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.
181 void __delete_from_page_cache(struct page *page, void *shadow)
183 struct address_space *mapping = page->mapping;
185 trace_mm_filemap_delete_from_page_cache(page);
187 * if we're uptodate, flush out into the cleancache, otherwise
188 * invalidate any existing cleancache entries. We can't leave
189 * stale data around in the cleancache once our page is gone
191 if (PageUptodate(page) && PageMappedToDisk(page))
192 cleancache_put_page(page);
194 cleancache_invalidate_page(mapping, page);
196 VM_BUG_ON_PAGE(page_mapped(page), page);
197 if (!IS_ENABLED(CONFIG_DEBUG_VM) && unlikely(page_mapped(page))) {
200 pr_alert("BUG: Bad page cache in process %s pfn:%05lx\n",
201 current->comm, page_to_pfn(page));
202 dump_page(page, "still mapped when deleted");
204 add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE);
206 mapcount = page_mapcount(page);
207 if (mapping_exiting(mapping) &&
208 page_count(page) >= mapcount + 2) {
210 * All vmas have already been torn down, so it's
211 * a good bet that actually the page is unmapped,
212 * and we'd prefer not to leak it: if we're wrong,
213 * some other bad page check should catch it later.
215 page_mapcount_reset(page);
216 atomic_sub(mapcount, &page->_count);
220 page_cache_tree_delete(mapping, page, shadow);
222 page->mapping = NULL;
223 /* Leave page->index set: truncation lookup relies upon it */
225 /* hugetlb pages do not participate in page cache accounting. */
227 __dec_zone_page_state(page, NR_FILE_PAGES);
228 if (PageSwapBacked(page))
229 __dec_zone_page_state(page, NR_SHMEM);
232 * At this point page must be either written or cleaned by truncate.
233 * Dirty page here signals a bug and loss of unwritten data.
235 * This fixes dirty accounting after removing the page entirely but
236 * leaves PageDirty set: it has no effect for truncated page and
237 * anyway will be cleared before returning page into buddy allocator.
239 if (WARN_ON_ONCE(PageDirty(page)))
240 account_page_cleaned(page, mapping, inode_to_wb(mapping->host));
244 * delete_from_page_cache - delete page from page cache
245 * @page: the page which the kernel is trying to remove from page cache
247 * This must be called only on pages that have been verified to be in the page
248 * cache and locked. It will never put the page into the free list, the caller
249 * has a reference on the page.
251 void delete_from_page_cache(struct page *page)
253 struct address_space *mapping = page->mapping;
256 void (*freepage)(struct page *);
258 BUG_ON(!PageLocked(page));
260 freepage = mapping->a_ops->freepage;
262 spin_lock_irqsave(&mapping->tree_lock, flags);
263 __delete_from_page_cache(page, NULL);
264 spin_unlock_irqrestore(&mapping->tree_lock, flags);
270 EXPORT_SYMBOL(delete_from_page_cache);
272 static int filemap_check_errors(struct address_space *mapping)
275 /* Check for outstanding write errors */
276 if (test_bit(AS_ENOSPC, &mapping->flags) &&
277 test_and_clear_bit(AS_ENOSPC, &mapping->flags))
279 if (test_bit(AS_EIO, &mapping->flags) &&
280 test_and_clear_bit(AS_EIO, &mapping->flags))
286 * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range
287 * @mapping: address space structure to write
288 * @start: offset in bytes where the range starts
289 * @end: offset in bytes where the range ends (inclusive)
290 * @sync_mode: enable synchronous operation
292 * Start writeback against all of a mapping's dirty pages that lie
293 * within the byte offsets <start, end> inclusive.
295 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
296 * opposed to a regular memory cleansing writeback. The difference between
297 * these two operations is that if a dirty page/buffer is encountered, it must
298 * be waited upon, and not just skipped over.
300 int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
301 loff_t end, int sync_mode)
304 struct writeback_control wbc = {
305 .sync_mode = sync_mode,
306 .nr_to_write = LONG_MAX,
307 .range_start = start,
311 if (!mapping_cap_writeback_dirty(mapping))
314 wbc_attach_fdatawrite_inode(&wbc, mapping->host);
315 ret = do_writepages(mapping, &wbc);
316 wbc_detach_inode(&wbc);
320 static inline int __filemap_fdatawrite(struct address_space *mapping,
323 return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode);
326 int filemap_fdatawrite(struct address_space *mapping)
328 return __filemap_fdatawrite(mapping, WB_SYNC_ALL);
330 EXPORT_SYMBOL(filemap_fdatawrite);
332 int filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
335 return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL);
337 EXPORT_SYMBOL(filemap_fdatawrite_range);
340 * filemap_flush - mostly a non-blocking flush
341 * @mapping: target address_space
343 * This is a mostly non-blocking flush. Not suitable for data-integrity
344 * purposes - I/O may not be started against all dirty pages.
346 int filemap_flush(struct address_space *mapping)
348 return __filemap_fdatawrite(mapping, WB_SYNC_NONE);
350 EXPORT_SYMBOL(filemap_flush);
352 static int __filemap_fdatawait_range(struct address_space *mapping,
353 loff_t start_byte, loff_t end_byte)
355 pgoff_t index = start_byte >> PAGE_SHIFT;
356 pgoff_t end = end_byte >> PAGE_SHIFT;
361 if (end_byte < start_byte)
364 pagevec_init(&pvec, 0);
365 while ((index <= end) &&
366 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
367 PAGECACHE_TAG_WRITEBACK,
368 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1)) != 0) {
371 for (i = 0; i < nr_pages; i++) {
372 struct page *page = pvec.pages[i];
374 /* until radix tree lookup accepts end_index */
375 if (page->index > end)
378 wait_on_page_writeback(page);
379 if (TestClearPageError(page))
382 pagevec_release(&pvec);
390 * filemap_fdatawait_range - wait for writeback to complete
391 * @mapping: address space structure to wait for
392 * @start_byte: offset in bytes where the range starts
393 * @end_byte: offset in bytes where the range ends (inclusive)
395 * Walk the list of under-writeback pages of the given address space
396 * in the given range and wait for all of them. Check error status of
397 * the address space and return it.
399 * Since the error status of the address space is cleared by this function,
400 * callers are responsible for checking the return value and handling and/or
401 * reporting the error.
403 int filemap_fdatawait_range(struct address_space *mapping, loff_t start_byte,
408 ret = __filemap_fdatawait_range(mapping, start_byte, end_byte);
409 ret2 = filemap_check_errors(mapping);
415 EXPORT_SYMBOL(filemap_fdatawait_range);
418 * filemap_fdatawait_keep_errors - wait for writeback without clearing errors
419 * @mapping: address space structure to wait for
421 * Walk the list of under-writeback pages of the given address space
422 * and wait for all of them. Unlike filemap_fdatawait(), this function
423 * does not clear error status of the address space.
425 * Use this function if callers don't handle errors themselves. Expected
426 * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2),
429 void filemap_fdatawait_keep_errors(struct address_space *mapping)
431 loff_t i_size = i_size_read(mapping->host);
436 __filemap_fdatawait_range(mapping, 0, i_size - 1);
440 * filemap_fdatawait - wait for all under-writeback pages to complete
441 * @mapping: address space structure to wait for
443 * Walk the list of under-writeback pages of the given address space
444 * and wait for all of them. Check error status of the address space
447 * Since the error status of the address space is cleared by this function,
448 * callers are responsible for checking the return value and handling and/or
449 * reporting the error.
451 int filemap_fdatawait(struct address_space *mapping)
453 loff_t i_size = i_size_read(mapping->host);
458 return filemap_fdatawait_range(mapping, 0, i_size - 1);
460 EXPORT_SYMBOL(filemap_fdatawait);
462 int filemap_write_and_wait(struct address_space *mapping)
466 if ((!dax_mapping(mapping) && mapping->nrpages) ||
467 (dax_mapping(mapping) && mapping->nrexceptional)) {
468 err = filemap_fdatawrite(mapping);
470 * Even if the above returned error, the pages may be
471 * written partially (e.g. -ENOSPC), so we wait for it.
472 * But the -EIO is special case, it may indicate the worst
473 * thing (e.g. bug) happened, so we avoid waiting for it.
476 int err2 = filemap_fdatawait(mapping);
481 err = filemap_check_errors(mapping);
485 EXPORT_SYMBOL(filemap_write_and_wait);
488 * filemap_write_and_wait_range - write out & wait on a file range
489 * @mapping: the address_space for the pages
490 * @lstart: offset in bytes where the range starts
491 * @lend: offset in bytes where the range ends (inclusive)
493 * Write out and wait upon file offsets lstart->lend, inclusive.
495 * Note that `lend' is inclusive (describes the last byte to be written) so
496 * that this function can be used to write to the very end-of-file (end = -1).
498 int filemap_write_and_wait_range(struct address_space *mapping,
499 loff_t lstart, loff_t lend)
503 if ((!dax_mapping(mapping) && mapping->nrpages) ||
504 (dax_mapping(mapping) && mapping->nrexceptional)) {
505 err = __filemap_fdatawrite_range(mapping, lstart, lend,
507 /* See comment of filemap_write_and_wait() */
509 int err2 = filemap_fdatawait_range(mapping,
515 err = filemap_check_errors(mapping);
519 EXPORT_SYMBOL(filemap_write_and_wait_range);
522 * replace_page_cache_page - replace a pagecache page with a new one
523 * @old: page to be replaced
524 * @new: page to replace with
525 * @gfp_mask: allocation mode
527 * This function replaces a page in the pagecache with a new one. On
528 * success it acquires the pagecache reference for the new page and
529 * drops it for the old page. Both the old and new pages must be
530 * locked. This function does not add the new page to the LRU, the
531 * caller must do that.
533 * The remove + add is atomic. The only way this function can fail is
534 * memory allocation failure.
536 int replace_page_cache_page(struct page *old, struct page *new, gfp_t gfp_mask)
540 VM_BUG_ON_PAGE(!PageLocked(old), old);
541 VM_BUG_ON_PAGE(!PageLocked(new), new);
542 VM_BUG_ON_PAGE(new->mapping, new);
544 error = radix_tree_preload(gfp_mask & ~__GFP_HIGHMEM);
546 struct address_space *mapping = old->mapping;
547 void (*freepage)(struct page *);
550 pgoff_t offset = old->index;
551 freepage = mapping->a_ops->freepage;
554 new->mapping = mapping;
557 spin_lock_irqsave(&mapping->tree_lock, flags);
558 __delete_from_page_cache(old, NULL);
559 error = radix_tree_insert(&mapping->page_tree, offset, new);
564 * hugetlb pages do not participate in page cache accounting.
567 __inc_zone_page_state(new, NR_FILE_PAGES);
568 if (PageSwapBacked(new))
569 __inc_zone_page_state(new, NR_SHMEM);
570 spin_unlock_irqrestore(&mapping->tree_lock, flags);
571 mem_cgroup_migrate(old, new);
572 radix_tree_preload_end();
580 EXPORT_SYMBOL_GPL(replace_page_cache_page);
582 static int page_cache_tree_insert(struct address_space *mapping,
583 struct page *page, void **shadowp)
585 struct radix_tree_node *node;
589 error = __radix_tree_create(&mapping->page_tree, page->index, 0,
596 p = radix_tree_deref_slot_protected(slot, &mapping->tree_lock);
597 if (!radix_tree_exceptional_entry(p))
600 mapping->nrexceptional--;
601 if (!dax_mapping(mapping)) {
605 workingset_node_shadows_dec(node);
607 /* DAX can replace empty locked entry with a hole */
609 (void *)(RADIX_TREE_EXCEPTIONAL_ENTRY |
610 RADIX_DAX_ENTRY_LOCK));
611 /* DAX accounts exceptional entries as normal pages */
613 workingset_node_pages_dec(node);
616 radix_tree_replace_slot(slot, page);
619 workingset_node_pages_inc(node);
621 * Don't track node that contains actual pages.
623 * Avoid acquiring the list_lru lock if already
624 * untracked. The list_empty() test is safe as
625 * node->private_list is protected by
626 * mapping->tree_lock.
628 if (!list_empty(&node->private_list))
629 list_lru_del(&workingset_shadow_nodes,
630 &node->private_list);
635 static int __add_to_page_cache_locked(struct page *page,
636 struct address_space *mapping,
637 pgoff_t offset, gfp_t gfp_mask,
640 int huge = PageHuge(page);
641 struct mem_cgroup *memcg;
644 VM_BUG_ON_PAGE(!PageLocked(page), page);
645 VM_BUG_ON_PAGE(PageSwapBacked(page), page);
648 error = mem_cgroup_try_charge(page, current->mm,
649 gfp_mask, &memcg, false);
654 error = radix_tree_maybe_preload(gfp_mask & ~__GFP_HIGHMEM);
657 mem_cgroup_cancel_charge(page, memcg, false);
662 page->mapping = mapping;
663 page->index = offset;
665 spin_lock_irq(&mapping->tree_lock);
666 error = page_cache_tree_insert(mapping, page, shadowp);
667 radix_tree_preload_end();
671 /* hugetlb pages do not participate in page cache accounting. */
673 __inc_zone_page_state(page, NR_FILE_PAGES);
674 spin_unlock_irq(&mapping->tree_lock);
676 mem_cgroup_commit_charge(page, memcg, false, false);
677 trace_mm_filemap_add_to_page_cache(page);
680 page->mapping = NULL;
681 /* Leave page->index set: truncation relies upon it */
682 spin_unlock_irq(&mapping->tree_lock);
684 mem_cgroup_cancel_charge(page, memcg, false);
690 * add_to_page_cache_locked - add a locked page to the pagecache
692 * @mapping: the page's address_space
693 * @offset: page index
694 * @gfp_mask: page allocation mode
696 * This function is used to add a page to the pagecache. It must be locked.
697 * This function does not add the page to the LRU. The caller must do that.
699 int add_to_page_cache_locked(struct page *page, struct address_space *mapping,
700 pgoff_t offset, gfp_t gfp_mask)
702 return __add_to_page_cache_locked(page, mapping, offset,
705 EXPORT_SYMBOL(add_to_page_cache_locked);
707 int add_to_page_cache_lru(struct page *page, struct address_space *mapping,
708 pgoff_t offset, gfp_t gfp_mask)
713 __SetPageLocked(page);
714 ret = __add_to_page_cache_locked(page, mapping, offset,
717 __ClearPageLocked(page);
720 * The page might have been evicted from cache only
721 * recently, in which case it should be activated like
722 * any other repeatedly accessed page.
724 if (shadow && workingset_refault(shadow)) {
726 workingset_activation(page);
728 ClearPageActive(page);
733 EXPORT_SYMBOL_GPL(add_to_page_cache_lru);
736 struct page *__page_cache_alloc(gfp_t gfp)
741 if (cpuset_do_page_mem_spread()) {
742 unsigned int cpuset_mems_cookie;
744 cpuset_mems_cookie = read_mems_allowed_begin();
745 n = cpuset_mem_spread_node();
746 page = __alloc_pages_node(n, gfp, 0);
747 } while (!page && read_mems_allowed_retry(cpuset_mems_cookie));
751 return alloc_pages(gfp, 0);
753 EXPORT_SYMBOL(__page_cache_alloc);
757 * In order to wait for pages to become available there must be
758 * waitqueues associated with pages. By using a hash table of
759 * waitqueues where the bucket discipline is to maintain all
760 * waiters on the same queue and wake all when any of the pages
761 * become available, and for the woken contexts to check to be
762 * sure the appropriate page became available, this saves space
763 * at a cost of "thundering herd" phenomena during rare hash
766 wait_queue_head_t *page_waitqueue(struct page *page)
768 const struct zone *zone = page_zone(page);
770 return &zone->wait_table[hash_ptr(page, zone->wait_table_bits)];
772 EXPORT_SYMBOL(page_waitqueue);
774 void wait_on_page_bit(struct page *page, int bit_nr)
776 DEFINE_WAIT_BIT(wait, &page->flags, bit_nr);
778 if (test_bit(bit_nr, &page->flags))
779 __wait_on_bit(page_waitqueue(page), &wait, bit_wait_io,
780 TASK_UNINTERRUPTIBLE);
782 EXPORT_SYMBOL(wait_on_page_bit);
784 int wait_on_page_bit_killable(struct page *page, int bit_nr)
786 DEFINE_WAIT_BIT(wait, &page->flags, bit_nr);
788 if (!test_bit(bit_nr, &page->flags))
791 return __wait_on_bit(page_waitqueue(page), &wait,
792 bit_wait_io, TASK_KILLABLE);
795 int wait_on_page_bit_killable_timeout(struct page *page,
796 int bit_nr, unsigned long timeout)
798 DEFINE_WAIT_BIT(wait, &page->flags, bit_nr);
800 wait.key.timeout = jiffies + timeout;
801 if (!test_bit(bit_nr, &page->flags))
803 return __wait_on_bit(page_waitqueue(page), &wait,
804 bit_wait_io_timeout, TASK_KILLABLE);
806 EXPORT_SYMBOL_GPL(wait_on_page_bit_killable_timeout);
809 * add_page_wait_queue - Add an arbitrary waiter to a page's wait queue
810 * @page: Page defining the wait queue of interest
811 * @waiter: Waiter to add to the queue
813 * Add an arbitrary @waiter to the wait queue for the nominated @page.
815 void add_page_wait_queue(struct page *page, wait_queue_t *waiter)
817 wait_queue_head_t *q = page_waitqueue(page);
820 spin_lock_irqsave(&q->lock, flags);
821 __add_wait_queue(q, waiter);
822 spin_unlock_irqrestore(&q->lock, flags);
824 EXPORT_SYMBOL_GPL(add_page_wait_queue);
827 * unlock_page - unlock a locked page
830 * Unlocks the page and wakes up sleepers in ___wait_on_page_locked().
831 * Also wakes sleepers in wait_on_page_writeback() because the wakeup
832 * mechanism between PageLocked pages and PageWriteback pages is shared.
833 * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep.
835 * The mb is necessary to enforce ordering between the clear_bit and the read
836 * of the waitqueue (to avoid SMP races with a parallel wait_on_page_locked()).
838 void unlock_page(struct page *page)
840 page = compound_head(page);
841 VM_BUG_ON_PAGE(!PageLocked(page), page);
842 clear_bit_unlock(PG_locked, &page->flags);
843 smp_mb__after_atomic();
844 wake_up_page(page, PG_locked);
846 EXPORT_SYMBOL(unlock_page);
849 * end_page_writeback - end writeback against a page
852 void end_page_writeback(struct page *page)
855 * TestClearPageReclaim could be used here but it is an atomic
856 * operation and overkill in this particular case. Failing to
857 * shuffle a page marked for immediate reclaim is too mild to
858 * justify taking an atomic operation penalty at the end of
859 * ever page writeback.
861 if (PageReclaim(page)) {
862 ClearPageReclaim(page);
863 rotate_reclaimable_page(page);
866 if (!test_clear_page_writeback(page))
869 smp_mb__after_atomic();
870 wake_up_page(page, PG_writeback);
872 EXPORT_SYMBOL(end_page_writeback);
875 * After completing I/O on a page, call this routine to update the page
876 * flags appropriately
878 void page_endio(struct page *page, int rw, int err)
882 SetPageUptodate(page);
884 ClearPageUptodate(page);
888 } else { /* rw == WRITE */
892 mapping_set_error(page->mapping, err);
894 end_page_writeback(page);
897 EXPORT_SYMBOL_GPL(page_endio);
900 * __lock_page - get a lock on the page, assuming we need to sleep to get it
901 * @page: the page to lock
903 void __lock_page(struct page *page)
905 struct page *page_head = compound_head(page);
906 DEFINE_WAIT_BIT(wait, &page_head->flags, PG_locked);
908 __wait_on_bit_lock(page_waitqueue(page_head), &wait, bit_wait_io,
909 TASK_UNINTERRUPTIBLE);
911 EXPORT_SYMBOL(__lock_page);
913 int __lock_page_killable(struct page *page)
915 struct page *page_head = compound_head(page);
916 DEFINE_WAIT_BIT(wait, &page_head->flags, PG_locked);
918 return __wait_on_bit_lock(page_waitqueue(page_head), &wait,
919 bit_wait_io, TASK_KILLABLE);
921 EXPORT_SYMBOL_GPL(__lock_page_killable);
925 * 1 - page is locked; mmap_sem is still held.
926 * 0 - page is not locked.
927 * mmap_sem has been released (up_read()), unless flags had both
928 * FAULT_FLAG_ALLOW_RETRY and FAULT_FLAG_RETRY_NOWAIT set, in
929 * which case mmap_sem is still held.
931 * If neither ALLOW_RETRY nor KILLABLE are set, will always return 1
932 * with the page locked and the mmap_sem unperturbed.
934 int __lock_page_or_retry(struct page *page, struct mm_struct *mm,
937 if (flags & FAULT_FLAG_ALLOW_RETRY) {
939 * CAUTION! In this case, mmap_sem is not released
940 * even though return 0.
942 if (flags & FAULT_FLAG_RETRY_NOWAIT)
945 up_read(&mm->mmap_sem);
946 if (flags & FAULT_FLAG_KILLABLE)
947 wait_on_page_locked_killable(page);
949 wait_on_page_locked(page);
952 if (flags & FAULT_FLAG_KILLABLE) {
955 ret = __lock_page_killable(page);
957 up_read(&mm->mmap_sem);
967 * page_cache_next_hole - find the next hole (not-present entry)
970 * @max_scan: maximum range to search
972 * Search the set [index, min(index+max_scan-1, MAX_INDEX)] for the
973 * lowest indexed hole.
975 * Returns: the index of the hole if found, otherwise returns an index
976 * outside of the set specified (in which case 'return - index >=
977 * max_scan' will be true). In rare cases of index wrap-around, 0 will
980 * page_cache_next_hole may be called under rcu_read_lock. However,
981 * like radix_tree_gang_lookup, this will not atomically search a
982 * snapshot of the tree at a single point in time. For example, if a
983 * hole is created at index 5, then subsequently a hole is created at
984 * index 10, page_cache_next_hole covering both indexes may return 10
985 * if called under rcu_read_lock.
987 pgoff_t page_cache_next_hole(struct address_space *mapping,
988 pgoff_t index, unsigned long max_scan)
992 for (i = 0; i < max_scan; i++) {
995 page = radix_tree_lookup(&mapping->page_tree, index);
996 if (!page || radix_tree_exceptional_entry(page))
1005 EXPORT_SYMBOL(page_cache_next_hole);
1008 * page_cache_prev_hole - find the prev hole (not-present entry)
1011 * @max_scan: maximum range to search
1013 * Search backwards in the range [max(index-max_scan+1, 0), index] for
1016 * Returns: the index of the hole if found, otherwise returns an index
1017 * outside of the set specified (in which case 'index - return >=
1018 * max_scan' will be true). In rare cases of wrap-around, ULONG_MAX
1021 * page_cache_prev_hole may be called under rcu_read_lock. However,
1022 * like radix_tree_gang_lookup, this will not atomically search a
1023 * snapshot of the tree at a single point in time. For example, if a
1024 * hole is created at index 10, then subsequently a hole is created at
1025 * index 5, page_cache_prev_hole covering both indexes may return 5 if
1026 * called under rcu_read_lock.
1028 pgoff_t page_cache_prev_hole(struct address_space *mapping,
1029 pgoff_t index, unsigned long max_scan)
1033 for (i = 0; i < max_scan; i++) {
1036 page = radix_tree_lookup(&mapping->page_tree, index);
1037 if (!page || radix_tree_exceptional_entry(page))
1040 if (index == ULONG_MAX)
1046 EXPORT_SYMBOL(page_cache_prev_hole);
1049 * find_get_entry - find and get a page cache entry
1050 * @mapping: the address_space to search
1051 * @offset: the page cache index
1053 * Looks up the page cache slot at @mapping & @offset. If there is a
1054 * page cache page, it is returned with an increased refcount.
1056 * If the slot holds a shadow entry of a previously evicted page, or a
1057 * swap entry from shmem/tmpfs, it is returned.
1059 * Otherwise, %NULL is returned.
1061 struct page *find_get_entry(struct address_space *mapping, pgoff_t offset)
1069 pagep = radix_tree_lookup_slot(&mapping->page_tree, offset);
1071 page = radix_tree_deref_slot(pagep);
1072 if (unlikely(!page))
1074 if (radix_tree_exception(page)) {
1075 if (radix_tree_deref_retry(page))
1078 * A shadow entry of a recently evicted page,
1079 * or a swap entry from shmem/tmpfs. Return
1080 * it without attempting to raise page count.
1084 if (!page_cache_get_speculative(page))
1088 * Has the page moved?
1089 * This is part of the lockless pagecache protocol. See
1090 * include/linux/pagemap.h for details.
1092 if (unlikely(page != *pagep)) {
1102 EXPORT_SYMBOL(find_get_entry);
1105 * find_lock_entry - locate, pin and lock a page cache entry
1106 * @mapping: the address_space to search
1107 * @offset: the page cache index
1109 * Looks up the page cache slot at @mapping & @offset. If there is a
1110 * page cache page, it is returned locked and with an increased
1113 * If the slot holds a shadow entry of a previously evicted page, or a
1114 * swap entry from shmem/tmpfs, it is returned.
1116 * Otherwise, %NULL is returned.
1118 * find_lock_entry() may sleep.
1120 struct page *find_lock_entry(struct address_space *mapping, pgoff_t offset)
1125 page = find_get_entry(mapping, offset);
1126 if (page && !radix_tree_exception(page)) {
1128 /* Has the page been truncated? */
1129 if (unlikely(page->mapping != mapping)) {
1134 VM_BUG_ON_PAGE(page->index != offset, page);
1138 EXPORT_SYMBOL(find_lock_entry);
1141 * pagecache_get_page - find and get a page reference
1142 * @mapping: the address_space to search
1143 * @offset: the page index
1144 * @fgp_flags: PCG flags
1145 * @gfp_mask: gfp mask to use for the page cache data page allocation
1147 * Looks up the page cache slot at @mapping & @offset.
1149 * PCG flags modify how the page is returned.
1151 * FGP_ACCESSED: the page will be marked accessed
1152 * FGP_LOCK: Page is return locked
1153 * FGP_CREAT: If page is not present then a new page is allocated using
1154 * @gfp_mask and added to the page cache and the VM's LRU
1155 * list. The page is returned locked and with an increased
1156 * refcount. Otherwise, %NULL is returned.
1158 * If FGP_LOCK or FGP_CREAT are specified then the function may sleep even
1159 * if the GFP flags specified for FGP_CREAT are atomic.
1161 * If there is a page cache page, it is returned with an increased refcount.
1163 struct page *pagecache_get_page(struct address_space *mapping, pgoff_t offset,
1164 int fgp_flags, gfp_t gfp_mask)
1169 page = find_get_entry(mapping, offset);
1170 if (radix_tree_exceptional_entry(page))
1175 if (fgp_flags & FGP_LOCK) {
1176 if (fgp_flags & FGP_NOWAIT) {
1177 if (!trylock_page(page)) {
1185 /* Has the page been truncated? */
1186 if (unlikely(page->mapping != mapping)) {
1191 VM_BUG_ON_PAGE(page->index != offset, page);
1194 if (page && (fgp_flags & FGP_ACCESSED))
1195 mark_page_accessed(page);
1198 if (!page && (fgp_flags & FGP_CREAT)) {
1200 if ((fgp_flags & FGP_WRITE) && mapping_cap_account_dirty(mapping))
1201 gfp_mask |= __GFP_WRITE;
1202 if (fgp_flags & FGP_NOFS)
1203 gfp_mask &= ~__GFP_FS;
1205 page = __page_cache_alloc(gfp_mask);
1209 if (WARN_ON_ONCE(!(fgp_flags & FGP_LOCK)))
1210 fgp_flags |= FGP_LOCK;
1212 /* Init accessed so avoid atomic mark_page_accessed later */
1213 if (fgp_flags & FGP_ACCESSED)
1214 __SetPageReferenced(page);
1216 err = add_to_page_cache_lru(page, mapping, offset,
1217 gfp_mask & GFP_RECLAIM_MASK);
1218 if (unlikely(err)) {
1228 EXPORT_SYMBOL(pagecache_get_page);
1231 * find_get_entries - gang pagecache lookup
1232 * @mapping: The address_space to search
1233 * @start: The starting page cache index
1234 * @nr_entries: The maximum number of entries
1235 * @entries: Where the resulting entries are placed
1236 * @indices: The cache indices corresponding to the entries in @entries
1238 * find_get_entries() will search for and return a group of up to
1239 * @nr_entries entries in the mapping. The entries are placed at
1240 * @entries. find_get_entries() takes a reference against any actual
1243 * The search returns a group of mapping-contiguous page cache entries
1244 * with ascending indexes. There may be holes in the indices due to
1245 * not-present pages.
1247 * Any shadow entries of evicted pages, or swap entries from
1248 * shmem/tmpfs, are included in the returned array.
1250 * find_get_entries() returns the number of pages and shadow entries
1253 unsigned find_get_entries(struct address_space *mapping,
1254 pgoff_t start, unsigned int nr_entries,
1255 struct page **entries, pgoff_t *indices)
1258 unsigned int ret = 0;
1259 struct radix_tree_iter iter;
1265 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) {
1268 page = radix_tree_deref_slot(slot);
1269 if (unlikely(!page))
1271 if (radix_tree_exception(page)) {
1272 if (radix_tree_deref_retry(page)) {
1273 slot = radix_tree_iter_retry(&iter);
1277 * A shadow entry of a recently evicted page, a swap
1278 * entry from shmem/tmpfs or a DAX entry. Return it
1279 * without attempting to raise page count.
1283 if (!page_cache_get_speculative(page))
1286 /* Has the page moved? */
1287 if (unlikely(page != *slot)) {
1292 indices[ret] = iter.index;
1293 entries[ret] = page;
1294 if (++ret == nr_entries)
1302 * find_get_pages - gang pagecache lookup
1303 * @mapping: The address_space to search
1304 * @start: The starting page index
1305 * @nr_pages: The maximum number of pages
1306 * @pages: Where the resulting pages are placed
1308 * find_get_pages() will search for and return a group of up to
1309 * @nr_pages pages in the mapping. The pages are placed at @pages.
1310 * find_get_pages() takes a reference against the returned pages.
1312 * The search returns a group of mapping-contiguous pages with ascending
1313 * indexes. There may be holes in the indices due to not-present pages.
1315 * find_get_pages() returns the number of pages which were found.
1317 unsigned find_get_pages(struct address_space *mapping, pgoff_t start,
1318 unsigned int nr_pages, struct page **pages)
1320 struct radix_tree_iter iter;
1324 if (unlikely(!nr_pages))
1328 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) {
1331 page = radix_tree_deref_slot(slot);
1332 if (unlikely(!page))
1335 if (radix_tree_exception(page)) {
1336 if (radix_tree_deref_retry(page)) {
1337 slot = radix_tree_iter_retry(&iter);
1341 * A shadow entry of a recently evicted page,
1342 * or a swap entry from shmem/tmpfs. Skip
1348 if (!page_cache_get_speculative(page))
1351 /* Has the page moved? */
1352 if (unlikely(page != *slot)) {
1358 if (++ret == nr_pages)
1367 * find_get_pages_contig - gang contiguous pagecache lookup
1368 * @mapping: The address_space to search
1369 * @index: The starting page index
1370 * @nr_pages: The maximum number of pages
1371 * @pages: Where the resulting pages are placed
1373 * find_get_pages_contig() works exactly like find_get_pages(), except
1374 * that the returned number of pages are guaranteed to be contiguous.
1376 * find_get_pages_contig() returns the number of pages which were found.
1378 unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index,
1379 unsigned int nr_pages, struct page **pages)
1381 struct radix_tree_iter iter;
1383 unsigned int ret = 0;
1385 if (unlikely(!nr_pages))
1389 radix_tree_for_each_contig(slot, &mapping->page_tree, &iter, index) {
1392 page = radix_tree_deref_slot(slot);
1393 /* The hole, there no reason to continue */
1394 if (unlikely(!page))
1397 if (radix_tree_exception(page)) {
1398 if (radix_tree_deref_retry(page)) {
1399 slot = radix_tree_iter_retry(&iter);
1403 * A shadow entry of a recently evicted page,
1404 * or a swap entry from shmem/tmpfs. Stop
1405 * looking for contiguous pages.
1410 if (!page_cache_get_speculative(page))
1413 /* Has the page moved? */
1414 if (unlikely(page != *slot)) {
1420 * must check mapping and index after taking the ref.
1421 * otherwise we can get both false positives and false
1422 * negatives, which is just confusing to the caller.
1424 if (page->mapping == NULL || page->index != iter.index) {
1430 if (++ret == nr_pages)
1436 EXPORT_SYMBOL(find_get_pages_contig);
1439 * find_get_pages_tag - find and return pages that match @tag
1440 * @mapping: the address_space to search
1441 * @index: the starting page index
1442 * @tag: the tag index
1443 * @nr_pages: the maximum number of pages
1444 * @pages: where the resulting pages are placed
1446 * Like find_get_pages, except we only return pages which are tagged with
1447 * @tag. We update @index to index the next page for the traversal.
1449 unsigned find_get_pages_tag(struct address_space *mapping, pgoff_t *index,
1450 int tag, unsigned int nr_pages, struct page **pages)
1452 struct radix_tree_iter iter;
1456 if (unlikely(!nr_pages))
1460 radix_tree_for_each_tagged(slot, &mapping->page_tree,
1461 &iter, *index, tag) {
1464 page = radix_tree_deref_slot(slot);
1465 if (unlikely(!page))
1468 if (radix_tree_exception(page)) {
1469 if (radix_tree_deref_retry(page)) {
1470 slot = radix_tree_iter_retry(&iter);
1474 * A shadow entry of a recently evicted page.
1476 * Those entries should never be tagged, but
1477 * this tree walk is lockless and the tags are
1478 * looked up in bulk, one radix tree node at a
1479 * time, so there is a sizable window for page
1480 * reclaim to evict a page we saw tagged.
1487 if (!page_cache_get_speculative(page))
1490 /* Has the page moved? */
1491 if (unlikely(page != *slot)) {
1497 if (++ret == nr_pages)
1504 *index = pages[ret - 1]->index + 1;
1508 EXPORT_SYMBOL(find_get_pages_tag);
1511 * find_get_entries_tag - find and return entries that match @tag
1512 * @mapping: the address_space to search
1513 * @start: the starting page cache index
1514 * @tag: the tag index
1515 * @nr_entries: the maximum number of entries
1516 * @entries: where the resulting entries are placed
1517 * @indices: the cache indices corresponding to the entries in @entries
1519 * Like find_get_entries, except we only return entries which are tagged with
1522 unsigned find_get_entries_tag(struct address_space *mapping, pgoff_t start,
1523 int tag, unsigned int nr_entries,
1524 struct page **entries, pgoff_t *indices)
1527 unsigned int ret = 0;
1528 struct radix_tree_iter iter;
1534 radix_tree_for_each_tagged(slot, &mapping->page_tree,
1535 &iter, start, tag) {
1538 page = radix_tree_deref_slot(slot);
1539 if (unlikely(!page))
1541 if (radix_tree_exception(page)) {
1542 if (radix_tree_deref_retry(page)) {
1543 slot = radix_tree_iter_retry(&iter);
1548 * A shadow entry of a recently evicted page, a swap
1549 * entry from shmem/tmpfs or a DAX entry. Return it
1550 * without attempting to raise page count.
1554 if (!page_cache_get_speculative(page))
1557 /* Has the page moved? */
1558 if (unlikely(page != *slot)) {
1563 indices[ret] = iter.index;
1564 entries[ret] = page;
1565 if (++ret == nr_entries)
1571 EXPORT_SYMBOL(find_get_entries_tag);
1574 * CD/DVDs are error prone. When a medium error occurs, the driver may fail
1575 * a _large_ part of the i/o request. Imagine the worst scenario:
1577 * ---R__________________________________________B__________
1578 * ^ reading here ^ bad block(assume 4k)
1580 * read(R) => miss => readahead(R...B) => media error => frustrating retries
1581 * => failing the whole request => read(R) => read(R+1) =>
1582 * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
1583 * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
1584 * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
1586 * It is going insane. Fix it by quickly scaling down the readahead size.
1588 static void shrink_readahead_size_eio(struct file *filp,
1589 struct file_ra_state *ra)
1595 * do_generic_file_read - generic file read routine
1596 * @filp: the file to read
1597 * @ppos: current file position
1598 * @iter: data destination
1599 * @written: already copied
1601 * This is a generic file read routine, and uses the
1602 * mapping->a_ops->readpage() function for the actual low-level stuff.
1604 * This is really ugly. But the goto's actually try to clarify some
1605 * of the logic when it comes to error handling etc.
1607 static ssize_t do_generic_file_read(struct file *filp, loff_t *ppos,
1608 struct iov_iter *iter, ssize_t written)
1610 struct address_space *mapping = filp->f_mapping;
1611 struct inode *inode = mapping->host;
1612 struct file_ra_state *ra = &filp->f_ra;
1616 unsigned long offset; /* offset into pagecache page */
1617 unsigned int prev_offset;
1620 index = *ppos >> PAGE_SHIFT;
1621 prev_index = ra->prev_pos >> PAGE_SHIFT;
1622 prev_offset = ra->prev_pos & (PAGE_SIZE-1);
1623 last_index = (*ppos + iter->count + PAGE_SIZE-1) >> PAGE_SHIFT;
1624 offset = *ppos & ~PAGE_MASK;
1630 unsigned long nr, ret;
1634 page = find_get_page(mapping, index);
1636 page_cache_sync_readahead(mapping,
1638 index, last_index - index);
1639 page = find_get_page(mapping, index);
1640 if (unlikely(page == NULL))
1641 goto no_cached_page;
1643 if (PageReadahead(page)) {
1644 page_cache_async_readahead(mapping,
1646 index, last_index - index);
1648 if (!PageUptodate(page)) {
1650 * See comment in do_read_cache_page on why
1651 * wait_on_page_locked is used to avoid unnecessarily
1652 * serialisations and why it's safe.
1654 wait_on_page_locked_killable(page);
1655 if (PageUptodate(page))
1658 if (inode->i_blkbits == PAGE_SHIFT ||
1659 !mapping->a_ops->is_partially_uptodate)
1660 goto page_not_up_to_date;
1661 if (!trylock_page(page))
1662 goto page_not_up_to_date;
1663 /* Did it get truncated before we got the lock? */
1665 goto page_not_up_to_date_locked;
1666 if (!mapping->a_ops->is_partially_uptodate(page,
1667 offset, iter->count))
1668 goto page_not_up_to_date_locked;
1673 * i_size must be checked after we know the page is Uptodate.
1675 * Checking i_size after the check allows us to calculate
1676 * the correct value for "nr", which means the zero-filled
1677 * part of the page is not copied back to userspace (unless
1678 * another truncate extends the file - this is desired though).
1681 isize = i_size_read(inode);
1682 end_index = (isize - 1) >> PAGE_SHIFT;
1683 if (unlikely(!isize || index > end_index)) {
1688 /* nr is the maximum number of bytes to copy from this page */
1690 if (index == end_index) {
1691 nr = ((isize - 1) & ~PAGE_MASK) + 1;
1699 /* If users can be writing to this page using arbitrary
1700 * virtual addresses, take care about potential aliasing
1701 * before reading the page on the kernel side.
1703 if (mapping_writably_mapped(mapping))
1704 flush_dcache_page(page);
1707 * When a sequential read accesses a page several times,
1708 * only mark it as accessed the first time.
1710 if (prev_index != index || offset != prev_offset)
1711 mark_page_accessed(page);
1715 * Ok, we have the page, and it's up-to-date, so
1716 * now we can copy it to user space...
1719 ret = copy_page_to_iter(page, offset, nr, iter);
1721 index += offset >> PAGE_SHIFT;
1722 offset &= ~PAGE_MASK;
1723 prev_offset = offset;
1727 if (!iov_iter_count(iter))
1735 page_not_up_to_date:
1736 /* Get exclusive access to the page ... */
1737 error = lock_page_killable(page);
1738 if (unlikely(error))
1739 goto readpage_error;
1741 page_not_up_to_date_locked:
1742 /* Did it get truncated before we got the lock? */
1743 if (!page->mapping) {
1749 /* Did somebody else fill it already? */
1750 if (PageUptodate(page)) {
1757 * A previous I/O error may have been due to temporary
1758 * failures, eg. multipath errors.
1759 * PG_error will be set again if readpage fails.
1761 ClearPageError(page);
1762 /* Start the actual read. The read will unlock the page. */
1763 error = mapping->a_ops->readpage(filp, page);
1765 if (unlikely(error)) {
1766 if (error == AOP_TRUNCATED_PAGE) {
1771 goto readpage_error;
1774 if (!PageUptodate(page)) {
1775 error = lock_page_killable(page);
1776 if (unlikely(error))
1777 goto readpage_error;
1778 if (!PageUptodate(page)) {
1779 if (page->mapping == NULL) {
1781 * invalidate_mapping_pages got it
1788 shrink_readahead_size_eio(filp, ra);
1790 goto readpage_error;
1798 /* UHHUH! A synchronous read error occurred. Report it */
1804 * Ok, it wasn't cached, so we need to create a new
1807 page = page_cache_alloc_cold(mapping);
1812 error = add_to_page_cache_lru(page, mapping, index,
1813 mapping_gfp_constraint(mapping, GFP_KERNEL));
1816 if (error == -EEXIST) {
1826 ra->prev_pos = prev_index;
1827 ra->prev_pos <<= PAGE_SHIFT;
1828 ra->prev_pos |= prev_offset;
1830 *ppos = ((loff_t)index << PAGE_SHIFT) + offset;
1831 file_accessed(filp);
1832 return written ? written : error;
1836 * generic_file_read_iter - generic filesystem read routine
1837 * @iocb: kernel I/O control block
1838 * @iter: destination for the data read
1840 * This is the "read_iter()" routine for all filesystems
1841 * that can use the page cache directly.
1844 generic_file_read_iter(struct kiocb *iocb, struct iov_iter *iter)
1846 struct file *file = iocb->ki_filp;
1848 loff_t *ppos = &iocb->ki_pos;
1850 size_t count = iov_iter_count(iter);
1853 goto out; /* skip atime */
1855 if (iocb->ki_flags & IOCB_DIRECT) {
1856 struct address_space *mapping = file->f_mapping;
1857 struct inode *inode = mapping->host;
1860 size = i_size_read(inode);
1861 retval = filemap_write_and_wait_range(mapping, pos,
1864 struct iov_iter data = *iter;
1865 retval = mapping->a_ops->direct_IO(iocb, &data, pos);
1869 *ppos = pos + retval;
1870 iov_iter_advance(iter, retval);
1874 * Btrfs can have a short DIO read if we encounter
1875 * compressed extents, so if there was an error, or if
1876 * we've already read everything we wanted to, or if
1877 * there was a short read because we hit EOF, go ahead
1878 * and return. Otherwise fallthrough to buffered io for
1879 * the rest of the read. Buffered reads will not work for
1880 * DAX files, so don't bother trying.
1882 if (retval < 0 || !iov_iter_count(iter) || *ppos >= size ||
1884 file_accessed(file);
1889 retval = do_generic_file_read(file, ppos, iter, retval);
1893 EXPORT_SYMBOL(generic_file_read_iter);
1897 * page_cache_read - adds requested page to the page cache if not already there
1898 * @file: file to read
1899 * @offset: page index
1900 * @gfp_mask: memory allocation flags
1902 * This adds the requested page to the page cache if it isn't already there,
1903 * and schedules an I/O to read in its contents from disk.
1905 static int page_cache_read(struct file *file, pgoff_t offset, gfp_t gfp_mask)
1907 struct address_space *mapping = file->f_mapping;
1912 page = __page_cache_alloc(gfp_mask|__GFP_COLD);
1916 ret = add_to_page_cache_lru(page, mapping, offset, gfp_mask & GFP_KERNEL);
1918 ret = mapping->a_ops->readpage(file, page);
1919 else if (ret == -EEXIST)
1920 ret = 0; /* losing race to add is OK */
1924 } while (ret == AOP_TRUNCATED_PAGE);
1929 #define MMAP_LOTSAMISS (100)
1932 * Synchronous readahead happens when we don't even find
1933 * a page in the page cache at all.
1935 static void do_sync_mmap_readahead(struct vm_area_struct *vma,
1936 struct file_ra_state *ra,
1940 struct address_space *mapping = file->f_mapping;
1942 /* If we don't want any read-ahead, don't bother */
1943 if (vma->vm_flags & VM_RAND_READ)
1948 if (vma->vm_flags & VM_SEQ_READ) {
1949 page_cache_sync_readahead(mapping, ra, file, offset,
1954 /* Avoid banging the cache line if not needed */
1955 if (ra->mmap_miss < MMAP_LOTSAMISS * 10)
1959 * Do we miss much more than hit in this file? If so,
1960 * stop bothering with read-ahead. It will only hurt.
1962 if (ra->mmap_miss > MMAP_LOTSAMISS)
1968 ra->start = max_t(long, 0, offset - ra->ra_pages / 2);
1969 ra->size = ra->ra_pages;
1970 ra->async_size = ra->ra_pages / 4;
1971 ra_submit(ra, mapping, file);
1975 * Asynchronous readahead happens when we find the page and PG_readahead,
1976 * so we want to possibly extend the readahead further..
1978 static void do_async_mmap_readahead(struct vm_area_struct *vma,
1979 struct file_ra_state *ra,
1984 struct address_space *mapping = file->f_mapping;
1986 /* If we don't want any read-ahead, don't bother */
1987 if (vma->vm_flags & VM_RAND_READ)
1989 if (ra->mmap_miss > 0)
1991 if (PageReadahead(page))
1992 page_cache_async_readahead(mapping, ra, file,
1993 page, offset, ra->ra_pages);
1997 * filemap_fault - read in file data for page fault handling
1998 * @vma: vma in which the fault was taken
1999 * @vmf: struct vm_fault containing details of the fault
2001 * filemap_fault() is invoked via the vma operations vector for a
2002 * mapped memory region to read in file data during a page fault.
2004 * The goto's are kind of ugly, but this streamlines the normal case of having
2005 * it in the page cache, and handles the special cases reasonably without
2006 * having a lot of duplicated code.
2008 * vma->vm_mm->mmap_sem must be held on entry.
2010 * If our return value has VM_FAULT_RETRY set, it's because
2011 * lock_page_or_retry() returned 0.
2012 * The mmap_sem has usually been released in this case.
2013 * See __lock_page_or_retry() for the exception.
2015 * If our return value does not have VM_FAULT_RETRY set, the mmap_sem
2016 * has not been released.
2018 * We never return with VM_FAULT_RETRY and a bit from VM_FAULT_ERROR set.
2020 int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
2023 struct file *file = vma->vm_file;
2024 struct address_space *mapping = file->f_mapping;
2025 struct file_ra_state *ra = &file->f_ra;
2026 struct inode *inode = mapping->host;
2027 pgoff_t offset = vmf->pgoff;
2032 size = round_up(i_size_read(inode), PAGE_SIZE);
2033 if (offset >= size >> PAGE_SHIFT)
2034 return VM_FAULT_SIGBUS;
2037 * Do we have something in the page cache already?
2039 page = find_get_page(mapping, offset);
2040 if (likely(page) && !(vmf->flags & FAULT_FLAG_TRIED)) {
2042 * We found the page, so try async readahead before
2043 * waiting for the lock.
2045 do_async_mmap_readahead(vma, ra, file, page, offset);
2047 /* No page in the page cache at all */
2048 do_sync_mmap_readahead(vma, ra, file, offset);
2049 count_vm_event(PGMAJFAULT);
2050 mem_cgroup_count_vm_event(vma->vm_mm, PGMAJFAULT);
2051 ret = VM_FAULT_MAJOR;
2053 page = find_get_page(mapping, offset);
2055 goto no_cached_page;
2058 if (!lock_page_or_retry(page, vma->vm_mm, vmf->flags)) {
2060 return ret | VM_FAULT_RETRY;
2063 /* Did it get truncated? */
2064 if (unlikely(page->mapping != mapping)) {
2069 VM_BUG_ON_PAGE(page->index != offset, page);
2072 * We have a locked page in the page cache, now we need to check
2073 * that it's up-to-date. If not, it is going to be due to an error.
2075 if (unlikely(!PageUptodate(page)))
2076 goto page_not_uptodate;
2079 * Found the page and have a reference on it.
2080 * We must recheck i_size under page lock.
2082 size = round_up(i_size_read(inode), PAGE_SIZE);
2083 if (unlikely(offset >= size >> PAGE_SHIFT)) {
2086 return VM_FAULT_SIGBUS;
2090 return ret | VM_FAULT_LOCKED;
2094 * We're only likely to ever get here if MADV_RANDOM is in
2097 error = page_cache_read(file, offset, vmf->gfp_mask);
2100 * The page we want has now been added to the page cache.
2101 * In the unlikely event that someone removed it in the
2102 * meantime, we'll just come back here and read it again.
2108 * An error return from page_cache_read can result if the
2109 * system is low on memory, or a problem occurs while trying
2112 if (error == -ENOMEM)
2113 return VM_FAULT_OOM;
2114 return VM_FAULT_SIGBUS;
2118 * Umm, take care of errors if the page isn't up-to-date.
2119 * Try to re-read it _once_. We do this synchronously,
2120 * because there really aren't any performance issues here
2121 * and we need to check for errors.
2123 ClearPageError(page);
2124 error = mapping->a_ops->readpage(file, page);
2126 wait_on_page_locked(page);
2127 if (!PageUptodate(page))
2132 if (!error || error == AOP_TRUNCATED_PAGE)
2135 /* Things didn't work out. Return zero to tell the mm layer so. */
2136 shrink_readahead_size_eio(file, ra);
2137 return VM_FAULT_SIGBUS;
2139 EXPORT_SYMBOL(filemap_fault);
2141 void filemap_map_pages(struct vm_area_struct *vma, struct vm_fault *vmf)
2143 struct radix_tree_iter iter;
2145 struct file *file = vma->vm_file;
2146 struct address_space *mapping = file->f_mapping;
2149 unsigned long address = (unsigned long) vmf->virtual_address;
2154 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, vmf->pgoff) {
2155 if (iter.index > vmf->max_pgoff)
2158 page = radix_tree_deref_slot(slot);
2159 if (unlikely(!page))
2161 if (radix_tree_exception(page)) {
2162 if (radix_tree_deref_retry(page)) {
2163 slot = radix_tree_iter_retry(&iter);
2169 if (!page_cache_get_speculative(page))
2172 /* Has the page moved? */
2173 if (unlikely(page != *slot)) {
2178 if (!PageUptodate(page) ||
2179 PageReadahead(page) ||
2182 if (!trylock_page(page))
2185 if (page->mapping != mapping || !PageUptodate(page))
2188 size = round_up(i_size_read(mapping->host), PAGE_SIZE);
2189 if (page->index >= size >> PAGE_SHIFT)
2192 pte = vmf->pte + page->index - vmf->pgoff;
2193 if (!pte_none(*pte))
2196 if (file->f_ra.mmap_miss > 0)
2197 file->f_ra.mmap_miss--;
2198 addr = address + (page->index - vmf->pgoff) * PAGE_SIZE;
2199 do_set_pte(vma, addr, page, pte, false, false);
2207 if (iter.index == vmf->max_pgoff)
2212 EXPORT_SYMBOL(filemap_map_pages);
2214 int filemap_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
2216 struct page *page = vmf->page;
2217 struct inode *inode = file_inode(vma->vm_file);
2218 int ret = VM_FAULT_LOCKED;
2220 sb_start_pagefault(inode->i_sb);
2221 file_update_time(vma->vm_file);
2223 if (page->mapping != inode->i_mapping) {
2225 ret = VM_FAULT_NOPAGE;
2229 * We mark the page dirty already here so that when freeze is in
2230 * progress, we are guaranteed that writeback during freezing will
2231 * see the dirty page and writeprotect it again.
2233 set_page_dirty(page);
2234 wait_for_stable_page(page);
2236 sb_end_pagefault(inode->i_sb);
2239 EXPORT_SYMBOL(filemap_page_mkwrite);
2241 const struct vm_operations_struct generic_file_vm_ops = {
2242 .fault = filemap_fault,
2243 .map_pages = filemap_map_pages,
2244 .page_mkwrite = filemap_page_mkwrite,
2247 /* This is used for a general mmap of a disk file */
2249 int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
2251 struct address_space *mapping = file->f_mapping;
2253 if (!mapping->a_ops->readpage)
2255 file_accessed(file);
2256 vma->vm_ops = &generic_file_vm_ops;
2261 * This is for filesystems which do not implement ->writepage.
2263 int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
2265 if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE))
2267 return generic_file_mmap(file, vma);
2270 int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
2274 int generic_file_readonly_mmap(struct file * file, struct vm_area_struct * vma)
2278 #endif /* CONFIG_MMU */
2280 EXPORT_SYMBOL(generic_file_mmap);
2281 EXPORT_SYMBOL(generic_file_readonly_mmap);
2283 static struct page *wait_on_page_read(struct page *page)
2285 if (!IS_ERR(page)) {
2286 wait_on_page_locked(page);
2287 if (!PageUptodate(page)) {
2289 page = ERR_PTR(-EIO);
2295 static struct page *do_read_cache_page(struct address_space *mapping,
2297 int (*filler)(void *, struct page *),
2304 page = find_get_page(mapping, index);
2306 page = __page_cache_alloc(gfp | __GFP_COLD);
2308 return ERR_PTR(-ENOMEM);
2309 err = add_to_page_cache_lru(page, mapping, index, gfp);
2310 if (unlikely(err)) {
2314 /* Presumably ENOMEM for radix tree node */
2315 return ERR_PTR(err);
2319 err = filler(data, page);
2322 return ERR_PTR(err);
2325 page = wait_on_page_read(page);
2330 if (PageUptodate(page))
2334 * Page is not up to date and may be locked due one of the following
2335 * case a: Page is being filled and the page lock is held
2336 * case b: Read/write error clearing the page uptodate status
2337 * case c: Truncation in progress (page locked)
2338 * case d: Reclaim in progress
2340 * Case a, the page will be up to date when the page is unlocked.
2341 * There is no need to serialise on the page lock here as the page
2342 * is pinned so the lock gives no additional protection. Even if the
2343 * the page is truncated, the data is still valid if PageUptodate as
2344 * it's a race vs truncate race.
2345 * Case b, the page will not be up to date
2346 * Case c, the page may be truncated but in itself, the data may still
2347 * be valid after IO completes as it's a read vs truncate race. The
2348 * operation must restart if the page is not uptodate on unlock but
2349 * otherwise serialising on page lock to stabilise the mapping gives
2350 * no additional guarantees to the caller as the page lock is
2351 * released before return.
2352 * Case d, similar to truncation. If reclaim holds the page lock, it
2353 * will be a race with remove_mapping that determines if the mapping
2354 * is valid on unlock but otherwise the data is valid and there is
2355 * no need to serialise with page lock.
2357 * As the page lock gives no additional guarantee, we optimistically
2358 * wait on the page to be unlocked and check if it's up to date and
2359 * use the page if it is. Otherwise, the page lock is required to
2360 * distinguish between the different cases. The motivation is that we
2361 * avoid spurious serialisations and wakeups when multiple processes
2362 * wait on the same page for IO to complete.
2364 wait_on_page_locked(page);
2365 if (PageUptodate(page))
2368 /* Distinguish between all the cases under the safety of the lock */
2371 /* Case c or d, restart the operation */
2372 if (!page->mapping) {
2378 /* Someone else locked and filled the page in a very small window */
2379 if (PageUptodate(page)) {
2386 mark_page_accessed(page);
2391 * read_cache_page - read into page cache, fill it if needed
2392 * @mapping: the page's address_space
2393 * @index: the page index
2394 * @filler: function to perform the read
2395 * @data: first arg to filler(data, page) function, often left as NULL
2397 * Read into the page cache. If a page already exists, and PageUptodate() is
2398 * not set, try to fill the page and wait for it to become unlocked.
2400 * If the page does not get brought uptodate, return -EIO.
2402 struct page *read_cache_page(struct address_space *mapping,
2404 int (*filler)(void *, struct page *),
2407 return do_read_cache_page(mapping, index, filler, data, mapping_gfp_mask(mapping));
2409 EXPORT_SYMBOL(read_cache_page);
2412 * read_cache_page_gfp - read into page cache, using specified page allocation flags.
2413 * @mapping: the page's address_space
2414 * @index: the page index
2415 * @gfp: the page allocator flags to use if allocating
2417 * This is the same as "read_mapping_page(mapping, index, NULL)", but with
2418 * any new page allocations done using the specified allocation flags.
2420 * If the page does not get brought uptodate, return -EIO.
2422 struct page *read_cache_page_gfp(struct address_space *mapping,
2426 filler_t *filler = (filler_t *)mapping->a_ops->readpage;
2428 return do_read_cache_page(mapping, index, filler, NULL, gfp);
2430 EXPORT_SYMBOL(read_cache_page_gfp);
2433 * Performs necessary checks before doing a write
2435 * Can adjust writing position or amount of bytes to write.
2436 * Returns appropriate error code that caller should return or
2437 * zero in case that write should be allowed.
2439 inline ssize_t generic_write_checks(struct kiocb *iocb, struct iov_iter *from)
2441 struct file *file = iocb->ki_filp;
2442 struct inode *inode = file->f_mapping->host;
2443 unsigned long limit = rlimit(RLIMIT_FSIZE);
2446 if (!iov_iter_count(from))
2449 /* FIXME: this is for backwards compatibility with 2.4 */
2450 if (iocb->ki_flags & IOCB_APPEND)
2451 iocb->ki_pos = i_size_read(inode);
2455 if (limit != RLIM_INFINITY) {
2456 if (iocb->ki_pos >= limit) {
2457 send_sig(SIGXFSZ, current, 0);
2460 iov_iter_truncate(from, limit - (unsigned long)pos);
2466 if (unlikely(pos + iov_iter_count(from) > MAX_NON_LFS &&
2467 !(file->f_flags & O_LARGEFILE))) {
2468 if (pos >= MAX_NON_LFS)
2470 iov_iter_truncate(from, MAX_NON_LFS - (unsigned long)pos);
2474 * Are we about to exceed the fs block limit ?
2476 * If we have written data it becomes a short write. If we have
2477 * exceeded without writing data we send a signal and return EFBIG.
2478 * Linus frestrict idea will clean these up nicely..
2480 if (unlikely(pos >= inode->i_sb->s_maxbytes))
2483 iov_iter_truncate(from, inode->i_sb->s_maxbytes - pos);
2484 return iov_iter_count(from);
2486 EXPORT_SYMBOL(generic_write_checks);
2488 int pagecache_write_begin(struct file *file, struct address_space *mapping,
2489 loff_t pos, unsigned len, unsigned flags,
2490 struct page **pagep, void **fsdata)
2492 const struct address_space_operations *aops = mapping->a_ops;
2494 return aops->write_begin(file, mapping, pos, len, flags,
2497 EXPORT_SYMBOL(pagecache_write_begin);
2499 int pagecache_write_end(struct file *file, struct address_space *mapping,
2500 loff_t pos, unsigned len, unsigned copied,
2501 struct page *page, void *fsdata)
2503 const struct address_space_operations *aops = mapping->a_ops;
2505 return aops->write_end(file, mapping, pos, len, copied, page, fsdata);
2507 EXPORT_SYMBOL(pagecache_write_end);
2510 generic_file_direct_write(struct kiocb *iocb, struct iov_iter *from, loff_t pos)
2512 struct file *file = iocb->ki_filp;
2513 struct address_space *mapping = file->f_mapping;
2514 struct inode *inode = mapping->host;
2518 struct iov_iter data;
2520 write_len = iov_iter_count(from);
2521 end = (pos + write_len - 1) >> PAGE_SHIFT;
2523 written = filemap_write_and_wait_range(mapping, pos, pos + write_len - 1);
2528 * After a write we want buffered reads to be sure to go to disk to get
2529 * the new data. We invalidate clean cached page from the region we're
2530 * about to write. We do this *before* the write so that we can return
2531 * without clobbering -EIOCBQUEUED from ->direct_IO().
2533 if (mapping->nrpages) {
2534 written = invalidate_inode_pages2_range(mapping,
2535 pos >> PAGE_SHIFT, end);
2537 * If a page can not be invalidated, return 0 to fall back
2538 * to buffered write.
2541 if (written == -EBUSY)
2548 written = mapping->a_ops->direct_IO(iocb, &data, pos);
2551 * Finally, try again to invalidate clean pages which might have been
2552 * cached by non-direct readahead, or faulted in by get_user_pages()
2553 * if the source of the write was an mmap'ed region of the file
2554 * we're writing. Either one is a pretty crazy thing to do,
2555 * so we don't support it 100%. If this invalidation
2556 * fails, tough, the write still worked...
2558 if (mapping->nrpages) {
2559 invalidate_inode_pages2_range(mapping,
2560 pos >> PAGE_SHIFT, end);
2565 iov_iter_advance(from, written);
2566 if (pos > i_size_read(inode) && !S_ISBLK(inode->i_mode)) {
2567 i_size_write(inode, pos);
2568 mark_inode_dirty(inode);
2575 EXPORT_SYMBOL(generic_file_direct_write);
2578 * Find or create a page at the given pagecache position. Return the locked
2579 * page. This function is specifically for buffered writes.
2581 struct page *grab_cache_page_write_begin(struct address_space *mapping,
2582 pgoff_t index, unsigned flags)
2585 int fgp_flags = FGP_LOCK|FGP_ACCESSED|FGP_WRITE|FGP_CREAT;
2587 if (flags & AOP_FLAG_NOFS)
2588 fgp_flags |= FGP_NOFS;
2590 page = pagecache_get_page(mapping, index, fgp_flags,
2591 mapping_gfp_mask(mapping));
2593 wait_for_stable_page(page);
2597 EXPORT_SYMBOL(grab_cache_page_write_begin);
2599 ssize_t generic_perform_write(struct file *file,
2600 struct iov_iter *i, loff_t pos)
2602 struct address_space *mapping = file->f_mapping;
2603 const struct address_space_operations *a_ops = mapping->a_ops;
2605 ssize_t written = 0;
2606 unsigned int flags = 0;
2609 * Copies from kernel address space cannot fail (NFSD is a big user).
2611 if (!iter_is_iovec(i))
2612 flags |= AOP_FLAG_UNINTERRUPTIBLE;
2616 unsigned long offset; /* Offset into pagecache page */
2617 unsigned long bytes; /* Bytes to write to page */
2618 size_t copied; /* Bytes copied from user */
2621 offset = (pos & (PAGE_SIZE - 1));
2622 bytes = min_t(unsigned long, PAGE_SIZE - offset,
2627 * Bring in the user page that we will copy from _first_.
2628 * Otherwise there's a nasty deadlock on copying from the
2629 * same page as we're writing to, without it being marked
2632 * Not only is this an optimisation, but it is also required
2633 * to check that the address is actually valid, when atomic
2634 * usercopies are used, below.
2636 if (unlikely(iov_iter_fault_in_readable(i, bytes))) {
2641 if (fatal_signal_pending(current)) {
2646 status = a_ops->write_begin(file, mapping, pos, bytes, flags,
2648 if (unlikely(status < 0))
2651 if (mapping_writably_mapped(mapping))
2652 flush_dcache_page(page);
2654 copied = iov_iter_copy_from_user_atomic(page, i, offset, bytes);
2655 flush_dcache_page(page);
2657 status = a_ops->write_end(file, mapping, pos, bytes, copied,
2659 if (unlikely(status < 0))
2665 iov_iter_advance(i, copied);
2666 if (unlikely(copied == 0)) {
2668 * If we were unable to copy any data at all, we must
2669 * fall back to a single segment length write.
2671 * If we didn't fallback here, we could livelock
2672 * because not all segments in the iov can be copied at
2673 * once without a pagefault.
2675 bytes = min_t(unsigned long, PAGE_SIZE - offset,
2676 iov_iter_single_seg_count(i));
2682 balance_dirty_pages_ratelimited(mapping);
2683 } while (iov_iter_count(i));
2685 return written ? written : status;
2687 EXPORT_SYMBOL(generic_perform_write);
2690 * __generic_file_write_iter - write data to a file
2691 * @iocb: IO state structure (file, offset, etc.)
2692 * @from: iov_iter with data to write
2694 * This function does all the work needed for actually writing data to a
2695 * file. It does all basic checks, removes SUID from the file, updates
2696 * modification times and calls proper subroutines depending on whether we
2697 * do direct IO or a standard buffered write.
2699 * It expects i_mutex to be grabbed unless we work on a block device or similar
2700 * object which does not need locking at all.
2702 * This function does *not* take care of syncing data in case of O_SYNC write.
2703 * A caller has to handle it. This is mainly due to the fact that we want to
2704 * avoid syncing under i_mutex.
2706 ssize_t __generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
2708 struct file *file = iocb->ki_filp;
2709 struct address_space * mapping = file->f_mapping;
2710 struct inode *inode = mapping->host;
2711 ssize_t written = 0;
2715 /* We can write back this queue in page reclaim */
2716 current->backing_dev_info = inode_to_bdi(inode);
2717 err = file_remove_privs(file);
2721 err = file_update_time(file);
2725 if (iocb->ki_flags & IOCB_DIRECT) {
2726 loff_t pos, endbyte;
2728 written = generic_file_direct_write(iocb, from, iocb->ki_pos);
2730 * If the write stopped short of completing, fall back to
2731 * buffered writes. Some filesystems do this for writes to
2732 * holes, for example. For DAX files, a buffered write will
2733 * not succeed (even if it did, DAX does not handle dirty
2734 * page-cache pages correctly).
2736 if (written < 0 || !iov_iter_count(from) || IS_DAX(inode))
2739 status = generic_perform_write(file, from, pos = iocb->ki_pos);
2741 * If generic_perform_write() returned a synchronous error
2742 * then we want to return the number of bytes which were
2743 * direct-written, or the error code if that was zero. Note
2744 * that this differs from normal direct-io semantics, which
2745 * will return -EFOO even if some bytes were written.
2747 if (unlikely(status < 0)) {
2752 * We need to ensure that the page cache pages are written to
2753 * disk and invalidated to preserve the expected O_DIRECT
2756 endbyte = pos + status - 1;
2757 err = filemap_write_and_wait_range(mapping, pos, endbyte);
2759 iocb->ki_pos = endbyte + 1;
2761 invalidate_mapping_pages(mapping,
2763 endbyte >> PAGE_SHIFT);
2766 * We don't know how much we wrote, so just return
2767 * the number of bytes which were direct-written
2771 written = generic_perform_write(file, from, iocb->ki_pos);
2772 if (likely(written > 0))
2773 iocb->ki_pos += written;
2776 current->backing_dev_info = NULL;
2777 return written ? written : err;
2779 EXPORT_SYMBOL(__generic_file_write_iter);
2782 * generic_file_write_iter - write data to a file
2783 * @iocb: IO state structure
2784 * @from: iov_iter with data to write
2786 * This is a wrapper around __generic_file_write_iter() to be used by most
2787 * filesystems. It takes care of syncing the file in case of O_SYNC file
2788 * and acquires i_mutex as needed.
2790 ssize_t generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
2792 struct file *file = iocb->ki_filp;
2793 struct inode *inode = file->f_mapping->host;
2797 ret = generic_write_checks(iocb, from);
2799 ret = __generic_file_write_iter(iocb, from);
2800 inode_unlock(inode);
2805 err = generic_write_sync(file, iocb->ki_pos - ret, ret);
2811 EXPORT_SYMBOL(generic_file_write_iter);
2814 * try_to_release_page() - release old fs-specific metadata on a page
2816 * @page: the page which the kernel is trying to free
2817 * @gfp_mask: memory allocation flags (and I/O mode)
2819 * The address_space is to try to release any data against the page
2820 * (presumably at page->private). If the release was successful, return `1'.
2821 * Otherwise return zero.
2823 * This may also be called if PG_fscache is set on a page, indicating that the
2824 * page is known to the local caching routines.
2826 * The @gfp_mask argument specifies whether I/O may be performed to release
2827 * this page (__GFP_IO), and whether the call may block (__GFP_RECLAIM & __GFP_FS).
2830 int try_to_release_page(struct page *page, gfp_t gfp_mask)
2832 struct address_space * const mapping = page->mapping;
2834 BUG_ON(!PageLocked(page));
2835 if (PageWriteback(page))
2838 if (mapping && mapping->a_ops->releasepage)
2839 return mapping->a_ops->releasepage(page, gfp_mask);
2840 return try_to_free_buffers(page);
2843 EXPORT_SYMBOL(try_to_release_page);