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(zone) (follow_page->mark_page_accessed)
99 * ->zone_lru_lock(zone) (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 int page_cache_tree_insert(struct address_space *mapping,
114 struct page *page, void **shadowp)
116 struct radix_tree_node *node;
120 error = __radix_tree_create(&mapping->page_tree, page->index, 0,
127 p = radix_tree_deref_slot_protected(slot, &mapping->tree_lock);
128 if (!radix_tree_exceptional_entry(p))
131 mapping->nrexceptional--;
132 if (!dax_mapping(mapping)) {
136 /* DAX can replace empty locked entry with a hole */
138 dax_radix_locked_entry(0, RADIX_DAX_EMPTY));
139 /* Wakeup waiters for exceptional entry lock */
140 dax_wake_mapping_entry_waiter(mapping, page->index, p,
144 __radix_tree_replace(&mapping->page_tree, node, slot, page,
145 workingset_update_node, mapping);
150 static void page_cache_tree_delete(struct address_space *mapping,
151 struct page *page, void *shadow)
155 /* hugetlb pages are represented by one entry in the radix tree */
156 nr = PageHuge(page) ? 1 : hpage_nr_pages(page);
158 VM_BUG_ON_PAGE(!PageLocked(page), page);
159 VM_BUG_ON_PAGE(PageTail(page), page);
160 VM_BUG_ON_PAGE(nr != 1 && shadow, page);
162 for (i = 0; i < nr; i++) {
163 struct radix_tree_node *node;
166 __radix_tree_lookup(&mapping->page_tree, page->index + i,
169 VM_BUG_ON_PAGE(!node && nr != 1, page);
171 radix_tree_clear_tags(&mapping->page_tree, node, slot);
172 __radix_tree_replace(&mapping->page_tree, node, slot, shadow,
173 workingset_update_node, mapping);
177 mapping->nrexceptional += nr;
179 * Make sure the nrexceptional update is committed before
180 * the nrpages update so that final truncate racing
181 * with reclaim does not see both counters 0 at the
182 * same time and miss a shadow entry.
186 mapping->nrpages -= nr;
190 * Delete a page from the page cache and free it. Caller has to make
191 * sure the page is locked and that nobody else uses it - or that usage
192 * is safe. The caller must hold the mapping's tree_lock.
194 void __delete_from_page_cache(struct page *page, void *shadow)
196 struct address_space *mapping = page->mapping;
197 int nr = hpage_nr_pages(page);
199 trace_mm_filemap_delete_from_page_cache(page);
201 * if we're uptodate, flush out into the cleancache, otherwise
202 * invalidate any existing cleancache entries. We can't leave
203 * stale data around in the cleancache once our page is gone
205 if (PageUptodate(page) && PageMappedToDisk(page))
206 cleancache_put_page(page);
208 cleancache_invalidate_page(mapping, page);
210 VM_BUG_ON_PAGE(PageTail(page), page);
211 VM_BUG_ON_PAGE(page_mapped(page), page);
212 if (!IS_ENABLED(CONFIG_DEBUG_VM) && unlikely(page_mapped(page))) {
215 pr_alert("BUG: Bad page cache in process %s pfn:%05lx\n",
216 current->comm, page_to_pfn(page));
217 dump_page(page, "still mapped when deleted");
219 add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE);
221 mapcount = page_mapcount(page);
222 if (mapping_exiting(mapping) &&
223 page_count(page) >= mapcount + 2) {
225 * All vmas have already been torn down, so it's
226 * a good bet that actually the page is unmapped,
227 * and we'd prefer not to leak it: if we're wrong,
228 * some other bad page check should catch it later.
230 page_mapcount_reset(page);
231 page_ref_sub(page, mapcount);
235 page_cache_tree_delete(mapping, page, shadow);
237 page->mapping = NULL;
238 /* Leave page->index set: truncation lookup relies upon it */
240 /* hugetlb pages do not participate in page cache accounting. */
242 __mod_node_page_state(page_pgdat(page), NR_FILE_PAGES, -nr);
243 if (PageSwapBacked(page)) {
244 __mod_node_page_state(page_pgdat(page), NR_SHMEM, -nr);
245 if (PageTransHuge(page))
246 __dec_node_page_state(page, NR_SHMEM_THPS);
248 VM_BUG_ON_PAGE(PageTransHuge(page) && !PageHuge(page), page);
252 * At this point page must be either written or cleaned by truncate.
253 * Dirty page here signals a bug and loss of unwritten data.
255 * This fixes dirty accounting after removing the page entirely but
256 * leaves PageDirty set: it has no effect for truncated page and
257 * anyway will be cleared before returning page into buddy allocator.
259 if (WARN_ON_ONCE(PageDirty(page)))
260 account_page_cleaned(page, mapping, inode_to_wb(mapping->host));
264 * delete_from_page_cache - delete page from page cache
265 * @page: the page which the kernel is trying to remove from page cache
267 * This must be called only on pages that have been verified to be in the page
268 * cache and locked. It will never put the page into the free list, the caller
269 * has a reference on the page.
271 void delete_from_page_cache(struct page *page)
273 struct address_space *mapping = page_mapping(page);
275 void (*freepage)(struct page *);
277 BUG_ON(!PageLocked(page));
279 freepage = mapping->a_ops->freepage;
281 spin_lock_irqsave(&mapping->tree_lock, flags);
282 __delete_from_page_cache(page, NULL);
283 spin_unlock_irqrestore(&mapping->tree_lock, flags);
288 if (PageTransHuge(page) && !PageHuge(page)) {
289 page_ref_sub(page, HPAGE_PMD_NR);
290 VM_BUG_ON_PAGE(page_count(page) <= 0, page);
295 EXPORT_SYMBOL(delete_from_page_cache);
297 int filemap_check_errors(struct address_space *mapping)
300 /* Check for outstanding write errors */
301 if (test_bit(AS_ENOSPC, &mapping->flags) &&
302 test_and_clear_bit(AS_ENOSPC, &mapping->flags))
304 if (test_bit(AS_EIO, &mapping->flags) &&
305 test_and_clear_bit(AS_EIO, &mapping->flags))
309 EXPORT_SYMBOL(filemap_check_errors);
312 * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range
313 * @mapping: address space structure to write
314 * @start: offset in bytes where the range starts
315 * @end: offset in bytes where the range ends (inclusive)
316 * @sync_mode: enable synchronous operation
318 * Start writeback against all of a mapping's dirty pages that lie
319 * within the byte offsets <start, end> inclusive.
321 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
322 * opposed to a regular memory cleansing writeback. The difference between
323 * these two operations is that if a dirty page/buffer is encountered, it must
324 * be waited upon, and not just skipped over.
326 int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
327 loff_t end, int sync_mode)
330 struct writeback_control wbc = {
331 .sync_mode = sync_mode,
332 .nr_to_write = LONG_MAX,
333 .range_start = start,
337 if (!mapping_cap_writeback_dirty(mapping))
340 wbc_attach_fdatawrite_inode(&wbc, mapping->host);
341 ret = do_writepages(mapping, &wbc);
342 wbc_detach_inode(&wbc);
346 static inline int __filemap_fdatawrite(struct address_space *mapping,
349 return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode);
352 int filemap_fdatawrite(struct address_space *mapping)
354 return __filemap_fdatawrite(mapping, WB_SYNC_ALL);
356 EXPORT_SYMBOL(filemap_fdatawrite);
358 int filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
361 return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL);
363 EXPORT_SYMBOL(filemap_fdatawrite_range);
366 * filemap_flush - mostly a non-blocking flush
367 * @mapping: target address_space
369 * This is a mostly non-blocking flush. Not suitable for data-integrity
370 * purposes - I/O may not be started against all dirty pages.
372 int filemap_flush(struct address_space *mapping)
374 return __filemap_fdatawrite(mapping, WB_SYNC_NONE);
376 EXPORT_SYMBOL(filemap_flush);
378 static int __filemap_fdatawait_range(struct address_space *mapping,
379 loff_t start_byte, loff_t end_byte)
381 pgoff_t index = start_byte >> PAGE_SHIFT;
382 pgoff_t end = end_byte >> PAGE_SHIFT;
387 if (end_byte < start_byte)
390 pagevec_init(&pvec, 0);
391 while ((index <= end) &&
392 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
393 PAGECACHE_TAG_WRITEBACK,
394 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1)) != 0) {
397 for (i = 0; i < nr_pages; i++) {
398 struct page *page = pvec.pages[i];
400 /* until radix tree lookup accepts end_index */
401 if (page->index > end)
404 wait_on_page_writeback(page);
405 if (TestClearPageError(page))
408 pagevec_release(&pvec);
416 * filemap_fdatawait_range - wait for writeback to complete
417 * @mapping: address space structure to wait for
418 * @start_byte: offset in bytes where the range starts
419 * @end_byte: offset in bytes where the range ends (inclusive)
421 * Walk the list of under-writeback pages of the given address space
422 * in the given range and wait for all of them. Check error status of
423 * the address space and return it.
425 * Since the error status of the address space is cleared by this function,
426 * callers are responsible for checking the return value and handling and/or
427 * reporting the error.
429 int filemap_fdatawait_range(struct address_space *mapping, loff_t start_byte,
434 ret = __filemap_fdatawait_range(mapping, start_byte, end_byte);
435 ret2 = filemap_check_errors(mapping);
441 EXPORT_SYMBOL(filemap_fdatawait_range);
444 * filemap_fdatawait_keep_errors - wait for writeback without clearing errors
445 * @mapping: address space structure to wait for
447 * Walk the list of under-writeback pages of the given address space
448 * and wait for all of them. Unlike filemap_fdatawait(), this function
449 * does not clear error status of the address space.
451 * Use this function if callers don't handle errors themselves. Expected
452 * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2),
455 void filemap_fdatawait_keep_errors(struct address_space *mapping)
457 loff_t i_size = i_size_read(mapping->host);
462 __filemap_fdatawait_range(mapping, 0, i_size - 1);
466 * filemap_fdatawait - wait for all under-writeback pages to complete
467 * @mapping: address space structure to wait for
469 * Walk the list of under-writeback pages of the given address space
470 * and wait for all of them. Check error status of the address space
473 * Since the error status of the address space is cleared by this function,
474 * callers are responsible for checking the return value and handling and/or
475 * reporting the error.
477 int filemap_fdatawait(struct address_space *mapping)
479 loff_t i_size = i_size_read(mapping->host);
484 return filemap_fdatawait_range(mapping, 0, i_size - 1);
486 EXPORT_SYMBOL(filemap_fdatawait);
488 int filemap_write_and_wait(struct address_space *mapping)
492 if ((!dax_mapping(mapping) && mapping->nrpages) ||
493 (dax_mapping(mapping) && mapping->nrexceptional)) {
494 err = filemap_fdatawrite(mapping);
496 * Even if the above returned error, the pages may be
497 * written partially (e.g. -ENOSPC), so we wait for it.
498 * But the -EIO is special case, it may indicate the worst
499 * thing (e.g. bug) happened, so we avoid waiting for it.
502 int err2 = filemap_fdatawait(mapping);
507 err = filemap_check_errors(mapping);
511 EXPORT_SYMBOL(filemap_write_and_wait);
514 * filemap_write_and_wait_range - write out & wait on a file range
515 * @mapping: the address_space for the pages
516 * @lstart: offset in bytes where the range starts
517 * @lend: offset in bytes where the range ends (inclusive)
519 * Write out and wait upon file offsets lstart->lend, inclusive.
521 * Note that `lend' is inclusive (describes the last byte to be written) so
522 * that this function can be used to write to the very end-of-file (end = -1).
524 int filemap_write_and_wait_range(struct address_space *mapping,
525 loff_t lstart, loff_t lend)
529 if ((!dax_mapping(mapping) && mapping->nrpages) ||
530 (dax_mapping(mapping) && mapping->nrexceptional)) {
531 err = __filemap_fdatawrite_range(mapping, lstart, lend,
533 /* See comment of filemap_write_and_wait() */
535 int err2 = filemap_fdatawait_range(mapping,
541 err = filemap_check_errors(mapping);
545 EXPORT_SYMBOL(filemap_write_and_wait_range);
548 * replace_page_cache_page - replace a pagecache page with a new one
549 * @old: page to be replaced
550 * @new: page to replace with
551 * @gfp_mask: allocation mode
553 * This function replaces a page in the pagecache with a new one. On
554 * success it acquires the pagecache reference for the new page and
555 * drops it for the old page. Both the old and new pages must be
556 * locked. This function does not add the new page to the LRU, the
557 * caller must do that.
559 * The remove + add is atomic. The only way this function can fail is
560 * memory allocation failure.
562 int replace_page_cache_page(struct page *old, struct page *new, gfp_t gfp_mask)
566 VM_BUG_ON_PAGE(!PageLocked(old), old);
567 VM_BUG_ON_PAGE(!PageLocked(new), new);
568 VM_BUG_ON_PAGE(new->mapping, new);
570 error = radix_tree_preload(gfp_mask & ~__GFP_HIGHMEM);
572 struct address_space *mapping = old->mapping;
573 void (*freepage)(struct page *);
576 pgoff_t offset = old->index;
577 freepage = mapping->a_ops->freepage;
580 new->mapping = mapping;
583 spin_lock_irqsave(&mapping->tree_lock, flags);
584 __delete_from_page_cache(old, NULL);
585 error = page_cache_tree_insert(mapping, new, NULL);
589 * hugetlb pages do not participate in page cache accounting.
592 __inc_node_page_state(new, NR_FILE_PAGES);
593 if (PageSwapBacked(new))
594 __inc_node_page_state(new, NR_SHMEM);
595 spin_unlock_irqrestore(&mapping->tree_lock, flags);
596 mem_cgroup_migrate(old, new);
597 radix_tree_preload_end();
605 EXPORT_SYMBOL_GPL(replace_page_cache_page);
607 static int __add_to_page_cache_locked(struct page *page,
608 struct address_space *mapping,
609 pgoff_t offset, gfp_t gfp_mask,
612 int huge = PageHuge(page);
613 struct mem_cgroup *memcg;
616 VM_BUG_ON_PAGE(!PageLocked(page), page);
617 VM_BUG_ON_PAGE(PageSwapBacked(page), page);
620 error = mem_cgroup_try_charge(page, current->mm,
621 gfp_mask, &memcg, false);
626 error = radix_tree_maybe_preload(gfp_mask & ~__GFP_HIGHMEM);
629 mem_cgroup_cancel_charge(page, memcg, false);
634 page->mapping = mapping;
635 page->index = offset;
637 spin_lock_irq(&mapping->tree_lock);
638 error = page_cache_tree_insert(mapping, page, shadowp);
639 radix_tree_preload_end();
643 /* hugetlb pages do not participate in page cache accounting. */
645 __inc_node_page_state(page, NR_FILE_PAGES);
646 spin_unlock_irq(&mapping->tree_lock);
648 mem_cgroup_commit_charge(page, memcg, false, false);
649 trace_mm_filemap_add_to_page_cache(page);
652 page->mapping = NULL;
653 /* Leave page->index set: truncation relies upon it */
654 spin_unlock_irq(&mapping->tree_lock);
656 mem_cgroup_cancel_charge(page, memcg, false);
662 * add_to_page_cache_locked - add a locked page to the pagecache
664 * @mapping: the page's address_space
665 * @offset: page index
666 * @gfp_mask: page allocation mode
668 * This function is used to add a page to the pagecache. It must be locked.
669 * This function does not add the page to the LRU. The caller must do that.
671 int add_to_page_cache_locked(struct page *page, struct address_space *mapping,
672 pgoff_t offset, gfp_t gfp_mask)
674 return __add_to_page_cache_locked(page, mapping, offset,
677 EXPORT_SYMBOL(add_to_page_cache_locked);
679 int add_to_page_cache_lru(struct page *page, struct address_space *mapping,
680 pgoff_t offset, gfp_t gfp_mask)
685 __SetPageLocked(page);
686 ret = __add_to_page_cache_locked(page, mapping, offset,
689 __ClearPageLocked(page);
692 * The page might have been evicted from cache only
693 * recently, in which case it should be activated like
694 * any other repeatedly accessed page.
695 * The exception is pages getting rewritten; evicting other
696 * data from the working set, only to cache data that will
697 * get overwritten with something else, is a waste of memory.
699 if (!(gfp_mask & __GFP_WRITE) &&
700 shadow && workingset_refault(shadow)) {
702 workingset_activation(page);
704 ClearPageActive(page);
709 EXPORT_SYMBOL_GPL(add_to_page_cache_lru);
712 struct page *__page_cache_alloc(gfp_t gfp)
717 if (cpuset_do_page_mem_spread()) {
718 unsigned int cpuset_mems_cookie;
720 cpuset_mems_cookie = read_mems_allowed_begin();
721 n = cpuset_mem_spread_node();
722 page = __alloc_pages_node(n, gfp, 0);
723 } while (!page && read_mems_allowed_retry(cpuset_mems_cookie));
727 return alloc_pages(gfp, 0);
729 EXPORT_SYMBOL(__page_cache_alloc);
733 * In order to wait for pages to become available there must be
734 * waitqueues associated with pages. By using a hash table of
735 * waitqueues where the bucket discipline is to maintain all
736 * waiters on the same queue and wake all when any of the pages
737 * become available, and for the woken contexts to check to be
738 * sure the appropriate page became available, this saves space
739 * at a cost of "thundering herd" phenomena during rare hash
742 #define PAGE_WAIT_TABLE_BITS 8
743 #define PAGE_WAIT_TABLE_SIZE (1 << PAGE_WAIT_TABLE_BITS)
744 static wait_queue_head_t page_wait_table[PAGE_WAIT_TABLE_SIZE] __cacheline_aligned;
746 static wait_queue_head_t *page_waitqueue(struct page *page)
748 return &page_wait_table[hash_ptr(page, PAGE_WAIT_TABLE_BITS)];
751 void __init pagecache_init(void)
755 for (i = 0; i < PAGE_WAIT_TABLE_SIZE; i++)
756 init_waitqueue_head(&page_wait_table[i]);
758 page_writeback_init();
761 struct wait_page_key {
767 struct wait_page_queue {
773 static int wake_page_function(wait_queue_t *wait, unsigned mode, int sync, void *arg)
775 struct wait_page_key *key = arg;
776 struct wait_page_queue *wait_page
777 = container_of(wait, struct wait_page_queue, wait);
779 if (wait_page->page != key->page)
783 if (wait_page->bit_nr != key->bit_nr)
785 if (test_bit(key->bit_nr, &key->page->flags))
788 return autoremove_wake_function(wait, mode, sync, key);
791 void wake_up_page_bit(struct page *page, int bit_nr)
793 wait_queue_head_t *q = page_waitqueue(page);
794 struct wait_page_key key;
801 spin_lock_irqsave(&q->lock, flags);
802 __wake_up_locked_key(q, TASK_NORMAL, &key);
804 * It is possible for other pages to have collided on the waitqueue
805 * hash, so in that case check for a page match. That prevents a long-
808 * It is still possible to miss a case here, when we woke page waiters
809 * and removed them from the waitqueue, but there are still other
812 if (!waitqueue_active(q) || !key.page_match) {
813 ClearPageWaiters(page);
815 * It's possible to miss clearing Waiters here, when we woke
816 * our page waiters, but the hashed waitqueue has waiters for
819 * That's okay, it's a rare case. The next waker will clear it.
822 spin_unlock_irqrestore(&q->lock, flags);
824 EXPORT_SYMBOL(wake_up_page_bit);
826 static inline int wait_on_page_bit_common(wait_queue_head_t *q,
827 struct page *page, int bit_nr, int state, bool lock)
829 struct wait_page_queue wait_page;
830 wait_queue_t *wait = &wait_page.wait;
834 wait->func = wake_page_function;
835 wait_page.page = page;
836 wait_page.bit_nr = bit_nr;
839 spin_lock_irq(&q->lock);
841 if (likely(list_empty(&wait->task_list))) {
843 __add_wait_queue_tail_exclusive(q, wait);
845 __add_wait_queue(q, wait);
846 SetPageWaiters(page);
849 set_current_state(state);
851 spin_unlock_irq(&q->lock);
853 if (likely(test_bit(bit_nr, &page->flags))) {
855 if (unlikely(signal_pending_state(state, current))) {
862 if (!test_and_set_bit_lock(bit_nr, &page->flags))
865 if (!test_bit(bit_nr, &page->flags))
870 finish_wait(q, wait);
873 * A signal could leave PageWaiters set. Clearing it here if
874 * !waitqueue_active would be possible (by open-coding finish_wait),
875 * but still fail to catch it in the case of wait hash collision. We
876 * already can fail to clear wait hash collision cases, so don't
877 * bother with signals either.
883 void wait_on_page_bit(struct page *page, int bit_nr)
885 wait_queue_head_t *q = page_waitqueue(page);
886 wait_on_page_bit_common(q, page, bit_nr, TASK_UNINTERRUPTIBLE, false);
888 EXPORT_SYMBOL(wait_on_page_bit);
890 int wait_on_page_bit_killable(struct page *page, int bit_nr)
892 wait_queue_head_t *q = page_waitqueue(page);
893 return wait_on_page_bit_common(q, page, bit_nr, TASK_KILLABLE, false);
897 * add_page_wait_queue - Add an arbitrary waiter to a page's wait queue
898 * @page: Page defining the wait queue of interest
899 * @waiter: Waiter to add to the queue
901 * Add an arbitrary @waiter to the wait queue for the nominated @page.
903 void add_page_wait_queue(struct page *page, wait_queue_t *waiter)
905 wait_queue_head_t *q = page_waitqueue(page);
908 spin_lock_irqsave(&q->lock, flags);
909 __add_wait_queue(q, waiter);
910 SetPageWaiters(page);
911 spin_unlock_irqrestore(&q->lock, flags);
913 EXPORT_SYMBOL_GPL(add_page_wait_queue);
916 * unlock_page - unlock a locked page
919 * Unlocks the page and wakes up sleepers in ___wait_on_page_locked().
920 * Also wakes sleepers in wait_on_page_writeback() because the wakeup
921 * mechanism between PageLocked pages and PageWriteback pages is shared.
922 * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep.
924 * The mb is necessary to enforce ordering between the clear_bit and the read
925 * of the waitqueue (to avoid SMP races with a parallel wait_on_page_locked()).
927 void unlock_page(struct page *page)
929 page = compound_head(page);
930 VM_BUG_ON_PAGE(!PageLocked(page), page);
931 clear_bit_unlock(PG_locked, &page->flags);
932 smp_mb__after_atomic();
933 wake_up_page(page, PG_locked);
935 EXPORT_SYMBOL(unlock_page);
938 * end_page_writeback - end writeback against a page
941 void end_page_writeback(struct page *page)
944 * TestClearPageReclaim could be used here but it is an atomic
945 * operation and overkill in this particular case. Failing to
946 * shuffle a page marked for immediate reclaim is too mild to
947 * justify taking an atomic operation penalty at the end of
948 * ever page writeback.
950 if (PageReclaim(page)) {
951 ClearPageReclaim(page);
952 rotate_reclaimable_page(page);
955 if (!test_clear_page_writeback(page))
958 smp_mb__after_atomic();
959 wake_up_page(page, PG_writeback);
961 EXPORT_SYMBOL(end_page_writeback);
964 * After completing I/O on a page, call this routine to update the page
965 * flags appropriately
967 void page_endio(struct page *page, bool is_write, int err)
971 SetPageUptodate(page);
973 ClearPageUptodate(page);
981 mapping_set_error(page->mapping, err);
983 end_page_writeback(page);
986 EXPORT_SYMBOL_GPL(page_endio);
989 * __lock_page - get a lock on the page, assuming we need to sleep to get it
990 * @page: the page to lock
992 void __lock_page(struct page *__page)
994 struct page *page = compound_head(__page);
995 wait_queue_head_t *q = page_waitqueue(page);
996 wait_on_page_bit_common(q, page, PG_locked, TASK_UNINTERRUPTIBLE, true);
998 EXPORT_SYMBOL(__lock_page);
1000 int __lock_page_killable(struct page *__page)
1002 struct page *page = compound_head(__page);
1003 wait_queue_head_t *q = page_waitqueue(page);
1004 return wait_on_page_bit_common(q, page, PG_locked, TASK_KILLABLE, true);
1006 EXPORT_SYMBOL_GPL(__lock_page_killable);
1010 * 1 - page is locked; mmap_sem is still held.
1011 * 0 - page is not locked.
1012 * mmap_sem has been released (up_read()), unless flags had both
1013 * FAULT_FLAG_ALLOW_RETRY and FAULT_FLAG_RETRY_NOWAIT set, in
1014 * which case mmap_sem is still held.
1016 * If neither ALLOW_RETRY nor KILLABLE are set, will always return 1
1017 * with the page locked and the mmap_sem unperturbed.
1019 int __lock_page_or_retry(struct page *page, struct mm_struct *mm,
1022 if (flags & FAULT_FLAG_ALLOW_RETRY) {
1024 * CAUTION! In this case, mmap_sem is not released
1025 * even though return 0.
1027 if (flags & FAULT_FLAG_RETRY_NOWAIT)
1030 up_read(&mm->mmap_sem);
1031 if (flags & FAULT_FLAG_KILLABLE)
1032 wait_on_page_locked_killable(page);
1034 wait_on_page_locked(page);
1037 if (flags & FAULT_FLAG_KILLABLE) {
1040 ret = __lock_page_killable(page);
1042 up_read(&mm->mmap_sem);
1052 * page_cache_next_hole - find the next hole (not-present entry)
1055 * @max_scan: maximum range to search
1057 * Search the set [index, min(index+max_scan-1, MAX_INDEX)] for the
1058 * lowest indexed hole.
1060 * Returns: the index of the hole if found, otherwise returns an index
1061 * outside of the set specified (in which case 'return - index >=
1062 * max_scan' will be true). In rare cases of index wrap-around, 0 will
1065 * page_cache_next_hole may be called under rcu_read_lock. However,
1066 * like radix_tree_gang_lookup, this will not atomically search a
1067 * snapshot of the tree at a single point in time. For example, if a
1068 * hole is created at index 5, then subsequently a hole is created at
1069 * index 10, page_cache_next_hole covering both indexes may return 10
1070 * if called under rcu_read_lock.
1072 pgoff_t page_cache_next_hole(struct address_space *mapping,
1073 pgoff_t index, unsigned long max_scan)
1077 for (i = 0; i < max_scan; i++) {
1080 page = radix_tree_lookup(&mapping->page_tree, index);
1081 if (!page || radix_tree_exceptional_entry(page))
1090 EXPORT_SYMBOL(page_cache_next_hole);
1093 * page_cache_prev_hole - find the prev hole (not-present entry)
1096 * @max_scan: maximum range to search
1098 * Search backwards in the range [max(index-max_scan+1, 0), index] for
1101 * Returns: the index of the hole if found, otherwise returns an index
1102 * outside of the set specified (in which case 'index - return >=
1103 * max_scan' will be true). In rare cases of wrap-around, ULONG_MAX
1106 * page_cache_prev_hole may be called under rcu_read_lock. However,
1107 * like radix_tree_gang_lookup, this will not atomically search a
1108 * snapshot of the tree at a single point in time. For example, if a
1109 * hole is created at index 10, then subsequently a hole is created at
1110 * index 5, page_cache_prev_hole covering both indexes may return 5 if
1111 * called under rcu_read_lock.
1113 pgoff_t page_cache_prev_hole(struct address_space *mapping,
1114 pgoff_t index, unsigned long max_scan)
1118 for (i = 0; i < max_scan; i++) {
1121 page = radix_tree_lookup(&mapping->page_tree, index);
1122 if (!page || radix_tree_exceptional_entry(page))
1125 if (index == ULONG_MAX)
1131 EXPORT_SYMBOL(page_cache_prev_hole);
1134 * find_get_entry - find and get a page cache entry
1135 * @mapping: the address_space to search
1136 * @offset: the page cache index
1138 * Looks up the page cache slot at @mapping & @offset. If there is a
1139 * page cache page, it is returned with an increased refcount.
1141 * If the slot holds a shadow entry of a previously evicted page, or a
1142 * swap entry from shmem/tmpfs, it is returned.
1144 * Otherwise, %NULL is returned.
1146 struct page *find_get_entry(struct address_space *mapping, pgoff_t offset)
1149 struct page *head, *page;
1154 pagep = radix_tree_lookup_slot(&mapping->page_tree, offset);
1156 page = radix_tree_deref_slot(pagep);
1157 if (unlikely(!page))
1159 if (radix_tree_exception(page)) {
1160 if (radix_tree_deref_retry(page))
1163 * A shadow entry of a recently evicted page,
1164 * or a swap entry from shmem/tmpfs. Return
1165 * it without attempting to raise page count.
1170 head = compound_head(page);
1171 if (!page_cache_get_speculative(head))
1174 /* The page was split under us? */
1175 if (compound_head(page) != head) {
1181 * Has the page moved?
1182 * This is part of the lockless pagecache protocol. See
1183 * include/linux/pagemap.h for details.
1185 if (unlikely(page != *pagep)) {
1195 EXPORT_SYMBOL(find_get_entry);
1198 * find_lock_entry - locate, pin and lock a page cache entry
1199 * @mapping: the address_space to search
1200 * @offset: the page cache index
1202 * Looks up the page cache slot at @mapping & @offset. If there is a
1203 * page cache page, it is returned locked and with an increased
1206 * If the slot holds a shadow entry of a previously evicted page, or a
1207 * swap entry from shmem/tmpfs, it is returned.
1209 * Otherwise, %NULL is returned.
1211 * find_lock_entry() may sleep.
1213 struct page *find_lock_entry(struct address_space *mapping, pgoff_t offset)
1218 page = find_get_entry(mapping, offset);
1219 if (page && !radix_tree_exception(page)) {
1221 /* Has the page been truncated? */
1222 if (unlikely(page_mapping(page) != mapping)) {
1227 VM_BUG_ON_PAGE(page_to_pgoff(page) != offset, page);
1231 EXPORT_SYMBOL(find_lock_entry);
1234 * pagecache_get_page - find and get a page reference
1235 * @mapping: the address_space to search
1236 * @offset: the page index
1237 * @fgp_flags: PCG flags
1238 * @gfp_mask: gfp mask to use for the page cache data page allocation
1240 * Looks up the page cache slot at @mapping & @offset.
1242 * PCG flags modify how the page is returned.
1244 * FGP_ACCESSED: the page will be marked accessed
1245 * FGP_LOCK: Page is return locked
1246 * FGP_CREAT: If page is not present then a new page is allocated using
1247 * @gfp_mask and added to the page cache and the VM's LRU
1248 * list. The page is returned locked and with an increased
1249 * refcount. Otherwise, %NULL is returned.
1251 * If FGP_LOCK or FGP_CREAT are specified then the function may sleep even
1252 * if the GFP flags specified for FGP_CREAT are atomic.
1254 * If there is a page cache page, it is returned with an increased refcount.
1256 struct page *pagecache_get_page(struct address_space *mapping, pgoff_t offset,
1257 int fgp_flags, gfp_t gfp_mask)
1262 page = find_get_entry(mapping, offset);
1263 if (radix_tree_exceptional_entry(page))
1268 if (fgp_flags & FGP_LOCK) {
1269 if (fgp_flags & FGP_NOWAIT) {
1270 if (!trylock_page(page)) {
1278 /* Has the page been truncated? */
1279 if (unlikely(page->mapping != mapping)) {
1284 VM_BUG_ON_PAGE(page->index != offset, page);
1287 if (page && (fgp_flags & FGP_ACCESSED))
1288 mark_page_accessed(page);
1291 if (!page && (fgp_flags & FGP_CREAT)) {
1293 if ((fgp_flags & FGP_WRITE) && mapping_cap_account_dirty(mapping))
1294 gfp_mask |= __GFP_WRITE;
1295 if (fgp_flags & FGP_NOFS)
1296 gfp_mask &= ~__GFP_FS;
1298 page = __page_cache_alloc(gfp_mask);
1302 if (WARN_ON_ONCE(!(fgp_flags & FGP_LOCK)))
1303 fgp_flags |= FGP_LOCK;
1305 /* Init accessed so avoid atomic mark_page_accessed later */
1306 if (fgp_flags & FGP_ACCESSED)
1307 __SetPageReferenced(page);
1309 err = add_to_page_cache_lru(page, mapping, offset,
1310 gfp_mask & GFP_RECLAIM_MASK);
1311 if (unlikely(err)) {
1321 EXPORT_SYMBOL(pagecache_get_page);
1324 * find_get_entries - gang pagecache lookup
1325 * @mapping: The address_space to search
1326 * @start: The starting page cache index
1327 * @nr_entries: The maximum number of entries
1328 * @entries: Where the resulting entries are placed
1329 * @indices: The cache indices corresponding to the entries in @entries
1331 * find_get_entries() will search for and return a group of up to
1332 * @nr_entries entries in the mapping. The entries are placed at
1333 * @entries. find_get_entries() takes a reference against any actual
1336 * The search returns a group of mapping-contiguous page cache entries
1337 * with ascending indexes. There may be holes in the indices due to
1338 * not-present pages.
1340 * Any shadow entries of evicted pages, or swap entries from
1341 * shmem/tmpfs, are included in the returned array.
1343 * find_get_entries() returns the number of pages and shadow entries
1346 unsigned find_get_entries(struct address_space *mapping,
1347 pgoff_t start, unsigned int nr_entries,
1348 struct page **entries, pgoff_t *indices)
1351 unsigned int ret = 0;
1352 struct radix_tree_iter iter;
1358 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) {
1359 struct page *head, *page;
1361 page = radix_tree_deref_slot(slot);
1362 if (unlikely(!page))
1364 if (radix_tree_exception(page)) {
1365 if (radix_tree_deref_retry(page)) {
1366 slot = radix_tree_iter_retry(&iter);
1370 * A shadow entry of a recently evicted page, a swap
1371 * entry from shmem/tmpfs or a DAX entry. Return it
1372 * without attempting to raise page count.
1377 head = compound_head(page);
1378 if (!page_cache_get_speculative(head))
1381 /* The page was split under us? */
1382 if (compound_head(page) != head) {
1387 /* Has the page moved? */
1388 if (unlikely(page != *slot)) {
1393 indices[ret] = iter.index;
1394 entries[ret] = page;
1395 if (++ret == nr_entries)
1403 * find_get_pages - gang pagecache lookup
1404 * @mapping: The address_space to search
1405 * @start: The starting page index
1406 * @nr_pages: The maximum number of pages
1407 * @pages: Where the resulting pages are placed
1409 * find_get_pages() will search for and return a group of up to
1410 * @nr_pages pages in the mapping. The pages are placed at @pages.
1411 * find_get_pages() takes a reference against the returned pages.
1413 * The search returns a group of mapping-contiguous pages with ascending
1414 * indexes. There may be holes in the indices due to not-present pages.
1416 * find_get_pages() returns the number of pages which were found.
1418 unsigned find_get_pages(struct address_space *mapping, pgoff_t start,
1419 unsigned int nr_pages, struct page **pages)
1421 struct radix_tree_iter iter;
1425 if (unlikely(!nr_pages))
1429 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) {
1430 struct page *head, *page;
1432 page = radix_tree_deref_slot(slot);
1433 if (unlikely(!page))
1436 if (radix_tree_exception(page)) {
1437 if (radix_tree_deref_retry(page)) {
1438 slot = radix_tree_iter_retry(&iter);
1442 * A shadow entry of a recently evicted page,
1443 * or a swap entry from shmem/tmpfs. Skip
1449 head = compound_head(page);
1450 if (!page_cache_get_speculative(head))
1453 /* The page was split under us? */
1454 if (compound_head(page) != head) {
1459 /* Has the page moved? */
1460 if (unlikely(page != *slot)) {
1466 if (++ret == nr_pages)
1475 * find_get_pages_contig - gang contiguous pagecache lookup
1476 * @mapping: The address_space to search
1477 * @index: The starting page index
1478 * @nr_pages: The maximum number of pages
1479 * @pages: Where the resulting pages are placed
1481 * find_get_pages_contig() works exactly like find_get_pages(), except
1482 * that the returned number of pages are guaranteed to be contiguous.
1484 * find_get_pages_contig() returns the number of pages which were found.
1486 unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index,
1487 unsigned int nr_pages, struct page **pages)
1489 struct radix_tree_iter iter;
1491 unsigned int ret = 0;
1493 if (unlikely(!nr_pages))
1497 radix_tree_for_each_contig(slot, &mapping->page_tree, &iter, index) {
1498 struct page *head, *page;
1500 page = radix_tree_deref_slot(slot);
1501 /* The hole, there no reason to continue */
1502 if (unlikely(!page))
1505 if (radix_tree_exception(page)) {
1506 if (radix_tree_deref_retry(page)) {
1507 slot = radix_tree_iter_retry(&iter);
1511 * A shadow entry of a recently evicted page,
1512 * or a swap entry from shmem/tmpfs. Stop
1513 * looking for contiguous pages.
1518 head = compound_head(page);
1519 if (!page_cache_get_speculative(head))
1522 /* The page was split under us? */
1523 if (compound_head(page) != head) {
1528 /* Has the page moved? */
1529 if (unlikely(page != *slot)) {
1535 * must check mapping and index after taking the ref.
1536 * otherwise we can get both false positives and false
1537 * negatives, which is just confusing to the caller.
1539 if (page->mapping == NULL || page_to_pgoff(page) != iter.index) {
1545 if (++ret == nr_pages)
1551 EXPORT_SYMBOL(find_get_pages_contig);
1554 * find_get_pages_tag - find and return pages that match @tag
1555 * @mapping: the address_space to search
1556 * @index: the starting page index
1557 * @tag: the tag index
1558 * @nr_pages: the maximum number of pages
1559 * @pages: where the resulting pages are placed
1561 * Like find_get_pages, except we only return pages which are tagged with
1562 * @tag. We update @index to index the next page for the traversal.
1564 unsigned find_get_pages_tag(struct address_space *mapping, pgoff_t *index,
1565 int tag, unsigned int nr_pages, struct page **pages)
1567 struct radix_tree_iter iter;
1571 if (unlikely(!nr_pages))
1575 radix_tree_for_each_tagged(slot, &mapping->page_tree,
1576 &iter, *index, tag) {
1577 struct page *head, *page;
1579 page = radix_tree_deref_slot(slot);
1580 if (unlikely(!page))
1583 if (radix_tree_exception(page)) {
1584 if (radix_tree_deref_retry(page)) {
1585 slot = radix_tree_iter_retry(&iter);
1589 * A shadow entry of a recently evicted page.
1591 * Those entries should never be tagged, but
1592 * this tree walk is lockless and the tags are
1593 * looked up in bulk, one radix tree node at a
1594 * time, so there is a sizable window for page
1595 * reclaim to evict a page we saw tagged.
1602 head = compound_head(page);
1603 if (!page_cache_get_speculative(head))
1606 /* The page was split under us? */
1607 if (compound_head(page) != head) {
1612 /* Has the page moved? */
1613 if (unlikely(page != *slot)) {
1619 if (++ret == nr_pages)
1626 *index = pages[ret - 1]->index + 1;
1630 EXPORT_SYMBOL(find_get_pages_tag);
1633 * find_get_entries_tag - find and return entries that match @tag
1634 * @mapping: the address_space to search
1635 * @start: the starting page cache index
1636 * @tag: the tag index
1637 * @nr_entries: the maximum number of entries
1638 * @entries: where the resulting entries are placed
1639 * @indices: the cache indices corresponding to the entries in @entries
1641 * Like find_get_entries, except we only return entries which are tagged with
1644 unsigned find_get_entries_tag(struct address_space *mapping, pgoff_t start,
1645 int tag, unsigned int nr_entries,
1646 struct page **entries, pgoff_t *indices)
1649 unsigned int ret = 0;
1650 struct radix_tree_iter iter;
1656 radix_tree_for_each_tagged(slot, &mapping->page_tree,
1657 &iter, start, tag) {
1658 struct page *head, *page;
1660 page = radix_tree_deref_slot(slot);
1661 if (unlikely(!page))
1663 if (radix_tree_exception(page)) {
1664 if (radix_tree_deref_retry(page)) {
1665 slot = radix_tree_iter_retry(&iter);
1670 * A shadow entry of a recently evicted page, a swap
1671 * entry from shmem/tmpfs or a DAX entry. Return it
1672 * without attempting to raise page count.
1677 head = compound_head(page);
1678 if (!page_cache_get_speculative(head))
1681 /* The page was split under us? */
1682 if (compound_head(page) != head) {
1687 /* Has the page moved? */
1688 if (unlikely(page != *slot)) {
1693 indices[ret] = iter.index;
1694 entries[ret] = page;
1695 if (++ret == nr_entries)
1701 EXPORT_SYMBOL(find_get_entries_tag);
1704 * CD/DVDs are error prone. When a medium error occurs, the driver may fail
1705 * a _large_ part of the i/o request. Imagine the worst scenario:
1707 * ---R__________________________________________B__________
1708 * ^ reading here ^ bad block(assume 4k)
1710 * read(R) => miss => readahead(R...B) => media error => frustrating retries
1711 * => failing the whole request => read(R) => read(R+1) =>
1712 * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
1713 * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
1714 * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
1716 * It is going insane. Fix it by quickly scaling down the readahead size.
1718 static void shrink_readahead_size_eio(struct file *filp,
1719 struct file_ra_state *ra)
1725 * do_generic_file_read - generic file read routine
1726 * @filp: the file to read
1727 * @ppos: current file position
1728 * @iter: data destination
1729 * @written: already copied
1731 * This is a generic file read routine, and uses the
1732 * mapping->a_ops->readpage() function for the actual low-level stuff.
1734 * This is really ugly. But the goto's actually try to clarify some
1735 * of the logic when it comes to error handling etc.
1737 static ssize_t do_generic_file_read(struct file *filp, loff_t *ppos,
1738 struct iov_iter *iter, ssize_t written)
1740 struct address_space *mapping = filp->f_mapping;
1741 struct inode *inode = mapping->host;
1742 struct file_ra_state *ra = &filp->f_ra;
1746 unsigned long offset; /* offset into pagecache page */
1747 unsigned int prev_offset;
1750 if (unlikely(*ppos >= inode->i_sb->s_maxbytes))
1752 iov_iter_truncate(iter, inode->i_sb->s_maxbytes);
1754 index = *ppos >> PAGE_SHIFT;
1755 prev_index = ra->prev_pos >> PAGE_SHIFT;
1756 prev_offset = ra->prev_pos & (PAGE_SIZE-1);
1757 last_index = (*ppos + iter->count + PAGE_SIZE-1) >> PAGE_SHIFT;
1758 offset = *ppos & ~PAGE_MASK;
1764 unsigned long nr, ret;
1768 page = find_get_page(mapping, index);
1770 page_cache_sync_readahead(mapping,
1772 index, last_index - index);
1773 page = find_get_page(mapping, index);
1774 if (unlikely(page == NULL))
1775 goto no_cached_page;
1777 if (PageReadahead(page)) {
1778 page_cache_async_readahead(mapping,
1780 index, last_index - index);
1782 if (!PageUptodate(page)) {
1784 * See comment in do_read_cache_page on why
1785 * wait_on_page_locked is used to avoid unnecessarily
1786 * serialisations and why it's safe.
1788 error = wait_on_page_locked_killable(page);
1789 if (unlikely(error))
1790 goto readpage_error;
1791 if (PageUptodate(page))
1794 if (inode->i_blkbits == PAGE_SHIFT ||
1795 !mapping->a_ops->is_partially_uptodate)
1796 goto page_not_up_to_date;
1797 /* pipes can't handle partially uptodate pages */
1798 if (unlikely(iter->type & ITER_PIPE))
1799 goto page_not_up_to_date;
1800 if (!trylock_page(page))
1801 goto page_not_up_to_date;
1802 /* Did it get truncated before we got the lock? */
1804 goto page_not_up_to_date_locked;
1805 if (!mapping->a_ops->is_partially_uptodate(page,
1806 offset, iter->count))
1807 goto page_not_up_to_date_locked;
1812 * i_size must be checked after we know the page is Uptodate.
1814 * Checking i_size after the check allows us to calculate
1815 * the correct value for "nr", which means the zero-filled
1816 * part of the page is not copied back to userspace (unless
1817 * another truncate extends the file - this is desired though).
1820 isize = i_size_read(inode);
1821 end_index = (isize - 1) >> PAGE_SHIFT;
1822 if (unlikely(!isize || index > end_index)) {
1827 /* nr is the maximum number of bytes to copy from this page */
1829 if (index == end_index) {
1830 nr = ((isize - 1) & ~PAGE_MASK) + 1;
1838 /* If users can be writing to this page using arbitrary
1839 * virtual addresses, take care about potential aliasing
1840 * before reading the page on the kernel side.
1842 if (mapping_writably_mapped(mapping))
1843 flush_dcache_page(page);
1846 * When a sequential read accesses a page several times,
1847 * only mark it as accessed the first time.
1849 if (prev_index != index || offset != prev_offset)
1850 mark_page_accessed(page);
1854 * Ok, we have the page, and it's up-to-date, so
1855 * now we can copy it to user space...
1858 ret = copy_page_to_iter(page, offset, nr, iter);
1860 index += offset >> PAGE_SHIFT;
1861 offset &= ~PAGE_MASK;
1862 prev_offset = offset;
1866 if (!iov_iter_count(iter))
1874 page_not_up_to_date:
1875 /* Get exclusive access to the page ... */
1876 error = lock_page_killable(page);
1877 if (unlikely(error))
1878 goto readpage_error;
1880 page_not_up_to_date_locked:
1881 /* Did it get truncated before we got the lock? */
1882 if (!page->mapping) {
1888 /* Did somebody else fill it already? */
1889 if (PageUptodate(page)) {
1896 * A previous I/O error may have been due to temporary
1897 * failures, eg. multipath errors.
1898 * PG_error will be set again if readpage fails.
1900 ClearPageError(page);
1901 /* Start the actual read. The read will unlock the page. */
1902 error = mapping->a_ops->readpage(filp, page);
1904 if (unlikely(error)) {
1905 if (error == AOP_TRUNCATED_PAGE) {
1910 goto readpage_error;
1913 if (!PageUptodate(page)) {
1914 error = lock_page_killable(page);
1915 if (unlikely(error))
1916 goto readpage_error;
1917 if (!PageUptodate(page)) {
1918 if (page->mapping == NULL) {
1920 * invalidate_mapping_pages got it
1927 shrink_readahead_size_eio(filp, ra);
1929 goto readpage_error;
1937 /* UHHUH! A synchronous read error occurred. Report it */
1943 * Ok, it wasn't cached, so we need to create a new
1946 page = page_cache_alloc_cold(mapping);
1951 error = add_to_page_cache_lru(page, mapping, index,
1952 mapping_gfp_constraint(mapping, GFP_KERNEL));
1955 if (error == -EEXIST) {
1965 ra->prev_pos = prev_index;
1966 ra->prev_pos <<= PAGE_SHIFT;
1967 ra->prev_pos |= prev_offset;
1969 *ppos = ((loff_t)index << PAGE_SHIFT) + offset;
1970 file_accessed(filp);
1971 return written ? written : error;
1975 * generic_file_read_iter - generic filesystem read routine
1976 * @iocb: kernel I/O control block
1977 * @iter: destination for the data read
1979 * This is the "read_iter()" routine for all filesystems
1980 * that can use the page cache directly.
1983 generic_file_read_iter(struct kiocb *iocb, struct iov_iter *iter)
1985 struct file *file = iocb->ki_filp;
1987 size_t count = iov_iter_count(iter);
1990 goto out; /* skip atime */
1992 if (iocb->ki_flags & IOCB_DIRECT) {
1993 struct address_space *mapping = file->f_mapping;
1994 struct inode *inode = mapping->host;
1995 struct iov_iter data = *iter;
1998 size = i_size_read(inode);
1999 retval = filemap_write_and_wait_range(mapping, iocb->ki_pos,
2000 iocb->ki_pos + count - 1);
2004 file_accessed(file);
2006 retval = mapping->a_ops->direct_IO(iocb, &data);
2008 iocb->ki_pos += retval;
2009 iov_iter_advance(iter, retval);
2013 * Btrfs can have a short DIO read if we encounter
2014 * compressed extents, so if there was an error, or if
2015 * we've already read everything we wanted to, or if
2016 * there was a short read because we hit EOF, go ahead
2017 * and return. Otherwise fallthrough to buffered io for
2018 * the rest of the read. Buffered reads will not work for
2019 * DAX files, so don't bother trying.
2021 if (retval < 0 || !iov_iter_count(iter) || iocb->ki_pos >= size ||
2026 retval = do_generic_file_read(file, &iocb->ki_pos, iter, retval);
2030 EXPORT_SYMBOL(generic_file_read_iter);
2034 * page_cache_read - adds requested page to the page cache if not already there
2035 * @file: file to read
2036 * @offset: page index
2037 * @gfp_mask: memory allocation flags
2039 * This adds the requested page to the page cache if it isn't already there,
2040 * and schedules an I/O to read in its contents from disk.
2042 static int page_cache_read(struct file *file, pgoff_t offset, gfp_t gfp_mask)
2044 struct address_space *mapping = file->f_mapping;
2049 page = __page_cache_alloc(gfp_mask|__GFP_COLD);
2053 ret = add_to_page_cache_lru(page, mapping, offset, gfp_mask & GFP_KERNEL);
2055 ret = mapping->a_ops->readpage(file, page);
2056 else if (ret == -EEXIST)
2057 ret = 0; /* losing race to add is OK */
2061 } while (ret == AOP_TRUNCATED_PAGE);
2066 #define MMAP_LOTSAMISS (100)
2069 * Synchronous readahead happens when we don't even find
2070 * a page in the page cache at all.
2072 static void do_sync_mmap_readahead(struct vm_area_struct *vma,
2073 struct file_ra_state *ra,
2077 struct address_space *mapping = file->f_mapping;
2079 /* If we don't want any read-ahead, don't bother */
2080 if (vma->vm_flags & VM_RAND_READ)
2085 if (vma->vm_flags & VM_SEQ_READ) {
2086 page_cache_sync_readahead(mapping, ra, file, offset,
2091 /* Avoid banging the cache line if not needed */
2092 if (ra->mmap_miss < MMAP_LOTSAMISS * 10)
2096 * Do we miss much more than hit in this file? If so,
2097 * stop bothering with read-ahead. It will only hurt.
2099 if (ra->mmap_miss > MMAP_LOTSAMISS)
2105 ra->start = max_t(long, 0, offset - ra->ra_pages / 2);
2106 ra->size = ra->ra_pages;
2107 ra->async_size = ra->ra_pages / 4;
2108 ra_submit(ra, mapping, file);
2112 * Asynchronous readahead happens when we find the page and PG_readahead,
2113 * so we want to possibly extend the readahead further..
2115 static void do_async_mmap_readahead(struct vm_area_struct *vma,
2116 struct file_ra_state *ra,
2121 struct address_space *mapping = file->f_mapping;
2123 /* If we don't want any read-ahead, don't bother */
2124 if (vma->vm_flags & VM_RAND_READ)
2126 if (ra->mmap_miss > 0)
2128 if (PageReadahead(page))
2129 page_cache_async_readahead(mapping, ra, file,
2130 page, offset, ra->ra_pages);
2134 * filemap_fault - read in file data for page fault handling
2135 * @vma: vma in which the fault was taken
2136 * @vmf: struct vm_fault containing details of the fault
2138 * filemap_fault() is invoked via the vma operations vector for a
2139 * mapped memory region to read in file data during a page fault.
2141 * The goto's are kind of ugly, but this streamlines the normal case of having
2142 * it in the page cache, and handles the special cases reasonably without
2143 * having a lot of duplicated code.
2145 * vma->vm_mm->mmap_sem must be held on entry.
2147 * If our return value has VM_FAULT_RETRY set, it's because
2148 * lock_page_or_retry() returned 0.
2149 * The mmap_sem has usually been released in this case.
2150 * See __lock_page_or_retry() for the exception.
2152 * If our return value does not have VM_FAULT_RETRY set, the mmap_sem
2153 * has not been released.
2155 * We never return with VM_FAULT_RETRY and a bit from VM_FAULT_ERROR set.
2157 int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
2160 struct file *file = vma->vm_file;
2161 struct address_space *mapping = file->f_mapping;
2162 struct file_ra_state *ra = &file->f_ra;
2163 struct inode *inode = mapping->host;
2164 pgoff_t offset = vmf->pgoff;
2169 size = round_up(i_size_read(inode), PAGE_SIZE);
2170 if (offset >= size >> PAGE_SHIFT)
2171 return VM_FAULT_SIGBUS;
2174 * Do we have something in the page cache already?
2176 page = find_get_page(mapping, offset);
2177 if (likely(page) && !(vmf->flags & FAULT_FLAG_TRIED)) {
2179 * We found the page, so try async readahead before
2180 * waiting for the lock.
2182 do_async_mmap_readahead(vma, ra, file, page, offset);
2184 /* No page in the page cache at all */
2185 do_sync_mmap_readahead(vma, ra, file, offset);
2186 count_vm_event(PGMAJFAULT);
2187 mem_cgroup_count_vm_event(vma->vm_mm, PGMAJFAULT);
2188 ret = VM_FAULT_MAJOR;
2190 page = find_get_page(mapping, offset);
2192 goto no_cached_page;
2195 if (!lock_page_or_retry(page, vma->vm_mm, vmf->flags)) {
2197 return ret | VM_FAULT_RETRY;
2200 /* Did it get truncated? */
2201 if (unlikely(page->mapping != mapping)) {
2206 VM_BUG_ON_PAGE(page->index != offset, page);
2209 * We have a locked page in the page cache, now we need to check
2210 * that it's up-to-date. If not, it is going to be due to an error.
2212 if (unlikely(!PageUptodate(page)))
2213 goto page_not_uptodate;
2216 * Found the page and have a reference on it.
2217 * We must recheck i_size under page lock.
2219 size = round_up(i_size_read(inode), PAGE_SIZE);
2220 if (unlikely(offset >= size >> PAGE_SHIFT)) {
2223 return VM_FAULT_SIGBUS;
2227 return ret | VM_FAULT_LOCKED;
2231 * We're only likely to ever get here if MADV_RANDOM is in
2234 error = page_cache_read(file, offset, vmf->gfp_mask);
2237 * The page we want has now been added to the page cache.
2238 * In the unlikely event that someone removed it in the
2239 * meantime, we'll just come back here and read it again.
2245 * An error return from page_cache_read can result if the
2246 * system is low on memory, or a problem occurs while trying
2249 if (error == -ENOMEM)
2250 return VM_FAULT_OOM;
2251 return VM_FAULT_SIGBUS;
2255 * Umm, take care of errors if the page isn't up-to-date.
2256 * Try to re-read it _once_. We do this synchronously,
2257 * because there really aren't any performance issues here
2258 * and we need to check for errors.
2260 ClearPageError(page);
2261 error = mapping->a_ops->readpage(file, page);
2263 wait_on_page_locked(page);
2264 if (!PageUptodate(page))
2269 if (!error || error == AOP_TRUNCATED_PAGE)
2272 /* Things didn't work out. Return zero to tell the mm layer so. */
2273 shrink_readahead_size_eio(file, ra);
2274 return VM_FAULT_SIGBUS;
2276 EXPORT_SYMBOL(filemap_fault);
2278 void filemap_map_pages(struct vm_fault *vmf,
2279 pgoff_t start_pgoff, pgoff_t end_pgoff)
2281 struct radix_tree_iter iter;
2283 struct file *file = vmf->vma->vm_file;
2284 struct address_space *mapping = file->f_mapping;
2285 pgoff_t last_pgoff = start_pgoff;
2287 struct page *head, *page;
2290 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter,
2292 if (iter.index > end_pgoff)
2295 page = radix_tree_deref_slot(slot);
2296 if (unlikely(!page))
2298 if (radix_tree_exception(page)) {
2299 if (radix_tree_deref_retry(page)) {
2300 slot = radix_tree_iter_retry(&iter);
2306 head = compound_head(page);
2307 if (!page_cache_get_speculative(head))
2310 /* The page was split under us? */
2311 if (compound_head(page) != head) {
2316 /* Has the page moved? */
2317 if (unlikely(page != *slot)) {
2322 if (!PageUptodate(page) ||
2323 PageReadahead(page) ||
2326 if (!trylock_page(page))
2329 if (page->mapping != mapping || !PageUptodate(page))
2332 size = round_up(i_size_read(mapping->host), PAGE_SIZE);
2333 if (page->index >= size >> PAGE_SHIFT)
2336 if (file->f_ra.mmap_miss > 0)
2337 file->f_ra.mmap_miss--;
2339 vmf->address += (iter.index - last_pgoff) << PAGE_SHIFT;
2341 vmf->pte += iter.index - last_pgoff;
2342 last_pgoff = iter.index;
2343 if (alloc_set_pte(vmf, NULL, page))
2352 /* Huge page is mapped? No need to proceed. */
2353 if (pmd_trans_huge(*vmf->pmd))
2355 if (iter.index == end_pgoff)
2360 EXPORT_SYMBOL(filemap_map_pages);
2362 int filemap_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
2364 struct page *page = vmf->page;
2365 struct inode *inode = file_inode(vma->vm_file);
2366 int ret = VM_FAULT_LOCKED;
2368 sb_start_pagefault(inode->i_sb);
2369 file_update_time(vma->vm_file);
2371 if (page->mapping != inode->i_mapping) {
2373 ret = VM_FAULT_NOPAGE;
2377 * We mark the page dirty already here so that when freeze is in
2378 * progress, we are guaranteed that writeback during freezing will
2379 * see the dirty page and writeprotect it again.
2381 set_page_dirty(page);
2382 wait_for_stable_page(page);
2384 sb_end_pagefault(inode->i_sb);
2387 EXPORT_SYMBOL(filemap_page_mkwrite);
2389 const struct vm_operations_struct generic_file_vm_ops = {
2390 .fault = filemap_fault,
2391 .map_pages = filemap_map_pages,
2392 .page_mkwrite = filemap_page_mkwrite,
2395 /* This is used for a general mmap of a disk file */
2397 int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
2399 struct address_space *mapping = file->f_mapping;
2401 if (!mapping->a_ops->readpage)
2403 file_accessed(file);
2404 vma->vm_ops = &generic_file_vm_ops;
2409 * This is for filesystems which do not implement ->writepage.
2411 int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
2413 if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE))
2415 return generic_file_mmap(file, vma);
2418 int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
2422 int generic_file_readonly_mmap(struct file * file, struct vm_area_struct * vma)
2426 #endif /* CONFIG_MMU */
2428 EXPORT_SYMBOL(generic_file_mmap);
2429 EXPORT_SYMBOL(generic_file_readonly_mmap);
2431 static struct page *wait_on_page_read(struct page *page)
2433 if (!IS_ERR(page)) {
2434 wait_on_page_locked(page);
2435 if (!PageUptodate(page)) {
2437 page = ERR_PTR(-EIO);
2443 static struct page *do_read_cache_page(struct address_space *mapping,
2445 int (*filler)(void *, struct page *),
2452 page = find_get_page(mapping, index);
2454 page = __page_cache_alloc(gfp | __GFP_COLD);
2456 return ERR_PTR(-ENOMEM);
2457 err = add_to_page_cache_lru(page, mapping, index, gfp);
2458 if (unlikely(err)) {
2462 /* Presumably ENOMEM for radix tree node */
2463 return ERR_PTR(err);
2467 err = filler(data, page);
2470 return ERR_PTR(err);
2473 page = wait_on_page_read(page);
2478 if (PageUptodate(page))
2482 * Page is not up to date and may be locked due one of the following
2483 * case a: Page is being filled and the page lock is held
2484 * case b: Read/write error clearing the page uptodate status
2485 * case c: Truncation in progress (page locked)
2486 * case d: Reclaim in progress
2488 * Case a, the page will be up to date when the page is unlocked.
2489 * There is no need to serialise on the page lock here as the page
2490 * is pinned so the lock gives no additional protection. Even if the
2491 * the page is truncated, the data is still valid if PageUptodate as
2492 * it's a race vs truncate race.
2493 * Case b, the page will not be up to date
2494 * Case c, the page may be truncated but in itself, the data may still
2495 * be valid after IO completes as it's a read vs truncate race. The
2496 * operation must restart if the page is not uptodate on unlock but
2497 * otherwise serialising on page lock to stabilise the mapping gives
2498 * no additional guarantees to the caller as the page lock is
2499 * released before return.
2500 * Case d, similar to truncation. If reclaim holds the page lock, it
2501 * will be a race with remove_mapping that determines if the mapping
2502 * is valid on unlock but otherwise the data is valid and there is
2503 * no need to serialise with page lock.
2505 * As the page lock gives no additional guarantee, we optimistically
2506 * wait on the page to be unlocked and check if it's up to date and
2507 * use the page if it is. Otherwise, the page lock is required to
2508 * distinguish between the different cases. The motivation is that we
2509 * avoid spurious serialisations and wakeups when multiple processes
2510 * wait on the same page for IO to complete.
2512 wait_on_page_locked(page);
2513 if (PageUptodate(page))
2516 /* Distinguish between all the cases under the safety of the lock */
2519 /* Case c or d, restart the operation */
2520 if (!page->mapping) {
2526 /* Someone else locked and filled the page in a very small window */
2527 if (PageUptodate(page)) {
2534 mark_page_accessed(page);
2539 * read_cache_page - read into page cache, fill it if needed
2540 * @mapping: the page's address_space
2541 * @index: the page index
2542 * @filler: function to perform the read
2543 * @data: first arg to filler(data, page) function, often left as NULL
2545 * Read into the page cache. If a page already exists, and PageUptodate() is
2546 * not set, try to fill the page and wait for it to become unlocked.
2548 * If the page does not get brought uptodate, return -EIO.
2550 struct page *read_cache_page(struct address_space *mapping,
2552 int (*filler)(void *, struct page *),
2555 return do_read_cache_page(mapping, index, filler, data, mapping_gfp_mask(mapping));
2557 EXPORT_SYMBOL(read_cache_page);
2560 * read_cache_page_gfp - read into page cache, using specified page allocation flags.
2561 * @mapping: the page's address_space
2562 * @index: the page index
2563 * @gfp: the page allocator flags to use if allocating
2565 * This is the same as "read_mapping_page(mapping, index, NULL)", but with
2566 * any new page allocations done using the specified allocation flags.
2568 * If the page does not get brought uptodate, return -EIO.
2570 struct page *read_cache_page_gfp(struct address_space *mapping,
2574 filler_t *filler = (filler_t *)mapping->a_ops->readpage;
2576 return do_read_cache_page(mapping, index, filler, NULL, gfp);
2578 EXPORT_SYMBOL(read_cache_page_gfp);
2581 * Performs necessary checks before doing a write
2583 * Can adjust writing position or amount of bytes to write.
2584 * Returns appropriate error code that caller should return or
2585 * zero in case that write should be allowed.
2587 inline ssize_t generic_write_checks(struct kiocb *iocb, struct iov_iter *from)
2589 struct file *file = iocb->ki_filp;
2590 struct inode *inode = file->f_mapping->host;
2591 unsigned long limit = rlimit(RLIMIT_FSIZE);
2594 if (!iov_iter_count(from))
2597 /* FIXME: this is for backwards compatibility with 2.4 */
2598 if (iocb->ki_flags & IOCB_APPEND)
2599 iocb->ki_pos = i_size_read(inode);
2603 if (limit != RLIM_INFINITY) {
2604 if (iocb->ki_pos >= limit) {
2605 send_sig(SIGXFSZ, current, 0);
2608 iov_iter_truncate(from, limit - (unsigned long)pos);
2614 if (unlikely(pos + iov_iter_count(from) > MAX_NON_LFS &&
2615 !(file->f_flags & O_LARGEFILE))) {
2616 if (pos >= MAX_NON_LFS)
2618 iov_iter_truncate(from, MAX_NON_LFS - (unsigned long)pos);
2622 * Are we about to exceed the fs block limit ?
2624 * If we have written data it becomes a short write. If we have
2625 * exceeded without writing data we send a signal and return EFBIG.
2626 * Linus frestrict idea will clean these up nicely..
2628 if (unlikely(pos >= inode->i_sb->s_maxbytes))
2631 iov_iter_truncate(from, inode->i_sb->s_maxbytes - pos);
2632 return iov_iter_count(from);
2634 EXPORT_SYMBOL(generic_write_checks);
2636 int pagecache_write_begin(struct file *file, struct address_space *mapping,
2637 loff_t pos, unsigned len, unsigned flags,
2638 struct page **pagep, void **fsdata)
2640 const struct address_space_operations *aops = mapping->a_ops;
2642 return aops->write_begin(file, mapping, pos, len, flags,
2645 EXPORT_SYMBOL(pagecache_write_begin);
2647 int pagecache_write_end(struct file *file, struct address_space *mapping,
2648 loff_t pos, unsigned len, unsigned copied,
2649 struct page *page, void *fsdata)
2651 const struct address_space_operations *aops = mapping->a_ops;
2653 return aops->write_end(file, mapping, pos, len, copied, page, fsdata);
2655 EXPORT_SYMBOL(pagecache_write_end);
2658 generic_file_direct_write(struct kiocb *iocb, struct iov_iter *from)
2660 struct file *file = iocb->ki_filp;
2661 struct address_space *mapping = file->f_mapping;
2662 struct inode *inode = mapping->host;
2663 loff_t pos = iocb->ki_pos;
2667 struct iov_iter data;
2669 write_len = iov_iter_count(from);
2670 end = (pos + write_len - 1) >> PAGE_SHIFT;
2672 written = filemap_write_and_wait_range(mapping, pos, pos + write_len - 1);
2677 * After a write we want buffered reads to be sure to go to disk to get
2678 * the new data. We invalidate clean cached page from the region we're
2679 * about to write. We do this *before* the write so that we can return
2680 * without clobbering -EIOCBQUEUED from ->direct_IO().
2682 if (mapping->nrpages) {
2683 written = invalidate_inode_pages2_range(mapping,
2684 pos >> PAGE_SHIFT, end);
2686 * If a page can not be invalidated, return 0 to fall back
2687 * to buffered write.
2690 if (written == -EBUSY)
2697 written = mapping->a_ops->direct_IO(iocb, &data);
2700 * Finally, try again to invalidate clean pages which might have been
2701 * cached by non-direct readahead, or faulted in by get_user_pages()
2702 * if the source of the write was an mmap'ed region of the file
2703 * we're writing. Either one is a pretty crazy thing to do,
2704 * so we don't support it 100%. If this invalidation
2705 * fails, tough, the write still worked...
2707 if (mapping->nrpages) {
2708 invalidate_inode_pages2_range(mapping,
2709 pos >> PAGE_SHIFT, end);
2714 iov_iter_advance(from, written);
2715 if (pos > i_size_read(inode) && !S_ISBLK(inode->i_mode)) {
2716 i_size_write(inode, pos);
2717 mark_inode_dirty(inode);
2724 EXPORT_SYMBOL(generic_file_direct_write);
2727 * Find or create a page at the given pagecache position. Return the locked
2728 * page. This function is specifically for buffered writes.
2730 struct page *grab_cache_page_write_begin(struct address_space *mapping,
2731 pgoff_t index, unsigned flags)
2734 int fgp_flags = FGP_LOCK|FGP_WRITE|FGP_CREAT;
2736 if (flags & AOP_FLAG_NOFS)
2737 fgp_flags |= FGP_NOFS;
2739 page = pagecache_get_page(mapping, index, fgp_flags,
2740 mapping_gfp_mask(mapping));
2742 wait_for_stable_page(page);
2746 EXPORT_SYMBOL(grab_cache_page_write_begin);
2748 ssize_t generic_perform_write(struct file *file,
2749 struct iov_iter *i, loff_t pos)
2751 struct address_space *mapping = file->f_mapping;
2752 const struct address_space_operations *a_ops = mapping->a_ops;
2754 ssize_t written = 0;
2755 unsigned int flags = 0;
2758 * Copies from kernel address space cannot fail (NFSD is a big user).
2760 if (!iter_is_iovec(i))
2761 flags |= AOP_FLAG_UNINTERRUPTIBLE;
2765 unsigned long offset; /* Offset into pagecache page */
2766 unsigned long bytes; /* Bytes to write to page */
2767 size_t copied; /* Bytes copied from user */
2770 offset = (pos & (PAGE_SIZE - 1));
2771 bytes = min_t(unsigned long, PAGE_SIZE - offset,
2776 * Bring in the user page that we will copy from _first_.
2777 * Otherwise there's a nasty deadlock on copying from the
2778 * same page as we're writing to, without it being marked
2781 * Not only is this an optimisation, but it is also required
2782 * to check that the address is actually valid, when atomic
2783 * usercopies are used, below.
2785 if (unlikely(iov_iter_fault_in_readable(i, bytes))) {
2790 if (fatal_signal_pending(current)) {
2795 status = a_ops->write_begin(file, mapping, pos, bytes, flags,
2797 if (unlikely(status < 0))
2800 if (mapping_writably_mapped(mapping))
2801 flush_dcache_page(page);
2803 copied = iov_iter_copy_from_user_atomic(page, i, offset, bytes);
2804 flush_dcache_page(page);
2806 status = a_ops->write_end(file, mapping, pos, bytes, copied,
2808 if (unlikely(status < 0))
2814 iov_iter_advance(i, copied);
2815 if (unlikely(copied == 0)) {
2817 * If we were unable to copy any data at all, we must
2818 * fall back to a single segment length write.
2820 * If we didn't fallback here, we could livelock
2821 * because not all segments in the iov can be copied at
2822 * once without a pagefault.
2824 bytes = min_t(unsigned long, PAGE_SIZE - offset,
2825 iov_iter_single_seg_count(i));
2831 balance_dirty_pages_ratelimited(mapping);
2832 } while (iov_iter_count(i));
2834 return written ? written : status;
2836 EXPORT_SYMBOL(generic_perform_write);
2839 * __generic_file_write_iter - write data to a file
2840 * @iocb: IO state structure (file, offset, etc.)
2841 * @from: iov_iter with data to write
2843 * This function does all the work needed for actually writing data to a
2844 * file. It does all basic checks, removes SUID from the file, updates
2845 * modification times and calls proper subroutines depending on whether we
2846 * do direct IO or a standard buffered write.
2848 * It expects i_mutex to be grabbed unless we work on a block device or similar
2849 * object which does not need locking at all.
2851 * This function does *not* take care of syncing data in case of O_SYNC write.
2852 * A caller has to handle it. This is mainly due to the fact that we want to
2853 * avoid syncing under i_mutex.
2855 ssize_t __generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
2857 struct file *file = iocb->ki_filp;
2858 struct address_space * mapping = file->f_mapping;
2859 struct inode *inode = mapping->host;
2860 ssize_t written = 0;
2864 /* We can write back this queue in page reclaim */
2865 current->backing_dev_info = inode_to_bdi(inode);
2866 err = file_remove_privs(file);
2870 err = file_update_time(file);
2874 if (iocb->ki_flags & IOCB_DIRECT) {
2875 loff_t pos, endbyte;
2877 written = generic_file_direct_write(iocb, from);
2879 * If the write stopped short of completing, fall back to
2880 * buffered writes. Some filesystems do this for writes to
2881 * holes, for example. For DAX files, a buffered write will
2882 * not succeed (even if it did, DAX does not handle dirty
2883 * page-cache pages correctly).
2885 if (written < 0 || !iov_iter_count(from) || IS_DAX(inode))
2888 status = generic_perform_write(file, from, pos = iocb->ki_pos);
2890 * If generic_perform_write() returned a synchronous error
2891 * then we want to return the number of bytes which were
2892 * direct-written, or the error code if that was zero. Note
2893 * that this differs from normal direct-io semantics, which
2894 * will return -EFOO even if some bytes were written.
2896 if (unlikely(status < 0)) {
2901 * We need to ensure that the page cache pages are written to
2902 * disk and invalidated to preserve the expected O_DIRECT
2905 endbyte = pos + status - 1;
2906 err = filemap_write_and_wait_range(mapping, pos, endbyte);
2908 iocb->ki_pos = endbyte + 1;
2910 invalidate_mapping_pages(mapping,
2912 endbyte >> PAGE_SHIFT);
2915 * We don't know how much we wrote, so just return
2916 * the number of bytes which were direct-written
2920 written = generic_perform_write(file, from, iocb->ki_pos);
2921 if (likely(written > 0))
2922 iocb->ki_pos += written;
2925 current->backing_dev_info = NULL;
2926 return written ? written : err;
2928 EXPORT_SYMBOL(__generic_file_write_iter);
2931 * generic_file_write_iter - write data to a file
2932 * @iocb: IO state structure
2933 * @from: iov_iter with data to write
2935 * This is a wrapper around __generic_file_write_iter() to be used by most
2936 * filesystems. It takes care of syncing the file in case of O_SYNC file
2937 * and acquires i_mutex as needed.
2939 ssize_t generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
2941 struct file *file = iocb->ki_filp;
2942 struct inode *inode = file->f_mapping->host;
2946 ret = generic_write_checks(iocb, from);
2948 ret = __generic_file_write_iter(iocb, from);
2949 inode_unlock(inode);
2952 ret = generic_write_sync(iocb, ret);
2955 EXPORT_SYMBOL(generic_file_write_iter);
2958 * try_to_release_page() - release old fs-specific metadata on a page
2960 * @page: the page which the kernel is trying to free
2961 * @gfp_mask: memory allocation flags (and I/O mode)
2963 * The address_space is to try to release any data against the page
2964 * (presumably at page->private). If the release was successful, return `1'.
2965 * Otherwise return zero.
2967 * This may also be called if PG_fscache is set on a page, indicating that the
2968 * page is known to the local caching routines.
2970 * The @gfp_mask argument specifies whether I/O may be performed to release
2971 * this page (__GFP_IO), and whether the call may block (__GFP_RECLAIM & __GFP_FS).
2974 int try_to_release_page(struct page *page, gfp_t gfp_mask)
2976 struct address_space * const mapping = page->mapping;
2978 BUG_ON(!PageLocked(page));
2979 if (PageWriteback(page))
2982 if (mapping && mapping->a_ops->releasepage)
2983 return mapping->a_ops->releasepage(page, gfp_mask);
2984 return try_to_free_buffers(page);
2987 EXPORT_SYMBOL(try_to_release_page);