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 tricky!
325 * Since there is no serialization what so ever against page_remove_rmap()
326 * the best this function can do is return a locked anon_vma that might
327 * have been relevant to this page.
329 * The page might have been remapped to a different anon_vma or the anon_vma
330 * returned may already be freed (and even reused).
332 * All users of this function must be very careful when walking the anon_vma
333 * chain and verify that the page in question is indeed mapped in it
334 * [ something equivalent to page_mapped_in_vma() ].
336 * Since anon_vma's slab is DESTROY_BY_RCU and we know from page_remove_rmap()
337 * that the anon_vma pointer from page->mapping is valid if there is a
338 * mapcount, we can dereference the anon_vma after observing those.
340 struct anon_vma
*page_lock_anon_vma(struct page
*page
)
342 struct anon_vma
*anon_vma
, *root_anon_vma
;
343 unsigned long anon_mapping
;
346 anon_mapping
= (unsigned long) ACCESS_ONCE(page
->mapping
);
347 if ((anon_mapping
& PAGE_MAPPING_FLAGS
) != PAGE_MAPPING_ANON
)
349 if (!page_mapped(page
))
352 anon_vma
= (struct anon_vma
*) (anon_mapping
- PAGE_MAPPING_ANON
);
353 root_anon_vma
= ACCESS_ONCE(anon_vma
->root
);
354 spin_lock(&root_anon_vma
->lock
);
357 * If this page is still mapped, then its anon_vma cannot have been
358 * freed. But if it has been unmapped, we have no security against
359 * the anon_vma structure being freed and reused (for another anon_vma:
360 * SLAB_DESTROY_BY_RCU guarantees that - so the spin_lock above cannot
361 * corrupt): with anon_vma_prepare() or anon_vma_fork() redirecting
362 * anon_vma->root before page_unlock_anon_vma() is called to unlock.
364 if (page_mapped(page
))
367 spin_unlock(&root_anon_vma
->lock
);
373 void page_unlock_anon_vma(struct anon_vma
*anon_vma
)
375 anon_vma_unlock(anon_vma
);
380 * At what user virtual address is page expected in @vma?
381 * Returns virtual address or -EFAULT if page's index/offset is not
382 * within the range mapped the @vma.
385 vma_address(struct page
*page
, struct vm_area_struct
*vma
)
387 pgoff_t pgoff
= page
->index
<< (PAGE_CACHE_SHIFT
- PAGE_SHIFT
);
388 unsigned long address
;
390 if (unlikely(is_vm_hugetlb_page(vma
)))
391 pgoff
= page
->index
<< huge_page_order(page_hstate(page
));
392 address
= vma
->vm_start
+ ((pgoff
- vma
->vm_pgoff
) << PAGE_SHIFT
);
393 if (unlikely(address
< vma
->vm_start
|| address
>= vma
->vm_end
)) {
394 /* page should be within @vma mapping range */
401 * At what user virtual address is page expected in vma?
402 * Caller should check the page is actually part of the vma.
404 unsigned long page_address_in_vma(struct page
*page
, struct vm_area_struct
*vma
)
406 if (PageAnon(page
)) {
407 struct anon_vma
*page__anon_vma
= page_anon_vma(page
);
409 * Note: swapoff's unuse_vma() is more efficient with this
410 * check, and needs it to match anon_vma when KSM is active.
412 if (!vma
->anon_vma
|| !page__anon_vma
||
413 vma
->anon_vma
->root
!= page__anon_vma
->root
)
415 } else if (page
->mapping
&& !(vma
->vm_flags
& VM_NONLINEAR
)) {
417 vma
->vm_file
->f_mapping
!= page
->mapping
)
421 return vma_address(page
, vma
);
425 * Check that @page is mapped at @address into @mm.
427 * If @sync is false, page_check_address may perform a racy check to avoid
428 * the page table lock when the pte is not present (helpful when reclaiming
429 * highly shared pages).
431 * On success returns with pte mapped and locked.
433 pte_t
*__page_check_address(struct page
*page
, struct mm_struct
*mm
,
434 unsigned long address
, spinlock_t
**ptlp
, int sync
)
442 if (unlikely(PageHuge(page
))) {
443 pte
= huge_pte_offset(mm
, address
);
444 ptl
= &mm
->page_table_lock
;
448 pgd
= pgd_offset(mm
, address
);
449 if (!pgd_present(*pgd
))
452 pud
= pud_offset(pgd
, address
);
453 if (!pud_present(*pud
))
456 pmd
= pmd_offset(pud
, address
);
457 if (!pmd_present(*pmd
))
459 if (pmd_trans_huge(*pmd
))
462 pte
= pte_offset_map(pmd
, address
);
463 /* Make a quick check before getting the lock */
464 if (!sync
&& !pte_present(*pte
)) {
469 ptl
= pte_lockptr(mm
, pmd
);
472 if (pte_present(*pte
) && page_to_pfn(page
) == pte_pfn(*pte
)) {
476 pte_unmap_unlock(pte
, ptl
);
481 * page_mapped_in_vma - check whether a page is really mapped in a VMA
482 * @page: the page to test
483 * @vma: the VMA to test
485 * Returns 1 if the page is mapped into the page tables of the VMA, 0
486 * if the page is not mapped into the page tables of this VMA. Only
487 * valid for normal file or anonymous VMAs.
489 int page_mapped_in_vma(struct page
*page
, struct vm_area_struct
*vma
)
491 unsigned long address
;
495 address
= vma_address(page
, vma
);
496 if (address
== -EFAULT
) /* out of vma range */
498 pte
= page_check_address(page
, vma
->vm_mm
, address
, &ptl
, 1);
499 if (!pte
) /* the page is not in this mm */
501 pte_unmap_unlock(pte
, ptl
);
507 * Subfunctions of page_referenced: page_referenced_one called
508 * repeatedly from either page_referenced_anon or page_referenced_file.
510 int page_referenced_one(struct page
*page
, struct vm_area_struct
*vma
,
511 unsigned long address
, unsigned int *mapcount
,
512 unsigned long *vm_flags
)
514 struct mm_struct
*mm
= vma
->vm_mm
;
517 if (unlikely(PageTransHuge(page
))) {
520 spin_lock(&mm
->page_table_lock
);
522 * rmap might return false positives; we must filter
523 * these out using page_check_address_pmd().
525 pmd
= page_check_address_pmd(page
, mm
, address
,
526 PAGE_CHECK_ADDRESS_PMD_FLAG
);
528 spin_unlock(&mm
->page_table_lock
);
532 if (vma
->vm_flags
& VM_LOCKED
) {
533 spin_unlock(&mm
->page_table_lock
);
534 *mapcount
= 0; /* break early from loop */
535 *vm_flags
|= VM_LOCKED
;
539 /* go ahead even if the pmd is pmd_trans_splitting() */
540 if (pmdp_clear_flush_young_notify(vma
, address
, pmd
))
542 spin_unlock(&mm
->page_table_lock
);
548 * rmap might return false positives; we must filter
549 * these out using page_check_address().
551 pte
= page_check_address(page
, mm
, address
, &ptl
, 0);
555 if (vma
->vm_flags
& VM_LOCKED
) {
556 pte_unmap_unlock(pte
, ptl
);
557 *mapcount
= 0; /* break early from loop */
558 *vm_flags
|= VM_LOCKED
;
562 if (ptep_clear_flush_young_notify(vma
, address
, pte
)) {
564 * Don't treat a reference through a sequentially read
565 * mapping as such. If the page has been used in
566 * another mapping, we will catch it; if this other
567 * mapping is already gone, the unmap path will have
568 * set PG_referenced or activated the page.
570 if (likely(!VM_SequentialReadHint(vma
)))
573 pte_unmap_unlock(pte
, ptl
);
576 /* Pretend the page is referenced if the task has the
577 swap token and is in the middle of a page fault. */
578 if (mm
!= current
->mm
&& has_swap_token(mm
) &&
579 rwsem_is_locked(&mm
->mmap_sem
))
585 *vm_flags
|= vma
->vm_flags
;
590 static int page_referenced_anon(struct page
*page
,
591 struct mem_cgroup
*mem_cont
,
592 unsigned long *vm_flags
)
594 unsigned int mapcount
;
595 struct anon_vma
*anon_vma
;
596 struct anon_vma_chain
*avc
;
599 anon_vma
= page_lock_anon_vma(page
);
603 mapcount
= page_mapcount(page
);
604 list_for_each_entry(avc
, &anon_vma
->head
, same_anon_vma
) {
605 struct vm_area_struct
*vma
= avc
->vma
;
606 unsigned long address
= vma_address(page
, vma
);
607 if (address
== -EFAULT
)
610 * If we are reclaiming on behalf of a cgroup, skip
611 * counting on behalf of references from different
614 if (mem_cont
&& !mm_match_cgroup(vma
->vm_mm
, mem_cont
))
616 referenced
+= page_referenced_one(page
, vma
, address
,
617 &mapcount
, vm_flags
);
622 page_unlock_anon_vma(anon_vma
);
627 * page_referenced_file - referenced check for object-based rmap
628 * @page: the page we're checking references on.
629 * @mem_cont: target memory controller
630 * @vm_flags: collect encountered vma->vm_flags who actually referenced the page
632 * For an object-based mapped page, find all the places it is mapped and
633 * check/clear the referenced flag. This is done by following the page->mapping
634 * pointer, then walking the chain of vmas it holds. It returns the number
635 * of references it found.
637 * This function is only called from page_referenced for object-based pages.
639 static int page_referenced_file(struct page
*page
,
640 struct mem_cgroup
*mem_cont
,
641 unsigned long *vm_flags
)
643 unsigned int mapcount
;
644 struct address_space
*mapping
= page
->mapping
;
645 pgoff_t pgoff
= page
->index
<< (PAGE_CACHE_SHIFT
- PAGE_SHIFT
);
646 struct vm_area_struct
*vma
;
647 struct prio_tree_iter iter
;
651 * The caller's checks on page->mapping and !PageAnon have made
652 * sure that this is a file page: the check for page->mapping
653 * excludes the case just before it gets set on an anon page.
655 BUG_ON(PageAnon(page
));
658 * The page lock not only makes sure that page->mapping cannot
659 * suddenly be NULLified by truncation, it makes sure that the
660 * structure at mapping cannot be freed and reused yet,
661 * so we can safely take mapping->i_mmap_mutex.
663 BUG_ON(!PageLocked(page
));
665 mutex_lock(&mapping
->i_mmap_mutex
);
668 * i_mmap_mutex does not stabilize mapcount at all, but mapcount
669 * is more likely to be accurate if we note it after spinning.
671 mapcount
= page_mapcount(page
);
673 vma_prio_tree_foreach(vma
, &iter
, &mapping
->i_mmap
, pgoff
, pgoff
) {
674 unsigned long address
= vma_address(page
, vma
);
675 if (address
== -EFAULT
)
678 * If we are reclaiming on behalf of a cgroup, skip
679 * counting on behalf of references from different
682 if (mem_cont
&& !mm_match_cgroup(vma
->vm_mm
, mem_cont
))
684 referenced
+= page_referenced_one(page
, vma
, address
,
685 &mapcount
, vm_flags
);
690 mutex_unlock(&mapping
->i_mmap_mutex
);
695 * page_referenced - test if the page was referenced
696 * @page: the page to test
697 * @is_locked: caller holds lock on the page
698 * @mem_cont: target memory controller
699 * @vm_flags: collect encountered vma->vm_flags who actually referenced the page
701 * Quick test_and_clear_referenced for all mappings to a page,
702 * returns the number of ptes which referenced the page.
704 int page_referenced(struct page
*page
,
706 struct mem_cgroup
*mem_cont
,
707 unsigned long *vm_flags
)
713 if (page_mapped(page
) && page_rmapping(page
)) {
714 if (!is_locked
&& (!PageAnon(page
) || PageKsm(page
))) {
715 we_locked
= trylock_page(page
);
721 if (unlikely(PageKsm(page
)))
722 referenced
+= page_referenced_ksm(page
, mem_cont
,
724 else if (PageAnon(page
))
725 referenced
+= page_referenced_anon(page
, mem_cont
,
727 else if (page
->mapping
)
728 referenced
+= page_referenced_file(page
, mem_cont
,
734 if (page_test_and_clear_young(page_to_pfn(page
)))
740 static int page_mkclean_one(struct page
*page
, struct vm_area_struct
*vma
,
741 unsigned long address
)
743 struct mm_struct
*mm
= vma
->vm_mm
;
748 pte
= page_check_address(page
, mm
, address
, &ptl
, 1);
752 if (pte_dirty(*pte
) || pte_write(*pte
)) {
755 flush_cache_page(vma
, address
, pte_pfn(*pte
));
756 entry
= ptep_clear_flush_notify(vma
, address
, pte
);
757 entry
= pte_wrprotect(entry
);
758 entry
= pte_mkclean(entry
);
759 set_pte_at(mm
, address
, pte
, entry
);
763 pte_unmap_unlock(pte
, ptl
);
768 static int page_mkclean_file(struct address_space
*mapping
, struct page
*page
)
770 pgoff_t pgoff
= page
->index
<< (PAGE_CACHE_SHIFT
- PAGE_SHIFT
);
771 struct vm_area_struct
*vma
;
772 struct prio_tree_iter iter
;
775 BUG_ON(PageAnon(page
));
777 mutex_lock(&mapping
->i_mmap_mutex
);
778 vma_prio_tree_foreach(vma
, &iter
, &mapping
->i_mmap
, pgoff
, pgoff
) {
779 if (vma
->vm_flags
& VM_SHARED
) {
780 unsigned long address
= vma_address(page
, vma
);
781 if (address
== -EFAULT
)
783 ret
+= page_mkclean_one(page
, vma
, address
);
786 mutex_unlock(&mapping
->i_mmap_mutex
);
790 int page_mkclean(struct page
*page
)
794 BUG_ON(!PageLocked(page
));
796 if (page_mapped(page
)) {
797 struct address_space
*mapping
= page_mapping(page
);
799 ret
= page_mkclean_file(mapping
, page
);
800 if (page_test_and_clear_dirty(page_to_pfn(page
), 1))
807 EXPORT_SYMBOL_GPL(page_mkclean
);
810 * page_move_anon_rmap - move a page to our anon_vma
811 * @page: the page to move to our anon_vma
812 * @vma: the vma the page belongs to
813 * @address: the user virtual address mapped
815 * When a page belongs exclusively to one process after a COW event,
816 * that page can be moved into the anon_vma that belongs to just that
817 * process, so the rmap code will not search the parent or sibling
820 void page_move_anon_rmap(struct page
*page
,
821 struct vm_area_struct
*vma
, unsigned long address
)
823 struct anon_vma
*anon_vma
= vma
->anon_vma
;
825 VM_BUG_ON(!PageLocked(page
));
826 VM_BUG_ON(!anon_vma
);
827 VM_BUG_ON(page
->index
!= linear_page_index(vma
, address
));
829 anon_vma
= (void *) anon_vma
+ PAGE_MAPPING_ANON
;
830 page
->mapping
= (struct address_space
*) anon_vma
;
834 * __page_set_anon_rmap - set up new anonymous rmap
835 * @page: Page to add to rmap
836 * @vma: VM area to add page to.
837 * @address: User virtual address of the mapping
838 * @exclusive: the page is exclusively owned by the current process
840 static void __page_set_anon_rmap(struct page
*page
,
841 struct vm_area_struct
*vma
, unsigned long address
, int exclusive
)
843 struct anon_vma
*anon_vma
= vma
->anon_vma
;
851 * If the page isn't exclusively mapped into this vma,
852 * we must use the _oldest_ possible anon_vma for the
856 anon_vma
= anon_vma
->root
;
858 anon_vma
= (void *) anon_vma
+ PAGE_MAPPING_ANON
;
859 page
->mapping
= (struct address_space
*) anon_vma
;
860 page
->index
= linear_page_index(vma
, address
);
864 * __page_check_anon_rmap - sanity check anonymous rmap addition
865 * @page: the page to add the mapping to
866 * @vma: the vm area in which the mapping is added
867 * @address: the user virtual address mapped
869 static void __page_check_anon_rmap(struct page
*page
,
870 struct vm_area_struct
*vma
, unsigned long address
)
872 #ifdef CONFIG_DEBUG_VM
874 * The page's anon-rmap details (mapping and index) are guaranteed to
875 * be set up correctly at this point.
877 * We have exclusion against page_add_anon_rmap because the caller
878 * always holds the page locked, except if called from page_dup_rmap,
879 * in which case the page is already known to be setup.
881 * We have exclusion against page_add_new_anon_rmap because those pages
882 * are initially only visible via the pagetables, and the pte is locked
883 * over the call to page_add_new_anon_rmap.
885 BUG_ON(page_anon_vma(page
)->root
!= vma
->anon_vma
->root
);
886 BUG_ON(page
->index
!= linear_page_index(vma
, address
));
891 * page_add_anon_rmap - add pte mapping to an anonymous page
892 * @page: the page to add the mapping to
893 * @vma: the vm area in which the mapping is added
894 * @address: the user virtual address mapped
896 * The caller needs to hold the pte lock, and the page must be locked in
897 * the anon_vma case: to serialize mapping,index checking after setting,
898 * and to ensure that PageAnon is not being upgraded racily to PageKsm
899 * (but PageKsm is never downgraded to PageAnon).
901 void page_add_anon_rmap(struct page
*page
,
902 struct vm_area_struct
*vma
, unsigned long address
)
904 do_page_add_anon_rmap(page
, vma
, address
, 0);
908 * Special version of the above for do_swap_page, which often runs
909 * into pages that are exclusively owned by the current process.
910 * Everybody else should continue to use page_add_anon_rmap above.
912 void do_page_add_anon_rmap(struct page
*page
,
913 struct vm_area_struct
*vma
, unsigned long address
, int exclusive
)
915 int first
= atomic_inc_and_test(&page
->_mapcount
);
917 if (!PageTransHuge(page
))
918 __inc_zone_page_state(page
, NR_ANON_PAGES
);
920 __inc_zone_page_state(page
,
921 NR_ANON_TRANSPARENT_HUGEPAGES
);
923 if (unlikely(PageKsm(page
)))
926 VM_BUG_ON(!PageLocked(page
));
927 VM_BUG_ON(address
< vma
->vm_start
|| address
>= vma
->vm_end
);
929 __page_set_anon_rmap(page
, vma
, address
, exclusive
);
931 __page_check_anon_rmap(page
, vma
, address
);
935 * page_add_new_anon_rmap - add pte mapping to a new anonymous page
936 * @page: the page to add the mapping to
937 * @vma: the vm area in which the mapping is added
938 * @address: the user virtual address mapped
940 * Same as page_add_anon_rmap but must only be called on *new* pages.
941 * This means the inc-and-test can be bypassed.
942 * Page does not have to be locked.
944 void page_add_new_anon_rmap(struct page
*page
,
945 struct vm_area_struct
*vma
, unsigned long address
)
947 VM_BUG_ON(address
< vma
->vm_start
|| address
>= vma
->vm_end
);
948 SetPageSwapBacked(page
);
949 atomic_set(&page
->_mapcount
, 0); /* increment count (starts at -1) */
950 if (!PageTransHuge(page
))
951 __inc_zone_page_state(page
, NR_ANON_PAGES
);
953 __inc_zone_page_state(page
, NR_ANON_TRANSPARENT_HUGEPAGES
);
954 __page_set_anon_rmap(page
, vma
, address
, 1);
955 if (page_evictable(page
, vma
))
956 lru_cache_add_lru(page
, LRU_ACTIVE_ANON
);
958 add_page_to_unevictable_list(page
);
962 * page_add_file_rmap - add pte mapping to a file page
963 * @page: the page to add the mapping to
965 * The caller needs to hold the pte lock.
967 void page_add_file_rmap(struct page
*page
)
969 if (atomic_inc_and_test(&page
->_mapcount
)) {
970 __inc_zone_page_state(page
, NR_FILE_MAPPED
);
971 mem_cgroup_inc_page_stat(page
, MEMCG_NR_FILE_MAPPED
);
976 * page_remove_rmap - take down pte mapping from a page
977 * @page: page to remove mapping from
979 * The caller needs to hold the pte lock.
981 void page_remove_rmap(struct page
*page
)
983 /* page still mapped by someone else? */
984 if (!atomic_add_negative(-1, &page
->_mapcount
))
988 * Now that the last pte has gone, s390 must transfer dirty
989 * flag from storage key to struct page. We can usually skip
990 * this if the page is anon, so about to be freed; but perhaps
991 * not if it's in swapcache - there might be another pte slot
992 * containing the swap entry, but page not yet written to swap.
994 if ((!PageAnon(page
) || PageSwapCache(page
)) &&
995 page_test_and_clear_dirty(page_to_pfn(page
), 1))
996 set_page_dirty(page
);
998 * Hugepages are not counted in NR_ANON_PAGES nor NR_FILE_MAPPED
999 * and not charged by memcg for now.
1001 if (unlikely(PageHuge(page
)))
1003 if (PageAnon(page
)) {
1004 mem_cgroup_uncharge_page(page
);
1005 if (!PageTransHuge(page
))
1006 __dec_zone_page_state(page
, NR_ANON_PAGES
);
1008 __dec_zone_page_state(page
,
1009 NR_ANON_TRANSPARENT_HUGEPAGES
);
1011 __dec_zone_page_state(page
, NR_FILE_MAPPED
);
1012 mem_cgroup_dec_page_stat(page
, MEMCG_NR_FILE_MAPPED
);
1015 * It would be tidy to reset the PageAnon mapping here,
1016 * but that might overwrite a racing page_add_anon_rmap
1017 * which increments mapcount after us but sets mapping
1018 * before us: so leave the reset to free_hot_cold_page,
1019 * and remember that it's only reliable while mapped.
1020 * Leaving it set also helps swapoff to reinstate ptes
1021 * faster for those pages still in swapcache.
1026 * Subfunctions of try_to_unmap: try_to_unmap_one called
1027 * repeatedly from either try_to_unmap_anon or try_to_unmap_file.
1029 int try_to_unmap_one(struct page
*page
, struct vm_area_struct
*vma
,
1030 unsigned long address
, enum ttu_flags flags
)
1032 struct mm_struct
*mm
= vma
->vm_mm
;
1036 int ret
= SWAP_AGAIN
;
1038 pte
= page_check_address(page
, mm
, address
, &ptl
, 0);
1043 * If the page is mlock()d, we cannot swap it out.
1044 * If it's recently referenced (perhaps page_referenced
1045 * skipped over this mm) then we should reactivate it.
1047 if (!(flags
& TTU_IGNORE_MLOCK
)) {
1048 if (vma
->vm_flags
& VM_LOCKED
)
1051 if (TTU_ACTION(flags
) == TTU_MUNLOCK
)
1054 if (!(flags
& TTU_IGNORE_ACCESS
)) {
1055 if (ptep_clear_flush_young_notify(vma
, address
, pte
)) {
1061 /* Nuke the page table entry. */
1062 flush_cache_page(vma
, address
, page_to_pfn(page
));
1063 pteval
= ptep_clear_flush_notify(vma
, address
, pte
);
1065 /* Move the dirty bit to the physical page now the pte is gone. */
1066 if (pte_dirty(pteval
))
1067 set_page_dirty(page
);
1069 /* Update high watermark before we lower rss */
1070 update_hiwater_rss(mm
);
1072 if (PageHWPoison(page
) && !(flags
& TTU_IGNORE_HWPOISON
)) {
1074 dec_mm_counter(mm
, MM_ANONPAGES
);
1076 dec_mm_counter(mm
, MM_FILEPAGES
);
1077 set_pte_at(mm
, address
, pte
,
1078 swp_entry_to_pte(make_hwpoison_entry(page
)));
1079 } else if (PageAnon(page
)) {
1080 swp_entry_t entry
= { .val
= page_private(page
) };
1082 if (PageSwapCache(page
)) {
1084 * Store the swap location in the pte.
1085 * See handle_pte_fault() ...
1087 if (swap_duplicate(entry
) < 0) {
1088 set_pte_at(mm
, address
, pte
, pteval
);
1092 if (list_empty(&mm
->mmlist
)) {
1093 spin_lock(&mmlist_lock
);
1094 if (list_empty(&mm
->mmlist
))
1095 list_add(&mm
->mmlist
, &init_mm
.mmlist
);
1096 spin_unlock(&mmlist_lock
);
1098 dec_mm_counter(mm
, MM_ANONPAGES
);
1099 inc_mm_counter(mm
, MM_SWAPENTS
);
1100 } else if (PAGE_MIGRATION
) {
1102 * Store the pfn of the page in a special migration
1103 * pte. do_swap_page() will wait until the migration
1104 * pte is removed and then restart fault handling.
1106 BUG_ON(TTU_ACTION(flags
) != TTU_MIGRATION
);
1107 entry
= make_migration_entry(page
, pte_write(pteval
));
1109 set_pte_at(mm
, address
, pte
, swp_entry_to_pte(entry
));
1110 BUG_ON(pte_file(*pte
));
1111 } else if (PAGE_MIGRATION
&& (TTU_ACTION(flags
) == TTU_MIGRATION
)) {
1112 /* Establish migration entry for a file page */
1114 entry
= make_migration_entry(page
, pte_write(pteval
));
1115 set_pte_at(mm
, address
, pte
, swp_entry_to_pte(entry
));
1117 dec_mm_counter(mm
, MM_FILEPAGES
);
1119 page_remove_rmap(page
);
1120 page_cache_release(page
);
1123 pte_unmap_unlock(pte
, ptl
);
1128 pte_unmap_unlock(pte
, ptl
);
1132 * We need mmap_sem locking, Otherwise VM_LOCKED check makes
1133 * unstable result and race. Plus, We can't wait here because
1134 * we now hold anon_vma->lock or mapping->i_mmap_mutex.
1135 * if trylock failed, the page remain in evictable lru and later
1136 * vmscan could retry to move the page to unevictable lru if the
1137 * page is actually mlocked.
1139 if (down_read_trylock(&vma
->vm_mm
->mmap_sem
)) {
1140 if (vma
->vm_flags
& VM_LOCKED
) {
1141 mlock_vma_page(page
);
1144 up_read(&vma
->vm_mm
->mmap_sem
);
1150 * objrmap doesn't work for nonlinear VMAs because the assumption that
1151 * offset-into-file correlates with offset-into-virtual-addresses does not hold.
1152 * Consequently, given a particular page and its ->index, we cannot locate the
1153 * ptes which are mapping that page without an exhaustive linear search.
1155 * So what this code does is a mini "virtual scan" of each nonlinear VMA which
1156 * maps the file to which the target page belongs. The ->vm_private_data field
1157 * holds the current cursor into that scan. Successive searches will circulate
1158 * around the vma's virtual address space.
1160 * So as more replacement pressure is applied to the pages in a nonlinear VMA,
1161 * more scanning pressure is placed against them as well. Eventually pages
1162 * will become fully unmapped and are eligible for eviction.
1164 * For very sparsely populated VMAs this is a little inefficient - chances are
1165 * there there won't be many ptes located within the scan cluster. In this case
1166 * maybe we could scan further - to the end of the pte page, perhaps.
1168 * Mlocked pages: check VM_LOCKED under mmap_sem held for read, if we can
1169 * acquire it without blocking. If vma locked, mlock the pages in the cluster,
1170 * rather than unmapping them. If we encounter the "check_page" that vmscan is
1171 * trying to unmap, return SWAP_MLOCK, else default SWAP_AGAIN.
1173 #define CLUSTER_SIZE min(32*PAGE_SIZE, PMD_SIZE)
1174 #define CLUSTER_MASK (~(CLUSTER_SIZE - 1))
1176 static int try_to_unmap_cluster(unsigned long cursor
, unsigned int *mapcount
,
1177 struct vm_area_struct
*vma
, struct page
*check_page
)
1179 struct mm_struct
*mm
= vma
->vm_mm
;
1187 unsigned long address
;
1189 int ret
= SWAP_AGAIN
;
1192 address
= (vma
->vm_start
+ cursor
) & CLUSTER_MASK
;
1193 end
= address
+ CLUSTER_SIZE
;
1194 if (address
< vma
->vm_start
)
1195 address
= vma
->vm_start
;
1196 if (end
> vma
->vm_end
)
1199 pgd
= pgd_offset(mm
, address
);
1200 if (!pgd_present(*pgd
))
1203 pud
= pud_offset(pgd
, address
);
1204 if (!pud_present(*pud
))
1207 pmd
= pmd_offset(pud
, address
);
1208 if (!pmd_present(*pmd
))
1212 * If we can acquire the mmap_sem for read, and vma is VM_LOCKED,
1213 * keep the sem while scanning the cluster for mlocking pages.
1215 if (down_read_trylock(&vma
->vm_mm
->mmap_sem
)) {
1216 locked_vma
= (vma
->vm_flags
& VM_LOCKED
);
1218 up_read(&vma
->vm_mm
->mmap_sem
); /* don't need it */
1221 pte
= pte_offset_map_lock(mm
, pmd
, address
, &ptl
);
1223 /* Update high watermark before we lower rss */
1224 update_hiwater_rss(mm
);
1226 for (; address
< end
; pte
++, address
+= PAGE_SIZE
) {
1227 if (!pte_present(*pte
))
1229 page
= vm_normal_page(vma
, address
, *pte
);
1230 BUG_ON(!page
|| PageAnon(page
));
1233 mlock_vma_page(page
); /* no-op if already mlocked */
1234 if (page
== check_page
)
1236 continue; /* don't unmap */
1239 if (ptep_clear_flush_young_notify(vma
, address
, pte
))
1242 /* Nuke the page table entry. */
1243 flush_cache_page(vma
, address
, pte_pfn(*pte
));
1244 pteval
= ptep_clear_flush_notify(vma
, address
, pte
);
1246 /* If nonlinear, store the file page offset in the pte. */
1247 if (page
->index
!= linear_page_index(vma
, address
))
1248 set_pte_at(mm
, address
, pte
, pgoff_to_pte(page
->index
));
1250 /* Move the dirty bit to the physical page now the pte is gone. */
1251 if (pte_dirty(pteval
))
1252 set_page_dirty(page
);
1254 page_remove_rmap(page
);
1255 page_cache_release(page
);
1256 dec_mm_counter(mm
, MM_FILEPAGES
);
1259 pte_unmap_unlock(pte
- 1, ptl
);
1261 up_read(&vma
->vm_mm
->mmap_sem
);
1265 bool is_vma_temporary_stack(struct vm_area_struct
*vma
)
1267 int maybe_stack
= vma
->vm_flags
& (VM_GROWSDOWN
| VM_GROWSUP
);
1272 if ((vma
->vm_flags
& VM_STACK_INCOMPLETE_SETUP
) ==
1273 VM_STACK_INCOMPLETE_SETUP
)
1280 * try_to_unmap_anon - unmap or unlock anonymous page using the object-based
1282 * @page: the page to unmap/unlock
1283 * @flags: action and flags
1285 * Find all the mappings of a page using the mapping pointer and the vma chains
1286 * contained in the anon_vma struct it points to.
1288 * This function is only called from try_to_unmap/try_to_munlock for
1290 * When called from try_to_munlock(), the mmap_sem of the mm containing the vma
1291 * where the page was found will be held for write. So, we won't recheck
1292 * vm_flags for that VMA. That should be OK, because that vma shouldn't be
1295 static int try_to_unmap_anon(struct page
*page
, enum ttu_flags flags
)
1297 struct anon_vma
*anon_vma
;
1298 struct anon_vma_chain
*avc
;
1299 int ret
= SWAP_AGAIN
;
1301 anon_vma
= page_lock_anon_vma(page
);
1305 list_for_each_entry(avc
, &anon_vma
->head
, same_anon_vma
) {
1306 struct vm_area_struct
*vma
= avc
->vma
;
1307 unsigned long address
;
1310 * During exec, a temporary VMA is setup and later moved.
1311 * The VMA is moved under the anon_vma lock but not the
1312 * page tables leading to a race where migration cannot
1313 * find the migration ptes. Rather than increasing the
1314 * locking requirements of exec(), migration skips
1315 * temporary VMAs until after exec() completes.
1317 if (PAGE_MIGRATION
&& (flags
& TTU_MIGRATION
) &&
1318 is_vma_temporary_stack(vma
))
1321 address
= vma_address(page
, vma
);
1322 if (address
== -EFAULT
)
1324 ret
= try_to_unmap_one(page
, vma
, address
, flags
);
1325 if (ret
!= SWAP_AGAIN
|| !page_mapped(page
))
1329 page_unlock_anon_vma(anon_vma
);
1334 * try_to_unmap_file - unmap/unlock file page using the object-based rmap method
1335 * @page: the page to unmap/unlock
1336 * @flags: action and flags
1338 * Find all the mappings of a page using the mapping pointer and the vma chains
1339 * contained in the address_space struct it points to.
1341 * This function is only called from try_to_unmap/try_to_munlock for
1342 * object-based pages.
1343 * When called from try_to_munlock(), the mmap_sem of the mm containing the vma
1344 * where the page was found will be held for write. So, we won't recheck
1345 * vm_flags for that VMA. That should be OK, because that vma shouldn't be
1348 static int try_to_unmap_file(struct page
*page
, enum ttu_flags flags
)
1350 struct address_space
*mapping
= page
->mapping
;
1351 pgoff_t pgoff
= page
->index
<< (PAGE_CACHE_SHIFT
- PAGE_SHIFT
);
1352 struct vm_area_struct
*vma
;
1353 struct prio_tree_iter iter
;
1354 int ret
= SWAP_AGAIN
;
1355 unsigned long cursor
;
1356 unsigned long max_nl_cursor
= 0;
1357 unsigned long max_nl_size
= 0;
1358 unsigned int mapcount
;
1360 mutex_lock(&mapping
->i_mmap_mutex
);
1361 vma_prio_tree_foreach(vma
, &iter
, &mapping
->i_mmap
, pgoff
, pgoff
) {
1362 unsigned long address
= vma_address(page
, vma
);
1363 if (address
== -EFAULT
)
1365 ret
= try_to_unmap_one(page
, vma
, address
, flags
);
1366 if (ret
!= SWAP_AGAIN
|| !page_mapped(page
))
1370 if (list_empty(&mapping
->i_mmap_nonlinear
))
1374 * We don't bother to try to find the munlocked page in nonlinears.
1375 * It's costly. Instead, later, page reclaim logic may call
1376 * try_to_unmap(TTU_MUNLOCK) and recover PG_mlocked lazily.
1378 if (TTU_ACTION(flags
) == TTU_MUNLOCK
)
1381 list_for_each_entry(vma
, &mapping
->i_mmap_nonlinear
,
1382 shared
.vm_set
.list
) {
1383 cursor
= (unsigned long) vma
->vm_private_data
;
1384 if (cursor
> max_nl_cursor
)
1385 max_nl_cursor
= cursor
;
1386 cursor
= vma
->vm_end
- vma
->vm_start
;
1387 if (cursor
> max_nl_size
)
1388 max_nl_size
= cursor
;
1391 if (max_nl_size
== 0) { /* all nonlinears locked or reserved ? */
1397 * We don't try to search for this page in the nonlinear vmas,
1398 * and page_referenced wouldn't have found it anyway. Instead
1399 * just walk the nonlinear vmas trying to age and unmap some.
1400 * The mapcount of the page we came in with is irrelevant,
1401 * but even so use it as a guide to how hard we should try?
1403 mapcount
= page_mapcount(page
);
1408 max_nl_size
= (max_nl_size
+ CLUSTER_SIZE
- 1) & CLUSTER_MASK
;
1409 if (max_nl_cursor
== 0)
1410 max_nl_cursor
= CLUSTER_SIZE
;
1413 list_for_each_entry(vma
, &mapping
->i_mmap_nonlinear
,
1414 shared
.vm_set
.list
) {
1415 cursor
= (unsigned long) vma
->vm_private_data
;
1416 while ( cursor
< max_nl_cursor
&&
1417 cursor
< vma
->vm_end
- vma
->vm_start
) {
1418 if (try_to_unmap_cluster(cursor
, &mapcount
,
1419 vma
, page
) == SWAP_MLOCK
)
1421 cursor
+= CLUSTER_SIZE
;
1422 vma
->vm_private_data
= (void *) cursor
;
1423 if ((int)mapcount
<= 0)
1426 vma
->vm_private_data
= (void *) max_nl_cursor
;
1429 max_nl_cursor
+= CLUSTER_SIZE
;
1430 } while (max_nl_cursor
<= max_nl_size
);
1433 * Don't loop forever (perhaps all the remaining pages are
1434 * in locked vmas). Reset cursor on all unreserved nonlinear
1435 * vmas, now forgetting on which ones it had fallen behind.
1437 list_for_each_entry(vma
, &mapping
->i_mmap_nonlinear
, shared
.vm_set
.list
)
1438 vma
->vm_private_data
= NULL
;
1440 mutex_unlock(&mapping
->i_mmap_mutex
);
1445 * try_to_unmap - try to remove all page table mappings to a page
1446 * @page: the page to get unmapped
1447 * @flags: action and flags
1449 * Tries to remove all the page table entries which are mapping this
1450 * page, used in the pageout path. Caller must hold the page lock.
1451 * Return values are:
1453 * SWAP_SUCCESS - we succeeded in removing all mappings
1454 * SWAP_AGAIN - we missed a mapping, try again later
1455 * SWAP_FAIL - the page is unswappable
1456 * SWAP_MLOCK - page is mlocked.
1458 int try_to_unmap(struct page
*page
, enum ttu_flags flags
)
1462 BUG_ON(!PageLocked(page
));
1463 VM_BUG_ON(!PageHuge(page
) && PageTransHuge(page
));
1465 if (unlikely(PageKsm(page
)))
1466 ret
= try_to_unmap_ksm(page
, flags
);
1467 else if (PageAnon(page
))
1468 ret
= try_to_unmap_anon(page
, flags
);
1470 ret
= try_to_unmap_file(page
, flags
);
1471 if (ret
!= SWAP_MLOCK
&& !page_mapped(page
))
1477 * try_to_munlock - try to munlock a page
1478 * @page: the page to be munlocked
1480 * Called from munlock code. Checks all of the VMAs mapping the page
1481 * to make sure nobody else has this page mlocked. The page will be
1482 * returned with PG_mlocked cleared if no other vmas have it mlocked.
1484 * Return values are:
1486 * SWAP_AGAIN - no vma is holding page mlocked, or,
1487 * SWAP_AGAIN - page mapped in mlocked vma -- couldn't acquire mmap sem
1488 * SWAP_FAIL - page cannot be located at present
1489 * SWAP_MLOCK - page is now mlocked.
1491 int try_to_munlock(struct page
*page
)
1493 VM_BUG_ON(!PageLocked(page
) || PageLRU(page
));
1495 if (unlikely(PageKsm(page
)))
1496 return try_to_unmap_ksm(page
, TTU_MUNLOCK
);
1497 else if (PageAnon(page
))
1498 return try_to_unmap_anon(page
, TTU_MUNLOCK
);
1500 return try_to_unmap_file(page
, TTU_MUNLOCK
);
1503 void __put_anon_vma(struct anon_vma
*anon_vma
)
1505 struct anon_vma
*root
= anon_vma
->root
;
1507 if (root
!= anon_vma
&& atomic_dec_and_test(&root
->refcount
))
1508 anon_vma_free(root
);
1510 anon_vma_free(anon_vma
);
1513 #ifdef CONFIG_MIGRATION
1515 * rmap_walk() and its helpers rmap_walk_anon() and rmap_walk_file():
1516 * Called by migrate.c to remove migration ptes, but might be used more later.
1518 static int rmap_walk_anon(struct page
*page
, int (*rmap_one
)(struct page
*,
1519 struct vm_area_struct
*, unsigned long, void *), void *arg
)
1521 struct anon_vma
*anon_vma
;
1522 struct anon_vma_chain
*avc
;
1523 int ret
= SWAP_AGAIN
;
1526 * Note: remove_migration_ptes() cannot use page_lock_anon_vma()
1527 * because that depends on page_mapped(); but not all its usages
1528 * are holding mmap_sem. Users without mmap_sem are required to
1529 * take a reference count to prevent the anon_vma disappearing
1531 anon_vma
= page_anon_vma(page
);
1534 anon_vma_lock(anon_vma
);
1535 list_for_each_entry(avc
, &anon_vma
->head
, same_anon_vma
) {
1536 struct vm_area_struct
*vma
= avc
->vma
;
1537 unsigned long address
= vma_address(page
, vma
);
1538 if (address
== -EFAULT
)
1540 ret
= rmap_one(page
, vma
, address
, arg
);
1541 if (ret
!= SWAP_AGAIN
)
1544 anon_vma_unlock(anon_vma
);
1548 static int rmap_walk_file(struct page
*page
, int (*rmap_one
)(struct page
*,
1549 struct vm_area_struct
*, unsigned long, void *), void *arg
)
1551 struct address_space
*mapping
= page
->mapping
;
1552 pgoff_t pgoff
= page
->index
<< (PAGE_CACHE_SHIFT
- PAGE_SHIFT
);
1553 struct vm_area_struct
*vma
;
1554 struct prio_tree_iter iter
;
1555 int ret
= SWAP_AGAIN
;
1559 mutex_lock(&mapping
->i_mmap_mutex
);
1560 vma_prio_tree_foreach(vma
, &iter
, &mapping
->i_mmap
, pgoff
, pgoff
) {
1561 unsigned long address
= vma_address(page
, vma
);
1562 if (address
== -EFAULT
)
1564 ret
= rmap_one(page
, vma
, address
, arg
);
1565 if (ret
!= SWAP_AGAIN
)
1569 * No nonlinear handling: being always shared, nonlinear vmas
1570 * never contain migration ptes. Decide what to do about this
1571 * limitation to linear when we need rmap_walk() on nonlinear.
1573 mutex_unlock(&mapping
->i_mmap_mutex
);
1577 int rmap_walk(struct page
*page
, int (*rmap_one
)(struct page
*,
1578 struct vm_area_struct
*, unsigned long, void *), void *arg
)
1580 VM_BUG_ON(!PageLocked(page
));
1582 if (unlikely(PageKsm(page
)))
1583 return rmap_walk_ksm(page
, rmap_one
, arg
);
1584 else if (PageAnon(page
))
1585 return rmap_walk_anon(page
, rmap_one
, arg
);
1587 return rmap_walk_file(page
, rmap_one
, arg
);
1589 #endif /* CONFIG_MIGRATION */
1591 #ifdef CONFIG_HUGETLB_PAGE
1593 * The following three functions are for anonymous (private mapped) hugepages.
1594 * Unlike common anonymous pages, anonymous hugepages have no accounting code
1595 * and no lru code, because we handle hugepages differently from common pages.
1597 static void __hugepage_set_anon_rmap(struct page
*page
,
1598 struct vm_area_struct
*vma
, unsigned long address
, int exclusive
)
1600 struct anon_vma
*anon_vma
= vma
->anon_vma
;
1607 anon_vma
= anon_vma
->root
;
1609 anon_vma
= (void *) anon_vma
+ PAGE_MAPPING_ANON
;
1610 page
->mapping
= (struct address_space
*) anon_vma
;
1611 page
->index
= linear_page_index(vma
, address
);
1614 void hugepage_add_anon_rmap(struct page
*page
,
1615 struct vm_area_struct
*vma
, unsigned long address
)
1617 struct anon_vma
*anon_vma
= vma
->anon_vma
;
1620 BUG_ON(!PageLocked(page
));
1622 BUG_ON(address
< vma
->vm_start
|| address
>= vma
->vm_end
);
1623 first
= atomic_inc_and_test(&page
->_mapcount
);
1625 __hugepage_set_anon_rmap(page
, vma
, address
, 0);
1628 void hugepage_add_new_anon_rmap(struct page
*page
,
1629 struct vm_area_struct
*vma
, unsigned long address
)
1631 BUG_ON(address
< vma
->vm_start
|| address
>= vma
->vm_end
);
1632 atomic_set(&page
->_mapcount
, 0);
1633 __hugepage_set_anon_rmap(page
, vma
, address
, 1);
1635 #endif /* CONFIG_HUGETLB_PAGE */