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/syscalls.h>
26 #include <linux/mman.h>
27 #include <linux/pagemap.h>
28 #include <linux/file.h>
29 #include <linux/uio.h>
30 #include <linux/error-injection.h>
31 #include <linux/hash.h>
32 #include <linux/writeback.h>
33 #include <linux/backing-dev.h>
34 #include <linux/pagevec.h>
35 #include <linux/security.h>
36 #include <linux/cpuset.h>
37 #include <linux/hugetlb.h>
38 #include <linux/memcontrol.h>
39 #include <linux/shmem_fs.h>
40 #include <linux/rmap.h>
41 #include <linux/delayacct.h>
42 #include <linux/psi.h>
43 #include <linux/ramfs.h>
44 #include <linux/page_idle.h>
45 #include <linux/migrate.h>
46 #include <linux/pipe_fs_i.h>
47 #include <linux/splice.h>
48 #include <asm/pgalloc.h>
49 #include <asm/tlbflush.h>
52 #define CREATE_TRACE_POINTS
53 #include <trace/events/filemap.h>
56 * FIXME: remove all knowledge of the buffer layer from the core VM
58 #include <linux/buffer_head.h> /* for try_to_free_buffers */
65 * Shared mappings implemented 30.11.1994. It's not fully working yet,
68 * Shared mappings now work. 15.8.1995 Bruno.
70 * finished 'unifying' the page and buffer cache and SMP-threaded the
71 * page-cache, 21.05.1999, Ingo Molnar <mingo@redhat.com>
73 * SMP-threaded pagemap-LRU 1999, Andrea Arcangeli <andrea@suse.de>
79 * ->i_mmap_rwsem (truncate_pagecache)
80 * ->private_lock (__free_pte->block_dirty_folio)
81 * ->swap_lock (exclusive_swap_page, others)
85 * ->invalidate_lock (acquired by fs in truncate path)
86 * ->i_mmap_rwsem (truncate->unmap_mapping_range)
90 * ->page_table_lock or pte_lock (various, mainly in memory.c)
91 * ->i_pages lock (arch-dependent flush_dcache_mmap_lock)
94 * ->invalidate_lock (filemap_fault)
95 * ->lock_page (filemap_fault, access_process_vm)
97 * ->i_rwsem (generic_perform_write)
98 * ->mmap_lock (fault_in_readable->do_page_fault)
101 * sb_lock (fs/fs-writeback.c)
102 * ->i_pages lock (__sync_single_inode)
105 * ->anon_vma.lock (vma_merge)
108 * ->page_table_lock or pte_lock (anon_vma_prepare and various)
110 * ->page_table_lock or pte_lock
111 * ->swap_lock (try_to_unmap_one)
112 * ->private_lock (try_to_unmap_one)
113 * ->i_pages lock (try_to_unmap_one)
114 * ->lruvec->lru_lock (follow_page->mark_page_accessed)
115 * ->lruvec->lru_lock (check_pte_range->isolate_lru_page)
116 * ->private_lock (page_remove_rmap->set_page_dirty)
117 * ->i_pages lock (page_remove_rmap->set_page_dirty)
118 * bdi.wb->list_lock (page_remove_rmap->set_page_dirty)
119 * ->inode->i_lock (page_remove_rmap->set_page_dirty)
120 * ->memcg->move_lock (page_remove_rmap->lock_page_memcg)
121 * bdi.wb->list_lock (zap_pte_range->set_page_dirty)
122 * ->inode->i_lock (zap_pte_range->set_page_dirty)
123 * ->private_lock (zap_pte_range->block_dirty_folio)
126 * ->tasklist_lock (memory_failure, collect_procs_ao)
129 static void page_cache_delete(struct address_space *mapping,
130 struct folio *folio, void *shadow)
132 XA_STATE(xas, &mapping->i_pages, folio->index);
135 mapping_set_update(&xas, mapping);
137 /* hugetlb pages are represented by a single entry in the xarray */
138 if (!folio_test_hugetlb(folio)) {
139 xas_set_order(&xas, folio->index, folio_order(folio));
140 nr = folio_nr_pages(folio);
143 VM_BUG_ON_FOLIO(!folio_test_locked(folio), folio);
145 xas_store(&xas, shadow);
146 xas_init_marks(&xas);
148 folio->mapping = NULL;
149 /* Leave page->index set: truncation lookup relies upon it */
150 mapping->nrpages -= nr;
153 static void filemap_unaccount_folio(struct address_space *mapping,
158 VM_BUG_ON_FOLIO(folio_mapped(folio), folio);
159 if (!IS_ENABLED(CONFIG_DEBUG_VM) && unlikely(folio_mapped(folio))) {
160 pr_alert("BUG: Bad page cache in process %s pfn:%05lx\n",
161 current->comm, folio_pfn(folio));
162 dump_page(&folio->page, "still mapped when deleted");
164 add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE);
166 if (mapping_exiting(mapping) && !folio_test_large(folio)) {
167 int mapcount = page_mapcount(&folio->page);
169 if (folio_ref_count(folio) >= mapcount + 2) {
171 * All vmas have already been torn down, so it's
172 * a good bet that actually the page is unmapped
173 * and we'd rather not leak it: if we're wrong,
174 * another bad page check should catch it later.
176 page_mapcount_reset(&folio->page);
177 folio_ref_sub(folio, mapcount);
182 /* hugetlb folios do not participate in page cache accounting. */
183 if (folio_test_hugetlb(folio))
186 nr = folio_nr_pages(folio);
188 __lruvec_stat_mod_folio(folio, NR_FILE_PAGES, -nr);
189 if (folio_test_swapbacked(folio)) {
190 __lruvec_stat_mod_folio(folio, NR_SHMEM, -nr);
191 if (folio_test_pmd_mappable(folio))
192 __lruvec_stat_mod_folio(folio, NR_SHMEM_THPS, -nr);
193 } else if (folio_test_pmd_mappable(folio)) {
194 __lruvec_stat_mod_folio(folio, NR_FILE_THPS, -nr);
195 filemap_nr_thps_dec(mapping);
199 * At this point folio must be either written or cleaned by
200 * truncate. Dirty folio here signals a bug and loss of
201 * unwritten data - on ordinary filesystems.
203 * But it's harmless on in-memory filesystems like tmpfs; and can
204 * occur when a driver which did get_user_pages() sets page dirty
205 * before putting it, while the inode is being finally evicted.
207 * Below fixes dirty accounting after removing the folio entirely
208 * but leaves the dirty flag set: it has no effect for truncated
209 * folio and anyway will be cleared before returning folio to
212 if (WARN_ON_ONCE(folio_test_dirty(folio) &&
213 mapping_can_writeback(mapping)))
214 folio_account_cleaned(folio, inode_to_wb(mapping->host));
218 * Delete a page from the page cache and free it. Caller has to make
219 * sure the page is locked and that nobody else uses it - or that usage
220 * is safe. The caller must hold the i_pages lock.
222 void __filemap_remove_folio(struct folio *folio, void *shadow)
224 struct address_space *mapping = folio->mapping;
226 trace_mm_filemap_delete_from_page_cache(folio);
227 filemap_unaccount_folio(mapping, folio);
228 page_cache_delete(mapping, folio, shadow);
231 void filemap_free_folio(struct address_space *mapping, struct folio *folio)
233 void (*free_folio)(struct folio *);
236 free_folio = mapping->a_ops->free_folio;
240 if (folio_test_large(folio) && !folio_test_hugetlb(folio))
241 refs = folio_nr_pages(folio);
242 folio_put_refs(folio, refs);
246 * filemap_remove_folio - Remove folio from page cache.
249 * This must be called only on folios that are locked and have been
250 * verified to be in the page cache. It will never put the folio into
251 * the free list because the caller has a reference on the page.
253 void filemap_remove_folio(struct folio *folio)
255 struct address_space *mapping = folio->mapping;
257 BUG_ON(!folio_test_locked(folio));
258 spin_lock(&mapping->host->i_lock);
259 xa_lock_irq(&mapping->i_pages);
260 __filemap_remove_folio(folio, NULL);
261 xa_unlock_irq(&mapping->i_pages);
262 if (mapping_shrinkable(mapping))
263 inode_add_lru(mapping->host);
264 spin_unlock(&mapping->host->i_lock);
266 filemap_free_folio(mapping, folio);
270 * page_cache_delete_batch - delete several folios from page cache
271 * @mapping: the mapping to which folios belong
272 * @fbatch: batch of folios to delete
274 * The function walks over mapping->i_pages and removes folios passed in
275 * @fbatch from the mapping. The function expects @fbatch to be sorted
276 * by page index and is optimised for it to be dense.
277 * It tolerates holes in @fbatch (mapping entries at those indices are not
280 * The function expects the i_pages lock to be held.
282 static void page_cache_delete_batch(struct address_space *mapping,
283 struct folio_batch *fbatch)
285 XA_STATE(xas, &mapping->i_pages, fbatch->folios[0]->index);
286 long total_pages = 0;
290 mapping_set_update(&xas, mapping);
291 xas_for_each(&xas, folio, ULONG_MAX) {
292 if (i >= folio_batch_count(fbatch))
295 /* A swap/dax/shadow entry got inserted? Skip it. */
296 if (xa_is_value(folio))
299 * A page got inserted in our range? Skip it. We have our
300 * pages locked so they are protected from being removed.
301 * If we see a page whose index is higher than ours, it
302 * means our page has been removed, which shouldn't be
303 * possible because we're holding the PageLock.
305 if (folio != fbatch->folios[i]) {
306 VM_BUG_ON_FOLIO(folio->index >
307 fbatch->folios[i]->index, folio);
311 WARN_ON_ONCE(!folio_test_locked(folio));
313 folio->mapping = NULL;
314 /* Leave folio->index set: truncation lookup relies on it */
317 xas_store(&xas, NULL);
318 total_pages += folio_nr_pages(folio);
320 mapping->nrpages -= total_pages;
323 void delete_from_page_cache_batch(struct address_space *mapping,
324 struct folio_batch *fbatch)
328 if (!folio_batch_count(fbatch))
331 spin_lock(&mapping->host->i_lock);
332 xa_lock_irq(&mapping->i_pages);
333 for (i = 0; i < folio_batch_count(fbatch); i++) {
334 struct folio *folio = fbatch->folios[i];
336 trace_mm_filemap_delete_from_page_cache(folio);
337 filemap_unaccount_folio(mapping, folio);
339 page_cache_delete_batch(mapping, fbatch);
340 xa_unlock_irq(&mapping->i_pages);
341 if (mapping_shrinkable(mapping))
342 inode_add_lru(mapping->host);
343 spin_unlock(&mapping->host->i_lock);
345 for (i = 0; i < folio_batch_count(fbatch); i++)
346 filemap_free_folio(mapping, fbatch->folios[i]);
349 int filemap_check_errors(struct address_space *mapping)
352 /* Check for outstanding write errors */
353 if (test_bit(AS_ENOSPC, &mapping->flags) &&
354 test_and_clear_bit(AS_ENOSPC, &mapping->flags))
356 if (test_bit(AS_EIO, &mapping->flags) &&
357 test_and_clear_bit(AS_EIO, &mapping->flags))
361 EXPORT_SYMBOL(filemap_check_errors);
363 static int filemap_check_and_keep_errors(struct address_space *mapping)
365 /* Check for outstanding write errors */
366 if (test_bit(AS_EIO, &mapping->flags))
368 if (test_bit(AS_ENOSPC, &mapping->flags))
374 * filemap_fdatawrite_wbc - start writeback on mapping dirty pages in range
375 * @mapping: address space structure to write
376 * @wbc: the writeback_control controlling the writeout
378 * Call writepages on the mapping using the provided wbc to control the
381 * Return: %0 on success, negative error code otherwise.
383 int filemap_fdatawrite_wbc(struct address_space *mapping,
384 struct writeback_control *wbc)
388 if (!mapping_can_writeback(mapping) ||
389 !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
392 wbc_attach_fdatawrite_inode(wbc, mapping->host);
393 ret = do_writepages(mapping, wbc);
394 wbc_detach_inode(wbc);
397 EXPORT_SYMBOL(filemap_fdatawrite_wbc);
400 * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range
401 * @mapping: address space structure to write
402 * @start: offset in bytes where the range starts
403 * @end: offset in bytes where the range ends (inclusive)
404 * @sync_mode: enable synchronous operation
406 * Start writeback against all of a mapping's dirty pages that lie
407 * within the byte offsets <start, end> inclusive.
409 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
410 * opposed to a regular memory cleansing writeback. The difference between
411 * these two operations is that if a dirty page/buffer is encountered, it must
412 * be waited upon, and not just skipped over.
414 * Return: %0 on success, negative error code otherwise.
416 int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
417 loff_t end, int sync_mode)
419 struct writeback_control wbc = {
420 .sync_mode = sync_mode,
421 .nr_to_write = LONG_MAX,
422 .range_start = start,
426 return filemap_fdatawrite_wbc(mapping, &wbc);
429 static inline int __filemap_fdatawrite(struct address_space *mapping,
432 return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode);
435 int filemap_fdatawrite(struct address_space *mapping)
437 return __filemap_fdatawrite(mapping, WB_SYNC_ALL);
439 EXPORT_SYMBOL(filemap_fdatawrite);
441 int filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
444 return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL);
446 EXPORT_SYMBOL(filemap_fdatawrite_range);
449 * filemap_flush - mostly a non-blocking flush
450 * @mapping: target address_space
452 * This is a mostly non-blocking flush. Not suitable for data-integrity
453 * purposes - I/O may not be started against all dirty pages.
455 * Return: %0 on success, negative error code otherwise.
457 int filemap_flush(struct address_space *mapping)
459 return __filemap_fdatawrite(mapping, WB_SYNC_NONE);
461 EXPORT_SYMBOL(filemap_flush);
464 * filemap_range_has_page - check if a page exists in range.
465 * @mapping: address space within which to check
466 * @start_byte: offset in bytes where the range starts
467 * @end_byte: offset in bytes where the range ends (inclusive)
469 * Find at least one page in the range supplied, usually used to check if
470 * direct writing in this range will trigger a writeback.
472 * Return: %true if at least one page exists in the specified range,
475 bool filemap_range_has_page(struct address_space *mapping,
476 loff_t start_byte, loff_t end_byte)
479 XA_STATE(xas, &mapping->i_pages, start_byte >> PAGE_SHIFT);
480 pgoff_t max = end_byte >> PAGE_SHIFT;
482 if (end_byte < start_byte)
487 folio = xas_find(&xas, max);
488 if (xas_retry(&xas, folio))
490 /* Shadow entries don't count */
491 if (xa_is_value(folio))
494 * We don't need to try to pin this page; we're about to
495 * release the RCU lock anyway. It is enough to know that
496 * there was a page here recently.
502 return folio != NULL;
504 EXPORT_SYMBOL(filemap_range_has_page);
506 static void __filemap_fdatawait_range(struct address_space *mapping,
507 loff_t start_byte, loff_t end_byte)
509 pgoff_t index = start_byte >> PAGE_SHIFT;
510 pgoff_t end = end_byte >> PAGE_SHIFT;
511 struct folio_batch fbatch;
514 folio_batch_init(&fbatch);
516 while (index <= end) {
519 nr_folios = filemap_get_folios_tag(mapping, &index, end,
520 PAGECACHE_TAG_WRITEBACK, &fbatch);
525 for (i = 0; i < nr_folios; i++) {
526 struct folio *folio = fbatch.folios[i];
528 folio_wait_writeback(folio);
529 folio_clear_error(folio);
531 folio_batch_release(&fbatch);
537 * filemap_fdatawait_range - wait for writeback to complete
538 * @mapping: address space structure to wait for
539 * @start_byte: offset in bytes where the range starts
540 * @end_byte: offset in bytes where the range ends (inclusive)
542 * Walk the list of under-writeback pages of the given address space
543 * in the given range and wait for all of them. Check error status of
544 * the address space and return it.
546 * Since the error status of the address space is cleared by this function,
547 * callers are responsible for checking the return value and handling and/or
548 * reporting the error.
550 * Return: error status of the address space.
552 int filemap_fdatawait_range(struct address_space *mapping, loff_t start_byte,
555 __filemap_fdatawait_range(mapping, start_byte, end_byte);
556 return filemap_check_errors(mapping);
558 EXPORT_SYMBOL(filemap_fdatawait_range);
561 * filemap_fdatawait_range_keep_errors - wait for writeback to complete
562 * @mapping: address space structure to wait for
563 * @start_byte: offset in bytes where the range starts
564 * @end_byte: offset in bytes where the range ends (inclusive)
566 * Walk the list of under-writeback pages of the given address space in the
567 * given range and wait for all of them. Unlike filemap_fdatawait_range(),
568 * this function does not clear error status of the address space.
570 * Use this function if callers don't handle errors themselves. Expected
571 * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2),
574 int filemap_fdatawait_range_keep_errors(struct address_space *mapping,
575 loff_t start_byte, loff_t end_byte)
577 __filemap_fdatawait_range(mapping, start_byte, end_byte);
578 return filemap_check_and_keep_errors(mapping);
580 EXPORT_SYMBOL(filemap_fdatawait_range_keep_errors);
583 * file_fdatawait_range - wait for writeback to complete
584 * @file: file pointing to address space structure to wait for
585 * @start_byte: offset in bytes where the range starts
586 * @end_byte: offset in bytes where the range ends (inclusive)
588 * Walk the list of under-writeback pages of the address space that file
589 * refers to, in the given range and wait for all of them. Check error
590 * status of the address space vs. the file->f_wb_err cursor and return it.
592 * Since the error status of the file is advanced by this function,
593 * callers are responsible for checking the return value and handling and/or
594 * reporting the error.
596 * Return: error status of the address space vs. the file->f_wb_err cursor.
598 int file_fdatawait_range(struct file *file, loff_t start_byte, loff_t end_byte)
600 struct address_space *mapping = file->f_mapping;
602 __filemap_fdatawait_range(mapping, start_byte, end_byte);
603 return file_check_and_advance_wb_err(file);
605 EXPORT_SYMBOL(file_fdatawait_range);
608 * filemap_fdatawait_keep_errors - wait for writeback without clearing errors
609 * @mapping: address space structure to wait for
611 * Walk the list of under-writeback pages of the given address space
612 * and wait for all of them. Unlike filemap_fdatawait(), this function
613 * does not clear error status of the address space.
615 * Use this function if callers don't handle errors themselves. Expected
616 * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2),
619 * Return: error status of the address space.
621 int filemap_fdatawait_keep_errors(struct address_space *mapping)
623 __filemap_fdatawait_range(mapping, 0, LLONG_MAX);
624 return filemap_check_and_keep_errors(mapping);
626 EXPORT_SYMBOL(filemap_fdatawait_keep_errors);
628 /* Returns true if writeback might be needed or already in progress. */
629 static bool mapping_needs_writeback(struct address_space *mapping)
631 return mapping->nrpages;
634 bool filemap_range_has_writeback(struct address_space *mapping,
635 loff_t start_byte, loff_t end_byte)
637 XA_STATE(xas, &mapping->i_pages, start_byte >> PAGE_SHIFT);
638 pgoff_t max = end_byte >> PAGE_SHIFT;
641 if (end_byte < start_byte)
645 xas_for_each(&xas, folio, max) {
646 if (xas_retry(&xas, folio))
648 if (xa_is_value(folio))
650 if (folio_test_dirty(folio) || folio_test_locked(folio) ||
651 folio_test_writeback(folio))
655 return folio != NULL;
657 EXPORT_SYMBOL_GPL(filemap_range_has_writeback);
660 * filemap_write_and_wait_range - write out & wait on a file range
661 * @mapping: the address_space for the pages
662 * @lstart: offset in bytes where the range starts
663 * @lend: offset in bytes where the range ends (inclusive)
665 * Write out and wait upon file offsets lstart->lend, inclusive.
667 * Note that @lend is inclusive (describes the last byte to be written) so
668 * that this function can be used to write to the very end-of-file (end = -1).
670 * Return: error status of the address space.
672 int filemap_write_and_wait_range(struct address_space *mapping,
673 loff_t lstart, loff_t lend)
680 if (mapping_needs_writeback(mapping)) {
681 err = __filemap_fdatawrite_range(mapping, lstart, lend,
684 * Even if the above returned error, the pages may be
685 * written partially (e.g. -ENOSPC), so we wait for it.
686 * But the -EIO is special case, it may indicate the worst
687 * thing (e.g. bug) happened, so we avoid waiting for it.
690 __filemap_fdatawait_range(mapping, lstart, lend);
692 err2 = filemap_check_errors(mapping);
697 EXPORT_SYMBOL(filemap_write_and_wait_range);
699 void __filemap_set_wb_err(struct address_space *mapping, int err)
701 errseq_t eseq = errseq_set(&mapping->wb_err, err);
703 trace_filemap_set_wb_err(mapping, eseq);
705 EXPORT_SYMBOL(__filemap_set_wb_err);
708 * file_check_and_advance_wb_err - report wb error (if any) that was previously
709 * and advance wb_err to current one
710 * @file: struct file on which the error is being reported
712 * When userland calls fsync (or something like nfsd does the equivalent), we
713 * want to report any writeback errors that occurred since the last fsync (or
714 * since the file was opened if there haven't been any).
716 * Grab the wb_err from the mapping. If it matches what we have in the file,
717 * then just quickly return 0. The file is all caught up.
719 * If it doesn't match, then take the mapping value, set the "seen" flag in
720 * it and try to swap it into place. If it works, or another task beat us
721 * to it with the new value, then update the f_wb_err and return the error
722 * portion. The error at this point must be reported via proper channels
723 * (a'la fsync, or NFS COMMIT operation, etc.).
725 * While we handle mapping->wb_err with atomic operations, the f_wb_err
726 * value is protected by the f_lock since we must ensure that it reflects
727 * the latest value swapped in for this file descriptor.
729 * Return: %0 on success, negative error code otherwise.
731 int file_check_and_advance_wb_err(struct file *file)
734 errseq_t old = READ_ONCE(file->f_wb_err);
735 struct address_space *mapping = file->f_mapping;
737 /* Locklessly handle the common case where nothing has changed */
738 if (errseq_check(&mapping->wb_err, old)) {
739 /* Something changed, must use slow path */
740 spin_lock(&file->f_lock);
741 old = file->f_wb_err;
742 err = errseq_check_and_advance(&mapping->wb_err,
744 trace_file_check_and_advance_wb_err(file, old);
745 spin_unlock(&file->f_lock);
749 * We're mostly using this function as a drop in replacement for
750 * filemap_check_errors. Clear AS_EIO/AS_ENOSPC to emulate the effect
751 * that the legacy code would have had on these flags.
753 clear_bit(AS_EIO, &mapping->flags);
754 clear_bit(AS_ENOSPC, &mapping->flags);
757 EXPORT_SYMBOL(file_check_and_advance_wb_err);
760 * file_write_and_wait_range - write out & wait on a file range
761 * @file: file pointing to address_space with pages
762 * @lstart: offset in bytes where the range starts
763 * @lend: offset in bytes where the range ends (inclusive)
765 * Write out and wait upon file offsets lstart->lend, inclusive.
767 * Note that @lend is inclusive (describes the last byte to be written) so
768 * that this function can be used to write to the very end-of-file (end = -1).
770 * After writing out and waiting on the data, we check and advance the
771 * f_wb_err cursor to the latest value, and return any errors detected there.
773 * Return: %0 on success, negative error code otherwise.
775 int file_write_and_wait_range(struct file *file, loff_t lstart, loff_t lend)
778 struct address_space *mapping = file->f_mapping;
783 if (mapping_needs_writeback(mapping)) {
784 err = __filemap_fdatawrite_range(mapping, lstart, lend,
786 /* See comment of filemap_write_and_wait() */
788 __filemap_fdatawait_range(mapping, lstart, lend);
790 err2 = file_check_and_advance_wb_err(file);
795 EXPORT_SYMBOL(file_write_and_wait_range);
798 * replace_page_cache_folio - replace a pagecache folio with a new one
799 * @old: folio to be replaced
800 * @new: folio to replace with
802 * This function replaces a folio in the pagecache with a new one. On
803 * success it acquires the pagecache reference for the new folio and
804 * drops it for the old folio. Both the old and new folios must be
805 * locked. This function does not add the new folio to the LRU, the
806 * caller must do that.
808 * The remove + add is atomic. This function cannot fail.
810 void replace_page_cache_folio(struct folio *old, struct folio *new)
812 struct address_space *mapping = old->mapping;
813 void (*free_folio)(struct folio *) = mapping->a_ops->free_folio;
814 pgoff_t offset = old->index;
815 XA_STATE(xas, &mapping->i_pages, offset);
817 VM_BUG_ON_FOLIO(!folio_test_locked(old), old);
818 VM_BUG_ON_FOLIO(!folio_test_locked(new), new);
819 VM_BUG_ON_FOLIO(new->mapping, new);
822 new->mapping = mapping;
825 mem_cgroup_migrate(old, new);
828 xas_store(&xas, new);
831 /* hugetlb pages do not participate in page cache accounting. */
832 if (!folio_test_hugetlb(old))
833 __lruvec_stat_sub_folio(old, NR_FILE_PAGES);
834 if (!folio_test_hugetlb(new))
835 __lruvec_stat_add_folio(new, NR_FILE_PAGES);
836 if (folio_test_swapbacked(old))
837 __lruvec_stat_sub_folio(old, NR_SHMEM);
838 if (folio_test_swapbacked(new))
839 __lruvec_stat_add_folio(new, NR_SHMEM);
840 xas_unlock_irq(&xas);
845 EXPORT_SYMBOL_GPL(replace_page_cache_folio);
847 noinline int __filemap_add_folio(struct address_space *mapping,
848 struct folio *folio, pgoff_t index, gfp_t gfp, void **shadowp)
850 XA_STATE(xas, &mapping->i_pages, index);
851 int huge = folio_test_hugetlb(folio);
852 bool charged = false;
855 VM_BUG_ON_FOLIO(!folio_test_locked(folio), folio);
856 VM_BUG_ON_FOLIO(folio_test_swapbacked(folio), folio);
857 mapping_set_update(&xas, mapping);
860 int error = mem_cgroup_charge(folio, NULL, gfp);
861 VM_BUG_ON_FOLIO(index & (folio_nr_pages(folio) - 1), folio);
865 xas_set_order(&xas, index, folio_order(folio));
866 nr = folio_nr_pages(folio);
869 gfp &= GFP_RECLAIM_MASK;
870 folio_ref_add(folio, nr);
871 folio->mapping = mapping;
872 folio->index = xas.xa_index;
875 unsigned int order = xa_get_order(xas.xa, xas.xa_index);
876 void *entry, *old = NULL;
878 if (order > folio_order(folio))
879 xas_split_alloc(&xas, xa_load(xas.xa, xas.xa_index),
882 xas_for_each_conflict(&xas, entry) {
884 if (!xa_is_value(entry)) {
885 xas_set_err(&xas, -EEXIST);
893 /* entry may have been split before we acquired lock */
894 order = xa_get_order(xas.xa, xas.xa_index);
895 if (order > folio_order(folio)) {
896 /* How to handle large swap entries? */
897 BUG_ON(shmem_mapping(mapping));
898 xas_split(&xas, old, order);
903 xas_store(&xas, folio);
907 mapping->nrpages += nr;
909 /* hugetlb pages do not participate in page cache accounting */
911 __lruvec_stat_mod_folio(folio, NR_FILE_PAGES, nr);
912 if (folio_test_pmd_mappable(folio))
913 __lruvec_stat_mod_folio(folio,
917 xas_unlock_irq(&xas);
918 } while (xas_nomem(&xas, gfp));
923 trace_mm_filemap_add_to_page_cache(folio);
927 mem_cgroup_uncharge(folio);
928 folio->mapping = NULL;
929 /* Leave page->index set: truncation relies upon it */
930 folio_put_refs(folio, nr);
931 return xas_error(&xas);
933 ALLOW_ERROR_INJECTION(__filemap_add_folio, ERRNO);
935 int filemap_add_folio(struct address_space *mapping, struct folio *folio,
936 pgoff_t index, gfp_t gfp)
941 __folio_set_locked(folio);
942 ret = __filemap_add_folio(mapping, folio, index, gfp, &shadow);
944 __folio_clear_locked(folio);
947 * The folio might have been evicted from cache only
948 * recently, in which case it should be activated like
949 * any other repeatedly accessed folio.
950 * The exception is folios getting rewritten; evicting other
951 * data from the working set, only to cache data that will
952 * get overwritten with something else, is a waste of memory.
954 WARN_ON_ONCE(folio_test_active(folio));
955 if (!(gfp & __GFP_WRITE) && shadow)
956 workingset_refault(folio, shadow);
957 folio_add_lru(folio);
961 EXPORT_SYMBOL_GPL(filemap_add_folio);
964 struct folio *filemap_alloc_folio(gfp_t gfp, unsigned int order)
969 if (cpuset_do_page_mem_spread()) {
970 unsigned int cpuset_mems_cookie;
972 cpuset_mems_cookie = read_mems_allowed_begin();
973 n = cpuset_mem_spread_node();
974 folio = __folio_alloc_node(gfp, order, n);
975 } while (!folio && read_mems_allowed_retry(cpuset_mems_cookie));
979 return folio_alloc(gfp, order);
981 EXPORT_SYMBOL(filemap_alloc_folio);
985 * filemap_invalidate_lock_two - lock invalidate_lock for two mappings
987 * Lock exclusively invalidate_lock of any passed mapping that is not NULL.
989 * @mapping1: the first mapping to lock
990 * @mapping2: the second mapping to lock
992 void filemap_invalidate_lock_two(struct address_space *mapping1,
993 struct address_space *mapping2)
995 if (mapping1 > mapping2)
996 swap(mapping1, mapping2);
998 down_write(&mapping1->invalidate_lock);
999 if (mapping2 && mapping1 != mapping2)
1000 down_write_nested(&mapping2->invalidate_lock, 1);
1002 EXPORT_SYMBOL(filemap_invalidate_lock_two);
1005 * filemap_invalidate_unlock_two - unlock invalidate_lock for two mappings
1007 * Unlock exclusive invalidate_lock of any passed mapping that is not NULL.
1009 * @mapping1: the first mapping to unlock
1010 * @mapping2: the second mapping to unlock
1012 void filemap_invalidate_unlock_two(struct address_space *mapping1,
1013 struct address_space *mapping2)
1016 up_write(&mapping1->invalidate_lock);
1017 if (mapping2 && mapping1 != mapping2)
1018 up_write(&mapping2->invalidate_lock);
1020 EXPORT_SYMBOL(filemap_invalidate_unlock_two);
1023 * In order to wait for pages to become available there must be
1024 * waitqueues associated with pages. By using a hash table of
1025 * waitqueues where the bucket discipline is to maintain all
1026 * waiters on the same queue and wake all when any of the pages
1027 * become available, and for the woken contexts to check to be
1028 * sure the appropriate page became available, this saves space
1029 * at a cost of "thundering herd" phenomena during rare hash
1032 #define PAGE_WAIT_TABLE_BITS 8
1033 #define PAGE_WAIT_TABLE_SIZE (1 << PAGE_WAIT_TABLE_BITS)
1034 static wait_queue_head_t folio_wait_table[PAGE_WAIT_TABLE_SIZE] __cacheline_aligned;
1036 static wait_queue_head_t *folio_waitqueue(struct folio *folio)
1038 return &folio_wait_table[hash_ptr(folio, PAGE_WAIT_TABLE_BITS)];
1041 void __init pagecache_init(void)
1045 for (i = 0; i < PAGE_WAIT_TABLE_SIZE; i++)
1046 init_waitqueue_head(&folio_wait_table[i]);
1048 page_writeback_init();
1052 * The page wait code treats the "wait->flags" somewhat unusually, because
1053 * we have multiple different kinds of waits, not just the usual "exclusive"
1058 * (a) no special bits set:
1060 * We're just waiting for the bit to be released, and when a waker
1061 * calls the wakeup function, we set WQ_FLAG_WOKEN and wake it up,
1062 * and remove it from the wait queue.
1064 * Simple and straightforward.
1066 * (b) WQ_FLAG_EXCLUSIVE:
1068 * The waiter is waiting to get the lock, and only one waiter should
1069 * be woken up to avoid any thundering herd behavior. We'll set the
1070 * WQ_FLAG_WOKEN bit, wake it up, and remove it from the wait queue.
1072 * This is the traditional exclusive wait.
1074 * (c) WQ_FLAG_EXCLUSIVE | WQ_FLAG_CUSTOM:
1076 * The waiter is waiting to get the bit, and additionally wants the
1077 * lock to be transferred to it for fair lock behavior. If the lock
1078 * cannot be taken, we stop walking the wait queue without waking
1081 * This is the "fair lock handoff" case, and in addition to setting
1082 * WQ_FLAG_WOKEN, we set WQ_FLAG_DONE to let the waiter easily see
1083 * that it now has the lock.
1085 static int wake_page_function(wait_queue_entry_t *wait, unsigned mode, int sync, void *arg)
1088 struct wait_page_key *key = arg;
1089 struct wait_page_queue *wait_page
1090 = container_of(wait, struct wait_page_queue, wait);
1092 if (!wake_page_match(wait_page, key))
1096 * If it's a lock handoff wait, we get the bit for it, and
1097 * stop walking (and do not wake it up) if we can't.
1099 flags = wait->flags;
1100 if (flags & WQ_FLAG_EXCLUSIVE) {
1101 if (test_bit(key->bit_nr, &key->folio->flags))
1103 if (flags & WQ_FLAG_CUSTOM) {
1104 if (test_and_set_bit(key->bit_nr, &key->folio->flags))
1106 flags |= WQ_FLAG_DONE;
1111 * We are holding the wait-queue lock, but the waiter that
1112 * is waiting for this will be checking the flags without
1115 * So update the flags atomically, and wake up the waiter
1116 * afterwards to avoid any races. This store-release pairs
1117 * with the load-acquire in folio_wait_bit_common().
1119 smp_store_release(&wait->flags, flags | WQ_FLAG_WOKEN);
1120 wake_up_state(wait->private, mode);
1123 * Ok, we have successfully done what we're waiting for,
1124 * and we can unconditionally remove the wait entry.
1126 * Note that this pairs with the "finish_wait()" in the
1127 * waiter, and has to be the absolute last thing we do.
1128 * After this list_del_init(&wait->entry) the wait entry
1129 * might be de-allocated and the process might even have
1132 list_del_init_careful(&wait->entry);
1133 return (flags & WQ_FLAG_EXCLUSIVE) != 0;
1136 static void folio_wake_bit(struct folio *folio, int bit_nr)
1138 wait_queue_head_t *q = folio_waitqueue(folio);
1139 struct wait_page_key key;
1140 unsigned long flags;
1141 wait_queue_entry_t bookmark;
1144 key.bit_nr = bit_nr;
1148 bookmark.private = NULL;
1149 bookmark.func = NULL;
1150 INIT_LIST_HEAD(&bookmark.entry);
1152 spin_lock_irqsave(&q->lock, flags);
1153 __wake_up_locked_key_bookmark(q, TASK_NORMAL, &key, &bookmark);
1155 while (bookmark.flags & WQ_FLAG_BOOKMARK) {
1157 * Take a breather from holding the lock,
1158 * allow pages that finish wake up asynchronously
1159 * to acquire the lock and remove themselves
1162 spin_unlock_irqrestore(&q->lock, flags);
1164 spin_lock_irqsave(&q->lock, flags);
1165 __wake_up_locked_key_bookmark(q, TASK_NORMAL, &key, &bookmark);
1169 * It's possible to miss clearing waiters here, when we woke our page
1170 * waiters, but the hashed waitqueue has waiters for other pages on it.
1171 * That's okay, it's a rare case. The next waker will clear it.
1173 * Note that, depending on the page pool (buddy, hugetlb, ZONE_DEVICE,
1174 * other), the flag may be cleared in the course of freeing the page;
1175 * but that is not required for correctness.
1177 if (!waitqueue_active(q) || !key.page_match)
1178 folio_clear_waiters(folio);
1180 spin_unlock_irqrestore(&q->lock, flags);
1183 static void folio_wake(struct folio *folio, int bit)
1185 if (!folio_test_waiters(folio))
1187 folio_wake_bit(folio, bit);
1191 * A choice of three behaviors for folio_wait_bit_common():
1194 EXCLUSIVE, /* Hold ref to page and take the bit when woken, like
1195 * __folio_lock() waiting on then setting PG_locked.
1197 SHARED, /* Hold ref to page and check the bit when woken, like
1198 * folio_wait_writeback() waiting on PG_writeback.
1200 DROP, /* Drop ref to page before wait, no check when woken,
1201 * like folio_put_wait_locked() on PG_locked.
1206 * Attempt to check (or get) the folio flag, and mark us done
1209 static inline bool folio_trylock_flag(struct folio *folio, int bit_nr,
1210 struct wait_queue_entry *wait)
1212 if (wait->flags & WQ_FLAG_EXCLUSIVE) {
1213 if (test_and_set_bit(bit_nr, &folio->flags))
1215 } else if (test_bit(bit_nr, &folio->flags))
1218 wait->flags |= WQ_FLAG_WOKEN | WQ_FLAG_DONE;
1222 /* How many times do we accept lock stealing from under a waiter? */
1223 int sysctl_page_lock_unfairness = 5;
1225 static inline int folio_wait_bit_common(struct folio *folio, int bit_nr,
1226 int state, enum behavior behavior)
1228 wait_queue_head_t *q = folio_waitqueue(folio);
1229 int unfairness = sysctl_page_lock_unfairness;
1230 struct wait_page_queue wait_page;
1231 wait_queue_entry_t *wait = &wait_page.wait;
1232 bool thrashing = false;
1233 unsigned long pflags;
1236 if (bit_nr == PG_locked &&
1237 !folio_test_uptodate(folio) && folio_test_workingset(folio)) {
1238 delayacct_thrashing_start(&in_thrashing);
1239 psi_memstall_enter(&pflags);
1244 wait->func = wake_page_function;
1245 wait_page.folio = folio;
1246 wait_page.bit_nr = bit_nr;
1250 if (behavior == EXCLUSIVE) {
1251 wait->flags = WQ_FLAG_EXCLUSIVE;
1252 if (--unfairness < 0)
1253 wait->flags |= WQ_FLAG_CUSTOM;
1257 * Do one last check whether we can get the
1258 * page bit synchronously.
1260 * Do the folio_set_waiters() marking before that
1261 * to let any waker we _just_ missed know they
1262 * need to wake us up (otherwise they'll never
1263 * even go to the slow case that looks at the
1264 * page queue), and add ourselves to the wait
1265 * queue if we need to sleep.
1267 * This part needs to be done under the queue
1268 * lock to avoid races.
1270 spin_lock_irq(&q->lock);
1271 folio_set_waiters(folio);
1272 if (!folio_trylock_flag(folio, bit_nr, wait))
1273 __add_wait_queue_entry_tail(q, wait);
1274 spin_unlock_irq(&q->lock);
1277 * From now on, all the logic will be based on
1278 * the WQ_FLAG_WOKEN and WQ_FLAG_DONE flag, to
1279 * see whether the page bit testing has already
1280 * been done by the wake function.
1282 * We can drop our reference to the folio.
1284 if (behavior == DROP)
1288 * Note that until the "finish_wait()", or until
1289 * we see the WQ_FLAG_WOKEN flag, we need to
1290 * be very careful with the 'wait->flags', because
1291 * we may race with a waker that sets them.
1296 set_current_state(state);
1298 /* Loop until we've been woken or interrupted */
1299 flags = smp_load_acquire(&wait->flags);
1300 if (!(flags & WQ_FLAG_WOKEN)) {
1301 if (signal_pending_state(state, current))
1308 /* If we were non-exclusive, we're done */
1309 if (behavior != EXCLUSIVE)
1312 /* If the waker got the lock for us, we're done */
1313 if (flags & WQ_FLAG_DONE)
1317 * Otherwise, if we're getting the lock, we need to
1318 * try to get it ourselves.
1320 * And if that fails, we'll have to retry this all.
1322 if (unlikely(test_and_set_bit(bit_nr, folio_flags(folio, 0))))
1325 wait->flags |= WQ_FLAG_DONE;
1330 * If a signal happened, this 'finish_wait()' may remove the last
1331 * waiter from the wait-queues, but the folio waiters bit will remain
1332 * set. That's ok. The next wakeup will take care of it, and trying
1333 * to do it here would be difficult and prone to races.
1335 finish_wait(q, wait);
1338 delayacct_thrashing_end(&in_thrashing);
1339 psi_memstall_leave(&pflags);
1343 * NOTE! The wait->flags weren't stable until we've done the
1344 * 'finish_wait()', and we could have exited the loop above due
1345 * to a signal, and had a wakeup event happen after the signal
1346 * test but before the 'finish_wait()'.
1348 * So only after the finish_wait() can we reliably determine
1349 * if we got woken up or not, so we can now figure out the final
1350 * return value based on that state without races.
1352 * Also note that WQ_FLAG_WOKEN is sufficient for a non-exclusive
1353 * waiter, but an exclusive one requires WQ_FLAG_DONE.
1355 if (behavior == EXCLUSIVE)
1356 return wait->flags & WQ_FLAG_DONE ? 0 : -EINTR;
1358 return wait->flags & WQ_FLAG_WOKEN ? 0 : -EINTR;
1361 #ifdef CONFIG_MIGRATION
1363 * migration_entry_wait_on_locked - Wait for a migration entry to be removed
1364 * @entry: migration swap entry.
1365 * @ptep: mapped pte pointer. Will return with the ptep unmapped. Only required
1366 * for pte entries, pass NULL for pmd entries.
1367 * @ptl: already locked ptl. This function will drop the lock.
1369 * Wait for a migration entry referencing the given page to be removed. This is
1370 * equivalent to put_and_wait_on_page_locked(page, TASK_UNINTERRUPTIBLE) except
1371 * this can be called without taking a reference on the page. Instead this
1372 * should be called while holding the ptl for the migration entry referencing
1375 * Returns after unmapping and unlocking the pte/ptl with pte_unmap_unlock().
1377 * This follows the same logic as folio_wait_bit_common() so see the comments
1380 void migration_entry_wait_on_locked(swp_entry_t entry, pte_t *ptep,
1383 struct wait_page_queue wait_page;
1384 wait_queue_entry_t *wait = &wait_page.wait;
1385 bool thrashing = false;
1386 unsigned long pflags;
1388 wait_queue_head_t *q;
1389 struct folio *folio = page_folio(pfn_swap_entry_to_page(entry));
1391 q = folio_waitqueue(folio);
1392 if (!folio_test_uptodate(folio) && folio_test_workingset(folio)) {
1393 delayacct_thrashing_start(&in_thrashing);
1394 psi_memstall_enter(&pflags);
1399 wait->func = wake_page_function;
1400 wait_page.folio = folio;
1401 wait_page.bit_nr = PG_locked;
1404 spin_lock_irq(&q->lock);
1405 folio_set_waiters(folio);
1406 if (!folio_trylock_flag(folio, PG_locked, wait))
1407 __add_wait_queue_entry_tail(q, wait);
1408 spin_unlock_irq(&q->lock);
1411 * If a migration entry exists for the page the migration path must hold
1412 * a valid reference to the page, and it must take the ptl to remove the
1413 * migration entry. So the page is valid until the ptl is dropped.
1416 pte_unmap_unlock(ptep, ptl);
1423 set_current_state(TASK_UNINTERRUPTIBLE);
1425 /* Loop until we've been woken or interrupted */
1426 flags = smp_load_acquire(&wait->flags);
1427 if (!(flags & WQ_FLAG_WOKEN)) {
1428 if (signal_pending_state(TASK_UNINTERRUPTIBLE, current))
1437 finish_wait(q, wait);
1440 delayacct_thrashing_end(&in_thrashing);
1441 psi_memstall_leave(&pflags);
1446 void folio_wait_bit(struct folio *folio, int bit_nr)
1448 folio_wait_bit_common(folio, bit_nr, TASK_UNINTERRUPTIBLE, SHARED);
1450 EXPORT_SYMBOL(folio_wait_bit);
1452 int folio_wait_bit_killable(struct folio *folio, int bit_nr)
1454 return folio_wait_bit_common(folio, bit_nr, TASK_KILLABLE, SHARED);
1456 EXPORT_SYMBOL(folio_wait_bit_killable);
1459 * folio_put_wait_locked - Drop a reference and wait for it to be unlocked
1460 * @folio: The folio to wait for.
1461 * @state: The sleep state (TASK_KILLABLE, TASK_UNINTERRUPTIBLE, etc).
1463 * The caller should hold a reference on @folio. They expect the page to
1464 * become unlocked relatively soon, but do not wish to hold up migration
1465 * (for example) by holding the reference while waiting for the folio to
1466 * come unlocked. After this function returns, the caller should not
1467 * dereference @folio.
1469 * Return: 0 if the folio was unlocked or -EINTR if interrupted by a signal.
1471 static int folio_put_wait_locked(struct folio *folio, int state)
1473 return folio_wait_bit_common(folio, PG_locked, state, DROP);
1477 * folio_add_wait_queue - Add an arbitrary waiter to a folio's wait queue
1478 * @folio: Folio defining the wait queue of interest
1479 * @waiter: Waiter to add to the queue
1481 * Add an arbitrary @waiter to the wait queue for the nominated @folio.
1483 void folio_add_wait_queue(struct folio *folio, wait_queue_entry_t *waiter)
1485 wait_queue_head_t *q = folio_waitqueue(folio);
1486 unsigned long flags;
1488 spin_lock_irqsave(&q->lock, flags);
1489 __add_wait_queue_entry_tail(q, waiter);
1490 folio_set_waiters(folio);
1491 spin_unlock_irqrestore(&q->lock, flags);
1493 EXPORT_SYMBOL_GPL(folio_add_wait_queue);
1495 #ifndef clear_bit_unlock_is_negative_byte
1498 * PG_waiters is the high bit in the same byte as PG_lock.
1500 * On x86 (and on many other architectures), we can clear PG_lock and
1501 * test the sign bit at the same time. But if the architecture does
1502 * not support that special operation, we just do this all by hand
1505 * The read of PG_waiters has to be after (or concurrently with) PG_locked
1506 * being cleared, but a memory barrier should be unnecessary since it is
1507 * in the same byte as PG_locked.
1509 static inline bool clear_bit_unlock_is_negative_byte(long nr, volatile void *mem)
1511 clear_bit_unlock(nr, mem);
1512 /* smp_mb__after_atomic(); */
1513 return test_bit(PG_waiters, mem);
1519 * folio_unlock - Unlock a locked folio.
1520 * @folio: The folio.
1522 * Unlocks the folio and wakes up any thread sleeping on the page lock.
1524 * Context: May be called from interrupt or process context. May not be
1525 * called from NMI context.
1527 void folio_unlock(struct folio *folio)
1529 /* Bit 7 allows x86 to check the byte's sign bit */
1530 BUILD_BUG_ON(PG_waiters != 7);
1531 BUILD_BUG_ON(PG_locked > 7);
1532 VM_BUG_ON_FOLIO(!folio_test_locked(folio), folio);
1533 if (clear_bit_unlock_is_negative_byte(PG_locked, folio_flags(folio, 0)))
1534 folio_wake_bit(folio, PG_locked);
1536 EXPORT_SYMBOL(folio_unlock);
1539 * folio_end_private_2 - Clear PG_private_2 and wake any waiters.
1540 * @folio: The folio.
1542 * Clear the PG_private_2 bit on a folio and wake up any sleepers waiting for
1543 * it. The folio reference held for PG_private_2 being set is released.
1545 * This is, for example, used when a netfs folio is being written to a local
1546 * disk cache, thereby allowing writes to the cache for the same folio to be
1549 void folio_end_private_2(struct folio *folio)
1551 VM_BUG_ON_FOLIO(!folio_test_private_2(folio), folio);
1552 clear_bit_unlock(PG_private_2, folio_flags(folio, 0));
1553 folio_wake_bit(folio, PG_private_2);
1556 EXPORT_SYMBOL(folio_end_private_2);
1559 * folio_wait_private_2 - Wait for PG_private_2 to be cleared on a folio.
1560 * @folio: The folio to wait on.
1562 * Wait for PG_private_2 (aka PG_fscache) to be cleared on a folio.
1564 void folio_wait_private_2(struct folio *folio)
1566 while (folio_test_private_2(folio))
1567 folio_wait_bit(folio, PG_private_2);
1569 EXPORT_SYMBOL(folio_wait_private_2);
1572 * folio_wait_private_2_killable - Wait for PG_private_2 to be cleared on a folio.
1573 * @folio: The folio to wait on.
1575 * Wait for PG_private_2 (aka PG_fscache) to be cleared on a folio or until a
1576 * fatal signal is received by the calling task.
1579 * - 0 if successful.
1580 * - -EINTR if a fatal signal was encountered.
1582 int folio_wait_private_2_killable(struct folio *folio)
1586 while (folio_test_private_2(folio)) {
1587 ret = folio_wait_bit_killable(folio, PG_private_2);
1594 EXPORT_SYMBOL(folio_wait_private_2_killable);
1597 * folio_end_writeback - End writeback against a folio.
1598 * @folio: The folio.
1600 void folio_end_writeback(struct folio *folio)
1603 * folio_test_clear_reclaim() could be used here but it is an
1604 * atomic operation and overkill in this particular case. Failing
1605 * to shuffle a folio marked for immediate reclaim is too mild
1606 * a gain to justify taking an atomic operation penalty at the
1607 * end of every folio writeback.
1609 if (folio_test_reclaim(folio)) {
1610 folio_clear_reclaim(folio);
1611 folio_rotate_reclaimable(folio);
1615 * Writeback does not hold a folio reference of its own, relying
1616 * on truncation to wait for the clearing of PG_writeback.
1617 * But here we must make sure that the folio is not freed and
1618 * reused before the folio_wake().
1621 if (!__folio_end_writeback(folio))
1624 smp_mb__after_atomic();
1625 folio_wake(folio, PG_writeback);
1626 acct_reclaim_writeback(folio);
1629 EXPORT_SYMBOL(folio_end_writeback);
1632 * __folio_lock - Get a lock on the folio, assuming we need to sleep to get it.
1633 * @folio: The folio to lock
1635 void __folio_lock(struct folio *folio)
1637 folio_wait_bit_common(folio, PG_locked, TASK_UNINTERRUPTIBLE,
1640 EXPORT_SYMBOL(__folio_lock);
1642 int __folio_lock_killable(struct folio *folio)
1644 return folio_wait_bit_common(folio, PG_locked, TASK_KILLABLE,
1647 EXPORT_SYMBOL_GPL(__folio_lock_killable);
1649 static int __folio_lock_async(struct folio *folio, struct wait_page_queue *wait)
1651 struct wait_queue_head *q = folio_waitqueue(folio);
1654 wait->folio = folio;
1655 wait->bit_nr = PG_locked;
1657 spin_lock_irq(&q->lock);
1658 __add_wait_queue_entry_tail(q, &wait->wait);
1659 folio_set_waiters(folio);
1660 ret = !folio_trylock(folio);
1662 * If we were successful now, we know we're still on the
1663 * waitqueue as we're still under the lock. This means it's
1664 * safe to remove and return success, we know the callback
1665 * isn't going to trigger.
1668 __remove_wait_queue(q, &wait->wait);
1671 spin_unlock_irq(&q->lock);
1677 * true - folio is locked; mmap_lock is still held.
1678 * false - folio is not locked.
1679 * mmap_lock has been released (mmap_read_unlock(), unless flags had both
1680 * FAULT_FLAG_ALLOW_RETRY and FAULT_FLAG_RETRY_NOWAIT set, in
1681 * which case mmap_lock is still held.
1683 * If neither ALLOW_RETRY nor KILLABLE are set, will always return true
1684 * with the folio locked and the mmap_lock unperturbed.
1686 bool __folio_lock_or_retry(struct folio *folio, struct mm_struct *mm,
1689 if (fault_flag_allow_retry_first(flags)) {
1691 * CAUTION! In this case, mmap_lock is not released
1692 * even though return 0.
1694 if (flags & FAULT_FLAG_RETRY_NOWAIT)
1697 mmap_read_unlock(mm);
1698 if (flags & FAULT_FLAG_KILLABLE)
1699 folio_wait_locked_killable(folio);
1701 folio_wait_locked(folio);
1704 if (flags & FAULT_FLAG_KILLABLE) {
1707 ret = __folio_lock_killable(folio);
1709 mmap_read_unlock(mm);
1713 __folio_lock(folio);
1720 * page_cache_next_miss() - Find the next gap in the page cache.
1721 * @mapping: Mapping.
1723 * @max_scan: Maximum range to search.
1725 * Search the range [index, min(index + max_scan - 1, ULONG_MAX)] for the
1726 * gap with the lowest index.
1728 * This function may be called under the rcu_read_lock. However, this will
1729 * not atomically search a snapshot of the cache at a single point in time.
1730 * For example, if a gap is created at index 5, then subsequently a gap is
1731 * created at index 10, page_cache_next_miss covering both indices may
1732 * return 10 if called under the rcu_read_lock.
1734 * Return: The index of the gap if found, otherwise an index outside the
1735 * range specified (in which case 'return - index >= max_scan' will be true).
1736 * In the rare case of index wrap-around, 0 will be returned.
1738 pgoff_t page_cache_next_miss(struct address_space *mapping,
1739 pgoff_t index, unsigned long max_scan)
1741 XA_STATE(xas, &mapping->i_pages, index);
1743 while (max_scan--) {
1744 void *entry = xas_next(&xas);
1745 if (!entry || xa_is_value(entry))
1747 if (xas.xa_index == 0)
1751 return xas.xa_index;
1753 EXPORT_SYMBOL(page_cache_next_miss);
1756 * page_cache_prev_miss() - Find the previous gap in the page cache.
1757 * @mapping: Mapping.
1759 * @max_scan: Maximum range to search.
1761 * Search the range [max(index - max_scan + 1, 0), index] for the
1762 * gap with the highest index.
1764 * This function may be called under the rcu_read_lock. However, this will
1765 * not atomically search a snapshot of the cache at a single point in time.
1766 * For example, if a gap is created at index 10, then subsequently a gap is
1767 * created at index 5, page_cache_prev_miss() covering both indices may
1768 * return 5 if called under the rcu_read_lock.
1770 * Return: The index of the gap if found, otherwise an index outside the
1771 * range specified (in which case 'index - return >= max_scan' will be true).
1772 * In the rare case of wrap-around, ULONG_MAX will be returned.
1774 pgoff_t page_cache_prev_miss(struct address_space *mapping,
1775 pgoff_t index, unsigned long max_scan)
1777 XA_STATE(xas, &mapping->i_pages, index);
1779 while (max_scan--) {
1780 void *entry = xas_prev(&xas);
1781 if (!entry || xa_is_value(entry))
1783 if (xas.xa_index == ULONG_MAX)
1787 return xas.xa_index;
1789 EXPORT_SYMBOL(page_cache_prev_miss);
1792 * Lockless page cache protocol:
1793 * On the lookup side:
1794 * 1. Load the folio from i_pages
1795 * 2. Increment the refcount if it's not zero
1796 * 3. If the folio is not found by xas_reload(), put the refcount and retry
1798 * On the removal side:
1799 * A. Freeze the page (by zeroing the refcount if nobody else has a reference)
1800 * B. Remove the page from i_pages
1801 * C. Return the page to the page allocator
1803 * This means that any page may have its reference count temporarily
1804 * increased by a speculative page cache (or fast GUP) lookup as it can
1805 * be allocated by another user before the RCU grace period expires.
1806 * Because the refcount temporarily acquired here may end up being the
1807 * last refcount on the page, any page allocation must be freeable by
1812 * filemap_get_entry - Get a page cache entry.
1813 * @mapping: the address_space to search
1814 * @index: The page cache index.
1816 * Looks up the page cache entry at @mapping & @index. If it is a folio,
1817 * it is returned with an increased refcount. If it is a shadow entry
1818 * of a previously evicted folio, or a swap entry from shmem/tmpfs,
1819 * it is returned without further action.
1821 * Return: The folio, swap or shadow entry, %NULL if nothing is found.
1823 void *filemap_get_entry(struct address_space *mapping, pgoff_t index)
1825 XA_STATE(xas, &mapping->i_pages, index);
1826 struct folio *folio;
1831 folio = xas_load(&xas);
1832 if (xas_retry(&xas, folio))
1835 * A shadow entry of a recently evicted page, or a swap entry from
1836 * shmem/tmpfs. Return it without attempting to raise page count.
1838 if (!folio || xa_is_value(folio))
1841 if (!folio_try_get_rcu(folio))
1844 if (unlikely(folio != xas_reload(&xas))) {
1855 * __filemap_get_folio - Find and get a reference to a folio.
1856 * @mapping: The address_space to search.
1857 * @index: The page index.
1858 * @fgp_flags: %FGP flags modify how the folio is returned.
1859 * @gfp: Memory allocation flags to use if %FGP_CREAT is specified.
1861 * Looks up the page cache entry at @mapping & @index.
1863 * @fgp_flags can be zero or more of these flags:
1865 * * %FGP_ACCESSED - The folio will be marked accessed.
1866 * * %FGP_LOCK - The folio is returned locked.
1867 * * %FGP_CREAT - If no page is present then a new page is allocated using
1868 * @gfp and added to the page cache and the VM's LRU list.
1869 * The page is returned locked and with an increased refcount.
1870 * * %FGP_FOR_MMAP - The caller wants to do its own locking dance if the
1871 * page is already in cache. If the page was allocated, unlock it before
1872 * returning so the caller can do the same dance.
1873 * * %FGP_WRITE - The page will be written to by the caller.
1874 * * %FGP_NOFS - __GFP_FS will get cleared in gfp.
1875 * * %FGP_NOWAIT - Don't get blocked by page lock.
1876 * * %FGP_STABLE - Wait for the folio to be stable (finished writeback)
1878 * If %FGP_LOCK or %FGP_CREAT are specified then the function may sleep even
1879 * if the %GFP flags specified for %FGP_CREAT are atomic.
1881 * If there is a page cache page, it is returned with an increased refcount.
1883 * Return: The found folio or an ERR_PTR() otherwise.
1885 struct folio *__filemap_get_folio(struct address_space *mapping, pgoff_t index,
1886 int fgp_flags, gfp_t gfp)
1888 struct folio *folio;
1891 folio = filemap_get_entry(mapping, index);
1892 if (xa_is_value(folio))
1897 if (fgp_flags & FGP_LOCK) {
1898 if (fgp_flags & FGP_NOWAIT) {
1899 if (!folio_trylock(folio)) {
1901 return ERR_PTR(-EAGAIN);
1907 /* Has the page been truncated? */
1908 if (unlikely(folio->mapping != mapping)) {
1909 folio_unlock(folio);
1913 VM_BUG_ON_FOLIO(!folio_contains(folio, index), folio);
1916 if (fgp_flags & FGP_ACCESSED)
1917 folio_mark_accessed(folio);
1918 else if (fgp_flags & FGP_WRITE) {
1919 /* Clear idle flag for buffer write */
1920 if (folio_test_idle(folio))
1921 folio_clear_idle(folio);
1924 if (fgp_flags & FGP_STABLE)
1925 folio_wait_stable(folio);
1927 if (!folio && (fgp_flags & FGP_CREAT)) {
1929 if ((fgp_flags & FGP_WRITE) && mapping_can_writeback(mapping))
1931 if (fgp_flags & FGP_NOFS)
1933 if (fgp_flags & FGP_NOWAIT) {
1935 gfp |= GFP_NOWAIT | __GFP_NOWARN;
1938 folio = filemap_alloc_folio(gfp, 0);
1940 return ERR_PTR(-ENOMEM);
1942 if (WARN_ON_ONCE(!(fgp_flags & (FGP_LOCK | FGP_FOR_MMAP))))
1943 fgp_flags |= FGP_LOCK;
1945 /* Init accessed so avoid atomic mark_page_accessed later */
1946 if (fgp_flags & FGP_ACCESSED)
1947 __folio_set_referenced(folio);
1949 err = filemap_add_folio(mapping, folio, index, gfp);
1950 if (unlikely(err)) {
1958 * filemap_add_folio locks the page, and for mmap
1959 * we expect an unlocked page.
1961 if (folio && (fgp_flags & FGP_FOR_MMAP))
1962 folio_unlock(folio);
1966 return ERR_PTR(-ENOENT);
1969 EXPORT_SYMBOL(__filemap_get_folio);
1971 static inline struct folio *find_get_entry(struct xa_state *xas, pgoff_t max,
1974 struct folio *folio;
1977 if (mark == XA_PRESENT)
1978 folio = xas_find(xas, max);
1980 folio = xas_find_marked(xas, max, mark);
1982 if (xas_retry(xas, folio))
1985 * A shadow entry of a recently evicted page, a swap
1986 * entry from shmem/tmpfs or a DAX entry. Return it
1987 * without attempting to raise page count.
1989 if (!folio || xa_is_value(folio))
1992 if (!folio_try_get_rcu(folio))
1995 if (unlikely(folio != xas_reload(xas))) {
2007 * find_get_entries - gang pagecache lookup
2008 * @mapping: The address_space to search
2009 * @start: The starting page cache index
2010 * @end: The final page index (inclusive).
2011 * @fbatch: Where the resulting entries are placed.
2012 * @indices: The cache indices corresponding to the entries in @entries
2014 * find_get_entries() will search for and return a batch of entries in
2015 * the mapping. The entries are placed in @fbatch. find_get_entries()
2016 * takes a reference on any actual folios it returns.
2018 * The entries have ascending indexes. The indices may not be consecutive
2019 * due to not-present entries or large folios.
2021 * Any shadow entries of evicted folios, or swap entries from
2022 * shmem/tmpfs, are included in the returned array.
2024 * Return: The number of entries which were found.
2026 unsigned find_get_entries(struct address_space *mapping, pgoff_t *start,
2027 pgoff_t end, struct folio_batch *fbatch, pgoff_t *indices)
2029 XA_STATE(xas, &mapping->i_pages, *start);
2030 struct folio *folio;
2033 while ((folio = find_get_entry(&xas, end, XA_PRESENT)) != NULL) {
2034 indices[fbatch->nr] = xas.xa_index;
2035 if (!folio_batch_add(fbatch, folio))
2040 if (folio_batch_count(fbatch)) {
2041 unsigned long nr = 1;
2042 int idx = folio_batch_count(fbatch) - 1;
2044 folio = fbatch->folios[idx];
2045 if (!xa_is_value(folio) && !folio_test_hugetlb(folio))
2046 nr = folio_nr_pages(folio);
2047 *start = indices[idx] + nr;
2049 return folio_batch_count(fbatch);
2053 * find_lock_entries - Find a batch of pagecache entries.
2054 * @mapping: The address_space to search.
2055 * @start: The starting page cache index.
2056 * @end: The final page index (inclusive).
2057 * @fbatch: Where the resulting entries are placed.
2058 * @indices: The cache indices of the entries in @fbatch.
2060 * find_lock_entries() will return a batch of entries from @mapping.
2061 * Swap, shadow and DAX entries are included. Folios are returned
2062 * locked and with an incremented refcount. Folios which are locked
2063 * by somebody else or under writeback are skipped. Folios which are
2064 * partially outside the range are not returned.
2066 * The entries have ascending indexes. The indices may not be consecutive
2067 * due to not-present entries, large folios, folios which could not be
2068 * locked or folios under writeback.
2070 * Return: The number of entries which were found.
2072 unsigned find_lock_entries(struct address_space *mapping, pgoff_t *start,
2073 pgoff_t end, struct folio_batch *fbatch, pgoff_t *indices)
2075 XA_STATE(xas, &mapping->i_pages, *start);
2076 struct folio *folio;
2079 while ((folio = find_get_entry(&xas, end, XA_PRESENT))) {
2080 if (!xa_is_value(folio)) {
2081 if (folio->index < *start)
2083 if (folio->index + folio_nr_pages(folio) - 1 > end)
2085 if (!folio_trylock(folio))
2087 if (folio->mapping != mapping ||
2088 folio_test_writeback(folio))
2090 VM_BUG_ON_FOLIO(!folio_contains(folio, xas.xa_index),
2093 indices[fbatch->nr] = xas.xa_index;
2094 if (!folio_batch_add(fbatch, folio))
2098 folio_unlock(folio);
2104 if (folio_batch_count(fbatch)) {
2105 unsigned long nr = 1;
2106 int idx = folio_batch_count(fbatch) - 1;
2108 folio = fbatch->folios[idx];
2109 if (!xa_is_value(folio) && !folio_test_hugetlb(folio))
2110 nr = folio_nr_pages(folio);
2111 *start = indices[idx] + nr;
2113 return folio_batch_count(fbatch);
2117 * filemap_get_folios - Get a batch of folios
2118 * @mapping: The address_space to search
2119 * @start: The starting page index
2120 * @end: The final page index (inclusive)
2121 * @fbatch: The batch to fill.
2123 * Search for and return a batch of folios in the mapping starting at
2124 * index @start and up to index @end (inclusive). The folios are returned
2125 * in @fbatch with an elevated reference count.
2127 * The first folio may start before @start; if it does, it will contain
2128 * @start. The final folio may extend beyond @end; if it does, it will
2129 * contain @end. The folios have ascending indices. There may be gaps
2130 * between the folios if there are indices which have no folio in the
2131 * page cache. If folios are added to or removed from the page cache
2132 * while this is running, they may or may not be found by this call.
2134 * Return: The number of folios which were found.
2135 * We also update @start to index the next folio for the traversal.
2137 unsigned filemap_get_folios(struct address_space *mapping, pgoff_t *start,
2138 pgoff_t end, struct folio_batch *fbatch)
2140 XA_STATE(xas, &mapping->i_pages, *start);
2141 struct folio *folio;
2144 while ((folio = find_get_entry(&xas, end, XA_PRESENT)) != NULL) {
2145 /* Skip over shadow, swap and DAX entries */
2146 if (xa_is_value(folio))
2148 if (!folio_batch_add(fbatch, folio)) {
2149 unsigned long nr = folio_nr_pages(folio);
2151 if (folio_test_hugetlb(folio))
2153 *start = folio->index + nr;
2159 * We come here when there is no page beyond @end. We take care to not
2160 * overflow the index @start as it confuses some of the callers. This
2161 * breaks the iteration when there is a page at index -1 but that is
2162 * already broken anyway.
2164 if (end == (pgoff_t)-1)
2165 *start = (pgoff_t)-1;
2171 return folio_batch_count(fbatch);
2173 EXPORT_SYMBOL(filemap_get_folios);
2176 bool folio_more_pages(struct folio *folio, pgoff_t index, pgoff_t max)
2178 if (!folio_test_large(folio) || folio_test_hugetlb(folio))
2182 return index < folio->index + folio_nr_pages(folio) - 1;
2186 * filemap_get_folios_contig - Get a batch of contiguous folios
2187 * @mapping: The address_space to search
2188 * @start: The starting page index
2189 * @end: The final page index (inclusive)
2190 * @fbatch: The batch to fill
2192 * filemap_get_folios_contig() works exactly like filemap_get_folios(),
2193 * except the returned folios are guaranteed to be contiguous. This may
2194 * not return all contiguous folios if the batch gets filled up.
2196 * Return: The number of folios found.
2197 * Also update @start to be positioned for traversal of the next folio.
2200 unsigned filemap_get_folios_contig(struct address_space *mapping,
2201 pgoff_t *start, pgoff_t end, struct folio_batch *fbatch)
2203 XA_STATE(xas, &mapping->i_pages, *start);
2205 struct folio *folio;
2209 for (folio = xas_load(&xas); folio && xas.xa_index <= end;
2210 folio = xas_next(&xas)) {
2211 if (xas_retry(&xas, folio))
2214 * If the entry has been swapped out, we can stop looking.
2215 * No current caller is looking for DAX entries.
2217 if (xa_is_value(folio))
2220 if (!folio_try_get_rcu(folio))
2223 if (unlikely(folio != xas_reload(&xas)))
2226 if (!folio_batch_add(fbatch, folio)) {
2227 nr = folio_nr_pages(folio);
2229 if (folio_test_hugetlb(folio))
2231 *start = folio->index + nr;
2243 nr = folio_batch_count(fbatch);
2246 folio = fbatch->folios[nr - 1];
2247 if (folio_test_hugetlb(folio))
2248 *start = folio->index + 1;
2250 *start = folio->index + folio_nr_pages(folio);
2254 return folio_batch_count(fbatch);
2256 EXPORT_SYMBOL(filemap_get_folios_contig);
2259 * filemap_get_folios_tag - Get a batch of folios matching @tag
2260 * @mapping: The address_space to search
2261 * @start: The starting page index
2262 * @end: The final page index (inclusive)
2263 * @tag: The tag index
2264 * @fbatch: The batch to fill
2266 * Same as filemap_get_folios(), but only returning folios tagged with @tag.
2268 * Return: The number of folios found.
2269 * Also update @start to index the next folio for traversal.
2271 unsigned filemap_get_folios_tag(struct address_space *mapping, pgoff_t *start,
2272 pgoff_t end, xa_mark_t tag, struct folio_batch *fbatch)
2274 XA_STATE(xas, &mapping->i_pages, *start);
2275 struct folio *folio;
2278 while ((folio = find_get_entry(&xas, end, tag)) != NULL) {
2280 * Shadow entries should never be tagged, but this iteration
2281 * is lockless so there is a window for page reclaim to evict
2282 * a page we saw tagged. Skip over it.
2284 if (xa_is_value(folio))
2286 if (!folio_batch_add(fbatch, folio)) {
2287 unsigned long nr = folio_nr_pages(folio);
2289 if (folio_test_hugetlb(folio))
2291 *start = folio->index + nr;
2296 * We come here when there is no page beyond @end. We take care to not
2297 * overflow the index @start as it confuses some of the callers. This
2298 * breaks the iteration when there is a page at index -1 but that is
2299 * already broke anyway.
2301 if (end == (pgoff_t)-1)
2302 *start = (pgoff_t)-1;
2308 return folio_batch_count(fbatch);
2310 EXPORT_SYMBOL(filemap_get_folios_tag);
2313 * CD/DVDs are error prone. When a medium error occurs, the driver may fail
2314 * a _large_ part of the i/o request. Imagine the worst scenario:
2316 * ---R__________________________________________B__________
2317 * ^ reading here ^ bad block(assume 4k)
2319 * read(R) => miss => readahead(R...B) => media error => frustrating retries
2320 * => failing the whole request => read(R) => read(R+1) =>
2321 * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
2322 * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
2323 * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
2325 * It is going insane. Fix it by quickly scaling down the readahead size.
2327 static void shrink_readahead_size_eio(struct file_ra_state *ra)
2333 * filemap_get_read_batch - Get a batch of folios for read
2335 * Get a batch of folios which represent a contiguous range of bytes in
2336 * the file. No exceptional entries will be returned. If @index is in
2337 * the middle of a folio, the entire folio will be returned. The last
2338 * folio in the batch may have the readahead flag set or the uptodate flag
2339 * clear so that the caller can take the appropriate action.
2341 static void filemap_get_read_batch(struct address_space *mapping,
2342 pgoff_t index, pgoff_t max, struct folio_batch *fbatch)
2344 XA_STATE(xas, &mapping->i_pages, index);
2345 struct folio *folio;
2348 for (folio = xas_load(&xas); folio; folio = xas_next(&xas)) {
2349 if (xas_retry(&xas, folio))
2351 if (xas.xa_index > max || xa_is_value(folio))
2353 if (xa_is_sibling(folio))
2355 if (!folio_try_get_rcu(folio))
2358 if (unlikely(folio != xas_reload(&xas)))
2361 if (!folio_batch_add(fbatch, folio))
2363 if (!folio_test_uptodate(folio))
2365 if (folio_test_readahead(folio))
2367 xas_advance(&xas, folio->index + folio_nr_pages(folio) - 1);
2377 static int filemap_read_folio(struct file *file, filler_t filler,
2378 struct folio *folio)
2380 bool workingset = folio_test_workingset(folio);
2381 unsigned long pflags;
2385 * A previous I/O error may have been due to temporary failures,
2386 * eg. multipath errors. PG_error will be set again if read_folio
2389 folio_clear_error(folio);
2391 /* Start the actual read. The read will unlock the page. */
2392 if (unlikely(workingset))
2393 psi_memstall_enter(&pflags);
2394 error = filler(file, folio);
2395 if (unlikely(workingset))
2396 psi_memstall_leave(&pflags);
2400 error = folio_wait_locked_killable(folio);
2403 if (folio_test_uptodate(folio))
2406 shrink_readahead_size_eio(&file->f_ra);
2410 static bool filemap_range_uptodate(struct address_space *mapping,
2411 loff_t pos, size_t count, struct folio *folio,
2414 if (folio_test_uptodate(folio))
2416 /* pipes can't handle partially uptodate pages */
2419 if (!mapping->a_ops->is_partially_uptodate)
2421 if (mapping->host->i_blkbits >= folio_shift(folio))
2424 if (folio_pos(folio) > pos) {
2425 count -= folio_pos(folio) - pos;
2428 pos -= folio_pos(folio);
2431 return mapping->a_ops->is_partially_uptodate(folio, pos, count);
2434 static int filemap_update_page(struct kiocb *iocb,
2435 struct address_space *mapping, size_t count,
2436 struct folio *folio, bool need_uptodate)
2440 if (iocb->ki_flags & IOCB_NOWAIT) {
2441 if (!filemap_invalidate_trylock_shared(mapping))
2444 filemap_invalidate_lock_shared(mapping);
2447 if (!folio_trylock(folio)) {
2449 if (iocb->ki_flags & (IOCB_NOWAIT | IOCB_NOIO))
2450 goto unlock_mapping;
2451 if (!(iocb->ki_flags & IOCB_WAITQ)) {
2452 filemap_invalidate_unlock_shared(mapping);
2454 * This is where we usually end up waiting for a
2455 * previously submitted readahead to finish.
2457 folio_put_wait_locked(folio, TASK_KILLABLE);
2458 return AOP_TRUNCATED_PAGE;
2460 error = __folio_lock_async(folio, iocb->ki_waitq);
2462 goto unlock_mapping;
2465 error = AOP_TRUNCATED_PAGE;
2466 if (!folio->mapping)
2470 if (filemap_range_uptodate(mapping, iocb->ki_pos, count, folio,
2475 if (iocb->ki_flags & (IOCB_NOIO | IOCB_NOWAIT | IOCB_WAITQ))
2478 error = filemap_read_folio(iocb->ki_filp, mapping->a_ops->read_folio,
2480 goto unlock_mapping;
2482 folio_unlock(folio);
2484 filemap_invalidate_unlock_shared(mapping);
2485 if (error == AOP_TRUNCATED_PAGE)
2490 static int filemap_create_folio(struct file *file,
2491 struct address_space *mapping, pgoff_t index,
2492 struct folio_batch *fbatch)
2494 struct folio *folio;
2497 folio = filemap_alloc_folio(mapping_gfp_mask(mapping), 0);
2502 * Protect against truncate / hole punch. Grabbing invalidate_lock
2503 * here assures we cannot instantiate and bring uptodate new
2504 * pagecache folios after evicting page cache during truncate
2505 * and before actually freeing blocks. Note that we could
2506 * release invalidate_lock after inserting the folio into
2507 * the page cache as the locked folio would then be enough to
2508 * synchronize with hole punching. But there are code paths
2509 * such as filemap_update_page() filling in partially uptodate
2510 * pages or ->readahead() that need to hold invalidate_lock
2511 * while mapping blocks for IO so let's hold the lock here as
2512 * well to keep locking rules simple.
2514 filemap_invalidate_lock_shared(mapping);
2515 error = filemap_add_folio(mapping, folio, index,
2516 mapping_gfp_constraint(mapping, GFP_KERNEL));
2517 if (error == -EEXIST)
2518 error = AOP_TRUNCATED_PAGE;
2522 error = filemap_read_folio(file, mapping->a_ops->read_folio, folio);
2526 filemap_invalidate_unlock_shared(mapping);
2527 folio_batch_add(fbatch, folio);
2530 filemap_invalidate_unlock_shared(mapping);
2535 static int filemap_readahead(struct kiocb *iocb, struct file *file,
2536 struct address_space *mapping, struct folio *folio,
2539 DEFINE_READAHEAD(ractl, file, &file->f_ra, mapping, folio->index);
2541 if (iocb->ki_flags & IOCB_NOIO)
2543 page_cache_async_ra(&ractl, folio, last_index - folio->index);
2547 static int filemap_get_pages(struct kiocb *iocb, size_t count,
2548 struct folio_batch *fbatch, bool need_uptodate)
2550 struct file *filp = iocb->ki_filp;
2551 struct address_space *mapping = filp->f_mapping;
2552 struct file_ra_state *ra = &filp->f_ra;
2553 pgoff_t index = iocb->ki_pos >> PAGE_SHIFT;
2555 struct folio *folio;
2558 /* "last_index" is the index of the page beyond the end of the read */
2559 last_index = DIV_ROUND_UP(iocb->ki_pos + count, PAGE_SIZE);
2561 if (fatal_signal_pending(current))
2564 filemap_get_read_batch(mapping, index, last_index - 1, fbatch);
2565 if (!folio_batch_count(fbatch)) {
2566 if (iocb->ki_flags & IOCB_NOIO)
2568 page_cache_sync_readahead(mapping, ra, filp, index,
2569 last_index - index);
2570 filemap_get_read_batch(mapping, index, last_index - 1, fbatch);
2572 if (!folio_batch_count(fbatch)) {
2573 if (iocb->ki_flags & (IOCB_NOWAIT | IOCB_WAITQ))
2575 err = filemap_create_folio(filp, mapping,
2576 iocb->ki_pos >> PAGE_SHIFT, fbatch);
2577 if (err == AOP_TRUNCATED_PAGE)
2582 folio = fbatch->folios[folio_batch_count(fbatch) - 1];
2583 if (folio_test_readahead(folio)) {
2584 err = filemap_readahead(iocb, filp, mapping, folio, last_index);
2588 if (!folio_test_uptodate(folio)) {
2589 if ((iocb->ki_flags & IOCB_WAITQ) &&
2590 folio_batch_count(fbatch) > 1)
2591 iocb->ki_flags |= IOCB_NOWAIT;
2592 err = filemap_update_page(iocb, mapping, count, folio,
2602 if (likely(--fbatch->nr))
2604 if (err == AOP_TRUNCATED_PAGE)
2609 static inline bool pos_same_folio(loff_t pos1, loff_t pos2, struct folio *folio)
2611 unsigned int shift = folio_shift(folio);
2613 return (pos1 >> shift == pos2 >> shift);
2617 * filemap_read - Read data from the page cache.
2618 * @iocb: The iocb to read.
2619 * @iter: Destination for the data.
2620 * @already_read: Number of bytes already read by the caller.
2622 * Copies data from the page cache. If the data is not currently present,
2623 * uses the readahead and read_folio address_space operations to fetch it.
2625 * Return: Total number of bytes copied, including those already read by
2626 * the caller. If an error happens before any bytes are copied, returns
2627 * a negative error number.
2629 ssize_t filemap_read(struct kiocb *iocb, struct iov_iter *iter,
2630 ssize_t already_read)
2632 struct file *filp = iocb->ki_filp;
2633 struct file_ra_state *ra = &filp->f_ra;
2634 struct address_space *mapping = filp->f_mapping;
2635 struct inode *inode = mapping->host;
2636 struct folio_batch fbatch;
2638 bool writably_mapped;
2639 loff_t isize, end_offset;
2641 if (unlikely(iocb->ki_pos >= inode->i_sb->s_maxbytes))
2643 if (unlikely(!iov_iter_count(iter)))
2646 iov_iter_truncate(iter, inode->i_sb->s_maxbytes);
2647 folio_batch_init(&fbatch);
2653 * If we've already successfully copied some data, then we
2654 * can no longer safely return -EIOCBQUEUED. Hence mark
2655 * an async read NOWAIT at that point.
2657 if ((iocb->ki_flags & IOCB_WAITQ) && already_read)
2658 iocb->ki_flags |= IOCB_NOWAIT;
2660 if (unlikely(iocb->ki_pos >= i_size_read(inode)))
2663 error = filemap_get_pages(iocb, iter->count, &fbatch,
2664 iov_iter_is_pipe(iter));
2669 * i_size must be checked after we know the pages are Uptodate.
2671 * Checking i_size after the check allows us to calculate
2672 * the correct value for "nr", which means the zero-filled
2673 * part of the page is not copied back to userspace (unless
2674 * another truncate extends the file - this is desired though).
2676 isize = i_size_read(inode);
2677 if (unlikely(iocb->ki_pos >= isize))
2679 end_offset = min_t(loff_t, isize, iocb->ki_pos + iter->count);
2682 * Once we start copying data, we don't want to be touching any
2683 * cachelines that might be contended:
2685 writably_mapped = mapping_writably_mapped(mapping);
2688 * When a read accesses the same folio several times, only
2689 * mark it as accessed the first time.
2691 if (!pos_same_folio(iocb->ki_pos, ra->prev_pos - 1,
2693 folio_mark_accessed(fbatch.folios[0]);
2695 for (i = 0; i < folio_batch_count(&fbatch); i++) {
2696 struct folio *folio = fbatch.folios[i];
2697 size_t fsize = folio_size(folio);
2698 size_t offset = iocb->ki_pos & (fsize - 1);
2699 size_t bytes = min_t(loff_t, end_offset - iocb->ki_pos,
2703 if (end_offset < folio_pos(folio))
2706 folio_mark_accessed(folio);
2708 * If users can be writing to this folio using arbitrary
2709 * virtual addresses, take care of potential aliasing
2710 * before reading the folio on the kernel side.
2712 if (writably_mapped)
2713 flush_dcache_folio(folio);
2715 copied = copy_folio_to_iter(folio, offset, bytes, iter);
2717 already_read += copied;
2718 iocb->ki_pos += copied;
2719 ra->prev_pos = iocb->ki_pos;
2721 if (copied < bytes) {
2727 for (i = 0; i < folio_batch_count(&fbatch); i++)
2728 folio_put(fbatch.folios[i]);
2729 folio_batch_init(&fbatch);
2730 } while (iov_iter_count(iter) && iocb->ki_pos < isize && !error);
2732 file_accessed(filp);
2734 return already_read ? already_read : error;
2736 EXPORT_SYMBOL_GPL(filemap_read);
2738 int kiocb_write_and_wait(struct kiocb *iocb, size_t count)
2740 struct address_space *mapping = iocb->ki_filp->f_mapping;
2741 loff_t pos = iocb->ki_pos;
2742 loff_t end = pos + count - 1;
2744 if (iocb->ki_flags & IOCB_NOWAIT) {
2745 if (filemap_range_needs_writeback(mapping, pos, end))
2750 return filemap_write_and_wait_range(mapping, pos, end);
2753 int kiocb_invalidate_pages(struct kiocb *iocb, size_t count)
2755 struct address_space *mapping = iocb->ki_filp->f_mapping;
2756 loff_t pos = iocb->ki_pos;
2757 loff_t end = pos + count - 1;
2760 if (iocb->ki_flags & IOCB_NOWAIT) {
2761 /* we could block if there are any pages in the range */
2762 if (filemap_range_has_page(mapping, pos, end))
2765 ret = filemap_write_and_wait_range(mapping, pos, end);
2771 * After a write we want buffered reads to be sure to go to disk to get
2772 * the new data. We invalidate clean cached page from the region we're
2773 * about to write. We do this *before* the write so that we can return
2774 * without clobbering -EIOCBQUEUED from ->direct_IO().
2776 return invalidate_inode_pages2_range(mapping, pos >> PAGE_SHIFT,
2781 * generic_file_read_iter - generic filesystem read routine
2782 * @iocb: kernel I/O control block
2783 * @iter: destination for the data read
2785 * This is the "read_iter()" routine for all filesystems
2786 * that can use the page cache directly.
2788 * The IOCB_NOWAIT flag in iocb->ki_flags indicates that -EAGAIN shall
2789 * be returned when no data can be read without waiting for I/O requests
2790 * to complete; it doesn't prevent readahead.
2792 * The IOCB_NOIO flag in iocb->ki_flags indicates that no new I/O
2793 * requests shall be made for the read or for readahead. When no data
2794 * can be read, -EAGAIN shall be returned. When readahead would be
2795 * triggered, a partial, possibly empty read shall be returned.
2798 * * number of bytes copied, even for partial reads
2799 * * negative error code (or 0 if IOCB_NOIO) if nothing was read
2802 generic_file_read_iter(struct kiocb *iocb, struct iov_iter *iter)
2804 size_t count = iov_iter_count(iter);
2808 return 0; /* skip atime */
2810 if (iocb->ki_flags & IOCB_DIRECT) {
2811 struct file *file = iocb->ki_filp;
2812 struct address_space *mapping = file->f_mapping;
2813 struct inode *inode = mapping->host;
2815 retval = kiocb_write_and_wait(iocb, count);
2818 file_accessed(file);
2820 retval = mapping->a_ops->direct_IO(iocb, iter);
2822 iocb->ki_pos += retval;
2825 if (retval != -EIOCBQUEUED)
2826 iov_iter_revert(iter, count - iov_iter_count(iter));
2829 * Btrfs can have a short DIO read if we encounter
2830 * compressed extents, so if there was an error, or if
2831 * we've already read everything we wanted to, or if
2832 * there was a short read because we hit EOF, go ahead
2833 * and return. Otherwise fallthrough to buffered io for
2834 * the rest of the read. Buffered reads will not work for
2835 * DAX files, so don't bother trying.
2837 if (retval < 0 || !count || IS_DAX(inode))
2839 if (iocb->ki_pos >= i_size_read(inode))
2843 return filemap_read(iocb, iter, retval);
2845 EXPORT_SYMBOL(generic_file_read_iter);
2848 * Splice subpages from a folio into a pipe.
2850 size_t splice_folio_into_pipe(struct pipe_inode_info *pipe,
2851 struct folio *folio, loff_t fpos, size_t size)
2854 size_t spliced = 0, offset = offset_in_folio(folio, fpos);
2856 page = folio_page(folio, offset / PAGE_SIZE);
2857 size = min(size, folio_size(folio) - offset);
2858 offset %= PAGE_SIZE;
2860 while (spliced < size &&
2861 !pipe_full(pipe->head, pipe->tail, pipe->max_usage)) {
2862 struct pipe_buffer *buf = pipe_head_buf(pipe);
2863 size_t part = min_t(size_t, PAGE_SIZE - offset, size - spliced);
2865 *buf = (struct pipe_buffer) {
2866 .ops = &page_cache_pipe_buf_ops,
2882 * Splice folios from the pagecache of a buffered (ie. non-O_DIRECT) file into
2885 ssize_t filemap_splice_read(struct file *in, loff_t *ppos,
2886 struct pipe_inode_info *pipe,
2887 size_t len, unsigned int flags)
2889 struct folio_batch fbatch;
2891 size_t total_spliced = 0, used, npages;
2892 loff_t isize, end_offset;
2893 bool writably_mapped;
2896 init_sync_kiocb(&iocb, in);
2897 iocb.ki_pos = *ppos;
2899 /* Work out how much data we can actually add into the pipe */
2900 used = pipe_occupancy(pipe->head, pipe->tail);
2901 npages = max_t(ssize_t, pipe->max_usage - used, 0);
2902 len = min_t(size_t, len, npages * PAGE_SIZE);
2904 folio_batch_init(&fbatch);
2909 if (*ppos >= i_size_read(file_inode(in)))
2912 iocb.ki_pos = *ppos;
2913 error = filemap_get_pages(&iocb, len, &fbatch, true);
2918 * i_size must be checked after we know the pages are Uptodate.
2920 * Checking i_size after the check allows us to calculate
2921 * the correct value for "nr", which means the zero-filled
2922 * part of the page is not copied back to userspace (unless
2923 * another truncate extends the file - this is desired though).
2925 isize = i_size_read(file_inode(in));
2926 if (unlikely(*ppos >= isize))
2928 end_offset = min_t(loff_t, isize, *ppos + len);
2931 * Once we start copying data, we don't want to be touching any
2932 * cachelines that might be contended:
2934 writably_mapped = mapping_writably_mapped(in->f_mapping);
2936 for (i = 0; i < folio_batch_count(&fbatch); i++) {
2937 struct folio *folio = fbatch.folios[i];
2940 if (folio_pos(folio) >= end_offset)
2942 folio_mark_accessed(folio);
2945 * If users can be writing to this folio using arbitrary
2946 * virtual addresses, take care of potential aliasing
2947 * before reading the folio on the kernel side.
2949 if (writably_mapped)
2950 flush_dcache_folio(folio);
2952 n = min_t(loff_t, len, isize - *ppos);
2953 n = splice_folio_into_pipe(pipe, folio, *ppos, n);
2959 in->f_ra.prev_pos = *ppos;
2960 if (pipe_full(pipe->head, pipe->tail, pipe->max_usage))
2964 folio_batch_release(&fbatch);
2968 folio_batch_release(&fbatch);
2971 return total_spliced ? total_spliced : error;
2973 EXPORT_SYMBOL(filemap_splice_read);
2975 static inline loff_t folio_seek_hole_data(struct xa_state *xas,
2976 struct address_space *mapping, struct folio *folio,
2977 loff_t start, loff_t end, bool seek_data)
2979 const struct address_space_operations *ops = mapping->a_ops;
2980 size_t offset, bsz = i_blocksize(mapping->host);
2982 if (xa_is_value(folio) || folio_test_uptodate(folio))
2983 return seek_data ? start : end;
2984 if (!ops->is_partially_uptodate)
2985 return seek_data ? end : start;
2990 if (unlikely(folio->mapping != mapping))
2993 offset = offset_in_folio(folio, start) & ~(bsz - 1);
2996 if (ops->is_partially_uptodate(folio, offset, bsz) ==
2999 start = (start + bsz) & ~(bsz - 1);
3001 } while (offset < folio_size(folio));
3003 folio_unlock(folio);
3008 static inline size_t seek_folio_size(struct xa_state *xas, struct folio *folio)
3010 if (xa_is_value(folio))
3011 return PAGE_SIZE << xa_get_order(xas->xa, xas->xa_index);
3012 return folio_size(folio);
3016 * mapping_seek_hole_data - Seek for SEEK_DATA / SEEK_HOLE in the page cache.
3017 * @mapping: Address space to search.
3018 * @start: First byte to consider.
3019 * @end: Limit of search (exclusive).
3020 * @whence: Either SEEK_HOLE or SEEK_DATA.
3022 * If the page cache knows which blocks contain holes and which blocks
3023 * contain data, your filesystem can use this function to implement
3024 * SEEK_HOLE and SEEK_DATA. This is useful for filesystems which are
3025 * entirely memory-based such as tmpfs, and filesystems which support
3026 * unwritten extents.
3028 * Return: The requested offset on success, or -ENXIO if @whence specifies
3029 * SEEK_DATA and there is no data after @start. There is an implicit hole
3030 * after @end - 1, so SEEK_HOLE returns @end if all the bytes between @start
3031 * and @end contain data.
3033 loff_t mapping_seek_hole_data(struct address_space *mapping, loff_t start,
3034 loff_t end, int whence)
3036 XA_STATE(xas, &mapping->i_pages, start >> PAGE_SHIFT);
3037 pgoff_t max = (end - 1) >> PAGE_SHIFT;
3038 bool seek_data = (whence == SEEK_DATA);
3039 struct folio *folio;
3045 while ((folio = find_get_entry(&xas, max, XA_PRESENT))) {
3046 loff_t pos = (u64)xas.xa_index << PAGE_SHIFT;
3055 seek_size = seek_folio_size(&xas, folio);
3056 pos = round_up((u64)pos + 1, seek_size);
3057 start = folio_seek_hole_data(&xas, mapping, folio, start, pos,
3063 if (seek_size > PAGE_SIZE)
3064 xas_set(&xas, pos >> PAGE_SHIFT);
3065 if (!xa_is_value(folio))
3072 if (folio && !xa_is_value(folio))
3080 #define MMAP_LOTSAMISS (100)
3082 * lock_folio_maybe_drop_mmap - lock the page, possibly dropping the mmap_lock
3083 * @vmf - the vm_fault for this fault.
3084 * @folio - the folio to lock.
3085 * @fpin - the pointer to the file we may pin (or is already pinned).
3087 * This works similar to lock_folio_or_retry in that it can drop the
3088 * mmap_lock. It differs in that it actually returns the folio locked
3089 * if it returns 1 and 0 if it couldn't lock the folio. If we did have
3090 * to drop the mmap_lock then fpin will point to the pinned file and
3091 * needs to be fput()'ed at a later point.
3093 static int lock_folio_maybe_drop_mmap(struct vm_fault *vmf, struct folio *folio,
3096 if (folio_trylock(folio))
3100 * NOTE! This will make us return with VM_FAULT_RETRY, but with
3101 * the mmap_lock still held. That's how FAULT_FLAG_RETRY_NOWAIT
3102 * is supposed to work. We have way too many special cases..
3104 if (vmf->flags & FAULT_FLAG_RETRY_NOWAIT)
3107 *fpin = maybe_unlock_mmap_for_io(vmf, *fpin);
3108 if (vmf->flags & FAULT_FLAG_KILLABLE) {
3109 if (__folio_lock_killable(folio)) {
3111 * We didn't have the right flags to drop the mmap_lock,
3112 * but all fault_handlers only check for fatal signals
3113 * if we return VM_FAULT_RETRY, so we need to drop the
3114 * mmap_lock here and return 0 if we don't have a fpin.
3117 mmap_read_unlock(vmf->vma->vm_mm);
3121 __folio_lock(folio);
3127 * Synchronous readahead happens when we don't even find a page in the page
3128 * cache at all. We don't want to perform IO under the mmap sem, so if we have
3129 * to drop the mmap sem we return the file that was pinned in order for us to do
3130 * that. If we didn't pin a file then we return NULL. The file that is
3131 * returned needs to be fput()'ed when we're done with it.
3133 static struct file *do_sync_mmap_readahead(struct vm_fault *vmf)
3135 struct file *file = vmf->vma->vm_file;
3136 struct file_ra_state *ra = &file->f_ra;
3137 struct address_space *mapping = file->f_mapping;
3138 DEFINE_READAHEAD(ractl, file, ra, mapping, vmf->pgoff);
3139 struct file *fpin = NULL;
3140 unsigned long vm_flags = vmf->vma->vm_flags;
3141 unsigned int mmap_miss;
3143 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
3144 /* Use the readahead code, even if readahead is disabled */
3145 if (vm_flags & VM_HUGEPAGE) {
3146 fpin = maybe_unlock_mmap_for_io(vmf, fpin);
3147 ractl._index &= ~((unsigned long)HPAGE_PMD_NR - 1);
3148 ra->size = HPAGE_PMD_NR;
3150 * Fetch two PMD folios, so we get the chance to actually
3151 * readahead, unless we've been told not to.
3153 if (!(vm_flags & VM_RAND_READ))
3155 ra->async_size = HPAGE_PMD_NR;
3156 page_cache_ra_order(&ractl, ra, HPAGE_PMD_ORDER);
3161 /* If we don't want any read-ahead, don't bother */
3162 if (vm_flags & VM_RAND_READ)
3167 if (vm_flags & VM_SEQ_READ) {
3168 fpin = maybe_unlock_mmap_for_io(vmf, fpin);
3169 page_cache_sync_ra(&ractl, ra->ra_pages);
3173 /* Avoid banging the cache line if not needed */
3174 mmap_miss = READ_ONCE(ra->mmap_miss);
3175 if (mmap_miss < MMAP_LOTSAMISS * 10)
3176 WRITE_ONCE(ra->mmap_miss, ++mmap_miss);
3179 * Do we miss much more than hit in this file? If so,
3180 * stop bothering with read-ahead. It will only hurt.
3182 if (mmap_miss > MMAP_LOTSAMISS)
3188 fpin = maybe_unlock_mmap_for_io(vmf, fpin);
3189 ra->start = max_t(long, 0, vmf->pgoff - ra->ra_pages / 2);
3190 ra->size = ra->ra_pages;
3191 ra->async_size = ra->ra_pages / 4;
3192 ractl._index = ra->start;
3193 page_cache_ra_order(&ractl, ra, 0);
3198 * Asynchronous readahead happens when we find the page and PG_readahead,
3199 * so we want to possibly extend the readahead further. We return the file that
3200 * was pinned if we have to drop the mmap_lock in order to do IO.
3202 static struct file *do_async_mmap_readahead(struct vm_fault *vmf,
3203 struct folio *folio)
3205 struct file *file = vmf->vma->vm_file;
3206 struct file_ra_state *ra = &file->f_ra;
3207 DEFINE_READAHEAD(ractl, file, ra, file->f_mapping, vmf->pgoff);
3208 struct file *fpin = NULL;
3209 unsigned int mmap_miss;
3211 /* If we don't want any read-ahead, don't bother */
3212 if (vmf->vma->vm_flags & VM_RAND_READ || !ra->ra_pages)
3215 mmap_miss = READ_ONCE(ra->mmap_miss);
3217 WRITE_ONCE(ra->mmap_miss, --mmap_miss);
3219 if (folio_test_readahead(folio)) {
3220 fpin = maybe_unlock_mmap_for_io(vmf, fpin);
3221 page_cache_async_ra(&ractl, folio, ra->ra_pages);
3227 * filemap_fault - read in file data for page fault handling
3228 * @vmf: struct vm_fault containing details of the fault
3230 * filemap_fault() is invoked via the vma operations vector for a
3231 * mapped memory region to read in file data during a page fault.
3233 * The goto's are kind of ugly, but this streamlines the normal case of having
3234 * it in the page cache, and handles the special cases reasonably without
3235 * having a lot of duplicated code.
3237 * vma->vm_mm->mmap_lock must be held on entry.
3239 * If our return value has VM_FAULT_RETRY set, it's because the mmap_lock
3240 * may be dropped before doing I/O or by lock_folio_maybe_drop_mmap().
3242 * If our return value does not have VM_FAULT_RETRY set, the mmap_lock
3243 * has not been released.
3245 * We never return with VM_FAULT_RETRY and a bit from VM_FAULT_ERROR set.
3247 * Return: bitwise-OR of %VM_FAULT_ codes.
3249 vm_fault_t filemap_fault(struct vm_fault *vmf)
3252 struct file *file = vmf->vma->vm_file;
3253 struct file *fpin = NULL;
3254 struct address_space *mapping = file->f_mapping;
3255 struct inode *inode = mapping->host;
3256 pgoff_t max_idx, index = vmf->pgoff;
3257 struct folio *folio;
3259 bool mapping_locked = false;
3261 max_idx = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
3262 if (unlikely(index >= max_idx))
3263 return VM_FAULT_SIGBUS;
3266 * Do we have something in the page cache already?
3268 folio = filemap_get_folio(mapping, index);
3269 if (likely(!IS_ERR(folio))) {
3271 * We found the page, so try async readahead before waiting for
3274 if (!(vmf->flags & FAULT_FLAG_TRIED))
3275 fpin = do_async_mmap_readahead(vmf, folio);
3276 if (unlikely(!folio_test_uptodate(folio))) {
3277 filemap_invalidate_lock_shared(mapping);
3278 mapping_locked = true;
3281 /* No page in the page cache at all */
3282 count_vm_event(PGMAJFAULT);
3283 count_memcg_event_mm(vmf->vma->vm_mm, PGMAJFAULT);
3284 ret = VM_FAULT_MAJOR;
3285 fpin = do_sync_mmap_readahead(vmf);
3288 * See comment in filemap_create_folio() why we need
3291 if (!mapping_locked) {
3292 filemap_invalidate_lock_shared(mapping);
3293 mapping_locked = true;
3295 folio = __filemap_get_folio(mapping, index,
3296 FGP_CREAT|FGP_FOR_MMAP,
3298 if (IS_ERR(folio)) {
3301 filemap_invalidate_unlock_shared(mapping);
3302 return VM_FAULT_OOM;
3306 if (!lock_folio_maybe_drop_mmap(vmf, folio, &fpin))
3309 /* Did it get truncated? */
3310 if (unlikely(folio->mapping != mapping)) {
3311 folio_unlock(folio);
3315 VM_BUG_ON_FOLIO(!folio_contains(folio, index), folio);
3318 * We have a locked page in the page cache, now we need to check
3319 * that it's up-to-date. If not, it is going to be due to an error.
3321 if (unlikely(!folio_test_uptodate(folio))) {
3323 * The page was in cache and uptodate and now it is not.
3324 * Strange but possible since we didn't hold the page lock all
3325 * the time. Let's drop everything get the invalidate lock and
3328 if (!mapping_locked) {
3329 folio_unlock(folio);
3333 goto page_not_uptodate;
3337 * We've made it this far and we had to drop our mmap_lock, now is the
3338 * time to return to the upper layer and have it re-find the vma and
3342 folio_unlock(folio);
3346 filemap_invalidate_unlock_shared(mapping);
3349 * Found the page and have a reference on it.
3350 * We must recheck i_size under page lock.
3352 max_idx = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
3353 if (unlikely(index >= max_idx)) {
3354 folio_unlock(folio);
3356 return VM_FAULT_SIGBUS;
3359 vmf->page = folio_file_page(folio, index);
3360 return ret | VM_FAULT_LOCKED;
3364 * Umm, take care of errors if the page isn't up-to-date.
3365 * Try to re-read it _once_. We do this synchronously,
3366 * because there really aren't any performance issues here
3367 * and we need to check for errors.
3369 fpin = maybe_unlock_mmap_for_io(vmf, fpin);
3370 error = filemap_read_folio(file, mapping->a_ops->read_folio, folio);
3375 if (!error || error == AOP_TRUNCATED_PAGE)
3377 filemap_invalidate_unlock_shared(mapping);
3379 return VM_FAULT_SIGBUS;
3383 * We dropped the mmap_lock, we need to return to the fault handler to
3384 * re-find the vma and come back and find our hopefully still populated
3390 filemap_invalidate_unlock_shared(mapping);
3393 return ret | VM_FAULT_RETRY;
3395 EXPORT_SYMBOL(filemap_fault);
3397 static bool filemap_map_pmd(struct vm_fault *vmf, struct folio *folio,
3400 struct mm_struct *mm = vmf->vma->vm_mm;
3402 /* Huge page is mapped? No need to proceed. */
3403 if (pmd_trans_huge(*vmf->pmd)) {
3404 folio_unlock(folio);
3409 if (pmd_none(*vmf->pmd) && folio_test_pmd_mappable(folio)) {
3410 struct page *page = folio_file_page(folio, start);
3411 vm_fault_t ret = do_set_pmd(vmf, page);
3413 /* The page is mapped successfully, reference consumed. */
3414 folio_unlock(folio);
3419 if (pmd_none(*vmf->pmd))
3420 pmd_install(mm, vmf->pmd, &vmf->prealloc_pte);
3422 /* See comment in handle_pte_fault() */
3423 if (pmd_devmap_trans_unstable(vmf->pmd)) {
3424 folio_unlock(folio);
3432 static struct folio *next_uptodate_page(struct folio *folio,
3433 struct address_space *mapping,
3434 struct xa_state *xas, pgoff_t end_pgoff)
3436 unsigned long max_idx;
3441 if (xas_retry(xas, folio))
3443 if (xa_is_value(folio))
3445 if (folio_test_locked(folio))
3447 if (!folio_try_get_rcu(folio))
3449 /* Has the page moved or been split? */
3450 if (unlikely(folio != xas_reload(xas)))
3452 if (!folio_test_uptodate(folio) || folio_test_readahead(folio))
3454 if (!folio_trylock(folio))
3456 if (folio->mapping != mapping)
3458 if (!folio_test_uptodate(folio))
3460 max_idx = DIV_ROUND_UP(i_size_read(mapping->host), PAGE_SIZE);
3461 if (xas->xa_index >= max_idx)
3465 folio_unlock(folio);
3468 } while ((folio = xas_next_entry(xas, end_pgoff)) != NULL);
3473 static inline struct folio *first_map_page(struct address_space *mapping,
3474 struct xa_state *xas,
3477 return next_uptodate_page(xas_find(xas, end_pgoff),
3478 mapping, xas, end_pgoff);
3481 static inline struct folio *next_map_page(struct address_space *mapping,
3482 struct xa_state *xas,
3485 return next_uptodate_page(xas_next_entry(xas, end_pgoff),
3486 mapping, xas, end_pgoff);
3489 vm_fault_t filemap_map_pages(struct vm_fault *vmf,
3490 pgoff_t start_pgoff, pgoff_t end_pgoff)
3492 struct vm_area_struct *vma = vmf->vma;
3493 struct file *file = vma->vm_file;
3494 struct address_space *mapping = file->f_mapping;
3495 pgoff_t last_pgoff = start_pgoff;
3497 XA_STATE(xas, &mapping->i_pages, start_pgoff);
3498 struct folio *folio;
3500 unsigned int mmap_miss = READ_ONCE(file->f_ra.mmap_miss);
3504 folio = first_map_page(mapping, &xas, end_pgoff);
3508 if (filemap_map_pmd(vmf, folio, start_pgoff)) {
3509 ret = VM_FAULT_NOPAGE;
3513 addr = vma->vm_start + ((start_pgoff - vma->vm_pgoff) << PAGE_SHIFT);
3514 vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, addr, &vmf->ptl);
3517 page = folio_file_page(folio, xas.xa_index);
3518 if (PageHWPoison(page))
3524 addr += (xas.xa_index - last_pgoff) << PAGE_SHIFT;
3525 vmf->pte += xas.xa_index - last_pgoff;
3526 last_pgoff = xas.xa_index;
3529 * NOTE: If there're PTE markers, we'll leave them to be
3530 * handled in the specific fault path, and it'll prohibit the
3531 * fault-around logic.
3533 if (!pte_none(*vmf->pte))
3536 /* We're about to handle the fault */
3537 if (vmf->address == addr)
3538 ret = VM_FAULT_NOPAGE;
3540 do_set_pte(vmf, page, addr);
3541 /* no need to invalidate: a not-present page won't be cached */
3542 update_mmu_cache(vma, addr, vmf->pte);
3543 if (folio_more_pages(folio, xas.xa_index, end_pgoff)) {
3545 folio_ref_inc(folio);
3548 folio_unlock(folio);
3551 if (folio_more_pages(folio, xas.xa_index, end_pgoff)) {
3555 folio_unlock(folio);
3557 } while ((folio = next_map_page(mapping, &xas, end_pgoff)) != NULL);
3558 pte_unmap_unlock(vmf->pte, vmf->ptl);
3561 WRITE_ONCE(file->f_ra.mmap_miss, mmap_miss);
3564 EXPORT_SYMBOL(filemap_map_pages);
3566 vm_fault_t filemap_page_mkwrite(struct vm_fault *vmf)
3568 struct address_space *mapping = vmf->vma->vm_file->f_mapping;
3569 struct folio *folio = page_folio(vmf->page);
3570 vm_fault_t ret = VM_FAULT_LOCKED;
3572 sb_start_pagefault(mapping->host->i_sb);
3573 file_update_time(vmf->vma->vm_file);
3575 if (folio->mapping != mapping) {
3576 folio_unlock(folio);
3577 ret = VM_FAULT_NOPAGE;
3581 * We mark the folio dirty already here so that when freeze is in
3582 * progress, we are guaranteed that writeback during freezing will
3583 * see the dirty folio and writeprotect it again.
3585 folio_mark_dirty(folio);
3586 folio_wait_stable(folio);
3588 sb_end_pagefault(mapping->host->i_sb);
3592 const struct vm_operations_struct generic_file_vm_ops = {
3593 .fault = filemap_fault,
3594 .map_pages = filemap_map_pages,
3595 .page_mkwrite = filemap_page_mkwrite,
3598 /* This is used for a general mmap of a disk file */
3600 int generic_file_mmap(struct file *file, struct vm_area_struct *vma)
3602 struct address_space *mapping = file->f_mapping;
3604 if (!mapping->a_ops->read_folio)
3606 file_accessed(file);
3607 vma->vm_ops = &generic_file_vm_ops;
3612 * This is for filesystems which do not implement ->writepage.
3614 int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
3616 if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE))
3618 return generic_file_mmap(file, vma);
3621 vm_fault_t filemap_page_mkwrite(struct vm_fault *vmf)
3623 return VM_FAULT_SIGBUS;
3625 int generic_file_mmap(struct file *file, struct vm_area_struct *vma)
3629 int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
3633 #endif /* CONFIG_MMU */
3635 EXPORT_SYMBOL(filemap_page_mkwrite);
3636 EXPORT_SYMBOL(generic_file_mmap);
3637 EXPORT_SYMBOL(generic_file_readonly_mmap);
3639 static struct folio *do_read_cache_folio(struct address_space *mapping,
3640 pgoff_t index, filler_t filler, struct file *file, gfp_t gfp)
3642 struct folio *folio;
3646 filler = mapping->a_ops->read_folio;
3648 folio = filemap_get_folio(mapping, index);
3649 if (IS_ERR(folio)) {
3650 folio = filemap_alloc_folio(gfp, 0);
3652 return ERR_PTR(-ENOMEM);
3653 err = filemap_add_folio(mapping, folio, index, gfp);
3654 if (unlikely(err)) {
3658 /* Presumably ENOMEM for xarray node */
3659 return ERR_PTR(err);
3664 if (folio_test_uptodate(folio))
3667 if (!folio_trylock(folio)) {
3668 folio_put_wait_locked(folio, TASK_UNINTERRUPTIBLE);
3672 /* Folio was truncated from mapping */
3673 if (!folio->mapping) {
3674 folio_unlock(folio);
3679 /* Someone else locked and filled the page in a very small window */
3680 if (folio_test_uptodate(folio)) {
3681 folio_unlock(folio);
3686 err = filemap_read_folio(file, filler, folio);
3689 if (err == AOP_TRUNCATED_PAGE)
3691 return ERR_PTR(err);
3695 folio_mark_accessed(folio);
3700 * read_cache_folio - Read into page cache, fill it if needed.
3701 * @mapping: The address_space to read from.
3702 * @index: The index to read.
3703 * @filler: Function to perform the read, or NULL to use aops->read_folio().
3704 * @file: Passed to filler function, may be NULL if not required.
3706 * Read one page into the page cache. If it succeeds, the folio returned
3707 * will contain @index, but it may not be the first page of the folio.
3709 * If the filler function returns an error, it will be returned to the
3712 * Context: May sleep. Expects mapping->invalidate_lock to be held.
3713 * Return: An uptodate folio on success, ERR_PTR() on failure.
3715 struct folio *read_cache_folio(struct address_space *mapping, pgoff_t index,
3716 filler_t filler, struct file *file)
3718 return do_read_cache_folio(mapping, index, filler, file,
3719 mapping_gfp_mask(mapping));
3721 EXPORT_SYMBOL(read_cache_folio);
3724 * mapping_read_folio_gfp - Read into page cache, using specified allocation flags.
3725 * @mapping: The address_space for the folio.
3726 * @index: The index that the allocated folio will contain.
3727 * @gfp: The page allocator flags to use if allocating.
3729 * This is the same as "read_cache_folio(mapping, index, NULL, NULL)", but with
3730 * any new memory allocations done using the specified allocation flags.
3732 * The most likely error from this function is EIO, but ENOMEM is
3733 * possible and so is EINTR. If ->read_folio returns another error,
3734 * that will be returned to the caller.
3736 * The function expects mapping->invalidate_lock to be already held.
3738 * Return: Uptodate folio on success, ERR_PTR() on failure.
3740 struct folio *mapping_read_folio_gfp(struct address_space *mapping,
3741 pgoff_t index, gfp_t gfp)
3743 return do_read_cache_folio(mapping, index, NULL, NULL, gfp);
3745 EXPORT_SYMBOL(mapping_read_folio_gfp);
3747 static struct page *do_read_cache_page(struct address_space *mapping,
3748 pgoff_t index, filler_t *filler, struct file *file, gfp_t gfp)
3750 struct folio *folio;
3752 folio = do_read_cache_folio(mapping, index, filler, file, gfp);
3754 return &folio->page;
3755 return folio_file_page(folio, index);
3758 struct page *read_cache_page(struct address_space *mapping,
3759 pgoff_t index, filler_t *filler, struct file *file)
3761 return do_read_cache_page(mapping, index, filler, file,
3762 mapping_gfp_mask(mapping));
3764 EXPORT_SYMBOL(read_cache_page);
3767 * read_cache_page_gfp - read into page cache, using specified page allocation flags.
3768 * @mapping: the page's address_space
3769 * @index: the page index
3770 * @gfp: the page allocator flags to use if allocating
3772 * This is the same as "read_mapping_page(mapping, index, NULL)", but with
3773 * any new page allocations done using the specified allocation flags.
3775 * If the page does not get brought uptodate, return -EIO.
3777 * The function expects mapping->invalidate_lock to be already held.
3779 * Return: up to date page on success, ERR_PTR() on failure.
3781 struct page *read_cache_page_gfp(struct address_space *mapping,
3785 return do_read_cache_page(mapping, index, NULL, NULL, gfp);
3787 EXPORT_SYMBOL(read_cache_page_gfp);
3790 * Warn about a page cache invalidation failure during a direct I/O write.
3792 static void dio_warn_stale_pagecache(struct file *filp)
3794 static DEFINE_RATELIMIT_STATE(_rs, 86400 * HZ, DEFAULT_RATELIMIT_BURST);
3798 errseq_set(&filp->f_mapping->wb_err, -EIO);
3799 if (__ratelimit(&_rs)) {
3800 path = file_path(filp, pathname, sizeof(pathname));
3803 pr_crit("Page cache invalidation failure on direct I/O. Possible data corruption due to collision with buffered I/O!\n");
3804 pr_crit("File: %s PID: %d Comm: %.20s\n", path, current->pid,
3809 void kiocb_invalidate_post_direct_write(struct kiocb *iocb, size_t count)
3811 struct address_space *mapping = iocb->ki_filp->f_mapping;
3813 if (mapping->nrpages &&
3814 invalidate_inode_pages2_range(mapping,
3815 iocb->ki_pos >> PAGE_SHIFT,
3816 (iocb->ki_pos + count - 1) >> PAGE_SHIFT))
3817 dio_warn_stale_pagecache(iocb->ki_filp);
3821 generic_file_direct_write(struct kiocb *iocb, struct iov_iter *from)
3823 struct address_space *mapping = iocb->ki_filp->f_mapping;
3824 size_t write_len = iov_iter_count(from);
3828 * If a page can not be invalidated, return 0 to fall back
3829 * to buffered write.
3831 written = kiocb_invalidate_pages(iocb, write_len);
3833 if (written == -EBUSY)
3838 written = mapping->a_ops->direct_IO(iocb, from);
3841 * Finally, try again to invalidate clean pages which might have been
3842 * cached by non-direct readahead, or faulted in by get_user_pages()
3843 * if the source of the write was an mmap'ed region of the file
3844 * we're writing. Either one is a pretty crazy thing to do,
3845 * so we don't support it 100%. If this invalidation
3846 * fails, tough, the write still worked...
3848 * Most of the time we do not need this since dio_complete() will do
3849 * the invalidation for us. However there are some file systems that
3850 * do not end up with dio_complete() being called, so let's not break
3851 * them by removing it completely.
3853 * Noticeable example is a blkdev_direct_IO().
3855 * Skip invalidation for async writes or if mapping has no pages.
3858 struct inode *inode = mapping->host;
3859 loff_t pos = iocb->ki_pos;
3861 kiocb_invalidate_post_direct_write(iocb, written);
3863 write_len -= written;
3864 if (pos > i_size_read(inode) && !S_ISBLK(inode->i_mode)) {
3865 i_size_write(inode, pos);
3866 mark_inode_dirty(inode);
3870 if (written != -EIOCBQUEUED)
3871 iov_iter_revert(from, write_len - iov_iter_count(from));
3874 EXPORT_SYMBOL(generic_file_direct_write);
3876 ssize_t generic_perform_write(struct kiocb *iocb, struct iov_iter *i)
3878 struct file *file = iocb->ki_filp;
3879 loff_t pos = iocb->ki_pos;
3880 struct address_space *mapping = file->f_mapping;
3881 const struct address_space_operations *a_ops = mapping->a_ops;
3883 ssize_t written = 0;
3887 unsigned long offset; /* Offset into pagecache page */
3888 unsigned long bytes; /* Bytes to write to page */
3889 size_t copied; /* Bytes copied from user */
3890 void *fsdata = NULL;
3892 offset = (pos & (PAGE_SIZE - 1));
3893 bytes = min_t(unsigned long, PAGE_SIZE - offset,
3898 * Bring in the user page that we will copy from _first_.
3899 * Otherwise there's a nasty deadlock on copying from the
3900 * same page as we're writing to, without it being marked
3903 if (unlikely(fault_in_iov_iter_readable(i, bytes) == bytes)) {
3908 if (fatal_signal_pending(current)) {
3913 status = a_ops->write_begin(file, mapping, pos, bytes,
3915 if (unlikely(status < 0))
3918 if (mapping_writably_mapped(mapping))
3919 flush_dcache_page(page);
3921 copied = copy_page_from_iter_atomic(page, offset, bytes, i);
3922 flush_dcache_page(page);
3924 status = a_ops->write_end(file, mapping, pos, bytes, copied,
3926 if (unlikely(status != copied)) {
3927 iov_iter_revert(i, copied - max(status, 0L));
3928 if (unlikely(status < 0))
3933 if (unlikely(status == 0)) {
3935 * A short copy made ->write_end() reject the
3936 * thing entirely. Might be memory poisoning
3937 * halfway through, might be a race with munmap,
3938 * might be severe memory pressure.
3947 balance_dirty_pages_ratelimited(mapping);
3948 } while (iov_iter_count(i));
3952 iocb->ki_pos += written;
3955 EXPORT_SYMBOL(generic_perform_write);
3958 * __generic_file_write_iter - write data to a file
3959 * @iocb: IO state structure (file, offset, etc.)
3960 * @from: iov_iter with data to write
3962 * This function does all the work needed for actually writing data to a
3963 * file. It does all basic checks, removes SUID from the file, updates
3964 * modification times and calls proper subroutines depending on whether we
3965 * do direct IO or a standard buffered write.
3967 * It expects i_rwsem to be grabbed unless we work on a block device or similar
3968 * object which does not need locking at all.
3970 * This function does *not* take care of syncing data in case of O_SYNC write.
3971 * A caller has to handle it. This is mainly due to the fact that we want to
3972 * avoid syncing under i_rwsem.
3975 * * number of bytes written, even for truncated writes
3976 * * negative error code if no data has been written at all
3978 ssize_t __generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
3980 struct file *file = iocb->ki_filp;
3981 struct address_space *mapping = file->f_mapping;
3982 struct inode *inode = mapping->host;
3985 ret = file_remove_privs(file);
3989 ret = file_update_time(file);
3993 if (iocb->ki_flags & IOCB_DIRECT) {
3994 ret = generic_file_direct_write(iocb, from);
3996 * If the write stopped short of completing, fall back to
3997 * buffered writes. Some filesystems do this for writes to
3998 * holes, for example. For DAX files, a buffered write will
3999 * not succeed (even if it did, DAX does not handle dirty
4000 * page-cache pages correctly).
4002 if (ret < 0 || !iov_iter_count(from) || IS_DAX(inode))
4004 return direct_write_fallback(iocb, from, ret,
4005 generic_perform_write(iocb, from));
4008 return generic_perform_write(iocb, from);
4010 EXPORT_SYMBOL(__generic_file_write_iter);
4013 * generic_file_write_iter - write data to a file
4014 * @iocb: IO state structure
4015 * @from: iov_iter with data to write
4017 * This is a wrapper around __generic_file_write_iter() to be used by most
4018 * filesystems. It takes care of syncing the file in case of O_SYNC file
4019 * and acquires i_rwsem as needed.
4021 * * negative error code if no data has been written at all of
4022 * vfs_fsync_range() failed for a synchronous write
4023 * * number of bytes written, even for truncated writes
4025 ssize_t generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
4027 struct file *file = iocb->ki_filp;
4028 struct inode *inode = file->f_mapping->host;
4032 ret = generic_write_checks(iocb, from);
4034 ret = __generic_file_write_iter(iocb, from);
4035 inode_unlock(inode);
4038 ret = generic_write_sync(iocb, ret);
4041 EXPORT_SYMBOL(generic_file_write_iter);
4044 * filemap_release_folio() - Release fs-specific metadata on a folio.
4045 * @folio: The folio which the kernel is trying to free.
4046 * @gfp: Memory allocation flags (and I/O mode).
4048 * The address_space is trying to release any data attached to a folio
4049 * (presumably at folio->private).
4051 * This will also be called if the private_2 flag is set on a page,
4052 * indicating that the folio has other metadata associated with it.
4054 * The @gfp argument specifies whether I/O may be performed to release
4055 * this page (__GFP_IO), and whether the call may block
4056 * (__GFP_RECLAIM & __GFP_FS).
4058 * Return: %true if the release was successful, otherwise %false.
4060 bool filemap_release_folio(struct folio *folio, gfp_t gfp)
4062 struct address_space * const mapping = folio->mapping;
4064 BUG_ON(!folio_test_locked(folio));
4065 if (folio_test_writeback(folio))
4068 if (mapping && mapping->a_ops->release_folio)
4069 return mapping->a_ops->release_folio(folio, gfp);
4070 return try_to_free_buffers(folio);
4072 EXPORT_SYMBOL(filemap_release_folio);
4074 #ifdef CONFIG_CACHESTAT_SYSCALL
4076 * filemap_cachestat() - compute the page cache statistics of a mapping
4077 * @mapping: The mapping to compute the statistics for.
4078 * @first_index: The starting page cache index.
4079 * @last_index: The final page index (inclusive).
4080 * @cs: the cachestat struct to write the result to.
4082 * This will query the page cache statistics of a mapping in the
4083 * page range of [first_index, last_index] (inclusive). The statistics
4084 * queried include: number of dirty pages, number of pages marked for
4085 * writeback, and the number of (recently) evicted pages.
4087 static void filemap_cachestat(struct address_space *mapping,
4088 pgoff_t first_index, pgoff_t last_index, struct cachestat *cs)
4090 XA_STATE(xas, &mapping->i_pages, first_index);
4091 struct folio *folio;
4094 xas_for_each(&xas, folio, last_index) {
4095 unsigned long nr_pages;
4096 pgoff_t folio_first_index, folio_last_index;
4098 if (xas_retry(&xas, folio))
4101 if (xa_is_value(folio)) {
4102 /* page is evicted */
4103 void *shadow = (void *)folio;
4104 bool workingset; /* not used */
4105 int order = xa_get_order(xas.xa, xas.xa_index);
4107 nr_pages = 1 << order;
4108 folio_first_index = round_down(xas.xa_index, 1 << order);
4109 folio_last_index = folio_first_index + nr_pages - 1;
4111 /* Folios might straddle the range boundaries, only count covered pages */
4112 if (folio_first_index < first_index)
4113 nr_pages -= first_index - folio_first_index;
4115 if (folio_last_index > last_index)
4116 nr_pages -= folio_last_index - last_index;
4118 cs->nr_evicted += nr_pages;
4120 #ifdef CONFIG_SWAP /* implies CONFIG_MMU */
4121 if (shmem_mapping(mapping)) {
4122 /* shmem file - in swap cache */
4123 swp_entry_t swp = radix_to_swp_entry(folio);
4125 shadow = get_shadow_from_swap_cache(swp);
4128 if (workingset_test_recent(shadow, true, &workingset))
4129 cs->nr_recently_evicted += nr_pages;
4134 nr_pages = folio_nr_pages(folio);
4135 folio_first_index = folio_pgoff(folio);
4136 folio_last_index = folio_first_index + nr_pages - 1;
4138 /* Folios might straddle the range boundaries, only count covered pages */
4139 if (folio_first_index < first_index)
4140 nr_pages -= first_index - folio_first_index;
4142 if (folio_last_index > last_index)
4143 nr_pages -= folio_last_index - last_index;
4145 /* page is in cache */
4146 cs->nr_cache += nr_pages;
4148 if (folio_test_dirty(folio))
4149 cs->nr_dirty += nr_pages;
4151 if (folio_test_writeback(folio))
4152 cs->nr_writeback += nr_pages;
4155 if (need_resched()) {
4164 * The cachestat(2) system call.
4166 * cachestat() returns the page cache statistics of a file in the
4167 * bytes range specified by `off` and `len`: number of cached pages,
4168 * number of dirty pages, number of pages marked for writeback,
4169 * number of evicted pages, and number of recently evicted pages.
4171 * An evicted page is a page that is previously in the page cache
4172 * but has been evicted since. A page is recently evicted if its last
4173 * eviction was recent enough that its reentry to the cache would
4174 * indicate that it is actively being used by the system, and that
4175 * there is memory pressure on the system.
4177 * `off` and `len` must be non-negative integers. If `len` > 0,
4178 * the queried range is [`off`, `off` + `len`]. If `len` == 0,
4179 * we will query in the range from `off` to the end of the file.
4181 * The `flags` argument is unused for now, but is included for future
4182 * extensibility. User should pass 0 (i.e no flag specified).
4184 * Currently, hugetlbfs is not supported.
4186 * Because the status of a page can change after cachestat() checks it
4187 * but before it returns to the application, the returned values may
4188 * contain stale information.
4192 * -EFAULT - cstat or cstat_range points to an illegal address
4193 * -EINVAL - invalid flags
4194 * -EBADF - invalid file descriptor
4195 * -EOPNOTSUPP - file descriptor is of a hugetlbfs file
4197 SYSCALL_DEFINE4(cachestat, unsigned int, fd,
4198 struct cachestat_range __user *, cstat_range,
4199 struct cachestat __user *, cstat, unsigned int, flags)
4201 struct fd f = fdget(fd);
4202 struct address_space *mapping;
4203 struct cachestat_range csr;
4204 struct cachestat cs;
4205 pgoff_t first_index, last_index;
4210 if (copy_from_user(&csr, cstat_range,
4211 sizeof(struct cachestat_range))) {
4216 /* hugetlbfs is not supported */
4217 if (is_file_hugepages(f.file)) {
4227 first_index = csr.off >> PAGE_SHIFT;
4229 csr.len == 0 ? ULONG_MAX : (csr.off + csr.len - 1) >> PAGE_SHIFT;
4230 memset(&cs, 0, sizeof(struct cachestat));
4231 mapping = f.file->f_mapping;
4232 filemap_cachestat(mapping, first_index, last_index, &cs);
4235 if (copy_to_user(cstat, &cs, sizeof(struct cachestat)))
4240 #endif /* CONFIG_CACHESTAT_SYSCALL */