[PATCH] I2C: Kill i2c_algorithm.id (6/7)
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / mm / filemap.c
blob88611928e71fc928a89b5b319044d759ea5f612d
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/config.h>
13 #include <linux/module.h>
14 #include <linux/slab.h>
15 #include <linux/compiler.h>
16 #include <linux/fs.h>
17 #include <linux/aio.h>
18 #include <linux/kernel_stat.h>
19 #include <linux/mm.h>
20 #include <linux/swap.h>
21 #include <linux/mman.h>
22 #include <linux/pagemap.h>
23 #include <linux/file.h>
24 #include <linux/uio.h>
25 #include <linux/hash.h>
26 #include <linux/writeback.h>
27 #include <linux/pagevec.h>
28 #include <linux/blkdev.h>
29 #include <linux/security.h>
30 #include <linux/syscalls.h>
31 #include "filemap.h"
33 * FIXME: remove all knowledge of the buffer layer from the core VM
35 #include <linux/buffer_head.h> /* for generic_osync_inode */
37 #include <asm/uaccess.h>
38 #include <asm/mman.h>
41 * Shared mappings implemented 30.11.1994. It's not fully working yet,
42 * though.
44 * Shared mappings now work. 15.8.1995 Bruno.
46 * finished 'unifying' the page and buffer cache and SMP-threaded the
47 * page-cache, 21.05.1999, Ingo Molnar <mingo@redhat.com>
49 * SMP-threaded pagemap-LRU 1999, Andrea Arcangeli <andrea@suse.de>
53 * Lock ordering:
55 * ->i_mmap_lock (vmtruncate)
56 * ->private_lock (__free_pte->__set_page_dirty_buffers)
57 * ->swap_lock (exclusive_swap_page, others)
58 * ->mapping->tree_lock
60 * ->i_sem
61 * ->i_mmap_lock (truncate->unmap_mapping_range)
63 * ->mmap_sem
64 * ->i_mmap_lock
65 * ->page_table_lock (various places, mainly in mmap.c)
66 * ->mapping->tree_lock (arch-dependent flush_dcache_mmap_lock)
68 * ->mmap_sem
69 * ->lock_page (access_process_vm)
71 * ->mmap_sem
72 * ->i_sem (msync)
74 * ->i_sem
75 * ->i_alloc_sem (various)
77 * ->inode_lock
78 * ->sb_lock (fs/fs-writeback.c)
79 * ->mapping->tree_lock (__sync_single_inode)
81 * ->i_mmap_lock
82 * ->anon_vma.lock (vma_adjust)
84 * ->anon_vma.lock
85 * ->page_table_lock (anon_vma_prepare and various)
87 * ->page_table_lock
88 * ->swap_lock (try_to_unmap_one)
89 * ->private_lock (try_to_unmap_one)
90 * ->tree_lock (try_to_unmap_one)
91 * ->zone.lru_lock (follow_page->mark_page_accessed)
92 * ->private_lock (page_remove_rmap->set_page_dirty)
93 * ->tree_lock (page_remove_rmap->set_page_dirty)
94 * ->inode_lock (page_remove_rmap->set_page_dirty)
95 * ->inode_lock (zap_pte_range->set_page_dirty)
96 * ->private_lock (zap_pte_range->__set_page_dirty_buffers)
98 * ->task->proc_lock
99 * ->dcache_lock (proc_pid_lookup)
103 * Remove a page from the page cache and free it. Caller has to make
104 * sure the page is locked and that nobody else uses it - or that usage
105 * is safe. The caller must hold a write_lock on the mapping's tree_lock.
107 void __remove_from_page_cache(struct page *page)
109 struct address_space *mapping = page->mapping;
111 radix_tree_delete(&mapping->page_tree, page->index);
112 page->mapping = NULL;
113 mapping->nrpages--;
114 pagecache_acct(-1);
117 void remove_from_page_cache(struct page *page)
119 struct address_space *mapping = page->mapping;
121 BUG_ON(!PageLocked(page));
123 write_lock_irq(&mapping->tree_lock);
124 __remove_from_page_cache(page);
125 write_unlock_irq(&mapping->tree_lock);
128 static int sync_page(void *word)
130 struct address_space *mapping;
131 struct page *page;
133 page = container_of((page_flags_t *)word, struct page, flags);
136 * page_mapping() is being called without PG_locked held.
137 * Some knowledge of the state and use of the page is used to
138 * reduce the requirements down to a memory barrier.
139 * The danger here is of a stale page_mapping() return value
140 * indicating a struct address_space different from the one it's
141 * associated with when it is associated with one.
142 * After smp_mb(), it's either the correct page_mapping() for
143 * the page, or an old page_mapping() and the page's own
144 * page_mapping() has gone NULL.
145 * The ->sync_page() address_space operation must tolerate
146 * page_mapping() going NULL. By an amazing coincidence,
147 * this comes about because none of the users of the page
148 * in the ->sync_page() methods make essential use of the
149 * page_mapping(), merely passing the page down to the backing
150 * device's unplug functions when it's non-NULL, which in turn
151 * ignore it for all cases but swap, where only page->private is
152 * of interest. When page_mapping() does go NULL, the entire
153 * call stack gracefully ignores the page and returns.
154 * -- wli
156 smp_mb();
157 mapping = page_mapping(page);
158 if (mapping && mapping->a_ops && mapping->a_ops->sync_page)
159 mapping->a_ops->sync_page(page);
160 io_schedule();
161 return 0;
165 * filemap_fdatawrite_range - start writeback against all of a mapping's
166 * dirty pages that lie within the byte offsets <start, end>
167 * @mapping: address space structure to write
168 * @start: offset in bytes where the range starts
169 * @end: offset in bytes where the range ends
170 * @sync_mode: enable synchronous operation
172 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
173 * opposed to a regular memory * cleansing writeback. The difference between
174 * these two operations is that if a dirty page/buffer is encountered, it must
175 * be waited upon, and not just skipped over.
177 static int __filemap_fdatawrite_range(struct address_space *mapping,
178 loff_t start, loff_t end, int sync_mode)
180 int ret;
181 struct writeback_control wbc = {
182 .sync_mode = sync_mode,
183 .nr_to_write = mapping->nrpages * 2,
184 .start = start,
185 .end = end,
188 if (!mapping_cap_writeback_dirty(mapping))
189 return 0;
191 ret = do_writepages(mapping, &wbc);
192 return ret;
195 static inline int __filemap_fdatawrite(struct address_space *mapping,
196 int sync_mode)
198 return __filemap_fdatawrite_range(mapping, 0, 0, sync_mode);
201 int filemap_fdatawrite(struct address_space *mapping)
203 return __filemap_fdatawrite(mapping, WB_SYNC_ALL);
205 EXPORT_SYMBOL(filemap_fdatawrite);
207 static int filemap_fdatawrite_range(struct address_space *mapping,
208 loff_t start, loff_t end)
210 return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL);
214 * This is a mostly non-blocking flush. Not suitable for data-integrity
215 * purposes - I/O may not be started against all dirty pages.
217 int filemap_flush(struct address_space *mapping)
219 return __filemap_fdatawrite(mapping, WB_SYNC_NONE);
221 EXPORT_SYMBOL(filemap_flush);
224 * Wait for writeback to complete against pages indexed by start->end
225 * inclusive
227 static int wait_on_page_writeback_range(struct address_space *mapping,
228 pgoff_t start, pgoff_t end)
230 struct pagevec pvec;
231 int nr_pages;
232 int ret = 0;
233 pgoff_t index;
235 if (end < start)
236 return 0;
238 pagevec_init(&pvec, 0);
239 index = start;
240 while ((index <= end) &&
241 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
242 PAGECACHE_TAG_WRITEBACK,
243 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1)) != 0) {
244 unsigned i;
246 for (i = 0; i < nr_pages; i++) {
247 struct page *page = pvec.pages[i];
249 /* until radix tree lookup accepts end_index */
250 if (page->index > end)
251 continue;
253 wait_on_page_writeback(page);
254 if (PageError(page))
255 ret = -EIO;
257 pagevec_release(&pvec);
258 cond_resched();
261 /* Check for outstanding write errors */
262 if (test_and_clear_bit(AS_ENOSPC, &mapping->flags))
263 ret = -ENOSPC;
264 if (test_and_clear_bit(AS_EIO, &mapping->flags))
265 ret = -EIO;
267 return ret;
271 * Write and wait upon all the pages in the passed range. This is a "data
272 * integrity" operation. It waits upon in-flight writeout before starting and
273 * waiting upon new writeout. If there was an IO error, return it.
275 * We need to re-take i_sem during the generic_osync_inode list walk because
276 * it is otherwise livelockable.
278 int sync_page_range(struct inode *inode, struct address_space *mapping,
279 loff_t pos, size_t count)
281 pgoff_t start = pos >> PAGE_CACHE_SHIFT;
282 pgoff_t end = (pos + count - 1) >> PAGE_CACHE_SHIFT;
283 int ret;
285 if (!mapping_cap_writeback_dirty(mapping) || !count)
286 return 0;
287 ret = filemap_fdatawrite_range(mapping, pos, pos + count - 1);
288 if (ret == 0) {
289 down(&inode->i_sem);
290 ret = generic_osync_inode(inode, mapping, OSYNC_METADATA);
291 up(&inode->i_sem);
293 if (ret == 0)
294 ret = wait_on_page_writeback_range(mapping, start, end);
295 return ret;
297 EXPORT_SYMBOL(sync_page_range);
300 * Note: Holding i_sem across sync_page_range_nolock is not a good idea
301 * as it forces O_SYNC writers to different parts of the same file
302 * to be serialised right until io completion.
304 int sync_page_range_nolock(struct inode *inode, struct address_space *mapping,
305 loff_t pos, size_t count)
307 pgoff_t start = pos >> PAGE_CACHE_SHIFT;
308 pgoff_t end = (pos + count - 1) >> PAGE_CACHE_SHIFT;
309 int ret;
311 if (!mapping_cap_writeback_dirty(mapping) || !count)
312 return 0;
313 ret = filemap_fdatawrite_range(mapping, pos, pos + count - 1);
314 if (ret == 0)
315 ret = generic_osync_inode(inode, mapping, OSYNC_METADATA);
316 if (ret == 0)
317 ret = wait_on_page_writeback_range(mapping, start, end);
318 return ret;
320 EXPORT_SYMBOL(sync_page_range_nolock);
323 * filemap_fdatawait - walk the list of under-writeback pages of the given
324 * address space and wait for all of them.
326 * @mapping: address space structure to wait for
328 int filemap_fdatawait(struct address_space *mapping)
330 loff_t i_size = i_size_read(mapping->host);
332 if (i_size == 0)
333 return 0;
335 return wait_on_page_writeback_range(mapping, 0,
336 (i_size - 1) >> PAGE_CACHE_SHIFT);
338 EXPORT_SYMBOL(filemap_fdatawait);
340 int filemap_write_and_wait(struct address_space *mapping)
342 int retval = 0;
344 if (mapping->nrpages) {
345 retval = filemap_fdatawrite(mapping);
346 if (retval == 0)
347 retval = filemap_fdatawait(mapping);
349 return retval;
352 int filemap_write_and_wait_range(struct address_space *mapping,
353 loff_t lstart, loff_t lend)
355 int retval = 0;
357 if (mapping->nrpages) {
358 retval = __filemap_fdatawrite_range(mapping, lstart, lend,
359 WB_SYNC_ALL);
360 if (retval == 0)
361 retval = wait_on_page_writeback_range(mapping,
362 lstart >> PAGE_CACHE_SHIFT,
363 lend >> PAGE_CACHE_SHIFT);
365 return retval;
369 * This function is used to add newly allocated pagecache pages:
370 * the page is new, so we can just run SetPageLocked() against it.
371 * The other page state flags were set by rmqueue().
373 * This function does not add the page to the LRU. The caller must do that.
375 int add_to_page_cache(struct page *page, struct address_space *mapping,
376 pgoff_t offset, int gfp_mask)
378 int error = radix_tree_preload(gfp_mask & ~__GFP_HIGHMEM);
380 if (error == 0) {
381 write_lock_irq(&mapping->tree_lock);
382 error = radix_tree_insert(&mapping->page_tree, offset, page);
383 if (!error) {
384 page_cache_get(page);
385 SetPageLocked(page);
386 page->mapping = mapping;
387 page->index = offset;
388 mapping->nrpages++;
389 pagecache_acct(1);
391 write_unlock_irq(&mapping->tree_lock);
392 radix_tree_preload_end();
394 return error;
397 EXPORT_SYMBOL(add_to_page_cache);
399 int add_to_page_cache_lru(struct page *page, struct address_space *mapping,
400 pgoff_t offset, int gfp_mask)
402 int ret = add_to_page_cache(page, mapping, offset, gfp_mask);
403 if (ret == 0)
404 lru_cache_add(page);
405 return ret;
409 * In order to wait for pages to become available there must be
410 * waitqueues associated with pages. By using a hash table of
411 * waitqueues where the bucket discipline is to maintain all
412 * waiters on the same queue and wake all when any of the pages
413 * become available, and for the woken contexts to check to be
414 * sure the appropriate page became available, this saves space
415 * at a cost of "thundering herd" phenomena during rare hash
416 * collisions.
418 static wait_queue_head_t *page_waitqueue(struct page *page)
420 const struct zone *zone = page_zone(page);
422 return &zone->wait_table[hash_ptr(page, zone->wait_table_bits)];
425 static inline void wake_up_page(struct page *page, int bit)
427 __wake_up_bit(page_waitqueue(page), &page->flags, bit);
430 void fastcall wait_on_page_bit(struct page *page, int bit_nr)
432 DEFINE_WAIT_BIT(wait, &page->flags, bit_nr);
434 if (test_bit(bit_nr, &page->flags))
435 __wait_on_bit(page_waitqueue(page), &wait, sync_page,
436 TASK_UNINTERRUPTIBLE);
438 EXPORT_SYMBOL(wait_on_page_bit);
441 * unlock_page() - unlock a locked page
443 * @page: the page
445 * Unlocks the page and wakes up sleepers in ___wait_on_page_locked().
446 * Also wakes sleepers in wait_on_page_writeback() because the wakeup
447 * mechananism between PageLocked pages and PageWriteback pages is shared.
448 * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep.
450 * The first mb is necessary to safely close the critical section opened by the
451 * TestSetPageLocked(), the second mb is necessary to enforce ordering between
452 * the clear_bit and the read of the waitqueue (to avoid SMP races with a
453 * parallel wait_on_page_locked()).
455 void fastcall unlock_page(struct page *page)
457 smp_mb__before_clear_bit();
458 if (!TestClearPageLocked(page))
459 BUG();
460 smp_mb__after_clear_bit();
461 wake_up_page(page, PG_locked);
463 EXPORT_SYMBOL(unlock_page);
466 * End writeback against a page.
468 void end_page_writeback(struct page *page)
470 if (!TestClearPageReclaim(page) || rotate_reclaimable_page(page)) {
471 if (!test_clear_page_writeback(page))
472 BUG();
474 smp_mb__after_clear_bit();
475 wake_up_page(page, PG_writeback);
477 EXPORT_SYMBOL(end_page_writeback);
480 * Get a lock on the page, assuming we need to sleep to get it.
482 * Ugly: running sync_page() in state TASK_UNINTERRUPTIBLE is scary. If some
483 * random driver's requestfn sets TASK_RUNNING, we could busywait. However
484 * chances are that on the second loop, the block layer's plug list is empty,
485 * so sync_page() will then return in state TASK_UNINTERRUPTIBLE.
487 void fastcall __lock_page(struct page *page)
489 DEFINE_WAIT_BIT(wait, &page->flags, PG_locked);
491 __wait_on_bit_lock(page_waitqueue(page), &wait, sync_page,
492 TASK_UNINTERRUPTIBLE);
494 EXPORT_SYMBOL(__lock_page);
497 * a rather lightweight function, finding and getting a reference to a
498 * hashed page atomically.
500 struct page * find_get_page(struct address_space *mapping, unsigned long offset)
502 struct page *page;
504 read_lock_irq(&mapping->tree_lock);
505 page = radix_tree_lookup(&mapping->page_tree, offset);
506 if (page)
507 page_cache_get(page);
508 read_unlock_irq(&mapping->tree_lock);
509 return page;
512 EXPORT_SYMBOL(find_get_page);
515 * Same as above, but trylock it instead of incrementing the count.
517 struct page *find_trylock_page(struct address_space *mapping, unsigned long offset)
519 struct page *page;
521 read_lock_irq(&mapping->tree_lock);
522 page = radix_tree_lookup(&mapping->page_tree, offset);
523 if (page && TestSetPageLocked(page))
524 page = NULL;
525 read_unlock_irq(&mapping->tree_lock);
526 return page;
529 EXPORT_SYMBOL(find_trylock_page);
532 * find_lock_page - locate, pin and lock a pagecache page
534 * @mapping: the address_space to search
535 * @offset: the page index
537 * Locates the desired pagecache page, locks it, increments its reference
538 * count and returns its address.
540 * Returns zero if the page was not present. find_lock_page() may sleep.
542 struct page *find_lock_page(struct address_space *mapping,
543 unsigned long offset)
545 struct page *page;
547 read_lock_irq(&mapping->tree_lock);
548 repeat:
549 page = radix_tree_lookup(&mapping->page_tree, offset);
550 if (page) {
551 page_cache_get(page);
552 if (TestSetPageLocked(page)) {
553 read_unlock_irq(&mapping->tree_lock);
554 lock_page(page);
555 read_lock_irq(&mapping->tree_lock);
557 /* Has the page been truncated while we slept? */
558 if (page->mapping != mapping || page->index != offset) {
559 unlock_page(page);
560 page_cache_release(page);
561 goto repeat;
565 read_unlock_irq(&mapping->tree_lock);
566 return page;
569 EXPORT_SYMBOL(find_lock_page);
572 * find_or_create_page - locate or add a pagecache page
574 * @mapping: the page's address_space
575 * @index: the page's index into the mapping
576 * @gfp_mask: page allocation mode
578 * Locates a page in the pagecache. If the page is not present, a new page
579 * is allocated using @gfp_mask and is added to the pagecache and to the VM's
580 * LRU list. The returned page is locked and has its reference count
581 * incremented.
583 * find_or_create_page() may sleep, even if @gfp_flags specifies an atomic
584 * allocation!
586 * find_or_create_page() returns the desired page's address, or zero on
587 * memory exhaustion.
589 struct page *find_or_create_page(struct address_space *mapping,
590 unsigned long index, unsigned int gfp_mask)
592 struct page *page, *cached_page = NULL;
593 int err;
594 repeat:
595 page = find_lock_page(mapping, index);
596 if (!page) {
597 if (!cached_page) {
598 cached_page = alloc_page(gfp_mask);
599 if (!cached_page)
600 return NULL;
602 err = add_to_page_cache_lru(cached_page, mapping,
603 index, gfp_mask);
604 if (!err) {
605 page = cached_page;
606 cached_page = NULL;
607 } else if (err == -EEXIST)
608 goto repeat;
610 if (cached_page)
611 page_cache_release(cached_page);
612 return page;
615 EXPORT_SYMBOL(find_or_create_page);
618 * find_get_pages - gang pagecache lookup
619 * @mapping: The address_space to search
620 * @start: The starting page index
621 * @nr_pages: The maximum number of pages
622 * @pages: Where the resulting pages are placed
624 * find_get_pages() will search for and return a group of up to
625 * @nr_pages pages in the mapping. The pages are placed at @pages.
626 * find_get_pages() takes a reference against the returned pages.
628 * The search returns a group of mapping-contiguous pages with ascending
629 * indexes. There may be holes in the indices due to not-present pages.
631 * find_get_pages() returns the number of pages which were found.
633 unsigned find_get_pages(struct address_space *mapping, pgoff_t start,
634 unsigned int nr_pages, struct page **pages)
636 unsigned int i;
637 unsigned int ret;
639 read_lock_irq(&mapping->tree_lock);
640 ret = radix_tree_gang_lookup(&mapping->page_tree,
641 (void **)pages, start, nr_pages);
642 for (i = 0; i < ret; i++)
643 page_cache_get(pages[i]);
644 read_unlock_irq(&mapping->tree_lock);
645 return ret;
649 * Like find_get_pages, except we only return pages which are tagged with
650 * `tag'. We update *index to index the next page for the traversal.
652 unsigned find_get_pages_tag(struct address_space *mapping, pgoff_t *index,
653 int tag, unsigned int nr_pages, struct page **pages)
655 unsigned int i;
656 unsigned int ret;
658 read_lock_irq(&mapping->tree_lock);
659 ret = radix_tree_gang_lookup_tag(&mapping->page_tree,
660 (void **)pages, *index, nr_pages, tag);
661 for (i = 0; i < ret; i++)
662 page_cache_get(pages[i]);
663 if (ret)
664 *index = pages[ret - 1]->index + 1;
665 read_unlock_irq(&mapping->tree_lock);
666 return ret;
670 * Same as grab_cache_page, but do not wait if the page is unavailable.
671 * This is intended for speculative data generators, where the data can
672 * be regenerated if the page couldn't be grabbed. This routine should
673 * be safe to call while holding the lock for another page.
675 * Clear __GFP_FS when allocating the page to avoid recursion into the fs
676 * and deadlock against the caller's locked page.
678 struct page *
679 grab_cache_page_nowait(struct address_space *mapping, unsigned long index)
681 struct page *page = find_get_page(mapping, index);
682 unsigned int gfp_mask;
684 if (page) {
685 if (!TestSetPageLocked(page))
686 return page;
687 page_cache_release(page);
688 return NULL;
690 gfp_mask = mapping_gfp_mask(mapping) & ~__GFP_FS;
691 page = alloc_pages(gfp_mask, 0);
692 if (page && add_to_page_cache_lru(page, mapping, index, gfp_mask)) {
693 page_cache_release(page);
694 page = NULL;
696 return page;
699 EXPORT_SYMBOL(grab_cache_page_nowait);
702 * This is a generic file read routine, and uses the
703 * mapping->a_ops->readpage() function for the actual low-level
704 * stuff.
706 * This is really ugly. But the goto's actually try to clarify some
707 * of the logic when it comes to error handling etc.
709 * Note the struct file* is only passed for the use of readpage. It may be
710 * NULL.
712 void do_generic_mapping_read(struct address_space *mapping,
713 struct file_ra_state *_ra,
714 struct file *filp,
715 loff_t *ppos,
716 read_descriptor_t *desc,
717 read_actor_t actor)
719 struct inode *inode = mapping->host;
720 unsigned long index;
721 unsigned long end_index;
722 unsigned long offset;
723 unsigned long last_index;
724 unsigned long next_index;
725 unsigned long prev_index;
726 loff_t isize;
727 struct page *cached_page;
728 int error;
729 struct file_ra_state ra = *_ra;
731 cached_page = NULL;
732 index = *ppos >> PAGE_CACHE_SHIFT;
733 next_index = index;
734 prev_index = ra.prev_page;
735 last_index = (*ppos + desc->count + PAGE_CACHE_SIZE-1) >> PAGE_CACHE_SHIFT;
736 offset = *ppos & ~PAGE_CACHE_MASK;
738 isize = i_size_read(inode);
739 if (!isize)
740 goto out;
742 end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
743 for (;;) {
744 struct page *page;
745 unsigned long nr, ret;
747 /* nr is the maximum number of bytes to copy from this page */
748 nr = PAGE_CACHE_SIZE;
749 if (index >= end_index) {
750 if (index > end_index)
751 goto out;
752 nr = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;
753 if (nr <= offset) {
754 goto out;
757 nr = nr - offset;
759 cond_resched();
760 if (index == next_index)
761 next_index = page_cache_readahead(mapping, &ra, filp,
762 index, last_index - index);
764 find_page:
765 page = find_get_page(mapping, index);
766 if (unlikely(page == NULL)) {
767 handle_ra_miss(mapping, &ra, index);
768 goto no_cached_page;
770 if (!PageUptodate(page))
771 goto page_not_up_to_date;
772 page_ok:
774 /* If users can be writing to this page using arbitrary
775 * virtual addresses, take care about potential aliasing
776 * before reading the page on the kernel side.
778 if (mapping_writably_mapped(mapping))
779 flush_dcache_page(page);
782 * When (part of) the same page is read multiple times
783 * in succession, only mark it as accessed the first time.
785 if (prev_index != index)
786 mark_page_accessed(page);
787 prev_index = index;
790 * Ok, we have the page, and it's up-to-date, so
791 * now we can copy it to user space...
793 * The actor routine returns how many bytes were actually used..
794 * NOTE! This may not be the same as how much of a user buffer
795 * we filled up (we may be padding etc), so we can only update
796 * "pos" here (the actor routine has to update the user buffer
797 * pointers and the remaining count).
799 ret = actor(desc, page, offset, nr);
800 offset += ret;
801 index += offset >> PAGE_CACHE_SHIFT;
802 offset &= ~PAGE_CACHE_MASK;
804 page_cache_release(page);
805 if (ret == nr && desc->count)
806 continue;
807 goto out;
809 page_not_up_to_date:
810 /* Get exclusive access to the page ... */
811 lock_page(page);
813 /* Did it get unhashed before we got the lock? */
814 if (!page->mapping) {
815 unlock_page(page);
816 page_cache_release(page);
817 continue;
820 /* Did somebody else fill it already? */
821 if (PageUptodate(page)) {
822 unlock_page(page);
823 goto page_ok;
826 readpage:
827 /* Start the actual read. The read will unlock the page. */
828 error = mapping->a_ops->readpage(filp, page);
830 if (unlikely(error))
831 goto readpage_error;
833 if (!PageUptodate(page)) {
834 lock_page(page);
835 if (!PageUptodate(page)) {
836 if (page->mapping == NULL) {
838 * invalidate_inode_pages got it
840 unlock_page(page);
841 page_cache_release(page);
842 goto find_page;
844 unlock_page(page);
845 error = -EIO;
846 goto readpage_error;
848 unlock_page(page);
852 * i_size must be checked after we have done ->readpage.
854 * Checking i_size after the readpage allows us to calculate
855 * the correct value for "nr", which means the zero-filled
856 * part of the page is not copied back to userspace (unless
857 * another truncate extends the file - this is desired though).
859 isize = i_size_read(inode);
860 end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
861 if (unlikely(!isize || index > end_index)) {
862 page_cache_release(page);
863 goto out;
866 /* nr is the maximum number of bytes to copy from this page */
867 nr = PAGE_CACHE_SIZE;
868 if (index == end_index) {
869 nr = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;
870 if (nr <= offset) {
871 page_cache_release(page);
872 goto out;
875 nr = nr - offset;
876 goto page_ok;
878 readpage_error:
879 /* UHHUH! A synchronous read error occurred. Report it */
880 desc->error = error;
881 page_cache_release(page);
882 goto out;
884 no_cached_page:
886 * Ok, it wasn't cached, so we need to create a new
887 * page..
889 if (!cached_page) {
890 cached_page = page_cache_alloc_cold(mapping);
891 if (!cached_page) {
892 desc->error = -ENOMEM;
893 goto out;
896 error = add_to_page_cache_lru(cached_page, mapping,
897 index, GFP_KERNEL);
898 if (error) {
899 if (error == -EEXIST)
900 goto find_page;
901 desc->error = error;
902 goto out;
904 page = cached_page;
905 cached_page = NULL;
906 goto readpage;
909 out:
910 *_ra = ra;
912 *ppos = ((loff_t) index << PAGE_CACHE_SHIFT) + offset;
913 if (cached_page)
914 page_cache_release(cached_page);
915 if (filp)
916 file_accessed(filp);
919 EXPORT_SYMBOL(do_generic_mapping_read);
921 int file_read_actor(read_descriptor_t *desc, struct page *page,
922 unsigned long offset, unsigned long size)
924 char *kaddr;
925 unsigned long left, count = desc->count;
927 if (size > count)
928 size = count;
931 * Faults on the destination of a read are common, so do it before
932 * taking the kmap.
934 if (!fault_in_pages_writeable(desc->arg.buf, size)) {
935 kaddr = kmap_atomic(page, KM_USER0);
936 left = __copy_to_user_inatomic(desc->arg.buf,
937 kaddr + offset, size);
938 kunmap_atomic(kaddr, KM_USER0);
939 if (left == 0)
940 goto success;
943 /* Do it the slow way */
944 kaddr = kmap(page);
945 left = __copy_to_user(desc->arg.buf, kaddr + offset, size);
946 kunmap(page);
948 if (left) {
949 size -= left;
950 desc->error = -EFAULT;
952 success:
953 desc->count = count - size;
954 desc->written += size;
955 desc->arg.buf += size;
956 return size;
960 * This is the "read()" routine for all filesystems
961 * that can use the page cache directly.
963 ssize_t
964 __generic_file_aio_read(struct kiocb *iocb, const struct iovec *iov,
965 unsigned long nr_segs, loff_t *ppos)
967 struct file *filp = iocb->ki_filp;
968 ssize_t retval;
969 unsigned long seg;
970 size_t count;
972 count = 0;
973 for (seg = 0; seg < nr_segs; seg++) {
974 const struct iovec *iv = &iov[seg];
977 * If any segment has a negative length, or the cumulative
978 * length ever wraps negative then return -EINVAL.
980 count += iv->iov_len;
981 if (unlikely((ssize_t)(count|iv->iov_len) < 0))
982 return -EINVAL;
983 if (access_ok(VERIFY_WRITE, iv->iov_base, iv->iov_len))
984 continue;
985 if (seg == 0)
986 return -EFAULT;
987 nr_segs = seg;
988 count -= iv->iov_len; /* This segment is no good */
989 break;
992 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */
993 if (filp->f_flags & O_DIRECT) {
994 loff_t pos = *ppos, size;
995 struct address_space *mapping;
996 struct inode *inode;
998 mapping = filp->f_mapping;
999 inode = mapping->host;
1000 retval = 0;
1001 if (!count)
1002 goto out; /* skip atime */
1003 size = i_size_read(inode);
1004 if (pos < size) {
1005 retval = generic_file_direct_IO(READ, iocb,
1006 iov, pos, nr_segs);
1007 if (retval > 0 && !is_sync_kiocb(iocb))
1008 retval = -EIOCBQUEUED;
1009 if (retval > 0)
1010 *ppos = pos + retval;
1012 file_accessed(filp);
1013 goto out;
1016 retval = 0;
1017 if (count) {
1018 for (seg = 0; seg < nr_segs; seg++) {
1019 read_descriptor_t desc;
1021 desc.written = 0;
1022 desc.arg.buf = iov[seg].iov_base;
1023 desc.count = iov[seg].iov_len;
1024 if (desc.count == 0)
1025 continue;
1026 desc.error = 0;
1027 do_generic_file_read(filp,ppos,&desc,file_read_actor);
1028 retval += desc.written;
1029 if (!retval) {
1030 retval = desc.error;
1031 break;
1035 out:
1036 return retval;
1039 EXPORT_SYMBOL(__generic_file_aio_read);
1041 ssize_t
1042 generic_file_aio_read(struct kiocb *iocb, char __user *buf, size_t count, loff_t pos)
1044 struct iovec local_iov = { .iov_base = buf, .iov_len = count };
1046 BUG_ON(iocb->ki_pos != pos);
1047 return __generic_file_aio_read(iocb, &local_iov, 1, &iocb->ki_pos);
1050 EXPORT_SYMBOL(generic_file_aio_read);
1052 ssize_t
1053 generic_file_read(struct file *filp, char __user *buf, size_t count, loff_t *ppos)
1055 struct iovec local_iov = { .iov_base = buf, .iov_len = count };
1056 struct kiocb kiocb;
1057 ssize_t ret;
1059 init_sync_kiocb(&kiocb, filp);
1060 ret = __generic_file_aio_read(&kiocb, &local_iov, 1, ppos);
1061 if (-EIOCBQUEUED == ret)
1062 ret = wait_on_sync_kiocb(&kiocb);
1063 return ret;
1066 EXPORT_SYMBOL(generic_file_read);
1068 int file_send_actor(read_descriptor_t * desc, struct page *page, unsigned long offset, unsigned long size)
1070 ssize_t written;
1071 unsigned long count = desc->count;
1072 struct file *file = desc->arg.data;
1074 if (size > count)
1075 size = count;
1077 written = file->f_op->sendpage(file, page, offset,
1078 size, &file->f_pos, size<count);
1079 if (written < 0) {
1080 desc->error = written;
1081 written = 0;
1083 desc->count = count - written;
1084 desc->written += written;
1085 return written;
1088 ssize_t generic_file_sendfile(struct file *in_file, loff_t *ppos,
1089 size_t count, read_actor_t actor, void *target)
1091 read_descriptor_t desc;
1093 if (!count)
1094 return 0;
1096 desc.written = 0;
1097 desc.count = count;
1098 desc.arg.data = target;
1099 desc.error = 0;
1101 do_generic_file_read(in_file, ppos, &desc, actor);
1102 if (desc.written)
1103 return desc.written;
1104 return desc.error;
1107 EXPORT_SYMBOL(generic_file_sendfile);
1109 static ssize_t
1110 do_readahead(struct address_space *mapping, struct file *filp,
1111 unsigned long index, unsigned long nr)
1113 if (!mapping || !mapping->a_ops || !mapping->a_ops->readpage)
1114 return -EINVAL;
1116 force_page_cache_readahead(mapping, filp, index,
1117 max_sane_readahead(nr));
1118 return 0;
1121 asmlinkage ssize_t sys_readahead(int fd, loff_t offset, size_t count)
1123 ssize_t ret;
1124 struct file *file;
1126 ret = -EBADF;
1127 file = fget(fd);
1128 if (file) {
1129 if (file->f_mode & FMODE_READ) {
1130 struct address_space *mapping = file->f_mapping;
1131 unsigned long start = offset >> PAGE_CACHE_SHIFT;
1132 unsigned long end = (offset + count - 1) >> PAGE_CACHE_SHIFT;
1133 unsigned long len = end - start + 1;
1134 ret = do_readahead(mapping, file, start, len);
1136 fput(file);
1138 return ret;
1141 #ifdef CONFIG_MMU
1143 * This adds the requested page to the page cache if it isn't already there,
1144 * and schedules an I/O to read in its contents from disk.
1146 static int FASTCALL(page_cache_read(struct file * file, unsigned long offset));
1147 static int fastcall page_cache_read(struct file * file, unsigned long offset)
1149 struct address_space *mapping = file->f_mapping;
1150 struct page *page;
1151 int error;
1153 page = page_cache_alloc_cold(mapping);
1154 if (!page)
1155 return -ENOMEM;
1157 error = add_to_page_cache_lru(page, mapping, offset, GFP_KERNEL);
1158 if (!error) {
1159 error = mapping->a_ops->readpage(file, page);
1160 page_cache_release(page);
1161 return error;
1165 * We arrive here in the unlikely event that someone
1166 * raced with us and added our page to the cache first
1167 * or we are out of memory for radix-tree nodes.
1169 page_cache_release(page);
1170 return error == -EEXIST ? 0 : error;
1173 #define MMAP_LOTSAMISS (100)
1176 * filemap_nopage() is invoked via the vma operations vector for a
1177 * mapped memory region to read in file data during a page fault.
1179 * The goto's are kind of ugly, but this streamlines the normal case of having
1180 * it in the page cache, and handles the special cases reasonably without
1181 * having a lot of duplicated code.
1183 struct page *filemap_nopage(struct vm_area_struct *area,
1184 unsigned long address, int *type)
1186 int error;
1187 struct file *file = area->vm_file;
1188 struct address_space *mapping = file->f_mapping;
1189 struct file_ra_state *ra = &file->f_ra;
1190 struct inode *inode = mapping->host;
1191 struct page *page;
1192 unsigned long size, pgoff;
1193 int did_readaround = 0, majmin = VM_FAULT_MINOR;
1195 pgoff = ((address-area->vm_start) >> PAGE_CACHE_SHIFT) + area->vm_pgoff;
1197 retry_all:
1198 size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1199 if (pgoff >= size)
1200 goto outside_data_content;
1202 /* If we don't want any read-ahead, don't bother */
1203 if (VM_RandomReadHint(area))
1204 goto no_cached_page;
1207 * The readahead code wants to be told about each and every page
1208 * so it can build and shrink its windows appropriately
1210 * For sequential accesses, we use the generic readahead logic.
1212 if (VM_SequentialReadHint(area))
1213 page_cache_readahead(mapping, ra, file, pgoff, 1);
1216 * Do we have something in the page cache already?
1218 retry_find:
1219 page = find_get_page(mapping, pgoff);
1220 if (!page) {
1221 unsigned long ra_pages;
1223 if (VM_SequentialReadHint(area)) {
1224 handle_ra_miss(mapping, ra, pgoff);
1225 goto no_cached_page;
1227 ra->mmap_miss++;
1230 * Do we miss much more than hit in this file? If so,
1231 * stop bothering with read-ahead. It will only hurt.
1233 if (ra->mmap_miss > ra->mmap_hit + MMAP_LOTSAMISS)
1234 goto no_cached_page;
1237 * To keep the pgmajfault counter straight, we need to
1238 * check did_readaround, as this is an inner loop.
1240 if (!did_readaround) {
1241 majmin = VM_FAULT_MAJOR;
1242 inc_page_state(pgmajfault);
1244 did_readaround = 1;
1245 ra_pages = max_sane_readahead(file->f_ra.ra_pages);
1246 if (ra_pages) {
1247 pgoff_t start = 0;
1249 if (pgoff > ra_pages / 2)
1250 start = pgoff - ra_pages / 2;
1251 do_page_cache_readahead(mapping, file, start, ra_pages);
1253 page = find_get_page(mapping, pgoff);
1254 if (!page)
1255 goto no_cached_page;
1258 if (!did_readaround)
1259 ra->mmap_hit++;
1262 * Ok, found a page in the page cache, now we need to check
1263 * that it's up-to-date.
1265 if (!PageUptodate(page))
1266 goto page_not_uptodate;
1268 success:
1270 * Found the page and have a reference on it.
1272 mark_page_accessed(page);
1273 if (type)
1274 *type = majmin;
1275 return page;
1277 outside_data_content:
1279 * An external ptracer can access pages that normally aren't
1280 * accessible..
1282 if (area->vm_mm == current->mm)
1283 return NULL;
1284 /* Fall through to the non-read-ahead case */
1285 no_cached_page:
1287 * We're only likely to ever get here if MADV_RANDOM is in
1288 * effect.
1290 error = page_cache_read(file, pgoff);
1291 grab_swap_token();
1294 * The page we want has now been added to the page cache.
1295 * In the unlikely event that someone removed it in the
1296 * meantime, we'll just come back here and read it again.
1298 if (error >= 0)
1299 goto retry_find;
1302 * An error return from page_cache_read can result if the
1303 * system is low on memory, or a problem occurs while trying
1304 * to schedule I/O.
1306 if (error == -ENOMEM)
1307 return NOPAGE_OOM;
1308 return NULL;
1310 page_not_uptodate:
1311 if (!did_readaround) {
1312 majmin = VM_FAULT_MAJOR;
1313 inc_page_state(pgmajfault);
1315 lock_page(page);
1317 /* Did it get unhashed while we waited for it? */
1318 if (!page->mapping) {
1319 unlock_page(page);
1320 page_cache_release(page);
1321 goto retry_all;
1324 /* Did somebody else get it up-to-date? */
1325 if (PageUptodate(page)) {
1326 unlock_page(page);
1327 goto success;
1330 if (!mapping->a_ops->readpage(file, page)) {
1331 wait_on_page_locked(page);
1332 if (PageUptodate(page))
1333 goto success;
1337 * Umm, take care of errors if the page isn't up-to-date.
1338 * Try to re-read it _once_. We do this synchronously,
1339 * because there really aren't any performance issues here
1340 * and we need to check for errors.
1342 lock_page(page);
1344 /* Somebody truncated the page on us? */
1345 if (!page->mapping) {
1346 unlock_page(page);
1347 page_cache_release(page);
1348 goto retry_all;
1351 /* Somebody else successfully read it in? */
1352 if (PageUptodate(page)) {
1353 unlock_page(page);
1354 goto success;
1356 ClearPageError(page);
1357 if (!mapping->a_ops->readpage(file, page)) {
1358 wait_on_page_locked(page);
1359 if (PageUptodate(page))
1360 goto success;
1364 * Things didn't work out. Return zero to tell the
1365 * mm layer so, possibly freeing the page cache page first.
1367 page_cache_release(page);
1368 return NULL;
1371 EXPORT_SYMBOL(filemap_nopage);
1373 static struct page * filemap_getpage(struct file *file, unsigned long pgoff,
1374 int nonblock)
1376 struct address_space *mapping = file->f_mapping;
1377 struct page *page;
1378 int error;
1381 * Do we have something in the page cache already?
1383 retry_find:
1384 page = find_get_page(mapping, pgoff);
1385 if (!page) {
1386 if (nonblock)
1387 return NULL;
1388 goto no_cached_page;
1392 * Ok, found a page in the page cache, now we need to check
1393 * that it's up-to-date.
1395 if (!PageUptodate(page)) {
1396 if (nonblock) {
1397 page_cache_release(page);
1398 return NULL;
1400 goto page_not_uptodate;
1403 success:
1405 * Found the page and have a reference on it.
1407 mark_page_accessed(page);
1408 return page;
1410 no_cached_page:
1411 error = page_cache_read(file, pgoff);
1414 * The page we want has now been added to the page cache.
1415 * In the unlikely event that someone removed it in the
1416 * meantime, we'll just come back here and read it again.
1418 if (error >= 0)
1419 goto retry_find;
1422 * An error return from page_cache_read can result if the
1423 * system is low on memory, or a problem occurs while trying
1424 * to schedule I/O.
1426 return NULL;
1428 page_not_uptodate:
1429 lock_page(page);
1431 /* Did it get unhashed while we waited for it? */
1432 if (!page->mapping) {
1433 unlock_page(page);
1434 goto err;
1437 /* Did somebody else get it up-to-date? */
1438 if (PageUptodate(page)) {
1439 unlock_page(page);
1440 goto success;
1443 if (!mapping->a_ops->readpage(file, page)) {
1444 wait_on_page_locked(page);
1445 if (PageUptodate(page))
1446 goto success;
1450 * Umm, take care of errors if the page isn't up-to-date.
1451 * Try to re-read it _once_. We do this synchronously,
1452 * because there really aren't any performance issues here
1453 * and we need to check for errors.
1455 lock_page(page);
1457 /* Somebody truncated the page on us? */
1458 if (!page->mapping) {
1459 unlock_page(page);
1460 goto err;
1462 /* Somebody else successfully read it in? */
1463 if (PageUptodate(page)) {
1464 unlock_page(page);
1465 goto success;
1468 ClearPageError(page);
1469 if (!mapping->a_ops->readpage(file, page)) {
1470 wait_on_page_locked(page);
1471 if (PageUptodate(page))
1472 goto success;
1476 * Things didn't work out. Return zero to tell the
1477 * mm layer so, possibly freeing the page cache page first.
1479 err:
1480 page_cache_release(page);
1482 return NULL;
1485 int filemap_populate(struct vm_area_struct *vma, unsigned long addr,
1486 unsigned long len, pgprot_t prot, unsigned long pgoff,
1487 int nonblock)
1489 struct file *file = vma->vm_file;
1490 struct address_space *mapping = file->f_mapping;
1491 struct inode *inode = mapping->host;
1492 unsigned long size;
1493 struct mm_struct *mm = vma->vm_mm;
1494 struct page *page;
1495 int err;
1497 if (!nonblock)
1498 force_page_cache_readahead(mapping, vma->vm_file,
1499 pgoff, len >> PAGE_CACHE_SHIFT);
1501 repeat:
1502 size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1503 if (pgoff + (len >> PAGE_CACHE_SHIFT) > size)
1504 return -EINVAL;
1506 page = filemap_getpage(file, pgoff, nonblock);
1508 /* XXX: This is wrong, a filesystem I/O error may have happened. Fix that as
1509 * done in shmem_populate calling shmem_getpage */
1510 if (!page && !nonblock)
1511 return -ENOMEM;
1513 if (page) {
1514 err = install_page(mm, vma, addr, page, prot);
1515 if (err) {
1516 page_cache_release(page);
1517 return err;
1519 } else {
1520 /* No page was found just because we can't read it in now (being
1521 * here implies nonblock != 0), but the page may exist, so set
1522 * the PTE to fault it in later. */
1523 err = install_file_pte(mm, vma, addr, pgoff, prot);
1524 if (err)
1525 return err;
1528 len -= PAGE_SIZE;
1529 addr += PAGE_SIZE;
1530 pgoff++;
1531 if (len)
1532 goto repeat;
1534 return 0;
1537 struct vm_operations_struct generic_file_vm_ops = {
1538 .nopage = filemap_nopage,
1539 .populate = filemap_populate,
1542 /* This is used for a general mmap of a disk file */
1544 int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
1546 struct address_space *mapping = file->f_mapping;
1548 if (!mapping->a_ops->readpage)
1549 return -ENOEXEC;
1550 file_accessed(file);
1551 vma->vm_ops = &generic_file_vm_ops;
1552 return 0;
1554 EXPORT_SYMBOL(filemap_populate);
1557 * This is for filesystems which do not implement ->writepage.
1559 int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
1561 if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE))
1562 return -EINVAL;
1563 return generic_file_mmap(file, vma);
1565 #else
1566 int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
1568 return -ENOSYS;
1570 int generic_file_readonly_mmap(struct file * file, struct vm_area_struct * vma)
1572 return -ENOSYS;
1574 #endif /* CONFIG_MMU */
1576 EXPORT_SYMBOL(generic_file_mmap);
1577 EXPORT_SYMBOL(generic_file_readonly_mmap);
1579 static inline struct page *__read_cache_page(struct address_space *mapping,
1580 unsigned long index,
1581 int (*filler)(void *,struct page*),
1582 void *data)
1584 struct page *page, *cached_page = NULL;
1585 int err;
1586 repeat:
1587 page = find_get_page(mapping, index);
1588 if (!page) {
1589 if (!cached_page) {
1590 cached_page = page_cache_alloc_cold(mapping);
1591 if (!cached_page)
1592 return ERR_PTR(-ENOMEM);
1594 err = add_to_page_cache_lru(cached_page, mapping,
1595 index, GFP_KERNEL);
1596 if (err == -EEXIST)
1597 goto repeat;
1598 if (err < 0) {
1599 /* Presumably ENOMEM for radix tree node */
1600 page_cache_release(cached_page);
1601 return ERR_PTR(err);
1603 page = cached_page;
1604 cached_page = NULL;
1605 err = filler(data, page);
1606 if (err < 0) {
1607 page_cache_release(page);
1608 page = ERR_PTR(err);
1611 if (cached_page)
1612 page_cache_release(cached_page);
1613 return page;
1617 * Read into the page cache. If a page already exists,
1618 * and PageUptodate() is not set, try to fill the page.
1620 struct page *read_cache_page(struct address_space *mapping,
1621 unsigned long index,
1622 int (*filler)(void *,struct page*),
1623 void *data)
1625 struct page *page;
1626 int err;
1628 retry:
1629 page = __read_cache_page(mapping, index, filler, data);
1630 if (IS_ERR(page))
1631 goto out;
1632 mark_page_accessed(page);
1633 if (PageUptodate(page))
1634 goto out;
1636 lock_page(page);
1637 if (!page->mapping) {
1638 unlock_page(page);
1639 page_cache_release(page);
1640 goto retry;
1642 if (PageUptodate(page)) {
1643 unlock_page(page);
1644 goto out;
1646 err = filler(data, page);
1647 if (err < 0) {
1648 page_cache_release(page);
1649 page = ERR_PTR(err);
1651 out:
1652 return page;
1655 EXPORT_SYMBOL(read_cache_page);
1658 * If the page was newly created, increment its refcount and add it to the
1659 * caller's lru-buffering pagevec. This function is specifically for
1660 * generic_file_write().
1662 static inline struct page *
1663 __grab_cache_page(struct address_space *mapping, unsigned long index,
1664 struct page **cached_page, struct pagevec *lru_pvec)
1666 int err;
1667 struct page *page;
1668 repeat:
1669 page = find_lock_page(mapping, index);
1670 if (!page) {
1671 if (!*cached_page) {
1672 *cached_page = page_cache_alloc(mapping);
1673 if (!*cached_page)
1674 return NULL;
1676 err = add_to_page_cache(*cached_page, mapping,
1677 index, GFP_KERNEL);
1678 if (err == -EEXIST)
1679 goto repeat;
1680 if (err == 0) {
1681 page = *cached_page;
1682 page_cache_get(page);
1683 if (!pagevec_add(lru_pvec, page))
1684 __pagevec_lru_add(lru_pvec);
1685 *cached_page = NULL;
1688 return page;
1692 * The logic we want is
1694 * if suid or (sgid and xgrp)
1695 * remove privs
1697 int remove_suid(struct dentry *dentry)
1699 mode_t mode = dentry->d_inode->i_mode;
1700 int kill = 0;
1701 int result = 0;
1703 /* suid always must be killed */
1704 if (unlikely(mode & S_ISUID))
1705 kill = ATTR_KILL_SUID;
1708 * sgid without any exec bits is just a mandatory locking mark; leave
1709 * it alone. If some exec bits are set, it's a real sgid; kill it.
1711 if (unlikely((mode & S_ISGID) && (mode & S_IXGRP)))
1712 kill |= ATTR_KILL_SGID;
1714 if (unlikely(kill && !capable(CAP_FSETID))) {
1715 struct iattr newattrs;
1717 newattrs.ia_valid = ATTR_FORCE | kill;
1718 result = notify_change(dentry, &newattrs);
1720 return result;
1722 EXPORT_SYMBOL(remove_suid);
1724 size_t
1725 __filemap_copy_from_user_iovec(char *vaddr,
1726 const struct iovec *iov, size_t base, size_t bytes)
1728 size_t copied = 0, left = 0;
1730 while (bytes) {
1731 char __user *buf = iov->iov_base + base;
1732 int copy = min(bytes, iov->iov_len - base);
1734 base = 0;
1735 left = __copy_from_user_inatomic(vaddr, buf, copy);
1736 copied += copy;
1737 bytes -= copy;
1738 vaddr += copy;
1739 iov++;
1741 if (unlikely(left)) {
1742 /* zero the rest of the target like __copy_from_user */
1743 if (bytes)
1744 memset(vaddr, 0, bytes);
1745 break;
1748 return copied - left;
1752 * Performs necessary checks before doing a write
1754 * Can adjust writing position aor amount of bytes to write.
1755 * Returns appropriate error code that caller should return or
1756 * zero in case that write should be allowed.
1758 inline int generic_write_checks(struct file *file, loff_t *pos, size_t *count, int isblk)
1760 struct inode *inode = file->f_mapping->host;
1761 unsigned long limit = current->signal->rlim[RLIMIT_FSIZE].rlim_cur;
1763 if (unlikely(*pos < 0))
1764 return -EINVAL;
1766 if (!isblk) {
1767 /* FIXME: this is for backwards compatibility with 2.4 */
1768 if (file->f_flags & O_APPEND)
1769 *pos = i_size_read(inode);
1771 if (limit != RLIM_INFINITY) {
1772 if (*pos >= limit) {
1773 send_sig(SIGXFSZ, current, 0);
1774 return -EFBIG;
1776 if (*count > limit - (typeof(limit))*pos) {
1777 *count = limit - (typeof(limit))*pos;
1783 * LFS rule
1785 if (unlikely(*pos + *count > MAX_NON_LFS &&
1786 !(file->f_flags & O_LARGEFILE))) {
1787 if (*pos >= MAX_NON_LFS) {
1788 send_sig(SIGXFSZ, current, 0);
1789 return -EFBIG;
1791 if (*count > MAX_NON_LFS - (unsigned long)*pos) {
1792 *count = MAX_NON_LFS - (unsigned long)*pos;
1797 * Are we about to exceed the fs block limit ?
1799 * If we have written data it becomes a short write. If we have
1800 * exceeded without writing data we send a signal and return EFBIG.
1801 * Linus frestrict idea will clean these up nicely..
1803 if (likely(!isblk)) {
1804 if (unlikely(*pos >= inode->i_sb->s_maxbytes)) {
1805 if (*count || *pos > inode->i_sb->s_maxbytes) {
1806 send_sig(SIGXFSZ, current, 0);
1807 return -EFBIG;
1809 /* zero-length writes at ->s_maxbytes are OK */
1812 if (unlikely(*pos + *count > inode->i_sb->s_maxbytes))
1813 *count = inode->i_sb->s_maxbytes - *pos;
1814 } else {
1815 loff_t isize;
1816 if (bdev_read_only(I_BDEV(inode)))
1817 return -EPERM;
1818 isize = i_size_read(inode);
1819 if (*pos >= isize) {
1820 if (*count || *pos > isize)
1821 return -ENOSPC;
1824 if (*pos + *count > isize)
1825 *count = isize - *pos;
1827 return 0;
1829 EXPORT_SYMBOL(generic_write_checks);
1831 ssize_t
1832 generic_file_direct_write(struct kiocb *iocb, const struct iovec *iov,
1833 unsigned long *nr_segs, loff_t pos, loff_t *ppos,
1834 size_t count, size_t ocount)
1836 struct file *file = iocb->ki_filp;
1837 struct address_space *mapping = file->f_mapping;
1838 struct inode *inode = mapping->host;
1839 ssize_t written;
1841 if (count != ocount)
1842 *nr_segs = iov_shorten((struct iovec *)iov, *nr_segs, count);
1844 written = generic_file_direct_IO(WRITE, iocb, iov, pos, *nr_segs);
1845 if (written > 0) {
1846 loff_t end = pos + written;
1847 if (end > i_size_read(inode) && !S_ISBLK(inode->i_mode)) {
1848 i_size_write(inode, end);
1849 mark_inode_dirty(inode);
1851 *ppos = end;
1855 * Sync the fs metadata but not the minor inode changes and
1856 * of course not the data as we did direct DMA for the IO.
1857 * i_sem is held, which protects generic_osync_inode() from
1858 * livelocking.
1860 if (written >= 0 && ((file->f_flags & O_SYNC) || IS_SYNC(inode))) {
1861 int err = generic_osync_inode(inode, mapping, OSYNC_METADATA);
1862 if (err < 0)
1863 written = err;
1865 if (written == count && !is_sync_kiocb(iocb))
1866 written = -EIOCBQUEUED;
1867 return written;
1869 EXPORT_SYMBOL(generic_file_direct_write);
1871 ssize_t
1872 generic_file_buffered_write(struct kiocb *iocb, const struct iovec *iov,
1873 unsigned long nr_segs, loff_t pos, loff_t *ppos,
1874 size_t count, ssize_t written)
1876 struct file *file = iocb->ki_filp;
1877 struct address_space * mapping = file->f_mapping;
1878 struct address_space_operations *a_ops = mapping->a_ops;
1879 struct inode *inode = mapping->host;
1880 long status = 0;
1881 struct page *page;
1882 struct page *cached_page = NULL;
1883 size_t bytes;
1884 struct pagevec lru_pvec;
1885 const struct iovec *cur_iov = iov; /* current iovec */
1886 size_t iov_base = 0; /* offset in the current iovec */
1887 char __user *buf;
1889 pagevec_init(&lru_pvec, 0);
1892 * handle partial DIO write. Adjust cur_iov if needed.
1894 if (likely(nr_segs == 1))
1895 buf = iov->iov_base + written;
1896 else {
1897 filemap_set_next_iovec(&cur_iov, &iov_base, written);
1898 buf = cur_iov->iov_base + iov_base;
1901 do {
1902 unsigned long index;
1903 unsigned long offset;
1904 unsigned long maxlen;
1905 size_t copied;
1907 offset = (pos & (PAGE_CACHE_SIZE -1)); /* Within page */
1908 index = pos >> PAGE_CACHE_SHIFT;
1909 bytes = PAGE_CACHE_SIZE - offset;
1910 if (bytes > count)
1911 bytes = count;
1914 * Bring in the user page that we will copy from _first_.
1915 * Otherwise there's a nasty deadlock on copying from the
1916 * same page as we're writing to, without it being marked
1917 * up-to-date.
1919 maxlen = cur_iov->iov_len - iov_base;
1920 if (maxlen > bytes)
1921 maxlen = bytes;
1922 fault_in_pages_readable(buf, maxlen);
1924 page = __grab_cache_page(mapping,index,&cached_page,&lru_pvec);
1925 if (!page) {
1926 status = -ENOMEM;
1927 break;
1930 status = a_ops->prepare_write(file, page, offset, offset+bytes);
1931 if (unlikely(status)) {
1932 loff_t isize = i_size_read(inode);
1934 * prepare_write() may have instantiated a few blocks
1935 * outside i_size. Trim these off again.
1937 unlock_page(page);
1938 page_cache_release(page);
1939 if (pos + bytes > isize)
1940 vmtruncate(inode, isize);
1941 break;
1943 if (likely(nr_segs == 1))
1944 copied = filemap_copy_from_user(page, offset,
1945 buf, bytes);
1946 else
1947 copied = filemap_copy_from_user_iovec(page, offset,
1948 cur_iov, iov_base, bytes);
1949 flush_dcache_page(page);
1950 status = a_ops->commit_write(file, page, offset, offset+bytes);
1951 if (likely(copied > 0)) {
1952 if (!status)
1953 status = copied;
1955 if (status >= 0) {
1956 written += status;
1957 count -= status;
1958 pos += status;
1959 buf += status;
1960 if (unlikely(nr_segs > 1)) {
1961 filemap_set_next_iovec(&cur_iov,
1962 &iov_base, status);
1963 if (count)
1964 buf = cur_iov->iov_base +
1965 iov_base;
1966 } else {
1967 iov_base += status;
1971 if (unlikely(copied != bytes))
1972 if (status >= 0)
1973 status = -EFAULT;
1974 unlock_page(page);
1975 mark_page_accessed(page);
1976 page_cache_release(page);
1977 if (status < 0)
1978 break;
1979 balance_dirty_pages_ratelimited(mapping);
1980 cond_resched();
1981 } while (count);
1982 *ppos = pos;
1984 if (cached_page)
1985 page_cache_release(cached_page);
1988 * For now, when the user asks for O_SYNC, we'll actually give O_DSYNC
1990 if (likely(status >= 0)) {
1991 if (unlikely((file->f_flags & O_SYNC) || IS_SYNC(inode))) {
1992 if (!a_ops->writepage || !is_sync_kiocb(iocb))
1993 status = generic_osync_inode(inode, mapping,
1994 OSYNC_METADATA|OSYNC_DATA);
1999 * If we get here for O_DIRECT writes then we must have fallen through
2000 * to buffered writes (block instantiation inside i_size). So we sync
2001 * the file data here, to try to honour O_DIRECT expectations.
2003 if (unlikely(file->f_flags & O_DIRECT) && written)
2004 status = filemap_write_and_wait(mapping);
2006 pagevec_lru_add(&lru_pvec);
2007 return written ? written : status;
2009 EXPORT_SYMBOL(generic_file_buffered_write);
2011 ssize_t
2012 __generic_file_aio_write_nolock(struct kiocb *iocb, const struct iovec *iov,
2013 unsigned long nr_segs, loff_t *ppos)
2015 struct file *file = iocb->ki_filp;
2016 struct address_space * mapping = file->f_mapping;
2017 size_t ocount; /* original count */
2018 size_t count; /* after file limit checks */
2019 struct inode *inode = mapping->host;
2020 unsigned long seg;
2021 loff_t pos;
2022 ssize_t written;
2023 ssize_t err;
2025 ocount = 0;
2026 for (seg = 0; seg < nr_segs; seg++) {
2027 const struct iovec *iv = &iov[seg];
2030 * If any segment has a negative length, or the cumulative
2031 * length ever wraps negative then return -EINVAL.
2033 ocount += iv->iov_len;
2034 if (unlikely((ssize_t)(ocount|iv->iov_len) < 0))
2035 return -EINVAL;
2036 if (access_ok(VERIFY_READ, iv->iov_base, iv->iov_len))
2037 continue;
2038 if (seg == 0)
2039 return -EFAULT;
2040 nr_segs = seg;
2041 ocount -= iv->iov_len; /* This segment is no good */
2042 break;
2045 count = ocount;
2046 pos = *ppos;
2048 vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
2050 /* We can write back this queue in page reclaim */
2051 current->backing_dev_info = mapping->backing_dev_info;
2052 written = 0;
2054 err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
2055 if (err)
2056 goto out;
2058 if (count == 0)
2059 goto out;
2061 err = remove_suid(file->f_dentry);
2062 if (err)
2063 goto out;
2065 inode_update_time(inode, 1);
2067 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */
2068 if (unlikely(file->f_flags & O_DIRECT)) {
2069 written = generic_file_direct_write(iocb, iov,
2070 &nr_segs, pos, ppos, count, ocount);
2071 if (written < 0 || written == count)
2072 goto out;
2074 * direct-io write to a hole: fall through to buffered I/O
2075 * for completing the rest of the request.
2077 pos += written;
2078 count -= written;
2081 written = generic_file_buffered_write(iocb, iov, nr_segs,
2082 pos, ppos, count, written);
2083 out:
2084 current->backing_dev_info = NULL;
2085 return written ? written : err;
2087 EXPORT_SYMBOL(generic_file_aio_write_nolock);
2089 ssize_t
2090 generic_file_aio_write_nolock(struct kiocb *iocb, const struct iovec *iov,
2091 unsigned long nr_segs, loff_t *ppos)
2093 struct file *file = iocb->ki_filp;
2094 struct address_space *mapping = file->f_mapping;
2095 struct inode *inode = mapping->host;
2096 ssize_t ret;
2097 loff_t pos = *ppos;
2099 ret = __generic_file_aio_write_nolock(iocb, iov, nr_segs, ppos);
2101 if (ret > 0 && ((file->f_flags & O_SYNC) || IS_SYNC(inode))) {
2102 int err;
2104 err = sync_page_range_nolock(inode, mapping, pos, ret);
2105 if (err < 0)
2106 ret = err;
2108 return ret;
2111 ssize_t
2112 __generic_file_write_nolock(struct file *file, const struct iovec *iov,
2113 unsigned long nr_segs, loff_t *ppos)
2115 struct kiocb kiocb;
2116 ssize_t ret;
2118 init_sync_kiocb(&kiocb, file);
2119 ret = __generic_file_aio_write_nolock(&kiocb, iov, nr_segs, ppos);
2120 if (ret == -EIOCBQUEUED)
2121 ret = wait_on_sync_kiocb(&kiocb);
2122 return ret;
2125 ssize_t
2126 generic_file_write_nolock(struct file *file, const struct iovec *iov,
2127 unsigned long nr_segs, loff_t *ppos)
2129 struct kiocb kiocb;
2130 ssize_t ret;
2132 init_sync_kiocb(&kiocb, file);
2133 ret = generic_file_aio_write_nolock(&kiocb, iov, nr_segs, ppos);
2134 if (-EIOCBQUEUED == ret)
2135 ret = wait_on_sync_kiocb(&kiocb);
2136 return ret;
2138 EXPORT_SYMBOL(generic_file_write_nolock);
2140 ssize_t generic_file_aio_write(struct kiocb *iocb, const char __user *buf,
2141 size_t count, loff_t pos)
2143 struct file *file = iocb->ki_filp;
2144 struct address_space *mapping = file->f_mapping;
2145 struct inode *inode = mapping->host;
2146 ssize_t ret;
2147 struct iovec local_iov = { .iov_base = (void __user *)buf,
2148 .iov_len = count };
2150 BUG_ON(iocb->ki_pos != pos);
2152 down(&inode->i_sem);
2153 ret = __generic_file_aio_write_nolock(iocb, &local_iov, 1,
2154 &iocb->ki_pos);
2155 up(&inode->i_sem);
2157 if (ret > 0 && ((file->f_flags & O_SYNC) || IS_SYNC(inode))) {
2158 ssize_t err;
2160 err = sync_page_range(inode, mapping, pos, ret);
2161 if (err < 0)
2162 ret = err;
2164 return ret;
2166 EXPORT_SYMBOL(generic_file_aio_write);
2168 ssize_t generic_file_write(struct file *file, const char __user *buf,
2169 size_t count, loff_t *ppos)
2171 struct address_space *mapping = file->f_mapping;
2172 struct inode *inode = mapping->host;
2173 ssize_t ret;
2174 struct iovec local_iov = { .iov_base = (void __user *)buf,
2175 .iov_len = count };
2177 down(&inode->i_sem);
2178 ret = __generic_file_write_nolock(file, &local_iov, 1, ppos);
2179 up(&inode->i_sem);
2181 if (ret > 0 && ((file->f_flags & O_SYNC) || IS_SYNC(inode))) {
2182 ssize_t err;
2184 err = sync_page_range(inode, mapping, *ppos - ret, ret);
2185 if (err < 0)
2186 ret = err;
2188 return ret;
2190 EXPORT_SYMBOL(generic_file_write);
2192 ssize_t generic_file_readv(struct file *filp, const struct iovec *iov,
2193 unsigned long nr_segs, loff_t *ppos)
2195 struct kiocb kiocb;
2196 ssize_t ret;
2198 init_sync_kiocb(&kiocb, filp);
2199 ret = __generic_file_aio_read(&kiocb, iov, nr_segs, ppos);
2200 if (-EIOCBQUEUED == ret)
2201 ret = wait_on_sync_kiocb(&kiocb);
2202 return ret;
2204 EXPORT_SYMBOL(generic_file_readv);
2206 ssize_t generic_file_writev(struct file *file, const struct iovec *iov,
2207 unsigned long nr_segs, loff_t *ppos)
2209 struct address_space *mapping = file->f_mapping;
2210 struct inode *inode = mapping->host;
2211 ssize_t ret;
2213 down(&inode->i_sem);
2214 ret = __generic_file_write_nolock(file, iov, nr_segs, ppos);
2215 up(&inode->i_sem);
2217 if (ret > 0 && ((file->f_flags & O_SYNC) || IS_SYNC(inode))) {
2218 int err;
2220 err = sync_page_range(inode, mapping, *ppos - ret, ret);
2221 if (err < 0)
2222 ret = err;
2224 return ret;
2226 EXPORT_SYMBOL(generic_file_writev);
2229 * Called under i_sem for writes to S_ISREG files. Returns -EIO if something
2230 * went wrong during pagecache shootdown.
2232 ssize_t
2233 generic_file_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov,
2234 loff_t offset, unsigned long nr_segs)
2236 struct file *file = iocb->ki_filp;
2237 struct address_space *mapping = file->f_mapping;
2238 ssize_t retval;
2239 size_t write_len = 0;
2242 * If it's a write, unmap all mmappings of the file up-front. This
2243 * will cause any pte dirty bits to be propagated into the pageframes
2244 * for the subsequent filemap_write_and_wait().
2246 if (rw == WRITE) {
2247 write_len = iov_length(iov, nr_segs);
2248 if (mapping_mapped(mapping))
2249 unmap_mapping_range(mapping, offset, write_len, 0);
2252 retval = filemap_write_and_wait(mapping);
2253 if (retval == 0) {
2254 retval = mapping->a_ops->direct_IO(rw, iocb, iov,
2255 offset, nr_segs);
2256 if (rw == WRITE && mapping->nrpages) {
2257 pgoff_t end = (offset + write_len - 1)
2258 >> PAGE_CACHE_SHIFT;
2259 int err = invalidate_inode_pages2_range(mapping,
2260 offset >> PAGE_CACHE_SHIFT, end);
2261 if (err)
2262 retval = err;
2265 return retval;
2267 EXPORT_SYMBOL_GPL(generic_file_direct_IO);