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/hugetlb.h>
35 #include <linux/memcontrol.h>
36 #include <linux/cleancache.h>
37 #include <linux/shmem_fs.h>
38 #include <linux/rmap.h>
39 #include <linux/delayacct.h>
40 #include <linux/psi.h>
43 #define CREATE_TRACE_POINTS
44 #include <trace/events/filemap.h>
47 * FIXME: remove all knowledge of the buffer layer from the core VM
49 #include <linux/buffer_head.h> /* for try_to_free_buffers */
54 * Shared mappings implemented 30.11.1994. It's not fully working yet,
57 * Shared mappings now work. 15.8.1995 Bruno.
59 * finished 'unifying' the page and buffer cache and SMP-threaded the
60 * page-cache, 21.05.1999, Ingo Molnar <mingo@redhat.com>
62 * SMP-threaded pagemap-LRU 1999, Andrea Arcangeli <andrea@suse.de>
68 * ->i_mmap_rwsem (truncate_pagecache)
69 * ->private_lock (__free_pte->__set_page_dirty_buffers)
70 * ->swap_lock (exclusive_swap_page, others)
74 * ->i_mmap_rwsem (truncate->unmap_mapping_range)
78 * ->page_table_lock or pte_lock (various, mainly in memory.c)
79 * ->i_pages lock (arch-dependent flush_dcache_mmap_lock)
82 * ->lock_page (access_process_vm)
84 * ->i_mutex (generic_perform_write)
85 * ->mmap_sem (fault_in_pages_readable->do_page_fault)
88 * sb_lock (fs/fs-writeback.c)
89 * ->i_pages lock (__sync_single_inode)
92 * ->anon_vma.lock (vma_adjust)
95 * ->page_table_lock or pte_lock (anon_vma_prepare and various)
97 * ->page_table_lock or pte_lock
98 * ->swap_lock (try_to_unmap_one)
99 * ->private_lock (try_to_unmap_one)
100 * ->i_pages lock (try_to_unmap_one)
101 * ->zone_lru_lock(zone) (follow_page->mark_page_accessed)
102 * ->zone_lru_lock(zone) (check_pte_range->isolate_lru_page)
103 * ->private_lock (page_remove_rmap->set_page_dirty)
104 * ->i_pages lock (page_remove_rmap->set_page_dirty)
105 * bdi.wb->list_lock (page_remove_rmap->set_page_dirty)
106 * ->inode->i_lock (page_remove_rmap->set_page_dirty)
107 * ->memcg->move_lock (page_remove_rmap->lock_page_memcg)
108 * bdi.wb->list_lock (zap_pte_range->set_page_dirty)
109 * ->inode->i_lock (zap_pte_range->set_page_dirty)
110 * ->private_lock (zap_pte_range->__set_page_dirty_buffers)
113 * ->tasklist_lock (memory_failure, collect_procs_ao)
116 static int page_cache_tree_insert(struct address_space *mapping,
117 struct page *page, void **shadowp)
119 struct radix_tree_node *node;
123 error = __radix_tree_create(&mapping->i_pages, page->index, 0,
130 p = radix_tree_deref_slot_protected(slot,
131 &mapping->i_pages.xa_lock);
132 if (!radix_tree_exceptional_entry(p))
135 mapping->nrexceptional--;
139 __radix_tree_replace(&mapping->i_pages, node, slot, page,
140 workingset_lookup_update(mapping));
145 static void page_cache_tree_delete(struct address_space *mapping,
146 struct page *page, void *shadow)
150 /* hugetlb pages are represented by one entry in the radix tree */
151 nr = PageHuge(page) ? 1 : hpage_nr_pages(page);
153 VM_BUG_ON_PAGE(!PageLocked(page), page);
154 VM_BUG_ON_PAGE(PageTail(page), page);
155 VM_BUG_ON_PAGE(nr != 1 && shadow, page);
157 for (i = 0; i < nr; i++) {
158 struct radix_tree_node *node;
161 __radix_tree_lookup(&mapping->i_pages, page->index + i,
164 VM_BUG_ON_PAGE(!node && nr != 1, page);
166 radix_tree_clear_tags(&mapping->i_pages, node, slot);
167 __radix_tree_replace(&mapping->i_pages, node, slot, shadow,
168 workingset_lookup_update(mapping));
171 page->mapping = NULL;
172 /* Leave page->index set: truncation lookup relies upon it */
175 mapping->nrexceptional += nr;
177 * Make sure the nrexceptional update is committed before
178 * the nrpages update so that final truncate racing
179 * with reclaim does not see both counters 0 at the
180 * same time and miss a shadow entry.
184 mapping->nrpages -= nr;
187 static void unaccount_page_cache_page(struct address_space *mapping,
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 /* hugetlb pages do not participate in page cache accounting. */
231 nr = hpage_nr_pages(page);
233 __mod_node_page_state(page_pgdat(page), NR_FILE_PAGES, -nr);
234 if (PageSwapBacked(page)) {
235 __mod_node_page_state(page_pgdat(page), NR_SHMEM, -nr);
236 if (PageTransHuge(page))
237 __dec_node_page_state(page, NR_SHMEM_THPS);
239 VM_BUG_ON_PAGE(PageTransHuge(page), page);
243 * At this point page must be either written or cleaned by
244 * truncate. Dirty page here signals a bug and loss of
247 * This fixes dirty accounting after removing the page entirely
248 * but leaves PageDirty set: it has no effect for truncated
249 * page and anyway will be cleared before returning page into
252 if (WARN_ON_ONCE(PageDirty(page)))
253 account_page_cleaned(page, mapping, inode_to_wb(mapping->host));
257 * Delete a page from the page cache and free it. Caller has to make
258 * sure the page is locked and that nobody else uses it - or that usage
259 * is safe. The caller must hold the i_pages lock.
261 void __delete_from_page_cache(struct page *page, void *shadow)
263 struct address_space *mapping = page->mapping;
265 trace_mm_filemap_delete_from_page_cache(page);
267 unaccount_page_cache_page(mapping, page);
268 page_cache_tree_delete(mapping, page, shadow);
271 static void page_cache_free_page(struct address_space *mapping,
274 void (*freepage)(struct page *);
276 freepage = mapping->a_ops->freepage;
280 if (PageTransHuge(page) && !PageHuge(page)) {
281 page_ref_sub(page, HPAGE_PMD_NR);
282 VM_BUG_ON_PAGE(page_count(page) <= 0, page);
289 * delete_from_page_cache - delete page from page cache
290 * @page: the page which the kernel is trying to remove from page cache
292 * This must be called only on pages that have been verified to be in the page
293 * cache and locked. It will never put the page into the free list, the caller
294 * has a reference on the page.
296 void delete_from_page_cache(struct page *page)
298 struct address_space *mapping = page_mapping(page);
301 BUG_ON(!PageLocked(page));
302 xa_lock_irqsave(&mapping->i_pages, flags);
303 __delete_from_page_cache(page, NULL);
304 xa_unlock_irqrestore(&mapping->i_pages, flags);
306 page_cache_free_page(mapping, page);
308 EXPORT_SYMBOL(delete_from_page_cache);
311 * page_cache_tree_delete_batch - delete several pages from page cache
312 * @mapping: the mapping to which pages belong
313 * @pvec: pagevec with pages to delete
315 * The function walks over mapping->i_pages and removes pages passed in @pvec
316 * from the mapping. The function expects @pvec to be sorted by page index.
317 * It tolerates holes in @pvec (mapping entries at those indices are not
318 * modified). The function expects only THP head pages to be present in the
319 * @pvec and takes care to delete all corresponding tail pages from the
322 * The function expects the i_pages lock to be held.
325 page_cache_tree_delete_batch(struct address_space *mapping,
326 struct pagevec *pvec)
328 struct radix_tree_iter iter;
331 int i = 0, tail_pages = 0;
335 start = pvec->pages[0]->index;
336 radix_tree_for_each_slot(slot, &mapping->i_pages, &iter, start) {
337 if (i >= pagevec_count(pvec) && !tail_pages)
339 page = radix_tree_deref_slot_protected(slot,
340 &mapping->i_pages.xa_lock);
341 if (radix_tree_exceptional_entry(page))
345 * Some page got inserted in our range? Skip it. We
346 * have our pages locked so they are protected from
349 if (page != pvec->pages[i])
351 WARN_ON_ONCE(!PageLocked(page));
352 if (PageTransHuge(page) && !PageHuge(page))
353 tail_pages = HPAGE_PMD_NR - 1;
354 page->mapping = NULL;
356 * Leave page->index set: truncation lookup relies
363 radix_tree_clear_tags(&mapping->i_pages, iter.node, slot);
364 __radix_tree_replace(&mapping->i_pages, iter.node, slot, NULL,
365 workingset_lookup_update(mapping));
368 mapping->nrpages -= total_pages;
371 void delete_from_page_cache_batch(struct address_space *mapping,
372 struct pagevec *pvec)
377 if (!pagevec_count(pvec))
380 xa_lock_irqsave(&mapping->i_pages, flags);
381 for (i = 0; i < pagevec_count(pvec); i++) {
382 trace_mm_filemap_delete_from_page_cache(pvec->pages[i]);
384 unaccount_page_cache_page(mapping, pvec->pages[i]);
386 page_cache_tree_delete_batch(mapping, pvec);
387 xa_unlock_irqrestore(&mapping->i_pages, flags);
389 for (i = 0; i < pagevec_count(pvec); i++)
390 page_cache_free_page(mapping, pvec->pages[i]);
393 int filemap_check_errors(struct address_space *mapping)
396 /* Check for outstanding write errors */
397 if (test_bit(AS_ENOSPC, &mapping->flags) &&
398 test_and_clear_bit(AS_ENOSPC, &mapping->flags))
400 if (test_bit(AS_EIO, &mapping->flags) &&
401 test_and_clear_bit(AS_EIO, &mapping->flags))
405 EXPORT_SYMBOL(filemap_check_errors);
407 static int filemap_check_and_keep_errors(struct address_space *mapping)
409 /* Check for outstanding write errors */
410 if (test_bit(AS_EIO, &mapping->flags))
412 if (test_bit(AS_ENOSPC, &mapping->flags))
418 * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range
419 * @mapping: address space structure to write
420 * @start: offset in bytes where the range starts
421 * @end: offset in bytes where the range ends (inclusive)
422 * @sync_mode: enable synchronous operation
424 * Start writeback against all of a mapping's dirty pages that lie
425 * within the byte offsets <start, end> inclusive.
427 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
428 * opposed to a regular memory cleansing writeback. The difference between
429 * these two operations is that if a dirty page/buffer is encountered, it must
430 * be waited upon, and not just skipped over.
432 int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
433 loff_t end, int sync_mode)
436 struct writeback_control wbc = {
437 .sync_mode = sync_mode,
438 .nr_to_write = LONG_MAX,
439 .range_start = start,
443 if (!mapping_cap_writeback_dirty(mapping))
446 wbc_attach_fdatawrite_inode(&wbc, mapping->host);
447 ret = do_writepages(mapping, &wbc);
448 wbc_detach_inode(&wbc);
452 static inline int __filemap_fdatawrite(struct address_space *mapping,
455 return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode);
458 int filemap_fdatawrite(struct address_space *mapping)
460 return __filemap_fdatawrite(mapping, WB_SYNC_ALL);
462 EXPORT_SYMBOL(filemap_fdatawrite);
464 int filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
467 return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL);
469 EXPORT_SYMBOL(filemap_fdatawrite_range);
472 * filemap_flush - mostly a non-blocking flush
473 * @mapping: target address_space
475 * This is a mostly non-blocking flush. Not suitable for data-integrity
476 * purposes - I/O may not be started against all dirty pages.
478 int filemap_flush(struct address_space *mapping)
480 return __filemap_fdatawrite(mapping, WB_SYNC_NONE);
482 EXPORT_SYMBOL(filemap_flush);
485 * filemap_range_has_page - check if a page exists in range.
486 * @mapping: address space within which to check
487 * @start_byte: offset in bytes where the range starts
488 * @end_byte: offset in bytes where the range ends (inclusive)
490 * Find at least one page in the range supplied, usually used to check if
491 * direct writing in this range will trigger a writeback.
493 bool filemap_range_has_page(struct address_space *mapping,
494 loff_t start_byte, loff_t end_byte)
496 pgoff_t index = start_byte >> PAGE_SHIFT;
497 pgoff_t end = end_byte >> PAGE_SHIFT;
500 if (end_byte < start_byte)
503 if (mapping->nrpages == 0)
506 if (!find_get_pages_range(mapping, &index, end, 1, &page))
511 EXPORT_SYMBOL(filemap_range_has_page);
513 static void __filemap_fdatawait_range(struct address_space *mapping,
514 loff_t start_byte, loff_t end_byte)
516 pgoff_t index = start_byte >> PAGE_SHIFT;
517 pgoff_t end = end_byte >> PAGE_SHIFT;
521 if (end_byte < start_byte)
525 while (index <= end) {
528 nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index,
529 end, PAGECACHE_TAG_WRITEBACK);
533 for (i = 0; i < nr_pages; i++) {
534 struct page *page = pvec.pages[i];
536 wait_on_page_writeback(page);
537 ClearPageError(page);
539 pagevec_release(&pvec);
545 * filemap_fdatawait_range - wait for writeback to complete
546 * @mapping: address space structure to wait for
547 * @start_byte: offset in bytes where the range starts
548 * @end_byte: offset in bytes where the range ends (inclusive)
550 * Walk the list of under-writeback pages of the given address space
551 * in the given range and wait for all of them. Check error status of
552 * the address space and return it.
554 * Since the error status of the address space is cleared by this function,
555 * callers are responsible for checking the return value and handling and/or
556 * reporting the error.
558 int filemap_fdatawait_range(struct address_space *mapping, loff_t start_byte,
561 __filemap_fdatawait_range(mapping, start_byte, end_byte);
562 return filemap_check_errors(mapping);
564 EXPORT_SYMBOL(filemap_fdatawait_range);
567 * file_fdatawait_range - wait for writeback to complete
568 * @file: file pointing to address space structure to wait for
569 * @start_byte: offset in bytes where the range starts
570 * @end_byte: offset in bytes where the range ends (inclusive)
572 * Walk the list of under-writeback pages of the address space that file
573 * refers to, in the given range and wait for all of them. Check error
574 * status of the address space vs. the file->f_wb_err cursor and return it.
576 * Since the error status of the file is advanced by this function,
577 * callers are responsible for checking the return value and handling and/or
578 * reporting the error.
580 int file_fdatawait_range(struct file *file, loff_t start_byte, loff_t end_byte)
582 struct address_space *mapping = file->f_mapping;
584 __filemap_fdatawait_range(mapping, start_byte, end_byte);
585 return file_check_and_advance_wb_err(file);
587 EXPORT_SYMBOL(file_fdatawait_range);
590 * filemap_fdatawait_keep_errors - wait for writeback without clearing errors
591 * @mapping: address space structure to wait for
593 * Walk the list of under-writeback pages of the given address space
594 * and wait for all of them. Unlike filemap_fdatawait(), this function
595 * does not clear error status of the address space.
597 * Use this function if callers don't handle errors themselves. Expected
598 * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2),
601 int filemap_fdatawait_keep_errors(struct address_space *mapping)
603 __filemap_fdatawait_range(mapping, 0, LLONG_MAX);
604 return filemap_check_and_keep_errors(mapping);
606 EXPORT_SYMBOL(filemap_fdatawait_keep_errors);
608 static bool mapping_needs_writeback(struct address_space *mapping)
610 return (!dax_mapping(mapping) && mapping->nrpages) ||
611 (dax_mapping(mapping) && mapping->nrexceptional);
614 int filemap_write_and_wait(struct address_space *mapping)
618 if (mapping_needs_writeback(mapping)) {
619 err = filemap_fdatawrite(mapping);
621 * Even if the above returned error, the pages may be
622 * written partially (e.g. -ENOSPC), so we wait for it.
623 * But the -EIO is special case, it may indicate the worst
624 * thing (e.g. bug) happened, so we avoid waiting for it.
627 int err2 = filemap_fdatawait(mapping);
631 /* Clear any previously stored errors */
632 filemap_check_errors(mapping);
635 err = filemap_check_errors(mapping);
639 EXPORT_SYMBOL(filemap_write_and_wait);
642 * filemap_write_and_wait_range - write out & wait on a file range
643 * @mapping: the address_space for the pages
644 * @lstart: offset in bytes where the range starts
645 * @lend: offset in bytes where the range ends (inclusive)
647 * Write out and wait upon file offsets lstart->lend, inclusive.
649 * Note that @lend is inclusive (describes the last byte to be written) so
650 * that this function can be used to write to the very end-of-file (end = -1).
652 int filemap_write_and_wait_range(struct address_space *mapping,
653 loff_t lstart, loff_t lend)
657 if (mapping_needs_writeback(mapping)) {
658 err = __filemap_fdatawrite_range(mapping, lstart, lend,
660 /* See comment of filemap_write_and_wait() */
662 int err2 = filemap_fdatawait_range(mapping,
667 /* Clear any previously stored errors */
668 filemap_check_errors(mapping);
671 err = filemap_check_errors(mapping);
675 EXPORT_SYMBOL(filemap_write_and_wait_range);
677 void __filemap_set_wb_err(struct address_space *mapping, int err)
679 errseq_t eseq = errseq_set(&mapping->wb_err, err);
681 trace_filemap_set_wb_err(mapping, eseq);
683 EXPORT_SYMBOL(__filemap_set_wb_err);
686 * file_check_and_advance_wb_err - report wb error (if any) that was previously
687 * and advance wb_err to current one
688 * @file: struct file on which the error is being reported
690 * When userland calls fsync (or something like nfsd does the equivalent), we
691 * want to report any writeback errors that occurred since the last fsync (or
692 * since the file was opened if there haven't been any).
694 * Grab the wb_err from the mapping. If it matches what we have in the file,
695 * then just quickly return 0. The file is all caught up.
697 * If it doesn't match, then take the mapping value, set the "seen" flag in
698 * it and try to swap it into place. If it works, or another task beat us
699 * to it with the new value, then update the f_wb_err and return the error
700 * portion. The error at this point must be reported via proper channels
701 * (a'la fsync, or NFS COMMIT operation, etc.).
703 * While we handle mapping->wb_err with atomic operations, the f_wb_err
704 * value is protected by the f_lock since we must ensure that it reflects
705 * the latest value swapped in for this file descriptor.
707 int file_check_and_advance_wb_err(struct file *file)
710 errseq_t old = READ_ONCE(file->f_wb_err);
711 struct address_space *mapping = file->f_mapping;
713 /* Locklessly handle the common case where nothing has changed */
714 if (errseq_check(&mapping->wb_err, old)) {
715 /* Something changed, must use slow path */
716 spin_lock(&file->f_lock);
717 old = file->f_wb_err;
718 err = errseq_check_and_advance(&mapping->wb_err,
720 trace_file_check_and_advance_wb_err(file, old);
721 spin_unlock(&file->f_lock);
725 * We're mostly using this function as a drop in replacement for
726 * filemap_check_errors. Clear AS_EIO/AS_ENOSPC to emulate the effect
727 * that the legacy code would have had on these flags.
729 clear_bit(AS_EIO, &mapping->flags);
730 clear_bit(AS_ENOSPC, &mapping->flags);
733 EXPORT_SYMBOL(file_check_and_advance_wb_err);
736 * file_write_and_wait_range - write out & wait on a file range
737 * @file: file pointing to address_space with pages
738 * @lstart: offset in bytes where the range starts
739 * @lend: offset in bytes where the range ends (inclusive)
741 * Write out and wait upon file offsets lstart->lend, inclusive.
743 * Note that @lend is inclusive (describes the last byte to be written) so
744 * that this function can be used to write to the very end-of-file (end = -1).
746 * After writing out and waiting on the data, we check and advance the
747 * f_wb_err cursor to the latest value, and return any errors detected there.
749 int file_write_and_wait_range(struct file *file, loff_t lstart, loff_t lend)
752 struct address_space *mapping = file->f_mapping;
754 if (mapping_needs_writeback(mapping)) {
755 err = __filemap_fdatawrite_range(mapping, lstart, lend,
757 /* See comment of filemap_write_and_wait() */
759 __filemap_fdatawait_range(mapping, lstart, lend);
761 err2 = file_check_and_advance_wb_err(file);
766 EXPORT_SYMBOL(file_write_and_wait_range);
769 * replace_page_cache_page - replace a pagecache page with a new one
770 * @old: page to be replaced
771 * @new: page to replace with
772 * @gfp_mask: allocation mode
774 * This function replaces a page in the pagecache with a new one. On
775 * success it acquires the pagecache reference for the new page and
776 * drops it for the old page. Both the old and new pages must be
777 * locked. This function does not add the new page to the LRU, the
778 * caller must do that.
780 * The remove + add is atomic. The only way this function can fail is
781 * memory allocation failure.
783 int replace_page_cache_page(struct page *old, struct page *new, gfp_t gfp_mask)
787 VM_BUG_ON_PAGE(!PageLocked(old), old);
788 VM_BUG_ON_PAGE(!PageLocked(new), new);
789 VM_BUG_ON_PAGE(new->mapping, new);
791 error = radix_tree_preload(gfp_mask & GFP_RECLAIM_MASK);
793 struct address_space *mapping = old->mapping;
794 void (*freepage)(struct page *);
797 pgoff_t offset = old->index;
798 freepage = mapping->a_ops->freepage;
801 new->mapping = mapping;
804 xa_lock_irqsave(&mapping->i_pages, flags);
805 __delete_from_page_cache(old, NULL);
806 error = page_cache_tree_insert(mapping, new, NULL);
810 * hugetlb pages do not participate in page cache accounting.
813 __inc_node_page_state(new, NR_FILE_PAGES);
814 if (PageSwapBacked(new))
815 __inc_node_page_state(new, NR_SHMEM);
816 xa_unlock_irqrestore(&mapping->i_pages, flags);
817 mem_cgroup_migrate(old, new);
818 radix_tree_preload_end();
826 EXPORT_SYMBOL_GPL(replace_page_cache_page);
828 static int __add_to_page_cache_locked(struct page *page,
829 struct address_space *mapping,
830 pgoff_t offset, gfp_t gfp_mask,
833 int huge = PageHuge(page);
834 struct mem_cgroup *memcg;
837 VM_BUG_ON_PAGE(!PageLocked(page), page);
838 VM_BUG_ON_PAGE(PageSwapBacked(page), page);
841 error = mem_cgroup_try_charge(page, current->mm,
842 gfp_mask, &memcg, false);
847 error = radix_tree_maybe_preload(gfp_mask & GFP_RECLAIM_MASK);
850 mem_cgroup_cancel_charge(page, memcg, false);
855 page->mapping = mapping;
856 page->index = offset;
858 xa_lock_irq(&mapping->i_pages);
859 error = page_cache_tree_insert(mapping, page, shadowp);
860 radix_tree_preload_end();
864 /* hugetlb pages do not participate in page cache accounting. */
866 __inc_node_page_state(page, NR_FILE_PAGES);
867 xa_unlock_irq(&mapping->i_pages);
869 mem_cgroup_commit_charge(page, memcg, false, false);
870 trace_mm_filemap_add_to_page_cache(page);
873 page->mapping = NULL;
874 /* Leave page->index set: truncation relies upon it */
875 xa_unlock_irq(&mapping->i_pages);
877 mem_cgroup_cancel_charge(page, memcg, false);
883 * add_to_page_cache_locked - add a locked page to the pagecache
885 * @mapping: the page's address_space
886 * @offset: page index
887 * @gfp_mask: page allocation mode
889 * This function is used to add a page to the pagecache. It must be locked.
890 * This function does not add the page to the LRU. The caller must do that.
892 int add_to_page_cache_locked(struct page *page, struct address_space *mapping,
893 pgoff_t offset, gfp_t gfp_mask)
895 return __add_to_page_cache_locked(page, mapping, offset,
898 EXPORT_SYMBOL(add_to_page_cache_locked);
900 int add_to_page_cache_lru(struct page *page, struct address_space *mapping,
901 pgoff_t offset, gfp_t gfp_mask)
906 __SetPageLocked(page);
907 ret = __add_to_page_cache_locked(page, mapping, offset,
910 __ClearPageLocked(page);
913 * The page might have been evicted from cache only
914 * recently, in which case it should be activated like
915 * any other repeatedly accessed page.
916 * The exception is pages getting rewritten; evicting other
917 * data from the working set, only to cache data that will
918 * get overwritten with something else, is a waste of memory.
920 WARN_ON_ONCE(PageActive(page));
921 if (!(gfp_mask & __GFP_WRITE) && shadow)
922 workingset_refault(page, shadow);
927 EXPORT_SYMBOL_GPL(add_to_page_cache_lru);
930 struct page *__page_cache_alloc(gfp_t gfp)
935 if (cpuset_do_page_mem_spread()) {
936 unsigned int cpuset_mems_cookie;
938 cpuset_mems_cookie = read_mems_allowed_begin();
939 n = cpuset_mem_spread_node();
940 page = __alloc_pages_node(n, gfp, 0);
941 } while (!page && read_mems_allowed_retry(cpuset_mems_cookie));
945 return alloc_pages(gfp, 0);
947 EXPORT_SYMBOL(__page_cache_alloc);
951 * In order to wait for pages to become available there must be
952 * waitqueues associated with pages. By using a hash table of
953 * waitqueues where the bucket discipline is to maintain all
954 * waiters on the same queue and wake all when any of the pages
955 * become available, and for the woken contexts to check to be
956 * sure the appropriate page became available, this saves space
957 * at a cost of "thundering herd" phenomena during rare hash
960 #define PAGE_WAIT_TABLE_BITS 8
961 #define PAGE_WAIT_TABLE_SIZE (1 << PAGE_WAIT_TABLE_BITS)
962 static wait_queue_head_t page_wait_table[PAGE_WAIT_TABLE_SIZE] __cacheline_aligned;
964 static wait_queue_head_t *page_waitqueue(struct page *page)
966 return &page_wait_table[hash_ptr(page, PAGE_WAIT_TABLE_BITS)];
969 void __init pagecache_init(void)
973 for (i = 0; i < PAGE_WAIT_TABLE_SIZE; i++)
974 init_waitqueue_head(&page_wait_table[i]);
976 page_writeback_init();
979 /* This has the same layout as wait_bit_key - see fs/cachefiles/rdwr.c */
980 struct wait_page_key {
986 struct wait_page_queue {
989 wait_queue_entry_t wait;
992 static int wake_page_function(wait_queue_entry_t *wait, unsigned mode, int sync, void *arg)
994 struct wait_page_key *key = arg;
995 struct wait_page_queue *wait_page
996 = container_of(wait, struct wait_page_queue, wait);
998 if (wait_page->page != key->page)
1000 key->page_match = 1;
1002 if (wait_page->bit_nr != key->bit_nr)
1005 /* Stop walking if it's locked */
1006 if (test_bit(key->bit_nr, &key->page->flags))
1009 return autoremove_wake_function(wait, mode, sync, key);
1012 static void wake_up_page_bit(struct page *page, int bit_nr)
1014 wait_queue_head_t *q = page_waitqueue(page);
1015 struct wait_page_key key;
1016 unsigned long flags;
1017 wait_queue_entry_t bookmark;
1020 key.bit_nr = bit_nr;
1024 bookmark.private = NULL;
1025 bookmark.func = NULL;
1026 INIT_LIST_HEAD(&bookmark.entry);
1028 spin_lock_irqsave(&q->lock, flags);
1029 __wake_up_locked_key_bookmark(q, TASK_NORMAL, &key, &bookmark);
1031 while (bookmark.flags & WQ_FLAG_BOOKMARK) {
1033 * Take a breather from holding the lock,
1034 * allow pages that finish wake up asynchronously
1035 * to acquire the lock and remove themselves
1038 spin_unlock_irqrestore(&q->lock, flags);
1040 spin_lock_irqsave(&q->lock, flags);
1041 __wake_up_locked_key_bookmark(q, TASK_NORMAL, &key, &bookmark);
1045 * It is possible for other pages to have collided on the waitqueue
1046 * hash, so in that case check for a page match. That prevents a long-
1049 * It is still possible to miss a case here, when we woke page waiters
1050 * and removed them from the waitqueue, but there are still other
1053 if (!waitqueue_active(q) || !key.page_match) {
1054 ClearPageWaiters(page);
1056 * It's possible to miss clearing Waiters here, when we woke
1057 * our page waiters, but the hashed waitqueue has waiters for
1058 * other pages on it.
1060 * That's okay, it's a rare case. The next waker will clear it.
1063 spin_unlock_irqrestore(&q->lock, flags);
1066 static void wake_up_page(struct page *page, int bit)
1068 if (!PageWaiters(page))
1070 wake_up_page_bit(page, bit);
1073 static inline int wait_on_page_bit_common(wait_queue_head_t *q,
1074 struct page *page, int bit_nr, int state, bool lock)
1076 struct wait_page_queue wait_page;
1077 wait_queue_entry_t *wait = &wait_page.wait;
1078 bool thrashing = false;
1079 unsigned long pflags;
1082 if (bit_nr == PG_locked &&
1083 !PageUptodate(page) && PageWorkingset(page)) {
1084 if (!PageSwapBacked(page))
1085 delayacct_thrashing_start();
1086 psi_memstall_enter(&pflags);
1091 wait->flags = lock ? WQ_FLAG_EXCLUSIVE : 0;
1092 wait->func = wake_page_function;
1093 wait_page.page = page;
1094 wait_page.bit_nr = bit_nr;
1097 spin_lock_irq(&q->lock);
1099 if (likely(list_empty(&wait->entry))) {
1100 __add_wait_queue_entry_tail(q, wait);
1101 SetPageWaiters(page);
1104 set_current_state(state);
1106 spin_unlock_irq(&q->lock);
1108 if (likely(test_bit(bit_nr, &page->flags))) {
1113 if (!test_and_set_bit_lock(bit_nr, &page->flags))
1116 if (!test_bit(bit_nr, &page->flags))
1120 if (unlikely(signal_pending_state(state, current))) {
1126 finish_wait(q, wait);
1129 if (!PageSwapBacked(page))
1130 delayacct_thrashing_end();
1131 psi_memstall_leave(&pflags);
1135 * A signal could leave PageWaiters set. Clearing it here if
1136 * !waitqueue_active would be possible (by open-coding finish_wait),
1137 * but still fail to catch it in the case of wait hash collision. We
1138 * already can fail to clear wait hash collision cases, so don't
1139 * bother with signals either.
1145 void wait_on_page_bit(struct page *page, int bit_nr)
1147 wait_queue_head_t *q = page_waitqueue(page);
1148 wait_on_page_bit_common(q, page, bit_nr, TASK_UNINTERRUPTIBLE, false);
1150 EXPORT_SYMBOL(wait_on_page_bit);
1152 int wait_on_page_bit_killable(struct page *page, int bit_nr)
1154 wait_queue_head_t *q = page_waitqueue(page);
1155 return wait_on_page_bit_common(q, page, bit_nr, TASK_KILLABLE, false);
1157 EXPORT_SYMBOL(wait_on_page_bit_killable);
1160 * add_page_wait_queue - Add an arbitrary waiter to a page's wait queue
1161 * @page: Page defining the wait queue of interest
1162 * @waiter: Waiter to add to the queue
1164 * Add an arbitrary @waiter to the wait queue for the nominated @page.
1166 void add_page_wait_queue(struct page *page, wait_queue_entry_t *waiter)
1168 wait_queue_head_t *q = page_waitqueue(page);
1169 unsigned long flags;
1171 spin_lock_irqsave(&q->lock, flags);
1172 __add_wait_queue_entry_tail(q, waiter);
1173 SetPageWaiters(page);
1174 spin_unlock_irqrestore(&q->lock, flags);
1176 EXPORT_SYMBOL_GPL(add_page_wait_queue);
1178 #ifndef clear_bit_unlock_is_negative_byte
1181 * PG_waiters is the high bit in the same byte as PG_lock.
1183 * On x86 (and on many other architectures), we can clear PG_lock and
1184 * test the sign bit at the same time. But if the architecture does
1185 * not support that special operation, we just do this all by hand
1188 * The read of PG_waiters has to be after (or concurrently with) PG_locked
1189 * being cleared, but a memory barrier should be unneccssary since it is
1190 * in the same byte as PG_locked.
1192 static inline bool clear_bit_unlock_is_negative_byte(long nr, volatile void *mem)
1194 clear_bit_unlock(nr, mem);
1195 /* smp_mb__after_atomic(); */
1196 return test_bit(PG_waiters, mem);
1202 * unlock_page - unlock a locked page
1205 * Unlocks the page and wakes up sleepers in ___wait_on_page_locked().
1206 * Also wakes sleepers in wait_on_page_writeback() because the wakeup
1207 * mechanism between PageLocked pages and PageWriteback pages is shared.
1208 * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep.
1210 * Note that this depends on PG_waiters being the sign bit in the byte
1211 * that contains PG_locked - thus the BUILD_BUG_ON(). That allows us to
1212 * clear the PG_locked bit and test PG_waiters at the same time fairly
1213 * portably (architectures that do LL/SC can test any bit, while x86 can
1214 * test the sign bit).
1216 void unlock_page(struct page *page)
1218 BUILD_BUG_ON(PG_waiters != 7);
1219 page = compound_head(page);
1220 VM_BUG_ON_PAGE(!PageLocked(page), page);
1221 if (clear_bit_unlock_is_negative_byte(PG_locked, &page->flags))
1222 wake_up_page_bit(page, PG_locked);
1224 EXPORT_SYMBOL(unlock_page);
1227 * end_page_writeback - end writeback against a page
1230 void end_page_writeback(struct page *page)
1233 * TestClearPageReclaim could be used here but it is an atomic
1234 * operation and overkill in this particular case. Failing to
1235 * shuffle a page marked for immediate reclaim is too mild to
1236 * justify taking an atomic operation penalty at the end of
1237 * ever page writeback.
1239 if (PageReclaim(page)) {
1240 ClearPageReclaim(page);
1241 rotate_reclaimable_page(page);
1244 if (!test_clear_page_writeback(page))
1247 smp_mb__after_atomic();
1248 wake_up_page(page, PG_writeback);
1250 EXPORT_SYMBOL(end_page_writeback);
1253 * After completing I/O on a page, call this routine to update the page
1254 * flags appropriately
1256 void page_endio(struct page *page, bool is_write, int err)
1260 SetPageUptodate(page);
1262 ClearPageUptodate(page);
1268 struct address_space *mapping;
1271 mapping = page_mapping(page);
1273 mapping_set_error(mapping, err);
1275 end_page_writeback(page);
1278 EXPORT_SYMBOL_GPL(page_endio);
1281 * __lock_page - get a lock on the page, assuming we need to sleep to get it
1282 * @__page: the page to lock
1284 void __lock_page(struct page *__page)
1286 struct page *page = compound_head(__page);
1287 wait_queue_head_t *q = page_waitqueue(page);
1288 wait_on_page_bit_common(q, page, PG_locked, TASK_UNINTERRUPTIBLE, true);
1290 EXPORT_SYMBOL(__lock_page);
1292 int __lock_page_killable(struct page *__page)
1294 struct page *page = compound_head(__page);
1295 wait_queue_head_t *q = page_waitqueue(page);
1296 return wait_on_page_bit_common(q, page, PG_locked, TASK_KILLABLE, true);
1298 EXPORT_SYMBOL_GPL(__lock_page_killable);
1302 * 1 - page is locked; mmap_sem is still held.
1303 * 0 - page is not locked.
1304 * mmap_sem has been released (up_read()), unless flags had both
1305 * FAULT_FLAG_ALLOW_RETRY and FAULT_FLAG_RETRY_NOWAIT set, in
1306 * which case mmap_sem is still held.
1308 * If neither ALLOW_RETRY nor KILLABLE are set, will always return 1
1309 * with the page locked and the mmap_sem unperturbed.
1311 int __lock_page_or_retry(struct page *page, struct mm_struct *mm,
1314 if (flags & FAULT_FLAG_ALLOW_RETRY) {
1316 * CAUTION! In this case, mmap_sem is not released
1317 * even though return 0.
1319 if (flags & FAULT_FLAG_RETRY_NOWAIT)
1322 up_read(&mm->mmap_sem);
1323 if (flags & FAULT_FLAG_KILLABLE)
1324 wait_on_page_locked_killable(page);
1326 wait_on_page_locked(page);
1329 if (flags & FAULT_FLAG_KILLABLE) {
1332 ret = __lock_page_killable(page);
1334 up_read(&mm->mmap_sem);
1344 * page_cache_next_hole - find the next hole (not-present entry)
1347 * @max_scan: maximum range to search
1349 * Search the set [index, min(index+max_scan-1, MAX_INDEX)] for the
1350 * lowest indexed hole.
1352 * Returns: the index of the hole if found, otherwise returns an index
1353 * outside of the set specified (in which case 'return - index >=
1354 * max_scan' will be true). In rare cases of index wrap-around, 0 will
1357 * page_cache_next_hole may be called under rcu_read_lock. However,
1358 * like radix_tree_gang_lookup, this will not atomically search a
1359 * snapshot of the tree at a single point in time. For example, if a
1360 * hole is created at index 5, then subsequently a hole is created at
1361 * index 10, page_cache_next_hole covering both indexes may return 10
1362 * if called under rcu_read_lock.
1364 pgoff_t page_cache_next_hole(struct address_space *mapping,
1365 pgoff_t index, unsigned long max_scan)
1369 for (i = 0; i < max_scan; i++) {
1372 page = radix_tree_lookup(&mapping->i_pages, index);
1373 if (!page || radix_tree_exceptional_entry(page))
1382 EXPORT_SYMBOL(page_cache_next_hole);
1385 * page_cache_prev_hole - find the prev hole (not-present entry)
1388 * @max_scan: maximum range to search
1390 * Search backwards in the range [max(index-max_scan+1, 0), index] for
1393 * Returns: the index of the hole if found, otherwise returns an index
1394 * outside of the set specified (in which case 'index - return >=
1395 * max_scan' will be true). In rare cases of wrap-around, ULONG_MAX
1398 * page_cache_prev_hole may be called under rcu_read_lock. However,
1399 * like radix_tree_gang_lookup, this will not atomically search a
1400 * snapshot of the tree at a single point in time. For example, if a
1401 * hole is created at index 10, then subsequently a hole is created at
1402 * index 5, page_cache_prev_hole covering both indexes may return 5 if
1403 * called under rcu_read_lock.
1405 pgoff_t page_cache_prev_hole(struct address_space *mapping,
1406 pgoff_t index, unsigned long max_scan)
1410 for (i = 0; i < max_scan; i++) {
1413 page = radix_tree_lookup(&mapping->i_pages, index);
1414 if (!page || radix_tree_exceptional_entry(page))
1417 if (index == ULONG_MAX)
1423 EXPORT_SYMBOL(page_cache_prev_hole);
1426 * find_get_entry - find and get a page cache entry
1427 * @mapping: the address_space to search
1428 * @offset: the page cache index
1430 * Looks up the page cache slot at @mapping & @offset. If there is a
1431 * page cache page, it is returned with an increased refcount.
1433 * If the slot holds a shadow entry of a previously evicted page, or a
1434 * swap entry from shmem/tmpfs, it is returned.
1436 * Otherwise, %NULL is returned.
1438 struct page *find_get_entry(struct address_space *mapping, pgoff_t offset)
1441 struct page *head, *page;
1446 pagep = radix_tree_lookup_slot(&mapping->i_pages, offset);
1448 page = radix_tree_deref_slot(pagep);
1449 if (unlikely(!page))
1451 if (radix_tree_exception(page)) {
1452 if (radix_tree_deref_retry(page))
1455 * A shadow entry of a recently evicted page,
1456 * or a swap entry from shmem/tmpfs. Return
1457 * it without attempting to raise page count.
1462 head = compound_head(page);
1463 if (!page_cache_get_speculative(head))
1466 /* The page was split under us? */
1467 if (compound_head(page) != head) {
1473 * Has the page moved?
1474 * This is part of the lockless pagecache protocol. See
1475 * include/linux/pagemap.h for details.
1477 if (unlikely(page != *pagep)) {
1487 EXPORT_SYMBOL(find_get_entry);
1490 * find_lock_entry - locate, pin and lock a page cache entry
1491 * @mapping: the address_space to search
1492 * @offset: the page cache index
1494 * Looks up the page cache slot at @mapping & @offset. If there is a
1495 * page cache page, it is returned locked and with an increased
1498 * If the slot holds a shadow entry of a previously evicted page, or a
1499 * swap entry from shmem/tmpfs, it is returned.
1501 * Otherwise, %NULL is returned.
1503 * find_lock_entry() may sleep.
1505 struct page *find_lock_entry(struct address_space *mapping, pgoff_t offset)
1510 page = find_get_entry(mapping, offset);
1511 if (page && !radix_tree_exception(page)) {
1513 /* Has the page been truncated? */
1514 if (unlikely(page_mapping(page) != mapping)) {
1519 VM_BUG_ON_PAGE(page_to_pgoff(page) != offset, page);
1523 EXPORT_SYMBOL(find_lock_entry);
1526 * pagecache_get_page - find and get a page reference
1527 * @mapping: the address_space to search
1528 * @offset: the page index
1529 * @fgp_flags: PCG flags
1530 * @gfp_mask: gfp mask to use for the page cache data page allocation
1532 * Looks up the page cache slot at @mapping & @offset.
1534 * PCG flags modify how the page is returned.
1536 * @fgp_flags can be:
1538 * - FGP_ACCESSED: the page will be marked accessed
1539 * - FGP_LOCK: Page is return locked
1540 * - FGP_CREAT: If page is not present then a new page is allocated using
1541 * @gfp_mask and added to the page cache and the VM's LRU
1542 * list. The page is returned locked and with an increased
1543 * refcount. Otherwise, NULL is returned.
1545 * If FGP_LOCK or FGP_CREAT are specified then the function may sleep even
1546 * if the GFP flags specified for FGP_CREAT are atomic.
1548 * If there is a page cache page, it is returned with an increased refcount.
1550 struct page *pagecache_get_page(struct address_space *mapping, pgoff_t offset,
1551 int fgp_flags, gfp_t gfp_mask)
1556 page = find_get_entry(mapping, offset);
1557 if (radix_tree_exceptional_entry(page))
1562 if (fgp_flags & FGP_LOCK) {
1563 if (fgp_flags & FGP_NOWAIT) {
1564 if (!trylock_page(page)) {
1572 /* Has the page been truncated? */
1573 if (unlikely(page->mapping != mapping)) {
1578 VM_BUG_ON_PAGE(page->index != offset, page);
1581 if (page && (fgp_flags & FGP_ACCESSED))
1582 mark_page_accessed(page);
1585 if (!page && (fgp_flags & FGP_CREAT)) {
1587 if ((fgp_flags & FGP_WRITE) && mapping_cap_account_dirty(mapping))
1588 gfp_mask |= __GFP_WRITE;
1589 if (fgp_flags & FGP_NOFS)
1590 gfp_mask &= ~__GFP_FS;
1592 page = __page_cache_alloc(gfp_mask);
1596 if (WARN_ON_ONCE(!(fgp_flags & FGP_LOCK)))
1597 fgp_flags |= FGP_LOCK;
1599 /* Init accessed so avoid atomic mark_page_accessed later */
1600 if (fgp_flags & FGP_ACCESSED)
1601 __SetPageReferenced(page);
1603 err = add_to_page_cache_lru(page, mapping, offset, gfp_mask);
1604 if (unlikely(err)) {
1614 EXPORT_SYMBOL(pagecache_get_page);
1617 * find_get_entries - gang pagecache lookup
1618 * @mapping: The address_space to search
1619 * @start: The starting page cache index
1620 * @nr_entries: The maximum number of entries
1621 * @entries: Where the resulting entries are placed
1622 * @indices: The cache indices corresponding to the entries in @entries
1624 * find_get_entries() will search for and return a group of up to
1625 * @nr_entries entries in the mapping. The entries are placed at
1626 * @entries. find_get_entries() takes a reference against any actual
1629 * The search returns a group of mapping-contiguous page cache entries
1630 * with ascending indexes. There may be holes in the indices due to
1631 * not-present pages.
1633 * Any shadow entries of evicted pages, or swap entries from
1634 * shmem/tmpfs, are included in the returned array.
1636 * find_get_entries() returns the number of pages and shadow entries
1639 unsigned find_get_entries(struct address_space *mapping,
1640 pgoff_t start, unsigned int nr_entries,
1641 struct page **entries, pgoff_t *indices)
1644 unsigned int ret = 0;
1645 struct radix_tree_iter iter;
1651 radix_tree_for_each_slot(slot, &mapping->i_pages, &iter, start) {
1652 struct page *head, *page;
1654 page = radix_tree_deref_slot(slot);
1655 if (unlikely(!page))
1657 if (radix_tree_exception(page)) {
1658 if (radix_tree_deref_retry(page)) {
1659 slot = radix_tree_iter_retry(&iter);
1663 * A shadow entry of a recently evicted page, a swap
1664 * entry from shmem/tmpfs or a DAX entry. Return it
1665 * without attempting to raise page count.
1670 head = compound_head(page);
1671 if (!page_cache_get_speculative(head))
1674 /* The page was split under us? */
1675 if (compound_head(page) != head) {
1680 /* Has the page moved? */
1681 if (unlikely(page != *slot)) {
1686 indices[ret] = iter.index;
1687 entries[ret] = page;
1688 if (++ret == nr_entries)
1696 * find_get_pages_range - gang pagecache lookup
1697 * @mapping: The address_space to search
1698 * @start: The starting page index
1699 * @end: The final page index (inclusive)
1700 * @nr_pages: The maximum number of pages
1701 * @pages: Where the resulting pages are placed
1703 * find_get_pages_range() will search for and return a group of up to @nr_pages
1704 * pages in the mapping starting at index @start and up to index @end
1705 * (inclusive). The pages are placed at @pages. find_get_pages_range() takes
1706 * a reference against the returned pages.
1708 * The search returns a group of mapping-contiguous pages with ascending
1709 * indexes. There may be holes in the indices due to not-present pages.
1710 * We also update @start to index the next page for the traversal.
1712 * find_get_pages_range() returns the number of pages which were found. If this
1713 * number is smaller than @nr_pages, the end of specified range has been
1716 unsigned find_get_pages_range(struct address_space *mapping, pgoff_t *start,
1717 pgoff_t end, unsigned int nr_pages,
1718 struct page **pages)
1720 struct radix_tree_iter iter;
1724 if (unlikely(!nr_pages))
1728 radix_tree_for_each_slot(slot, &mapping->i_pages, &iter, *start) {
1729 struct page *head, *page;
1731 if (iter.index > end)
1734 page = radix_tree_deref_slot(slot);
1735 if (unlikely(!page))
1738 if (radix_tree_exception(page)) {
1739 if (radix_tree_deref_retry(page)) {
1740 slot = radix_tree_iter_retry(&iter);
1744 * A shadow entry of a recently evicted page,
1745 * or a swap entry from shmem/tmpfs. Skip
1751 head = compound_head(page);
1752 if (!page_cache_get_speculative(head))
1755 /* The page was split under us? */
1756 if (compound_head(page) != head) {
1761 /* Has the page moved? */
1762 if (unlikely(page != *slot)) {
1768 if (++ret == nr_pages) {
1769 *start = pages[ret - 1]->index + 1;
1775 * We come here when there is no page beyond @end. We take care to not
1776 * overflow the index @start as it confuses some of the callers. This
1777 * breaks the iteration when there is page at index -1 but that is
1778 * already broken anyway.
1780 if (end == (pgoff_t)-1)
1781 *start = (pgoff_t)-1;
1791 * find_get_pages_contig - gang contiguous pagecache lookup
1792 * @mapping: The address_space to search
1793 * @index: The starting page index
1794 * @nr_pages: The maximum number of pages
1795 * @pages: Where the resulting pages are placed
1797 * find_get_pages_contig() works exactly like find_get_pages(), except
1798 * that the returned number of pages are guaranteed to be contiguous.
1800 * find_get_pages_contig() returns the number of pages which were found.
1802 unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index,
1803 unsigned int nr_pages, struct page **pages)
1805 struct radix_tree_iter iter;
1807 unsigned int ret = 0;
1809 if (unlikely(!nr_pages))
1813 radix_tree_for_each_contig(slot, &mapping->i_pages, &iter, index) {
1814 struct page *head, *page;
1816 page = radix_tree_deref_slot(slot);
1817 /* The hole, there no reason to continue */
1818 if (unlikely(!page))
1821 if (radix_tree_exception(page)) {
1822 if (radix_tree_deref_retry(page)) {
1823 slot = radix_tree_iter_retry(&iter);
1827 * A shadow entry of a recently evicted page,
1828 * or a swap entry from shmem/tmpfs. Stop
1829 * looking for contiguous pages.
1834 head = compound_head(page);
1835 if (!page_cache_get_speculative(head))
1838 /* The page was split under us? */
1839 if (compound_head(page) != head) {
1844 /* Has the page moved? */
1845 if (unlikely(page != *slot)) {
1851 * must check mapping and index after taking the ref.
1852 * otherwise we can get both false positives and false
1853 * negatives, which is just confusing to the caller.
1855 if (page->mapping == NULL || page_to_pgoff(page) != iter.index) {
1861 if (++ret == nr_pages)
1867 EXPORT_SYMBOL(find_get_pages_contig);
1870 * find_get_pages_range_tag - find and return pages in given range matching @tag
1871 * @mapping: the address_space to search
1872 * @index: the starting page index
1873 * @end: The final page index (inclusive)
1874 * @tag: the tag index
1875 * @nr_pages: the maximum number of pages
1876 * @pages: where the resulting pages are placed
1878 * Like find_get_pages, except we only return pages which are tagged with
1879 * @tag. We update @index to index the next page for the traversal.
1881 unsigned find_get_pages_range_tag(struct address_space *mapping, pgoff_t *index,
1882 pgoff_t end, int tag, unsigned int nr_pages,
1883 struct page **pages)
1885 struct radix_tree_iter iter;
1889 if (unlikely(!nr_pages))
1893 radix_tree_for_each_tagged(slot, &mapping->i_pages, &iter, *index, tag) {
1894 struct page *head, *page;
1896 if (iter.index > end)
1899 page = radix_tree_deref_slot(slot);
1900 if (unlikely(!page))
1903 if (radix_tree_exception(page)) {
1904 if (radix_tree_deref_retry(page)) {
1905 slot = radix_tree_iter_retry(&iter);
1909 * A shadow entry of a recently evicted page.
1911 * Those entries should never be tagged, but
1912 * this tree walk is lockless and the tags are
1913 * looked up in bulk, one radix tree node at a
1914 * time, so there is a sizable window for page
1915 * reclaim to evict a page we saw tagged.
1922 head = compound_head(page);
1923 if (!page_cache_get_speculative(head))
1926 /* The page was split under us? */
1927 if (compound_head(page) != head) {
1932 /* Has the page moved? */
1933 if (unlikely(page != *slot)) {
1939 if (++ret == nr_pages) {
1940 *index = pages[ret - 1]->index + 1;
1946 * We come here when we got at @end. We take care to not overflow the
1947 * index @index as it confuses some of the callers. This breaks the
1948 * iteration when there is page at index -1 but that is already broken
1951 if (end == (pgoff_t)-1)
1952 *index = (pgoff_t)-1;
1960 EXPORT_SYMBOL(find_get_pages_range_tag);
1963 * find_get_entries_tag - find and return entries that match @tag
1964 * @mapping: the address_space to search
1965 * @start: the starting page cache index
1966 * @tag: the tag index
1967 * @nr_entries: the maximum number of entries
1968 * @entries: where the resulting entries are placed
1969 * @indices: the cache indices corresponding to the entries in @entries
1971 * Like find_get_entries, except we only return entries which are tagged with
1974 unsigned find_get_entries_tag(struct address_space *mapping, pgoff_t start,
1975 int tag, unsigned int nr_entries,
1976 struct page **entries, pgoff_t *indices)
1979 unsigned int ret = 0;
1980 struct radix_tree_iter iter;
1986 radix_tree_for_each_tagged(slot, &mapping->i_pages, &iter, start, tag) {
1987 struct page *head, *page;
1989 page = radix_tree_deref_slot(slot);
1990 if (unlikely(!page))
1992 if (radix_tree_exception(page)) {
1993 if (radix_tree_deref_retry(page)) {
1994 slot = radix_tree_iter_retry(&iter);
1999 * A shadow entry of a recently evicted page, a swap
2000 * entry from shmem/tmpfs or a DAX entry. Return it
2001 * without attempting to raise page count.
2006 head = compound_head(page);
2007 if (!page_cache_get_speculative(head))
2010 /* The page was split under us? */
2011 if (compound_head(page) != head) {
2016 /* Has the page moved? */
2017 if (unlikely(page != *slot)) {
2022 indices[ret] = iter.index;
2023 entries[ret] = page;
2024 if (++ret == nr_entries)
2030 EXPORT_SYMBOL(find_get_entries_tag);
2033 * CD/DVDs are error prone. When a medium error occurs, the driver may fail
2034 * a _large_ part of the i/o request. Imagine the worst scenario:
2036 * ---R__________________________________________B__________
2037 * ^ reading here ^ bad block(assume 4k)
2039 * read(R) => miss => readahead(R...B) => media error => frustrating retries
2040 * => failing the whole request => read(R) => read(R+1) =>
2041 * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
2042 * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
2043 * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
2045 * It is going insane. Fix it by quickly scaling down the readahead size.
2047 static void shrink_readahead_size_eio(struct file *filp,
2048 struct file_ra_state *ra)
2054 * generic_file_buffered_read - generic file read routine
2055 * @iocb: the iocb to read
2056 * @iter: data destination
2057 * @written: already copied
2059 * This is a generic file read routine, and uses the
2060 * mapping->a_ops->readpage() function for the actual low-level stuff.
2062 * This is really ugly. But the goto's actually try to clarify some
2063 * of the logic when it comes to error handling etc.
2065 static ssize_t generic_file_buffered_read(struct kiocb *iocb,
2066 struct iov_iter *iter, ssize_t written)
2068 struct file *filp = iocb->ki_filp;
2069 struct address_space *mapping = filp->f_mapping;
2070 struct inode *inode = mapping->host;
2071 struct file_ra_state *ra = &filp->f_ra;
2072 loff_t *ppos = &iocb->ki_pos;
2076 unsigned long offset; /* offset into pagecache page */
2077 unsigned int prev_offset;
2080 if (unlikely(*ppos >= inode->i_sb->s_maxbytes))
2082 iov_iter_truncate(iter, inode->i_sb->s_maxbytes);
2084 index = *ppos >> PAGE_SHIFT;
2085 prev_index = ra->prev_pos >> PAGE_SHIFT;
2086 prev_offset = ra->prev_pos & (PAGE_SIZE-1);
2087 last_index = (*ppos + iter->count + PAGE_SIZE-1) >> PAGE_SHIFT;
2088 offset = *ppos & ~PAGE_MASK;
2094 unsigned long nr, ret;
2098 if (fatal_signal_pending(current)) {
2103 page = find_get_page(mapping, index);
2105 if (iocb->ki_flags & IOCB_NOWAIT)
2107 page_cache_sync_readahead(mapping,
2109 index, last_index - index);
2110 page = find_get_page(mapping, index);
2111 if (unlikely(page == NULL))
2112 goto no_cached_page;
2114 if (PageReadahead(page)) {
2115 page_cache_async_readahead(mapping,
2117 index, last_index - index);
2119 if (!PageUptodate(page)) {
2120 if (iocb->ki_flags & IOCB_NOWAIT) {
2126 * See comment in do_read_cache_page on why
2127 * wait_on_page_locked is used to avoid unnecessarily
2128 * serialisations and why it's safe.
2130 error = wait_on_page_locked_killable(page);
2131 if (unlikely(error))
2132 goto readpage_error;
2133 if (PageUptodate(page))
2136 if (inode->i_blkbits == PAGE_SHIFT ||
2137 !mapping->a_ops->is_partially_uptodate)
2138 goto page_not_up_to_date;
2139 /* pipes can't handle partially uptodate pages */
2140 if (unlikely(iter->type & ITER_PIPE))
2141 goto page_not_up_to_date;
2142 if (!trylock_page(page))
2143 goto page_not_up_to_date;
2144 /* Did it get truncated before we got the lock? */
2146 goto page_not_up_to_date_locked;
2147 if (!mapping->a_ops->is_partially_uptodate(page,
2148 offset, iter->count))
2149 goto page_not_up_to_date_locked;
2154 * i_size must be checked after we know the page is Uptodate.
2156 * Checking i_size after the check allows us to calculate
2157 * the correct value for "nr", which means the zero-filled
2158 * part of the page is not copied back to userspace (unless
2159 * another truncate extends the file - this is desired though).
2162 isize = i_size_read(inode);
2163 end_index = (isize - 1) >> PAGE_SHIFT;
2164 if (unlikely(!isize || index > end_index)) {
2169 /* nr is the maximum number of bytes to copy from this page */
2171 if (index == end_index) {
2172 nr = ((isize - 1) & ~PAGE_MASK) + 1;
2180 /* If users can be writing to this page using arbitrary
2181 * virtual addresses, take care about potential aliasing
2182 * before reading the page on the kernel side.
2184 if (mapping_writably_mapped(mapping))
2185 flush_dcache_page(page);
2188 * When a sequential read accesses a page several times,
2189 * only mark it as accessed the first time.
2191 if (prev_index != index || offset != prev_offset)
2192 mark_page_accessed(page);
2196 * Ok, we have the page, and it's up-to-date, so
2197 * now we can copy it to user space...
2200 ret = copy_page_to_iter(page, offset, nr, iter);
2202 index += offset >> PAGE_SHIFT;
2203 offset &= ~PAGE_MASK;
2204 prev_offset = offset;
2208 if (!iov_iter_count(iter))
2216 page_not_up_to_date:
2217 /* Get exclusive access to the page ... */
2218 error = lock_page_killable(page);
2219 if (unlikely(error))
2220 goto readpage_error;
2222 page_not_up_to_date_locked:
2223 /* Did it get truncated before we got the lock? */
2224 if (!page->mapping) {
2230 /* Did somebody else fill it already? */
2231 if (PageUptodate(page)) {
2238 * A previous I/O error may have been due to temporary
2239 * failures, eg. multipath errors.
2240 * PG_error will be set again if readpage fails.
2242 ClearPageError(page);
2243 /* Start the actual read. The read will unlock the page. */
2244 error = mapping->a_ops->readpage(filp, page);
2246 if (unlikely(error)) {
2247 if (error == AOP_TRUNCATED_PAGE) {
2252 goto readpage_error;
2255 if (!PageUptodate(page)) {
2256 error = lock_page_killable(page);
2257 if (unlikely(error))
2258 goto readpage_error;
2259 if (!PageUptodate(page)) {
2260 if (page->mapping == NULL) {
2262 * invalidate_mapping_pages got it
2269 shrink_readahead_size_eio(filp, ra);
2271 goto readpage_error;
2279 /* UHHUH! A synchronous read error occurred. Report it */
2285 * Ok, it wasn't cached, so we need to create a new
2288 page = page_cache_alloc(mapping);
2293 error = add_to_page_cache_lru(page, mapping, index,
2294 mapping_gfp_constraint(mapping, GFP_KERNEL));
2297 if (error == -EEXIST) {
2309 ra->prev_pos = prev_index;
2310 ra->prev_pos <<= PAGE_SHIFT;
2311 ra->prev_pos |= prev_offset;
2313 *ppos = ((loff_t)index << PAGE_SHIFT) + offset;
2314 file_accessed(filp);
2315 return written ? written : error;
2319 * generic_file_read_iter - generic filesystem read routine
2320 * @iocb: kernel I/O control block
2321 * @iter: destination for the data read
2323 * This is the "read_iter()" routine for all filesystems
2324 * that can use the page cache directly.
2327 generic_file_read_iter(struct kiocb *iocb, struct iov_iter *iter)
2329 size_t count = iov_iter_count(iter);
2333 goto out; /* skip atime */
2335 if (iocb->ki_flags & IOCB_DIRECT) {
2336 struct file *file = iocb->ki_filp;
2337 struct address_space *mapping = file->f_mapping;
2338 struct inode *inode = mapping->host;
2341 size = i_size_read(inode);
2342 if (iocb->ki_flags & IOCB_NOWAIT) {
2343 if (filemap_range_has_page(mapping, iocb->ki_pos,
2344 iocb->ki_pos + count - 1))
2347 retval = filemap_write_and_wait_range(mapping,
2349 iocb->ki_pos + count - 1);
2354 file_accessed(file);
2356 retval = mapping->a_ops->direct_IO(iocb, iter);
2358 iocb->ki_pos += retval;
2361 iov_iter_revert(iter, count - iov_iter_count(iter));
2364 * Btrfs can have a short DIO read if we encounter
2365 * compressed extents, so if there was an error, or if
2366 * we've already read everything we wanted to, or if
2367 * there was a short read because we hit EOF, go ahead
2368 * and return. Otherwise fallthrough to buffered io for
2369 * the rest of the read. Buffered reads will not work for
2370 * DAX files, so don't bother trying.
2372 if (retval < 0 || !count || iocb->ki_pos >= size ||
2377 retval = generic_file_buffered_read(iocb, iter, retval);
2381 EXPORT_SYMBOL(generic_file_read_iter);
2385 * page_cache_read - adds requested page to the page cache if not already there
2386 * @file: file to read
2387 * @offset: page index
2388 * @gfp_mask: memory allocation flags
2390 * This adds the requested page to the page cache if it isn't already there,
2391 * and schedules an I/O to read in its contents from disk.
2393 static int page_cache_read(struct file *file, pgoff_t offset, gfp_t gfp_mask)
2395 struct address_space *mapping = file->f_mapping;
2400 page = __page_cache_alloc(gfp_mask);
2404 ret = add_to_page_cache_lru(page, mapping, offset, gfp_mask);
2406 ret = mapping->a_ops->readpage(file, page);
2407 else if (ret == -EEXIST)
2408 ret = 0; /* losing race to add is OK */
2412 } while (ret == AOP_TRUNCATED_PAGE);
2417 #define MMAP_LOTSAMISS (100)
2420 * Synchronous readahead happens when we don't even find
2421 * a page in the page cache at all.
2423 static void do_sync_mmap_readahead(struct vm_area_struct *vma,
2424 struct file_ra_state *ra,
2428 struct address_space *mapping = file->f_mapping;
2430 /* If we don't want any read-ahead, don't bother */
2431 if (vma->vm_flags & VM_RAND_READ)
2436 if (vma->vm_flags & VM_SEQ_READ) {
2437 page_cache_sync_readahead(mapping, ra, file, offset,
2442 /* Avoid banging the cache line if not needed */
2443 if (ra->mmap_miss < MMAP_LOTSAMISS * 10)
2447 * Do we miss much more than hit in this file? If so,
2448 * stop bothering with read-ahead. It will only hurt.
2450 if (ra->mmap_miss > MMAP_LOTSAMISS)
2456 ra->start = max_t(long, 0, offset - ra->ra_pages / 2);
2457 ra->size = ra->ra_pages;
2458 ra->async_size = ra->ra_pages / 4;
2459 ra_submit(ra, mapping, file);
2463 * Asynchronous readahead happens when we find the page and PG_readahead,
2464 * so we want to possibly extend the readahead further..
2466 static void do_async_mmap_readahead(struct vm_area_struct *vma,
2467 struct file_ra_state *ra,
2472 struct address_space *mapping = file->f_mapping;
2474 /* If we don't want any read-ahead, don't bother */
2475 if (vma->vm_flags & VM_RAND_READ)
2477 if (ra->mmap_miss > 0)
2479 if (PageReadahead(page))
2480 page_cache_async_readahead(mapping, ra, file,
2481 page, offset, ra->ra_pages);
2485 * filemap_fault - read in file data for page fault handling
2486 * @vmf: struct vm_fault containing details of the fault
2488 * filemap_fault() is invoked via the vma operations vector for a
2489 * mapped memory region to read in file data during a page fault.
2491 * The goto's are kind of ugly, but this streamlines the normal case of having
2492 * it in the page cache, and handles the special cases reasonably without
2493 * having a lot of duplicated code.
2495 * vma->vm_mm->mmap_sem must be held on entry.
2497 * If our return value has VM_FAULT_RETRY set, it's because
2498 * lock_page_or_retry() returned 0.
2499 * The mmap_sem has usually been released in this case.
2500 * See __lock_page_or_retry() for the exception.
2502 * If our return value does not have VM_FAULT_RETRY set, the mmap_sem
2503 * has not been released.
2505 * We never return with VM_FAULT_RETRY and a bit from VM_FAULT_ERROR set.
2507 vm_fault_t filemap_fault(struct vm_fault *vmf)
2510 struct file *file = vmf->vma->vm_file;
2511 struct address_space *mapping = file->f_mapping;
2512 struct file_ra_state *ra = &file->f_ra;
2513 struct inode *inode = mapping->host;
2514 pgoff_t offset = vmf->pgoff;
2519 max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
2520 if (unlikely(offset >= max_off))
2521 return VM_FAULT_SIGBUS;
2524 * Do we have something in the page cache already?
2526 page = find_get_page(mapping, offset);
2527 if (likely(page) && !(vmf->flags & FAULT_FLAG_TRIED)) {
2529 * We found the page, so try async readahead before
2530 * waiting for the lock.
2532 do_async_mmap_readahead(vmf->vma, ra, file, page, offset);
2534 /* No page in the page cache at all */
2535 do_sync_mmap_readahead(vmf->vma, ra, file, offset);
2536 count_vm_event(PGMAJFAULT);
2537 count_memcg_event_mm(vmf->vma->vm_mm, PGMAJFAULT);
2538 ret = VM_FAULT_MAJOR;
2540 page = find_get_page(mapping, offset);
2542 goto no_cached_page;
2545 if (!lock_page_or_retry(page, vmf->vma->vm_mm, vmf->flags)) {
2547 return ret | VM_FAULT_RETRY;
2550 /* Did it get truncated? */
2551 if (unlikely(page->mapping != mapping)) {
2556 VM_BUG_ON_PAGE(page->index != offset, page);
2559 * We have a locked page in the page cache, now we need to check
2560 * that it's up-to-date. If not, it is going to be due to an error.
2562 if (unlikely(!PageUptodate(page)))
2563 goto page_not_uptodate;
2566 * Found the page and have a reference on it.
2567 * We must recheck i_size under page lock.
2569 max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
2570 if (unlikely(offset >= max_off)) {
2573 return VM_FAULT_SIGBUS;
2577 return ret | VM_FAULT_LOCKED;
2581 * We're only likely to ever get here if MADV_RANDOM is in
2584 error = page_cache_read(file, offset, vmf->gfp_mask);
2587 * The page we want has now been added to the page cache.
2588 * In the unlikely event that someone removed it in the
2589 * meantime, we'll just come back here and read it again.
2595 * An error return from page_cache_read can result if the
2596 * system is low on memory, or a problem occurs while trying
2599 return vmf_error(error);
2603 * Umm, take care of errors if the page isn't up-to-date.
2604 * Try to re-read it _once_. We do this synchronously,
2605 * because there really aren't any performance issues here
2606 * and we need to check for errors.
2608 ClearPageError(page);
2609 error = mapping->a_ops->readpage(file, page);
2611 wait_on_page_locked(page);
2612 if (!PageUptodate(page))
2617 if (!error || error == AOP_TRUNCATED_PAGE)
2620 /* Things didn't work out. Return zero to tell the mm layer so. */
2621 shrink_readahead_size_eio(file, ra);
2622 return VM_FAULT_SIGBUS;
2624 EXPORT_SYMBOL(filemap_fault);
2626 void filemap_map_pages(struct vm_fault *vmf,
2627 pgoff_t start_pgoff, pgoff_t end_pgoff)
2629 struct radix_tree_iter iter;
2631 struct file *file = vmf->vma->vm_file;
2632 struct address_space *mapping = file->f_mapping;
2633 pgoff_t last_pgoff = start_pgoff;
2634 unsigned long max_idx;
2635 struct page *head, *page;
2638 radix_tree_for_each_slot(slot, &mapping->i_pages, &iter, start_pgoff) {
2639 if (iter.index > end_pgoff)
2642 page = radix_tree_deref_slot(slot);
2643 if (unlikely(!page))
2645 if (radix_tree_exception(page)) {
2646 if (radix_tree_deref_retry(page)) {
2647 slot = radix_tree_iter_retry(&iter);
2653 head = compound_head(page);
2654 if (!page_cache_get_speculative(head))
2657 /* The page was split under us? */
2658 if (compound_head(page) != head) {
2663 /* Has the page moved? */
2664 if (unlikely(page != *slot)) {
2669 if (!PageUptodate(page) ||
2670 PageReadahead(page) ||
2673 if (!trylock_page(page))
2676 if (page->mapping != mapping || !PageUptodate(page))
2679 max_idx = DIV_ROUND_UP(i_size_read(mapping->host), PAGE_SIZE);
2680 if (page->index >= max_idx)
2683 if (file->f_ra.mmap_miss > 0)
2684 file->f_ra.mmap_miss--;
2686 vmf->address += (iter.index - last_pgoff) << PAGE_SHIFT;
2688 vmf->pte += iter.index - last_pgoff;
2689 last_pgoff = iter.index;
2690 if (alloc_set_pte(vmf, NULL, page))
2699 /* Huge page is mapped? No need to proceed. */
2700 if (pmd_trans_huge(*vmf->pmd))
2702 if (iter.index == end_pgoff)
2707 EXPORT_SYMBOL(filemap_map_pages);
2709 vm_fault_t filemap_page_mkwrite(struct vm_fault *vmf)
2711 struct page *page = vmf->page;
2712 struct inode *inode = file_inode(vmf->vma->vm_file);
2713 vm_fault_t ret = VM_FAULT_LOCKED;
2715 sb_start_pagefault(inode->i_sb);
2716 file_update_time(vmf->vma->vm_file);
2718 if (page->mapping != inode->i_mapping) {
2720 ret = VM_FAULT_NOPAGE;
2724 * We mark the page dirty already here so that when freeze is in
2725 * progress, we are guaranteed that writeback during freezing will
2726 * see the dirty page and writeprotect it again.
2728 set_page_dirty(page);
2729 wait_for_stable_page(page);
2731 sb_end_pagefault(inode->i_sb);
2735 const struct vm_operations_struct generic_file_vm_ops = {
2736 .fault = filemap_fault,
2737 .map_pages = filemap_map_pages,
2738 .page_mkwrite = filemap_page_mkwrite,
2741 /* This is used for a general mmap of a disk file */
2743 int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
2745 struct address_space *mapping = file->f_mapping;
2747 if (!mapping->a_ops->readpage)
2749 file_accessed(file);
2750 vma->vm_ops = &generic_file_vm_ops;
2755 * This is for filesystems which do not implement ->writepage.
2757 int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
2759 if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE))
2761 return generic_file_mmap(file, vma);
2764 vm_fault_t filemap_page_mkwrite(struct vm_fault *vmf)
2766 return VM_FAULT_SIGBUS;
2768 int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
2772 int generic_file_readonly_mmap(struct file * file, struct vm_area_struct * vma)
2776 #endif /* CONFIG_MMU */
2778 EXPORT_SYMBOL(filemap_page_mkwrite);
2779 EXPORT_SYMBOL(generic_file_mmap);
2780 EXPORT_SYMBOL(generic_file_readonly_mmap);
2782 static struct page *wait_on_page_read(struct page *page)
2784 if (!IS_ERR(page)) {
2785 wait_on_page_locked(page);
2786 if (!PageUptodate(page)) {
2788 page = ERR_PTR(-EIO);
2794 static struct page *do_read_cache_page(struct address_space *mapping,
2796 int (*filler)(void *, struct page *),
2803 page = find_get_page(mapping, index);
2805 page = __page_cache_alloc(gfp);
2807 return ERR_PTR(-ENOMEM);
2808 err = add_to_page_cache_lru(page, mapping, index, gfp);
2809 if (unlikely(err)) {
2813 /* Presumably ENOMEM for radix tree node */
2814 return ERR_PTR(err);
2818 err = filler(data, page);
2821 return ERR_PTR(err);
2824 page = wait_on_page_read(page);
2829 if (PageUptodate(page))
2833 * Page is not up to date and may be locked due one of the following
2834 * case a: Page is being filled and the page lock is held
2835 * case b: Read/write error clearing the page uptodate status
2836 * case c: Truncation in progress (page locked)
2837 * case d: Reclaim in progress
2839 * Case a, the page will be up to date when the page is unlocked.
2840 * There is no need to serialise on the page lock here as the page
2841 * is pinned so the lock gives no additional protection. Even if the
2842 * the page is truncated, the data is still valid if PageUptodate as
2843 * it's a race vs truncate race.
2844 * Case b, the page will not be up to date
2845 * Case c, the page may be truncated but in itself, the data may still
2846 * be valid after IO completes as it's a read vs truncate race. The
2847 * operation must restart if the page is not uptodate on unlock but
2848 * otherwise serialising on page lock to stabilise the mapping gives
2849 * no additional guarantees to the caller as the page lock is
2850 * released before return.
2851 * Case d, similar to truncation. If reclaim holds the page lock, it
2852 * will be a race with remove_mapping that determines if the mapping
2853 * is valid on unlock but otherwise the data is valid and there is
2854 * no need to serialise with page lock.
2856 * As the page lock gives no additional guarantee, we optimistically
2857 * wait on the page to be unlocked and check if it's up to date and
2858 * use the page if it is. Otherwise, the page lock is required to
2859 * distinguish between the different cases. The motivation is that we
2860 * avoid spurious serialisations and wakeups when multiple processes
2861 * wait on the same page for IO to complete.
2863 wait_on_page_locked(page);
2864 if (PageUptodate(page))
2867 /* Distinguish between all the cases under the safety of the lock */
2870 /* Case c or d, restart the operation */
2871 if (!page->mapping) {
2877 /* Someone else locked and filled the page in a very small window */
2878 if (PageUptodate(page)) {
2885 mark_page_accessed(page);
2890 * read_cache_page - read into page cache, fill it if needed
2891 * @mapping: the page's address_space
2892 * @index: the page index
2893 * @filler: function to perform the read
2894 * @data: first arg to filler(data, page) function, often left as NULL
2896 * Read into the page cache. If a page already exists, and PageUptodate() is
2897 * not set, try to fill the page and wait for it to become unlocked.
2899 * If the page does not get brought uptodate, return -EIO.
2901 struct page *read_cache_page(struct address_space *mapping,
2903 int (*filler)(void *, struct page *),
2906 return do_read_cache_page(mapping, index, filler, data, mapping_gfp_mask(mapping));
2908 EXPORT_SYMBOL(read_cache_page);
2911 * read_cache_page_gfp - read into page cache, using specified page allocation flags.
2912 * @mapping: the page's address_space
2913 * @index: the page index
2914 * @gfp: the page allocator flags to use if allocating
2916 * This is the same as "read_mapping_page(mapping, index, NULL)", but with
2917 * any new page allocations done using the specified allocation flags.
2919 * If the page does not get brought uptodate, return -EIO.
2921 struct page *read_cache_page_gfp(struct address_space *mapping,
2925 filler_t *filler = (filler_t *)mapping->a_ops->readpage;
2927 return do_read_cache_page(mapping, index, filler, NULL, gfp);
2929 EXPORT_SYMBOL(read_cache_page_gfp);
2932 * Performs necessary checks before doing a write
2934 * Can adjust writing position or amount of bytes to write.
2935 * Returns appropriate error code that caller should return or
2936 * zero in case that write should be allowed.
2938 inline ssize_t generic_write_checks(struct kiocb *iocb, struct iov_iter *from)
2940 struct file *file = iocb->ki_filp;
2941 struct inode *inode = file->f_mapping->host;
2942 unsigned long limit = rlimit(RLIMIT_FSIZE);
2945 if (!iov_iter_count(from))
2948 /* FIXME: this is for backwards compatibility with 2.4 */
2949 if (iocb->ki_flags & IOCB_APPEND)
2950 iocb->ki_pos = i_size_read(inode);
2954 if ((iocb->ki_flags & IOCB_NOWAIT) && !(iocb->ki_flags & IOCB_DIRECT))
2957 if (limit != RLIM_INFINITY) {
2958 if (iocb->ki_pos >= limit) {
2959 send_sig(SIGXFSZ, current, 0);
2962 iov_iter_truncate(from, limit - (unsigned long)pos);
2968 if (unlikely(pos + iov_iter_count(from) > MAX_NON_LFS &&
2969 !(file->f_flags & O_LARGEFILE))) {
2970 if (pos >= MAX_NON_LFS)
2972 iov_iter_truncate(from, MAX_NON_LFS - (unsigned long)pos);
2976 * Are we about to exceed the fs block limit ?
2978 * If we have written data it becomes a short write. If we have
2979 * exceeded without writing data we send a signal and return EFBIG.
2980 * Linus frestrict idea will clean these up nicely..
2982 if (unlikely(pos >= inode->i_sb->s_maxbytes))
2985 iov_iter_truncate(from, inode->i_sb->s_maxbytes - pos);
2986 return iov_iter_count(from);
2988 EXPORT_SYMBOL(generic_write_checks);
2990 int pagecache_write_begin(struct file *file, struct address_space *mapping,
2991 loff_t pos, unsigned len, unsigned flags,
2992 struct page **pagep, void **fsdata)
2994 const struct address_space_operations *aops = mapping->a_ops;
2996 return aops->write_begin(file, mapping, pos, len, flags,
2999 EXPORT_SYMBOL(pagecache_write_begin);
3001 int pagecache_write_end(struct file *file, struct address_space *mapping,
3002 loff_t pos, unsigned len, unsigned copied,
3003 struct page *page, void *fsdata)
3005 const struct address_space_operations *aops = mapping->a_ops;
3007 return aops->write_end(file, mapping, pos, len, copied, page, fsdata);
3009 EXPORT_SYMBOL(pagecache_write_end);
3012 generic_file_direct_write(struct kiocb *iocb, struct iov_iter *from)
3014 struct file *file = iocb->ki_filp;
3015 struct address_space *mapping = file->f_mapping;
3016 struct inode *inode = mapping->host;
3017 loff_t pos = iocb->ki_pos;
3022 write_len = iov_iter_count(from);
3023 end = (pos + write_len - 1) >> PAGE_SHIFT;
3025 if (iocb->ki_flags & IOCB_NOWAIT) {
3026 /* If there are pages to writeback, return */
3027 if (filemap_range_has_page(inode->i_mapping, pos,
3031 written = filemap_write_and_wait_range(mapping, pos,
3032 pos + write_len - 1);
3038 * After a write we want buffered reads to be sure to go to disk to get
3039 * the new data. We invalidate clean cached page from the region we're
3040 * about to write. We do this *before* the write so that we can return
3041 * without clobbering -EIOCBQUEUED from ->direct_IO().
3043 written = invalidate_inode_pages2_range(mapping,
3044 pos >> PAGE_SHIFT, end);
3046 * If a page can not be invalidated, return 0 to fall back
3047 * to buffered write.
3050 if (written == -EBUSY)
3055 written = mapping->a_ops->direct_IO(iocb, from);
3058 * Finally, try again to invalidate clean pages which might have been
3059 * cached by non-direct readahead, or faulted in by get_user_pages()
3060 * if the source of the write was an mmap'ed region of the file
3061 * we're writing. Either one is a pretty crazy thing to do,
3062 * so we don't support it 100%. If this invalidation
3063 * fails, tough, the write still worked...
3065 * Most of the time we do not need this since dio_complete() will do
3066 * the invalidation for us. However there are some file systems that
3067 * do not end up with dio_complete() being called, so let's not break
3068 * them by removing it completely
3070 if (mapping->nrpages)
3071 invalidate_inode_pages2_range(mapping,
3072 pos >> PAGE_SHIFT, end);
3076 write_len -= written;
3077 if (pos > i_size_read(inode) && !S_ISBLK(inode->i_mode)) {
3078 i_size_write(inode, pos);
3079 mark_inode_dirty(inode);
3083 iov_iter_revert(from, write_len - iov_iter_count(from));
3087 EXPORT_SYMBOL(generic_file_direct_write);
3090 * Find or create a page at the given pagecache position. Return the locked
3091 * page. This function is specifically for buffered writes.
3093 struct page *grab_cache_page_write_begin(struct address_space *mapping,
3094 pgoff_t index, unsigned flags)
3097 int fgp_flags = FGP_LOCK|FGP_WRITE|FGP_CREAT;
3099 if (flags & AOP_FLAG_NOFS)
3100 fgp_flags |= FGP_NOFS;
3102 page = pagecache_get_page(mapping, index, fgp_flags,
3103 mapping_gfp_mask(mapping));
3105 wait_for_stable_page(page);
3109 EXPORT_SYMBOL(grab_cache_page_write_begin);
3111 ssize_t generic_perform_write(struct file *file,
3112 struct iov_iter *i, loff_t pos)
3114 struct address_space *mapping = file->f_mapping;
3115 const struct address_space_operations *a_ops = mapping->a_ops;
3117 ssize_t written = 0;
3118 unsigned int flags = 0;
3122 unsigned long offset; /* Offset into pagecache page */
3123 unsigned long bytes; /* Bytes to write to page */
3124 size_t copied; /* Bytes copied from user */
3127 offset = (pos & (PAGE_SIZE - 1));
3128 bytes = min_t(unsigned long, PAGE_SIZE - offset,
3133 * Bring in the user page that we will copy from _first_.
3134 * Otherwise there's a nasty deadlock on copying from the
3135 * same page as we're writing to, without it being marked
3138 * Not only is this an optimisation, but it is also required
3139 * to check that the address is actually valid, when atomic
3140 * usercopies are used, below.
3142 if (unlikely(iov_iter_fault_in_readable(i, bytes))) {
3147 if (fatal_signal_pending(current)) {
3152 status = a_ops->write_begin(file, mapping, pos, bytes, flags,
3154 if (unlikely(status < 0))
3157 if (mapping_writably_mapped(mapping))
3158 flush_dcache_page(page);
3160 copied = iov_iter_copy_from_user_atomic(page, i, offset, bytes);
3161 flush_dcache_page(page);
3163 status = a_ops->write_end(file, mapping, pos, bytes, copied,
3165 if (unlikely(status < 0))
3171 iov_iter_advance(i, copied);
3172 if (unlikely(copied == 0)) {
3174 * If we were unable to copy any data at all, we must
3175 * fall back to a single segment length write.
3177 * If we didn't fallback here, we could livelock
3178 * because not all segments in the iov can be copied at
3179 * once without a pagefault.
3181 bytes = min_t(unsigned long, PAGE_SIZE - offset,
3182 iov_iter_single_seg_count(i));
3188 balance_dirty_pages_ratelimited(mapping);
3189 } while (iov_iter_count(i));
3191 return written ? written : status;
3193 EXPORT_SYMBOL(generic_perform_write);
3196 * __generic_file_write_iter - write data to a file
3197 * @iocb: IO state structure (file, offset, etc.)
3198 * @from: iov_iter with data to write
3200 * This function does all the work needed for actually writing data to a
3201 * file. It does all basic checks, removes SUID from the file, updates
3202 * modification times and calls proper subroutines depending on whether we
3203 * do direct IO or a standard buffered write.
3205 * It expects i_mutex to be grabbed unless we work on a block device or similar
3206 * object which does not need locking at all.
3208 * This function does *not* take care of syncing data in case of O_SYNC write.
3209 * A caller has to handle it. This is mainly due to the fact that we want to
3210 * avoid syncing under i_mutex.
3212 ssize_t __generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
3214 struct file *file = iocb->ki_filp;
3215 struct address_space * mapping = file->f_mapping;
3216 struct inode *inode = mapping->host;
3217 ssize_t written = 0;
3221 /* We can write back this queue in page reclaim */
3222 current->backing_dev_info = inode_to_bdi(inode);
3223 err = file_remove_privs(file);
3227 err = file_update_time(file);
3231 if (iocb->ki_flags & IOCB_DIRECT) {
3232 loff_t pos, endbyte;
3234 written = generic_file_direct_write(iocb, from);
3236 * If the write stopped short of completing, fall back to
3237 * buffered writes. Some filesystems do this for writes to
3238 * holes, for example. For DAX files, a buffered write will
3239 * not succeed (even if it did, DAX does not handle dirty
3240 * page-cache pages correctly).
3242 if (written < 0 || !iov_iter_count(from) || IS_DAX(inode))
3245 status = generic_perform_write(file, from, pos = iocb->ki_pos);
3247 * If generic_perform_write() returned a synchronous error
3248 * then we want to return the number of bytes which were
3249 * direct-written, or the error code if that was zero. Note
3250 * that this differs from normal direct-io semantics, which
3251 * will return -EFOO even if some bytes were written.
3253 if (unlikely(status < 0)) {
3258 * We need to ensure that the page cache pages are written to
3259 * disk and invalidated to preserve the expected O_DIRECT
3262 endbyte = pos + status - 1;
3263 err = filemap_write_and_wait_range(mapping, pos, endbyte);
3265 iocb->ki_pos = endbyte + 1;
3267 invalidate_mapping_pages(mapping,
3269 endbyte >> PAGE_SHIFT);
3272 * We don't know how much we wrote, so just return
3273 * the number of bytes which were direct-written
3277 written = generic_perform_write(file, from, iocb->ki_pos);
3278 if (likely(written > 0))
3279 iocb->ki_pos += written;
3282 current->backing_dev_info = NULL;
3283 return written ? written : err;
3285 EXPORT_SYMBOL(__generic_file_write_iter);
3288 * generic_file_write_iter - write data to a file
3289 * @iocb: IO state structure
3290 * @from: iov_iter with data to write
3292 * This is a wrapper around __generic_file_write_iter() to be used by most
3293 * filesystems. It takes care of syncing the file in case of O_SYNC file
3294 * and acquires i_mutex as needed.
3296 ssize_t generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
3298 struct file *file = iocb->ki_filp;
3299 struct inode *inode = file->f_mapping->host;
3303 ret = generic_write_checks(iocb, from);
3305 ret = __generic_file_write_iter(iocb, from);
3306 inode_unlock(inode);
3309 ret = generic_write_sync(iocb, ret);
3312 EXPORT_SYMBOL(generic_file_write_iter);
3315 * try_to_release_page() - release old fs-specific metadata on a page
3317 * @page: the page which the kernel is trying to free
3318 * @gfp_mask: memory allocation flags (and I/O mode)
3320 * The address_space is to try to release any data against the page
3321 * (presumably at page->private). If the release was successful, return '1'.
3322 * Otherwise return zero.
3324 * This may also be called if PG_fscache is set on a page, indicating that the
3325 * page is known to the local caching routines.
3327 * The @gfp_mask argument specifies whether I/O may be performed to release
3328 * this page (__GFP_IO), and whether the call may block (__GFP_RECLAIM & __GFP_FS).
3331 int try_to_release_page(struct page *page, gfp_t gfp_mask)
3333 struct address_space * const mapping = page->mapping;
3335 BUG_ON(!PageLocked(page));
3336 if (PageWriteback(page))
3339 if (mapping && mapping->a_ops->releasepage)
3340 return mapping->a_ops->releasepage(page, gfp_mask);
3341 return try_to_free_buffers(page);
3344 EXPORT_SYMBOL(try_to_release_page);