1 // SPDX-License-Identifier: GPL-2.0-only
5 * Copyright (C) 1994-1999 Linus Torvalds
9 * This file handles the generic file mmap semantics used by
10 * most "normal" filesystems (but you don't /have/ to use this:
11 * the NFS filesystem used to do this differently, for example)
13 #include <linux/export.h>
14 #include <linux/compiler.h>
15 #include <linux/dax.h>
17 #include <linux/sched/signal.h>
18 #include <linux/uaccess.h>
19 #include <linux/capability.h>
20 #include <linux/kernel_stat.h>
21 #include <linux/gfp.h>
23 #include <linux/swap.h>
24 #include <linux/swapops.h>
25 #include <linux/mman.h>
26 #include <linux/pagemap.h>
27 #include <linux/file.h>
28 #include <linux/uio.h>
29 #include <linux/error-injection.h>
30 #include <linux/hash.h>
31 #include <linux/writeback.h>
32 #include <linux/backing-dev.h>
33 #include <linux/pagevec.h>
34 #include <linux/security.h>
35 #include <linux/cpuset.h>
36 #include <linux/hugetlb.h>
37 #include <linux/memcontrol.h>
38 #include <linux/shmem_fs.h>
39 #include <linux/rmap.h>
40 #include <linux/delayacct.h>
41 #include <linux/psi.h>
42 #include <linux/ramfs.h>
43 #include <linux/page_idle.h>
44 #include <linux/migrate.h>
45 #include <asm/pgalloc.h>
46 #include <asm/tlbflush.h>
49 #define CREATE_TRACE_POINTS
50 #include <trace/events/filemap.h>
53 * FIXME: remove all knowledge of the buffer layer from the core VM
55 #include <linux/buffer_head.h> /* for try_to_free_buffers */
60 * Shared mappings implemented 30.11.1994. It's not fully working yet,
63 * Shared mappings now work. 15.8.1995 Bruno.
65 * finished 'unifying' the page and buffer cache and SMP-threaded the
66 * page-cache, 21.05.1999, Ingo Molnar <mingo@redhat.com>
68 * SMP-threaded pagemap-LRU 1999, Andrea Arcangeli <andrea@suse.de>
74 * ->i_mmap_rwsem (truncate_pagecache)
75 * ->private_lock (__free_pte->block_dirty_folio)
76 * ->swap_lock (exclusive_swap_page, others)
80 * ->invalidate_lock (acquired by fs in truncate path)
81 * ->i_mmap_rwsem (truncate->unmap_mapping_range)
85 * ->page_table_lock or pte_lock (various, mainly in memory.c)
86 * ->i_pages lock (arch-dependent flush_dcache_mmap_lock)
89 * ->invalidate_lock (filemap_fault)
90 * ->lock_page (filemap_fault, access_process_vm)
92 * ->i_rwsem (generic_perform_write)
93 * ->mmap_lock (fault_in_readable->do_page_fault)
96 * sb_lock (fs/fs-writeback.c)
97 * ->i_pages lock (__sync_single_inode)
100 * ->anon_vma.lock (vma_adjust)
103 * ->page_table_lock or pte_lock (anon_vma_prepare and various)
105 * ->page_table_lock or pte_lock
106 * ->swap_lock (try_to_unmap_one)
107 * ->private_lock (try_to_unmap_one)
108 * ->i_pages lock (try_to_unmap_one)
109 * ->lruvec->lru_lock (follow_page->mark_page_accessed)
110 * ->lruvec->lru_lock (check_pte_range->isolate_lru_page)
111 * ->private_lock (page_remove_rmap->set_page_dirty)
112 * ->i_pages lock (page_remove_rmap->set_page_dirty)
113 * bdi.wb->list_lock (page_remove_rmap->set_page_dirty)
114 * ->inode->i_lock (page_remove_rmap->set_page_dirty)
115 * ->memcg->move_lock (page_remove_rmap->lock_page_memcg)
116 * bdi.wb->list_lock (zap_pte_range->set_page_dirty)
117 * ->inode->i_lock (zap_pte_range->set_page_dirty)
118 * ->private_lock (zap_pte_range->block_dirty_folio)
121 * ->tasklist_lock (memory_failure, collect_procs_ao)
124 static void page_cache_delete(struct address_space *mapping,
125 struct folio *folio, void *shadow)
127 XA_STATE(xas, &mapping->i_pages, folio->index);
130 mapping_set_update(&xas, mapping);
132 /* hugetlb pages are represented by a single entry in the xarray */
133 if (!folio_test_hugetlb(folio)) {
134 xas_set_order(&xas, folio->index, folio_order(folio));
135 nr = folio_nr_pages(folio);
138 VM_BUG_ON_FOLIO(!folio_test_locked(folio), folio);
140 xas_store(&xas, shadow);
141 xas_init_marks(&xas);
143 folio->mapping = NULL;
144 /* Leave page->index set: truncation lookup relies upon it */
145 mapping->nrpages -= nr;
148 static void filemap_unaccount_folio(struct address_space *mapping,
153 VM_BUG_ON_FOLIO(folio_mapped(folio), folio);
154 if (!IS_ENABLED(CONFIG_DEBUG_VM) && unlikely(folio_mapped(folio))) {
155 pr_alert("BUG: Bad page cache in process %s pfn:%05lx\n",
156 current->comm, folio_pfn(folio));
157 dump_page(&folio->page, "still mapped when deleted");
159 add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE);
161 if (mapping_exiting(mapping) && !folio_test_large(folio)) {
162 int mapcount = page_mapcount(&folio->page);
164 if (folio_ref_count(folio) >= mapcount + 2) {
166 * All vmas have already been torn down, so it's
167 * a good bet that actually the page is unmapped
168 * and we'd rather not leak it: if we're wrong,
169 * another bad page check should catch it later.
171 page_mapcount_reset(&folio->page);
172 folio_ref_sub(folio, mapcount);
177 /* hugetlb folios do not participate in page cache accounting. */
178 if (folio_test_hugetlb(folio))
181 nr = folio_nr_pages(folio);
183 __lruvec_stat_mod_folio(folio, NR_FILE_PAGES, -nr);
184 if (folio_test_swapbacked(folio)) {
185 __lruvec_stat_mod_folio(folio, NR_SHMEM, -nr);
186 if (folio_test_pmd_mappable(folio))
187 __lruvec_stat_mod_folio(folio, NR_SHMEM_THPS, -nr);
188 } else if (folio_test_pmd_mappable(folio)) {
189 __lruvec_stat_mod_folio(folio, NR_FILE_THPS, -nr);
190 filemap_nr_thps_dec(mapping);
194 * At this point folio must be either written or cleaned by
195 * truncate. Dirty folio here signals a bug and loss of
198 * This fixes dirty accounting after removing the folio entirely
199 * but leaves the dirty flag set: it has no effect for truncated
200 * folio and anyway will be cleared before returning folio to
203 if (WARN_ON_ONCE(folio_test_dirty(folio)))
204 folio_account_cleaned(folio, mapping,
205 inode_to_wb(mapping->host));
209 * Delete a page from the page cache and free it. Caller has to make
210 * sure the page is locked and that nobody else uses it - or that usage
211 * is safe. The caller must hold the i_pages lock.
213 void __filemap_remove_folio(struct folio *folio, void *shadow)
215 struct address_space *mapping = folio->mapping;
217 trace_mm_filemap_delete_from_page_cache(folio);
218 filemap_unaccount_folio(mapping, folio);
219 page_cache_delete(mapping, folio, shadow);
222 void filemap_free_folio(struct address_space *mapping, struct folio *folio)
224 void (*freepage)(struct page *);
227 freepage = mapping->a_ops->freepage;
229 freepage(&folio->page);
231 if (folio_test_large(folio) && !folio_test_hugetlb(folio))
232 refs = folio_nr_pages(folio);
233 folio_put_refs(folio, refs);
237 * filemap_remove_folio - Remove folio from page cache.
240 * This must be called only on folios that are locked and have been
241 * verified to be in the page cache. It will never put the folio into
242 * the free list because the caller has a reference on the page.
244 void filemap_remove_folio(struct folio *folio)
246 struct address_space *mapping = folio->mapping;
248 BUG_ON(!folio_test_locked(folio));
249 spin_lock(&mapping->host->i_lock);
250 xa_lock_irq(&mapping->i_pages);
251 __filemap_remove_folio(folio, NULL);
252 xa_unlock_irq(&mapping->i_pages);
253 if (mapping_shrinkable(mapping))
254 inode_add_lru(mapping->host);
255 spin_unlock(&mapping->host->i_lock);
257 filemap_free_folio(mapping, folio);
261 * page_cache_delete_batch - delete several folios from page cache
262 * @mapping: the mapping to which folios belong
263 * @fbatch: batch of folios to delete
265 * The function walks over mapping->i_pages and removes folios passed in
266 * @fbatch from the mapping. The function expects @fbatch to be sorted
267 * by page index and is optimised for it to be dense.
268 * It tolerates holes in @fbatch (mapping entries at those indices are not
271 * The function expects the i_pages lock to be held.
273 static void page_cache_delete_batch(struct address_space *mapping,
274 struct folio_batch *fbatch)
276 XA_STATE(xas, &mapping->i_pages, fbatch->folios[0]->index);
277 long total_pages = 0;
281 mapping_set_update(&xas, mapping);
282 xas_for_each(&xas, folio, ULONG_MAX) {
283 if (i >= folio_batch_count(fbatch))
286 /* A swap/dax/shadow entry got inserted? Skip it. */
287 if (xa_is_value(folio))
290 * A page got inserted in our range? Skip it. We have our
291 * pages locked so they are protected from being removed.
292 * If we see a page whose index is higher than ours, it
293 * means our page has been removed, which shouldn't be
294 * possible because we're holding the PageLock.
296 if (folio != fbatch->folios[i]) {
297 VM_BUG_ON_FOLIO(folio->index >
298 fbatch->folios[i]->index, folio);
302 WARN_ON_ONCE(!folio_test_locked(folio));
304 folio->mapping = NULL;
305 /* Leave folio->index set: truncation lookup relies on it */
308 xas_store(&xas, NULL);
309 total_pages += folio_nr_pages(folio);
311 mapping->nrpages -= total_pages;
314 void delete_from_page_cache_batch(struct address_space *mapping,
315 struct folio_batch *fbatch)
319 if (!folio_batch_count(fbatch))
322 spin_lock(&mapping->host->i_lock);
323 xa_lock_irq(&mapping->i_pages);
324 for (i = 0; i < folio_batch_count(fbatch); i++) {
325 struct folio *folio = fbatch->folios[i];
327 trace_mm_filemap_delete_from_page_cache(folio);
328 filemap_unaccount_folio(mapping, folio);
330 page_cache_delete_batch(mapping, fbatch);
331 xa_unlock_irq(&mapping->i_pages);
332 if (mapping_shrinkable(mapping))
333 inode_add_lru(mapping->host);
334 spin_unlock(&mapping->host->i_lock);
336 for (i = 0; i < folio_batch_count(fbatch); i++)
337 filemap_free_folio(mapping, fbatch->folios[i]);
340 int filemap_check_errors(struct address_space *mapping)
343 /* Check for outstanding write errors */
344 if (test_bit(AS_ENOSPC, &mapping->flags) &&
345 test_and_clear_bit(AS_ENOSPC, &mapping->flags))
347 if (test_bit(AS_EIO, &mapping->flags) &&
348 test_and_clear_bit(AS_EIO, &mapping->flags))
352 EXPORT_SYMBOL(filemap_check_errors);
354 static int filemap_check_and_keep_errors(struct address_space *mapping)
356 /* Check for outstanding write errors */
357 if (test_bit(AS_EIO, &mapping->flags))
359 if (test_bit(AS_ENOSPC, &mapping->flags))
365 * filemap_fdatawrite_wbc - start writeback on mapping dirty pages in range
366 * @mapping: address space structure to write
367 * @wbc: the writeback_control controlling the writeout
369 * Call writepages on the mapping using the provided wbc to control the
372 * Return: %0 on success, negative error code otherwise.
374 int filemap_fdatawrite_wbc(struct address_space *mapping,
375 struct writeback_control *wbc)
379 if (!mapping_can_writeback(mapping) ||
380 !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
383 wbc_attach_fdatawrite_inode(wbc, mapping->host);
384 ret = do_writepages(mapping, wbc);
385 wbc_detach_inode(wbc);
388 EXPORT_SYMBOL(filemap_fdatawrite_wbc);
391 * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range
392 * @mapping: address space structure to write
393 * @start: offset in bytes where the range starts
394 * @end: offset in bytes where the range ends (inclusive)
395 * @sync_mode: enable synchronous operation
397 * Start writeback against all of a mapping's dirty pages that lie
398 * within the byte offsets <start, end> inclusive.
400 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
401 * opposed to a regular memory cleansing writeback. The difference between
402 * these two operations is that if a dirty page/buffer is encountered, it must
403 * be waited upon, and not just skipped over.
405 * Return: %0 on success, negative error code otherwise.
407 int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
408 loff_t end, int sync_mode)
410 struct writeback_control wbc = {
411 .sync_mode = sync_mode,
412 .nr_to_write = LONG_MAX,
413 .range_start = start,
417 return filemap_fdatawrite_wbc(mapping, &wbc);
420 static inline int __filemap_fdatawrite(struct address_space *mapping,
423 return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode);
426 int filemap_fdatawrite(struct address_space *mapping)
428 return __filemap_fdatawrite(mapping, WB_SYNC_ALL);
430 EXPORT_SYMBOL(filemap_fdatawrite);
432 int filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
435 return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL);
437 EXPORT_SYMBOL(filemap_fdatawrite_range);
440 * filemap_flush - mostly a non-blocking flush
441 * @mapping: target address_space
443 * This is a mostly non-blocking flush. Not suitable for data-integrity
444 * purposes - I/O may not be started against all dirty pages.
446 * Return: %0 on success, negative error code otherwise.
448 int filemap_flush(struct address_space *mapping)
450 return __filemap_fdatawrite(mapping, WB_SYNC_NONE);
452 EXPORT_SYMBOL(filemap_flush);
455 * filemap_range_has_page - check if a page exists in range.
456 * @mapping: address space within which to check
457 * @start_byte: offset in bytes where the range starts
458 * @end_byte: offset in bytes where the range ends (inclusive)
460 * Find at least one page in the range supplied, usually used to check if
461 * direct writing in this range will trigger a writeback.
463 * Return: %true if at least one page exists in the specified range,
466 bool filemap_range_has_page(struct address_space *mapping,
467 loff_t start_byte, loff_t end_byte)
470 XA_STATE(xas, &mapping->i_pages, start_byte >> PAGE_SHIFT);
471 pgoff_t max = end_byte >> PAGE_SHIFT;
473 if (end_byte < start_byte)
478 page = xas_find(&xas, max);
479 if (xas_retry(&xas, page))
481 /* Shadow entries don't count */
482 if (xa_is_value(page))
485 * We don't need to try to pin this page; we're about to
486 * release the RCU lock anyway. It is enough to know that
487 * there was a page here recently.
495 EXPORT_SYMBOL(filemap_range_has_page);
497 static void __filemap_fdatawait_range(struct address_space *mapping,
498 loff_t start_byte, loff_t end_byte)
500 pgoff_t index = start_byte >> PAGE_SHIFT;
501 pgoff_t end = end_byte >> PAGE_SHIFT;
505 if (end_byte < start_byte)
509 while (index <= end) {
512 nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index,
513 end, PAGECACHE_TAG_WRITEBACK);
517 for (i = 0; i < nr_pages; i++) {
518 struct page *page = pvec.pages[i];
520 wait_on_page_writeback(page);
521 ClearPageError(page);
523 pagevec_release(&pvec);
529 * filemap_fdatawait_range - wait for writeback to complete
530 * @mapping: address space structure to wait for
531 * @start_byte: offset in bytes where the range starts
532 * @end_byte: offset in bytes where the range ends (inclusive)
534 * Walk the list of under-writeback pages of the given address space
535 * in the given range and wait for all of them. Check error status of
536 * the address space and return it.
538 * Since the error status of the address space is cleared by this function,
539 * callers are responsible for checking the return value and handling and/or
540 * reporting the error.
542 * Return: error status of the address space.
544 int filemap_fdatawait_range(struct address_space *mapping, loff_t start_byte,
547 __filemap_fdatawait_range(mapping, start_byte, end_byte);
548 return filemap_check_errors(mapping);
550 EXPORT_SYMBOL(filemap_fdatawait_range);
553 * filemap_fdatawait_range_keep_errors - wait for writeback to complete
554 * @mapping: address space structure to wait for
555 * @start_byte: offset in bytes where the range starts
556 * @end_byte: offset in bytes where the range ends (inclusive)
558 * Walk the list of under-writeback pages of the given address space in the
559 * given range and wait for all of them. Unlike filemap_fdatawait_range(),
560 * this function does not clear error status of the address space.
562 * Use this function if callers don't handle errors themselves. Expected
563 * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2),
566 int filemap_fdatawait_range_keep_errors(struct address_space *mapping,
567 loff_t start_byte, loff_t end_byte)
569 __filemap_fdatawait_range(mapping, start_byte, end_byte);
570 return filemap_check_and_keep_errors(mapping);
572 EXPORT_SYMBOL(filemap_fdatawait_range_keep_errors);
575 * file_fdatawait_range - wait for writeback to complete
576 * @file: file pointing to address space structure to wait for
577 * @start_byte: offset in bytes where the range starts
578 * @end_byte: offset in bytes where the range ends (inclusive)
580 * Walk the list of under-writeback pages of the address space that file
581 * refers to, in the given range and wait for all of them. Check error
582 * status of the address space vs. the file->f_wb_err cursor and return it.
584 * Since the error status of the file is advanced by this function,
585 * callers are responsible for checking the return value and handling and/or
586 * reporting the error.
588 * Return: error status of the address space vs. the file->f_wb_err cursor.
590 int file_fdatawait_range(struct file *file, loff_t start_byte, loff_t end_byte)
592 struct address_space *mapping = file->f_mapping;
594 __filemap_fdatawait_range(mapping, start_byte, end_byte);
595 return file_check_and_advance_wb_err(file);
597 EXPORT_SYMBOL(file_fdatawait_range);
600 * filemap_fdatawait_keep_errors - wait for writeback without clearing errors
601 * @mapping: address space structure to wait for
603 * Walk the list of under-writeback pages of the given address space
604 * and wait for all of them. Unlike filemap_fdatawait(), this function
605 * does not clear error status of the address space.
607 * Use this function if callers don't handle errors themselves. Expected
608 * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2),
611 * Return: error status of the address space.
613 int filemap_fdatawait_keep_errors(struct address_space *mapping)
615 __filemap_fdatawait_range(mapping, 0, LLONG_MAX);
616 return filemap_check_and_keep_errors(mapping);
618 EXPORT_SYMBOL(filemap_fdatawait_keep_errors);
620 /* Returns true if writeback might be needed or already in progress. */
621 static bool mapping_needs_writeback(struct address_space *mapping)
623 return mapping->nrpages;
626 bool filemap_range_has_writeback(struct address_space *mapping,
627 loff_t start_byte, loff_t end_byte)
629 XA_STATE(xas, &mapping->i_pages, start_byte >> PAGE_SHIFT);
630 pgoff_t max = end_byte >> PAGE_SHIFT;
633 if (end_byte < start_byte)
637 xas_for_each(&xas, page, max) {
638 if (xas_retry(&xas, page))
640 if (xa_is_value(page))
642 if (PageDirty(page) || PageLocked(page) || PageWriteback(page))
648 EXPORT_SYMBOL_GPL(filemap_range_has_writeback);
651 * filemap_write_and_wait_range - write out & wait on a file range
652 * @mapping: the address_space for the pages
653 * @lstart: offset in bytes where the range starts
654 * @lend: offset in bytes where the range ends (inclusive)
656 * Write out and wait upon file offsets lstart->lend, inclusive.
658 * Note that @lend is inclusive (describes the last byte to be written) so
659 * that this function can be used to write to the very end-of-file (end = -1).
661 * Return: error status of the address space.
663 int filemap_write_and_wait_range(struct address_space *mapping,
664 loff_t lstart, loff_t lend)
668 if (mapping_needs_writeback(mapping)) {
669 err = __filemap_fdatawrite_range(mapping, lstart, lend,
672 * Even if the above returned error, the pages may be
673 * written partially (e.g. -ENOSPC), so we wait for it.
674 * But the -EIO is special case, it may indicate the worst
675 * thing (e.g. bug) happened, so we avoid waiting for it.
678 int err2 = filemap_fdatawait_range(mapping,
683 /* Clear any previously stored errors */
684 filemap_check_errors(mapping);
687 err = filemap_check_errors(mapping);
691 EXPORT_SYMBOL(filemap_write_and_wait_range);
693 void __filemap_set_wb_err(struct address_space *mapping, int err)
695 errseq_t eseq = errseq_set(&mapping->wb_err, err);
697 trace_filemap_set_wb_err(mapping, eseq);
699 EXPORT_SYMBOL(__filemap_set_wb_err);
702 * file_check_and_advance_wb_err - report wb error (if any) that was previously
703 * and advance wb_err to current one
704 * @file: struct file on which the error is being reported
706 * When userland calls fsync (or something like nfsd does the equivalent), we
707 * want to report any writeback errors that occurred since the last fsync (or
708 * since the file was opened if there haven't been any).
710 * Grab the wb_err from the mapping. If it matches what we have in the file,
711 * then just quickly return 0. The file is all caught up.
713 * If it doesn't match, then take the mapping value, set the "seen" flag in
714 * it and try to swap it into place. If it works, or another task beat us
715 * to it with the new value, then update the f_wb_err and return the error
716 * portion. The error at this point must be reported via proper channels
717 * (a'la fsync, or NFS COMMIT operation, etc.).
719 * While we handle mapping->wb_err with atomic operations, the f_wb_err
720 * value is protected by the f_lock since we must ensure that it reflects
721 * the latest value swapped in for this file descriptor.
723 * Return: %0 on success, negative error code otherwise.
725 int file_check_and_advance_wb_err(struct file *file)
728 errseq_t old = READ_ONCE(file->f_wb_err);
729 struct address_space *mapping = file->f_mapping;
731 /* Locklessly handle the common case where nothing has changed */
732 if (errseq_check(&mapping->wb_err, old)) {
733 /* Something changed, must use slow path */
734 spin_lock(&file->f_lock);
735 old = file->f_wb_err;
736 err = errseq_check_and_advance(&mapping->wb_err,
738 trace_file_check_and_advance_wb_err(file, old);
739 spin_unlock(&file->f_lock);
743 * We're mostly using this function as a drop in replacement for
744 * filemap_check_errors. Clear AS_EIO/AS_ENOSPC to emulate the effect
745 * that the legacy code would have had on these flags.
747 clear_bit(AS_EIO, &mapping->flags);
748 clear_bit(AS_ENOSPC, &mapping->flags);
751 EXPORT_SYMBOL(file_check_and_advance_wb_err);
754 * file_write_and_wait_range - write out & wait on a file range
755 * @file: file pointing to address_space with pages
756 * @lstart: offset in bytes where the range starts
757 * @lend: offset in bytes where the range ends (inclusive)
759 * Write out and wait upon file offsets lstart->lend, inclusive.
761 * Note that @lend is inclusive (describes the last byte to be written) so
762 * that this function can be used to write to the very end-of-file (end = -1).
764 * After writing out and waiting on the data, we check and advance the
765 * f_wb_err cursor to the latest value, and return any errors detected there.
767 * Return: %0 on success, negative error code otherwise.
769 int file_write_and_wait_range(struct file *file, loff_t lstart, loff_t lend)
772 struct address_space *mapping = file->f_mapping;
774 if (mapping_needs_writeback(mapping)) {
775 err = __filemap_fdatawrite_range(mapping, lstart, lend,
777 /* See comment of filemap_write_and_wait() */
779 __filemap_fdatawait_range(mapping, lstart, lend);
781 err2 = file_check_and_advance_wb_err(file);
786 EXPORT_SYMBOL(file_write_and_wait_range);
789 * replace_page_cache_page - replace a pagecache page with a new one
790 * @old: page to be replaced
791 * @new: page to replace with
793 * This function replaces a page in the pagecache with a new one. On
794 * success it acquires the pagecache reference for the new page and
795 * drops it for the old page. Both the old and new pages must be
796 * locked. This function does not add the new page to the LRU, the
797 * caller must do that.
799 * The remove + add is atomic. This function cannot fail.
801 void replace_page_cache_page(struct page *old, struct page *new)
803 struct folio *fold = page_folio(old);
804 struct folio *fnew = page_folio(new);
805 struct address_space *mapping = old->mapping;
806 void (*freepage)(struct page *) = mapping->a_ops->freepage;
807 pgoff_t offset = old->index;
808 XA_STATE(xas, &mapping->i_pages, offset);
810 VM_BUG_ON_PAGE(!PageLocked(old), old);
811 VM_BUG_ON_PAGE(!PageLocked(new), new);
812 VM_BUG_ON_PAGE(new->mapping, new);
815 new->mapping = mapping;
818 mem_cgroup_migrate(fold, fnew);
821 xas_store(&xas, new);
824 /* hugetlb pages do not participate in page cache accounting. */
826 __dec_lruvec_page_state(old, NR_FILE_PAGES);
828 __inc_lruvec_page_state(new, NR_FILE_PAGES);
829 if (PageSwapBacked(old))
830 __dec_lruvec_page_state(old, NR_SHMEM);
831 if (PageSwapBacked(new))
832 __inc_lruvec_page_state(new, NR_SHMEM);
833 xas_unlock_irq(&xas);
838 EXPORT_SYMBOL_GPL(replace_page_cache_page);
840 noinline int __filemap_add_folio(struct address_space *mapping,
841 struct folio *folio, pgoff_t index, gfp_t gfp, void **shadowp)
843 XA_STATE(xas, &mapping->i_pages, index);
844 int huge = folio_test_hugetlb(folio);
845 bool charged = false;
848 VM_BUG_ON_FOLIO(!folio_test_locked(folio), folio);
849 VM_BUG_ON_FOLIO(folio_test_swapbacked(folio), folio);
850 mapping_set_update(&xas, mapping);
853 int error = mem_cgroup_charge(folio, NULL, gfp);
854 VM_BUG_ON_FOLIO(index & (folio_nr_pages(folio) - 1), folio);
858 xas_set_order(&xas, index, folio_order(folio));
859 nr = folio_nr_pages(folio);
862 gfp &= GFP_RECLAIM_MASK;
863 folio_ref_add(folio, nr);
864 folio->mapping = mapping;
865 folio->index = xas.xa_index;
868 unsigned int order = xa_get_order(xas.xa, xas.xa_index);
869 void *entry, *old = NULL;
871 if (order > folio_order(folio))
872 xas_split_alloc(&xas, xa_load(xas.xa, xas.xa_index),
875 xas_for_each_conflict(&xas, entry) {
877 if (!xa_is_value(entry)) {
878 xas_set_err(&xas, -EEXIST);
886 /* entry may have been split before we acquired lock */
887 order = xa_get_order(xas.xa, xas.xa_index);
888 if (order > folio_order(folio)) {
889 /* How to handle large swap entries? */
890 BUG_ON(shmem_mapping(mapping));
891 xas_split(&xas, old, order);
896 xas_store(&xas, folio);
900 mapping->nrpages += nr;
902 /* hugetlb pages do not participate in page cache accounting */
904 __lruvec_stat_mod_folio(folio, NR_FILE_PAGES, nr);
905 if (folio_test_pmd_mappable(folio))
906 __lruvec_stat_mod_folio(folio,
910 xas_unlock_irq(&xas);
911 } while (xas_nomem(&xas, gfp));
916 trace_mm_filemap_add_to_page_cache(folio);
920 mem_cgroup_uncharge(folio);
921 folio->mapping = NULL;
922 /* Leave page->index set: truncation relies upon it */
923 folio_put_refs(folio, nr);
924 return xas_error(&xas);
926 ALLOW_ERROR_INJECTION(__filemap_add_folio, ERRNO);
929 * add_to_page_cache_locked - add a locked page to the pagecache
931 * @mapping: the page's address_space
932 * @offset: page index
933 * @gfp_mask: page allocation mode
935 * This function is used to add a page to the pagecache. It must be locked.
936 * This function does not add the page to the LRU. The caller must do that.
938 * Return: %0 on success, negative error code otherwise.
940 int add_to_page_cache_locked(struct page *page, struct address_space *mapping,
941 pgoff_t offset, gfp_t gfp_mask)
943 return __filemap_add_folio(mapping, page_folio(page), offset,
946 EXPORT_SYMBOL(add_to_page_cache_locked);
948 int filemap_add_folio(struct address_space *mapping, struct folio *folio,
949 pgoff_t index, gfp_t gfp)
954 __folio_set_locked(folio);
955 ret = __filemap_add_folio(mapping, folio, index, gfp, &shadow);
957 __folio_clear_locked(folio);
960 * The folio might have been evicted from cache only
961 * recently, in which case it should be activated like
962 * any other repeatedly accessed folio.
963 * The exception is folios getting rewritten; evicting other
964 * data from the working set, only to cache data that will
965 * get overwritten with something else, is a waste of memory.
967 WARN_ON_ONCE(folio_test_active(folio));
968 if (!(gfp & __GFP_WRITE) && shadow)
969 workingset_refault(folio, shadow);
970 folio_add_lru(folio);
974 EXPORT_SYMBOL_GPL(filemap_add_folio);
977 struct folio *filemap_alloc_folio(gfp_t gfp, unsigned int order)
982 if (cpuset_do_page_mem_spread()) {
983 unsigned int cpuset_mems_cookie;
985 cpuset_mems_cookie = read_mems_allowed_begin();
986 n = cpuset_mem_spread_node();
987 folio = __folio_alloc_node(gfp, order, n);
988 } while (!folio && read_mems_allowed_retry(cpuset_mems_cookie));
992 return folio_alloc(gfp, order);
994 EXPORT_SYMBOL(filemap_alloc_folio);
998 * filemap_invalidate_lock_two - lock invalidate_lock for two mappings
1000 * Lock exclusively invalidate_lock of any passed mapping that is not NULL.
1002 * @mapping1: the first mapping to lock
1003 * @mapping2: the second mapping to lock
1005 void filemap_invalidate_lock_two(struct address_space *mapping1,
1006 struct address_space *mapping2)
1008 if (mapping1 > mapping2)
1009 swap(mapping1, mapping2);
1011 down_write(&mapping1->invalidate_lock);
1012 if (mapping2 && mapping1 != mapping2)
1013 down_write_nested(&mapping2->invalidate_lock, 1);
1015 EXPORT_SYMBOL(filemap_invalidate_lock_two);
1018 * filemap_invalidate_unlock_two - unlock invalidate_lock for two mappings
1020 * Unlock exclusive invalidate_lock of any passed mapping that is not NULL.
1022 * @mapping1: the first mapping to unlock
1023 * @mapping2: the second mapping to unlock
1025 void filemap_invalidate_unlock_two(struct address_space *mapping1,
1026 struct address_space *mapping2)
1029 up_write(&mapping1->invalidate_lock);
1030 if (mapping2 && mapping1 != mapping2)
1031 up_write(&mapping2->invalidate_lock);
1033 EXPORT_SYMBOL(filemap_invalidate_unlock_two);
1036 * In order to wait for pages to become available there must be
1037 * waitqueues associated with pages. By using a hash table of
1038 * waitqueues where the bucket discipline is to maintain all
1039 * waiters on the same queue and wake all when any of the pages
1040 * become available, and for the woken contexts to check to be
1041 * sure the appropriate page became available, this saves space
1042 * at a cost of "thundering herd" phenomena during rare hash
1045 #define PAGE_WAIT_TABLE_BITS 8
1046 #define PAGE_WAIT_TABLE_SIZE (1 << PAGE_WAIT_TABLE_BITS)
1047 static wait_queue_head_t folio_wait_table[PAGE_WAIT_TABLE_SIZE] __cacheline_aligned;
1049 static wait_queue_head_t *folio_waitqueue(struct folio *folio)
1051 return &folio_wait_table[hash_ptr(folio, PAGE_WAIT_TABLE_BITS)];
1054 void __init pagecache_init(void)
1058 for (i = 0; i < PAGE_WAIT_TABLE_SIZE; i++)
1059 init_waitqueue_head(&folio_wait_table[i]);
1061 page_writeback_init();
1064 * tmpfs uses the ZERO_PAGE for reading holes: it is up-to-date,
1065 * and splice's page_cache_pipe_buf_confirm() needs to see that.
1067 SetPageUptodate(ZERO_PAGE(0));
1071 * The page wait code treats the "wait->flags" somewhat unusually, because
1072 * we have multiple different kinds of waits, not just the usual "exclusive"
1077 * (a) no special bits set:
1079 * We're just waiting for the bit to be released, and when a waker
1080 * calls the wakeup function, we set WQ_FLAG_WOKEN and wake it up,
1081 * and remove it from the wait queue.
1083 * Simple and straightforward.
1085 * (b) WQ_FLAG_EXCLUSIVE:
1087 * The waiter is waiting to get the lock, and only one waiter should
1088 * be woken up to avoid any thundering herd behavior. We'll set the
1089 * WQ_FLAG_WOKEN bit, wake it up, and remove it from the wait queue.
1091 * This is the traditional exclusive wait.
1093 * (c) WQ_FLAG_EXCLUSIVE | WQ_FLAG_CUSTOM:
1095 * The waiter is waiting to get the bit, and additionally wants the
1096 * lock to be transferred to it for fair lock behavior. If the lock
1097 * cannot be taken, we stop walking the wait queue without waking
1100 * This is the "fair lock handoff" case, and in addition to setting
1101 * WQ_FLAG_WOKEN, we set WQ_FLAG_DONE to let the waiter easily see
1102 * that it now has the lock.
1104 static int wake_page_function(wait_queue_entry_t *wait, unsigned mode, int sync, void *arg)
1107 struct wait_page_key *key = arg;
1108 struct wait_page_queue *wait_page
1109 = container_of(wait, struct wait_page_queue, wait);
1111 if (!wake_page_match(wait_page, key))
1115 * If it's a lock handoff wait, we get the bit for it, and
1116 * stop walking (and do not wake it up) if we can't.
1118 flags = wait->flags;
1119 if (flags & WQ_FLAG_EXCLUSIVE) {
1120 if (test_bit(key->bit_nr, &key->folio->flags))
1122 if (flags & WQ_FLAG_CUSTOM) {
1123 if (test_and_set_bit(key->bit_nr, &key->folio->flags))
1125 flags |= WQ_FLAG_DONE;
1130 * We are holding the wait-queue lock, but the waiter that
1131 * is waiting for this will be checking the flags without
1134 * So update the flags atomically, and wake up the waiter
1135 * afterwards to avoid any races. This store-release pairs
1136 * with the load-acquire in folio_wait_bit_common().
1138 smp_store_release(&wait->flags, flags | WQ_FLAG_WOKEN);
1139 wake_up_state(wait->private, mode);
1142 * Ok, we have successfully done what we're waiting for,
1143 * and we can unconditionally remove the wait entry.
1145 * Note that this pairs with the "finish_wait()" in the
1146 * waiter, and has to be the absolute last thing we do.
1147 * After this list_del_init(&wait->entry) the wait entry
1148 * might be de-allocated and the process might even have
1151 list_del_init_careful(&wait->entry);
1152 return (flags & WQ_FLAG_EXCLUSIVE) != 0;
1155 static void folio_wake_bit(struct folio *folio, int bit_nr)
1157 wait_queue_head_t *q = folio_waitqueue(folio);
1158 struct wait_page_key key;
1159 unsigned long flags;
1160 wait_queue_entry_t bookmark;
1163 key.bit_nr = bit_nr;
1167 bookmark.private = NULL;
1168 bookmark.func = NULL;
1169 INIT_LIST_HEAD(&bookmark.entry);
1171 spin_lock_irqsave(&q->lock, flags);
1172 __wake_up_locked_key_bookmark(q, TASK_NORMAL, &key, &bookmark);
1174 while (bookmark.flags & WQ_FLAG_BOOKMARK) {
1176 * Take a breather from holding the lock,
1177 * allow pages that finish wake up asynchronously
1178 * to acquire the lock and remove themselves
1181 spin_unlock_irqrestore(&q->lock, flags);
1183 spin_lock_irqsave(&q->lock, flags);
1184 __wake_up_locked_key_bookmark(q, TASK_NORMAL, &key, &bookmark);
1188 * It's possible to miss clearing waiters here, when we woke our page
1189 * waiters, but the hashed waitqueue has waiters for other pages on it.
1190 * That's okay, it's a rare case. The next waker will clear it.
1192 * Note that, depending on the page pool (buddy, hugetlb, ZONE_DEVICE,
1193 * other), the flag may be cleared in the course of freeing the page;
1194 * but that is not required for correctness.
1196 if (!waitqueue_active(q) || !key.page_match)
1197 folio_clear_waiters(folio);
1199 spin_unlock_irqrestore(&q->lock, flags);
1202 static void folio_wake(struct folio *folio, int bit)
1204 if (!folio_test_waiters(folio))
1206 folio_wake_bit(folio, bit);
1210 * A choice of three behaviors for folio_wait_bit_common():
1213 EXCLUSIVE, /* Hold ref to page and take the bit when woken, like
1214 * __folio_lock() waiting on then setting PG_locked.
1216 SHARED, /* Hold ref to page and check the bit when woken, like
1217 * folio_wait_writeback() waiting on PG_writeback.
1219 DROP, /* Drop ref to page before wait, no check when woken,
1220 * like folio_put_wait_locked() on PG_locked.
1225 * Attempt to check (or get) the folio flag, and mark us done
1228 static inline bool folio_trylock_flag(struct folio *folio, int bit_nr,
1229 struct wait_queue_entry *wait)
1231 if (wait->flags & WQ_FLAG_EXCLUSIVE) {
1232 if (test_and_set_bit(bit_nr, &folio->flags))
1234 } else if (test_bit(bit_nr, &folio->flags))
1237 wait->flags |= WQ_FLAG_WOKEN | WQ_FLAG_DONE;
1241 /* How many times do we accept lock stealing from under a waiter? */
1242 int sysctl_page_lock_unfairness = 5;
1244 static inline int folio_wait_bit_common(struct folio *folio, int bit_nr,
1245 int state, enum behavior behavior)
1247 wait_queue_head_t *q = folio_waitqueue(folio);
1248 int unfairness = sysctl_page_lock_unfairness;
1249 struct wait_page_queue wait_page;
1250 wait_queue_entry_t *wait = &wait_page.wait;
1251 bool thrashing = false;
1252 bool delayacct = false;
1253 unsigned long pflags;
1255 if (bit_nr == PG_locked &&
1256 !folio_test_uptodate(folio) && folio_test_workingset(folio)) {
1257 if (!folio_test_swapbacked(folio)) {
1258 delayacct_thrashing_start();
1261 psi_memstall_enter(&pflags);
1266 wait->func = wake_page_function;
1267 wait_page.folio = folio;
1268 wait_page.bit_nr = bit_nr;
1272 if (behavior == EXCLUSIVE) {
1273 wait->flags = WQ_FLAG_EXCLUSIVE;
1274 if (--unfairness < 0)
1275 wait->flags |= WQ_FLAG_CUSTOM;
1279 * Do one last check whether we can get the
1280 * page bit synchronously.
1282 * Do the folio_set_waiters() marking before that
1283 * to let any waker we _just_ missed know they
1284 * need to wake us up (otherwise they'll never
1285 * even go to the slow case that looks at the
1286 * page queue), and add ourselves to the wait
1287 * queue if we need to sleep.
1289 * This part needs to be done under the queue
1290 * lock to avoid races.
1292 spin_lock_irq(&q->lock);
1293 folio_set_waiters(folio);
1294 if (!folio_trylock_flag(folio, bit_nr, wait))
1295 __add_wait_queue_entry_tail(q, wait);
1296 spin_unlock_irq(&q->lock);
1299 * From now on, all the logic will be based on
1300 * the WQ_FLAG_WOKEN and WQ_FLAG_DONE flag, to
1301 * see whether the page bit testing has already
1302 * been done by the wake function.
1304 * We can drop our reference to the folio.
1306 if (behavior == DROP)
1310 * Note that until the "finish_wait()", or until
1311 * we see the WQ_FLAG_WOKEN flag, we need to
1312 * be very careful with the 'wait->flags', because
1313 * we may race with a waker that sets them.
1318 set_current_state(state);
1320 /* Loop until we've been woken or interrupted */
1321 flags = smp_load_acquire(&wait->flags);
1322 if (!(flags & WQ_FLAG_WOKEN)) {
1323 if (signal_pending_state(state, current))
1330 /* If we were non-exclusive, we're done */
1331 if (behavior != EXCLUSIVE)
1334 /* If the waker got the lock for us, we're done */
1335 if (flags & WQ_FLAG_DONE)
1339 * Otherwise, if we're getting the lock, we need to
1340 * try to get it ourselves.
1342 * And if that fails, we'll have to retry this all.
1344 if (unlikely(test_and_set_bit(bit_nr, folio_flags(folio, 0))))
1347 wait->flags |= WQ_FLAG_DONE;
1352 * If a signal happened, this 'finish_wait()' may remove the last
1353 * waiter from the wait-queues, but the folio waiters bit will remain
1354 * set. That's ok. The next wakeup will take care of it, and trying
1355 * to do it here would be difficult and prone to races.
1357 finish_wait(q, wait);
1361 delayacct_thrashing_end();
1362 psi_memstall_leave(&pflags);
1366 * NOTE! The wait->flags weren't stable until we've done the
1367 * 'finish_wait()', and we could have exited the loop above due
1368 * to a signal, and had a wakeup event happen after the signal
1369 * test but before the 'finish_wait()'.
1371 * So only after the finish_wait() can we reliably determine
1372 * if we got woken up or not, so we can now figure out the final
1373 * return value based on that state without races.
1375 * Also note that WQ_FLAG_WOKEN is sufficient for a non-exclusive
1376 * waiter, but an exclusive one requires WQ_FLAG_DONE.
1378 if (behavior == EXCLUSIVE)
1379 return wait->flags & WQ_FLAG_DONE ? 0 : -EINTR;
1381 return wait->flags & WQ_FLAG_WOKEN ? 0 : -EINTR;
1384 #ifdef CONFIG_MIGRATION
1386 * migration_entry_wait_on_locked - Wait for a migration entry to be removed
1387 * @entry: migration swap entry.
1388 * @ptep: mapped pte pointer. Will return with the ptep unmapped. Only required
1389 * for pte entries, pass NULL for pmd entries.
1390 * @ptl: already locked ptl. This function will drop the lock.
1392 * Wait for a migration entry referencing the given page to be removed. This is
1393 * equivalent to put_and_wait_on_page_locked(page, TASK_UNINTERRUPTIBLE) except
1394 * this can be called without taking a reference on the page. Instead this
1395 * should be called while holding the ptl for the migration entry referencing
1398 * Returns after unmapping and unlocking the pte/ptl with pte_unmap_unlock().
1400 * This follows the same logic as folio_wait_bit_common() so see the comments
1403 void migration_entry_wait_on_locked(swp_entry_t entry, pte_t *ptep,
1406 struct wait_page_queue wait_page;
1407 wait_queue_entry_t *wait = &wait_page.wait;
1408 bool thrashing = false;
1409 bool delayacct = false;
1410 unsigned long pflags;
1411 wait_queue_head_t *q;
1412 struct folio *folio = page_folio(pfn_swap_entry_to_page(entry));
1414 q = folio_waitqueue(folio);
1415 if (!folio_test_uptodate(folio) && folio_test_workingset(folio)) {
1416 if (!folio_test_swapbacked(folio)) {
1417 delayacct_thrashing_start();
1420 psi_memstall_enter(&pflags);
1425 wait->func = wake_page_function;
1426 wait_page.folio = folio;
1427 wait_page.bit_nr = PG_locked;
1430 spin_lock_irq(&q->lock);
1431 folio_set_waiters(folio);
1432 if (!folio_trylock_flag(folio, PG_locked, wait))
1433 __add_wait_queue_entry_tail(q, wait);
1434 spin_unlock_irq(&q->lock);
1437 * If a migration entry exists for the page the migration path must hold
1438 * a valid reference to the page, and it must take the ptl to remove the
1439 * migration entry. So the page is valid until the ptl is dropped.
1442 pte_unmap_unlock(ptep, ptl);
1449 set_current_state(TASK_UNINTERRUPTIBLE);
1451 /* Loop until we've been woken or interrupted */
1452 flags = smp_load_acquire(&wait->flags);
1453 if (!(flags & WQ_FLAG_WOKEN)) {
1454 if (signal_pending_state(TASK_UNINTERRUPTIBLE, current))
1463 finish_wait(q, wait);
1467 delayacct_thrashing_end();
1468 psi_memstall_leave(&pflags);
1473 void folio_wait_bit(struct folio *folio, int bit_nr)
1475 folio_wait_bit_common(folio, bit_nr, TASK_UNINTERRUPTIBLE, SHARED);
1477 EXPORT_SYMBOL(folio_wait_bit);
1479 int folio_wait_bit_killable(struct folio *folio, int bit_nr)
1481 return folio_wait_bit_common(folio, bit_nr, TASK_KILLABLE, SHARED);
1483 EXPORT_SYMBOL(folio_wait_bit_killable);
1486 * folio_put_wait_locked - Drop a reference and wait for it to be unlocked
1487 * @folio: The folio to wait for.
1488 * @state: The sleep state (TASK_KILLABLE, TASK_UNINTERRUPTIBLE, etc).
1490 * The caller should hold a reference on @folio. They expect the page to
1491 * become unlocked relatively soon, but do not wish to hold up migration
1492 * (for example) by holding the reference while waiting for the folio to
1493 * come unlocked. After this function returns, the caller should not
1494 * dereference @folio.
1496 * Return: 0 if the folio was unlocked or -EINTR if interrupted by a signal.
1498 int folio_put_wait_locked(struct folio *folio, int state)
1500 return folio_wait_bit_common(folio, PG_locked, state, DROP);
1504 * folio_add_wait_queue - Add an arbitrary waiter to a folio's wait queue
1505 * @folio: Folio defining the wait queue of interest
1506 * @waiter: Waiter to add to the queue
1508 * Add an arbitrary @waiter to the wait queue for the nominated @folio.
1510 void folio_add_wait_queue(struct folio *folio, wait_queue_entry_t *waiter)
1512 wait_queue_head_t *q = folio_waitqueue(folio);
1513 unsigned long flags;
1515 spin_lock_irqsave(&q->lock, flags);
1516 __add_wait_queue_entry_tail(q, waiter);
1517 folio_set_waiters(folio);
1518 spin_unlock_irqrestore(&q->lock, flags);
1520 EXPORT_SYMBOL_GPL(folio_add_wait_queue);
1522 #ifndef clear_bit_unlock_is_negative_byte
1525 * PG_waiters is the high bit in the same byte as PG_lock.
1527 * On x86 (and on many other architectures), we can clear PG_lock and
1528 * test the sign bit at the same time. But if the architecture does
1529 * not support that special operation, we just do this all by hand
1532 * The read of PG_waiters has to be after (or concurrently with) PG_locked
1533 * being cleared, but a memory barrier should be unnecessary since it is
1534 * in the same byte as PG_locked.
1536 static inline bool clear_bit_unlock_is_negative_byte(long nr, volatile void *mem)
1538 clear_bit_unlock(nr, mem);
1539 /* smp_mb__after_atomic(); */
1540 return test_bit(PG_waiters, mem);
1546 * folio_unlock - Unlock a locked folio.
1547 * @folio: The folio.
1549 * Unlocks the folio and wakes up any thread sleeping on the page lock.
1551 * Context: May be called from interrupt or process context. May not be
1552 * called from NMI context.
1554 void folio_unlock(struct folio *folio)
1556 /* Bit 7 allows x86 to check the byte's sign bit */
1557 BUILD_BUG_ON(PG_waiters != 7);
1558 BUILD_BUG_ON(PG_locked > 7);
1559 VM_BUG_ON_FOLIO(!folio_test_locked(folio), folio);
1560 if (clear_bit_unlock_is_negative_byte(PG_locked, folio_flags(folio, 0)))
1561 folio_wake_bit(folio, PG_locked);
1563 EXPORT_SYMBOL(folio_unlock);
1566 * folio_end_private_2 - Clear PG_private_2 and wake any waiters.
1567 * @folio: The folio.
1569 * Clear the PG_private_2 bit on a folio and wake up any sleepers waiting for
1570 * it. The folio reference held for PG_private_2 being set is released.
1572 * This is, for example, used when a netfs folio is being written to a local
1573 * disk cache, thereby allowing writes to the cache for the same folio to be
1576 void folio_end_private_2(struct folio *folio)
1578 VM_BUG_ON_FOLIO(!folio_test_private_2(folio), folio);
1579 clear_bit_unlock(PG_private_2, folio_flags(folio, 0));
1580 folio_wake_bit(folio, PG_private_2);
1583 EXPORT_SYMBOL(folio_end_private_2);
1586 * folio_wait_private_2 - Wait for PG_private_2 to be cleared on a folio.
1587 * @folio: The folio to wait on.
1589 * Wait for PG_private_2 (aka PG_fscache) to be cleared on a folio.
1591 void folio_wait_private_2(struct folio *folio)
1593 while (folio_test_private_2(folio))
1594 folio_wait_bit(folio, PG_private_2);
1596 EXPORT_SYMBOL(folio_wait_private_2);
1599 * folio_wait_private_2_killable - Wait for PG_private_2 to be cleared on a folio.
1600 * @folio: The folio to wait on.
1602 * Wait for PG_private_2 (aka PG_fscache) to be cleared on a folio or until a
1603 * fatal signal is received by the calling task.
1606 * - 0 if successful.
1607 * - -EINTR if a fatal signal was encountered.
1609 int folio_wait_private_2_killable(struct folio *folio)
1613 while (folio_test_private_2(folio)) {
1614 ret = folio_wait_bit_killable(folio, PG_private_2);
1621 EXPORT_SYMBOL(folio_wait_private_2_killable);
1624 * folio_end_writeback - End writeback against a folio.
1625 * @folio: The folio.
1627 void folio_end_writeback(struct folio *folio)
1630 * folio_test_clear_reclaim() could be used here but it is an
1631 * atomic operation and overkill in this particular case. Failing
1632 * to shuffle a folio marked for immediate reclaim is too mild
1633 * a gain to justify taking an atomic operation penalty at the
1634 * end of every folio writeback.
1636 if (folio_test_reclaim(folio)) {
1637 folio_clear_reclaim(folio);
1638 folio_rotate_reclaimable(folio);
1642 * Writeback does not hold a folio reference of its own, relying
1643 * on truncation to wait for the clearing of PG_writeback.
1644 * But here we must make sure that the folio is not freed and
1645 * reused before the folio_wake().
1648 if (!__folio_end_writeback(folio))
1651 smp_mb__after_atomic();
1652 folio_wake(folio, PG_writeback);
1653 acct_reclaim_writeback(folio);
1656 EXPORT_SYMBOL(folio_end_writeback);
1659 * After completing I/O on a page, call this routine to update the page
1660 * flags appropriately
1662 void page_endio(struct page *page, bool is_write, int err)
1666 SetPageUptodate(page);
1668 ClearPageUptodate(page);
1674 struct address_space *mapping;
1677 mapping = page_mapping(page);
1679 mapping_set_error(mapping, err);
1681 end_page_writeback(page);
1684 EXPORT_SYMBOL_GPL(page_endio);
1687 * __folio_lock - Get a lock on the folio, assuming we need to sleep to get it.
1688 * @folio: The folio to lock
1690 void __folio_lock(struct folio *folio)
1692 folio_wait_bit_common(folio, PG_locked, TASK_UNINTERRUPTIBLE,
1695 EXPORT_SYMBOL(__folio_lock);
1697 int __folio_lock_killable(struct folio *folio)
1699 return folio_wait_bit_common(folio, PG_locked, TASK_KILLABLE,
1702 EXPORT_SYMBOL_GPL(__folio_lock_killable);
1704 static int __folio_lock_async(struct folio *folio, struct wait_page_queue *wait)
1706 struct wait_queue_head *q = folio_waitqueue(folio);
1709 wait->folio = folio;
1710 wait->bit_nr = PG_locked;
1712 spin_lock_irq(&q->lock);
1713 __add_wait_queue_entry_tail(q, &wait->wait);
1714 folio_set_waiters(folio);
1715 ret = !folio_trylock(folio);
1717 * If we were successful now, we know we're still on the
1718 * waitqueue as we're still under the lock. This means it's
1719 * safe to remove and return success, we know the callback
1720 * isn't going to trigger.
1723 __remove_wait_queue(q, &wait->wait);
1726 spin_unlock_irq(&q->lock);
1732 * true - folio is locked; mmap_lock is still held.
1733 * false - folio is not locked.
1734 * mmap_lock has been released (mmap_read_unlock(), unless flags had both
1735 * FAULT_FLAG_ALLOW_RETRY and FAULT_FLAG_RETRY_NOWAIT set, in
1736 * which case mmap_lock is still held.
1738 * If neither ALLOW_RETRY nor KILLABLE are set, will always return true
1739 * with the folio locked and the mmap_lock unperturbed.
1741 bool __folio_lock_or_retry(struct folio *folio, struct mm_struct *mm,
1744 if (fault_flag_allow_retry_first(flags)) {
1746 * CAUTION! In this case, mmap_lock is not released
1747 * even though return 0.
1749 if (flags & FAULT_FLAG_RETRY_NOWAIT)
1752 mmap_read_unlock(mm);
1753 if (flags & FAULT_FLAG_KILLABLE)
1754 folio_wait_locked_killable(folio);
1756 folio_wait_locked(folio);
1759 if (flags & FAULT_FLAG_KILLABLE) {
1762 ret = __folio_lock_killable(folio);
1764 mmap_read_unlock(mm);
1768 __folio_lock(folio);
1775 * page_cache_next_miss() - Find the next gap in the page cache.
1776 * @mapping: Mapping.
1778 * @max_scan: Maximum range to search.
1780 * Search the range [index, min(index + max_scan - 1, ULONG_MAX)] for the
1781 * gap with the lowest index.
1783 * This function may be called under the rcu_read_lock. However, this will
1784 * not atomically search a snapshot of the cache at a single point in time.
1785 * For example, if a gap is created at index 5, then subsequently a gap is
1786 * created at index 10, page_cache_next_miss covering both indices may
1787 * return 10 if called under the rcu_read_lock.
1789 * Return: The index of the gap if found, otherwise an index outside the
1790 * range specified (in which case 'return - index >= max_scan' will be true).
1791 * In the rare case of index wrap-around, 0 will be returned.
1793 pgoff_t page_cache_next_miss(struct address_space *mapping,
1794 pgoff_t index, unsigned long max_scan)
1796 XA_STATE(xas, &mapping->i_pages, index);
1798 while (max_scan--) {
1799 void *entry = xas_next(&xas);
1800 if (!entry || xa_is_value(entry))
1802 if (xas.xa_index == 0)
1806 return xas.xa_index;
1808 EXPORT_SYMBOL(page_cache_next_miss);
1811 * page_cache_prev_miss() - Find the previous gap in the page cache.
1812 * @mapping: Mapping.
1814 * @max_scan: Maximum range to search.
1816 * Search the range [max(index - max_scan + 1, 0), index] for the
1817 * gap with the highest index.
1819 * This function may be called under the rcu_read_lock. However, this will
1820 * not atomically search a snapshot of the cache at a single point in time.
1821 * For example, if a gap is created at index 10, then subsequently a gap is
1822 * created at index 5, page_cache_prev_miss() covering both indices may
1823 * return 5 if called under the rcu_read_lock.
1825 * Return: The index of the gap if found, otherwise an index outside the
1826 * range specified (in which case 'index - return >= max_scan' will be true).
1827 * In the rare case of wrap-around, ULONG_MAX will be returned.
1829 pgoff_t page_cache_prev_miss(struct address_space *mapping,
1830 pgoff_t index, unsigned long max_scan)
1832 XA_STATE(xas, &mapping->i_pages, index);
1834 while (max_scan--) {
1835 void *entry = xas_prev(&xas);
1836 if (!entry || xa_is_value(entry))
1838 if (xas.xa_index == ULONG_MAX)
1842 return xas.xa_index;
1844 EXPORT_SYMBOL(page_cache_prev_miss);
1847 * Lockless page cache protocol:
1848 * On the lookup side:
1849 * 1. Load the folio from i_pages
1850 * 2. Increment the refcount if it's not zero
1851 * 3. If the folio is not found by xas_reload(), put the refcount and retry
1853 * On the removal side:
1854 * A. Freeze the page (by zeroing the refcount if nobody else has a reference)
1855 * B. Remove the page from i_pages
1856 * C. Return the page to the page allocator
1858 * This means that any page may have its reference count temporarily
1859 * increased by a speculative page cache (or fast GUP) lookup as it can
1860 * be allocated by another user before the RCU grace period expires.
1861 * Because the refcount temporarily acquired here may end up being the
1862 * last refcount on the page, any page allocation must be freeable by
1867 * mapping_get_entry - Get a page cache entry.
1868 * @mapping: the address_space to search
1869 * @index: The page cache index.
1871 * Looks up the page cache entry at @mapping & @index. If it is a folio,
1872 * it is returned with an increased refcount. If it is a shadow entry
1873 * of a previously evicted folio, or a swap entry from shmem/tmpfs,
1874 * it is returned without further action.
1876 * Return: The folio, swap or shadow entry, %NULL if nothing is found.
1878 static void *mapping_get_entry(struct address_space *mapping, pgoff_t index)
1880 XA_STATE(xas, &mapping->i_pages, index);
1881 struct folio *folio;
1886 folio = xas_load(&xas);
1887 if (xas_retry(&xas, folio))
1890 * A shadow entry of a recently evicted page, or a swap entry from
1891 * shmem/tmpfs. Return it without attempting to raise page count.
1893 if (!folio || xa_is_value(folio))
1896 if (!folio_try_get_rcu(folio))
1899 if (unlikely(folio != xas_reload(&xas))) {
1910 * __filemap_get_folio - Find and get a reference to a folio.
1911 * @mapping: The address_space to search.
1912 * @index: The page index.
1913 * @fgp_flags: %FGP flags modify how the folio is returned.
1914 * @gfp: Memory allocation flags to use if %FGP_CREAT is specified.
1916 * Looks up the page cache entry at @mapping & @index.
1918 * @fgp_flags can be zero or more of these flags:
1920 * * %FGP_ACCESSED - The folio will be marked accessed.
1921 * * %FGP_LOCK - The folio is returned locked.
1922 * * %FGP_ENTRY - If there is a shadow / swap / DAX entry, return it
1923 * instead of allocating a new folio to replace it.
1924 * * %FGP_CREAT - If no page is present then a new page is allocated using
1925 * @gfp and added to the page cache and the VM's LRU list.
1926 * The page is returned locked and with an increased refcount.
1927 * * %FGP_FOR_MMAP - The caller wants to do its own locking dance if the
1928 * page is already in cache. If the page was allocated, unlock it before
1929 * returning so the caller can do the same dance.
1930 * * %FGP_WRITE - The page will be written to by the caller.
1931 * * %FGP_NOFS - __GFP_FS will get cleared in gfp.
1932 * * %FGP_NOWAIT - Don't get blocked by page lock.
1933 * * %FGP_STABLE - Wait for the folio to be stable (finished writeback)
1935 * If %FGP_LOCK or %FGP_CREAT are specified then the function may sleep even
1936 * if the %GFP flags specified for %FGP_CREAT are atomic.
1938 * If there is a page cache page, it is returned with an increased refcount.
1940 * Return: The found folio or %NULL otherwise.
1942 struct folio *__filemap_get_folio(struct address_space *mapping, pgoff_t index,
1943 int fgp_flags, gfp_t gfp)
1945 struct folio *folio;
1948 folio = mapping_get_entry(mapping, index);
1949 if (xa_is_value(folio)) {
1950 if (fgp_flags & FGP_ENTRY)
1957 if (fgp_flags & FGP_LOCK) {
1958 if (fgp_flags & FGP_NOWAIT) {
1959 if (!folio_trylock(folio)) {
1967 /* Has the page been truncated? */
1968 if (unlikely(folio->mapping != mapping)) {
1969 folio_unlock(folio);
1973 VM_BUG_ON_FOLIO(!folio_contains(folio, index), folio);
1976 if (fgp_flags & FGP_ACCESSED)
1977 folio_mark_accessed(folio);
1978 else if (fgp_flags & FGP_WRITE) {
1979 /* Clear idle flag for buffer write */
1980 if (folio_test_idle(folio))
1981 folio_clear_idle(folio);
1984 if (fgp_flags & FGP_STABLE)
1985 folio_wait_stable(folio);
1987 if (!folio && (fgp_flags & FGP_CREAT)) {
1989 if ((fgp_flags & FGP_WRITE) && mapping_can_writeback(mapping))
1991 if (fgp_flags & FGP_NOFS)
1994 folio = filemap_alloc_folio(gfp, 0);
1998 if (WARN_ON_ONCE(!(fgp_flags & (FGP_LOCK | FGP_FOR_MMAP))))
1999 fgp_flags |= FGP_LOCK;
2001 /* Init accessed so avoid atomic mark_page_accessed later */
2002 if (fgp_flags & FGP_ACCESSED)
2003 __folio_set_referenced(folio);
2005 err = filemap_add_folio(mapping, folio, index, gfp);
2006 if (unlikely(err)) {
2014 * filemap_add_folio locks the page, and for mmap
2015 * we expect an unlocked page.
2017 if (folio && (fgp_flags & FGP_FOR_MMAP))
2018 folio_unlock(folio);
2023 EXPORT_SYMBOL(__filemap_get_folio);
2025 static inline struct folio *find_get_entry(struct xa_state *xas, pgoff_t max,
2028 struct folio *folio;
2031 if (mark == XA_PRESENT)
2032 folio = xas_find(xas, max);
2034 folio = xas_find_marked(xas, max, mark);
2036 if (xas_retry(xas, folio))
2039 * A shadow entry of a recently evicted page, a swap
2040 * entry from shmem/tmpfs or a DAX entry. Return it
2041 * without attempting to raise page count.
2043 if (!folio || xa_is_value(folio))
2046 if (!folio_try_get_rcu(folio))
2049 if (unlikely(folio != xas_reload(xas))) {
2061 * find_get_entries - gang pagecache lookup
2062 * @mapping: The address_space to search
2063 * @start: The starting page cache index
2064 * @end: The final page index (inclusive).
2065 * @fbatch: Where the resulting entries are placed.
2066 * @indices: The cache indices corresponding to the entries in @entries
2068 * find_get_entries() will search for and return a batch of entries in
2069 * the mapping. The entries are placed in @fbatch. find_get_entries()
2070 * takes a reference on any actual folios it returns.
2072 * The entries have ascending indexes. The indices may not be consecutive
2073 * due to not-present entries or large folios.
2075 * Any shadow entries of evicted folios, or swap entries from
2076 * shmem/tmpfs, are included in the returned array.
2078 * Return: The number of entries which were found.
2080 unsigned find_get_entries(struct address_space *mapping, pgoff_t start,
2081 pgoff_t end, struct folio_batch *fbatch, pgoff_t *indices)
2083 XA_STATE(xas, &mapping->i_pages, start);
2084 struct folio *folio;
2087 while ((folio = find_get_entry(&xas, end, XA_PRESENT)) != NULL) {
2088 indices[fbatch->nr] = xas.xa_index;
2089 if (!folio_batch_add(fbatch, folio))
2094 return folio_batch_count(fbatch);
2098 * find_lock_entries - Find a batch of pagecache entries.
2099 * @mapping: The address_space to search.
2100 * @start: The starting page cache index.
2101 * @end: The final page index (inclusive).
2102 * @fbatch: Where the resulting entries are placed.
2103 * @indices: The cache indices of the entries in @fbatch.
2105 * find_lock_entries() will return a batch of entries from @mapping.
2106 * Swap, shadow and DAX entries are included. Folios are returned
2107 * locked and with an incremented refcount. Folios which are locked
2108 * by somebody else or under writeback are skipped. Folios which are
2109 * partially outside the range are not returned.
2111 * The entries have ascending indexes. The indices may not be consecutive
2112 * due to not-present entries, large folios, folios which could not be
2113 * locked or folios under writeback.
2115 * Return: The number of entries which were found.
2117 unsigned find_lock_entries(struct address_space *mapping, pgoff_t start,
2118 pgoff_t end, struct folio_batch *fbatch, pgoff_t *indices)
2120 XA_STATE(xas, &mapping->i_pages, start);
2121 struct folio *folio;
2124 while ((folio = find_get_entry(&xas, end, XA_PRESENT))) {
2125 if (!xa_is_value(folio)) {
2126 if (folio->index < start)
2128 if (folio->index + folio_nr_pages(folio) - 1 > end)
2130 if (!folio_trylock(folio))
2132 if (folio->mapping != mapping ||
2133 folio_test_writeback(folio))
2135 VM_BUG_ON_FOLIO(!folio_contains(folio, xas.xa_index),
2138 indices[fbatch->nr] = xas.xa_index;
2139 if (!folio_batch_add(fbatch, folio))
2143 folio_unlock(folio);
2149 return folio_batch_count(fbatch);
2153 bool folio_more_pages(struct folio *folio, pgoff_t index, pgoff_t max)
2155 if (!folio_test_large(folio) || folio_test_hugetlb(folio))
2159 return index < folio->index + folio_nr_pages(folio) - 1;
2163 * find_get_pages_range - gang pagecache lookup
2164 * @mapping: The address_space to search
2165 * @start: The starting page index
2166 * @end: The final page index (inclusive)
2167 * @nr_pages: The maximum number of pages
2168 * @pages: Where the resulting pages are placed
2170 * find_get_pages_range() will search for and return a group of up to @nr_pages
2171 * pages in the mapping starting at index @start and up to index @end
2172 * (inclusive). The pages are placed at @pages. find_get_pages_range() takes
2173 * a reference against the returned pages.
2175 * The search returns a group of mapping-contiguous pages with ascending
2176 * indexes. There may be holes in the indices due to not-present pages.
2177 * We also update @start to index the next page for the traversal.
2179 * Return: the number of pages which were found. If this number is
2180 * smaller than @nr_pages, the end of specified range has been
2183 unsigned find_get_pages_range(struct address_space *mapping, pgoff_t *start,
2184 pgoff_t end, unsigned int nr_pages,
2185 struct page **pages)
2187 XA_STATE(xas, &mapping->i_pages, *start);
2188 struct folio *folio;
2191 if (unlikely(!nr_pages))
2195 while ((folio = find_get_entry(&xas, end, XA_PRESENT))) {
2196 /* Skip over shadow, swap and DAX entries */
2197 if (xa_is_value(folio))
2201 pages[ret] = folio_file_page(folio, xas.xa_index);
2202 if (++ret == nr_pages) {
2203 *start = xas.xa_index + 1;
2206 if (folio_more_pages(folio, xas.xa_index, end)) {
2208 folio_ref_inc(folio);
2214 * We come here when there is no page beyond @end. We take care to not
2215 * overflow the index @start as it confuses some of the callers. This
2216 * breaks the iteration when there is a page at index -1 but that is
2217 * already broken anyway.
2219 if (end == (pgoff_t)-1)
2220 *start = (pgoff_t)-1;
2230 * find_get_pages_contig - gang contiguous pagecache lookup
2231 * @mapping: The address_space to search
2232 * @index: The starting page index
2233 * @nr_pages: The maximum number of pages
2234 * @pages: Where the resulting pages are placed
2236 * find_get_pages_contig() works exactly like find_get_pages_range(),
2237 * except that the returned number of pages are guaranteed to be
2240 * Return: the number of pages which were found.
2242 unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index,
2243 unsigned int nr_pages, struct page **pages)
2245 XA_STATE(xas, &mapping->i_pages, index);
2246 struct folio *folio;
2247 unsigned int ret = 0;
2249 if (unlikely(!nr_pages))
2253 for (folio = xas_load(&xas); folio; folio = xas_next(&xas)) {
2254 if (xas_retry(&xas, folio))
2257 * If the entry has been swapped out, we can stop looking.
2258 * No current caller is looking for DAX entries.
2260 if (xa_is_value(folio))
2263 if (!folio_try_get_rcu(folio))
2266 if (unlikely(folio != xas_reload(&xas)))
2270 pages[ret] = folio_file_page(folio, xas.xa_index);
2271 if (++ret == nr_pages)
2273 if (folio_more_pages(folio, xas.xa_index, ULONG_MAX)) {
2275 folio_ref_inc(folio);
2287 EXPORT_SYMBOL(find_get_pages_contig);
2290 * find_get_pages_range_tag - Find and return head pages matching @tag.
2291 * @mapping: the address_space to search
2292 * @index: the starting page index
2293 * @end: The final page index (inclusive)
2294 * @tag: the tag index
2295 * @nr_pages: the maximum number of pages
2296 * @pages: where the resulting pages are placed
2298 * Like find_get_pages_range(), except we only return head pages which are
2299 * tagged with @tag. @index is updated to the index immediately after the
2300 * last page we return, ready for the next iteration.
2302 * Return: the number of pages which were found.
2304 unsigned find_get_pages_range_tag(struct address_space *mapping, pgoff_t *index,
2305 pgoff_t end, xa_mark_t tag, unsigned int nr_pages,
2306 struct page **pages)
2308 XA_STATE(xas, &mapping->i_pages, *index);
2309 struct folio *folio;
2312 if (unlikely(!nr_pages))
2316 while ((folio = find_get_entry(&xas, end, tag))) {
2318 * Shadow entries should never be tagged, but this iteration
2319 * is lockless so there is a window for page reclaim to evict
2320 * a page we saw tagged. Skip over it.
2322 if (xa_is_value(folio))
2325 pages[ret] = &folio->page;
2326 if (++ret == nr_pages) {
2327 *index = folio->index + folio_nr_pages(folio);
2333 * We come here when we got to @end. We take care to not overflow the
2334 * index @index as it confuses some of the callers. This breaks the
2335 * iteration when there is a page at index -1 but that is already
2338 if (end == (pgoff_t)-1)
2339 *index = (pgoff_t)-1;
2347 EXPORT_SYMBOL(find_get_pages_range_tag);
2350 * CD/DVDs are error prone. When a medium error occurs, the driver may fail
2351 * a _large_ part of the i/o request. Imagine the worst scenario:
2353 * ---R__________________________________________B__________
2354 * ^ reading here ^ bad block(assume 4k)
2356 * read(R) => miss => readahead(R...B) => media error => frustrating retries
2357 * => failing the whole request => read(R) => read(R+1) =>
2358 * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
2359 * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
2360 * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
2362 * It is going insane. Fix it by quickly scaling down the readahead size.
2364 static void shrink_readahead_size_eio(struct file_ra_state *ra)
2370 * filemap_get_read_batch - Get a batch of folios for read
2372 * Get a batch of folios which represent a contiguous range of bytes in
2373 * the file. No exceptional entries will be returned. If @index is in
2374 * the middle of a folio, the entire folio will be returned. The last
2375 * folio in the batch may have the readahead flag set or the uptodate flag
2376 * clear so that the caller can take the appropriate action.
2378 static void filemap_get_read_batch(struct address_space *mapping,
2379 pgoff_t index, pgoff_t max, struct folio_batch *fbatch)
2381 XA_STATE(xas, &mapping->i_pages, index);
2382 struct folio *folio;
2385 for (folio = xas_load(&xas); folio; folio = xas_next(&xas)) {
2386 if (xas_retry(&xas, folio))
2388 if (xas.xa_index > max || xa_is_value(folio))
2390 if (!folio_try_get_rcu(folio))
2393 if (unlikely(folio != xas_reload(&xas)))
2396 if (!folio_batch_add(fbatch, folio))
2398 if (!folio_test_uptodate(folio))
2400 if (folio_test_readahead(folio))
2402 xas_advance(&xas, folio->index + folio_nr_pages(folio) - 1);
2412 static int filemap_read_folio(struct file *file, struct address_space *mapping,
2413 struct folio *folio)
2418 * A previous I/O error may have been due to temporary failures,
2419 * eg. multipath errors. PG_error will be set again if readpage
2422 folio_clear_error(folio);
2423 /* Start the actual read. The read will unlock the page. */
2424 error = mapping->a_ops->readpage(file, &folio->page);
2428 error = folio_wait_locked_killable(folio);
2431 if (folio_test_uptodate(folio))
2433 shrink_readahead_size_eio(&file->f_ra);
2437 static bool filemap_range_uptodate(struct address_space *mapping,
2438 loff_t pos, struct iov_iter *iter, struct folio *folio)
2442 if (folio_test_uptodate(folio))
2444 /* pipes can't handle partially uptodate pages */
2445 if (iov_iter_is_pipe(iter))
2447 if (!mapping->a_ops->is_partially_uptodate)
2449 if (mapping->host->i_blkbits >= folio_shift(folio))
2452 count = iter->count;
2453 if (folio_pos(folio) > pos) {
2454 count -= folio_pos(folio) - pos;
2457 pos -= folio_pos(folio);
2460 return mapping->a_ops->is_partially_uptodate(folio, pos, count);
2463 static int filemap_update_page(struct kiocb *iocb,
2464 struct address_space *mapping, struct iov_iter *iter,
2465 struct folio *folio)
2469 if (iocb->ki_flags & IOCB_NOWAIT) {
2470 if (!filemap_invalidate_trylock_shared(mapping))
2473 filemap_invalidate_lock_shared(mapping);
2476 if (!folio_trylock(folio)) {
2478 if (iocb->ki_flags & (IOCB_NOWAIT | IOCB_NOIO))
2479 goto unlock_mapping;
2480 if (!(iocb->ki_flags & IOCB_WAITQ)) {
2481 filemap_invalidate_unlock_shared(mapping);
2483 * This is where we usually end up waiting for a
2484 * previously submitted readahead to finish.
2486 folio_put_wait_locked(folio, TASK_KILLABLE);
2487 return AOP_TRUNCATED_PAGE;
2489 error = __folio_lock_async(folio, iocb->ki_waitq);
2491 goto unlock_mapping;
2494 error = AOP_TRUNCATED_PAGE;
2495 if (!folio->mapping)
2499 if (filemap_range_uptodate(mapping, iocb->ki_pos, iter, folio))
2503 if (iocb->ki_flags & (IOCB_NOIO | IOCB_NOWAIT | IOCB_WAITQ))
2506 error = filemap_read_folio(iocb->ki_filp, mapping, folio);
2507 goto unlock_mapping;
2509 folio_unlock(folio);
2511 filemap_invalidate_unlock_shared(mapping);
2512 if (error == AOP_TRUNCATED_PAGE)
2517 static int filemap_create_folio(struct file *file,
2518 struct address_space *mapping, pgoff_t index,
2519 struct folio_batch *fbatch)
2521 struct folio *folio;
2524 folio = filemap_alloc_folio(mapping_gfp_mask(mapping), 0);
2529 * Protect against truncate / hole punch. Grabbing invalidate_lock
2530 * here assures we cannot instantiate and bring uptodate new
2531 * pagecache folios after evicting page cache during truncate
2532 * and before actually freeing blocks. Note that we could
2533 * release invalidate_lock after inserting the folio into
2534 * the page cache as the locked folio would then be enough to
2535 * synchronize with hole punching. But there are code paths
2536 * such as filemap_update_page() filling in partially uptodate
2537 * pages or ->readpages() that need to hold invalidate_lock
2538 * while mapping blocks for IO so let's hold the lock here as
2539 * well to keep locking rules simple.
2541 filemap_invalidate_lock_shared(mapping);
2542 error = filemap_add_folio(mapping, folio, index,
2543 mapping_gfp_constraint(mapping, GFP_KERNEL));
2544 if (error == -EEXIST)
2545 error = AOP_TRUNCATED_PAGE;
2549 error = filemap_read_folio(file, mapping, folio);
2553 filemap_invalidate_unlock_shared(mapping);
2554 folio_batch_add(fbatch, folio);
2557 filemap_invalidate_unlock_shared(mapping);
2562 static int filemap_readahead(struct kiocb *iocb, struct file *file,
2563 struct address_space *mapping, struct folio *folio,
2566 DEFINE_READAHEAD(ractl, file, &file->f_ra, mapping, folio->index);
2568 if (iocb->ki_flags & IOCB_NOIO)
2570 page_cache_async_ra(&ractl, folio, last_index - folio->index);
2574 static int filemap_get_pages(struct kiocb *iocb, struct iov_iter *iter,
2575 struct folio_batch *fbatch)
2577 struct file *filp = iocb->ki_filp;
2578 struct address_space *mapping = filp->f_mapping;
2579 struct file_ra_state *ra = &filp->f_ra;
2580 pgoff_t index = iocb->ki_pos >> PAGE_SHIFT;
2582 struct folio *folio;
2585 last_index = DIV_ROUND_UP(iocb->ki_pos + iter->count, PAGE_SIZE);
2587 if (fatal_signal_pending(current))
2590 filemap_get_read_batch(mapping, index, last_index, fbatch);
2591 if (!folio_batch_count(fbatch)) {
2592 if (iocb->ki_flags & IOCB_NOIO)
2594 page_cache_sync_readahead(mapping, ra, filp, index,
2595 last_index - index);
2596 filemap_get_read_batch(mapping, index, last_index, fbatch);
2598 if (!folio_batch_count(fbatch)) {
2599 if (iocb->ki_flags & (IOCB_NOWAIT | IOCB_WAITQ))
2601 err = filemap_create_folio(filp, mapping,
2602 iocb->ki_pos >> PAGE_SHIFT, fbatch);
2603 if (err == AOP_TRUNCATED_PAGE)
2608 folio = fbatch->folios[folio_batch_count(fbatch) - 1];
2609 if (folio_test_readahead(folio)) {
2610 err = filemap_readahead(iocb, filp, mapping, folio, last_index);
2614 if (!folio_test_uptodate(folio)) {
2615 if ((iocb->ki_flags & IOCB_WAITQ) &&
2616 folio_batch_count(fbatch) > 1)
2617 iocb->ki_flags |= IOCB_NOWAIT;
2618 err = filemap_update_page(iocb, mapping, iter, folio);
2627 if (likely(--fbatch->nr))
2629 if (err == AOP_TRUNCATED_PAGE)
2635 * filemap_read - Read data from the page cache.
2636 * @iocb: The iocb to read.
2637 * @iter: Destination for the data.
2638 * @already_read: Number of bytes already read by the caller.
2640 * Copies data from the page cache. If the data is not currently present,
2641 * uses the readahead and readpage address_space operations to fetch it.
2643 * Return: Total number of bytes copied, including those already read by
2644 * the caller. If an error happens before any bytes are copied, returns
2645 * a negative error number.
2647 ssize_t filemap_read(struct kiocb *iocb, struct iov_iter *iter,
2648 ssize_t already_read)
2650 struct file *filp = iocb->ki_filp;
2651 struct file_ra_state *ra = &filp->f_ra;
2652 struct address_space *mapping = filp->f_mapping;
2653 struct inode *inode = mapping->host;
2654 struct folio_batch fbatch;
2656 bool writably_mapped;
2657 loff_t isize, end_offset;
2659 if (unlikely(iocb->ki_pos >= inode->i_sb->s_maxbytes))
2661 if (unlikely(!iov_iter_count(iter)))
2664 iov_iter_truncate(iter, inode->i_sb->s_maxbytes);
2665 folio_batch_init(&fbatch);
2671 * If we've already successfully copied some data, then we
2672 * can no longer safely return -EIOCBQUEUED. Hence mark
2673 * an async read NOWAIT at that point.
2675 if ((iocb->ki_flags & IOCB_WAITQ) && already_read)
2676 iocb->ki_flags |= IOCB_NOWAIT;
2678 if (unlikely(iocb->ki_pos >= i_size_read(inode)))
2681 error = filemap_get_pages(iocb, iter, &fbatch);
2686 * i_size must be checked after we know the pages are Uptodate.
2688 * Checking i_size after the check allows us to calculate
2689 * the correct value for "nr", which means the zero-filled
2690 * part of the page is not copied back to userspace (unless
2691 * another truncate extends the file - this is desired though).
2693 isize = i_size_read(inode);
2694 if (unlikely(iocb->ki_pos >= isize))
2696 end_offset = min_t(loff_t, isize, iocb->ki_pos + iter->count);
2699 * Once we start copying data, we don't want to be touching any
2700 * cachelines that might be contended:
2702 writably_mapped = mapping_writably_mapped(mapping);
2705 * When a sequential read accesses a page several times, only
2706 * mark it as accessed the first time.
2708 if (iocb->ki_pos >> PAGE_SHIFT !=
2709 ra->prev_pos >> PAGE_SHIFT)
2710 folio_mark_accessed(fbatch.folios[0]);
2712 for (i = 0; i < folio_batch_count(&fbatch); i++) {
2713 struct folio *folio = fbatch.folios[i];
2714 size_t fsize = folio_size(folio);
2715 size_t offset = iocb->ki_pos & (fsize - 1);
2716 size_t bytes = min_t(loff_t, end_offset - iocb->ki_pos,
2720 if (end_offset < folio_pos(folio))
2723 folio_mark_accessed(folio);
2725 * If users can be writing to this folio using arbitrary
2726 * virtual addresses, take care of potential aliasing
2727 * before reading the folio on the kernel side.
2729 if (writably_mapped)
2730 flush_dcache_folio(folio);
2732 copied = copy_folio_to_iter(folio, offset, bytes, iter);
2734 already_read += copied;
2735 iocb->ki_pos += copied;
2736 ra->prev_pos = iocb->ki_pos;
2738 if (copied < bytes) {
2744 for (i = 0; i < folio_batch_count(&fbatch); i++)
2745 folio_put(fbatch.folios[i]);
2746 folio_batch_init(&fbatch);
2747 } while (iov_iter_count(iter) && iocb->ki_pos < isize && !error);
2749 file_accessed(filp);
2751 return already_read ? already_read : error;
2753 EXPORT_SYMBOL_GPL(filemap_read);
2756 * generic_file_read_iter - generic filesystem read routine
2757 * @iocb: kernel I/O control block
2758 * @iter: destination for the data read
2760 * This is the "read_iter()" routine for all filesystems
2761 * that can use the page cache directly.
2763 * The IOCB_NOWAIT flag in iocb->ki_flags indicates that -EAGAIN shall
2764 * be returned when no data can be read without waiting for I/O requests
2765 * to complete; it doesn't prevent readahead.
2767 * The IOCB_NOIO flag in iocb->ki_flags indicates that no new I/O
2768 * requests shall be made for the read or for readahead. When no data
2769 * can be read, -EAGAIN shall be returned. When readahead would be
2770 * triggered, a partial, possibly empty read shall be returned.
2773 * * number of bytes copied, even for partial reads
2774 * * negative error code (or 0 if IOCB_NOIO) if nothing was read
2777 generic_file_read_iter(struct kiocb *iocb, struct iov_iter *iter)
2779 size_t count = iov_iter_count(iter);
2783 return 0; /* skip atime */
2785 if (iocb->ki_flags & IOCB_DIRECT) {
2786 struct file *file = iocb->ki_filp;
2787 struct address_space *mapping = file->f_mapping;
2788 struct inode *inode = mapping->host;
2790 if (iocb->ki_flags & IOCB_NOWAIT) {
2791 if (filemap_range_needs_writeback(mapping, iocb->ki_pos,
2792 iocb->ki_pos + count - 1))
2795 retval = filemap_write_and_wait_range(mapping,
2797 iocb->ki_pos + count - 1);
2802 file_accessed(file);
2804 retval = mapping->a_ops->direct_IO(iocb, iter);
2806 iocb->ki_pos += retval;
2809 if (retval != -EIOCBQUEUED)
2810 iov_iter_revert(iter, count - iov_iter_count(iter));
2813 * Btrfs can have a short DIO read if we encounter
2814 * compressed extents, so if there was an error, or if
2815 * we've already read everything we wanted to, or if
2816 * there was a short read because we hit EOF, go ahead
2817 * and return. Otherwise fallthrough to buffered io for
2818 * the rest of the read. Buffered reads will not work for
2819 * DAX files, so don't bother trying.
2821 if (retval < 0 || !count || IS_DAX(inode))
2823 if (iocb->ki_pos >= i_size_read(inode))
2827 return filemap_read(iocb, iter, retval);
2829 EXPORT_SYMBOL(generic_file_read_iter);
2831 static inline loff_t folio_seek_hole_data(struct xa_state *xas,
2832 struct address_space *mapping, struct folio *folio,
2833 loff_t start, loff_t end, bool seek_data)
2835 const struct address_space_operations *ops = mapping->a_ops;
2836 size_t offset, bsz = i_blocksize(mapping->host);
2838 if (xa_is_value(folio) || folio_test_uptodate(folio))
2839 return seek_data ? start : end;
2840 if (!ops->is_partially_uptodate)
2841 return seek_data ? end : start;
2846 if (unlikely(folio->mapping != mapping))
2849 offset = offset_in_folio(folio, start) & ~(bsz - 1);
2852 if (ops->is_partially_uptodate(folio, offset, bsz) ==
2855 start = (start + bsz) & ~(bsz - 1);
2857 } while (offset < folio_size(folio));
2859 folio_unlock(folio);
2864 static inline size_t seek_folio_size(struct xa_state *xas, struct folio *folio)
2866 if (xa_is_value(folio))
2867 return PAGE_SIZE << xa_get_order(xas->xa, xas->xa_index);
2868 return folio_size(folio);
2872 * mapping_seek_hole_data - Seek for SEEK_DATA / SEEK_HOLE in the page cache.
2873 * @mapping: Address space to search.
2874 * @start: First byte to consider.
2875 * @end: Limit of search (exclusive).
2876 * @whence: Either SEEK_HOLE or SEEK_DATA.
2878 * If the page cache knows which blocks contain holes and which blocks
2879 * contain data, your filesystem can use this function to implement
2880 * SEEK_HOLE and SEEK_DATA. This is useful for filesystems which are
2881 * entirely memory-based such as tmpfs, and filesystems which support
2882 * unwritten extents.
2884 * Return: The requested offset on success, or -ENXIO if @whence specifies
2885 * SEEK_DATA and there is no data after @start. There is an implicit hole
2886 * after @end - 1, so SEEK_HOLE returns @end if all the bytes between @start
2887 * and @end contain data.
2889 loff_t mapping_seek_hole_data(struct address_space *mapping, loff_t start,
2890 loff_t end, int whence)
2892 XA_STATE(xas, &mapping->i_pages, start >> PAGE_SHIFT);
2893 pgoff_t max = (end - 1) >> PAGE_SHIFT;
2894 bool seek_data = (whence == SEEK_DATA);
2895 struct folio *folio;
2901 while ((folio = find_get_entry(&xas, max, XA_PRESENT))) {
2902 loff_t pos = (u64)xas.xa_index << PAGE_SHIFT;
2911 seek_size = seek_folio_size(&xas, folio);
2912 pos = round_up((u64)pos + 1, seek_size);
2913 start = folio_seek_hole_data(&xas, mapping, folio, start, pos,
2919 if (seek_size > PAGE_SIZE)
2920 xas_set(&xas, pos >> PAGE_SHIFT);
2921 if (!xa_is_value(folio))
2928 if (folio && !xa_is_value(folio))
2936 #define MMAP_LOTSAMISS (100)
2938 * lock_folio_maybe_drop_mmap - lock the page, possibly dropping the mmap_lock
2939 * @vmf - the vm_fault for this fault.
2940 * @folio - the folio to lock.
2941 * @fpin - the pointer to the file we may pin (or is already pinned).
2943 * This works similar to lock_folio_or_retry in that it can drop the
2944 * mmap_lock. It differs in that it actually returns the folio locked
2945 * if it returns 1 and 0 if it couldn't lock the folio. If we did have
2946 * to drop the mmap_lock then fpin will point to the pinned file and
2947 * needs to be fput()'ed at a later point.
2949 static int lock_folio_maybe_drop_mmap(struct vm_fault *vmf, struct folio *folio,
2952 if (folio_trylock(folio))
2956 * NOTE! This will make us return with VM_FAULT_RETRY, but with
2957 * the mmap_lock still held. That's how FAULT_FLAG_RETRY_NOWAIT
2958 * is supposed to work. We have way too many special cases..
2960 if (vmf->flags & FAULT_FLAG_RETRY_NOWAIT)
2963 *fpin = maybe_unlock_mmap_for_io(vmf, *fpin);
2964 if (vmf->flags & FAULT_FLAG_KILLABLE) {
2965 if (__folio_lock_killable(folio)) {
2967 * We didn't have the right flags to drop the mmap_lock,
2968 * but all fault_handlers only check for fatal signals
2969 * if we return VM_FAULT_RETRY, so we need to drop the
2970 * mmap_lock here and return 0 if we don't have a fpin.
2973 mmap_read_unlock(vmf->vma->vm_mm);
2977 __folio_lock(folio);
2983 * Synchronous readahead happens when we don't even find a page in the page
2984 * cache at all. We don't want to perform IO under the mmap sem, so if we have
2985 * to drop the mmap sem we return the file that was pinned in order for us to do
2986 * that. If we didn't pin a file then we return NULL. The file that is
2987 * returned needs to be fput()'ed when we're done with it.
2989 static struct file *do_sync_mmap_readahead(struct vm_fault *vmf)
2991 struct file *file = vmf->vma->vm_file;
2992 struct file_ra_state *ra = &file->f_ra;
2993 struct address_space *mapping = file->f_mapping;
2994 DEFINE_READAHEAD(ractl, file, ra, mapping, vmf->pgoff);
2995 struct file *fpin = NULL;
2996 unsigned int mmap_miss;
2998 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
2999 /* Use the readahead code, even if readahead is disabled */
3000 if (vmf->vma->vm_flags & VM_HUGEPAGE) {
3001 fpin = maybe_unlock_mmap_for_io(vmf, fpin);
3002 ractl._index &= ~((unsigned long)HPAGE_PMD_NR - 1);
3003 ra->size = HPAGE_PMD_NR;
3005 * Fetch two PMD folios, so we get the chance to actually
3006 * readahead, unless we've been told not to.
3008 if (!(vmf->vma->vm_flags & VM_RAND_READ))
3010 ra->async_size = HPAGE_PMD_NR;
3011 page_cache_ra_order(&ractl, ra, HPAGE_PMD_ORDER);
3016 /* If we don't want any read-ahead, don't bother */
3017 if (vmf->vma->vm_flags & VM_RAND_READ)
3022 if (vmf->vma->vm_flags & VM_SEQ_READ) {
3023 fpin = maybe_unlock_mmap_for_io(vmf, fpin);
3024 page_cache_sync_ra(&ractl, ra->ra_pages);
3028 /* Avoid banging the cache line if not needed */
3029 mmap_miss = READ_ONCE(ra->mmap_miss);
3030 if (mmap_miss < MMAP_LOTSAMISS * 10)
3031 WRITE_ONCE(ra->mmap_miss, ++mmap_miss);
3034 * Do we miss much more than hit in this file? If so,
3035 * stop bothering with read-ahead. It will only hurt.
3037 if (mmap_miss > MMAP_LOTSAMISS)
3043 fpin = maybe_unlock_mmap_for_io(vmf, fpin);
3044 ra->start = max_t(long, 0, vmf->pgoff - ra->ra_pages / 2);
3045 ra->size = ra->ra_pages;
3046 ra->async_size = ra->ra_pages / 4;
3047 ractl._index = ra->start;
3048 page_cache_ra_order(&ractl, ra, 0);
3053 * Asynchronous readahead happens when we find the page and PG_readahead,
3054 * so we want to possibly extend the readahead further. We return the file that
3055 * was pinned if we have to drop the mmap_lock in order to do IO.
3057 static struct file *do_async_mmap_readahead(struct vm_fault *vmf,
3058 struct folio *folio)
3060 struct file *file = vmf->vma->vm_file;
3061 struct file_ra_state *ra = &file->f_ra;
3062 DEFINE_READAHEAD(ractl, file, ra, file->f_mapping, vmf->pgoff);
3063 struct file *fpin = NULL;
3064 unsigned int mmap_miss;
3066 /* If we don't want any read-ahead, don't bother */
3067 if (vmf->vma->vm_flags & VM_RAND_READ || !ra->ra_pages)
3070 mmap_miss = READ_ONCE(ra->mmap_miss);
3072 WRITE_ONCE(ra->mmap_miss, --mmap_miss);
3074 if (folio_test_readahead(folio)) {
3075 fpin = maybe_unlock_mmap_for_io(vmf, fpin);
3076 page_cache_async_ra(&ractl, folio, ra->ra_pages);
3082 * filemap_fault - read in file data for page fault handling
3083 * @vmf: struct vm_fault containing details of the fault
3085 * filemap_fault() is invoked via the vma operations vector for a
3086 * mapped memory region to read in file data during a page fault.
3088 * The goto's are kind of ugly, but this streamlines the normal case of having
3089 * it in the page cache, and handles the special cases reasonably without
3090 * having a lot of duplicated code.
3092 * vma->vm_mm->mmap_lock must be held on entry.
3094 * If our return value has VM_FAULT_RETRY set, it's because the mmap_lock
3095 * may be dropped before doing I/O or by lock_folio_maybe_drop_mmap().
3097 * If our return value does not have VM_FAULT_RETRY set, the mmap_lock
3098 * has not been released.
3100 * We never return with VM_FAULT_RETRY and a bit from VM_FAULT_ERROR set.
3102 * Return: bitwise-OR of %VM_FAULT_ codes.
3104 vm_fault_t filemap_fault(struct vm_fault *vmf)
3107 struct file *file = vmf->vma->vm_file;
3108 struct file *fpin = NULL;
3109 struct address_space *mapping = file->f_mapping;
3110 struct inode *inode = mapping->host;
3111 pgoff_t max_idx, index = vmf->pgoff;
3112 struct folio *folio;
3114 bool mapping_locked = false;
3116 max_idx = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
3117 if (unlikely(index >= max_idx))
3118 return VM_FAULT_SIGBUS;
3121 * Do we have something in the page cache already?
3123 folio = filemap_get_folio(mapping, index);
3124 if (likely(folio)) {
3126 * We found the page, so try async readahead before waiting for
3129 if (!(vmf->flags & FAULT_FLAG_TRIED))
3130 fpin = do_async_mmap_readahead(vmf, folio);
3131 if (unlikely(!folio_test_uptodate(folio))) {
3132 filemap_invalidate_lock_shared(mapping);
3133 mapping_locked = true;
3136 /* No page in the page cache at all */
3137 count_vm_event(PGMAJFAULT);
3138 count_memcg_event_mm(vmf->vma->vm_mm, PGMAJFAULT);
3139 ret = VM_FAULT_MAJOR;
3140 fpin = do_sync_mmap_readahead(vmf);
3143 * See comment in filemap_create_folio() why we need
3146 if (!mapping_locked) {
3147 filemap_invalidate_lock_shared(mapping);
3148 mapping_locked = true;
3150 folio = __filemap_get_folio(mapping, index,
3151 FGP_CREAT|FGP_FOR_MMAP,
3156 filemap_invalidate_unlock_shared(mapping);
3157 return VM_FAULT_OOM;
3161 if (!lock_folio_maybe_drop_mmap(vmf, folio, &fpin))
3164 /* Did it get truncated? */
3165 if (unlikely(folio->mapping != mapping)) {
3166 folio_unlock(folio);
3170 VM_BUG_ON_FOLIO(!folio_contains(folio, index), folio);
3173 * We have a locked page in the page cache, now we need to check
3174 * that it's up-to-date. If not, it is going to be due to an error.
3176 if (unlikely(!folio_test_uptodate(folio))) {
3178 * The page was in cache and uptodate and now it is not.
3179 * Strange but possible since we didn't hold the page lock all
3180 * the time. Let's drop everything get the invalidate lock and
3183 if (!mapping_locked) {
3184 folio_unlock(folio);
3188 goto page_not_uptodate;
3192 * We've made it this far and we had to drop our mmap_lock, now is the
3193 * time to return to the upper layer and have it re-find the vma and
3197 folio_unlock(folio);
3201 filemap_invalidate_unlock_shared(mapping);
3204 * Found the page and have a reference on it.
3205 * We must recheck i_size under page lock.
3207 max_idx = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
3208 if (unlikely(index >= max_idx)) {
3209 folio_unlock(folio);
3211 return VM_FAULT_SIGBUS;
3214 vmf->page = folio_file_page(folio, index);
3215 return ret | VM_FAULT_LOCKED;
3219 * Umm, take care of errors if the page isn't up-to-date.
3220 * Try to re-read it _once_. We do this synchronously,
3221 * because there really aren't any performance issues here
3222 * and we need to check for errors.
3224 fpin = maybe_unlock_mmap_for_io(vmf, fpin);
3225 error = filemap_read_folio(file, mapping, folio);
3230 if (!error || error == AOP_TRUNCATED_PAGE)
3232 filemap_invalidate_unlock_shared(mapping);
3234 return VM_FAULT_SIGBUS;
3238 * We dropped the mmap_lock, we need to return to the fault handler to
3239 * re-find the vma and come back and find our hopefully still populated
3245 filemap_invalidate_unlock_shared(mapping);
3248 return ret | VM_FAULT_RETRY;
3250 EXPORT_SYMBOL(filemap_fault);
3252 static bool filemap_map_pmd(struct vm_fault *vmf, struct page *page)
3254 struct mm_struct *mm = vmf->vma->vm_mm;
3256 /* Huge page is mapped? No need to proceed. */
3257 if (pmd_trans_huge(*vmf->pmd)) {
3263 if (pmd_none(*vmf->pmd) && PageTransHuge(page)) {
3264 vm_fault_t ret = do_set_pmd(vmf, page);
3266 /* The page is mapped successfully, reference consumed. */
3272 if (pmd_none(*vmf->pmd))
3273 pmd_install(mm, vmf->pmd, &vmf->prealloc_pte);
3275 /* See comment in handle_pte_fault() */
3276 if (pmd_devmap_trans_unstable(vmf->pmd)) {
3285 static struct folio *next_uptodate_page(struct folio *folio,
3286 struct address_space *mapping,
3287 struct xa_state *xas, pgoff_t end_pgoff)
3289 unsigned long max_idx;
3294 if (xas_retry(xas, folio))
3296 if (xa_is_value(folio))
3298 if (folio_test_locked(folio))
3300 if (!folio_try_get_rcu(folio))
3302 /* Has the page moved or been split? */
3303 if (unlikely(folio != xas_reload(xas)))
3305 if (!folio_test_uptodate(folio) || folio_test_readahead(folio))
3307 if (!folio_trylock(folio))
3309 if (folio->mapping != mapping)
3311 if (!folio_test_uptodate(folio))
3313 max_idx = DIV_ROUND_UP(i_size_read(mapping->host), PAGE_SIZE);
3314 if (xas->xa_index >= max_idx)
3318 folio_unlock(folio);
3321 } while ((folio = xas_next_entry(xas, end_pgoff)) != NULL);
3326 static inline struct folio *first_map_page(struct address_space *mapping,
3327 struct xa_state *xas,
3330 return next_uptodate_page(xas_find(xas, end_pgoff),
3331 mapping, xas, end_pgoff);
3334 static inline struct folio *next_map_page(struct address_space *mapping,
3335 struct xa_state *xas,
3338 return next_uptodate_page(xas_next_entry(xas, end_pgoff),
3339 mapping, xas, end_pgoff);
3342 vm_fault_t filemap_map_pages(struct vm_fault *vmf,
3343 pgoff_t start_pgoff, pgoff_t end_pgoff)
3345 struct vm_area_struct *vma = vmf->vma;
3346 struct file *file = vma->vm_file;
3347 struct address_space *mapping = file->f_mapping;
3348 pgoff_t last_pgoff = start_pgoff;
3350 XA_STATE(xas, &mapping->i_pages, start_pgoff);
3351 struct folio *folio;
3353 unsigned int mmap_miss = READ_ONCE(file->f_ra.mmap_miss);
3357 folio = first_map_page(mapping, &xas, end_pgoff);
3361 if (filemap_map_pmd(vmf, &folio->page)) {
3362 ret = VM_FAULT_NOPAGE;
3366 addr = vma->vm_start + ((start_pgoff - vma->vm_pgoff) << PAGE_SHIFT);
3367 vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, addr, &vmf->ptl);
3370 page = folio_file_page(folio, xas.xa_index);
3371 if (PageHWPoison(page))
3377 addr += (xas.xa_index - last_pgoff) << PAGE_SHIFT;
3378 vmf->pte += xas.xa_index - last_pgoff;
3379 last_pgoff = xas.xa_index;
3381 if (!pte_none(*vmf->pte))
3384 /* We're about to handle the fault */
3385 if (vmf->address == addr)
3386 ret = VM_FAULT_NOPAGE;
3388 do_set_pte(vmf, page, addr);
3389 /* no need to invalidate: a not-present page won't be cached */
3390 update_mmu_cache(vma, addr, vmf->pte);
3391 if (folio_more_pages(folio, xas.xa_index, end_pgoff)) {
3393 folio_ref_inc(folio);
3396 folio_unlock(folio);
3399 if (folio_more_pages(folio, xas.xa_index, end_pgoff)) {
3403 folio_unlock(folio);
3405 } while ((folio = next_map_page(mapping, &xas, end_pgoff)) != NULL);
3406 pte_unmap_unlock(vmf->pte, vmf->ptl);
3409 WRITE_ONCE(file->f_ra.mmap_miss, mmap_miss);
3412 EXPORT_SYMBOL(filemap_map_pages);
3414 vm_fault_t filemap_page_mkwrite(struct vm_fault *vmf)
3416 struct address_space *mapping = vmf->vma->vm_file->f_mapping;
3417 struct folio *folio = page_folio(vmf->page);
3418 vm_fault_t ret = VM_FAULT_LOCKED;
3420 sb_start_pagefault(mapping->host->i_sb);
3421 file_update_time(vmf->vma->vm_file);
3423 if (folio->mapping != mapping) {
3424 folio_unlock(folio);
3425 ret = VM_FAULT_NOPAGE;
3429 * We mark the folio dirty already here so that when freeze is in
3430 * progress, we are guaranteed that writeback during freezing will
3431 * see the dirty folio and writeprotect it again.
3433 folio_mark_dirty(folio);
3434 folio_wait_stable(folio);
3436 sb_end_pagefault(mapping->host->i_sb);
3440 const struct vm_operations_struct generic_file_vm_ops = {
3441 .fault = filemap_fault,
3442 .map_pages = filemap_map_pages,
3443 .page_mkwrite = filemap_page_mkwrite,
3446 /* This is used for a general mmap of a disk file */
3448 int generic_file_mmap(struct file *file, struct vm_area_struct *vma)
3450 struct address_space *mapping = file->f_mapping;
3452 if (!mapping->a_ops->readpage)
3454 file_accessed(file);
3455 vma->vm_ops = &generic_file_vm_ops;
3460 * This is for filesystems which do not implement ->writepage.
3462 int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
3464 if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE))
3466 return generic_file_mmap(file, vma);
3469 vm_fault_t filemap_page_mkwrite(struct vm_fault *vmf)
3471 return VM_FAULT_SIGBUS;
3473 int generic_file_mmap(struct file *file, struct vm_area_struct *vma)
3477 int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
3481 #endif /* CONFIG_MMU */
3483 EXPORT_SYMBOL(filemap_page_mkwrite);
3484 EXPORT_SYMBOL(generic_file_mmap);
3485 EXPORT_SYMBOL(generic_file_readonly_mmap);
3487 static struct folio *do_read_cache_folio(struct address_space *mapping,
3488 pgoff_t index, filler_t filler, void *data, gfp_t gfp)
3490 struct folio *folio;
3493 folio = filemap_get_folio(mapping, index);
3495 folio = filemap_alloc_folio(gfp, 0);
3497 return ERR_PTR(-ENOMEM);
3498 err = filemap_add_folio(mapping, folio, index, gfp);
3499 if (unlikely(err)) {
3503 /* Presumably ENOMEM for xarray node */
3504 return ERR_PTR(err);
3509 err = filler(data, &folio->page);
3511 err = mapping->a_ops->readpage(data, &folio->page);
3515 return ERR_PTR(err);
3518 folio_wait_locked(folio);
3519 if (!folio_test_uptodate(folio)) {
3521 return ERR_PTR(-EIO);
3526 if (folio_test_uptodate(folio))
3529 if (!folio_trylock(folio)) {
3530 folio_put_wait_locked(folio, TASK_UNINTERRUPTIBLE);
3534 /* Folio was truncated from mapping */
3535 if (!folio->mapping) {
3536 folio_unlock(folio);
3541 /* Someone else locked and filled the page in a very small window */
3542 if (folio_test_uptodate(folio)) {
3543 folio_unlock(folio);
3548 * A previous I/O error may have been due to temporary
3550 * Clear page error before actual read, PG_error will be
3551 * set again if read page fails.
3553 folio_clear_error(folio);
3557 folio_mark_accessed(folio);
3562 * read_cache_folio - read into page cache, fill it if needed
3563 * @mapping: the page's address_space
3564 * @index: the page index
3565 * @filler: function to perform the read
3566 * @data: first arg to filler(data, page) function, often left as NULL
3568 * Read into the page cache. If a page already exists, and PageUptodate() is
3569 * not set, try to fill the page and wait for it to become unlocked.
3571 * If the page does not get brought uptodate, return -EIO.
3573 * The function expects mapping->invalidate_lock to be already held.
3575 * Return: up to date page on success, ERR_PTR() on failure.
3577 struct folio *read_cache_folio(struct address_space *mapping, pgoff_t index,
3578 filler_t filler, void *data)
3580 return do_read_cache_folio(mapping, index, filler, data,
3581 mapping_gfp_mask(mapping));
3583 EXPORT_SYMBOL(read_cache_folio);
3585 static struct page *do_read_cache_page(struct address_space *mapping,
3586 pgoff_t index, filler_t *filler, void *data, gfp_t gfp)
3588 struct folio *folio;
3590 folio = do_read_cache_folio(mapping, index, filler, data, gfp);
3592 return &folio->page;
3593 return folio_file_page(folio, index);
3596 struct page *read_cache_page(struct address_space *mapping,
3597 pgoff_t index, filler_t *filler, void *data)
3599 return do_read_cache_page(mapping, index, filler, data,
3600 mapping_gfp_mask(mapping));
3602 EXPORT_SYMBOL(read_cache_page);
3605 * read_cache_page_gfp - read into page cache, using specified page allocation flags.
3606 * @mapping: the page's address_space
3607 * @index: the page index
3608 * @gfp: the page allocator flags to use if allocating
3610 * This is the same as "read_mapping_page(mapping, index, NULL)", but with
3611 * any new page allocations done using the specified allocation flags.
3613 * If the page does not get brought uptodate, return -EIO.
3615 * The function expects mapping->invalidate_lock to be already held.
3617 * Return: up to date page on success, ERR_PTR() on failure.
3619 struct page *read_cache_page_gfp(struct address_space *mapping,
3623 return do_read_cache_page(mapping, index, NULL, NULL, gfp);
3625 EXPORT_SYMBOL(read_cache_page_gfp);
3627 int pagecache_write_begin(struct file *file, struct address_space *mapping,
3628 loff_t pos, unsigned len, unsigned flags,
3629 struct page **pagep, void **fsdata)
3631 const struct address_space_operations *aops = mapping->a_ops;
3633 return aops->write_begin(file, mapping, pos, len, flags,
3636 EXPORT_SYMBOL(pagecache_write_begin);
3638 int pagecache_write_end(struct file *file, struct address_space *mapping,
3639 loff_t pos, unsigned len, unsigned copied,
3640 struct page *page, void *fsdata)
3642 const struct address_space_operations *aops = mapping->a_ops;
3644 return aops->write_end(file, mapping, pos, len, copied, page, fsdata);
3646 EXPORT_SYMBOL(pagecache_write_end);
3649 * Warn about a page cache invalidation failure during a direct I/O write.
3651 void dio_warn_stale_pagecache(struct file *filp)
3653 static DEFINE_RATELIMIT_STATE(_rs, 86400 * HZ, DEFAULT_RATELIMIT_BURST);
3657 errseq_set(&filp->f_mapping->wb_err, -EIO);
3658 if (__ratelimit(&_rs)) {
3659 path = file_path(filp, pathname, sizeof(pathname));
3662 pr_crit("Page cache invalidation failure on direct I/O. Possible data corruption due to collision with buffered I/O!\n");
3663 pr_crit("File: %s PID: %d Comm: %.20s\n", path, current->pid,
3669 generic_file_direct_write(struct kiocb *iocb, struct iov_iter *from)
3671 struct file *file = iocb->ki_filp;
3672 struct address_space *mapping = file->f_mapping;
3673 struct inode *inode = mapping->host;
3674 loff_t pos = iocb->ki_pos;
3679 write_len = iov_iter_count(from);
3680 end = (pos + write_len - 1) >> PAGE_SHIFT;
3682 if (iocb->ki_flags & IOCB_NOWAIT) {
3683 /* If there are pages to writeback, return */
3684 if (filemap_range_has_page(file->f_mapping, pos,
3685 pos + write_len - 1))
3688 written = filemap_write_and_wait_range(mapping, pos,
3689 pos + write_len - 1);
3695 * After a write we want buffered reads to be sure to go to disk to get
3696 * the new data. We invalidate clean cached page from the region we're
3697 * about to write. We do this *before* the write so that we can return
3698 * without clobbering -EIOCBQUEUED from ->direct_IO().
3700 written = invalidate_inode_pages2_range(mapping,
3701 pos >> PAGE_SHIFT, end);
3703 * If a page can not be invalidated, return 0 to fall back
3704 * to buffered write.
3707 if (written == -EBUSY)
3712 written = mapping->a_ops->direct_IO(iocb, from);
3715 * Finally, try again to invalidate clean pages which might have been
3716 * cached by non-direct readahead, or faulted in by get_user_pages()
3717 * if the source of the write was an mmap'ed region of the file
3718 * we're writing. Either one is a pretty crazy thing to do,
3719 * so we don't support it 100%. If this invalidation
3720 * fails, tough, the write still worked...
3722 * Most of the time we do not need this since dio_complete() will do
3723 * the invalidation for us. However there are some file systems that
3724 * do not end up with dio_complete() being called, so let's not break
3725 * them by removing it completely.
3727 * Noticeable example is a blkdev_direct_IO().
3729 * Skip invalidation for async writes or if mapping has no pages.
3731 if (written > 0 && mapping->nrpages &&
3732 invalidate_inode_pages2_range(mapping, pos >> PAGE_SHIFT, end))
3733 dio_warn_stale_pagecache(file);
3737 write_len -= written;
3738 if (pos > i_size_read(inode) && !S_ISBLK(inode->i_mode)) {
3739 i_size_write(inode, pos);
3740 mark_inode_dirty(inode);
3744 if (written != -EIOCBQUEUED)
3745 iov_iter_revert(from, write_len - iov_iter_count(from));
3749 EXPORT_SYMBOL(generic_file_direct_write);
3751 ssize_t generic_perform_write(struct file *file,
3752 struct iov_iter *i, loff_t pos)
3754 struct address_space *mapping = file->f_mapping;
3755 const struct address_space_operations *a_ops = mapping->a_ops;
3757 ssize_t written = 0;
3758 unsigned int flags = 0;
3762 unsigned long offset; /* Offset into pagecache page */
3763 unsigned long bytes; /* Bytes to write to page */
3764 size_t copied; /* Bytes copied from user */
3767 offset = (pos & (PAGE_SIZE - 1));
3768 bytes = min_t(unsigned long, PAGE_SIZE - offset,
3773 * Bring in the user page that we will copy from _first_.
3774 * Otherwise there's a nasty deadlock on copying from the
3775 * same page as we're writing to, without it being marked
3778 if (unlikely(fault_in_iov_iter_readable(i, bytes))) {
3783 if (fatal_signal_pending(current)) {
3788 status = a_ops->write_begin(file, mapping, pos, bytes, flags,
3790 if (unlikely(status < 0))
3793 if (mapping_writably_mapped(mapping))
3794 flush_dcache_page(page);
3796 copied = copy_page_from_iter_atomic(page, offset, bytes, i);
3797 flush_dcache_page(page);
3799 status = a_ops->write_end(file, mapping, pos, bytes, copied,
3801 if (unlikely(status != copied)) {
3802 iov_iter_revert(i, copied - max(status, 0L));
3803 if (unlikely(status < 0))
3808 if (unlikely(status == 0)) {
3810 * A short copy made ->write_end() reject the
3811 * thing entirely. Might be memory poisoning
3812 * halfway through, might be a race with munmap,
3813 * might be severe memory pressure.
3822 balance_dirty_pages_ratelimited(mapping);
3823 } while (iov_iter_count(i));
3825 return written ? written : status;
3827 EXPORT_SYMBOL(generic_perform_write);
3830 * __generic_file_write_iter - write data to a file
3831 * @iocb: IO state structure (file, offset, etc.)
3832 * @from: iov_iter with data to write
3834 * This function does all the work needed for actually writing data to a
3835 * file. It does all basic checks, removes SUID from the file, updates
3836 * modification times and calls proper subroutines depending on whether we
3837 * do direct IO or a standard buffered write.
3839 * It expects i_rwsem to be grabbed unless we work on a block device or similar
3840 * object which does not need locking at all.
3842 * This function does *not* take care of syncing data in case of O_SYNC write.
3843 * A caller has to handle it. This is mainly due to the fact that we want to
3844 * avoid syncing under i_rwsem.
3847 * * number of bytes written, even for truncated writes
3848 * * negative error code if no data has been written at all
3850 ssize_t __generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
3852 struct file *file = iocb->ki_filp;
3853 struct address_space *mapping = file->f_mapping;
3854 struct inode *inode = mapping->host;
3855 ssize_t written = 0;
3859 /* We can write back this queue in page reclaim */
3860 current->backing_dev_info = inode_to_bdi(inode);
3861 err = file_remove_privs(file);
3865 err = file_update_time(file);
3869 if (iocb->ki_flags & IOCB_DIRECT) {
3870 loff_t pos, endbyte;
3872 written = generic_file_direct_write(iocb, from);
3874 * If the write stopped short of completing, fall back to
3875 * buffered writes. Some filesystems do this for writes to
3876 * holes, for example. For DAX files, a buffered write will
3877 * not succeed (even if it did, DAX does not handle dirty
3878 * page-cache pages correctly).
3880 if (written < 0 || !iov_iter_count(from) || IS_DAX(inode))
3883 status = generic_perform_write(file, from, pos = iocb->ki_pos);
3885 * If generic_perform_write() returned a synchronous error
3886 * then we want to return the number of bytes which were
3887 * direct-written, or the error code if that was zero. Note
3888 * that this differs from normal direct-io semantics, which
3889 * will return -EFOO even if some bytes were written.
3891 if (unlikely(status < 0)) {
3896 * We need to ensure that the page cache pages are written to
3897 * disk and invalidated to preserve the expected O_DIRECT
3900 endbyte = pos + status - 1;
3901 err = filemap_write_and_wait_range(mapping, pos, endbyte);
3903 iocb->ki_pos = endbyte + 1;
3905 invalidate_mapping_pages(mapping,
3907 endbyte >> PAGE_SHIFT);
3910 * We don't know how much we wrote, so just return
3911 * the number of bytes which were direct-written
3915 written = generic_perform_write(file, from, iocb->ki_pos);
3916 if (likely(written > 0))
3917 iocb->ki_pos += written;
3920 current->backing_dev_info = NULL;
3921 return written ? written : err;
3923 EXPORT_SYMBOL(__generic_file_write_iter);
3926 * generic_file_write_iter - write data to a file
3927 * @iocb: IO state structure
3928 * @from: iov_iter with data to write
3930 * This is a wrapper around __generic_file_write_iter() to be used by most
3931 * filesystems. It takes care of syncing the file in case of O_SYNC file
3932 * and acquires i_rwsem as needed.
3934 * * negative error code if no data has been written at all of
3935 * vfs_fsync_range() failed for a synchronous write
3936 * * number of bytes written, even for truncated writes
3938 ssize_t generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
3940 struct file *file = iocb->ki_filp;
3941 struct inode *inode = file->f_mapping->host;
3945 ret = generic_write_checks(iocb, from);
3947 ret = __generic_file_write_iter(iocb, from);
3948 inode_unlock(inode);
3951 ret = generic_write_sync(iocb, ret);
3954 EXPORT_SYMBOL(generic_file_write_iter);
3957 * filemap_release_folio() - Release fs-specific metadata on a folio.
3958 * @folio: The folio which the kernel is trying to free.
3959 * @gfp: Memory allocation flags (and I/O mode).
3961 * The address_space is trying to release any data attached to a folio
3962 * (presumably at folio->private).
3964 * This will also be called if the private_2 flag is set on a page,
3965 * indicating that the folio has other metadata associated with it.
3967 * The @gfp argument specifies whether I/O may be performed to release
3968 * this page (__GFP_IO), and whether the call may block
3969 * (__GFP_RECLAIM & __GFP_FS).
3971 * Return: %true if the release was successful, otherwise %false.
3973 bool filemap_release_folio(struct folio *folio, gfp_t gfp)
3975 struct address_space * const mapping = folio->mapping;
3977 BUG_ON(!folio_test_locked(folio));
3978 if (folio_test_writeback(folio))
3981 if (mapping && mapping->a_ops->releasepage)
3982 return mapping->a_ops->releasepage(&folio->page, gfp);
3983 return try_to_free_buffers(&folio->page);
3985 EXPORT_SYMBOL(filemap_release_folio);