Loading...
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/module.h>
13#include <linux/compiler.h>
14#include <linux/fs.h>
15#include <linux/uaccess.h>
16#include <linux/aio.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/syscalls.h>
33#include <linux/cpuset.h>
34#include <linux/hardirq.h> /* for BUG_ON(!in_atomic()) only */
35#include <linux/memcontrol.h>
36#include <linux/cleancache.h>
37#include "internal.h"
38
39/*
40 * FIXME: remove all knowledge of the buffer layer from the core VM
41 */
42#include <linux/buffer_head.h> /* for try_to_free_buffers */
43
44#include <asm/mman.h>
45
46/*
47 * Shared mappings implemented 30.11.1994. It's not fully working yet,
48 * though.
49 *
50 * Shared mappings now work. 15.8.1995 Bruno.
51 *
52 * finished 'unifying' the page and buffer cache and SMP-threaded the
53 * page-cache, 21.05.1999, Ingo Molnar <mingo@redhat.com>
54 *
55 * SMP-threaded pagemap-LRU 1999, Andrea Arcangeli <andrea@suse.de>
56 */
57
58/*
59 * Lock ordering:
60 *
61 * ->i_mmap_mutex (truncate_pagecache)
62 * ->private_lock (__free_pte->__set_page_dirty_buffers)
63 * ->swap_lock (exclusive_swap_page, others)
64 * ->mapping->tree_lock
65 *
66 * ->i_mutex
67 * ->i_mmap_mutex (truncate->unmap_mapping_range)
68 *
69 * ->mmap_sem
70 * ->i_mmap_mutex
71 * ->page_table_lock or pte_lock (various, mainly in memory.c)
72 * ->mapping->tree_lock (arch-dependent flush_dcache_mmap_lock)
73 *
74 * ->mmap_sem
75 * ->lock_page (access_process_vm)
76 *
77 * ->i_mutex (generic_file_buffered_write)
78 * ->mmap_sem (fault_in_pages_readable->do_page_fault)
79 *
80 * bdi->wb.list_lock
81 * sb_lock (fs/fs-writeback.c)
82 * ->mapping->tree_lock (__sync_single_inode)
83 *
84 * ->i_mmap_mutex
85 * ->anon_vma.lock (vma_adjust)
86 *
87 * ->anon_vma.lock
88 * ->page_table_lock or pte_lock (anon_vma_prepare and various)
89 *
90 * ->page_table_lock or pte_lock
91 * ->swap_lock (try_to_unmap_one)
92 * ->private_lock (try_to_unmap_one)
93 * ->tree_lock (try_to_unmap_one)
94 * ->zone.lru_lock (follow_page->mark_page_accessed)
95 * ->zone.lru_lock (check_pte_range->isolate_lru_page)
96 * ->private_lock (page_remove_rmap->set_page_dirty)
97 * ->tree_lock (page_remove_rmap->set_page_dirty)
98 * bdi.wb->list_lock (page_remove_rmap->set_page_dirty)
99 * ->inode->i_lock (page_remove_rmap->set_page_dirty)
100 * bdi.wb->list_lock (zap_pte_range->set_page_dirty)
101 * ->inode->i_lock (zap_pte_range->set_page_dirty)
102 * ->private_lock (zap_pte_range->__set_page_dirty_buffers)
103 *
104 * (code doesn't rely on that order, so you could switch it around)
105 * ->tasklist_lock (memory_failure, collect_procs_ao)
106 * ->i_mmap_mutex
107 */
108
109/*
110 * Delete a page from the page cache and free it. Caller has to make
111 * sure the page is locked and that nobody else uses it - or that usage
112 * is safe. The caller must hold the mapping's tree_lock.
113 */
114void __delete_from_page_cache(struct page *page)
115{
116 struct address_space *mapping = page->mapping;
117
118 /*
119 * if we're uptodate, flush out into the cleancache, otherwise
120 * invalidate any existing cleancache entries. We can't leave
121 * stale data around in the cleancache once our page is gone
122 */
123 if (PageUptodate(page) && PageMappedToDisk(page))
124 cleancache_put_page(page);
125 else
126 cleancache_flush_page(mapping, page);
127
128 radix_tree_delete(&mapping->page_tree, page->index);
129 page->mapping = NULL;
130 /* Leave page->index set: truncation lookup relies upon it */
131 mapping->nrpages--;
132 __dec_zone_page_state(page, NR_FILE_PAGES);
133 if (PageSwapBacked(page))
134 __dec_zone_page_state(page, NR_SHMEM);
135 BUG_ON(page_mapped(page));
136
137 /*
138 * Some filesystems seem to re-dirty the page even after
139 * the VM has canceled the dirty bit (eg ext3 journaling).
140 *
141 * Fix it up by doing a final dirty accounting check after
142 * having removed the page entirely.
143 */
144 if (PageDirty(page) && mapping_cap_account_dirty(mapping)) {
145 dec_zone_page_state(page, NR_FILE_DIRTY);
146 dec_bdi_stat(mapping->backing_dev_info, BDI_RECLAIMABLE);
147 }
148}
149
150/**
151 * delete_from_page_cache - delete page from page cache
152 * @page: the page which the kernel is trying to remove from page cache
153 *
154 * This must be called only on pages that have been verified to be in the page
155 * cache and locked. It will never put the page into the free list, the caller
156 * has a reference on the page.
157 */
158void delete_from_page_cache(struct page *page)
159{
160 struct address_space *mapping = page->mapping;
161 void (*freepage)(struct page *);
162
163 BUG_ON(!PageLocked(page));
164
165 freepage = mapping->a_ops->freepage;
166 spin_lock_irq(&mapping->tree_lock);
167 __delete_from_page_cache(page);
168 spin_unlock_irq(&mapping->tree_lock);
169 mem_cgroup_uncharge_cache_page(page);
170
171 if (freepage)
172 freepage(page);
173 page_cache_release(page);
174}
175EXPORT_SYMBOL(delete_from_page_cache);
176
177static int sleep_on_page(void *word)
178{
179 io_schedule();
180 return 0;
181}
182
183static int sleep_on_page_killable(void *word)
184{
185 sleep_on_page(word);
186 return fatal_signal_pending(current) ? -EINTR : 0;
187}
188
189/**
190 * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range
191 * @mapping: address space structure to write
192 * @start: offset in bytes where the range starts
193 * @end: offset in bytes where the range ends (inclusive)
194 * @sync_mode: enable synchronous operation
195 *
196 * Start writeback against all of a mapping's dirty pages that lie
197 * within the byte offsets <start, end> inclusive.
198 *
199 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
200 * opposed to a regular memory cleansing writeback. The difference between
201 * these two operations is that if a dirty page/buffer is encountered, it must
202 * be waited upon, and not just skipped over.
203 */
204int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
205 loff_t end, int sync_mode)
206{
207 int ret;
208 struct writeback_control wbc = {
209 .sync_mode = sync_mode,
210 .nr_to_write = LONG_MAX,
211 .range_start = start,
212 .range_end = end,
213 };
214
215 if (!mapping_cap_writeback_dirty(mapping))
216 return 0;
217
218 ret = do_writepages(mapping, &wbc);
219 return ret;
220}
221
222static inline int __filemap_fdatawrite(struct address_space *mapping,
223 int sync_mode)
224{
225 return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode);
226}
227
228int filemap_fdatawrite(struct address_space *mapping)
229{
230 return __filemap_fdatawrite(mapping, WB_SYNC_ALL);
231}
232EXPORT_SYMBOL(filemap_fdatawrite);
233
234int filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
235 loff_t end)
236{
237 return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL);
238}
239EXPORT_SYMBOL(filemap_fdatawrite_range);
240
241/**
242 * filemap_flush - mostly a non-blocking flush
243 * @mapping: target address_space
244 *
245 * This is a mostly non-blocking flush. Not suitable for data-integrity
246 * purposes - I/O may not be started against all dirty pages.
247 */
248int filemap_flush(struct address_space *mapping)
249{
250 return __filemap_fdatawrite(mapping, WB_SYNC_NONE);
251}
252EXPORT_SYMBOL(filemap_flush);
253
254/**
255 * filemap_fdatawait_range - wait for writeback to complete
256 * @mapping: address space structure to wait for
257 * @start_byte: offset in bytes where the range starts
258 * @end_byte: offset in bytes where the range ends (inclusive)
259 *
260 * Walk the list of under-writeback pages of the given address space
261 * in the given range and wait for all of them.
262 */
263int filemap_fdatawait_range(struct address_space *mapping, loff_t start_byte,
264 loff_t end_byte)
265{
266 pgoff_t index = start_byte >> PAGE_CACHE_SHIFT;
267 pgoff_t end = end_byte >> PAGE_CACHE_SHIFT;
268 struct pagevec pvec;
269 int nr_pages;
270 int ret = 0;
271
272 if (end_byte < start_byte)
273 return 0;
274
275 pagevec_init(&pvec, 0);
276 while ((index <= end) &&
277 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
278 PAGECACHE_TAG_WRITEBACK,
279 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1)) != 0) {
280 unsigned i;
281
282 for (i = 0; i < nr_pages; i++) {
283 struct page *page = pvec.pages[i];
284
285 /* until radix tree lookup accepts end_index */
286 if (page->index > end)
287 continue;
288
289 wait_on_page_writeback(page);
290 if (TestClearPageError(page))
291 ret = -EIO;
292 }
293 pagevec_release(&pvec);
294 cond_resched();
295 }
296
297 /* Check for outstanding write errors */
298 if (test_and_clear_bit(AS_ENOSPC, &mapping->flags))
299 ret = -ENOSPC;
300 if (test_and_clear_bit(AS_EIO, &mapping->flags))
301 ret = -EIO;
302
303 return ret;
304}
305EXPORT_SYMBOL(filemap_fdatawait_range);
306
307/**
308 * filemap_fdatawait - wait for all under-writeback pages to complete
309 * @mapping: address space structure to wait for
310 *
311 * Walk the list of under-writeback pages of the given address space
312 * and wait for all of them.
313 */
314int filemap_fdatawait(struct address_space *mapping)
315{
316 loff_t i_size = i_size_read(mapping->host);
317
318 if (i_size == 0)
319 return 0;
320
321 return filemap_fdatawait_range(mapping, 0, i_size - 1);
322}
323EXPORT_SYMBOL(filemap_fdatawait);
324
325int filemap_write_and_wait(struct address_space *mapping)
326{
327 int err = 0;
328
329 if (mapping->nrpages) {
330 err = filemap_fdatawrite(mapping);
331 /*
332 * Even if the above returned error, the pages may be
333 * written partially (e.g. -ENOSPC), so we wait for it.
334 * But the -EIO is special case, it may indicate the worst
335 * thing (e.g. bug) happened, so we avoid waiting for it.
336 */
337 if (err != -EIO) {
338 int err2 = filemap_fdatawait(mapping);
339 if (!err)
340 err = err2;
341 }
342 }
343 return err;
344}
345EXPORT_SYMBOL(filemap_write_and_wait);
346
347/**
348 * filemap_write_and_wait_range - write out & wait on a file range
349 * @mapping: the address_space for the pages
350 * @lstart: offset in bytes where the range starts
351 * @lend: offset in bytes where the range ends (inclusive)
352 *
353 * Write out and wait upon file offsets lstart->lend, inclusive.
354 *
355 * Note that `lend' is inclusive (describes the last byte to be written) so
356 * that this function can be used to write to the very end-of-file (end = -1).
357 */
358int filemap_write_and_wait_range(struct address_space *mapping,
359 loff_t lstart, loff_t lend)
360{
361 int err = 0;
362
363 if (mapping->nrpages) {
364 err = __filemap_fdatawrite_range(mapping, lstart, lend,
365 WB_SYNC_ALL);
366 /* See comment of filemap_write_and_wait() */
367 if (err != -EIO) {
368 int err2 = filemap_fdatawait_range(mapping,
369 lstart, lend);
370 if (!err)
371 err = err2;
372 }
373 }
374 return err;
375}
376EXPORT_SYMBOL(filemap_write_and_wait_range);
377
378/**
379 * replace_page_cache_page - replace a pagecache page with a new one
380 * @old: page to be replaced
381 * @new: page to replace with
382 * @gfp_mask: allocation mode
383 *
384 * This function replaces a page in the pagecache with a new one. On
385 * success it acquires the pagecache reference for the new page and
386 * drops it for the old page. Both the old and new pages must be
387 * locked. This function does not add the new page to the LRU, the
388 * caller must do that.
389 *
390 * The remove + add is atomic. The only way this function can fail is
391 * memory allocation failure.
392 */
393int replace_page_cache_page(struct page *old, struct page *new, gfp_t gfp_mask)
394{
395 int error;
396 struct mem_cgroup *memcg = NULL;
397
398 VM_BUG_ON(!PageLocked(old));
399 VM_BUG_ON(!PageLocked(new));
400 VM_BUG_ON(new->mapping);
401
402 /*
403 * This is not page migration, but prepare_migration and
404 * end_migration does enough work for charge replacement.
405 *
406 * In the longer term we probably want a specialized function
407 * for moving the charge from old to new in a more efficient
408 * manner.
409 */
410 error = mem_cgroup_prepare_migration(old, new, &memcg, gfp_mask);
411 if (error)
412 return error;
413
414 error = radix_tree_preload(gfp_mask & ~__GFP_HIGHMEM);
415 if (!error) {
416 struct address_space *mapping = old->mapping;
417 void (*freepage)(struct page *);
418
419 pgoff_t offset = old->index;
420 freepage = mapping->a_ops->freepage;
421
422 page_cache_get(new);
423 new->mapping = mapping;
424 new->index = offset;
425
426 spin_lock_irq(&mapping->tree_lock);
427 __delete_from_page_cache(old);
428 error = radix_tree_insert(&mapping->page_tree, offset, new);
429 BUG_ON(error);
430 mapping->nrpages++;
431 __inc_zone_page_state(new, NR_FILE_PAGES);
432 if (PageSwapBacked(new))
433 __inc_zone_page_state(new, NR_SHMEM);
434 spin_unlock_irq(&mapping->tree_lock);
435 radix_tree_preload_end();
436 if (freepage)
437 freepage(old);
438 page_cache_release(old);
439 mem_cgroup_end_migration(memcg, old, new, true);
440 } else {
441 mem_cgroup_end_migration(memcg, old, new, false);
442 }
443
444 return error;
445}
446EXPORT_SYMBOL_GPL(replace_page_cache_page);
447
448/**
449 * add_to_page_cache_locked - add a locked page to the pagecache
450 * @page: page to add
451 * @mapping: the page's address_space
452 * @offset: page index
453 * @gfp_mask: page allocation mode
454 *
455 * This function is used to add a page to the pagecache. It must be locked.
456 * This function does not add the page to the LRU. The caller must do that.
457 */
458int add_to_page_cache_locked(struct page *page, struct address_space *mapping,
459 pgoff_t offset, gfp_t gfp_mask)
460{
461 int error;
462
463 VM_BUG_ON(!PageLocked(page));
464 VM_BUG_ON(PageSwapBacked(page));
465
466 error = mem_cgroup_cache_charge(page, current->mm,
467 gfp_mask & GFP_RECLAIM_MASK);
468 if (error)
469 goto out;
470
471 error = radix_tree_preload(gfp_mask & ~__GFP_HIGHMEM);
472 if (error == 0) {
473 page_cache_get(page);
474 page->mapping = mapping;
475 page->index = offset;
476
477 spin_lock_irq(&mapping->tree_lock);
478 error = radix_tree_insert(&mapping->page_tree, offset, page);
479 if (likely(!error)) {
480 mapping->nrpages++;
481 __inc_zone_page_state(page, NR_FILE_PAGES);
482 spin_unlock_irq(&mapping->tree_lock);
483 } else {
484 page->mapping = NULL;
485 /* Leave page->index set: truncation relies upon it */
486 spin_unlock_irq(&mapping->tree_lock);
487 mem_cgroup_uncharge_cache_page(page);
488 page_cache_release(page);
489 }
490 radix_tree_preload_end();
491 } else
492 mem_cgroup_uncharge_cache_page(page);
493out:
494 return error;
495}
496EXPORT_SYMBOL(add_to_page_cache_locked);
497
498int add_to_page_cache_lru(struct page *page, struct address_space *mapping,
499 pgoff_t offset, gfp_t gfp_mask)
500{
501 int ret;
502
503 ret = add_to_page_cache(page, mapping, offset, gfp_mask);
504 if (ret == 0)
505 lru_cache_add_file(page);
506 return ret;
507}
508EXPORT_SYMBOL_GPL(add_to_page_cache_lru);
509
510#ifdef CONFIG_NUMA
511struct page *__page_cache_alloc(gfp_t gfp)
512{
513 int n;
514 struct page *page;
515
516 if (cpuset_do_page_mem_spread()) {
517 get_mems_allowed();
518 n = cpuset_mem_spread_node();
519 page = alloc_pages_exact_node(n, gfp, 0);
520 put_mems_allowed();
521 return page;
522 }
523 return alloc_pages(gfp, 0);
524}
525EXPORT_SYMBOL(__page_cache_alloc);
526#endif
527
528/*
529 * In order to wait for pages to become available there must be
530 * waitqueues associated with pages. By using a hash table of
531 * waitqueues where the bucket discipline is to maintain all
532 * waiters on the same queue and wake all when any of the pages
533 * become available, and for the woken contexts to check to be
534 * sure the appropriate page became available, this saves space
535 * at a cost of "thundering herd" phenomena during rare hash
536 * collisions.
537 */
538static wait_queue_head_t *page_waitqueue(struct page *page)
539{
540 const struct zone *zone = page_zone(page);
541
542 return &zone->wait_table[hash_ptr(page, zone->wait_table_bits)];
543}
544
545static inline void wake_up_page(struct page *page, int bit)
546{
547 __wake_up_bit(page_waitqueue(page), &page->flags, bit);
548}
549
550void wait_on_page_bit(struct page *page, int bit_nr)
551{
552 DEFINE_WAIT_BIT(wait, &page->flags, bit_nr);
553
554 if (test_bit(bit_nr, &page->flags))
555 __wait_on_bit(page_waitqueue(page), &wait, sleep_on_page,
556 TASK_UNINTERRUPTIBLE);
557}
558EXPORT_SYMBOL(wait_on_page_bit);
559
560int wait_on_page_bit_killable(struct page *page, int bit_nr)
561{
562 DEFINE_WAIT_BIT(wait, &page->flags, bit_nr);
563
564 if (!test_bit(bit_nr, &page->flags))
565 return 0;
566
567 return __wait_on_bit(page_waitqueue(page), &wait,
568 sleep_on_page_killable, TASK_KILLABLE);
569}
570
571/**
572 * add_page_wait_queue - Add an arbitrary waiter to a page's wait queue
573 * @page: Page defining the wait queue of interest
574 * @waiter: Waiter to add to the queue
575 *
576 * Add an arbitrary @waiter to the wait queue for the nominated @page.
577 */
578void add_page_wait_queue(struct page *page, wait_queue_t *waiter)
579{
580 wait_queue_head_t *q = page_waitqueue(page);
581 unsigned long flags;
582
583 spin_lock_irqsave(&q->lock, flags);
584 __add_wait_queue(q, waiter);
585 spin_unlock_irqrestore(&q->lock, flags);
586}
587EXPORT_SYMBOL_GPL(add_page_wait_queue);
588
589/**
590 * unlock_page - unlock a locked page
591 * @page: the page
592 *
593 * Unlocks the page and wakes up sleepers in ___wait_on_page_locked().
594 * Also wakes sleepers in wait_on_page_writeback() because the wakeup
595 * mechananism between PageLocked pages and PageWriteback pages is shared.
596 * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep.
597 *
598 * The mb is necessary to enforce ordering between the clear_bit and the read
599 * of the waitqueue (to avoid SMP races with a parallel wait_on_page_locked()).
600 */
601void unlock_page(struct page *page)
602{
603 VM_BUG_ON(!PageLocked(page));
604 clear_bit_unlock(PG_locked, &page->flags);
605 smp_mb__after_clear_bit();
606 wake_up_page(page, PG_locked);
607}
608EXPORT_SYMBOL(unlock_page);
609
610/**
611 * end_page_writeback - end writeback against a page
612 * @page: the page
613 */
614void end_page_writeback(struct page *page)
615{
616 if (TestClearPageReclaim(page))
617 rotate_reclaimable_page(page);
618
619 if (!test_clear_page_writeback(page))
620 BUG();
621
622 smp_mb__after_clear_bit();
623 wake_up_page(page, PG_writeback);
624}
625EXPORT_SYMBOL(end_page_writeback);
626
627/**
628 * __lock_page - get a lock on the page, assuming we need to sleep to get it
629 * @page: the page to lock
630 */
631void __lock_page(struct page *page)
632{
633 DEFINE_WAIT_BIT(wait, &page->flags, PG_locked);
634
635 __wait_on_bit_lock(page_waitqueue(page), &wait, sleep_on_page,
636 TASK_UNINTERRUPTIBLE);
637}
638EXPORT_SYMBOL(__lock_page);
639
640int __lock_page_killable(struct page *page)
641{
642 DEFINE_WAIT_BIT(wait, &page->flags, PG_locked);
643
644 return __wait_on_bit_lock(page_waitqueue(page), &wait,
645 sleep_on_page_killable, TASK_KILLABLE);
646}
647EXPORT_SYMBOL_GPL(__lock_page_killable);
648
649int __lock_page_or_retry(struct page *page, struct mm_struct *mm,
650 unsigned int flags)
651{
652 if (flags & FAULT_FLAG_ALLOW_RETRY) {
653 /*
654 * CAUTION! In this case, mmap_sem is not released
655 * even though return 0.
656 */
657 if (flags & FAULT_FLAG_RETRY_NOWAIT)
658 return 0;
659
660 up_read(&mm->mmap_sem);
661 if (flags & FAULT_FLAG_KILLABLE)
662 wait_on_page_locked_killable(page);
663 else
664 wait_on_page_locked(page);
665 return 0;
666 } else {
667 if (flags & FAULT_FLAG_KILLABLE) {
668 int ret;
669
670 ret = __lock_page_killable(page);
671 if (ret) {
672 up_read(&mm->mmap_sem);
673 return 0;
674 }
675 } else
676 __lock_page(page);
677 return 1;
678 }
679}
680
681/**
682 * find_get_page - find and get a page reference
683 * @mapping: the address_space to search
684 * @offset: the page index
685 *
686 * Is there a pagecache struct page at the given (mapping, offset) tuple?
687 * If yes, increment its refcount and return it; if no, return NULL.
688 */
689struct page *find_get_page(struct address_space *mapping, pgoff_t offset)
690{
691 void **pagep;
692 struct page *page;
693
694 rcu_read_lock();
695repeat:
696 page = NULL;
697 pagep = radix_tree_lookup_slot(&mapping->page_tree, offset);
698 if (pagep) {
699 page = radix_tree_deref_slot(pagep);
700 if (unlikely(!page))
701 goto out;
702 if (radix_tree_exception(page)) {
703 if (radix_tree_deref_retry(page))
704 goto repeat;
705 /*
706 * Otherwise, shmem/tmpfs must be storing a swap entry
707 * here as an exceptional entry: so return it without
708 * attempting to raise page count.
709 */
710 goto out;
711 }
712 if (!page_cache_get_speculative(page))
713 goto repeat;
714
715 /*
716 * Has the page moved?
717 * This is part of the lockless pagecache protocol. See
718 * include/linux/pagemap.h for details.
719 */
720 if (unlikely(page != *pagep)) {
721 page_cache_release(page);
722 goto repeat;
723 }
724 }
725out:
726 rcu_read_unlock();
727
728 return page;
729}
730EXPORT_SYMBOL(find_get_page);
731
732/**
733 * find_lock_page - locate, pin and lock a pagecache page
734 * @mapping: the address_space to search
735 * @offset: the page index
736 *
737 * Locates the desired pagecache page, locks it, increments its reference
738 * count and returns its address.
739 *
740 * Returns zero if the page was not present. find_lock_page() may sleep.
741 */
742struct page *find_lock_page(struct address_space *mapping, pgoff_t offset)
743{
744 struct page *page;
745
746repeat:
747 page = find_get_page(mapping, offset);
748 if (page && !radix_tree_exception(page)) {
749 lock_page(page);
750 /* Has the page been truncated? */
751 if (unlikely(page->mapping != mapping)) {
752 unlock_page(page);
753 page_cache_release(page);
754 goto repeat;
755 }
756 VM_BUG_ON(page->index != offset);
757 }
758 return page;
759}
760EXPORT_SYMBOL(find_lock_page);
761
762/**
763 * find_or_create_page - locate or add a pagecache page
764 * @mapping: the page's address_space
765 * @index: the page's index into the mapping
766 * @gfp_mask: page allocation mode
767 *
768 * Locates a page in the pagecache. If the page is not present, a new page
769 * is allocated using @gfp_mask and is added to the pagecache and to the VM's
770 * LRU list. The returned page is locked and has its reference count
771 * incremented.
772 *
773 * find_or_create_page() may sleep, even if @gfp_flags specifies an atomic
774 * allocation!
775 *
776 * find_or_create_page() returns the desired page's address, or zero on
777 * memory exhaustion.
778 */
779struct page *find_or_create_page(struct address_space *mapping,
780 pgoff_t index, gfp_t gfp_mask)
781{
782 struct page *page;
783 int err;
784repeat:
785 page = find_lock_page(mapping, index);
786 if (!page) {
787 page = __page_cache_alloc(gfp_mask);
788 if (!page)
789 return NULL;
790 /*
791 * We want a regular kernel memory (not highmem or DMA etc)
792 * allocation for the radix tree nodes, but we need to honour
793 * the context-specific requirements the caller has asked for.
794 * GFP_RECLAIM_MASK collects those requirements.
795 */
796 err = add_to_page_cache_lru(page, mapping, index,
797 (gfp_mask & GFP_RECLAIM_MASK));
798 if (unlikely(err)) {
799 page_cache_release(page);
800 page = NULL;
801 if (err == -EEXIST)
802 goto repeat;
803 }
804 }
805 return page;
806}
807EXPORT_SYMBOL(find_or_create_page);
808
809/**
810 * find_get_pages - gang pagecache lookup
811 * @mapping: The address_space to search
812 * @start: The starting page index
813 * @nr_pages: The maximum number of pages
814 * @pages: Where the resulting pages are placed
815 *
816 * find_get_pages() will search for and return a group of up to
817 * @nr_pages pages in the mapping. The pages are placed at @pages.
818 * find_get_pages() takes a reference against the returned pages.
819 *
820 * The search returns a group of mapping-contiguous pages with ascending
821 * indexes. There may be holes in the indices due to not-present pages.
822 *
823 * find_get_pages() returns the number of pages which were found.
824 */
825unsigned find_get_pages(struct address_space *mapping, pgoff_t start,
826 unsigned int nr_pages, struct page **pages)
827{
828 unsigned int i;
829 unsigned int ret;
830 unsigned int nr_found, nr_skip;
831
832 rcu_read_lock();
833restart:
834 nr_found = radix_tree_gang_lookup_slot(&mapping->page_tree,
835 (void ***)pages, NULL, start, nr_pages);
836 ret = 0;
837 nr_skip = 0;
838 for (i = 0; i < nr_found; i++) {
839 struct page *page;
840repeat:
841 page = radix_tree_deref_slot((void **)pages[i]);
842 if (unlikely(!page))
843 continue;
844
845 if (radix_tree_exception(page)) {
846 if (radix_tree_deref_retry(page)) {
847 /*
848 * Transient condition which can only trigger
849 * when entry at index 0 moves out of or back
850 * to root: none yet gotten, safe to restart.
851 */
852 WARN_ON(start | i);
853 goto restart;
854 }
855 /*
856 * Otherwise, shmem/tmpfs must be storing a swap entry
857 * here as an exceptional entry: so skip over it -
858 * we only reach this from invalidate_mapping_pages().
859 */
860 nr_skip++;
861 continue;
862 }
863
864 if (!page_cache_get_speculative(page))
865 goto repeat;
866
867 /* Has the page moved? */
868 if (unlikely(page != *((void **)pages[i]))) {
869 page_cache_release(page);
870 goto repeat;
871 }
872
873 pages[ret] = page;
874 ret++;
875 }
876
877 /*
878 * If all entries were removed before we could secure them,
879 * try again, because callers stop trying once 0 is returned.
880 */
881 if (unlikely(!ret && nr_found > nr_skip))
882 goto restart;
883 rcu_read_unlock();
884 return ret;
885}
886
887/**
888 * find_get_pages_contig - gang contiguous pagecache lookup
889 * @mapping: The address_space to search
890 * @index: The starting page index
891 * @nr_pages: The maximum number of pages
892 * @pages: Where the resulting pages are placed
893 *
894 * find_get_pages_contig() works exactly like find_get_pages(), except
895 * that the returned number of pages are guaranteed to be contiguous.
896 *
897 * find_get_pages_contig() returns the number of pages which were found.
898 */
899unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index,
900 unsigned int nr_pages, struct page **pages)
901{
902 unsigned int i;
903 unsigned int ret;
904 unsigned int nr_found;
905
906 rcu_read_lock();
907restart:
908 nr_found = radix_tree_gang_lookup_slot(&mapping->page_tree,
909 (void ***)pages, NULL, index, nr_pages);
910 ret = 0;
911 for (i = 0; i < nr_found; i++) {
912 struct page *page;
913repeat:
914 page = radix_tree_deref_slot((void **)pages[i]);
915 if (unlikely(!page))
916 continue;
917
918 if (radix_tree_exception(page)) {
919 if (radix_tree_deref_retry(page)) {
920 /*
921 * Transient condition which can only trigger
922 * when entry at index 0 moves out of or back
923 * to root: none yet gotten, safe to restart.
924 */
925 goto restart;
926 }
927 /*
928 * Otherwise, shmem/tmpfs must be storing a swap entry
929 * here as an exceptional entry: so stop looking for
930 * contiguous pages.
931 */
932 break;
933 }
934
935 if (!page_cache_get_speculative(page))
936 goto repeat;
937
938 /* Has the page moved? */
939 if (unlikely(page != *((void **)pages[i]))) {
940 page_cache_release(page);
941 goto repeat;
942 }
943
944 /*
945 * must check mapping and index after taking the ref.
946 * otherwise we can get both false positives and false
947 * negatives, which is just confusing to the caller.
948 */
949 if (page->mapping == NULL || page->index != index) {
950 page_cache_release(page);
951 break;
952 }
953
954 pages[ret] = page;
955 ret++;
956 index++;
957 }
958 rcu_read_unlock();
959 return ret;
960}
961EXPORT_SYMBOL(find_get_pages_contig);
962
963/**
964 * find_get_pages_tag - find and return pages that match @tag
965 * @mapping: the address_space to search
966 * @index: the starting page index
967 * @tag: the tag index
968 * @nr_pages: the maximum number of pages
969 * @pages: where the resulting pages are placed
970 *
971 * Like find_get_pages, except we only return pages which are tagged with
972 * @tag. We update @index to index the next page for the traversal.
973 */
974unsigned find_get_pages_tag(struct address_space *mapping, pgoff_t *index,
975 int tag, unsigned int nr_pages, struct page **pages)
976{
977 unsigned int i;
978 unsigned int ret;
979 unsigned int nr_found;
980
981 rcu_read_lock();
982restart:
983 nr_found = radix_tree_gang_lookup_tag_slot(&mapping->page_tree,
984 (void ***)pages, *index, nr_pages, tag);
985 ret = 0;
986 for (i = 0; i < nr_found; i++) {
987 struct page *page;
988repeat:
989 page = radix_tree_deref_slot((void **)pages[i]);
990 if (unlikely(!page))
991 continue;
992
993 if (radix_tree_exception(page)) {
994 if (radix_tree_deref_retry(page)) {
995 /*
996 * Transient condition which can only trigger
997 * when entry at index 0 moves out of or back
998 * to root: none yet gotten, safe to restart.
999 */
1000 goto restart;
1001 }
1002 /*
1003 * This function is never used on a shmem/tmpfs
1004 * mapping, so a swap entry won't be found here.
1005 */
1006 BUG();
1007 }
1008
1009 if (!page_cache_get_speculative(page))
1010 goto repeat;
1011
1012 /* Has the page moved? */
1013 if (unlikely(page != *((void **)pages[i]))) {
1014 page_cache_release(page);
1015 goto repeat;
1016 }
1017
1018 pages[ret] = page;
1019 ret++;
1020 }
1021
1022 /*
1023 * If all entries were removed before we could secure them,
1024 * try again, because callers stop trying once 0 is returned.
1025 */
1026 if (unlikely(!ret && nr_found))
1027 goto restart;
1028 rcu_read_unlock();
1029
1030 if (ret)
1031 *index = pages[ret - 1]->index + 1;
1032
1033 return ret;
1034}
1035EXPORT_SYMBOL(find_get_pages_tag);
1036
1037/**
1038 * grab_cache_page_nowait - returns locked page at given index in given cache
1039 * @mapping: target address_space
1040 * @index: the page index
1041 *
1042 * Same as grab_cache_page(), but do not wait if the page is unavailable.
1043 * This is intended for speculative data generators, where the data can
1044 * be regenerated if the page couldn't be grabbed. This routine should
1045 * be safe to call while holding the lock for another page.
1046 *
1047 * Clear __GFP_FS when allocating the page to avoid recursion into the fs
1048 * and deadlock against the caller's locked page.
1049 */
1050struct page *
1051grab_cache_page_nowait(struct address_space *mapping, pgoff_t index)
1052{
1053 struct page *page = find_get_page(mapping, index);
1054
1055 if (page) {
1056 if (trylock_page(page))
1057 return page;
1058 page_cache_release(page);
1059 return NULL;
1060 }
1061 page = __page_cache_alloc(mapping_gfp_mask(mapping) & ~__GFP_FS);
1062 if (page && add_to_page_cache_lru(page, mapping, index, GFP_NOFS)) {
1063 page_cache_release(page);
1064 page = NULL;
1065 }
1066 return page;
1067}
1068EXPORT_SYMBOL(grab_cache_page_nowait);
1069
1070/*
1071 * CD/DVDs are error prone. When a medium error occurs, the driver may fail
1072 * a _large_ part of the i/o request. Imagine the worst scenario:
1073 *
1074 * ---R__________________________________________B__________
1075 * ^ reading here ^ bad block(assume 4k)
1076 *
1077 * read(R) => miss => readahead(R...B) => media error => frustrating retries
1078 * => failing the whole request => read(R) => read(R+1) =>
1079 * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
1080 * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
1081 * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
1082 *
1083 * It is going insane. Fix it by quickly scaling down the readahead size.
1084 */
1085static void shrink_readahead_size_eio(struct file *filp,
1086 struct file_ra_state *ra)
1087{
1088 ra->ra_pages /= 4;
1089}
1090
1091/**
1092 * do_generic_file_read - generic file read routine
1093 * @filp: the file to read
1094 * @ppos: current file position
1095 * @desc: read_descriptor
1096 * @actor: read method
1097 *
1098 * This is a generic file read routine, and uses the
1099 * mapping->a_ops->readpage() function for the actual low-level stuff.
1100 *
1101 * This is really ugly. But the goto's actually try to clarify some
1102 * of the logic when it comes to error handling etc.
1103 */
1104static void do_generic_file_read(struct file *filp, loff_t *ppos,
1105 read_descriptor_t *desc, read_actor_t actor)
1106{
1107 struct address_space *mapping = filp->f_mapping;
1108 struct inode *inode = mapping->host;
1109 struct file_ra_state *ra = &filp->f_ra;
1110 pgoff_t index;
1111 pgoff_t last_index;
1112 pgoff_t prev_index;
1113 unsigned long offset; /* offset into pagecache page */
1114 unsigned int prev_offset;
1115 int error;
1116
1117 index = *ppos >> PAGE_CACHE_SHIFT;
1118 prev_index = ra->prev_pos >> PAGE_CACHE_SHIFT;
1119 prev_offset = ra->prev_pos & (PAGE_CACHE_SIZE-1);
1120 last_index = (*ppos + desc->count + PAGE_CACHE_SIZE-1) >> PAGE_CACHE_SHIFT;
1121 offset = *ppos & ~PAGE_CACHE_MASK;
1122
1123 for (;;) {
1124 struct page *page;
1125 pgoff_t end_index;
1126 loff_t isize;
1127 unsigned long nr, ret;
1128
1129 cond_resched();
1130find_page:
1131 page = find_get_page(mapping, index);
1132 if (!page) {
1133 page_cache_sync_readahead(mapping,
1134 ra, filp,
1135 index, last_index - index);
1136 page = find_get_page(mapping, index);
1137 if (unlikely(page == NULL))
1138 goto no_cached_page;
1139 }
1140 if (PageReadahead(page)) {
1141 page_cache_async_readahead(mapping,
1142 ra, filp, page,
1143 index, last_index - index);
1144 }
1145 if (!PageUptodate(page)) {
1146 if (inode->i_blkbits == PAGE_CACHE_SHIFT ||
1147 !mapping->a_ops->is_partially_uptodate)
1148 goto page_not_up_to_date;
1149 if (!trylock_page(page))
1150 goto page_not_up_to_date;
1151 /* Did it get truncated before we got the lock? */
1152 if (!page->mapping)
1153 goto page_not_up_to_date_locked;
1154 if (!mapping->a_ops->is_partially_uptodate(page,
1155 desc, offset))
1156 goto page_not_up_to_date_locked;
1157 unlock_page(page);
1158 }
1159page_ok:
1160 /*
1161 * i_size must be checked after we know the page is Uptodate.
1162 *
1163 * Checking i_size after the check allows us to calculate
1164 * the correct value for "nr", which means the zero-filled
1165 * part of the page is not copied back to userspace (unless
1166 * another truncate extends the file - this is desired though).
1167 */
1168
1169 isize = i_size_read(inode);
1170 end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
1171 if (unlikely(!isize || index > end_index)) {
1172 page_cache_release(page);
1173 goto out;
1174 }
1175
1176 /* nr is the maximum number of bytes to copy from this page */
1177 nr = PAGE_CACHE_SIZE;
1178 if (index == end_index) {
1179 nr = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;
1180 if (nr <= offset) {
1181 page_cache_release(page);
1182 goto out;
1183 }
1184 }
1185 nr = nr - offset;
1186
1187 /* If users can be writing to this page using arbitrary
1188 * virtual addresses, take care about potential aliasing
1189 * before reading the page on the kernel side.
1190 */
1191 if (mapping_writably_mapped(mapping))
1192 flush_dcache_page(page);
1193
1194 /*
1195 * When a sequential read accesses a page several times,
1196 * only mark it as accessed the first time.
1197 */
1198 if (prev_index != index || offset != prev_offset)
1199 mark_page_accessed(page);
1200 prev_index = index;
1201
1202 /*
1203 * Ok, we have the page, and it's up-to-date, so
1204 * now we can copy it to user space...
1205 *
1206 * The actor routine returns how many bytes were actually used..
1207 * NOTE! This may not be the same as how much of a user buffer
1208 * we filled up (we may be padding etc), so we can only update
1209 * "pos" here (the actor routine has to update the user buffer
1210 * pointers and the remaining count).
1211 */
1212 ret = actor(desc, page, offset, nr);
1213 offset += ret;
1214 index += offset >> PAGE_CACHE_SHIFT;
1215 offset &= ~PAGE_CACHE_MASK;
1216 prev_offset = offset;
1217
1218 page_cache_release(page);
1219 if (ret == nr && desc->count)
1220 continue;
1221 goto out;
1222
1223page_not_up_to_date:
1224 /* Get exclusive access to the page ... */
1225 error = lock_page_killable(page);
1226 if (unlikely(error))
1227 goto readpage_error;
1228
1229page_not_up_to_date_locked:
1230 /* Did it get truncated before we got the lock? */
1231 if (!page->mapping) {
1232 unlock_page(page);
1233 page_cache_release(page);
1234 continue;
1235 }
1236
1237 /* Did somebody else fill it already? */
1238 if (PageUptodate(page)) {
1239 unlock_page(page);
1240 goto page_ok;
1241 }
1242
1243readpage:
1244 /*
1245 * A previous I/O error may have been due to temporary
1246 * failures, eg. multipath errors.
1247 * PG_error will be set again if readpage fails.
1248 */
1249 ClearPageError(page);
1250 /* Start the actual read. The read will unlock the page. */
1251 error = mapping->a_ops->readpage(filp, page);
1252
1253 if (unlikely(error)) {
1254 if (error == AOP_TRUNCATED_PAGE) {
1255 page_cache_release(page);
1256 goto find_page;
1257 }
1258 goto readpage_error;
1259 }
1260
1261 if (!PageUptodate(page)) {
1262 error = lock_page_killable(page);
1263 if (unlikely(error))
1264 goto readpage_error;
1265 if (!PageUptodate(page)) {
1266 if (page->mapping == NULL) {
1267 /*
1268 * invalidate_mapping_pages got it
1269 */
1270 unlock_page(page);
1271 page_cache_release(page);
1272 goto find_page;
1273 }
1274 unlock_page(page);
1275 shrink_readahead_size_eio(filp, ra);
1276 error = -EIO;
1277 goto readpage_error;
1278 }
1279 unlock_page(page);
1280 }
1281
1282 goto page_ok;
1283
1284readpage_error:
1285 /* UHHUH! A synchronous read error occurred. Report it */
1286 desc->error = error;
1287 page_cache_release(page);
1288 goto out;
1289
1290no_cached_page:
1291 /*
1292 * Ok, it wasn't cached, so we need to create a new
1293 * page..
1294 */
1295 page = page_cache_alloc_cold(mapping);
1296 if (!page) {
1297 desc->error = -ENOMEM;
1298 goto out;
1299 }
1300 error = add_to_page_cache_lru(page, mapping,
1301 index, GFP_KERNEL);
1302 if (error) {
1303 page_cache_release(page);
1304 if (error == -EEXIST)
1305 goto find_page;
1306 desc->error = error;
1307 goto out;
1308 }
1309 goto readpage;
1310 }
1311
1312out:
1313 ra->prev_pos = prev_index;
1314 ra->prev_pos <<= PAGE_CACHE_SHIFT;
1315 ra->prev_pos |= prev_offset;
1316
1317 *ppos = ((loff_t)index << PAGE_CACHE_SHIFT) + offset;
1318 file_accessed(filp);
1319}
1320
1321int file_read_actor(read_descriptor_t *desc, struct page *page,
1322 unsigned long offset, unsigned long size)
1323{
1324 char *kaddr;
1325 unsigned long left, count = desc->count;
1326
1327 if (size > count)
1328 size = count;
1329
1330 /*
1331 * Faults on the destination of a read are common, so do it before
1332 * taking the kmap.
1333 */
1334 if (!fault_in_pages_writeable(desc->arg.buf, size)) {
1335 kaddr = kmap_atomic(page, KM_USER0);
1336 left = __copy_to_user_inatomic(desc->arg.buf,
1337 kaddr + offset, size);
1338 kunmap_atomic(kaddr, KM_USER0);
1339 if (left == 0)
1340 goto success;
1341 }
1342
1343 /* Do it the slow way */
1344 kaddr = kmap(page);
1345 left = __copy_to_user(desc->arg.buf, kaddr + offset, size);
1346 kunmap(page);
1347
1348 if (left) {
1349 size -= left;
1350 desc->error = -EFAULT;
1351 }
1352success:
1353 desc->count = count - size;
1354 desc->written += size;
1355 desc->arg.buf += size;
1356 return size;
1357}
1358
1359/*
1360 * Performs necessary checks before doing a write
1361 * @iov: io vector request
1362 * @nr_segs: number of segments in the iovec
1363 * @count: number of bytes to write
1364 * @access_flags: type of access: %VERIFY_READ or %VERIFY_WRITE
1365 *
1366 * Adjust number of segments and amount of bytes to write (nr_segs should be
1367 * properly initialized first). Returns appropriate error code that caller
1368 * should return or zero in case that write should be allowed.
1369 */
1370int generic_segment_checks(const struct iovec *iov,
1371 unsigned long *nr_segs, size_t *count, int access_flags)
1372{
1373 unsigned long seg;
1374 size_t cnt = 0;
1375 for (seg = 0; seg < *nr_segs; seg++) {
1376 const struct iovec *iv = &iov[seg];
1377
1378 /*
1379 * If any segment has a negative length, or the cumulative
1380 * length ever wraps negative then return -EINVAL.
1381 */
1382 cnt += iv->iov_len;
1383 if (unlikely((ssize_t)(cnt|iv->iov_len) < 0))
1384 return -EINVAL;
1385 if (access_ok(access_flags, iv->iov_base, iv->iov_len))
1386 continue;
1387 if (seg == 0)
1388 return -EFAULT;
1389 *nr_segs = seg;
1390 cnt -= iv->iov_len; /* This segment is no good */
1391 break;
1392 }
1393 *count = cnt;
1394 return 0;
1395}
1396EXPORT_SYMBOL(generic_segment_checks);
1397
1398/**
1399 * generic_file_aio_read - generic filesystem read routine
1400 * @iocb: kernel I/O control block
1401 * @iov: io vector request
1402 * @nr_segs: number of segments in the iovec
1403 * @pos: current file position
1404 *
1405 * This is the "read()" routine for all filesystems
1406 * that can use the page cache directly.
1407 */
1408ssize_t
1409generic_file_aio_read(struct kiocb *iocb, const struct iovec *iov,
1410 unsigned long nr_segs, loff_t pos)
1411{
1412 struct file *filp = iocb->ki_filp;
1413 ssize_t retval;
1414 unsigned long seg = 0;
1415 size_t count;
1416 loff_t *ppos = &iocb->ki_pos;
1417 struct blk_plug plug;
1418
1419 count = 0;
1420 retval = generic_segment_checks(iov, &nr_segs, &count, VERIFY_WRITE);
1421 if (retval)
1422 return retval;
1423
1424 blk_start_plug(&plug);
1425
1426 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */
1427 if (filp->f_flags & O_DIRECT) {
1428 loff_t size;
1429 struct address_space *mapping;
1430 struct inode *inode;
1431
1432 mapping = filp->f_mapping;
1433 inode = mapping->host;
1434 if (!count)
1435 goto out; /* skip atime */
1436 size = i_size_read(inode);
1437 if (pos < size) {
1438 retval = filemap_write_and_wait_range(mapping, pos,
1439 pos + iov_length(iov, nr_segs) - 1);
1440 if (!retval) {
1441 retval = mapping->a_ops->direct_IO(READ, iocb,
1442 iov, pos, nr_segs);
1443 }
1444 if (retval > 0) {
1445 *ppos = pos + retval;
1446 count -= retval;
1447 }
1448
1449 /*
1450 * Btrfs can have a short DIO read if we encounter
1451 * compressed extents, so if there was an error, or if
1452 * we've already read everything we wanted to, or if
1453 * there was a short read because we hit EOF, go ahead
1454 * and return. Otherwise fallthrough to buffered io for
1455 * the rest of the read.
1456 */
1457 if (retval < 0 || !count || *ppos >= size) {
1458 file_accessed(filp);
1459 goto out;
1460 }
1461 }
1462 }
1463
1464 count = retval;
1465 for (seg = 0; seg < nr_segs; seg++) {
1466 read_descriptor_t desc;
1467 loff_t offset = 0;
1468
1469 /*
1470 * If we did a short DIO read we need to skip the section of the
1471 * iov that we've already read data into.
1472 */
1473 if (count) {
1474 if (count > iov[seg].iov_len) {
1475 count -= iov[seg].iov_len;
1476 continue;
1477 }
1478 offset = count;
1479 count = 0;
1480 }
1481
1482 desc.written = 0;
1483 desc.arg.buf = iov[seg].iov_base + offset;
1484 desc.count = iov[seg].iov_len - offset;
1485 if (desc.count == 0)
1486 continue;
1487 desc.error = 0;
1488 do_generic_file_read(filp, ppos, &desc, file_read_actor);
1489 retval += desc.written;
1490 if (desc.error) {
1491 retval = retval ?: desc.error;
1492 break;
1493 }
1494 if (desc.count > 0)
1495 break;
1496 }
1497out:
1498 blk_finish_plug(&plug);
1499 return retval;
1500}
1501EXPORT_SYMBOL(generic_file_aio_read);
1502
1503static ssize_t
1504do_readahead(struct address_space *mapping, struct file *filp,
1505 pgoff_t index, unsigned long nr)
1506{
1507 if (!mapping || !mapping->a_ops || !mapping->a_ops->readpage)
1508 return -EINVAL;
1509
1510 force_page_cache_readahead(mapping, filp, index, nr);
1511 return 0;
1512}
1513
1514SYSCALL_DEFINE(readahead)(int fd, loff_t offset, size_t count)
1515{
1516 ssize_t ret;
1517 struct file *file;
1518
1519 ret = -EBADF;
1520 file = fget(fd);
1521 if (file) {
1522 if (file->f_mode & FMODE_READ) {
1523 struct address_space *mapping = file->f_mapping;
1524 pgoff_t start = offset >> PAGE_CACHE_SHIFT;
1525 pgoff_t end = (offset + count - 1) >> PAGE_CACHE_SHIFT;
1526 unsigned long len = end - start + 1;
1527 ret = do_readahead(mapping, file, start, len);
1528 }
1529 fput(file);
1530 }
1531 return ret;
1532}
1533#ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
1534asmlinkage long SyS_readahead(long fd, loff_t offset, long count)
1535{
1536 return SYSC_readahead((int) fd, offset, (size_t) count);
1537}
1538SYSCALL_ALIAS(sys_readahead, SyS_readahead);
1539#endif
1540
1541#ifdef CONFIG_MMU
1542/**
1543 * page_cache_read - adds requested page to the page cache if not already there
1544 * @file: file to read
1545 * @offset: page index
1546 *
1547 * This adds the requested page to the page cache if it isn't already there,
1548 * and schedules an I/O to read in its contents from disk.
1549 */
1550static int page_cache_read(struct file *file, pgoff_t offset)
1551{
1552 struct address_space *mapping = file->f_mapping;
1553 struct page *page;
1554 int ret;
1555
1556 do {
1557 page = page_cache_alloc_cold(mapping);
1558 if (!page)
1559 return -ENOMEM;
1560
1561 ret = add_to_page_cache_lru(page, mapping, offset, GFP_KERNEL);
1562 if (ret == 0)
1563 ret = mapping->a_ops->readpage(file, page);
1564 else if (ret == -EEXIST)
1565 ret = 0; /* losing race to add is OK */
1566
1567 page_cache_release(page);
1568
1569 } while (ret == AOP_TRUNCATED_PAGE);
1570
1571 return ret;
1572}
1573
1574#define MMAP_LOTSAMISS (100)
1575
1576/*
1577 * Synchronous readahead happens when we don't even find
1578 * a page in the page cache at all.
1579 */
1580static void do_sync_mmap_readahead(struct vm_area_struct *vma,
1581 struct file_ra_state *ra,
1582 struct file *file,
1583 pgoff_t offset)
1584{
1585 unsigned long ra_pages;
1586 struct address_space *mapping = file->f_mapping;
1587
1588 /* If we don't want any read-ahead, don't bother */
1589 if (VM_RandomReadHint(vma))
1590 return;
1591 if (!ra->ra_pages)
1592 return;
1593
1594 if (VM_SequentialReadHint(vma)) {
1595 page_cache_sync_readahead(mapping, ra, file, offset,
1596 ra->ra_pages);
1597 return;
1598 }
1599
1600 /* Avoid banging the cache line if not needed */
1601 if (ra->mmap_miss < MMAP_LOTSAMISS * 10)
1602 ra->mmap_miss++;
1603
1604 /*
1605 * Do we miss much more than hit in this file? If so,
1606 * stop bothering with read-ahead. It will only hurt.
1607 */
1608 if (ra->mmap_miss > MMAP_LOTSAMISS)
1609 return;
1610
1611 /*
1612 * mmap read-around
1613 */
1614 ra_pages = max_sane_readahead(ra->ra_pages);
1615 ra->start = max_t(long, 0, offset - ra_pages / 2);
1616 ra->size = ra_pages;
1617 ra->async_size = ra_pages / 4;
1618 ra_submit(ra, mapping, file);
1619}
1620
1621/*
1622 * Asynchronous readahead happens when we find the page and PG_readahead,
1623 * so we want to possibly extend the readahead further..
1624 */
1625static void do_async_mmap_readahead(struct vm_area_struct *vma,
1626 struct file_ra_state *ra,
1627 struct file *file,
1628 struct page *page,
1629 pgoff_t offset)
1630{
1631 struct address_space *mapping = file->f_mapping;
1632
1633 /* If we don't want any read-ahead, don't bother */
1634 if (VM_RandomReadHint(vma))
1635 return;
1636 if (ra->mmap_miss > 0)
1637 ra->mmap_miss--;
1638 if (PageReadahead(page))
1639 page_cache_async_readahead(mapping, ra, file,
1640 page, offset, ra->ra_pages);
1641}
1642
1643/**
1644 * filemap_fault - read in file data for page fault handling
1645 * @vma: vma in which the fault was taken
1646 * @vmf: struct vm_fault containing details of the fault
1647 *
1648 * filemap_fault() is invoked via the vma operations vector for a
1649 * mapped memory region to read in file data during a page fault.
1650 *
1651 * The goto's are kind of ugly, but this streamlines the normal case of having
1652 * it in the page cache, and handles the special cases reasonably without
1653 * having a lot of duplicated code.
1654 */
1655int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1656{
1657 int error;
1658 struct file *file = vma->vm_file;
1659 struct address_space *mapping = file->f_mapping;
1660 struct file_ra_state *ra = &file->f_ra;
1661 struct inode *inode = mapping->host;
1662 pgoff_t offset = vmf->pgoff;
1663 struct page *page;
1664 pgoff_t size;
1665 int ret = 0;
1666
1667 size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1668 if (offset >= size)
1669 return VM_FAULT_SIGBUS;
1670
1671 /*
1672 * Do we have something in the page cache already?
1673 */
1674 page = find_get_page(mapping, offset);
1675 if (likely(page)) {
1676 /*
1677 * We found the page, so try async readahead before
1678 * waiting for the lock.
1679 */
1680 do_async_mmap_readahead(vma, ra, file, page, offset);
1681 } else {
1682 /* No page in the page cache at all */
1683 do_sync_mmap_readahead(vma, ra, file, offset);
1684 count_vm_event(PGMAJFAULT);
1685 mem_cgroup_count_vm_event(vma->vm_mm, PGMAJFAULT);
1686 ret = VM_FAULT_MAJOR;
1687retry_find:
1688 page = find_get_page(mapping, offset);
1689 if (!page)
1690 goto no_cached_page;
1691 }
1692
1693 if (!lock_page_or_retry(page, vma->vm_mm, vmf->flags)) {
1694 page_cache_release(page);
1695 return ret | VM_FAULT_RETRY;
1696 }
1697
1698 /* Did it get truncated? */
1699 if (unlikely(page->mapping != mapping)) {
1700 unlock_page(page);
1701 put_page(page);
1702 goto retry_find;
1703 }
1704 VM_BUG_ON(page->index != offset);
1705
1706 /*
1707 * We have a locked page in the page cache, now we need to check
1708 * that it's up-to-date. If not, it is going to be due to an error.
1709 */
1710 if (unlikely(!PageUptodate(page)))
1711 goto page_not_uptodate;
1712
1713 /*
1714 * Found the page and have a reference on it.
1715 * We must recheck i_size under page lock.
1716 */
1717 size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1718 if (unlikely(offset >= size)) {
1719 unlock_page(page);
1720 page_cache_release(page);
1721 return VM_FAULT_SIGBUS;
1722 }
1723
1724 vmf->page = page;
1725 return ret | VM_FAULT_LOCKED;
1726
1727no_cached_page:
1728 /*
1729 * We're only likely to ever get here if MADV_RANDOM is in
1730 * effect.
1731 */
1732 error = page_cache_read(file, offset);
1733
1734 /*
1735 * The page we want has now been added to the page cache.
1736 * In the unlikely event that someone removed it in the
1737 * meantime, we'll just come back here and read it again.
1738 */
1739 if (error >= 0)
1740 goto retry_find;
1741
1742 /*
1743 * An error return from page_cache_read can result if the
1744 * system is low on memory, or a problem occurs while trying
1745 * to schedule I/O.
1746 */
1747 if (error == -ENOMEM)
1748 return VM_FAULT_OOM;
1749 return VM_FAULT_SIGBUS;
1750
1751page_not_uptodate:
1752 /*
1753 * Umm, take care of errors if the page isn't up-to-date.
1754 * Try to re-read it _once_. We do this synchronously,
1755 * because there really aren't any performance issues here
1756 * and we need to check for errors.
1757 */
1758 ClearPageError(page);
1759 error = mapping->a_ops->readpage(file, page);
1760 if (!error) {
1761 wait_on_page_locked(page);
1762 if (!PageUptodate(page))
1763 error = -EIO;
1764 }
1765 page_cache_release(page);
1766
1767 if (!error || error == AOP_TRUNCATED_PAGE)
1768 goto retry_find;
1769
1770 /* Things didn't work out. Return zero to tell the mm layer so. */
1771 shrink_readahead_size_eio(file, ra);
1772 return VM_FAULT_SIGBUS;
1773}
1774EXPORT_SYMBOL(filemap_fault);
1775
1776const struct vm_operations_struct generic_file_vm_ops = {
1777 .fault = filemap_fault,
1778};
1779
1780/* This is used for a general mmap of a disk file */
1781
1782int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
1783{
1784 struct address_space *mapping = file->f_mapping;
1785
1786 if (!mapping->a_ops->readpage)
1787 return -ENOEXEC;
1788 file_accessed(file);
1789 vma->vm_ops = &generic_file_vm_ops;
1790 vma->vm_flags |= VM_CAN_NONLINEAR;
1791 return 0;
1792}
1793
1794/*
1795 * This is for filesystems which do not implement ->writepage.
1796 */
1797int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
1798{
1799 if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE))
1800 return -EINVAL;
1801 return generic_file_mmap(file, vma);
1802}
1803#else
1804int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
1805{
1806 return -ENOSYS;
1807}
1808int generic_file_readonly_mmap(struct file * file, struct vm_area_struct * vma)
1809{
1810 return -ENOSYS;
1811}
1812#endif /* CONFIG_MMU */
1813
1814EXPORT_SYMBOL(generic_file_mmap);
1815EXPORT_SYMBOL(generic_file_readonly_mmap);
1816
1817static struct page *__read_cache_page(struct address_space *mapping,
1818 pgoff_t index,
1819 int (*filler)(void *, struct page *),
1820 void *data,
1821 gfp_t gfp)
1822{
1823 struct page *page;
1824 int err;
1825repeat:
1826 page = find_get_page(mapping, index);
1827 if (!page) {
1828 page = __page_cache_alloc(gfp | __GFP_COLD);
1829 if (!page)
1830 return ERR_PTR(-ENOMEM);
1831 err = add_to_page_cache_lru(page, mapping, index, GFP_KERNEL);
1832 if (unlikely(err)) {
1833 page_cache_release(page);
1834 if (err == -EEXIST)
1835 goto repeat;
1836 /* Presumably ENOMEM for radix tree node */
1837 return ERR_PTR(err);
1838 }
1839 err = filler(data, page);
1840 if (err < 0) {
1841 page_cache_release(page);
1842 page = ERR_PTR(err);
1843 }
1844 }
1845 return page;
1846}
1847
1848static struct page *do_read_cache_page(struct address_space *mapping,
1849 pgoff_t index,
1850 int (*filler)(void *, struct page *),
1851 void *data,
1852 gfp_t gfp)
1853
1854{
1855 struct page *page;
1856 int err;
1857
1858retry:
1859 page = __read_cache_page(mapping, index, filler, data, gfp);
1860 if (IS_ERR(page))
1861 return page;
1862 if (PageUptodate(page))
1863 goto out;
1864
1865 lock_page(page);
1866 if (!page->mapping) {
1867 unlock_page(page);
1868 page_cache_release(page);
1869 goto retry;
1870 }
1871 if (PageUptodate(page)) {
1872 unlock_page(page);
1873 goto out;
1874 }
1875 err = filler(data, page);
1876 if (err < 0) {
1877 page_cache_release(page);
1878 return ERR_PTR(err);
1879 }
1880out:
1881 mark_page_accessed(page);
1882 return page;
1883}
1884
1885/**
1886 * read_cache_page_async - read into page cache, fill it if needed
1887 * @mapping: the page's address_space
1888 * @index: the page index
1889 * @filler: function to perform the read
1890 * @data: first arg to filler(data, page) function, often left as NULL
1891 *
1892 * Same as read_cache_page, but don't wait for page to become unlocked
1893 * after submitting it to the filler.
1894 *
1895 * Read into the page cache. If a page already exists, and PageUptodate() is
1896 * not set, try to fill the page but don't wait for it to become unlocked.
1897 *
1898 * If the page does not get brought uptodate, return -EIO.
1899 */
1900struct page *read_cache_page_async(struct address_space *mapping,
1901 pgoff_t index,
1902 int (*filler)(void *, struct page *),
1903 void *data)
1904{
1905 return do_read_cache_page(mapping, index, filler, data, mapping_gfp_mask(mapping));
1906}
1907EXPORT_SYMBOL(read_cache_page_async);
1908
1909static struct page *wait_on_page_read(struct page *page)
1910{
1911 if (!IS_ERR(page)) {
1912 wait_on_page_locked(page);
1913 if (!PageUptodate(page)) {
1914 page_cache_release(page);
1915 page = ERR_PTR(-EIO);
1916 }
1917 }
1918 return page;
1919}
1920
1921/**
1922 * read_cache_page_gfp - read into page cache, using specified page allocation flags.
1923 * @mapping: the page's address_space
1924 * @index: the page index
1925 * @gfp: the page allocator flags to use if allocating
1926 *
1927 * This is the same as "read_mapping_page(mapping, index, NULL)", but with
1928 * any new page allocations done using the specified allocation flags. Note
1929 * that the Radix tree operations will still use GFP_KERNEL, so you can't
1930 * expect to do this atomically or anything like that - but you can pass in
1931 * other page requirements.
1932 *
1933 * If the page does not get brought uptodate, return -EIO.
1934 */
1935struct page *read_cache_page_gfp(struct address_space *mapping,
1936 pgoff_t index,
1937 gfp_t gfp)
1938{
1939 filler_t *filler = (filler_t *)mapping->a_ops->readpage;
1940
1941 return wait_on_page_read(do_read_cache_page(mapping, index, filler, NULL, gfp));
1942}
1943EXPORT_SYMBOL(read_cache_page_gfp);
1944
1945/**
1946 * read_cache_page - read into page cache, fill it if needed
1947 * @mapping: the page's address_space
1948 * @index: the page index
1949 * @filler: function to perform the read
1950 * @data: first arg to filler(data, page) function, often left as NULL
1951 *
1952 * Read into the page cache. If a page already exists, and PageUptodate() is
1953 * not set, try to fill the page then wait for it to become unlocked.
1954 *
1955 * If the page does not get brought uptodate, return -EIO.
1956 */
1957struct page *read_cache_page(struct address_space *mapping,
1958 pgoff_t index,
1959 int (*filler)(void *, struct page *),
1960 void *data)
1961{
1962 return wait_on_page_read(read_cache_page_async(mapping, index, filler, data));
1963}
1964EXPORT_SYMBOL(read_cache_page);
1965
1966/*
1967 * The logic we want is
1968 *
1969 * if suid or (sgid and xgrp)
1970 * remove privs
1971 */
1972int should_remove_suid(struct dentry *dentry)
1973{
1974 mode_t mode = dentry->d_inode->i_mode;
1975 int kill = 0;
1976
1977 /* suid always must be killed */
1978 if (unlikely(mode & S_ISUID))
1979 kill = ATTR_KILL_SUID;
1980
1981 /*
1982 * sgid without any exec bits is just a mandatory locking mark; leave
1983 * it alone. If some exec bits are set, it's a real sgid; kill it.
1984 */
1985 if (unlikely((mode & S_ISGID) && (mode & S_IXGRP)))
1986 kill |= ATTR_KILL_SGID;
1987
1988 if (unlikely(kill && !capable(CAP_FSETID) && S_ISREG(mode)))
1989 return kill;
1990
1991 return 0;
1992}
1993EXPORT_SYMBOL(should_remove_suid);
1994
1995static int __remove_suid(struct dentry *dentry, int kill)
1996{
1997 struct iattr newattrs;
1998
1999 newattrs.ia_valid = ATTR_FORCE | kill;
2000 return notify_change(dentry, &newattrs);
2001}
2002
2003int file_remove_suid(struct file *file)
2004{
2005 struct dentry *dentry = file->f_path.dentry;
2006 struct inode *inode = dentry->d_inode;
2007 int killsuid;
2008 int killpriv;
2009 int error = 0;
2010
2011 /* Fast path for nothing security related */
2012 if (IS_NOSEC(inode))
2013 return 0;
2014
2015 killsuid = should_remove_suid(dentry);
2016 killpriv = security_inode_need_killpriv(dentry);
2017
2018 if (killpriv < 0)
2019 return killpriv;
2020 if (killpriv)
2021 error = security_inode_killpriv(dentry);
2022 if (!error && killsuid)
2023 error = __remove_suid(dentry, killsuid);
2024 if (!error && (inode->i_sb->s_flags & MS_NOSEC))
2025 inode->i_flags |= S_NOSEC;
2026
2027 return error;
2028}
2029EXPORT_SYMBOL(file_remove_suid);
2030
2031static size_t __iovec_copy_from_user_inatomic(char *vaddr,
2032 const struct iovec *iov, size_t base, size_t bytes)
2033{
2034 size_t copied = 0, left = 0;
2035
2036 while (bytes) {
2037 char __user *buf = iov->iov_base + base;
2038 int copy = min(bytes, iov->iov_len - base);
2039
2040 base = 0;
2041 left = __copy_from_user_inatomic(vaddr, buf, copy);
2042 copied += copy;
2043 bytes -= copy;
2044 vaddr += copy;
2045 iov++;
2046
2047 if (unlikely(left))
2048 break;
2049 }
2050 return copied - left;
2051}
2052
2053/*
2054 * Copy as much as we can into the page and return the number of bytes which
2055 * were successfully copied. If a fault is encountered then return the number of
2056 * bytes which were copied.
2057 */
2058size_t iov_iter_copy_from_user_atomic(struct page *page,
2059 struct iov_iter *i, unsigned long offset, size_t bytes)
2060{
2061 char *kaddr;
2062 size_t copied;
2063
2064 BUG_ON(!in_atomic());
2065 kaddr = kmap_atomic(page, KM_USER0);
2066 if (likely(i->nr_segs == 1)) {
2067 int left;
2068 char __user *buf = i->iov->iov_base + i->iov_offset;
2069 left = __copy_from_user_inatomic(kaddr + offset, buf, bytes);
2070 copied = bytes - left;
2071 } else {
2072 copied = __iovec_copy_from_user_inatomic(kaddr + offset,
2073 i->iov, i->iov_offset, bytes);
2074 }
2075 kunmap_atomic(kaddr, KM_USER0);
2076
2077 return copied;
2078}
2079EXPORT_SYMBOL(iov_iter_copy_from_user_atomic);
2080
2081/*
2082 * This has the same sideeffects and return value as
2083 * iov_iter_copy_from_user_atomic().
2084 * The difference is that it attempts to resolve faults.
2085 * Page must not be locked.
2086 */
2087size_t iov_iter_copy_from_user(struct page *page,
2088 struct iov_iter *i, unsigned long offset, size_t bytes)
2089{
2090 char *kaddr;
2091 size_t copied;
2092
2093 kaddr = kmap(page);
2094 if (likely(i->nr_segs == 1)) {
2095 int left;
2096 char __user *buf = i->iov->iov_base + i->iov_offset;
2097 left = __copy_from_user(kaddr + offset, buf, bytes);
2098 copied = bytes - left;
2099 } else {
2100 copied = __iovec_copy_from_user_inatomic(kaddr + offset,
2101 i->iov, i->iov_offset, bytes);
2102 }
2103 kunmap(page);
2104 return copied;
2105}
2106EXPORT_SYMBOL(iov_iter_copy_from_user);
2107
2108void iov_iter_advance(struct iov_iter *i, size_t bytes)
2109{
2110 BUG_ON(i->count < bytes);
2111
2112 if (likely(i->nr_segs == 1)) {
2113 i->iov_offset += bytes;
2114 i->count -= bytes;
2115 } else {
2116 const struct iovec *iov = i->iov;
2117 size_t base = i->iov_offset;
2118
2119 /*
2120 * The !iov->iov_len check ensures we skip over unlikely
2121 * zero-length segments (without overruning the iovec).
2122 */
2123 while (bytes || unlikely(i->count && !iov->iov_len)) {
2124 int copy;
2125
2126 copy = min(bytes, iov->iov_len - base);
2127 BUG_ON(!i->count || i->count < copy);
2128 i->count -= copy;
2129 bytes -= copy;
2130 base += copy;
2131 if (iov->iov_len == base) {
2132 iov++;
2133 base = 0;
2134 }
2135 }
2136 i->iov = iov;
2137 i->iov_offset = base;
2138 }
2139}
2140EXPORT_SYMBOL(iov_iter_advance);
2141
2142/*
2143 * Fault in the first iovec of the given iov_iter, to a maximum length
2144 * of bytes. Returns 0 on success, or non-zero if the memory could not be
2145 * accessed (ie. because it is an invalid address).
2146 *
2147 * writev-intensive code may want this to prefault several iovecs -- that
2148 * would be possible (callers must not rely on the fact that _only_ the
2149 * first iovec will be faulted with the current implementation).
2150 */
2151int iov_iter_fault_in_readable(struct iov_iter *i, size_t bytes)
2152{
2153 char __user *buf = i->iov->iov_base + i->iov_offset;
2154 bytes = min(bytes, i->iov->iov_len - i->iov_offset);
2155 return fault_in_pages_readable(buf, bytes);
2156}
2157EXPORT_SYMBOL(iov_iter_fault_in_readable);
2158
2159/*
2160 * Return the count of just the current iov_iter segment.
2161 */
2162size_t iov_iter_single_seg_count(struct iov_iter *i)
2163{
2164 const struct iovec *iov = i->iov;
2165 if (i->nr_segs == 1)
2166 return i->count;
2167 else
2168 return min(i->count, iov->iov_len - i->iov_offset);
2169}
2170EXPORT_SYMBOL(iov_iter_single_seg_count);
2171
2172/*
2173 * Performs necessary checks before doing a write
2174 *
2175 * Can adjust writing position or amount of bytes to write.
2176 * Returns appropriate error code that caller should return or
2177 * zero in case that write should be allowed.
2178 */
2179inline int generic_write_checks(struct file *file, loff_t *pos, size_t *count, int isblk)
2180{
2181 struct inode *inode = file->f_mapping->host;
2182 unsigned long limit = rlimit(RLIMIT_FSIZE);
2183
2184 if (unlikely(*pos < 0))
2185 return -EINVAL;
2186
2187 if (!isblk) {
2188 /* FIXME: this is for backwards compatibility with 2.4 */
2189 if (file->f_flags & O_APPEND)
2190 *pos = i_size_read(inode);
2191
2192 if (limit != RLIM_INFINITY) {
2193 if (*pos >= limit) {
2194 send_sig(SIGXFSZ, current, 0);
2195 return -EFBIG;
2196 }
2197 if (*count > limit - (typeof(limit))*pos) {
2198 *count = limit - (typeof(limit))*pos;
2199 }
2200 }
2201 }
2202
2203 /*
2204 * LFS rule
2205 */
2206 if (unlikely(*pos + *count > MAX_NON_LFS &&
2207 !(file->f_flags & O_LARGEFILE))) {
2208 if (*pos >= MAX_NON_LFS) {
2209 return -EFBIG;
2210 }
2211 if (*count > MAX_NON_LFS - (unsigned long)*pos) {
2212 *count = MAX_NON_LFS - (unsigned long)*pos;
2213 }
2214 }
2215
2216 /*
2217 * Are we about to exceed the fs block limit ?
2218 *
2219 * If we have written data it becomes a short write. If we have
2220 * exceeded without writing data we send a signal and return EFBIG.
2221 * Linus frestrict idea will clean these up nicely..
2222 */
2223 if (likely(!isblk)) {
2224 if (unlikely(*pos >= inode->i_sb->s_maxbytes)) {
2225 if (*count || *pos > inode->i_sb->s_maxbytes) {
2226 return -EFBIG;
2227 }
2228 /* zero-length writes at ->s_maxbytes are OK */
2229 }
2230
2231 if (unlikely(*pos + *count > inode->i_sb->s_maxbytes))
2232 *count = inode->i_sb->s_maxbytes - *pos;
2233 } else {
2234#ifdef CONFIG_BLOCK
2235 loff_t isize;
2236 if (bdev_read_only(I_BDEV(inode)))
2237 return -EPERM;
2238 isize = i_size_read(inode);
2239 if (*pos >= isize) {
2240 if (*count || *pos > isize)
2241 return -ENOSPC;
2242 }
2243
2244 if (*pos + *count > isize)
2245 *count = isize - *pos;
2246#else
2247 return -EPERM;
2248#endif
2249 }
2250 return 0;
2251}
2252EXPORT_SYMBOL(generic_write_checks);
2253
2254int pagecache_write_begin(struct file *file, struct address_space *mapping,
2255 loff_t pos, unsigned len, unsigned flags,
2256 struct page **pagep, void **fsdata)
2257{
2258 const struct address_space_operations *aops = mapping->a_ops;
2259
2260 return aops->write_begin(file, mapping, pos, len, flags,
2261 pagep, fsdata);
2262}
2263EXPORT_SYMBOL(pagecache_write_begin);
2264
2265int pagecache_write_end(struct file *file, struct address_space *mapping,
2266 loff_t pos, unsigned len, unsigned copied,
2267 struct page *page, void *fsdata)
2268{
2269 const struct address_space_operations *aops = mapping->a_ops;
2270
2271 mark_page_accessed(page);
2272 return aops->write_end(file, mapping, pos, len, copied, page, fsdata);
2273}
2274EXPORT_SYMBOL(pagecache_write_end);
2275
2276ssize_t
2277generic_file_direct_write(struct kiocb *iocb, const struct iovec *iov,
2278 unsigned long *nr_segs, loff_t pos, loff_t *ppos,
2279 size_t count, size_t ocount)
2280{
2281 struct file *file = iocb->ki_filp;
2282 struct address_space *mapping = file->f_mapping;
2283 struct inode *inode = mapping->host;
2284 ssize_t written;
2285 size_t write_len;
2286 pgoff_t end;
2287
2288 if (count != ocount)
2289 *nr_segs = iov_shorten((struct iovec *)iov, *nr_segs, count);
2290
2291 write_len = iov_length(iov, *nr_segs);
2292 end = (pos + write_len - 1) >> PAGE_CACHE_SHIFT;
2293
2294 written = filemap_write_and_wait_range(mapping, pos, pos + write_len - 1);
2295 if (written)
2296 goto out;
2297
2298 /*
2299 * After a write we want buffered reads to be sure to go to disk to get
2300 * the new data. We invalidate clean cached page from the region we're
2301 * about to write. We do this *before* the write so that we can return
2302 * without clobbering -EIOCBQUEUED from ->direct_IO().
2303 */
2304 if (mapping->nrpages) {
2305 written = invalidate_inode_pages2_range(mapping,
2306 pos >> PAGE_CACHE_SHIFT, end);
2307 /*
2308 * If a page can not be invalidated, return 0 to fall back
2309 * to buffered write.
2310 */
2311 if (written) {
2312 if (written == -EBUSY)
2313 return 0;
2314 goto out;
2315 }
2316 }
2317
2318 written = mapping->a_ops->direct_IO(WRITE, iocb, iov, pos, *nr_segs);
2319
2320 /*
2321 * Finally, try again to invalidate clean pages which might have been
2322 * cached by non-direct readahead, or faulted in by get_user_pages()
2323 * if the source of the write was an mmap'ed region of the file
2324 * we're writing. Either one is a pretty crazy thing to do,
2325 * so we don't support it 100%. If this invalidation
2326 * fails, tough, the write still worked...
2327 */
2328 if (mapping->nrpages) {
2329 invalidate_inode_pages2_range(mapping,
2330 pos >> PAGE_CACHE_SHIFT, end);
2331 }
2332
2333 if (written > 0) {
2334 pos += written;
2335 if (pos > i_size_read(inode) && !S_ISBLK(inode->i_mode)) {
2336 i_size_write(inode, pos);
2337 mark_inode_dirty(inode);
2338 }
2339 *ppos = pos;
2340 }
2341out:
2342 return written;
2343}
2344EXPORT_SYMBOL(generic_file_direct_write);
2345
2346/*
2347 * Find or create a page at the given pagecache position. Return the locked
2348 * page. This function is specifically for buffered writes.
2349 */
2350struct page *grab_cache_page_write_begin(struct address_space *mapping,
2351 pgoff_t index, unsigned flags)
2352{
2353 int status;
2354 struct page *page;
2355 gfp_t gfp_notmask = 0;
2356 if (flags & AOP_FLAG_NOFS)
2357 gfp_notmask = __GFP_FS;
2358repeat:
2359 page = find_lock_page(mapping, index);
2360 if (page)
2361 goto found;
2362
2363 page = __page_cache_alloc(mapping_gfp_mask(mapping) & ~gfp_notmask);
2364 if (!page)
2365 return NULL;
2366 status = add_to_page_cache_lru(page, mapping, index,
2367 GFP_KERNEL & ~gfp_notmask);
2368 if (unlikely(status)) {
2369 page_cache_release(page);
2370 if (status == -EEXIST)
2371 goto repeat;
2372 return NULL;
2373 }
2374found:
2375 wait_on_page_writeback(page);
2376 return page;
2377}
2378EXPORT_SYMBOL(grab_cache_page_write_begin);
2379
2380static ssize_t generic_perform_write(struct file *file,
2381 struct iov_iter *i, loff_t pos)
2382{
2383 struct address_space *mapping = file->f_mapping;
2384 const struct address_space_operations *a_ops = mapping->a_ops;
2385 long status = 0;
2386 ssize_t written = 0;
2387 unsigned int flags = 0;
2388
2389 /*
2390 * Copies from kernel address space cannot fail (NFSD is a big user).
2391 */
2392 if (segment_eq(get_fs(), KERNEL_DS))
2393 flags |= AOP_FLAG_UNINTERRUPTIBLE;
2394
2395 do {
2396 struct page *page;
2397 unsigned long offset; /* Offset into pagecache page */
2398 unsigned long bytes; /* Bytes to write to page */
2399 size_t copied; /* Bytes copied from user */
2400 void *fsdata;
2401
2402 offset = (pos & (PAGE_CACHE_SIZE - 1));
2403 bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset,
2404 iov_iter_count(i));
2405
2406again:
2407
2408 /*
2409 * Bring in the user page that we will copy from _first_.
2410 * Otherwise there's a nasty deadlock on copying from the
2411 * same page as we're writing to, without it being marked
2412 * up-to-date.
2413 *
2414 * Not only is this an optimisation, but it is also required
2415 * to check that the address is actually valid, when atomic
2416 * usercopies are used, below.
2417 */
2418 if (unlikely(iov_iter_fault_in_readable(i, bytes))) {
2419 status = -EFAULT;
2420 break;
2421 }
2422
2423 status = a_ops->write_begin(file, mapping, pos, bytes, flags,
2424 &page, &fsdata);
2425 if (unlikely(status))
2426 break;
2427
2428 if (mapping_writably_mapped(mapping))
2429 flush_dcache_page(page);
2430
2431 pagefault_disable();
2432 copied = iov_iter_copy_from_user_atomic(page, i, offset, bytes);
2433 pagefault_enable();
2434 flush_dcache_page(page);
2435
2436 mark_page_accessed(page);
2437 status = a_ops->write_end(file, mapping, pos, bytes, copied,
2438 page, fsdata);
2439 if (unlikely(status < 0))
2440 break;
2441 copied = status;
2442
2443 cond_resched();
2444
2445 iov_iter_advance(i, copied);
2446 if (unlikely(copied == 0)) {
2447 /*
2448 * If we were unable to copy any data at all, we must
2449 * fall back to a single segment length write.
2450 *
2451 * If we didn't fallback here, we could livelock
2452 * because not all segments in the iov can be copied at
2453 * once without a pagefault.
2454 */
2455 bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset,
2456 iov_iter_single_seg_count(i));
2457 goto again;
2458 }
2459 pos += copied;
2460 written += copied;
2461
2462 balance_dirty_pages_ratelimited(mapping);
2463
2464 } while (iov_iter_count(i));
2465
2466 return written ? written : status;
2467}
2468
2469ssize_t
2470generic_file_buffered_write(struct kiocb *iocb, const struct iovec *iov,
2471 unsigned long nr_segs, loff_t pos, loff_t *ppos,
2472 size_t count, ssize_t written)
2473{
2474 struct file *file = iocb->ki_filp;
2475 ssize_t status;
2476 struct iov_iter i;
2477
2478 iov_iter_init(&i, iov, nr_segs, count, written);
2479 status = generic_perform_write(file, &i, pos);
2480
2481 if (likely(status >= 0)) {
2482 written += status;
2483 *ppos = pos + status;
2484 }
2485
2486 return written ? written : status;
2487}
2488EXPORT_SYMBOL(generic_file_buffered_write);
2489
2490/**
2491 * __generic_file_aio_write - write data to a file
2492 * @iocb: IO state structure (file, offset, etc.)
2493 * @iov: vector with data to write
2494 * @nr_segs: number of segments in the vector
2495 * @ppos: position where to write
2496 *
2497 * This function does all the work needed for actually writing data to a
2498 * file. It does all basic checks, removes SUID from the file, updates
2499 * modification times and calls proper subroutines depending on whether we
2500 * do direct IO or a standard buffered write.
2501 *
2502 * It expects i_mutex to be grabbed unless we work on a block device or similar
2503 * object which does not need locking at all.
2504 *
2505 * This function does *not* take care of syncing data in case of O_SYNC write.
2506 * A caller has to handle it. This is mainly due to the fact that we want to
2507 * avoid syncing under i_mutex.
2508 */
2509ssize_t __generic_file_aio_write(struct kiocb *iocb, const struct iovec *iov,
2510 unsigned long nr_segs, loff_t *ppos)
2511{
2512 struct file *file = iocb->ki_filp;
2513 struct address_space * mapping = file->f_mapping;
2514 size_t ocount; /* original count */
2515 size_t count; /* after file limit checks */
2516 struct inode *inode = mapping->host;
2517 loff_t pos;
2518 ssize_t written;
2519 ssize_t err;
2520
2521 ocount = 0;
2522 err = generic_segment_checks(iov, &nr_segs, &ocount, VERIFY_READ);
2523 if (err)
2524 return err;
2525
2526 count = ocount;
2527 pos = *ppos;
2528
2529 vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
2530
2531 /* We can write back this queue in page reclaim */
2532 current->backing_dev_info = mapping->backing_dev_info;
2533 written = 0;
2534
2535 err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
2536 if (err)
2537 goto out;
2538
2539 if (count == 0)
2540 goto out;
2541
2542 err = file_remove_suid(file);
2543 if (err)
2544 goto out;
2545
2546 file_update_time(file);
2547
2548 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */
2549 if (unlikely(file->f_flags & O_DIRECT)) {
2550 loff_t endbyte;
2551 ssize_t written_buffered;
2552
2553 written = generic_file_direct_write(iocb, iov, &nr_segs, pos,
2554 ppos, count, ocount);
2555 if (written < 0 || written == count)
2556 goto out;
2557 /*
2558 * direct-io write to a hole: fall through to buffered I/O
2559 * for completing the rest of the request.
2560 */
2561 pos += written;
2562 count -= written;
2563 written_buffered = generic_file_buffered_write(iocb, iov,
2564 nr_segs, pos, ppos, count,
2565 written);
2566 /*
2567 * If generic_file_buffered_write() retuned a synchronous error
2568 * then we want to return the number of bytes which were
2569 * direct-written, or the error code if that was zero. Note
2570 * that this differs from normal direct-io semantics, which
2571 * will return -EFOO even if some bytes were written.
2572 */
2573 if (written_buffered < 0) {
2574 err = written_buffered;
2575 goto out;
2576 }
2577
2578 /*
2579 * We need to ensure that the page cache pages are written to
2580 * disk and invalidated to preserve the expected O_DIRECT
2581 * semantics.
2582 */
2583 endbyte = pos + written_buffered - written - 1;
2584 err = filemap_write_and_wait_range(file->f_mapping, pos, endbyte);
2585 if (err == 0) {
2586 written = written_buffered;
2587 invalidate_mapping_pages(mapping,
2588 pos >> PAGE_CACHE_SHIFT,
2589 endbyte >> PAGE_CACHE_SHIFT);
2590 } else {
2591 /*
2592 * We don't know how much we wrote, so just return
2593 * the number of bytes which were direct-written
2594 */
2595 }
2596 } else {
2597 written = generic_file_buffered_write(iocb, iov, nr_segs,
2598 pos, ppos, count, written);
2599 }
2600out:
2601 current->backing_dev_info = NULL;
2602 return written ? written : err;
2603}
2604EXPORT_SYMBOL(__generic_file_aio_write);
2605
2606/**
2607 * generic_file_aio_write - write data to a file
2608 * @iocb: IO state structure
2609 * @iov: vector with data to write
2610 * @nr_segs: number of segments in the vector
2611 * @pos: position in file where to write
2612 *
2613 * This is a wrapper around __generic_file_aio_write() to be used by most
2614 * filesystems. It takes care of syncing the file in case of O_SYNC file
2615 * and acquires i_mutex as needed.
2616 */
2617ssize_t generic_file_aio_write(struct kiocb *iocb, const struct iovec *iov,
2618 unsigned long nr_segs, loff_t pos)
2619{
2620 struct file *file = iocb->ki_filp;
2621 struct inode *inode = file->f_mapping->host;
2622 struct blk_plug plug;
2623 ssize_t ret;
2624
2625 BUG_ON(iocb->ki_pos != pos);
2626
2627 mutex_lock(&inode->i_mutex);
2628 blk_start_plug(&plug);
2629 ret = __generic_file_aio_write(iocb, iov, nr_segs, &iocb->ki_pos);
2630 mutex_unlock(&inode->i_mutex);
2631
2632 if (ret > 0 || ret == -EIOCBQUEUED) {
2633 ssize_t err;
2634
2635 err = generic_write_sync(file, pos, ret);
2636 if (err < 0 && ret > 0)
2637 ret = err;
2638 }
2639 blk_finish_plug(&plug);
2640 return ret;
2641}
2642EXPORT_SYMBOL(generic_file_aio_write);
2643
2644/**
2645 * try_to_release_page() - release old fs-specific metadata on a page
2646 *
2647 * @page: the page which the kernel is trying to free
2648 * @gfp_mask: memory allocation flags (and I/O mode)
2649 *
2650 * The address_space is to try to release any data against the page
2651 * (presumably at page->private). If the release was successful, return `1'.
2652 * Otherwise return zero.
2653 *
2654 * This may also be called if PG_fscache is set on a page, indicating that the
2655 * page is known to the local caching routines.
2656 *
2657 * The @gfp_mask argument specifies whether I/O may be performed to release
2658 * this page (__GFP_IO), and whether the call may block (__GFP_WAIT & __GFP_FS).
2659 *
2660 */
2661int try_to_release_page(struct page *page, gfp_t gfp_mask)
2662{
2663 struct address_space * const mapping = page->mapping;
2664
2665 BUG_ON(!PageLocked(page));
2666 if (PageWriteback(page))
2667 return 0;
2668
2669 if (mapping && mapping->a_ops->releasepage)
2670 return mapping->a_ops->releasepage(page, gfp_mask);
2671 return try_to_free_buffers(page);
2672}
2673
2674EXPORT_SYMBOL(try_to_release_page);
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * linux/mm/filemap.c
4 *
5 * Copyright (C) 1994-1999 Linus Torvalds
6 */
7
8/*
9 * This file handles the generic file mmap semantics used by
10 * most "normal" filesystems (but you don't /have/ to use this:
11 * the NFS filesystem used to do this differently, for example)
12 */
13#include <linux/export.h>
14#include <linux/compiler.h>
15#include <linux/dax.h>
16#include <linux/fs.h>
17#include <linux/sched/signal.h>
18#include <linux/uaccess.h>
19#include <linux/capability.h>
20#include <linux/kernel_stat.h>
21#include <linux/gfp.h>
22#include <linux/mm.h>
23#include <linux/swap.h>
24#include <linux/mman.h>
25#include <linux/pagemap.h>
26#include <linux/file.h>
27#include <linux/uio.h>
28#include <linux/error-injection.h>
29#include <linux/hash.h>
30#include <linux/writeback.h>
31#include <linux/backing-dev.h>
32#include <linux/pagevec.h>
33#include <linux/blkdev.h>
34#include <linux/security.h>
35#include <linux/cpuset.h>
36#include <linux/hugetlb.h>
37#include <linux/memcontrol.h>
38#include <linux/cleancache.h>
39#include <linux/shmem_fs.h>
40#include <linux/rmap.h>
41#include <linux/delayacct.h>
42#include <linux/psi.h>
43#include <linux/ramfs.h>
44#include "internal.h"
45
46#define CREATE_TRACE_POINTS
47#include <trace/events/filemap.h>
48
49/*
50 * FIXME: remove all knowledge of the buffer layer from the core VM
51 */
52#include <linux/buffer_head.h> /* for try_to_free_buffers */
53
54#include <asm/mman.h>
55
56/*
57 * Shared mappings implemented 30.11.1994. It's not fully working yet,
58 * though.
59 *
60 * Shared mappings now work. 15.8.1995 Bruno.
61 *
62 * finished 'unifying' the page and buffer cache and SMP-threaded the
63 * page-cache, 21.05.1999, Ingo Molnar <mingo@redhat.com>
64 *
65 * SMP-threaded pagemap-LRU 1999, Andrea Arcangeli <andrea@suse.de>
66 */
67
68/*
69 * Lock ordering:
70 *
71 * ->i_mmap_rwsem (truncate_pagecache)
72 * ->private_lock (__free_pte->__set_page_dirty_buffers)
73 * ->swap_lock (exclusive_swap_page, others)
74 * ->i_pages lock
75 *
76 * ->i_mutex
77 * ->i_mmap_rwsem (truncate->unmap_mapping_range)
78 *
79 * ->mmap_sem
80 * ->i_mmap_rwsem
81 * ->page_table_lock or pte_lock (various, mainly in memory.c)
82 * ->i_pages lock (arch-dependent flush_dcache_mmap_lock)
83 *
84 * ->mmap_sem
85 * ->lock_page (access_process_vm)
86 *
87 * ->i_mutex (generic_perform_write)
88 * ->mmap_sem (fault_in_pages_readable->do_page_fault)
89 *
90 * bdi->wb.list_lock
91 * sb_lock (fs/fs-writeback.c)
92 * ->i_pages lock (__sync_single_inode)
93 *
94 * ->i_mmap_rwsem
95 * ->anon_vma.lock (vma_adjust)
96 *
97 * ->anon_vma.lock
98 * ->page_table_lock or pte_lock (anon_vma_prepare and various)
99 *
100 * ->page_table_lock or pte_lock
101 * ->swap_lock (try_to_unmap_one)
102 * ->private_lock (try_to_unmap_one)
103 * ->i_pages lock (try_to_unmap_one)
104 * ->pgdat->lru_lock (follow_page->mark_page_accessed)
105 * ->pgdat->lru_lock (check_pte_range->isolate_lru_page)
106 * ->private_lock (page_remove_rmap->set_page_dirty)
107 * ->i_pages lock (page_remove_rmap->set_page_dirty)
108 * bdi.wb->list_lock (page_remove_rmap->set_page_dirty)
109 * ->inode->i_lock (page_remove_rmap->set_page_dirty)
110 * ->memcg->move_lock (page_remove_rmap->lock_page_memcg)
111 * bdi.wb->list_lock (zap_pte_range->set_page_dirty)
112 * ->inode->i_lock (zap_pte_range->set_page_dirty)
113 * ->private_lock (zap_pte_range->__set_page_dirty_buffers)
114 *
115 * ->i_mmap_rwsem
116 * ->tasklist_lock (memory_failure, collect_procs_ao)
117 */
118
119static void page_cache_delete(struct address_space *mapping,
120 struct page *page, void *shadow)
121{
122 XA_STATE(xas, &mapping->i_pages, page->index);
123 unsigned int nr = 1;
124
125 mapping_set_update(&xas, mapping);
126
127 /* hugetlb pages are represented by a single entry in the xarray */
128 if (!PageHuge(page)) {
129 xas_set_order(&xas, page->index, compound_order(page));
130 nr = compound_nr(page);
131 }
132
133 VM_BUG_ON_PAGE(!PageLocked(page), page);
134 VM_BUG_ON_PAGE(PageTail(page), page);
135 VM_BUG_ON_PAGE(nr != 1 && shadow, page);
136
137 xas_store(&xas, shadow);
138 xas_init_marks(&xas);
139
140 page->mapping = NULL;
141 /* Leave page->index set: truncation lookup relies upon it */
142
143 if (shadow) {
144 mapping->nrexceptional += nr;
145 /*
146 * Make sure the nrexceptional update is committed before
147 * the nrpages update so that final truncate racing
148 * with reclaim does not see both counters 0 at the
149 * same time and miss a shadow entry.
150 */
151 smp_wmb();
152 }
153 mapping->nrpages -= nr;
154}
155
156static void unaccount_page_cache_page(struct address_space *mapping,
157 struct page *page)
158{
159 int nr;
160
161 /*
162 * if we're uptodate, flush out into the cleancache, otherwise
163 * invalidate any existing cleancache entries. We can't leave
164 * stale data around in the cleancache once our page is gone
165 */
166 if (PageUptodate(page) && PageMappedToDisk(page))
167 cleancache_put_page(page);
168 else
169 cleancache_invalidate_page(mapping, page);
170
171 VM_BUG_ON_PAGE(PageTail(page), page);
172 VM_BUG_ON_PAGE(page_mapped(page), page);
173 if (!IS_ENABLED(CONFIG_DEBUG_VM) && unlikely(page_mapped(page))) {
174 int mapcount;
175
176 pr_alert("BUG: Bad page cache in process %s pfn:%05lx\n",
177 current->comm, page_to_pfn(page));
178 dump_page(page, "still mapped when deleted");
179 dump_stack();
180 add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE);
181
182 mapcount = page_mapcount(page);
183 if (mapping_exiting(mapping) &&
184 page_count(page) >= mapcount + 2) {
185 /*
186 * All vmas have already been torn down, so it's
187 * a good bet that actually the page is unmapped,
188 * and we'd prefer not to leak it: if we're wrong,
189 * some other bad page check should catch it later.
190 */
191 page_mapcount_reset(page);
192 page_ref_sub(page, mapcount);
193 }
194 }
195
196 /* hugetlb pages do not participate in page cache accounting. */
197 if (PageHuge(page))
198 return;
199
200 nr = hpage_nr_pages(page);
201
202 __mod_node_page_state(page_pgdat(page), NR_FILE_PAGES, -nr);
203 if (PageSwapBacked(page)) {
204 __mod_node_page_state(page_pgdat(page), NR_SHMEM, -nr);
205 if (PageTransHuge(page))
206 __dec_node_page_state(page, NR_SHMEM_THPS);
207 } else if (PageTransHuge(page)) {
208 __dec_node_page_state(page, NR_FILE_THPS);
209 filemap_nr_thps_dec(mapping);
210 }
211
212 /*
213 * At this point page must be either written or cleaned by
214 * truncate. Dirty page here signals a bug and loss of
215 * unwritten data.
216 *
217 * This fixes dirty accounting after removing the page entirely
218 * but leaves PageDirty set: it has no effect for truncated
219 * page and anyway will be cleared before returning page into
220 * buddy allocator.
221 */
222 if (WARN_ON_ONCE(PageDirty(page)))
223 account_page_cleaned(page, mapping, inode_to_wb(mapping->host));
224}
225
226/*
227 * Delete a page from the page cache and free it. Caller has to make
228 * sure the page is locked and that nobody else uses it - or that usage
229 * is safe. The caller must hold the i_pages lock.
230 */
231void __delete_from_page_cache(struct page *page, void *shadow)
232{
233 struct address_space *mapping = page->mapping;
234
235 trace_mm_filemap_delete_from_page_cache(page);
236
237 unaccount_page_cache_page(mapping, page);
238 page_cache_delete(mapping, page, shadow);
239}
240
241static void page_cache_free_page(struct address_space *mapping,
242 struct page *page)
243{
244 void (*freepage)(struct page *);
245
246 freepage = mapping->a_ops->freepage;
247 if (freepage)
248 freepage(page);
249
250 if (PageTransHuge(page) && !PageHuge(page)) {
251 page_ref_sub(page, HPAGE_PMD_NR);
252 VM_BUG_ON_PAGE(page_count(page) <= 0, page);
253 } else {
254 put_page(page);
255 }
256}
257
258/**
259 * delete_from_page_cache - delete page from page cache
260 * @page: the page which the kernel is trying to remove from page cache
261 *
262 * This must be called only on pages that have been verified to be in the page
263 * cache and locked. It will never put the page into the free list, the caller
264 * has a reference on the page.
265 */
266void delete_from_page_cache(struct page *page)
267{
268 struct address_space *mapping = page_mapping(page);
269 unsigned long flags;
270
271 BUG_ON(!PageLocked(page));
272 xa_lock_irqsave(&mapping->i_pages, flags);
273 __delete_from_page_cache(page, NULL);
274 xa_unlock_irqrestore(&mapping->i_pages, flags);
275
276 page_cache_free_page(mapping, page);
277}
278EXPORT_SYMBOL(delete_from_page_cache);
279
280/*
281 * page_cache_delete_batch - delete several pages from page cache
282 * @mapping: the mapping to which pages belong
283 * @pvec: pagevec with pages to delete
284 *
285 * The function walks over mapping->i_pages and removes pages passed in @pvec
286 * from the mapping. The function expects @pvec to be sorted by page index
287 * and is optimised for it to be dense.
288 * It tolerates holes in @pvec (mapping entries at those indices are not
289 * modified). The function expects only THP head pages to be present in the
290 * @pvec.
291 *
292 * The function expects the i_pages lock to be held.
293 */
294static void page_cache_delete_batch(struct address_space *mapping,
295 struct pagevec *pvec)
296{
297 XA_STATE(xas, &mapping->i_pages, pvec->pages[0]->index);
298 int total_pages = 0;
299 int i = 0;
300 struct page *page;
301
302 mapping_set_update(&xas, mapping);
303 xas_for_each(&xas, page, ULONG_MAX) {
304 if (i >= pagevec_count(pvec))
305 break;
306
307 /* A swap/dax/shadow entry got inserted? Skip it. */
308 if (xa_is_value(page))
309 continue;
310 /*
311 * A page got inserted in our range? Skip it. We have our
312 * pages locked so they are protected from being removed.
313 * If we see a page whose index is higher than ours, it
314 * means our page has been removed, which shouldn't be
315 * possible because we're holding the PageLock.
316 */
317 if (page != pvec->pages[i]) {
318 VM_BUG_ON_PAGE(page->index > pvec->pages[i]->index,
319 page);
320 continue;
321 }
322
323 WARN_ON_ONCE(!PageLocked(page));
324
325 if (page->index == xas.xa_index)
326 page->mapping = NULL;
327 /* Leave page->index set: truncation lookup relies on it */
328
329 /*
330 * Move to the next page in the vector if this is a regular
331 * page or the index is of the last sub-page of this compound
332 * page.
333 */
334 if (page->index + compound_nr(page) - 1 == xas.xa_index)
335 i++;
336 xas_store(&xas, NULL);
337 total_pages++;
338 }
339 mapping->nrpages -= total_pages;
340}
341
342void delete_from_page_cache_batch(struct address_space *mapping,
343 struct pagevec *pvec)
344{
345 int i;
346 unsigned long flags;
347
348 if (!pagevec_count(pvec))
349 return;
350
351 xa_lock_irqsave(&mapping->i_pages, flags);
352 for (i = 0; i < pagevec_count(pvec); i++) {
353 trace_mm_filemap_delete_from_page_cache(pvec->pages[i]);
354
355 unaccount_page_cache_page(mapping, pvec->pages[i]);
356 }
357 page_cache_delete_batch(mapping, pvec);
358 xa_unlock_irqrestore(&mapping->i_pages, flags);
359
360 for (i = 0; i < pagevec_count(pvec); i++)
361 page_cache_free_page(mapping, pvec->pages[i]);
362}
363
364int filemap_check_errors(struct address_space *mapping)
365{
366 int ret = 0;
367 /* Check for outstanding write errors */
368 if (test_bit(AS_ENOSPC, &mapping->flags) &&
369 test_and_clear_bit(AS_ENOSPC, &mapping->flags))
370 ret = -ENOSPC;
371 if (test_bit(AS_EIO, &mapping->flags) &&
372 test_and_clear_bit(AS_EIO, &mapping->flags))
373 ret = -EIO;
374 return ret;
375}
376EXPORT_SYMBOL(filemap_check_errors);
377
378static int filemap_check_and_keep_errors(struct address_space *mapping)
379{
380 /* Check for outstanding write errors */
381 if (test_bit(AS_EIO, &mapping->flags))
382 return -EIO;
383 if (test_bit(AS_ENOSPC, &mapping->flags))
384 return -ENOSPC;
385 return 0;
386}
387
388/**
389 * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range
390 * @mapping: address space structure to write
391 * @start: offset in bytes where the range starts
392 * @end: offset in bytes where the range ends (inclusive)
393 * @sync_mode: enable synchronous operation
394 *
395 * Start writeback against all of a mapping's dirty pages that lie
396 * within the byte offsets <start, end> inclusive.
397 *
398 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
399 * opposed to a regular memory cleansing writeback. The difference between
400 * these two operations is that if a dirty page/buffer is encountered, it must
401 * be waited upon, and not just skipped over.
402 *
403 * Return: %0 on success, negative error code otherwise.
404 */
405int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
406 loff_t end, int sync_mode)
407{
408 int ret;
409 struct writeback_control wbc = {
410 .sync_mode = sync_mode,
411 .nr_to_write = LONG_MAX,
412 .range_start = start,
413 .range_end = end,
414 };
415
416 if (!mapping_cap_writeback_dirty(mapping) ||
417 !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
418 return 0;
419
420 wbc_attach_fdatawrite_inode(&wbc, mapping->host);
421 ret = do_writepages(mapping, &wbc);
422 wbc_detach_inode(&wbc);
423 return ret;
424}
425
426static inline int __filemap_fdatawrite(struct address_space *mapping,
427 int sync_mode)
428{
429 return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode);
430}
431
432int filemap_fdatawrite(struct address_space *mapping)
433{
434 return __filemap_fdatawrite(mapping, WB_SYNC_ALL);
435}
436EXPORT_SYMBOL(filemap_fdatawrite);
437
438int filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
439 loff_t end)
440{
441 return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL);
442}
443EXPORT_SYMBOL(filemap_fdatawrite_range);
444
445/**
446 * filemap_flush - mostly a non-blocking flush
447 * @mapping: target address_space
448 *
449 * This is a mostly non-blocking flush. Not suitable for data-integrity
450 * purposes - I/O may not be started against all dirty pages.
451 *
452 * Return: %0 on success, negative error code otherwise.
453 */
454int filemap_flush(struct address_space *mapping)
455{
456 return __filemap_fdatawrite(mapping, WB_SYNC_NONE);
457}
458EXPORT_SYMBOL(filemap_flush);
459
460/**
461 * filemap_range_has_page - check if a page exists in range.
462 * @mapping: address space within which to check
463 * @start_byte: offset in bytes where the range starts
464 * @end_byte: offset in bytes where the range ends (inclusive)
465 *
466 * Find at least one page in the range supplied, usually used to check if
467 * direct writing in this range will trigger a writeback.
468 *
469 * Return: %true if at least one page exists in the specified range,
470 * %false otherwise.
471 */
472bool filemap_range_has_page(struct address_space *mapping,
473 loff_t start_byte, loff_t end_byte)
474{
475 struct page *page;
476 XA_STATE(xas, &mapping->i_pages, start_byte >> PAGE_SHIFT);
477 pgoff_t max = end_byte >> PAGE_SHIFT;
478
479 if (end_byte < start_byte)
480 return false;
481
482 rcu_read_lock();
483 for (;;) {
484 page = xas_find(&xas, max);
485 if (xas_retry(&xas, page))
486 continue;
487 /* Shadow entries don't count */
488 if (xa_is_value(page))
489 continue;
490 /*
491 * We don't need to try to pin this page; we're about to
492 * release the RCU lock anyway. It is enough to know that
493 * there was a page here recently.
494 */
495 break;
496 }
497 rcu_read_unlock();
498
499 return page != NULL;
500}
501EXPORT_SYMBOL(filemap_range_has_page);
502
503static void __filemap_fdatawait_range(struct address_space *mapping,
504 loff_t start_byte, loff_t end_byte)
505{
506 pgoff_t index = start_byte >> PAGE_SHIFT;
507 pgoff_t end = end_byte >> PAGE_SHIFT;
508 struct pagevec pvec;
509 int nr_pages;
510
511 if (end_byte < start_byte)
512 return;
513
514 pagevec_init(&pvec);
515 while (index <= end) {
516 unsigned i;
517
518 nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index,
519 end, PAGECACHE_TAG_WRITEBACK);
520 if (!nr_pages)
521 break;
522
523 for (i = 0; i < nr_pages; i++) {
524 struct page *page = pvec.pages[i];
525
526 wait_on_page_writeback(page);
527 ClearPageError(page);
528 }
529 pagevec_release(&pvec);
530 cond_resched();
531 }
532}
533
534/**
535 * filemap_fdatawait_range - wait for writeback to complete
536 * @mapping: address space structure to wait for
537 * @start_byte: offset in bytes where the range starts
538 * @end_byte: offset in bytes where the range ends (inclusive)
539 *
540 * Walk the list of under-writeback pages of the given address space
541 * in the given range and wait for all of them. Check error status of
542 * the address space and return it.
543 *
544 * Since the error status of the address space is cleared by this function,
545 * callers are responsible for checking the return value and handling and/or
546 * reporting the error.
547 *
548 * Return: error status of the address space.
549 */
550int filemap_fdatawait_range(struct address_space *mapping, loff_t start_byte,
551 loff_t end_byte)
552{
553 __filemap_fdatawait_range(mapping, start_byte, end_byte);
554 return filemap_check_errors(mapping);
555}
556EXPORT_SYMBOL(filemap_fdatawait_range);
557
558/**
559 * filemap_fdatawait_range_keep_errors - wait for writeback to complete
560 * @mapping: address space structure to wait for
561 * @start_byte: offset in bytes where the range starts
562 * @end_byte: offset in bytes where the range ends (inclusive)
563 *
564 * Walk the list of under-writeback pages of the given address space in the
565 * given range and wait for all of them. Unlike filemap_fdatawait_range(),
566 * this function does not clear error status of the address space.
567 *
568 * Use this function if callers don't handle errors themselves. Expected
569 * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2),
570 * fsfreeze(8)
571 */
572int filemap_fdatawait_range_keep_errors(struct address_space *mapping,
573 loff_t start_byte, loff_t end_byte)
574{
575 __filemap_fdatawait_range(mapping, start_byte, end_byte);
576 return filemap_check_and_keep_errors(mapping);
577}
578EXPORT_SYMBOL(filemap_fdatawait_range_keep_errors);
579
580/**
581 * file_fdatawait_range - wait for writeback to complete
582 * @file: file pointing to address space structure to wait for
583 * @start_byte: offset in bytes where the range starts
584 * @end_byte: offset in bytes where the range ends (inclusive)
585 *
586 * Walk the list of under-writeback pages of the address space that file
587 * refers to, in the given range and wait for all of them. Check error
588 * status of the address space vs. the file->f_wb_err cursor and return it.
589 *
590 * Since the error status of the file is advanced by this function,
591 * callers are responsible for checking the return value and handling and/or
592 * reporting the error.
593 *
594 * Return: error status of the address space vs. the file->f_wb_err cursor.
595 */
596int file_fdatawait_range(struct file *file, loff_t start_byte, loff_t end_byte)
597{
598 struct address_space *mapping = file->f_mapping;
599
600 __filemap_fdatawait_range(mapping, start_byte, end_byte);
601 return file_check_and_advance_wb_err(file);
602}
603EXPORT_SYMBOL(file_fdatawait_range);
604
605/**
606 * filemap_fdatawait_keep_errors - wait for writeback without clearing errors
607 * @mapping: address space structure to wait for
608 *
609 * Walk the list of under-writeback pages of the given address space
610 * and wait for all of them. Unlike filemap_fdatawait(), this function
611 * does not clear error status of the address space.
612 *
613 * Use this function if callers don't handle errors themselves. Expected
614 * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2),
615 * fsfreeze(8)
616 *
617 * Return: error status of the address space.
618 */
619int filemap_fdatawait_keep_errors(struct address_space *mapping)
620{
621 __filemap_fdatawait_range(mapping, 0, LLONG_MAX);
622 return filemap_check_and_keep_errors(mapping);
623}
624EXPORT_SYMBOL(filemap_fdatawait_keep_errors);
625
626/* Returns true if writeback might be needed or already in progress. */
627static bool mapping_needs_writeback(struct address_space *mapping)
628{
629 if (dax_mapping(mapping))
630 return mapping->nrexceptional;
631
632 return mapping->nrpages;
633}
634
635int filemap_write_and_wait(struct address_space *mapping)
636{
637 int err = 0;
638
639 if (mapping_needs_writeback(mapping)) {
640 err = filemap_fdatawrite(mapping);
641 /*
642 * Even if the above returned error, the pages may be
643 * written partially (e.g. -ENOSPC), so we wait for it.
644 * But the -EIO is special case, it may indicate the worst
645 * thing (e.g. bug) happened, so we avoid waiting for it.
646 */
647 if (err != -EIO) {
648 int err2 = filemap_fdatawait(mapping);
649 if (!err)
650 err = err2;
651 } else {
652 /* Clear any previously stored errors */
653 filemap_check_errors(mapping);
654 }
655 } else {
656 err = filemap_check_errors(mapping);
657 }
658 return err;
659}
660EXPORT_SYMBOL(filemap_write_and_wait);
661
662/**
663 * filemap_write_and_wait_range - write out & wait on a file range
664 * @mapping: the address_space for the pages
665 * @lstart: offset in bytes where the range starts
666 * @lend: offset in bytes where the range ends (inclusive)
667 *
668 * Write out and wait upon file offsets lstart->lend, inclusive.
669 *
670 * Note that @lend is inclusive (describes the last byte to be written) so
671 * that this function can be used to write to the very end-of-file (end = -1).
672 *
673 * Return: error status of the address space.
674 */
675int filemap_write_and_wait_range(struct address_space *mapping,
676 loff_t lstart, loff_t lend)
677{
678 int err = 0;
679
680 if (mapping_needs_writeback(mapping)) {
681 err = __filemap_fdatawrite_range(mapping, lstart, lend,
682 WB_SYNC_ALL);
683 /* See comment of filemap_write_and_wait() */
684 if (err != -EIO) {
685 int err2 = filemap_fdatawait_range(mapping,
686 lstart, lend);
687 if (!err)
688 err = err2;
689 } else {
690 /* Clear any previously stored errors */
691 filemap_check_errors(mapping);
692 }
693 } else {
694 err = filemap_check_errors(mapping);
695 }
696 return err;
697}
698EXPORT_SYMBOL(filemap_write_and_wait_range);
699
700void __filemap_set_wb_err(struct address_space *mapping, int err)
701{
702 errseq_t eseq = errseq_set(&mapping->wb_err, err);
703
704 trace_filemap_set_wb_err(mapping, eseq);
705}
706EXPORT_SYMBOL(__filemap_set_wb_err);
707
708/**
709 * file_check_and_advance_wb_err - report wb error (if any) that was previously
710 * and advance wb_err to current one
711 * @file: struct file on which the error is being reported
712 *
713 * When userland calls fsync (or something like nfsd does the equivalent), we
714 * want to report any writeback errors that occurred since the last fsync (or
715 * since the file was opened if there haven't been any).
716 *
717 * Grab the wb_err from the mapping. If it matches what we have in the file,
718 * then just quickly return 0. The file is all caught up.
719 *
720 * If it doesn't match, then take the mapping value, set the "seen" flag in
721 * it and try to swap it into place. If it works, or another task beat us
722 * to it with the new value, then update the f_wb_err and return the error
723 * portion. The error at this point must be reported via proper channels
724 * (a'la fsync, or NFS COMMIT operation, etc.).
725 *
726 * While we handle mapping->wb_err with atomic operations, the f_wb_err
727 * value is protected by the f_lock since we must ensure that it reflects
728 * the latest value swapped in for this file descriptor.
729 *
730 * Return: %0 on success, negative error code otherwise.
731 */
732int file_check_and_advance_wb_err(struct file *file)
733{
734 int err = 0;
735 errseq_t old = READ_ONCE(file->f_wb_err);
736 struct address_space *mapping = file->f_mapping;
737
738 /* Locklessly handle the common case where nothing has changed */
739 if (errseq_check(&mapping->wb_err, old)) {
740 /* Something changed, must use slow path */
741 spin_lock(&file->f_lock);
742 old = file->f_wb_err;
743 err = errseq_check_and_advance(&mapping->wb_err,
744 &file->f_wb_err);
745 trace_file_check_and_advance_wb_err(file, old);
746 spin_unlock(&file->f_lock);
747 }
748
749 /*
750 * We're mostly using this function as a drop in replacement for
751 * filemap_check_errors. Clear AS_EIO/AS_ENOSPC to emulate the effect
752 * that the legacy code would have had on these flags.
753 */
754 clear_bit(AS_EIO, &mapping->flags);
755 clear_bit(AS_ENOSPC, &mapping->flags);
756 return err;
757}
758EXPORT_SYMBOL(file_check_and_advance_wb_err);
759
760/**
761 * file_write_and_wait_range - write out & wait on a file range
762 * @file: file pointing to address_space with pages
763 * @lstart: offset in bytes where the range starts
764 * @lend: offset in bytes where the range ends (inclusive)
765 *
766 * Write out and wait upon file offsets lstart->lend, inclusive.
767 *
768 * Note that @lend is inclusive (describes the last byte to be written) so
769 * that this function can be used to write to the very end-of-file (end = -1).
770 *
771 * After writing out and waiting on the data, we check and advance the
772 * f_wb_err cursor to the latest value, and return any errors detected there.
773 *
774 * Return: %0 on success, negative error code otherwise.
775 */
776int file_write_and_wait_range(struct file *file, loff_t lstart, loff_t lend)
777{
778 int err = 0, err2;
779 struct address_space *mapping = file->f_mapping;
780
781 if (mapping_needs_writeback(mapping)) {
782 err = __filemap_fdatawrite_range(mapping, lstart, lend,
783 WB_SYNC_ALL);
784 /* See comment of filemap_write_and_wait() */
785 if (err != -EIO)
786 __filemap_fdatawait_range(mapping, lstart, lend);
787 }
788 err2 = file_check_and_advance_wb_err(file);
789 if (!err)
790 err = err2;
791 return err;
792}
793EXPORT_SYMBOL(file_write_and_wait_range);
794
795/**
796 * replace_page_cache_page - replace a pagecache page with a new one
797 * @old: page to be replaced
798 * @new: page to replace with
799 * @gfp_mask: allocation mode
800 *
801 * This function replaces a page in the pagecache with a new one. On
802 * success it acquires the pagecache reference for the new page and
803 * drops it for the old page. Both the old and new pages must be
804 * locked. This function does not add the new page to the LRU, the
805 * caller must do that.
806 *
807 * The remove + add is atomic. This function cannot fail.
808 *
809 * Return: %0
810 */
811int replace_page_cache_page(struct page *old, struct page *new, gfp_t gfp_mask)
812{
813 struct address_space *mapping = old->mapping;
814 void (*freepage)(struct page *) = mapping->a_ops->freepage;
815 pgoff_t offset = old->index;
816 XA_STATE(xas, &mapping->i_pages, offset);
817 unsigned long flags;
818
819 VM_BUG_ON_PAGE(!PageLocked(old), old);
820 VM_BUG_ON_PAGE(!PageLocked(new), new);
821 VM_BUG_ON_PAGE(new->mapping, new);
822
823 get_page(new);
824 new->mapping = mapping;
825 new->index = offset;
826
827 xas_lock_irqsave(&xas, flags);
828 xas_store(&xas, new);
829
830 old->mapping = NULL;
831 /* hugetlb pages do not participate in page cache accounting. */
832 if (!PageHuge(old))
833 __dec_node_page_state(new, NR_FILE_PAGES);
834 if (!PageHuge(new))
835 __inc_node_page_state(new, NR_FILE_PAGES);
836 if (PageSwapBacked(old))
837 __dec_node_page_state(new, NR_SHMEM);
838 if (PageSwapBacked(new))
839 __inc_node_page_state(new, NR_SHMEM);
840 xas_unlock_irqrestore(&xas, flags);
841 mem_cgroup_migrate(old, new);
842 if (freepage)
843 freepage(old);
844 put_page(old);
845
846 return 0;
847}
848EXPORT_SYMBOL_GPL(replace_page_cache_page);
849
850static int __add_to_page_cache_locked(struct page *page,
851 struct address_space *mapping,
852 pgoff_t offset, gfp_t gfp_mask,
853 void **shadowp)
854{
855 XA_STATE(xas, &mapping->i_pages, offset);
856 int huge = PageHuge(page);
857 struct mem_cgroup *memcg;
858 int error;
859 void *old;
860
861 VM_BUG_ON_PAGE(!PageLocked(page), page);
862 VM_BUG_ON_PAGE(PageSwapBacked(page), page);
863 mapping_set_update(&xas, mapping);
864
865 if (!huge) {
866 error = mem_cgroup_try_charge(page, current->mm,
867 gfp_mask, &memcg, false);
868 if (error)
869 return error;
870 }
871
872 get_page(page);
873 page->mapping = mapping;
874 page->index = offset;
875
876 do {
877 xas_lock_irq(&xas);
878 old = xas_load(&xas);
879 if (old && !xa_is_value(old))
880 xas_set_err(&xas, -EEXIST);
881 xas_store(&xas, page);
882 if (xas_error(&xas))
883 goto unlock;
884
885 if (xa_is_value(old)) {
886 mapping->nrexceptional--;
887 if (shadowp)
888 *shadowp = old;
889 }
890 mapping->nrpages++;
891
892 /* hugetlb pages do not participate in page cache accounting */
893 if (!huge)
894 __inc_node_page_state(page, NR_FILE_PAGES);
895unlock:
896 xas_unlock_irq(&xas);
897 } while (xas_nomem(&xas, gfp_mask & GFP_RECLAIM_MASK));
898
899 if (xas_error(&xas))
900 goto error;
901
902 if (!huge)
903 mem_cgroup_commit_charge(page, memcg, false, false);
904 trace_mm_filemap_add_to_page_cache(page);
905 return 0;
906error:
907 page->mapping = NULL;
908 /* Leave page->index set: truncation relies upon it */
909 if (!huge)
910 mem_cgroup_cancel_charge(page, memcg, false);
911 put_page(page);
912 return xas_error(&xas);
913}
914ALLOW_ERROR_INJECTION(__add_to_page_cache_locked, ERRNO);
915
916/**
917 * add_to_page_cache_locked - add a locked page to the pagecache
918 * @page: page to add
919 * @mapping: the page's address_space
920 * @offset: page index
921 * @gfp_mask: page allocation mode
922 *
923 * This function is used to add a page to the pagecache. It must be locked.
924 * This function does not add the page to the LRU. The caller must do that.
925 *
926 * Return: %0 on success, negative error code otherwise.
927 */
928int add_to_page_cache_locked(struct page *page, struct address_space *mapping,
929 pgoff_t offset, gfp_t gfp_mask)
930{
931 return __add_to_page_cache_locked(page, mapping, offset,
932 gfp_mask, NULL);
933}
934EXPORT_SYMBOL(add_to_page_cache_locked);
935
936int add_to_page_cache_lru(struct page *page, struct address_space *mapping,
937 pgoff_t offset, gfp_t gfp_mask)
938{
939 void *shadow = NULL;
940 int ret;
941
942 __SetPageLocked(page);
943 ret = __add_to_page_cache_locked(page, mapping, offset,
944 gfp_mask, &shadow);
945 if (unlikely(ret))
946 __ClearPageLocked(page);
947 else {
948 /*
949 * The page might have been evicted from cache only
950 * recently, in which case it should be activated like
951 * any other repeatedly accessed page.
952 * The exception is pages getting rewritten; evicting other
953 * data from the working set, only to cache data that will
954 * get overwritten with something else, is a waste of memory.
955 */
956 WARN_ON_ONCE(PageActive(page));
957 if (!(gfp_mask & __GFP_WRITE) && shadow)
958 workingset_refault(page, shadow);
959 lru_cache_add(page);
960 }
961 return ret;
962}
963EXPORT_SYMBOL_GPL(add_to_page_cache_lru);
964
965#ifdef CONFIG_NUMA
966struct page *__page_cache_alloc(gfp_t gfp)
967{
968 int n;
969 struct page *page;
970
971 if (cpuset_do_page_mem_spread()) {
972 unsigned int cpuset_mems_cookie;
973 do {
974 cpuset_mems_cookie = read_mems_allowed_begin();
975 n = cpuset_mem_spread_node();
976 page = __alloc_pages_node(n, gfp, 0);
977 } while (!page && read_mems_allowed_retry(cpuset_mems_cookie));
978
979 return page;
980 }
981 return alloc_pages(gfp, 0);
982}
983EXPORT_SYMBOL(__page_cache_alloc);
984#endif
985
986/*
987 * In order to wait for pages to become available there must be
988 * waitqueues associated with pages. By using a hash table of
989 * waitqueues where the bucket discipline is to maintain all
990 * waiters on the same queue and wake all when any of the pages
991 * become available, and for the woken contexts to check to be
992 * sure the appropriate page became available, this saves space
993 * at a cost of "thundering herd" phenomena during rare hash
994 * collisions.
995 */
996#define PAGE_WAIT_TABLE_BITS 8
997#define PAGE_WAIT_TABLE_SIZE (1 << PAGE_WAIT_TABLE_BITS)
998static wait_queue_head_t page_wait_table[PAGE_WAIT_TABLE_SIZE] __cacheline_aligned;
999
1000static wait_queue_head_t *page_waitqueue(struct page *page)
1001{
1002 return &page_wait_table[hash_ptr(page, PAGE_WAIT_TABLE_BITS)];
1003}
1004
1005void __init pagecache_init(void)
1006{
1007 int i;
1008
1009 for (i = 0; i < PAGE_WAIT_TABLE_SIZE; i++)
1010 init_waitqueue_head(&page_wait_table[i]);
1011
1012 page_writeback_init();
1013}
1014
1015/* This has the same layout as wait_bit_key - see fs/cachefiles/rdwr.c */
1016struct wait_page_key {
1017 struct page *page;
1018 int bit_nr;
1019 int page_match;
1020};
1021
1022struct wait_page_queue {
1023 struct page *page;
1024 int bit_nr;
1025 wait_queue_entry_t wait;
1026};
1027
1028static int wake_page_function(wait_queue_entry_t *wait, unsigned mode, int sync, void *arg)
1029{
1030 struct wait_page_key *key = arg;
1031 struct wait_page_queue *wait_page
1032 = container_of(wait, struct wait_page_queue, wait);
1033
1034 if (wait_page->page != key->page)
1035 return 0;
1036 key->page_match = 1;
1037
1038 if (wait_page->bit_nr != key->bit_nr)
1039 return 0;
1040
1041 /*
1042 * Stop walking if it's locked.
1043 * Is this safe if put_and_wait_on_page_locked() is in use?
1044 * Yes: the waker must hold a reference to this page, and if PG_locked
1045 * has now already been set by another task, that task must also hold
1046 * a reference to the *same usage* of this page; so there is no need
1047 * to walk on to wake even the put_and_wait_on_page_locked() callers.
1048 */
1049 if (test_bit(key->bit_nr, &key->page->flags))
1050 return -1;
1051
1052 return autoremove_wake_function(wait, mode, sync, key);
1053}
1054
1055static void wake_up_page_bit(struct page *page, int bit_nr)
1056{
1057 wait_queue_head_t *q = page_waitqueue(page);
1058 struct wait_page_key key;
1059 unsigned long flags;
1060 wait_queue_entry_t bookmark;
1061
1062 key.page = page;
1063 key.bit_nr = bit_nr;
1064 key.page_match = 0;
1065
1066 bookmark.flags = 0;
1067 bookmark.private = NULL;
1068 bookmark.func = NULL;
1069 INIT_LIST_HEAD(&bookmark.entry);
1070
1071 spin_lock_irqsave(&q->lock, flags);
1072 __wake_up_locked_key_bookmark(q, TASK_NORMAL, &key, &bookmark);
1073
1074 while (bookmark.flags & WQ_FLAG_BOOKMARK) {
1075 /*
1076 * Take a breather from holding the lock,
1077 * allow pages that finish wake up asynchronously
1078 * to acquire the lock and remove themselves
1079 * from wait queue
1080 */
1081 spin_unlock_irqrestore(&q->lock, flags);
1082 cpu_relax();
1083 spin_lock_irqsave(&q->lock, flags);
1084 __wake_up_locked_key_bookmark(q, TASK_NORMAL, &key, &bookmark);
1085 }
1086
1087 /*
1088 * It is possible for other pages to have collided on the waitqueue
1089 * hash, so in that case check for a page match. That prevents a long-
1090 * term waiter
1091 *
1092 * It is still possible to miss a case here, when we woke page waiters
1093 * and removed them from the waitqueue, but there are still other
1094 * page waiters.
1095 */
1096 if (!waitqueue_active(q) || !key.page_match) {
1097 ClearPageWaiters(page);
1098 /*
1099 * It's possible to miss clearing Waiters here, when we woke
1100 * our page waiters, but the hashed waitqueue has waiters for
1101 * other pages on it.
1102 *
1103 * That's okay, it's a rare case. The next waker will clear it.
1104 */
1105 }
1106 spin_unlock_irqrestore(&q->lock, flags);
1107}
1108
1109static void wake_up_page(struct page *page, int bit)
1110{
1111 if (!PageWaiters(page))
1112 return;
1113 wake_up_page_bit(page, bit);
1114}
1115
1116/*
1117 * A choice of three behaviors for wait_on_page_bit_common():
1118 */
1119enum behavior {
1120 EXCLUSIVE, /* Hold ref to page and take the bit when woken, like
1121 * __lock_page() waiting on then setting PG_locked.
1122 */
1123 SHARED, /* Hold ref to page and check the bit when woken, like
1124 * wait_on_page_writeback() waiting on PG_writeback.
1125 */
1126 DROP, /* Drop ref to page before wait, no check when woken,
1127 * like put_and_wait_on_page_locked() on PG_locked.
1128 */
1129};
1130
1131static inline int wait_on_page_bit_common(wait_queue_head_t *q,
1132 struct page *page, int bit_nr, int state, enum behavior behavior)
1133{
1134 struct wait_page_queue wait_page;
1135 wait_queue_entry_t *wait = &wait_page.wait;
1136 bool bit_is_set;
1137 bool thrashing = false;
1138 bool delayacct = false;
1139 unsigned long pflags;
1140 int ret = 0;
1141
1142 if (bit_nr == PG_locked &&
1143 !PageUptodate(page) && PageWorkingset(page)) {
1144 if (!PageSwapBacked(page)) {
1145 delayacct_thrashing_start();
1146 delayacct = true;
1147 }
1148 psi_memstall_enter(&pflags);
1149 thrashing = true;
1150 }
1151
1152 init_wait(wait);
1153 wait->flags = behavior == EXCLUSIVE ? WQ_FLAG_EXCLUSIVE : 0;
1154 wait->func = wake_page_function;
1155 wait_page.page = page;
1156 wait_page.bit_nr = bit_nr;
1157
1158 for (;;) {
1159 spin_lock_irq(&q->lock);
1160
1161 if (likely(list_empty(&wait->entry))) {
1162 __add_wait_queue_entry_tail(q, wait);
1163 SetPageWaiters(page);
1164 }
1165
1166 set_current_state(state);
1167
1168 spin_unlock_irq(&q->lock);
1169
1170 bit_is_set = test_bit(bit_nr, &page->flags);
1171 if (behavior == DROP)
1172 put_page(page);
1173
1174 if (likely(bit_is_set))
1175 io_schedule();
1176
1177 if (behavior == EXCLUSIVE) {
1178 if (!test_and_set_bit_lock(bit_nr, &page->flags))
1179 break;
1180 } else if (behavior == SHARED) {
1181 if (!test_bit(bit_nr, &page->flags))
1182 break;
1183 }
1184
1185 if (signal_pending_state(state, current)) {
1186 ret = -EINTR;
1187 break;
1188 }
1189
1190 if (behavior == DROP) {
1191 /*
1192 * We can no longer safely access page->flags:
1193 * even if CONFIG_MEMORY_HOTREMOVE is not enabled,
1194 * there is a risk of waiting forever on a page reused
1195 * for something that keeps it locked indefinitely.
1196 * But best check for -EINTR above before breaking.
1197 */
1198 break;
1199 }
1200 }
1201
1202 finish_wait(q, wait);
1203
1204 if (thrashing) {
1205 if (delayacct)
1206 delayacct_thrashing_end();
1207 psi_memstall_leave(&pflags);
1208 }
1209
1210 /*
1211 * A signal could leave PageWaiters set. Clearing it here if
1212 * !waitqueue_active would be possible (by open-coding finish_wait),
1213 * but still fail to catch it in the case of wait hash collision. We
1214 * already can fail to clear wait hash collision cases, so don't
1215 * bother with signals either.
1216 */
1217
1218 return ret;
1219}
1220
1221void wait_on_page_bit(struct page *page, int bit_nr)
1222{
1223 wait_queue_head_t *q = page_waitqueue(page);
1224 wait_on_page_bit_common(q, page, bit_nr, TASK_UNINTERRUPTIBLE, SHARED);
1225}
1226EXPORT_SYMBOL(wait_on_page_bit);
1227
1228int wait_on_page_bit_killable(struct page *page, int bit_nr)
1229{
1230 wait_queue_head_t *q = page_waitqueue(page);
1231 return wait_on_page_bit_common(q, page, bit_nr, TASK_KILLABLE, SHARED);
1232}
1233EXPORT_SYMBOL(wait_on_page_bit_killable);
1234
1235/**
1236 * put_and_wait_on_page_locked - Drop a reference and wait for it to be unlocked
1237 * @page: The page to wait for.
1238 *
1239 * The caller should hold a reference on @page. They expect the page to
1240 * become unlocked relatively soon, but do not wish to hold up migration
1241 * (for example) by holding the reference while waiting for the page to
1242 * come unlocked. After this function returns, the caller should not
1243 * dereference @page.
1244 */
1245void put_and_wait_on_page_locked(struct page *page)
1246{
1247 wait_queue_head_t *q;
1248
1249 page = compound_head(page);
1250 q = page_waitqueue(page);
1251 wait_on_page_bit_common(q, page, PG_locked, TASK_UNINTERRUPTIBLE, DROP);
1252}
1253
1254/**
1255 * add_page_wait_queue - Add an arbitrary waiter to a page's wait queue
1256 * @page: Page defining the wait queue of interest
1257 * @waiter: Waiter to add to the queue
1258 *
1259 * Add an arbitrary @waiter to the wait queue for the nominated @page.
1260 */
1261void add_page_wait_queue(struct page *page, wait_queue_entry_t *waiter)
1262{
1263 wait_queue_head_t *q = page_waitqueue(page);
1264 unsigned long flags;
1265
1266 spin_lock_irqsave(&q->lock, flags);
1267 __add_wait_queue_entry_tail(q, waiter);
1268 SetPageWaiters(page);
1269 spin_unlock_irqrestore(&q->lock, flags);
1270}
1271EXPORT_SYMBOL_GPL(add_page_wait_queue);
1272
1273#ifndef clear_bit_unlock_is_negative_byte
1274
1275/*
1276 * PG_waiters is the high bit in the same byte as PG_lock.
1277 *
1278 * On x86 (and on many other architectures), we can clear PG_lock and
1279 * test the sign bit at the same time. But if the architecture does
1280 * not support that special operation, we just do this all by hand
1281 * instead.
1282 *
1283 * The read of PG_waiters has to be after (or concurrently with) PG_locked
1284 * being cleared, but a memory barrier should be unneccssary since it is
1285 * in the same byte as PG_locked.
1286 */
1287static inline bool clear_bit_unlock_is_negative_byte(long nr, volatile void *mem)
1288{
1289 clear_bit_unlock(nr, mem);
1290 /* smp_mb__after_atomic(); */
1291 return test_bit(PG_waiters, mem);
1292}
1293
1294#endif
1295
1296/**
1297 * unlock_page - unlock a locked page
1298 * @page: the page
1299 *
1300 * Unlocks the page and wakes up sleepers in ___wait_on_page_locked().
1301 * Also wakes sleepers in wait_on_page_writeback() because the wakeup
1302 * mechanism between PageLocked pages and PageWriteback pages is shared.
1303 * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep.
1304 *
1305 * Note that this depends on PG_waiters being the sign bit in the byte
1306 * that contains PG_locked - thus the BUILD_BUG_ON(). That allows us to
1307 * clear the PG_locked bit and test PG_waiters at the same time fairly
1308 * portably (architectures that do LL/SC can test any bit, while x86 can
1309 * test the sign bit).
1310 */
1311void unlock_page(struct page *page)
1312{
1313 BUILD_BUG_ON(PG_waiters != 7);
1314 page = compound_head(page);
1315 VM_BUG_ON_PAGE(!PageLocked(page), page);
1316 if (clear_bit_unlock_is_negative_byte(PG_locked, &page->flags))
1317 wake_up_page_bit(page, PG_locked);
1318}
1319EXPORT_SYMBOL(unlock_page);
1320
1321/**
1322 * end_page_writeback - end writeback against a page
1323 * @page: the page
1324 */
1325void end_page_writeback(struct page *page)
1326{
1327 /*
1328 * TestClearPageReclaim could be used here but it is an atomic
1329 * operation and overkill in this particular case. Failing to
1330 * shuffle a page marked for immediate reclaim is too mild to
1331 * justify taking an atomic operation penalty at the end of
1332 * ever page writeback.
1333 */
1334 if (PageReclaim(page)) {
1335 ClearPageReclaim(page);
1336 rotate_reclaimable_page(page);
1337 }
1338
1339 if (!test_clear_page_writeback(page))
1340 BUG();
1341
1342 smp_mb__after_atomic();
1343 wake_up_page(page, PG_writeback);
1344}
1345EXPORT_SYMBOL(end_page_writeback);
1346
1347/*
1348 * After completing I/O on a page, call this routine to update the page
1349 * flags appropriately
1350 */
1351void page_endio(struct page *page, bool is_write, int err)
1352{
1353 if (!is_write) {
1354 if (!err) {
1355 SetPageUptodate(page);
1356 } else {
1357 ClearPageUptodate(page);
1358 SetPageError(page);
1359 }
1360 unlock_page(page);
1361 } else {
1362 if (err) {
1363 struct address_space *mapping;
1364
1365 SetPageError(page);
1366 mapping = page_mapping(page);
1367 if (mapping)
1368 mapping_set_error(mapping, err);
1369 }
1370 end_page_writeback(page);
1371 }
1372}
1373EXPORT_SYMBOL_GPL(page_endio);
1374
1375/**
1376 * __lock_page - get a lock on the page, assuming we need to sleep to get it
1377 * @__page: the page to lock
1378 */
1379void __lock_page(struct page *__page)
1380{
1381 struct page *page = compound_head(__page);
1382 wait_queue_head_t *q = page_waitqueue(page);
1383 wait_on_page_bit_common(q, page, PG_locked, TASK_UNINTERRUPTIBLE,
1384 EXCLUSIVE);
1385}
1386EXPORT_SYMBOL(__lock_page);
1387
1388int __lock_page_killable(struct page *__page)
1389{
1390 struct page *page = compound_head(__page);
1391 wait_queue_head_t *q = page_waitqueue(page);
1392 return wait_on_page_bit_common(q, page, PG_locked, TASK_KILLABLE,
1393 EXCLUSIVE);
1394}
1395EXPORT_SYMBOL_GPL(__lock_page_killable);
1396
1397/*
1398 * Return values:
1399 * 1 - page is locked; mmap_sem is still held.
1400 * 0 - page is not locked.
1401 * mmap_sem has been released (up_read()), unless flags had both
1402 * FAULT_FLAG_ALLOW_RETRY and FAULT_FLAG_RETRY_NOWAIT set, in
1403 * which case mmap_sem is still held.
1404 *
1405 * If neither ALLOW_RETRY nor KILLABLE are set, will always return 1
1406 * with the page locked and the mmap_sem unperturbed.
1407 */
1408int __lock_page_or_retry(struct page *page, struct mm_struct *mm,
1409 unsigned int flags)
1410{
1411 if (flags & FAULT_FLAG_ALLOW_RETRY) {
1412 /*
1413 * CAUTION! In this case, mmap_sem is not released
1414 * even though return 0.
1415 */
1416 if (flags & FAULT_FLAG_RETRY_NOWAIT)
1417 return 0;
1418
1419 up_read(&mm->mmap_sem);
1420 if (flags & FAULT_FLAG_KILLABLE)
1421 wait_on_page_locked_killable(page);
1422 else
1423 wait_on_page_locked(page);
1424 return 0;
1425 } else {
1426 if (flags & FAULT_FLAG_KILLABLE) {
1427 int ret;
1428
1429 ret = __lock_page_killable(page);
1430 if (ret) {
1431 up_read(&mm->mmap_sem);
1432 return 0;
1433 }
1434 } else
1435 __lock_page(page);
1436 return 1;
1437 }
1438}
1439
1440/**
1441 * page_cache_next_miss() - Find the next gap in the page cache.
1442 * @mapping: Mapping.
1443 * @index: Index.
1444 * @max_scan: Maximum range to search.
1445 *
1446 * Search the range [index, min(index + max_scan - 1, ULONG_MAX)] for the
1447 * gap with the lowest index.
1448 *
1449 * This function may be called under the rcu_read_lock. However, this will
1450 * not atomically search a snapshot of the cache at a single point in time.
1451 * For example, if a gap is created at index 5, then subsequently a gap is
1452 * created at index 10, page_cache_next_miss covering both indices may
1453 * return 10 if called under the rcu_read_lock.
1454 *
1455 * Return: The index of the gap if found, otherwise an index outside the
1456 * range specified (in which case 'return - index >= max_scan' will be true).
1457 * In the rare case of index wrap-around, 0 will be returned.
1458 */
1459pgoff_t page_cache_next_miss(struct address_space *mapping,
1460 pgoff_t index, unsigned long max_scan)
1461{
1462 XA_STATE(xas, &mapping->i_pages, index);
1463
1464 while (max_scan--) {
1465 void *entry = xas_next(&xas);
1466 if (!entry || xa_is_value(entry))
1467 break;
1468 if (xas.xa_index == 0)
1469 break;
1470 }
1471
1472 return xas.xa_index;
1473}
1474EXPORT_SYMBOL(page_cache_next_miss);
1475
1476/**
1477 * page_cache_prev_miss() - Find the previous gap in the page cache.
1478 * @mapping: Mapping.
1479 * @index: Index.
1480 * @max_scan: Maximum range to search.
1481 *
1482 * Search the range [max(index - max_scan + 1, 0), index] for the
1483 * gap with the highest index.
1484 *
1485 * This function may be called under the rcu_read_lock. However, this will
1486 * not atomically search a snapshot of the cache at a single point in time.
1487 * For example, if a gap is created at index 10, then subsequently a gap is
1488 * created at index 5, page_cache_prev_miss() covering both indices may
1489 * return 5 if called under the rcu_read_lock.
1490 *
1491 * Return: The index of the gap if found, otherwise an index outside the
1492 * range specified (in which case 'index - return >= max_scan' will be true).
1493 * In the rare case of wrap-around, ULONG_MAX will be returned.
1494 */
1495pgoff_t page_cache_prev_miss(struct address_space *mapping,
1496 pgoff_t index, unsigned long max_scan)
1497{
1498 XA_STATE(xas, &mapping->i_pages, index);
1499
1500 while (max_scan--) {
1501 void *entry = xas_prev(&xas);
1502 if (!entry || xa_is_value(entry))
1503 break;
1504 if (xas.xa_index == ULONG_MAX)
1505 break;
1506 }
1507
1508 return xas.xa_index;
1509}
1510EXPORT_SYMBOL(page_cache_prev_miss);
1511
1512/**
1513 * find_get_entry - find and get a page cache entry
1514 * @mapping: the address_space to search
1515 * @offset: the page cache index
1516 *
1517 * Looks up the page cache slot at @mapping & @offset. If there is a
1518 * page cache page, it is returned with an increased refcount.
1519 *
1520 * If the slot holds a shadow entry of a previously evicted page, or a
1521 * swap entry from shmem/tmpfs, it is returned.
1522 *
1523 * Return: the found page or shadow entry, %NULL if nothing is found.
1524 */
1525struct page *find_get_entry(struct address_space *mapping, pgoff_t offset)
1526{
1527 XA_STATE(xas, &mapping->i_pages, offset);
1528 struct page *page;
1529
1530 rcu_read_lock();
1531repeat:
1532 xas_reset(&xas);
1533 page = xas_load(&xas);
1534 if (xas_retry(&xas, page))
1535 goto repeat;
1536 /*
1537 * A shadow entry of a recently evicted page, or a swap entry from
1538 * shmem/tmpfs. Return it without attempting to raise page count.
1539 */
1540 if (!page || xa_is_value(page))
1541 goto out;
1542
1543 if (!page_cache_get_speculative(page))
1544 goto repeat;
1545
1546 /*
1547 * Has the page moved or been split?
1548 * This is part of the lockless pagecache protocol. See
1549 * include/linux/pagemap.h for details.
1550 */
1551 if (unlikely(page != xas_reload(&xas))) {
1552 put_page(page);
1553 goto repeat;
1554 }
1555 page = find_subpage(page, offset);
1556out:
1557 rcu_read_unlock();
1558
1559 return page;
1560}
1561EXPORT_SYMBOL(find_get_entry);
1562
1563/**
1564 * find_lock_entry - locate, pin and lock a page cache entry
1565 * @mapping: the address_space to search
1566 * @offset: the page cache index
1567 *
1568 * Looks up the page cache slot at @mapping & @offset. If there is a
1569 * page cache page, it is returned locked and with an increased
1570 * refcount.
1571 *
1572 * If the slot holds a shadow entry of a previously evicted page, or a
1573 * swap entry from shmem/tmpfs, it is returned.
1574 *
1575 * find_lock_entry() may sleep.
1576 *
1577 * Return: the found page or shadow entry, %NULL if nothing is found.
1578 */
1579struct page *find_lock_entry(struct address_space *mapping, pgoff_t offset)
1580{
1581 struct page *page;
1582
1583repeat:
1584 page = find_get_entry(mapping, offset);
1585 if (page && !xa_is_value(page)) {
1586 lock_page(page);
1587 /* Has the page been truncated? */
1588 if (unlikely(page_mapping(page) != mapping)) {
1589 unlock_page(page);
1590 put_page(page);
1591 goto repeat;
1592 }
1593 VM_BUG_ON_PAGE(page_to_pgoff(page) != offset, page);
1594 }
1595 return page;
1596}
1597EXPORT_SYMBOL(find_lock_entry);
1598
1599/**
1600 * pagecache_get_page - find and get a page reference
1601 * @mapping: the address_space to search
1602 * @offset: the page index
1603 * @fgp_flags: PCG flags
1604 * @gfp_mask: gfp mask to use for the page cache data page allocation
1605 *
1606 * Looks up the page cache slot at @mapping & @offset.
1607 *
1608 * PCG flags modify how the page is returned.
1609 *
1610 * @fgp_flags can be:
1611 *
1612 * - FGP_ACCESSED: the page will be marked accessed
1613 * - FGP_LOCK: Page is return locked
1614 * - FGP_CREAT: If page is not present then a new page is allocated using
1615 * @gfp_mask and added to the page cache and the VM's LRU
1616 * list. The page is returned locked and with an increased
1617 * refcount.
1618 * - FGP_FOR_MMAP: Similar to FGP_CREAT, only we want to allow the caller to do
1619 * its own locking dance if the page is already in cache, or unlock the page
1620 * before returning if we had to add the page to pagecache.
1621 *
1622 * If FGP_LOCK or FGP_CREAT are specified then the function may sleep even
1623 * if the GFP flags specified for FGP_CREAT are atomic.
1624 *
1625 * If there is a page cache page, it is returned with an increased refcount.
1626 *
1627 * Return: the found page or %NULL otherwise.
1628 */
1629struct page *pagecache_get_page(struct address_space *mapping, pgoff_t offset,
1630 int fgp_flags, gfp_t gfp_mask)
1631{
1632 struct page *page;
1633
1634repeat:
1635 page = find_get_entry(mapping, offset);
1636 if (xa_is_value(page))
1637 page = NULL;
1638 if (!page)
1639 goto no_page;
1640
1641 if (fgp_flags & FGP_LOCK) {
1642 if (fgp_flags & FGP_NOWAIT) {
1643 if (!trylock_page(page)) {
1644 put_page(page);
1645 return NULL;
1646 }
1647 } else {
1648 lock_page(page);
1649 }
1650
1651 /* Has the page been truncated? */
1652 if (unlikely(compound_head(page)->mapping != mapping)) {
1653 unlock_page(page);
1654 put_page(page);
1655 goto repeat;
1656 }
1657 VM_BUG_ON_PAGE(page->index != offset, page);
1658 }
1659
1660 if (fgp_flags & FGP_ACCESSED)
1661 mark_page_accessed(page);
1662
1663no_page:
1664 if (!page && (fgp_flags & FGP_CREAT)) {
1665 int err;
1666 if ((fgp_flags & FGP_WRITE) && mapping_cap_account_dirty(mapping))
1667 gfp_mask |= __GFP_WRITE;
1668 if (fgp_flags & FGP_NOFS)
1669 gfp_mask &= ~__GFP_FS;
1670
1671 page = __page_cache_alloc(gfp_mask);
1672 if (!page)
1673 return NULL;
1674
1675 if (WARN_ON_ONCE(!(fgp_flags & (FGP_LOCK | FGP_FOR_MMAP))))
1676 fgp_flags |= FGP_LOCK;
1677
1678 /* Init accessed so avoid atomic mark_page_accessed later */
1679 if (fgp_flags & FGP_ACCESSED)
1680 __SetPageReferenced(page);
1681
1682 err = add_to_page_cache_lru(page, mapping, offset, gfp_mask);
1683 if (unlikely(err)) {
1684 put_page(page);
1685 page = NULL;
1686 if (err == -EEXIST)
1687 goto repeat;
1688 }
1689
1690 /*
1691 * add_to_page_cache_lru locks the page, and for mmap we expect
1692 * an unlocked page.
1693 */
1694 if (page && (fgp_flags & FGP_FOR_MMAP))
1695 unlock_page(page);
1696 }
1697
1698 return page;
1699}
1700EXPORT_SYMBOL(pagecache_get_page);
1701
1702/**
1703 * find_get_entries - gang pagecache lookup
1704 * @mapping: The address_space to search
1705 * @start: The starting page cache index
1706 * @nr_entries: The maximum number of entries
1707 * @entries: Where the resulting entries are placed
1708 * @indices: The cache indices corresponding to the entries in @entries
1709 *
1710 * find_get_entries() will search for and return a group of up to
1711 * @nr_entries entries in the mapping. The entries are placed at
1712 * @entries. find_get_entries() takes a reference against any actual
1713 * pages it returns.
1714 *
1715 * The search returns a group of mapping-contiguous page cache entries
1716 * with ascending indexes. There may be holes in the indices due to
1717 * not-present pages.
1718 *
1719 * Any shadow entries of evicted pages, or swap entries from
1720 * shmem/tmpfs, are included in the returned array.
1721 *
1722 * Return: the number of pages and shadow entries which were found.
1723 */
1724unsigned find_get_entries(struct address_space *mapping,
1725 pgoff_t start, unsigned int nr_entries,
1726 struct page **entries, pgoff_t *indices)
1727{
1728 XA_STATE(xas, &mapping->i_pages, start);
1729 struct page *page;
1730 unsigned int ret = 0;
1731
1732 if (!nr_entries)
1733 return 0;
1734
1735 rcu_read_lock();
1736 xas_for_each(&xas, page, ULONG_MAX) {
1737 if (xas_retry(&xas, page))
1738 continue;
1739 /*
1740 * A shadow entry of a recently evicted page, a swap
1741 * entry from shmem/tmpfs or a DAX entry. Return it
1742 * without attempting to raise page count.
1743 */
1744 if (xa_is_value(page))
1745 goto export;
1746
1747 if (!page_cache_get_speculative(page))
1748 goto retry;
1749
1750 /* Has the page moved or been split? */
1751 if (unlikely(page != xas_reload(&xas)))
1752 goto put_page;
1753 page = find_subpage(page, xas.xa_index);
1754
1755export:
1756 indices[ret] = xas.xa_index;
1757 entries[ret] = page;
1758 if (++ret == nr_entries)
1759 break;
1760 continue;
1761put_page:
1762 put_page(page);
1763retry:
1764 xas_reset(&xas);
1765 }
1766 rcu_read_unlock();
1767 return ret;
1768}
1769
1770/**
1771 * find_get_pages_range - gang pagecache lookup
1772 * @mapping: The address_space to search
1773 * @start: The starting page index
1774 * @end: The final page index (inclusive)
1775 * @nr_pages: The maximum number of pages
1776 * @pages: Where the resulting pages are placed
1777 *
1778 * find_get_pages_range() will search for and return a group of up to @nr_pages
1779 * pages in the mapping starting at index @start and up to index @end
1780 * (inclusive). The pages are placed at @pages. find_get_pages_range() takes
1781 * a reference against the returned pages.
1782 *
1783 * The search returns a group of mapping-contiguous pages with ascending
1784 * indexes. There may be holes in the indices due to not-present pages.
1785 * We also update @start to index the next page for the traversal.
1786 *
1787 * Return: the number of pages which were found. If this number is
1788 * smaller than @nr_pages, the end of specified range has been
1789 * reached.
1790 */
1791unsigned find_get_pages_range(struct address_space *mapping, pgoff_t *start,
1792 pgoff_t end, unsigned int nr_pages,
1793 struct page **pages)
1794{
1795 XA_STATE(xas, &mapping->i_pages, *start);
1796 struct page *page;
1797 unsigned ret = 0;
1798
1799 if (unlikely(!nr_pages))
1800 return 0;
1801
1802 rcu_read_lock();
1803 xas_for_each(&xas, page, end) {
1804 if (xas_retry(&xas, page))
1805 continue;
1806 /* Skip over shadow, swap and DAX entries */
1807 if (xa_is_value(page))
1808 continue;
1809
1810 if (!page_cache_get_speculative(page))
1811 goto retry;
1812
1813 /* Has the page moved or been split? */
1814 if (unlikely(page != xas_reload(&xas)))
1815 goto put_page;
1816
1817 pages[ret] = find_subpage(page, xas.xa_index);
1818 if (++ret == nr_pages) {
1819 *start = xas.xa_index + 1;
1820 goto out;
1821 }
1822 continue;
1823put_page:
1824 put_page(page);
1825retry:
1826 xas_reset(&xas);
1827 }
1828
1829 /*
1830 * We come here when there is no page beyond @end. We take care to not
1831 * overflow the index @start as it confuses some of the callers. This
1832 * breaks the iteration when there is a page at index -1 but that is
1833 * already broken anyway.
1834 */
1835 if (end == (pgoff_t)-1)
1836 *start = (pgoff_t)-1;
1837 else
1838 *start = end + 1;
1839out:
1840 rcu_read_unlock();
1841
1842 return ret;
1843}
1844
1845/**
1846 * find_get_pages_contig - gang contiguous pagecache lookup
1847 * @mapping: The address_space to search
1848 * @index: The starting page index
1849 * @nr_pages: The maximum number of pages
1850 * @pages: Where the resulting pages are placed
1851 *
1852 * find_get_pages_contig() works exactly like find_get_pages(), except
1853 * that the returned number of pages are guaranteed to be contiguous.
1854 *
1855 * Return: the number of pages which were found.
1856 */
1857unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index,
1858 unsigned int nr_pages, struct page **pages)
1859{
1860 XA_STATE(xas, &mapping->i_pages, index);
1861 struct page *page;
1862 unsigned int ret = 0;
1863
1864 if (unlikely(!nr_pages))
1865 return 0;
1866
1867 rcu_read_lock();
1868 for (page = xas_load(&xas); page; page = xas_next(&xas)) {
1869 if (xas_retry(&xas, page))
1870 continue;
1871 /*
1872 * If the entry has been swapped out, we can stop looking.
1873 * No current caller is looking for DAX entries.
1874 */
1875 if (xa_is_value(page))
1876 break;
1877
1878 if (!page_cache_get_speculative(page))
1879 goto retry;
1880
1881 /* Has the page moved or been split? */
1882 if (unlikely(page != xas_reload(&xas)))
1883 goto put_page;
1884
1885 pages[ret] = find_subpage(page, xas.xa_index);
1886 if (++ret == nr_pages)
1887 break;
1888 continue;
1889put_page:
1890 put_page(page);
1891retry:
1892 xas_reset(&xas);
1893 }
1894 rcu_read_unlock();
1895 return ret;
1896}
1897EXPORT_SYMBOL(find_get_pages_contig);
1898
1899/**
1900 * find_get_pages_range_tag - find and return pages in given range matching @tag
1901 * @mapping: the address_space to search
1902 * @index: the starting page index
1903 * @end: The final page index (inclusive)
1904 * @tag: the tag index
1905 * @nr_pages: the maximum number of pages
1906 * @pages: where the resulting pages are placed
1907 *
1908 * Like find_get_pages, except we only return pages which are tagged with
1909 * @tag. We update @index to index the next page for the traversal.
1910 *
1911 * Return: the number of pages which were found.
1912 */
1913unsigned find_get_pages_range_tag(struct address_space *mapping, pgoff_t *index,
1914 pgoff_t end, xa_mark_t tag, unsigned int nr_pages,
1915 struct page **pages)
1916{
1917 XA_STATE(xas, &mapping->i_pages, *index);
1918 struct page *page;
1919 unsigned ret = 0;
1920
1921 if (unlikely(!nr_pages))
1922 return 0;
1923
1924 rcu_read_lock();
1925 xas_for_each_marked(&xas, page, end, tag) {
1926 if (xas_retry(&xas, page))
1927 continue;
1928 /*
1929 * Shadow entries should never be tagged, but this iteration
1930 * is lockless so there is a window for page reclaim to evict
1931 * a page we saw tagged. Skip over it.
1932 */
1933 if (xa_is_value(page))
1934 continue;
1935
1936 if (!page_cache_get_speculative(page))
1937 goto retry;
1938
1939 /* Has the page moved or been split? */
1940 if (unlikely(page != xas_reload(&xas)))
1941 goto put_page;
1942
1943 pages[ret] = find_subpage(page, xas.xa_index);
1944 if (++ret == nr_pages) {
1945 *index = xas.xa_index + 1;
1946 goto out;
1947 }
1948 continue;
1949put_page:
1950 put_page(page);
1951retry:
1952 xas_reset(&xas);
1953 }
1954
1955 /*
1956 * We come here when we got to @end. We take care to not overflow the
1957 * index @index as it confuses some of the callers. This breaks the
1958 * iteration when there is a page at index -1 but that is already
1959 * broken anyway.
1960 */
1961 if (end == (pgoff_t)-1)
1962 *index = (pgoff_t)-1;
1963 else
1964 *index = end + 1;
1965out:
1966 rcu_read_unlock();
1967
1968 return ret;
1969}
1970EXPORT_SYMBOL(find_get_pages_range_tag);
1971
1972/*
1973 * CD/DVDs are error prone. When a medium error occurs, the driver may fail
1974 * a _large_ part of the i/o request. Imagine the worst scenario:
1975 *
1976 * ---R__________________________________________B__________
1977 * ^ reading here ^ bad block(assume 4k)
1978 *
1979 * read(R) => miss => readahead(R...B) => media error => frustrating retries
1980 * => failing the whole request => read(R) => read(R+1) =>
1981 * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
1982 * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
1983 * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
1984 *
1985 * It is going insane. Fix it by quickly scaling down the readahead size.
1986 */
1987static void shrink_readahead_size_eio(struct file *filp,
1988 struct file_ra_state *ra)
1989{
1990 ra->ra_pages /= 4;
1991}
1992
1993/**
1994 * generic_file_buffered_read - generic file read routine
1995 * @iocb: the iocb to read
1996 * @iter: data destination
1997 * @written: already copied
1998 *
1999 * This is a generic file read routine, and uses the
2000 * mapping->a_ops->readpage() function for the actual low-level stuff.
2001 *
2002 * This is really ugly. But the goto's actually try to clarify some
2003 * of the logic when it comes to error handling etc.
2004 *
2005 * Return:
2006 * * total number of bytes copied, including those the were already @written
2007 * * negative error code if nothing was copied
2008 */
2009static ssize_t generic_file_buffered_read(struct kiocb *iocb,
2010 struct iov_iter *iter, ssize_t written)
2011{
2012 struct file *filp = iocb->ki_filp;
2013 struct address_space *mapping = filp->f_mapping;
2014 struct inode *inode = mapping->host;
2015 struct file_ra_state *ra = &filp->f_ra;
2016 loff_t *ppos = &iocb->ki_pos;
2017 pgoff_t index;
2018 pgoff_t last_index;
2019 pgoff_t prev_index;
2020 unsigned long offset; /* offset into pagecache page */
2021 unsigned int prev_offset;
2022 int error = 0;
2023
2024 if (unlikely(*ppos >= inode->i_sb->s_maxbytes))
2025 return 0;
2026 iov_iter_truncate(iter, inode->i_sb->s_maxbytes);
2027
2028 index = *ppos >> PAGE_SHIFT;
2029 prev_index = ra->prev_pos >> PAGE_SHIFT;
2030 prev_offset = ra->prev_pos & (PAGE_SIZE-1);
2031 last_index = (*ppos + iter->count + PAGE_SIZE-1) >> PAGE_SHIFT;
2032 offset = *ppos & ~PAGE_MASK;
2033
2034 for (;;) {
2035 struct page *page;
2036 pgoff_t end_index;
2037 loff_t isize;
2038 unsigned long nr, ret;
2039
2040 cond_resched();
2041find_page:
2042 if (fatal_signal_pending(current)) {
2043 error = -EINTR;
2044 goto out;
2045 }
2046
2047 page = find_get_page(mapping, index);
2048 if (!page) {
2049 if (iocb->ki_flags & IOCB_NOWAIT)
2050 goto would_block;
2051 page_cache_sync_readahead(mapping,
2052 ra, filp,
2053 index, last_index - index);
2054 page = find_get_page(mapping, index);
2055 if (unlikely(page == NULL))
2056 goto no_cached_page;
2057 }
2058 if (PageReadahead(page)) {
2059 page_cache_async_readahead(mapping,
2060 ra, filp, page,
2061 index, last_index - index);
2062 }
2063 if (!PageUptodate(page)) {
2064 if (iocb->ki_flags & IOCB_NOWAIT) {
2065 put_page(page);
2066 goto would_block;
2067 }
2068
2069 /*
2070 * See comment in do_read_cache_page on why
2071 * wait_on_page_locked is used to avoid unnecessarily
2072 * serialisations and why it's safe.
2073 */
2074 error = wait_on_page_locked_killable(page);
2075 if (unlikely(error))
2076 goto readpage_error;
2077 if (PageUptodate(page))
2078 goto page_ok;
2079
2080 if (inode->i_blkbits == PAGE_SHIFT ||
2081 !mapping->a_ops->is_partially_uptodate)
2082 goto page_not_up_to_date;
2083 /* pipes can't handle partially uptodate pages */
2084 if (unlikely(iov_iter_is_pipe(iter)))
2085 goto page_not_up_to_date;
2086 if (!trylock_page(page))
2087 goto page_not_up_to_date;
2088 /* Did it get truncated before we got the lock? */
2089 if (!page->mapping)
2090 goto page_not_up_to_date_locked;
2091 if (!mapping->a_ops->is_partially_uptodate(page,
2092 offset, iter->count))
2093 goto page_not_up_to_date_locked;
2094 unlock_page(page);
2095 }
2096page_ok:
2097 /*
2098 * i_size must be checked after we know the page is Uptodate.
2099 *
2100 * Checking i_size after the check allows us to calculate
2101 * the correct value for "nr", which means the zero-filled
2102 * part of the page is not copied back to userspace (unless
2103 * another truncate extends the file - this is desired though).
2104 */
2105
2106 isize = i_size_read(inode);
2107 end_index = (isize - 1) >> PAGE_SHIFT;
2108 if (unlikely(!isize || index > end_index)) {
2109 put_page(page);
2110 goto out;
2111 }
2112
2113 /* nr is the maximum number of bytes to copy from this page */
2114 nr = PAGE_SIZE;
2115 if (index == end_index) {
2116 nr = ((isize - 1) & ~PAGE_MASK) + 1;
2117 if (nr <= offset) {
2118 put_page(page);
2119 goto out;
2120 }
2121 }
2122 nr = nr - offset;
2123
2124 /* If users can be writing to this page using arbitrary
2125 * virtual addresses, take care about potential aliasing
2126 * before reading the page on the kernel side.
2127 */
2128 if (mapping_writably_mapped(mapping))
2129 flush_dcache_page(page);
2130
2131 /*
2132 * When a sequential read accesses a page several times,
2133 * only mark it as accessed the first time.
2134 */
2135 if (prev_index != index || offset != prev_offset)
2136 mark_page_accessed(page);
2137 prev_index = index;
2138
2139 /*
2140 * Ok, we have the page, and it's up-to-date, so
2141 * now we can copy it to user space...
2142 */
2143
2144 ret = copy_page_to_iter(page, offset, nr, iter);
2145 offset += ret;
2146 index += offset >> PAGE_SHIFT;
2147 offset &= ~PAGE_MASK;
2148 prev_offset = offset;
2149
2150 put_page(page);
2151 written += ret;
2152 if (!iov_iter_count(iter))
2153 goto out;
2154 if (ret < nr) {
2155 error = -EFAULT;
2156 goto out;
2157 }
2158 continue;
2159
2160page_not_up_to_date:
2161 /* Get exclusive access to the page ... */
2162 error = lock_page_killable(page);
2163 if (unlikely(error))
2164 goto readpage_error;
2165
2166page_not_up_to_date_locked:
2167 /* Did it get truncated before we got the lock? */
2168 if (!page->mapping) {
2169 unlock_page(page);
2170 put_page(page);
2171 continue;
2172 }
2173
2174 /* Did somebody else fill it already? */
2175 if (PageUptodate(page)) {
2176 unlock_page(page);
2177 goto page_ok;
2178 }
2179
2180readpage:
2181 /*
2182 * A previous I/O error may have been due to temporary
2183 * failures, eg. multipath errors.
2184 * PG_error will be set again if readpage fails.
2185 */
2186 ClearPageError(page);
2187 /* Start the actual read. The read will unlock the page. */
2188 error = mapping->a_ops->readpage(filp, page);
2189
2190 if (unlikely(error)) {
2191 if (error == AOP_TRUNCATED_PAGE) {
2192 put_page(page);
2193 error = 0;
2194 goto find_page;
2195 }
2196 goto readpage_error;
2197 }
2198
2199 if (!PageUptodate(page)) {
2200 error = lock_page_killable(page);
2201 if (unlikely(error))
2202 goto readpage_error;
2203 if (!PageUptodate(page)) {
2204 if (page->mapping == NULL) {
2205 /*
2206 * invalidate_mapping_pages got it
2207 */
2208 unlock_page(page);
2209 put_page(page);
2210 goto find_page;
2211 }
2212 unlock_page(page);
2213 shrink_readahead_size_eio(filp, ra);
2214 error = -EIO;
2215 goto readpage_error;
2216 }
2217 unlock_page(page);
2218 }
2219
2220 goto page_ok;
2221
2222readpage_error:
2223 /* UHHUH! A synchronous read error occurred. Report it */
2224 put_page(page);
2225 goto out;
2226
2227no_cached_page:
2228 /*
2229 * Ok, it wasn't cached, so we need to create a new
2230 * page..
2231 */
2232 page = page_cache_alloc(mapping);
2233 if (!page) {
2234 error = -ENOMEM;
2235 goto out;
2236 }
2237 error = add_to_page_cache_lru(page, mapping, index,
2238 mapping_gfp_constraint(mapping, GFP_KERNEL));
2239 if (error) {
2240 put_page(page);
2241 if (error == -EEXIST) {
2242 error = 0;
2243 goto find_page;
2244 }
2245 goto out;
2246 }
2247 goto readpage;
2248 }
2249
2250would_block:
2251 error = -EAGAIN;
2252out:
2253 ra->prev_pos = prev_index;
2254 ra->prev_pos <<= PAGE_SHIFT;
2255 ra->prev_pos |= prev_offset;
2256
2257 *ppos = ((loff_t)index << PAGE_SHIFT) + offset;
2258 file_accessed(filp);
2259 return written ? written : error;
2260}
2261
2262/**
2263 * generic_file_read_iter - generic filesystem read routine
2264 * @iocb: kernel I/O control block
2265 * @iter: destination for the data read
2266 *
2267 * This is the "read_iter()" routine for all filesystems
2268 * that can use the page cache directly.
2269 * Return:
2270 * * number of bytes copied, even for partial reads
2271 * * negative error code if nothing was read
2272 */
2273ssize_t
2274generic_file_read_iter(struct kiocb *iocb, struct iov_iter *iter)
2275{
2276 size_t count = iov_iter_count(iter);
2277 ssize_t retval = 0;
2278
2279 if (!count)
2280 goto out; /* skip atime */
2281
2282 if (iocb->ki_flags & IOCB_DIRECT) {
2283 struct file *file = iocb->ki_filp;
2284 struct address_space *mapping = file->f_mapping;
2285 struct inode *inode = mapping->host;
2286 loff_t size;
2287
2288 size = i_size_read(inode);
2289 if (iocb->ki_flags & IOCB_NOWAIT) {
2290 if (filemap_range_has_page(mapping, iocb->ki_pos,
2291 iocb->ki_pos + count - 1))
2292 return -EAGAIN;
2293 } else {
2294 retval = filemap_write_and_wait_range(mapping,
2295 iocb->ki_pos,
2296 iocb->ki_pos + count - 1);
2297 if (retval < 0)
2298 goto out;
2299 }
2300
2301 file_accessed(file);
2302
2303 retval = mapping->a_ops->direct_IO(iocb, iter);
2304 if (retval >= 0) {
2305 iocb->ki_pos += retval;
2306 count -= retval;
2307 }
2308 iov_iter_revert(iter, count - iov_iter_count(iter));
2309
2310 /*
2311 * Btrfs can have a short DIO read if we encounter
2312 * compressed extents, so if there was an error, or if
2313 * we've already read everything we wanted to, or if
2314 * there was a short read because we hit EOF, go ahead
2315 * and return. Otherwise fallthrough to buffered io for
2316 * the rest of the read. Buffered reads will not work for
2317 * DAX files, so don't bother trying.
2318 */
2319 if (retval < 0 || !count || iocb->ki_pos >= size ||
2320 IS_DAX(inode))
2321 goto out;
2322 }
2323
2324 retval = generic_file_buffered_read(iocb, iter, retval);
2325out:
2326 return retval;
2327}
2328EXPORT_SYMBOL(generic_file_read_iter);
2329
2330#ifdef CONFIG_MMU
2331#define MMAP_LOTSAMISS (100)
2332static struct file *maybe_unlock_mmap_for_io(struct vm_fault *vmf,
2333 struct file *fpin)
2334{
2335 int flags = vmf->flags;
2336
2337 if (fpin)
2338 return fpin;
2339
2340 /*
2341 * FAULT_FLAG_RETRY_NOWAIT means we don't want to wait on page locks or
2342 * anything, so we only pin the file and drop the mmap_sem if only
2343 * FAULT_FLAG_ALLOW_RETRY is set.
2344 */
2345 if ((flags & (FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_RETRY_NOWAIT)) ==
2346 FAULT_FLAG_ALLOW_RETRY) {
2347 fpin = get_file(vmf->vma->vm_file);
2348 up_read(&vmf->vma->vm_mm->mmap_sem);
2349 }
2350 return fpin;
2351}
2352
2353/*
2354 * lock_page_maybe_drop_mmap - lock the page, possibly dropping the mmap_sem
2355 * @vmf - the vm_fault for this fault.
2356 * @page - the page to lock.
2357 * @fpin - the pointer to the file we may pin (or is already pinned).
2358 *
2359 * This works similar to lock_page_or_retry in that it can drop the mmap_sem.
2360 * It differs in that it actually returns the page locked if it returns 1 and 0
2361 * if it couldn't lock the page. If we did have to drop the mmap_sem then fpin
2362 * will point to the pinned file and needs to be fput()'ed at a later point.
2363 */
2364static int lock_page_maybe_drop_mmap(struct vm_fault *vmf, struct page *page,
2365 struct file **fpin)
2366{
2367 if (trylock_page(page))
2368 return 1;
2369
2370 /*
2371 * NOTE! This will make us return with VM_FAULT_RETRY, but with
2372 * the mmap_sem still held. That's how FAULT_FLAG_RETRY_NOWAIT
2373 * is supposed to work. We have way too many special cases..
2374 */
2375 if (vmf->flags & FAULT_FLAG_RETRY_NOWAIT)
2376 return 0;
2377
2378 *fpin = maybe_unlock_mmap_for_io(vmf, *fpin);
2379 if (vmf->flags & FAULT_FLAG_KILLABLE) {
2380 if (__lock_page_killable(page)) {
2381 /*
2382 * We didn't have the right flags to drop the mmap_sem,
2383 * but all fault_handlers only check for fatal signals
2384 * if we return VM_FAULT_RETRY, so we need to drop the
2385 * mmap_sem here and return 0 if we don't have a fpin.
2386 */
2387 if (*fpin == NULL)
2388 up_read(&vmf->vma->vm_mm->mmap_sem);
2389 return 0;
2390 }
2391 } else
2392 __lock_page(page);
2393 return 1;
2394}
2395
2396
2397/*
2398 * Synchronous readahead happens when we don't even find a page in the page
2399 * cache at all. We don't want to perform IO under the mmap sem, so if we have
2400 * to drop the mmap sem we return the file that was pinned in order for us to do
2401 * that. If we didn't pin a file then we return NULL. The file that is
2402 * returned needs to be fput()'ed when we're done with it.
2403 */
2404static struct file *do_sync_mmap_readahead(struct vm_fault *vmf)
2405{
2406 struct file *file = vmf->vma->vm_file;
2407 struct file_ra_state *ra = &file->f_ra;
2408 struct address_space *mapping = file->f_mapping;
2409 struct file *fpin = NULL;
2410 pgoff_t offset = vmf->pgoff;
2411
2412 /* If we don't want any read-ahead, don't bother */
2413 if (vmf->vma->vm_flags & VM_RAND_READ)
2414 return fpin;
2415 if (!ra->ra_pages)
2416 return fpin;
2417
2418 if (vmf->vma->vm_flags & VM_SEQ_READ) {
2419 fpin = maybe_unlock_mmap_for_io(vmf, fpin);
2420 page_cache_sync_readahead(mapping, ra, file, offset,
2421 ra->ra_pages);
2422 return fpin;
2423 }
2424
2425 /* Avoid banging the cache line if not needed */
2426 if (ra->mmap_miss < MMAP_LOTSAMISS * 10)
2427 ra->mmap_miss++;
2428
2429 /*
2430 * Do we miss much more than hit in this file? If so,
2431 * stop bothering with read-ahead. It will only hurt.
2432 */
2433 if (ra->mmap_miss > MMAP_LOTSAMISS)
2434 return fpin;
2435
2436 /*
2437 * mmap read-around
2438 */
2439 fpin = maybe_unlock_mmap_for_io(vmf, fpin);
2440 ra->start = max_t(long, 0, offset - ra->ra_pages / 2);
2441 ra->size = ra->ra_pages;
2442 ra->async_size = ra->ra_pages / 4;
2443 ra_submit(ra, mapping, file);
2444 return fpin;
2445}
2446
2447/*
2448 * Asynchronous readahead happens when we find the page and PG_readahead,
2449 * so we want to possibly extend the readahead further. We return the file that
2450 * was pinned if we have to drop the mmap_sem in order to do IO.
2451 */
2452static struct file *do_async_mmap_readahead(struct vm_fault *vmf,
2453 struct page *page)
2454{
2455 struct file *file = vmf->vma->vm_file;
2456 struct file_ra_state *ra = &file->f_ra;
2457 struct address_space *mapping = file->f_mapping;
2458 struct file *fpin = NULL;
2459 pgoff_t offset = vmf->pgoff;
2460
2461 /* If we don't want any read-ahead, don't bother */
2462 if (vmf->vma->vm_flags & VM_RAND_READ)
2463 return fpin;
2464 if (ra->mmap_miss > 0)
2465 ra->mmap_miss--;
2466 if (PageReadahead(page)) {
2467 fpin = maybe_unlock_mmap_for_io(vmf, fpin);
2468 page_cache_async_readahead(mapping, ra, file,
2469 page, offset, ra->ra_pages);
2470 }
2471 return fpin;
2472}
2473
2474/**
2475 * filemap_fault - read in file data for page fault handling
2476 * @vmf: struct vm_fault containing details of the fault
2477 *
2478 * filemap_fault() is invoked via the vma operations vector for a
2479 * mapped memory region to read in file data during a page fault.
2480 *
2481 * The goto's are kind of ugly, but this streamlines the normal case of having
2482 * it in the page cache, and handles the special cases reasonably without
2483 * having a lot of duplicated code.
2484 *
2485 * vma->vm_mm->mmap_sem must be held on entry.
2486 *
2487 * If our return value has VM_FAULT_RETRY set, it's because the mmap_sem
2488 * may be dropped before doing I/O or by lock_page_maybe_drop_mmap().
2489 *
2490 * If our return value does not have VM_FAULT_RETRY set, the mmap_sem
2491 * has not been released.
2492 *
2493 * We never return with VM_FAULT_RETRY and a bit from VM_FAULT_ERROR set.
2494 *
2495 * Return: bitwise-OR of %VM_FAULT_ codes.
2496 */
2497vm_fault_t filemap_fault(struct vm_fault *vmf)
2498{
2499 int error;
2500 struct file *file = vmf->vma->vm_file;
2501 struct file *fpin = NULL;
2502 struct address_space *mapping = file->f_mapping;
2503 struct file_ra_state *ra = &file->f_ra;
2504 struct inode *inode = mapping->host;
2505 pgoff_t offset = vmf->pgoff;
2506 pgoff_t max_off;
2507 struct page *page;
2508 vm_fault_t ret = 0;
2509
2510 max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
2511 if (unlikely(offset >= max_off))
2512 return VM_FAULT_SIGBUS;
2513
2514 /*
2515 * Do we have something in the page cache already?
2516 */
2517 page = find_get_page(mapping, offset);
2518 if (likely(page) && !(vmf->flags & FAULT_FLAG_TRIED)) {
2519 /*
2520 * We found the page, so try async readahead before
2521 * waiting for the lock.
2522 */
2523 fpin = do_async_mmap_readahead(vmf, page);
2524 } else if (!page) {
2525 /* No page in the page cache at all */
2526 count_vm_event(PGMAJFAULT);
2527 count_memcg_event_mm(vmf->vma->vm_mm, PGMAJFAULT);
2528 ret = VM_FAULT_MAJOR;
2529 fpin = do_sync_mmap_readahead(vmf);
2530retry_find:
2531 page = pagecache_get_page(mapping, offset,
2532 FGP_CREAT|FGP_FOR_MMAP,
2533 vmf->gfp_mask);
2534 if (!page) {
2535 if (fpin)
2536 goto out_retry;
2537 return vmf_error(-ENOMEM);
2538 }
2539 }
2540
2541 if (!lock_page_maybe_drop_mmap(vmf, page, &fpin))
2542 goto out_retry;
2543
2544 /* Did it get truncated? */
2545 if (unlikely(compound_head(page)->mapping != mapping)) {
2546 unlock_page(page);
2547 put_page(page);
2548 goto retry_find;
2549 }
2550 VM_BUG_ON_PAGE(page_to_pgoff(page) != offset, page);
2551
2552 /*
2553 * We have a locked page in the page cache, now we need to check
2554 * that it's up-to-date. If not, it is going to be due to an error.
2555 */
2556 if (unlikely(!PageUptodate(page)))
2557 goto page_not_uptodate;
2558
2559 /*
2560 * We've made it this far and we had to drop our mmap_sem, now is the
2561 * time to return to the upper layer and have it re-find the vma and
2562 * redo the fault.
2563 */
2564 if (fpin) {
2565 unlock_page(page);
2566 goto out_retry;
2567 }
2568
2569 /*
2570 * Found the page and have a reference on it.
2571 * We must recheck i_size under page lock.
2572 */
2573 max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
2574 if (unlikely(offset >= max_off)) {
2575 unlock_page(page);
2576 put_page(page);
2577 return VM_FAULT_SIGBUS;
2578 }
2579
2580 vmf->page = page;
2581 return ret | VM_FAULT_LOCKED;
2582
2583page_not_uptodate:
2584 /*
2585 * Umm, take care of errors if the page isn't up-to-date.
2586 * Try to re-read it _once_. We do this synchronously,
2587 * because there really aren't any performance issues here
2588 * and we need to check for errors.
2589 */
2590 ClearPageError(page);
2591 fpin = maybe_unlock_mmap_for_io(vmf, fpin);
2592 error = mapping->a_ops->readpage(file, page);
2593 if (!error) {
2594 wait_on_page_locked(page);
2595 if (!PageUptodate(page))
2596 error = -EIO;
2597 }
2598 if (fpin)
2599 goto out_retry;
2600 put_page(page);
2601
2602 if (!error || error == AOP_TRUNCATED_PAGE)
2603 goto retry_find;
2604
2605 /* Things didn't work out. Return zero to tell the mm layer so. */
2606 shrink_readahead_size_eio(file, ra);
2607 return VM_FAULT_SIGBUS;
2608
2609out_retry:
2610 /*
2611 * We dropped the mmap_sem, we need to return to the fault handler to
2612 * re-find the vma and come back and find our hopefully still populated
2613 * page.
2614 */
2615 if (page)
2616 put_page(page);
2617 if (fpin)
2618 fput(fpin);
2619 return ret | VM_FAULT_RETRY;
2620}
2621EXPORT_SYMBOL(filemap_fault);
2622
2623void filemap_map_pages(struct vm_fault *vmf,
2624 pgoff_t start_pgoff, pgoff_t end_pgoff)
2625{
2626 struct file *file = vmf->vma->vm_file;
2627 struct address_space *mapping = file->f_mapping;
2628 pgoff_t last_pgoff = start_pgoff;
2629 unsigned long max_idx;
2630 XA_STATE(xas, &mapping->i_pages, start_pgoff);
2631 struct page *page;
2632
2633 rcu_read_lock();
2634 xas_for_each(&xas, page, end_pgoff) {
2635 if (xas_retry(&xas, page))
2636 continue;
2637 if (xa_is_value(page))
2638 goto next;
2639
2640 /*
2641 * Check for a locked page first, as a speculative
2642 * reference may adversely influence page migration.
2643 */
2644 if (PageLocked(page))
2645 goto next;
2646 if (!page_cache_get_speculative(page))
2647 goto next;
2648
2649 /* Has the page moved or been split? */
2650 if (unlikely(page != xas_reload(&xas)))
2651 goto skip;
2652 page = find_subpage(page, xas.xa_index);
2653
2654 if (!PageUptodate(page) ||
2655 PageReadahead(page) ||
2656 PageHWPoison(page))
2657 goto skip;
2658 if (!trylock_page(page))
2659 goto skip;
2660
2661 if (page->mapping != mapping || !PageUptodate(page))
2662 goto unlock;
2663
2664 max_idx = DIV_ROUND_UP(i_size_read(mapping->host), PAGE_SIZE);
2665 if (page->index >= max_idx)
2666 goto unlock;
2667
2668 if (file->f_ra.mmap_miss > 0)
2669 file->f_ra.mmap_miss--;
2670
2671 vmf->address += (xas.xa_index - last_pgoff) << PAGE_SHIFT;
2672 if (vmf->pte)
2673 vmf->pte += xas.xa_index - last_pgoff;
2674 last_pgoff = xas.xa_index;
2675 if (alloc_set_pte(vmf, NULL, page))
2676 goto unlock;
2677 unlock_page(page);
2678 goto next;
2679unlock:
2680 unlock_page(page);
2681skip:
2682 put_page(page);
2683next:
2684 /* Huge page is mapped? No need to proceed. */
2685 if (pmd_trans_huge(*vmf->pmd))
2686 break;
2687 }
2688 rcu_read_unlock();
2689}
2690EXPORT_SYMBOL(filemap_map_pages);
2691
2692vm_fault_t filemap_page_mkwrite(struct vm_fault *vmf)
2693{
2694 struct page *page = vmf->page;
2695 struct inode *inode = file_inode(vmf->vma->vm_file);
2696 vm_fault_t ret = VM_FAULT_LOCKED;
2697
2698 sb_start_pagefault(inode->i_sb);
2699 file_update_time(vmf->vma->vm_file);
2700 lock_page(page);
2701 if (page->mapping != inode->i_mapping) {
2702 unlock_page(page);
2703 ret = VM_FAULT_NOPAGE;
2704 goto out;
2705 }
2706 /*
2707 * We mark the page dirty already here so that when freeze is in
2708 * progress, we are guaranteed that writeback during freezing will
2709 * see the dirty page and writeprotect it again.
2710 */
2711 set_page_dirty(page);
2712 wait_for_stable_page(page);
2713out:
2714 sb_end_pagefault(inode->i_sb);
2715 return ret;
2716}
2717
2718const struct vm_operations_struct generic_file_vm_ops = {
2719 .fault = filemap_fault,
2720 .map_pages = filemap_map_pages,
2721 .page_mkwrite = filemap_page_mkwrite,
2722};
2723
2724/* This is used for a general mmap of a disk file */
2725
2726int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
2727{
2728 struct address_space *mapping = file->f_mapping;
2729
2730 if (!mapping->a_ops->readpage)
2731 return -ENOEXEC;
2732 file_accessed(file);
2733 vma->vm_ops = &generic_file_vm_ops;
2734 return 0;
2735}
2736
2737/*
2738 * This is for filesystems which do not implement ->writepage.
2739 */
2740int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
2741{
2742 if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE))
2743 return -EINVAL;
2744 return generic_file_mmap(file, vma);
2745}
2746#else
2747vm_fault_t filemap_page_mkwrite(struct vm_fault *vmf)
2748{
2749 return VM_FAULT_SIGBUS;
2750}
2751int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
2752{
2753 return -ENOSYS;
2754}
2755int generic_file_readonly_mmap(struct file * file, struct vm_area_struct * vma)
2756{
2757 return -ENOSYS;
2758}
2759#endif /* CONFIG_MMU */
2760
2761EXPORT_SYMBOL(filemap_page_mkwrite);
2762EXPORT_SYMBOL(generic_file_mmap);
2763EXPORT_SYMBOL(generic_file_readonly_mmap);
2764
2765static struct page *wait_on_page_read(struct page *page)
2766{
2767 if (!IS_ERR(page)) {
2768 wait_on_page_locked(page);
2769 if (!PageUptodate(page)) {
2770 put_page(page);
2771 page = ERR_PTR(-EIO);
2772 }
2773 }
2774 return page;
2775}
2776
2777static struct page *do_read_cache_page(struct address_space *mapping,
2778 pgoff_t index,
2779 int (*filler)(void *, struct page *),
2780 void *data,
2781 gfp_t gfp)
2782{
2783 struct page *page;
2784 int err;
2785repeat:
2786 page = find_get_page(mapping, index);
2787 if (!page) {
2788 page = __page_cache_alloc(gfp);
2789 if (!page)
2790 return ERR_PTR(-ENOMEM);
2791 err = add_to_page_cache_lru(page, mapping, index, gfp);
2792 if (unlikely(err)) {
2793 put_page(page);
2794 if (err == -EEXIST)
2795 goto repeat;
2796 /* Presumably ENOMEM for xarray node */
2797 return ERR_PTR(err);
2798 }
2799
2800filler:
2801 if (filler)
2802 err = filler(data, page);
2803 else
2804 err = mapping->a_ops->readpage(data, page);
2805
2806 if (err < 0) {
2807 put_page(page);
2808 return ERR_PTR(err);
2809 }
2810
2811 page = wait_on_page_read(page);
2812 if (IS_ERR(page))
2813 return page;
2814 goto out;
2815 }
2816 if (PageUptodate(page))
2817 goto out;
2818
2819 /*
2820 * Page is not up to date and may be locked due one of the following
2821 * case a: Page is being filled and the page lock is held
2822 * case b: Read/write error clearing the page uptodate status
2823 * case c: Truncation in progress (page locked)
2824 * case d: Reclaim in progress
2825 *
2826 * Case a, the page will be up to date when the page is unlocked.
2827 * There is no need to serialise on the page lock here as the page
2828 * is pinned so the lock gives no additional protection. Even if the
2829 * the page is truncated, the data is still valid if PageUptodate as
2830 * it's a race vs truncate race.
2831 * Case b, the page will not be up to date
2832 * Case c, the page may be truncated but in itself, the data may still
2833 * be valid after IO completes as it's a read vs truncate race. The
2834 * operation must restart if the page is not uptodate on unlock but
2835 * otherwise serialising on page lock to stabilise the mapping gives
2836 * no additional guarantees to the caller as the page lock is
2837 * released before return.
2838 * Case d, similar to truncation. If reclaim holds the page lock, it
2839 * will be a race with remove_mapping that determines if the mapping
2840 * is valid on unlock but otherwise the data is valid and there is
2841 * no need to serialise with page lock.
2842 *
2843 * As the page lock gives no additional guarantee, we optimistically
2844 * wait on the page to be unlocked and check if it's up to date and
2845 * use the page if it is. Otherwise, the page lock is required to
2846 * distinguish between the different cases. The motivation is that we
2847 * avoid spurious serialisations and wakeups when multiple processes
2848 * wait on the same page for IO to complete.
2849 */
2850 wait_on_page_locked(page);
2851 if (PageUptodate(page))
2852 goto out;
2853
2854 /* Distinguish between all the cases under the safety of the lock */
2855 lock_page(page);
2856
2857 /* Case c or d, restart the operation */
2858 if (!page->mapping) {
2859 unlock_page(page);
2860 put_page(page);
2861 goto repeat;
2862 }
2863
2864 /* Someone else locked and filled the page in a very small window */
2865 if (PageUptodate(page)) {
2866 unlock_page(page);
2867 goto out;
2868 }
2869 goto filler;
2870
2871out:
2872 mark_page_accessed(page);
2873 return page;
2874}
2875
2876/**
2877 * read_cache_page - read into page cache, fill it if needed
2878 * @mapping: the page's address_space
2879 * @index: the page index
2880 * @filler: function to perform the read
2881 * @data: first arg to filler(data, page) function, often left as NULL
2882 *
2883 * Read into the page cache. If a page already exists, and PageUptodate() is
2884 * not set, try to fill the page and wait for it to become unlocked.
2885 *
2886 * If the page does not get brought uptodate, return -EIO.
2887 *
2888 * Return: up to date page on success, ERR_PTR() on failure.
2889 */
2890struct page *read_cache_page(struct address_space *mapping,
2891 pgoff_t index,
2892 int (*filler)(void *, struct page *),
2893 void *data)
2894{
2895 return do_read_cache_page(mapping, index, filler, data,
2896 mapping_gfp_mask(mapping));
2897}
2898EXPORT_SYMBOL(read_cache_page);
2899
2900/**
2901 * read_cache_page_gfp - read into page cache, using specified page allocation flags.
2902 * @mapping: the page's address_space
2903 * @index: the page index
2904 * @gfp: the page allocator flags to use if allocating
2905 *
2906 * This is the same as "read_mapping_page(mapping, index, NULL)", but with
2907 * any new page allocations done using the specified allocation flags.
2908 *
2909 * If the page does not get brought uptodate, return -EIO.
2910 *
2911 * Return: up to date page on success, ERR_PTR() on failure.
2912 */
2913struct page *read_cache_page_gfp(struct address_space *mapping,
2914 pgoff_t index,
2915 gfp_t gfp)
2916{
2917 return do_read_cache_page(mapping, index, NULL, NULL, gfp);
2918}
2919EXPORT_SYMBOL(read_cache_page_gfp);
2920
2921/*
2922 * Don't operate on ranges the page cache doesn't support, and don't exceed the
2923 * LFS limits. If pos is under the limit it becomes a short access. If it
2924 * exceeds the limit we return -EFBIG.
2925 */
2926static int generic_write_check_limits(struct file *file, loff_t pos,
2927 loff_t *count)
2928{
2929 struct inode *inode = file->f_mapping->host;
2930 loff_t max_size = inode->i_sb->s_maxbytes;
2931 loff_t limit = rlimit(RLIMIT_FSIZE);
2932
2933 if (limit != RLIM_INFINITY) {
2934 if (pos >= limit) {
2935 send_sig(SIGXFSZ, current, 0);
2936 return -EFBIG;
2937 }
2938 *count = min(*count, limit - pos);
2939 }
2940
2941 if (!(file->f_flags & O_LARGEFILE))
2942 max_size = MAX_NON_LFS;
2943
2944 if (unlikely(pos >= max_size))
2945 return -EFBIG;
2946
2947 *count = min(*count, max_size - pos);
2948
2949 return 0;
2950}
2951
2952/*
2953 * Performs necessary checks before doing a write
2954 *
2955 * Can adjust writing position or amount of bytes to write.
2956 * Returns appropriate error code that caller should return or
2957 * zero in case that write should be allowed.
2958 */
2959inline ssize_t generic_write_checks(struct kiocb *iocb, struct iov_iter *from)
2960{
2961 struct file *file = iocb->ki_filp;
2962 struct inode *inode = file->f_mapping->host;
2963 loff_t count;
2964 int ret;
2965
2966 if (IS_SWAPFILE(inode))
2967 return -ETXTBSY;
2968
2969 if (!iov_iter_count(from))
2970 return 0;
2971
2972 /* FIXME: this is for backwards compatibility with 2.4 */
2973 if (iocb->ki_flags & IOCB_APPEND)
2974 iocb->ki_pos = i_size_read(inode);
2975
2976 if ((iocb->ki_flags & IOCB_NOWAIT) && !(iocb->ki_flags & IOCB_DIRECT))
2977 return -EINVAL;
2978
2979 count = iov_iter_count(from);
2980 ret = generic_write_check_limits(file, iocb->ki_pos, &count);
2981 if (ret)
2982 return ret;
2983
2984 iov_iter_truncate(from, count);
2985 return iov_iter_count(from);
2986}
2987EXPORT_SYMBOL(generic_write_checks);
2988
2989/*
2990 * Performs necessary checks before doing a clone.
2991 *
2992 * Can adjust amount of bytes to clone via @req_count argument.
2993 * Returns appropriate error code that caller should return or
2994 * zero in case the clone should be allowed.
2995 */
2996int generic_remap_checks(struct file *file_in, loff_t pos_in,
2997 struct file *file_out, loff_t pos_out,
2998 loff_t *req_count, unsigned int remap_flags)
2999{
3000 struct inode *inode_in = file_in->f_mapping->host;
3001 struct inode *inode_out = file_out->f_mapping->host;
3002 uint64_t count = *req_count;
3003 uint64_t bcount;
3004 loff_t size_in, size_out;
3005 loff_t bs = inode_out->i_sb->s_blocksize;
3006 int ret;
3007
3008 /* The start of both ranges must be aligned to an fs block. */
3009 if (!IS_ALIGNED(pos_in, bs) || !IS_ALIGNED(pos_out, bs))
3010 return -EINVAL;
3011
3012 /* Ensure offsets don't wrap. */
3013 if (pos_in + count < pos_in || pos_out + count < pos_out)
3014 return -EINVAL;
3015
3016 size_in = i_size_read(inode_in);
3017 size_out = i_size_read(inode_out);
3018
3019 /* Dedupe requires both ranges to be within EOF. */
3020 if ((remap_flags & REMAP_FILE_DEDUP) &&
3021 (pos_in >= size_in || pos_in + count > size_in ||
3022 pos_out >= size_out || pos_out + count > size_out))
3023 return -EINVAL;
3024
3025 /* Ensure the infile range is within the infile. */
3026 if (pos_in >= size_in)
3027 return -EINVAL;
3028 count = min(count, size_in - (uint64_t)pos_in);
3029
3030 ret = generic_write_check_limits(file_out, pos_out, &count);
3031 if (ret)
3032 return ret;
3033
3034 /*
3035 * If the user wanted us to link to the infile's EOF, round up to the
3036 * next block boundary for this check.
3037 *
3038 * Otherwise, make sure the count is also block-aligned, having
3039 * already confirmed the starting offsets' block alignment.
3040 */
3041 if (pos_in + count == size_in) {
3042 bcount = ALIGN(size_in, bs) - pos_in;
3043 } else {
3044 if (!IS_ALIGNED(count, bs))
3045 count = ALIGN_DOWN(count, bs);
3046 bcount = count;
3047 }
3048
3049 /* Don't allow overlapped cloning within the same file. */
3050 if (inode_in == inode_out &&
3051 pos_out + bcount > pos_in &&
3052 pos_out < pos_in + bcount)
3053 return -EINVAL;
3054
3055 /*
3056 * We shortened the request but the caller can't deal with that, so
3057 * bounce the request back to userspace.
3058 */
3059 if (*req_count != count && !(remap_flags & REMAP_FILE_CAN_SHORTEN))
3060 return -EINVAL;
3061
3062 *req_count = count;
3063 return 0;
3064}
3065
3066
3067/*
3068 * Performs common checks before doing a file copy/clone
3069 * from @file_in to @file_out.
3070 */
3071int generic_file_rw_checks(struct file *file_in, struct file *file_out)
3072{
3073 struct inode *inode_in = file_inode(file_in);
3074 struct inode *inode_out = file_inode(file_out);
3075
3076 /* Don't copy dirs, pipes, sockets... */
3077 if (S_ISDIR(inode_in->i_mode) || S_ISDIR(inode_out->i_mode))
3078 return -EISDIR;
3079 if (!S_ISREG(inode_in->i_mode) || !S_ISREG(inode_out->i_mode))
3080 return -EINVAL;
3081
3082 if (!(file_in->f_mode & FMODE_READ) ||
3083 !(file_out->f_mode & FMODE_WRITE) ||
3084 (file_out->f_flags & O_APPEND))
3085 return -EBADF;
3086
3087 return 0;
3088}
3089
3090/*
3091 * Performs necessary checks before doing a file copy
3092 *
3093 * Can adjust amount of bytes to copy via @req_count argument.
3094 * Returns appropriate error code that caller should return or
3095 * zero in case the copy should be allowed.
3096 */
3097int generic_copy_file_checks(struct file *file_in, loff_t pos_in,
3098 struct file *file_out, loff_t pos_out,
3099 size_t *req_count, unsigned int flags)
3100{
3101 struct inode *inode_in = file_inode(file_in);
3102 struct inode *inode_out = file_inode(file_out);
3103 uint64_t count = *req_count;
3104 loff_t size_in;
3105 int ret;
3106
3107 ret = generic_file_rw_checks(file_in, file_out);
3108 if (ret)
3109 return ret;
3110
3111 /* Don't touch certain kinds of inodes */
3112 if (IS_IMMUTABLE(inode_out))
3113 return -EPERM;
3114
3115 if (IS_SWAPFILE(inode_in) || IS_SWAPFILE(inode_out))
3116 return -ETXTBSY;
3117
3118 /* Ensure offsets don't wrap. */
3119 if (pos_in + count < pos_in || pos_out + count < pos_out)
3120 return -EOVERFLOW;
3121
3122 /* Shorten the copy to EOF */
3123 size_in = i_size_read(inode_in);
3124 if (pos_in >= size_in)
3125 count = 0;
3126 else
3127 count = min(count, size_in - (uint64_t)pos_in);
3128
3129 ret = generic_write_check_limits(file_out, pos_out, &count);
3130 if (ret)
3131 return ret;
3132
3133 /* Don't allow overlapped copying within the same file. */
3134 if (inode_in == inode_out &&
3135 pos_out + count > pos_in &&
3136 pos_out < pos_in + count)
3137 return -EINVAL;
3138
3139 *req_count = count;
3140 return 0;
3141}
3142
3143int pagecache_write_begin(struct file *file, struct address_space *mapping,
3144 loff_t pos, unsigned len, unsigned flags,
3145 struct page **pagep, void **fsdata)
3146{
3147 const struct address_space_operations *aops = mapping->a_ops;
3148
3149 return aops->write_begin(file, mapping, pos, len, flags,
3150 pagep, fsdata);
3151}
3152EXPORT_SYMBOL(pagecache_write_begin);
3153
3154int pagecache_write_end(struct file *file, struct address_space *mapping,
3155 loff_t pos, unsigned len, unsigned copied,
3156 struct page *page, void *fsdata)
3157{
3158 const struct address_space_operations *aops = mapping->a_ops;
3159
3160 return aops->write_end(file, mapping, pos, len, copied, page, fsdata);
3161}
3162EXPORT_SYMBOL(pagecache_write_end);
3163
3164ssize_t
3165generic_file_direct_write(struct kiocb *iocb, struct iov_iter *from)
3166{
3167 struct file *file = iocb->ki_filp;
3168 struct address_space *mapping = file->f_mapping;
3169 struct inode *inode = mapping->host;
3170 loff_t pos = iocb->ki_pos;
3171 ssize_t written;
3172 size_t write_len;
3173 pgoff_t end;
3174
3175 write_len = iov_iter_count(from);
3176 end = (pos + write_len - 1) >> PAGE_SHIFT;
3177
3178 if (iocb->ki_flags & IOCB_NOWAIT) {
3179 /* If there are pages to writeback, return */
3180 if (filemap_range_has_page(inode->i_mapping, pos,
3181 pos + write_len - 1))
3182 return -EAGAIN;
3183 } else {
3184 written = filemap_write_and_wait_range(mapping, pos,
3185 pos + write_len - 1);
3186 if (written)
3187 goto out;
3188 }
3189
3190 /*
3191 * After a write we want buffered reads to be sure to go to disk to get
3192 * the new data. We invalidate clean cached page from the region we're
3193 * about to write. We do this *before* the write so that we can return
3194 * without clobbering -EIOCBQUEUED from ->direct_IO().
3195 */
3196 written = invalidate_inode_pages2_range(mapping,
3197 pos >> PAGE_SHIFT, end);
3198 /*
3199 * If a page can not be invalidated, return 0 to fall back
3200 * to buffered write.
3201 */
3202 if (written) {
3203 if (written == -EBUSY)
3204 return 0;
3205 goto out;
3206 }
3207
3208 written = mapping->a_ops->direct_IO(iocb, from);
3209
3210 /*
3211 * Finally, try again to invalidate clean pages which might have been
3212 * cached by non-direct readahead, or faulted in by get_user_pages()
3213 * if the source of the write was an mmap'ed region of the file
3214 * we're writing. Either one is a pretty crazy thing to do,
3215 * so we don't support it 100%. If this invalidation
3216 * fails, tough, the write still worked...
3217 *
3218 * Most of the time we do not need this since dio_complete() will do
3219 * the invalidation for us. However there are some file systems that
3220 * do not end up with dio_complete() being called, so let's not break
3221 * them by removing it completely
3222 */
3223 if (mapping->nrpages)
3224 invalidate_inode_pages2_range(mapping,
3225 pos >> PAGE_SHIFT, end);
3226
3227 if (written > 0) {
3228 pos += written;
3229 write_len -= written;
3230 if (pos > i_size_read(inode) && !S_ISBLK(inode->i_mode)) {
3231 i_size_write(inode, pos);
3232 mark_inode_dirty(inode);
3233 }
3234 iocb->ki_pos = pos;
3235 }
3236 iov_iter_revert(from, write_len - iov_iter_count(from));
3237out:
3238 return written;
3239}
3240EXPORT_SYMBOL(generic_file_direct_write);
3241
3242/*
3243 * Find or create a page at the given pagecache position. Return the locked
3244 * page. This function is specifically for buffered writes.
3245 */
3246struct page *grab_cache_page_write_begin(struct address_space *mapping,
3247 pgoff_t index, unsigned flags)
3248{
3249 struct page *page;
3250 int fgp_flags = FGP_LOCK|FGP_WRITE|FGP_CREAT;
3251
3252 if (flags & AOP_FLAG_NOFS)
3253 fgp_flags |= FGP_NOFS;
3254
3255 page = pagecache_get_page(mapping, index, fgp_flags,
3256 mapping_gfp_mask(mapping));
3257 if (page)
3258 wait_for_stable_page(page);
3259
3260 return page;
3261}
3262EXPORT_SYMBOL(grab_cache_page_write_begin);
3263
3264ssize_t generic_perform_write(struct file *file,
3265 struct iov_iter *i, loff_t pos)
3266{
3267 struct address_space *mapping = file->f_mapping;
3268 const struct address_space_operations *a_ops = mapping->a_ops;
3269 long status = 0;
3270 ssize_t written = 0;
3271 unsigned int flags = 0;
3272
3273 do {
3274 struct page *page;
3275 unsigned long offset; /* Offset into pagecache page */
3276 unsigned long bytes; /* Bytes to write to page */
3277 size_t copied; /* Bytes copied from user */
3278 void *fsdata;
3279
3280 offset = (pos & (PAGE_SIZE - 1));
3281 bytes = min_t(unsigned long, PAGE_SIZE - offset,
3282 iov_iter_count(i));
3283
3284again:
3285 /*
3286 * Bring in the user page that we will copy from _first_.
3287 * Otherwise there's a nasty deadlock on copying from the
3288 * same page as we're writing to, without it being marked
3289 * up-to-date.
3290 *
3291 * Not only is this an optimisation, but it is also required
3292 * to check that the address is actually valid, when atomic
3293 * usercopies are used, below.
3294 */
3295 if (unlikely(iov_iter_fault_in_readable(i, bytes))) {
3296 status = -EFAULT;
3297 break;
3298 }
3299
3300 if (fatal_signal_pending(current)) {
3301 status = -EINTR;
3302 break;
3303 }
3304
3305 status = a_ops->write_begin(file, mapping, pos, bytes, flags,
3306 &page, &fsdata);
3307 if (unlikely(status < 0))
3308 break;
3309
3310 if (mapping_writably_mapped(mapping))
3311 flush_dcache_page(page);
3312
3313 copied = iov_iter_copy_from_user_atomic(page, i, offset, bytes);
3314 flush_dcache_page(page);
3315
3316 status = a_ops->write_end(file, mapping, pos, bytes, copied,
3317 page, fsdata);
3318 if (unlikely(status < 0))
3319 break;
3320 copied = status;
3321
3322 cond_resched();
3323
3324 iov_iter_advance(i, copied);
3325 if (unlikely(copied == 0)) {
3326 /*
3327 * If we were unable to copy any data at all, we must
3328 * fall back to a single segment length write.
3329 *
3330 * If we didn't fallback here, we could livelock
3331 * because not all segments in the iov can be copied at
3332 * once without a pagefault.
3333 */
3334 bytes = min_t(unsigned long, PAGE_SIZE - offset,
3335 iov_iter_single_seg_count(i));
3336 goto again;
3337 }
3338 pos += copied;
3339 written += copied;
3340
3341 balance_dirty_pages_ratelimited(mapping);
3342 } while (iov_iter_count(i));
3343
3344 return written ? written : status;
3345}
3346EXPORT_SYMBOL(generic_perform_write);
3347
3348/**
3349 * __generic_file_write_iter - write data to a file
3350 * @iocb: IO state structure (file, offset, etc.)
3351 * @from: iov_iter with data to write
3352 *
3353 * This function does all the work needed for actually writing data to a
3354 * file. It does all basic checks, removes SUID from the file, updates
3355 * modification times and calls proper subroutines depending on whether we
3356 * do direct IO or a standard buffered write.
3357 *
3358 * It expects i_mutex to be grabbed unless we work on a block device or similar
3359 * object which does not need locking at all.
3360 *
3361 * This function does *not* take care of syncing data in case of O_SYNC write.
3362 * A caller has to handle it. This is mainly due to the fact that we want to
3363 * avoid syncing under i_mutex.
3364 *
3365 * Return:
3366 * * number of bytes written, even for truncated writes
3367 * * negative error code if no data has been written at all
3368 */
3369ssize_t __generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
3370{
3371 struct file *file = iocb->ki_filp;
3372 struct address_space * mapping = file->f_mapping;
3373 struct inode *inode = mapping->host;
3374 ssize_t written = 0;
3375 ssize_t err;
3376 ssize_t status;
3377
3378 /* We can write back this queue in page reclaim */
3379 current->backing_dev_info = inode_to_bdi(inode);
3380 err = file_remove_privs(file);
3381 if (err)
3382 goto out;
3383
3384 err = file_update_time(file);
3385 if (err)
3386 goto out;
3387
3388 if (iocb->ki_flags & IOCB_DIRECT) {
3389 loff_t pos, endbyte;
3390
3391 written = generic_file_direct_write(iocb, from);
3392 /*
3393 * If the write stopped short of completing, fall back to
3394 * buffered writes. Some filesystems do this for writes to
3395 * holes, for example. For DAX files, a buffered write will
3396 * not succeed (even if it did, DAX does not handle dirty
3397 * page-cache pages correctly).
3398 */
3399 if (written < 0 || !iov_iter_count(from) || IS_DAX(inode))
3400 goto out;
3401
3402 status = generic_perform_write(file, from, pos = iocb->ki_pos);
3403 /*
3404 * If generic_perform_write() returned a synchronous error
3405 * then we want to return the number of bytes which were
3406 * direct-written, or the error code if that was zero. Note
3407 * that this differs from normal direct-io semantics, which
3408 * will return -EFOO even if some bytes were written.
3409 */
3410 if (unlikely(status < 0)) {
3411 err = status;
3412 goto out;
3413 }
3414 /*
3415 * We need to ensure that the page cache pages are written to
3416 * disk and invalidated to preserve the expected O_DIRECT
3417 * semantics.
3418 */
3419 endbyte = pos + status - 1;
3420 err = filemap_write_and_wait_range(mapping, pos, endbyte);
3421 if (err == 0) {
3422 iocb->ki_pos = endbyte + 1;
3423 written += status;
3424 invalidate_mapping_pages(mapping,
3425 pos >> PAGE_SHIFT,
3426 endbyte >> PAGE_SHIFT);
3427 } else {
3428 /*
3429 * We don't know how much we wrote, so just return
3430 * the number of bytes which were direct-written
3431 */
3432 }
3433 } else {
3434 written = generic_perform_write(file, from, iocb->ki_pos);
3435 if (likely(written > 0))
3436 iocb->ki_pos += written;
3437 }
3438out:
3439 current->backing_dev_info = NULL;
3440 return written ? written : err;
3441}
3442EXPORT_SYMBOL(__generic_file_write_iter);
3443
3444/**
3445 * generic_file_write_iter - write data to a file
3446 * @iocb: IO state structure
3447 * @from: iov_iter with data to write
3448 *
3449 * This is a wrapper around __generic_file_write_iter() to be used by most
3450 * filesystems. It takes care of syncing the file in case of O_SYNC file
3451 * and acquires i_mutex as needed.
3452 * Return:
3453 * * negative error code if no data has been written at all of
3454 * vfs_fsync_range() failed for a synchronous write
3455 * * number of bytes written, even for truncated writes
3456 */
3457ssize_t generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
3458{
3459 struct file *file = iocb->ki_filp;
3460 struct inode *inode = file->f_mapping->host;
3461 ssize_t ret;
3462
3463 inode_lock(inode);
3464 ret = generic_write_checks(iocb, from);
3465 if (ret > 0)
3466 ret = __generic_file_write_iter(iocb, from);
3467 inode_unlock(inode);
3468
3469 if (ret > 0)
3470 ret = generic_write_sync(iocb, ret);
3471 return ret;
3472}
3473EXPORT_SYMBOL(generic_file_write_iter);
3474
3475/**
3476 * try_to_release_page() - release old fs-specific metadata on a page
3477 *
3478 * @page: the page which the kernel is trying to free
3479 * @gfp_mask: memory allocation flags (and I/O mode)
3480 *
3481 * The address_space is to try to release any data against the page
3482 * (presumably at page->private).
3483 *
3484 * This may also be called if PG_fscache is set on a page, indicating that the
3485 * page is known to the local caching routines.
3486 *
3487 * The @gfp_mask argument specifies whether I/O may be performed to release
3488 * this page (__GFP_IO), and whether the call may block (__GFP_RECLAIM & __GFP_FS).
3489 *
3490 * Return: %1 if the release was successful, otherwise return zero.
3491 */
3492int try_to_release_page(struct page *page, gfp_t gfp_mask)
3493{
3494 struct address_space * const mapping = page->mapping;
3495
3496 BUG_ON(!PageLocked(page));
3497 if (PageWriteback(page))
3498 return 0;
3499
3500 if (mapping && mapping->a_ops->releasepage)
3501 return mapping->a_ops->releasepage(page, gfp_mask);
3502 return try_to_free_buffers(page);
3503}
3504
3505EXPORT_SYMBOL(try_to_release_page);