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/security.h>
34 #include <linux/cpuset.h>
35 #include <linux/hugetlb.h>
36 #include <linux/memcontrol.h>
37 #include <linux/cleancache.h>
38 #include <linux/shmem_fs.h>
39 #include <linux/rmap.h>
40 #include <linux/delayacct.h>
41 #include <linux/psi.h>
42 #include <linux/ramfs.h>
43 #include <linux/page_idle.h>
44 #include <asm/pgalloc.h>
45 #include <asm/tlbflush.h>
48 #define CREATE_TRACE_POINTS
49 #include <trace/events/filemap.h>
52 * FIXME: remove all knowledge of the buffer layer from the core VM
54 #include <linux/buffer_head.h> /* for try_to_free_buffers */
59 * Shared mappings implemented 30.11.1994. It's not fully working yet,
62 * Shared mappings now work. 15.8.1995 Bruno.
64 * finished 'unifying' the page and buffer cache and SMP-threaded the
65 * page-cache, 21.05.1999, Ingo Molnar <mingo@redhat.com>
67 * SMP-threaded pagemap-LRU 1999, Andrea Arcangeli <andrea@suse.de>
73 * ->i_mmap_rwsem (truncate_pagecache)
74 * ->private_lock (__free_pte->__set_page_dirty_buffers)
75 * ->swap_lock (exclusive_swap_page, others)
79 * ->invalidate_lock (acquired by fs in truncate path)
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 * ->invalidate_lock (filemap_fault)
89 * ->lock_page (filemap_fault, access_process_vm)
91 * ->i_rwsem (generic_perform_write)
92 * ->mmap_lock (fault_in_readable->do_page_fault)
95 * sb_lock (fs/fs-writeback.c)
96 * ->i_pages lock (__sync_single_inode)
99 * ->anon_vma.lock (vma_adjust)
102 * ->page_table_lock or pte_lock (anon_vma_prepare and various)
104 * ->page_table_lock or pte_lock
105 * ->swap_lock (try_to_unmap_one)
106 * ->private_lock (try_to_unmap_one)
107 * ->i_pages lock (try_to_unmap_one)
108 * ->lruvec->lru_lock (follow_page->mark_page_accessed)
109 * ->lruvec->lru_lock (check_pte_range->isolate_lru_page)
110 * ->private_lock (page_remove_rmap->set_page_dirty)
111 * ->i_pages lock (page_remove_rmap->set_page_dirty)
112 * bdi.wb->list_lock (page_remove_rmap->set_page_dirty)
113 * ->inode->i_lock (page_remove_rmap->set_page_dirty)
114 * ->memcg->move_lock (page_remove_rmap->lock_page_memcg)
115 * bdi.wb->list_lock (zap_pte_range->set_page_dirty)
116 * ->inode->i_lock (zap_pte_range->set_page_dirty)
117 * ->private_lock (zap_pte_range->__set_page_dirty_buffers)
120 * ->tasklist_lock (memory_failure, collect_procs_ao)
123 static void page_cache_delete(struct address_space *mapping,
124 struct page *page, void *shadow)
126 XA_STATE(xas, &mapping->i_pages, page->index);
129 mapping_set_update(&xas, mapping);
131 /* hugetlb pages are represented by a single entry in the xarray */
132 if (!PageHuge(page)) {
133 xas_set_order(&xas, page->index, compound_order(page));
134 nr = compound_nr(page);
137 VM_BUG_ON_PAGE(!PageLocked(page), page);
138 VM_BUG_ON_PAGE(PageTail(page), page);
139 VM_BUG_ON_PAGE(nr != 1 && shadow, page);
141 xas_store(&xas, shadow);
142 xas_init_marks(&xas);
144 page->mapping = NULL;
145 /* Leave page->index set: truncation lookup relies upon it */
146 mapping->nrpages -= nr;
149 static void unaccount_page_cache_page(struct address_space *mapping,
155 * if we're uptodate, flush out into the cleancache, otherwise
156 * invalidate any existing cleancache entries. We can't leave
157 * stale data around in the cleancache once our page is gone
159 if (PageUptodate(page) && PageMappedToDisk(page))
160 cleancache_put_page(page);
162 cleancache_invalidate_page(mapping, page);
164 VM_BUG_ON_PAGE(PageTail(page), page);
165 VM_BUG_ON_PAGE(page_mapped(page), page);
166 if (!IS_ENABLED(CONFIG_DEBUG_VM) && unlikely(page_mapped(page))) {
169 pr_alert("BUG: Bad page cache in process %s pfn:%05lx\n",
170 current->comm, page_to_pfn(page));
171 dump_page(page, "still mapped when deleted");
173 add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE);
175 mapcount = page_mapcount(page);
176 if (mapping_exiting(mapping) &&
177 page_count(page) >= mapcount + 2) {
179 * All vmas have already been torn down, so it's
180 * a good bet that actually the page is unmapped,
181 * and we'd prefer not to leak it: if we're wrong,
182 * some other bad page check should catch it later.
184 page_mapcount_reset(page);
185 page_ref_sub(page, mapcount);
189 /* hugetlb pages do not participate in page cache accounting. */
193 nr = thp_nr_pages(page);
195 __mod_lruvec_page_state(page, NR_FILE_PAGES, -nr);
196 if (PageSwapBacked(page)) {
197 __mod_lruvec_page_state(page, NR_SHMEM, -nr);
198 if (PageTransHuge(page))
199 __mod_lruvec_page_state(page, NR_SHMEM_THPS, -nr);
200 } else if (PageTransHuge(page)) {
201 __mod_lruvec_page_state(page, NR_FILE_THPS, -nr);
202 filemap_nr_thps_dec(mapping);
206 * At this point page must be either written or cleaned by
207 * truncate. Dirty page here signals a bug and loss of
210 * This fixes dirty accounting after removing the page entirely
211 * but leaves PageDirty set: it has no effect for truncated
212 * page and anyway will be cleared before returning page into
215 if (WARN_ON_ONCE(PageDirty(page)))
216 account_page_cleaned(page, mapping, inode_to_wb(mapping->host));
220 * Delete a page from the page cache and free it. Caller has to make
221 * sure the page is locked and that nobody else uses it - or that usage
222 * is safe. The caller must hold the i_pages lock.
224 void __delete_from_page_cache(struct page *page, void *shadow)
226 struct address_space *mapping = page->mapping;
228 trace_mm_filemap_delete_from_page_cache(page);
230 unaccount_page_cache_page(mapping, page);
231 page_cache_delete(mapping, page, shadow);
234 static void page_cache_free_page(struct address_space *mapping,
237 void (*freepage)(struct page *);
239 freepage = mapping->a_ops->freepage;
243 if (PageTransHuge(page) && !PageHuge(page)) {
244 page_ref_sub(page, thp_nr_pages(page));
245 VM_BUG_ON_PAGE(page_count(page) <= 0, page);
252 * delete_from_page_cache - delete page from page cache
253 * @page: the page which the kernel is trying to remove from page cache
255 * This must be called only on pages that have been verified to be in the page
256 * cache and locked. It will never put the page into the free list, the caller
257 * has a reference on the page.
259 void delete_from_page_cache(struct page *page)
261 struct address_space *mapping = page_mapping(page);
263 BUG_ON(!PageLocked(page));
264 xa_lock_irq(&mapping->i_pages);
265 __delete_from_page_cache(page, NULL);
266 xa_unlock_irq(&mapping->i_pages);
268 page_cache_free_page(mapping, page);
270 EXPORT_SYMBOL(delete_from_page_cache);
273 * page_cache_delete_batch - delete several pages from page cache
274 * @mapping: the mapping to which pages belong
275 * @pvec: pagevec with pages to delete
277 * The function walks over mapping->i_pages and removes pages passed in @pvec
278 * from the mapping. The function expects @pvec to be sorted by page index
279 * and is optimised for it to be dense.
280 * It tolerates holes in @pvec (mapping entries at those indices are not
281 * modified). The function expects only THP head pages to be present in the
284 * The function expects the i_pages lock to be held.
286 static void page_cache_delete_batch(struct address_space *mapping,
287 struct pagevec *pvec)
289 XA_STATE(xas, &mapping->i_pages, pvec->pages[0]->index);
294 mapping_set_update(&xas, mapping);
295 xas_for_each(&xas, page, ULONG_MAX) {
296 if (i >= pagevec_count(pvec))
299 /* A swap/dax/shadow entry got inserted? Skip it. */
300 if (xa_is_value(page))
303 * A page got inserted in our range? Skip it. We have our
304 * pages locked so they are protected from being removed.
305 * If we see a page whose index is higher than ours, it
306 * means our page has been removed, which shouldn't be
307 * possible because we're holding the PageLock.
309 if (page != pvec->pages[i]) {
310 VM_BUG_ON_PAGE(page->index > pvec->pages[i]->index,
315 WARN_ON_ONCE(!PageLocked(page));
317 if (page->index == xas.xa_index)
318 page->mapping = NULL;
319 /* Leave page->index set: truncation lookup relies on it */
322 * Move to the next page in the vector if this is a regular
323 * page or the index is of the last sub-page of this compound
326 if (page->index + compound_nr(page) - 1 == xas.xa_index)
328 xas_store(&xas, NULL);
331 mapping->nrpages -= total_pages;
334 void delete_from_page_cache_batch(struct address_space *mapping,
335 struct pagevec *pvec)
339 if (!pagevec_count(pvec))
342 xa_lock_irq(&mapping->i_pages);
343 for (i = 0; i < pagevec_count(pvec); i++) {
344 trace_mm_filemap_delete_from_page_cache(pvec->pages[i]);
346 unaccount_page_cache_page(mapping, pvec->pages[i]);
348 page_cache_delete_batch(mapping, pvec);
349 xa_unlock_irq(&mapping->i_pages);
351 for (i = 0; i < pagevec_count(pvec); i++)
352 page_cache_free_page(mapping, pvec->pages[i]);
355 int filemap_check_errors(struct address_space *mapping)
358 /* Check for outstanding write errors */
359 if (test_bit(AS_ENOSPC, &mapping->flags) &&
360 test_and_clear_bit(AS_ENOSPC, &mapping->flags))
362 if (test_bit(AS_EIO, &mapping->flags) &&
363 test_and_clear_bit(AS_EIO, &mapping->flags))
367 EXPORT_SYMBOL(filemap_check_errors);
369 static int filemap_check_and_keep_errors(struct address_space *mapping)
371 /* Check for outstanding write errors */
372 if (test_bit(AS_EIO, &mapping->flags))
374 if (test_bit(AS_ENOSPC, &mapping->flags))
380 * filemap_fdatawrite_wbc - start writeback on mapping dirty pages in range
381 * @mapping: address space structure to write
382 * @wbc: the writeback_control controlling the writeout
384 * Call writepages on the mapping using the provided wbc to control the
387 * Return: %0 on success, negative error code otherwise.
389 int filemap_fdatawrite_wbc(struct address_space *mapping,
390 struct writeback_control *wbc)
394 if (!mapping_can_writeback(mapping) ||
395 !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
398 wbc_attach_fdatawrite_inode(wbc, mapping->host);
399 ret = do_writepages(mapping, wbc);
400 wbc_detach_inode(wbc);
403 EXPORT_SYMBOL(filemap_fdatawrite_wbc);
406 * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range
407 * @mapping: address space structure to write
408 * @start: offset in bytes where the range starts
409 * @end: offset in bytes where the range ends (inclusive)
410 * @sync_mode: enable synchronous operation
412 * Start writeback against all of a mapping's dirty pages that lie
413 * within the byte offsets <start, end> inclusive.
415 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
416 * opposed to a regular memory cleansing writeback. The difference between
417 * these two operations is that if a dirty page/buffer is encountered, it must
418 * be waited upon, and not just skipped over.
420 * Return: %0 on success, negative error code otherwise.
422 int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
423 loff_t end, int sync_mode)
425 struct writeback_control wbc = {
426 .sync_mode = sync_mode,
427 .nr_to_write = LONG_MAX,
428 .range_start = start,
432 return filemap_fdatawrite_wbc(mapping, &wbc);
435 static inline int __filemap_fdatawrite(struct address_space *mapping,
438 return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode);
441 int filemap_fdatawrite(struct address_space *mapping)
443 return __filemap_fdatawrite(mapping, WB_SYNC_ALL);
445 EXPORT_SYMBOL(filemap_fdatawrite);
447 int filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
450 return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL);
452 EXPORT_SYMBOL(filemap_fdatawrite_range);
455 * filemap_flush - mostly a non-blocking flush
456 * @mapping: target address_space
458 * This is a mostly non-blocking flush. Not suitable for data-integrity
459 * purposes - I/O may not be started against all dirty pages.
461 * Return: %0 on success, negative error code otherwise.
463 int filemap_flush(struct address_space *mapping)
465 return __filemap_fdatawrite(mapping, WB_SYNC_NONE);
467 EXPORT_SYMBOL(filemap_flush);
470 * filemap_range_has_page - check if a page exists in range.
471 * @mapping: address space within which to check
472 * @start_byte: offset in bytes where the range starts
473 * @end_byte: offset in bytes where the range ends (inclusive)
475 * Find at least one page in the range supplied, usually used to check if
476 * direct writing in this range will trigger a writeback.
478 * Return: %true if at least one page exists in the specified range,
481 bool filemap_range_has_page(struct address_space *mapping,
482 loff_t start_byte, loff_t end_byte)
485 XA_STATE(xas, &mapping->i_pages, start_byte >> PAGE_SHIFT);
486 pgoff_t max = end_byte >> PAGE_SHIFT;
488 if (end_byte < start_byte)
493 page = xas_find(&xas, max);
494 if (xas_retry(&xas, page))
496 /* Shadow entries don't count */
497 if (xa_is_value(page))
500 * We don't need to try to pin this page; we're about to
501 * release the RCU lock anyway. It is enough to know that
502 * there was a page here recently.
510 EXPORT_SYMBOL(filemap_range_has_page);
512 static void __filemap_fdatawait_range(struct address_space *mapping,
513 loff_t start_byte, loff_t end_byte)
515 pgoff_t index = start_byte >> PAGE_SHIFT;
516 pgoff_t end = end_byte >> PAGE_SHIFT;
520 if (end_byte < start_byte)
524 while (index <= end) {
527 nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index,
528 end, PAGECACHE_TAG_WRITEBACK);
532 for (i = 0; i < nr_pages; i++) {
533 struct page *page = pvec.pages[i];
535 wait_on_page_writeback(page);
536 ClearPageError(page);
538 pagevec_release(&pvec);
544 * filemap_fdatawait_range - wait for writeback to complete
545 * @mapping: address space structure to wait for
546 * @start_byte: offset in bytes where the range starts
547 * @end_byte: offset in bytes where the range ends (inclusive)
549 * Walk the list of under-writeback pages of the given address space
550 * in the given range and wait for all of them. Check error status of
551 * the address space and return it.
553 * Since the error status of the address space is cleared by this function,
554 * callers are responsible for checking the return value and handling and/or
555 * reporting the error.
557 * Return: error status of the address space.
559 int filemap_fdatawait_range(struct address_space *mapping, loff_t start_byte,
562 __filemap_fdatawait_range(mapping, start_byte, end_byte);
563 return filemap_check_errors(mapping);
565 EXPORT_SYMBOL(filemap_fdatawait_range);
568 * filemap_fdatawait_range_keep_errors - wait for writeback to complete
569 * @mapping: address space structure to wait for
570 * @start_byte: offset in bytes where the range starts
571 * @end_byte: offset in bytes where the range ends (inclusive)
573 * Walk the list of under-writeback pages of the given address space in the
574 * given range and wait for all of them. Unlike filemap_fdatawait_range(),
575 * this function does not clear error status of the address space.
577 * Use this function if callers don't handle errors themselves. Expected
578 * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2),
581 int filemap_fdatawait_range_keep_errors(struct address_space *mapping,
582 loff_t start_byte, loff_t end_byte)
584 __filemap_fdatawait_range(mapping, start_byte, end_byte);
585 return filemap_check_and_keep_errors(mapping);
587 EXPORT_SYMBOL(filemap_fdatawait_range_keep_errors);
590 * file_fdatawait_range - wait for writeback to complete
591 * @file: file pointing to address space structure to wait for
592 * @start_byte: offset in bytes where the range starts
593 * @end_byte: offset in bytes where the range ends (inclusive)
595 * Walk the list of under-writeback pages of the address space that file
596 * refers to, in the given range and wait for all of them. Check error
597 * status of the address space vs. the file->f_wb_err cursor and return it.
599 * Since the error status of the file is advanced by this function,
600 * callers are responsible for checking the return value and handling and/or
601 * reporting the error.
603 * Return: error status of the address space vs. the file->f_wb_err cursor.
605 int file_fdatawait_range(struct file *file, loff_t start_byte, loff_t end_byte)
607 struct address_space *mapping = file->f_mapping;
609 __filemap_fdatawait_range(mapping, start_byte, end_byte);
610 return file_check_and_advance_wb_err(file);
612 EXPORT_SYMBOL(file_fdatawait_range);
615 * filemap_fdatawait_keep_errors - wait for writeback without clearing errors
616 * @mapping: address space structure to wait for
618 * Walk the list of under-writeback pages of the given address space
619 * and wait for all of them. Unlike filemap_fdatawait(), this function
620 * does not clear error status of the address space.
622 * Use this function if callers don't handle errors themselves. Expected
623 * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2),
626 * Return: error status of the address space.
628 int filemap_fdatawait_keep_errors(struct address_space *mapping)
630 __filemap_fdatawait_range(mapping, 0, LLONG_MAX);
631 return filemap_check_and_keep_errors(mapping);
633 EXPORT_SYMBOL(filemap_fdatawait_keep_errors);
635 /* Returns true if writeback might be needed or already in progress. */
636 static bool mapping_needs_writeback(struct address_space *mapping)
638 return mapping->nrpages;
641 static bool filemap_range_has_writeback(struct address_space *mapping,
642 loff_t start_byte, loff_t end_byte)
644 XA_STATE(xas, &mapping->i_pages, start_byte >> PAGE_SHIFT);
645 pgoff_t max = end_byte >> PAGE_SHIFT;
648 if (end_byte < start_byte)
652 xas_for_each(&xas, page, max) {
653 if (xas_retry(&xas, page))
655 if (xa_is_value(page))
657 if (PageDirty(page) || PageLocked(page) || PageWriteback(page))
666 * filemap_range_needs_writeback - check if range potentially needs writeback
667 * @mapping: address space within which to check
668 * @start_byte: offset in bytes where the range starts
669 * @end_byte: offset in bytes where the range ends (inclusive)
671 * Find at least one page in the range supplied, usually used to check if
672 * direct writing in this range will trigger a writeback. Used by O_DIRECT
673 * read/write with IOCB_NOWAIT, to see if the caller needs to do
674 * filemap_write_and_wait_range() before proceeding.
676 * Return: %true if the caller should do filemap_write_and_wait_range() before
677 * doing O_DIRECT to a page in this range, %false otherwise.
679 bool filemap_range_needs_writeback(struct address_space *mapping,
680 loff_t start_byte, loff_t end_byte)
682 if (!mapping_needs_writeback(mapping))
684 if (!mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) &&
685 !mapping_tagged(mapping, PAGECACHE_TAG_WRITEBACK))
687 return filemap_range_has_writeback(mapping, start_byte, end_byte);
689 EXPORT_SYMBOL_GPL(filemap_range_needs_writeback);
692 * filemap_write_and_wait_range - write out & wait on a file range
693 * @mapping: the address_space for the pages
694 * @lstart: offset in bytes where the range starts
695 * @lend: offset in bytes where the range ends (inclusive)
697 * Write out and wait upon file offsets lstart->lend, inclusive.
699 * Note that @lend is inclusive (describes the last byte to be written) so
700 * that this function can be used to write to the very end-of-file (end = -1).
702 * Return: error status of the address space.
704 int filemap_write_and_wait_range(struct address_space *mapping,
705 loff_t lstart, loff_t lend)
709 if (mapping_needs_writeback(mapping)) {
710 err = __filemap_fdatawrite_range(mapping, lstart, lend,
713 * Even if the above returned error, the pages may be
714 * written partially (e.g. -ENOSPC), so we wait for it.
715 * But the -EIO is special case, it may indicate the worst
716 * thing (e.g. bug) happened, so we avoid waiting for it.
719 int err2 = filemap_fdatawait_range(mapping,
724 /* Clear any previously stored errors */
725 filemap_check_errors(mapping);
728 err = filemap_check_errors(mapping);
732 EXPORT_SYMBOL(filemap_write_and_wait_range);
734 void __filemap_set_wb_err(struct address_space *mapping, int err)
736 errseq_t eseq = errseq_set(&mapping->wb_err, err);
738 trace_filemap_set_wb_err(mapping, eseq);
740 EXPORT_SYMBOL(__filemap_set_wb_err);
743 * file_check_and_advance_wb_err - report wb error (if any) that was previously
744 * and advance wb_err to current one
745 * @file: struct file on which the error is being reported
747 * When userland calls fsync (or something like nfsd does the equivalent), we
748 * want to report any writeback errors that occurred since the last fsync (or
749 * since the file was opened if there haven't been any).
751 * Grab the wb_err from the mapping. If it matches what we have in the file,
752 * then just quickly return 0. The file is all caught up.
754 * If it doesn't match, then take the mapping value, set the "seen" flag in
755 * it and try to swap it into place. If it works, or another task beat us
756 * to it with the new value, then update the f_wb_err and return the error
757 * portion. The error at this point must be reported via proper channels
758 * (a'la fsync, or NFS COMMIT operation, etc.).
760 * While we handle mapping->wb_err with atomic operations, the f_wb_err
761 * value is protected by the f_lock since we must ensure that it reflects
762 * the latest value swapped in for this file descriptor.
764 * Return: %0 on success, negative error code otherwise.
766 int file_check_and_advance_wb_err(struct file *file)
769 errseq_t old = READ_ONCE(file->f_wb_err);
770 struct address_space *mapping = file->f_mapping;
772 /* Locklessly handle the common case where nothing has changed */
773 if (errseq_check(&mapping->wb_err, old)) {
774 /* Something changed, must use slow path */
775 spin_lock(&file->f_lock);
776 old = file->f_wb_err;
777 err = errseq_check_and_advance(&mapping->wb_err,
779 trace_file_check_and_advance_wb_err(file, old);
780 spin_unlock(&file->f_lock);
784 * We're mostly using this function as a drop in replacement for
785 * filemap_check_errors. Clear AS_EIO/AS_ENOSPC to emulate the effect
786 * that the legacy code would have had on these flags.
788 clear_bit(AS_EIO, &mapping->flags);
789 clear_bit(AS_ENOSPC, &mapping->flags);
792 EXPORT_SYMBOL(file_check_and_advance_wb_err);
795 * file_write_and_wait_range - write out & wait on a file range
796 * @file: file pointing to address_space with pages
797 * @lstart: offset in bytes where the range starts
798 * @lend: offset in bytes where the range ends (inclusive)
800 * Write out and wait upon file offsets lstart->lend, inclusive.
802 * Note that @lend is inclusive (describes the last byte to be written) so
803 * that this function can be used to write to the very end-of-file (end = -1).
805 * After writing out and waiting on the data, we check and advance the
806 * f_wb_err cursor to the latest value, and return any errors detected there.
808 * Return: %0 on success, negative error code otherwise.
810 int file_write_and_wait_range(struct file *file, loff_t lstart, loff_t lend)
813 struct address_space *mapping = file->f_mapping;
815 if (mapping_needs_writeback(mapping)) {
816 err = __filemap_fdatawrite_range(mapping, lstart, lend,
818 /* See comment of filemap_write_and_wait() */
820 __filemap_fdatawait_range(mapping, lstart, lend);
822 err2 = file_check_and_advance_wb_err(file);
827 EXPORT_SYMBOL(file_write_and_wait_range);
830 * replace_page_cache_page - replace a pagecache page with a new one
831 * @old: page to be replaced
832 * @new: page to replace with
834 * This function replaces a page in the pagecache with a new one. On
835 * success it acquires the pagecache reference for the new page and
836 * drops it for the old page. Both the old and new pages must be
837 * locked. This function does not add the new page to the LRU, the
838 * caller must do that.
840 * The remove + add is atomic. This function cannot fail.
842 void replace_page_cache_page(struct page *old, struct page *new)
844 struct folio *fold = page_folio(old);
845 struct folio *fnew = page_folio(new);
846 struct address_space *mapping = old->mapping;
847 void (*freepage)(struct page *) = mapping->a_ops->freepage;
848 pgoff_t offset = old->index;
849 XA_STATE(xas, &mapping->i_pages, offset);
851 VM_BUG_ON_PAGE(!PageLocked(old), old);
852 VM_BUG_ON_PAGE(!PageLocked(new), new);
853 VM_BUG_ON_PAGE(new->mapping, new);
856 new->mapping = mapping;
859 mem_cgroup_migrate(fold, fnew);
862 xas_store(&xas, new);
865 /* hugetlb pages do not participate in page cache accounting. */
867 __dec_lruvec_page_state(old, NR_FILE_PAGES);
869 __inc_lruvec_page_state(new, NR_FILE_PAGES);
870 if (PageSwapBacked(old))
871 __dec_lruvec_page_state(old, NR_SHMEM);
872 if (PageSwapBacked(new))
873 __inc_lruvec_page_state(new, NR_SHMEM);
874 xas_unlock_irq(&xas);
879 EXPORT_SYMBOL_GPL(replace_page_cache_page);
881 noinline int __filemap_add_folio(struct address_space *mapping,
882 struct folio *folio, pgoff_t index, gfp_t gfp, void **shadowp)
884 XA_STATE(xas, &mapping->i_pages, index);
885 int huge = folio_test_hugetlb(folio);
887 bool charged = false;
889 VM_BUG_ON_FOLIO(!folio_test_locked(folio), folio);
890 VM_BUG_ON_FOLIO(folio_test_swapbacked(folio), folio);
891 mapping_set_update(&xas, mapping);
894 folio->mapping = mapping;
895 folio->index = index;
898 error = mem_cgroup_charge(folio, NULL, gfp);
899 VM_BUG_ON_FOLIO(index & (folio_nr_pages(folio) - 1), folio);
905 gfp &= GFP_RECLAIM_MASK;
908 unsigned int order = xa_get_order(xas.xa, xas.xa_index);
909 void *entry, *old = NULL;
911 if (order > folio_order(folio))
912 xas_split_alloc(&xas, xa_load(xas.xa, xas.xa_index),
915 xas_for_each_conflict(&xas, entry) {
917 if (!xa_is_value(entry)) {
918 xas_set_err(&xas, -EEXIST);
926 /* entry may have been split before we acquired lock */
927 order = xa_get_order(xas.xa, xas.xa_index);
928 if (order > folio_order(folio)) {
929 xas_split(&xas, old, order);
934 xas_store(&xas, folio);
940 /* hugetlb pages do not participate in page cache accounting */
942 __lruvec_stat_add_folio(folio, NR_FILE_PAGES);
944 xas_unlock_irq(&xas);
945 } while (xas_nomem(&xas, gfp));
947 if (xas_error(&xas)) {
948 error = xas_error(&xas);
950 mem_cgroup_uncharge(folio);
954 trace_mm_filemap_add_to_page_cache(&folio->page);
957 folio->mapping = NULL;
958 /* Leave page->index set: truncation relies upon it */
962 ALLOW_ERROR_INJECTION(__filemap_add_folio, ERRNO);
965 * add_to_page_cache_locked - add a locked page to the pagecache
967 * @mapping: the page's address_space
968 * @offset: page index
969 * @gfp_mask: page allocation mode
971 * This function is used to add a page to the pagecache. It must be locked.
972 * This function does not add the page to the LRU. The caller must do that.
974 * Return: %0 on success, negative error code otherwise.
976 int add_to_page_cache_locked(struct page *page, struct address_space *mapping,
977 pgoff_t offset, gfp_t gfp_mask)
979 return __filemap_add_folio(mapping, page_folio(page), offset,
982 EXPORT_SYMBOL(add_to_page_cache_locked);
984 int filemap_add_folio(struct address_space *mapping, struct folio *folio,
985 pgoff_t index, gfp_t gfp)
990 __folio_set_locked(folio);
991 ret = __filemap_add_folio(mapping, folio, index, gfp, &shadow);
993 __folio_clear_locked(folio);
996 * The folio might have been evicted from cache only
997 * recently, in which case it should be activated like
998 * any other repeatedly accessed folio.
999 * The exception is folios getting rewritten; evicting other
1000 * data from the working set, only to cache data that will
1001 * get overwritten with something else, is a waste of memory.
1003 WARN_ON_ONCE(folio_test_active(folio));
1004 if (!(gfp & __GFP_WRITE) && shadow)
1005 workingset_refault(folio, shadow);
1006 folio_add_lru(folio);
1010 EXPORT_SYMBOL_GPL(filemap_add_folio);
1013 struct folio *filemap_alloc_folio(gfp_t gfp, unsigned int order)
1016 struct folio *folio;
1018 if (cpuset_do_page_mem_spread()) {
1019 unsigned int cpuset_mems_cookie;
1021 cpuset_mems_cookie = read_mems_allowed_begin();
1022 n = cpuset_mem_spread_node();
1023 folio = __folio_alloc_node(gfp, order, n);
1024 } while (!folio && read_mems_allowed_retry(cpuset_mems_cookie));
1028 return folio_alloc(gfp, order);
1030 EXPORT_SYMBOL(filemap_alloc_folio);
1034 * filemap_invalidate_lock_two - lock invalidate_lock for two mappings
1036 * Lock exclusively invalidate_lock of any passed mapping that is not NULL.
1038 * @mapping1: the first mapping to lock
1039 * @mapping2: the second mapping to lock
1041 void filemap_invalidate_lock_two(struct address_space *mapping1,
1042 struct address_space *mapping2)
1044 if (mapping1 > mapping2)
1045 swap(mapping1, mapping2);
1047 down_write(&mapping1->invalidate_lock);
1048 if (mapping2 && mapping1 != mapping2)
1049 down_write_nested(&mapping2->invalidate_lock, 1);
1051 EXPORT_SYMBOL(filemap_invalidate_lock_two);
1054 * filemap_invalidate_unlock_two - unlock invalidate_lock for two mappings
1056 * Unlock exclusive invalidate_lock of any passed mapping that is not NULL.
1058 * @mapping1: the first mapping to unlock
1059 * @mapping2: the second mapping to unlock
1061 void filemap_invalidate_unlock_two(struct address_space *mapping1,
1062 struct address_space *mapping2)
1065 up_write(&mapping1->invalidate_lock);
1066 if (mapping2 && mapping1 != mapping2)
1067 up_write(&mapping2->invalidate_lock);
1069 EXPORT_SYMBOL(filemap_invalidate_unlock_two);
1072 * In order to wait for pages to become available there must be
1073 * waitqueues associated with pages. By using a hash table of
1074 * waitqueues where the bucket discipline is to maintain all
1075 * waiters on the same queue and wake all when any of the pages
1076 * become available, and for the woken contexts to check to be
1077 * sure the appropriate page became available, this saves space
1078 * at a cost of "thundering herd" phenomena during rare hash
1081 #define PAGE_WAIT_TABLE_BITS 8
1082 #define PAGE_WAIT_TABLE_SIZE (1 << PAGE_WAIT_TABLE_BITS)
1083 static wait_queue_head_t folio_wait_table[PAGE_WAIT_TABLE_SIZE] __cacheline_aligned;
1085 static wait_queue_head_t *folio_waitqueue(struct folio *folio)
1087 return &folio_wait_table[hash_ptr(folio, PAGE_WAIT_TABLE_BITS)];
1090 void __init pagecache_init(void)
1094 for (i = 0; i < PAGE_WAIT_TABLE_SIZE; i++)
1095 init_waitqueue_head(&folio_wait_table[i]);
1097 page_writeback_init();
1101 * The page wait code treats the "wait->flags" somewhat unusually, because
1102 * we have multiple different kinds of waits, not just the usual "exclusive"
1107 * (a) no special bits set:
1109 * We're just waiting for the bit to be released, and when a waker
1110 * calls the wakeup function, we set WQ_FLAG_WOKEN and wake it up,
1111 * and remove it from the wait queue.
1113 * Simple and straightforward.
1115 * (b) WQ_FLAG_EXCLUSIVE:
1117 * The waiter is waiting to get the lock, and only one waiter should
1118 * be woken up to avoid any thundering herd behavior. We'll set the
1119 * WQ_FLAG_WOKEN bit, wake it up, and remove it from the wait queue.
1121 * This is the traditional exclusive wait.
1123 * (c) WQ_FLAG_EXCLUSIVE | WQ_FLAG_CUSTOM:
1125 * The waiter is waiting to get the bit, and additionally wants the
1126 * lock to be transferred to it for fair lock behavior. If the lock
1127 * cannot be taken, we stop walking the wait queue without waking
1130 * This is the "fair lock handoff" case, and in addition to setting
1131 * WQ_FLAG_WOKEN, we set WQ_FLAG_DONE to let the waiter easily see
1132 * that it now has the lock.
1134 static int wake_page_function(wait_queue_entry_t *wait, unsigned mode, int sync, void *arg)
1137 struct wait_page_key *key = arg;
1138 struct wait_page_queue *wait_page
1139 = container_of(wait, struct wait_page_queue, wait);
1141 if (!wake_page_match(wait_page, key))
1145 * If it's a lock handoff wait, we get the bit for it, and
1146 * stop walking (and do not wake it up) if we can't.
1148 flags = wait->flags;
1149 if (flags & WQ_FLAG_EXCLUSIVE) {
1150 if (test_bit(key->bit_nr, &key->folio->flags))
1152 if (flags & WQ_FLAG_CUSTOM) {
1153 if (test_and_set_bit(key->bit_nr, &key->folio->flags))
1155 flags |= WQ_FLAG_DONE;
1160 * We are holding the wait-queue lock, but the waiter that
1161 * is waiting for this will be checking the flags without
1164 * So update the flags atomically, and wake up the waiter
1165 * afterwards to avoid any races. This store-release pairs
1166 * with the load-acquire in folio_wait_bit_common().
1168 smp_store_release(&wait->flags, flags | WQ_FLAG_WOKEN);
1169 wake_up_state(wait->private, mode);
1172 * Ok, we have successfully done what we're waiting for,
1173 * and we can unconditionally remove the wait entry.
1175 * Note that this pairs with the "finish_wait()" in the
1176 * waiter, and has to be the absolute last thing we do.
1177 * After this list_del_init(&wait->entry) the wait entry
1178 * might be de-allocated and the process might even have
1181 list_del_init_careful(&wait->entry);
1182 return (flags & WQ_FLAG_EXCLUSIVE) != 0;
1185 static void folio_wake_bit(struct folio *folio, int bit_nr)
1187 wait_queue_head_t *q = folio_waitqueue(folio);
1188 struct wait_page_key key;
1189 unsigned long flags;
1190 wait_queue_entry_t bookmark;
1193 key.bit_nr = bit_nr;
1197 bookmark.private = NULL;
1198 bookmark.func = NULL;
1199 INIT_LIST_HEAD(&bookmark.entry);
1201 spin_lock_irqsave(&q->lock, flags);
1202 __wake_up_locked_key_bookmark(q, TASK_NORMAL, &key, &bookmark);
1204 while (bookmark.flags & WQ_FLAG_BOOKMARK) {
1206 * Take a breather from holding the lock,
1207 * allow pages that finish wake up asynchronously
1208 * to acquire the lock and remove themselves
1211 spin_unlock_irqrestore(&q->lock, flags);
1213 spin_lock_irqsave(&q->lock, flags);
1214 __wake_up_locked_key_bookmark(q, TASK_NORMAL, &key, &bookmark);
1218 * It is possible for other pages to have collided on the waitqueue
1219 * hash, so in that case check for a page match. That prevents a long-
1222 * It is still possible to miss a case here, when we woke page waiters
1223 * and removed them from the waitqueue, but there are still other
1226 if (!waitqueue_active(q) || !key.page_match) {
1227 folio_clear_waiters(folio);
1229 * It's possible to miss clearing Waiters here, when we woke
1230 * our page waiters, but the hashed waitqueue has waiters for
1231 * other pages on it.
1233 * That's okay, it's a rare case. The next waker will clear it.
1236 spin_unlock_irqrestore(&q->lock, flags);
1239 static void folio_wake(struct folio *folio, int bit)
1241 if (!folio_test_waiters(folio))
1243 folio_wake_bit(folio, bit);
1247 * A choice of three behaviors for folio_wait_bit_common():
1250 EXCLUSIVE, /* Hold ref to page and take the bit when woken, like
1251 * __folio_lock() waiting on then setting PG_locked.
1253 SHARED, /* Hold ref to page and check the bit when woken, like
1254 * wait_on_page_writeback() waiting on PG_writeback.
1256 DROP, /* Drop ref to page before wait, no check when woken,
1257 * like put_and_wait_on_page_locked() on PG_locked.
1262 * Attempt to check (or get) the folio flag, and mark us done
1265 static inline bool folio_trylock_flag(struct folio *folio, int bit_nr,
1266 struct wait_queue_entry *wait)
1268 if (wait->flags & WQ_FLAG_EXCLUSIVE) {
1269 if (test_and_set_bit(bit_nr, &folio->flags))
1271 } else if (test_bit(bit_nr, &folio->flags))
1274 wait->flags |= WQ_FLAG_WOKEN | WQ_FLAG_DONE;
1278 /* How many times do we accept lock stealing from under a waiter? */
1279 int sysctl_page_lock_unfairness = 5;
1281 static inline int folio_wait_bit_common(struct folio *folio, int bit_nr,
1282 int state, enum behavior behavior)
1284 wait_queue_head_t *q = folio_waitqueue(folio);
1285 int unfairness = sysctl_page_lock_unfairness;
1286 struct wait_page_queue wait_page;
1287 wait_queue_entry_t *wait = &wait_page.wait;
1288 bool thrashing = false;
1289 bool delayacct = false;
1290 unsigned long pflags;
1292 if (bit_nr == PG_locked &&
1293 !folio_test_uptodate(folio) && folio_test_workingset(folio)) {
1294 if (!folio_test_swapbacked(folio)) {
1295 delayacct_thrashing_start();
1298 psi_memstall_enter(&pflags);
1303 wait->func = wake_page_function;
1304 wait_page.folio = folio;
1305 wait_page.bit_nr = bit_nr;
1309 if (behavior == EXCLUSIVE) {
1310 wait->flags = WQ_FLAG_EXCLUSIVE;
1311 if (--unfairness < 0)
1312 wait->flags |= WQ_FLAG_CUSTOM;
1316 * Do one last check whether we can get the
1317 * page bit synchronously.
1319 * Do the folio_set_waiters() marking before that
1320 * to let any waker we _just_ missed know they
1321 * need to wake us up (otherwise they'll never
1322 * even go to the slow case that looks at the
1323 * page queue), and add ourselves to the wait
1324 * queue if we need to sleep.
1326 * This part needs to be done under the queue
1327 * lock to avoid races.
1329 spin_lock_irq(&q->lock);
1330 folio_set_waiters(folio);
1331 if (!folio_trylock_flag(folio, bit_nr, wait))
1332 __add_wait_queue_entry_tail(q, wait);
1333 spin_unlock_irq(&q->lock);
1336 * From now on, all the logic will be based on
1337 * the WQ_FLAG_WOKEN and WQ_FLAG_DONE flag, to
1338 * see whether the page bit testing has already
1339 * been done by the wake function.
1341 * We can drop our reference to the folio.
1343 if (behavior == DROP)
1347 * Note that until the "finish_wait()", or until
1348 * we see the WQ_FLAG_WOKEN flag, we need to
1349 * be very careful with the 'wait->flags', because
1350 * we may race with a waker that sets them.
1355 set_current_state(state);
1357 /* Loop until we've been woken or interrupted */
1358 flags = smp_load_acquire(&wait->flags);
1359 if (!(flags & WQ_FLAG_WOKEN)) {
1360 if (signal_pending_state(state, current))
1367 /* If we were non-exclusive, we're done */
1368 if (behavior != EXCLUSIVE)
1371 /* If the waker got the lock for us, we're done */
1372 if (flags & WQ_FLAG_DONE)
1376 * Otherwise, if we're getting the lock, we need to
1377 * try to get it ourselves.
1379 * And if that fails, we'll have to retry this all.
1381 if (unlikely(test_and_set_bit(bit_nr, folio_flags(folio, 0))))
1384 wait->flags |= WQ_FLAG_DONE;
1389 * If a signal happened, this 'finish_wait()' may remove the last
1390 * waiter from the wait-queues, but the folio waiters bit will remain
1391 * set. That's ok. The next wakeup will take care of it, and trying
1392 * to do it here would be difficult and prone to races.
1394 finish_wait(q, wait);
1398 delayacct_thrashing_end();
1399 psi_memstall_leave(&pflags);
1403 * NOTE! The wait->flags weren't stable until we've done the
1404 * 'finish_wait()', and we could have exited the loop above due
1405 * to a signal, and had a wakeup event happen after the signal
1406 * test but before the 'finish_wait()'.
1408 * So only after the finish_wait() can we reliably determine
1409 * if we got woken up or not, so we can now figure out the final
1410 * return value based on that state without races.
1412 * Also note that WQ_FLAG_WOKEN is sufficient for a non-exclusive
1413 * waiter, but an exclusive one requires WQ_FLAG_DONE.
1415 if (behavior == EXCLUSIVE)
1416 return wait->flags & WQ_FLAG_DONE ? 0 : -EINTR;
1418 return wait->flags & WQ_FLAG_WOKEN ? 0 : -EINTR;
1421 void folio_wait_bit(struct folio *folio, int bit_nr)
1423 folio_wait_bit_common(folio, bit_nr, TASK_UNINTERRUPTIBLE, SHARED);
1425 EXPORT_SYMBOL(folio_wait_bit);
1427 int folio_wait_bit_killable(struct folio *folio, int bit_nr)
1429 return folio_wait_bit_common(folio, bit_nr, TASK_KILLABLE, SHARED);
1431 EXPORT_SYMBOL(folio_wait_bit_killable);
1434 * put_and_wait_on_page_locked - Drop a reference and wait for it to be unlocked
1435 * @page: The page to wait for.
1436 * @state: The sleep state (TASK_KILLABLE, TASK_UNINTERRUPTIBLE, etc).
1438 * The caller should hold a reference on @page. They expect the page to
1439 * become unlocked relatively soon, but do not wish to hold up migration
1440 * (for example) by holding the reference while waiting for the page to
1441 * come unlocked. After this function returns, the caller should not
1442 * dereference @page.
1444 * Return: 0 if the page was unlocked or -EINTR if interrupted by a signal.
1446 int put_and_wait_on_page_locked(struct page *page, int state)
1448 return folio_wait_bit_common(page_folio(page), PG_locked, state,
1453 * folio_add_wait_queue - Add an arbitrary waiter to a folio's wait queue
1454 * @folio: Folio defining the wait queue of interest
1455 * @waiter: Waiter to add to the queue
1457 * Add an arbitrary @waiter to the wait queue for the nominated @folio.
1459 void folio_add_wait_queue(struct folio *folio, wait_queue_entry_t *waiter)
1461 wait_queue_head_t *q = folio_waitqueue(folio);
1462 unsigned long flags;
1464 spin_lock_irqsave(&q->lock, flags);
1465 __add_wait_queue_entry_tail(q, waiter);
1466 folio_set_waiters(folio);
1467 spin_unlock_irqrestore(&q->lock, flags);
1469 EXPORT_SYMBOL_GPL(folio_add_wait_queue);
1471 #ifndef clear_bit_unlock_is_negative_byte
1474 * PG_waiters is the high bit in the same byte as PG_lock.
1476 * On x86 (and on many other architectures), we can clear PG_lock and
1477 * test the sign bit at the same time. But if the architecture does
1478 * not support that special operation, we just do this all by hand
1481 * The read of PG_waiters has to be after (or concurrently with) PG_locked
1482 * being cleared, but a memory barrier should be unnecessary since it is
1483 * in the same byte as PG_locked.
1485 static inline bool clear_bit_unlock_is_negative_byte(long nr, volatile void *mem)
1487 clear_bit_unlock(nr, mem);
1488 /* smp_mb__after_atomic(); */
1489 return test_bit(PG_waiters, mem);
1495 * folio_unlock - Unlock a locked folio.
1496 * @folio: The folio.
1498 * Unlocks the folio and wakes up any thread sleeping on the page lock.
1500 * Context: May be called from interrupt or process context. May not be
1501 * called from NMI context.
1503 void folio_unlock(struct folio *folio)
1505 /* Bit 7 allows x86 to check the byte's sign bit */
1506 BUILD_BUG_ON(PG_waiters != 7);
1507 BUILD_BUG_ON(PG_locked > 7);
1508 VM_BUG_ON_FOLIO(!folio_test_locked(folio), folio);
1509 if (clear_bit_unlock_is_negative_byte(PG_locked, folio_flags(folio, 0)))
1510 folio_wake_bit(folio, PG_locked);
1512 EXPORT_SYMBOL(folio_unlock);
1515 * folio_end_private_2 - Clear PG_private_2 and wake any waiters.
1516 * @folio: The folio.
1518 * Clear the PG_private_2 bit on a folio and wake up any sleepers waiting for
1519 * it. The folio reference held for PG_private_2 being set is released.
1521 * This is, for example, used when a netfs folio is being written to a local
1522 * disk cache, thereby allowing writes to the cache for the same folio to be
1525 void folio_end_private_2(struct folio *folio)
1527 VM_BUG_ON_FOLIO(!folio_test_private_2(folio), folio);
1528 clear_bit_unlock(PG_private_2, folio_flags(folio, 0));
1529 folio_wake_bit(folio, PG_private_2);
1532 EXPORT_SYMBOL(folio_end_private_2);
1535 * folio_wait_private_2 - Wait for PG_private_2 to be cleared on a folio.
1536 * @folio: The folio to wait on.
1538 * Wait for PG_private_2 (aka PG_fscache) to be cleared on a folio.
1540 void folio_wait_private_2(struct folio *folio)
1542 while (folio_test_private_2(folio))
1543 folio_wait_bit(folio, PG_private_2);
1545 EXPORT_SYMBOL(folio_wait_private_2);
1548 * folio_wait_private_2_killable - Wait for PG_private_2 to be cleared on a folio.
1549 * @folio: The folio to wait on.
1551 * Wait for PG_private_2 (aka PG_fscache) to be cleared on a folio or until a
1552 * fatal signal is received by the calling task.
1555 * - 0 if successful.
1556 * - -EINTR if a fatal signal was encountered.
1558 int folio_wait_private_2_killable(struct folio *folio)
1562 while (folio_test_private_2(folio)) {
1563 ret = folio_wait_bit_killable(folio, PG_private_2);
1570 EXPORT_SYMBOL(folio_wait_private_2_killable);
1573 * folio_end_writeback - End writeback against a folio.
1574 * @folio: The folio.
1576 void folio_end_writeback(struct folio *folio)
1579 * folio_test_clear_reclaim() could be used here but it is an
1580 * atomic operation and overkill in this particular case. Failing
1581 * to shuffle a folio marked for immediate reclaim is too mild
1582 * a gain to justify taking an atomic operation penalty at the
1583 * end of every folio writeback.
1585 if (folio_test_reclaim(folio)) {
1586 folio_clear_reclaim(folio);
1587 folio_rotate_reclaimable(folio);
1591 * Writeback does not hold a folio reference of its own, relying
1592 * on truncation to wait for the clearing of PG_writeback.
1593 * But here we must make sure that the folio is not freed and
1594 * reused before the folio_wake().
1597 if (!__folio_end_writeback(folio))
1600 smp_mb__after_atomic();
1601 folio_wake(folio, PG_writeback);
1602 acct_reclaim_writeback(folio);
1605 EXPORT_SYMBOL(folio_end_writeback);
1608 * After completing I/O on a page, call this routine to update the page
1609 * flags appropriately
1611 void page_endio(struct page *page, bool is_write, int err)
1615 SetPageUptodate(page);
1617 ClearPageUptodate(page);
1623 struct address_space *mapping;
1626 mapping = page_mapping(page);
1628 mapping_set_error(mapping, err);
1630 end_page_writeback(page);
1633 EXPORT_SYMBOL_GPL(page_endio);
1636 * __folio_lock - Get a lock on the folio, assuming we need to sleep to get it.
1637 * @folio: The folio to lock
1639 void __folio_lock(struct folio *folio)
1641 folio_wait_bit_common(folio, PG_locked, TASK_UNINTERRUPTIBLE,
1644 EXPORT_SYMBOL(__folio_lock);
1646 int __folio_lock_killable(struct folio *folio)
1648 return folio_wait_bit_common(folio, PG_locked, TASK_KILLABLE,
1651 EXPORT_SYMBOL_GPL(__folio_lock_killable);
1653 static int __folio_lock_async(struct folio *folio, struct wait_page_queue *wait)
1655 struct wait_queue_head *q = folio_waitqueue(folio);
1658 wait->folio = folio;
1659 wait->bit_nr = PG_locked;
1661 spin_lock_irq(&q->lock);
1662 __add_wait_queue_entry_tail(q, &wait->wait);
1663 folio_set_waiters(folio);
1664 ret = !folio_trylock(folio);
1666 * If we were successful now, we know we're still on the
1667 * waitqueue as we're still under the lock. This means it's
1668 * safe to remove and return success, we know the callback
1669 * isn't going to trigger.
1672 __remove_wait_queue(q, &wait->wait);
1675 spin_unlock_irq(&q->lock);
1681 * true - folio is locked; mmap_lock is still held.
1682 * false - folio is not locked.
1683 * mmap_lock has been released (mmap_read_unlock(), unless flags had both
1684 * FAULT_FLAG_ALLOW_RETRY and FAULT_FLAG_RETRY_NOWAIT set, in
1685 * which case mmap_lock is still held.
1687 * If neither ALLOW_RETRY nor KILLABLE are set, will always return true
1688 * with the folio locked and the mmap_lock unperturbed.
1690 bool __folio_lock_or_retry(struct folio *folio, struct mm_struct *mm,
1693 if (fault_flag_allow_retry_first(flags)) {
1695 * CAUTION! In this case, mmap_lock is not released
1696 * even though return 0.
1698 if (flags & FAULT_FLAG_RETRY_NOWAIT)
1701 mmap_read_unlock(mm);
1702 if (flags & FAULT_FLAG_KILLABLE)
1703 folio_wait_locked_killable(folio);
1705 folio_wait_locked(folio);
1708 if (flags & FAULT_FLAG_KILLABLE) {
1711 ret = __folio_lock_killable(folio);
1713 mmap_read_unlock(mm);
1717 __folio_lock(folio);
1724 * page_cache_next_miss() - Find the next gap in the page cache.
1725 * @mapping: Mapping.
1727 * @max_scan: Maximum range to search.
1729 * Search the range [index, min(index + max_scan - 1, ULONG_MAX)] for the
1730 * gap with the lowest index.
1732 * This function may be called under the rcu_read_lock. However, this will
1733 * not atomically search a snapshot of the cache at a single point in time.
1734 * For example, if a gap is created at index 5, then subsequently a gap is
1735 * created at index 10, page_cache_next_miss covering both indices may
1736 * return 10 if called under the rcu_read_lock.
1738 * Return: The index of the gap if found, otherwise an index outside the
1739 * range specified (in which case 'return - index >= max_scan' will be true).
1740 * In the rare case of index wrap-around, 0 will be returned.
1742 pgoff_t page_cache_next_miss(struct address_space *mapping,
1743 pgoff_t index, unsigned long max_scan)
1745 XA_STATE(xas, &mapping->i_pages, index);
1747 while (max_scan--) {
1748 void *entry = xas_next(&xas);
1749 if (!entry || xa_is_value(entry))
1751 if (xas.xa_index == 0)
1755 return xas.xa_index;
1757 EXPORT_SYMBOL(page_cache_next_miss);
1760 * page_cache_prev_miss() - Find the previous gap in the page cache.
1761 * @mapping: Mapping.
1763 * @max_scan: Maximum range to search.
1765 * Search the range [max(index - max_scan + 1, 0), index] for the
1766 * gap with the highest index.
1768 * This function may be called under the rcu_read_lock. However, this will
1769 * not atomically search a snapshot of the cache at a single point in time.
1770 * For example, if a gap is created at index 10, then subsequently a gap is
1771 * created at index 5, page_cache_prev_miss() covering both indices may
1772 * return 5 if called under the rcu_read_lock.
1774 * Return: The index of the gap if found, otherwise an index outside the
1775 * range specified (in which case 'index - return >= max_scan' will be true).
1776 * In the rare case of wrap-around, ULONG_MAX will be returned.
1778 pgoff_t page_cache_prev_miss(struct address_space *mapping,
1779 pgoff_t index, unsigned long max_scan)
1781 XA_STATE(xas, &mapping->i_pages, index);
1783 while (max_scan--) {
1784 void *entry = xas_prev(&xas);
1785 if (!entry || xa_is_value(entry))
1787 if (xas.xa_index == ULONG_MAX)
1791 return xas.xa_index;
1793 EXPORT_SYMBOL(page_cache_prev_miss);
1796 * Lockless page cache protocol:
1797 * On the lookup side:
1798 * 1. Load the folio from i_pages
1799 * 2. Increment the refcount if it's not zero
1800 * 3. If the folio is not found by xas_reload(), put the refcount and retry
1802 * On the removal side:
1803 * A. Freeze the page (by zeroing the refcount if nobody else has a reference)
1804 * B. Remove the page from i_pages
1805 * C. Return the page to the page allocator
1807 * This means that any page may have its reference count temporarily
1808 * increased by a speculative page cache (or fast GUP) lookup as it can
1809 * be allocated by another user before the RCU grace period expires.
1810 * Because the refcount temporarily acquired here may end up being the
1811 * last refcount on the page, any page allocation must be freeable by
1816 * mapping_get_entry - Get a page cache entry.
1817 * @mapping: the address_space to search
1818 * @index: The page cache index.
1820 * Looks up the page cache entry at @mapping & @index. If it is a folio,
1821 * it is returned with an increased refcount. If it is a shadow entry
1822 * of a previously evicted folio, or a swap entry from shmem/tmpfs,
1823 * it is returned without further action.
1825 * Return: The folio, swap or shadow entry, %NULL if nothing is found.
1827 static void *mapping_get_entry(struct address_space *mapping, pgoff_t index)
1829 XA_STATE(xas, &mapping->i_pages, index);
1830 struct folio *folio;
1835 folio = xas_load(&xas);
1836 if (xas_retry(&xas, folio))
1839 * A shadow entry of a recently evicted page, or a swap entry from
1840 * shmem/tmpfs. Return it without attempting to raise page count.
1842 if (!folio || xa_is_value(folio))
1845 if (!folio_try_get_rcu(folio))
1848 if (unlikely(folio != xas_reload(&xas))) {
1859 * __filemap_get_folio - Find and get a reference to a folio.
1860 * @mapping: The address_space to search.
1861 * @index: The page index.
1862 * @fgp_flags: %FGP flags modify how the folio is returned.
1863 * @gfp: Memory allocation flags to use if %FGP_CREAT is specified.
1865 * Looks up the page cache entry at @mapping & @index.
1867 * @fgp_flags can be zero or more of these flags:
1869 * * %FGP_ACCESSED - The folio will be marked accessed.
1870 * * %FGP_LOCK - The folio is returned locked.
1871 * * %FGP_ENTRY - If there is a shadow / swap / DAX entry, return it
1872 * instead of allocating a new folio to replace it.
1873 * * %FGP_CREAT - If no page is present then a new page is allocated using
1874 * @gfp and added to the page cache and the VM's LRU list.
1875 * The page is returned locked and with an increased refcount.
1876 * * %FGP_FOR_MMAP - The caller wants to do its own locking dance if the
1877 * page is already in cache. If the page was allocated, unlock it before
1878 * returning so the caller can do the same dance.
1879 * * %FGP_WRITE - The page will be written to by the caller.
1880 * * %FGP_NOFS - __GFP_FS will get cleared in gfp.
1881 * * %FGP_NOWAIT - Don't get blocked by page lock.
1882 * * %FGP_STABLE - Wait for the folio to be stable (finished writeback)
1884 * If %FGP_LOCK or %FGP_CREAT are specified then the function may sleep even
1885 * if the %GFP flags specified for %FGP_CREAT are atomic.
1887 * If there is a page cache page, it is returned with an increased refcount.
1889 * Return: The found folio or %NULL otherwise.
1891 struct folio *__filemap_get_folio(struct address_space *mapping, pgoff_t index,
1892 int fgp_flags, gfp_t gfp)
1894 struct folio *folio;
1897 folio = mapping_get_entry(mapping, index);
1898 if (xa_is_value(folio)) {
1899 if (fgp_flags & FGP_ENTRY)
1906 if (fgp_flags & FGP_LOCK) {
1907 if (fgp_flags & FGP_NOWAIT) {
1908 if (!folio_trylock(folio)) {
1916 /* Has the page been truncated? */
1917 if (unlikely(folio->mapping != mapping)) {
1918 folio_unlock(folio);
1922 VM_BUG_ON_FOLIO(!folio_contains(folio, index), folio);
1925 if (fgp_flags & FGP_ACCESSED)
1926 folio_mark_accessed(folio);
1927 else if (fgp_flags & FGP_WRITE) {
1928 /* Clear idle flag for buffer write */
1929 if (folio_test_idle(folio))
1930 folio_clear_idle(folio);
1933 if (fgp_flags & FGP_STABLE)
1934 folio_wait_stable(folio);
1936 if (!folio && (fgp_flags & FGP_CREAT)) {
1938 if ((fgp_flags & FGP_WRITE) && mapping_can_writeback(mapping))
1940 if (fgp_flags & FGP_NOFS)
1943 folio = filemap_alloc_folio(gfp, 0);
1947 if (WARN_ON_ONCE(!(fgp_flags & (FGP_LOCK | FGP_FOR_MMAP))))
1948 fgp_flags |= FGP_LOCK;
1950 /* Init accessed so avoid atomic mark_page_accessed later */
1951 if (fgp_flags & FGP_ACCESSED)
1952 __folio_set_referenced(folio);
1954 err = filemap_add_folio(mapping, folio, index, gfp);
1955 if (unlikely(err)) {
1963 * filemap_add_folio locks the page, and for mmap
1964 * we expect an unlocked page.
1966 if (folio && (fgp_flags & FGP_FOR_MMAP))
1967 folio_unlock(folio);
1972 EXPORT_SYMBOL(__filemap_get_folio);
1974 static inline struct page *find_get_entry(struct xa_state *xas, pgoff_t max,
1980 if (mark == XA_PRESENT)
1981 page = xas_find(xas, max);
1983 page = xas_find_marked(xas, max, mark);
1985 if (xas_retry(xas, page))
1988 * A shadow entry of a recently evicted page, a swap
1989 * entry from shmem/tmpfs or a DAX entry. Return it
1990 * without attempting to raise page count.
1992 if (!page || xa_is_value(page))
1995 if (!page_cache_get_speculative(page))
1998 /* Has the page moved or been split? */
1999 if (unlikely(page != xas_reload(xas))) {
2011 * find_get_entries - gang pagecache lookup
2012 * @mapping: The address_space to search
2013 * @start: The starting page cache index
2014 * @end: The final page index (inclusive).
2015 * @pvec: Where the resulting entries are placed.
2016 * @indices: The cache indices corresponding to the entries in @entries
2018 * find_get_entries() will search for and return a batch of entries in
2019 * the mapping. The entries are placed in @pvec. find_get_entries()
2020 * takes a reference on any actual pages it returns.
2022 * The search returns a group of mapping-contiguous page cache entries
2023 * with ascending indexes. There may be holes in the indices due to
2024 * not-present pages.
2026 * Any shadow entries of evicted pages, or swap entries from
2027 * shmem/tmpfs, are included in the returned array.
2029 * If it finds a Transparent Huge Page, head or tail, find_get_entries()
2030 * stops at that page: the caller is likely to have a better way to handle
2031 * the compound page as a whole, and then skip its extent, than repeatedly
2032 * calling find_get_entries() to return all its tails.
2034 * Return: the number of pages and shadow entries which were found.
2036 unsigned find_get_entries(struct address_space *mapping, pgoff_t start,
2037 pgoff_t end, struct pagevec *pvec, pgoff_t *indices)
2039 XA_STATE(xas, &mapping->i_pages, start);
2041 unsigned int ret = 0;
2042 unsigned nr_entries = PAGEVEC_SIZE;
2045 while ((page = find_get_entry(&xas, end, XA_PRESENT))) {
2047 * Terminate early on finding a THP, to allow the caller to
2048 * handle it all at once; but continue if this is hugetlbfs.
2050 if (!xa_is_value(page) && PageTransHuge(page) &&
2052 page = find_subpage(page, xas.xa_index);
2053 nr_entries = ret + 1;
2056 indices[ret] = xas.xa_index;
2057 pvec->pages[ret] = page;
2058 if (++ret == nr_entries)
2068 * find_lock_entries - Find a batch of pagecache entries.
2069 * @mapping: The address_space to search.
2070 * @start: The starting page cache index.
2071 * @end: The final page index (inclusive).
2072 * @pvec: Where the resulting entries are placed.
2073 * @indices: The cache indices of the entries in @pvec.
2075 * find_lock_entries() will return a batch of entries from @mapping.
2076 * Swap, shadow and DAX entries are included. Pages are returned
2077 * locked and with an incremented refcount. Pages which are locked by
2078 * somebody else or under writeback are skipped. Only the head page of
2079 * a THP is returned. Pages which are partially outside the range are
2082 * The entries have ascending indexes. The indices may not be consecutive
2083 * due to not-present entries, THP pages, pages which could not be locked
2084 * or pages under writeback.
2086 * Return: The number of entries which were found.
2088 unsigned find_lock_entries(struct address_space *mapping, pgoff_t start,
2089 pgoff_t end, struct pagevec *pvec, pgoff_t *indices)
2091 XA_STATE(xas, &mapping->i_pages, start);
2095 while ((page = find_get_entry(&xas, end, XA_PRESENT))) {
2096 if (!xa_is_value(page)) {
2097 if (page->index < start)
2099 if (page->index + thp_nr_pages(page) - 1 > end)
2101 if (!trylock_page(page))
2103 if (page->mapping != mapping || PageWriteback(page))
2105 VM_BUG_ON_PAGE(!thp_contains(page, xas.xa_index),
2108 indices[pvec->nr] = xas.xa_index;
2109 if (!pagevec_add(pvec, page))
2117 if (!xa_is_value(page) && PageTransHuge(page)) {
2118 unsigned int nr_pages = thp_nr_pages(page);
2120 /* Final THP may cross MAX_LFS_FILESIZE on 32-bit */
2121 xas_set(&xas, page->index + nr_pages);
2122 if (xas.xa_index < nr_pages)
2128 return pagevec_count(pvec);
2132 * find_get_pages_range - gang pagecache lookup
2133 * @mapping: The address_space to search
2134 * @start: The starting page index
2135 * @end: The final page index (inclusive)
2136 * @nr_pages: The maximum number of pages
2137 * @pages: Where the resulting pages are placed
2139 * find_get_pages_range() will search for and return a group of up to @nr_pages
2140 * pages in the mapping starting at index @start and up to index @end
2141 * (inclusive). The pages are placed at @pages. find_get_pages_range() takes
2142 * a reference against the returned pages.
2144 * The search returns a group of mapping-contiguous pages with ascending
2145 * indexes. There may be holes in the indices due to not-present pages.
2146 * We also update @start to index the next page for the traversal.
2148 * Return: the number of pages which were found. If this number is
2149 * smaller than @nr_pages, the end of specified range has been
2152 unsigned find_get_pages_range(struct address_space *mapping, pgoff_t *start,
2153 pgoff_t end, unsigned int nr_pages,
2154 struct page **pages)
2156 XA_STATE(xas, &mapping->i_pages, *start);
2160 if (unlikely(!nr_pages))
2164 while ((page = find_get_entry(&xas, end, XA_PRESENT))) {
2165 /* Skip over shadow, swap and DAX entries */
2166 if (xa_is_value(page))
2169 pages[ret] = find_subpage(page, xas.xa_index);
2170 if (++ret == nr_pages) {
2171 *start = xas.xa_index + 1;
2177 * We come here when there is no page beyond @end. We take care to not
2178 * overflow the index @start as it confuses some of the callers. This
2179 * breaks the iteration when there is a page at index -1 but that is
2180 * already broken anyway.
2182 if (end == (pgoff_t)-1)
2183 *start = (pgoff_t)-1;
2193 * find_get_pages_contig - gang contiguous pagecache lookup
2194 * @mapping: The address_space to search
2195 * @index: The starting page index
2196 * @nr_pages: The maximum number of pages
2197 * @pages: Where the resulting pages are placed
2199 * find_get_pages_contig() works exactly like find_get_pages(), except
2200 * that the returned number of pages are guaranteed to be contiguous.
2202 * Return: the number of pages which were found.
2204 unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index,
2205 unsigned int nr_pages, struct page **pages)
2207 XA_STATE(xas, &mapping->i_pages, index);
2209 unsigned int ret = 0;
2211 if (unlikely(!nr_pages))
2215 for (page = xas_load(&xas); page; page = xas_next(&xas)) {
2216 if (xas_retry(&xas, page))
2219 * If the entry has been swapped out, we can stop looking.
2220 * No current caller is looking for DAX entries.
2222 if (xa_is_value(page))
2225 if (!page_cache_get_speculative(page))
2228 /* Has the page moved or been split? */
2229 if (unlikely(page != xas_reload(&xas)))
2232 pages[ret] = find_subpage(page, xas.xa_index);
2233 if (++ret == nr_pages)
2244 EXPORT_SYMBOL(find_get_pages_contig);
2247 * find_get_pages_range_tag - Find and return head pages matching @tag.
2248 * @mapping: the address_space to search
2249 * @index: the starting page index
2250 * @end: The final page index (inclusive)
2251 * @tag: the tag index
2252 * @nr_pages: the maximum number of pages
2253 * @pages: where the resulting pages are placed
2255 * Like find_get_pages(), except we only return head pages which are tagged
2256 * with @tag. @index is updated to the index immediately after the last
2257 * page we return, ready for the next iteration.
2259 * Return: the number of pages which were found.
2261 unsigned find_get_pages_range_tag(struct address_space *mapping, pgoff_t *index,
2262 pgoff_t end, xa_mark_t tag, unsigned int nr_pages,
2263 struct page **pages)
2265 XA_STATE(xas, &mapping->i_pages, *index);
2269 if (unlikely(!nr_pages))
2273 while ((page = find_get_entry(&xas, end, tag))) {
2275 * Shadow entries should never be tagged, but this iteration
2276 * is lockless so there is a window for page reclaim to evict
2277 * a page we saw tagged. Skip over it.
2279 if (xa_is_value(page))
2283 if (++ret == nr_pages) {
2284 *index = page->index + thp_nr_pages(page);
2290 * We come here when we got to @end. We take care to not overflow the
2291 * index @index as it confuses some of the callers. This breaks the
2292 * iteration when there is a page at index -1 but that is already
2295 if (end == (pgoff_t)-1)
2296 *index = (pgoff_t)-1;
2304 EXPORT_SYMBOL(find_get_pages_range_tag);
2307 * CD/DVDs are error prone. When a medium error occurs, the driver may fail
2308 * a _large_ part of the i/o request. Imagine the worst scenario:
2310 * ---R__________________________________________B__________
2311 * ^ reading here ^ bad block(assume 4k)
2313 * read(R) => miss => readahead(R...B) => media error => frustrating retries
2314 * => failing the whole request => read(R) => read(R+1) =>
2315 * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
2316 * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
2317 * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
2319 * It is going insane. Fix it by quickly scaling down the readahead size.
2321 static void shrink_readahead_size_eio(struct file_ra_state *ra)
2327 * filemap_get_read_batch - Get a batch of pages for read
2329 * Get a batch of pages which represent a contiguous range of bytes
2330 * in the file. No tail pages will be returned. If @index is in the
2331 * middle of a THP, the entire THP will be returned. The last page in
2332 * the batch may have Readahead set or be not Uptodate so that the
2333 * caller can take the appropriate action.
2335 static void filemap_get_read_batch(struct address_space *mapping,
2336 pgoff_t index, pgoff_t max, struct pagevec *pvec)
2338 XA_STATE(xas, &mapping->i_pages, index);
2342 for (head = xas_load(&xas); head; head = xas_next(&xas)) {
2343 if (xas_retry(&xas, head))
2345 if (xas.xa_index > max || xa_is_value(head))
2347 if (!page_cache_get_speculative(head))
2350 /* Has the page moved or been split? */
2351 if (unlikely(head != xas_reload(&xas)))
2354 if (!pagevec_add(pvec, head))
2356 if (!PageUptodate(head))
2358 if (PageReadahead(head))
2360 xas.xa_index = head->index + thp_nr_pages(head) - 1;
2361 xas.xa_offset = (xas.xa_index >> xas.xa_shift) & XA_CHUNK_MASK;
2371 static int filemap_read_page(struct file *file, struct address_space *mapping,
2377 * A previous I/O error may have been due to temporary failures,
2378 * eg. multipath errors. PG_error will be set again if readpage
2381 ClearPageError(page);
2382 /* Start the actual read. The read will unlock the page. */
2383 error = mapping->a_ops->readpage(file, page);
2387 error = wait_on_page_locked_killable(page);
2390 if (PageUptodate(page))
2392 shrink_readahead_size_eio(&file->f_ra);
2396 static bool filemap_range_uptodate(struct address_space *mapping,
2397 loff_t pos, struct iov_iter *iter, struct page *page)
2401 if (PageUptodate(page))
2403 /* pipes can't handle partially uptodate pages */
2404 if (iov_iter_is_pipe(iter))
2406 if (!mapping->a_ops->is_partially_uptodate)
2408 if (mapping->host->i_blkbits >= (PAGE_SHIFT + thp_order(page)))
2411 count = iter->count;
2412 if (page_offset(page) > pos) {
2413 count -= page_offset(page) - pos;
2416 pos -= page_offset(page);
2419 return mapping->a_ops->is_partially_uptodate(page, pos, count);
2422 static int filemap_update_page(struct kiocb *iocb,
2423 struct address_space *mapping, struct iov_iter *iter,
2426 struct folio *folio = page_folio(page);
2429 if (iocb->ki_flags & IOCB_NOWAIT) {
2430 if (!filemap_invalidate_trylock_shared(mapping))
2433 filemap_invalidate_lock_shared(mapping);
2436 if (!folio_trylock(folio)) {
2438 if (iocb->ki_flags & (IOCB_NOWAIT | IOCB_NOIO))
2439 goto unlock_mapping;
2440 if (!(iocb->ki_flags & IOCB_WAITQ)) {
2441 filemap_invalidate_unlock_shared(mapping);
2442 put_and_wait_on_page_locked(&folio->page, TASK_KILLABLE);
2443 return AOP_TRUNCATED_PAGE;
2445 error = __folio_lock_async(folio, iocb->ki_waitq);
2447 goto unlock_mapping;
2450 error = AOP_TRUNCATED_PAGE;
2451 if (!folio->mapping)
2455 if (filemap_range_uptodate(mapping, iocb->ki_pos, iter, &folio->page))
2459 if (iocb->ki_flags & (IOCB_NOIO | IOCB_NOWAIT | IOCB_WAITQ))
2462 error = filemap_read_page(iocb->ki_filp, mapping, &folio->page);
2463 goto unlock_mapping;
2465 folio_unlock(folio);
2467 filemap_invalidate_unlock_shared(mapping);
2468 if (error == AOP_TRUNCATED_PAGE)
2473 static int filemap_create_page(struct file *file,
2474 struct address_space *mapping, pgoff_t index,
2475 struct pagevec *pvec)
2480 page = page_cache_alloc(mapping);
2485 * Protect against truncate / hole punch. Grabbing invalidate_lock here
2486 * assures we cannot instantiate and bring uptodate new pagecache pages
2487 * after evicting page cache during truncate and before actually
2488 * freeing blocks. Note that we could release invalidate_lock after
2489 * inserting the page into page cache as the locked page would then be
2490 * enough to synchronize with hole punching. But there are code paths
2491 * such as filemap_update_page() filling in partially uptodate pages or
2492 * ->readpages() that need to hold invalidate_lock while mapping blocks
2493 * for IO so let's hold the lock here as well to keep locking rules
2496 filemap_invalidate_lock_shared(mapping);
2497 error = add_to_page_cache_lru(page, mapping, index,
2498 mapping_gfp_constraint(mapping, GFP_KERNEL));
2499 if (error == -EEXIST)
2500 error = AOP_TRUNCATED_PAGE;
2504 error = filemap_read_page(file, mapping, page);
2508 filemap_invalidate_unlock_shared(mapping);
2509 pagevec_add(pvec, page);
2512 filemap_invalidate_unlock_shared(mapping);
2517 static int filemap_readahead(struct kiocb *iocb, struct file *file,
2518 struct address_space *mapping, struct page *page,
2521 if (iocb->ki_flags & IOCB_NOIO)
2523 page_cache_async_readahead(mapping, &file->f_ra, file, page,
2524 page->index, last_index - page->index);
2528 static int filemap_get_pages(struct kiocb *iocb, struct iov_iter *iter,
2529 struct pagevec *pvec)
2531 struct file *filp = iocb->ki_filp;
2532 struct address_space *mapping = filp->f_mapping;
2533 struct file_ra_state *ra = &filp->f_ra;
2534 pgoff_t index = iocb->ki_pos >> PAGE_SHIFT;
2539 last_index = DIV_ROUND_UP(iocb->ki_pos + iter->count, PAGE_SIZE);
2541 if (fatal_signal_pending(current))
2544 filemap_get_read_batch(mapping, index, last_index, pvec);
2545 if (!pagevec_count(pvec)) {
2546 if (iocb->ki_flags & IOCB_NOIO)
2548 page_cache_sync_readahead(mapping, ra, filp, index,
2549 last_index - index);
2550 filemap_get_read_batch(mapping, index, last_index, pvec);
2552 if (!pagevec_count(pvec)) {
2553 if (iocb->ki_flags & (IOCB_NOWAIT | IOCB_WAITQ))
2555 err = filemap_create_page(filp, mapping,
2556 iocb->ki_pos >> PAGE_SHIFT, pvec);
2557 if (err == AOP_TRUNCATED_PAGE)
2562 page = pvec->pages[pagevec_count(pvec) - 1];
2563 if (PageReadahead(page)) {
2564 err = filemap_readahead(iocb, filp, mapping, page, last_index);
2568 if (!PageUptodate(page)) {
2569 if ((iocb->ki_flags & IOCB_WAITQ) && pagevec_count(pvec) > 1)
2570 iocb->ki_flags |= IOCB_NOWAIT;
2571 err = filemap_update_page(iocb, mapping, iter, page);
2580 if (likely(--pvec->nr))
2582 if (err == AOP_TRUNCATED_PAGE)
2588 * filemap_read - Read data from the page cache.
2589 * @iocb: The iocb to read.
2590 * @iter: Destination for the data.
2591 * @already_read: Number of bytes already read by the caller.
2593 * Copies data from the page cache. If the data is not currently present,
2594 * uses the readahead and readpage address_space operations to fetch it.
2596 * Return: Total number of bytes copied, including those already read by
2597 * the caller. If an error happens before any bytes are copied, returns
2598 * a negative error number.
2600 ssize_t filemap_read(struct kiocb *iocb, struct iov_iter *iter,
2601 ssize_t already_read)
2603 struct file *filp = iocb->ki_filp;
2604 struct file_ra_state *ra = &filp->f_ra;
2605 struct address_space *mapping = filp->f_mapping;
2606 struct inode *inode = mapping->host;
2607 struct pagevec pvec;
2609 bool writably_mapped;
2610 loff_t isize, end_offset;
2612 if (unlikely(iocb->ki_pos >= inode->i_sb->s_maxbytes))
2614 if (unlikely(!iov_iter_count(iter)))
2617 iov_iter_truncate(iter, inode->i_sb->s_maxbytes);
2618 pagevec_init(&pvec);
2624 * If we've already successfully copied some data, then we
2625 * can no longer safely return -EIOCBQUEUED. Hence mark
2626 * an async read NOWAIT at that point.
2628 if ((iocb->ki_flags & IOCB_WAITQ) && already_read)
2629 iocb->ki_flags |= IOCB_NOWAIT;
2631 if (unlikely(iocb->ki_pos >= i_size_read(inode)))
2634 error = filemap_get_pages(iocb, iter, &pvec);
2639 * i_size must be checked after we know the pages are Uptodate.
2641 * Checking i_size after the check allows us to calculate
2642 * the correct value for "nr", which means the zero-filled
2643 * part of the page is not copied back to userspace (unless
2644 * another truncate extends the file - this is desired though).
2646 isize = i_size_read(inode);
2647 if (unlikely(iocb->ki_pos >= isize))
2649 end_offset = min_t(loff_t, isize, iocb->ki_pos + iter->count);
2652 * Once we start copying data, we don't want to be touching any
2653 * cachelines that might be contended:
2655 writably_mapped = mapping_writably_mapped(mapping);
2658 * When a sequential read accesses a page several times, only
2659 * mark it as accessed the first time.
2661 if (iocb->ki_pos >> PAGE_SHIFT !=
2662 ra->prev_pos >> PAGE_SHIFT)
2663 mark_page_accessed(pvec.pages[0]);
2665 for (i = 0; i < pagevec_count(&pvec); i++) {
2666 struct page *page = pvec.pages[i];
2667 size_t page_size = thp_size(page);
2668 size_t offset = iocb->ki_pos & (page_size - 1);
2669 size_t bytes = min_t(loff_t, end_offset - iocb->ki_pos,
2670 page_size - offset);
2673 if (end_offset < page_offset(page))
2676 mark_page_accessed(page);
2678 * If users can be writing to this page using arbitrary
2679 * virtual addresses, take care about potential aliasing
2680 * before reading the page on the kernel side.
2682 if (writably_mapped) {
2685 for (j = 0; j < thp_nr_pages(page); j++)
2686 flush_dcache_page(page + j);
2689 copied = copy_page_to_iter(page, offset, bytes, iter);
2691 already_read += copied;
2692 iocb->ki_pos += copied;
2693 ra->prev_pos = iocb->ki_pos;
2695 if (copied < bytes) {
2701 for (i = 0; i < pagevec_count(&pvec); i++)
2702 put_page(pvec.pages[i]);
2703 pagevec_reinit(&pvec);
2704 } while (iov_iter_count(iter) && iocb->ki_pos < isize && !error);
2706 file_accessed(filp);
2708 return already_read ? already_read : error;
2710 EXPORT_SYMBOL_GPL(filemap_read);
2713 * generic_file_read_iter - generic filesystem read routine
2714 * @iocb: kernel I/O control block
2715 * @iter: destination for the data read
2717 * This is the "read_iter()" routine for all filesystems
2718 * that can use the page cache directly.
2720 * The IOCB_NOWAIT flag in iocb->ki_flags indicates that -EAGAIN shall
2721 * be returned when no data can be read without waiting for I/O requests
2722 * to complete; it doesn't prevent readahead.
2724 * The IOCB_NOIO flag in iocb->ki_flags indicates that no new I/O
2725 * requests shall be made for the read or for readahead. When no data
2726 * can be read, -EAGAIN shall be returned. When readahead would be
2727 * triggered, a partial, possibly empty read shall be returned.
2730 * * number of bytes copied, even for partial reads
2731 * * negative error code (or 0 if IOCB_NOIO) if nothing was read
2734 generic_file_read_iter(struct kiocb *iocb, struct iov_iter *iter)
2736 size_t count = iov_iter_count(iter);
2740 return 0; /* skip atime */
2742 if (iocb->ki_flags & IOCB_DIRECT) {
2743 struct file *file = iocb->ki_filp;
2744 struct address_space *mapping = file->f_mapping;
2745 struct inode *inode = mapping->host;
2747 if (iocb->ki_flags & IOCB_NOWAIT) {
2748 if (filemap_range_needs_writeback(mapping, iocb->ki_pos,
2749 iocb->ki_pos + count - 1))
2752 retval = filemap_write_and_wait_range(mapping,
2754 iocb->ki_pos + count - 1);
2759 file_accessed(file);
2761 retval = mapping->a_ops->direct_IO(iocb, iter);
2763 iocb->ki_pos += retval;
2766 if (retval != -EIOCBQUEUED)
2767 iov_iter_revert(iter, count - iov_iter_count(iter));
2770 * Btrfs can have a short DIO read if we encounter
2771 * compressed extents, so if there was an error, or if
2772 * we've already read everything we wanted to, or if
2773 * there was a short read because we hit EOF, go ahead
2774 * and return. Otherwise fallthrough to buffered io for
2775 * the rest of the read. Buffered reads will not work for
2776 * DAX files, so don't bother trying.
2778 if (retval < 0 || !count || IS_DAX(inode))
2780 if (iocb->ki_pos >= i_size_read(inode))
2784 return filemap_read(iocb, iter, retval);
2786 EXPORT_SYMBOL(generic_file_read_iter);
2788 static inline loff_t page_seek_hole_data(struct xa_state *xas,
2789 struct address_space *mapping, struct page *page,
2790 loff_t start, loff_t end, bool seek_data)
2792 const struct address_space_operations *ops = mapping->a_ops;
2793 size_t offset, bsz = i_blocksize(mapping->host);
2795 if (xa_is_value(page) || PageUptodate(page))
2796 return seek_data ? start : end;
2797 if (!ops->is_partially_uptodate)
2798 return seek_data ? end : start;
2803 if (unlikely(page->mapping != mapping))
2806 offset = offset_in_thp(page, start) & ~(bsz - 1);
2809 if (ops->is_partially_uptodate(page, offset, bsz) == seek_data)
2811 start = (start + bsz) & ~(bsz - 1);
2813 } while (offset < thp_size(page));
2821 unsigned int seek_page_size(struct xa_state *xas, struct page *page)
2823 if (xa_is_value(page))
2824 return PAGE_SIZE << xa_get_order(xas->xa, xas->xa_index);
2825 return thp_size(page);
2829 * mapping_seek_hole_data - Seek for SEEK_DATA / SEEK_HOLE in the page cache.
2830 * @mapping: Address space to search.
2831 * @start: First byte to consider.
2832 * @end: Limit of search (exclusive).
2833 * @whence: Either SEEK_HOLE or SEEK_DATA.
2835 * If the page cache knows which blocks contain holes and which blocks
2836 * contain data, your filesystem can use this function to implement
2837 * SEEK_HOLE and SEEK_DATA. This is useful for filesystems which are
2838 * entirely memory-based such as tmpfs, and filesystems which support
2839 * unwritten extents.
2841 * Return: The requested offset on success, or -ENXIO if @whence specifies
2842 * SEEK_DATA and there is no data after @start. There is an implicit hole
2843 * after @end - 1, so SEEK_HOLE returns @end if all the bytes between @start
2844 * and @end contain data.
2846 loff_t mapping_seek_hole_data(struct address_space *mapping, loff_t start,
2847 loff_t end, int whence)
2849 XA_STATE(xas, &mapping->i_pages, start >> PAGE_SHIFT);
2850 pgoff_t max = (end - 1) >> PAGE_SHIFT;
2851 bool seek_data = (whence == SEEK_DATA);
2858 while ((page = find_get_entry(&xas, max, XA_PRESENT))) {
2859 loff_t pos = (u64)xas.xa_index << PAGE_SHIFT;
2860 unsigned int seek_size;
2868 seek_size = seek_page_size(&xas, page);
2869 pos = round_up(pos + 1, seek_size);
2870 start = page_seek_hole_data(&xas, mapping, page, start, pos,
2876 if (seek_size > PAGE_SIZE)
2877 xas_set(&xas, pos >> PAGE_SHIFT);
2878 if (!xa_is_value(page))
2885 if (page && !xa_is_value(page))
2893 #define MMAP_LOTSAMISS (100)
2895 * lock_page_maybe_drop_mmap - lock the page, possibly dropping the mmap_lock
2896 * @vmf - the vm_fault for this fault.
2897 * @page - the page to lock.
2898 * @fpin - the pointer to the file we may pin (or is already pinned).
2900 * This works similar to lock_page_or_retry in that it can drop the mmap_lock.
2901 * It differs in that it actually returns the page locked if it returns 1 and 0
2902 * if it couldn't lock the page. If we did have to drop the mmap_lock then fpin
2903 * will point to the pinned file and needs to be fput()'ed at a later point.
2905 static int lock_page_maybe_drop_mmap(struct vm_fault *vmf, struct page *page,
2908 struct folio *folio = page_folio(page);
2910 if (folio_trylock(folio))
2914 * NOTE! This will make us return with VM_FAULT_RETRY, but with
2915 * the mmap_lock still held. That's how FAULT_FLAG_RETRY_NOWAIT
2916 * is supposed to work. We have way too many special cases..
2918 if (vmf->flags & FAULT_FLAG_RETRY_NOWAIT)
2921 *fpin = maybe_unlock_mmap_for_io(vmf, *fpin);
2922 if (vmf->flags & FAULT_FLAG_KILLABLE) {
2923 if (__folio_lock_killable(folio)) {
2925 * We didn't have the right flags to drop the mmap_lock,
2926 * but all fault_handlers only check for fatal signals
2927 * if we return VM_FAULT_RETRY, so we need to drop the
2928 * mmap_lock here and return 0 if we don't have a fpin.
2931 mmap_read_unlock(vmf->vma->vm_mm);
2935 __folio_lock(folio);
2941 * Synchronous readahead happens when we don't even find a page in the page
2942 * cache at all. We don't want to perform IO under the mmap sem, so if we have
2943 * to drop the mmap sem we return the file that was pinned in order for us to do
2944 * that. If we didn't pin a file then we return NULL. The file that is
2945 * returned needs to be fput()'ed when we're done with it.
2947 static struct file *do_sync_mmap_readahead(struct vm_fault *vmf)
2949 struct file *file = vmf->vma->vm_file;
2950 struct file_ra_state *ra = &file->f_ra;
2951 struct address_space *mapping = file->f_mapping;
2952 DEFINE_READAHEAD(ractl, file, ra, mapping, vmf->pgoff);
2953 struct file *fpin = NULL;
2954 unsigned int mmap_miss;
2956 /* If we don't want any read-ahead, don't bother */
2957 if (vmf->vma->vm_flags & VM_RAND_READ)
2962 if (vmf->vma->vm_flags & VM_SEQ_READ) {
2963 fpin = maybe_unlock_mmap_for_io(vmf, fpin);
2964 page_cache_sync_ra(&ractl, ra->ra_pages);
2968 /* Avoid banging the cache line if not needed */
2969 mmap_miss = READ_ONCE(ra->mmap_miss);
2970 if (mmap_miss < MMAP_LOTSAMISS * 10)
2971 WRITE_ONCE(ra->mmap_miss, ++mmap_miss);
2974 * Do we miss much more than hit in this file? If so,
2975 * stop bothering with read-ahead. It will only hurt.
2977 if (mmap_miss > MMAP_LOTSAMISS)
2983 fpin = maybe_unlock_mmap_for_io(vmf, fpin);
2984 ra->start = max_t(long, 0, vmf->pgoff - ra->ra_pages / 2);
2985 ra->size = ra->ra_pages;
2986 ra->async_size = ra->ra_pages / 4;
2987 ractl._index = ra->start;
2988 do_page_cache_ra(&ractl, ra->size, ra->async_size);
2993 * Asynchronous readahead happens when we find the page and PG_readahead,
2994 * so we want to possibly extend the readahead further. We return the file that
2995 * was pinned if we have to drop the mmap_lock in order to do IO.
2997 static struct file *do_async_mmap_readahead(struct vm_fault *vmf,
3000 struct file *file = vmf->vma->vm_file;
3001 struct file_ra_state *ra = &file->f_ra;
3002 struct address_space *mapping = file->f_mapping;
3003 struct file *fpin = NULL;
3004 unsigned int mmap_miss;
3005 pgoff_t offset = vmf->pgoff;
3007 /* If we don't want any read-ahead, don't bother */
3008 if (vmf->vma->vm_flags & VM_RAND_READ || !ra->ra_pages)
3010 mmap_miss = READ_ONCE(ra->mmap_miss);
3012 WRITE_ONCE(ra->mmap_miss, --mmap_miss);
3013 if (PageReadahead(page)) {
3014 fpin = maybe_unlock_mmap_for_io(vmf, fpin);
3015 page_cache_async_readahead(mapping, ra, file,
3016 page, offset, ra->ra_pages);
3022 * filemap_fault - read in file data for page fault handling
3023 * @vmf: struct vm_fault containing details of the fault
3025 * filemap_fault() is invoked via the vma operations vector for a
3026 * mapped memory region to read in file data during a page fault.
3028 * The goto's are kind of ugly, but this streamlines the normal case of having
3029 * it in the page cache, and handles the special cases reasonably without
3030 * having a lot of duplicated code.
3032 * vma->vm_mm->mmap_lock must be held on entry.
3034 * If our return value has VM_FAULT_RETRY set, it's because the mmap_lock
3035 * may be dropped before doing I/O or by lock_page_maybe_drop_mmap().
3037 * If our return value does not have VM_FAULT_RETRY set, the mmap_lock
3038 * has not been released.
3040 * We never return with VM_FAULT_RETRY and a bit from VM_FAULT_ERROR set.
3042 * Return: bitwise-OR of %VM_FAULT_ codes.
3044 vm_fault_t filemap_fault(struct vm_fault *vmf)
3047 struct file *file = vmf->vma->vm_file;
3048 struct file *fpin = NULL;
3049 struct address_space *mapping = file->f_mapping;
3050 struct inode *inode = mapping->host;
3051 pgoff_t offset = vmf->pgoff;
3055 bool mapping_locked = false;
3057 max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
3058 if (unlikely(offset >= max_off))
3059 return VM_FAULT_SIGBUS;
3062 * Do we have something in the page cache already?
3064 page = find_get_page(mapping, offset);
3067 * We found the page, so try async readahead before waiting for
3070 if (!(vmf->flags & FAULT_FLAG_TRIED))
3071 fpin = do_async_mmap_readahead(vmf, page);
3072 if (unlikely(!PageUptodate(page))) {
3073 filemap_invalidate_lock_shared(mapping);
3074 mapping_locked = true;
3077 /* No page in the page cache at all */
3078 count_vm_event(PGMAJFAULT);
3079 count_memcg_event_mm(vmf->vma->vm_mm, PGMAJFAULT);
3080 ret = VM_FAULT_MAJOR;
3081 fpin = do_sync_mmap_readahead(vmf);
3084 * See comment in filemap_create_page() why we need
3087 if (!mapping_locked) {
3088 filemap_invalidate_lock_shared(mapping);
3089 mapping_locked = true;
3091 page = pagecache_get_page(mapping, offset,
3092 FGP_CREAT|FGP_FOR_MMAP,
3097 filemap_invalidate_unlock_shared(mapping);
3098 return VM_FAULT_OOM;
3102 if (!lock_page_maybe_drop_mmap(vmf, page, &fpin))
3105 /* Did it get truncated? */
3106 if (unlikely(compound_head(page)->mapping != mapping)) {
3111 VM_BUG_ON_PAGE(page_to_pgoff(page) != offset, page);
3114 * We have a locked page in the page cache, now we need to check
3115 * that it's up-to-date. If not, it is going to be due to an error.
3117 if (unlikely(!PageUptodate(page))) {
3119 * The page was in cache and uptodate and now it is not.
3120 * Strange but possible since we didn't hold the page lock all
3121 * the time. Let's drop everything get the invalidate lock and
3124 if (!mapping_locked) {
3129 goto page_not_uptodate;
3133 * We've made it this far and we had to drop our mmap_lock, now is the
3134 * time to return to the upper layer and have it re-find the vma and
3142 filemap_invalidate_unlock_shared(mapping);
3145 * Found the page and have a reference on it.
3146 * We must recheck i_size under page lock.
3148 max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
3149 if (unlikely(offset >= max_off)) {
3152 return VM_FAULT_SIGBUS;
3156 return ret | VM_FAULT_LOCKED;
3160 * Umm, take care of errors if the page isn't up-to-date.
3161 * Try to re-read it _once_. We do this synchronously,
3162 * because there really aren't any performance issues here
3163 * and we need to check for errors.
3165 fpin = maybe_unlock_mmap_for_io(vmf, fpin);
3166 error = filemap_read_page(file, mapping, page);
3171 if (!error || error == AOP_TRUNCATED_PAGE)
3173 filemap_invalidate_unlock_shared(mapping);
3175 return VM_FAULT_SIGBUS;
3179 * We dropped the mmap_lock, we need to return to the fault handler to
3180 * re-find the vma and come back and find our hopefully still populated
3186 filemap_invalidate_unlock_shared(mapping);
3189 return ret | VM_FAULT_RETRY;
3191 EXPORT_SYMBOL(filemap_fault);
3193 static bool filemap_map_pmd(struct vm_fault *vmf, struct page *page)
3195 struct mm_struct *mm = vmf->vma->vm_mm;
3197 /* Huge page is mapped? No need to proceed. */
3198 if (pmd_trans_huge(*vmf->pmd)) {
3204 if (pmd_none(*vmf->pmd) && PageTransHuge(page)) {
3205 vm_fault_t ret = do_set_pmd(vmf, page);
3207 /* The page is mapped successfully, reference consumed. */
3213 if (pmd_none(*vmf->pmd))
3214 pmd_install(mm, vmf->pmd, &vmf->prealloc_pte);
3216 /* See comment in handle_pte_fault() */
3217 if (pmd_devmap_trans_unstable(vmf->pmd)) {
3226 static struct page *next_uptodate_page(struct page *page,
3227 struct address_space *mapping,
3228 struct xa_state *xas, pgoff_t end_pgoff)
3230 unsigned long max_idx;
3235 if (xas_retry(xas, page))
3237 if (xa_is_value(page))
3239 if (PageLocked(page))
3241 if (!page_cache_get_speculative(page))
3243 /* Has the page moved or been split? */
3244 if (unlikely(page != xas_reload(xas)))
3246 if (!PageUptodate(page) || PageReadahead(page))
3248 if (PageHWPoison(page))
3250 if (!trylock_page(page))
3252 if (page->mapping != mapping)
3254 if (!PageUptodate(page))
3256 max_idx = DIV_ROUND_UP(i_size_read(mapping->host), PAGE_SIZE);
3257 if (xas->xa_index >= max_idx)
3264 } while ((page = xas_next_entry(xas, end_pgoff)) != NULL);
3269 static inline struct page *first_map_page(struct address_space *mapping,
3270 struct xa_state *xas,
3273 return next_uptodate_page(xas_find(xas, end_pgoff),
3274 mapping, xas, end_pgoff);
3277 static inline struct page *next_map_page(struct address_space *mapping,
3278 struct xa_state *xas,
3281 return next_uptodate_page(xas_next_entry(xas, end_pgoff),
3282 mapping, xas, end_pgoff);
3285 vm_fault_t filemap_map_pages(struct vm_fault *vmf,
3286 pgoff_t start_pgoff, pgoff_t end_pgoff)
3288 struct vm_area_struct *vma = vmf->vma;
3289 struct file *file = vma->vm_file;
3290 struct address_space *mapping = file->f_mapping;
3291 pgoff_t last_pgoff = start_pgoff;
3293 XA_STATE(xas, &mapping->i_pages, start_pgoff);
3294 struct page *head, *page;
3295 unsigned int mmap_miss = READ_ONCE(file->f_ra.mmap_miss);
3299 head = first_map_page(mapping, &xas, end_pgoff);
3303 if (filemap_map_pmd(vmf, head)) {
3304 ret = VM_FAULT_NOPAGE;
3308 addr = vma->vm_start + ((start_pgoff - vma->vm_pgoff) << PAGE_SHIFT);
3309 vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, addr, &vmf->ptl);
3311 page = find_subpage(head, xas.xa_index);
3312 if (PageHWPoison(page))
3318 addr += (xas.xa_index - last_pgoff) << PAGE_SHIFT;
3319 vmf->pte += xas.xa_index - last_pgoff;
3320 last_pgoff = xas.xa_index;
3322 if (!pte_none(*vmf->pte))
3325 /* We're about to handle the fault */
3326 if (vmf->address == addr)
3327 ret = VM_FAULT_NOPAGE;
3329 do_set_pte(vmf, page, addr);
3330 /* no need to invalidate: a not-present page won't be cached */
3331 update_mmu_cache(vma, addr, vmf->pte);
3337 } while ((head = next_map_page(mapping, &xas, end_pgoff)) != NULL);
3338 pte_unmap_unlock(vmf->pte, vmf->ptl);
3341 WRITE_ONCE(file->f_ra.mmap_miss, mmap_miss);
3344 EXPORT_SYMBOL(filemap_map_pages);
3346 vm_fault_t filemap_page_mkwrite(struct vm_fault *vmf)
3348 struct address_space *mapping = vmf->vma->vm_file->f_mapping;
3349 struct page *page = vmf->page;
3350 vm_fault_t ret = VM_FAULT_LOCKED;
3352 sb_start_pagefault(mapping->host->i_sb);
3353 file_update_time(vmf->vma->vm_file);
3355 if (page->mapping != mapping) {
3357 ret = VM_FAULT_NOPAGE;
3361 * We mark the page dirty already here so that when freeze is in
3362 * progress, we are guaranteed that writeback during freezing will
3363 * see the dirty page and writeprotect it again.
3365 set_page_dirty(page);
3366 wait_for_stable_page(page);
3368 sb_end_pagefault(mapping->host->i_sb);
3372 const struct vm_operations_struct generic_file_vm_ops = {
3373 .fault = filemap_fault,
3374 .map_pages = filemap_map_pages,
3375 .page_mkwrite = filemap_page_mkwrite,
3378 /* This is used for a general mmap of a disk file */
3380 int generic_file_mmap(struct file *file, struct vm_area_struct *vma)
3382 struct address_space *mapping = file->f_mapping;
3384 if (!mapping->a_ops->readpage)
3386 file_accessed(file);
3387 vma->vm_ops = &generic_file_vm_ops;
3392 * This is for filesystems which do not implement ->writepage.
3394 int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
3396 if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE))
3398 return generic_file_mmap(file, vma);
3401 vm_fault_t filemap_page_mkwrite(struct vm_fault *vmf)
3403 return VM_FAULT_SIGBUS;
3405 int generic_file_mmap(struct file *file, struct vm_area_struct *vma)
3409 int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
3413 #endif /* CONFIG_MMU */
3415 EXPORT_SYMBOL(filemap_page_mkwrite);
3416 EXPORT_SYMBOL(generic_file_mmap);
3417 EXPORT_SYMBOL(generic_file_readonly_mmap);
3419 static struct page *wait_on_page_read(struct page *page)
3421 if (!IS_ERR(page)) {
3422 wait_on_page_locked(page);
3423 if (!PageUptodate(page)) {
3425 page = ERR_PTR(-EIO);
3431 static struct page *do_read_cache_page(struct address_space *mapping,
3433 int (*filler)(void *, struct page *),
3440 page = find_get_page(mapping, index);
3442 page = __page_cache_alloc(gfp);
3444 return ERR_PTR(-ENOMEM);
3445 err = add_to_page_cache_lru(page, mapping, index, gfp);
3446 if (unlikely(err)) {
3450 /* Presumably ENOMEM for xarray node */
3451 return ERR_PTR(err);
3456 err = filler(data, page);
3458 err = mapping->a_ops->readpage(data, page);
3462 return ERR_PTR(err);
3465 page = wait_on_page_read(page);
3470 if (PageUptodate(page))
3474 * Page is not up to date and may be locked due to one of the following
3475 * case a: Page is being filled and the page lock is held
3476 * case b: Read/write error clearing the page uptodate status
3477 * case c: Truncation in progress (page locked)
3478 * case d: Reclaim in progress
3480 * Case a, the page will be up to date when the page is unlocked.
3481 * There is no need to serialise on the page lock here as the page
3482 * is pinned so the lock gives no additional protection. Even if the
3483 * page is truncated, the data is still valid if PageUptodate as
3484 * it's a race vs truncate race.
3485 * Case b, the page will not be up to date
3486 * Case c, the page may be truncated but in itself, the data may still
3487 * be valid after IO completes as it's a read vs truncate race. The
3488 * operation must restart if the page is not uptodate on unlock but
3489 * otherwise serialising on page lock to stabilise the mapping gives
3490 * no additional guarantees to the caller as the page lock is
3491 * released before return.
3492 * Case d, similar to truncation. If reclaim holds the page lock, it
3493 * will be a race with remove_mapping that determines if the mapping
3494 * is valid on unlock but otherwise the data is valid and there is
3495 * no need to serialise with page lock.
3497 * As the page lock gives no additional guarantee, we optimistically
3498 * wait on the page to be unlocked and check if it's up to date and
3499 * use the page if it is. Otherwise, the page lock is required to
3500 * distinguish between the different cases. The motivation is that we
3501 * avoid spurious serialisations and wakeups when multiple processes
3502 * wait on the same page for IO to complete.
3504 wait_on_page_locked(page);
3505 if (PageUptodate(page))
3508 /* Distinguish between all the cases under the safety of the lock */
3511 /* Case c or d, restart the operation */
3512 if (!page->mapping) {
3518 /* Someone else locked and filled the page in a very small window */
3519 if (PageUptodate(page)) {
3525 * A previous I/O error may have been due to temporary
3527 * Clear page error before actual read, PG_error will be
3528 * set again if read page fails.
3530 ClearPageError(page);
3534 mark_page_accessed(page);
3539 * read_cache_page - read into page cache, fill it if needed
3540 * @mapping: the page's address_space
3541 * @index: the page index
3542 * @filler: function to perform the read
3543 * @data: first arg to filler(data, page) function, often left as NULL
3545 * Read into the page cache. If a page already exists, and PageUptodate() is
3546 * not set, try to fill the page and wait for it to become unlocked.
3548 * If the page does not get brought uptodate, return -EIO.
3550 * The function expects mapping->invalidate_lock to be already held.
3552 * Return: up to date page on success, ERR_PTR() on failure.
3554 struct page *read_cache_page(struct address_space *mapping,
3556 int (*filler)(void *, struct page *),
3559 return do_read_cache_page(mapping, index, filler, data,
3560 mapping_gfp_mask(mapping));
3562 EXPORT_SYMBOL(read_cache_page);
3565 * read_cache_page_gfp - read into page cache, using specified page allocation flags.
3566 * @mapping: the page's address_space
3567 * @index: the page index
3568 * @gfp: the page allocator flags to use if allocating
3570 * This is the same as "read_mapping_page(mapping, index, NULL)", but with
3571 * any new page allocations done using the specified allocation flags.
3573 * If the page does not get brought uptodate, return -EIO.
3575 * The function expects mapping->invalidate_lock to be already held.
3577 * Return: up to date page on success, ERR_PTR() on failure.
3579 struct page *read_cache_page_gfp(struct address_space *mapping,
3583 return do_read_cache_page(mapping, index, NULL, NULL, gfp);
3585 EXPORT_SYMBOL(read_cache_page_gfp);
3587 int pagecache_write_begin(struct file *file, struct address_space *mapping,
3588 loff_t pos, unsigned len, unsigned flags,
3589 struct page **pagep, void **fsdata)
3591 const struct address_space_operations *aops = mapping->a_ops;
3593 return aops->write_begin(file, mapping, pos, len, flags,
3596 EXPORT_SYMBOL(pagecache_write_begin);
3598 int pagecache_write_end(struct file *file, struct address_space *mapping,
3599 loff_t pos, unsigned len, unsigned copied,
3600 struct page *page, void *fsdata)
3602 const struct address_space_operations *aops = mapping->a_ops;
3604 return aops->write_end(file, mapping, pos, len, copied, page, fsdata);
3606 EXPORT_SYMBOL(pagecache_write_end);
3609 * Warn about a page cache invalidation failure during a direct I/O write.
3611 void dio_warn_stale_pagecache(struct file *filp)
3613 static DEFINE_RATELIMIT_STATE(_rs, 86400 * HZ, DEFAULT_RATELIMIT_BURST);
3617 errseq_set(&filp->f_mapping->wb_err, -EIO);
3618 if (__ratelimit(&_rs)) {
3619 path = file_path(filp, pathname, sizeof(pathname));
3622 pr_crit("Page cache invalidation failure on direct I/O. Possible data corruption due to collision with buffered I/O!\n");
3623 pr_crit("File: %s PID: %d Comm: %.20s\n", path, current->pid,
3629 generic_file_direct_write(struct kiocb *iocb, struct iov_iter *from)
3631 struct file *file = iocb->ki_filp;
3632 struct address_space *mapping = file->f_mapping;
3633 struct inode *inode = mapping->host;
3634 loff_t pos = iocb->ki_pos;
3639 write_len = iov_iter_count(from);
3640 end = (pos + write_len - 1) >> PAGE_SHIFT;
3642 if (iocb->ki_flags & IOCB_NOWAIT) {
3643 /* If there are pages to writeback, return */
3644 if (filemap_range_has_page(file->f_mapping, pos,
3645 pos + write_len - 1))
3648 written = filemap_write_and_wait_range(mapping, pos,
3649 pos + write_len - 1);
3655 * After a write we want buffered reads to be sure to go to disk to get
3656 * the new data. We invalidate clean cached page from the region we're
3657 * about to write. We do this *before* the write so that we can return
3658 * without clobbering -EIOCBQUEUED from ->direct_IO().
3660 written = invalidate_inode_pages2_range(mapping,
3661 pos >> PAGE_SHIFT, end);
3663 * If a page can not be invalidated, return 0 to fall back
3664 * to buffered write.
3667 if (written == -EBUSY)
3672 written = mapping->a_ops->direct_IO(iocb, from);
3675 * Finally, try again to invalidate clean pages which might have been
3676 * cached by non-direct readahead, or faulted in by get_user_pages()
3677 * if the source of the write was an mmap'ed region of the file
3678 * we're writing. Either one is a pretty crazy thing to do,
3679 * so we don't support it 100%. If this invalidation
3680 * fails, tough, the write still worked...
3682 * Most of the time we do not need this since dio_complete() will do
3683 * the invalidation for us. However there are some file systems that
3684 * do not end up with dio_complete() being called, so let's not break
3685 * them by removing it completely.
3687 * Noticeable example is a blkdev_direct_IO().
3689 * Skip invalidation for async writes or if mapping has no pages.
3691 if (written > 0 && mapping->nrpages &&
3692 invalidate_inode_pages2_range(mapping, pos >> PAGE_SHIFT, end))
3693 dio_warn_stale_pagecache(file);
3697 write_len -= written;
3698 if (pos > i_size_read(inode) && !S_ISBLK(inode->i_mode)) {
3699 i_size_write(inode, pos);
3700 mark_inode_dirty(inode);
3704 if (written != -EIOCBQUEUED)
3705 iov_iter_revert(from, write_len - iov_iter_count(from));
3709 EXPORT_SYMBOL(generic_file_direct_write);
3711 ssize_t generic_perform_write(struct file *file,
3712 struct iov_iter *i, loff_t pos)
3714 struct address_space *mapping = file->f_mapping;
3715 const struct address_space_operations *a_ops = mapping->a_ops;
3717 ssize_t written = 0;
3718 unsigned int flags = 0;
3722 unsigned long offset; /* Offset into pagecache page */
3723 unsigned long bytes; /* Bytes to write to page */
3724 size_t copied; /* Bytes copied from user */
3727 offset = (pos & (PAGE_SIZE - 1));
3728 bytes = min_t(unsigned long, PAGE_SIZE - offset,
3733 * Bring in the user page that we will copy from _first_.
3734 * Otherwise there's a nasty deadlock on copying from the
3735 * same page as we're writing to, without it being marked
3738 if (unlikely(fault_in_iov_iter_readable(i, bytes))) {
3743 if (fatal_signal_pending(current)) {
3748 status = a_ops->write_begin(file, mapping, pos, bytes, flags,
3750 if (unlikely(status < 0))
3753 if (mapping_writably_mapped(mapping))
3754 flush_dcache_page(page);
3756 copied = copy_page_from_iter_atomic(page, offset, bytes, i);
3757 flush_dcache_page(page);
3759 status = a_ops->write_end(file, mapping, pos, bytes, copied,
3761 if (unlikely(status != copied)) {
3762 iov_iter_revert(i, copied - max(status, 0L));
3763 if (unlikely(status < 0))
3768 if (unlikely(status == 0)) {
3770 * A short copy made ->write_end() reject the
3771 * thing entirely. Might be memory poisoning
3772 * halfway through, might be a race with munmap,
3773 * might be severe memory pressure.
3782 balance_dirty_pages_ratelimited(mapping);
3783 } while (iov_iter_count(i));
3785 return written ? written : status;
3787 EXPORT_SYMBOL(generic_perform_write);
3790 * __generic_file_write_iter - write data to a file
3791 * @iocb: IO state structure (file, offset, etc.)
3792 * @from: iov_iter with data to write
3794 * This function does all the work needed for actually writing data to a
3795 * file. It does all basic checks, removes SUID from the file, updates
3796 * modification times and calls proper subroutines depending on whether we
3797 * do direct IO or a standard buffered write.
3799 * It expects i_rwsem to be grabbed unless we work on a block device or similar
3800 * object which does not need locking at all.
3802 * This function does *not* take care of syncing data in case of O_SYNC write.
3803 * A caller has to handle it. This is mainly due to the fact that we want to
3804 * avoid syncing under i_rwsem.
3807 * * number of bytes written, even for truncated writes
3808 * * negative error code if no data has been written at all
3810 ssize_t __generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
3812 struct file *file = iocb->ki_filp;
3813 struct address_space *mapping = file->f_mapping;
3814 struct inode *inode = mapping->host;
3815 ssize_t written = 0;
3819 /* We can write back this queue in page reclaim */
3820 current->backing_dev_info = inode_to_bdi(inode);
3821 err = file_remove_privs(file);
3825 err = file_update_time(file);
3829 if (iocb->ki_flags & IOCB_DIRECT) {
3830 loff_t pos, endbyte;
3832 written = generic_file_direct_write(iocb, from);
3834 * If the write stopped short of completing, fall back to
3835 * buffered writes. Some filesystems do this for writes to
3836 * holes, for example. For DAX files, a buffered write will
3837 * not succeed (even if it did, DAX does not handle dirty
3838 * page-cache pages correctly).
3840 if (written < 0 || !iov_iter_count(from) || IS_DAX(inode))
3843 status = generic_perform_write(file, from, pos = iocb->ki_pos);
3845 * If generic_perform_write() returned a synchronous error
3846 * then we want to return the number of bytes which were
3847 * direct-written, or the error code if that was zero. Note
3848 * that this differs from normal direct-io semantics, which
3849 * will return -EFOO even if some bytes were written.
3851 if (unlikely(status < 0)) {
3856 * We need to ensure that the page cache pages are written to
3857 * disk and invalidated to preserve the expected O_DIRECT
3860 endbyte = pos + status - 1;
3861 err = filemap_write_and_wait_range(mapping, pos, endbyte);
3863 iocb->ki_pos = endbyte + 1;
3865 invalidate_mapping_pages(mapping,
3867 endbyte >> PAGE_SHIFT);
3870 * We don't know how much we wrote, so just return
3871 * the number of bytes which were direct-written
3875 written = generic_perform_write(file, from, iocb->ki_pos);
3876 if (likely(written > 0))
3877 iocb->ki_pos += written;
3880 current->backing_dev_info = NULL;
3881 return written ? written : err;
3883 EXPORT_SYMBOL(__generic_file_write_iter);
3886 * generic_file_write_iter - write data to a file
3887 * @iocb: IO state structure
3888 * @from: iov_iter with data to write
3890 * This is a wrapper around __generic_file_write_iter() to be used by most
3891 * filesystems. It takes care of syncing the file in case of O_SYNC file
3892 * and acquires i_rwsem as needed.
3894 * * negative error code if no data has been written at all of
3895 * vfs_fsync_range() failed for a synchronous write
3896 * * number of bytes written, even for truncated writes
3898 ssize_t generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
3900 struct file *file = iocb->ki_filp;
3901 struct inode *inode = file->f_mapping->host;
3905 ret = generic_write_checks(iocb, from);
3907 ret = __generic_file_write_iter(iocb, from);
3908 inode_unlock(inode);
3911 ret = generic_write_sync(iocb, ret);
3914 EXPORT_SYMBOL(generic_file_write_iter);
3917 * try_to_release_page() - release old fs-specific metadata on a page
3919 * @page: the page which the kernel is trying to free
3920 * @gfp_mask: memory allocation flags (and I/O mode)
3922 * The address_space is to try to release any data against the page
3923 * (presumably at page->private).
3925 * This may also be called if PG_fscache is set on a page, indicating that the
3926 * page is known to the local caching routines.
3928 * The @gfp_mask argument specifies whether I/O may be performed to release
3929 * this page (__GFP_IO), and whether the call may block (__GFP_RECLAIM & __GFP_FS).
3931 * Return: %1 if the release was successful, otherwise return zero.
3933 int try_to_release_page(struct page *page, gfp_t gfp_mask)
3935 struct address_space * const mapping = page->mapping;
3937 BUG_ON(!PageLocked(page));
3938 if (PageWriteback(page))
3941 if (mapping && mapping->a_ops->releasepage)
3942 return mapping->a_ops->releasepage(page, gfp_mask);
3943 return try_to_free_buffers(page);
3946 EXPORT_SYMBOL(try_to_release_page);