1 // SPDX-License-Identifier: GPL-2.0-only
5 * Copyright (C) 1994-1999 Linus Torvalds
9 * This file handles the generic file mmap semantics used by
10 * most "normal" filesystems (but you don't /have/ to use this:
11 * the NFS filesystem used to do this differently, for example)
13 #include <linux/export.h>
14 #include <linux/compiler.h>
15 #include <linux/dax.h>
17 #include <linux/sched/signal.h>
18 #include <linux/uaccess.h>
19 #include <linux/capability.h>
20 #include <linux/kernel_stat.h>
21 #include <linux/gfp.h>
23 #include <linux/swap.h>
24 #include <linux/mman.h>
25 #include <linux/pagemap.h>
26 #include <linux/file.h>
27 #include <linux/uio.h>
28 #include <linux/error-injection.h>
29 #include <linux/hash.h>
30 #include <linux/writeback.h>
31 #include <linux/backing-dev.h>
32 #include <linux/pagevec.h>
33 #include <linux/blkdev.h>
34 #include <linux/security.h>
35 #include <linux/cpuset.h>
36 #include <linux/hugetlb.h>
37 #include <linux/memcontrol.h>
38 #include <linux/cleancache.h>
39 #include <linux/shmem_fs.h>
40 #include <linux/rmap.h>
41 #include <linux/delayacct.h>
42 #include <linux/psi.h>
43 #include <linux/ramfs.h>
44 #include <linux/page_idle.h>
45 #include <asm/pgalloc.h>
46 #include <asm/tlbflush.h>
49 #define CREATE_TRACE_POINTS
50 #include <trace/events/filemap.h>
53 * FIXME: remove all knowledge of the buffer layer from the core VM
55 #include <linux/buffer_head.h> /* for try_to_free_buffers */
60 * Shared mappings implemented 30.11.1994. It's not fully working yet,
63 * Shared mappings now work. 15.8.1995 Bruno.
65 * finished 'unifying' the page and buffer cache and SMP-threaded the
66 * page-cache, 21.05.1999, Ingo Molnar <mingo@redhat.com>
68 * SMP-threaded pagemap-LRU 1999, Andrea Arcangeli <andrea@suse.de>
74 * ->i_mmap_rwsem (truncate_pagecache)
75 * ->private_lock (__free_pte->__set_page_dirty_buffers)
76 * ->swap_lock (exclusive_swap_page, others)
80 * ->invalidate_lock (acquired by fs in truncate path)
81 * ->i_mmap_rwsem (truncate->unmap_mapping_range)
85 * ->page_table_lock or pte_lock (various, mainly in memory.c)
86 * ->i_pages lock (arch-dependent flush_dcache_mmap_lock)
89 * ->invalidate_lock (filemap_fault)
90 * ->lock_page (filemap_fault, access_process_vm)
92 * ->i_rwsem (generic_perform_write)
93 * ->mmap_lock (fault_in_pages_readable->do_page_fault)
96 * sb_lock (fs/fs-writeback.c)
97 * ->i_pages lock (__sync_single_inode)
100 * ->anon_vma.lock (vma_adjust)
103 * ->page_table_lock or pte_lock (anon_vma_prepare and various)
105 * ->page_table_lock or pte_lock
106 * ->swap_lock (try_to_unmap_one)
107 * ->private_lock (try_to_unmap_one)
108 * ->i_pages lock (try_to_unmap_one)
109 * ->lruvec->lru_lock (follow_page->mark_page_accessed)
110 * ->lruvec->lru_lock (check_pte_range->isolate_lru_page)
111 * ->private_lock (page_remove_rmap->set_page_dirty)
112 * ->i_pages lock (page_remove_rmap->set_page_dirty)
113 * bdi.wb->list_lock (page_remove_rmap->set_page_dirty)
114 * ->inode->i_lock (page_remove_rmap->set_page_dirty)
115 * ->memcg->move_lock (page_remove_rmap->lock_page_memcg)
116 * bdi.wb->list_lock (zap_pte_range->set_page_dirty)
117 * ->inode->i_lock (zap_pte_range->set_page_dirty)
118 * ->private_lock (zap_pte_range->__set_page_dirty_buffers)
121 * ->tasklist_lock (memory_failure, collect_procs_ao)
124 static void page_cache_delete(struct address_space *mapping,
125 struct page *page, void *shadow)
127 XA_STATE(xas, &mapping->i_pages, page->index);
130 mapping_set_update(&xas, mapping);
132 /* hugetlb pages are represented by a single entry in the xarray */
133 if (!PageHuge(page)) {
134 xas_set_order(&xas, page->index, compound_order(page));
135 nr = compound_nr(page);
138 VM_BUG_ON_PAGE(!PageLocked(page), page);
139 VM_BUG_ON_PAGE(PageTail(page), page);
140 VM_BUG_ON_PAGE(nr != 1 && shadow, page);
142 xas_store(&xas, shadow);
143 xas_init_marks(&xas);
145 page->mapping = NULL;
146 /* Leave page->index set: truncation lookup relies upon it */
147 mapping->nrpages -= nr;
150 static void unaccount_page_cache_page(struct address_space *mapping,
156 * if we're uptodate, flush out into the cleancache, otherwise
157 * invalidate any existing cleancache entries. We can't leave
158 * stale data around in the cleancache once our page is gone
160 if (PageUptodate(page) && PageMappedToDisk(page))
161 cleancache_put_page(page);
163 cleancache_invalidate_page(mapping, page);
165 VM_BUG_ON_PAGE(PageTail(page), page);
166 VM_BUG_ON_PAGE(page_mapped(page), page);
167 if (!IS_ENABLED(CONFIG_DEBUG_VM) && unlikely(page_mapped(page))) {
170 pr_alert("BUG: Bad page cache in process %s pfn:%05lx\n",
171 current->comm, page_to_pfn(page));
172 dump_page(page, "still mapped when deleted");
174 add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE);
176 mapcount = page_mapcount(page);
177 if (mapping_exiting(mapping) &&
178 page_count(page) >= mapcount + 2) {
180 * All vmas have already been torn down, so it's
181 * a good bet that actually the page is unmapped,
182 * and we'd prefer not to leak it: if we're wrong,
183 * some other bad page check should catch it later.
185 page_mapcount_reset(page);
186 page_ref_sub(page, mapcount);
190 /* hugetlb pages do not participate in page cache accounting. */
194 nr = thp_nr_pages(page);
196 __mod_lruvec_page_state(page, NR_FILE_PAGES, -nr);
197 if (PageSwapBacked(page)) {
198 __mod_lruvec_page_state(page, NR_SHMEM, -nr);
199 if (PageTransHuge(page))
200 __mod_lruvec_page_state(page, NR_SHMEM_THPS, -nr);
201 } else if (PageTransHuge(page)) {
202 __mod_lruvec_page_state(page, NR_FILE_THPS, -nr);
203 filemap_nr_thps_dec(mapping);
207 * At this point page must be either written or cleaned by
208 * truncate. Dirty page here signals a bug and loss of
211 * This fixes dirty accounting after removing the page entirely
212 * but leaves PageDirty set: it has no effect for truncated
213 * page and anyway will be cleared before returning page into
216 if (WARN_ON_ONCE(PageDirty(page)))
217 account_page_cleaned(page, mapping, inode_to_wb(mapping->host));
221 * Delete a page from the page cache and free it. Caller has to make
222 * sure the page is locked and that nobody else uses it - or that usage
223 * is safe. The caller must hold the i_pages lock.
225 void __delete_from_page_cache(struct page *page, void *shadow)
227 struct address_space *mapping = page->mapping;
229 trace_mm_filemap_delete_from_page_cache(page);
231 unaccount_page_cache_page(mapping, page);
232 page_cache_delete(mapping, page, shadow);
235 static void page_cache_free_page(struct address_space *mapping,
238 void (*freepage)(struct page *);
240 freepage = mapping->a_ops->freepage;
244 if (PageTransHuge(page) && !PageHuge(page)) {
245 page_ref_sub(page, thp_nr_pages(page));
246 VM_BUG_ON_PAGE(page_count(page) <= 0, page);
253 * delete_from_page_cache - delete page from page cache
254 * @page: the page which the kernel is trying to remove from page cache
256 * This must be called only on pages that have been verified to be in the page
257 * cache and locked. It will never put the page into the free list, the caller
258 * has a reference on the page.
260 void delete_from_page_cache(struct page *page)
262 struct address_space *mapping = page_mapping(page);
264 BUG_ON(!PageLocked(page));
265 xa_lock_irq(&mapping->i_pages);
266 __delete_from_page_cache(page, NULL);
267 xa_unlock_irq(&mapping->i_pages);
269 page_cache_free_page(mapping, page);
271 EXPORT_SYMBOL(delete_from_page_cache);
274 * page_cache_delete_batch - delete several pages from page cache
275 * @mapping: the mapping to which pages belong
276 * @pvec: pagevec with pages to delete
278 * The function walks over mapping->i_pages and removes pages passed in @pvec
279 * from the mapping. The function expects @pvec to be sorted by page index
280 * and is optimised for it to be dense.
281 * It tolerates holes in @pvec (mapping entries at those indices are not
282 * modified). The function expects only THP head pages to be present in the
285 * The function expects the i_pages lock to be held.
287 static void page_cache_delete_batch(struct address_space *mapping,
288 struct pagevec *pvec)
290 XA_STATE(xas, &mapping->i_pages, pvec->pages[0]->index);
295 mapping_set_update(&xas, mapping);
296 xas_for_each(&xas, page, ULONG_MAX) {
297 if (i >= pagevec_count(pvec))
300 /* A swap/dax/shadow entry got inserted? Skip it. */
301 if (xa_is_value(page))
304 * A page got inserted in our range? Skip it. We have our
305 * pages locked so they are protected from being removed.
306 * If we see a page whose index is higher than ours, it
307 * means our page has been removed, which shouldn't be
308 * possible because we're holding the PageLock.
310 if (page != pvec->pages[i]) {
311 VM_BUG_ON_PAGE(page->index > pvec->pages[i]->index,
316 WARN_ON_ONCE(!PageLocked(page));
318 if (page->index == xas.xa_index)
319 page->mapping = NULL;
320 /* Leave page->index set: truncation lookup relies on it */
323 * Move to the next page in the vector if this is a regular
324 * page or the index is of the last sub-page of this compound
327 if (page->index + compound_nr(page) - 1 == xas.xa_index)
329 xas_store(&xas, NULL);
332 mapping->nrpages -= total_pages;
335 void delete_from_page_cache_batch(struct address_space *mapping,
336 struct pagevec *pvec)
340 if (!pagevec_count(pvec))
343 xa_lock_irq(&mapping->i_pages);
344 for (i = 0; i < pagevec_count(pvec); i++) {
345 trace_mm_filemap_delete_from_page_cache(pvec->pages[i]);
347 unaccount_page_cache_page(mapping, pvec->pages[i]);
349 page_cache_delete_batch(mapping, pvec);
350 xa_unlock_irq(&mapping->i_pages);
352 for (i = 0; i < pagevec_count(pvec); i++)
353 page_cache_free_page(mapping, pvec->pages[i]);
356 int filemap_check_errors(struct address_space *mapping)
359 /* Check for outstanding write errors */
360 if (test_bit(AS_ENOSPC, &mapping->flags) &&
361 test_and_clear_bit(AS_ENOSPC, &mapping->flags))
363 if (test_bit(AS_EIO, &mapping->flags) &&
364 test_and_clear_bit(AS_EIO, &mapping->flags))
368 EXPORT_SYMBOL(filemap_check_errors);
370 static int filemap_check_and_keep_errors(struct address_space *mapping)
372 /* Check for outstanding write errors */
373 if (test_bit(AS_EIO, &mapping->flags))
375 if (test_bit(AS_ENOSPC, &mapping->flags))
381 * filemap_fdatawrite_wbc - start writeback on mapping dirty pages in range
382 * @mapping: address space structure to write
383 * @wbc: the writeback_control controlling the writeout
385 * Call writepages on the mapping using the provided wbc to control the
388 * Return: %0 on success, negative error code otherwise.
390 int filemap_fdatawrite_wbc(struct address_space *mapping,
391 struct writeback_control *wbc)
395 if (!mapping_can_writeback(mapping) ||
396 !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
399 wbc_attach_fdatawrite_inode(wbc, mapping->host);
400 ret = do_writepages(mapping, wbc);
401 wbc_detach_inode(wbc);
404 EXPORT_SYMBOL(filemap_fdatawrite_wbc);
407 * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range
408 * @mapping: address space structure to write
409 * @start: offset in bytes where the range starts
410 * @end: offset in bytes where the range ends (inclusive)
411 * @sync_mode: enable synchronous operation
413 * Start writeback against all of a mapping's dirty pages that lie
414 * within the byte offsets <start, end> inclusive.
416 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
417 * opposed to a regular memory cleansing writeback. The difference between
418 * these two operations is that if a dirty page/buffer is encountered, it must
419 * be waited upon, and not just skipped over.
421 * Return: %0 on success, negative error code otherwise.
423 int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
424 loff_t end, int sync_mode)
426 struct writeback_control wbc = {
427 .sync_mode = sync_mode,
428 .nr_to_write = LONG_MAX,
429 .range_start = start,
433 return filemap_fdatawrite_wbc(mapping, &wbc);
436 static inline int __filemap_fdatawrite(struct address_space *mapping,
439 return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode);
442 int filemap_fdatawrite(struct address_space *mapping)
444 return __filemap_fdatawrite(mapping, WB_SYNC_ALL);
446 EXPORT_SYMBOL(filemap_fdatawrite);
448 int filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
451 return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL);
453 EXPORT_SYMBOL(filemap_fdatawrite_range);
456 * filemap_flush - mostly a non-blocking flush
457 * @mapping: target address_space
459 * This is a mostly non-blocking flush. Not suitable for data-integrity
460 * purposes - I/O may not be started against all dirty pages.
462 * Return: %0 on success, negative error code otherwise.
464 int filemap_flush(struct address_space *mapping)
466 return __filemap_fdatawrite(mapping, WB_SYNC_NONE);
468 EXPORT_SYMBOL(filemap_flush);
471 * filemap_range_has_page - check if a page exists in range.
472 * @mapping: address space within which to check
473 * @start_byte: offset in bytes where the range starts
474 * @end_byte: offset in bytes where the range ends (inclusive)
476 * Find at least one page in the range supplied, usually used to check if
477 * direct writing in this range will trigger a writeback.
479 * Return: %true if at least one page exists in the specified range,
482 bool filemap_range_has_page(struct address_space *mapping,
483 loff_t start_byte, loff_t end_byte)
486 XA_STATE(xas, &mapping->i_pages, start_byte >> PAGE_SHIFT);
487 pgoff_t max = end_byte >> PAGE_SHIFT;
489 if (end_byte < start_byte)
494 page = xas_find(&xas, max);
495 if (xas_retry(&xas, page))
497 /* Shadow entries don't count */
498 if (xa_is_value(page))
501 * We don't need to try to pin this page; we're about to
502 * release the RCU lock anyway. It is enough to know that
503 * there was a page here recently.
511 EXPORT_SYMBOL(filemap_range_has_page);
513 static void __filemap_fdatawait_range(struct address_space *mapping,
514 loff_t start_byte, loff_t end_byte)
516 pgoff_t index = start_byte >> PAGE_SHIFT;
517 pgoff_t end = end_byte >> PAGE_SHIFT;
521 if (end_byte < start_byte)
525 while (index <= end) {
528 nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index,
529 end, PAGECACHE_TAG_WRITEBACK);
533 for (i = 0; i < nr_pages; i++) {
534 struct page *page = pvec.pages[i];
536 wait_on_page_writeback(page);
537 ClearPageError(page);
539 pagevec_release(&pvec);
545 * filemap_fdatawait_range - wait for writeback to complete
546 * @mapping: address space structure to wait for
547 * @start_byte: offset in bytes where the range starts
548 * @end_byte: offset in bytes where the range ends (inclusive)
550 * Walk the list of under-writeback pages of the given address space
551 * in the given range and wait for all of them. Check error status of
552 * the address space and return it.
554 * Since the error status of the address space is cleared by this function,
555 * callers are responsible for checking the return value and handling and/or
556 * reporting the error.
558 * Return: error status of the address space.
560 int filemap_fdatawait_range(struct address_space *mapping, loff_t start_byte,
563 __filemap_fdatawait_range(mapping, start_byte, end_byte);
564 return filemap_check_errors(mapping);
566 EXPORT_SYMBOL(filemap_fdatawait_range);
569 * filemap_fdatawait_range_keep_errors - wait for writeback to complete
570 * @mapping: address space structure to wait for
571 * @start_byte: offset in bytes where the range starts
572 * @end_byte: offset in bytes where the range ends (inclusive)
574 * Walk the list of under-writeback pages of the given address space in the
575 * given range and wait for all of them. Unlike filemap_fdatawait_range(),
576 * this function does not clear error status of the address space.
578 * Use this function if callers don't handle errors themselves. Expected
579 * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2),
582 int filemap_fdatawait_range_keep_errors(struct address_space *mapping,
583 loff_t start_byte, loff_t end_byte)
585 __filemap_fdatawait_range(mapping, start_byte, end_byte);
586 return filemap_check_and_keep_errors(mapping);
588 EXPORT_SYMBOL(filemap_fdatawait_range_keep_errors);
591 * file_fdatawait_range - wait for writeback to complete
592 * @file: file pointing to address space structure to wait for
593 * @start_byte: offset in bytes where the range starts
594 * @end_byte: offset in bytes where the range ends (inclusive)
596 * Walk the list of under-writeback pages of the address space that file
597 * refers to, in the given range and wait for all of them. Check error
598 * status of the address space vs. the file->f_wb_err cursor and return it.
600 * Since the error status of the file is advanced by this function,
601 * callers are responsible for checking the return value and handling and/or
602 * reporting the error.
604 * Return: error status of the address space vs. the file->f_wb_err cursor.
606 int file_fdatawait_range(struct file *file, loff_t start_byte, loff_t end_byte)
608 struct address_space *mapping = file->f_mapping;
610 __filemap_fdatawait_range(mapping, start_byte, end_byte);
611 return file_check_and_advance_wb_err(file);
613 EXPORT_SYMBOL(file_fdatawait_range);
616 * filemap_fdatawait_keep_errors - wait for writeback without clearing errors
617 * @mapping: address space structure to wait for
619 * Walk the list of under-writeback pages of the given address space
620 * and wait for all of them. Unlike filemap_fdatawait(), this function
621 * does not clear error status of the address space.
623 * Use this function if callers don't handle errors themselves. Expected
624 * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2),
627 * Return: error status of the address space.
629 int filemap_fdatawait_keep_errors(struct address_space *mapping)
631 __filemap_fdatawait_range(mapping, 0, LLONG_MAX);
632 return filemap_check_and_keep_errors(mapping);
634 EXPORT_SYMBOL(filemap_fdatawait_keep_errors);
636 /* Returns true if writeback might be needed or already in progress. */
637 static bool mapping_needs_writeback(struct address_space *mapping)
639 return mapping->nrpages;
643 * filemap_range_needs_writeback - check if range potentially needs writeback
644 * @mapping: address space within which to check
645 * @start_byte: offset in bytes where the range starts
646 * @end_byte: offset in bytes where the range ends (inclusive)
648 * Find at least one page in the range supplied, usually used to check if
649 * direct writing in this range will trigger a writeback. Used by O_DIRECT
650 * read/write with IOCB_NOWAIT, to see if the caller needs to do
651 * filemap_write_and_wait_range() before proceeding.
653 * Return: %true if the caller should do filemap_write_and_wait_range() before
654 * doing O_DIRECT to a page in this range, %false otherwise.
656 bool filemap_range_needs_writeback(struct address_space *mapping,
657 loff_t start_byte, loff_t end_byte)
659 XA_STATE(xas, &mapping->i_pages, start_byte >> PAGE_SHIFT);
660 pgoff_t max = end_byte >> PAGE_SHIFT;
663 if (!mapping_needs_writeback(mapping))
665 if (!mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) &&
666 !mapping_tagged(mapping, PAGECACHE_TAG_WRITEBACK))
668 if (end_byte < start_byte)
672 xas_for_each(&xas, page, max) {
673 if (xas_retry(&xas, page))
675 if (xa_is_value(page))
677 if (PageDirty(page) || PageLocked(page) || PageWriteback(page))
683 EXPORT_SYMBOL_GPL(filemap_range_needs_writeback);
686 * filemap_write_and_wait_range - write out & wait on a file range
687 * @mapping: the address_space for the pages
688 * @lstart: offset in bytes where the range starts
689 * @lend: offset in bytes where the range ends (inclusive)
691 * Write out and wait upon file offsets lstart->lend, inclusive.
693 * Note that @lend is inclusive (describes the last byte to be written) so
694 * that this function can be used to write to the very end-of-file (end = -1).
696 * Return: error status of the address space.
698 int filemap_write_and_wait_range(struct address_space *mapping,
699 loff_t lstart, loff_t lend)
703 if (mapping_needs_writeback(mapping)) {
704 err = __filemap_fdatawrite_range(mapping, lstart, lend,
707 * Even if the above returned error, the pages may be
708 * written partially (e.g. -ENOSPC), so we wait for it.
709 * But the -EIO is special case, it may indicate the worst
710 * thing (e.g. bug) happened, so we avoid waiting for it.
713 int err2 = filemap_fdatawait_range(mapping,
718 /* Clear any previously stored errors */
719 filemap_check_errors(mapping);
722 err = filemap_check_errors(mapping);
726 EXPORT_SYMBOL(filemap_write_and_wait_range);
728 void __filemap_set_wb_err(struct address_space *mapping, int err)
730 errseq_t eseq = errseq_set(&mapping->wb_err, err);
732 trace_filemap_set_wb_err(mapping, eseq);
734 EXPORT_SYMBOL(__filemap_set_wb_err);
737 * file_check_and_advance_wb_err - report wb error (if any) that was previously
738 * and advance wb_err to current one
739 * @file: struct file on which the error is being reported
741 * When userland calls fsync (or something like nfsd does the equivalent), we
742 * want to report any writeback errors that occurred since the last fsync (or
743 * since the file was opened if there haven't been any).
745 * Grab the wb_err from the mapping. If it matches what we have in the file,
746 * then just quickly return 0. The file is all caught up.
748 * If it doesn't match, then take the mapping value, set the "seen" flag in
749 * it and try to swap it into place. If it works, or another task beat us
750 * to it with the new value, then update the f_wb_err and return the error
751 * portion. The error at this point must be reported via proper channels
752 * (a'la fsync, or NFS COMMIT operation, etc.).
754 * While we handle mapping->wb_err with atomic operations, the f_wb_err
755 * value is protected by the f_lock since we must ensure that it reflects
756 * the latest value swapped in for this file descriptor.
758 * Return: %0 on success, negative error code otherwise.
760 int file_check_and_advance_wb_err(struct file *file)
763 errseq_t old = READ_ONCE(file->f_wb_err);
764 struct address_space *mapping = file->f_mapping;
766 /* Locklessly handle the common case where nothing has changed */
767 if (errseq_check(&mapping->wb_err, old)) {
768 /* Something changed, must use slow path */
769 spin_lock(&file->f_lock);
770 old = file->f_wb_err;
771 err = errseq_check_and_advance(&mapping->wb_err,
773 trace_file_check_and_advance_wb_err(file, old);
774 spin_unlock(&file->f_lock);
778 * We're mostly using this function as a drop in replacement for
779 * filemap_check_errors. Clear AS_EIO/AS_ENOSPC to emulate the effect
780 * that the legacy code would have had on these flags.
782 clear_bit(AS_EIO, &mapping->flags);
783 clear_bit(AS_ENOSPC, &mapping->flags);
786 EXPORT_SYMBOL(file_check_and_advance_wb_err);
789 * file_write_and_wait_range - write out & wait on a file range
790 * @file: file pointing to address_space with pages
791 * @lstart: offset in bytes where the range starts
792 * @lend: offset in bytes where the range ends (inclusive)
794 * Write out and wait upon file offsets lstart->lend, inclusive.
796 * Note that @lend is inclusive (describes the last byte to be written) so
797 * that this function can be used to write to the very end-of-file (end = -1).
799 * After writing out and waiting on the data, we check and advance the
800 * f_wb_err cursor to the latest value, and return any errors detected there.
802 * Return: %0 on success, negative error code otherwise.
804 int file_write_and_wait_range(struct file *file, loff_t lstart, loff_t lend)
807 struct address_space *mapping = file->f_mapping;
809 if (mapping_needs_writeback(mapping)) {
810 err = __filemap_fdatawrite_range(mapping, lstart, lend,
812 /* See comment of filemap_write_and_wait() */
814 __filemap_fdatawait_range(mapping, lstart, lend);
816 err2 = file_check_and_advance_wb_err(file);
821 EXPORT_SYMBOL(file_write_and_wait_range);
824 * replace_page_cache_page - replace a pagecache page with a new one
825 * @old: page to be replaced
826 * @new: page to replace with
828 * This function replaces a page in the pagecache with a new one. On
829 * success it acquires the pagecache reference for the new page and
830 * drops it for the old page. Both the old and new pages must be
831 * locked. This function does not add the new page to the LRU, the
832 * caller must do that.
834 * The remove + add is atomic. This function cannot fail.
836 void replace_page_cache_page(struct page *old, struct page *new)
838 struct address_space *mapping = old->mapping;
839 void (*freepage)(struct page *) = mapping->a_ops->freepage;
840 pgoff_t offset = old->index;
841 XA_STATE(xas, &mapping->i_pages, offset);
843 VM_BUG_ON_PAGE(!PageLocked(old), old);
844 VM_BUG_ON_PAGE(!PageLocked(new), new);
845 VM_BUG_ON_PAGE(new->mapping, new);
848 new->mapping = mapping;
851 mem_cgroup_migrate(old, new);
854 xas_store(&xas, new);
857 /* hugetlb pages do not participate in page cache accounting. */
859 __dec_lruvec_page_state(old, NR_FILE_PAGES);
861 __inc_lruvec_page_state(new, NR_FILE_PAGES);
862 if (PageSwapBacked(old))
863 __dec_lruvec_page_state(old, NR_SHMEM);
864 if (PageSwapBacked(new))
865 __inc_lruvec_page_state(new, NR_SHMEM);
866 xas_unlock_irq(&xas);
871 EXPORT_SYMBOL_GPL(replace_page_cache_page);
873 noinline int __add_to_page_cache_locked(struct page *page,
874 struct address_space *mapping,
875 pgoff_t offset, gfp_t gfp,
878 XA_STATE(xas, &mapping->i_pages, offset);
879 int huge = PageHuge(page);
881 bool charged = false;
883 VM_BUG_ON_PAGE(!PageLocked(page), page);
884 VM_BUG_ON_PAGE(PageSwapBacked(page), page);
885 mapping_set_update(&xas, mapping);
888 page->mapping = mapping;
889 page->index = offset;
892 error = mem_cgroup_charge(page, NULL, gfp);
898 gfp &= GFP_RECLAIM_MASK;
901 unsigned int order = xa_get_order(xas.xa, xas.xa_index);
902 void *entry, *old = NULL;
904 if (order > thp_order(page))
905 xas_split_alloc(&xas, xa_load(xas.xa, xas.xa_index),
908 xas_for_each_conflict(&xas, entry) {
910 if (!xa_is_value(entry)) {
911 xas_set_err(&xas, -EEXIST);
919 /* entry may have been split before we acquired lock */
920 order = xa_get_order(xas.xa, xas.xa_index);
921 if (order > thp_order(page)) {
922 xas_split(&xas, old, order);
927 xas_store(&xas, page);
933 /* hugetlb pages do not participate in page cache accounting */
935 __inc_lruvec_page_state(page, NR_FILE_PAGES);
937 xas_unlock_irq(&xas);
938 } while (xas_nomem(&xas, gfp));
940 if (xas_error(&xas)) {
941 error = xas_error(&xas);
943 mem_cgroup_uncharge(page);
947 trace_mm_filemap_add_to_page_cache(page);
950 page->mapping = NULL;
951 /* Leave page->index set: truncation relies upon it */
955 ALLOW_ERROR_INJECTION(__add_to_page_cache_locked, ERRNO);
958 * add_to_page_cache_locked - add a locked page to the pagecache
960 * @mapping: the page's address_space
961 * @offset: page index
962 * @gfp_mask: page allocation mode
964 * This function is used to add a page to the pagecache. It must be locked.
965 * This function does not add the page to the LRU. The caller must do that.
967 * Return: %0 on success, negative error code otherwise.
969 int add_to_page_cache_locked(struct page *page, struct address_space *mapping,
970 pgoff_t offset, gfp_t gfp_mask)
972 return __add_to_page_cache_locked(page, mapping, offset,
975 EXPORT_SYMBOL(add_to_page_cache_locked);
977 int add_to_page_cache_lru(struct page *page, struct address_space *mapping,
978 pgoff_t offset, gfp_t gfp_mask)
983 __SetPageLocked(page);
984 ret = __add_to_page_cache_locked(page, mapping, offset,
987 __ClearPageLocked(page);
990 * The page might have been evicted from cache only
991 * recently, in which case it should be activated like
992 * any other repeatedly accessed page.
993 * The exception is pages getting rewritten; evicting other
994 * data from the working set, only to cache data that will
995 * get overwritten with something else, is a waste of memory.
997 WARN_ON_ONCE(PageActive(page));
998 if (!(gfp_mask & __GFP_WRITE) && shadow)
999 workingset_refault(page, shadow);
1000 lru_cache_add(page);
1004 EXPORT_SYMBOL_GPL(add_to_page_cache_lru);
1007 struct page *__page_cache_alloc(gfp_t gfp)
1012 if (cpuset_do_page_mem_spread()) {
1013 unsigned int cpuset_mems_cookie;
1015 cpuset_mems_cookie = read_mems_allowed_begin();
1016 n = cpuset_mem_spread_node();
1017 page = __alloc_pages_node(n, gfp, 0);
1018 } while (!page && read_mems_allowed_retry(cpuset_mems_cookie));
1022 return alloc_pages(gfp, 0);
1024 EXPORT_SYMBOL(__page_cache_alloc);
1028 * filemap_invalidate_lock_two - lock invalidate_lock for two mappings
1030 * Lock exclusively invalidate_lock of any passed mapping that is not NULL.
1032 * @mapping1: the first mapping to lock
1033 * @mapping2: the second mapping to lock
1035 void filemap_invalidate_lock_two(struct address_space *mapping1,
1036 struct address_space *mapping2)
1038 if (mapping1 > mapping2)
1039 swap(mapping1, mapping2);
1041 down_write(&mapping1->invalidate_lock);
1042 if (mapping2 && mapping1 != mapping2)
1043 down_write_nested(&mapping2->invalidate_lock, 1);
1045 EXPORT_SYMBOL(filemap_invalidate_lock_two);
1048 * filemap_invalidate_unlock_two - unlock invalidate_lock for two mappings
1050 * Unlock exclusive invalidate_lock of any passed mapping that is not NULL.
1052 * @mapping1: the first mapping to unlock
1053 * @mapping2: the second mapping to unlock
1055 void filemap_invalidate_unlock_two(struct address_space *mapping1,
1056 struct address_space *mapping2)
1059 up_write(&mapping1->invalidate_lock);
1060 if (mapping2 && mapping1 != mapping2)
1061 up_write(&mapping2->invalidate_lock);
1063 EXPORT_SYMBOL(filemap_invalidate_unlock_two);
1066 * In order to wait for pages to become available there must be
1067 * waitqueues associated with pages. By using a hash table of
1068 * waitqueues where the bucket discipline is to maintain all
1069 * waiters on the same queue and wake all when any of the pages
1070 * become available, and for the woken contexts to check to be
1071 * sure the appropriate page became available, this saves space
1072 * at a cost of "thundering herd" phenomena during rare hash
1075 #define PAGE_WAIT_TABLE_BITS 8
1076 #define PAGE_WAIT_TABLE_SIZE (1 << PAGE_WAIT_TABLE_BITS)
1077 static wait_queue_head_t page_wait_table[PAGE_WAIT_TABLE_SIZE] __cacheline_aligned;
1079 static wait_queue_head_t *page_waitqueue(struct page *page)
1081 return &page_wait_table[hash_ptr(page, PAGE_WAIT_TABLE_BITS)];
1084 void __init pagecache_init(void)
1088 for (i = 0; i < PAGE_WAIT_TABLE_SIZE; i++)
1089 init_waitqueue_head(&page_wait_table[i]);
1091 page_writeback_init();
1095 * The page wait code treats the "wait->flags" somewhat unusually, because
1096 * we have multiple different kinds of waits, not just the usual "exclusive"
1101 * (a) no special bits set:
1103 * We're just waiting for the bit to be released, and when a waker
1104 * calls the wakeup function, we set WQ_FLAG_WOKEN and wake it up,
1105 * and remove it from the wait queue.
1107 * Simple and straightforward.
1109 * (b) WQ_FLAG_EXCLUSIVE:
1111 * The waiter is waiting to get the lock, and only one waiter should
1112 * be woken up to avoid any thundering herd behavior. We'll set the
1113 * WQ_FLAG_WOKEN bit, wake it up, and remove it from the wait queue.
1115 * This is the traditional exclusive wait.
1117 * (c) WQ_FLAG_EXCLUSIVE | WQ_FLAG_CUSTOM:
1119 * The waiter is waiting to get the bit, and additionally wants the
1120 * lock to be transferred to it for fair lock behavior. If the lock
1121 * cannot be taken, we stop walking the wait queue without waking
1124 * This is the "fair lock handoff" case, and in addition to setting
1125 * WQ_FLAG_WOKEN, we set WQ_FLAG_DONE to let the waiter easily see
1126 * that it now has the lock.
1128 static int wake_page_function(wait_queue_entry_t *wait, unsigned mode, int sync, void *arg)
1131 struct wait_page_key *key = arg;
1132 struct wait_page_queue *wait_page
1133 = container_of(wait, struct wait_page_queue, wait);
1135 if (!wake_page_match(wait_page, key))
1139 * If it's a lock handoff wait, we get the bit for it, and
1140 * stop walking (and do not wake it up) if we can't.
1142 flags = wait->flags;
1143 if (flags & WQ_FLAG_EXCLUSIVE) {
1144 if (test_bit(key->bit_nr, &key->page->flags))
1146 if (flags & WQ_FLAG_CUSTOM) {
1147 if (test_and_set_bit(key->bit_nr, &key->page->flags))
1149 flags |= WQ_FLAG_DONE;
1154 * We are holding the wait-queue lock, but the waiter that
1155 * is waiting for this will be checking the flags without
1158 * So update the flags atomically, and wake up the waiter
1159 * afterwards to avoid any races. This store-release pairs
1160 * with the load-acquire in wait_on_page_bit_common().
1162 smp_store_release(&wait->flags, flags | WQ_FLAG_WOKEN);
1163 wake_up_state(wait->private, mode);
1166 * Ok, we have successfully done what we're waiting for,
1167 * and we can unconditionally remove the wait entry.
1169 * Note that this pairs with the "finish_wait()" in the
1170 * waiter, and has to be the absolute last thing we do.
1171 * After this list_del_init(&wait->entry) the wait entry
1172 * might be de-allocated and the process might even have
1175 list_del_init_careful(&wait->entry);
1176 return (flags & WQ_FLAG_EXCLUSIVE) != 0;
1179 static void wake_up_page_bit(struct page *page, int bit_nr)
1181 wait_queue_head_t *q = page_waitqueue(page);
1182 struct wait_page_key key;
1183 unsigned long flags;
1184 wait_queue_entry_t bookmark;
1187 key.bit_nr = bit_nr;
1191 bookmark.private = NULL;
1192 bookmark.func = NULL;
1193 INIT_LIST_HEAD(&bookmark.entry);
1195 spin_lock_irqsave(&q->lock, flags);
1196 __wake_up_locked_key_bookmark(q, TASK_NORMAL, &key, &bookmark);
1198 while (bookmark.flags & WQ_FLAG_BOOKMARK) {
1200 * Take a breather from holding the lock,
1201 * allow pages that finish wake up asynchronously
1202 * to acquire the lock and remove themselves
1205 spin_unlock_irqrestore(&q->lock, flags);
1207 spin_lock_irqsave(&q->lock, flags);
1208 __wake_up_locked_key_bookmark(q, TASK_NORMAL, &key, &bookmark);
1212 * It is possible for other pages to have collided on the waitqueue
1213 * hash, so in that case check for a page match. That prevents a long-
1216 * It is still possible to miss a case here, when we woke page waiters
1217 * and removed them from the waitqueue, but there are still other
1220 if (!waitqueue_active(q) || !key.page_match) {
1221 ClearPageWaiters(page);
1223 * It's possible to miss clearing Waiters here, when we woke
1224 * our page waiters, but the hashed waitqueue has waiters for
1225 * other pages on it.
1227 * That's okay, it's a rare case. The next waker will clear it.
1230 spin_unlock_irqrestore(&q->lock, flags);
1233 static void wake_up_page(struct page *page, int bit)
1235 if (!PageWaiters(page))
1237 wake_up_page_bit(page, bit);
1241 * A choice of three behaviors for wait_on_page_bit_common():
1244 EXCLUSIVE, /* Hold ref to page and take the bit when woken, like
1245 * __lock_page() waiting on then setting PG_locked.
1247 SHARED, /* Hold ref to page and check the bit when woken, like
1248 * wait_on_page_writeback() waiting on PG_writeback.
1250 DROP, /* Drop ref to page before wait, no check when woken,
1251 * like put_and_wait_on_page_locked() on PG_locked.
1256 * Attempt to check (or get) the page bit, and mark us done
1259 static inline bool trylock_page_bit_common(struct page *page, int bit_nr,
1260 struct wait_queue_entry *wait)
1262 if (wait->flags & WQ_FLAG_EXCLUSIVE) {
1263 if (test_and_set_bit(bit_nr, &page->flags))
1265 } else if (test_bit(bit_nr, &page->flags))
1268 wait->flags |= WQ_FLAG_WOKEN | WQ_FLAG_DONE;
1272 /* How many times do we accept lock stealing from under a waiter? */
1273 int sysctl_page_lock_unfairness = 5;
1275 static inline int wait_on_page_bit_common(wait_queue_head_t *q,
1276 struct page *page, int bit_nr, int state, enum behavior behavior)
1278 int unfairness = sysctl_page_lock_unfairness;
1279 struct wait_page_queue wait_page;
1280 wait_queue_entry_t *wait = &wait_page.wait;
1281 bool thrashing = false;
1282 bool delayacct = false;
1283 unsigned long pflags;
1285 if (bit_nr == PG_locked &&
1286 !PageUptodate(page) && PageWorkingset(page)) {
1287 if (!PageSwapBacked(page)) {
1288 delayacct_thrashing_start();
1291 psi_memstall_enter(&pflags);
1296 wait->func = wake_page_function;
1297 wait_page.page = page;
1298 wait_page.bit_nr = bit_nr;
1302 if (behavior == EXCLUSIVE) {
1303 wait->flags = WQ_FLAG_EXCLUSIVE;
1304 if (--unfairness < 0)
1305 wait->flags |= WQ_FLAG_CUSTOM;
1309 * Do one last check whether we can get the
1310 * page bit synchronously.
1312 * Do the SetPageWaiters() marking before that
1313 * to let any waker we _just_ missed know they
1314 * need to wake us up (otherwise they'll never
1315 * even go to the slow case that looks at the
1316 * page queue), and add ourselves to the wait
1317 * queue if we need to sleep.
1319 * This part needs to be done under the queue
1320 * lock to avoid races.
1322 spin_lock_irq(&q->lock);
1323 SetPageWaiters(page);
1324 if (!trylock_page_bit_common(page, bit_nr, wait))
1325 __add_wait_queue_entry_tail(q, wait);
1326 spin_unlock_irq(&q->lock);
1329 * From now on, all the logic will be based on
1330 * the WQ_FLAG_WOKEN and WQ_FLAG_DONE flag, to
1331 * see whether the page bit testing has already
1332 * been done by the wake function.
1334 * We can drop our reference to the page.
1336 if (behavior == DROP)
1340 * Note that until the "finish_wait()", or until
1341 * we see the WQ_FLAG_WOKEN flag, we need to
1342 * be very careful with the 'wait->flags', because
1343 * we may race with a waker that sets them.
1348 set_current_state(state);
1350 /* Loop until we've been woken or interrupted */
1351 flags = smp_load_acquire(&wait->flags);
1352 if (!(flags & WQ_FLAG_WOKEN)) {
1353 if (signal_pending_state(state, current))
1360 /* If we were non-exclusive, we're done */
1361 if (behavior != EXCLUSIVE)
1364 /* If the waker got the lock for us, we're done */
1365 if (flags & WQ_FLAG_DONE)
1369 * Otherwise, if we're getting the lock, we need to
1370 * try to get it ourselves.
1372 * And if that fails, we'll have to retry this all.
1374 if (unlikely(test_and_set_bit(bit_nr, &page->flags)))
1377 wait->flags |= WQ_FLAG_DONE;
1382 * If a signal happened, this 'finish_wait()' may remove the last
1383 * waiter from the wait-queues, but the PageWaiters bit will remain
1384 * set. That's ok. The next wakeup will take care of it, and trying
1385 * to do it here would be difficult and prone to races.
1387 finish_wait(q, wait);
1391 delayacct_thrashing_end();
1392 psi_memstall_leave(&pflags);
1396 * NOTE! The wait->flags weren't stable until we've done the
1397 * 'finish_wait()', and we could have exited the loop above due
1398 * to a signal, and had a wakeup event happen after the signal
1399 * test but before the 'finish_wait()'.
1401 * So only after the finish_wait() can we reliably determine
1402 * if we got woken up or not, so we can now figure out the final
1403 * return value based on that state without races.
1405 * Also note that WQ_FLAG_WOKEN is sufficient for a non-exclusive
1406 * waiter, but an exclusive one requires WQ_FLAG_DONE.
1408 if (behavior == EXCLUSIVE)
1409 return wait->flags & WQ_FLAG_DONE ? 0 : -EINTR;
1411 return wait->flags & WQ_FLAG_WOKEN ? 0 : -EINTR;
1414 void wait_on_page_bit(struct page *page, int bit_nr)
1416 wait_queue_head_t *q = page_waitqueue(page);
1417 wait_on_page_bit_common(q, page, bit_nr, TASK_UNINTERRUPTIBLE, SHARED);
1419 EXPORT_SYMBOL(wait_on_page_bit);
1421 int wait_on_page_bit_killable(struct page *page, int bit_nr)
1423 wait_queue_head_t *q = page_waitqueue(page);
1424 return wait_on_page_bit_common(q, page, bit_nr, TASK_KILLABLE, SHARED);
1426 EXPORT_SYMBOL(wait_on_page_bit_killable);
1429 * put_and_wait_on_page_locked - Drop a reference and wait for it to be unlocked
1430 * @page: The page to wait for.
1431 * @state: The sleep state (TASK_KILLABLE, TASK_UNINTERRUPTIBLE, etc).
1433 * The caller should hold a reference on @page. They expect the page to
1434 * become unlocked relatively soon, but do not wish to hold up migration
1435 * (for example) by holding the reference while waiting for the page to
1436 * come unlocked. After this function returns, the caller should not
1437 * dereference @page.
1439 * Return: 0 if the page was unlocked or -EINTR if interrupted by a signal.
1441 int put_and_wait_on_page_locked(struct page *page, int state)
1443 wait_queue_head_t *q;
1445 page = compound_head(page);
1446 q = page_waitqueue(page);
1447 return wait_on_page_bit_common(q, page, PG_locked, state, DROP);
1451 * add_page_wait_queue - Add an arbitrary waiter to a page's wait queue
1452 * @page: Page defining the wait queue of interest
1453 * @waiter: Waiter to add to the queue
1455 * Add an arbitrary @waiter to the wait queue for the nominated @page.
1457 void add_page_wait_queue(struct page *page, wait_queue_entry_t *waiter)
1459 wait_queue_head_t *q = page_waitqueue(page);
1460 unsigned long flags;
1462 spin_lock_irqsave(&q->lock, flags);
1463 __add_wait_queue_entry_tail(q, waiter);
1464 SetPageWaiters(page);
1465 spin_unlock_irqrestore(&q->lock, flags);
1467 EXPORT_SYMBOL_GPL(add_page_wait_queue);
1469 #ifndef clear_bit_unlock_is_negative_byte
1472 * PG_waiters is the high bit in the same byte as PG_lock.
1474 * On x86 (and on many other architectures), we can clear PG_lock and
1475 * test the sign bit at the same time. But if the architecture does
1476 * not support that special operation, we just do this all by hand
1479 * The read of PG_waiters has to be after (or concurrently with) PG_locked
1480 * being cleared, but a memory barrier should be unnecessary since it is
1481 * in the same byte as PG_locked.
1483 static inline bool clear_bit_unlock_is_negative_byte(long nr, volatile void *mem)
1485 clear_bit_unlock(nr, mem);
1486 /* smp_mb__after_atomic(); */
1487 return test_bit(PG_waiters, mem);
1493 * unlock_page - unlock a locked page
1496 * Unlocks the page and wakes up sleepers in wait_on_page_locked().
1497 * Also wakes sleepers in wait_on_page_writeback() because the wakeup
1498 * mechanism between PageLocked pages and PageWriteback pages is shared.
1499 * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep.
1501 * Note that this depends on PG_waiters being the sign bit in the byte
1502 * that contains PG_locked - thus the BUILD_BUG_ON(). That allows us to
1503 * clear the PG_locked bit and test PG_waiters at the same time fairly
1504 * portably (architectures that do LL/SC can test any bit, while x86 can
1505 * test the sign bit).
1507 void unlock_page(struct page *page)
1509 BUILD_BUG_ON(PG_waiters != 7);
1510 page = compound_head(page);
1511 VM_BUG_ON_PAGE(!PageLocked(page), page);
1512 if (clear_bit_unlock_is_negative_byte(PG_locked, &page->flags))
1513 wake_up_page_bit(page, PG_locked);
1515 EXPORT_SYMBOL(unlock_page);
1518 * end_page_private_2 - Clear PG_private_2 and release any waiters
1521 * Clear the PG_private_2 bit on a page and wake up any sleepers waiting for
1522 * this. The page ref held for PG_private_2 being set is released.
1524 * This is, for example, used when a netfs page is being written to a local
1525 * disk cache, thereby allowing writes to the cache for the same page to be
1528 void end_page_private_2(struct page *page)
1530 page = compound_head(page);
1531 VM_BUG_ON_PAGE(!PagePrivate2(page), page);
1532 clear_bit_unlock(PG_private_2, &page->flags);
1533 wake_up_page_bit(page, PG_private_2);
1536 EXPORT_SYMBOL(end_page_private_2);
1539 * wait_on_page_private_2 - Wait for PG_private_2 to be cleared on a page
1540 * @page: The page to wait on
1542 * Wait for PG_private_2 (aka PG_fscache) to be cleared on a page.
1544 void wait_on_page_private_2(struct page *page)
1546 page = compound_head(page);
1547 while (PagePrivate2(page))
1548 wait_on_page_bit(page, PG_private_2);
1550 EXPORT_SYMBOL(wait_on_page_private_2);
1553 * wait_on_page_private_2_killable - Wait for PG_private_2 to be cleared on a page
1554 * @page: The page to wait on
1556 * Wait for PG_private_2 (aka PG_fscache) to be cleared on a page or until a
1557 * fatal signal is received by the calling task.
1560 * - 0 if successful.
1561 * - -EINTR if a fatal signal was encountered.
1563 int wait_on_page_private_2_killable(struct page *page)
1567 page = compound_head(page);
1568 while (PagePrivate2(page)) {
1569 ret = wait_on_page_bit_killable(page, PG_private_2);
1576 EXPORT_SYMBOL(wait_on_page_private_2_killable);
1579 * end_page_writeback - end writeback against a page
1582 void end_page_writeback(struct page *page)
1585 * TestClearPageReclaim could be used here but it is an atomic
1586 * operation and overkill in this particular case. Failing to
1587 * shuffle a page marked for immediate reclaim is too mild to
1588 * justify taking an atomic operation penalty at the end of
1589 * ever page writeback.
1591 if (PageReclaim(page)) {
1592 ClearPageReclaim(page);
1593 rotate_reclaimable_page(page);
1597 * Writeback does not hold a page reference of its own, relying
1598 * on truncation to wait for the clearing of PG_writeback.
1599 * But here we must make sure that the page is not freed and
1600 * reused before the wake_up_page().
1603 if (!test_clear_page_writeback(page))
1606 smp_mb__after_atomic();
1607 wake_up_page(page, PG_writeback);
1610 EXPORT_SYMBOL(end_page_writeback);
1613 * After completing I/O on a page, call this routine to update the page
1614 * flags appropriately
1616 void page_endio(struct page *page, bool is_write, int err)
1620 SetPageUptodate(page);
1622 ClearPageUptodate(page);
1628 struct address_space *mapping;
1631 mapping = page_mapping(page);
1633 mapping_set_error(mapping, err);
1635 end_page_writeback(page);
1638 EXPORT_SYMBOL_GPL(page_endio);
1641 * __lock_page - get a lock on the page, assuming we need to sleep to get it
1642 * @__page: the page to lock
1644 void __lock_page(struct page *__page)
1646 struct page *page = compound_head(__page);
1647 wait_queue_head_t *q = page_waitqueue(page);
1648 wait_on_page_bit_common(q, page, PG_locked, TASK_UNINTERRUPTIBLE,
1651 EXPORT_SYMBOL(__lock_page);
1653 int __lock_page_killable(struct page *__page)
1655 struct page *page = compound_head(__page);
1656 wait_queue_head_t *q = page_waitqueue(page);
1657 return wait_on_page_bit_common(q, page, PG_locked, TASK_KILLABLE,
1660 EXPORT_SYMBOL_GPL(__lock_page_killable);
1662 int __lock_page_async(struct page *page, struct wait_page_queue *wait)
1664 struct wait_queue_head *q = page_waitqueue(page);
1668 wait->bit_nr = PG_locked;
1670 spin_lock_irq(&q->lock);
1671 __add_wait_queue_entry_tail(q, &wait->wait);
1672 SetPageWaiters(page);
1673 ret = !trylock_page(page);
1675 * If we were successful now, we know we're still on the
1676 * waitqueue as we're still under the lock. This means it's
1677 * safe to remove and return success, we know the callback
1678 * isn't going to trigger.
1681 __remove_wait_queue(q, &wait->wait);
1684 spin_unlock_irq(&q->lock);
1690 * 1 - page is locked; mmap_lock is still held.
1691 * 0 - page is not locked.
1692 * mmap_lock has been released (mmap_read_unlock(), unless flags had both
1693 * FAULT_FLAG_ALLOW_RETRY and FAULT_FLAG_RETRY_NOWAIT set, in
1694 * which case mmap_lock is still held.
1696 * If neither ALLOW_RETRY nor KILLABLE are set, will always return 1
1697 * with the page locked and the mmap_lock unperturbed.
1699 int __lock_page_or_retry(struct page *page, struct mm_struct *mm,
1702 if (fault_flag_allow_retry_first(flags)) {
1704 * CAUTION! In this case, mmap_lock is not released
1705 * even though return 0.
1707 if (flags & FAULT_FLAG_RETRY_NOWAIT)
1710 mmap_read_unlock(mm);
1711 if (flags & FAULT_FLAG_KILLABLE)
1712 wait_on_page_locked_killable(page);
1714 wait_on_page_locked(page);
1717 if (flags & FAULT_FLAG_KILLABLE) {
1720 ret = __lock_page_killable(page);
1722 mmap_read_unlock(mm);
1733 * page_cache_next_miss() - Find the next gap in the page cache.
1734 * @mapping: Mapping.
1736 * @max_scan: Maximum range to search.
1738 * Search the range [index, min(index + max_scan - 1, ULONG_MAX)] for the
1739 * gap with the lowest index.
1741 * This function may be called under the rcu_read_lock. However, this will
1742 * not atomically search a snapshot of the cache at a single point in time.
1743 * For example, if a gap is created at index 5, then subsequently a gap is
1744 * created at index 10, page_cache_next_miss covering both indices may
1745 * return 10 if called under the rcu_read_lock.
1747 * Return: The index of the gap if found, otherwise an index outside the
1748 * range specified (in which case 'return - index >= max_scan' will be true).
1749 * In the rare case of index wrap-around, 0 will be returned.
1751 pgoff_t page_cache_next_miss(struct address_space *mapping,
1752 pgoff_t index, unsigned long max_scan)
1754 XA_STATE(xas, &mapping->i_pages, index);
1756 while (max_scan--) {
1757 void *entry = xas_next(&xas);
1758 if (!entry || xa_is_value(entry))
1760 if (xas.xa_index == 0)
1764 return xas.xa_index;
1766 EXPORT_SYMBOL(page_cache_next_miss);
1769 * page_cache_prev_miss() - Find the previous gap in the page cache.
1770 * @mapping: Mapping.
1772 * @max_scan: Maximum range to search.
1774 * Search the range [max(index - max_scan + 1, 0), index] for the
1775 * gap with the highest index.
1777 * This function may be called under the rcu_read_lock. However, this will
1778 * not atomically search a snapshot of the cache at a single point in time.
1779 * For example, if a gap is created at index 10, then subsequently a gap is
1780 * created at index 5, page_cache_prev_miss() covering both indices may
1781 * return 5 if called under the rcu_read_lock.
1783 * Return: The index of the gap if found, otherwise an index outside the
1784 * range specified (in which case 'index - return >= max_scan' will be true).
1785 * In the rare case of wrap-around, ULONG_MAX will be returned.
1787 pgoff_t page_cache_prev_miss(struct address_space *mapping,
1788 pgoff_t index, unsigned long max_scan)
1790 XA_STATE(xas, &mapping->i_pages, index);
1792 while (max_scan--) {
1793 void *entry = xas_prev(&xas);
1794 if (!entry || xa_is_value(entry))
1796 if (xas.xa_index == ULONG_MAX)
1800 return xas.xa_index;
1802 EXPORT_SYMBOL(page_cache_prev_miss);
1805 * Lockless page cache protocol:
1806 * On the lookup side:
1807 * 1. Load the folio from i_pages
1808 * 2. Increment the refcount if it's not zero
1809 * 3. If the folio is not found by xas_reload(), put the refcount and retry
1811 * On the removal side:
1812 * A. Freeze the page (by zeroing the refcount if nobody else has a reference)
1813 * B. Remove the page from i_pages
1814 * C. Return the page to the page allocator
1816 * This means that any page may have its reference count temporarily
1817 * increased by a speculative page cache (or fast GUP) lookup as it can
1818 * be allocated by another user before the RCU grace period expires.
1819 * Because the refcount temporarily acquired here may end up being the
1820 * last refcount on the page, any page allocation must be freeable by
1825 * mapping_get_entry - Get a page cache entry.
1826 * @mapping: the address_space to search
1827 * @index: The page cache index.
1829 * Looks up the page cache slot at @mapping & @index. If there is a
1830 * page cache page, the head page is returned with an increased refcount.
1832 * If the slot holds a shadow entry of a previously evicted page, or a
1833 * swap entry from shmem/tmpfs, it is returned.
1835 * Return: The head page or shadow entry, %NULL if nothing is found.
1837 static struct page *mapping_get_entry(struct address_space *mapping,
1840 XA_STATE(xas, &mapping->i_pages, index);
1846 page = xas_load(&xas);
1847 if (xas_retry(&xas, page))
1850 * A shadow entry of a recently evicted page, or a swap entry from
1851 * shmem/tmpfs. Return it without attempting to raise page count.
1853 if (!page || xa_is_value(page))
1856 if (!page_cache_get_speculative(page))
1860 * Has the page moved or been split?
1861 * This is part of the lockless pagecache protocol. See
1862 * include/linux/pagemap.h for details.
1864 if (unlikely(page != xas_reload(&xas))) {
1875 * pagecache_get_page - Find and get a reference to a page.
1876 * @mapping: The address_space to search.
1877 * @index: The page index.
1878 * @fgp_flags: %FGP flags modify how the page is returned.
1879 * @gfp_mask: Memory allocation flags to use if %FGP_CREAT is specified.
1881 * Looks up the page cache entry at @mapping & @index.
1883 * @fgp_flags can be zero or more of these flags:
1885 * * %FGP_ACCESSED - The page will be marked accessed.
1886 * * %FGP_LOCK - The page is returned locked.
1887 * * %FGP_HEAD - If the page is present and a THP, return the head page
1888 * rather than the exact page specified by the index.
1889 * * %FGP_ENTRY - If there is a shadow / swap / DAX entry, return it
1890 * instead of allocating a new page to replace it.
1891 * * %FGP_CREAT - If no page is present then a new page is allocated using
1892 * @gfp_mask and added to the page cache and the VM's LRU list.
1893 * The page is returned locked and with an increased refcount.
1894 * * %FGP_FOR_MMAP - The caller wants to do its own locking dance if the
1895 * page is already in cache. If the page was allocated, unlock it before
1896 * returning so the caller can do the same dance.
1897 * * %FGP_WRITE - The page will be written
1898 * * %FGP_NOFS - __GFP_FS will get cleared in gfp mask
1899 * * %FGP_NOWAIT - Don't get blocked by page lock
1901 * If %FGP_LOCK or %FGP_CREAT are specified then the function may sleep even
1902 * if the %GFP flags specified for %FGP_CREAT are atomic.
1904 * If there is a page cache page, it is returned with an increased refcount.
1906 * Return: The found page or %NULL otherwise.
1908 struct page *pagecache_get_page(struct address_space *mapping, pgoff_t index,
1909 int fgp_flags, gfp_t gfp_mask)
1914 page = mapping_get_entry(mapping, index);
1915 if (xa_is_value(page)) {
1916 if (fgp_flags & FGP_ENTRY)
1923 if (fgp_flags & FGP_LOCK) {
1924 if (fgp_flags & FGP_NOWAIT) {
1925 if (!trylock_page(page)) {
1933 /* Has the page been truncated? */
1934 if (unlikely(page->mapping != mapping)) {
1939 VM_BUG_ON_PAGE(!thp_contains(page, index), page);
1942 if (fgp_flags & FGP_ACCESSED)
1943 mark_page_accessed(page);
1944 else if (fgp_flags & FGP_WRITE) {
1945 /* Clear idle flag for buffer write */
1946 if (page_is_idle(page))
1947 clear_page_idle(page);
1949 if (!(fgp_flags & FGP_HEAD))
1950 page = find_subpage(page, index);
1953 if (!page && (fgp_flags & FGP_CREAT)) {
1955 if ((fgp_flags & FGP_WRITE) && mapping_can_writeback(mapping))
1956 gfp_mask |= __GFP_WRITE;
1957 if (fgp_flags & FGP_NOFS)
1958 gfp_mask &= ~__GFP_FS;
1960 page = __page_cache_alloc(gfp_mask);
1964 if (WARN_ON_ONCE(!(fgp_flags & (FGP_LOCK | FGP_FOR_MMAP))))
1965 fgp_flags |= FGP_LOCK;
1967 /* Init accessed so avoid atomic mark_page_accessed later */
1968 if (fgp_flags & FGP_ACCESSED)
1969 __SetPageReferenced(page);
1971 err = add_to_page_cache_lru(page, mapping, index, gfp_mask);
1972 if (unlikely(err)) {
1980 * add_to_page_cache_lru locks the page, and for mmap we expect
1983 if (page && (fgp_flags & FGP_FOR_MMAP))
1989 EXPORT_SYMBOL(pagecache_get_page);
1991 static inline struct page *find_get_entry(struct xa_state *xas, pgoff_t max,
1997 if (mark == XA_PRESENT)
1998 page = xas_find(xas, max);
2000 page = xas_find_marked(xas, max, mark);
2002 if (xas_retry(xas, page))
2005 * A shadow entry of a recently evicted page, a swap
2006 * entry from shmem/tmpfs or a DAX entry. Return it
2007 * without attempting to raise page count.
2009 if (!page || xa_is_value(page))
2012 if (!page_cache_get_speculative(page))
2015 /* Has the page moved or been split? */
2016 if (unlikely(page != xas_reload(xas))) {
2028 * find_get_entries - gang pagecache lookup
2029 * @mapping: The address_space to search
2030 * @start: The starting page cache index
2031 * @end: The final page index (inclusive).
2032 * @pvec: Where the resulting entries are placed.
2033 * @indices: The cache indices corresponding to the entries in @entries
2035 * find_get_entries() will search for and return a batch of entries in
2036 * the mapping. The entries are placed in @pvec. find_get_entries()
2037 * takes a reference on any actual pages it returns.
2039 * The search returns a group of mapping-contiguous page cache entries
2040 * with ascending indexes. There may be holes in the indices due to
2041 * not-present pages.
2043 * Any shadow entries of evicted pages, or swap entries from
2044 * shmem/tmpfs, are included in the returned array.
2046 * If it finds a Transparent Huge Page, head or tail, find_get_entries()
2047 * stops at that page: the caller is likely to have a better way to handle
2048 * the compound page as a whole, and then skip its extent, than repeatedly
2049 * calling find_get_entries() to return all its tails.
2051 * Return: the number of pages and shadow entries which were found.
2053 unsigned find_get_entries(struct address_space *mapping, pgoff_t start,
2054 pgoff_t end, struct pagevec *pvec, pgoff_t *indices)
2056 XA_STATE(xas, &mapping->i_pages, start);
2058 unsigned int ret = 0;
2059 unsigned nr_entries = PAGEVEC_SIZE;
2062 while ((page = find_get_entry(&xas, end, XA_PRESENT))) {
2064 * Terminate early on finding a THP, to allow the caller to
2065 * handle it all at once; but continue if this is hugetlbfs.
2067 if (!xa_is_value(page) && PageTransHuge(page) &&
2069 page = find_subpage(page, xas.xa_index);
2070 nr_entries = ret + 1;
2073 indices[ret] = xas.xa_index;
2074 pvec->pages[ret] = page;
2075 if (++ret == nr_entries)
2085 * find_lock_entries - Find a batch of pagecache entries.
2086 * @mapping: The address_space to search.
2087 * @start: The starting page cache index.
2088 * @end: The final page index (inclusive).
2089 * @pvec: Where the resulting entries are placed.
2090 * @indices: The cache indices of the entries in @pvec.
2092 * find_lock_entries() will return a batch of entries from @mapping.
2093 * Swap, shadow and DAX entries are included. Pages are returned
2094 * locked and with an incremented refcount. Pages which are locked by
2095 * somebody else or under writeback are skipped. Only the head page of
2096 * a THP is returned. Pages which are partially outside the range are
2099 * The entries have ascending indexes. The indices may not be consecutive
2100 * due to not-present entries, THP pages, pages which could not be locked
2101 * or pages under writeback.
2103 * Return: The number of entries which were found.
2105 unsigned find_lock_entries(struct address_space *mapping, pgoff_t start,
2106 pgoff_t end, struct pagevec *pvec, pgoff_t *indices)
2108 XA_STATE(xas, &mapping->i_pages, start);
2112 while ((page = find_get_entry(&xas, end, XA_PRESENT))) {
2113 if (!xa_is_value(page)) {
2114 if (page->index < start)
2116 VM_BUG_ON_PAGE(page->index != xas.xa_index, page);
2117 if (page->index + thp_nr_pages(page) - 1 > end)
2119 if (!trylock_page(page))
2121 if (page->mapping != mapping || PageWriteback(page))
2123 VM_BUG_ON_PAGE(!thp_contains(page, xas.xa_index),
2126 indices[pvec->nr] = xas.xa_index;
2127 if (!pagevec_add(pvec, page))
2135 if (!xa_is_value(page) && PageTransHuge(page)) {
2136 unsigned int nr_pages = thp_nr_pages(page);
2138 /* Final THP may cross MAX_LFS_FILESIZE on 32-bit */
2139 xas_set(&xas, page->index + nr_pages);
2140 if (xas.xa_index < nr_pages)
2146 return pagevec_count(pvec);
2150 * find_get_pages_range - gang pagecache lookup
2151 * @mapping: The address_space to search
2152 * @start: The starting page index
2153 * @end: The final page index (inclusive)
2154 * @nr_pages: The maximum number of pages
2155 * @pages: Where the resulting pages are placed
2157 * find_get_pages_range() will search for and return a group of up to @nr_pages
2158 * pages in the mapping starting at index @start and up to index @end
2159 * (inclusive). The pages are placed at @pages. find_get_pages_range() takes
2160 * a reference against the returned pages.
2162 * The search returns a group of mapping-contiguous pages with ascending
2163 * indexes. There may be holes in the indices due to not-present pages.
2164 * We also update @start to index the next page for the traversal.
2166 * Return: the number of pages which were found. If this number is
2167 * smaller than @nr_pages, the end of specified range has been
2170 unsigned find_get_pages_range(struct address_space *mapping, pgoff_t *start,
2171 pgoff_t end, unsigned int nr_pages,
2172 struct page **pages)
2174 XA_STATE(xas, &mapping->i_pages, *start);
2178 if (unlikely(!nr_pages))
2182 while ((page = find_get_entry(&xas, end, XA_PRESENT))) {
2183 /* Skip over shadow, swap and DAX entries */
2184 if (xa_is_value(page))
2187 pages[ret] = find_subpage(page, xas.xa_index);
2188 if (++ret == nr_pages) {
2189 *start = xas.xa_index + 1;
2195 * We come here when there is no page beyond @end. We take care to not
2196 * overflow the index @start as it confuses some of the callers. This
2197 * breaks the iteration when there is a page at index -1 but that is
2198 * already broken anyway.
2200 if (end == (pgoff_t)-1)
2201 *start = (pgoff_t)-1;
2211 * find_get_pages_contig - gang contiguous pagecache lookup
2212 * @mapping: The address_space to search
2213 * @index: The starting page index
2214 * @nr_pages: The maximum number of pages
2215 * @pages: Where the resulting pages are placed
2217 * find_get_pages_contig() works exactly like find_get_pages(), except
2218 * that the returned number of pages are guaranteed to be contiguous.
2220 * Return: the number of pages which were found.
2222 unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index,
2223 unsigned int nr_pages, struct page **pages)
2225 XA_STATE(xas, &mapping->i_pages, index);
2227 unsigned int ret = 0;
2229 if (unlikely(!nr_pages))
2233 for (page = xas_load(&xas); page; page = xas_next(&xas)) {
2234 if (xas_retry(&xas, page))
2237 * If the entry has been swapped out, we can stop looking.
2238 * No current caller is looking for DAX entries.
2240 if (xa_is_value(page))
2243 if (!page_cache_get_speculative(page))
2246 /* Has the page moved or been split? */
2247 if (unlikely(page != xas_reload(&xas)))
2250 pages[ret] = find_subpage(page, xas.xa_index);
2251 if (++ret == nr_pages)
2262 EXPORT_SYMBOL(find_get_pages_contig);
2265 * find_get_pages_range_tag - Find and return head pages matching @tag.
2266 * @mapping: the address_space to search
2267 * @index: the starting page index
2268 * @end: The final page index (inclusive)
2269 * @tag: the tag index
2270 * @nr_pages: the maximum number of pages
2271 * @pages: where the resulting pages are placed
2273 * Like find_get_pages(), except we only return head pages which are tagged
2274 * with @tag. @index is updated to the index immediately after the last
2275 * page we return, ready for the next iteration.
2277 * Return: the number of pages which were found.
2279 unsigned find_get_pages_range_tag(struct address_space *mapping, pgoff_t *index,
2280 pgoff_t end, xa_mark_t tag, unsigned int nr_pages,
2281 struct page **pages)
2283 XA_STATE(xas, &mapping->i_pages, *index);
2287 if (unlikely(!nr_pages))
2291 while ((page = find_get_entry(&xas, end, tag))) {
2293 * Shadow entries should never be tagged, but this iteration
2294 * is lockless so there is a window for page reclaim to evict
2295 * a page we saw tagged. Skip over it.
2297 if (xa_is_value(page))
2301 if (++ret == nr_pages) {
2302 *index = page->index + thp_nr_pages(page);
2308 * We come here when we got to @end. We take care to not overflow the
2309 * index @index as it confuses some of the callers. This breaks the
2310 * iteration when there is a page at index -1 but that is already
2313 if (end == (pgoff_t)-1)
2314 *index = (pgoff_t)-1;
2322 EXPORT_SYMBOL(find_get_pages_range_tag);
2325 * CD/DVDs are error prone. When a medium error occurs, the driver may fail
2326 * a _large_ part of the i/o request. Imagine the worst scenario:
2328 * ---R__________________________________________B__________
2329 * ^ reading here ^ bad block(assume 4k)
2331 * read(R) => miss => readahead(R...B) => media error => frustrating retries
2332 * => failing the whole request => read(R) => read(R+1) =>
2333 * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
2334 * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
2335 * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
2337 * It is going insane. Fix it by quickly scaling down the readahead size.
2339 static void shrink_readahead_size_eio(struct file_ra_state *ra)
2345 * filemap_get_read_batch - Get a batch of pages for read
2347 * Get a batch of pages which represent a contiguous range of bytes
2348 * in the file. No tail pages will be returned. If @index is in the
2349 * middle of a THP, the entire THP will be returned. The last page in
2350 * the batch may have Readahead set or be not Uptodate so that the
2351 * caller can take the appropriate action.
2353 static void filemap_get_read_batch(struct address_space *mapping,
2354 pgoff_t index, pgoff_t max, struct pagevec *pvec)
2356 XA_STATE(xas, &mapping->i_pages, index);
2360 for (head = xas_load(&xas); head; head = xas_next(&xas)) {
2361 if (xas_retry(&xas, head))
2363 if (xas.xa_index > max || xa_is_value(head))
2365 if (!page_cache_get_speculative(head))
2368 /* Has the page moved or been split? */
2369 if (unlikely(head != xas_reload(&xas)))
2372 if (!pagevec_add(pvec, head))
2374 if (!PageUptodate(head))
2376 if (PageReadahead(head))
2378 xas.xa_index = head->index + thp_nr_pages(head) - 1;
2379 xas.xa_offset = (xas.xa_index >> xas.xa_shift) & XA_CHUNK_MASK;
2389 static int filemap_read_page(struct file *file, struct address_space *mapping,
2395 * A previous I/O error may have been due to temporary failures,
2396 * eg. multipath errors. PG_error will be set again if readpage
2399 ClearPageError(page);
2400 /* Start the actual read. The read will unlock the page. */
2401 error = mapping->a_ops->readpage(file, page);
2405 error = wait_on_page_locked_killable(page);
2408 if (PageUptodate(page))
2410 shrink_readahead_size_eio(&file->f_ra);
2414 static bool filemap_range_uptodate(struct address_space *mapping,
2415 loff_t pos, struct iov_iter *iter, struct page *page)
2419 if (PageUptodate(page))
2421 /* pipes can't handle partially uptodate pages */
2422 if (iov_iter_is_pipe(iter))
2424 if (!mapping->a_ops->is_partially_uptodate)
2426 if (mapping->host->i_blkbits >= (PAGE_SHIFT + thp_order(page)))
2429 count = iter->count;
2430 if (page_offset(page) > pos) {
2431 count -= page_offset(page) - pos;
2434 pos -= page_offset(page);
2437 return mapping->a_ops->is_partially_uptodate(page, pos, count);
2440 static int filemap_update_page(struct kiocb *iocb,
2441 struct address_space *mapping, struct iov_iter *iter,
2446 if (iocb->ki_flags & IOCB_NOWAIT) {
2447 if (!filemap_invalidate_trylock_shared(mapping))
2450 filemap_invalidate_lock_shared(mapping);
2453 if (!trylock_page(page)) {
2455 if (iocb->ki_flags & (IOCB_NOWAIT | IOCB_NOIO))
2456 goto unlock_mapping;
2457 if (!(iocb->ki_flags & IOCB_WAITQ)) {
2458 filemap_invalidate_unlock_shared(mapping);
2459 put_and_wait_on_page_locked(page, TASK_KILLABLE);
2460 return AOP_TRUNCATED_PAGE;
2462 error = __lock_page_async(page, iocb->ki_waitq);
2464 goto unlock_mapping;
2467 error = AOP_TRUNCATED_PAGE;
2472 if (filemap_range_uptodate(mapping, iocb->ki_pos, iter, page))
2476 if (iocb->ki_flags & (IOCB_NOIO | IOCB_NOWAIT | IOCB_WAITQ))
2479 error = filemap_read_page(iocb->ki_filp, mapping, page);
2480 goto unlock_mapping;
2484 filemap_invalidate_unlock_shared(mapping);
2485 if (error == AOP_TRUNCATED_PAGE)
2490 static int filemap_create_page(struct file *file,
2491 struct address_space *mapping, pgoff_t index,
2492 struct pagevec *pvec)
2497 page = page_cache_alloc(mapping);
2502 * Protect against truncate / hole punch. Grabbing invalidate_lock here
2503 * assures we cannot instantiate and bring uptodate new pagecache pages
2504 * after evicting page cache during truncate and before actually
2505 * freeing blocks. Note that we could release invalidate_lock after
2506 * inserting the page into page cache as the locked page would then be
2507 * enough to synchronize with hole punching. But there are code paths
2508 * such as filemap_update_page() filling in partially uptodate pages or
2509 * ->readpages() that need to hold invalidate_lock while mapping blocks
2510 * for IO so let's hold the lock here as well to keep locking rules
2513 filemap_invalidate_lock_shared(mapping);
2514 error = add_to_page_cache_lru(page, mapping, index,
2515 mapping_gfp_constraint(mapping, GFP_KERNEL));
2516 if (error == -EEXIST)
2517 error = AOP_TRUNCATED_PAGE;
2521 error = filemap_read_page(file, mapping, page);
2525 filemap_invalidate_unlock_shared(mapping);
2526 pagevec_add(pvec, page);
2529 filemap_invalidate_unlock_shared(mapping);
2534 static int filemap_readahead(struct kiocb *iocb, struct file *file,
2535 struct address_space *mapping, struct page *page,
2538 if (iocb->ki_flags & IOCB_NOIO)
2540 page_cache_async_readahead(mapping, &file->f_ra, file, page,
2541 page->index, last_index - page->index);
2545 static int filemap_get_pages(struct kiocb *iocb, struct iov_iter *iter,
2546 struct pagevec *pvec)
2548 struct file *filp = iocb->ki_filp;
2549 struct address_space *mapping = filp->f_mapping;
2550 struct file_ra_state *ra = &filp->f_ra;
2551 pgoff_t index = iocb->ki_pos >> PAGE_SHIFT;
2556 last_index = DIV_ROUND_UP(iocb->ki_pos + iter->count, PAGE_SIZE);
2558 if (fatal_signal_pending(current))
2561 filemap_get_read_batch(mapping, index, last_index, pvec);
2562 if (!pagevec_count(pvec)) {
2563 if (iocb->ki_flags & IOCB_NOIO)
2565 page_cache_sync_readahead(mapping, ra, filp, index,
2566 last_index - index);
2567 filemap_get_read_batch(mapping, index, last_index, pvec);
2569 if (!pagevec_count(pvec)) {
2570 if (iocb->ki_flags & (IOCB_NOWAIT | IOCB_WAITQ))
2572 err = filemap_create_page(filp, mapping,
2573 iocb->ki_pos >> PAGE_SHIFT, pvec);
2574 if (err == AOP_TRUNCATED_PAGE)
2579 page = pvec->pages[pagevec_count(pvec) - 1];
2580 if (PageReadahead(page)) {
2581 err = filemap_readahead(iocb, filp, mapping, page, last_index);
2585 if (!PageUptodate(page)) {
2586 if ((iocb->ki_flags & IOCB_WAITQ) && pagevec_count(pvec) > 1)
2587 iocb->ki_flags |= IOCB_NOWAIT;
2588 err = filemap_update_page(iocb, mapping, iter, page);
2597 if (likely(--pvec->nr))
2599 if (err == AOP_TRUNCATED_PAGE)
2605 * filemap_read - Read data from the page cache.
2606 * @iocb: The iocb to read.
2607 * @iter: Destination for the data.
2608 * @already_read: Number of bytes already read by the caller.
2610 * Copies data from the page cache. If the data is not currently present,
2611 * uses the readahead and readpage address_space operations to fetch it.
2613 * Return: Total number of bytes copied, including those already read by
2614 * the caller. If an error happens before any bytes are copied, returns
2615 * a negative error number.
2617 ssize_t filemap_read(struct kiocb *iocb, struct iov_iter *iter,
2618 ssize_t already_read)
2620 struct file *filp = iocb->ki_filp;
2621 struct file_ra_state *ra = &filp->f_ra;
2622 struct address_space *mapping = filp->f_mapping;
2623 struct inode *inode = mapping->host;
2624 struct pagevec pvec;
2626 bool writably_mapped;
2627 loff_t isize, end_offset;
2629 if (unlikely(iocb->ki_pos >= inode->i_sb->s_maxbytes))
2631 if (unlikely(!iov_iter_count(iter)))
2634 iov_iter_truncate(iter, inode->i_sb->s_maxbytes);
2635 pagevec_init(&pvec);
2641 * If we've already successfully copied some data, then we
2642 * can no longer safely return -EIOCBQUEUED. Hence mark
2643 * an async read NOWAIT at that point.
2645 if ((iocb->ki_flags & IOCB_WAITQ) && already_read)
2646 iocb->ki_flags |= IOCB_NOWAIT;
2648 error = filemap_get_pages(iocb, iter, &pvec);
2653 * i_size must be checked after we know the pages are Uptodate.
2655 * Checking i_size after the check allows us to calculate
2656 * the correct value for "nr", which means the zero-filled
2657 * part of the page is not copied back to userspace (unless
2658 * another truncate extends the file - this is desired though).
2660 isize = i_size_read(inode);
2661 if (unlikely(iocb->ki_pos >= isize))
2663 end_offset = min_t(loff_t, isize, iocb->ki_pos + iter->count);
2666 * Once we start copying data, we don't want to be touching any
2667 * cachelines that might be contended:
2669 writably_mapped = mapping_writably_mapped(mapping);
2672 * When a sequential read accesses a page several times, only
2673 * mark it as accessed the first time.
2675 if (iocb->ki_pos >> PAGE_SHIFT !=
2676 ra->prev_pos >> PAGE_SHIFT)
2677 mark_page_accessed(pvec.pages[0]);
2679 for (i = 0; i < pagevec_count(&pvec); i++) {
2680 struct page *page = pvec.pages[i];
2681 size_t page_size = thp_size(page);
2682 size_t offset = iocb->ki_pos & (page_size - 1);
2683 size_t bytes = min_t(loff_t, end_offset - iocb->ki_pos,
2684 page_size - offset);
2687 if (end_offset < page_offset(page))
2690 mark_page_accessed(page);
2692 * If users can be writing to this page using arbitrary
2693 * virtual addresses, take care about potential aliasing
2694 * before reading the page on the kernel side.
2696 if (writably_mapped) {
2699 for (j = 0; j < thp_nr_pages(page); j++)
2700 flush_dcache_page(page + j);
2703 copied = copy_page_to_iter(page, offset, bytes, iter);
2705 already_read += copied;
2706 iocb->ki_pos += copied;
2707 ra->prev_pos = iocb->ki_pos;
2709 if (copied < bytes) {
2715 for (i = 0; i < pagevec_count(&pvec); i++)
2716 put_page(pvec.pages[i]);
2717 pagevec_reinit(&pvec);
2718 } while (iov_iter_count(iter) && iocb->ki_pos < isize && !error);
2720 file_accessed(filp);
2722 return already_read ? already_read : error;
2724 EXPORT_SYMBOL_GPL(filemap_read);
2727 * generic_file_read_iter - generic filesystem read routine
2728 * @iocb: kernel I/O control block
2729 * @iter: destination for the data read
2731 * This is the "read_iter()" routine for all filesystems
2732 * that can use the page cache directly.
2734 * The IOCB_NOWAIT flag in iocb->ki_flags indicates that -EAGAIN shall
2735 * be returned when no data can be read without waiting for I/O requests
2736 * to complete; it doesn't prevent readahead.
2738 * The IOCB_NOIO flag in iocb->ki_flags indicates that no new I/O
2739 * requests shall be made for the read or for readahead. When no data
2740 * can be read, -EAGAIN shall be returned. When readahead would be
2741 * triggered, a partial, possibly empty read shall be returned.
2744 * * number of bytes copied, even for partial reads
2745 * * negative error code (or 0 if IOCB_NOIO) if nothing was read
2748 generic_file_read_iter(struct kiocb *iocb, struct iov_iter *iter)
2750 size_t count = iov_iter_count(iter);
2754 return 0; /* skip atime */
2756 if (iocb->ki_flags & IOCB_DIRECT) {
2757 struct file *file = iocb->ki_filp;
2758 struct address_space *mapping = file->f_mapping;
2759 struct inode *inode = mapping->host;
2762 size = i_size_read(inode);
2763 if (iocb->ki_flags & IOCB_NOWAIT) {
2764 if (filemap_range_needs_writeback(mapping, iocb->ki_pos,
2765 iocb->ki_pos + count - 1))
2768 retval = filemap_write_and_wait_range(mapping,
2770 iocb->ki_pos + count - 1);
2775 file_accessed(file);
2777 retval = mapping->a_ops->direct_IO(iocb, iter);
2779 iocb->ki_pos += retval;
2782 if (retval != -EIOCBQUEUED)
2783 iov_iter_revert(iter, count - iov_iter_count(iter));
2786 * Btrfs can have a short DIO read if we encounter
2787 * compressed extents, so if there was an error, or if
2788 * we've already read everything we wanted to, or if
2789 * there was a short read because we hit EOF, go ahead
2790 * and return. Otherwise fallthrough to buffered io for
2791 * the rest of the read. Buffered reads will not work for
2792 * DAX files, so don't bother trying.
2794 if (retval < 0 || !count || iocb->ki_pos >= size ||
2799 return filemap_read(iocb, iter, retval);
2801 EXPORT_SYMBOL(generic_file_read_iter);
2803 static inline loff_t page_seek_hole_data(struct xa_state *xas,
2804 struct address_space *mapping, struct page *page,
2805 loff_t start, loff_t end, bool seek_data)
2807 const struct address_space_operations *ops = mapping->a_ops;
2808 size_t offset, bsz = i_blocksize(mapping->host);
2810 if (xa_is_value(page) || PageUptodate(page))
2811 return seek_data ? start : end;
2812 if (!ops->is_partially_uptodate)
2813 return seek_data ? end : start;
2818 if (unlikely(page->mapping != mapping))
2821 offset = offset_in_thp(page, start) & ~(bsz - 1);
2824 if (ops->is_partially_uptodate(page, offset, bsz) == seek_data)
2826 start = (start + bsz) & ~(bsz - 1);
2828 } while (offset < thp_size(page));
2836 unsigned int seek_page_size(struct xa_state *xas, struct page *page)
2838 if (xa_is_value(page))
2839 return PAGE_SIZE << xa_get_order(xas->xa, xas->xa_index);
2840 return thp_size(page);
2844 * mapping_seek_hole_data - Seek for SEEK_DATA / SEEK_HOLE in the page cache.
2845 * @mapping: Address space to search.
2846 * @start: First byte to consider.
2847 * @end: Limit of search (exclusive).
2848 * @whence: Either SEEK_HOLE or SEEK_DATA.
2850 * If the page cache knows which blocks contain holes and which blocks
2851 * contain data, your filesystem can use this function to implement
2852 * SEEK_HOLE and SEEK_DATA. This is useful for filesystems which are
2853 * entirely memory-based such as tmpfs, and filesystems which support
2854 * unwritten extents.
2856 * Return: The requested offset on success, or -ENXIO if @whence specifies
2857 * SEEK_DATA and there is no data after @start. There is an implicit hole
2858 * after @end - 1, so SEEK_HOLE returns @end if all the bytes between @start
2859 * and @end contain data.
2861 loff_t mapping_seek_hole_data(struct address_space *mapping, loff_t start,
2862 loff_t end, int whence)
2864 XA_STATE(xas, &mapping->i_pages, start >> PAGE_SHIFT);
2865 pgoff_t max = (end - 1) >> PAGE_SHIFT;
2866 bool seek_data = (whence == SEEK_DATA);
2873 while ((page = find_get_entry(&xas, max, XA_PRESENT))) {
2874 loff_t pos = (u64)xas.xa_index << PAGE_SHIFT;
2875 unsigned int seek_size;
2883 seek_size = seek_page_size(&xas, page);
2884 pos = round_up(pos + 1, seek_size);
2885 start = page_seek_hole_data(&xas, mapping, page, start, pos,
2891 if (seek_size > PAGE_SIZE)
2892 xas_set(&xas, pos >> PAGE_SHIFT);
2893 if (!xa_is_value(page))
2900 if (page && !xa_is_value(page))
2908 #define MMAP_LOTSAMISS (100)
2910 * lock_page_maybe_drop_mmap - lock the page, possibly dropping the mmap_lock
2911 * @vmf - the vm_fault for this fault.
2912 * @page - the page to lock.
2913 * @fpin - the pointer to the file we may pin (or is already pinned).
2915 * This works similar to lock_page_or_retry in that it can drop the mmap_lock.
2916 * It differs in that it actually returns the page locked if it returns 1 and 0
2917 * if it couldn't lock the page. If we did have to drop the mmap_lock then fpin
2918 * will point to the pinned file and needs to be fput()'ed at a later point.
2920 static int lock_page_maybe_drop_mmap(struct vm_fault *vmf, struct page *page,
2923 if (trylock_page(page))
2927 * NOTE! This will make us return with VM_FAULT_RETRY, but with
2928 * the mmap_lock still held. That's how FAULT_FLAG_RETRY_NOWAIT
2929 * is supposed to work. We have way too many special cases..
2931 if (vmf->flags & FAULT_FLAG_RETRY_NOWAIT)
2934 *fpin = maybe_unlock_mmap_for_io(vmf, *fpin);
2935 if (vmf->flags & FAULT_FLAG_KILLABLE) {
2936 if (__lock_page_killable(page)) {
2938 * We didn't have the right flags to drop the mmap_lock,
2939 * but all fault_handlers only check for fatal signals
2940 * if we return VM_FAULT_RETRY, so we need to drop the
2941 * mmap_lock here and return 0 if we don't have a fpin.
2944 mmap_read_unlock(vmf->vma->vm_mm);
2954 * Synchronous readahead happens when we don't even find a page in the page
2955 * cache at all. We don't want to perform IO under the mmap sem, so if we have
2956 * to drop the mmap sem we return the file that was pinned in order for us to do
2957 * that. If we didn't pin a file then we return NULL. The file that is
2958 * returned needs to be fput()'ed when we're done with it.
2960 static struct file *do_sync_mmap_readahead(struct vm_fault *vmf)
2962 struct file *file = vmf->vma->vm_file;
2963 struct file_ra_state *ra = &file->f_ra;
2964 struct address_space *mapping = file->f_mapping;
2965 DEFINE_READAHEAD(ractl, file, ra, mapping, vmf->pgoff);
2966 struct file *fpin = NULL;
2967 unsigned int mmap_miss;
2969 /* If we don't want any read-ahead, don't bother */
2970 if (vmf->vma->vm_flags & VM_RAND_READ)
2975 if (vmf->vma->vm_flags & VM_SEQ_READ) {
2976 fpin = maybe_unlock_mmap_for_io(vmf, fpin);
2977 page_cache_sync_ra(&ractl, ra->ra_pages);
2981 /* Avoid banging the cache line if not needed */
2982 mmap_miss = READ_ONCE(ra->mmap_miss);
2983 if (mmap_miss < MMAP_LOTSAMISS * 10)
2984 WRITE_ONCE(ra->mmap_miss, ++mmap_miss);
2987 * Do we miss much more than hit in this file? If so,
2988 * stop bothering with read-ahead. It will only hurt.
2990 if (mmap_miss > MMAP_LOTSAMISS)
2996 fpin = maybe_unlock_mmap_for_io(vmf, fpin);
2997 ra->start = max_t(long, 0, vmf->pgoff - ra->ra_pages / 2);
2998 ra->size = ra->ra_pages;
2999 ra->async_size = ra->ra_pages / 4;
3000 ractl._index = ra->start;
3001 do_page_cache_ra(&ractl, ra->size, ra->async_size);
3006 * Asynchronous readahead happens when we find the page and PG_readahead,
3007 * so we want to possibly extend the readahead further. We return the file that
3008 * was pinned if we have to drop the mmap_lock in order to do IO.
3010 static struct file *do_async_mmap_readahead(struct vm_fault *vmf,
3013 struct file *file = vmf->vma->vm_file;
3014 struct file_ra_state *ra = &file->f_ra;
3015 struct address_space *mapping = file->f_mapping;
3016 struct file *fpin = NULL;
3017 unsigned int mmap_miss;
3018 pgoff_t offset = vmf->pgoff;
3020 /* If we don't want any read-ahead, don't bother */
3021 if (vmf->vma->vm_flags & VM_RAND_READ || !ra->ra_pages)
3023 mmap_miss = READ_ONCE(ra->mmap_miss);
3025 WRITE_ONCE(ra->mmap_miss, --mmap_miss);
3026 if (PageReadahead(page)) {
3027 fpin = maybe_unlock_mmap_for_io(vmf, fpin);
3028 page_cache_async_readahead(mapping, ra, file,
3029 page, offset, ra->ra_pages);
3035 * filemap_fault - read in file data for page fault handling
3036 * @vmf: struct vm_fault containing details of the fault
3038 * filemap_fault() is invoked via the vma operations vector for a
3039 * mapped memory region to read in file data during a page fault.
3041 * The goto's are kind of ugly, but this streamlines the normal case of having
3042 * it in the page cache, and handles the special cases reasonably without
3043 * having a lot of duplicated code.
3045 * vma->vm_mm->mmap_lock must be held on entry.
3047 * If our return value has VM_FAULT_RETRY set, it's because the mmap_lock
3048 * may be dropped before doing I/O or by lock_page_maybe_drop_mmap().
3050 * If our return value does not have VM_FAULT_RETRY set, the mmap_lock
3051 * has not been released.
3053 * We never return with VM_FAULT_RETRY and a bit from VM_FAULT_ERROR set.
3055 * Return: bitwise-OR of %VM_FAULT_ codes.
3057 vm_fault_t filemap_fault(struct vm_fault *vmf)
3060 struct file *file = vmf->vma->vm_file;
3061 struct file *fpin = NULL;
3062 struct address_space *mapping = file->f_mapping;
3063 struct inode *inode = mapping->host;
3064 pgoff_t offset = vmf->pgoff;
3068 bool mapping_locked = false;
3070 max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
3071 if (unlikely(offset >= max_off))
3072 return VM_FAULT_SIGBUS;
3075 * Do we have something in the page cache already?
3077 page = find_get_page(mapping, offset);
3080 * We found the page, so try async readahead before waiting for
3083 if (!(vmf->flags & FAULT_FLAG_TRIED))
3084 fpin = do_async_mmap_readahead(vmf, page);
3085 if (unlikely(!PageUptodate(page))) {
3086 filemap_invalidate_lock_shared(mapping);
3087 mapping_locked = true;
3090 /* No page in the page cache at all */
3091 count_vm_event(PGMAJFAULT);
3092 count_memcg_event_mm(vmf->vma->vm_mm, PGMAJFAULT);
3093 ret = VM_FAULT_MAJOR;
3094 fpin = do_sync_mmap_readahead(vmf);
3097 * See comment in filemap_create_page() why we need
3100 if (!mapping_locked) {
3101 filemap_invalidate_lock_shared(mapping);
3102 mapping_locked = true;
3104 page = pagecache_get_page(mapping, offset,
3105 FGP_CREAT|FGP_FOR_MMAP,
3110 filemap_invalidate_unlock_shared(mapping);
3111 return VM_FAULT_OOM;
3115 if (!lock_page_maybe_drop_mmap(vmf, page, &fpin))
3118 /* Did it get truncated? */
3119 if (unlikely(compound_head(page)->mapping != mapping)) {
3124 VM_BUG_ON_PAGE(page_to_pgoff(page) != offset, page);
3127 * We have a locked page in the page cache, now we need to check
3128 * that it's up-to-date. If not, it is going to be due to an error.
3130 if (unlikely(!PageUptodate(page))) {
3132 * The page was in cache and uptodate and now it is not.
3133 * Strange but possible since we didn't hold the page lock all
3134 * the time. Let's drop everything get the invalidate lock and
3137 if (!mapping_locked) {
3142 goto page_not_uptodate;
3146 * We've made it this far and we had to drop our mmap_lock, now is the
3147 * time to return to the upper layer and have it re-find the vma and
3155 filemap_invalidate_unlock_shared(mapping);
3158 * Found the page and have a reference on it.
3159 * We must recheck i_size under page lock.
3161 max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
3162 if (unlikely(offset >= max_off)) {
3165 return VM_FAULT_SIGBUS;
3169 return ret | VM_FAULT_LOCKED;
3173 * Umm, take care of errors if the page isn't up-to-date.
3174 * Try to re-read it _once_. We do this synchronously,
3175 * because there really aren't any performance issues here
3176 * and we need to check for errors.
3178 fpin = maybe_unlock_mmap_for_io(vmf, fpin);
3179 error = filemap_read_page(file, mapping, page);
3184 if (!error || error == AOP_TRUNCATED_PAGE)
3186 filemap_invalidate_unlock_shared(mapping);
3188 return VM_FAULT_SIGBUS;
3192 * We dropped the mmap_lock, we need to return to the fault handler to
3193 * re-find the vma and come back and find our hopefully still populated
3199 filemap_invalidate_unlock_shared(mapping);
3202 return ret | VM_FAULT_RETRY;
3204 EXPORT_SYMBOL(filemap_fault);
3206 static bool filemap_map_pmd(struct vm_fault *vmf, struct page *page)
3208 struct mm_struct *mm = vmf->vma->vm_mm;
3210 /* Huge page is mapped? No need to proceed. */
3211 if (pmd_trans_huge(*vmf->pmd)) {
3217 if (pmd_none(*vmf->pmd) && PageTransHuge(page)) {
3218 vm_fault_t ret = do_set_pmd(vmf, page);
3220 /* The page is mapped successfully, reference consumed. */
3226 if (pmd_none(*vmf->pmd)) {
3227 vmf->ptl = pmd_lock(mm, vmf->pmd);
3228 if (likely(pmd_none(*vmf->pmd))) {
3230 pmd_populate(mm, vmf->pmd, vmf->prealloc_pte);
3231 vmf->prealloc_pte = NULL;
3233 spin_unlock(vmf->ptl);
3236 /* See comment in handle_pte_fault() */
3237 if (pmd_devmap_trans_unstable(vmf->pmd)) {
3246 static struct page *next_uptodate_page(struct page *page,
3247 struct address_space *mapping,
3248 struct xa_state *xas, pgoff_t end_pgoff)
3250 unsigned long max_idx;
3255 if (xas_retry(xas, page))
3257 if (xa_is_value(page))
3259 if (PageLocked(page))
3261 if (!page_cache_get_speculative(page))
3263 /* Has the page moved or been split? */
3264 if (unlikely(page != xas_reload(xas)))
3266 if (!PageUptodate(page) || PageReadahead(page))
3268 if (PageHWPoison(page))
3270 if (!trylock_page(page))
3272 if (page->mapping != mapping)
3274 if (!PageUptodate(page))
3276 max_idx = DIV_ROUND_UP(i_size_read(mapping->host), PAGE_SIZE);
3277 if (xas->xa_index >= max_idx)
3284 } while ((page = xas_next_entry(xas, end_pgoff)) != NULL);
3289 static inline struct page *first_map_page(struct address_space *mapping,
3290 struct xa_state *xas,
3293 return next_uptodate_page(xas_find(xas, end_pgoff),
3294 mapping, xas, end_pgoff);
3297 static inline struct page *next_map_page(struct address_space *mapping,
3298 struct xa_state *xas,
3301 return next_uptodate_page(xas_next_entry(xas, end_pgoff),
3302 mapping, xas, end_pgoff);
3305 vm_fault_t filemap_map_pages(struct vm_fault *vmf,
3306 pgoff_t start_pgoff, pgoff_t end_pgoff)
3308 struct vm_area_struct *vma = vmf->vma;
3309 struct file *file = vma->vm_file;
3310 struct address_space *mapping = file->f_mapping;
3311 pgoff_t last_pgoff = start_pgoff;
3313 XA_STATE(xas, &mapping->i_pages, start_pgoff);
3314 struct page *head, *page;
3315 unsigned int mmap_miss = READ_ONCE(file->f_ra.mmap_miss);
3319 head = first_map_page(mapping, &xas, end_pgoff);
3323 if (filemap_map_pmd(vmf, head)) {
3324 ret = VM_FAULT_NOPAGE;
3328 addr = vma->vm_start + ((start_pgoff - vma->vm_pgoff) << PAGE_SHIFT);
3329 vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, addr, &vmf->ptl);
3331 page = find_subpage(head, xas.xa_index);
3332 if (PageHWPoison(page))
3338 addr += (xas.xa_index - last_pgoff) << PAGE_SHIFT;
3339 vmf->pte += xas.xa_index - last_pgoff;
3340 last_pgoff = xas.xa_index;
3342 if (!pte_none(*vmf->pte))
3345 /* We're about to handle the fault */
3346 if (vmf->address == addr)
3347 ret = VM_FAULT_NOPAGE;
3349 do_set_pte(vmf, page, addr);
3350 /* no need to invalidate: a not-present page won't be cached */
3351 update_mmu_cache(vma, addr, vmf->pte);
3357 } while ((head = next_map_page(mapping, &xas, end_pgoff)) != NULL);
3358 pte_unmap_unlock(vmf->pte, vmf->ptl);
3361 WRITE_ONCE(file->f_ra.mmap_miss, mmap_miss);
3364 EXPORT_SYMBOL(filemap_map_pages);
3366 vm_fault_t filemap_page_mkwrite(struct vm_fault *vmf)
3368 struct address_space *mapping = vmf->vma->vm_file->f_mapping;
3369 struct page *page = vmf->page;
3370 vm_fault_t ret = VM_FAULT_LOCKED;
3372 sb_start_pagefault(mapping->host->i_sb);
3373 file_update_time(vmf->vma->vm_file);
3375 if (page->mapping != mapping) {
3377 ret = VM_FAULT_NOPAGE;
3381 * We mark the page dirty already here so that when freeze is in
3382 * progress, we are guaranteed that writeback during freezing will
3383 * see the dirty page and writeprotect it again.
3385 set_page_dirty(page);
3386 wait_for_stable_page(page);
3388 sb_end_pagefault(mapping->host->i_sb);
3392 const struct vm_operations_struct generic_file_vm_ops = {
3393 .fault = filemap_fault,
3394 .map_pages = filemap_map_pages,
3395 .page_mkwrite = filemap_page_mkwrite,
3398 /* This is used for a general mmap of a disk file */
3400 int generic_file_mmap(struct file *file, struct vm_area_struct *vma)
3402 struct address_space *mapping = file->f_mapping;
3404 if (!mapping->a_ops->readpage)
3406 file_accessed(file);
3407 vma->vm_ops = &generic_file_vm_ops;
3412 * This is for filesystems which do not implement ->writepage.
3414 int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
3416 if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE))
3418 return generic_file_mmap(file, vma);
3421 vm_fault_t filemap_page_mkwrite(struct vm_fault *vmf)
3423 return VM_FAULT_SIGBUS;
3425 int generic_file_mmap(struct file *file, struct vm_area_struct *vma)
3429 int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
3433 #endif /* CONFIG_MMU */
3435 EXPORT_SYMBOL(filemap_page_mkwrite);
3436 EXPORT_SYMBOL(generic_file_mmap);
3437 EXPORT_SYMBOL(generic_file_readonly_mmap);
3439 static struct page *wait_on_page_read(struct page *page)
3441 if (!IS_ERR(page)) {
3442 wait_on_page_locked(page);
3443 if (!PageUptodate(page)) {
3445 page = ERR_PTR(-EIO);
3451 static struct page *do_read_cache_page(struct address_space *mapping,
3453 int (*filler)(void *, struct page *),
3460 page = find_get_page(mapping, index);
3462 page = __page_cache_alloc(gfp);
3464 return ERR_PTR(-ENOMEM);
3465 err = add_to_page_cache_lru(page, mapping, index, gfp);
3466 if (unlikely(err)) {
3470 /* Presumably ENOMEM for xarray node */
3471 return ERR_PTR(err);
3476 err = filler(data, page);
3478 err = mapping->a_ops->readpage(data, page);
3482 return ERR_PTR(err);
3485 page = wait_on_page_read(page);
3490 if (PageUptodate(page))
3494 * Page is not up to date and may be locked due to one of the following
3495 * case a: Page is being filled and the page lock is held
3496 * case b: Read/write error clearing the page uptodate status
3497 * case c: Truncation in progress (page locked)
3498 * case d: Reclaim in progress
3500 * Case a, the page will be up to date when the page is unlocked.
3501 * There is no need to serialise on the page lock here as the page
3502 * is pinned so the lock gives no additional protection. Even if the
3503 * page is truncated, the data is still valid if PageUptodate as
3504 * it's a race vs truncate race.
3505 * Case b, the page will not be up to date
3506 * Case c, the page may be truncated but in itself, the data may still
3507 * be valid after IO completes as it's a read vs truncate race. The
3508 * operation must restart if the page is not uptodate on unlock but
3509 * otherwise serialising on page lock to stabilise the mapping gives
3510 * no additional guarantees to the caller as the page lock is
3511 * released before return.
3512 * Case d, similar to truncation. If reclaim holds the page lock, it
3513 * will be a race with remove_mapping that determines if the mapping
3514 * is valid on unlock but otherwise the data is valid and there is
3515 * no need to serialise with page lock.
3517 * As the page lock gives no additional guarantee, we optimistically
3518 * wait on the page to be unlocked and check if it's up to date and
3519 * use the page if it is. Otherwise, the page lock is required to
3520 * distinguish between the different cases. The motivation is that we
3521 * avoid spurious serialisations and wakeups when multiple processes
3522 * wait on the same page for IO to complete.
3524 wait_on_page_locked(page);
3525 if (PageUptodate(page))
3528 /* Distinguish between all the cases under the safety of the lock */
3531 /* Case c or d, restart the operation */
3532 if (!page->mapping) {
3538 /* Someone else locked and filled the page in a very small window */
3539 if (PageUptodate(page)) {
3545 * A previous I/O error may have been due to temporary
3547 * Clear page error before actual read, PG_error will be
3548 * set again if read page fails.
3550 ClearPageError(page);
3554 mark_page_accessed(page);
3559 * read_cache_page - read into page cache, fill it if needed
3560 * @mapping: the page's address_space
3561 * @index: the page index
3562 * @filler: function to perform the read
3563 * @data: first arg to filler(data, page) function, often left as NULL
3565 * Read into the page cache. If a page already exists, and PageUptodate() is
3566 * not set, try to fill the page and wait for it to become unlocked.
3568 * If the page does not get brought uptodate, return -EIO.
3570 * The function expects mapping->invalidate_lock to be already held.
3572 * Return: up to date page on success, ERR_PTR() on failure.
3574 struct page *read_cache_page(struct address_space *mapping,
3576 int (*filler)(void *, struct page *),
3579 return do_read_cache_page(mapping, index, filler, data,
3580 mapping_gfp_mask(mapping));
3582 EXPORT_SYMBOL(read_cache_page);
3585 * read_cache_page_gfp - read into page cache, using specified page allocation flags.
3586 * @mapping: the page's address_space
3587 * @index: the page index
3588 * @gfp: the page allocator flags to use if allocating
3590 * This is the same as "read_mapping_page(mapping, index, NULL)", but with
3591 * any new page allocations done using the specified allocation flags.
3593 * If the page does not get brought uptodate, return -EIO.
3595 * The function expects mapping->invalidate_lock to be already held.
3597 * Return: up to date page on success, ERR_PTR() on failure.
3599 struct page *read_cache_page_gfp(struct address_space *mapping,
3603 return do_read_cache_page(mapping, index, NULL, NULL, gfp);
3605 EXPORT_SYMBOL(read_cache_page_gfp);
3607 int pagecache_write_begin(struct file *file, struct address_space *mapping,
3608 loff_t pos, unsigned len, unsigned flags,
3609 struct page **pagep, void **fsdata)
3611 const struct address_space_operations *aops = mapping->a_ops;
3613 return aops->write_begin(file, mapping, pos, len, flags,
3616 EXPORT_SYMBOL(pagecache_write_begin);
3618 int pagecache_write_end(struct file *file, struct address_space *mapping,
3619 loff_t pos, unsigned len, unsigned copied,
3620 struct page *page, void *fsdata)
3622 const struct address_space_operations *aops = mapping->a_ops;
3624 return aops->write_end(file, mapping, pos, len, copied, page, fsdata);
3626 EXPORT_SYMBOL(pagecache_write_end);
3629 * Warn about a page cache invalidation failure during a direct I/O write.
3631 void dio_warn_stale_pagecache(struct file *filp)
3633 static DEFINE_RATELIMIT_STATE(_rs, 86400 * HZ, DEFAULT_RATELIMIT_BURST);
3637 errseq_set(&filp->f_mapping->wb_err, -EIO);
3638 if (__ratelimit(&_rs)) {
3639 path = file_path(filp, pathname, sizeof(pathname));
3642 pr_crit("Page cache invalidation failure on direct I/O. Possible data corruption due to collision with buffered I/O!\n");
3643 pr_crit("File: %s PID: %d Comm: %.20s\n", path, current->pid,
3649 generic_file_direct_write(struct kiocb *iocb, struct iov_iter *from)
3651 struct file *file = iocb->ki_filp;
3652 struct address_space *mapping = file->f_mapping;
3653 struct inode *inode = mapping->host;
3654 loff_t pos = iocb->ki_pos;
3659 write_len = iov_iter_count(from);
3660 end = (pos + write_len - 1) >> PAGE_SHIFT;
3662 if (iocb->ki_flags & IOCB_NOWAIT) {
3663 /* If there are pages to writeback, return */
3664 if (filemap_range_has_page(file->f_mapping, pos,
3665 pos + write_len - 1))
3668 written = filemap_write_and_wait_range(mapping, pos,
3669 pos + write_len - 1);
3675 * After a write we want buffered reads to be sure to go to disk to get
3676 * the new data. We invalidate clean cached page from the region we're
3677 * about to write. We do this *before* the write so that we can return
3678 * without clobbering -EIOCBQUEUED from ->direct_IO().
3680 written = invalidate_inode_pages2_range(mapping,
3681 pos >> PAGE_SHIFT, end);
3683 * If a page can not be invalidated, return 0 to fall back
3684 * to buffered write.
3687 if (written == -EBUSY)
3692 written = mapping->a_ops->direct_IO(iocb, from);
3695 * Finally, try again to invalidate clean pages which might have been
3696 * cached by non-direct readahead, or faulted in by get_user_pages()
3697 * if the source of the write was an mmap'ed region of the file
3698 * we're writing. Either one is a pretty crazy thing to do,
3699 * so we don't support it 100%. If this invalidation
3700 * fails, tough, the write still worked...
3702 * Most of the time we do not need this since dio_complete() will do
3703 * the invalidation for us. However there are some file systems that
3704 * do not end up with dio_complete() being called, so let's not break
3705 * them by removing it completely.
3707 * Noticeable example is a blkdev_direct_IO().
3709 * Skip invalidation for async writes or if mapping has no pages.
3711 if (written > 0 && mapping->nrpages &&
3712 invalidate_inode_pages2_range(mapping, pos >> PAGE_SHIFT, end))
3713 dio_warn_stale_pagecache(file);
3717 write_len -= written;
3718 if (pos > i_size_read(inode) && !S_ISBLK(inode->i_mode)) {
3719 i_size_write(inode, pos);
3720 mark_inode_dirty(inode);
3724 if (written != -EIOCBQUEUED)
3725 iov_iter_revert(from, write_len - iov_iter_count(from));
3729 EXPORT_SYMBOL(generic_file_direct_write);
3732 * Find or create a page at the given pagecache position. Return the locked
3733 * page. This function is specifically for buffered writes.
3735 struct page *grab_cache_page_write_begin(struct address_space *mapping,
3736 pgoff_t index, unsigned flags)
3739 int fgp_flags = FGP_LOCK|FGP_WRITE|FGP_CREAT;
3741 if (flags & AOP_FLAG_NOFS)
3742 fgp_flags |= FGP_NOFS;
3744 page = pagecache_get_page(mapping, index, fgp_flags,
3745 mapping_gfp_mask(mapping));
3747 wait_for_stable_page(page);
3751 EXPORT_SYMBOL(grab_cache_page_write_begin);
3753 ssize_t generic_perform_write(struct file *file,
3754 struct iov_iter *i, loff_t pos)
3756 struct address_space *mapping = file->f_mapping;
3757 const struct address_space_operations *a_ops = mapping->a_ops;
3759 ssize_t written = 0;
3760 unsigned int flags = 0;
3764 unsigned long offset; /* Offset into pagecache page */
3765 unsigned long bytes; /* Bytes to write to page */
3766 size_t copied; /* Bytes copied from user */
3769 offset = (pos & (PAGE_SIZE - 1));
3770 bytes = min_t(unsigned long, PAGE_SIZE - offset,
3775 * Bring in the user page that we will copy from _first_.
3776 * Otherwise there's a nasty deadlock on copying from the
3777 * same page as we're writing to, without it being marked
3780 if (unlikely(iov_iter_fault_in_readable(i, bytes))) {
3785 if (fatal_signal_pending(current)) {
3790 status = a_ops->write_begin(file, mapping, pos, bytes, flags,
3792 if (unlikely(status < 0))
3795 if (mapping_writably_mapped(mapping))
3796 flush_dcache_page(page);
3798 copied = copy_page_from_iter_atomic(page, offset, bytes, i);
3799 flush_dcache_page(page);
3801 status = a_ops->write_end(file, mapping, pos, bytes, copied,
3803 if (unlikely(status != copied)) {
3804 iov_iter_revert(i, copied - max(status, 0L));
3805 if (unlikely(status < 0))
3810 if (unlikely(status == 0)) {
3812 * A short copy made ->write_end() reject the
3813 * thing entirely. Might be memory poisoning
3814 * halfway through, might be a race with munmap,
3815 * might be severe memory pressure.
3824 balance_dirty_pages_ratelimited(mapping);
3825 } while (iov_iter_count(i));
3827 return written ? written : status;
3829 EXPORT_SYMBOL(generic_perform_write);
3832 * __generic_file_write_iter - write data to a file
3833 * @iocb: IO state structure (file, offset, etc.)
3834 * @from: iov_iter with data to write
3836 * This function does all the work needed for actually writing data to a
3837 * file. It does all basic checks, removes SUID from the file, updates
3838 * modification times and calls proper subroutines depending on whether we
3839 * do direct IO or a standard buffered write.
3841 * It expects i_rwsem to be grabbed unless we work on a block device or similar
3842 * object which does not need locking at all.
3844 * This function does *not* take care of syncing data in case of O_SYNC write.
3845 * A caller has to handle it. This is mainly due to the fact that we want to
3846 * avoid syncing under i_rwsem.
3849 * * number of bytes written, even for truncated writes
3850 * * negative error code if no data has been written at all
3852 ssize_t __generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
3854 struct file *file = iocb->ki_filp;
3855 struct address_space *mapping = file->f_mapping;
3856 struct inode *inode = mapping->host;
3857 ssize_t written = 0;
3861 /* We can write back this queue in page reclaim */
3862 current->backing_dev_info = inode_to_bdi(inode);
3863 err = file_remove_privs(file);
3867 err = file_update_time(file);
3871 if (iocb->ki_flags & IOCB_DIRECT) {
3872 loff_t pos, endbyte;
3874 written = generic_file_direct_write(iocb, from);
3876 * If the write stopped short of completing, fall back to
3877 * buffered writes. Some filesystems do this for writes to
3878 * holes, for example. For DAX files, a buffered write will
3879 * not succeed (even if it did, DAX does not handle dirty
3880 * page-cache pages correctly).
3882 if (written < 0 || !iov_iter_count(from) || IS_DAX(inode))
3885 status = generic_perform_write(file, from, pos = iocb->ki_pos);
3887 * If generic_perform_write() returned a synchronous error
3888 * then we want to return the number of bytes which were
3889 * direct-written, or the error code if that was zero. Note
3890 * that this differs from normal direct-io semantics, which
3891 * will return -EFOO even if some bytes were written.
3893 if (unlikely(status < 0)) {
3898 * We need to ensure that the page cache pages are written to
3899 * disk and invalidated to preserve the expected O_DIRECT
3902 endbyte = pos + status - 1;
3903 err = filemap_write_and_wait_range(mapping, pos, endbyte);
3905 iocb->ki_pos = endbyte + 1;
3907 invalidate_mapping_pages(mapping,
3909 endbyte >> PAGE_SHIFT);
3912 * We don't know how much we wrote, so just return
3913 * the number of bytes which were direct-written
3917 written = generic_perform_write(file, from, iocb->ki_pos);
3918 if (likely(written > 0))
3919 iocb->ki_pos += written;
3922 current->backing_dev_info = NULL;
3923 return written ? written : err;
3925 EXPORT_SYMBOL(__generic_file_write_iter);
3928 * generic_file_write_iter - write data to a file
3929 * @iocb: IO state structure
3930 * @from: iov_iter with data to write
3932 * This is a wrapper around __generic_file_write_iter() to be used by most
3933 * filesystems. It takes care of syncing the file in case of O_SYNC file
3934 * and acquires i_rwsem as needed.
3936 * * negative error code if no data has been written at all of
3937 * vfs_fsync_range() failed for a synchronous write
3938 * * number of bytes written, even for truncated writes
3940 ssize_t generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
3942 struct file *file = iocb->ki_filp;
3943 struct inode *inode = file->f_mapping->host;
3947 ret = generic_write_checks(iocb, from);
3949 ret = __generic_file_write_iter(iocb, from);
3950 inode_unlock(inode);
3953 ret = generic_write_sync(iocb, ret);
3956 EXPORT_SYMBOL(generic_file_write_iter);
3959 * try_to_release_page() - release old fs-specific metadata on a page
3961 * @page: the page which the kernel is trying to free
3962 * @gfp_mask: memory allocation flags (and I/O mode)
3964 * The address_space is to try to release any data against the page
3965 * (presumably at page->private).
3967 * This may also be called if PG_fscache is set on a page, indicating that the
3968 * page is known to the local caching routines.
3970 * The @gfp_mask argument specifies whether I/O may be performed to release
3971 * this page (__GFP_IO), and whether the call may block (__GFP_RECLAIM & __GFP_FS).
3973 * Return: %1 if the release was successful, otherwise return zero.
3975 int try_to_release_page(struct page *page, gfp_t gfp_mask)
3977 struct address_space * const mapping = page->mapping;
3979 BUG_ON(!PageLocked(page));
3980 if (PageWriteback(page))
3983 if (mapping && mapping->a_ops->releasepage)
3984 return mapping->a_ops->releasepage(page, gfp_mask);
3985 return try_to_free_buffers(page);
3988 EXPORT_SYMBOL(try_to_release_page);