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 static 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);
825 static void wake_up_page(struct page *page, int bit)
827 if (!PageWaiters(page))
829 wake_up_page_bit(page, bit);
832 static inline int wait_on_page_bit_common(wait_queue_head_t *q,
833 struct page *page, int bit_nr, int state, bool lock)
835 struct wait_page_queue wait_page;
836 wait_queue_t *wait = &wait_page.wait;
840 wait->func = wake_page_function;
841 wait_page.page = page;
842 wait_page.bit_nr = bit_nr;
845 spin_lock_irq(&q->lock);
847 if (likely(list_empty(&wait->task_list))) {
849 __add_wait_queue_tail_exclusive(q, wait);
851 __add_wait_queue(q, wait);
852 SetPageWaiters(page);
855 set_current_state(state);
857 spin_unlock_irq(&q->lock);
859 if (likely(test_bit(bit_nr, &page->flags))) {
861 if (unlikely(signal_pending_state(state, current))) {
868 if (!test_and_set_bit_lock(bit_nr, &page->flags))
871 if (!test_bit(bit_nr, &page->flags))
876 finish_wait(q, wait);
879 * A signal could leave PageWaiters set. Clearing it here if
880 * !waitqueue_active would be possible (by open-coding finish_wait),
881 * but still fail to catch it in the case of wait hash collision. We
882 * already can fail to clear wait hash collision cases, so don't
883 * bother with signals either.
889 void wait_on_page_bit(struct page *page, int bit_nr)
891 wait_queue_head_t *q = page_waitqueue(page);
892 wait_on_page_bit_common(q, page, bit_nr, TASK_UNINTERRUPTIBLE, false);
894 EXPORT_SYMBOL(wait_on_page_bit);
896 int wait_on_page_bit_killable(struct page *page, int bit_nr)
898 wait_queue_head_t *q = page_waitqueue(page);
899 return wait_on_page_bit_common(q, page, bit_nr, TASK_KILLABLE, false);
903 * add_page_wait_queue - Add an arbitrary waiter to a page's wait queue
904 * @page: Page defining the wait queue of interest
905 * @waiter: Waiter to add to the queue
907 * Add an arbitrary @waiter to the wait queue for the nominated @page.
909 void add_page_wait_queue(struct page *page, wait_queue_t *waiter)
911 wait_queue_head_t *q = page_waitqueue(page);
914 spin_lock_irqsave(&q->lock, flags);
915 __add_wait_queue(q, waiter);
916 SetPageWaiters(page);
917 spin_unlock_irqrestore(&q->lock, flags);
919 EXPORT_SYMBOL_GPL(add_page_wait_queue);
921 #ifndef clear_bit_unlock_is_negative_byte
924 * PG_waiters is the high bit in the same byte as PG_lock.
926 * On x86 (and on many other architectures), we can clear PG_lock and
927 * test the sign bit at the same time. But if the architecture does
928 * not support that special operation, we just do this all by hand
931 * The read of PG_waiters has to be after (or concurrently with) PG_locked
932 * being cleared, but a memory barrier should be unneccssary since it is
933 * in the same byte as PG_locked.
935 static inline bool clear_bit_unlock_is_negative_byte(long nr, volatile void *mem)
937 clear_bit_unlock(nr, mem);
938 /* smp_mb__after_atomic(); */
939 return test_bit(PG_waiters, mem);
945 * unlock_page - unlock a locked page
948 * Unlocks the page and wakes up sleepers in ___wait_on_page_locked().
949 * Also wakes sleepers in wait_on_page_writeback() because the wakeup
950 * mechanism between PageLocked pages and PageWriteback pages is shared.
951 * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep.
953 * Note that this depends on PG_waiters being the sign bit in the byte
954 * that contains PG_locked - thus the BUILD_BUG_ON(). That allows us to
955 * clear the PG_locked bit and test PG_waiters at the same time fairly
956 * portably (architectures that do LL/SC can test any bit, while x86 can
957 * test the sign bit).
959 void unlock_page(struct page *page)
961 BUILD_BUG_ON(PG_waiters != 7);
962 page = compound_head(page);
963 VM_BUG_ON_PAGE(!PageLocked(page), page);
964 if (clear_bit_unlock_is_negative_byte(PG_locked, &page->flags))
965 wake_up_page_bit(page, PG_locked);
967 EXPORT_SYMBOL(unlock_page);
970 * end_page_writeback - end writeback against a page
973 void end_page_writeback(struct page *page)
976 * TestClearPageReclaim could be used here but it is an atomic
977 * operation and overkill in this particular case. Failing to
978 * shuffle a page marked for immediate reclaim is too mild to
979 * justify taking an atomic operation penalty at the end of
980 * ever page writeback.
982 if (PageReclaim(page)) {
983 ClearPageReclaim(page);
984 rotate_reclaimable_page(page);
987 if (!test_clear_page_writeback(page))
990 smp_mb__after_atomic();
991 wake_up_page(page, PG_writeback);
993 EXPORT_SYMBOL(end_page_writeback);
996 * After completing I/O on a page, call this routine to update the page
997 * flags appropriately
999 void page_endio(struct page *page, bool is_write, int err)
1003 SetPageUptodate(page);
1005 ClearPageUptodate(page);
1011 struct address_space *mapping;
1014 mapping = page_mapping(page);
1016 mapping_set_error(mapping, err);
1018 end_page_writeback(page);
1021 EXPORT_SYMBOL_GPL(page_endio);
1024 * __lock_page - get a lock on the page, assuming we need to sleep to get it
1025 * @__page: the page to lock
1027 void __lock_page(struct page *__page)
1029 struct page *page = compound_head(__page);
1030 wait_queue_head_t *q = page_waitqueue(page);
1031 wait_on_page_bit_common(q, page, PG_locked, TASK_UNINTERRUPTIBLE, true);
1033 EXPORT_SYMBOL(__lock_page);
1035 int __lock_page_killable(struct page *__page)
1037 struct page *page = compound_head(__page);
1038 wait_queue_head_t *q = page_waitqueue(page);
1039 return wait_on_page_bit_common(q, page, PG_locked, TASK_KILLABLE, true);
1041 EXPORT_SYMBOL_GPL(__lock_page_killable);
1045 * 1 - page is locked; mmap_sem is still held.
1046 * 0 - page is not locked.
1047 * mmap_sem has been released (up_read()), unless flags had both
1048 * FAULT_FLAG_ALLOW_RETRY and FAULT_FLAG_RETRY_NOWAIT set, in
1049 * which case mmap_sem is still held.
1051 * If neither ALLOW_RETRY nor KILLABLE are set, will always return 1
1052 * with the page locked and the mmap_sem unperturbed.
1054 int __lock_page_or_retry(struct page *page, struct mm_struct *mm,
1057 if (flags & FAULT_FLAG_ALLOW_RETRY) {
1059 * CAUTION! In this case, mmap_sem is not released
1060 * even though return 0.
1062 if (flags & FAULT_FLAG_RETRY_NOWAIT)
1065 up_read(&mm->mmap_sem);
1066 if (flags & FAULT_FLAG_KILLABLE)
1067 wait_on_page_locked_killable(page);
1069 wait_on_page_locked(page);
1072 if (flags & FAULT_FLAG_KILLABLE) {
1075 ret = __lock_page_killable(page);
1077 up_read(&mm->mmap_sem);
1087 * page_cache_next_hole - find the next hole (not-present entry)
1090 * @max_scan: maximum range to search
1092 * Search the set [index, min(index+max_scan-1, MAX_INDEX)] for the
1093 * lowest indexed hole.
1095 * Returns: the index of the hole if found, otherwise returns an index
1096 * outside of the set specified (in which case 'return - index >=
1097 * max_scan' will be true). In rare cases of index wrap-around, 0 will
1100 * page_cache_next_hole may be called under rcu_read_lock. However,
1101 * like radix_tree_gang_lookup, this will not atomically search a
1102 * snapshot of the tree at a single point in time. For example, if a
1103 * hole is created at index 5, then subsequently a hole is created at
1104 * index 10, page_cache_next_hole covering both indexes may return 10
1105 * if called under rcu_read_lock.
1107 pgoff_t page_cache_next_hole(struct address_space *mapping,
1108 pgoff_t index, unsigned long max_scan)
1112 for (i = 0; i < max_scan; i++) {
1115 page = radix_tree_lookup(&mapping->page_tree, index);
1116 if (!page || radix_tree_exceptional_entry(page))
1125 EXPORT_SYMBOL(page_cache_next_hole);
1128 * page_cache_prev_hole - find the prev hole (not-present entry)
1131 * @max_scan: maximum range to search
1133 * Search backwards in the range [max(index-max_scan+1, 0), index] for
1136 * Returns: the index of the hole if found, otherwise returns an index
1137 * outside of the set specified (in which case 'index - return >=
1138 * max_scan' will be true). In rare cases of wrap-around, ULONG_MAX
1141 * page_cache_prev_hole may be called under rcu_read_lock. However,
1142 * like radix_tree_gang_lookup, this will not atomically search a
1143 * snapshot of the tree at a single point in time. For example, if a
1144 * hole is created at index 10, then subsequently a hole is created at
1145 * index 5, page_cache_prev_hole covering both indexes may return 5 if
1146 * called under rcu_read_lock.
1148 pgoff_t page_cache_prev_hole(struct address_space *mapping,
1149 pgoff_t index, unsigned long max_scan)
1153 for (i = 0; i < max_scan; i++) {
1156 page = radix_tree_lookup(&mapping->page_tree, index);
1157 if (!page || radix_tree_exceptional_entry(page))
1160 if (index == ULONG_MAX)
1166 EXPORT_SYMBOL(page_cache_prev_hole);
1169 * find_get_entry - find and get a page cache entry
1170 * @mapping: the address_space to search
1171 * @offset: the page cache index
1173 * Looks up the page cache slot at @mapping & @offset. If there is a
1174 * page cache page, it is returned with an increased refcount.
1176 * If the slot holds a shadow entry of a previously evicted page, or a
1177 * swap entry from shmem/tmpfs, it is returned.
1179 * Otherwise, %NULL is returned.
1181 struct page *find_get_entry(struct address_space *mapping, pgoff_t offset)
1184 struct page *head, *page;
1189 pagep = radix_tree_lookup_slot(&mapping->page_tree, offset);
1191 page = radix_tree_deref_slot(pagep);
1192 if (unlikely(!page))
1194 if (radix_tree_exception(page)) {
1195 if (radix_tree_deref_retry(page))
1198 * A shadow entry of a recently evicted page,
1199 * or a swap entry from shmem/tmpfs. Return
1200 * it without attempting to raise page count.
1205 head = compound_head(page);
1206 if (!page_cache_get_speculative(head))
1209 /* The page was split under us? */
1210 if (compound_head(page) != head) {
1216 * Has the page moved?
1217 * This is part of the lockless pagecache protocol. See
1218 * include/linux/pagemap.h for details.
1220 if (unlikely(page != *pagep)) {
1230 EXPORT_SYMBOL(find_get_entry);
1233 * find_lock_entry - locate, pin and lock a page cache entry
1234 * @mapping: the address_space to search
1235 * @offset: the page cache index
1237 * Looks up the page cache slot at @mapping & @offset. If there is a
1238 * page cache page, it is returned locked and with an increased
1241 * If the slot holds a shadow entry of a previously evicted page, or a
1242 * swap entry from shmem/tmpfs, it is returned.
1244 * Otherwise, %NULL is returned.
1246 * find_lock_entry() may sleep.
1248 struct page *find_lock_entry(struct address_space *mapping, pgoff_t offset)
1253 page = find_get_entry(mapping, offset);
1254 if (page && !radix_tree_exception(page)) {
1256 /* Has the page been truncated? */
1257 if (unlikely(page_mapping(page) != mapping)) {
1262 VM_BUG_ON_PAGE(page_to_pgoff(page) != offset, page);
1266 EXPORT_SYMBOL(find_lock_entry);
1269 * pagecache_get_page - find and get a page reference
1270 * @mapping: the address_space to search
1271 * @offset: the page index
1272 * @fgp_flags: PCG flags
1273 * @gfp_mask: gfp mask to use for the page cache data page allocation
1275 * Looks up the page cache slot at @mapping & @offset.
1277 * PCG flags modify how the page is returned.
1279 * FGP_ACCESSED: the page will be marked accessed
1280 * FGP_LOCK: Page is return locked
1281 * FGP_CREAT: If page is not present then a new page is allocated using
1282 * @gfp_mask and added to the page cache and the VM's LRU
1283 * list. The page is returned locked and with an increased
1284 * refcount. Otherwise, %NULL is returned.
1286 * If FGP_LOCK or FGP_CREAT are specified then the function may sleep even
1287 * if the GFP flags specified for FGP_CREAT are atomic.
1289 * If there is a page cache page, it is returned with an increased refcount.
1291 struct page *pagecache_get_page(struct address_space *mapping, pgoff_t offset,
1292 int fgp_flags, gfp_t gfp_mask)
1297 page = find_get_entry(mapping, offset);
1298 if (radix_tree_exceptional_entry(page))
1303 if (fgp_flags & FGP_LOCK) {
1304 if (fgp_flags & FGP_NOWAIT) {
1305 if (!trylock_page(page)) {
1313 /* Has the page been truncated? */
1314 if (unlikely(page->mapping != mapping)) {
1319 VM_BUG_ON_PAGE(page->index != offset, page);
1322 if (page && (fgp_flags & FGP_ACCESSED))
1323 mark_page_accessed(page);
1326 if (!page && (fgp_flags & FGP_CREAT)) {
1328 if ((fgp_flags & FGP_WRITE) && mapping_cap_account_dirty(mapping))
1329 gfp_mask |= __GFP_WRITE;
1330 if (fgp_flags & FGP_NOFS)
1331 gfp_mask &= ~__GFP_FS;
1333 page = __page_cache_alloc(gfp_mask);
1337 if (WARN_ON_ONCE(!(fgp_flags & FGP_LOCK)))
1338 fgp_flags |= FGP_LOCK;
1340 /* Init accessed so avoid atomic mark_page_accessed later */
1341 if (fgp_flags & FGP_ACCESSED)
1342 __SetPageReferenced(page);
1344 err = add_to_page_cache_lru(page, mapping, offset,
1345 gfp_mask & GFP_RECLAIM_MASK);
1346 if (unlikely(err)) {
1356 EXPORT_SYMBOL(pagecache_get_page);
1359 * find_get_entries - gang pagecache lookup
1360 * @mapping: The address_space to search
1361 * @start: The starting page cache index
1362 * @nr_entries: The maximum number of entries
1363 * @entries: Where the resulting entries are placed
1364 * @indices: The cache indices corresponding to the entries in @entries
1366 * find_get_entries() will search for and return a group of up to
1367 * @nr_entries entries in the mapping. The entries are placed at
1368 * @entries. find_get_entries() takes a reference against any actual
1371 * The search returns a group of mapping-contiguous page cache entries
1372 * with ascending indexes. There may be holes in the indices due to
1373 * not-present pages.
1375 * Any shadow entries of evicted pages, or swap entries from
1376 * shmem/tmpfs, are included in the returned array.
1378 * find_get_entries() returns the number of pages and shadow entries
1381 unsigned find_get_entries(struct address_space *mapping,
1382 pgoff_t start, unsigned int nr_entries,
1383 struct page **entries, pgoff_t *indices)
1386 unsigned int ret = 0;
1387 struct radix_tree_iter iter;
1393 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) {
1394 struct page *head, *page;
1396 page = radix_tree_deref_slot(slot);
1397 if (unlikely(!page))
1399 if (radix_tree_exception(page)) {
1400 if (radix_tree_deref_retry(page)) {
1401 slot = radix_tree_iter_retry(&iter);
1405 * A shadow entry of a recently evicted page, a swap
1406 * entry from shmem/tmpfs or a DAX entry. Return it
1407 * without attempting to raise page count.
1412 head = compound_head(page);
1413 if (!page_cache_get_speculative(head))
1416 /* The page was split under us? */
1417 if (compound_head(page) != head) {
1422 /* Has the page moved? */
1423 if (unlikely(page != *slot)) {
1428 indices[ret] = iter.index;
1429 entries[ret] = page;
1430 if (++ret == nr_entries)
1438 * find_get_pages - gang pagecache lookup
1439 * @mapping: The address_space to search
1440 * @start: The starting page index
1441 * @nr_pages: The maximum number of pages
1442 * @pages: Where the resulting pages are placed
1444 * find_get_pages() will search for and return a group of up to
1445 * @nr_pages pages in the mapping. The pages are placed at @pages.
1446 * find_get_pages() takes a reference against the returned pages.
1448 * The search returns a group of mapping-contiguous pages with ascending
1449 * indexes. There may be holes in the indices due to not-present pages.
1451 * find_get_pages() returns the number of pages which were found.
1453 unsigned find_get_pages(struct address_space *mapping, pgoff_t start,
1454 unsigned int nr_pages, struct page **pages)
1456 struct radix_tree_iter iter;
1460 if (unlikely(!nr_pages))
1464 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) {
1465 struct page *head, *page;
1467 page = radix_tree_deref_slot(slot);
1468 if (unlikely(!page))
1471 if (radix_tree_exception(page)) {
1472 if (radix_tree_deref_retry(page)) {
1473 slot = radix_tree_iter_retry(&iter);
1477 * A shadow entry of a recently evicted page,
1478 * or a swap entry from shmem/tmpfs. Skip
1484 head = compound_head(page);
1485 if (!page_cache_get_speculative(head))
1488 /* The page was split under us? */
1489 if (compound_head(page) != head) {
1494 /* Has the page moved? */
1495 if (unlikely(page != *slot)) {
1501 if (++ret == nr_pages)
1510 * find_get_pages_contig - gang contiguous pagecache lookup
1511 * @mapping: The address_space to search
1512 * @index: The starting page index
1513 * @nr_pages: The maximum number of pages
1514 * @pages: Where the resulting pages are placed
1516 * find_get_pages_contig() works exactly like find_get_pages(), except
1517 * that the returned number of pages are guaranteed to be contiguous.
1519 * find_get_pages_contig() returns the number of pages which were found.
1521 unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index,
1522 unsigned int nr_pages, struct page **pages)
1524 struct radix_tree_iter iter;
1526 unsigned int ret = 0;
1528 if (unlikely(!nr_pages))
1532 radix_tree_for_each_contig(slot, &mapping->page_tree, &iter, index) {
1533 struct page *head, *page;
1535 page = radix_tree_deref_slot(slot);
1536 /* The hole, there no reason to continue */
1537 if (unlikely(!page))
1540 if (radix_tree_exception(page)) {
1541 if (radix_tree_deref_retry(page)) {
1542 slot = radix_tree_iter_retry(&iter);
1546 * A shadow entry of a recently evicted page,
1547 * or a swap entry from shmem/tmpfs. Stop
1548 * looking for contiguous pages.
1553 head = compound_head(page);
1554 if (!page_cache_get_speculative(head))
1557 /* The page was split under us? */
1558 if (compound_head(page) != head) {
1563 /* Has the page moved? */
1564 if (unlikely(page != *slot)) {
1570 * must check mapping and index after taking the ref.
1571 * otherwise we can get both false positives and false
1572 * negatives, which is just confusing to the caller.
1574 if (page->mapping == NULL || page_to_pgoff(page) != iter.index) {
1580 if (++ret == nr_pages)
1586 EXPORT_SYMBOL(find_get_pages_contig);
1589 * find_get_pages_tag - find and return pages that match @tag
1590 * @mapping: the address_space to search
1591 * @index: the starting page index
1592 * @tag: the tag index
1593 * @nr_pages: the maximum number of pages
1594 * @pages: where the resulting pages are placed
1596 * Like find_get_pages, except we only return pages which are tagged with
1597 * @tag. We update @index to index the next page for the traversal.
1599 unsigned find_get_pages_tag(struct address_space *mapping, pgoff_t *index,
1600 int tag, unsigned int nr_pages, struct page **pages)
1602 struct radix_tree_iter iter;
1606 if (unlikely(!nr_pages))
1610 radix_tree_for_each_tagged(slot, &mapping->page_tree,
1611 &iter, *index, tag) {
1612 struct page *head, *page;
1614 page = radix_tree_deref_slot(slot);
1615 if (unlikely(!page))
1618 if (radix_tree_exception(page)) {
1619 if (radix_tree_deref_retry(page)) {
1620 slot = radix_tree_iter_retry(&iter);
1624 * A shadow entry of a recently evicted page.
1626 * Those entries should never be tagged, but
1627 * this tree walk is lockless and the tags are
1628 * looked up in bulk, one radix tree node at a
1629 * time, so there is a sizable window for page
1630 * reclaim to evict a page we saw tagged.
1637 head = compound_head(page);
1638 if (!page_cache_get_speculative(head))
1641 /* The page was split under us? */
1642 if (compound_head(page) != head) {
1647 /* Has the page moved? */
1648 if (unlikely(page != *slot)) {
1654 if (++ret == nr_pages)
1661 *index = pages[ret - 1]->index + 1;
1665 EXPORT_SYMBOL(find_get_pages_tag);
1668 * find_get_entries_tag - find and return entries that match @tag
1669 * @mapping: the address_space to search
1670 * @start: the starting page cache index
1671 * @tag: the tag index
1672 * @nr_entries: the maximum number of entries
1673 * @entries: where the resulting entries are placed
1674 * @indices: the cache indices corresponding to the entries in @entries
1676 * Like find_get_entries, except we only return entries which are tagged with
1679 unsigned find_get_entries_tag(struct address_space *mapping, pgoff_t start,
1680 int tag, unsigned int nr_entries,
1681 struct page **entries, pgoff_t *indices)
1684 unsigned int ret = 0;
1685 struct radix_tree_iter iter;
1691 radix_tree_for_each_tagged(slot, &mapping->page_tree,
1692 &iter, start, tag) {
1693 struct page *head, *page;
1695 page = radix_tree_deref_slot(slot);
1696 if (unlikely(!page))
1698 if (radix_tree_exception(page)) {
1699 if (radix_tree_deref_retry(page)) {
1700 slot = radix_tree_iter_retry(&iter);
1705 * A shadow entry of a recently evicted page, a swap
1706 * entry from shmem/tmpfs or a DAX entry. Return it
1707 * without attempting to raise page count.
1712 head = compound_head(page);
1713 if (!page_cache_get_speculative(head))
1716 /* The page was split under us? */
1717 if (compound_head(page) != head) {
1722 /* Has the page moved? */
1723 if (unlikely(page != *slot)) {
1728 indices[ret] = iter.index;
1729 entries[ret] = page;
1730 if (++ret == nr_entries)
1736 EXPORT_SYMBOL(find_get_entries_tag);
1739 * CD/DVDs are error prone. When a medium error occurs, the driver may fail
1740 * a _large_ part of the i/o request. Imagine the worst scenario:
1742 * ---R__________________________________________B__________
1743 * ^ reading here ^ bad block(assume 4k)
1745 * read(R) => miss => readahead(R...B) => media error => frustrating retries
1746 * => failing the whole request => read(R) => read(R+1) =>
1747 * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
1748 * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
1749 * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
1751 * It is going insane. Fix it by quickly scaling down the readahead size.
1753 static void shrink_readahead_size_eio(struct file *filp,
1754 struct file_ra_state *ra)
1760 * do_generic_file_read - generic file read routine
1761 * @filp: the file to read
1762 * @ppos: current file position
1763 * @iter: data destination
1764 * @written: already copied
1766 * This is a generic file read routine, and uses the
1767 * mapping->a_ops->readpage() function for the actual low-level stuff.
1769 * This is really ugly. But the goto's actually try to clarify some
1770 * of the logic when it comes to error handling etc.
1772 static ssize_t do_generic_file_read(struct file *filp, loff_t *ppos,
1773 struct iov_iter *iter, ssize_t written)
1775 struct address_space *mapping = filp->f_mapping;
1776 struct inode *inode = mapping->host;
1777 struct file_ra_state *ra = &filp->f_ra;
1781 unsigned long offset; /* offset into pagecache page */
1782 unsigned int prev_offset;
1785 if (unlikely(*ppos >= inode->i_sb->s_maxbytes))
1787 iov_iter_truncate(iter, inode->i_sb->s_maxbytes);
1789 index = *ppos >> PAGE_SHIFT;
1790 prev_index = ra->prev_pos >> PAGE_SHIFT;
1791 prev_offset = ra->prev_pos & (PAGE_SIZE-1);
1792 last_index = (*ppos + iter->count + PAGE_SIZE-1) >> PAGE_SHIFT;
1793 offset = *ppos & ~PAGE_MASK;
1799 unsigned long nr, ret;
1803 if (fatal_signal_pending(current)) {
1808 page = find_get_page(mapping, index);
1810 page_cache_sync_readahead(mapping,
1812 index, last_index - index);
1813 page = find_get_page(mapping, index);
1814 if (unlikely(page == NULL))
1815 goto no_cached_page;
1817 if (PageReadahead(page)) {
1818 page_cache_async_readahead(mapping,
1820 index, last_index - index);
1822 if (!PageUptodate(page)) {
1824 * See comment in do_read_cache_page on why
1825 * wait_on_page_locked is used to avoid unnecessarily
1826 * serialisations and why it's safe.
1828 error = wait_on_page_locked_killable(page);
1829 if (unlikely(error))
1830 goto readpage_error;
1831 if (PageUptodate(page))
1834 if (inode->i_blkbits == PAGE_SHIFT ||
1835 !mapping->a_ops->is_partially_uptodate)
1836 goto page_not_up_to_date;
1837 /* pipes can't handle partially uptodate pages */
1838 if (unlikely(iter->type & ITER_PIPE))
1839 goto page_not_up_to_date;
1840 if (!trylock_page(page))
1841 goto page_not_up_to_date;
1842 /* Did it get truncated before we got the lock? */
1844 goto page_not_up_to_date_locked;
1845 if (!mapping->a_ops->is_partially_uptodate(page,
1846 offset, iter->count))
1847 goto page_not_up_to_date_locked;
1852 * i_size must be checked after we know the page is Uptodate.
1854 * Checking i_size after the check allows us to calculate
1855 * the correct value for "nr", which means the zero-filled
1856 * part of the page is not copied back to userspace (unless
1857 * another truncate extends the file - this is desired though).
1860 isize = i_size_read(inode);
1861 end_index = (isize - 1) >> PAGE_SHIFT;
1862 if (unlikely(!isize || index > end_index)) {
1867 /* nr is the maximum number of bytes to copy from this page */
1869 if (index == end_index) {
1870 nr = ((isize - 1) & ~PAGE_MASK) + 1;
1878 /* If users can be writing to this page using arbitrary
1879 * virtual addresses, take care about potential aliasing
1880 * before reading the page on the kernel side.
1882 if (mapping_writably_mapped(mapping))
1883 flush_dcache_page(page);
1886 * When a sequential read accesses a page several times,
1887 * only mark it as accessed the first time.
1889 if (prev_index != index || offset != prev_offset)
1890 mark_page_accessed(page);
1894 * Ok, we have the page, and it's up-to-date, so
1895 * now we can copy it to user space...
1898 ret = copy_page_to_iter(page, offset, nr, iter);
1900 index += offset >> PAGE_SHIFT;
1901 offset &= ~PAGE_MASK;
1902 prev_offset = offset;
1906 if (!iov_iter_count(iter))
1914 page_not_up_to_date:
1915 /* Get exclusive access to the page ... */
1916 error = lock_page_killable(page);
1917 if (unlikely(error))
1918 goto readpage_error;
1920 page_not_up_to_date_locked:
1921 /* Did it get truncated before we got the lock? */
1922 if (!page->mapping) {
1928 /* Did somebody else fill it already? */
1929 if (PageUptodate(page)) {
1936 * A previous I/O error may have been due to temporary
1937 * failures, eg. multipath errors.
1938 * PG_error will be set again if readpage fails.
1940 ClearPageError(page);
1941 /* Start the actual read. The read will unlock the page. */
1942 error = mapping->a_ops->readpage(filp, page);
1944 if (unlikely(error)) {
1945 if (error == AOP_TRUNCATED_PAGE) {
1950 goto readpage_error;
1953 if (!PageUptodate(page)) {
1954 error = lock_page_killable(page);
1955 if (unlikely(error))
1956 goto readpage_error;
1957 if (!PageUptodate(page)) {
1958 if (page->mapping == NULL) {
1960 * invalidate_mapping_pages got it
1967 shrink_readahead_size_eio(filp, ra);
1969 goto readpage_error;
1977 /* UHHUH! A synchronous read error occurred. Report it */
1983 * Ok, it wasn't cached, so we need to create a new
1986 page = page_cache_alloc_cold(mapping);
1991 error = add_to_page_cache_lru(page, mapping, index,
1992 mapping_gfp_constraint(mapping, GFP_KERNEL));
1995 if (error == -EEXIST) {
2005 ra->prev_pos = prev_index;
2006 ra->prev_pos <<= PAGE_SHIFT;
2007 ra->prev_pos |= prev_offset;
2009 *ppos = ((loff_t)index << PAGE_SHIFT) + offset;
2010 file_accessed(filp);
2011 return written ? written : error;
2015 * generic_file_read_iter - generic filesystem read routine
2016 * @iocb: kernel I/O control block
2017 * @iter: destination for the data read
2019 * This is the "read_iter()" routine for all filesystems
2020 * that can use the page cache directly.
2023 generic_file_read_iter(struct kiocb *iocb, struct iov_iter *iter)
2025 struct file *file = iocb->ki_filp;
2027 size_t count = iov_iter_count(iter);
2030 goto out; /* skip atime */
2032 if (iocb->ki_flags & IOCB_DIRECT) {
2033 struct address_space *mapping = file->f_mapping;
2034 struct inode *inode = mapping->host;
2035 struct iov_iter data = *iter;
2038 size = i_size_read(inode);
2039 retval = filemap_write_and_wait_range(mapping, iocb->ki_pos,
2040 iocb->ki_pos + count - 1);
2044 file_accessed(file);
2046 retval = mapping->a_ops->direct_IO(iocb, &data);
2048 iocb->ki_pos += retval;
2049 iov_iter_advance(iter, retval);
2053 * Btrfs can have a short DIO read if we encounter
2054 * compressed extents, so if there was an error, or if
2055 * we've already read everything we wanted to, or if
2056 * there was a short read because we hit EOF, go ahead
2057 * and return. Otherwise fallthrough to buffered io for
2058 * the rest of the read. Buffered reads will not work for
2059 * DAX files, so don't bother trying.
2061 if (retval < 0 || !iov_iter_count(iter) || iocb->ki_pos >= size ||
2066 retval = do_generic_file_read(file, &iocb->ki_pos, iter, retval);
2070 EXPORT_SYMBOL(generic_file_read_iter);
2074 * page_cache_read - adds requested page to the page cache if not already there
2075 * @file: file to read
2076 * @offset: page index
2077 * @gfp_mask: memory allocation flags
2079 * This adds the requested page to the page cache if it isn't already there,
2080 * and schedules an I/O to read in its contents from disk.
2082 static int page_cache_read(struct file *file, pgoff_t offset, gfp_t gfp_mask)
2084 struct address_space *mapping = file->f_mapping;
2089 page = __page_cache_alloc(gfp_mask|__GFP_COLD);
2093 ret = add_to_page_cache_lru(page, mapping, offset, gfp_mask & GFP_KERNEL);
2095 ret = mapping->a_ops->readpage(file, page);
2096 else if (ret == -EEXIST)
2097 ret = 0; /* losing race to add is OK */
2101 } while (ret == AOP_TRUNCATED_PAGE);
2106 #define MMAP_LOTSAMISS (100)
2109 * Synchronous readahead happens when we don't even find
2110 * a page in the page cache at all.
2112 static void do_sync_mmap_readahead(struct vm_area_struct *vma,
2113 struct file_ra_state *ra,
2117 struct address_space *mapping = file->f_mapping;
2119 /* If we don't want any read-ahead, don't bother */
2120 if (vma->vm_flags & VM_RAND_READ)
2125 if (vma->vm_flags & VM_SEQ_READ) {
2126 page_cache_sync_readahead(mapping, ra, file, offset,
2131 /* Avoid banging the cache line if not needed */
2132 if (ra->mmap_miss < MMAP_LOTSAMISS * 10)
2136 * Do we miss much more than hit in this file? If so,
2137 * stop bothering with read-ahead. It will only hurt.
2139 if (ra->mmap_miss > MMAP_LOTSAMISS)
2145 ra->start = max_t(long, 0, offset - ra->ra_pages / 2);
2146 ra->size = ra->ra_pages;
2147 ra->async_size = ra->ra_pages / 4;
2148 ra_submit(ra, mapping, file);
2152 * Asynchronous readahead happens when we find the page and PG_readahead,
2153 * so we want to possibly extend the readahead further..
2155 static void do_async_mmap_readahead(struct vm_area_struct *vma,
2156 struct file_ra_state *ra,
2161 struct address_space *mapping = file->f_mapping;
2163 /* If we don't want any read-ahead, don't bother */
2164 if (vma->vm_flags & VM_RAND_READ)
2166 if (ra->mmap_miss > 0)
2168 if (PageReadahead(page))
2169 page_cache_async_readahead(mapping, ra, file,
2170 page, offset, ra->ra_pages);
2174 * filemap_fault - read in file data for page fault handling
2175 * @vmf: struct vm_fault containing details of the fault
2177 * filemap_fault() is invoked via the vma operations vector for a
2178 * mapped memory region to read in file data during a page fault.
2180 * The goto's are kind of ugly, but this streamlines the normal case of having
2181 * it in the page cache, and handles the special cases reasonably without
2182 * having a lot of duplicated code.
2184 * vma->vm_mm->mmap_sem must be held on entry.
2186 * If our return value has VM_FAULT_RETRY set, it's because
2187 * lock_page_or_retry() returned 0.
2188 * The mmap_sem has usually been released in this case.
2189 * See __lock_page_or_retry() for the exception.
2191 * If our return value does not have VM_FAULT_RETRY set, the mmap_sem
2192 * has not been released.
2194 * We never return with VM_FAULT_RETRY and a bit from VM_FAULT_ERROR set.
2196 int filemap_fault(struct vm_fault *vmf)
2199 struct file *file = vmf->vma->vm_file;
2200 struct address_space *mapping = file->f_mapping;
2201 struct file_ra_state *ra = &file->f_ra;
2202 struct inode *inode = mapping->host;
2203 pgoff_t offset = vmf->pgoff;
2208 size = round_up(i_size_read(inode), PAGE_SIZE);
2209 if (offset >= size >> PAGE_SHIFT)
2210 return VM_FAULT_SIGBUS;
2213 * Do we have something in the page cache already?
2215 page = find_get_page(mapping, offset);
2216 if (likely(page) && !(vmf->flags & FAULT_FLAG_TRIED)) {
2218 * We found the page, so try async readahead before
2219 * waiting for the lock.
2221 do_async_mmap_readahead(vmf->vma, ra, file, page, offset);
2223 /* No page in the page cache at all */
2224 do_sync_mmap_readahead(vmf->vma, ra, file, offset);
2225 count_vm_event(PGMAJFAULT);
2226 mem_cgroup_count_vm_event(vmf->vma->vm_mm, PGMAJFAULT);
2227 ret = VM_FAULT_MAJOR;
2229 page = find_get_page(mapping, offset);
2231 goto no_cached_page;
2234 if (!lock_page_or_retry(page, vmf->vma->vm_mm, vmf->flags)) {
2236 return ret | VM_FAULT_RETRY;
2239 /* Did it get truncated? */
2240 if (unlikely(page->mapping != mapping)) {
2245 VM_BUG_ON_PAGE(page->index != offset, page);
2248 * We have a locked page in the page cache, now we need to check
2249 * that it's up-to-date. If not, it is going to be due to an error.
2251 if (unlikely(!PageUptodate(page)))
2252 goto page_not_uptodate;
2255 * Found the page and have a reference on it.
2256 * We must recheck i_size under page lock.
2258 size = round_up(i_size_read(inode), PAGE_SIZE);
2259 if (unlikely(offset >= size >> PAGE_SHIFT)) {
2262 return VM_FAULT_SIGBUS;
2266 return ret | VM_FAULT_LOCKED;
2270 * We're only likely to ever get here if MADV_RANDOM is in
2273 error = page_cache_read(file, offset, vmf->gfp_mask);
2276 * The page we want has now been added to the page cache.
2277 * In the unlikely event that someone removed it in the
2278 * meantime, we'll just come back here and read it again.
2284 * An error return from page_cache_read can result if the
2285 * system is low on memory, or a problem occurs while trying
2288 if (error == -ENOMEM)
2289 return VM_FAULT_OOM;
2290 return VM_FAULT_SIGBUS;
2294 * Umm, take care of errors if the page isn't up-to-date.
2295 * Try to re-read it _once_. We do this synchronously,
2296 * because there really aren't any performance issues here
2297 * and we need to check for errors.
2299 ClearPageError(page);
2300 error = mapping->a_ops->readpage(file, page);
2302 wait_on_page_locked(page);
2303 if (!PageUptodate(page))
2308 if (!error || error == AOP_TRUNCATED_PAGE)
2311 /* Things didn't work out. Return zero to tell the mm layer so. */
2312 shrink_readahead_size_eio(file, ra);
2313 return VM_FAULT_SIGBUS;
2315 EXPORT_SYMBOL(filemap_fault);
2317 void filemap_map_pages(struct vm_fault *vmf,
2318 pgoff_t start_pgoff, pgoff_t end_pgoff)
2320 struct radix_tree_iter iter;
2322 struct file *file = vmf->vma->vm_file;
2323 struct address_space *mapping = file->f_mapping;
2324 pgoff_t last_pgoff = start_pgoff;
2326 struct page *head, *page;
2329 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter,
2331 if (iter.index > end_pgoff)
2334 page = radix_tree_deref_slot(slot);
2335 if (unlikely(!page))
2337 if (radix_tree_exception(page)) {
2338 if (radix_tree_deref_retry(page)) {
2339 slot = radix_tree_iter_retry(&iter);
2345 head = compound_head(page);
2346 if (!page_cache_get_speculative(head))
2349 /* The page was split under us? */
2350 if (compound_head(page) != head) {
2355 /* Has the page moved? */
2356 if (unlikely(page != *slot)) {
2361 if (!PageUptodate(page) ||
2362 PageReadahead(page) ||
2365 if (!trylock_page(page))
2368 if (page->mapping != mapping || !PageUptodate(page))
2371 size = round_up(i_size_read(mapping->host), PAGE_SIZE);
2372 if (page->index >= size >> PAGE_SHIFT)
2375 if (file->f_ra.mmap_miss > 0)
2376 file->f_ra.mmap_miss--;
2378 vmf->address += (iter.index - last_pgoff) << PAGE_SHIFT;
2380 vmf->pte += iter.index - last_pgoff;
2381 last_pgoff = iter.index;
2382 if (alloc_set_pte(vmf, NULL, page))
2391 /* Huge page is mapped? No need to proceed. */
2392 if (pmd_trans_huge(*vmf->pmd))
2394 if (iter.index == end_pgoff)
2399 EXPORT_SYMBOL(filemap_map_pages);
2401 int filemap_page_mkwrite(struct vm_fault *vmf)
2403 struct page *page = vmf->page;
2404 struct inode *inode = file_inode(vmf->vma->vm_file);
2405 int ret = VM_FAULT_LOCKED;
2407 sb_start_pagefault(inode->i_sb);
2408 file_update_time(vmf->vma->vm_file);
2410 if (page->mapping != inode->i_mapping) {
2412 ret = VM_FAULT_NOPAGE;
2416 * We mark the page dirty already here so that when freeze is in
2417 * progress, we are guaranteed that writeback during freezing will
2418 * see the dirty page and writeprotect it again.
2420 set_page_dirty(page);
2421 wait_for_stable_page(page);
2423 sb_end_pagefault(inode->i_sb);
2426 EXPORT_SYMBOL(filemap_page_mkwrite);
2428 const struct vm_operations_struct generic_file_vm_ops = {
2429 .fault = filemap_fault,
2430 .map_pages = filemap_map_pages,
2431 .page_mkwrite = filemap_page_mkwrite,
2434 /* This is used for a general mmap of a disk file */
2436 int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
2438 struct address_space *mapping = file->f_mapping;
2440 if (!mapping->a_ops->readpage)
2442 file_accessed(file);
2443 vma->vm_ops = &generic_file_vm_ops;
2448 * This is for filesystems which do not implement ->writepage.
2450 int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
2452 if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE))
2454 return generic_file_mmap(file, vma);
2457 int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
2461 int generic_file_readonly_mmap(struct file * file, struct vm_area_struct * vma)
2465 #endif /* CONFIG_MMU */
2467 EXPORT_SYMBOL(generic_file_mmap);
2468 EXPORT_SYMBOL(generic_file_readonly_mmap);
2470 static struct page *wait_on_page_read(struct page *page)
2472 if (!IS_ERR(page)) {
2473 wait_on_page_locked(page);
2474 if (!PageUptodate(page)) {
2476 page = ERR_PTR(-EIO);
2482 static struct page *do_read_cache_page(struct address_space *mapping,
2484 int (*filler)(void *, struct page *),
2491 page = find_get_page(mapping, index);
2493 page = __page_cache_alloc(gfp | __GFP_COLD);
2495 return ERR_PTR(-ENOMEM);
2496 err = add_to_page_cache_lru(page, mapping, index, gfp);
2497 if (unlikely(err)) {
2501 /* Presumably ENOMEM for radix tree node */
2502 return ERR_PTR(err);
2506 err = filler(data, page);
2509 return ERR_PTR(err);
2512 page = wait_on_page_read(page);
2517 if (PageUptodate(page))
2521 * Page is not up to date and may be locked due one of the following
2522 * case a: Page is being filled and the page lock is held
2523 * case b: Read/write error clearing the page uptodate status
2524 * case c: Truncation in progress (page locked)
2525 * case d: Reclaim in progress
2527 * Case a, the page will be up to date when the page is unlocked.
2528 * There is no need to serialise on the page lock here as the page
2529 * is pinned so the lock gives no additional protection. Even if the
2530 * the page is truncated, the data is still valid if PageUptodate as
2531 * it's a race vs truncate race.
2532 * Case b, the page will not be up to date
2533 * Case c, the page may be truncated but in itself, the data may still
2534 * be valid after IO completes as it's a read vs truncate race. The
2535 * operation must restart if the page is not uptodate on unlock but
2536 * otherwise serialising on page lock to stabilise the mapping gives
2537 * no additional guarantees to the caller as the page lock is
2538 * released before return.
2539 * Case d, similar to truncation. If reclaim holds the page lock, it
2540 * will be a race with remove_mapping that determines if the mapping
2541 * is valid on unlock but otherwise the data is valid and there is
2542 * no need to serialise with page lock.
2544 * As the page lock gives no additional guarantee, we optimistically
2545 * wait on the page to be unlocked and check if it's up to date and
2546 * use the page if it is. Otherwise, the page lock is required to
2547 * distinguish between the different cases. The motivation is that we
2548 * avoid spurious serialisations and wakeups when multiple processes
2549 * wait on the same page for IO to complete.
2551 wait_on_page_locked(page);
2552 if (PageUptodate(page))
2555 /* Distinguish between all the cases under the safety of the lock */
2558 /* Case c or d, restart the operation */
2559 if (!page->mapping) {
2565 /* Someone else locked and filled the page in a very small window */
2566 if (PageUptodate(page)) {
2573 mark_page_accessed(page);
2578 * read_cache_page - read into page cache, fill it if needed
2579 * @mapping: the page's address_space
2580 * @index: the page index
2581 * @filler: function to perform the read
2582 * @data: first arg to filler(data, page) function, often left as NULL
2584 * Read into the page cache. If a page already exists, and PageUptodate() is
2585 * not set, try to fill the page and wait for it to become unlocked.
2587 * If the page does not get brought uptodate, return -EIO.
2589 struct page *read_cache_page(struct address_space *mapping,
2591 int (*filler)(void *, struct page *),
2594 return do_read_cache_page(mapping, index, filler, data, mapping_gfp_mask(mapping));
2596 EXPORT_SYMBOL(read_cache_page);
2599 * read_cache_page_gfp - read into page cache, using specified page allocation flags.
2600 * @mapping: the page's address_space
2601 * @index: the page index
2602 * @gfp: the page allocator flags to use if allocating
2604 * This is the same as "read_mapping_page(mapping, index, NULL)", but with
2605 * any new page allocations done using the specified allocation flags.
2607 * If the page does not get brought uptodate, return -EIO.
2609 struct page *read_cache_page_gfp(struct address_space *mapping,
2613 filler_t *filler = (filler_t *)mapping->a_ops->readpage;
2615 return do_read_cache_page(mapping, index, filler, NULL, gfp);
2617 EXPORT_SYMBOL(read_cache_page_gfp);
2620 * Performs necessary checks before doing a write
2622 * Can adjust writing position or amount of bytes to write.
2623 * Returns appropriate error code that caller should return or
2624 * zero in case that write should be allowed.
2626 inline ssize_t generic_write_checks(struct kiocb *iocb, struct iov_iter *from)
2628 struct file *file = iocb->ki_filp;
2629 struct inode *inode = file->f_mapping->host;
2630 unsigned long limit = rlimit(RLIMIT_FSIZE);
2633 if (!iov_iter_count(from))
2636 /* FIXME: this is for backwards compatibility with 2.4 */
2637 if (iocb->ki_flags & IOCB_APPEND)
2638 iocb->ki_pos = i_size_read(inode);
2642 if (limit != RLIM_INFINITY) {
2643 if (iocb->ki_pos >= limit) {
2644 send_sig(SIGXFSZ, current, 0);
2647 iov_iter_truncate(from, limit - (unsigned long)pos);
2653 if (unlikely(pos + iov_iter_count(from) > MAX_NON_LFS &&
2654 !(file->f_flags & O_LARGEFILE))) {
2655 if (pos >= MAX_NON_LFS)
2657 iov_iter_truncate(from, MAX_NON_LFS - (unsigned long)pos);
2661 * Are we about to exceed the fs block limit ?
2663 * If we have written data it becomes a short write. If we have
2664 * exceeded without writing data we send a signal and return EFBIG.
2665 * Linus frestrict idea will clean these up nicely..
2667 if (unlikely(pos >= inode->i_sb->s_maxbytes))
2670 iov_iter_truncate(from, inode->i_sb->s_maxbytes - pos);
2671 return iov_iter_count(from);
2673 EXPORT_SYMBOL(generic_write_checks);
2675 int pagecache_write_begin(struct file *file, struct address_space *mapping,
2676 loff_t pos, unsigned len, unsigned flags,
2677 struct page **pagep, void **fsdata)
2679 const struct address_space_operations *aops = mapping->a_ops;
2681 return aops->write_begin(file, mapping, pos, len, flags,
2684 EXPORT_SYMBOL(pagecache_write_begin);
2686 int pagecache_write_end(struct file *file, struct address_space *mapping,
2687 loff_t pos, unsigned len, unsigned copied,
2688 struct page *page, void *fsdata)
2690 const struct address_space_operations *aops = mapping->a_ops;
2692 return aops->write_end(file, mapping, pos, len, copied, page, fsdata);
2694 EXPORT_SYMBOL(pagecache_write_end);
2697 generic_file_direct_write(struct kiocb *iocb, struct iov_iter *from)
2699 struct file *file = iocb->ki_filp;
2700 struct address_space *mapping = file->f_mapping;
2701 struct inode *inode = mapping->host;
2702 loff_t pos = iocb->ki_pos;
2706 struct iov_iter data;
2708 write_len = iov_iter_count(from);
2709 end = (pos + write_len - 1) >> PAGE_SHIFT;
2711 written = filemap_write_and_wait_range(mapping, pos, pos + write_len - 1);
2716 * After a write we want buffered reads to be sure to go to disk to get
2717 * the new data. We invalidate clean cached page from the region we're
2718 * about to write. We do this *before* the write so that we can return
2719 * without clobbering -EIOCBQUEUED from ->direct_IO().
2721 if (mapping->nrpages) {
2722 written = invalidate_inode_pages2_range(mapping,
2723 pos >> PAGE_SHIFT, end);
2725 * If a page can not be invalidated, return 0 to fall back
2726 * to buffered write.
2729 if (written == -EBUSY)
2736 written = mapping->a_ops->direct_IO(iocb, &data);
2739 * Finally, try again to invalidate clean pages which might have been
2740 * cached by non-direct readahead, or faulted in by get_user_pages()
2741 * if the source of the write was an mmap'ed region of the file
2742 * we're writing. Either one is a pretty crazy thing to do,
2743 * so we don't support it 100%. If this invalidation
2744 * fails, tough, the write still worked...
2746 if (mapping->nrpages) {
2747 invalidate_inode_pages2_range(mapping,
2748 pos >> PAGE_SHIFT, end);
2753 iov_iter_advance(from, written);
2754 if (pos > i_size_read(inode) && !S_ISBLK(inode->i_mode)) {
2755 i_size_write(inode, pos);
2756 mark_inode_dirty(inode);
2763 EXPORT_SYMBOL(generic_file_direct_write);
2766 * Find or create a page at the given pagecache position. Return the locked
2767 * page. This function is specifically for buffered writes.
2769 struct page *grab_cache_page_write_begin(struct address_space *mapping,
2770 pgoff_t index, unsigned flags)
2773 int fgp_flags = FGP_LOCK|FGP_WRITE|FGP_CREAT;
2775 if (flags & AOP_FLAG_NOFS)
2776 fgp_flags |= FGP_NOFS;
2778 page = pagecache_get_page(mapping, index, fgp_flags,
2779 mapping_gfp_mask(mapping));
2781 wait_for_stable_page(page);
2785 EXPORT_SYMBOL(grab_cache_page_write_begin);
2787 ssize_t generic_perform_write(struct file *file,
2788 struct iov_iter *i, loff_t pos)
2790 struct address_space *mapping = file->f_mapping;
2791 const struct address_space_operations *a_ops = mapping->a_ops;
2793 ssize_t written = 0;
2794 unsigned int flags = 0;
2797 * Copies from kernel address space cannot fail (NFSD is a big user).
2799 if (!iter_is_iovec(i))
2800 flags |= AOP_FLAG_UNINTERRUPTIBLE;
2804 unsigned long offset; /* Offset into pagecache page */
2805 unsigned long bytes; /* Bytes to write to page */
2806 size_t copied; /* Bytes copied from user */
2809 offset = (pos & (PAGE_SIZE - 1));
2810 bytes = min_t(unsigned long, PAGE_SIZE - offset,
2815 * Bring in the user page that we will copy from _first_.
2816 * Otherwise there's a nasty deadlock on copying from the
2817 * same page as we're writing to, without it being marked
2820 * Not only is this an optimisation, but it is also required
2821 * to check that the address is actually valid, when atomic
2822 * usercopies are used, below.
2824 if (unlikely(iov_iter_fault_in_readable(i, bytes))) {
2829 if (fatal_signal_pending(current)) {
2834 status = a_ops->write_begin(file, mapping, pos, bytes, flags,
2836 if (unlikely(status < 0))
2839 if (mapping_writably_mapped(mapping))
2840 flush_dcache_page(page);
2842 copied = iov_iter_copy_from_user_atomic(page, i, offset, bytes);
2843 flush_dcache_page(page);
2845 status = a_ops->write_end(file, mapping, pos, bytes, copied,
2847 if (unlikely(status < 0))
2853 iov_iter_advance(i, copied);
2854 if (unlikely(copied == 0)) {
2856 * If we were unable to copy any data at all, we must
2857 * fall back to a single segment length write.
2859 * If we didn't fallback here, we could livelock
2860 * because not all segments in the iov can be copied at
2861 * once without a pagefault.
2863 bytes = min_t(unsigned long, PAGE_SIZE - offset,
2864 iov_iter_single_seg_count(i));
2870 balance_dirty_pages_ratelimited(mapping);
2871 } while (iov_iter_count(i));
2873 return written ? written : status;
2875 EXPORT_SYMBOL(generic_perform_write);
2878 * __generic_file_write_iter - write data to a file
2879 * @iocb: IO state structure (file, offset, etc.)
2880 * @from: iov_iter with data to write
2882 * This function does all the work needed for actually writing data to a
2883 * file. It does all basic checks, removes SUID from the file, updates
2884 * modification times and calls proper subroutines depending on whether we
2885 * do direct IO or a standard buffered write.
2887 * It expects i_mutex to be grabbed unless we work on a block device or similar
2888 * object which does not need locking at all.
2890 * This function does *not* take care of syncing data in case of O_SYNC write.
2891 * A caller has to handle it. This is mainly due to the fact that we want to
2892 * avoid syncing under i_mutex.
2894 ssize_t __generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
2896 struct file *file = iocb->ki_filp;
2897 struct address_space * mapping = file->f_mapping;
2898 struct inode *inode = mapping->host;
2899 ssize_t written = 0;
2903 /* We can write back this queue in page reclaim */
2904 current->backing_dev_info = inode_to_bdi(inode);
2905 err = file_remove_privs(file);
2909 err = file_update_time(file);
2913 if (iocb->ki_flags & IOCB_DIRECT) {
2914 loff_t pos, endbyte;
2916 written = generic_file_direct_write(iocb, from);
2918 * If the write stopped short of completing, fall back to
2919 * buffered writes. Some filesystems do this for writes to
2920 * holes, for example. For DAX files, a buffered write will
2921 * not succeed (even if it did, DAX does not handle dirty
2922 * page-cache pages correctly).
2924 if (written < 0 || !iov_iter_count(from) || IS_DAX(inode))
2927 status = generic_perform_write(file, from, pos = iocb->ki_pos);
2929 * If generic_perform_write() returned a synchronous error
2930 * then we want to return the number of bytes which were
2931 * direct-written, or the error code if that was zero. Note
2932 * that this differs from normal direct-io semantics, which
2933 * will return -EFOO even if some bytes were written.
2935 if (unlikely(status < 0)) {
2940 * We need to ensure that the page cache pages are written to
2941 * disk and invalidated to preserve the expected O_DIRECT
2944 endbyte = pos + status - 1;
2945 err = filemap_write_and_wait_range(mapping, pos, endbyte);
2947 iocb->ki_pos = endbyte + 1;
2949 invalidate_mapping_pages(mapping,
2951 endbyte >> PAGE_SHIFT);
2954 * We don't know how much we wrote, so just return
2955 * the number of bytes which were direct-written
2959 written = generic_perform_write(file, from, iocb->ki_pos);
2960 if (likely(written > 0))
2961 iocb->ki_pos += written;
2964 current->backing_dev_info = NULL;
2965 return written ? written : err;
2967 EXPORT_SYMBOL(__generic_file_write_iter);
2970 * generic_file_write_iter - write data to a file
2971 * @iocb: IO state structure
2972 * @from: iov_iter with data to write
2974 * This is a wrapper around __generic_file_write_iter() to be used by most
2975 * filesystems. It takes care of syncing the file in case of O_SYNC file
2976 * and acquires i_mutex as needed.
2978 ssize_t generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
2980 struct file *file = iocb->ki_filp;
2981 struct inode *inode = file->f_mapping->host;
2985 ret = generic_write_checks(iocb, from);
2987 ret = __generic_file_write_iter(iocb, from);
2988 inode_unlock(inode);
2991 ret = generic_write_sync(iocb, ret);
2994 EXPORT_SYMBOL(generic_file_write_iter);
2997 * try_to_release_page() - release old fs-specific metadata on a page
2999 * @page: the page which the kernel is trying to free
3000 * @gfp_mask: memory allocation flags (and I/O mode)
3002 * The address_space is to try to release any data against the page
3003 * (presumably at page->private). If the release was successful, return `1'.
3004 * Otherwise return zero.
3006 * This may also be called if PG_fscache is set on a page, indicating that the
3007 * page is known to the local caching routines.
3009 * The @gfp_mask argument specifies whether I/O may be performed to release
3010 * this page (__GFP_IO), and whether the call may block (__GFP_RECLAIM & __GFP_FS).
3013 int try_to_release_page(struct page *page, gfp_t gfp_mask)
3015 struct address_space * const mapping = page->mapping;
3017 BUG_ON(!PageLocked(page));
3018 if (PageWriteback(page))
3021 if (mapping && mapping->a_ops->releasepage)
3022 return mapping->a_ops->releasepage(page, gfp_mask);
3023 return try_to_free_buffers(page);
3026 EXPORT_SYMBOL(try_to_release_page);