1 // SPDX-License-Identifier: GPL-2.0-only
5 * Copyright (C) 1994-1999 Linus Torvalds
9 * This file handles the generic file mmap semantics used by
10 * most "normal" filesystems (but you don't /have/ to use this:
11 * the NFS filesystem used to do this differently, for example)
13 #include <linux/export.h>
14 #include <linux/compiler.h>
15 #include <linux/dax.h>
17 #include <linux/sched/signal.h>
18 #include <linux/uaccess.h>
19 #include <linux/capability.h>
20 #include <linux/kernel_stat.h>
21 #include <linux/gfp.h>
23 #include <linux/swap.h>
24 #include <linux/mman.h>
25 #include <linux/pagemap.h>
26 #include <linux/file.h>
27 #include <linux/uio.h>
28 #include <linux/error-injection.h>
29 #include <linux/hash.h>
30 #include <linux/writeback.h>
31 #include <linux/backing-dev.h>
32 #include <linux/pagevec.h>
33 #include <linux/security.h>
34 #include <linux/cpuset.h>
35 #include <linux/hugetlb.h>
36 #include <linux/memcontrol.h>
37 #include <linux/cleancache.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 <asm/pgalloc.h>
45 #include <asm/tlbflush.h>
48 #define CREATE_TRACE_POINTS
49 #include <trace/events/filemap.h>
52 * FIXME: remove all knowledge of the buffer layer from the core VM
54 #include <linux/buffer_head.h> /* for try_to_free_buffers */
59 * Shared mappings implemented 30.11.1994. It's not fully working yet,
62 * Shared mappings now work. 15.8.1995 Bruno.
64 * finished 'unifying' the page and buffer cache and SMP-threaded the
65 * page-cache, 21.05.1999, Ingo Molnar <mingo@redhat.com>
67 * SMP-threaded pagemap-LRU 1999, Andrea Arcangeli <andrea@suse.de>
73 * ->i_mmap_rwsem (truncate_pagecache)
74 * ->private_lock (__free_pte->__set_page_dirty_buffers)
75 * ->swap_lock (exclusive_swap_page, others)
79 * ->invalidate_lock (acquired by fs in truncate path)
80 * ->i_mmap_rwsem (truncate->unmap_mapping_range)
84 * ->page_table_lock or pte_lock (various, mainly in memory.c)
85 * ->i_pages lock (arch-dependent flush_dcache_mmap_lock)
88 * ->invalidate_lock (filemap_fault)
89 * ->lock_page (filemap_fault, access_process_vm)
91 * ->i_rwsem (generic_perform_write)
92 * ->mmap_lock (fault_in_readable->do_page_fault)
95 * sb_lock (fs/fs-writeback.c)
96 * ->i_pages lock (__sync_single_inode)
99 * ->anon_vma.lock (vma_adjust)
102 * ->page_table_lock or pte_lock (anon_vma_prepare and various)
104 * ->page_table_lock or pte_lock
105 * ->swap_lock (try_to_unmap_one)
106 * ->private_lock (try_to_unmap_one)
107 * ->i_pages lock (try_to_unmap_one)
108 * ->lruvec->lru_lock (follow_page->mark_page_accessed)
109 * ->lruvec->lru_lock (check_pte_range->isolate_lru_page)
110 * ->private_lock (page_remove_rmap->set_page_dirty)
111 * ->i_pages lock (page_remove_rmap->set_page_dirty)
112 * bdi.wb->list_lock (page_remove_rmap->set_page_dirty)
113 * ->inode->i_lock (page_remove_rmap->set_page_dirty)
114 * ->memcg->move_lock (page_remove_rmap->lock_page_memcg)
115 * bdi.wb->list_lock (zap_pte_range->set_page_dirty)
116 * ->inode->i_lock (zap_pte_range->set_page_dirty)
117 * ->private_lock (zap_pte_range->__set_page_dirty_buffers)
120 * ->tasklist_lock (memory_failure, collect_procs_ao)
123 static void page_cache_delete(struct address_space *mapping,
124 struct folio *folio, void *shadow)
126 XA_STATE(xas, &mapping->i_pages, folio->index);
129 mapping_set_update(&xas, mapping);
131 /* hugetlb pages are represented by a single entry in the xarray */
132 if (!folio_test_hugetlb(folio)) {
133 xas_set_order(&xas, folio->index, folio_order(folio));
134 nr = folio_nr_pages(folio);
137 VM_BUG_ON_FOLIO(!folio_test_locked(folio), folio);
139 xas_store(&xas, shadow);
140 xas_init_marks(&xas);
142 folio->mapping = NULL;
143 /* Leave page->index set: truncation lookup relies upon it */
144 mapping->nrpages -= nr;
147 static void filemap_unaccount_folio(struct address_space *mapping,
153 * if we're uptodate, flush out into the cleancache, otherwise
154 * invalidate any existing cleancache entries. We can't leave
155 * stale data around in the cleancache once our page is gone
157 if (folio_test_uptodate(folio) && folio_test_mappedtodisk(folio))
158 cleancache_put_page(&folio->page);
160 cleancache_invalidate_page(mapping, &folio->page);
162 VM_BUG_ON_FOLIO(folio_mapped(folio), folio);
163 if (!IS_ENABLED(CONFIG_DEBUG_VM) && unlikely(folio_mapped(folio))) {
166 pr_alert("BUG: Bad page cache in process %s pfn:%05lx\n",
167 current->comm, folio_pfn(folio));
168 dump_page(&folio->page, "still mapped when deleted");
170 add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE);
172 mapcount = page_mapcount(&folio->page);
173 if (mapping_exiting(mapping) &&
174 folio_ref_count(folio) >= mapcount + 2) {
176 * All vmas have already been torn down, so it's
177 * a good bet that actually the folio is unmapped,
178 * and we'd prefer not to leak it: if we're wrong,
179 * some other bad page check should catch it later.
181 page_mapcount_reset(&folio->page);
182 folio_ref_sub(folio, mapcount);
186 /* hugetlb folios do not participate in page cache accounting. */
187 if (folio_test_hugetlb(folio))
190 nr = folio_nr_pages(folio);
192 __lruvec_stat_mod_folio(folio, NR_FILE_PAGES, -nr);
193 if (folio_test_swapbacked(folio)) {
194 __lruvec_stat_mod_folio(folio, NR_SHMEM, -nr);
195 if (folio_test_pmd_mappable(folio))
196 __lruvec_stat_mod_folio(folio, NR_SHMEM_THPS, -nr);
197 } else if (folio_test_pmd_mappable(folio)) {
198 __lruvec_stat_mod_folio(folio, NR_FILE_THPS, -nr);
199 filemap_nr_thps_dec(mapping);
203 * At this point folio must be either written or cleaned by
204 * truncate. Dirty folio here signals a bug and loss of
207 * This fixes dirty accounting after removing the folio entirely
208 * but leaves the dirty flag set: it has no effect for truncated
209 * folio and anyway will be cleared before returning folio to
212 if (WARN_ON_ONCE(folio_test_dirty(folio)))
213 folio_account_cleaned(folio, mapping,
214 inode_to_wb(mapping->host));
218 * Delete a page from the page cache and free it. Caller has to make
219 * sure the page is locked and that nobody else uses it - or that usage
220 * is safe. The caller must hold the i_pages lock.
222 void __filemap_remove_folio(struct folio *folio, void *shadow)
224 struct address_space *mapping = folio->mapping;
226 trace_mm_filemap_delete_from_page_cache(folio);
227 filemap_unaccount_folio(mapping, folio);
228 page_cache_delete(mapping, folio, shadow);
231 void filemap_free_folio(struct address_space *mapping, struct folio *folio)
233 void (*freepage)(struct page *);
235 freepage = mapping->a_ops->freepage;
237 freepage(&folio->page);
239 if (folio_test_large(folio) && !folio_test_hugetlb(folio)) {
240 folio_ref_sub(folio, folio_nr_pages(folio));
241 VM_BUG_ON_FOLIO(folio_ref_count(folio) <= 0, folio);
248 * filemap_remove_folio - Remove folio from page cache.
251 * This must be called only on folios that are locked and have been
252 * verified to be in the page cache. It will never put the folio into
253 * the free list because the caller has a reference on the page.
255 void filemap_remove_folio(struct folio *folio)
257 struct address_space *mapping = folio->mapping;
259 BUG_ON(!folio_test_locked(folio));
260 spin_lock(&mapping->host->i_lock);
261 xa_lock_irq(&mapping->i_pages);
262 __filemap_remove_folio(folio, NULL);
263 xa_unlock_irq(&mapping->i_pages);
264 if (mapping_shrinkable(mapping))
265 inode_add_lru(mapping->host);
266 spin_unlock(&mapping->host->i_lock);
268 filemap_free_folio(mapping, folio);
272 * page_cache_delete_batch - delete several folios from page cache
273 * @mapping: the mapping to which folios belong
274 * @fbatch: batch of folios to delete
276 * The function walks over mapping->i_pages and removes folios passed in
277 * @fbatch from the mapping. The function expects @fbatch to be sorted
278 * by page index and is optimised for it to be dense.
279 * It tolerates holes in @fbatch (mapping entries at those indices are not
282 * The function expects the i_pages lock to be held.
284 static void page_cache_delete_batch(struct address_space *mapping,
285 struct folio_batch *fbatch)
287 XA_STATE(xas, &mapping->i_pages, fbatch->folios[0]->index);
288 long total_pages = 0;
292 mapping_set_update(&xas, mapping);
293 xas_for_each(&xas, folio, ULONG_MAX) {
294 if (i >= folio_batch_count(fbatch))
297 /* A swap/dax/shadow entry got inserted? Skip it. */
298 if (xa_is_value(folio))
301 * A page got inserted in our range? Skip it. We have our
302 * pages locked so they are protected from being removed.
303 * If we see a page whose index is higher than ours, it
304 * means our page has been removed, which shouldn't be
305 * possible because we're holding the PageLock.
307 if (folio != fbatch->folios[i]) {
308 VM_BUG_ON_FOLIO(folio->index >
309 fbatch->folios[i]->index, folio);
313 WARN_ON_ONCE(!folio_test_locked(folio));
315 folio->mapping = NULL;
316 /* Leave folio->index set: truncation lookup relies on it */
319 xas_store(&xas, NULL);
320 total_pages += folio_nr_pages(folio);
322 mapping->nrpages -= total_pages;
325 void delete_from_page_cache_batch(struct address_space *mapping,
326 struct folio_batch *fbatch)
330 if (!folio_batch_count(fbatch))
333 spin_lock(&mapping->host->i_lock);
334 xa_lock_irq(&mapping->i_pages);
335 for (i = 0; i < folio_batch_count(fbatch); i++) {
336 struct folio *folio = fbatch->folios[i];
338 trace_mm_filemap_delete_from_page_cache(folio);
339 filemap_unaccount_folio(mapping, folio);
341 page_cache_delete_batch(mapping, fbatch);
342 xa_unlock_irq(&mapping->i_pages);
343 if (mapping_shrinkable(mapping))
344 inode_add_lru(mapping->host);
345 spin_unlock(&mapping->host->i_lock);
347 for (i = 0; i < folio_batch_count(fbatch); i++)
348 filemap_free_folio(mapping, fbatch->folios[i]);
351 int filemap_check_errors(struct address_space *mapping)
354 /* Check for outstanding write errors */
355 if (test_bit(AS_ENOSPC, &mapping->flags) &&
356 test_and_clear_bit(AS_ENOSPC, &mapping->flags))
358 if (test_bit(AS_EIO, &mapping->flags) &&
359 test_and_clear_bit(AS_EIO, &mapping->flags))
363 EXPORT_SYMBOL(filemap_check_errors);
365 static int filemap_check_and_keep_errors(struct address_space *mapping)
367 /* Check for outstanding write errors */
368 if (test_bit(AS_EIO, &mapping->flags))
370 if (test_bit(AS_ENOSPC, &mapping->flags))
376 * filemap_fdatawrite_wbc - start writeback on mapping dirty pages in range
377 * @mapping: address space structure to write
378 * @wbc: the writeback_control controlling the writeout
380 * Call writepages on the mapping using the provided wbc to control the
383 * Return: %0 on success, negative error code otherwise.
385 int filemap_fdatawrite_wbc(struct address_space *mapping,
386 struct writeback_control *wbc)
390 if (!mapping_can_writeback(mapping) ||
391 !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
394 wbc_attach_fdatawrite_inode(wbc, mapping->host);
395 ret = do_writepages(mapping, wbc);
396 wbc_detach_inode(wbc);
399 EXPORT_SYMBOL(filemap_fdatawrite_wbc);
402 * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range
403 * @mapping: address space structure to write
404 * @start: offset in bytes where the range starts
405 * @end: offset in bytes where the range ends (inclusive)
406 * @sync_mode: enable synchronous operation
408 * Start writeback against all of a mapping's dirty pages that lie
409 * within the byte offsets <start, end> inclusive.
411 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
412 * opposed to a regular memory cleansing writeback. The difference between
413 * these two operations is that if a dirty page/buffer is encountered, it must
414 * be waited upon, and not just skipped over.
416 * Return: %0 on success, negative error code otherwise.
418 int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
419 loff_t end, int sync_mode)
421 struct writeback_control wbc = {
422 .sync_mode = sync_mode,
423 .nr_to_write = LONG_MAX,
424 .range_start = start,
428 return filemap_fdatawrite_wbc(mapping, &wbc);
431 static inline int __filemap_fdatawrite(struct address_space *mapping,
434 return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode);
437 int filemap_fdatawrite(struct address_space *mapping)
439 return __filemap_fdatawrite(mapping, WB_SYNC_ALL);
441 EXPORT_SYMBOL(filemap_fdatawrite);
443 int filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
446 return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL);
448 EXPORT_SYMBOL(filemap_fdatawrite_range);
451 * filemap_flush - mostly a non-blocking flush
452 * @mapping: target address_space
454 * This is a mostly non-blocking flush. Not suitable for data-integrity
455 * purposes - I/O may not be started against all dirty pages.
457 * Return: %0 on success, negative error code otherwise.
459 int filemap_flush(struct address_space *mapping)
461 return __filemap_fdatawrite(mapping, WB_SYNC_NONE);
463 EXPORT_SYMBOL(filemap_flush);
466 * filemap_range_has_page - check if a page exists in range.
467 * @mapping: address space within which to check
468 * @start_byte: offset in bytes where the range starts
469 * @end_byte: offset in bytes where the range ends (inclusive)
471 * Find at least one page in the range supplied, usually used to check if
472 * direct writing in this range will trigger a writeback.
474 * Return: %true if at least one page exists in the specified range,
477 bool filemap_range_has_page(struct address_space *mapping,
478 loff_t start_byte, loff_t end_byte)
481 XA_STATE(xas, &mapping->i_pages, start_byte >> PAGE_SHIFT);
482 pgoff_t max = end_byte >> PAGE_SHIFT;
484 if (end_byte < start_byte)
489 page = xas_find(&xas, max);
490 if (xas_retry(&xas, page))
492 /* Shadow entries don't count */
493 if (xa_is_value(page))
496 * We don't need to try to pin this page; we're about to
497 * release the RCU lock anyway. It is enough to know that
498 * there was a page here recently.
506 EXPORT_SYMBOL(filemap_range_has_page);
508 static void __filemap_fdatawait_range(struct address_space *mapping,
509 loff_t start_byte, loff_t end_byte)
511 pgoff_t index = start_byte >> PAGE_SHIFT;
512 pgoff_t end = end_byte >> PAGE_SHIFT;
516 if (end_byte < start_byte)
520 while (index <= end) {
523 nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index,
524 end, PAGECACHE_TAG_WRITEBACK);
528 for (i = 0; i < nr_pages; i++) {
529 struct page *page = pvec.pages[i];
531 wait_on_page_writeback(page);
532 ClearPageError(page);
534 pagevec_release(&pvec);
540 * filemap_fdatawait_range - wait for writeback to complete
541 * @mapping: address space structure to wait for
542 * @start_byte: offset in bytes where the range starts
543 * @end_byte: offset in bytes where the range ends (inclusive)
545 * Walk the list of under-writeback pages of the given address space
546 * in the given range and wait for all of them. Check error status of
547 * the address space and return it.
549 * Since the error status of the address space is cleared by this function,
550 * callers are responsible for checking the return value and handling and/or
551 * reporting the error.
553 * Return: error status of the address space.
555 int filemap_fdatawait_range(struct address_space *mapping, loff_t start_byte,
558 __filemap_fdatawait_range(mapping, start_byte, end_byte);
559 return filemap_check_errors(mapping);
561 EXPORT_SYMBOL(filemap_fdatawait_range);
564 * filemap_fdatawait_range_keep_errors - wait for writeback to complete
565 * @mapping: address space structure to wait for
566 * @start_byte: offset in bytes where the range starts
567 * @end_byte: offset in bytes where the range ends (inclusive)
569 * Walk the list of under-writeback pages of the given address space in the
570 * given range and wait for all of them. Unlike filemap_fdatawait_range(),
571 * this function does not clear error status of the address space.
573 * Use this function if callers don't handle errors themselves. Expected
574 * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2),
577 int filemap_fdatawait_range_keep_errors(struct address_space *mapping,
578 loff_t start_byte, loff_t end_byte)
580 __filemap_fdatawait_range(mapping, start_byte, end_byte);
581 return filemap_check_and_keep_errors(mapping);
583 EXPORT_SYMBOL(filemap_fdatawait_range_keep_errors);
586 * file_fdatawait_range - wait for writeback to complete
587 * @file: file pointing to address space structure to wait for
588 * @start_byte: offset in bytes where the range starts
589 * @end_byte: offset in bytes where the range ends (inclusive)
591 * Walk the list of under-writeback pages of the address space that file
592 * refers to, in the given range and wait for all of them. Check error
593 * status of the address space vs. the file->f_wb_err cursor and return it.
595 * Since the error status of the file is advanced by this function,
596 * callers are responsible for checking the return value and handling and/or
597 * reporting the error.
599 * Return: error status of the address space vs. the file->f_wb_err cursor.
601 int file_fdatawait_range(struct file *file, loff_t start_byte, loff_t end_byte)
603 struct address_space *mapping = file->f_mapping;
605 __filemap_fdatawait_range(mapping, start_byte, end_byte);
606 return file_check_and_advance_wb_err(file);
608 EXPORT_SYMBOL(file_fdatawait_range);
611 * filemap_fdatawait_keep_errors - wait for writeback without clearing errors
612 * @mapping: address space structure to wait for
614 * Walk the list of under-writeback pages of the given address space
615 * and wait for all of them. Unlike filemap_fdatawait(), this function
616 * does not clear error status of the address space.
618 * Use this function if callers don't handle errors themselves. Expected
619 * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2),
622 * Return: error status of the address space.
624 int filemap_fdatawait_keep_errors(struct address_space *mapping)
626 __filemap_fdatawait_range(mapping, 0, LLONG_MAX);
627 return filemap_check_and_keep_errors(mapping);
629 EXPORT_SYMBOL(filemap_fdatawait_keep_errors);
631 /* Returns true if writeback might be needed or already in progress. */
632 static bool mapping_needs_writeback(struct address_space *mapping)
634 return mapping->nrpages;
637 bool filemap_range_has_writeback(struct address_space *mapping,
638 loff_t start_byte, loff_t end_byte)
640 XA_STATE(xas, &mapping->i_pages, start_byte >> PAGE_SHIFT);
641 pgoff_t max = end_byte >> PAGE_SHIFT;
644 if (end_byte < start_byte)
648 xas_for_each(&xas, page, max) {
649 if (xas_retry(&xas, page))
651 if (xa_is_value(page))
653 if (PageDirty(page) || PageLocked(page) || PageWriteback(page))
659 EXPORT_SYMBOL_GPL(filemap_range_has_writeback);
662 * filemap_write_and_wait_range - write out & wait on a file range
663 * @mapping: the address_space for the pages
664 * @lstart: offset in bytes where the range starts
665 * @lend: offset in bytes where the range ends (inclusive)
667 * Write out and wait upon file offsets lstart->lend, inclusive.
669 * Note that @lend is inclusive (describes the last byte to be written) so
670 * that this function can be used to write to the very end-of-file (end = -1).
672 * Return: error status of the address space.
674 int filemap_write_and_wait_range(struct address_space *mapping,
675 loff_t lstart, loff_t lend)
679 if (mapping_needs_writeback(mapping)) {
680 err = __filemap_fdatawrite_range(mapping, lstart, lend,
683 * Even if the above returned error, the pages may be
684 * written partially (e.g. -ENOSPC), so we wait for it.
685 * But the -EIO is special case, it may indicate the worst
686 * thing (e.g. bug) happened, so we avoid waiting for it.
689 int err2 = filemap_fdatawait_range(mapping,
694 /* Clear any previously stored errors */
695 filemap_check_errors(mapping);
698 err = filemap_check_errors(mapping);
702 EXPORT_SYMBOL(filemap_write_and_wait_range);
704 void __filemap_set_wb_err(struct address_space *mapping, int err)
706 errseq_t eseq = errseq_set(&mapping->wb_err, err);
708 trace_filemap_set_wb_err(mapping, eseq);
710 EXPORT_SYMBOL(__filemap_set_wb_err);
713 * file_check_and_advance_wb_err - report wb error (if any) that was previously
714 * and advance wb_err to current one
715 * @file: struct file on which the error is being reported
717 * When userland calls fsync (or something like nfsd does the equivalent), we
718 * want to report any writeback errors that occurred since the last fsync (or
719 * since the file was opened if there haven't been any).
721 * Grab the wb_err from the mapping. If it matches what we have in the file,
722 * then just quickly return 0. The file is all caught up.
724 * If it doesn't match, then take the mapping value, set the "seen" flag in
725 * it and try to swap it into place. If it works, or another task beat us
726 * to it with the new value, then update the f_wb_err and return the error
727 * portion. The error at this point must be reported via proper channels
728 * (a'la fsync, or NFS COMMIT operation, etc.).
730 * While we handle mapping->wb_err with atomic operations, the f_wb_err
731 * value is protected by the f_lock since we must ensure that it reflects
732 * the latest value swapped in for this file descriptor.
734 * Return: %0 on success, negative error code otherwise.
736 int file_check_and_advance_wb_err(struct file *file)
739 errseq_t old = READ_ONCE(file->f_wb_err);
740 struct address_space *mapping = file->f_mapping;
742 /* Locklessly handle the common case where nothing has changed */
743 if (errseq_check(&mapping->wb_err, old)) {
744 /* Something changed, must use slow path */
745 spin_lock(&file->f_lock);
746 old = file->f_wb_err;
747 err = errseq_check_and_advance(&mapping->wb_err,
749 trace_file_check_and_advance_wb_err(file, old);
750 spin_unlock(&file->f_lock);
754 * We're mostly using this function as a drop in replacement for
755 * filemap_check_errors. Clear AS_EIO/AS_ENOSPC to emulate the effect
756 * that the legacy code would have had on these flags.
758 clear_bit(AS_EIO, &mapping->flags);
759 clear_bit(AS_ENOSPC, &mapping->flags);
762 EXPORT_SYMBOL(file_check_and_advance_wb_err);
765 * file_write_and_wait_range - write out & wait on a file range
766 * @file: file pointing to address_space with pages
767 * @lstart: offset in bytes where the range starts
768 * @lend: offset in bytes where the range ends (inclusive)
770 * Write out and wait upon file offsets lstart->lend, inclusive.
772 * Note that @lend is inclusive (describes the last byte to be written) so
773 * that this function can be used to write to the very end-of-file (end = -1).
775 * After writing out and waiting on the data, we check and advance the
776 * f_wb_err cursor to the latest value, and return any errors detected there.
778 * Return: %0 on success, negative error code otherwise.
780 int file_write_and_wait_range(struct file *file, loff_t lstart, loff_t lend)
783 struct address_space *mapping = file->f_mapping;
785 if (mapping_needs_writeback(mapping)) {
786 err = __filemap_fdatawrite_range(mapping, lstart, lend,
788 /* See comment of filemap_write_and_wait() */
790 __filemap_fdatawait_range(mapping, lstart, lend);
792 err2 = file_check_and_advance_wb_err(file);
797 EXPORT_SYMBOL(file_write_and_wait_range);
800 * replace_page_cache_page - replace a pagecache page with a new one
801 * @old: page to be replaced
802 * @new: page to replace with
804 * This function replaces a page in the pagecache with a new one. On
805 * success it acquires the pagecache reference for the new page and
806 * drops it for the old page. Both the old and new pages must be
807 * locked. This function does not add the new page to the LRU, the
808 * caller must do that.
810 * The remove + add is atomic. This function cannot fail.
812 void replace_page_cache_page(struct page *old, struct page *new)
814 struct folio *fold = page_folio(old);
815 struct folio *fnew = page_folio(new);
816 struct address_space *mapping = old->mapping;
817 void (*freepage)(struct page *) = mapping->a_ops->freepage;
818 pgoff_t offset = old->index;
819 XA_STATE(xas, &mapping->i_pages, offset);
821 VM_BUG_ON_PAGE(!PageLocked(old), old);
822 VM_BUG_ON_PAGE(!PageLocked(new), new);
823 VM_BUG_ON_PAGE(new->mapping, new);
826 new->mapping = mapping;
829 mem_cgroup_migrate(fold, fnew);
832 xas_store(&xas, new);
835 /* hugetlb pages do not participate in page cache accounting. */
837 __dec_lruvec_page_state(old, NR_FILE_PAGES);
839 __inc_lruvec_page_state(new, NR_FILE_PAGES);
840 if (PageSwapBacked(old))
841 __dec_lruvec_page_state(old, NR_SHMEM);
842 if (PageSwapBacked(new))
843 __inc_lruvec_page_state(new, NR_SHMEM);
844 xas_unlock_irq(&xas);
849 EXPORT_SYMBOL_GPL(replace_page_cache_page);
851 noinline int __filemap_add_folio(struct address_space *mapping,
852 struct folio *folio, pgoff_t index, gfp_t gfp, void **shadowp)
854 XA_STATE(xas, &mapping->i_pages, index);
855 int huge = folio_test_hugetlb(folio);
857 bool charged = false;
859 VM_BUG_ON_FOLIO(!folio_test_locked(folio), folio);
860 VM_BUG_ON_FOLIO(folio_test_swapbacked(folio), folio);
861 mapping_set_update(&xas, mapping);
864 folio->mapping = mapping;
865 folio->index = index;
868 error = mem_cgroup_charge(folio, NULL, gfp);
869 VM_BUG_ON_FOLIO(index & (folio_nr_pages(folio) - 1), folio);
875 gfp &= GFP_RECLAIM_MASK;
878 unsigned int order = xa_get_order(xas.xa, xas.xa_index);
879 void *entry, *old = NULL;
881 if (order > folio_order(folio))
882 xas_split_alloc(&xas, xa_load(xas.xa, xas.xa_index),
885 xas_for_each_conflict(&xas, entry) {
887 if (!xa_is_value(entry)) {
888 xas_set_err(&xas, -EEXIST);
896 /* entry may have been split before we acquired lock */
897 order = xa_get_order(xas.xa, xas.xa_index);
898 if (order > folio_order(folio)) {
899 xas_split(&xas, old, order);
904 xas_store(&xas, folio);
910 /* hugetlb pages do not participate in page cache accounting */
912 __lruvec_stat_add_folio(folio, NR_FILE_PAGES);
914 xas_unlock_irq(&xas);
915 } while (xas_nomem(&xas, gfp));
917 if (xas_error(&xas)) {
918 error = xas_error(&xas);
920 mem_cgroup_uncharge(folio);
924 trace_mm_filemap_add_to_page_cache(folio);
927 folio->mapping = NULL;
928 /* Leave page->index set: truncation relies upon it */
932 ALLOW_ERROR_INJECTION(__filemap_add_folio, ERRNO);
935 * add_to_page_cache_locked - add a locked page to the pagecache
937 * @mapping: the page's address_space
938 * @offset: page index
939 * @gfp_mask: page allocation mode
941 * This function is used to add a page to the pagecache. It must be locked.
942 * This function does not add the page to the LRU. The caller must do that.
944 * Return: %0 on success, negative error code otherwise.
946 int add_to_page_cache_locked(struct page *page, struct address_space *mapping,
947 pgoff_t offset, gfp_t gfp_mask)
949 return __filemap_add_folio(mapping, page_folio(page), offset,
952 EXPORT_SYMBOL(add_to_page_cache_locked);
954 int filemap_add_folio(struct address_space *mapping, struct folio *folio,
955 pgoff_t index, gfp_t gfp)
960 __folio_set_locked(folio);
961 ret = __filemap_add_folio(mapping, folio, index, gfp, &shadow);
963 __folio_clear_locked(folio);
966 * The folio might have been evicted from cache only
967 * recently, in which case it should be activated like
968 * any other repeatedly accessed folio.
969 * The exception is folios getting rewritten; evicting other
970 * data from the working set, only to cache data that will
971 * get overwritten with something else, is a waste of memory.
973 WARN_ON_ONCE(folio_test_active(folio));
974 if (!(gfp & __GFP_WRITE) && shadow)
975 workingset_refault(folio, shadow);
976 folio_add_lru(folio);
980 EXPORT_SYMBOL_GPL(filemap_add_folio);
983 struct folio *filemap_alloc_folio(gfp_t gfp, unsigned int order)
988 if (cpuset_do_page_mem_spread()) {
989 unsigned int cpuset_mems_cookie;
991 cpuset_mems_cookie = read_mems_allowed_begin();
992 n = cpuset_mem_spread_node();
993 folio = __folio_alloc_node(gfp, order, n);
994 } while (!folio && read_mems_allowed_retry(cpuset_mems_cookie));
998 return folio_alloc(gfp, order);
1000 EXPORT_SYMBOL(filemap_alloc_folio);
1004 * filemap_invalidate_lock_two - lock invalidate_lock for two mappings
1006 * Lock exclusively invalidate_lock of any passed mapping that is not NULL.
1008 * @mapping1: the first mapping to lock
1009 * @mapping2: the second mapping to lock
1011 void filemap_invalidate_lock_two(struct address_space *mapping1,
1012 struct address_space *mapping2)
1014 if (mapping1 > mapping2)
1015 swap(mapping1, mapping2);
1017 down_write(&mapping1->invalidate_lock);
1018 if (mapping2 && mapping1 != mapping2)
1019 down_write_nested(&mapping2->invalidate_lock, 1);
1021 EXPORT_SYMBOL(filemap_invalidate_lock_two);
1024 * filemap_invalidate_unlock_two - unlock invalidate_lock for two mappings
1026 * Unlock exclusive invalidate_lock of any passed mapping that is not NULL.
1028 * @mapping1: the first mapping to unlock
1029 * @mapping2: the second mapping to unlock
1031 void filemap_invalidate_unlock_two(struct address_space *mapping1,
1032 struct address_space *mapping2)
1035 up_write(&mapping1->invalidate_lock);
1036 if (mapping2 && mapping1 != mapping2)
1037 up_write(&mapping2->invalidate_lock);
1039 EXPORT_SYMBOL(filemap_invalidate_unlock_two);
1042 * In order to wait for pages to become available there must be
1043 * waitqueues associated with pages. By using a hash table of
1044 * waitqueues where the bucket discipline is to maintain all
1045 * waiters on the same queue and wake all when any of the pages
1046 * become available, and for the woken contexts to check to be
1047 * sure the appropriate page became available, this saves space
1048 * at a cost of "thundering herd" phenomena during rare hash
1051 #define PAGE_WAIT_TABLE_BITS 8
1052 #define PAGE_WAIT_TABLE_SIZE (1 << PAGE_WAIT_TABLE_BITS)
1053 static wait_queue_head_t folio_wait_table[PAGE_WAIT_TABLE_SIZE] __cacheline_aligned;
1055 static wait_queue_head_t *folio_waitqueue(struct folio *folio)
1057 return &folio_wait_table[hash_ptr(folio, PAGE_WAIT_TABLE_BITS)];
1060 void __init pagecache_init(void)
1064 for (i = 0; i < PAGE_WAIT_TABLE_SIZE; i++)
1065 init_waitqueue_head(&folio_wait_table[i]);
1067 page_writeback_init();
1071 * The page wait code treats the "wait->flags" somewhat unusually, because
1072 * we have multiple different kinds of waits, not just the usual "exclusive"
1077 * (a) no special bits set:
1079 * We're just waiting for the bit to be released, and when a waker
1080 * calls the wakeup function, we set WQ_FLAG_WOKEN and wake it up,
1081 * and remove it from the wait queue.
1083 * Simple and straightforward.
1085 * (b) WQ_FLAG_EXCLUSIVE:
1087 * The waiter is waiting to get the lock, and only one waiter should
1088 * be woken up to avoid any thundering herd behavior. We'll set the
1089 * WQ_FLAG_WOKEN bit, wake it up, and remove it from the wait queue.
1091 * This is the traditional exclusive wait.
1093 * (c) WQ_FLAG_EXCLUSIVE | WQ_FLAG_CUSTOM:
1095 * The waiter is waiting to get the bit, and additionally wants the
1096 * lock to be transferred to it for fair lock behavior. If the lock
1097 * cannot be taken, we stop walking the wait queue without waking
1100 * This is the "fair lock handoff" case, and in addition to setting
1101 * WQ_FLAG_WOKEN, we set WQ_FLAG_DONE to let the waiter easily see
1102 * that it now has the lock.
1104 static int wake_page_function(wait_queue_entry_t *wait, unsigned mode, int sync, void *arg)
1107 struct wait_page_key *key = arg;
1108 struct wait_page_queue *wait_page
1109 = container_of(wait, struct wait_page_queue, wait);
1111 if (!wake_page_match(wait_page, key))
1115 * If it's a lock handoff wait, we get the bit for it, and
1116 * stop walking (and do not wake it up) if we can't.
1118 flags = wait->flags;
1119 if (flags & WQ_FLAG_EXCLUSIVE) {
1120 if (test_bit(key->bit_nr, &key->folio->flags))
1122 if (flags & WQ_FLAG_CUSTOM) {
1123 if (test_and_set_bit(key->bit_nr, &key->folio->flags))
1125 flags |= WQ_FLAG_DONE;
1130 * We are holding the wait-queue lock, but the waiter that
1131 * is waiting for this will be checking the flags without
1134 * So update the flags atomically, and wake up the waiter
1135 * afterwards to avoid any races. This store-release pairs
1136 * with the load-acquire in folio_wait_bit_common().
1138 smp_store_release(&wait->flags, flags | WQ_FLAG_WOKEN);
1139 wake_up_state(wait->private, mode);
1142 * Ok, we have successfully done what we're waiting for,
1143 * and we can unconditionally remove the wait entry.
1145 * Note that this pairs with the "finish_wait()" in the
1146 * waiter, and has to be the absolute last thing we do.
1147 * After this list_del_init(&wait->entry) the wait entry
1148 * might be de-allocated and the process might even have
1151 list_del_init_careful(&wait->entry);
1152 return (flags & WQ_FLAG_EXCLUSIVE) != 0;
1155 static void folio_wake_bit(struct folio *folio, int bit_nr)
1157 wait_queue_head_t *q = folio_waitqueue(folio);
1158 struct wait_page_key key;
1159 unsigned long flags;
1160 wait_queue_entry_t bookmark;
1163 key.bit_nr = bit_nr;
1167 bookmark.private = NULL;
1168 bookmark.func = NULL;
1169 INIT_LIST_HEAD(&bookmark.entry);
1171 spin_lock_irqsave(&q->lock, flags);
1172 __wake_up_locked_key_bookmark(q, TASK_NORMAL, &key, &bookmark);
1174 while (bookmark.flags & WQ_FLAG_BOOKMARK) {
1176 * Take a breather from holding the lock,
1177 * allow pages that finish wake up asynchronously
1178 * to acquire the lock and remove themselves
1181 spin_unlock_irqrestore(&q->lock, flags);
1183 spin_lock_irqsave(&q->lock, flags);
1184 __wake_up_locked_key_bookmark(q, TASK_NORMAL, &key, &bookmark);
1188 * It is possible for other pages to have collided on the waitqueue
1189 * hash, so in that case check for a page match. That prevents a long-
1192 * It is still possible to miss a case here, when we woke page waiters
1193 * and removed them from the waitqueue, but there are still other
1196 if (!waitqueue_active(q) || !key.page_match) {
1197 folio_clear_waiters(folio);
1199 * It's possible to miss clearing Waiters here, when we woke
1200 * our page waiters, but the hashed waitqueue has waiters for
1201 * other pages on it.
1203 * That's okay, it's a rare case. The next waker will clear it.
1206 spin_unlock_irqrestore(&q->lock, flags);
1209 static void folio_wake(struct folio *folio, int bit)
1211 if (!folio_test_waiters(folio))
1213 folio_wake_bit(folio, bit);
1217 * A choice of three behaviors for folio_wait_bit_common():
1220 EXCLUSIVE, /* Hold ref to page and take the bit when woken, like
1221 * __folio_lock() waiting on then setting PG_locked.
1223 SHARED, /* Hold ref to page and check the bit when woken, like
1224 * folio_wait_writeback() waiting on PG_writeback.
1226 DROP, /* Drop ref to page before wait, no check when woken,
1227 * like folio_put_wait_locked() on PG_locked.
1232 * Attempt to check (or get) the folio flag, and mark us done
1235 static inline bool folio_trylock_flag(struct folio *folio, int bit_nr,
1236 struct wait_queue_entry *wait)
1238 if (wait->flags & WQ_FLAG_EXCLUSIVE) {
1239 if (test_and_set_bit(bit_nr, &folio->flags))
1241 } else if (test_bit(bit_nr, &folio->flags))
1244 wait->flags |= WQ_FLAG_WOKEN | WQ_FLAG_DONE;
1248 /* How many times do we accept lock stealing from under a waiter? */
1249 int sysctl_page_lock_unfairness = 5;
1251 static inline int folio_wait_bit_common(struct folio *folio, int bit_nr,
1252 int state, enum behavior behavior)
1254 wait_queue_head_t *q = folio_waitqueue(folio);
1255 int unfairness = sysctl_page_lock_unfairness;
1256 struct wait_page_queue wait_page;
1257 wait_queue_entry_t *wait = &wait_page.wait;
1258 bool thrashing = false;
1259 bool delayacct = false;
1260 unsigned long pflags;
1262 if (bit_nr == PG_locked &&
1263 !folio_test_uptodate(folio) && folio_test_workingset(folio)) {
1264 if (!folio_test_swapbacked(folio)) {
1265 delayacct_thrashing_start();
1268 psi_memstall_enter(&pflags);
1273 wait->func = wake_page_function;
1274 wait_page.folio = folio;
1275 wait_page.bit_nr = bit_nr;
1279 if (behavior == EXCLUSIVE) {
1280 wait->flags = WQ_FLAG_EXCLUSIVE;
1281 if (--unfairness < 0)
1282 wait->flags |= WQ_FLAG_CUSTOM;
1286 * Do one last check whether we can get the
1287 * page bit synchronously.
1289 * Do the folio_set_waiters() marking before that
1290 * to let any waker we _just_ missed know they
1291 * need to wake us up (otherwise they'll never
1292 * even go to the slow case that looks at the
1293 * page queue), and add ourselves to the wait
1294 * queue if we need to sleep.
1296 * This part needs to be done under the queue
1297 * lock to avoid races.
1299 spin_lock_irq(&q->lock);
1300 folio_set_waiters(folio);
1301 if (!folio_trylock_flag(folio, bit_nr, wait))
1302 __add_wait_queue_entry_tail(q, wait);
1303 spin_unlock_irq(&q->lock);
1306 * From now on, all the logic will be based on
1307 * the WQ_FLAG_WOKEN and WQ_FLAG_DONE flag, to
1308 * see whether the page bit testing has already
1309 * been done by the wake function.
1311 * We can drop our reference to the folio.
1313 if (behavior == DROP)
1317 * Note that until the "finish_wait()", or until
1318 * we see the WQ_FLAG_WOKEN flag, we need to
1319 * be very careful with the 'wait->flags', because
1320 * we may race with a waker that sets them.
1325 set_current_state(state);
1327 /* Loop until we've been woken or interrupted */
1328 flags = smp_load_acquire(&wait->flags);
1329 if (!(flags & WQ_FLAG_WOKEN)) {
1330 if (signal_pending_state(state, current))
1337 /* If we were non-exclusive, we're done */
1338 if (behavior != EXCLUSIVE)
1341 /* If the waker got the lock for us, we're done */
1342 if (flags & WQ_FLAG_DONE)
1346 * Otherwise, if we're getting the lock, we need to
1347 * try to get it ourselves.
1349 * And if that fails, we'll have to retry this all.
1351 if (unlikely(test_and_set_bit(bit_nr, folio_flags(folio, 0))))
1354 wait->flags |= WQ_FLAG_DONE;
1359 * If a signal happened, this 'finish_wait()' may remove the last
1360 * waiter from the wait-queues, but the folio waiters bit will remain
1361 * set. That's ok. The next wakeup will take care of it, and trying
1362 * to do it here would be difficult and prone to races.
1364 finish_wait(q, wait);
1368 delayacct_thrashing_end();
1369 psi_memstall_leave(&pflags);
1373 * NOTE! The wait->flags weren't stable until we've done the
1374 * 'finish_wait()', and we could have exited the loop above due
1375 * to a signal, and had a wakeup event happen after the signal
1376 * test but before the 'finish_wait()'.
1378 * So only after the finish_wait() can we reliably determine
1379 * if we got woken up or not, so we can now figure out the final
1380 * return value based on that state without races.
1382 * Also note that WQ_FLAG_WOKEN is sufficient for a non-exclusive
1383 * waiter, but an exclusive one requires WQ_FLAG_DONE.
1385 if (behavior == EXCLUSIVE)
1386 return wait->flags & WQ_FLAG_DONE ? 0 : -EINTR;
1388 return wait->flags & WQ_FLAG_WOKEN ? 0 : -EINTR;
1391 void folio_wait_bit(struct folio *folio, int bit_nr)
1393 folio_wait_bit_common(folio, bit_nr, TASK_UNINTERRUPTIBLE, SHARED);
1395 EXPORT_SYMBOL(folio_wait_bit);
1397 int folio_wait_bit_killable(struct folio *folio, int bit_nr)
1399 return folio_wait_bit_common(folio, bit_nr, TASK_KILLABLE, SHARED);
1401 EXPORT_SYMBOL(folio_wait_bit_killable);
1404 * folio_put_wait_locked - Drop a reference and wait for it to be unlocked
1405 * @folio: The folio to wait for.
1406 * @state: The sleep state (TASK_KILLABLE, TASK_UNINTERRUPTIBLE, etc).
1408 * The caller should hold a reference on @folio. They expect the page to
1409 * become unlocked relatively soon, but do not wish to hold up migration
1410 * (for example) by holding the reference while waiting for the folio to
1411 * come unlocked. After this function returns, the caller should not
1412 * dereference @folio.
1414 * Return: 0 if the folio was unlocked or -EINTR if interrupted by a signal.
1416 int folio_put_wait_locked(struct folio *folio, int state)
1418 return folio_wait_bit_common(folio, PG_locked, state, DROP);
1422 * folio_add_wait_queue - Add an arbitrary waiter to a folio's wait queue
1423 * @folio: Folio defining the wait queue of interest
1424 * @waiter: Waiter to add to the queue
1426 * Add an arbitrary @waiter to the wait queue for the nominated @folio.
1428 void folio_add_wait_queue(struct folio *folio, wait_queue_entry_t *waiter)
1430 wait_queue_head_t *q = folio_waitqueue(folio);
1431 unsigned long flags;
1433 spin_lock_irqsave(&q->lock, flags);
1434 __add_wait_queue_entry_tail(q, waiter);
1435 folio_set_waiters(folio);
1436 spin_unlock_irqrestore(&q->lock, flags);
1438 EXPORT_SYMBOL_GPL(folio_add_wait_queue);
1440 #ifndef clear_bit_unlock_is_negative_byte
1443 * PG_waiters is the high bit in the same byte as PG_lock.
1445 * On x86 (and on many other architectures), we can clear PG_lock and
1446 * test the sign bit at the same time. But if the architecture does
1447 * not support that special operation, we just do this all by hand
1450 * The read of PG_waiters has to be after (or concurrently with) PG_locked
1451 * being cleared, but a memory barrier should be unnecessary since it is
1452 * in the same byte as PG_locked.
1454 static inline bool clear_bit_unlock_is_negative_byte(long nr, volatile void *mem)
1456 clear_bit_unlock(nr, mem);
1457 /* smp_mb__after_atomic(); */
1458 return test_bit(PG_waiters, mem);
1464 * folio_unlock - Unlock a locked folio.
1465 * @folio: The folio.
1467 * Unlocks the folio and wakes up any thread sleeping on the page lock.
1469 * Context: May be called from interrupt or process context. May not be
1470 * called from NMI context.
1472 void folio_unlock(struct folio *folio)
1474 /* Bit 7 allows x86 to check the byte's sign bit */
1475 BUILD_BUG_ON(PG_waiters != 7);
1476 BUILD_BUG_ON(PG_locked > 7);
1477 VM_BUG_ON_FOLIO(!folio_test_locked(folio), folio);
1478 if (clear_bit_unlock_is_negative_byte(PG_locked, folio_flags(folio, 0)))
1479 folio_wake_bit(folio, PG_locked);
1481 EXPORT_SYMBOL(folio_unlock);
1484 * folio_end_private_2 - Clear PG_private_2 and wake any waiters.
1485 * @folio: The folio.
1487 * Clear the PG_private_2 bit on a folio and wake up any sleepers waiting for
1488 * it. The folio reference held for PG_private_2 being set is released.
1490 * This is, for example, used when a netfs folio is being written to a local
1491 * disk cache, thereby allowing writes to the cache for the same folio to be
1494 void folio_end_private_2(struct folio *folio)
1496 VM_BUG_ON_FOLIO(!folio_test_private_2(folio), folio);
1497 clear_bit_unlock(PG_private_2, folio_flags(folio, 0));
1498 folio_wake_bit(folio, PG_private_2);
1501 EXPORT_SYMBOL(folio_end_private_2);
1504 * folio_wait_private_2 - Wait for PG_private_2 to be cleared on a folio.
1505 * @folio: The folio to wait on.
1507 * Wait for PG_private_2 (aka PG_fscache) to be cleared on a folio.
1509 void folio_wait_private_2(struct folio *folio)
1511 while (folio_test_private_2(folio))
1512 folio_wait_bit(folio, PG_private_2);
1514 EXPORT_SYMBOL(folio_wait_private_2);
1517 * folio_wait_private_2_killable - Wait for PG_private_2 to be cleared on a folio.
1518 * @folio: The folio to wait on.
1520 * Wait for PG_private_2 (aka PG_fscache) to be cleared on a folio or until a
1521 * fatal signal is received by the calling task.
1524 * - 0 if successful.
1525 * - -EINTR if a fatal signal was encountered.
1527 int folio_wait_private_2_killable(struct folio *folio)
1531 while (folio_test_private_2(folio)) {
1532 ret = folio_wait_bit_killable(folio, PG_private_2);
1539 EXPORT_SYMBOL(folio_wait_private_2_killable);
1542 * folio_end_writeback - End writeback against a folio.
1543 * @folio: The folio.
1545 void folio_end_writeback(struct folio *folio)
1548 * folio_test_clear_reclaim() could be used here but it is an
1549 * atomic operation and overkill in this particular case. Failing
1550 * to shuffle a folio marked for immediate reclaim is too mild
1551 * a gain to justify taking an atomic operation penalty at the
1552 * end of every folio writeback.
1554 if (folio_test_reclaim(folio)) {
1555 folio_clear_reclaim(folio);
1556 folio_rotate_reclaimable(folio);
1560 * Writeback does not hold a folio reference of its own, relying
1561 * on truncation to wait for the clearing of PG_writeback.
1562 * But here we must make sure that the folio is not freed and
1563 * reused before the folio_wake().
1566 if (!__folio_end_writeback(folio))
1569 smp_mb__after_atomic();
1570 folio_wake(folio, PG_writeback);
1571 acct_reclaim_writeback(folio);
1574 EXPORT_SYMBOL(folio_end_writeback);
1577 * After completing I/O on a page, call this routine to update the page
1578 * flags appropriately
1580 void page_endio(struct page *page, bool is_write, int err)
1584 SetPageUptodate(page);
1586 ClearPageUptodate(page);
1592 struct address_space *mapping;
1595 mapping = page_mapping(page);
1597 mapping_set_error(mapping, err);
1599 end_page_writeback(page);
1602 EXPORT_SYMBOL_GPL(page_endio);
1605 * __folio_lock - Get a lock on the folio, assuming we need to sleep to get it.
1606 * @folio: The folio to lock
1608 void __folio_lock(struct folio *folio)
1610 folio_wait_bit_common(folio, PG_locked, TASK_UNINTERRUPTIBLE,
1613 EXPORT_SYMBOL(__folio_lock);
1615 int __folio_lock_killable(struct folio *folio)
1617 return folio_wait_bit_common(folio, PG_locked, TASK_KILLABLE,
1620 EXPORT_SYMBOL_GPL(__folio_lock_killable);
1622 static int __folio_lock_async(struct folio *folio, struct wait_page_queue *wait)
1624 struct wait_queue_head *q = folio_waitqueue(folio);
1627 wait->folio = folio;
1628 wait->bit_nr = PG_locked;
1630 spin_lock_irq(&q->lock);
1631 __add_wait_queue_entry_tail(q, &wait->wait);
1632 folio_set_waiters(folio);
1633 ret = !folio_trylock(folio);
1635 * If we were successful now, we know we're still on the
1636 * waitqueue as we're still under the lock. This means it's
1637 * safe to remove and return success, we know the callback
1638 * isn't going to trigger.
1641 __remove_wait_queue(q, &wait->wait);
1644 spin_unlock_irq(&q->lock);
1650 * true - folio is locked; mmap_lock is still held.
1651 * false - folio is not locked.
1652 * mmap_lock has been released (mmap_read_unlock(), unless flags had both
1653 * FAULT_FLAG_ALLOW_RETRY and FAULT_FLAG_RETRY_NOWAIT set, in
1654 * which case mmap_lock is still held.
1656 * If neither ALLOW_RETRY nor KILLABLE are set, will always return true
1657 * with the folio locked and the mmap_lock unperturbed.
1659 bool __folio_lock_or_retry(struct folio *folio, struct mm_struct *mm,
1662 if (fault_flag_allow_retry_first(flags)) {
1664 * CAUTION! In this case, mmap_lock is not released
1665 * even though return 0.
1667 if (flags & FAULT_FLAG_RETRY_NOWAIT)
1670 mmap_read_unlock(mm);
1671 if (flags & FAULT_FLAG_KILLABLE)
1672 folio_wait_locked_killable(folio);
1674 folio_wait_locked(folio);
1677 if (flags & FAULT_FLAG_KILLABLE) {
1680 ret = __folio_lock_killable(folio);
1682 mmap_read_unlock(mm);
1686 __folio_lock(folio);
1693 * page_cache_next_miss() - Find the next gap in the page cache.
1694 * @mapping: Mapping.
1696 * @max_scan: Maximum range to search.
1698 * Search the range [index, min(index + max_scan - 1, ULONG_MAX)] for the
1699 * gap with the lowest index.
1701 * This function may be called under the rcu_read_lock. However, this will
1702 * not atomically search a snapshot of the cache at a single point in time.
1703 * For example, if a gap is created at index 5, then subsequently a gap is
1704 * created at index 10, page_cache_next_miss covering both indices may
1705 * return 10 if called under the rcu_read_lock.
1707 * Return: The index of the gap if found, otherwise an index outside the
1708 * range specified (in which case 'return - index >= max_scan' will be true).
1709 * In the rare case of index wrap-around, 0 will be returned.
1711 pgoff_t page_cache_next_miss(struct address_space *mapping,
1712 pgoff_t index, unsigned long max_scan)
1714 XA_STATE(xas, &mapping->i_pages, index);
1716 while (max_scan--) {
1717 void *entry = xas_next(&xas);
1718 if (!entry || xa_is_value(entry))
1720 if (xas.xa_index == 0)
1724 return xas.xa_index;
1726 EXPORT_SYMBOL(page_cache_next_miss);
1729 * page_cache_prev_miss() - Find the previous gap in the page cache.
1730 * @mapping: Mapping.
1732 * @max_scan: Maximum range to search.
1734 * Search the range [max(index - max_scan + 1, 0), index] for the
1735 * gap with the highest index.
1737 * This function may be called under the rcu_read_lock. However, this will
1738 * not atomically search a snapshot of the cache at a single point in time.
1739 * For example, if a gap is created at index 10, then subsequently a gap is
1740 * created at index 5, page_cache_prev_miss() covering both indices may
1741 * return 5 if called under the rcu_read_lock.
1743 * Return: The index of the gap if found, otherwise an index outside the
1744 * range specified (in which case 'index - return >= max_scan' will be true).
1745 * In the rare case of wrap-around, ULONG_MAX will be returned.
1747 pgoff_t page_cache_prev_miss(struct address_space *mapping,
1748 pgoff_t index, unsigned long max_scan)
1750 XA_STATE(xas, &mapping->i_pages, index);
1752 while (max_scan--) {
1753 void *entry = xas_prev(&xas);
1754 if (!entry || xa_is_value(entry))
1756 if (xas.xa_index == ULONG_MAX)
1760 return xas.xa_index;
1762 EXPORT_SYMBOL(page_cache_prev_miss);
1765 * Lockless page cache protocol:
1766 * On the lookup side:
1767 * 1. Load the folio from i_pages
1768 * 2. Increment the refcount if it's not zero
1769 * 3. If the folio is not found by xas_reload(), put the refcount and retry
1771 * On the removal side:
1772 * A. Freeze the page (by zeroing the refcount if nobody else has a reference)
1773 * B. Remove the page from i_pages
1774 * C. Return the page to the page allocator
1776 * This means that any page may have its reference count temporarily
1777 * increased by a speculative page cache (or fast GUP) lookup as it can
1778 * be allocated by another user before the RCU grace period expires.
1779 * Because the refcount temporarily acquired here may end up being the
1780 * last refcount on the page, any page allocation must be freeable by
1785 * mapping_get_entry - Get a page cache entry.
1786 * @mapping: the address_space to search
1787 * @index: The page cache index.
1789 * Looks up the page cache entry at @mapping & @index. If it is a folio,
1790 * it is returned with an increased refcount. If it is a shadow entry
1791 * of a previously evicted folio, or a swap entry from shmem/tmpfs,
1792 * it is returned without further action.
1794 * Return: The folio, swap or shadow entry, %NULL if nothing is found.
1796 static void *mapping_get_entry(struct address_space *mapping, pgoff_t index)
1798 XA_STATE(xas, &mapping->i_pages, index);
1799 struct folio *folio;
1804 folio = xas_load(&xas);
1805 if (xas_retry(&xas, folio))
1808 * A shadow entry of a recently evicted page, or a swap entry from
1809 * shmem/tmpfs. Return it without attempting to raise page count.
1811 if (!folio || xa_is_value(folio))
1814 if (!folio_try_get_rcu(folio))
1817 if (unlikely(folio != xas_reload(&xas))) {
1828 * __filemap_get_folio - Find and get a reference to a folio.
1829 * @mapping: The address_space to search.
1830 * @index: The page index.
1831 * @fgp_flags: %FGP flags modify how the folio is returned.
1832 * @gfp: Memory allocation flags to use if %FGP_CREAT is specified.
1834 * Looks up the page cache entry at @mapping & @index.
1836 * @fgp_flags can be zero or more of these flags:
1838 * * %FGP_ACCESSED - The folio will be marked accessed.
1839 * * %FGP_LOCK - The folio is returned locked.
1840 * * %FGP_ENTRY - If there is a shadow / swap / DAX entry, return it
1841 * instead of allocating a new folio to replace it.
1842 * * %FGP_CREAT - If no page is present then a new page is allocated using
1843 * @gfp and added to the page cache and the VM's LRU list.
1844 * The page is returned locked and with an increased refcount.
1845 * * %FGP_FOR_MMAP - The caller wants to do its own locking dance if the
1846 * page is already in cache. If the page was allocated, unlock it before
1847 * returning so the caller can do the same dance.
1848 * * %FGP_WRITE - The page will be written to by the caller.
1849 * * %FGP_NOFS - __GFP_FS will get cleared in gfp.
1850 * * %FGP_NOWAIT - Don't get blocked by page lock.
1851 * * %FGP_STABLE - Wait for the folio to be stable (finished writeback)
1853 * If %FGP_LOCK or %FGP_CREAT are specified then the function may sleep even
1854 * if the %GFP flags specified for %FGP_CREAT are atomic.
1856 * If there is a page cache page, it is returned with an increased refcount.
1858 * Return: The found folio or %NULL otherwise.
1860 struct folio *__filemap_get_folio(struct address_space *mapping, pgoff_t index,
1861 int fgp_flags, gfp_t gfp)
1863 struct folio *folio;
1866 folio = mapping_get_entry(mapping, index);
1867 if (xa_is_value(folio)) {
1868 if (fgp_flags & FGP_ENTRY)
1875 if (fgp_flags & FGP_LOCK) {
1876 if (fgp_flags & FGP_NOWAIT) {
1877 if (!folio_trylock(folio)) {
1885 /* Has the page been truncated? */
1886 if (unlikely(folio->mapping != mapping)) {
1887 folio_unlock(folio);
1891 VM_BUG_ON_FOLIO(!folio_contains(folio, index), folio);
1894 if (fgp_flags & FGP_ACCESSED)
1895 folio_mark_accessed(folio);
1896 else if (fgp_flags & FGP_WRITE) {
1897 /* Clear idle flag for buffer write */
1898 if (folio_test_idle(folio))
1899 folio_clear_idle(folio);
1902 if (fgp_flags & FGP_STABLE)
1903 folio_wait_stable(folio);
1905 if (!folio && (fgp_flags & FGP_CREAT)) {
1907 if ((fgp_flags & FGP_WRITE) && mapping_can_writeback(mapping))
1909 if (fgp_flags & FGP_NOFS)
1912 folio = filemap_alloc_folio(gfp, 0);
1916 if (WARN_ON_ONCE(!(fgp_flags & (FGP_LOCK | FGP_FOR_MMAP))))
1917 fgp_flags |= FGP_LOCK;
1919 /* Init accessed so avoid atomic mark_page_accessed later */
1920 if (fgp_flags & FGP_ACCESSED)
1921 __folio_set_referenced(folio);
1923 err = filemap_add_folio(mapping, folio, index, gfp);
1924 if (unlikely(err)) {
1932 * filemap_add_folio locks the page, and for mmap
1933 * we expect an unlocked page.
1935 if (folio && (fgp_flags & FGP_FOR_MMAP))
1936 folio_unlock(folio);
1941 EXPORT_SYMBOL(__filemap_get_folio);
1943 static inline struct folio *find_get_entry(struct xa_state *xas, pgoff_t max,
1946 struct folio *folio;
1949 if (mark == XA_PRESENT)
1950 folio = xas_find(xas, max);
1952 folio = xas_find_marked(xas, max, mark);
1954 if (xas_retry(xas, folio))
1957 * A shadow entry of a recently evicted page, a swap
1958 * entry from shmem/tmpfs or a DAX entry. Return it
1959 * without attempting to raise page count.
1961 if (!folio || xa_is_value(folio))
1964 if (!folio_try_get_rcu(folio))
1967 if (unlikely(folio != xas_reload(xas))) {
1979 * find_get_entries - gang pagecache lookup
1980 * @mapping: The address_space to search
1981 * @start: The starting page cache index
1982 * @end: The final page index (inclusive).
1983 * @fbatch: Where the resulting entries are placed.
1984 * @indices: The cache indices corresponding to the entries in @entries
1986 * find_get_entries() will search for and return a batch of entries in
1987 * the mapping. The entries are placed in @fbatch. find_get_entries()
1988 * takes a reference on any actual folios it returns.
1990 * The entries have ascending indexes. The indices may not be consecutive
1991 * due to not-present entries or large folios.
1993 * Any shadow entries of evicted folios, or swap entries from
1994 * shmem/tmpfs, are included in the returned array.
1996 * Return: The number of entries which were found.
1998 unsigned find_get_entries(struct address_space *mapping, pgoff_t start,
1999 pgoff_t end, struct folio_batch *fbatch, pgoff_t *indices)
2001 XA_STATE(xas, &mapping->i_pages, start);
2002 struct folio *folio;
2005 while ((folio = find_get_entry(&xas, end, XA_PRESENT)) != NULL) {
2006 indices[fbatch->nr] = xas.xa_index;
2007 if (!folio_batch_add(fbatch, folio))
2012 return folio_batch_count(fbatch);
2016 * find_lock_entries - Find a batch of pagecache entries.
2017 * @mapping: The address_space to search.
2018 * @start: The starting page cache index.
2019 * @end: The final page index (inclusive).
2020 * @fbatch: Where the resulting entries are placed.
2021 * @indices: The cache indices of the entries in @fbatch.
2023 * find_lock_entries() will return a batch of entries from @mapping.
2024 * Swap, shadow and DAX entries are included. Folios are returned
2025 * locked and with an incremented refcount. Folios which are locked
2026 * by somebody else or under writeback are skipped. Folios which are
2027 * partially outside the range are not returned.
2029 * The entries have ascending indexes. The indices may not be consecutive
2030 * due to not-present entries, large folios, folios which could not be
2031 * locked or folios under writeback.
2033 * Return: The number of entries which were found.
2035 unsigned find_lock_entries(struct address_space *mapping, pgoff_t start,
2036 pgoff_t end, struct folio_batch *fbatch, pgoff_t *indices)
2038 XA_STATE(xas, &mapping->i_pages, start);
2039 struct folio *folio;
2042 while ((folio = find_get_entry(&xas, end, XA_PRESENT))) {
2043 if (!xa_is_value(folio)) {
2044 if (folio->index < start)
2046 if (folio->index + folio_nr_pages(folio) - 1 > end)
2048 if (!folio_trylock(folio))
2050 if (folio->mapping != mapping ||
2051 folio_test_writeback(folio))
2053 VM_BUG_ON_FOLIO(!folio_contains(folio, xas.xa_index),
2056 indices[fbatch->nr] = xas.xa_index;
2057 if (!folio_batch_add(fbatch, folio))
2061 folio_unlock(folio);
2067 return folio_batch_count(fbatch);
2071 bool folio_more_pages(struct folio *folio, pgoff_t index, pgoff_t max)
2073 if (!folio_test_large(folio) || folio_test_hugetlb(folio))
2077 return index < folio->index + folio_nr_pages(folio) - 1;
2081 * find_get_pages_range - gang pagecache lookup
2082 * @mapping: The address_space to search
2083 * @start: The starting page index
2084 * @end: The final page index (inclusive)
2085 * @nr_pages: The maximum number of pages
2086 * @pages: Where the resulting pages are placed
2088 * find_get_pages_range() will search for and return a group of up to @nr_pages
2089 * pages in the mapping starting at index @start and up to index @end
2090 * (inclusive). The pages are placed at @pages. find_get_pages_range() takes
2091 * a reference against the returned pages.
2093 * The search returns a group of mapping-contiguous pages with ascending
2094 * indexes. There may be holes in the indices due to not-present pages.
2095 * We also update @start to index the next page for the traversal.
2097 * Return: the number of pages which were found. If this number is
2098 * smaller than @nr_pages, the end of specified range has been
2101 unsigned find_get_pages_range(struct address_space *mapping, pgoff_t *start,
2102 pgoff_t end, unsigned int nr_pages,
2103 struct page **pages)
2105 XA_STATE(xas, &mapping->i_pages, *start);
2106 struct folio *folio;
2109 if (unlikely(!nr_pages))
2113 while ((folio = find_get_entry(&xas, end, XA_PRESENT))) {
2114 /* Skip over shadow, swap and DAX entries */
2115 if (xa_is_value(folio))
2119 pages[ret] = folio_file_page(folio, xas.xa_index);
2120 if (++ret == nr_pages) {
2121 *start = xas.xa_index + 1;
2124 if (folio_more_pages(folio, xas.xa_index, end)) {
2126 folio_ref_inc(folio);
2132 * We come here when there is no page beyond @end. We take care to not
2133 * overflow the index @start as it confuses some of the callers. This
2134 * breaks the iteration when there is a page at index -1 but that is
2135 * already broken anyway.
2137 if (end == (pgoff_t)-1)
2138 *start = (pgoff_t)-1;
2148 * find_get_pages_contig - gang contiguous pagecache lookup
2149 * @mapping: The address_space to search
2150 * @index: The starting page index
2151 * @nr_pages: The maximum number of pages
2152 * @pages: Where the resulting pages are placed
2154 * find_get_pages_contig() works exactly like find_get_pages(), except
2155 * that the returned number of pages are guaranteed to be contiguous.
2157 * Return: the number of pages which were found.
2159 unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index,
2160 unsigned int nr_pages, struct page **pages)
2162 XA_STATE(xas, &mapping->i_pages, index);
2163 struct folio *folio;
2164 unsigned int ret = 0;
2166 if (unlikely(!nr_pages))
2170 for (folio = xas_load(&xas); folio; folio = xas_next(&xas)) {
2171 if (xas_retry(&xas, folio))
2174 * If the entry has been swapped out, we can stop looking.
2175 * No current caller is looking for DAX entries.
2177 if (xa_is_value(folio))
2180 if (!folio_try_get_rcu(folio))
2183 if (unlikely(folio != xas_reload(&xas)))
2187 pages[ret] = folio_file_page(folio, xas.xa_index);
2188 if (++ret == nr_pages)
2190 if (folio_more_pages(folio, xas.xa_index, ULONG_MAX)) {
2192 folio_ref_inc(folio);
2204 EXPORT_SYMBOL(find_get_pages_contig);
2207 * find_get_pages_range_tag - Find and return head pages matching @tag.
2208 * @mapping: the address_space to search
2209 * @index: the starting page index
2210 * @end: The final page index (inclusive)
2211 * @tag: the tag index
2212 * @nr_pages: the maximum number of pages
2213 * @pages: where the resulting pages are placed
2215 * Like find_get_pages(), except we only return head pages which are tagged
2216 * with @tag. @index is updated to the index immediately after the last
2217 * page we return, ready for the next iteration.
2219 * Return: the number of pages which were found.
2221 unsigned find_get_pages_range_tag(struct address_space *mapping, pgoff_t *index,
2222 pgoff_t end, xa_mark_t tag, unsigned int nr_pages,
2223 struct page **pages)
2225 XA_STATE(xas, &mapping->i_pages, *index);
2226 struct folio *folio;
2229 if (unlikely(!nr_pages))
2233 while ((folio = find_get_entry(&xas, end, tag))) {
2235 * Shadow entries should never be tagged, but this iteration
2236 * is lockless so there is a window for page reclaim to evict
2237 * a page we saw tagged. Skip over it.
2239 if (xa_is_value(folio))
2242 pages[ret] = &folio->page;
2243 if (++ret == nr_pages) {
2244 *index = folio->index + folio_nr_pages(folio);
2250 * We come here when we got to @end. We take care to not overflow the
2251 * index @index as it confuses some of the callers. This breaks the
2252 * iteration when there is a page at index -1 but that is already
2255 if (end == (pgoff_t)-1)
2256 *index = (pgoff_t)-1;
2264 EXPORT_SYMBOL(find_get_pages_range_tag);
2267 * CD/DVDs are error prone. When a medium error occurs, the driver may fail
2268 * a _large_ part of the i/o request. Imagine the worst scenario:
2270 * ---R__________________________________________B__________
2271 * ^ reading here ^ bad block(assume 4k)
2273 * read(R) => miss => readahead(R...B) => media error => frustrating retries
2274 * => failing the whole request => read(R) => read(R+1) =>
2275 * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
2276 * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
2277 * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
2279 * It is going insane. Fix it by quickly scaling down the readahead size.
2281 static void shrink_readahead_size_eio(struct file_ra_state *ra)
2287 * filemap_get_read_batch - Get a batch of folios for read
2289 * Get a batch of folios which represent a contiguous range of bytes in
2290 * the file. No exceptional entries will be returned. If @index is in
2291 * the middle of a folio, the entire folio will be returned. The last
2292 * folio in the batch may have the readahead flag set or the uptodate flag
2293 * clear so that the caller can take the appropriate action.
2295 static void filemap_get_read_batch(struct address_space *mapping,
2296 pgoff_t index, pgoff_t max, struct folio_batch *fbatch)
2298 XA_STATE(xas, &mapping->i_pages, index);
2299 struct folio *folio;
2302 for (folio = xas_load(&xas); folio; folio = xas_next(&xas)) {
2303 if (xas_retry(&xas, folio))
2305 if (xas.xa_index > max || xa_is_value(folio))
2307 if (!folio_try_get_rcu(folio))
2310 if (unlikely(folio != xas_reload(&xas)))
2313 if (!folio_batch_add(fbatch, folio))
2315 if (!folio_test_uptodate(folio))
2317 if (folio_test_readahead(folio))
2319 xas_advance(&xas, folio->index + folio_nr_pages(folio) - 1);
2329 static int filemap_read_folio(struct file *file, struct address_space *mapping,
2330 struct folio *folio)
2335 * A previous I/O error may have been due to temporary failures,
2336 * eg. multipath errors. PG_error will be set again if readpage
2339 folio_clear_error(folio);
2340 /* Start the actual read. The read will unlock the page. */
2341 error = mapping->a_ops->readpage(file, &folio->page);
2345 error = folio_wait_locked_killable(folio);
2348 if (folio_test_uptodate(folio))
2350 shrink_readahead_size_eio(&file->f_ra);
2354 static bool filemap_range_uptodate(struct address_space *mapping,
2355 loff_t pos, struct iov_iter *iter, struct folio *folio)
2359 if (folio_test_uptodate(folio))
2361 /* pipes can't handle partially uptodate pages */
2362 if (iov_iter_is_pipe(iter))
2364 if (!mapping->a_ops->is_partially_uptodate)
2366 if (mapping->host->i_blkbits >= folio_shift(folio))
2369 count = iter->count;
2370 if (folio_pos(folio) > pos) {
2371 count -= folio_pos(folio) - pos;
2374 pos -= folio_pos(folio);
2377 return mapping->a_ops->is_partially_uptodate(&folio->page, pos, count);
2380 static int filemap_update_page(struct kiocb *iocb,
2381 struct address_space *mapping, struct iov_iter *iter,
2382 struct folio *folio)
2386 if (iocb->ki_flags & IOCB_NOWAIT) {
2387 if (!filemap_invalidate_trylock_shared(mapping))
2390 filemap_invalidate_lock_shared(mapping);
2393 if (!folio_trylock(folio)) {
2395 if (iocb->ki_flags & (IOCB_NOWAIT | IOCB_NOIO))
2396 goto unlock_mapping;
2397 if (!(iocb->ki_flags & IOCB_WAITQ)) {
2398 filemap_invalidate_unlock_shared(mapping);
2400 * This is where we usually end up waiting for a
2401 * previously submitted readahead to finish.
2403 folio_put_wait_locked(folio, TASK_KILLABLE);
2404 return AOP_TRUNCATED_PAGE;
2406 error = __folio_lock_async(folio, iocb->ki_waitq);
2408 goto unlock_mapping;
2411 error = AOP_TRUNCATED_PAGE;
2412 if (!folio->mapping)
2416 if (filemap_range_uptodate(mapping, iocb->ki_pos, iter, folio))
2420 if (iocb->ki_flags & (IOCB_NOIO | IOCB_NOWAIT | IOCB_WAITQ))
2423 error = filemap_read_folio(iocb->ki_filp, mapping, folio);
2424 goto unlock_mapping;
2426 folio_unlock(folio);
2428 filemap_invalidate_unlock_shared(mapping);
2429 if (error == AOP_TRUNCATED_PAGE)
2434 static int filemap_create_folio(struct file *file,
2435 struct address_space *mapping, pgoff_t index,
2436 struct folio_batch *fbatch)
2438 struct folio *folio;
2441 folio = filemap_alloc_folio(mapping_gfp_mask(mapping), 0);
2446 * Protect against truncate / hole punch. Grabbing invalidate_lock
2447 * here assures we cannot instantiate and bring uptodate new
2448 * pagecache folios after evicting page cache during truncate
2449 * and before actually freeing blocks. Note that we could
2450 * release invalidate_lock after inserting the folio into
2451 * the page cache as the locked folio would then be enough to
2452 * synchronize with hole punching. But there are code paths
2453 * such as filemap_update_page() filling in partially uptodate
2454 * pages or ->readpages() that need to hold invalidate_lock
2455 * while mapping blocks for IO so let's hold the lock here as
2456 * well to keep locking rules simple.
2458 filemap_invalidate_lock_shared(mapping);
2459 error = filemap_add_folio(mapping, folio, index,
2460 mapping_gfp_constraint(mapping, GFP_KERNEL));
2461 if (error == -EEXIST)
2462 error = AOP_TRUNCATED_PAGE;
2466 error = filemap_read_folio(file, mapping, folio);
2470 filemap_invalidate_unlock_shared(mapping);
2471 folio_batch_add(fbatch, folio);
2474 filemap_invalidate_unlock_shared(mapping);
2479 static int filemap_readahead(struct kiocb *iocb, struct file *file,
2480 struct address_space *mapping, struct folio *folio,
2483 DEFINE_READAHEAD(ractl, file, &file->f_ra, mapping, folio->index);
2485 if (iocb->ki_flags & IOCB_NOIO)
2487 page_cache_async_ra(&ractl, folio, last_index - folio->index);
2491 static int filemap_get_pages(struct kiocb *iocb, struct iov_iter *iter,
2492 struct folio_batch *fbatch)
2494 struct file *filp = iocb->ki_filp;
2495 struct address_space *mapping = filp->f_mapping;
2496 struct file_ra_state *ra = &filp->f_ra;
2497 pgoff_t index = iocb->ki_pos >> PAGE_SHIFT;
2499 struct folio *folio;
2502 last_index = DIV_ROUND_UP(iocb->ki_pos + iter->count, PAGE_SIZE);
2504 if (fatal_signal_pending(current))
2507 filemap_get_read_batch(mapping, index, last_index, fbatch);
2508 if (!folio_batch_count(fbatch)) {
2509 if (iocb->ki_flags & IOCB_NOIO)
2511 page_cache_sync_readahead(mapping, ra, filp, index,
2512 last_index - index);
2513 filemap_get_read_batch(mapping, index, last_index, fbatch);
2515 if (!folio_batch_count(fbatch)) {
2516 if (iocb->ki_flags & (IOCB_NOWAIT | IOCB_WAITQ))
2518 err = filemap_create_folio(filp, mapping,
2519 iocb->ki_pos >> PAGE_SHIFT, fbatch);
2520 if (err == AOP_TRUNCATED_PAGE)
2525 folio = fbatch->folios[folio_batch_count(fbatch) - 1];
2526 if (folio_test_readahead(folio)) {
2527 err = filemap_readahead(iocb, filp, mapping, folio, last_index);
2531 if (!folio_test_uptodate(folio)) {
2532 if ((iocb->ki_flags & IOCB_WAITQ) &&
2533 folio_batch_count(fbatch) > 1)
2534 iocb->ki_flags |= IOCB_NOWAIT;
2535 err = filemap_update_page(iocb, mapping, iter, folio);
2544 if (likely(--fbatch->nr))
2546 if (err == AOP_TRUNCATED_PAGE)
2552 * filemap_read - Read data from the page cache.
2553 * @iocb: The iocb to read.
2554 * @iter: Destination for the data.
2555 * @already_read: Number of bytes already read by the caller.
2557 * Copies data from the page cache. If the data is not currently present,
2558 * uses the readahead and readpage address_space operations to fetch it.
2560 * Return: Total number of bytes copied, including those already read by
2561 * the caller. If an error happens before any bytes are copied, returns
2562 * a negative error number.
2564 ssize_t filemap_read(struct kiocb *iocb, struct iov_iter *iter,
2565 ssize_t already_read)
2567 struct file *filp = iocb->ki_filp;
2568 struct file_ra_state *ra = &filp->f_ra;
2569 struct address_space *mapping = filp->f_mapping;
2570 struct inode *inode = mapping->host;
2571 struct folio_batch fbatch;
2573 bool writably_mapped;
2574 loff_t isize, end_offset;
2576 if (unlikely(iocb->ki_pos >= inode->i_sb->s_maxbytes))
2578 if (unlikely(!iov_iter_count(iter)))
2581 iov_iter_truncate(iter, inode->i_sb->s_maxbytes);
2582 folio_batch_init(&fbatch);
2588 * If we've already successfully copied some data, then we
2589 * can no longer safely return -EIOCBQUEUED. Hence mark
2590 * an async read NOWAIT at that point.
2592 if ((iocb->ki_flags & IOCB_WAITQ) && already_read)
2593 iocb->ki_flags |= IOCB_NOWAIT;
2595 if (unlikely(iocb->ki_pos >= i_size_read(inode)))
2598 error = filemap_get_pages(iocb, iter, &fbatch);
2603 * i_size must be checked after we know the pages are Uptodate.
2605 * Checking i_size after the check allows us to calculate
2606 * the correct value for "nr", which means the zero-filled
2607 * part of the page is not copied back to userspace (unless
2608 * another truncate extends the file - this is desired though).
2610 isize = i_size_read(inode);
2611 if (unlikely(iocb->ki_pos >= isize))
2613 end_offset = min_t(loff_t, isize, iocb->ki_pos + iter->count);
2616 * Once we start copying data, we don't want to be touching any
2617 * cachelines that might be contended:
2619 writably_mapped = mapping_writably_mapped(mapping);
2622 * When a sequential read accesses a page several times, only
2623 * mark it as accessed the first time.
2625 if (iocb->ki_pos >> PAGE_SHIFT !=
2626 ra->prev_pos >> PAGE_SHIFT)
2627 folio_mark_accessed(fbatch.folios[0]);
2629 for (i = 0; i < folio_batch_count(&fbatch); i++) {
2630 struct folio *folio = fbatch.folios[i];
2631 size_t fsize = folio_size(folio);
2632 size_t offset = iocb->ki_pos & (fsize - 1);
2633 size_t bytes = min_t(loff_t, end_offset - iocb->ki_pos,
2637 if (end_offset < folio_pos(folio))
2640 folio_mark_accessed(folio);
2642 * If users can be writing to this folio using arbitrary
2643 * virtual addresses, take care of potential aliasing
2644 * before reading the folio on the kernel side.
2646 if (writably_mapped)
2647 flush_dcache_folio(folio);
2649 copied = copy_folio_to_iter(folio, offset, bytes, iter);
2651 already_read += copied;
2652 iocb->ki_pos += copied;
2653 ra->prev_pos = iocb->ki_pos;
2655 if (copied < bytes) {
2661 for (i = 0; i < folio_batch_count(&fbatch); i++)
2662 folio_put(fbatch.folios[i]);
2663 folio_batch_init(&fbatch);
2664 } while (iov_iter_count(iter) && iocb->ki_pos < isize && !error);
2666 file_accessed(filp);
2668 return already_read ? already_read : error;
2670 EXPORT_SYMBOL_GPL(filemap_read);
2673 * generic_file_read_iter - generic filesystem read routine
2674 * @iocb: kernel I/O control block
2675 * @iter: destination for the data read
2677 * This is the "read_iter()" routine for all filesystems
2678 * that can use the page cache directly.
2680 * The IOCB_NOWAIT flag in iocb->ki_flags indicates that -EAGAIN shall
2681 * be returned when no data can be read without waiting for I/O requests
2682 * to complete; it doesn't prevent readahead.
2684 * The IOCB_NOIO flag in iocb->ki_flags indicates that no new I/O
2685 * requests shall be made for the read or for readahead. When no data
2686 * can be read, -EAGAIN shall be returned. When readahead would be
2687 * triggered, a partial, possibly empty read shall be returned.
2690 * * number of bytes copied, even for partial reads
2691 * * negative error code (or 0 if IOCB_NOIO) if nothing was read
2694 generic_file_read_iter(struct kiocb *iocb, struct iov_iter *iter)
2696 size_t count = iov_iter_count(iter);
2700 return 0; /* skip atime */
2702 if (iocb->ki_flags & IOCB_DIRECT) {
2703 struct file *file = iocb->ki_filp;
2704 struct address_space *mapping = file->f_mapping;
2705 struct inode *inode = mapping->host;
2707 if (iocb->ki_flags & IOCB_NOWAIT) {
2708 if (filemap_range_needs_writeback(mapping, iocb->ki_pos,
2709 iocb->ki_pos + count - 1))
2712 retval = filemap_write_and_wait_range(mapping,
2714 iocb->ki_pos + count - 1);
2719 file_accessed(file);
2721 retval = mapping->a_ops->direct_IO(iocb, iter);
2723 iocb->ki_pos += retval;
2726 if (retval != -EIOCBQUEUED)
2727 iov_iter_revert(iter, count - iov_iter_count(iter));
2730 * Btrfs can have a short DIO read if we encounter
2731 * compressed extents, so if there was an error, or if
2732 * we've already read everything we wanted to, or if
2733 * there was a short read because we hit EOF, go ahead
2734 * and return. Otherwise fallthrough to buffered io for
2735 * the rest of the read. Buffered reads will not work for
2736 * DAX files, so don't bother trying.
2738 if (retval < 0 || !count || IS_DAX(inode))
2740 if (iocb->ki_pos >= i_size_read(inode))
2744 return filemap_read(iocb, iter, retval);
2746 EXPORT_SYMBOL(generic_file_read_iter);
2748 static inline loff_t folio_seek_hole_data(struct xa_state *xas,
2749 struct address_space *mapping, struct folio *folio,
2750 loff_t start, loff_t end, bool seek_data)
2752 const struct address_space_operations *ops = mapping->a_ops;
2753 size_t offset, bsz = i_blocksize(mapping->host);
2755 if (xa_is_value(folio) || folio_test_uptodate(folio))
2756 return seek_data ? start : end;
2757 if (!ops->is_partially_uptodate)
2758 return seek_data ? end : start;
2763 if (unlikely(folio->mapping != mapping))
2766 offset = offset_in_folio(folio, start) & ~(bsz - 1);
2769 if (ops->is_partially_uptodate(&folio->page, offset, bsz) ==
2772 start = (start + bsz) & ~(bsz - 1);
2774 } while (offset < folio_size(folio));
2776 folio_unlock(folio);
2781 static inline size_t seek_folio_size(struct xa_state *xas, struct folio *folio)
2783 if (xa_is_value(folio))
2784 return PAGE_SIZE << xa_get_order(xas->xa, xas->xa_index);
2785 return folio_size(folio);
2789 * mapping_seek_hole_data - Seek for SEEK_DATA / SEEK_HOLE in the page cache.
2790 * @mapping: Address space to search.
2791 * @start: First byte to consider.
2792 * @end: Limit of search (exclusive).
2793 * @whence: Either SEEK_HOLE or SEEK_DATA.
2795 * If the page cache knows which blocks contain holes and which blocks
2796 * contain data, your filesystem can use this function to implement
2797 * SEEK_HOLE and SEEK_DATA. This is useful for filesystems which are
2798 * entirely memory-based such as tmpfs, and filesystems which support
2799 * unwritten extents.
2801 * Return: The requested offset on success, or -ENXIO if @whence specifies
2802 * SEEK_DATA and there is no data after @start. There is an implicit hole
2803 * after @end - 1, so SEEK_HOLE returns @end if all the bytes between @start
2804 * and @end contain data.
2806 loff_t mapping_seek_hole_data(struct address_space *mapping, loff_t start,
2807 loff_t end, int whence)
2809 XA_STATE(xas, &mapping->i_pages, start >> PAGE_SHIFT);
2810 pgoff_t max = (end - 1) >> PAGE_SHIFT;
2811 bool seek_data = (whence == SEEK_DATA);
2812 struct folio *folio;
2818 while ((folio = find_get_entry(&xas, max, XA_PRESENT))) {
2819 loff_t pos = (u64)xas.xa_index << PAGE_SHIFT;
2828 seek_size = seek_folio_size(&xas, folio);
2829 pos = round_up((u64)pos + 1, seek_size);
2830 start = folio_seek_hole_data(&xas, mapping, folio, start, pos,
2836 if (seek_size > PAGE_SIZE)
2837 xas_set(&xas, pos >> PAGE_SHIFT);
2838 if (!xa_is_value(folio))
2845 if (folio && !xa_is_value(folio))
2853 #define MMAP_LOTSAMISS (100)
2855 * lock_folio_maybe_drop_mmap - lock the page, possibly dropping the mmap_lock
2856 * @vmf - the vm_fault for this fault.
2857 * @folio - the folio to lock.
2858 * @fpin - the pointer to the file we may pin (or is already pinned).
2860 * This works similar to lock_folio_or_retry in that it can drop the
2861 * mmap_lock. It differs in that it actually returns the folio locked
2862 * if it returns 1 and 0 if it couldn't lock the folio. If we did have
2863 * to drop the mmap_lock then fpin will point to the pinned file and
2864 * needs to be fput()'ed at a later point.
2866 static int lock_folio_maybe_drop_mmap(struct vm_fault *vmf, struct folio *folio,
2869 if (folio_trylock(folio))
2873 * NOTE! This will make us return with VM_FAULT_RETRY, but with
2874 * the mmap_lock still held. That's how FAULT_FLAG_RETRY_NOWAIT
2875 * is supposed to work. We have way too many special cases..
2877 if (vmf->flags & FAULT_FLAG_RETRY_NOWAIT)
2880 *fpin = maybe_unlock_mmap_for_io(vmf, *fpin);
2881 if (vmf->flags & FAULT_FLAG_KILLABLE) {
2882 if (__folio_lock_killable(folio)) {
2884 * We didn't have the right flags to drop the mmap_lock,
2885 * but all fault_handlers only check for fatal signals
2886 * if we return VM_FAULT_RETRY, so we need to drop the
2887 * mmap_lock here and return 0 if we don't have a fpin.
2890 mmap_read_unlock(vmf->vma->vm_mm);
2894 __folio_lock(folio);
2900 * Synchronous readahead happens when we don't even find a page in the page
2901 * cache at all. We don't want to perform IO under the mmap sem, so if we have
2902 * to drop the mmap sem we return the file that was pinned in order for us to do
2903 * that. If we didn't pin a file then we return NULL. The file that is
2904 * returned needs to be fput()'ed when we're done with it.
2906 static struct file *do_sync_mmap_readahead(struct vm_fault *vmf)
2908 struct file *file = vmf->vma->vm_file;
2909 struct file_ra_state *ra = &file->f_ra;
2910 struct address_space *mapping = file->f_mapping;
2911 DEFINE_READAHEAD(ractl, file, ra, mapping, vmf->pgoff);
2912 struct file *fpin = NULL;
2913 unsigned int mmap_miss;
2915 /* If we don't want any read-ahead, don't bother */
2916 if (vmf->vma->vm_flags & VM_RAND_READ)
2921 if (vmf->vma->vm_flags & VM_SEQ_READ) {
2922 fpin = maybe_unlock_mmap_for_io(vmf, fpin);
2923 page_cache_sync_ra(&ractl, ra->ra_pages);
2927 /* Avoid banging the cache line if not needed */
2928 mmap_miss = READ_ONCE(ra->mmap_miss);
2929 if (mmap_miss < MMAP_LOTSAMISS * 10)
2930 WRITE_ONCE(ra->mmap_miss, ++mmap_miss);
2933 * Do we miss much more than hit in this file? If so,
2934 * stop bothering with read-ahead. It will only hurt.
2936 if (mmap_miss > MMAP_LOTSAMISS)
2942 fpin = maybe_unlock_mmap_for_io(vmf, fpin);
2943 ra->start = max_t(long, 0, vmf->pgoff - ra->ra_pages / 2);
2944 ra->size = ra->ra_pages;
2945 ra->async_size = ra->ra_pages / 4;
2946 ractl._index = ra->start;
2947 do_page_cache_ra(&ractl, ra->size, ra->async_size);
2952 * Asynchronous readahead happens when we find the page and PG_readahead,
2953 * so we want to possibly extend the readahead further. We return the file that
2954 * was pinned if we have to drop the mmap_lock in order to do IO.
2956 static struct file *do_async_mmap_readahead(struct vm_fault *vmf,
2957 struct folio *folio)
2959 struct file *file = vmf->vma->vm_file;
2960 struct file_ra_state *ra = &file->f_ra;
2961 DEFINE_READAHEAD(ractl, file, ra, file->f_mapping, vmf->pgoff);
2962 struct file *fpin = NULL;
2963 unsigned int mmap_miss;
2965 /* If we don't want any read-ahead, don't bother */
2966 if (vmf->vma->vm_flags & VM_RAND_READ || !ra->ra_pages)
2969 mmap_miss = READ_ONCE(ra->mmap_miss);
2971 WRITE_ONCE(ra->mmap_miss, --mmap_miss);
2973 if (folio_test_readahead(folio)) {
2974 fpin = maybe_unlock_mmap_for_io(vmf, fpin);
2975 page_cache_async_ra(&ractl, folio, ra->ra_pages);
2981 * filemap_fault - read in file data for page fault handling
2982 * @vmf: struct vm_fault containing details of the fault
2984 * filemap_fault() is invoked via the vma operations vector for a
2985 * mapped memory region to read in file data during a page fault.
2987 * The goto's are kind of ugly, but this streamlines the normal case of having
2988 * it in the page cache, and handles the special cases reasonably without
2989 * having a lot of duplicated code.
2991 * vma->vm_mm->mmap_lock must be held on entry.
2993 * If our return value has VM_FAULT_RETRY set, it's because the mmap_lock
2994 * may be dropped before doing I/O or by lock_folio_maybe_drop_mmap().
2996 * If our return value does not have VM_FAULT_RETRY set, the mmap_lock
2997 * has not been released.
2999 * We never return with VM_FAULT_RETRY and a bit from VM_FAULT_ERROR set.
3001 * Return: bitwise-OR of %VM_FAULT_ codes.
3003 vm_fault_t filemap_fault(struct vm_fault *vmf)
3006 struct file *file = vmf->vma->vm_file;
3007 struct file *fpin = NULL;
3008 struct address_space *mapping = file->f_mapping;
3009 struct inode *inode = mapping->host;
3010 pgoff_t max_idx, index = vmf->pgoff;
3011 struct folio *folio;
3013 bool mapping_locked = false;
3015 max_idx = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
3016 if (unlikely(index >= max_idx))
3017 return VM_FAULT_SIGBUS;
3020 * Do we have something in the page cache already?
3022 folio = filemap_get_folio(mapping, index);
3023 if (likely(folio)) {
3025 * We found the page, so try async readahead before waiting for
3028 if (!(vmf->flags & FAULT_FLAG_TRIED))
3029 fpin = do_async_mmap_readahead(vmf, folio);
3030 if (unlikely(!folio_test_uptodate(folio))) {
3031 filemap_invalidate_lock_shared(mapping);
3032 mapping_locked = true;
3035 /* No page in the page cache at all */
3036 count_vm_event(PGMAJFAULT);
3037 count_memcg_event_mm(vmf->vma->vm_mm, PGMAJFAULT);
3038 ret = VM_FAULT_MAJOR;
3039 fpin = do_sync_mmap_readahead(vmf);
3042 * See comment in filemap_create_folio() why we need
3045 if (!mapping_locked) {
3046 filemap_invalidate_lock_shared(mapping);
3047 mapping_locked = true;
3049 folio = __filemap_get_folio(mapping, index,
3050 FGP_CREAT|FGP_FOR_MMAP,
3055 filemap_invalidate_unlock_shared(mapping);
3056 return VM_FAULT_OOM;
3060 if (!lock_folio_maybe_drop_mmap(vmf, folio, &fpin))
3063 /* Did it get truncated? */
3064 if (unlikely(folio->mapping != mapping)) {
3065 folio_unlock(folio);
3069 VM_BUG_ON_FOLIO(!folio_contains(folio, index), folio);
3072 * We have a locked page in the page cache, now we need to check
3073 * that it's up-to-date. If not, it is going to be due to an error.
3075 if (unlikely(!folio_test_uptodate(folio))) {
3077 * The page was in cache and uptodate and now it is not.
3078 * Strange but possible since we didn't hold the page lock all
3079 * the time. Let's drop everything get the invalidate lock and
3082 if (!mapping_locked) {
3083 folio_unlock(folio);
3087 goto page_not_uptodate;
3091 * We've made it this far and we had to drop our mmap_lock, now is the
3092 * time to return to the upper layer and have it re-find the vma and
3096 folio_unlock(folio);
3100 filemap_invalidate_unlock_shared(mapping);
3103 * Found the page and have a reference on it.
3104 * We must recheck i_size under page lock.
3106 max_idx = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
3107 if (unlikely(index >= max_idx)) {
3108 folio_unlock(folio);
3110 return VM_FAULT_SIGBUS;
3113 vmf->page = folio_file_page(folio, index);
3114 return ret | VM_FAULT_LOCKED;
3118 * Umm, take care of errors if the page isn't up-to-date.
3119 * Try to re-read it _once_. We do this synchronously,
3120 * because there really aren't any performance issues here
3121 * and we need to check for errors.
3123 fpin = maybe_unlock_mmap_for_io(vmf, fpin);
3124 error = filemap_read_folio(file, mapping, folio);
3129 if (!error || error == AOP_TRUNCATED_PAGE)
3131 filemap_invalidate_unlock_shared(mapping);
3133 return VM_FAULT_SIGBUS;
3137 * We dropped the mmap_lock, we need to return to the fault handler to
3138 * re-find the vma and come back and find our hopefully still populated
3144 filemap_invalidate_unlock_shared(mapping);
3147 return ret | VM_FAULT_RETRY;
3149 EXPORT_SYMBOL(filemap_fault);
3151 static bool filemap_map_pmd(struct vm_fault *vmf, struct page *page)
3153 struct mm_struct *mm = vmf->vma->vm_mm;
3155 /* Huge page is mapped? No need to proceed. */
3156 if (pmd_trans_huge(*vmf->pmd)) {
3162 if (pmd_none(*vmf->pmd) && PageTransHuge(page)) {
3163 vm_fault_t ret = do_set_pmd(vmf, page);
3165 /* The page is mapped successfully, reference consumed. */
3171 if (pmd_none(*vmf->pmd))
3172 pmd_install(mm, vmf->pmd, &vmf->prealloc_pte);
3174 /* See comment in handle_pte_fault() */
3175 if (pmd_devmap_trans_unstable(vmf->pmd)) {
3184 static struct folio *next_uptodate_page(struct folio *folio,
3185 struct address_space *mapping,
3186 struct xa_state *xas, pgoff_t end_pgoff)
3188 unsigned long max_idx;
3193 if (xas_retry(xas, folio))
3195 if (xa_is_value(folio))
3197 if (folio_test_locked(folio))
3199 if (!folio_try_get_rcu(folio))
3201 /* Has the page moved or been split? */
3202 if (unlikely(folio != xas_reload(xas)))
3204 if (!folio_test_uptodate(folio) || folio_test_readahead(folio))
3206 if (!folio_trylock(folio))
3208 if (folio->mapping != mapping)
3210 if (!folio_test_uptodate(folio))
3212 max_idx = DIV_ROUND_UP(i_size_read(mapping->host), PAGE_SIZE);
3213 if (xas->xa_index >= max_idx)
3217 folio_unlock(folio);
3220 } while ((folio = xas_next_entry(xas, end_pgoff)) != NULL);
3225 static inline struct folio *first_map_page(struct address_space *mapping,
3226 struct xa_state *xas,
3229 return next_uptodate_page(xas_find(xas, end_pgoff),
3230 mapping, xas, end_pgoff);
3233 static inline struct folio *next_map_page(struct address_space *mapping,
3234 struct xa_state *xas,
3237 return next_uptodate_page(xas_next_entry(xas, end_pgoff),
3238 mapping, xas, end_pgoff);
3241 vm_fault_t filemap_map_pages(struct vm_fault *vmf,
3242 pgoff_t start_pgoff, pgoff_t end_pgoff)
3244 struct vm_area_struct *vma = vmf->vma;
3245 struct file *file = vma->vm_file;
3246 struct address_space *mapping = file->f_mapping;
3247 pgoff_t last_pgoff = start_pgoff;
3249 XA_STATE(xas, &mapping->i_pages, start_pgoff);
3250 struct folio *folio;
3252 unsigned int mmap_miss = READ_ONCE(file->f_ra.mmap_miss);
3256 folio = first_map_page(mapping, &xas, end_pgoff);
3260 if (filemap_map_pmd(vmf, &folio->page)) {
3261 ret = VM_FAULT_NOPAGE;
3265 addr = vma->vm_start + ((start_pgoff - vma->vm_pgoff) << PAGE_SHIFT);
3266 vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, addr, &vmf->ptl);
3269 page = folio_file_page(folio, xas.xa_index);
3270 if (PageHWPoison(page))
3276 addr += (xas.xa_index - last_pgoff) << PAGE_SHIFT;
3277 vmf->pte += xas.xa_index - last_pgoff;
3278 last_pgoff = xas.xa_index;
3280 if (!pte_none(*vmf->pte))
3283 /* We're about to handle the fault */
3284 if (vmf->address == addr)
3285 ret = VM_FAULT_NOPAGE;
3287 do_set_pte(vmf, page, addr);
3288 /* no need to invalidate: a not-present page won't be cached */
3289 update_mmu_cache(vma, addr, vmf->pte);
3290 if (folio_more_pages(folio, xas.xa_index, end_pgoff)) {
3292 folio_ref_inc(folio);
3295 folio_unlock(folio);
3298 if (folio_more_pages(folio, xas.xa_index, end_pgoff)) {
3302 folio_unlock(folio);
3304 } while ((folio = next_map_page(mapping, &xas, end_pgoff)) != NULL);
3305 pte_unmap_unlock(vmf->pte, vmf->ptl);
3308 WRITE_ONCE(file->f_ra.mmap_miss, mmap_miss);
3311 EXPORT_SYMBOL(filemap_map_pages);
3313 vm_fault_t filemap_page_mkwrite(struct vm_fault *vmf)
3315 struct address_space *mapping = vmf->vma->vm_file->f_mapping;
3316 struct folio *folio = page_folio(vmf->page);
3317 vm_fault_t ret = VM_FAULT_LOCKED;
3319 sb_start_pagefault(mapping->host->i_sb);
3320 file_update_time(vmf->vma->vm_file);
3322 if (folio->mapping != mapping) {
3323 folio_unlock(folio);
3324 ret = VM_FAULT_NOPAGE;
3328 * We mark the folio dirty already here so that when freeze is in
3329 * progress, we are guaranteed that writeback during freezing will
3330 * see the dirty folio and writeprotect it again.
3332 folio_mark_dirty(folio);
3333 folio_wait_stable(folio);
3335 sb_end_pagefault(mapping->host->i_sb);
3339 const struct vm_operations_struct generic_file_vm_ops = {
3340 .fault = filemap_fault,
3341 .map_pages = filemap_map_pages,
3342 .page_mkwrite = filemap_page_mkwrite,
3345 /* This is used for a general mmap of a disk file */
3347 int generic_file_mmap(struct file *file, struct vm_area_struct *vma)
3349 struct address_space *mapping = file->f_mapping;
3351 if (!mapping->a_ops->readpage)
3353 file_accessed(file);
3354 vma->vm_ops = &generic_file_vm_ops;
3359 * This is for filesystems which do not implement ->writepage.
3361 int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
3363 if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE))
3365 return generic_file_mmap(file, vma);
3368 vm_fault_t filemap_page_mkwrite(struct vm_fault *vmf)
3370 return VM_FAULT_SIGBUS;
3372 int generic_file_mmap(struct file *file, struct vm_area_struct *vma)
3376 int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
3380 #endif /* CONFIG_MMU */
3382 EXPORT_SYMBOL(filemap_page_mkwrite);
3383 EXPORT_SYMBOL(generic_file_mmap);
3384 EXPORT_SYMBOL(generic_file_readonly_mmap);
3386 static struct folio *do_read_cache_folio(struct address_space *mapping,
3387 pgoff_t index, filler_t filler, void *data, gfp_t gfp)
3389 struct folio *folio;
3392 folio = filemap_get_folio(mapping, index);
3394 folio = filemap_alloc_folio(gfp, 0);
3396 return ERR_PTR(-ENOMEM);
3397 err = filemap_add_folio(mapping, folio, index, gfp);
3398 if (unlikely(err)) {
3402 /* Presumably ENOMEM for xarray node */
3403 return ERR_PTR(err);
3408 err = filler(data, &folio->page);
3410 err = mapping->a_ops->readpage(data, &folio->page);
3414 return ERR_PTR(err);
3417 folio_wait_locked(folio);
3418 if (!folio_test_uptodate(folio)) {
3420 return ERR_PTR(-EIO);
3425 if (folio_test_uptodate(folio))
3428 if (!folio_trylock(folio)) {
3429 folio_put_wait_locked(folio, TASK_UNINTERRUPTIBLE);
3433 /* Folio was truncated from mapping */
3434 if (!folio->mapping) {
3435 folio_unlock(folio);
3440 /* Someone else locked and filled the page in a very small window */
3441 if (folio_test_uptodate(folio)) {
3442 folio_unlock(folio);
3447 * A previous I/O error may have been due to temporary
3449 * Clear page error before actual read, PG_error will be
3450 * set again if read page fails.
3452 folio_clear_error(folio);
3456 folio_mark_accessed(folio);
3461 * read_cache_folio - read into page cache, fill it if needed
3462 * @mapping: the page's address_space
3463 * @index: the page index
3464 * @filler: function to perform the read
3465 * @data: first arg to filler(data, page) function, often left as NULL
3467 * Read into the page cache. If a page already exists, and PageUptodate() is
3468 * not set, try to fill the page and wait for it to become unlocked.
3470 * If the page does not get brought uptodate, return -EIO.
3472 * The function expects mapping->invalidate_lock to be already held.
3474 * Return: up to date page on success, ERR_PTR() on failure.
3476 struct folio *read_cache_folio(struct address_space *mapping, pgoff_t index,
3477 filler_t filler, void *data)
3479 return do_read_cache_folio(mapping, index, filler, data,
3480 mapping_gfp_mask(mapping));
3482 EXPORT_SYMBOL(read_cache_folio);
3484 static struct page *do_read_cache_page(struct address_space *mapping,
3485 pgoff_t index, filler_t *filler, void *data, gfp_t gfp)
3487 struct folio *folio;
3489 folio = do_read_cache_folio(mapping, index, filler, data, gfp);
3491 return &folio->page;
3492 return folio_file_page(folio, index);
3495 struct page *read_cache_page(struct address_space *mapping,
3496 pgoff_t index, filler_t *filler, void *data)
3498 return do_read_cache_page(mapping, index, filler, data,
3499 mapping_gfp_mask(mapping));
3501 EXPORT_SYMBOL(read_cache_page);
3504 * read_cache_page_gfp - read into page cache, using specified page allocation flags.
3505 * @mapping: the page's address_space
3506 * @index: the page index
3507 * @gfp: the page allocator flags to use if allocating
3509 * This is the same as "read_mapping_page(mapping, index, NULL)", but with
3510 * any new page allocations done using the specified allocation flags.
3512 * If the page does not get brought uptodate, return -EIO.
3514 * The function expects mapping->invalidate_lock to be already held.
3516 * Return: up to date page on success, ERR_PTR() on failure.
3518 struct page *read_cache_page_gfp(struct address_space *mapping,
3522 return do_read_cache_page(mapping, index, NULL, NULL, gfp);
3524 EXPORT_SYMBOL(read_cache_page_gfp);
3526 int pagecache_write_begin(struct file *file, struct address_space *mapping,
3527 loff_t pos, unsigned len, unsigned flags,
3528 struct page **pagep, void **fsdata)
3530 const struct address_space_operations *aops = mapping->a_ops;
3532 return aops->write_begin(file, mapping, pos, len, flags,
3535 EXPORT_SYMBOL(pagecache_write_begin);
3537 int pagecache_write_end(struct file *file, struct address_space *mapping,
3538 loff_t pos, unsigned len, unsigned copied,
3539 struct page *page, void *fsdata)
3541 const struct address_space_operations *aops = mapping->a_ops;
3543 return aops->write_end(file, mapping, pos, len, copied, page, fsdata);
3545 EXPORT_SYMBOL(pagecache_write_end);
3548 * Warn about a page cache invalidation failure during a direct I/O write.
3550 void dio_warn_stale_pagecache(struct file *filp)
3552 static DEFINE_RATELIMIT_STATE(_rs, 86400 * HZ, DEFAULT_RATELIMIT_BURST);
3556 errseq_set(&filp->f_mapping->wb_err, -EIO);
3557 if (__ratelimit(&_rs)) {
3558 path = file_path(filp, pathname, sizeof(pathname));
3561 pr_crit("Page cache invalidation failure on direct I/O. Possible data corruption due to collision with buffered I/O!\n");
3562 pr_crit("File: %s PID: %d Comm: %.20s\n", path, current->pid,
3568 generic_file_direct_write(struct kiocb *iocb, struct iov_iter *from)
3570 struct file *file = iocb->ki_filp;
3571 struct address_space *mapping = file->f_mapping;
3572 struct inode *inode = mapping->host;
3573 loff_t pos = iocb->ki_pos;
3578 write_len = iov_iter_count(from);
3579 end = (pos + write_len - 1) >> PAGE_SHIFT;
3581 if (iocb->ki_flags & IOCB_NOWAIT) {
3582 /* If there are pages to writeback, return */
3583 if (filemap_range_has_page(file->f_mapping, pos,
3584 pos + write_len - 1))
3587 written = filemap_write_and_wait_range(mapping, pos,
3588 pos + write_len - 1);
3594 * After a write we want buffered reads to be sure to go to disk to get
3595 * the new data. We invalidate clean cached page from the region we're
3596 * about to write. We do this *before* the write so that we can return
3597 * without clobbering -EIOCBQUEUED from ->direct_IO().
3599 written = invalidate_inode_pages2_range(mapping,
3600 pos >> PAGE_SHIFT, end);
3602 * If a page can not be invalidated, return 0 to fall back
3603 * to buffered write.
3606 if (written == -EBUSY)
3611 written = mapping->a_ops->direct_IO(iocb, from);
3614 * Finally, try again to invalidate clean pages which might have been
3615 * cached by non-direct readahead, or faulted in by get_user_pages()
3616 * if the source of the write was an mmap'ed region of the file
3617 * we're writing. Either one is a pretty crazy thing to do,
3618 * so we don't support it 100%. If this invalidation
3619 * fails, tough, the write still worked...
3621 * Most of the time we do not need this since dio_complete() will do
3622 * the invalidation for us. However there are some file systems that
3623 * do not end up with dio_complete() being called, so let's not break
3624 * them by removing it completely.
3626 * Noticeable example is a blkdev_direct_IO().
3628 * Skip invalidation for async writes or if mapping has no pages.
3630 if (written > 0 && mapping->nrpages &&
3631 invalidate_inode_pages2_range(mapping, pos >> PAGE_SHIFT, end))
3632 dio_warn_stale_pagecache(file);
3636 write_len -= written;
3637 if (pos > i_size_read(inode) && !S_ISBLK(inode->i_mode)) {
3638 i_size_write(inode, pos);
3639 mark_inode_dirty(inode);
3643 if (written != -EIOCBQUEUED)
3644 iov_iter_revert(from, write_len - iov_iter_count(from));
3648 EXPORT_SYMBOL(generic_file_direct_write);
3650 ssize_t generic_perform_write(struct file *file,
3651 struct iov_iter *i, loff_t pos)
3653 struct address_space *mapping = file->f_mapping;
3654 const struct address_space_operations *a_ops = mapping->a_ops;
3656 ssize_t written = 0;
3657 unsigned int flags = 0;
3661 unsigned long offset; /* Offset into pagecache page */
3662 unsigned long bytes; /* Bytes to write to page */
3663 size_t copied; /* Bytes copied from user */
3666 offset = (pos & (PAGE_SIZE - 1));
3667 bytes = min_t(unsigned long, PAGE_SIZE - offset,
3672 * Bring in the user page that we will copy from _first_.
3673 * Otherwise there's a nasty deadlock on copying from the
3674 * same page as we're writing to, without it being marked
3677 if (unlikely(fault_in_iov_iter_readable(i, bytes))) {
3682 if (fatal_signal_pending(current)) {
3687 status = a_ops->write_begin(file, mapping, pos, bytes, flags,
3689 if (unlikely(status < 0))
3692 if (mapping_writably_mapped(mapping))
3693 flush_dcache_page(page);
3695 copied = copy_page_from_iter_atomic(page, offset, bytes, i);
3696 flush_dcache_page(page);
3698 status = a_ops->write_end(file, mapping, pos, bytes, copied,
3700 if (unlikely(status != copied)) {
3701 iov_iter_revert(i, copied - max(status, 0L));
3702 if (unlikely(status < 0))
3707 if (unlikely(status == 0)) {
3709 * A short copy made ->write_end() reject the
3710 * thing entirely. Might be memory poisoning
3711 * halfway through, might be a race with munmap,
3712 * might be severe memory pressure.
3721 balance_dirty_pages_ratelimited(mapping);
3722 } while (iov_iter_count(i));
3724 return written ? written : status;
3726 EXPORT_SYMBOL(generic_perform_write);
3729 * __generic_file_write_iter - write data to a file
3730 * @iocb: IO state structure (file, offset, etc.)
3731 * @from: iov_iter with data to write
3733 * This function does all the work needed for actually writing data to a
3734 * file. It does all basic checks, removes SUID from the file, updates
3735 * modification times and calls proper subroutines depending on whether we
3736 * do direct IO or a standard buffered write.
3738 * It expects i_rwsem to be grabbed unless we work on a block device or similar
3739 * object which does not need locking at all.
3741 * This function does *not* take care of syncing data in case of O_SYNC write.
3742 * A caller has to handle it. This is mainly due to the fact that we want to
3743 * avoid syncing under i_rwsem.
3746 * * number of bytes written, even for truncated writes
3747 * * negative error code if no data has been written at all
3749 ssize_t __generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
3751 struct file *file = iocb->ki_filp;
3752 struct address_space *mapping = file->f_mapping;
3753 struct inode *inode = mapping->host;
3754 ssize_t written = 0;
3758 /* We can write back this queue in page reclaim */
3759 current->backing_dev_info = inode_to_bdi(inode);
3760 err = file_remove_privs(file);
3764 err = file_update_time(file);
3768 if (iocb->ki_flags & IOCB_DIRECT) {
3769 loff_t pos, endbyte;
3771 written = generic_file_direct_write(iocb, from);
3773 * If the write stopped short of completing, fall back to
3774 * buffered writes. Some filesystems do this for writes to
3775 * holes, for example. For DAX files, a buffered write will
3776 * not succeed (even if it did, DAX does not handle dirty
3777 * page-cache pages correctly).
3779 if (written < 0 || !iov_iter_count(from) || IS_DAX(inode))
3782 status = generic_perform_write(file, from, pos = iocb->ki_pos);
3784 * If generic_perform_write() returned a synchronous error
3785 * then we want to return the number of bytes which were
3786 * direct-written, or the error code if that was zero. Note
3787 * that this differs from normal direct-io semantics, which
3788 * will return -EFOO even if some bytes were written.
3790 if (unlikely(status < 0)) {
3795 * We need to ensure that the page cache pages are written to
3796 * disk and invalidated to preserve the expected O_DIRECT
3799 endbyte = pos + status - 1;
3800 err = filemap_write_and_wait_range(mapping, pos, endbyte);
3802 iocb->ki_pos = endbyte + 1;
3804 invalidate_mapping_pages(mapping,
3806 endbyte >> PAGE_SHIFT);
3809 * We don't know how much we wrote, so just return
3810 * the number of bytes which were direct-written
3814 written = generic_perform_write(file, from, iocb->ki_pos);
3815 if (likely(written > 0))
3816 iocb->ki_pos += written;
3819 current->backing_dev_info = NULL;
3820 return written ? written : err;
3822 EXPORT_SYMBOL(__generic_file_write_iter);
3825 * generic_file_write_iter - write data to a file
3826 * @iocb: IO state structure
3827 * @from: iov_iter with data to write
3829 * This is a wrapper around __generic_file_write_iter() to be used by most
3830 * filesystems. It takes care of syncing the file in case of O_SYNC file
3831 * and acquires i_rwsem as needed.
3833 * * negative error code if no data has been written at all of
3834 * vfs_fsync_range() failed for a synchronous write
3835 * * number of bytes written, even for truncated writes
3837 ssize_t generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
3839 struct file *file = iocb->ki_filp;
3840 struct inode *inode = file->f_mapping->host;
3844 ret = generic_write_checks(iocb, from);
3846 ret = __generic_file_write_iter(iocb, from);
3847 inode_unlock(inode);
3850 ret = generic_write_sync(iocb, ret);
3853 EXPORT_SYMBOL(generic_file_write_iter);
3856 * filemap_release_folio() - Release fs-specific metadata on a folio.
3857 * @folio: The folio which the kernel is trying to free.
3858 * @gfp: Memory allocation flags (and I/O mode).
3860 * The address_space is trying to release any data attached to a folio
3861 * (presumably at folio->private).
3863 * This will also be called if the private_2 flag is set on a page,
3864 * indicating that the folio has other metadata associated with it.
3866 * The @gfp argument specifies whether I/O may be performed to release
3867 * this page (__GFP_IO), and whether the call may block
3868 * (__GFP_RECLAIM & __GFP_FS).
3870 * Return: %true if the release was successful, otherwise %false.
3872 bool filemap_release_folio(struct folio *folio, gfp_t gfp)
3874 struct address_space * const mapping = folio->mapping;
3876 BUG_ON(!folio_test_locked(folio));
3877 if (folio_test_writeback(folio))
3880 if (mapping && mapping->a_ops->releasepage)
3881 return mapping->a_ops->releasepage(&folio->page, gfp);
3882 return try_to_free_buffers(&folio->page);
3884 EXPORT_SYMBOL(filemap_release_folio);