1#ifndef _LINUX_PAGEMAP_H
2#define _LINUX_PAGEMAP_H
3
4/*
5 * Copyright 1995 Linus Torvalds
6 */
7#include <linux/mm.h>
8#include <linux/fs.h>
9#include <linux/list.h>
10#include <linux/highmem.h>
11#include <linux/compiler.h>
12#include <asm/uaccess.h>
13#include <linux/gfp.h>
14#include <linux/bitops.h>
15#include <linux/hardirq.h> /* for in_interrupt() */
16#include <linux/hugetlb_inline.h>
17
18/*
19 * Bits in mapping->flags.  The lower __GFP_BITS_SHIFT bits are the page
20 * allocation mode flags.
21 */
22enum mapping_flags {
23	AS_EIO		= __GFP_BITS_SHIFT + 0,	/* IO error on async write */
24	AS_ENOSPC	= __GFP_BITS_SHIFT + 1,	/* ENOSPC on async write */
25	AS_MM_ALL_LOCKS	= __GFP_BITS_SHIFT + 2,	/* under mm_take_all_locks() */
26	AS_UNEVICTABLE	= __GFP_BITS_SHIFT + 3,	/* e.g., ramdisk, SHM_LOCK */
27	AS_EXITING	= __GFP_BITS_SHIFT + 4, /* final truncate in progress */
28};
29
30static inline void mapping_set_error(struct address_space *mapping, int error)
31{
32	if (unlikely(error)) {
33		if (error == -ENOSPC)
34			set_bit(AS_ENOSPC, &mapping->flags);
35		else
36			set_bit(AS_EIO, &mapping->flags);
37	}
38}
39
40static inline void mapping_set_unevictable(struct address_space *mapping)
41{
42	set_bit(AS_UNEVICTABLE, &mapping->flags);
43}
44
45static inline void mapping_clear_unevictable(struct address_space *mapping)
46{
47	clear_bit(AS_UNEVICTABLE, &mapping->flags);
48}
49
50static inline int mapping_unevictable(struct address_space *mapping)
51{
52	if (mapping)
53		return test_bit(AS_UNEVICTABLE, &mapping->flags);
54	return !!mapping;
55}
56
57static inline void mapping_set_exiting(struct address_space *mapping)
58{
59	set_bit(AS_EXITING, &mapping->flags);
60}
61
62static inline int mapping_exiting(struct address_space *mapping)
63{
64	return test_bit(AS_EXITING, &mapping->flags);
65}
66
67static inline gfp_t mapping_gfp_mask(struct address_space * mapping)
68{
69	return (__force gfp_t)mapping->flags & __GFP_BITS_MASK;
70}
71
72/*
73 * This is non-atomic.  Only to be used before the mapping is activated.
74 * Probably needs a barrier...
75 */
76static inline void mapping_set_gfp_mask(struct address_space *m, gfp_t mask)
77{
78	m->flags = (m->flags & ~(__force unsigned long)__GFP_BITS_MASK) |
79				(__force unsigned long)mask;
80}
81
82/*
83 * The page cache can be done in larger chunks than
84 * one page, because it allows for more efficient
85 * throughput (it can then be mapped into user
86 * space in smaller chunks for same flexibility).
87 *
88 * Or rather, it _will_ be done in larger chunks.
89 */
90#define PAGE_CACHE_SHIFT	PAGE_SHIFT
91#define PAGE_CACHE_SIZE		PAGE_SIZE
92#define PAGE_CACHE_MASK		PAGE_MASK
93#define PAGE_CACHE_ALIGN(addr)	(((addr)+PAGE_CACHE_SIZE-1)&PAGE_CACHE_MASK)
94
95#define page_cache_get(page)		get_page(page)
96#define page_cache_release(page)	put_page(page)
97void release_pages(struct page **pages, int nr, bool cold);
98
99/*
100 * speculatively take a reference to a page.
101 * If the page is free (_count == 0), then _count is untouched, and 0
102 * is returned. Otherwise, _count is incremented by 1 and 1 is returned.
103 *
104 * This function must be called inside the same rcu_read_lock() section as has
105 * been used to lookup the page in the pagecache radix-tree (or page table):
106 * this allows allocators to use a synchronize_rcu() to stabilize _count.
107 *
108 * Unless an RCU grace period has passed, the count of all pages coming out
109 * of the allocator must be considered unstable. page_count may return higher
110 * than expected, and put_page must be able to do the right thing when the
111 * page has been finished with, no matter what it is subsequently allocated
112 * for (because put_page is what is used here to drop an invalid speculative
113 * reference).
114 *
115 * This is the interesting part of the lockless pagecache (and lockless
116 * get_user_pages) locking protocol, where the lookup-side (eg. find_get_page)
117 * has the following pattern:
118 * 1. find page in radix tree
119 * 2. conditionally increment refcount
120 * 3. check the page is still in pagecache (if no, goto 1)
121 *
122 * Remove-side that cares about stability of _count (eg. reclaim) has the
123 * following (with tree_lock held for write):
124 * A. atomically check refcount is correct and set it to 0 (atomic_cmpxchg)
125 * B. remove page from pagecache
126 * C. free the page
127 *
128 * There are 2 critical interleavings that matter:
129 * - 2 runs before A: in this case, A sees elevated refcount and bails out
130 * - A runs before 2: in this case, 2 sees zero refcount and retries;
131 *   subsequently, B will complete and 1 will find no page, causing the
132 *   lookup to return NULL.
133 *
134 * It is possible that between 1 and 2, the page is removed then the exact same
135 * page is inserted into the same position in pagecache. That's OK: the
136 * old find_get_page using tree_lock could equally have run before or after
137 * such a re-insertion, depending on order that locks are granted.
138 *
139 * Lookups racing against pagecache insertion isn't a big problem: either 1
140 * will find the page or it will not. Likewise, the old find_get_page could run
141 * either before the insertion or afterwards, depending on timing.
142 */
143static inline int page_cache_get_speculative(struct page *page)
144{
145	VM_BUG_ON(in_interrupt());
146
147#ifdef CONFIG_TINY_RCU
148# ifdef CONFIG_PREEMPT_COUNT
149	VM_BUG_ON(!in_atomic());
150# endif
151	/*
152	 * Preempt must be disabled here - we rely on rcu_read_lock doing
153	 * this for us.
154	 *
155	 * Pagecache won't be truncated from interrupt context, so if we have
156	 * found a page in the radix tree here, we have pinned its refcount by
157	 * disabling preempt, and hence no need for the "speculative get" that
158	 * SMP requires.
159	 */
160	VM_BUG_ON_PAGE(page_count(page) == 0, page);
161	atomic_inc(&page->_count);
162
163#else
164	if (unlikely(!get_page_unless_zero(page))) {
165		/*
166		 * Either the page has been freed, or will be freed.
167		 * In either case, retry here and the caller should
168		 * do the right thing (see comments above).
169		 */
170		return 0;
171	}
172#endif
173	VM_BUG_ON_PAGE(PageTail(page), page);
174
175	return 1;
176}
177
178/*
179 * Same as above, but add instead of inc (could just be merged)
180 */
181static inline int page_cache_add_speculative(struct page *page, int count)
182{
183	VM_BUG_ON(in_interrupt());
184
185#if !defined(CONFIG_SMP) && defined(CONFIG_TREE_RCU)
186# ifdef CONFIG_PREEMPT_COUNT
187	VM_BUG_ON(!in_atomic());
188# endif
189	VM_BUG_ON_PAGE(page_count(page) == 0, page);
190	atomic_add(count, &page->_count);
191
192#else
193	if (unlikely(!atomic_add_unless(&page->_count, count, 0)))
194		return 0;
195#endif
196	VM_BUG_ON_PAGE(PageCompound(page) && page != compound_head(page), page);
197
198	return 1;
199}
200
201static inline int page_freeze_refs(struct page *page, int count)
202{
203	return likely(atomic_cmpxchg(&page->_count, count, 0) == count);
204}
205
206static inline void page_unfreeze_refs(struct page *page, int count)
207{
208	VM_BUG_ON_PAGE(page_count(page) != 0, page);
209	VM_BUG_ON(count == 0);
210
211	atomic_set(&page->_count, count);
212}
213
214#ifdef CONFIG_NUMA
215extern struct page *__page_cache_alloc(gfp_t gfp);
216#else
217static inline struct page *__page_cache_alloc(gfp_t gfp)
218{
219	return alloc_pages(gfp, 0);
220}
221#endif
222
223static inline struct page *page_cache_alloc(struct address_space *x)
224{
225	return __page_cache_alloc(mapping_gfp_mask(x));
226}
227
228static inline struct page *page_cache_alloc_cold(struct address_space *x)
229{
230	return __page_cache_alloc(mapping_gfp_mask(x)|__GFP_COLD);
231}
232
233static inline struct page *page_cache_alloc_readahead(struct address_space *x)
234{
235	return __page_cache_alloc(mapping_gfp_mask(x) |
236				  __GFP_COLD | __GFP_NORETRY | __GFP_NOWARN);
237}
238
239typedef int filler_t(void *, struct page *);
240
241pgoff_t page_cache_next_hole(struct address_space *mapping,
242			     pgoff_t index, unsigned long max_scan);
243pgoff_t page_cache_prev_hole(struct address_space *mapping,
244			     pgoff_t index, unsigned long max_scan);
245
246#define FGP_ACCESSED		0x00000001
247#define FGP_LOCK		0x00000002
248#define FGP_CREAT		0x00000004
249#define FGP_WRITE		0x00000008
250#define FGP_NOFS		0x00000010
251#define FGP_NOWAIT		0x00000020
252
253struct page *pagecache_get_page(struct address_space *mapping, pgoff_t offset,
254		int fgp_flags, gfp_t cache_gfp_mask);
255
256/**
257 * find_get_page - find and get a page reference
258 * @mapping: the address_space to search
259 * @offset: the page index
260 *
261 * Looks up the page cache slot at @mapping & @offset.  If there is a
262 * page cache page, it is returned with an increased refcount.
263 *
264 * Otherwise, %NULL is returned.
265 */
266static inline struct page *find_get_page(struct address_space *mapping,
267					pgoff_t offset)
268{
269	return pagecache_get_page(mapping, offset, 0, 0);
270}
271
272static inline struct page *find_get_page_flags(struct address_space *mapping,
273					pgoff_t offset, int fgp_flags)
274{
275	return pagecache_get_page(mapping, offset, fgp_flags, 0);
276}
277
278/**
279 * find_lock_page - locate, pin and lock a pagecache page
280 * pagecache_get_page - find and get a page reference
281 * @mapping: the address_space to search
282 * @offset: the page index
283 *
284 * Looks up the page cache slot at @mapping & @offset.  If there is a
285 * page cache page, it is returned locked and with an increased
286 * refcount.
287 *
288 * Otherwise, %NULL is returned.
289 *
290 * find_lock_page() may sleep.
291 */
292static inline struct page *find_lock_page(struct address_space *mapping,
293					pgoff_t offset)
294{
295	return pagecache_get_page(mapping, offset, FGP_LOCK, 0);
296}
297
298/**
299 * find_or_create_page - locate or add a pagecache page
300 * @mapping: the page's address_space
301 * @index: the page's index into the mapping
302 * @gfp_mask: page allocation mode
303 *
304 * Looks up the page cache slot at @mapping & @offset.  If there is a
305 * page cache page, it is returned locked and with an increased
306 * refcount.
307 *
308 * If the page is not present, a new page is allocated using @gfp_mask
309 * and added to the page cache and the VM's LRU list.  The page is
310 * returned locked and with an increased refcount.
311 *
312 * On memory exhaustion, %NULL is returned.
313 *
314 * find_or_create_page() may sleep, even if @gfp_flags specifies an
315 * atomic allocation!
316 */
317static inline struct page *find_or_create_page(struct address_space *mapping,
318					pgoff_t offset, gfp_t gfp_mask)
319{
320	return pagecache_get_page(mapping, offset,
321					FGP_LOCK|FGP_ACCESSED|FGP_CREAT,
322					gfp_mask);
323}
324
325/**
326 * grab_cache_page_nowait - returns locked page at given index in given cache
327 * @mapping: target address_space
328 * @index: the page index
329 *
330 * Same as grab_cache_page(), but do not wait if the page is unavailable.
331 * This is intended for speculative data generators, where the data can
332 * be regenerated if the page couldn't be grabbed.  This routine should
333 * be safe to call while holding the lock for another page.
334 *
335 * Clear __GFP_FS when allocating the page to avoid recursion into the fs
336 * and deadlock against the caller's locked page.
337 */
338static inline struct page *grab_cache_page_nowait(struct address_space *mapping,
339				pgoff_t index)
340{
341	return pagecache_get_page(mapping, index,
342			FGP_LOCK|FGP_CREAT|FGP_NOFS|FGP_NOWAIT,
343			mapping_gfp_mask(mapping));
344}
345
346struct page *find_get_entry(struct address_space *mapping, pgoff_t offset);
347struct page *find_lock_entry(struct address_space *mapping, pgoff_t offset);
348unsigned find_get_entries(struct address_space *mapping, pgoff_t start,
349			  unsigned int nr_entries, struct page **entries,
350			  pgoff_t *indices);
351unsigned find_get_pages(struct address_space *mapping, pgoff_t start,
352			unsigned int nr_pages, struct page **pages);
353unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t start,
354			       unsigned int nr_pages, struct page **pages);
355unsigned find_get_pages_tag(struct address_space *mapping, pgoff_t *index,
356			int tag, unsigned int nr_pages, struct page **pages);
357
358struct page *grab_cache_page_write_begin(struct address_space *mapping,
359			pgoff_t index, unsigned flags);
360
361/*
362 * Returns locked page at given index in given cache, creating it if needed.
363 */
364static inline struct page *grab_cache_page(struct address_space *mapping,
365								pgoff_t index)
366{
367	return find_or_create_page(mapping, index, mapping_gfp_mask(mapping));
368}
369
370extern struct page * read_cache_page(struct address_space *mapping,
371				pgoff_t index, filler_t *filler, void *data);
372extern struct page * read_cache_page_gfp(struct address_space *mapping,
373				pgoff_t index, gfp_t gfp_mask);
374extern int read_cache_pages(struct address_space *mapping,
375		struct list_head *pages, filler_t *filler, void *data);
376
377static inline struct page *read_mapping_page(struct address_space *mapping,
378				pgoff_t index, void *data)
379{
380	filler_t *filler = (filler_t *)mapping->a_ops->readpage;
381	return read_cache_page(mapping, index, filler, data);
382}
383
384/*
385 * Get the offset in PAGE_SIZE.
386 * (TODO: hugepage should have ->index in PAGE_SIZE)
387 */
388static inline pgoff_t page_to_pgoff(struct page *page)
389{
390	if (unlikely(PageHeadHuge(page)))
391		return page->index << compound_order(page);
392	else
393		return page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT);
394}
395
396/*
397 * Return byte-offset into filesystem object for page.
398 */
399static inline loff_t page_offset(struct page *page)
400{
401	return ((loff_t)page->index) << PAGE_CACHE_SHIFT;
402}
403
404static inline loff_t page_file_offset(struct page *page)
405{
406	return ((loff_t)page_file_index(page)) << PAGE_CACHE_SHIFT;
407}
408
409extern pgoff_t linear_hugepage_index(struct vm_area_struct *vma,
410				     unsigned long address);
411
412static inline pgoff_t linear_page_index(struct vm_area_struct *vma,
413					unsigned long address)
414{
415	pgoff_t pgoff;
416	if (unlikely(is_vm_hugetlb_page(vma)))
417		return linear_hugepage_index(vma, address);
418	pgoff = (address - vma->vm_start) >> PAGE_SHIFT;
419	pgoff += vma->vm_pgoff;
420	return pgoff >> (PAGE_CACHE_SHIFT - PAGE_SHIFT);
421}
422
423extern void __lock_page(struct page *page);
424extern int __lock_page_killable(struct page *page);
425extern int __lock_page_or_retry(struct page *page, struct mm_struct *mm,
426				unsigned int flags);
427extern void unlock_page(struct page *page);
428
429static inline void __set_page_locked(struct page *page)
430{
431	__set_bit(PG_locked, &page->flags);
432}
433
434static inline void __clear_page_locked(struct page *page)
435{
436	__clear_bit(PG_locked, &page->flags);
437}
438
439static inline int trylock_page(struct page *page)
440{
441	return (likely(!test_and_set_bit_lock(PG_locked, &page->flags)));
442}
443
444/*
445 * lock_page may only be called if we have the page's inode pinned.
446 */
447static inline void lock_page(struct page *page)
448{
449	might_sleep();
450	if (!trylock_page(page))
451		__lock_page(page);
452}
453
454/*
455 * lock_page_killable is like lock_page but can be interrupted by fatal
456 * signals.  It returns 0 if it locked the page and -EINTR if it was
457 * killed while waiting.
458 */
459static inline int lock_page_killable(struct page *page)
460{
461	might_sleep();
462	if (!trylock_page(page))
463		return __lock_page_killable(page);
464	return 0;
465}
466
467/*
468 * lock_page_or_retry - Lock the page, unless this would block and the
469 * caller indicated that it can handle a retry.
470 *
471 * Return value and mmap_sem implications depend on flags; see
472 * __lock_page_or_retry().
473 */
474static inline int lock_page_or_retry(struct page *page, struct mm_struct *mm,
475				     unsigned int flags)
476{
477	might_sleep();
478	return trylock_page(page) || __lock_page_or_retry(page, mm, flags);
479}
480
481/*
482 * This is exported only for wait_on_page_locked/wait_on_page_writeback,
483 * and for filesystems which need to wait on PG_private.
484 */
485extern void wait_on_page_bit(struct page *page, int bit_nr);
486
487extern int wait_on_page_bit_killable(struct page *page, int bit_nr);
488extern int wait_on_page_bit_killable_timeout(struct page *page,
489					     int bit_nr, unsigned long timeout);
490
491static inline int wait_on_page_locked_killable(struct page *page)
492{
493	if (PageLocked(page))
494		return wait_on_page_bit_killable(page, PG_locked);
495	return 0;
496}
497
498extern wait_queue_head_t *page_waitqueue(struct page *page);
499static inline void wake_up_page(struct page *page, int bit)
500{
501	__wake_up_bit(page_waitqueue(page), &page->flags, bit);
502}
503
504/*
505 * Wait for a page to be unlocked.
506 *
507 * This must be called with the caller "holding" the page,
508 * ie with increased "page->count" so that the page won't
509 * go away during the wait..
510 */
511static inline void wait_on_page_locked(struct page *page)
512{
513	if (PageLocked(page))
514		wait_on_page_bit(page, PG_locked);
515}
516
517/*
518 * Wait for a page to complete writeback
519 */
520static inline void wait_on_page_writeback(struct page *page)
521{
522	if (PageWriteback(page))
523		wait_on_page_bit(page, PG_writeback);
524}
525
526extern void end_page_writeback(struct page *page);
527void wait_for_stable_page(struct page *page);
528
529void page_endio(struct page *page, int rw, int err);
530
531/*
532 * Add an arbitrary waiter to a page's wait queue
533 */
534extern void add_page_wait_queue(struct page *page, wait_queue_t *waiter);
535
536/*
537 * Fault a userspace page into pagetables.  Return non-zero on a fault.
538 *
539 * This assumes that two userspace pages are always sufficient.  That's
540 * not true if PAGE_CACHE_SIZE > PAGE_SIZE.
541 */
542static inline int fault_in_pages_writeable(char __user *uaddr, int size)
543{
544	int ret;
545
546	if (unlikely(size == 0))
547		return 0;
548
549	/*
550	 * Writing zeroes into userspace here is OK, because we know that if
551	 * the zero gets there, we'll be overwriting it.
552	 */
553	ret = __put_user(0, uaddr);
554	if (ret == 0) {
555		char __user *end = uaddr + size - 1;
556
557		/*
558		 * If the page was already mapped, this will get a cache miss
559		 * for sure, so try to avoid doing it.
560		 */
561		if (((unsigned long)uaddr & PAGE_MASK) !=
562				((unsigned long)end & PAGE_MASK))
563			ret = __put_user(0, end);
564	}
565	return ret;
566}
567
568static inline int fault_in_pages_readable(const char __user *uaddr, int size)
569{
570	volatile char c;
571	int ret;
572
573	if (unlikely(size == 0))
574		return 0;
575
576	ret = __get_user(c, uaddr);
577	if (ret == 0) {
578		const char __user *end = uaddr + size - 1;
579
580		if (((unsigned long)uaddr & PAGE_MASK) !=
581				((unsigned long)end & PAGE_MASK)) {
582			ret = __get_user(c, end);
583			(void)c;
584		}
585	}
586	return ret;
587}
588
589/*
590 * Multipage variants of the above prefault helpers, useful if more than
591 * PAGE_SIZE of data needs to be prefaulted. These are separate from the above
592 * functions (which only handle up to PAGE_SIZE) to avoid clobbering the
593 * filemap.c hotpaths.
594 */
595static inline int fault_in_multipages_writeable(char __user *uaddr, int size)
596{
597	int ret = 0;
598	char __user *end = uaddr + size - 1;
599
600	if (unlikely(size == 0))
601		return ret;
602
603	/*
604	 * Writing zeroes into userspace here is OK, because we know that if
605	 * the zero gets there, we'll be overwriting it.
606	 */
607	while (uaddr <= end) {
608		ret = __put_user(0, uaddr);
609		if (ret != 0)
610			return ret;
611		uaddr += PAGE_SIZE;
612	}
613
614	/* Check whether the range spilled into the next page. */
615	if (((unsigned long)uaddr & PAGE_MASK) ==
616			((unsigned long)end & PAGE_MASK))
617		ret = __put_user(0, end);
618
619	return ret;
620}
621
622static inline int fault_in_multipages_readable(const char __user *uaddr,
623					       int size)
624{
625	volatile char c;
626	int ret = 0;
627	const char __user *end = uaddr + size - 1;
628
629	if (unlikely(size == 0))
630		return ret;
631
632	while (uaddr <= end) {
633		ret = __get_user(c, uaddr);
634		if (ret != 0)
635			return ret;
636		uaddr += PAGE_SIZE;
637	}
638
639	/* Check whether the range spilled into the next page. */
640	if (((unsigned long)uaddr & PAGE_MASK) ==
641			((unsigned long)end & PAGE_MASK)) {
642		ret = __get_user(c, end);
643		(void)c;
644	}
645
646	return ret;
647}
648
649int add_to_page_cache_locked(struct page *page, struct address_space *mapping,
650				pgoff_t index, gfp_t gfp_mask);
651int add_to_page_cache_lru(struct page *page, struct address_space *mapping,
652				pgoff_t index, gfp_t gfp_mask);
653extern void delete_from_page_cache(struct page *page);
654extern void __delete_from_page_cache(struct page *page, void *shadow);
655int replace_page_cache_page(struct page *old, struct page *new, gfp_t gfp_mask);
656
657/*
658 * Like add_to_page_cache_locked, but used to add newly allocated pages:
659 * the page is new, so we can just run __set_page_locked() against it.
660 */
661static inline int add_to_page_cache(struct page *page,
662		struct address_space *mapping, pgoff_t offset, gfp_t gfp_mask)
663{
664	int error;
665
666	__set_page_locked(page);
667	error = add_to_page_cache_locked(page, mapping, offset, gfp_mask);
668	if (unlikely(error))
669		__clear_page_locked(page);
670	return error;
671}
672
673#endif /* _LINUX_PAGEMAP_H */
674