4 * Copyright (C) 1994-1999 Linus Torvalds
8 * This file handles the generic file mmap semantics used by
9 * most "normal" filesystems (but you don't /have/ to use this:
10 * the NFS filesystem used to do this differently, for example)
12 #include <linux/export.h>
13 #include <linux/compiler.h>
14 #include <linux/dax.h>
16 #include <linux/sched/signal.h>
17 #include <linux/uaccess.h>
18 #include <linux/capability.h>
19 #include <linux/kernel_stat.h>
20 #include <linux/gfp.h>
22 #include <linux/swap.h>
23 #include <linux/mman.h>
24 #include <linux/pagemap.h>
25 #include <linux/file.h>
26 #include <linux/uio.h>
27 #include <linux/hash.h>
28 #include <linux/writeback.h>
29 #include <linux/backing-dev.h>
30 #include <linux/pagevec.h>
31 #include <linux/blkdev.h>
32 #include <linux/security.h>
33 #include <linux/cpuset.h>
34 #include <linux/hardirq.h> /* for BUG_ON(!in_atomic()) only */
35 #include <linux/hugetlb.h>
36 #include <linux/memcontrol.h>
37 #include <linux/cleancache.h>
38 #include <linux/rmap.h>
41 #define CREATE_TRACE_POINTS
42 #include <trace/events/filemap.h>
45 * FIXME: remove all knowledge of the buffer layer from the core VM
47 #include <linux/buffer_head.h> /* for try_to_free_buffers */
52 * Shared mappings implemented 30.11.1994. It's not fully working yet,
55 * Shared mappings now work. 15.8.1995 Bruno.
57 * finished 'unifying' the page and buffer cache and SMP-threaded the
58 * page-cache, 21.05.1999, Ingo Molnar <mingo@redhat.com>
60 * SMP-threaded pagemap-LRU 1999, Andrea Arcangeli <andrea@suse.de>
66 * ->i_mmap_rwsem (truncate_pagecache)
67 * ->private_lock (__free_pte->__set_page_dirty_buffers)
68 * ->swap_lock (exclusive_swap_page, others)
69 * ->mapping->tree_lock
72 * ->i_mmap_rwsem (truncate->unmap_mapping_range)
76 * ->page_table_lock or pte_lock (various, mainly in memory.c)
77 * ->mapping->tree_lock (arch-dependent flush_dcache_mmap_lock)
80 * ->lock_page (access_process_vm)
82 * ->i_mutex (generic_perform_write)
83 * ->mmap_sem (fault_in_pages_readable->do_page_fault)
86 * sb_lock (fs/fs-writeback.c)
87 * ->mapping->tree_lock (__sync_single_inode)
90 * ->anon_vma.lock (vma_adjust)
93 * ->page_table_lock or pte_lock (anon_vma_prepare and various)
95 * ->page_table_lock or pte_lock
96 * ->swap_lock (try_to_unmap_one)
97 * ->private_lock (try_to_unmap_one)
98 * ->tree_lock (try_to_unmap_one)
99 * ->zone_lru_lock(zone) (follow_page->mark_page_accessed)
100 * ->zone_lru_lock(zone) (check_pte_range->isolate_lru_page)
101 * ->private_lock (page_remove_rmap->set_page_dirty)
102 * ->tree_lock (page_remove_rmap->set_page_dirty)
103 * bdi.wb->list_lock (page_remove_rmap->set_page_dirty)
104 * ->inode->i_lock (page_remove_rmap->set_page_dirty)
105 * ->memcg->move_lock (page_remove_rmap->lock_page_memcg)
106 * bdi.wb->list_lock (zap_pte_range->set_page_dirty)
107 * ->inode->i_lock (zap_pte_range->set_page_dirty)
108 * ->private_lock (zap_pte_range->__set_page_dirty_buffers)
111 * ->tasklist_lock (memory_failure, collect_procs_ao)
114 static int page_cache_tree_insert(struct address_space *mapping,
115 struct page *page, void **shadowp)
117 struct radix_tree_node *node;
121 error = __radix_tree_create(&mapping->page_tree, page->index, 0,
128 p = radix_tree_deref_slot_protected(slot, &mapping->tree_lock);
129 if (!radix_tree_exceptional_entry(p))
132 mapping->nrexceptional--;
133 if (!dax_mapping(mapping)) {
137 /* DAX can replace empty locked entry with a hole */
139 dax_radix_locked_entry(0, RADIX_DAX_EMPTY));
140 /* Wakeup waiters for exceptional entry lock */
141 dax_wake_mapping_entry_waiter(mapping, page->index, p,
145 __radix_tree_replace(&mapping->page_tree, node, slot, page,
146 workingset_update_node, mapping);
151 static void page_cache_tree_delete(struct address_space *mapping,
152 struct page *page, void *shadow)
156 /* hugetlb pages are represented by one entry in the radix tree */
157 nr = PageHuge(page) ? 1 : hpage_nr_pages(page);
159 VM_BUG_ON_PAGE(!PageLocked(page), page);
160 VM_BUG_ON_PAGE(PageTail(page), page);
161 VM_BUG_ON_PAGE(nr != 1 && shadow, page);
163 for (i = 0; i < nr; i++) {
164 struct radix_tree_node *node;
167 __radix_tree_lookup(&mapping->page_tree, page->index + i,
170 VM_BUG_ON_PAGE(!node && nr != 1, page);
172 radix_tree_clear_tags(&mapping->page_tree, node, slot);
173 __radix_tree_replace(&mapping->page_tree, node, slot, shadow,
174 workingset_update_node, mapping);
178 mapping->nrexceptional += nr;
180 * Make sure the nrexceptional update is committed before
181 * the nrpages update so that final truncate racing
182 * with reclaim does not see both counters 0 at the
183 * same time and miss a shadow entry.
187 mapping->nrpages -= nr;
191 * Delete a page from the page cache and free it. Caller has to make
192 * sure the page is locked and that nobody else uses it - or that usage
193 * is safe. The caller must hold the mapping's tree_lock.
195 void __delete_from_page_cache(struct page *page, void *shadow)
197 struct address_space *mapping = page->mapping;
198 int nr = hpage_nr_pages(page);
200 trace_mm_filemap_delete_from_page_cache(page);
202 * if we're uptodate, flush out into the cleancache, otherwise
203 * invalidate any existing cleancache entries. We can't leave
204 * stale data around in the cleancache once our page is gone
206 if (PageUptodate(page) && PageMappedToDisk(page))
207 cleancache_put_page(page);
209 cleancache_invalidate_page(mapping, page);
211 VM_BUG_ON_PAGE(PageTail(page), page);
212 VM_BUG_ON_PAGE(page_mapped(page), page);
213 if (!IS_ENABLED(CONFIG_DEBUG_VM) && unlikely(page_mapped(page))) {
216 pr_alert("BUG: Bad page cache in process %s pfn:%05lx\n",
217 current->comm, page_to_pfn(page));
218 dump_page(page, "still mapped when deleted");
220 add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE);
222 mapcount = page_mapcount(page);
223 if (mapping_exiting(mapping) &&
224 page_count(page) >= mapcount + 2) {
226 * All vmas have already been torn down, so it's
227 * a good bet that actually the page is unmapped,
228 * and we'd prefer not to leak it: if we're wrong,
229 * some other bad page check should catch it later.
231 page_mapcount_reset(page);
232 page_ref_sub(page, mapcount);
236 page_cache_tree_delete(mapping, page, shadow);
238 page->mapping = NULL;
239 /* Leave page->index set: truncation lookup relies upon it */
241 /* hugetlb pages do not participate in page cache accounting. */
243 __mod_node_page_state(page_pgdat(page), NR_FILE_PAGES, -nr);
244 if (PageSwapBacked(page)) {
245 __mod_node_page_state(page_pgdat(page), NR_SHMEM, -nr);
246 if (PageTransHuge(page))
247 __dec_node_page_state(page, NR_SHMEM_THPS);
249 VM_BUG_ON_PAGE(PageTransHuge(page) && !PageHuge(page), page);
253 * At this point page must be either written or cleaned by truncate.
254 * Dirty page here signals a bug and loss of unwritten data.
256 * This fixes dirty accounting after removing the page entirely but
257 * leaves PageDirty set: it has no effect for truncated page and
258 * anyway will be cleared before returning page into buddy allocator.
260 if (WARN_ON_ONCE(PageDirty(page)))
261 account_page_cleaned(page, mapping, inode_to_wb(mapping->host));
265 * delete_from_page_cache - delete page from page cache
266 * @page: the page which the kernel is trying to remove from page cache
268 * This must be called only on pages that have been verified to be in the page
269 * cache and locked. It will never put the page into the free list, the caller
270 * has a reference on the page.
272 void delete_from_page_cache(struct page *page)
274 struct address_space *mapping = page_mapping(page);
276 void (*freepage)(struct page *);
278 BUG_ON(!PageLocked(page));
280 freepage = mapping->a_ops->freepage;
282 spin_lock_irqsave(&mapping->tree_lock, flags);
283 __delete_from_page_cache(page, NULL);
284 spin_unlock_irqrestore(&mapping->tree_lock, flags);
289 if (PageTransHuge(page) && !PageHuge(page)) {
290 page_ref_sub(page, HPAGE_PMD_NR);
291 VM_BUG_ON_PAGE(page_count(page) <= 0, page);
296 EXPORT_SYMBOL(delete_from_page_cache);
298 int filemap_check_errors(struct address_space *mapping)
301 /* Check for outstanding write errors */
302 if (test_bit(AS_ENOSPC, &mapping->flags) &&
303 test_and_clear_bit(AS_ENOSPC, &mapping->flags))
305 if (test_bit(AS_EIO, &mapping->flags) &&
306 test_and_clear_bit(AS_EIO, &mapping->flags))
310 EXPORT_SYMBOL(filemap_check_errors);
312 static int filemap_check_and_keep_errors(struct address_space *mapping)
314 /* Check for outstanding write errors */
315 if (test_bit(AS_EIO, &mapping->flags))
317 if (test_bit(AS_ENOSPC, &mapping->flags))
323 * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range
324 * @mapping: address space structure to write
325 * @start: offset in bytes where the range starts
326 * @end: offset in bytes where the range ends (inclusive)
327 * @sync_mode: enable synchronous operation
329 * Start writeback against all of a mapping's dirty pages that lie
330 * within the byte offsets <start, end> inclusive.
332 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
333 * opposed to a regular memory cleansing writeback. The difference between
334 * these two operations is that if a dirty page/buffer is encountered, it must
335 * be waited upon, and not just skipped over.
337 int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
338 loff_t end, int sync_mode)
341 struct writeback_control wbc = {
342 .sync_mode = sync_mode,
343 .nr_to_write = LONG_MAX,
344 .range_start = start,
348 if (!mapping_cap_writeback_dirty(mapping))
351 wbc_attach_fdatawrite_inode(&wbc, mapping->host);
352 ret = do_writepages(mapping, &wbc);
353 wbc_detach_inode(&wbc);
357 static inline int __filemap_fdatawrite(struct address_space *mapping,
360 return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode);
363 int filemap_fdatawrite(struct address_space *mapping)
365 return __filemap_fdatawrite(mapping, WB_SYNC_ALL);
367 EXPORT_SYMBOL(filemap_fdatawrite);
369 int filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
372 return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL);
374 EXPORT_SYMBOL(filemap_fdatawrite_range);
377 * filemap_flush - mostly a non-blocking flush
378 * @mapping: target address_space
380 * This is a mostly non-blocking flush. Not suitable for data-integrity
381 * purposes - I/O may not be started against all dirty pages.
383 int filemap_flush(struct address_space *mapping)
385 return __filemap_fdatawrite(mapping, WB_SYNC_NONE);
387 EXPORT_SYMBOL(filemap_flush);
389 static int __filemap_fdatawait_range(struct address_space *mapping,
390 loff_t start_byte, loff_t end_byte)
392 pgoff_t index = start_byte >> PAGE_SHIFT;
393 pgoff_t end = end_byte >> PAGE_SHIFT;
398 if (end_byte < start_byte)
401 pagevec_init(&pvec, 0);
402 while ((index <= end) &&
403 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
404 PAGECACHE_TAG_WRITEBACK,
405 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1)) != 0) {
408 for (i = 0; i < nr_pages; i++) {
409 struct page *page = pvec.pages[i];
411 /* until radix tree lookup accepts end_index */
412 if (page->index > end)
415 wait_on_page_writeback(page);
416 if (TestClearPageError(page))
419 pagevec_release(&pvec);
427 * filemap_fdatawait_range - wait for writeback to complete
428 * @mapping: address space structure to wait for
429 * @start_byte: offset in bytes where the range starts
430 * @end_byte: offset in bytes where the range ends (inclusive)
432 * Walk the list of under-writeback pages of the given address space
433 * in the given range and wait for all of them. Check error status of
434 * the address space and return it.
436 * Since the error status of the address space is cleared by this function,
437 * callers are responsible for checking the return value and handling and/or
438 * reporting the error.
440 int filemap_fdatawait_range(struct address_space *mapping, loff_t start_byte,
445 ret = __filemap_fdatawait_range(mapping, start_byte, end_byte);
446 ret2 = filemap_check_errors(mapping);
452 EXPORT_SYMBOL(filemap_fdatawait_range);
455 * filemap_fdatawait_keep_errors - wait for writeback without clearing errors
456 * @mapping: address space structure to wait for
458 * Walk the list of under-writeback pages of the given address space
459 * and wait for all of them. Unlike filemap_fdatawait(), this function
460 * does not clear error status of the address space.
462 * Use this function if callers don't handle errors themselves. Expected
463 * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2),
466 int filemap_fdatawait_keep_errors(struct address_space *mapping)
468 loff_t i_size = i_size_read(mapping->host);
473 __filemap_fdatawait_range(mapping, 0, i_size - 1);
474 return filemap_check_and_keep_errors(mapping);
476 EXPORT_SYMBOL(filemap_fdatawait_keep_errors);
479 * filemap_fdatawait - wait for all under-writeback pages to complete
480 * @mapping: address space structure to wait for
482 * Walk the list of under-writeback pages of the given address space
483 * and wait for all of them. Check error status of the address space
486 * Since the error status of the address space is cleared by this function,
487 * callers are responsible for checking the return value and handling and/or
488 * reporting the error.
490 int filemap_fdatawait(struct address_space *mapping)
492 loff_t i_size = i_size_read(mapping->host);
497 return filemap_fdatawait_range(mapping, 0, i_size - 1);
499 EXPORT_SYMBOL(filemap_fdatawait);
501 int filemap_write_and_wait(struct address_space *mapping)
505 if ((!dax_mapping(mapping) && mapping->nrpages) ||
506 (dax_mapping(mapping) && mapping->nrexceptional)) {
507 err = filemap_fdatawrite(mapping);
509 * Even if the above returned error, the pages may be
510 * written partially (e.g. -ENOSPC), so we wait for it.
511 * But the -EIO is special case, it may indicate the worst
512 * thing (e.g. bug) happened, so we avoid waiting for it.
515 int err2 = filemap_fdatawait(mapping);
519 /* Clear any previously stored errors */
520 filemap_check_errors(mapping);
523 err = filemap_check_errors(mapping);
527 EXPORT_SYMBOL(filemap_write_and_wait);
530 * filemap_write_and_wait_range - write out & wait on a file range
531 * @mapping: the address_space for the pages
532 * @lstart: offset in bytes where the range starts
533 * @lend: offset in bytes where the range ends (inclusive)
535 * Write out and wait upon file offsets lstart->lend, inclusive.
537 * Note that @lend is inclusive (describes the last byte to be written) so
538 * that this function can be used to write to the very end-of-file (end = -1).
540 int filemap_write_and_wait_range(struct address_space *mapping,
541 loff_t lstart, loff_t lend)
545 if ((!dax_mapping(mapping) && mapping->nrpages) ||
546 (dax_mapping(mapping) && mapping->nrexceptional)) {
547 err = __filemap_fdatawrite_range(mapping, lstart, lend,
549 /* See comment of filemap_write_and_wait() */
551 int err2 = filemap_fdatawait_range(mapping,
556 /* Clear any previously stored errors */
557 filemap_check_errors(mapping);
560 err = filemap_check_errors(mapping);
564 EXPORT_SYMBOL(filemap_write_and_wait_range);
567 * replace_page_cache_page - replace a pagecache page with a new one
568 * @old: page to be replaced
569 * @new: page to replace with
570 * @gfp_mask: allocation mode
572 * This function replaces a page in the pagecache with a new one. On
573 * success it acquires the pagecache reference for the new page and
574 * drops it for the old page. Both the old and new pages must be
575 * locked. This function does not add the new page to the LRU, the
576 * caller must do that.
578 * The remove + add is atomic. The only way this function can fail is
579 * memory allocation failure.
581 int replace_page_cache_page(struct page *old, struct page *new, gfp_t gfp_mask)
585 VM_BUG_ON_PAGE(!PageLocked(old), old);
586 VM_BUG_ON_PAGE(!PageLocked(new), new);
587 VM_BUG_ON_PAGE(new->mapping, new);
589 error = radix_tree_preload(gfp_mask & ~__GFP_HIGHMEM);
591 struct address_space *mapping = old->mapping;
592 void (*freepage)(struct page *);
595 pgoff_t offset = old->index;
596 freepage = mapping->a_ops->freepage;
599 new->mapping = mapping;
602 spin_lock_irqsave(&mapping->tree_lock, flags);
603 __delete_from_page_cache(old, NULL);
604 error = page_cache_tree_insert(mapping, new, NULL);
608 * hugetlb pages do not participate in page cache accounting.
611 __inc_node_page_state(new, NR_FILE_PAGES);
612 if (PageSwapBacked(new))
613 __inc_node_page_state(new, NR_SHMEM);
614 spin_unlock_irqrestore(&mapping->tree_lock, flags);
615 mem_cgroup_migrate(old, new);
616 radix_tree_preload_end();
624 EXPORT_SYMBOL_GPL(replace_page_cache_page);
626 static int __add_to_page_cache_locked(struct page *page,
627 struct address_space *mapping,
628 pgoff_t offset, gfp_t gfp_mask,
631 int huge = PageHuge(page);
632 struct mem_cgroup *memcg;
635 VM_BUG_ON_PAGE(!PageLocked(page), page);
636 VM_BUG_ON_PAGE(PageSwapBacked(page), page);
639 error = mem_cgroup_try_charge(page, current->mm,
640 gfp_mask, &memcg, false);
645 error = radix_tree_maybe_preload(gfp_mask & ~__GFP_HIGHMEM);
648 mem_cgroup_cancel_charge(page, memcg, false);
653 page->mapping = mapping;
654 page->index = offset;
656 spin_lock_irq(&mapping->tree_lock);
657 error = page_cache_tree_insert(mapping, page, shadowp);
658 radix_tree_preload_end();
662 /* hugetlb pages do not participate in page cache accounting. */
664 __inc_node_page_state(page, NR_FILE_PAGES);
665 spin_unlock_irq(&mapping->tree_lock);
667 mem_cgroup_commit_charge(page, memcg, false, false);
668 trace_mm_filemap_add_to_page_cache(page);
671 page->mapping = NULL;
672 /* Leave page->index set: truncation relies upon it */
673 spin_unlock_irq(&mapping->tree_lock);
675 mem_cgroup_cancel_charge(page, memcg, false);
681 * add_to_page_cache_locked - add a locked page to the pagecache
683 * @mapping: the page's address_space
684 * @offset: page index
685 * @gfp_mask: page allocation mode
687 * This function is used to add a page to the pagecache. It must be locked.
688 * This function does not add the page to the LRU. The caller must do that.
690 int add_to_page_cache_locked(struct page *page, struct address_space *mapping,
691 pgoff_t offset, gfp_t gfp_mask)
693 return __add_to_page_cache_locked(page, mapping, offset,
696 EXPORT_SYMBOL(add_to_page_cache_locked);
698 int add_to_page_cache_lru(struct page *page, struct address_space *mapping,
699 pgoff_t offset, gfp_t gfp_mask)
704 __SetPageLocked(page);
705 ret = __add_to_page_cache_locked(page, mapping, offset,
708 __ClearPageLocked(page);
711 * The page might have been evicted from cache only
712 * recently, in which case it should be activated like
713 * any other repeatedly accessed page.
714 * The exception is pages getting rewritten; evicting other
715 * data from the working set, only to cache data that will
716 * get overwritten with something else, is a waste of memory.
718 if (!(gfp_mask & __GFP_WRITE) &&
719 shadow && workingset_refault(shadow)) {
721 workingset_activation(page);
723 ClearPageActive(page);
728 EXPORT_SYMBOL_GPL(add_to_page_cache_lru);
731 struct page *__page_cache_alloc(gfp_t gfp)
736 if (cpuset_do_page_mem_spread()) {
737 unsigned int cpuset_mems_cookie;
739 cpuset_mems_cookie = read_mems_allowed_begin();
740 n = cpuset_mem_spread_node();
741 page = __alloc_pages_node(n, gfp, 0);
742 } while (!page && read_mems_allowed_retry(cpuset_mems_cookie));
746 return alloc_pages(gfp, 0);
748 EXPORT_SYMBOL(__page_cache_alloc);
752 * In order to wait for pages to become available there must be
753 * waitqueues associated with pages. By using a hash table of
754 * waitqueues where the bucket discipline is to maintain all
755 * waiters on the same queue and wake all when any of the pages
756 * become available, and for the woken contexts to check to be
757 * sure the appropriate page became available, this saves space
758 * at a cost of "thundering herd" phenomena during rare hash
761 #define PAGE_WAIT_TABLE_BITS 8
762 #define PAGE_WAIT_TABLE_SIZE (1 << PAGE_WAIT_TABLE_BITS)
763 static wait_queue_head_t page_wait_table[PAGE_WAIT_TABLE_SIZE] __cacheline_aligned;
765 static wait_queue_head_t *page_waitqueue(struct page *page)
767 return &page_wait_table[hash_ptr(page, PAGE_WAIT_TABLE_BITS)];
770 void __init pagecache_init(void)
774 for (i = 0; i < PAGE_WAIT_TABLE_SIZE; i++)
775 init_waitqueue_head(&page_wait_table[i]);
777 page_writeback_init();
780 struct wait_page_key {
786 struct wait_page_queue {
792 static int wake_page_function(wait_queue_t *wait, unsigned mode, int sync, void *arg)
794 struct wait_page_key *key = arg;
795 struct wait_page_queue *wait_page
796 = container_of(wait, struct wait_page_queue, wait);
798 if (wait_page->page != key->page)
802 if (wait_page->bit_nr != key->bit_nr)
804 if (test_bit(key->bit_nr, &key->page->flags))
807 return autoremove_wake_function(wait, mode, sync, key);
810 static void wake_up_page_bit(struct page *page, int bit_nr)
812 wait_queue_head_t *q = page_waitqueue(page);
813 struct wait_page_key key;
820 spin_lock_irqsave(&q->lock, flags);
821 __wake_up_locked_key(q, TASK_NORMAL, &key);
823 * It is possible for other pages to have collided on the waitqueue
824 * hash, so in that case check for a page match. That prevents a long-
827 * It is still possible to miss a case here, when we woke page waiters
828 * and removed them from the waitqueue, but there are still other
831 if (!waitqueue_active(q) || !key.page_match) {
832 ClearPageWaiters(page);
834 * It's possible to miss clearing Waiters here, when we woke
835 * our page waiters, but the hashed waitqueue has waiters for
838 * That's okay, it's a rare case. The next waker will clear it.
841 spin_unlock_irqrestore(&q->lock, flags);
844 static void wake_up_page(struct page *page, int bit)
846 if (!PageWaiters(page))
848 wake_up_page_bit(page, bit);
851 static inline int wait_on_page_bit_common(wait_queue_head_t *q,
852 struct page *page, int bit_nr, int state, bool lock)
854 struct wait_page_queue wait_page;
855 wait_queue_t *wait = &wait_page.wait;
859 wait->func = wake_page_function;
860 wait_page.page = page;
861 wait_page.bit_nr = bit_nr;
864 spin_lock_irq(&q->lock);
866 if (likely(list_empty(&wait->task_list))) {
868 __add_wait_queue_tail_exclusive(q, wait);
870 __add_wait_queue(q, wait);
871 SetPageWaiters(page);
874 set_current_state(state);
876 spin_unlock_irq(&q->lock);
878 if (likely(test_bit(bit_nr, &page->flags))) {
880 if (unlikely(signal_pending_state(state, current))) {
887 if (!test_and_set_bit_lock(bit_nr, &page->flags))
890 if (!test_bit(bit_nr, &page->flags))
895 finish_wait(q, wait);
898 * A signal could leave PageWaiters set. Clearing it here if
899 * !waitqueue_active would be possible (by open-coding finish_wait),
900 * but still fail to catch it in the case of wait hash collision. We
901 * already can fail to clear wait hash collision cases, so don't
902 * bother with signals either.
908 void wait_on_page_bit(struct page *page, int bit_nr)
910 wait_queue_head_t *q = page_waitqueue(page);
911 wait_on_page_bit_common(q, page, bit_nr, TASK_UNINTERRUPTIBLE, false);
913 EXPORT_SYMBOL(wait_on_page_bit);
915 int wait_on_page_bit_killable(struct page *page, int bit_nr)
917 wait_queue_head_t *q = page_waitqueue(page);
918 return wait_on_page_bit_common(q, page, bit_nr, TASK_KILLABLE, false);
922 * add_page_wait_queue - Add an arbitrary waiter to a page's wait queue
923 * @page: Page defining the wait queue of interest
924 * @waiter: Waiter to add to the queue
926 * Add an arbitrary @waiter to the wait queue for the nominated @page.
928 void add_page_wait_queue(struct page *page, wait_queue_t *waiter)
930 wait_queue_head_t *q = page_waitqueue(page);
933 spin_lock_irqsave(&q->lock, flags);
934 __add_wait_queue(q, waiter);
935 SetPageWaiters(page);
936 spin_unlock_irqrestore(&q->lock, flags);
938 EXPORT_SYMBOL_GPL(add_page_wait_queue);
940 #ifndef clear_bit_unlock_is_negative_byte
943 * PG_waiters is the high bit in the same byte as PG_lock.
945 * On x86 (and on many other architectures), we can clear PG_lock and
946 * test the sign bit at the same time. But if the architecture does
947 * not support that special operation, we just do this all by hand
950 * The read of PG_waiters has to be after (or concurrently with) PG_locked
951 * being cleared, but a memory barrier should be unneccssary since it is
952 * in the same byte as PG_locked.
954 static inline bool clear_bit_unlock_is_negative_byte(long nr, volatile void *mem)
956 clear_bit_unlock(nr, mem);
957 /* smp_mb__after_atomic(); */
958 return test_bit(PG_waiters, mem);
964 * unlock_page - unlock a locked page
967 * Unlocks the page and wakes up sleepers in ___wait_on_page_locked().
968 * Also wakes sleepers in wait_on_page_writeback() because the wakeup
969 * mechanism between PageLocked pages and PageWriteback pages is shared.
970 * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep.
972 * Note that this depends on PG_waiters being the sign bit in the byte
973 * that contains PG_locked - thus the BUILD_BUG_ON(). That allows us to
974 * clear the PG_locked bit and test PG_waiters at the same time fairly
975 * portably (architectures that do LL/SC can test any bit, while x86 can
976 * test the sign bit).
978 void unlock_page(struct page *page)
980 BUILD_BUG_ON(PG_waiters != 7);
981 page = compound_head(page);
982 VM_BUG_ON_PAGE(!PageLocked(page), page);
983 if (clear_bit_unlock_is_negative_byte(PG_locked, &page->flags))
984 wake_up_page_bit(page, PG_locked);
986 EXPORT_SYMBOL(unlock_page);
989 * end_page_writeback - end writeback against a page
992 void end_page_writeback(struct page *page)
995 * TestClearPageReclaim could be used here but it is an atomic
996 * operation and overkill in this particular case. Failing to
997 * shuffle a page marked for immediate reclaim is too mild to
998 * justify taking an atomic operation penalty at the end of
999 * ever page writeback.
1001 if (PageReclaim(page)) {
1002 ClearPageReclaim(page);
1003 rotate_reclaimable_page(page);
1006 if (!test_clear_page_writeback(page))
1009 smp_mb__after_atomic();
1010 wake_up_page(page, PG_writeback);
1012 EXPORT_SYMBOL(end_page_writeback);
1015 * After completing I/O on a page, call this routine to update the page
1016 * flags appropriately
1018 void page_endio(struct page *page, bool is_write, int err)
1022 SetPageUptodate(page);
1024 ClearPageUptodate(page);
1030 struct address_space *mapping;
1033 mapping = page_mapping(page);
1035 mapping_set_error(mapping, err);
1037 end_page_writeback(page);
1040 EXPORT_SYMBOL_GPL(page_endio);
1043 * __lock_page - get a lock on the page, assuming we need to sleep to get it
1044 * @__page: the page to lock
1046 void __lock_page(struct page *__page)
1048 struct page *page = compound_head(__page);
1049 wait_queue_head_t *q = page_waitqueue(page);
1050 wait_on_page_bit_common(q, page, PG_locked, TASK_UNINTERRUPTIBLE, true);
1052 EXPORT_SYMBOL(__lock_page);
1054 int __lock_page_killable(struct page *__page)
1056 struct page *page = compound_head(__page);
1057 wait_queue_head_t *q = page_waitqueue(page);
1058 return wait_on_page_bit_common(q, page, PG_locked, TASK_KILLABLE, true);
1060 EXPORT_SYMBOL_GPL(__lock_page_killable);
1064 * 1 - page is locked; mmap_sem is still held.
1065 * 0 - page is not locked.
1066 * mmap_sem has been released (up_read()), unless flags had both
1067 * FAULT_FLAG_ALLOW_RETRY and FAULT_FLAG_RETRY_NOWAIT set, in
1068 * which case mmap_sem is still held.
1070 * If neither ALLOW_RETRY nor KILLABLE are set, will always return 1
1071 * with the page locked and the mmap_sem unperturbed.
1073 int __lock_page_or_retry(struct page *page, struct mm_struct *mm,
1076 if (flags & FAULT_FLAG_ALLOW_RETRY) {
1078 * CAUTION! In this case, mmap_sem is not released
1079 * even though return 0.
1081 if (flags & FAULT_FLAG_RETRY_NOWAIT)
1084 up_read(&mm->mmap_sem);
1085 if (flags & FAULT_FLAG_KILLABLE)
1086 wait_on_page_locked_killable(page);
1088 wait_on_page_locked(page);
1091 if (flags & FAULT_FLAG_KILLABLE) {
1094 ret = __lock_page_killable(page);
1096 up_read(&mm->mmap_sem);
1106 * page_cache_next_hole - find the next hole (not-present entry)
1109 * @max_scan: maximum range to search
1111 * Search the set [index, min(index+max_scan-1, MAX_INDEX)] for the
1112 * lowest indexed hole.
1114 * Returns: the index of the hole if found, otherwise returns an index
1115 * outside of the set specified (in which case 'return - index >=
1116 * max_scan' will be true). In rare cases of index wrap-around, 0 will
1119 * page_cache_next_hole may be called under rcu_read_lock. However,
1120 * like radix_tree_gang_lookup, this will not atomically search a
1121 * snapshot of the tree at a single point in time. For example, if a
1122 * hole is created at index 5, then subsequently a hole is created at
1123 * index 10, page_cache_next_hole covering both indexes may return 10
1124 * if called under rcu_read_lock.
1126 pgoff_t page_cache_next_hole(struct address_space *mapping,
1127 pgoff_t index, unsigned long max_scan)
1131 for (i = 0; i < max_scan; i++) {
1134 page = radix_tree_lookup(&mapping->page_tree, index);
1135 if (!page || radix_tree_exceptional_entry(page))
1144 EXPORT_SYMBOL(page_cache_next_hole);
1147 * page_cache_prev_hole - find the prev hole (not-present entry)
1150 * @max_scan: maximum range to search
1152 * Search backwards in the range [max(index-max_scan+1, 0), index] for
1155 * Returns: the index of the hole if found, otherwise returns an index
1156 * outside of the set specified (in which case 'index - return >=
1157 * max_scan' will be true). In rare cases of wrap-around, ULONG_MAX
1160 * page_cache_prev_hole may be called under rcu_read_lock. However,
1161 * like radix_tree_gang_lookup, this will not atomically search a
1162 * snapshot of the tree at a single point in time. For example, if a
1163 * hole is created at index 10, then subsequently a hole is created at
1164 * index 5, page_cache_prev_hole covering both indexes may return 5 if
1165 * called under rcu_read_lock.
1167 pgoff_t page_cache_prev_hole(struct address_space *mapping,
1168 pgoff_t index, unsigned long max_scan)
1172 for (i = 0; i < max_scan; i++) {
1175 page = radix_tree_lookup(&mapping->page_tree, index);
1176 if (!page || radix_tree_exceptional_entry(page))
1179 if (index == ULONG_MAX)
1185 EXPORT_SYMBOL(page_cache_prev_hole);
1188 * find_get_entry - find and get a page cache entry
1189 * @mapping: the address_space to search
1190 * @offset: the page cache index
1192 * Looks up the page cache slot at @mapping & @offset. If there is a
1193 * page cache page, it is returned with an increased refcount.
1195 * If the slot holds a shadow entry of a previously evicted page, or a
1196 * swap entry from shmem/tmpfs, it is returned.
1198 * Otherwise, %NULL is returned.
1200 struct page *find_get_entry(struct address_space *mapping, pgoff_t offset)
1203 struct page *head, *page;
1208 pagep = radix_tree_lookup_slot(&mapping->page_tree, offset);
1210 page = radix_tree_deref_slot(pagep);
1211 if (unlikely(!page))
1213 if (radix_tree_exception(page)) {
1214 if (radix_tree_deref_retry(page))
1217 * A shadow entry of a recently evicted page,
1218 * or a swap entry from shmem/tmpfs. Return
1219 * it without attempting to raise page count.
1224 head = compound_head(page);
1225 if (!page_cache_get_speculative(head))
1228 /* The page was split under us? */
1229 if (compound_head(page) != head) {
1235 * Has the page moved?
1236 * This is part of the lockless pagecache protocol. See
1237 * include/linux/pagemap.h for details.
1239 if (unlikely(page != *pagep)) {
1249 EXPORT_SYMBOL(find_get_entry);
1252 * find_lock_entry - locate, pin and lock a page cache entry
1253 * @mapping: the address_space to search
1254 * @offset: the page cache index
1256 * Looks up the page cache slot at @mapping & @offset. If there is a
1257 * page cache page, it is returned locked and with an increased
1260 * If the slot holds a shadow entry of a previously evicted page, or a
1261 * swap entry from shmem/tmpfs, it is returned.
1263 * Otherwise, %NULL is returned.
1265 * find_lock_entry() may sleep.
1267 struct page *find_lock_entry(struct address_space *mapping, pgoff_t offset)
1272 page = find_get_entry(mapping, offset);
1273 if (page && !radix_tree_exception(page)) {
1275 /* Has the page been truncated? */
1276 if (unlikely(page_mapping(page) != mapping)) {
1281 VM_BUG_ON_PAGE(page_to_pgoff(page) != offset, page);
1285 EXPORT_SYMBOL(find_lock_entry);
1288 * pagecache_get_page - find and get a page reference
1289 * @mapping: the address_space to search
1290 * @offset: the page index
1291 * @fgp_flags: PCG flags
1292 * @gfp_mask: gfp mask to use for the page cache data page allocation
1294 * Looks up the page cache slot at @mapping & @offset.
1296 * PCG flags modify how the page is returned.
1298 * @fgp_flags can be:
1300 * - FGP_ACCESSED: the page will be marked accessed
1301 * - FGP_LOCK: Page is return locked
1302 * - FGP_CREAT: If page is not present then a new page is allocated using
1303 * @gfp_mask and added to the page cache and the VM's LRU
1304 * list. The page is returned locked and with an increased
1305 * refcount. Otherwise, NULL is returned.
1307 * If FGP_LOCK or FGP_CREAT are specified then the function may sleep even
1308 * if the GFP flags specified for FGP_CREAT are atomic.
1310 * If there is a page cache page, it is returned with an increased refcount.
1312 struct page *pagecache_get_page(struct address_space *mapping, pgoff_t offset,
1313 int fgp_flags, gfp_t gfp_mask)
1318 page = find_get_entry(mapping, offset);
1319 if (radix_tree_exceptional_entry(page))
1324 if (fgp_flags & FGP_LOCK) {
1325 if (fgp_flags & FGP_NOWAIT) {
1326 if (!trylock_page(page)) {
1334 /* Has the page been truncated? */
1335 if (unlikely(page->mapping != mapping)) {
1340 VM_BUG_ON_PAGE(page->index != offset, page);
1343 if (page && (fgp_flags & FGP_ACCESSED))
1344 mark_page_accessed(page);
1347 if (!page && (fgp_flags & FGP_CREAT)) {
1349 if ((fgp_flags & FGP_WRITE) && mapping_cap_account_dirty(mapping))
1350 gfp_mask |= __GFP_WRITE;
1351 if (fgp_flags & FGP_NOFS)
1352 gfp_mask &= ~__GFP_FS;
1354 page = __page_cache_alloc(gfp_mask);
1358 if (WARN_ON_ONCE(!(fgp_flags & FGP_LOCK)))
1359 fgp_flags |= FGP_LOCK;
1361 /* Init accessed so avoid atomic mark_page_accessed later */
1362 if (fgp_flags & FGP_ACCESSED)
1363 __SetPageReferenced(page);
1365 err = add_to_page_cache_lru(page, mapping, offset,
1366 gfp_mask & GFP_RECLAIM_MASK);
1367 if (unlikely(err)) {
1377 EXPORT_SYMBOL(pagecache_get_page);
1380 * find_get_entries - gang pagecache lookup
1381 * @mapping: The address_space to search
1382 * @start: The starting page cache index
1383 * @nr_entries: The maximum number of entries
1384 * @entries: Where the resulting entries are placed
1385 * @indices: The cache indices corresponding to the entries in @entries
1387 * find_get_entries() will search for and return a group of up to
1388 * @nr_entries entries in the mapping. The entries are placed at
1389 * @entries. find_get_entries() takes a reference against any actual
1392 * The search returns a group of mapping-contiguous page cache entries
1393 * with ascending indexes. There may be holes in the indices due to
1394 * not-present pages.
1396 * Any shadow entries of evicted pages, or swap entries from
1397 * shmem/tmpfs, are included in the returned array.
1399 * find_get_entries() returns the number of pages and shadow entries
1402 unsigned find_get_entries(struct address_space *mapping,
1403 pgoff_t start, unsigned int nr_entries,
1404 struct page **entries, pgoff_t *indices)
1407 unsigned int ret = 0;
1408 struct radix_tree_iter iter;
1414 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) {
1415 struct page *head, *page;
1417 page = radix_tree_deref_slot(slot);
1418 if (unlikely(!page))
1420 if (radix_tree_exception(page)) {
1421 if (radix_tree_deref_retry(page)) {
1422 slot = radix_tree_iter_retry(&iter);
1426 * A shadow entry of a recently evicted page, a swap
1427 * entry from shmem/tmpfs or a DAX entry. Return it
1428 * without attempting to raise page count.
1433 head = compound_head(page);
1434 if (!page_cache_get_speculative(head))
1437 /* The page was split under us? */
1438 if (compound_head(page) != head) {
1443 /* Has the page moved? */
1444 if (unlikely(page != *slot)) {
1449 indices[ret] = iter.index;
1450 entries[ret] = page;
1451 if (++ret == nr_entries)
1459 * find_get_pages - gang pagecache lookup
1460 * @mapping: The address_space to search
1461 * @start: The starting page index
1462 * @nr_pages: The maximum number of pages
1463 * @pages: Where the resulting pages are placed
1465 * find_get_pages() will search for and return a group of up to
1466 * @nr_pages pages in the mapping. The pages are placed at @pages.
1467 * find_get_pages() takes a reference against the returned pages.
1469 * The search returns a group of mapping-contiguous pages with ascending
1470 * indexes. There may be holes in the indices due to not-present pages.
1472 * find_get_pages() returns the number of pages which were found.
1474 unsigned find_get_pages(struct address_space *mapping, pgoff_t start,
1475 unsigned int nr_pages, struct page **pages)
1477 struct radix_tree_iter iter;
1481 if (unlikely(!nr_pages))
1485 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) {
1486 struct page *head, *page;
1488 page = radix_tree_deref_slot(slot);
1489 if (unlikely(!page))
1492 if (radix_tree_exception(page)) {
1493 if (radix_tree_deref_retry(page)) {
1494 slot = radix_tree_iter_retry(&iter);
1498 * A shadow entry of a recently evicted page,
1499 * or a swap entry from shmem/tmpfs. Skip
1505 head = compound_head(page);
1506 if (!page_cache_get_speculative(head))
1509 /* The page was split under us? */
1510 if (compound_head(page) != head) {
1515 /* Has the page moved? */
1516 if (unlikely(page != *slot)) {
1522 if (++ret == nr_pages)
1531 * find_get_pages_contig - gang contiguous pagecache lookup
1532 * @mapping: The address_space to search
1533 * @index: The starting page index
1534 * @nr_pages: The maximum number of pages
1535 * @pages: Where the resulting pages are placed
1537 * find_get_pages_contig() works exactly like find_get_pages(), except
1538 * that the returned number of pages are guaranteed to be contiguous.
1540 * find_get_pages_contig() returns the number of pages which were found.
1542 unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index,
1543 unsigned int nr_pages, struct page **pages)
1545 struct radix_tree_iter iter;
1547 unsigned int ret = 0;
1549 if (unlikely(!nr_pages))
1553 radix_tree_for_each_contig(slot, &mapping->page_tree, &iter, index) {
1554 struct page *head, *page;
1556 page = radix_tree_deref_slot(slot);
1557 /* The hole, there no reason to continue */
1558 if (unlikely(!page))
1561 if (radix_tree_exception(page)) {
1562 if (radix_tree_deref_retry(page)) {
1563 slot = radix_tree_iter_retry(&iter);
1567 * A shadow entry of a recently evicted page,
1568 * or a swap entry from shmem/tmpfs. Stop
1569 * looking for contiguous pages.
1574 head = compound_head(page);
1575 if (!page_cache_get_speculative(head))
1578 /* The page was split under us? */
1579 if (compound_head(page) != head) {
1584 /* Has the page moved? */
1585 if (unlikely(page != *slot)) {
1591 * must check mapping and index after taking the ref.
1592 * otherwise we can get both false positives and false
1593 * negatives, which is just confusing to the caller.
1595 if (page->mapping == NULL || page_to_pgoff(page) != iter.index) {
1601 if (++ret == nr_pages)
1607 EXPORT_SYMBOL(find_get_pages_contig);
1610 * find_get_pages_tag - find and return pages that match @tag
1611 * @mapping: the address_space to search
1612 * @index: the starting page index
1613 * @tag: the tag index
1614 * @nr_pages: the maximum number of pages
1615 * @pages: where the resulting pages are placed
1617 * Like find_get_pages, except we only return pages which are tagged with
1618 * @tag. We update @index to index the next page for the traversal.
1620 unsigned find_get_pages_tag(struct address_space *mapping, pgoff_t *index,
1621 int tag, unsigned int nr_pages, struct page **pages)
1623 struct radix_tree_iter iter;
1627 if (unlikely(!nr_pages))
1631 radix_tree_for_each_tagged(slot, &mapping->page_tree,
1632 &iter, *index, tag) {
1633 struct page *head, *page;
1635 page = radix_tree_deref_slot(slot);
1636 if (unlikely(!page))
1639 if (radix_tree_exception(page)) {
1640 if (radix_tree_deref_retry(page)) {
1641 slot = radix_tree_iter_retry(&iter);
1645 * A shadow entry of a recently evicted page.
1647 * Those entries should never be tagged, but
1648 * this tree walk is lockless and the tags are
1649 * looked up in bulk, one radix tree node at a
1650 * time, so there is a sizable window for page
1651 * reclaim to evict a page we saw tagged.
1658 head = compound_head(page);
1659 if (!page_cache_get_speculative(head))
1662 /* The page was split under us? */
1663 if (compound_head(page) != head) {
1668 /* Has the page moved? */
1669 if (unlikely(page != *slot)) {
1675 if (++ret == nr_pages)
1682 *index = pages[ret - 1]->index + 1;
1686 EXPORT_SYMBOL(find_get_pages_tag);
1689 * find_get_entries_tag - find and return entries that match @tag
1690 * @mapping: the address_space to search
1691 * @start: the starting page cache index
1692 * @tag: the tag index
1693 * @nr_entries: the maximum number of entries
1694 * @entries: where the resulting entries are placed
1695 * @indices: the cache indices corresponding to the entries in @entries
1697 * Like find_get_entries, except we only return entries which are tagged with
1700 unsigned find_get_entries_tag(struct address_space *mapping, pgoff_t start,
1701 int tag, unsigned int nr_entries,
1702 struct page **entries, pgoff_t *indices)
1705 unsigned int ret = 0;
1706 struct radix_tree_iter iter;
1712 radix_tree_for_each_tagged(slot, &mapping->page_tree,
1713 &iter, start, tag) {
1714 struct page *head, *page;
1716 page = radix_tree_deref_slot(slot);
1717 if (unlikely(!page))
1719 if (radix_tree_exception(page)) {
1720 if (radix_tree_deref_retry(page)) {
1721 slot = radix_tree_iter_retry(&iter);
1726 * A shadow entry of a recently evicted page, a swap
1727 * entry from shmem/tmpfs or a DAX entry. Return it
1728 * without attempting to raise page count.
1733 head = compound_head(page);
1734 if (!page_cache_get_speculative(head))
1737 /* The page was split under us? */
1738 if (compound_head(page) != head) {
1743 /* Has the page moved? */
1744 if (unlikely(page != *slot)) {
1749 indices[ret] = iter.index;
1750 entries[ret] = page;
1751 if (++ret == nr_entries)
1757 EXPORT_SYMBOL(find_get_entries_tag);
1760 * CD/DVDs are error prone. When a medium error occurs, the driver may fail
1761 * a _large_ part of the i/o request. Imagine the worst scenario:
1763 * ---R__________________________________________B__________
1764 * ^ reading here ^ bad block(assume 4k)
1766 * read(R) => miss => readahead(R...B) => media error => frustrating retries
1767 * => failing the whole request => read(R) => read(R+1) =>
1768 * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
1769 * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
1770 * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
1772 * It is going insane. Fix it by quickly scaling down the readahead size.
1774 static void shrink_readahead_size_eio(struct file *filp,
1775 struct file_ra_state *ra)
1781 * do_generic_file_read - generic file read routine
1782 * @filp: the file to read
1783 * @ppos: current file position
1784 * @iter: data destination
1785 * @written: already copied
1787 * This is a generic file read routine, and uses the
1788 * mapping->a_ops->readpage() function for the actual low-level stuff.
1790 * This is really ugly. But the goto's actually try to clarify some
1791 * of the logic when it comes to error handling etc.
1793 static ssize_t do_generic_file_read(struct file *filp, loff_t *ppos,
1794 struct iov_iter *iter, ssize_t written)
1796 struct address_space *mapping = filp->f_mapping;
1797 struct inode *inode = mapping->host;
1798 struct file_ra_state *ra = &filp->f_ra;
1802 unsigned long offset; /* offset into pagecache page */
1803 unsigned int prev_offset;
1806 if (unlikely(*ppos >= inode->i_sb->s_maxbytes))
1808 iov_iter_truncate(iter, inode->i_sb->s_maxbytes);
1810 index = *ppos >> PAGE_SHIFT;
1811 prev_index = ra->prev_pos >> PAGE_SHIFT;
1812 prev_offset = ra->prev_pos & (PAGE_SIZE-1);
1813 last_index = (*ppos + iter->count + PAGE_SIZE-1) >> PAGE_SHIFT;
1814 offset = *ppos & ~PAGE_MASK;
1820 unsigned long nr, ret;
1824 if (fatal_signal_pending(current)) {
1829 page = find_get_page(mapping, index);
1831 page_cache_sync_readahead(mapping,
1833 index, last_index - index);
1834 page = find_get_page(mapping, index);
1835 if (unlikely(page == NULL))
1836 goto no_cached_page;
1838 if (PageReadahead(page)) {
1839 page_cache_async_readahead(mapping,
1841 index, last_index - index);
1843 if (!PageUptodate(page)) {
1845 * See comment in do_read_cache_page on why
1846 * wait_on_page_locked is used to avoid unnecessarily
1847 * serialisations and why it's safe.
1849 error = wait_on_page_locked_killable(page);
1850 if (unlikely(error))
1851 goto readpage_error;
1852 if (PageUptodate(page))
1855 if (inode->i_blkbits == PAGE_SHIFT ||
1856 !mapping->a_ops->is_partially_uptodate)
1857 goto page_not_up_to_date;
1858 /* pipes can't handle partially uptodate pages */
1859 if (unlikely(iter->type & ITER_PIPE))
1860 goto page_not_up_to_date;
1861 if (!trylock_page(page))
1862 goto page_not_up_to_date;
1863 /* Did it get truncated before we got the lock? */
1865 goto page_not_up_to_date_locked;
1866 if (!mapping->a_ops->is_partially_uptodate(page,
1867 offset, iter->count))
1868 goto page_not_up_to_date_locked;
1873 * i_size must be checked after we know the page is Uptodate.
1875 * Checking i_size after the check allows us to calculate
1876 * the correct value for "nr", which means the zero-filled
1877 * part of the page is not copied back to userspace (unless
1878 * another truncate extends the file - this is desired though).
1881 isize = i_size_read(inode);
1882 end_index = (isize - 1) >> PAGE_SHIFT;
1883 if (unlikely(!isize || index > end_index)) {
1888 /* nr is the maximum number of bytes to copy from this page */
1890 if (index == end_index) {
1891 nr = ((isize - 1) & ~PAGE_MASK) + 1;
1899 /* If users can be writing to this page using arbitrary
1900 * virtual addresses, take care about potential aliasing
1901 * before reading the page on the kernel side.
1903 if (mapping_writably_mapped(mapping))
1904 flush_dcache_page(page);
1907 * When a sequential read accesses a page several times,
1908 * only mark it as accessed the first time.
1910 if (prev_index != index || offset != prev_offset)
1911 mark_page_accessed(page);
1915 * Ok, we have the page, and it's up-to-date, so
1916 * now we can copy it to user space...
1919 ret = copy_page_to_iter(page, offset, nr, iter);
1921 index += offset >> PAGE_SHIFT;
1922 offset &= ~PAGE_MASK;
1923 prev_offset = offset;
1927 if (!iov_iter_count(iter))
1935 page_not_up_to_date:
1936 /* Get exclusive access to the page ... */
1937 error = lock_page_killable(page);
1938 if (unlikely(error))
1939 goto readpage_error;
1941 page_not_up_to_date_locked:
1942 /* Did it get truncated before we got the lock? */
1943 if (!page->mapping) {
1949 /* Did somebody else fill it already? */
1950 if (PageUptodate(page)) {
1957 * A previous I/O error may have been due to temporary
1958 * failures, eg. multipath errors.
1959 * PG_error will be set again if readpage fails.
1961 ClearPageError(page);
1962 /* Start the actual read. The read will unlock the page. */
1963 error = mapping->a_ops->readpage(filp, page);
1965 if (unlikely(error)) {
1966 if (error == AOP_TRUNCATED_PAGE) {
1971 goto readpage_error;
1974 if (!PageUptodate(page)) {
1975 error = lock_page_killable(page);
1976 if (unlikely(error))
1977 goto readpage_error;
1978 if (!PageUptodate(page)) {
1979 if (page->mapping == NULL) {
1981 * invalidate_mapping_pages got it
1988 shrink_readahead_size_eio(filp, ra);
1990 goto readpage_error;
1998 /* UHHUH! A synchronous read error occurred. Report it */
2004 * Ok, it wasn't cached, so we need to create a new
2007 page = page_cache_alloc_cold(mapping);
2012 error = add_to_page_cache_lru(page, mapping, index,
2013 mapping_gfp_constraint(mapping, GFP_KERNEL));
2016 if (error == -EEXIST) {
2026 ra->prev_pos = prev_index;
2027 ra->prev_pos <<= PAGE_SHIFT;
2028 ra->prev_pos |= prev_offset;
2030 *ppos = ((loff_t)index << PAGE_SHIFT) + offset;
2031 file_accessed(filp);
2032 return written ? written : error;
2036 * generic_file_read_iter - generic filesystem read routine
2037 * @iocb: kernel I/O control block
2038 * @iter: destination for the data read
2040 * This is the "read_iter()" routine for all filesystems
2041 * that can use the page cache directly.
2044 generic_file_read_iter(struct kiocb *iocb, struct iov_iter *iter)
2046 struct file *file = iocb->ki_filp;
2048 size_t count = iov_iter_count(iter);
2051 goto out; /* skip atime */
2053 if (iocb->ki_flags & IOCB_DIRECT) {
2054 struct address_space *mapping = file->f_mapping;
2055 struct inode *inode = mapping->host;
2058 size = i_size_read(inode);
2059 retval = filemap_write_and_wait_range(mapping, iocb->ki_pos,
2060 iocb->ki_pos + count - 1);
2064 file_accessed(file);
2066 retval = mapping->a_ops->direct_IO(iocb, iter);
2068 iocb->ki_pos += retval;
2071 iov_iter_revert(iter, count - iov_iter_count(iter));
2074 * Btrfs can have a short DIO read if we encounter
2075 * compressed extents, so if there was an error, or if
2076 * we've already read everything we wanted to, or if
2077 * there was a short read because we hit EOF, go ahead
2078 * and return. Otherwise fallthrough to buffered io for
2079 * the rest of the read. Buffered reads will not work for
2080 * DAX files, so don't bother trying.
2082 if (retval < 0 || !count || iocb->ki_pos >= size ||
2087 retval = do_generic_file_read(file, &iocb->ki_pos, iter, retval);
2091 EXPORT_SYMBOL(generic_file_read_iter);
2095 * page_cache_read - adds requested page to the page cache if not already there
2096 * @file: file to read
2097 * @offset: page index
2098 * @gfp_mask: memory allocation flags
2100 * This adds the requested page to the page cache if it isn't already there,
2101 * and schedules an I/O to read in its contents from disk.
2103 static int page_cache_read(struct file *file, pgoff_t offset, gfp_t gfp_mask)
2105 struct address_space *mapping = file->f_mapping;
2110 page = __page_cache_alloc(gfp_mask|__GFP_COLD);
2114 ret = add_to_page_cache_lru(page, mapping, offset, gfp_mask & GFP_KERNEL);
2116 ret = mapping->a_ops->readpage(file, page);
2117 else if (ret == -EEXIST)
2118 ret = 0; /* losing race to add is OK */
2122 } while (ret == AOP_TRUNCATED_PAGE);
2127 #define MMAP_LOTSAMISS (100)
2130 * Synchronous readahead happens when we don't even find
2131 * a page in the page cache at all.
2133 static void do_sync_mmap_readahead(struct vm_area_struct *vma,
2134 struct file_ra_state *ra,
2138 struct address_space *mapping = file->f_mapping;
2140 /* If we don't want any read-ahead, don't bother */
2141 if (vma->vm_flags & VM_RAND_READ)
2146 if (vma->vm_flags & VM_SEQ_READ) {
2147 page_cache_sync_readahead(mapping, ra, file, offset,
2152 /* Avoid banging the cache line if not needed */
2153 if (ra->mmap_miss < MMAP_LOTSAMISS * 10)
2157 * Do we miss much more than hit in this file? If so,
2158 * stop bothering with read-ahead. It will only hurt.
2160 if (ra->mmap_miss > MMAP_LOTSAMISS)
2166 ra->start = max_t(long, 0, offset - ra->ra_pages / 2);
2167 ra->size = ra->ra_pages;
2168 ra->async_size = ra->ra_pages / 4;
2169 ra_submit(ra, mapping, file);
2173 * Asynchronous readahead happens when we find the page and PG_readahead,
2174 * so we want to possibly extend the readahead further..
2176 static void do_async_mmap_readahead(struct vm_area_struct *vma,
2177 struct file_ra_state *ra,
2182 struct address_space *mapping = file->f_mapping;
2184 /* If we don't want any read-ahead, don't bother */
2185 if (vma->vm_flags & VM_RAND_READ)
2187 if (ra->mmap_miss > 0)
2189 if (PageReadahead(page))
2190 page_cache_async_readahead(mapping, ra, file,
2191 page, offset, ra->ra_pages);
2195 * filemap_fault - read in file data for page fault handling
2196 * @vmf: struct vm_fault containing details of the fault
2198 * filemap_fault() is invoked via the vma operations vector for a
2199 * mapped memory region to read in file data during a page fault.
2201 * The goto's are kind of ugly, but this streamlines the normal case of having
2202 * it in the page cache, and handles the special cases reasonably without
2203 * having a lot of duplicated code.
2205 * vma->vm_mm->mmap_sem must be held on entry.
2207 * If our return value has VM_FAULT_RETRY set, it's because
2208 * lock_page_or_retry() returned 0.
2209 * The mmap_sem has usually been released in this case.
2210 * See __lock_page_or_retry() for the exception.
2212 * If our return value does not have VM_FAULT_RETRY set, the mmap_sem
2213 * has not been released.
2215 * We never return with VM_FAULT_RETRY and a bit from VM_FAULT_ERROR set.
2217 int filemap_fault(struct vm_fault *vmf)
2220 struct file *file = vmf->vma->vm_file;
2221 struct address_space *mapping = file->f_mapping;
2222 struct file_ra_state *ra = &file->f_ra;
2223 struct inode *inode = mapping->host;
2224 pgoff_t offset = vmf->pgoff;
2229 max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
2230 if (unlikely(offset >= max_off))
2231 return VM_FAULT_SIGBUS;
2234 * Do we have something in the page cache already?
2236 page = find_get_page(mapping, offset);
2237 if (likely(page) && !(vmf->flags & FAULT_FLAG_TRIED)) {
2239 * We found the page, so try async readahead before
2240 * waiting for the lock.
2242 do_async_mmap_readahead(vmf->vma, ra, file, page, offset);
2244 /* No page in the page cache at all */
2245 do_sync_mmap_readahead(vmf->vma, ra, file, offset);
2246 count_vm_event(PGMAJFAULT);
2247 mem_cgroup_count_vm_event(vmf->vma->vm_mm, PGMAJFAULT);
2248 ret = VM_FAULT_MAJOR;
2250 page = find_get_page(mapping, offset);
2252 goto no_cached_page;
2255 if (!lock_page_or_retry(page, vmf->vma->vm_mm, vmf->flags)) {
2257 return ret | VM_FAULT_RETRY;
2260 /* Did it get truncated? */
2261 if (unlikely(page->mapping != mapping)) {
2266 VM_BUG_ON_PAGE(page->index != offset, page);
2269 * We have a locked page in the page cache, now we need to check
2270 * that it's up-to-date. If not, it is going to be due to an error.
2272 if (unlikely(!PageUptodate(page)))
2273 goto page_not_uptodate;
2276 * Found the page and have a reference on it.
2277 * We must recheck i_size under page lock.
2279 max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
2280 if (unlikely(offset >= max_off)) {
2283 return VM_FAULT_SIGBUS;
2287 return ret | VM_FAULT_LOCKED;
2291 * We're only likely to ever get here if MADV_RANDOM is in
2294 error = page_cache_read(file, offset, vmf->gfp_mask);
2297 * The page we want has now been added to the page cache.
2298 * In the unlikely event that someone removed it in the
2299 * meantime, we'll just come back here and read it again.
2305 * An error return from page_cache_read can result if the
2306 * system is low on memory, or a problem occurs while trying
2309 if (error == -ENOMEM)
2310 return VM_FAULT_OOM;
2311 return VM_FAULT_SIGBUS;
2315 * Umm, take care of errors if the page isn't up-to-date.
2316 * Try to re-read it _once_. We do this synchronously,
2317 * because there really aren't any performance issues here
2318 * and we need to check for errors.
2320 ClearPageError(page);
2321 error = mapping->a_ops->readpage(file, page);
2323 wait_on_page_locked(page);
2324 if (!PageUptodate(page))
2329 if (!error || error == AOP_TRUNCATED_PAGE)
2332 /* Things didn't work out. Return zero to tell the mm layer so. */
2333 shrink_readahead_size_eio(file, ra);
2334 return VM_FAULT_SIGBUS;
2336 EXPORT_SYMBOL(filemap_fault);
2338 void filemap_map_pages(struct vm_fault *vmf,
2339 pgoff_t start_pgoff, pgoff_t end_pgoff)
2341 struct radix_tree_iter iter;
2343 struct file *file = vmf->vma->vm_file;
2344 struct address_space *mapping = file->f_mapping;
2345 pgoff_t last_pgoff = start_pgoff;
2346 unsigned long max_idx;
2347 struct page *head, *page;
2350 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter,
2352 if (iter.index > end_pgoff)
2355 page = radix_tree_deref_slot(slot);
2356 if (unlikely(!page))
2358 if (radix_tree_exception(page)) {
2359 if (radix_tree_deref_retry(page)) {
2360 slot = radix_tree_iter_retry(&iter);
2366 head = compound_head(page);
2367 if (!page_cache_get_speculative(head))
2370 /* The page was split under us? */
2371 if (compound_head(page) != head) {
2376 /* Has the page moved? */
2377 if (unlikely(page != *slot)) {
2382 if (!PageUptodate(page) ||
2383 PageReadahead(page) ||
2386 if (!trylock_page(page))
2389 if (page->mapping != mapping || !PageUptodate(page))
2392 max_idx = DIV_ROUND_UP(i_size_read(mapping->host), PAGE_SIZE);
2393 if (page->index >= max_idx)
2396 if (file->f_ra.mmap_miss > 0)
2397 file->f_ra.mmap_miss--;
2399 vmf->address += (iter.index - last_pgoff) << PAGE_SHIFT;
2401 vmf->pte += iter.index - last_pgoff;
2402 last_pgoff = iter.index;
2403 if (alloc_set_pte(vmf, NULL, page))
2412 /* Huge page is mapped? No need to proceed. */
2413 if (pmd_trans_huge(*vmf->pmd))
2415 if (iter.index == end_pgoff)
2420 EXPORT_SYMBOL(filemap_map_pages);
2422 int filemap_page_mkwrite(struct vm_fault *vmf)
2424 struct page *page = vmf->page;
2425 struct inode *inode = file_inode(vmf->vma->vm_file);
2426 int ret = VM_FAULT_LOCKED;
2428 sb_start_pagefault(inode->i_sb);
2429 file_update_time(vmf->vma->vm_file);
2431 if (page->mapping != inode->i_mapping) {
2433 ret = VM_FAULT_NOPAGE;
2437 * We mark the page dirty already here so that when freeze is in
2438 * progress, we are guaranteed that writeback during freezing will
2439 * see the dirty page and writeprotect it again.
2441 set_page_dirty(page);
2442 wait_for_stable_page(page);
2444 sb_end_pagefault(inode->i_sb);
2447 EXPORT_SYMBOL(filemap_page_mkwrite);
2449 const struct vm_operations_struct generic_file_vm_ops = {
2450 .fault = filemap_fault,
2451 .map_pages = filemap_map_pages,
2452 .page_mkwrite = filemap_page_mkwrite,
2455 /* This is used for a general mmap of a disk file */
2457 int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
2459 struct address_space *mapping = file->f_mapping;
2461 if (!mapping->a_ops->readpage)
2463 file_accessed(file);
2464 vma->vm_ops = &generic_file_vm_ops;
2469 * This is for filesystems which do not implement ->writepage.
2471 int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
2473 if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE))
2475 return generic_file_mmap(file, vma);
2478 int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
2482 int generic_file_readonly_mmap(struct file * file, struct vm_area_struct * vma)
2486 #endif /* CONFIG_MMU */
2488 EXPORT_SYMBOL(generic_file_mmap);
2489 EXPORT_SYMBOL(generic_file_readonly_mmap);
2491 static struct page *wait_on_page_read(struct page *page)
2493 if (!IS_ERR(page)) {
2494 wait_on_page_locked(page);
2495 if (!PageUptodate(page)) {
2497 page = ERR_PTR(-EIO);
2503 static struct page *do_read_cache_page(struct address_space *mapping,
2505 int (*filler)(void *, struct page *),
2512 page = find_get_page(mapping, index);
2514 page = __page_cache_alloc(gfp | __GFP_COLD);
2516 return ERR_PTR(-ENOMEM);
2517 err = add_to_page_cache_lru(page, mapping, index, gfp);
2518 if (unlikely(err)) {
2522 /* Presumably ENOMEM for radix tree node */
2523 return ERR_PTR(err);
2527 err = filler(data, page);
2530 return ERR_PTR(err);
2533 page = wait_on_page_read(page);
2538 if (PageUptodate(page))
2542 * Page is not up to date and may be locked due one of the following
2543 * case a: Page is being filled and the page lock is held
2544 * case b: Read/write error clearing the page uptodate status
2545 * case c: Truncation in progress (page locked)
2546 * case d: Reclaim in progress
2548 * Case a, the page will be up to date when the page is unlocked.
2549 * There is no need to serialise on the page lock here as the page
2550 * is pinned so the lock gives no additional protection. Even if the
2551 * the page is truncated, the data is still valid if PageUptodate as
2552 * it's a race vs truncate race.
2553 * Case b, the page will not be up to date
2554 * Case c, the page may be truncated but in itself, the data may still
2555 * be valid after IO completes as it's a read vs truncate race. The
2556 * operation must restart if the page is not uptodate on unlock but
2557 * otherwise serialising on page lock to stabilise the mapping gives
2558 * no additional guarantees to the caller as the page lock is
2559 * released before return.
2560 * Case d, similar to truncation. If reclaim holds the page lock, it
2561 * will be a race with remove_mapping that determines if the mapping
2562 * is valid on unlock but otherwise the data is valid and there is
2563 * no need to serialise with page lock.
2565 * As the page lock gives no additional guarantee, we optimistically
2566 * wait on the page to be unlocked and check if it's up to date and
2567 * use the page if it is. Otherwise, the page lock is required to
2568 * distinguish between the different cases. The motivation is that we
2569 * avoid spurious serialisations and wakeups when multiple processes
2570 * wait on the same page for IO to complete.
2572 wait_on_page_locked(page);
2573 if (PageUptodate(page))
2576 /* Distinguish between all the cases under the safety of the lock */
2579 /* Case c or d, restart the operation */
2580 if (!page->mapping) {
2586 /* Someone else locked and filled the page in a very small window */
2587 if (PageUptodate(page)) {
2594 mark_page_accessed(page);
2599 * read_cache_page - read into page cache, fill it if needed
2600 * @mapping: the page's address_space
2601 * @index: the page index
2602 * @filler: function to perform the read
2603 * @data: first arg to filler(data, page) function, often left as NULL
2605 * Read into the page cache. If a page already exists, and PageUptodate() is
2606 * not set, try to fill the page and wait for it to become unlocked.
2608 * If the page does not get brought uptodate, return -EIO.
2610 struct page *read_cache_page(struct address_space *mapping,
2612 int (*filler)(void *, struct page *),
2615 return do_read_cache_page(mapping, index, filler, data, mapping_gfp_mask(mapping));
2617 EXPORT_SYMBOL(read_cache_page);
2620 * read_cache_page_gfp - read into page cache, using specified page allocation flags.
2621 * @mapping: the page's address_space
2622 * @index: the page index
2623 * @gfp: the page allocator flags to use if allocating
2625 * This is the same as "read_mapping_page(mapping, index, NULL)", but with
2626 * any new page allocations done using the specified allocation flags.
2628 * If the page does not get brought uptodate, return -EIO.
2630 struct page *read_cache_page_gfp(struct address_space *mapping,
2634 filler_t *filler = (filler_t *)mapping->a_ops->readpage;
2636 return do_read_cache_page(mapping, index, filler, NULL, gfp);
2638 EXPORT_SYMBOL(read_cache_page_gfp);
2641 * Performs necessary checks before doing a write
2643 * Can adjust writing position or amount of bytes to write.
2644 * Returns appropriate error code that caller should return or
2645 * zero in case that write should be allowed.
2647 inline ssize_t generic_write_checks(struct kiocb *iocb, struct iov_iter *from)
2649 struct file *file = iocb->ki_filp;
2650 struct inode *inode = file->f_mapping->host;
2651 unsigned long limit = rlimit(RLIMIT_FSIZE);
2654 if (!iov_iter_count(from))
2657 /* FIXME: this is for backwards compatibility with 2.4 */
2658 if (iocb->ki_flags & IOCB_APPEND)
2659 iocb->ki_pos = i_size_read(inode);
2663 if (limit != RLIM_INFINITY) {
2664 if (iocb->ki_pos >= limit) {
2665 send_sig(SIGXFSZ, current, 0);
2668 iov_iter_truncate(from, limit - (unsigned long)pos);
2674 if (unlikely(pos + iov_iter_count(from) > MAX_NON_LFS &&
2675 !(file->f_flags & O_LARGEFILE))) {
2676 if (pos >= MAX_NON_LFS)
2678 iov_iter_truncate(from, MAX_NON_LFS - (unsigned long)pos);
2682 * Are we about to exceed the fs block limit ?
2684 * If we have written data it becomes a short write. If we have
2685 * exceeded without writing data we send a signal and return EFBIG.
2686 * Linus frestrict idea will clean these up nicely..
2688 if (unlikely(pos >= inode->i_sb->s_maxbytes))
2691 iov_iter_truncate(from, inode->i_sb->s_maxbytes - pos);
2692 return iov_iter_count(from);
2694 EXPORT_SYMBOL(generic_write_checks);
2696 int pagecache_write_begin(struct file *file, struct address_space *mapping,
2697 loff_t pos, unsigned len, unsigned flags,
2698 struct page **pagep, void **fsdata)
2700 const struct address_space_operations *aops = mapping->a_ops;
2702 return aops->write_begin(file, mapping, pos, len, flags,
2705 EXPORT_SYMBOL(pagecache_write_begin);
2707 int pagecache_write_end(struct file *file, struct address_space *mapping,
2708 loff_t pos, unsigned len, unsigned copied,
2709 struct page *page, void *fsdata)
2711 const struct address_space_operations *aops = mapping->a_ops;
2713 return aops->write_end(file, mapping, pos, len, copied, page, fsdata);
2715 EXPORT_SYMBOL(pagecache_write_end);
2718 generic_file_direct_write(struct kiocb *iocb, struct iov_iter *from)
2720 struct file *file = iocb->ki_filp;
2721 struct address_space *mapping = file->f_mapping;
2722 struct inode *inode = mapping->host;
2723 loff_t pos = iocb->ki_pos;
2728 write_len = iov_iter_count(from);
2729 end = (pos + write_len - 1) >> PAGE_SHIFT;
2731 written = filemap_write_and_wait_range(mapping, pos, pos + write_len - 1);
2736 * After a write we want buffered reads to be sure to go to disk to get
2737 * the new data. We invalidate clean cached page from the region we're
2738 * about to write. We do this *before* the write so that we can return
2739 * without clobbering -EIOCBQUEUED from ->direct_IO().
2741 written = invalidate_inode_pages2_range(mapping,
2742 pos >> PAGE_SHIFT, end);
2744 * If a page can not be invalidated, return 0 to fall back
2745 * to buffered write.
2748 if (written == -EBUSY)
2753 written = mapping->a_ops->direct_IO(iocb, from);
2756 * Finally, try again to invalidate clean pages which might have been
2757 * cached by non-direct readahead, or faulted in by get_user_pages()
2758 * if the source of the write was an mmap'ed region of the file
2759 * we're writing. Either one is a pretty crazy thing to do,
2760 * so we don't support it 100%. If this invalidation
2761 * fails, tough, the write still worked...
2763 invalidate_inode_pages2_range(mapping,
2764 pos >> PAGE_SHIFT, end);
2768 write_len -= written;
2769 if (pos > i_size_read(inode) && !S_ISBLK(inode->i_mode)) {
2770 i_size_write(inode, pos);
2771 mark_inode_dirty(inode);
2775 iov_iter_revert(from, write_len - iov_iter_count(from));
2779 EXPORT_SYMBOL(generic_file_direct_write);
2782 * Find or create a page at the given pagecache position. Return the locked
2783 * page. This function is specifically for buffered writes.
2785 struct page *grab_cache_page_write_begin(struct address_space *mapping,
2786 pgoff_t index, unsigned flags)
2789 int fgp_flags = FGP_LOCK|FGP_WRITE|FGP_CREAT;
2791 if (flags & AOP_FLAG_NOFS)
2792 fgp_flags |= FGP_NOFS;
2794 page = pagecache_get_page(mapping, index, fgp_flags,
2795 mapping_gfp_mask(mapping));
2797 wait_for_stable_page(page);
2801 EXPORT_SYMBOL(grab_cache_page_write_begin);
2803 ssize_t generic_perform_write(struct file *file,
2804 struct iov_iter *i, loff_t pos)
2806 struct address_space *mapping = file->f_mapping;
2807 const struct address_space_operations *a_ops = mapping->a_ops;
2809 ssize_t written = 0;
2810 unsigned int flags = 0;
2814 unsigned long offset; /* Offset into pagecache page */
2815 unsigned long bytes; /* Bytes to write to page */
2816 size_t copied; /* Bytes copied from user */
2819 offset = (pos & (PAGE_SIZE - 1));
2820 bytes = min_t(unsigned long, PAGE_SIZE - offset,
2825 * Bring in the user page that we will copy from _first_.
2826 * Otherwise there's a nasty deadlock on copying from the
2827 * same page as we're writing to, without it being marked
2830 * Not only is this an optimisation, but it is also required
2831 * to check that the address is actually valid, when atomic
2832 * usercopies are used, below.
2834 if (unlikely(iov_iter_fault_in_readable(i, bytes))) {
2839 if (fatal_signal_pending(current)) {
2844 status = a_ops->write_begin(file, mapping, pos, bytes, flags,
2846 if (unlikely(status < 0))
2849 if (mapping_writably_mapped(mapping))
2850 flush_dcache_page(page);
2852 copied = iov_iter_copy_from_user_atomic(page, i, offset, bytes);
2853 flush_dcache_page(page);
2855 status = a_ops->write_end(file, mapping, pos, bytes, copied,
2857 if (unlikely(status < 0))
2863 iov_iter_advance(i, copied);
2864 if (unlikely(copied == 0)) {
2866 * If we were unable to copy any data at all, we must
2867 * fall back to a single segment length write.
2869 * If we didn't fallback here, we could livelock
2870 * because not all segments in the iov can be copied at
2871 * once without a pagefault.
2873 bytes = min_t(unsigned long, PAGE_SIZE - offset,
2874 iov_iter_single_seg_count(i));
2880 balance_dirty_pages_ratelimited(mapping);
2881 } while (iov_iter_count(i));
2883 return written ? written : status;
2885 EXPORT_SYMBOL(generic_perform_write);
2888 * __generic_file_write_iter - write data to a file
2889 * @iocb: IO state structure (file, offset, etc.)
2890 * @from: iov_iter with data to write
2892 * This function does all the work needed for actually writing data to a
2893 * file. It does all basic checks, removes SUID from the file, updates
2894 * modification times and calls proper subroutines depending on whether we
2895 * do direct IO or a standard buffered write.
2897 * It expects i_mutex to be grabbed unless we work on a block device or similar
2898 * object which does not need locking at all.
2900 * This function does *not* take care of syncing data in case of O_SYNC write.
2901 * A caller has to handle it. This is mainly due to the fact that we want to
2902 * avoid syncing under i_mutex.
2904 ssize_t __generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
2906 struct file *file = iocb->ki_filp;
2907 struct address_space * mapping = file->f_mapping;
2908 struct inode *inode = mapping->host;
2909 ssize_t written = 0;
2913 /* We can write back this queue in page reclaim */
2914 current->backing_dev_info = inode_to_bdi(inode);
2915 err = file_remove_privs(file);
2919 err = file_update_time(file);
2923 if (iocb->ki_flags & IOCB_DIRECT) {
2924 loff_t pos, endbyte;
2926 written = generic_file_direct_write(iocb, from);
2928 * If the write stopped short of completing, fall back to
2929 * buffered writes. Some filesystems do this for writes to
2930 * holes, for example. For DAX files, a buffered write will
2931 * not succeed (even if it did, DAX does not handle dirty
2932 * page-cache pages correctly).
2934 if (written < 0 || !iov_iter_count(from) || IS_DAX(inode))
2937 status = generic_perform_write(file, from, pos = iocb->ki_pos);
2939 * If generic_perform_write() returned a synchronous error
2940 * then we want to return the number of bytes which were
2941 * direct-written, or the error code if that was zero. Note
2942 * that this differs from normal direct-io semantics, which
2943 * will return -EFOO even if some bytes were written.
2945 if (unlikely(status < 0)) {
2950 * We need to ensure that the page cache pages are written to
2951 * disk and invalidated to preserve the expected O_DIRECT
2954 endbyte = pos + status - 1;
2955 err = filemap_write_and_wait_range(mapping, pos, endbyte);
2957 iocb->ki_pos = endbyte + 1;
2959 invalidate_mapping_pages(mapping,
2961 endbyte >> PAGE_SHIFT);
2964 * We don't know how much we wrote, so just return
2965 * the number of bytes which were direct-written
2969 written = generic_perform_write(file, from, iocb->ki_pos);
2970 if (likely(written > 0))
2971 iocb->ki_pos += written;
2974 current->backing_dev_info = NULL;
2975 return written ? written : err;
2977 EXPORT_SYMBOL(__generic_file_write_iter);
2980 * generic_file_write_iter - write data to a file
2981 * @iocb: IO state structure
2982 * @from: iov_iter with data to write
2984 * This is a wrapper around __generic_file_write_iter() to be used by most
2985 * filesystems. It takes care of syncing the file in case of O_SYNC file
2986 * and acquires i_mutex as needed.
2988 ssize_t generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
2990 struct file *file = iocb->ki_filp;
2991 struct inode *inode = file->f_mapping->host;
2995 ret = generic_write_checks(iocb, from);
2997 ret = __generic_file_write_iter(iocb, from);
2998 inode_unlock(inode);
3001 ret = generic_write_sync(iocb, ret);
3004 EXPORT_SYMBOL(generic_file_write_iter);
3007 * try_to_release_page() - release old fs-specific metadata on a page
3009 * @page: the page which the kernel is trying to free
3010 * @gfp_mask: memory allocation flags (and I/O mode)
3012 * The address_space is to try to release any data against the page
3013 * (presumably at page->private). If the release was successful, return '1'.
3014 * Otherwise return zero.
3016 * This may also be called if PG_fscache is set on a page, indicating that the
3017 * page is known to the local caching routines.
3019 * The @gfp_mask argument specifies whether I/O may be performed to release
3020 * this page (__GFP_IO), and whether the call may block (__GFP_RECLAIM & __GFP_FS).
3023 int try_to_release_page(struct page *page, gfp_t gfp_mask)
3025 struct address_space * const mapping = page->mapping;
3027 BUG_ON(!PageLocked(page));
3028 if (PageWriteback(page))
3031 if (mapping && mapping->a_ops->releasepage)
3032 return mapping->a_ops->releasepage(page, gfp_mask);
3033 return try_to_free_buffers(page);
3036 EXPORT_SYMBOL(try_to_release_page);