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
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 = page_mapcount(&folio->page);
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 page_mapcount_reset(&folio->page);
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 bool huge = folio_test_hugetlb(folio);
856 bool charged = false;
859 VM_BUG_ON_FOLIO(!folio_test_locked(folio), folio);
860 VM_BUG_ON_FOLIO(folio_test_swapbacked(folio), folio);
861 mapping_set_update(&xas, mapping);
864 int error = mem_cgroup_charge(folio, NULL, gfp);
870 VM_BUG_ON_FOLIO(index & (folio_nr_pages(folio) - 1), folio);
871 xas_set_order(&xas, index, folio_order(folio));
872 nr = folio_nr_pages(folio);
874 gfp &= GFP_RECLAIM_MASK;
875 folio_ref_add(folio, nr);
876 folio->mapping = mapping;
877 folio->index = xas.xa_index;
880 unsigned int order = xa_get_order(xas.xa, xas.xa_index);
881 void *entry, *old = NULL;
883 if (order > folio_order(folio))
884 xas_split_alloc(&xas, xa_load(xas.xa, xas.xa_index),
887 xas_for_each_conflict(&xas, entry) {
889 if (!xa_is_value(entry)) {
890 xas_set_err(&xas, -EEXIST);
898 /* entry may have been split before we acquired lock */
899 order = xa_get_order(xas.xa, xas.xa_index);
900 if (order > folio_order(folio)) {
901 /* How to handle large swap entries? */
902 BUG_ON(shmem_mapping(mapping));
903 xas_split(&xas, old, order);
908 xas_store(&xas, folio);
912 mapping->nrpages += nr;
914 /* hugetlb pages do not participate in page cache accounting */
916 __lruvec_stat_mod_folio(folio, NR_FILE_PAGES, nr);
917 if (folio_test_pmd_mappable(folio))
918 __lruvec_stat_mod_folio(folio,
922 xas_unlock_irq(&xas);
923 } while (xas_nomem(&xas, gfp));
928 trace_mm_filemap_add_to_page_cache(folio);
932 mem_cgroup_uncharge(folio);
933 folio->mapping = NULL;
934 /* Leave page->index set: truncation relies upon it */
935 folio_put_refs(folio, nr);
936 return xas_error(&xas);
938 ALLOW_ERROR_INJECTION(__filemap_add_folio, ERRNO);
940 int filemap_add_folio(struct address_space *mapping, struct folio *folio,
941 pgoff_t index, gfp_t gfp)
946 __folio_set_locked(folio);
947 ret = __filemap_add_folio(mapping, folio, index, gfp, &shadow);
949 __folio_clear_locked(folio);
952 * The folio might have been evicted from cache only
953 * recently, in which case it should be activated like
954 * any other repeatedly accessed folio.
955 * The exception is folios getting rewritten; evicting other
956 * data from the working set, only to cache data that will
957 * get overwritten with something else, is a waste of memory.
959 WARN_ON_ONCE(folio_test_active(folio));
960 if (!(gfp & __GFP_WRITE) && shadow)
961 workingset_refault(folio, shadow);
962 folio_add_lru(folio);
966 EXPORT_SYMBOL_GPL(filemap_add_folio);
969 struct folio *filemap_alloc_folio(gfp_t gfp, unsigned int order)
974 if (cpuset_do_page_mem_spread()) {
975 unsigned int cpuset_mems_cookie;
977 cpuset_mems_cookie = read_mems_allowed_begin();
978 n = cpuset_mem_spread_node();
979 folio = __folio_alloc_node(gfp, order, n);
980 } while (!folio && read_mems_allowed_retry(cpuset_mems_cookie));
984 return folio_alloc(gfp, order);
986 EXPORT_SYMBOL(filemap_alloc_folio);
990 * filemap_invalidate_lock_two - lock invalidate_lock for two mappings
992 * Lock exclusively invalidate_lock of any passed mapping that is not NULL.
994 * @mapping1: the first mapping to lock
995 * @mapping2: the second mapping to lock
997 void filemap_invalidate_lock_two(struct address_space *mapping1,
998 struct address_space *mapping2)
1000 if (mapping1 > mapping2)
1001 swap(mapping1, mapping2);
1003 down_write(&mapping1->invalidate_lock);
1004 if (mapping2 && mapping1 != mapping2)
1005 down_write_nested(&mapping2->invalidate_lock, 1);
1007 EXPORT_SYMBOL(filemap_invalidate_lock_two);
1010 * filemap_invalidate_unlock_two - unlock invalidate_lock for two mappings
1012 * Unlock exclusive invalidate_lock of any passed mapping that is not NULL.
1014 * @mapping1: the first mapping to unlock
1015 * @mapping2: the second mapping to unlock
1017 void filemap_invalidate_unlock_two(struct address_space *mapping1,
1018 struct address_space *mapping2)
1021 up_write(&mapping1->invalidate_lock);
1022 if (mapping2 && mapping1 != mapping2)
1023 up_write(&mapping2->invalidate_lock);
1025 EXPORT_SYMBOL(filemap_invalidate_unlock_two);
1028 * In order to wait for pages to become available there must be
1029 * waitqueues associated with pages. By using a hash table of
1030 * waitqueues where the bucket discipline is to maintain all
1031 * waiters on the same queue and wake all when any of the pages
1032 * become available, and for the woken contexts to check to be
1033 * sure the appropriate page became available, this saves space
1034 * at a cost of "thundering herd" phenomena during rare hash
1037 #define PAGE_WAIT_TABLE_BITS 8
1038 #define PAGE_WAIT_TABLE_SIZE (1 << PAGE_WAIT_TABLE_BITS)
1039 static wait_queue_head_t folio_wait_table[PAGE_WAIT_TABLE_SIZE] __cacheline_aligned;
1041 static wait_queue_head_t *folio_waitqueue(struct folio *folio)
1043 return &folio_wait_table[hash_ptr(folio, PAGE_WAIT_TABLE_BITS)];
1046 void __init pagecache_init(void)
1050 for (i = 0; i < PAGE_WAIT_TABLE_SIZE; i++)
1051 init_waitqueue_head(&folio_wait_table[i]);
1053 page_writeback_init();
1057 * The page wait code treats the "wait->flags" somewhat unusually, because
1058 * we have multiple different kinds of waits, not just the usual "exclusive"
1063 * (a) no special bits set:
1065 * We're just waiting for the bit to be released, and when a waker
1066 * calls the wakeup function, we set WQ_FLAG_WOKEN and wake it up,
1067 * and remove it from the wait queue.
1069 * Simple and straightforward.
1071 * (b) WQ_FLAG_EXCLUSIVE:
1073 * The waiter is waiting to get the lock, and only one waiter should
1074 * be woken up to avoid any thundering herd behavior. We'll set the
1075 * WQ_FLAG_WOKEN bit, wake it up, and remove it from the wait queue.
1077 * This is the traditional exclusive wait.
1079 * (c) WQ_FLAG_EXCLUSIVE | WQ_FLAG_CUSTOM:
1081 * The waiter is waiting to get the bit, and additionally wants the
1082 * lock to be transferred to it for fair lock behavior. If the lock
1083 * cannot be taken, we stop walking the wait queue without waking
1086 * This is the "fair lock handoff" case, and in addition to setting
1087 * WQ_FLAG_WOKEN, we set WQ_FLAG_DONE to let the waiter easily see
1088 * that it now has the lock.
1090 static int wake_page_function(wait_queue_entry_t *wait, unsigned mode, int sync, void *arg)
1093 struct wait_page_key *key = arg;
1094 struct wait_page_queue *wait_page
1095 = container_of(wait, struct wait_page_queue, wait);
1097 if (!wake_page_match(wait_page, key))
1101 * If it's a lock handoff wait, we get the bit for it, and
1102 * stop walking (and do not wake it up) if we can't.
1104 flags = wait->flags;
1105 if (flags & WQ_FLAG_EXCLUSIVE) {
1106 if (test_bit(key->bit_nr, &key->folio->flags))
1108 if (flags & WQ_FLAG_CUSTOM) {
1109 if (test_and_set_bit(key->bit_nr, &key->folio->flags))
1111 flags |= WQ_FLAG_DONE;
1116 * We are holding the wait-queue lock, but the waiter that
1117 * is waiting for this will be checking the flags without
1120 * So update the flags atomically, and wake up the waiter
1121 * afterwards to avoid any races. This store-release pairs
1122 * with the load-acquire in folio_wait_bit_common().
1124 smp_store_release(&wait->flags, flags | WQ_FLAG_WOKEN);
1125 wake_up_state(wait->private, mode);
1128 * Ok, we have successfully done what we're waiting for,
1129 * and we can unconditionally remove the wait entry.
1131 * Note that this pairs with the "finish_wait()" in the
1132 * waiter, and has to be the absolute last thing we do.
1133 * After this list_del_init(&wait->entry) the wait entry
1134 * might be de-allocated and the process might even have
1137 list_del_init_careful(&wait->entry);
1138 return (flags & WQ_FLAG_EXCLUSIVE) != 0;
1141 static void folio_wake_bit(struct folio *folio, int bit_nr)
1143 wait_queue_head_t *q = folio_waitqueue(folio);
1144 struct wait_page_key key;
1145 unsigned long flags;
1148 key.bit_nr = bit_nr;
1151 spin_lock_irqsave(&q->lock, flags);
1152 __wake_up_locked_key(q, TASK_NORMAL, &key);
1155 * It's possible to miss clearing waiters here, when we woke our page
1156 * waiters, but the hashed waitqueue has waiters for other pages on it.
1157 * That's okay, it's a rare case. The next waker will clear it.
1159 * Note that, depending on the page pool (buddy, hugetlb, ZONE_DEVICE,
1160 * other), the flag may be cleared in the course of freeing the page;
1161 * but that is not required for correctness.
1163 if (!waitqueue_active(q) || !key.page_match)
1164 folio_clear_waiters(folio);
1166 spin_unlock_irqrestore(&q->lock, flags);
1170 * A choice of three behaviors for folio_wait_bit_common():
1173 EXCLUSIVE, /* Hold ref to page and take the bit when woken, like
1174 * __folio_lock() waiting on then setting PG_locked.
1176 SHARED, /* Hold ref to page and check the bit when woken, like
1177 * folio_wait_writeback() waiting on PG_writeback.
1179 DROP, /* Drop ref to page before wait, no check when woken,
1180 * like folio_put_wait_locked() on PG_locked.
1185 * Attempt to check (or get) the folio flag, and mark us done
1188 static inline bool folio_trylock_flag(struct folio *folio, int bit_nr,
1189 struct wait_queue_entry *wait)
1191 if (wait->flags & WQ_FLAG_EXCLUSIVE) {
1192 if (test_and_set_bit(bit_nr, &folio->flags))
1194 } else if (test_bit(bit_nr, &folio->flags))
1197 wait->flags |= WQ_FLAG_WOKEN | WQ_FLAG_DONE;
1201 /* How many times do we accept lock stealing from under a waiter? */
1202 int sysctl_page_lock_unfairness = 5;
1204 static inline int folio_wait_bit_common(struct folio *folio, int bit_nr,
1205 int state, enum behavior behavior)
1207 wait_queue_head_t *q = folio_waitqueue(folio);
1208 int unfairness = sysctl_page_lock_unfairness;
1209 struct wait_page_queue wait_page;
1210 wait_queue_entry_t *wait = &wait_page.wait;
1211 bool thrashing = false;
1212 unsigned long pflags;
1215 if (bit_nr == PG_locked &&
1216 !folio_test_uptodate(folio) && folio_test_workingset(folio)) {
1217 delayacct_thrashing_start(&in_thrashing);
1218 psi_memstall_enter(&pflags);
1223 wait->func = wake_page_function;
1224 wait_page.folio = folio;
1225 wait_page.bit_nr = bit_nr;
1229 if (behavior == EXCLUSIVE) {
1230 wait->flags = WQ_FLAG_EXCLUSIVE;
1231 if (--unfairness < 0)
1232 wait->flags |= WQ_FLAG_CUSTOM;
1236 * Do one last check whether we can get the
1237 * page bit synchronously.
1239 * Do the folio_set_waiters() marking before that
1240 * to let any waker we _just_ missed know they
1241 * need to wake us up (otherwise they'll never
1242 * even go to the slow case that looks at the
1243 * page queue), and add ourselves to the wait
1244 * queue if we need to sleep.
1246 * This part needs to be done under the queue
1247 * lock to avoid races.
1249 spin_lock_irq(&q->lock);
1250 folio_set_waiters(folio);
1251 if (!folio_trylock_flag(folio, bit_nr, wait))
1252 __add_wait_queue_entry_tail(q, wait);
1253 spin_unlock_irq(&q->lock);
1256 * From now on, all the logic will be based on
1257 * the WQ_FLAG_WOKEN and WQ_FLAG_DONE flag, to
1258 * see whether the page bit testing has already
1259 * been done by the wake function.
1261 * We can drop our reference to the folio.
1263 if (behavior == DROP)
1267 * Note that until the "finish_wait()", or until
1268 * we see the WQ_FLAG_WOKEN flag, we need to
1269 * be very careful with the 'wait->flags', because
1270 * we may race with a waker that sets them.
1275 set_current_state(state);
1277 /* Loop until we've been woken or interrupted */
1278 flags = smp_load_acquire(&wait->flags);
1279 if (!(flags & WQ_FLAG_WOKEN)) {
1280 if (signal_pending_state(state, current))
1287 /* If we were non-exclusive, we're done */
1288 if (behavior != EXCLUSIVE)
1291 /* If the waker got the lock for us, we're done */
1292 if (flags & WQ_FLAG_DONE)
1296 * Otherwise, if we're getting the lock, we need to
1297 * try to get it ourselves.
1299 * And if that fails, we'll have to retry this all.
1301 if (unlikely(test_and_set_bit(bit_nr, folio_flags(folio, 0))))
1304 wait->flags |= WQ_FLAG_DONE;
1309 * If a signal happened, this 'finish_wait()' may remove the last
1310 * waiter from the wait-queues, but the folio waiters bit will remain
1311 * set. That's ok. The next wakeup will take care of it, and trying
1312 * to do it here would be difficult and prone to races.
1314 finish_wait(q, wait);
1317 delayacct_thrashing_end(&in_thrashing);
1318 psi_memstall_leave(&pflags);
1322 * NOTE! The wait->flags weren't stable until we've done the
1323 * 'finish_wait()', and we could have exited the loop above due
1324 * to a signal, and had a wakeup event happen after the signal
1325 * test but before the 'finish_wait()'.
1327 * So only after the finish_wait() can we reliably determine
1328 * if we got woken up or not, so we can now figure out the final
1329 * return value based on that state without races.
1331 * Also note that WQ_FLAG_WOKEN is sufficient for a non-exclusive
1332 * waiter, but an exclusive one requires WQ_FLAG_DONE.
1334 if (behavior == EXCLUSIVE)
1335 return wait->flags & WQ_FLAG_DONE ? 0 : -EINTR;
1337 return wait->flags & WQ_FLAG_WOKEN ? 0 : -EINTR;
1340 #ifdef CONFIG_MIGRATION
1342 * migration_entry_wait_on_locked - Wait for a migration entry to be removed
1343 * @entry: migration swap entry.
1344 * @ptl: already locked ptl. This function will drop the lock.
1346 * Wait for a migration entry referencing the given page to be removed. This is
1347 * equivalent to put_and_wait_on_page_locked(page, TASK_UNINTERRUPTIBLE) except
1348 * this can be called without taking a reference on the page. Instead this
1349 * should be called while holding the ptl for the migration entry referencing
1352 * Returns after unlocking the ptl.
1354 * This follows the same logic as folio_wait_bit_common() so see the comments
1357 void migration_entry_wait_on_locked(swp_entry_t entry, spinlock_t *ptl)
1360 struct wait_page_queue wait_page;
1361 wait_queue_entry_t *wait = &wait_page.wait;
1362 bool thrashing = false;
1363 unsigned long pflags;
1365 wait_queue_head_t *q;
1366 struct folio *folio = pfn_swap_entry_folio(entry);
1368 q = folio_waitqueue(folio);
1369 if (!folio_test_uptodate(folio) && folio_test_workingset(folio)) {
1370 delayacct_thrashing_start(&in_thrashing);
1371 psi_memstall_enter(&pflags);
1376 wait->func = wake_page_function;
1377 wait_page.folio = folio;
1378 wait_page.bit_nr = PG_locked;
1381 spin_lock_irq(&q->lock);
1382 folio_set_waiters(folio);
1383 if (!folio_trylock_flag(folio, PG_locked, wait))
1384 __add_wait_queue_entry_tail(q, wait);
1385 spin_unlock_irq(&q->lock);
1388 * If a migration entry exists for the page the migration path must hold
1389 * a valid reference to the page, and it must take the ptl to remove the
1390 * migration entry. So the page is valid until the ptl is dropped.
1397 set_current_state(TASK_UNINTERRUPTIBLE);
1399 /* Loop until we've been woken or interrupted */
1400 flags = smp_load_acquire(&wait->flags);
1401 if (!(flags & WQ_FLAG_WOKEN)) {
1402 if (signal_pending_state(TASK_UNINTERRUPTIBLE, current))
1411 finish_wait(q, wait);
1414 delayacct_thrashing_end(&in_thrashing);
1415 psi_memstall_leave(&pflags);
1420 void folio_wait_bit(struct folio *folio, int bit_nr)
1422 folio_wait_bit_common(folio, bit_nr, TASK_UNINTERRUPTIBLE, SHARED);
1424 EXPORT_SYMBOL(folio_wait_bit);
1426 int folio_wait_bit_killable(struct folio *folio, int bit_nr)
1428 return folio_wait_bit_common(folio, bit_nr, TASK_KILLABLE, SHARED);
1430 EXPORT_SYMBOL(folio_wait_bit_killable);
1433 * folio_put_wait_locked - Drop a reference and wait for it to be unlocked
1434 * @folio: The folio to wait for.
1435 * @state: The sleep state (TASK_KILLABLE, TASK_UNINTERRUPTIBLE, etc).
1437 * The caller should hold a reference on @folio. They expect the page to
1438 * become unlocked relatively soon, but do not wish to hold up migration
1439 * (for example) by holding the reference while waiting for the folio to
1440 * come unlocked. After this function returns, the caller should not
1441 * dereference @folio.
1443 * Return: 0 if the folio was unlocked or -EINTR if interrupted by a signal.
1445 static int folio_put_wait_locked(struct folio *folio, int state)
1447 return folio_wait_bit_common(folio, PG_locked, state, DROP);
1451 * folio_add_wait_queue - Add an arbitrary waiter to a folio's wait queue
1452 * @folio: Folio defining the wait queue of interest
1453 * @waiter: Waiter to add to the queue
1455 * Add an arbitrary @waiter to the wait queue for the nominated @folio.
1457 void folio_add_wait_queue(struct folio *folio, wait_queue_entry_t *waiter)
1459 wait_queue_head_t *q = folio_waitqueue(folio);
1460 unsigned long flags;
1462 spin_lock_irqsave(&q->lock, flags);
1463 __add_wait_queue_entry_tail(q, waiter);
1464 folio_set_waiters(folio);
1465 spin_unlock_irqrestore(&q->lock, flags);
1467 EXPORT_SYMBOL_GPL(folio_add_wait_queue);
1470 * folio_unlock - Unlock a locked folio.
1471 * @folio: The folio.
1473 * Unlocks the folio and wakes up any thread sleeping on the page lock.
1475 * Context: May be called from interrupt or process context. May not be
1476 * called from NMI context.
1478 void folio_unlock(struct folio *folio)
1480 /* Bit 7 allows x86 to check the byte's sign bit */
1481 BUILD_BUG_ON(PG_waiters != 7);
1482 BUILD_BUG_ON(PG_locked > 7);
1483 VM_BUG_ON_FOLIO(!folio_test_locked(folio), folio);
1484 if (folio_xor_flags_has_waiters(folio, 1 << PG_locked))
1485 folio_wake_bit(folio, PG_locked);
1487 EXPORT_SYMBOL(folio_unlock);
1490 * folio_end_read - End read on a folio.
1491 * @folio: The folio.
1492 * @success: True if all reads completed successfully.
1494 * When all reads against a folio have completed, filesystems should
1495 * call this function to let the pagecache know that no more reads
1496 * are outstanding. This will unlock the folio and wake up any thread
1497 * sleeping on the lock. The folio will also be marked uptodate if all
1500 * Context: May be called from interrupt or process context. May not be
1501 * called from NMI context.
1503 void folio_end_read(struct folio *folio, bool success)
1505 unsigned long mask = 1 << PG_locked;
1507 /* Must be in bottom byte for x86 to work */
1508 BUILD_BUG_ON(PG_uptodate > 7);
1509 VM_BUG_ON_FOLIO(!folio_test_locked(folio), folio);
1510 VM_BUG_ON_FOLIO(folio_test_uptodate(folio), folio);
1512 if (likely(success))
1513 mask |= 1 << PG_uptodate;
1514 if (folio_xor_flags_has_waiters(folio, mask))
1515 folio_wake_bit(folio, PG_locked);
1517 EXPORT_SYMBOL(folio_end_read);
1520 * folio_end_private_2 - Clear PG_private_2 and wake any waiters.
1521 * @folio: The folio.
1523 * Clear the PG_private_2 bit on a folio and wake up any sleepers waiting for
1524 * it. The folio reference held for PG_private_2 being set is released.
1526 * This is, for example, used when a netfs folio is being written to a local
1527 * disk cache, thereby allowing writes to the cache for the same folio to be
1530 void folio_end_private_2(struct folio *folio)
1532 VM_BUG_ON_FOLIO(!folio_test_private_2(folio), folio);
1533 clear_bit_unlock(PG_private_2, folio_flags(folio, 0));
1534 folio_wake_bit(folio, PG_private_2);
1537 EXPORT_SYMBOL(folio_end_private_2);
1540 * folio_wait_private_2 - Wait for PG_private_2 to be cleared on a folio.
1541 * @folio: The folio to wait on.
1543 * Wait for PG_private_2 (aka PG_fscache) to be cleared on a folio.
1545 void folio_wait_private_2(struct folio *folio)
1547 while (folio_test_private_2(folio))
1548 folio_wait_bit(folio, PG_private_2);
1550 EXPORT_SYMBOL(folio_wait_private_2);
1553 * folio_wait_private_2_killable - Wait for PG_private_2 to be cleared on a folio.
1554 * @folio: The folio to wait on.
1556 * Wait for PG_private_2 (aka PG_fscache) to be cleared on a folio or until a
1557 * fatal signal is received by the calling task.
1560 * - 0 if successful.
1561 * - -EINTR if a fatal signal was encountered.
1563 int folio_wait_private_2_killable(struct folio *folio)
1567 while (folio_test_private_2(folio)) {
1568 ret = folio_wait_bit_killable(folio, PG_private_2);
1575 EXPORT_SYMBOL(folio_wait_private_2_killable);
1578 * folio_end_writeback - End writeback against a folio.
1579 * @folio: The folio.
1581 * The folio must actually be under writeback.
1583 * Context: May be called from process or interrupt context.
1585 void folio_end_writeback(struct folio *folio)
1587 VM_BUG_ON_FOLIO(!folio_test_writeback(folio), folio);
1590 * folio_test_clear_reclaim() could be used here but it is an
1591 * atomic operation and overkill in this particular case. Failing
1592 * to shuffle a folio marked for immediate reclaim is too mild
1593 * a gain to justify taking an atomic operation penalty at the
1594 * end of every folio writeback.
1596 if (folio_test_reclaim(folio)) {
1597 folio_clear_reclaim(folio);
1598 folio_rotate_reclaimable(folio);
1602 * Writeback does not hold a folio reference of its own, relying
1603 * on truncation to wait for the clearing of PG_writeback.
1604 * But here we must make sure that the folio is not freed and
1605 * reused before the folio_wake_bit().
1608 if (__folio_end_writeback(folio))
1609 folio_wake_bit(folio, PG_writeback);
1610 acct_reclaim_writeback(folio);
1613 EXPORT_SYMBOL(folio_end_writeback);
1616 * __folio_lock - Get a lock on the folio, assuming we need to sleep to get it.
1617 * @folio: The folio to lock
1619 void __folio_lock(struct folio *folio)
1621 folio_wait_bit_common(folio, PG_locked, TASK_UNINTERRUPTIBLE,
1624 EXPORT_SYMBOL(__folio_lock);
1626 int __folio_lock_killable(struct folio *folio)
1628 return folio_wait_bit_common(folio, PG_locked, TASK_KILLABLE,
1631 EXPORT_SYMBOL_GPL(__folio_lock_killable);
1633 static int __folio_lock_async(struct folio *folio, struct wait_page_queue *wait)
1635 struct wait_queue_head *q = folio_waitqueue(folio);
1638 wait->folio = folio;
1639 wait->bit_nr = PG_locked;
1641 spin_lock_irq(&q->lock);
1642 __add_wait_queue_entry_tail(q, &wait->wait);
1643 folio_set_waiters(folio);
1644 ret = !folio_trylock(folio);
1646 * If we were successful now, we know we're still on the
1647 * waitqueue as we're still under the lock. This means it's
1648 * safe to remove and return success, we know the callback
1649 * isn't going to trigger.
1652 __remove_wait_queue(q, &wait->wait);
1655 spin_unlock_irq(&q->lock);
1661 * 0 - folio is locked.
1662 * non-zero - folio is not locked.
1663 * mmap_lock or per-VMA lock has been released (mmap_read_unlock() or
1664 * vma_end_read()), unless flags had both FAULT_FLAG_ALLOW_RETRY and
1665 * FAULT_FLAG_RETRY_NOWAIT set, in which case the lock is still held.
1667 * If neither ALLOW_RETRY nor KILLABLE are set, will always return 0
1668 * with the folio locked and the mmap_lock/per-VMA lock is left unperturbed.
1670 vm_fault_t __folio_lock_or_retry(struct folio *folio, struct vm_fault *vmf)
1672 unsigned int flags = vmf->flags;
1674 if (fault_flag_allow_retry_first(flags)) {
1676 * CAUTION! In this case, mmap_lock/per-VMA lock is not
1677 * released even though returning VM_FAULT_RETRY.
1679 if (flags & FAULT_FLAG_RETRY_NOWAIT)
1680 return VM_FAULT_RETRY;
1682 release_fault_lock(vmf);
1683 if (flags & FAULT_FLAG_KILLABLE)
1684 folio_wait_locked_killable(folio);
1686 folio_wait_locked(folio);
1687 return VM_FAULT_RETRY;
1689 if (flags & FAULT_FLAG_KILLABLE) {
1692 ret = __folio_lock_killable(folio);
1694 release_fault_lock(vmf);
1695 return VM_FAULT_RETRY;
1698 __folio_lock(folio);
1705 * page_cache_next_miss() - Find the next gap in the page cache.
1706 * @mapping: Mapping.
1708 * @max_scan: Maximum range to search.
1710 * Search the range [index, min(index + max_scan - 1, ULONG_MAX)] for the
1711 * gap with the lowest index.
1713 * This function may be called under the rcu_read_lock. However, this will
1714 * not atomically search a snapshot of the cache at a single point in time.
1715 * For example, if a gap is created at index 5, then subsequently a gap is
1716 * created at index 10, page_cache_next_miss covering both indices may
1717 * return 10 if called under the rcu_read_lock.
1719 * Return: The index of the gap if found, otherwise an index outside the
1720 * range specified (in which case 'return - index >= max_scan' will be true).
1721 * In the rare case of index wrap-around, 0 will be returned.
1723 pgoff_t page_cache_next_miss(struct address_space *mapping,
1724 pgoff_t index, unsigned long max_scan)
1726 XA_STATE(xas, &mapping->i_pages, index);
1728 while (max_scan--) {
1729 void *entry = xas_next(&xas);
1730 if (!entry || xa_is_value(entry))
1732 if (xas.xa_index == 0)
1736 return xas.xa_index;
1738 EXPORT_SYMBOL(page_cache_next_miss);
1741 * page_cache_prev_miss() - Find the previous gap in the page cache.
1742 * @mapping: Mapping.
1744 * @max_scan: Maximum range to search.
1746 * Search the range [max(index - max_scan + 1, 0), index] for the
1747 * gap with the highest index.
1749 * This function may be called under the rcu_read_lock. However, this will
1750 * not atomically search a snapshot of the cache at a single point in time.
1751 * For example, if a gap is created at index 10, then subsequently a gap is
1752 * created at index 5, page_cache_prev_miss() covering both indices may
1753 * return 5 if called under the rcu_read_lock.
1755 * Return: The index of the gap if found, otherwise an index outside the
1756 * range specified (in which case 'index - return >= max_scan' will be true).
1757 * In the rare case of wrap-around, ULONG_MAX will be returned.
1759 pgoff_t page_cache_prev_miss(struct address_space *mapping,
1760 pgoff_t index, unsigned long max_scan)
1762 XA_STATE(xas, &mapping->i_pages, index);
1764 while (max_scan--) {
1765 void *entry = xas_prev(&xas);
1766 if (!entry || xa_is_value(entry))
1768 if (xas.xa_index == ULONG_MAX)
1772 return xas.xa_index;
1774 EXPORT_SYMBOL(page_cache_prev_miss);
1777 * Lockless page cache protocol:
1778 * On the lookup side:
1779 * 1. Load the folio from i_pages
1780 * 2. Increment the refcount if it's not zero
1781 * 3. If the folio is not found by xas_reload(), put the refcount and retry
1783 * On the removal side:
1784 * A. Freeze the page (by zeroing the refcount if nobody else has a reference)
1785 * B. Remove the page from i_pages
1786 * C. Return the page to the page allocator
1788 * This means that any page may have its reference count temporarily
1789 * increased by a speculative page cache (or fast GUP) lookup as it can
1790 * be allocated by another user before the RCU grace period expires.
1791 * Because the refcount temporarily acquired here may end up being the
1792 * last refcount on the page, any page allocation must be freeable by
1797 * filemap_get_entry - Get a page cache entry.
1798 * @mapping: the address_space to search
1799 * @index: The page cache index.
1801 * Looks up the page cache entry at @mapping & @index. If it is a folio,
1802 * it is returned with an increased refcount. If it is a shadow entry
1803 * of a previously evicted folio, or a swap entry from shmem/tmpfs,
1804 * it is returned without further action.
1806 * Return: The folio, swap or shadow entry, %NULL if nothing is found.
1808 void *filemap_get_entry(struct address_space *mapping, pgoff_t index)
1810 XA_STATE(xas, &mapping->i_pages, index);
1811 struct folio *folio;
1816 folio = xas_load(&xas);
1817 if (xas_retry(&xas, folio))
1820 * A shadow entry of a recently evicted page, or a swap entry from
1821 * shmem/tmpfs. Return it without attempting to raise page count.
1823 if (!folio || xa_is_value(folio))
1826 if (!folio_try_get_rcu(folio))
1829 if (unlikely(folio != xas_reload(&xas))) {
1840 * __filemap_get_folio - Find and get a reference to a folio.
1841 * @mapping: The address_space to search.
1842 * @index: The page index.
1843 * @fgp_flags: %FGP flags modify how the folio is returned.
1844 * @gfp: Memory allocation flags to use if %FGP_CREAT is specified.
1846 * Looks up the page cache entry at @mapping & @index.
1848 * If %FGP_LOCK or %FGP_CREAT are specified then the function may sleep even
1849 * if the %GFP flags specified for %FGP_CREAT are atomic.
1851 * If this function returns a folio, it is returned with an increased refcount.
1853 * Return: The found folio or an ERR_PTR() otherwise.
1855 struct folio *__filemap_get_folio(struct address_space *mapping, pgoff_t index,
1856 fgf_t fgp_flags, gfp_t gfp)
1858 struct folio *folio;
1861 folio = filemap_get_entry(mapping, index);
1862 if (xa_is_value(folio))
1867 if (fgp_flags & FGP_LOCK) {
1868 if (fgp_flags & FGP_NOWAIT) {
1869 if (!folio_trylock(folio)) {
1871 return ERR_PTR(-EAGAIN);
1877 /* Has the page been truncated? */
1878 if (unlikely(folio->mapping != mapping)) {
1879 folio_unlock(folio);
1883 VM_BUG_ON_FOLIO(!folio_contains(folio, index), folio);
1886 if (fgp_flags & FGP_ACCESSED)
1887 folio_mark_accessed(folio);
1888 else if (fgp_flags & FGP_WRITE) {
1889 /* Clear idle flag for buffer write */
1890 if (folio_test_idle(folio))
1891 folio_clear_idle(folio);
1894 if (fgp_flags & FGP_STABLE)
1895 folio_wait_stable(folio);
1897 if (!folio && (fgp_flags & FGP_CREAT)) {
1898 unsigned order = FGF_GET_ORDER(fgp_flags);
1901 if ((fgp_flags & FGP_WRITE) && mapping_can_writeback(mapping))
1903 if (fgp_flags & FGP_NOFS)
1905 if (fgp_flags & FGP_NOWAIT) {
1907 gfp |= GFP_NOWAIT | __GFP_NOWARN;
1909 if (WARN_ON_ONCE(!(fgp_flags & (FGP_LOCK | FGP_FOR_MMAP))))
1910 fgp_flags |= FGP_LOCK;
1912 if (!mapping_large_folio_support(mapping))
1914 if (order > MAX_PAGECACHE_ORDER)
1915 order = MAX_PAGECACHE_ORDER;
1916 /* If we're not aligned, allocate a smaller folio */
1917 if (index & ((1UL << order) - 1))
1918 order = __ffs(index);
1921 gfp_t alloc_gfp = gfp;
1925 alloc_gfp |= __GFP_NORETRY | __GFP_NOWARN;
1926 folio = filemap_alloc_folio(alloc_gfp, order);
1930 /* Init accessed so avoid atomic mark_page_accessed later */
1931 if (fgp_flags & FGP_ACCESSED)
1932 __folio_set_referenced(folio);
1934 err = filemap_add_folio(mapping, folio, index, gfp);
1939 } while (order-- > 0);
1944 return ERR_PTR(err);
1946 * filemap_add_folio locks the page, and for mmap
1947 * we expect an unlocked page.
1949 if (folio && (fgp_flags & FGP_FOR_MMAP))
1950 folio_unlock(folio);
1954 return ERR_PTR(-ENOENT);
1957 EXPORT_SYMBOL(__filemap_get_folio);
1959 static inline struct folio *find_get_entry(struct xa_state *xas, pgoff_t max,
1962 struct folio *folio;
1965 if (mark == XA_PRESENT)
1966 folio = xas_find(xas, max);
1968 folio = xas_find_marked(xas, max, mark);
1970 if (xas_retry(xas, folio))
1973 * A shadow entry of a recently evicted page, a swap
1974 * entry from shmem/tmpfs or a DAX entry. Return it
1975 * without attempting to raise page count.
1977 if (!folio || xa_is_value(folio))
1980 if (!folio_try_get_rcu(folio))
1983 if (unlikely(folio != xas_reload(xas))) {
1995 * find_get_entries - gang pagecache lookup
1996 * @mapping: The address_space to search
1997 * @start: The starting page cache index
1998 * @end: The final page index (inclusive).
1999 * @fbatch: Where the resulting entries are placed.
2000 * @indices: The cache indices corresponding to the entries in @entries
2002 * find_get_entries() will search for and return a batch of entries in
2003 * the mapping. The entries are placed in @fbatch. find_get_entries()
2004 * takes a reference on any actual folios it returns.
2006 * The entries have ascending indexes. The indices may not be consecutive
2007 * due to not-present entries or large folios.
2009 * Any shadow entries of evicted folios, or swap entries from
2010 * shmem/tmpfs, are included in the returned array.
2012 * Return: The number of entries which were found.
2014 unsigned find_get_entries(struct address_space *mapping, pgoff_t *start,
2015 pgoff_t end, struct folio_batch *fbatch, pgoff_t *indices)
2017 XA_STATE(xas, &mapping->i_pages, *start);
2018 struct folio *folio;
2021 while ((folio = find_get_entry(&xas, end, XA_PRESENT)) != NULL) {
2022 indices[fbatch->nr] = xas.xa_index;
2023 if (!folio_batch_add(fbatch, folio))
2028 if (folio_batch_count(fbatch)) {
2029 unsigned long nr = 1;
2030 int idx = folio_batch_count(fbatch) - 1;
2032 folio = fbatch->folios[idx];
2033 if (!xa_is_value(folio))
2034 nr = folio_nr_pages(folio);
2035 *start = indices[idx] + nr;
2037 return folio_batch_count(fbatch);
2041 * find_lock_entries - Find a batch of pagecache entries.
2042 * @mapping: The address_space to search.
2043 * @start: The starting page cache index.
2044 * @end: The final page index (inclusive).
2045 * @fbatch: Where the resulting entries are placed.
2046 * @indices: The cache indices of the entries in @fbatch.
2048 * find_lock_entries() will return a batch of entries from @mapping.
2049 * Swap, shadow and DAX entries are included. Folios are returned
2050 * locked and with an incremented refcount. Folios which are locked
2051 * by somebody else or under writeback are skipped. Folios which are
2052 * partially outside the range are not returned.
2054 * The entries have ascending indexes. The indices may not be consecutive
2055 * due to not-present entries, large folios, folios which could not be
2056 * locked or folios under writeback.
2058 * Return: The number of entries which were found.
2060 unsigned find_lock_entries(struct address_space *mapping, pgoff_t *start,
2061 pgoff_t end, struct folio_batch *fbatch, pgoff_t *indices)
2063 XA_STATE(xas, &mapping->i_pages, *start);
2064 struct folio *folio;
2067 while ((folio = find_get_entry(&xas, end, XA_PRESENT))) {
2068 if (!xa_is_value(folio)) {
2069 if (folio->index < *start)
2071 if (folio_next_index(folio) - 1 > end)
2073 if (!folio_trylock(folio))
2075 if (folio->mapping != mapping ||
2076 folio_test_writeback(folio))
2078 VM_BUG_ON_FOLIO(!folio_contains(folio, xas.xa_index),
2081 indices[fbatch->nr] = xas.xa_index;
2082 if (!folio_batch_add(fbatch, folio))
2086 folio_unlock(folio);
2092 if (folio_batch_count(fbatch)) {
2093 unsigned long nr = 1;
2094 int idx = folio_batch_count(fbatch) - 1;
2096 folio = fbatch->folios[idx];
2097 if (!xa_is_value(folio))
2098 nr = folio_nr_pages(folio);
2099 *start = indices[idx] + nr;
2101 return folio_batch_count(fbatch);
2105 * filemap_get_folios - Get a batch of folios
2106 * @mapping: The address_space to search
2107 * @start: The starting page index
2108 * @end: The final page index (inclusive)
2109 * @fbatch: The batch to fill.
2111 * Search for and return a batch of folios in the mapping starting at
2112 * index @start and up to index @end (inclusive). The folios are returned
2113 * in @fbatch with an elevated reference count.
2115 * Return: The number of folios which were found.
2116 * We also update @start to index the next folio for the traversal.
2118 unsigned filemap_get_folios(struct address_space *mapping, pgoff_t *start,
2119 pgoff_t end, struct folio_batch *fbatch)
2121 return filemap_get_folios_tag(mapping, start, end, XA_PRESENT, fbatch);
2123 EXPORT_SYMBOL(filemap_get_folios);
2126 * filemap_get_folios_contig - Get a batch of contiguous folios
2127 * @mapping: The address_space to search
2128 * @start: The starting page index
2129 * @end: The final page index (inclusive)
2130 * @fbatch: The batch to fill
2132 * filemap_get_folios_contig() works exactly like filemap_get_folios(),
2133 * except the returned folios are guaranteed to be contiguous. This may
2134 * not return all contiguous folios if the batch gets filled up.
2136 * Return: The number of folios found.
2137 * Also update @start to be positioned for traversal of the next folio.
2140 unsigned filemap_get_folios_contig(struct address_space *mapping,
2141 pgoff_t *start, pgoff_t end, struct folio_batch *fbatch)
2143 XA_STATE(xas, &mapping->i_pages, *start);
2145 struct folio *folio;
2149 for (folio = xas_load(&xas); folio && xas.xa_index <= end;
2150 folio = xas_next(&xas)) {
2151 if (xas_retry(&xas, folio))
2154 * If the entry has been swapped out, we can stop looking.
2155 * No current caller is looking for DAX entries.
2157 if (xa_is_value(folio))
2160 if (!folio_try_get_rcu(folio))
2163 if (unlikely(folio != xas_reload(&xas)))
2166 if (!folio_batch_add(fbatch, folio)) {
2167 nr = folio_nr_pages(folio);
2168 *start = folio->index + nr;
2180 nr = folio_batch_count(fbatch);
2183 folio = fbatch->folios[nr - 1];
2184 *start = folio_next_index(folio);
2188 return folio_batch_count(fbatch);
2190 EXPORT_SYMBOL(filemap_get_folios_contig);
2193 * filemap_get_folios_tag - Get a batch of folios matching @tag
2194 * @mapping: The address_space to search
2195 * @start: The starting page index
2196 * @end: The final page index (inclusive)
2197 * @tag: The tag index
2198 * @fbatch: The batch to fill
2200 * The first folio may start before @start; if it does, it will contain
2201 * @start. The final folio may extend beyond @end; if it does, it will
2202 * contain @end. The folios have ascending indices. There may be gaps
2203 * between the folios if there are indices which have no folio in the
2204 * page cache. If folios are added to or removed from the page cache
2205 * while this is running, they may or may not be found by this call.
2206 * Only returns folios that are tagged with @tag.
2208 * Return: The number of folios found.
2209 * Also update @start to index the next folio for traversal.
2211 unsigned filemap_get_folios_tag(struct address_space *mapping, pgoff_t *start,
2212 pgoff_t end, xa_mark_t tag, struct folio_batch *fbatch)
2214 XA_STATE(xas, &mapping->i_pages, *start);
2215 struct folio *folio;
2218 while ((folio = find_get_entry(&xas, end, tag)) != NULL) {
2220 * Shadow entries should never be tagged, but this iteration
2221 * is lockless so there is a window for page reclaim to evict
2222 * a page we saw tagged. Skip over it.
2224 if (xa_is_value(folio))
2226 if (!folio_batch_add(fbatch, folio)) {
2227 unsigned long nr = folio_nr_pages(folio);
2228 *start = folio->index + nr;
2233 * We come here when there is no page beyond @end. We take care to not
2234 * overflow the index @start as it confuses some of the callers. This
2235 * breaks the iteration when there is a page at index -1 but that is
2236 * already broke anyway.
2238 if (end == (pgoff_t)-1)
2239 *start = (pgoff_t)-1;
2245 return folio_batch_count(fbatch);
2247 EXPORT_SYMBOL(filemap_get_folios_tag);
2250 * CD/DVDs are error prone. When a medium error occurs, the driver may fail
2251 * a _large_ part of the i/o request. Imagine the worst scenario:
2253 * ---R__________________________________________B__________
2254 * ^ reading here ^ bad block(assume 4k)
2256 * read(R) => miss => readahead(R...B) => media error => frustrating retries
2257 * => failing the whole request => read(R) => read(R+1) =>
2258 * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
2259 * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
2260 * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
2262 * It is going insane. Fix it by quickly scaling down the readahead size.
2264 static void shrink_readahead_size_eio(struct file_ra_state *ra)
2270 * filemap_get_read_batch - Get a batch of folios for read
2272 * Get a batch of folios which represent a contiguous range of bytes in
2273 * the file. No exceptional entries will be returned. If @index is in
2274 * the middle of a folio, the entire folio will be returned. The last
2275 * folio in the batch may have the readahead flag set or the uptodate flag
2276 * clear so that the caller can take the appropriate action.
2278 static void filemap_get_read_batch(struct address_space *mapping,
2279 pgoff_t index, pgoff_t max, struct folio_batch *fbatch)
2281 XA_STATE(xas, &mapping->i_pages, index);
2282 struct folio *folio;
2285 for (folio = xas_load(&xas); folio; folio = xas_next(&xas)) {
2286 if (xas_retry(&xas, folio))
2288 if (xas.xa_index > max || xa_is_value(folio))
2290 if (xa_is_sibling(folio))
2292 if (!folio_try_get_rcu(folio))
2295 if (unlikely(folio != xas_reload(&xas)))
2298 if (!folio_batch_add(fbatch, folio))
2300 if (!folio_test_uptodate(folio))
2302 if (folio_test_readahead(folio))
2304 xas_advance(&xas, folio_next_index(folio) - 1);
2314 static int filemap_read_folio(struct file *file, filler_t filler,
2315 struct folio *folio)
2317 bool workingset = folio_test_workingset(folio);
2318 unsigned long pflags;
2322 * A previous I/O error may have been due to temporary failures,
2323 * eg. multipath errors. PG_error will be set again if read_folio
2326 folio_clear_error(folio);
2328 /* Start the actual read. The read will unlock the page. */
2329 if (unlikely(workingset))
2330 psi_memstall_enter(&pflags);
2331 error = filler(file, folio);
2332 if (unlikely(workingset))
2333 psi_memstall_leave(&pflags);
2337 error = folio_wait_locked_killable(folio);
2340 if (folio_test_uptodate(folio))
2343 shrink_readahead_size_eio(&file->f_ra);
2347 static bool filemap_range_uptodate(struct address_space *mapping,
2348 loff_t pos, size_t count, struct folio *folio,
2351 if (folio_test_uptodate(folio))
2353 /* pipes can't handle partially uptodate pages */
2356 if (!mapping->a_ops->is_partially_uptodate)
2358 if (mapping->host->i_blkbits >= folio_shift(folio))
2361 if (folio_pos(folio) > pos) {
2362 count -= folio_pos(folio) - pos;
2365 pos -= folio_pos(folio);
2368 return mapping->a_ops->is_partially_uptodate(folio, pos, count);
2371 static int filemap_update_page(struct kiocb *iocb,
2372 struct address_space *mapping, size_t count,
2373 struct folio *folio, bool need_uptodate)
2377 if (iocb->ki_flags & IOCB_NOWAIT) {
2378 if (!filemap_invalidate_trylock_shared(mapping))
2381 filemap_invalidate_lock_shared(mapping);
2384 if (!folio_trylock(folio)) {
2386 if (iocb->ki_flags & (IOCB_NOWAIT | IOCB_NOIO))
2387 goto unlock_mapping;
2388 if (!(iocb->ki_flags & IOCB_WAITQ)) {
2389 filemap_invalidate_unlock_shared(mapping);
2391 * This is where we usually end up waiting for a
2392 * previously submitted readahead to finish.
2394 folio_put_wait_locked(folio, TASK_KILLABLE);
2395 return AOP_TRUNCATED_PAGE;
2397 error = __folio_lock_async(folio, iocb->ki_waitq);
2399 goto unlock_mapping;
2402 error = AOP_TRUNCATED_PAGE;
2403 if (!folio->mapping)
2407 if (filemap_range_uptodate(mapping, iocb->ki_pos, count, folio,
2412 if (iocb->ki_flags & (IOCB_NOIO | IOCB_NOWAIT | IOCB_WAITQ))
2415 error = filemap_read_folio(iocb->ki_filp, mapping->a_ops->read_folio,
2417 goto unlock_mapping;
2419 folio_unlock(folio);
2421 filemap_invalidate_unlock_shared(mapping);
2422 if (error == AOP_TRUNCATED_PAGE)
2427 static int filemap_create_folio(struct file *file,
2428 struct address_space *mapping, pgoff_t index,
2429 struct folio_batch *fbatch)
2431 struct folio *folio;
2434 folio = filemap_alloc_folio(mapping_gfp_mask(mapping), 0);
2439 * Protect against truncate / hole punch. Grabbing invalidate_lock
2440 * here assures we cannot instantiate and bring uptodate new
2441 * pagecache folios after evicting page cache during truncate
2442 * and before actually freeing blocks. Note that we could
2443 * release invalidate_lock after inserting the folio into
2444 * the page cache as the locked folio would then be enough to
2445 * synchronize with hole punching. But there are code paths
2446 * such as filemap_update_page() filling in partially uptodate
2447 * pages or ->readahead() that need to hold invalidate_lock
2448 * while mapping blocks for IO so let's hold the lock here as
2449 * well to keep locking rules simple.
2451 filemap_invalidate_lock_shared(mapping);
2452 error = filemap_add_folio(mapping, folio, index,
2453 mapping_gfp_constraint(mapping, GFP_KERNEL));
2454 if (error == -EEXIST)
2455 error = AOP_TRUNCATED_PAGE;
2459 error = filemap_read_folio(file, mapping->a_ops->read_folio, folio);
2463 filemap_invalidate_unlock_shared(mapping);
2464 folio_batch_add(fbatch, folio);
2467 filemap_invalidate_unlock_shared(mapping);
2472 static int filemap_readahead(struct kiocb *iocb, struct file *file,
2473 struct address_space *mapping, struct folio *folio,
2476 DEFINE_READAHEAD(ractl, file, &file->f_ra, mapping, folio->index);
2478 if (iocb->ki_flags & IOCB_NOIO)
2480 page_cache_async_ra(&ractl, folio, last_index - folio->index);
2484 static int filemap_get_pages(struct kiocb *iocb, size_t count,
2485 struct folio_batch *fbatch, bool need_uptodate)
2487 struct file *filp = iocb->ki_filp;
2488 struct address_space *mapping = filp->f_mapping;
2489 struct file_ra_state *ra = &filp->f_ra;
2490 pgoff_t index = iocb->ki_pos >> PAGE_SHIFT;
2492 struct folio *folio;
2495 /* "last_index" is the index of the page beyond the end of the read */
2496 last_index = DIV_ROUND_UP(iocb->ki_pos + count, PAGE_SIZE);
2498 if (fatal_signal_pending(current))
2501 filemap_get_read_batch(mapping, index, last_index - 1, fbatch);
2502 if (!folio_batch_count(fbatch)) {
2503 if (iocb->ki_flags & IOCB_NOIO)
2505 page_cache_sync_readahead(mapping, ra, filp, index,
2506 last_index - index);
2507 filemap_get_read_batch(mapping, index, last_index - 1, fbatch);
2509 if (!folio_batch_count(fbatch)) {
2510 if (iocb->ki_flags & (IOCB_NOWAIT | IOCB_WAITQ))
2512 err = filemap_create_folio(filp, mapping,
2513 iocb->ki_pos >> PAGE_SHIFT, fbatch);
2514 if (err == AOP_TRUNCATED_PAGE)
2519 folio = fbatch->folios[folio_batch_count(fbatch) - 1];
2520 if (folio_test_readahead(folio)) {
2521 err = filemap_readahead(iocb, filp, mapping, folio, last_index);
2525 if (!folio_test_uptodate(folio)) {
2526 if ((iocb->ki_flags & IOCB_WAITQ) &&
2527 folio_batch_count(fbatch) > 1)
2528 iocb->ki_flags |= IOCB_NOWAIT;
2529 err = filemap_update_page(iocb, mapping, count, folio,
2539 if (likely(--fbatch->nr))
2541 if (err == AOP_TRUNCATED_PAGE)
2546 static inline bool pos_same_folio(loff_t pos1, loff_t pos2, struct folio *folio)
2548 unsigned int shift = folio_shift(folio);
2550 return (pos1 >> shift == pos2 >> shift);
2554 * filemap_read - Read data from the page cache.
2555 * @iocb: The iocb to read.
2556 * @iter: Destination for the data.
2557 * @already_read: Number of bytes already read by the caller.
2559 * Copies data from the page cache. If the data is not currently present,
2560 * uses the readahead and read_folio address_space operations to fetch it.
2562 * Return: Total number of bytes copied, including those already read by
2563 * the caller. If an error happens before any bytes are copied, returns
2564 * a negative error number.
2566 ssize_t filemap_read(struct kiocb *iocb, struct iov_iter *iter,
2567 ssize_t already_read)
2569 struct file *filp = iocb->ki_filp;
2570 struct file_ra_state *ra = &filp->f_ra;
2571 struct address_space *mapping = filp->f_mapping;
2572 struct inode *inode = mapping->host;
2573 struct folio_batch fbatch;
2575 bool writably_mapped;
2576 loff_t isize, end_offset;
2577 loff_t last_pos = ra->prev_pos;
2579 if (unlikely(iocb->ki_pos >= inode->i_sb->s_maxbytes))
2581 if (unlikely(!iov_iter_count(iter)))
2584 iov_iter_truncate(iter, inode->i_sb->s_maxbytes);
2585 folio_batch_init(&fbatch);
2591 * If we've already successfully copied some data, then we
2592 * can no longer safely return -EIOCBQUEUED. Hence mark
2593 * an async read NOWAIT at that point.
2595 if ((iocb->ki_flags & IOCB_WAITQ) && already_read)
2596 iocb->ki_flags |= IOCB_NOWAIT;
2598 if (unlikely(iocb->ki_pos >= i_size_read(inode)))
2601 error = filemap_get_pages(iocb, iter->count, &fbatch, false);
2606 * i_size must be checked after we know the pages are Uptodate.
2608 * Checking i_size after the check allows us to calculate
2609 * the correct value for "nr", which means the zero-filled
2610 * part of the page is not copied back to userspace (unless
2611 * another truncate extends the file - this is desired though).
2613 isize = i_size_read(inode);
2614 if (unlikely(iocb->ki_pos >= isize))
2616 end_offset = min_t(loff_t, isize, iocb->ki_pos + iter->count);
2619 * Once we start copying data, we don't want to be touching any
2620 * cachelines that might be contended:
2622 writably_mapped = mapping_writably_mapped(mapping);
2625 * When a read accesses the same folio several times, only
2626 * mark it as accessed the first time.
2628 if (!pos_same_folio(iocb->ki_pos, last_pos - 1,
2630 folio_mark_accessed(fbatch.folios[0]);
2632 for (i = 0; i < folio_batch_count(&fbatch); i++) {
2633 struct folio *folio = fbatch.folios[i];
2634 size_t fsize = folio_size(folio);
2635 size_t offset = iocb->ki_pos & (fsize - 1);
2636 size_t bytes = min_t(loff_t, end_offset - iocb->ki_pos,
2640 if (end_offset < folio_pos(folio))
2643 folio_mark_accessed(folio);
2645 * If users can be writing to this folio using arbitrary
2646 * virtual addresses, take care of potential aliasing
2647 * before reading the folio on the kernel side.
2649 if (writably_mapped)
2650 flush_dcache_folio(folio);
2652 copied = copy_folio_to_iter(folio, offset, bytes, iter);
2654 already_read += copied;
2655 iocb->ki_pos += copied;
2656 last_pos = iocb->ki_pos;
2658 if (copied < bytes) {
2664 for (i = 0; i < folio_batch_count(&fbatch); i++)
2665 folio_put(fbatch.folios[i]);
2666 folio_batch_init(&fbatch);
2667 } while (iov_iter_count(iter) && iocb->ki_pos < isize && !error);
2669 file_accessed(filp);
2670 ra->prev_pos = last_pos;
2671 return already_read ? already_read : error;
2673 EXPORT_SYMBOL_GPL(filemap_read);
2675 int kiocb_write_and_wait(struct kiocb *iocb, size_t count)
2677 struct address_space *mapping = iocb->ki_filp->f_mapping;
2678 loff_t pos = iocb->ki_pos;
2679 loff_t end = pos + count - 1;
2681 if (iocb->ki_flags & IOCB_NOWAIT) {
2682 if (filemap_range_needs_writeback(mapping, pos, end))
2687 return filemap_write_and_wait_range(mapping, pos, end);
2689 EXPORT_SYMBOL_GPL(kiocb_write_and_wait);
2691 int kiocb_invalidate_pages(struct kiocb *iocb, size_t count)
2693 struct address_space *mapping = iocb->ki_filp->f_mapping;
2694 loff_t pos = iocb->ki_pos;
2695 loff_t end = pos + count - 1;
2698 if (iocb->ki_flags & IOCB_NOWAIT) {
2699 /* we could block if there are any pages in the range */
2700 if (filemap_range_has_page(mapping, pos, end))
2703 ret = filemap_write_and_wait_range(mapping, pos, end);
2709 * After a write we want buffered reads to be sure to go to disk to get
2710 * the new data. We invalidate clean cached page from the region we're
2711 * about to write. We do this *before* the write so that we can return
2712 * without clobbering -EIOCBQUEUED from ->direct_IO().
2714 return invalidate_inode_pages2_range(mapping, pos >> PAGE_SHIFT,
2717 EXPORT_SYMBOL_GPL(kiocb_invalidate_pages);
2720 * generic_file_read_iter - generic filesystem read routine
2721 * @iocb: kernel I/O control block
2722 * @iter: destination for the data read
2724 * This is the "read_iter()" routine for all filesystems
2725 * that can use the page cache directly.
2727 * The IOCB_NOWAIT flag in iocb->ki_flags indicates that -EAGAIN shall
2728 * be returned when no data can be read without waiting for I/O requests
2729 * to complete; it doesn't prevent readahead.
2731 * The IOCB_NOIO flag in iocb->ki_flags indicates that no new I/O
2732 * requests shall be made for the read or for readahead. When no data
2733 * can be read, -EAGAIN shall be returned. When readahead would be
2734 * triggered, a partial, possibly empty read shall be returned.
2737 * * number of bytes copied, even for partial reads
2738 * * negative error code (or 0 if IOCB_NOIO) if nothing was read
2741 generic_file_read_iter(struct kiocb *iocb, struct iov_iter *iter)
2743 size_t count = iov_iter_count(iter);
2747 return 0; /* skip atime */
2749 if (iocb->ki_flags & IOCB_DIRECT) {
2750 struct file *file = iocb->ki_filp;
2751 struct address_space *mapping = file->f_mapping;
2752 struct inode *inode = mapping->host;
2754 retval = kiocb_write_and_wait(iocb, count);
2757 file_accessed(file);
2759 retval = mapping->a_ops->direct_IO(iocb, iter);
2761 iocb->ki_pos += retval;
2764 if (retval != -EIOCBQUEUED)
2765 iov_iter_revert(iter, count - iov_iter_count(iter));
2768 * Btrfs can have a short DIO read if we encounter
2769 * compressed extents, so if there was an error, or if
2770 * we've already read everything we wanted to, or if
2771 * there was a short read because we hit EOF, go ahead
2772 * and return. Otherwise fallthrough to buffered io for
2773 * the rest of the read. Buffered reads will not work for
2774 * DAX files, so don't bother trying.
2776 if (retval < 0 || !count || IS_DAX(inode))
2778 if (iocb->ki_pos >= i_size_read(inode))
2782 return filemap_read(iocb, iter, retval);
2784 EXPORT_SYMBOL(generic_file_read_iter);
2787 * Splice subpages from a folio into a pipe.
2789 size_t splice_folio_into_pipe(struct pipe_inode_info *pipe,
2790 struct folio *folio, loff_t fpos, size_t size)
2793 size_t spliced = 0, offset = offset_in_folio(folio, fpos);
2795 page = folio_page(folio, offset / PAGE_SIZE);
2796 size = min(size, folio_size(folio) - offset);
2797 offset %= PAGE_SIZE;
2799 while (spliced < size &&
2800 !pipe_full(pipe->head, pipe->tail, pipe->max_usage)) {
2801 struct pipe_buffer *buf = pipe_head_buf(pipe);
2802 size_t part = min_t(size_t, PAGE_SIZE - offset, size - spliced);
2804 *buf = (struct pipe_buffer) {
2805 .ops = &page_cache_pipe_buf_ops,
2821 * filemap_splice_read - Splice data from a file's pagecache into a pipe
2822 * @in: The file to read from
2823 * @ppos: Pointer to the file position to read from
2824 * @pipe: The pipe to splice into
2825 * @len: The amount to splice
2826 * @flags: The SPLICE_F_* flags
2828 * This function gets folios from a file's pagecache and splices them into the
2829 * pipe. Readahead will be called as necessary to fill more folios. This may
2830 * be used for blockdevs also.
2832 * Return: On success, the number of bytes read will be returned and *@ppos
2833 * will be updated if appropriate; 0 will be returned if there is no more data
2834 * to be read; -EAGAIN will be returned if the pipe had no space, and some
2835 * other negative error code will be returned on error. A short read may occur
2836 * if the pipe has insufficient space, we reach the end of the data or we hit a
2839 ssize_t filemap_splice_read(struct file *in, loff_t *ppos,
2840 struct pipe_inode_info *pipe,
2841 size_t len, unsigned int flags)
2843 struct folio_batch fbatch;
2845 size_t total_spliced = 0, used, npages;
2846 loff_t isize, end_offset;
2847 bool writably_mapped;
2850 if (unlikely(*ppos >= in->f_mapping->host->i_sb->s_maxbytes))
2853 init_sync_kiocb(&iocb, in);
2854 iocb.ki_pos = *ppos;
2856 /* Work out how much data we can actually add into the pipe */
2857 used = pipe_occupancy(pipe->head, pipe->tail);
2858 npages = max_t(ssize_t, pipe->max_usage - used, 0);
2859 len = min_t(size_t, len, npages * PAGE_SIZE);
2861 folio_batch_init(&fbatch);
2866 if (*ppos >= i_size_read(in->f_mapping->host))
2869 iocb.ki_pos = *ppos;
2870 error = filemap_get_pages(&iocb, len, &fbatch, true);
2875 * i_size must be checked after we know the pages are Uptodate.
2877 * Checking i_size after the check allows us to calculate
2878 * the correct value for "nr", which means the zero-filled
2879 * part of the page is not copied back to userspace (unless
2880 * another truncate extends the file - this is desired though).
2882 isize = i_size_read(in->f_mapping->host);
2883 if (unlikely(*ppos >= isize))
2885 end_offset = min_t(loff_t, isize, *ppos + len);
2888 * Once we start copying data, we don't want to be touching any
2889 * cachelines that might be contended:
2891 writably_mapped = mapping_writably_mapped(in->f_mapping);
2893 for (i = 0; i < folio_batch_count(&fbatch); i++) {
2894 struct folio *folio = fbatch.folios[i];
2897 if (folio_pos(folio) >= end_offset)
2899 folio_mark_accessed(folio);
2902 * If users can be writing to this folio using arbitrary
2903 * virtual addresses, take care of potential aliasing
2904 * before reading the folio on the kernel side.
2906 if (writably_mapped)
2907 flush_dcache_folio(folio);
2909 n = min_t(loff_t, len, isize - *ppos);
2910 n = splice_folio_into_pipe(pipe, folio, *ppos, n);
2916 in->f_ra.prev_pos = *ppos;
2917 if (pipe_full(pipe->head, pipe->tail, pipe->max_usage))
2921 folio_batch_release(&fbatch);
2925 folio_batch_release(&fbatch);
2928 return total_spliced ? total_spliced : error;
2930 EXPORT_SYMBOL(filemap_splice_read);
2932 static inline loff_t folio_seek_hole_data(struct xa_state *xas,
2933 struct address_space *mapping, struct folio *folio,
2934 loff_t start, loff_t end, bool seek_data)
2936 const struct address_space_operations *ops = mapping->a_ops;
2937 size_t offset, bsz = i_blocksize(mapping->host);
2939 if (xa_is_value(folio) || folio_test_uptodate(folio))
2940 return seek_data ? start : end;
2941 if (!ops->is_partially_uptodate)
2942 return seek_data ? end : start;
2947 if (unlikely(folio->mapping != mapping))
2950 offset = offset_in_folio(folio, start) & ~(bsz - 1);
2953 if (ops->is_partially_uptodate(folio, offset, bsz) ==
2956 start = (start + bsz) & ~(bsz - 1);
2958 } while (offset < folio_size(folio));
2960 folio_unlock(folio);
2965 static inline size_t seek_folio_size(struct xa_state *xas, struct folio *folio)
2967 if (xa_is_value(folio))
2968 return PAGE_SIZE << xa_get_order(xas->xa, xas->xa_index);
2969 return folio_size(folio);
2973 * mapping_seek_hole_data - Seek for SEEK_DATA / SEEK_HOLE in the page cache.
2974 * @mapping: Address space to search.
2975 * @start: First byte to consider.
2976 * @end: Limit of search (exclusive).
2977 * @whence: Either SEEK_HOLE or SEEK_DATA.
2979 * If the page cache knows which blocks contain holes and which blocks
2980 * contain data, your filesystem can use this function to implement
2981 * SEEK_HOLE and SEEK_DATA. This is useful for filesystems which are
2982 * entirely memory-based such as tmpfs, and filesystems which support
2983 * unwritten extents.
2985 * Return: The requested offset on success, or -ENXIO if @whence specifies
2986 * SEEK_DATA and there is no data after @start. There is an implicit hole
2987 * after @end - 1, so SEEK_HOLE returns @end if all the bytes between @start
2988 * and @end contain data.
2990 loff_t mapping_seek_hole_data(struct address_space *mapping, loff_t start,
2991 loff_t end, int whence)
2993 XA_STATE(xas, &mapping->i_pages, start >> PAGE_SHIFT);
2994 pgoff_t max = (end - 1) >> PAGE_SHIFT;
2995 bool seek_data = (whence == SEEK_DATA);
2996 struct folio *folio;
3002 while ((folio = find_get_entry(&xas, max, XA_PRESENT))) {
3003 loff_t pos = (u64)xas.xa_index << PAGE_SHIFT;
3012 seek_size = seek_folio_size(&xas, folio);
3013 pos = round_up((u64)pos + 1, seek_size);
3014 start = folio_seek_hole_data(&xas, mapping, folio, start, pos,
3020 if (seek_size > PAGE_SIZE)
3021 xas_set(&xas, pos >> PAGE_SHIFT);
3022 if (!xa_is_value(folio))
3029 if (folio && !xa_is_value(folio))
3037 #define MMAP_LOTSAMISS (100)
3039 * lock_folio_maybe_drop_mmap - lock the page, possibly dropping the mmap_lock
3040 * @vmf - the vm_fault for this fault.
3041 * @folio - the folio to lock.
3042 * @fpin - the pointer to the file we may pin (or is already pinned).
3044 * This works similar to lock_folio_or_retry in that it can drop the
3045 * mmap_lock. It differs in that it actually returns the folio locked
3046 * if it returns 1 and 0 if it couldn't lock the folio. If we did have
3047 * to drop the mmap_lock then fpin will point to the pinned file and
3048 * needs to be fput()'ed at a later point.
3050 static int lock_folio_maybe_drop_mmap(struct vm_fault *vmf, struct folio *folio,
3053 if (folio_trylock(folio))
3057 * NOTE! This will make us return with VM_FAULT_RETRY, but with
3058 * the fault lock still held. That's how FAULT_FLAG_RETRY_NOWAIT
3059 * is supposed to work. We have way too many special cases..
3061 if (vmf->flags & FAULT_FLAG_RETRY_NOWAIT)
3064 *fpin = maybe_unlock_mmap_for_io(vmf, *fpin);
3065 if (vmf->flags & FAULT_FLAG_KILLABLE) {
3066 if (__folio_lock_killable(folio)) {
3068 * We didn't have the right flags to drop the
3069 * fault lock, but all fault_handlers only check
3070 * for fatal signals if we return VM_FAULT_RETRY,
3071 * so we need to drop the fault lock here and
3072 * return 0 if we don't have a fpin.
3075 release_fault_lock(vmf);
3079 __folio_lock(folio);
3085 * Synchronous readahead happens when we don't even find a page in the page
3086 * cache at all. We don't want to perform IO under the mmap sem, so if we have
3087 * to drop the mmap sem we return the file that was pinned in order for us to do
3088 * that. If we didn't pin a file then we return NULL. The file that is
3089 * returned needs to be fput()'ed when we're done with it.
3091 static struct file *do_sync_mmap_readahead(struct vm_fault *vmf)
3093 struct file *file = vmf->vma->vm_file;
3094 struct file_ra_state *ra = &file->f_ra;
3095 struct address_space *mapping = file->f_mapping;
3096 DEFINE_READAHEAD(ractl, file, ra, mapping, vmf->pgoff);
3097 struct file *fpin = NULL;
3098 unsigned long vm_flags = vmf->vma->vm_flags;
3099 unsigned int mmap_miss;
3101 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
3102 /* Use the readahead code, even if readahead is disabled */
3103 if (vm_flags & VM_HUGEPAGE) {
3104 fpin = maybe_unlock_mmap_for_io(vmf, fpin);
3105 ractl._index &= ~((unsigned long)HPAGE_PMD_NR - 1);
3106 ra->size = HPAGE_PMD_NR;
3108 * Fetch two PMD folios, so we get the chance to actually
3109 * readahead, unless we've been told not to.
3111 if (!(vm_flags & VM_RAND_READ))
3113 ra->async_size = HPAGE_PMD_NR;
3114 page_cache_ra_order(&ractl, ra, HPAGE_PMD_ORDER);
3119 /* If we don't want any read-ahead, don't bother */
3120 if (vm_flags & VM_RAND_READ)
3125 if (vm_flags & VM_SEQ_READ) {
3126 fpin = maybe_unlock_mmap_for_io(vmf, fpin);
3127 page_cache_sync_ra(&ractl, ra->ra_pages);
3131 /* Avoid banging the cache line if not needed */
3132 mmap_miss = READ_ONCE(ra->mmap_miss);
3133 if (mmap_miss < MMAP_LOTSAMISS * 10)
3134 WRITE_ONCE(ra->mmap_miss, ++mmap_miss);
3137 * Do we miss much more than hit in this file? If so,
3138 * stop bothering with read-ahead. It will only hurt.
3140 if (mmap_miss > MMAP_LOTSAMISS)
3146 fpin = maybe_unlock_mmap_for_io(vmf, fpin);
3147 ra->start = max_t(long, 0, vmf->pgoff - ra->ra_pages / 2);
3148 ra->size = ra->ra_pages;
3149 ra->async_size = ra->ra_pages / 4;
3150 ractl._index = ra->start;
3151 page_cache_ra_order(&ractl, ra, 0);
3156 * Asynchronous readahead happens when we find the page and PG_readahead,
3157 * so we want to possibly extend the readahead further. We return the file that
3158 * was pinned if we have to drop the mmap_lock in order to do IO.
3160 static struct file *do_async_mmap_readahead(struct vm_fault *vmf,
3161 struct folio *folio)
3163 struct file *file = vmf->vma->vm_file;
3164 struct file_ra_state *ra = &file->f_ra;
3165 DEFINE_READAHEAD(ractl, file, ra, file->f_mapping, vmf->pgoff);
3166 struct file *fpin = NULL;
3167 unsigned int mmap_miss;
3169 /* If we don't want any read-ahead, don't bother */
3170 if (vmf->vma->vm_flags & VM_RAND_READ || !ra->ra_pages)
3173 mmap_miss = READ_ONCE(ra->mmap_miss);
3175 WRITE_ONCE(ra->mmap_miss, --mmap_miss);
3177 if (folio_test_readahead(folio)) {
3178 fpin = maybe_unlock_mmap_for_io(vmf, fpin);
3179 page_cache_async_ra(&ractl, folio, ra->ra_pages);
3184 static vm_fault_t filemap_fault_recheck_pte_none(struct vm_fault *vmf)
3186 struct vm_area_struct *vma = vmf->vma;
3191 * We might have COW'ed a pagecache folio and might now have an mlocked
3192 * anon folio mapped. The original pagecache folio is not mlocked and
3193 * might have been evicted. During a read+clear/modify/write update of
3194 * the PTE, such as done in do_numa_page()/change_pte_range(), we
3195 * temporarily clear the PTE under PT lock and might detect it here as
3196 * "none" when not holding the PT lock.
3198 * Not rechecking the PTE under PT lock could result in an unexpected
3199 * major fault in an mlock'ed region. Recheck only for this special
3200 * scenario while holding the PT lock, to not degrade non-mlocked
3201 * scenarios. Recheck the PTE without PT lock firstly, thereby reducing
3202 * the number of times we hold PT lock.
3204 if (!(vma->vm_flags & VM_LOCKED))
3207 if (!(vmf->flags & FAULT_FLAG_ORIG_PTE_VALID))
3210 ptep = pte_offset_map(vmf->pmd, vmf->address);
3211 if (unlikely(!ptep))
3212 return VM_FAULT_NOPAGE;
3214 if (unlikely(!pte_none(ptep_get_lockless(ptep)))) {
3215 ret = VM_FAULT_NOPAGE;
3217 spin_lock(vmf->ptl);
3218 if (unlikely(!pte_none(ptep_get(ptep))))
3219 ret = VM_FAULT_NOPAGE;
3220 spin_unlock(vmf->ptl);
3227 * filemap_fault - read in file data for page fault handling
3228 * @vmf: struct vm_fault containing details of the fault
3230 * filemap_fault() is invoked via the vma operations vector for a
3231 * mapped memory region to read in file data during a page fault.
3233 * The goto's are kind of ugly, but this streamlines the normal case of having
3234 * it in the page cache, and handles the special cases reasonably without
3235 * having a lot of duplicated code.
3237 * vma->vm_mm->mmap_lock must be held on entry.
3239 * If our return value has VM_FAULT_RETRY set, it's because the mmap_lock
3240 * may be dropped before doing I/O or by lock_folio_maybe_drop_mmap().
3242 * If our return value does not have VM_FAULT_RETRY set, the mmap_lock
3243 * has not been released.
3245 * We never return with VM_FAULT_RETRY and a bit from VM_FAULT_ERROR set.
3247 * Return: bitwise-OR of %VM_FAULT_ codes.
3249 vm_fault_t filemap_fault(struct vm_fault *vmf)
3252 struct file *file = vmf->vma->vm_file;
3253 struct file *fpin = NULL;
3254 struct address_space *mapping = file->f_mapping;
3255 struct inode *inode = mapping->host;
3256 pgoff_t max_idx, index = vmf->pgoff;
3257 struct folio *folio;
3259 bool mapping_locked = false;
3261 max_idx = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
3262 if (unlikely(index >= max_idx))
3263 return VM_FAULT_SIGBUS;
3266 * Do we have something in the page cache already?
3268 folio = filemap_get_folio(mapping, index);
3269 if (likely(!IS_ERR(folio))) {
3271 * We found the page, so try async readahead before waiting for
3274 if (!(vmf->flags & FAULT_FLAG_TRIED))
3275 fpin = do_async_mmap_readahead(vmf, folio);
3276 if (unlikely(!folio_test_uptodate(folio))) {
3277 filemap_invalidate_lock_shared(mapping);
3278 mapping_locked = true;
3281 ret = filemap_fault_recheck_pte_none(vmf);
3285 /* No page in the page cache at all */
3286 count_vm_event(PGMAJFAULT);
3287 count_memcg_event_mm(vmf->vma->vm_mm, PGMAJFAULT);
3288 ret = VM_FAULT_MAJOR;
3289 fpin = do_sync_mmap_readahead(vmf);
3292 * See comment in filemap_create_folio() why we need
3295 if (!mapping_locked) {
3296 filemap_invalidate_lock_shared(mapping);
3297 mapping_locked = true;
3299 folio = __filemap_get_folio(mapping, index,
3300 FGP_CREAT|FGP_FOR_MMAP,
3302 if (IS_ERR(folio)) {
3305 filemap_invalidate_unlock_shared(mapping);
3306 return VM_FAULT_OOM;
3310 if (!lock_folio_maybe_drop_mmap(vmf, folio, &fpin))
3313 /* Did it get truncated? */
3314 if (unlikely(folio->mapping != mapping)) {
3315 folio_unlock(folio);
3319 VM_BUG_ON_FOLIO(!folio_contains(folio, index), folio);
3322 * We have a locked folio in the page cache, now we need to check
3323 * that it's up-to-date. If not, it is going to be due to an error,
3324 * or because readahead was otherwise unable to retrieve it.
3326 if (unlikely(!folio_test_uptodate(folio))) {
3328 * If the invalidate lock is not held, the folio was in cache
3329 * and uptodate and now it is not. Strange but possible since we
3330 * didn't hold the page lock all the time. Let's drop
3331 * everything, get the invalidate lock and try again.
3333 if (!mapping_locked) {
3334 folio_unlock(folio);
3340 * OK, the folio is really not uptodate. This can be because the
3341 * VMA has the VM_RAND_READ flag set, or because an error
3342 * arose. Let's read it in directly.
3344 goto page_not_uptodate;
3348 * We've made it this far and we had to drop our mmap_lock, now is the
3349 * time to return to the upper layer and have it re-find the vma and
3353 folio_unlock(folio);
3357 filemap_invalidate_unlock_shared(mapping);
3360 * Found the page and have a reference on it.
3361 * We must recheck i_size under page lock.
3363 max_idx = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
3364 if (unlikely(index >= max_idx)) {
3365 folio_unlock(folio);
3367 return VM_FAULT_SIGBUS;
3370 vmf->page = folio_file_page(folio, index);
3371 return ret | VM_FAULT_LOCKED;
3375 * Umm, take care of errors if the page isn't up-to-date.
3376 * Try to re-read it _once_. We do this synchronously,
3377 * because there really aren't any performance issues here
3378 * and we need to check for errors.
3380 fpin = maybe_unlock_mmap_for_io(vmf, fpin);
3381 error = filemap_read_folio(file, mapping->a_ops->read_folio, folio);
3386 if (!error || error == AOP_TRUNCATED_PAGE)
3388 filemap_invalidate_unlock_shared(mapping);
3390 return VM_FAULT_SIGBUS;
3394 * We dropped the mmap_lock, we need to return to the fault handler to
3395 * re-find the vma and come back and find our hopefully still populated
3401 filemap_invalidate_unlock_shared(mapping);
3404 return ret | VM_FAULT_RETRY;
3406 EXPORT_SYMBOL(filemap_fault);
3408 static bool filemap_map_pmd(struct vm_fault *vmf, struct folio *folio,
3411 struct mm_struct *mm = vmf->vma->vm_mm;
3413 /* Huge page is mapped? No need to proceed. */
3414 if (pmd_trans_huge(*vmf->pmd)) {
3415 folio_unlock(folio);
3420 if (pmd_none(*vmf->pmd) && folio_test_pmd_mappable(folio)) {
3421 struct page *page = folio_file_page(folio, start);
3422 vm_fault_t ret = do_set_pmd(vmf, page);
3424 /* The page is mapped successfully, reference consumed. */
3425 folio_unlock(folio);
3430 if (pmd_none(*vmf->pmd) && vmf->prealloc_pte)
3431 pmd_install(mm, vmf->pmd, &vmf->prealloc_pte);
3436 static struct folio *next_uptodate_folio(struct xa_state *xas,
3437 struct address_space *mapping, pgoff_t end_pgoff)
3439 struct folio *folio = xas_next_entry(xas, end_pgoff);
3440 unsigned long max_idx;
3445 if (xas_retry(xas, folio))
3447 if (xa_is_value(folio))
3449 if (folio_test_locked(folio))
3451 if (!folio_try_get_rcu(folio))
3453 /* Has the page moved or been split? */
3454 if (unlikely(folio != xas_reload(xas)))
3456 if (!folio_test_uptodate(folio) || folio_test_readahead(folio))
3458 if (!folio_trylock(folio))
3460 if (folio->mapping != mapping)
3462 if (!folio_test_uptodate(folio))
3464 max_idx = DIV_ROUND_UP(i_size_read(mapping->host), PAGE_SIZE);
3465 if (xas->xa_index >= max_idx)
3469 folio_unlock(folio);
3472 } while ((folio = xas_next_entry(xas, end_pgoff)) != NULL);
3478 * Map page range [start_page, start_page + nr_pages) of folio.
3479 * start_page is gotten from start by folio_page(folio, start)
3481 static vm_fault_t filemap_map_folio_range(struct vm_fault *vmf,
3482 struct folio *folio, unsigned long start,
3483 unsigned long addr, unsigned int nr_pages,
3484 unsigned int *mmap_miss)
3487 struct page *page = folio_page(folio, start);
3488 unsigned int count = 0;
3489 pte_t *old_ptep = vmf->pte;
3492 if (PageHWPoison(page + count))
3498 * NOTE: If there're PTE markers, we'll leave them to be
3499 * handled in the specific fault path, and it'll prohibit the
3500 * fault-around logic.
3502 if (!pte_none(ptep_get(&vmf->pte[count])))
3509 set_pte_range(vmf, folio, page, count, addr);
3510 folio_ref_add(folio, count);
3511 if (in_range(vmf->address, addr, count * PAGE_SIZE))
3512 ret = VM_FAULT_NOPAGE;
3518 addr += count * PAGE_SIZE;
3520 } while (--nr_pages > 0);
3523 set_pte_range(vmf, folio, page, count, addr);
3524 folio_ref_add(folio, count);
3525 if (in_range(vmf->address, addr, count * PAGE_SIZE))
3526 ret = VM_FAULT_NOPAGE;
3529 vmf->pte = old_ptep;
3534 static vm_fault_t filemap_map_order0_folio(struct vm_fault *vmf,
3535 struct folio *folio, unsigned long addr,
3536 unsigned int *mmap_miss)
3539 struct page *page = &folio->page;
3541 if (PageHWPoison(page))
3547 * NOTE: If there're PTE markers, we'll leave them to be
3548 * handled in the specific fault path, and it'll prohibit
3549 * the fault-around logic.
3551 if (!pte_none(ptep_get(vmf->pte)))
3554 if (vmf->address == addr)
3555 ret = VM_FAULT_NOPAGE;
3557 set_pte_range(vmf, folio, page, 1, addr);
3558 folio_ref_inc(folio);
3563 vm_fault_t filemap_map_pages(struct vm_fault *vmf,
3564 pgoff_t start_pgoff, pgoff_t end_pgoff)
3566 struct vm_area_struct *vma = vmf->vma;
3567 struct file *file = vma->vm_file;
3568 struct address_space *mapping = file->f_mapping;
3569 pgoff_t last_pgoff = start_pgoff;
3571 XA_STATE(xas, &mapping->i_pages, start_pgoff);
3572 struct folio *folio;
3574 unsigned int nr_pages = 0, mmap_miss = 0, mmap_miss_saved;
3577 folio = next_uptodate_folio(&xas, mapping, end_pgoff);
3581 if (filemap_map_pmd(vmf, folio, start_pgoff)) {
3582 ret = VM_FAULT_NOPAGE;
3586 addr = vma->vm_start + ((start_pgoff - vma->vm_pgoff) << PAGE_SHIFT);
3587 vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, addr, &vmf->ptl);
3589 folio_unlock(folio);
3596 addr += (xas.xa_index - last_pgoff) << PAGE_SHIFT;
3597 vmf->pte += xas.xa_index - last_pgoff;
3598 last_pgoff = xas.xa_index;
3599 end = folio_next_index(folio) - 1;
3600 nr_pages = min(end, end_pgoff) - xas.xa_index + 1;
3602 if (!folio_test_large(folio))
3603 ret |= filemap_map_order0_folio(vmf,
3604 folio, addr, &mmap_miss);
3606 ret |= filemap_map_folio_range(vmf, folio,
3607 xas.xa_index - folio->index, addr,
3608 nr_pages, &mmap_miss);
3610 folio_unlock(folio);
3612 } while ((folio = next_uptodate_folio(&xas, mapping, end_pgoff)) != NULL);
3613 pte_unmap_unlock(vmf->pte, vmf->ptl);
3617 mmap_miss_saved = READ_ONCE(file->f_ra.mmap_miss);
3618 if (mmap_miss >= mmap_miss_saved)
3619 WRITE_ONCE(file->f_ra.mmap_miss, 0);
3621 WRITE_ONCE(file->f_ra.mmap_miss, mmap_miss_saved - mmap_miss);
3625 EXPORT_SYMBOL(filemap_map_pages);
3627 vm_fault_t filemap_page_mkwrite(struct vm_fault *vmf)
3629 struct address_space *mapping = vmf->vma->vm_file->f_mapping;
3630 struct folio *folio = page_folio(vmf->page);
3631 vm_fault_t ret = VM_FAULT_LOCKED;
3633 sb_start_pagefault(mapping->host->i_sb);
3634 file_update_time(vmf->vma->vm_file);
3636 if (folio->mapping != mapping) {
3637 folio_unlock(folio);
3638 ret = VM_FAULT_NOPAGE;
3642 * We mark the folio dirty already here so that when freeze is in
3643 * progress, we are guaranteed that writeback during freezing will
3644 * see the dirty folio and writeprotect it again.
3646 folio_mark_dirty(folio);
3647 folio_wait_stable(folio);
3649 sb_end_pagefault(mapping->host->i_sb);
3653 const struct vm_operations_struct generic_file_vm_ops = {
3654 .fault = filemap_fault,
3655 .map_pages = filemap_map_pages,
3656 .page_mkwrite = filemap_page_mkwrite,
3659 /* This is used for a general mmap of a disk file */
3661 int generic_file_mmap(struct file *file, struct vm_area_struct *vma)
3663 struct address_space *mapping = file->f_mapping;
3665 if (!mapping->a_ops->read_folio)
3667 file_accessed(file);
3668 vma->vm_ops = &generic_file_vm_ops;
3673 * This is for filesystems which do not implement ->writepage.
3675 int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
3677 if (vma_is_shared_maywrite(vma))
3679 return generic_file_mmap(file, vma);
3682 vm_fault_t filemap_page_mkwrite(struct vm_fault *vmf)
3684 return VM_FAULT_SIGBUS;
3686 int generic_file_mmap(struct file *file, struct vm_area_struct *vma)
3690 int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
3694 #endif /* CONFIG_MMU */
3696 EXPORT_SYMBOL(filemap_page_mkwrite);
3697 EXPORT_SYMBOL(generic_file_mmap);
3698 EXPORT_SYMBOL(generic_file_readonly_mmap);
3700 static struct folio *do_read_cache_folio(struct address_space *mapping,
3701 pgoff_t index, filler_t filler, struct file *file, gfp_t gfp)
3703 struct folio *folio;
3707 filler = mapping->a_ops->read_folio;
3709 folio = filemap_get_folio(mapping, index);
3710 if (IS_ERR(folio)) {
3711 folio = filemap_alloc_folio(gfp, 0);
3713 return ERR_PTR(-ENOMEM);
3714 err = filemap_add_folio(mapping, folio, index, gfp);
3715 if (unlikely(err)) {
3719 /* Presumably ENOMEM for xarray node */
3720 return ERR_PTR(err);
3725 if (folio_test_uptodate(folio))
3728 if (!folio_trylock(folio)) {
3729 folio_put_wait_locked(folio, TASK_UNINTERRUPTIBLE);
3733 /* Folio was truncated from mapping */
3734 if (!folio->mapping) {
3735 folio_unlock(folio);
3740 /* Someone else locked and filled the page in a very small window */
3741 if (folio_test_uptodate(folio)) {
3742 folio_unlock(folio);
3747 err = filemap_read_folio(file, filler, folio);
3750 if (err == AOP_TRUNCATED_PAGE)
3752 return ERR_PTR(err);
3756 folio_mark_accessed(folio);
3761 * read_cache_folio - Read into page cache, fill it if needed.
3762 * @mapping: The address_space to read from.
3763 * @index: The index to read.
3764 * @filler: Function to perform the read, or NULL to use aops->read_folio().
3765 * @file: Passed to filler function, may be NULL if not required.
3767 * Read one page into the page cache. If it succeeds, the folio returned
3768 * will contain @index, but it may not be the first page of the folio.
3770 * If the filler function returns an error, it will be returned to the
3773 * Context: May sleep. Expects mapping->invalidate_lock to be held.
3774 * Return: An uptodate folio on success, ERR_PTR() on failure.
3776 struct folio *read_cache_folio(struct address_space *mapping, pgoff_t index,
3777 filler_t filler, struct file *file)
3779 return do_read_cache_folio(mapping, index, filler, file,
3780 mapping_gfp_mask(mapping));
3782 EXPORT_SYMBOL(read_cache_folio);
3785 * mapping_read_folio_gfp - Read into page cache, using specified allocation flags.
3786 * @mapping: The address_space for the folio.
3787 * @index: The index that the allocated folio will contain.
3788 * @gfp: The page allocator flags to use if allocating.
3790 * This is the same as "read_cache_folio(mapping, index, NULL, NULL)", but with
3791 * any new memory allocations done using the specified allocation flags.
3793 * The most likely error from this function is EIO, but ENOMEM is
3794 * possible and so is EINTR. If ->read_folio returns another error,
3795 * that will be returned to the caller.
3797 * The function expects mapping->invalidate_lock to be already held.
3799 * Return: Uptodate folio on success, ERR_PTR() on failure.
3801 struct folio *mapping_read_folio_gfp(struct address_space *mapping,
3802 pgoff_t index, gfp_t gfp)
3804 return do_read_cache_folio(mapping, index, NULL, NULL, gfp);
3806 EXPORT_SYMBOL(mapping_read_folio_gfp);
3808 static struct page *do_read_cache_page(struct address_space *mapping,
3809 pgoff_t index, filler_t *filler, struct file *file, gfp_t gfp)
3811 struct folio *folio;
3813 folio = do_read_cache_folio(mapping, index, filler, file, gfp);
3815 return &folio->page;
3816 return folio_file_page(folio, index);
3819 struct page *read_cache_page(struct address_space *mapping,
3820 pgoff_t index, filler_t *filler, struct file *file)
3822 return do_read_cache_page(mapping, index, filler, file,
3823 mapping_gfp_mask(mapping));
3825 EXPORT_SYMBOL(read_cache_page);
3828 * read_cache_page_gfp - read into page cache, using specified page allocation flags.
3829 * @mapping: the page's address_space
3830 * @index: the page index
3831 * @gfp: the page allocator flags to use if allocating
3833 * This is the same as "read_mapping_page(mapping, index, NULL)", but with
3834 * any new page allocations done using the specified allocation flags.
3836 * If the page does not get brought uptodate, return -EIO.
3838 * The function expects mapping->invalidate_lock to be already held.
3840 * Return: up to date page on success, ERR_PTR() on failure.
3842 struct page *read_cache_page_gfp(struct address_space *mapping,
3846 return do_read_cache_page(mapping, index, NULL, NULL, gfp);
3848 EXPORT_SYMBOL(read_cache_page_gfp);
3851 * Warn about a page cache invalidation failure during a direct I/O write.
3853 static void dio_warn_stale_pagecache(struct file *filp)
3855 static DEFINE_RATELIMIT_STATE(_rs, 86400 * HZ, DEFAULT_RATELIMIT_BURST);
3859 errseq_set(&filp->f_mapping->wb_err, -EIO);
3860 if (__ratelimit(&_rs)) {
3861 path = file_path(filp, pathname, sizeof(pathname));
3864 pr_crit("Page cache invalidation failure on direct I/O. Possible data corruption due to collision with buffered I/O!\n");
3865 pr_crit("File: %s PID: %d Comm: %.20s\n", path, current->pid,
3870 void kiocb_invalidate_post_direct_write(struct kiocb *iocb, size_t count)
3872 struct address_space *mapping = iocb->ki_filp->f_mapping;
3874 if (mapping->nrpages &&
3875 invalidate_inode_pages2_range(mapping,
3876 iocb->ki_pos >> PAGE_SHIFT,
3877 (iocb->ki_pos + count - 1) >> PAGE_SHIFT))
3878 dio_warn_stale_pagecache(iocb->ki_filp);
3882 generic_file_direct_write(struct kiocb *iocb, struct iov_iter *from)
3884 struct address_space *mapping = iocb->ki_filp->f_mapping;
3885 size_t write_len = iov_iter_count(from);
3889 * If a page can not be invalidated, return 0 to fall back
3890 * to buffered write.
3892 written = kiocb_invalidate_pages(iocb, write_len);
3894 if (written == -EBUSY)
3899 written = mapping->a_ops->direct_IO(iocb, from);
3902 * Finally, try again to invalidate clean pages which might have been
3903 * cached by non-direct readahead, or faulted in by get_user_pages()
3904 * if the source of the write was an mmap'ed region of the file
3905 * we're writing. Either one is a pretty crazy thing to do,
3906 * so we don't support it 100%. If this invalidation
3907 * fails, tough, the write still worked...
3909 * Most of the time we do not need this since dio_complete() will do
3910 * the invalidation for us. However there are some file systems that
3911 * do not end up with dio_complete() being called, so let's not break
3912 * them by removing it completely.
3914 * Noticeable example is a blkdev_direct_IO().
3916 * Skip invalidation for async writes or if mapping has no pages.
3919 struct inode *inode = mapping->host;
3920 loff_t pos = iocb->ki_pos;
3922 kiocb_invalidate_post_direct_write(iocb, written);
3924 write_len -= written;
3925 if (pos > i_size_read(inode) && !S_ISBLK(inode->i_mode)) {
3926 i_size_write(inode, pos);
3927 mark_inode_dirty(inode);
3931 if (written != -EIOCBQUEUED)
3932 iov_iter_revert(from, write_len - iov_iter_count(from));
3935 EXPORT_SYMBOL(generic_file_direct_write);
3937 ssize_t generic_perform_write(struct kiocb *iocb, struct iov_iter *i)
3939 struct file *file = iocb->ki_filp;
3940 loff_t pos = iocb->ki_pos;
3941 struct address_space *mapping = file->f_mapping;
3942 const struct address_space_operations *a_ops = mapping->a_ops;
3944 ssize_t written = 0;
3948 unsigned long offset; /* Offset into pagecache page */
3949 unsigned long bytes; /* Bytes to write to page */
3950 size_t copied; /* Bytes copied from user */
3951 void *fsdata = NULL;
3953 offset = (pos & (PAGE_SIZE - 1));
3954 bytes = min_t(unsigned long, PAGE_SIZE - offset,
3959 * Bring in the user page that we will copy from _first_.
3960 * Otherwise there's a nasty deadlock on copying from the
3961 * same page as we're writing to, without it being marked
3964 if (unlikely(fault_in_iov_iter_readable(i, bytes) == bytes)) {
3969 if (fatal_signal_pending(current)) {
3974 status = a_ops->write_begin(file, mapping, pos, bytes,
3976 if (unlikely(status < 0))
3979 if (mapping_writably_mapped(mapping))
3980 flush_dcache_page(page);
3982 copied = copy_page_from_iter_atomic(page, offset, bytes, i);
3983 flush_dcache_page(page);
3985 status = a_ops->write_end(file, mapping, pos, bytes, copied,
3987 if (unlikely(status != copied)) {
3988 iov_iter_revert(i, copied - max(status, 0L));
3989 if (unlikely(status < 0))
3994 if (unlikely(status == 0)) {
3996 * A short copy made ->write_end() reject the
3997 * thing entirely. Might be memory poisoning
3998 * halfway through, might be a race with munmap,
3999 * might be severe memory pressure.
4008 balance_dirty_pages_ratelimited(mapping);
4009 } while (iov_iter_count(i));
4013 iocb->ki_pos += written;
4016 EXPORT_SYMBOL(generic_perform_write);
4019 * __generic_file_write_iter - write data to a file
4020 * @iocb: IO state structure (file, offset, etc.)
4021 * @from: iov_iter with data to write
4023 * This function does all the work needed for actually writing data to a
4024 * file. It does all basic checks, removes SUID from the file, updates
4025 * modification times and calls proper subroutines depending on whether we
4026 * do direct IO or a standard buffered write.
4028 * It expects i_rwsem to be grabbed unless we work on a block device or similar
4029 * object which does not need locking at all.
4031 * This function does *not* take care of syncing data in case of O_SYNC write.
4032 * A caller has to handle it. This is mainly due to the fact that we want to
4033 * avoid syncing under i_rwsem.
4036 * * number of bytes written, even for truncated writes
4037 * * negative error code if no data has been written at all
4039 ssize_t __generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
4041 struct file *file = iocb->ki_filp;
4042 struct address_space *mapping = file->f_mapping;
4043 struct inode *inode = mapping->host;
4046 ret = file_remove_privs(file);
4050 ret = file_update_time(file);
4054 if (iocb->ki_flags & IOCB_DIRECT) {
4055 ret = generic_file_direct_write(iocb, from);
4057 * If the write stopped short of completing, fall back to
4058 * buffered writes. Some filesystems do this for writes to
4059 * holes, for example. For DAX files, a buffered write will
4060 * not succeed (even if it did, DAX does not handle dirty
4061 * page-cache pages correctly).
4063 if (ret < 0 || !iov_iter_count(from) || IS_DAX(inode))
4065 return direct_write_fallback(iocb, from, ret,
4066 generic_perform_write(iocb, from));
4069 return generic_perform_write(iocb, from);
4071 EXPORT_SYMBOL(__generic_file_write_iter);
4074 * generic_file_write_iter - write data to a file
4075 * @iocb: IO state structure
4076 * @from: iov_iter with data to write
4078 * This is a wrapper around __generic_file_write_iter() to be used by most
4079 * filesystems. It takes care of syncing the file in case of O_SYNC file
4080 * and acquires i_rwsem as needed.
4082 * * negative error code if no data has been written at all of
4083 * vfs_fsync_range() failed for a synchronous write
4084 * * number of bytes written, even for truncated writes
4086 ssize_t generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
4088 struct file *file = iocb->ki_filp;
4089 struct inode *inode = file->f_mapping->host;
4093 ret = generic_write_checks(iocb, from);
4095 ret = __generic_file_write_iter(iocb, from);
4096 inode_unlock(inode);
4099 ret = generic_write_sync(iocb, ret);
4102 EXPORT_SYMBOL(generic_file_write_iter);
4105 * filemap_release_folio() - Release fs-specific metadata on a folio.
4106 * @folio: The folio which the kernel is trying to free.
4107 * @gfp: Memory allocation flags (and I/O mode).
4109 * The address_space is trying to release any data attached to a folio
4110 * (presumably at folio->private).
4112 * This will also be called if the private_2 flag is set on a page,
4113 * indicating that the folio has other metadata associated with it.
4115 * The @gfp argument specifies whether I/O may be performed to release
4116 * this page (__GFP_IO), and whether the call may block
4117 * (__GFP_RECLAIM & __GFP_FS).
4119 * Return: %true if the release was successful, otherwise %false.
4121 bool filemap_release_folio(struct folio *folio, gfp_t gfp)
4123 struct address_space * const mapping = folio->mapping;
4125 BUG_ON(!folio_test_locked(folio));
4126 if (!folio_needs_release(folio))
4128 if (folio_test_writeback(folio))
4131 if (mapping && mapping->a_ops->release_folio)
4132 return mapping->a_ops->release_folio(folio, gfp);
4133 return try_to_free_buffers(folio);
4135 EXPORT_SYMBOL(filemap_release_folio);
4137 #ifdef CONFIG_CACHESTAT_SYSCALL
4139 * filemap_cachestat() - compute the page cache statistics of a mapping
4140 * @mapping: The mapping to compute the statistics for.
4141 * @first_index: The starting page cache index.
4142 * @last_index: The final page index (inclusive).
4143 * @cs: the cachestat struct to write the result to.
4145 * This will query the page cache statistics of a mapping in the
4146 * page range of [first_index, last_index] (inclusive). The statistics
4147 * queried include: number of dirty pages, number of pages marked for
4148 * writeback, and the number of (recently) evicted pages.
4150 static void filemap_cachestat(struct address_space *mapping,
4151 pgoff_t first_index, pgoff_t last_index, struct cachestat *cs)
4153 XA_STATE(xas, &mapping->i_pages, first_index);
4154 struct folio *folio;
4157 xas_for_each(&xas, folio, last_index) {
4159 unsigned long nr_pages;
4160 pgoff_t folio_first_index, folio_last_index;
4163 * Don't deref the folio. It is not pinned, and might
4164 * get freed (and reused) underneath us.
4166 * We *could* pin it, but that would be expensive for
4167 * what should be a fast and lightweight syscall.
4169 * Instead, derive all information of interest from
4170 * the rcu-protected xarray.
4173 if (xas_retry(&xas, folio))
4176 order = xa_get_order(xas.xa, xas.xa_index);
4177 nr_pages = 1 << order;
4178 folio_first_index = round_down(xas.xa_index, 1 << order);
4179 folio_last_index = folio_first_index + nr_pages - 1;
4181 /* Folios might straddle the range boundaries, only count covered pages */
4182 if (folio_first_index < first_index)
4183 nr_pages -= first_index - folio_first_index;
4185 if (folio_last_index > last_index)
4186 nr_pages -= folio_last_index - last_index;
4188 if (xa_is_value(folio)) {
4189 /* page is evicted */
4190 void *shadow = (void *)folio;
4191 bool workingset; /* not used */
4193 cs->nr_evicted += nr_pages;
4195 #ifdef CONFIG_SWAP /* implies CONFIG_MMU */
4196 if (shmem_mapping(mapping)) {
4197 /* shmem file - in swap cache */
4198 swp_entry_t swp = radix_to_swp_entry(folio);
4200 shadow = get_shadow_from_swap_cache(swp);
4203 if (workingset_test_recent(shadow, true, &workingset))
4204 cs->nr_recently_evicted += nr_pages;
4209 /* page is in cache */
4210 cs->nr_cache += nr_pages;
4212 if (xas_get_mark(&xas, PAGECACHE_TAG_DIRTY))
4213 cs->nr_dirty += nr_pages;
4215 if (xas_get_mark(&xas, PAGECACHE_TAG_WRITEBACK))
4216 cs->nr_writeback += nr_pages;
4219 if (need_resched()) {
4228 * The cachestat(2) system call.
4230 * cachestat() returns the page cache statistics of a file in the
4231 * bytes range specified by `off` and `len`: number of cached pages,
4232 * number of dirty pages, number of pages marked for writeback,
4233 * number of evicted pages, and number of recently evicted pages.
4235 * An evicted page is a page that is previously in the page cache
4236 * but has been evicted since. A page is recently evicted if its last
4237 * eviction was recent enough that its reentry to the cache would
4238 * indicate that it is actively being used by the system, and that
4239 * there is memory pressure on the system.
4241 * `off` and `len` must be non-negative integers. If `len` > 0,
4242 * the queried range is [`off`, `off` + `len`]. If `len` == 0,
4243 * we will query in the range from `off` to the end of the file.
4245 * The `flags` argument is unused for now, but is included for future
4246 * extensibility. User should pass 0 (i.e no flag specified).
4248 * Currently, hugetlbfs is not supported.
4250 * Because the status of a page can change after cachestat() checks it
4251 * but before it returns to the application, the returned values may
4252 * contain stale information.
4256 * -EFAULT - cstat or cstat_range points to an illegal address
4257 * -EINVAL - invalid flags
4258 * -EBADF - invalid file descriptor
4259 * -EOPNOTSUPP - file descriptor is of a hugetlbfs file
4261 SYSCALL_DEFINE4(cachestat, unsigned int, fd,
4262 struct cachestat_range __user *, cstat_range,
4263 struct cachestat __user *, cstat, unsigned int, flags)
4265 struct fd f = fdget(fd);
4266 struct address_space *mapping;
4267 struct cachestat_range csr;
4268 struct cachestat cs;
4269 pgoff_t first_index, last_index;
4274 if (copy_from_user(&csr, cstat_range,
4275 sizeof(struct cachestat_range))) {
4280 /* hugetlbfs is not supported */
4281 if (is_file_hugepages(f.file)) {
4291 first_index = csr.off >> PAGE_SHIFT;
4293 csr.len == 0 ? ULONG_MAX : (csr.off + csr.len - 1) >> PAGE_SHIFT;
4294 memset(&cs, 0, sizeof(struct cachestat));
4295 mapping = f.file->f_mapping;
4296 filemap_cachestat(mapping, first_index, last_index, &cs);
4299 if (copy_to_user(cstat, &cs, sizeof(struct cachestat)))
4304 #endif /* CONFIG_CACHESTAT_SYSCALL */