xref: /openbmc/linux/mm/filemap.c (revision e868d61272caa648214046a096e5a6bfc068dc8c)
1 /*
2  *	linux/mm/filemap.c
3  *
4  * Copyright (C) 1994-1999  Linus Torvalds
5  */
6 
7 /*
8  * This file handles the generic file mmap semantics used by
9  * most "normal" filesystems (but you don't /have/ to use this:
10  * the NFS filesystem used to do this differently, for example)
11  */
12 #include <linux/module.h>
13 #include <linux/slab.h>
14 #include <linux/compiler.h>
15 #include <linux/fs.h>
16 #include <linux/uaccess.h>
17 #include <linux/aio.h>
18 #include <linux/capability.h>
19 #include <linux/kernel_stat.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/pagevec.h>
29 #include <linux/blkdev.h>
30 #include <linux/security.h>
31 #include <linux/syscalls.h>
32 #include <linux/cpuset.h>
33 #include "filemap.h"
34 #include "internal.h"
35 
36 /*
37  * FIXME: remove all knowledge of the buffer layer from the core VM
38  */
39 #include <linux/buffer_head.h> /* for generic_osync_inode */
40 
41 #include <asm/mman.h>
42 
43 static ssize_t
44 generic_file_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov,
45 	loff_t offset, unsigned long nr_segs);
46 
47 /*
48  * Shared mappings implemented 30.11.1994. It's not fully working yet,
49  * though.
50  *
51  * Shared mappings now work. 15.8.1995  Bruno.
52  *
53  * finished 'unifying' the page and buffer cache and SMP-threaded the
54  * page-cache, 21.05.1999, Ingo Molnar <mingo@redhat.com>
55  *
56  * SMP-threaded pagemap-LRU 1999, Andrea Arcangeli <andrea@suse.de>
57  */
58 
59 /*
60  * Lock ordering:
61  *
62  *  ->i_mmap_lock		(vmtruncate)
63  *    ->private_lock		(__free_pte->__set_page_dirty_buffers)
64  *      ->swap_lock		(exclusive_swap_page, others)
65  *        ->mapping->tree_lock
66  *
67  *  ->i_mutex
68  *    ->i_mmap_lock		(truncate->unmap_mapping_range)
69  *
70  *  ->mmap_sem
71  *    ->i_mmap_lock
72  *      ->page_table_lock or pte_lock	(various, mainly in memory.c)
73  *        ->mapping->tree_lock	(arch-dependent flush_dcache_mmap_lock)
74  *
75  *  ->mmap_sem
76  *    ->lock_page		(access_process_vm)
77  *
78  *  ->i_mutex			(generic_file_buffered_write)
79  *    ->mmap_sem		(fault_in_pages_readable->do_page_fault)
80  *
81  *  ->i_mutex
82  *    ->i_alloc_sem             (various)
83  *
84  *  ->inode_lock
85  *    ->sb_lock			(fs/fs-writeback.c)
86  *    ->mapping->tree_lock	(__sync_single_inode)
87  *
88  *  ->i_mmap_lock
89  *    ->anon_vma.lock		(vma_adjust)
90  *
91  *  ->anon_vma.lock
92  *    ->page_table_lock or pte_lock	(anon_vma_prepare and various)
93  *
94  *  ->page_table_lock or pte_lock
95  *    ->swap_lock		(try_to_unmap_one)
96  *    ->private_lock		(try_to_unmap_one)
97  *    ->tree_lock		(try_to_unmap_one)
98  *    ->zone.lru_lock		(follow_page->mark_page_accessed)
99  *    ->zone.lru_lock		(check_pte_range->isolate_lru_page)
100  *    ->private_lock		(page_remove_rmap->set_page_dirty)
101  *    ->tree_lock		(page_remove_rmap->set_page_dirty)
102  *    ->inode_lock		(page_remove_rmap->set_page_dirty)
103  *    ->inode_lock		(zap_pte_range->set_page_dirty)
104  *    ->private_lock		(zap_pte_range->__set_page_dirty_buffers)
105  *
106  *  ->task->proc_lock
107  *    ->dcache_lock		(proc_pid_lookup)
108  */
109 
110 /*
111  * Remove a page from the page cache and free it. Caller has to make
112  * sure the page is locked and that nobody else uses it - or that usage
113  * is safe.  The caller must hold a write_lock on the mapping's tree_lock.
114  */
115 void __remove_from_page_cache(struct page *page)
116 {
117 	struct address_space *mapping = page->mapping;
118 
119 	radix_tree_delete(&mapping->page_tree, page->index);
120 	page->mapping = NULL;
121 	mapping->nrpages--;
122 	__dec_zone_page_state(page, NR_FILE_PAGES);
123 }
124 
125 void remove_from_page_cache(struct page *page)
126 {
127 	struct address_space *mapping = page->mapping;
128 
129 	BUG_ON(!PageLocked(page));
130 
131 	write_lock_irq(&mapping->tree_lock);
132 	__remove_from_page_cache(page);
133 	write_unlock_irq(&mapping->tree_lock);
134 }
135 
136 static int sync_page(void *word)
137 {
138 	struct address_space *mapping;
139 	struct page *page;
140 
141 	page = container_of((unsigned long *)word, struct page, flags);
142 
143 	/*
144 	 * page_mapping() is being called without PG_locked held.
145 	 * Some knowledge of the state and use of the page is used to
146 	 * reduce the requirements down to a memory barrier.
147 	 * The danger here is of a stale page_mapping() return value
148 	 * indicating a struct address_space different from the one it's
149 	 * associated with when it is associated with one.
150 	 * After smp_mb(), it's either the correct page_mapping() for
151 	 * the page, or an old page_mapping() and the page's own
152 	 * page_mapping() has gone NULL.
153 	 * The ->sync_page() address_space operation must tolerate
154 	 * page_mapping() going NULL. By an amazing coincidence,
155 	 * this comes about because none of the users of the page
156 	 * in the ->sync_page() methods make essential use of the
157 	 * page_mapping(), merely passing the page down to the backing
158 	 * device's unplug functions when it's non-NULL, which in turn
159 	 * ignore it for all cases but swap, where only page_private(page) is
160 	 * of interest. When page_mapping() does go NULL, the entire
161 	 * call stack gracefully ignores the page and returns.
162 	 * -- wli
163 	 */
164 	smp_mb();
165 	mapping = page_mapping(page);
166 	if (mapping && mapping->a_ops && mapping->a_ops->sync_page)
167 		mapping->a_ops->sync_page(page);
168 	io_schedule();
169 	return 0;
170 }
171 
172 /**
173  * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range
174  * @mapping:	address space structure to write
175  * @start:	offset in bytes where the range starts
176  * @end:	offset in bytes where the range ends (inclusive)
177  * @sync_mode:	enable synchronous operation
178  *
179  * Start writeback against all of a mapping's dirty pages that lie
180  * within the byte offsets <start, end> inclusive.
181  *
182  * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
183  * opposed to a regular memory cleansing writeback.  The difference between
184  * these two operations is that if a dirty page/buffer is encountered, it must
185  * be waited upon, and not just skipped over.
186  */
187 int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
188 				loff_t end, int sync_mode)
189 {
190 	int ret;
191 	struct writeback_control wbc = {
192 		.sync_mode = sync_mode,
193 		.nr_to_write = mapping->nrpages * 2,
194 		.range_start = start,
195 		.range_end = end,
196 	};
197 
198 	if (!mapping_cap_writeback_dirty(mapping))
199 		return 0;
200 
201 	ret = do_writepages(mapping, &wbc);
202 	return ret;
203 }
204 
205 static inline int __filemap_fdatawrite(struct address_space *mapping,
206 	int sync_mode)
207 {
208 	return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode);
209 }
210 
211 int filemap_fdatawrite(struct address_space *mapping)
212 {
213 	return __filemap_fdatawrite(mapping, WB_SYNC_ALL);
214 }
215 EXPORT_SYMBOL(filemap_fdatawrite);
216 
217 static int filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
218 				loff_t end)
219 {
220 	return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL);
221 }
222 
223 /**
224  * filemap_flush - mostly a non-blocking flush
225  * @mapping:	target address_space
226  *
227  * This is a mostly non-blocking flush.  Not suitable for data-integrity
228  * purposes - I/O may not be started against all dirty pages.
229  */
230 int filemap_flush(struct address_space *mapping)
231 {
232 	return __filemap_fdatawrite(mapping, WB_SYNC_NONE);
233 }
234 EXPORT_SYMBOL(filemap_flush);
235 
236 /**
237  * wait_on_page_writeback_range - wait for writeback to complete
238  * @mapping:	target address_space
239  * @start:	beginning page index
240  * @end:	ending page index
241  *
242  * Wait for writeback to complete against pages indexed by start->end
243  * inclusive
244  */
245 int wait_on_page_writeback_range(struct address_space *mapping,
246 				pgoff_t start, pgoff_t end)
247 {
248 	struct pagevec pvec;
249 	int nr_pages;
250 	int ret = 0;
251 	pgoff_t index;
252 
253 	if (end < start)
254 		return 0;
255 
256 	pagevec_init(&pvec, 0);
257 	index = start;
258 	while ((index <= end) &&
259 			(nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
260 			PAGECACHE_TAG_WRITEBACK,
261 			min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1)) != 0) {
262 		unsigned i;
263 
264 		for (i = 0; i < nr_pages; i++) {
265 			struct page *page = pvec.pages[i];
266 
267 			/* until radix tree lookup accepts end_index */
268 			if (page->index > end)
269 				continue;
270 
271 			wait_on_page_writeback(page);
272 			if (PageError(page))
273 				ret = -EIO;
274 		}
275 		pagevec_release(&pvec);
276 		cond_resched();
277 	}
278 
279 	/* Check for outstanding write errors */
280 	if (test_and_clear_bit(AS_ENOSPC, &mapping->flags))
281 		ret = -ENOSPC;
282 	if (test_and_clear_bit(AS_EIO, &mapping->flags))
283 		ret = -EIO;
284 
285 	return ret;
286 }
287 
288 /**
289  * sync_page_range - write and wait on all pages in the passed range
290  * @inode:	target inode
291  * @mapping:	target address_space
292  * @pos:	beginning offset in pages to write
293  * @count:	number of bytes to write
294  *
295  * Write and wait upon all the pages in the passed range.  This is a "data
296  * integrity" operation.  It waits upon in-flight writeout before starting and
297  * waiting upon new writeout.  If there was an IO error, return it.
298  *
299  * We need to re-take i_mutex during the generic_osync_inode list walk because
300  * it is otherwise livelockable.
301  */
302 int sync_page_range(struct inode *inode, struct address_space *mapping,
303 			loff_t pos, loff_t count)
304 {
305 	pgoff_t start = pos >> PAGE_CACHE_SHIFT;
306 	pgoff_t end = (pos + count - 1) >> PAGE_CACHE_SHIFT;
307 	int ret;
308 
309 	if (!mapping_cap_writeback_dirty(mapping) || !count)
310 		return 0;
311 	ret = filemap_fdatawrite_range(mapping, pos, pos + count - 1);
312 	if (ret == 0) {
313 		mutex_lock(&inode->i_mutex);
314 		ret = generic_osync_inode(inode, mapping, OSYNC_METADATA);
315 		mutex_unlock(&inode->i_mutex);
316 	}
317 	if (ret == 0)
318 		ret = wait_on_page_writeback_range(mapping, start, end);
319 	return ret;
320 }
321 EXPORT_SYMBOL(sync_page_range);
322 
323 /**
324  * sync_page_range_nolock
325  * @inode:	target inode
326  * @mapping:	target address_space
327  * @pos:	beginning offset in pages to write
328  * @count:	number of bytes to write
329  *
330  * Note: Holding i_mutex across sync_page_range_nolock() is not a good idea
331  * as it forces O_SYNC writers to different parts of the same file
332  * to be serialised right until io completion.
333  */
334 int sync_page_range_nolock(struct inode *inode, struct address_space *mapping,
335 			   loff_t pos, loff_t count)
336 {
337 	pgoff_t start = pos >> PAGE_CACHE_SHIFT;
338 	pgoff_t end = (pos + count - 1) >> PAGE_CACHE_SHIFT;
339 	int ret;
340 
341 	if (!mapping_cap_writeback_dirty(mapping) || !count)
342 		return 0;
343 	ret = filemap_fdatawrite_range(mapping, pos, pos + count - 1);
344 	if (ret == 0)
345 		ret = generic_osync_inode(inode, mapping, OSYNC_METADATA);
346 	if (ret == 0)
347 		ret = wait_on_page_writeback_range(mapping, start, end);
348 	return ret;
349 }
350 EXPORT_SYMBOL(sync_page_range_nolock);
351 
352 /**
353  * filemap_fdatawait - wait for all under-writeback pages to complete
354  * @mapping: address space structure to wait for
355  *
356  * Walk the list of under-writeback pages of the given address space
357  * and wait for all of them.
358  */
359 int filemap_fdatawait(struct address_space *mapping)
360 {
361 	loff_t i_size = i_size_read(mapping->host);
362 
363 	if (i_size == 0)
364 		return 0;
365 
366 	return wait_on_page_writeback_range(mapping, 0,
367 				(i_size - 1) >> PAGE_CACHE_SHIFT);
368 }
369 EXPORT_SYMBOL(filemap_fdatawait);
370 
371 int filemap_write_and_wait(struct address_space *mapping)
372 {
373 	int err = 0;
374 
375 	if (mapping->nrpages) {
376 		err = filemap_fdatawrite(mapping);
377 		/*
378 		 * Even if the above returned error, the pages may be
379 		 * written partially (e.g. -ENOSPC), so we wait for it.
380 		 * But the -EIO is special case, it may indicate the worst
381 		 * thing (e.g. bug) happened, so we avoid waiting for it.
382 		 */
383 		if (err != -EIO) {
384 			int err2 = filemap_fdatawait(mapping);
385 			if (!err)
386 				err = err2;
387 		}
388 	}
389 	return err;
390 }
391 EXPORT_SYMBOL(filemap_write_and_wait);
392 
393 /**
394  * filemap_write_and_wait_range - write out & wait on a file range
395  * @mapping:	the address_space for the pages
396  * @lstart:	offset in bytes where the range starts
397  * @lend:	offset in bytes where the range ends (inclusive)
398  *
399  * Write out and wait upon file offsets lstart->lend, inclusive.
400  *
401  * Note that `lend' is inclusive (describes the last byte to be written) so
402  * that this function can be used to write to the very end-of-file (end = -1).
403  */
404 int filemap_write_and_wait_range(struct address_space *mapping,
405 				 loff_t lstart, loff_t lend)
406 {
407 	int err = 0;
408 
409 	if (mapping->nrpages) {
410 		err = __filemap_fdatawrite_range(mapping, lstart, lend,
411 						 WB_SYNC_ALL);
412 		/* See comment of filemap_write_and_wait() */
413 		if (err != -EIO) {
414 			int err2 = wait_on_page_writeback_range(mapping,
415 						lstart >> PAGE_CACHE_SHIFT,
416 						lend >> PAGE_CACHE_SHIFT);
417 			if (!err)
418 				err = err2;
419 		}
420 	}
421 	return err;
422 }
423 
424 /**
425  * add_to_page_cache - add newly allocated pagecache pages
426  * @page:	page to add
427  * @mapping:	the page's address_space
428  * @offset:	page index
429  * @gfp_mask:	page allocation mode
430  *
431  * This function is used to add newly allocated pagecache pages;
432  * the page is new, so we can just run SetPageLocked() against it.
433  * The other page state flags were set by rmqueue().
434  *
435  * This function does not add the page to the LRU.  The caller must do that.
436  */
437 int add_to_page_cache(struct page *page, struct address_space *mapping,
438 		pgoff_t offset, gfp_t gfp_mask)
439 {
440 	int error = radix_tree_preload(gfp_mask & ~__GFP_HIGHMEM);
441 
442 	if (error == 0) {
443 		write_lock_irq(&mapping->tree_lock);
444 		error = radix_tree_insert(&mapping->page_tree, offset, page);
445 		if (!error) {
446 			page_cache_get(page);
447 			SetPageLocked(page);
448 			page->mapping = mapping;
449 			page->index = offset;
450 			mapping->nrpages++;
451 			__inc_zone_page_state(page, NR_FILE_PAGES);
452 		}
453 		write_unlock_irq(&mapping->tree_lock);
454 		radix_tree_preload_end();
455 	}
456 	return error;
457 }
458 EXPORT_SYMBOL(add_to_page_cache);
459 
460 int add_to_page_cache_lru(struct page *page, struct address_space *mapping,
461 				pgoff_t offset, gfp_t gfp_mask)
462 {
463 	int ret = add_to_page_cache(page, mapping, offset, gfp_mask);
464 	if (ret == 0)
465 		lru_cache_add(page);
466 	return ret;
467 }
468 
469 #ifdef CONFIG_NUMA
470 struct page *__page_cache_alloc(gfp_t gfp)
471 {
472 	if (cpuset_do_page_mem_spread()) {
473 		int n = cpuset_mem_spread_node();
474 		return alloc_pages_node(n, gfp, 0);
475 	}
476 	return alloc_pages(gfp, 0);
477 }
478 EXPORT_SYMBOL(__page_cache_alloc);
479 #endif
480 
481 static int __sleep_on_page_lock(void *word)
482 {
483 	io_schedule();
484 	return 0;
485 }
486 
487 /*
488  * In order to wait for pages to become available there must be
489  * waitqueues associated with pages. By using a hash table of
490  * waitqueues where the bucket discipline is to maintain all
491  * waiters on the same queue and wake all when any of the pages
492  * become available, and for the woken contexts to check to be
493  * sure the appropriate page became available, this saves space
494  * at a cost of "thundering herd" phenomena during rare hash
495  * collisions.
496  */
497 static wait_queue_head_t *page_waitqueue(struct page *page)
498 {
499 	const struct zone *zone = page_zone(page);
500 
501 	return &zone->wait_table[hash_ptr(page, zone->wait_table_bits)];
502 }
503 
504 static inline void wake_up_page(struct page *page, int bit)
505 {
506 	__wake_up_bit(page_waitqueue(page), &page->flags, bit);
507 }
508 
509 void fastcall wait_on_page_bit(struct page *page, int bit_nr)
510 {
511 	DEFINE_WAIT_BIT(wait, &page->flags, bit_nr);
512 
513 	if (test_bit(bit_nr, &page->flags))
514 		__wait_on_bit(page_waitqueue(page), &wait, sync_page,
515 							TASK_UNINTERRUPTIBLE);
516 }
517 EXPORT_SYMBOL(wait_on_page_bit);
518 
519 /**
520  * unlock_page - unlock a locked page
521  * @page: the page
522  *
523  * Unlocks the page and wakes up sleepers in ___wait_on_page_locked().
524  * Also wakes sleepers in wait_on_page_writeback() because the wakeup
525  * mechananism between PageLocked pages and PageWriteback pages is shared.
526  * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep.
527  *
528  * The first mb is necessary to safely close the critical section opened by the
529  * TestSetPageLocked(), the second mb is necessary to enforce ordering between
530  * the clear_bit and the read of the waitqueue (to avoid SMP races with a
531  * parallel wait_on_page_locked()).
532  */
533 void fastcall unlock_page(struct page *page)
534 {
535 	smp_mb__before_clear_bit();
536 	if (!TestClearPageLocked(page))
537 		BUG();
538 	smp_mb__after_clear_bit();
539 	wake_up_page(page, PG_locked);
540 }
541 EXPORT_SYMBOL(unlock_page);
542 
543 /**
544  * end_page_writeback - end writeback against a page
545  * @page: the page
546  */
547 void end_page_writeback(struct page *page)
548 {
549 	if (!TestClearPageReclaim(page) || rotate_reclaimable_page(page)) {
550 		if (!test_clear_page_writeback(page))
551 			BUG();
552 	}
553 	smp_mb__after_clear_bit();
554 	wake_up_page(page, PG_writeback);
555 }
556 EXPORT_SYMBOL(end_page_writeback);
557 
558 /**
559  * __lock_page - get a lock on the page, assuming we need to sleep to get it
560  * @page: the page to lock
561  *
562  * Ugly. Running sync_page() in state TASK_UNINTERRUPTIBLE is scary.  If some
563  * random driver's requestfn sets TASK_RUNNING, we could busywait.  However
564  * chances are that on the second loop, the block layer's plug list is empty,
565  * so sync_page() will then return in state TASK_UNINTERRUPTIBLE.
566  */
567 void fastcall __lock_page(struct page *page)
568 {
569 	DEFINE_WAIT_BIT(wait, &page->flags, PG_locked);
570 
571 	__wait_on_bit_lock(page_waitqueue(page), &wait, sync_page,
572 							TASK_UNINTERRUPTIBLE);
573 }
574 EXPORT_SYMBOL(__lock_page);
575 
576 /*
577  * Variant of lock_page that does not require the caller to hold a reference
578  * on the page's mapping.
579  */
580 void fastcall __lock_page_nosync(struct page *page)
581 {
582 	DEFINE_WAIT_BIT(wait, &page->flags, PG_locked);
583 	__wait_on_bit_lock(page_waitqueue(page), &wait, __sleep_on_page_lock,
584 							TASK_UNINTERRUPTIBLE);
585 }
586 
587 /**
588  * find_get_page - find and get a page reference
589  * @mapping: the address_space to search
590  * @offset: the page index
591  *
592  * Is there a pagecache struct page at the given (mapping, offset) tuple?
593  * If yes, increment its refcount and return it; if no, return NULL.
594  */
595 struct page * find_get_page(struct address_space *mapping, unsigned long offset)
596 {
597 	struct page *page;
598 
599 	read_lock_irq(&mapping->tree_lock);
600 	page = radix_tree_lookup(&mapping->page_tree, offset);
601 	if (page)
602 		page_cache_get(page);
603 	read_unlock_irq(&mapping->tree_lock);
604 	return page;
605 }
606 EXPORT_SYMBOL(find_get_page);
607 
608 /**
609  * find_lock_page - locate, pin and lock a pagecache page
610  * @mapping: the address_space to search
611  * @offset: the page index
612  *
613  * Locates the desired pagecache page, locks it, increments its reference
614  * count and returns its address.
615  *
616  * Returns zero if the page was not present. find_lock_page() may sleep.
617  */
618 struct page *find_lock_page(struct address_space *mapping,
619 				unsigned long offset)
620 {
621 	struct page *page;
622 
623 	read_lock_irq(&mapping->tree_lock);
624 repeat:
625 	page = radix_tree_lookup(&mapping->page_tree, offset);
626 	if (page) {
627 		page_cache_get(page);
628 		if (TestSetPageLocked(page)) {
629 			read_unlock_irq(&mapping->tree_lock);
630 			__lock_page(page);
631 			read_lock_irq(&mapping->tree_lock);
632 
633 			/* Has the page been truncated while we slept? */
634 			if (unlikely(page->mapping != mapping ||
635 				     page->index != offset)) {
636 				unlock_page(page);
637 				page_cache_release(page);
638 				goto repeat;
639 			}
640 		}
641 	}
642 	read_unlock_irq(&mapping->tree_lock);
643 	return page;
644 }
645 EXPORT_SYMBOL(find_lock_page);
646 
647 /**
648  * find_or_create_page - locate or add a pagecache page
649  * @mapping: the page's address_space
650  * @index: the page's index into the mapping
651  * @gfp_mask: page allocation mode
652  *
653  * Locates a page in the pagecache.  If the page is not present, a new page
654  * is allocated using @gfp_mask and is added to the pagecache and to the VM's
655  * LRU list.  The returned page is locked and has its reference count
656  * incremented.
657  *
658  * find_or_create_page() may sleep, even if @gfp_flags specifies an atomic
659  * allocation!
660  *
661  * find_or_create_page() returns the desired page's address, or zero on
662  * memory exhaustion.
663  */
664 struct page *find_or_create_page(struct address_space *mapping,
665 		unsigned long index, gfp_t gfp_mask)
666 {
667 	struct page *page, *cached_page = NULL;
668 	int err;
669 repeat:
670 	page = find_lock_page(mapping, index);
671 	if (!page) {
672 		if (!cached_page) {
673 			cached_page = alloc_page(gfp_mask);
674 			if (!cached_page)
675 				return NULL;
676 		}
677 		err = add_to_page_cache_lru(cached_page, mapping,
678 					index, gfp_mask);
679 		if (!err) {
680 			page = cached_page;
681 			cached_page = NULL;
682 		} else if (err == -EEXIST)
683 			goto repeat;
684 	}
685 	if (cached_page)
686 		page_cache_release(cached_page);
687 	return page;
688 }
689 EXPORT_SYMBOL(find_or_create_page);
690 
691 /**
692  * find_get_pages - gang pagecache lookup
693  * @mapping:	The address_space to search
694  * @start:	The starting page index
695  * @nr_pages:	The maximum number of pages
696  * @pages:	Where the resulting pages are placed
697  *
698  * find_get_pages() will search for and return a group of up to
699  * @nr_pages pages in the mapping.  The pages are placed at @pages.
700  * find_get_pages() takes a reference against the returned pages.
701  *
702  * The search returns a group of mapping-contiguous pages with ascending
703  * indexes.  There may be holes in the indices due to not-present pages.
704  *
705  * find_get_pages() returns the number of pages which were found.
706  */
707 unsigned find_get_pages(struct address_space *mapping, pgoff_t start,
708 			    unsigned int nr_pages, struct page **pages)
709 {
710 	unsigned int i;
711 	unsigned int ret;
712 
713 	read_lock_irq(&mapping->tree_lock);
714 	ret = radix_tree_gang_lookup(&mapping->page_tree,
715 				(void **)pages, start, nr_pages);
716 	for (i = 0; i < ret; i++)
717 		page_cache_get(pages[i]);
718 	read_unlock_irq(&mapping->tree_lock);
719 	return ret;
720 }
721 
722 /**
723  * find_get_pages_contig - gang contiguous pagecache lookup
724  * @mapping:	The address_space to search
725  * @index:	The starting page index
726  * @nr_pages:	The maximum number of pages
727  * @pages:	Where the resulting pages are placed
728  *
729  * find_get_pages_contig() works exactly like find_get_pages(), except
730  * that the returned number of pages are guaranteed to be contiguous.
731  *
732  * find_get_pages_contig() returns the number of pages which were found.
733  */
734 unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index,
735 			       unsigned int nr_pages, struct page **pages)
736 {
737 	unsigned int i;
738 	unsigned int ret;
739 
740 	read_lock_irq(&mapping->tree_lock);
741 	ret = radix_tree_gang_lookup(&mapping->page_tree,
742 				(void **)pages, index, nr_pages);
743 	for (i = 0; i < ret; i++) {
744 		if (pages[i]->mapping == NULL || pages[i]->index != index)
745 			break;
746 
747 		page_cache_get(pages[i]);
748 		index++;
749 	}
750 	read_unlock_irq(&mapping->tree_lock);
751 	return i;
752 }
753 EXPORT_SYMBOL(find_get_pages_contig);
754 
755 /**
756  * find_get_pages_tag - find and return pages that match @tag
757  * @mapping:	the address_space to search
758  * @index:	the starting page index
759  * @tag:	the tag index
760  * @nr_pages:	the maximum number of pages
761  * @pages:	where the resulting pages are placed
762  *
763  * Like find_get_pages, except we only return pages which are tagged with
764  * @tag.   We update @index to index the next page for the traversal.
765  */
766 unsigned find_get_pages_tag(struct address_space *mapping, pgoff_t *index,
767 			int tag, unsigned int nr_pages, struct page **pages)
768 {
769 	unsigned int i;
770 	unsigned int ret;
771 
772 	read_lock_irq(&mapping->tree_lock);
773 	ret = radix_tree_gang_lookup_tag(&mapping->page_tree,
774 				(void **)pages, *index, nr_pages, tag);
775 	for (i = 0; i < ret; i++)
776 		page_cache_get(pages[i]);
777 	if (ret)
778 		*index = pages[ret - 1]->index + 1;
779 	read_unlock_irq(&mapping->tree_lock);
780 	return ret;
781 }
782 EXPORT_SYMBOL(find_get_pages_tag);
783 
784 /**
785  * grab_cache_page_nowait - returns locked page at given index in given cache
786  * @mapping: target address_space
787  * @index: the page index
788  *
789  * Same as grab_cache_page(), but do not wait if the page is unavailable.
790  * This is intended for speculative data generators, where the data can
791  * be regenerated if the page couldn't be grabbed.  This routine should
792  * be safe to call while holding the lock for another page.
793  *
794  * Clear __GFP_FS when allocating the page to avoid recursion into the fs
795  * and deadlock against the caller's locked page.
796  */
797 struct page *
798 grab_cache_page_nowait(struct address_space *mapping, unsigned long index)
799 {
800 	struct page *page = find_get_page(mapping, index);
801 
802 	if (page) {
803 		if (!TestSetPageLocked(page))
804 			return page;
805 		page_cache_release(page);
806 		return NULL;
807 	}
808 	page = __page_cache_alloc(mapping_gfp_mask(mapping) & ~__GFP_FS);
809 	if (page && add_to_page_cache_lru(page, mapping, index, GFP_KERNEL)) {
810 		page_cache_release(page);
811 		page = NULL;
812 	}
813 	return page;
814 }
815 EXPORT_SYMBOL(grab_cache_page_nowait);
816 
817 /*
818  * CD/DVDs are error prone. When a medium error occurs, the driver may fail
819  * a _large_ part of the i/o request. Imagine the worst scenario:
820  *
821  *      ---R__________________________________________B__________
822  *         ^ reading here                             ^ bad block(assume 4k)
823  *
824  * read(R) => miss => readahead(R...B) => media error => frustrating retries
825  * => failing the whole request => read(R) => read(R+1) =>
826  * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
827  * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
828  * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
829  *
830  * It is going insane. Fix it by quickly scaling down the readahead size.
831  */
832 static void shrink_readahead_size_eio(struct file *filp,
833 					struct file_ra_state *ra)
834 {
835 	if (!ra->ra_pages)
836 		return;
837 
838 	ra->ra_pages /= 4;
839 }
840 
841 /**
842  * do_generic_mapping_read - generic file read routine
843  * @mapping:	address_space to be read
844  * @_ra:	file's readahead state
845  * @filp:	the file to read
846  * @ppos:	current file position
847  * @desc:	read_descriptor
848  * @actor:	read method
849  *
850  * This is a generic file read routine, and uses the
851  * mapping->a_ops->readpage() function for the actual low-level stuff.
852  *
853  * This is really ugly. But the goto's actually try to clarify some
854  * of the logic when it comes to error handling etc.
855  *
856  * Note the struct file* is only passed for the use of readpage.
857  * It may be NULL.
858  */
859 void do_generic_mapping_read(struct address_space *mapping,
860 			     struct file_ra_state *_ra,
861 			     struct file *filp,
862 			     loff_t *ppos,
863 			     read_descriptor_t *desc,
864 			     read_actor_t actor)
865 {
866 	struct inode *inode = mapping->host;
867 	unsigned long index;
868 	unsigned long end_index;
869 	unsigned long offset;
870 	unsigned long last_index;
871 	unsigned long next_index;
872 	unsigned long prev_index;
873 	unsigned int prev_offset;
874 	loff_t isize;
875 	struct page *cached_page;
876 	int error;
877 	struct file_ra_state ra = *_ra;
878 
879 	cached_page = NULL;
880 	index = *ppos >> PAGE_CACHE_SHIFT;
881 	next_index = index;
882 	prev_index = ra.prev_index;
883 	prev_offset = ra.prev_offset;
884 	last_index = (*ppos + desc->count + PAGE_CACHE_SIZE-1) >> PAGE_CACHE_SHIFT;
885 	offset = *ppos & ~PAGE_CACHE_MASK;
886 
887 	isize = i_size_read(inode);
888 	if (!isize)
889 		goto out;
890 
891 	end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
892 	for (;;) {
893 		struct page *page;
894 		unsigned long nr, ret;
895 
896 		/* nr is the maximum number of bytes to copy from this page */
897 		nr = PAGE_CACHE_SIZE;
898 		if (index >= end_index) {
899 			if (index > end_index)
900 				goto out;
901 			nr = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;
902 			if (nr <= offset) {
903 				goto out;
904 			}
905 		}
906 		nr = nr - offset;
907 
908 		cond_resched();
909 		if (index == next_index)
910 			next_index = page_cache_readahead(mapping, &ra, filp,
911 					index, last_index - index);
912 
913 find_page:
914 		page = find_get_page(mapping, index);
915 		if (unlikely(page == NULL)) {
916 			handle_ra_miss(mapping, &ra, index);
917 			goto no_cached_page;
918 		}
919 		if (!PageUptodate(page))
920 			goto page_not_up_to_date;
921 page_ok:
922 
923 		/* If users can be writing to this page using arbitrary
924 		 * virtual addresses, take care about potential aliasing
925 		 * before reading the page on the kernel side.
926 		 */
927 		if (mapping_writably_mapped(mapping))
928 			flush_dcache_page(page);
929 
930 		/*
931 		 * When a sequential read accesses a page several times,
932 		 * only mark it as accessed the first time.
933 		 */
934 		if (prev_index != index || offset != prev_offset)
935 			mark_page_accessed(page);
936 		prev_index = index;
937 
938 		/*
939 		 * Ok, we have the page, and it's up-to-date, so
940 		 * now we can copy it to user space...
941 		 *
942 		 * The actor routine returns how many bytes were actually used..
943 		 * NOTE! This may not be the same as how much of a user buffer
944 		 * we filled up (we may be padding etc), so we can only update
945 		 * "pos" here (the actor routine has to update the user buffer
946 		 * pointers and the remaining count).
947 		 */
948 		ret = actor(desc, page, offset, nr);
949 		offset += ret;
950 		index += offset >> PAGE_CACHE_SHIFT;
951 		offset &= ~PAGE_CACHE_MASK;
952 		prev_offset = offset;
953 		ra.prev_offset = offset;
954 
955 		page_cache_release(page);
956 		if (ret == nr && desc->count)
957 			continue;
958 		goto out;
959 
960 page_not_up_to_date:
961 		/* Get exclusive access to the page ... */
962 		lock_page(page);
963 
964 		/* Did it get truncated before we got the lock? */
965 		if (!page->mapping) {
966 			unlock_page(page);
967 			page_cache_release(page);
968 			continue;
969 		}
970 
971 		/* Did somebody else fill it already? */
972 		if (PageUptodate(page)) {
973 			unlock_page(page);
974 			goto page_ok;
975 		}
976 
977 readpage:
978 		/* Start the actual read. The read will unlock the page. */
979 		error = mapping->a_ops->readpage(filp, page);
980 
981 		if (unlikely(error)) {
982 			if (error == AOP_TRUNCATED_PAGE) {
983 				page_cache_release(page);
984 				goto find_page;
985 			}
986 			goto readpage_error;
987 		}
988 
989 		if (!PageUptodate(page)) {
990 			lock_page(page);
991 			if (!PageUptodate(page)) {
992 				if (page->mapping == NULL) {
993 					/*
994 					 * invalidate_inode_pages got it
995 					 */
996 					unlock_page(page);
997 					page_cache_release(page);
998 					goto find_page;
999 				}
1000 				unlock_page(page);
1001 				error = -EIO;
1002 				shrink_readahead_size_eio(filp, &ra);
1003 				goto readpage_error;
1004 			}
1005 			unlock_page(page);
1006 		}
1007 
1008 		/*
1009 		 * i_size must be checked after we have done ->readpage.
1010 		 *
1011 		 * Checking i_size after the readpage allows us to calculate
1012 		 * the correct value for "nr", which means the zero-filled
1013 		 * part of the page is not copied back to userspace (unless
1014 		 * another truncate extends the file - this is desired though).
1015 		 */
1016 		isize = i_size_read(inode);
1017 		end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
1018 		if (unlikely(!isize || index > end_index)) {
1019 			page_cache_release(page);
1020 			goto out;
1021 		}
1022 
1023 		/* nr is the maximum number of bytes to copy from this page */
1024 		nr = PAGE_CACHE_SIZE;
1025 		if (index == end_index) {
1026 			nr = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;
1027 			if (nr <= offset) {
1028 				page_cache_release(page);
1029 				goto out;
1030 			}
1031 		}
1032 		nr = nr - offset;
1033 		goto page_ok;
1034 
1035 readpage_error:
1036 		/* UHHUH! A synchronous read error occurred. Report it */
1037 		desc->error = error;
1038 		page_cache_release(page);
1039 		goto out;
1040 
1041 no_cached_page:
1042 		/*
1043 		 * Ok, it wasn't cached, so we need to create a new
1044 		 * page..
1045 		 */
1046 		if (!cached_page) {
1047 			cached_page = page_cache_alloc_cold(mapping);
1048 			if (!cached_page) {
1049 				desc->error = -ENOMEM;
1050 				goto out;
1051 			}
1052 		}
1053 		error = add_to_page_cache_lru(cached_page, mapping,
1054 						index, GFP_KERNEL);
1055 		if (error) {
1056 			if (error == -EEXIST)
1057 				goto find_page;
1058 			desc->error = error;
1059 			goto out;
1060 		}
1061 		page = cached_page;
1062 		cached_page = NULL;
1063 		goto readpage;
1064 	}
1065 
1066 out:
1067 	*_ra = ra;
1068 
1069 	*ppos = ((loff_t) index << PAGE_CACHE_SHIFT) + offset;
1070 	if (cached_page)
1071 		page_cache_release(cached_page);
1072 	if (filp)
1073 		file_accessed(filp);
1074 }
1075 EXPORT_SYMBOL(do_generic_mapping_read);
1076 
1077 int file_read_actor(read_descriptor_t *desc, struct page *page,
1078 			unsigned long offset, unsigned long size)
1079 {
1080 	char *kaddr;
1081 	unsigned long left, count = desc->count;
1082 
1083 	if (size > count)
1084 		size = count;
1085 
1086 	/*
1087 	 * Faults on the destination of a read are common, so do it before
1088 	 * taking the kmap.
1089 	 */
1090 	if (!fault_in_pages_writeable(desc->arg.buf, size)) {
1091 		kaddr = kmap_atomic(page, KM_USER0);
1092 		left = __copy_to_user_inatomic(desc->arg.buf,
1093 						kaddr + offset, size);
1094 		kunmap_atomic(kaddr, KM_USER0);
1095 		if (left == 0)
1096 			goto success;
1097 	}
1098 
1099 	/* Do it the slow way */
1100 	kaddr = kmap(page);
1101 	left = __copy_to_user(desc->arg.buf, kaddr + offset, size);
1102 	kunmap(page);
1103 
1104 	if (left) {
1105 		size -= left;
1106 		desc->error = -EFAULT;
1107 	}
1108 success:
1109 	desc->count = count - size;
1110 	desc->written += size;
1111 	desc->arg.buf += size;
1112 	return size;
1113 }
1114 
1115 /*
1116  * Performs necessary checks before doing a write
1117  * @iov:	io vector request
1118  * @nr_segs:	number of segments in the iovec
1119  * @count:	number of bytes to write
1120  * @access_flags: type of access: %VERIFY_READ or %VERIFY_WRITE
1121  *
1122  * Adjust number of segments and amount of bytes to write (nr_segs should be
1123  * properly initialized first). Returns appropriate error code that caller
1124  * should return or zero in case that write should be allowed.
1125  */
1126 int generic_segment_checks(const struct iovec *iov,
1127 			unsigned long *nr_segs, size_t *count, int access_flags)
1128 {
1129 	unsigned long   seg;
1130 	size_t cnt = 0;
1131 	for (seg = 0; seg < *nr_segs; seg++) {
1132 		const struct iovec *iv = &iov[seg];
1133 
1134 		/*
1135 		 * If any segment has a negative length, or the cumulative
1136 		 * length ever wraps negative then return -EINVAL.
1137 		 */
1138 		cnt += iv->iov_len;
1139 		if (unlikely((ssize_t)(cnt|iv->iov_len) < 0))
1140 			return -EINVAL;
1141 		if (access_ok(access_flags, iv->iov_base, iv->iov_len))
1142 			continue;
1143 		if (seg == 0)
1144 			return -EFAULT;
1145 		*nr_segs = seg;
1146 		cnt -= iv->iov_len;	/* This segment is no good */
1147 		break;
1148 	}
1149 	*count = cnt;
1150 	return 0;
1151 }
1152 EXPORT_SYMBOL(generic_segment_checks);
1153 
1154 /**
1155  * generic_file_aio_read - generic filesystem read routine
1156  * @iocb:	kernel I/O control block
1157  * @iov:	io vector request
1158  * @nr_segs:	number of segments in the iovec
1159  * @pos:	current file position
1160  *
1161  * This is the "read()" routine for all filesystems
1162  * that can use the page cache directly.
1163  */
1164 ssize_t
1165 generic_file_aio_read(struct kiocb *iocb, const struct iovec *iov,
1166 		unsigned long nr_segs, loff_t pos)
1167 {
1168 	struct file *filp = iocb->ki_filp;
1169 	ssize_t retval;
1170 	unsigned long seg;
1171 	size_t count;
1172 	loff_t *ppos = &iocb->ki_pos;
1173 
1174 	count = 0;
1175 	retval = generic_segment_checks(iov, &nr_segs, &count, VERIFY_WRITE);
1176 	if (retval)
1177 		return retval;
1178 
1179 	/* coalesce the iovecs and go direct-to-BIO for O_DIRECT */
1180 	if (filp->f_flags & O_DIRECT) {
1181 		loff_t size;
1182 		struct address_space *mapping;
1183 		struct inode *inode;
1184 
1185 		mapping = filp->f_mapping;
1186 		inode = mapping->host;
1187 		retval = 0;
1188 		if (!count)
1189 			goto out; /* skip atime */
1190 		size = i_size_read(inode);
1191 		if (pos < size) {
1192 			retval = generic_file_direct_IO(READ, iocb,
1193 						iov, pos, nr_segs);
1194 			if (retval > 0)
1195 				*ppos = pos + retval;
1196 		}
1197 		if (likely(retval != 0)) {
1198 			file_accessed(filp);
1199 			goto out;
1200 		}
1201 	}
1202 
1203 	retval = 0;
1204 	if (count) {
1205 		for (seg = 0; seg < nr_segs; seg++) {
1206 			read_descriptor_t desc;
1207 
1208 			desc.written = 0;
1209 			desc.arg.buf = iov[seg].iov_base;
1210 			desc.count = iov[seg].iov_len;
1211 			if (desc.count == 0)
1212 				continue;
1213 			desc.error = 0;
1214 			do_generic_file_read(filp,ppos,&desc,file_read_actor);
1215 			retval += desc.written;
1216 			if (desc.error) {
1217 				retval = retval ?: desc.error;
1218 				break;
1219 			}
1220 		}
1221 	}
1222 out:
1223 	return retval;
1224 }
1225 EXPORT_SYMBOL(generic_file_aio_read);
1226 
1227 int file_send_actor(read_descriptor_t * desc, struct page *page, unsigned long offset, unsigned long size)
1228 {
1229 	ssize_t written;
1230 	unsigned long count = desc->count;
1231 	struct file *file = desc->arg.data;
1232 
1233 	if (size > count)
1234 		size = count;
1235 
1236 	written = file->f_op->sendpage(file, page, offset,
1237 				       size, &file->f_pos, size<count);
1238 	if (written < 0) {
1239 		desc->error = written;
1240 		written = 0;
1241 	}
1242 	desc->count = count - written;
1243 	desc->written += written;
1244 	return written;
1245 }
1246 
1247 ssize_t generic_file_sendfile(struct file *in_file, loff_t *ppos,
1248 			 size_t count, read_actor_t actor, void *target)
1249 {
1250 	read_descriptor_t desc;
1251 
1252 	if (!count)
1253 		return 0;
1254 
1255 	desc.written = 0;
1256 	desc.count = count;
1257 	desc.arg.data = target;
1258 	desc.error = 0;
1259 
1260 	do_generic_file_read(in_file, ppos, &desc, actor);
1261 	if (desc.written)
1262 		return desc.written;
1263 	return desc.error;
1264 }
1265 EXPORT_SYMBOL(generic_file_sendfile);
1266 
1267 static ssize_t
1268 do_readahead(struct address_space *mapping, struct file *filp,
1269 	     unsigned long index, unsigned long nr)
1270 {
1271 	if (!mapping || !mapping->a_ops || !mapping->a_ops->readpage)
1272 		return -EINVAL;
1273 
1274 	force_page_cache_readahead(mapping, filp, index,
1275 					max_sane_readahead(nr));
1276 	return 0;
1277 }
1278 
1279 asmlinkage ssize_t sys_readahead(int fd, loff_t offset, size_t count)
1280 {
1281 	ssize_t ret;
1282 	struct file *file;
1283 
1284 	ret = -EBADF;
1285 	file = fget(fd);
1286 	if (file) {
1287 		if (file->f_mode & FMODE_READ) {
1288 			struct address_space *mapping = file->f_mapping;
1289 			unsigned long start = offset >> PAGE_CACHE_SHIFT;
1290 			unsigned long end = (offset + count - 1) >> PAGE_CACHE_SHIFT;
1291 			unsigned long len = end - start + 1;
1292 			ret = do_readahead(mapping, file, start, len);
1293 		}
1294 		fput(file);
1295 	}
1296 	return ret;
1297 }
1298 
1299 #ifdef CONFIG_MMU
1300 static int FASTCALL(page_cache_read(struct file * file, unsigned long offset));
1301 /**
1302  * page_cache_read - adds requested page to the page cache if not already there
1303  * @file:	file to read
1304  * @offset:	page index
1305  *
1306  * This adds the requested page to the page cache if it isn't already there,
1307  * and schedules an I/O to read in its contents from disk.
1308  */
1309 static int fastcall page_cache_read(struct file * file, unsigned long offset)
1310 {
1311 	struct address_space *mapping = file->f_mapping;
1312 	struct page *page;
1313 	int ret;
1314 
1315 	do {
1316 		page = page_cache_alloc_cold(mapping);
1317 		if (!page)
1318 			return -ENOMEM;
1319 
1320 		ret = add_to_page_cache_lru(page, mapping, offset, GFP_KERNEL);
1321 		if (ret == 0)
1322 			ret = mapping->a_ops->readpage(file, page);
1323 		else if (ret == -EEXIST)
1324 			ret = 0; /* losing race to add is OK */
1325 
1326 		page_cache_release(page);
1327 
1328 	} while (ret == AOP_TRUNCATED_PAGE);
1329 
1330 	return ret;
1331 }
1332 
1333 #define MMAP_LOTSAMISS  (100)
1334 
1335 /**
1336  * filemap_nopage - read in file data for page fault handling
1337  * @area:	the applicable vm_area
1338  * @address:	target address to read in
1339  * @type:	returned with VM_FAULT_{MINOR,MAJOR} if not %NULL
1340  *
1341  * filemap_nopage() is invoked via the vma operations vector for a
1342  * mapped memory region to read in file data during a page fault.
1343  *
1344  * The goto's are kind of ugly, but this streamlines the normal case of having
1345  * it in the page cache, and handles the special cases reasonably without
1346  * having a lot of duplicated code.
1347  */
1348 struct page *filemap_nopage(struct vm_area_struct *area,
1349 				unsigned long address, int *type)
1350 {
1351 	int error;
1352 	struct file *file = area->vm_file;
1353 	struct address_space *mapping = file->f_mapping;
1354 	struct file_ra_state *ra = &file->f_ra;
1355 	struct inode *inode = mapping->host;
1356 	struct page *page;
1357 	unsigned long size, pgoff;
1358 	int did_readaround = 0, majmin = VM_FAULT_MINOR;
1359 
1360 	pgoff = ((address-area->vm_start) >> PAGE_CACHE_SHIFT) + area->vm_pgoff;
1361 
1362 retry_all:
1363 	size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1364 	if (pgoff >= size)
1365 		goto outside_data_content;
1366 
1367 	/* If we don't want any read-ahead, don't bother */
1368 	if (VM_RandomReadHint(area))
1369 		goto no_cached_page;
1370 
1371 	/*
1372 	 * The readahead code wants to be told about each and every page
1373 	 * so it can build and shrink its windows appropriately
1374 	 *
1375 	 * For sequential accesses, we use the generic readahead logic.
1376 	 */
1377 	if (VM_SequentialReadHint(area))
1378 		page_cache_readahead(mapping, ra, file, pgoff, 1);
1379 
1380 	/*
1381 	 * Do we have something in the page cache already?
1382 	 */
1383 retry_find:
1384 	page = find_get_page(mapping, pgoff);
1385 	if (!page) {
1386 		unsigned long ra_pages;
1387 
1388 		if (VM_SequentialReadHint(area)) {
1389 			handle_ra_miss(mapping, ra, pgoff);
1390 			goto no_cached_page;
1391 		}
1392 		ra->mmap_miss++;
1393 
1394 		/*
1395 		 * Do we miss much more than hit in this file? If so,
1396 		 * stop bothering with read-ahead. It will only hurt.
1397 		 */
1398 		if (ra->mmap_miss > ra->mmap_hit + MMAP_LOTSAMISS)
1399 			goto no_cached_page;
1400 
1401 		/*
1402 		 * To keep the pgmajfault counter straight, we need to
1403 		 * check did_readaround, as this is an inner loop.
1404 		 */
1405 		if (!did_readaround) {
1406 			majmin = VM_FAULT_MAJOR;
1407 			count_vm_event(PGMAJFAULT);
1408 		}
1409 		did_readaround = 1;
1410 		ra_pages = max_sane_readahead(file->f_ra.ra_pages);
1411 		if (ra_pages) {
1412 			pgoff_t start = 0;
1413 
1414 			if (pgoff > ra_pages / 2)
1415 				start = pgoff - ra_pages / 2;
1416 			do_page_cache_readahead(mapping, file, start, ra_pages);
1417 		}
1418 		page = find_get_page(mapping, pgoff);
1419 		if (!page)
1420 			goto no_cached_page;
1421 	}
1422 
1423 	if (!did_readaround)
1424 		ra->mmap_hit++;
1425 
1426 	/*
1427 	 * Ok, found a page in the page cache, now we need to check
1428 	 * that it's up-to-date.
1429 	 */
1430 	if (!PageUptodate(page))
1431 		goto page_not_uptodate;
1432 
1433 success:
1434 	/*
1435 	 * Found the page and have a reference on it.
1436 	 */
1437 	mark_page_accessed(page);
1438 	if (type)
1439 		*type = majmin;
1440 	return page;
1441 
1442 outside_data_content:
1443 	/*
1444 	 * An external ptracer can access pages that normally aren't
1445 	 * accessible..
1446 	 */
1447 	if (area->vm_mm == current->mm)
1448 		return NOPAGE_SIGBUS;
1449 	/* Fall through to the non-read-ahead case */
1450 no_cached_page:
1451 	/*
1452 	 * We're only likely to ever get here if MADV_RANDOM is in
1453 	 * effect.
1454 	 */
1455 	error = page_cache_read(file, pgoff);
1456 
1457 	/*
1458 	 * The page we want has now been added to the page cache.
1459 	 * In the unlikely event that someone removed it in the
1460 	 * meantime, we'll just come back here and read it again.
1461 	 */
1462 	if (error >= 0)
1463 		goto retry_find;
1464 
1465 	/*
1466 	 * An error return from page_cache_read can result if the
1467 	 * system is low on memory, or a problem occurs while trying
1468 	 * to schedule I/O.
1469 	 */
1470 	if (error == -ENOMEM)
1471 		return NOPAGE_OOM;
1472 	return NOPAGE_SIGBUS;
1473 
1474 page_not_uptodate:
1475 	if (!did_readaround) {
1476 		majmin = VM_FAULT_MAJOR;
1477 		count_vm_event(PGMAJFAULT);
1478 	}
1479 
1480 	/*
1481 	 * Umm, take care of errors if the page isn't up-to-date.
1482 	 * Try to re-read it _once_. We do this synchronously,
1483 	 * because there really aren't any performance issues here
1484 	 * and we need to check for errors.
1485 	 */
1486 	lock_page(page);
1487 
1488 	/* Somebody truncated the page on us? */
1489 	if (!page->mapping) {
1490 		unlock_page(page);
1491 		page_cache_release(page);
1492 		goto retry_all;
1493 	}
1494 
1495 	/* Somebody else successfully read it in? */
1496 	if (PageUptodate(page)) {
1497 		unlock_page(page);
1498 		goto success;
1499 	}
1500 	ClearPageError(page);
1501 	error = mapping->a_ops->readpage(file, page);
1502 	if (!error) {
1503 		wait_on_page_locked(page);
1504 		if (PageUptodate(page))
1505 			goto success;
1506 	} else if (error == AOP_TRUNCATED_PAGE) {
1507 		page_cache_release(page);
1508 		goto retry_find;
1509 	}
1510 
1511 	/*
1512 	 * Things didn't work out. Return zero to tell the
1513 	 * mm layer so, possibly freeing the page cache page first.
1514 	 */
1515 	shrink_readahead_size_eio(file, ra);
1516 	page_cache_release(page);
1517 	return NOPAGE_SIGBUS;
1518 }
1519 EXPORT_SYMBOL(filemap_nopage);
1520 
1521 static struct page * filemap_getpage(struct file *file, unsigned long pgoff,
1522 					int nonblock)
1523 {
1524 	struct address_space *mapping = file->f_mapping;
1525 	struct page *page;
1526 	int error;
1527 
1528 	/*
1529 	 * Do we have something in the page cache already?
1530 	 */
1531 retry_find:
1532 	page = find_get_page(mapping, pgoff);
1533 	if (!page) {
1534 		if (nonblock)
1535 			return NULL;
1536 		goto no_cached_page;
1537 	}
1538 
1539 	/*
1540 	 * Ok, found a page in the page cache, now we need to check
1541 	 * that it's up-to-date.
1542 	 */
1543 	if (!PageUptodate(page)) {
1544 		if (nonblock) {
1545 			page_cache_release(page);
1546 			return NULL;
1547 		}
1548 		goto page_not_uptodate;
1549 	}
1550 
1551 success:
1552 	/*
1553 	 * Found the page and have a reference on it.
1554 	 */
1555 	mark_page_accessed(page);
1556 	return page;
1557 
1558 no_cached_page:
1559 	error = page_cache_read(file, pgoff);
1560 
1561 	/*
1562 	 * The page we want has now been added to the page cache.
1563 	 * In the unlikely event that someone removed it in the
1564 	 * meantime, we'll just come back here and read it again.
1565 	 */
1566 	if (error >= 0)
1567 		goto retry_find;
1568 
1569 	/*
1570 	 * An error return from page_cache_read can result if the
1571 	 * system is low on memory, or a problem occurs while trying
1572 	 * to schedule I/O.
1573 	 */
1574 	return NULL;
1575 
1576 page_not_uptodate:
1577 	lock_page(page);
1578 
1579 	/* Did it get truncated while we waited for it? */
1580 	if (!page->mapping) {
1581 		unlock_page(page);
1582 		goto err;
1583 	}
1584 
1585 	/* Did somebody else get it up-to-date? */
1586 	if (PageUptodate(page)) {
1587 		unlock_page(page);
1588 		goto success;
1589 	}
1590 
1591 	error = mapping->a_ops->readpage(file, page);
1592 	if (!error) {
1593 		wait_on_page_locked(page);
1594 		if (PageUptodate(page))
1595 			goto success;
1596 	} else if (error == AOP_TRUNCATED_PAGE) {
1597 		page_cache_release(page);
1598 		goto retry_find;
1599 	}
1600 
1601 	/*
1602 	 * Umm, take care of errors if the page isn't up-to-date.
1603 	 * Try to re-read it _once_. We do this synchronously,
1604 	 * because there really aren't any performance issues here
1605 	 * and we need to check for errors.
1606 	 */
1607 	lock_page(page);
1608 
1609 	/* Somebody truncated the page on us? */
1610 	if (!page->mapping) {
1611 		unlock_page(page);
1612 		goto err;
1613 	}
1614 	/* Somebody else successfully read it in? */
1615 	if (PageUptodate(page)) {
1616 		unlock_page(page);
1617 		goto success;
1618 	}
1619 
1620 	ClearPageError(page);
1621 	error = mapping->a_ops->readpage(file, page);
1622 	if (!error) {
1623 		wait_on_page_locked(page);
1624 		if (PageUptodate(page))
1625 			goto success;
1626 	} else if (error == AOP_TRUNCATED_PAGE) {
1627 		page_cache_release(page);
1628 		goto retry_find;
1629 	}
1630 
1631 	/*
1632 	 * Things didn't work out. Return zero to tell the
1633 	 * mm layer so, possibly freeing the page cache page first.
1634 	 */
1635 err:
1636 	page_cache_release(page);
1637 
1638 	return NULL;
1639 }
1640 
1641 int filemap_populate(struct vm_area_struct *vma, unsigned long addr,
1642 		unsigned long len, pgprot_t prot, unsigned long pgoff,
1643 		int nonblock)
1644 {
1645 	struct file *file = vma->vm_file;
1646 	struct address_space *mapping = file->f_mapping;
1647 	struct inode *inode = mapping->host;
1648 	unsigned long size;
1649 	struct mm_struct *mm = vma->vm_mm;
1650 	struct page *page;
1651 	int err;
1652 
1653 	if (!nonblock)
1654 		force_page_cache_readahead(mapping, vma->vm_file,
1655 					pgoff, len >> PAGE_CACHE_SHIFT);
1656 
1657 repeat:
1658 	size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1659 	if (pgoff + (len >> PAGE_CACHE_SHIFT) > size)
1660 		return -EINVAL;
1661 
1662 	page = filemap_getpage(file, pgoff, nonblock);
1663 
1664 	/* XXX: This is wrong, a filesystem I/O error may have happened. Fix that as
1665 	 * done in shmem_populate calling shmem_getpage */
1666 	if (!page && !nonblock)
1667 		return -ENOMEM;
1668 
1669 	if (page) {
1670 		err = install_page(mm, vma, addr, page, prot);
1671 		if (err) {
1672 			page_cache_release(page);
1673 			return err;
1674 		}
1675 	} else if (vma->vm_flags & VM_NONLINEAR) {
1676 		/* No page was found just because we can't read it in now (being
1677 		 * here implies nonblock != 0), but the page may exist, so set
1678 		 * the PTE to fault it in later. */
1679 		err = install_file_pte(mm, vma, addr, pgoff, prot);
1680 		if (err)
1681 			return err;
1682 	}
1683 
1684 	len -= PAGE_SIZE;
1685 	addr += PAGE_SIZE;
1686 	pgoff++;
1687 	if (len)
1688 		goto repeat;
1689 
1690 	return 0;
1691 }
1692 EXPORT_SYMBOL(filemap_populate);
1693 
1694 struct vm_operations_struct generic_file_vm_ops = {
1695 	.nopage		= filemap_nopage,
1696 	.populate	= filemap_populate,
1697 };
1698 
1699 /* This is used for a general mmap of a disk file */
1700 
1701 int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
1702 {
1703 	struct address_space *mapping = file->f_mapping;
1704 
1705 	if (!mapping->a_ops->readpage)
1706 		return -ENOEXEC;
1707 	file_accessed(file);
1708 	vma->vm_ops = &generic_file_vm_ops;
1709 	return 0;
1710 }
1711 
1712 /*
1713  * This is for filesystems which do not implement ->writepage.
1714  */
1715 int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
1716 {
1717 	if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE))
1718 		return -EINVAL;
1719 	return generic_file_mmap(file, vma);
1720 }
1721 #else
1722 int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
1723 {
1724 	return -ENOSYS;
1725 }
1726 int generic_file_readonly_mmap(struct file * file, struct vm_area_struct * vma)
1727 {
1728 	return -ENOSYS;
1729 }
1730 #endif /* CONFIG_MMU */
1731 
1732 EXPORT_SYMBOL(generic_file_mmap);
1733 EXPORT_SYMBOL(generic_file_readonly_mmap);
1734 
1735 static struct page *__read_cache_page(struct address_space *mapping,
1736 				unsigned long index,
1737 				int (*filler)(void *,struct page*),
1738 				void *data)
1739 {
1740 	struct page *page, *cached_page = NULL;
1741 	int err;
1742 repeat:
1743 	page = find_get_page(mapping, index);
1744 	if (!page) {
1745 		if (!cached_page) {
1746 			cached_page = page_cache_alloc_cold(mapping);
1747 			if (!cached_page)
1748 				return ERR_PTR(-ENOMEM);
1749 		}
1750 		err = add_to_page_cache_lru(cached_page, mapping,
1751 					index, GFP_KERNEL);
1752 		if (err == -EEXIST)
1753 			goto repeat;
1754 		if (err < 0) {
1755 			/* Presumably ENOMEM for radix tree node */
1756 			page_cache_release(cached_page);
1757 			return ERR_PTR(err);
1758 		}
1759 		page = cached_page;
1760 		cached_page = NULL;
1761 		err = filler(data, page);
1762 		if (err < 0) {
1763 			page_cache_release(page);
1764 			page = ERR_PTR(err);
1765 		}
1766 	}
1767 	if (cached_page)
1768 		page_cache_release(cached_page);
1769 	return page;
1770 }
1771 
1772 /*
1773  * Same as read_cache_page, but don't wait for page to become unlocked
1774  * after submitting it to the filler.
1775  */
1776 struct page *read_cache_page_async(struct address_space *mapping,
1777 				unsigned long index,
1778 				int (*filler)(void *,struct page*),
1779 				void *data)
1780 {
1781 	struct page *page;
1782 	int err;
1783 
1784 retry:
1785 	page = __read_cache_page(mapping, index, filler, data);
1786 	if (IS_ERR(page))
1787 		return page;
1788 	mark_page_accessed(page);
1789 	if (PageUptodate(page))
1790 		goto out;
1791 
1792 	lock_page(page);
1793 	if (!page->mapping) {
1794 		unlock_page(page);
1795 		page_cache_release(page);
1796 		goto retry;
1797 	}
1798 	if (PageUptodate(page)) {
1799 		unlock_page(page);
1800 		goto out;
1801 	}
1802 	err = filler(data, page);
1803 	if (err < 0) {
1804 		page_cache_release(page);
1805 		return ERR_PTR(err);
1806 	}
1807 out:
1808 	mark_page_accessed(page);
1809 	return page;
1810 }
1811 EXPORT_SYMBOL(read_cache_page_async);
1812 
1813 /**
1814  * read_cache_page - read into page cache, fill it if needed
1815  * @mapping:	the page's address_space
1816  * @index:	the page index
1817  * @filler:	function to perform the read
1818  * @data:	destination for read data
1819  *
1820  * Read into the page cache. If a page already exists, and PageUptodate() is
1821  * not set, try to fill the page then wait for it to become unlocked.
1822  *
1823  * If the page does not get brought uptodate, return -EIO.
1824  */
1825 struct page *read_cache_page(struct address_space *mapping,
1826 				unsigned long index,
1827 				int (*filler)(void *,struct page*),
1828 				void *data)
1829 {
1830 	struct page *page;
1831 
1832 	page = read_cache_page_async(mapping, index, filler, data);
1833 	if (IS_ERR(page))
1834 		goto out;
1835 	wait_on_page_locked(page);
1836 	if (!PageUptodate(page)) {
1837 		page_cache_release(page);
1838 		page = ERR_PTR(-EIO);
1839 	}
1840  out:
1841 	return page;
1842 }
1843 EXPORT_SYMBOL(read_cache_page);
1844 
1845 /*
1846  * If the page was newly created, increment its refcount and add it to the
1847  * caller's lru-buffering pagevec.  This function is specifically for
1848  * generic_file_write().
1849  */
1850 static inline struct page *
1851 __grab_cache_page(struct address_space *mapping, unsigned long index,
1852 			struct page **cached_page, struct pagevec *lru_pvec)
1853 {
1854 	int err;
1855 	struct page *page;
1856 repeat:
1857 	page = find_lock_page(mapping, index);
1858 	if (!page) {
1859 		if (!*cached_page) {
1860 			*cached_page = page_cache_alloc(mapping);
1861 			if (!*cached_page)
1862 				return NULL;
1863 		}
1864 		err = add_to_page_cache(*cached_page, mapping,
1865 					index, GFP_KERNEL);
1866 		if (err == -EEXIST)
1867 			goto repeat;
1868 		if (err == 0) {
1869 			page = *cached_page;
1870 			page_cache_get(page);
1871 			if (!pagevec_add(lru_pvec, page))
1872 				__pagevec_lru_add(lru_pvec);
1873 			*cached_page = NULL;
1874 		}
1875 	}
1876 	return page;
1877 }
1878 
1879 /*
1880  * The logic we want is
1881  *
1882  *	if suid or (sgid and xgrp)
1883  *		remove privs
1884  */
1885 int should_remove_suid(struct dentry *dentry)
1886 {
1887 	mode_t mode = dentry->d_inode->i_mode;
1888 	int kill = 0;
1889 
1890 	/* suid always must be killed */
1891 	if (unlikely(mode & S_ISUID))
1892 		kill = ATTR_KILL_SUID;
1893 
1894 	/*
1895 	 * sgid without any exec bits is just a mandatory locking mark; leave
1896 	 * it alone.  If some exec bits are set, it's a real sgid; kill it.
1897 	 */
1898 	if (unlikely((mode & S_ISGID) && (mode & S_IXGRP)))
1899 		kill |= ATTR_KILL_SGID;
1900 
1901 	if (unlikely(kill && !capable(CAP_FSETID)))
1902 		return kill;
1903 
1904 	return 0;
1905 }
1906 EXPORT_SYMBOL(should_remove_suid);
1907 
1908 int __remove_suid(struct dentry *dentry, int kill)
1909 {
1910 	struct iattr newattrs;
1911 
1912 	newattrs.ia_valid = ATTR_FORCE | kill;
1913 	return notify_change(dentry, &newattrs);
1914 }
1915 
1916 int remove_suid(struct dentry *dentry)
1917 {
1918 	int kill = should_remove_suid(dentry);
1919 
1920 	if (unlikely(kill))
1921 		return __remove_suid(dentry, kill);
1922 
1923 	return 0;
1924 }
1925 EXPORT_SYMBOL(remove_suid);
1926 
1927 size_t
1928 __filemap_copy_from_user_iovec_inatomic(char *vaddr,
1929 			const struct iovec *iov, size_t base, size_t bytes)
1930 {
1931 	size_t copied = 0, left = 0;
1932 
1933 	while (bytes) {
1934 		char __user *buf = iov->iov_base + base;
1935 		int copy = min(bytes, iov->iov_len - base);
1936 
1937 		base = 0;
1938 		left = __copy_from_user_inatomic_nocache(vaddr, buf, copy);
1939 		copied += copy;
1940 		bytes -= copy;
1941 		vaddr += copy;
1942 		iov++;
1943 
1944 		if (unlikely(left))
1945 			break;
1946 	}
1947 	return copied - left;
1948 }
1949 
1950 /*
1951  * Performs necessary checks before doing a write
1952  *
1953  * Can adjust writing position or amount of bytes to write.
1954  * Returns appropriate error code that caller should return or
1955  * zero in case that write should be allowed.
1956  */
1957 inline int generic_write_checks(struct file *file, loff_t *pos, size_t *count, int isblk)
1958 {
1959 	struct inode *inode = file->f_mapping->host;
1960 	unsigned long limit = current->signal->rlim[RLIMIT_FSIZE].rlim_cur;
1961 
1962         if (unlikely(*pos < 0))
1963                 return -EINVAL;
1964 
1965 	if (!isblk) {
1966 		/* FIXME: this is for backwards compatibility with 2.4 */
1967 		if (file->f_flags & O_APPEND)
1968                         *pos = i_size_read(inode);
1969 
1970 		if (limit != RLIM_INFINITY) {
1971 			if (*pos >= limit) {
1972 				send_sig(SIGXFSZ, current, 0);
1973 				return -EFBIG;
1974 			}
1975 			if (*count > limit - (typeof(limit))*pos) {
1976 				*count = limit - (typeof(limit))*pos;
1977 			}
1978 		}
1979 	}
1980 
1981 	/*
1982 	 * LFS rule
1983 	 */
1984 	if (unlikely(*pos + *count > MAX_NON_LFS &&
1985 				!(file->f_flags & O_LARGEFILE))) {
1986 		if (*pos >= MAX_NON_LFS) {
1987 			send_sig(SIGXFSZ, current, 0);
1988 			return -EFBIG;
1989 		}
1990 		if (*count > MAX_NON_LFS - (unsigned long)*pos) {
1991 			*count = MAX_NON_LFS - (unsigned long)*pos;
1992 		}
1993 	}
1994 
1995 	/*
1996 	 * Are we about to exceed the fs block limit ?
1997 	 *
1998 	 * If we have written data it becomes a short write.  If we have
1999 	 * exceeded without writing data we send a signal and return EFBIG.
2000 	 * Linus frestrict idea will clean these up nicely..
2001 	 */
2002 	if (likely(!isblk)) {
2003 		if (unlikely(*pos >= inode->i_sb->s_maxbytes)) {
2004 			if (*count || *pos > inode->i_sb->s_maxbytes) {
2005 				send_sig(SIGXFSZ, current, 0);
2006 				return -EFBIG;
2007 			}
2008 			/* zero-length writes at ->s_maxbytes are OK */
2009 		}
2010 
2011 		if (unlikely(*pos + *count > inode->i_sb->s_maxbytes))
2012 			*count = inode->i_sb->s_maxbytes - *pos;
2013 	} else {
2014 #ifdef CONFIG_BLOCK
2015 		loff_t isize;
2016 		if (bdev_read_only(I_BDEV(inode)))
2017 			return -EPERM;
2018 		isize = i_size_read(inode);
2019 		if (*pos >= isize) {
2020 			if (*count || *pos > isize)
2021 				return -ENOSPC;
2022 		}
2023 
2024 		if (*pos + *count > isize)
2025 			*count = isize - *pos;
2026 #else
2027 		return -EPERM;
2028 #endif
2029 	}
2030 	return 0;
2031 }
2032 EXPORT_SYMBOL(generic_write_checks);
2033 
2034 ssize_t
2035 generic_file_direct_write(struct kiocb *iocb, const struct iovec *iov,
2036 		unsigned long *nr_segs, loff_t pos, loff_t *ppos,
2037 		size_t count, size_t ocount)
2038 {
2039 	struct file	*file = iocb->ki_filp;
2040 	struct address_space *mapping = file->f_mapping;
2041 	struct inode	*inode = mapping->host;
2042 	ssize_t		written;
2043 
2044 	if (count != ocount)
2045 		*nr_segs = iov_shorten((struct iovec *)iov, *nr_segs, count);
2046 
2047 	written = generic_file_direct_IO(WRITE, iocb, iov, pos, *nr_segs);
2048 	if (written > 0) {
2049 		loff_t end = pos + written;
2050 		if (end > i_size_read(inode) && !S_ISBLK(inode->i_mode)) {
2051 			i_size_write(inode,  end);
2052 			mark_inode_dirty(inode);
2053 		}
2054 		*ppos = end;
2055 	}
2056 
2057 	/*
2058 	 * Sync the fs metadata but not the minor inode changes and
2059 	 * of course not the data as we did direct DMA for the IO.
2060 	 * i_mutex is held, which protects generic_osync_inode() from
2061 	 * livelocking.  AIO O_DIRECT ops attempt to sync metadata here.
2062 	 */
2063 	if ((written >= 0 || written == -EIOCBQUEUED) &&
2064 	    ((file->f_flags & O_SYNC) || IS_SYNC(inode))) {
2065 		int err = generic_osync_inode(inode, mapping, OSYNC_METADATA);
2066 		if (err < 0)
2067 			written = err;
2068 	}
2069 	return written;
2070 }
2071 EXPORT_SYMBOL(generic_file_direct_write);
2072 
2073 ssize_t
2074 generic_file_buffered_write(struct kiocb *iocb, const struct iovec *iov,
2075 		unsigned long nr_segs, loff_t pos, loff_t *ppos,
2076 		size_t count, ssize_t written)
2077 {
2078 	struct file *file = iocb->ki_filp;
2079 	struct address_space * mapping = file->f_mapping;
2080 	const struct address_space_operations *a_ops = mapping->a_ops;
2081 	struct inode 	*inode = mapping->host;
2082 	long		status = 0;
2083 	struct page	*page;
2084 	struct page	*cached_page = NULL;
2085 	size_t		bytes;
2086 	struct pagevec	lru_pvec;
2087 	const struct iovec *cur_iov = iov; /* current iovec */
2088 	size_t		iov_base = 0;	   /* offset in the current iovec */
2089 	char __user	*buf;
2090 
2091 	pagevec_init(&lru_pvec, 0);
2092 
2093 	/*
2094 	 * handle partial DIO write.  Adjust cur_iov if needed.
2095 	 */
2096 	if (likely(nr_segs == 1))
2097 		buf = iov->iov_base + written;
2098 	else {
2099 		filemap_set_next_iovec(&cur_iov, &iov_base, written);
2100 		buf = cur_iov->iov_base + iov_base;
2101 	}
2102 
2103 	do {
2104 		unsigned long index;
2105 		unsigned long offset;
2106 		size_t copied;
2107 
2108 		offset = (pos & (PAGE_CACHE_SIZE -1)); /* Within page */
2109 		index = pos >> PAGE_CACHE_SHIFT;
2110 		bytes = PAGE_CACHE_SIZE - offset;
2111 
2112 		/* Limit the size of the copy to the caller's write size */
2113 		bytes = min(bytes, count);
2114 
2115 		/* We only need to worry about prefaulting when writes are from
2116 		 * user-space.  NFSd uses vfs_writev with several non-aligned
2117 		 * segments in the vector, and limiting to one segment a time is
2118 		 * a noticeable performance for re-write
2119 		 */
2120 		if (!segment_eq(get_fs(), KERNEL_DS)) {
2121 			/*
2122 			 * Limit the size of the copy to that of the current
2123 			 * segment, because fault_in_pages_readable() doesn't
2124 			 * know how to walk segments.
2125 			 */
2126 			bytes = min(bytes, cur_iov->iov_len - iov_base);
2127 
2128 			/*
2129 			 * Bring in the user page that we will copy from
2130 			 * _first_.  Otherwise there's a nasty deadlock on
2131 			 * copying from the same page as we're writing to,
2132 			 * without it being marked up-to-date.
2133 			 */
2134 			fault_in_pages_readable(buf, bytes);
2135 		}
2136 		page = __grab_cache_page(mapping,index,&cached_page,&lru_pvec);
2137 		if (!page) {
2138 			status = -ENOMEM;
2139 			break;
2140 		}
2141 
2142 		if (unlikely(bytes == 0)) {
2143 			status = 0;
2144 			copied = 0;
2145 			goto zero_length_segment;
2146 		}
2147 
2148 		status = a_ops->prepare_write(file, page, offset, offset+bytes);
2149 		if (unlikely(status)) {
2150 			loff_t isize = i_size_read(inode);
2151 
2152 			if (status != AOP_TRUNCATED_PAGE)
2153 				unlock_page(page);
2154 			page_cache_release(page);
2155 			if (status == AOP_TRUNCATED_PAGE)
2156 				continue;
2157 			/*
2158 			 * prepare_write() may have instantiated a few blocks
2159 			 * outside i_size.  Trim these off again.
2160 			 */
2161 			if (pos + bytes > isize)
2162 				vmtruncate(inode, isize);
2163 			break;
2164 		}
2165 		if (likely(nr_segs == 1))
2166 			copied = filemap_copy_from_user(page, offset,
2167 							buf, bytes);
2168 		else
2169 			copied = filemap_copy_from_user_iovec(page, offset,
2170 						cur_iov, iov_base, bytes);
2171 		flush_dcache_page(page);
2172 		status = a_ops->commit_write(file, page, offset, offset+bytes);
2173 		if (status == AOP_TRUNCATED_PAGE) {
2174 			page_cache_release(page);
2175 			continue;
2176 		}
2177 zero_length_segment:
2178 		if (likely(copied >= 0)) {
2179 			if (!status)
2180 				status = copied;
2181 
2182 			if (status >= 0) {
2183 				written += status;
2184 				count -= status;
2185 				pos += status;
2186 				buf += status;
2187 				if (unlikely(nr_segs > 1)) {
2188 					filemap_set_next_iovec(&cur_iov,
2189 							&iov_base, status);
2190 					if (count)
2191 						buf = cur_iov->iov_base +
2192 							iov_base;
2193 				} else {
2194 					iov_base += status;
2195 				}
2196 			}
2197 		}
2198 		if (unlikely(copied != bytes))
2199 			if (status >= 0)
2200 				status = -EFAULT;
2201 		unlock_page(page);
2202 		mark_page_accessed(page);
2203 		page_cache_release(page);
2204 		if (status < 0)
2205 			break;
2206 		balance_dirty_pages_ratelimited(mapping);
2207 		cond_resched();
2208 	} while (count);
2209 	*ppos = pos;
2210 
2211 	if (cached_page)
2212 		page_cache_release(cached_page);
2213 
2214 	/*
2215 	 * For now, when the user asks for O_SYNC, we'll actually give O_DSYNC
2216 	 */
2217 	if (likely(status >= 0)) {
2218 		if (unlikely((file->f_flags & O_SYNC) || IS_SYNC(inode))) {
2219 			if (!a_ops->writepage || !is_sync_kiocb(iocb))
2220 				status = generic_osync_inode(inode, mapping,
2221 						OSYNC_METADATA|OSYNC_DATA);
2222 		}
2223   	}
2224 
2225 	/*
2226 	 * If we get here for O_DIRECT writes then we must have fallen through
2227 	 * to buffered writes (block instantiation inside i_size).  So we sync
2228 	 * the file data here, to try to honour O_DIRECT expectations.
2229 	 */
2230 	if (unlikely(file->f_flags & O_DIRECT) && written)
2231 		status = filemap_write_and_wait(mapping);
2232 
2233 	pagevec_lru_add(&lru_pvec);
2234 	return written ? written : status;
2235 }
2236 EXPORT_SYMBOL(generic_file_buffered_write);
2237 
2238 static ssize_t
2239 __generic_file_aio_write_nolock(struct kiocb *iocb, const struct iovec *iov,
2240 				unsigned long nr_segs, loff_t *ppos)
2241 {
2242 	struct file *file = iocb->ki_filp;
2243 	struct address_space * mapping = file->f_mapping;
2244 	size_t ocount;		/* original count */
2245 	size_t count;		/* after file limit checks */
2246 	struct inode 	*inode = mapping->host;
2247 	loff_t		pos;
2248 	ssize_t		written;
2249 	ssize_t		err;
2250 
2251 	ocount = 0;
2252 	err = generic_segment_checks(iov, &nr_segs, &ocount, VERIFY_READ);
2253 	if (err)
2254 		return err;
2255 
2256 	count = ocount;
2257 	pos = *ppos;
2258 
2259 	vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
2260 
2261 	/* We can write back this queue in page reclaim */
2262 	current->backing_dev_info = mapping->backing_dev_info;
2263 	written = 0;
2264 
2265 	err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
2266 	if (err)
2267 		goto out;
2268 
2269 	if (count == 0)
2270 		goto out;
2271 
2272 	err = remove_suid(file->f_path.dentry);
2273 	if (err)
2274 		goto out;
2275 
2276 	file_update_time(file);
2277 
2278 	/* coalesce the iovecs and go direct-to-BIO for O_DIRECT */
2279 	if (unlikely(file->f_flags & O_DIRECT)) {
2280 		loff_t endbyte;
2281 		ssize_t written_buffered;
2282 
2283 		written = generic_file_direct_write(iocb, iov, &nr_segs, pos,
2284 							ppos, count, ocount);
2285 		if (written < 0 || written == count)
2286 			goto out;
2287 		/*
2288 		 * direct-io write to a hole: fall through to buffered I/O
2289 		 * for completing the rest of the request.
2290 		 */
2291 		pos += written;
2292 		count -= written;
2293 		written_buffered = generic_file_buffered_write(iocb, iov,
2294 						nr_segs, pos, ppos, count,
2295 						written);
2296 		/*
2297 		 * If generic_file_buffered_write() retuned a synchronous error
2298 		 * then we want to return the number of bytes which were
2299 		 * direct-written, or the error code if that was zero.  Note
2300 		 * that this differs from normal direct-io semantics, which
2301 		 * will return -EFOO even if some bytes were written.
2302 		 */
2303 		if (written_buffered < 0) {
2304 			err = written_buffered;
2305 			goto out;
2306 		}
2307 
2308 		/*
2309 		 * We need to ensure that the page cache pages are written to
2310 		 * disk and invalidated to preserve the expected O_DIRECT
2311 		 * semantics.
2312 		 */
2313 		endbyte = pos + written_buffered - written - 1;
2314 		err = do_sync_mapping_range(file->f_mapping, pos, endbyte,
2315 					    SYNC_FILE_RANGE_WAIT_BEFORE|
2316 					    SYNC_FILE_RANGE_WRITE|
2317 					    SYNC_FILE_RANGE_WAIT_AFTER);
2318 		if (err == 0) {
2319 			written = written_buffered;
2320 			invalidate_mapping_pages(mapping,
2321 						 pos >> PAGE_CACHE_SHIFT,
2322 						 endbyte >> PAGE_CACHE_SHIFT);
2323 		} else {
2324 			/*
2325 			 * We don't know how much we wrote, so just return
2326 			 * the number of bytes which were direct-written
2327 			 */
2328 		}
2329 	} else {
2330 		written = generic_file_buffered_write(iocb, iov, nr_segs,
2331 				pos, ppos, count, written);
2332 	}
2333 out:
2334 	current->backing_dev_info = NULL;
2335 	return written ? written : err;
2336 }
2337 
2338 ssize_t generic_file_aio_write_nolock(struct kiocb *iocb,
2339 		const struct iovec *iov, unsigned long nr_segs, loff_t pos)
2340 {
2341 	struct file *file = iocb->ki_filp;
2342 	struct address_space *mapping = file->f_mapping;
2343 	struct inode *inode = mapping->host;
2344 	ssize_t ret;
2345 
2346 	BUG_ON(iocb->ki_pos != pos);
2347 
2348 	ret = __generic_file_aio_write_nolock(iocb, iov, nr_segs,
2349 			&iocb->ki_pos);
2350 
2351 	if (ret > 0 && ((file->f_flags & O_SYNC) || IS_SYNC(inode))) {
2352 		ssize_t err;
2353 
2354 		err = sync_page_range_nolock(inode, mapping, pos, ret);
2355 		if (err < 0)
2356 			ret = err;
2357 	}
2358 	return ret;
2359 }
2360 EXPORT_SYMBOL(generic_file_aio_write_nolock);
2361 
2362 ssize_t generic_file_aio_write(struct kiocb *iocb, const struct iovec *iov,
2363 		unsigned long nr_segs, loff_t pos)
2364 {
2365 	struct file *file = iocb->ki_filp;
2366 	struct address_space *mapping = file->f_mapping;
2367 	struct inode *inode = mapping->host;
2368 	ssize_t ret;
2369 
2370 	BUG_ON(iocb->ki_pos != pos);
2371 
2372 	mutex_lock(&inode->i_mutex);
2373 	ret = __generic_file_aio_write_nolock(iocb, iov, nr_segs,
2374 			&iocb->ki_pos);
2375 	mutex_unlock(&inode->i_mutex);
2376 
2377 	if (ret > 0 && ((file->f_flags & O_SYNC) || IS_SYNC(inode))) {
2378 		ssize_t err;
2379 
2380 		err = sync_page_range(inode, mapping, pos, ret);
2381 		if (err < 0)
2382 			ret = err;
2383 	}
2384 	return ret;
2385 }
2386 EXPORT_SYMBOL(generic_file_aio_write);
2387 
2388 /*
2389  * Called under i_mutex for writes to S_ISREG files.   Returns -EIO if something
2390  * went wrong during pagecache shootdown.
2391  */
2392 static ssize_t
2393 generic_file_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov,
2394 	loff_t offset, unsigned long nr_segs)
2395 {
2396 	struct file *file = iocb->ki_filp;
2397 	struct address_space *mapping = file->f_mapping;
2398 	ssize_t retval;
2399 	size_t write_len;
2400 	pgoff_t end = 0; /* silence gcc */
2401 
2402 	/*
2403 	 * If it's a write, unmap all mmappings of the file up-front.  This
2404 	 * will cause any pte dirty bits to be propagated into the pageframes
2405 	 * for the subsequent filemap_write_and_wait().
2406 	 */
2407 	if (rw == WRITE) {
2408 		write_len = iov_length(iov, nr_segs);
2409 		end = (offset + write_len - 1) >> PAGE_CACHE_SHIFT;
2410 	       	if (mapping_mapped(mapping))
2411 			unmap_mapping_range(mapping, offset, write_len, 0);
2412 	}
2413 
2414 	retval = filemap_write_and_wait(mapping);
2415 	if (retval)
2416 		goto out;
2417 
2418 	/*
2419 	 * After a write we want buffered reads to be sure to go to disk to get
2420 	 * the new data.  We invalidate clean cached page from the region we're
2421 	 * about to write.  We do this *before* the write so that we can return
2422 	 * -EIO without clobbering -EIOCBQUEUED from ->direct_IO().
2423 	 */
2424 	if (rw == WRITE && mapping->nrpages) {
2425 		retval = invalidate_inode_pages2_range(mapping,
2426 					offset >> PAGE_CACHE_SHIFT, end);
2427 		if (retval)
2428 			goto out;
2429 	}
2430 
2431 	retval = mapping->a_ops->direct_IO(rw, iocb, iov, offset, nr_segs);
2432 	if (retval)
2433 		goto out;
2434 
2435 	/*
2436 	 * Finally, try again to invalidate clean pages which might have been
2437 	 * faulted in by get_user_pages() if the source of the write was an
2438 	 * mmap()ed region of the file we're writing.  That's a pretty crazy
2439 	 * thing to do, so we don't support it 100%.  If this invalidation
2440 	 * fails and we have -EIOCBQUEUED we ignore the failure.
2441 	 */
2442 	if (rw == WRITE && mapping->nrpages) {
2443 		int err = invalidate_inode_pages2_range(mapping,
2444 					      offset >> PAGE_CACHE_SHIFT, end);
2445 		if (err && retval >= 0)
2446 			retval = err;
2447 	}
2448 out:
2449 	return retval;
2450 }
2451 
2452 /**
2453  * try_to_release_page() - release old fs-specific metadata on a page
2454  *
2455  * @page: the page which the kernel is trying to free
2456  * @gfp_mask: memory allocation flags (and I/O mode)
2457  *
2458  * The address_space is to try to release any data against the page
2459  * (presumably at page->private).  If the release was successful, return `1'.
2460  * Otherwise return zero.
2461  *
2462  * The @gfp_mask argument specifies whether I/O may be performed to release
2463  * this page (__GFP_IO), and whether the call may block (__GFP_WAIT).
2464  *
2465  * NOTE: @gfp_mask may go away, and this function may become non-blocking.
2466  */
2467 int try_to_release_page(struct page *page, gfp_t gfp_mask)
2468 {
2469 	struct address_space * const mapping = page->mapping;
2470 
2471 	BUG_ON(!PageLocked(page));
2472 	if (PageWriteback(page))
2473 		return 0;
2474 
2475 	if (mapping && mapping->a_ops->releasepage)
2476 		return mapping->a_ops->releasepage(page, gfp_mask);
2477 	return try_to_free_buffers(page);
2478 }
2479 
2480 EXPORT_SYMBOL(try_to_release_page);
2481