2 * mm/rmap.c - physical to virtual reverse mappings
4 * Copyright 2001, Rik van Riel <riel@conectiva.com.br>
5 * Released under the General Public License (GPL).
7 * Simple, low overhead reverse mapping scheme.
8 * Please try to keep this thing as modular as possible.
10 * Provides methods for unmapping each kind of mapped page:
11 * the anon methods track anonymous pages, and
12 * the file methods track pages belonging to an inode.
14 * Original design by Rik van Riel <riel@conectiva.com.br> 2001
15 * File methods by Dave McCracken <dmccr@us.ibm.com> 2003, 2004
16 * Anonymous methods by Andrea Arcangeli <andrea@suse.de> 2004
17 * Contributions by Hugh Dickins 2003, 2004
21 * Lock ordering in mm:
23 * inode->i_mutex (while writing or truncating, not reading or faulting)
24 * inode->i_alloc_sem (vmtruncate_range)
26 * page->flags PG_locked (lock_page)
27 * mapping->i_mmap_lock
29 * mm->page_table_lock or pte_lock
30 * zone->lru_lock (in mark_page_accessed, isolate_lru_page)
31 * swap_lock (in swap_duplicate, swap_info_get)
32 * mmlist_lock (in mmput, drain_mmlist and others)
33 * mapping->private_lock (in __set_page_dirty_buffers)
34 * inode_lock (in set_page_dirty's __mark_inode_dirty)
35 * sb_lock (within inode_lock in fs/fs-writeback.c)
36 * mapping->tree_lock (widely used, in set_page_dirty,
37 * in arch-dependent flush_dcache_mmap_lock,
38 * within inode_lock in __sync_single_inode)
40 * (code doesn't rely on that order so it could be switched around)
42 * anon_vma->lock (memory_failure, collect_procs_anon)
47 #include <linux/pagemap.h>
48 #include <linux/swap.h>
49 #include <linux/swapops.h>
50 #include <linux/slab.h>
51 #include <linux/init.h>
52 #include <linux/ksm.h>
53 #include <linux/rmap.h>
54 #include <linux/rcupdate.h>
55 #include <linux/module.h>
56 #include <linux/memcontrol.h>
57 #include <linux/mmu_notifier.h>
58 #include <linux/migrate.h>
59 #include <linux/hugetlb.h>
61 #include <asm/tlbflush.h>
65 static struct kmem_cache
*anon_vma_cachep
;
66 static struct kmem_cache
*anon_vma_chain_cachep
;
68 static inline struct anon_vma
*anon_vma_alloc(void)
70 return kmem_cache_alloc(anon_vma_cachep
, GFP_KERNEL
);
73 void anon_vma_free(struct anon_vma
*anon_vma
)
75 kmem_cache_free(anon_vma_cachep
, anon_vma
);
78 static inline struct anon_vma_chain
*anon_vma_chain_alloc(void)
80 return kmem_cache_alloc(anon_vma_chain_cachep
, GFP_KERNEL
);
83 void anon_vma_chain_free(struct anon_vma_chain
*anon_vma_chain
)
85 kmem_cache_free(anon_vma_chain_cachep
, anon_vma_chain
);
89 * anon_vma_prepare - attach an anon_vma to a memory region
90 * @vma: the memory region in question
92 * This makes sure the memory mapping described by 'vma' has
93 * an 'anon_vma' attached to it, so that we can associate the
94 * anonymous pages mapped into it with that anon_vma.
96 * The common case will be that we already have one, but if
97 * if not we either need to find an adjacent mapping that we
98 * can re-use the anon_vma from (very common when the only
99 * reason for splitting a vma has been mprotect()), or we
100 * allocate a new one.
102 * Anon-vma allocations are very subtle, because we may have
103 * optimistically looked up an anon_vma in page_lock_anon_vma()
104 * and that may actually touch the spinlock even in the newly
105 * allocated vma (it depends on RCU to make sure that the
106 * anon_vma isn't actually destroyed).
108 * As a result, we need to do proper anon_vma locking even
109 * for the new allocation. At the same time, we do not want
110 * to do any locking for the common case of already having
113 * This must be called with the mmap_sem held for reading.
115 int anon_vma_prepare(struct vm_area_struct
*vma
)
117 struct anon_vma
*anon_vma
= vma
->anon_vma
;
118 struct anon_vma_chain
*avc
;
121 if (unlikely(!anon_vma
)) {
122 struct mm_struct
*mm
= vma
->vm_mm
;
123 struct anon_vma
*allocated
;
125 avc
= anon_vma_chain_alloc();
129 anon_vma
= find_mergeable_anon_vma(vma
);
132 anon_vma
= anon_vma_alloc();
133 if (unlikely(!anon_vma
))
134 goto out_enomem_free_avc
;
135 allocated
= anon_vma
;
137 * This VMA had no anon_vma yet. This anon_vma is
138 * the root of any anon_vma tree that might form.
140 anon_vma
->root
= anon_vma
;
143 anon_vma_lock(anon_vma
);
144 /* page_table_lock to protect against threads */
145 spin_lock(&mm
->page_table_lock
);
146 if (likely(!vma
->anon_vma
)) {
147 vma
->anon_vma
= anon_vma
;
148 avc
->anon_vma
= anon_vma
;
150 list_add(&avc
->same_vma
, &vma
->anon_vma_chain
);
151 list_add_tail(&avc
->same_anon_vma
, &anon_vma
->head
);
155 spin_unlock(&mm
->page_table_lock
);
156 anon_vma_unlock(anon_vma
);
158 if (unlikely(allocated
))
159 anon_vma_free(allocated
);
161 anon_vma_chain_free(avc
);
166 anon_vma_chain_free(avc
);
171 static void anon_vma_chain_link(struct vm_area_struct
*vma
,
172 struct anon_vma_chain
*avc
,
173 struct anon_vma
*anon_vma
)
176 avc
->anon_vma
= anon_vma
;
177 list_add(&avc
->same_vma
, &vma
->anon_vma_chain
);
179 anon_vma_lock(anon_vma
);
180 list_add_tail(&avc
->same_anon_vma
, &anon_vma
->head
);
181 anon_vma_unlock(anon_vma
);
185 * Attach the anon_vmas from src to dst.
186 * Returns 0 on success, -ENOMEM on failure.
188 int anon_vma_clone(struct vm_area_struct
*dst
, struct vm_area_struct
*src
)
190 struct anon_vma_chain
*avc
, *pavc
;
192 list_for_each_entry_reverse(pavc
, &src
->anon_vma_chain
, same_vma
) {
193 avc
= anon_vma_chain_alloc();
196 anon_vma_chain_link(dst
, avc
, pavc
->anon_vma
);
201 unlink_anon_vmas(dst
);
206 * Attach vma to its own anon_vma, as well as to the anon_vmas that
207 * the corresponding VMA in the parent process is attached to.
208 * Returns 0 on success, non-zero on failure.
210 int anon_vma_fork(struct vm_area_struct
*vma
, struct vm_area_struct
*pvma
)
212 struct anon_vma_chain
*avc
;
213 struct anon_vma
*anon_vma
;
215 /* Don't bother if the parent process has no anon_vma here. */
220 * First, attach the new VMA to the parent VMA's anon_vmas,
221 * so rmap can find non-COWed pages in child processes.
223 if (anon_vma_clone(vma
, pvma
))
226 /* Then add our own anon_vma. */
227 anon_vma
= anon_vma_alloc();
230 avc
= anon_vma_chain_alloc();
232 goto out_error_free_anon_vma
;
235 * The root anon_vma's spinlock is the lock actually used when we
236 * lock any of the anon_vmas in this anon_vma tree.
238 anon_vma
->root
= pvma
->anon_vma
->root
;
240 * With KSM refcounts, an anon_vma can stay around longer than the
241 * process it belongs to. The root anon_vma needs to be pinned
242 * until this anon_vma is freed, because the lock lives in the root.
244 get_anon_vma(anon_vma
->root
);
245 /* Mark this anon_vma as the one where our new (COWed) pages go. */
246 vma
->anon_vma
= anon_vma
;
247 anon_vma_chain_link(vma
, avc
, anon_vma
);
251 out_error_free_anon_vma
:
252 anon_vma_free(anon_vma
);
254 unlink_anon_vmas(vma
);
258 static void anon_vma_unlink(struct anon_vma_chain
*anon_vma_chain
)
260 struct anon_vma
*anon_vma
= anon_vma_chain
->anon_vma
;
263 /* If anon_vma_fork fails, we can get an empty anon_vma_chain. */
267 anon_vma_lock(anon_vma
);
268 list_del(&anon_vma_chain
->same_anon_vma
);
270 /* We must garbage collect the anon_vma if it's empty */
271 empty
= list_empty(&anon_vma
->head
) && !anonvma_external_refcount(anon_vma
);
272 anon_vma_unlock(anon_vma
);
275 /* We no longer need the root anon_vma */
276 if (anon_vma
->root
!= anon_vma
)
277 drop_anon_vma(anon_vma
->root
);
278 anon_vma_free(anon_vma
);
282 void unlink_anon_vmas(struct vm_area_struct
*vma
)
284 struct anon_vma_chain
*avc
, *next
;
287 * Unlink each anon_vma chained to the VMA. This list is ordered
288 * from newest to oldest, ensuring the root anon_vma gets freed last.
290 list_for_each_entry_safe(avc
, next
, &vma
->anon_vma_chain
, same_vma
) {
291 anon_vma_unlink(avc
);
292 list_del(&avc
->same_vma
);
293 anon_vma_chain_free(avc
);
297 static void anon_vma_ctor(void *data
)
299 struct anon_vma
*anon_vma
= data
;
301 spin_lock_init(&anon_vma
->lock
);
302 anonvma_external_refcount_init(anon_vma
);
303 INIT_LIST_HEAD(&anon_vma
->head
);
306 void __init
anon_vma_init(void)
308 anon_vma_cachep
= kmem_cache_create("anon_vma", sizeof(struct anon_vma
),
309 0, SLAB_DESTROY_BY_RCU
|SLAB_PANIC
, anon_vma_ctor
);
310 anon_vma_chain_cachep
= KMEM_CACHE(anon_vma_chain
, SLAB_PANIC
);
314 * Getting a lock on a stable anon_vma from a page off the LRU is
315 * tricky: page_lock_anon_vma rely on RCU to guard against the races.
317 struct anon_vma
*page_lock_anon_vma(struct page
*page
)
319 struct anon_vma
*anon_vma
;
320 unsigned long anon_mapping
;
323 anon_mapping
= (unsigned long) ACCESS_ONCE(page
->mapping
);
324 if ((anon_mapping
& PAGE_MAPPING_FLAGS
) != PAGE_MAPPING_ANON
)
326 if (!page_mapped(page
))
329 anon_vma
= (struct anon_vma
*) (anon_mapping
- PAGE_MAPPING_ANON
);
330 anon_vma_lock(anon_vma
);
337 void page_unlock_anon_vma(struct anon_vma
*anon_vma
)
339 anon_vma_unlock(anon_vma
);
344 * At what user virtual address is page expected in @vma?
345 * Returns virtual address or -EFAULT if page's index/offset is not
346 * within the range mapped the @vma.
348 static inline unsigned long
349 vma_address(struct page
*page
, struct vm_area_struct
*vma
)
351 pgoff_t pgoff
= page
->index
<< (PAGE_CACHE_SHIFT
- PAGE_SHIFT
);
352 unsigned long address
;
354 if (unlikely(is_vm_hugetlb_page(vma
)))
355 pgoff
= page
->index
<< huge_page_order(page_hstate(page
));
356 address
= vma
->vm_start
+ ((pgoff
- vma
->vm_pgoff
) << PAGE_SHIFT
);
357 if (unlikely(address
< vma
->vm_start
|| address
>= vma
->vm_end
)) {
358 /* page should be within @vma mapping range */
365 * At what user virtual address is page expected in vma?
366 * Caller should check the page is actually part of the vma.
368 unsigned long page_address_in_vma(struct page
*page
, struct vm_area_struct
*vma
)
370 if (PageAnon(page
)) {
371 if (vma
->anon_vma
->root
!= page_anon_vma(page
)->root
)
373 } else if (page
->mapping
&& !(vma
->vm_flags
& VM_NONLINEAR
)) {
375 vma
->vm_file
->f_mapping
!= page
->mapping
)
379 return vma_address(page
, vma
);
383 * Check that @page is mapped at @address into @mm.
385 * If @sync is false, page_check_address may perform a racy check to avoid
386 * the page table lock when the pte is not present (helpful when reclaiming
387 * highly shared pages).
389 * On success returns with pte mapped and locked.
391 pte_t
*page_check_address(struct page
*page
, struct mm_struct
*mm
,
392 unsigned long address
, spinlock_t
**ptlp
, int sync
)
400 if (unlikely(PageHuge(page
))) {
401 pte
= huge_pte_offset(mm
, address
);
402 ptl
= &mm
->page_table_lock
;
406 pgd
= pgd_offset(mm
, address
);
407 if (!pgd_present(*pgd
))
410 pud
= pud_offset(pgd
, address
);
411 if (!pud_present(*pud
))
414 pmd
= pmd_offset(pud
, address
);
415 if (!pmd_present(*pmd
))
418 pte
= pte_offset_map(pmd
, address
);
419 /* Make a quick check before getting the lock */
420 if (!sync
&& !pte_present(*pte
)) {
425 ptl
= pte_lockptr(mm
, pmd
);
428 if (pte_present(*pte
) && page_to_pfn(page
) == pte_pfn(*pte
)) {
432 pte_unmap_unlock(pte
, ptl
);
437 * page_mapped_in_vma - check whether a page is really mapped in a VMA
438 * @page: the page to test
439 * @vma: the VMA to test
441 * Returns 1 if the page is mapped into the page tables of the VMA, 0
442 * if the page is not mapped into the page tables of this VMA. Only
443 * valid for normal file or anonymous VMAs.
445 int page_mapped_in_vma(struct page
*page
, struct vm_area_struct
*vma
)
447 unsigned long address
;
451 address
= vma_address(page
, vma
);
452 if (address
== -EFAULT
) /* out of vma range */
454 pte
= page_check_address(page
, vma
->vm_mm
, address
, &ptl
, 1);
455 if (!pte
) /* the page is not in this mm */
457 pte_unmap_unlock(pte
, ptl
);
463 * Subfunctions of page_referenced: page_referenced_one called
464 * repeatedly from either page_referenced_anon or page_referenced_file.
466 int page_referenced_one(struct page
*page
, struct vm_area_struct
*vma
,
467 unsigned long address
, unsigned int *mapcount
,
468 unsigned long *vm_flags
)
470 struct mm_struct
*mm
= vma
->vm_mm
;
475 pte
= page_check_address(page
, mm
, address
, &ptl
, 0);
480 * Don't want to elevate referenced for mlocked page that gets this far,
481 * in order that it progresses to try_to_unmap and is moved to the
484 if (vma
->vm_flags
& VM_LOCKED
) {
485 *mapcount
= 1; /* break early from loop */
486 *vm_flags
|= VM_LOCKED
;
490 if (ptep_clear_flush_young_notify(vma
, address
, pte
)) {
492 * Don't treat a reference through a sequentially read
493 * mapping as such. If the page has been used in
494 * another mapping, we will catch it; if this other
495 * mapping is already gone, the unmap path will have
496 * set PG_referenced or activated the page.
498 if (likely(!VM_SequentialReadHint(vma
)))
502 /* Pretend the page is referenced if the task has the
503 swap token and is in the middle of a page fault. */
504 if (mm
!= current
->mm
&& has_swap_token(mm
) &&
505 rwsem_is_locked(&mm
->mmap_sem
))
510 pte_unmap_unlock(pte
, ptl
);
513 *vm_flags
|= vma
->vm_flags
;
518 static int page_referenced_anon(struct page
*page
,
519 struct mem_cgroup
*mem_cont
,
520 unsigned long *vm_flags
)
522 unsigned int mapcount
;
523 struct anon_vma
*anon_vma
;
524 struct anon_vma_chain
*avc
;
527 anon_vma
= page_lock_anon_vma(page
);
531 mapcount
= page_mapcount(page
);
532 list_for_each_entry(avc
, &anon_vma
->head
, same_anon_vma
) {
533 struct vm_area_struct
*vma
= avc
->vma
;
534 unsigned long address
= vma_address(page
, vma
);
535 if (address
== -EFAULT
)
538 * If we are reclaiming on behalf of a cgroup, skip
539 * counting on behalf of references from different
542 if (mem_cont
&& !mm_match_cgroup(vma
->vm_mm
, mem_cont
))
544 referenced
+= page_referenced_one(page
, vma
, address
,
545 &mapcount
, vm_flags
);
550 page_unlock_anon_vma(anon_vma
);
555 * page_referenced_file - referenced check for object-based rmap
556 * @page: the page we're checking references on.
557 * @mem_cont: target memory controller
558 * @vm_flags: collect encountered vma->vm_flags who actually referenced the page
560 * For an object-based mapped page, find all the places it is mapped and
561 * check/clear the referenced flag. This is done by following the page->mapping
562 * pointer, then walking the chain of vmas it holds. It returns the number
563 * of references it found.
565 * This function is only called from page_referenced for object-based pages.
567 static int page_referenced_file(struct page
*page
,
568 struct mem_cgroup
*mem_cont
,
569 unsigned long *vm_flags
)
571 unsigned int mapcount
;
572 struct address_space
*mapping
= page
->mapping
;
573 pgoff_t pgoff
= page
->index
<< (PAGE_CACHE_SHIFT
- PAGE_SHIFT
);
574 struct vm_area_struct
*vma
;
575 struct prio_tree_iter iter
;
579 * The caller's checks on page->mapping and !PageAnon have made
580 * sure that this is a file page: the check for page->mapping
581 * excludes the case just before it gets set on an anon page.
583 BUG_ON(PageAnon(page
));
586 * The page lock not only makes sure that page->mapping cannot
587 * suddenly be NULLified by truncation, it makes sure that the
588 * structure at mapping cannot be freed and reused yet,
589 * so we can safely take mapping->i_mmap_lock.
591 BUG_ON(!PageLocked(page
));
593 spin_lock(&mapping
->i_mmap_lock
);
596 * i_mmap_lock does not stabilize mapcount at all, but mapcount
597 * is more likely to be accurate if we note it after spinning.
599 mapcount
= page_mapcount(page
);
601 vma_prio_tree_foreach(vma
, &iter
, &mapping
->i_mmap
, pgoff
, pgoff
) {
602 unsigned long address
= vma_address(page
, vma
);
603 if (address
== -EFAULT
)
606 * If we are reclaiming on behalf of a cgroup, skip
607 * counting on behalf of references from different
610 if (mem_cont
&& !mm_match_cgroup(vma
->vm_mm
, mem_cont
))
612 referenced
+= page_referenced_one(page
, vma
, address
,
613 &mapcount
, vm_flags
);
618 spin_unlock(&mapping
->i_mmap_lock
);
623 * page_referenced - test if the page was referenced
624 * @page: the page to test
625 * @is_locked: caller holds lock on the page
626 * @mem_cont: target memory controller
627 * @vm_flags: collect encountered vma->vm_flags who actually referenced the page
629 * Quick test_and_clear_referenced for all mappings to a page,
630 * returns the number of ptes which referenced the page.
632 int page_referenced(struct page
*page
,
634 struct mem_cgroup
*mem_cont
,
635 unsigned long *vm_flags
)
641 if (page_mapped(page
) && page_rmapping(page
)) {
642 if (!is_locked
&& (!PageAnon(page
) || PageKsm(page
))) {
643 we_locked
= trylock_page(page
);
649 if (unlikely(PageKsm(page
)))
650 referenced
+= page_referenced_ksm(page
, mem_cont
,
652 else if (PageAnon(page
))
653 referenced
+= page_referenced_anon(page
, mem_cont
,
655 else if (page
->mapping
)
656 referenced
+= page_referenced_file(page
, mem_cont
,
662 if (page_test_and_clear_young(page
))
668 static int page_mkclean_one(struct page
*page
, struct vm_area_struct
*vma
,
669 unsigned long address
)
671 struct mm_struct
*mm
= vma
->vm_mm
;
676 pte
= page_check_address(page
, mm
, address
, &ptl
, 1);
680 if (pte_dirty(*pte
) || pte_write(*pte
)) {
683 flush_cache_page(vma
, address
, pte_pfn(*pte
));
684 entry
= ptep_clear_flush_notify(vma
, address
, pte
);
685 entry
= pte_wrprotect(entry
);
686 entry
= pte_mkclean(entry
);
687 set_pte_at(mm
, address
, pte
, entry
);
691 pte_unmap_unlock(pte
, ptl
);
696 static int page_mkclean_file(struct address_space
*mapping
, struct page
*page
)
698 pgoff_t pgoff
= page
->index
<< (PAGE_CACHE_SHIFT
- PAGE_SHIFT
);
699 struct vm_area_struct
*vma
;
700 struct prio_tree_iter iter
;
703 BUG_ON(PageAnon(page
));
705 spin_lock(&mapping
->i_mmap_lock
);
706 vma_prio_tree_foreach(vma
, &iter
, &mapping
->i_mmap
, pgoff
, pgoff
) {
707 if (vma
->vm_flags
& VM_SHARED
) {
708 unsigned long address
= vma_address(page
, vma
);
709 if (address
== -EFAULT
)
711 ret
+= page_mkclean_one(page
, vma
, address
);
714 spin_unlock(&mapping
->i_mmap_lock
);
718 int page_mkclean(struct page
*page
)
722 BUG_ON(!PageLocked(page
));
724 if (page_mapped(page
)) {
725 struct address_space
*mapping
= page_mapping(page
);
727 ret
= page_mkclean_file(mapping
, page
);
728 if (page_test_dirty(page
)) {
729 page_clear_dirty(page
);
737 EXPORT_SYMBOL_GPL(page_mkclean
);
740 * page_move_anon_rmap - move a page to our anon_vma
741 * @page: the page to move to our anon_vma
742 * @vma: the vma the page belongs to
743 * @address: the user virtual address mapped
745 * When a page belongs exclusively to one process after a COW event,
746 * that page can be moved into the anon_vma that belongs to just that
747 * process, so the rmap code will not search the parent or sibling
750 void page_move_anon_rmap(struct page
*page
,
751 struct vm_area_struct
*vma
, unsigned long address
)
753 struct anon_vma
*anon_vma
= vma
->anon_vma
;
755 VM_BUG_ON(!PageLocked(page
));
756 VM_BUG_ON(!anon_vma
);
757 VM_BUG_ON(page
->index
!= linear_page_index(vma
, address
));
759 anon_vma
= (void *) anon_vma
+ PAGE_MAPPING_ANON
;
760 page
->mapping
= (struct address_space
*) anon_vma
;
764 * __page_set_anon_rmap - setup new anonymous rmap
765 * @page: the page to add the mapping to
766 * @vma: the vm area in which the mapping is added
767 * @address: the user virtual address mapped
768 * @exclusive: the page is exclusively owned by the current process
770 static void __page_set_anon_rmap(struct page
*page
,
771 struct vm_area_struct
*vma
, unsigned long address
, int exclusive
)
773 struct anon_vma
*anon_vma
= vma
->anon_vma
;
778 * If the page isn't exclusively mapped into this vma,
779 * we must use the _oldest_ possible anon_vma for the
785 anon_vma
= anon_vma
->root
;
788 * In this case, swapped-out-but-not-discarded swap-cache
789 * is remapped. So, no need to update page->mapping here.
790 * We convice anon_vma poitned by page->mapping is not obsolete
791 * because vma->anon_vma is necessary to be a family of it.
797 anon_vma
= (void *) anon_vma
+ PAGE_MAPPING_ANON
;
798 page
->mapping
= (struct address_space
*) anon_vma
;
799 page
->index
= linear_page_index(vma
, address
);
803 * __page_check_anon_rmap - sanity check anonymous rmap addition
804 * @page: the page to add the mapping to
805 * @vma: the vm area in which the mapping is added
806 * @address: the user virtual address mapped
808 static void __page_check_anon_rmap(struct page
*page
,
809 struct vm_area_struct
*vma
, unsigned long address
)
811 #ifdef CONFIG_DEBUG_VM
813 * The page's anon-rmap details (mapping and index) are guaranteed to
814 * be set up correctly at this point.
816 * We have exclusion against page_add_anon_rmap because the caller
817 * always holds the page locked, except if called from page_dup_rmap,
818 * in which case the page is already known to be setup.
820 * We have exclusion against page_add_new_anon_rmap because those pages
821 * are initially only visible via the pagetables, and the pte is locked
822 * over the call to page_add_new_anon_rmap.
824 BUG_ON(page_anon_vma(page
)->root
!= vma
->anon_vma
->root
);
825 BUG_ON(page
->index
!= linear_page_index(vma
, address
));
830 * page_add_anon_rmap - add pte mapping to an anonymous page
831 * @page: the page to add the mapping to
832 * @vma: the vm area in which the mapping is added
833 * @address: the user virtual address mapped
835 * The caller needs to hold the pte lock, and the page must be locked in
836 * the anon_vma case: to serialize mapping,index checking after setting,
837 * and to ensure that PageAnon is not being upgraded racily to PageKsm
838 * (but PageKsm is never downgraded to PageAnon).
840 void page_add_anon_rmap(struct page
*page
,
841 struct vm_area_struct
*vma
, unsigned long address
)
843 do_page_add_anon_rmap(page
, vma
, address
, 0);
847 * Special version of the above for do_swap_page, which often runs
848 * into pages that are exclusively owned by the current process.
849 * Everybody else should continue to use page_add_anon_rmap above.
851 void do_page_add_anon_rmap(struct page
*page
,
852 struct vm_area_struct
*vma
, unsigned long address
, int exclusive
)
854 int first
= atomic_inc_and_test(&page
->_mapcount
);
856 __inc_zone_page_state(page
, NR_ANON_PAGES
);
857 if (unlikely(PageKsm(page
)))
860 VM_BUG_ON(!PageLocked(page
));
861 VM_BUG_ON(address
< vma
->vm_start
|| address
>= vma
->vm_end
);
863 __page_set_anon_rmap(page
, vma
, address
, exclusive
);
865 __page_check_anon_rmap(page
, vma
, address
);
869 * page_add_new_anon_rmap - add pte mapping to a new anonymous page
870 * @page: the page to add the mapping to
871 * @vma: the vm area in which the mapping is added
872 * @address: the user virtual address mapped
874 * Same as page_add_anon_rmap but must only be called on *new* pages.
875 * This means the inc-and-test can be bypassed.
876 * Page does not have to be locked.
878 void page_add_new_anon_rmap(struct page
*page
,
879 struct vm_area_struct
*vma
, unsigned long address
)
881 VM_BUG_ON(address
< vma
->vm_start
|| address
>= vma
->vm_end
);
882 SetPageSwapBacked(page
);
883 atomic_set(&page
->_mapcount
, 0); /* increment count (starts at -1) */
884 __inc_zone_page_state(page
, NR_ANON_PAGES
);
885 __page_set_anon_rmap(page
, vma
, address
, 1);
886 if (page_evictable(page
, vma
))
887 lru_cache_add_lru(page
, LRU_ACTIVE_ANON
);
889 add_page_to_unevictable_list(page
);
893 * page_add_file_rmap - add pte mapping to a file page
894 * @page: the page to add the mapping to
896 * The caller needs to hold the pte lock.
898 void page_add_file_rmap(struct page
*page
)
900 if (atomic_inc_and_test(&page
->_mapcount
)) {
901 __inc_zone_page_state(page
, NR_FILE_MAPPED
);
902 mem_cgroup_update_file_mapped(page
, 1);
907 * page_remove_rmap - take down pte mapping from a page
908 * @page: page to remove mapping from
910 * The caller needs to hold the pte lock.
912 void page_remove_rmap(struct page
*page
)
914 /* page still mapped by someone else? */
915 if (!atomic_add_negative(-1, &page
->_mapcount
))
919 * Now that the last pte has gone, s390 must transfer dirty
920 * flag from storage key to struct page. We can usually skip
921 * this if the page is anon, so about to be freed; but perhaps
922 * not if it's in swapcache - there might be another pte slot
923 * containing the swap entry, but page not yet written to swap.
925 if ((!PageAnon(page
) || PageSwapCache(page
)) && page_test_dirty(page
)) {
926 page_clear_dirty(page
);
927 set_page_dirty(page
);
930 * Hugepages are not counted in NR_ANON_PAGES nor NR_FILE_MAPPED
931 * and not charged by memcg for now.
933 if (unlikely(PageHuge(page
)))
935 if (PageAnon(page
)) {
936 mem_cgroup_uncharge_page(page
);
937 __dec_zone_page_state(page
, NR_ANON_PAGES
);
939 __dec_zone_page_state(page
, NR_FILE_MAPPED
);
940 mem_cgroup_update_file_mapped(page
, -1);
943 * It would be tidy to reset the PageAnon mapping here,
944 * but that might overwrite a racing page_add_anon_rmap
945 * which increments mapcount after us but sets mapping
946 * before us: so leave the reset to free_hot_cold_page,
947 * and remember that it's only reliable while mapped.
948 * Leaving it set also helps swapoff to reinstate ptes
949 * faster for those pages still in swapcache.
954 * Subfunctions of try_to_unmap: try_to_unmap_one called
955 * repeatedly from either try_to_unmap_anon or try_to_unmap_file.
957 int try_to_unmap_one(struct page
*page
, struct vm_area_struct
*vma
,
958 unsigned long address
, enum ttu_flags flags
)
960 struct mm_struct
*mm
= vma
->vm_mm
;
964 int ret
= SWAP_AGAIN
;
966 pte
= page_check_address(page
, mm
, address
, &ptl
, 0);
971 * If the page is mlock()d, we cannot swap it out.
972 * If it's recently referenced (perhaps page_referenced
973 * skipped over this mm) then we should reactivate it.
975 if (!(flags
& TTU_IGNORE_MLOCK
)) {
976 if (vma
->vm_flags
& VM_LOCKED
)
979 if (TTU_ACTION(flags
) == TTU_MUNLOCK
)
982 if (!(flags
& TTU_IGNORE_ACCESS
)) {
983 if (ptep_clear_flush_young_notify(vma
, address
, pte
)) {
989 /* Nuke the page table entry. */
990 flush_cache_page(vma
, address
, page_to_pfn(page
));
991 pteval
= ptep_clear_flush_notify(vma
, address
, pte
);
993 /* Move the dirty bit to the physical page now the pte is gone. */
994 if (pte_dirty(pteval
))
995 set_page_dirty(page
);
997 /* Update high watermark before we lower rss */
998 update_hiwater_rss(mm
);
1000 if (PageHWPoison(page
) && !(flags
& TTU_IGNORE_HWPOISON
)) {
1002 dec_mm_counter(mm
, MM_ANONPAGES
);
1004 dec_mm_counter(mm
, MM_FILEPAGES
);
1005 set_pte_at(mm
, address
, pte
,
1006 swp_entry_to_pte(make_hwpoison_entry(page
)));
1007 } else if (PageAnon(page
)) {
1008 swp_entry_t entry
= { .val
= page_private(page
) };
1010 if (PageSwapCache(page
)) {
1012 * Store the swap location in the pte.
1013 * See handle_pte_fault() ...
1015 if (swap_duplicate(entry
) < 0) {
1016 set_pte_at(mm
, address
, pte
, pteval
);
1020 if (list_empty(&mm
->mmlist
)) {
1021 spin_lock(&mmlist_lock
);
1022 if (list_empty(&mm
->mmlist
))
1023 list_add(&mm
->mmlist
, &init_mm
.mmlist
);
1024 spin_unlock(&mmlist_lock
);
1026 dec_mm_counter(mm
, MM_ANONPAGES
);
1027 inc_mm_counter(mm
, MM_SWAPENTS
);
1028 } else if (PAGE_MIGRATION
) {
1030 * Store the pfn of the page in a special migration
1031 * pte. do_swap_page() will wait until the migration
1032 * pte is removed and then restart fault handling.
1034 BUG_ON(TTU_ACTION(flags
) != TTU_MIGRATION
);
1035 entry
= make_migration_entry(page
, pte_write(pteval
));
1037 set_pte_at(mm
, address
, pte
, swp_entry_to_pte(entry
));
1038 BUG_ON(pte_file(*pte
));
1039 } else if (PAGE_MIGRATION
&& (TTU_ACTION(flags
) == TTU_MIGRATION
)) {
1040 /* Establish migration entry for a file page */
1042 entry
= make_migration_entry(page
, pte_write(pteval
));
1043 set_pte_at(mm
, address
, pte
, swp_entry_to_pte(entry
));
1045 dec_mm_counter(mm
, MM_FILEPAGES
);
1047 page_remove_rmap(page
);
1048 page_cache_release(page
);
1051 pte_unmap_unlock(pte
, ptl
);
1056 pte_unmap_unlock(pte
, ptl
);
1060 * We need mmap_sem locking, Otherwise VM_LOCKED check makes
1061 * unstable result and race. Plus, We can't wait here because
1062 * we now hold anon_vma->lock or mapping->i_mmap_lock.
1063 * if trylock failed, the page remain in evictable lru and later
1064 * vmscan could retry to move the page to unevictable lru if the
1065 * page is actually mlocked.
1067 if (down_read_trylock(&vma
->vm_mm
->mmap_sem
)) {
1068 if (vma
->vm_flags
& VM_LOCKED
) {
1069 mlock_vma_page(page
);
1072 up_read(&vma
->vm_mm
->mmap_sem
);
1078 * objrmap doesn't work for nonlinear VMAs because the assumption that
1079 * offset-into-file correlates with offset-into-virtual-addresses does not hold.
1080 * Consequently, given a particular page and its ->index, we cannot locate the
1081 * ptes which are mapping that page without an exhaustive linear search.
1083 * So what this code does is a mini "virtual scan" of each nonlinear VMA which
1084 * maps the file to which the target page belongs. The ->vm_private_data field
1085 * holds the current cursor into that scan. Successive searches will circulate
1086 * around the vma's virtual address space.
1088 * So as more replacement pressure is applied to the pages in a nonlinear VMA,
1089 * more scanning pressure is placed against them as well. Eventually pages
1090 * will become fully unmapped and are eligible for eviction.
1092 * For very sparsely populated VMAs this is a little inefficient - chances are
1093 * there there won't be many ptes located within the scan cluster. In this case
1094 * maybe we could scan further - to the end of the pte page, perhaps.
1096 * Mlocked pages: check VM_LOCKED under mmap_sem held for read, if we can
1097 * acquire it without blocking. If vma locked, mlock the pages in the cluster,
1098 * rather than unmapping them. If we encounter the "check_page" that vmscan is
1099 * trying to unmap, return SWAP_MLOCK, else default SWAP_AGAIN.
1101 #define CLUSTER_SIZE min(32*PAGE_SIZE, PMD_SIZE)
1102 #define CLUSTER_MASK (~(CLUSTER_SIZE - 1))
1104 static int try_to_unmap_cluster(unsigned long cursor
, unsigned int *mapcount
,
1105 struct vm_area_struct
*vma
, struct page
*check_page
)
1107 struct mm_struct
*mm
= vma
->vm_mm
;
1115 unsigned long address
;
1117 int ret
= SWAP_AGAIN
;
1120 address
= (vma
->vm_start
+ cursor
) & CLUSTER_MASK
;
1121 end
= address
+ CLUSTER_SIZE
;
1122 if (address
< vma
->vm_start
)
1123 address
= vma
->vm_start
;
1124 if (end
> vma
->vm_end
)
1127 pgd
= pgd_offset(mm
, address
);
1128 if (!pgd_present(*pgd
))
1131 pud
= pud_offset(pgd
, address
);
1132 if (!pud_present(*pud
))
1135 pmd
= pmd_offset(pud
, address
);
1136 if (!pmd_present(*pmd
))
1140 * If we can acquire the mmap_sem for read, and vma is VM_LOCKED,
1141 * keep the sem while scanning the cluster for mlocking pages.
1143 if (down_read_trylock(&vma
->vm_mm
->mmap_sem
)) {
1144 locked_vma
= (vma
->vm_flags
& VM_LOCKED
);
1146 up_read(&vma
->vm_mm
->mmap_sem
); /* don't need it */
1149 pte
= pte_offset_map_lock(mm
, pmd
, address
, &ptl
);
1151 /* Update high watermark before we lower rss */
1152 update_hiwater_rss(mm
);
1154 for (; address
< end
; pte
++, address
+= PAGE_SIZE
) {
1155 if (!pte_present(*pte
))
1157 page
= vm_normal_page(vma
, address
, *pte
);
1158 BUG_ON(!page
|| PageAnon(page
));
1161 mlock_vma_page(page
); /* no-op if already mlocked */
1162 if (page
== check_page
)
1164 continue; /* don't unmap */
1167 if (ptep_clear_flush_young_notify(vma
, address
, pte
))
1170 /* Nuke the page table entry. */
1171 flush_cache_page(vma
, address
, pte_pfn(*pte
));
1172 pteval
= ptep_clear_flush_notify(vma
, address
, pte
);
1174 /* If nonlinear, store the file page offset in the pte. */
1175 if (page
->index
!= linear_page_index(vma
, address
))
1176 set_pte_at(mm
, address
, pte
, pgoff_to_pte(page
->index
));
1178 /* Move the dirty bit to the physical page now the pte is gone. */
1179 if (pte_dirty(pteval
))
1180 set_page_dirty(page
);
1182 page_remove_rmap(page
);
1183 page_cache_release(page
);
1184 dec_mm_counter(mm
, MM_FILEPAGES
);
1187 pte_unmap_unlock(pte
- 1, ptl
);
1189 up_read(&vma
->vm_mm
->mmap_sem
);
1193 static bool is_vma_temporary_stack(struct vm_area_struct
*vma
)
1195 int maybe_stack
= vma
->vm_flags
& (VM_GROWSDOWN
| VM_GROWSUP
);
1200 if ((vma
->vm_flags
& VM_STACK_INCOMPLETE_SETUP
) ==
1201 VM_STACK_INCOMPLETE_SETUP
)
1208 * try_to_unmap_anon - unmap or unlock anonymous page using the object-based
1210 * @page: the page to unmap/unlock
1211 * @flags: action and flags
1213 * Find all the mappings of a page using the mapping pointer and the vma chains
1214 * contained in the anon_vma struct it points to.
1216 * This function is only called from try_to_unmap/try_to_munlock for
1218 * When called from try_to_munlock(), the mmap_sem of the mm containing the vma
1219 * where the page was found will be held for write. So, we won't recheck
1220 * vm_flags for that VMA. That should be OK, because that vma shouldn't be
1223 static int try_to_unmap_anon(struct page
*page
, enum ttu_flags flags
)
1225 struct anon_vma
*anon_vma
;
1226 struct anon_vma_chain
*avc
;
1227 int ret
= SWAP_AGAIN
;
1229 anon_vma
= page_lock_anon_vma(page
);
1233 list_for_each_entry(avc
, &anon_vma
->head
, same_anon_vma
) {
1234 struct vm_area_struct
*vma
= avc
->vma
;
1235 unsigned long address
;
1238 * During exec, a temporary VMA is setup and later moved.
1239 * The VMA is moved under the anon_vma lock but not the
1240 * page tables leading to a race where migration cannot
1241 * find the migration ptes. Rather than increasing the
1242 * locking requirements of exec(), migration skips
1243 * temporary VMAs until after exec() completes.
1245 if (PAGE_MIGRATION
&& (flags
& TTU_MIGRATION
) &&
1246 is_vma_temporary_stack(vma
))
1249 address
= vma_address(page
, vma
);
1250 if (address
== -EFAULT
)
1252 ret
= try_to_unmap_one(page
, vma
, address
, flags
);
1253 if (ret
!= SWAP_AGAIN
|| !page_mapped(page
))
1257 page_unlock_anon_vma(anon_vma
);
1262 * try_to_unmap_file - unmap/unlock file page using the object-based rmap method
1263 * @page: the page to unmap/unlock
1264 * @flags: action and flags
1266 * Find all the mappings of a page using the mapping pointer and the vma chains
1267 * contained in the address_space struct it points to.
1269 * This function is only called from try_to_unmap/try_to_munlock for
1270 * object-based pages.
1271 * When called from try_to_munlock(), the mmap_sem of the mm containing the vma
1272 * where the page was found will be held for write. So, we won't recheck
1273 * vm_flags for that VMA. That should be OK, because that vma shouldn't be
1276 static int try_to_unmap_file(struct page
*page
, enum ttu_flags flags
)
1278 struct address_space
*mapping
= page
->mapping
;
1279 pgoff_t pgoff
= page
->index
<< (PAGE_CACHE_SHIFT
- PAGE_SHIFT
);
1280 struct vm_area_struct
*vma
;
1281 struct prio_tree_iter iter
;
1282 int ret
= SWAP_AGAIN
;
1283 unsigned long cursor
;
1284 unsigned long max_nl_cursor
= 0;
1285 unsigned long max_nl_size
= 0;
1286 unsigned int mapcount
;
1288 spin_lock(&mapping
->i_mmap_lock
);
1289 vma_prio_tree_foreach(vma
, &iter
, &mapping
->i_mmap
, pgoff
, pgoff
) {
1290 unsigned long address
= vma_address(page
, vma
);
1291 if (address
== -EFAULT
)
1293 ret
= try_to_unmap_one(page
, vma
, address
, flags
);
1294 if (ret
!= SWAP_AGAIN
|| !page_mapped(page
))
1298 if (list_empty(&mapping
->i_mmap_nonlinear
))
1302 * We don't bother to try to find the munlocked page in nonlinears.
1303 * It's costly. Instead, later, page reclaim logic may call
1304 * try_to_unmap(TTU_MUNLOCK) and recover PG_mlocked lazily.
1306 if (TTU_ACTION(flags
) == TTU_MUNLOCK
)
1309 list_for_each_entry(vma
, &mapping
->i_mmap_nonlinear
,
1310 shared
.vm_set
.list
) {
1311 cursor
= (unsigned long) vma
->vm_private_data
;
1312 if (cursor
> max_nl_cursor
)
1313 max_nl_cursor
= cursor
;
1314 cursor
= vma
->vm_end
- vma
->vm_start
;
1315 if (cursor
> max_nl_size
)
1316 max_nl_size
= cursor
;
1319 if (max_nl_size
== 0) { /* all nonlinears locked or reserved ? */
1325 * We don't try to search for this page in the nonlinear vmas,
1326 * and page_referenced wouldn't have found it anyway. Instead
1327 * just walk the nonlinear vmas trying to age and unmap some.
1328 * The mapcount of the page we came in with is irrelevant,
1329 * but even so use it as a guide to how hard we should try?
1331 mapcount
= page_mapcount(page
);
1334 cond_resched_lock(&mapping
->i_mmap_lock
);
1336 max_nl_size
= (max_nl_size
+ CLUSTER_SIZE
- 1) & CLUSTER_MASK
;
1337 if (max_nl_cursor
== 0)
1338 max_nl_cursor
= CLUSTER_SIZE
;
1341 list_for_each_entry(vma
, &mapping
->i_mmap_nonlinear
,
1342 shared
.vm_set
.list
) {
1343 cursor
= (unsigned long) vma
->vm_private_data
;
1344 while ( cursor
< max_nl_cursor
&&
1345 cursor
< vma
->vm_end
- vma
->vm_start
) {
1346 if (try_to_unmap_cluster(cursor
, &mapcount
,
1347 vma
, page
) == SWAP_MLOCK
)
1349 cursor
+= CLUSTER_SIZE
;
1350 vma
->vm_private_data
= (void *) cursor
;
1351 if ((int)mapcount
<= 0)
1354 vma
->vm_private_data
= (void *) max_nl_cursor
;
1356 cond_resched_lock(&mapping
->i_mmap_lock
);
1357 max_nl_cursor
+= CLUSTER_SIZE
;
1358 } while (max_nl_cursor
<= max_nl_size
);
1361 * Don't loop forever (perhaps all the remaining pages are
1362 * in locked vmas). Reset cursor on all unreserved nonlinear
1363 * vmas, now forgetting on which ones it had fallen behind.
1365 list_for_each_entry(vma
, &mapping
->i_mmap_nonlinear
, shared
.vm_set
.list
)
1366 vma
->vm_private_data
= NULL
;
1368 spin_unlock(&mapping
->i_mmap_lock
);
1373 * try_to_unmap - try to remove all page table mappings to a page
1374 * @page: the page to get unmapped
1375 * @flags: action and flags
1377 * Tries to remove all the page table entries which are mapping this
1378 * page, used in the pageout path. Caller must hold the page lock.
1379 * Return values are:
1381 * SWAP_SUCCESS - we succeeded in removing all mappings
1382 * SWAP_AGAIN - we missed a mapping, try again later
1383 * SWAP_FAIL - the page is unswappable
1384 * SWAP_MLOCK - page is mlocked.
1386 int try_to_unmap(struct page
*page
, enum ttu_flags flags
)
1390 BUG_ON(!PageLocked(page
));
1392 if (unlikely(PageKsm(page
)))
1393 ret
= try_to_unmap_ksm(page
, flags
);
1394 else if (PageAnon(page
))
1395 ret
= try_to_unmap_anon(page
, flags
);
1397 ret
= try_to_unmap_file(page
, flags
);
1398 if (ret
!= SWAP_MLOCK
&& !page_mapped(page
))
1404 * try_to_munlock - try to munlock a page
1405 * @page: the page to be munlocked
1407 * Called from munlock code. Checks all of the VMAs mapping the page
1408 * to make sure nobody else has this page mlocked. The page will be
1409 * returned with PG_mlocked cleared if no other vmas have it mlocked.
1411 * Return values are:
1413 * SWAP_AGAIN - no vma is holding page mlocked, or,
1414 * SWAP_AGAIN - page mapped in mlocked vma -- couldn't acquire mmap sem
1415 * SWAP_FAIL - page cannot be located at present
1416 * SWAP_MLOCK - page is now mlocked.
1418 int try_to_munlock(struct page
*page
)
1420 VM_BUG_ON(!PageLocked(page
) || PageLRU(page
));
1422 if (unlikely(PageKsm(page
)))
1423 return try_to_unmap_ksm(page
, TTU_MUNLOCK
);
1424 else if (PageAnon(page
))
1425 return try_to_unmap_anon(page
, TTU_MUNLOCK
);
1427 return try_to_unmap_file(page
, TTU_MUNLOCK
);
1430 #if defined(CONFIG_KSM) || defined(CONFIG_MIGRATION)
1432 * Drop an anon_vma refcount, freeing the anon_vma and anon_vma->root
1433 * if necessary. Be careful to do all the tests under the lock. Once
1434 * we know we are the last user, nobody else can get a reference and we
1435 * can do the freeing without the lock.
1437 void drop_anon_vma(struct anon_vma
*anon_vma
)
1439 BUG_ON(atomic_read(&anon_vma
->external_refcount
) <= 0);
1440 if (atomic_dec_and_lock(&anon_vma
->external_refcount
, &anon_vma
->root
->lock
)) {
1441 struct anon_vma
*root
= anon_vma
->root
;
1442 int empty
= list_empty(&anon_vma
->head
);
1443 int last_root_user
= 0;
1447 * The refcount on a non-root anon_vma got dropped. Drop
1448 * the refcount on the root and check if we need to free it.
1450 if (empty
&& anon_vma
!= root
) {
1451 BUG_ON(atomic_read(&root
->external_refcount
) <= 0);
1452 last_root_user
= atomic_dec_and_test(&root
->external_refcount
);
1453 root_empty
= list_empty(&root
->head
);
1455 anon_vma_unlock(anon_vma
);
1458 anon_vma_free(anon_vma
);
1459 if (root_empty
&& last_root_user
)
1460 anon_vma_free(root
);
1466 #ifdef CONFIG_MIGRATION
1468 * rmap_walk() and its helpers rmap_walk_anon() and rmap_walk_file():
1469 * Called by migrate.c to remove migration ptes, but might be used more later.
1471 static int rmap_walk_anon(struct page
*page
, int (*rmap_one
)(struct page
*,
1472 struct vm_area_struct
*, unsigned long, void *), void *arg
)
1474 struct anon_vma
*anon_vma
;
1475 struct anon_vma_chain
*avc
;
1476 int ret
= SWAP_AGAIN
;
1479 * Note: remove_migration_ptes() cannot use page_lock_anon_vma()
1480 * because that depends on page_mapped(); but not all its usages
1481 * are holding mmap_sem. Users without mmap_sem are required to
1482 * take a reference count to prevent the anon_vma disappearing
1484 anon_vma
= page_anon_vma(page
);
1487 anon_vma_lock(anon_vma
);
1488 list_for_each_entry(avc
, &anon_vma
->head
, same_anon_vma
) {
1489 struct vm_area_struct
*vma
= avc
->vma
;
1490 unsigned long address
= vma_address(page
, vma
);
1491 if (address
== -EFAULT
)
1493 ret
= rmap_one(page
, vma
, address
, arg
);
1494 if (ret
!= SWAP_AGAIN
)
1497 anon_vma_unlock(anon_vma
);
1501 static int rmap_walk_file(struct page
*page
, int (*rmap_one
)(struct page
*,
1502 struct vm_area_struct
*, unsigned long, void *), void *arg
)
1504 struct address_space
*mapping
= page
->mapping
;
1505 pgoff_t pgoff
= page
->index
<< (PAGE_CACHE_SHIFT
- PAGE_SHIFT
);
1506 struct vm_area_struct
*vma
;
1507 struct prio_tree_iter iter
;
1508 int ret
= SWAP_AGAIN
;
1512 spin_lock(&mapping
->i_mmap_lock
);
1513 vma_prio_tree_foreach(vma
, &iter
, &mapping
->i_mmap
, pgoff
, pgoff
) {
1514 unsigned long address
= vma_address(page
, vma
);
1515 if (address
== -EFAULT
)
1517 ret
= rmap_one(page
, vma
, address
, arg
);
1518 if (ret
!= SWAP_AGAIN
)
1522 * No nonlinear handling: being always shared, nonlinear vmas
1523 * never contain migration ptes. Decide what to do about this
1524 * limitation to linear when we need rmap_walk() on nonlinear.
1526 spin_unlock(&mapping
->i_mmap_lock
);
1530 int rmap_walk(struct page
*page
, int (*rmap_one
)(struct page
*,
1531 struct vm_area_struct
*, unsigned long, void *), void *arg
)
1533 VM_BUG_ON(!PageLocked(page
));
1535 if (unlikely(PageKsm(page
)))
1536 return rmap_walk_ksm(page
, rmap_one
, arg
);
1537 else if (PageAnon(page
))
1538 return rmap_walk_anon(page
, rmap_one
, arg
);
1540 return rmap_walk_file(page
, rmap_one
, arg
);
1542 #endif /* CONFIG_MIGRATION */
1544 #ifdef CONFIG_HUGETLB_PAGE
1546 * The following three functions are for anonymous (private mapped) hugepages.
1547 * Unlike common anonymous pages, anonymous hugepages have no accounting code
1548 * and no lru code, because we handle hugepages differently from common pages.
1550 static void __hugepage_set_anon_rmap(struct page
*page
,
1551 struct vm_area_struct
*vma
, unsigned long address
, int exclusive
)
1553 struct anon_vma
*anon_vma
= vma
->anon_vma
;
1556 struct anon_vma_chain
*avc
;
1557 avc
= list_entry(vma
->anon_vma_chain
.prev
,
1558 struct anon_vma_chain
, same_vma
);
1559 anon_vma
= avc
->anon_vma
;
1561 anon_vma
= (void *) anon_vma
+ PAGE_MAPPING_ANON
;
1562 page
->mapping
= (struct address_space
*) anon_vma
;
1563 page
->index
= linear_page_index(vma
, address
);
1566 void hugepage_add_anon_rmap(struct page
*page
,
1567 struct vm_area_struct
*vma
, unsigned long address
)
1569 struct anon_vma
*anon_vma
= vma
->anon_vma
;
1572 BUG_ON(address
< vma
->vm_start
|| address
>= vma
->vm_end
);
1573 first
= atomic_inc_and_test(&page
->_mapcount
);
1575 __hugepage_set_anon_rmap(page
, vma
, address
, 0);
1578 void hugepage_add_new_anon_rmap(struct page
*page
,
1579 struct vm_area_struct
*vma
, unsigned long address
)
1581 BUG_ON(address
< vma
->vm_start
|| address
>= vma
->vm_end
);
1582 atomic_set(&page
->_mapcount
, 0);
1583 __hugepage_set_anon_rmap(page
, vma
, address
, 1);
1585 #endif /* CONFIG_HUGETLB_PAGE */