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/uaccess.h>
17 #include <linux/capability.h>
18 #include <linux/kernel_stat.h>
19 #include <linux/gfp.h>
21 #include <linux/swap.h>
22 #include <linux/mman.h>
23 #include <linux/pagemap.h>
24 #include <linux/file.h>
25 #include <linux/uio.h>
26 #include <linux/hash.h>
27 #include <linux/writeback.h>
28 #include <linux/backing-dev.h>
29 #include <linux/pagevec.h>
30 #include <linux/blkdev.h>
31 #include <linux/security.h>
32 #include <linux/cpuset.h>
33 #include <linux/hardirq.h> /* for BUG_ON(!in_atomic()) only */
34 #include <linux/hugetlb.h>
35 #include <linux/memcontrol.h>
36 #include <linux/cleancache.h>
37 #include <linux/rmap.h>
40 #define CREATE_TRACE_POINTS
41 #include <trace/events/filemap.h>
44 * FIXME: remove all knowledge of the buffer layer from the core VM
46 #include <linux/buffer_head.h> /* for try_to_free_buffers */
51 * Shared mappings implemented 30.11.1994. It's not fully working yet,
54 * Shared mappings now work. 15.8.1995 Bruno.
56 * finished 'unifying' the page and buffer cache and SMP-threaded the
57 * page-cache, 21.05.1999, Ingo Molnar <mingo@redhat.com>
59 * SMP-threaded pagemap-LRU 1999, Andrea Arcangeli <andrea@suse.de>
65 * ->i_mmap_rwsem (truncate_pagecache)
66 * ->private_lock (__free_pte->__set_page_dirty_buffers)
67 * ->swap_lock (exclusive_swap_page, others)
68 * ->mapping->tree_lock
71 * ->i_mmap_rwsem (truncate->unmap_mapping_range)
75 * ->page_table_lock or pte_lock (various, mainly in memory.c)
76 * ->mapping->tree_lock (arch-dependent flush_dcache_mmap_lock)
79 * ->lock_page (access_process_vm)
81 * ->i_mutex (generic_perform_write)
82 * ->mmap_sem (fault_in_pages_readable->do_page_fault)
85 * sb_lock (fs/fs-writeback.c)
86 * ->mapping->tree_lock (__sync_single_inode)
89 * ->anon_vma.lock (vma_adjust)
92 * ->page_table_lock or pte_lock (anon_vma_prepare and various)
94 * ->page_table_lock or pte_lock
95 * ->swap_lock (try_to_unmap_one)
96 * ->private_lock (try_to_unmap_one)
97 * ->tree_lock (try_to_unmap_one)
98 * ->zone.lru_lock (follow_page->mark_page_accessed)
99 * ->zone.lru_lock (check_pte_range->isolate_lru_page)
100 * ->private_lock (page_remove_rmap->set_page_dirty)
101 * ->tree_lock (page_remove_rmap->set_page_dirty)
102 * bdi.wb->list_lock (page_remove_rmap->set_page_dirty)
103 * ->inode->i_lock (page_remove_rmap->set_page_dirty)
104 * ->memcg->move_lock (page_remove_rmap->lock_page_memcg)
105 * bdi.wb->list_lock (zap_pte_range->set_page_dirty)
106 * ->inode->i_lock (zap_pte_range->set_page_dirty)
107 * ->private_lock (zap_pte_range->__set_page_dirty_buffers)
110 * ->tasklist_lock (memory_failure, collect_procs_ao)
113 static void page_cache_tree_delete(struct address_space *mapping,
114 struct page *page, void *shadow)
116 struct radix_tree_node *node;
122 VM_BUG_ON(!PageLocked(page));
124 __radix_tree_lookup(&mapping->page_tree, page->index, &node, &slot);
127 mapping->nrexceptional++;
129 * Make sure the nrexceptional update is committed before
130 * the nrpages update so that final truncate racing
131 * with reclaim does not see both counters 0 at the
132 * same time and miss a shadow entry.
139 /* Clear direct pointer tags in root node */
140 mapping->page_tree.gfp_mask &= __GFP_BITS_MASK;
141 radix_tree_replace_slot(slot, shadow);
145 /* Clear tree tags for the removed page */
147 offset = index & RADIX_TREE_MAP_MASK;
148 for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++) {
149 if (test_bit(offset, node->tags[tag]))
150 radix_tree_tag_clear(&mapping->page_tree, index, tag);
153 /* Delete page, swap shadow entry */
154 radix_tree_replace_slot(slot, shadow);
155 workingset_node_pages_dec(node);
157 workingset_node_shadows_inc(node);
159 if (__radix_tree_delete_node(&mapping->page_tree, node))
163 * Track node that only contains shadow entries.
165 * Avoid acquiring the list_lru lock if already tracked. The
166 * list_empty() test is safe as node->private_list is
167 * protected by mapping->tree_lock.
169 if (!workingset_node_pages(node) &&
170 list_empty(&node->private_list)) {
171 node->private_data = mapping;
172 list_lru_add(&workingset_shadow_nodes, &node->private_list);
177 * Delete a page from the page cache and free it. Caller has to make
178 * sure the page is locked and that nobody else uses it - or that usage
179 * is safe. The caller must hold the mapping's tree_lock and
182 void __delete_from_page_cache(struct page *page, void *shadow)
184 struct address_space *mapping = page->mapping;
186 trace_mm_filemap_delete_from_page_cache(page);
188 * if we're uptodate, flush out into the cleancache, otherwise
189 * invalidate any existing cleancache entries. We can't leave
190 * stale data around in the cleancache once our page is gone
192 if (PageUptodate(page) && PageMappedToDisk(page))
193 cleancache_put_page(page);
195 cleancache_invalidate_page(mapping, page);
197 VM_BUG_ON_PAGE(page_mapped(page), page);
198 if (!IS_ENABLED(CONFIG_DEBUG_VM) && unlikely(page_mapped(page))) {
201 pr_alert("BUG: Bad page cache in process %s pfn:%05lx\n",
202 current->comm, page_to_pfn(page));
203 dump_page(page, "still mapped when deleted");
205 add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE);
207 mapcount = page_mapcount(page);
208 if (mapping_exiting(mapping) &&
209 page_count(page) >= mapcount + 2) {
211 * All vmas have already been torn down, so it's
212 * a good bet that actually the page is unmapped,
213 * and we'd prefer not to leak it: if we're wrong,
214 * some other bad page check should catch it later.
216 page_mapcount_reset(page);
217 atomic_sub(mapcount, &page->_count);
221 page_cache_tree_delete(mapping, page, shadow);
223 page->mapping = NULL;
224 /* Leave page->index set: truncation lookup relies upon it */
226 /* hugetlb pages do not participate in page cache accounting. */
228 __dec_zone_page_state(page, NR_FILE_PAGES);
229 if (PageSwapBacked(page))
230 __dec_zone_page_state(page, NR_SHMEM);
233 * At this point page must be either written or cleaned by truncate.
234 * Dirty page here signals a bug and loss of unwritten data.
236 * This fixes dirty accounting after removing the page entirely but
237 * leaves PageDirty set: it has no effect for truncated page and
238 * anyway will be cleared before returning page into buddy allocator.
240 if (WARN_ON_ONCE(PageDirty(page)))
241 account_page_cleaned(page, mapping, inode_to_wb(mapping->host));
245 * delete_from_page_cache - delete page from page cache
246 * @page: the page which the kernel is trying to remove from page cache
248 * This must be called only on pages that have been verified to be in the page
249 * cache and locked. It will never put the page into the free list, the caller
250 * has a reference on the page.
252 void delete_from_page_cache(struct page *page)
254 struct address_space *mapping = page->mapping;
257 void (*freepage)(struct page *);
259 BUG_ON(!PageLocked(page));
261 freepage = mapping->a_ops->freepage;
263 lock_page_memcg(page);
264 spin_lock_irqsave(&mapping->tree_lock, flags);
265 __delete_from_page_cache(page, NULL);
266 spin_unlock_irqrestore(&mapping->tree_lock, flags);
267 unlock_page_memcg(page);
271 page_cache_release(page);
273 EXPORT_SYMBOL(delete_from_page_cache);
275 static int filemap_check_errors(struct address_space *mapping)
278 /* Check for outstanding write errors */
279 if (test_bit(AS_ENOSPC, &mapping->flags) &&
280 test_and_clear_bit(AS_ENOSPC, &mapping->flags))
282 if (test_bit(AS_EIO, &mapping->flags) &&
283 test_and_clear_bit(AS_EIO, &mapping->flags))
289 * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range
290 * @mapping: address space structure to write
291 * @start: offset in bytes where the range starts
292 * @end: offset in bytes where the range ends (inclusive)
293 * @sync_mode: enable synchronous operation
295 * Start writeback against all of a mapping's dirty pages that lie
296 * within the byte offsets <start, end> inclusive.
298 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
299 * opposed to a regular memory cleansing writeback. The difference between
300 * these two operations is that if a dirty page/buffer is encountered, it must
301 * be waited upon, and not just skipped over.
303 int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
304 loff_t end, int sync_mode)
307 struct writeback_control wbc = {
308 .sync_mode = sync_mode,
309 .nr_to_write = LONG_MAX,
310 .range_start = start,
314 if (!mapping_cap_writeback_dirty(mapping))
317 wbc_attach_fdatawrite_inode(&wbc, mapping->host);
318 ret = do_writepages(mapping, &wbc);
319 wbc_detach_inode(&wbc);
323 static inline int __filemap_fdatawrite(struct address_space *mapping,
326 return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode);
329 int filemap_fdatawrite(struct address_space *mapping)
331 return __filemap_fdatawrite(mapping, WB_SYNC_ALL);
333 EXPORT_SYMBOL(filemap_fdatawrite);
335 int filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
338 return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL);
340 EXPORT_SYMBOL(filemap_fdatawrite_range);
343 * filemap_flush - mostly a non-blocking flush
344 * @mapping: target address_space
346 * This is a mostly non-blocking flush. Not suitable for data-integrity
347 * purposes - I/O may not be started against all dirty pages.
349 int filemap_flush(struct address_space *mapping)
351 return __filemap_fdatawrite(mapping, WB_SYNC_NONE);
353 EXPORT_SYMBOL(filemap_flush);
355 static int __filemap_fdatawait_range(struct address_space *mapping,
356 loff_t start_byte, loff_t end_byte)
358 pgoff_t index = start_byte >> PAGE_CACHE_SHIFT;
359 pgoff_t end = end_byte >> PAGE_CACHE_SHIFT;
364 if (end_byte < start_byte)
367 pagevec_init(&pvec, 0);
368 while ((index <= end) &&
369 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
370 PAGECACHE_TAG_WRITEBACK,
371 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1)) != 0) {
374 for (i = 0; i < nr_pages; i++) {
375 struct page *page = pvec.pages[i];
377 /* until radix tree lookup accepts end_index */
378 if (page->index > end)
381 wait_on_page_writeback(page);
382 if (TestClearPageError(page))
385 pagevec_release(&pvec);
393 * filemap_fdatawait_range - wait for writeback to complete
394 * @mapping: address space structure to wait for
395 * @start_byte: offset in bytes where the range starts
396 * @end_byte: offset in bytes where the range ends (inclusive)
398 * Walk the list of under-writeback pages of the given address space
399 * in the given range and wait for all of them. Check error status of
400 * the address space and return it.
402 * Since the error status of the address space is cleared by this function,
403 * callers are responsible for checking the return value and handling and/or
404 * reporting the error.
406 int filemap_fdatawait_range(struct address_space *mapping, loff_t start_byte,
411 ret = __filemap_fdatawait_range(mapping, start_byte, end_byte);
412 ret2 = filemap_check_errors(mapping);
418 EXPORT_SYMBOL(filemap_fdatawait_range);
421 * filemap_fdatawait_keep_errors - wait for writeback without clearing errors
422 * @mapping: address space structure to wait for
424 * Walk the list of under-writeback pages of the given address space
425 * and wait for all of them. Unlike filemap_fdatawait(), this function
426 * does not clear error status of the address space.
428 * Use this function if callers don't handle errors themselves. Expected
429 * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2),
432 void filemap_fdatawait_keep_errors(struct address_space *mapping)
434 loff_t i_size = i_size_read(mapping->host);
439 __filemap_fdatawait_range(mapping, 0, i_size - 1);
443 * filemap_fdatawait - wait for all under-writeback pages to complete
444 * @mapping: address space structure to wait for
446 * Walk the list of under-writeback pages of the given address space
447 * and wait for all of them. Check error status of the address space
450 * Since the error status of the address space is cleared by this function,
451 * callers are responsible for checking the return value and handling and/or
452 * reporting the error.
454 int filemap_fdatawait(struct address_space *mapping)
456 loff_t i_size = i_size_read(mapping->host);
461 return filemap_fdatawait_range(mapping, 0, i_size - 1);
463 EXPORT_SYMBOL(filemap_fdatawait);
465 int filemap_write_and_wait(struct address_space *mapping)
469 if ((!dax_mapping(mapping) && mapping->nrpages) ||
470 (dax_mapping(mapping) && mapping->nrexceptional)) {
471 err = filemap_fdatawrite(mapping);
473 * Even if the above returned error, the pages may be
474 * written partially (e.g. -ENOSPC), so we wait for it.
475 * But the -EIO is special case, it may indicate the worst
476 * thing (e.g. bug) happened, so we avoid waiting for it.
479 int err2 = filemap_fdatawait(mapping);
484 err = filemap_check_errors(mapping);
488 EXPORT_SYMBOL(filemap_write_and_wait);
491 * filemap_write_and_wait_range - write out & wait on a file range
492 * @mapping: the address_space for the pages
493 * @lstart: offset in bytes where the range starts
494 * @lend: offset in bytes where the range ends (inclusive)
496 * Write out and wait upon file offsets lstart->lend, inclusive.
498 * Note that `lend' is inclusive (describes the last byte to be written) so
499 * that this function can be used to write to the very end-of-file (end = -1).
501 int filemap_write_and_wait_range(struct address_space *mapping,
502 loff_t lstart, loff_t lend)
506 if ((!dax_mapping(mapping) && mapping->nrpages) ||
507 (dax_mapping(mapping) && mapping->nrexceptional)) {
508 err = __filemap_fdatawrite_range(mapping, lstart, lend,
510 /* See comment of filemap_write_and_wait() */
512 int err2 = filemap_fdatawait_range(mapping,
518 err = filemap_check_errors(mapping);
522 EXPORT_SYMBOL(filemap_write_and_wait_range);
525 * replace_page_cache_page - replace a pagecache page with a new one
526 * @old: page to be replaced
527 * @new: page to replace with
528 * @gfp_mask: allocation mode
530 * This function replaces a page in the pagecache with a new one. On
531 * success it acquires the pagecache reference for the new page and
532 * drops it for the old page. Both the old and new pages must be
533 * locked. This function does not add the new page to the LRU, the
534 * caller must do that.
536 * The remove + add is atomic. The only way this function can fail is
537 * memory allocation failure.
539 int replace_page_cache_page(struct page *old, struct page *new, gfp_t gfp_mask)
543 VM_BUG_ON_PAGE(!PageLocked(old), old);
544 VM_BUG_ON_PAGE(!PageLocked(new), new);
545 VM_BUG_ON_PAGE(new->mapping, new);
547 error = radix_tree_preload(gfp_mask & ~__GFP_HIGHMEM);
549 struct address_space *mapping = old->mapping;
550 void (*freepage)(struct page *);
553 pgoff_t offset = old->index;
554 freepage = mapping->a_ops->freepage;
557 new->mapping = mapping;
560 lock_page_memcg(old);
561 spin_lock_irqsave(&mapping->tree_lock, flags);
562 __delete_from_page_cache(old, NULL);
563 error = radix_tree_insert(&mapping->page_tree, offset, new);
568 * hugetlb pages do not participate in page cache accounting.
571 __inc_zone_page_state(new, NR_FILE_PAGES);
572 if (PageSwapBacked(new))
573 __inc_zone_page_state(new, NR_SHMEM);
574 spin_unlock_irqrestore(&mapping->tree_lock, flags);
575 unlock_page_memcg(old);
576 mem_cgroup_migrate(old, new);
577 radix_tree_preload_end();
580 page_cache_release(old);
585 EXPORT_SYMBOL_GPL(replace_page_cache_page);
587 static int page_cache_tree_insert(struct address_space *mapping,
588 struct page *page, void **shadowp)
590 struct radix_tree_node *node;
594 error = __radix_tree_create(&mapping->page_tree, page->index,
601 p = radix_tree_deref_slot_protected(slot, &mapping->tree_lock);
602 if (!radix_tree_exceptional_entry(p))
605 if (WARN_ON(dax_mapping(mapping)))
610 mapping->nrexceptional--;
612 workingset_node_shadows_dec(node);
614 radix_tree_replace_slot(slot, page);
617 workingset_node_pages_inc(node);
619 * Don't track node that contains actual pages.
621 * Avoid acquiring the list_lru lock if already
622 * untracked. The list_empty() test is safe as
623 * node->private_list is protected by
624 * mapping->tree_lock.
626 if (!list_empty(&node->private_list))
627 list_lru_del(&workingset_shadow_nodes,
628 &node->private_list);
633 static int __add_to_page_cache_locked(struct page *page,
634 struct address_space *mapping,
635 pgoff_t offset, gfp_t gfp_mask,
638 int huge = PageHuge(page);
639 struct mem_cgroup *memcg;
642 VM_BUG_ON_PAGE(!PageLocked(page), page);
643 VM_BUG_ON_PAGE(PageSwapBacked(page), page);
646 error = mem_cgroup_try_charge(page, current->mm,
647 gfp_mask, &memcg, false);
652 error = radix_tree_maybe_preload(gfp_mask & ~__GFP_HIGHMEM);
655 mem_cgroup_cancel_charge(page, memcg, false);
659 page_cache_get(page);
660 page->mapping = mapping;
661 page->index = offset;
663 spin_lock_irq(&mapping->tree_lock);
664 error = page_cache_tree_insert(mapping, page, shadowp);
665 radix_tree_preload_end();
669 /* hugetlb pages do not participate in page cache accounting. */
671 __inc_zone_page_state(page, NR_FILE_PAGES);
672 spin_unlock_irq(&mapping->tree_lock);
674 mem_cgroup_commit_charge(page, memcg, false, false);
675 trace_mm_filemap_add_to_page_cache(page);
678 page->mapping = NULL;
679 /* Leave page->index set: truncation relies upon it */
680 spin_unlock_irq(&mapping->tree_lock);
682 mem_cgroup_cancel_charge(page, memcg, false);
683 page_cache_release(page);
688 * add_to_page_cache_locked - add a locked page to the pagecache
690 * @mapping: the page's address_space
691 * @offset: page index
692 * @gfp_mask: page allocation mode
694 * This function is used to add a page to the pagecache. It must be locked.
695 * This function does not add the page to the LRU. The caller must do that.
697 int add_to_page_cache_locked(struct page *page, struct address_space *mapping,
698 pgoff_t offset, gfp_t gfp_mask)
700 return __add_to_page_cache_locked(page, mapping, offset,
703 EXPORT_SYMBOL(add_to_page_cache_locked);
705 int add_to_page_cache_lru(struct page *page, struct address_space *mapping,
706 pgoff_t offset, gfp_t gfp_mask)
711 __SetPageLocked(page);
712 ret = __add_to_page_cache_locked(page, mapping, offset,
715 __ClearPageLocked(page);
718 * The page might have been evicted from cache only
719 * recently, in which case it should be activated like
720 * any other repeatedly accessed page.
722 if (shadow && workingset_refault(shadow)) {
724 workingset_activation(page);
726 ClearPageActive(page);
731 EXPORT_SYMBOL_GPL(add_to_page_cache_lru);
734 struct page *__page_cache_alloc(gfp_t gfp)
739 if (cpuset_do_page_mem_spread()) {
740 unsigned int cpuset_mems_cookie;
742 cpuset_mems_cookie = read_mems_allowed_begin();
743 n = cpuset_mem_spread_node();
744 page = __alloc_pages_node(n, gfp, 0);
745 } while (!page && read_mems_allowed_retry(cpuset_mems_cookie));
749 return alloc_pages(gfp, 0);
751 EXPORT_SYMBOL(__page_cache_alloc);
755 * In order to wait for pages to become available there must be
756 * waitqueues associated with pages. By using a hash table of
757 * waitqueues where the bucket discipline is to maintain all
758 * waiters on the same queue and wake all when any of the pages
759 * become available, and for the woken contexts to check to be
760 * sure the appropriate page became available, this saves space
761 * at a cost of "thundering herd" phenomena during rare hash
764 wait_queue_head_t *page_waitqueue(struct page *page)
766 const struct zone *zone = page_zone(page);
768 return &zone->wait_table[hash_ptr(page, zone->wait_table_bits)];
770 EXPORT_SYMBOL(page_waitqueue);
772 void wait_on_page_bit(struct page *page, int bit_nr)
774 DEFINE_WAIT_BIT(wait, &page->flags, bit_nr);
776 if (test_bit(bit_nr, &page->flags))
777 __wait_on_bit(page_waitqueue(page), &wait, bit_wait_io,
778 TASK_UNINTERRUPTIBLE);
780 EXPORT_SYMBOL(wait_on_page_bit);
782 int wait_on_page_bit_killable(struct page *page, int bit_nr)
784 DEFINE_WAIT_BIT(wait, &page->flags, bit_nr);
786 if (!test_bit(bit_nr, &page->flags))
789 return __wait_on_bit(page_waitqueue(page), &wait,
790 bit_wait_io, TASK_KILLABLE);
793 int wait_on_page_bit_killable_timeout(struct page *page,
794 int bit_nr, unsigned long timeout)
796 DEFINE_WAIT_BIT(wait, &page->flags, bit_nr);
798 wait.key.timeout = jiffies + timeout;
799 if (!test_bit(bit_nr, &page->flags))
801 return __wait_on_bit(page_waitqueue(page), &wait,
802 bit_wait_io_timeout, TASK_KILLABLE);
804 EXPORT_SYMBOL_GPL(wait_on_page_bit_killable_timeout);
807 * add_page_wait_queue - Add an arbitrary waiter to a page's wait queue
808 * @page: Page defining the wait queue of interest
809 * @waiter: Waiter to add to the queue
811 * Add an arbitrary @waiter to the wait queue for the nominated @page.
813 void add_page_wait_queue(struct page *page, wait_queue_t *waiter)
815 wait_queue_head_t *q = page_waitqueue(page);
818 spin_lock_irqsave(&q->lock, flags);
819 __add_wait_queue(q, waiter);
820 spin_unlock_irqrestore(&q->lock, flags);
822 EXPORT_SYMBOL_GPL(add_page_wait_queue);
825 * unlock_page - unlock a locked page
828 * Unlocks the page and wakes up sleepers in ___wait_on_page_locked().
829 * Also wakes sleepers in wait_on_page_writeback() because the wakeup
830 * mechanism between PageLocked pages and PageWriteback pages is shared.
831 * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep.
833 * The mb is necessary to enforce ordering between the clear_bit and the read
834 * of the waitqueue (to avoid SMP races with a parallel wait_on_page_locked()).
836 void unlock_page(struct page *page)
838 page = compound_head(page);
839 VM_BUG_ON_PAGE(!PageLocked(page), page);
840 clear_bit_unlock(PG_locked, &page->flags);
841 smp_mb__after_atomic();
842 wake_up_page(page, PG_locked);
844 EXPORT_SYMBOL(unlock_page);
847 * end_page_writeback - end writeback against a page
850 void end_page_writeback(struct page *page)
853 * TestClearPageReclaim could be used here but it is an atomic
854 * operation and overkill in this particular case. Failing to
855 * shuffle a page marked for immediate reclaim is too mild to
856 * justify taking an atomic operation penalty at the end of
857 * ever page writeback.
859 if (PageReclaim(page)) {
860 ClearPageReclaim(page);
861 rotate_reclaimable_page(page);
864 if (!test_clear_page_writeback(page))
867 smp_mb__after_atomic();
868 wake_up_page(page, PG_writeback);
870 EXPORT_SYMBOL(end_page_writeback);
873 * After completing I/O on a page, call this routine to update the page
874 * flags appropriately
876 void page_endio(struct page *page, int rw, int err)
880 SetPageUptodate(page);
882 ClearPageUptodate(page);
886 } else { /* rw == WRITE */
890 mapping_set_error(page->mapping, err);
892 end_page_writeback(page);
895 EXPORT_SYMBOL_GPL(page_endio);
898 * __lock_page - get a lock on the page, assuming we need to sleep to get it
899 * @page: the page to lock
901 void __lock_page(struct page *page)
903 struct page *page_head = compound_head(page);
904 DEFINE_WAIT_BIT(wait, &page_head->flags, PG_locked);
906 __wait_on_bit_lock(page_waitqueue(page_head), &wait, bit_wait_io,
907 TASK_UNINTERRUPTIBLE);
909 EXPORT_SYMBOL(__lock_page);
911 int __lock_page_killable(struct page *page)
913 struct page *page_head = compound_head(page);
914 DEFINE_WAIT_BIT(wait, &page_head->flags, PG_locked);
916 return __wait_on_bit_lock(page_waitqueue(page_head), &wait,
917 bit_wait_io, TASK_KILLABLE);
919 EXPORT_SYMBOL_GPL(__lock_page_killable);
923 * 1 - page is locked; mmap_sem is still held.
924 * 0 - page is not locked.
925 * mmap_sem has been released (up_read()), unless flags had both
926 * FAULT_FLAG_ALLOW_RETRY and FAULT_FLAG_RETRY_NOWAIT set, in
927 * which case mmap_sem is still held.
929 * If neither ALLOW_RETRY nor KILLABLE are set, will always return 1
930 * with the page locked and the mmap_sem unperturbed.
932 int __lock_page_or_retry(struct page *page, struct mm_struct *mm,
935 if (flags & FAULT_FLAG_ALLOW_RETRY) {
937 * CAUTION! In this case, mmap_sem is not released
938 * even though return 0.
940 if (flags & FAULT_FLAG_RETRY_NOWAIT)
943 up_read(&mm->mmap_sem);
944 if (flags & FAULT_FLAG_KILLABLE)
945 wait_on_page_locked_killable(page);
947 wait_on_page_locked(page);
950 if (flags & FAULT_FLAG_KILLABLE) {
953 ret = __lock_page_killable(page);
955 up_read(&mm->mmap_sem);
965 * page_cache_next_hole - find the next hole (not-present entry)
968 * @max_scan: maximum range to search
970 * Search the set [index, min(index+max_scan-1, MAX_INDEX)] for the
971 * lowest indexed hole.
973 * Returns: the index of the hole if found, otherwise returns an index
974 * outside of the set specified (in which case 'return - index >=
975 * max_scan' will be true). In rare cases of index wrap-around, 0 will
978 * page_cache_next_hole may be called under rcu_read_lock. However,
979 * like radix_tree_gang_lookup, this will not atomically search a
980 * snapshot of the tree at a single point in time. For example, if a
981 * hole is created at index 5, then subsequently a hole is created at
982 * index 10, page_cache_next_hole covering both indexes may return 10
983 * if called under rcu_read_lock.
985 pgoff_t page_cache_next_hole(struct address_space *mapping,
986 pgoff_t index, unsigned long max_scan)
990 for (i = 0; i < max_scan; i++) {
993 page = radix_tree_lookup(&mapping->page_tree, index);
994 if (!page || radix_tree_exceptional_entry(page))
1003 EXPORT_SYMBOL(page_cache_next_hole);
1006 * page_cache_prev_hole - find the prev hole (not-present entry)
1009 * @max_scan: maximum range to search
1011 * Search backwards in the range [max(index-max_scan+1, 0), index] for
1014 * Returns: the index of the hole if found, otherwise returns an index
1015 * outside of the set specified (in which case 'index - return >=
1016 * max_scan' will be true). In rare cases of wrap-around, ULONG_MAX
1019 * page_cache_prev_hole may be called under rcu_read_lock. However,
1020 * like radix_tree_gang_lookup, this will not atomically search a
1021 * snapshot of the tree at a single point in time. For example, if a
1022 * hole is created at index 10, then subsequently a hole is created at
1023 * index 5, page_cache_prev_hole covering both indexes may return 5 if
1024 * called under rcu_read_lock.
1026 pgoff_t page_cache_prev_hole(struct address_space *mapping,
1027 pgoff_t index, unsigned long max_scan)
1031 for (i = 0; i < max_scan; i++) {
1034 page = radix_tree_lookup(&mapping->page_tree, index);
1035 if (!page || radix_tree_exceptional_entry(page))
1038 if (index == ULONG_MAX)
1044 EXPORT_SYMBOL(page_cache_prev_hole);
1047 * find_get_entry - find and get a page cache entry
1048 * @mapping: the address_space to search
1049 * @offset: the page cache index
1051 * Looks up the page cache slot at @mapping & @offset. If there is a
1052 * page cache page, it is returned with an increased refcount.
1054 * If the slot holds a shadow entry of a previously evicted page, or a
1055 * swap entry from shmem/tmpfs, it is returned.
1057 * Otherwise, %NULL is returned.
1059 struct page *find_get_entry(struct address_space *mapping, pgoff_t offset)
1067 pagep = radix_tree_lookup_slot(&mapping->page_tree, offset);
1069 page = radix_tree_deref_slot(pagep);
1070 if (unlikely(!page))
1072 if (radix_tree_exception(page)) {
1073 if (radix_tree_deref_retry(page))
1076 * A shadow entry of a recently evicted page,
1077 * or a swap entry from shmem/tmpfs. Return
1078 * it without attempting to raise page count.
1082 if (!page_cache_get_speculative(page))
1086 * Has the page moved?
1087 * This is part of the lockless pagecache protocol. See
1088 * include/linux/pagemap.h for details.
1090 if (unlikely(page != *pagep)) {
1091 page_cache_release(page);
1100 EXPORT_SYMBOL(find_get_entry);
1103 * find_lock_entry - locate, pin and lock a page cache entry
1104 * @mapping: the address_space to search
1105 * @offset: the page cache index
1107 * Looks up the page cache slot at @mapping & @offset. If there is a
1108 * page cache page, it is returned locked and with an increased
1111 * If the slot holds a shadow entry of a previously evicted page, or a
1112 * swap entry from shmem/tmpfs, it is returned.
1114 * Otherwise, %NULL is returned.
1116 * find_lock_entry() may sleep.
1118 struct page *find_lock_entry(struct address_space *mapping, pgoff_t offset)
1123 page = find_get_entry(mapping, offset);
1124 if (page && !radix_tree_exception(page)) {
1126 /* Has the page been truncated? */
1127 if (unlikely(page->mapping != mapping)) {
1129 page_cache_release(page);
1132 VM_BUG_ON_PAGE(page->index != offset, page);
1136 EXPORT_SYMBOL(find_lock_entry);
1139 * pagecache_get_page - find and get a page reference
1140 * @mapping: the address_space to search
1141 * @offset: the page index
1142 * @fgp_flags: PCG flags
1143 * @gfp_mask: gfp mask to use for the page cache data page allocation
1145 * Looks up the page cache slot at @mapping & @offset.
1147 * PCG flags modify how the page is returned.
1149 * FGP_ACCESSED: the page will be marked accessed
1150 * FGP_LOCK: Page is return locked
1151 * FGP_CREAT: If page is not present then a new page is allocated using
1152 * @gfp_mask and added to the page cache and the VM's LRU
1153 * list. The page is returned locked and with an increased
1154 * refcount. Otherwise, %NULL is returned.
1156 * If FGP_LOCK or FGP_CREAT are specified then the function may sleep even
1157 * if the GFP flags specified for FGP_CREAT are atomic.
1159 * If there is a page cache page, it is returned with an increased refcount.
1161 struct page *pagecache_get_page(struct address_space *mapping, pgoff_t offset,
1162 int fgp_flags, gfp_t gfp_mask)
1167 page = find_get_entry(mapping, offset);
1168 if (radix_tree_exceptional_entry(page))
1173 if (fgp_flags & FGP_LOCK) {
1174 if (fgp_flags & FGP_NOWAIT) {
1175 if (!trylock_page(page)) {
1176 page_cache_release(page);
1183 /* Has the page been truncated? */
1184 if (unlikely(page->mapping != mapping)) {
1186 page_cache_release(page);
1189 VM_BUG_ON_PAGE(page->index != offset, page);
1192 if (page && (fgp_flags & FGP_ACCESSED))
1193 mark_page_accessed(page);
1196 if (!page && (fgp_flags & FGP_CREAT)) {
1198 if ((fgp_flags & FGP_WRITE) && mapping_cap_account_dirty(mapping))
1199 gfp_mask |= __GFP_WRITE;
1200 if (fgp_flags & FGP_NOFS)
1201 gfp_mask &= ~__GFP_FS;
1203 page = __page_cache_alloc(gfp_mask);
1207 if (WARN_ON_ONCE(!(fgp_flags & FGP_LOCK)))
1208 fgp_flags |= FGP_LOCK;
1210 /* Init accessed so avoid atomic mark_page_accessed later */
1211 if (fgp_flags & FGP_ACCESSED)
1212 __SetPageReferenced(page);
1214 err = add_to_page_cache_lru(page, mapping, offset,
1215 gfp_mask & GFP_RECLAIM_MASK);
1216 if (unlikely(err)) {
1217 page_cache_release(page);
1226 EXPORT_SYMBOL(pagecache_get_page);
1229 * find_get_entries - gang pagecache lookup
1230 * @mapping: The address_space to search
1231 * @start: The starting page cache index
1232 * @nr_entries: The maximum number of entries
1233 * @entries: Where the resulting entries are placed
1234 * @indices: The cache indices corresponding to the entries in @entries
1236 * find_get_entries() will search for and return a group of up to
1237 * @nr_entries entries in the mapping. The entries are placed at
1238 * @entries. find_get_entries() takes a reference against any actual
1241 * The search returns a group of mapping-contiguous page cache entries
1242 * with ascending indexes. There may be holes in the indices due to
1243 * not-present pages.
1245 * Any shadow entries of evicted pages, or swap entries from
1246 * shmem/tmpfs, are included in the returned array.
1248 * find_get_entries() returns the number of pages and shadow entries
1251 unsigned find_get_entries(struct address_space *mapping,
1252 pgoff_t start, unsigned int nr_entries,
1253 struct page **entries, pgoff_t *indices)
1256 unsigned int ret = 0;
1257 struct radix_tree_iter iter;
1264 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) {
1267 page = radix_tree_deref_slot(slot);
1268 if (unlikely(!page))
1270 if (radix_tree_exception(page)) {
1271 if (radix_tree_deref_retry(page))
1274 * A shadow entry of a recently evicted page, a swap
1275 * entry from shmem/tmpfs or a DAX entry. Return it
1276 * without attempting to raise page count.
1280 if (!page_cache_get_speculative(page))
1283 /* Has the page moved? */
1284 if (unlikely(page != *slot)) {
1285 page_cache_release(page);
1289 indices[ret] = iter.index;
1290 entries[ret] = page;
1291 if (++ret == nr_entries)
1299 * find_get_pages - gang pagecache lookup
1300 * @mapping: The address_space to search
1301 * @start: The starting page index
1302 * @nr_pages: The maximum number of pages
1303 * @pages: Where the resulting pages are placed
1305 * find_get_pages() will search for and return a group of up to
1306 * @nr_pages pages in the mapping. The pages are placed at @pages.
1307 * find_get_pages() takes a reference against the returned pages.
1309 * The search returns a group of mapping-contiguous pages with ascending
1310 * indexes. There may be holes in the indices due to not-present pages.
1312 * find_get_pages() returns the number of pages which were found.
1314 unsigned find_get_pages(struct address_space *mapping, pgoff_t start,
1315 unsigned int nr_pages, struct page **pages)
1317 struct radix_tree_iter iter;
1321 if (unlikely(!nr_pages))
1326 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) {
1329 page = radix_tree_deref_slot(slot);
1330 if (unlikely(!page))
1333 if (radix_tree_exception(page)) {
1334 if (radix_tree_deref_retry(page)) {
1336 * Transient condition which can only trigger
1337 * when entry at index 0 moves out of or back
1338 * to root: none yet gotten, safe to restart.
1340 WARN_ON(iter.index);
1344 * A shadow entry of a recently evicted page,
1345 * or a swap entry from shmem/tmpfs. Skip
1351 if (!page_cache_get_speculative(page))
1354 /* Has the page moved? */
1355 if (unlikely(page != *slot)) {
1356 page_cache_release(page);
1361 if (++ret == nr_pages)
1370 * find_get_pages_contig - gang contiguous pagecache lookup
1371 * @mapping: The address_space to search
1372 * @index: The starting page index
1373 * @nr_pages: The maximum number of pages
1374 * @pages: Where the resulting pages are placed
1376 * find_get_pages_contig() works exactly like find_get_pages(), except
1377 * that the returned number of pages are guaranteed to be contiguous.
1379 * find_get_pages_contig() returns the number of pages which were found.
1381 unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index,
1382 unsigned int nr_pages, struct page **pages)
1384 struct radix_tree_iter iter;
1386 unsigned int ret = 0;
1388 if (unlikely(!nr_pages))
1393 radix_tree_for_each_contig(slot, &mapping->page_tree, &iter, index) {
1396 page = radix_tree_deref_slot(slot);
1397 /* The hole, there no reason to continue */
1398 if (unlikely(!page))
1401 if (radix_tree_exception(page)) {
1402 if (radix_tree_deref_retry(page)) {
1404 * Transient condition which can only trigger
1405 * when entry at index 0 moves out of or back
1406 * to root: none yet gotten, safe to restart.
1411 * A shadow entry of a recently evicted page,
1412 * or a swap entry from shmem/tmpfs. Stop
1413 * looking for contiguous pages.
1418 if (!page_cache_get_speculative(page))
1421 /* Has the page moved? */
1422 if (unlikely(page != *slot)) {
1423 page_cache_release(page);
1428 * must check mapping and index after taking the ref.
1429 * otherwise we can get both false positives and false
1430 * negatives, which is just confusing to the caller.
1432 if (page->mapping == NULL || page->index != iter.index) {
1433 page_cache_release(page);
1438 if (++ret == nr_pages)
1444 EXPORT_SYMBOL(find_get_pages_contig);
1447 * find_get_pages_tag - find and return pages that match @tag
1448 * @mapping: the address_space to search
1449 * @index: the starting page index
1450 * @tag: the tag index
1451 * @nr_pages: the maximum number of pages
1452 * @pages: where the resulting pages are placed
1454 * Like find_get_pages, except we only return pages which are tagged with
1455 * @tag. We update @index to index the next page for the traversal.
1457 unsigned find_get_pages_tag(struct address_space *mapping, pgoff_t *index,
1458 int tag, unsigned int nr_pages, struct page **pages)
1460 struct radix_tree_iter iter;
1464 if (unlikely(!nr_pages))
1469 radix_tree_for_each_tagged(slot, &mapping->page_tree,
1470 &iter, *index, tag) {
1473 page = radix_tree_deref_slot(slot);
1474 if (unlikely(!page))
1477 if (radix_tree_exception(page)) {
1478 if (radix_tree_deref_retry(page)) {
1480 * Transient condition which can only trigger
1481 * when entry at index 0 moves out of or back
1482 * to root: none yet gotten, safe to restart.
1487 * A shadow entry of a recently evicted page.
1489 * Those entries should never be tagged, but
1490 * this tree walk is lockless and the tags are
1491 * looked up in bulk, one radix tree node at a
1492 * time, so there is a sizable window for page
1493 * reclaim to evict a page we saw tagged.
1500 if (!page_cache_get_speculative(page))
1503 /* Has the page moved? */
1504 if (unlikely(page != *slot)) {
1505 page_cache_release(page);
1510 if (++ret == nr_pages)
1517 *index = pages[ret - 1]->index + 1;
1521 EXPORT_SYMBOL(find_get_pages_tag);
1524 * find_get_entries_tag - find and return entries that match @tag
1525 * @mapping: the address_space to search
1526 * @start: the starting page cache index
1527 * @tag: the tag index
1528 * @nr_entries: the maximum number of entries
1529 * @entries: where the resulting entries are placed
1530 * @indices: the cache indices corresponding to the entries in @entries
1532 * Like find_get_entries, except we only return entries which are tagged with
1535 unsigned find_get_entries_tag(struct address_space *mapping, pgoff_t start,
1536 int tag, unsigned int nr_entries,
1537 struct page **entries, pgoff_t *indices)
1540 unsigned int ret = 0;
1541 struct radix_tree_iter iter;
1548 radix_tree_for_each_tagged(slot, &mapping->page_tree,
1549 &iter, start, tag) {
1552 page = radix_tree_deref_slot(slot);
1553 if (unlikely(!page))
1555 if (radix_tree_exception(page)) {
1556 if (radix_tree_deref_retry(page)) {
1558 * Transient condition which can only trigger
1559 * when entry at index 0 moves out of or back
1560 * to root: none yet gotten, safe to restart.
1566 * A shadow entry of a recently evicted page, a swap
1567 * entry from shmem/tmpfs or a DAX entry. Return it
1568 * without attempting to raise page count.
1572 if (!page_cache_get_speculative(page))
1575 /* Has the page moved? */
1576 if (unlikely(page != *slot)) {
1577 page_cache_release(page);
1581 indices[ret] = iter.index;
1582 entries[ret] = page;
1583 if (++ret == nr_entries)
1589 EXPORT_SYMBOL(find_get_entries_tag);
1592 * CD/DVDs are error prone. When a medium error occurs, the driver may fail
1593 * a _large_ part of the i/o request. Imagine the worst scenario:
1595 * ---R__________________________________________B__________
1596 * ^ reading here ^ bad block(assume 4k)
1598 * read(R) => miss => readahead(R...B) => media error => frustrating retries
1599 * => failing the whole request => read(R) => read(R+1) =>
1600 * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
1601 * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
1602 * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
1604 * It is going insane. Fix it by quickly scaling down the readahead size.
1606 static void shrink_readahead_size_eio(struct file *filp,
1607 struct file_ra_state *ra)
1613 * do_generic_file_read - generic file read routine
1614 * @filp: the file to read
1615 * @ppos: current file position
1616 * @iter: data destination
1617 * @written: already copied
1619 * This is a generic file read routine, and uses the
1620 * mapping->a_ops->readpage() function for the actual low-level stuff.
1622 * This is really ugly. But the goto's actually try to clarify some
1623 * of the logic when it comes to error handling etc.
1625 static ssize_t do_generic_file_read(struct file *filp, loff_t *ppos,
1626 struct iov_iter *iter, ssize_t written)
1628 struct address_space *mapping = filp->f_mapping;
1629 struct inode *inode = mapping->host;
1630 struct file_ra_state *ra = &filp->f_ra;
1634 unsigned long offset; /* offset into pagecache page */
1635 unsigned int prev_offset;
1638 index = *ppos >> PAGE_CACHE_SHIFT;
1639 prev_index = ra->prev_pos >> PAGE_CACHE_SHIFT;
1640 prev_offset = ra->prev_pos & (PAGE_CACHE_SIZE-1);
1641 last_index = (*ppos + iter->count + PAGE_CACHE_SIZE-1) >> PAGE_CACHE_SHIFT;
1642 offset = *ppos & ~PAGE_CACHE_MASK;
1648 unsigned long nr, ret;
1652 page = find_get_page(mapping, index);
1654 page_cache_sync_readahead(mapping,
1656 index, last_index - index);
1657 page = find_get_page(mapping, index);
1658 if (unlikely(page == NULL))
1659 goto no_cached_page;
1661 if (PageReadahead(page)) {
1662 page_cache_async_readahead(mapping,
1664 index, last_index - index);
1666 if (!PageUptodate(page)) {
1668 * See comment in do_read_cache_page on why
1669 * wait_on_page_locked is used to avoid unnecessarily
1670 * serialisations and why it's safe.
1672 wait_on_page_locked_killable(page);
1673 if (PageUptodate(page))
1676 if (inode->i_blkbits == PAGE_CACHE_SHIFT ||
1677 !mapping->a_ops->is_partially_uptodate)
1678 goto page_not_up_to_date;
1679 if (!trylock_page(page))
1680 goto page_not_up_to_date;
1681 /* Did it get truncated before we got the lock? */
1683 goto page_not_up_to_date_locked;
1684 if (!mapping->a_ops->is_partially_uptodate(page,
1685 offset, iter->count))
1686 goto page_not_up_to_date_locked;
1691 * i_size must be checked after we know the page is Uptodate.
1693 * Checking i_size after the check allows us to calculate
1694 * the correct value for "nr", which means the zero-filled
1695 * part of the page is not copied back to userspace (unless
1696 * another truncate extends the file - this is desired though).
1699 isize = i_size_read(inode);
1700 end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
1701 if (unlikely(!isize || index > end_index)) {
1702 page_cache_release(page);
1706 /* nr is the maximum number of bytes to copy from this page */
1707 nr = PAGE_CACHE_SIZE;
1708 if (index == end_index) {
1709 nr = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;
1711 page_cache_release(page);
1717 /* If users can be writing to this page using arbitrary
1718 * virtual addresses, take care about potential aliasing
1719 * before reading the page on the kernel side.
1721 if (mapping_writably_mapped(mapping))
1722 flush_dcache_page(page);
1725 * When a sequential read accesses a page several times,
1726 * only mark it as accessed the first time.
1728 if (prev_index != index || offset != prev_offset)
1729 mark_page_accessed(page);
1733 * Ok, we have the page, and it's up-to-date, so
1734 * now we can copy it to user space...
1737 ret = copy_page_to_iter(page, offset, nr, iter);
1739 index += offset >> PAGE_CACHE_SHIFT;
1740 offset &= ~PAGE_CACHE_MASK;
1741 prev_offset = offset;
1743 page_cache_release(page);
1745 if (!iov_iter_count(iter))
1753 page_not_up_to_date:
1754 /* Get exclusive access to the page ... */
1755 error = lock_page_killable(page);
1756 if (unlikely(error))
1757 goto readpage_error;
1759 page_not_up_to_date_locked:
1760 /* Did it get truncated before we got the lock? */
1761 if (!page->mapping) {
1763 page_cache_release(page);
1767 /* Did somebody else fill it already? */
1768 if (PageUptodate(page)) {
1775 * A previous I/O error may have been due to temporary
1776 * failures, eg. multipath errors.
1777 * PG_error will be set again if readpage fails.
1779 ClearPageError(page);
1780 /* Start the actual read. The read will unlock the page. */
1781 error = mapping->a_ops->readpage(filp, page);
1783 if (unlikely(error)) {
1784 if (error == AOP_TRUNCATED_PAGE) {
1785 page_cache_release(page);
1789 goto readpage_error;
1792 if (!PageUptodate(page)) {
1793 error = lock_page_killable(page);
1794 if (unlikely(error))
1795 goto readpage_error;
1796 if (!PageUptodate(page)) {
1797 if (page->mapping == NULL) {
1799 * invalidate_mapping_pages got it
1802 page_cache_release(page);
1806 shrink_readahead_size_eio(filp, ra);
1808 goto readpage_error;
1816 /* UHHUH! A synchronous read error occurred. Report it */
1817 page_cache_release(page);
1822 * Ok, it wasn't cached, so we need to create a new
1825 page = page_cache_alloc_cold(mapping);
1830 error = add_to_page_cache_lru(page, mapping, index,
1831 mapping_gfp_constraint(mapping, GFP_KERNEL));
1833 page_cache_release(page);
1834 if (error == -EEXIST) {
1844 ra->prev_pos = prev_index;
1845 ra->prev_pos <<= PAGE_CACHE_SHIFT;
1846 ra->prev_pos |= prev_offset;
1848 *ppos = ((loff_t)index << PAGE_CACHE_SHIFT) + offset;
1849 file_accessed(filp);
1850 return written ? written : error;
1854 * generic_file_read_iter - generic filesystem read routine
1855 * @iocb: kernel I/O control block
1856 * @iter: destination for the data read
1858 * This is the "read_iter()" routine for all filesystems
1859 * that can use the page cache directly.
1862 generic_file_read_iter(struct kiocb *iocb, struct iov_iter *iter)
1864 struct file *file = iocb->ki_filp;
1866 loff_t *ppos = &iocb->ki_pos;
1869 if (iocb->ki_flags & IOCB_DIRECT) {
1870 struct address_space *mapping = file->f_mapping;
1871 struct inode *inode = mapping->host;
1872 size_t count = iov_iter_count(iter);
1876 goto out; /* skip atime */
1877 size = i_size_read(inode);
1878 retval = filemap_write_and_wait_range(mapping, pos,
1881 struct iov_iter data = *iter;
1882 retval = mapping->a_ops->direct_IO(iocb, &data, pos);
1886 *ppos = pos + retval;
1887 iov_iter_advance(iter, retval);
1891 * Btrfs can have a short DIO read if we encounter
1892 * compressed extents, so if there was an error, or if
1893 * we've already read everything we wanted to, or if
1894 * there was a short read because we hit EOF, go ahead
1895 * and return. Otherwise fallthrough to buffered io for
1896 * the rest of the read. Buffered reads will not work for
1897 * DAX files, so don't bother trying.
1899 if (retval < 0 || !iov_iter_count(iter) || *ppos >= size ||
1901 file_accessed(file);
1906 retval = do_generic_file_read(file, ppos, iter, retval);
1910 EXPORT_SYMBOL(generic_file_read_iter);
1914 * page_cache_read - adds requested page to the page cache if not already there
1915 * @file: file to read
1916 * @offset: page index
1917 * @gfp_mask: memory allocation flags
1919 * This adds the requested page to the page cache if it isn't already there,
1920 * and schedules an I/O to read in its contents from disk.
1922 static int page_cache_read(struct file *file, pgoff_t offset, gfp_t gfp_mask)
1924 struct address_space *mapping = file->f_mapping;
1929 page = __page_cache_alloc(gfp_mask|__GFP_COLD);
1933 ret = add_to_page_cache_lru(page, mapping, offset, gfp_mask & GFP_KERNEL);
1935 ret = mapping->a_ops->readpage(file, page);
1936 else if (ret == -EEXIST)
1937 ret = 0; /* losing race to add is OK */
1939 page_cache_release(page);
1941 } while (ret == AOP_TRUNCATED_PAGE);
1946 #define MMAP_LOTSAMISS (100)
1949 * Synchronous readahead happens when we don't even find
1950 * a page in the page cache at all.
1952 static void do_sync_mmap_readahead(struct vm_area_struct *vma,
1953 struct file_ra_state *ra,
1957 struct address_space *mapping = file->f_mapping;
1959 /* If we don't want any read-ahead, don't bother */
1960 if (vma->vm_flags & VM_RAND_READ)
1965 if (vma->vm_flags & VM_SEQ_READ) {
1966 page_cache_sync_readahead(mapping, ra, file, offset,
1971 /* Avoid banging the cache line if not needed */
1972 if (ra->mmap_miss < MMAP_LOTSAMISS * 10)
1976 * Do we miss much more than hit in this file? If so,
1977 * stop bothering with read-ahead. It will only hurt.
1979 if (ra->mmap_miss > MMAP_LOTSAMISS)
1985 ra->start = max_t(long, 0, offset - ra->ra_pages / 2);
1986 ra->size = ra->ra_pages;
1987 ra->async_size = ra->ra_pages / 4;
1988 ra_submit(ra, mapping, file);
1992 * Asynchronous readahead happens when we find the page and PG_readahead,
1993 * so we want to possibly extend the readahead further..
1995 static void do_async_mmap_readahead(struct vm_area_struct *vma,
1996 struct file_ra_state *ra,
2001 struct address_space *mapping = file->f_mapping;
2003 /* If we don't want any read-ahead, don't bother */
2004 if (vma->vm_flags & VM_RAND_READ)
2006 if (ra->mmap_miss > 0)
2008 if (PageReadahead(page))
2009 page_cache_async_readahead(mapping, ra, file,
2010 page, offset, ra->ra_pages);
2014 * filemap_fault - read in file data for page fault handling
2015 * @vma: vma in which the fault was taken
2016 * @vmf: struct vm_fault containing details of the fault
2018 * filemap_fault() is invoked via the vma operations vector for a
2019 * mapped memory region to read in file data during a page fault.
2021 * The goto's are kind of ugly, but this streamlines the normal case of having
2022 * it in the page cache, and handles the special cases reasonably without
2023 * having a lot of duplicated code.
2025 * vma->vm_mm->mmap_sem must be held on entry.
2027 * If our return value has VM_FAULT_RETRY set, it's because
2028 * lock_page_or_retry() returned 0.
2029 * The mmap_sem has usually been released in this case.
2030 * See __lock_page_or_retry() for the exception.
2032 * If our return value does not have VM_FAULT_RETRY set, the mmap_sem
2033 * has not been released.
2035 * We never return with VM_FAULT_RETRY and a bit from VM_FAULT_ERROR set.
2037 int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
2040 struct file *file = vma->vm_file;
2041 struct address_space *mapping = file->f_mapping;
2042 struct file_ra_state *ra = &file->f_ra;
2043 struct inode *inode = mapping->host;
2044 pgoff_t offset = vmf->pgoff;
2049 size = round_up(i_size_read(inode), PAGE_CACHE_SIZE);
2050 if (offset >= size >> PAGE_CACHE_SHIFT)
2051 return VM_FAULT_SIGBUS;
2054 * Do we have something in the page cache already?
2056 page = find_get_page(mapping, offset);
2057 if (likely(page) && !(vmf->flags & FAULT_FLAG_TRIED)) {
2059 * We found the page, so try async readahead before
2060 * waiting for the lock.
2062 do_async_mmap_readahead(vma, ra, file, page, offset);
2064 /* No page in the page cache at all */
2065 do_sync_mmap_readahead(vma, ra, file, offset);
2066 count_vm_event(PGMAJFAULT);
2067 mem_cgroup_count_vm_event(vma->vm_mm, PGMAJFAULT);
2068 ret = VM_FAULT_MAJOR;
2070 page = find_get_page(mapping, offset);
2072 goto no_cached_page;
2075 if (!lock_page_or_retry(page, vma->vm_mm, vmf->flags)) {
2076 page_cache_release(page);
2077 return ret | VM_FAULT_RETRY;
2080 /* Did it get truncated? */
2081 if (unlikely(page->mapping != mapping)) {
2086 VM_BUG_ON_PAGE(page->index != offset, page);
2089 * We have a locked page in the page cache, now we need to check
2090 * that it's up-to-date. If not, it is going to be due to an error.
2092 if (unlikely(!PageUptodate(page)))
2093 goto page_not_uptodate;
2096 * Found the page and have a reference on it.
2097 * We must recheck i_size under page lock.
2099 size = round_up(i_size_read(inode), PAGE_CACHE_SIZE);
2100 if (unlikely(offset >= size >> PAGE_CACHE_SHIFT)) {
2102 page_cache_release(page);
2103 return VM_FAULT_SIGBUS;
2107 return ret | VM_FAULT_LOCKED;
2111 * We're only likely to ever get here if MADV_RANDOM is in
2114 error = page_cache_read(file, offset, vmf->gfp_mask);
2117 * The page we want has now been added to the page cache.
2118 * In the unlikely event that someone removed it in the
2119 * meantime, we'll just come back here and read it again.
2125 * An error return from page_cache_read can result if the
2126 * system is low on memory, or a problem occurs while trying
2129 if (error == -ENOMEM)
2130 return VM_FAULT_OOM;
2131 return VM_FAULT_SIGBUS;
2135 * Umm, take care of errors if the page isn't up-to-date.
2136 * Try to re-read it _once_. We do this synchronously,
2137 * because there really aren't any performance issues here
2138 * and we need to check for errors.
2140 ClearPageError(page);
2141 error = mapping->a_ops->readpage(file, page);
2143 wait_on_page_locked(page);
2144 if (!PageUptodate(page))
2147 page_cache_release(page);
2149 if (!error || error == AOP_TRUNCATED_PAGE)
2152 /* Things didn't work out. Return zero to tell the mm layer so. */
2153 shrink_readahead_size_eio(file, ra);
2154 return VM_FAULT_SIGBUS;
2156 EXPORT_SYMBOL(filemap_fault);
2158 void filemap_map_pages(struct vm_area_struct *vma, struct vm_fault *vmf)
2160 struct radix_tree_iter iter;
2162 struct file *file = vma->vm_file;
2163 struct address_space *mapping = file->f_mapping;
2166 unsigned long address = (unsigned long) vmf->virtual_address;
2171 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, vmf->pgoff) {
2172 if (iter.index > vmf->max_pgoff)
2175 page = radix_tree_deref_slot(slot);
2176 if (unlikely(!page))
2178 if (radix_tree_exception(page)) {
2179 if (radix_tree_deref_retry(page))
2185 if (!page_cache_get_speculative(page))
2188 /* Has the page moved? */
2189 if (unlikely(page != *slot)) {
2190 page_cache_release(page);
2194 if (!PageUptodate(page) ||
2195 PageReadahead(page) ||
2198 if (!trylock_page(page))
2201 if (page->mapping != mapping || !PageUptodate(page))
2204 size = round_up(i_size_read(mapping->host), PAGE_CACHE_SIZE);
2205 if (page->index >= size >> PAGE_CACHE_SHIFT)
2208 pte = vmf->pte + page->index - vmf->pgoff;
2209 if (!pte_none(*pte))
2212 if (file->f_ra.mmap_miss > 0)
2213 file->f_ra.mmap_miss--;
2214 addr = address + (page->index - vmf->pgoff) * PAGE_SIZE;
2215 do_set_pte(vma, addr, page, pte, false, false);
2221 page_cache_release(page);
2223 if (iter.index == vmf->max_pgoff)
2228 EXPORT_SYMBOL(filemap_map_pages);
2230 int filemap_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
2232 struct page *page = vmf->page;
2233 struct inode *inode = file_inode(vma->vm_file);
2234 int ret = VM_FAULT_LOCKED;
2236 sb_start_pagefault(inode->i_sb);
2237 file_update_time(vma->vm_file);
2239 if (page->mapping != inode->i_mapping) {
2241 ret = VM_FAULT_NOPAGE;
2245 * We mark the page dirty already here so that when freeze is in
2246 * progress, we are guaranteed that writeback during freezing will
2247 * see the dirty page and writeprotect it again.
2249 set_page_dirty(page);
2250 wait_for_stable_page(page);
2252 sb_end_pagefault(inode->i_sb);
2255 EXPORT_SYMBOL(filemap_page_mkwrite);
2257 const struct vm_operations_struct generic_file_vm_ops = {
2258 .fault = filemap_fault,
2259 .map_pages = filemap_map_pages,
2260 .page_mkwrite = filemap_page_mkwrite,
2263 /* This is used for a general mmap of a disk file */
2265 int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
2267 struct address_space *mapping = file->f_mapping;
2269 if (!mapping->a_ops->readpage)
2271 file_accessed(file);
2272 vma->vm_ops = &generic_file_vm_ops;
2277 * This is for filesystems which do not implement ->writepage.
2279 int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
2281 if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE))
2283 return generic_file_mmap(file, vma);
2286 int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
2290 int generic_file_readonly_mmap(struct file * file, struct vm_area_struct * vma)
2294 #endif /* CONFIG_MMU */
2296 EXPORT_SYMBOL(generic_file_mmap);
2297 EXPORT_SYMBOL(generic_file_readonly_mmap);
2299 static struct page *wait_on_page_read(struct page *page)
2301 if (!IS_ERR(page)) {
2302 wait_on_page_locked(page);
2303 if (!PageUptodate(page)) {
2304 page_cache_release(page);
2305 page = ERR_PTR(-EIO);
2311 static struct page *do_read_cache_page(struct address_space *mapping,
2313 int (*filler)(void *, struct page *),
2320 page = find_get_page(mapping, index);
2322 page = __page_cache_alloc(gfp | __GFP_COLD);
2324 return ERR_PTR(-ENOMEM);
2325 err = add_to_page_cache_lru(page, mapping, index, gfp);
2326 if (unlikely(err)) {
2327 page_cache_release(page);
2330 /* Presumably ENOMEM for radix tree node */
2331 return ERR_PTR(err);
2335 err = filler(data, page);
2337 page_cache_release(page);
2338 return ERR_PTR(err);
2341 page = wait_on_page_read(page);
2346 if (PageUptodate(page))
2350 * Page is not up to date and may be locked due one of the following
2351 * case a: Page is being filled and the page lock is held
2352 * case b: Read/write error clearing the page uptodate status
2353 * case c: Truncation in progress (page locked)
2354 * case d: Reclaim in progress
2356 * Case a, the page will be up to date when the page is unlocked.
2357 * There is no need to serialise on the page lock here as the page
2358 * is pinned so the lock gives no additional protection. Even if the
2359 * the page is truncated, the data is still valid if PageUptodate as
2360 * it's a race vs truncate race.
2361 * Case b, the page will not be up to date
2362 * Case c, the page may be truncated but in itself, the data may still
2363 * be valid after IO completes as it's a read vs truncate race. The
2364 * operation must restart if the page is not uptodate on unlock but
2365 * otherwise serialising on page lock to stabilise the mapping gives
2366 * no additional guarantees to the caller as the page lock is
2367 * released before return.
2368 * Case d, similar to truncation. If reclaim holds the page lock, it
2369 * will be a race with remove_mapping that determines if the mapping
2370 * is valid on unlock but otherwise the data is valid and there is
2371 * no need to serialise with page lock.
2373 * As the page lock gives no additional guarantee, we optimistically
2374 * wait on the page to be unlocked and check if it's up to date and
2375 * use the page if it is. Otherwise, the page lock is required to
2376 * distinguish between the different cases. The motivation is that we
2377 * avoid spurious serialisations and wakeups when multiple processes
2378 * wait on the same page for IO to complete.
2380 wait_on_page_locked(page);
2381 if (PageUptodate(page))
2384 /* Distinguish between all the cases under the safety of the lock */
2387 /* Case c or d, restart the operation */
2388 if (!page->mapping) {
2390 page_cache_release(page);
2394 /* Someone else locked and filled the page in a very small window */
2395 if (PageUptodate(page)) {
2402 mark_page_accessed(page);
2407 * read_cache_page - read into page cache, fill it if needed
2408 * @mapping: the page's address_space
2409 * @index: the page index
2410 * @filler: function to perform the read
2411 * @data: first arg to filler(data, page) function, often left as NULL
2413 * Read into the page cache. If a page already exists, and PageUptodate() is
2414 * not set, try to fill the page and wait for it to become unlocked.
2416 * If the page does not get brought uptodate, return -EIO.
2418 struct page *read_cache_page(struct address_space *mapping,
2420 int (*filler)(void *, struct page *),
2423 return do_read_cache_page(mapping, index, filler, data, mapping_gfp_mask(mapping));
2425 EXPORT_SYMBOL(read_cache_page);
2428 * read_cache_page_gfp - read into page cache, using specified page allocation flags.
2429 * @mapping: the page's address_space
2430 * @index: the page index
2431 * @gfp: the page allocator flags to use if allocating
2433 * This is the same as "read_mapping_page(mapping, index, NULL)", but with
2434 * any new page allocations done using the specified allocation flags.
2436 * If the page does not get brought uptodate, return -EIO.
2438 struct page *read_cache_page_gfp(struct address_space *mapping,
2442 filler_t *filler = (filler_t *)mapping->a_ops->readpage;
2444 return do_read_cache_page(mapping, index, filler, NULL, gfp);
2446 EXPORT_SYMBOL(read_cache_page_gfp);
2449 * Performs necessary checks before doing a write
2451 * Can adjust writing position or amount of bytes to write.
2452 * Returns appropriate error code that caller should return or
2453 * zero in case that write should be allowed.
2455 inline ssize_t generic_write_checks(struct kiocb *iocb, struct iov_iter *from)
2457 struct file *file = iocb->ki_filp;
2458 struct inode *inode = file->f_mapping->host;
2459 unsigned long limit = rlimit(RLIMIT_FSIZE);
2462 if (!iov_iter_count(from))
2465 /* FIXME: this is for backwards compatibility with 2.4 */
2466 if (iocb->ki_flags & IOCB_APPEND)
2467 iocb->ki_pos = i_size_read(inode);
2471 if (limit != RLIM_INFINITY) {
2472 if (iocb->ki_pos >= limit) {
2473 send_sig(SIGXFSZ, current, 0);
2476 iov_iter_truncate(from, limit - (unsigned long)pos);
2482 if (unlikely(pos + iov_iter_count(from) > MAX_NON_LFS &&
2483 !(file->f_flags & O_LARGEFILE))) {
2484 if (pos >= MAX_NON_LFS)
2486 iov_iter_truncate(from, MAX_NON_LFS - (unsigned long)pos);
2490 * Are we about to exceed the fs block limit ?
2492 * If we have written data it becomes a short write. If we have
2493 * exceeded without writing data we send a signal and return EFBIG.
2494 * Linus frestrict idea will clean these up nicely..
2496 if (unlikely(pos >= inode->i_sb->s_maxbytes))
2499 iov_iter_truncate(from, inode->i_sb->s_maxbytes - pos);
2500 return iov_iter_count(from);
2502 EXPORT_SYMBOL(generic_write_checks);
2504 int pagecache_write_begin(struct file *file, struct address_space *mapping,
2505 loff_t pos, unsigned len, unsigned flags,
2506 struct page **pagep, void **fsdata)
2508 const struct address_space_operations *aops = mapping->a_ops;
2510 return aops->write_begin(file, mapping, pos, len, flags,
2513 EXPORT_SYMBOL(pagecache_write_begin);
2515 int pagecache_write_end(struct file *file, struct address_space *mapping,
2516 loff_t pos, unsigned len, unsigned copied,
2517 struct page *page, void *fsdata)
2519 const struct address_space_operations *aops = mapping->a_ops;
2521 return aops->write_end(file, mapping, pos, len, copied, page, fsdata);
2523 EXPORT_SYMBOL(pagecache_write_end);
2526 generic_file_direct_write(struct kiocb *iocb, struct iov_iter *from, loff_t pos)
2528 struct file *file = iocb->ki_filp;
2529 struct address_space *mapping = file->f_mapping;
2530 struct inode *inode = mapping->host;
2534 struct iov_iter data;
2536 write_len = iov_iter_count(from);
2537 end = (pos + write_len - 1) >> PAGE_CACHE_SHIFT;
2539 written = filemap_write_and_wait_range(mapping, pos, pos + write_len - 1);
2544 * After a write we want buffered reads to be sure to go to disk to get
2545 * the new data. We invalidate clean cached page from the region we're
2546 * about to write. We do this *before* the write so that we can return
2547 * without clobbering -EIOCBQUEUED from ->direct_IO().
2549 if (mapping->nrpages) {
2550 written = invalidate_inode_pages2_range(mapping,
2551 pos >> PAGE_CACHE_SHIFT, end);
2553 * If a page can not be invalidated, return 0 to fall back
2554 * to buffered write.
2557 if (written == -EBUSY)
2564 written = mapping->a_ops->direct_IO(iocb, &data, pos);
2567 * Finally, try again to invalidate clean pages which might have been
2568 * cached by non-direct readahead, or faulted in by get_user_pages()
2569 * if the source of the write was an mmap'ed region of the file
2570 * we're writing. Either one is a pretty crazy thing to do,
2571 * so we don't support it 100%. If this invalidation
2572 * fails, tough, the write still worked...
2574 if (mapping->nrpages) {
2575 invalidate_inode_pages2_range(mapping,
2576 pos >> PAGE_CACHE_SHIFT, end);
2581 iov_iter_advance(from, written);
2582 if (pos > i_size_read(inode) && !S_ISBLK(inode->i_mode)) {
2583 i_size_write(inode, pos);
2584 mark_inode_dirty(inode);
2591 EXPORT_SYMBOL(generic_file_direct_write);
2594 * Find or create a page at the given pagecache position. Return the locked
2595 * page. This function is specifically for buffered writes.
2597 struct page *grab_cache_page_write_begin(struct address_space *mapping,
2598 pgoff_t index, unsigned flags)
2601 int fgp_flags = FGP_LOCK|FGP_ACCESSED|FGP_WRITE|FGP_CREAT;
2603 if (flags & AOP_FLAG_NOFS)
2604 fgp_flags |= FGP_NOFS;
2606 page = pagecache_get_page(mapping, index, fgp_flags,
2607 mapping_gfp_mask(mapping));
2609 wait_for_stable_page(page);
2613 EXPORT_SYMBOL(grab_cache_page_write_begin);
2615 ssize_t generic_perform_write(struct file *file,
2616 struct iov_iter *i, loff_t pos)
2618 struct address_space *mapping = file->f_mapping;
2619 const struct address_space_operations *a_ops = mapping->a_ops;
2621 ssize_t written = 0;
2622 unsigned int flags = 0;
2625 * Copies from kernel address space cannot fail (NFSD is a big user).
2627 if (!iter_is_iovec(i))
2628 flags |= AOP_FLAG_UNINTERRUPTIBLE;
2632 unsigned long offset; /* Offset into pagecache page */
2633 unsigned long bytes; /* Bytes to write to page */
2634 size_t copied; /* Bytes copied from user */
2637 offset = (pos & (PAGE_CACHE_SIZE - 1));
2638 bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset,
2643 * Bring in the user page that we will copy from _first_.
2644 * Otherwise there's a nasty deadlock on copying from the
2645 * same page as we're writing to, without it being marked
2648 * Not only is this an optimisation, but it is also required
2649 * to check that the address is actually valid, when atomic
2650 * usercopies are used, below.
2652 if (unlikely(iov_iter_fault_in_readable(i, bytes))) {
2657 if (fatal_signal_pending(current)) {
2662 status = a_ops->write_begin(file, mapping, pos, bytes, flags,
2664 if (unlikely(status < 0))
2667 if (mapping_writably_mapped(mapping))
2668 flush_dcache_page(page);
2670 copied = iov_iter_copy_from_user_atomic(page, i, offset, bytes);
2671 flush_dcache_page(page);
2673 status = a_ops->write_end(file, mapping, pos, bytes, copied,
2675 if (unlikely(status < 0))
2681 iov_iter_advance(i, copied);
2682 if (unlikely(copied == 0)) {
2684 * If we were unable to copy any data at all, we must
2685 * fall back to a single segment length write.
2687 * If we didn't fallback here, we could livelock
2688 * because not all segments in the iov can be copied at
2689 * once without a pagefault.
2691 bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset,
2692 iov_iter_single_seg_count(i));
2698 balance_dirty_pages_ratelimited(mapping);
2699 } while (iov_iter_count(i));
2701 return written ? written : status;
2703 EXPORT_SYMBOL(generic_perform_write);
2706 * __generic_file_write_iter - write data to a file
2707 * @iocb: IO state structure (file, offset, etc.)
2708 * @from: iov_iter with data to write
2710 * This function does all the work needed for actually writing data to a
2711 * file. It does all basic checks, removes SUID from the file, updates
2712 * modification times and calls proper subroutines depending on whether we
2713 * do direct IO or a standard buffered write.
2715 * It expects i_mutex to be grabbed unless we work on a block device or similar
2716 * object which does not need locking at all.
2718 * This function does *not* take care of syncing data in case of O_SYNC write.
2719 * A caller has to handle it. This is mainly due to the fact that we want to
2720 * avoid syncing under i_mutex.
2722 ssize_t __generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
2724 struct file *file = iocb->ki_filp;
2725 struct address_space * mapping = file->f_mapping;
2726 struct inode *inode = mapping->host;
2727 ssize_t written = 0;
2731 /* We can write back this queue in page reclaim */
2732 current->backing_dev_info = inode_to_bdi(inode);
2733 err = file_remove_privs(file);
2737 err = file_update_time(file);
2741 if (iocb->ki_flags & IOCB_DIRECT) {
2742 loff_t pos, endbyte;
2744 written = generic_file_direct_write(iocb, from, iocb->ki_pos);
2746 * If the write stopped short of completing, fall back to
2747 * buffered writes. Some filesystems do this for writes to
2748 * holes, for example. For DAX files, a buffered write will
2749 * not succeed (even if it did, DAX does not handle dirty
2750 * page-cache pages correctly).
2752 if (written < 0 || !iov_iter_count(from) || IS_DAX(inode))
2755 status = generic_perform_write(file, from, pos = iocb->ki_pos);
2757 * If generic_perform_write() returned a synchronous error
2758 * then we want to return the number of bytes which were
2759 * direct-written, or the error code if that was zero. Note
2760 * that this differs from normal direct-io semantics, which
2761 * will return -EFOO even if some bytes were written.
2763 if (unlikely(status < 0)) {
2768 * We need to ensure that the page cache pages are written to
2769 * disk and invalidated to preserve the expected O_DIRECT
2772 endbyte = pos + status - 1;
2773 err = filemap_write_and_wait_range(mapping, pos, endbyte);
2775 iocb->ki_pos = endbyte + 1;
2777 invalidate_mapping_pages(mapping,
2778 pos >> PAGE_CACHE_SHIFT,
2779 endbyte >> PAGE_CACHE_SHIFT);
2782 * We don't know how much we wrote, so just return
2783 * the number of bytes which were direct-written
2787 written = generic_perform_write(file, from, iocb->ki_pos);
2788 if (likely(written > 0))
2789 iocb->ki_pos += written;
2792 current->backing_dev_info = NULL;
2793 return written ? written : err;
2795 EXPORT_SYMBOL(__generic_file_write_iter);
2798 * generic_file_write_iter - write data to a file
2799 * @iocb: IO state structure
2800 * @from: iov_iter with data to write
2802 * This is a wrapper around __generic_file_write_iter() to be used by most
2803 * filesystems. It takes care of syncing the file in case of O_SYNC file
2804 * and acquires i_mutex as needed.
2806 ssize_t generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
2808 struct file *file = iocb->ki_filp;
2809 struct inode *inode = file->f_mapping->host;
2813 ret = generic_write_checks(iocb, from);
2815 ret = __generic_file_write_iter(iocb, from);
2816 inode_unlock(inode);
2821 err = generic_write_sync(file, iocb->ki_pos - ret, ret);
2827 EXPORT_SYMBOL(generic_file_write_iter);
2830 * try_to_release_page() - release old fs-specific metadata on a page
2832 * @page: the page which the kernel is trying to free
2833 * @gfp_mask: memory allocation flags (and I/O mode)
2835 * The address_space is to try to release any data against the page
2836 * (presumably at page->private). If the release was successful, return `1'.
2837 * Otherwise return zero.
2839 * This may also be called if PG_fscache is set on a page, indicating that the
2840 * page is known to the local caching routines.
2842 * The @gfp_mask argument specifies whether I/O may be performed to release
2843 * this page (__GFP_IO), and whether the call may block (__GFP_RECLAIM & __GFP_FS).
2846 int try_to_release_page(struct page *page, gfp_t gfp_mask)
2848 struct address_space * const mapping = page->mapping;
2850 BUG_ON(!PageLocked(page));
2851 if (PageWriteback(page))
2854 if (mapping && mapping->a_ops->releasepage)
2855 return mapping->a_ops->releasepage(page, gfp_mask);
2856 return try_to_free_buffers(page);
2859 EXPORT_SYMBOL(try_to_release_page);