4 * Copyright (C) 1994-1999 Linus Torvalds
8 * This file handles the generic file mmap semantics used by
9 * most "normal" filesystems (but you don't /have/ to use this:
10 * the NFS filesystem used to do this differently, for example)
12 #include <linux/export.h>
13 #include <linux/compiler.h>
14 #include <linux/dax.h>
16 #include <linux/sched/signal.h>
17 #include <linux/uaccess.h>
18 #include <linux/capability.h>
19 #include <linux/kernel_stat.h>
20 #include <linux/gfp.h>
22 #include <linux/swap.h>
23 #include <linux/mman.h>
24 #include <linux/pagemap.h>
25 #include <linux/file.h>
26 #include <linux/uio.h>
27 #include <linux/hash.h>
28 #include <linux/writeback.h>
29 #include <linux/backing-dev.h>
30 #include <linux/pagevec.h>
31 #include <linux/blkdev.h>
32 #include <linux/security.h>
33 #include <linux/cpuset.h>
34 #include <linux/hardirq.h> /* for BUG_ON(!in_atomic()) only */
35 #include <linux/hugetlb.h>
36 #include <linux/memcontrol.h>
37 #include <linux/cleancache.h>
38 #include <linux/rmap.h>
41 #define CREATE_TRACE_POINTS
42 #include <trace/events/filemap.h>
45 * FIXME: remove all knowledge of the buffer layer from the core VM
47 #include <linux/buffer_head.h> /* for try_to_free_buffers */
52 * Shared mappings implemented 30.11.1994. It's not fully working yet,
55 * Shared mappings now work. 15.8.1995 Bruno.
57 * finished 'unifying' the page and buffer cache and SMP-threaded the
58 * page-cache, 21.05.1999, Ingo Molnar <mingo@redhat.com>
60 * SMP-threaded pagemap-LRU 1999, Andrea Arcangeli <andrea@suse.de>
66 * ->i_mmap_rwsem (truncate_pagecache)
67 * ->private_lock (__free_pte->__set_page_dirty_buffers)
68 * ->swap_lock (exclusive_swap_page, others)
69 * ->mapping->tree_lock
72 * ->i_mmap_rwsem (truncate->unmap_mapping_range)
76 * ->page_table_lock or pte_lock (various, mainly in memory.c)
77 * ->mapping->tree_lock (arch-dependent flush_dcache_mmap_lock)
80 * ->lock_page (access_process_vm)
82 * ->i_mutex (generic_perform_write)
83 * ->mmap_sem (fault_in_pages_readable->do_page_fault)
86 * sb_lock (fs/fs-writeback.c)
87 * ->mapping->tree_lock (__sync_single_inode)
90 * ->anon_vma.lock (vma_adjust)
93 * ->page_table_lock or pte_lock (anon_vma_prepare and various)
95 * ->page_table_lock or pte_lock
96 * ->swap_lock (try_to_unmap_one)
97 * ->private_lock (try_to_unmap_one)
98 * ->tree_lock (try_to_unmap_one)
99 * ->zone_lru_lock(zone) (follow_page->mark_page_accessed)
100 * ->zone_lru_lock(zone) (check_pte_range->isolate_lru_page)
101 * ->private_lock (page_remove_rmap->set_page_dirty)
102 * ->tree_lock (page_remove_rmap->set_page_dirty)
103 * bdi.wb->list_lock (page_remove_rmap->set_page_dirty)
104 * ->inode->i_lock (page_remove_rmap->set_page_dirty)
105 * ->memcg->move_lock (page_remove_rmap->lock_page_memcg)
106 * bdi.wb->list_lock (zap_pte_range->set_page_dirty)
107 * ->inode->i_lock (zap_pte_range->set_page_dirty)
108 * ->private_lock (zap_pte_range->__set_page_dirty_buffers)
111 * ->tasklist_lock (memory_failure, collect_procs_ao)
114 static int page_cache_tree_insert(struct address_space *mapping,
115 struct page *page, void **shadowp)
117 struct radix_tree_node *node;
121 error = __radix_tree_create(&mapping->page_tree, page->index, 0,
128 p = radix_tree_deref_slot_protected(slot, &mapping->tree_lock);
129 if (!radix_tree_exceptional_entry(p))
132 mapping->nrexceptional--;
133 if (!dax_mapping(mapping)) {
137 /* DAX can replace empty locked entry with a hole */
139 dax_radix_locked_entry(0, RADIX_DAX_EMPTY));
140 /* Wakeup waiters for exceptional entry lock */
141 dax_wake_mapping_entry_waiter(mapping, page->index, p,
145 __radix_tree_replace(&mapping->page_tree, node, slot, page,
146 workingset_update_node, mapping);
151 static void page_cache_tree_delete(struct address_space *mapping,
152 struct page *page, void *shadow)
156 /* hugetlb pages are represented by one entry in the radix tree */
157 nr = PageHuge(page) ? 1 : hpage_nr_pages(page);
159 VM_BUG_ON_PAGE(!PageLocked(page), page);
160 VM_BUG_ON_PAGE(PageTail(page), page);
161 VM_BUG_ON_PAGE(nr != 1 && shadow, page);
163 for (i = 0; i < nr; i++) {
164 struct radix_tree_node *node;
167 __radix_tree_lookup(&mapping->page_tree, page->index + i,
170 VM_BUG_ON_PAGE(!node && nr != 1, page);
172 radix_tree_clear_tags(&mapping->page_tree, node, slot);
173 __radix_tree_replace(&mapping->page_tree, node, slot, shadow,
174 workingset_update_node, mapping);
178 mapping->nrexceptional += nr;
180 * Make sure the nrexceptional update is committed before
181 * the nrpages update so that final truncate racing
182 * with reclaim does not see both counters 0 at the
183 * same time and miss a shadow entry.
187 mapping->nrpages -= nr;
191 * Delete a page from the page cache and free it. Caller has to make
192 * sure the page is locked and that nobody else uses it - or that usage
193 * is safe. The caller must hold the mapping's tree_lock.
195 void __delete_from_page_cache(struct page *page, void *shadow)
197 struct address_space *mapping = page->mapping;
198 int nr = hpage_nr_pages(page);
200 trace_mm_filemap_delete_from_page_cache(page);
202 * if we're uptodate, flush out into the cleancache, otherwise
203 * invalidate any existing cleancache entries. We can't leave
204 * stale data around in the cleancache once our page is gone
206 if (PageUptodate(page) && PageMappedToDisk(page))
207 cleancache_put_page(page);
209 cleancache_invalidate_page(mapping, page);
211 VM_BUG_ON_PAGE(PageTail(page), page);
212 VM_BUG_ON_PAGE(page_mapped(page), page);
213 if (!IS_ENABLED(CONFIG_DEBUG_VM) && unlikely(page_mapped(page))) {
216 pr_alert("BUG: Bad page cache in process %s pfn:%05lx\n",
217 current->comm, page_to_pfn(page));
218 dump_page(page, "still mapped when deleted");
220 add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE);
222 mapcount = page_mapcount(page);
223 if (mapping_exiting(mapping) &&
224 page_count(page) >= mapcount + 2) {
226 * All vmas have already been torn down, so it's
227 * a good bet that actually the page is unmapped,
228 * and we'd prefer not to leak it: if we're wrong,
229 * some other bad page check should catch it later.
231 page_mapcount_reset(page);
232 page_ref_sub(page, mapcount);
236 page_cache_tree_delete(mapping, page, shadow);
238 page->mapping = NULL;
239 /* Leave page->index set: truncation lookup relies upon it */
241 /* hugetlb pages do not participate in page cache accounting. */
245 __mod_node_page_state(page_pgdat(page), NR_FILE_PAGES, -nr);
246 if (PageSwapBacked(page)) {
247 __mod_node_page_state(page_pgdat(page), NR_SHMEM, -nr);
248 if (PageTransHuge(page))
249 __dec_node_page_state(page, NR_SHMEM_THPS);
251 VM_BUG_ON_PAGE(PageTransHuge(page), page);
255 * At this point page must be either written or cleaned by truncate.
256 * Dirty page here signals a bug and loss of unwritten data.
258 * This fixes dirty accounting after removing the page entirely but
259 * leaves PageDirty set: it has no effect for truncated page and
260 * anyway will be cleared before returning page into buddy allocator.
262 if (WARN_ON_ONCE(PageDirty(page)))
263 account_page_cleaned(page, mapping, inode_to_wb(mapping->host));
267 * delete_from_page_cache - delete page from page cache
268 * @page: the page which the kernel is trying to remove from page cache
270 * This must be called only on pages that have been verified to be in the page
271 * cache and locked. It will never put the page into the free list, the caller
272 * has a reference on the page.
274 void delete_from_page_cache(struct page *page)
276 struct address_space *mapping = page_mapping(page);
278 void (*freepage)(struct page *);
280 BUG_ON(!PageLocked(page));
282 freepage = mapping->a_ops->freepage;
284 spin_lock_irqsave(&mapping->tree_lock, flags);
285 __delete_from_page_cache(page, NULL);
286 spin_unlock_irqrestore(&mapping->tree_lock, flags);
291 if (PageTransHuge(page) && !PageHuge(page)) {
292 page_ref_sub(page, HPAGE_PMD_NR);
293 VM_BUG_ON_PAGE(page_count(page) <= 0, page);
298 EXPORT_SYMBOL(delete_from_page_cache);
300 int filemap_check_errors(struct address_space *mapping)
303 /* Check for outstanding write errors */
304 if (test_bit(AS_ENOSPC, &mapping->flags) &&
305 test_and_clear_bit(AS_ENOSPC, &mapping->flags))
307 if (test_bit(AS_EIO, &mapping->flags) &&
308 test_and_clear_bit(AS_EIO, &mapping->flags))
312 EXPORT_SYMBOL(filemap_check_errors);
314 static int filemap_check_and_keep_errors(struct address_space *mapping)
316 /* Check for outstanding write errors */
317 if (test_bit(AS_EIO, &mapping->flags))
319 if (test_bit(AS_ENOSPC, &mapping->flags))
325 * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range
326 * @mapping: address space structure to write
327 * @start: offset in bytes where the range starts
328 * @end: offset in bytes where the range ends (inclusive)
329 * @sync_mode: enable synchronous operation
331 * Start writeback against all of a mapping's dirty pages that lie
332 * within the byte offsets <start, end> inclusive.
334 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
335 * opposed to a regular memory cleansing writeback. The difference between
336 * these two operations is that if a dirty page/buffer is encountered, it must
337 * be waited upon, and not just skipped over.
339 int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
340 loff_t end, int sync_mode)
343 struct writeback_control wbc = {
344 .sync_mode = sync_mode,
345 .nr_to_write = LONG_MAX,
346 .range_start = start,
350 if (!mapping_cap_writeback_dirty(mapping))
353 wbc_attach_fdatawrite_inode(&wbc, mapping->host);
354 ret = do_writepages(mapping, &wbc);
355 wbc_detach_inode(&wbc);
359 static inline int __filemap_fdatawrite(struct address_space *mapping,
362 return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode);
365 int filemap_fdatawrite(struct address_space *mapping)
367 return __filemap_fdatawrite(mapping, WB_SYNC_ALL);
369 EXPORT_SYMBOL(filemap_fdatawrite);
371 int filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
374 return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL);
376 EXPORT_SYMBOL(filemap_fdatawrite_range);
379 * filemap_flush - mostly a non-blocking flush
380 * @mapping: target address_space
382 * This is a mostly non-blocking flush. Not suitable for data-integrity
383 * purposes - I/O may not be started against all dirty pages.
385 int filemap_flush(struct address_space *mapping)
387 return __filemap_fdatawrite(mapping, WB_SYNC_NONE);
389 EXPORT_SYMBOL(filemap_flush);
392 * filemap_range_has_page - check if a page exists in range.
393 * @mapping: address space within which to check
394 * @start_byte: offset in bytes where the range starts
395 * @end_byte: offset in bytes where the range ends (inclusive)
397 * Find at least one page in the range supplied, usually used to check if
398 * direct writing in this range will trigger a writeback.
400 bool filemap_range_has_page(struct address_space *mapping,
401 loff_t start_byte, loff_t end_byte)
403 pgoff_t index = start_byte >> PAGE_SHIFT;
404 pgoff_t end = end_byte >> PAGE_SHIFT;
408 if (end_byte < start_byte)
411 if (mapping->nrpages == 0)
414 pagevec_init(&pvec, 0);
415 if (!pagevec_lookup(&pvec, mapping, index, 1))
417 ret = (pvec.pages[0]->index <= end);
418 pagevec_release(&pvec);
421 EXPORT_SYMBOL(filemap_range_has_page);
423 static void __filemap_fdatawait_range(struct address_space *mapping,
424 loff_t start_byte, loff_t end_byte)
426 pgoff_t index = start_byte >> PAGE_SHIFT;
427 pgoff_t end = end_byte >> PAGE_SHIFT;
431 if (end_byte < start_byte)
434 pagevec_init(&pvec, 0);
435 while ((index <= end) &&
436 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
437 PAGECACHE_TAG_WRITEBACK,
438 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1)) != 0) {
441 for (i = 0; i < nr_pages; i++) {
442 struct page *page = pvec.pages[i];
444 /* until radix tree lookup accepts end_index */
445 if (page->index > end)
448 wait_on_page_writeback(page);
449 ClearPageError(page);
451 pagevec_release(&pvec);
457 * filemap_fdatawait_range - wait for writeback to complete
458 * @mapping: address space structure to wait for
459 * @start_byte: offset in bytes where the range starts
460 * @end_byte: offset in bytes where the range ends (inclusive)
462 * Walk the list of under-writeback pages of the given address space
463 * in the given range and wait for all of them. Check error status of
464 * the address space and return it.
466 * Since the error status of the address space is cleared by this function,
467 * callers are responsible for checking the return value and handling and/or
468 * reporting the error.
470 int filemap_fdatawait_range(struct address_space *mapping, loff_t start_byte,
473 __filemap_fdatawait_range(mapping, start_byte, end_byte);
474 return filemap_check_errors(mapping);
476 EXPORT_SYMBOL(filemap_fdatawait_range);
479 * file_fdatawait_range - wait for writeback to complete
480 * @file: file pointing to address space structure to wait for
481 * @start_byte: offset in bytes where the range starts
482 * @end_byte: offset in bytes where the range ends (inclusive)
484 * Walk the list of under-writeback pages of the address space that file
485 * refers to, in the given range and wait for all of them. Check error
486 * status of the address space vs. the file->f_wb_err cursor and return it.
488 * Since the error status of the file is advanced by this function,
489 * callers are responsible for checking the return value and handling and/or
490 * reporting the error.
492 int file_fdatawait_range(struct file *file, loff_t start_byte, loff_t end_byte)
494 struct address_space *mapping = file->f_mapping;
496 __filemap_fdatawait_range(mapping, start_byte, end_byte);
497 return file_check_and_advance_wb_err(file);
499 EXPORT_SYMBOL(file_fdatawait_range);
502 * filemap_fdatawait_keep_errors - wait for writeback without clearing errors
503 * @mapping: address space structure to wait for
505 * Walk the list of under-writeback pages of the given address space
506 * and wait for all of them. Unlike filemap_fdatawait(), this function
507 * does not clear error status of the address space.
509 * Use this function if callers don't handle errors themselves. Expected
510 * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2),
513 int filemap_fdatawait_keep_errors(struct address_space *mapping)
515 loff_t i_size = i_size_read(mapping->host);
520 __filemap_fdatawait_range(mapping, 0, i_size - 1);
521 return filemap_check_and_keep_errors(mapping);
523 EXPORT_SYMBOL(filemap_fdatawait_keep_errors);
526 * filemap_fdatawait - wait for all under-writeback pages to complete
527 * @mapping: address space structure to wait for
529 * Walk the list of under-writeback pages of the given address space
530 * and wait for all of them. Check error status of the address space
533 * Since the error status of the address space is cleared by this function,
534 * callers are responsible for checking the return value and handling and/or
535 * reporting the error.
537 int filemap_fdatawait(struct address_space *mapping)
539 loff_t i_size = i_size_read(mapping->host);
544 return filemap_fdatawait_range(mapping, 0, i_size - 1);
546 EXPORT_SYMBOL(filemap_fdatawait);
548 static bool mapping_needs_writeback(struct address_space *mapping)
550 return (!dax_mapping(mapping) && mapping->nrpages) ||
551 (dax_mapping(mapping) && mapping->nrexceptional);
554 int filemap_write_and_wait(struct address_space *mapping)
558 if (mapping_needs_writeback(mapping)) {
559 err = filemap_fdatawrite(mapping);
561 * Even if the above returned error, the pages may be
562 * written partially (e.g. -ENOSPC), so we wait for it.
563 * But the -EIO is special case, it may indicate the worst
564 * thing (e.g. bug) happened, so we avoid waiting for it.
567 int err2 = filemap_fdatawait(mapping);
571 /* Clear any previously stored errors */
572 filemap_check_errors(mapping);
575 err = filemap_check_errors(mapping);
579 EXPORT_SYMBOL(filemap_write_and_wait);
582 * filemap_write_and_wait_range - write out & wait on a file range
583 * @mapping: the address_space for the pages
584 * @lstart: offset in bytes where the range starts
585 * @lend: offset in bytes where the range ends (inclusive)
587 * Write out and wait upon file offsets lstart->lend, inclusive.
589 * Note that @lend is inclusive (describes the last byte to be written) so
590 * that this function can be used to write to the very end-of-file (end = -1).
592 int filemap_write_and_wait_range(struct address_space *mapping,
593 loff_t lstart, loff_t lend)
597 if (mapping_needs_writeback(mapping)) {
598 err = __filemap_fdatawrite_range(mapping, lstart, lend,
600 /* See comment of filemap_write_and_wait() */
602 int err2 = filemap_fdatawait_range(mapping,
607 /* Clear any previously stored errors */
608 filemap_check_errors(mapping);
611 err = filemap_check_errors(mapping);
615 EXPORT_SYMBOL(filemap_write_and_wait_range);
617 void __filemap_set_wb_err(struct address_space *mapping, int err)
619 errseq_t eseq = errseq_set(&mapping->wb_err, err);
621 trace_filemap_set_wb_err(mapping, eseq);
623 EXPORT_SYMBOL(__filemap_set_wb_err);
626 * file_check_and_advance_wb_err - report wb error (if any) that was previously
627 * and advance wb_err to current one
628 * @file: struct file on which the error is being reported
630 * When userland calls fsync (or something like nfsd does the equivalent), we
631 * want to report any writeback errors that occurred since the last fsync (or
632 * since the file was opened if there haven't been any).
634 * Grab the wb_err from the mapping. If it matches what we have in the file,
635 * then just quickly return 0. The file is all caught up.
637 * If it doesn't match, then take the mapping value, set the "seen" flag in
638 * it and try to swap it into place. If it works, or another task beat us
639 * to it with the new value, then update the f_wb_err and return the error
640 * portion. The error at this point must be reported via proper channels
641 * (a'la fsync, or NFS COMMIT operation, etc.).
643 * While we handle mapping->wb_err with atomic operations, the f_wb_err
644 * value is protected by the f_lock since we must ensure that it reflects
645 * the latest value swapped in for this file descriptor.
647 int file_check_and_advance_wb_err(struct file *file)
650 errseq_t old = READ_ONCE(file->f_wb_err);
651 struct address_space *mapping = file->f_mapping;
653 /* Locklessly handle the common case where nothing has changed */
654 if (errseq_check(&mapping->wb_err, old)) {
655 /* Something changed, must use slow path */
656 spin_lock(&file->f_lock);
657 old = file->f_wb_err;
658 err = errseq_check_and_advance(&mapping->wb_err,
660 trace_file_check_and_advance_wb_err(file, old);
661 spin_unlock(&file->f_lock);
665 EXPORT_SYMBOL(file_check_and_advance_wb_err);
668 * file_write_and_wait_range - write out & wait on a file range
669 * @file: file pointing to address_space with pages
670 * @lstart: offset in bytes where the range starts
671 * @lend: offset in bytes where the range ends (inclusive)
673 * Write out and wait upon file offsets lstart->lend, inclusive.
675 * Note that @lend is inclusive (describes the last byte to be written) so
676 * that this function can be used to write to the very end-of-file (end = -1).
678 * After writing out and waiting on the data, we check and advance the
679 * f_wb_err cursor to the latest value, and return any errors detected there.
681 int file_write_and_wait_range(struct file *file, loff_t lstart, loff_t lend)
684 struct address_space *mapping = file->f_mapping;
686 if (mapping_needs_writeback(mapping)) {
687 err = __filemap_fdatawrite_range(mapping, lstart, lend,
689 /* See comment of filemap_write_and_wait() */
691 __filemap_fdatawait_range(mapping, lstart, lend);
693 err2 = file_check_and_advance_wb_err(file);
698 EXPORT_SYMBOL(file_write_and_wait_range);
701 * replace_page_cache_page - replace a pagecache page with a new one
702 * @old: page to be replaced
703 * @new: page to replace with
704 * @gfp_mask: allocation mode
706 * This function replaces a page in the pagecache with a new one. On
707 * success it acquires the pagecache reference for the new page and
708 * drops it for the old page. Both the old and new pages must be
709 * locked. This function does not add the new page to the LRU, the
710 * caller must do that.
712 * The remove + add is atomic. The only way this function can fail is
713 * memory allocation failure.
715 int replace_page_cache_page(struct page *old, struct page *new, gfp_t gfp_mask)
719 VM_BUG_ON_PAGE(!PageLocked(old), old);
720 VM_BUG_ON_PAGE(!PageLocked(new), new);
721 VM_BUG_ON_PAGE(new->mapping, new);
723 error = radix_tree_preload(gfp_mask & ~__GFP_HIGHMEM);
725 struct address_space *mapping = old->mapping;
726 void (*freepage)(struct page *);
729 pgoff_t offset = old->index;
730 freepage = mapping->a_ops->freepage;
733 new->mapping = mapping;
736 spin_lock_irqsave(&mapping->tree_lock, flags);
737 __delete_from_page_cache(old, NULL);
738 error = page_cache_tree_insert(mapping, new, NULL);
742 * hugetlb pages do not participate in page cache accounting.
745 __inc_node_page_state(new, NR_FILE_PAGES);
746 if (PageSwapBacked(new))
747 __inc_node_page_state(new, NR_SHMEM);
748 spin_unlock_irqrestore(&mapping->tree_lock, flags);
749 mem_cgroup_migrate(old, new);
750 radix_tree_preload_end();
758 EXPORT_SYMBOL_GPL(replace_page_cache_page);
760 static int __add_to_page_cache_locked(struct page *page,
761 struct address_space *mapping,
762 pgoff_t offset, gfp_t gfp_mask,
765 int huge = PageHuge(page);
766 struct mem_cgroup *memcg;
769 VM_BUG_ON_PAGE(!PageLocked(page), page);
770 VM_BUG_ON_PAGE(PageSwapBacked(page), page);
773 error = mem_cgroup_try_charge(page, current->mm,
774 gfp_mask, &memcg, false);
779 error = radix_tree_maybe_preload(gfp_mask & ~__GFP_HIGHMEM);
782 mem_cgroup_cancel_charge(page, memcg, false);
787 page->mapping = mapping;
788 page->index = offset;
790 spin_lock_irq(&mapping->tree_lock);
791 error = page_cache_tree_insert(mapping, page, shadowp);
792 radix_tree_preload_end();
796 /* hugetlb pages do not participate in page cache accounting. */
798 __inc_node_page_state(page, NR_FILE_PAGES);
799 spin_unlock_irq(&mapping->tree_lock);
801 mem_cgroup_commit_charge(page, memcg, false, false);
802 trace_mm_filemap_add_to_page_cache(page);
805 page->mapping = NULL;
806 /* Leave page->index set: truncation relies upon it */
807 spin_unlock_irq(&mapping->tree_lock);
809 mem_cgroup_cancel_charge(page, memcg, false);
815 * add_to_page_cache_locked - add a locked page to the pagecache
817 * @mapping: the page's address_space
818 * @offset: page index
819 * @gfp_mask: page allocation mode
821 * This function is used to add a page to the pagecache. It must be locked.
822 * This function does not add the page to the LRU. The caller must do that.
824 int add_to_page_cache_locked(struct page *page, struct address_space *mapping,
825 pgoff_t offset, gfp_t gfp_mask)
827 return __add_to_page_cache_locked(page, mapping, offset,
830 EXPORT_SYMBOL(add_to_page_cache_locked);
832 int add_to_page_cache_lru(struct page *page, struct address_space *mapping,
833 pgoff_t offset, gfp_t gfp_mask)
838 __SetPageLocked(page);
839 ret = __add_to_page_cache_locked(page, mapping, offset,
842 __ClearPageLocked(page);
845 * The page might have been evicted from cache only
846 * recently, in which case it should be activated like
847 * any other repeatedly accessed page.
848 * The exception is pages getting rewritten; evicting other
849 * data from the working set, only to cache data that will
850 * get overwritten with something else, is a waste of memory.
852 if (!(gfp_mask & __GFP_WRITE) &&
853 shadow && workingset_refault(shadow)) {
855 workingset_activation(page);
857 ClearPageActive(page);
862 EXPORT_SYMBOL_GPL(add_to_page_cache_lru);
865 struct page *__page_cache_alloc(gfp_t gfp)
870 if (cpuset_do_page_mem_spread()) {
871 unsigned int cpuset_mems_cookie;
873 cpuset_mems_cookie = read_mems_allowed_begin();
874 n = cpuset_mem_spread_node();
875 page = __alloc_pages_node(n, gfp, 0);
876 } while (!page && read_mems_allowed_retry(cpuset_mems_cookie));
880 return alloc_pages(gfp, 0);
882 EXPORT_SYMBOL(__page_cache_alloc);
886 * In order to wait for pages to become available there must be
887 * waitqueues associated with pages. By using a hash table of
888 * waitqueues where the bucket discipline is to maintain all
889 * waiters on the same queue and wake all when any of the pages
890 * become available, and for the woken contexts to check to be
891 * sure the appropriate page became available, this saves space
892 * at a cost of "thundering herd" phenomena during rare hash
895 #define PAGE_WAIT_TABLE_BITS 8
896 #define PAGE_WAIT_TABLE_SIZE (1 << PAGE_WAIT_TABLE_BITS)
897 static wait_queue_head_t page_wait_table[PAGE_WAIT_TABLE_SIZE] __cacheline_aligned;
899 static wait_queue_head_t *page_waitqueue(struct page *page)
901 return &page_wait_table[hash_ptr(page, PAGE_WAIT_TABLE_BITS)];
904 void __init pagecache_init(void)
908 for (i = 0; i < PAGE_WAIT_TABLE_SIZE; i++)
909 init_waitqueue_head(&page_wait_table[i]);
911 page_writeback_init();
914 struct wait_page_key {
920 struct wait_page_queue {
923 wait_queue_entry_t wait;
926 static int wake_page_function(wait_queue_entry_t *wait, unsigned mode, int sync, void *arg)
928 struct wait_page_key *key = arg;
929 struct wait_page_queue *wait_page
930 = container_of(wait, struct wait_page_queue, wait);
932 if (wait_page->page != key->page)
936 if (wait_page->bit_nr != key->bit_nr)
938 if (test_bit(key->bit_nr, &key->page->flags))
941 return autoremove_wake_function(wait, mode, sync, key);
944 static void wake_up_page_bit(struct page *page, int bit_nr)
946 wait_queue_head_t *q = page_waitqueue(page);
947 struct wait_page_key key;
954 spin_lock_irqsave(&q->lock, flags);
955 __wake_up_locked_key(q, TASK_NORMAL, &key);
957 * It is possible for other pages to have collided on the waitqueue
958 * hash, so in that case check for a page match. That prevents a long-
961 * It is still possible to miss a case here, when we woke page waiters
962 * and removed them from the waitqueue, but there are still other
965 if (!waitqueue_active(q) || !key.page_match) {
966 ClearPageWaiters(page);
968 * It's possible to miss clearing Waiters here, when we woke
969 * our page waiters, but the hashed waitqueue has waiters for
972 * That's okay, it's a rare case. The next waker will clear it.
975 spin_unlock_irqrestore(&q->lock, flags);
978 static void wake_up_page(struct page *page, int bit)
980 if (!PageWaiters(page))
982 wake_up_page_bit(page, bit);
985 static inline int wait_on_page_bit_common(wait_queue_head_t *q,
986 struct page *page, int bit_nr, int state, bool lock)
988 struct wait_page_queue wait_page;
989 wait_queue_entry_t *wait = &wait_page.wait;
993 wait->func = wake_page_function;
994 wait_page.page = page;
995 wait_page.bit_nr = bit_nr;
998 spin_lock_irq(&q->lock);
1000 if (likely(list_empty(&wait->entry))) {
1002 __add_wait_queue_entry_tail_exclusive(q, wait);
1004 __add_wait_queue(q, wait);
1005 SetPageWaiters(page);
1008 set_current_state(state);
1010 spin_unlock_irq(&q->lock);
1012 if (likely(test_bit(bit_nr, &page->flags))) {
1014 if (unlikely(signal_pending_state(state, current))) {
1021 if (!test_and_set_bit_lock(bit_nr, &page->flags))
1024 if (!test_bit(bit_nr, &page->flags))
1029 finish_wait(q, wait);
1032 * A signal could leave PageWaiters set. Clearing it here if
1033 * !waitqueue_active would be possible (by open-coding finish_wait),
1034 * but still fail to catch it in the case of wait hash collision. We
1035 * already can fail to clear wait hash collision cases, so don't
1036 * bother with signals either.
1042 void wait_on_page_bit(struct page *page, int bit_nr)
1044 wait_queue_head_t *q = page_waitqueue(page);
1045 wait_on_page_bit_common(q, page, bit_nr, TASK_UNINTERRUPTIBLE, false);
1047 EXPORT_SYMBOL(wait_on_page_bit);
1049 int wait_on_page_bit_killable(struct page *page, int bit_nr)
1051 wait_queue_head_t *q = page_waitqueue(page);
1052 return wait_on_page_bit_common(q, page, bit_nr, TASK_KILLABLE, false);
1056 * add_page_wait_queue - Add an arbitrary waiter to a page's wait queue
1057 * @page: Page defining the wait queue of interest
1058 * @waiter: Waiter to add to the queue
1060 * Add an arbitrary @waiter to the wait queue for the nominated @page.
1062 void add_page_wait_queue(struct page *page, wait_queue_entry_t *waiter)
1064 wait_queue_head_t *q = page_waitqueue(page);
1065 unsigned long flags;
1067 spin_lock_irqsave(&q->lock, flags);
1068 __add_wait_queue(q, waiter);
1069 SetPageWaiters(page);
1070 spin_unlock_irqrestore(&q->lock, flags);
1072 EXPORT_SYMBOL_GPL(add_page_wait_queue);
1074 #ifndef clear_bit_unlock_is_negative_byte
1077 * PG_waiters is the high bit in the same byte as PG_lock.
1079 * On x86 (and on many other architectures), we can clear PG_lock and
1080 * test the sign bit at the same time. But if the architecture does
1081 * not support that special operation, we just do this all by hand
1084 * The read of PG_waiters has to be after (or concurrently with) PG_locked
1085 * being cleared, but a memory barrier should be unneccssary since it is
1086 * in the same byte as PG_locked.
1088 static inline bool clear_bit_unlock_is_negative_byte(long nr, volatile void *mem)
1090 clear_bit_unlock(nr, mem);
1091 /* smp_mb__after_atomic(); */
1092 return test_bit(PG_waiters, mem);
1098 * unlock_page - unlock a locked page
1101 * Unlocks the page and wakes up sleepers in ___wait_on_page_locked().
1102 * Also wakes sleepers in wait_on_page_writeback() because the wakeup
1103 * mechanism between PageLocked pages and PageWriteback pages is shared.
1104 * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep.
1106 * Note that this depends on PG_waiters being the sign bit in the byte
1107 * that contains PG_locked - thus the BUILD_BUG_ON(). That allows us to
1108 * clear the PG_locked bit and test PG_waiters at the same time fairly
1109 * portably (architectures that do LL/SC can test any bit, while x86 can
1110 * test the sign bit).
1112 void unlock_page(struct page *page)
1114 BUILD_BUG_ON(PG_waiters != 7);
1115 page = compound_head(page);
1116 VM_BUG_ON_PAGE(!PageLocked(page), page);
1117 if (clear_bit_unlock_is_negative_byte(PG_locked, &page->flags))
1118 wake_up_page_bit(page, PG_locked);
1120 EXPORT_SYMBOL(unlock_page);
1123 * end_page_writeback - end writeback against a page
1126 void end_page_writeback(struct page *page)
1129 * TestClearPageReclaim could be used here but it is an atomic
1130 * operation and overkill in this particular case. Failing to
1131 * shuffle a page marked for immediate reclaim is too mild to
1132 * justify taking an atomic operation penalty at the end of
1133 * ever page writeback.
1135 if (PageReclaim(page)) {
1136 ClearPageReclaim(page);
1137 rotate_reclaimable_page(page);
1140 if (!test_clear_page_writeback(page))
1143 smp_mb__after_atomic();
1144 wake_up_page(page, PG_writeback);
1146 EXPORT_SYMBOL(end_page_writeback);
1149 * After completing I/O on a page, call this routine to update the page
1150 * flags appropriately
1152 void page_endio(struct page *page, bool is_write, int err)
1156 SetPageUptodate(page);
1158 ClearPageUptodate(page);
1164 struct address_space *mapping;
1167 mapping = page_mapping(page);
1169 mapping_set_error(mapping, err);
1171 end_page_writeback(page);
1174 EXPORT_SYMBOL_GPL(page_endio);
1177 * __lock_page - get a lock on the page, assuming we need to sleep to get it
1178 * @__page: the page to lock
1180 void __lock_page(struct page *__page)
1182 struct page *page = compound_head(__page);
1183 wait_queue_head_t *q = page_waitqueue(page);
1184 wait_on_page_bit_common(q, page, PG_locked, TASK_UNINTERRUPTIBLE, true);
1186 EXPORT_SYMBOL(__lock_page);
1188 int __lock_page_killable(struct page *__page)
1190 struct page *page = compound_head(__page);
1191 wait_queue_head_t *q = page_waitqueue(page);
1192 return wait_on_page_bit_common(q, page, PG_locked, TASK_KILLABLE, true);
1194 EXPORT_SYMBOL_GPL(__lock_page_killable);
1198 * 1 - page is locked; mmap_sem is still held.
1199 * 0 - page is not locked.
1200 * mmap_sem has been released (up_read()), unless flags had both
1201 * FAULT_FLAG_ALLOW_RETRY and FAULT_FLAG_RETRY_NOWAIT set, in
1202 * which case mmap_sem is still held.
1204 * If neither ALLOW_RETRY nor KILLABLE are set, will always return 1
1205 * with the page locked and the mmap_sem unperturbed.
1207 int __lock_page_or_retry(struct page *page, struct mm_struct *mm,
1210 if (flags & FAULT_FLAG_ALLOW_RETRY) {
1212 * CAUTION! In this case, mmap_sem is not released
1213 * even though return 0.
1215 if (flags & FAULT_FLAG_RETRY_NOWAIT)
1218 up_read(&mm->mmap_sem);
1219 if (flags & FAULT_FLAG_KILLABLE)
1220 wait_on_page_locked_killable(page);
1222 wait_on_page_locked(page);
1225 if (flags & FAULT_FLAG_KILLABLE) {
1228 ret = __lock_page_killable(page);
1230 up_read(&mm->mmap_sem);
1240 * page_cache_next_hole - find the next hole (not-present entry)
1243 * @max_scan: maximum range to search
1245 * Search the set [index, min(index+max_scan-1, MAX_INDEX)] for the
1246 * lowest indexed hole.
1248 * Returns: the index of the hole if found, otherwise returns an index
1249 * outside of the set specified (in which case 'return - index >=
1250 * max_scan' will be true). In rare cases of index wrap-around, 0 will
1253 * page_cache_next_hole may be called under rcu_read_lock. However,
1254 * like radix_tree_gang_lookup, this will not atomically search a
1255 * snapshot of the tree at a single point in time. For example, if a
1256 * hole is created at index 5, then subsequently a hole is created at
1257 * index 10, page_cache_next_hole covering both indexes may return 10
1258 * if called under rcu_read_lock.
1260 pgoff_t page_cache_next_hole(struct address_space *mapping,
1261 pgoff_t index, unsigned long max_scan)
1265 for (i = 0; i < max_scan; i++) {
1268 page = radix_tree_lookup(&mapping->page_tree, index);
1269 if (!page || radix_tree_exceptional_entry(page))
1278 EXPORT_SYMBOL(page_cache_next_hole);
1281 * page_cache_prev_hole - find the prev hole (not-present entry)
1284 * @max_scan: maximum range to search
1286 * Search backwards in the range [max(index-max_scan+1, 0), index] for
1289 * Returns: the index of the hole if found, otherwise returns an index
1290 * outside of the set specified (in which case 'index - return >=
1291 * max_scan' will be true). In rare cases of wrap-around, ULONG_MAX
1294 * page_cache_prev_hole may be called under rcu_read_lock. However,
1295 * like radix_tree_gang_lookup, this will not atomically search a
1296 * snapshot of the tree at a single point in time. For example, if a
1297 * hole is created at index 10, then subsequently a hole is created at
1298 * index 5, page_cache_prev_hole covering both indexes may return 5 if
1299 * called under rcu_read_lock.
1301 pgoff_t page_cache_prev_hole(struct address_space *mapping,
1302 pgoff_t index, unsigned long max_scan)
1306 for (i = 0; i < max_scan; i++) {
1309 page = radix_tree_lookup(&mapping->page_tree, index);
1310 if (!page || radix_tree_exceptional_entry(page))
1313 if (index == ULONG_MAX)
1319 EXPORT_SYMBOL(page_cache_prev_hole);
1322 * find_get_entry - find and get a page cache entry
1323 * @mapping: the address_space to search
1324 * @offset: the page cache index
1326 * Looks up the page cache slot at @mapping & @offset. If there is a
1327 * page cache page, it is returned with an increased refcount.
1329 * If the slot holds a shadow entry of a previously evicted page, or a
1330 * swap entry from shmem/tmpfs, it is returned.
1332 * Otherwise, %NULL is returned.
1334 struct page *find_get_entry(struct address_space *mapping, pgoff_t offset)
1337 struct page *head, *page;
1342 pagep = radix_tree_lookup_slot(&mapping->page_tree, offset);
1344 page = radix_tree_deref_slot(pagep);
1345 if (unlikely(!page))
1347 if (radix_tree_exception(page)) {
1348 if (radix_tree_deref_retry(page))
1351 * A shadow entry of a recently evicted page,
1352 * or a swap entry from shmem/tmpfs. Return
1353 * it without attempting to raise page count.
1358 head = compound_head(page);
1359 if (!page_cache_get_speculative(head))
1362 /* The page was split under us? */
1363 if (compound_head(page) != head) {
1369 * Has the page moved?
1370 * This is part of the lockless pagecache protocol. See
1371 * include/linux/pagemap.h for details.
1373 if (unlikely(page != *pagep)) {
1383 EXPORT_SYMBOL(find_get_entry);
1386 * find_lock_entry - locate, pin and lock a page cache entry
1387 * @mapping: the address_space to search
1388 * @offset: the page cache index
1390 * Looks up the page cache slot at @mapping & @offset. If there is a
1391 * page cache page, it is returned locked and with an increased
1394 * If the slot holds a shadow entry of a previously evicted page, or a
1395 * swap entry from shmem/tmpfs, it is returned.
1397 * Otherwise, %NULL is returned.
1399 * find_lock_entry() may sleep.
1401 struct page *find_lock_entry(struct address_space *mapping, pgoff_t offset)
1406 page = find_get_entry(mapping, offset);
1407 if (page && !radix_tree_exception(page)) {
1409 /* Has the page been truncated? */
1410 if (unlikely(page_mapping(page) != mapping)) {
1415 VM_BUG_ON_PAGE(page_to_pgoff(page) != offset, page);
1419 EXPORT_SYMBOL(find_lock_entry);
1422 * pagecache_get_page - find and get a page reference
1423 * @mapping: the address_space to search
1424 * @offset: the page index
1425 * @fgp_flags: PCG flags
1426 * @gfp_mask: gfp mask to use for the page cache data page allocation
1428 * Looks up the page cache slot at @mapping & @offset.
1430 * PCG flags modify how the page is returned.
1432 * @fgp_flags can be:
1434 * - FGP_ACCESSED: the page will be marked accessed
1435 * - FGP_LOCK: Page is return locked
1436 * - FGP_CREAT: If page is not present then a new page is allocated using
1437 * @gfp_mask and added to the page cache and the VM's LRU
1438 * list. The page is returned locked and with an increased
1439 * refcount. Otherwise, NULL is returned.
1441 * If FGP_LOCK or FGP_CREAT are specified then the function may sleep even
1442 * if the GFP flags specified for FGP_CREAT are atomic.
1444 * If there is a page cache page, it is returned with an increased refcount.
1446 struct page *pagecache_get_page(struct address_space *mapping, pgoff_t offset,
1447 int fgp_flags, gfp_t gfp_mask)
1452 page = find_get_entry(mapping, offset);
1453 if (radix_tree_exceptional_entry(page))
1458 if (fgp_flags & FGP_LOCK) {
1459 if (fgp_flags & FGP_NOWAIT) {
1460 if (!trylock_page(page)) {
1468 /* Has the page been truncated? */
1469 if (unlikely(page->mapping != mapping)) {
1474 VM_BUG_ON_PAGE(page->index != offset, page);
1477 if (page && (fgp_flags & FGP_ACCESSED))
1478 mark_page_accessed(page);
1481 if (!page && (fgp_flags & FGP_CREAT)) {
1483 if ((fgp_flags & FGP_WRITE) && mapping_cap_account_dirty(mapping))
1484 gfp_mask |= __GFP_WRITE;
1485 if (fgp_flags & FGP_NOFS)
1486 gfp_mask &= ~__GFP_FS;
1488 page = __page_cache_alloc(gfp_mask);
1492 if (WARN_ON_ONCE(!(fgp_flags & FGP_LOCK)))
1493 fgp_flags |= FGP_LOCK;
1495 /* Init accessed so avoid atomic mark_page_accessed later */
1496 if (fgp_flags & FGP_ACCESSED)
1497 __SetPageReferenced(page);
1499 err = add_to_page_cache_lru(page, mapping, offset,
1500 gfp_mask & GFP_RECLAIM_MASK);
1501 if (unlikely(err)) {
1511 EXPORT_SYMBOL(pagecache_get_page);
1514 * find_get_entries - gang pagecache lookup
1515 * @mapping: The address_space to search
1516 * @start: The starting page cache index
1517 * @nr_entries: The maximum number of entries
1518 * @entries: Where the resulting entries are placed
1519 * @indices: The cache indices corresponding to the entries in @entries
1521 * find_get_entries() will search for and return a group of up to
1522 * @nr_entries entries in the mapping. The entries are placed at
1523 * @entries. find_get_entries() takes a reference against any actual
1526 * The search returns a group of mapping-contiguous page cache entries
1527 * with ascending indexes. There may be holes in the indices due to
1528 * not-present pages.
1530 * Any shadow entries of evicted pages, or swap entries from
1531 * shmem/tmpfs, are included in the returned array.
1533 * find_get_entries() returns the number of pages and shadow entries
1536 unsigned find_get_entries(struct address_space *mapping,
1537 pgoff_t start, unsigned int nr_entries,
1538 struct page **entries, pgoff_t *indices)
1541 unsigned int ret = 0;
1542 struct radix_tree_iter iter;
1548 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) {
1549 struct page *head, *page;
1551 page = radix_tree_deref_slot(slot);
1552 if (unlikely(!page))
1554 if (radix_tree_exception(page)) {
1555 if (radix_tree_deref_retry(page)) {
1556 slot = radix_tree_iter_retry(&iter);
1560 * A shadow entry of a recently evicted page, a swap
1561 * entry from shmem/tmpfs or a DAX entry. Return it
1562 * without attempting to raise page count.
1567 head = compound_head(page);
1568 if (!page_cache_get_speculative(head))
1571 /* The page was split under us? */
1572 if (compound_head(page) != head) {
1577 /* Has the page moved? */
1578 if (unlikely(page != *slot)) {
1583 indices[ret] = iter.index;
1584 entries[ret] = page;
1585 if (++ret == nr_entries)
1593 * find_get_pages - gang pagecache lookup
1594 * @mapping: The address_space to search
1595 * @start: The starting page index
1596 * @nr_pages: The maximum number of pages
1597 * @pages: Where the resulting pages are placed
1599 * find_get_pages() will search for and return a group of up to
1600 * @nr_pages pages in the mapping. The pages are placed at @pages.
1601 * find_get_pages() takes a reference against the returned pages.
1603 * The search returns a group of mapping-contiguous pages with ascending
1604 * indexes. There may be holes in the indices due to not-present pages.
1606 * find_get_pages() returns the number of pages which were found.
1608 unsigned find_get_pages(struct address_space *mapping, pgoff_t start,
1609 unsigned int nr_pages, struct page **pages)
1611 struct radix_tree_iter iter;
1615 if (unlikely(!nr_pages))
1619 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) {
1620 struct page *head, *page;
1622 page = radix_tree_deref_slot(slot);
1623 if (unlikely(!page))
1626 if (radix_tree_exception(page)) {
1627 if (radix_tree_deref_retry(page)) {
1628 slot = radix_tree_iter_retry(&iter);
1632 * A shadow entry of a recently evicted page,
1633 * or a swap entry from shmem/tmpfs. Skip
1639 head = compound_head(page);
1640 if (!page_cache_get_speculative(head))
1643 /* The page was split under us? */
1644 if (compound_head(page) != head) {
1649 /* Has the page moved? */
1650 if (unlikely(page != *slot)) {
1656 if (++ret == nr_pages)
1665 * find_get_pages_contig - gang contiguous pagecache lookup
1666 * @mapping: The address_space to search
1667 * @index: The starting page index
1668 * @nr_pages: The maximum number of pages
1669 * @pages: Where the resulting pages are placed
1671 * find_get_pages_contig() works exactly like find_get_pages(), except
1672 * that the returned number of pages are guaranteed to be contiguous.
1674 * find_get_pages_contig() returns the number of pages which were found.
1676 unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index,
1677 unsigned int nr_pages, struct page **pages)
1679 struct radix_tree_iter iter;
1681 unsigned int ret = 0;
1683 if (unlikely(!nr_pages))
1687 radix_tree_for_each_contig(slot, &mapping->page_tree, &iter, index) {
1688 struct page *head, *page;
1690 page = radix_tree_deref_slot(slot);
1691 /* The hole, there no reason to continue */
1692 if (unlikely(!page))
1695 if (radix_tree_exception(page)) {
1696 if (radix_tree_deref_retry(page)) {
1697 slot = radix_tree_iter_retry(&iter);
1701 * A shadow entry of a recently evicted page,
1702 * or a swap entry from shmem/tmpfs. Stop
1703 * looking for contiguous pages.
1708 head = compound_head(page);
1709 if (!page_cache_get_speculative(head))
1712 /* The page was split under us? */
1713 if (compound_head(page) != head) {
1718 /* Has the page moved? */
1719 if (unlikely(page != *slot)) {
1725 * must check mapping and index after taking the ref.
1726 * otherwise we can get both false positives and false
1727 * negatives, which is just confusing to the caller.
1729 if (page->mapping == NULL || page_to_pgoff(page) != iter.index) {
1735 if (++ret == nr_pages)
1741 EXPORT_SYMBOL(find_get_pages_contig);
1744 * find_get_pages_tag - find and return pages that match @tag
1745 * @mapping: the address_space to search
1746 * @index: the starting page index
1747 * @tag: the tag index
1748 * @nr_pages: the maximum number of pages
1749 * @pages: where the resulting pages are placed
1751 * Like find_get_pages, except we only return pages which are tagged with
1752 * @tag. We update @index to index the next page for the traversal.
1754 unsigned find_get_pages_tag(struct address_space *mapping, pgoff_t *index,
1755 int tag, unsigned int nr_pages, struct page **pages)
1757 struct radix_tree_iter iter;
1761 if (unlikely(!nr_pages))
1765 radix_tree_for_each_tagged(slot, &mapping->page_tree,
1766 &iter, *index, tag) {
1767 struct page *head, *page;
1769 page = radix_tree_deref_slot(slot);
1770 if (unlikely(!page))
1773 if (radix_tree_exception(page)) {
1774 if (radix_tree_deref_retry(page)) {
1775 slot = radix_tree_iter_retry(&iter);
1779 * A shadow entry of a recently evicted page.
1781 * Those entries should never be tagged, but
1782 * this tree walk is lockless and the tags are
1783 * looked up in bulk, one radix tree node at a
1784 * time, so there is a sizable window for page
1785 * reclaim to evict a page we saw tagged.
1792 head = compound_head(page);
1793 if (!page_cache_get_speculative(head))
1796 /* The page was split under us? */
1797 if (compound_head(page) != head) {
1802 /* Has the page moved? */
1803 if (unlikely(page != *slot)) {
1809 if (++ret == nr_pages)
1816 *index = pages[ret - 1]->index + 1;
1820 EXPORT_SYMBOL(find_get_pages_tag);
1823 * find_get_entries_tag - find and return entries that match @tag
1824 * @mapping: the address_space to search
1825 * @start: the starting page cache index
1826 * @tag: the tag index
1827 * @nr_entries: the maximum number of entries
1828 * @entries: where the resulting entries are placed
1829 * @indices: the cache indices corresponding to the entries in @entries
1831 * Like find_get_entries, except we only return entries which are tagged with
1834 unsigned find_get_entries_tag(struct address_space *mapping, pgoff_t start,
1835 int tag, unsigned int nr_entries,
1836 struct page **entries, pgoff_t *indices)
1839 unsigned int ret = 0;
1840 struct radix_tree_iter iter;
1846 radix_tree_for_each_tagged(slot, &mapping->page_tree,
1847 &iter, start, tag) {
1848 struct page *head, *page;
1850 page = radix_tree_deref_slot(slot);
1851 if (unlikely(!page))
1853 if (radix_tree_exception(page)) {
1854 if (radix_tree_deref_retry(page)) {
1855 slot = radix_tree_iter_retry(&iter);
1860 * A shadow entry of a recently evicted page, a swap
1861 * entry from shmem/tmpfs or a DAX entry. Return it
1862 * without attempting to raise page count.
1867 head = compound_head(page);
1868 if (!page_cache_get_speculative(head))
1871 /* The page was split under us? */
1872 if (compound_head(page) != head) {
1877 /* Has the page moved? */
1878 if (unlikely(page != *slot)) {
1883 indices[ret] = iter.index;
1884 entries[ret] = page;
1885 if (++ret == nr_entries)
1891 EXPORT_SYMBOL(find_get_entries_tag);
1894 * CD/DVDs are error prone. When a medium error occurs, the driver may fail
1895 * a _large_ part of the i/o request. Imagine the worst scenario:
1897 * ---R__________________________________________B__________
1898 * ^ reading here ^ bad block(assume 4k)
1900 * read(R) => miss => readahead(R...B) => media error => frustrating retries
1901 * => failing the whole request => read(R) => read(R+1) =>
1902 * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
1903 * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
1904 * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
1906 * It is going insane. Fix it by quickly scaling down the readahead size.
1908 static void shrink_readahead_size_eio(struct file *filp,
1909 struct file_ra_state *ra)
1915 * do_generic_file_read - generic file read routine
1916 * @filp: the file to read
1917 * @ppos: current file position
1918 * @iter: data destination
1919 * @written: already copied
1921 * This is a generic file read routine, and uses the
1922 * mapping->a_ops->readpage() function for the actual low-level stuff.
1924 * This is really ugly. But the goto's actually try to clarify some
1925 * of the logic when it comes to error handling etc.
1927 static ssize_t do_generic_file_read(struct file *filp, loff_t *ppos,
1928 struct iov_iter *iter, ssize_t written)
1930 struct address_space *mapping = filp->f_mapping;
1931 struct inode *inode = mapping->host;
1932 struct file_ra_state *ra = &filp->f_ra;
1936 unsigned long offset; /* offset into pagecache page */
1937 unsigned int prev_offset;
1940 if (unlikely(*ppos >= inode->i_sb->s_maxbytes))
1942 iov_iter_truncate(iter, inode->i_sb->s_maxbytes);
1944 index = *ppos >> PAGE_SHIFT;
1945 prev_index = ra->prev_pos >> PAGE_SHIFT;
1946 prev_offset = ra->prev_pos & (PAGE_SIZE-1);
1947 last_index = (*ppos + iter->count + PAGE_SIZE-1) >> PAGE_SHIFT;
1948 offset = *ppos & ~PAGE_MASK;
1954 unsigned long nr, ret;
1958 if (fatal_signal_pending(current)) {
1963 page = find_get_page(mapping, index);
1965 page_cache_sync_readahead(mapping,
1967 index, last_index - index);
1968 page = find_get_page(mapping, index);
1969 if (unlikely(page == NULL))
1970 goto no_cached_page;
1972 if (PageReadahead(page)) {
1973 page_cache_async_readahead(mapping,
1975 index, last_index - index);
1977 if (!PageUptodate(page)) {
1979 * See comment in do_read_cache_page on why
1980 * wait_on_page_locked is used to avoid unnecessarily
1981 * serialisations and why it's safe.
1983 error = wait_on_page_locked_killable(page);
1984 if (unlikely(error))
1985 goto readpage_error;
1986 if (PageUptodate(page))
1989 if (inode->i_blkbits == PAGE_SHIFT ||
1990 !mapping->a_ops->is_partially_uptodate)
1991 goto page_not_up_to_date;
1992 /* pipes can't handle partially uptodate pages */
1993 if (unlikely(iter->type & ITER_PIPE))
1994 goto page_not_up_to_date;
1995 if (!trylock_page(page))
1996 goto page_not_up_to_date;
1997 /* Did it get truncated before we got the lock? */
1999 goto page_not_up_to_date_locked;
2000 if (!mapping->a_ops->is_partially_uptodate(page,
2001 offset, iter->count))
2002 goto page_not_up_to_date_locked;
2007 * i_size must be checked after we know the page is Uptodate.
2009 * Checking i_size after the check allows us to calculate
2010 * the correct value for "nr", which means the zero-filled
2011 * part of the page is not copied back to userspace (unless
2012 * another truncate extends the file - this is desired though).
2015 isize = i_size_read(inode);
2016 end_index = (isize - 1) >> PAGE_SHIFT;
2017 if (unlikely(!isize || index > end_index)) {
2022 /* nr is the maximum number of bytes to copy from this page */
2024 if (index == end_index) {
2025 nr = ((isize - 1) & ~PAGE_MASK) + 1;
2033 /* If users can be writing to this page using arbitrary
2034 * virtual addresses, take care about potential aliasing
2035 * before reading the page on the kernel side.
2037 if (mapping_writably_mapped(mapping))
2038 flush_dcache_page(page);
2041 * When a sequential read accesses a page several times,
2042 * only mark it as accessed the first time.
2044 if (prev_index != index || offset != prev_offset)
2045 mark_page_accessed(page);
2049 * Ok, we have the page, and it's up-to-date, so
2050 * now we can copy it to user space...
2053 ret = copy_page_to_iter(page, offset, nr, iter);
2055 index += offset >> PAGE_SHIFT;
2056 offset &= ~PAGE_MASK;
2057 prev_offset = offset;
2061 if (!iov_iter_count(iter))
2069 page_not_up_to_date:
2070 /* Get exclusive access to the page ... */
2071 error = lock_page_killable(page);
2072 if (unlikely(error))
2073 goto readpage_error;
2075 page_not_up_to_date_locked:
2076 /* Did it get truncated before we got the lock? */
2077 if (!page->mapping) {
2083 /* Did somebody else fill it already? */
2084 if (PageUptodate(page)) {
2091 * A previous I/O error may have been due to temporary
2092 * failures, eg. multipath errors.
2093 * PG_error will be set again if readpage fails.
2095 ClearPageError(page);
2096 /* Start the actual read. The read will unlock the page. */
2097 error = mapping->a_ops->readpage(filp, page);
2099 if (unlikely(error)) {
2100 if (error == AOP_TRUNCATED_PAGE) {
2105 goto readpage_error;
2108 if (!PageUptodate(page)) {
2109 error = lock_page_killable(page);
2110 if (unlikely(error))
2111 goto readpage_error;
2112 if (!PageUptodate(page)) {
2113 if (page->mapping == NULL) {
2115 * invalidate_mapping_pages got it
2122 shrink_readahead_size_eio(filp, ra);
2124 goto readpage_error;
2132 /* UHHUH! A synchronous read error occurred. Report it */
2138 * Ok, it wasn't cached, so we need to create a new
2141 page = page_cache_alloc_cold(mapping);
2146 error = add_to_page_cache_lru(page, mapping, index,
2147 mapping_gfp_constraint(mapping, GFP_KERNEL));
2150 if (error == -EEXIST) {
2160 ra->prev_pos = prev_index;
2161 ra->prev_pos <<= PAGE_SHIFT;
2162 ra->prev_pos |= prev_offset;
2164 *ppos = ((loff_t)index << PAGE_SHIFT) + offset;
2165 file_accessed(filp);
2166 return written ? written : error;
2170 * generic_file_read_iter - generic filesystem read routine
2171 * @iocb: kernel I/O control block
2172 * @iter: destination for the data read
2174 * This is the "read_iter()" routine for all filesystems
2175 * that can use the page cache directly.
2178 generic_file_read_iter(struct kiocb *iocb, struct iov_iter *iter)
2180 struct file *file = iocb->ki_filp;
2182 size_t count = iov_iter_count(iter);
2185 goto out; /* skip atime */
2187 if (iocb->ki_flags & IOCB_DIRECT) {
2188 struct address_space *mapping = file->f_mapping;
2189 struct inode *inode = mapping->host;
2192 size = i_size_read(inode);
2193 if (iocb->ki_flags & IOCB_NOWAIT) {
2194 if (filemap_range_has_page(mapping, iocb->ki_pos,
2195 iocb->ki_pos + count - 1))
2198 retval = filemap_write_and_wait_range(mapping,
2200 iocb->ki_pos + count - 1);
2205 file_accessed(file);
2207 retval = mapping->a_ops->direct_IO(iocb, iter);
2209 iocb->ki_pos += retval;
2212 iov_iter_revert(iter, count - iov_iter_count(iter));
2215 * Btrfs can have a short DIO read if we encounter
2216 * compressed extents, so if there was an error, or if
2217 * we've already read everything we wanted to, or if
2218 * there was a short read because we hit EOF, go ahead
2219 * and return. Otherwise fallthrough to buffered io for
2220 * the rest of the read. Buffered reads will not work for
2221 * DAX files, so don't bother trying.
2223 if (retval < 0 || !count || iocb->ki_pos >= size ||
2228 retval = do_generic_file_read(file, &iocb->ki_pos, iter, retval);
2232 EXPORT_SYMBOL(generic_file_read_iter);
2236 * page_cache_read - adds requested page to the page cache if not already there
2237 * @file: file to read
2238 * @offset: page index
2239 * @gfp_mask: memory allocation flags
2241 * This adds the requested page to the page cache if it isn't already there,
2242 * and schedules an I/O to read in its contents from disk.
2244 static int page_cache_read(struct file *file, pgoff_t offset, gfp_t gfp_mask)
2246 struct address_space *mapping = file->f_mapping;
2251 page = __page_cache_alloc(gfp_mask|__GFP_COLD);
2255 ret = add_to_page_cache_lru(page, mapping, offset, gfp_mask & GFP_KERNEL);
2257 ret = mapping->a_ops->readpage(file, page);
2258 else if (ret == -EEXIST)
2259 ret = 0; /* losing race to add is OK */
2263 } while (ret == AOP_TRUNCATED_PAGE);
2268 #define MMAP_LOTSAMISS (100)
2271 * Synchronous readahead happens when we don't even find
2272 * a page in the page cache at all.
2274 static void do_sync_mmap_readahead(struct vm_area_struct *vma,
2275 struct file_ra_state *ra,
2279 struct address_space *mapping = file->f_mapping;
2281 /* If we don't want any read-ahead, don't bother */
2282 if (vma->vm_flags & VM_RAND_READ)
2287 if (vma->vm_flags & VM_SEQ_READ) {
2288 page_cache_sync_readahead(mapping, ra, file, offset,
2293 /* Avoid banging the cache line if not needed */
2294 if (ra->mmap_miss < MMAP_LOTSAMISS * 10)
2298 * Do we miss much more than hit in this file? If so,
2299 * stop bothering with read-ahead. It will only hurt.
2301 if (ra->mmap_miss > MMAP_LOTSAMISS)
2307 ra->start = max_t(long, 0, offset - ra->ra_pages / 2);
2308 ra->size = ra->ra_pages;
2309 ra->async_size = ra->ra_pages / 4;
2310 ra_submit(ra, mapping, file);
2314 * Asynchronous readahead happens when we find the page and PG_readahead,
2315 * so we want to possibly extend the readahead further..
2317 static void do_async_mmap_readahead(struct vm_area_struct *vma,
2318 struct file_ra_state *ra,
2323 struct address_space *mapping = file->f_mapping;
2325 /* If we don't want any read-ahead, don't bother */
2326 if (vma->vm_flags & VM_RAND_READ)
2328 if (ra->mmap_miss > 0)
2330 if (PageReadahead(page))
2331 page_cache_async_readahead(mapping, ra, file,
2332 page, offset, ra->ra_pages);
2336 * filemap_fault - read in file data for page fault handling
2337 * @vmf: struct vm_fault containing details of the fault
2339 * filemap_fault() is invoked via the vma operations vector for a
2340 * mapped memory region to read in file data during a page fault.
2342 * The goto's are kind of ugly, but this streamlines the normal case of having
2343 * it in the page cache, and handles the special cases reasonably without
2344 * having a lot of duplicated code.
2346 * vma->vm_mm->mmap_sem must be held on entry.
2348 * If our return value has VM_FAULT_RETRY set, it's because
2349 * lock_page_or_retry() returned 0.
2350 * The mmap_sem has usually been released in this case.
2351 * See __lock_page_or_retry() for the exception.
2353 * If our return value does not have VM_FAULT_RETRY set, the mmap_sem
2354 * has not been released.
2356 * We never return with VM_FAULT_RETRY and a bit from VM_FAULT_ERROR set.
2358 int filemap_fault(struct vm_fault *vmf)
2361 struct file *file = vmf->vma->vm_file;
2362 struct address_space *mapping = file->f_mapping;
2363 struct file_ra_state *ra = &file->f_ra;
2364 struct inode *inode = mapping->host;
2365 pgoff_t offset = vmf->pgoff;
2370 max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
2371 if (unlikely(offset >= max_off))
2372 return VM_FAULT_SIGBUS;
2375 * Do we have something in the page cache already?
2377 page = find_get_page(mapping, offset);
2378 if (likely(page) && !(vmf->flags & FAULT_FLAG_TRIED)) {
2380 * We found the page, so try async readahead before
2381 * waiting for the lock.
2383 do_async_mmap_readahead(vmf->vma, ra, file, page, offset);
2385 /* No page in the page cache at all */
2386 do_sync_mmap_readahead(vmf->vma, ra, file, offset);
2387 count_vm_event(PGMAJFAULT);
2388 count_memcg_event_mm(vmf->vma->vm_mm, PGMAJFAULT);
2389 ret = VM_FAULT_MAJOR;
2391 page = find_get_page(mapping, offset);
2393 goto no_cached_page;
2396 if (!lock_page_or_retry(page, vmf->vma->vm_mm, vmf->flags)) {
2398 return ret | VM_FAULT_RETRY;
2401 /* Did it get truncated? */
2402 if (unlikely(page->mapping != mapping)) {
2407 VM_BUG_ON_PAGE(page->index != offset, page);
2410 * We have a locked page in the page cache, now we need to check
2411 * that it's up-to-date. If not, it is going to be due to an error.
2413 if (unlikely(!PageUptodate(page)))
2414 goto page_not_uptodate;
2417 * Found the page and have a reference on it.
2418 * We must recheck i_size under page lock.
2420 max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
2421 if (unlikely(offset >= max_off)) {
2424 return VM_FAULT_SIGBUS;
2428 return ret | VM_FAULT_LOCKED;
2432 * We're only likely to ever get here if MADV_RANDOM is in
2435 error = page_cache_read(file, offset, vmf->gfp_mask);
2438 * The page we want has now been added to the page cache.
2439 * In the unlikely event that someone removed it in the
2440 * meantime, we'll just come back here and read it again.
2446 * An error return from page_cache_read can result if the
2447 * system is low on memory, or a problem occurs while trying
2450 if (error == -ENOMEM)
2451 return VM_FAULT_OOM;
2452 return VM_FAULT_SIGBUS;
2456 * Umm, take care of errors if the page isn't up-to-date.
2457 * Try to re-read it _once_. We do this synchronously,
2458 * because there really aren't any performance issues here
2459 * and we need to check for errors.
2461 ClearPageError(page);
2462 error = mapping->a_ops->readpage(file, page);
2464 wait_on_page_locked(page);
2465 if (!PageUptodate(page))
2470 if (!error || error == AOP_TRUNCATED_PAGE)
2473 /* Things didn't work out. Return zero to tell the mm layer so. */
2474 shrink_readahead_size_eio(file, ra);
2475 return VM_FAULT_SIGBUS;
2477 EXPORT_SYMBOL(filemap_fault);
2479 void filemap_map_pages(struct vm_fault *vmf,
2480 pgoff_t start_pgoff, pgoff_t end_pgoff)
2482 struct radix_tree_iter iter;
2484 struct file *file = vmf->vma->vm_file;
2485 struct address_space *mapping = file->f_mapping;
2486 pgoff_t last_pgoff = start_pgoff;
2487 unsigned long max_idx;
2488 struct page *head, *page;
2491 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter,
2493 if (iter.index > end_pgoff)
2496 page = radix_tree_deref_slot(slot);
2497 if (unlikely(!page))
2499 if (radix_tree_exception(page)) {
2500 if (radix_tree_deref_retry(page)) {
2501 slot = radix_tree_iter_retry(&iter);
2507 head = compound_head(page);
2508 if (!page_cache_get_speculative(head))
2511 /* The page was split under us? */
2512 if (compound_head(page) != head) {
2517 /* Has the page moved? */
2518 if (unlikely(page != *slot)) {
2523 if (!PageUptodate(page) ||
2524 PageReadahead(page) ||
2527 if (!trylock_page(page))
2530 if (page->mapping != mapping || !PageUptodate(page))
2533 max_idx = DIV_ROUND_UP(i_size_read(mapping->host), PAGE_SIZE);
2534 if (page->index >= max_idx)
2537 if (file->f_ra.mmap_miss > 0)
2538 file->f_ra.mmap_miss--;
2540 vmf->address += (iter.index - last_pgoff) << PAGE_SHIFT;
2542 vmf->pte += iter.index - last_pgoff;
2543 last_pgoff = iter.index;
2544 if (alloc_set_pte(vmf, NULL, page))
2553 /* Huge page is mapped? No need to proceed. */
2554 if (pmd_trans_huge(*vmf->pmd))
2556 if (iter.index == end_pgoff)
2561 EXPORT_SYMBOL(filemap_map_pages);
2563 int filemap_page_mkwrite(struct vm_fault *vmf)
2565 struct page *page = vmf->page;
2566 struct inode *inode = file_inode(vmf->vma->vm_file);
2567 int ret = VM_FAULT_LOCKED;
2569 sb_start_pagefault(inode->i_sb);
2570 file_update_time(vmf->vma->vm_file);
2572 if (page->mapping != inode->i_mapping) {
2574 ret = VM_FAULT_NOPAGE;
2578 * We mark the page dirty already here so that when freeze is in
2579 * progress, we are guaranteed that writeback during freezing will
2580 * see the dirty page and writeprotect it again.
2582 set_page_dirty(page);
2583 wait_for_stable_page(page);
2585 sb_end_pagefault(inode->i_sb);
2588 EXPORT_SYMBOL(filemap_page_mkwrite);
2590 const struct vm_operations_struct generic_file_vm_ops = {
2591 .fault = filemap_fault,
2592 .map_pages = filemap_map_pages,
2593 .page_mkwrite = filemap_page_mkwrite,
2596 /* This is used for a general mmap of a disk file */
2598 int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
2600 struct address_space *mapping = file->f_mapping;
2602 if (!mapping->a_ops->readpage)
2604 file_accessed(file);
2605 vma->vm_ops = &generic_file_vm_ops;
2610 * This is for filesystems which do not implement ->writepage.
2612 int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
2614 if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE))
2616 return generic_file_mmap(file, vma);
2619 int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
2623 int generic_file_readonly_mmap(struct file * file, struct vm_area_struct * vma)
2627 #endif /* CONFIG_MMU */
2629 EXPORT_SYMBOL(generic_file_mmap);
2630 EXPORT_SYMBOL(generic_file_readonly_mmap);
2632 static struct page *wait_on_page_read(struct page *page)
2634 if (!IS_ERR(page)) {
2635 wait_on_page_locked(page);
2636 if (!PageUptodate(page)) {
2638 page = ERR_PTR(-EIO);
2644 static struct page *do_read_cache_page(struct address_space *mapping,
2646 int (*filler)(void *, struct page *),
2653 page = find_get_page(mapping, index);
2655 page = __page_cache_alloc(gfp | __GFP_COLD);
2657 return ERR_PTR(-ENOMEM);
2658 err = add_to_page_cache_lru(page, mapping, index, gfp);
2659 if (unlikely(err)) {
2663 /* Presumably ENOMEM for radix tree node */
2664 return ERR_PTR(err);
2668 err = filler(data, page);
2671 return ERR_PTR(err);
2674 page = wait_on_page_read(page);
2679 if (PageUptodate(page))
2683 * Page is not up to date and may be locked due one of the following
2684 * case a: Page is being filled and the page lock is held
2685 * case b: Read/write error clearing the page uptodate status
2686 * case c: Truncation in progress (page locked)
2687 * case d: Reclaim in progress
2689 * Case a, the page will be up to date when the page is unlocked.
2690 * There is no need to serialise on the page lock here as the page
2691 * is pinned so the lock gives no additional protection. Even if the
2692 * the page is truncated, the data is still valid if PageUptodate as
2693 * it's a race vs truncate race.
2694 * Case b, the page will not be up to date
2695 * Case c, the page may be truncated but in itself, the data may still
2696 * be valid after IO completes as it's a read vs truncate race. The
2697 * operation must restart if the page is not uptodate on unlock but
2698 * otherwise serialising on page lock to stabilise the mapping gives
2699 * no additional guarantees to the caller as the page lock is
2700 * released before return.
2701 * Case d, similar to truncation. If reclaim holds the page lock, it
2702 * will be a race with remove_mapping that determines if the mapping
2703 * is valid on unlock but otherwise the data is valid and there is
2704 * no need to serialise with page lock.
2706 * As the page lock gives no additional guarantee, we optimistically
2707 * wait on the page to be unlocked and check if it's up to date and
2708 * use the page if it is. Otherwise, the page lock is required to
2709 * distinguish between the different cases. The motivation is that we
2710 * avoid spurious serialisations and wakeups when multiple processes
2711 * wait on the same page for IO to complete.
2713 wait_on_page_locked(page);
2714 if (PageUptodate(page))
2717 /* Distinguish between all the cases under the safety of the lock */
2720 /* Case c or d, restart the operation */
2721 if (!page->mapping) {
2727 /* Someone else locked and filled the page in a very small window */
2728 if (PageUptodate(page)) {
2735 mark_page_accessed(page);
2740 * read_cache_page - read into page cache, fill it if needed
2741 * @mapping: the page's address_space
2742 * @index: the page index
2743 * @filler: function to perform the read
2744 * @data: first arg to filler(data, page) function, often left as NULL
2746 * Read into the page cache. If a page already exists, and PageUptodate() is
2747 * not set, try to fill the page and wait for it to become unlocked.
2749 * If the page does not get brought uptodate, return -EIO.
2751 struct page *read_cache_page(struct address_space *mapping,
2753 int (*filler)(void *, struct page *),
2756 return do_read_cache_page(mapping, index, filler, data, mapping_gfp_mask(mapping));
2758 EXPORT_SYMBOL(read_cache_page);
2761 * read_cache_page_gfp - read into page cache, using specified page allocation flags.
2762 * @mapping: the page's address_space
2763 * @index: the page index
2764 * @gfp: the page allocator flags to use if allocating
2766 * This is the same as "read_mapping_page(mapping, index, NULL)", but with
2767 * any new page allocations done using the specified allocation flags.
2769 * If the page does not get brought uptodate, return -EIO.
2771 struct page *read_cache_page_gfp(struct address_space *mapping,
2775 filler_t *filler = (filler_t *)mapping->a_ops->readpage;
2777 return do_read_cache_page(mapping, index, filler, NULL, gfp);
2779 EXPORT_SYMBOL(read_cache_page_gfp);
2782 * Performs necessary checks before doing a write
2784 * Can adjust writing position or amount of bytes to write.
2785 * Returns appropriate error code that caller should return or
2786 * zero in case that write should be allowed.
2788 inline ssize_t generic_write_checks(struct kiocb *iocb, struct iov_iter *from)
2790 struct file *file = iocb->ki_filp;
2791 struct inode *inode = file->f_mapping->host;
2792 unsigned long limit = rlimit(RLIMIT_FSIZE);
2795 if (!iov_iter_count(from))
2798 /* FIXME: this is for backwards compatibility with 2.4 */
2799 if (iocb->ki_flags & IOCB_APPEND)
2800 iocb->ki_pos = i_size_read(inode);
2804 if ((iocb->ki_flags & IOCB_NOWAIT) && !(iocb->ki_flags & IOCB_DIRECT))
2807 if (limit != RLIM_INFINITY) {
2808 if (iocb->ki_pos >= limit) {
2809 send_sig(SIGXFSZ, current, 0);
2812 iov_iter_truncate(from, limit - (unsigned long)pos);
2818 if (unlikely(pos + iov_iter_count(from) > MAX_NON_LFS &&
2819 !(file->f_flags & O_LARGEFILE))) {
2820 if (pos >= MAX_NON_LFS)
2822 iov_iter_truncate(from, MAX_NON_LFS - (unsigned long)pos);
2826 * Are we about to exceed the fs block limit ?
2828 * If we have written data it becomes a short write. If we have
2829 * exceeded without writing data we send a signal and return EFBIG.
2830 * Linus frestrict idea will clean these up nicely..
2832 if (unlikely(pos >= inode->i_sb->s_maxbytes))
2835 iov_iter_truncate(from, inode->i_sb->s_maxbytes - pos);
2836 return iov_iter_count(from);
2838 EXPORT_SYMBOL(generic_write_checks);
2840 int pagecache_write_begin(struct file *file, struct address_space *mapping,
2841 loff_t pos, unsigned len, unsigned flags,
2842 struct page **pagep, void **fsdata)
2844 const struct address_space_operations *aops = mapping->a_ops;
2846 return aops->write_begin(file, mapping, pos, len, flags,
2849 EXPORT_SYMBOL(pagecache_write_begin);
2851 int pagecache_write_end(struct file *file, struct address_space *mapping,
2852 loff_t pos, unsigned len, unsigned copied,
2853 struct page *page, void *fsdata)
2855 const struct address_space_operations *aops = mapping->a_ops;
2857 return aops->write_end(file, mapping, pos, len, copied, page, fsdata);
2859 EXPORT_SYMBOL(pagecache_write_end);
2862 generic_file_direct_write(struct kiocb *iocb, struct iov_iter *from)
2864 struct file *file = iocb->ki_filp;
2865 struct address_space *mapping = file->f_mapping;
2866 struct inode *inode = mapping->host;
2867 loff_t pos = iocb->ki_pos;
2872 write_len = iov_iter_count(from);
2873 end = (pos + write_len - 1) >> PAGE_SHIFT;
2875 if (iocb->ki_flags & IOCB_NOWAIT) {
2876 /* If there are pages to writeback, return */
2877 if (filemap_range_has_page(inode->i_mapping, pos,
2878 pos + iov_iter_count(from)))
2881 written = filemap_write_and_wait_range(mapping, pos,
2882 pos + write_len - 1);
2888 * After a write we want buffered reads to be sure to go to disk to get
2889 * the new data. We invalidate clean cached page from the region we're
2890 * about to write. We do this *before* the write so that we can return
2891 * without clobbering -EIOCBQUEUED from ->direct_IO().
2893 written = invalidate_inode_pages2_range(mapping,
2894 pos >> PAGE_SHIFT, end);
2896 * If a page can not be invalidated, return 0 to fall back
2897 * to buffered write.
2900 if (written == -EBUSY)
2905 written = mapping->a_ops->direct_IO(iocb, from);
2908 * Finally, try again to invalidate clean pages which might have been
2909 * cached by non-direct readahead, or faulted in by get_user_pages()
2910 * if the source of the write was an mmap'ed region of the file
2911 * we're writing. Either one is a pretty crazy thing to do,
2912 * so we don't support it 100%. If this invalidation
2913 * fails, tough, the write still worked...
2915 invalidate_inode_pages2_range(mapping,
2916 pos >> PAGE_SHIFT, end);
2920 write_len -= written;
2921 if (pos > i_size_read(inode) && !S_ISBLK(inode->i_mode)) {
2922 i_size_write(inode, pos);
2923 mark_inode_dirty(inode);
2927 iov_iter_revert(from, write_len - iov_iter_count(from));
2931 EXPORT_SYMBOL(generic_file_direct_write);
2934 * Find or create a page at the given pagecache position. Return the locked
2935 * page. This function is specifically for buffered writes.
2937 struct page *grab_cache_page_write_begin(struct address_space *mapping,
2938 pgoff_t index, unsigned flags)
2941 int fgp_flags = FGP_LOCK|FGP_WRITE|FGP_CREAT;
2943 if (flags & AOP_FLAG_NOFS)
2944 fgp_flags |= FGP_NOFS;
2946 page = pagecache_get_page(mapping, index, fgp_flags,
2947 mapping_gfp_mask(mapping));
2949 wait_for_stable_page(page);
2953 EXPORT_SYMBOL(grab_cache_page_write_begin);
2955 ssize_t generic_perform_write(struct file *file,
2956 struct iov_iter *i, loff_t pos)
2958 struct address_space *mapping = file->f_mapping;
2959 const struct address_space_operations *a_ops = mapping->a_ops;
2961 ssize_t written = 0;
2962 unsigned int flags = 0;
2966 unsigned long offset; /* Offset into pagecache page */
2967 unsigned long bytes; /* Bytes to write to page */
2968 size_t copied; /* Bytes copied from user */
2971 offset = (pos & (PAGE_SIZE - 1));
2972 bytes = min_t(unsigned long, PAGE_SIZE - offset,
2977 * Bring in the user page that we will copy from _first_.
2978 * Otherwise there's a nasty deadlock on copying from the
2979 * same page as we're writing to, without it being marked
2982 * Not only is this an optimisation, but it is also required
2983 * to check that the address is actually valid, when atomic
2984 * usercopies are used, below.
2986 if (unlikely(iov_iter_fault_in_readable(i, bytes))) {
2991 if (fatal_signal_pending(current)) {
2996 status = a_ops->write_begin(file, mapping, pos, bytes, flags,
2998 if (unlikely(status < 0))
3001 if (mapping_writably_mapped(mapping))
3002 flush_dcache_page(page);
3004 copied = iov_iter_copy_from_user_atomic(page, i, offset, bytes);
3005 flush_dcache_page(page);
3007 status = a_ops->write_end(file, mapping, pos, bytes, copied,
3009 if (unlikely(status < 0))
3015 iov_iter_advance(i, copied);
3016 if (unlikely(copied == 0)) {
3018 * If we were unable to copy any data at all, we must
3019 * fall back to a single segment length write.
3021 * If we didn't fallback here, we could livelock
3022 * because not all segments in the iov can be copied at
3023 * once without a pagefault.
3025 bytes = min_t(unsigned long, PAGE_SIZE - offset,
3026 iov_iter_single_seg_count(i));
3032 balance_dirty_pages_ratelimited(mapping);
3033 } while (iov_iter_count(i));
3035 return written ? written : status;
3037 EXPORT_SYMBOL(generic_perform_write);
3040 * __generic_file_write_iter - write data to a file
3041 * @iocb: IO state structure (file, offset, etc.)
3042 * @from: iov_iter with data to write
3044 * This function does all the work needed for actually writing data to a
3045 * file. It does all basic checks, removes SUID from the file, updates
3046 * modification times and calls proper subroutines depending on whether we
3047 * do direct IO or a standard buffered write.
3049 * It expects i_mutex to be grabbed unless we work on a block device or similar
3050 * object which does not need locking at all.
3052 * This function does *not* take care of syncing data in case of O_SYNC write.
3053 * A caller has to handle it. This is mainly due to the fact that we want to
3054 * avoid syncing under i_mutex.
3056 ssize_t __generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
3058 struct file *file = iocb->ki_filp;
3059 struct address_space * mapping = file->f_mapping;
3060 struct inode *inode = mapping->host;
3061 ssize_t written = 0;
3065 /* We can write back this queue in page reclaim */
3066 current->backing_dev_info = inode_to_bdi(inode);
3067 err = file_remove_privs(file);
3071 err = file_update_time(file);
3075 if (iocb->ki_flags & IOCB_DIRECT) {
3076 loff_t pos, endbyte;
3078 written = generic_file_direct_write(iocb, from);
3080 * If the write stopped short of completing, fall back to
3081 * buffered writes. Some filesystems do this for writes to
3082 * holes, for example. For DAX files, a buffered write will
3083 * not succeed (even if it did, DAX does not handle dirty
3084 * page-cache pages correctly).
3086 if (written < 0 || !iov_iter_count(from) || IS_DAX(inode))
3089 status = generic_perform_write(file, from, pos = iocb->ki_pos);
3091 * If generic_perform_write() returned a synchronous error
3092 * then we want to return the number of bytes which were
3093 * direct-written, or the error code if that was zero. Note
3094 * that this differs from normal direct-io semantics, which
3095 * will return -EFOO even if some bytes were written.
3097 if (unlikely(status < 0)) {
3102 * We need to ensure that the page cache pages are written to
3103 * disk and invalidated to preserve the expected O_DIRECT
3106 endbyte = pos + status - 1;
3107 err = filemap_write_and_wait_range(mapping, pos, endbyte);
3109 iocb->ki_pos = endbyte + 1;
3111 invalidate_mapping_pages(mapping,
3113 endbyte >> PAGE_SHIFT);
3116 * We don't know how much we wrote, so just return
3117 * the number of bytes which were direct-written
3121 written = generic_perform_write(file, from, iocb->ki_pos);
3122 if (likely(written > 0))
3123 iocb->ki_pos += written;
3126 current->backing_dev_info = NULL;
3127 return written ? written : err;
3129 EXPORT_SYMBOL(__generic_file_write_iter);
3132 * generic_file_write_iter - write data to a file
3133 * @iocb: IO state structure
3134 * @from: iov_iter with data to write
3136 * This is a wrapper around __generic_file_write_iter() to be used by most
3137 * filesystems. It takes care of syncing the file in case of O_SYNC file
3138 * and acquires i_mutex as needed.
3140 ssize_t generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
3142 struct file *file = iocb->ki_filp;
3143 struct inode *inode = file->f_mapping->host;
3147 ret = generic_write_checks(iocb, from);
3149 ret = __generic_file_write_iter(iocb, from);
3150 inode_unlock(inode);
3153 ret = generic_write_sync(iocb, ret);
3156 EXPORT_SYMBOL(generic_file_write_iter);
3159 * try_to_release_page() - release old fs-specific metadata on a page
3161 * @page: the page which the kernel is trying to free
3162 * @gfp_mask: memory allocation flags (and I/O mode)
3164 * The address_space is to try to release any data against the page
3165 * (presumably at page->private). If the release was successful, return '1'.
3166 * Otherwise return zero.
3168 * This may also be called if PG_fscache is set on a page, indicating that the
3169 * page is known to the local caching routines.
3171 * The @gfp_mask argument specifies whether I/O may be performed to release
3172 * this page (__GFP_IO), and whether the call may block (__GFP_RECLAIM & __GFP_FS).
3175 int try_to_release_page(struct page *page, gfp_t gfp_mask)
3177 struct address_space * const mapping = page->mapping;
3179 BUG_ON(!PageLocked(page));
3180 if (PageWriteback(page))
3183 if (mapping && mapping->a_ops->releasepage)
3184 return mapping->a_ops->releasepage(page, gfp_mask);
3185 return try_to_free_buffers(page);
3188 EXPORT_SYMBOL(try_to_release_page);