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 workingset_node_shadows_dec(node);
138 /* DAX can replace empty locked entry with a hole */
140 (void *)(RADIX_TREE_EXCEPTIONAL_ENTRY |
141 RADIX_DAX_ENTRY_LOCK));
142 /* DAX accounts exceptional entries as normal pages */
144 workingset_node_pages_dec(node);
145 /* Wakeup waiters for exceptional entry lock */
146 dax_wake_mapping_entry_waiter(mapping, page->index,
150 radix_tree_replace_slot(slot, page);
153 workingset_node_pages_inc(node);
155 * Don't track node that contains actual pages.
157 * Avoid acquiring the list_lru lock if already
158 * untracked. The list_empty() test is safe as
159 * node->private_list is protected by
160 * mapping->tree_lock.
162 if (!list_empty(&node->private_list))
163 list_lru_del(&workingset_shadow_nodes,
164 &node->private_list);
169 static void page_cache_tree_delete(struct address_space *mapping,
170 struct page *page, void *shadow)
172 int i, nr = PageHuge(page) ? 1 : hpage_nr_pages(page);
174 VM_BUG_ON_PAGE(!PageLocked(page), page);
175 VM_BUG_ON_PAGE(PageTail(page), page);
176 VM_BUG_ON_PAGE(nr != 1 && shadow, page);
178 for (i = 0; i < nr; i++) {
179 struct radix_tree_node *node;
182 __radix_tree_lookup(&mapping->page_tree, page->index + i,
185 radix_tree_clear_tags(&mapping->page_tree, node, slot);
188 VM_BUG_ON_PAGE(nr != 1, page);
190 * We need a node to properly account shadow
191 * entries. Don't plant any without. XXX
196 radix_tree_replace_slot(slot, shadow);
201 workingset_node_pages_dec(node);
203 workingset_node_shadows_inc(node);
205 if (__radix_tree_delete_node(&mapping->page_tree, node))
209 * Track node that only contains shadow entries. DAX mappings
210 * contain no shadow entries and may contain other exceptional
211 * entries so skip those.
213 * Avoid acquiring the list_lru lock if already tracked.
214 * The list_empty() test is safe as node->private_list is
215 * protected by mapping->tree_lock.
217 if (!dax_mapping(mapping) && !workingset_node_pages(node) &&
218 list_empty(&node->private_list)) {
219 node->private_data = mapping;
220 list_lru_add(&workingset_shadow_nodes,
221 &node->private_list);
226 mapping->nrexceptional += nr;
228 * Make sure the nrexceptional update is committed before
229 * the nrpages update so that final truncate racing
230 * with reclaim does not see both counters 0 at the
231 * same time and miss a shadow entry.
235 mapping->nrpages -= nr;
239 * Delete a page from the page cache and free it. Caller has to make
240 * sure the page is locked and that nobody else uses it - or that usage
241 * is safe. The caller must hold the mapping's tree_lock.
243 void __delete_from_page_cache(struct page *page, void *shadow)
245 struct address_space *mapping = page->mapping;
246 int nr = hpage_nr_pages(page);
248 trace_mm_filemap_delete_from_page_cache(page);
250 * if we're uptodate, flush out into the cleancache, otherwise
251 * invalidate any existing cleancache entries. We can't leave
252 * stale data around in the cleancache once our page is gone
254 if (PageUptodate(page) && PageMappedToDisk(page))
255 cleancache_put_page(page);
257 cleancache_invalidate_page(mapping, page);
259 VM_BUG_ON_PAGE(PageTail(page), page);
260 VM_BUG_ON_PAGE(page_mapped(page), page);
261 if (!IS_ENABLED(CONFIG_DEBUG_VM) && unlikely(page_mapped(page))) {
264 pr_alert("BUG: Bad page cache in process %s pfn:%05lx\n",
265 current->comm, page_to_pfn(page));
266 dump_page(page, "still mapped when deleted");
268 add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE);
270 mapcount = page_mapcount(page);
271 if (mapping_exiting(mapping) &&
272 page_count(page) >= mapcount + 2) {
274 * All vmas have already been torn down, so it's
275 * a good bet that actually the page is unmapped,
276 * and we'd prefer not to leak it: if we're wrong,
277 * some other bad page check should catch it later.
279 page_mapcount_reset(page);
280 page_ref_sub(page, mapcount);
284 page_cache_tree_delete(mapping, page, shadow);
286 page->mapping = NULL;
287 /* Leave page->index set: truncation lookup relies upon it */
289 /* hugetlb pages do not participate in page cache accounting. */
291 __mod_node_page_state(page_pgdat(page), NR_FILE_PAGES, -nr);
292 if (PageSwapBacked(page)) {
293 __mod_node_page_state(page_pgdat(page), NR_SHMEM, -nr);
294 if (PageTransHuge(page))
295 __dec_node_page_state(page, NR_SHMEM_THPS);
297 VM_BUG_ON_PAGE(PageTransHuge(page) && !PageHuge(page), page);
301 * At this point page must be either written or cleaned by truncate.
302 * Dirty page here signals a bug and loss of unwritten data.
304 * This fixes dirty accounting after removing the page entirely but
305 * leaves PageDirty set: it has no effect for truncated page and
306 * anyway will be cleared before returning page into buddy allocator.
308 if (WARN_ON_ONCE(PageDirty(page)))
309 account_page_cleaned(page, mapping, inode_to_wb(mapping->host));
313 * delete_from_page_cache - delete page from page cache
314 * @page: the page which the kernel is trying to remove from page cache
316 * This must be called only on pages that have been verified to be in the page
317 * cache and locked. It will never put the page into the free list, the caller
318 * has a reference on the page.
320 void delete_from_page_cache(struct page *page)
322 struct address_space *mapping = page_mapping(page);
324 void (*freepage)(struct page *);
326 BUG_ON(!PageLocked(page));
328 freepage = mapping->a_ops->freepage;
330 spin_lock_irqsave(&mapping->tree_lock, flags);
331 __delete_from_page_cache(page, NULL);
332 spin_unlock_irqrestore(&mapping->tree_lock, flags);
337 if (PageTransHuge(page) && !PageHuge(page)) {
338 page_ref_sub(page, HPAGE_PMD_NR);
339 VM_BUG_ON_PAGE(page_count(page) <= 0, page);
344 EXPORT_SYMBOL(delete_from_page_cache);
346 int filemap_check_errors(struct address_space *mapping)
349 /* Check for outstanding write errors */
350 if (test_bit(AS_ENOSPC, &mapping->flags) &&
351 test_and_clear_bit(AS_ENOSPC, &mapping->flags))
353 if (test_bit(AS_EIO, &mapping->flags) &&
354 test_and_clear_bit(AS_EIO, &mapping->flags))
358 EXPORT_SYMBOL(filemap_check_errors);
361 * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range
362 * @mapping: address space structure to write
363 * @start: offset in bytes where the range starts
364 * @end: offset in bytes where the range ends (inclusive)
365 * @sync_mode: enable synchronous operation
367 * Start writeback against all of a mapping's dirty pages that lie
368 * within the byte offsets <start, end> inclusive.
370 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
371 * opposed to a regular memory cleansing writeback. The difference between
372 * these two operations is that if a dirty page/buffer is encountered, it must
373 * be waited upon, and not just skipped over.
375 int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
376 loff_t end, int sync_mode)
379 struct writeback_control wbc = {
380 .sync_mode = sync_mode,
381 .nr_to_write = LONG_MAX,
382 .range_start = start,
386 if (!mapping_cap_writeback_dirty(mapping) ||
387 !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
390 wbc_attach_fdatawrite_inode(&wbc, mapping->host);
391 ret = do_writepages(mapping, &wbc);
392 wbc_detach_inode(&wbc);
396 static inline int __filemap_fdatawrite(struct address_space *mapping,
399 return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode);
402 int filemap_fdatawrite(struct address_space *mapping)
404 return __filemap_fdatawrite(mapping, WB_SYNC_ALL);
406 EXPORT_SYMBOL(filemap_fdatawrite);
408 int filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
411 return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL);
413 EXPORT_SYMBOL(filemap_fdatawrite_range);
416 * filemap_flush - mostly a non-blocking flush
417 * @mapping: target address_space
419 * This is a mostly non-blocking flush. Not suitable for data-integrity
420 * purposes - I/O may not be started against all dirty pages.
422 int filemap_flush(struct address_space *mapping)
424 return __filemap_fdatawrite(mapping, WB_SYNC_NONE);
426 EXPORT_SYMBOL(filemap_flush);
428 static int __filemap_fdatawait_range(struct address_space *mapping,
429 loff_t start_byte, loff_t end_byte)
431 pgoff_t index = start_byte >> PAGE_SHIFT;
432 pgoff_t end = end_byte >> PAGE_SHIFT;
437 if (end_byte < start_byte)
440 pagevec_init(&pvec, 0);
441 while ((index <= end) &&
442 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
443 PAGECACHE_TAG_WRITEBACK,
444 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1)) != 0) {
447 for (i = 0; i < nr_pages; i++) {
448 struct page *page = pvec.pages[i];
450 /* until radix tree lookup accepts end_index */
451 if (page->index > end)
454 wait_on_page_writeback(page);
455 if (TestClearPageError(page))
458 pagevec_release(&pvec);
466 * filemap_fdatawait_range - wait for writeback to complete
467 * @mapping: address space structure to wait for
468 * @start_byte: offset in bytes where the range starts
469 * @end_byte: offset in bytes where the range ends (inclusive)
471 * Walk the list of under-writeback pages of the given address space
472 * in the given range and wait for all of them. Check error status of
473 * the address space and return it.
475 * Since the error status of the address space is cleared by this function,
476 * callers are responsible for checking the return value and handling and/or
477 * reporting the error.
479 int filemap_fdatawait_range(struct address_space *mapping, loff_t start_byte,
484 ret = __filemap_fdatawait_range(mapping, start_byte, end_byte);
485 ret2 = filemap_check_errors(mapping);
491 EXPORT_SYMBOL(filemap_fdatawait_range);
494 * filemap_fdatawait_keep_errors - wait for writeback without clearing errors
495 * @mapping: address space structure to wait for
497 * Walk the list of under-writeback pages of the given address space
498 * and wait for all of them. Unlike filemap_fdatawait(), this function
499 * does not clear error status of the address space.
501 * Use this function if callers don't handle errors themselves. Expected
502 * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2),
505 void filemap_fdatawait_keep_errors(struct address_space *mapping)
507 loff_t i_size = i_size_read(mapping->host);
512 __filemap_fdatawait_range(mapping, 0, i_size - 1);
516 * filemap_fdatawait - wait for all under-writeback pages to complete
517 * @mapping: address space structure to wait for
519 * Walk the list of under-writeback pages of the given address space
520 * and wait for all of them. Check error status of the address space
523 * Since the error status of the address space is cleared by this function,
524 * callers are responsible for checking the return value and handling and/or
525 * reporting the error.
527 int filemap_fdatawait(struct address_space *mapping)
529 loff_t i_size = i_size_read(mapping->host);
534 return filemap_fdatawait_range(mapping, 0, i_size - 1);
536 EXPORT_SYMBOL(filemap_fdatawait);
538 int filemap_write_and_wait(struct address_space *mapping)
542 if ((!dax_mapping(mapping) && mapping->nrpages) ||
543 (dax_mapping(mapping) && mapping->nrexceptional)) {
544 err = filemap_fdatawrite(mapping);
546 * Even if the above returned error, the pages may be
547 * written partially (e.g. -ENOSPC), so we wait for it.
548 * But the -EIO is special case, it may indicate the worst
549 * thing (e.g. bug) happened, so we avoid waiting for it.
552 int err2 = filemap_fdatawait(mapping);
557 err = filemap_check_errors(mapping);
561 EXPORT_SYMBOL(filemap_write_and_wait);
564 * filemap_write_and_wait_range - write out & wait on a file range
565 * @mapping: the address_space for the pages
566 * @lstart: offset in bytes where the range starts
567 * @lend: offset in bytes where the range ends (inclusive)
569 * Write out and wait upon file offsets lstart->lend, inclusive.
571 * Note that `lend' is inclusive (describes the last byte to be written) so
572 * that this function can be used to write to the very end-of-file (end = -1).
574 int filemap_write_and_wait_range(struct address_space *mapping,
575 loff_t lstart, loff_t lend)
579 if ((!dax_mapping(mapping) && mapping->nrpages) ||
580 (dax_mapping(mapping) && mapping->nrexceptional)) {
581 err = __filemap_fdatawrite_range(mapping, lstart, lend,
583 /* See comment of filemap_write_and_wait() */
585 int err2 = filemap_fdatawait_range(mapping,
591 err = filemap_check_errors(mapping);
595 EXPORT_SYMBOL(filemap_write_and_wait_range);
598 * replace_page_cache_page - replace a pagecache page with a new one
599 * @old: page to be replaced
600 * @new: page to replace with
601 * @gfp_mask: allocation mode
603 * This function replaces a page in the pagecache with a new one. On
604 * success it acquires the pagecache reference for the new page and
605 * drops it for the old page. Both the old and new pages must be
606 * locked. This function does not add the new page to the LRU, the
607 * caller must do that.
609 * The remove + add is atomic. The only way this function can fail is
610 * memory allocation failure.
612 int replace_page_cache_page(struct page *old, struct page *new, gfp_t gfp_mask)
616 VM_BUG_ON_PAGE(!PageLocked(old), old);
617 VM_BUG_ON_PAGE(!PageLocked(new), new);
618 VM_BUG_ON_PAGE(new->mapping, new);
620 error = radix_tree_preload(gfp_mask & GFP_RECLAIM_MASK);
622 struct address_space *mapping = old->mapping;
623 void (*freepage)(struct page *);
626 pgoff_t offset = old->index;
627 freepage = mapping->a_ops->freepage;
630 new->mapping = mapping;
633 spin_lock_irqsave(&mapping->tree_lock, flags);
634 __delete_from_page_cache(old, NULL);
635 error = page_cache_tree_insert(mapping, new, NULL);
639 * hugetlb pages do not participate in page cache accounting.
642 __inc_node_page_state(new, NR_FILE_PAGES);
643 if (PageSwapBacked(new))
644 __inc_node_page_state(new, NR_SHMEM);
645 spin_unlock_irqrestore(&mapping->tree_lock, flags);
646 mem_cgroup_migrate(old, new);
647 radix_tree_preload_end();
655 EXPORT_SYMBOL_GPL(replace_page_cache_page);
657 static int __add_to_page_cache_locked(struct page *page,
658 struct address_space *mapping,
659 pgoff_t offset, gfp_t gfp_mask,
662 int huge = PageHuge(page);
663 struct mem_cgroup *memcg;
666 VM_BUG_ON_PAGE(!PageLocked(page), page);
667 VM_BUG_ON_PAGE(PageSwapBacked(page), page);
670 error = mem_cgroup_try_charge(page, current->mm,
671 gfp_mask, &memcg, false);
676 error = radix_tree_maybe_preload(gfp_mask & GFP_RECLAIM_MASK);
679 mem_cgroup_cancel_charge(page, memcg, false);
684 page->mapping = mapping;
685 page->index = offset;
687 spin_lock_irq(&mapping->tree_lock);
688 error = page_cache_tree_insert(mapping, page, shadowp);
689 radix_tree_preload_end();
693 /* hugetlb pages do not participate in page cache accounting. */
695 __inc_node_page_state(page, NR_FILE_PAGES);
696 spin_unlock_irq(&mapping->tree_lock);
698 mem_cgroup_commit_charge(page, memcg, false, false);
699 trace_mm_filemap_add_to_page_cache(page);
702 page->mapping = NULL;
703 /* Leave page->index set: truncation relies upon it */
704 spin_unlock_irq(&mapping->tree_lock);
706 mem_cgroup_cancel_charge(page, memcg, false);
712 * add_to_page_cache_locked - add a locked page to the pagecache
714 * @mapping: the page's address_space
715 * @offset: page index
716 * @gfp_mask: page allocation mode
718 * This function is used to add a page to the pagecache. It must be locked.
719 * This function does not add the page to the LRU. The caller must do that.
721 int add_to_page_cache_locked(struct page *page, struct address_space *mapping,
722 pgoff_t offset, gfp_t gfp_mask)
724 return __add_to_page_cache_locked(page, mapping, offset,
727 EXPORT_SYMBOL(add_to_page_cache_locked);
729 int add_to_page_cache_lru(struct page *page, struct address_space *mapping,
730 pgoff_t offset, gfp_t gfp_mask)
735 __SetPageLocked(page);
736 ret = __add_to_page_cache_locked(page, mapping, offset,
739 __ClearPageLocked(page);
742 * The page might have been evicted from cache only
743 * recently, in which case it should be activated like
744 * any other repeatedly accessed page.
745 * The exception is pages getting rewritten; evicting other
746 * data from the working set, only to cache data that will
747 * get overwritten with something else, is a waste of memory.
749 if (!(gfp_mask & __GFP_WRITE) &&
750 shadow && workingset_refault(shadow)) {
752 workingset_activation(page);
754 ClearPageActive(page);
759 EXPORT_SYMBOL_GPL(add_to_page_cache_lru);
762 struct page *__page_cache_alloc(gfp_t gfp)
767 if (cpuset_do_page_mem_spread()) {
768 unsigned int cpuset_mems_cookie;
770 cpuset_mems_cookie = read_mems_allowed_begin();
771 n = cpuset_mem_spread_node();
772 page = __alloc_pages_node(n, gfp, 0);
773 } while (!page && read_mems_allowed_retry(cpuset_mems_cookie));
777 return alloc_pages(gfp, 0);
779 EXPORT_SYMBOL(__page_cache_alloc);
783 * In order to wait for pages to become available there must be
784 * waitqueues associated with pages. By using a hash table of
785 * waitqueues where the bucket discipline is to maintain all
786 * waiters on the same queue and wake all when any of the pages
787 * become available, and for the woken contexts to check to be
788 * sure the appropriate page became available, this saves space
789 * at a cost of "thundering herd" phenomena during rare hash
792 wait_queue_head_t *page_waitqueue(struct page *page)
794 return bit_waitqueue(page, 0);
796 EXPORT_SYMBOL(page_waitqueue);
798 void wait_on_page_bit(struct page *page, int bit_nr)
800 DEFINE_WAIT_BIT(wait, &page->flags, bit_nr);
802 if (test_bit(bit_nr, &page->flags))
803 __wait_on_bit(page_waitqueue(page), &wait, bit_wait_io,
804 TASK_UNINTERRUPTIBLE);
806 EXPORT_SYMBOL(wait_on_page_bit);
808 int wait_on_page_bit_killable(struct page *page, int bit_nr)
810 DEFINE_WAIT_BIT(wait, &page->flags, bit_nr);
812 if (!test_bit(bit_nr, &page->flags))
815 return __wait_on_bit(page_waitqueue(page), &wait,
816 bit_wait_io, TASK_KILLABLE);
819 int wait_on_page_bit_killable_timeout(struct page *page,
820 int bit_nr, unsigned long timeout)
822 DEFINE_WAIT_BIT(wait, &page->flags, bit_nr);
824 wait.key.timeout = jiffies + timeout;
825 if (!test_bit(bit_nr, &page->flags))
827 return __wait_on_bit(page_waitqueue(page), &wait,
828 bit_wait_io_timeout, TASK_KILLABLE);
830 EXPORT_SYMBOL_GPL(wait_on_page_bit_killable_timeout);
833 * add_page_wait_queue - Add an arbitrary waiter to a page's wait queue
834 * @page: Page defining the wait queue of interest
835 * @waiter: Waiter to add to the queue
837 * Add an arbitrary @waiter to the wait queue for the nominated @page.
839 void add_page_wait_queue(struct page *page, wait_queue_t *waiter)
841 wait_queue_head_t *q = page_waitqueue(page);
844 spin_lock_irqsave(&q->lock, flags);
845 __add_wait_queue(q, waiter);
846 spin_unlock_irqrestore(&q->lock, flags);
848 EXPORT_SYMBOL_GPL(add_page_wait_queue);
851 * unlock_page - unlock a locked page
854 * Unlocks the page and wakes up sleepers in ___wait_on_page_locked().
855 * Also wakes sleepers in wait_on_page_writeback() because the wakeup
856 * mechanism between PageLocked pages and PageWriteback pages is shared.
857 * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep.
859 * The mb is necessary to enforce ordering between the clear_bit and the read
860 * of the waitqueue (to avoid SMP races with a parallel wait_on_page_locked()).
862 void unlock_page(struct page *page)
864 page = compound_head(page);
865 VM_BUG_ON_PAGE(!PageLocked(page), page);
866 clear_bit_unlock(PG_locked, &page->flags);
867 smp_mb__after_atomic();
868 wake_up_page(page, PG_locked);
870 EXPORT_SYMBOL(unlock_page);
873 * end_page_writeback - end writeback against a page
876 void end_page_writeback(struct page *page)
879 * TestClearPageReclaim could be used here but it is an atomic
880 * operation and overkill in this particular case. Failing to
881 * shuffle a page marked for immediate reclaim is too mild to
882 * justify taking an atomic operation penalty at the end of
883 * ever page writeback.
885 if (PageReclaim(page)) {
886 ClearPageReclaim(page);
887 rotate_reclaimable_page(page);
890 if (!test_clear_page_writeback(page))
893 smp_mb__after_atomic();
894 wake_up_page(page, PG_writeback);
896 EXPORT_SYMBOL(end_page_writeback);
899 * After completing I/O on a page, call this routine to update the page
900 * flags appropriately
902 void page_endio(struct page *page, bool is_write, int err)
906 SetPageUptodate(page);
908 ClearPageUptodate(page);
914 struct address_space *mapping;
917 mapping = page_mapping(page);
919 mapping_set_error(mapping, err);
921 end_page_writeback(page);
924 EXPORT_SYMBOL_GPL(page_endio);
927 * __lock_page - get a lock on the page, assuming we need to sleep to get it
928 * @page: the page to lock
930 void __lock_page(struct page *page)
932 struct page *page_head = compound_head(page);
933 DEFINE_WAIT_BIT(wait, &page_head->flags, PG_locked);
935 __wait_on_bit_lock(page_waitqueue(page_head), &wait, bit_wait_io,
936 TASK_UNINTERRUPTIBLE);
938 EXPORT_SYMBOL(__lock_page);
940 int __lock_page_killable(struct page *page)
942 struct page *page_head = compound_head(page);
943 DEFINE_WAIT_BIT(wait, &page_head->flags, PG_locked);
945 return __wait_on_bit_lock(page_waitqueue(page_head), &wait,
946 bit_wait_io, TASK_KILLABLE);
948 EXPORT_SYMBOL_GPL(__lock_page_killable);
952 * 1 - page is locked; mmap_sem is still held.
953 * 0 - page is not locked.
954 * mmap_sem has been released (up_read()), unless flags had both
955 * FAULT_FLAG_ALLOW_RETRY and FAULT_FLAG_RETRY_NOWAIT set, in
956 * which case mmap_sem is still held.
958 * If neither ALLOW_RETRY nor KILLABLE are set, will always return 1
959 * with the page locked and the mmap_sem unperturbed.
961 int __lock_page_or_retry(struct page *page, struct mm_struct *mm,
964 if (flags & FAULT_FLAG_ALLOW_RETRY) {
966 * CAUTION! In this case, mmap_sem is not released
967 * even though return 0.
969 if (flags & FAULT_FLAG_RETRY_NOWAIT)
972 up_read(&mm->mmap_sem);
973 if (flags & FAULT_FLAG_KILLABLE)
974 wait_on_page_locked_killable(page);
976 wait_on_page_locked(page);
979 if (flags & FAULT_FLAG_KILLABLE) {
982 ret = __lock_page_killable(page);
984 up_read(&mm->mmap_sem);
994 * page_cache_next_hole - find the next hole (not-present entry)
997 * @max_scan: maximum range to search
999 * Search the set [index, min(index+max_scan-1, MAX_INDEX)] for the
1000 * lowest indexed hole.
1002 * Returns: the index of the hole if found, otherwise returns an index
1003 * outside of the set specified (in which case 'return - index >=
1004 * max_scan' will be true). In rare cases of index wrap-around, 0 will
1007 * page_cache_next_hole may be called under rcu_read_lock. However,
1008 * like radix_tree_gang_lookup, this will not atomically search a
1009 * snapshot of the tree at a single point in time. For example, if a
1010 * hole is created at index 5, then subsequently a hole is created at
1011 * index 10, page_cache_next_hole covering both indexes may return 10
1012 * if called under rcu_read_lock.
1014 pgoff_t page_cache_next_hole(struct address_space *mapping,
1015 pgoff_t index, unsigned long max_scan)
1019 for (i = 0; i < max_scan; i++) {
1022 page = radix_tree_lookup(&mapping->page_tree, index);
1023 if (!page || radix_tree_exceptional_entry(page))
1032 EXPORT_SYMBOL(page_cache_next_hole);
1035 * page_cache_prev_hole - find the prev hole (not-present entry)
1038 * @max_scan: maximum range to search
1040 * Search backwards in the range [max(index-max_scan+1, 0), index] for
1043 * Returns: the index of the hole if found, otherwise returns an index
1044 * outside of the set specified (in which case 'index - return >=
1045 * max_scan' will be true). In rare cases of wrap-around, ULONG_MAX
1048 * page_cache_prev_hole may be called under rcu_read_lock. However,
1049 * like radix_tree_gang_lookup, this will not atomically search a
1050 * snapshot of the tree at a single point in time. For example, if a
1051 * hole is created at index 10, then subsequently a hole is created at
1052 * index 5, page_cache_prev_hole covering both indexes may return 5 if
1053 * called under rcu_read_lock.
1055 pgoff_t page_cache_prev_hole(struct address_space *mapping,
1056 pgoff_t index, unsigned long max_scan)
1060 for (i = 0; i < max_scan; i++) {
1063 page = radix_tree_lookup(&mapping->page_tree, index);
1064 if (!page || radix_tree_exceptional_entry(page))
1067 if (index == ULONG_MAX)
1073 EXPORT_SYMBOL(page_cache_prev_hole);
1076 * find_get_entry - find and get a page cache entry
1077 * @mapping: the address_space to search
1078 * @offset: the page cache index
1080 * Looks up the page cache slot at @mapping & @offset. If there is a
1081 * page cache page, it is returned with an increased refcount.
1083 * If the slot holds a shadow entry of a previously evicted page, or a
1084 * swap entry from shmem/tmpfs, it is returned.
1086 * Otherwise, %NULL is returned.
1088 struct page *find_get_entry(struct address_space *mapping, pgoff_t offset)
1091 struct page *head, *page;
1096 pagep = radix_tree_lookup_slot(&mapping->page_tree, offset);
1098 page = radix_tree_deref_slot(pagep);
1099 if (unlikely(!page))
1101 if (radix_tree_exception(page)) {
1102 if (radix_tree_deref_retry(page))
1105 * A shadow entry of a recently evicted page,
1106 * or a swap entry from shmem/tmpfs. Return
1107 * it without attempting to raise page count.
1112 head = compound_head(page);
1113 if (!page_cache_get_speculative(head))
1116 /* The page was split under us? */
1117 if (compound_head(page) != head) {
1123 * Has the page moved?
1124 * This is part of the lockless pagecache protocol. See
1125 * include/linux/pagemap.h for details.
1127 if (unlikely(page != *pagep)) {
1137 EXPORT_SYMBOL(find_get_entry);
1140 * find_lock_entry - locate, pin and lock a page cache entry
1141 * @mapping: the address_space to search
1142 * @offset: the page cache index
1144 * Looks up the page cache slot at @mapping & @offset. If there is a
1145 * page cache page, it is returned locked and with an increased
1148 * If the slot holds a shadow entry of a previously evicted page, or a
1149 * swap entry from shmem/tmpfs, it is returned.
1151 * Otherwise, %NULL is returned.
1153 * find_lock_entry() may sleep.
1155 struct page *find_lock_entry(struct address_space *mapping, pgoff_t offset)
1160 page = find_get_entry(mapping, offset);
1161 if (page && !radix_tree_exception(page)) {
1163 /* Has the page been truncated? */
1164 if (unlikely(page_mapping(page) != mapping)) {
1169 VM_BUG_ON_PAGE(page_to_pgoff(page) != offset, page);
1173 EXPORT_SYMBOL(find_lock_entry);
1176 * pagecache_get_page - find and get a page reference
1177 * @mapping: the address_space to search
1178 * @offset: the page index
1179 * @fgp_flags: PCG flags
1180 * @gfp_mask: gfp mask to use for the page cache data page allocation
1182 * Looks up the page cache slot at @mapping & @offset.
1184 * PCG flags modify how the page is returned.
1186 * FGP_ACCESSED: the page will be marked accessed
1187 * FGP_LOCK: Page is return locked
1188 * FGP_CREAT: If page is not present then a new page is allocated using
1189 * @gfp_mask and added to the page cache and the VM's LRU
1190 * list. The page is returned locked and with an increased
1191 * refcount. Otherwise, %NULL is returned.
1193 * If FGP_LOCK or FGP_CREAT are specified then the function may sleep even
1194 * if the GFP flags specified for FGP_CREAT are atomic.
1196 * If there is a page cache page, it is returned with an increased refcount.
1198 struct page *pagecache_get_page(struct address_space *mapping, pgoff_t offset,
1199 int fgp_flags, gfp_t gfp_mask)
1204 page = find_get_entry(mapping, offset);
1205 if (radix_tree_exceptional_entry(page))
1210 if (fgp_flags & FGP_LOCK) {
1211 if (fgp_flags & FGP_NOWAIT) {
1212 if (!trylock_page(page)) {
1220 /* Has the page been truncated? */
1221 if (unlikely(page->mapping != mapping)) {
1226 VM_BUG_ON_PAGE(page->index != offset, page);
1229 if (page && (fgp_flags & FGP_ACCESSED))
1230 mark_page_accessed(page);
1233 if (!page && (fgp_flags & FGP_CREAT)) {
1235 if ((fgp_flags & FGP_WRITE) && mapping_cap_account_dirty(mapping))
1236 gfp_mask |= __GFP_WRITE;
1237 if (fgp_flags & FGP_NOFS)
1238 gfp_mask &= ~__GFP_FS;
1240 page = __page_cache_alloc(gfp_mask);
1244 if (WARN_ON_ONCE(!(fgp_flags & FGP_LOCK)))
1245 fgp_flags |= FGP_LOCK;
1247 /* Init accessed so avoid atomic mark_page_accessed later */
1248 if (fgp_flags & FGP_ACCESSED)
1249 __SetPageReferenced(page);
1251 err = add_to_page_cache_lru(page, mapping, offset, gfp_mask);
1252 if (unlikely(err)) {
1262 EXPORT_SYMBOL(pagecache_get_page);
1265 * find_get_entries - gang pagecache lookup
1266 * @mapping: The address_space to search
1267 * @start: The starting page cache index
1268 * @nr_entries: The maximum number of entries
1269 * @entries: Where the resulting entries are placed
1270 * @indices: The cache indices corresponding to the entries in @entries
1272 * find_get_entries() will search for and return a group of up to
1273 * @nr_entries entries in the mapping. The entries are placed at
1274 * @entries. find_get_entries() takes a reference against any actual
1277 * The search returns a group of mapping-contiguous page cache entries
1278 * with ascending indexes. There may be holes in the indices due to
1279 * not-present pages.
1281 * Any shadow entries of evicted pages, or swap entries from
1282 * shmem/tmpfs, are included in the returned array.
1284 * find_get_entries() returns the number of pages and shadow entries
1287 unsigned find_get_entries(struct address_space *mapping,
1288 pgoff_t start, unsigned int nr_entries,
1289 struct page **entries, pgoff_t *indices)
1292 unsigned int ret = 0;
1293 struct radix_tree_iter iter;
1299 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) {
1300 struct page *head, *page;
1302 page = radix_tree_deref_slot(slot);
1303 if (unlikely(!page))
1305 if (radix_tree_exception(page)) {
1306 if (radix_tree_deref_retry(page)) {
1307 slot = radix_tree_iter_retry(&iter);
1311 * A shadow entry of a recently evicted page, a swap
1312 * entry from shmem/tmpfs or a DAX entry. Return it
1313 * without attempting to raise page count.
1318 head = compound_head(page);
1319 if (!page_cache_get_speculative(head))
1322 /* The page was split under us? */
1323 if (compound_head(page) != head) {
1328 /* Has the page moved? */
1329 if (unlikely(page != *slot)) {
1334 indices[ret] = iter.index;
1335 entries[ret] = page;
1336 if (++ret == nr_entries)
1344 * find_get_pages - gang pagecache lookup
1345 * @mapping: The address_space to search
1346 * @start: The starting page index
1347 * @nr_pages: The maximum number of pages
1348 * @pages: Where the resulting pages are placed
1350 * find_get_pages() will search for and return a group of up to
1351 * @nr_pages pages in the mapping. The pages are placed at @pages.
1352 * find_get_pages() takes a reference against the returned pages.
1354 * The search returns a group of mapping-contiguous pages with ascending
1355 * indexes. There may be holes in the indices due to not-present pages.
1357 * find_get_pages() returns the number of pages which were found.
1359 unsigned find_get_pages(struct address_space *mapping, pgoff_t start,
1360 unsigned int nr_pages, struct page **pages)
1362 struct radix_tree_iter iter;
1366 if (unlikely(!nr_pages))
1370 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) {
1371 struct page *head, *page;
1373 page = radix_tree_deref_slot(slot);
1374 if (unlikely(!page))
1377 if (radix_tree_exception(page)) {
1378 if (radix_tree_deref_retry(page)) {
1379 slot = radix_tree_iter_retry(&iter);
1383 * A shadow entry of a recently evicted page,
1384 * or a swap entry from shmem/tmpfs. Skip
1390 head = compound_head(page);
1391 if (!page_cache_get_speculative(head))
1394 /* The page was split under us? */
1395 if (compound_head(page) != head) {
1400 /* Has the page moved? */
1401 if (unlikely(page != *slot)) {
1407 if (++ret == nr_pages)
1416 * find_get_pages_contig - gang contiguous pagecache lookup
1417 * @mapping: The address_space to search
1418 * @index: The starting page index
1419 * @nr_pages: The maximum number of pages
1420 * @pages: Where the resulting pages are placed
1422 * find_get_pages_contig() works exactly like find_get_pages(), except
1423 * that the returned number of pages are guaranteed to be contiguous.
1425 * find_get_pages_contig() returns the number of pages which were found.
1427 unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index,
1428 unsigned int nr_pages, struct page **pages)
1430 struct radix_tree_iter iter;
1432 unsigned int ret = 0;
1434 if (unlikely(!nr_pages))
1438 radix_tree_for_each_contig(slot, &mapping->page_tree, &iter, index) {
1439 struct page *head, *page;
1441 page = radix_tree_deref_slot(slot);
1442 /* The hole, there no reason to continue */
1443 if (unlikely(!page))
1446 if (radix_tree_exception(page)) {
1447 if (radix_tree_deref_retry(page)) {
1448 slot = radix_tree_iter_retry(&iter);
1452 * A shadow entry of a recently evicted page,
1453 * or a swap entry from shmem/tmpfs. Stop
1454 * looking for contiguous pages.
1459 head = compound_head(page);
1460 if (!page_cache_get_speculative(head))
1463 /* The page was split under us? */
1464 if (compound_head(page) != head) {
1469 /* Has the page moved? */
1470 if (unlikely(page != *slot)) {
1476 * must check mapping and index after taking the ref.
1477 * otherwise we can get both false positives and false
1478 * negatives, which is just confusing to the caller.
1480 if (page->mapping == NULL || page_to_pgoff(page) != iter.index) {
1486 if (++ret == nr_pages)
1492 EXPORT_SYMBOL(find_get_pages_contig);
1495 * find_get_pages_tag - find and return pages that match @tag
1496 * @mapping: the address_space to search
1497 * @index: the starting page index
1498 * @tag: the tag index
1499 * @nr_pages: the maximum number of pages
1500 * @pages: where the resulting pages are placed
1502 * Like find_get_pages, except we only return pages which are tagged with
1503 * @tag. We update @index to index the next page for the traversal.
1505 unsigned find_get_pages_tag(struct address_space *mapping, pgoff_t *index,
1506 int tag, unsigned int nr_pages, struct page **pages)
1508 struct radix_tree_iter iter;
1512 if (unlikely(!nr_pages))
1516 radix_tree_for_each_tagged(slot, &mapping->page_tree,
1517 &iter, *index, tag) {
1518 struct page *head, *page;
1520 page = radix_tree_deref_slot(slot);
1521 if (unlikely(!page))
1524 if (radix_tree_exception(page)) {
1525 if (radix_tree_deref_retry(page)) {
1526 slot = radix_tree_iter_retry(&iter);
1530 * A shadow entry of a recently evicted page.
1532 * Those entries should never be tagged, but
1533 * this tree walk is lockless and the tags are
1534 * looked up in bulk, one radix tree node at a
1535 * time, so there is a sizable window for page
1536 * reclaim to evict a page we saw tagged.
1543 head = compound_head(page);
1544 if (!page_cache_get_speculative(head))
1547 /* The page was split under us? */
1548 if (compound_head(page) != head) {
1553 /* Has the page moved? */
1554 if (unlikely(page != *slot)) {
1560 if (++ret == nr_pages)
1567 *index = pages[ret - 1]->index + 1;
1571 EXPORT_SYMBOL(find_get_pages_tag);
1574 * find_get_entries_tag - find and return entries that match @tag
1575 * @mapping: the address_space to search
1576 * @start: the starting page cache index
1577 * @tag: the tag index
1578 * @nr_entries: the maximum number of entries
1579 * @entries: where the resulting entries are placed
1580 * @indices: the cache indices corresponding to the entries in @entries
1582 * Like find_get_entries, except we only return entries which are tagged with
1585 unsigned find_get_entries_tag(struct address_space *mapping, pgoff_t start,
1586 int tag, unsigned int nr_entries,
1587 struct page **entries, pgoff_t *indices)
1590 unsigned int ret = 0;
1591 struct radix_tree_iter iter;
1597 radix_tree_for_each_tagged(slot, &mapping->page_tree,
1598 &iter, start, tag) {
1599 struct page *head, *page;
1601 page = radix_tree_deref_slot(slot);
1602 if (unlikely(!page))
1604 if (radix_tree_exception(page)) {
1605 if (radix_tree_deref_retry(page)) {
1606 slot = radix_tree_iter_retry(&iter);
1611 * A shadow entry of a recently evicted page, a swap
1612 * entry from shmem/tmpfs or a DAX entry. Return it
1613 * without attempting to raise page count.
1618 head = compound_head(page);
1619 if (!page_cache_get_speculative(head))
1622 /* The page was split under us? */
1623 if (compound_head(page) != head) {
1628 /* Has the page moved? */
1629 if (unlikely(page != *slot)) {
1634 indices[ret] = iter.index;
1635 entries[ret] = page;
1636 if (++ret == nr_entries)
1642 EXPORT_SYMBOL(find_get_entries_tag);
1645 * CD/DVDs are error prone. When a medium error occurs, the driver may fail
1646 * a _large_ part of the i/o request. Imagine the worst scenario:
1648 * ---R__________________________________________B__________
1649 * ^ reading here ^ bad block(assume 4k)
1651 * read(R) => miss => readahead(R...B) => media error => frustrating retries
1652 * => failing the whole request => read(R) => read(R+1) =>
1653 * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
1654 * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
1655 * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
1657 * It is going insane. Fix it by quickly scaling down the readahead size.
1659 static void shrink_readahead_size_eio(struct file *filp,
1660 struct file_ra_state *ra)
1666 * do_generic_file_read - generic file read routine
1667 * @filp: the file to read
1668 * @ppos: current file position
1669 * @iter: data destination
1670 * @written: already copied
1672 * This is a generic file read routine, and uses the
1673 * mapping->a_ops->readpage() function for the actual low-level stuff.
1675 * This is really ugly. But the goto's actually try to clarify some
1676 * of the logic when it comes to error handling etc.
1678 static ssize_t do_generic_file_read(struct file *filp, loff_t *ppos,
1679 struct iov_iter *iter, ssize_t written)
1681 struct address_space *mapping = filp->f_mapping;
1682 struct inode *inode = mapping->host;
1683 struct file_ra_state *ra = &filp->f_ra;
1687 unsigned long offset; /* offset into pagecache page */
1688 unsigned int prev_offset;
1691 if (unlikely(*ppos >= inode->i_sb->s_maxbytes))
1693 iov_iter_truncate(iter, inode->i_sb->s_maxbytes);
1695 index = *ppos >> PAGE_SHIFT;
1696 prev_index = ra->prev_pos >> PAGE_SHIFT;
1697 prev_offset = ra->prev_pos & (PAGE_SIZE-1);
1698 last_index = (*ppos + iter->count + PAGE_SIZE-1) >> PAGE_SHIFT;
1699 offset = *ppos & ~PAGE_MASK;
1705 unsigned long nr, ret;
1709 if (fatal_signal_pending(current)) {
1714 page = find_get_page(mapping, index);
1716 page_cache_sync_readahead(mapping,
1718 index, last_index - index);
1719 page = find_get_page(mapping, index);
1720 if (unlikely(page == NULL))
1721 goto no_cached_page;
1723 if (PageReadahead(page)) {
1724 page_cache_async_readahead(mapping,
1726 index, last_index - index);
1728 if (!PageUptodate(page)) {
1730 * See comment in do_read_cache_page on why
1731 * wait_on_page_locked is used to avoid unnecessarily
1732 * serialisations and why it's safe.
1734 error = wait_on_page_locked_killable(page);
1735 if (unlikely(error))
1736 goto readpage_error;
1737 if (PageUptodate(page))
1740 if (inode->i_blkbits == PAGE_SHIFT ||
1741 !mapping->a_ops->is_partially_uptodate)
1742 goto page_not_up_to_date;
1743 /* pipes can't handle partially uptodate pages */
1744 if (unlikely(iter->type & ITER_PIPE))
1745 goto page_not_up_to_date;
1746 if (!trylock_page(page))
1747 goto page_not_up_to_date;
1748 /* Did it get truncated before we got the lock? */
1750 goto page_not_up_to_date_locked;
1751 if (!mapping->a_ops->is_partially_uptodate(page,
1752 offset, iter->count))
1753 goto page_not_up_to_date_locked;
1758 * i_size must be checked after we know the page is Uptodate.
1760 * Checking i_size after the check allows us to calculate
1761 * the correct value for "nr", which means the zero-filled
1762 * part of the page is not copied back to userspace (unless
1763 * another truncate extends the file - this is desired though).
1766 isize = i_size_read(inode);
1767 end_index = (isize - 1) >> PAGE_SHIFT;
1768 if (unlikely(!isize || index > end_index)) {
1773 /* nr is the maximum number of bytes to copy from this page */
1775 if (index == end_index) {
1776 nr = ((isize - 1) & ~PAGE_MASK) + 1;
1784 /* If users can be writing to this page using arbitrary
1785 * virtual addresses, take care about potential aliasing
1786 * before reading the page on the kernel side.
1788 if (mapping_writably_mapped(mapping))
1789 flush_dcache_page(page);
1792 * When a sequential read accesses a page several times,
1793 * only mark it as accessed the first time.
1795 if (prev_index != index || offset != prev_offset)
1796 mark_page_accessed(page);
1800 * Ok, we have the page, and it's up-to-date, so
1801 * now we can copy it to user space...
1804 ret = copy_page_to_iter(page, offset, nr, iter);
1806 index += offset >> PAGE_SHIFT;
1807 offset &= ~PAGE_MASK;
1808 prev_offset = offset;
1812 if (!iov_iter_count(iter))
1820 page_not_up_to_date:
1821 /* Get exclusive access to the page ... */
1822 error = lock_page_killable(page);
1823 if (unlikely(error))
1824 goto readpage_error;
1826 page_not_up_to_date_locked:
1827 /* Did it get truncated before we got the lock? */
1828 if (!page->mapping) {
1834 /* Did somebody else fill it already? */
1835 if (PageUptodate(page)) {
1842 * A previous I/O error may have been due to temporary
1843 * failures, eg. multipath errors.
1844 * PG_error will be set again if readpage fails.
1846 ClearPageError(page);
1847 /* Start the actual read. The read will unlock the page. */
1848 error = mapping->a_ops->readpage(filp, page);
1850 if (unlikely(error)) {
1851 if (error == AOP_TRUNCATED_PAGE) {
1856 goto readpage_error;
1859 if (!PageUptodate(page)) {
1860 error = lock_page_killable(page);
1861 if (unlikely(error))
1862 goto readpage_error;
1863 if (!PageUptodate(page)) {
1864 if (page->mapping == NULL) {
1866 * invalidate_mapping_pages got it
1873 shrink_readahead_size_eio(filp, ra);
1875 goto readpage_error;
1883 /* UHHUH! A synchronous read error occurred. Report it */
1889 * Ok, it wasn't cached, so we need to create a new
1892 page = page_cache_alloc_cold(mapping);
1897 error = add_to_page_cache_lru(page, mapping, index,
1898 mapping_gfp_constraint(mapping, GFP_KERNEL));
1901 if (error == -EEXIST) {
1911 ra->prev_pos = prev_index;
1912 ra->prev_pos <<= PAGE_SHIFT;
1913 ra->prev_pos |= prev_offset;
1915 *ppos = ((loff_t)index << PAGE_SHIFT) + offset;
1916 file_accessed(filp);
1917 return written ? written : error;
1921 * generic_file_read_iter - generic filesystem read routine
1922 * @iocb: kernel I/O control block
1923 * @iter: destination for the data read
1925 * This is the "read_iter()" routine for all filesystems
1926 * that can use the page cache directly.
1929 generic_file_read_iter(struct kiocb *iocb, struct iov_iter *iter)
1931 struct file *file = iocb->ki_filp;
1933 size_t count = iov_iter_count(iter);
1936 goto out; /* skip atime */
1938 if (iocb->ki_flags & IOCB_DIRECT) {
1939 struct address_space *mapping = file->f_mapping;
1940 struct inode *inode = mapping->host;
1941 struct iov_iter data = *iter;
1944 size = i_size_read(inode);
1945 retval = filemap_write_and_wait_range(mapping, iocb->ki_pos,
1946 iocb->ki_pos + count - 1);
1950 file_accessed(file);
1952 retval = mapping->a_ops->direct_IO(iocb, &data);
1954 iocb->ki_pos += retval;
1955 iov_iter_advance(iter, retval);
1959 * Btrfs can have a short DIO read if we encounter
1960 * compressed extents, so if there was an error, or if
1961 * we've already read everything we wanted to, or if
1962 * there was a short read because we hit EOF, go ahead
1963 * and return. Otherwise fallthrough to buffered io for
1964 * the rest of the read. Buffered reads will not work for
1965 * DAX files, so don't bother trying.
1967 if (retval < 0 || !iov_iter_count(iter) || iocb->ki_pos >= size ||
1972 retval = do_generic_file_read(file, &iocb->ki_pos, iter, retval);
1976 EXPORT_SYMBOL(generic_file_read_iter);
1980 * page_cache_read - adds requested page to the page cache if not already there
1981 * @file: file to read
1982 * @offset: page index
1983 * @gfp_mask: memory allocation flags
1985 * This adds the requested page to the page cache if it isn't already there,
1986 * and schedules an I/O to read in its contents from disk.
1988 static int page_cache_read(struct file *file, pgoff_t offset, gfp_t gfp_mask)
1990 struct address_space *mapping = file->f_mapping;
1995 page = __page_cache_alloc(gfp_mask|__GFP_COLD);
1999 ret = add_to_page_cache_lru(page, mapping, offset, gfp_mask);
2001 ret = mapping->a_ops->readpage(file, page);
2002 else if (ret == -EEXIST)
2003 ret = 0; /* losing race to add is OK */
2007 } while (ret == AOP_TRUNCATED_PAGE);
2012 #define MMAP_LOTSAMISS (100)
2015 * Synchronous readahead happens when we don't even find
2016 * a page in the page cache at all.
2018 static void do_sync_mmap_readahead(struct vm_area_struct *vma,
2019 struct file_ra_state *ra,
2023 struct address_space *mapping = file->f_mapping;
2025 /* If we don't want any read-ahead, don't bother */
2026 if (vma->vm_flags & VM_RAND_READ)
2031 if (vma->vm_flags & VM_SEQ_READ) {
2032 page_cache_sync_readahead(mapping, ra, file, offset,
2037 /* Avoid banging the cache line if not needed */
2038 if (ra->mmap_miss < MMAP_LOTSAMISS * 10)
2042 * Do we miss much more than hit in this file? If so,
2043 * stop bothering with read-ahead. It will only hurt.
2045 if (ra->mmap_miss > MMAP_LOTSAMISS)
2051 ra->start = max_t(long, 0, offset - ra->ra_pages / 2);
2052 ra->size = ra->ra_pages;
2053 ra->async_size = ra->ra_pages / 4;
2054 ra_submit(ra, mapping, file);
2058 * Asynchronous readahead happens when we find the page and PG_readahead,
2059 * so we want to possibly extend the readahead further..
2061 static void do_async_mmap_readahead(struct vm_area_struct *vma,
2062 struct file_ra_state *ra,
2067 struct address_space *mapping = file->f_mapping;
2069 /* If we don't want any read-ahead, don't bother */
2070 if (vma->vm_flags & VM_RAND_READ)
2072 if (ra->mmap_miss > 0)
2074 if (PageReadahead(page))
2075 page_cache_async_readahead(mapping, ra, file,
2076 page, offset, ra->ra_pages);
2080 * filemap_fault - read in file data for page fault handling
2081 * @vma: vma in which the fault was taken
2082 * @vmf: struct vm_fault containing details of the fault
2084 * filemap_fault() is invoked via the vma operations vector for a
2085 * mapped memory region to read in file data during a page fault.
2087 * The goto's are kind of ugly, but this streamlines the normal case of having
2088 * it in the page cache, and handles the special cases reasonably without
2089 * having a lot of duplicated code.
2091 * vma->vm_mm->mmap_sem must be held on entry.
2093 * If our return value has VM_FAULT_RETRY set, it's because
2094 * lock_page_or_retry() returned 0.
2095 * The mmap_sem has usually been released in this case.
2096 * See __lock_page_or_retry() for the exception.
2098 * If our return value does not have VM_FAULT_RETRY set, the mmap_sem
2099 * has not been released.
2101 * We never return with VM_FAULT_RETRY and a bit from VM_FAULT_ERROR set.
2103 int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
2106 struct file *file = vma->vm_file;
2107 struct address_space *mapping = file->f_mapping;
2108 struct file_ra_state *ra = &file->f_ra;
2109 struct inode *inode = mapping->host;
2110 pgoff_t offset = vmf->pgoff;
2115 size = round_up(i_size_read(inode), PAGE_SIZE);
2116 if (offset >= size >> PAGE_SHIFT)
2117 return VM_FAULT_SIGBUS;
2120 * Do we have something in the page cache already?
2122 page = find_get_page(mapping, offset);
2123 if (likely(page) && !(vmf->flags & FAULT_FLAG_TRIED)) {
2125 * We found the page, so try async readahead before
2126 * waiting for the lock.
2128 do_async_mmap_readahead(vma, ra, file, page, offset);
2130 /* No page in the page cache at all */
2131 do_sync_mmap_readahead(vma, ra, file, offset);
2132 count_vm_event(PGMAJFAULT);
2133 mem_cgroup_count_vm_event(vma->vm_mm, PGMAJFAULT);
2134 ret = VM_FAULT_MAJOR;
2136 page = find_get_page(mapping, offset);
2138 goto no_cached_page;
2141 if (!lock_page_or_retry(page, vma->vm_mm, vmf->flags)) {
2143 return ret | VM_FAULT_RETRY;
2146 /* Did it get truncated? */
2147 if (unlikely(page->mapping != mapping)) {
2152 VM_BUG_ON_PAGE(page->index != offset, page);
2155 * We have a locked page in the page cache, now we need to check
2156 * that it's up-to-date. If not, it is going to be due to an error.
2158 if (unlikely(!PageUptodate(page)))
2159 goto page_not_uptodate;
2162 * Found the page and have a reference on it.
2163 * We must recheck i_size under page lock.
2165 size = round_up(i_size_read(inode), PAGE_SIZE);
2166 if (unlikely(offset >= size >> PAGE_SHIFT)) {
2169 return VM_FAULT_SIGBUS;
2173 return ret | VM_FAULT_LOCKED;
2177 * We're only likely to ever get here if MADV_RANDOM is in
2180 error = page_cache_read(file, offset, vmf->gfp_mask);
2183 * The page we want has now been added to the page cache.
2184 * In the unlikely event that someone removed it in the
2185 * meantime, we'll just come back here and read it again.
2191 * An error return from page_cache_read can result if the
2192 * system is low on memory, or a problem occurs while trying
2195 if (error == -ENOMEM)
2196 return VM_FAULT_OOM;
2197 return VM_FAULT_SIGBUS;
2201 * Umm, take care of errors if the page isn't up-to-date.
2202 * Try to re-read it _once_. We do this synchronously,
2203 * because there really aren't any performance issues here
2204 * and we need to check for errors.
2206 ClearPageError(page);
2207 error = mapping->a_ops->readpage(file, page);
2209 wait_on_page_locked(page);
2210 if (!PageUptodate(page))
2215 if (!error || error == AOP_TRUNCATED_PAGE)
2218 /* Things didn't work out. Return zero to tell the mm layer so. */
2219 shrink_readahead_size_eio(file, ra);
2220 return VM_FAULT_SIGBUS;
2222 EXPORT_SYMBOL(filemap_fault);
2224 void filemap_map_pages(struct fault_env *fe,
2225 pgoff_t start_pgoff, pgoff_t end_pgoff)
2227 struct radix_tree_iter iter;
2229 struct file *file = fe->vma->vm_file;
2230 struct address_space *mapping = file->f_mapping;
2231 pgoff_t last_pgoff = start_pgoff;
2233 struct page *head, *page;
2236 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter,
2238 if (iter.index > end_pgoff)
2241 page = radix_tree_deref_slot(slot);
2242 if (unlikely(!page))
2244 if (radix_tree_exception(page)) {
2245 if (radix_tree_deref_retry(page)) {
2246 slot = radix_tree_iter_retry(&iter);
2252 head = compound_head(page);
2253 if (!page_cache_get_speculative(head))
2256 /* The page was split under us? */
2257 if (compound_head(page) != head) {
2262 /* Has the page moved? */
2263 if (unlikely(page != *slot)) {
2268 if (!PageUptodate(page) ||
2269 PageReadahead(page) ||
2272 if (!trylock_page(page))
2275 if (page->mapping != mapping || !PageUptodate(page))
2278 size = round_up(i_size_read(mapping->host), PAGE_SIZE);
2279 if (page->index >= size >> PAGE_SHIFT)
2282 if (file->f_ra.mmap_miss > 0)
2283 file->f_ra.mmap_miss--;
2285 fe->address += (iter.index - last_pgoff) << PAGE_SHIFT;
2287 fe->pte += iter.index - last_pgoff;
2288 last_pgoff = iter.index;
2289 if (alloc_set_pte(fe, NULL, page))
2298 /* Huge page is mapped? No need to proceed. */
2299 if (pmd_trans_huge(*fe->pmd))
2301 if (iter.index == end_pgoff)
2306 EXPORT_SYMBOL(filemap_map_pages);
2308 int filemap_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
2310 struct page *page = vmf->page;
2311 struct inode *inode = file_inode(vma->vm_file);
2312 int ret = VM_FAULT_LOCKED;
2314 sb_start_pagefault(inode->i_sb);
2315 file_update_time(vma->vm_file);
2317 if (page->mapping != inode->i_mapping) {
2319 ret = VM_FAULT_NOPAGE;
2323 * We mark the page dirty already here so that when freeze is in
2324 * progress, we are guaranteed that writeback during freezing will
2325 * see the dirty page and writeprotect it again.
2327 set_page_dirty(page);
2328 wait_for_stable_page(page);
2330 sb_end_pagefault(inode->i_sb);
2333 EXPORT_SYMBOL(filemap_page_mkwrite);
2335 const struct vm_operations_struct generic_file_vm_ops = {
2336 .fault = filemap_fault,
2337 .map_pages = filemap_map_pages,
2338 .page_mkwrite = filemap_page_mkwrite,
2341 /* This is used for a general mmap of a disk file */
2343 int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
2345 struct address_space *mapping = file->f_mapping;
2347 if (!mapping->a_ops->readpage)
2349 file_accessed(file);
2350 vma->vm_ops = &generic_file_vm_ops;
2355 * This is for filesystems which do not implement ->writepage.
2357 int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
2359 if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE))
2361 return generic_file_mmap(file, vma);
2364 int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
2368 int generic_file_readonly_mmap(struct file * file, struct vm_area_struct * vma)
2372 #endif /* CONFIG_MMU */
2374 EXPORT_SYMBOL(generic_file_mmap);
2375 EXPORT_SYMBOL(generic_file_readonly_mmap);
2377 static struct page *wait_on_page_read(struct page *page)
2379 if (!IS_ERR(page)) {
2380 wait_on_page_locked(page);
2381 if (!PageUptodate(page)) {
2383 page = ERR_PTR(-EIO);
2389 static struct page *do_read_cache_page(struct address_space *mapping,
2391 int (*filler)(void *, struct page *),
2398 page = find_get_page(mapping, index);
2400 page = __page_cache_alloc(gfp | __GFP_COLD);
2402 return ERR_PTR(-ENOMEM);
2403 err = add_to_page_cache_lru(page, mapping, index, gfp);
2404 if (unlikely(err)) {
2408 /* Presumably ENOMEM for radix tree node */
2409 return ERR_PTR(err);
2413 err = filler(data, page);
2416 return ERR_PTR(err);
2419 page = wait_on_page_read(page);
2424 if (PageUptodate(page))
2428 * Page is not up to date and may be locked due one of the following
2429 * case a: Page is being filled and the page lock is held
2430 * case b: Read/write error clearing the page uptodate status
2431 * case c: Truncation in progress (page locked)
2432 * case d: Reclaim in progress
2434 * Case a, the page will be up to date when the page is unlocked.
2435 * There is no need to serialise on the page lock here as the page
2436 * is pinned so the lock gives no additional protection. Even if the
2437 * the page is truncated, the data is still valid if PageUptodate as
2438 * it's a race vs truncate race.
2439 * Case b, the page will not be up to date
2440 * Case c, the page may be truncated but in itself, the data may still
2441 * be valid after IO completes as it's a read vs truncate race. The
2442 * operation must restart if the page is not uptodate on unlock but
2443 * otherwise serialising on page lock to stabilise the mapping gives
2444 * no additional guarantees to the caller as the page lock is
2445 * released before return.
2446 * Case d, similar to truncation. If reclaim holds the page lock, it
2447 * will be a race with remove_mapping that determines if the mapping
2448 * is valid on unlock but otherwise the data is valid and there is
2449 * no need to serialise with page lock.
2451 * As the page lock gives no additional guarantee, we optimistically
2452 * wait on the page to be unlocked and check if it's up to date and
2453 * use the page if it is. Otherwise, the page lock is required to
2454 * distinguish between the different cases. The motivation is that we
2455 * avoid spurious serialisations and wakeups when multiple processes
2456 * wait on the same page for IO to complete.
2458 wait_on_page_locked(page);
2459 if (PageUptodate(page))
2462 /* Distinguish between all the cases under the safety of the lock */
2465 /* Case c or d, restart the operation */
2466 if (!page->mapping) {
2472 /* Someone else locked and filled the page in a very small window */
2473 if (PageUptodate(page)) {
2479 * A previous I/O error may have been due to temporary
2481 * Clear page error before actual read, PG_error will be
2482 * set again if read page fails.
2484 ClearPageError(page);
2488 mark_page_accessed(page);
2493 * read_cache_page - read into page cache, fill it if needed
2494 * @mapping: the page's address_space
2495 * @index: the page index
2496 * @filler: function to perform the read
2497 * @data: first arg to filler(data, page) function, often left as NULL
2499 * Read into the page cache. If a page already exists, and PageUptodate() is
2500 * not set, try to fill the page and wait for it to become unlocked.
2502 * If the page does not get brought uptodate, return -EIO.
2504 struct page *read_cache_page(struct address_space *mapping,
2506 int (*filler)(void *, struct page *),
2509 return do_read_cache_page(mapping, index, filler, data, mapping_gfp_mask(mapping));
2511 EXPORT_SYMBOL(read_cache_page);
2514 * read_cache_page_gfp - read into page cache, using specified page allocation flags.
2515 * @mapping: the page's address_space
2516 * @index: the page index
2517 * @gfp: the page allocator flags to use if allocating
2519 * This is the same as "read_mapping_page(mapping, index, NULL)", but with
2520 * any new page allocations done using the specified allocation flags.
2522 * If the page does not get brought uptodate, return -EIO.
2524 struct page *read_cache_page_gfp(struct address_space *mapping,
2528 filler_t *filler = (filler_t *)mapping->a_ops->readpage;
2530 return do_read_cache_page(mapping, index, filler, NULL, gfp);
2532 EXPORT_SYMBOL(read_cache_page_gfp);
2535 * Performs necessary checks before doing a write
2537 * Can adjust writing position or amount of bytes to write.
2538 * Returns appropriate error code that caller should return or
2539 * zero in case that write should be allowed.
2541 inline ssize_t generic_write_checks(struct kiocb *iocb, struct iov_iter *from)
2543 struct file *file = iocb->ki_filp;
2544 struct inode *inode = file->f_mapping->host;
2545 unsigned long limit = rlimit(RLIMIT_FSIZE);
2548 if (!iov_iter_count(from))
2551 /* FIXME: this is for backwards compatibility with 2.4 */
2552 if (iocb->ki_flags & IOCB_APPEND)
2553 iocb->ki_pos = i_size_read(inode);
2557 if (limit != RLIM_INFINITY) {
2558 if (iocb->ki_pos >= limit) {
2559 send_sig(SIGXFSZ, current, 0);
2562 iov_iter_truncate(from, limit - (unsigned long)pos);
2568 if (unlikely(pos + iov_iter_count(from) > MAX_NON_LFS &&
2569 !(file->f_flags & O_LARGEFILE))) {
2570 if (pos >= MAX_NON_LFS)
2572 iov_iter_truncate(from, MAX_NON_LFS - (unsigned long)pos);
2576 * Are we about to exceed the fs block limit ?
2578 * If we have written data it becomes a short write. If we have
2579 * exceeded without writing data we send a signal and return EFBIG.
2580 * Linus frestrict idea will clean these up nicely..
2582 if (unlikely(pos >= inode->i_sb->s_maxbytes))
2585 iov_iter_truncate(from, inode->i_sb->s_maxbytes - pos);
2586 return iov_iter_count(from);
2588 EXPORT_SYMBOL(generic_write_checks);
2590 int pagecache_write_begin(struct file *file, struct address_space *mapping,
2591 loff_t pos, unsigned len, unsigned flags,
2592 struct page **pagep, void **fsdata)
2594 const struct address_space_operations *aops = mapping->a_ops;
2596 return aops->write_begin(file, mapping, pos, len, flags,
2599 EXPORT_SYMBOL(pagecache_write_begin);
2601 int pagecache_write_end(struct file *file, struct address_space *mapping,
2602 loff_t pos, unsigned len, unsigned copied,
2603 struct page *page, void *fsdata)
2605 const struct address_space_operations *aops = mapping->a_ops;
2607 return aops->write_end(file, mapping, pos, len, copied, page, fsdata);
2609 EXPORT_SYMBOL(pagecache_write_end);
2612 generic_file_direct_write(struct kiocb *iocb, struct iov_iter *from)
2614 struct file *file = iocb->ki_filp;
2615 struct address_space *mapping = file->f_mapping;
2616 struct inode *inode = mapping->host;
2617 loff_t pos = iocb->ki_pos;
2621 struct iov_iter data;
2623 write_len = iov_iter_count(from);
2624 end = (pos + write_len - 1) >> PAGE_SHIFT;
2626 written = filemap_write_and_wait_range(mapping, pos, pos + write_len - 1);
2631 * After a write we want buffered reads to be sure to go to disk to get
2632 * the new data. We invalidate clean cached page from the region we're
2633 * about to write. We do this *before* the write so that we can return
2634 * without clobbering -EIOCBQUEUED from ->direct_IO().
2636 if (mapping->nrpages) {
2637 written = invalidate_inode_pages2_range(mapping,
2638 pos >> PAGE_SHIFT, end);
2640 * If a page can not be invalidated, return 0 to fall back
2641 * to buffered write.
2644 if (written == -EBUSY)
2651 written = mapping->a_ops->direct_IO(iocb, &data);
2654 * Finally, try again to invalidate clean pages which might have been
2655 * cached by non-direct readahead, or faulted in by get_user_pages()
2656 * if the source of the write was an mmap'ed region of the file
2657 * we're writing. Either one is a pretty crazy thing to do,
2658 * so we don't support it 100%. If this invalidation
2659 * fails, tough, the write still worked...
2661 if (mapping->nrpages) {
2662 invalidate_inode_pages2_range(mapping,
2663 pos >> PAGE_SHIFT, end);
2668 iov_iter_advance(from, written);
2669 if (pos > i_size_read(inode) && !S_ISBLK(inode->i_mode)) {
2670 i_size_write(inode, pos);
2671 mark_inode_dirty(inode);
2678 EXPORT_SYMBOL(generic_file_direct_write);
2681 * Find or create a page at the given pagecache position. Return the locked
2682 * page. This function is specifically for buffered writes.
2684 struct page *grab_cache_page_write_begin(struct address_space *mapping,
2685 pgoff_t index, unsigned flags)
2688 int fgp_flags = FGP_LOCK|FGP_WRITE|FGP_CREAT;
2690 if (flags & AOP_FLAG_NOFS)
2691 fgp_flags |= FGP_NOFS;
2693 page = pagecache_get_page(mapping, index, fgp_flags,
2694 mapping_gfp_mask(mapping));
2696 wait_for_stable_page(page);
2700 EXPORT_SYMBOL(grab_cache_page_write_begin);
2702 ssize_t generic_perform_write(struct file *file,
2703 struct iov_iter *i, loff_t pos)
2705 struct address_space *mapping = file->f_mapping;
2706 const struct address_space_operations *a_ops = mapping->a_ops;
2708 ssize_t written = 0;
2709 unsigned int flags = 0;
2712 * Copies from kernel address space cannot fail (NFSD is a big user).
2714 if (!iter_is_iovec(i))
2715 flags |= AOP_FLAG_UNINTERRUPTIBLE;
2719 unsigned long offset; /* Offset into pagecache page */
2720 unsigned long bytes; /* Bytes to write to page */
2721 size_t copied; /* Bytes copied from user */
2724 offset = (pos & (PAGE_SIZE - 1));
2725 bytes = min_t(unsigned long, PAGE_SIZE - offset,
2730 * Bring in the user page that we will copy from _first_.
2731 * Otherwise there's a nasty deadlock on copying from the
2732 * same page as we're writing to, without it being marked
2735 * Not only is this an optimisation, but it is also required
2736 * to check that the address is actually valid, when atomic
2737 * usercopies are used, below.
2739 if (unlikely(iov_iter_fault_in_readable(i, bytes))) {
2744 if (fatal_signal_pending(current)) {
2749 status = a_ops->write_begin(file, mapping, pos, bytes, flags,
2751 if (unlikely(status < 0))
2754 if (mapping_writably_mapped(mapping))
2755 flush_dcache_page(page);
2757 copied = iov_iter_copy_from_user_atomic(page, i, offset, bytes);
2758 flush_dcache_page(page);
2760 status = a_ops->write_end(file, mapping, pos, bytes, copied,
2762 if (unlikely(status < 0))
2768 iov_iter_advance(i, copied);
2769 if (unlikely(copied == 0)) {
2771 * If we were unable to copy any data at all, we must
2772 * fall back to a single segment length write.
2774 * If we didn't fallback here, we could livelock
2775 * because not all segments in the iov can be copied at
2776 * once without a pagefault.
2778 bytes = min_t(unsigned long, PAGE_SIZE - offset,
2779 iov_iter_single_seg_count(i));
2785 balance_dirty_pages_ratelimited(mapping);
2786 } while (iov_iter_count(i));
2788 return written ? written : status;
2790 EXPORT_SYMBOL(generic_perform_write);
2793 * __generic_file_write_iter - write data to a file
2794 * @iocb: IO state structure (file, offset, etc.)
2795 * @from: iov_iter with data to write
2797 * This function does all the work needed for actually writing data to a
2798 * file. It does all basic checks, removes SUID from the file, updates
2799 * modification times and calls proper subroutines depending on whether we
2800 * do direct IO or a standard buffered write.
2802 * It expects i_mutex to be grabbed unless we work on a block device or similar
2803 * object which does not need locking at all.
2805 * This function does *not* take care of syncing data in case of O_SYNC write.
2806 * A caller has to handle it. This is mainly due to the fact that we want to
2807 * avoid syncing under i_mutex.
2809 ssize_t __generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
2811 struct file *file = iocb->ki_filp;
2812 struct address_space * mapping = file->f_mapping;
2813 struct inode *inode = mapping->host;
2814 ssize_t written = 0;
2818 /* We can write back this queue in page reclaim */
2819 current->backing_dev_info = inode_to_bdi(inode);
2820 err = file_remove_privs(file);
2824 err = file_update_time(file);
2828 if (iocb->ki_flags & IOCB_DIRECT) {
2829 loff_t pos, endbyte;
2831 written = generic_file_direct_write(iocb, from);
2833 * If the write stopped short of completing, fall back to
2834 * buffered writes. Some filesystems do this for writes to
2835 * holes, for example. For DAX files, a buffered write will
2836 * not succeed (even if it did, DAX does not handle dirty
2837 * page-cache pages correctly).
2839 if (written < 0 || !iov_iter_count(from) || IS_DAX(inode))
2842 status = generic_perform_write(file, from, pos = iocb->ki_pos);
2844 * If generic_perform_write() returned a synchronous error
2845 * then we want to return the number of bytes which were
2846 * direct-written, or the error code if that was zero. Note
2847 * that this differs from normal direct-io semantics, which
2848 * will return -EFOO even if some bytes were written.
2850 if (unlikely(status < 0)) {
2855 * We need to ensure that the page cache pages are written to
2856 * disk and invalidated to preserve the expected O_DIRECT
2859 endbyte = pos + status - 1;
2860 err = filemap_write_and_wait_range(mapping, pos, endbyte);
2862 iocb->ki_pos = endbyte + 1;
2864 invalidate_mapping_pages(mapping,
2866 endbyte >> PAGE_SHIFT);
2869 * We don't know how much we wrote, so just return
2870 * the number of bytes which were direct-written
2874 written = generic_perform_write(file, from, iocb->ki_pos);
2875 if (likely(written > 0))
2876 iocb->ki_pos += written;
2879 current->backing_dev_info = NULL;
2880 return written ? written : err;
2882 EXPORT_SYMBOL(__generic_file_write_iter);
2885 * generic_file_write_iter - write data to a file
2886 * @iocb: IO state structure
2887 * @from: iov_iter with data to write
2889 * This is a wrapper around __generic_file_write_iter() to be used by most
2890 * filesystems. It takes care of syncing the file in case of O_SYNC file
2891 * and acquires i_mutex as needed.
2893 ssize_t generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
2895 struct file *file = iocb->ki_filp;
2896 struct inode *inode = file->f_mapping->host;
2900 ret = generic_write_checks(iocb, from);
2902 ret = __generic_file_write_iter(iocb, from);
2903 inode_unlock(inode);
2906 ret = generic_write_sync(iocb, ret);
2909 EXPORT_SYMBOL(generic_file_write_iter);
2912 * try_to_release_page() - release old fs-specific metadata on a page
2914 * @page: the page which the kernel is trying to free
2915 * @gfp_mask: memory allocation flags (and I/O mode)
2917 * The address_space is to try to release any data against the page
2918 * (presumably at page->private). If the release was successful, return `1'.
2919 * Otherwise return zero.
2921 * This may also be called if PG_fscache is set on a page, indicating that the
2922 * page is known to the local caching routines.
2924 * The @gfp_mask argument specifies whether I/O may be performed to release
2925 * this page (__GFP_IO), and whether the call may block (__GFP_RECLAIM & __GFP_FS).
2928 int try_to_release_page(struct page *page, gfp_t gfp_mask)
2930 struct address_space * const mapping = page->mapping;
2932 BUG_ON(!PageLocked(page));
2933 if (PageWriteback(page))
2936 if (mapping && mapping->a_ops->releasepage)
2937 return mapping->a_ops->releasepage(page, gfp_mask);
2938 return try_to_free_buffers(page);
2941 EXPORT_SYMBOL(try_to_release_page);