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/swapops.h>
25 #include <linux/mman.h>
26 #include <linux/pagemap.h>
27 #include <linux/file.h>
28 #include <linux/uio.h>
29 #include <linux/error-injection.h>
30 #include <linux/hash.h>
31 #include <linux/writeback.h>
32 #include <linux/backing-dev.h>
33 #include <linux/pagevec.h>
34 #include <linux/security.h>
35 #include <linux/cpuset.h>
36 #include <linux/hugetlb.h>
37 #include <linux/memcontrol.h>
38 #include <linux/shmem_fs.h>
39 #include <linux/rmap.h>
40 #include <linux/delayacct.h>
41 #include <linux/psi.h>
42 #include <linux/ramfs.h>
43 #include <linux/page_idle.h>
44 #include <linux/migrate.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_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 folio *folio, void *shadow)
127 XA_STATE(xas, &mapping->i_pages, folio->index);
130 mapping_set_update(&xas, mapping);
132 /* hugetlb pages are represented by a single entry in the xarray */
133 if (!folio_test_hugetlb(folio)) {
134 xas_set_order(&xas, folio->index, folio_order(folio));
135 nr = folio_nr_pages(folio);
138 VM_BUG_ON_FOLIO(!folio_test_locked(folio), folio);
140 xas_store(&xas, shadow);
141 xas_init_marks(&xas);
143 folio->mapping = NULL;
144 /* Leave page->index set: truncation lookup relies upon it */
145 mapping->nrpages -= nr;
148 static void filemap_unaccount_folio(struct address_space *mapping,
153 VM_BUG_ON_FOLIO(folio_mapped(folio), folio);
154 if (!IS_ENABLED(CONFIG_DEBUG_VM) && unlikely(folio_mapped(folio))) {
157 pr_alert("BUG: Bad page cache in process %s pfn:%05lx\n",
158 current->comm, folio_pfn(folio));
159 dump_page(&folio->page, "still mapped when deleted");
161 add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE);
163 mapcount = page_mapcount(&folio->page);
164 if (mapping_exiting(mapping) &&
165 folio_ref_count(folio) >= mapcount + 2) {
167 * All vmas have already been torn down, so it's
168 * a good bet that actually the folio is unmapped,
169 * and we'd prefer not to leak it: if we're wrong,
170 * some other bad page check should catch it later.
172 page_mapcount_reset(&folio->page);
173 folio_ref_sub(folio, mapcount);
177 /* hugetlb folios do not participate in page cache accounting. */
178 if (folio_test_hugetlb(folio))
181 nr = folio_nr_pages(folio);
183 __lruvec_stat_mod_folio(folio, NR_FILE_PAGES, -nr);
184 if (folio_test_swapbacked(folio)) {
185 __lruvec_stat_mod_folio(folio, NR_SHMEM, -nr);
186 if (folio_test_pmd_mappable(folio))
187 __lruvec_stat_mod_folio(folio, NR_SHMEM_THPS, -nr);
188 } else if (folio_test_pmd_mappable(folio)) {
189 __lruvec_stat_mod_folio(folio, NR_FILE_THPS, -nr);
190 filemap_nr_thps_dec(mapping);
194 * At this point folio must be either written or cleaned by
195 * truncate. Dirty folio here signals a bug and loss of
198 * This fixes dirty accounting after removing the folio entirely
199 * but leaves the dirty flag set: it has no effect for truncated
200 * folio and anyway will be cleared before returning folio to
203 if (WARN_ON_ONCE(folio_test_dirty(folio)))
204 folio_account_cleaned(folio, mapping,
205 inode_to_wb(mapping->host));
209 * Delete a page from the page cache and free it. Caller has to make
210 * sure the page is locked and that nobody else uses it - or that usage
211 * is safe. The caller must hold the i_pages lock.
213 void __filemap_remove_folio(struct folio *folio, void *shadow)
215 struct address_space *mapping = folio->mapping;
217 trace_mm_filemap_delete_from_page_cache(folio);
218 filemap_unaccount_folio(mapping, folio);
219 page_cache_delete(mapping, folio, shadow);
222 void filemap_free_folio(struct address_space *mapping, struct folio *folio)
224 void (*freepage)(struct page *);
226 freepage = mapping->a_ops->freepage;
228 freepage(&folio->page);
230 if (folio_test_large(folio) && !folio_test_hugetlb(folio)) {
231 folio_ref_sub(folio, folio_nr_pages(folio));
232 VM_BUG_ON_FOLIO(folio_ref_count(folio) <= 0, folio);
239 * filemap_remove_folio - Remove folio from page cache.
242 * This must be called only on folios that are locked and have been
243 * verified to be in the page cache. It will never put the folio into
244 * the free list because the caller has a reference on the page.
246 void filemap_remove_folio(struct folio *folio)
248 struct address_space *mapping = folio->mapping;
250 BUG_ON(!folio_test_locked(folio));
251 spin_lock(&mapping->host->i_lock);
252 xa_lock_irq(&mapping->i_pages);
253 __filemap_remove_folio(folio, NULL);
254 xa_unlock_irq(&mapping->i_pages);
255 if (mapping_shrinkable(mapping))
256 inode_add_lru(mapping->host);
257 spin_unlock(&mapping->host->i_lock);
259 filemap_free_folio(mapping, folio);
263 * page_cache_delete_batch - delete several folios from page cache
264 * @mapping: the mapping to which folios belong
265 * @fbatch: batch of folios to delete
267 * The function walks over mapping->i_pages and removes folios passed in
268 * @fbatch from the mapping. The function expects @fbatch to be sorted
269 * by page index and is optimised for it to be dense.
270 * It tolerates holes in @fbatch (mapping entries at those indices are not
273 * The function expects the i_pages lock to be held.
275 static void page_cache_delete_batch(struct address_space *mapping,
276 struct folio_batch *fbatch)
278 XA_STATE(xas, &mapping->i_pages, fbatch->folios[0]->index);
279 long total_pages = 0;
283 mapping_set_update(&xas, mapping);
284 xas_for_each(&xas, folio, ULONG_MAX) {
285 if (i >= folio_batch_count(fbatch))
288 /* A swap/dax/shadow entry got inserted? Skip it. */
289 if (xa_is_value(folio))
292 * A page got inserted in our range? Skip it. We have our
293 * pages locked so they are protected from being removed.
294 * If we see a page whose index is higher than ours, it
295 * means our page has been removed, which shouldn't be
296 * possible because we're holding the PageLock.
298 if (folio != fbatch->folios[i]) {
299 VM_BUG_ON_FOLIO(folio->index >
300 fbatch->folios[i]->index, folio);
304 WARN_ON_ONCE(!folio_test_locked(folio));
306 folio->mapping = NULL;
307 /* Leave folio->index set: truncation lookup relies on it */
310 xas_store(&xas, NULL);
311 total_pages += folio_nr_pages(folio);
313 mapping->nrpages -= total_pages;
316 void delete_from_page_cache_batch(struct address_space *mapping,
317 struct folio_batch *fbatch)
321 if (!folio_batch_count(fbatch))
324 spin_lock(&mapping->host->i_lock);
325 xa_lock_irq(&mapping->i_pages);
326 for (i = 0; i < folio_batch_count(fbatch); i++) {
327 struct folio *folio = fbatch->folios[i];
329 trace_mm_filemap_delete_from_page_cache(folio);
330 filemap_unaccount_folio(mapping, folio);
332 page_cache_delete_batch(mapping, fbatch);
333 xa_unlock_irq(&mapping->i_pages);
334 if (mapping_shrinkable(mapping))
335 inode_add_lru(mapping->host);
336 spin_unlock(&mapping->host->i_lock);
338 for (i = 0; i < folio_batch_count(fbatch); i++)
339 filemap_free_folio(mapping, fbatch->folios[i]);
342 int filemap_check_errors(struct address_space *mapping)
345 /* Check for outstanding write errors */
346 if (test_bit(AS_ENOSPC, &mapping->flags) &&
347 test_and_clear_bit(AS_ENOSPC, &mapping->flags))
349 if (test_bit(AS_EIO, &mapping->flags) &&
350 test_and_clear_bit(AS_EIO, &mapping->flags))
354 EXPORT_SYMBOL(filemap_check_errors);
356 static int filemap_check_and_keep_errors(struct address_space *mapping)
358 /* Check for outstanding write errors */
359 if (test_bit(AS_EIO, &mapping->flags))
361 if (test_bit(AS_ENOSPC, &mapping->flags))
367 * filemap_fdatawrite_wbc - start writeback on mapping dirty pages in range
368 * @mapping: address space structure to write
369 * @wbc: the writeback_control controlling the writeout
371 * Call writepages on the mapping using the provided wbc to control the
374 * Return: %0 on success, negative error code otherwise.
376 int filemap_fdatawrite_wbc(struct address_space *mapping,
377 struct writeback_control *wbc)
381 if (!mapping_can_writeback(mapping) ||
382 !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
385 wbc_attach_fdatawrite_inode(wbc, mapping->host);
386 ret = do_writepages(mapping, wbc);
387 wbc_detach_inode(wbc);
390 EXPORT_SYMBOL(filemap_fdatawrite_wbc);
393 * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range
394 * @mapping: address space structure to write
395 * @start: offset in bytes where the range starts
396 * @end: offset in bytes where the range ends (inclusive)
397 * @sync_mode: enable synchronous operation
399 * Start writeback against all of a mapping's dirty pages that lie
400 * within the byte offsets <start, end> inclusive.
402 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
403 * opposed to a regular memory cleansing writeback. The difference between
404 * these two operations is that if a dirty page/buffer is encountered, it must
405 * be waited upon, and not just skipped over.
407 * Return: %0 on success, negative error code otherwise.
409 int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
410 loff_t end, int sync_mode)
412 struct writeback_control wbc = {
413 .sync_mode = sync_mode,
414 .nr_to_write = LONG_MAX,
415 .range_start = start,
419 return filemap_fdatawrite_wbc(mapping, &wbc);
422 static inline int __filemap_fdatawrite(struct address_space *mapping,
425 return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode);
428 int filemap_fdatawrite(struct address_space *mapping)
430 return __filemap_fdatawrite(mapping, WB_SYNC_ALL);
432 EXPORT_SYMBOL(filemap_fdatawrite);
434 int filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
437 return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL);
439 EXPORT_SYMBOL(filemap_fdatawrite_range);
442 * filemap_flush - mostly a non-blocking flush
443 * @mapping: target address_space
445 * This is a mostly non-blocking flush. Not suitable for data-integrity
446 * purposes - I/O may not be started against all dirty pages.
448 * Return: %0 on success, negative error code otherwise.
450 int filemap_flush(struct address_space *mapping)
452 return __filemap_fdatawrite(mapping, WB_SYNC_NONE);
454 EXPORT_SYMBOL(filemap_flush);
457 * filemap_range_has_page - check if a page exists in range.
458 * @mapping: address space within which to check
459 * @start_byte: offset in bytes where the range starts
460 * @end_byte: offset in bytes where the range ends (inclusive)
462 * Find at least one page in the range supplied, usually used to check if
463 * direct writing in this range will trigger a writeback.
465 * Return: %true if at least one page exists in the specified range,
468 bool filemap_range_has_page(struct address_space *mapping,
469 loff_t start_byte, loff_t end_byte)
472 XA_STATE(xas, &mapping->i_pages, start_byte >> PAGE_SHIFT);
473 pgoff_t max = end_byte >> PAGE_SHIFT;
475 if (end_byte < start_byte)
480 page = xas_find(&xas, max);
481 if (xas_retry(&xas, page))
483 /* Shadow entries don't count */
484 if (xa_is_value(page))
487 * We don't need to try to pin this page; we're about to
488 * release the RCU lock anyway. It is enough to know that
489 * there was a page here recently.
497 EXPORT_SYMBOL(filemap_range_has_page);
499 static void __filemap_fdatawait_range(struct address_space *mapping,
500 loff_t start_byte, loff_t end_byte)
502 pgoff_t index = start_byte >> PAGE_SHIFT;
503 pgoff_t end = end_byte >> PAGE_SHIFT;
507 if (end_byte < start_byte)
511 while (index <= end) {
514 nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index,
515 end, PAGECACHE_TAG_WRITEBACK);
519 for (i = 0; i < nr_pages; i++) {
520 struct page *page = pvec.pages[i];
522 wait_on_page_writeback(page);
523 ClearPageError(page);
525 pagevec_release(&pvec);
531 * filemap_fdatawait_range - wait for writeback to complete
532 * @mapping: address space structure to wait for
533 * @start_byte: offset in bytes where the range starts
534 * @end_byte: offset in bytes where the range ends (inclusive)
536 * Walk the list of under-writeback pages of the given address space
537 * in the given range and wait for all of them. Check error status of
538 * the address space and return it.
540 * Since the error status of the address space is cleared by this function,
541 * callers are responsible for checking the return value and handling and/or
542 * reporting the error.
544 * Return: error status of the address space.
546 int filemap_fdatawait_range(struct address_space *mapping, loff_t start_byte,
549 __filemap_fdatawait_range(mapping, start_byte, end_byte);
550 return filemap_check_errors(mapping);
552 EXPORT_SYMBOL(filemap_fdatawait_range);
555 * filemap_fdatawait_range_keep_errors - wait for writeback to complete
556 * @mapping: address space structure to wait for
557 * @start_byte: offset in bytes where the range starts
558 * @end_byte: offset in bytes where the range ends (inclusive)
560 * Walk the list of under-writeback pages of the given address space in the
561 * given range and wait for all of them. Unlike filemap_fdatawait_range(),
562 * this function does not clear error status of the address space.
564 * Use this function if callers don't handle errors themselves. Expected
565 * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2),
568 int filemap_fdatawait_range_keep_errors(struct address_space *mapping,
569 loff_t start_byte, loff_t end_byte)
571 __filemap_fdatawait_range(mapping, start_byte, end_byte);
572 return filemap_check_and_keep_errors(mapping);
574 EXPORT_SYMBOL(filemap_fdatawait_range_keep_errors);
577 * file_fdatawait_range - wait for writeback to complete
578 * @file: file pointing to address space structure to wait for
579 * @start_byte: offset in bytes where the range starts
580 * @end_byte: offset in bytes where the range ends (inclusive)
582 * Walk the list of under-writeback pages of the address space that file
583 * refers to, in the given range and wait for all of them. Check error
584 * status of the address space vs. the file->f_wb_err cursor and return it.
586 * Since the error status of the file is advanced by this function,
587 * callers are responsible for checking the return value and handling and/or
588 * reporting the error.
590 * Return: error status of the address space vs. the file->f_wb_err cursor.
592 int file_fdatawait_range(struct file *file, loff_t start_byte, loff_t end_byte)
594 struct address_space *mapping = file->f_mapping;
596 __filemap_fdatawait_range(mapping, start_byte, end_byte);
597 return file_check_and_advance_wb_err(file);
599 EXPORT_SYMBOL(file_fdatawait_range);
602 * filemap_fdatawait_keep_errors - wait for writeback without clearing errors
603 * @mapping: address space structure to wait for
605 * Walk the list of under-writeback pages of the given address space
606 * and wait for all of them. Unlike filemap_fdatawait(), this function
607 * does not clear error status of the address space.
609 * Use this function if callers don't handle errors themselves. Expected
610 * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2),
613 * Return: error status of the address space.
615 int filemap_fdatawait_keep_errors(struct address_space *mapping)
617 __filemap_fdatawait_range(mapping, 0, LLONG_MAX);
618 return filemap_check_and_keep_errors(mapping);
620 EXPORT_SYMBOL(filemap_fdatawait_keep_errors);
622 /* Returns true if writeback might be needed or already in progress. */
623 static bool mapping_needs_writeback(struct address_space *mapping)
625 return mapping->nrpages;
628 bool filemap_range_has_writeback(struct address_space *mapping,
629 loff_t start_byte, loff_t end_byte)
631 XA_STATE(xas, &mapping->i_pages, start_byte >> PAGE_SHIFT);
632 pgoff_t max = end_byte >> PAGE_SHIFT;
635 if (end_byte < start_byte)
639 xas_for_each(&xas, page, max) {
640 if (xas_retry(&xas, page))
642 if (xa_is_value(page))
644 if (PageDirty(page) || PageLocked(page) || PageWriteback(page))
650 EXPORT_SYMBOL_GPL(filemap_range_has_writeback);
653 * filemap_write_and_wait_range - write out & wait on a file range
654 * @mapping: the address_space for the pages
655 * @lstart: offset in bytes where the range starts
656 * @lend: offset in bytes where the range ends (inclusive)
658 * Write out and wait upon file offsets lstart->lend, inclusive.
660 * Note that @lend is inclusive (describes the last byte to be written) so
661 * that this function can be used to write to the very end-of-file (end = -1).
663 * Return: error status of the address space.
665 int filemap_write_and_wait_range(struct address_space *mapping,
666 loff_t lstart, loff_t lend)
670 if (mapping_needs_writeback(mapping)) {
671 err = __filemap_fdatawrite_range(mapping, lstart, lend,
674 * Even if the above returned error, the pages may be
675 * written partially (e.g. -ENOSPC), so we wait for it.
676 * But the -EIO is special case, it may indicate the worst
677 * thing (e.g. bug) happened, so we avoid waiting for it.
680 int err2 = filemap_fdatawait_range(mapping,
685 /* Clear any previously stored errors */
686 filemap_check_errors(mapping);
689 err = filemap_check_errors(mapping);
693 EXPORT_SYMBOL(filemap_write_and_wait_range);
695 void __filemap_set_wb_err(struct address_space *mapping, int err)
697 errseq_t eseq = errseq_set(&mapping->wb_err, err);
699 trace_filemap_set_wb_err(mapping, eseq);
701 EXPORT_SYMBOL(__filemap_set_wb_err);
704 * file_check_and_advance_wb_err - report wb error (if any) that was previously
705 * and advance wb_err to current one
706 * @file: struct file on which the error is being reported
708 * When userland calls fsync (or something like nfsd does the equivalent), we
709 * want to report any writeback errors that occurred since the last fsync (or
710 * since the file was opened if there haven't been any).
712 * Grab the wb_err from the mapping. If it matches what we have in the file,
713 * then just quickly return 0. The file is all caught up.
715 * If it doesn't match, then take the mapping value, set the "seen" flag in
716 * it and try to swap it into place. If it works, or another task beat us
717 * to it with the new value, then update the f_wb_err and return the error
718 * portion. The error at this point must be reported via proper channels
719 * (a'la fsync, or NFS COMMIT operation, etc.).
721 * While we handle mapping->wb_err with atomic operations, the f_wb_err
722 * value is protected by the f_lock since we must ensure that it reflects
723 * the latest value swapped in for this file descriptor.
725 * Return: %0 on success, negative error code otherwise.
727 int file_check_and_advance_wb_err(struct file *file)
730 errseq_t old = READ_ONCE(file->f_wb_err);
731 struct address_space *mapping = file->f_mapping;
733 /* Locklessly handle the common case where nothing has changed */
734 if (errseq_check(&mapping->wb_err, old)) {
735 /* Something changed, must use slow path */
736 spin_lock(&file->f_lock);
737 old = file->f_wb_err;
738 err = errseq_check_and_advance(&mapping->wb_err,
740 trace_file_check_and_advance_wb_err(file, old);
741 spin_unlock(&file->f_lock);
745 * We're mostly using this function as a drop in replacement for
746 * filemap_check_errors. Clear AS_EIO/AS_ENOSPC to emulate the effect
747 * that the legacy code would have had on these flags.
749 clear_bit(AS_EIO, &mapping->flags);
750 clear_bit(AS_ENOSPC, &mapping->flags);
753 EXPORT_SYMBOL(file_check_and_advance_wb_err);
756 * file_write_and_wait_range - write out & wait on a file range
757 * @file: file pointing to address_space with pages
758 * @lstart: offset in bytes where the range starts
759 * @lend: offset in bytes where the range ends (inclusive)
761 * Write out and wait upon file offsets lstart->lend, inclusive.
763 * Note that @lend is inclusive (describes the last byte to be written) so
764 * that this function can be used to write to the very end-of-file (end = -1).
766 * After writing out and waiting on the data, we check and advance the
767 * f_wb_err cursor to the latest value, and return any errors detected there.
769 * Return: %0 on success, negative error code otherwise.
771 int file_write_and_wait_range(struct file *file, loff_t lstart, loff_t lend)
774 struct address_space *mapping = file->f_mapping;
776 if (mapping_needs_writeback(mapping)) {
777 err = __filemap_fdatawrite_range(mapping, lstart, lend,
779 /* See comment of filemap_write_and_wait() */
781 __filemap_fdatawait_range(mapping, lstart, lend);
783 err2 = file_check_and_advance_wb_err(file);
788 EXPORT_SYMBOL(file_write_and_wait_range);
791 * replace_page_cache_page - replace a pagecache page with a new one
792 * @old: page to be replaced
793 * @new: page to replace with
795 * This function replaces a page in the pagecache with a new one. On
796 * success it acquires the pagecache reference for the new page and
797 * drops it for the old page. Both the old and new pages must be
798 * locked. This function does not add the new page to the LRU, the
799 * caller must do that.
801 * The remove + add is atomic. This function cannot fail.
803 void replace_page_cache_page(struct page *old, struct page *new)
805 struct folio *fold = page_folio(old);
806 struct folio *fnew = page_folio(new);
807 struct address_space *mapping = old->mapping;
808 void (*freepage)(struct page *) = mapping->a_ops->freepage;
809 pgoff_t offset = old->index;
810 XA_STATE(xas, &mapping->i_pages, offset);
812 VM_BUG_ON_PAGE(!PageLocked(old), old);
813 VM_BUG_ON_PAGE(!PageLocked(new), new);
814 VM_BUG_ON_PAGE(new->mapping, new);
817 new->mapping = mapping;
820 mem_cgroup_migrate(fold, fnew);
823 xas_store(&xas, new);
826 /* hugetlb pages do not participate in page cache accounting. */
828 __dec_lruvec_page_state(old, NR_FILE_PAGES);
830 __inc_lruvec_page_state(new, NR_FILE_PAGES);
831 if (PageSwapBacked(old))
832 __dec_lruvec_page_state(old, NR_SHMEM);
833 if (PageSwapBacked(new))
834 __inc_lruvec_page_state(new, NR_SHMEM);
835 xas_unlock_irq(&xas);
840 EXPORT_SYMBOL_GPL(replace_page_cache_page);
842 noinline int __filemap_add_folio(struct address_space *mapping,
843 struct folio *folio, pgoff_t index, gfp_t gfp, void **shadowp)
845 XA_STATE(xas, &mapping->i_pages, index);
846 int huge = folio_test_hugetlb(folio);
848 bool charged = false;
850 VM_BUG_ON_FOLIO(!folio_test_locked(folio), folio);
851 VM_BUG_ON_FOLIO(folio_test_swapbacked(folio), folio);
852 mapping_set_update(&xas, mapping);
855 folio->mapping = mapping;
856 folio->index = index;
859 error = mem_cgroup_charge(folio, NULL, gfp);
860 VM_BUG_ON_FOLIO(index & (folio_nr_pages(folio) - 1), folio);
866 gfp &= GFP_RECLAIM_MASK;
869 unsigned int order = xa_get_order(xas.xa, xas.xa_index);
870 void *entry, *old = NULL;
872 if (order > folio_order(folio))
873 xas_split_alloc(&xas, xa_load(xas.xa, xas.xa_index),
876 xas_for_each_conflict(&xas, entry) {
878 if (!xa_is_value(entry)) {
879 xas_set_err(&xas, -EEXIST);
887 /* entry may have been split before we acquired lock */
888 order = xa_get_order(xas.xa, xas.xa_index);
889 if (order > folio_order(folio)) {
890 xas_split(&xas, old, order);
895 xas_store(&xas, folio);
901 /* hugetlb pages do not participate in page cache accounting */
903 __lruvec_stat_add_folio(folio, NR_FILE_PAGES);
905 xas_unlock_irq(&xas);
906 } while (xas_nomem(&xas, gfp));
908 if (xas_error(&xas)) {
909 error = xas_error(&xas);
911 mem_cgroup_uncharge(folio);
915 trace_mm_filemap_add_to_page_cache(folio);
918 folio->mapping = NULL;
919 /* Leave page->index set: truncation relies upon it */
923 ALLOW_ERROR_INJECTION(__filemap_add_folio, ERRNO);
926 * add_to_page_cache_locked - add a locked page to the pagecache
928 * @mapping: the page's address_space
929 * @offset: page index
930 * @gfp_mask: page allocation mode
932 * This function is used to add a page to the pagecache. It must be locked.
933 * This function does not add the page to the LRU. The caller must do that.
935 * Return: %0 on success, negative error code otherwise.
937 int add_to_page_cache_locked(struct page *page, struct address_space *mapping,
938 pgoff_t offset, gfp_t gfp_mask)
940 return __filemap_add_folio(mapping, page_folio(page), offset,
943 EXPORT_SYMBOL(add_to_page_cache_locked);
945 int filemap_add_folio(struct address_space *mapping, struct folio *folio,
946 pgoff_t index, gfp_t gfp)
951 __folio_set_locked(folio);
952 ret = __filemap_add_folio(mapping, folio, index, gfp, &shadow);
954 __folio_clear_locked(folio);
957 * The folio might have been evicted from cache only
958 * recently, in which case it should be activated like
959 * any other repeatedly accessed folio.
960 * The exception is folios getting rewritten; evicting other
961 * data from the working set, only to cache data that will
962 * get overwritten with something else, is a waste of memory.
964 WARN_ON_ONCE(folio_test_active(folio));
965 if (!(gfp & __GFP_WRITE) && shadow)
966 workingset_refault(folio, shadow);
967 folio_add_lru(folio);
971 EXPORT_SYMBOL_GPL(filemap_add_folio);
974 struct folio *filemap_alloc_folio(gfp_t gfp, unsigned int order)
979 if (cpuset_do_page_mem_spread()) {
980 unsigned int cpuset_mems_cookie;
982 cpuset_mems_cookie = read_mems_allowed_begin();
983 n = cpuset_mem_spread_node();
984 folio = __folio_alloc_node(gfp, order, n);
985 } while (!folio && read_mems_allowed_retry(cpuset_mems_cookie));
989 return folio_alloc(gfp, order);
991 EXPORT_SYMBOL(filemap_alloc_folio);
995 * filemap_invalidate_lock_two - lock invalidate_lock for two mappings
997 * Lock exclusively invalidate_lock of any passed mapping that is not NULL.
999 * @mapping1: the first mapping to lock
1000 * @mapping2: the second mapping to lock
1002 void filemap_invalidate_lock_two(struct address_space *mapping1,
1003 struct address_space *mapping2)
1005 if (mapping1 > mapping2)
1006 swap(mapping1, mapping2);
1008 down_write(&mapping1->invalidate_lock);
1009 if (mapping2 && mapping1 != mapping2)
1010 down_write_nested(&mapping2->invalidate_lock, 1);
1012 EXPORT_SYMBOL(filemap_invalidate_lock_two);
1015 * filemap_invalidate_unlock_two - unlock invalidate_lock for two mappings
1017 * Unlock exclusive invalidate_lock of any passed mapping that is not NULL.
1019 * @mapping1: the first mapping to unlock
1020 * @mapping2: the second mapping to unlock
1022 void filemap_invalidate_unlock_two(struct address_space *mapping1,
1023 struct address_space *mapping2)
1026 up_write(&mapping1->invalidate_lock);
1027 if (mapping2 && mapping1 != mapping2)
1028 up_write(&mapping2->invalidate_lock);
1030 EXPORT_SYMBOL(filemap_invalidate_unlock_two);
1033 * In order to wait for pages to become available there must be
1034 * waitqueues associated with pages. By using a hash table of
1035 * waitqueues where the bucket discipline is to maintain all
1036 * waiters on the same queue and wake all when any of the pages
1037 * become available, and for the woken contexts to check to be
1038 * sure the appropriate page became available, this saves space
1039 * at a cost of "thundering herd" phenomena during rare hash
1042 #define PAGE_WAIT_TABLE_BITS 8
1043 #define PAGE_WAIT_TABLE_SIZE (1 << PAGE_WAIT_TABLE_BITS)
1044 static wait_queue_head_t folio_wait_table[PAGE_WAIT_TABLE_SIZE] __cacheline_aligned;
1046 static wait_queue_head_t *folio_waitqueue(struct folio *folio)
1048 return &folio_wait_table[hash_ptr(folio, PAGE_WAIT_TABLE_BITS)];
1051 void __init pagecache_init(void)
1055 for (i = 0; i < PAGE_WAIT_TABLE_SIZE; i++)
1056 init_waitqueue_head(&folio_wait_table[i]);
1058 page_writeback_init();
1062 * The page wait code treats the "wait->flags" somewhat unusually, because
1063 * we have multiple different kinds of waits, not just the usual "exclusive"
1068 * (a) no special bits set:
1070 * We're just waiting for the bit to be released, and when a waker
1071 * calls the wakeup function, we set WQ_FLAG_WOKEN and wake it up,
1072 * and remove it from the wait queue.
1074 * Simple and straightforward.
1076 * (b) WQ_FLAG_EXCLUSIVE:
1078 * The waiter is waiting to get the lock, and only one waiter should
1079 * be woken up to avoid any thundering herd behavior. We'll set the
1080 * WQ_FLAG_WOKEN bit, wake it up, and remove it from the wait queue.
1082 * This is the traditional exclusive wait.
1084 * (c) WQ_FLAG_EXCLUSIVE | WQ_FLAG_CUSTOM:
1086 * The waiter is waiting to get the bit, and additionally wants the
1087 * lock to be transferred to it for fair lock behavior. If the lock
1088 * cannot be taken, we stop walking the wait queue without waking
1091 * This is the "fair lock handoff" case, and in addition to setting
1092 * WQ_FLAG_WOKEN, we set WQ_FLAG_DONE to let the waiter easily see
1093 * that it now has the lock.
1095 static int wake_page_function(wait_queue_entry_t *wait, unsigned mode, int sync, void *arg)
1098 struct wait_page_key *key = arg;
1099 struct wait_page_queue *wait_page
1100 = container_of(wait, struct wait_page_queue, wait);
1102 if (!wake_page_match(wait_page, key))
1106 * If it's a lock handoff wait, we get the bit for it, and
1107 * stop walking (and do not wake it up) if we can't.
1109 flags = wait->flags;
1110 if (flags & WQ_FLAG_EXCLUSIVE) {
1111 if (test_bit(key->bit_nr, &key->folio->flags))
1113 if (flags & WQ_FLAG_CUSTOM) {
1114 if (test_and_set_bit(key->bit_nr, &key->folio->flags))
1116 flags |= WQ_FLAG_DONE;
1121 * We are holding the wait-queue lock, but the waiter that
1122 * is waiting for this will be checking the flags without
1125 * So update the flags atomically, and wake up the waiter
1126 * afterwards to avoid any races. This store-release pairs
1127 * with the load-acquire in folio_wait_bit_common().
1129 smp_store_release(&wait->flags, flags | WQ_FLAG_WOKEN);
1130 wake_up_state(wait->private, mode);
1133 * Ok, we have successfully done what we're waiting for,
1134 * and we can unconditionally remove the wait entry.
1136 * Note that this pairs with the "finish_wait()" in the
1137 * waiter, and has to be the absolute last thing we do.
1138 * After this list_del_init(&wait->entry) the wait entry
1139 * might be de-allocated and the process might even have
1142 list_del_init_careful(&wait->entry);
1143 return (flags & WQ_FLAG_EXCLUSIVE) != 0;
1146 static void folio_wake_bit(struct folio *folio, int bit_nr)
1148 wait_queue_head_t *q = folio_waitqueue(folio);
1149 struct wait_page_key key;
1150 unsigned long flags;
1151 wait_queue_entry_t bookmark;
1154 key.bit_nr = bit_nr;
1158 bookmark.private = NULL;
1159 bookmark.func = NULL;
1160 INIT_LIST_HEAD(&bookmark.entry);
1162 spin_lock_irqsave(&q->lock, flags);
1163 __wake_up_locked_key_bookmark(q, TASK_NORMAL, &key, &bookmark);
1165 while (bookmark.flags & WQ_FLAG_BOOKMARK) {
1167 * Take a breather from holding the lock,
1168 * allow pages that finish wake up asynchronously
1169 * to acquire the lock and remove themselves
1172 spin_unlock_irqrestore(&q->lock, flags);
1174 spin_lock_irqsave(&q->lock, flags);
1175 __wake_up_locked_key_bookmark(q, TASK_NORMAL, &key, &bookmark);
1179 * It is possible for other pages to have collided on the waitqueue
1180 * hash, so in that case check for a page match. That prevents a long-
1183 * It is still possible to miss a case here, when we woke page waiters
1184 * and removed them from the waitqueue, but there are still other
1187 if (!waitqueue_active(q) || !key.page_match) {
1188 folio_clear_waiters(folio);
1190 * It's possible to miss clearing Waiters here, when we woke
1191 * our page waiters, but the hashed waitqueue has waiters for
1192 * other pages on it.
1194 * That's okay, it's a rare case. The next waker will clear it.
1197 spin_unlock_irqrestore(&q->lock, flags);
1200 static void folio_wake(struct folio *folio, int bit)
1202 if (!folio_test_waiters(folio))
1204 folio_wake_bit(folio, bit);
1208 * A choice of three behaviors for folio_wait_bit_common():
1211 EXCLUSIVE, /* Hold ref to page and take the bit when woken, like
1212 * __folio_lock() waiting on then setting PG_locked.
1214 SHARED, /* Hold ref to page and check the bit when woken, like
1215 * folio_wait_writeback() waiting on PG_writeback.
1217 DROP, /* Drop ref to page before wait, no check when woken,
1218 * like folio_put_wait_locked() on PG_locked.
1223 * Attempt to check (or get) the folio flag, and mark us done
1226 static inline bool folio_trylock_flag(struct folio *folio, int bit_nr,
1227 struct wait_queue_entry *wait)
1229 if (wait->flags & WQ_FLAG_EXCLUSIVE) {
1230 if (test_and_set_bit(bit_nr, &folio->flags))
1232 } else if (test_bit(bit_nr, &folio->flags))
1235 wait->flags |= WQ_FLAG_WOKEN | WQ_FLAG_DONE;
1239 /* How many times do we accept lock stealing from under a waiter? */
1240 int sysctl_page_lock_unfairness = 5;
1242 static inline int folio_wait_bit_common(struct folio *folio, int bit_nr,
1243 int state, enum behavior behavior)
1245 wait_queue_head_t *q = folio_waitqueue(folio);
1246 int unfairness = sysctl_page_lock_unfairness;
1247 struct wait_page_queue wait_page;
1248 wait_queue_entry_t *wait = &wait_page.wait;
1249 bool thrashing = false;
1250 bool delayacct = false;
1251 unsigned long pflags;
1253 if (bit_nr == PG_locked &&
1254 !folio_test_uptodate(folio) && folio_test_workingset(folio)) {
1255 if (!folio_test_swapbacked(folio)) {
1256 delayacct_thrashing_start();
1259 psi_memstall_enter(&pflags);
1264 wait->func = wake_page_function;
1265 wait_page.folio = folio;
1266 wait_page.bit_nr = bit_nr;
1270 if (behavior == EXCLUSIVE) {
1271 wait->flags = WQ_FLAG_EXCLUSIVE;
1272 if (--unfairness < 0)
1273 wait->flags |= WQ_FLAG_CUSTOM;
1277 * Do one last check whether we can get the
1278 * page bit synchronously.
1280 * Do the folio_set_waiters() marking before that
1281 * to let any waker we _just_ missed know they
1282 * need to wake us up (otherwise they'll never
1283 * even go to the slow case that looks at the
1284 * page queue), and add ourselves to the wait
1285 * queue if we need to sleep.
1287 * This part needs to be done under the queue
1288 * lock to avoid races.
1290 spin_lock_irq(&q->lock);
1291 folio_set_waiters(folio);
1292 if (!folio_trylock_flag(folio, bit_nr, wait))
1293 __add_wait_queue_entry_tail(q, wait);
1294 spin_unlock_irq(&q->lock);
1297 * From now on, all the logic will be based on
1298 * the WQ_FLAG_WOKEN and WQ_FLAG_DONE flag, to
1299 * see whether the page bit testing has already
1300 * been done by the wake function.
1302 * We can drop our reference to the folio.
1304 if (behavior == DROP)
1308 * Note that until the "finish_wait()", or until
1309 * we see the WQ_FLAG_WOKEN flag, we need to
1310 * be very careful with the 'wait->flags', because
1311 * we may race with a waker that sets them.
1316 set_current_state(state);
1318 /* Loop until we've been woken or interrupted */
1319 flags = smp_load_acquire(&wait->flags);
1320 if (!(flags & WQ_FLAG_WOKEN)) {
1321 if (signal_pending_state(state, current))
1328 /* If we were non-exclusive, we're done */
1329 if (behavior != EXCLUSIVE)
1332 /* If the waker got the lock for us, we're done */
1333 if (flags & WQ_FLAG_DONE)
1337 * Otherwise, if we're getting the lock, we need to
1338 * try to get it ourselves.
1340 * And if that fails, we'll have to retry this all.
1342 if (unlikely(test_and_set_bit(bit_nr, folio_flags(folio, 0))))
1345 wait->flags |= WQ_FLAG_DONE;
1350 * If a signal happened, this 'finish_wait()' may remove the last
1351 * waiter from the wait-queues, but the folio waiters bit will remain
1352 * set. That's ok. The next wakeup will take care of it, and trying
1353 * to do it here would be difficult and prone to races.
1355 finish_wait(q, wait);
1359 delayacct_thrashing_end();
1360 psi_memstall_leave(&pflags);
1364 * NOTE! The wait->flags weren't stable until we've done the
1365 * 'finish_wait()', and we could have exited the loop above due
1366 * to a signal, and had a wakeup event happen after the signal
1367 * test but before the 'finish_wait()'.
1369 * So only after the finish_wait() can we reliably determine
1370 * if we got woken up or not, so we can now figure out the final
1371 * return value based on that state without races.
1373 * Also note that WQ_FLAG_WOKEN is sufficient for a non-exclusive
1374 * waiter, but an exclusive one requires WQ_FLAG_DONE.
1376 if (behavior == EXCLUSIVE)
1377 return wait->flags & WQ_FLAG_DONE ? 0 : -EINTR;
1379 return wait->flags & WQ_FLAG_WOKEN ? 0 : -EINTR;
1382 #ifdef CONFIG_MIGRATION
1384 * migration_entry_wait_on_locked - Wait for a migration entry to be removed
1385 * @entry: migration swap entry.
1386 * @ptep: mapped pte pointer. Will return with the ptep unmapped. Only required
1387 * for pte entries, pass NULL for pmd entries.
1388 * @ptl: already locked ptl. This function will drop the lock.
1390 * Wait for a migration entry referencing the given page to be removed. This is
1391 * equivalent to put_and_wait_on_page_locked(page, TASK_UNINTERRUPTIBLE) except
1392 * this can be called without taking a reference on the page. Instead this
1393 * should be called while holding the ptl for the migration entry referencing
1396 * Returns after unmapping and unlocking the pte/ptl with pte_unmap_unlock().
1398 * This follows the same logic as folio_wait_bit_common() so see the comments
1401 void migration_entry_wait_on_locked(swp_entry_t entry, pte_t *ptep,
1404 struct wait_page_queue wait_page;
1405 wait_queue_entry_t *wait = &wait_page.wait;
1406 bool thrashing = false;
1407 bool delayacct = false;
1408 unsigned long pflags;
1409 wait_queue_head_t *q;
1410 struct folio *folio = page_folio(pfn_swap_entry_to_page(entry));
1412 q = folio_waitqueue(folio);
1413 if (!folio_test_uptodate(folio) && folio_test_workingset(folio)) {
1414 if (!folio_test_swapbacked(folio)) {
1415 delayacct_thrashing_start();
1418 psi_memstall_enter(&pflags);
1423 wait->func = wake_page_function;
1424 wait_page.folio = folio;
1425 wait_page.bit_nr = PG_locked;
1428 spin_lock_irq(&q->lock);
1429 folio_set_waiters(folio);
1430 if (!folio_trylock_flag(folio, PG_locked, wait))
1431 __add_wait_queue_entry_tail(q, wait);
1432 spin_unlock_irq(&q->lock);
1435 * If a migration entry exists for the page the migration path must hold
1436 * a valid reference to the page, and it must take the ptl to remove the
1437 * migration entry. So the page is valid until the ptl is dropped.
1440 pte_unmap_unlock(ptep, ptl);
1447 set_current_state(TASK_UNINTERRUPTIBLE);
1449 /* Loop until we've been woken or interrupted */
1450 flags = smp_load_acquire(&wait->flags);
1451 if (!(flags & WQ_FLAG_WOKEN)) {
1452 if (signal_pending_state(TASK_UNINTERRUPTIBLE, current))
1461 finish_wait(q, wait);
1465 delayacct_thrashing_end();
1466 psi_memstall_leave(&pflags);
1471 void folio_wait_bit(struct folio *folio, int bit_nr)
1473 folio_wait_bit_common(folio, bit_nr, TASK_UNINTERRUPTIBLE, SHARED);
1475 EXPORT_SYMBOL(folio_wait_bit);
1477 int folio_wait_bit_killable(struct folio *folio, int bit_nr)
1479 return folio_wait_bit_common(folio, bit_nr, TASK_KILLABLE, SHARED);
1481 EXPORT_SYMBOL(folio_wait_bit_killable);
1484 * folio_put_wait_locked - Drop a reference and wait for it to be unlocked
1485 * @folio: The folio to wait for.
1486 * @state: The sleep state (TASK_KILLABLE, TASK_UNINTERRUPTIBLE, etc).
1488 * The caller should hold a reference on @folio. They expect the page to
1489 * become unlocked relatively soon, but do not wish to hold up migration
1490 * (for example) by holding the reference while waiting for the folio to
1491 * come unlocked. After this function returns, the caller should not
1492 * dereference @folio.
1494 * Return: 0 if the folio was unlocked or -EINTR if interrupted by a signal.
1496 int folio_put_wait_locked(struct folio *folio, int state)
1498 return folio_wait_bit_common(folio, PG_locked, state, DROP);
1502 * folio_add_wait_queue - Add an arbitrary waiter to a folio's wait queue
1503 * @folio: Folio defining the wait queue of interest
1504 * @waiter: Waiter to add to the queue
1506 * Add an arbitrary @waiter to the wait queue for the nominated @folio.
1508 void folio_add_wait_queue(struct folio *folio, wait_queue_entry_t *waiter)
1510 wait_queue_head_t *q = folio_waitqueue(folio);
1511 unsigned long flags;
1513 spin_lock_irqsave(&q->lock, flags);
1514 __add_wait_queue_entry_tail(q, waiter);
1515 folio_set_waiters(folio);
1516 spin_unlock_irqrestore(&q->lock, flags);
1518 EXPORT_SYMBOL_GPL(folio_add_wait_queue);
1520 #ifndef clear_bit_unlock_is_negative_byte
1523 * PG_waiters is the high bit in the same byte as PG_lock.
1525 * On x86 (and on many other architectures), we can clear PG_lock and
1526 * test the sign bit at the same time. But if the architecture does
1527 * not support that special operation, we just do this all by hand
1530 * The read of PG_waiters has to be after (or concurrently with) PG_locked
1531 * being cleared, but a memory barrier should be unnecessary since it is
1532 * in the same byte as PG_locked.
1534 static inline bool clear_bit_unlock_is_negative_byte(long nr, volatile void *mem)
1536 clear_bit_unlock(nr, mem);
1537 /* smp_mb__after_atomic(); */
1538 return test_bit(PG_waiters, mem);
1544 * folio_unlock - Unlock a locked folio.
1545 * @folio: The folio.
1547 * Unlocks the folio and wakes up any thread sleeping on the page lock.
1549 * Context: May be called from interrupt or process context. May not be
1550 * called from NMI context.
1552 void folio_unlock(struct folio *folio)
1554 /* Bit 7 allows x86 to check the byte's sign bit */
1555 BUILD_BUG_ON(PG_waiters != 7);
1556 BUILD_BUG_ON(PG_locked > 7);
1557 VM_BUG_ON_FOLIO(!folio_test_locked(folio), folio);
1558 if (clear_bit_unlock_is_negative_byte(PG_locked, folio_flags(folio, 0)))
1559 folio_wake_bit(folio, PG_locked);
1561 EXPORT_SYMBOL(folio_unlock);
1564 * folio_end_private_2 - Clear PG_private_2 and wake any waiters.
1565 * @folio: The folio.
1567 * Clear the PG_private_2 bit on a folio and wake up any sleepers waiting for
1568 * it. The folio reference held for PG_private_2 being set is released.
1570 * This is, for example, used when a netfs folio is being written to a local
1571 * disk cache, thereby allowing writes to the cache for the same folio to be
1574 void folio_end_private_2(struct folio *folio)
1576 VM_BUG_ON_FOLIO(!folio_test_private_2(folio), folio);
1577 clear_bit_unlock(PG_private_2, folio_flags(folio, 0));
1578 folio_wake_bit(folio, PG_private_2);
1581 EXPORT_SYMBOL(folio_end_private_2);
1584 * folio_wait_private_2 - Wait for PG_private_2 to be cleared on a folio.
1585 * @folio: The folio to wait on.
1587 * Wait for PG_private_2 (aka PG_fscache) to be cleared on a folio.
1589 void folio_wait_private_2(struct folio *folio)
1591 while (folio_test_private_2(folio))
1592 folio_wait_bit(folio, PG_private_2);
1594 EXPORT_SYMBOL(folio_wait_private_2);
1597 * folio_wait_private_2_killable - Wait for PG_private_2 to be cleared on a folio.
1598 * @folio: The folio to wait on.
1600 * Wait for PG_private_2 (aka PG_fscache) to be cleared on a folio or until a
1601 * fatal signal is received by the calling task.
1604 * - 0 if successful.
1605 * - -EINTR if a fatal signal was encountered.
1607 int folio_wait_private_2_killable(struct folio *folio)
1611 while (folio_test_private_2(folio)) {
1612 ret = folio_wait_bit_killable(folio, PG_private_2);
1619 EXPORT_SYMBOL(folio_wait_private_2_killable);
1622 * folio_end_writeback - End writeback against a folio.
1623 * @folio: The folio.
1625 void folio_end_writeback(struct folio *folio)
1628 * folio_test_clear_reclaim() could be used here but it is an
1629 * atomic operation and overkill in this particular case. Failing
1630 * to shuffle a folio marked for immediate reclaim is too mild
1631 * a gain to justify taking an atomic operation penalty at the
1632 * end of every folio writeback.
1634 if (folio_test_reclaim(folio)) {
1635 folio_clear_reclaim(folio);
1636 folio_rotate_reclaimable(folio);
1640 * Writeback does not hold a folio reference of its own, relying
1641 * on truncation to wait for the clearing of PG_writeback.
1642 * But here we must make sure that the folio is not freed and
1643 * reused before the folio_wake().
1646 if (!__folio_end_writeback(folio))
1649 smp_mb__after_atomic();
1650 folio_wake(folio, PG_writeback);
1651 acct_reclaim_writeback(folio);
1654 EXPORT_SYMBOL(folio_end_writeback);
1657 * After completing I/O on a page, call this routine to update the page
1658 * flags appropriately
1660 void page_endio(struct page *page, bool is_write, int err)
1664 SetPageUptodate(page);
1666 ClearPageUptodate(page);
1672 struct address_space *mapping;
1675 mapping = page_mapping(page);
1677 mapping_set_error(mapping, err);
1679 end_page_writeback(page);
1682 EXPORT_SYMBOL_GPL(page_endio);
1685 * __folio_lock - Get a lock on the folio, assuming we need to sleep to get it.
1686 * @folio: The folio to lock
1688 void __folio_lock(struct folio *folio)
1690 folio_wait_bit_common(folio, PG_locked, TASK_UNINTERRUPTIBLE,
1693 EXPORT_SYMBOL(__folio_lock);
1695 int __folio_lock_killable(struct folio *folio)
1697 return folio_wait_bit_common(folio, PG_locked, TASK_KILLABLE,
1700 EXPORT_SYMBOL_GPL(__folio_lock_killable);
1702 static int __folio_lock_async(struct folio *folio, struct wait_page_queue *wait)
1704 struct wait_queue_head *q = folio_waitqueue(folio);
1707 wait->folio = folio;
1708 wait->bit_nr = PG_locked;
1710 spin_lock_irq(&q->lock);
1711 __add_wait_queue_entry_tail(q, &wait->wait);
1712 folio_set_waiters(folio);
1713 ret = !folio_trylock(folio);
1715 * If we were successful now, we know we're still on the
1716 * waitqueue as we're still under the lock. This means it's
1717 * safe to remove and return success, we know the callback
1718 * isn't going to trigger.
1721 __remove_wait_queue(q, &wait->wait);
1724 spin_unlock_irq(&q->lock);
1730 * true - folio is locked; mmap_lock is still held.
1731 * false - folio is not locked.
1732 * mmap_lock has been released (mmap_read_unlock(), unless flags had both
1733 * FAULT_FLAG_ALLOW_RETRY and FAULT_FLAG_RETRY_NOWAIT set, in
1734 * which case mmap_lock is still held.
1736 * If neither ALLOW_RETRY nor KILLABLE are set, will always return true
1737 * with the folio locked and the mmap_lock unperturbed.
1739 bool __folio_lock_or_retry(struct folio *folio, struct mm_struct *mm,
1742 if (fault_flag_allow_retry_first(flags)) {
1744 * CAUTION! In this case, mmap_lock is not released
1745 * even though return 0.
1747 if (flags & FAULT_FLAG_RETRY_NOWAIT)
1750 mmap_read_unlock(mm);
1751 if (flags & FAULT_FLAG_KILLABLE)
1752 folio_wait_locked_killable(folio);
1754 folio_wait_locked(folio);
1757 if (flags & FAULT_FLAG_KILLABLE) {
1760 ret = __folio_lock_killable(folio);
1762 mmap_read_unlock(mm);
1766 __folio_lock(folio);
1773 * page_cache_next_miss() - Find the next gap in the page cache.
1774 * @mapping: Mapping.
1776 * @max_scan: Maximum range to search.
1778 * Search the range [index, min(index + max_scan - 1, ULONG_MAX)] for the
1779 * gap with the lowest index.
1781 * This function may be called under the rcu_read_lock. However, this will
1782 * not atomically search a snapshot of the cache at a single point in time.
1783 * For example, if a gap is created at index 5, then subsequently a gap is
1784 * created at index 10, page_cache_next_miss covering both indices may
1785 * return 10 if called under the rcu_read_lock.
1787 * Return: The index of the gap if found, otherwise an index outside the
1788 * range specified (in which case 'return - index >= max_scan' will be true).
1789 * In the rare case of index wrap-around, 0 will be returned.
1791 pgoff_t page_cache_next_miss(struct address_space *mapping,
1792 pgoff_t index, unsigned long max_scan)
1794 XA_STATE(xas, &mapping->i_pages, index);
1796 while (max_scan--) {
1797 void *entry = xas_next(&xas);
1798 if (!entry || xa_is_value(entry))
1800 if (xas.xa_index == 0)
1804 return xas.xa_index;
1806 EXPORT_SYMBOL(page_cache_next_miss);
1809 * page_cache_prev_miss() - Find the previous gap in the page cache.
1810 * @mapping: Mapping.
1812 * @max_scan: Maximum range to search.
1814 * Search the range [max(index - max_scan + 1, 0), index] for the
1815 * gap with the highest index.
1817 * This function may be called under the rcu_read_lock. However, this will
1818 * not atomically search a snapshot of the cache at a single point in time.
1819 * For example, if a gap is created at index 10, then subsequently a gap is
1820 * created at index 5, page_cache_prev_miss() covering both indices may
1821 * return 5 if called under the rcu_read_lock.
1823 * Return: The index of the gap if found, otherwise an index outside the
1824 * range specified (in which case 'index - return >= max_scan' will be true).
1825 * In the rare case of wrap-around, ULONG_MAX will be returned.
1827 pgoff_t page_cache_prev_miss(struct address_space *mapping,
1828 pgoff_t index, unsigned long max_scan)
1830 XA_STATE(xas, &mapping->i_pages, index);
1832 while (max_scan--) {
1833 void *entry = xas_prev(&xas);
1834 if (!entry || xa_is_value(entry))
1836 if (xas.xa_index == ULONG_MAX)
1840 return xas.xa_index;
1842 EXPORT_SYMBOL(page_cache_prev_miss);
1845 * Lockless page cache protocol:
1846 * On the lookup side:
1847 * 1. Load the folio from i_pages
1848 * 2. Increment the refcount if it's not zero
1849 * 3. If the folio is not found by xas_reload(), put the refcount and retry
1851 * On the removal side:
1852 * A. Freeze the page (by zeroing the refcount if nobody else has a reference)
1853 * B. Remove the page from i_pages
1854 * C. Return the page to the page allocator
1856 * This means that any page may have its reference count temporarily
1857 * increased by a speculative page cache (or fast GUP) lookup as it can
1858 * be allocated by another user before the RCU grace period expires.
1859 * Because the refcount temporarily acquired here may end up being the
1860 * last refcount on the page, any page allocation must be freeable by
1865 * mapping_get_entry - Get a page cache entry.
1866 * @mapping: the address_space to search
1867 * @index: The page cache index.
1869 * Looks up the page cache entry at @mapping & @index. If it is a folio,
1870 * it is returned with an increased refcount. If it is a shadow entry
1871 * of a previously evicted folio, or a swap entry from shmem/tmpfs,
1872 * it is returned without further action.
1874 * Return: The folio, swap or shadow entry, %NULL if nothing is found.
1876 static void *mapping_get_entry(struct address_space *mapping, pgoff_t index)
1878 XA_STATE(xas, &mapping->i_pages, index);
1879 struct folio *folio;
1884 folio = xas_load(&xas);
1885 if (xas_retry(&xas, folio))
1888 * A shadow entry of a recently evicted page, or a swap entry from
1889 * shmem/tmpfs. Return it without attempting to raise page count.
1891 if (!folio || xa_is_value(folio))
1894 if (!folio_try_get_rcu(folio))
1897 if (unlikely(folio != xas_reload(&xas))) {
1908 * __filemap_get_folio - Find and get a reference to a folio.
1909 * @mapping: The address_space to search.
1910 * @index: The page index.
1911 * @fgp_flags: %FGP flags modify how the folio is returned.
1912 * @gfp: Memory allocation flags to use if %FGP_CREAT is specified.
1914 * Looks up the page cache entry at @mapping & @index.
1916 * @fgp_flags can be zero or more of these flags:
1918 * * %FGP_ACCESSED - The folio will be marked accessed.
1919 * * %FGP_LOCK - The folio is returned locked.
1920 * * %FGP_ENTRY - If there is a shadow / swap / DAX entry, return it
1921 * instead of allocating a new folio to replace it.
1922 * * %FGP_CREAT - If no page is present then a new page is allocated using
1923 * @gfp and added to the page cache and the VM's LRU list.
1924 * The page is returned locked and with an increased refcount.
1925 * * %FGP_FOR_MMAP - The caller wants to do its own locking dance if the
1926 * page is already in cache. If the page was allocated, unlock it before
1927 * returning so the caller can do the same dance.
1928 * * %FGP_WRITE - The page will be written to by the caller.
1929 * * %FGP_NOFS - __GFP_FS will get cleared in gfp.
1930 * * %FGP_NOWAIT - Don't get blocked by page lock.
1931 * * %FGP_STABLE - Wait for the folio to be stable (finished writeback)
1933 * If %FGP_LOCK or %FGP_CREAT are specified then the function may sleep even
1934 * if the %GFP flags specified for %FGP_CREAT are atomic.
1936 * If there is a page cache page, it is returned with an increased refcount.
1938 * Return: The found folio or %NULL otherwise.
1940 struct folio *__filemap_get_folio(struct address_space *mapping, pgoff_t index,
1941 int fgp_flags, gfp_t gfp)
1943 struct folio *folio;
1946 folio = mapping_get_entry(mapping, index);
1947 if (xa_is_value(folio)) {
1948 if (fgp_flags & FGP_ENTRY)
1955 if (fgp_flags & FGP_LOCK) {
1956 if (fgp_flags & FGP_NOWAIT) {
1957 if (!folio_trylock(folio)) {
1965 /* Has the page been truncated? */
1966 if (unlikely(folio->mapping != mapping)) {
1967 folio_unlock(folio);
1971 VM_BUG_ON_FOLIO(!folio_contains(folio, index), folio);
1974 if (fgp_flags & FGP_ACCESSED)
1975 folio_mark_accessed(folio);
1976 else if (fgp_flags & FGP_WRITE) {
1977 /* Clear idle flag for buffer write */
1978 if (folio_test_idle(folio))
1979 folio_clear_idle(folio);
1982 if (fgp_flags & FGP_STABLE)
1983 folio_wait_stable(folio);
1985 if (!folio && (fgp_flags & FGP_CREAT)) {
1987 if ((fgp_flags & FGP_WRITE) && mapping_can_writeback(mapping))
1989 if (fgp_flags & FGP_NOFS)
1992 folio = filemap_alloc_folio(gfp, 0);
1996 if (WARN_ON_ONCE(!(fgp_flags & (FGP_LOCK | FGP_FOR_MMAP))))
1997 fgp_flags |= FGP_LOCK;
1999 /* Init accessed so avoid atomic mark_page_accessed later */
2000 if (fgp_flags & FGP_ACCESSED)
2001 __folio_set_referenced(folio);
2003 err = filemap_add_folio(mapping, folio, index, gfp);
2004 if (unlikely(err)) {
2012 * filemap_add_folio locks the page, and for mmap
2013 * we expect an unlocked page.
2015 if (folio && (fgp_flags & FGP_FOR_MMAP))
2016 folio_unlock(folio);
2021 EXPORT_SYMBOL(__filemap_get_folio);
2023 static inline struct folio *find_get_entry(struct xa_state *xas, pgoff_t max,
2026 struct folio *folio;
2029 if (mark == XA_PRESENT)
2030 folio = xas_find(xas, max);
2032 folio = xas_find_marked(xas, max, mark);
2034 if (xas_retry(xas, folio))
2037 * A shadow entry of a recently evicted page, a swap
2038 * entry from shmem/tmpfs or a DAX entry. Return it
2039 * without attempting to raise page count.
2041 if (!folio || xa_is_value(folio))
2044 if (!folio_try_get_rcu(folio))
2047 if (unlikely(folio != xas_reload(xas))) {
2059 * find_get_entries - gang pagecache lookup
2060 * @mapping: The address_space to search
2061 * @start: The starting page cache index
2062 * @end: The final page index (inclusive).
2063 * @fbatch: Where the resulting entries are placed.
2064 * @indices: The cache indices corresponding to the entries in @entries
2066 * find_get_entries() will search for and return a batch of entries in
2067 * the mapping. The entries are placed in @fbatch. find_get_entries()
2068 * takes a reference on any actual folios it returns.
2070 * The entries have ascending indexes. The indices may not be consecutive
2071 * due to not-present entries or large folios.
2073 * Any shadow entries of evicted folios, or swap entries from
2074 * shmem/tmpfs, are included in the returned array.
2076 * Return: The number of entries which were found.
2078 unsigned find_get_entries(struct address_space *mapping, pgoff_t start,
2079 pgoff_t end, struct folio_batch *fbatch, pgoff_t *indices)
2081 XA_STATE(xas, &mapping->i_pages, start);
2082 struct folio *folio;
2085 while ((folio = find_get_entry(&xas, end, XA_PRESENT)) != NULL) {
2086 indices[fbatch->nr] = xas.xa_index;
2087 if (!folio_batch_add(fbatch, folio))
2092 return folio_batch_count(fbatch);
2096 * find_lock_entries - Find a batch of pagecache entries.
2097 * @mapping: The address_space to search.
2098 * @start: The starting page cache index.
2099 * @end: The final page index (inclusive).
2100 * @fbatch: Where the resulting entries are placed.
2101 * @indices: The cache indices of the entries in @fbatch.
2103 * find_lock_entries() will return a batch of entries from @mapping.
2104 * Swap, shadow and DAX entries are included. Folios are returned
2105 * locked and with an incremented refcount. Folios which are locked
2106 * by somebody else or under writeback are skipped. Folios which are
2107 * partially outside the range are not returned.
2109 * The entries have ascending indexes. The indices may not be consecutive
2110 * due to not-present entries, large folios, folios which could not be
2111 * locked or folios under writeback.
2113 * Return: The number of entries which were found.
2115 unsigned find_lock_entries(struct address_space *mapping, pgoff_t start,
2116 pgoff_t end, struct folio_batch *fbatch, pgoff_t *indices)
2118 XA_STATE(xas, &mapping->i_pages, start);
2119 struct folio *folio;
2122 while ((folio = find_get_entry(&xas, end, XA_PRESENT))) {
2123 if (!xa_is_value(folio)) {
2124 if (folio->index < start)
2126 if (folio->index + folio_nr_pages(folio) - 1 > end)
2128 if (!folio_trylock(folio))
2130 if (folio->mapping != mapping ||
2131 folio_test_writeback(folio))
2133 VM_BUG_ON_FOLIO(!folio_contains(folio, xas.xa_index),
2136 indices[fbatch->nr] = xas.xa_index;
2137 if (!folio_batch_add(fbatch, folio))
2141 folio_unlock(folio);
2147 return folio_batch_count(fbatch);
2151 bool folio_more_pages(struct folio *folio, pgoff_t index, pgoff_t max)
2153 if (!folio_test_large(folio) || folio_test_hugetlb(folio))
2157 return index < folio->index + folio_nr_pages(folio) - 1;
2161 * find_get_pages_range - gang pagecache lookup
2162 * @mapping: The address_space to search
2163 * @start: The starting page index
2164 * @end: The final page index (inclusive)
2165 * @nr_pages: The maximum number of pages
2166 * @pages: Where the resulting pages are placed
2168 * find_get_pages_range() will search for and return a group of up to @nr_pages
2169 * pages in the mapping starting at index @start and up to index @end
2170 * (inclusive). The pages are placed at @pages. find_get_pages_range() takes
2171 * a reference against the returned pages.
2173 * The search returns a group of mapping-contiguous pages with ascending
2174 * indexes. There may be holes in the indices due to not-present pages.
2175 * We also update @start to index the next page for the traversal.
2177 * Return: the number of pages which were found. If this number is
2178 * smaller than @nr_pages, the end of specified range has been
2181 unsigned find_get_pages_range(struct address_space *mapping, pgoff_t *start,
2182 pgoff_t end, unsigned int nr_pages,
2183 struct page **pages)
2185 XA_STATE(xas, &mapping->i_pages, *start);
2186 struct folio *folio;
2189 if (unlikely(!nr_pages))
2193 while ((folio = find_get_entry(&xas, end, XA_PRESENT))) {
2194 /* Skip over shadow, swap and DAX entries */
2195 if (xa_is_value(folio))
2199 pages[ret] = folio_file_page(folio, xas.xa_index);
2200 if (++ret == nr_pages) {
2201 *start = xas.xa_index + 1;
2204 if (folio_more_pages(folio, xas.xa_index, end)) {
2206 folio_ref_inc(folio);
2212 * We come here when there is no page beyond @end. We take care to not
2213 * overflow the index @start as it confuses some of the callers. This
2214 * breaks the iteration when there is a page at index -1 but that is
2215 * already broken anyway.
2217 if (end == (pgoff_t)-1)
2218 *start = (pgoff_t)-1;
2228 * find_get_pages_contig - gang contiguous pagecache lookup
2229 * @mapping: The address_space to search
2230 * @index: The starting page index
2231 * @nr_pages: The maximum number of pages
2232 * @pages: Where the resulting pages are placed
2234 * find_get_pages_contig() works exactly like find_get_pages(), except
2235 * that the returned number of pages are guaranteed to be contiguous.
2237 * Return: the number of pages which were found.
2239 unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index,
2240 unsigned int nr_pages, struct page **pages)
2242 XA_STATE(xas, &mapping->i_pages, index);
2243 struct folio *folio;
2244 unsigned int ret = 0;
2246 if (unlikely(!nr_pages))
2250 for (folio = xas_load(&xas); folio; folio = xas_next(&xas)) {
2251 if (xas_retry(&xas, folio))
2254 * If the entry has been swapped out, we can stop looking.
2255 * No current caller is looking for DAX entries.
2257 if (xa_is_value(folio))
2260 if (!folio_try_get_rcu(folio))
2263 if (unlikely(folio != xas_reload(&xas)))
2267 pages[ret] = folio_file_page(folio, xas.xa_index);
2268 if (++ret == nr_pages)
2270 if (folio_more_pages(folio, xas.xa_index, ULONG_MAX)) {
2272 folio_ref_inc(folio);
2284 EXPORT_SYMBOL(find_get_pages_contig);
2287 * find_get_pages_range_tag - Find and return head pages matching @tag.
2288 * @mapping: the address_space to search
2289 * @index: the starting page index
2290 * @end: The final page index (inclusive)
2291 * @tag: the tag index
2292 * @nr_pages: the maximum number of pages
2293 * @pages: where the resulting pages are placed
2295 * Like find_get_pages(), except we only return head pages which are tagged
2296 * with @tag. @index is updated to the index immediately after the last
2297 * page we return, ready for the next iteration.
2299 * Return: the number of pages which were found.
2301 unsigned find_get_pages_range_tag(struct address_space *mapping, pgoff_t *index,
2302 pgoff_t end, xa_mark_t tag, unsigned int nr_pages,
2303 struct page **pages)
2305 XA_STATE(xas, &mapping->i_pages, *index);
2306 struct folio *folio;
2309 if (unlikely(!nr_pages))
2313 while ((folio = find_get_entry(&xas, end, tag))) {
2315 * Shadow entries should never be tagged, but this iteration
2316 * is lockless so there is a window for page reclaim to evict
2317 * a page we saw tagged. Skip over it.
2319 if (xa_is_value(folio))
2322 pages[ret] = &folio->page;
2323 if (++ret == nr_pages) {
2324 *index = folio->index + folio_nr_pages(folio);
2330 * We come here when we got to @end. We take care to not overflow the
2331 * index @index as it confuses some of the callers. This breaks the
2332 * iteration when there is a page at index -1 but that is already
2335 if (end == (pgoff_t)-1)
2336 *index = (pgoff_t)-1;
2344 EXPORT_SYMBOL(find_get_pages_range_tag);
2347 * CD/DVDs are error prone. When a medium error occurs, the driver may fail
2348 * a _large_ part of the i/o request. Imagine the worst scenario:
2350 * ---R__________________________________________B__________
2351 * ^ reading here ^ bad block(assume 4k)
2353 * read(R) => miss => readahead(R...B) => media error => frustrating retries
2354 * => failing the whole request => read(R) => read(R+1) =>
2355 * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
2356 * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
2357 * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
2359 * It is going insane. Fix it by quickly scaling down the readahead size.
2361 static void shrink_readahead_size_eio(struct file_ra_state *ra)
2367 * filemap_get_read_batch - Get a batch of folios for read
2369 * Get a batch of folios which represent a contiguous range of bytes in
2370 * the file. No exceptional entries will be returned. If @index is in
2371 * the middle of a folio, the entire folio will be returned. The last
2372 * folio in the batch may have the readahead flag set or the uptodate flag
2373 * clear so that the caller can take the appropriate action.
2375 static void filemap_get_read_batch(struct address_space *mapping,
2376 pgoff_t index, pgoff_t max, struct folio_batch *fbatch)
2378 XA_STATE(xas, &mapping->i_pages, index);
2379 struct folio *folio;
2382 for (folio = xas_load(&xas); folio; folio = xas_next(&xas)) {
2383 if (xas_retry(&xas, folio))
2385 if (xas.xa_index > max || xa_is_value(folio))
2387 if (!folio_try_get_rcu(folio))
2390 if (unlikely(folio != xas_reload(&xas)))
2393 if (!folio_batch_add(fbatch, folio))
2395 if (!folio_test_uptodate(folio))
2397 if (folio_test_readahead(folio))
2399 xas_advance(&xas, folio->index + folio_nr_pages(folio) - 1);
2409 static int filemap_read_folio(struct file *file, struct address_space *mapping,
2410 struct folio *folio)
2415 * A previous I/O error may have been due to temporary failures,
2416 * eg. multipath errors. PG_error will be set again if readpage
2419 folio_clear_error(folio);
2420 /* Start the actual read. The read will unlock the page. */
2421 error = mapping->a_ops->readpage(file, &folio->page);
2425 error = folio_wait_locked_killable(folio);
2428 if (folio_test_uptodate(folio))
2430 shrink_readahead_size_eio(&file->f_ra);
2434 static bool filemap_range_uptodate(struct address_space *mapping,
2435 loff_t pos, struct iov_iter *iter, struct folio *folio)
2439 if (folio_test_uptodate(folio))
2441 /* pipes can't handle partially uptodate pages */
2442 if (iov_iter_is_pipe(iter))
2444 if (!mapping->a_ops->is_partially_uptodate)
2446 if (mapping->host->i_blkbits >= folio_shift(folio))
2449 count = iter->count;
2450 if (folio_pos(folio) > pos) {
2451 count -= folio_pos(folio) - pos;
2454 pos -= folio_pos(folio);
2457 return mapping->a_ops->is_partially_uptodate(&folio->page, pos, count);
2460 static int filemap_update_page(struct kiocb *iocb,
2461 struct address_space *mapping, struct iov_iter *iter,
2462 struct folio *folio)
2466 if (iocb->ki_flags & IOCB_NOWAIT) {
2467 if (!filemap_invalidate_trylock_shared(mapping))
2470 filemap_invalidate_lock_shared(mapping);
2473 if (!folio_trylock(folio)) {
2475 if (iocb->ki_flags & (IOCB_NOWAIT | IOCB_NOIO))
2476 goto unlock_mapping;
2477 if (!(iocb->ki_flags & IOCB_WAITQ)) {
2478 filemap_invalidate_unlock_shared(mapping);
2480 * This is where we usually end up waiting for a
2481 * previously submitted readahead to finish.
2483 folio_put_wait_locked(folio, TASK_KILLABLE);
2484 return AOP_TRUNCATED_PAGE;
2486 error = __folio_lock_async(folio, iocb->ki_waitq);
2488 goto unlock_mapping;
2491 error = AOP_TRUNCATED_PAGE;
2492 if (!folio->mapping)
2496 if (filemap_range_uptodate(mapping, iocb->ki_pos, iter, folio))
2500 if (iocb->ki_flags & (IOCB_NOIO | IOCB_NOWAIT | IOCB_WAITQ))
2503 error = filemap_read_folio(iocb->ki_filp, mapping, folio);
2504 goto unlock_mapping;
2506 folio_unlock(folio);
2508 filemap_invalidate_unlock_shared(mapping);
2509 if (error == AOP_TRUNCATED_PAGE)
2514 static int filemap_create_folio(struct file *file,
2515 struct address_space *mapping, pgoff_t index,
2516 struct folio_batch *fbatch)
2518 struct folio *folio;
2521 folio = filemap_alloc_folio(mapping_gfp_mask(mapping), 0);
2526 * Protect against truncate / hole punch. Grabbing invalidate_lock
2527 * here assures we cannot instantiate and bring uptodate new
2528 * pagecache folios after evicting page cache during truncate
2529 * and before actually freeing blocks. Note that we could
2530 * release invalidate_lock after inserting the folio into
2531 * the page cache as the locked folio would then be enough to
2532 * synchronize with hole punching. But there are code paths
2533 * such as filemap_update_page() filling in partially uptodate
2534 * pages or ->readpages() that need to hold invalidate_lock
2535 * while mapping blocks for IO so let's hold the lock here as
2536 * well to keep locking rules simple.
2538 filemap_invalidate_lock_shared(mapping);
2539 error = filemap_add_folio(mapping, folio, index,
2540 mapping_gfp_constraint(mapping, GFP_KERNEL));
2541 if (error == -EEXIST)
2542 error = AOP_TRUNCATED_PAGE;
2546 error = filemap_read_folio(file, mapping, folio);
2550 filemap_invalidate_unlock_shared(mapping);
2551 folio_batch_add(fbatch, folio);
2554 filemap_invalidate_unlock_shared(mapping);
2559 static int filemap_readahead(struct kiocb *iocb, struct file *file,
2560 struct address_space *mapping, struct folio *folio,
2563 DEFINE_READAHEAD(ractl, file, &file->f_ra, mapping, folio->index);
2565 if (iocb->ki_flags & IOCB_NOIO)
2567 page_cache_async_ra(&ractl, folio, last_index - folio->index);
2571 static int filemap_get_pages(struct kiocb *iocb, struct iov_iter *iter,
2572 struct folio_batch *fbatch)
2574 struct file *filp = iocb->ki_filp;
2575 struct address_space *mapping = filp->f_mapping;
2576 struct file_ra_state *ra = &filp->f_ra;
2577 pgoff_t index = iocb->ki_pos >> PAGE_SHIFT;
2579 struct folio *folio;
2582 last_index = DIV_ROUND_UP(iocb->ki_pos + iter->count, PAGE_SIZE);
2584 if (fatal_signal_pending(current))
2587 filemap_get_read_batch(mapping, index, last_index, fbatch);
2588 if (!folio_batch_count(fbatch)) {
2589 if (iocb->ki_flags & IOCB_NOIO)
2591 page_cache_sync_readahead(mapping, ra, filp, index,
2592 last_index - index);
2593 filemap_get_read_batch(mapping, index, last_index, fbatch);
2595 if (!folio_batch_count(fbatch)) {
2596 if (iocb->ki_flags & (IOCB_NOWAIT | IOCB_WAITQ))
2598 err = filemap_create_folio(filp, mapping,
2599 iocb->ki_pos >> PAGE_SHIFT, fbatch);
2600 if (err == AOP_TRUNCATED_PAGE)
2605 folio = fbatch->folios[folio_batch_count(fbatch) - 1];
2606 if (folio_test_readahead(folio)) {
2607 err = filemap_readahead(iocb, filp, mapping, folio, last_index);
2611 if (!folio_test_uptodate(folio)) {
2612 if ((iocb->ki_flags & IOCB_WAITQ) &&
2613 folio_batch_count(fbatch) > 1)
2614 iocb->ki_flags |= IOCB_NOWAIT;
2615 err = filemap_update_page(iocb, mapping, iter, folio);
2624 if (likely(--fbatch->nr))
2626 if (err == AOP_TRUNCATED_PAGE)
2632 * filemap_read - Read data from the page cache.
2633 * @iocb: The iocb to read.
2634 * @iter: Destination for the data.
2635 * @already_read: Number of bytes already read by the caller.
2637 * Copies data from the page cache. If the data is not currently present,
2638 * uses the readahead and readpage address_space operations to fetch it.
2640 * Return: Total number of bytes copied, including those already read by
2641 * the caller. If an error happens before any bytes are copied, returns
2642 * a negative error number.
2644 ssize_t filemap_read(struct kiocb *iocb, struct iov_iter *iter,
2645 ssize_t already_read)
2647 struct file *filp = iocb->ki_filp;
2648 struct file_ra_state *ra = &filp->f_ra;
2649 struct address_space *mapping = filp->f_mapping;
2650 struct inode *inode = mapping->host;
2651 struct folio_batch fbatch;
2653 bool writably_mapped;
2654 loff_t isize, end_offset;
2656 if (unlikely(iocb->ki_pos >= inode->i_sb->s_maxbytes))
2658 if (unlikely(!iov_iter_count(iter)))
2661 iov_iter_truncate(iter, inode->i_sb->s_maxbytes);
2662 folio_batch_init(&fbatch);
2668 * If we've already successfully copied some data, then we
2669 * can no longer safely return -EIOCBQUEUED. Hence mark
2670 * an async read NOWAIT at that point.
2672 if ((iocb->ki_flags & IOCB_WAITQ) && already_read)
2673 iocb->ki_flags |= IOCB_NOWAIT;
2675 if (unlikely(iocb->ki_pos >= i_size_read(inode)))
2678 error = filemap_get_pages(iocb, iter, &fbatch);
2683 * i_size must be checked after we know the pages are Uptodate.
2685 * Checking i_size after the check allows us to calculate
2686 * the correct value for "nr", which means the zero-filled
2687 * part of the page is not copied back to userspace (unless
2688 * another truncate extends the file - this is desired though).
2690 isize = i_size_read(inode);
2691 if (unlikely(iocb->ki_pos >= isize))
2693 end_offset = min_t(loff_t, isize, iocb->ki_pos + iter->count);
2696 * Once we start copying data, we don't want to be touching any
2697 * cachelines that might be contended:
2699 writably_mapped = mapping_writably_mapped(mapping);
2702 * When a sequential read accesses a page several times, only
2703 * mark it as accessed the first time.
2705 if (iocb->ki_pos >> PAGE_SHIFT !=
2706 ra->prev_pos >> PAGE_SHIFT)
2707 folio_mark_accessed(fbatch.folios[0]);
2709 for (i = 0; i < folio_batch_count(&fbatch); i++) {
2710 struct folio *folio = fbatch.folios[i];
2711 size_t fsize = folio_size(folio);
2712 size_t offset = iocb->ki_pos & (fsize - 1);
2713 size_t bytes = min_t(loff_t, end_offset - iocb->ki_pos,
2717 if (end_offset < folio_pos(folio))
2720 folio_mark_accessed(folio);
2722 * If users can be writing to this folio using arbitrary
2723 * virtual addresses, take care of potential aliasing
2724 * before reading the folio on the kernel side.
2726 if (writably_mapped)
2727 flush_dcache_folio(folio);
2729 copied = copy_folio_to_iter(folio, offset, bytes, iter);
2731 already_read += copied;
2732 iocb->ki_pos += copied;
2733 ra->prev_pos = iocb->ki_pos;
2735 if (copied < bytes) {
2741 for (i = 0; i < folio_batch_count(&fbatch); i++)
2742 folio_put(fbatch.folios[i]);
2743 folio_batch_init(&fbatch);
2744 } while (iov_iter_count(iter) && iocb->ki_pos < isize && !error);
2746 file_accessed(filp);
2748 return already_read ? already_read : error;
2750 EXPORT_SYMBOL_GPL(filemap_read);
2753 * generic_file_read_iter - generic filesystem read routine
2754 * @iocb: kernel I/O control block
2755 * @iter: destination for the data read
2757 * This is the "read_iter()" routine for all filesystems
2758 * that can use the page cache directly.
2760 * The IOCB_NOWAIT flag in iocb->ki_flags indicates that -EAGAIN shall
2761 * be returned when no data can be read without waiting for I/O requests
2762 * to complete; it doesn't prevent readahead.
2764 * The IOCB_NOIO flag in iocb->ki_flags indicates that no new I/O
2765 * requests shall be made for the read or for readahead. When no data
2766 * can be read, -EAGAIN shall be returned. When readahead would be
2767 * triggered, a partial, possibly empty read shall be returned.
2770 * * number of bytes copied, even for partial reads
2771 * * negative error code (or 0 if IOCB_NOIO) if nothing was read
2774 generic_file_read_iter(struct kiocb *iocb, struct iov_iter *iter)
2776 size_t count = iov_iter_count(iter);
2780 return 0; /* skip atime */
2782 if (iocb->ki_flags & IOCB_DIRECT) {
2783 struct file *file = iocb->ki_filp;
2784 struct address_space *mapping = file->f_mapping;
2785 struct inode *inode = mapping->host;
2787 if (iocb->ki_flags & IOCB_NOWAIT) {
2788 if (filemap_range_needs_writeback(mapping, iocb->ki_pos,
2789 iocb->ki_pos + count - 1))
2792 retval = filemap_write_and_wait_range(mapping,
2794 iocb->ki_pos + count - 1);
2799 file_accessed(file);
2801 retval = mapping->a_ops->direct_IO(iocb, iter);
2803 iocb->ki_pos += retval;
2806 if (retval != -EIOCBQUEUED)
2807 iov_iter_revert(iter, count - iov_iter_count(iter));
2810 * Btrfs can have a short DIO read if we encounter
2811 * compressed extents, so if there was an error, or if
2812 * we've already read everything we wanted to, or if
2813 * there was a short read because we hit EOF, go ahead
2814 * and return. Otherwise fallthrough to buffered io for
2815 * the rest of the read. Buffered reads will not work for
2816 * DAX files, so don't bother trying.
2818 if (retval < 0 || !count || IS_DAX(inode))
2820 if (iocb->ki_pos >= i_size_read(inode))
2824 return filemap_read(iocb, iter, retval);
2826 EXPORT_SYMBOL(generic_file_read_iter);
2828 static inline loff_t folio_seek_hole_data(struct xa_state *xas,
2829 struct address_space *mapping, struct folio *folio,
2830 loff_t start, loff_t end, bool seek_data)
2832 const struct address_space_operations *ops = mapping->a_ops;
2833 size_t offset, bsz = i_blocksize(mapping->host);
2835 if (xa_is_value(folio) || folio_test_uptodate(folio))
2836 return seek_data ? start : end;
2837 if (!ops->is_partially_uptodate)
2838 return seek_data ? end : start;
2843 if (unlikely(folio->mapping != mapping))
2846 offset = offset_in_folio(folio, start) & ~(bsz - 1);
2849 if (ops->is_partially_uptodate(&folio->page, offset, bsz) ==
2852 start = (start + bsz) & ~(bsz - 1);
2854 } while (offset < folio_size(folio));
2856 folio_unlock(folio);
2861 static inline size_t seek_folio_size(struct xa_state *xas, struct folio *folio)
2863 if (xa_is_value(folio))
2864 return PAGE_SIZE << xa_get_order(xas->xa, xas->xa_index);
2865 return folio_size(folio);
2869 * mapping_seek_hole_data - Seek for SEEK_DATA / SEEK_HOLE in the page cache.
2870 * @mapping: Address space to search.
2871 * @start: First byte to consider.
2872 * @end: Limit of search (exclusive).
2873 * @whence: Either SEEK_HOLE or SEEK_DATA.
2875 * If the page cache knows which blocks contain holes and which blocks
2876 * contain data, your filesystem can use this function to implement
2877 * SEEK_HOLE and SEEK_DATA. This is useful for filesystems which are
2878 * entirely memory-based such as tmpfs, and filesystems which support
2879 * unwritten extents.
2881 * Return: The requested offset on success, or -ENXIO if @whence specifies
2882 * SEEK_DATA and there is no data after @start. There is an implicit hole
2883 * after @end - 1, so SEEK_HOLE returns @end if all the bytes between @start
2884 * and @end contain data.
2886 loff_t mapping_seek_hole_data(struct address_space *mapping, loff_t start,
2887 loff_t end, int whence)
2889 XA_STATE(xas, &mapping->i_pages, start >> PAGE_SHIFT);
2890 pgoff_t max = (end - 1) >> PAGE_SHIFT;
2891 bool seek_data = (whence == SEEK_DATA);
2892 struct folio *folio;
2898 while ((folio = find_get_entry(&xas, max, XA_PRESENT))) {
2899 loff_t pos = (u64)xas.xa_index << PAGE_SHIFT;
2908 seek_size = seek_folio_size(&xas, folio);
2909 pos = round_up((u64)pos + 1, seek_size);
2910 start = folio_seek_hole_data(&xas, mapping, folio, start, pos,
2916 if (seek_size > PAGE_SIZE)
2917 xas_set(&xas, pos >> PAGE_SHIFT);
2918 if (!xa_is_value(folio))
2925 if (folio && !xa_is_value(folio))
2933 #define MMAP_LOTSAMISS (100)
2935 * lock_folio_maybe_drop_mmap - lock the page, possibly dropping the mmap_lock
2936 * @vmf - the vm_fault for this fault.
2937 * @folio - the folio to lock.
2938 * @fpin - the pointer to the file we may pin (or is already pinned).
2940 * This works similar to lock_folio_or_retry in that it can drop the
2941 * mmap_lock. It differs in that it actually returns the folio locked
2942 * if it returns 1 and 0 if it couldn't lock the folio. If we did have
2943 * to drop the mmap_lock then fpin will point to the pinned file and
2944 * needs to be fput()'ed at a later point.
2946 static int lock_folio_maybe_drop_mmap(struct vm_fault *vmf, struct folio *folio,
2949 if (folio_trylock(folio))
2953 * NOTE! This will make us return with VM_FAULT_RETRY, but with
2954 * the mmap_lock still held. That's how FAULT_FLAG_RETRY_NOWAIT
2955 * is supposed to work. We have way too many special cases..
2957 if (vmf->flags & FAULT_FLAG_RETRY_NOWAIT)
2960 *fpin = maybe_unlock_mmap_for_io(vmf, *fpin);
2961 if (vmf->flags & FAULT_FLAG_KILLABLE) {
2962 if (__folio_lock_killable(folio)) {
2964 * We didn't have the right flags to drop the mmap_lock,
2965 * but all fault_handlers only check for fatal signals
2966 * if we return VM_FAULT_RETRY, so we need to drop the
2967 * mmap_lock here and return 0 if we don't have a fpin.
2970 mmap_read_unlock(vmf->vma->vm_mm);
2974 __folio_lock(folio);
2980 * Synchronous readahead happens when we don't even find a page in the page
2981 * cache at all. We don't want to perform IO under the mmap sem, so if we have
2982 * to drop the mmap sem we return the file that was pinned in order for us to do
2983 * that. If we didn't pin a file then we return NULL. The file that is
2984 * returned needs to be fput()'ed when we're done with it.
2986 static struct file *do_sync_mmap_readahead(struct vm_fault *vmf)
2988 struct file *file = vmf->vma->vm_file;
2989 struct file_ra_state *ra = &file->f_ra;
2990 struct address_space *mapping = file->f_mapping;
2991 DEFINE_READAHEAD(ractl, file, ra, mapping, vmf->pgoff);
2992 struct file *fpin = NULL;
2993 unsigned int mmap_miss;
2995 /* If we don't want any read-ahead, don't bother */
2996 if (vmf->vma->vm_flags & VM_RAND_READ)
3001 if (vmf->vma->vm_flags & VM_SEQ_READ) {
3002 fpin = maybe_unlock_mmap_for_io(vmf, fpin);
3003 page_cache_sync_ra(&ractl, ra->ra_pages);
3007 /* Avoid banging the cache line if not needed */
3008 mmap_miss = READ_ONCE(ra->mmap_miss);
3009 if (mmap_miss < MMAP_LOTSAMISS * 10)
3010 WRITE_ONCE(ra->mmap_miss, ++mmap_miss);
3013 * Do we miss much more than hit in this file? If so,
3014 * stop bothering with read-ahead. It will only hurt.
3016 if (mmap_miss > MMAP_LOTSAMISS)
3022 fpin = maybe_unlock_mmap_for_io(vmf, fpin);
3023 ra->start = max_t(long, 0, vmf->pgoff - ra->ra_pages / 2);
3024 ra->size = ra->ra_pages;
3025 ra->async_size = ra->ra_pages / 4;
3026 ractl._index = ra->start;
3027 do_page_cache_ra(&ractl, ra->size, ra->async_size);
3032 * Asynchronous readahead happens when we find the page and PG_readahead,
3033 * so we want to possibly extend the readahead further. We return the file that
3034 * was pinned if we have to drop the mmap_lock in order to do IO.
3036 static struct file *do_async_mmap_readahead(struct vm_fault *vmf,
3037 struct folio *folio)
3039 struct file *file = vmf->vma->vm_file;
3040 struct file_ra_state *ra = &file->f_ra;
3041 DEFINE_READAHEAD(ractl, file, ra, file->f_mapping, vmf->pgoff);
3042 struct file *fpin = NULL;
3043 unsigned int mmap_miss;
3045 /* If we don't want any read-ahead, don't bother */
3046 if (vmf->vma->vm_flags & VM_RAND_READ || !ra->ra_pages)
3049 mmap_miss = READ_ONCE(ra->mmap_miss);
3051 WRITE_ONCE(ra->mmap_miss, --mmap_miss);
3053 if (folio_test_readahead(folio)) {
3054 fpin = maybe_unlock_mmap_for_io(vmf, fpin);
3055 page_cache_async_ra(&ractl, folio, ra->ra_pages);
3061 * filemap_fault - read in file data for page fault handling
3062 * @vmf: struct vm_fault containing details of the fault
3064 * filemap_fault() is invoked via the vma operations vector for a
3065 * mapped memory region to read in file data during a page fault.
3067 * The goto's are kind of ugly, but this streamlines the normal case of having
3068 * it in the page cache, and handles the special cases reasonably without
3069 * having a lot of duplicated code.
3071 * vma->vm_mm->mmap_lock must be held on entry.
3073 * If our return value has VM_FAULT_RETRY set, it's because the mmap_lock
3074 * may be dropped before doing I/O or by lock_folio_maybe_drop_mmap().
3076 * If our return value does not have VM_FAULT_RETRY set, the mmap_lock
3077 * has not been released.
3079 * We never return with VM_FAULT_RETRY and a bit from VM_FAULT_ERROR set.
3081 * Return: bitwise-OR of %VM_FAULT_ codes.
3083 vm_fault_t filemap_fault(struct vm_fault *vmf)
3086 struct file *file = vmf->vma->vm_file;
3087 struct file *fpin = NULL;
3088 struct address_space *mapping = file->f_mapping;
3089 struct inode *inode = mapping->host;
3090 pgoff_t max_idx, index = vmf->pgoff;
3091 struct folio *folio;
3093 bool mapping_locked = false;
3095 max_idx = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
3096 if (unlikely(index >= max_idx))
3097 return VM_FAULT_SIGBUS;
3100 * Do we have something in the page cache already?
3102 folio = filemap_get_folio(mapping, index);
3103 if (likely(folio)) {
3105 * We found the page, so try async readahead before waiting for
3108 if (!(vmf->flags & FAULT_FLAG_TRIED))
3109 fpin = do_async_mmap_readahead(vmf, folio);
3110 if (unlikely(!folio_test_uptodate(folio))) {
3111 filemap_invalidate_lock_shared(mapping);
3112 mapping_locked = true;
3115 /* No page in the page cache at all */
3116 count_vm_event(PGMAJFAULT);
3117 count_memcg_event_mm(vmf->vma->vm_mm, PGMAJFAULT);
3118 ret = VM_FAULT_MAJOR;
3119 fpin = do_sync_mmap_readahead(vmf);
3122 * See comment in filemap_create_folio() why we need
3125 if (!mapping_locked) {
3126 filemap_invalidate_lock_shared(mapping);
3127 mapping_locked = true;
3129 folio = __filemap_get_folio(mapping, index,
3130 FGP_CREAT|FGP_FOR_MMAP,
3135 filemap_invalidate_unlock_shared(mapping);
3136 return VM_FAULT_OOM;
3140 if (!lock_folio_maybe_drop_mmap(vmf, folio, &fpin))
3143 /* Did it get truncated? */
3144 if (unlikely(folio->mapping != mapping)) {
3145 folio_unlock(folio);
3149 VM_BUG_ON_FOLIO(!folio_contains(folio, index), folio);
3152 * We have a locked page in the page cache, now we need to check
3153 * that it's up-to-date. If not, it is going to be due to an error.
3155 if (unlikely(!folio_test_uptodate(folio))) {
3157 * The page was in cache and uptodate and now it is not.
3158 * Strange but possible since we didn't hold the page lock all
3159 * the time. Let's drop everything get the invalidate lock and
3162 if (!mapping_locked) {
3163 folio_unlock(folio);
3167 goto page_not_uptodate;
3171 * We've made it this far and we had to drop our mmap_lock, now is the
3172 * time to return to the upper layer and have it re-find the vma and
3176 folio_unlock(folio);
3180 filemap_invalidate_unlock_shared(mapping);
3183 * Found the page and have a reference on it.
3184 * We must recheck i_size under page lock.
3186 max_idx = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
3187 if (unlikely(index >= max_idx)) {
3188 folio_unlock(folio);
3190 return VM_FAULT_SIGBUS;
3193 vmf->page = folio_file_page(folio, index);
3194 return ret | VM_FAULT_LOCKED;
3198 * Umm, take care of errors if the page isn't up-to-date.
3199 * Try to re-read it _once_. We do this synchronously,
3200 * because there really aren't any performance issues here
3201 * and we need to check for errors.
3203 fpin = maybe_unlock_mmap_for_io(vmf, fpin);
3204 error = filemap_read_folio(file, mapping, folio);
3209 if (!error || error == AOP_TRUNCATED_PAGE)
3211 filemap_invalidate_unlock_shared(mapping);
3213 return VM_FAULT_SIGBUS;
3217 * We dropped the mmap_lock, we need to return to the fault handler to
3218 * re-find the vma and come back and find our hopefully still populated
3224 filemap_invalidate_unlock_shared(mapping);
3227 return ret | VM_FAULT_RETRY;
3229 EXPORT_SYMBOL(filemap_fault);
3231 static bool filemap_map_pmd(struct vm_fault *vmf, struct page *page)
3233 struct mm_struct *mm = vmf->vma->vm_mm;
3235 /* Huge page is mapped? No need to proceed. */
3236 if (pmd_trans_huge(*vmf->pmd)) {
3242 if (pmd_none(*vmf->pmd) && PageTransHuge(page)) {
3243 vm_fault_t ret = do_set_pmd(vmf, page);
3245 /* The page is mapped successfully, reference consumed. */
3251 if (pmd_none(*vmf->pmd))
3252 pmd_install(mm, vmf->pmd, &vmf->prealloc_pte);
3254 /* See comment in handle_pte_fault() */
3255 if (pmd_devmap_trans_unstable(vmf->pmd)) {
3264 static struct folio *next_uptodate_page(struct folio *folio,
3265 struct address_space *mapping,
3266 struct xa_state *xas, pgoff_t end_pgoff)
3268 unsigned long max_idx;
3273 if (xas_retry(xas, folio))
3275 if (xa_is_value(folio))
3277 if (folio_test_locked(folio))
3279 if (!folio_try_get_rcu(folio))
3281 /* Has the page moved or been split? */
3282 if (unlikely(folio != xas_reload(xas)))
3284 if (!folio_test_uptodate(folio) || folio_test_readahead(folio))
3286 if (!folio_trylock(folio))
3288 if (folio->mapping != mapping)
3290 if (!folio_test_uptodate(folio))
3292 max_idx = DIV_ROUND_UP(i_size_read(mapping->host), PAGE_SIZE);
3293 if (xas->xa_index >= max_idx)
3297 folio_unlock(folio);
3300 } while ((folio = xas_next_entry(xas, end_pgoff)) != NULL);
3305 static inline struct folio *first_map_page(struct address_space *mapping,
3306 struct xa_state *xas,
3309 return next_uptodate_page(xas_find(xas, end_pgoff),
3310 mapping, xas, end_pgoff);
3313 static inline struct folio *next_map_page(struct address_space *mapping,
3314 struct xa_state *xas,
3317 return next_uptodate_page(xas_next_entry(xas, end_pgoff),
3318 mapping, xas, end_pgoff);
3321 vm_fault_t filemap_map_pages(struct vm_fault *vmf,
3322 pgoff_t start_pgoff, pgoff_t end_pgoff)
3324 struct vm_area_struct *vma = vmf->vma;
3325 struct file *file = vma->vm_file;
3326 struct address_space *mapping = file->f_mapping;
3327 pgoff_t last_pgoff = start_pgoff;
3329 XA_STATE(xas, &mapping->i_pages, start_pgoff);
3330 struct folio *folio;
3332 unsigned int mmap_miss = READ_ONCE(file->f_ra.mmap_miss);
3336 folio = first_map_page(mapping, &xas, end_pgoff);
3340 if (filemap_map_pmd(vmf, &folio->page)) {
3341 ret = VM_FAULT_NOPAGE;
3345 addr = vma->vm_start + ((start_pgoff - vma->vm_pgoff) << PAGE_SHIFT);
3346 vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, addr, &vmf->ptl);
3349 page = folio_file_page(folio, xas.xa_index);
3350 if (PageHWPoison(page))
3356 addr += (xas.xa_index - last_pgoff) << PAGE_SHIFT;
3357 vmf->pte += xas.xa_index - last_pgoff;
3358 last_pgoff = xas.xa_index;
3360 if (!pte_none(*vmf->pte))
3363 /* We're about to handle the fault */
3364 if (vmf->address == addr)
3365 ret = VM_FAULT_NOPAGE;
3367 do_set_pte(vmf, page, addr);
3368 /* no need to invalidate: a not-present page won't be cached */
3369 update_mmu_cache(vma, addr, vmf->pte);
3370 if (folio_more_pages(folio, xas.xa_index, end_pgoff)) {
3372 folio_ref_inc(folio);
3375 folio_unlock(folio);
3378 if (folio_more_pages(folio, xas.xa_index, end_pgoff)) {
3382 folio_unlock(folio);
3384 } while ((folio = next_map_page(mapping, &xas, end_pgoff)) != NULL);
3385 pte_unmap_unlock(vmf->pte, vmf->ptl);
3388 WRITE_ONCE(file->f_ra.mmap_miss, mmap_miss);
3391 EXPORT_SYMBOL(filemap_map_pages);
3393 vm_fault_t filemap_page_mkwrite(struct vm_fault *vmf)
3395 struct address_space *mapping = vmf->vma->vm_file->f_mapping;
3396 struct folio *folio = page_folio(vmf->page);
3397 vm_fault_t ret = VM_FAULT_LOCKED;
3399 sb_start_pagefault(mapping->host->i_sb);
3400 file_update_time(vmf->vma->vm_file);
3402 if (folio->mapping != mapping) {
3403 folio_unlock(folio);
3404 ret = VM_FAULT_NOPAGE;
3408 * We mark the folio dirty already here so that when freeze is in
3409 * progress, we are guaranteed that writeback during freezing will
3410 * see the dirty folio and writeprotect it again.
3412 folio_mark_dirty(folio);
3413 folio_wait_stable(folio);
3415 sb_end_pagefault(mapping->host->i_sb);
3419 const struct vm_operations_struct generic_file_vm_ops = {
3420 .fault = filemap_fault,
3421 .map_pages = filemap_map_pages,
3422 .page_mkwrite = filemap_page_mkwrite,
3425 /* This is used for a general mmap of a disk file */
3427 int generic_file_mmap(struct file *file, struct vm_area_struct *vma)
3429 struct address_space *mapping = file->f_mapping;
3431 if (!mapping->a_ops->readpage)
3433 file_accessed(file);
3434 vma->vm_ops = &generic_file_vm_ops;
3439 * This is for filesystems which do not implement ->writepage.
3441 int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
3443 if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE))
3445 return generic_file_mmap(file, vma);
3448 vm_fault_t filemap_page_mkwrite(struct vm_fault *vmf)
3450 return VM_FAULT_SIGBUS;
3452 int generic_file_mmap(struct file *file, struct vm_area_struct *vma)
3456 int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
3460 #endif /* CONFIG_MMU */
3462 EXPORT_SYMBOL(filemap_page_mkwrite);
3463 EXPORT_SYMBOL(generic_file_mmap);
3464 EXPORT_SYMBOL(generic_file_readonly_mmap);
3466 static struct folio *do_read_cache_folio(struct address_space *mapping,
3467 pgoff_t index, filler_t filler, void *data, gfp_t gfp)
3469 struct folio *folio;
3472 folio = filemap_get_folio(mapping, index);
3474 folio = filemap_alloc_folio(gfp, 0);
3476 return ERR_PTR(-ENOMEM);
3477 err = filemap_add_folio(mapping, folio, index, gfp);
3478 if (unlikely(err)) {
3482 /* Presumably ENOMEM for xarray node */
3483 return ERR_PTR(err);
3488 err = filler(data, &folio->page);
3490 err = mapping->a_ops->readpage(data, &folio->page);
3494 return ERR_PTR(err);
3497 folio_wait_locked(folio);
3498 if (!folio_test_uptodate(folio)) {
3500 return ERR_PTR(-EIO);
3505 if (folio_test_uptodate(folio))
3508 if (!folio_trylock(folio)) {
3509 folio_put_wait_locked(folio, TASK_UNINTERRUPTIBLE);
3513 /* Folio was truncated from mapping */
3514 if (!folio->mapping) {
3515 folio_unlock(folio);
3520 /* Someone else locked and filled the page in a very small window */
3521 if (folio_test_uptodate(folio)) {
3522 folio_unlock(folio);
3527 * A previous I/O error may have been due to temporary
3529 * Clear page error before actual read, PG_error will be
3530 * set again if read page fails.
3532 folio_clear_error(folio);
3536 folio_mark_accessed(folio);
3541 * read_cache_folio - read into page cache, fill it if needed
3542 * @mapping: the page's address_space
3543 * @index: the page index
3544 * @filler: function to perform the read
3545 * @data: first arg to filler(data, page) function, often left as NULL
3547 * Read into the page cache. If a page already exists, and PageUptodate() is
3548 * not set, try to fill the page and wait for it to become unlocked.
3550 * If the page does not get brought uptodate, return -EIO.
3552 * The function expects mapping->invalidate_lock to be already held.
3554 * Return: up to date page on success, ERR_PTR() on failure.
3556 struct folio *read_cache_folio(struct address_space *mapping, pgoff_t index,
3557 filler_t filler, void *data)
3559 return do_read_cache_folio(mapping, index, filler, data,
3560 mapping_gfp_mask(mapping));
3562 EXPORT_SYMBOL(read_cache_folio);
3564 static struct page *do_read_cache_page(struct address_space *mapping,
3565 pgoff_t index, filler_t *filler, void *data, gfp_t gfp)
3567 struct folio *folio;
3569 folio = do_read_cache_folio(mapping, index, filler, data, gfp);
3571 return &folio->page;
3572 return folio_file_page(folio, index);
3575 struct page *read_cache_page(struct address_space *mapping,
3576 pgoff_t index, filler_t *filler, void *data)
3578 return do_read_cache_page(mapping, index, filler, data,
3579 mapping_gfp_mask(mapping));
3581 EXPORT_SYMBOL(read_cache_page);
3584 * read_cache_page_gfp - read into page cache, using specified page allocation flags.
3585 * @mapping: the page's address_space
3586 * @index: the page index
3587 * @gfp: the page allocator flags to use if allocating
3589 * This is the same as "read_mapping_page(mapping, index, NULL)", but with
3590 * any new page allocations done using the specified allocation flags.
3592 * If the page does not get brought uptodate, return -EIO.
3594 * The function expects mapping->invalidate_lock to be already held.
3596 * Return: up to date page on success, ERR_PTR() on failure.
3598 struct page *read_cache_page_gfp(struct address_space *mapping,
3602 return do_read_cache_page(mapping, index, NULL, NULL, gfp);
3604 EXPORT_SYMBOL(read_cache_page_gfp);
3606 int pagecache_write_begin(struct file *file, struct address_space *mapping,
3607 loff_t pos, unsigned len, unsigned flags,
3608 struct page **pagep, void **fsdata)
3610 const struct address_space_operations *aops = mapping->a_ops;
3612 return aops->write_begin(file, mapping, pos, len, flags,
3615 EXPORT_SYMBOL(pagecache_write_begin);
3617 int pagecache_write_end(struct file *file, struct address_space *mapping,
3618 loff_t pos, unsigned len, unsigned copied,
3619 struct page *page, void *fsdata)
3621 const struct address_space_operations *aops = mapping->a_ops;
3623 return aops->write_end(file, mapping, pos, len, copied, page, fsdata);
3625 EXPORT_SYMBOL(pagecache_write_end);
3628 * Warn about a page cache invalidation failure during a direct I/O write.
3630 void dio_warn_stale_pagecache(struct file *filp)
3632 static DEFINE_RATELIMIT_STATE(_rs, 86400 * HZ, DEFAULT_RATELIMIT_BURST);
3636 errseq_set(&filp->f_mapping->wb_err, -EIO);
3637 if (__ratelimit(&_rs)) {
3638 path = file_path(filp, pathname, sizeof(pathname));
3641 pr_crit("Page cache invalidation failure on direct I/O. Possible data corruption due to collision with buffered I/O!\n");
3642 pr_crit("File: %s PID: %d Comm: %.20s\n", path, current->pid,
3648 generic_file_direct_write(struct kiocb *iocb, struct iov_iter *from)
3650 struct file *file = iocb->ki_filp;
3651 struct address_space *mapping = file->f_mapping;
3652 struct inode *inode = mapping->host;
3653 loff_t pos = iocb->ki_pos;
3658 write_len = iov_iter_count(from);
3659 end = (pos + write_len - 1) >> PAGE_SHIFT;
3661 if (iocb->ki_flags & IOCB_NOWAIT) {
3662 /* If there are pages to writeback, return */
3663 if (filemap_range_has_page(file->f_mapping, pos,
3664 pos + write_len - 1))
3667 written = filemap_write_and_wait_range(mapping, pos,
3668 pos + write_len - 1);
3674 * After a write we want buffered reads to be sure to go to disk to get
3675 * the new data. We invalidate clean cached page from the region we're
3676 * about to write. We do this *before* the write so that we can return
3677 * without clobbering -EIOCBQUEUED from ->direct_IO().
3679 written = invalidate_inode_pages2_range(mapping,
3680 pos >> PAGE_SHIFT, end);
3682 * If a page can not be invalidated, return 0 to fall back
3683 * to buffered write.
3686 if (written == -EBUSY)
3691 written = mapping->a_ops->direct_IO(iocb, from);
3694 * Finally, try again to invalidate clean pages which might have been
3695 * cached by non-direct readahead, or faulted in by get_user_pages()
3696 * if the source of the write was an mmap'ed region of the file
3697 * we're writing. Either one is a pretty crazy thing to do,
3698 * so we don't support it 100%. If this invalidation
3699 * fails, tough, the write still worked...
3701 * Most of the time we do not need this since dio_complete() will do
3702 * the invalidation for us. However there are some file systems that
3703 * do not end up with dio_complete() being called, so let's not break
3704 * them by removing it completely.
3706 * Noticeable example is a blkdev_direct_IO().
3708 * Skip invalidation for async writes or if mapping has no pages.
3710 if (written > 0 && mapping->nrpages &&
3711 invalidate_inode_pages2_range(mapping, pos >> PAGE_SHIFT, end))
3712 dio_warn_stale_pagecache(file);
3716 write_len -= written;
3717 if (pos > i_size_read(inode) && !S_ISBLK(inode->i_mode)) {
3718 i_size_write(inode, pos);
3719 mark_inode_dirty(inode);
3723 if (written != -EIOCBQUEUED)
3724 iov_iter_revert(from, write_len - iov_iter_count(from));
3728 EXPORT_SYMBOL(generic_file_direct_write);
3730 ssize_t generic_perform_write(struct file *file,
3731 struct iov_iter *i, loff_t pos)
3733 struct address_space *mapping = file->f_mapping;
3734 const struct address_space_operations *a_ops = mapping->a_ops;
3736 ssize_t written = 0;
3737 unsigned int flags = 0;
3741 unsigned long offset; /* Offset into pagecache page */
3742 unsigned long bytes; /* Bytes to write to page */
3743 size_t copied; /* Bytes copied from user */
3746 offset = (pos & (PAGE_SIZE - 1));
3747 bytes = min_t(unsigned long, PAGE_SIZE - offset,
3752 * Bring in the user page that we will copy from _first_.
3753 * Otherwise there's a nasty deadlock on copying from the
3754 * same page as we're writing to, without it being marked
3757 if (unlikely(fault_in_iov_iter_readable(i, bytes))) {
3762 if (fatal_signal_pending(current)) {
3767 status = a_ops->write_begin(file, mapping, pos, bytes, flags,
3769 if (unlikely(status < 0))
3772 if (mapping_writably_mapped(mapping))
3773 flush_dcache_page(page);
3775 copied = copy_page_from_iter_atomic(page, offset, bytes, i);
3776 flush_dcache_page(page);
3778 status = a_ops->write_end(file, mapping, pos, bytes, copied,
3780 if (unlikely(status != copied)) {
3781 iov_iter_revert(i, copied - max(status, 0L));
3782 if (unlikely(status < 0))
3787 if (unlikely(status == 0)) {
3789 * A short copy made ->write_end() reject the
3790 * thing entirely. Might be memory poisoning
3791 * halfway through, might be a race with munmap,
3792 * might be severe memory pressure.
3801 balance_dirty_pages_ratelimited(mapping);
3802 } while (iov_iter_count(i));
3804 return written ? written : status;
3806 EXPORT_SYMBOL(generic_perform_write);
3809 * __generic_file_write_iter - write data to a file
3810 * @iocb: IO state structure (file, offset, etc.)
3811 * @from: iov_iter with data to write
3813 * This function does all the work needed for actually writing data to a
3814 * file. It does all basic checks, removes SUID from the file, updates
3815 * modification times and calls proper subroutines depending on whether we
3816 * do direct IO or a standard buffered write.
3818 * It expects i_rwsem to be grabbed unless we work on a block device or similar
3819 * object which does not need locking at all.
3821 * This function does *not* take care of syncing data in case of O_SYNC write.
3822 * A caller has to handle it. This is mainly due to the fact that we want to
3823 * avoid syncing under i_rwsem.
3826 * * number of bytes written, even for truncated writes
3827 * * negative error code if no data has been written at all
3829 ssize_t __generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
3831 struct file *file = iocb->ki_filp;
3832 struct address_space *mapping = file->f_mapping;
3833 struct inode *inode = mapping->host;
3834 ssize_t written = 0;
3838 /* We can write back this queue in page reclaim */
3839 current->backing_dev_info = inode_to_bdi(inode);
3840 err = file_remove_privs(file);
3844 err = file_update_time(file);
3848 if (iocb->ki_flags & IOCB_DIRECT) {
3849 loff_t pos, endbyte;
3851 written = generic_file_direct_write(iocb, from);
3853 * If the write stopped short of completing, fall back to
3854 * buffered writes. Some filesystems do this for writes to
3855 * holes, for example. For DAX files, a buffered write will
3856 * not succeed (even if it did, DAX does not handle dirty
3857 * page-cache pages correctly).
3859 if (written < 0 || !iov_iter_count(from) || IS_DAX(inode))
3862 status = generic_perform_write(file, from, pos = iocb->ki_pos);
3864 * If generic_perform_write() returned a synchronous error
3865 * then we want to return the number of bytes which were
3866 * direct-written, or the error code if that was zero. Note
3867 * that this differs from normal direct-io semantics, which
3868 * will return -EFOO even if some bytes were written.
3870 if (unlikely(status < 0)) {
3875 * We need to ensure that the page cache pages are written to
3876 * disk and invalidated to preserve the expected O_DIRECT
3879 endbyte = pos + status - 1;
3880 err = filemap_write_and_wait_range(mapping, pos, endbyte);
3882 iocb->ki_pos = endbyte + 1;
3884 invalidate_mapping_pages(mapping,
3886 endbyte >> PAGE_SHIFT);
3889 * We don't know how much we wrote, so just return
3890 * the number of bytes which were direct-written
3894 written = generic_perform_write(file, from, iocb->ki_pos);
3895 if (likely(written > 0))
3896 iocb->ki_pos += written;
3899 current->backing_dev_info = NULL;
3900 return written ? written : err;
3902 EXPORT_SYMBOL(__generic_file_write_iter);
3905 * generic_file_write_iter - write data to a file
3906 * @iocb: IO state structure
3907 * @from: iov_iter with data to write
3909 * This is a wrapper around __generic_file_write_iter() to be used by most
3910 * filesystems. It takes care of syncing the file in case of O_SYNC file
3911 * and acquires i_rwsem as needed.
3913 * * negative error code if no data has been written at all of
3914 * vfs_fsync_range() failed for a synchronous write
3915 * * number of bytes written, even for truncated writes
3917 ssize_t generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
3919 struct file *file = iocb->ki_filp;
3920 struct inode *inode = file->f_mapping->host;
3924 ret = generic_write_checks(iocb, from);
3926 ret = __generic_file_write_iter(iocb, from);
3927 inode_unlock(inode);
3930 ret = generic_write_sync(iocb, ret);
3933 EXPORT_SYMBOL(generic_file_write_iter);
3936 * filemap_release_folio() - Release fs-specific metadata on a folio.
3937 * @folio: The folio which the kernel is trying to free.
3938 * @gfp: Memory allocation flags (and I/O mode).
3940 * The address_space is trying to release any data attached to a folio
3941 * (presumably at folio->private).
3943 * This will also be called if the private_2 flag is set on a page,
3944 * indicating that the folio has other metadata associated with it.
3946 * The @gfp argument specifies whether I/O may be performed to release
3947 * this page (__GFP_IO), and whether the call may block
3948 * (__GFP_RECLAIM & __GFP_FS).
3950 * Return: %true if the release was successful, otherwise %false.
3952 bool filemap_release_folio(struct folio *folio, gfp_t gfp)
3954 struct address_space * const mapping = folio->mapping;
3956 BUG_ON(!folio_test_locked(folio));
3957 if (folio_test_writeback(folio))
3960 if (mapping && mapping->a_ops->releasepage)
3961 return mapping->a_ops->releasepage(&folio->page, gfp);
3962 return try_to_free_buffers(&folio->page);
3964 EXPORT_SYMBOL(filemap_release_folio);