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--;
136 __radix_tree_replace(&mapping->page_tree, node, slot, page,
137 workingset_update_node, mapping);
142 static void page_cache_tree_delete(struct address_space *mapping,
143 struct page *page, void *shadow)
147 /* hugetlb pages are represented by one entry in the radix tree */
148 nr = PageHuge(page) ? 1 : hpage_nr_pages(page);
150 VM_BUG_ON_PAGE(!PageLocked(page), page);
151 VM_BUG_ON_PAGE(PageTail(page), page);
152 VM_BUG_ON_PAGE(nr != 1 && shadow, page);
154 for (i = 0; i < nr; i++) {
155 struct radix_tree_node *node;
158 __radix_tree_lookup(&mapping->page_tree, page->index + i,
161 VM_BUG_ON_PAGE(!node && nr != 1, page);
163 radix_tree_clear_tags(&mapping->page_tree, node, slot);
164 __radix_tree_replace(&mapping->page_tree, node, slot, shadow,
165 workingset_update_node, mapping);
169 mapping->nrexceptional += nr;
171 * Make sure the nrexceptional update is committed before
172 * the nrpages update so that final truncate racing
173 * with reclaim does not see both counters 0 at the
174 * same time and miss a shadow entry.
178 mapping->nrpages -= nr;
182 * Delete a page from the page cache and free it. Caller has to make
183 * sure the page is locked and that nobody else uses it - or that usage
184 * is safe. The caller must hold the mapping's tree_lock.
186 void __delete_from_page_cache(struct page *page, void *shadow)
188 struct address_space *mapping = page->mapping;
189 int nr = hpage_nr_pages(page);
191 trace_mm_filemap_delete_from_page_cache(page);
193 * if we're uptodate, flush out into the cleancache, otherwise
194 * invalidate any existing cleancache entries. We can't leave
195 * stale data around in the cleancache once our page is gone
197 if (PageUptodate(page) && PageMappedToDisk(page))
198 cleancache_put_page(page);
200 cleancache_invalidate_page(mapping, page);
202 VM_BUG_ON_PAGE(PageTail(page), page);
203 VM_BUG_ON_PAGE(page_mapped(page), page);
204 if (!IS_ENABLED(CONFIG_DEBUG_VM) && unlikely(page_mapped(page))) {
207 pr_alert("BUG: Bad page cache in process %s pfn:%05lx\n",
208 current->comm, page_to_pfn(page));
209 dump_page(page, "still mapped when deleted");
211 add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE);
213 mapcount = page_mapcount(page);
214 if (mapping_exiting(mapping) &&
215 page_count(page) >= mapcount + 2) {
217 * All vmas have already been torn down, so it's
218 * a good bet that actually the page is unmapped,
219 * and we'd prefer not to leak it: if we're wrong,
220 * some other bad page check should catch it later.
222 page_mapcount_reset(page);
223 page_ref_sub(page, mapcount);
227 page_cache_tree_delete(mapping, page, shadow);
229 page->mapping = NULL;
230 /* Leave page->index set: truncation lookup relies upon it */
232 /* hugetlb pages do not participate in page cache accounting. */
236 __mod_node_page_state(page_pgdat(page), NR_FILE_PAGES, -nr);
237 if (PageSwapBacked(page)) {
238 __mod_node_page_state(page_pgdat(page), NR_SHMEM, -nr);
239 if (PageTransHuge(page))
240 __dec_node_page_state(page, NR_SHMEM_THPS);
242 VM_BUG_ON_PAGE(PageTransHuge(page), page);
246 * At this point page must be either written or cleaned by truncate.
247 * Dirty page here signals a bug and loss of unwritten data.
249 * This fixes dirty accounting after removing the page entirely but
250 * leaves PageDirty set: it has no effect for truncated page and
251 * anyway will be cleared before returning page into buddy allocator.
253 if (WARN_ON_ONCE(PageDirty(page)))
254 account_page_cleaned(page, mapping, inode_to_wb(mapping->host));
258 * delete_from_page_cache - delete page from page cache
259 * @page: the page which the kernel is trying to remove from page cache
261 * This must be called only on pages that have been verified to be in the page
262 * cache and locked. It will never put the page into the free list, the caller
263 * has a reference on the page.
265 void delete_from_page_cache(struct page *page)
267 struct address_space *mapping = page_mapping(page);
269 void (*freepage)(struct page *);
271 BUG_ON(!PageLocked(page));
273 freepage = mapping->a_ops->freepage;
275 spin_lock_irqsave(&mapping->tree_lock, flags);
276 __delete_from_page_cache(page, NULL);
277 spin_unlock_irqrestore(&mapping->tree_lock, flags);
282 if (PageTransHuge(page) && !PageHuge(page)) {
283 page_ref_sub(page, HPAGE_PMD_NR);
284 VM_BUG_ON_PAGE(page_count(page) <= 0, page);
289 EXPORT_SYMBOL(delete_from_page_cache);
291 int filemap_check_errors(struct address_space *mapping)
294 /* Check for outstanding write errors */
295 if (test_bit(AS_ENOSPC, &mapping->flags) &&
296 test_and_clear_bit(AS_ENOSPC, &mapping->flags))
298 if (test_bit(AS_EIO, &mapping->flags) &&
299 test_and_clear_bit(AS_EIO, &mapping->flags))
303 EXPORT_SYMBOL(filemap_check_errors);
305 static int filemap_check_and_keep_errors(struct address_space *mapping)
307 /* Check for outstanding write errors */
308 if (test_bit(AS_EIO, &mapping->flags))
310 if (test_bit(AS_ENOSPC, &mapping->flags))
316 * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range
317 * @mapping: address space structure to write
318 * @start: offset in bytes where the range starts
319 * @end: offset in bytes where the range ends (inclusive)
320 * @sync_mode: enable synchronous operation
322 * Start writeback against all of a mapping's dirty pages that lie
323 * within the byte offsets <start, end> inclusive.
325 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
326 * opposed to a regular memory cleansing writeback. The difference between
327 * these two operations is that if a dirty page/buffer is encountered, it must
328 * be waited upon, and not just skipped over.
330 int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
331 loff_t end, int sync_mode)
334 struct writeback_control wbc = {
335 .sync_mode = sync_mode,
336 .nr_to_write = LONG_MAX,
337 .range_start = start,
341 if (!mapping_cap_writeback_dirty(mapping))
344 wbc_attach_fdatawrite_inode(&wbc, mapping->host);
345 ret = do_writepages(mapping, &wbc);
346 wbc_detach_inode(&wbc);
350 static inline int __filemap_fdatawrite(struct address_space *mapping,
353 return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode);
356 int filemap_fdatawrite(struct address_space *mapping)
358 return __filemap_fdatawrite(mapping, WB_SYNC_ALL);
360 EXPORT_SYMBOL(filemap_fdatawrite);
362 int filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
365 return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL);
367 EXPORT_SYMBOL(filemap_fdatawrite_range);
370 * filemap_flush - mostly a non-blocking flush
371 * @mapping: target address_space
373 * This is a mostly non-blocking flush. Not suitable for data-integrity
374 * purposes - I/O may not be started against all dirty pages.
376 int filemap_flush(struct address_space *mapping)
378 return __filemap_fdatawrite(mapping, WB_SYNC_NONE);
380 EXPORT_SYMBOL(filemap_flush);
383 * filemap_range_has_page - check if a page exists in range.
384 * @mapping: address space within which to check
385 * @start_byte: offset in bytes where the range starts
386 * @end_byte: offset in bytes where the range ends (inclusive)
388 * Find at least one page in the range supplied, usually used to check if
389 * direct writing in this range will trigger a writeback.
391 bool filemap_range_has_page(struct address_space *mapping,
392 loff_t start_byte, loff_t end_byte)
394 pgoff_t index = start_byte >> PAGE_SHIFT;
395 pgoff_t end = end_byte >> PAGE_SHIFT;
398 if (end_byte < start_byte)
401 if (mapping->nrpages == 0)
404 if (!find_get_pages_range(mapping, &index, end, 1, &page))
409 EXPORT_SYMBOL(filemap_range_has_page);
411 static void __filemap_fdatawait_range(struct address_space *mapping,
412 loff_t start_byte, loff_t end_byte)
414 pgoff_t index = start_byte >> PAGE_SHIFT;
415 pgoff_t end = end_byte >> PAGE_SHIFT;
419 if (end_byte < start_byte)
422 pagevec_init(&pvec, 0);
423 while ((index <= end) &&
424 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
425 PAGECACHE_TAG_WRITEBACK,
426 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1)) != 0) {
429 for (i = 0; i < nr_pages; i++) {
430 struct page *page = pvec.pages[i];
432 /* until radix tree lookup accepts end_index */
433 if (page->index > end)
436 wait_on_page_writeback(page);
437 ClearPageError(page);
439 pagevec_release(&pvec);
445 * filemap_fdatawait_range - wait for writeback to complete
446 * @mapping: address space structure to wait for
447 * @start_byte: offset in bytes where the range starts
448 * @end_byte: offset in bytes where the range ends (inclusive)
450 * Walk the list of under-writeback pages of the given address space
451 * in the given range and wait for all of them. Check error status of
452 * the address space and return it.
454 * Since the error status of the address space is cleared by this function,
455 * callers are responsible for checking the return value and handling and/or
456 * reporting the error.
458 int filemap_fdatawait_range(struct address_space *mapping, loff_t start_byte,
461 __filemap_fdatawait_range(mapping, start_byte, end_byte);
462 return filemap_check_errors(mapping);
464 EXPORT_SYMBOL(filemap_fdatawait_range);
467 * file_fdatawait_range - wait for writeback to complete
468 * @file: file pointing to address space structure to wait for
469 * @start_byte: offset in bytes where the range starts
470 * @end_byte: offset in bytes where the range ends (inclusive)
472 * Walk the list of under-writeback pages of the address space that file
473 * refers to, in the given range and wait for all of them. Check error
474 * status of the address space vs. the file->f_wb_err cursor and return it.
476 * Since the error status of the file is advanced by this function,
477 * callers are responsible for checking the return value and handling and/or
478 * reporting the error.
480 int file_fdatawait_range(struct file *file, loff_t start_byte, loff_t end_byte)
482 struct address_space *mapping = file->f_mapping;
484 __filemap_fdatawait_range(mapping, start_byte, end_byte);
485 return file_check_and_advance_wb_err(file);
487 EXPORT_SYMBOL(file_fdatawait_range);
490 * filemap_fdatawait_keep_errors - wait for writeback without clearing errors
491 * @mapping: address space structure to wait for
493 * Walk the list of under-writeback pages of the given address space
494 * and wait for all of them. Unlike filemap_fdatawait(), this function
495 * does not clear error status of the address space.
497 * Use this function if callers don't handle errors themselves. Expected
498 * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2),
501 int filemap_fdatawait_keep_errors(struct address_space *mapping)
503 __filemap_fdatawait_range(mapping, 0, LLONG_MAX);
504 return filemap_check_and_keep_errors(mapping);
506 EXPORT_SYMBOL(filemap_fdatawait_keep_errors);
508 static bool mapping_needs_writeback(struct address_space *mapping)
510 return (!dax_mapping(mapping) && mapping->nrpages) ||
511 (dax_mapping(mapping) && mapping->nrexceptional);
514 int filemap_write_and_wait(struct address_space *mapping)
518 if (mapping_needs_writeback(mapping)) {
519 err = filemap_fdatawrite(mapping);
521 * Even if the above returned error, the pages may be
522 * written partially (e.g. -ENOSPC), so we wait for it.
523 * But the -EIO is special case, it may indicate the worst
524 * thing (e.g. bug) happened, so we avoid waiting for it.
527 int err2 = filemap_fdatawait(mapping);
531 /* Clear any previously stored errors */
532 filemap_check_errors(mapping);
535 err = filemap_check_errors(mapping);
539 EXPORT_SYMBOL(filemap_write_and_wait);
542 * filemap_write_and_wait_range - write out & wait on a file range
543 * @mapping: the address_space for the pages
544 * @lstart: offset in bytes where the range starts
545 * @lend: offset in bytes where the range ends (inclusive)
547 * Write out and wait upon file offsets lstart->lend, inclusive.
549 * Note that @lend is inclusive (describes the last byte to be written) so
550 * that this function can be used to write to the very end-of-file (end = -1).
552 int filemap_write_and_wait_range(struct address_space *mapping,
553 loff_t lstart, loff_t lend)
557 if (mapping_needs_writeback(mapping)) {
558 err = __filemap_fdatawrite_range(mapping, lstart, lend,
560 /* See comment of filemap_write_and_wait() */
562 int err2 = filemap_fdatawait_range(mapping,
567 /* Clear any previously stored errors */
568 filemap_check_errors(mapping);
571 err = filemap_check_errors(mapping);
575 EXPORT_SYMBOL(filemap_write_and_wait_range);
577 void __filemap_set_wb_err(struct address_space *mapping, int err)
579 errseq_t eseq = errseq_set(&mapping->wb_err, err);
581 trace_filemap_set_wb_err(mapping, eseq);
583 EXPORT_SYMBOL(__filemap_set_wb_err);
586 * file_check_and_advance_wb_err - report wb error (if any) that was previously
587 * and advance wb_err to current one
588 * @file: struct file on which the error is being reported
590 * When userland calls fsync (or something like nfsd does the equivalent), we
591 * want to report any writeback errors that occurred since the last fsync (or
592 * since the file was opened if there haven't been any).
594 * Grab the wb_err from the mapping. If it matches what we have in the file,
595 * then just quickly return 0. The file is all caught up.
597 * If it doesn't match, then take the mapping value, set the "seen" flag in
598 * it and try to swap it into place. If it works, or another task beat us
599 * to it with the new value, then update the f_wb_err and return the error
600 * portion. The error at this point must be reported via proper channels
601 * (a'la fsync, or NFS COMMIT operation, etc.).
603 * While we handle mapping->wb_err with atomic operations, the f_wb_err
604 * value is protected by the f_lock since we must ensure that it reflects
605 * the latest value swapped in for this file descriptor.
607 int file_check_and_advance_wb_err(struct file *file)
610 errseq_t old = READ_ONCE(file->f_wb_err);
611 struct address_space *mapping = file->f_mapping;
613 /* Locklessly handle the common case where nothing has changed */
614 if (errseq_check(&mapping->wb_err, old)) {
615 /* Something changed, must use slow path */
616 spin_lock(&file->f_lock);
617 old = file->f_wb_err;
618 err = errseq_check_and_advance(&mapping->wb_err,
620 trace_file_check_and_advance_wb_err(file, old);
621 spin_unlock(&file->f_lock);
625 EXPORT_SYMBOL(file_check_and_advance_wb_err);
628 * file_write_and_wait_range - write out & wait on a file range
629 * @file: file pointing to address_space with pages
630 * @lstart: offset in bytes where the range starts
631 * @lend: offset in bytes where the range ends (inclusive)
633 * Write out and wait upon file offsets lstart->lend, inclusive.
635 * Note that @lend is inclusive (describes the last byte to be written) so
636 * that this function can be used to write to the very end-of-file (end = -1).
638 * After writing out and waiting on the data, we check and advance the
639 * f_wb_err cursor to the latest value, and return any errors detected there.
641 int file_write_and_wait_range(struct file *file, loff_t lstart, loff_t lend)
644 struct address_space *mapping = file->f_mapping;
646 if (mapping_needs_writeback(mapping)) {
647 err = __filemap_fdatawrite_range(mapping, lstart, lend,
649 /* See comment of filemap_write_and_wait() */
651 __filemap_fdatawait_range(mapping, lstart, lend);
653 err2 = file_check_and_advance_wb_err(file);
658 EXPORT_SYMBOL(file_write_and_wait_range);
661 * replace_page_cache_page - replace a pagecache page with a new one
662 * @old: page to be replaced
663 * @new: page to replace with
664 * @gfp_mask: allocation mode
666 * This function replaces a page in the pagecache with a new one. On
667 * success it acquires the pagecache reference for the new page and
668 * drops it for the old page. Both the old and new pages must be
669 * locked. This function does not add the new page to the LRU, the
670 * caller must do that.
672 * The remove + add is atomic. The only way this function can fail is
673 * memory allocation failure.
675 int replace_page_cache_page(struct page *old, struct page *new, gfp_t gfp_mask)
679 VM_BUG_ON_PAGE(!PageLocked(old), old);
680 VM_BUG_ON_PAGE(!PageLocked(new), new);
681 VM_BUG_ON_PAGE(new->mapping, new);
683 error = radix_tree_preload(gfp_mask & ~__GFP_HIGHMEM);
685 struct address_space *mapping = old->mapping;
686 void (*freepage)(struct page *);
689 pgoff_t offset = old->index;
690 freepage = mapping->a_ops->freepage;
693 new->mapping = mapping;
696 spin_lock_irqsave(&mapping->tree_lock, flags);
697 __delete_from_page_cache(old, NULL);
698 error = page_cache_tree_insert(mapping, new, NULL);
702 * hugetlb pages do not participate in page cache accounting.
705 __inc_node_page_state(new, NR_FILE_PAGES);
706 if (PageSwapBacked(new))
707 __inc_node_page_state(new, NR_SHMEM);
708 spin_unlock_irqrestore(&mapping->tree_lock, flags);
709 mem_cgroup_migrate(old, new);
710 radix_tree_preload_end();
718 EXPORT_SYMBOL_GPL(replace_page_cache_page);
720 static int __add_to_page_cache_locked(struct page *page,
721 struct address_space *mapping,
722 pgoff_t offset, gfp_t gfp_mask,
725 int huge = PageHuge(page);
726 struct mem_cgroup *memcg;
729 VM_BUG_ON_PAGE(!PageLocked(page), page);
730 VM_BUG_ON_PAGE(PageSwapBacked(page), page);
733 error = mem_cgroup_try_charge(page, current->mm,
734 gfp_mask, &memcg, false);
739 error = radix_tree_maybe_preload(gfp_mask & ~__GFP_HIGHMEM);
742 mem_cgroup_cancel_charge(page, memcg, false);
747 page->mapping = mapping;
748 page->index = offset;
750 spin_lock_irq(&mapping->tree_lock);
751 error = page_cache_tree_insert(mapping, page, shadowp);
752 radix_tree_preload_end();
756 /* hugetlb pages do not participate in page cache accounting. */
758 __inc_node_page_state(page, NR_FILE_PAGES);
759 spin_unlock_irq(&mapping->tree_lock);
761 mem_cgroup_commit_charge(page, memcg, false, false);
762 trace_mm_filemap_add_to_page_cache(page);
765 page->mapping = NULL;
766 /* Leave page->index set: truncation relies upon it */
767 spin_unlock_irq(&mapping->tree_lock);
769 mem_cgroup_cancel_charge(page, memcg, false);
775 * add_to_page_cache_locked - add a locked page to the pagecache
777 * @mapping: the page's address_space
778 * @offset: page index
779 * @gfp_mask: page allocation mode
781 * This function is used to add a page to the pagecache. It must be locked.
782 * This function does not add the page to the LRU. The caller must do that.
784 int add_to_page_cache_locked(struct page *page, struct address_space *mapping,
785 pgoff_t offset, gfp_t gfp_mask)
787 return __add_to_page_cache_locked(page, mapping, offset,
790 EXPORT_SYMBOL(add_to_page_cache_locked);
792 int add_to_page_cache_lru(struct page *page, struct address_space *mapping,
793 pgoff_t offset, gfp_t gfp_mask)
798 __SetPageLocked(page);
799 ret = __add_to_page_cache_locked(page, mapping, offset,
802 __ClearPageLocked(page);
805 * The page might have been evicted from cache only
806 * recently, in which case it should be activated like
807 * any other repeatedly accessed page.
808 * The exception is pages getting rewritten; evicting other
809 * data from the working set, only to cache data that will
810 * get overwritten with something else, is a waste of memory.
812 if (!(gfp_mask & __GFP_WRITE) &&
813 shadow && workingset_refault(shadow)) {
815 workingset_activation(page);
817 ClearPageActive(page);
822 EXPORT_SYMBOL_GPL(add_to_page_cache_lru);
825 struct page *__page_cache_alloc(gfp_t gfp)
830 if (cpuset_do_page_mem_spread()) {
831 unsigned int cpuset_mems_cookie;
833 cpuset_mems_cookie = read_mems_allowed_begin();
834 n = cpuset_mem_spread_node();
835 page = __alloc_pages_node(n, gfp, 0);
836 } while (!page && read_mems_allowed_retry(cpuset_mems_cookie));
840 return alloc_pages(gfp, 0);
842 EXPORT_SYMBOL(__page_cache_alloc);
846 * In order to wait for pages to become available there must be
847 * waitqueues associated with pages. By using a hash table of
848 * waitqueues where the bucket discipline is to maintain all
849 * waiters on the same queue and wake all when any of the pages
850 * become available, and for the woken contexts to check to be
851 * sure the appropriate page became available, this saves space
852 * at a cost of "thundering herd" phenomena during rare hash
855 #define PAGE_WAIT_TABLE_BITS 8
856 #define PAGE_WAIT_TABLE_SIZE (1 << PAGE_WAIT_TABLE_BITS)
857 static wait_queue_head_t page_wait_table[PAGE_WAIT_TABLE_SIZE] __cacheline_aligned;
859 static wait_queue_head_t *page_waitqueue(struct page *page)
861 return &page_wait_table[hash_ptr(page, PAGE_WAIT_TABLE_BITS)];
864 void __init pagecache_init(void)
868 for (i = 0; i < PAGE_WAIT_TABLE_SIZE; i++)
869 init_waitqueue_head(&page_wait_table[i]);
871 page_writeback_init();
874 /* This has the same layout as wait_bit_key - see fs/cachefiles/rdwr.c */
875 struct wait_page_key {
881 struct wait_page_queue {
884 wait_queue_entry_t wait;
887 static int wake_page_function(wait_queue_entry_t *wait, unsigned mode, int sync, void *arg)
889 struct wait_page_key *key = arg;
890 struct wait_page_queue *wait_page
891 = container_of(wait, struct wait_page_queue, wait);
893 if (wait_page->page != key->page)
897 if (wait_page->bit_nr != key->bit_nr)
900 /* Stop walking if it's locked */
901 if (test_bit(key->bit_nr, &key->page->flags))
904 return autoremove_wake_function(wait, mode, sync, key);
907 static void wake_up_page_bit(struct page *page, int bit_nr)
909 wait_queue_head_t *q = page_waitqueue(page);
910 struct wait_page_key key;
912 wait_queue_entry_t bookmark;
919 bookmark.private = NULL;
920 bookmark.func = NULL;
921 INIT_LIST_HEAD(&bookmark.entry);
923 spin_lock_irqsave(&q->lock, flags);
924 __wake_up_locked_key_bookmark(q, TASK_NORMAL, &key, &bookmark);
926 while (bookmark.flags & WQ_FLAG_BOOKMARK) {
928 * Take a breather from holding the lock,
929 * allow pages that finish wake up asynchronously
930 * to acquire the lock and remove themselves
933 spin_unlock_irqrestore(&q->lock, flags);
935 spin_lock_irqsave(&q->lock, flags);
936 __wake_up_locked_key_bookmark(q, TASK_NORMAL, &key, &bookmark);
940 * It is possible for other pages to have collided on the waitqueue
941 * hash, so in that case check for a page match. That prevents a long-
944 * It is still possible to miss a case here, when we woke page waiters
945 * and removed them from the waitqueue, but there are still other
948 if (!waitqueue_active(q) || !key.page_match) {
949 ClearPageWaiters(page);
951 * It's possible to miss clearing Waiters here, when we woke
952 * our page waiters, but the hashed waitqueue has waiters for
955 * That's okay, it's a rare case. The next waker will clear it.
958 spin_unlock_irqrestore(&q->lock, flags);
961 static void wake_up_page(struct page *page, int bit)
963 if (!PageWaiters(page))
965 wake_up_page_bit(page, bit);
968 static inline int wait_on_page_bit_common(wait_queue_head_t *q,
969 struct page *page, int bit_nr, int state, bool lock)
971 struct wait_page_queue wait_page;
972 wait_queue_entry_t *wait = &wait_page.wait;
976 wait->flags = lock ? WQ_FLAG_EXCLUSIVE : 0;
977 wait->func = wake_page_function;
978 wait_page.page = page;
979 wait_page.bit_nr = bit_nr;
982 spin_lock_irq(&q->lock);
984 if (likely(list_empty(&wait->entry))) {
985 __add_wait_queue_entry_tail(q, wait);
986 SetPageWaiters(page);
989 set_current_state(state);
991 spin_unlock_irq(&q->lock);
993 if (likely(test_bit(bit_nr, &page->flags))) {
998 if (!test_and_set_bit_lock(bit_nr, &page->flags))
1001 if (!test_bit(bit_nr, &page->flags))
1005 if (unlikely(signal_pending_state(state, current))) {
1011 finish_wait(q, wait);
1014 * A signal could leave PageWaiters set. Clearing it here if
1015 * !waitqueue_active would be possible (by open-coding finish_wait),
1016 * but still fail to catch it in the case of wait hash collision. We
1017 * already can fail to clear wait hash collision cases, so don't
1018 * bother with signals either.
1024 void wait_on_page_bit(struct page *page, int bit_nr)
1026 wait_queue_head_t *q = page_waitqueue(page);
1027 wait_on_page_bit_common(q, page, bit_nr, TASK_UNINTERRUPTIBLE, false);
1029 EXPORT_SYMBOL(wait_on_page_bit);
1031 int wait_on_page_bit_killable(struct page *page, int bit_nr)
1033 wait_queue_head_t *q = page_waitqueue(page);
1034 return wait_on_page_bit_common(q, page, bit_nr, TASK_KILLABLE, false);
1038 * add_page_wait_queue - Add an arbitrary waiter to a page's wait queue
1039 * @page: Page defining the wait queue of interest
1040 * @waiter: Waiter to add to the queue
1042 * Add an arbitrary @waiter to the wait queue for the nominated @page.
1044 void add_page_wait_queue(struct page *page, wait_queue_entry_t *waiter)
1046 wait_queue_head_t *q = page_waitqueue(page);
1047 unsigned long flags;
1049 spin_lock_irqsave(&q->lock, flags);
1050 __add_wait_queue_entry_tail(q, waiter);
1051 SetPageWaiters(page);
1052 spin_unlock_irqrestore(&q->lock, flags);
1054 EXPORT_SYMBOL_GPL(add_page_wait_queue);
1056 #ifndef clear_bit_unlock_is_negative_byte
1059 * PG_waiters is the high bit in the same byte as PG_lock.
1061 * On x86 (and on many other architectures), we can clear PG_lock and
1062 * test the sign bit at the same time. But if the architecture does
1063 * not support that special operation, we just do this all by hand
1066 * The read of PG_waiters has to be after (or concurrently with) PG_locked
1067 * being cleared, but a memory barrier should be unneccssary since it is
1068 * in the same byte as PG_locked.
1070 static inline bool clear_bit_unlock_is_negative_byte(long nr, volatile void *mem)
1072 clear_bit_unlock(nr, mem);
1073 /* smp_mb__after_atomic(); */
1074 return test_bit(PG_waiters, mem);
1080 * unlock_page - unlock a locked page
1083 * Unlocks the page and wakes up sleepers in ___wait_on_page_locked().
1084 * Also wakes sleepers in wait_on_page_writeback() because the wakeup
1085 * mechanism between PageLocked pages and PageWriteback pages is shared.
1086 * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep.
1088 * Note that this depends on PG_waiters being the sign bit in the byte
1089 * that contains PG_locked - thus the BUILD_BUG_ON(). That allows us to
1090 * clear the PG_locked bit and test PG_waiters at the same time fairly
1091 * portably (architectures that do LL/SC can test any bit, while x86 can
1092 * test the sign bit).
1094 void unlock_page(struct page *page)
1096 BUILD_BUG_ON(PG_waiters != 7);
1097 page = compound_head(page);
1098 VM_BUG_ON_PAGE(!PageLocked(page), page);
1099 if (clear_bit_unlock_is_negative_byte(PG_locked, &page->flags))
1100 wake_up_page_bit(page, PG_locked);
1102 EXPORT_SYMBOL(unlock_page);
1105 * end_page_writeback - end writeback against a page
1108 void end_page_writeback(struct page *page)
1111 * TestClearPageReclaim could be used here but it is an atomic
1112 * operation and overkill in this particular case. Failing to
1113 * shuffle a page marked for immediate reclaim is too mild to
1114 * justify taking an atomic operation penalty at the end of
1115 * ever page writeback.
1117 if (PageReclaim(page)) {
1118 ClearPageReclaim(page);
1119 rotate_reclaimable_page(page);
1122 if (!test_clear_page_writeback(page))
1125 smp_mb__after_atomic();
1126 wake_up_page(page, PG_writeback);
1128 EXPORT_SYMBOL(end_page_writeback);
1131 * After completing I/O on a page, call this routine to update the page
1132 * flags appropriately
1134 void page_endio(struct page *page, bool is_write, int err)
1138 SetPageUptodate(page);
1140 ClearPageUptodate(page);
1146 struct address_space *mapping;
1149 mapping = page_mapping(page);
1151 mapping_set_error(mapping, err);
1153 end_page_writeback(page);
1156 EXPORT_SYMBOL_GPL(page_endio);
1159 * __lock_page - get a lock on the page, assuming we need to sleep to get it
1160 * @__page: the page to lock
1162 void __lock_page(struct page *__page)
1164 struct page *page = compound_head(__page);
1165 wait_queue_head_t *q = page_waitqueue(page);
1166 wait_on_page_bit_common(q, page, PG_locked, TASK_UNINTERRUPTIBLE, true);
1168 EXPORT_SYMBOL(__lock_page);
1170 int __lock_page_killable(struct page *__page)
1172 struct page *page = compound_head(__page);
1173 wait_queue_head_t *q = page_waitqueue(page);
1174 return wait_on_page_bit_common(q, page, PG_locked, TASK_KILLABLE, true);
1176 EXPORT_SYMBOL_GPL(__lock_page_killable);
1180 * 1 - page is locked; mmap_sem is still held.
1181 * 0 - page is not locked.
1182 * mmap_sem has been released (up_read()), unless flags had both
1183 * FAULT_FLAG_ALLOW_RETRY and FAULT_FLAG_RETRY_NOWAIT set, in
1184 * which case mmap_sem is still held.
1186 * If neither ALLOW_RETRY nor KILLABLE are set, will always return 1
1187 * with the page locked and the mmap_sem unperturbed.
1189 int __lock_page_or_retry(struct page *page, struct mm_struct *mm,
1192 if (flags & FAULT_FLAG_ALLOW_RETRY) {
1194 * CAUTION! In this case, mmap_sem is not released
1195 * even though return 0.
1197 if (flags & FAULT_FLAG_RETRY_NOWAIT)
1200 up_read(&mm->mmap_sem);
1201 if (flags & FAULT_FLAG_KILLABLE)
1202 wait_on_page_locked_killable(page);
1204 wait_on_page_locked(page);
1207 if (flags & FAULT_FLAG_KILLABLE) {
1210 ret = __lock_page_killable(page);
1212 up_read(&mm->mmap_sem);
1222 * page_cache_next_hole - find the next hole (not-present entry)
1225 * @max_scan: maximum range to search
1227 * Search the set [index, min(index+max_scan-1, MAX_INDEX)] for the
1228 * lowest indexed hole.
1230 * Returns: the index of the hole if found, otherwise returns an index
1231 * outside of the set specified (in which case 'return - index >=
1232 * max_scan' will be true). In rare cases of index wrap-around, 0 will
1235 * page_cache_next_hole may be called under rcu_read_lock. However,
1236 * like radix_tree_gang_lookup, this will not atomically search a
1237 * snapshot of the tree at a single point in time. For example, if a
1238 * hole is created at index 5, then subsequently a hole is created at
1239 * index 10, page_cache_next_hole covering both indexes may return 10
1240 * if called under rcu_read_lock.
1242 pgoff_t page_cache_next_hole(struct address_space *mapping,
1243 pgoff_t index, unsigned long max_scan)
1247 for (i = 0; i < max_scan; i++) {
1250 page = radix_tree_lookup(&mapping->page_tree, index);
1251 if (!page || radix_tree_exceptional_entry(page))
1260 EXPORT_SYMBOL(page_cache_next_hole);
1263 * page_cache_prev_hole - find the prev hole (not-present entry)
1266 * @max_scan: maximum range to search
1268 * Search backwards in the range [max(index-max_scan+1, 0), index] for
1271 * Returns: the index of the hole if found, otherwise returns an index
1272 * outside of the set specified (in which case 'index - return >=
1273 * max_scan' will be true). In rare cases of wrap-around, ULONG_MAX
1276 * page_cache_prev_hole may be called under rcu_read_lock. However,
1277 * like radix_tree_gang_lookup, this will not atomically search a
1278 * snapshot of the tree at a single point in time. For example, if a
1279 * hole is created at index 10, then subsequently a hole is created at
1280 * index 5, page_cache_prev_hole covering both indexes may return 5 if
1281 * called under rcu_read_lock.
1283 pgoff_t page_cache_prev_hole(struct address_space *mapping,
1284 pgoff_t index, unsigned long max_scan)
1288 for (i = 0; i < max_scan; i++) {
1291 page = radix_tree_lookup(&mapping->page_tree, index);
1292 if (!page || radix_tree_exceptional_entry(page))
1295 if (index == ULONG_MAX)
1301 EXPORT_SYMBOL(page_cache_prev_hole);
1304 * find_get_entry - find and get a page cache entry
1305 * @mapping: the address_space to search
1306 * @offset: the page cache index
1308 * Looks up the page cache slot at @mapping & @offset. If there is a
1309 * page cache page, it is returned with an increased refcount.
1311 * If the slot holds a shadow entry of a previously evicted page, or a
1312 * swap entry from shmem/tmpfs, it is returned.
1314 * Otherwise, %NULL is returned.
1316 struct page *find_get_entry(struct address_space *mapping, pgoff_t offset)
1319 struct page *head, *page;
1324 pagep = radix_tree_lookup_slot(&mapping->page_tree, offset);
1326 page = radix_tree_deref_slot(pagep);
1327 if (unlikely(!page))
1329 if (radix_tree_exception(page)) {
1330 if (radix_tree_deref_retry(page))
1333 * A shadow entry of a recently evicted page,
1334 * or a swap entry from shmem/tmpfs. Return
1335 * it without attempting to raise page count.
1340 head = compound_head(page);
1341 if (!page_cache_get_speculative(head))
1344 /* The page was split under us? */
1345 if (compound_head(page) != head) {
1351 * Has the page moved?
1352 * This is part of the lockless pagecache protocol. See
1353 * include/linux/pagemap.h for details.
1355 if (unlikely(page != *pagep)) {
1365 EXPORT_SYMBOL(find_get_entry);
1368 * find_lock_entry - locate, pin and lock a page cache entry
1369 * @mapping: the address_space to search
1370 * @offset: the page cache index
1372 * Looks up the page cache slot at @mapping & @offset. If there is a
1373 * page cache page, it is returned locked and with an increased
1376 * If the slot holds a shadow entry of a previously evicted page, or a
1377 * swap entry from shmem/tmpfs, it is returned.
1379 * Otherwise, %NULL is returned.
1381 * find_lock_entry() may sleep.
1383 struct page *find_lock_entry(struct address_space *mapping, pgoff_t offset)
1388 page = find_get_entry(mapping, offset);
1389 if (page && !radix_tree_exception(page)) {
1391 /* Has the page been truncated? */
1392 if (unlikely(page_mapping(page) != mapping)) {
1397 VM_BUG_ON_PAGE(page_to_pgoff(page) != offset, page);
1401 EXPORT_SYMBOL(find_lock_entry);
1404 * pagecache_get_page - find and get a page reference
1405 * @mapping: the address_space to search
1406 * @offset: the page index
1407 * @fgp_flags: PCG flags
1408 * @gfp_mask: gfp mask to use for the page cache data page allocation
1410 * Looks up the page cache slot at @mapping & @offset.
1412 * PCG flags modify how the page is returned.
1414 * @fgp_flags can be:
1416 * - FGP_ACCESSED: the page will be marked accessed
1417 * - FGP_LOCK: Page is return locked
1418 * - FGP_CREAT: If page is not present then a new page is allocated using
1419 * @gfp_mask and added to the page cache and the VM's LRU
1420 * list. The page is returned locked and with an increased
1421 * refcount. Otherwise, NULL is returned.
1423 * If FGP_LOCK or FGP_CREAT are specified then the function may sleep even
1424 * if the GFP flags specified for FGP_CREAT are atomic.
1426 * If there is a page cache page, it is returned with an increased refcount.
1428 struct page *pagecache_get_page(struct address_space *mapping, pgoff_t offset,
1429 int fgp_flags, gfp_t gfp_mask)
1434 page = find_get_entry(mapping, offset);
1435 if (radix_tree_exceptional_entry(page))
1440 if (fgp_flags & FGP_LOCK) {
1441 if (fgp_flags & FGP_NOWAIT) {
1442 if (!trylock_page(page)) {
1450 /* Has the page been truncated? */
1451 if (unlikely(page->mapping != mapping)) {
1456 VM_BUG_ON_PAGE(page->index != offset, page);
1459 if (page && (fgp_flags & FGP_ACCESSED))
1460 mark_page_accessed(page);
1463 if (!page && (fgp_flags & FGP_CREAT)) {
1465 if ((fgp_flags & FGP_WRITE) && mapping_cap_account_dirty(mapping))
1466 gfp_mask |= __GFP_WRITE;
1467 if (fgp_flags & FGP_NOFS)
1468 gfp_mask &= ~__GFP_FS;
1470 page = __page_cache_alloc(gfp_mask);
1474 if (WARN_ON_ONCE(!(fgp_flags & FGP_LOCK)))
1475 fgp_flags |= FGP_LOCK;
1477 /* Init accessed so avoid atomic mark_page_accessed later */
1478 if (fgp_flags & FGP_ACCESSED)
1479 __SetPageReferenced(page);
1481 err = add_to_page_cache_lru(page, mapping, offset,
1482 gfp_mask & GFP_RECLAIM_MASK);
1483 if (unlikely(err)) {
1493 EXPORT_SYMBOL(pagecache_get_page);
1496 * find_get_entries - gang pagecache lookup
1497 * @mapping: The address_space to search
1498 * @start: The starting page cache index
1499 * @nr_entries: The maximum number of entries
1500 * @entries: Where the resulting entries are placed
1501 * @indices: The cache indices corresponding to the entries in @entries
1503 * find_get_entries() will search for and return a group of up to
1504 * @nr_entries entries in the mapping. The entries are placed at
1505 * @entries. find_get_entries() takes a reference against any actual
1508 * The search returns a group of mapping-contiguous page cache entries
1509 * with ascending indexes. There may be holes in the indices due to
1510 * not-present pages.
1512 * Any shadow entries of evicted pages, or swap entries from
1513 * shmem/tmpfs, are included in the returned array.
1515 * find_get_entries() returns the number of pages and shadow entries
1518 unsigned find_get_entries(struct address_space *mapping,
1519 pgoff_t start, unsigned int nr_entries,
1520 struct page **entries, pgoff_t *indices)
1523 unsigned int ret = 0;
1524 struct radix_tree_iter iter;
1530 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) {
1531 struct page *head, *page;
1533 page = radix_tree_deref_slot(slot);
1534 if (unlikely(!page))
1536 if (radix_tree_exception(page)) {
1537 if (radix_tree_deref_retry(page)) {
1538 slot = radix_tree_iter_retry(&iter);
1542 * A shadow entry of a recently evicted page, a swap
1543 * entry from shmem/tmpfs or a DAX entry. Return it
1544 * without attempting to raise page count.
1549 head = compound_head(page);
1550 if (!page_cache_get_speculative(head))
1553 /* The page was split under us? */
1554 if (compound_head(page) != head) {
1559 /* Has the page moved? */
1560 if (unlikely(page != *slot)) {
1565 indices[ret] = iter.index;
1566 entries[ret] = page;
1567 if (++ret == nr_entries)
1575 * find_get_pages_range - gang pagecache lookup
1576 * @mapping: The address_space to search
1577 * @start: The starting page index
1578 * @end: The final page index (inclusive)
1579 * @nr_pages: The maximum number of pages
1580 * @pages: Where the resulting pages are placed
1582 * find_get_pages_range() will search for and return a group of up to @nr_pages
1583 * pages in the mapping starting at index @start and up to index @end
1584 * (inclusive). The pages are placed at @pages. find_get_pages_range() takes
1585 * a reference against the returned pages.
1587 * The search returns a group of mapping-contiguous pages with ascending
1588 * indexes. There may be holes in the indices due to not-present pages.
1589 * We also update @start to index the next page for the traversal.
1591 * find_get_pages_range() returns the number of pages which were found. If this
1592 * number is smaller than @nr_pages, the end of specified range has been
1595 unsigned find_get_pages_range(struct address_space *mapping, pgoff_t *start,
1596 pgoff_t end, unsigned int nr_pages,
1597 struct page **pages)
1599 struct radix_tree_iter iter;
1603 if (unlikely(!nr_pages))
1607 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, *start) {
1608 struct page *head, *page;
1610 if (iter.index > end)
1613 page = radix_tree_deref_slot(slot);
1614 if (unlikely(!page))
1617 if (radix_tree_exception(page)) {
1618 if (radix_tree_deref_retry(page)) {
1619 slot = radix_tree_iter_retry(&iter);
1623 * A shadow entry of a recently evicted page,
1624 * or a swap entry from shmem/tmpfs. Skip
1630 head = compound_head(page);
1631 if (!page_cache_get_speculative(head))
1634 /* The page was split under us? */
1635 if (compound_head(page) != head) {
1640 /* Has the page moved? */
1641 if (unlikely(page != *slot)) {
1647 if (++ret == nr_pages) {
1648 *start = pages[ret - 1]->index + 1;
1654 * We come here when there is no page beyond @end. We take care to not
1655 * overflow the index @start as it confuses some of the callers. This
1656 * breaks the iteration when there is page at index -1 but that is
1657 * already broken anyway.
1659 if (end == (pgoff_t)-1)
1660 *start = (pgoff_t)-1;
1670 * find_get_pages_contig - gang contiguous pagecache lookup
1671 * @mapping: The address_space to search
1672 * @index: The starting page index
1673 * @nr_pages: The maximum number of pages
1674 * @pages: Where the resulting pages are placed
1676 * find_get_pages_contig() works exactly like find_get_pages(), except
1677 * that the returned number of pages are guaranteed to be contiguous.
1679 * find_get_pages_contig() returns the number of pages which were found.
1681 unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index,
1682 unsigned int nr_pages, struct page **pages)
1684 struct radix_tree_iter iter;
1686 unsigned int ret = 0;
1688 if (unlikely(!nr_pages))
1692 radix_tree_for_each_contig(slot, &mapping->page_tree, &iter, index) {
1693 struct page *head, *page;
1695 page = radix_tree_deref_slot(slot);
1696 /* The hole, there no reason to continue */
1697 if (unlikely(!page))
1700 if (radix_tree_exception(page)) {
1701 if (radix_tree_deref_retry(page)) {
1702 slot = radix_tree_iter_retry(&iter);
1706 * A shadow entry of a recently evicted page,
1707 * or a swap entry from shmem/tmpfs. Stop
1708 * looking for contiguous pages.
1713 head = compound_head(page);
1714 if (!page_cache_get_speculative(head))
1717 /* The page was split under us? */
1718 if (compound_head(page) != head) {
1723 /* Has the page moved? */
1724 if (unlikely(page != *slot)) {
1730 * must check mapping and index after taking the ref.
1731 * otherwise we can get both false positives and false
1732 * negatives, which is just confusing to the caller.
1734 if (page->mapping == NULL || page_to_pgoff(page) != iter.index) {
1740 if (++ret == nr_pages)
1746 EXPORT_SYMBOL(find_get_pages_contig);
1749 * find_get_pages_tag - find and return pages that match @tag
1750 * @mapping: the address_space to search
1751 * @index: the starting page index
1752 * @tag: the tag index
1753 * @nr_pages: the maximum number of pages
1754 * @pages: where the resulting pages are placed
1756 * Like find_get_pages, except we only return pages which are tagged with
1757 * @tag. We update @index to index the next page for the traversal.
1759 unsigned find_get_pages_tag(struct address_space *mapping, pgoff_t *index,
1760 int tag, unsigned int nr_pages, struct page **pages)
1762 struct radix_tree_iter iter;
1766 if (unlikely(!nr_pages))
1770 radix_tree_for_each_tagged(slot, &mapping->page_tree,
1771 &iter, *index, tag) {
1772 struct page *head, *page;
1774 page = radix_tree_deref_slot(slot);
1775 if (unlikely(!page))
1778 if (radix_tree_exception(page)) {
1779 if (radix_tree_deref_retry(page)) {
1780 slot = radix_tree_iter_retry(&iter);
1784 * A shadow entry of a recently evicted page.
1786 * Those entries should never be tagged, but
1787 * this tree walk is lockless and the tags are
1788 * looked up in bulk, one radix tree node at a
1789 * time, so there is a sizable window for page
1790 * reclaim to evict a page we saw tagged.
1797 head = compound_head(page);
1798 if (!page_cache_get_speculative(head))
1801 /* The page was split under us? */
1802 if (compound_head(page) != head) {
1807 /* Has the page moved? */
1808 if (unlikely(page != *slot)) {
1814 if (++ret == nr_pages)
1821 *index = pages[ret - 1]->index + 1;
1825 EXPORT_SYMBOL(find_get_pages_tag);
1828 * find_get_entries_tag - find and return entries that match @tag
1829 * @mapping: the address_space to search
1830 * @start: the starting page cache index
1831 * @tag: the tag index
1832 * @nr_entries: the maximum number of entries
1833 * @entries: where the resulting entries are placed
1834 * @indices: the cache indices corresponding to the entries in @entries
1836 * Like find_get_entries, except we only return entries which are tagged with
1839 unsigned find_get_entries_tag(struct address_space *mapping, pgoff_t start,
1840 int tag, unsigned int nr_entries,
1841 struct page **entries, pgoff_t *indices)
1844 unsigned int ret = 0;
1845 struct radix_tree_iter iter;
1851 radix_tree_for_each_tagged(slot, &mapping->page_tree,
1852 &iter, start, tag) {
1853 struct page *head, *page;
1855 page = radix_tree_deref_slot(slot);
1856 if (unlikely(!page))
1858 if (radix_tree_exception(page)) {
1859 if (radix_tree_deref_retry(page)) {
1860 slot = radix_tree_iter_retry(&iter);
1865 * A shadow entry of a recently evicted page, a swap
1866 * entry from shmem/tmpfs or a DAX entry. Return it
1867 * without attempting to raise page count.
1872 head = compound_head(page);
1873 if (!page_cache_get_speculative(head))
1876 /* The page was split under us? */
1877 if (compound_head(page) != head) {
1882 /* Has the page moved? */
1883 if (unlikely(page != *slot)) {
1888 indices[ret] = iter.index;
1889 entries[ret] = page;
1890 if (++ret == nr_entries)
1896 EXPORT_SYMBOL(find_get_entries_tag);
1899 * CD/DVDs are error prone. When a medium error occurs, the driver may fail
1900 * a _large_ part of the i/o request. Imagine the worst scenario:
1902 * ---R__________________________________________B__________
1903 * ^ reading here ^ bad block(assume 4k)
1905 * read(R) => miss => readahead(R...B) => media error => frustrating retries
1906 * => failing the whole request => read(R) => read(R+1) =>
1907 * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
1908 * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
1909 * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
1911 * It is going insane. Fix it by quickly scaling down the readahead size.
1913 static void shrink_readahead_size_eio(struct file *filp,
1914 struct file_ra_state *ra)
1920 * do_generic_file_read - generic file read routine
1921 * @filp: the file to read
1922 * @ppos: current file position
1923 * @iter: data destination
1924 * @written: already copied
1926 * This is a generic file read routine, and uses the
1927 * mapping->a_ops->readpage() function for the actual low-level stuff.
1929 * This is really ugly. But the goto's actually try to clarify some
1930 * of the logic when it comes to error handling etc.
1932 static ssize_t do_generic_file_read(struct file *filp, loff_t *ppos,
1933 struct iov_iter *iter, ssize_t written)
1935 struct address_space *mapping = filp->f_mapping;
1936 struct inode *inode = mapping->host;
1937 struct file_ra_state *ra = &filp->f_ra;
1941 unsigned long offset; /* offset into pagecache page */
1942 unsigned int prev_offset;
1945 if (unlikely(*ppos >= inode->i_sb->s_maxbytes))
1947 iov_iter_truncate(iter, inode->i_sb->s_maxbytes);
1949 index = *ppos >> PAGE_SHIFT;
1950 prev_index = ra->prev_pos >> PAGE_SHIFT;
1951 prev_offset = ra->prev_pos & (PAGE_SIZE-1);
1952 last_index = (*ppos + iter->count + PAGE_SIZE-1) >> PAGE_SHIFT;
1953 offset = *ppos & ~PAGE_MASK;
1959 unsigned long nr, ret;
1963 if (fatal_signal_pending(current)) {
1968 page = find_get_page(mapping, index);
1970 page_cache_sync_readahead(mapping,
1972 index, last_index - index);
1973 page = find_get_page(mapping, index);
1974 if (unlikely(page == NULL))
1975 goto no_cached_page;
1977 if (PageReadahead(page)) {
1978 page_cache_async_readahead(mapping,
1980 index, last_index - index);
1982 if (!PageUptodate(page)) {
1984 * See comment in do_read_cache_page on why
1985 * wait_on_page_locked is used to avoid unnecessarily
1986 * serialisations and why it's safe.
1988 error = wait_on_page_locked_killable(page);
1989 if (unlikely(error))
1990 goto readpage_error;
1991 if (PageUptodate(page))
1994 if (inode->i_blkbits == PAGE_SHIFT ||
1995 !mapping->a_ops->is_partially_uptodate)
1996 goto page_not_up_to_date;
1997 /* pipes can't handle partially uptodate pages */
1998 if (unlikely(iter->type & ITER_PIPE))
1999 goto page_not_up_to_date;
2000 if (!trylock_page(page))
2001 goto page_not_up_to_date;
2002 /* Did it get truncated before we got the lock? */
2004 goto page_not_up_to_date_locked;
2005 if (!mapping->a_ops->is_partially_uptodate(page,
2006 offset, iter->count))
2007 goto page_not_up_to_date_locked;
2012 * i_size must be checked after we know the page is Uptodate.
2014 * Checking i_size after the check allows us to calculate
2015 * the correct value for "nr", which means the zero-filled
2016 * part of the page is not copied back to userspace (unless
2017 * another truncate extends the file - this is desired though).
2020 isize = i_size_read(inode);
2021 end_index = (isize - 1) >> PAGE_SHIFT;
2022 if (unlikely(!isize || index > end_index)) {
2027 /* nr is the maximum number of bytes to copy from this page */
2029 if (index == end_index) {
2030 nr = ((isize - 1) & ~PAGE_MASK) + 1;
2038 /* If users can be writing to this page using arbitrary
2039 * virtual addresses, take care about potential aliasing
2040 * before reading the page on the kernel side.
2042 if (mapping_writably_mapped(mapping))
2043 flush_dcache_page(page);
2046 * When a sequential read accesses a page several times,
2047 * only mark it as accessed the first time.
2049 if (prev_index != index || offset != prev_offset)
2050 mark_page_accessed(page);
2054 * Ok, we have the page, and it's up-to-date, so
2055 * now we can copy it to user space...
2058 ret = copy_page_to_iter(page, offset, nr, iter);
2060 index += offset >> PAGE_SHIFT;
2061 offset &= ~PAGE_MASK;
2062 prev_offset = offset;
2066 if (!iov_iter_count(iter))
2074 page_not_up_to_date:
2075 /* Get exclusive access to the page ... */
2076 error = lock_page_killable(page);
2077 if (unlikely(error))
2078 goto readpage_error;
2080 page_not_up_to_date_locked:
2081 /* Did it get truncated before we got the lock? */
2082 if (!page->mapping) {
2088 /* Did somebody else fill it already? */
2089 if (PageUptodate(page)) {
2096 * A previous I/O error may have been due to temporary
2097 * failures, eg. multipath errors.
2098 * PG_error will be set again if readpage fails.
2100 ClearPageError(page);
2101 /* Start the actual read. The read will unlock the page. */
2102 error = mapping->a_ops->readpage(filp, page);
2104 if (unlikely(error)) {
2105 if (error == AOP_TRUNCATED_PAGE) {
2110 goto readpage_error;
2113 if (!PageUptodate(page)) {
2114 error = lock_page_killable(page);
2115 if (unlikely(error))
2116 goto readpage_error;
2117 if (!PageUptodate(page)) {
2118 if (page->mapping == NULL) {
2120 * invalidate_mapping_pages got it
2127 shrink_readahead_size_eio(filp, ra);
2129 goto readpage_error;
2137 /* UHHUH! A synchronous read error occurred. Report it */
2143 * Ok, it wasn't cached, so we need to create a new
2146 page = page_cache_alloc_cold(mapping);
2151 error = add_to_page_cache_lru(page, mapping, index,
2152 mapping_gfp_constraint(mapping, GFP_KERNEL));
2155 if (error == -EEXIST) {
2165 ra->prev_pos = prev_index;
2166 ra->prev_pos <<= PAGE_SHIFT;
2167 ra->prev_pos |= prev_offset;
2169 *ppos = ((loff_t)index << PAGE_SHIFT) + offset;
2170 file_accessed(filp);
2171 return written ? written : error;
2175 * generic_file_read_iter - generic filesystem read routine
2176 * @iocb: kernel I/O control block
2177 * @iter: destination for the data read
2179 * This is the "read_iter()" routine for all filesystems
2180 * that can use the page cache directly.
2183 generic_file_read_iter(struct kiocb *iocb, struct iov_iter *iter)
2185 struct file *file = iocb->ki_filp;
2187 size_t count = iov_iter_count(iter);
2190 goto out; /* skip atime */
2192 if (iocb->ki_flags & IOCB_DIRECT) {
2193 struct address_space *mapping = file->f_mapping;
2194 struct inode *inode = mapping->host;
2197 size = i_size_read(inode);
2198 if (iocb->ki_flags & IOCB_NOWAIT) {
2199 if (filemap_range_has_page(mapping, iocb->ki_pos,
2200 iocb->ki_pos + count - 1))
2203 retval = filemap_write_and_wait_range(mapping,
2205 iocb->ki_pos + count - 1);
2210 file_accessed(file);
2212 retval = mapping->a_ops->direct_IO(iocb, iter);
2214 iocb->ki_pos += retval;
2217 iov_iter_revert(iter, count - iov_iter_count(iter));
2220 * Btrfs can have a short DIO read if we encounter
2221 * compressed extents, so if there was an error, or if
2222 * we've already read everything we wanted to, or if
2223 * there was a short read because we hit EOF, go ahead
2224 * and return. Otherwise fallthrough to buffered io for
2225 * the rest of the read. Buffered reads will not work for
2226 * DAX files, so don't bother trying.
2228 if (retval < 0 || !count || iocb->ki_pos >= size ||
2233 retval = do_generic_file_read(file, &iocb->ki_pos, iter, retval);
2237 EXPORT_SYMBOL(generic_file_read_iter);
2241 * page_cache_read - adds requested page to the page cache if not already there
2242 * @file: file to read
2243 * @offset: page index
2244 * @gfp_mask: memory allocation flags
2246 * This adds the requested page to the page cache if it isn't already there,
2247 * and schedules an I/O to read in its contents from disk.
2249 static int page_cache_read(struct file *file, pgoff_t offset, gfp_t gfp_mask)
2251 struct address_space *mapping = file->f_mapping;
2256 page = __page_cache_alloc(gfp_mask|__GFP_COLD);
2260 ret = add_to_page_cache_lru(page, mapping, offset, gfp_mask & GFP_KERNEL);
2262 ret = mapping->a_ops->readpage(file, page);
2263 else if (ret == -EEXIST)
2264 ret = 0; /* losing race to add is OK */
2268 } while (ret == AOP_TRUNCATED_PAGE);
2273 #define MMAP_LOTSAMISS (100)
2276 * Synchronous readahead happens when we don't even find
2277 * a page in the page cache at all.
2279 static void do_sync_mmap_readahead(struct vm_area_struct *vma,
2280 struct file_ra_state *ra,
2284 struct address_space *mapping = file->f_mapping;
2286 /* If we don't want any read-ahead, don't bother */
2287 if (vma->vm_flags & VM_RAND_READ)
2292 if (vma->vm_flags & VM_SEQ_READ) {
2293 page_cache_sync_readahead(mapping, ra, file, offset,
2298 /* Avoid banging the cache line if not needed */
2299 if (ra->mmap_miss < MMAP_LOTSAMISS * 10)
2303 * Do we miss much more than hit in this file? If so,
2304 * stop bothering with read-ahead. It will only hurt.
2306 if (ra->mmap_miss > MMAP_LOTSAMISS)
2312 ra->start = max_t(long, 0, offset - ra->ra_pages / 2);
2313 ra->size = ra->ra_pages;
2314 ra->async_size = ra->ra_pages / 4;
2315 ra_submit(ra, mapping, file);
2319 * Asynchronous readahead happens when we find the page and PG_readahead,
2320 * so we want to possibly extend the readahead further..
2322 static void do_async_mmap_readahead(struct vm_area_struct *vma,
2323 struct file_ra_state *ra,
2328 struct address_space *mapping = file->f_mapping;
2330 /* If we don't want any read-ahead, don't bother */
2331 if (vma->vm_flags & VM_RAND_READ)
2333 if (ra->mmap_miss > 0)
2335 if (PageReadahead(page))
2336 page_cache_async_readahead(mapping, ra, file,
2337 page, offset, ra->ra_pages);
2341 * filemap_fault - read in file data for page fault handling
2342 * @vmf: struct vm_fault containing details of the fault
2344 * filemap_fault() is invoked via the vma operations vector for a
2345 * mapped memory region to read in file data during a page fault.
2347 * The goto's are kind of ugly, but this streamlines the normal case of having
2348 * it in the page cache, and handles the special cases reasonably without
2349 * having a lot of duplicated code.
2351 * vma->vm_mm->mmap_sem must be held on entry.
2353 * If our return value has VM_FAULT_RETRY set, it's because
2354 * lock_page_or_retry() returned 0.
2355 * The mmap_sem has usually been released in this case.
2356 * See __lock_page_or_retry() for the exception.
2358 * If our return value does not have VM_FAULT_RETRY set, the mmap_sem
2359 * has not been released.
2361 * We never return with VM_FAULT_RETRY and a bit from VM_FAULT_ERROR set.
2363 int filemap_fault(struct vm_fault *vmf)
2366 struct file *file = vmf->vma->vm_file;
2367 struct address_space *mapping = file->f_mapping;
2368 struct file_ra_state *ra = &file->f_ra;
2369 struct inode *inode = mapping->host;
2370 pgoff_t offset = vmf->pgoff;
2375 max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
2376 if (unlikely(offset >= max_off))
2377 return VM_FAULT_SIGBUS;
2380 * Do we have something in the page cache already?
2382 page = find_get_page(mapping, offset);
2383 if (likely(page) && !(vmf->flags & FAULT_FLAG_TRIED)) {
2385 * We found the page, so try async readahead before
2386 * waiting for the lock.
2388 do_async_mmap_readahead(vmf->vma, ra, file, page, offset);
2390 /* No page in the page cache at all */
2391 do_sync_mmap_readahead(vmf->vma, ra, file, offset);
2392 count_vm_event(PGMAJFAULT);
2393 count_memcg_event_mm(vmf->vma->vm_mm, PGMAJFAULT);
2394 ret = VM_FAULT_MAJOR;
2396 page = find_get_page(mapping, offset);
2398 goto no_cached_page;
2401 if (!lock_page_or_retry(page, vmf->vma->vm_mm, vmf->flags)) {
2403 return ret | VM_FAULT_RETRY;
2406 /* Did it get truncated? */
2407 if (unlikely(page->mapping != mapping)) {
2412 VM_BUG_ON_PAGE(page->index != offset, page);
2415 * We have a locked page in the page cache, now we need to check
2416 * that it's up-to-date. If not, it is going to be due to an error.
2418 if (unlikely(!PageUptodate(page)))
2419 goto page_not_uptodate;
2422 * Found the page and have a reference on it.
2423 * We must recheck i_size under page lock.
2425 max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
2426 if (unlikely(offset >= max_off)) {
2429 return VM_FAULT_SIGBUS;
2433 return ret | VM_FAULT_LOCKED;
2437 * We're only likely to ever get here if MADV_RANDOM is in
2440 error = page_cache_read(file, offset, vmf->gfp_mask);
2443 * The page we want has now been added to the page cache.
2444 * In the unlikely event that someone removed it in the
2445 * meantime, we'll just come back here and read it again.
2451 * An error return from page_cache_read can result if the
2452 * system is low on memory, or a problem occurs while trying
2455 if (error == -ENOMEM)
2456 return VM_FAULT_OOM;
2457 return VM_FAULT_SIGBUS;
2461 * Umm, take care of errors if the page isn't up-to-date.
2462 * Try to re-read it _once_. We do this synchronously,
2463 * because there really aren't any performance issues here
2464 * and we need to check for errors.
2466 ClearPageError(page);
2467 error = mapping->a_ops->readpage(file, page);
2469 wait_on_page_locked(page);
2470 if (!PageUptodate(page))
2475 if (!error || error == AOP_TRUNCATED_PAGE)
2478 /* Things didn't work out. Return zero to tell the mm layer so. */
2479 shrink_readahead_size_eio(file, ra);
2480 return VM_FAULT_SIGBUS;
2482 EXPORT_SYMBOL(filemap_fault);
2484 void filemap_map_pages(struct vm_fault *vmf,
2485 pgoff_t start_pgoff, pgoff_t end_pgoff)
2487 struct radix_tree_iter iter;
2489 struct file *file = vmf->vma->vm_file;
2490 struct address_space *mapping = file->f_mapping;
2491 pgoff_t last_pgoff = start_pgoff;
2492 unsigned long max_idx;
2493 struct page *head, *page;
2496 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter,
2498 if (iter.index > end_pgoff)
2501 page = radix_tree_deref_slot(slot);
2502 if (unlikely(!page))
2504 if (radix_tree_exception(page)) {
2505 if (radix_tree_deref_retry(page)) {
2506 slot = radix_tree_iter_retry(&iter);
2512 head = compound_head(page);
2513 if (!page_cache_get_speculative(head))
2516 /* The page was split under us? */
2517 if (compound_head(page) != head) {
2522 /* Has the page moved? */
2523 if (unlikely(page != *slot)) {
2528 if (!PageUptodate(page) ||
2529 PageReadahead(page) ||
2532 if (!trylock_page(page))
2535 if (page->mapping != mapping || !PageUptodate(page))
2538 max_idx = DIV_ROUND_UP(i_size_read(mapping->host), PAGE_SIZE);
2539 if (page->index >= max_idx)
2542 if (file->f_ra.mmap_miss > 0)
2543 file->f_ra.mmap_miss--;
2545 vmf->address += (iter.index - last_pgoff) << PAGE_SHIFT;
2547 vmf->pte += iter.index - last_pgoff;
2548 last_pgoff = iter.index;
2549 if (alloc_set_pte(vmf, NULL, page))
2558 /* Huge page is mapped? No need to proceed. */
2559 if (pmd_trans_huge(*vmf->pmd))
2561 if (iter.index == end_pgoff)
2566 EXPORT_SYMBOL(filemap_map_pages);
2568 int filemap_page_mkwrite(struct vm_fault *vmf)
2570 struct page *page = vmf->page;
2571 struct inode *inode = file_inode(vmf->vma->vm_file);
2572 int ret = VM_FAULT_LOCKED;
2574 sb_start_pagefault(inode->i_sb);
2575 file_update_time(vmf->vma->vm_file);
2577 if (page->mapping != inode->i_mapping) {
2579 ret = VM_FAULT_NOPAGE;
2583 * We mark the page dirty already here so that when freeze is in
2584 * progress, we are guaranteed that writeback during freezing will
2585 * see the dirty page and writeprotect it again.
2587 set_page_dirty(page);
2588 wait_for_stable_page(page);
2590 sb_end_pagefault(inode->i_sb);
2593 EXPORT_SYMBOL(filemap_page_mkwrite);
2595 const struct vm_operations_struct generic_file_vm_ops = {
2596 .fault = filemap_fault,
2597 .map_pages = filemap_map_pages,
2598 .page_mkwrite = filemap_page_mkwrite,
2601 /* This is used for a general mmap of a disk file */
2603 int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
2605 struct address_space *mapping = file->f_mapping;
2607 if (!mapping->a_ops->readpage)
2609 file_accessed(file);
2610 vma->vm_ops = &generic_file_vm_ops;
2615 * This is for filesystems which do not implement ->writepage.
2617 int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
2619 if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE))
2621 return generic_file_mmap(file, vma);
2624 int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
2628 int generic_file_readonly_mmap(struct file * file, struct vm_area_struct * vma)
2632 #endif /* CONFIG_MMU */
2634 EXPORT_SYMBOL(generic_file_mmap);
2635 EXPORT_SYMBOL(generic_file_readonly_mmap);
2637 static struct page *wait_on_page_read(struct page *page)
2639 if (!IS_ERR(page)) {
2640 wait_on_page_locked(page);
2641 if (!PageUptodate(page)) {
2643 page = ERR_PTR(-EIO);
2649 static struct page *do_read_cache_page(struct address_space *mapping,
2651 int (*filler)(void *, struct page *),
2658 page = find_get_page(mapping, index);
2660 page = __page_cache_alloc(gfp | __GFP_COLD);
2662 return ERR_PTR(-ENOMEM);
2663 err = add_to_page_cache_lru(page, mapping, index, gfp);
2664 if (unlikely(err)) {
2668 /* Presumably ENOMEM for radix tree node */
2669 return ERR_PTR(err);
2673 err = filler(data, page);
2676 return ERR_PTR(err);
2679 page = wait_on_page_read(page);
2684 if (PageUptodate(page))
2688 * Page is not up to date and may be locked due one of the following
2689 * case a: Page is being filled and the page lock is held
2690 * case b: Read/write error clearing the page uptodate status
2691 * case c: Truncation in progress (page locked)
2692 * case d: Reclaim in progress
2694 * Case a, the page will be up to date when the page is unlocked.
2695 * There is no need to serialise on the page lock here as the page
2696 * is pinned so the lock gives no additional protection. Even if the
2697 * the page is truncated, the data is still valid if PageUptodate as
2698 * it's a race vs truncate race.
2699 * Case b, the page will not be up to date
2700 * Case c, the page may be truncated but in itself, the data may still
2701 * be valid after IO completes as it's a read vs truncate race. The
2702 * operation must restart if the page is not uptodate on unlock but
2703 * otherwise serialising on page lock to stabilise the mapping gives
2704 * no additional guarantees to the caller as the page lock is
2705 * released before return.
2706 * Case d, similar to truncation. If reclaim holds the page lock, it
2707 * will be a race with remove_mapping that determines if the mapping
2708 * is valid on unlock but otherwise the data is valid and there is
2709 * no need to serialise with page lock.
2711 * As the page lock gives no additional guarantee, we optimistically
2712 * wait on the page to be unlocked and check if it's up to date and
2713 * use the page if it is. Otherwise, the page lock is required to
2714 * distinguish between the different cases. The motivation is that we
2715 * avoid spurious serialisations and wakeups when multiple processes
2716 * wait on the same page for IO to complete.
2718 wait_on_page_locked(page);
2719 if (PageUptodate(page))
2722 /* Distinguish between all the cases under the safety of the lock */
2725 /* Case c or d, restart the operation */
2726 if (!page->mapping) {
2732 /* Someone else locked and filled the page in a very small window */
2733 if (PageUptodate(page)) {
2740 mark_page_accessed(page);
2745 * read_cache_page - read into page cache, fill it if needed
2746 * @mapping: the page's address_space
2747 * @index: the page index
2748 * @filler: function to perform the read
2749 * @data: first arg to filler(data, page) function, often left as NULL
2751 * Read into the page cache. If a page already exists, and PageUptodate() is
2752 * not set, try to fill the page and wait for it to become unlocked.
2754 * If the page does not get brought uptodate, return -EIO.
2756 struct page *read_cache_page(struct address_space *mapping,
2758 int (*filler)(void *, struct page *),
2761 return do_read_cache_page(mapping, index, filler, data, mapping_gfp_mask(mapping));
2763 EXPORT_SYMBOL(read_cache_page);
2766 * read_cache_page_gfp - read into page cache, using specified page allocation flags.
2767 * @mapping: the page's address_space
2768 * @index: the page index
2769 * @gfp: the page allocator flags to use if allocating
2771 * This is the same as "read_mapping_page(mapping, index, NULL)", but with
2772 * any new page allocations done using the specified allocation flags.
2774 * If the page does not get brought uptodate, return -EIO.
2776 struct page *read_cache_page_gfp(struct address_space *mapping,
2780 filler_t *filler = (filler_t *)mapping->a_ops->readpage;
2782 return do_read_cache_page(mapping, index, filler, NULL, gfp);
2784 EXPORT_SYMBOL(read_cache_page_gfp);
2787 * Performs necessary checks before doing a write
2789 * Can adjust writing position or amount of bytes to write.
2790 * Returns appropriate error code that caller should return or
2791 * zero in case that write should be allowed.
2793 inline ssize_t generic_write_checks(struct kiocb *iocb, struct iov_iter *from)
2795 struct file *file = iocb->ki_filp;
2796 struct inode *inode = file->f_mapping->host;
2797 unsigned long limit = rlimit(RLIMIT_FSIZE);
2800 if (!iov_iter_count(from))
2803 /* FIXME: this is for backwards compatibility with 2.4 */
2804 if (iocb->ki_flags & IOCB_APPEND)
2805 iocb->ki_pos = i_size_read(inode);
2809 if ((iocb->ki_flags & IOCB_NOWAIT) && !(iocb->ki_flags & IOCB_DIRECT))
2812 if (limit != RLIM_INFINITY) {
2813 if (iocb->ki_pos >= limit) {
2814 send_sig(SIGXFSZ, current, 0);
2817 iov_iter_truncate(from, limit - (unsigned long)pos);
2823 if (unlikely(pos + iov_iter_count(from) > MAX_NON_LFS &&
2824 !(file->f_flags & O_LARGEFILE))) {
2825 if (pos >= MAX_NON_LFS)
2827 iov_iter_truncate(from, MAX_NON_LFS - (unsigned long)pos);
2831 * Are we about to exceed the fs block limit ?
2833 * If we have written data it becomes a short write. If we have
2834 * exceeded without writing data we send a signal and return EFBIG.
2835 * Linus frestrict idea will clean these up nicely..
2837 if (unlikely(pos >= inode->i_sb->s_maxbytes))
2840 iov_iter_truncate(from, inode->i_sb->s_maxbytes - pos);
2841 return iov_iter_count(from);
2843 EXPORT_SYMBOL(generic_write_checks);
2845 int pagecache_write_begin(struct file *file, struct address_space *mapping,
2846 loff_t pos, unsigned len, unsigned flags,
2847 struct page **pagep, void **fsdata)
2849 const struct address_space_operations *aops = mapping->a_ops;
2851 return aops->write_begin(file, mapping, pos, len, flags,
2854 EXPORT_SYMBOL(pagecache_write_begin);
2856 int pagecache_write_end(struct file *file, struct address_space *mapping,
2857 loff_t pos, unsigned len, unsigned copied,
2858 struct page *page, void *fsdata)
2860 const struct address_space_operations *aops = mapping->a_ops;
2862 return aops->write_end(file, mapping, pos, len, copied, page, fsdata);
2864 EXPORT_SYMBOL(pagecache_write_end);
2867 generic_file_direct_write(struct kiocb *iocb, struct iov_iter *from)
2869 struct file *file = iocb->ki_filp;
2870 struct address_space *mapping = file->f_mapping;
2871 struct inode *inode = mapping->host;
2872 loff_t pos = iocb->ki_pos;
2877 write_len = iov_iter_count(from);
2878 end = (pos + write_len - 1) >> PAGE_SHIFT;
2880 if (iocb->ki_flags & IOCB_NOWAIT) {
2881 /* If there are pages to writeback, return */
2882 if (filemap_range_has_page(inode->i_mapping, pos,
2883 pos + iov_iter_count(from)))
2886 written = filemap_write_and_wait_range(mapping, pos,
2887 pos + write_len - 1);
2893 * After a write we want buffered reads to be sure to go to disk to get
2894 * the new data. We invalidate clean cached page from the region we're
2895 * about to write. We do this *before* the write so that we can return
2896 * without clobbering -EIOCBQUEUED from ->direct_IO().
2898 written = invalidate_inode_pages2_range(mapping,
2899 pos >> PAGE_SHIFT, end);
2901 * If a page can not be invalidated, return 0 to fall back
2902 * to buffered write.
2905 if (written == -EBUSY)
2910 written = mapping->a_ops->direct_IO(iocb, from);
2913 * Finally, try again to invalidate clean pages which might have been
2914 * cached by non-direct readahead, or faulted in by get_user_pages()
2915 * if the source of the write was an mmap'ed region of the file
2916 * we're writing. Either one is a pretty crazy thing to do,
2917 * so we don't support it 100%. If this invalidation
2918 * fails, tough, the write still worked...
2920 invalidate_inode_pages2_range(mapping,
2921 pos >> PAGE_SHIFT, end);
2925 write_len -= written;
2926 if (pos > i_size_read(inode) && !S_ISBLK(inode->i_mode)) {
2927 i_size_write(inode, pos);
2928 mark_inode_dirty(inode);
2932 iov_iter_revert(from, write_len - iov_iter_count(from));
2936 EXPORT_SYMBOL(generic_file_direct_write);
2939 * Find or create a page at the given pagecache position. Return the locked
2940 * page. This function is specifically for buffered writes.
2942 struct page *grab_cache_page_write_begin(struct address_space *mapping,
2943 pgoff_t index, unsigned flags)
2946 int fgp_flags = FGP_LOCK|FGP_WRITE|FGP_CREAT;
2948 if (flags & AOP_FLAG_NOFS)
2949 fgp_flags |= FGP_NOFS;
2951 page = pagecache_get_page(mapping, index, fgp_flags,
2952 mapping_gfp_mask(mapping));
2954 wait_for_stable_page(page);
2958 EXPORT_SYMBOL(grab_cache_page_write_begin);
2960 ssize_t generic_perform_write(struct file *file,
2961 struct iov_iter *i, loff_t pos)
2963 struct address_space *mapping = file->f_mapping;
2964 const struct address_space_operations *a_ops = mapping->a_ops;
2966 ssize_t written = 0;
2967 unsigned int flags = 0;
2971 unsigned long offset; /* Offset into pagecache page */
2972 unsigned long bytes; /* Bytes to write to page */
2973 size_t copied; /* Bytes copied from user */
2976 offset = (pos & (PAGE_SIZE - 1));
2977 bytes = min_t(unsigned long, PAGE_SIZE - offset,
2982 * Bring in the user page that we will copy from _first_.
2983 * Otherwise there's a nasty deadlock on copying from the
2984 * same page as we're writing to, without it being marked
2987 * Not only is this an optimisation, but it is also required
2988 * to check that the address is actually valid, when atomic
2989 * usercopies are used, below.
2991 if (unlikely(iov_iter_fault_in_readable(i, bytes))) {
2996 if (fatal_signal_pending(current)) {
3001 status = a_ops->write_begin(file, mapping, pos, bytes, flags,
3003 if (unlikely(status < 0))
3006 if (mapping_writably_mapped(mapping))
3007 flush_dcache_page(page);
3009 copied = iov_iter_copy_from_user_atomic(page, i, offset, bytes);
3010 flush_dcache_page(page);
3012 status = a_ops->write_end(file, mapping, pos, bytes, copied,
3014 if (unlikely(status < 0))
3020 iov_iter_advance(i, copied);
3021 if (unlikely(copied == 0)) {
3023 * If we were unable to copy any data at all, we must
3024 * fall back to a single segment length write.
3026 * If we didn't fallback here, we could livelock
3027 * because not all segments in the iov can be copied at
3028 * once without a pagefault.
3030 bytes = min_t(unsigned long, PAGE_SIZE - offset,
3031 iov_iter_single_seg_count(i));
3037 balance_dirty_pages_ratelimited(mapping);
3038 } while (iov_iter_count(i));
3040 return written ? written : status;
3042 EXPORT_SYMBOL(generic_perform_write);
3045 * __generic_file_write_iter - write data to a file
3046 * @iocb: IO state structure (file, offset, etc.)
3047 * @from: iov_iter with data to write
3049 * This function does all the work needed for actually writing data to a
3050 * file. It does all basic checks, removes SUID from the file, updates
3051 * modification times and calls proper subroutines depending on whether we
3052 * do direct IO or a standard buffered write.
3054 * It expects i_mutex to be grabbed unless we work on a block device or similar
3055 * object which does not need locking at all.
3057 * This function does *not* take care of syncing data in case of O_SYNC write.
3058 * A caller has to handle it. This is mainly due to the fact that we want to
3059 * avoid syncing under i_mutex.
3061 ssize_t __generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
3063 struct file *file = iocb->ki_filp;
3064 struct address_space * mapping = file->f_mapping;
3065 struct inode *inode = mapping->host;
3066 ssize_t written = 0;
3070 /* We can write back this queue in page reclaim */
3071 current->backing_dev_info = inode_to_bdi(inode);
3072 err = file_remove_privs(file);
3076 err = file_update_time(file);
3080 if (iocb->ki_flags & IOCB_DIRECT) {
3081 loff_t pos, endbyte;
3083 written = generic_file_direct_write(iocb, from);
3085 * If the write stopped short of completing, fall back to
3086 * buffered writes. Some filesystems do this for writes to
3087 * holes, for example. For DAX files, a buffered write will
3088 * not succeed (even if it did, DAX does not handle dirty
3089 * page-cache pages correctly).
3091 if (written < 0 || !iov_iter_count(from) || IS_DAX(inode))
3094 status = generic_perform_write(file, from, pos = iocb->ki_pos);
3096 * If generic_perform_write() returned a synchronous error
3097 * then we want to return the number of bytes which were
3098 * direct-written, or the error code if that was zero. Note
3099 * that this differs from normal direct-io semantics, which
3100 * will return -EFOO even if some bytes were written.
3102 if (unlikely(status < 0)) {
3107 * We need to ensure that the page cache pages are written to
3108 * disk and invalidated to preserve the expected O_DIRECT
3111 endbyte = pos + status - 1;
3112 err = filemap_write_and_wait_range(mapping, pos, endbyte);
3114 iocb->ki_pos = endbyte + 1;
3116 invalidate_mapping_pages(mapping,
3118 endbyte >> PAGE_SHIFT);
3121 * We don't know how much we wrote, so just return
3122 * the number of bytes which were direct-written
3126 written = generic_perform_write(file, from, iocb->ki_pos);
3127 if (likely(written > 0))
3128 iocb->ki_pos += written;
3131 current->backing_dev_info = NULL;
3132 return written ? written : err;
3134 EXPORT_SYMBOL(__generic_file_write_iter);
3137 * generic_file_write_iter - write data to a file
3138 * @iocb: IO state structure
3139 * @from: iov_iter with data to write
3141 * This is a wrapper around __generic_file_write_iter() to be used by most
3142 * filesystems. It takes care of syncing the file in case of O_SYNC file
3143 * and acquires i_mutex as needed.
3145 ssize_t generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
3147 struct file *file = iocb->ki_filp;
3148 struct inode *inode = file->f_mapping->host;
3152 ret = generic_write_checks(iocb, from);
3154 ret = __generic_file_write_iter(iocb, from);
3155 inode_unlock(inode);
3158 ret = generic_write_sync(iocb, ret);
3161 EXPORT_SYMBOL(generic_file_write_iter);
3164 * try_to_release_page() - release old fs-specific metadata on a page
3166 * @page: the page which the kernel is trying to free
3167 * @gfp_mask: memory allocation flags (and I/O mode)
3169 * The address_space is to try to release any data against the page
3170 * (presumably at page->private). If the release was successful, return '1'.
3171 * Otherwise return zero.
3173 * This may also be called if PG_fscache is set on a page, indicating that the
3174 * page is known to the local caching routines.
3176 * The @gfp_mask argument specifies whether I/O may be performed to release
3177 * this page (__GFP_IO), and whether the call may block (__GFP_RECLAIM & __GFP_FS).
3180 int try_to_release_page(struct page *page, gfp_t gfp_mask)
3182 struct address_space * const mapping = page->mapping;
3184 BUG_ON(!PageLocked(page));
3185 if (PageWriteback(page))
3188 if (mapping && mapping->a_ops->releasepage)
3189 return mapping->a_ops->releasepage(page, gfp_mask);
3190 return try_to_free_buffers(page);
3193 EXPORT_SYMBOL(try_to_release_page);