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. */
245 __mod_node_page_state(page_pgdat(page), NR_FILE_PAGES, -nr);
246 if (PageSwapBacked(page)) {
247 __mod_node_page_state(page_pgdat(page), NR_SHMEM, -nr);
248 if (PageTransHuge(page))
249 __dec_node_page_state(page, NR_SHMEM_THPS);
251 VM_BUG_ON_PAGE(PageTransHuge(page), page);
255 * At this point page must be either written or cleaned by truncate.
256 * Dirty page here signals a bug and loss of unwritten data.
258 * This fixes dirty accounting after removing the page entirely but
259 * leaves PageDirty set: it has no effect for truncated page and
260 * anyway will be cleared before returning page into buddy allocator.
262 if (WARN_ON_ONCE(PageDirty(page)))
263 account_page_cleaned(page, mapping, inode_to_wb(mapping->host));
267 * delete_from_page_cache - delete page from page cache
268 * @page: the page which the kernel is trying to remove from page cache
270 * This must be called only on pages that have been verified to be in the page
271 * cache and locked. It will never put the page into the free list, the caller
272 * has a reference on the page.
274 void delete_from_page_cache(struct page *page)
276 struct address_space *mapping = page_mapping(page);
278 void (*freepage)(struct page *);
280 BUG_ON(!PageLocked(page));
282 freepage = mapping->a_ops->freepage;
284 spin_lock_irqsave(&mapping->tree_lock, flags);
285 __delete_from_page_cache(page, NULL);
286 spin_unlock_irqrestore(&mapping->tree_lock, flags);
291 if (PageTransHuge(page) && !PageHuge(page)) {
292 page_ref_sub(page, HPAGE_PMD_NR);
293 VM_BUG_ON_PAGE(page_count(page) <= 0, page);
298 EXPORT_SYMBOL(delete_from_page_cache);
300 int filemap_check_errors(struct address_space *mapping)
303 /* Check for outstanding write errors */
304 if (test_bit(AS_ENOSPC, &mapping->flags) &&
305 test_and_clear_bit(AS_ENOSPC, &mapping->flags))
307 if (test_bit(AS_EIO, &mapping->flags) &&
308 test_and_clear_bit(AS_EIO, &mapping->flags))
312 EXPORT_SYMBOL(filemap_check_errors);
314 static int filemap_check_and_keep_errors(struct address_space *mapping)
316 /* Check for outstanding write errors */
317 if (test_bit(AS_EIO, &mapping->flags))
319 if (test_bit(AS_ENOSPC, &mapping->flags))
325 * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range
326 * @mapping: address space structure to write
327 * @start: offset in bytes where the range starts
328 * @end: offset in bytes where the range ends (inclusive)
329 * @sync_mode: enable synchronous operation
331 * Start writeback against all of a mapping's dirty pages that lie
332 * within the byte offsets <start, end> inclusive.
334 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
335 * opposed to a regular memory cleansing writeback. The difference between
336 * these two operations is that if a dirty page/buffer is encountered, it must
337 * be waited upon, and not just skipped over.
339 int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
340 loff_t end, int sync_mode)
343 struct writeback_control wbc = {
344 .sync_mode = sync_mode,
345 .nr_to_write = LONG_MAX,
346 .range_start = start,
350 if (!mapping_cap_writeback_dirty(mapping))
353 wbc_attach_fdatawrite_inode(&wbc, mapping->host);
354 ret = do_writepages(mapping, &wbc);
355 wbc_detach_inode(&wbc);
359 static inline int __filemap_fdatawrite(struct address_space *mapping,
362 return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode);
365 int filemap_fdatawrite(struct address_space *mapping)
367 return __filemap_fdatawrite(mapping, WB_SYNC_ALL);
369 EXPORT_SYMBOL(filemap_fdatawrite);
371 int filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
374 return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL);
376 EXPORT_SYMBOL(filemap_fdatawrite_range);
379 * filemap_flush - mostly a non-blocking flush
380 * @mapping: target address_space
382 * This is a mostly non-blocking flush. Not suitable for data-integrity
383 * purposes - I/O may not be started against all dirty pages.
385 int filemap_flush(struct address_space *mapping)
387 return __filemap_fdatawrite(mapping, WB_SYNC_NONE);
389 EXPORT_SYMBOL(filemap_flush);
392 * filemap_range_has_page - check if a page exists in range.
393 * @mapping: address space within which to check
394 * @start_byte: offset in bytes where the range starts
395 * @end_byte: offset in bytes where the range ends (inclusive)
397 * Find at least one page in the range supplied, usually used to check if
398 * direct writing in this range will trigger a writeback.
400 bool filemap_range_has_page(struct address_space *mapping,
401 loff_t start_byte, loff_t end_byte)
403 pgoff_t index = start_byte >> PAGE_SHIFT;
404 pgoff_t end = end_byte >> PAGE_SHIFT;
408 if (end_byte < start_byte)
411 if (mapping->nrpages == 0)
414 pagevec_init(&pvec, 0);
415 if (!pagevec_lookup(&pvec, mapping, index, 1))
417 ret = (pvec.pages[0]->index <= end);
418 pagevec_release(&pvec);
421 EXPORT_SYMBOL(filemap_range_has_page);
423 static void __filemap_fdatawait_range(struct address_space *mapping,
424 loff_t start_byte, loff_t end_byte)
426 pgoff_t index = start_byte >> PAGE_SHIFT;
427 pgoff_t end = end_byte >> PAGE_SHIFT;
431 if (end_byte < start_byte)
434 pagevec_init(&pvec, 0);
435 while ((index <= end) &&
436 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
437 PAGECACHE_TAG_WRITEBACK,
438 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1)) != 0) {
441 for (i = 0; i < nr_pages; i++) {
442 struct page *page = pvec.pages[i];
444 /* until radix tree lookup accepts end_index */
445 if (page->index > end)
448 wait_on_page_writeback(page);
449 ClearPageError(page);
451 pagevec_release(&pvec);
457 * filemap_fdatawait_range - wait for writeback to complete
458 * @mapping: address space structure to wait for
459 * @start_byte: offset in bytes where the range starts
460 * @end_byte: offset in bytes where the range ends (inclusive)
462 * Walk the list of under-writeback pages of the given address space
463 * in the given range and wait for all of them. Check error status of
464 * the address space and return it.
466 * Since the error status of the address space is cleared by this function,
467 * callers are responsible for checking the return value and handling and/or
468 * reporting the error.
470 int filemap_fdatawait_range(struct address_space *mapping, loff_t start_byte,
473 __filemap_fdatawait_range(mapping, start_byte, end_byte);
474 return filemap_check_errors(mapping);
476 EXPORT_SYMBOL(filemap_fdatawait_range);
479 * filemap_fdatawait_keep_errors - wait for writeback without clearing errors
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. Unlike filemap_fdatawait(), this function
484 * does not clear error status of the address space.
486 * Use this function if callers don't handle errors themselves. Expected
487 * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2),
490 int filemap_fdatawait_keep_errors(struct address_space *mapping)
492 loff_t i_size = i_size_read(mapping->host);
497 __filemap_fdatawait_range(mapping, 0, i_size - 1);
498 return filemap_check_and_keep_errors(mapping);
500 EXPORT_SYMBOL(filemap_fdatawait_keep_errors);
503 * filemap_fdatawait - wait for all under-writeback pages to complete
504 * @mapping: address space structure to wait for
506 * Walk the list of under-writeback pages of the given address space
507 * and wait for all of them. Check error status of the address space
510 * Since the error status of the address space is cleared by this function,
511 * callers are responsible for checking the return value and handling and/or
512 * reporting the error.
514 int filemap_fdatawait(struct address_space *mapping)
516 loff_t i_size = i_size_read(mapping->host);
521 return filemap_fdatawait_range(mapping, 0, i_size - 1);
523 EXPORT_SYMBOL(filemap_fdatawait);
525 static bool mapping_needs_writeback(struct address_space *mapping)
527 return (!dax_mapping(mapping) && mapping->nrpages) ||
528 (dax_mapping(mapping) && mapping->nrexceptional);
531 int filemap_write_and_wait(struct address_space *mapping)
535 if (mapping_needs_writeback(mapping)) {
536 err = filemap_fdatawrite(mapping);
538 * Even if the above returned error, the pages may be
539 * written partially (e.g. -ENOSPC), so we wait for it.
540 * But the -EIO is special case, it may indicate the worst
541 * thing (e.g. bug) happened, so we avoid waiting for it.
544 int err2 = filemap_fdatawait(mapping);
548 /* Clear any previously stored errors */
549 filemap_check_errors(mapping);
552 err = filemap_check_errors(mapping);
556 EXPORT_SYMBOL(filemap_write_and_wait);
559 * filemap_write_and_wait_range - write out & wait on a file range
560 * @mapping: the address_space for the pages
561 * @lstart: offset in bytes where the range starts
562 * @lend: offset in bytes where the range ends (inclusive)
564 * Write out and wait upon file offsets lstart->lend, inclusive.
566 * Note that @lend is inclusive (describes the last byte to be written) so
567 * that this function can be used to write to the very end-of-file (end = -1).
569 int filemap_write_and_wait_range(struct address_space *mapping,
570 loff_t lstart, loff_t lend)
574 if (mapping_needs_writeback(mapping)) {
575 err = __filemap_fdatawrite_range(mapping, lstart, lend,
577 /* See comment of filemap_write_and_wait() */
579 int err2 = filemap_fdatawait_range(mapping,
584 /* Clear any previously stored errors */
585 filemap_check_errors(mapping);
588 err = filemap_check_errors(mapping);
592 EXPORT_SYMBOL(filemap_write_and_wait_range);
594 void __filemap_set_wb_err(struct address_space *mapping, int err)
596 errseq_t eseq = errseq_set(&mapping->wb_err, err);
598 trace_filemap_set_wb_err(mapping, eseq);
600 EXPORT_SYMBOL(__filemap_set_wb_err);
603 * file_check_and_advance_wb_err - report wb error (if any) that was previously
604 * and advance wb_err to current one
605 * @file: struct file on which the error is being reported
607 * When userland calls fsync (or something like nfsd does the equivalent), we
608 * want to report any writeback errors that occurred since the last fsync (or
609 * since the file was opened if there haven't been any).
611 * Grab the wb_err from the mapping. If it matches what we have in the file,
612 * then just quickly return 0. The file is all caught up.
614 * If it doesn't match, then take the mapping value, set the "seen" flag in
615 * it and try to swap it into place. If it works, or another task beat us
616 * to it with the new value, then update the f_wb_err and return the error
617 * portion. The error at this point must be reported via proper channels
618 * (a'la fsync, or NFS COMMIT operation, etc.).
620 * While we handle mapping->wb_err with atomic operations, the f_wb_err
621 * value is protected by the f_lock since we must ensure that it reflects
622 * the latest value swapped in for this file descriptor.
624 int file_check_and_advance_wb_err(struct file *file)
627 errseq_t old = READ_ONCE(file->f_wb_err);
628 struct address_space *mapping = file->f_mapping;
630 /* Locklessly handle the common case where nothing has changed */
631 if (errseq_check(&mapping->wb_err, old)) {
632 /* Something changed, must use slow path */
633 spin_lock(&file->f_lock);
634 old = file->f_wb_err;
635 err = errseq_check_and_advance(&mapping->wb_err,
637 trace_file_check_and_advance_wb_err(file, old);
638 spin_unlock(&file->f_lock);
642 EXPORT_SYMBOL(file_check_and_advance_wb_err);
645 * file_write_and_wait_range - write out & wait on a file range
646 * @file: file pointing to address_space with pages
647 * @lstart: offset in bytes where the range starts
648 * @lend: offset in bytes where the range ends (inclusive)
650 * Write out and wait upon file offsets lstart->lend, inclusive.
652 * Note that @lend is inclusive (describes the last byte to be written) so
653 * that this function can be used to write to the very end-of-file (end = -1).
655 * After writing out and waiting on the data, we check and advance the
656 * f_wb_err cursor to the latest value, and return any errors detected there.
658 int file_write_and_wait_range(struct file *file, loff_t lstart, loff_t lend)
661 struct address_space *mapping = file->f_mapping;
663 if (mapping_needs_writeback(mapping)) {
664 err = __filemap_fdatawrite_range(mapping, lstart, lend,
666 /* See comment of filemap_write_and_wait() */
668 __filemap_fdatawait_range(mapping, lstart, lend);
670 err2 = file_check_and_advance_wb_err(file);
675 EXPORT_SYMBOL(file_write_and_wait_range);
678 * replace_page_cache_page - replace a pagecache page with a new one
679 * @old: page to be replaced
680 * @new: page to replace with
681 * @gfp_mask: allocation mode
683 * This function replaces a page in the pagecache with a new one. On
684 * success it acquires the pagecache reference for the new page and
685 * drops it for the old page. Both the old and new pages must be
686 * locked. This function does not add the new page to the LRU, the
687 * caller must do that.
689 * The remove + add is atomic. The only way this function can fail is
690 * memory allocation failure.
692 int replace_page_cache_page(struct page *old, struct page *new, gfp_t gfp_mask)
696 VM_BUG_ON_PAGE(!PageLocked(old), old);
697 VM_BUG_ON_PAGE(!PageLocked(new), new);
698 VM_BUG_ON_PAGE(new->mapping, new);
700 error = radix_tree_preload(gfp_mask & ~__GFP_HIGHMEM);
702 struct address_space *mapping = old->mapping;
703 void (*freepage)(struct page *);
706 pgoff_t offset = old->index;
707 freepage = mapping->a_ops->freepage;
710 new->mapping = mapping;
713 spin_lock_irqsave(&mapping->tree_lock, flags);
714 __delete_from_page_cache(old, NULL);
715 error = page_cache_tree_insert(mapping, new, NULL);
719 * hugetlb pages do not participate in page cache accounting.
722 __inc_node_page_state(new, NR_FILE_PAGES);
723 if (PageSwapBacked(new))
724 __inc_node_page_state(new, NR_SHMEM);
725 spin_unlock_irqrestore(&mapping->tree_lock, flags);
726 mem_cgroup_migrate(old, new);
727 radix_tree_preload_end();
735 EXPORT_SYMBOL_GPL(replace_page_cache_page);
737 static int __add_to_page_cache_locked(struct page *page,
738 struct address_space *mapping,
739 pgoff_t offset, gfp_t gfp_mask,
742 int huge = PageHuge(page);
743 struct mem_cgroup *memcg;
746 VM_BUG_ON_PAGE(!PageLocked(page), page);
747 VM_BUG_ON_PAGE(PageSwapBacked(page), page);
750 error = mem_cgroup_try_charge(page, current->mm,
751 gfp_mask, &memcg, false);
756 error = radix_tree_maybe_preload(gfp_mask & ~__GFP_HIGHMEM);
759 mem_cgroup_cancel_charge(page, memcg, false);
764 page->mapping = mapping;
765 page->index = offset;
767 spin_lock_irq(&mapping->tree_lock);
768 error = page_cache_tree_insert(mapping, page, shadowp);
769 radix_tree_preload_end();
773 /* hugetlb pages do not participate in page cache accounting. */
775 __inc_node_page_state(page, NR_FILE_PAGES);
776 spin_unlock_irq(&mapping->tree_lock);
778 mem_cgroup_commit_charge(page, memcg, false, false);
779 trace_mm_filemap_add_to_page_cache(page);
782 page->mapping = NULL;
783 /* Leave page->index set: truncation relies upon it */
784 spin_unlock_irq(&mapping->tree_lock);
786 mem_cgroup_cancel_charge(page, memcg, false);
792 * add_to_page_cache_locked - add a locked page to the pagecache
794 * @mapping: the page's address_space
795 * @offset: page index
796 * @gfp_mask: page allocation mode
798 * This function is used to add a page to the pagecache. It must be locked.
799 * This function does not add the page to the LRU. The caller must do that.
801 int add_to_page_cache_locked(struct page *page, struct address_space *mapping,
802 pgoff_t offset, gfp_t gfp_mask)
804 return __add_to_page_cache_locked(page, mapping, offset,
807 EXPORT_SYMBOL(add_to_page_cache_locked);
809 int add_to_page_cache_lru(struct page *page, struct address_space *mapping,
810 pgoff_t offset, gfp_t gfp_mask)
815 __SetPageLocked(page);
816 ret = __add_to_page_cache_locked(page, mapping, offset,
819 __ClearPageLocked(page);
822 * The page might have been evicted from cache only
823 * recently, in which case it should be activated like
824 * any other repeatedly accessed page.
825 * The exception is pages getting rewritten; evicting other
826 * data from the working set, only to cache data that will
827 * get overwritten with something else, is a waste of memory.
829 if (!(gfp_mask & __GFP_WRITE) &&
830 shadow && workingset_refault(shadow)) {
832 workingset_activation(page);
834 ClearPageActive(page);
839 EXPORT_SYMBOL_GPL(add_to_page_cache_lru);
842 struct page *__page_cache_alloc(gfp_t gfp)
847 if (cpuset_do_page_mem_spread()) {
848 unsigned int cpuset_mems_cookie;
850 cpuset_mems_cookie = read_mems_allowed_begin();
851 n = cpuset_mem_spread_node();
852 page = __alloc_pages_node(n, gfp, 0);
853 } while (!page && read_mems_allowed_retry(cpuset_mems_cookie));
857 return alloc_pages(gfp, 0);
859 EXPORT_SYMBOL(__page_cache_alloc);
863 * In order to wait for pages to become available there must be
864 * waitqueues associated with pages. By using a hash table of
865 * waitqueues where the bucket discipline is to maintain all
866 * waiters on the same queue and wake all when any of the pages
867 * become available, and for the woken contexts to check to be
868 * sure the appropriate page became available, this saves space
869 * at a cost of "thundering herd" phenomena during rare hash
872 #define PAGE_WAIT_TABLE_BITS 8
873 #define PAGE_WAIT_TABLE_SIZE (1 << PAGE_WAIT_TABLE_BITS)
874 static wait_queue_head_t page_wait_table[PAGE_WAIT_TABLE_SIZE] __cacheline_aligned;
876 static wait_queue_head_t *page_waitqueue(struct page *page)
878 return &page_wait_table[hash_ptr(page, PAGE_WAIT_TABLE_BITS)];
881 void __init pagecache_init(void)
885 for (i = 0; i < PAGE_WAIT_TABLE_SIZE; i++)
886 init_waitqueue_head(&page_wait_table[i]);
888 page_writeback_init();
891 struct wait_page_key {
897 struct wait_page_queue {
900 wait_queue_entry_t wait;
903 static int wake_page_function(wait_queue_entry_t *wait, unsigned mode, int sync, void *arg)
905 struct wait_page_key *key = arg;
906 struct wait_page_queue *wait_page
907 = container_of(wait, struct wait_page_queue, wait);
909 if (wait_page->page != key->page)
913 if (wait_page->bit_nr != key->bit_nr)
915 if (test_bit(key->bit_nr, &key->page->flags))
918 return autoremove_wake_function(wait, mode, sync, key);
921 static void wake_up_page_bit(struct page *page, int bit_nr)
923 wait_queue_head_t *q = page_waitqueue(page);
924 struct wait_page_key key;
931 spin_lock_irqsave(&q->lock, flags);
932 __wake_up_locked_key(q, TASK_NORMAL, &key);
934 * It is possible for other pages to have collided on the waitqueue
935 * hash, so in that case check for a page match. That prevents a long-
938 * It is still possible to miss a case here, when we woke page waiters
939 * and removed them from the waitqueue, but there are still other
942 if (!waitqueue_active(q) || !key.page_match) {
943 ClearPageWaiters(page);
945 * It's possible to miss clearing Waiters here, when we woke
946 * our page waiters, but the hashed waitqueue has waiters for
949 * That's okay, it's a rare case. The next waker will clear it.
952 spin_unlock_irqrestore(&q->lock, flags);
955 static void wake_up_page(struct page *page, int bit)
957 if (!PageWaiters(page))
959 wake_up_page_bit(page, bit);
962 static inline int wait_on_page_bit_common(wait_queue_head_t *q,
963 struct page *page, int bit_nr, int state, bool lock)
965 struct wait_page_queue wait_page;
966 wait_queue_entry_t *wait = &wait_page.wait;
970 wait->func = wake_page_function;
971 wait_page.page = page;
972 wait_page.bit_nr = bit_nr;
975 spin_lock_irq(&q->lock);
977 if (likely(list_empty(&wait->entry))) {
979 __add_wait_queue_entry_tail_exclusive(q, wait);
981 __add_wait_queue(q, wait);
982 SetPageWaiters(page);
985 set_current_state(state);
987 spin_unlock_irq(&q->lock);
989 if (likely(test_bit(bit_nr, &page->flags))) {
991 if (unlikely(signal_pending_state(state, current))) {
998 if (!test_and_set_bit_lock(bit_nr, &page->flags))
1001 if (!test_bit(bit_nr, &page->flags))
1006 finish_wait(q, wait);
1009 * A signal could leave PageWaiters set. Clearing it here if
1010 * !waitqueue_active would be possible (by open-coding finish_wait),
1011 * but still fail to catch it in the case of wait hash collision. We
1012 * already can fail to clear wait hash collision cases, so don't
1013 * bother with signals either.
1019 void wait_on_page_bit(struct page *page, int bit_nr)
1021 wait_queue_head_t *q = page_waitqueue(page);
1022 wait_on_page_bit_common(q, page, bit_nr, TASK_UNINTERRUPTIBLE, false);
1024 EXPORT_SYMBOL(wait_on_page_bit);
1026 int wait_on_page_bit_killable(struct page *page, int bit_nr)
1028 wait_queue_head_t *q = page_waitqueue(page);
1029 return wait_on_page_bit_common(q, page, bit_nr, TASK_KILLABLE, false);
1033 * add_page_wait_queue - Add an arbitrary waiter to a page's wait queue
1034 * @page: Page defining the wait queue of interest
1035 * @waiter: Waiter to add to the queue
1037 * Add an arbitrary @waiter to the wait queue for the nominated @page.
1039 void add_page_wait_queue(struct page *page, wait_queue_entry_t *waiter)
1041 wait_queue_head_t *q = page_waitqueue(page);
1042 unsigned long flags;
1044 spin_lock_irqsave(&q->lock, flags);
1045 __add_wait_queue(q, waiter);
1046 SetPageWaiters(page);
1047 spin_unlock_irqrestore(&q->lock, flags);
1049 EXPORT_SYMBOL_GPL(add_page_wait_queue);
1051 #ifndef clear_bit_unlock_is_negative_byte
1054 * PG_waiters is the high bit in the same byte as PG_lock.
1056 * On x86 (and on many other architectures), we can clear PG_lock and
1057 * test the sign bit at the same time. But if the architecture does
1058 * not support that special operation, we just do this all by hand
1061 * The read of PG_waiters has to be after (or concurrently with) PG_locked
1062 * being cleared, but a memory barrier should be unneccssary since it is
1063 * in the same byte as PG_locked.
1065 static inline bool clear_bit_unlock_is_negative_byte(long nr, volatile void *mem)
1067 clear_bit_unlock(nr, mem);
1068 /* smp_mb__after_atomic(); */
1069 return test_bit(PG_waiters, mem);
1075 * unlock_page - unlock a locked page
1078 * Unlocks the page and wakes up sleepers in ___wait_on_page_locked().
1079 * Also wakes sleepers in wait_on_page_writeback() because the wakeup
1080 * mechanism between PageLocked pages and PageWriteback pages is shared.
1081 * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep.
1083 * Note that this depends on PG_waiters being the sign bit in the byte
1084 * that contains PG_locked - thus the BUILD_BUG_ON(). That allows us to
1085 * clear the PG_locked bit and test PG_waiters at the same time fairly
1086 * portably (architectures that do LL/SC can test any bit, while x86 can
1087 * test the sign bit).
1089 void unlock_page(struct page *page)
1091 BUILD_BUG_ON(PG_waiters != 7);
1092 page = compound_head(page);
1093 VM_BUG_ON_PAGE(!PageLocked(page), page);
1094 if (clear_bit_unlock_is_negative_byte(PG_locked, &page->flags))
1095 wake_up_page_bit(page, PG_locked);
1097 EXPORT_SYMBOL(unlock_page);
1100 * end_page_writeback - end writeback against a page
1103 void end_page_writeback(struct page *page)
1106 * TestClearPageReclaim could be used here but it is an atomic
1107 * operation and overkill in this particular case. Failing to
1108 * shuffle a page marked for immediate reclaim is too mild to
1109 * justify taking an atomic operation penalty at the end of
1110 * ever page writeback.
1112 if (PageReclaim(page)) {
1113 ClearPageReclaim(page);
1114 rotate_reclaimable_page(page);
1117 if (!test_clear_page_writeback(page))
1120 smp_mb__after_atomic();
1121 wake_up_page(page, PG_writeback);
1123 EXPORT_SYMBOL(end_page_writeback);
1126 * After completing I/O on a page, call this routine to update the page
1127 * flags appropriately
1129 void page_endio(struct page *page, bool is_write, int err)
1133 SetPageUptodate(page);
1135 ClearPageUptodate(page);
1141 struct address_space *mapping;
1144 mapping = page_mapping(page);
1146 mapping_set_error(mapping, err);
1148 end_page_writeback(page);
1151 EXPORT_SYMBOL_GPL(page_endio);
1154 * __lock_page - get a lock on the page, assuming we need to sleep to get it
1155 * @__page: the page to lock
1157 void __lock_page(struct page *__page)
1159 struct page *page = compound_head(__page);
1160 wait_queue_head_t *q = page_waitqueue(page);
1161 wait_on_page_bit_common(q, page, PG_locked, TASK_UNINTERRUPTIBLE, true);
1163 EXPORT_SYMBOL(__lock_page);
1165 int __lock_page_killable(struct page *__page)
1167 struct page *page = compound_head(__page);
1168 wait_queue_head_t *q = page_waitqueue(page);
1169 return wait_on_page_bit_common(q, page, PG_locked, TASK_KILLABLE, true);
1171 EXPORT_SYMBOL_GPL(__lock_page_killable);
1175 * 1 - page is locked; mmap_sem is still held.
1176 * 0 - page is not locked.
1177 * mmap_sem has been released (up_read()), unless flags had both
1178 * FAULT_FLAG_ALLOW_RETRY and FAULT_FLAG_RETRY_NOWAIT set, in
1179 * which case mmap_sem is still held.
1181 * If neither ALLOW_RETRY nor KILLABLE are set, will always return 1
1182 * with the page locked and the mmap_sem unperturbed.
1184 int __lock_page_or_retry(struct page *page, struct mm_struct *mm,
1187 if (flags & FAULT_FLAG_ALLOW_RETRY) {
1189 * CAUTION! In this case, mmap_sem is not released
1190 * even though return 0.
1192 if (flags & FAULT_FLAG_RETRY_NOWAIT)
1195 up_read(&mm->mmap_sem);
1196 if (flags & FAULT_FLAG_KILLABLE)
1197 wait_on_page_locked_killable(page);
1199 wait_on_page_locked(page);
1202 if (flags & FAULT_FLAG_KILLABLE) {
1205 ret = __lock_page_killable(page);
1207 up_read(&mm->mmap_sem);
1217 * page_cache_next_hole - find the next hole (not-present entry)
1220 * @max_scan: maximum range to search
1222 * Search the set [index, min(index+max_scan-1, MAX_INDEX)] for the
1223 * lowest indexed hole.
1225 * Returns: the index of the hole if found, otherwise returns an index
1226 * outside of the set specified (in which case 'return - index >=
1227 * max_scan' will be true). In rare cases of index wrap-around, 0 will
1230 * page_cache_next_hole may be called under rcu_read_lock. However,
1231 * like radix_tree_gang_lookup, this will not atomically search a
1232 * snapshot of the tree at a single point in time. For example, if a
1233 * hole is created at index 5, then subsequently a hole is created at
1234 * index 10, page_cache_next_hole covering both indexes may return 10
1235 * if called under rcu_read_lock.
1237 pgoff_t page_cache_next_hole(struct address_space *mapping,
1238 pgoff_t index, unsigned long max_scan)
1242 for (i = 0; i < max_scan; i++) {
1245 page = radix_tree_lookup(&mapping->page_tree, index);
1246 if (!page || radix_tree_exceptional_entry(page))
1255 EXPORT_SYMBOL(page_cache_next_hole);
1258 * page_cache_prev_hole - find the prev hole (not-present entry)
1261 * @max_scan: maximum range to search
1263 * Search backwards in the range [max(index-max_scan+1, 0), index] for
1266 * Returns: the index of the hole if found, otherwise returns an index
1267 * outside of the set specified (in which case 'index - return >=
1268 * max_scan' will be true). In rare cases of wrap-around, ULONG_MAX
1271 * page_cache_prev_hole may be called under rcu_read_lock. However,
1272 * like radix_tree_gang_lookup, this will not atomically search a
1273 * snapshot of the tree at a single point in time. For example, if a
1274 * hole is created at index 10, then subsequently a hole is created at
1275 * index 5, page_cache_prev_hole covering both indexes may return 5 if
1276 * called under rcu_read_lock.
1278 pgoff_t page_cache_prev_hole(struct address_space *mapping,
1279 pgoff_t index, unsigned long max_scan)
1283 for (i = 0; i < max_scan; i++) {
1286 page = radix_tree_lookup(&mapping->page_tree, index);
1287 if (!page || radix_tree_exceptional_entry(page))
1290 if (index == ULONG_MAX)
1296 EXPORT_SYMBOL(page_cache_prev_hole);
1299 * find_get_entry - find and get a page cache entry
1300 * @mapping: the address_space to search
1301 * @offset: the page cache index
1303 * Looks up the page cache slot at @mapping & @offset. If there is a
1304 * page cache page, it is returned with an increased refcount.
1306 * If the slot holds a shadow entry of a previously evicted page, or a
1307 * swap entry from shmem/tmpfs, it is returned.
1309 * Otherwise, %NULL is returned.
1311 struct page *find_get_entry(struct address_space *mapping, pgoff_t offset)
1314 struct page *head, *page;
1319 pagep = radix_tree_lookup_slot(&mapping->page_tree, offset);
1321 page = radix_tree_deref_slot(pagep);
1322 if (unlikely(!page))
1324 if (radix_tree_exception(page)) {
1325 if (radix_tree_deref_retry(page))
1328 * A shadow entry of a recently evicted page,
1329 * or a swap entry from shmem/tmpfs. Return
1330 * it without attempting to raise page count.
1335 head = compound_head(page);
1336 if (!page_cache_get_speculative(head))
1339 /* The page was split under us? */
1340 if (compound_head(page) != head) {
1346 * Has the page moved?
1347 * This is part of the lockless pagecache protocol. See
1348 * include/linux/pagemap.h for details.
1350 if (unlikely(page != *pagep)) {
1360 EXPORT_SYMBOL(find_get_entry);
1363 * find_lock_entry - locate, pin and lock a page cache entry
1364 * @mapping: the address_space to search
1365 * @offset: the page cache index
1367 * Looks up the page cache slot at @mapping & @offset. If there is a
1368 * page cache page, it is returned locked and with an increased
1371 * If the slot holds a shadow entry of a previously evicted page, or a
1372 * swap entry from shmem/tmpfs, it is returned.
1374 * Otherwise, %NULL is returned.
1376 * find_lock_entry() may sleep.
1378 struct page *find_lock_entry(struct address_space *mapping, pgoff_t offset)
1383 page = find_get_entry(mapping, offset);
1384 if (page && !radix_tree_exception(page)) {
1386 /* Has the page been truncated? */
1387 if (unlikely(page_mapping(page) != mapping)) {
1392 VM_BUG_ON_PAGE(page_to_pgoff(page) != offset, page);
1396 EXPORT_SYMBOL(find_lock_entry);
1399 * pagecache_get_page - find and get a page reference
1400 * @mapping: the address_space to search
1401 * @offset: the page index
1402 * @fgp_flags: PCG flags
1403 * @gfp_mask: gfp mask to use for the page cache data page allocation
1405 * Looks up the page cache slot at @mapping & @offset.
1407 * PCG flags modify how the page is returned.
1409 * @fgp_flags can be:
1411 * - FGP_ACCESSED: the page will be marked accessed
1412 * - FGP_LOCK: Page is return locked
1413 * - FGP_CREAT: If page is not present then a new page is allocated using
1414 * @gfp_mask and added to the page cache and the VM's LRU
1415 * list. The page is returned locked and with an increased
1416 * refcount. Otherwise, NULL is returned.
1418 * If FGP_LOCK or FGP_CREAT are specified then the function may sleep even
1419 * if the GFP flags specified for FGP_CREAT are atomic.
1421 * If there is a page cache page, it is returned with an increased refcount.
1423 struct page *pagecache_get_page(struct address_space *mapping, pgoff_t offset,
1424 int fgp_flags, gfp_t gfp_mask)
1429 page = find_get_entry(mapping, offset);
1430 if (radix_tree_exceptional_entry(page))
1435 if (fgp_flags & FGP_LOCK) {
1436 if (fgp_flags & FGP_NOWAIT) {
1437 if (!trylock_page(page)) {
1445 /* Has the page been truncated? */
1446 if (unlikely(page->mapping != mapping)) {
1451 VM_BUG_ON_PAGE(page->index != offset, page);
1454 if (page && (fgp_flags & FGP_ACCESSED))
1455 mark_page_accessed(page);
1458 if (!page && (fgp_flags & FGP_CREAT)) {
1460 if ((fgp_flags & FGP_WRITE) && mapping_cap_account_dirty(mapping))
1461 gfp_mask |= __GFP_WRITE;
1462 if (fgp_flags & FGP_NOFS)
1463 gfp_mask &= ~__GFP_FS;
1465 page = __page_cache_alloc(gfp_mask);
1469 if (WARN_ON_ONCE(!(fgp_flags & FGP_LOCK)))
1470 fgp_flags |= FGP_LOCK;
1472 /* Init accessed so avoid atomic mark_page_accessed later */
1473 if (fgp_flags & FGP_ACCESSED)
1474 __SetPageReferenced(page);
1476 err = add_to_page_cache_lru(page, mapping, offset,
1477 gfp_mask & GFP_RECLAIM_MASK);
1478 if (unlikely(err)) {
1488 EXPORT_SYMBOL(pagecache_get_page);
1491 * find_get_entries - gang pagecache lookup
1492 * @mapping: The address_space to search
1493 * @start: The starting page cache index
1494 * @nr_entries: The maximum number of entries
1495 * @entries: Where the resulting entries are placed
1496 * @indices: The cache indices corresponding to the entries in @entries
1498 * find_get_entries() will search for and return a group of up to
1499 * @nr_entries entries in the mapping. The entries are placed at
1500 * @entries. find_get_entries() takes a reference against any actual
1503 * The search returns a group of mapping-contiguous page cache entries
1504 * with ascending indexes. There may be holes in the indices due to
1505 * not-present pages.
1507 * Any shadow entries of evicted pages, or swap entries from
1508 * shmem/tmpfs, are included in the returned array.
1510 * find_get_entries() returns the number of pages and shadow entries
1513 unsigned find_get_entries(struct address_space *mapping,
1514 pgoff_t start, unsigned int nr_entries,
1515 struct page **entries, pgoff_t *indices)
1518 unsigned int ret = 0;
1519 struct radix_tree_iter iter;
1525 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) {
1526 struct page *head, *page;
1528 page = radix_tree_deref_slot(slot);
1529 if (unlikely(!page))
1531 if (radix_tree_exception(page)) {
1532 if (radix_tree_deref_retry(page)) {
1533 slot = radix_tree_iter_retry(&iter);
1537 * A shadow entry of a recently evicted page, a swap
1538 * entry from shmem/tmpfs or a DAX entry. Return it
1539 * without attempting to raise page count.
1544 head = compound_head(page);
1545 if (!page_cache_get_speculative(head))
1548 /* The page was split under us? */
1549 if (compound_head(page) != head) {
1554 /* Has the page moved? */
1555 if (unlikely(page != *slot)) {
1560 indices[ret] = iter.index;
1561 entries[ret] = page;
1562 if (++ret == nr_entries)
1570 * find_get_pages - gang pagecache lookup
1571 * @mapping: The address_space to search
1572 * @start: The starting page index
1573 * @nr_pages: The maximum number of pages
1574 * @pages: Where the resulting pages are placed
1576 * find_get_pages() will search for and return a group of up to
1577 * @nr_pages pages in the mapping. The pages are placed at @pages.
1578 * find_get_pages() takes a reference against the returned pages.
1580 * The search returns a group of mapping-contiguous pages with ascending
1581 * indexes. There may be holes in the indices due to not-present pages.
1583 * find_get_pages() returns the number of pages which were found.
1585 unsigned find_get_pages(struct address_space *mapping, pgoff_t start,
1586 unsigned int nr_pages, struct page **pages)
1588 struct radix_tree_iter iter;
1592 if (unlikely(!nr_pages))
1596 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) {
1597 struct page *head, *page;
1599 page = radix_tree_deref_slot(slot);
1600 if (unlikely(!page))
1603 if (radix_tree_exception(page)) {
1604 if (radix_tree_deref_retry(page)) {
1605 slot = radix_tree_iter_retry(&iter);
1609 * A shadow entry of a recently evicted page,
1610 * or a swap entry from shmem/tmpfs. Skip
1616 head = compound_head(page);
1617 if (!page_cache_get_speculative(head))
1620 /* The page was split under us? */
1621 if (compound_head(page) != head) {
1626 /* Has the page moved? */
1627 if (unlikely(page != *slot)) {
1633 if (++ret == nr_pages)
1642 * find_get_pages_contig - gang contiguous pagecache lookup
1643 * @mapping: The address_space to search
1644 * @index: The starting page index
1645 * @nr_pages: The maximum number of pages
1646 * @pages: Where the resulting pages are placed
1648 * find_get_pages_contig() works exactly like find_get_pages(), except
1649 * that the returned number of pages are guaranteed to be contiguous.
1651 * find_get_pages_contig() returns the number of pages which were found.
1653 unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index,
1654 unsigned int nr_pages, struct page **pages)
1656 struct radix_tree_iter iter;
1658 unsigned int ret = 0;
1660 if (unlikely(!nr_pages))
1664 radix_tree_for_each_contig(slot, &mapping->page_tree, &iter, index) {
1665 struct page *head, *page;
1667 page = radix_tree_deref_slot(slot);
1668 /* The hole, there no reason to continue */
1669 if (unlikely(!page))
1672 if (radix_tree_exception(page)) {
1673 if (radix_tree_deref_retry(page)) {
1674 slot = radix_tree_iter_retry(&iter);
1678 * A shadow entry of a recently evicted page,
1679 * or a swap entry from shmem/tmpfs. Stop
1680 * looking for contiguous pages.
1685 head = compound_head(page);
1686 if (!page_cache_get_speculative(head))
1689 /* The page was split under us? */
1690 if (compound_head(page) != head) {
1695 /* Has the page moved? */
1696 if (unlikely(page != *slot)) {
1702 * must check mapping and index after taking the ref.
1703 * otherwise we can get both false positives and false
1704 * negatives, which is just confusing to the caller.
1706 if (page->mapping == NULL || page_to_pgoff(page) != iter.index) {
1712 if (++ret == nr_pages)
1718 EXPORT_SYMBOL(find_get_pages_contig);
1721 * find_get_pages_tag - find and return pages that match @tag
1722 * @mapping: the address_space to search
1723 * @index: the starting page index
1724 * @tag: the tag index
1725 * @nr_pages: the maximum number of pages
1726 * @pages: where the resulting pages are placed
1728 * Like find_get_pages, except we only return pages which are tagged with
1729 * @tag. We update @index to index the next page for the traversal.
1731 unsigned find_get_pages_tag(struct address_space *mapping, pgoff_t *index,
1732 int tag, unsigned int nr_pages, struct page **pages)
1734 struct radix_tree_iter iter;
1738 if (unlikely(!nr_pages))
1742 radix_tree_for_each_tagged(slot, &mapping->page_tree,
1743 &iter, *index, tag) {
1744 struct page *head, *page;
1746 page = radix_tree_deref_slot(slot);
1747 if (unlikely(!page))
1750 if (radix_tree_exception(page)) {
1751 if (radix_tree_deref_retry(page)) {
1752 slot = radix_tree_iter_retry(&iter);
1756 * A shadow entry of a recently evicted page.
1758 * Those entries should never be tagged, but
1759 * this tree walk is lockless and the tags are
1760 * looked up in bulk, one radix tree node at a
1761 * time, so there is a sizable window for page
1762 * reclaim to evict a page we saw tagged.
1769 head = compound_head(page);
1770 if (!page_cache_get_speculative(head))
1773 /* The page was split under us? */
1774 if (compound_head(page) != head) {
1779 /* Has the page moved? */
1780 if (unlikely(page != *slot)) {
1786 if (++ret == nr_pages)
1793 *index = pages[ret - 1]->index + 1;
1797 EXPORT_SYMBOL(find_get_pages_tag);
1800 * find_get_entries_tag - find and return entries that match @tag
1801 * @mapping: the address_space to search
1802 * @start: the starting page cache index
1803 * @tag: the tag index
1804 * @nr_entries: the maximum number of entries
1805 * @entries: where the resulting entries are placed
1806 * @indices: the cache indices corresponding to the entries in @entries
1808 * Like find_get_entries, except we only return entries which are tagged with
1811 unsigned find_get_entries_tag(struct address_space *mapping, pgoff_t start,
1812 int tag, unsigned int nr_entries,
1813 struct page **entries, pgoff_t *indices)
1816 unsigned int ret = 0;
1817 struct radix_tree_iter iter;
1823 radix_tree_for_each_tagged(slot, &mapping->page_tree,
1824 &iter, start, tag) {
1825 struct page *head, *page;
1827 page = radix_tree_deref_slot(slot);
1828 if (unlikely(!page))
1830 if (radix_tree_exception(page)) {
1831 if (radix_tree_deref_retry(page)) {
1832 slot = radix_tree_iter_retry(&iter);
1837 * A shadow entry of a recently evicted page, a swap
1838 * entry from shmem/tmpfs or a DAX entry. Return it
1839 * without attempting to raise page count.
1844 head = compound_head(page);
1845 if (!page_cache_get_speculative(head))
1848 /* The page was split under us? */
1849 if (compound_head(page) != head) {
1854 /* Has the page moved? */
1855 if (unlikely(page != *slot)) {
1860 indices[ret] = iter.index;
1861 entries[ret] = page;
1862 if (++ret == nr_entries)
1868 EXPORT_SYMBOL(find_get_entries_tag);
1871 * CD/DVDs are error prone. When a medium error occurs, the driver may fail
1872 * a _large_ part of the i/o request. Imagine the worst scenario:
1874 * ---R__________________________________________B__________
1875 * ^ reading here ^ bad block(assume 4k)
1877 * read(R) => miss => readahead(R...B) => media error => frustrating retries
1878 * => failing the whole request => read(R) => read(R+1) =>
1879 * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
1880 * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
1881 * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
1883 * It is going insane. Fix it by quickly scaling down the readahead size.
1885 static void shrink_readahead_size_eio(struct file *filp,
1886 struct file_ra_state *ra)
1892 * do_generic_file_read - generic file read routine
1893 * @filp: the file to read
1894 * @ppos: current file position
1895 * @iter: data destination
1896 * @written: already copied
1898 * This is a generic file read routine, and uses the
1899 * mapping->a_ops->readpage() function for the actual low-level stuff.
1901 * This is really ugly. But the goto's actually try to clarify some
1902 * of the logic when it comes to error handling etc.
1904 static ssize_t do_generic_file_read(struct file *filp, loff_t *ppos,
1905 struct iov_iter *iter, ssize_t written)
1907 struct address_space *mapping = filp->f_mapping;
1908 struct inode *inode = mapping->host;
1909 struct file_ra_state *ra = &filp->f_ra;
1913 unsigned long offset; /* offset into pagecache page */
1914 unsigned int prev_offset;
1917 if (unlikely(*ppos >= inode->i_sb->s_maxbytes))
1919 iov_iter_truncate(iter, inode->i_sb->s_maxbytes);
1921 index = *ppos >> PAGE_SHIFT;
1922 prev_index = ra->prev_pos >> PAGE_SHIFT;
1923 prev_offset = ra->prev_pos & (PAGE_SIZE-1);
1924 last_index = (*ppos + iter->count + PAGE_SIZE-1) >> PAGE_SHIFT;
1925 offset = *ppos & ~PAGE_MASK;
1931 unsigned long nr, ret;
1935 if (fatal_signal_pending(current)) {
1940 page = find_get_page(mapping, index);
1942 page_cache_sync_readahead(mapping,
1944 index, last_index - index);
1945 page = find_get_page(mapping, index);
1946 if (unlikely(page == NULL))
1947 goto no_cached_page;
1949 if (PageReadahead(page)) {
1950 page_cache_async_readahead(mapping,
1952 index, last_index - index);
1954 if (!PageUptodate(page)) {
1956 * See comment in do_read_cache_page on why
1957 * wait_on_page_locked is used to avoid unnecessarily
1958 * serialisations and why it's safe.
1960 error = wait_on_page_locked_killable(page);
1961 if (unlikely(error))
1962 goto readpage_error;
1963 if (PageUptodate(page))
1966 if (inode->i_blkbits == PAGE_SHIFT ||
1967 !mapping->a_ops->is_partially_uptodate)
1968 goto page_not_up_to_date;
1969 /* pipes can't handle partially uptodate pages */
1970 if (unlikely(iter->type & ITER_PIPE))
1971 goto page_not_up_to_date;
1972 if (!trylock_page(page))
1973 goto page_not_up_to_date;
1974 /* Did it get truncated before we got the lock? */
1976 goto page_not_up_to_date_locked;
1977 if (!mapping->a_ops->is_partially_uptodate(page,
1978 offset, iter->count))
1979 goto page_not_up_to_date_locked;
1984 * i_size must be checked after we know the page is Uptodate.
1986 * Checking i_size after the check allows us to calculate
1987 * the correct value for "nr", which means the zero-filled
1988 * part of the page is not copied back to userspace (unless
1989 * another truncate extends the file - this is desired though).
1992 isize = i_size_read(inode);
1993 end_index = (isize - 1) >> PAGE_SHIFT;
1994 if (unlikely(!isize || index > end_index)) {
1999 /* nr is the maximum number of bytes to copy from this page */
2001 if (index == end_index) {
2002 nr = ((isize - 1) & ~PAGE_MASK) + 1;
2010 /* If users can be writing to this page using arbitrary
2011 * virtual addresses, take care about potential aliasing
2012 * before reading the page on the kernel side.
2014 if (mapping_writably_mapped(mapping))
2015 flush_dcache_page(page);
2018 * When a sequential read accesses a page several times,
2019 * only mark it as accessed the first time.
2021 if (prev_index != index || offset != prev_offset)
2022 mark_page_accessed(page);
2026 * Ok, we have the page, and it's up-to-date, so
2027 * now we can copy it to user space...
2030 ret = copy_page_to_iter(page, offset, nr, iter);
2032 index += offset >> PAGE_SHIFT;
2033 offset &= ~PAGE_MASK;
2034 prev_offset = offset;
2038 if (!iov_iter_count(iter))
2046 page_not_up_to_date:
2047 /* Get exclusive access to the page ... */
2048 error = lock_page_killable(page);
2049 if (unlikely(error))
2050 goto readpage_error;
2052 page_not_up_to_date_locked:
2053 /* Did it get truncated before we got the lock? */
2054 if (!page->mapping) {
2060 /* Did somebody else fill it already? */
2061 if (PageUptodate(page)) {
2068 * A previous I/O error may have been due to temporary
2069 * failures, eg. multipath errors.
2070 * PG_error will be set again if readpage fails.
2072 ClearPageError(page);
2073 /* Start the actual read. The read will unlock the page. */
2074 error = mapping->a_ops->readpage(filp, page);
2076 if (unlikely(error)) {
2077 if (error == AOP_TRUNCATED_PAGE) {
2082 goto readpage_error;
2085 if (!PageUptodate(page)) {
2086 error = lock_page_killable(page);
2087 if (unlikely(error))
2088 goto readpage_error;
2089 if (!PageUptodate(page)) {
2090 if (page->mapping == NULL) {
2092 * invalidate_mapping_pages got it
2099 shrink_readahead_size_eio(filp, ra);
2101 goto readpage_error;
2109 /* UHHUH! A synchronous read error occurred. Report it */
2115 * Ok, it wasn't cached, so we need to create a new
2118 page = page_cache_alloc_cold(mapping);
2123 error = add_to_page_cache_lru(page, mapping, index,
2124 mapping_gfp_constraint(mapping, GFP_KERNEL));
2127 if (error == -EEXIST) {
2137 ra->prev_pos = prev_index;
2138 ra->prev_pos <<= PAGE_SHIFT;
2139 ra->prev_pos |= prev_offset;
2141 *ppos = ((loff_t)index << PAGE_SHIFT) + offset;
2142 file_accessed(filp);
2143 return written ? written : error;
2147 * generic_file_read_iter - generic filesystem read routine
2148 * @iocb: kernel I/O control block
2149 * @iter: destination for the data read
2151 * This is the "read_iter()" routine for all filesystems
2152 * that can use the page cache directly.
2155 generic_file_read_iter(struct kiocb *iocb, struct iov_iter *iter)
2157 struct file *file = iocb->ki_filp;
2159 size_t count = iov_iter_count(iter);
2162 goto out; /* skip atime */
2164 if (iocb->ki_flags & IOCB_DIRECT) {
2165 struct address_space *mapping = file->f_mapping;
2166 struct inode *inode = mapping->host;
2169 size = i_size_read(inode);
2170 if (iocb->ki_flags & IOCB_NOWAIT) {
2171 if (filemap_range_has_page(mapping, iocb->ki_pos,
2172 iocb->ki_pos + count - 1))
2175 retval = filemap_write_and_wait_range(mapping,
2177 iocb->ki_pos + count - 1);
2182 file_accessed(file);
2184 retval = mapping->a_ops->direct_IO(iocb, iter);
2186 iocb->ki_pos += retval;
2189 iov_iter_revert(iter, count - iov_iter_count(iter));
2192 * Btrfs can have a short DIO read if we encounter
2193 * compressed extents, so if there was an error, or if
2194 * we've already read everything we wanted to, or if
2195 * there was a short read because we hit EOF, go ahead
2196 * and return. Otherwise fallthrough to buffered io for
2197 * the rest of the read. Buffered reads will not work for
2198 * DAX files, so don't bother trying.
2200 if (retval < 0 || !count || iocb->ki_pos >= size ||
2205 retval = do_generic_file_read(file, &iocb->ki_pos, iter, retval);
2209 EXPORT_SYMBOL(generic_file_read_iter);
2213 * page_cache_read - adds requested page to the page cache if not already there
2214 * @file: file to read
2215 * @offset: page index
2216 * @gfp_mask: memory allocation flags
2218 * This adds the requested page to the page cache if it isn't already there,
2219 * and schedules an I/O to read in its contents from disk.
2221 static int page_cache_read(struct file *file, pgoff_t offset, gfp_t gfp_mask)
2223 struct address_space *mapping = file->f_mapping;
2228 page = __page_cache_alloc(gfp_mask|__GFP_COLD);
2232 ret = add_to_page_cache_lru(page, mapping, offset, gfp_mask & GFP_KERNEL);
2234 ret = mapping->a_ops->readpage(file, page);
2235 else if (ret == -EEXIST)
2236 ret = 0; /* losing race to add is OK */
2240 } while (ret == AOP_TRUNCATED_PAGE);
2245 #define MMAP_LOTSAMISS (100)
2248 * Synchronous readahead happens when we don't even find
2249 * a page in the page cache at all.
2251 static void do_sync_mmap_readahead(struct vm_area_struct *vma,
2252 struct file_ra_state *ra,
2256 struct address_space *mapping = file->f_mapping;
2258 /* If we don't want any read-ahead, don't bother */
2259 if (vma->vm_flags & VM_RAND_READ)
2264 if (vma->vm_flags & VM_SEQ_READ) {
2265 page_cache_sync_readahead(mapping, ra, file, offset,
2270 /* Avoid banging the cache line if not needed */
2271 if (ra->mmap_miss < MMAP_LOTSAMISS * 10)
2275 * Do we miss much more than hit in this file? If so,
2276 * stop bothering with read-ahead. It will only hurt.
2278 if (ra->mmap_miss > MMAP_LOTSAMISS)
2284 ra->start = max_t(long, 0, offset - ra->ra_pages / 2);
2285 ra->size = ra->ra_pages;
2286 ra->async_size = ra->ra_pages / 4;
2287 ra_submit(ra, mapping, file);
2291 * Asynchronous readahead happens when we find the page and PG_readahead,
2292 * so we want to possibly extend the readahead further..
2294 static void do_async_mmap_readahead(struct vm_area_struct *vma,
2295 struct file_ra_state *ra,
2300 struct address_space *mapping = file->f_mapping;
2302 /* If we don't want any read-ahead, don't bother */
2303 if (vma->vm_flags & VM_RAND_READ)
2305 if (ra->mmap_miss > 0)
2307 if (PageReadahead(page))
2308 page_cache_async_readahead(mapping, ra, file,
2309 page, offset, ra->ra_pages);
2313 * filemap_fault - read in file data for page fault handling
2314 * @vmf: struct vm_fault containing details of the fault
2316 * filemap_fault() is invoked via the vma operations vector for a
2317 * mapped memory region to read in file data during a page fault.
2319 * The goto's are kind of ugly, but this streamlines the normal case of having
2320 * it in the page cache, and handles the special cases reasonably without
2321 * having a lot of duplicated code.
2323 * vma->vm_mm->mmap_sem must be held on entry.
2325 * If our return value has VM_FAULT_RETRY set, it's because
2326 * lock_page_or_retry() returned 0.
2327 * The mmap_sem has usually been released in this case.
2328 * See __lock_page_or_retry() for the exception.
2330 * If our return value does not have VM_FAULT_RETRY set, the mmap_sem
2331 * has not been released.
2333 * We never return with VM_FAULT_RETRY and a bit from VM_FAULT_ERROR set.
2335 int filemap_fault(struct vm_fault *vmf)
2338 struct file *file = vmf->vma->vm_file;
2339 struct address_space *mapping = file->f_mapping;
2340 struct file_ra_state *ra = &file->f_ra;
2341 struct inode *inode = mapping->host;
2342 pgoff_t offset = vmf->pgoff;
2347 max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
2348 if (unlikely(offset >= max_off))
2349 return VM_FAULT_SIGBUS;
2352 * Do we have something in the page cache already?
2354 page = find_get_page(mapping, offset);
2355 if (likely(page) && !(vmf->flags & FAULT_FLAG_TRIED)) {
2357 * We found the page, so try async readahead before
2358 * waiting for the lock.
2360 do_async_mmap_readahead(vmf->vma, ra, file, page, offset);
2362 /* No page in the page cache at all */
2363 do_sync_mmap_readahead(vmf->vma, ra, file, offset);
2364 count_vm_event(PGMAJFAULT);
2365 count_memcg_event_mm(vmf->vma->vm_mm, PGMAJFAULT);
2366 ret = VM_FAULT_MAJOR;
2368 page = find_get_page(mapping, offset);
2370 goto no_cached_page;
2373 if (!lock_page_or_retry(page, vmf->vma->vm_mm, vmf->flags)) {
2375 return ret | VM_FAULT_RETRY;
2378 /* Did it get truncated? */
2379 if (unlikely(page->mapping != mapping)) {
2384 VM_BUG_ON_PAGE(page->index != offset, page);
2387 * We have a locked page in the page cache, now we need to check
2388 * that it's up-to-date. If not, it is going to be due to an error.
2390 if (unlikely(!PageUptodate(page)))
2391 goto page_not_uptodate;
2394 * Found the page and have a reference on it.
2395 * We must recheck i_size under page lock.
2397 max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
2398 if (unlikely(offset >= max_off)) {
2401 return VM_FAULT_SIGBUS;
2405 return ret | VM_FAULT_LOCKED;
2409 * We're only likely to ever get here if MADV_RANDOM is in
2412 error = page_cache_read(file, offset, vmf->gfp_mask);
2415 * The page we want has now been added to the page cache.
2416 * In the unlikely event that someone removed it in the
2417 * meantime, we'll just come back here and read it again.
2423 * An error return from page_cache_read can result if the
2424 * system is low on memory, or a problem occurs while trying
2427 if (error == -ENOMEM)
2428 return VM_FAULT_OOM;
2429 return VM_FAULT_SIGBUS;
2433 * Umm, take care of errors if the page isn't up-to-date.
2434 * Try to re-read it _once_. We do this synchronously,
2435 * because there really aren't any performance issues here
2436 * and we need to check for errors.
2438 ClearPageError(page);
2439 error = mapping->a_ops->readpage(file, page);
2441 wait_on_page_locked(page);
2442 if (!PageUptodate(page))
2447 if (!error || error == AOP_TRUNCATED_PAGE)
2450 /* Things didn't work out. Return zero to tell the mm layer so. */
2451 shrink_readahead_size_eio(file, ra);
2452 return VM_FAULT_SIGBUS;
2454 EXPORT_SYMBOL(filemap_fault);
2456 void filemap_map_pages(struct vm_fault *vmf,
2457 pgoff_t start_pgoff, pgoff_t end_pgoff)
2459 struct radix_tree_iter iter;
2461 struct file *file = vmf->vma->vm_file;
2462 struct address_space *mapping = file->f_mapping;
2463 pgoff_t last_pgoff = start_pgoff;
2464 unsigned long max_idx;
2465 struct page *head, *page;
2468 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter,
2470 if (iter.index > end_pgoff)
2473 page = radix_tree_deref_slot(slot);
2474 if (unlikely(!page))
2476 if (radix_tree_exception(page)) {
2477 if (radix_tree_deref_retry(page)) {
2478 slot = radix_tree_iter_retry(&iter);
2484 head = compound_head(page);
2485 if (!page_cache_get_speculative(head))
2488 /* The page was split under us? */
2489 if (compound_head(page) != head) {
2494 /* Has the page moved? */
2495 if (unlikely(page != *slot)) {
2500 if (!PageUptodate(page) ||
2501 PageReadahead(page) ||
2504 if (!trylock_page(page))
2507 if (page->mapping != mapping || !PageUptodate(page))
2510 max_idx = DIV_ROUND_UP(i_size_read(mapping->host), PAGE_SIZE);
2511 if (page->index >= max_idx)
2514 if (file->f_ra.mmap_miss > 0)
2515 file->f_ra.mmap_miss--;
2517 vmf->address += (iter.index - last_pgoff) << PAGE_SHIFT;
2519 vmf->pte += iter.index - last_pgoff;
2520 last_pgoff = iter.index;
2521 if (alloc_set_pte(vmf, NULL, page))
2530 /* Huge page is mapped? No need to proceed. */
2531 if (pmd_trans_huge(*vmf->pmd))
2533 if (iter.index == end_pgoff)
2538 EXPORT_SYMBOL(filemap_map_pages);
2540 int filemap_page_mkwrite(struct vm_fault *vmf)
2542 struct page *page = vmf->page;
2543 struct inode *inode = file_inode(vmf->vma->vm_file);
2544 int ret = VM_FAULT_LOCKED;
2546 sb_start_pagefault(inode->i_sb);
2547 file_update_time(vmf->vma->vm_file);
2549 if (page->mapping != inode->i_mapping) {
2551 ret = VM_FAULT_NOPAGE;
2555 * We mark the page dirty already here so that when freeze is in
2556 * progress, we are guaranteed that writeback during freezing will
2557 * see the dirty page and writeprotect it again.
2559 set_page_dirty(page);
2560 wait_for_stable_page(page);
2562 sb_end_pagefault(inode->i_sb);
2565 EXPORT_SYMBOL(filemap_page_mkwrite);
2567 const struct vm_operations_struct generic_file_vm_ops = {
2568 .fault = filemap_fault,
2569 .map_pages = filemap_map_pages,
2570 .page_mkwrite = filemap_page_mkwrite,
2573 /* This is used for a general mmap of a disk file */
2575 int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
2577 struct address_space *mapping = file->f_mapping;
2579 if (!mapping->a_ops->readpage)
2581 file_accessed(file);
2582 vma->vm_ops = &generic_file_vm_ops;
2587 * This is for filesystems which do not implement ->writepage.
2589 int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
2591 if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE))
2593 return generic_file_mmap(file, vma);
2596 int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
2600 int generic_file_readonly_mmap(struct file * file, struct vm_area_struct * vma)
2604 #endif /* CONFIG_MMU */
2606 EXPORT_SYMBOL(generic_file_mmap);
2607 EXPORT_SYMBOL(generic_file_readonly_mmap);
2609 static struct page *wait_on_page_read(struct page *page)
2611 if (!IS_ERR(page)) {
2612 wait_on_page_locked(page);
2613 if (!PageUptodate(page)) {
2615 page = ERR_PTR(-EIO);
2621 static struct page *do_read_cache_page(struct address_space *mapping,
2623 int (*filler)(void *, struct page *),
2630 page = find_get_page(mapping, index);
2632 page = __page_cache_alloc(gfp | __GFP_COLD);
2634 return ERR_PTR(-ENOMEM);
2635 err = add_to_page_cache_lru(page, mapping, index, gfp);
2636 if (unlikely(err)) {
2640 /* Presumably ENOMEM for radix tree node */
2641 return ERR_PTR(err);
2645 err = filler(data, page);
2648 return ERR_PTR(err);
2651 page = wait_on_page_read(page);
2656 if (PageUptodate(page))
2660 * Page is not up to date and may be locked due one of the following
2661 * case a: Page is being filled and the page lock is held
2662 * case b: Read/write error clearing the page uptodate status
2663 * case c: Truncation in progress (page locked)
2664 * case d: Reclaim in progress
2666 * Case a, the page will be up to date when the page is unlocked.
2667 * There is no need to serialise on the page lock here as the page
2668 * is pinned so the lock gives no additional protection. Even if the
2669 * the page is truncated, the data is still valid if PageUptodate as
2670 * it's a race vs truncate race.
2671 * Case b, the page will not be up to date
2672 * Case c, the page may be truncated but in itself, the data may still
2673 * be valid after IO completes as it's a read vs truncate race. The
2674 * operation must restart if the page is not uptodate on unlock but
2675 * otherwise serialising on page lock to stabilise the mapping gives
2676 * no additional guarantees to the caller as the page lock is
2677 * released before return.
2678 * Case d, similar to truncation. If reclaim holds the page lock, it
2679 * will be a race with remove_mapping that determines if the mapping
2680 * is valid on unlock but otherwise the data is valid and there is
2681 * no need to serialise with page lock.
2683 * As the page lock gives no additional guarantee, we optimistically
2684 * wait on the page to be unlocked and check if it's up to date and
2685 * use the page if it is. Otherwise, the page lock is required to
2686 * distinguish between the different cases. The motivation is that we
2687 * avoid spurious serialisations and wakeups when multiple processes
2688 * wait on the same page for IO to complete.
2690 wait_on_page_locked(page);
2691 if (PageUptodate(page))
2694 /* Distinguish between all the cases under the safety of the lock */
2697 /* Case c or d, restart the operation */
2698 if (!page->mapping) {
2704 /* Someone else locked and filled the page in a very small window */
2705 if (PageUptodate(page)) {
2712 mark_page_accessed(page);
2717 * read_cache_page - read into page cache, fill it if needed
2718 * @mapping: the page's address_space
2719 * @index: the page index
2720 * @filler: function to perform the read
2721 * @data: first arg to filler(data, page) function, often left as NULL
2723 * Read into the page cache. If a page already exists, and PageUptodate() is
2724 * not set, try to fill the page and wait for it to become unlocked.
2726 * If the page does not get brought uptodate, return -EIO.
2728 struct page *read_cache_page(struct address_space *mapping,
2730 int (*filler)(void *, struct page *),
2733 return do_read_cache_page(mapping, index, filler, data, mapping_gfp_mask(mapping));
2735 EXPORT_SYMBOL(read_cache_page);
2738 * read_cache_page_gfp - read into page cache, using specified page allocation flags.
2739 * @mapping: the page's address_space
2740 * @index: the page index
2741 * @gfp: the page allocator flags to use if allocating
2743 * This is the same as "read_mapping_page(mapping, index, NULL)", but with
2744 * any new page allocations done using the specified allocation flags.
2746 * If the page does not get brought uptodate, return -EIO.
2748 struct page *read_cache_page_gfp(struct address_space *mapping,
2752 filler_t *filler = (filler_t *)mapping->a_ops->readpage;
2754 return do_read_cache_page(mapping, index, filler, NULL, gfp);
2756 EXPORT_SYMBOL(read_cache_page_gfp);
2759 * Performs necessary checks before doing a write
2761 * Can adjust writing position or amount of bytes to write.
2762 * Returns appropriate error code that caller should return or
2763 * zero in case that write should be allowed.
2765 inline ssize_t generic_write_checks(struct kiocb *iocb, struct iov_iter *from)
2767 struct file *file = iocb->ki_filp;
2768 struct inode *inode = file->f_mapping->host;
2769 unsigned long limit = rlimit(RLIMIT_FSIZE);
2772 if (!iov_iter_count(from))
2775 /* FIXME: this is for backwards compatibility with 2.4 */
2776 if (iocb->ki_flags & IOCB_APPEND)
2777 iocb->ki_pos = i_size_read(inode);
2781 if ((iocb->ki_flags & IOCB_NOWAIT) && !(iocb->ki_flags & IOCB_DIRECT))
2784 if (limit != RLIM_INFINITY) {
2785 if (iocb->ki_pos >= limit) {
2786 send_sig(SIGXFSZ, current, 0);
2789 iov_iter_truncate(from, limit - (unsigned long)pos);
2795 if (unlikely(pos + iov_iter_count(from) > MAX_NON_LFS &&
2796 !(file->f_flags & O_LARGEFILE))) {
2797 if (pos >= MAX_NON_LFS)
2799 iov_iter_truncate(from, MAX_NON_LFS - (unsigned long)pos);
2803 * Are we about to exceed the fs block limit ?
2805 * If we have written data it becomes a short write. If we have
2806 * exceeded without writing data we send a signal and return EFBIG.
2807 * Linus frestrict idea will clean these up nicely..
2809 if (unlikely(pos >= inode->i_sb->s_maxbytes))
2812 iov_iter_truncate(from, inode->i_sb->s_maxbytes - pos);
2813 return iov_iter_count(from);
2815 EXPORT_SYMBOL(generic_write_checks);
2817 int pagecache_write_begin(struct file *file, struct address_space *mapping,
2818 loff_t pos, unsigned len, unsigned flags,
2819 struct page **pagep, void **fsdata)
2821 const struct address_space_operations *aops = mapping->a_ops;
2823 return aops->write_begin(file, mapping, pos, len, flags,
2826 EXPORT_SYMBOL(pagecache_write_begin);
2828 int pagecache_write_end(struct file *file, struct address_space *mapping,
2829 loff_t pos, unsigned len, unsigned copied,
2830 struct page *page, void *fsdata)
2832 const struct address_space_operations *aops = mapping->a_ops;
2834 return aops->write_end(file, mapping, pos, len, copied, page, fsdata);
2836 EXPORT_SYMBOL(pagecache_write_end);
2839 generic_file_direct_write(struct kiocb *iocb, struct iov_iter *from)
2841 struct file *file = iocb->ki_filp;
2842 struct address_space *mapping = file->f_mapping;
2843 struct inode *inode = mapping->host;
2844 loff_t pos = iocb->ki_pos;
2849 write_len = iov_iter_count(from);
2850 end = (pos + write_len - 1) >> PAGE_SHIFT;
2852 if (iocb->ki_flags & IOCB_NOWAIT) {
2853 /* If there are pages to writeback, return */
2854 if (filemap_range_has_page(inode->i_mapping, pos,
2855 pos + iov_iter_count(from)))
2858 written = filemap_write_and_wait_range(mapping, pos,
2859 pos + write_len - 1);
2865 * After a write we want buffered reads to be sure to go to disk to get
2866 * the new data. We invalidate clean cached page from the region we're
2867 * about to write. We do this *before* the write so that we can return
2868 * without clobbering -EIOCBQUEUED from ->direct_IO().
2870 written = invalidate_inode_pages2_range(mapping,
2871 pos >> PAGE_SHIFT, end);
2873 * If a page can not be invalidated, return 0 to fall back
2874 * to buffered write.
2877 if (written == -EBUSY)
2882 written = mapping->a_ops->direct_IO(iocb, from);
2885 * Finally, try again to invalidate clean pages which might have been
2886 * cached by non-direct readahead, or faulted in by get_user_pages()
2887 * if the source of the write was an mmap'ed region of the file
2888 * we're writing. Either one is a pretty crazy thing to do,
2889 * so we don't support it 100%. If this invalidation
2890 * fails, tough, the write still worked...
2892 invalidate_inode_pages2_range(mapping,
2893 pos >> PAGE_SHIFT, end);
2897 write_len -= written;
2898 if (pos > i_size_read(inode) && !S_ISBLK(inode->i_mode)) {
2899 i_size_write(inode, pos);
2900 mark_inode_dirty(inode);
2904 iov_iter_revert(from, write_len - iov_iter_count(from));
2908 EXPORT_SYMBOL(generic_file_direct_write);
2911 * Find or create a page at the given pagecache position. Return the locked
2912 * page. This function is specifically for buffered writes.
2914 struct page *grab_cache_page_write_begin(struct address_space *mapping,
2915 pgoff_t index, unsigned flags)
2918 int fgp_flags = FGP_LOCK|FGP_WRITE|FGP_CREAT;
2920 if (flags & AOP_FLAG_NOFS)
2921 fgp_flags |= FGP_NOFS;
2923 page = pagecache_get_page(mapping, index, fgp_flags,
2924 mapping_gfp_mask(mapping));
2926 wait_for_stable_page(page);
2930 EXPORT_SYMBOL(grab_cache_page_write_begin);
2932 ssize_t generic_perform_write(struct file *file,
2933 struct iov_iter *i, loff_t pos)
2935 struct address_space *mapping = file->f_mapping;
2936 const struct address_space_operations *a_ops = mapping->a_ops;
2938 ssize_t written = 0;
2939 unsigned int flags = 0;
2943 unsigned long offset; /* Offset into pagecache page */
2944 unsigned long bytes; /* Bytes to write to page */
2945 size_t copied; /* Bytes copied from user */
2948 offset = (pos & (PAGE_SIZE - 1));
2949 bytes = min_t(unsigned long, PAGE_SIZE - offset,
2954 * Bring in the user page that we will copy from _first_.
2955 * Otherwise there's a nasty deadlock on copying from the
2956 * same page as we're writing to, without it being marked
2959 * Not only is this an optimisation, but it is also required
2960 * to check that the address is actually valid, when atomic
2961 * usercopies are used, below.
2963 if (unlikely(iov_iter_fault_in_readable(i, bytes))) {
2968 if (fatal_signal_pending(current)) {
2973 status = a_ops->write_begin(file, mapping, pos, bytes, flags,
2975 if (unlikely(status < 0))
2978 if (mapping_writably_mapped(mapping))
2979 flush_dcache_page(page);
2981 copied = iov_iter_copy_from_user_atomic(page, i, offset, bytes);
2982 flush_dcache_page(page);
2984 status = a_ops->write_end(file, mapping, pos, bytes, copied,
2986 if (unlikely(status < 0))
2992 iov_iter_advance(i, copied);
2993 if (unlikely(copied == 0)) {
2995 * If we were unable to copy any data at all, we must
2996 * fall back to a single segment length write.
2998 * If we didn't fallback here, we could livelock
2999 * because not all segments in the iov can be copied at
3000 * once without a pagefault.
3002 bytes = min_t(unsigned long, PAGE_SIZE - offset,
3003 iov_iter_single_seg_count(i));
3009 balance_dirty_pages_ratelimited(mapping);
3010 } while (iov_iter_count(i));
3012 return written ? written : status;
3014 EXPORT_SYMBOL(generic_perform_write);
3017 * __generic_file_write_iter - write data to a file
3018 * @iocb: IO state structure (file, offset, etc.)
3019 * @from: iov_iter with data to write
3021 * This function does all the work needed for actually writing data to a
3022 * file. It does all basic checks, removes SUID from the file, updates
3023 * modification times and calls proper subroutines depending on whether we
3024 * do direct IO or a standard buffered write.
3026 * It expects i_mutex to be grabbed unless we work on a block device or similar
3027 * object which does not need locking at all.
3029 * This function does *not* take care of syncing data in case of O_SYNC write.
3030 * A caller has to handle it. This is mainly due to the fact that we want to
3031 * avoid syncing under i_mutex.
3033 ssize_t __generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
3035 struct file *file = iocb->ki_filp;
3036 struct address_space * mapping = file->f_mapping;
3037 struct inode *inode = mapping->host;
3038 ssize_t written = 0;
3042 /* We can write back this queue in page reclaim */
3043 current->backing_dev_info = inode_to_bdi(inode);
3044 err = file_remove_privs(file);
3048 err = file_update_time(file);
3052 if (iocb->ki_flags & IOCB_DIRECT) {
3053 loff_t pos, endbyte;
3055 written = generic_file_direct_write(iocb, from);
3057 * If the write stopped short of completing, fall back to
3058 * buffered writes. Some filesystems do this for writes to
3059 * holes, for example. For DAX files, a buffered write will
3060 * not succeed (even if it did, DAX does not handle dirty
3061 * page-cache pages correctly).
3063 if (written < 0 || !iov_iter_count(from) || IS_DAX(inode))
3066 status = generic_perform_write(file, from, pos = iocb->ki_pos);
3068 * If generic_perform_write() returned a synchronous error
3069 * then we want to return the number of bytes which were
3070 * direct-written, or the error code if that was zero. Note
3071 * that this differs from normal direct-io semantics, which
3072 * will return -EFOO even if some bytes were written.
3074 if (unlikely(status < 0)) {
3079 * We need to ensure that the page cache pages are written to
3080 * disk and invalidated to preserve the expected O_DIRECT
3083 endbyte = pos + status - 1;
3084 err = filemap_write_and_wait_range(mapping, pos, endbyte);
3086 iocb->ki_pos = endbyte + 1;
3088 invalidate_mapping_pages(mapping,
3090 endbyte >> PAGE_SHIFT);
3093 * We don't know how much we wrote, so just return
3094 * the number of bytes which were direct-written
3098 written = generic_perform_write(file, from, iocb->ki_pos);
3099 if (likely(written > 0))
3100 iocb->ki_pos += written;
3103 current->backing_dev_info = NULL;
3104 return written ? written : err;
3106 EXPORT_SYMBOL(__generic_file_write_iter);
3109 * generic_file_write_iter - write data to a file
3110 * @iocb: IO state structure
3111 * @from: iov_iter with data to write
3113 * This is a wrapper around __generic_file_write_iter() to be used by most
3114 * filesystems. It takes care of syncing the file in case of O_SYNC file
3115 * and acquires i_mutex as needed.
3117 ssize_t generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
3119 struct file *file = iocb->ki_filp;
3120 struct inode *inode = file->f_mapping->host;
3124 ret = generic_write_checks(iocb, from);
3126 ret = __generic_file_write_iter(iocb, from);
3127 inode_unlock(inode);
3130 ret = generic_write_sync(iocb, ret);
3133 EXPORT_SYMBOL(generic_file_write_iter);
3136 * try_to_release_page() - release old fs-specific metadata on a page
3138 * @page: the page which the kernel is trying to free
3139 * @gfp_mask: memory allocation flags (and I/O mode)
3141 * The address_space is to try to release any data against the page
3142 * (presumably at page->private). If the release was successful, return '1'.
3143 * Otherwise return zero.
3145 * This may also be called if PG_fscache is set on a page, indicating that the
3146 * page is known to the local caching routines.
3148 * The @gfp_mask argument specifies whether I/O may be performed to release
3149 * this page (__GFP_IO), and whether the call may block (__GFP_RECLAIM & __GFP_FS).
3152 int try_to_release_page(struct page *page, gfp_t gfp_mask)
3154 struct address_space * const mapping = page->mapping;
3156 BUG_ON(!PageLocked(page));
3157 if (PageWriteback(page))
3160 if (mapping && mapping->a_ops->releasepage)
3161 return mapping->a_ops->releasepage(page, gfp_mask);
3162 return try_to_free_buffers(page);
3165 EXPORT_SYMBOL(try_to_release_page);