Staging: rt2870: Add USB ID for Buffalo Airstation WLI-UC-GN
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / mm / filemap.c
blob9701a501f7696b065b75c645945b386630009bc5
1 /*
2 * linux/mm/filemap.c
4 * Copyright (C) 1994-1999 Linus Torvalds
5 */
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)
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/mm_inline.h> /* for page_is_file_cache() */
37 #include "internal.h"
40 * FIXME: remove all knowledge of the buffer layer from the core VM
42 #include <linux/buffer_head.h> /* for try_to_free_buffers */
44 #include <asm/mman.h>
47 * Shared mappings implemented 30.11.1994. It's not fully working yet,
48 * though.
50 * Shared mappings now work. 15.8.1995 Bruno.
52 * finished 'unifying' the page and buffer cache and SMP-threaded the
53 * page-cache, 21.05.1999, Ingo Molnar <mingo@redhat.com>
55 * SMP-threaded pagemap-LRU 1999, Andrea Arcangeli <andrea@suse.de>
59 * Lock ordering:
61 * ->i_mmap_lock (truncate_pagecache)
62 * ->private_lock (__free_pte->__set_page_dirty_buffers)
63 * ->swap_lock (exclusive_swap_page, others)
64 * ->mapping->tree_lock
66 * ->i_mutex
67 * ->i_mmap_lock (truncate->unmap_mapping_range)
69 * ->mmap_sem
70 * ->i_mmap_lock
71 * ->page_table_lock or pte_lock (various, mainly in memory.c)
72 * ->mapping->tree_lock (arch-dependent flush_dcache_mmap_lock)
74 * ->mmap_sem
75 * ->lock_page (access_process_vm)
77 * ->i_mutex (generic_file_buffered_write)
78 * ->mmap_sem (fault_in_pages_readable->do_page_fault)
80 * ->i_mutex
81 * ->i_alloc_sem (various)
83 * ->inode_lock
84 * ->sb_lock (fs/fs-writeback.c)
85 * ->mapping->tree_lock (__sync_single_inode)
87 * ->i_mmap_lock
88 * ->anon_vma.lock (vma_adjust)
90 * ->anon_vma.lock
91 * ->page_table_lock or pte_lock (anon_vma_prepare and various)
93 * ->page_table_lock or pte_lock
94 * ->swap_lock (try_to_unmap_one)
95 * ->private_lock (try_to_unmap_one)
96 * ->tree_lock (try_to_unmap_one)
97 * ->zone.lru_lock (follow_page->mark_page_accessed)
98 * ->zone.lru_lock (check_pte_range->isolate_lru_page)
99 * ->private_lock (page_remove_rmap->set_page_dirty)
100 * ->tree_lock (page_remove_rmap->set_page_dirty)
101 * ->inode_lock (page_remove_rmap->set_page_dirty)
102 * ->inode_lock (zap_pte_range->set_page_dirty)
103 * ->private_lock (zap_pte_range->__set_page_dirty_buffers)
105 * ->task->proc_lock
106 * ->dcache_lock (proc_pid_lookup)
108 * (code doesn't rely on that order, so you could switch it around)
109 * ->tasklist_lock (memory_failure, collect_procs_ao)
110 * ->i_mmap_lock
114 * Remove a page from the page cache and free it. Caller has to make
115 * sure the page is locked and that nobody else uses it - or that usage
116 * is safe. The caller must hold the mapping's tree_lock.
118 void __remove_from_page_cache(struct page *page)
120 struct address_space *mapping = page->mapping;
122 radix_tree_delete(&mapping->page_tree, page->index);
123 page->mapping = NULL;
124 mapping->nrpages--;
125 __dec_zone_page_state(page, NR_FILE_PAGES);
126 if (PageSwapBacked(page))
127 __dec_zone_page_state(page, NR_SHMEM);
128 BUG_ON(page_mapped(page));
131 * Some filesystems seem to re-dirty the page even after
132 * the VM has canceled the dirty bit (eg ext3 journaling).
134 * Fix it up by doing a final dirty accounting check after
135 * having removed the page entirely.
137 if (PageDirty(page) && mapping_cap_account_dirty(mapping)) {
138 dec_zone_page_state(page, NR_FILE_DIRTY);
139 dec_bdi_stat(mapping->backing_dev_info, BDI_RECLAIMABLE);
143 void remove_from_page_cache(struct page *page)
145 struct address_space *mapping = page->mapping;
147 BUG_ON(!PageLocked(page));
149 spin_lock_irq(&mapping->tree_lock);
150 __remove_from_page_cache(page);
151 spin_unlock_irq(&mapping->tree_lock);
152 mem_cgroup_uncharge_cache_page(page);
154 EXPORT_SYMBOL(remove_from_page_cache);
156 static int sync_page(void *word)
158 struct address_space *mapping;
159 struct page *page;
161 page = container_of((unsigned long *)word, struct page, flags);
164 * page_mapping() is being called without PG_locked held.
165 * Some knowledge of the state and use of the page is used to
166 * reduce the requirements down to a memory barrier.
167 * The danger here is of a stale page_mapping() return value
168 * indicating a struct address_space different from the one it's
169 * associated with when it is associated with one.
170 * After smp_mb(), it's either the correct page_mapping() for
171 * the page, or an old page_mapping() and the page's own
172 * page_mapping() has gone NULL.
173 * The ->sync_page() address_space operation must tolerate
174 * page_mapping() going NULL. By an amazing coincidence,
175 * this comes about because none of the users of the page
176 * in the ->sync_page() methods make essential use of the
177 * page_mapping(), merely passing the page down to the backing
178 * device's unplug functions when it's non-NULL, which in turn
179 * ignore it for all cases but swap, where only page_private(page) is
180 * of interest. When page_mapping() does go NULL, the entire
181 * call stack gracefully ignores the page and returns.
182 * -- wli
184 smp_mb();
185 mapping = page_mapping(page);
186 if (mapping && mapping->a_ops && mapping->a_ops->sync_page)
187 mapping->a_ops->sync_page(page);
188 io_schedule();
189 return 0;
192 static int sync_page_killable(void *word)
194 sync_page(word);
195 return fatal_signal_pending(current) ? -EINTR : 0;
199 * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range
200 * @mapping: address space structure to write
201 * @start: offset in bytes where the range starts
202 * @end: offset in bytes where the range ends (inclusive)
203 * @sync_mode: enable synchronous operation
205 * Start writeback against all of a mapping's dirty pages that lie
206 * within the byte offsets <start, end> inclusive.
208 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
209 * opposed to a regular memory cleansing writeback. The difference between
210 * these two operations is that if a dirty page/buffer is encountered, it must
211 * be waited upon, and not just skipped over.
213 int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
214 loff_t end, int sync_mode)
216 int ret;
217 struct writeback_control wbc = {
218 .sync_mode = sync_mode,
219 .nr_to_write = LONG_MAX,
220 .range_start = start,
221 .range_end = end,
224 if (!mapping_cap_writeback_dirty(mapping))
225 return 0;
227 ret = do_writepages(mapping, &wbc);
228 return ret;
231 static inline int __filemap_fdatawrite(struct address_space *mapping,
232 int sync_mode)
234 return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode);
237 int filemap_fdatawrite(struct address_space *mapping)
239 return __filemap_fdatawrite(mapping, WB_SYNC_ALL);
241 EXPORT_SYMBOL(filemap_fdatawrite);
243 int filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
244 loff_t end)
246 return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL);
248 EXPORT_SYMBOL(filemap_fdatawrite_range);
251 * filemap_flush - mostly a non-blocking flush
252 * @mapping: target address_space
254 * This is a mostly non-blocking flush. Not suitable for data-integrity
255 * purposes - I/O may not be started against all dirty pages.
257 int filemap_flush(struct address_space *mapping)
259 return __filemap_fdatawrite(mapping, WB_SYNC_NONE);
261 EXPORT_SYMBOL(filemap_flush);
264 * filemap_fdatawait_range - wait for writeback to complete
265 * @mapping: address space structure to wait for
266 * @start_byte: offset in bytes where the range starts
267 * @end_byte: offset in bytes where the range ends (inclusive)
269 * Walk the list of under-writeback pages of the given address space
270 * in the given range and wait for all of them.
272 int filemap_fdatawait_range(struct address_space *mapping, loff_t start_byte,
273 loff_t end_byte)
275 pgoff_t index = start_byte >> PAGE_CACHE_SHIFT;
276 pgoff_t end = end_byte >> PAGE_CACHE_SHIFT;
277 struct pagevec pvec;
278 int nr_pages;
279 int ret = 0;
281 if (end_byte < start_byte)
282 return 0;
284 pagevec_init(&pvec, 0);
285 while ((index <= end) &&
286 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
287 PAGECACHE_TAG_WRITEBACK,
288 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1)) != 0) {
289 unsigned i;
291 for (i = 0; i < nr_pages; i++) {
292 struct page *page = pvec.pages[i];
294 /* until radix tree lookup accepts end_index */
295 if (page->index > end)
296 continue;
298 wait_on_page_writeback(page);
299 if (PageError(page))
300 ret = -EIO;
302 pagevec_release(&pvec);
303 cond_resched();
306 /* Check for outstanding write errors */
307 if (test_and_clear_bit(AS_ENOSPC, &mapping->flags))
308 ret = -ENOSPC;
309 if (test_and_clear_bit(AS_EIO, &mapping->flags))
310 ret = -EIO;
312 return ret;
314 EXPORT_SYMBOL(filemap_fdatawait_range);
317 * filemap_fdatawait - wait for all under-writeback pages to complete
318 * @mapping: address space structure to wait for
320 * Walk the list of under-writeback pages of the given address space
321 * and wait for all of them.
323 int filemap_fdatawait(struct address_space *mapping)
325 loff_t i_size = i_size_read(mapping->host);
327 if (i_size == 0)
328 return 0;
330 return filemap_fdatawait_range(mapping, 0, i_size - 1);
332 EXPORT_SYMBOL(filemap_fdatawait);
334 int filemap_write_and_wait(struct address_space *mapping)
336 int err = 0;
338 if (mapping->nrpages) {
339 err = filemap_fdatawrite(mapping);
341 * Even if the above returned error, the pages may be
342 * written partially (e.g. -ENOSPC), so we wait for it.
343 * But the -EIO is special case, it may indicate the worst
344 * thing (e.g. bug) happened, so we avoid waiting for it.
346 if (err != -EIO) {
347 int err2 = filemap_fdatawait(mapping);
348 if (!err)
349 err = err2;
352 return err;
354 EXPORT_SYMBOL(filemap_write_and_wait);
357 * filemap_write_and_wait_range - write out & wait on a file range
358 * @mapping: the address_space for the pages
359 * @lstart: offset in bytes where the range starts
360 * @lend: offset in bytes where the range ends (inclusive)
362 * Write out and wait upon file offsets lstart->lend, inclusive.
364 * Note that `lend' is inclusive (describes the last byte to be written) so
365 * that this function can be used to write to the very end-of-file (end = -1).
367 int filemap_write_and_wait_range(struct address_space *mapping,
368 loff_t lstart, loff_t lend)
370 int err = 0;
372 if (mapping->nrpages) {
373 err = __filemap_fdatawrite_range(mapping, lstart, lend,
374 WB_SYNC_ALL);
375 /* See comment of filemap_write_and_wait() */
376 if (err != -EIO) {
377 int err2 = filemap_fdatawait_range(mapping,
378 lstart, lend);
379 if (!err)
380 err = err2;
383 return err;
385 EXPORT_SYMBOL(filemap_write_and_wait_range);
388 * add_to_page_cache_locked - add a locked page to the pagecache
389 * @page: page to add
390 * @mapping: the page's address_space
391 * @offset: page index
392 * @gfp_mask: page allocation mode
394 * This function is used to add a page to the pagecache. It must be locked.
395 * This function does not add the page to the LRU. The caller must do that.
397 int add_to_page_cache_locked(struct page *page, struct address_space *mapping,
398 pgoff_t offset, gfp_t gfp_mask)
400 int error;
402 VM_BUG_ON(!PageLocked(page));
404 error = mem_cgroup_cache_charge(page, current->mm,
405 gfp_mask & GFP_RECLAIM_MASK);
406 if (error)
407 goto out;
409 error = radix_tree_preload(gfp_mask & ~__GFP_HIGHMEM);
410 if (error == 0) {
411 page_cache_get(page);
412 page->mapping = mapping;
413 page->index = offset;
415 spin_lock_irq(&mapping->tree_lock);
416 error = radix_tree_insert(&mapping->page_tree, offset, page);
417 if (likely(!error)) {
418 mapping->nrpages++;
419 __inc_zone_page_state(page, NR_FILE_PAGES);
420 if (PageSwapBacked(page))
421 __inc_zone_page_state(page, NR_SHMEM);
422 spin_unlock_irq(&mapping->tree_lock);
423 } else {
424 page->mapping = NULL;
425 spin_unlock_irq(&mapping->tree_lock);
426 mem_cgroup_uncharge_cache_page(page);
427 page_cache_release(page);
429 radix_tree_preload_end();
430 } else
431 mem_cgroup_uncharge_cache_page(page);
432 out:
433 return error;
435 EXPORT_SYMBOL(add_to_page_cache_locked);
437 int add_to_page_cache_lru(struct page *page, struct address_space *mapping,
438 pgoff_t offset, gfp_t gfp_mask)
440 int ret;
443 * Splice_read and readahead add shmem/tmpfs pages into the page cache
444 * before shmem_readpage has a chance to mark them as SwapBacked: they
445 * need to go on the anon lru below, and mem_cgroup_cache_charge
446 * (called in add_to_page_cache) needs to know where they're going too.
448 if (mapping_cap_swap_backed(mapping))
449 SetPageSwapBacked(page);
451 ret = add_to_page_cache(page, mapping, offset, gfp_mask);
452 if (ret == 0) {
453 if (page_is_file_cache(page))
454 lru_cache_add_file(page);
455 else
456 lru_cache_add_anon(page);
458 return ret;
460 EXPORT_SYMBOL_GPL(add_to_page_cache_lru);
462 #ifdef CONFIG_NUMA
463 struct page *__page_cache_alloc(gfp_t gfp)
465 int n;
466 struct page *page;
468 if (cpuset_do_page_mem_spread()) {
469 get_mems_allowed();
470 n = cpuset_mem_spread_node();
471 page = alloc_pages_exact_node(n, gfp, 0);
472 put_mems_allowed();
473 return page;
475 return alloc_pages(gfp, 0);
477 EXPORT_SYMBOL(__page_cache_alloc);
478 #endif
480 static int __sleep_on_page_lock(void *word)
482 io_schedule();
483 return 0;
487 * In order to wait for pages to become available there must be
488 * waitqueues associated with pages. By using a hash table of
489 * waitqueues where the bucket discipline is to maintain all
490 * waiters on the same queue and wake all when any of the pages
491 * become available, and for the woken contexts to check to be
492 * sure the appropriate page became available, this saves space
493 * at a cost of "thundering herd" phenomena during rare hash
494 * collisions.
496 static wait_queue_head_t *page_waitqueue(struct page *page)
498 const struct zone *zone = page_zone(page);
500 return &zone->wait_table[hash_ptr(page, zone->wait_table_bits)];
503 static inline void wake_up_page(struct page *page, int bit)
505 __wake_up_bit(page_waitqueue(page), &page->flags, bit);
508 void wait_on_page_bit(struct page *page, int bit_nr)
510 DEFINE_WAIT_BIT(wait, &page->flags, bit_nr);
512 if (test_bit(bit_nr, &page->flags))
513 __wait_on_bit(page_waitqueue(page), &wait, sync_page,
514 TASK_UNINTERRUPTIBLE);
516 EXPORT_SYMBOL(wait_on_page_bit);
519 * add_page_wait_queue - Add an arbitrary waiter to a page's wait queue
520 * @page: Page defining the wait queue of interest
521 * @waiter: Waiter to add to the queue
523 * Add an arbitrary @waiter to the wait queue for the nominated @page.
525 void add_page_wait_queue(struct page *page, wait_queue_t *waiter)
527 wait_queue_head_t *q = page_waitqueue(page);
528 unsigned long flags;
530 spin_lock_irqsave(&q->lock, flags);
531 __add_wait_queue(q, waiter);
532 spin_unlock_irqrestore(&q->lock, flags);
534 EXPORT_SYMBOL_GPL(add_page_wait_queue);
537 * unlock_page - unlock a locked page
538 * @page: the page
540 * Unlocks the page and wakes up sleepers in ___wait_on_page_locked().
541 * Also wakes sleepers in wait_on_page_writeback() because the wakeup
542 * mechananism between PageLocked pages and PageWriteback pages is shared.
543 * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep.
545 * The mb is necessary to enforce ordering between the clear_bit and the read
546 * of the waitqueue (to avoid SMP races with a parallel wait_on_page_locked()).
548 void unlock_page(struct page *page)
550 VM_BUG_ON(!PageLocked(page));
551 clear_bit_unlock(PG_locked, &page->flags);
552 smp_mb__after_clear_bit();
553 wake_up_page(page, PG_locked);
555 EXPORT_SYMBOL(unlock_page);
558 * end_page_writeback - end writeback against a page
559 * @page: the page
561 void end_page_writeback(struct page *page)
563 if (TestClearPageReclaim(page))
564 rotate_reclaimable_page(page);
566 if (!test_clear_page_writeback(page))
567 BUG();
569 smp_mb__after_clear_bit();
570 wake_up_page(page, PG_writeback);
572 EXPORT_SYMBOL(end_page_writeback);
575 * __lock_page - get a lock on the page, assuming we need to sleep to get it
576 * @page: the page to lock
578 * Ugly. Running sync_page() in state TASK_UNINTERRUPTIBLE is scary. If some
579 * random driver's requestfn sets TASK_RUNNING, we could busywait. However
580 * chances are that on the second loop, the block layer's plug list is empty,
581 * so sync_page() will then return in state TASK_UNINTERRUPTIBLE.
583 void __lock_page(struct page *page)
585 DEFINE_WAIT_BIT(wait, &page->flags, PG_locked);
587 __wait_on_bit_lock(page_waitqueue(page), &wait, sync_page,
588 TASK_UNINTERRUPTIBLE);
590 EXPORT_SYMBOL(__lock_page);
592 int __lock_page_killable(struct page *page)
594 DEFINE_WAIT_BIT(wait, &page->flags, PG_locked);
596 return __wait_on_bit_lock(page_waitqueue(page), &wait,
597 sync_page_killable, TASK_KILLABLE);
599 EXPORT_SYMBOL_GPL(__lock_page_killable);
602 * __lock_page_nosync - get a lock on the page, without calling sync_page()
603 * @page: the page to lock
605 * Variant of lock_page that does not require the caller to hold a reference
606 * on the page's mapping.
608 void __lock_page_nosync(struct page *page)
610 DEFINE_WAIT_BIT(wait, &page->flags, PG_locked);
611 __wait_on_bit_lock(page_waitqueue(page), &wait, __sleep_on_page_lock,
612 TASK_UNINTERRUPTIBLE);
616 * find_get_page - find and get a page reference
617 * @mapping: the address_space to search
618 * @offset: the page index
620 * Is there a pagecache struct page at the given (mapping, offset) tuple?
621 * If yes, increment its refcount and return it; if no, return NULL.
623 struct page *find_get_page(struct address_space *mapping, pgoff_t offset)
625 void **pagep;
626 struct page *page;
628 rcu_read_lock();
629 repeat:
630 page = NULL;
631 pagep = radix_tree_lookup_slot(&mapping->page_tree, offset);
632 if (pagep) {
633 page = radix_tree_deref_slot(pagep);
634 if (unlikely(!page))
635 goto out;
636 if (radix_tree_deref_retry(page))
637 goto repeat;
639 if (!page_cache_get_speculative(page))
640 goto repeat;
643 * Has the page moved?
644 * This is part of the lockless pagecache protocol. See
645 * include/linux/pagemap.h for details.
647 if (unlikely(page != *pagep)) {
648 page_cache_release(page);
649 goto repeat;
652 out:
653 rcu_read_unlock();
655 return page;
657 EXPORT_SYMBOL(find_get_page);
660 * find_lock_page - locate, pin and lock a pagecache page
661 * @mapping: the address_space to search
662 * @offset: the page index
664 * Locates the desired pagecache page, locks it, increments its reference
665 * count and returns its address.
667 * Returns zero if the page was not present. find_lock_page() may sleep.
669 struct page *find_lock_page(struct address_space *mapping, pgoff_t offset)
671 struct page *page;
673 repeat:
674 page = find_get_page(mapping, offset);
675 if (page) {
676 lock_page(page);
677 /* Has the page been truncated? */
678 if (unlikely(page->mapping != mapping)) {
679 unlock_page(page);
680 page_cache_release(page);
681 goto repeat;
683 VM_BUG_ON(page->index != offset);
685 return page;
687 EXPORT_SYMBOL(find_lock_page);
690 * find_or_create_page - locate or add a pagecache page
691 * @mapping: the page's address_space
692 * @index: the page's index into the mapping
693 * @gfp_mask: page allocation mode
695 * Locates a page in the pagecache. If the page is not present, a new page
696 * is allocated using @gfp_mask and is added to the pagecache and to the VM's
697 * LRU list. The returned page is locked and has its reference count
698 * incremented.
700 * find_or_create_page() may sleep, even if @gfp_flags specifies an atomic
701 * allocation!
703 * find_or_create_page() returns the desired page's address, or zero on
704 * memory exhaustion.
706 struct page *find_or_create_page(struct address_space *mapping,
707 pgoff_t index, gfp_t gfp_mask)
709 struct page *page;
710 int err;
711 repeat:
712 page = find_lock_page(mapping, index);
713 if (!page) {
714 page = __page_cache_alloc(gfp_mask);
715 if (!page)
716 return NULL;
718 * We want a regular kernel memory (not highmem or DMA etc)
719 * allocation for the radix tree nodes, but we need to honour
720 * the context-specific requirements the caller has asked for.
721 * GFP_RECLAIM_MASK collects those requirements.
723 err = add_to_page_cache_lru(page, mapping, index,
724 (gfp_mask & GFP_RECLAIM_MASK));
725 if (unlikely(err)) {
726 page_cache_release(page);
727 page = NULL;
728 if (err == -EEXIST)
729 goto repeat;
732 return page;
734 EXPORT_SYMBOL(find_or_create_page);
737 * find_get_pages - gang pagecache lookup
738 * @mapping: The address_space to search
739 * @start: The starting page index
740 * @nr_pages: The maximum number of pages
741 * @pages: Where the resulting pages are placed
743 * find_get_pages() will search for and return a group of up to
744 * @nr_pages pages in the mapping. The pages are placed at @pages.
745 * find_get_pages() takes a reference against the returned pages.
747 * The search returns a group of mapping-contiguous pages with ascending
748 * indexes. There may be holes in the indices due to not-present pages.
750 * find_get_pages() returns the number of pages which were found.
752 unsigned find_get_pages(struct address_space *mapping, pgoff_t start,
753 unsigned int nr_pages, struct page **pages)
755 unsigned int i;
756 unsigned int ret;
757 unsigned int nr_found;
759 rcu_read_lock();
760 restart:
761 nr_found = radix_tree_gang_lookup_slot(&mapping->page_tree,
762 (void ***)pages, start, nr_pages);
763 ret = 0;
764 for (i = 0; i < nr_found; i++) {
765 struct page *page;
766 repeat:
767 page = radix_tree_deref_slot((void **)pages[i]);
768 if (unlikely(!page))
769 continue;
770 if (radix_tree_deref_retry(page)) {
771 if (ret)
772 start = pages[ret-1]->index;
773 goto restart;
776 if (!page_cache_get_speculative(page))
777 goto repeat;
779 /* Has the page moved? */
780 if (unlikely(page != *((void **)pages[i]))) {
781 page_cache_release(page);
782 goto repeat;
785 pages[ret] = page;
786 ret++;
788 rcu_read_unlock();
789 return ret;
793 * find_get_pages_contig - gang contiguous pagecache lookup
794 * @mapping: The address_space to search
795 * @index: The starting page index
796 * @nr_pages: The maximum number of pages
797 * @pages: Where the resulting pages are placed
799 * find_get_pages_contig() works exactly like find_get_pages(), except
800 * that the returned number of pages are guaranteed to be contiguous.
802 * find_get_pages_contig() returns the number of pages which were found.
804 unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index,
805 unsigned int nr_pages, struct page **pages)
807 unsigned int i;
808 unsigned int ret;
809 unsigned int nr_found;
811 rcu_read_lock();
812 restart:
813 nr_found = radix_tree_gang_lookup_slot(&mapping->page_tree,
814 (void ***)pages, index, nr_pages);
815 ret = 0;
816 for (i = 0; i < nr_found; i++) {
817 struct page *page;
818 repeat:
819 page = radix_tree_deref_slot((void **)pages[i]);
820 if (unlikely(!page))
821 continue;
822 if (radix_tree_deref_retry(page))
823 goto restart;
825 if (page->mapping == NULL || page->index != index)
826 break;
828 if (!page_cache_get_speculative(page))
829 goto repeat;
831 /* Has the page moved? */
832 if (unlikely(page != *((void **)pages[i]))) {
833 page_cache_release(page);
834 goto repeat;
837 pages[ret] = page;
838 ret++;
839 index++;
841 rcu_read_unlock();
842 return ret;
844 EXPORT_SYMBOL(find_get_pages_contig);
847 * find_get_pages_tag - find and return pages that match @tag
848 * @mapping: the address_space to search
849 * @index: the starting page index
850 * @tag: the tag index
851 * @nr_pages: the maximum number of pages
852 * @pages: where the resulting pages are placed
854 * Like find_get_pages, except we only return pages which are tagged with
855 * @tag. We update @index to index the next page for the traversal.
857 unsigned find_get_pages_tag(struct address_space *mapping, pgoff_t *index,
858 int tag, unsigned int nr_pages, struct page **pages)
860 unsigned int i;
861 unsigned int ret;
862 unsigned int nr_found;
864 rcu_read_lock();
865 restart:
866 nr_found = radix_tree_gang_lookup_tag_slot(&mapping->page_tree,
867 (void ***)pages, *index, nr_pages, tag);
868 ret = 0;
869 for (i = 0; i < nr_found; i++) {
870 struct page *page;
871 repeat:
872 page = radix_tree_deref_slot((void **)pages[i]);
873 if (unlikely(!page))
874 continue;
875 if (radix_tree_deref_retry(page))
876 goto restart;
878 if (!page_cache_get_speculative(page))
879 goto repeat;
881 /* Has the page moved? */
882 if (unlikely(page != *((void **)pages[i]))) {
883 page_cache_release(page);
884 goto repeat;
887 pages[ret] = page;
888 ret++;
890 rcu_read_unlock();
892 if (ret)
893 *index = pages[ret - 1]->index + 1;
895 return ret;
897 EXPORT_SYMBOL(find_get_pages_tag);
900 * grab_cache_page_nowait - returns locked page at given index in given cache
901 * @mapping: target address_space
902 * @index: the page index
904 * Same as grab_cache_page(), but do not wait if the page is unavailable.
905 * This is intended for speculative data generators, where the data can
906 * be regenerated if the page couldn't be grabbed. This routine should
907 * be safe to call while holding the lock for another page.
909 * Clear __GFP_FS when allocating the page to avoid recursion into the fs
910 * and deadlock against the caller's locked page.
912 struct page *
913 grab_cache_page_nowait(struct address_space *mapping, pgoff_t index)
915 struct page *page = find_get_page(mapping, index);
917 if (page) {
918 if (trylock_page(page))
919 return page;
920 page_cache_release(page);
921 return NULL;
923 page = __page_cache_alloc(mapping_gfp_mask(mapping) & ~__GFP_FS);
924 if (page && add_to_page_cache_lru(page, mapping, index, GFP_NOFS)) {
925 page_cache_release(page);
926 page = NULL;
928 return page;
930 EXPORT_SYMBOL(grab_cache_page_nowait);
933 * CD/DVDs are error prone. When a medium error occurs, the driver may fail
934 * a _large_ part of the i/o request. Imagine the worst scenario:
936 * ---R__________________________________________B__________
937 * ^ reading here ^ bad block(assume 4k)
939 * read(R) => miss => readahead(R...B) => media error => frustrating retries
940 * => failing the whole request => read(R) => read(R+1) =>
941 * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
942 * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
943 * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
945 * It is going insane. Fix it by quickly scaling down the readahead size.
947 static void shrink_readahead_size_eio(struct file *filp,
948 struct file_ra_state *ra)
950 ra->ra_pages /= 4;
954 * do_generic_file_read - generic file read routine
955 * @filp: the file to read
956 * @ppos: current file position
957 * @desc: read_descriptor
958 * @actor: read method
960 * This is a generic file read routine, and uses the
961 * mapping->a_ops->readpage() function for the actual low-level stuff.
963 * This is really ugly. But the goto's actually try to clarify some
964 * of the logic when it comes to error handling etc.
966 static void do_generic_file_read(struct file *filp, loff_t *ppos,
967 read_descriptor_t *desc, read_actor_t actor)
969 struct address_space *mapping = filp->f_mapping;
970 struct inode *inode = mapping->host;
971 struct file_ra_state *ra = &filp->f_ra;
972 pgoff_t index;
973 pgoff_t last_index;
974 pgoff_t prev_index;
975 unsigned long offset; /* offset into pagecache page */
976 unsigned int prev_offset;
977 int error;
979 index = *ppos >> PAGE_CACHE_SHIFT;
980 prev_index = ra->prev_pos >> PAGE_CACHE_SHIFT;
981 prev_offset = ra->prev_pos & (PAGE_CACHE_SIZE-1);
982 last_index = (*ppos + desc->count + PAGE_CACHE_SIZE-1) >> PAGE_CACHE_SHIFT;
983 offset = *ppos & ~PAGE_CACHE_MASK;
985 for (;;) {
986 struct page *page;
987 pgoff_t end_index;
988 loff_t isize;
989 unsigned long nr, ret;
991 cond_resched();
992 find_page:
993 page = find_get_page(mapping, index);
994 if (!page) {
995 page_cache_sync_readahead(mapping,
996 ra, filp,
997 index, last_index - index);
998 page = find_get_page(mapping, index);
999 if (unlikely(page == NULL))
1000 goto no_cached_page;
1002 if (PageReadahead(page)) {
1003 page_cache_async_readahead(mapping,
1004 ra, filp, page,
1005 index, last_index - index);
1007 if (!PageUptodate(page)) {
1008 if (inode->i_blkbits == PAGE_CACHE_SHIFT ||
1009 !mapping->a_ops->is_partially_uptodate)
1010 goto page_not_up_to_date;
1011 if (!trylock_page(page))
1012 goto page_not_up_to_date;
1013 /* Did it get truncated before we got the lock? */
1014 if (!page->mapping)
1015 goto page_not_up_to_date_locked;
1016 if (!mapping->a_ops->is_partially_uptodate(page,
1017 desc, offset))
1018 goto page_not_up_to_date_locked;
1019 unlock_page(page);
1021 page_ok:
1023 * i_size must be checked after we know the page is Uptodate.
1025 * Checking i_size after the check allows us to calculate
1026 * the correct value for "nr", which means the zero-filled
1027 * part of the page is not copied back to userspace (unless
1028 * another truncate extends the file - this is desired though).
1031 isize = i_size_read(inode);
1032 end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
1033 if (unlikely(!isize || index > end_index)) {
1034 page_cache_release(page);
1035 goto out;
1038 /* nr is the maximum number of bytes to copy from this page */
1039 nr = PAGE_CACHE_SIZE;
1040 if (index == end_index) {
1041 nr = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;
1042 if (nr <= offset) {
1043 page_cache_release(page);
1044 goto out;
1047 nr = nr - offset;
1049 /* If users can be writing to this page using arbitrary
1050 * virtual addresses, take care about potential aliasing
1051 * before reading the page on the kernel side.
1053 if (mapping_writably_mapped(mapping))
1054 flush_dcache_page(page);
1057 * When a sequential read accesses a page several times,
1058 * only mark it as accessed the first time.
1060 if (prev_index != index || offset != prev_offset)
1061 mark_page_accessed(page);
1062 prev_index = index;
1065 * Ok, we have the page, and it's up-to-date, so
1066 * now we can copy it to user space...
1068 * The actor routine returns how many bytes were actually used..
1069 * NOTE! This may not be the same as how much of a user buffer
1070 * we filled up (we may be padding etc), so we can only update
1071 * "pos" here (the actor routine has to update the user buffer
1072 * pointers and the remaining count).
1074 ret = actor(desc, page, offset, nr);
1075 offset += ret;
1076 index += offset >> PAGE_CACHE_SHIFT;
1077 offset &= ~PAGE_CACHE_MASK;
1078 prev_offset = offset;
1080 page_cache_release(page);
1081 if (ret == nr && desc->count)
1082 continue;
1083 goto out;
1085 page_not_up_to_date:
1086 /* Get exclusive access to the page ... */
1087 error = lock_page_killable(page);
1088 if (unlikely(error))
1089 goto readpage_error;
1091 page_not_up_to_date_locked:
1092 /* Did it get truncated before we got the lock? */
1093 if (!page->mapping) {
1094 unlock_page(page);
1095 page_cache_release(page);
1096 continue;
1099 /* Did somebody else fill it already? */
1100 if (PageUptodate(page)) {
1101 unlock_page(page);
1102 goto page_ok;
1105 readpage:
1107 * A previous I/O error may have been due to temporary
1108 * failures, eg. multipath errors.
1109 * PG_error will be set again if readpage fails.
1111 ClearPageError(page);
1112 /* Start the actual read. The read will unlock the page. */
1113 error = mapping->a_ops->readpage(filp, page);
1115 if (unlikely(error)) {
1116 if (error == AOP_TRUNCATED_PAGE) {
1117 page_cache_release(page);
1118 goto find_page;
1120 goto readpage_error;
1123 if (!PageUptodate(page)) {
1124 error = lock_page_killable(page);
1125 if (unlikely(error))
1126 goto readpage_error;
1127 if (!PageUptodate(page)) {
1128 if (page->mapping == NULL) {
1130 * invalidate_mapping_pages got it
1132 unlock_page(page);
1133 page_cache_release(page);
1134 goto find_page;
1136 unlock_page(page);
1137 shrink_readahead_size_eio(filp, ra);
1138 error = -EIO;
1139 goto readpage_error;
1141 unlock_page(page);
1144 goto page_ok;
1146 readpage_error:
1147 /* UHHUH! A synchronous read error occurred. Report it */
1148 desc->error = error;
1149 page_cache_release(page);
1150 goto out;
1152 no_cached_page:
1154 * Ok, it wasn't cached, so we need to create a new
1155 * page..
1157 page = page_cache_alloc_cold(mapping);
1158 if (!page) {
1159 desc->error = -ENOMEM;
1160 goto out;
1162 error = add_to_page_cache_lru(page, mapping,
1163 index, GFP_KERNEL);
1164 if (error) {
1165 page_cache_release(page);
1166 if (error == -EEXIST)
1167 goto find_page;
1168 desc->error = error;
1169 goto out;
1171 goto readpage;
1174 out:
1175 ra->prev_pos = prev_index;
1176 ra->prev_pos <<= PAGE_CACHE_SHIFT;
1177 ra->prev_pos |= prev_offset;
1179 *ppos = ((loff_t)index << PAGE_CACHE_SHIFT) + offset;
1180 file_accessed(filp);
1183 int file_read_actor(read_descriptor_t *desc, struct page *page,
1184 unsigned long offset, unsigned long size)
1186 char *kaddr;
1187 unsigned long left, count = desc->count;
1189 if (size > count)
1190 size = count;
1193 * Faults on the destination of a read are common, so do it before
1194 * taking the kmap.
1196 if (!fault_in_pages_writeable(desc->arg.buf, size)) {
1197 kaddr = kmap_atomic(page, KM_USER0);
1198 left = __copy_to_user_inatomic(desc->arg.buf,
1199 kaddr + offset, size);
1200 kunmap_atomic(kaddr, KM_USER0);
1201 if (left == 0)
1202 goto success;
1205 /* Do it the slow way */
1206 kaddr = kmap(page);
1207 left = __copy_to_user(desc->arg.buf, kaddr + offset, size);
1208 kunmap(page);
1210 if (left) {
1211 size -= left;
1212 desc->error = -EFAULT;
1214 success:
1215 desc->count = count - size;
1216 desc->written += size;
1217 desc->arg.buf += size;
1218 return size;
1222 * Performs necessary checks before doing a write
1223 * @iov: io vector request
1224 * @nr_segs: number of segments in the iovec
1225 * @count: number of bytes to write
1226 * @access_flags: type of access: %VERIFY_READ or %VERIFY_WRITE
1228 * Adjust number of segments and amount of bytes to write (nr_segs should be
1229 * properly initialized first). Returns appropriate error code that caller
1230 * should return or zero in case that write should be allowed.
1232 int generic_segment_checks(const struct iovec *iov,
1233 unsigned long *nr_segs, size_t *count, int access_flags)
1235 unsigned long seg;
1236 size_t cnt = 0;
1237 for (seg = 0; seg < *nr_segs; seg++) {
1238 const struct iovec *iv = &iov[seg];
1241 * If any segment has a negative length, or the cumulative
1242 * length ever wraps negative then return -EINVAL.
1244 cnt += iv->iov_len;
1245 if (unlikely((ssize_t)(cnt|iv->iov_len) < 0))
1246 return -EINVAL;
1247 if (access_ok(access_flags, iv->iov_base, iv->iov_len))
1248 continue;
1249 if (seg == 0)
1250 return -EFAULT;
1251 *nr_segs = seg;
1252 cnt -= iv->iov_len; /* This segment is no good */
1253 break;
1255 *count = cnt;
1256 return 0;
1258 EXPORT_SYMBOL(generic_segment_checks);
1261 * generic_file_aio_read - generic filesystem read routine
1262 * @iocb: kernel I/O control block
1263 * @iov: io vector request
1264 * @nr_segs: number of segments in the iovec
1265 * @pos: current file position
1267 * This is the "read()" routine for all filesystems
1268 * that can use the page cache directly.
1270 ssize_t
1271 generic_file_aio_read(struct kiocb *iocb, const struct iovec *iov,
1272 unsigned long nr_segs, loff_t pos)
1274 struct file *filp = iocb->ki_filp;
1275 ssize_t retval;
1276 unsigned long seg = 0;
1277 size_t count;
1278 loff_t *ppos = &iocb->ki_pos;
1280 count = 0;
1281 retval = generic_segment_checks(iov, &nr_segs, &count, VERIFY_WRITE);
1282 if (retval)
1283 return retval;
1285 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */
1286 if (filp->f_flags & O_DIRECT) {
1287 loff_t size;
1288 struct address_space *mapping;
1289 struct inode *inode;
1291 mapping = filp->f_mapping;
1292 inode = mapping->host;
1293 if (!count)
1294 goto out; /* skip atime */
1295 size = i_size_read(inode);
1296 if (pos < size) {
1297 retval = filemap_write_and_wait_range(mapping, pos,
1298 pos + iov_length(iov, nr_segs) - 1);
1299 if (!retval) {
1300 retval = mapping->a_ops->direct_IO(READ, iocb,
1301 iov, pos, nr_segs);
1303 if (retval > 0) {
1304 *ppos = pos + retval;
1305 count -= retval;
1309 * Btrfs can have a short DIO read if we encounter
1310 * compressed extents, so if there was an error, or if
1311 * we've already read everything we wanted to, or if
1312 * there was a short read because we hit EOF, go ahead
1313 * and return. Otherwise fallthrough to buffered io for
1314 * the rest of the read.
1316 if (retval < 0 || !count || *ppos >= size) {
1317 file_accessed(filp);
1318 goto out;
1323 count = retval;
1324 for (seg = 0; seg < nr_segs; seg++) {
1325 read_descriptor_t desc;
1326 loff_t offset = 0;
1329 * If we did a short DIO read we need to skip the section of the
1330 * iov that we've already read data into.
1332 if (count) {
1333 if (count > iov[seg].iov_len) {
1334 count -= iov[seg].iov_len;
1335 continue;
1337 offset = count;
1338 count = 0;
1341 desc.written = 0;
1342 desc.arg.buf = iov[seg].iov_base + offset;
1343 desc.count = iov[seg].iov_len - offset;
1344 if (desc.count == 0)
1345 continue;
1346 desc.error = 0;
1347 do_generic_file_read(filp, ppos, &desc, file_read_actor);
1348 retval += desc.written;
1349 if (desc.error) {
1350 retval = retval ?: desc.error;
1351 break;
1353 if (desc.count > 0)
1354 break;
1356 out:
1357 return retval;
1359 EXPORT_SYMBOL(generic_file_aio_read);
1361 static ssize_t
1362 do_readahead(struct address_space *mapping, struct file *filp,
1363 pgoff_t index, unsigned long nr)
1365 if (!mapping || !mapping->a_ops || !mapping->a_ops->readpage)
1366 return -EINVAL;
1368 force_page_cache_readahead(mapping, filp, index, nr);
1369 return 0;
1372 SYSCALL_DEFINE(readahead)(int fd, loff_t offset, size_t count)
1374 ssize_t ret;
1375 struct file *file;
1377 ret = -EBADF;
1378 file = fget(fd);
1379 if (file) {
1380 if (file->f_mode & FMODE_READ) {
1381 struct address_space *mapping = file->f_mapping;
1382 pgoff_t start = offset >> PAGE_CACHE_SHIFT;
1383 pgoff_t end = (offset + count - 1) >> PAGE_CACHE_SHIFT;
1384 unsigned long len = end - start + 1;
1385 ret = do_readahead(mapping, file, start, len);
1387 fput(file);
1389 return ret;
1391 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
1392 asmlinkage long SyS_readahead(long fd, loff_t offset, long count)
1394 return SYSC_readahead((int) fd, offset, (size_t) count);
1396 SYSCALL_ALIAS(sys_readahead, SyS_readahead);
1397 #endif
1399 #ifdef CONFIG_MMU
1401 * page_cache_read - adds requested page to the page cache if not already there
1402 * @file: file to read
1403 * @offset: page index
1405 * This adds the requested page to the page cache if it isn't already there,
1406 * and schedules an I/O to read in its contents from disk.
1408 static int page_cache_read(struct file *file, pgoff_t offset)
1410 struct address_space *mapping = file->f_mapping;
1411 struct page *page;
1412 int ret;
1414 do {
1415 page = page_cache_alloc_cold(mapping);
1416 if (!page)
1417 return -ENOMEM;
1419 ret = add_to_page_cache_lru(page, mapping, offset, GFP_KERNEL);
1420 if (ret == 0)
1421 ret = mapping->a_ops->readpage(file, page);
1422 else if (ret == -EEXIST)
1423 ret = 0; /* losing race to add is OK */
1425 page_cache_release(page);
1427 } while (ret == AOP_TRUNCATED_PAGE);
1429 return ret;
1432 #define MMAP_LOTSAMISS (100)
1435 * Synchronous readahead happens when we don't even find
1436 * a page in the page cache at all.
1438 static void do_sync_mmap_readahead(struct vm_area_struct *vma,
1439 struct file_ra_state *ra,
1440 struct file *file,
1441 pgoff_t offset)
1443 unsigned long ra_pages;
1444 struct address_space *mapping = file->f_mapping;
1446 /* If we don't want any read-ahead, don't bother */
1447 if (VM_RandomReadHint(vma))
1448 return;
1450 if (VM_SequentialReadHint(vma) ||
1451 offset - 1 == (ra->prev_pos >> PAGE_CACHE_SHIFT)) {
1452 page_cache_sync_readahead(mapping, ra, file, offset,
1453 ra->ra_pages);
1454 return;
1457 if (ra->mmap_miss < INT_MAX)
1458 ra->mmap_miss++;
1461 * Do we miss much more than hit in this file? If so,
1462 * stop bothering with read-ahead. It will only hurt.
1464 if (ra->mmap_miss > MMAP_LOTSAMISS)
1465 return;
1468 * mmap read-around
1470 ra_pages = max_sane_readahead(ra->ra_pages);
1471 if (ra_pages) {
1472 ra->start = max_t(long, 0, offset - ra_pages/2);
1473 ra->size = ra_pages;
1474 ra->async_size = 0;
1475 ra_submit(ra, mapping, file);
1480 * Asynchronous readahead happens when we find the page and PG_readahead,
1481 * so we want to possibly extend the readahead further..
1483 static void do_async_mmap_readahead(struct vm_area_struct *vma,
1484 struct file_ra_state *ra,
1485 struct file *file,
1486 struct page *page,
1487 pgoff_t offset)
1489 struct address_space *mapping = file->f_mapping;
1491 /* If we don't want any read-ahead, don't bother */
1492 if (VM_RandomReadHint(vma))
1493 return;
1494 if (ra->mmap_miss > 0)
1495 ra->mmap_miss--;
1496 if (PageReadahead(page))
1497 page_cache_async_readahead(mapping, ra, file,
1498 page, offset, ra->ra_pages);
1502 * filemap_fault - read in file data for page fault handling
1503 * @vma: vma in which the fault was taken
1504 * @vmf: struct vm_fault containing details of the fault
1506 * filemap_fault() is invoked via the vma operations vector for a
1507 * mapped memory region to read in file data during a page fault.
1509 * The goto's are kind of ugly, but this streamlines the normal case of having
1510 * it in the page cache, and handles the special cases reasonably without
1511 * having a lot of duplicated code.
1513 int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1515 int error;
1516 struct file *file = vma->vm_file;
1517 struct address_space *mapping = file->f_mapping;
1518 struct file_ra_state *ra = &file->f_ra;
1519 struct inode *inode = mapping->host;
1520 pgoff_t offset = vmf->pgoff;
1521 struct page *page;
1522 pgoff_t size;
1523 int ret = 0;
1525 size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1526 if (offset >= size)
1527 return VM_FAULT_SIGBUS;
1530 * Do we have something in the page cache already?
1532 page = find_get_page(mapping, offset);
1533 if (likely(page)) {
1535 * We found the page, so try async readahead before
1536 * waiting for the lock.
1538 do_async_mmap_readahead(vma, ra, file, page, offset);
1539 lock_page(page);
1541 /* Did it get truncated? */
1542 if (unlikely(page->mapping != mapping)) {
1543 unlock_page(page);
1544 put_page(page);
1545 goto no_cached_page;
1547 } else {
1548 /* No page in the page cache at all */
1549 do_sync_mmap_readahead(vma, ra, file, offset);
1550 count_vm_event(PGMAJFAULT);
1551 ret = VM_FAULT_MAJOR;
1552 retry_find:
1553 page = find_lock_page(mapping, offset);
1554 if (!page)
1555 goto no_cached_page;
1559 * We have a locked page in the page cache, now we need to check
1560 * that it's up-to-date. If not, it is going to be due to an error.
1562 if (unlikely(!PageUptodate(page)))
1563 goto page_not_uptodate;
1566 * Found the page and have a reference on it.
1567 * We must recheck i_size under page lock.
1569 size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1570 if (unlikely(offset >= size)) {
1571 unlock_page(page);
1572 page_cache_release(page);
1573 return VM_FAULT_SIGBUS;
1576 ra->prev_pos = (loff_t)offset << PAGE_CACHE_SHIFT;
1577 vmf->page = page;
1578 return ret | VM_FAULT_LOCKED;
1580 no_cached_page:
1582 * We're only likely to ever get here if MADV_RANDOM is in
1583 * effect.
1585 error = page_cache_read(file, offset);
1588 * The page we want has now been added to the page cache.
1589 * In the unlikely event that someone removed it in the
1590 * meantime, we'll just come back here and read it again.
1592 if (error >= 0)
1593 goto retry_find;
1596 * An error return from page_cache_read can result if the
1597 * system is low on memory, or a problem occurs while trying
1598 * to schedule I/O.
1600 if (error == -ENOMEM)
1601 return VM_FAULT_OOM;
1602 return VM_FAULT_SIGBUS;
1604 page_not_uptodate:
1606 * Umm, take care of errors if the page isn't up-to-date.
1607 * Try to re-read it _once_. We do this synchronously,
1608 * because there really aren't any performance issues here
1609 * and we need to check for errors.
1611 ClearPageError(page);
1612 error = mapping->a_ops->readpage(file, page);
1613 if (!error) {
1614 wait_on_page_locked(page);
1615 if (!PageUptodate(page))
1616 error = -EIO;
1618 page_cache_release(page);
1620 if (!error || error == AOP_TRUNCATED_PAGE)
1621 goto retry_find;
1623 /* Things didn't work out. Return zero to tell the mm layer so. */
1624 shrink_readahead_size_eio(file, ra);
1625 return VM_FAULT_SIGBUS;
1627 EXPORT_SYMBOL(filemap_fault);
1629 const struct vm_operations_struct generic_file_vm_ops = {
1630 .fault = filemap_fault,
1633 /* This is used for a general mmap of a disk file */
1635 int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
1637 struct address_space *mapping = file->f_mapping;
1639 if (!mapping->a_ops->readpage)
1640 return -ENOEXEC;
1641 file_accessed(file);
1642 vma->vm_ops = &generic_file_vm_ops;
1643 vma->vm_flags |= VM_CAN_NONLINEAR;
1644 return 0;
1648 * This is for filesystems which do not implement ->writepage.
1650 int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
1652 if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE))
1653 return -EINVAL;
1654 return generic_file_mmap(file, vma);
1656 #else
1657 int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
1659 return -ENOSYS;
1661 int generic_file_readonly_mmap(struct file * file, struct vm_area_struct * vma)
1663 return -ENOSYS;
1665 #endif /* CONFIG_MMU */
1667 EXPORT_SYMBOL(generic_file_mmap);
1668 EXPORT_SYMBOL(generic_file_readonly_mmap);
1670 static struct page *__read_cache_page(struct address_space *mapping,
1671 pgoff_t index,
1672 int (*filler)(void *,struct page*),
1673 void *data,
1674 gfp_t gfp)
1676 struct page *page;
1677 int err;
1678 repeat:
1679 page = find_get_page(mapping, index);
1680 if (!page) {
1681 page = __page_cache_alloc(gfp | __GFP_COLD);
1682 if (!page)
1683 return ERR_PTR(-ENOMEM);
1684 err = add_to_page_cache_lru(page, mapping, index, GFP_KERNEL);
1685 if (unlikely(err)) {
1686 page_cache_release(page);
1687 if (err == -EEXIST)
1688 goto repeat;
1689 /* Presumably ENOMEM for radix tree node */
1690 return ERR_PTR(err);
1692 err = filler(data, page);
1693 if (err < 0) {
1694 page_cache_release(page);
1695 page = ERR_PTR(err);
1698 return page;
1701 static struct page *do_read_cache_page(struct address_space *mapping,
1702 pgoff_t index,
1703 int (*filler)(void *,struct page*),
1704 void *data,
1705 gfp_t gfp)
1708 struct page *page;
1709 int err;
1711 retry:
1712 page = __read_cache_page(mapping, index, filler, data, gfp);
1713 if (IS_ERR(page))
1714 return page;
1715 if (PageUptodate(page))
1716 goto out;
1718 lock_page(page);
1719 if (!page->mapping) {
1720 unlock_page(page);
1721 page_cache_release(page);
1722 goto retry;
1724 if (PageUptodate(page)) {
1725 unlock_page(page);
1726 goto out;
1728 err = filler(data, page);
1729 if (err < 0) {
1730 page_cache_release(page);
1731 return ERR_PTR(err);
1733 out:
1734 mark_page_accessed(page);
1735 return page;
1739 * read_cache_page_async - read into page cache, fill it if needed
1740 * @mapping: the page's address_space
1741 * @index: the page index
1742 * @filler: function to perform the read
1743 * @data: destination for read data
1745 * Same as read_cache_page, but don't wait for page to become unlocked
1746 * after submitting it to the filler.
1748 * Read into the page cache. If a page already exists, and PageUptodate() is
1749 * not set, try to fill the page but don't wait for it to become unlocked.
1751 * If the page does not get brought uptodate, return -EIO.
1753 struct page *read_cache_page_async(struct address_space *mapping,
1754 pgoff_t index,
1755 int (*filler)(void *,struct page*),
1756 void *data)
1758 return do_read_cache_page(mapping, index, filler, data, mapping_gfp_mask(mapping));
1760 EXPORT_SYMBOL(read_cache_page_async);
1762 static struct page *wait_on_page_read(struct page *page)
1764 if (!IS_ERR(page)) {
1765 wait_on_page_locked(page);
1766 if (!PageUptodate(page)) {
1767 page_cache_release(page);
1768 page = ERR_PTR(-EIO);
1771 return page;
1775 * read_cache_page_gfp - read into page cache, using specified page allocation flags.
1776 * @mapping: the page's address_space
1777 * @index: the page index
1778 * @gfp: the page allocator flags to use if allocating
1780 * This is the same as "read_mapping_page(mapping, index, NULL)", but with
1781 * any new page allocations done using the specified allocation flags. Note
1782 * that the Radix tree operations will still use GFP_KERNEL, so you can't
1783 * expect to do this atomically or anything like that - but you can pass in
1784 * other page requirements.
1786 * If the page does not get brought uptodate, return -EIO.
1788 struct page *read_cache_page_gfp(struct address_space *mapping,
1789 pgoff_t index,
1790 gfp_t gfp)
1792 filler_t *filler = (filler_t *)mapping->a_ops->readpage;
1794 return wait_on_page_read(do_read_cache_page(mapping, index, filler, NULL, gfp));
1796 EXPORT_SYMBOL(read_cache_page_gfp);
1799 * read_cache_page - read into page cache, fill it if needed
1800 * @mapping: the page's address_space
1801 * @index: the page index
1802 * @filler: function to perform the read
1803 * @data: destination for read data
1805 * Read into the page cache. If a page already exists, and PageUptodate() is
1806 * not set, try to fill the page then wait for it to become unlocked.
1808 * If the page does not get brought uptodate, return -EIO.
1810 struct page *read_cache_page(struct address_space *mapping,
1811 pgoff_t index,
1812 int (*filler)(void *,struct page*),
1813 void *data)
1815 return wait_on_page_read(read_cache_page_async(mapping, index, filler, data));
1817 EXPORT_SYMBOL(read_cache_page);
1820 * The logic we want is
1822 * if suid or (sgid and xgrp)
1823 * remove privs
1825 int should_remove_suid(struct dentry *dentry)
1827 mode_t mode = dentry->d_inode->i_mode;
1828 int kill = 0;
1830 /* suid always must be killed */
1831 if (unlikely(mode & S_ISUID))
1832 kill = ATTR_KILL_SUID;
1835 * sgid without any exec bits is just a mandatory locking mark; leave
1836 * it alone. If some exec bits are set, it's a real sgid; kill it.
1838 if (unlikely((mode & S_ISGID) && (mode & S_IXGRP)))
1839 kill |= ATTR_KILL_SGID;
1841 if (unlikely(kill && !capable(CAP_FSETID) && S_ISREG(mode)))
1842 return kill;
1844 return 0;
1846 EXPORT_SYMBOL(should_remove_suid);
1848 static int __remove_suid(struct dentry *dentry, int kill)
1850 struct iattr newattrs;
1852 newattrs.ia_valid = ATTR_FORCE | kill;
1853 return notify_change(dentry, &newattrs);
1856 int file_remove_suid(struct file *file)
1858 struct dentry *dentry = file->f_path.dentry;
1859 int killsuid = should_remove_suid(dentry);
1860 int killpriv = security_inode_need_killpriv(dentry);
1861 int error = 0;
1863 if (killpriv < 0)
1864 return killpriv;
1865 if (killpriv)
1866 error = security_inode_killpriv(dentry);
1867 if (!error && killsuid)
1868 error = __remove_suid(dentry, killsuid);
1870 return error;
1872 EXPORT_SYMBOL(file_remove_suid);
1874 static size_t __iovec_copy_from_user_inatomic(char *vaddr,
1875 const struct iovec *iov, size_t base, size_t bytes)
1877 size_t copied = 0, left = 0;
1879 while (bytes) {
1880 char __user *buf = iov->iov_base + base;
1881 int copy = min(bytes, iov->iov_len - base);
1883 base = 0;
1884 left = __copy_from_user_inatomic(vaddr, buf, copy);
1885 copied += copy;
1886 bytes -= copy;
1887 vaddr += copy;
1888 iov++;
1890 if (unlikely(left))
1891 break;
1893 return copied - left;
1897 * Copy as much as we can into the page and return the number of bytes which
1898 * were successfully copied. If a fault is encountered then return the number of
1899 * bytes which were copied.
1901 size_t iov_iter_copy_from_user_atomic(struct page *page,
1902 struct iov_iter *i, unsigned long offset, size_t bytes)
1904 char *kaddr;
1905 size_t copied;
1907 BUG_ON(!in_atomic());
1908 kaddr = kmap_atomic(page, KM_USER0);
1909 if (likely(i->nr_segs == 1)) {
1910 int left;
1911 char __user *buf = i->iov->iov_base + i->iov_offset;
1912 left = __copy_from_user_inatomic(kaddr + offset, buf, bytes);
1913 copied = bytes - left;
1914 } else {
1915 copied = __iovec_copy_from_user_inatomic(kaddr + offset,
1916 i->iov, i->iov_offset, bytes);
1918 kunmap_atomic(kaddr, KM_USER0);
1920 return copied;
1922 EXPORT_SYMBOL(iov_iter_copy_from_user_atomic);
1925 * This has the same sideeffects and return value as
1926 * iov_iter_copy_from_user_atomic().
1927 * The difference is that it attempts to resolve faults.
1928 * Page must not be locked.
1930 size_t iov_iter_copy_from_user(struct page *page,
1931 struct iov_iter *i, unsigned long offset, size_t bytes)
1933 char *kaddr;
1934 size_t copied;
1936 kaddr = kmap(page);
1937 if (likely(i->nr_segs == 1)) {
1938 int left;
1939 char __user *buf = i->iov->iov_base + i->iov_offset;
1940 left = __copy_from_user(kaddr + offset, buf, bytes);
1941 copied = bytes - left;
1942 } else {
1943 copied = __iovec_copy_from_user_inatomic(kaddr + offset,
1944 i->iov, i->iov_offset, bytes);
1946 kunmap(page);
1947 return copied;
1949 EXPORT_SYMBOL(iov_iter_copy_from_user);
1951 void iov_iter_advance(struct iov_iter *i, size_t bytes)
1953 BUG_ON(i->count < bytes);
1955 if (likely(i->nr_segs == 1)) {
1956 i->iov_offset += bytes;
1957 i->count -= bytes;
1958 } else {
1959 const struct iovec *iov = i->iov;
1960 size_t base = i->iov_offset;
1963 * The !iov->iov_len check ensures we skip over unlikely
1964 * zero-length segments (without overruning the iovec).
1966 while (bytes || unlikely(i->count && !iov->iov_len)) {
1967 int copy;
1969 copy = min(bytes, iov->iov_len - base);
1970 BUG_ON(!i->count || i->count < copy);
1971 i->count -= copy;
1972 bytes -= copy;
1973 base += copy;
1974 if (iov->iov_len == base) {
1975 iov++;
1976 base = 0;
1979 i->iov = iov;
1980 i->iov_offset = base;
1983 EXPORT_SYMBOL(iov_iter_advance);
1986 * Fault in the first iovec of the given iov_iter, to a maximum length
1987 * of bytes. Returns 0 on success, or non-zero if the memory could not be
1988 * accessed (ie. because it is an invalid address).
1990 * writev-intensive code may want this to prefault several iovecs -- that
1991 * would be possible (callers must not rely on the fact that _only_ the
1992 * first iovec will be faulted with the current implementation).
1994 int iov_iter_fault_in_readable(struct iov_iter *i, size_t bytes)
1996 char __user *buf = i->iov->iov_base + i->iov_offset;
1997 bytes = min(bytes, i->iov->iov_len - i->iov_offset);
1998 return fault_in_pages_readable(buf, bytes);
2000 EXPORT_SYMBOL(iov_iter_fault_in_readable);
2003 * Return the count of just the current iov_iter segment.
2005 size_t iov_iter_single_seg_count(struct iov_iter *i)
2007 const struct iovec *iov = i->iov;
2008 if (i->nr_segs == 1)
2009 return i->count;
2010 else
2011 return min(i->count, iov->iov_len - i->iov_offset);
2013 EXPORT_SYMBOL(iov_iter_single_seg_count);
2016 * Performs necessary checks before doing a write
2018 * Can adjust writing position or amount of bytes to write.
2019 * Returns appropriate error code that caller should return or
2020 * zero in case that write should be allowed.
2022 inline int generic_write_checks(struct file *file, loff_t *pos, size_t *count, int isblk)
2024 struct inode *inode = file->f_mapping->host;
2025 unsigned long limit = rlimit(RLIMIT_FSIZE);
2027 if (unlikely(*pos < 0))
2028 return -EINVAL;
2030 if (!isblk) {
2031 /* FIXME: this is for backwards compatibility with 2.4 */
2032 if (file->f_flags & O_APPEND)
2033 *pos = i_size_read(inode);
2035 if (limit != RLIM_INFINITY) {
2036 if (*pos >= limit) {
2037 send_sig(SIGXFSZ, current, 0);
2038 return -EFBIG;
2040 if (*count > limit - (typeof(limit))*pos) {
2041 *count = limit - (typeof(limit))*pos;
2047 * LFS rule
2049 if (unlikely(*pos + *count > MAX_NON_LFS &&
2050 !(file->f_flags & O_LARGEFILE))) {
2051 if (*pos >= MAX_NON_LFS) {
2052 return -EFBIG;
2054 if (*count > MAX_NON_LFS - (unsigned long)*pos) {
2055 *count = MAX_NON_LFS - (unsigned long)*pos;
2060 * Are we about to exceed the fs block limit ?
2062 * If we have written data it becomes a short write. If we have
2063 * exceeded without writing data we send a signal and return EFBIG.
2064 * Linus frestrict idea will clean these up nicely..
2066 if (likely(!isblk)) {
2067 if (unlikely(*pos >= inode->i_sb->s_maxbytes)) {
2068 if (*count || *pos > inode->i_sb->s_maxbytes) {
2069 return -EFBIG;
2071 /* zero-length writes at ->s_maxbytes are OK */
2074 if (unlikely(*pos + *count > inode->i_sb->s_maxbytes))
2075 *count = inode->i_sb->s_maxbytes - *pos;
2076 } else {
2077 #ifdef CONFIG_BLOCK
2078 loff_t isize;
2079 if (bdev_read_only(I_BDEV(inode)))
2080 return -EPERM;
2081 isize = i_size_read(inode);
2082 if (*pos >= isize) {
2083 if (*count || *pos > isize)
2084 return -ENOSPC;
2087 if (*pos + *count > isize)
2088 *count = isize - *pos;
2089 #else
2090 return -EPERM;
2091 #endif
2093 return 0;
2095 EXPORT_SYMBOL(generic_write_checks);
2097 int pagecache_write_begin(struct file *file, struct address_space *mapping,
2098 loff_t pos, unsigned len, unsigned flags,
2099 struct page **pagep, void **fsdata)
2101 const struct address_space_operations *aops = mapping->a_ops;
2103 return aops->write_begin(file, mapping, pos, len, flags,
2104 pagep, fsdata);
2106 EXPORT_SYMBOL(pagecache_write_begin);
2108 int pagecache_write_end(struct file *file, struct address_space *mapping,
2109 loff_t pos, unsigned len, unsigned copied,
2110 struct page *page, void *fsdata)
2112 const struct address_space_operations *aops = mapping->a_ops;
2114 mark_page_accessed(page);
2115 return aops->write_end(file, mapping, pos, len, copied, page, fsdata);
2117 EXPORT_SYMBOL(pagecache_write_end);
2119 ssize_t
2120 generic_file_direct_write(struct kiocb *iocb, const struct iovec *iov,
2121 unsigned long *nr_segs, loff_t pos, loff_t *ppos,
2122 size_t count, size_t ocount)
2124 struct file *file = iocb->ki_filp;
2125 struct address_space *mapping = file->f_mapping;
2126 struct inode *inode = mapping->host;
2127 ssize_t written;
2128 size_t write_len;
2129 pgoff_t end;
2131 if (count != ocount)
2132 *nr_segs = iov_shorten((struct iovec *)iov, *nr_segs, count);
2134 write_len = iov_length(iov, *nr_segs);
2135 end = (pos + write_len - 1) >> PAGE_CACHE_SHIFT;
2137 written = filemap_write_and_wait_range(mapping, pos, pos + write_len - 1);
2138 if (written)
2139 goto out;
2142 * After a write we want buffered reads to be sure to go to disk to get
2143 * the new data. We invalidate clean cached page from the region we're
2144 * about to write. We do this *before* the write so that we can return
2145 * without clobbering -EIOCBQUEUED from ->direct_IO().
2147 if (mapping->nrpages) {
2148 written = invalidate_inode_pages2_range(mapping,
2149 pos >> PAGE_CACHE_SHIFT, end);
2151 * If a page can not be invalidated, return 0 to fall back
2152 * to buffered write.
2154 if (written) {
2155 if (written == -EBUSY)
2156 return 0;
2157 goto out;
2161 written = mapping->a_ops->direct_IO(WRITE, iocb, iov, pos, *nr_segs);
2164 * Finally, try again to invalidate clean pages which might have been
2165 * cached by non-direct readahead, or faulted in by get_user_pages()
2166 * if the source of the write was an mmap'ed region of the file
2167 * we're writing. Either one is a pretty crazy thing to do,
2168 * so we don't support it 100%. If this invalidation
2169 * fails, tough, the write still worked...
2171 if (mapping->nrpages) {
2172 invalidate_inode_pages2_range(mapping,
2173 pos >> PAGE_CACHE_SHIFT, end);
2176 if (written > 0) {
2177 loff_t end = pos + written;
2178 if (end > i_size_read(inode) && !S_ISBLK(inode->i_mode)) {
2179 i_size_write(inode, end);
2180 mark_inode_dirty(inode);
2182 *ppos = end;
2184 out:
2185 return written;
2187 EXPORT_SYMBOL(generic_file_direct_write);
2190 * Find or create a page at the given pagecache position. Return the locked
2191 * page. This function is specifically for buffered writes.
2193 struct page *grab_cache_page_write_begin(struct address_space *mapping,
2194 pgoff_t index, unsigned flags)
2196 int status;
2197 struct page *page;
2198 gfp_t gfp_notmask = 0;
2199 if (flags & AOP_FLAG_NOFS)
2200 gfp_notmask = __GFP_FS;
2201 repeat:
2202 page = find_lock_page(mapping, index);
2203 if (likely(page))
2204 return page;
2206 page = __page_cache_alloc(mapping_gfp_mask(mapping) & ~gfp_notmask);
2207 if (!page)
2208 return NULL;
2209 status = add_to_page_cache_lru(page, mapping, index,
2210 GFP_KERNEL & ~gfp_notmask);
2211 if (unlikely(status)) {
2212 page_cache_release(page);
2213 if (status == -EEXIST)
2214 goto repeat;
2215 return NULL;
2217 return page;
2219 EXPORT_SYMBOL(grab_cache_page_write_begin);
2221 static ssize_t generic_perform_write(struct file *file,
2222 struct iov_iter *i, loff_t pos)
2224 struct address_space *mapping = file->f_mapping;
2225 const struct address_space_operations *a_ops = mapping->a_ops;
2226 long status = 0;
2227 ssize_t written = 0;
2228 unsigned int flags = 0;
2231 * Copies from kernel address space cannot fail (NFSD is a big user).
2233 if (segment_eq(get_fs(), KERNEL_DS))
2234 flags |= AOP_FLAG_UNINTERRUPTIBLE;
2236 do {
2237 struct page *page;
2238 unsigned long offset; /* Offset into pagecache page */
2239 unsigned long bytes; /* Bytes to write to page */
2240 size_t copied; /* Bytes copied from user */
2241 void *fsdata;
2243 offset = (pos & (PAGE_CACHE_SIZE - 1));
2244 bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset,
2245 iov_iter_count(i));
2247 again:
2250 * Bring in the user page that we will copy from _first_.
2251 * Otherwise there's a nasty deadlock on copying from the
2252 * same page as we're writing to, without it being marked
2253 * up-to-date.
2255 * Not only is this an optimisation, but it is also required
2256 * to check that the address is actually valid, when atomic
2257 * usercopies are used, below.
2259 if (unlikely(iov_iter_fault_in_readable(i, bytes))) {
2260 status = -EFAULT;
2261 break;
2264 status = a_ops->write_begin(file, mapping, pos, bytes, flags,
2265 &page, &fsdata);
2266 if (unlikely(status))
2267 break;
2269 if (mapping_writably_mapped(mapping))
2270 flush_dcache_page(page);
2272 pagefault_disable();
2273 copied = iov_iter_copy_from_user_atomic(page, i, offset, bytes);
2274 pagefault_enable();
2275 flush_dcache_page(page);
2277 mark_page_accessed(page);
2278 status = a_ops->write_end(file, mapping, pos, bytes, copied,
2279 page, fsdata);
2280 if (unlikely(status < 0))
2281 break;
2282 copied = status;
2284 cond_resched();
2286 iov_iter_advance(i, copied);
2287 if (unlikely(copied == 0)) {
2289 * If we were unable to copy any data at all, we must
2290 * fall back to a single segment length write.
2292 * If we didn't fallback here, we could livelock
2293 * because not all segments in the iov can be copied at
2294 * once without a pagefault.
2296 bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset,
2297 iov_iter_single_seg_count(i));
2298 goto again;
2300 pos += copied;
2301 written += copied;
2303 balance_dirty_pages_ratelimited(mapping);
2305 } while (iov_iter_count(i));
2307 return written ? written : status;
2310 ssize_t
2311 generic_file_buffered_write(struct kiocb *iocb, const struct iovec *iov,
2312 unsigned long nr_segs, loff_t pos, loff_t *ppos,
2313 size_t count, ssize_t written)
2315 struct file *file = iocb->ki_filp;
2316 ssize_t status;
2317 struct iov_iter i;
2319 iov_iter_init(&i, iov, nr_segs, count, written);
2320 status = generic_perform_write(file, &i, pos);
2322 if (likely(status >= 0)) {
2323 written += status;
2324 *ppos = pos + status;
2327 return written ? written : status;
2329 EXPORT_SYMBOL(generic_file_buffered_write);
2332 * __generic_file_aio_write - write data to a file
2333 * @iocb: IO state structure (file, offset, etc.)
2334 * @iov: vector with data to write
2335 * @nr_segs: number of segments in the vector
2336 * @ppos: position where to write
2338 * This function does all the work needed for actually writing data to a
2339 * file. It does all basic checks, removes SUID from the file, updates
2340 * modification times and calls proper subroutines depending on whether we
2341 * do direct IO or a standard buffered write.
2343 * It expects i_mutex to be grabbed unless we work on a block device or similar
2344 * object which does not need locking at all.
2346 * This function does *not* take care of syncing data in case of O_SYNC write.
2347 * A caller has to handle it. This is mainly due to the fact that we want to
2348 * avoid syncing under i_mutex.
2350 ssize_t __generic_file_aio_write(struct kiocb *iocb, const struct iovec *iov,
2351 unsigned long nr_segs, loff_t *ppos)
2353 struct file *file = iocb->ki_filp;
2354 struct address_space * mapping = file->f_mapping;
2355 size_t ocount; /* original count */
2356 size_t count; /* after file limit checks */
2357 struct inode *inode = mapping->host;
2358 loff_t pos;
2359 ssize_t written;
2360 ssize_t err;
2362 ocount = 0;
2363 err = generic_segment_checks(iov, &nr_segs, &ocount, VERIFY_READ);
2364 if (err)
2365 return err;
2367 count = ocount;
2368 pos = *ppos;
2370 vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
2372 /* We can write back this queue in page reclaim */
2373 current->backing_dev_info = mapping->backing_dev_info;
2374 written = 0;
2376 err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
2377 if (err)
2378 goto out;
2380 if (count == 0)
2381 goto out;
2383 err = file_remove_suid(file);
2384 if (err)
2385 goto out;
2387 file_update_time(file);
2389 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */
2390 if (unlikely(file->f_flags & O_DIRECT)) {
2391 loff_t endbyte;
2392 ssize_t written_buffered;
2394 written = generic_file_direct_write(iocb, iov, &nr_segs, pos,
2395 ppos, count, ocount);
2396 if (written < 0 || written == count)
2397 goto out;
2399 * direct-io write to a hole: fall through to buffered I/O
2400 * for completing the rest of the request.
2402 pos += written;
2403 count -= written;
2404 written_buffered = generic_file_buffered_write(iocb, iov,
2405 nr_segs, pos, ppos, count,
2406 written);
2408 * If generic_file_buffered_write() retuned a synchronous error
2409 * then we want to return the number of bytes which were
2410 * direct-written, or the error code if that was zero. Note
2411 * that this differs from normal direct-io semantics, which
2412 * will return -EFOO even if some bytes were written.
2414 if (written_buffered < 0) {
2415 err = written_buffered;
2416 goto out;
2420 * We need to ensure that the page cache pages are written to
2421 * disk and invalidated to preserve the expected O_DIRECT
2422 * semantics.
2424 endbyte = pos + written_buffered - written - 1;
2425 err = filemap_write_and_wait_range(file->f_mapping, pos, endbyte);
2426 if (err == 0) {
2427 written = written_buffered;
2428 invalidate_mapping_pages(mapping,
2429 pos >> PAGE_CACHE_SHIFT,
2430 endbyte >> PAGE_CACHE_SHIFT);
2431 } else {
2433 * We don't know how much we wrote, so just return
2434 * the number of bytes which were direct-written
2437 } else {
2438 written = generic_file_buffered_write(iocb, iov, nr_segs,
2439 pos, ppos, count, written);
2441 out:
2442 current->backing_dev_info = NULL;
2443 return written ? written : err;
2445 EXPORT_SYMBOL(__generic_file_aio_write);
2448 * generic_file_aio_write - write data to a file
2449 * @iocb: IO state structure
2450 * @iov: vector with data to write
2451 * @nr_segs: number of segments in the vector
2452 * @pos: position in file where to write
2454 * This is a wrapper around __generic_file_aio_write() to be used by most
2455 * filesystems. It takes care of syncing the file in case of O_SYNC file
2456 * and acquires i_mutex as needed.
2458 ssize_t generic_file_aio_write(struct kiocb *iocb, const struct iovec *iov,
2459 unsigned long nr_segs, loff_t pos)
2461 struct file *file = iocb->ki_filp;
2462 struct inode *inode = file->f_mapping->host;
2463 ssize_t ret;
2465 BUG_ON(iocb->ki_pos != pos);
2467 mutex_lock(&inode->i_mutex);
2468 ret = __generic_file_aio_write(iocb, iov, nr_segs, &iocb->ki_pos);
2469 mutex_unlock(&inode->i_mutex);
2471 if (ret > 0 || ret == -EIOCBQUEUED) {
2472 ssize_t err;
2474 err = generic_write_sync(file, pos, ret);
2475 if (err < 0 && ret > 0)
2476 ret = err;
2478 return ret;
2480 EXPORT_SYMBOL(generic_file_aio_write);
2483 * try_to_release_page() - release old fs-specific metadata on a page
2485 * @page: the page which the kernel is trying to free
2486 * @gfp_mask: memory allocation flags (and I/O mode)
2488 * The address_space is to try to release any data against the page
2489 * (presumably at page->private). If the release was successful, return `1'.
2490 * Otherwise return zero.
2492 * This may also be called if PG_fscache is set on a page, indicating that the
2493 * page is known to the local caching routines.
2495 * The @gfp_mask argument specifies whether I/O may be performed to release
2496 * this page (__GFP_IO), and whether the call may block (__GFP_WAIT & __GFP_FS).
2499 int try_to_release_page(struct page *page, gfp_t gfp_mask)
2501 struct address_space * const mapping = page->mapping;
2503 BUG_ON(!PageLocked(page));
2504 if (PageWriteback(page))
2505 return 0;
2507 if (mapping && mapping->a_ops->releasepage)
2508 return mapping->a_ops->releasepage(page, gfp_mask);
2509 return try_to_free_buffers(page);
2512 EXPORT_SYMBOL(try_to_release_page);