1 // SPDX-License-Identifier: GPL-2.0-only
5 * Copyright (C) 1994-1999 Linus Torvalds
9 * This file handles the generic file mmap semantics used by
10 * most "normal" filesystems (but you don't /have/ to use this:
11 * the NFS filesystem used to do this differently, for example)
13 #include <linux/export.h>
14 #include <linux/compiler.h>
15 #include <linux/dax.h>
17 #include <linux/sched/signal.h>
18 #include <linux/uaccess.h>
19 #include <linux/capability.h>
20 #include <linux/kernel_stat.h>
21 #include <linux/gfp.h>
23 #include <linux/swap.h>
24 #include <linux/swapops.h>
25 #include <linux/syscalls.h>
26 #include <linux/mman.h>
27 #include <linux/pagemap.h>
28 #include <linux/file.h>
29 #include <linux/uio.h>
30 #include <linux/error-injection.h>
31 #include <linux/hash.h>
32 #include <linux/writeback.h>
33 #include <linux/backing-dev.h>
34 #include <linux/pagevec.h>
35 #include <linux/security.h>
36 #include <linux/cpuset.h>
37 #include <linux/hugetlb.h>
38 #include <linux/memcontrol.h>
39 #include <linux/shmem_fs.h>
40 #include <linux/rmap.h>
41 #include <linux/delayacct.h>
42 #include <linux/psi.h>
43 #include <linux/ramfs.h>
44 #include <linux/page_idle.h>
45 #include <linux/migrate.h>
46 #include <linux/pipe_fs_i.h>
47 #include <linux/splice.h>
48 #include <linux/rcupdate_wait.h>
49 #include <asm/pgalloc.h>
50 #include <asm/tlbflush.h>
53 #define CREATE_TRACE_POINTS
54 #include <trace/events/filemap.h>
57 * FIXME: remove all knowledge of the buffer layer from the core VM
59 #include <linux/buffer_head.h> /* for try_to_free_buffers */
66 * Shared mappings implemented 30.11.1994. It's not fully working yet,
69 * Shared mappings now work. 15.8.1995 Bruno.
71 * finished 'unifying' the page and buffer cache and SMP-threaded the
72 * page-cache, 21.05.1999, Ingo Molnar <mingo@redhat.com>
74 * SMP-threaded pagemap-LRU 1999, Andrea Arcangeli <andrea@suse.de>
80 * ->i_mmap_rwsem (truncate_pagecache)
81 * ->private_lock (__free_pte->block_dirty_folio)
82 * ->swap_lock (exclusive_swap_page, others)
86 * ->invalidate_lock (acquired by fs in truncate path)
87 * ->i_mmap_rwsem (truncate->unmap_mapping_range)
91 * ->page_table_lock or pte_lock (various, mainly in memory.c)
92 * ->i_pages lock (arch-dependent flush_dcache_mmap_lock)
95 * ->invalidate_lock (filemap_fault)
96 * ->lock_page (filemap_fault, access_process_vm)
98 * ->i_rwsem (generic_perform_write)
99 * ->mmap_lock (fault_in_readable->do_page_fault)
102 * sb_lock (fs/fs-writeback.c)
103 * ->i_pages lock (__sync_single_inode)
106 * ->anon_vma.lock (vma_merge)
109 * ->page_table_lock or pte_lock (anon_vma_prepare and various)
111 * ->page_table_lock or pte_lock
112 * ->swap_lock (try_to_unmap_one)
113 * ->private_lock (try_to_unmap_one)
114 * ->i_pages lock (try_to_unmap_one)
115 * ->lruvec->lru_lock (follow_page->mark_page_accessed)
116 * ->lruvec->lru_lock (check_pte_range->isolate_lru_page)
117 * ->private_lock (folio_remove_rmap_pte->set_page_dirty)
118 * ->i_pages lock (folio_remove_rmap_pte->set_page_dirty)
119 * bdi.wb->list_lock (folio_remove_rmap_pte->set_page_dirty)
120 * ->inode->i_lock (folio_remove_rmap_pte->set_page_dirty)
121 * ->memcg->move_lock (folio_remove_rmap_pte->folio_memcg_lock)
122 * bdi.wb->list_lock (zap_pte_range->set_page_dirty)
123 * ->inode->i_lock (zap_pte_range->set_page_dirty)
124 * ->private_lock (zap_pte_range->block_dirty_folio)
127 static void mapping_set_update(struct xa_state *xas,
128 struct address_space *mapping)
130 if (dax_mapping(mapping) || shmem_mapping(mapping))
132 xas_set_update(xas, workingset_update_node);
133 xas_set_lru(xas, &shadow_nodes);
136 static void page_cache_delete(struct address_space *mapping,
137 struct folio *folio, void *shadow)
139 XA_STATE(xas, &mapping->i_pages, folio->index);
142 mapping_set_update(&xas, mapping);
144 xas_set_order(&xas, folio->index, folio_order(folio));
145 nr = folio_nr_pages(folio);
147 VM_BUG_ON_FOLIO(!folio_test_locked(folio), folio);
149 xas_store(&xas, shadow);
150 xas_init_marks(&xas);
152 folio->mapping = NULL;
153 /* Leave page->index set: truncation lookup relies upon it */
154 mapping->nrpages -= nr;
157 static void filemap_unaccount_folio(struct address_space *mapping,
162 VM_BUG_ON_FOLIO(folio_mapped(folio), folio);
163 if (!IS_ENABLED(CONFIG_DEBUG_VM) && unlikely(folio_mapped(folio))) {
164 pr_alert("BUG: Bad page cache in process %s pfn:%05lx\n",
165 current->comm, folio_pfn(folio));
166 dump_page(&folio->page, "still mapped when deleted");
168 add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE);
170 if (mapping_exiting(mapping) && !folio_test_large(folio)) {
171 int mapcount = folio_mapcount(folio);
173 if (folio_ref_count(folio) >= mapcount + 2) {
175 * All vmas have already been torn down, so it's
176 * a good bet that actually the page is unmapped
177 * and we'd rather not leak it: if we're wrong,
178 * another bad page check should catch it later.
180 atomic_set(&folio->_mapcount, -1);
181 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
205 * unwritten data - on ordinary filesystems.
207 * But it's harmless on in-memory filesystems like tmpfs; and can
208 * occur when a driver which did get_user_pages() sets page dirty
209 * before putting it, while the inode is being finally evicted.
211 * Below fixes dirty accounting after removing the folio entirely
212 * but leaves the dirty flag set: it has no effect for truncated
213 * folio and anyway will be cleared before returning folio to
216 if (WARN_ON_ONCE(folio_test_dirty(folio) &&
217 mapping_can_writeback(mapping)))
218 folio_account_cleaned(folio, inode_to_wb(mapping->host));
222 * Delete a page from the page cache and free it. Caller has to make
223 * sure the page is locked and that nobody else uses it - or that usage
224 * is safe. The caller must hold the i_pages lock.
226 void __filemap_remove_folio(struct folio *folio, void *shadow)
228 struct address_space *mapping = folio->mapping;
230 trace_mm_filemap_delete_from_page_cache(folio);
231 filemap_unaccount_folio(mapping, folio);
232 page_cache_delete(mapping, folio, shadow);
235 void filemap_free_folio(struct address_space *mapping, struct folio *folio)
237 void (*free_folio)(struct folio *);
240 free_folio = mapping->a_ops->free_folio;
244 if (folio_test_large(folio))
245 refs = folio_nr_pages(folio);
246 folio_put_refs(folio, refs);
250 * filemap_remove_folio - Remove folio from page cache.
253 * This must be called only on folios that are locked and have been
254 * verified to be in the page cache. It will never put the folio into
255 * the free list because the caller has a reference on the page.
257 void filemap_remove_folio(struct folio *folio)
259 struct address_space *mapping = folio->mapping;
261 BUG_ON(!folio_test_locked(folio));
262 spin_lock(&mapping->host->i_lock);
263 xa_lock_irq(&mapping->i_pages);
264 __filemap_remove_folio(folio, NULL);
265 xa_unlock_irq(&mapping->i_pages);
266 if (mapping_shrinkable(mapping))
267 inode_add_lru(mapping->host);
268 spin_unlock(&mapping->host->i_lock);
270 filemap_free_folio(mapping, folio);
274 * page_cache_delete_batch - delete several folios from page cache
275 * @mapping: the mapping to which folios belong
276 * @fbatch: batch of folios to delete
278 * The function walks over mapping->i_pages and removes folios passed in
279 * @fbatch from the mapping. The function expects @fbatch to be sorted
280 * by page index and is optimised for it to be dense.
281 * It tolerates holes in @fbatch (mapping entries at those indices are not
284 * The function expects the i_pages lock to be held.
286 static void page_cache_delete_batch(struct address_space *mapping,
287 struct folio_batch *fbatch)
289 XA_STATE(xas, &mapping->i_pages, fbatch->folios[0]->index);
290 long total_pages = 0;
294 mapping_set_update(&xas, mapping);
295 xas_for_each(&xas, folio, ULONG_MAX) {
296 if (i >= folio_batch_count(fbatch))
299 /* A swap/dax/shadow entry got inserted? Skip it. */
300 if (xa_is_value(folio))
303 * A page got inserted in our range? Skip it. We have our
304 * pages locked so they are protected from being removed.
305 * If we see a page whose index is higher than ours, it
306 * means our page has been removed, which shouldn't be
307 * possible because we're holding the PageLock.
309 if (folio != fbatch->folios[i]) {
310 VM_BUG_ON_FOLIO(folio->index >
311 fbatch->folios[i]->index, folio);
315 WARN_ON_ONCE(!folio_test_locked(folio));
317 folio->mapping = NULL;
318 /* Leave folio->index set: truncation lookup relies on it */
321 xas_store(&xas, NULL);
322 total_pages += folio_nr_pages(folio);
324 mapping->nrpages -= total_pages;
327 void delete_from_page_cache_batch(struct address_space *mapping,
328 struct folio_batch *fbatch)
332 if (!folio_batch_count(fbatch))
335 spin_lock(&mapping->host->i_lock);
336 xa_lock_irq(&mapping->i_pages);
337 for (i = 0; i < folio_batch_count(fbatch); i++) {
338 struct folio *folio = fbatch->folios[i];
340 trace_mm_filemap_delete_from_page_cache(folio);
341 filemap_unaccount_folio(mapping, folio);
343 page_cache_delete_batch(mapping, fbatch);
344 xa_unlock_irq(&mapping->i_pages);
345 if (mapping_shrinkable(mapping))
346 inode_add_lru(mapping->host);
347 spin_unlock(&mapping->host->i_lock);
349 for (i = 0; i < folio_batch_count(fbatch); i++)
350 filemap_free_folio(mapping, fbatch->folios[i]);
353 int filemap_check_errors(struct address_space *mapping)
356 /* Check for outstanding write errors */
357 if (test_bit(AS_ENOSPC, &mapping->flags) &&
358 test_and_clear_bit(AS_ENOSPC, &mapping->flags))
360 if (test_bit(AS_EIO, &mapping->flags) &&
361 test_and_clear_bit(AS_EIO, &mapping->flags))
365 EXPORT_SYMBOL(filemap_check_errors);
367 static int filemap_check_and_keep_errors(struct address_space *mapping)
369 /* Check for outstanding write errors */
370 if (test_bit(AS_EIO, &mapping->flags))
372 if (test_bit(AS_ENOSPC, &mapping->flags))
378 * filemap_fdatawrite_wbc - start writeback on mapping dirty pages in range
379 * @mapping: address space structure to write
380 * @wbc: the writeback_control controlling the writeout
382 * Call writepages on the mapping using the provided wbc to control the
385 * Return: %0 on success, negative error code otherwise.
387 int filemap_fdatawrite_wbc(struct address_space *mapping,
388 struct writeback_control *wbc)
392 if (!mapping_can_writeback(mapping) ||
393 !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
396 wbc_attach_fdatawrite_inode(wbc, mapping->host);
397 ret = do_writepages(mapping, wbc);
398 wbc_detach_inode(wbc);
401 EXPORT_SYMBOL(filemap_fdatawrite_wbc);
404 * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range
405 * @mapping: address space structure to write
406 * @start: offset in bytes where the range starts
407 * @end: offset in bytes where the range ends (inclusive)
408 * @sync_mode: enable synchronous operation
410 * Start writeback against all of a mapping's dirty pages that lie
411 * within the byte offsets <start, end> inclusive.
413 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
414 * opposed to a regular memory cleansing writeback. The difference between
415 * these two operations is that if a dirty page/buffer is encountered, it must
416 * be waited upon, and not just skipped over.
418 * Return: %0 on success, negative error code otherwise.
420 int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
421 loff_t end, int sync_mode)
423 struct writeback_control wbc = {
424 .sync_mode = sync_mode,
425 .nr_to_write = LONG_MAX,
426 .range_start = start,
430 return filemap_fdatawrite_wbc(mapping, &wbc);
433 static inline int __filemap_fdatawrite(struct address_space *mapping,
436 return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode);
439 int filemap_fdatawrite(struct address_space *mapping)
441 return __filemap_fdatawrite(mapping, WB_SYNC_ALL);
443 EXPORT_SYMBOL(filemap_fdatawrite);
445 int filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
448 return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL);
450 EXPORT_SYMBOL(filemap_fdatawrite_range);
453 * filemap_flush - mostly a non-blocking flush
454 * @mapping: target address_space
456 * This is a mostly non-blocking flush. Not suitable for data-integrity
457 * purposes - I/O may not be started against all dirty pages.
459 * Return: %0 on success, negative error code otherwise.
461 int filemap_flush(struct address_space *mapping)
463 return __filemap_fdatawrite(mapping, WB_SYNC_NONE);
465 EXPORT_SYMBOL(filemap_flush);
468 * filemap_range_has_page - check if a page exists in range.
469 * @mapping: address space within which to check
470 * @start_byte: offset in bytes where the range starts
471 * @end_byte: offset in bytes where the range ends (inclusive)
473 * Find at least one page in the range supplied, usually used to check if
474 * direct writing in this range will trigger a writeback.
476 * Return: %true if at least one page exists in the specified range,
479 bool filemap_range_has_page(struct address_space *mapping,
480 loff_t start_byte, loff_t end_byte)
483 XA_STATE(xas, &mapping->i_pages, start_byte >> PAGE_SHIFT);
484 pgoff_t max = end_byte >> PAGE_SHIFT;
486 if (end_byte < start_byte)
491 folio = xas_find(&xas, max);
492 if (xas_retry(&xas, folio))
494 /* Shadow entries don't count */
495 if (xa_is_value(folio))
498 * We don't need to try to pin this page; we're about to
499 * release the RCU lock anyway. It is enough to know that
500 * there was a page here recently.
506 return folio != NULL;
508 EXPORT_SYMBOL(filemap_range_has_page);
510 static void __filemap_fdatawait_range(struct address_space *mapping,
511 loff_t start_byte, loff_t end_byte)
513 pgoff_t index = start_byte >> PAGE_SHIFT;
514 pgoff_t end = end_byte >> PAGE_SHIFT;
515 struct folio_batch fbatch;
518 folio_batch_init(&fbatch);
520 while (index <= end) {
523 nr_folios = filemap_get_folios_tag(mapping, &index, end,
524 PAGECACHE_TAG_WRITEBACK, &fbatch);
529 for (i = 0; i < nr_folios; i++) {
530 struct folio *folio = fbatch.folios[i];
532 folio_wait_writeback(folio);
533 folio_clear_error(folio);
535 folio_batch_release(&fbatch);
541 * filemap_fdatawait_range - wait for writeback to complete
542 * @mapping: address space structure to wait for
543 * @start_byte: offset in bytes where the range starts
544 * @end_byte: offset in bytes where the range ends (inclusive)
546 * Walk the list of under-writeback pages of the given address space
547 * in the given range and wait for all of them. Check error status of
548 * the address space and return it.
550 * Since the error status of the address space is cleared by this function,
551 * callers are responsible for checking the return value and handling and/or
552 * reporting the error.
554 * Return: error status of the address space.
556 int filemap_fdatawait_range(struct address_space *mapping, loff_t start_byte,
559 __filemap_fdatawait_range(mapping, start_byte, end_byte);
560 return filemap_check_errors(mapping);
562 EXPORT_SYMBOL(filemap_fdatawait_range);
565 * filemap_fdatawait_range_keep_errors - wait for writeback to complete
566 * @mapping: address space structure to wait for
567 * @start_byte: offset in bytes where the range starts
568 * @end_byte: offset in bytes where the range ends (inclusive)
570 * Walk the list of under-writeback pages of the given address space in the
571 * given range and wait for all of them. Unlike filemap_fdatawait_range(),
572 * this function does not clear error status of the address space.
574 * Use this function if callers don't handle errors themselves. Expected
575 * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2),
578 int filemap_fdatawait_range_keep_errors(struct address_space *mapping,
579 loff_t start_byte, loff_t end_byte)
581 __filemap_fdatawait_range(mapping, start_byte, end_byte);
582 return filemap_check_and_keep_errors(mapping);
584 EXPORT_SYMBOL(filemap_fdatawait_range_keep_errors);
587 * file_fdatawait_range - wait for writeback to complete
588 * @file: file pointing to address space structure to wait for
589 * @start_byte: offset in bytes where the range starts
590 * @end_byte: offset in bytes where the range ends (inclusive)
592 * Walk the list of under-writeback pages of the address space that file
593 * refers to, in the given range and wait for all of them. Check error
594 * status of the address space vs. the file->f_wb_err cursor and return it.
596 * Since the error status of the file is advanced by this function,
597 * callers are responsible for checking the return value and handling and/or
598 * reporting the error.
600 * Return: error status of the address space vs. the file->f_wb_err cursor.
602 int file_fdatawait_range(struct file *file, loff_t start_byte, loff_t end_byte)
604 struct address_space *mapping = file->f_mapping;
606 __filemap_fdatawait_range(mapping, start_byte, end_byte);
607 return file_check_and_advance_wb_err(file);
609 EXPORT_SYMBOL(file_fdatawait_range);
612 * filemap_fdatawait_keep_errors - wait for writeback without clearing errors
613 * @mapping: address space structure to wait for
615 * Walk the list of under-writeback pages of the given address space
616 * and wait for all of them. Unlike filemap_fdatawait(), this function
617 * does not clear error status of the address space.
619 * Use this function if callers don't handle errors themselves. Expected
620 * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2),
623 * Return: error status of the address space.
625 int filemap_fdatawait_keep_errors(struct address_space *mapping)
627 __filemap_fdatawait_range(mapping, 0, LLONG_MAX);
628 return filemap_check_and_keep_errors(mapping);
630 EXPORT_SYMBOL(filemap_fdatawait_keep_errors);
632 /* Returns true if writeback might be needed or already in progress. */
633 static bool mapping_needs_writeback(struct address_space *mapping)
635 return mapping->nrpages;
638 bool filemap_range_has_writeback(struct address_space *mapping,
639 loff_t start_byte, loff_t end_byte)
641 XA_STATE(xas, &mapping->i_pages, start_byte >> PAGE_SHIFT);
642 pgoff_t max = end_byte >> PAGE_SHIFT;
645 if (end_byte < start_byte)
649 xas_for_each(&xas, folio, max) {
650 if (xas_retry(&xas, folio))
652 if (xa_is_value(folio))
654 if (folio_test_dirty(folio) || folio_test_locked(folio) ||
655 folio_test_writeback(folio))
659 return folio != NULL;
661 EXPORT_SYMBOL_GPL(filemap_range_has_writeback);
664 * filemap_write_and_wait_range - write out & wait on a file range
665 * @mapping: the address_space for the pages
666 * @lstart: offset in bytes where the range starts
667 * @lend: offset in bytes where the range ends (inclusive)
669 * Write out and wait upon file offsets lstart->lend, inclusive.
671 * Note that @lend is inclusive (describes the last byte to be written) so
672 * that this function can be used to write to the very end-of-file (end = -1).
674 * Return: error status of the address space.
676 int filemap_write_and_wait_range(struct address_space *mapping,
677 loff_t lstart, loff_t lend)
684 if (mapping_needs_writeback(mapping)) {
685 err = __filemap_fdatawrite_range(mapping, lstart, lend,
688 * Even if the above returned error, the pages may be
689 * written partially (e.g. -ENOSPC), so we wait for it.
690 * But the -EIO is special case, it may indicate the worst
691 * thing (e.g. bug) happened, so we avoid waiting for it.
694 __filemap_fdatawait_range(mapping, lstart, lend);
696 err2 = filemap_check_errors(mapping);
701 EXPORT_SYMBOL(filemap_write_and_wait_range);
703 void __filemap_set_wb_err(struct address_space *mapping, int err)
705 errseq_t eseq = errseq_set(&mapping->wb_err, err);
707 trace_filemap_set_wb_err(mapping, eseq);
709 EXPORT_SYMBOL(__filemap_set_wb_err);
712 * file_check_and_advance_wb_err - report wb error (if any) that was previously
713 * and advance wb_err to current one
714 * @file: struct file on which the error is being reported
716 * When userland calls fsync (or something like nfsd does the equivalent), we
717 * want to report any writeback errors that occurred since the last fsync (or
718 * since the file was opened if there haven't been any).
720 * Grab the wb_err from the mapping. If it matches what we have in the file,
721 * then just quickly return 0. The file is all caught up.
723 * If it doesn't match, then take the mapping value, set the "seen" flag in
724 * it and try to swap it into place. If it works, or another task beat us
725 * to it with the new value, then update the f_wb_err and return the error
726 * portion. The error at this point must be reported via proper channels
727 * (a'la fsync, or NFS COMMIT operation, etc.).
729 * While we handle mapping->wb_err with atomic operations, the f_wb_err
730 * value is protected by the f_lock since we must ensure that it reflects
731 * the latest value swapped in for this file descriptor.
733 * Return: %0 on success, negative error code otherwise.
735 int file_check_and_advance_wb_err(struct file *file)
738 errseq_t old = READ_ONCE(file->f_wb_err);
739 struct address_space *mapping = file->f_mapping;
741 /* Locklessly handle the common case where nothing has changed */
742 if (errseq_check(&mapping->wb_err, old)) {
743 /* Something changed, must use slow path */
744 spin_lock(&file->f_lock);
745 old = file->f_wb_err;
746 err = errseq_check_and_advance(&mapping->wb_err,
748 trace_file_check_and_advance_wb_err(file, old);
749 spin_unlock(&file->f_lock);
753 * We're mostly using this function as a drop in replacement for
754 * filemap_check_errors. Clear AS_EIO/AS_ENOSPC to emulate the effect
755 * that the legacy code would have had on these flags.
757 clear_bit(AS_EIO, &mapping->flags);
758 clear_bit(AS_ENOSPC, &mapping->flags);
761 EXPORT_SYMBOL(file_check_and_advance_wb_err);
764 * file_write_and_wait_range - write out & wait on a file range
765 * @file: file pointing to address_space with pages
766 * @lstart: offset in bytes where the range starts
767 * @lend: offset in bytes where the range ends (inclusive)
769 * Write out and wait upon file offsets lstart->lend, inclusive.
771 * Note that @lend is inclusive (describes the last byte to be written) so
772 * that this function can be used to write to the very end-of-file (end = -1).
774 * After writing out and waiting on the data, we check and advance the
775 * f_wb_err cursor to the latest value, and return any errors detected there.
777 * Return: %0 on success, negative error code otherwise.
779 int file_write_and_wait_range(struct file *file, loff_t lstart, loff_t lend)
782 struct address_space *mapping = file->f_mapping;
787 if (mapping_needs_writeback(mapping)) {
788 err = __filemap_fdatawrite_range(mapping, lstart, lend,
790 /* See comment of filemap_write_and_wait() */
792 __filemap_fdatawait_range(mapping, lstart, lend);
794 err2 = file_check_and_advance_wb_err(file);
799 EXPORT_SYMBOL(file_write_and_wait_range);
802 * replace_page_cache_folio - replace a pagecache folio with a new one
803 * @old: folio to be replaced
804 * @new: folio to replace with
806 * This function replaces a folio in the pagecache with a new one. On
807 * success it acquires the pagecache reference for the new folio and
808 * drops it for the old folio. Both the old and new folios must be
809 * locked. This function does not add the new folio to the LRU, the
810 * caller must do that.
812 * The remove + add is atomic. This function cannot fail.
814 void replace_page_cache_folio(struct folio *old, struct folio *new)
816 struct address_space *mapping = old->mapping;
817 void (*free_folio)(struct folio *) = mapping->a_ops->free_folio;
818 pgoff_t offset = old->index;
819 XA_STATE(xas, &mapping->i_pages, offset);
821 VM_BUG_ON_FOLIO(!folio_test_locked(old), old);
822 VM_BUG_ON_FOLIO(!folio_test_locked(new), new);
823 VM_BUG_ON_FOLIO(new->mapping, new);
826 new->mapping = mapping;
829 mem_cgroup_replace_folio(old, new);
832 xas_store(&xas, new);
835 /* hugetlb pages do not participate in page cache accounting. */
836 if (!folio_test_hugetlb(old))
837 __lruvec_stat_sub_folio(old, NR_FILE_PAGES);
838 if (!folio_test_hugetlb(new))
839 __lruvec_stat_add_folio(new, NR_FILE_PAGES);
840 if (folio_test_swapbacked(old))
841 __lruvec_stat_sub_folio(old, NR_SHMEM);
842 if (folio_test_swapbacked(new))
843 __lruvec_stat_add_folio(new, NR_SHMEM);
844 xas_unlock_irq(&xas);
849 EXPORT_SYMBOL_GPL(replace_page_cache_folio);
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 void *alloced_shadow = NULL;
856 int alloced_order = 0;
860 VM_BUG_ON_FOLIO(!folio_test_locked(folio), folio);
861 VM_BUG_ON_FOLIO(folio_test_swapbacked(folio), folio);
862 VM_BUG_ON_FOLIO(folio_order(folio) < mapping_min_folio_order(mapping),
864 mapping_set_update(&xas, mapping);
866 VM_BUG_ON_FOLIO(index & (folio_nr_pages(folio) - 1), folio);
867 xas_set_order(&xas, index, folio_order(folio));
868 huge = folio_test_hugetlb(folio);
869 nr = folio_nr_pages(folio);
871 gfp &= GFP_RECLAIM_MASK;
872 folio_ref_add(folio, nr);
873 folio->mapping = mapping;
874 folio->index = xas.xa_index;
877 int order = -1, split_order = 0;
878 void *entry, *old = NULL;
881 xas_for_each_conflict(&xas, entry) {
883 if (!xa_is_value(entry)) {
884 xas_set_err(&xas, -EEXIST);
888 * If a larger entry exists,
889 * it will be the first and only entry iterated.
892 order = xas_get_order(&xas);
895 /* entry may have changed before we re-acquire the lock */
896 if (alloced_order && (old != alloced_shadow || order != alloced_order)) {
902 if (order > 0 && order > folio_order(folio)) {
903 /* How to handle large swap entries? */
904 BUG_ON(shmem_mapping(mapping));
905 if (!alloced_order) {
909 xas_split(&xas, old, order);
916 xas_store(&xas, folio);
920 mapping->nrpages += nr;
922 /* hugetlb pages do not participate in page cache accounting */
924 __lruvec_stat_mod_folio(folio, NR_FILE_PAGES, nr);
925 if (folio_test_pmd_mappable(folio))
926 __lruvec_stat_mod_folio(folio,
931 xas_unlock_irq(&xas);
933 /* split needed, alloc here and retry. */
935 xas_split_alloc(&xas, old, split_order, gfp);
938 alloced_shadow = old;
939 alloced_order = split_order;
944 if (!xas_nomem(&xas, gfp))
951 trace_mm_filemap_add_to_page_cache(folio);
954 folio->mapping = NULL;
955 /* Leave page->index set: truncation relies upon it */
956 folio_put_refs(folio, nr);
957 return xas_error(&xas);
959 ALLOW_ERROR_INJECTION(__filemap_add_folio, ERRNO);
961 int filemap_add_folio(struct address_space *mapping, struct folio *folio,
962 pgoff_t index, gfp_t gfp)
967 ret = mem_cgroup_charge(folio, NULL, gfp);
971 __folio_set_locked(folio);
972 ret = __filemap_add_folio(mapping, folio, index, gfp, &shadow);
974 mem_cgroup_uncharge(folio);
975 __folio_clear_locked(folio);
978 * The folio might have been evicted from cache only
979 * recently, in which case it should be activated like
980 * any other repeatedly accessed folio.
981 * The exception is folios getting rewritten; evicting other
982 * data from the working set, only to cache data that will
983 * get overwritten with something else, is a waste of memory.
985 WARN_ON_ONCE(folio_test_active(folio));
986 if (!(gfp & __GFP_WRITE) && shadow)
987 workingset_refault(folio, shadow);
988 folio_add_lru(folio);
992 EXPORT_SYMBOL_GPL(filemap_add_folio);
995 struct folio *filemap_alloc_folio_noprof(gfp_t gfp, unsigned int order)
1000 if (cpuset_do_page_mem_spread()) {
1001 unsigned int cpuset_mems_cookie;
1003 cpuset_mems_cookie = read_mems_allowed_begin();
1004 n = cpuset_mem_spread_node();
1005 folio = __folio_alloc_node_noprof(gfp, order, n);
1006 } while (!folio && read_mems_allowed_retry(cpuset_mems_cookie));
1010 return folio_alloc_noprof(gfp, order);
1012 EXPORT_SYMBOL(filemap_alloc_folio_noprof);
1016 * filemap_invalidate_lock_two - lock invalidate_lock for two mappings
1018 * Lock exclusively invalidate_lock of any passed mapping that is not NULL.
1020 * @mapping1: the first mapping to lock
1021 * @mapping2: the second mapping to lock
1023 void filemap_invalidate_lock_two(struct address_space *mapping1,
1024 struct address_space *mapping2)
1026 if (mapping1 > mapping2)
1027 swap(mapping1, mapping2);
1029 down_write(&mapping1->invalidate_lock);
1030 if (mapping2 && mapping1 != mapping2)
1031 down_write_nested(&mapping2->invalidate_lock, 1);
1033 EXPORT_SYMBOL(filemap_invalidate_lock_two);
1036 * filemap_invalidate_unlock_two - unlock invalidate_lock for two mappings
1038 * Unlock exclusive invalidate_lock of any passed mapping that is not NULL.
1040 * @mapping1: the first mapping to unlock
1041 * @mapping2: the second mapping to unlock
1043 void filemap_invalidate_unlock_two(struct address_space *mapping1,
1044 struct address_space *mapping2)
1047 up_write(&mapping1->invalidate_lock);
1048 if (mapping2 && mapping1 != mapping2)
1049 up_write(&mapping2->invalidate_lock);
1051 EXPORT_SYMBOL(filemap_invalidate_unlock_two);
1054 * In order to wait for pages to become available there must be
1055 * waitqueues associated with pages. By using a hash table of
1056 * waitqueues where the bucket discipline is to maintain all
1057 * waiters on the same queue and wake all when any of the pages
1058 * become available, and for the woken contexts to check to be
1059 * sure the appropriate page became available, this saves space
1060 * at a cost of "thundering herd" phenomena during rare hash
1063 #define PAGE_WAIT_TABLE_BITS 8
1064 #define PAGE_WAIT_TABLE_SIZE (1 << PAGE_WAIT_TABLE_BITS)
1065 static wait_queue_head_t folio_wait_table[PAGE_WAIT_TABLE_SIZE] __cacheline_aligned;
1067 static wait_queue_head_t *folio_waitqueue(struct folio *folio)
1069 return &folio_wait_table[hash_ptr(folio, PAGE_WAIT_TABLE_BITS)];
1072 void __init pagecache_init(void)
1076 for (i = 0; i < PAGE_WAIT_TABLE_SIZE; i++)
1077 init_waitqueue_head(&folio_wait_table[i]);
1079 page_writeback_init();
1083 * The page wait code treats the "wait->flags" somewhat unusually, because
1084 * we have multiple different kinds of waits, not just the usual "exclusive"
1089 * (a) no special bits set:
1091 * We're just waiting for the bit to be released, and when a waker
1092 * calls the wakeup function, we set WQ_FLAG_WOKEN and wake it up,
1093 * and remove it from the wait queue.
1095 * Simple and straightforward.
1097 * (b) WQ_FLAG_EXCLUSIVE:
1099 * The waiter is waiting to get the lock, and only one waiter should
1100 * be woken up to avoid any thundering herd behavior. We'll set the
1101 * WQ_FLAG_WOKEN bit, wake it up, and remove it from the wait queue.
1103 * This is the traditional exclusive wait.
1105 * (c) WQ_FLAG_EXCLUSIVE | WQ_FLAG_CUSTOM:
1107 * The waiter is waiting to get the bit, and additionally wants the
1108 * lock to be transferred to it for fair lock behavior. If the lock
1109 * cannot be taken, we stop walking the wait queue without waking
1112 * This is the "fair lock handoff" case, and in addition to setting
1113 * WQ_FLAG_WOKEN, we set WQ_FLAG_DONE to let the waiter easily see
1114 * that it now has the lock.
1116 static int wake_page_function(wait_queue_entry_t *wait, unsigned mode, int sync, void *arg)
1119 struct wait_page_key *key = arg;
1120 struct wait_page_queue *wait_page
1121 = container_of(wait, struct wait_page_queue, wait);
1123 if (!wake_page_match(wait_page, key))
1127 * If it's a lock handoff wait, we get the bit for it, and
1128 * stop walking (and do not wake it up) if we can't.
1130 flags = wait->flags;
1131 if (flags & WQ_FLAG_EXCLUSIVE) {
1132 if (test_bit(key->bit_nr, &key->folio->flags))
1134 if (flags & WQ_FLAG_CUSTOM) {
1135 if (test_and_set_bit(key->bit_nr, &key->folio->flags))
1137 flags |= WQ_FLAG_DONE;
1142 * We are holding the wait-queue lock, but the waiter that
1143 * is waiting for this will be checking the flags without
1146 * So update the flags atomically, and wake up the waiter
1147 * afterwards to avoid any races. This store-release pairs
1148 * with the load-acquire in folio_wait_bit_common().
1150 smp_store_release(&wait->flags, flags | WQ_FLAG_WOKEN);
1151 wake_up_state(wait->private, mode);
1154 * Ok, we have successfully done what we're waiting for,
1155 * and we can unconditionally remove the wait entry.
1157 * Note that this pairs with the "finish_wait()" in the
1158 * waiter, and has to be the absolute last thing we do.
1159 * After this list_del_init(&wait->entry) the wait entry
1160 * might be de-allocated and the process might even have
1163 list_del_init_careful(&wait->entry);
1164 return (flags & WQ_FLAG_EXCLUSIVE) != 0;
1167 static void folio_wake_bit(struct folio *folio, int bit_nr)
1169 wait_queue_head_t *q = folio_waitqueue(folio);
1170 struct wait_page_key key;
1171 unsigned long flags;
1174 key.bit_nr = bit_nr;
1177 spin_lock_irqsave(&q->lock, flags);
1178 __wake_up_locked_key(q, TASK_NORMAL, &key);
1181 * It's possible to miss clearing waiters here, when we woke our page
1182 * waiters, but the hashed waitqueue has waiters for other pages on it.
1183 * That's okay, it's a rare case. The next waker will clear it.
1185 * Note that, depending on the page pool (buddy, hugetlb, ZONE_DEVICE,
1186 * other), the flag may be cleared in the course of freeing the page;
1187 * but that is not required for correctness.
1189 if (!waitqueue_active(q) || !key.page_match)
1190 folio_clear_waiters(folio);
1192 spin_unlock_irqrestore(&q->lock, flags);
1196 * A choice of three behaviors for folio_wait_bit_common():
1199 EXCLUSIVE, /* Hold ref to page and take the bit when woken, like
1200 * __folio_lock() waiting on then setting PG_locked.
1202 SHARED, /* Hold ref to page and check the bit when woken, like
1203 * folio_wait_writeback() waiting on PG_writeback.
1205 DROP, /* Drop ref to page before wait, no check when woken,
1206 * like folio_put_wait_locked() on PG_locked.
1211 * Attempt to check (or get) the folio flag, and mark us done
1214 static inline bool folio_trylock_flag(struct folio *folio, int bit_nr,
1215 struct wait_queue_entry *wait)
1217 if (wait->flags & WQ_FLAG_EXCLUSIVE) {
1218 if (test_and_set_bit(bit_nr, &folio->flags))
1220 } else if (test_bit(bit_nr, &folio->flags))
1223 wait->flags |= WQ_FLAG_WOKEN | WQ_FLAG_DONE;
1227 /* How many times do we accept lock stealing from under a waiter? */
1228 int sysctl_page_lock_unfairness = 5;
1230 static inline int folio_wait_bit_common(struct folio *folio, int bit_nr,
1231 int state, enum behavior behavior)
1233 wait_queue_head_t *q = folio_waitqueue(folio);
1234 int unfairness = sysctl_page_lock_unfairness;
1235 struct wait_page_queue wait_page;
1236 wait_queue_entry_t *wait = &wait_page.wait;
1237 bool thrashing = false;
1238 unsigned long pflags;
1241 if (bit_nr == PG_locked &&
1242 !folio_test_uptodate(folio) && folio_test_workingset(folio)) {
1243 delayacct_thrashing_start(&in_thrashing);
1244 psi_memstall_enter(&pflags);
1249 wait->func = wake_page_function;
1250 wait_page.folio = folio;
1251 wait_page.bit_nr = bit_nr;
1255 if (behavior == EXCLUSIVE) {
1256 wait->flags = WQ_FLAG_EXCLUSIVE;
1257 if (--unfairness < 0)
1258 wait->flags |= WQ_FLAG_CUSTOM;
1262 * Do one last check whether we can get the
1263 * page bit synchronously.
1265 * Do the folio_set_waiters() marking before that
1266 * to let any waker we _just_ missed know they
1267 * need to wake us up (otherwise they'll never
1268 * even go to the slow case that looks at the
1269 * page queue), and add ourselves to the wait
1270 * queue if we need to sleep.
1272 * This part needs to be done under the queue
1273 * lock to avoid races.
1275 spin_lock_irq(&q->lock);
1276 folio_set_waiters(folio);
1277 if (!folio_trylock_flag(folio, bit_nr, wait))
1278 __add_wait_queue_entry_tail(q, wait);
1279 spin_unlock_irq(&q->lock);
1282 * From now on, all the logic will be based on
1283 * the WQ_FLAG_WOKEN and WQ_FLAG_DONE flag, to
1284 * see whether the page bit testing has already
1285 * been done by the wake function.
1287 * We can drop our reference to the folio.
1289 if (behavior == DROP)
1293 * Note that until the "finish_wait()", or until
1294 * we see the WQ_FLAG_WOKEN flag, we need to
1295 * be very careful with the 'wait->flags', because
1296 * we may race with a waker that sets them.
1301 set_current_state(state);
1303 /* Loop until we've been woken or interrupted */
1304 flags = smp_load_acquire(&wait->flags);
1305 if (!(flags & WQ_FLAG_WOKEN)) {
1306 if (signal_pending_state(state, current))
1313 /* If we were non-exclusive, we're done */
1314 if (behavior != EXCLUSIVE)
1317 /* If the waker got the lock for us, we're done */
1318 if (flags & WQ_FLAG_DONE)
1322 * Otherwise, if we're getting the lock, we need to
1323 * try to get it ourselves.
1325 * And if that fails, we'll have to retry this all.
1327 if (unlikely(test_and_set_bit(bit_nr, folio_flags(folio, 0))))
1330 wait->flags |= WQ_FLAG_DONE;
1335 * If a signal happened, this 'finish_wait()' may remove the last
1336 * waiter from the wait-queues, but the folio waiters bit will remain
1337 * set. That's ok. The next wakeup will take care of it, and trying
1338 * to do it here would be difficult and prone to races.
1340 finish_wait(q, wait);
1343 delayacct_thrashing_end(&in_thrashing);
1344 psi_memstall_leave(&pflags);
1348 * NOTE! The wait->flags weren't stable until we've done the
1349 * 'finish_wait()', and we could have exited the loop above due
1350 * to a signal, and had a wakeup event happen after the signal
1351 * test but before the 'finish_wait()'.
1353 * So only after the finish_wait() can we reliably determine
1354 * if we got woken up or not, so we can now figure out the final
1355 * return value based on that state without races.
1357 * Also note that WQ_FLAG_WOKEN is sufficient for a non-exclusive
1358 * waiter, but an exclusive one requires WQ_FLAG_DONE.
1360 if (behavior == EXCLUSIVE)
1361 return wait->flags & WQ_FLAG_DONE ? 0 : -EINTR;
1363 return wait->flags & WQ_FLAG_WOKEN ? 0 : -EINTR;
1366 #ifdef CONFIG_MIGRATION
1368 * migration_entry_wait_on_locked - Wait for a migration entry to be removed
1369 * @entry: migration swap entry.
1370 * @ptl: already locked ptl. This function will drop the lock.
1372 * Wait for a migration entry referencing the given page to be removed. This is
1373 * equivalent to put_and_wait_on_page_locked(page, TASK_UNINTERRUPTIBLE) except
1374 * this can be called without taking a reference on the page. Instead this
1375 * should be called while holding the ptl for the migration entry referencing
1378 * Returns after unlocking the ptl.
1380 * This follows the same logic as folio_wait_bit_common() so see the comments
1383 void migration_entry_wait_on_locked(swp_entry_t entry, spinlock_t *ptl)
1386 struct wait_page_queue wait_page;
1387 wait_queue_entry_t *wait = &wait_page.wait;
1388 bool thrashing = false;
1389 unsigned long pflags;
1391 wait_queue_head_t *q;
1392 struct folio *folio = pfn_swap_entry_folio(entry);
1394 q = folio_waitqueue(folio);
1395 if (!folio_test_uptodate(folio) && folio_test_workingset(folio)) {
1396 delayacct_thrashing_start(&in_thrashing);
1397 psi_memstall_enter(&pflags);
1402 wait->func = wake_page_function;
1403 wait_page.folio = folio;
1404 wait_page.bit_nr = PG_locked;
1407 spin_lock_irq(&q->lock);
1408 folio_set_waiters(folio);
1409 if (!folio_trylock_flag(folio, PG_locked, wait))
1410 __add_wait_queue_entry_tail(q, wait);
1411 spin_unlock_irq(&q->lock);
1414 * If a migration entry exists for the page the migration path must hold
1415 * a valid reference to the page, and it must take the ptl to remove the
1416 * migration entry. So the page is valid until the ptl is dropped.
1423 set_current_state(TASK_UNINTERRUPTIBLE);
1425 /* Loop until we've been woken or interrupted */
1426 flags = smp_load_acquire(&wait->flags);
1427 if (!(flags & WQ_FLAG_WOKEN)) {
1428 if (signal_pending_state(TASK_UNINTERRUPTIBLE, current))
1437 finish_wait(q, wait);
1440 delayacct_thrashing_end(&in_thrashing);
1441 psi_memstall_leave(&pflags);
1446 void folio_wait_bit(struct folio *folio, int bit_nr)
1448 folio_wait_bit_common(folio, bit_nr, TASK_UNINTERRUPTIBLE, SHARED);
1450 EXPORT_SYMBOL(folio_wait_bit);
1452 int folio_wait_bit_killable(struct folio *folio, int bit_nr)
1454 return folio_wait_bit_common(folio, bit_nr, TASK_KILLABLE, SHARED);
1456 EXPORT_SYMBOL(folio_wait_bit_killable);
1459 * folio_put_wait_locked - Drop a reference and wait for it to be unlocked
1460 * @folio: The folio to wait for.
1461 * @state: The sleep state (TASK_KILLABLE, TASK_UNINTERRUPTIBLE, etc).
1463 * The caller should hold a reference on @folio. They expect the page to
1464 * become unlocked relatively soon, but do not wish to hold up migration
1465 * (for example) by holding the reference while waiting for the folio to
1466 * come unlocked. After this function returns, the caller should not
1467 * dereference @folio.
1469 * Return: 0 if the folio was unlocked or -EINTR if interrupted by a signal.
1471 static int folio_put_wait_locked(struct folio *folio, int state)
1473 return folio_wait_bit_common(folio, PG_locked, state, DROP);
1477 * folio_add_wait_queue - Add an arbitrary waiter to a folio's wait queue
1478 * @folio: Folio defining the wait queue of interest
1479 * @waiter: Waiter to add to the queue
1481 * Add an arbitrary @waiter to the wait queue for the nominated @folio.
1483 void folio_add_wait_queue(struct folio *folio, wait_queue_entry_t *waiter)
1485 wait_queue_head_t *q = folio_waitqueue(folio);
1486 unsigned long flags;
1488 spin_lock_irqsave(&q->lock, flags);
1489 __add_wait_queue_entry_tail(q, waiter);
1490 folio_set_waiters(folio);
1491 spin_unlock_irqrestore(&q->lock, flags);
1493 EXPORT_SYMBOL_GPL(folio_add_wait_queue);
1496 * folio_unlock - Unlock a locked folio.
1497 * @folio: The folio.
1499 * Unlocks the folio and wakes up any thread sleeping on the page lock.
1501 * Context: May be called from interrupt or process context. May not be
1502 * called from NMI context.
1504 void folio_unlock(struct folio *folio)
1506 /* Bit 7 allows x86 to check the byte's sign bit */
1507 BUILD_BUG_ON(PG_waiters != 7);
1508 BUILD_BUG_ON(PG_locked > 7);
1509 VM_BUG_ON_FOLIO(!folio_test_locked(folio), folio);
1510 if (folio_xor_flags_has_waiters(folio, 1 << PG_locked))
1511 folio_wake_bit(folio, PG_locked);
1513 EXPORT_SYMBOL(folio_unlock);
1516 * folio_end_read - End read on a folio.
1517 * @folio: The folio.
1518 * @success: True if all reads completed successfully.
1520 * When all reads against a folio have completed, filesystems should
1521 * call this function to let the pagecache know that no more reads
1522 * are outstanding. This will unlock the folio and wake up any thread
1523 * sleeping on the lock. The folio will also be marked uptodate if all
1526 * Context: May be called from interrupt or process context. May not be
1527 * called from NMI context.
1529 void folio_end_read(struct folio *folio, bool success)
1531 unsigned long mask = 1 << PG_locked;
1533 /* Must be in bottom byte for x86 to work */
1534 BUILD_BUG_ON(PG_uptodate > 7);
1535 VM_BUG_ON_FOLIO(!folio_test_locked(folio), folio);
1536 VM_BUG_ON_FOLIO(folio_test_uptodate(folio), folio);
1538 if (likely(success))
1539 mask |= 1 << PG_uptodate;
1540 if (folio_xor_flags_has_waiters(folio, mask))
1541 folio_wake_bit(folio, PG_locked);
1543 EXPORT_SYMBOL(folio_end_read);
1546 * folio_end_private_2 - Clear PG_private_2 and wake any waiters.
1547 * @folio: The folio.
1549 * Clear the PG_private_2 bit on a folio and wake up any sleepers waiting for
1550 * it. The folio reference held for PG_private_2 being set is released.
1552 * This is, for example, used when a netfs folio is being written to a local
1553 * disk cache, thereby allowing writes to the cache for the same folio to be
1556 void folio_end_private_2(struct folio *folio)
1558 VM_BUG_ON_FOLIO(!folio_test_private_2(folio), folio);
1559 clear_bit_unlock(PG_private_2, folio_flags(folio, 0));
1560 folio_wake_bit(folio, PG_private_2);
1563 EXPORT_SYMBOL(folio_end_private_2);
1566 * folio_wait_private_2 - Wait for PG_private_2 to be cleared on a folio.
1567 * @folio: The folio to wait on.
1569 * Wait for PG_private_2 to be cleared on a folio.
1571 void folio_wait_private_2(struct folio *folio)
1573 while (folio_test_private_2(folio))
1574 folio_wait_bit(folio, PG_private_2);
1576 EXPORT_SYMBOL(folio_wait_private_2);
1579 * folio_wait_private_2_killable - Wait for PG_private_2 to be cleared on a folio.
1580 * @folio: The folio to wait on.
1582 * Wait for PG_private_2 to be cleared on a folio or until a fatal signal is
1583 * received by the calling task.
1586 * - 0 if successful.
1587 * - -EINTR if a fatal signal was encountered.
1589 int folio_wait_private_2_killable(struct folio *folio)
1593 while (folio_test_private_2(folio)) {
1594 ret = folio_wait_bit_killable(folio, PG_private_2);
1601 EXPORT_SYMBOL(folio_wait_private_2_killable);
1604 * folio_end_writeback - End writeback against a folio.
1605 * @folio: The folio.
1607 * The folio must actually be under writeback.
1609 * Context: May be called from process or interrupt context.
1611 void folio_end_writeback(struct folio *folio)
1613 VM_BUG_ON_FOLIO(!folio_test_writeback(folio), folio);
1616 * folio_test_clear_reclaim() could be used here but it is an
1617 * atomic operation and overkill in this particular case. Failing
1618 * to shuffle a folio marked for immediate reclaim is too mild
1619 * a gain to justify taking an atomic operation penalty at the
1620 * end of every folio writeback.
1622 if (folio_test_reclaim(folio)) {
1623 folio_clear_reclaim(folio);
1624 folio_rotate_reclaimable(folio);
1628 * Writeback does not hold a folio reference of its own, relying
1629 * on truncation to wait for the clearing of PG_writeback.
1630 * But here we must make sure that the folio is not freed and
1631 * reused before the folio_wake_bit().
1634 if (__folio_end_writeback(folio))
1635 folio_wake_bit(folio, PG_writeback);
1636 acct_reclaim_writeback(folio);
1639 EXPORT_SYMBOL(folio_end_writeback);
1642 * __folio_lock - Get a lock on the folio, assuming we need to sleep to get it.
1643 * @folio: The folio to lock
1645 void __folio_lock(struct folio *folio)
1647 folio_wait_bit_common(folio, PG_locked, TASK_UNINTERRUPTIBLE,
1650 EXPORT_SYMBOL(__folio_lock);
1652 int __folio_lock_killable(struct folio *folio)
1654 return folio_wait_bit_common(folio, PG_locked, TASK_KILLABLE,
1657 EXPORT_SYMBOL_GPL(__folio_lock_killable);
1659 static int __folio_lock_async(struct folio *folio, struct wait_page_queue *wait)
1661 struct wait_queue_head *q = folio_waitqueue(folio);
1664 wait->folio = folio;
1665 wait->bit_nr = PG_locked;
1667 spin_lock_irq(&q->lock);
1668 __add_wait_queue_entry_tail(q, &wait->wait);
1669 folio_set_waiters(folio);
1670 ret = !folio_trylock(folio);
1672 * If we were successful now, we know we're still on the
1673 * waitqueue as we're still under the lock. This means it's
1674 * safe to remove and return success, we know the callback
1675 * isn't going to trigger.
1678 __remove_wait_queue(q, &wait->wait);
1681 spin_unlock_irq(&q->lock);
1687 * 0 - folio is locked.
1688 * non-zero - folio is not locked.
1689 * mmap_lock or per-VMA lock has been released (mmap_read_unlock() or
1690 * vma_end_read()), unless flags had both FAULT_FLAG_ALLOW_RETRY and
1691 * FAULT_FLAG_RETRY_NOWAIT set, in which case the lock is still held.
1693 * If neither ALLOW_RETRY nor KILLABLE are set, will always return 0
1694 * with the folio locked and the mmap_lock/per-VMA lock is left unperturbed.
1696 vm_fault_t __folio_lock_or_retry(struct folio *folio, struct vm_fault *vmf)
1698 unsigned int flags = vmf->flags;
1700 if (fault_flag_allow_retry_first(flags)) {
1702 * CAUTION! In this case, mmap_lock/per-VMA lock is not
1703 * released even though returning VM_FAULT_RETRY.
1705 if (flags & FAULT_FLAG_RETRY_NOWAIT)
1706 return VM_FAULT_RETRY;
1708 release_fault_lock(vmf);
1709 if (flags & FAULT_FLAG_KILLABLE)
1710 folio_wait_locked_killable(folio);
1712 folio_wait_locked(folio);
1713 return VM_FAULT_RETRY;
1715 if (flags & FAULT_FLAG_KILLABLE) {
1718 ret = __folio_lock_killable(folio);
1720 release_fault_lock(vmf);
1721 return VM_FAULT_RETRY;
1724 __folio_lock(folio);
1731 * page_cache_next_miss() - Find the next gap in the page cache.
1732 * @mapping: Mapping.
1734 * @max_scan: Maximum range to search.
1736 * Search the range [index, min(index + max_scan - 1, ULONG_MAX)] for the
1737 * gap with the lowest index.
1739 * This function may be called under the rcu_read_lock. However, this will
1740 * not atomically search a snapshot of the cache at a single point in time.
1741 * For example, if a gap is created at index 5, then subsequently a gap is
1742 * created at index 10, page_cache_next_miss covering both indices may
1743 * return 10 if called under the rcu_read_lock.
1745 * Return: The index of the gap if found, otherwise an index outside the
1746 * range specified (in which case 'return - index >= max_scan' will be true).
1747 * In the rare case of index wrap-around, 0 will be returned.
1749 pgoff_t page_cache_next_miss(struct address_space *mapping,
1750 pgoff_t index, unsigned long max_scan)
1752 XA_STATE(xas, &mapping->i_pages, index);
1754 while (max_scan--) {
1755 void *entry = xas_next(&xas);
1756 if (!entry || xa_is_value(entry))
1757 return xas.xa_index;
1758 if (xas.xa_index == 0)
1762 return index + max_scan;
1764 EXPORT_SYMBOL(page_cache_next_miss);
1767 * page_cache_prev_miss() - Find the previous gap in the page cache.
1768 * @mapping: Mapping.
1770 * @max_scan: Maximum range to search.
1772 * Search the range [max(index - max_scan + 1, 0), index] for the
1773 * gap with the highest index.
1775 * This function may be called under the rcu_read_lock. However, this will
1776 * not atomically search a snapshot of the cache at a single point in time.
1777 * For example, if a gap is created at index 10, then subsequently a gap is
1778 * created at index 5, page_cache_prev_miss() covering both indices may
1779 * return 5 if called under the rcu_read_lock.
1781 * Return: The index of the gap if found, otherwise an index outside the
1782 * range specified (in which case 'index - return >= max_scan' will be true).
1783 * In the rare case of wrap-around, ULONG_MAX will be returned.
1785 pgoff_t page_cache_prev_miss(struct address_space *mapping,
1786 pgoff_t index, unsigned long max_scan)
1788 XA_STATE(xas, &mapping->i_pages, index);
1790 while (max_scan--) {
1791 void *entry = xas_prev(&xas);
1792 if (!entry || xa_is_value(entry))
1794 if (xas.xa_index == ULONG_MAX)
1798 return xas.xa_index;
1800 EXPORT_SYMBOL(page_cache_prev_miss);
1803 * Lockless page cache protocol:
1804 * On the lookup side:
1805 * 1. Load the folio from i_pages
1806 * 2. Increment the refcount if it's not zero
1807 * 3. If the folio is not found by xas_reload(), put the refcount and retry
1809 * On the removal side:
1810 * A. Freeze the page (by zeroing the refcount if nobody else has a reference)
1811 * B. Remove the page from i_pages
1812 * C. Return the page to the page allocator
1814 * This means that any page may have its reference count temporarily
1815 * increased by a speculative page cache (or GUP-fast) lookup as it can
1816 * be allocated by another user before the RCU grace period expires.
1817 * Because the refcount temporarily acquired here may end up being the
1818 * last refcount on the page, any page allocation must be freeable by
1823 * filemap_get_entry - Get a page cache entry.
1824 * @mapping: the address_space to search
1825 * @index: The page cache index.
1827 * Looks up the page cache entry at @mapping & @index. If it is a folio,
1828 * it is returned with an increased refcount. If it is a shadow entry
1829 * of a previously evicted folio, or a swap entry from shmem/tmpfs,
1830 * it is returned without further action.
1832 * Return: The folio, swap or shadow entry, %NULL if nothing is found.
1834 void *filemap_get_entry(struct address_space *mapping, pgoff_t index)
1836 XA_STATE(xas, &mapping->i_pages, index);
1837 struct folio *folio;
1842 folio = xas_load(&xas);
1843 if (xas_retry(&xas, folio))
1846 * A shadow entry of a recently evicted page, or a swap entry from
1847 * shmem/tmpfs. Return it without attempting to raise page count.
1849 if (!folio || xa_is_value(folio))
1852 if (!folio_try_get(folio))
1855 if (unlikely(folio != xas_reload(&xas))) {
1866 * __filemap_get_folio - Find and get a reference to a folio.
1867 * @mapping: The address_space to search.
1868 * @index: The page index.
1869 * @fgp_flags: %FGP flags modify how the folio is returned.
1870 * @gfp: Memory allocation flags to use if %FGP_CREAT is specified.
1872 * Looks up the page cache entry at @mapping & @index.
1874 * If %FGP_LOCK or %FGP_CREAT are specified then the function may sleep even
1875 * if the %GFP flags specified for %FGP_CREAT are atomic.
1877 * If this function returns a folio, it is returned with an increased refcount.
1879 * Return: The found folio or an ERR_PTR() otherwise.
1881 struct folio *__filemap_get_folio(struct address_space *mapping, pgoff_t index,
1882 fgf_t fgp_flags, gfp_t gfp)
1884 struct folio *folio;
1887 folio = filemap_get_entry(mapping, index);
1888 if (xa_is_value(folio))
1893 if (fgp_flags & FGP_LOCK) {
1894 if (fgp_flags & FGP_NOWAIT) {
1895 if (!folio_trylock(folio)) {
1897 return ERR_PTR(-EAGAIN);
1903 /* Has the page been truncated? */
1904 if (unlikely(folio->mapping != mapping)) {
1905 folio_unlock(folio);
1909 VM_BUG_ON_FOLIO(!folio_contains(folio, index), folio);
1912 if (fgp_flags & FGP_ACCESSED)
1913 folio_mark_accessed(folio);
1914 else if (fgp_flags & FGP_WRITE) {
1915 /* Clear idle flag for buffer write */
1916 if (folio_test_idle(folio))
1917 folio_clear_idle(folio);
1920 if (fgp_flags & FGP_STABLE)
1921 folio_wait_stable(folio);
1923 if (!folio && (fgp_flags & FGP_CREAT)) {
1924 unsigned int min_order = mapping_min_folio_order(mapping);
1925 unsigned int order = max(min_order, FGF_GET_ORDER(fgp_flags));
1927 index = mapping_align_index(mapping, index);
1929 if ((fgp_flags & FGP_WRITE) && mapping_can_writeback(mapping))
1931 if (fgp_flags & FGP_NOFS)
1933 if (fgp_flags & FGP_NOWAIT) {
1935 gfp |= GFP_NOWAIT | __GFP_NOWARN;
1937 if (WARN_ON_ONCE(!(fgp_flags & (FGP_LOCK | FGP_FOR_MMAP))))
1938 fgp_flags |= FGP_LOCK;
1940 if (order > mapping_max_folio_order(mapping))
1941 order = mapping_max_folio_order(mapping);
1942 /* If we're not aligned, allocate a smaller folio */
1943 if (index & ((1UL << order) - 1))
1944 order = __ffs(index);
1947 gfp_t alloc_gfp = gfp;
1950 if (order > min_order)
1951 alloc_gfp |= __GFP_NORETRY | __GFP_NOWARN;
1952 folio = filemap_alloc_folio(alloc_gfp, order);
1956 /* Init accessed so avoid atomic mark_page_accessed later */
1957 if (fgp_flags & FGP_ACCESSED)
1958 __folio_set_referenced(folio);
1960 err = filemap_add_folio(mapping, folio, index, gfp);
1965 } while (order-- > min_order);
1970 return ERR_PTR(err);
1972 * filemap_add_folio locks the page, and for mmap
1973 * we expect an unlocked page.
1975 if (folio && (fgp_flags & FGP_FOR_MMAP))
1976 folio_unlock(folio);
1980 return ERR_PTR(-ENOENT);
1983 EXPORT_SYMBOL(__filemap_get_folio);
1985 static inline struct folio *find_get_entry(struct xa_state *xas, pgoff_t max,
1988 struct folio *folio;
1991 if (mark == XA_PRESENT)
1992 folio = xas_find(xas, max);
1994 folio = xas_find_marked(xas, max, mark);
1996 if (xas_retry(xas, folio))
1999 * A shadow entry of a recently evicted page, a swap
2000 * entry from shmem/tmpfs or a DAX entry. Return it
2001 * without attempting to raise page count.
2003 if (!folio || xa_is_value(folio))
2006 if (!folio_try_get(folio))
2009 if (unlikely(folio != xas_reload(xas))) {
2021 * find_get_entries - gang pagecache lookup
2022 * @mapping: The address_space to search
2023 * @start: The starting page cache index
2024 * @end: The final page index (inclusive).
2025 * @fbatch: Where the resulting entries are placed.
2026 * @indices: The cache indices corresponding to the entries in @entries
2028 * find_get_entries() will search for and return a batch of entries in
2029 * the mapping. The entries are placed in @fbatch. find_get_entries()
2030 * takes a reference on any actual folios it returns.
2032 * The entries have ascending indexes. The indices may not be consecutive
2033 * due to not-present entries or large folios.
2035 * Any shadow entries of evicted folios, or swap entries from
2036 * shmem/tmpfs, are included in the returned array.
2038 * Return: The number of entries which were found.
2040 unsigned find_get_entries(struct address_space *mapping, pgoff_t *start,
2041 pgoff_t end, struct folio_batch *fbatch, pgoff_t *indices)
2043 XA_STATE(xas, &mapping->i_pages, *start);
2044 struct folio *folio;
2047 while ((folio = find_get_entry(&xas, end, XA_PRESENT)) != NULL) {
2048 indices[fbatch->nr] = xas.xa_index;
2049 if (!folio_batch_add(fbatch, folio))
2054 if (folio_batch_count(fbatch)) {
2055 unsigned long nr = 1;
2056 int idx = folio_batch_count(fbatch) - 1;
2058 folio = fbatch->folios[idx];
2059 if (!xa_is_value(folio))
2060 nr = folio_nr_pages(folio);
2061 *start = indices[idx] + nr;
2063 return folio_batch_count(fbatch);
2067 * find_lock_entries - Find a batch of pagecache entries.
2068 * @mapping: The address_space to search.
2069 * @start: The starting page cache index.
2070 * @end: The final page index (inclusive).
2071 * @fbatch: Where the resulting entries are placed.
2072 * @indices: The cache indices of the entries in @fbatch.
2074 * find_lock_entries() will return a batch of entries from @mapping.
2075 * Swap, shadow and DAX entries are included. Folios are returned
2076 * locked and with an incremented refcount. Folios which are locked
2077 * by somebody else or under writeback are skipped. Folios which are
2078 * partially outside the range are not returned.
2080 * The entries have ascending indexes. The indices may not be consecutive
2081 * due to not-present entries, large folios, folios which could not be
2082 * locked or folios under writeback.
2084 * Return: The number of entries which were found.
2086 unsigned find_lock_entries(struct address_space *mapping, pgoff_t *start,
2087 pgoff_t end, struct folio_batch *fbatch, pgoff_t *indices)
2089 XA_STATE(xas, &mapping->i_pages, *start);
2090 struct folio *folio;
2093 while ((folio = find_get_entry(&xas, end, XA_PRESENT))) {
2094 if (!xa_is_value(folio)) {
2095 if (folio->index < *start)
2097 if (folio_next_index(folio) - 1 > end)
2099 if (!folio_trylock(folio))
2101 if (folio->mapping != mapping ||
2102 folio_test_writeback(folio))
2104 VM_BUG_ON_FOLIO(!folio_contains(folio, xas.xa_index),
2107 indices[fbatch->nr] = xas.xa_index;
2108 if (!folio_batch_add(fbatch, folio))
2112 folio_unlock(folio);
2118 if (folio_batch_count(fbatch)) {
2119 unsigned long nr = 1;
2120 int idx = folio_batch_count(fbatch) - 1;
2122 folio = fbatch->folios[idx];
2123 if (!xa_is_value(folio))
2124 nr = folio_nr_pages(folio);
2125 *start = indices[idx] + nr;
2127 return folio_batch_count(fbatch);
2131 * filemap_get_folios - Get a batch of folios
2132 * @mapping: The address_space to search
2133 * @start: The starting page index
2134 * @end: The final page index (inclusive)
2135 * @fbatch: The batch to fill.
2137 * Search for and return a batch of folios in the mapping starting at
2138 * index @start and up to index @end (inclusive). The folios are returned
2139 * in @fbatch with an elevated reference count.
2141 * Return: The number of folios which were found.
2142 * We also update @start to index the next folio for the traversal.
2144 unsigned filemap_get_folios(struct address_space *mapping, pgoff_t *start,
2145 pgoff_t end, struct folio_batch *fbatch)
2147 return filemap_get_folios_tag(mapping, start, end, XA_PRESENT, fbatch);
2149 EXPORT_SYMBOL(filemap_get_folios);
2152 * filemap_get_folios_contig - Get a batch of contiguous folios
2153 * @mapping: The address_space to search
2154 * @start: The starting page index
2155 * @end: The final page index (inclusive)
2156 * @fbatch: The batch to fill
2158 * filemap_get_folios_contig() works exactly like filemap_get_folios(),
2159 * except the returned folios are guaranteed to be contiguous. This may
2160 * not return all contiguous folios if the batch gets filled up.
2162 * Return: The number of folios found.
2163 * Also update @start to be positioned for traversal of the next folio.
2166 unsigned filemap_get_folios_contig(struct address_space *mapping,
2167 pgoff_t *start, pgoff_t end, struct folio_batch *fbatch)
2169 XA_STATE(xas, &mapping->i_pages, *start);
2171 struct folio *folio;
2175 for (folio = xas_load(&xas); folio && xas.xa_index <= end;
2176 folio = xas_next(&xas)) {
2177 if (xas_retry(&xas, folio))
2180 * If the entry has been swapped out, we can stop looking.
2181 * No current caller is looking for DAX entries.
2183 if (xa_is_value(folio))
2186 if (!folio_try_get(folio))
2189 if (unlikely(folio != xas_reload(&xas)))
2192 if (!folio_batch_add(fbatch, folio)) {
2193 nr = folio_nr_pages(folio);
2194 *start = folio->index + nr;
2206 nr = folio_batch_count(fbatch);
2209 folio = fbatch->folios[nr - 1];
2210 *start = folio_next_index(folio);
2214 return folio_batch_count(fbatch);
2216 EXPORT_SYMBOL(filemap_get_folios_contig);
2219 * filemap_get_folios_tag - Get a batch of folios matching @tag
2220 * @mapping: The address_space to search
2221 * @start: The starting page index
2222 * @end: The final page index (inclusive)
2223 * @tag: The tag index
2224 * @fbatch: The batch to fill
2226 * The first folio may start before @start; if it does, it will contain
2227 * @start. The final folio may extend beyond @end; if it does, it will
2228 * contain @end. The folios have ascending indices. There may be gaps
2229 * between the folios if there are indices which have no folio in the
2230 * page cache. If folios are added to or removed from the page cache
2231 * while this is running, they may or may not be found by this call.
2232 * Only returns folios that are tagged with @tag.
2234 * Return: The number of folios found.
2235 * Also update @start to index the next folio for traversal.
2237 unsigned filemap_get_folios_tag(struct address_space *mapping, pgoff_t *start,
2238 pgoff_t end, xa_mark_t tag, struct folio_batch *fbatch)
2240 XA_STATE(xas, &mapping->i_pages, *start);
2241 struct folio *folio;
2244 while ((folio = find_get_entry(&xas, end, tag)) != NULL) {
2246 * Shadow entries should never be tagged, but this iteration
2247 * is lockless so there is a window for page reclaim to evict
2248 * a page we saw tagged. Skip over it.
2250 if (xa_is_value(folio))
2252 if (!folio_batch_add(fbatch, folio)) {
2253 unsigned long nr = folio_nr_pages(folio);
2254 *start = folio->index + nr;
2259 * We come here when there is no page beyond @end. We take care to not
2260 * overflow the index @start as it confuses some of the callers. This
2261 * breaks the iteration when there is a page at index -1 but that is
2262 * already broke anyway.
2264 if (end == (pgoff_t)-1)
2265 *start = (pgoff_t)-1;
2271 return folio_batch_count(fbatch);
2273 EXPORT_SYMBOL(filemap_get_folios_tag);
2276 * CD/DVDs are error prone. When a medium error occurs, the driver may fail
2277 * a _large_ part of the i/o request. Imagine the worst scenario:
2279 * ---R__________________________________________B__________
2280 * ^ reading here ^ bad block(assume 4k)
2282 * read(R) => miss => readahead(R...B) => media error => frustrating retries
2283 * => failing the whole request => read(R) => read(R+1) =>
2284 * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
2285 * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
2286 * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
2288 * It is going insane. Fix it by quickly scaling down the readahead size.
2290 static void shrink_readahead_size_eio(struct file_ra_state *ra)
2296 * filemap_get_read_batch - Get a batch of folios for read
2298 * Get a batch of folios which represent a contiguous range of bytes in
2299 * the file. No exceptional entries will be returned. If @index is in
2300 * the middle of a folio, the entire folio will be returned. The last
2301 * folio in the batch may have the readahead flag set or the uptodate flag
2302 * clear so that the caller can take the appropriate action.
2304 static void filemap_get_read_batch(struct address_space *mapping,
2305 pgoff_t index, pgoff_t max, struct folio_batch *fbatch)
2307 XA_STATE(xas, &mapping->i_pages, index);
2308 struct folio *folio;
2311 for (folio = xas_load(&xas); folio; folio = xas_next(&xas)) {
2312 if (xas_retry(&xas, folio))
2314 if (xas.xa_index > max || xa_is_value(folio))
2316 if (xa_is_sibling(folio))
2318 if (!folio_try_get(folio))
2321 if (unlikely(folio != xas_reload(&xas)))
2324 if (!folio_batch_add(fbatch, folio))
2326 if (!folio_test_uptodate(folio))
2328 if (folio_test_readahead(folio))
2330 xas_advance(&xas, folio_next_index(folio) - 1);
2340 static int filemap_read_folio(struct file *file, filler_t filler,
2341 struct folio *folio)
2343 bool workingset = folio_test_workingset(folio);
2344 unsigned long pflags;
2348 * A previous I/O error may have been due to temporary failures,
2349 * eg. multipath errors. PG_error will be set again if read_folio
2352 folio_clear_error(folio);
2354 /* Start the actual read. The read will unlock the page. */
2355 if (unlikely(workingset))
2356 psi_memstall_enter(&pflags);
2357 error = filler(file, folio);
2358 if (unlikely(workingset))
2359 psi_memstall_leave(&pflags);
2363 error = folio_wait_locked_killable(folio);
2366 if (folio_test_uptodate(folio))
2369 shrink_readahead_size_eio(&file->f_ra);
2373 static bool filemap_range_uptodate(struct address_space *mapping,
2374 loff_t pos, size_t count, struct folio *folio,
2377 if (folio_test_uptodate(folio))
2379 /* pipes can't handle partially uptodate pages */
2382 if (!mapping->a_ops->is_partially_uptodate)
2384 if (mapping->host->i_blkbits >= folio_shift(folio))
2387 if (folio_pos(folio) > pos) {
2388 count -= folio_pos(folio) - pos;
2391 pos -= folio_pos(folio);
2394 return mapping->a_ops->is_partially_uptodate(folio, pos, count);
2397 static int filemap_update_page(struct kiocb *iocb,
2398 struct address_space *mapping, size_t count,
2399 struct folio *folio, bool need_uptodate)
2403 if (iocb->ki_flags & IOCB_NOWAIT) {
2404 if (!filemap_invalidate_trylock_shared(mapping))
2407 filemap_invalidate_lock_shared(mapping);
2410 if (!folio_trylock(folio)) {
2412 if (iocb->ki_flags & (IOCB_NOWAIT | IOCB_NOIO))
2413 goto unlock_mapping;
2414 if (!(iocb->ki_flags & IOCB_WAITQ)) {
2415 filemap_invalidate_unlock_shared(mapping);
2417 * This is where we usually end up waiting for a
2418 * previously submitted readahead to finish.
2420 folio_put_wait_locked(folio, TASK_KILLABLE);
2421 return AOP_TRUNCATED_PAGE;
2423 error = __folio_lock_async(folio, iocb->ki_waitq);
2425 goto unlock_mapping;
2428 error = AOP_TRUNCATED_PAGE;
2429 if (!folio->mapping)
2433 if (filemap_range_uptodate(mapping, iocb->ki_pos, count, folio,
2438 if (iocb->ki_flags & (IOCB_NOIO | IOCB_NOWAIT | IOCB_WAITQ))
2441 error = filemap_read_folio(iocb->ki_filp, mapping->a_ops->read_folio,
2443 goto unlock_mapping;
2445 folio_unlock(folio);
2447 filemap_invalidate_unlock_shared(mapping);
2448 if (error == AOP_TRUNCATED_PAGE)
2453 static int filemap_create_folio(struct file *file,
2454 struct address_space *mapping, loff_t pos,
2455 struct folio_batch *fbatch)
2457 struct folio *folio;
2459 unsigned int min_order = mapping_min_folio_order(mapping);
2462 folio = filemap_alloc_folio(mapping_gfp_mask(mapping), min_order);
2467 * Protect against truncate / hole punch. Grabbing invalidate_lock
2468 * here assures we cannot instantiate and bring uptodate new
2469 * pagecache folios after evicting page cache during truncate
2470 * and before actually freeing blocks. Note that we could
2471 * release invalidate_lock after inserting the folio into
2472 * the page cache as the locked folio would then be enough to
2473 * synchronize with hole punching. But there are code paths
2474 * such as filemap_update_page() filling in partially uptodate
2475 * pages or ->readahead() that need to hold invalidate_lock
2476 * while mapping blocks for IO so let's hold the lock here as
2477 * well to keep locking rules simple.
2479 filemap_invalidate_lock_shared(mapping);
2480 index = (pos >> (PAGE_SHIFT + min_order)) << min_order;
2481 error = filemap_add_folio(mapping, folio, index,
2482 mapping_gfp_constraint(mapping, GFP_KERNEL));
2483 if (error == -EEXIST)
2484 error = AOP_TRUNCATED_PAGE;
2488 error = filemap_read_folio(file, mapping->a_ops->read_folio, folio);
2492 filemap_invalidate_unlock_shared(mapping);
2493 folio_batch_add(fbatch, folio);
2496 filemap_invalidate_unlock_shared(mapping);
2501 static int filemap_readahead(struct kiocb *iocb, struct file *file,
2502 struct address_space *mapping, struct folio *folio,
2505 DEFINE_READAHEAD(ractl, file, &file->f_ra, mapping, folio->index);
2507 if (iocb->ki_flags & IOCB_NOIO)
2509 page_cache_async_ra(&ractl, folio, last_index - folio->index);
2513 static int filemap_get_pages(struct kiocb *iocb, size_t count,
2514 struct folio_batch *fbatch, bool need_uptodate)
2516 struct file *filp = iocb->ki_filp;
2517 struct address_space *mapping = filp->f_mapping;
2518 struct file_ra_state *ra = &filp->f_ra;
2519 pgoff_t index = iocb->ki_pos >> PAGE_SHIFT;
2521 struct folio *folio;
2524 /* "last_index" is the index of the page beyond the end of the read */
2525 last_index = DIV_ROUND_UP(iocb->ki_pos + count, PAGE_SIZE);
2527 if (fatal_signal_pending(current))
2530 filemap_get_read_batch(mapping, index, last_index - 1, fbatch);
2531 if (!folio_batch_count(fbatch)) {
2532 if (iocb->ki_flags & IOCB_NOIO)
2534 page_cache_sync_readahead(mapping, ra, filp, index,
2535 last_index - index);
2536 filemap_get_read_batch(mapping, index, last_index - 1, fbatch);
2538 if (!folio_batch_count(fbatch)) {
2539 if (iocb->ki_flags & (IOCB_NOWAIT | IOCB_WAITQ))
2541 err = filemap_create_folio(filp, mapping, iocb->ki_pos, fbatch);
2542 if (err == AOP_TRUNCATED_PAGE)
2547 folio = fbatch->folios[folio_batch_count(fbatch) - 1];
2548 if (folio_test_readahead(folio)) {
2549 err = filemap_readahead(iocb, filp, mapping, folio, last_index);
2553 if (!folio_test_uptodate(folio)) {
2554 if ((iocb->ki_flags & IOCB_WAITQ) &&
2555 folio_batch_count(fbatch) > 1)
2556 iocb->ki_flags |= IOCB_NOWAIT;
2557 err = filemap_update_page(iocb, mapping, count, folio,
2567 if (likely(--fbatch->nr))
2569 if (err == AOP_TRUNCATED_PAGE)
2574 static inline bool pos_same_folio(loff_t pos1, loff_t pos2, struct folio *folio)
2576 unsigned int shift = folio_shift(folio);
2578 return (pos1 >> shift == pos2 >> shift);
2582 * filemap_read - Read data from the page cache.
2583 * @iocb: The iocb to read.
2584 * @iter: Destination for the data.
2585 * @already_read: Number of bytes already read by the caller.
2587 * Copies data from the page cache. If the data is not currently present,
2588 * uses the readahead and read_folio address_space operations to fetch it.
2590 * Return: Total number of bytes copied, including those already read by
2591 * the caller. If an error happens before any bytes are copied, returns
2592 * a negative error number.
2594 ssize_t filemap_read(struct kiocb *iocb, struct iov_iter *iter,
2595 ssize_t already_read)
2597 struct file *filp = iocb->ki_filp;
2598 struct file_ra_state *ra = &filp->f_ra;
2599 struct address_space *mapping = filp->f_mapping;
2600 struct inode *inode = mapping->host;
2601 struct folio_batch fbatch;
2603 bool writably_mapped;
2604 loff_t isize, end_offset;
2605 loff_t last_pos = ra->prev_pos;
2607 if (unlikely(iocb->ki_pos >= inode->i_sb->s_maxbytes))
2609 if (unlikely(!iov_iter_count(iter)))
2612 iov_iter_truncate(iter, inode->i_sb->s_maxbytes);
2613 folio_batch_init(&fbatch);
2619 * If we've already successfully copied some data, then we
2620 * can no longer safely return -EIOCBQUEUED. Hence mark
2621 * an async read NOWAIT at that point.
2623 if ((iocb->ki_flags & IOCB_WAITQ) && already_read)
2624 iocb->ki_flags |= IOCB_NOWAIT;
2626 if (unlikely(iocb->ki_pos >= i_size_read(inode)))
2629 error = filemap_get_pages(iocb, iter->count, &fbatch, false);
2634 * i_size must be checked after we know the pages are Uptodate.
2636 * Checking i_size after the check allows us to calculate
2637 * the correct value for "nr", which means the zero-filled
2638 * part of the page is not copied back to userspace (unless
2639 * another truncate extends the file - this is desired though).
2641 isize = i_size_read(inode);
2642 if (unlikely(iocb->ki_pos >= isize))
2644 end_offset = min_t(loff_t, isize, iocb->ki_pos + iter->count);
2647 * Once we start copying data, we don't want to be touching any
2648 * cachelines that might be contended:
2650 writably_mapped = mapping_writably_mapped(mapping);
2653 * When a read accesses the same folio several times, only
2654 * mark it as accessed the first time.
2656 if (!pos_same_folio(iocb->ki_pos, last_pos - 1,
2658 folio_mark_accessed(fbatch.folios[0]);
2660 for (i = 0; i < folio_batch_count(&fbatch); i++) {
2661 struct folio *folio = fbatch.folios[i];
2662 size_t fsize = folio_size(folio);
2663 size_t offset = iocb->ki_pos & (fsize - 1);
2664 size_t bytes = min_t(loff_t, end_offset - iocb->ki_pos,
2668 if (end_offset < folio_pos(folio))
2671 folio_mark_accessed(folio);
2673 * If users can be writing to this folio using arbitrary
2674 * virtual addresses, take care of potential aliasing
2675 * before reading the folio on the kernel side.
2677 if (writably_mapped)
2678 flush_dcache_folio(folio);
2680 copied = copy_folio_to_iter(folio, offset, bytes, iter);
2682 already_read += copied;
2683 iocb->ki_pos += copied;
2684 last_pos = iocb->ki_pos;
2686 if (copied < bytes) {
2692 for (i = 0; i < folio_batch_count(&fbatch); i++)
2693 folio_put(fbatch.folios[i]);
2694 folio_batch_init(&fbatch);
2695 } while (iov_iter_count(iter) && iocb->ki_pos < isize && !error);
2697 file_accessed(filp);
2698 ra->prev_pos = last_pos;
2699 return already_read ? already_read : error;
2701 EXPORT_SYMBOL_GPL(filemap_read);
2703 int kiocb_write_and_wait(struct kiocb *iocb, size_t count)
2705 struct address_space *mapping = iocb->ki_filp->f_mapping;
2706 loff_t pos = iocb->ki_pos;
2707 loff_t end = pos + count - 1;
2709 if (iocb->ki_flags & IOCB_NOWAIT) {
2710 if (filemap_range_needs_writeback(mapping, pos, end))
2715 return filemap_write_and_wait_range(mapping, pos, end);
2717 EXPORT_SYMBOL_GPL(kiocb_write_and_wait);
2719 int filemap_invalidate_pages(struct address_space *mapping,
2720 loff_t pos, loff_t end, bool nowait)
2725 /* we could block if there are any pages in the range */
2726 if (filemap_range_has_page(mapping, pos, end))
2729 ret = filemap_write_and_wait_range(mapping, pos, end);
2735 * After a write we want buffered reads to be sure to go to disk to get
2736 * the new data. We invalidate clean cached page from the region we're
2737 * about to write. We do this *before* the write so that we can return
2738 * without clobbering -EIOCBQUEUED from ->direct_IO().
2740 return invalidate_inode_pages2_range(mapping, pos >> PAGE_SHIFT,
2744 int kiocb_invalidate_pages(struct kiocb *iocb, size_t count)
2746 struct address_space *mapping = iocb->ki_filp->f_mapping;
2748 return filemap_invalidate_pages(mapping, iocb->ki_pos,
2749 iocb->ki_pos + count - 1,
2750 iocb->ki_flags & IOCB_NOWAIT);
2752 EXPORT_SYMBOL_GPL(kiocb_invalidate_pages);
2755 * generic_file_read_iter - generic filesystem read routine
2756 * @iocb: kernel I/O control block
2757 * @iter: destination for the data read
2759 * This is the "read_iter()" routine for all filesystems
2760 * that can use the page cache directly.
2762 * The IOCB_NOWAIT flag in iocb->ki_flags indicates that -EAGAIN shall
2763 * be returned when no data can be read without waiting for I/O requests
2764 * to complete; it doesn't prevent readahead.
2766 * The IOCB_NOIO flag in iocb->ki_flags indicates that no new I/O
2767 * requests shall be made for the read or for readahead. When no data
2768 * can be read, -EAGAIN shall be returned. When readahead would be
2769 * triggered, a partial, possibly empty read shall be returned.
2772 * * number of bytes copied, even for partial reads
2773 * * negative error code (or 0 if IOCB_NOIO) if nothing was read
2776 generic_file_read_iter(struct kiocb *iocb, struct iov_iter *iter)
2778 size_t count = iov_iter_count(iter);
2782 return 0; /* skip atime */
2784 if (iocb->ki_flags & IOCB_DIRECT) {
2785 struct file *file = iocb->ki_filp;
2786 struct address_space *mapping = file->f_mapping;
2787 struct inode *inode = mapping->host;
2789 retval = kiocb_write_and_wait(iocb, count);
2792 file_accessed(file);
2794 retval = mapping->a_ops->direct_IO(iocb, iter);
2796 iocb->ki_pos += retval;
2799 if (retval != -EIOCBQUEUED)
2800 iov_iter_revert(iter, count - iov_iter_count(iter));
2803 * Btrfs can have a short DIO read if we encounter
2804 * compressed extents, so if there was an error, or if
2805 * we've already read everything we wanted to, or if
2806 * there was a short read because we hit EOF, go ahead
2807 * and return. Otherwise fallthrough to buffered io for
2808 * the rest of the read. Buffered reads will not work for
2809 * DAX files, so don't bother trying.
2811 if (retval < 0 || !count || IS_DAX(inode))
2813 if (iocb->ki_pos >= i_size_read(inode))
2817 return filemap_read(iocb, iter, retval);
2819 EXPORT_SYMBOL(generic_file_read_iter);
2822 * Splice subpages from a folio into a pipe.
2824 size_t splice_folio_into_pipe(struct pipe_inode_info *pipe,
2825 struct folio *folio, loff_t fpos, size_t size)
2828 size_t spliced = 0, offset = offset_in_folio(folio, fpos);
2830 page = folio_page(folio, offset / PAGE_SIZE);
2831 size = min(size, folio_size(folio) - offset);
2832 offset %= PAGE_SIZE;
2834 while (spliced < size &&
2835 !pipe_full(pipe->head, pipe->tail, pipe->max_usage)) {
2836 struct pipe_buffer *buf = pipe_head_buf(pipe);
2837 size_t part = min_t(size_t, PAGE_SIZE - offset, size - spliced);
2839 *buf = (struct pipe_buffer) {
2840 .ops = &page_cache_pipe_buf_ops,
2856 * filemap_splice_read - Splice data from a file's pagecache into a pipe
2857 * @in: The file to read from
2858 * @ppos: Pointer to the file position to read from
2859 * @pipe: The pipe to splice into
2860 * @len: The amount to splice
2861 * @flags: The SPLICE_F_* flags
2863 * This function gets folios from a file's pagecache and splices them into the
2864 * pipe. Readahead will be called as necessary to fill more folios. This may
2865 * be used for blockdevs also.
2867 * Return: On success, the number of bytes read will be returned and *@ppos
2868 * will be updated if appropriate; 0 will be returned if there is no more data
2869 * to be read; -EAGAIN will be returned if the pipe had no space, and some
2870 * other negative error code will be returned on error. A short read may occur
2871 * if the pipe has insufficient space, we reach the end of the data or we hit a
2874 ssize_t filemap_splice_read(struct file *in, loff_t *ppos,
2875 struct pipe_inode_info *pipe,
2876 size_t len, unsigned int flags)
2878 struct folio_batch fbatch;
2880 size_t total_spliced = 0, used, npages;
2881 loff_t isize, end_offset;
2882 bool writably_mapped;
2885 if (unlikely(*ppos >= in->f_mapping->host->i_sb->s_maxbytes))
2888 init_sync_kiocb(&iocb, in);
2889 iocb.ki_pos = *ppos;
2891 /* Work out how much data we can actually add into the pipe */
2892 used = pipe_occupancy(pipe->head, pipe->tail);
2893 npages = max_t(ssize_t, pipe->max_usage - used, 0);
2894 len = min_t(size_t, len, npages * PAGE_SIZE);
2896 folio_batch_init(&fbatch);
2901 if (*ppos >= i_size_read(in->f_mapping->host))
2904 iocb.ki_pos = *ppos;
2905 error = filemap_get_pages(&iocb, len, &fbatch, true);
2910 * i_size must be checked after we know the pages are Uptodate.
2912 * Checking i_size after the check allows us to calculate
2913 * the correct value for "nr", which means the zero-filled
2914 * part of the page is not copied back to userspace (unless
2915 * another truncate extends the file - this is desired though).
2917 isize = i_size_read(in->f_mapping->host);
2918 if (unlikely(*ppos >= isize))
2920 end_offset = min_t(loff_t, isize, *ppos + len);
2923 * Once we start copying data, we don't want to be touching any
2924 * cachelines that might be contended:
2926 writably_mapped = mapping_writably_mapped(in->f_mapping);
2928 for (i = 0; i < folio_batch_count(&fbatch); i++) {
2929 struct folio *folio = fbatch.folios[i];
2932 if (folio_pos(folio) >= end_offset)
2934 folio_mark_accessed(folio);
2937 * If users can be writing to this folio using arbitrary
2938 * virtual addresses, take care of potential aliasing
2939 * before reading the folio on the kernel side.
2941 if (writably_mapped)
2942 flush_dcache_folio(folio);
2944 n = min_t(loff_t, len, isize - *ppos);
2945 n = splice_folio_into_pipe(pipe, folio, *ppos, n);
2951 in->f_ra.prev_pos = *ppos;
2952 if (pipe_full(pipe->head, pipe->tail, pipe->max_usage))
2956 folio_batch_release(&fbatch);
2960 folio_batch_release(&fbatch);
2963 return total_spliced ? total_spliced : error;
2965 EXPORT_SYMBOL(filemap_splice_read);
2967 static inline loff_t folio_seek_hole_data(struct xa_state *xas,
2968 struct address_space *mapping, struct folio *folio,
2969 loff_t start, loff_t end, bool seek_data)
2971 const struct address_space_operations *ops = mapping->a_ops;
2972 size_t offset, bsz = i_blocksize(mapping->host);
2974 if (xa_is_value(folio) || folio_test_uptodate(folio))
2975 return seek_data ? start : end;
2976 if (!ops->is_partially_uptodate)
2977 return seek_data ? end : start;
2982 if (unlikely(folio->mapping != mapping))
2985 offset = offset_in_folio(folio, start) & ~(bsz - 1);
2988 if (ops->is_partially_uptodate(folio, offset, bsz) ==
2991 start = (start + bsz) & ~(bsz - 1);
2993 } while (offset < folio_size(folio));
2995 folio_unlock(folio);
3000 static inline size_t seek_folio_size(struct xa_state *xas, struct folio *folio)
3002 if (xa_is_value(folio))
3003 return PAGE_SIZE << xa_get_order(xas->xa, xas->xa_index);
3004 return folio_size(folio);
3008 * mapping_seek_hole_data - Seek for SEEK_DATA / SEEK_HOLE in the page cache.
3009 * @mapping: Address space to search.
3010 * @start: First byte to consider.
3011 * @end: Limit of search (exclusive).
3012 * @whence: Either SEEK_HOLE or SEEK_DATA.
3014 * If the page cache knows which blocks contain holes and which blocks
3015 * contain data, your filesystem can use this function to implement
3016 * SEEK_HOLE and SEEK_DATA. This is useful for filesystems which are
3017 * entirely memory-based such as tmpfs, and filesystems which support
3018 * unwritten extents.
3020 * Return: The requested offset on success, or -ENXIO if @whence specifies
3021 * SEEK_DATA and there is no data after @start. There is an implicit hole
3022 * after @end - 1, so SEEK_HOLE returns @end if all the bytes between @start
3023 * and @end contain data.
3025 loff_t mapping_seek_hole_data(struct address_space *mapping, loff_t start,
3026 loff_t end, int whence)
3028 XA_STATE(xas, &mapping->i_pages, start >> PAGE_SHIFT);
3029 pgoff_t max = (end - 1) >> PAGE_SHIFT;
3030 bool seek_data = (whence == SEEK_DATA);
3031 struct folio *folio;
3037 while ((folio = find_get_entry(&xas, max, XA_PRESENT))) {
3038 loff_t pos = (u64)xas.xa_index << PAGE_SHIFT;
3047 seek_size = seek_folio_size(&xas, folio);
3048 pos = round_up((u64)pos + 1, seek_size);
3049 start = folio_seek_hole_data(&xas, mapping, folio, start, pos,
3055 if (seek_size > PAGE_SIZE)
3056 xas_set(&xas, pos >> PAGE_SHIFT);
3057 if (!xa_is_value(folio))
3064 if (folio && !xa_is_value(folio))
3072 #define MMAP_LOTSAMISS (100)
3074 * lock_folio_maybe_drop_mmap - lock the page, possibly dropping the mmap_lock
3075 * @vmf - the vm_fault for this fault.
3076 * @folio - the folio to lock.
3077 * @fpin - the pointer to the file we may pin (or is already pinned).
3079 * This works similar to lock_folio_or_retry in that it can drop the
3080 * mmap_lock. It differs in that it actually returns the folio locked
3081 * if it returns 1 and 0 if it couldn't lock the folio. If we did have
3082 * to drop the mmap_lock then fpin will point to the pinned file and
3083 * needs to be fput()'ed at a later point.
3085 static int lock_folio_maybe_drop_mmap(struct vm_fault *vmf, struct folio *folio,
3088 if (folio_trylock(folio))
3092 * NOTE! This will make us return with VM_FAULT_RETRY, but with
3093 * the fault lock still held. That's how FAULT_FLAG_RETRY_NOWAIT
3094 * is supposed to work. We have way too many special cases..
3096 if (vmf->flags & FAULT_FLAG_RETRY_NOWAIT)
3099 *fpin = maybe_unlock_mmap_for_io(vmf, *fpin);
3100 if (vmf->flags & FAULT_FLAG_KILLABLE) {
3101 if (__folio_lock_killable(folio)) {
3103 * We didn't have the right flags to drop the
3104 * fault lock, but all fault_handlers only check
3105 * for fatal signals if we return VM_FAULT_RETRY,
3106 * so we need to drop the fault lock here and
3107 * return 0 if we don't have a fpin.
3110 release_fault_lock(vmf);
3114 __folio_lock(folio);
3120 * Synchronous readahead happens when we don't even find a page in the page
3121 * cache at all. We don't want to perform IO under the mmap sem, so if we have
3122 * to drop the mmap sem we return the file that was pinned in order for us to do
3123 * that. If we didn't pin a file then we return NULL. The file that is
3124 * returned needs to be fput()'ed when we're done with it.
3126 static struct file *do_sync_mmap_readahead(struct vm_fault *vmf)
3128 struct file *file = vmf->vma->vm_file;
3129 struct file_ra_state *ra = &file->f_ra;
3130 struct address_space *mapping = file->f_mapping;
3131 DEFINE_READAHEAD(ractl, file, ra, mapping, vmf->pgoff);
3132 struct file *fpin = NULL;
3133 unsigned long vm_flags = vmf->vma->vm_flags;
3134 unsigned int mmap_miss;
3136 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
3137 /* Use the readahead code, even if readahead is disabled */
3138 if ((vm_flags & VM_HUGEPAGE) && HPAGE_PMD_ORDER <= MAX_PAGECACHE_ORDER) {
3139 fpin = maybe_unlock_mmap_for_io(vmf, fpin);
3140 ractl._index &= ~((unsigned long)HPAGE_PMD_NR - 1);
3141 ra->size = HPAGE_PMD_NR;
3143 * Fetch two PMD folios, so we get the chance to actually
3144 * readahead, unless we've been told not to.
3146 if (!(vm_flags & VM_RAND_READ))
3148 ra->async_size = HPAGE_PMD_NR;
3149 page_cache_ra_order(&ractl, ra, HPAGE_PMD_ORDER);
3154 /* If we don't want any read-ahead, don't bother */
3155 if (vm_flags & VM_RAND_READ)
3160 if (vm_flags & VM_SEQ_READ) {
3161 fpin = maybe_unlock_mmap_for_io(vmf, fpin);
3162 page_cache_sync_ra(&ractl, ra->ra_pages);
3166 /* Avoid banging the cache line if not needed */
3167 mmap_miss = READ_ONCE(ra->mmap_miss);
3168 if (mmap_miss < MMAP_LOTSAMISS * 10)
3169 WRITE_ONCE(ra->mmap_miss, ++mmap_miss);
3172 * Do we miss much more than hit in this file? If so,
3173 * stop bothering with read-ahead. It will only hurt.
3175 if (mmap_miss > MMAP_LOTSAMISS)
3181 fpin = maybe_unlock_mmap_for_io(vmf, fpin);
3182 ra->start = max_t(long, 0, vmf->pgoff - ra->ra_pages / 2);
3183 ra->size = ra->ra_pages;
3184 ra->async_size = ra->ra_pages / 4;
3185 ractl._index = ra->start;
3186 page_cache_ra_order(&ractl, ra, 0);
3191 * Asynchronous readahead happens when we find the page and PG_readahead,
3192 * so we want to possibly extend the readahead further. We return the file that
3193 * was pinned if we have to drop the mmap_lock in order to do IO.
3195 static struct file *do_async_mmap_readahead(struct vm_fault *vmf,
3196 struct folio *folio)
3198 struct file *file = vmf->vma->vm_file;
3199 struct file_ra_state *ra = &file->f_ra;
3200 DEFINE_READAHEAD(ractl, file, ra, file->f_mapping, vmf->pgoff);
3201 struct file *fpin = NULL;
3202 unsigned int mmap_miss;
3204 /* If we don't want any read-ahead, don't bother */
3205 if (vmf->vma->vm_flags & VM_RAND_READ || !ra->ra_pages)
3208 mmap_miss = READ_ONCE(ra->mmap_miss);
3210 WRITE_ONCE(ra->mmap_miss, --mmap_miss);
3212 if (folio_test_readahead(folio)) {
3213 fpin = maybe_unlock_mmap_for_io(vmf, fpin);
3214 page_cache_async_ra(&ractl, folio, ra->ra_pages);
3219 static vm_fault_t filemap_fault_recheck_pte_none(struct vm_fault *vmf)
3221 struct vm_area_struct *vma = vmf->vma;
3226 * We might have COW'ed a pagecache folio and might now have an mlocked
3227 * anon folio mapped. The original pagecache folio is not mlocked and
3228 * might have been evicted. During a read+clear/modify/write update of
3229 * the PTE, such as done in do_numa_page()/change_pte_range(), we
3230 * temporarily clear the PTE under PT lock and might detect it here as
3231 * "none" when not holding the PT lock.
3233 * Not rechecking the PTE under PT lock could result in an unexpected
3234 * major fault in an mlock'ed region. Recheck only for this special
3235 * scenario while holding the PT lock, to not degrade non-mlocked
3236 * scenarios. Recheck the PTE without PT lock firstly, thereby reducing
3237 * the number of times we hold PT lock.
3239 if (!(vma->vm_flags & VM_LOCKED))
3242 if (!(vmf->flags & FAULT_FLAG_ORIG_PTE_VALID))
3245 ptep = pte_offset_map_nolock(vma->vm_mm, vmf->pmd, vmf->address,
3247 if (unlikely(!ptep))
3248 return VM_FAULT_NOPAGE;
3250 if (unlikely(!pte_none(ptep_get_lockless(ptep)))) {
3251 ret = VM_FAULT_NOPAGE;
3253 spin_lock(vmf->ptl);
3254 if (unlikely(!pte_none(ptep_get(ptep))))
3255 ret = VM_FAULT_NOPAGE;
3256 spin_unlock(vmf->ptl);
3263 * filemap_fault - read in file data for page fault handling
3264 * @vmf: struct vm_fault containing details of the fault
3266 * filemap_fault() is invoked via the vma operations vector for a
3267 * mapped memory region to read in file data during a page fault.
3269 * The goto's are kind of ugly, but this streamlines the normal case of having
3270 * it in the page cache, and handles the special cases reasonably without
3271 * having a lot of duplicated code.
3273 * vma->vm_mm->mmap_lock must be held on entry.
3275 * If our return value has VM_FAULT_RETRY set, it's because the mmap_lock
3276 * may be dropped before doing I/O or by lock_folio_maybe_drop_mmap().
3278 * If our return value does not have VM_FAULT_RETRY set, the mmap_lock
3279 * has not been released.
3281 * We never return with VM_FAULT_RETRY and a bit from VM_FAULT_ERROR set.
3283 * Return: bitwise-OR of %VM_FAULT_ codes.
3285 vm_fault_t filemap_fault(struct vm_fault *vmf)
3288 struct file *file = vmf->vma->vm_file;
3289 struct file *fpin = NULL;
3290 struct address_space *mapping = file->f_mapping;
3291 struct inode *inode = mapping->host;
3292 pgoff_t max_idx, index = vmf->pgoff;
3293 struct folio *folio;
3295 bool mapping_locked = false;
3297 max_idx = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
3298 if (unlikely(index >= max_idx))
3299 return VM_FAULT_SIGBUS;
3302 * Do we have something in the page cache already?
3304 folio = filemap_get_folio(mapping, index);
3305 if (likely(!IS_ERR(folio))) {
3307 * We found the page, so try async readahead before waiting for
3310 if (!(vmf->flags & FAULT_FLAG_TRIED))
3311 fpin = do_async_mmap_readahead(vmf, folio);
3312 if (unlikely(!folio_test_uptodate(folio))) {
3313 filemap_invalidate_lock_shared(mapping);
3314 mapping_locked = true;
3317 ret = filemap_fault_recheck_pte_none(vmf);
3321 /* No page in the page cache at all */
3322 count_vm_event(PGMAJFAULT);
3323 count_memcg_event_mm(vmf->vma->vm_mm, PGMAJFAULT);
3324 ret = VM_FAULT_MAJOR;
3325 fpin = do_sync_mmap_readahead(vmf);
3328 * See comment in filemap_create_folio() why we need
3331 if (!mapping_locked) {
3332 filemap_invalidate_lock_shared(mapping);
3333 mapping_locked = true;
3335 folio = __filemap_get_folio(mapping, index,
3336 FGP_CREAT|FGP_FOR_MMAP,
3338 if (IS_ERR(folio)) {
3341 filemap_invalidate_unlock_shared(mapping);
3342 return VM_FAULT_OOM;
3346 if (!lock_folio_maybe_drop_mmap(vmf, folio, &fpin))
3349 /* Did it get truncated? */
3350 if (unlikely(folio->mapping != mapping)) {
3351 folio_unlock(folio);
3355 VM_BUG_ON_FOLIO(!folio_contains(folio, index), folio);
3358 * We have a locked folio in the page cache, now we need to check
3359 * that it's up-to-date. If not, it is going to be due to an error,
3360 * or because readahead was otherwise unable to retrieve it.
3362 if (unlikely(!folio_test_uptodate(folio))) {
3364 * If the invalidate lock is not held, the folio was in cache
3365 * and uptodate and now it is not. Strange but possible since we
3366 * didn't hold the page lock all the time. Let's drop
3367 * everything, get the invalidate lock and try again.
3369 if (!mapping_locked) {
3370 folio_unlock(folio);
3376 * OK, the folio is really not uptodate. This can be because the
3377 * VMA has the VM_RAND_READ flag set, or because an error
3378 * arose. Let's read it in directly.
3380 goto page_not_uptodate;
3384 * We've made it this far and we had to drop our mmap_lock, now is the
3385 * time to return to the upper layer and have it re-find the vma and
3389 folio_unlock(folio);
3393 filemap_invalidate_unlock_shared(mapping);
3396 * Found the page and have a reference on it.
3397 * We must recheck i_size under page lock.
3399 max_idx = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
3400 if (unlikely(index >= max_idx)) {
3401 folio_unlock(folio);
3403 return VM_FAULT_SIGBUS;
3406 vmf->page = folio_file_page(folio, index);
3407 return ret | VM_FAULT_LOCKED;
3411 * Umm, take care of errors if the page isn't up-to-date.
3412 * Try to re-read it _once_. We do this synchronously,
3413 * because there really aren't any performance issues here
3414 * and we need to check for errors.
3416 fpin = maybe_unlock_mmap_for_io(vmf, fpin);
3417 error = filemap_read_folio(file, mapping->a_ops->read_folio, folio);
3422 if (!error || error == AOP_TRUNCATED_PAGE)
3424 filemap_invalidate_unlock_shared(mapping);
3426 return VM_FAULT_SIGBUS;
3430 * We dropped the mmap_lock, we need to return to the fault handler to
3431 * re-find the vma and come back and find our hopefully still populated
3437 filemap_invalidate_unlock_shared(mapping);
3440 return ret | VM_FAULT_RETRY;
3442 EXPORT_SYMBOL(filemap_fault);
3444 static bool filemap_map_pmd(struct vm_fault *vmf, struct folio *folio,
3447 struct mm_struct *mm = vmf->vma->vm_mm;
3449 /* Huge page is mapped? No need to proceed. */
3450 if (pmd_trans_huge(*vmf->pmd)) {
3451 folio_unlock(folio);
3456 if (pmd_none(*vmf->pmd) && folio_test_pmd_mappable(folio)) {
3457 struct page *page = folio_file_page(folio, start);
3458 vm_fault_t ret = do_set_pmd(vmf, page);
3460 /* The page is mapped successfully, reference consumed. */
3461 folio_unlock(folio);
3466 if (pmd_none(*vmf->pmd) && vmf->prealloc_pte)
3467 pmd_install(mm, vmf->pmd, &vmf->prealloc_pte);
3472 static struct folio *next_uptodate_folio(struct xa_state *xas,
3473 struct address_space *mapping, pgoff_t end_pgoff)
3475 struct folio *folio = xas_next_entry(xas, end_pgoff);
3476 unsigned long max_idx;
3481 if (xas_retry(xas, folio))
3483 if (xa_is_value(folio))
3485 if (folio_test_locked(folio))
3487 if (!folio_try_get(folio))
3489 /* Has the page moved or been split? */
3490 if (unlikely(folio != xas_reload(xas)))
3492 if (!folio_test_uptodate(folio) || folio_test_readahead(folio))
3494 if (!folio_trylock(folio))
3496 if (folio->mapping != mapping)
3498 if (!folio_test_uptodate(folio))
3500 max_idx = DIV_ROUND_UP(i_size_read(mapping->host), PAGE_SIZE);
3501 if (xas->xa_index >= max_idx)
3505 folio_unlock(folio);
3508 } while ((folio = xas_next_entry(xas, end_pgoff)) != NULL);
3514 * Map page range [start_page, start_page + nr_pages) of folio.
3515 * start_page is gotten from start by folio_page(folio, start)
3517 static vm_fault_t filemap_map_folio_range(struct vm_fault *vmf,
3518 struct folio *folio, unsigned long start,
3519 unsigned long addr, unsigned int nr_pages,
3520 unsigned long *rss, unsigned int *mmap_miss)
3523 struct page *page = folio_page(folio, start);
3524 unsigned int count = 0;
3525 pte_t *old_ptep = vmf->pte;
3528 if (PageHWPoison(page + count))
3532 * If there are too many folios that are recently evicted
3533 * in a file, they will probably continue to be evicted.
3534 * In such situation, read-ahead is only a waste of IO.
3535 * Don't decrease mmap_miss in this scenario to make sure
3536 * we can stop read-ahead.
3538 if (!folio_test_workingset(folio))
3542 * NOTE: If there're PTE markers, we'll leave them to be
3543 * handled in the specific fault path, and it'll prohibit the
3544 * fault-around logic.
3546 if (!pte_none(ptep_get(&vmf->pte[count])))
3553 set_pte_range(vmf, folio, page, count, addr);
3555 folio_ref_add(folio, count);
3556 if (in_range(vmf->address, addr, count * PAGE_SIZE))
3557 ret = VM_FAULT_NOPAGE;
3563 addr += count * PAGE_SIZE;
3565 } while (--nr_pages > 0);
3568 set_pte_range(vmf, folio, page, count, addr);
3570 folio_ref_add(folio, count);
3571 if (in_range(vmf->address, addr, count * PAGE_SIZE))
3572 ret = VM_FAULT_NOPAGE;
3575 vmf->pte = old_ptep;
3580 static vm_fault_t filemap_map_order0_folio(struct vm_fault *vmf,
3581 struct folio *folio, unsigned long addr,
3582 unsigned long *rss, unsigned int *mmap_miss)
3585 struct page *page = &folio->page;
3587 if (PageHWPoison(page))
3590 /* See comment of filemap_map_folio_range() */
3591 if (!folio_test_workingset(folio))
3595 * NOTE: If there're PTE markers, we'll leave them to be
3596 * handled in the specific fault path, and it'll prohibit
3597 * the fault-around logic.
3599 if (!pte_none(ptep_get(vmf->pte)))
3602 if (vmf->address == addr)
3603 ret = VM_FAULT_NOPAGE;
3605 set_pte_range(vmf, folio, page, 1, addr);
3607 folio_ref_inc(folio);
3612 vm_fault_t filemap_map_pages(struct vm_fault *vmf,
3613 pgoff_t start_pgoff, pgoff_t end_pgoff)
3615 struct vm_area_struct *vma = vmf->vma;
3616 struct file *file = vma->vm_file;
3617 struct address_space *mapping = file->f_mapping;
3618 pgoff_t file_end, last_pgoff = start_pgoff;
3620 XA_STATE(xas, &mapping->i_pages, start_pgoff);
3621 struct folio *folio;
3623 unsigned long rss = 0;
3624 unsigned int nr_pages = 0, mmap_miss = 0, mmap_miss_saved, folio_type;
3627 folio = next_uptodate_folio(&xas, mapping, end_pgoff);
3631 if (filemap_map_pmd(vmf, folio, start_pgoff)) {
3632 ret = VM_FAULT_NOPAGE;
3636 addr = vma->vm_start + ((start_pgoff - vma->vm_pgoff) << PAGE_SHIFT);
3637 vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, addr, &vmf->ptl);
3639 folio_unlock(folio);
3644 file_end = DIV_ROUND_UP(i_size_read(mapping->host), PAGE_SIZE) - 1;
3645 if (end_pgoff > file_end)
3646 end_pgoff = file_end;
3648 folio_type = mm_counter_file(folio);
3652 addr += (xas.xa_index - last_pgoff) << PAGE_SHIFT;
3653 vmf->pte += xas.xa_index - last_pgoff;
3654 last_pgoff = xas.xa_index;
3655 end = folio_next_index(folio) - 1;
3656 nr_pages = min(end, end_pgoff) - xas.xa_index + 1;
3658 if (!folio_test_large(folio))
3659 ret |= filemap_map_order0_folio(vmf,
3660 folio, addr, &rss, &mmap_miss);
3662 ret |= filemap_map_folio_range(vmf, folio,
3663 xas.xa_index - folio->index, addr,
3664 nr_pages, &rss, &mmap_miss);
3666 folio_unlock(folio);
3668 } while ((folio = next_uptodate_folio(&xas, mapping, end_pgoff)) != NULL);
3669 add_mm_counter(vma->vm_mm, folio_type, rss);
3670 pte_unmap_unlock(vmf->pte, vmf->ptl);
3674 mmap_miss_saved = READ_ONCE(file->f_ra.mmap_miss);
3675 if (mmap_miss >= mmap_miss_saved)
3676 WRITE_ONCE(file->f_ra.mmap_miss, 0);
3678 WRITE_ONCE(file->f_ra.mmap_miss, mmap_miss_saved - mmap_miss);
3682 EXPORT_SYMBOL(filemap_map_pages);
3684 vm_fault_t filemap_page_mkwrite(struct vm_fault *vmf)
3686 struct address_space *mapping = vmf->vma->vm_file->f_mapping;
3687 struct folio *folio = page_folio(vmf->page);
3688 vm_fault_t ret = VM_FAULT_LOCKED;
3690 sb_start_pagefault(mapping->host->i_sb);
3691 file_update_time(vmf->vma->vm_file);
3693 if (folio->mapping != mapping) {
3694 folio_unlock(folio);
3695 ret = VM_FAULT_NOPAGE;
3699 * We mark the folio dirty already here so that when freeze is in
3700 * progress, we are guaranteed that writeback during freezing will
3701 * see the dirty folio and writeprotect it again.
3703 folio_mark_dirty(folio);
3704 folio_wait_stable(folio);
3706 sb_end_pagefault(mapping->host->i_sb);
3710 const struct vm_operations_struct generic_file_vm_ops = {
3711 .fault = filemap_fault,
3712 .map_pages = filemap_map_pages,
3713 .page_mkwrite = filemap_page_mkwrite,
3716 /* This is used for a general mmap of a disk file */
3718 int generic_file_mmap(struct file *file, struct vm_area_struct *vma)
3720 struct address_space *mapping = file->f_mapping;
3722 if (!mapping->a_ops->read_folio)
3724 file_accessed(file);
3725 vma->vm_ops = &generic_file_vm_ops;
3730 * This is for filesystems which do not implement ->writepage.
3732 int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
3734 if (vma_is_shared_maywrite(vma))
3736 return generic_file_mmap(file, vma);
3739 vm_fault_t filemap_page_mkwrite(struct vm_fault *vmf)
3741 return VM_FAULT_SIGBUS;
3743 int generic_file_mmap(struct file *file, struct vm_area_struct *vma)
3747 int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
3751 #endif /* CONFIG_MMU */
3753 EXPORT_SYMBOL(filemap_page_mkwrite);
3754 EXPORT_SYMBOL(generic_file_mmap);
3755 EXPORT_SYMBOL(generic_file_readonly_mmap);
3757 static struct folio *do_read_cache_folio(struct address_space *mapping,
3758 pgoff_t index, filler_t filler, struct file *file, gfp_t gfp)
3760 struct folio *folio;
3764 filler = mapping->a_ops->read_folio;
3766 folio = filemap_get_folio(mapping, index);
3767 if (IS_ERR(folio)) {
3768 folio = filemap_alloc_folio(gfp,
3769 mapping_min_folio_order(mapping));
3771 return ERR_PTR(-ENOMEM);
3772 index = mapping_align_index(mapping, index);
3773 err = filemap_add_folio(mapping, folio, index, gfp);
3774 if (unlikely(err)) {
3778 /* Presumably ENOMEM for xarray node */
3779 return ERR_PTR(err);
3784 if (folio_test_uptodate(folio))
3787 if (!folio_trylock(folio)) {
3788 folio_put_wait_locked(folio, TASK_UNINTERRUPTIBLE);
3792 /* Folio was truncated from mapping */
3793 if (!folio->mapping) {
3794 folio_unlock(folio);
3799 /* Someone else locked and filled the page in a very small window */
3800 if (folio_test_uptodate(folio)) {
3801 folio_unlock(folio);
3806 err = filemap_read_folio(file, filler, folio);
3809 if (err == AOP_TRUNCATED_PAGE)
3811 return ERR_PTR(err);
3815 folio_mark_accessed(folio);
3820 * read_cache_folio - Read into page cache, fill it if needed.
3821 * @mapping: The address_space to read from.
3822 * @index: The index to read.
3823 * @filler: Function to perform the read, or NULL to use aops->read_folio().
3824 * @file: Passed to filler function, may be NULL if not required.
3826 * Read one page into the page cache. If it succeeds, the folio returned
3827 * will contain @index, but it may not be the first page of the folio.
3829 * If the filler function returns an error, it will be returned to the
3832 * Context: May sleep. Expects mapping->invalidate_lock to be held.
3833 * Return: An uptodate folio on success, ERR_PTR() on failure.
3835 struct folio *read_cache_folio(struct address_space *mapping, pgoff_t index,
3836 filler_t filler, struct file *file)
3838 return do_read_cache_folio(mapping, index, filler, file,
3839 mapping_gfp_mask(mapping));
3841 EXPORT_SYMBOL(read_cache_folio);
3844 * mapping_read_folio_gfp - Read into page cache, using specified allocation flags.
3845 * @mapping: The address_space for the folio.
3846 * @index: The index that the allocated folio will contain.
3847 * @gfp: The page allocator flags to use if allocating.
3849 * This is the same as "read_cache_folio(mapping, index, NULL, NULL)", but with
3850 * any new memory allocations done using the specified allocation flags.
3852 * The most likely error from this function is EIO, but ENOMEM is
3853 * possible and so is EINTR. If ->read_folio returns another error,
3854 * that will be returned to the caller.
3856 * The function expects mapping->invalidate_lock to be already held.
3858 * Return: Uptodate folio on success, ERR_PTR() on failure.
3860 struct folio *mapping_read_folio_gfp(struct address_space *mapping,
3861 pgoff_t index, gfp_t gfp)
3863 return do_read_cache_folio(mapping, index, NULL, NULL, gfp);
3865 EXPORT_SYMBOL(mapping_read_folio_gfp);
3867 static struct page *do_read_cache_page(struct address_space *mapping,
3868 pgoff_t index, filler_t *filler, struct file *file, gfp_t gfp)
3870 struct folio *folio;
3872 folio = do_read_cache_folio(mapping, index, filler, file, gfp);
3874 return &folio->page;
3875 return folio_file_page(folio, index);
3878 struct page *read_cache_page(struct address_space *mapping,
3879 pgoff_t index, filler_t *filler, struct file *file)
3881 return do_read_cache_page(mapping, index, filler, file,
3882 mapping_gfp_mask(mapping));
3884 EXPORT_SYMBOL(read_cache_page);
3887 * read_cache_page_gfp - read into page cache, using specified page allocation flags.
3888 * @mapping: the page's address_space
3889 * @index: the page index
3890 * @gfp: the page allocator flags to use if allocating
3892 * This is the same as "read_mapping_page(mapping, index, NULL)", but with
3893 * any new page allocations done using the specified allocation flags.
3895 * If the page does not get brought uptodate, return -EIO.
3897 * The function expects mapping->invalidate_lock to be already held.
3899 * Return: up to date page on success, ERR_PTR() on failure.
3901 struct page *read_cache_page_gfp(struct address_space *mapping,
3905 return do_read_cache_page(mapping, index, NULL, NULL, gfp);
3907 EXPORT_SYMBOL(read_cache_page_gfp);
3910 * Warn about a page cache invalidation failure during a direct I/O write.
3912 static void dio_warn_stale_pagecache(struct file *filp)
3914 static DEFINE_RATELIMIT_STATE(_rs, 86400 * HZ, DEFAULT_RATELIMIT_BURST);
3918 errseq_set(&filp->f_mapping->wb_err, -EIO);
3919 if (__ratelimit(&_rs)) {
3920 path = file_path(filp, pathname, sizeof(pathname));
3923 pr_crit("Page cache invalidation failure on direct I/O. Possible data corruption due to collision with buffered I/O!\n");
3924 pr_crit("File: %s PID: %d Comm: %.20s\n", path, current->pid,
3929 void kiocb_invalidate_post_direct_write(struct kiocb *iocb, size_t count)
3931 struct address_space *mapping = iocb->ki_filp->f_mapping;
3933 if (mapping->nrpages &&
3934 invalidate_inode_pages2_range(mapping,
3935 iocb->ki_pos >> PAGE_SHIFT,
3936 (iocb->ki_pos + count - 1) >> PAGE_SHIFT))
3937 dio_warn_stale_pagecache(iocb->ki_filp);
3941 generic_file_direct_write(struct kiocb *iocb, struct iov_iter *from)
3943 struct address_space *mapping = iocb->ki_filp->f_mapping;
3944 size_t write_len = iov_iter_count(from);
3948 * If a page can not be invalidated, return 0 to fall back
3949 * to buffered write.
3951 written = kiocb_invalidate_pages(iocb, write_len);
3953 if (written == -EBUSY)
3958 written = mapping->a_ops->direct_IO(iocb, from);
3961 * Finally, try again to invalidate clean pages which might have been
3962 * cached by non-direct readahead, or faulted in by get_user_pages()
3963 * if the source of the write was an mmap'ed region of the file
3964 * we're writing. Either one is a pretty crazy thing to do,
3965 * so we don't support it 100%. If this invalidation
3966 * fails, tough, the write still worked...
3968 * Most of the time we do not need this since dio_complete() will do
3969 * the invalidation for us. However there are some file systems that
3970 * do not end up with dio_complete() being called, so let's not break
3971 * them by removing it completely.
3973 * Noticeable example is a blkdev_direct_IO().
3975 * Skip invalidation for async writes or if mapping has no pages.
3978 struct inode *inode = mapping->host;
3979 loff_t pos = iocb->ki_pos;
3981 kiocb_invalidate_post_direct_write(iocb, written);
3983 write_len -= written;
3984 if (pos > i_size_read(inode) && !S_ISBLK(inode->i_mode)) {
3985 i_size_write(inode, pos);
3986 mark_inode_dirty(inode);
3990 if (written != -EIOCBQUEUED)
3991 iov_iter_revert(from, write_len - iov_iter_count(from));
3994 EXPORT_SYMBOL(generic_file_direct_write);
3996 ssize_t generic_perform_write(struct kiocb *iocb, struct iov_iter *i)
3998 struct file *file = iocb->ki_filp;
3999 loff_t pos = iocb->ki_pos;
4000 struct address_space *mapping = file->f_mapping;
4001 const struct address_space_operations *a_ops = mapping->a_ops;
4002 size_t chunk = mapping_max_folio_size(mapping);
4004 ssize_t written = 0;
4007 struct folio *folio;
4008 size_t offset; /* Offset into folio */
4009 size_t bytes; /* Bytes to write to folio */
4010 size_t copied; /* Bytes copied from user */
4011 void *fsdata = NULL;
4013 bytes = iov_iter_count(i);
4015 offset = pos & (chunk - 1);
4016 bytes = min(chunk - offset, bytes);
4017 balance_dirty_pages_ratelimited(mapping);
4020 * Bring in the user page that we will copy from _first_.
4021 * Otherwise there's a nasty deadlock on copying from the
4022 * same page as we're writing to, without it being marked
4025 if (unlikely(fault_in_iov_iter_readable(i, bytes) == bytes)) {
4030 if (fatal_signal_pending(current)) {
4035 status = a_ops->write_begin(file, mapping, pos, bytes,
4037 if (unlikely(status < 0))
4040 offset = offset_in_folio(folio, pos);
4041 if (bytes > folio_size(folio) - offset)
4042 bytes = folio_size(folio) - offset;
4044 if (mapping_writably_mapped(mapping))
4045 flush_dcache_folio(folio);
4047 copied = copy_folio_from_iter_atomic(folio, offset, bytes, i);
4048 flush_dcache_folio(folio);
4050 status = a_ops->write_end(file, mapping, pos, bytes, copied,
4052 if (unlikely(status != copied)) {
4053 iov_iter_revert(i, copied - max(status, 0L));
4054 if (unlikely(status < 0))
4059 if (unlikely(status == 0)) {
4061 * A short copy made ->write_end() reject the
4062 * thing entirely. Might be memory poisoning
4063 * halfway through, might be a race with munmap,
4064 * might be severe memory pressure.
4066 if (chunk > PAGE_SIZE)
4076 } while (iov_iter_count(i));
4080 iocb->ki_pos += written;
4083 EXPORT_SYMBOL(generic_perform_write);
4086 * __generic_file_write_iter - write data to a file
4087 * @iocb: IO state structure (file, offset, etc.)
4088 * @from: iov_iter with data to write
4090 * This function does all the work needed for actually writing data to a
4091 * file. It does all basic checks, removes SUID from the file, updates
4092 * modification times and calls proper subroutines depending on whether we
4093 * do direct IO or a standard buffered write.
4095 * It expects i_rwsem to be grabbed unless we work on a block device or similar
4096 * object which does not need locking at all.
4098 * This function does *not* take care of syncing data in case of O_SYNC write.
4099 * A caller has to handle it. This is mainly due to the fact that we want to
4100 * avoid syncing under i_rwsem.
4103 * * number of bytes written, even for truncated writes
4104 * * negative error code if no data has been written at all
4106 ssize_t __generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
4108 struct file *file = iocb->ki_filp;
4109 struct address_space *mapping = file->f_mapping;
4110 struct inode *inode = mapping->host;
4113 ret = file_remove_privs(file);
4117 ret = file_update_time(file);
4121 if (iocb->ki_flags & IOCB_DIRECT) {
4122 ret = generic_file_direct_write(iocb, from);
4124 * If the write stopped short of completing, fall back to
4125 * buffered writes. Some filesystems do this for writes to
4126 * holes, for example. For DAX files, a buffered write will
4127 * not succeed (even if it did, DAX does not handle dirty
4128 * page-cache pages correctly).
4130 if (ret < 0 || !iov_iter_count(from) || IS_DAX(inode))
4132 return direct_write_fallback(iocb, from, ret,
4133 generic_perform_write(iocb, from));
4136 return generic_perform_write(iocb, from);
4138 EXPORT_SYMBOL(__generic_file_write_iter);
4141 * generic_file_write_iter - write data to a file
4142 * @iocb: IO state structure
4143 * @from: iov_iter with data to write
4145 * This is a wrapper around __generic_file_write_iter() to be used by most
4146 * filesystems. It takes care of syncing the file in case of O_SYNC file
4147 * and acquires i_rwsem as needed.
4149 * * negative error code if no data has been written at all of
4150 * vfs_fsync_range() failed for a synchronous write
4151 * * number of bytes written, even for truncated writes
4153 ssize_t generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
4155 struct file *file = iocb->ki_filp;
4156 struct inode *inode = file->f_mapping->host;
4160 ret = generic_write_checks(iocb, from);
4162 ret = __generic_file_write_iter(iocb, from);
4163 inode_unlock(inode);
4166 ret = generic_write_sync(iocb, ret);
4169 EXPORT_SYMBOL(generic_file_write_iter);
4172 * filemap_release_folio() - Release fs-specific metadata on a folio.
4173 * @folio: The folio which the kernel is trying to free.
4174 * @gfp: Memory allocation flags (and I/O mode).
4176 * The address_space is trying to release any data attached to a folio
4177 * (presumably at folio->private).
4179 * This will also be called if the private_2 flag is set on a page,
4180 * indicating that the folio has other metadata associated with it.
4182 * The @gfp argument specifies whether I/O may be performed to release
4183 * this page (__GFP_IO), and whether the call may block
4184 * (__GFP_RECLAIM & __GFP_FS).
4186 * Return: %true if the release was successful, otherwise %false.
4188 bool filemap_release_folio(struct folio *folio, gfp_t gfp)
4190 struct address_space * const mapping = folio->mapping;
4192 BUG_ON(!folio_test_locked(folio));
4193 if (!folio_needs_release(folio))
4195 if (folio_test_writeback(folio))
4198 if (mapping && mapping->a_ops->release_folio)
4199 return mapping->a_ops->release_folio(folio, gfp);
4200 return try_to_free_buffers(folio);
4202 EXPORT_SYMBOL(filemap_release_folio);
4205 * filemap_invalidate_inode - Invalidate/forcibly write back a range of an inode's pagecache
4206 * @inode: The inode to flush
4207 * @flush: Set to write back rather than simply invalidate.
4208 * @start: First byte to in range.
4209 * @end: Last byte in range (inclusive), or LLONG_MAX for everything from start
4212 * Invalidate all the folios on an inode that contribute to the specified
4213 * range, possibly writing them back first. Whilst the operation is
4214 * undertaken, the invalidate lock is held to prevent new folios from being
4217 int filemap_invalidate_inode(struct inode *inode, bool flush,
4218 loff_t start, loff_t end)
4220 struct address_space *mapping = inode->i_mapping;
4221 pgoff_t first = start >> PAGE_SHIFT;
4222 pgoff_t last = end >> PAGE_SHIFT;
4223 pgoff_t nr = end == LLONG_MAX ? ULONG_MAX : last - first + 1;
4225 if (!mapping || !mapping->nrpages || end < start)
4228 /* Prevent new folios from being added to the inode. */
4229 filemap_invalidate_lock(mapping);
4231 if (!mapping->nrpages)
4234 unmap_mapping_pages(mapping, first, nr, false);
4236 /* Write back the data if we're asked to. */
4238 struct writeback_control wbc = {
4239 .sync_mode = WB_SYNC_ALL,
4240 .nr_to_write = LONG_MAX,
4241 .range_start = start,
4245 filemap_fdatawrite_wbc(mapping, &wbc);
4248 /* Wait for writeback to complete on all folios and discard. */
4249 invalidate_inode_pages2_range(mapping, start / PAGE_SIZE, end / PAGE_SIZE);
4252 filemap_invalidate_unlock(mapping);
4254 return filemap_check_errors(mapping);
4256 EXPORT_SYMBOL_GPL(filemap_invalidate_inode);
4258 #ifdef CONFIG_CACHESTAT_SYSCALL
4260 * filemap_cachestat() - compute the page cache statistics of a mapping
4261 * @mapping: The mapping to compute the statistics for.
4262 * @first_index: The starting page cache index.
4263 * @last_index: The final page index (inclusive).
4264 * @cs: the cachestat struct to write the result to.
4266 * This will query the page cache statistics of a mapping in the
4267 * page range of [first_index, last_index] (inclusive). The statistics
4268 * queried include: number of dirty pages, number of pages marked for
4269 * writeback, and the number of (recently) evicted pages.
4271 static void filemap_cachestat(struct address_space *mapping,
4272 pgoff_t first_index, pgoff_t last_index, struct cachestat *cs)
4274 XA_STATE(xas, &mapping->i_pages, first_index);
4275 struct folio *folio;
4277 /* Flush stats (and potentially sleep) outside the RCU read section. */
4278 mem_cgroup_flush_stats_ratelimited(NULL);
4281 xas_for_each(&xas, folio, last_index) {
4283 unsigned long nr_pages;
4284 pgoff_t folio_first_index, folio_last_index;
4287 * Don't deref the folio. It is not pinned, and might
4288 * get freed (and reused) underneath us.
4290 * We *could* pin it, but that would be expensive for
4291 * what should be a fast and lightweight syscall.
4293 * Instead, derive all information of interest from
4294 * the rcu-protected xarray.
4297 if (xas_retry(&xas, folio))
4300 order = xa_get_order(xas.xa, xas.xa_index);
4301 nr_pages = 1 << order;
4302 folio_first_index = round_down(xas.xa_index, 1 << order);
4303 folio_last_index = folio_first_index + nr_pages - 1;
4305 /* Folios might straddle the range boundaries, only count covered pages */
4306 if (folio_first_index < first_index)
4307 nr_pages -= first_index - folio_first_index;
4309 if (folio_last_index > last_index)
4310 nr_pages -= folio_last_index - last_index;
4312 if (xa_is_value(folio)) {
4313 /* page is evicted */
4314 void *shadow = (void *)folio;
4315 bool workingset; /* not used */
4317 cs->nr_evicted += nr_pages;
4319 #ifdef CONFIG_SWAP /* implies CONFIG_MMU */
4320 if (shmem_mapping(mapping)) {
4321 /* shmem file - in swap cache */
4322 swp_entry_t swp = radix_to_swp_entry(folio);
4324 /* swapin error results in poisoned entry */
4325 if (non_swap_entry(swp))
4329 * Getting a swap entry from the shmem
4330 * inode means we beat
4331 * shmem_unuse(). rcu_read_lock()
4332 * ensures swapoff waits for us before
4333 * freeing the swapper space. However,
4334 * we can race with swapping and
4335 * invalidation, so there might not be
4336 * a shadow in the swapcache (yet).
4338 shadow = get_shadow_from_swap_cache(swp);
4343 if (workingset_test_recent(shadow, true, &workingset, false))
4344 cs->nr_recently_evicted += nr_pages;
4349 /* page is in cache */
4350 cs->nr_cache += nr_pages;
4352 if (xas_get_mark(&xas, PAGECACHE_TAG_DIRTY))
4353 cs->nr_dirty += nr_pages;
4355 if (xas_get_mark(&xas, PAGECACHE_TAG_WRITEBACK))
4356 cs->nr_writeback += nr_pages;
4359 if (need_resched()) {
4368 * The cachestat(2) system call.
4370 * cachestat() returns the page cache statistics of a file in the
4371 * bytes range specified by `off` and `len`: number of cached pages,
4372 * number of dirty pages, number of pages marked for writeback,
4373 * number of evicted pages, and number of recently evicted pages.
4375 * An evicted page is a page that is previously in the page cache
4376 * but has been evicted since. A page is recently evicted if its last
4377 * eviction was recent enough that its reentry to the cache would
4378 * indicate that it is actively being used by the system, and that
4379 * there is memory pressure on the system.
4381 * `off` and `len` must be non-negative integers. If `len` > 0,
4382 * the queried range is [`off`, `off` + `len`]. If `len` == 0,
4383 * we will query in the range from `off` to the end of the file.
4385 * The `flags` argument is unused for now, but is included for future
4386 * extensibility. User should pass 0 (i.e no flag specified).
4388 * Currently, hugetlbfs is not supported.
4390 * Because the status of a page can change after cachestat() checks it
4391 * but before it returns to the application, the returned values may
4392 * contain stale information.
4396 * -EFAULT - cstat or cstat_range points to an illegal address
4397 * -EINVAL - invalid flags
4398 * -EBADF - invalid file descriptor
4399 * -EOPNOTSUPP - file descriptor is of a hugetlbfs file
4401 SYSCALL_DEFINE4(cachestat, unsigned int, fd,
4402 struct cachestat_range __user *, cstat_range,
4403 struct cachestat __user *, cstat, unsigned int, flags)
4405 struct fd f = fdget(fd);
4406 struct address_space *mapping;
4407 struct cachestat_range csr;
4408 struct cachestat cs;
4409 pgoff_t first_index, last_index;
4414 if (copy_from_user(&csr, cstat_range,
4415 sizeof(struct cachestat_range))) {
4420 /* hugetlbfs is not supported */
4421 if (is_file_hugepages(f.file)) {
4431 first_index = csr.off >> PAGE_SHIFT;
4433 csr.len == 0 ? ULONG_MAX : (csr.off + csr.len - 1) >> PAGE_SHIFT;
4434 memset(&cs, 0, sizeof(struct cachestat));
4435 mapping = f.file->f_mapping;
4436 filemap_cachestat(mapping, first_index, last_index, &cs);
4439 if (copy_to_user(cstat, &cs, sizeof(struct cachestat)))
4444 #endif /* CONFIG_CACHESTAT_SYSCALL */