]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blob - mm/filemap.c
ASoC: max98504: Add missing MAX98504 on SND_SOC_ALL_CODECS
[mirror_ubuntu-bionic-kernel.git] / mm / filemap.c
1 /*
2 * linux/mm/filemap.c
3 *
4 * Copyright (C) 1994-1999 Linus Torvalds
5 */
6
7 /*
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)
11 */
12 #include <linux/export.h>
13 #include <linux/compiler.h>
14 #include <linux/dax.h>
15 #include <linux/fs.h>
16 #include <linux/uaccess.h>
17 #include <linux/capability.h>
18 #include <linux/kernel_stat.h>
19 #include <linux/gfp.h>
20 #include <linux/mm.h>
21 #include <linux/swap.h>
22 #include <linux/mman.h>
23 #include <linux/pagemap.h>
24 #include <linux/file.h>
25 #include <linux/uio.h>
26 #include <linux/hash.h>
27 #include <linux/writeback.h>
28 #include <linux/backing-dev.h>
29 #include <linux/pagevec.h>
30 #include <linux/blkdev.h>
31 #include <linux/security.h>
32 #include <linux/cpuset.h>
33 #include <linux/hardirq.h> /* for BUG_ON(!in_atomic()) only */
34 #include <linux/hugetlb.h>
35 #include <linux/memcontrol.h>
36 #include <linux/cleancache.h>
37 #include <linux/rmap.h>
38 #include "internal.h"
39
40 #define CREATE_TRACE_POINTS
41 #include <trace/events/filemap.h>
42
43 /*
44 * FIXME: remove all knowledge of the buffer layer from the core VM
45 */
46 #include <linux/buffer_head.h> /* for try_to_free_buffers */
47
48 #include <asm/mman.h>
49
50 /*
51 * Shared mappings implemented 30.11.1994. It's not fully working yet,
52 * though.
53 *
54 * Shared mappings now work. 15.8.1995 Bruno.
55 *
56 * finished 'unifying' the page and buffer cache and SMP-threaded the
57 * page-cache, 21.05.1999, Ingo Molnar <mingo@redhat.com>
58 *
59 * SMP-threaded pagemap-LRU 1999, Andrea Arcangeli <andrea@suse.de>
60 */
61
62 /*
63 * Lock ordering:
64 *
65 * ->i_mmap_rwsem (truncate_pagecache)
66 * ->private_lock (__free_pte->__set_page_dirty_buffers)
67 * ->swap_lock (exclusive_swap_page, others)
68 * ->mapping->tree_lock
69 *
70 * ->i_mutex
71 * ->i_mmap_rwsem (truncate->unmap_mapping_range)
72 *
73 * ->mmap_sem
74 * ->i_mmap_rwsem
75 * ->page_table_lock or pte_lock (various, mainly in memory.c)
76 * ->mapping->tree_lock (arch-dependent flush_dcache_mmap_lock)
77 *
78 * ->mmap_sem
79 * ->lock_page (access_process_vm)
80 *
81 * ->i_mutex (generic_perform_write)
82 * ->mmap_sem (fault_in_pages_readable->do_page_fault)
83 *
84 * bdi->wb.list_lock
85 * sb_lock (fs/fs-writeback.c)
86 * ->mapping->tree_lock (__sync_single_inode)
87 *
88 * ->i_mmap_rwsem
89 * ->anon_vma.lock (vma_adjust)
90 *
91 * ->anon_vma.lock
92 * ->page_table_lock or pte_lock (anon_vma_prepare and various)
93 *
94 * ->page_table_lock or pte_lock
95 * ->swap_lock (try_to_unmap_one)
96 * ->private_lock (try_to_unmap_one)
97 * ->tree_lock (try_to_unmap_one)
98 * ->zone_lru_lock(zone) (follow_page->mark_page_accessed)
99 * ->zone_lru_lock(zone) (check_pte_range->isolate_lru_page)
100 * ->private_lock (page_remove_rmap->set_page_dirty)
101 * ->tree_lock (page_remove_rmap->set_page_dirty)
102 * bdi.wb->list_lock (page_remove_rmap->set_page_dirty)
103 * ->inode->i_lock (page_remove_rmap->set_page_dirty)
104 * ->memcg->move_lock (page_remove_rmap->lock_page_memcg)
105 * bdi.wb->list_lock (zap_pte_range->set_page_dirty)
106 * ->inode->i_lock (zap_pte_range->set_page_dirty)
107 * ->private_lock (zap_pte_range->__set_page_dirty_buffers)
108 *
109 * ->i_mmap_rwsem
110 * ->tasklist_lock (memory_failure, collect_procs_ao)
111 */
112
113 static int page_cache_tree_insert(struct address_space *mapping,
114 struct page *page, void **shadowp)
115 {
116 struct radix_tree_node *node;
117 void **slot;
118 int error;
119
120 error = __radix_tree_create(&mapping->page_tree, page->index, 0,
121 &node, &slot);
122 if (error)
123 return error;
124 if (*slot) {
125 void *p;
126
127 p = radix_tree_deref_slot_protected(slot, &mapping->tree_lock);
128 if (!radix_tree_exceptional_entry(p))
129 return -EEXIST;
130
131 mapping->nrexceptional--;
132 if (!dax_mapping(mapping)) {
133 if (shadowp)
134 *shadowp = p;
135 if (node)
136 workingset_node_shadows_dec(node);
137 } else {
138 /* DAX can replace empty locked entry with a hole */
139 WARN_ON_ONCE(p !=
140 (void *)(RADIX_TREE_EXCEPTIONAL_ENTRY |
141 RADIX_DAX_ENTRY_LOCK));
142 /* DAX accounts exceptional entries as normal pages */
143 if (node)
144 workingset_node_pages_dec(node);
145 /* Wakeup waiters for exceptional entry lock */
146 dax_wake_mapping_entry_waiter(mapping, page->index,
147 false);
148 }
149 }
150 radix_tree_replace_slot(slot, page);
151 mapping->nrpages++;
152 if (node) {
153 workingset_node_pages_inc(node);
154 /*
155 * Don't track node that contains actual pages.
156 *
157 * Avoid acquiring the list_lru lock if already
158 * untracked. The list_empty() test is safe as
159 * node->private_list is protected by
160 * mapping->tree_lock.
161 */
162 if (!list_empty(&node->private_list))
163 list_lru_del(&workingset_shadow_nodes,
164 &node->private_list);
165 }
166 return 0;
167 }
168
169 static void page_cache_tree_delete(struct address_space *mapping,
170 struct page *page, void *shadow)
171 {
172 int i, nr = PageHuge(page) ? 1 : hpage_nr_pages(page);
173
174 VM_BUG_ON_PAGE(!PageLocked(page), page);
175 VM_BUG_ON_PAGE(PageTail(page), page);
176 VM_BUG_ON_PAGE(nr != 1 && shadow, page);
177
178 for (i = 0; i < nr; i++) {
179 struct radix_tree_node *node;
180 void **slot;
181
182 __radix_tree_lookup(&mapping->page_tree, page->index + i,
183 &node, &slot);
184
185 radix_tree_clear_tags(&mapping->page_tree, node, slot);
186
187 if (!node) {
188 VM_BUG_ON_PAGE(nr != 1, page);
189 /*
190 * We need a node to properly account shadow
191 * entries. Don't plant any without. XXX
192 */
193 shadow = NULL;
194 }
195
196 radix_tree_replace_slot(slot, shadow);
197
198 if (!node)
199 break;
200
201 workingset_node_pages_dec(node);
202 if (shadow)
203 workingset_node_shadows_inc(node);
204 else
205 if (__radix_tree_delete_node(&mapping->page_tree, node))
206 continue;
207
208 /*
209 * Track node that only contains shadow entries. DAX mappings
210 * contain no shadow entries and may contain other exceptional
211 * entries so skip those.
212 *
213 * Avoid acquiring the list_lru lock if already tracked.
214 * The list_empty() test is safe as node->private_list is
215 * protected by mapping->tree_lock.
216 */
217 if (!dax_mapping(mapping) && !workingset_node_pages(node) &&
218 list_empty(&node->private_list)) {
219 node->private_data = mapping;
220 list_lru_add(&workingset_shadow_nodes,
221 &node->private_list);
222 }
223 }
224
225 if (shadow) {
226 mapping->nrexceptional += nr;
227 /*
228 * Make sure the nrexceptional update is committed before
229 * the nrpages update so that final truncate racing
230 * with reclaim does not see both counters 0 at the
231 * same time and miss a shadow entry.
232 */
233 smp_wmb();
234 }
235 mapping->nrpages -= nr;
236 }
237
238 /*
239 * Delete a page from the page cache and free it. Caller has to make
240 * sure the page is locked and that nobody else uses it - or that usage
241 * is safe. The caller must hold the mapping's tree_lock.
242 */
243 void __delete_from_page_cache(struct page *page, void *shadow)
244 {
245 struct address_space *mapping = page->mapping;
246 int nr = hpage_nr_pages(page);
247
248 trace_mm_filemap_delete_from_page_cache(page);
249 /*
250 * if we're uptodate, flush out into the cleancache, otherwise
251 * invalidate any existing cleancache entries. We can't leave
252 * stale data around in the cleancache once our page is gone
253 */
254 if (PageUptodate(page) && PageMappedToDisk(page))
255 cleancache_put_page(page);
256 else
257 cleancache_invalidate_page(mapping, page);
258
259 VM_BUG_ON_PAGE(PageTail(page), page);
260 VM_BUG_ON_PAGE(page_mapped(page), page);
261 if (!IS_ENABLED(CONFIG_DEBUG_VM) && unlikely(page_mapped(page))) {
262 int mapcount;
263
264 pr_alert("BUG: Bad page cache in process %s pfn:%05lx\n",
265 current->comm, page_to_pfn(page));
266 dump_page(page, "still mapped when deleted");
267 dump_stack();
268 add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE);
269
270 mapcount = page_mapcount(page);
271 if (mapping_exiting(mapping) &&
272 page_count(page) >= mapcount + 2) {
273 /*
274 * All vmas have already been torn down, so it's
275 * a good bet that actually the page is unmapped,
276 * and we'd prefer not to leak it: if we're wrong,
277 * some other bad page check should catch it later.
278 */
279 page_mapcount_reset(page);
280 page_ref_sub(page, mapcount);
281 }
282 }
283
284 page_cache_tree_delete(mapping, page, shadow);
285
286 page->mapping = NULL;
287 /* Leave page->index set: truncation lookup relies upon it */
288
289 /* hugetlb pages do not participate in page cache accounting. */
290 if (!PageHuge(page))
291 __mod_node_page_state(page_pgdat(page), NR_FILE_PAGES, -nr);
292 if (PageSwapBacked(page)) {
293 __mod_node_page_state(page_pgdat(page), NR_SHMEM, -nr);
294 if (PageTransHuge(page))
295 __dec_node_page_state(page, NR_SHMEM_THPS);
296 } else {
297 VM_BUG_ON_PAGE(PageTransHuge(page) && !PageHuge(page), page);
298 }
299
300 /*
301 * At this point page must be either written or cleaned by truncate.
302 * Dirty page here signals a bug and loss of unwritten data.
303 *
304 * This fixes dirty accounting after removing the page entirely but
305 * leaves PageDirty set: it has no effect for truncated page and
306 * anyway will be cleared before returning page into buddy allocator.
307 */
308 if (WARN_ON_ONCE(PageDirty(page)))
309 account_page_cleaned(page, mapping, inode_to_wb(mapping->host));
310 }
311
312 /**
313 * delete_from_page_cache - delete page from page cache
314 * @page: the page which the kernel is trying to remove from page cache
315 *
316 * This must be called only on pages that have been verified to be in the page
317 * cache and locked. It will never put the page into the free list, the caller
318 * has a reference on the page.
319 */
320 void delete_from_page_cache(struct page *page)
321 {
322 struct address_space *mapping = page_mapping(page);
323 unsigned long flags;
324 void (*freepage)(struct page *);
325
326 BUG_ON(!PageLocked(page));
327
328 freepage = mapping->a_ops->freepage;
329
330 spin_lock_irqsave(&mapping->tree_lock, flags);
331 __delete_from_page_cache(page, NULL);
332 spin_unlock_irqrestore(&mapping->tree_lock, flags);
333
334 if (freepage)
335 freepage(page);
336
337 if (PageTransHuge(page) && !PageHuge(page)) {
338 page_ref_sub(page, HPAGE_PMD_NR);
339 VM_BUG_ON_PAGE(page_count(page) <= 0, page);
340 } else {
341 put_page(page);
342 }
343 }
344 EXPORT_SYMBOL(delete_from_page_cache);
345
346 int filemap_check_errors(struct address_space *mapping)
347 {
348 int ret = 0;
349 /* Check for outstanding write errors */
350 if (test_bit(AS_ENOSPC, &mapping->flags) &&
351 test_and_clear_bit(AS_ENOSPC, &mapping->flags))
352 ret = -ENOSPC;
353 if (test_bit(AS_EIO, &mapping->flags) &&
354 test_and_clear_bit(AS_EIO, &mapping->flags))
355 ret = -EIO;
356 return ret;
357 }
358 EXPORT_SYMBOL(filemap_check_errors);
359
360 /**
361 * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range
362 * @mapping: address space structure to write
363 * @start: offset in bytes where the range starts
364 * @end: offset in bytes where the range ends (inclusive)
365 * @sync_mode: enable synchronous operation
366 *
367 * Start writeback against all of a mapping's dirty pages that lie
368 * within the byte offsets <start, end> inclusive.
369 *
370 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
371 * opposed to a regular memory cleansing writeback. The difference between
372 * these two operations is that if a dirty page/buffer is encountered, it must
373 * be waited upon, and not just skipped over.
374 */
375 int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
376 loff_t end, int sync_mode)
377 {
378 int ret;
379 struct writeback_control wbc = {
380 .sync_mode = sync_mode,
381 .nr_to_write = LONG_MAX,
382 .range_start = start,
383 .range_end = end,
384 };
385
386 if (!mapping_cap_writeback_dirty(mapping))
387 return 0;
388
389 wbc_attach_fdatawrite_inode(&wbc, mapping->host);
390 ret = do_writepages(mapping, &wbc);
391 wbc_detach_inode(&wbc);
392 return ret;
393 }
394
395 static inline int __filemap_fdatawrite(struct address_space *mapping,
396 int sync_mode)
397 {
398 return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode);
399 }
400
401 int filemap_fdatawrite(struct address_space *mapping)
402 {
403 return __filemap_fdatawrite(mapping, WB_SYNC_ALL);
404 }
405 EXPORT_SYMBOL(filemap_fdatawrite);
406
407 int filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
408 loff_t end)
409 {
410 return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL);
411 }
412 EXPORT_SYMBOL(filemap_fdatawrite_range);
413
414 /**
415 * filemap_flush - mostly a non-blocking flush
416 * @mapping: target address_space
417 *
418 * This is a mostly non-blocking flush. Not suitable for data-integrity
419 * purposes - I/O may not be started against all dirty pages.
420 */
421 int filemap_flush(struct address_space *mapping)
422 {
423 return __filemap_fdatawrite(mapping, WB_SYNC_NONE);
424 }
425 EXPORT_SYMBOL(filemap_flush);
426
427 static int __filemap_fdatawait_range(struct address_space *mapping,
428 loff_t start_byte, loff_t end_byte)
429 {
430 pgoff_t index = start_byte >> PAGE_SHIFT;
431 pgoff_t end = end_byte >> PAGE_SHIFT;
432 struct pagevec pvec;
433 int nr_pages;
434 int ret = 0;
435
436 if (end_byte < start_byte)
437 goto out;
438
439 pagevec_init(&pvec, 0);
440 while ((index <= end) &&
441 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
442 PAGECACHE_TAG_WRITEBACK,
443 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1)) != 0) {
444 unsigned i;
445
446 for (i = 0; i < nr_pages; i++) {
447 struct page *page = pvec.pages[i];
448
449 /* until radix tree lookup accepts end_index */
450 if (page->index > end)
451 continue;
452
453 wait_on_page_writeback(page);
454 if (TestClearPageError(page))
455 ret = -EIO;
456 }
457 pagevec_release(&pvec);
458 cond_resched();
459 }
460 out:
461 return ret;
462 }
463
464 /**
465 * filemap_fdatawait_range - wait for writeback to complete
466 * @mapping: address space structure to wait for
467 * @start_byte: offset in bytes where the range starts
468 * @end_byte: offset in bytes where the range ends (inclusive)
469 *
470 * Walk the list of under-writeback pages of the given address space
471 * in the given range and wait for all of them. Check error status of
472 * the address space and return it.
473 *
474 * Since the error status of the address space is cleared by this function,
475 * callers are responsible for checking the return value and handling and/or
476 * reporting the error.
477 */
478 int filemap_fdatawait_range(struct address_space *mapping, loff_t start_byte,
479 loff_t end_byte)
480 {
481 int ret, ret2;
482
483 ret = __filemap_fdatawait_range(mapping, start_byte, end_byte);
484 ret2 = filemap_check_errors(mapping);
485 if (!ret)
486 ret = ret2;
487
488 return ret;
489 }
490 EXPORT_SYMBOL(filemap_fdatawait_range);
491
492 /**
493 * filemap_fdatawait_keep_errors - wait for writeback without clearing errors
494 * @mapping: address space structure to wait for
495 *
496 * Walk the list of under-writeback pages of the given address space
497 * and wait for all of them. Unlike filemap_fdatawait(), this function
498 * does not clear error status of the address space.
499 *
500 * Use this function if callers don't handle errors themselves. Expected
501 * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2),
502 * fsfreeze(8)
503 */
504 void filemap_fdatawait_keep_errors(struct address_space *mapping)
505 {
506 loff_t i_size = i_size_read(mapping->host);
507
508 if (i_size == 0)
509 return;
510
511 __filemap_fdatawait_range(mapping, 0, i_size - 1);
512 }
513
514 /**
515 * filemap_fdatawait - wait for all under-writeback pages to complete
516 * @mapping: address space structure to wait for
517 *
518 * Walk the list of under-writeback pages of the given address space
519 * and wait for all of them. Check error status of the address space
520 * and return it.
521 *
522 * Since the error status of the address space is cleared by this function,
523 * callers are responsible for checking the return value and handling and/or
524 * reporting the error.
525 */
526 int filemap_fdatawait(struct address_space *mapping)
527 {
528 loff_t i_size = i_size_read(mapping->host);
529
530 if (i_size == 0)
531 return 0;
532
533 return filemap_fdatawait_range(mapping, 0, i_size - 1);
534 }
535 EXPORT_SYMBOL(filemap_fdatawait);
536
537 int filemap_write_and_wait(struct address_space *mapping)
538 {
539 int err = 0;
540
541 if ((!dax_mapping(mapping) && mapping->nrpages) ||
542 (dax_mapping(mapping) && mapping->nrexceptional)) {
543 err = filemap_fdatawrite(mapping);
544 /*
545 * Even if the above returned error, the pages may be
546 * written partially (e.g. -ENOSPC), so we wait for it.
547 * But the -EIO is special case, it may indicate the worst
548 * thing (e.g. bug) happened, so we avoid waiting for it.
549 */
550 if (err != -EIO) {
551 int err2 = filemap_fdatawait(mapping);
552 if (!err)
553 err = err2;
554 }
555 } else {
556 err = filemap_check_errors(mapping);
557 }
558 return err;
559 }
560 EXPORT_SYMBOL(filemap_write_and_wait);
561
562 /**
563 * filemap_write_and_wait_range - write out & wait on a file range
564 * @mapping: the address_space for the pages
565 * @lstart: offset in bytes where the range starts
566 * @lend: offset in bytes where the range ends (inclusive)
567 *
568 * Write out and wait upon file offsets lstart->lend, inclusive.
569 *
570 * Note that `lend' is inclusive (describes the last byte to be written) so
571 * that this function can be used to write to the very end-of-file (end = -1).
572 */
573 int filemap_write_and_wait_range(struct address_space *mapping,
574 loff_t lstart, loff_t lend)
575 {
576 int err = 0;
577
578 if ((!dax_mapping(mapping) && mapping->nrpages) ||
579 (dax_mapping(mapping) && mapping->nrexceptional)) {
580 err = __filemap_fdatawrite_range(mapping, lstart, lend,
581 WB_SYNC_ALL);
582 /* See comment of filemap_write_and_wait() */
583 if (err != -EIO) {
584 int err2 = filemap_fdatawait_range(mapping,
585 lstart, lend);
586 if (!err)
587 err = err2;
588 }
589 } else {
590 err = filemap_check_errors(mapping);
591 }
592 return err;
593 }
594 EXPORT_SYMBOL(filemap_write_and_wait_range);
595
596 /**
597 * replace_page_cache_page - replace a pagecache page with a new one
598 * @old: page to be replaced
599 * @new: page to replace with
600 * @gfp_mask: allocation mode
601 *
602 * This function replaces a page in the pagecache with a new one. On
603 * success it acquires the pagecache reference for the new page and
604 * drops it for the old page. Both the old and new pages must be
605 * locked. This function does not add the new page to the LRU, the
606 * caller must do that.
607 *
608 * The remove + add is atomic. The only way this function can fail is
609 * memory allocation failure.
610 */
611 int replace_page_cache_page(struct page *old, struct page *new, gfp_t gfp_mask)
612 {
613 int error;
614
615 VM_BUG_ON_PAGE(!PageLocked(old), old);
616 VM_BUG_ON_PAGE(!PageLocked(new), new);
617 VM_BUG_ON_PAGE(new->mapping, new);
618
619 error = radix_tree_preload(gfp_mask & ~__GFP_HIGHMEM);
620 if (!error) {
621 struct address_space *mapping = old->mapping;
622 void (*freepage)(struct page *);
623 unsigned long flags;
624
625 pgoff_t offset = old->index;
626 freepage = mapping->a_ops->freepage;
627
628 get_page(new);
629 new->mapping = mapping;
630 new->index = offset;
631
632 spin_lock_irqsave(&mapping->tree_lock, flags);
633 __delete_from_page_cache(old, NULL);
634 error = page_cache_tree_insert(mapping, new, NULL);
635 BUG_ON(error);
636
637 /*
638 * hugetlb pages do not participate in page cache accounting.
639 */
640 if (!PageHuge(new))
641 __inc_node_page_state(new, NR_FILE_PAGES);
642 if (PageSwapBacked(new))
643 __inc_node_page_state(new, NR_SHMEM);
644 spin_unlock_irqrestore(&mapping->tree_lock, flags);
645 mem_cgroup_migrate(old, new);
646 radix_tree_preload_end();
647 if (freepage)
648 freepage(old);
649 put_page(old);
650 }
651
652 return error;
653 }
654 EXPORT_SYMBOL_GPL(replace_page_cache_page);
655
656 static int __add_to_page_cache_locked(struct page *page,
657 struct address_space *mapping,
658 pgoff_t offset, gfp_t gfp_mask,
659 void **shadowp)
660 {
661 int huge = PageHuge(page);
662 struct mem_cgroup *memcg;
663 int error;
664
665 VM_BUG_ON_PAGE(!PageLocked(page), page);
666 VM_BUG_ON_PAGE(PageSwapBacked(page), page);
667
668 if (!huge) {
669 error = mem_cgroup_try_charge(page, current->mm,
670 gfp_mask, &memcg, false);
671 if (error)
672 return error;
673 }
674
675 error = radix_tree_maybe_preload(gfp_mask & ~__GFP_HIGHMEM);
676 if (error) {
677 if (!huge)
678 mem_cgroup_cancel_charge(page, memcg, false);
679 return error;
680 }
681
682 get_page(page);
683 page->mapping = mapping;
684 page->index = offset;
685
686 spin_lock_irq(&mapping->tree_lock);
687 error = page_cache_tree_insert(mapping, page, shadowp);
688 radix_tree_preload_end();
689 if (unlikely(error))
690 goto err_insert;
691
692 /* hugetlb pages do not participate in page cache accounting. */
693 if (!huge)
694 __inc_node_page_state(page, NR_FILE_PAGES);
695 spin_unlock_irq(&mapping->tree_lock);
696 if (!huge)
697 mem_cgroup_commit_charge(page, memcg, false, false);
698 trace_mm_filemap_add_to_page_cache(page);
699 return 0;
700 err_insert:
701 page->mapping = NULL;
702 /* Leave page->index set: truncation relies upon it */
703 spin_unlock_irq(&mapping->tree_lock);
704 if (!huge)
705 mem_cgroup_cancel_charge(page, memcg, false);
706 put_page(page);
707 return error;
708 }
709
710 /**
711 * add_to_page_cache_locked - add a locked page to the pagecache
712 * @page: page to add
713 * @mapping: the page's address_space
714 * @offset: page index
715 * @gfp_mask: page allocation mode
716 *
717 * This function is used to add a page to the pagecache. It must be locked.
718 * This function does not add the page to the LRU. The caller must do that.
719 */
720 int add_to_page_cache_locked(struct page *page, struct address_space *mapping,
721 pgoff_t offset, gfp_t gfp_mask)
722 {
723 return __add_to_page_cache_locked(page, mapping, offset,
724 gfp_mask, NULL);
725 }
726 EXPORT_SYMBOL(add_to_page_cache_locked);
727
728 int add_to_page_cache_lru(struct page *page, struct address_space *mapping,
729 pgoff_t offset, gfp_t gfp_mask)
730 {
731 void *shadow = NULL;
732 int ret;
733
734 __SetPageLocked(page);
735 ret = __add_to_page_cache_locked(page, mapping, offset,
736 gfp_mask, &shadow);
737 if (unlikely(ret))
738 __ClearPageLocked(page);
739 else {
740 /*
741 * The page might have been evicted from cache only
742 * recently, in which case it should be activated like
743 * any other repeatedly accessed page.
744 * The exception is pages getting rewritten; evicting other
745 * data from the working set, only to cache data that will
746 * get overwritten with something else, is a waste of memory.
747 */
748 if (!(gfp_mask & __GFP_WRITE) &&
749 shadow && workingset_refault(shadow)) {
750 SetPageActive(page);
751 workingset_activation(page);
752 } else
753 ClearPageActive(page);
754 lru_cache_add(page);
755 }
756 return ret;
757 }
758 EXPORT_SYMBOL_GPL(add_to_page_cache_lru);
759
760 #ifdef CONFIG_NUMA
761 struct page *__page_cache_alloc(gfp_t gfp)
762 {
763 int n;
764 struct page *page;
765
766 if (cpuset_do_page_mem_spread()) {
767 unsigned int cpuset_mems_cookie;
768 do {
769 cpuset_mems_cookie = read_mems_allowed_begin();
770 n = cpuset_mem_spread_node();
771 page = __alloc_pages_node(n, gfp, 0);
772 } while (!page && read_mems_allowed_retry(cpuset_mems_cookie));
773
774 return page;
775 }
776 return alloc_pages(gfp, 0);
777 }
778 EXPORT_SYMBOL(__page_cache_alloc);
779 #endif
780
781 /*
782 * In order to wait for pages to become available there must be
783 * waitqueues associated with pages. By using a hash table of
784 * waitqueues where the bucket discipline is to maintain all
785 * waiters on the same queue and wake all when any of the pages
786 * become available, and for the woken contexts to check to be
787 * sure the appropriate page became available, this saves space
788 * at a cost of "thundering herd" phenomena during rare hash
789 * collisions.
790 */
791 wait_queue_head_t *page_waitqueue(struct page *page)
792 {
793 const struct zone *zone = page_zone(page);
794
795 return &zone->wait_table[hash_ptr(page, zone->wait_table_bits)];
796 }
797 EXPORT_SYMBOL(page_waitqueue);
798
799 void wait_on_page_bit(struct page *page, int bit_nr)
800 {
801 DEFINE_WAIT_BIT(wait, &page->flags, bit_nr);
802
803 if (test_bit(bit_nr, &page->flags))
804 __wait_on_bit(page_waitqueue(page), &wait, bit_wait_io,
805 TASK_UNINTERRUPTIBLE);
806 }
807 EXPORT_SYMBOL(wait_on_page_bit);
808
809 int wait_on_page_bit_killable(struct page *page, int bit_nr)
810 {
811 DEFINE_WAIT_BIT(wait, &page->flags, bit_nr);
812
813 if (!test_bit(bit_nr, &page->flags))
814 return 0;
815
816 return __wait_on_bit(page_waitqueue(page), &wait,
817 bit_wait_io, TASK_KILLABLE);
818 }
819
820 int wait_on_page_bit_killable_timeout(struct page *page,
821 int bit_nr, unsigned long timeout)
822 {
823 DEFINE_WAIT_BIT(wait, &page->flags, bit_nr);
824
825 wait.key.timeout = jiffies + timeout;
826 if (!test_bit(bit_nr, &page->flags))
827 return 0;
828 return __wait_on_bit(page_waitqueue(page), &wait,
829 bit_wait_io_timeout, TASK_KILLABLE);
830 }
831 EXPORT_SYMBOL_GPL(wait_on_page_bit_killable_timeout);
832
833 /**
834 * add_page_wait_queue - Add an arbitrary waiter to a page's wait queue
835 * @page: Page defining the wait queue of interest
836 * @waiter: Waiter to add to the queue
837 *
838 * Add an arbitrary @waiter to the wait queue for the nominated @page.
839 */
840 void add_page_wait_queue(struct page *page, wait_queue_t *waiter)
841 {
842 wait_queue_head_t *q = page_waitqueue(page);
843 unsigned long flags;
844
845 spin_lock_irqsave(&q->lock, flags);
846 __add_wait_queue(q, waiter);
847 spin_unlock_irqrestore(&q->lock, flags);
848 }
849 EXPORT_SYMBOL_GPL(add_page_wait_queue);
850
851 /**
852 * unlock_page - unlock a locked page
853 * @page: the page
854 *
855 * Unlocks the page and wakes up sleepers in ___wait_on_page_locked().
856 * Also wakes sleepers in wait_on_page_writeback() because the wakeup
857 * mechanism between PageLocked pages and PageWriteback pages is shared.
858 * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep.
859 *
860 * The mb is necessary to enforce ordering between the clear_bit and the read
861 * of the waitqueue (to avoid SMP races with a parallel wait_on_page_locked()).
862 */
863 void unlock_page(struct page *page)
864 {
865 page = compound_head(page);
866 VM_BUG_ON_PAGE(!PageLocked(page), page);
867 clear_bit_unlock(PG_locked, &page->flags);
868 smp_mb__after_atomic();
869 wake_up_page(page, PG_locked);
870 }
871 EXPORT_SYMBOL(unlock_page);
872
873 /**
874 * end_page_writeback - end writeback against a page
875 * @page: the page
876 */
877 void end_page_writeback(struct page *page)
878 {
879 /*
880 * TestClearPageReclaim could be used here but it is an atomic
881 * operation and overkill in this particular case. Failing to
882 * shuffle a page marked for immediate reclaim is too mild to
883 * justify taking an atomic operation penalty at the end of
884 * ever page writeback.
885 */
886 if (PageReclaim(page)) {
887 ClearPageReclaim(page);
888 rotate_reclaimable_page(page);
889 }
890
891 if (!test_clear_page_writeback(page))
892 BUG();
893
894 smp_mb__after_atomic();
895 wake_up_page(page, PG_writeback);
896 }
897 EXPORT_SYMBOL(end_page_writeback);
898
899 /*
900 * After completing I/O on a page, call this routine to update the page
901 * flags appropriately
902 */
903 void page_endio(struct page *page, bool is_write, int err)
904 {
905 if (!is_write) {
906 if (!err) {
907 SetPageUptodate(page);
908 } else {
909 ClearPageUptodate(page);
910 SetPageError(page);
911 }
912 unlock_page(page);
913 } else {
914 if (err) {
915 SetPageError(page);
916 if (page->mapping)
917 mapping_set_error(page->mapping, err);
918 }
919 end_page_writeback(page);
920 }
921 }
922 EXPORT_SYMBOL_GPL(page_endio);
923
924 /**
925 * __lock_page - get a lock on the page, assuming we need to sleep to get it
926 * @page: the page to lock
927 */
928 void __lock_page(struct page *page)
929 {
930 struct page *page_head = compound_head(page);
931 DEFINE_WAIT_BIT(wait, &page_head->flags, PG_locked);
932
933 __wait_on_bit_lock(page_waitqueue(page_head), &wait, bit_wait_io,
934 TASK_UNINTERRUPTIBLE);
935 }
936 EXPORT_SYMBOL(__lock_page);
937
938 int __lock_page_killable(struct page *page)
939 {
940 struct page *page_head = compound_head(page);
941 DEFINE_WAIT_BIT(wait, &page_head->flags, PG_locked);
942
943 return __wait_on_bit_lock(page_waitqueue(page_head), &wait,
944 bit_wait_io, TASK_KILLABLE);
945 }
946 EXPORT_SYMBOL_GPL(__lock_page_killable);
947
948 /*
949 * Return values:
950 * 1 - page is locked; mmap_sem is still held.
951 * 0 - page is not locked.
952 * mmap_sem has been released (up_read()), unless flags had both
953 * FAULT_FLAG_ALLOW_RETRY and FAULT_FLAG_RETRY_NOWAIT set, in
954 * which case mmap_sem is still held.
955 *
956 * If neither ALLOW_RETRY nor KILLABLE are set, will always return 1
957 * with the page locked and the mmap_sem unperturbed.
958 */
959 int __lock_page_or_retry(struct page *page, struct mm_struct *mm,
960 unsigned int flags)
961 {
962 if (flags & FAULT_FLAG_ALLOW_RETRY) {
963 /*
964 * CAUTION! In this case, mmap_sem is not released
965 * even though return 0.
966 */
967 if (flags & FAULT_FLAG_RETRY_NOWAIT)
968 return 0;
969
970 up_read(&mm->mmap_sem);
971 if (flags & FAULT_FLAG_KILLABLE)
972 wait_on_page_locked_killable(page);
973 else
974 wait_on_page_locked(page);
975 return 0;
976 } else {
977 if (flags & FAULT_FLAG_KILLABLE) {
978 int ret;
979
980 ret = __lock_page_killable(page);
981 if (ret) {
982 up_read(&mm->mmap_sem);
983 return 0;
984 }
985 } else
986 __lock_page(page);
987 return 1;
988 }
989 }
990
991 /**
992 * page_cache_next_hole - find the next hole (not-present entry)
993 * @mapping: mapping
994 * @index: index
995 * @max_scan: maximum range to search
996 *
997 * Search the set [index, min(index+max_scan-1, MAX_INDEX)] for the
998 * lowest indexed hole.
999 *
1000 * Returns: the index of the hole if found, otherwise returns an index
1001 * outside of the set specified (in which case 'return - index >=
1002 * max_scan' will be true). In rare cases of index wrap-around, 0 will
1003 * be returned.
1004 *
1005 * page_cache_next_hole may be called under rcu_read_lock. However,
1006 * like radix_tree_gang_lookup, this will not atomically search a
1007 * snapshot of the tree at a single point in time. For example, if a
1008 * hole is created at index 5, then subsequently a hole is created at
1009 * index 10, page_cache_next_hole covering both indexes may return 10
1010 * if called under rcu_read_lock.
1011 */
1012 pgoff_t page_cache_next_hole(struct address_space *mapping,
1013 pgoff_t index, unsigned long max_scan)
1014 {
1015 unsigned long i;
1016
1017 for (i = 0; i < max_scan; i++) {
1018 struct page *page;
1019
1020 page = radix_tree_lookup(&mapping->page_tree, index);
1021 if (!page || radix_tree_exceptional_entry(page))
1022 break;
1023 index++;
1024 if (index == 0)
1025 break;
1026 }
1027
1028 return index;
1029 }
1030 EXPORT_SYMBOL(page_cache_next_hole);
1031
1032 /**
1033 * page_cache_prev_hole - find the prev hole (not-present entry)
1034 * @mapping: mapping
1035 * @index: index
1036 * @max_scan: maximum range to search
1037 *
1038 * Search backwards in the range [max(index-max_scan+1, 0), index] for
1039 * the first hole.
1040 *
1041 * Returns: the index of the hole if found, otherwise returns an index
1042 * outside of the set specified (in which case 'index - return >=
1043 * max_scan' will be true). In rare cases of wrap-around, ULONG_MAX
1044 * will be returned.
1045 *
1046 * page_cache_prev_hole may be called under rcu_read_lock. However,
1047 * like radix_tree_gang_lookup, this will not atomically search a
1048 * snapshot of the tree at a single point in time. For example, if a
1049 * hole is created at index 10, then subsequently a hole is created at
1050 * index 5, page_cache_prev_hole covering both indexes may return 5 if
1051 * called under rcu_read_lock.
1052 */
1053 pgoff_t page_cache_prev_hole(struct address_space *mapping,
1054 pgoff_t index, unsigned long max_scan)
1055 {
1056 unsigned long i;
1057
1058 for (i = 0; i < max_scan; i++) {
1059 struct page *page;
1060
1061 page = radix_tree_lookup(&mapping->page_tree, index);
1062 if (!page || radix_tree_exceptional_entry(page))
1063 break;
1064 index--;
1065 if (index == ULONG_MAX)
1066 break;
1067 }
1068
1069 return index;
1070 }
1071 EXPORT_SYMBOL(page_cache_prev_hole);
1072
1073 /**
1074 * find_get_entry - find and get a page cache entry
1075 * @mapping: the address_space to search
1076 * @offset: the page cache index
1077 *
1078 * Looks up the page cache slot at @mapping & @offset. If there is a
1079 * page cache page, it is returned with an increased refcount.
1080 *
1081 * If the slot holds a shadow entry of a previously evicted page, or a
1082 * swap entry from shmem/tmpfs, it is returned.
1083 *
1084 * Otherwise, %NULL is returned.
1085 */
1086 struct page *find_get_entry(struct address_space *mapping, pgoff_t offset)
1087 {
1088 void **pagep;
1089 struct page *head, *page;
1090
1091 rcu_read_lock();
1092 repeat:
1093 page = NULL;
1094 pagep = radix_tree_lookup_slot(&mapping->page_tree, offset);
1095 if (pagep) {
1096 page = radix_tree_deref_slot(pagep);
1097 if (unlikely(!page))
1098 goto out;
1099 if (radix_tree_exception(page)) {
1100 if (radix_tree_deref_retry(page))
1101 goto repeat;
1102 /*
1103 * A shadow entry of a recently evicted page,
1104 * or a swap entry from shmem/tmpfs. Return
1105 * it without attempting to raise page count.
1106 */
1107 goto out;
1108 }
1109
1110 head = compound_head(page);
1111 if (!page_cache_get_speculative(head))
1112 goto repeat;
1113
1114 /* The page was split under us? */
1115 if (compound_head(page) != head) {
1116 put_page(head);
1117 goto repeat;
1118 }
1119
1120 /*
1121 * Has the page moved?
1122 * This is part of the lockless pagecache protocol. See
1123 * include/linux/pagemap.h for details.
1124 */
1125 if (unlikely(page != *pagep)) {
1126 put_page(head);
1127 goto repeat;
1128 }
1129 }
1130 out:
1131 rcu_read_unlock();
1132
1133 return page;
1134 }
1135 EXPORT_SYMBOL(find_get_entry);
1136
1137 /**
1138 * find_lock_entry - locate, pin and lock a page cache entry
1139 * @mapping: the address_space to search
1140 * @offset: the page cache index
1141 *
1142 * Looks up the page cache slot at @mapping & @offset. If there is a
1143 * page cache page, it is returned locked and with an increased
1144 * refcount.
1145 *
1146 * If the slot holds a shadow entry of a previously evicted page, or a
1147 * swap entry from shmem/tmpfs, it is returned.
1148 *
1149 * Otherwise, %NULL is returned.
1150 *
1151 * find_lock_entry() may sleep.
1152 */
1153 struct page *find_lock_entry(struct address_space *mapping, pgoff_t offset)
1154 {
1155 struct page *page;
1156
1157 repeat:
1158 page = find_get_entry(mapping, offset);
1159 if (page && !radix_tree_exception(page)) {
1160 lock_page(page);
1161 /* Has the page been truncated? */
1162 if (unlikely(page_mapping(page) != mapping)) {
1163 unlock_page(page);
1164 put_page(page);
1165 goto repeat;
1166 }
1167 VM_BUG_ON_PAGE(page_to_pgoff(page) != offset, page);
1168 }
1169 return page;
1170 }
1171 EXPORT_SYMBOL(find_lock_entry);
1172
1173 /**
1174 * pagecache_get_page - find and get a page reference
1175 * @mapping: the address_space to search
1176 * @offset: the page index
1177 * @fgp_flags: PCG flags
1178 * @gfp_mask: gfp mask to use for the page cache data page allocation
1179 *
1180 * Looks up the page cache slot at @mapping & @offset.
1181 *
1182 * PCG flags modify how the page is returned.
1183 *
1184 * FGP_ACCESSED: the page will be marked accessed
1185 * FGP_LOCK: Page is return locked
1186 * FGP_CREAT: If page is not present then a new page is allocated using
1187 * @gfp_mask and added to the page cache and the VM's LRU
1188 * list. The page is returned locked and with an increased
1189 * refcount. Otherwise, %NULL is returned.
1190 *
1191 * If FGP_LOCK or FGP_CREAT are specified then the function may sleep even
1192 * if the GFP flags specified for FGP_CREAT are atomic.
1193 *
1194 * If there is a page cache page, it is returned with an increased refcount.
1195 */
1196 struct page *pagecache_get_page(struct address_space *mapping, pgoff_t offset,
1197 int fgp_flags, gfp_t gfp_mask)
1198 {
1199 struct page *page;
1200
1201 repeat:
1202 page = find_get_entry(mapping, offset);
1203 if (radix_tree_exceptional_entry(page))
1204 page = NULL;
1205 if (!page)
1206 goto no_page;
1207
1208 if (fgp_flags & FGP_LOCK) {
1209 if (fgp_flags & FGP_NOWAIT) {
1210 if (!trylock_page(page)) {
1211 put_page(page);
1212 return NULL;
1213 }
1214 } else {
1215 lock_page(page);
1216 }
1217
1218 /* Has the page been truncated? */
1219 if (unlikely(page->mapping != mapping)) {
1220 unlock_page(page);
1221 put_page(page);
1222 goto repeat;
1223 }
1224 VM_BUG_ON_PAGE(page->index != offset, page);
1225 }
1226
1227 if (page && (fgp_flags & FGP_ACCESSED))
1228 mark_page_accessed(page);
1229
1230 no_page:
1231 if (!page && (fgp_flags & FGP_CREAT)) {
1232 int err;
1233 if ((fgp_flags & FGP_WRITE) && mapping_cap_account_dirty(mapping))
1234 gfp_mask |= __GFP_WRITE;
1235 if (fgp_flags & FGP_NOFS)
1236 gfp_mask &= ~__GFP_FS;
1237
1238 page = __page_cache_alloc(gfp_mask);
1239 if (!page)
1240 return NULL;
1241
1242 if (WARN_ON_ONCE(!(fgp_flags & FGP_LOCK)))
1243 fgp_flags |= FGP_LOCK;
1244
1245 /* Init accessed so avoid atomic mark_page_accessed later */
1246 if (fgp_flags & FGP_ACCESSED)
1247 __SetPageReferenced(page);
1248
1249 err = add_to_page_cache_lru(page, mapping, offset,
1250 gfp_mask & GFP_RECLAIM_MASK);
1251 if (unlikely(err)) {
1252 put_page(page);
1253 page = NULL;
1254 if (err == -EEXIST)
1255 goto repeat;
1256 }
1257 }
1258
1259 return page;
1260 }
1261 EXPORT_SYMBOL(pagecache_get_page);
1262
1263 /**
1264 * find_get_entries - gang pagecache lookup
1265 * @mapping: The address_space to search
1266 * @start: The starting page cache index
1267 * @nr_entries: The maximum number of entries
1268 * @entries: Where the resulting entries are placed
1269 * @indices: The cache indices corresponding to the entries in @entries
1270 *
1271 * find_get_entries() will search for and return a group of up to
1272 * @nr_entries entries in the mapping. The entries are placed at
1273 * @entries. find_get_entries() takes a reference against any actual
1274 * pages it returns.
1275 *
1276 * The search returns a group of mapping-contiguous page cache entries
1277 * with ascending indexes. There may be holes in the indices due to
1278 * not-present pages.
1279 *
1280 * Any shadow entries of evicted pages, or swap entries from
1281 * shmem/tmpfs, are included in the returned array.
1282 *
1283 * find_get_entries() returns the number of pages and shadow entries
1284 * which were found.
1285 */
1286 unsigned find_get_entries(struct address_space *mapping,
1287 pgoff_t start, unsigned int nr_entries,
1288 struct page **entries, pgoff_t *indices)
1289 {
1290 void **slot;
1291 unsigned int ret = 0;
1292 struct radix_tree_iter iter;
1293
1294 if (!nr_entries)
1295 return 0;
1296
1297 rcu_read_lock();
1298 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) {
1299 struct page *head, *page;
1300 repeat:
1301 page = radix_tree_deref_slot(slot);
1302 if (unlikely(!page))
1303 continue;
1304 if (radix_tree_exception(page)) {
1305 if (radix_tree_deref_retry(page)) {
1306 slot = radix_tree_iter_retry(&iter);
1307 continue;
1308 }
1309 /*
1310 * A shadow entry of a recently evicted page, a swap
1311 * entry from shmem/tmpfs or a DAX entry. Return it
1312 * without attempting to raise page count.
1313 */
1314 goto export;
1315 }
1316
1317 head = compound_head(page);
1318 if (!page_cache_get_speculative(head))
1319 goto repeat;
1320
1321 /* The page was split under us? */
1322 if (compound_head(page) != head) {
1323 put_page(head);
1324 goto repeat;
1325 }
1326
1327 /* Has the page moved? */
1328 if (unlikely(page != *slot)) {
1329 put_page(head);
1330 goto repeat;
1331 }
1332 export:
1333 indices[ret] = iter.index;
1334 entries[ret] = page;
1335 if (++ret == nr_entries)
1336 break;
1337 }
1338 rcu_read_unlock();
1339 return ret;
1340 }
1341
1342 /**
1343 * find_get_pages - gang pagecache lookup
1344 * @mapping: The address_space to search
1345 * @start: The starting page index
1346 * @nr_pages: The maximum number of pages
1347 * @pages: Where the resulting pages are placed
1348 *
1349 * find_get_pages() will search for and return a group of up to
1350 * @nr_pages pages in the mapping. The pages are placed at @pages.
1351 * find_get_pages() takes a reference against the returned pages.
1352 *
1353 * The search returns a group of mapping-contiguous pages with ascending
1354 * indexes. There may be holes in the indices due to not-present pages.
1355 *
1356 * find_get_pages() returns the number of pages which were found.
1357 */
1358 unsigned find_get_pages(struct address_space *mapping, pgoff_t start,
1359 unsigned int nr_pages, struct page **pages)
1360 {
1361 struct radix_tree_iter iter;
1362 void **slot;
1363 unsigned ret = 0;
1364
1365 if (unlikely(!nr_pages))
1366 return 0;
1367
1368 rcu_read_lock();
1369 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) {
1370 struct page *head, *page;
1371 repeat:
1372 page = radix_tree_deref_slot(slot);
1373 if (unlikely(!page))
1374 continue;
1375
1376 if (radix_tree_exception(page)) {
1377 if (radix_tree_deref_retry(page)) {
1378 slot = radix_tree_iter_retry(&iter);
1379 continue;
1380 }
1381 /*
1382 * A shadow entry of a recently evicted page,
1383 * or a swap entry from shmem/tmpfs. Skip
1384 * over it.
1385 */
1386 continue;
1387 }
1388
1389 head = compound_head(page);
1390 if (!page_cache_get_speculative(head))
1391 goto repeat;
1392
1393 /* The page was split under us? */
1394 if (compound_head(page) != head) {
1395 put_page(head);
1396 goto repeat;
1397 }
1398
1399 /* Has the page moved? */
1400 if (unlikely(page != *slot)) {
1401 put_page(head);
1402 goto repeat;
1403 }
1404
1405 pages[ret] = page;
1406 if (++ret == nr_pages)
1407 break;
1408 }
1409
1410 rcu_read_unlock();
1411 return ret;
1412 }
1413
1414 /**
1415 * find_get_pages_contig - gang contiguous pagecache lookup
1416 * @mapping: The address_space to search
1417 * @index: The starting page index
1418 * @nr_pages: The maximum number of pages
1419 * @pages: Where the resulting pages are placed
1420 *
1421 * find_get_pages_contig() works exactly like find_get_pages(), except
1422 * that the returned number of pages are guaranteed to be contiguous.
1423 *
1424 * find_get_pages_contig() returns the number of pages which were found.
1425 */
1426 unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index,
1427 unsigned int nr_pages, struct page **pages)
1428 {
1429 struct radix_tree_iter iter;
1430 void **slot;
1431 unsigned int ret = 0;
1432
1433 if (unlikely(!nr_pages))
1434 return 0;
1435
1436 rcu_read_lock();
1437 radix_tree_for_each_contig(slot, &mapping->page_tree, &iter, index) {
1438 struct page *head, *page;
1439 repeat:
1440 page = radix_tree_deref_slot(slot);
1441 /* The hole, there no reason to continue */
1442 if (unlikely(!page))
1443 break;
1444
1445 if (radix_tree_exception(page)) {
1446 if (radix_tree_deref_retry(page)) {
1447 slot = radix_tree_iter_retry(&iter);
1448 continue;
1449 }
1450 /*
1451 * A shadow entry of a recently evicted page,
1452 * or a swap entry from shmem/tmpfs. Stop
1453 * looking for contiguous pages.
1454 */
1455 break;
1456 }
1457
1458 head = compound_head(page);
1459 if (!page_cache_get_speculative(head))
1460 goto repeat;
1461
1462 /* The page was split under us? */
1463 if (compound_head(page) != head) {
1464 put_page(head);
1465 goto repeat;
1466 }
1467
1468 /* Has the page moved? */
1469 if (unlikely(page != *slot)) {
1470 put_page(head);
1471 goto repeat;
1472 }
1473
1474 /*
1475 * must check mapping and index after taking the ref.
1476 * otherwise we can get both false positives and false
1477 * negatives, which is just confusing to the caller.
1478 */
1479 if (page->mapping == NULL || page_to_pgoff(page) != iter.index) {
1480 put_page(page);
1481 break;
1482 }
1483
1484 pages[ret] = page;
1485 if (++ret == nr_pages)
1486 break;
1487 }
1488 rcu_read_unlock();
1489 return ret;
1490 }
1491 EXPORT_SYMBOL(find_get_pages_contig);
1492
1493 /**
1494 * find_get_pages_tag - find and return pages that match @tag
1495 * @mapping: the address_space to search
1496 * @index: the starting page index
1497 * @tag: the tag index
1498 * @nr_pages: the maximum number of pages
1499 * @pages: where the resulting pages are placed
1500 *
1501 * Like find_get_pages, except we only return pages which are tagged with
1502 * @tag. We update @index to index the next page for the traversal.
1503 */
1504 unsigned find_get_pages_tag(struct address_space *mapping, pgoff_t *index,
1505 int tag, unsigned int nr_pages, struct page **pages)
1506 {
1507 struct radix_tree_iter iter;
1508 void **slot;
1509 unsigned ret = 0;
1510
1511 if (unlikely(!nr_pages))
1512 return 0;
1513
1514 rcu_read_lock();
1515 radix_tree_for_each_tagged(slot, &mapping->page_tree,
1516 &iter, *index, tag) {
1517 struct page *head, *page;
1518 repeat:
1519 page = radix_tree_deref_slot(slot);
1520 if (unlikely(!page))
1521 continue;
1522
1523 if (radix_tree_exception(page)) {
1524 if (radix_tree_deref_retry(page)) {
1525 slot = radix_tree_iter_retry(&iter);
1526 continue;
1527 }
1528 /*
1529 * A shadow entry of a recently evicted page.
1530 *
1531 * Those entries should never be tagged, but
1532 * this tree walk is lockless and the tags are
1533 * looked up in bulk, one radix tree node at a
1534 * time, so there is a sizable window for page
1535 * reclaim to evict a page we saw tagged.
1536 *
1537 * Skip over it.
1538 */
1539 continue;
1540 }
1541
1542 head = compound_head(page);
1543 if (!page_cache_get_speculative(head))
1544 goto repeat;
1545
1546 /* The page was split under us? */
1547 if (compound_head(page) != head) {
1548 put_page(head);
1549 goto repeat;
1550 }
1551
1552 /* Has the page moved? */
1553 if (unlikely(page != *slot)) {
1554 put_page(head);
1555 goto repeat;
1556 }
1557
1558 pages[ret] = page;
1559 if (++ret == nr_pages)
1560 break;
1561 }
1562
1563 rcu_read_unlock();
1564
1565 if (ret)
1566 *index = pages[ret - 1]->index + 1;
1567
1568 return ret;
1569 }
1570 EXPORT_SYMBOL(find_get_pages_tag);
1571
1572 /**
1573 * find_get_entries_tag - find and return entries that match @tag
1574 * @mapping: the address_space to search
1575 * @start: the starting page cache index
1576 * @tag: the tag index
1577 * @nr_entries: the maximum number of entries
1578 * @entries: where the resulting entries are placed
1579 * @indices: the cache indices corresponding to the entries in @entries
1580 *
1581 * Like find_get_entries, except we only return entries which are tagged with
1582 * @tag.
1583 */
1584 unsigned find_get_entries_tag(struct address_space *mapping, pgoff_t start,
1585 int tag, unsigned int nr_entries,
1586 struct page **entries, pgoff_t *indices)
1587 {
1588 void **slot;
1589 unsigned int ret = 0;
1590 struct radix_tree_iter iter;
1591
1592 if (!nr_entries)
1593 return 0;
1594
1595 rcu_read_lock();
1596 radix_tree_for_each_tagged(slot, &mapping->page_tree,
1597 &iter, start, tag) {
1598 struct page *head, *page;
1599 repeat:
1600 page = radix_tree_deref_slot(slot);
1601 if (unlikely(!page))
1602 continue;
1603 if (radix_tree_exception(page)) {
1604 if (radix_tree_deref_retry(page)) {
1605 slot = radix_tree_iter_retry(&iter);
1606 continue;
1607 }
1608
1609 /*
1610 * A shadow entry of a recently evicted page, a swap
1611 * entry from shmem/tmpfs or a DAX entry. Return it
1612 * without attempting to raise page count.
1613 */
1614 goto export;
1615 }
1616
1617 head = compound_head(page);
1618 if (!page_cache_get_speculative(head))
1619 goto repeat;
1620
1621 /* The page was split under us? */
1622 if (compound_head(page) != head) {
1623 put_page(head);
1624 goto repeat;
1625 }
1626
1627 /* Has the page moved? */
1628 if (unlikely(page != *slot)) {
1629 put_page(head);
1630 goto repeat;
1631 }
1632 export:
1633 indices[ret] = iter.index;
1634 entries[ret] = page;
1635 if (++ret == nr_entries)
1636 break;
1637 }
1638 rcu_read_unlock();
1639 return ret;
1640 }
1641 EXPORT_SYMBOL(find_get_entries_tag);
1642
1643 /*
1644 * CD/DVDs are error prone. When a medium error occurs, the driver may fail
1645 * a _large_ part of the i/o request. Imagine the worst scenario:
1646 *
1647 * ---R__________________________________________B__________
1648 * ^ reading here ^ bad block(assume 4k)
1649 *
1650 * read(R) => miss => readahead(R...B) => media error => frustrating retries
1651 * => failing the whole request => read(R) => read(R+1) =>
1652 * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
1653 * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
1654 * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
1655 *
1656 * It is going insane. Fix it by quickly scaling down the readahead size.
1657 */
1658 static void shrink_readahead_size_eio(struct file *filp,
1659 struct file_ra_state *ra)
1660 {
1661 ra->ra_pages /= 4;
1662 }
1663
1664 /**
1665 * do_generic_file_read - generic file read routine
1666 * @filp: the file to read
1667 * @ppos: current file position
1668 * @iter: data destination
1669 * @written: already copied
1670 *
1671 * This is a generic file read routine, and uses the
1672 * mapping->a_ops->readpage() function for the actual low-level stuff.
1673 *
1674 * This is really ugly. But the goto's actually try to clarify some
1675 * of the logic when it comes to error handling etc.
1676 */
1677 static ssize_t do_generic_file_read(struct file *filp, loff_t *ppos,
1678 struct iov_iter *iter, ssize_t written)
1679 {
1680 struct address_space *mapping = filp->f_mapping;
1681 struct inode *inode = mapping->host;
1682 struct file_ra_state *ra = &filp->f_ra;
1683 pgoff_t index;
1684 pgoff_t last_index;
1685 pgoff_t prev_index;
1686 unsigned long offset; /* offset into pagecache page */
1687 unsigned int prev_offset;
1688 int error = 0;
1689
1690 if (unlikely(*ppos >= inode->i_sb->s_maxbytes))
1691 return -EINVAL;
1692 iov_iter_truncate(iter, inode->i_sb->s_maxbytes);
1693
1694 index = *ppos >> PAGE_SHIFT;
1695 prev_index = ra->prev_pos >> PAGE_SHIFT;
1696 prev_offset = ra->prev_pos & (PAGE_SIZE-1);
1697 last_index = (*ppos + iter->count + PAGE_SIZE-1) >> PAGE_SHIFT;
1698 offset = *ppos & ~PAGE_MASK;
1699
1700 for (;;) {
1701 struct page *page;
1702 pgoff_t end_index;
1703 loff_t isize;
1704 unsigned long nr, ret;
1705
1706 cond_resched();
1707 find_page:
1708 page = find_get_page(mapping, index);
1709 if (!page) {
1710 page_cache_sync_readahead(mapping,
1711 ra, filp,
1712 index, last_index - index);
1713 page = find_get_page(mapping, index);
1714 if (unlikely(page == NULL))
1715 goto no_cached_page;
1716 }
1717 if (PageReadahead(page)) {
1718 page_cache_async_readahead(mapping,
1719 ra, filp, page,
1720 index, last_index - index);
1721 }
1722 if (!PageUptodate(page)) {
1723 /*
1724 * See comment in do_read_cache_page on why
1725 * wait_on_page_locked is used to avoid unnecessarily
1726 * serialisations and why it's safe.
1727 */
1728 error = wait_on_page_locked_killable(page);
1729 if (unlikely(error))
1730 goto readpage_error;
1731 if (PageUptodate(page))
1732 goto page_ok;
1733
1734 if (inode->i_blkbits == PAGE_SHIFT ||
1735 !mapping->a_ops->is_partially_uptodate)
1736 goto page_not_up_to_date;
1737 if (!trylock_page(page))
1738 goto page_not_up_to_date;
1739 /* Did it get truncated before we got the lock? */
1740 if (!page->mapping)
1741 goto page_not_up_to_date_locked;
1742 if (!mapping->a_ops->is_partially_uptodate(page,
1743 offset, iter->count))
1744 goto page_not_up_to_date_locked;
1745 unlock_page(page);
1746 }
1747 page_ok:
1748 /*
1749 * i_size must be checked after we know the page is Uptodate.
1750 *
1751 * Checking i_size after the check allows us to calculate
1752 * the correct value for "nr", which means the zero-filled
1753 * part of the page is not copied back to userspace (unless
1754 * another truncate extends the file - this is desired though).
1755 */
1756
1757 isize = i_size_read(inode);
1758 end_index = (isize - 1) >> PAGE_SHIFT;
1759 if (unlikely(!isize || index > end_index)) {
1760 put_page(page);
1761 goto out;
1762 }
1763
1764 /* nr is the maximum number of bytes to copy from this page */
1765 nr = PAGE_SIZE;
1766 if (index == end_index) {
1767 nr = ((isize - 1) & ~PAGE_MASK) + 1;
1768 if (nr <= offset) {
1769 put_page(page);
1770 goto out;
1771 }
1772 }
1773 nr = nr - offset;
1774
1775 /* If users can be writing to this page using arbitrary
1776 * virtual addresses, take care about potential aliasing
1777 * before reading the page on the kernel side.
1778 */
1779 if (mapping_writably_mapped(mapping))
1780 flush_dcache_page(page);
1781
1782 /*
1783 * When a sequential read accesses a page several times,
1784 * only mark it as accessed the first time.
1785 */
1786 if (prev_index != index || offset != prev_offset)
1787 mark_page_accessed(page);
1788 prev_index = index;
1789
1790 /*
1791 * Ok, we have the page, and it's up-to-date, so
1792 * now we can copy it to user space...
1793 */
1794
1795 ret = copy_page_to_iter(page, offset, nr, iter);
1796 offset += ret;
1797 index += offset >> PAGE_SHIFT;
1798 offset &= ~PAGE_MASK;
1799 prev_offset = offset;
1800
1801 put_page(page);
1802 written += ret;
1803 if (!iov_iter_count(iter))
1804 goto out;
1805 if (ret < nr) {
1806 error = -EFAULT;
1807 goto out;
1808 }
1809 continue;
1810
1811 page_not_up_to_date:
1812 /* Get exclusive access to the page ... */
1813 error = lock_page_killable(page);
1814 if (unlikely(error))
1815 goto readpage_error;
1816
1817 page_not_up_to_date_locked:
1818 /* Did it get truncated before we got the lock? */
1819 if (!page->mapping) {
1820 unlock_page(page);
1821 put_page(page);
1822 continue;
1823 }
1824
1825 /* Did somebody else fill it already? */
1826 if (PageUptodate(page)) {
1827 unlock_page(page);
1828 goto page_ok;
1829 }
1830
1831 readpage:
1832 /*
1833 * A previous I/O error may have been due to temporary
1834 * failures, eg. multipath errors.
1835 * PG_error will be set again if readpage fails.
1836 */
1837 ClearPageError(page);
1838 /* Start the actual read. The read will unlock the page. */
1839 error = mapping->a_ops->readpage(filp, page);
1840
1841 if (unlikely(error)) {
1842 if (error == AOP_TRUNCATED_PAGE) {
1843 put_page(page);
1844 error = 0;
1845 goto find_page;
1846 }
1847 goto readpage_error;
1848 }
1849
1850 if (!PageUptodate(page)) {
1851 error = lock_page_killable(page);
1852 if (unlikely(error))
1853 goto readpage_error;
1854 if (!PageUptodate(page)) {
1855 if (page->mapping == NULL) {
1856 /*
1857 * invalidate_mapping_pages got it
1858 */
1859 unlock_page(page);
1860 put_page(page);
1861 goto find_page;
1862 }
1863 unlock_page(page);
1864 shrink_readahead_size_eio(filp, ra);
1865 error = -EIO;
1866 goto readpage_error;
1867 }
1868 unlock_page(page);
1869 }
1870
1871 goto page_ok;
1872
1873 readpage_error:
1874 /* UHHUH! A synchronous read error occurred. Report it */
1875 put_page(page);
1876 goto out;
1877
1878 no_cached_page:
1879 /*
1880 * Ok, it wasn't cached, so we need to create a new
1881 * page..
1882 */
1883 page = page_cache_alloc_cold(mapping);
1884 if (!page) {
1885 error = -ENOMEM;
1886 goto out;
1887 }
1888 error = add_to_page_cache_lru(page, mapping, index,
1889 mapping_gfp_constraint(mapping, GFP_KERNEL));
1890 if (error) {
1891 put_page(page);
1892 if (error == -EEXIST) {
1893 error = 0;
1894 goto find_page;
1895 }
1896 goto out;
1897 }
1898 goto readpage;
1899 }
1900
1901 out:
1902 ra->prev_pos = prev_index;
1903 ra->prev_pos <<= PAGE_SHIFT;
1904 ra->prev_pos |= prev_offset;
1905
1906 *ppos = ((loff_t)index << PAGE_SHIFT) + offset;
1907 file_accessed(filp);
1908 return written ? written : error;
1909 }
1910
1911 /**
1912 * generic_file_read_iter - generic filesystem read routine
1913 * @iocb: kernel I/O control block
1914 * @iter: destination for the data read
1915 *
1916 * This is the "read_iter()" routine for all filesystems
1917 * that can use the page cache directly.
1918 */
1919 ssize_t
1920 generic_file_read_iter(struct kiocb *iocb, struct iov_iter *iter)
1921 {
1922 struct file *file = iocb->ki_filp;
1923 ssize_t retval = 0;
1924 size_t count = iov_iter_count(iter);
1925
1926 if (!count)
1927 goto out; /* skip atime */
1928
1929 if (iocb->ki_flags & IOCB_DIRECT) {
1930 struct address_space *mapping = file->f_mapping;
1931 struct inode *inode = mapping->host;
1932 struct iov_iter data = *iter;
1933 loff_t size;
1934
1935 size = i_size_read(inode);
1936 retval = filemap_write_and_wait_range(mapping, iocb->ki_pos,
1937 iocb->ki_pos + count - 1);
1938 if (retval < 0)
1939 goto out;
1940
1941 file_accessed(file);
1942
1943 retval = mapping->a_ops->direct_IO(iocb, &data);
1944 if (retval >= 0) {
1945 iocb->ki_pos += retval;
1946 iov_iter_advance(iter, retval);
1947 }
1948
1949 /*
1950 * Btrfs can have a short DIO read if we encounter
1951 * compressed extents, so if there was an error, or if
1952 * we've already read everything we wanted to, or if
1953 * there was a short read because we hit EOF, go ahead
1954 * and return. Otherwise fallthrough to buffered io for
1955 * the rest of the read. Buffered reads will not work for
1956 * DAX files, so don't bother trying.
1957 */
1958 if (retval < 0 || !iov_iter_count(iter) || iocb->ki_pos >= size ||
1959 IS_DAX(inode))
1960 goto out;
1961 }
1962
1963 retval = do_generic_file_read(file, &iocb->ki_pos, iter, retval);
1964 out:
1965 return retval;
1966 }
1967 EXPORT_SYMBOL(generic_file_read_iter);
1968
1969 #ifdef CONFIG_MMU
1970 /**
1971 * page_cache_read - adds requested page to the page cache if not already there
1972 * @file: file to read
1973 * @offset: page index
1974 * @gfp_mask: memory allocation flags
1975 *
1976 * This adds the requested page to the page cache if it isn't already there,
1977 * and schedules an I/O to read in its contents from disk.
1978 */
1979 static int page_cache_read(struct file *file, pgoff_t offset, gfp_t gfp_mask)
1980 {
1981 struct address_space *mapping = file->f_mapping;
1982 struct page *page;
1983 int ret;
1984
1985 do {
1986 page = __page_cache_alloc(gfp_mask|__GFP_COLD);
1987 if (!page)
1988 return -ENOMEM;
1989
1990 ret = add_to_page_cache_lru(page, mapping, offset, gfp_mask & GFP_KERNEL);
1991 if (ret == 0)
1992 ret = mapping->a_ops->readpage(file, page);
1993 else if (ret == -EEXIST)
1994 ret = 0; /* losing race to add is OK */
1995
1996 put_page(page);
1997
1998 } while (ret == AOP_TRUNCATED_PAGE);
1999
2000 return ret;
2001 }
2002
2003 #define MMAP_LOTSAMISS (100)
2004
2005 /*
2006 * Synchronous readahead happens when we don't even find
2007 * a page in the page cache at all.
2008 */
2009 static void do_sync_mmap_readahead(struct vm_area_struct *vma,
2010 struct file_ra_state *ra,
2011 struct file *file,
2012 pgoff_t offset)
2013 {
2014 struct address_space *mapping = file->f_mapping;
2015
2016 /* If we don't want any read-ahead, don't bother */
2017 if (vma->vm_flags & VM_RAND_READ)
2018 return;
2019 if (!ra->ra_pages)
2020 return;
2021
2022 if (vma->vm_flags & VM_SEQ_READ) {
2023 page_cache_sync_readahead(mapping, ra, file, offset,
2024 ra->ra_pages);
2025 return;
2026 }
2027
2028 /* Avoid banging the cache line if not needed */
2029 if (ra->mmap_miss < MMAP_LOTSAMISS * 10)
2030 ra->mmap_miss++;
2031
2032 /*
2033 * Do we miss much more than hit in this file? If so,
2034 * stop bothering with read-ahead. It will only hurt.
2035 */
2036 if (ra->mmap_miss > MMAP_LOTSAMISS)
2037 return;
2038
2039 /*
2040 * mmap read-around
2041 */
2042 ra->start = max_t(long, 0, offset - ra->ra_pages / 2);
2043 ra->size = ra->ra_pages;
2044 ra->async_size = ra->ra_pages / 4;
2045 ra_submit(ra, mapping, file);
2046 }
2047
2048 /*
2049 * Asynchronous readahead happens when we find the page and PG_readahead,
2050 * so we want to possibly extend the readahead further..
2051 */
2052 static void do_async_mmap_readahead(struct vm_area_struct *vma,
2053 struct file_ra_state *ra,
2054 struct file *file,
2055 struct page *page,
2056 pgoff_t offset)
2057 {
2058 struct address_space *mapping = file->f_mapping;
2059
2060 /* If we don't want any read-ahead, don't bother */
2061 if (vma->vm_flags & VM_RAND_READ)
2062 return;
2063 if (ra->mmap_miss > 0)
2064 ra->mmap_miss--;
2065 if (PageReadahead(page))
2066 page_cache_async_readahead(mapping, ra, file,
2067 page, offset, ra->ra_pages);
2068 }
2069
2070 /**
2071 * filemap_fault - read in file data for page fault handling
2072 * @vma: vma in which the fault was taken
2073 * @vmf: struct vm_fault containing details of the fault
2074 *
2075 * filemap_fault() is invoked via the vma operations vector for a
2076 * mapped memory region to read in file data during a page fault.
2077 *
2078 * The goto's are kind of ugly, but this streamlines the normal case of having
2079 * it in the page cache, and handles the special cases reasonably without
2080 * having a lot of duplicated code.
2081 *
2082 * vma->vm_mm->mmap_sem must be held on entry.
2083 *
2084 * If our return value has VM_FAULT_RETRY set, it's because
2085 * lock_page_or_retry() returned 0.
2086 * The mmap_sem has usually been released in this case.
2087 * See __lock_page_or_retry() for the exception.
2088 *
2089 * If our return value does not have VM_FAULT_RETRY set, the mmap_sem
2090 * has not been released.
2091 *
2092 * We never return with VM_FAULT_RETRY and a bit from VM_FAULT_ERROR set.
2093 */
2094 int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
2095 {
2096 int error;
2097 struct file *file = vma->vm_file;
2098 struct address_space *mapping = file->f_mapping;
2099 struct file_ra_state *ra = &file->f_ra;
2100 struct inode *inode = mapping->host;
2101 pgoff_t offset = vmf->pgoff;
2102 struct page *page;
2103 loff_t size;
2104 int ret = 0;
2105
2106 size = round_up(i_size_read(inode), PAGE_SIZE);
2107 if (offset >= size >> PAGE_SHIFT)
2108 return VM_FAULT_SIGBUS;
2109
2110 /*
2111 * Do we have something in the page cache already?
2112 */
2113 page = find_get_page(mapping, offset);
2114 if (likely(page) && !(vmf->flags & FAULT_FLAG_TRIED)) {
2115 /*
2116 * We found the page, so try async readahead before
2117 * waiting for the lock.
2118 */
2119 do_async_mmap_readahead(vma, ra, file, page, offset);
2120 } else if (!page) {
2121 /* No page in the page cache at all */
2122 do_sync_mmap_readahead(vma, ra, file, offset);
2123 count_vm_event(PGMAJFAULT);
2124 mem_cgroup_count_vm_event(vma->vm_mm, PGMAJFAULT);
2125 ret = VM_FAULT_MAJOR;
2126 retry_find:
2127 page = find_get_page(mapping, offset);
2128 if (!page)
2129 goto no_cached_page;
2130 }
2131
2132 if (!lock_page_or_retry(page, vma->vm_mm, vmf->flags)) {
2133 put_page(page);
2134 return ret | VM_FAULT_RETRY;
2135 }
2136
2137 /* Did it get truncated? */
2138 if (unlikely(page->mapping != mapping)) {
2139 unlock_page(page);
2140 put_page(page);
2141 goto retry_find;
2142 }
2143 VM_BUG_ON_PAGE(page->index != offset, page);
2144
2145 /*
2146 * We have a locked page in the page cache, now we need to check
2147 * that it's up-to-date. If not, it is going to be due to an error.
2148 */
2149 if (unlikely(!PageUptodate(page)))
2150 goto page_not_uptodate;
2151
2152 /*
2153 * Found the page and have a reference on it.
2154 * We must recheck i_size under page lock.
2155 */
2156 size = round_up(i_size_read(inode), PAGE_SIZE);
2157 if (unlikely(offset >= size >> PAGE_SHIFT)) {
2158 unlock_page(page);
2159 put_page(page);
2160 return VM_FAULT_SIGBUS;
2161 }
2162
2163 vmf->page = page;
2164 return ret | VM_FAULT_LOCKED;
2165
2166 no_cached_page:
2167 /*
2168 * We're only likely to ever get here if MADV_RANDOM is in
2169 * effect.
2170 */
2171 error = page_cache_read(file, offset, vmf->gfp_mask);
2172
2173 /*
2174 * The page we want has now been added to the page cache.
2175 * In the unlikely event that someone removed it in the
2176 * meantime, we'll just come back here and read it again.
2177 */
2178 if (error >= 0)
2179 goto retry_find;
2180
2181 /*
2182 * An error return from page_cache_read can result if the
2183 * system is low on memory, or a problem occurs while trying
2184 * to schedule I/O.
2185 */
2186 if (error == -ENOMEM)
2187 return VM_FAULT_OOM;
2188 return VM_FAULT_SIGBUS;
2189
2190 page_not_uptodate:
2191 /*
2192 * Umm, take care of errors if the page isn't up-to-date.
2193 * Try to re-read it _once_. We do this synchronously,
2194 * because there really aren't any performance issues here
2195 * and we need to check for errors.
2196 */
2197 ClearPageError(page);
2198 error = mapping->a_ops->readpage(file, page);
2199 if (!error) {
2200 wait_on_page_locked(page);
2201 if (!PageUptodate(page))
2202 error = -EIO;
2203 }
2204 put_page(page);
2205
2206 if (!error || error == AOP_TRUNCATED_PAGE)
2207 goto retry_find;
2208
2209 /* Things didn't work out. Return zero to tell the mm layer so. */
2210 shrink_readahead_size_eio(file, ra);
2211 return VM_FAULT_SIGBUS;
2212 }
2213 EXPORT_SYMBOL(filemap_fault);
2214
2215 void filemap_map_pages(struct fault_env *fe,
2216 pgoff_t start_pgoff, pgoff_t end_pgoff)
2217 {
2218 struct radix_tree_iter iter;
2219 void **slot;
2220 struct file *file = fe->vma->vm_file;
2221 struct address_space *mapping = file->f_mapping;
2222 pgoff_t last_pgoff = start_pgoff;
2223 loff_t size;
2224 struct page *head, *page;
2225
2226 rcu_read_lock();
2227 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter,
2228 start_pgoff) {
2229 if (iter.index > end_pgoff)
2230 break;
2231 repeat:
2232 page = radix_tree_deref_slot(slot);
2233 if (unlikely(!page))
2234 goto next;
2235 if (radix_tree_exception(page)) {
2236 if (radix_tree_deref_retry(page)) {
2237 slot = radix_tree_iter_retry(&iter);
2238 continue;
2239 }
2240 goto next;
2241 }
2242
2243 head = compound_head(page);
2244 if (!page_cache_get_speculative(head))
2245 goto repeat;
2246
2247 /* The page was split under us? */
2248 if (compound_head(page) != head) {
2249 put_page(head);
2250 goto repeat;
2251 }
2252
2253 /* Has the page moved? */
2254 if (unlikely(page != *slot)) {
2255 put_page(head);
2256 goto repeat;
2257 }
2258
2259 if (!PageUptodate(page) ||
2260 PageReadahead(page) ||
2261 PageHWPoison(page))
2262 goto skip;
2263 if (!trylock_page(page))
2264 goto skip;
2265
2266 if (page->mapping != mapping || !PageUptodate(page))
2267 goto unlock;
2268
2269 size = round_up(i_size_read(mapping->host), PAGE_SIZE);
2270 if (page->index >= size >> PAGE_SHIFT)
2271 goto unlock;
2272
2273 if (file->f_ra.mmap_miss > 0)
2274 file->f_ra.mmap_miss--;
2275
2276 fe->address += (iter.index - last_pgoff) << PAGE_SHIFT;
2277 if (fe->pte)
2278 fe->pte += iter.index - last_pgoff;
2279 last_pgoff = iter.index;
2280 if (alloc_set_pte(fe, NULL, page))
2281 goto unlock;
2282 unlock_page(page);
2283 goto next;
2284 unlock:
2285 unlock_page(page);
2286 skip:
2287 put_page(page);
2288 next:
2289 /* Huge page is mapped? No need to proceed. */
2290 if (pmd_trans_huge(*fe->pmd))
2291 break;
2292 if (iter.index == end_pgoff)
2293 break;
2294 }
2295 rcu_read_unlock();
2296 }
2297 EXPORT_SYMBOL(filemap_map_pages);
2298
2299 int filemap_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
2300 {
2301 struct page *page = vmf->page;
2302 struct inode *inode = file_inode(vma->vm_file);
2303 int ret = VM_FAULT_LOCKED;
2304
2305 sb_start_pagefault(inode->i_sb);
2306 file_update_time(vma->vm_file);
2307 lock_page(page);
2308 if (page->mapping != inode->i_mapping) {
2309 unlock_page(page);
2310 ret = VM_FAULT_NOPAGE;
2311 goto out;
2312 }
2313 /*
2314 * We mark the page dirty already here so that when freeze is in
2315 * progress, we are guaranteed that writeback during freezing will
2316 * see the dirty page and writeprotect it again.
2317 */
2318 set_page_dirty(page);
2319 wait_for_stable_page(page);
2320 out:
2321 sb_end_pagefault(inode->i_sb);
2322 return ret;
2323 }
2324 EXPORT_SYMBOL(filemap_page_mkwrite);
2325
2326 const struct vm_operations_struct generic_file_vm_ops = {
2327 .fault = filemap_fault,
2328 .map_pages = filemap_map_pages,
2329 .page_mkwrite = filemap_page_mkwrite,
2330 };
2331
2332 /* This is used for a general mmap of a disk file */
2333
2334 int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
2335 {
2336 struct address_space *mapping = file->f_mapping;
2337
2338 if (!mapping->a_ops->readpage)
2339 return -ENOEXEC;
2340 file_accessed(file);
2341 vma->vm_ops = &generic_file_vm_ops;
2342 return 0;
2343 }
2344
2345 /*
2346 * This is for filesystems which do not implement ->writepage.
2347 */
2348 int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
2349 {
2350 if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE))
2351 return -EINVAL;
2352 return generic_file_mmap(file, vma);
2353 }
2354 #else
2355 int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
2356 {
2357 return -ENOSYS;
2358 }
2359 int generic_file_readonly_mmap(struct file * file, struct vm_area_struct * vma)
2360 {
2361 return -ENOSYS;
2362 }
2363 #endif /* CONFIG_MMU */
2364
2365 EXPORT_SYMBOL(generic_file_mmap);
2366 EXPORT_SYMBOL(generic_file_readonly_mmap);
2367
2368 static struct page *wait_on_page_read(struct page *page)
2369 {
2370 if (!IS_ERR(page)) {
2371 wait_on_page_locked(page);
2372 if (!PageUptodate(page)) {
2373 put_page(page);
2374 page = ERR_PTR(-EIO);
2375 }
2376 }
2377 return page;
2378 }
2379
2380 static struct page *do_read_cache_page(struct address_space *mapping,
2381 pgoff_t index,
2382 int (*filler)(void *, struct page *),
2383 void *data,
2384 gfp_t gfp)
2385 {
2386 struct page *page;
2387 int err;
2388 repeat:
2389 page = find_get_page(mapping, index);
2390 if (!page) {
2391 page = __page_cache_alloc(gfp | __GFP_COLD);
2392 if (!page)
2393 return ERR_PTR(-ENOMEM);
2394 err = add_to_page_cache_lru(page, mapping, index, gfp);
2395 if (unlikely(err)) {
2396 put_page(page);
2397 if (err == -EEXIST)
2398 goto repeat;
2399 /* Presumably ENOMEM for radix tree node */
2400 return ERR_PTR(err);
2401 }
2402
2403 filler:
2404 err = filler(data, page);
2405 if (err < 0) {
2406 put_page(page);
2407 return ERR_PTR(err);
2408 }
2409
2410 page = wait_on_page_read(page);
2411 if (IS_ERR(page))
2412 return page;
2413 goto out;
2414 }
2415 if (PageUptodate(page))
2416 goto out;
2417
2418 /*
2419 * Page is not up to date and may be locked due one of the following
2420 * case a: Page is being filled and the page lock is held
2421 * case b: Read/write error clearing the page uptodate status
2422 * case c: Truncation in progress (page locked)
2423 * case d: Reclaim in progress
2424 *
2425 * Case a, the page will be up to date when the page is unlocked.
2426 * There is no need to serialise on the page lock here as the page
2427 * is pinned so the lock gives no additional protection. Even if the
2428 * the page is truncated, the data is still valid if PageUptodate as
2429 * it's a race vs truncate race.
2430 * Case b, the page will not be up to date
2431 * Case c, the page may be truncated but in itself, the data may still
2432 * be valid after IO completes as it's a read vs truncate race. The
2433 * operation must restart if the page is not uptodate on unlock but
2434 * otherwise serialising on page lock to stabilise the mapping gives
2435 * no additional guarantees to the caller as the page lock is
2436 * released before return.
2437 * Case d, similar to truncation. If reclaim holds the page lock, it
2438 * will be a race with remove_mapping that determines if the mapping
2439 * is valid on unlock but otherwise the data is valid and there is
2440 * no need to serialise with page lock.
2441 *
2442 * As the page lock gives no additional guarantee, we optimistically
2443 * wait on the page to be unlocked and check if it's up to date and
2444 * use the page if it is. Otherwise, the page lock is required to
2445 * distinguish between the different cases. The motivation is that we
2446 * avoid spurious serialisations and wakeups when multiple processes
2447 * wait on the same page for IO to complete.
2448 */
2449 wait_on_page_locked(page);
2450 if (PageUptodate(page))
2451 goto out;
2452
2453 /* Distinguish between all the cases under the safety of the lock */
2454 lock_page(page);
2455
2456 /* Case c or d, restart the operation */
2457 if (!page->mapping) {
2458 unlock_page(page);
2459 put_page(page);
2460 goto repeat;
2461 }
2462
2463 /* Someone else locked and filled the page in a very small window */
2464 if (PageUptodate(page)) {
2465 unlock_page(page);
2466 goto out;
2467 }
2468 goto filler;
2469
2470 out:
2471 mark_page_accessed(page);
2472 return page;
2473 }
2474
2475 /**
2476 * read_cache_page - read into page cache, fill it if needed
2477 * @mapping: the page's address_space
2478 * @index: the page index
2479 * @filler: function to perform the read
2480 * @data: first arg to filler(data, page) function, often left as NULL
2481 *
2482 * Read into the page cache. If a page already exists, and PageUptodate() is
2483 * not set, try to fill the page and wait for it to become unlocked.
2484 *
2485 * If the page does not get brought uptodate, return -EIO.
2486 */
2487 struct page *read_cache_page(struct address_space *mapping,
2488 pgoff_t index,
2489 int (*filler)(void *, struct page *),
2490 void *data)
2491 {
2492 return do_read_cache_page(mapping, index, filler, data, mapping_gfp_mask(mapping));
2493 }
2494 EXPORT_SYMBOL(read_cache_page);
2495
2496 /**
2497 * read_cache_page_gfp - read into page cache, using specified page allocation flags.
2498 * @mapping: the page's address_space
2499 * @index: the page index
2500 * @gfp: the page allocator flags to use if allocating
2501 *
2502 * This is the same as "read_mapping_page(mapping, index, NULL)", but with
2503 * any new page allocations done using the specified allocation flags.
2504 *
2505 * If the page does not get brought uptodate, return -EIO.
2506 */
2507 struct page *read_cache_page_gfp(struct address_space *mapping,
2508 pgoff_t index,
2509 gfp_t gfp)
2510 {
2511 filler_t *filler = (filler_t *)mapping->a_ops->readpage;
2512
2513 return do_read_cache_page(mapping, index, filler, NULL, gfp);
2514 }
2515 EXPORT_SYMBOL(read_cache_page_gfp);
2516
2517 /*
2518 * Performs necessary checks before doing a write
2519 *
2520 * Can adjust writing position or amount of bytes to write.
2521 * Returns appropriate error code that caller should return or
2522 * zero in case that write should be allowed.
2523 */
2524 inline ssize_t generic_write_checks(struct kiocb *iocb, struct iov_iter *from)
2525 {
2526 struct file *file = iocb->ki_filp;
2527 struct inode *inode = file->f_mapping->host;
2528 unsigned long limit = rlimit(RLIMIT_FSIZE);
2529 loff_t pos;
2530
2531 if (!iov_iter_count(from))
2532 return 0;
2533
2534 /* FIXME: this is for backwards compatibility with 2.4 */
2535 if (iocb->ki_flags & IOCB_APPEND)
2536 iocb->ki_pos = i_size_read(inode);
2537
2538 pos = iocb->ki_pos;
2539
2540 if (limit != RLIM_INFINITY) {
2541 if (iocb->ki_pos >= limit) {
2542 send_sig(SIGXFSZ, current, 0);
2543 return -EFBIG;
2544 }
2545 iov_iter_truncate(from, limit - (unsigned long)pos);
2546 }
2547
2548 /*
2549 * LFS rule
2550 */
2551 if (unlikely(pos + iov_iter_count(from) > MAX_NON_LFS &&
2552 !(file->f_flags & O_LARGEFILE))) {
2553 if (pos >= MAX_NON_LFS)
2554 return -EFBIG;
2555 iov_iter_truncate(from, MAX_NON_LFS - (unsigned long)pos);
2556 }
2557
2558 /*
2559 * Are we about to exceed the fs block limit ?
2560 *
2561 * If we have written data it becomes a short write. If we have
2562 * exceeded without writing data we send a signal and return EFBIG.
2563 * Linus frestrict idea will clean these up nicely..
2564 */
2565 if (unlikely(pos >= inode->i_sb->s_maxbytes))
2566 return -EFBIG;
2567
2568 iov_iter_truncate(from, inode->i_sb->s_maxbytes - pos);
2569 return iov_iter_count(from);
2570 }
2571 EXPORT_SYMBOL(generic_write_checks);
2572
2573 int pagecache_write_begin(struct file *file, struct address_space *mapping,
2574 loff_t pos, unsigned len, unsigned flags,
2575 struct page **pagep, void **fsdata)
2576 {
2577 const struct address_space_operations *aops = mapping->a_ops;
2578
2579 return aops->write_begin(file, mapping, pos, len, flags,
2580 pagep, fsdata);
2581 }
2582 EXPORT_SYMBOL(pagecache_write_begin);
2583
2584 int pagecache_write_end(struct file *file, struct address_space *mapping,
2585 loff_t pos, unsigned len, unsigned copied,
2586 struct page *page, void *fsdata)
2587 {
2588 const struct address_space_operations *aops = mapping->a_ops;
2589
2590 return aops->write_end(file, mapping, pos, len, copied, page, fsdata);
2591 }
2592 EXPORT_SYMBOL(pagecache_write_end);
2593
2594 ssize_t
2595 generic_file_direct_write(struct kiocb *iocb, struct iov_iter *from)
2596 {
2597 struct file *file = iocb->ki_filp;
2598 struct address_space *mapping = file->f_mapping;
2599 struct inode *inode = mapping->host;
2600 loff_t pos = iocb->ki_pos;
2601 ssize_t written;
2602 size_t write_len;
2603 pgoff_t end;
2604 struct iov_iter data;
2605
2606 write_len = iov_iter_count(from);
2607 end = (pos + write_len - 1) >> PAGE_SHIFT;
2608
2609 written = filemap_write_and_wait_range(mapping, pos, pos + write_len - 1);
2610 if (written)
2611 goto out;
2612
2613 /*
2614 * After a write we want buffered reads to be sure to go to disk to get
2615 * the new data. We invalidate clean cached page from the region we're
2616 * about to write. We do this *before* the write so that we can return
2617 * without clobbering -EIOCBQUEUED from ->direct_IO().
2618 */
2619 if (mapping->nrpages) {
2620 written = invalidate_inode_pages2_range(mapping,
2621 pos >> PAGE_SHIFT, end);
2622 /*
2623 * If a page can not be invalidated, return 0 to fall back
2624 * to buffered write.
2625 */
2626 if (written) {
2627 if (written == -EBUSY)
2628 return 0;
2629 goto out;
2630 }
2631 }
2632
2633 data = *from;
2634 written = mapping->a_ops->direct_IO(iocb, &data);
2635
2636 /*
2637 * Finally, try again to invalidate clean pages which might have been
2638 * cached by non-direct readahead, or faulted in by get_user_pages()
2639 * if the source of the write was an mmap'ed region of the file
2640 * we're writing. Either one is a pretty crazy thing to do,
2641 * so we don't support it 100%. If this invalidation
2642 * fails, tough, the write still worked...
2643 */
2644 if (mapping->nrpages) {
2645 invalidate_inode_pages2_range(mapping,
2646 pos >> PAGE_SHIFT, end);
2647 }
2648
2649 if (written > 0) {
2650 pos += written;
2651 iov_iter_advance(from, written);
2652 if (pos > i_size_read(inode) && !S_ISBLK(inode->i_mode)) {
2653 i_size_write(inode, pos);
2654 mark_inode_dirty(inode);
2655 }
2656 iocb->ki_pos = pos;
2657 }
2658 out:
2659 return written;
2660 }
2661 EXPORT_SYMBOL(generic_file_direct_write);
2662
2663 /*
2664 * Find or create a page at the given pagecache position. Return the locked
2665 * page. This function is specifically for buffered writes.
2666 */
2667 struct page *grab_cache_page_write_begin(struct address_space *mapping,
2668 pgoff_t index, unsigned flags)
2669 {
2670 struct page *page;
2671 int fgp_flags = FGP_LOCK|FGP_WRITE|FGP_CREAT;
2672
2673 if (flags & AOP_FLAG_NOFS)
2674 fgp_flags |= FGP_NOFS;
2675
2676 page = pagecache_get_page(mapping, index, fgp_flags,
2677 mapping_gfp_mask(mapping));
2678 if (page)
2679 wait_for_stable_page(page);
2680
2681 return page;
2682 }
2683 EXPORT_SYMBOL(grab_cache_page_write_begin);
2684
2685 ssize_t generic_perform_write(struct file *file,
2686 struct iov_iter *i, loff_t pos)
2687 {
2688 struct address_space *mapping = file->f_mapping;
2689 const struct address_space_operations *a_ops = mapping->a_ops;
2690 long status = 0;
2691 ssize_t written = 0;
2692 unsigned int flags = 0;
2693
2694 /*
2695 * Copies from kernel address space cannot fail (NFSD is a big user).
2696 */
2697 if (!iter_is_iovec(i))
2698 flags |= AOP_FLAG_UNINTERRUPTIBLE;
2699
2700 do {
2701 struct page *page;
2702 unsigned long offset; /* Offset into pagecache page */
2703 unsigned long bytes; /* Bytes to write to page */
2704 size_t copied; /* Bytes copied from user */
2705 void *fsdata;
2706
2707 offset = (pos & (PAGE_SIZE - 1));
2708 bytes = min_t(unsigned long, PAGE_SIZE - offset,
2709 iov_iter_count(i));
2710
2711 again:
2712 /*
2713 * Bring in the user page that we will copy from _first_.
2714 * Otherwise there's a nasty deadlock on copying from the
2715 * same page as we're writing to, without it being marked
2716 * up-to-date.
2717 *
2718 * Not only is this an optimisation, but it is also required
2719 * to check that the address is actually valid, when atomic
2720 * usercopies are used, below.
2721 */
2722 if (unlikely(iov_iter_fault_in_readable(i, bytes))) {
2723 status = -EFAULT;
2724 break;
2725 }
2726
2727 if (fatal_signal_pending(current)) {
2728 status = -EINTR;
2729 break;
2730 }
2731
2732 status = a_ops->write_begin(file, mapping, pos, bytes, flags,
2733 &page, &fsdata);
2734 if (unlikely(status < 0))
2735 break;
2736
2737 if (mapping_writably_mapped(mapping))
2738 flush_dcache_page(page);
2739
2740 copied = iov_iter_copy_from_user_atomic(page, i, offset, bytes);
2741 flush_dcache_page(page);
2742
2743 status = a_ops->write_end(file, mapping, pos, bytes, copied,
2744 page, fsdata);
2745 if (unlikely(status < 0))
2746 break;
2747 copied = status;
2748
2749 cond_resched();
2750
2751 iov_iter_advance(i, copied);
2752 if (unlikely(copied == 0)) {
2753 /*
2754 * If we were unable to copy any data at all, we must
2755 * fall back to a single segment length write.
2756 *
2757 * If we didn't fallback here, we could livelock
2758 * because not all segments in the iov can be copied at
2759 * once without a pagefault.
2760 */
2761 bytes = min_t(unsigned long, PAGE_SIZE - offset,
2762 iov_iter_single_seg_count(i));
2763 goto again;
2764 }
2765 pos += copied;
2766 written += copied;
2767
2768 balance_dirty_pages_ratelimited(mapping);
2769 } while (iov_iter_count(i));
2770
2771 return written ? written : status;
2772 }
2773 EXPORT_SYMBOL(generic_perform_write);
2774
2775 /**
2776 * __generic_file_write_iter - write data to a file
2777 * @iocb: IO state structure (file, offset, etc.)
2778 * @from: iov_iter with data to write
2779 *
2780 * This function does all the work needed for actually writing data to a
2781 * file. It does all basic checks, removes SUID from the file, updates
2782 * modification times and calls proper subroutines depending on whether we
2783 * do direct IO or a standard buffered write.
2784 *
2785 * It expects i_mutex to be grabbed unless we work on a block device or similar
2786 * object which does not need locking at all.
2787 *
2788 * This function does *not* take care of syncing data in case of O_SYNC write.
2789 * A caller has to handle it. This is mainly due to the fact that we want to
2790 * avoid syncing under i_mutex.
2791 */
2792 ssize_t __generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
2793 {
2794 struct file *file = iocb->ki_filp;
2795 struct address_space * mapping = file->f_mapping;
2796 struct inode *inode = mapping->host;
2797 ssize_t written = 0;
2798 ssize_t err;
2799 ssize_t status;
2800
2801 /* We can write back this queue in page reclaim */
2802 current->backing_dev_info = inode_to_bdi(inode);
2803 err = file_remove_privs(file);
2804 if (err)
2805 goto out;
2806
2807 err = file_update_time(file);
2808 if (err)
2809 goto out;
2810
2811 if (iocb->ki_flags & IOCB_DIRECT) {
2812 loff_t pos, endbyte;
2813
2814 written = generic_file_direct_write(iocb, from);
2815 /*
2816 * If the write stopped short of completing, fall back to
2817 * buffered writes. Some filesystems do this for writes to
2818 * holes, for example. For DAX files, a buffered write will
2819 * not succeed (even if it did, DAX does not handle dirty
2820 * page-cache pages correctly).
2821 */
2822 if (written < 0 || !iov_iter_count(from) || IS_DAX(inode))
2823 goto out;
2824
2825 status = generic_perform_write(file, from, pos = iocb->ki_pos);
2826 /*
2827 * If generic_perform_write() returned a synchronous error
2828 * then we want to return the number of bytes which were
2829 * direct-written, or the error code if that was zero. Note
2830 * that this differs from normal direct-io semantics, which
2831 * will return -EFOO even if some bytes were written.
2832 */
2833 if (unlikely(status < 0)) {
2834 err = status;
2835 goto out;
2836 }
2837 /*
2838 * We need to ensure that the page cache pages are written to
2839 * disk and invalidated to preserve the expected O_DIRECT
2840 * semantics.
2841 */
2842 endbyte = pos + status - 1;
2843 err = filemap_write_and_wait_range(mapping, pos, endbyte);
2844 if (err == 0) {
2845 iocb->ki_pos = endbyte + 1;
2846 written += status;
2847 invalidate_mapping_pages(mapping,
2848 pos >> PAGE_SHIFT,
2849 endbyte >> PAGE_SHIFT);
2850 } else {
2851 /*
2852 * We don't know how much we wrote, so just return
2853 * the number of bytes which were direct-written
2854 */
2855 }
2856 } else {
2857 written = generic_perform_write(file, from, iocb->ki_pos);
2858 if (likely(written > 0))
2859 iocb->ki_pos += written;
2860 }
2861 out:
2862 current->backing_dev_info = NULL;
2863 return written ? written : err;
2864 }
2865 EXPORT_SYMBOL(__generic_file_write_iter);
2866
2867 /**
2868 * generic_file_write_iter - write data to a file
2869 * @iocb: IO state structure
2870 * @from: iov_iter with data to write
2871 *
2872 * This is a wrapper around __generic_file_write_iter() to be used by most
2873 * filesystems. It takes care of syncing the file in case of O_SYNC file
2874 * and acquires i_mutex as needed.
2875 */
2876 ssize_t generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
2877 {
2878 struct file *file = iocb->ki_filp;
2879 struct inode *inode = file->f_mapping->host;
2880 ssize_t ret;
2881
2882 inode_lock(inode);
2883 ret = generic_write_checks(iocb, from);
2884 if (ret > 0)
2885 ret = __generic_file_write_iter(iocb, from);
2886 inode_unlock(inode);
2887
2888 if (ret > 0)
2889 ret = generic_write_sync(iocb, ret);
2890 return ret;
2891 }
2892 EXPORT_SYMBOL(generic_file_write_iter);
2893
2894 /**
2895 * try_to_release_page() - release old fs-specific metadata on a page
2896 *
2897 * @page: the page which the kernel is trying to free
2898 * @gfp_mask: memory allocation flags (and I/O mode)
2899 *
2900 * The address_space is to try to release any data against the page
2901 * (presumably at page->private). If the release was successful, return `1'.
2902 * Otherwise return zero.
2903 *
2904 * This may also be called if PG_fscache is set on a page, indicating that the
2905 * page is known to the local caching routines.
2906 *
2907 * The @gfp_mask argument specifies whether I/O may be performed to release
2908 * this page (__GFP_IO), and whether the call may block (__GFP_RECLAIM & __GFP_FS).
2909 *
2910 */
2911 int try_to_release_page(struct page *page, gfp_t gfp_mask)
2912 {
2913 struct address_space * const mapping = page->mapping;
2914
2915 BUG_ON(!PageLocked(page));
2916 if (PageWriteback(page))
2917 return 0;
2918
2919 if (mapping && mapping->a_ops->releasepage)
2920 return mapping->a_ops->releasepage(page, gfp_mask);
2921 return try_to_free_buffers(page);
2922 }
2923
2924 EXPORT_SYMBOL(try_to_release_page);