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)
25 * page->flags PG_locked (lock_page)
26 * mapping->i_mmap_mutex
28 * mm->page_table_lock or pte_lock
29 * zone->lru_lock (in mark_page_accessed, isolate_lru_page)
30 * swap_lock (in swap_duplicate, swap_info_get)
31 * mmlist_lock (in mmput, drain_mmlist and others)
32 * mapping->private_lock (in __set_page_dirty_buffers)
33 * inode->i_lock (in set_page_dirty's __mark_inode_dirty)
34 * bdi.wb->list_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 bdi.wb->list_lock in __sync_single_inode)
40 * anon_vma->rwsem,mapping->i_mutex (memory_failure, collect_procs_anon)
46 #include <linux/pagemap.h>
47 #include <linux/swap.h>
48 #include <linux/swapops.h>
49 #include <linux/slab.h>
50 #include <linux/init.h>
51 #include <linux/ksm.h>
52 #include <linux/rmap.h>
53 #include <linux/rcupdate.h>
54 #include <linux/export.h>
55 #include <linux/memcontrol.h>
56 #include <linux/mmu_notifier.h>
57 #include <linux/migrate.h>
58 #include <linux/hugetlb.h>
59 #include <linux/backing-dev.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 struct anon_vma *anon_vma;
72 anon_vma = kmem_cache_alloc(anon_vma_cachep, GFP_KERNEL);
74 atomic_set(&anon_vma->refcount, 1);
76 * Initialise the anon_vma root to point to itself. If called
77 * from fork, the root will be reset to the parents anon_vma.
79 anon_vma->root = anon_vma;
85 static inline void anon_vma_free(struct anon_vma *anon_vma)
87 VM_BUG_ON(atomic_read(&anon_vma->refcount));
90 * Synchronize against page_lock_anon_vma_read() such that
91 * we can safely hold the lock without the anon_vma getting
94 * Relies on the full mb implied by the atomic_dec_and_test() from
95 * put_anon_vma() against the acquire barrier implied by
96 * down_read_trylock() from page_lock_anon_vma_read(). This orders:
98 * page_lock_anon_vma_read() VS put_anon_vma()
99 * down_read_trylock() atomic_dec_and_test()
101 * atomic_read() rwsem_is_locked()
103 * LOCK should suffice since the actual taking of the lock must
104 * happen _before_ what follows.
106 if (rwsem_is_locked(&anon_vma->root->rwsem)) {
107 anon_vma_lock_write(anon_vma);
108 anon_vma_unlock_write(anon_vma);
111 kmem_cache_free(anon_vma_cachep, anon_vma);
114 static inline struct anon_vma_chain *anon_vma_chain_alloc(gfp_t gfp)
116 return kmem_cache_alloc(anon_vma_chain_cachep, gfp);
119 static void anon_vma_chain_free(struct anon_vma_chain *anon_vma_chain)
121 kmem_cache_free(anon_vma_chain_cachep, anon_vma_chain);
124 static void anon_vma_chain_link(struct vm_area_struct *vma,
125 struct anon_vma_chain *avc,
126 struct anon_vma *anon_vma)
129 avc->anon_vma = anon_vma;
130 list_add(&avc->same_vma, &vma->anon_vma_chain);
131 anon_vma_interval_tree_insert(avc, &anon_vma->rb_root);
135 * anon_vma_prepare - attach an anon_vma to a memory region
136 * @vma: the memory region in question
138 * This makes sure the memory mapping described by 'vma' has
139 * an 'anon_vma' attached to it, so that we can associate the
140 * anonymous pages mapped into it with that anon_vma.
142 * The common case will be that we already have one, but if
143 * not we either need to find an adjacent mapping that we
144 * can re-use the anon_vma from (very common when the only
145 * reason for splitting a vma has been mprotect()), or we
146 * allocate a new one.
148 * Anon-vma allocations are very subtle, because we may have
149 * optimistically looked up an anon_vma in page_lock_anon_vma_read()
150 * and that may actually touch the spinlock even in the newly
151 * allocated vma (it depends on RCU to make sure that the
152 * anon_vma isn't actually destroyed).
154 * As a result, we need to do proper anon_vma locking even
155 * for the new allocation. At the same time, we do not want
156 * to do any locking for the common case of already having
159 * This must be called with the mmap_sem held for reading.
161 int anon_vma_prepare(struct vm_area_struct *vma)
163 struct anon_vma *anon_vma = vma->anon_vma;
164 struct anon_vma_chain *avc;
167 if (unlikely(!anon_vma)) {
168 struct mm_struct *mm = vma->vm_mm;
169 struct anon_vma *allocated;
171 avc = anon_vma_chain_alloc(GFP_KERNEL);
175 anon_vma = find_mergeable_anon_vma(vma);
178 anon_vma = anon_vma_alloc();
179 if (unlikely(!anon_vma))
180 goto out_enomem_free_avc;
181 allocated = anon_vma;
184 anon_vma_lock_write(anon_vma);
185 /* page_table_lock to protect against threads */
186 spin_lock(&mm->page_table_lock);
187 if (likely(!vma->anon_vma)) {
188 vma->anon_vma = anon_vma;
189 anon_vma_chain_link(vma, avc, anon_vma);
193 spin_unlock(&mm->page_table_lock);
194 anon_vma_unlock_write(anon_vma);
196 if (unlikely(allocated))
197 put_anon_vma(allocated);
199 anon_vma_chain_free(avc);
204 anon_vma_chain_free(avc);
210 * This is a useful helper function for locking the anon_vma root as
211 * we traverse the vma->anon_vma_chain, looping over anon_vma's that
214 * Such anon_vma's should have the same root, so you'd expect to see
215 * just a single mutex_lock for the whole traversal.
217 static inline struct anon_vma *lock_anon_vma_root(struct anon_vma *root, struct anon_vma *anon_vma)
219 struct anon_vma *new_root = anon_vma->root;
220 if (new_root != root) {
221 if (WARN_ON_ONCE(root))
222 up_write(&root->rwsem);
224 down_write(&root->rwsem);
229 static inline void unlock_anon_vma_root(struct anon_vma *root)
232 up_write(&root->rwsem);
236 * Attach the anon_vmas from src to dst.
237 * Returns 0 on success, -ENOMEM on failure.
239 int anon_vma_clone(struct vm_area_struct *dst, struct vm_area_struct *src)
241 struct anon_vma_chain *avc, *pavc;
242 struct anon_vma *root = NULL;
244 list_for_each_entry_reverse(pavc, &src->anon_vma_chain, same_vma) {
245 struct anon_vma *anon_vma;
247 avc = anon_vma_chain_alloc(GFP_NOWAIT | __GFP_NOWARN);
248 if (unlikely(!avc)) {
249 unlock_anon_vma_root(root);
251 avc = anon_vma_chain_alloc(GFP_KERNEL);
255 anon_vma = pavc->anon_vma;
256 root = lock_anon_vma_root(root, anon_vma);
257 anon_vma_chain_link(dst, avc, anon_vma);
259 unlock_anon_vma_root(root);
263 unlink_anon_vmas(dst);
268 * Attach vma to its own anon_vma, as well as to the anon_vmas that
269 * the corresponding VMA in the parent process is attached to.
270 * Returns 0 on success, non-zero on failure.
272 int anon_vma_fork(struct vm_area_struct *vma, struct vm_area_struct *pvma)
274 struct anon_vma_chain *avc;
275 struct anon_vma *anon_vma;
277 /* Don't bother if the parent process has no anon_vma here. */
282 * First, attach the new VMA to the parent VMA's anon_vmas,
283 * so rmap can find non-COWed pages in child processes.
285 if (anon_vma_clone(vma, pvma))
288 /* Then add our own anon_vma. */
289 anon_vma = anon_vma_alloc();
292 avc = anon_vma_chain_alloc(GFP_KERNEL);
294 goto out_error_free_anon_vma;
297 * The root anon_vma's spinlock is the lock actually used when we
298 * lock any of the anon_vmas in this anon_vma tree.
300 anon_vma->root = pvma->anon_vma->root;
302 * With refcounts, an anon_vma can stay around longer than the
303 * process it belongs to. The root anon_vma needs to be pinned until
304 * this anon_vma is freed, because the lock lives in the root.
306 get_anon_vma(anon_vma->root);
307 /* Mark this anon_vma as the one where our new (COWed) pages go. */
308 vma->anon_vma = anon_vma;
309 anon_vma_lock_write(anon_vma);
310 anon_vma_chain_link(vma, avc, anon_vma);
311 anon_vma_unlock_write(anon_vma);
315 out_error_free_anon_vma:
316 put_anon_vma(anon_vma);
318 unlink_anon_vmas(vma);
322 void unlink_anon_vmas(struct vm_area_struct *vma)
324 struct anon_vma_chain *avc, *next;
325 struct anon_vma *root = NULL;
328 * Unlink each anon_vma chained to the VMA. This list is ordered
329 * from newest to oldest, ensuring the root anon_vma gets freed last.
331 list_for_each_entry_safe(avc, next, &vma->anon_vma_chain, same_vma) {
332 struct anon_vma *anon_vma = avc->anon_vma;
334 root = lock_anon_vma_root(root, anon_vma);
335 anon_vma_interval_tree_remove(avc, &anon_vma->rb_root);
338 * Leave empty anon_vmas on the list - we'll need
339 * to free them outside the lock.
341 if (RB_EMPTY_ROOT(&anon_vma->rb_root))
344 list_del(&avc->same_vma);
345 anon_vma_chain_free(avc);
347 unlock_anon_vma_root(root);
350 * Iterate the list once more, it now only contains empty and unlinked
351 * anon_vmas, destroy them. Could not do before due to __put_anon_vma()
352 * needing to write-acquire the anon_vma->root->rwsem.
354 list_for_each_entry_safe(avc, next, &vma->anon_vma_chain, same_vma) {
355 struct anon_vma *anon_vma = avc->anon_vma;
357 put_anon_vma(anon_vma);
359 list_del(&avc->same_vma);
360 anon_vma_chain_free(avc);
364 static void anon_vma_ctor(void *data)
366 struct anon_vma *anon_vma = data;
368 init_rwsem(&anon_vma->rwsem);
369 atomic_set(&anon_vma->refcount, 0);
370 anon_vma->rb_root = RB_ROOT;
373 void __init anon_vma_init(void)
375 anon_vma_cachep = kmem_cache_create("anon_vma", sizeof(struct anon_vma),
376 0, SLAB_DESTROY_BY_RCU|SLAB_PANIC, anon_vma_ctor);
377 anon_vma_chain_cachep = KMEM_CACHE(anon_vma_chain, SLAB_PANIC);
381 * Getting a lock on a stable anon_vma from a page off the LRU is tricky!
383 * Since there is no serialization what so ever against page_remove_rmap()
384 * the best this function can do is return a locked anon_vma that might
385 * have been relevant to this page.
387 * The page might have been remapped to a different anon_vma or the anon_vma
388 * returned may already be freed (and even reused).
390 * In case it was remapped to a different anon_vma, the new anon_vma will be a
391 * child of the old anon_vma, and the anon_vma lifetime rules will therefore
392 * ensure that any anon_vma obtained from the page will still be valid for as
393 * long as we observe page_mapped() [ hence all those page_mapped() tests ].
395 * All users of this function must be very careful when walking the anon_vma
396 * chain and verify that the page in question is indeed mapped in it
397 * [ something equivalent to page_mapped_in_vma() ].
399 * Since anon_vma's slab is DESTROY_BY_RCU and we know from page_remove_rmap()
400 * that the anon_vma pointer from page->mapping is valid if there is a
401 * mapcount, we can dereference the anon_vma after observing those.
403 struct anon_vma *page_get_anon_vma(struct page *page)
405 struct anon_vma *anon_vma = NULL;
406 unsigned long anon_mapping;
409 anon_mapping = (unsigned long) ACCESS_ONCE(page->mapping);
410 if ((anon_mapping & PAGE_MAPPING_FLAGS) != PAGE_MAPPING_ANON)
412 if (!page_mapped(page))
415 anon_vma = (struct anon_vma *) (anon_mapping - PAGE_MAPPING_ANON);
416 if (!atomic_inc_not_zero(&anon_vma->refcount)) {
422 * If this page is still mapped, then its anon_vma cannot have been
423 * freed. But if it has been unmapped, we have no security against the
424 * anon_vma structure being freed and reused (for another anon_vma:
425 * SLAB_DESTROY_BY_RCU guarantees that - so the atomic_inc_not_zero()
426 * above cannot corrupt).
428 if (!page_mapped(page)) {
429 put_anon_vma(anon_vma);
439 * Similar to page_get_anon_vma() except it locks the anon_vma.
441 * Its a little more complex as it tries to keep the fast path to a single
442 * atomic op -- the trylock. If we fail the trylock, we fall back to getting a
443 * reference like with page_get_anon_vma() and then block on the mutex.
445 struct anon_vma *page_lock_anon_vma_read(struct page *page)
447 struct anon_vma *anon_vma = NULL;
448 struct anon_vma *root_anon_vma;
449 unsigned long anon_mapping;
452 anon_mapping = (unsigned long) ACCESS_ONCE(page->mapping);
453 if ((anon_mapping & PAGE_MAPPING_FLAGS) != PAGE_MAPPING_ANON)
455 if (!page_mapped(page))
458 anon_vma = (struct anon_vma *) (anon_mapping - PAGE_MAPPING_ANON);
459 root_anon_vma = ACCESS_ONCE(anon_vma->root);
460 if (down_read_trylock(&root_anon_vma->rwsem)) {
462 * If the page is still mapped, then this anon_vma is still
463 * its anon_vma, and holding the mutex ensures that it will
464 * not go away, see anon_vma_free().
466 if (!page_mapped(page)) {
467 up_read(&root_anon_vma->rwsem);
473 /* trylock failed, we got to sleep */
474 if (!atomic_inc_not_zero(&anon_vma->refcount)) {
479 if (!page_mapped(page)) {
480 put_anon_vma(anon_vma);
485 /* we pinned the anon_vma, its safe to sleep */
487 anon_vma_lock_read(anon_vma);
489 if (atomic_dec_and_test(&anon_vma->refcount)) {
491 * Oops, we held the last refcount, release the lock
492 * and bail -- can't simply use put_anon_vma() because
493 * we'll deadlock on the anon_vma_lock_write() recursion.
495 anon_vma_unlock_read(anon_vma);
496 __put_anon_vma(anon_vma);
507 void page_unlock_anon_vma_read(struct anon_vma *anon_vma)
509 anon_vma_unlock_read(anon_vma);
513 * At what user virtual address is page expected in @vma?
515 static inline unsigned long
516 __vma_address(struct page *page, struct vm_area_struct *vma)
518 pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT);
520 if (unlikely(is_vm_hugetlb_page(vma)))
521 pgoff = page->index << huge_page_order(page_hstate(page));
523 return vma->vm_start + ((pgoff - vma->vm_pgoff) << PAGE_SHIFT);
527 vma_address(struct page *page, struct vm_area_struct *vma)
529 unsigned long address = __vma_address(page, vma);
531 /* page should be within @vma mapping range */
532 VM_BUG_ON(address < vma->vm_start || address >= vma->vm_end);
538 * At what user virtual address is page expected in vma?
539 * Caller should check the page is actually part of the vma.
541 unsigned long page_address_in_vma(struct page *page, struct vm_area_struct *vma)
543 unsigned long address;
544 if (PageAnon(page)) {
545 struct anon_vma *page__anon_vma = page_anon_vma(page);
547 * Note: swapoff's unuse_vma() is more efficient with this
548 * check, and needs it to match anon_vma when KSM is active.
550 if (!vma->anon_vma || !page__anon_vma ||
551 vma->anon_vma->root != page__anon_vma->root)
553 } else if (page->mapping && !(vma->vm_flags & VM_NONLINEAR)) {
555 vma->vm_file->f_mapping != page->mapping)
559 address = __vma_address(page, vma);
560 if (unlikely(address < vma->vm_start || address >= vma->vm_end))
565 pmd_t *mm_find_pmd(struct mm_struct *mm, unsigned long address)
571 pgd = pgd_offset(mm, address);
572 if (!pgd_present(*pgd))
575 pud = pud_offset(pgd, address);
576 if (!pud_present(*pud))
579 pmd = pmd_offset(pud, address);
580 if (!pmd_present(*pmd))
587 * Check that @page is mapped at @address into @mm.
589 * If @sync is false, page_check_address may perform a racy check to avoid
590 * the page table lock when the pte is not present (helpful when reclaiming
591 * highly shared pages).
593 * On success returns with pte mapped and locked.
595 pte_t *__page_check_address(struct page *page, struct mm_struct *mm,
596 unsigned long address, spinlock_t **ptlp, int sync)
602 if (unlikely(PageHuge(page))) {
603 /* when pud is not present, pte will be NULL */
604 pte = huge_pte_offset(mm, address);
608 ptl = &mm->page_table_lock;
612 pmd = mm_find_pmd(mm, address);
616 if (pmd_trans_huge(*pmd))
619 pte = pte_offset_map(pmd, address);
620 /* Make a quick check before getting the lock */
621 if (!sync && !pte_present(*pte)) {
626 ptl = pte_lockptr(mm, pmd);
629 if (pte_present(*pte) && page_to_pfn(page) == pte_pfn(*pte)) {
633 pte_unmap_unlock(pte, ptl);
638 * page_mapped_in_vma - check whether a page is really mapped in a VMA
639 * @page: the page to test
640 * @vma: the VMA to test
642 * Returns 1 if the page is mapped into the page tables of the VMA, 0
643 * if the page is not mapped into the page tables of this VMA. Only
644 * valid for normal file or anonymous VMAs.
646 int page_mapped_in_vma(struct page *page, struct vm_area_struct *vma)
648 unsigned long address;
652 address = __vma_address(page, vma);
653 if (unlikely(address < vma->vm_start || address >= vma->vm_end))
655 pte = page_check_address(page, vma->vm_mm, address, &ptl, 1);
656 if (!pte) /* the page is not in this mm */
658 pte_unmap_unlock(pte, ptl);
664 * Subfunctions of page_referenced: page_referenced_one called
665 * repeatedly from either page_referenced_anon or page_referenced_file.
667 int page_referenced_one(struct page *page, struct vm_area_struct *vma,
668 unsigned long address, unsigned int *mapcount,
669 unsigned long *vm_flags)
671 struct mm_struct *mm = vma->vm_mm;
674 if (unlikely(PageTransHuge(page))) {
677 spin_lock(&mm->page_table_lock);
679 * rmap might return false positives; we must filter
680 * these out using page_check_address_pmd().
682 pmd = page_check_address_pmd(page, mm, address,
683 PAGE_CHECK_ADDRESS_PMD_FLAG);
685 spin_unlock(&mm->page_table_lock);
689 if (vma->vm_flags & VM_LOCKED) {
690 spin_unlock(&mm->page_table_lock);
691 *mapcount = 0; /* break early from loop */
692 *vm_flags |= VM_LOCKED;
696 /* go ahead even if the pmd is pmd_trans_splitting() */
697 if (pmdp_clear_flush_young_notify(vma, address, pmd))
699 spin_unlock(&mm->page_table_lock);
705 * rmap might return false positives; we must filter
706 * these out using page_check_address().
708 pte = page_check_address(page, mm, address, &ptl, 0);
712 if (vma->vm_flags & VM_LOCKED) {
713 pte_unmap_unlock(pte, ptl);
714 *mapcount = 0; /* break early from loop */
715 *vm_flags |= VM_LOCKED;
719 if (ptep_clear_flush_young_notify(vma, address, pte)) {
721 * Don't treat a reference through a sequentially read
722 * mapping as such. If the page has been used in
723 * another mapping, we will catch it; if this other
724 * mapping is already gone, the unmap path will have
725 * set PG_referenced or activated the page.
727 if (likely(!VM_SequentialReadHint(vma)))
730 pte_unmap_unlock(pte, ptl);
736 *vm_flags |= vma->vm_flags;
741 static int page_referenced_anon(struct page *page,
742 struct mem_cgroup *memcg,
743 unsigned long *vm_flags)
745 unsigned int mapcount;
746 struct anon_vma *anon_vma;
748 struct anon_vma_chain *avc;
751 anon_vma = page_lock_anon_vma_read(page);
755 mapcount = page_mapcount(page);
756 pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT);
757 anon_vma_interval_tree_foreach(avc, &anon_vma->rb_root, pgoff, pgoff) {
758 struct vm_area_struct *vma = avc->vma;
759 unsigned long address = vma_address(page, vma);
761 * If we are reclaiming on behalf of a cgroup, skip
762 * counting on behalf of references from different
765 if (memcg && !mm_match_cgroup(vma->vm_mm, memcg))
767 referenced += page_referenced_one(page, vma, address,
768 &mapcount, vm_flags);
773 page_unlock_anon_vma_read(anon_vma);
778 * page_referenced_file - referenced check for object-based rmap
779 * @page: the page we're checking references on.
780 * @memcg: target memory control group
781 * @vm_flags: collect encountered vma->vm_flags who actually referenced the page
783 * For an object-based mapped page, find all the places it is mapped and
784 * check/clear the referenced flag. This is done by following the page->mapping
785 * pointer, then walking the chain of vmas it holds. It returns the number
786 * of references it found.
788 * This function is only called from page_referenced for object-based pages.
790 static int page_referenced_file(struct page *page,
791 struct mem_cgroup *memcg,
792 unsigned long *vm_flags)
794 unsigned int mapcount;
795 struct address_space *mapping = page->mapping;
796 pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT);
797 struct vm_area_struct *vma;
801 * The caller's checks on page->mapping and !PageAnon have made
802 * sure that this is a file page: the check for page->mapping
803 * excludes the case just before it gets set on an anon page.
805 BUG_ON(PageAnon(page));
808 * The page lock not only makes sure that page->mapping cannot
809 * suddenly be NULLified by truncation, it makes sure that the
810 * structure at mapping cannot be freed and reused yet,
811 * so we can safely take mapping->i_mmap_mutex.
813 BUG_ON(!PageLocked(page));
815 mutex_lock(&mapping->i_mmap_mutex);
818 * i_mmap_mutex does not stabilize mapcount at all, but mapcount
819 * is more likely to be accurate if we note it after spinning.
821 mapcount = page_mapcount(page);
823 vma_interval_tree_foreach(vma, &mapping->i_mmap, pgoff, pgoff) {
824 unsigned long address = vma_address(page, vma);
826 * If we are reclaiming on behalf of a cgroup, skip
827 * counting on behalf of references from different
830 if (memcg && !mm_match_cgroup(vma->vm_mm, memcg))
832 referenced += page_referenced_one(page, vma, address,
833 &mapcount, vm_flags);
838 mutex_unlock(&mapping->i_mmap_mutex);
843 * page_referenced - test if the page was referenced
844 * @page: the page to test
845 * @is_locked: caller holds lock on the page
846 * @memcg: target memory cgroup
847 * @vm_flags: collect encountered vma->vm_flags who actually referenced the page
849 * Quick test_and_clear_referenced for all mappings to a page,
850 * returns the number of ptes which referenced the page.
852 int page_referenced(struct page *page,
854 struct mem_cgroup *memcg,
855 unsigned long *vm_flags)
861 if (page_mapped(page) && page_rmapping(page)) {
862 if (!is_locked && (!PageAnon(page) || PageKsm(page))) {
863 we_locked = trylock_page(page);
869 if (unlikely(PageKsm(page)))
870 referenced += page_referenced_ksm(page, memcg,
872 else if (PageAnon(page))
873 referenced += page_referenced_anon(page, memcg,
875 else if (page->mapping)
876 referenced += page_referenced_file(page, memcg,
881 if (page_test_and_clear_young(page_to_pfn(page)))
888 static int page_mkclean_one(struct page *page, struct vm_area_struct *vma,
889 unsigned long address)
891 struct mm_struct *mm = vma->vm_mm;
896 pte = page_check_address(page, mm, address, &ptl, 1);
900 if (pte_dirty(*pte) || pte_write(*pte)) {
903 flush_cache_page(vma, address, pte_pfn(*pte));
904 entry = ptep_clear_flush(vma, address, pte);
905 entry = pte_wrprotect(entry);
906 entry = pte_mkclean(entry);
907 set_pte_at(mm, address, pte, entry);
911 pte_unmap_unlock(pte, ptl);
914 mmu_notifier_invalidate_page(mm, address);
919 static int page_mkclean_file(struct address_space *mapping, struct page *page)
921 pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT);
922 struct vm_area_struct *vma;
925 BUG_ON(PageAnon(page));
927 mutex_lock(&mapping->i_mmap_mutex);
928 vma_interval_tree_foreach(vma, &mapping->i_mmap, pgoff, pgoff) {
929 if (vma->vm_flags & VM_SHARED) {
930 unsigned long address = vma_address(page, vma);
931 ret += page_mkclean_one(page, vma, address);
934 mutex_unlock(&mapping->i_mmap_mutex);
938 int page_mkclean(struct page *page)
942 BUG_ON(!PageLocked(page));
944 if (page_mapped(page)) {
945 struct address_space *mapping = page_mapping(page);
947 ret = page_mkclean_file(mapping, page);
952 EXPORT_SYMBOL_GPL(page_mkclean);
955 * page_move_anon_rmap - move a page to our anon_vma
956 * @page: the page to move to our anon_vma
957 * @vma: the vma the page belongs to
958 * @address: the user virtual address mapped
960 * When a page belongs exclusively to one process after a COW event,
961 * that page can be moved into the anon_vma that belongs to just that
962 * process, so the rmap code will not search the parent or sibling
965 void page_move_anon_rmap(struct page *page,
966 struct vm_area_struct *vma, unsigned long address)
968 struct anon_vma *anon_vma = vma->anon_vma;
970 VM_BUG_ON(!PageLocked(page));
971 VM_BUG_ON(!anon_vma);
972 VM_BUG_ON(page->index != linear_page_index(vma, address));
974 anon_vma = (void *) anon_vma + PAGE_MAPPING_ANON;
975 page->mapping = (struct address_space *) anon_vma;
979 * __page_set_anon_rmap - set up new anonymous rmap
980 * @page: Page to add to rmap
981 * @vma: VM area to add page to.
982 * @address: User virtual address of the mapping
983 * @exclusive: the page is exclusively owned by the current process
985 static void __page_set_anon_rmap(struct page *page,
986 struct vm_area_struct *vma, unsigned long address, int exclusive)
988 struct anon_vma *anon_vma = vma->anon_vma;
996 * If the page isn't exclusively mapped into this vma,
997 * we must use the _oldest_ possible anon_vma for the
1001 anon_vma = anon_vma->root;
1003 anon_vma = (void *) anon_vma + PAGE_MAPPING_ANON;
1004 page->mapping = (struct address_space *) anon_vma;
1005 page->index = linear_page_index(vma, address);
1009 * __page_check_anon_rmap - sanity check anonymous rmap addition
1010 * @page: the page to add the mapping to
1011 * @vma: the vm area in which the mapping is added
1012 * @address: the user virtual address mapped
1014 static void __page_check_anon_rmap(struct page *page,
1015 struct vm_area_struct *vma, unsigned long address)
1017 #ifdef CONFIG_DEBUG_VM
1019 * The page's anon-rmap details (mapping and index) are guaranteed to
1020 * be set up correctly at this point.
1022 * We have exclusion against page_add_anon_rmap because the caller
1023 * always holds the page locked, except if called from page_dup_rmap,
1024 * in which case the page is already known to be setup.
1026 * We have exclusion against page_add_new_anon_rmap because those pages
1027 * are initially only visible via the pagetables, and the pte is locked
1028 * over the call to page_add_new_anon_rmap.
1030 BUG_ON(page_anon_vma(page)->root != vma->anon_vma->root);
1031 BUG_ON(page->index != linear_page_index(vma, address));
1036 * page_add_anon_rmap - add pte mapping to an anonymous page
1037 * @page: the page to add the mapping to
1038 * @vma: the vm area in which the mapping is added
1039 * @address: the user virtual address mapped
1041 * The caller needs to hold the pte lock, and the page must be locked in
1042 * the anon_vma case: to serialize mapping,index checking after setting,
1043 * and to ensure that PageAnon is not being upgraded racily to PageKsm
1044 * (but PageKsm is never downgraded to PageAnon).
1046 void page_add_anon_rmap(struct page *page,
1047 struct vm_area_struct *vma, unsigned long address)
1049 do_page_add_anon_rmap(page, vma, address, 0);
1053 * Special version of the above for do_swap_page, which often runs
1054 * into pages that are exclusively owned by the current process.
1055 * Everybody else should continue to use page_add_anon_rmap above.
1057 void do_page_add_anon_rmap(struct page *page,
1058 struct vm_area_struct *vma, unsigned long address, int exclusive)
1060 int first = atomic_inc_and_test(&page->_mapcount);
1062 if (!PageTransHuge(page))
1063 __inc_zone_page_state(page, NR_ANON_PAGES);
1065 __inc_zone_page_state(page,
1066 NR_ANON_TRANSPARENT_HUGEPAGES);
1068 if (unlikely(PageKsm(page)))
1071 VM_BUG_ON(!PageLocked(page));
1072 /* address might be in next vma when migration races vma_adjust */
1074 __page_set_anon_rmap(page, vma, address, exclusive);
1076 __page_check_anon_rmap(page, vma, address);
1080 * page_add_new_anon_rmap - add pte mapping to a new anonymous page
1081 * @page: the page to add the mapping to
1082 * @vma: the vm area in which the mapping is added
1083 * @address: the user virtual address mapped
1085 * Same as page_add_anon_rmap but must only be called on *new* pages.
1086 * This means the inc-and-test can be bypassed.
1087 * Page does not have to be locked.
1089 void page_add_new_anon_rmap(struct page *page,
1090 struct vm_area_struct *vma, unsigned long address)
1092 VM_BUG_ON(address < vma->vm_start || address >= vma->vm_end);
1093 SetPageSwapBacked(page);
1094 atomic_set(&page->_mapcount, 0); /* increment count (starts at -1) */
1095 if (!PageTransHuge(page))
1096 __inc_zone_page_state(page, NR_ANON_PAGES);
1098 __inc_zone_page_state(page, NR_ANON_TRANSPARENT_HUGEPAGES);
1099 __page_set_anon_rmap(page, vma, address, 1);
1100 if (!mlocked_vma_newpage(vma, page))
1101 lru_cache_add_lru(page, LRU_ACTIVE_ANON);
1103 add_page_to_unevictable_list(page);
1107 * page_add_file_rmap - add pte mapping to a file page
1108 * @page: the page to add the mapping to
1110 * The caller needs to hold the pte lock.
1112 void page_add_file_rmap(struct page *page)
1115 unsigned long flags;
1117 mem_cgroup_begin_update_page_stat(page, &locked, &flags);
1118 if (atomic_inc_and_test(&page->_mapcount)) {
1119 __inc_zone_page_state(page, NR_FILE_MAPPED);
1120 mem_cgroup_inc_page_stat(page, MEMCG_NR_FILE_MAPPED);
1122 mem_cgroup_end_update_page_stat(page, &locked, &flags);
1126 * page_remove_rmap - take down pte mapping from a page
1127 * @page: page to remove mapping from
1129 * The caller needs to hold the pte lock.
1131 void page_remove_rmap(struct page *page)
1133 bool anon = PageAnon(page);
1135 unsigned long flags;
1138 * The anon case has no mem_cgroup page_stat to update; but may
1139 * uncharge_page() below, where the lock ordering can deadlock if
1140 * we hold the lock against page_stat move: so avoid it on anon.
1143 mem_cgroup_begin_update_page_stat(page, &locked, &flags);
1145 /* page still mapped by someone else? */
1146 if (!atomic_add_negative(-1, &page->_mapcount))
1150 * Hugepages are not counted in NR_ANON_PAGES nor NR_FILE_MAPPED
1151 * and not charged by memcg for now.
1153 if (unlikely(PageHuge(page)))
1156 mem_cgroup_uncharge_page(page);
1157 if (!PageTransHuge(page))
1158 __dec_zone_page_state(page, NR_ANON_PAGES);
1160 __dec_zone_page_state(page,
1161 NR_ANON_TRANSPARENT_HUGEPAGES);
1163 __dec_zone_page_state(page, NR_FILE_MAPPED);
1164 mem_cgroup_dec_page_stat(page, MEMCG_NR_FILE_MAPPED);
1165 mem_cgroup_end_update_page_stat(page, &locked, &flags);
1167 if (unlikely(PageMlocked(page)))
1168 clear_page_mlock(page);
1170 * It would be tidy to reset the PageAnon mapping here,
1171 * but that might overwrite a racing page_add_anon_rmap
1172 * which increments mapcount after us but sets mapping
1173 * before us: so leave the reset to free_hot_cold_page,
1174 * and remember that it's only reliable while mapped.
1175 * Leaving it set also helps swapoff to reinstate ptes
1176 * faster for those pages still in swapcache.
1181 mem_cgroup_end_update_page_stat(page, &locked, &flags);
1185 * Subfunctions of try_to_unmap: try_to_unmap_one called
1186 * repeatedly from try_to_unmap_ksm, try_to_unmap_anon or try_to_unmap_file.
1188 int try_to_unmap_one(struct page *page, struct vm_area_struct *vma,
1189 unsigned long address, enum ttu_flags flags)
1191 struct mm_struct *mm = vma->vm_mm;
1195 int ret = SWAP_AGAIN;
1197 pte = page_check_address(page, mm, address, &ptl, 0);
1202 * If the page is mlock()d, we cannot swap it out.
1203 * If it's recently referenced (perhaps page_referenced
1204 * skipped over this mm) then we should reactivate it.
1206 if (!(flags & TTU_IGNORE_MLOCK)) {
1207 if (vma->vm_flags & VM_LOCKED)
1210 if (TTU_ACTION(flags) == TTU_MUNLOCK)
1213 if (!(flags & TTU_IGNORE_ACCESS)) {
1214 if (ptep_clear_flush_young_notify(vma, address, pte)) {
1220 /* Nuke the page table entry. */
1221 flush_cache_page(vma, address, page_to_pfn(page));
1222 pteval = ptep_clear_flush(vma, address, pte);
1224 /* Move the dirty bit to the physical page now the pte is gone. */
1225 if (pte_dirty(pteval))
1226 set_page_dirty(page);
1228 /* Update high watermark before we lower rss */
1229 update_hiwater_rss(mm);
1231 if (PageHWPoison(page) && !(flags & TTU_IGNORE_HWPOISON)) {
1232 if (!PageHuge(page)) {
1234 dec_mm_counter(mm, MM_ANONPAGES);
1236 dec_mm_counter(mm, MM_FILEPAGES);
1238 set_pte_at(mm, address, pte,
1239 swp_entry_to_pte(make_hwpoison_entry(page)));
1240 } else if (PageAnon(page)) {
1241 swp_entry_t entry = { .val = page_private(page) };
1243 if (PageSwapCache(page)) {
1245 * Store the swap location in the pte.
1246 * See handle_pte_fault() ...
1248 if (swap_duplicate(entry) < 0) {
1249 set_pte_at(mm, address, pte, pteval);
1253 if (list_empty(&mm->mmlist)) {
1254 spin_lock(&mmlist_lock);
1255 if (list_empty(&mm->mmlist))
1256 list_add(&mm->mmlist, &init_mm.mmlist);
1257 spin_unlock(&mmlist_lock);
1259 dec_mm_counter(mm, MM_ANONPAGES);
1260 inc_mm_counter(mm, MM_SWAPENTS);
1261 } else if (IS_ENABLED(CONFIG_MIGRATION)) {
1263 * Store the pfn of the page in a special migration
1264 * pte. do_swap_page() will wait until the migration
1265 * pte is removed and then restart fault handling.
1267 BUG_ON(TTU_ACTION(flags) != TTU_MIGRATION);
1268 entry = make_migration_entry(page, pte_write(pteval));
1270 set_pte_at(mm, address, pte, swp_entry_to_pte(entry));
1271 BUG_ON(pte_file(*pte));
1272 } else if (IS_ENABLED(CONFIG_MIGRATION) &&
1273 (TTU_ACTION(flags) == TTU_MIGRATION)) {
1274 /* Establish migration entry for a file page */
1276 entry = make_migration_entry(page, pte_write(pteval));
1277 set_pte_at(mm, address, pte, swp_entry_to_pte(entry));
1279 dec_mm_counter(mm, MM_FILEPAGES);
1281 page_remove_rmap(page);
1282 page_cache_release(page);
1285 pte_unmap_unlock(pte, ptl);
1286 if (ret != SWAP_FAIL)
1287 mmu_notifier_invalidate_page(mm, address);
1292 pte_unmap_unlock(pte, ptl);
1296 * We need mmap_sem locking, Otherwise VM_LOCKED check makes
1297 * unstable result and race. Plus, We can't wait here because
1298 * we now hold anon_vma->rwsem or mapping->i_mmap_mutex.
1299 * if trylock failed, the page remain in evictable lru and later
1300 * vmscan could retry to move the page to unevictable lru if the
1301 * page is actually mlocked.
1303 if (down_read_trylock(&vma->vm_mm->mmap_sem)) {
1304 if (vma->vm_flags & VM_LOCKED) {
1305 mlock_vma_page(page);
1308 up_read(&vma->vm_mm->mmap_sem);
1314 * objrmap doesn't work for nonlinear VMAs because the assumption that
1315 * offset-into-file correlates with offset-into-virtual-addresses does not hold.
1316 * Consequently, given a particular page and its ->index, we cannot locate the
1317 * ptes which are mapping that page without an exhaustive linear search.
1319 * So what this code does is a mini "virtual scan" of each nonlinear VMA which
1320 * maps the file to which the target page belongs. The ->vm_private_data field
1321 * holds the current cursor into that scan. Successive searches will circulate
1322 * around the vma's virtual address space.
1324 * So as more replacement pressure is applied to the pages in a nonlinear VMA,
1325 * more scanning pressure is placed against them as well. Eventually pages
1326 * will become fully unmapped and are eligible for eviction.
1328 * For very sparsely populated VMAs this is a little inefficient - chances are
1329 * there there won't be many ptes located within the scan cluster. In this case
1330 * maybe we could scan further - to the end of the pte page, perhaps.
1332 * Mlocked pages: check VM_LOCKED under mmap_sem held for read, if we can
1333 * acquire it without blocking. If vma locked, mlock the pages in the cluster,
1334 * rather than unmapping them. If we encounter the "check_page" that vmscan is
1335 * trying to unmap, return SWAP_MLOCK, else default SWAP_AGAIN.
1337 #define CLUSTER_SIZE min(32*PAGE_SIZE, PMD_SIZE)
1338 #define CLUSTER_MASK (~(CLUSTER_SIZE - 1))
1340 static int try_to_unmap_cluster(unsigned long cursor, unsigned int *mapcount,
1341 struct vm_area_struct *vma, struct page *check_page)
1343 struct mm_struct *mm = vma->vm_mm;
1349 unsigned long address;
1350 unsigned long mmun_start; /* For mmu_notifiers */
1351 unsigned long mmun_end; /* For mmu_notifiers */
1353 int ret = SWAP_AGAIN;
1356 address = (vma->vm_start + cursor) & CLUSTER_MASK;
1357 end = address + CLUSTER_SIZE;
1358 if (address < vma->vm_start)
1359 address = vma->vm_start;
1360 if (end > vma->vm_end)
1363 pmd = mm_find_pmd(mm, address);
1367 mmun_start = address;
1369 mmu_notifier_invalidate_range_start(mm, mmun_start, mmun_end);
1372 * If we can acquire the mmap_sem for read, and vma is VM_LOCKED,
1373 * keep the sem while scanning the cluster for mlocking pages.
1375 if (down_read_trylock(&vma->vm_mm->mmap_sem)) {
1376 locked_vma = (vma->vm_flags & VM_LOCKED);
1378 up_read(&vma->vm_mm->mmap_sem); /* don't need it */
1381 pte = pte_offset_map_lock(mm, pmd, address, &ptl);
1383 /* Update high watermark before we lower rss */
1384 update_hiwater_rss(mm);
1386 for (; address < end; pte++, address += PAGE_SIZE) {
1387 if (!pte_present(*pte))
1389 page = vm_normal_page(vma, address, *pte);
1390 BUG_ON(!page || PageAnon(page));
1393 mlock_vma_page(page); /* no-op if already mlocked */
1394 if (page == check_page)
1396 continue; /* don't unmap */
1399 if (ptep_clear_flush_young_notify(vma, address, pte))
1402 /* Nuke the page table entry. */
1403 flush_cache_page(vma, address, pte_pfn(*pte));
1404 pteval = ptep_clear_flush(vma, address, pte);
1406 /* If nonlinear, store the file page offset in the pte. */
1407 if (page->index != linear_page_index(vma, address))
1408 set_pte_at(mm, address, pte, pgoff_to_pte(page->index));
1410 /* Move the dirty bit to the physical page now the pte is gone. */
1411 if (pte_dirty(pteval))
1412 set_page_dirty(page);
1414 page_remove_rmap(page);
1415 page_cache_release(page);
1416 dec_mm_counter(mm, MM_FILEPAGES);
1419 pte_unmap_unlock(pte - 1, ptl);
1420 mmu_notifier_invalidate_range_end(mm, mmun_start, mmun_end);
1422 up_read(&vma->vm_mm->mmap_sem);
1426 bool is_vma_temporary_stack(struct vm_area_struct *vma)
1428 int maybe_stack = vma->vm_flags & (VM_GROWSDOWN | VM_GROWSUP);
1433 if ((vma->vm_flags & VM_STACK_INCOMPLETE_SETUP) ==
1434 VM_STACK_INCOMPLETE_SETUP)
1441 * try_to_unmap_anon - unmap or unlock anonymous page using the object-based
1443 * @page: the page to unmap/unlock
1444 * @flags: action and flags
1446 * Find all the mappings of a page using the mapping pointer and the vma chains
1447 * contained in the anon_vma struct it points to.
1449 * This function is only called from try_to_unmap/try_to_munlock for
1451 * When called from try_to_munlock(), the mmap_sem of the mm containing the vma
1452 * where the page was found will be held for write. So, we won't recheck
1453 * vm_flags for that VMA. That should be OK, because that vma shouldn't be
1456 static int try_to_unmap_anon(struct page *page, enum ttu_flags flags)
1458 struct anon_vma *anon_vma;
1460 struct anon_vma_chain *avc;
1461 int ret = SWAP_AGAIN;
1463 anon_vma = page_lock_anon_vma_read(page);
1467 pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT);
1468 anon_vma_interval_tree_foreach(avc, &anon_vma->rb_root, pgoff, pgoff) {
1469 struct vm_area_struct *vma = avc->vma;
1470 unsigned long address;
1473 * During exec, a temporary VMA is setup and later moved.
1474 * The VMA is moved under the anon_vma lock but not the
1475 * page tables leading to a race where migration cannot
1476 * find the migration ptes. Rather than increasing the
1477 * locking requirements of exec(), migration skips
1478 * temporary VMAs until after exec() completes.
1480 if (IS_ENABLED(CONFIG_MIGRATION) && (flags & TTU_MIGRATION) &&
1481 is_vma_temporary_stack(vma))
1484 address = vma_address(page, vma);
1485 ret = try_to_unmap_one(page, vma, address, flags);
1486 if (ret != SWAP_AGAIN || !page_mapped(page))
1490 page_unlock_anon_vma_read(anon_vma);
1495 * try_to_unmap_file - unmap/unlock file page using the object-based rmap method
1496 * @page: the page to unmap/unlock
1497 * @flags: action and flags
1499 * Find all the mappings of a page using the mapping pointer and the vma chains
1500 * contained in the address_space struct it points to.
1502 * This function is only called from try_to_unmap/try_to_munlock for
1503 * object-based pages.
1504 * When called from try_to_munlock(), the mmap_sem of the mm containing the vma
1505 * where the page was found will be held for write. So, we won't recheck
1506 * vm_flags for that VMA. That should be OK, because that vma shouldn't be
1509 static int try_to_unmap_file(struct page *page, enum ttu_flags flags)
1511 struct address_space *mapping = page->mapping;
1512 pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT);
1513 struct vm_area_struct *vma;
1514 int ret = SWAP_AGAIN;
1515 unsigned long cursor;
1516 unsigned long max_nl_cursor = 0;
1517 unsigned long max_nl_size = 0;
1518 unsigned int mapcount;
1521 pgoff = page->index << compound_order(page);
1523 mutex_lock(&mapping->i_mmap_mutex);
1524 vma_interval_tree_foreach(vma, &mapping->i_mmap, pgoff, pgoff) {
1525 unsigned long address = vma_address(page, vma);
1526 ret = try_to_unmap_one(page, vma, address, flags);
1527 if (ret != SWAP_AGAIN || !page_mapped(page))
1531 if (list_empty(&mapping->i_mmap_nonlinear))
1535 * We don't bother to try to find the munlocked page in nonlinears.
1536 * It's costly. Instead, later, page reclaim logic may call
1537 * try_to_unmap(TTU_MUNLOCK) and recover PG_mlocked lazily.
1539 if (TTU_ACTION(flags) == TTU_MUNLOCK)
1542 list_for_each_entry(vma, &mapping->i_mmap_nonlinear,
1544 cursor = (unsigned long) vma->vm_private_data;
1545 if (cursor > max_nl_cursor)
1546 max_nl_cursor = cursor;
1547 cursor = vma->vm_end - vma->vm_start;
1548 if (cursor > max_nl_size)
1549 max_nl_size = cursor;
1552 if (max_nl_size == 0) { /* all nonlinears locked or reserved ? */
1558 * We don't try to search for this page in the nonlinear vmas,
1559 * and page_referenced wouldn't have found it anyway. Instead
1560 * just walk the nonlinear vmas trying to age and unmap some.
1561 * The mapcount of the page we came in with is irrelevant,
1562 * but even so use it as a guide to how hard we should try?
1564 mapcount = page_mapcount(page);
1569 max_nl_size = (max_nl_size + CLUSTER_SIZE - 1) & CLUSTER_MASK;
1570 if (max_nl_cursor == 0)
1571 max_nl_cursor = CLUSTER_SIZE;
1574 list_for_each_entry(vma, &mapping->i_mmap_nonlinear,
1576 cursor = (unsigned long) vma->vm_private_data;
1577 while ( cursor < max_nl_cursor &&
1578 cursor < vma->vm_end - vma->vm_start) {
1579 if (try_to_unmap_cluster(cursor, &mapcount,
1580 vma, page) == SWAP_MLOCK)
1582 cursor += CLUSTER_SIZE;
1583 vma->vm_private_data = (void *) cursor;
1584 if ((int)mapcount <= 0)
1587 vma->vm_private_data = (void *) max_nl_cursor;
1590 max_nl_cursor += CLUSTER_SIZE;
1591 } while (max_nl_cursor <= max_nl_size);
1594 * Don't loop forever (perhaps all the remaining pages are
1595 * in locked vmas). Reset cursor on all unreserved nonlinear
1596 * vmas, now forgetting on which ones it had fallen behind.
1598 list_for_each_entry(vma, &mapping->i_mmap_nonlinear, shared.nonlinear)
1599 vma->vm_private_data = NULL;
1601 mutex_unlock(&mapping->i_mmap_mutex);
1606 * try_to_unmap - try to remove all page table mappings to a page
1607 * @page: the page to get unmapped
1608 * @flags: action and flags
1610 * Tries to remove all the page table entries which are mapping this
1611 * page, used in the pageout path. Caller must hold the page lock.
1612 * Return values are:
1614 * SWAP_SUCCESS - we succeeded in removing all mappings
1615 * SWAP_AGAIN - we missed a mapping, try again later
1616 * SWAP_FAIL - the page is unswappable
1617 * SWAP_MLOCK - page is mlocked.
1619 int try_to_unmap(struct page *page, enum ttu_flags flags)
1623 BUG_ON(!PageLocked(page));
1624 VM_BUG_ON(!PageHuge(page) && PageTransHuge(page));
1626 if (unlikely(PageKsm(page)))
1627 ret = try_to_unmap_ksm(page, flags);
1628 else if (PageAnon(page))
1629 ret = try_to_unmap_anon(page, flags);
1631 ret = try_to_unmap_file(page, flags);
1632 if (ret != SWAP_MLOCK && !page_mapped(page))
1638 * try_to_munlock - try to munlock a page
1639 * @page: the page to be munlocked
1641 * Called from munlock code. Checks all of the VMAs mapping the page
1642 * to make sure nobody else has this page mlocked. The page will be
1643 * returned with PG_mlocked cleared if no other vmas have it mlocked.
1645 * Return values are:
1647 * SWAP_AGAIN - no vma is holding page mlocked, or,
1648 * SWAP_AGAIN - page mapped in mlocked vma -- couldn't acquire mmap sem
1649 * SWAP_FAIL - page cannot be located at present
1650 * SWAP_MLOCK - page is now mlocked.
1652 int try_to_munlock(struct page *page)
1654 VM_BUG_ON(!PageLocked(page) || PageLRU(page));
1656 if (unlikely(PageKsm(page)))
1657 return try_to_unmap_ksm(page, TTU_MUNLOCK);
1658 else if (PageAnon(page))
1659 return try_to_unmap_anon(page, TTU_MUNLOCK);
1661 return try_to_unmap_file(page, TTU_MUNLOCK);
1664 void __put_anon_vma(struct anon_vma *anon_vma)
1666 struct anon_vma *root = anon_vma->root;
1668 if (root != anon_vma && atomic_dec_and_test(&root->refcount))
1669 anon_vma_free(root);
1671 anon_vma_free(anon_vma);
1674 #ifdef CONFIG_MIGRATION
1676 * rmap_walk() and its helpers rmap_walk_anon() and rmap_walk_file():
1677 * Called by migrate.c to remove migration ptes, but might be used more later.
1679 static int rmap_walk_anon(struct page *page, int (*rmap_one)(struct page *,
1680 struct vm_area_struct *, unsigned long, void *), void *arg)
1682 struct anon_vma *anon_vma;
1683 pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT);
1684 struct anon_vma_chain *avc;
1685 int ret = SWAP_AGAIN;
1688 * Note: remove_migration_ptes() cannot use page_lock_anon_vma_read()
1689 * because that depends on page_mapped(); but not all its usages
1690 * are holding mmap_sem. Users without mmap_sem are required to
1691 * take a reference count to prevent the anon_vma disappearing
1693 anon_vma = page_anon_vma(page);
1696 anon_vma_lock_read(anon_vma);
1697 anon_vma_interval_tree_foreach(avc, &anon_vma->rb_root, pgoff, pgoff) {
1698 struct vm_area_struct *vma = avc->vma;
1699 unsigned long address = vma_address(page, vma);
1700 ret = rmap_one(page, vma, address, arg);
1701 if (ret != SWAP_AGAIN)
1704 anon_vma_unlock_read(anon_vma);
1708 static int rmap_walk_file(struct page *page, int (*rmap_one)(struct page *,
1709 struct vm_area_struct *, unsigned long, void *), void *arg)
1711 struct address_space *mapping = page->mapping;
1712 pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT);
1713 struct vm_area_struct *vma;
1714 int ret = SWAP_AGAIN;
1718 mutex_lock(&mapping->i_mmap_mutex);
1719 vma_interval_tree_foreach(vma, &mapping->i_mmap, pgoff, pgoff) {
1720 unsigned long address = vma_address(page, vma);
1721 ret = rmap_one(page, vma, address, arg);
1722 if (ret != SWAP_AGAIN)
1726 * No nonlinear handling: being always shared, nonlinear vmas
1727 * never contain migration ptes. Decide what to do about this
1728 * limitation to linear when we need rmap_walk() on nonlinear.
1730 mutex_unlock(&mapping->i_mmap_mutex);
1734 int rmap_walk(struct page *page, int (*rmap_one)(struct page *,
1735 struct vm_area_struct *, unsigned long, void *), void *arg)
1737 VM_BUG_ON(!PageLocked(page));
1739 if (unlikely(PageKsm(page)))
1740 return rmap_walk_ksm(page, rmap_one, arg);
1741 else if (PageAnon(page))
1742 return rmap_walk_anon(page, rmap_one, arg);
1744 return rmap_walk_file(page, rmap_one, arg);
1746 #endif /* CONFIG_MIGRATION */
1748 #ifdef CONFIG_HUGETLB_PAGE
1750 * The following three functions are for anonymous (private mapped) hugepages.
1751 * Unlike common anonymous pages, anonymous hugepages have no accounting code
1752 * and no lru code, because we handle hugepages differently from common pages.
1754 static void __hugepage_set_anon_rmap(struct page *page,
1755 struct vm_area_struct *vma, unsigned long address, int exclusive)
1757 struct anon_vma *anon_vma = vma->anon_vma;
1764 anon_vma = anon_vma->root;
1766 anon_vma = (void *) anon_vma + PAGE_MAPPING_ANON;
1767 page->mapping = (struct address_space *) anon_vma;
1768 page->index = linear_page_index(vma, address);
1771 void hugepage_add_anon_rmap(struct page *page,
1772 struct vm_area_struct *vma, unsigned long address)
1774 struct anon_vma *anon_vma = vma->anon_vma;
1777 BUG_ON(!PageLocked(page));
1779 /* address might be in next vma when migration races vma_adjust */
1780 first = atomic_inc_and_test(&page->_mapcount);
1782 __hugepage_set_anon_rmap(page, vma, address, 0);
1785 void hugepage_add_new_anon_rmap(struct page *page,
1786 struct vm_area_struct *vma, unsigned long address)
1788 BUG_ON(address < vma->vm_start || address >= vma->vm_end);
1789 atomic_set(&page->_mapcount, 0);
1790 __hugepage_set_anon_rmap(page, vma, address, 1);
1792 #endif /* CONFIG_HUGETLB_PAGE */