1 // SPDX-License-Identifier: GPL-2.0-only
5 * Copyright (C) 1994-1999 Linus Torvalds
9 * This file handles the generic file mmap semantics used by
10 * most "normal" filesystems (but you don't /have/ to use this:
11 * the NFS filesystem used to do this differently, for example)
13 #include <linux/export.h>
14 #include <linux/compiler.h>
15 #include <linux/dax.h>
17 #include <linux/sched/signal.h>
18 #include <linux/uaccess.h>
19 #include <linux/capability.h>
20 #include <linux/kernel_stat.h>
21 #include <linux/gfp.h>
23 #include <linux/swap.h>
24 #include <linux/swapops.h>
25 #include <linux/mman.h>
26 #include <linux/pagemap.h>
27 #include <linux/file.h>
28 #include <linux/uio.h>
29 #include <linux/error-injection.h>
30 #include <linux/hash.h>
31 #include <linux/writeback.h>
32 #include <linux/backing-dev.h>
33 #include <linux/pagevec.h>
34 #include <linux/security.h>
35 #include <linux/cpuset.h>
36 #include <linux/hugetlb.h>
37 #include <linux/memcontrol.h>
38 #include <linux/shmem_fs.h>
39 #include <linux/rmap.h>
40 #include <linux/delayacct.h>
41 #include <linux/psi.h>
42 #include <linux/ramfs.h>
43 #include <linux/page_idle.h>
44 #include <linux/migrate.h>
45 #include <asm/pgalloc.h>
46 #include <asm/tlbflush.h>
49 #define CREATE_TRACE_POINTS
50 #include <trace/events/filemap.h>
53 * FIXME: remove all knowledge of the buffer layer from the core VM
55 #include <linux/buffer_head.h> /* for try_to_free_buffers */
60 * Shared mappings implemented 30.11.1994. It's not fully working yet,
63 * Shared mappings now work. 15.8.1995 Bruno.
65 * finished 'unifying' the page and buffer cache and SMP-threaded the
66 * page-cache, 21.05.1999, Ingo Molnar <mingo@redhat.com>
68 * SMP-threaded pagemap-LRU 1999, Andrea Arcangeli <andrea@suse.de>
74 * ->i_mmap_rwsem (truncate_pagecache)
75 * ->private_lock (__free_pte->__set_page_dirty_buffers)
76 * ->swap_lock (exclusive_swap_page, others)
80 * ->invalidate_lock (acquired by fs in truncate path)
81 * ->i_mmap_rwsem (truncate->unmap_mapping_range)
85 * ->page_table_lock or pte_lock (various, mainly in memory.c)
86 * ->i_pages lock (arch-dependent flush_dcache_mmap_lock)
89 * ->invalidate_lock (filemap_fault)
90 * ->lock_page (filemap_fault, access_process_vm)
92 * ->i_rwsem (generic_perform_write)
93 * ->mmap_lock (fault_in_readable->do_page_fault)
96 * sb_lock (fs/fs-writeback.c)
97 * ->i_pages lock (__sync_single_inode)
100 * ->anon_vma.lock (vma_adjust)
103 * ->page_table_lock or pte_lock (anon_vma_prepare and various)
105 * ->page_table_lock or pte_lock
106 * ->swap_lock (try_to_unmap_one)
107 * ->private_lock (try_to_unmap_one)
108 * ->i_pages lock (try_to_unmap_one)
109 * ->lruvec->lru_lock (follow_page->mark_page_accessed)
110 * ->lruvec->lru_lock (check_pte_range->isolate_lru_page)
111 * ->private_lock (page_remove_rmap->set_page_dirty)
112 * ->i_pages lock (page_remove_rmap->set_page_dirty)
113 * bdi.wb->list_lock (page_remove_rmap->set_page_dirty)
114 * ->inode->i_lock (page_remove_rmap->set_page_dirty)
115 * ->memcg->move_lock (page_remove_rmap->lock_page_memcg)
116 * bdi.wb->list_lock (zap_pte_range->set_page_dirty)
117 * ->inode->i_lock (zap_pte_range->set_page_dirty)
118 * ->private_lock (zap_pte_range->__set_page_dirty_buffers)
121 * ->tasklist_lock (memory_failure, collect_procs_ao)
124 static void page_cache_delete(struct address_space *mapping,
125 struct folio *folio, void *shadow)
127 XA_STATE(xas, &mapping->i_pages, folio->index);
130 mapping_set_update(&xas, mapping);
132 /* hugetlb pages are represented by a single entry in the xarray */
133 if (!folio_test_hugetlb(folio)) {
134 xas_set_order(&xas, folio->index, folio_order(folio));
135 nr = folio_nr_pages(folio);
138 VM_BUG_ON_FOLIO(!folio_test_locked(folio), folio);
140 xas_store(&xas, shadow);
141 xas_init_marks(&xas);
143 folio->mapping = NULL;
144 /* Leave page->index set: truncation lookup relies upon it */
145 mapping->nrpages -= nr;
148 static void filemap_unaccount_folio(struct address_space *mapping,
153 VM_BUG_ON_FOLIO(folio_mapped(folio), folio);
154 if (!IS_ENABLED(CONFIG_DEBUG_VM) && unlikely(folio_mapped(folio))) {
157 pr_alert("BUG: Bad page cache in process %s pfn:%05lx\n",
158 current->comm, folio_pfn(folio));
159 dump_page(&folio->page, "still mapped when deleted");
161 add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE);
163 mapcount = page_mapcount(&folio->page);
164 if (mapping_exiting(mapping) &&
165 folio_ref_count(folio) >= mapcount + 2) {
167 * All vmas have already been torn down, so it's
168 * a good bet that actually the folio is unmapped,
169 * and we'd prefer not to leak it: if we're wrong,
170 * some other bad page check should catch it later.
172 page_mapcount_reset(&folio->page);
173 folio_ref_sub(folio, mapcount);
177 /* hugetlb folios do not participate in page cache accounting. */
178 if (folio_test_hugetlb(folio))
181 nr = folio_nr_pages(folio);
183 __lruvec_stat_mod_folio(folio, NR_FILE_PAGES, -nr);
184 if (folio_test_swapbacked(folio)) {
185 __lruvec_stat_mod_folio(folio, NR_SHMEM, -nr);
186 if (folio_test_pmd_mappable(folio))
187 __lruvec_stat_mod_folio(folio, NR_SHMEM_THPS, -nr);
188 } else if (folio_test_pmd_mappable(folio)) {
189 __lruvec_stat_mod_folio(folio, NR_FILE_THPS, -nr);
190 filemap_nr_thps_dec(mapping);
194 * At this point folio must be either written or cleaned by
195 * truncate. Dirty folio here signals a bug and loss of
198 * This fixes dirty accounting after removing the folio entirely
199 * but leaves the dirty flag set: it has no effect for truncated
200 * folio and anyway will be cleared before returning folio to
203 if (WARN_ON_ONCE(folio_test_dirty(folio)))
204 folio_account_cleaned(folio, mapping,
205 inode_to_wb(mapping->host));
209 * Delete a page from the page cache and free it. Caller has to make
210 * sure the page is locked and that nobody else uses it - or that usage
211 * is safe. The caller must hold the i_pages lock.
213 void __filemap_remove_folio(struct folio *folio, void *shadow)
215 struct address_space *mapping = folio->mapping;
217 trace_mm_filemap_delete_from_page_cache(folio);
218 filemap_unaccount_folio(mapping, folio);
219 page_cache_delete(mapping, folio, shadow);
222 void filemap_free_folio(struct address_space *mapping, struct folio *folio)
224 void (*freepage)(struct page *);
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);
846 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 folio->mapping = mapping;
854 folio->index = index;
857 error = mem_cgroup_charge(folio, NULL, gfp);
858 VM_BUG_ON_FOLIO(index & (folio_nr_pages(folio) - 1), folio);
864 gfp &= GFP_RECLAIM_MASK;
867 unsigned int order = xa_get_order(xas.xa, xas.xa_index);
868 void *entry, *old = NULL;
870 if (order > folio_order(folio))
871 xas_split_alloc(&xas, xa_load(xas.xa, xas.xa_index),
874 xas_for_each_conflict(&xas, entry) {
876 if (!xa_is_value(entry)) {
877 xas_set_err(&xas, -EEXIST);
885 /* entry may have been split before we acquired lock */
886 order = xa_get_order(xas.xa, xas.xa_index);
887 if (order > folio_order(folio)) {
888 xas_split(&xas, old, order);
893 xas_store(&xas, folio);
899 /* hugetlb pages do not participate in page cache accounting */
901 __lruvec_stat_add_folio(folio, NR_FILE_PAGES);
903 xas_unlock_irq(&xas);
904 } while (xas_nomem(&xas, gfp));
906 if (xas_error(&xas)) {
907 error = xas_error(&xas);
909 mem_cgroup_uncharge(folio);
913 trace_mm_filemap_add_to_page_cache(folio);
916 folio->mapping = NULL;
917 /* Leave page->index set: truncation relies upon it */
921 ALLOW_ERROR_INJECTION(__filemap_add_folio, ERRNO);
924 * add_to_page_cache_locked - add a locked page to the pagecache
926 * @mapping: the page's address_space
927 * @offset: page index
928 * @gfp_mask: page allocation mode
930 * This function is used to add a page to the pagecache. It must be locked.
931 * This function does not add the page to the LRU. The caller must do that.
933 * Return: %0 on success, negative error code otherwise.
935 int add_to_page_cache_locked(struct page *page, struct address_space *mapping,
936 pgoff_t offset, gfp_t gfp_mask)
938 return __filemap_add_folio(mapping, page_folio(page), offset,
941 EXPORT_SYMBOL(add_to_page_cache_locked);
943 int filemap_add_folio(struct address_space *mapping, struct folio *folio,
944 pgoff_t index, gfp_t gfp)
949 __folio_set_locked(folio);
950 ret = __filemap_add_folio(mapping, folio, index, gfp, &shadow);
952 __folio_clear_locked(folio);
955 * The folio might have been evicted from cache only
956 * recently, in which case it should be activated like
957 * any other repeatedly accessed folio.
958 * The exception is folios getting rewritten; evicting other
959 * data from the working set, only to cache data that will
960 * get overwritten with something else, is a waste of memory.
962 WARN_ON_ONCE(folio_test_active(folio));
963 if (!(gfp & __GFP_WRITE) && shadow)
964 workingset_refault(folio, shadow);
965 folio_add_lru(folio);
969 EXPORT_SYMBOL_GPL(filemap_add_folio);
972 struct folio *filemap_alloc_folio(gfp_t gfp, unsigned int order)
977 if (cpuset_do_page_mem_spread()) {
978 unsigned int cpuset_mems_cookie;
980 cpuset_mems_cookie = read_mems_allowed_begin();
981 n = cpuset_mem_spread_node();
982 folio = __folio_alloc_node(gfp, order, n);
983 } while (!folio && read_mems_allowed_retry(cpuset_mems_cookie));
987 return folio_alloc(gfp, order);
989 EXPORT_SYMBOL(filemap_alloc_folio);
993 * filemap_invalidate_lock_two - lock invalidate_lock for two mappings
995 * Lock exclusively invalidate_lock of any passed mapping that is not NULL.
997 * @mapping1: the first mapping to lock
998 * @mapping2: the second mapping to lock
1000 void filemap_invalidate_lock_two(struct address_space *mapping1,
1001 struct address_space *mapping2)
1003 if (mapping1 > mapping2)
1004 swap(mapping1, mapping2);
1006 down_write(&mapping1->invalidate_lock);
1007 if (mapping2 && mapping1 != mapping2)
1008 down_write_nested(&mapping2->invalidate_lock, 1);
1010 EXPORT_SYMBOL(filemap_invalidate_lock_two);
1013 * filemap_invalidate_unlock_two - unlock invalidate_lock for two mappings
1015 * Unlock exclusive invalidate_lock of any passed mapping that is not NULL.
1017 * @mapping1: the first mapping to unlock
1018 * @mapping2: the second mapping to unlock
1020 void filemap_invalidate_unlock_two(struct address_space *mapping1,
1021 struct address_space *mapping2)
1024 up_write(&mapping1->invalidate_lock);
1025 if (mapping2 && mapping1 != mapping2)
1026 up_write(&mapping2->invalidate_lock);
1028 EXPORT_SYMBOL(filemap_invalidate_unlock_two);
1031 * In order to wait for pages to become available there must be
1032 * waitqueues associated with pages. By using a hash table of
1033 * waitqueues where the bucket discipline is to maintain all
1034 * waiters on the same queue and wake all when any of the pages
1035 * become available, and for the woken contexts to check to be
1036 * sure the appropriate page became available, this saves space
1037 * at a cost of "thundering herd" phenomena during rare hash
1040 #define PAGE_WAIT_TABLE_BITS 8
1041 #define PAGE_WAIT_TABLE_SIZE (1 << PAGE_WAIT_TABLE_BITS)
1042 static wait_queue_head_t folio_wait_table[PAGE_WAIT_TABLE_SIZE] __cacheline_aligned;
1044 static wait_queue_head_t *folio_waitqueue(struct folio *folio)
1046 return &folio_wait_table[hash_ptr(folio, PAGE_WAIT_TABLE_BITS)];
1049 void __init pagecache_init(void)
1053 for (i = 0; i < PAGE_WAIT_TABLE_SIZE; i++)
1054 init_waitqueue_head(&folio_wait_table[i]);
1056 page_writeback_init();
1059 * tmpfs uses the ZERO_PAGE for reading holes: it is up-to-date,
1060 * and splice's page_cache_pipe_buf_confirm() needs to see that.
1062 SetPageUptodate(ZERO_PAGE(0));
1066 * The page wait code treats the "wait->flags" somewhat unusually, because
1067 * we have multiple different kinds of waits, not just the usual "exclusive"
1072 * (a) no special bits set:
1074 * We're just waiting for the bit to be released, and when a waker
1075 * calls the wakeup function, we set WQ_FLAG_WOKEN and wake it up,
1076 * and remove it from the wait queue.
1078 * Simple and straightforward.
1080 * (b) WQ_FLAG_EXCLUSIVE:
1082 * The waiter is waiting to get the lock, and only one waiter should
1083 * be woken up to avoid any thundering herd behavior. We'll set the
1084 * WQ_FLAG_WOKEN bit, wake it up, and remove it from the wait queue.
1086 * This is the traditional exclusive wait.
1088 * (c) WQ_FLAG_EXCLUSIVE | WQ_FLAG_CUSTOM:
1090 * The waiter is waiting to get the bit, and additionally wants the
1091 * lock to be transferred to it for fair lock behavior. If the lock
1092 * cannot be taken, we stop walking the wait queue without waking
1095 * This is the "fair lock handoff" case, and in addition to setting
1096 * WQ_FLAG_WOKEN, we set WQ_FLAG_DONE to let the waiter easily see
1097 * that it now has the lock.
1099 static int wake_page_function(wait_queue_entry_t *wait, unsigned mode, int sync, void *arg)
1102 struct wait_page_key *key = arg;
1103 struct wait_page_queue *wait_page
1104 = container_of(wait, struct wait_page_queue, wait);
1106 if (!wake_page_match(wait_page, key))
1110 * If it's a lock handoff wait, we get the bit for it, and
1111 * stop walking (and do not wake it up) if we can't.
1113 flags = wait->flags;
1114 if (flags & WQ_FLAG_EXCLUSIVE) {
1115 if (test_bit(key->bit_nr, &key->folio->flags))
1117 if (flags & WQ_FLAG_CUSTOM) {
1118 if (test_and_set_bit(key->bit_nr, &key->folio->flags))
1120 flags |= WQ_FLAG_DONE;
1125 * We are holding the wait-queue lock, but the waiter that
1126 * is waiting for this will be checking the flags without
1129 * So update the flags atomically, and wake up the waiter
1130 * afterwards to avoid any races. This store-release pairs
1131 * with the load-acquire in folio_wait_bit_common().
1133 smp_store_release(&wait->flags, flags | WQ_FLAG_WOKEN);
1134 wake_up_state(wait->private, mode);
1137 * Ok, we have successfully done what we're waiting for,
1138 * and we can unconditionally remove the wait entry.
1140 * Note that this pairs with the "finish_wait()" in the
1141 * waiter, and has to be the absolute last thing we do.
1142 * After this list_del_init(&wait->entry) the wait entry
1143 * might be de-allocated and the process might even have
1146 list_del_init_careful(&wait->entry);
1147 return (flags & WQ_FLAG_EXCLUSIVE) != 0;
1150 static void folio_wake_bit(struct folio *folio, int bit_nr)
1152 wait_queue_head_t *q = folio_waitqueue(folio);
1153 struct wait_page_key key;
1154 unsigned long flags;
1155 wait_queue_entry_t bookmark;
1158 key.bit_nr = bit_nr;
1162 bookmark.private = NULL;
1163 bookmark.func = NULL;
1164 INIT_LIST_HEAD(&bookmark.entry);
1166 spin_lock_irqsave(&q->lock, flags);
1167 __wake_up_locked_key_bookmark(q, TASK_NORMAL, &key, &bookmark);
1169 while (bookmark.flags & WQ_FLAG_BOOKMARK) {
1171 * Take a breather from holding the lock,
1172 * allow pages that finish wake up asynchronously
1173 * to acquire the lock and remove themselves
1176 spin_unlock_irqrestore(&q->lock, flags);
1178 spin_lock_irqsave(&q->lock, flags);
1179 __wake_up_locked_key_bookmark(q, TASK_NORMAL, &key, &bookmark);
1183 * It is possible for other pages to have collided on the waitqueue
1184 * hash, so in that case check for a page match. That prevents a long-
1187 * It is still possible to miss a case here, when we woke page waiters
1188 * and removed them from the waitqueue, but there are still other
1191 if (!waitqueue_active(q) || !key.page_match) {
1192 folio_clear_waiters(folio);
1194 * It's possible to miss clearing Waiters here, when we woke
1195 * our page waiters, but the hashed waitqueue has waiters for
1196 * other pages on it.
1198 * That's okay, it's a rare case. The next waker will clear it.
1201 spin_unlock_irqrestore(&q->lock, flags);
1204 static void folio_wake(struct folio *folio, int bit)
1206 if (!folio_test_waiters(folio))
1208 folio_wake_bit(folio, bit);
1212 * A choice of three behaviors for folio_wait_bit_common():
1215 EXCLUSIVE, /* Hold ref to page and take the bit when woken, like
1216 * __folio_lock() waiting on then setting PG_locked.
1218 SHARED, /* Hold ref to page and check the bit when woken, like
1219 * folio_wait_writeback() waiting on PG_writeback.
1221 DROP, /* Drop ref to page before wait, no check when woken,
1222 * like folio_put_wait_locked() on PG_locked.
1227 * Attempt to check (or get) the folio flag, and mark us done
1230 static inline bool folio_trylock_flag(struct folio *folio, int bit_nr,
1231 struct wait_queue_entry *wait)
1233 if (wait->flags & WQ_FLAG_EXCLUSIVE) {
1234 if (test_and_set_bit(bit_nr, &folio->flags))
1236 } else if (test_bit(bit_nr, &folio->flags))
1239 wait->flags |= WQ_FLAG_WOKEN | WQ_FLAG_DONE;
1243 /* How many times do we accept lock stealing from under a waiter? */
1244 int sysctl_page_lock_unfairness = 5;
1246 static inline int folio_wait_bit_common(struct folio *folio, int bit_nr,
1247 int state, enum behavior behavior)
1249 wait_queue_head_t *q = folio_waitqueue(folio);
1250 int unfairness = sysctl_page_lock_unfairness;
1251 struct wait_page_queue wait_page;
1252 wait_queue_entry_t *wait = &wait_page.wait;
1253 bool thrashing = false;
1254 bool delayacct = false;
1255 unsigned long pflags;
1257 if (bit_nr == PG_locked &&
1258 !folio_test_uptodate(folio) && folio_test_workingset(folio)) {
1259 if (!folio_test_swapbacked(folio)) {
1260 delayacct_thrashing_start();
1263 psi_memstall_enter(&pflags);
1268 wait->func = wake_page_function;
1269 wait_page.folio = folio;
1270 wait_page.bit_nr = bit_nr;
1274 if (behavior == EXCLUSIVE) {
1275 wait->flags = WQ_FLAG_EXCLUSIVE;
1276 if (--unfairness < 0)
1277 wait->flags |= WQ_FLAG_CUSTOM;
1281 * Do one last check whether we can get the
1282 * page bit synchronously.
1284 * Do the folio_set_waiters() marking before that
1285 * to let any waker we _just_ missed know they
1286 * need to wake us up (otherwise they'll never
1287 * even go to the slow case that looks at the
1288 * page queue), and add ourselves to the wait
1289 * queue if we need to sleep.
1291 * This part needs to be done under the queue
1292 * lock to avoid races.
1294 spin_lock_irq(&q->lock);
1295 folio_set_waiters(folio);
1296 if (!folio_trylock_flag(folio, bit_nr, wait))
1297 __add_wait_queue_entry_tail(q, wait);
1298 spin_unlock_irq(&q->lock);
1301 * From now on, all the logic will be based on
1302 * the WQ_FLAG_WOKEN and WQ_FLAG_DONE flag, to
1303 * see whether the page bit testing has already
1304 * been done by the wake function.
1306 * We can drop our reference to the folio.
1308 if (behavior == DROP)
1312 * Note that until the "finish_wait()", or until
1313 * we see the WQ_FLAG_WOKEN flag, we need to
1314 * be very careful with the 'wait->flags', because
1315 * we may race with a waker that sets them.
1320 set_current_state(state);
1322 /* Loop until we've been woken or interrupted */
1323 flags = smp_load_acquire(&wait->flags);
1324 if (!(flags & WQ_FLAG_WOKEN)) {
1325 if (signal_pending_state(state, current))
1332 /* If we were non-exclusive, we're done */
1333 if (behavior != EXCLUSIVE)
1336 /* If the waker got the lock for us, we're done */
1337 if (flags & WQ_FLAG_DONE)
1341 * Otherwise, if we're getting the lock, we need to
1342 * try to get it ourselves.
1344 * And if that fails, we'll have to retry this all.
1346 if (unlikely(test_and_set_bit(bit_nr, folio_flags(folio, 0))))
1349 wait->flags |= WQ_FLAG_DONE;
1354 * If a signal happened, this 'finish_wait()' may remove the last
1355 * waiter from the wait-queues, but the folio waiters bit will remain
1356 * set. That's ok. The next wakeup will take care of it, and trying
1357 * to do it here would be difficult and prone to races.
1359 finish_wait(q, wait);
1363 delayacct_thrashing_end();
1364 psi_memstall_leave(&pflags);
1368 * NOTE! The wait->flags weren't stable until we've done the
1369 * 'finish_wait()', and we could have exited the loop above due
1370 * to a signal, and had a wakeup event happen after the signal
1371 * test but before the 'finish_wait()'.
1373 * So only after the finish_wait() can we reliably determine
1374 * if we got woken up or not, so we can now figure out the final
1375 * return value based on that state without races.
1377 * Also note that WQ_FLAG_WOKEN is sufficient for a non-exclusive
1378 * waiter, but an exclusive one requires WQ_FLAG_DONE.
1380 if (behavior == EXCLUSIVE)
1381 return wait->flags & WQ_FLAG_DONE ? 0 : -EINTR;
1383 return wait->flags & WQ_FLAG_WOKEN ? 0 : -EINTR;
1386 #ifdef CONFIG_MIGRATION
1388 * migration_entry_wait_on_locked - Wait for a migration entry to be removed
1389 * @entry: migration swap entry.
1390 * @ptep: mapped pte pointer. Will return with the ptep unmapped. Only required
1391 * for pte entries, pass NULL for pmd entries.
1392 * @ptl: already locked ptl. This function will drop the lock.
1394 * Wait for a migration entry referencing the given page to be removed. This is
1395 * equivalent to put_and_wait_on_page_locked(page, TASK_UNINTERRUPTIBLE) except
1396 * this can be called without taking a reference on the page. Instead this
1397 * should be called while holding the ptl for the migration entry referencing
1400 * Returns after unmapping and unlocking the pte/ptl with pte_unmap_unlock().
1402 * This follows the same logic as folio_wait_bit_common() so see the comments
1405 void migration_entry_wait_on_locked(swp_entry_t entry, pte_t *ptep,
1408 struct wait_page_queue wait_page;
1409 wait_queue_entry_t *wait = &wait_page.wait;
1410 bool thrashing = false;
1411 bool delayacct = false;
1412 unsigned long pflags;
1413 wait_queue_head_t *q;
1414 struct folio *folio = page_folio(pfn_swap_entry_to_page(entry));
1416 q = folio_waitqueue(folio);
1417 if (!folio_test_uptodate(folio) && folio_test_workingset(folio)) {
1418 if (!folio_test_swapbacked(folio)) {
1419 delayacct_thrashing_start();
1422 psi_memstall_enter(&pflags);
1427 wait->func = wake_page_function;
1428 wait_page.folio = folio;
1429 wait_page.bit_nr = PG_locked;
1432 spin_lock_irq(&q->lock);
1433 folio_set_waiters(folio);
1434 if (!folio_trylock_flag(folio, PG_locked, wait))
1435 __add_wait_queue_entry_tail(q, wait);
1436 spin_unlock_irq(&q->lock);
1439 * If a migration entry exists for the page the migration path must hold
1440 * a valid reference to the page, and it must take the ptl to remove the
1441 * migration entry. So the page is valid until the ptl is dropped.
1444 pte_unmap_unlock(ptep, ptl);
1451 set_current_state(TASK_UNINTERRUPTIBLE);
1453 /* Loop until we've been woken or interrupted */
1454 flags = smp_load_acquire(&wait->flags);
1455 if (!(flags & WQ_FLAG_WOKEN)) {
1456 if (signal_pending_state(TASK_UNINTERRUPTIBLE, current))
1465 finish_wait(q, wait);
1469 delayacct_thrashing_end();
1470 psi_memstall_leave(&pflags);
1475 void folio_wait_bit(struct folio *folio, int bit_nr)
1477 folio_wait_bit_common(folio, bit_nr, TASK_UNINTERRUPTIBLE, SHARED);
1479 EXPORT_SYMBOL(folio_wait_bit);
1481 int folio_wait_bit_killable(struct folio *folio, int bit_nr)
1483 return folio_wait_bit_common(folio, bit_nr, TASK_KILLABLE, SHARED);
1485 EXPORT_SYMBOL(folio_wait_bit_killable);
1488 * folio_put_wait_locked - Drop a reference and wait for it to be unlocked
1489 * @folio: The folio to wait for.
1490 * @state: The sleep state (TASK_KILLABLE, TASK_UNINTERRUPTIBLE, etc).
1492 * The caller should hold a reference on @folio. They expect the page to
1493 * become unlocked relatively soon, but do not wish to hold up migration
1494 * (for example) by holding the reference while waiting for the folio to
1495 * come unlocked. After this function returns, the caller should not
1496 * dereference @folio.
1498 * Return: 0 if the folio was unlocked or -EINTR if interrupted by a signal.
1500 int folio_put_wait_locked(struct folio *folio, int state)
1502 return folio_wait_bit_common(folio, PG_locked, state, DROP);
1506 * folio_add_wait_queue - Add an arbitrary waiter to a folio's wait queue
1507 * @folio: Folio defining the wait queue of interest
1508 * @waiter: Waiter to add to the queue
1510 * Add an arbitrary @waiter to the wait queue for the nominated @folio.
1512 void folio_add_wait_queue(struct folio *folio, wait_queue_entry_t *waiter)
1514 wait_queue_head_t *q = folio_waitqueue(folio);
1515 unsigned long flags;
1517 spin_lock_irqsave(&q->lock, flags);
1518 __add_wait_queue_entry_tail(q, waiter);
1519 folio_set_waiters(folio);
1520 spin_unlock_irqrestore(&q->lock, flags);
1522 EXPORT_SYMBOL_GPL(folio_add_wait_queue);
1524 #ifndef clear_bit_unlock_is_negative_byte
1527 * PG_waiters is the high bit in the same byte as PG_lock.
1529 * On x86 (and on many other architectures), we can clear PG_lock and
1530 * test the sign bit at the same time. But if the architecture does
1531 * not support that special operation, we just do this all by hand
1534 * The read of PG_waiters has to be after (or concurrently with) PG_locked
1535 * being cleared, but a memory barrier should be unnecessary since it is
1536 * in the same byte as PG_locked.
1538 static inline bool clear_bit_unlock_is_negative_byte(long nr, volatile void *mem)
1540 clear_bit_unlock(nr, mem);
1541 /* smp_mb__after_atomic(); */
1542 return test_bit(PG_waiters, mem);
1548 * folio_unlock - Unlock a locked folio.
1549 * @folio: The folio.
1551 * Unlocks the folio and wakes up any thread sleeping on the page lock.
1553 * Context: May be called from interrupt or process context. May not be
1554 * called from NMI context.
1556 void folio_unlock(struct folio *folio)
1558 /* Bit 7 allows x86 to check the byte's sign bit */
1559 BUILD_BUG_ON(PG_waiters != 7);
1560 BUILD_BUG_ON(PG_locked > 7);
1561 VM_BUG_ON_FOLIO(!folio_test_locked(folio), folio);
1562 if (clear_bit_unlock_is_negative_byte(PG_locked, folio_flags(folio, 0)))
1563 folio_wake_bit(folio, PG_locked);
1565 EXPORT_SYMBOL(folio_unlock);
1568 * folio_end_private_2 - Clear PG_private_2 and wake any waiters.
1569 * @folio: The folio.
1571 * Clear the PG_private_2 bit on a folio and wake up any sleepers waiting for
1572 * it. The folio reference held for PG_private_2 being set is released.
1574 * This is, for example, used when a netfs folio is being written to a local
1575 * disk cache, thereby allowing writes to the cache for the same folio to be
1578 void folio_end_private_2(struct folio *folio)
1580 VM_BUG_ON_FOLIO(!folio_test_private_2(folio), folio);
1581 clear_bit_unlock(PG_private_2, folio_flags(folio, 0));
1582 folio_wake_bit(folio, PG_private_2);
1585 EXPORT_SYMBOL(folio_end_private_2);
1588 * folio_wait_private_2 - Wait for PG_private_2 to be cleared on a folio.
1589 * @folio: The folio to wait on.
1591 * Wait for PG_private_2 (aka PG_fscache) to be cleared on a folio.
1593 void folio_wait_private_2(struct folio *folio)
1595 while (folio_test_private_2(folio))
1596 folio_wait_bit(folio, PG_private_2);
1598 EXPORT_SYMBOL(folio_wait_private_2);
1601 * folio_wait_private_2_killable - Wait for PG_private_2 to be cleared on a folio.
1602 * @folio: The folio to wait on.
1604 * Wait for PG_private_2 (aka PG_fscache) to be cleared on a folio or until a
1605 * fatal signal is received by the calling task.
1608 * - 0 if successful.
1609 * - -EINTR if a fatal signal was encountered.
1611 int folio_wait_private_2_killable(struct folio *folio)
1615 while (folio_test_private_2(folio)) {
1616 ret = folio_wait_bit_killable(folio, PG_private_2);
1623 EXPORT_SYMBOL(folio_wait_private_2_killable);
1626 * folio_end_writeback - End writeback against a folio.
1627 * @folio: The folio.
1629 void folio_end_writeback(struct folio *folio)
1632 * folio_test_clear_reclaim() could be used here but it is an
1633 * atomic operation and overkill in this particular case. Failing
1634 * to shuffle a folio marked for immediate reclaim is too mild
1635 * a gain to justify taking an atomic operation penalty at the
1636 * end of every folio writeback.
1638 if (folio_test_reclaim(folio)) {
1639 folio_clear_reclaim(folio);
1640 folio_rotate_reclaimable(folio);
1644 * Writeback does not hold a folio reference of its own, relying
1645 * on truncation to wait for the clearing of PG_writeback.
1646 * But here we must make sure that the folio is not freed and
1647 * reused before the folio_wake().
1650 if (!__folio_end_writeback(folio))
1653 smp_mb__after_atomic();
1654 folio_wake(folio, PG_writeback);
1655 acct_reclaim_writeback(folio);
1658 EXPORT_SYMBOL(folio_end_writeback);
1661 * After completing I/O on a page, call this routine to update the page
1662 * flags appropriately
1664 void page_endio(struct page *page, bool is_write, int err)
1668 SetPageUptodate(page);
1670 ClearPageUptodate(page);
1676 struct address_space *mapping;
1679 mapping = page_mapping(page);
1681 mapping_set_error(mapping, err);
1683 end_page_writeback(page);
1686 EXPORT_SYMBOL_GPL(page_endio);
1689 * __folio_lock - Get a lock on the folio, assuming we need to sleep to get it.
1690 * @folio: The folio to lock
1692 void __folio_lock(struct folio *folio)
1694 folio_wait_bit_common(folio, PG_locked, TASK_UNINTERRUPTIBLE,
1697 EXPORT_SYMBOL(__folio_lock);
1699 int __folio_lock_killable(struct folio *folio)
1701 return folio_wait_bit_common(folio, PG_locked, TASK_KILLABLE,
1704 EXPORT_SYMBOL_GPL(__folio_lock_killable);
1706 static int __folio_lock_async(struct folio *folio, struct wait_page_queue *wait)
1708 struct wait_queue_head *q = folio_waitqueue(folio);
1711 wait->folio = folio;
1712 wait->bit_nr = PG_locked;
1714 spin_lock_irq(&q->lock);
1715 __add_wait_queue_entry_tail(q, &wait->wait);
1716 folio_set_waiters(folio);
1717 ret = !folio_trylock(folio);
1719 * If we were successful now, we know we're still on the
1720 * waitqueue as we're still under the lock. This means it's
1721 * safe to remove and return success, we know the callback
1722 * isn't going to trigger.
1725 __remove_wait_queue(q, &wait->wait);
1728 spin_unlock_irq(&q->lock);
1734 * true - folio is locked; mmap_lock is still held.
1735 * false - folio is not locked.
1736 * mmap_lock has been released (mmap_read_unlock(), unless flags had both
1737 * FAULT_FLAG_ALLOW_RETRY and FAULT_FLAG_RETRY_NOWAIT set, in
1738 * which case mmap_lock is still held.
1740 * If neither ALLOW_RETRY nor KILLABLE are set, will always return true
1741 * with the folio locked and the mmap_lock unperturbed.
1743 bool __folio_lock_or_retry(struct folio *folio, struct mm_struct *mm,
1746 if (fault_flag_allow_retry_first(flags)) {
1748 * CAUTION! In this case, mmap_lock is not released
1749 * even though return 0.
1751 if (flags & FAULT_FLAG_RETRY_NOWAIT)
1754 mmap_read_unlock(mm);
1755 if (flags & FAULT_FLAG_KILLABLE)
1756 folio_wait_locked_killable(folio);
1758 folio_wait_locked(folio);
1761 if (flags & FAULT_FLAG_KILLABLE) {
1764 ret = __folio_lock_killable(folio);
1766 mmap_read_unlock(mm);
1770 __folio_lock(folio);
1777 * page_cache_next_miss() - Find the next gap in the page cache.
1778 * @mapping: Mapping.
1780 * @max_scan: Maximum range to search.
1782 * Search the range [index, min(index + max_scan - 1, ULONG_MAX)] for the
1783 * gap with the lowest index.
1785 * This function may be called under the rcu_read_lock. However, this will
1786 * not atomically search a snapshot of the cache at a single point in time.
1787 * For example, if a gap is created at index 5, then subsequently a gap is
1788 * created at index 10, page_cache_next_miss covering both indices may
1789 * return 10 if called under the rcu_read_lock.
1791 * Return: The index of the gap if found, otherwise an index outside the
1792 * range specified (in which case 'return - index >= max_scan' will be true).
1793 * In the rare case of index wrap-around, 0 will be returned.
1795 pgoff_t page_cache_next_miss(struct address_space *mapping,
1796 pgoff_t index, unsigned long max_scan)
1798 XA_STATE(xas, &mapping->i_pages, index);
1800 while (max_scan--) {
1801 void *entry = xas_next(&xas);
1802 if (!entry || xa_is_value(entry))
1804 if (xas.xa_index == 0)
1808 return xas.xa_index;
1810 EXPORT_SYMBOL(page_cache_next_miss);
1813 * page_cache_prev_miss() - Find the previous gap in the page cache.
1814 * @mapping: Mapping.
1816 * @max_scan: Maximum range to search.
1818 * Search the range [max(index - max_scan + 1, 0), index] for the
1819 * gap with the highest index.
1821 * This function may be called under the rcu_read_lock. However, this will
1822 * not atomically search a snapshot of the cache at a single point in time.
1823 * For example, if a gap is created at index 10, then subsequently a gap is
1824 * created at index 5, page_cache_prev_miss() covering both indices may
1825 * return 5 if called under the rcu_read_lock.
1827 * Return: The index of the gap if found, otherwise an index outside the
1828 * range specified (in which case 'index - return >= max_scan' will be true).
1829 * In the rare case of wrap-around, ULONG_MAX will be returned.
1831 pgoff_t page_cache_prev_miss(struct address_space *mapping,
1832 pgoff_t index, unsigned long max_scan)
1834 XA_STATE(xas, &mapping->i_pages, index);
1836 while (max_scan--) {
1837 void *entry = xas_prev(&xas);
1838 if (!entry || xa_is_value(entry))
1840 if (xas.xa_index == ULONG_MAX)
1844 return xas.xa_index;
1846 EXPORT_SYMBOL(page_cache_prev_miss);
1849 * Lockless page cache protocol:
1850 * On the lookup side:
1851 * 1. Load the folio from i_pages
1852 * 2. Increment the refcount if it's not zero
1853 * 3. If the folio is not found by xas_reload(), put the refcount and retry
1855 * On the removal side:
1856 * A. Freeze the page (by zeroing the refcount if nobody else has a reference)
1857 * B. Remove the page from i_pages
1858 * C. Return the page to the page allocator
1860 * This means that any page may have its reference count temporarily
1861 * increased by a speculative page cache (or fast GUP) lookup as it can
1862 * be allocated by another user before the RCU grace period expires.
1863 * Because the refcount temporarily acquired here may end up being the
1864 * last refcount on the page, any page allocation must be freeable by
1869 * mapping_get_entry - Get a page cache entry.
1870 * @mapping: the address_space to search
1871 * @index: The page cache index.
1873 * Looks up the page cache entry at @mapping & @index. If it is a folio,
1874 * it is returned with an increased refcount. If it is a shadow entry
1875 * of a previously evicted folio, or a swap entry from shmem/tmpfs,
1876 * it is returned without further action.
1878 * Return: The folio, swap or shadow entry, %NULL if nothing is found.
1880 static void *mapping_get_entry(struct address_space *mapping, pgoff_t index)
1882 XA_STATE(xas, &mapping->i_pages, index);
1883 struct folio *folio;
1888 folio = xas_load(&xas);
1889 if (xas_retry(&xas, folio))
1892 * A shadow entry of a recently evicted page, or a swap entry from
1893 * shmem/tmpfs. Return it without attempting to raise page count.
1895 if (!folio || xa_is_value(folio))
1898 if (!folio_try_get_rcu(folio))
1901 if (unlikely(folio != xas_reload(&xas))) {
1912 * __filemap_get_folio - Find and get a reference to a folio.
1913 * @mapping: The address_space to search.
1914 * @index: The page index.
1915 * @fgp_flags: %FGP flags modify how the folio is returned.
1916 * @gfp: Memory allocation flags to use if %FGP_CREAT is specified.
1918 * Looks up the page cache entry at @mapping & @index.
1920 * @fgp_flags can be zero or more of these flags:
1922 * * %FGP_ACCESSED - The folio will be marked accessed.
1923 * * %FGP_LOCK - The folio is returned locked.
1924 * * %FGP_ENTRY - If there is a shadow / swap / DAX entry, return it
1925 * instead of allocating a new folio to replace it.
1926 * * %FGP_CREAT - If no page is present then a new page is allocated using
1927 * @gfp and added to the page cache and the VM's LRU list.
1928 * The page is returned locked and with an increased refcount.
1929 * * %FGP_FOR_MMAP - The caller wants to do its own locking dance if the
1930 * page is already in cache. If the page was allocated, unlock it before
1931 * returning so the caller can do the same dance.
1932 * * %FGP_WRITE - The page will be written to by the caller.
1933 * * %FGP_NOFS - __GFP_FS will get cleared in gfp.
1934 * * %FGP_NOWAIT - Don't get blocked by page lock.
1935 * * %FGP_STABLE - Wait for the folio to be stable (finished writeback)
1937 * If %FGP_LOCK or %FGP_CREAT are specified then the function may sleep even
1938 * if the %GFP flags specified for %FGP_CREAT are atomic.
1940 * If there is a page cache page, it is returned with an increased refcount.
1942 * Return: The found folio or %NULL otherwise.
1944 struct folio *__filemap_get_folio(struct address_space *mapping, pgoff_t index,
1945 int fgp_flags, gfp_t gfp)
1947 struct folio *folio;
1950 folio = mapping_get_entry(mapping, index);
1951 if (xa_is_value(folio)) {
1952 if (fgp_flags & FGP_ENTRY)
1959 if (fgp_flags & FGP_LOCK) {
1960 if (fgp_flags & FGP_NOWAIT) {
1961 if (!folio_trylock(folio)) {
1969 /* Has the page been truncated? */
1970 if (unlikely(folio->mapping != mapping)) {
1971 folio_unlock(folio);
1975 VM_BUG_ON_FOLIO(!folio_contains(folio, index), folio);
1978 if (fgp_flags & FGP_ACCESSED)
1979 folio_mark_accessed(folio);
1980 else if (fgp_flags & FGP_WRITE) {
1981 /* Clear idle flag for buffer write */
1982 if (folio_test_idle(folio))
1983 folio_clear_idle(folio);
1986 if (fgp_flags & FGP_STABLE)
1987 folio_wait_stable(folio);
1989 if (!folio && (fgp_flags & FGP_CREAT)) {
1991 if ((fgp_flags & FGP_WRITE) && mapping_can_writeback(mapping))
1993 if (fgp_flags & FGP_NOFS)
1996 folio = filemap_alloc_folio(gfp, 0);
2000 if (WARN_ON_ONCE(!(fgp_flags & (FGP_LOCK | FGP_FOR_MMAP))))
2001 fgp_flags |= FGP_LOCK;
2003 /* Init accessed so avoid atomic mark_page_accessed later */
2004 if (fgp_flags & FGP_ACCESSED)
2005 __folio_set_referenced(folio);
2007 err = filemap_add_folio(mapping, folio, index, gfp);
2008 if (unlikely(err)) {
2016 * filemap_add_folio locks the page, and for mmap
2017 * we expect an unlocked page.
2019 if (folio && (fgp_flags & FGP_FOR_MMAP))
2020 folio_unlock(folio);
2025 EXPORT_SYMBOL(__filemap_get_folio);
2027 static inline struct folio *find_get_entry(struct xa_state *xas, pgoff_t max,
2030 struct folio *folio;
2033 if (mark == XA_PRESENT)
2034 folio = xas_find(xas, max);
2036 folio = xas_find_marked(xas, max, mark);
2038 if (xas_retry(xas, folio))
2041 * A shadow entry of a recently evicted page, a swap
2042 * entry from shmem/tmpfs or a DAX entry. Return it
2043 * without attempting to raise page count.
2045 if (!folio || xa_is_value(folio))
2048 if (!folio_try_get_rcu(folio))
2051 if (unlikely(folio != xas_reload(xas))) {
2063 * find_get_entries - gang pagecache lookup
2064 * @mapping: The address_space to search
2065 * @start: The starting page cache index
2066 * @end: The final page index (inclusive).
2067 * @fbatch: Where the resulting entries are placed.
2068 * @indices: The cache indices corresponding to the entries in @entries
2070 * find_get_entries() will search for and return a batch of entries in
2071 * the mapping. The entries are placed in @fbatch. find_get_entries()
2072 * takes a reference on any actual folios it returns.
2074 * The entries have ascending indexes. The indices may not be consecutive
2075 * due to not-present entries or large folios.
2077 * Any shadow entries of evicted folios, or swap entries from
2078 * shmem/tmpfs, are included in the returned array.
2080 * Return: The number of entries which were found.
2082 unsigned find_get_entries(struct address_space *mapping, pgoff_t start,
2083 pgoff_t end, struct folio_batch *fbatch, pgoff_t *indices)
2085 XA_STATE(xas, &mapping->i_pages, start);
2086 struct folio *folio;
2089 while ((folio = find_get_entry(&xas, end, XA_PRESENT)) != NULL) {
2090 indices[fbatch->nr] = xas.xa_index;
2091 if (!folio_batch_add(fbatch, folio))
2096 return folio_batch_count(fbatch);
2100 * find_lock_entries - Find a batch of pagecache entries.
2101 * @mapping: The address_space to search.
2102 * @start: The starting page cache index.
2103 * @end: The final page index (inclusive).
2104 * @fbatch: Where the resulting entries are placed.
2105 * @indices: The cache indices of the entries in @fbatch.
2107 * find_lock_entries() will return a batch of entries from @mapping.
2108 * Swap, shadow and DAX entries are included. Folios are returned
2109 * locked and with an incremented refcount. Folios which are locked
2110 * by somebody else or under writeback are skipped. Folios which are
2111 * partially outside the range are not returned.
2113 * The entries have ascending indexes. The indices may not be consecutive
2114 * due to not-present entries, large folios, folios which could not be
2115 * locked or folios under writeback.
2117 * Return: The number of entries which were found.
2119 unsigned find_lock_entries(struct address_space *mapping, pgoff_t start,
2120 pgoff_t end, struct folio_batch *fbatch, pgoff_t *indices)
2122 XA_STATE(xas, &mapping->i_pages, start);
2123 struct folio *folio;
2126 while ((folio = find_get_entry(&xas, end, XA_PRESENT))) {
2127 if (!xa_is_value(folio)) {
2128 if (folio->index < start)
2130 if (folio->index + folio_nr_pages(folio) - 1 > end)
2132 if (!folio_trylock(folio))
2134 if (folio->mapping != mapping ||
2135 folio_test_writeback(folio))
2137 VM_BUG_ON_FOLIO(!folio_contains(folio, xas.xa_index),
2140 indices[fbatch->nr] = xas.xa_index;
2141 if (!folio_batch_add(fbatch, folio))
2145 folio_unlock(folio);
2151 return folio_batch_count(fbatch);
2155 bool folio_more_pages(struct folio *folio, pgoff_t index, pgoff_t max)
2157 if (!folio_test_large(folio) || folio_test_hugetlb(folio))
2161 return index < folio->index + folio_nr_pages(folio) - 1;
2165 * find_get_pages_range - gang pagecache lookup
2166 * @mapping: The address_space to search
2167 * @start: The starting page index
2168 * @end: The final page index (inclusive)
2169 * @nr_pages: The maximum number of pages
2170 * @pages: Where the resulting pages are placed
2172 * find_get_pages_range() will search for and return a group of up to @nr_pages
2173 * pages in the mapping starting at index @start and up to index @end
2174 * (inclusive). The pages are placed at @pages. find_get_pages_range() takes
2175 * a reference against the returned pages.
2177 * The search returns a group of mapping-contiguous pages with ascending
2178 * indexes. There may be holes in the indices due to not-present pages.
2179 * We also update @start to index the next page for the traversal.
2181 * Return: the number of pages which were found. If this number is
2182 * smaller than @nr_pages, the end of specified range has been
2185 unsigned find_get_pages_range(struct address_space *mapping, pgoff_t *start,
2186 pgoff_t end, unsigned int nr_pages,
2187 struct page **pages)
2189 XA_STATE(xas, &mapping->i_pages, *start);
2190 struct folio *folio;
2193 if (unlikely(!nr_pages))
2197 while ((folio = find_get_entry(&xas, end, XA_PRESENT))) {
2198 /* Skip over shadow, swap and DAX entries */
2199 if (xa_is_value(folio))
2203 pages[ret] = folio_file_page(folio, xas.xa_index);
2204 if (++ret == nr_pages) {
2205 *start = xas.xa_index + 1;
2208 if (folio_more_pages(folio, xas.xa_index, end)) {
2210 folio_ref_inc(folio);
2216 * We come here when there is no page beyond @end. We take care to not
2217 * overflow the index @start as it confuses some of the callers. This
2218 * breaks the iteration when there is a page at index -1 but that is
2219 * already broken anyway.
2221 if (end == (pgoff_t)-1)
2222 *start = (pgoff_t)-1;
2232 * find_get_pages_contig - gang contiguous pagecache lookup
2233 * @mapping: The address_space to search
2234 * @index: The starting page index
2235 * @nr_pages: The maximum number of pages
2236 * @pages: Where the resulting pages are placed
2238 * find_get_pages_contig() works exactly like find_get_pages_range(),
2239 * except that the returned number of pages are guaranteed to be
2242 * Return: the number of pages which were found.
2244 unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index,
2245 unsigned int nr_pages, struct page **pages)
2247 XA_STATE(xas, &mapping->i_pages, index);
2248 struct folio *folio;
2249 unsigned int ret = 0;
2251 if (unlikely(!nr_pages))
2255 for (folio = xas_load(&xas); folio; folio = xas_next(&xas)) {
2256 if (xas_retry(&xas, folio))
2259 * If the entry has been swapped out, we can stop looking.
2260 * No current caller is looking for DAX entries.
2262 if (xa_is_value(folio))
2265 if (!folio_try_get_rcu(folio))
2268 if (unlikely(folio != xas_reload(&xas)))
2272 pages[ret] = folio_file_page(folio, xas.xa_index);
2273 if (++ret == nr_pages)
2275 if (folio_more_pages(folio, xas.xa_index, ULONG_MAX)) {
2277 folio_ref_inc(folio);
2289 EXPORT_SYMBOL(find_get_pages_contig);
2292 * find_get_pages_range_tag - Find and return head pages matching @tag.
2293 * @mapping: the address_space to search
2294 * @index: the starting page index
2295 * @end: The final page index (inclusive)
2296 * @tag: the tag index
2297 * @nr_pages: the maximum number of pages
2298 * @pages: where the resulting pages are placed
2300 * Like find_get_pages_range(), except we only return head pages which are
2301 * tagged with @tag. @index is updated to the index immediately after the
2302 * last page we return, ready for the next iteration.
2304 * Return: the number of pages which were found.
2306 unsigned find_get_pages_range_tag(struct address_space *mapping, pgoff_t *index,
2307 pgoff_t end, xa_mark_t tag, unsigned int nr_pages,
2308 struct page **pages)
2310 XA_STATE(xas, &mapping->i_pages, *index);
2311 struct folio *folio;
2314 if (unlikely(!nr_pages))
2318 while ((folio = find_get_entry(&xas, end, tag))) {
2320 * Shadow entries should never be tagged, but this iteration
2321 * is lockless so there is a window for page reclaim to evict
2322 * a page we saw tagged. Skip over it.
2324 if (xa_is_value(folio))
2327 pages[ret] = &folio->page;
2328 if (++ret == nr_pages) {
2329 *index = folio->index + folio_nr_pages(folio);
2335 * We come here when we got to @end. We take care to not overflow the
2336 * index @index as it confuses some of the callers. This breaks the
2337 * iteration when there is a page at index -1 but that is already
2340 if (end == (pgoff_t)-1)
2341 *index = (pgoff_t)-1;
2349 EXPORT_SYMBOL(find_get_pages_range_tag);
2352 * CD/DVDs are error prone. When a medium error occurs, the driver may fail
2353 * a _large_ part of the i/o request. Imagine the worst scenario:
2355 * ---R__________________________________________B__________
2356 * ^ reading here ^ bad block(assume 4k)
2358 * read(R) => miss => readahead(R...B) => media error => frustrating retries
2359 * => failing the whole request => read(R) => read(R+1) =>
2360 * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
2361 * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
2362 * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
2364 * It is going insane. Fix it by quickly scaling down the readahead size.
2366 static void shrink_readahead_size_eio(struct file_ra_state *ra)
2372 * filemap_get_read_batch - Get a batch of folios for read
2374 * Get a batch of folios which represent a contiguous range of bytes in
2375 * the file. No exceptional entries will be returned. If @index is in
2376 * the middle of a folio, the entire folio will be returned. The last
2377 * folio in the batch may have the readahead flag set or the uptodate flag
2378 * clear so that the caller can take the appropriate action.
2380 static void filemap_get_read_batch(struct address_space *mapping,
2381 pgoff_t index, pgoff_t max, struct folio_batch *fbatch)
2383 XA_STATE(xas, &mapping->i_pages, index);
2384 struct folio *folio;
2387 for (folio = xas_load(&xas); folio; folio = xas_next(&xas)) {
2388 if (xas_retry(&xas, folio))
2390 if (xas.xa_index > max || xa_is_value(folio))
2392 if (!folio_try_get_rcu(folio))
2395 if (unlikely(folio != xas_reload(&xas)))
2398 if (!folio_batch_add(fbatch, folio))
2400 if (!folio_test_uptodate(folio))
2402 if (folio_test_readahead(folio))
2404 xas_advance(&xas, folio->index + folio_nr_pages(folio) - 1);
2414 static int filemap_read_folio(struct file *file, struct address_space *mapping,
2415 struct folio *folio)
2420 * A previous I/O error may have been due to temporary failures,
2421 * eg. multipath errors. PG_error will be set again if readpage
2424 folio_clear_error(folio);
2425 /* Start the actual read. The read will unlock the page. */
2426 error = mapping->a_ops->readpage(file, &folio->page);
2430 error = folio_wait_locked_killable(folio);
2433 if (folio_test_uptodate(folio))
2435 shrink_readahead_size_eio(&file->f_ra);
2439 static bool filemap_range_uptodate(struct address_space *mapping,
2440 loff_t pos, struct iov_iter *iter, struct folio *folio)
2444 if (folio_test_uptodate(folio))
2446 /* pipes can't handle partially uptodate pages */
2447 if (iov_iter_is_pipe(iter))
2449 if (!mapping->a_ops->is_partially_uptodate)
2451 if (mapping->host->i_blkbits >= folio_shift(folio))
2454 count = iter->count;
2455 if (folio_pos(folio) > pos) {
2456 count -= folio_pos(folio) - pos;
2459 pos -= folio_pos(folio);
2462 return mapping->a_ops->is_partially_uptodate(&folio->page, pos, count);
2465 static int filemap_update_page(struct kiocb *iocb,
2466 struct address_space *mapping, struct iov_iter *iter,
2467 struct folio *folio)
2471 if (iocb->ki_flags & IOCB_NOWAIT) {
2472 if (!filemap_invalidate_trylock_shared(mapping))
2475 filemap_invalidate_lock_shared(mapping);
2478 if (!folio_trylock(folio)) {
2480 if (iocb->ki_flags & (IOCB_NOWAIT | IOCB_NOIO))
2481 goto unlock_mapping;
2482 if (!(iocb->ki_flags & IOCB_WAITQ)) {
2483 filemap_invalidate_unlock_shared(mapping);
2485 * This is where we usually end up waiting for a
2486 * previously submitted readahead to finish.
2488 folio_put_wait_locked(folio, TASK_KILLABLE);
2489 return AOP_TRUNCATED_PAGE;
2491 error = __folio_lock_async(folio, iocb->ki_waitq);
2493 goto unlock_mapping;
2496 error = AOP_TRUNCATED_PAGE;
2497 if (!folio->mapping)
2501 if (filemap_range_uptodate(mapping, iocb->ki_pos, iter, folio))
2505 if (iocb->ki_flags & (IOCB_NOIO | IOCB_NOWAIT | IOCB_WAITQ))
2508 error = filemap_read_folio(iocb->ki_filp, mapping, folio);
2509 goto unlock_mapping;
2511 folio_unlock(folio);
2513 filemap_invalidate_unlock_shared(mapping);
2514 if (error == AOP_TRUNCATED_PAGE)
2519 static int filemap_create_folio(struct file *file,
2520 struct address_space *mapping, pgoff_t index,
2521 struct folio_batch *fbatch)
2523 struct folio *folio;
2526 folio = filemap_alloc_folio(mapping_gfp_mask(mapping), 0);
2531 * Protect against truncate / hole punch. Grabbing invalidate_lock
2532 * here assures we cannot instantiate and bring uptodate new
2533 * pagecache folios after evicting page cache during truncate
2534 * and before actually freeing blocks. Note that we could
2535 * release invalidate_lock after inserting the folio into
2536 * the page cache as the locked folio would then be enough to
2537 * synchronize with hole punching. But there are code paths
2538 * such as filemap_update_page() filling in partially uptodate
2539 * pages or ->readpages() that need to hold invalidate_lock
2540 * while mapping blocks for IO so let's hold the lock here as
2541 * well to keep locking rules simple.
2543 filemap_invalidate_lock_shared(mapping);
2544 error = filemap_add_folio(mapping, folio, index,
2545 mapping_gfp_constraint(mapping, GFP_KERNEL));
2546 if (error == -EEXIST)
2547 error = AOP_TRUNCATED_PAGE;
2551 error = filemap_read_folio(file, mapping, folio);
2555 filemap_invalidate_unlock_shared(mapping);
2556 folio_batch_add(fbatch, folio);
2559 filemap_invalidate_unlock_shared(mapping);
2564 static int filemap_readahead(struct kiocb *iocb, struct file *file,
2565 struct address_space *mapping, struct folio *folio,
2568 DEFINE_READAHEAD(ractl, file, &file->f_ra, mapping, folio->index);
2570 if (iocb->ki_flags & IOCB_NOIO)
2572 page_cache_async_ra(&ractl, folio, last_index - folio->index);
2576 static int filemap_get_pages(struct kiocb *iocb, struct iov_iter *iter,
2577 struct folio_batch *fbatch)
2579 struct file *filp = iocb->ki_filp;
2580 struct address_space *mapping = filp->f_mapping;
2581 struct file_ra_state *ra = &filp->f_ra;
2582 pgoff_t index = iocb->ki_pos >> PAGE_SHIFT;
2584 struct folio *folio;
2587 last_index = DIV_ROUND_UP(iocb->ki_pos + iter->count, PAGE_SIZE);
2589 if (fatal_signal_pending(current))
2592 filemap_get_read_batch(mapping, index, last_index, fbatch);
2593 if (!folio_batch_count(fbatch)) {
2594 if (iocb->ki_flags & IOCB_NOIO)
2596 page_cache_sync_readahead(mapping, ra, filp, index,
2597 last_index - index);
2598 filemap_get_read_batch(mapping, index, last_index, fbatch);
2600 if (!folio_batch_count(fbatch)) {
2601 if (iocb->ki_flags & (IOCB_NOWAIT | IOCB_WAITQ))
2603 err = filemap_create_folio(filp, mapping,
2604 iocb->ki_pos >> PAGE_SHIFT, fbatch);
2605 if (err == AOP_TRUNCATED_PAGE)
2610 folio = fbatch->folios[folio_batch_count(fbatch) - 1];
2611 if (folio_test_readahead(folio)) {
2612 err = filemap_readahead(iocb, filp, mapping, folio, last_index);
2616 if (!folio_test_uptodate(folio)) {
2617 if ((iocb->ki_flags & IOCB_WAITQ) &&
2618 folio_batch_count(fbatch) > 1)
2619 iocb->ki_flags |= IOCB_NOWAIT;
2620 err = filemap_update_page(iocb, mapping, iter, folio);
2629 if (likely(--fbatch->nr))
2631 if (err == AOP_TRUNCATED_PAGE)
2637 * filemap_read - Read data from the page cache.
2638 * @iocb: The iocb to read.
2639 * @iter: Destination for the data.
2640 * @already_read: Number of bytes already read by the caller.
2642 * Copies data from the page cache. If the data is not currently present,
2643 * uses the readahead and readpage address_space operations to fetch it.
2645 * Return: Total number of bytes copied, including those already read by
2646 * the caller. If an error happens before any bytes are copied, returns
2647 * a negative error number.
2649 ssize_t filemap_read(struct kiocb *iocb, struct iov_iter *iter,
2650 ssize_t already_read)
2652 struct file *filp = iocb->ki_filp;
2653 struct file_ra_state *ra = &filp->f_ra;
2654 struct address_space *mapping = filp->f_mapping;
2655 struct inode *inode = mapping->host;
2656 struct folio_batch fbatch;
2658 bool writably_mapped;
2659 loff_t isize, end_offset;
2661 if (unlikely(iocb->ki_pos >= inode->i_sb->s_maxbytes))
2663 if (unlikely(!iov_iter_count(iter)))
2666 iov_iter_truncate(iter, inode->i_sb->s_maxbytes);
2667 folio_batch_init(&fbatch);
2673 * If we've already successfully copied some data, then we
2674 * can no longer safely return -EIOCBQUEUED. Hence mark
2675 * an async read NOWAIT at that point.
2677 if ((iocb->ki_flags & IOCB_WAITQ) && already_read)
2678 iocb->ki_flags |= IOCB_NOWAIT;
2680 if (unlikely(iocb->ki_pos >= i_size_read(inode)))
2683 error = filemap_get_pages(iocb, iter, &fbatch);
2688 * i_size must be checked after we know the pages are Uptodate.
2690 * Checking i_size after the check allows us to calculate
2691 * the correct value for "nr", which means the zero-filled
2692 * part of the page is not copied back to userspace (unless
2693 * another truncate extends the file - this is desired though).
2695 isize = i_size_read(inode);
2696 if (unlikely(iocb->ki_pos >= isize))
2698 end_offset = min_t(loff_t, isize, iocb->ki_pos + iter->count);
2701 * Once we start copying data, we don't want to be touching any
2702 * cachelines that might be contended:
2704 writably_mapped = mapping_writably_mapped(mapping);
2707 * When a sequential read accesses a page several times, only
2708 * mark it as accessed the first time.
2710 if (iocb->ki_pos >> PAGE_SHIFT !=
2711 ra->prev_pos >> PAGE_SHIFT)
2712 folio_mark_accessed(fbatch.folios[0]);
2714 for (i = 0; i < folio_batch_count(&fbatch); i++) {
2715 struct folio *folio = fbatch.folios[i];
2716 size_t fsize = folio_size(folio);
2717 size_t offset = iocb->ki_pos & (fsize - 1);
2718 size_t bytes = min_t(loff_t, end_offset - iocb->ki_pos,
2722 if (end_offset < folio_pos(folio))
2725 folio_mark_accessed(folio);
2727 * If users can be writing to this folio using arbitrary
2728 * virtual addresses, take care of potential aliasing
2729 * before reading the folio on the kernel side.
2731 if (writably_mapped)
2732 flush_dcache_folio(folio);
2734 copied = copy_folio_to_iter(folio, offset, bytes, iter);
2736 already_read += copied;
2737 iocb->ki_pos += copied;
2738 ra->prev_pos = iocb->ki_pos;
2740 if (copied < bytes) {
2746 for (i = 0; i < folio_batch_count(&fbatch); i++)
2747 folio_put(fbatch.folios[i]);
2748 folio_batch_init(&fbatch);
2749 } while (iov_iter_count(iter) && iocb->ki_pos < isize && !error);
2751 file_accessed(filp);
2753 return already_read ? already_read : error;
2755 EXPORT_SYMBOL_GPL(filemap_read);
2758 * generic_file_read_iter - generic filesystem read routine
2759 * @iocb: kernel I/O control block
2760 * @iter: destination for the data read
2762 * This is the "read_iter()" routine for all filesystems
2763 * that can use the page cache directly.
2765 * The IOCB_NOWAIT flag in iocb->ki_flags indicates that -EAGAIN shall
2766 * be returned when no data can be read without waiting for I/O requests
2767 * to complete; it doesn't prevent readahead.
2769 * The IOCB_NOIO flag in iocb->ki_flags indicates that no new I/O
2770 * requests shall be made for the read or for readahead. When no data
2771 * can be read, -EAGAIN shall be returned. When readahead would be
2772 * triggered, a partial, possibly empty read shall be returned.
2775 * * number of bytes copied, even for partial reads
2776 * * negative error code (or 0 if IOCB_NOIO) if nothing was read
2779 generic_file_read_iter(struct kiocb *iocb, struct iov_iter *iter)
2781 size_t count = iov_iter_count(iter);
2785 return 0; /* skip atime */
2787 if (iocb->ki_flags & IOCB_DIRECT) {
2788 struct file *file = iocb->ki_filp;
2789 struct address_space *mapping = file->f_mapping;
2790 struct inode *inode = mapping->host;
2792 if (iocb->ki_flags & IOCB_NOWAIT) {
2793 if (filemap_range_needs_writeback(mapping, iocb->ki_pos,
2794 iocb->ki_pos + count - 1))
2797 retval = filemap_write_and_wait_range(mapping,
2799 iocb->ki_pos + count - 1);
2804 file_accessed(file);
2806 retval = mapping->a_ops->direct_IO(iocb, iter);
2808 iocb->ki_pos += retval;
2811 if (retval != -EIOCBQUEUED)
2812 iov_iter_revert(iter, count - iov_iter_count(iter));
2815 * Btrfs can have a short DIO read if we encounter
2816 * compressed extents, so if there was an error, or if
2817 * we've already read everything we wanted to, or if
2818 * there was a short read because we hit EOF, go ahead
2819 * and return. Otherwise fallthrough to buffered io for
2820 * the rest of the read. Buffered reads will not work for
2821 * DAX files, so don't bother trying.
2823 if (retval < 0 || !count || IS_DAX(inode))
2825 if (iocb->ki_pos >= i_size_read(inode))
2829 return filemap_read(iocb, iter, retval);
2831 EXPORT_SYMBOL(generic_file_read_iter);
2833 static inline loff_t folio_seek_hole_data(struct xa_state *xas,
2834 struct address_space *mapping, struct folio *folio,
2835 loff_t start, loff_t end, bool seek_data)
2837 const struct address_space_operations *ops = mapping->a_ops;
2838 size_t offset, bsz = i_blocksize(mapping->host);
2840 if (xa_is_value(folio) || folio_test_uptodate(folio))
2841 return seek_data ? start : end;
2842 if (!ops->is_partially_uptodate)
2843 return seek_data ? end : start;
2848 if (unlikely(folio->mapping != mapping))
2851 offset = offset_in_folio(folio, start) & ~(bsz - 1);
2854 if (ops->is_partially_uptodate(&folio->page, offset, bsz) ==
2857 start = (start + bsz) & ~(bsz - 1);
2859 } while (offset < folio_size(folio));
2861 folio_unlock(folio);
2866 static inline size_t seek_folio_size(struct xa_state *xas, struct folio *folio)
2868 if (xa_is_value(folio))
2869 return PAGE_SIZE << xa_get_order(xas->xa, xas->xa_index);
2870 return folio_size(folio);
2874 * mapping_seek_hole_data - Seek for SEEK_DATA / SEEK_HOLE in the page cache.
2875 * @mapping: Address space to search.
2876 * @start: First byte to consider.
2877 * @end: Limit of search (exclusive).
2878 * @whence: Either SEEK_HOLE or SEEK_DATA.
2880 * If the page cache knows which blocks contain holes and which blocks
2881 * contain data, your filesystem can use this function to implement
2882 * SEEK_HOLE and SEEK_DATA. This is useful for filesystems which are
2883 * entirely memory-based such as tmpfs, and filesystems which support
2884 * unwritten extents.
2886 * Return: The requested offset on success, or -ENXIO if @whence specifies
2887 * SEEK_DATA and there is no data after @start. There is an implicit hole
2888 * after @end - 1, so SEEK_HOLE returns @end if all the bytes between @start
2889 * and @end contain data.
2891 loff_t mapping_seek_hole_data(struct address_space *mapping, loff_t start,
2892 loff_t end, int whence)
2894 XA_STATE(xas, &mapping->i_pages, start >> PAGE_SHIFT);
2895 pgoff_t max = (end - 1) >> PAGE_SHIFT;
2896 bool seek_data = (whence == SEEK_DATA);
2897 struct folio *folio;
2903 while ((folio = find_get_entry(&xas, max, XA_PRESENT))) {
2904 loff_t pos = (u64)xas.xa_index << PAGE_SHIFT;
2913 seek_size = seek_folio_size(&xas, folio);
2914 pos = round_up((u64)pos + 1, seek_size);
2915 start = folio_seek_hole_data(&xas, mapping, folio, start, pos,
2921 if (seek_size > PAGE_SIZE)
2922 xas_set(&xas, pos >> PAGE_SHIFT);
2923 if (!xa_is_value(folio))
2930 if (folio && !xa_is_value(folio))
2938 #define MMAP_LOTSAMISS (100)
2940 * lock_folio_maybe_drop_mmap - lock the page, possibly dropping the mmap_lock
2941 * @vmf - the vm_fault for this fault.
2942 * @folio - the folio to lock.
2943 * @fpin - the pointer to the file we may pin (or is already pinned).
2945 * This works similar to lock_folio_or_retry in that it can drop the
2946 * mmap_lock. It differs in that it actually returns the folio locked
2947 * if it returns 1 and 0 if it couldn't lock the folio. If we did have
2948 * to drop the mmap_lock then fpin will point to the pinned file and
2949 * needs to be fput()'ed at a later point.
2951 static int lock_folio_maybe_drop_mmap(struct vm_fault *vmf, struct folio *folio,
2954 if (folio_trylock(folio))
2958 * NOTE! This will make us return with VM_FAULT_RETRY, but with
2959 * the mmap_lock still held. That's how FAULT_FLAG_RETRY_NOWAIT
2960 * is supposed to work. We have way too many special cases..
2962 if (vmf->flags & FAULT_FLAG_RETRY_NOWAIT)
2965 *fpin = maybe_unlock_mmap_for_io(vmf, *fpin);
2966 if (vmf->flags & FAULT_FLAG_KILLABLE) {
2967 if (__folio_lock_killable(folio)) {
2969 * We didn't have the right flags to drop the mmap_lock,
2970 * but all fault_handlers only check for fatal signals
2971 * if we return VM_FAULT_RETRY, so we need to drop the
2972 * mmap_lock here and return 0 if we don't have a fpin.
2975 mmap_read_unlock(vmf->vma->vm_mm);
2979 __folio_lock(folio);
2985 * Synchronous readahead happens when we don't even find a page in the page
2986 * cache at all. We don't want to perform IO under the mmap sem, so if we have
2987 * to drop the mmap sem we return the file that was pinned in order for us to do
2988 * that. If we didn't pin a file then we return NULL. The file that is
2989 * returned needs to be fput()'ed when we're done with it.
2991 static struct file *do_sync_mmap_readahead(struct vm_fault *vmf)
2993 struct file *file = vmf->vma->vm_file;
2994 struct file_ra_state *ra = &file->f_ra;
2995 struct address_space *mapping = file->f_mapping;
2996 DEFINE_READAHEAD(ractl, file, ra, mapping, vmf->pgoff);
2997 struct file *fpin = NULL;
2998 unsigned int mmap_miss;
3000 /* If we don't want any read-ahead, don't bother */
3001 if (vmf->vma->vm_flags & VM_RAND_READ)
3006 if (vmf->vma->vm_flags & VM_SEQ_READ) {
3007 fpin = maybe_unlock_mmap_for_io(vmf, fpin);
3008 page_cache_sync_ra(&ractl, ra->ra_pages);
3012 /* Avoid banging the cache line if not needed */
3013 mmap_miss = READ_ONCE(ra->mmap_miss);
3014 if (mmap_miss < MMAP_LOTSAMISS * 10)
3015 WRITE_ONCE(ra->mmap_miss, ++mmap_miss);
3018 * Do we miss much more than hit in this file? If so,
3019 * stop bothering with read-ahead. It will only hurt.
3021 if (mmap_miss > MMAP_LOTSAMISS)
3027 fpin = maybe_unlock_mmap_for_io(vmf, fpin);
3028 ra->start = max_t(long, 0, vmf->pgoff - ra->ra_pages / 2);
3029 ra->size = ra->ra_pages;
3030 ra->async_size = ra->ra_pages / 4;
3031 ractl._index = ra->start;
3032 do_page_cache_ra(&ractl, ra->size, ra->async_size);
3037 * Asynchronous readahead happens when we find the page and PG_readahead,
3038 * so we want to possibly extend the readahead further. We return the file that
3039 * was pinned if we have to drop the mmap_lock in order to do IO.
3041 static struct file *do_async_mmap_readahead(struct vm_fault *vmf,
3042 struct folio *folio)
3044 struct file *file = vmf->vma->vm_file;
3045 struct file_ra_state *ra = &file->f_ra;
3046 DEFINE_READAHEAD(ractl, file, ra, file->f_mapping, vmf->pgoff);
3047 struct file *fpin = NULL;
3048 unsigned int mmap_miss;
3050 /* If we don't want any read-ahead, don't bother */
3051 if (vmf->vma->vm_flags & VM_RAND_READ || !ra->ra_pages)
3054 mmap_miss = READ_ONCE(ra->mmap_miss);
3056 WRITE_ONCE(ra->mmap_miss, --mmap_miss);
3058 if (folio_test_readahead(folio)) {
3059 fpin = maybe_unlock_mmap_for_io(vmf, fpin);
3060 page_cache_async_ra(&ractl, folio, ra->ra_pages);
3066 * filemap_fault - read in file data for page fault handling
3067 * @vmf: struct vm_fault containing details of the fault
3069 * filemap_fault() is invoked via the vma operations vector for a
3070 * mapped memory region to read in file data during a page fault.
3072 * The goto's are kind of ugly, but this streamlines the normal case of having
3073 * it in the page cache, and handles the special cases reasonably without
3074 * having a lot of duplicated code.
3076 * vma->vm_mm->mmap_lock must be held on entry.
3078 * If our return value has VM_FAULT_RETRY set, it's because the mmap_lock
3079 * may be dropped before doing I/O or by lock_folio_maybe_drop_mmap().
3081 * If our return value does not have VM_FAULT_RETRY set, the mmap_lock
3082 * has not been released.
3084 * We never return with VM_FAULT_RETRY and a bit from VM_FAULT_ERROR set.
3086 * Return: bitwise-OR of %VM_FAULT_ codes.
3088 vm_fault_t filemap_fault(struct vm_fault *vmf)
3091 struct file *file = vmf->vma->vm_file;
3092 struct file *fpin = NULL;
3093 struct address_space *mapping = file->f_mapping;
3094 struct inode *inode = mapping->host;
3095 pgoff_t max_idx, index = vmf->pgoff;
3096 struct folio *folio;
3098 bool mapping_locked = false;
3100 max_idx = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
3101 if (unlikely(index >= max_idx))
3102 return VM_FAULT_SIGBUS;
3105 * Do we have something in the page cache already?
3107 folio = filemap_get_folio(mapping, index);
3108 if (likely(folio)) {
3110 * We found the page, so try async readahead before waiting for
3113 if (!(vmf->flags & FAULT_FLAG_TRIED))
3114 fpin = do_async_mmap_readahead(vmf, folio);
3115 if (unlikely(!folio_test_uptodate(folio))) {
3116 filemap_invalidate_lock_shared(mapping);
3117 mapping_locked = true;
3120 /* No page in the page cache at all */
3121 count_vm_event(PGMAJFAULT);
3122 count_memcg_event_mm(vmf->vma->vm_mm, PGMAJFAULT);
3123 ret = VM_FAULT_MAJOR;
3124 fpin = do_sync_mmap_readahead(vmf);
3127 * See comment in filemap_create_folio() why we need
3130 if (!mapping_locked) {
3131 filemap_invalidate_lock_shared(mapping);
3132 mapping_locked = true;
3134 folio = __filemap_get_folio(mapping, index,
3135 FGP_CREAT|FGP_FOR_MMAP,
3140 filemap_invalidate_unlock_shared(mapping);
3141 return VM_FAULT_OOM;
3145 if (!lock_folio_maybe_drop_mmap(vmf, folio, &fpin))
3148 /* Did it get truncated? */
3149 if (unlikely(folio->mapping != mapping)) {
3150 folio_unlock(folio);
3154 VM_BUG_ON_FOLIO(!folio_contains(folio, index), folio);
3157 * We have a locked page in the page cache, now we need to check
3158 * that it's up-to-date. If not, it is going to be due to an error.
3160 if (unlikely(!folio_test_uptodate(folio))) {
3162 * The page was in cache and uptodate and now it is not.
3163 * Strange but possible since we didn't hold the page lock all
3164 * the time. Let's drop everything get the invalidate lock and
3167 if (!mapping_locked) {
3168 folio_unlock(folio);
3172 goto page_not_uptodate;
3176 * We've made it this far and we had to drop our mmap_lock, now is the
3177 * time to return to the upper layer and have it re-find the vma and
3181 folio_unlock(folio);
3185 filemap_invalidate_unlock_shared(mapping);
3188 * Found the page and have a reference on it.
3189 * We must recheck i_size under page lock.
3191 max_idx = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
3192 if (unlikely(index >= max_idx)) {
3193 folio_unlock(folio);
3195 return VM_FAULT_SIGBUS;
3198 vmf->page = folio_file_page(folio, index);
3199 return ret | VM_FAULT_LOCKED;
3203 * Umm, take care of errors if the page isn't up-to-date.
3204 * Try to re-read it _once_. We do this synchronously,
3205 * because there really aren't any performance issues here
3206 * and we need to check for errors.
3208 fpin = maybe_unlock_mmap_for_io(vmf, fpin);
3209 error = filemap_read_folio(file, mapping, folio);
3214 if (!error || error == AOP_TRUNCATED_PAGE)
3216 filemap_invalidate_unlock_shared(mapping);
3218 return VM_FAULT_SIGBUS;
3222 * We dropped the mmap_lock, we need to return to the fault handler to
3223 * re-find the vma and come back and find our hopefully still populated
3229 filemap_invalidate_unlock_shared(mapping);
3232 return ret | VM_FAULT_RETRY;
3234 EXPORT_SYMBOL(filemap_fault);
3236 static bool filemap_map_pmd(struct vm_fault *vmf, struct page *page)
3238 struct mm_struct *mm = vmf->vma->vm_mm;
3240 /* Huge page is mapped? No need to proceed. */
3241 if (pmd_trans_huge(*vmf->pmd)) {
3247 if (pmd_none(*vmf->pmd) && PageTransHuge(page)) {
3248 vm_fault_t ret = do_set_pmd(vmf, page);
3250 /* The page is mapped successfully, reference consumed. */
3256 if (pmd_none(*vmf->pmd))
3257 pmd_install(mm, vmf->pmd, &vmf->prealloc_pte);
3259 /* See comment in handle_pte_fault() */
3260 if (pmd_devmap_trans_unstable(vmf->pmd)) {
3269 static struct folio *next_uptodate_page(struct folio *folio,
3270 struct address_space *mapping,
3271 struct xa_state *xas, pgoff_t end_pgoff)
3273 unsigned long max_idx;
3278 if (xas_retry(xas, folio))
3280 if (xa_is_value(folio))
3282 if (folio_test_locked(folio))
3284 if (!folio_try_get_rcu(folio))
3286 /* Has the page moved or been split? */
3287 if (unlikely(folio != xas_reload(xas)))
3289 if (!folio_test_uptodate(folio) || folio_test_readahead(folio))
3291 if (!folio_trylock(folio))
3293 if (folio->mapping != mapping)
3295 if (!folio_test_uptodate(folio))
3297 max_idx = DIV_ROUND_UP(i_size_read(mapping->host), PAGE_SIZE);
3298 if (xas->xa_index >= max_idx)
3302 folio_unlock(folio);
3305 } while ((folio = xas_next_entry(xas, end_pgoff)) != NULL);
3310 static inline struct folio *first_map_page(struct address_space *mapping,
3311 struct xa_state *xas,
3314 return next_uptodate_page(xas_find(xas, end_pgoff),
3315 mapping, xas, end_pgoff);
3318 static inline struct folio *next_map_page(struct address_space *mapping,
3319 struct xa_state *xas,
3322 return next_uptodate_page(xas_next_entry(xas, end_pgoff),
3323 mapping, xas, end_pgoff);
3326 vm_fault_t filemap_map_pages(struct vm_fault *vmf,
3327 pgoff_t start_pgoff, pgoff_t end_pgoff)
3329 struct vm_area_struct *vma = vmf->vma;
3330 struct file *file = vma->vm_file;
3331 struct address_space *mapping = file->f_mapping;
3332 pgoff_t last_pgoff = start_pgoff;
3334 XA_STATE(xas, &mapping->i_pages, start_pgoff);
3335 struct folio *folio;
3337 unsigned int mmap_miss = READ_ONCE(file->f_ra.mmap_miss);
3341 folio = first_map_page(mapping, &xas, end_pgoff);
3345 if (filemap_map_pmd(vmf, &folio->page)) {
3346 ret = VM_FAULT_NOPAGE;
3350 addr = vma->vm_start + ((start_pgoff - vma->vm_pgoff) << PAGE_SHIFT);
3351 vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, addr, &vmf->ptl);
3354 page = folio_file_page(folio, xas.xa_index);
3355 if (PageHWPoison(page))
3361 addr += (xas.xa_index - last_pgoff) << PAGE_SHIFT;
3362 vmf->pte += xas.xa_index - last_pgoff;
3363 last_pgoff = xas.xa_index;
3365 if (!pte_none(*vmf->pte))
3368 /* We're about to handle the fault */
3369 if (vmf->address == addr)
3370 ret = VM_FAULT_NOPAGE;
3372 do_set_pte(vmf, page, addr);
3373 /* no need to invalidate: a not-present page won't be cached */
3374 update_mmu_cache(vma, addr, vmf->pte);
3375 if (folio_more_pages(folio, xas.xa_index, end_pgoff)) {
3377 folio_ref_inc(folio);
3380 folio_unlock(folio);
3383 if (folio_more_pages(folio, xas.xa_index, end_pgoff)) {
3387 folio_unlock(folio);
3389 } while ((folio = next_map_page(mapping, &xas, end_pgoff)) != NULL);
3390 pte_unmap_unlock(vmf->pte, vmf->ptl);
3393 WRITE_ONCE(file->f_ra.mmap_miss, mmap_miss);
3396 EXPORT_SYMBOL(filemap_map_pages);
3398 vm_fault_t filemap_page_mkwrite(struct vm_fault *vmf)
3400 struct address_space *mapping = vmf->vma->vm_file->f_mapping;
3401 struct folio *folio = page_folio(vmf->page);
3402 vm_fault_t ret = VM_FAULT_LOCKED;
3404 sb_start_pagefault(mapping->host->i_sb);
3405 file_update_time(vmf->vma->vm_file);
3407 if (folio->mapping != mapping) {
3408 folio_unlock(folio);
3409 ret = VM_FAULT_NOPAGE;
3413 * We mark the folio dirty already here so that when freeze is in
3414 * progress, we are guaranteed that writeback during freezing will
3415 * see the dirty folio and writeprotect it again.
3417 folio_mark_dirty(folio);
3418 folio_wait_stable(folio);
3420 sb_end_pagefault(mapping->host->i_sb);
3424 const struct vm_operations_struct generic_file_vm_ops = {
3425 .fault = filemap_fault,
3426 .map_pages = filemap_map_pages,
3427 .page_mkwrite = filemap_page_mkwrite,
3430 /* This is used for a general mmap of a disk file */
3432 int generic_file_mmap(struct file *file, struct vm_area_struct *vma)
3434 struct address_space *mapping = file->f_mapping;
3436 if (!mapping->a_ops->readpage)
3438 file_accessed(file);
3439 vma->vm_ops = &generic_file_vm_ops;
3444 * This is for filesystems which do not implement ->writepage.
3446 int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
3448 if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE))
3450 return generic_file_mmap(file, vma);
3453 vm_fault_t filemap_page_mkwrite(struct vm_fault *vmf)
3455 return VM_FAULT_SIGBUS;
3457 int generic_file_mmap(struct file *file, struct vm_area_struct *vma)
3461 int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
3465 #endif /* CONFIG_MMU */
3467 EXPORT_SYMBOL(filemap_page_mkwrite);
3468 EXPORT_SYMBOL(generic_file_mmap);
3469 EXPORT_SYMBOL(generic_file_readonly_mmap);
3471 static struct folio *do_read_cache_folio(struct address_space *mapping,
3472 pgoff_t index, filler_t filler, void *data, gfp_t gfp)
3474 struct folio *folio;
3477 folio = filemap_get_folio(mapping, index);
3479 folio = filemap_alloc_folio(gfp, 0);
3481 return ERR_PTR(-ENOMEM);
3482 err = filemap_add_folio(mapping, folio, index, gfp);
3483 if (unlikely(err)) {
3487 /* Presumably ENOMEM for xarray node */
3488 return ERR_PTR(err);
3493 err = filler(data, &folio->page);
3495 err = mapping->a_ops->readpage(data, &folio->page);
3499 return ERR_PTR(err);
3502 folio_wait_locked(folio);
3503 if (!folio_test_uptodate(folio)) {
3505 return ERR_PTR(-EIO);
3510 if (folio_test_uptodate(folio))
3513 if (!folio_trylock(folio)) {
3514 folio_put_wait_locked(folio, TASK_UNINTERRUPTIBLE);
3518 /* Folio was truncated from mapping */
3519 if (!folio->mapping) {
3520 folio_unlock(folio);
3525 /* Someone else locked and filled the page in a very small window */
3526 if (folio_test_uptodate(folio)) {
3527 folio_unlock(folio);
3532 * A previous I/O error may have been due to temporary
3534 * Clear page error before actual read, PG_error will be
3535 * set again if read page fails.
3537 folio_clear_error(folio);
3541 folio_mark_accessed(folio);
3546 * read_cache_folio - read into page cache, fill it if needed
3547 * @mapping: the page's address_space
3548 * @index: the page index
3549 * @filler: function to perform the read
3550 * @data: first arg to filler(data, page) function, often left as NULL
3552 * Read into the page cache. If a page already exists, and PageUptodate() is
3553 * not set, try to fill the page and wait for it to become unlocked.
3555 * If the page does not get brought uptodate, return -EIO.
3557 * The function expects mapping->invalidate_lock to be already held.
3559 * Return: up to date page on success, ERR_PTR() on failure.
3561 struct folio *read_cache_folio(struct address_space *mapping, pgoff_t index,
3562 filler_t filler, void *data)
3564 return do_read_cache_folio(mapping, index, filler, data,
3565 mapping_gfp_mask(mapping));
3567 EXPORT_SYMBOL(read_cache_folio);
3569 static struct page *do_read_cache_page(struct address_space *mapping,
3570 pgoff_t index, filler_t *filler, void *data, gfp_t gfp)
3572 struct folio *folio;
3574 folio = do_read_cache_folio(mapping, index, filler, data, gfp);
3576 return &folio->page;
3577 return folio_file_page(folio, index);
3580 struct page *read_cache_page(struct address_space *mapping,
3581 pgoff_t index, filler_t *filler, void *data)
3583 return do_read_cache_page(mapping, index, filler, data,
3584 mapping_gfp_mask(mapping));
3586 EXPORT_SYMBOL(read_cache_page);
3589 * read_cache_page_gfp - read into page cache, using specified page allocation flags.
3590 * @mapping: the page's address_space
3591 * @index: the page index
3592 * @gfp: the page allocator flags to use if allocating
3594 * This is the same as "read_mapping_page(mapping, index, NULL)", but with
3595 * any new page allocations done using the specified allocation flags.
3597 * If the page does not get brought uptodate, return -EIO.
3599 * The function expects mapping->invalidate_lock to be already held.
3601 * Return: up to date page on success, ERR_PTR() on failure.
3603 struct page *read_cache_page_gfp(struct address_space *mapping,
3607 return do_read_cache_page(mapping, index, NULL, NULL, gfp);
3609 EXPORT_SYMBOL(read_cache_page_gfp);
3611 int pagecache_write_begin(struct file *file, struct address_space *mapping,
3612 loff_t pos, unsigned len, unsigned flags,
3613 struct page **pagep, void **fsdata)
3615 const struct address_space_operations *aops = mapping->a_ops;
3617 return aops->write_begin(file, mapping, pos, len, flags,
3620 EXPORT_SYMBOL(pagecache_write_begin);
3622 int pagecache_write_end(struct file *file, struct address_space *mapping,
3623 loff_t pos, unsigned len, unsigned copied,
3624 struct page *page, void *fsdata)
3626 const struct address_space_operations *aops = mapping->a_ops;
3628 return aops->write_end(file, mapping, pos, len, copied, page, fsdata);
3630 EXPORT_SYMBOL(pagecache_write_end);
3633 * Warn about a page cache invalidation failure during a direct I/O write.
3635 void dio_warn_stale_pagecache(struct file *filp)
3637 static DEFINE_RATELIMIT_STATE(_rs, 86400 * HZ, DEFAULT_RATELIMIT_BURST);
3641 errseq_set(&filp->f_mapping->wb_err, -EIO);
3642 if (__ratelimit(&_rs)) {
3643 path = file_path(filp, pathname, sizeof(pathname));
3646 pr_crit("Page cache invalidation failure on direct I/O. Possible data corruption due to collision with buffered I/O!\n");
3647 pr_crit("File: %s PID: %d Comm: %.20s\n", path, current->pid,
3653 generic_file_direct_write(struct kiocb *iocb, struct iov_iter *from)
3655 struct file *file = iocb->ki_filp;
3656 struct address_space *mapping = file->f_mapping;
3657 struct inode *inode = mapping->host;
3658 loff_t pos = iocb->ki_pos;
3663 write_len = iov_iter_count(from);
3664 end = (pos + write_len - 1) >> PAGE_SHIFT;
3666 if (iocb->ki_flags & IOCB_NOWAIT) {
3667 /* If there are pages to writeback, return */
3668 if (filemap_range_has_page(file->f_mapping, pos,
3669 pos + write_len - 1))
3672 written = filemap_write_and_wait_range(mapping, pos,
3673 pos + write_len - 1);
3679 * After a write we want buffered reads to be sure to go to disk to get
3680 * the new data. We invalidate clean cached page from the region we're
3681 * about to write. We do this *before* the write so that we can return
3682 * without clobbering -EIOCBQUEUED from ->direct_IO().
3684 written = invalidate_inode_pages2_range(mapping,
3685 pos >> PAGE_SHIFT, end);
3687 * If a page can not be invalidated, return 0 to fall back
3688 * to buffered write.
3691 if (written == -EBUSY)
3696 written = mapping->a_ops->direct_IO(iocb, from);
3699 * Finally, try again to invalidate clean pages which might have been
3700 * cached by non-direct readahead, or faulted in by get_user_pages()
3701 * if the source of the write was an mmap'ed region of the file
3702 * we're writing. Either one is a pretty crazy thing to do,
3703 * so we don't support it 100%. If this invalidation
3704 * fails, tough, the write still worked...
3706 * Most of the time we do not need this since dio_complete() will do
3707 * the invalidation for us. However there are some file systems that
3708 * do not end up with dio_complete() being called, so let's not break
3709 * them by removing it completely.
3711 * Noticeable example is a blkdev_direct_IO().
3713 * Skip invalidation for async writes or if mapping has no pages.
3715 if (written > 0 && mapping->nrpages &&
3716 invalidate_inode_pages2_range(mapping, pos >> PAGE_SHIFT, end))
3717 dio_warn_stale_pagecache(file);
3721 write_len -= written;
3722 if (pos > i_size_read(inode) && !S_ISBLK(inode->i_mode)) {
3723 i_size_write(inode, pos);
3724 mark_inode_dirty(inode);
3728 if (written != -EIOCBQUEUED)
3729 iov_iter_revert(from, write_len - iov_iter_count(from));
3733 EXPORT_SYMBOL(generic_file_direct_write);
3735 ssize_t generic_perform_write(struct file *file,
3736 struct iov_iter *i, loff_t pos)
3738 struct address_space *mapping = file->f_mapping;
3739 const struct address_space_operations *a_ops = mapping->a_ops;
3741 ssize_t written = 0;
3742 unsigned int flags = 0;
3746 unsigned long offset; /* Offset into pagecache page */
3747 unsigned long bytes; /* Bytes to write to page */
3748 size_t copied; /* Bytes copied from user */
3751 offset = (pos & (PAGE_SIZE - 1));
3752 bytes = min_t(unsigned long, PAGE_SIZE - offset,
3757 * Bring in the user page that we will copy from _first_.
3758 * Otherwise there's a nasty deadlock on copying from the
3759 * same page as we're writing to, without it being marked
3762 if (unlikely(fault_in_iov_iter_readable(i, bytes))) {
3767 if (fatal_signal_pending(current)) {
3772 status = a_ops->write_begin(file, mapping, pos, bytes, flags,
3774 if (unlikely(status < 0))
3777 if (mapping_writably_mapped(mapping))
3778 flush_dcache_page(page);
3780 copied = copy_page_from_iter_atomic(page, offset, bytes, i);
3781 flush_dcache_page(page);
3783 status = a_ops->write_end(file, mapping, pos, bytes, copied,
3785 if (unlikely(status != copied)) {
3786 iov_iter_revert(i, copied - max(status, 0L));
3787 if (unlikely(status < 0))
3792 if (unlikely(status == 0)) {
3794 * A short copy made ->write_end() reject the
3795 * thing entirely. Might be memory poisoning
3796 * halfway through, might be a race with munmap,
3797 * might be severe memory pressure.
3806 balance_dirty_pages_ratelimited(mapping);
3807 } while (iov_iter_count(i));
3809 return written ? written : status;
3811 EXPORT_SYMBOL(generic_perform_write);
3814 * __generic_file_write_iter - write data to a file
3815 * @iocb: IO state structure (file, offset, etc.)
3816 * @from: iov_iter with data to write
3818 * This function does all the work needed for actually writing data to a
3819 * file. It does all basic checks, removes SUID from the file, updates
3820 * modification times and calls proper subroutines depending on whether we
3821 * do direct IO or a standard buffered write.
3823 * It expects i_rwsem to be grabbed unless we work on a block device or similar
3824 * object which does not need locking at all.
3826 * This function does *not* take care of syncing data in case of O_SYNC write.
3827 * A caller has to handle it. This is mainly due to the fact that we want to
3828 * avoid syncing under i_rwsem.
3831 * * number of bytes written, even for truncated writes
3832 * * negative error code if no data has been written at all
3834 ssize_t __generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
3836 struct file *file = iocb->ki_filp;
3837 struct address_space *mapping = file->f_mapping;
3838 struct inode *inode = mapping->host;
3839 ssize_t written = 0;
3843 /* We can write back this queue in page reclaim */
3844 current->backing_dev_info = inode_to_bdi(inode);
3845 err = file_remove_privs(file);
3849 err = file_update_time(file);
3853 if (iocb->ki_flags & IOCB_DIRECT) {
3854 loff_t pos, endbyte;
3856 written = generic_file_direct_write(iocb, from);
3858 * If the write stopped short of completing, fall back to
3859 * buffered writes. Some filesystems do this for writes to
3860 * holes, for example. For DAX files, a buffered write will
3861 * not succeed (even if it did, DAX does not handle dirty
3862 * page-cache pages correctly).
3864 if (written < 0 || !iov_iter_count(from) || IS_DAX(inode))
3867 status = generic_perform_write(file, from, pos = iocb->ki_pos);
3869 * If generic_perform_write() returned a synchronous error
3870 * then we want to return the number of bytes which were
3871 * direct-written, or the error code if that was zero. Note
3872 * that this differs from normal direct-io semantics, which
3873 * will return -EFOO even if some bytes were written.
3875 if (unlikely(status < 0)) {
3880 * We need to ensure that the page cache pages are written to
3881 * disk and invalidated to preserve the expected O_DIRECT
3884 endbyte = pos + status - 1;
3885 err = filemap_write_and_wait_range(mapping, pos, endbyte);
3887 iocb->ki_pos = endbyte + 1;
3889 invalidate_mapping_pages(mapping,
3891 endbyte >> PAGE_SHIFT);
3894 * We don't know how much we wrote, so just return
3895 * the number of bytes which were direct-written
3899 written = generic_perform_write(file, from, iocb->ki_pos);
3900 if (likely(written > 0))
3901 iocb->ki_pos += written;
3904 current->backing_dev_info = NULL;
3905 return written ? written : err;
3907 EXPORT_SYMBOL(__generic_file_write_iter);
3910 * generic_file_write_iter - write data to a file
3911 * @iocb: IO state structure
3912 * @from: iov_iter with data to write
3914 * This is a wrapper around __generic_file_write_iter() to be used by most
3915 * filesystems. It takes care of syncing the file in case of O_SYNC file
3916 * and acquires i_rwsem as needed.
3918 * * negative error code if no data has been written at all of
3919 * vfs_fsync_range() failed for a synchronous write
3920 * * number of bytes written, even for truncated writes
3922 ssize_t generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
3924 struct file *file = iocb->ki_filp;
3925 struct inode *inode = file->f_mapping->host;
3929 ret = generic_write_checks(iocb, from);
3931 ret = __generic_file_write_iter(iocb, from);
3932 inode_unlock(inode);
3935 ret = generic_write_sync(iocb, ret);
3938 EXPORT_SYMBOL(generic_file_write_iter);
3941 * filemap_release_folio() - Release fs-specific metadata on a folio.
3942 * @folio: The folio which the kernel is trying to free.
3943 * @gfp: Memory allocation flags (and I/O mode).
3945 * The address_space is trying to release any data attached to a folio
3946 * (presumably at folio->private).
3948 * This will also be called if the private_2 flag is set on a page,
3949 * indicating that the folio has other metadata associated with it.
3951 * The @gfp argument specifies whether I/O may be performed to release
3952 * this page (__GFP_IO), and whether the call may block
3953 * (__GFP_RECLAIM & __GFP_FS).
3955 * Return: %true if the release was successful, otherwise %false.
3957 bool filemap_release_folio(struct folio *folio, gfp_t gfp)
3959 struct address_space * const mapping = folio->mapping;
3961 BUG_ON(!folio_test_locked(folio));
3962 if (folio_test_writeback(folio))
3965 if (mapping && mapping->a_ops->releasepage)
3966 return mapping->a_ops->releasepage(&folio->page, gfp);
3967 return try_to_free_buffers(&folio->page);
3969 EXPORT_SYMBOL(filemap_release_folio);