
967 unsigned int cpuset_mems_cookie;
969 cpuset_mems_cookie = read_mems_allowed_begin();
970 n = cpuset_mem_spread_node();
971 page = __alloc_pages_node(n, gfp, 0);
972 } while (!page && read_mems_allowed_retry(cpuset_mems_cookie));
976 return alloc_pages(gfp, 0);
978 EXPORT_SYMBOL(__page_cache_alloc);
982 * In order to wait for pages to become available there must be
983 * waitqueues associated with pages. By using a hash table of
984 * waitqueues where the bucket discipline is to maintain all
985 * waiters on the same queue and wake all when any of the pages
986 * become available, and for the woken contexts to check to be
987 * sure the appropriate page became available, this saves space
988 * at a cost of "thundering herd" phenomena during rare hash
991 #define PAGE_WAIT_TABLE_BITS 8
992 #define PAGE_WAIT_TABLE_SIZE (1 << PAGE_WAIT_TABLE_BITS)
993 static wait_queue_head_t page_wait_table[PAGE_WAIT_TABLE_SIZE] __cacheline_aligned;
995 static wait_queue_head_t *page_waitqueue(struct page *page)
997 return &page_wait_table[hash_ptr(page, PAGE_WAIT_TABLE_BITS)];
1000 void __init pagecache_init(void)
1004 for (i = 0; i < PAGE_WAIT_TABLE_SIZE; i++)
1005 init_waitqueue_head(&page_wait_table[i]);
1007 page_writeback_init();
1010 /* This has the same layout as wait_bit_key - see fs/cachefiles/rdwr.c */
1011 struct wait_page_key {
1017 struct wait_page_queue {
1020 wait_queue_entry_t wait;
1023 static int wake_page_function(wait_queue_entry_t *wait, unsigned mode, int sync, void *arg)
1025 struct wait_page_key *key = arg;
1026 struct wait_page_queue *wait_page
1027 = container_of(wait, struct wait_page_queue, wait);
1029 if (wait_page->page != key->page)
1031 key->page_match = 1;
1033 if (wait_page->bit_nr != key->bit_nr)
1037 * Stop walking if it's locked.
1038 * Is this safe if put_and_wait_on_page_locked() is in use?
1039 * Yes: the waker must hold a reference to this page, and if PG_locked
1040 * has now already been set by another task, that task must also hold
1041 * a reference to the *same usage* of this page; so there is no need
1042 * to walk on to wake even the put_and_wait_on_page_locked() callers.
1044 if (test_bit(key->bit_nr, &key->page->flags))
1047 return autoremove_wake_function(wait, mode, sync, key);
1050 static void wake_up_page_bit(struct page *page, int bit_nr)
1052 wait_queue_head_t *q = page_waitqueue(page);
1053 struct wait_page_key key;
1054 unsigned long flags;
1055 wait_queue_entry_t bookmark;
1058 key.bit_nr = bit_nr;
1062 bookmark.private = NULL;
1063 bookmark.func = NULL;
1064 INIT_LIST_HEAD(&bookmark.entry);
1066 spin_lock_irqsave(&q->lock, flags);
1067 __wake_up_locked_key_bookmark(q, TASK_NORMAL, &key, &bookmark);
1069 while (bookmark.flags & WQ_FLAG_BOOKMARK) {
1071 * Take a breather from holding the lock,
1072 * allow pages that finish wake up asynchronously
1073 * to acquire the lock and remove themselves
1076 spin_unlock_irqrestore(&q->lock, flags);
1078 spin_lock_irqsave(&q->lock, flags);
1079 __wake_up_locked_key_bookmark(q, TASK_NORMAL, &key, &bookmark);
1083 * It is possible for other pages to have collided on the waitqueue
1084 * hash, so in that case check for a page match. That prevents a long-
1087 * It is still possible to miss a case here, when we woke page waiters
1088 * and removed them from the waitqueue, but there are still other
1091 if (!waitqueue_active(q) || !key.page_match) {
1092 ClearPageWaiters(page);
1094 * It's possible to miss clearing Waiters here, when we woke
1095 * our page waiters, but the hashed waitqueue has waiters for
1096 * other pages on it.
1098 * That's okay, it's a rare case. The next waker will clear it.
1101 spin_unlock_irqrestore(&q->lock, flags);
1104 static void wake_up_page(struct page *page, int bit)
1106 if (!PageWaiters(page))
1108 wake_up_page_bit(page, bit);
1112 * A choice of three behaviors for wait_on_page_bit_common():
1115 EXCLUSIVE, /* Hold ref to page and take the bit when woken, like
1116 * __lock_page() waiting on then setting PG_locked.
1118 SHARED, /* Hold ref to page and check the bit when woken, like
1119 * wait_on_page_writeback() waiting on PG_writeback.
1121 DROP, /* Drop ref to page before wait, no check when woken,
1122 * like put_and_wait_on_page_locked() on PG_locked.
1126 static inline int wait_on_page_bit_common(wait_queue_head_t *q,
1127 struct page *page, int bit_nr, int state, enum behavior behavior)
1129 struct wait_page_queue wait_page;
1130 wait_queue_entry_t *wait = &wait_page.wait;
1132 bool thrashing = false;
1133 bool delayacct = false;
1134 unsigned long pflags;
1137 if (bit_nr == PG_locked &&
1138 !PageUptodate(page) && PageWorkingset(page)) {
1139 if (!PageSwapBacked(page)) {
1140 delayacct_thrashing_start();
1143 psi_memstall_enter(&pflags);
1148 wait->flags = behavior == EXCLUSIVE ? WQ_FLAG_EXCLUSIVE : 0;
1149 wait->func = wake_page_function;
1150 wait_page.page = page;
1151 wait_page.bit_nr = bit_nr;
1154 spin_lock_irq(&q->lock);
1156 if (likely(list_empty(&wait->entry))) {
1157 __add_wait_queue_entry_tail(q, wait);
1158 SetPageWaiters(page);
1161 set_current_state(state);
1163 spin_unlock_irq(&q->lock);
1165 bit_is_set = test_bit(bit_nr, &page->flags);
1166 if (behavior == DROP)
1169 if (likely(bit_is_set))
1172 if (behavior == EXCLUSIVE) {
1173 if (!test_and_set_bit_lock(bit_nr, &page->flags))
1175 } else if (behavior == SHARED) {
1176 if (!test_bit(bit_nr, &page->flags))
1180 if (signal_pending_state(state, current)) {
1185 if (behavior == DROP) {
1187 * We can no longer safely access page->flags:
1188 * even if CONFIG_MEMORY_HOTREMOVE is not enabled,
1189 * there is a risk of waiting forever on a page reused
1190 * for something that keeps it locked indefinitely.
1191 * But best check for -EINTR above before breaking.
1197 finish_wait(q, wait);
1201 delayacct_thrashing_end();
1202 psi_memstall_leave(&pflags);
1206 * A signal could leave PageWaiters set. Clearing it here if
1207 * !waitqueue_active would be possible (by open-coding finish_wait),
1208 * but still fail to catch it in the case of wait hash collision. We
1209 * already can fail to clear wait hash collision cases, so don't
1210 * bother with signals either.
1216 void wait_on_page_bit(struct page *page, int bit_nr)
1218 wait_queue_head_t *q = page_waitqueue(page);
1219 wait_on_page_bit_common(q, page, bit_nr, TASK_UNINTERRUPTIBLE, SHARED);
1221 EXPORT_SYMBOL(wait_on_page_bit);
1223 int wait_on_page_bit_killable(struct page *page, int bit_nr)
1225 wait_queue_head_t *q = page_waitqueue(page);
1226 return wait_on_page_bit_common(q, page, bit_nr, TASK_KILLABLE, SHARED);
1228 EXPORT_SYMBOL(wait_on_page_bit_killable);
1231 * put_and_wait_on_page_locked - Drop a reference and wait for it to be unlocked
1232 * @page: The page to wait for.
1234 * The caller should hold a reference on @page. They expect the page to
1235 * become unlocked relatively soon, but do not wish to hold up migration
1236 * (for example) by holding the reference while waiting for the page to
1237 * come unlocked. After this function returns, the caller should not
1238 * dereference @page.
1240 void put_and_wait_on_page_locked(struct page *page)
1242 wait_queue_head_t *q;
1244 page = compound_head(page);
1245 q = page_waitqueue(page);
1246 wait_on_page_bit_common(q, page, PG_locked, TASK_UNINTERRUPTIBLE, DROP);
1250 * add_page_wait_queue - Add an arbitrary waiter to a page's wait queue
1251 * @page: Page defining the wait queue of interest
1252 * @waiter: Waiter to add to the queue
1254 * Add an arbitrary @waiter to the wait queue for the nominated @page.
1256 void add_page_wait_queue(struct page *page, wait_queue_entry_t *waiter)
1258 wait_queue_head_t *q = page_waitqueue(page);
1259 unsigned long flags;
1261 spin_lock_irqsave(&q->lock, flags);
1262 __add_wait_queue_entry_tail(q, waiter);
1263 SetPageWaiters(page);
1264 spin_unlock_irqrestore(&q->lock, flags);
1266 EXPORT_SYMBOL_GPL(add_page_wait_queue);
1268 #ifndef clear_bit_unlock_is_negative_byte
1271 * PG_waiters is the high bit in the same byte as PG_lock.
1273 * On x86 (and on many other architectures), we can clear PG_lock and
1274 * test the sign bit at the same time. But if the architecture does
1275 * not support that special operation, we just do this all by hand
1278 * The read of PG_waiters has to be after (or concurrently with) PG_locked
1279 * being cleared, but a memory barrier should be unneccssary since it is
1280 * in the same byte as PG_locked.
1282 static inline bool clear_bit_unlock_is_negative_byte(long nr, volatile void *mem)
1284 clear_bit_unlock(nr, mem);
1285 /* smp_mb__after_atomic(); */
1286 return test_bit(PG_waiters, mem);
1292 * unlock_page - unlock a locked page
1295 * Unlocks the page and wakes up sleepers in ___wait_on_page_locked().
1296 * Also wakes sleepers in wait_on_page_writeback() because the wakeup
1297 * mechanism between PageLocked pages and PageWriteback pages is shared.
1298 * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep.
1300 * Note that this depends on PG_waiters being the sign bit in the byte
1301 * that contains PG_locked - thus the BUILD_BUG_ON(). That allows us to
1302 * clear the PG_locked bit and test PG_waiters at the same time fairly
1303 * portably (architectures that do LL/SC can test any bit, while x86 can
1304 * test the sign bit).
1306 void unlock_page(struct page *page)
1308 BUILD_BUG_ON(PG_waiters != 7);
1309 page = compound_head(page);
1310 VM_BUG_ON_PAGE(!PageLocked(page), page);
1311 if (clear_bit_unlock_is_negative_byte(PG_locked, &page->flags))
1312 wake_up_page_bit(page, PG_locked);
1314 EXPORT_SYMBOL(unlock_page);
1317 * end_page_writeback - end writeback against a page
1320 void end_page_writeback(struct page *page)
1323 * TestClearPageReclaim could be used here but it is an atomic
1324 * operation and overkill in this particular case. Failing to
1325 * shuffle a page marked for immediate reclaim is too mild to
1326 * justify taking an atomic operation penalty at the end of
1327 * ever page writeback.
1329 if (PageReclaim(page)) {
1330 ClearPageReclaim(page);
1331 rotate_reclaimable_page(page);
1334 if (!test_clear_page_writeback(page))
1337 smp_mb__after_atomic();
1338 wake_up_page(page, PG_writeback);
1340 EXPORT_SYMBOL(end_page_writeback);
1343 * After completing I/O on a page, call this routine to update the page
1344 * flags appropriately
1346 void page_endio(struct page *page, bool is_write, int err)
1350 SetPageUptodate(page);
1352 ClearPageUptodate(page);
1358 struct address_space *mapping;
1361 mapping = page_mapping(page);
1363 mapping_set_error(mapping, err);
1365 end_page_writeback(page);
1368 EXPORT_SYMBOL_GPL(page_endio);
1371 * __lock_page - get a lock on the page, assuming we need to sleep to get it
1372 * @__page: the page to lock
1374 void __lock_page(struct page *__page)
1376 struct page *page = compound_head(__page);
1377 wait_queue_head_t *q = page_waitqueue(page);
1378 wait_on_page_bit_common(q, page, PG_locked, TASK_UNINTERRUPTIBLE,
1381 EXPORT_SYMBOL(__lock_page);
1383 int __lock_page_killable(struct page *__page)
1385 struct page *page = compound_head(__page);
1386 wait_queue_head_t *q = page_waitqueue(page);
1387 return wait_on_page_bit_common(q, page, PG_locked, TASK_KILLABLE,
1390 EXPORT_SYMBOL_GPL(__lock_page_killable);
1394 * 1 - page is locked; mmap_sem is still held.
1395 * 0 - page is not locked.
1396 * mmap_sem has been released (up_read()), unless flags had both
1397 * FAULT_FLAG_ALLOW_RETRY and FAULT_FLAG_RETRY_NOWAIT set, in
1398 * which case mmap_sem is still held.
1400 * If neither ALLOW_RETRY nor KILLABLE are set, will always return 1
1401 * with the page locked and the mmap_sem unperturbed.
1403 int __lock_page_or_retry(struct page *page, struct mm_struct *mm,
1406 if (flags & FAULT_FLAG_ALLOW_RETRY) {
1408 * CAUTION! In this case, mmap_sem is not released
1409 * even though return 0.
1411 if (flags & FAULT_FLAG_RETRY_NOWAIT)
1414 up_read(&mm->mmap_sem);
1415 if (flags & FAULT_FLAG_KILLABLE)
1416 wait_on_page_locked_killable(page);
1418 wait_on_page_locked(page);
1421 if (flags & FAULT_FLAG_KILLABLE) {
1424 ret = __lock_page_killable(page);
1426 up_read(&mm->mmap_sem);
1436 * page_cache_next_miss() - Find the next gap in the page cache.
1437 * @mapping: Mapping.
1439 * @max_scan: Maximum range to search.
1441 * Search the range [index, min(index + max_scan - 1, ULONG_MAX)] for the
1442 * gap with the lowest index.
1444 * This function may be called under the rcu_read_lock. However, this will
1445 * not atomically search a snapshot of the cache at a single point in time.
1446 * For example, if a gap is created at index 5, then subsequently a gap is
1447 * created at index 10, page_cache_next_miss covering both indices may
1448 * return 10 if called under the rcu_read_lock.
1450 * Return: The index of the gap if found, otherwise an index outside the
1451 * range specified (in which case 'return - index >= max_scan' will be true).
1452 * In the rare case of index wrap-around, 0 will be returned.
1454 pgoff_t page_cache_next_miss(struct address_space *mapping,
1455 pgoff_t index, unsigned long max_scan)
1457 XA_STATE(xas, &mapping->i_pages, index);
1459 while (max_scan--) {
1460 void *entry = xas_next(&xas);
1461 if (!entry || xa_is_value(entry))
1463 if (xas.xa_index == 0)
1467 return xas.xa_index;
1469 EXPORT_SYMBOL(page_cache_next_miss);
1472 * page_cache_prev_miss() - Find the previous gap in the page cache.
1473 * @mapping: Mapping.
1475 * @max_scan: Maximum range to search.
1477 * Search the range [max(index - max_scan + 1, 0), index] for the
1478 * gap with the highest index.
1480 * This function may be called under the rcu_read_lock. However, this will
1481 * not atomically search a snapshot of the cache at a single point in time.
1482 * For example, if a gap is created at index 10, then subsequently a gap is
1483 * created at index 5, page_cache_prev_miss() covering both indices may
1484 * return 5 if called under the rcu_read_lock.
1486 * Return: The index of the gap if found, otherwise an index outside the
1487 * range specified (in which case 'index - return >= max_scan' will be true).
1488 * In the rare case of wrap-around, ULONG_MAX will be returned.
1490 pgoff_t page_cache_prev_miss(struct address_space *mapping,
1491 pgoff_t index, unsigned long max_scan)
1493 XA_STATE(xas, &mapping->i_pages, index);
1495 while (max_scan--) {
1496 void *entry = xas_prev(&xas);
1497 if (!entry || xa_is_value(entry))
1499 if (xas.xa_index == ULONG_MAX)
1503 return xas.xa_index;
1505 EXPORT_SYMBOL(page_cache_prev_miss);
1508 * find_get_entry - find and get a page cache entry
1509 * @mapping: the address_space to search
1510 * @offset: the page cache index
1512 * Looks up the page cache slot at @mapping & @offset. If there is a
1513 * page cache page, it is returned with an increased refcount.
1515 * If the slot holds a shadow entry of a previously evicted page, or a
1516 * swap entry from shmem/tmpfs, it is returned.
1518 * Return: the found page or shadow entry, %NULL if nothing is found.
1520 struct page *find_get_entry(struct address_space *mapping, pgoff_t offset)
1522 XA_STATE(xas, &mapping->i_pages, offset);
1523 struct page *head, *page;
1528 page = xas_load(&xas);
1529 if (xas_retry(&xas, page))
1532 * A shadow entry of a recently evicted page, or a swap entry from
1533 * shmem/tmpfs. Return it without attempting to raise page count.
1535 if (!page || xa_is_value(page))
1538 head = compound_head(page);
1539 if (!page_cache_get_speculative(head))
1542 /* The page was split under us? */
1543 if (compound_head(page) != head) {
1549 * Has the page moved?
1550 * This is part of the lockless pagecache protocol. See
1551 * include/linux/pagemap.h for details.
1553 if (unlikely(page != xas_reload(&xas))) {
1562 EXPORT_SYMBOL(find_get_entry);
1565 * find_lock_entry - locate, pin and lock a page cache entry
1566 * @mapping: the address_space to search
1567 * @offset: the page cache index
1569 * Looks up the page cache slot at @mapping & @offset. If there is a
1570 * page cache page, it is returned locked and with an increased
1573 * If the slot holds a shadow entry of a previously evicted page, or a
1574 * swap entry from shmem/tmpfs, it is returned.
1576 * find_lock_entry() may sleep.
1578 * Return: the found page or shadow entry, %NULL if nothing is found.
1580 struct page *find_lock_entry(struct address_space *mapping, pgoff_t offset)
1585 page = find_get_entry(mapping, offset);
1586 if (page && !xa_is_value(page)) {
1588 /* Has the page been truncated? */
1589 if (unlikely(page_mapping(page) != mapping)) {
1594 VM_BUG_ON_PAGE(page_to_pgoff(page) != offset, page);
1598 EXPORT_SYMBOL(find_lock_entry);
1601 * pagecache_get_page - find and get a page reference
1602 * @mapping: the address_space to search
1603 * @offset: the page index
1604 * @fgp_flags: PCG flags
1605 * @gfp_mask: gfp mask to use for the page cache data page allocation
1607 * Looks up the page cache slot at @mapping & @offset.
1609 * PCG flags modify how the page is returned.
1611 * @fgp_flags can be:
1613 * - FGP_ACCESSED: the page will be marked accessed
1614 * - FGP_LOCK: Page is return locked
1615 * - FGP_CREAT: If page is not present then a new page is allocated using
1616 * @gfp_mask and added to the page cache and the VM's LRU
1617 * list. The page is returned locked and with an increased
1619 * - FGP_FOR_MMAP: Similar to FGP_CREAT, only we want to allow the caller to do
1620 * its own locking dance if the page is already in cache, or unlock the page
1621 * before returning if we had to add the page to pagecache.
1623 * If FGP_LOCK or FGP_CREAT are specified then the function may sleep even
1624 * if the GFP flags specified for FGP_CREAT are atomic.
1626 * If there is a page cache page, it is returned with an increased refcount.
1628 * Return: the found page or %NULL otherwise.
1630 struct page *pagecache_get_page(struct address_space *mapping, pgoff_t offset,
1631 int fgp_flags, gfp_t gfp_mask)
1636 page = find_get_entry(mapping, offset);
1637 if (xa_is_value(page))
1642 if (fgp_flags & FGP_LOCK) {
1643 if (fgp_flags & FGP_NOWAIT) {
1644 if (!trylock_page(page)) {
1652 /* Has the page been truncated? */
1653 if (unlikely(page->mapping != mapping)) {
1658 VM_BUG_ON_PAGE(page->index != offset, page);
1661 if (fgp_flags & FGP_ACCESSED)
1662 mark_page_accessed(page);
1665 if (!page && (fgp_flags & FGP_CREAT)) {
1667 if ((fgp_flags & FGP_WRITE) && mapping_cap_account_dirty(mapping))
1668 gfp_mask |= __GFP_WRITE;
1669 if (fgp_flags & FGP_NOFS)
1670 gfp_mask &= ~__GFP_FS;
1672 page = __page_cache_alloc(gfp_mask);
1676 if (WARN_ON_ONCE(!(fgp_flags & (FGP_LOCK | FGP_FOR_MMAP))))
1677 fgp_flags |= FGP_LOCK;
1679 /* Init accessed so avoid atomic mark_page_accessed later */
1680 if (fgp_flags & FGP_ACCESSED)
1681 __SetPageReferenced(page);
1683 err = add_to_page_cache_lru(page, mapping, offset, gfp_mask);
1684 if (unlikely(err)) {
1692 * add_to_page_cache_lru locks the page, and for mmap we expect
1695 if (page && (fgp_flags & FGP_FOR_MMAP))
1701 EXPORT_SYMBOL(pagecache_get_page);
1704 * find_get_entries - gang pagecache lookup
1705 * @mapping: The address_space to search
1706 * @start: The starting page cache index
1707 * @nr_entries: The maximum number of entries
1708 * @entries: Where the resulting entries are placed
1709 * @indices: The cache indices corresponding to the entries in @entries
1711 * find_get_entries() will search for and return a group of up to
1712 * @nr_entries entries in the mapping. The entries are placed at
1713 * @entries. find_get_entries() takes a reference against any actual
1716 * The search returns a group of mapping-contiguous page cache entries
1717 * with ascending indexes. There may be holes in the indices due to
1718 * not-present pages.
1720 * Any shadow entries of evicted pages, or swap entries from
1721 * shmem/tmpfs, are included in the returned array.
1723 * Return: the number of pages and shadow entries which were found.
1725 unsigned find_get_entries(struct address_space *mapping,
1726 pgoff_t start, unsigned int nr_entries,
1727 struct page **entries, pgoff_t *indices)
1729 XA_STATE(xas, &mapping->i_pages, start);
1731 unsigned int ret = 0;
1737 xas_for_each(&xas, page, ULONG_MAX) {
1739 if (xas_retry(&xas, page))
1742 * A shadow entry of a recently evicted page, a swap
1743 * entry from shmem/tmpfs or a DAX entry. Return it
1744 * without attempting to raise page count.
1746 if (xa_is_value(page))
1749 head = compound_head(page);
1750 if (!page_cache_get_speculative(head))
1753 /* The page was split under us? */
1754 if (compound_head(page) != head)
1757 /* Has the page moved? */
1758 if (unlikely(page != xas_reload(&xas)))
1762 indices[ret] = xas.xa_index;
1763 entries[ret] = page;
1764 if (++ret == nr_entries)
1777 * find_get_pages_range - gang pagecache lookup
1778 * @mapping: The address_space to search
1779 * @start: The starting page index
1780 * @end: The final page index (inclusive)
1781 * @nr_pages: The maximum number of pages
1782 * @pages: Where the resulting pages are placed
1784 * find_get_pages_range() will search for and return a group of up to @nr_pages
1785 * pages in the mapping starting at index @start and up to index @end
1786 * (inclusive). The pages are placed at @pages. find_get_pages_range() takes
1787 * a reference against the returned pages.
1789 * The search returns a group of mapping-contiguous pages with ascending
1790 * indexes. There may be holes in the indices due to not-present pages.
1791 * We also update @start to index the next page for the traversal.
1793 * Return: the number of pages which were found. If this number is
1794 * smaller than @nr_pages, the end of specified range has been
1797 unsigned find_get_pages_range(struct address_space *mapping, pgoff_t *start,
1798 pgoff_t end, unsigned int nr_pages,
1799 struct page **pages)
1801 XA_STATE(xas, &mapping->i_pages, *start);
1805 if (unlikely(!nr_pages))
1809 xas_for_each(&xas, page, end) {
1811 if (xas_retry(&xas, page))
1813 /* Skip over shadow, swap and DAX entries */
1814 if (xa_is_value(page))
1817 head = compound_head(page);
1818 if (!page_cache_get_speculative(head))
1821 /* The page was split under us? */
1822 if (compound_head(page) != head)
1825 /* Has the page moved? */
1826 if (unlikely(page != xas_reload(&xas)))
1830 if (++ret == nr_pages) {
1831 *start = xas.xa_index + 1;
1842 * We come here when there is no page beyond @end. We take care to not
1843 * overflow the index @start as it confuses some of the callers. This
1844 * breaks the iteration when there is a page at index -1 but that is
1845 * already broken anyway.
1847 if (end == (pgoff_t)-1)
1848 *start = (pgoff_t)-1;
1858 * find_get_pages_contig - gang contiguous pagecache lookup
1859 * @mapping: The address_space to search
1860 * @index: The starting page index
1861 * @nr_pages: The maximum number of pages
1862 * @pages: Where the resulting pages are placed
1864 * find_get_pages_contig() works exactly like find_get_pages(), except
1865 * that the returned number of pages are guaranteed to be contiguous.
1867 * Return: the number of pages which were found.
1869 unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index,
1870 unsigned int nr_pages, struct page **pages)
1872 XA_STATE(xas, &mapping->i_pages, index);
1874 unsigned int ret = 0;
1876 if (unlikely(!nr_pages))
1880 for (page = xas_load(&xas); page; page = xas_next(&xas)) {
1882 if (xas_retry(&xas, page))
1885 * If the entry has been swapped out, we can stop looking.
1886 * No current caller is looking for DAX entries.
1888 if (xa_is_value(page))
1891 head = compound_head(page);
1892 if (!page_cache_get_speculative(head))
1895 /* The page was split under us? */
1896 if (compound_head(page) != head)
1899 /* Has the page moved? */
1900 if (unlikely(page != xas_reload(&xas)))
1904 if (++ret == nr_pages)
1915 EXPORT_SYMBOL(find_get_pages_contig);
1918 * find_get_pages_range_tag - find and return pages in given range matching @tag
1919 * @mapping: the address_space to search
1920 * @index: the starting page index
1921 * @end: The final page index (inclusive)
1922 * @tag: the tag index
1923 * @nr_pages: the maximum number of pages
1924 * @pages: where the resulting pages are placed
1926 * Like find_get_pages, except we only return pages which are tagged with
1927 * @tag. We update @index to index the next page for the traversal.
1929 * Return: the number of pages which were found.
1931 unsigned find_get_pages_range_tag(struct address_space *mapping, pgoff_t *index,
1932 pgoff_t end, xa_mark_t tag, unsigned int nr_pages,
1933 struct page **pages)
1935 XA_STATE(xas, &mapping->i_pages, *index);
1939 if (unlikely(!nr_pages))
1943 xas_for_each_marked(&xas, page, end, tag) {
1945 if (xas_retry(&xas, page))
1948 * Shadow entries should never be tagged, but this iteration
1949 * is lockless so there is a window for page reclaim to evict
1950 * a page we saw tagged. Skip over it.
1952 if (xa_is_value(page))
1955 head = compound_head(page);
1956 if (!page_cache_get_speculative(head))
1959 /* The page was split under us? */
1960 if (compound_head(page) != head)
1963 /* Has the page moved? */
1964 if (unlikely(page != xas_reload(&xas)))
1968 if (++ret == nr_pages) {
1969 *index = xas.xa_index + 1;
1980 * We come here when we got to @end. We take care to not overflow the
1981 * index @index as it confuses some of the callers. This breaks the
1982 * iteration when there is a page at index -1 but that is already
1985 if (end == (pgoff_t)-1)
1986 *index = (pgoff_t)-1;
1994 EXPORT_SYMBOL(find_get_pages_range_tag);
1997 * CD/DVDs are error prone. When a medium error occurs, the driver may fail
1998 * a _large_ part of the i/o request. Imagine the worst scenario:
2000 * ---R__________________________________________B__________
2001 * ^ reading here ^ bad block(assume 4k)
2003 * read(R) => miss => readahead(R...B) => media error => frustrating retries
2004 * => failing the whole request => read(R) => read(R+1) =>
2005 * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
2006 * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
2007 * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
2009 * It is going insane. Fix it by quickly scaling down the readahead size.
2011 static void shrink_readahead_size_eio(struct file *filp,
2012 struct file_ra_state *ra)
2018 * generic_file_buffered_read - generic file read routine
2019 * @iocb: the iocb to read
2020 * @iter: data destination
2021 * @written: already copied
2023 * This is a generic file read routine, and uses the
2024 * mapping->a_ops->readpage() function for the actual low-level stuff.
2026 * This is really ugly. But the goto's actually try to clarify some
2027 * of the logic when it comes to error handling etc.
2030 * * total number of bytes copied, including those the were already @written
2031 * * negative error code if nothing was copied
2033 static ssize_t generic_file_buffered_read(struct kiocb *iocb,
2034 struct iov_iter *iter, ssize_t written)
2036 struct file *filp = iocb->ki_filp;
2037 struct address_space *mapping = filp->f_mapping;
2038 struct inode *inode = mapping->host;
2039 struct file_ra_state *ra = &filp->f_ra;
2040 loff_t *ppos = &iocb->ki_pos;
2044 unsigned long offset; /* offset into pagecache page */
2045 unsigned int prev_offset;
2048 if (unlikely(*ppos >= inode->i_sb->s_maxbytes))
2050 iov_iter_truncate(iter, inode->i_sb->s_maxbytes);
2052 index = *ppos >> PAGE_SHIFT;
2053 prev_index = ra->prev_pos >> PAGE_SHIFT;
2054 prev_offset = ra->prev_pos & (PAGE_SIZE-1);
2055 last_index = (*ppos + iter->count + PAGE_SIZE-1) >> PAGE_SHIFT;
2056 offset = *ppos & ~PAGE_MASK;
2062 unsigned long nr, ret;
2066 if (fatal_signal_pending(current)) {
2071 page = find_get_page(mapping, index);
2073 if (iocb->ki_flags & IOCB_NOWAIT)
2075 page_cache_sync_readahead(mapping,
2077 index, last_index - index);
2078 page = find_get_page(mapping, index);
2079 if (unlikely(page == NULL))
2080 goto no_cached_page;
2082 if (PageReadahead(page)) {
2083 page_cache_async_readahead(mapping,
2085 index, last_index - index);
2087 if (!PageUptodate(page)) {
2088 if (iocb->ki_flags & IOCB_NOWAIT) {
2094 * See comment in do_read_cache_page on why
2095 * wait_on_page_locked is used to avoid unnecessarily
2096 * serialisations and why it's safe.
2098 error = wait_on_page_locked_killable(page);
2099 if (unlikely(error))
2100 goto readpage_error;
2101 if (PageUptodate(page))
2104 if (inode->i_blkbits == PAGE_SHIFT ||
2105 !mapping->a_ops->is_partially_uptodate)
2106 goto page_not_up_to_date;
2107 /* pipes can't handle partially uptodate pages */
2108 if (unlikely(iov_iter_is_pipe(iter)))
2109 goto page_not_up_to_date;
2110 if (!trylock_page(page))
2111 goto page_not_up_to_date;
2112 /* Did it get truncated before we got the lock? */
2114 goto page_not_up_to_date_locked;
2115 if (!mapping->a_ops->is_partially_uptodate(page,
2116 offset, iter->count))
2117 goto page_not_up_to_date_locked;
2122 * i_size must be checked after we know the page is Uptodate.
2124 * Checking i_size after the check allows us to calculate
2125 * the correct value for "nr", which means the zero-filled
2126 * part of the page is not copied back to userspace (unless
2127 * another truncate extends the file - this is desired though).
2130 isize = i_size_read(inode);
2131 end_index = (isize - 1) >> PAGE_SHIFT;
2132 if (unlikely(!isize || index > end_index)) {
2137 /* nr is the maximum number of bytes to copy from this page */
2139 if (index == end_index) {
2140 nr = ((isize - 1) & ~PAGE_MASK) + 1;
2148 /* If users can be writing to this page using arbitrary
2149 * virtual addresses, take care about potential aliasing
2150 * before reading the page on the kernel side.
2152 if (mapping_writably_mapped(mapping))
2153 flush_dcache_page(page);
2156 * When a sequential read accesses a page several times,
2157 * only mark it as accessed the first time.
2159 if (prev_index != index || offset != prev_offset)
2160 mark_page_accessed(page);
2164 * Ok, we have the page, and it's up-to-date, so
2165 * now we can copy it to user space...
2168 ret = copy_page_to_iter(page, offset, nr, iter);
2170 index += offset >> PAGE_SHIFT;
2171 offset &= ~PAGE_MASK;
2172 prev_offset = offset;
2176 if (!iov_iter_count(iter))
2184 page_not_up_to_date:
2185 /* Get exclusive access to the page ... */
2186 error = lock_page_killable(page);
2187 if (unlikely(error))
2188 goto readpage_error;
2190 page_not_up_to_date_locked:
2191 /* Did it get truncated before we got the lock? */
2192 if (!page->mapping) {
2198 /* Did somebody else fill it already? */
2199 if (PageUptodate(page)) {
2206 * A previous I/O error may have been due to temporary
2207 * failures, eg. multipath errors.
2208 * PG_error will be set again if readpage fails.
2210 ClearPageError(page);
2211 /* Start the actual read. The read will unlock the page. */
2212 error = mapping->a_ops->readpage(filp, page);
2214 if (unlikely(error)) {
2215 if (error == AOP_TRUNCATED_PAGE) {
2220 goto readpage_error;
2223 if (!PageUptodate(page)) {
2224 error = lock_page_killable(page);
2225 if (unlikely(error))
2226 goto readpage_error;
2227 if (!PageUptodate(page)) {
2228 if (page->mapping == NULL) {
2230 * invalidate_mapping_pages got it
2237 shrink_readahead_size_eio(filp, ra);
2239 goto readpage_error;
2247 /* UHHUH! A synchronous read error occurred. Report it */
2253 * Ok, it wasn't cached, so we need to create a new
2256 page = page_cache_alloc(mapping);
2261 error = add_to_page_cache_lru(page, mapping, index,
2262 mapping_gfp_constraint(mapping, GFP_KERNEL));
2265 if (error == -EEXIST) {
2277 ra->prev_pos = prev_index;
2278 ra->prev_pos <<= PAGE_SHIFT;
2279 ra->prev_pos |= prev_offset;
2281 *ppos = ((loff_t)index << PAGE_SHIFT) + offset;
2282 file_accessed(filp);
2283 return written ? written : error;
2287 * generic_file_read_iter - generic filesystem read routine
2288 * @iocb: kernel I/O control block
2289 * @iter: destination for the data read
2291 * This is the "read_iter()" routine for all filesystems
2292 * that can use the page cache directly.
2294 * * number of bytes copied, even for partial reads
2295 * * negative error code if nothing was read
2298 generic_file_read_iter(struct kiocb *iocb, struct iov_iter *iter)
2300 size_t count = iov_iter_count(iter);
2304 goto out; /* skip atime */
2306 if (iocb->ki_flags & IOCB_DIRECT) {
2307 struct file *file = iocb->ki_filp;
2308 struct address_space *mapping = file->f_mapping;
2309 struct inode *inode = mapping->host;
2312 size = i_size_read(inode);
2313 if (iocb->ki_flags & IOCB_NOWAIT) {
2314 if (filemap_range_has_page(mapping, iocb->ki_pos,
2315 iocb->ki_pos + count - 1))
2318 retval = filemap_write_and_wait_range(mapping,
2320 iocb->ki_pos + count - 1);
2325 file_accessed(file);
2327 retval = mapping->a_ops->direct_IO(iocb, iter);
2329 iocb->ki_pos += retval;
2332 iov_iter_revert(iter, count - iov_iter_count(iter));
2335 * Btrfs can have a short DIO read if we encounter
2336 * compressed extents, so if there was an error, or if
2337 * we've already read everything we wanted to, or if
2338 * there was a short read because we hit EOF, go ahead
2339 * and return. Otherwise fallthrough to buffered io for
2340 * the rest of the read. Buffered reads will not work for
2341 * DAX files, so don't bother trying.
2343 if (retval < 0 || !count || iocb->ki_pos >= size ||
2348 retval = generic_file_buffered_read(iocb, iter, retval);
2352 EXPORT_SYMBOL(generic_file_read_iter);
2355 #define MMAP_LOTSAMISS (100)
2356 static struct file *maybe_unlock_mmap_for_io(struct vm_fault *vmf,
2359 int flags = vmf->flags;
2365 * FAULT_FLAG_RETRY_NOWAIT means we don't want to wait on page locks or
2366 * anything, so we only pin the file and drop the mmap_sem if only
2367 * FAULT_FLAG_ALLOW_RETRY is set.
2369 if ((flags & (FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_RETRY_NOWAIT)) ==
2370 FAULT_FLAG_ALLOW_RETRY) {
2371 fpin = get_file(vmf->vma->vm_file);
2372 up_read(&vmf->vma->vm_mm->mmap_sem);
2378 * lock_page_maybe_drop_mmap - lock the page, possibly dropping the mmap_sem
2379 * @vmf - the vm_fault for this fault.
2380 * @page - the page to lock.
2381 * @fpin - the pointer to the file we may pin (or is already pinned).
2383 * This works similar to lock_page_or_retry in that it can drop the mmap_sem.
2384 * It differs in that it actually returns the page locked if it returns 1 and 0
2385 * if it couldn't lock the page. If we did have to drop the mmap_sem then fpin
2386 * will point to the pinned file and needs to be fput()'ed at a later point.
2388 static int lock_page_maybe_drop_mmap(struct vm_fault *vmf, struct page *page,
2391 if (trylock_page(page))
2395 * NOTE! This will make us return with VM_FAULT_RETRY, but with
2396 * the mmap_sem still held. That's how FAULT_FLAG_RETRY_NOWAIT
2397 * is supposed to work. We have way too many special cases..
2399 if (vmf->flags & FAULT_FLAG_RETRY_NOWAIT)
2402 *fpin = maybe_unlock_mmap_for_io(vmf, *fpin);
2403 if (vmf->flags & FAULT_FLAG_KILLABLE) {
2404 if (__lock_page_killable(page)) {
2406 * We didn't have the right flags to drop the mmap_sem,
2407 * but all fault_handlers only check for fatal signals
2408 * if we return VM_FAULT_RETRY, so we need to drop the
2409 * mmap_sem here and return 0 if we don't have a fpin.
2412 up_read(&vmf->vma->vm_mm->mmap_sem);
2422 * Synchronous readahead happens when we don't even find a page in the page
2423 * cache at all. We don't want to perform IO under the mmap sem, so if we have
2424 * to drop the mmap sem we return the file that was pinned in order for us to do
2425 * that. If we didn't pin a file then we return NULL. The file that is
2426 * returned needs to be fput()'ed when we're done with it.
2428 static struct file *do_sync_mmap_readahead(struct vm_fault *vmf)
2430 struct file *file = vmf->vma->vm_file;
2431 struct file_ra_state *ra = &file->f_ra;
2432 struct address_space *mapping = file->f_mapping;
2433 struct file *fpin = NULL;
2434 pgoff_t offset = vmf->pgoff;
2436 /* If we don't want any read-ahead, don't bother */
2437 if (vmf->vma->vm_flags & VM_RAND_READ)
2442 if (vmf->vma->vm_flags & VM_SEQ_READ) {
2443 fpin = maybe_unlock_mmap_for_io(vmf, fpin);
2444 page_cache_sync_readahead(mapping, ra, file, offset,
2449 /* Avoid banging the cache line if not needed */
2450 if (ra->mmap_miss < MMAP_LOTSAMISS * 10)
2454 * Do we miss much more than hit in this file? If so,
2455 * stop bothering with read-ahead. It will only hurt.
2457 if (ra->mmap_miss > MMAP_LOTSAMISS)
2463 fpin = maybe_unlock_mmap_for_io(vmf, fpin);
2464 ra->start = max_t(long, 0, offset - ra->ra_pages / 2);
2465 ra->size = ra->ra_pages;
2466 ra->async_size = ra->ra_pages / 4;
2467 ra_submit(ra, mapping, file);
2472 * Asynchronous readahead happens when we find the page and PG_readahead,
2473 * so we want to possibly extend the readahead further. We return the file that
2474 * was pinned if we have to drop the mmap_sem in order to do IO.
2476 static struct file *do_async_mmap_readahead(struct vm_fault *vmf,
2479 struct file *file = vmf->vma->vm_file;
2480 struct file_ra_state *ra = &file->f_ra;
2481 struct address_space *mapping = file->f_mapping;
2482 struct file *fpin = NULL;
2483 pgoff_t offset = vmf->pgoff;
2485 /* If we don't want any read-ahead, don't bother */
2486 if (vmf->vma->vm_flags & VM_RAND_READ)
2488 if (ra->mmap_miss > 0)
2490 if (PageReadahead(page)) {
2491 fpin = maybe_unlock_mmap_for_io(vmf, fpin);
2492 page_cache_async_readahead(mapping, ra, file,
2493 page, offset, ra->ra_pages);
2499 * filemap_fault - read in file data for page fault handling
2500 * @vmf: struct vm_fault containing details of the fault
2502 * filemap_fault() is invoked via the vma operations vector for a
2503 * mapped memory region to read in file data during a page fault.
2505 * The goto's are kind of ugly, but this streamlines the normal case of having
2506 * it in the page cache, and handles the special cases reasonably without
2507 * having a lot of duplicated code.
2509 * vma->vm_mm->mmap_sem must be held on entry.
2511 * If our return value has VM_FAULT_RETRY set, it's because the mmap_sem
2512 * may be dropped before doing I/O or by lock_page_maybe_drop_mmap().
2514 * If our return value does not have VM_FAULT_RETRY set, the mmap_sem
2515 * has not been released.
2517 * We never return with VM_FAULT_RETRY and a bit from VM_FAULT_ERROR set.
2519 * Return: bitwise-OR of %VM_FAULT_ codes.
2521 vm_fault_t filemap_fault(struct vm_fault *vmf)
2524 struct file *file = vmf->vma->vm_file;
2525 struct file *fpin = NULL;
2526 struct address_space *mapping = file->f_mapping;
2527 struct file_ra_state *ra = &file->f_ra;
2528 struct inode *inode = mapping->host;
2529 pgoff_t offset = vmf->pgoff;
2534 max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
2535 if (unlikely(offset >= max_off))
2536 return VM_FAULT_SIGBUS;
2539 * Do we have something in the page cache already?
2541 page = find_get_page(mapping, offset);
2542 if (likely(page) && !(vmf->flags & FAULT_FLAG_TRIED)) {
2544 * We found the page, so try async readahead before
2545 * waiting for the lock.
2547 fpin = do_async_mmap_readahead(vmf, page);
2549 /* No page in the page cache at all */
2550 count_vm_event(PGMAJFAULT);
2551 count_memcg_event_mm(vmf->vma->vm_mm, PGMAJFAULT);
2552 ret = VM_FAULT_MAJOR;
2553 fpin = do_sync_mmap_readahead(vmf);
2555 page = pagecache_get_page(mapping, offset,
2556 FGP_CREAT|FGP_FOR_MMAP,
2561 return vmf_error(-ENOMEM);
2565 if (!lock_page_maybe_drop_mmap(vmf, page, &fpin))
2568 /* Did it get truncated? */
2569 if (unlikely(page->mapping != mapping)) {
2574 VM_BUG_ON_PAGE(page->index != offset, page);
2577 * We have a locked page in the page cache, now we need to check
2578 * that it's up-to-date. If not, it is going to be due to an error.
2580 if (unlikely(!PageUptodate(page)))
2581 goto page_not_uptodate;
2584 * We've made it this far and we had to drop our mmap_sem, now is the
2585 * time to return to the upper layer and have it re-find the vma and
2594 * Found the page and have a reference on it.
2595 * We must recheck i_size under page lock.
2597 max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
2598 if (unlikely(offset >= max_off)) {
2601 return VM_FAULT_SIGBUS;
2605 return ret | VM_FAULT_LOCKED;
2609 * Umm, take care of errors if the page isn't up-to-date.
2610 * Try to re-read it _once_. We do this synchronously,
2611 * because there really aren't any performance issues here
2612 * and we need to check for errors.
2614 ClearPageError(page);
2615 fpin = maybe_unlock_mmap_for_io(vmf, fpin);
2616 error = mapping->a_ops->readpage(file, page);
2618 wait_on_page_locked(page);
2619 if (!PageUptodate(page))
2626 if (!error || error == AOP_TRUNCATED_PAGE)
2629 /* Things didn't work out. Return zero to tell the mm layer so. */
2630 shrink_readahead_size_eio(file, ra);
2631 return VM_FAULT_SIGBUS;
2635 * We dropped the mmap_sem, we need to return to the fault handler to
2636 * re-find the vma and come back and find our hopefully still populated
2643 return ret | VM_FAULT_RETRY;
2645 EXPORT_SYMBOL(filemap_fault);
2647 void filemap_map_pages(struct vm_fault *vmf,
2648 pgoff_t start_pgoff, pgoff_t end_pgoff)
2650 struct file *file = vmf->vma->vm_file;
2651 struct address_space *mapping = file->f_mapping;
2652 pgoff_t last_pgoff = start_pgoff;
2653 unsigned long max_idx;
2654 XA_STATE(xas, &mapping->i_pages, start_pgoff);
2655 struct page *head, *page;
2658 xas_for_each(&xas, page, end_pgoff) {
2659 if (xas_retry(&xas, page))
2661 if (xa_is_value(page))
2664 head = compound_head(page);
2667 * Check for a locked page first, as a speculative
2668 * reference may adversely influence page migration.
2670 if (PageLocked(head))
2672 if (!page_cache_get_speculative(head))
2675 /* The page was split under us? */
2676 if (compound_head(page) != head)
2679 /* Has the page moved? */
2680 if (unlikely(page != xas_reload(&xas)))
2683 if (!PageUptodate(page) ||
2684 PageReadahead(page) ||
2687 if (!trylock_page(page))
2690 if (page->mapping != mapping || !PageUptodate(page))
2693 max_idx = DIV_ROUND_UP(i_size_read(mapping->host), PAGE_SIZE);
2694 if (page->index >= max_idx)
2697 if (file->f_ra.mmap_miss > 0)
2698 file->f_ra.mmap_miss--;
2700 vmf->address += (xas.xa_index - last_pgoff) << PAGE_SHIFT;
2702 vmf->pte += xas.xa_index - last_pgoff;
2703 last_pgoff = xas.xa_index;
2704 if (alloc_set_pte(vmf, NULL, page))
2713 /* Huge page is mapped? No need to proceed. */
2714 if (pmd_trans_huge(*vmf->pmd))
2719 EXPORT_SYMBOL(filemap_map_pages);
2721 vm_fault_t filemap_page_mkwrite(struct vm_fault *vmf)
2723 struct page *page = vmf->page;
2724 struct inode *inode = file_inode(vmf->vma->vm_file);
2725 vm_fault_t ret = VM_FAULT_LOCKED;
2727 sb_start_pagefault(inode->i_sb);
2728 file_update_time(vmf->vma->vm_file);
2730 if (page->mapping != inode->i_mapping) {
2732 ret = VM_FAULT_NOPAGE;
2736 * We mark the page dirty already here so that when freeze is in
2737 * progress, we are guaranteed that writeback during freezing will
2738 * see the dirty page and writeprotect it again.
2740 set_page_dirty(page);
2741 wait_for_stable_page(page);
2743 sb_end_pagefault(inode->i_sb);
2747 const struct vm_operations_struct generic_file_vm_ops = {
2748 .fault = filemap_fault,
2749 .map_pages = filemap_map_pages,
2750 .page_mkwrite = filemap_page_mkwrite,
2753 /* This is used for a general mmap of a disk file */
2755 int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
2757 struct address_space *mapping = file->f_mapping;
2759 if (!mapping->a_ops->readpage)
2761 file_accessed(file);
2762 vma->vm_ops = &generic_file_vm_ops;
2767 * This is for filesystems which do not implement ->writepage.
2769 int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
2771 if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE))
2773 return generic_file_mmap(file, vma);
2776 vm_fault_t filemap_page_mkwrite(struct vm_fault *vmf)
2778 return VM_FAULT_SIGBUS;
2780 int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
2784 int generic_file_readonly_mmap(struct file * file, struct vm_area_struct * vma)
2788 #endif /* CONFIG_MMU */
2790 EXPORT_SYMBOL(filemap_page_mkwrite);
2791 EXPORT_SYMBOL(generic_file_mmap);
2792 EXPORT_SYMBOL(generic_file_readonly_mmap);
2794 static struct page *wait_on_page_read(struct page *page)
2796 if (!IS_ERR(page)) {
2797 wait_on_page_locked(page);
2798 if (!PageUptodate(page)) {
2800 page = ERR_PTR(-EIO);
2806 static struct page *do_read_cache_page(struct address_space *mapping,
2808 int (*filler)(void *, struct page *),
2815 page = find_get_page(mapping, index);
2817 page = __page_cache_alloc(gfp);
2819 return ERR_PTR(-ENOMEM);
2820 err = add_to_page_cache_lru(page, mapping, index, gfp);
2821 if (unlikely(err)) {
2825 /* Presumably ENOMEM for xarray node */
2826 return ERR_PTR(err);
2831 err = filler(data, page);
2833 err = mapping->a_ops->readpage(data, page);
2837 return ERR_PTR(err);
2840 page = wait_on_page_read(page);
2845 if (PageUptodate(page))
2849 * Page is not up to date and may be locked due one of the following
2850 * case a: Page is being filled and the page lock is held
2851 * case b: Read/write error clearing the page uptodate status
2852 * case c: Truncation in progress (page locked)
2853 * case d: Reclaim in progress
2855 * Case a, the page will be up to date when the page is unlocked.
2856 * There is no need to serialise on the page lock here as the page
2857 * is pinned so the lock gives no additional protection. Even if the
2858 * the page is truncated, the data is still valid if PageUptodate as
2859 * it's a race vs truncate race.
2860 * Case b, the page will not be up to date
2861 * Case c, the page may be truncated but in itself, the data may still
2862 * be valid after IO completes as it's a read vs truncate race. The
2863 * operation must restart if the page is not uptodate on unlock but
2864 * otherwise serialising on page lock to stabilise the mapping gives
2865 * no additional guarantees to the caller as the page lock is
2866 * released before return.
2867 * Case d, similar to truncation. If reclaim holds the page lock, it
2868 * will be a race with remove_mapping that determines if the mapping
2869 * is valid on unlock but otherwise the data is valid and there is
2870 * no need to serialise with page lock.
2872 * As the page lock gives no additional guarantee, we optimistically
2873 * wait on the page to be unlocked and check if it's up to date and
2874 * use the page if it is. Otherwise, the page lock is required to
2875 * distinguish between the different cases. The motivation is that we
2876 * avoid spurious serialisations and wakeups when multiple processes
2877 * wait on the same page for IO to complete.
2879 wait_on_page_locked(page);
2880 if (PageUptodate(page))
2883 /* Distinguish between all the cases under the safety of the lock */
2886 /* Case c or d, restart the operation */
2887 if (!page->mapping) {
2893 /* Someone else locked and filled the page in a very small window */
2894 if (PageUptodate(page)) {
2901 mark_page_accessed(page);
2906 * read_cache_page - read into page cache, fill it if needed
2907 * @mapping: the page's address_space
2908 * @index: the page index
2909 * @filler: function to perform the read
2910 * @data: first arg to filler(data, page) function, often left as NULL
2912 * Read into the page cache. If a page already exists, and PageUptodate() is
2913 * not set, try to fill the page and wait for it to become unlocked.
2915 * If the page does not get brought uptodate, return -EIO.
2917 * Return: up to date page on success, ERR_PTR() on failure.
2919 struct page *read_cache_page(struct address_space *mapping,
2921 int (*filler)(void *, struct page *),
2924 return do_read_cache_page(mapping, index, filler, data,
2925 mapping_gfp_mask(mapping));
2927 EXPORT_SYMBOL(read_cache_page);
2930 * read_cache_page_gfp - read into page cache, using specified page allocation flags.
2931 * @mapping: the page's address_space
2932 * @index: the page index
2933 * @gfp: the page allocator flags to use if allocating
2935 * This is the same as "read_mapping_page(mapping, index, NULL)", but with
2936 * any new page allocations done using the specified allocation flags.
2938 * If the page does not get brought uptodate, return -EIO.
2940 * Return: up to date page on success, ERR_PTR() on failure.
2942 struct page *read_cache_page_gfp(struct address_space *mapping,
2946 return do_read_cache_page(mapping, index, NULL, NULL, gfp);
2948 EXPORT_SYMBOL(read_cache_page_gfp);
2951 * Don't operate on ranges the page cache doesn't support, and don't exceed the
2952 * LFS limits. If pos is under the limit it becomes a short access. If it
2953 * exceeds the limit we return -EFBIG.
2955 static int generic_write_check_limits(struct file *file, loff_t pos,
2958 struct inode *inode = file->f_mapping->host;
2959 loff_t max_size = inode->i_sb->s_maxbytes;
2960 loff_t limit = rlimit(RLIMIT_FSIZE);
2962 if (limit != RLIM_INFINITY) {
2964 send_sig(SIGXFSZ, current, 0);
2967 *count = min(*count, limit - pos);
2970 if (!(file->f_flags & O_LARGEFILE))
2971 max_size = MAX_NON_LFS;
2973 if (unlikely(pos >= max_size))
2976 *count = min(*count, max_size - pos);
2982 * Performs necessary checks before doing a write
2984 * Can adjust writing position or amount of bytes to write.
2985 * Returns appropriate error code that caller should return or
2986 * zero in case that write should be allowed.
2988 inline ssize_t generic_write_checks(struct kiocb *iocb, struct iov_iter *from)
2990 struct file *file = iocb->ki_filp;
2991 struct inode *inode = file->f_mapping->host;
2995 if (IS_SWAPFILE(inode))
2998 if (!iov_iter_count(from))
3001 /* FIXME: this is for backwards compatibility with 2.4 */
3002 if (iocb->ki_flags & IOCB_APPEND)
3003 iocb->ki_pos = i_size_read(inode);
3005 if ((iocb->ki_flags & IOCB_NOWAIT) && !(iocb->ki_flags & IOCB_DIRECT))
3008 count = iov_iter_count(from);
3009 ret = generic_write_check_limits(file, iocb->ki_pos, &count);
3013 iov_iter_truncate(from, count);
3014 return iov_iter_count(from);
3016 EXPORT_SYMBOL(generic_write_checks);
3019 * Performs necessary checks before doing a clone.
3021 * Can adjust amount of bytes to clone via @req_count argument.
3022 * Returns appropriate error code that caller should return or
3023 * zero in case the clone should be allowed.
3025 int generic_remap_checks(struct file *file_in, loff_t pos_in,
3026 struct file *file_out, loff_t pos_out,
3027 loff_t *req_count, unsigned int remap_flags)
3029 struct inode *inode_in = file_in->f_mapping->host;
3030 struct inode *inode_out = file_out->f_mapping->host;
3031 uint64_t count = *req_count;
3033 loff_t size_in, size_out;
3034 loff_t bs = inode_out->i_sb->s_blocksize;
3037 /* The start of both ranges must be aligned to an fs block. */
3038 if (!IS_ALIGNED(pos_in, bs) || !IS_ALIGNED(pos_out, bs))
3041 /* Ensure offsets don't wrap. */
3042 if (pos_in + count < pos_in || pos_out + count < pos_out)
3045 size_in = i_size_read(inode_in);
3046 size_out = i_size_read(inode_out);
3048 /* Dedupe requires both ranges to be within EOF. */
3049 if ((remap_flags & REMAP_FILE_DEDUP) &&
3050 (pos_in >= size_in || pos_in + count > size_in ||
3051 pos_out >= size_out || pos_out + count > size_out))
3054 /* Ensure the infile range is within the infile. */
3055 if (pos_in >= size_in)
3057 count = min(count, size_in - (uint64_t)pos_in);
3059 ret = generic_write_check_limits(file_out, pos_out, &count);
3064 * If the user wanted us to link to the infile's EOF, round up to the
3065 * next block boundary for this check.
3067 * Otherwise, make sure the count is also block-aligned, having
3068 * already confirmed the starting offsets' block alignment.
3070 if (pos_in + count == size_in) {
3071 bcount = ALIGN(size_in, bs) - pos_in;
3073 if (!IS_ALIGNED(count, bs))
3074 count = ALIGN_DOWN(count, bs);
3078 /* Don't allow overlapped cloning within the same file. */
3079 if (inode_in == inode_out &&
3080 pos_out + bcount > pos_in &&
3081 pos_out < pos_in + bcount)
3085 * We shortened the request but the caller can't deal with that, so
3086 * bounce the request back to userspace.
3088 if (*req_count != count && !(remap_flags & REMAP_FILE_CAN_SHORTEN))
3097 * Performs common checks before doing a file copy/clone
3098 * from @file_in to @file_out.
3100 int generic_file_rw_checks(struct file *file_in, struct file *file_out)
3102 struct inode *inode_in = file_inode(file_in);
3103 struct inode *inode_out = file_inode(file_out);
3105 /* Don't copy dirs, pipes, sockets... */
3106 if (S_ISDIR(inode_in->i_mode) || S_ISDIR(inode_out->i_mode))
3108 if (!S_ISREG(inode_in->i_mode) || !S_ISREG(inode_out->i_mode))
3111 if (!(file_in->f_mode & FMODE_READ) ||
3112 !(file_out->f_mode & FMODE_WRITE) ||
3113 (file_out->f_flags & O_APPEND))
3120 * Performs necessary checks before doing a file copy
3122 * Can adjust amount of bytes to copy via @req_count argument.
3123 * Returns appropriate error code that caller should return or
3124 * zero in case the copy should be allowed.
3126 int generic_copy_file_checks(struct file *file_in, loff_t pos_in,
3127 struct file *file_out, loff_t pos_out,
3128 size_t *req_count, unsigned int flags)
3130 struct inode *inode_in = file_inode(file_in);
3131 struct inode *inode_out = file_inode(file_out);
3132 uint64_t count = *req_count;
3136 ret = generic_file_rw_checks(file_in, file_out);
3140 /* Don't touch certain kinds of inodes */
3141 if (IS_IMMUTABLE(inode_out))
3144 if (IS_SWAPFILE(inode_in) || IS_SWAPFILE(inode_out))
3147 /* Ensure offsets don't wrap. */
3148 if (pos_in + count < pos_in || pos_out + count < pos_out)
3151 /* Shorten the copy to EOF */
3152 size_in = i_size_read(inode_in);
3153 if (pos_in >= size_in)
3156 count = min(count, size_in - (uint64_t)pos_in);
3158 ret = generic_write_check_limits(file_out, pos_out, &count);
3162 /* Don't allow overlapped copying within the same file. */
3163 if (inode_in == inode_out &&
3164 pos_out + count > pos_in &&
3165 pos_out < pos_in + count)
3172 int pagecache_write_begin(struct file *file, struct address_space *mapping,
3173 loff_t pos, unsigned len, unsigned flags,
3174 struct page **pagep, void **fsdata)
3176 const struct address_space_operations *aops = mapping->a_ops;
3178 return aops->write_begin(file, mapping, pos, len, flags,
3181 EXPORT_SYMBOL(pagecache_write_begin);
3183 int pagecache_write_end(struct file *file, struct address_space *mapping,
3184 loff_t pos, unsigned len, unsigned copied,
3185 struct page *page, void *fsdata)
3187 const struct address_space_operations *aops = mapping->a_ops;
3189 return aops->write_end(file, mapping, pos, len, copied, page, fsdata);
3191 EXPORT_SYMBOL(pagecache_write_end);
3194 generic_file_direct_write(struct kiocb *iocb, struct iov_iter *from)
3196 struct file *file = iocb->ki_filp;
3197 struct address_space *mapping = file->f_mapping;
3198 struct inode *inode = mapping->host;
3199 loff_t pos = iocb->ki_pos;
3204 write_len = iov_iter_count(from);
3205 end = (pos + write_len - 1) >> PAGE_SHIFT;
3207 if (iocb->ki_flags & IOCB_NOWAIT) {
3208 /* If there are pages to writeback, return */
3209 if (filemap_range_has_page(inode->i_mapping, pos,
3210 pos + write_len - 1))
3213 written = filemap_write_and_wait_range(mapping, pos,
3214 pos + write_len - 1);
3220 * After a write we want buffered reads to be sure to go to disk to get
3221 * the new data. We invalidate clean cached page from the region we're
3222 * about to write. We do this *before* the write so that we can return
3223 * without clobbering -EIOCBQUEUED from ->direct_IO().
3225 written = invalidate_inode_pages2_range(mapping,
3226 pos >> PAGE_SHIFT, end);
3228 * If a page can not be invalidated, return 0 to fall back
3229 * to buffered write.
3232 if (written == -EBUSY)
3237 written = mapping->a_ops->direct_IO(iocb, from);
3240 * Finally, try again to invalidate clean pages which might have been
3241 * cached by non-direct readahead, or faulted in by get_user_pages()
3242 * if the source of the write was an mmap'ed region of the file
3243 * we're writing. Either one is a pretty crazy thing to do,
3244 * so we don't support it 100%. If this invalidation
3245 * fails, tough, the write still worked...
3247 * Most of the time we do not need this since dio_complete() will do
3248 * the invalidation for us. However there are some file systems that
3249 * do not end up with dio_complete() being called, so let's not break
3250 * them by removing it completely
3252 if (mapping->nrpages)
3253 invalidate_inode_pages2_range(mapping,
3254 pos >> PAGE_SHIFT, end);
3258 write_len -= written;
3259 if (pos > i_size_read(inode) && !S_ISBLK(inode->i_mode)) {
3260 i_size_write(inode, pos);
3261 mark_inode_dirty(inode);
3265 iov_iter_revert(from, write_len - iov_iter_count(from));
3269 EXPORT_SYMBOL(generic_file_direct_write);
3272 * Find or create a page at the given pagecache position. Return the locked
3273 * page. This function is specifically for buffered writes.
3275 struct page *grab_cache_page_write_begin(struct address_space *mapping,
3276 pgoff_t index, unsigned flags)
3279 int fgp_flags = FGP_LOCK|FGP_WRITE|FGP_CREAT;
3281 if (flags & AOP_FLAG_NOFS)
3282 fgp_flags |= FGP_NOFS;
3284 page = pagecache_get_page(mapping, index, fgp_flags,
3285 mapping_gfp_mask(mapping));
3287 wait_for_stable_page(page);
3291 EXPORT_SYMBOL(grab_cache_page_write_begin);
3293 ssize_t generic_perform_write(struct file *file,
3294 struct iov_iter *i, loff_t pos)
3296 struct address_space *mapping = file->f_mapping;
3297 const struct address_space_operations *a_ops = mapping->a_ops;
3299 ssize_t written = 0;
3300 unsigned int flags = 0;
3304 unsigned long offset; /* Offset into pagecache page */
3305 unsigned long bytes; /* Bytes to write to page */
3306 size_t copied; /* Bytes copied from user */
3309 offset = (pos & (PAGE_SIZE - 1));
3310 bytes = min_t(unsigned long, PAGE_SIZE - offset,
3315 * Bring in the user page that we will copy from _first_.
3316 * Otherwise there's a nasty deadlock on copying from the
3317 * same page as we're writing to, without it being marked
3320 * Not only is this an optimisation, but it is also required
3321 * to check that the address is actually valid, when atomic
3322 * usercopies are used, below.
3324 if (unlikely(iov_iter_fault_in_readable(i, bytes))) {
3329 if (fatal_signal_pending(current)) {
3334 status = a_ops->write_begin(file, mapping, pos, bytes, flags,
3336 if (unlikely(status < 0))
3339 if (mapping_writably_mapped(mapping))
3340 flush_dcache_page(page);
3342 copied = iov_iter_copy_from_user_atomic(page, i, offset, bytes);
3343 flush_dcache_page(page);
3345 status = a_ops->write_end(file, mapping, pos, bytes, copied,
3347 if (unlikely(status < 0))
3353 iov_iter_advance(i, copied);
3354 if (unlikely(copied == 0)) {
3356 * If we were unable to copy any data at all, we must
3357 * fall back to a single segment length write.
3359 * If we didn't fallback here, we could livelock
3360 * because not all segments in the iov can be copied at
3361 * once without a pagefault.
3363 bytes = min_t(unsigned long, PAGE_SIZE - offset,
3364 iov_iter_single_seg_count(i));
3370 balance_dirty_pages_ratelimited(mapping);
3371 } while (iov_iter_count(i));
3373 return written ? written : status;
3375 EXPORT_SYMBOL(generic_perform_write);
3378 * __generic_file_write_iter - write data to a file
3379 * @iocb: IO state structure (file, offset, etc.)
3380 * @from: iov_iter with data to write
3382 * This function does all the work needed for actually writing data to a
3383 * file. It does all basic checks, removes SUID from the file, updates
3384 * modification times and calls proper subroutines depending on whether we
3385 * do direct IO or a standard buffered write.
3387 * It expects i_mutex to be grabbed unless we work on a block device or similar
3388 * object which does not need locking at all.
3390 * This function does *not* take care of syncing data in case of O_SYNC write.
3391 * A caller has to handle it. This is mainly due to the fact that we want to
3392 * avoid syncing under i_mutex.
3395 * * number of bytes written, even for truncated writes
3396 * * negative error code if no data has been written at all
3398 ssize_t __generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
3400 struct file *file = iocb->ki_filp;
3401 struct address_space * mapping = file->f_mapping;
3402 struct inode *inode = mapping->host;
3403 ssize_t written = 0;
3407 /* We can write back this queue in page reclaim */
3408 current->backing_dev_info = inode_to_bdi(inode);
3409 err = file_remove_privs(file);
3413 err = file_update_time(file);
3417 if (iocb->ki_flags & IOCB_DIRECT) {
3418 loff_t pos, endbyte;
3420 written = generic_file_direct_write(iocb, from);
3422 * If the write stopped short of completing, fall back to
3423 * buffered writes. Some filesystems do this for writes to
3424 * holes, for example. For DAX files, a buffered write will
3425 * not succeed (even if it did, DAX does not handle dirty
3426 * page-cache pages correctly).
3428 if (written < 0 || !iov_iter_count(from) || IS_DAX(inode))
3431 status = generic_perform_write(file, from, pos = iocb->ki_pos);
3433 * If generic_perform_write() returned a synchronous error
3434 * then we want to return the number of bytes which were
3435 * direct-written, or the error code if that was zero. Note
3436 * that this differs from normal direct-io semantics, which
3437 * will return -EFOO even if some bytes were written.
3439 if (unlikely(status < 0)) {
3444 * We need to ensure that the page cache pages are written to
3445 * disk and invalidated to preserve the expected O_DIRECT
3448 endbyte = pos + status - 1;
3449 err = filemap_write_and_wait_range(mapping, pos, endbyte);
3451 iocb->ki_pos = endbyte + 1;
3453 invalidate_mapping_pages(mapping,
3455 endbyte >> PAGE_SHIFT);
3458 * We don't know how much we wrote, so just return
3459 * the number of bytes which were direct-written
3463 written = generic_perform_write(file, from, iocb->ki_pos);
3464 if (likely(written > 0))
3465 iocb->ki_pos += written;
3468 current->backing_dev_info = NULL;
3469 return written ? written : err;
3471 EXPORT_SYMBOL(__generic_file_write_iter);
3474 * generic_file_write_iter - write data to a file
3475 * @iocb: IO state structure
3476 * @from: iov_iter with data to write
3478 * This is a wrapper around __generic_file_write_iter() to be used by most
3479 * filesystems. It takes care of syncing the file in case of O_SYNC file
3480 * and acquires i_mutex as needed.
3482 * * negative error code if no data has been written at all of
3483 * vfs_fsync_range() failed for a synchronous write
3484 * * number of bytes written, even for truncated writes
3486 ssize_t generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
3488 struct file *file = iocb->ki_filp;
3489 struct inode *inode = file->f_mapping->host;
3493 ret = generic_write_checks(iocb, from);
3495 ret = __generic_file_write_iter(iocb, from);
3496 inode_unlock(inode);
3499 ret = generic_write_sync(iocb, ret);
3502 EXPORT_SYMBOL(generic_file_write_iter);
3505 * try_to_release_page() - release old fs-specific metadata on a page
3507 * @page: the page which the kernel is trying to free
3508 * @gfp_mask: memory allocation flags (and I/O mode)
3510 * The address_space is to try to release any data against the page
3511 * (presumably at page->private).
3513 * This may also be called if PG_fscache is set on a page, indicating that the
3514 * page is known to the local caching routines.
3516 * The @gfp_mask argument specifies whether I/O may be performed to release
3517 * this page (__GFP_IO), and whether the call may block (__GFP_RECLAIM & __GFP_FS).
3519 * Return: %1 if the release was successful, otherwise return zero.
3521 int try_to_release_page(struct page *page, gfp_t gfp_mask)
3523 struct address_space * const mapping = page->mapping;
3525 BUG_ON(!PageLocked(page));
3526 if (PageWriteback(page))
3529 if (mapping && mapping->a_ops->releasepage)
3530 return mapping->a_ops->releasepage(page, gfp_mask);
3531 return try_to_free_buffers(page);
3534 EXPORT_SYMBOL(try_to_release_page);