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/hugetlb.h>
35 #include <linux/memcontrol.h>
36 #include <linux/cleancache.h>
37 #include <linux/shmem_fs.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)
72 * ->i_mmap_rwsem (truncate->unmap_mapping_range)
76 * ->page_table_lock or pte_lock (various, mainly in memory.c)
77 * ->i_pages 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 * ->i_pages 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 * ->i_pages 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 * ->i_pages 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 void page_cache_delete(struct address_space *mapping,
115 struct page *page, void *shadow)
117 XA_STATE(xas, &mapping->i_pages, page->index);
120 mapping_set_update(&xas, mapping);
122 /* hugetlb pages are represented by a single entry in the xarray */
123 if (!PageHuge(page)) {
124 xas_set_order(&xas, page->index, compound_order(page));
125 nr = 1U << compound_order(page);
128 VM_BUG_ON_PAGE(!PageLocked(page), page);
129 VM_BUG_ON_PAGE(PageTail(page), page);
130 VM_BUG_ON_PAGE(nr != 1 && shadow, page);
132 xas_store(&xas, shadow);
133 xas_init_marks(&xas);
135 page->mapping = NULL;
136 /* Leave page->index set: truncation lookup relies upon it */
139 mapping->nrexceptional += nr;
141 * Make sure the nrexceptional update is committed before
142 * the nrpages update so that final truncate racing
143 * with reclaim does not see both counters 0 at the
144 * same time and miss a shadow entry.
148 mapping->nrpages -= nr;
151 static void unaccount_page_cache_page(struct address_space *mapping,
157 * if we're uptodate, flush out into the cleancache, otherwise
158 * invalidate any existing cleancache entries. We can't leave
159 * stale data around in the cleancache once our page is gone
161 if (PageUptodate(page) && PageMappedToDisk(page))
162 cleancache_put_page(page);
164 cleancache_invalidate_page(mapping, page);
166 VM_BUG_ON_PAGE(PageTail(page), page);
167 VM_BUG_ON_PAGE(page_mapped(page), page);
168 if (!IS_ENABLED(CONFIG_DEBUG_VM) && unlikely(page_mapped(page))) {
171 pr_alert("BUG: Bad page cache in process %s pfn:%05lx\n",
172 current->comm, page_to_pfn(page));
173 dump_page(page, "still mapped when deleted");
175 add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE);
177 mapcount = page_mapcount(page);
178 if (mapping_exiting(mapping) &&
179 page_count(page) >= mapcount + 2) {
181 * All vmas have already been torn down, so it's
182 * a good bet that actually the page is unmapped,
183 * and we'd prefer not to leak it: if we're wrong,
184 * some other bad page check should catch it later.
186 page_mapcount_reset(page);
187 page_ref_sub(page, mapcount);
191 /* hugetlb pages do not participate in page cache accounting. */
195 nr = hpage_nr_pages(page);
197 __mod_node_page_state(page_pgdat(page), NR_FILE_PAGES, -nr);
198 if (PageSwapBacked(page)) {
199 __mod_node_page_state(page_pgdat(page), NR_SHMEM, -nr);
200 if (PageTransHuge(page))
201 __dec_node_page_state(page, NR_SHMEM_THPS);
203 VM_BUG_ON_PAGE(PageTransHuge(page), page);
207 * At this point page must be either written or cleaned by
208 * truncate. Dirty page here signals a bug and loss of
211 * This fixes dirty accounting after removing the page entirely
212 * but leaves PageDirty set: it has no effect for truncated
213 * page and anyway will be cleared before returning page into
216 if (WARN_ON_ONCE(PageDirty(page)))
217 account_page_cleaned(page, mapping, inode_to_wb(mapping->host));
221 * Delete a page from the page cache and free it. Caller has to make
222 * sure the page is locked and that nobody else uses it - or that usage
223 * is safe. The caller must hold the i_pages lock.
225 void __delete_from_page_cache(struct page *page, void *shadow)
227 struct address_space *mapping = page->mapping;
229 trace_mm_filemap_delete_from_page_cache(page);
231 unaccount_page_cache_page(mapping, page);
232 page_cache_delete(mapping, page, shadow);
235 static void page_cache_free_page(struct address_space *mapping,
238 void (*freepage)(struct page *);
240 freepage = mapping->a_ops->freepage;
244 if (PageTransHuge(page) && !PageHuge(page)) {
245 page_ref_sub(page, HPAGE_PMD_NR);
246 VM_BUG_ON_PAGE(page_count(page) <= 0, page);
253 * delete_from_page_cache - delete page from page cache
254 * @page: the page which the kernel is trying to remove from page cache
256 * This must be called only on pages that have been verified to be in the page
257 * cache and locked. It will never put the page into the free list, the caller
258 * has a reference on the page.
260 void delete_from_page_cache(struct page *page)
262 struct address_space *mapping = page_mapping(page);
265 BUG_ON(!PageLocked(page));
266 xa_lock_irqsave(&mapping->i_pages, flags);
267 __delete_from_page_cache(page, NULL);
268 xa_unlock_irqrestore(&mapping->i_pages, flags);
270 page_cache_free_page(mapping, page);
272 EXPORT_SYMBOL(delete_from_page_cache);
275 * page_cache_tree_delete_batch - delete several pages from page cache
276 * @mapping: the mapping to which pages belong
277 * @pvec: pagevec with pages to delete
279 * The function walks over mapping->i_pages and removes pages passed in @pvec
280 * from the mapping. The function expects @pvec to be sorted by page index.
281 * It tolerates holes in @pvec (mapping entries at those indices are not
282 * modified). The function expects only THP head pages to be present in the
283 * @pvec and takes care to delete all corresponding tail pages from the
286 * The function expects the i_pages lock to be held.
289 page_cache_tree_delete_batch(struct address_space *mapping,
290 struct pagevec *pvec)
292 struct radix_tree_iter iter;
295 int i = 0, tail_pages = 0;
299 start = pvec->pages[0]->index;
300 radix_tree_for_each_slot(slot, &mapping->i_pages, &iter, start) {
301 if (i >= pagevec_count(pvec) && !tail_pages)
303 page = radix_tree_deref_slot_protected(slot,
304 &mapping->i_pages.xa_lock);
305 if (xa_is_value(page))
309 * Some page got inserted in our range? Skip it. We
310 * have our pages locked so they are protected from
313 if (page != pvec->pages[i])
315 WARN_ON_ONCE(!PageLocked(page));
316 if (PageTransHuge(page) && !PageHuge(page))
317 tail_pages = HPAGE_PMD_NR - 1;
318 page->mapping = NULL;
320 * Leave page->index set: truncation lookup relies
327 radix_tree_clear_tags(&mapping->i_pages, iter.node, slot);
328 __radix_tree_replace(&mapping->i_pages, iter.node, slot, NULL,
329 workingset_lookup_update(mapping));
332 mapping->nrpages -= total_pages;
335 void delete_from_page_cache_batch(struct address_space *mapping,
336 struct pagevec *pvec)
341 if (!pagevec_count(pvec))
344 xa_lock_irqsave(&mapping->i_pages, flags);
345 for (i = 0; i < pagevec_count(pvec); i++) {
346 trace_mm_filemap_delete_from_page_cache(pvec->pages[i]);
348 unaccount_page_cache_page(mapping, pvec->pages[i]);
350 page_cache_tree_delete_batch(mapping, pvec);
351 xa_unlock_irqrestore(&mapping->i_pages, flags);
353 for (i = 0; i < pagevec_count(pvec); i++)
354 page_cache_free_page(mapping, pvec->pages[i]);
357 int filemap_check_errors(struct address_space *mapping)
360 /* Check for outstanding write errors */
361 if (test_bit(AS_ENOSPC, &mapping->flags) &&
362 test_and_clear_bit(AS_ENOSPC, &mapping->flags))
364 if (test_bit(AS_EIO, &mapping->flags) &&
365 test_and_clear_bit(AS_EIO, &mapping->flags))
369 EXPORT_SYMBOL(filemap_check_errors);
371 static int filemap_check_and_keep_errors(struct address_space *mapping)
373 /* Check for outstanding write errors */
374 if (test_bit(AS_EIO, &mapping->flags))
376 if (test_bit(AS_ENOSPC, &mapping->flags))
382 * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range
383 * @mapping: address space structure to write
384 * @start: offset in bytes where the range starts
385 * @end: offset in bytes where the range ends (inclusive)
386 * @sync_mode: enable synchronous operation
388 * Start writeback against all of a mapping's dirty pages that lie
389 * within the byte offsets <start, end> inclusive.
391 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
392 * opposed to a regular memory cleansing writeback. The difference between
393 * these two operations is that if a dirty page/buffer is encountered, it must
394 * be waited upon, and not just skipped over.
396 int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
397 loff_t end, int sync_mode)
400 struct writeback_control wbc = {
401 .sync_mode = sync_mode,
402 .nr_to_write = LONG_MAX,
403 .range_start = start,
407 if (!mapping_cap_writeback_dirty(mapping))
410 wbc_attach_fdatawrite_inode(&wbc, mapping->host);
411 ret = do_writepages(mapping, &wbc);
412 wbc_detach_inode(&wbc);
416 static inline int __filemap_fdatawrite(struct address_space *mapping,
419 return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode);
422 int filemap_fdatawrite(struct address_space *mapping)
424 return __filemap_fdatawrite(mapping, WB_SYNC_ALL);
426 EXPORT_SYMBOL(filemap_fdatawrite);
428 int filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
431 return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL);
433 EXPORT_SYMBOL(filemap_fdatawrite_range);
436 * filemap_flush - mostly a non-blocking flush
437 * @mapping: target address_space
439 * This is a mostly non-blocking flush. Not suitable for data-integrity
440 * purposes - I/O may not be started against all dirty pages.
442 int filemap_flush(struct address_space *mapping)
444 return __filemap_fdatawrite(mapping, WB_SYNC_NONE);
446 EXPORT_SYMBOL(filemap_flush);
449 * filemap_range_has_page - check if a page exists in range.
450 * @mapping: address space within which to check
451 * @start_byte: offset in bytes where the range starts
452 * @end_byte: offset in bytes where the range ends (inclusive)
454 * Find at least one page in the range supplied, usually used to check if
455 * direct writing in this range will trigger a writeback.
457 bool filemap_range_has_page(struct address_space *mapping,
458 loff_t start_byte, loff_t end_byte)
460 pgoff_t index = start_byte >> PAGE_SHIFT;
461 pgoff_t end = end_byte >> PAGE_SHIFT;
464 if (end_byte < start_byte)
467 if (mapping->nrpages == 0)
470 if (!find_get_pages_range(mapping, &index, end, 1, &page))
475 EXPORT_SYMBOL(filemap_range_has_page);
477 static void __filemap_fdatawait_range(struct address_space *mapping,
478 loff_t start_byte, loff_t end_byte)
480 pgoff_t index = start_byte >> PAGE_SHIFT;
481 pgoff_t end = end_byte >> PAGE_SHIFT;
485 if (end_byte < start_byte)
489 while (index <= end) {
492 nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index,
493 end, PAGECACHE_TAG_WRITEBACK);
497 for (i = 0; i < nr_pages; i++) {
498 struct page *page = pvec.pages[i];
500 wait_on_page_writeback(page);
501 ClearPageError(page);
503 pagevec_release(&pvec);
509 * filemap_fdatawait_range - wait for writeback to complete
510 * @mapping: address space structure to wait for
511 * @start_byte: offset in bytes where the range starts
512 * @end_byte: offset in bytes where the range ends (inclusive)
514 * Walk the list of under-writeback pages of the given address space
515 * in the given range and wait for all of them. Check error status of
516 * the address space and return it.
518 * Since the error status of the address space is cleared by this function,
519 * callers are responsible for checking the return value and handling and/or
520 * reporting the error.
522 int filemap_fdatawait_range(struct address_space *mapping, loff_t start_byte,
525 __filemap_fdatawait_range(mapping, start_byte, end_byte);
526 return filemap_check_errors(mapping);
528 EXPORT_SYMBOL(filemap_fdatawait_range);
531 * file_fdatawait_range - wait for writeback to complete
532 * @file: file pointing to address space structure to wait for
533 * @start_byte: offset in bytes where the range starts
534 * @end_byte: offset in bytes where the range ends (inclusive)
536 * Walk the list of under-writeback pages of the address space that file
537 * refers to, in the given range and wait for all of them. Check error
538 * status of the address space vs. the file->f_wb_err cursor and return it.
540 * Since the error status of the file is advanced by this function,
541 * callers are responsible for checking the return value and handling and/or
542 * reporting the error.
544 int file_fdatawait_range(struct file *file, loff_t start_byte, loff_t end_byte)
546 struct address_space *mapping = file->f_mapping;
548 __filemap_fdatawait_range(mapping, start_byte, end_byte);
549 return file_check_and_advance_wb_err(file);
551 EXPORT_SYMBOL(file_fdatawait_range);
554 * filemap_fdatawait_keep_errors - wait for writeback without clearing errors
555 * @mapping: address space structure to wait for
557 * Walk the list of under-writeback pages of the given address space
558 * and wait for all of them. Unlike filemap_fdatawait(), this function
559 * does not clear error status of the address space.
561 * Use this function if callers don't handle errors themselves. Expected
562 * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2),
565 int filemap_fdatawait_keep_errors(struct address_space *mapping)
567 __filemap_fdatawait_range(mapping, 0, LLONG_MAX);
568 return filemap_check_and_keep_errors(mapping);
570 EXPORT_SYMBOL(filemap_fdatawait_keep_errors);
572 static bool mapping_needs_writeback(struct address_space *mapping)
574 return (!dax_mapping(mapping) && mapping->nrpages) ||
575 (dax_mapping(mapping) && mapping->nrexceptional);
578 int filemap_write_and_wait(struct address_space *mapping)
582 if (mapping_needs_writeback(mapping)) {
583 err = filemap_fdatawrite(mapping);
585 * Even if the above returned error, the pages may be
586 * written partially (e.g. -ENOSPC), so we wait for it.
587 * But the -EIO is special case, it may indicate the worst
588 * thing (e.g. bug) happened, so we avoid waiting for it.
591 int err2 = filemap_fdatawait(mapping);
595 /* Clear any previously stored errors */
596 filemap_check_errors(mapping);
599 err = filemap_check_errors(mapping);
603 EXPORT_SYMBOL(filemap_write_and_wait);
606 * filemap_write_and_wait_range - write out & wait on a file range
607 * @mapping: the address_space for the pages
608 * @lstart: offset in bytes where the range starts
609 * @lend: offset in bytes where the range ends (inclusive)
611 * Write out and wait upon file offsets lstart->lend, inclusive.
613 * Note that @lend is inclusive (describes the last byte to be written) so
614 * that this function can be used to write to the very end-of-file (end = -1).
616 int filemap_write_and_wait_range(struct address_space *mapping,
617 loff_t lstart, loff_t lend)
621 if (mapping_needs_writeback(mapping)) {
622 err = __filemap_fdatawrite_range(mapping, lstart, lend,
624 /* See comment of filemap_write_and_wait() */
626 int err2 = filemap_fdatawait_range(mapping,
631 /* Clear any previously stored errors */
632 filemap_check_errors(mapping);
635 err = filemap_check_errors(mapping);
639 EXPORT_SYMBOL(filemap_write_and_wait_range);
641 void __filemap_set_wb_err(struct address_space *mapping, int err)
643 errseq_t eseq = errseq_set(&mapping->wb_err, err);
645 trace_filemap_set_wb_err(mapping, eseq);
647 EXPORT_SYMBOL(__filemap_set_wb_err);
650 * file_check_and_advance_wb_err - report wb error (if any) that was previously
651 * and advance wb_err to current one
652 * @file: struct file on which the error is being reported
654 * When userland calls fsync (or something like nfsd does the equivalent), we
655 * want to report any writeback errors that occurred since the last fsync (or
656 * since the file was opened if there haven't been any).
658 * Grab the wb_err from the mapping. If it matches what we have in the file,
659 * then just quickly return 0. The file is all caught up.
661 * If it doesn't match, then take the mapping value, set the "seen" flag in
662 * it and try to swap it into place. If it works, or another task beat us
663 * to it with the new value, then update the f_wb_err and return the error
664 * portion. The error at this point must be reported via proper channels
665 * (a'la fsync, or NFS COMMIT operation, etc.).
667 * While we handle mapping->wb_err with atomic operations, the f_wb_err
668 * value is protected by the f_lock since we must ensure that it reflects
669 * the latest value swapped in for this file descriptor.
671 int file_check_and_advance_wb_err(struct file *file)
674 errseq_t old = READ_ONCE(file->f_wb_err);
675 struct address_space *mapping = file->f_mapping;
677 /* Locklessly handle the common case where nothing has changed */
678 if (errseq_check(&mapping->wb_err, old)) {
679 /* Something changed, must use slow path */
680 spin_lock(&file->f_lock);
681 old = file->f_wb_err;
682 err = errseq_check_and_advance(&mapping->wb_err,
684 trace_file_check_and_advance_wb_err(file, old);
685 spin_unlock(&file->f_lock);
689 * We're mostly using this function as a drop in replacement for
690 * filemap_check_errors. Clear AS_EIO/AS_ENOSPC to emulate the effect
691 * that the legacy code would have had on these flags.
693 clear_bit(AS_EIO, &mapping->flags);
694 clear_bit(AS_ENOSPC, &mapping->flags);
697 EXPORT_SYMBOL(file_check_and_advance_wb_err);
700 * file_write_and_wait_range - write out & wait on a file range
701 * @file: file pointing to address_space with pages
702 * @lstart: offset in bytes where the range starts
703 * @lend: offset in bytes where the range ends (inclusive)
705 * Write out and wait upon file offsets lstart->lend, inclusive.
707 * Note that @lend is inclusive (describes the last byte to be written) so
708 * that this function can be used to write to the very end-of-file (end = -1).
710 * After writing out and waiting on the data, we check and advance the
711 * f_wb_err cursor to the latest value, and return any errors detected there.
713 int file_write_and_wait_range(struct file *file, loff_t lstart, loff_t lend)
716 struct address_space *mapping = file->f_mapping;
718 if (mapping_needs_writeback(mapping)) {
719 err = __filemap_fdatawrite_range(mapping, lstart, lend,
721 /* See comment of filemap_write_and_wait() */
723 __filemap_fdatawait_range(mapping, lstart, lend);
725 err2 = file_check_and_advance_wb_err(file);
730 EXPORT_SYMBOL(file_write_and_wait_range);
733 * replace_page_cache_page - replace a pagecache page with a new one
734 * @old: page to be replaced
735 * @new: page to replace with
736 * @gfp_mask: allocation mode
738 * This function replaces a page in the pagecache with a new one. On
739 * success it acquires the pagecache reference for the new page and
740 * drops it for the old page. Both the old and new pages must be
741 * locked. This function does not add the new page to the LRU, the
742 * caller must do that.
744 * The remove + add is atomic. This function cannot fail.
746 int replace_page_cache_page(struct page *old, struct page *new, gfp_t gfp_mask)
748 struct address_space *mapping = old->mapping;
749 void (*freepage)(struct page *) = mapping->a_ops->freepage;
750 pgoff_t offset = old->index;
751 XA_STATE(xas, &mapping->i_pages, offset);
754 VM_BUG_ON_PAGE(!PageLocked(old), old);
755 VM_BUG_ON_PAGE(!PageLocked(new), new);
756 VM_BUG_ON_PAGE(new->mapping, new);
759 new->mapping = mapping;
762 xas_lock_irqsave(&xas, flags);
763 xas_store(&xas, new);
766 /* hugetlb pages do not participate in page cache accounting. */
768 __dec_node_page_state(new, NR_FILE_PAGES);
770 __inc_node_page_state(new, NR_FILE_PAGES);
771 if (PageSwapBacked(old))
772 __dec_node_page_state(new, NR_SHMEM);
773 if (PageSwapBacked(new))
774 __inc_node_page_state(new, NR_SHMEM);
775 xas_unlock_irqrestore(&xas, flags);
776 mem_cgroup_migrate(old, new);
783 EXPORT_SYMBOL_GPL(replace_page_cache_page);
785 static int __add_to_page_cache_locked(struct page *page,
786 struct address_space *mapping,
787 pgoff_t offset, gfp_t gfp_mask,
790 XA_STATE(xas, &mapping->i_pages, offset);
791 int huge = PageHuge(page);
792 struct mem_cgroup *memcg;
796 VM_BUG_ON_PAGE(!PageLocked(page), page);
797 VM_BUG_ON_PAGE(PageSwapBacked(page), page);
798 mapping_set_update(&xas, mapping);
801 error = mem_cgroup_try_charge(page, current->mm,
802 gfp_mask, &memcg, false);
808 page->mapping = mapping;
809 page->index = offset;
813 old = xas_load(&xas);
814 if (old && !xa_is_value(old))
815 xas_set_err(&xas, -EEXIST);
816 xas_store(&xas, page);
820 if (xa_is_value(old)) {
821 mapping->nrexceptional--;
827 /* hugetlb pages do not participate in page cache accounting */
829 __inc_node_page_state(page, NR_FILE_PAGES);
831 xas_unlock_irq(&xas);
832 } while (xas_nomem(&xas, gfp_mask & GFP_RECLAIM_MASK));
838 mem_cgroup_commit_charge(page, memcg, false, false);
839 trace_mm_filemap_add_to_page_cache(page);
842 page->mapping = NULL;
843 /* Leave page->index set: truncation relies upon it */
845 mem_cgroup_cancel_charge(page, memcg, false);
847 return xas_error(&xas);
851 * add_to_page_cache_locked - add a locked page to the pagecache
853 * @mapping: the page's address_space
854 * @offset: page index
855 * @gfp_mask: page allocation mode
857 * This function is used to add a page to the pagecache. It must be locked.
858 * This function does not add the page to the LRU. The caller must do that.
860 int add_to_page_cache_locked(struct page *page, struct address_space *mapping,
861 pgoff_t offset, gfp_t gfp_mask)
863 return __add_to_page_cache_locked(page, mapping, offset,
866 EXPORT_SYMBOL(add_to_page_cache_locked);
868 int add_to_page_cache_lru(struct page *page, struct address_space *mapping,
869 pgoff_t offset, gfp_t gfp_mask)
874 __SetPageLocked(page);
875 ret = __add_to_page_cache_locked(page, mapping, offset,
878 __ClearPageLocked(page);
881 * The page might have been evicted from cache only
882 * recently, in which case it should be activated like
883 * any other repeatedly accessed page.
884 * The exception is pages getting rewritten; evicting other
885 * data from the working set, only to cache data that will
886 * get overwritten with something else, is a waste of memory.
888 if (!(gfp_mask & __GFP_WRITE) &&
889 shadow && workingset_refault(shadow)) {
891 workingset_activation(page);
893 ClearPageActive(page);
898 EXPORT_SYMBOL_GPL(add_to_page_cache_lru);
901 struct page *__page_cache_alloc(gfp_t gfp)
906 if (cpuset_do_page_mem_spread()) {
907 unsigned int cpuset_mems_cookie;
909 cpuset_mems_cookie = read_mems_allowed_begin();
910 n = cpuset_mem_spread_node();
911 page = __alloc_pages_node(n, gfp, 0);
912 } while (!page && read_mems_allowed_retry(cpuset_mems_cookie));
916 return alloc_pages(gfp, 0);
918 EXPORT_SYMBOL(__page_cache_alloc);
922 * In order to wait for pages to become available there must be
923 * waitqueues associated with pages. By using a hash table of
924 * waitqueues where the bucket discipline is to maintain all
925 * waiters on the same queue and wake all when any of the pages
926 * become available, and for the woken contexts to check to be
927 * sure the appropriate page became available, this saves space
928 * at a cost of "thundering herd" phenomena during rare hash
931 #define PAGE_WAIT_TABLE_BITS 8
932 #define PAGE_WAIT_TABLE_SIZE (1 << PAGE_WAIT_TABLE_BITS)
933 static wait_queue_head_t page_wait_table[PAGE_WAIT_TABLE_SIZE] __cacheline_aligned;
935 static wait_queue_head_t *page_waitqueue(struct page *page)
937 return &page_wait_table[hash_ptr(page, PAGE_WAIT_TABLE_BITS)];
940 void __init pagecache_init(void)
944 for (i = 0; i < PAGE_WAIT_TABLE_SIZE; i++)
945 init_waitqueue_head(&page_wait_table[i]);
947 page_writeback_init();
950 /* This has the same layout as wait_bit_key - see fs/cachefiles/rdwr.c */
951 struct wait_page_key {
957 struct wait_page_queue {
960 wait_queue_entry_t wait;
963 static int wake_page_function(wait_queue_entry_t *wait, unsigned mode, int sync, void *arg)
965 struct wait_page_key *key = arg;
966 struct wait_page_queue *wait_page
967 = container_of(wait, struct wait_page_queue, wait);
969 if (wait_page->page != key->page)
973 if (wait_page->bit_nr != key->bit_nr)
976 /* Stop walking if it's locked */
977 if (test_bit(key->bit_nr, &key->page->flags))
980 return autoremove_wake_function(wait, mode, sync, key);
983 static void wake_up_page_bit(struct page *page, int bit_nr)
985 wait_queue_head_t *q = page_waitqueue(page);
986 struct wait_page_key key;
988 wait_queue_entry_t bookmark;
995 bookmark.private = NULL;
996 bookmark.func = NULL;
997 INIT_LIST_HEAD(&bookmark.entry);
999 spin_lock_irqsave(&q->lock, flags);
1000 __wake_up_locked_key_bookmark(q, TASK_NORMAL, &key, &bookmark);
1002 while (bookmark.flags & WQ_FLAG_BOOKMARK) {
1004 * Take a breather from holding the lock,
1005 * allow pages that finish wake up asynchronously
1006 * to acquire the lock and remove themselves
1009 spin_unlock_irqrestore(&q->lock, flags);
1011 spin_lock_irqsave(&q->lock, flags);
1012 __wake_up_locked_key_bookmark(q, TASK_NORMAL, &key, &bookmark);
1016 * It is possible for other pages to have collided on the waitqueue
1017 * hash, so in that case check for a page match. That prevents a long-
1020 * It is still possible to miss a case here, when we woke page waiters
1021 * and removed them from the waitqueue, but there are still other
1024 if (!waitqueue_active(q) || !key.page_match) {
1025 ClearPageWaiters(page);
1027 * It's possible to miss clearing Waiters here, when we woke
1028 * our page waiters, but the hashed waitqueue has waiters for
1029 * other pages on it.
1031 * That's okay, it's a rare case. The next waker will clear it.
1034 spin_unlock_irqrestore(&q->lock, flags);
1037 static void wake_up_page(struct page *page, int bit)
1039 if (!PageWaiters(page))
1041 wake_up_page_bit(page, bit);
1044 static inline int wait_on_page_bit_common(wait_queue_head_t *q,
1045 struct page *page, int bit_nr, int state, bool lock)
1047 struct wait_page_queue wait_page;
1048 wait_queue_entry_t *wait = &wait_page.wait;
1052 wait->flags = lock ? WQ_FLAG_EXCLUSIVE : 0;
1053 wait->func = wake_page_function;
1054 wait_page.page = page;
1055 wait_page.bit_nr = bit_nr;
1058 spin_lock_irq(&q->lock);
1060 if (likely(list_empty(&wait->entry))) {
1061 __add_wait_queue_entry_tail(q, wait);
1062 SetPageWaiters(page);
1065 set_current_state(state);
1067 spin_unlock_irq(&q->lock);
1069 if (likely(test_bit(bit_nr, &page->flags))) {
1074 if (!test_and_set_bit_lock(bit_nr, &page->flags))
1077 if (!test_bit(bit_nr, &page->flags))
1081 if (unlikely(signal_pending_state(state, current))) {
1087 finish_wait(q, wait);
1090 * A signal could leave PageWaiters set. Clearing it here if
1091 * !waitqueue_active would be possible (by open-coding finish_wait),
1092 * but still fail to catch it in the case of wait hash collision. We
1093 * already can fail to clear wait hash collision cases, so don't
1094 * bother with signals either.
1100 void wait_on_page_bit(struct page *page, int bit_nr)
1102 wait_queue_head_t *q = page_waitqueue(page);
1103 wait_on_page_bit_common(q, page, bit_nr, TASK_UNINTERRUPTIBLE, false);
1105 EXPORT_SYMBOL(wait_on_page_bit);
1107 int wait_on_page_bit_killable(struct page *page, int bit_nr)
1109 wait_queue_head_t *q = page_waitqueue(page);
1110 return wait_on_page_bit_common(q, page, bit_nr, TASK_KILLABLE, false);
1112 EXPORT_SYMBOL(wait_on_page_bit_killable);
1115 * add_page_wait_queue - Add an arbitrary waiter to a page's wait queue
1116 * @page: Page defining the wait queue of interest
1117 * @waiter: Waiter to add to the queue
1119 * Add an arbitrary @waiter to the wait queue for the nominated @page.
1121 void add_page_wait_queue(struct page *page, wait_queue_entry_t *waiter)
1123 wait_queue_head_t *q = page_waitqueue(page);
1124 unsigned long flags;
1126 spin_lock_irqsave(&q->lock, flags);
1127 __add_wait_queue_entry_tail(q, waiter);
1128 SetPageWaiters(page);
1129 spin_unlock_irqrestore(&q->lock, flags);
1131 EXPORT_SYMBOL_GPL(add_page_wait_queue);
1133 #ifndef clear_bit_unlock_is_negative_byte
1136 * PG_waiters is the high bit in the same byte as PG_lock.
1138 * On x86 (and on many other architectures), we can clear PG_lock and
1139 * test the sign bit at the same time. But if the architecture does
1140 * not support that special operation, we just do this all by hand
1143 * The read of PG_waiters has to be after (or concurrently with) PG_locked
1144 * being cleared, but a memory barrier should be unneccssary since it is
1145 * in the same byte as PG_locked.
1147 static inline bool clear_bit_unlock_is_negative_byte(long nr, volatile void *mem)
1149 clear_bit_unlock(nr, mem);
1150 /* smp_mb__after_atomic(); */
1151 return test_bit(PG_waiters, mem);
1157 * unlock_page - unlock a locked page
1160 * Unlocks the page and wakes up sleepers in ___wait_on_page_locked().
1161 * Also wakes sleepers in wait_on_page_writeback() because the wakeup
1162 * mechanism between PageLocked pages and PageWriteback pages is shared.
1163 * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep.
1165 * Note that this depends on PG_waiters being the sign bit in the byte
1166 * that contains PG_locked - thus the BUILD_BUG_ON(). That allows us to
1167 * clear the PG_locked bit and test PG_waiters at the same time fairly
1168 * portably (architectures that do LL/SC can test any bit, while x86 can
1169 * test the sign bit).
1171 void unlock_page(struct page *page)
1173 BUILD_BUG_ON(PG_waiters != 7);
1174 page = compound_head(page);
1175 VM_BUG_ON_PAGE(!PageLocked(page), page);
1176 if (clear_bit_unlock_is_negative_byte(PG_locked, &page->flags))
1177 wake_up_page_bit(page, PG_locked);
1179 EXPORT_SYMBOL(unlock_page);
1182 * end_page_writeback - end writeback against a page
1185 void end_page_writeback(struct page *page)
1188 * TestClearPageReclaim could be used here but it is an atomic
1189 * operation and overkill in this particular case. Failing to
1190 * shuffle a page marked for immediate reclaim is too mild to
1191 * justify taking an atomic operation penalty at the end of
1192 * ever page writeback.
1194 if (PageReclaim(page)) {
1195 ClearPageReclaim(page);
1196 rotate_reclaimable_page(page);
1199 if (!test_clear_page_writeback(page))
1202 smp_mb__after_atomic();
1203 wake_up_page(page, PG_writeback);
1205 EXPORT_SYMBOL(end_page_writeback);
1208 * After completing I/O on a page, call this routine to update the page
1209 * flags appropriately
1211 void page_endio(struct page *page, bool is_write, int err)
1215 SetPageUptodate(page);
1217 ClearPageUptodate(page);
1223 struct address_space *mapping;
1226 mapping = page_mapping(page);
1228 mapping_set_error(mapping, err);
1230 end_page_writeback(page);
1233 EXPORT_SYMBOL_GPL(page_endio);
1236 * __lock_page - get a lock on the page, assuming we need to sleep to get it
1237 * @__page: the page to lock
1239 void __lock_page(struct page *__page)
1241 struct page *page = compound_head(__page);
1242 wait_queue_head_t *q = page_waitqueue(page);
1243 wait_on_page_bit_common(q, page, PG_locked, TASK_UNINTERRUPTIBLE, true);
1245 EXPORT_SYMBOL(__lock_page);
1247 int __lock_page_killable(struct page *__page)
1249 struct page *page = compound_head(__page);
1250 wait_queue_head_t *q = page_waitqueue(page);
1251 return wait_on_page_bit_common(q, page, PG_locked, TASK_KILLABLE, true);
1253 EXPORT_SYMBOL_GPL(__lock_page_killable);
1257 * 1 - page is locked; mmap_sem is still held.
1258 * 0 - page is not locked.
1259 * mmap_sem has been released (up_read()), unless flags had both
1260 * FAULT_FLAG_ALLOW_RETRY and FAULT_FLAG_RETRY_NOWAIT set, in
1261 * which case mmap_sem is still held.
1263 * If neither ALLOW_RETRY nor KILLABLE are set, will always return 1
1264 * with the page locked and the mmap_sem unperturbed.
1266 int __lock_page_or_retry(struct page *page, struct mm_struct *mm,
1269 if (flags & FAULT_FLAG_ALLOW_RETRY) {
1271 * CAUTION! In this case, mmap_sem is not released
1272 * even though return 0.
1274 if (flags & FAULT_FLAG_RETRY_NOWAIT)
1277 up_read(&mm->mmap_sem);
1278 if (flags & FAULT_FLAG_KILLABLE)
1279 wait_on_page_locked_killable(page);
1281 wait_on_page_locked(page);
1284 if (flags & FAULT_FLAG_KILLABLE) {
1287 ret = __lock_page_killable(page);
1289 up_read(&mm->mmap_sem);
1299 * page_cache_next_miss() - Find the next gap in the page cache.
1300 * @mapping: Mapping.
1302 * @max_scan: Maximum range to search.
1304 * Search the range [index, min(index + max_scan - 1, ULONG_MAX)] for the
1305 * gap with the lowest index.
1307 * This function may be called under the rcu_read_lock. However, this will
1308 * not atomically search a snapshot of the cache at a single point in time.
1309 * For example, if a gap is created at index 5, then subsequently a gap is
1310 * created at index 10, page_cache_next_miss covering both indices may
1311 * return 10 if called under the rcu_read_lock.
1313 * Return: The index of the gap if found, otherwise an index outside the
1314 * range specified (in which case 'return - index >= max_scan' will be true).
1315 * In the rare case of index wrap-around, 0 will be returned.
1317 pgoff_t page_cache_next_miss(struct address_space *mapping,
1318 pgoff_t index, unsigned long max_scan)
1320 XA_STATE(xas, &mapping->i_pages, index);
1322 while (max_scan--) {
1323 void *entry = xas_next(&xas);
1324 if (!entry || xa_is_value(entry))
1326 if (xas.xa_index == 0)
1330 return xas.xa_index;
1332 EXPORT_SYMBOL(page_cache_next_miss);
1335 * page_cache_prev_miss() - Find the next gap in the page cache.
1336 * @mapping: Mapping.
1338 * @max_scan: Maximum range to search.
1340 * Search the range [max(index - max_scan + 1, 0), index] for the
1341 * gap with the highest index.
1343 * This function may be called under the rcu_read_lock. However, this will
1344 * not atomically search a snapshot of the cache at a single point in time.
1345 * For example, if a gap is created at index 10, then subsequently a gap is
1346 * created at index 5, page_cache_prev_miss() covering both indices may
1347 * return 5 if called under the rcu_read_lock.
1349 * Return: The index of the gap if found, otherwise an index outside the
1350 * range specified (in which case 'index - return >= max_scan' will be true).
1351 * In the rare case of wrap-around, ULONG_MAX will be returned.
1353 pgoff_t page_cache_prev_miss(struct address_space *mapping,
1354 pgoff_t index, unsigned long max_scan)
1356 XA_STATE(xas, &mapping->i_pages, index);
1358 while (max_scan--) {
1359 void *entry = xas_prev(&xas);
1360 if (!entry || xa_is_value(entry))
1362 if (xas.xa_index == ULONG_MAX)
1366 return xas.xa_index;
1368 EXPORT_SYMBOL(page_cache_prev_miss);
1371 * find_get_entry - find and get a page cache entry
1372 * @mapping: the address_space to search
1373 * @offset: the page cache index
1375 * Looks up the page cache slot at @mapping & @offset. If there is a
1376 * page cache page, it is returned with an increased refcount.
1378 * If the slot holds a shadow entry of a previously evicted page, or a
1379 * swap entry from shmem/tmpfs, it is returned.
1381 * Otherwise, %NULL is returned.
1383 struct page *find_get_entry(struct address_space *mapping, pgoff_t offset)
1385 XA_STATE(xas, &mapping->i_pages, offset);
1386 struct page *head, *page;
1391 page = xas_load(&xas);
1392 if (xas_retry(&xas, page))
1395 * A shadow entry of a recently evicted page, or a swap entry from
1396 * shmem/tmpfs. Return it without attempting to raise page count.
1398 if (!page || xa_is_value(page))
1401 head = compound_head(page);
1402 if (!page_cache_get_speculative(head))
1405 /* The page was split under us? */
1406 if (compound_head(page) != head) {
1412 * Has the page moved?
1413 * This is part of the lockless pagecache protocol. See
1414 * include/linux/pagemap.h for details.
1416 if (unlikely(page != xas_reload(&xas))) {
1425 EXPORT_SYMBOL(find_get_entry);
1428 * find_lock_entry - locate, pin and lock a page cache entry
1429 * @mapping: the address_space to search
1430 * @offset: the page cache index
1432 * Looks up the page cache slot at @mapping & @offset. If there is a
1433 * page cache page, it is returned locked and with an increased
1436 * If the slot holds a shadow entry of a previously evicted page, or a
1437 * swap entry from shmem/tmpfs, it is returned.
1439 * Otherwise, %NULL is returned.
1441 * find_lock_entry() may sleep.
1443 struct page *find_lock_entry(struct address_space *mapping, pgoff_t offset)
1448 page = find_get_entry(mapping, offset);
1449 if (page && !xa_is_value(page)) {
1451 /* Has the page been truncated? */
1452 if (unlikely(page_mapping(page) != mapping)) {
1457 VM_BUG_ON_PAGE(page_to_pgoff(page) != offset, page);
1461 EXPORT_SYMBOL(find_lock_entry);
1464 * pagecache_get_page - find and get a page reference
1465 * @mapping: the address_space to search
1466 * @offset: the page index
1467 * @fgp_flags: PCG flags
1468 * @gfp_mask: gfp mask to use for the page cache data page allocation
1470 * Looks up the page cache slot at @mapping & @offset.
1472 * PCG flags modify how the page is returned.
1474 * @fgp_flags can be:
1476 * - FGP_ACCESSED: the page will be marked accessed
1477 * - FGP_LOCK: Page is return locked
1478 * - FGP_CREAT: If page is not present then a new page is allocated using
1479 * @gfp_mask and added to the page cache and the VM's LRU
1480 * list. The page is returned locked and with an increased
1481 * refcount. Otherwise, NULL is returned.
1483 * If FGP_LOCK or FGP_CREAT are specified then the function may sleep even
1484 * if the GFP flags specified for FGP_CREAT are atomic.
1486 * If there is a page cache page, it is returned with an increased refcount.
1488 struct page *pagecache_get_page(struct address_space *mapping, pgoff_t offset,
1489 int fgp_flags, gfp_t gfp_mask)
1494 page = find_get_entry(mapping, offset);
1495 if (xa_is_value(page))
1500 if (fgp_flags & FGP_LOCK) {
1501 if (fgp_flags & FGP_NOWAIT) {
1502 if (!trylock_page(page)) {
1510 /* Has the page been truncated? */
1511 if (unlikely(page->mapping != mapping)) {
1516 VM_BUG_ON_PAGE(page->index != offset, page);
1519 if (page && (fgp_flags & FGP_ACCESSED))
1520 mark_page_accessed(page);
1523 if (!page && (fgp_flags & FGP_CREAT)) {
1525 if ((fgp_flags & FGP_WRITE) && mapping_cap_account_dirty(mapping))
1526 gfp_mask |= __GFP_WRITE;
1527 if (fgp_flags & FGP_NOFS)
1528 gfp_mask &= ~__GFP_FS;
1530 page = __page_cache_alloc(gfp_mask);
1534 if (WARN_ON_ONCE(!(fgp_flags & FGP_LOCK)))
1535 fgp_flags |= FGP_LOCK;
1537 /* Init accessed so avoid atomic mark_page_accessed later */
1538 if (fgp_flags & FGP_ACCESSED)
1539 __SetPageReferenced(page);
1541 err = add_to_page_cache_lru(page, mapping, offset, gfp_mask);
1542 if (unlikely(err)) {
1552 EXPORT_SYMBOL(pagecache_get_page);
1555 * find_get_entries - gang pagecache lookup
1556 * @mapping: The address_space to search
1557 * @start: The starting page cache index
1558 * @nr_entries: The maximum number of entries
1559 * @entries: Where the resulting entries are placed
1560 * @indices: The cache indices corresponding to the entries in @entries
1562 * find_get_entries() will search for and return a group of up to
1563 * @nr_entries entries in the mapping. The entries are placed at
1564 * @entries. find_get_entries() takes a reference against any actual
1567 * The search returns a group of mapping-contiguous page cache entries
1568 * with ascending indexes. There may be holes in the indices due to
1569 * not-present pages.
1571 * Any shadow entries of evicted pages, or swap entries from
1572 * shmem/tmpfs, are included in the returned array.
1574 * find_get_entries() returns the number of pages and shadow entries
1577 unsigned find_get_entries(struct address_space *mapping,
1578 pgoff_t start, unsigned int nr_entries,
1579 struct page **entries, pgoff_t *indices)
1581 XA_STATE(xas, &mapping->i_pages, start);
1583 unsigned int ret = 0;
1589 xas_for_each(&xas, page, ULONG_MAX) {
1591 if (xas_retry(&xas, page))
1594 * A shadow entry of a recently evicted page, a swap
1595 * entry from shmem/tmpfs or a DAX entry. Return it
1596 * without attempting to raise page count.
1598 if (xa_is_value(page))
1601 head = compound_head(page);
1602 if (!page_cache_get_speculative(head))
1605 /* The page was split under us? */
1606 if (compound_head(page) != head)
1609 /* Has the page moved? */
1610 if (unlikely(page != xas_reload(&xas)))
1614 indices[ret] = xas.xa_index;
1615 entries[ret] = page;
1616 if (++ret == nr_entries)
1629 * find_get_pages_range - gang pagecache lookup
1630 * @mapping: The address_space to search
1631 * @start: The starting page index
1632 * @end: The final page index (inclusive)
1633 * @nr_pages: The maximum number of pages
1634 * @pages: Where the resulting pages are placed
1636 * find_get_pages_range() will search for and return a group of up to @nr_pages
1637 * pages in the mapping starting at index @start and up to index @end
1638 * (inclusive). The pages are placed at @pages. find_get_pages_range() takes
1639 * a reference against the returned pages.
1641 * The search returns a group of mapping-contiguous pages with ascending
1642 * indexes. There may be holes in the indices due to not-present pages.
1643 * We also update @start to index the next page for the traversal.
1645 * find_get_pages_range() returns the number of pages which were found. If this
1646 * number is smaller than @nr_pages, the end of specified range has been
1649 unsigned find_get_pages_range(struct address_space *mapping, pgoff_t *start,
1650 pgoff_t end, unsigned int nr_pages,
1651 struct page **pages)
1653 XA_STATE(xas, &mapping->i_pages, *start);
1657 if (unlikely(!nr_pages))
1661 xas_for_each(&xas, page, end) {
1663 if (xas_retry(&xas, page))
1665 /* Skip over shadow, swap and DAX entries */
1666 if (xa_is_value(page))
1669 head = compound_head(page);
1670 if (!page_cache_get_speculative(head))
1673 /* The page was split under us? */
1674 if (compound_head(page) != head)
1677 /* Has the page moved? */
1678 if (unlikely(page != xas_reload(&xas)))
1682 if (++ret == nr_pages) {
1683 *start = page->index + 1;
1694 * We come here when there is no page beyond @end. We take care to not
1695 * overflow the index @start as it confuses some of the callers. This
1696 * breaks the iteration when there is a page at index -1 but that is
1697 * already broken anyway.
1699 if (end == (pgoff_t)-1)
1700 *start = (pgoff_t)-1;
1710 * find_get_pages_contig - gang contiguous pagecache lookup
1711 * @mapping: The address_space to search
1712 * @index: The starting page index
1713 * @nr_pages: The maximum number of pages
1714 * @pages: Where the resulting pages are placed
1716 * find_get_pages_contig() works exactly like find_get_pages(), except
1717 * that the returned number of pages are guaranteed to be contiguous.
1719 * find_get_pages_contig() returns the number of pages which were found.
1721 unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index,
1722 unsigned int nr_pages, struct page **pages)
1724 XA_STATE(xas, &mapping->i_pages, index);
1726 unsigned int ret = 0;
1728 if (unlikely(!nr_pages))
1732 for (page = xas_load(&xas); page; page = xas_next(&xas)) {
1734 if (xas_retry(&xas, page))
1737 * If the entry has been swapped out, we can stop looking.
1738 * No current caller is looking for DAX entries.
1740 if (xa_is_value(page))
1743 head = compound_head(page);
1744 if (!page_cache_get_speculative(head))
1747 /* The page was split under us? */
1748 if (compound_head(page) != head)
1751 /* Has the page moved? */
1752 if (unlikely(page != xas_reload(&xas)))
1756 * must check mapping and index after taking the ref.
1757 * otherwise we can get both false positives and false
1758 * negatives, which is just confusing to the caller.
1760 if (!page->mapping || page_to_pgoff(page) != xas.xa_index) {
1766 if (++ret == nr_pages)
1777 EXPORT_SYMBOL(find_get_pages_contig);
1780 * find_get_pages_range_tag - find and return pages in given range matching @tag
1781 * @mapping: the address_space to search
1782 * @index: the starting page index
1783 * @end: The final page index (inclusive)
1784 * @tag: the tag index
1785 * @nr_pages: the maximum number of pages
1786 * @pages: where the resulting pages are placed
1788 * Like find_get_pages, except we only return pages which are tagged with
1789 * @tag. We update @index to index the next page for the traversal.
1791 unsigned find_get_pages_range_tag(struct address_space *mapping, pgoff_t *index,
1792 pgoff_t end, xa_mark_t tag, unsigned int nr_pages,
1793 struct page **pages)
1795 XA_STATE(xas, &mapping->i_pages, *index);
1799 if (unlikely(!nr_pages))
1803 xas_for_each_marked(&xas, page, end, tag) {
1805 if (xas_retry(&xas, page))
1808 * Shadow entries should never be tagged, but this iteration
1809 * is lockless so there is a window for page reclaim to evict
1810 * a page we saw tagged. Skip over it.
1812 if (xa_is_value(page))
1815 head = compound_head(page);
1816 if (!page_cache_get_speculative(head))
1819 /* The page was split under us? */
1820 if (compound_head(page) != head)
1823 /* Has the page moved? */
1824 if (unlikely(page != xas_reload(&xas)))
1828 if (++ret == nr_pages) {
1829 *index = page->index + 1;
1840 * We come here when we got to @end. We take care to not overflow the
1841 * index @index as it confuses some of the callers. This breaks the
1842 * iteration when there is a page at index -1 but that is already
1845 if (end == (pgoff_t)-1)
1846 *index = (pgoff_t)-1;
1854 EXPORT_SYMBOL(find_get_pages_range_tag);
1857 * find_get_entries_tag - find and return entries that match @tag
1858 * @mapping: the address_space to search
1859 * @start: the starting page cache index
1860 * @tag: the tag index
1861 * @nr_entries: the maximum number of entries
1862 * @entries: where the resulting entries are placed
1863 * @indices: the cache indices corresponding to the entries in @entries
1865 * Like find_get_entries, except we only return entries which are tagged with
1868 unsigned find_get_entries_tag(struct address_space *mapping, pgoff_t start,
1869 xa_mark_t tag, unsigned int nr_entries,
1870 struct page **entries, pgoff_t *indices)
1872 XA_STATE(xas, &mapping->i_pages, start);
1874 unsigned int ret = 0;
1880 xas_for_each_marked(&xas, page, ULONG_MAX, tag) {
1882 if (xas_retry(&xas, page))
1885 * A shadow entry of a recently evicted page, a swap
1886 * entry from shmem/tmpfs or a DAX entry. Return it
1887 * without attempting to raise page count.
1889 if (xa_is_value(page))
1892 head = compound_head(page);
1893 if (!page_cache_get_speculative(head))
1896 /* The page was split under us? */
1897 if (compound_head(page) != head)
1900 /* Has the page moved? */
1901 if (unlikely(page != xas_reload(&xas)))
1905 indices[ret] = xas.xa_index;
1906 entries[ret] = page;
1907 if (++ret == nr_entries)
1918 EXPORT_SYMBOL(find_get_entries_tag);
1921 * CD/DVDs are error prone. When a medium error occurs, the driver may fail
1922 * a _large_ part of the i/o request. Imagine the worst scenario:
1924 * ---R__________________________________________B__________
1925 * ^ reading here ^ bad block(assume 4k)
1927 * read(R) => miss => readahead(R...B) => media error => frustrating retries
1928 * => failing the whole request => read(R) => read(R+1) =>
1929 * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
1930 * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
1931 * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
1933 * It is going insane. Fix it by quickly scaling down the readahead size.
1935 static void shrink_readahead_size_eio(struct file *filp,
1936 struct file_ra_state *ra)
1942 * generic_file_buffered_read - generic file read routine
1943 * @iocb: the iocb to read
1944 * @iter: data destination
1945 * @written: already copied
1947 * This is a generic file read routine, and uses the
1948 * mapping->a_ops->readpage() function for the actual low-level stuff.
1950 * This is really ugly. But the goto's actually try to clarify some
1951 * of the logic when it comes to error handling etc.
1953 static ssize_t generic_file_buffered_read(struct kiocb *iocb,
1954 struct iov_iter *iter, ssize_t written)
1956 struct file *filp = iocb->ki_filp;
1957 struct address_space *mapping = filp->f_mapping;
1958 struct inode *inode = mapping->host;
1959 struct file_ra_state *ra = &filp->f_ra;
1960 loff_t *ppos = &iocb->ki_pos;
1964 unsigned long offset; /* offset into pagecache page */
1965 unsigned int prev_offset;
1968 if (unlikely(*ppos >= inode->i_sb->s_maxbytes))
1970 iov_iter_truncate(iter, inode->i_sb->s_maxbytes);
1972 index = *ppos >> PAGE_SHIFT;
1973 prev_index = ra->prev_pos >> PAGE_SHIFT;
1974 prev_offset = ra->prev_pos & (PAGE_SIZE-1);
1975 last_index = (*ppos + iter->count + PAGE_SIZE-1) >> PAGE_SHIFT;
1976 offset = *ppos & ~PAGE_MASK;
1982 unsigned long nr, ret;
1986 if (fatal_signal_pending(current)) {
1991 page = find_get_page(mapping, index);
1993 if (iocb->ki_flags & IOCB_NOWAIT)
1995 page_cache_sync_readahead(mapping,
1997 index, last_index - index);
1998 page = find_get_page(mapping, index);
1999 if (unlikely(page == NULL))
2000 goto no_cached_page;
2002 if (PageReadahead(page)) {
2003 page_cache_async_readahead(mapping,
2005 index, last_index - index);
2007 if (!PageUptodate(page)) {
2008 if (iocb->ki_flags & IOCB_NOWAIT) {
2014 * See comment in do_read_cache_page on why
2015 * wait_on_page_locked is used to avoid unnecessarily
2016 * serialisations and why it's safe.
2018 error = wait_on_page_locked_killable(page);
2019 if (unlikely(error))
2020 goto readpage_error;
2021 if (PageUptodate(page))
2024 if (inode->i_blkbits == PAGE_SHIFT ||
2025 !mapping->a_ops->is_partially_uptodate)
2026 goto page_not_up_to_date;
2027 /* pipes can't handle partially uptodate pages */
2028 if (unlikely(iter->type & ITER_PIPE))
2029 goto page_not_up_to_date;
2030 if (!trylock_page(page))
2031 goto page_not_up_to_date;
2032 /* Did it get truncated before we got the lock? */
2034 goto page_not_up_to_date_locked;
2035 if (!mapping->a_ops->is_partially_uptodate(page,
2036 offset, iter->count))
2037 goto page_not_up_to_date_locked;
2042 * i_size must be checked after we know the page is Uptodate.
2044 * Checking i_size after the check allows us to calculate
2045 * the correct value for "nr", which means the zero-filled
2046 * part of the page is not copied back to userspace (unless
2047 * another truncate extends the file - this is desired though).
2050 isize = i_size_read(inode);
2051 end_index = (isize - 1) >> PAGE_SHIFT;
2052 if (unlikely(!isize || index > end_index)) {
2057 /* nr is the maximum number of bytes to copy from this page */
2059 if (index == end_index) {
2060 nr = ((isize - 1) & ~PAGE_MASK) + 1;
2068 /* If users can be writing to this page using arbitrary
2069 * virtual addresses, take care about potential aliasing
2070 * before reading the page on the kernel side.
2072 if (mapping_writably_mapped(mapping))
2073 flush_dcache_page(page);
2076 * When a sequential read accesses a page several times,
2077 * only mark it as accessed the first time.
2079 if (prev_index != index || offset != prev_offset)
2080 mark_page_accessed(page);
2084 * Ok, we have the page, and it's up-to-date, so
2085 * now we can copy it to user space...
2088 ret = copy_page_to_iter(page, offset, nr, iter);
2090 index += offset >> PAGE_SHIFT;
2091 offset &= ~PAGE_MASK;
2092 prev_offset = offset;
2096 if (!iov_iter_count(iter))
2104 page_not_up_to_date:
2105 /* Get exclusive access to the page ... */
2106 error = lock_page_killable(page);
2107 if (unlikely(error))
2108 goto readpage_error;
2110 page_not_up_to_date_locked:
2111 /* Did it get truncated before we got the lock? */
2112 if (!page->mapping) {
2118 /* Did somebody else fill it already? */
2119 if (PageUptodate(page)) {
2126 * A previous I/O error may have been due to temporary
2127 * failures, eg. multipath errors.
2128 * PG_error will be set again if readpage fails.
2130 ClearPageError(page);
2131 /* Start the actual read. The read will unlock the page. */
2132 error = mapping->a_ops->readpage(filp, page);
2134 if (unlikely(error)) {
2135 if (error == AOP_TRUNCATED_PAGE) {
2140 goto readpage_error;
2143 if (!PageUptodate(page)) {
2144 error = lock_page_killable(page);
2145 if (unlikely(error))
2146 goto readpage_error;
2147 if (!PageUptodate(page)) {
2148 if (page->mapping == NULL) {
2150 * invalidate_mapping_pages got it
2157 shrink_readahead_size_eio(filp, ra);
2159 goto readpage_error;
2167 /* UHHUH! A synchronous read error occurred. Report it */
2173 * Ok, it wasn't cached, so we need to create a new
2176 page = page_cache_alloc(mapping);
2181 error = add_to_page_cache_lru(page, mapping, index,
2182 mapping_gfp_constraint(mapping, GFP_KERNEL));
2185 if (error == -EEXIST) {
2197 ra->prev_pos = prev_index;
2198 ra->prev_pos <<= PAGE_SHIFT;
2199 ra->prev_pos |= prev_offset;
2201 *ppos = ((loff_t)index << PAGE_SHIFT) + offset;
2202 file_accessed(filp);
2203 return written ? written : error;
2207 * generic_file_read_iter - generic filesystem read routine
2208 * @iocb: kernel I/O control block
2209 * @iter: destination for the data read
2211 * This is the "read_iter()" routine for all filesystems
2212 * that can use the page cache directly.
2215 generic_file_read_iter(struct kiocb *iocb, struct iov_iter *iter)
2217 size_t count = iov_iter_count(iter);
2221 goto out; /* skip atime */
2223 if (iocb->ki_flags & IOCB_DIRECT) {
2224 struct file *file = iocb->ki_filp;
2225 struct address_space *mapping = file->f_mapping;
2226 struct inode *inode = mapping->host;
2229 size = i_size_read(inode);
2230 if (iocb->ki_flags & IOCB_NOWAIT) {
2231 if (filemap_range_has_page(mapping, iocb->ki_pos,
2232 iocb->ki_pos + count - 1))
2235 retval = filemap_write_and_wait_range(mapping,
2237 iocb->ki_pos + count - 1);
2242 file_accessed(file);
2244 retval = mapping->a_ops->direct_IO(iocb, iter);
2246 iocb->ki_pos += retval;
2249 iov_iter_revert(iter, count - iov_iter_count(iter));
2252 * Btrfs can have a short DIO read if we encounter
2253 * compressed extents, so if there was an error, or if
2254 * we've already read everything we wanted to, or if
2255 * there was a short read because we hit EOF, go ahead
2256 * and return. Otherwise fallthrough to buffered io for
2257 * the rest of the read. Buffered reads will not work for
2258 * DAX files, so don't bother trying.
2260 if (retval < 0 || !count || iocb->ki_pos >= size ||
2265 retval = generic_file_buffered_read(iocb, iter, retval);
2269 EXPORT_SYMBOL(generic_file_read_iter);
2273 * page_cache_read - adds requested page to the page cache if not already there
2274 * @file: file to read
2275 * @offset: page index
2276 * @gfp_mask: memory allocation flags
2278 * This adds the requested page to the page cache if it isn't already there,
2279 * and schedules an I/O to read in its contents from disk.
2281 static int page_cache_read(struct file *file, pgoff_t offset, gfp_t gfp_mask)
2283 struct address_space *mapping = file->f_mapping;
2288 page = __page_cache_alloc(gfp_mask);
2292 ret = add_to_page_cache_lru(page, mapping, offset, gfp_mask);
2294 ret = mapping->a_ops->readpage(file, page);
2295 else if (ret == -EEXIST)
2296 ret = 0; /* losing race to add is OK */
2300 } while (ret == AOP_TRUNCATED_PAGE);
2305 #define MMAP_LOTSAMISS (100)
2308 * Synchronous readahead happens when we don't even find
2309 * a page in the page cache at all.
2311 static void do_sync_mmap_readahead(struct vm_area_struct *vma,
2312 struct file_ra_state *ra,
2316 struct address_space *mapping = file->f_mapping;
2318 /* If we don't want any read-ahead, don't bother */
2319 if (vma->vm_flags & VM_RAND_READ)
2324 if (vma->vm_flags & VM_SEQ_READ) {
2325 page_cache_sync_readahead(mapping, ra, file, offset,
2330 /* Avoid banging the cache line if not needed */
2331 if (ra->mmap_miss < MMAP_LOTSAMISS * 10)
2335 * Do we miss much more than hit in this file? If so,
2336 * stop bothering with read-ahead. It will only hurt.
2338 if (ra->mmap_miss > MMAP_LOTSAMISS)
2344 ra->start = max_t(long, 0, offset - ra->ra_pages / 2);
2345 ra->size = ra->ra_pages;
2346 ra->async_size = ra->ra_pages / 4;
2347 ra_submit(ra, mapping, file);
2351 * Asynchronous readahead happens when we find the page and PG_readahead,
2352 * so we want to possibly extend the readahead further..
2354 static void do_async_mmap_readahead(struct vm_area_struct *vma,
2355 struct file_ra_state *ra,
2360 struct address_space *mapping = file->f_mapping;
2362 /* If we don't want any read-ahead, don't bother */
2363 if (vma->vm_flags & VM_RAND_READ)
2365 if (ra->mmap_miss > 0)
2367 if (PageReadahead(page))
2368 page_cache_async_readahead(mapping, ra, file,
2369 page, offset, ra->ra_pages);
2373 * filemap_fault - read in file data for page fault handling
2374 * @vmf: struct vm_fault containing details of the fault
2376 * filemap_fault() is invoked via the vma operations vector for a
2377 * mapped memory region to read in file data during a page fault.
2379 * The goto's are kind of ugly, but this streamlines the normal case of having
2380 * it in the page cache, and handles the special cases reasonably without
2381 * having a lot of duplicated code.
2383 * vma->vm_mm->mmap_sem must be held on entry.
2385 * If our return value has VM_FAULT_RETRY set, it's because
2386 * lock_page_or_retry() returned 0.
2387 * The mmap_sem has usually been released in this case.
2388 * See __lock_page_or_retry() for the exception.
2390 * If our return value does not have VM_FAULT_RETRY set, the mmap_sem
2391 * has not been released.
2393 * We never return with VM_FAULT_RETRY and a bit from VM_FAULT_ERROR set.
2395 vm_fault_t filemap_fault(struct vm_fault *vmf)
2398 struct file *file = vmf->vma->vm_file;
2399 struct address_space *mapping = file->f_mapping;
2400 struct file_ra_state *ra = &file->f_ra;
2401 struct inode *inode = mapping->host;
2402 pgoff_t offset = vmf->pgoff;
2407 max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
2408 if (unlikely(offset >= max_off))
2409 return VM_FAULT_SIGBUS;
2412 * Do we have something in the page cache already?
2414 page = find_get_page(mapping, offset);
2415 if (likely(page) && !(vmf->flags & FAULT_FLAG_TRIED)) {
2417 * We found the page, so try async readahead before
2418 * waiting for the lock.
2420 do_async_mmap_readahead(vmf->vma, ra, file, page, offset);
2422 /* No page in the page cache at all */
2423 do_sync_mmap_readahead(vmf->vma, ra, file, offset);
2424 count_vm_event(PGMAJFAULT);
2425 count_memcg_event_mm(vmf->vma->vm_mm, PGMAJFAULT);
2426 ret = VM_FAULT_MAJOR;
2428 page = find_get_page(mapping, offset);
2430 goto no_cached_page;
2433 if (!lock_page_or_retry(page, vmf->vma->vm_mm, vmf->flags)) {
2435 return ret | VM_FAULT_RETRY;
2438 /* Did it get truncated? */
2439 if (unlikely(page->mapping != mapping)) {
2444 VM_BUG_ON_PAGE(page->index != offset, page);
2447 * We have a locked page in the page cache, now we need to check
2448 * that it's up-to-date. If not, it is going to be due to an error.
2450 if (unlikely(!PageUptodate(page)))
2451 goto page_not_uptodate;
2454 * Found the page and have a reference on it.
2455 * We must recheck i_size under page lock.
2457 max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
2458 if (unlikely(offset >= max_off)) {
2461 return VM_FAULT_SIGBUS;
2465 return ret | VM_FAULT_LOCKED;
2469 * We're only likely to ever get here if MADV_RANDOM is in
2472 error = page_cache_read(file, offset, vmf->gfp_mask);
2475 * The page we want has now been added to the page cache.
2476 * In the unlikely event that someone removed it in the
2477 * meantime, we'll just come back here and read it again.
2483 * An error return from page_cache_read can result if the
2484 * system is low on memory, or a problem occurs while trying
2487 if (error == -ENOMEM)
2488 return VM_FAULT_OOM;
2489 return VM_FAULT_SIGBUS;
2493 * Umm, take care of errors if the page isn't up-to-date.
2494 * Try to re-read it _once_. We do this synchronously,
2495 * because there really aren't any performance issues here
2496 * and we need to check for errors.
2498 ClearPageError(page);
2499 error = mapping->a_ops->readpage(file, page);
2501 wait_on_page_locked(page);
2502 if (!PageUptodate(page))
2507 if (!error || error == AOP_TRUNCATED_PAGE)
2510 /* Things didn't work out. Return zero to tell the mm layer so. */
2511 shrink_readahead_size_eio(file, ra);
2512 return VM_FAULT_SIGBUS;
2514 EXPORT_SYMBOL(filemap_fault);
2516 void filemap_map_pages(struct vm_fault *vmf,
2517 pgoff_t start_pgoff, pgoff_t end_pgoff)
2519 struct file *file = vmf->vma->vm_file;
2520 struct address_space *mapping = file->f_mapping;
2521 pgoff_t last_pgoff = start_pgoff;
2522 unsigned long max_idx;
2523 XA_STATE(xas, &mapping->i_pages, start_pgoff);
2524 struct page *head, *page;
2527 xas_for_each(&xas, page, end_pgoff) {
2528 if (xas_retry(&xas, page))
2530 if (xa_is_value(page))
2533 head = compound_head(page);
2534 if (!page_cache_get_speculative(head))
2537 /* The page was split under us? */
2538 if (compound_head(page) != head)
2541 /* Has the page moved? */
2542 if (unlikely(page != xas_reload(&xas)))
2545 if (!PageUptodate(page) ||
2546 PageReadahead(page) ||
2549 if (!trylock_page(page))
2552 if (page->mapping != mapping || !PageUptodate(page))
2555 max_idx = DIV_ROUND_UP(i_size_read(mapping->host), PAGE_SIZE);
2556 if (page->index >= max_idx)
2559 if (file->f_ra.mmap_miss > 0)
2560 file->f_ra.mmap_miss--;
2562 vmf->address += (xas.xa_index - last_pgoff) << PAGE_SHIFT;
2564 vmf->pte += xas.xa_index - last_pgoff;
2565 last_pgoff = xas.xa_index;
2566 if (alloc_set_pte(vmf, NULL, page))
2575 /* Huge page is mapped? No need to proceed. */
2576 if (pmd_trans_huge(*vmf->pmd))
2581 EXPORT_SYMBOL(filemap_map_pages);
2583 vm_fault_t filemap_page_mkwrite(struct vm_fault *vmf)
2585 struct page *page = vmf->page;
2586 struct inode *inode = file_inode(vmf->vma->vm_file);
2587 vm_fault_t ret = VM_FAULT_LOCKED;
2589 sb_start_pagefault(inode->i_sb);
2590 file_update_time(vmf->vma->vm_file);
2592 if (page->mapping != inode->i_mapping) {
2594 ret = VM_FAULT_NOPAGE;
2598 * We mark the page dirty already here so that when freeze is in
2599 * progress, we are guaranteed that writeback during freezing will
2600 * see the dirty page and writeprotect it again.
2602 set_page_dirty(page);
2603 wait_for_stable_page(page);
2605 sb_end_pagefault(inode->i_sb);
2609 const struct vm_operations_struct generic_file_vm_ops = {
2610 .fault = filemap_fault,
2611 .map_pages = filemap_map_pages,
2612 .page_mkwrite = filemap_page_mkwrite,
2615 /* This is used for a general mmap of a disk file */
2617 int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
2619 struct address_space *mapping = file->f_mapping;
2621 if (!mapping->a_ops->readpage)
2623 file_accessed(file);
2624 vma->vm_ops = &generic_file_vm_ops;
2629 * This is for filesystems which do not implement ->writepage.
2631 int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
2633 if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE))
2635 return generic_file_mmap(file, vma);
2638 int filemap_page_mkwrite(struct vm_fault *vmf)
2642 int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
2646 int generic_file_readonly_mmap(struct file * file, struct vm_area_struct * vma)
2650 #endif /* CONFIG_MMU */
2652 EXPORT_SYMBOL(filemap_page_mkwrite);
2653 EXPORT_SYMBOL(generic_file_mmap);
2654 EXPORT_SYMBOL(generic_file_readonly_mmap);
2656 static struct page *wait_on_page_read(struct page *page)
2658 if (!IS_ERR(page)) {
2659 wait_on_page_locked(page);
2660 if (!PageUptodate(page)) {
2662 page = ERR_PTR(-EIO);
2668 static struct page *do_read_cache_page(struct address_space *mapping,
2670 int (*filler)(void *, struct page *),
2677 page = find_get_page(mapping, index);
2679 page = __page_cache_alloc(gfp);
2681 return ERR_PTR(-ENOMEM);
2682 err = add_to_page_cache_lru(page, mapping, index, gfp);
2683 if (unlikely(err)) {
2687 /* Presumably ENOMEM for radix tree node */
2688 return ERR_PTR(err);
2692 err = filler(data, page);
2695 return ERR_PTR(err);
2698 page = wait_on_page_read(page);
2703 if (PageUptodate(page))
2707 * Page is not up to date and may be locked due one of the following
2708 * case a: Page is being filled and the page lock is held
2709 * case b: Read/write error clearing the page uptodate status
2710 * case c: Truncation in progress (page locked)
2711 * case d: Reclaim in progress
2713 * Case a, the page will be up to date when the page is unlocked.
2714 * There is no need to serialise on the page lock here as the page
2715 * is pinned so the lock gives no additional protection. Even if the
2716 * the page is truncated, the data is still valid if PageUptodate as
2717 * it's a race vs truncate race.
2718 * Case b, the page will not be up to date
2719 * Case c, the page may be truncated but in itself, the data may still
2720 * be valid after IO completes as it's a read vs truncate race. The
2721 * operation must restart if the page is not uptodate on unlock but
2722 * otherwise serialising on page lock to stabilise the mapping gives
2723 * no additional guarantees to the caller as the page lock is
2724 * released before return.
2725 * Case d, similar to truncation. If reclaim holds the page lock, it
2726 * will be a race with remove_mapping that determines if the mapping
2727 * is valid on unlock but otherwise the data is valid and there is
2728 * no need to serialise with page lock.
2730 * As the page lock gives no additional guarantee, we optimistically
2731 * wait on the page to be unlocked and check if it's up to date and
2732 * use the page if it is. Otherwise, the page lock is required to
2733 * distinguish between the different cases. The motivation is that we
2734 * avoid spurious serialisations and wakeups when multiple processes
2735 * wait on the same page for IO to complete.
2737 wait_on_page_locked(page);
2738 if (PageUptodate(page))
2741 /* Distinguish between all the cases under the safety of the lock */
2744 /* Case c or d, restart the operation */
2745 if (!page->mapping) {
2751 /* Someone else locked and filled the page in a very small window */
2752 if (PageUptodate(page)) {
2759 mark_page_accessed(page);
2764 * read_cache_page - read into page cache, fill it if needed
2765 * @mapping: the page's address_space
2766 * @index: the page index
2767 * @filler: function to perform the read
2768 * @data: first arg to filler(data, page) function, often left as NULL
2770 * Read into the page cache. If a page already exists, and PageUptodate() is
2771 * not set, try to fill the page and wait for it to become unlocked.
2773 * If the page does not get brought uptodate, return -EIO.
2775 struct page *read_cache_page(struct address_space *mapping,
2777 int (*filler)(void *, struct page *),
2780 return do_read_cache_page(mapping, index, filler, data, mapping_gfp_mask(mapping));
2782 EXPORT_SYMBOL(read_cache_page);
2785 * read_cache_page_gfp - read into page cache, using specified page allocation flags.
2786 * @mapping: the page's address_space
2787 * @index: the page index
2788 * @gfp: the page allocator flags to use if allocating
2790 * This is the same as "read_mapping_page(mapping, index, NULL)", but with
2791 * any new page allocations done using the specified allocation flags.
2793 * If the page does not get brought uptodate, return -EIO.
2795 struct page *read_cache_page_gfp(struct address_space *mapping,
2799 filler_t *filler = (filler_t *)mapping->a_ops->readpage;
2801 return do_read_cache_page(mapping, index, filler, NULL, gfp);
2803 EXPORT_SYMBOL(read_cache_page_gfp);
2806 * Performs necessary checks before doing a write
2808 * Can adjust writing position or amount of bytes to write.
2809 * Returns appropriate error code that caller should return or
2810 * zero in case that write should be allowed.
2812 inline ssize_t generic_write_checks(struct kiocb *iocb, struct iov_iter *from)
2814 struct file *file = iocb->ki_filp;
2815 struct inode *inode = file->f_mapping->host;
2816 unsigned long limit = rlimit(RLIMIT_FSIZE);
2819 if (!iov_iter_count(from))
2822 /* FIXME: this is for backwards compatibility with 2.4 */
2823 if (iocb->ki_flags & IOCB_APPEND)
2824 iocb->ki_pos = i_size_read(inode);
2828 if ((iocb->ki_flags & IOCB_NOWAIT) && !(iocb->ki_flags & IOCB_DIRECT))
2831 if (limit != RLIM_INFINITY) {
2832 if (iocb->ki_pos >= limit) {
2833 send_sig(SIGXFSZ, current, 0);
2836 iov_iter_truncate(from, limit - (unsigned long)pos);
2842 if (unlikely(pos + iov_iter_count(from) > MAX_NON_LFS &&
2843 !(file->f_flags & O_LARGEFILE))) {
2844 if (pos >= MAX_NON_LFS)
2846 iov_iter_truncate(from, MAX_NON_LFS - (unsigned long)pos);
2850 * Are we about to exceed the fs block limit ?
2852 * If we have written data it becomes a short write. If we have
2853 * exceeded without writing data we send a signal and return EFBIG.
2854 * Linus frestrict idea will clean these up nicely..
2856 if (unlikely(pos >= inode->i_sb->s_maxbytes))
2859 iov_iter_truncate(from, inode->i_sb->s_maxbytes - pos);
2860 return iov_iter_count(from);
2862 EXPORT_SYMBOL(generic_write_checks);
2864 int pagecache_write_begin(struct file *file, struct address_space *mapping,
2865 loff_t pos, unsigned len, unsigned flags,
2866 struct page **pagep, void **fsdata)
2868 const struct address_space_operations *aops = mapping->a_ops;
2870 return aops->write_begin(file, mapping, pos, len, flags,
2873 EXPORT_SYMBOL(pagecache_write_begin);
2875 int pagecache_write_end(struct file *file, struct address_space *mapping,
2876 loff_t pos, unsigned len, unsigned copied,
2877 struct page *page, void *fsdata)
2879 const struct address_space_operations *aops = mapping->a_ops;
2881 return aops->write_end(file, mapping, pos, len, copied, page, fsdata);
2883 EXPORT_SYMBOL(pagecache_write_end);
2886 generic_file_direct_write(struct kiocb *iocb, struct iov_iter *from)
2888 struct file *file = iocb->ki_filp;
2889 struct address_space *mapping = file->f_mapping;
2890 struct inode *inode = mapping->host;
2891 loff_t pos = iocb->ki_pos;
2896 write_len = iov_iter_count(from);
2897 end = (pos + write_len - 1) >> PAGE_SHIFT;
2899 if (iocb->ki_flags & IOCB_NOWAIT) {
2900 /* If there are pages to writeback, return */
2901 if (filemap_range_has_page(inode->i_mapping, pos,
2902 pos + iov_iter_count(from)))
2905 written = filemap_write_and_wait_range(mapping, pos,
2906 pos + write_len - 1);
2912 * After a write we want buffered reads to be sure to go to disk to get
2913 * the new data. We invalidate clean cached page from the region we're
2914 * about to write. We do this *before* the write so that we can return
2915 * without clobbering -EIOCBQUEUED from ->direct_IO().
2917 written = invalidate_inode_pages2_range(mapping,
2918 pos >> PAGE_SHIFT, end);
2920 * If a page can not be invalidated, return 0 to fall back
2921 * to buffered write.
2924 if (written == -EBUSY)
2929 written = mapping->a_ops->direct_IO(iocb, from);
2932 * Finally, try again to invalidate clean pages which might have been
2933 * cached by non-direct readahead, or faulted in by get_user_pages()
2934 * if the source of the write was an mmap'ed region of the file
2935 * we're writing. Either one is a pretty crazy thing to do,
2936 * so we don't support it 100%. If this invalidation
2937 * fails, tough, the write still worked...
2939 * Most of the time we do not need this since dio_complete() will do
2940 * the invalidation for us. However there are some file systems that
2941 * do not end up with dio_complete() being called, so let's not break
2942 * them by removing it completely
2944 if (mapping->nrpages)
2945 invalidate_inode_pages2_range(mapping,
2946 pos >> PAGE_SHIFT, end);
2950 write_len -= written;
2951 if (pos > i_size_read(inode) && !S_ISBLK(inode->i_mode)) {
2952 i_size_write(inode, pos);
2953 mark_inode_dirty(inode);
2957 iov_iter_revert(from, write_len - iov_iter_count(from));
2961 EXPORT_SYMBOL(generic_file_direct_write);
2964 * Find or create a page at the given pagecache position. Return the locked
2965 * page. This function is specifically for buffered writes.
2967 struct page *grab_cache_page_write_begin(struct address_space *mapping,
2968 pgoff_t index, unsigned flags)
2971 int fgp_flags = FGP_LOCK|FGP_WRITE|FGP_CREAT;
2973 if (flags & AOP_FLAG_NOFS)
2974 fgp_flags |= FGP_NOFS;
2976 page = pagecache_get_page(mapping, index, fgp_flags,
2977 mapping_gfp_mask(mapping));
2979 wait_for_stable_page(page);
2983 EXPORT_SYMBOL(grab_cache_page_write_begin);
2985 ssize_t generic_perform_write(struct file *file,
2986 struct iov_iter *i, loff_t pos)
2988 struct address_space *mapping = file->f_mapping;
2989 const struct address_space_operations *a_ops = mapping->a_ops;
2991 ssize_t written = 0;
2992 unsigned int flags = 0;
2996 unsigned long offset; /* Offset into pagecache page */
2997 unsigned long bytes; /* Bytes to write to page */
2998 size_t copied; /* Bytes copied from user */
3001 offset = (pos & (PAGE_SIZE - 1));
3002 bytes = min_t(unsigned long, PAGE_SIZE - offset,
3007 * Bring in the user page that we will copy from _first_.
3008 * Otherwise there's a nasty deadlock on copying from the
3009 * same page as we're writing to, without it being marked
3012 * Not only is this an optimisation, but it is also required
3013 * to check that the address is actually valid, when atomic
3014 * usercopies are used, below.
3016 if (unlikely(iov_iter_fault_in_readable(i, bytes))) {
3021 if (fatal_signal_pending(current)) {
3026 status = a_ops->write_begin(file, mapping, pos, bytes, flags,
3028 if (unlikely(status < 0))
3031 if (mapping_writably_mapped(mapping))
3032 flush_dcache_page(page);
3034 copied = iov_iter_copy_from_user_atomic(page, i, offset, bytes);
3035 flush_dcache_page(page);
3037 status = a_ops->write_end(file, mapping, pos, bytes, copied,
3039 if (unlikely(status < 0))
3045 iov_iter_advance(i, copied);
3046 if (unlikely(copied == 0)) {
3048 * If we were unable to copy any data at all, we must
3049 * fall back to a single segment length write.
3051 * If we didn't fallback here, we could livelock
3052 * because not all segments in the iov can be copied at
3053 * once without a pagefault.
3055 bytes = min_t(unsigned long, PAGE_SIZE - offset,
3056 iov_iter_single_seg_count(i));
3062 balance_dirty_pages_ratelimited(mapping);
3063 } while (iov_iter_count(i));
3065 return written ? written : status;
3067 EXPORT_SYMBOL(generic_perform_write);
3070 * __generic_file_write_iter - write data to a file
3071 * @iocb: IO state structure (file, offset, etc.)
3072 * @from: iov_iter with data to write
3074 * This function does all the work needed for actually writing data to a
3075 * file. It does all basic checks, removes SUID from the file, updates
3076 * modification times and calls proper subroutines depending on whether we
3077 * do direct IO or a standard buffered write.
3079 * It expects i_mutex to be grabbed unless we work on a block device or similar
3080 * object which does not need locking at all.
3082 * This function does *not* take care of syncing data in case of O_SYNC write.
3083 * A caller has to handle it. This is mainly due to the fact that we want to
3084 * avoid syncing under i_mutex.
3086 ssize_t __generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
3088 struct file *file = iocb->ki_filp;
3089 struct address_space * mapping = file->f_mapping;
3090 struct inode *inode = mapping->host;
3091 ssize_t written = 0;
3095 /* We can write back this queue in page reclaim */
3096 current->backing_dev_info = inode_to_bdi(inode);
3097 err = file_remove_privs(file);
3101 err = file_update_time(file);
3105 if (iocb->ki_flags & IOCB_DIRECT) {
3106 loff_t pos, endbyte;
3108 written = generic_file_direct_write(iocb, from);
3110 * If the write stopped short of completing, fall back to
3111 * buffered writes. Some filesystems do this for writes to
3112 * holes, for example. For DAX files, a buffered write will
3113 * not succeed (even if it did, DAX does not handle dirty
3114 * page-cache pages correctly).
3116 if (written < 0 || !iov_iter_count(from) || IS_DAX(inode))
3119 status = generic_perform_write(file, from, pos = iocb->ki_pos);
3121 * If generic_perform_write() returned a synchronous error
3122 * then we want to return the number of bytes which were
3123 * direct-written, or the error code if that was zero. Note
3124 * that this differs from normal direct-io semantics, which
3125 * will return -EFOO even if some bytes were written.
3127 if (unlikely(status < 0)) {
3132 * We need to ensure that the page cache pages are written to
3133 * disk and invalidated to preserve the expected O_DIRECT
3136 endbyte = pos + status - 1;
3137 err = filemap_write_and_wait_range(mapping, pos, endbyte);
3139 iocb->ki_pos = endbyte + 1;
3141 invalidate_mapping_pages(mapping,
3143 endbyte >> PAGE_SHIFT);
3146 * We don't know how much we wrote, so just return
3147 * the number of bytes which were direct-written
3151 written = generic_perform_write(file, from, iocb->ki_pos);
3152 if (likely(written > 0))
3153 iocb->ki_pos += written;
3156 current->backing_dev_info = NULL;
3157 return written ? written : err;
3159 EXPORT_SYMBOL(__generic_file_write_iter);
3162 * generic_file_write_iter - write data to a file
3163 * @iocb: IO state structure
3164 * @from: iov_iter with data to write
3166 * This is a wrapper around __generic_file_write_iter() to be used by most
3167 * filesystems. It takes care of syncing the file in case of O_SYNC file
3168 * and acquires i_mutex as needed.
3170 ssize_t generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
3172 struct file *file = iocb->ki_filp;
3173 struct inode *inode = file->f_mapping->host;
3177 ret = generic_write_checks(iocb, from);
3179 ret = __generic_file_write_iter(iocb, from);
3180 inode_unlock(inode);
3183 ret = generic_write_sync(iocb, ret);
3186 EXPORT_SYMBOL(generic_file_write_iter);
3189 * try_to_release_page() - release old fs-specific metadata on a page
3191 * @page: the page which the kernel is trying to free
3192 * @gfp_mask: memory allocation flags (and I/O mode)
3194 * The address_space is to try to release any data against the page
3195 * (presumably at page->private). If the release was successful, return '1'.
3196 * Otherwise return zero.
3198 * This may also be called if PG_fscache is set on a page, indicating that the
3199 * page is known to the local caching routines.
3201 * The @gfp_mask argument specifies whether I/O may be performed to release
3202 * this page (__GFP_IO), and whether the call may block (__GFP_RECLAIM & __GFP_FS).
3205 int try_to_release_page(struct page *page, gfp_t gfp_mask)
3207 struct address_space * const mapping = page->mapping;
3209 BUG_ON(!PageLocked(page));
3210 if (PageWriteback(page))
3213 if (mapping && mapping->a_ops->releasepage)
3214 return mapping->a_ops->releasepage(page, gfp_mask);
3215 return try_to_free_buffers(page);
3218 EXPORT_SYMBOL(try_to_release_page);