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_mutex
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->i_lock (in set_page_dirty's __mark_inode_dirty)
35 * inode_wb_list_lock (in set_page_dirty's __mark_inode_dirty)
36 * sb_lock (within inode_lock in fs/fs-writeback.c)
37 * mapping->tree_lock (widely used, in set_page_dirty,
38 * in arch-dependent flush_dcache_mmap_lock,
39 * within inode_wb_list_lock in __sync_single_inode)
41 * (code doesn't rely on that order so it could be switched around)
43 * anon_vma->lock (memory_failure, collect_procs_anon)
48 #include <linux/pagemap.h>
49 #include <linux/swap.h>
50 #include <linux/swapops.h>
51 #include <linux/slab.h>
52 #include <linux/init.h>
53 #include <linux/ksm.h>
54 #include <linux/rmap.h>
55 #include <linux/rcupdate.h>
56 #include <linux/module.h>
57 #include <linux/memcontrol.h>
58 #include <linux/mmu_notifier.h>
59 #include <linux/migrate.h>
60 #include <linux/hugetlb.h>
62 #include <asm/tlbflush.h>
66 static struct kmem_cache
*anon_vma_cachep
;
67 static struct kmem_cache
*anon_vma_chain_cachep
;
69 static inline struct anon_vma
*anon_vma_alloc(void)
71 struct anon_vma
*anon_vma
;
73 anon_vma
= kmem_cache_alloc(anon_vma_cachep
, GFP_KERNEL
);
75 atomic_set(&anon_vma
->refcount
, 1);
77 * Initialise the anon_vma root to point to itself. If called
78 * from fork, the root will be reset to the parents anon_vma.
80 anon_vma
->root
= anon_vma
;
86 static inline void anon_vma_free(struct anon_vma
*anon_vma
)
88 VM_BUG_ON(atomic_read(&anon_vma
->refcount
));
89 kmem_cache_free(anon_vma_cachep
, anon_vma
);
92 static inline struct anon_vma_chain
*anon_vma_chain_alloc(void)
94 return kmem_cache_alloc(anon_vma_chain_cachep
, GFP_KERNEL
);
97 static void anon_vma_chain_free(struct anon_vma_chain
*anon_vma_chain
)
99 kmem_cache_free(anon_vma_chain_cachep
, anon_vma_chain
);
103 * anon_vma_prepare - attach an anon_vma to a memory region
104 * @vma: the memory region in question
106 * This makes sure the memory mapping described by 'vma' has
107 * an 'anon_vma' attached to it, so that we can associate the
108 * anonymous pages mapped into it with that anon_vma.
110 * The common case will be that we already have one, but if
111 * not we either need to find an adjacent mapping that we
112 * can re-use the anon_vma from (very common when the only
113 * reason for splitting a vma has been mprotect()), or we
114 * allocate a new one.
116 * Anon-vma allocations are very subtle, because we may have
117 * optimistically looked up an anon_vma in page_lock_anon_vma()
118 * and that may actually touch the spinlock even in the newly
119 * allocated vma (it depends on RCU to make sure that the
120 * anon_vma isn't actually destroyed).
122 * As a result, we need to do proper anon_vma locking even
123 * for the new allocation. At the same time, we do not want
124 * to do any locking for the common case of already having
127 * This must be called with the mmap_sem held for reading.
129 int anon_vma_prepare(struct vm_area_struct
*vma
)
131 struct anon_vma
*anon_vma
= vma
->anon_vma
;
132 struct anon_vma_chain
*avc
;
135 if (unlikely(!anon_vma
)) {
136 struct mm_struct
*mm
= vma
->vm_mm
;
137 struct anon_vma
*allocated
;
139 avc
= anon_vma_chain_alloc();
143 anon_vma
= find_mergeable_anon_vma(vma
);
146 anon_vma
= anon_vma_alloc();
147 if (unlikely(!anon_vma
))
148 goto out_enomem_free_avc
;
149 allocated
= anon_vma
;
152 anon_vma_lock(anon_vma
);
153 /* page_table_lock to protect against threads */
154 spin_lock(&mm
->page_table_lock
);
155 if (likely(!vma
->anon_vma
)) {
156 vma
->anon_vma
= anon_vma
;
157 avc
->anon_vma
= anon_vma
;
159 list_add(&avc
->same_vma
, &vma
->anon_vma_chain
);
160 list_add_tail(&avc
->same_anon_vma
, &anon_vma
->head
);
164 spin_unlock(&mm
->page_table_lock
);
165 anon_vma_unlock(anon_vma
);
167 if (unlikely(allocated
))
168 put_anon_vma(allocated
);
170 anon_vma_chain_free(avc
);
175 anon_vma_chain_free(avc
);
180 static void anon_vma_chain_link(struct vm_area_struct
*vma
,
181 struct anon_vma_chain
*avc
,
182 struct anon_vma
*anon_vma
)
185 avc
->anon_vma
= anon_vma
;
186 list_add(&avc
->same_vma
, &vma
->anon_vma_chain
);
188 anon_vma_lock(anon_vma
);
190 * It's critical to add new vmas to the tail of the anon_vma,
191 * see comment in huge_memory.c:__split_huge_page().
193 list_add_tail(&avc
->same_anon_vma
, &anon_vma
->head
);
194 anon_vma_unlock(anon_vma
);
198 * Attach the anon_vmas from src to dst.
199 * Returns 0 on success, -ENOMEM on failure.
201 int anon_vma_clone(struct vm_area_struct
*dst
, struct vm_area_struct
*src
)
203 struct anon_vma_chain
*avc
, *pavc
;
205 list_for_each_entry_reverse(pavc
, &src
->anon_vma_chain
, same_vma
) {
206 avc
= anon_vma_chain_alloc();
209 anon_vma_chain_link(dst
, avc
, pavc
->anon_vma
);
214 unlink_anon_vmas(dst
);
219 * Attach vma to its own anon_vma, as well as to the anon_vmas that
220 * the corresponding VMA in the parent process is attached to.
221 * Returns 0 on success, non-zero on failure.
223 int anon_vma_fork(struct vm_area_struct
*vma
, struct vm_area_struct
*pvma
)
225 struct anon_vma_chain
*avc
;
226 struct anon_vma
*anon_vma
;
228 /* Don't bother if the parent process has no anon_vma here. */
233 * First, attach the new VMA to the parent VMA's anon_vmas,
234 * so rmap can find non-COWed pages in child processes.
236 if (anon_vma_clone(vma
, pvma
))
239 /* Then add our own anon_vma. */
240 anon_vma
= anon_vma_alloc();
243 avc
= anon_vma_chain_alloc();
245 goto out_error_free_anon_vma
;
248 * The root anon_vma's spinlock is the lock actually used when we
249 * lock any of the anon_vmas in this anon_vma tree.
251 anon_vma
->root
= pvma
->anon_vma
->root
;
253 * With refcounts, an anon_vma can stay around longer than the
254 * process it belongs to. The root anon_vma needs to be pinned until
255 * this anon_vma is freed, because the lock lives in the root.
257 get_anon_vma(anon_vma
->root
);
258 /* Mark this anon_vma as the one where our new (COWed) pages go. */
259 vma
->anon_vma
= anon_vma
;
260 anon_vma_chain_link(vma
, avc
, anon_vma
);
264 out_error_free_anon_vma
:
265 put_anon_vma(anon_vma
);
267 unlink_anon_vmas(vma
);
271 static void anon_vma_unlink(struct anon_vma_chain
*anon_vma_chain
)
273 struct anon_vma
*anon_vma
= anon_vma_chain
->anon_vma
;
276 /* If anon_vma_fork fails, we can get an empty anon_vma_chain. */
280 anon_vma_lock(anon_vma
);
281 list_del(&anon_vma_chain
->same_anon_vma
);
283 /* We must garbage collect the anon_vma if it's empty */
284 empty
= list_empty(&anon_vma
->head
);
285 anon_vma_unlock(anon_vma
);
288 put_anon_vma(anon_vma
);
291 void unlink_anon_vmas(struct vm_area_struct
*vma
)
293 struct anon_vma_chain
*avc
, *next
;
296 * Unlink each anon_vma chained to the VMA. This list is ordered
297 * from newest to oldest, ensuring the root anon_vma gets freed last.
299 list_for_each_entry_safe(avc
, next
, &vma
->anon_vma_chain
, same_vma
) {
300 anon_vma_unlink(avc
);
301 list_del(&avc
->same_vma
);
302 anon_vma_chain_free(avc
);
306 static void anon_vma_ctor(void *data
)
308 struct anon_vma
*anon_vma
= data
;
310 spin_lock_init(&anon_vma
->lock
);
311 atomic_set(&anon_vma
->refcount
, 0);
312 INIT_LIST_HEAD(&anon_vma
->head
);
315 void __init
anon_vma_init(void)
317 anon_vma_cachep
= kmem_cache_create("anon_vma", sizeof(struct anon_vma
),
318 0, SLAB_DESTROY_BY_RCU
|SLAB_PANIC
, anon_vma_ctor
);
319 anon_vma_chain_cachep
= KMEM_CACHE(anon_vma_chain
, SLAB_PANIC
);
323 * Getting a lock on a stable anon_vma from a page off the LRU is
324 * tricky: page_lock_anon_vma rely on RCU to guard against the races.
326 struct anon_vma
*page_lock_anon_vma(struct page
*page
)
328 struct anon_vma
*anon_vma
, *root_anon_vma
;
329 unsigned long anon_mapping
;
332 anon_mapping
= (unsigned long) ACCESS_ONCE(page
->mapping
);
333 if ((anon_mapping
& PAGE_MAPPING_FLAGS
) != PAGE_MAPPING_ANON
)
335 if (!page_mapped(page
))
338 anon_vma
= (struct anon_vma
*) (anon_mapping
- PAGE_MAPPING_ANON
);
339 root_anon_vma
= ACCESS_ONCE(anon_vma
->root
);
340 spin_lock(&root_anon_vma
->lock
);
343 * If this page is still mapped, then its anon_vma cannot have been
344 * freed. But if it has been unmapped, we have no security against
345 * the anon_vma structure being freed and reused (for another anon_vma:
346 * SLAB_DESTROY_BY_RCU guarantees that - so the spin_lock above cannot
347 * corrupt): with anon_vma_prepare() or anon_vma_fork() redirecting
348 * anon_vma->root before page_unlock_anon_vma() is called to unlock.
350 if (page_mapped(page
))
353 spin_unlock(&root_anon_vma
->lock
);
359 void page_unlock_anon_vma(struct anon_vma
*anon_vma
)
361 anon_vma_unlock(anon_vma
);
366 * At what user virtual address is page expected in @vma?
367 * Returns virtual address or -EFAULT if page's index/offset is not
368 * within the range mapped the @vma.
371 vma_address(struct page
*page
, struct vm_area_struct
*vma
)
373 pgoff_t pgoff
= page
->index
<< (PAGE_CACHE_SHIFT
- PAGE_SHIFT
);
374 unsigned long address
;
376 if (unlikely(is_vm_hugetlb_page(vma
)))
377 pgoff
= page
->index
<< huge_page_order(page_hstate(page
));
378 address
= vma
->vm_start
+ ((pgoff
- vma
->vm_pgoff
) << PAGE_SHIFT
);
379 if (unlikely(address
< vma
->vm_start
|| address
>= vma
->vm_end
)) {
380 /* page should be within @vma mapping range */
387 * At what user virtual address is page expected in vma?
388 * Caller should check the page is actually part of the vma.
390 unsigned long page_address_in_vma(struct page
*page
, struct vm_area_struct
*vma
)
392 if (PageAnon(page
)) {
393 struct anon_vma
*page__anon_vma
= page_anon_vma(page
);
395 * Note: swapoff's unuse_vma() is more efficient with this
396 * check, and needs it to match anon_vma when KSM is active.
398 if (!vma
->anon_vma
|| !page__anon_vma
||
399 vma
->anon_vma
->root
!= page__anon_vma
->root
)
401 } else if (page
->mapping
&& !(vma
->vm_flags
& VM_NONLINEAR
)) {
403 vma
->vm_file
->f_mapping
!= page
->mapping
)
407 return vma_address(page
, vma
);
411 * Check that @page is mapped at @address into @mm.
413 * If @sync is false, page_check_address may perform a racy check to avoid
414 * the page table lock when the pte is not present (helpful when reclaiming
415 * highly shared pages).
417 * On success returns with pte mapped and locked.
419 pte_t
*__page_check_address(struct page
*page
, struct mm_struct
*mm
,
420 unsigned long address
, spinlock_t
**ptlp
, int sync
)
428 if (unlikely(PageHuge(page
))) {
429 pte
= huge_pte_offset(mm
, address
);
430 ptl
= &mm
->page_table_lock
;
434 pgd
= pgd_offset(mm
, address
);
435 if (!pgd_present(*pgd
))
438 pud
= pud_offset(pgd
, address
);
439 if (!pud_present(*pud
))
442 pmd
= pmd_offset(pud
, address
);
443 if (!pmd_present(*pmd
))
445 if (pmd_trans_huge(*pmd
))
448 pte
= pte_offset_map(pmd
, address
);
449 /* Make a quick check before getting the lock */
450 if (!sync
&& !pte_present(*pte
)) {
455 ptl
= pte_lockptr(mm
, pmd
);
458 if (pte_present(*pte
) && page_to_pfn(page
) == pte_pfn(*pte
)) {
462 pte_unmap_unlock(pte
, ptl
);
467 * page_mapped_in_vma - check whether a page is really mapped in a VMA
468 * @page: the page to test
469 * @vma: the VMA to test
471 * Returns 1 if the page is mapped into the page tables of the VMA, 0
472 * if the page is not mapped into the page tables of this VMA. Only
473 * valid for normal file or anonymous VMAs.
475 int page_mapped_in_vma(struct page
*page
, struct vm_area_struct
*vma
)
477 unsigned long address
;
481 address
= vma_address(page
, vma
);
482 if (address
== -EFAULT
) /* out of vma range */
484 pte
= page_check_address(page
, vma
->vm_mm
, address
, &ptl
, 1);
485 if (!pte
) /* the page is not in this mm */
487 pte_unmap_unlock(pte
, ptl
);
493 * Subfunctions of page_referenced: page_referenced_one called
494 * repeatedly from either page_referenced_anon or page_referenced_file.
496 int page_referenced_one(struct page
*page
, struct vm_area_struct
*vma
,
497 unsigned long address
, unsigned int *mapcount
,
498 unsigned long *vm_flags
)
500 struct mm_struct
*mm
= vma
->vm_mm
;
503 if (unlikely(PageTransHuge(page
))) {
506 spin_lock(&mm
->page_table_lock
);
508 * rmap might return false positives; we must filter
509 * these out using page_check_address_pmd().
511 pmd
= page_check_address_pmd(page
, mm
, address
,
512 PAGE_CHECK_ADDRESS_PMD_FLAG
);
514 spin_unlock(&mm
->page_table_lock
);
518 if (vma
->vm_flags
& VM_LOCKED
) {
519 spin_unlock(&mm
->page_table_lock
);
520 *mapcount
= 0; /* break early from loop */
521 *vm_flags
|= VM_LOCKED
;
525 /* go ahead even if the pmd is pmd_trans_splitting() */
526 if (pmdp_clear_flush_young_notify(vma
, address
, pmd
))
528 spin_unlock(&mm
->page_table_lock
);
534 * rmap might return false positives; we must filter
535 * these out using page_check_address().
537 pte
= page_check_address(page
, mm
, address
, &ptl
, 0);
541 if (vma
->vm_flags
& VM_LOCKED
) {
542 pte_unmap_unlock(pte
, ptl
);
543 *mapcount
= 0; /* break early from loop */
544 *vm_flags
|= VM_LOCKED
;
548 if (ptep_clear_flush_young_notify(vma
, address
, pte
)) {
550 * Don't treat a reference through a sequentially read
551 * mapping as such. If the page has been used in
552 * another mapping, we will catch it; if this other
553 * mapping is already gone, the unmap path will have
554 * set PG_referenced or activated the page.
556 if (likely(!VM_SequentialReadHint(vma
)))
559 pte_unmap_unlock(pte
, ptl
);
562 /* Pretend the page is referenced if the task has the
563 swap token and is in the middle of a page fault. */
564 if (mm
!= current
->mm
&& has_swap_token(mm
) &&
565 rwsem_is_locked(&mm
->mmap_sem
))
571 *vm_flags
|= vma
->vm_flags
;
576 static int page_referenced_anon(struct page
*page
,
577 struct mem_cgroup
*mem_cont
,
578 unsigned long *vm_flags
)
580 unsigned int mapcount
;
581 struct anon_vma
*anon_vma
;
582 struct anon_vma_chain
*avc
;
585 anon_vma
= page_lock_anon_vma(page
);
589 mapcount
= page_mapcount(page
);
590 list_for_each_entry(avc
, &anon_vma
->head
, same_anon_vma
) {
591 struct vm_area_struct
*vma
= avc
->vma
;
592 unsigned long address
= vma_address(page
, vma
);
593 if (address
== -EFAULT
)
596 * If we are reclaiming on behalf of a cgroup, skip
597 * counting on behalf of references from different
600 if (mem_cont
&& !mm_match_cgroup(vma
->vm_mm
, mem_cont
))
602 referenced
+= page_referenced_one(page
, vma
, address
,
603 &mapcount
, vm_flags
);
608 page_unlock_anon_vma(anon_vma
);
613 * page_referenced_file - referenced check for object-based rmap
614 * @page: the page we're checking references on.
615 * @mem_cont: target memory controller
616 * @vm_flags: collect encountered vma->vm_flags who actually referenced the page
618 * For an object-based mapped page, find all the places it is mapped and
619 * check/clear the referenced flag. This is done by following the page->mapping
620 * pointer, then walking the chain of vmas it holds. It returns the number
621 * of references it found.
623 * This function is only called from page_referenced for object-based pages.
625 static int page_referenced_file(struct page
*page
,
626 struct mem_cgroup
*mem_cont
,
627 unsigned long *vm_flags
)
629 unsigned int mapcount
;
630 struct address_space
*mapping
= page
->mapping
;
631 pgoff_t pgoff
= page
->index
<< (PAGE_CACHE_SHIFT
- PAGE_SHIFT
);
632 struct vm_area_struct
*vma
;
633 struct prio_tree_iter iter
;
637 * The caller's checks on page->mapping and !PageAnon have made
638 * sure that this is a file page: the check for page->mapping
639 * excludes the case just before it gets set on an anon page.
641 BUG_ON(PageAnon(page
));
644 * The page lock not only makes sure that page->mapping cannot
645 * suddenly be NULLified by truncation, it makes sure that the
646 * structure at mapping cannot be freed and reused yet,
647 * so we can safely take mapping->i_mmap_mutex.
649 BUG_ON(!PageLocked(page
));
651 mutex_lock(&mapping
->i_mmap_mutex
);
654 * i_mmap_mutex does not stabilize mapcount at all, but mapcount
655 * is more likely to be accurate if we note it after spinning.
657 mapcount
= page_mapcount(page
);
659 vma_prio_tree_foreach(vma
, &iter
, &mapping
->i_mmap
, pgoff
, pgoff
) {
660 unsigned long address
= vma_address(page
, vma
);
661 if (address
== -EFAULT
)
664 * If we are reclaiming on behalf of a cgroup, skip
665 * counting on behalf of references from different
668 if (mem_cont
&& !mm_match_cgroup(vma
->vm_mm
, mem_cont
))
670 referenced
+= page_referenced_one(page
, vma
, address
,
671 &mapcount
, vm_flags
);
676 mutex_unlock(&mapping
->i_mmap_mutex
);
681 * page_referenced - test if the page was referenced
682 * @page: the page to test
683 * @is_locked: caller holds lock on the page
684 * @mem_cont: target memory controller
685 * @vm_flags: collect encountered vma->vm_flags who actually referenced the page
687 * Quick test_and_clear_referenced for all mappings to a page,
688 * returns the number of ptes which referenced the page.
690 int page_referenced(struct page
*page
,
692 struct mem_cgroup
*mem_cont
,
693 unsigned long *vm_flags
)
699 if (page_mapped(page
) && page_rmapping(page
)) {
700 if (!is_locked
&& (!PageAnon(page
) || PageKsm(page
))) {
701 we_locked
= trylock_page(page
);
707 if (unlikely(PageKsm(page
)))
708 referenced
+= page_referenced_ksm(page
, mem_cont
,
710 else if (PageAnon(page
))
711 referenced
+= page_referenced_anon(page
, mem_cont
,
713 else if (page
->mapping
)
714 referenced
+= page_referenced_file(page
, mem_cont
,
720 if (page_test_and_clear_young(page_to_pfn(page
)))
726 static int page_mkclean_one(struct page
*page
, struct vm_area_struct
*vma
,
727 unsigned long address
)
729 struct mm_struct
*mm
= vma
->vm_mm
;
734 pte
= page_check_address(page
, mm
, address
, &ptl
, 1);
738 if (pte_dirty(*pte
) || pte_write(*pte
)) {
741 flush_cache_page(vma
, address
, pte_pfn(*pte
));
742 entry
= ptep_clear_flush_notify(vma
, address
, pte
);
743 entry
= pte_wrprotect(entry
);
744 entry
= pte_mkclean(entry
);
745 set_pte_at(mm
, address
, pte
, entry
);
749 pte_unmap_unlock(pte
, ptl
);
754 static int page_mkclean_file(struct address_space
*mapping
, struct page
*page
)
756 pgoff_t pgoff
= page
->index
<< (PAGE_CACHE_SHIFT
- PAGE_SHIFT
);
757 struct vm_area_struct
*vma
;
758 struct prio_tree_iter iter
;
761 BUG_ON(PageAnon(page
));
763 mutex_lock(&mapping
->i_mmap_mutex
);
764 vma_prio_tree_foreach(vma
, &iter
, &mapping
->i_mmap
, pgoff
, pgoff
) {
765 if (vma
->vm_flags
& VM_SHARED
) {
766 unsigned long address
= vma_address(page
, vma
);
767 if (address
== -EFAULT
)
769 ret
+= page_mkclean_one(page
, vma
, address
);
772 mutex_unlock(&mapping
->i_mmap_mutex
);
776 int page_mkclean(struct page
*page
)
780 BUG_ON(!PageLocked(page
));
782 if (page_mapped(page
)) {
783 struct address_space
*mapping
= page_mapping(page
);
785 ret
= page_mkclean_file(mapping
, page
);
786 if (page_test_and_clear_dirty(page_to_pfn(page
), 1))
793 EXPORT_SYMBOL_GPL(page_mkclean
);
796 * page_move_anon_rmap - move a page to our anon_vma
797 * @page: the page to move to our anon_vma
798 * @vma: the vma the page belongs to
799 * @address: the user virtual address mapped
801 * When a page belongs exclusively to one process after a COW event,
802 * that page can be moved into the anon_vma that belongs to just that
803 * process, so the rmap code will not search the parent or sibling
806 void page_move_anon_rmap(struct page
*page
,
807 struct vm_area_struct
*vma
, unsigned long address
)
809 struct anon_vma
*anon_vma
= vma
->anon_vma
;
811 VM_BUG_ON(!PageLocked(page
));
812 VM_BUG_ON(!anon_vma
);
813 VM_BUG_ON(page
->index
!= linear_page_index(vma
, address
));
815 anon_vma
= (void *) anon_vma
+ PAGE_MAPPING_ANON
;
816 page
->mapping
= (struct address_space
*) anon_vma
;
820 * __page_set_anon_rmap - set up new anonymous rmap
821 * @page: Page to add to rmap
822 * @vma: VM area to add page to.
823 * @address: User virtual address of the mapping
824 * @exclusive: the page is exclusively owned by the current process
826 static void __page_set_anon_rmap(struct page
*page
,
827 struct vm_area_struct
*vma
, unsigned long address
, int exclusive
)
829 struct anon_vma
*anon_vma
= vma
->anon_vma
;
837 * If the page isn't exclusively mapped into this vma,
838 * we must use the _oldest_ possible anon_vma for the
842 anon_vma
= anon_vma
->root
;
844 anon_vma
= (void *) anon_vma
+ PAGE_MAPPING_ANON
;
845 page
->mapping
= (struct address_space
*) anon_vma
;
846 page
->index
= linear_page_index(vma
, address
);
850 * __page_check_anon_rmap - sanity check anonymous rmap addition
851 * @page: the page to add the mapping to
852 * @vma: the vm area in which the mapping is added
853 * @address: the user virtual address mapped
855 static void __page_check_anon_rmap(struct page
*page
,
856 struct vm_area_struct
*vma
, unsigned long address
)
858 #ifdef CONFIG_DEBUG_VM
860 * The page's anon-rmap details (mapping and index) are guaranteed to
861 * be set up correctly at this point.
863 * We have exclusion against page_add_anon_rmap because the caller
864 * always holds the page locked, except if called from page_dup_rmap,
865 * in which case the page is already known to be setup.
867 * We have exclusion against page_add_new_anon_rmap because those pages
868 * are initially only visible via the pagetables, and the pte is locked
869 * over the call to page_add_new_anon_rmap.
871 BUG_ON(page_anon_vma(page
)->root
!= vma
->anon_vma
->root
);
872 BUG_ON(page
->index
!= linear_page_index(vma
, address
));
877 * page_add_anon_rmap - add pte mapping to an anonymous page
878 * @page: the page to add the mapping to
879 * @vma: the vm area in which the mapping is added
880 * @address: the user virtual address mapped
882 * The caller needs to hold the pte lock, and the page must be locked in
883 * the anon_vma case: to serialize mapping,index checking after setting,
884 * and to ensure that PageAnon is not being upgraded racily to PageKsm
885 * (but PageKsm is never downgraded to PageAnon).
887 void page_add_anon_rmap(struct page
*page
,
888 struct vm_area_struct
*vma
, unsigned long address
)
890 do_page_add_anon_rmap(page
, vma
, address
, 0);
894 * Special version of the above for do_swap_page, which often runs
895 * into pages that are exclusively owned by the current process.
896 * Everybody else should continue to use page_add_anon_rmap above.
898 void do_page_add_anon_rmap(struct page
*page
,
899 struct vm_area_struct
*vma
, unsigned long address
, int exclusive
)
901 int first
= atomic_inc_and_test(&page
->_mapcount
);
903 if (!PageTransHuge(page
))
904 __inc_zone_page_state(page
, NR_ANON_PAGES
);
906 __inc_zone_page_state(page
,
907 NR_ANON_TRANSPARENT_HUGEPAGES
);
909 if (unlikely(PageKsm(page
)))
912 VM_BUG_ON(!PageLocked(page
));
913 VM_BUG_ON(address
< vma
->vm_start
|| address
>= vma
->vm_end
);
915 __page_set_anon_rmap(page
, vma
, address
, exclusive
);
917 __page_check_anon_rmap(page
, vma
, address
);
921 * page_add_new_anon_rmap - add pte mapping to a new anonymous page
922 * @page: the page to add the mapping to
923 * @vma: the vm area in which the mapping is added
924 * @address: the user virtual address mapped
926 * Same as page_add_anon_rmap but must only be called on *new* pages.
927 * This means the inc-and-test can be bypassed.
928 * Page does not have to be locked.
930 void page_add_new_anon_rmap(struct page
*page
,
931 struct vm_area_struct
*vma
, unsigned long address
)
933 VM_BUG_ON(address
< vma
->vm_start
|| address
>= vma
->vm_end
);
934 SetPageSwapBacked(page
);
935 atomic_set(&page
->_mapcount
, 0); /* increment count (starts at -1) */
936 if (!PageTransHuge(page
))
937 __inc_zone_page_state(page
, NR_ANON_PAGES
);
939 __inc_zone_page_state(page
, NR_ANON_TRANSPARENT_HUGEPAGES
);
940 __page_set_anon_rmap(page
, vma
, address
, 1);
941 if (page_evictable(page
, vma
))
942 lru_cache_add_lru(page
, LRU_ACTIVE_ANON
);
944 add_page_to_unevictable_list(page
);
948 * page_add_file_rmap - add pte mapping to a file page
949 * @page: the page to add the mapping to
951 * The caller needs to hold the pte lock.
953 void page_add_file_rmap(struct page
*page
)
955 if (atomic_inc_and_test(&page
->_mapcount
)) {
956 __inc_zone_page_state(page
, NR_FILE_MAPPED
);
957 mem_cgroup_inc_page_stat(page
, MEMCG_NR_FILE_MAPPED
);
962 * page_remove_rmap - take down pte mapping from a page
963 * @page: page to remove mapping from
965 * The caller needs to hold the pte lock.
967 void page_remove_rmap(struct page
*page
)
969 /* page still mapped by someone else? */
970 if (!atomic_add_negative(-1, &page
->_mapcount
))
974 * Now that the last pte has gone, s390 must transfer dirty
975 * flag from storage key to struct page. We can usually skip
976 * this if the page is anon, so about to be freed; but perhaps
977 * not if it's in swapcache - there might be another pte slot
978 * containing the swap entry, but page not yet written to swap.
980 if ((!PageAnon(page
) || PageSwapCache(page
)) &&
981 page_test_and_clear_dirty(page_to_pfn(page
), 1))
982 set_page_dirty(page
);
984 * Hugepages are not counted in NR_ANON_PAGES nor NR_FILE_MAPPED
985 * and not charged by memcg for now.
987 if (unlikely(PageHuge(page
)))
989 if (PageAnon(page
)) {
990 mem_cgroup_uncharge_page(page
);
991 if (!PageTransHuge(page
))
992 __dec_zone_page_state(page
, NR_ANON_PAGES
);
994 __dec_zone_page_state(page
,
995 NR_ANON_TRANSPARENT_HUGEPAGES
);
997 __dec_zone_page_state(page
, NR_FILE_MAPPED
);
998 mem_cgroup_dec_page_stat(page
, MEMCG_NR_FILE_MAPPED
);
1001 * It would be tidy to reset the PageAnon mapping here,
1002 * but that might overwrite a racing page_add_anon_rmap
1003 * which increments mapcount after us but sets mapping
1004 * before us: so leave the reset to free_hot_cold_page,
1005 * and remember that it's only reliable while mapped.
1006 * Leaving it set also helps swapoff to reinstate ptes
1007 * faster for those pages still in swapcache.
1012 * Subfunctions of try_to_unmap: try_to_unmap_one called
1013 * repeatedly from either try_to_unmap_anon or try_to_unmap_file.
1015 int try_to_unmap_one(struct page
*page
, struct vm_area_struct
*vma
,
1016 unsigned long address
, enum ttu_flags flags
)
1018 struct mm_struct
*mm
= vma
->vm_mm
;
1022 int ret
= SWAP_AGAIN
;
1024 pte
= page_check_address(page
, mm
, address
, &ptl
, 0);
1029 * If the page is mlock()d, we cannot swap it out.
1030 * If it's recently referenced (perhaps page_referenced
1031 * skipped over this mm) then we should reactivate it.
1033 if (!(flags
& TTU_IGNORE_MLOCK
)) {
1034 if (vma
->vm_flags
& VM_LOCKED
)
1037 if (TTU_ACTION(flags
) == TTU_MUNLOCK
)
1040 if (!(flags
& TTU_IGNORE_ACCESS
)) {
1041 if (ptep_clear_flush_young_notify(vma
, address
, pte
)) {
1047 /* Nuke the page table entry. */
1048 flush_cache_page(vma
, address
, page_to_pfn(page
));
1049 pteval
= ptep_clear_flush_notify(vma
, address
, pte
);
1051 /* Move the dirty bit to the physical page now the pte is gone. */
1052 if (pte_dirty(pteval
))
1053 set_page_dirty(page
);
1055 /* Update high watermark before we lower rss */
1056 update_hiwater_rss(mm
);
1058 if (PageHWPoison(page
) && !(flags
& TTU_IGNORE_HWPOISON
)) {
1060 dec_mm_counter(mm
, MM_ANONPAGES
);
1062 dec_mm_counter(mm
, MM_FILEPAGES
);
1063 set_pte_at(mm
, address
, pte
,
1064 swp_entry_to_pte(make_hwpoison_entry(page
)));
1065 } else if (PageAnon(page
)) {
1066 swp_entry_t entry
= { .val
= page_private(page
) };
1068 if (PageSwapCache(page
)) {
1070 * Store the swap location in the pte.
1071 * See handle_pte_fault() ...
1073 if (swap_duplicate(entry
) < 0) {
1074 set_pte_at(mm
, address
, pte
, pteval
);
1078 if (list_empty(&mm
->mmlist
)) {
1079 spin_lock(&mmlist_lock
);
1080 if (list_empty(&mm
->mmlist
))
1081 list_add(&mm
->mmlist
, &init_mm
.mmlist
);
1082 spin_unlock(&mmlist_lock
);
1084 dec_mm_counter(mm
, MM_ANONPAGES
);
1085 inc_mm_counter(mm
, MM_SWAPENTS
);
1086 } else if (PAGE_MIGRATION
) {
1088 * Store the pfn of the page in a special migration
1089 * pte. do_swap_page() will wait until the migration
1090 * pte is removed and then restart fault handling.
1092 BUG_ON(TTU_ACTION(flags
) != TTU_MIGRATION
);
1093 entry
= make_migration_entry(page
, pte_write(pteval
));
1095 set_pte_at(mm
, address
, pte
, swp_entry_to_pte(entry
));
1096 BUG_ON(pte_file(*pte
));
1097 } else if (PAGE_MIGRATION
&& (TTU_ACTION(flags
) == TTU_MIGRATION
)) {
1098 /* Establish migration entry for a file page */
1100 entry
= make_migration_entry(page
, pte_write(pteval
));
1101 set_pte_at(mm
, address
, pte
, swp_entry_to_pte(entry
));
1103 dec_mm_counter(mm
, MM_FILEPAGES
);
1105 page_remove_rmap(page
);
1106 page_cache_release(page
);
1109 pte_unmap_unlock(pte
, ptl
);
1114 pte_unmap_unlock(pte
, ptl
);
1118 * We need mmap_sem locking, Otherwise VM_LOCKED check makes
1119 * unstable result and race. Plus, We can't wait here because
1120 * we now hold anon_vma->lock or mapping->i_mmap_mutex.
1121 * if trylock failed, the page remain in evictable lru and later
1122 * vmscan could retry to move the page to unevictable lru if the
1123 * page is actually mlocked.
1125 if (down_read_trylock(&vma
->vm_mm
->mmap_sem
)) {
1126 if (vma
->vm_flags
& VM_LOCKED
) {
1127 mlock_vma_page(page
);
1130 up_read(&vma
->vm_mm
->mmap_sem
);
1136 * objrmap doesn't work for nonlinear VMAs because the assumption that
1137 * offset-into-file correlates with offset-into-virtual-addresses does not hold.
1138 * Consequently, given a particular page and its ->index, we cannot locate the
1139 * ptes which are mapping that page without an exhaustive linear search.
1141 * So what this code does is a mini "virtual scan" of each nonlinear VMA which
1142 * maps the file to which the target page belongs. The ->vm_private_data field
1143 * holds the current cursor into that scan. Successive searches will circulate
1144 * around the vma's virtual address space.
1146 * So as more replacement pressure is applied to the pages in a nonlinear VMA,
1147 * more scanning pressure is placed against them as well. Eventually pages
1148 * will become fully unmapped and are eligible for eviction.
1150 * For very sparsely populated VMAs this is a little inefficient - chances are
1151 * there there won't be many ptes located within the scan cluster. In this case
1152 * maybe we could scan further - to the end of the pte page, perhaps.
1154 * Mlocked pages: check VM_LOCKED under mmap_sem held for read, if we can
1155 * acquire it without blocking. If vma locked, mlock the pages in the cluster,
1156 * rather than unmapping them. If we encounter the "check_page" that vmscan is
1157 * trying to unmap, return SWAP_MLOCK, else default SWAP_AGAIN.
1159 #define CLUSTER_SIZE min(32*PAGE_SIZE, PMD_SIZE)
1160 #define CLUSTER_MASK (~(CLUSTER_SIZE - 1))
1162 static int try_to_unmap_cluster(unsigned long cursor
, unsigned int *mapcount
,
1163 struct vm_area_struct
*vma
, struct page
*check_page
)
1165 struct mm_struct
*mm
= vma
->vm_mm
;
1173 unsigned long address
;
1175 int ret
= SWAP_AGAIN
;
1178 address
= (vma
->vm_start
+ cursor
) & CLUSTER_MASK
;
1179 end
= address
+ CLUSTER_SIZE
;
1180 if (address
< vma
->vm_start
)
1181 address
= vma
->vm_start
;
1182 if (end
> vma
->vm_end
)
1185 pgd
= pgd_offset(mm
, address
);
1186 if (!pgd_present(*pgd
))
1189 pud
= pud_offset(pgd
, address
);
1190 if (!pud_present(*pud
))
1193 pmd
= pmd_offset(pud
, address
);
1194 if (!pmd_present(*pmd
))
1198 * If we can acquire the mmap_sem for read, and vma is VM_LOCKED,
1199 * keep the sem while scanning the cluster for mlocking pages.
1201 if (down_read_trylock(&vma
->vm_mm
->mmap_sem
)) {
1202 locked_vma
= (vma
->vm_flags
& VM_LOCKED
);
1204 up_read(&vma
->vm_mm
->mmap_sem
); /* don't need it */
1207 pte
= pte_offset_map_lock(mm
, pmd
, address
, &ptl
);
1209 /* Update high watermark before we lower rss */
1210 update_hiwater_rss(mm
);
1212 for (; address
< end
; pte
++, address
+= PAGE_SIZE
) {
1213 if (!pte_present(*pte
))
1215 page
= vm_normal_page(vma
, address
, *pte
);
1216 BUG_ON(!page
|| PageAnon(page
));
1219 mlock_vma_page(page
); /* no-op if already mlocked */
1220 if (page
== check_page
)
1222 continue; /* don't unmap */
1225 if (ptep_clear_flush_young_notify(vma
, address
, pte
))
1228 /* Nuke the page table entry. */
1229 flush_cache_page(vma
, address
, pte_pfn(*pte
));
1230 pteval
= ptep_clear_flush_notify(vma
, address
, pte
);
1232 /* If nonlinear, store the file page offset in the pte. */
1233 if (page
->index
!= linear_page_index(vma
, address
))
1234 set_pte_at(mm
, address
, pte
, pgoff_to_pte(page
->index
));
1236 /* Move the dirty bit to the physical page now the pte is gone. */
1237 if (pte_dirty(pteval
))
1238 set_page_dirty(page
);
1240 page_remove_rmap(page
);
1241 page_cache_release(page
);
1242 dec_mm_counter(mm
, MM_FILEPAGES
);
1245 pte_unmap_unlock(pte
- 1, ptl
);
1247 up_read(&vma
->vm_mm
->mmap_sem
);
1251 bool is_vma_temporary_stack(struct vm_area_struct
*vma
)
1253 int maybe_stack
= vma
->vm_flags
& (VM_GROWSDOWN
| VM_GROWSUP
);
1258 if ((vma
->vm_flags
& VM_STACK_INCOMPLETE_SETUP
) ==
1259 VM_STACK_INCOMPLETE_SETUP
)
1266 * try_to_unmap_anon - unmap or unlock anonymous page using the object-based
1268 * @page: the page to unmap/unlock
1269 * @flags: action and flags
1271 * Find all the mappings of a page using the mapping pointer and the vma chains
1272 * contained in the anon_vma struct it points to.
1274 * This function is only called from try_to_unmap/try_to_munlock for
1276 * When called from try_to_munlock(), the mmap_sem of the mm containing the vma
1277 * where the page was found will be held for write. So, we won't recheck
1278 * vm_flags for that VMA. That should be OK, because that vma shouldn't be
1281 static int try_to_unmap_anon(struct page
*page
, enum ttu_flags flags
)
1283 struct anon_vma
*anon_vma
;
1284 struct anon_vma_chain
*avc
;
1285 int ret
= SWAP_AGAIN
;
1287 anon_vma
= page_lock_anon_vma(page
);
1291 list_for_each_entry(avc
, &anon_vma
->head
, same_anon_vma
) {
1292 struct vm_area_struct
*vma
= avc
->vma
;
1293 unsigned long address
;
1296 * During exec, a temporary VMA is setup and later moved.
1297 * The VMA is moved under the anon_vma lock but not the
1298 * page tables leading to a race where migration cannot
1299 * find the migration ptes. Rather than increasing the
1300 * locking requirements of exec(), migration skips
1301 * temporary VMAs until after exec() completes.
1303 if (PAGE_MIGRATION
&& (flags
& TTU_MIGRATION
) &&
1304 is_vma_temporary_stack(vma
))
1307 address
= vma_address(page
, vma
);
1308 if (address
== -EFAULT
)
1310 ret
= try_to_unmap_one(page
, vma
, address
, flags
);
1311 if (ret
!= SWAP_AGAIN
|| !page_mapped(page
))
1315 page_unlock_anon_vma(anon_vma
);
1320 * try_to_unmap_file - unmap/unlock file page using the object-based rmap method
1321 * @page: the page to unmap/unlock
1322 * @flags: action and flags
1324 * Find all the mappings of a page using the mapping pointer and the vma chains
1325 * contained in the address_space struct it points to.
1327 * This function is only called from try_to_unmap/try_to_munlock for
1328 * object-based pages.
1329 * When called from try_to_munlock(), the mmap_sem of the mm containing the vma
1330 * where the page was found will be held for write. So, we won't recheck
1331 * vm_flags for that VMA. That should be OK, because that vma shouldn't be
1334 static int try_to_unmap_file(struct page
*page
, enum ttu_flags flags
)
1336 struct address_space
*mapping
= page
->mapping
;
1337 pgoff_t pgoff
= page
->index
<< (PAGE_CACHE_SHIFT
- PAGE_SHIFT
);
1338 struct vm_area_struct
*vma
;
1339 struct prio_tree_iter iter
;
1340 int ret
= SWAP_AGAIN
;
1341 unsigned long cursor
;
1342 unsigned long max_nl_cursor
= 0;
1343 unsigned long max_nl_size
= 0;
1344 unsigned int mapcount
;
1346 mutex_lock(&mapping
->i_mmap_mutex
);
1347 vma_prio_tree_foreach(vma
, &iter
, &mapping
->i_mmap
, pgoff
, pgoff
) {
1348 unsigned long address
= vma_address(page
, vma
);
1349 if (address
== -EFAULT
)
1351 ret
= try_to_unmap_one(page
, vma
, address
, flags
);
1352 if (ret
!= SWAP_AGAIN
|| !page_mapped(page
))
1356 if (list_empty(&mapping
->i_mmap_nonlinear
))
1360 * We don't bother to try to find the munlocked page in nonlinears.
1361 * It's costly. Instead, later, page reclaim logic may call
1362 * try_to_unmap(TTU_MUNLOCK) and recover PG_mlocked lazily.
1364 if (TTU_ACTION(flags
) == TTU_MUNLOCK
)
1367 list_for_each_entry(vma
, &mapping
->i_mmap_nonlinear
,
1368 shared
.vm_set
.list
) {
1369 cursor
= (unsigned long) vma
->vm_private_data
;
1370 if (cursor
> max_nl_cursor
)
1371 max_nl_cursor
= cursor
;
1372 cursor
= vma
->vm_end
- vma
->vm_start
;
1373 if (cursor
> max_nl_size
)
1374 max_nl_size
= cursor
;
1377 if (max_nl_size
== 0) { /* all nonlinears locked or reserved ? */
1383 * We don't try to search for this page in the nonlinear vmas,
1384 * and page_referenced wouldn't have found it anyway. Instead
1385 * just walk the nonlinear vmas trying to age and unmap some.
1386 * The mapcount of the page we came in with is irrelevant,
1387 * but even so use it as a guide to how hard we should try?
1389 mapcount
= page_mapcount(page
);
1394 max_nl_size
= (max_nl_size
+ CLUSTER_SIZE
- 1) & CLUSTER_MASK
;
1395 if (max_nl_cursor
== 0)
1396 max_nl_cursor
= CLUSTER_SIZE
;
1399 list_for_each_entry(vma
, &mapping
->i_mmap_nonlinear
,
1400 shared
.vm_set
.list
) {
1401 cursor
= (unsigned long) vma
->vm_private_data
;
1402 while ( cursor
< max_nl_cursor
&&
1403 cursor
< vma
->vm_end
- vma
->vm_start
) {
1404 if (try_to_unmap_cluster(cursor
, &mapcount
,
1405 vma
, page
) == SWAP_MLOCK
)
1407 cursor
+= CLUSTER_SIZE
;
1408 vma
->vm_private_data
= (void *) cursor
;
1409 if ((int)mapcount
<= 0)
1412 vma
->vm_private_data
= (void *) max_nl_cursor
;
1415 max_nl_cursor
+= CLUSTER_SIZE
;
1416 } while (max_nl_cursor
<= max_nl_size
);
1419 * Don't loop forever (perhaps all the remaining pages are
1420 * in locked vmas). Reset cursor on all unreserved nonlinear
1421 * vmas, now forgetting on which ones it had fallen behind.
1423 list_for_each_entry(vma
, &mapping
->i_mmap_nonlinear
, shared
.vm_set
.list
)
1424 vma
->vm_private_data
= NULL
;
1426 mutex_unlock(&mapping
->i_mmap_mutex
);
1431 * try_to_unmap - try to remove all page table mappings to a page
1432 * @page: the page to get unmapped
1433 * @flags: action and flags
1435 * Tries to remove all the page table entries which are mapping this
1436 * page, used in the pageout path. Caller must hold the page lock.
1437 * Return values are:
1439 * SWAP_SUCCESS - we succeeded in removing all mappings
1440 * SWAP_AGAIN - we missed a mapping, try again later
1441 * SWAP_FAIL - the page is unswappable
1442 * SWAP_MLOCK - page is mlocked.
1444 int try_to_unmap(struct page
*page
, enum ttu_flags flags
)
1448 BUG_ON(!PageLocked(page
));
1449 VM_BUG_ON(!PageHuge(page
) && PageTransHuge(page
));
1451 if (unlikely(PageKsm(page
)))
1452 ret
= try_to_unmap_ksm(page
, flags
);
1453 else if (PageAnon(page
))
1454 ret
= try_to_unmap_anon(page
, flags
);
1456 ret
= try_to_unmap_file(page
, flags
);
1457 if (ret
!= SWAP_MLOCK
&& !page_mapped(page
))
1463 * try_to_munlock - try to munlock a page
1464 * @page: the page to be munlocked
1466 * Called from munlock code. Checks all of the VMAs mapping the page
1467 * to make sure nobody else has this page mlocked. The page will be
1468 * returned with PG_mlocked cleared if no other vmas have it mlocked.
1470 * Return values are:
1472 * SWAP_AGAIN - no vma is holding page mlocked, or,
1473 * SWAP_AGAIN - page mapped in mlocked vma -- couldn't acquire mmap sem
1474 * SWAP_FAIL - page cannot be located at present
1475 * SWAP_MLOCK - page is now mlocked.
1477 int try_to_munlock(struct page
*page
)
1479 VM_BUG_ON(!PageLocked(page
) || PageLRU(page
));
1481 if (unlikely(PageKsm(page
)))
1482 return try_to_unmap_ksm(page
, TTU_MUNLOCK
);
1483 else if (PageAnon(page
))
1484 return try_to_unmap_anon(page
, TTU_MUNLOCK
);
1486 return try_to_unmap_file(page
, TTU_MUNLOCK
);
1489 void __put_anon_vma(struct anon_vma
*anon_vma
)
1491 struct anon_vma
*root
= anon_vma
->root
;
1493 if (root
!= anon_vma
&& atomic_dec_and_test(&root
->refcount
))
1494 anon_vma_free(root
);
1496 anon_vma_free(anon_vma
);
1499 #ifdef CONFIG_MIGRATION
1501 * rmap_walk() and its helpers rmap_walk_anon() and rmap_walk_file():
1502 * Called by migrate.c to remove migration ptes, but might be used more later.
1504 static int rmap_walk_anon(struct page
*page
, int (*rmap_one
)(struct page
*,
1505 struct vm_area_struct
*, unsigned long, void *), void *arg
)
1507 struct anon_vma
*anon_vma
;
1508 struct anon_vma_chain
*avc
;
1509 int ret
= SWAP_AGAIN
;
1512 * Note: remove_migration_ptes() cannot use page_lock_anon_vma()
1513 * because that depends on page_mapped(); but not all its usages
1514 * are holding mmap_sem. Users without mmap_sem are required to
1515 * take a reference count to prevent the anon_vma disappearing
1517 anon_vma
= page_anon_vma(page
);
1520 anon_vma_lock(anon_vma
);
1521 list_for_each_entry(avc
, &anon_vma
->head
, same_anon_vma
) {
1522 struct vm_area_struct
*vma
= avc
->vma
;
1523 unsigned long address
= vma_address(page
, vma
);
1524 if (address
== -EFAULT
)
1526 ret
= rmap_one(page
, vma
, address
, arg
);
1527 if (ret
!= SWAP_AGAIN
)
1530 anon_vma_unlock(anon_vma
);
1534 static int rmap_walk_file(struct page
*page
, int (*rmap_one
)(struct page
*,
1535 struct vm_area_struct
*, unsigned long, void *), void *arg
)
1537 struct address_space
*mapping
= page
->mapping
;
1538 pgoff_t pgoff
= page
->index
<< (PAGE_CACHE_SHIFT
- PAGE_SHIFT
);
1539 struct vm_area_struct
*vma
;
1540 struct prio_tree_iter iter
;
1541 int ret
= SWAP_AGAIN
;
1545 mutex_lock(&mapping
->i_mmap_mutex
);
1546 vma_prio_tree_foreach(vma
, &iter
, &mapping
->i_mmap
, pgoff
, pgoff
) {
1547 unsigned long address
= vma_address(page
, vma
);
1548 if (address
== -EFAULT
)
1550 ret
= rmap_one(page
, vma
, address
, arg
);
1551 if (ret
!= SWAP_AGAIN
)
1555 * No nonlinear handling: being always shared, nonlinear vmas
1556 * never contain migration ptes. Decide what to do about this
1557 * limitation to linear when we need rmap_walk() on nonlinear.
1559 mutex_unlock(&mapping
->i_mmap_mutex
);
1563 int rmap_walk(struct page
*page
, int (*rmap_one
)(struct page
*,
1564 struct vm_area_struct
*, unsigned long, void *), void *arg
)
1566 VM_BUG_ON(!PageLocked(page
));
1568 if (unlikely(PageKsm(page
)))
1569 return rmap_walk_ksm(page
, rmap_one
, arg
);
1570 else if (PageAnon(page
))
1571 return rmap_walk_anon(page
, rmap_one
, arg
);
1573 return rmap_walk_file(page
, rmap_one
, arg
);
1575 #endif /* CONFIG_MIGRATION */
1577 #ifdef CONFIG_HUGETLB_PAGE
1579 * The following three functions are for anonymous (private mapped) hugepages.
1580 * Unlike common anonymous pages, anonymous hugepages have no accounting code
1581 * and no lru code, because we handle hugepages differently from common pages.
1583 static void __hugepage_set_anon_rmap(struct page
*page
,
1584 struct vm_area_struct
*vma
, unsigned long address
, int exclusive
)
1586 struct anon_vma
*anon_vma
= vma
->anon_vma
;
1593 anon_vma
= anon_vma
->root
;
1595 anon_vma
= (void *) anon_vma
+ PAGE_MAPPING_ANON
;
1596 page
->mapping
= (struct address_space
*) anon_vma
;
1597 page
->index
= linear_page_index(vma
, address
);
1600 void hugepage_add_anon_rmap(struct page
*page
,
1601 struct vm_area_struct
*vma
, unsigned long address
)
1603 struct anon_vma
*anon_vma
= vma
->anon_vma
;
1606 BUG_ON(!PageLocked(page
));
1608 BUG_ON(address
< vma
->vm_start
|| address
>= vma
->vm_end
);
1609 first
= atomic_inc_and_test(&page
->_mapcount
);
1611 __hugepage_set_anon_rmap(page
, vma
, address
, 0);
1614 void hugepage_add_new_anon_rmap(struct page
*page
,
1615 struct vm_area_struct
*vma
, unsigned long address
)
1617 BUG_ON(address
< vma
->vm_start
|| address
>= vma
->vm_end
);
1618 atomic_set(&page
->_mapcount
, 0);
1619 __hugepage_set_anon_rmap(page
, vma
, address
, 1);
1621 #endif /* CONFIG_HUGETLB_PAGE */