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->mutex,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() 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 * mutex_trylock() from page_lock_anon_vma(). This orders:
98 * page_lock_anon_vma() VS put_anon_vma()
99 * mutex_trylock() atomic_dec_and_test()
101 * atomic_read() mutex_is_locked()
103 * LOCK should suffice since the actual taking of the lock must
104 * happen _before_ what follows.
106 if (mutex_is_locked(&anon_vma->root->mutex)) {
107 anon_vma_lock(anon_vma);
108 anon_vma_unlock(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()
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(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(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 mutex_unlock(&root->mutex);
224 mutex_lock(&root->mutex);
229 static inline void unlock_anon_vma_root(struct anon_vma *root)
232 mutex_unlock(&root->mutex);
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(anon_vma);
310 anon_vma_chain_link(vma, avc, anon_vma);
311 anon_vma_unlock(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 acquire the anon_vma->root->mutex.
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 mutex_init(&anon_vma->mutex);
369 atomic_set(&anon_vma->refcount, 0);
371 atomic_set(&anon_vma->swapra_miss, 0);
373 anon_vma->rb_root = RB_ROOT;
376 void __init anon_vma_init(void)
378 anon_vma_cachep = kmem_cache_create("anon_vma", sizeof(struct anon_vma),
379 0, SLAB_DESTROY_BY_RCU|SLAB_PANIC, anon_vma_ctor);
380 anon_vma_chain_cachep = KMEM_CACHE(anon_vma_chain, SLAB_PANIC);
384 * Getting a lock on a stable anon_vma from a page off the LRU is tricky!
386 * Since there is no serialization what so ever against page_remove_rmap()
387 * the best this function can do is return a locked anon_vma that might
388 * have been relevant to this page.
390 * The page might have been remapped to a different anon_vma or the anon_vma
391 * returned may already be freed (and even reused).
393 * In case it was remapped to a different anon_vma, the new anon_vma will be a
394 * child of the old anon_vma, and the anon_vma lifetime rules will therefore
395 * ensure that any anon_vma obtained from the page will still be valid for as
396 * long as we observe page_mapped() [ hence all those page_mapped() tests ].
398 * All users of this function must be very careful when walking the anon_vma
399 * chain and verify that the page in question is indeed mapped in it
400 * [ something equivalent to page_mapped_in_vma() ].
402 * Since anon_vma's slab is DESTROY_BY_RCU and we know from page_remove_rmap()
403 * that the anon_vma pointer from page->mapping is valid if there is a
404 * mapcount, we can dereference the anon_vma after observing those.
406 struct anon_vma *page_get_anon_vma(struct page *page)
408 struct anon_vma *anon_vma = NULL;
409 unsigned long anon_mapping;
412 anon_mapping = (unsigned long) ACCESS_ONCE(page->mapping);
413 if ((anon_mapping & PAGE_MAPPING_FLAGS) != PAGE_MAPPING_ANON)
415 if (!page_mapped(page))
418 anon_vma = (struct anon_vma *) (anon_mapping - PAGE_MAPPING_ANON);
419 if (!atomic_inc_not_zero(&anon_vma->refcount)) {
425 * If this page is still mapped, then its anon_vma cannot have been
426 * freed. But if it has been unmapped, we have no security against the
427 * anon_vma structure being freed and reused (for another anon_vma:
428 * SLAB_DESTROY_BY_RCU guarantees that - so the atomic_inc_not_zero()
429 * above cannot corrupt).
431 if (!page_mapped(page)) {
432 put_anon_vma(anon_vma);
442 * Similar to page_get_anon_vma() except it locks the anon_vma.
444 * Its a little more complex as it tries to keep the fast path to a single
445 * atomic op -- the trylock. If we fail the trylock, we fall back to getting a
446 * reference like with page_get_anon_vma() and then block on the mutex.
448 struct anon_vma *page_lock_anon_vma(struct page *page)
450 struct anon_vma *anon_vma = NULL;
451 struct anon_vma *root_anon_vma;
452 unsigned long anon_mapping;
455 anon_mapping = (unsigned long) ACCESS_ONCE(page->mapping);
456 if ((anon_mapping & PAGE_MAPPING_FLAGS) != PAGE_MAPPING_ANON)
458 if (!page_mapped(page))
461 anon_vma = (struct anon_vma *) (anon_mapping - PAGE_MAPPING_ANON);
462 root_anon_vma = ACCESS_ONCE(anon_vma->root);
463 if (mutex_trylock(&root_anon_vma->mutex)) {
465 * If the page is still mapped, then this anon_vma is still
466 * its anon_vma, and holding the mutex ensures that it will
467 * not go away, see anon_vma_free().
469 if (!page_mapped(page)) {
470 mutex_unlock(&root_anon_vma->mutex);
476 /* trylock failed, we got to sleep */
477 if (!atomic_inc_not_zero(&anon_vma->refcount)) {
482 if (!page_mapped(page)) {
483 put_anon_vma(anon_vma);
488 /* we pinned the anon_vma, its safe to sleep */
490 anon_vma_lock(anon_vma);
492 if (atomic_dec_and_test(&anon_vma->refcount)) {
494 * Oops, we held the last refcount, release the lock
495 * and bail -- can't simply use put_anon_vma() because
496 * we'll deadlock on the anon_vma_lock() recursion.
498 anon_vma_unlock(anon_vma);
499 __put_anon_vma(anon_vma);
510 void page_unlock_anon_vma(struct anon_vma *anon_vma)
512 anon_vma_unlock(anon_vma);
516 * At what user virtual address is page expected in @vma?
518 static inline unsigned long
519 __vma_address(struct page *page, struct vm_area_struct *vma)
521 pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT);
523 if (unlikely(is_vm_hugetlb_page(vma)))
524 pgoff = page->index << huge_page_order(page_hstate(page));
526 return vma->vm_start + ((pgoff - vma->vm_pgoff) << PAGE_SHIFT);
530 vma_address(struct page *page, struct vm_area_struct *vma)
532 unsigned long address = __vma_address(page, vma);
534 /* page should be within @vma mapping range */
535 VM_BUG_ON(address < vma->vm_start || address >= vma->vm_end);
541 * At what user virtual address is page expected in vma?
542 * Caller should check the page is actually part of the vma.
544 unsigned long page_address_in_vma(struct page *page, struct vm_area_struct *vma)
546 unsigned long address;
547 if (PageAnon(page)) {
548 struct anon_vma *page__anon_vma = page_anon_vma(page);
550 * Note: swapoff's unuse_vma() is more efficient with this
551 * check, and needs it to match anon_vma when KSM is active.
553 if (!vma->anon_vma || !page__anon_vma ||
554 vma->anon_vma->root != page__anon_vma->root)
556 } else if (page->mapping && !(vma->vm_flags & VM_NONLINEAR)) {
558 vma->vm_file->f_mapping != page->mapping)
562 address = __vma_address(page, vma);
563 if (unlikely(address < vma->vm_start || address >= vma->vm_end))
568 pmd_t *mm_find_pmd(struct mm_struct *mm, unsigned long address)
574 pgd = pgd_offset(mm, address);
575 if (!pgd_present(*pgd))
578 pud = pud_offset(pgd, address);
579 if (!pud_present(*pud))
582 pmd = pmd_offset(pud, address);
583 if (!pmd_present(*pmd))
590 * Check that @page is mapped at @address into @mm.
592 * If @sync is false, page_check_address may perform a racy check to avoid
593 * the page table lock when the pte is not present (helpful when reclaiming
594 * highly shared pages).
596 * On success returns with pte mapped and locked.
598 pte_t *__page_check_address(struct page *page, struct mm_struct *mm,
599 unsigned long address, spinlock_t **ptlp, int sync)
605 if (unlikely(PageHuge(page))) {
606 pte = huge_pte_offset(mm, address);
607 ptl = &mm->page_table_lock;
611 pmd = mm_find_pmd(mm, address);
615 if (pmd_trans_huge(*pmd))
618 pte = pte_offset_map(pmd, address);
619 /* Make a quick check before getting the lock */
620 if (!sync && !pte_present(*pte)) {
625 ptl = pte_lockptr(mm, pmd);
628 if (pte_present(*pte) && page_to_pfn(page) == pte_pfn(*pte)) {
632 pte_unmap_unlock(pte, ptl);
637 * page_mapped_in_vma - check whether a page is really mapped in a VMA
638 * @page: the page to test
639 * @vma: the VMA to test
641 * Returns 1 if the page is mapped into the page tables of the VMA, 0
642 * if the page is not mapped into the page tables of this VMA. Only
643 * valid for normal file or anonymous VMAs.
645 int page_mapped_in_vma(struct page *page, struct vm_area_struct *vma)
647 unsigned long address;
651 address = __vma_address(page, vma);
652 if (unlikely(address < vma->vm_start || address >= vma->vm_end))
654 pte = page_check_address(page, vma->vm_mm, address, &ptl, 1);
655 if (!pte) /* the page is not in this mm */
657 pte_unmap_unlock(pte, ptl);
663 * Subfunctions of page_referenced: page_referenced_one called
664 * repeatedly from either page_referenced_anon or page_referenced_file.
666 int page_referenced_one(struct page *page, struct vm_area_struct *vma,
667 unsigned long address, unsigned int *mapcount,
668 unsigned long *vm_flags)
670 struct mm_struct *mm = vma->vm_mm;
673 if (unlikely(PageTransHuge(page))) {
676 spin_lock(&mm->page_table_lock);
678 * rmap might return false positives; we must filter
679 * these out using page_check_address_pmd().
681 pmd = page_check_address_pmd(page, mm, address,
682 PAGE_CHECK_ADDRESS_PMD_FLAG);
684 spin_unlock(&mm->page_table_lock);
688 if (vma->vm_flags & VM_LOCKED) {
689 spin_unlock(&mm->page_table_lock);
690 *mapcount = 0; /* break early from loop */
691 *vm_flags |= VM_LOCKED;
695 /* go ahead even if the pmd is pmd_trans_splitting() */
696 if (pmdp_clear_flush_young_notify(vma, address, pmd))
698 spin_unlock(&mm->page_table_lock);
704 * rmap might return false positives; we must filter
705 * these out using page_check_address().
707 pte = page_check_address(page, mm, address, &ptl, 0);
711 if (vma->vm_flags & VM_LOCKED) {
712 pte_unmap_unlock(pte, ptl);
713 *mapcount = 0; /* break early from loop */
714 *vm_flags |= VM_LOCKED;
718 if (ptep_clear_flush_young_notify(vma, address, pte)) {
720 * Don't treat a reference through a sequentially read
721 * mapping as such. If the page has been used in
722 * another mapping, we will catch it; if this other
723 * mapping is already gone, the unmap path will have
724 * set PG_referenced or activated the page.
726 if (likely(!VM_SequentialReadHint(vma)))
729 pte_unmap_unlock(pte, ptl);
735 *vm_flags |= vma->vm_flags;
740 static int page_referenced_anon(struct page *page,
741 struct mem_cgroup *memcg,
742 unsigned long *vm_flags)
744 unsigned int mapcount;
745 struct anon_vma *anon_vma;
747 struct anon_vma_chain *avc;
750 anon_vma = page_lock_anon_vma(page);
754 mapcount = page_mapcount(page);
755 pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT);
756 anon_vma_interval_tree_foreach(avc, &anon_vma->rb_root, pgoff, pgoff) {
757 struct vm_area_struct *vma = avc->vma;
758 unsigned long address = vma_address(page, vma);
760 * If we are reclaiming on behalf of a cgroup, skip
761 * counting on behalf of references from different
764 if (memcg && !mm_match_cgroup(vma->vm_mm, memcg))
766 referenced += page_referenced_one(page, vma, address,
767 &mapcount, vm_flags);
772 page_unlock_anon_vma(anon_vma);
777 * page_referenced_file - referenced check for object-based rmap
778 * @page: the page we're checking references on.
779 * @memcg: target memory control group
780 * @vm_flags: collect encountered vma->vm_flags who actually referenced the page
782 * For an object-based mapped page, find all the places it is mapped and
783 * check/clear the referenced flag. This is done by following the page->mapping
784 * pointer, then walking the chain of vmas it holds. It returns the number
785 * of references it found.
787 * This function is only called from page_referenced for object-based pages.
789 static int page_referenced_file(struct page *page,
790 struct mem_cgroup *memcg,
791 unsigned long *vm_flags)
793 unsigned int mapcount;
794 struct address_space *mapping = page->mapping;
795 pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT);
796 struct vm_area_struct *vma;
800 * The caller's checks on page->mapping and !PageAnon have made
801 * sure that this is a file page: the check for page->mapping
802 * excludes the case just before it gets set on an anon page.
804 BUG_ON(PageAnon(page));
807 * The page lock not only makes sure that page->mapping cannot
808 * suddenly be NULLified by truncation, it makes sure that the
809 * structure at mapping cannot be freed and reused yet,
810 * so we can safely take mapping->i_mmap_mutex.
812 BUG_ON(!PageLocked(page));
814 mutex_lock(&mapping->i_mmap_mutex);
817 * i_mmap_mutex does not stabilize mapcount at all, but mapcount
818 * is more likely to be accurate if we note it after spinning.
820 mapcount = page_mapcount(page);
822 vma_interval_tree_foreach(vma, &mapping->i_mmap, pgoff, pgoff) {
823 unsigned long address = vma_address(page, vma);
825 * If we are reclaiming on behalf of a cgroup, skip
826 * counting on behalf of references from different
829 if (memcg && !mm_match_cgroup(vma->vm_mm, memcg))
831 referenced += page_referenced_one(page, vma, address,
832 &mapcount, vm_flags);
837 mutex_unlock(&mapping->i_mmap_mutex);
842 * page_referenced - test if the page was referenced
843 * @page: the page to test
844 * @is_locked: caller holds lock on the page
845 * @memcg: target memory cgroup
846 * @vm_flags: collect encountered vma->vm_flags who actually referenced the page
848 * Quick test_and_clear_referenced for all mappings to a page,
849 * returns the number of ptes which referenced the page.
851 int page_referenced(struct page *page,
853 struct mem_cgroup *memcg,
854 unsigned long *vm_flags)
860 if (page_mapped(page) && page_rmapping(page)) {
861 if (!is_locked && (!PageAnon(page) || PageKsm(page))) {
862 we_locked = trylock_page(page);
868 if (unlikely(PageKsm(page)))
869 referenced += page_referenced_ksm(page, memcg,
871 else if (PageAnon(page))
872 referenced += page_referenced_anon(page, memcg,
874 else if (page->mapping)
875 referenced += page_referenced_file(page, memcg,
880 if (page_test_and_clear_young(page_to_pfn(page)))
887 static int page_mkclean_one(struct page *page, struct vm_area_struct *vma,
888 unsigned long address)
890 struct mm_struct *mm = vma->vm_mm;
895 pte = page_check_address(page, mm, address, &ptl, 1);
899 if (pte_dirty(*pte) || pte_write(*pte)) {
902 flush_cache_page(vma, address, pte_pfn(*pte));
903 entry = ptep_clear_flush(vma, address, pte);
904 entry = pte_wrprotect(entry);
905 entry = pte_mkclean(entry);
906 set_pte_at(mm, address, pte, entry);
910 pte_unmap_unlock(pte, ptl);
913 mmu_notifier_invalidate_page(mm, address);
918 static int page_mkclean_file(struct address_space *mapping, struct page *page)
920 pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT);
921 struct vm_area_struct *vma;
924 BUG_ON(PageAnon(page));
926 mutex_lock(&mapping->i_mmap_mutex);
927 vma_interval_tree_foreach(vma, &mapping->i_mmap, pgoff, pgoff) {
928 if (vma->vm_flags & VM_SHARED) {
929 unsigned long address = vma_address(page, vma);
930 ret += page_mkclean_one(page, vma, address);
933 mutex_unlock(&mapping->i_mmap_mutex);
937 int page_mkclean(struct page *page)
941 BUG_ON(!PageLocked(page));
943 if (page_mapped(page)) {
944 struct address_space *mapping = page_mapping(page);
946 ret = page_mkclean_file(mapping, page);
951 EXPORT_SYMBOL_GPL(page_mkclean);
954 * page_move_anon_rmap - move a page to our anon_vma
955 * @page: the page to move to our anon_vma
956 * @vma: the vma the page belongs to
957 * @address: the user virtual address mapped
959 * When a page belongs exclusively to one process after a COW event,
960 * that page can be moved into the anon_vma that belongs to just that
961 * process, so the rmap code will not search the parent or sibling
964 void page_move_anon_rmap(struct page *page,
965 struct vm_area_struct *vma, unsigned long address)
967 struct anon_vma *anon_vma = vma->anon_vma;
969 VM_BUG_ON(!PageLocked(page));
970 VM_BUG_ON(!anon_vma);
971 VM_BUG_ON(page->index != linear_page_index(vma, address));
973 anon_vma = (void *) anon_vma + PAGE_MAPPING_ANON;
974 page->mapping = (struct address_space *) anon_vma;
978 * __page_set_anon_rmap - set up new anonymous rmap
979 * @page: Page to add to rmap
980 * @vma: VM area to add page to.
981 * @address: User virtual address of the mapping
982 * @exclusive: the page is exclusively owned by the current process
984 static void __page_set_anon_rmap(struct page *page,
985 struct vm_area_struct *vma, unsigned long address, int exclusive)
987 struct anon_vma *anon_vma = vma->anon_vma;
995 * If the page isn't exclusively mapped into this vma,
996 * we must use the _oldest_ possible anon_vma for the
1000 anon_vma = anon_vma->root;
1002 anon_vma = (void *) anon_vma + PAGE_MAPPING_ANON;
1003 page->mapping = (struct address_space *) anon_vma;
1004 page->index = linear_page_index(vma, address);
1008 * __page_check_anon_rmap - sanity check anonymous rmap addition
1009 * @page: the page to add the mapping to
1010 * @vma: the vm area in which the mapping is added
1011 * @address: the user virtual address mapped
1013 static void __page_check_anon_rmap(struct page *page,
1014 struct vm_area_struct *vma, unsigned long address)
1016 #ifdef CONFIG_DEBUG_VM
1018 * The page's anon-rmap details (mapping and index) are guaranteed to
1019 * be set up correctly at this point.
1021 * We have exclusion against page_add_anon_rmap because the caller
1022 * always holds the page locked, except if called from page_dup_rmap,
1023 * in which case the page is already known to be setup.
1025 * We have exclusion against page_add_new_anon_rmap because those pages
1026 * are initially only visible via the pagetables, and the pte is locked
1027 * over the call to page_add_new_anon_rmap.
1029 BUG_ON(page_anon_vma(page)->root != vma->anon_vma->root);
1030 BUG_ON(page->index != linear_page_index(vma, address));
1035 * page_add_anon_rmap - add pte mapping to an anonymous page
1036 * @page: the page to add the mapping to
1037 * @vma: the vm area in which the mapping is added
1038 * @address: the user virtual address mapped
1040 * The caller needs to hold the pte lock, and the page must be locked in
1041 * the anon_vma case: to serialize mapping,index checking after setting,
1042 * and to ensure that PageAnon is not being upgraded racily to PageKsm
1043 * (but PageKsm is never downgraded to PageAnon).
1045 void page_add_anon_rmap(struct page *page,
1046 struct vm_area_struct *vma, unsigned long address)
1048 do_page_add_anon_rmap(page, vma, address, 0);
1052 * Special version of the above for do_swap_page, which often runs
1053 * into pages that are exclusively owned by the current process.
1054 * Everybody else should continue to use page_add_anon_rmap above.
1056 void do_page_add_anon_rmap(struct page *page,
1057 struct vm_area_struct *vma, unsigned long address, int exclusive)
1059 int first = atomic_inc_and_test(&page->_mapcount);
1061 if (!PageTransHuge(page))
1062 __inc_zone_page_state(page, NR_ANON_PAGES);
1064 __inc_zone_page_state(page,
1065 NR_ANON_TRANSPARENT_HUGEPAGES);
1067 if (unlikely(PageKsm(page)))
1070 VM_BUG_ON(!PageLocked(page));
1071 /* address might be in next vma when migration races vma_adjust */
1073 __page_set_anon_rmap(page, vma, address, exclusive);
1075 __page_check_anon_rmap(page, vma, address);
1079 * page_add_new_anon_rmap - add pte mapping to a new anonymous page
1080 * @page: the page to add the mapping to
1081 * @vma: the vm area in which the mapping is added
1082 * @address: the user virtual address mapped
1084 * Same as page_add_anon_rmap but must only be called on *new* pages.
1085 * This means the inc-and-test can be bypassed.
1086 * Page does not have to be locked.
1088 void page_add_new_anon_rmap(struct page *page,
1089 struct vm_area_struct *vma, unsigned long address)
1091 VM_BUG_ON(address < vma->vm_start || address >= vma->vm_end);
1092 SetPageSwapBacked(page);
1093 atomic_set(&page->_mapcount, 0); /* increment count (starts at -1) */
1094 if (!PageTransHuge(page))
1095 __inc_zone_page_state(page, NR_ANON_PAGES);
1097 __inc_zone_page_state(page, NR_ANON_TRANSPARENT_HUGEPAGES);
1098 __page_set_anon_rmap(page, vma, address, 1);
1099 if (!mlocked_vma_newpage(vma, page))
1100 lru_cache_add_lru(page, LRU_ACTIVE_ANON);
1102 add_page_to_unevictable_list(page);
1106 * page_add_file_rmap - add pte mapping to a file page
1107 * @page: the page to add the mapping to
1109 * The caller needs to hold the pte lock.
1111 void page_add_file_rmap(struct page *page)
1114 unsigned long flags;
1116 mem_cgroup_begin_update_page_stat(page, &locked, &flags);
1117 if (atomic_inc_and_test(&page->_mapcount)) {
1118 __inc_zone_page_state(page, NR_FILE_MAPPED);
1119 mem_cgroup_inc_page_stat(page, MEMCG_NR_FILE_MAPPED);
1121 mem_cgroup_end_update_page_stat(page, &locked, &flags);
1125 * page_remove_rmap - take down pte mapping from a page
1126 * @page: page to remove mapping from
1128 * The caller needs to hold the pte lock.
1130 void page_remove_rmap(struct page *page)
1132 struct address_space *mapping = page_mapping(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 * Now that the last pte has gone, s390 must transfer dirty
1151 * flag from storage key to struct page. We can usually skip
1152 * this if the page is anon, so about to be freed; but perhaps
1153 * not if it's in swapcache - there might be another pte slot
1154 * containing the swap entry, but page not yet written to swap.
1156 * And we can skip it on file pages, so long as the filesystem
1157 * participates in dirty tracking; but need to catch shm and tmpfs
1158 * and ramfs pages which have been modified since creation by read
1161 * Note that mapping must be decided above, before decrementing
1162 * mapcount (which luckily provides a barrier): once page is unmapped,
1163 * it could be truncated and page->mapping reset to NULL at any moment.
1164 * Note also that we are relying on page_mapping(page) to set mapping
1165 * to &swapper_space when PageSwapCache(page).
1167 if (mapping && !mapping_cap_account_dirty(mapping) &&
1168 page_test_and_clear_dirty(page_to_pfn(page), 1))
1169 set_page_dirty(page);
1171 * Hugepages are not counted in NR_ANON_PAGES nor NR_FILE_MAPPED
1172 * and not charged by memcg for now.
1174 if (unlikely(PageHuge(page)))
1177 mem_cgroup_uncharge_page(page);
1178 if (!PageTransHuge(page))
1179 __dec_zone_page_state(page, NR_ANON_PAGES);
1181 __dec_zone_page_state(page,
1182 NR_ANON_TRANSPARENT_HUGEPAGES);
1184 __dec_zone_page_state(page, NR_FILE_MAPPED);
1185 mem_cgroup_dec_page_stat(page, MEMCG_NR_FILE_MAPPED);
1186 mem_cgroup_end_update_page_stat(page, &locked, &flags);
1188 if (unlikely(PageMlocked(page)))
1189 clear_page_mlock(page);
1191 * It would be tidy to reset the PageAnon mapping here,
1192 * but that might overwrite a racing page_add_anon_rmap
1193 * which increments mapcount after us but sets mapping
1194 * before us: so leave the reset to free_hot_cold_page,
1195 * and remember that it's only reliable while mapped.
1196 * Leaving it set also helps swapoff to reinstate ptes
1197 * faster for those pages still in swapcache.
1202 mem_cgroup_end_update_page_stat(page, &locked, &flags);
1206 * Subfunctions of try_to_unmap: try_to_unmap_one called
1207 * repeatedly from try_to_unmap_ksm, try_to_unmap_anon or try_to_unmap_file.
1209 int try_to_unmap_one(struct page *page, struct vm_area_struct *vma,
1210 unsigned long address, enum ttu_flags flags)
1212 struct mm_struct *mm = vma->vm_mm;
1216 int ret = SWAP_AGAIN;
1218 pte = page_check_address(page, mm, address, &ptl, 0);
1223 * If the page is mlock()d, we cannot swap it out.
1224 * If it's recently referenced (perhaps page_referenced
1225 * skipped over this mm) then we should reactivate it.
1227 if (!(flags & TTU_IGNORE_MLOCK)) {
1228 if (vma->vm_flags & VM_LOCKED)
1231 if (TTU_ACTION(flags) == TTU_MUNLOCK)
1234 if (!(flags & TTU_IGNORE_ACCESS)) {
1235 if (ptep_clear_flush_young_notify(vma, address, pte)) {
1241 /* Nuke the page table entry. */
1242 flush_cache_page(vma, address, page_to_pfn(page));
1243 pteval = ptep_clear_flush(vma, address, pte);
1245 /* Move the dirty bit to the physical page now the pte is gone. */
1246 if (pte_dirty(pteval))
1247 set_page_dirty(page);
1249 /* Update high watermark before we lower rss */
1250 update_hiwater_rss(mm);
1252 if (PageHWPoison(page) && !(flags & TTU_IGNORE_HWPOISON)) {
1254 dec_mm_counter(mm, MM_ANONPAGES);
1256 dec_mm_counter(mm, MM_FILEPAGES);
1257 set_pte_at(mm, address, pte,
1258 swp_entry_to_pte(make_hwpoison_entry(page)));
1259 } else if (PageAnon(page)) {
1260 swp_entry_t entry = { .val = page_private(page) };
1262 if (PageSwapCache(page)) {
1264 * Store the swap location in the pte.
1265 * See handle_pte_fault() ...
1267 if (swap_duplicate(entry) < 0) {
1268 set_pte_at(mm, address, pte, pteval);
1272 if (list_empty(&mm->mmlist)) {
1273 spin_lock(&mmlist_lock);
1274 if (list_empty(&mm->mmlist))
1275 list_add(&mm->mmlist, &init_mm.mmlist);
1276 spin_unlock(&mmlist_lock);
1278 dec_mm_counter(mm, MM_ANONPAGES);
1279 inc_mm_counter(mm, MM_SWAPENTS);
1280 } else if (IS_ENABLED(CONFIG_MIGRATION)) {
1282 * Store the pfn of the page in a special migration
1283 * pte. do_swap_page() will wait until the migration
1284 * pte is removed and then restart fault handling.
1286 BUG_ON(TTU_ACTION(flags) != TTU_MIGRATION);
1287 entry = make_migration_entry(page, pte_write(pteval));
1289 set_pte_at(mm, address, pte, swp_entry_to_pte(entry));
1290 BUG_ON(pte_file(*pte));
1291 } else if (IS_ENABLED(CONFIG_MIGRATION) &&
1292 (TTU_ACTION(flags) == TTU_MIGRATION)) {
1293 /* Establish migration entry for a file page */
1295 entry = make_migration_entry(page, pte_write(pteval));
1296 set_pte_at(mm, address, pte, swp_entry_to_pte(entry));
1298 dec_mm_counter(mm, MM_FILEPAGES);
1300 page_remove_rmap(page);
1301 page_cache_release(page);
1304 pte_unmap_unlock(pte, ptl);
1305 if (ret != SWAP_FAIL)
1306 mmu_notifier_invalidate_page(mm, address);
1311 pte_unmap_unlock(pte, ptl);
1315 * We need mmap_sem locking, Otherwise VM_LOCKED check makes
1316 * unstable result and race. Plus, We can't wait here because
1317 * we now hold anon_vma->mutex or mapping->i_mmap_mutex.
1318 * if trylock failed, the page remain in evictable lru and later
1319 * vmscan could retry to move the page to unevictable lru if the
1320 * page is actually mlocked.
1322 if (down_read_trylock(&vma->vm_mm->mmap_sem)) {
1323 if (vma->vm_flags & VM_LOCKED) {
1324 mlock_vma_page(page);
1327 up_read(&vma->vm_mm->mmap_sem);
1333 * objrmap doesn't work for nonlinear VMAs because the assumption that
1334 * offset-into-file correlates with offset-into-virtual-addresses does not hold.
1335 * Consequently, given a particular page and its ->index, we cannot locate the
1336 * ptes which are mapping that page without an exhaustive linear search.
1338 * So what this code does is a mini "virtual scan" of each nonlinear VMA which
1339 * maps the file to which the target page belongs. The ->vm_private_data field
1340 * holds the current cursor into that scan. Successive searches will circulate
1341 * around the vma's virtual address space.
1343 * So as more replacement pressure is applied to the pages in a nonlinear VMA,
1344 * more scanning pressure is placed against them as well. Eventually pages
1345 * will become fully unmapped and are eligible for eviction.
1347 * For very sparsely populated VMAs this is a little inefficient - chances are
1348 * there there won't be many ptes located within the scan cluster. In this case
1349 * maybe we could scan further - to the end of the pte page, perhaps.
1351 * Mlocked pages: check VM_LOCKED under mmap_sem held for read, if we can
1352 * acquire it without blocking. If vma locked, mlock the pages in the cluster,
1353 * rather than unmapping them. If we encounter the "check_page" that vmscan is
1354 * trying to unmap, return SWAP_MLOCK, else default SWAP_AGAIN.
1356 #define CLUSTER_SIZE min(32*PAGE_SIZE, PMD_SIZE)
1357 #define CLUSTER_MASK (~(CLUSTER_SIZE - 1))
1359 static int try_to_unmap_cluster(unsigned long cursor, unsigned int *mapcount,
1360 struct vm_area_struct *vma, struct page *check_page)
1362 struct mm_struct *mm = vma->vm_mm;
1368 unsigned long address;
1369 unsigned long mmun_start; /* For mmu_notifiers */
1370 unsigned long mmun_end; /* For mmu_notifiers */
1372 int ret = SWAP_AGAIN;
1375 address = (vma->vm_start + cursor) & CLUSTER_MASK;
1376 end = address + CLUSTER_SIZE;
1377 if (address < vma->vm_start)
1378 address = vma->vm_start;
1379 if (end > vma->vm_end)
1382 pmd = mm_find_pmd(mm, address);
1386 mmun_start = address;
1388 mmu_notifier_invalidate_range_start(mm, mmun_start, mmun_end);
1391 * If we can acquire the mmap_sem for read, and vma is VM_LOCKED,
1392 * keep the sem while scanning the cluster for mlocking pages.
1394 if (down_read_trylock(&vma->vm_mm->mmap_sem)) {
1395 locked_vma = (vma->vm_flags & VM_LOCKED);
1397 up_read(&vma->vm_mm->mmap_sem); /* don't need it */
1400 pte = pte_offset_map_lock(mm, pmd, address, &ptl);
1402 /* Update high watermark before we lower rss */
1403 update_hiwater_rss(mm);
1405 for (; address < end; pte++, address += PAGE_SIZE) {
1406 if (!pte_present(*pte))
1408 page = vm_normal_page(vma, address, *pte);
1409 BUG_ON(!page || PageAnon(page));
1412 mlock_vma_page(page); /* no-op if already mlocked */
1413 if (page == check_page)
1415 continue; /* don't unmap */
1418 if (ptep_clear_flush_young_notify(vma, address, pte))
1421 /* Nuke the page table entry. */
1422 flush_cache_page(vma, address, pte_pfn(*pte));
1423 pteval = ptep_clear_flush(vma, address, pte);
1425 /* If nonlinear, store the file page offset in the pte. */
1426 if (page->index != linear_page_index(vma, address))
1427 set_pte_at(mm, address, pte, pgoff_to_pte(page->index));
1429 /* Move the dirty bit to the physical page now the pte is gone. */
1430 if (pte_dirty(pteval))
1431 set_page_dirty(page);
1433 page_remove_rmap(page);
1434 page_cache_release(page);
1435 dec_mm_counter(mm, MM_FILEPAGES);
1438 pte_unmap_unlock(pte - 1, ptl);
1439 mmu_notifier_invalidate_range_end(mm, mmun_start, mmun_end);
1441 up_read(&vma->vm_mm->mmap_sem);
1445 bool is_vma_temporary_stack(struct vm_area_struct *vma)
1447 int maybe_stack = vma->vm_flags & (VM_GROWSDOWN | VM_GROWSUP);
1452 if ((vma->vm_flags & VM_STACK_INCOMPLETE_SETUP) ==
1453 VM_STACK_INCOMPLETE_SETUP)
1460 * try_to_unmap_anon - unmap or unlock anonymous page using the object-based
1462 * @page: the page to unmap/unlock
1463 * @flags: action and flags
1465 * Find all the mappings of a page using the mapping pointer and the vma chains
1466 * contained in the anon_vma struct it points to.
1468 * This function is only called from try_to_unmap/try_to_munlock for
1470 * When called from try_to_munlock(), the mmap_sem of the mm containing the vma
1471 * where the page was found will be held for write. So, we won't recheck
1472 * vm_flags for that VMA. That should be OK, because that vma shouldn't be
1475 static int try_to_unmap_anon(struct page *page, enum ttu_flags flags)
1477 struct anon_vma *anon_vma;
1479 struct anon_vma_chain *avc;
1480 int ret = SWAP_AGAIN;
1482 anon_vma = page_lock_anon_vma(page);
1486 pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT);
1487 anon_vma_interval_tree_foreach(avc, &anon_vma->rb_root, pgoff, pgoff) {
1488 struct vm_area_struct *vma = avc->vma;
1489 unsigned long address;
1492 * During exec, a temporary VMA is setup and later moved.
1493 * The VMA is moved under the anon_vma lock but not the
1494 * page tables leading to a race where migration cannot
1495 * find the migration ptes. Rather than increasing the
1496 * locking requirements of exec(), migration skips
1497 * temporary VMAs until after exec() completes.
1499 if (IS_ENABLED(CONFIG_MIGRATION) && (flags & TTU_MIGRATION) &&
1500 is_vma_temporary_stack(vma))
1503 address = vma_address(page, vma);
1504 ret = try_to_unmap_one(page, vma, address, flags);
1505 if (ret != SWAP_AGAIN || !page_mapped(page))
1509 page_unlock_anon_vma(anon_vma);
1514 * try_to_unmap_file - unmap/unlock file page using the object-based rmap method
1515 * @page: the page to unmap/unlock
1516 * @flags: action and flags
1518 * Find all the mappings of a page using the mapping pointer and the vma chains
1519 * contained in the address_space struct it points to.
1521 * This function is only called from try_to_unmap/try_to_munlock for
1522 * object-based pages.
1523 * When called from try_to_munlock(), the mmap_sem of the mm containing the vma
1524 * where the page was found will be held for write. So, we won't recheck
1525 * vm_flags for that VMA. That should be OK, because that vma shouldn't be
1528 static int try_to_unmap_file(struct page *page, enum ttu_flags flags)
1530 struct address_space *mapping = page->mapping;
1531 pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT);
1532 struct vm_area_struct *vma;
1533 int ret = SWAP_AGAIN;
1534 unsigned long cursor;
1535 unsigned long max_nl_cursor = 0;
1536 unsigned long max_nl_size = 0;
1537 unsigned int mapcount;
1539 mutex_lock(&mapping->i_mmap_mutex);
1540 vma_interval_tree_foreach(vma, &mapping->i_mmap, pgoff, pgoff) {
1541 unsigned long address = vma_address(page, vma);
1542 ret = try_to_unmap_one(page, vma, address, flags);
1543 if (ret != SWAP_AGAIN || !page_mapped(page))
1547 if (list_empty(&mapping->i_mmap_nonlinear))
1551 * We don't bother to try to find the munlocked page in nonlinears.
1552 * It's costly. Instead, later, page reclaim logic may call
1553 * try_to_unmap(TTU_MUNLOCK) and recover PG_mlocked lazily.
1555 if (TTU_ACTION(flags) == TTU_MUNLOCK)
1558 list_for_each_entry(vma, &mapping->i_mmap_nonlinear,
1560 cursor = (unsigned long) vma->vm_private_data;
1561 if (cursor > max_nl_cursor)
1562 max_nl_cursor = cursor;
1563 cursor = vma->vm_end - vma->vm_start;
1564 if (cursor > max_nl_size)
1565 max_nl_size = cursor;
1568 if (max_nl_size == 0) { /* all nonlinears locked or reserved ? */
1574 * We don't try to search for this page in the nonlinear vmas,
1575 * and page_referenced wouldn't have found it anyway. Instead
1576 * just walk the nonlinear vmas trying to age and unmap some.
1577 * The mapcount of the page we came in with is irrelevant,
1578 * but even so use it as a guide to how hard we should try?
1580 mapcount = page_mapcount(page);
1585 max_nl_size = (max_nl_size + CLUSTER_SIZE - 1) & CLUSTER_MASK;
1586 if (max_nl_cursor == 0)
1587 max_nl_cursor = CLUSTER_SIZE;
1590 list_for_each_entry(vma, &mapping->i_mmap_nonlinear,
1592 cursor = (unsigned long) vma->vm_private_data;
1593 while ( cursor < max_nl_cursor &&
1594 cursor < vma->vm_end - vma->vm_start) {
1595 if (try_to_unmap_cluster(cursor, &mapcount,
1596 vma, page) == SWAP_MLOCK)
1598 cursor += CLUSTER_SIZE;
1599 vma->vm_private_data = (void *) cursor;
1600 if ((int)mapcount <= 0)
1603 vma->vm_private_data = (void *) max_nl_cursor;
1606 max_nl_cursor += CLUSTER_SIZE;
1607 } while (max_nl_cursor <= max_nl_size);
1610 * Don't loop forever (perhaps all the remaining pages are
1611 * in locked vmas). Reset cursor on all unreserved nonlinear
1612 * vmas, now forgetting on which ones it had fallen behind.
1614 list_for_each_entry(vma, &mapping->i_mmap_nonlinear, shared.nonlinear)
1615 vma->vm_private_data = NULL;
1617 mutex_unlock(&mapping->i_mmap_mutex);
1622 * try_to_unmap - try to remove all page table mappings to a page
1623 * @page: the page to get unmapped
1624 * @flags: action and flags
1626 * Tries to remove all the page table entries which are mapping this
1627 * page, used in the pageout path. Caller must hold the page lock.
1628 * Return values are:
1630 * SWAP_SUCCESS - we succeeded in removing all mappings
1631 * SWAP_AGAIN - we missed a mapping, try again later
1632 * SWAP_FAIL - the page is unswappable
1633 * SWAP_MLOCK - page is mlocked.
1635 int try_to_unmap(struct page *page, enum ttu_flags flags)
1639 BUG_ON(!PageLocked(page));
1640 VM_BUG_ON(!PageHuge(page) && PageTransHuge(page));
1642 if (unlikely(PageKsm(page)))
1643 ret = try_to_unmap_ksm(page, flags);
1644 else if (PageAnon(page))
1645 ret = try_to_unmap_anon(page, flags);
1647 ret = try_to_unmap_file(page, flags);
1648 if (ret != SWAP_MLOCK && !page_mapped(page))
1654 * try_to_munlock - try to munlock a page
1655 * @page: the page to be munlocked
1657 * Called from munlock code. Checks all of the VMAs mapping the page
1658 * to make sure nobody else has this page mlocked. The page will be
1659 * returned with PG_mlocked cleared if no other vmas have it mlocked.
1661 * Return values are:
1663 * SWAP_AGAIN - no vma is holding page mlocked, or,
1664 * SWAP_AGAIN - page mapped in mlocked vma -- couldn't acquire mmap sem
1665 * SWAP_FAIL - page cannot be located at present
1666 * SWAP_MLOCK - page is now mlocked.
1668 int try_to_munlock(struct page *page)
1670 VM_BUG_ON(!PageLocked(page) || PageLRU(page));
1672 if (unlikely(PageKsm(page)))
1673 return try_to_unmap_ksm(page, TTU_MUNLOCK);
1674 else if (PageAnon(page))
1675 return try_to_unmap_anon(page, TTU_MUNLOCK);
1677 return try_to_unmap_file(page, TTU_MUNLOCK);
1680 void __put_anon_vma(struct anon_vma *anon_vma)
1682 struct anon_vma *root = anon_vma->root;
1684 if (root != anon_vma && atomic_dec_and_test(&root->refcount))
1685 anon_vma_free(root);
1687 anon_vma_free(anon_vma);
1690 #ifdef CONFIG_MIGRATION
1692 * rmap_walk() and its helpers rmap_walk_anon() and rmap_walk_file():
1693 * Called by migrate.c to remove migration ptes, but might be used more later.
1695 static int rmap_walk_anon(struct page *page, int (*rmap_one)(struct page *,
1696 struct vm_area_struct *, unsigned long, void *), void *arg)
1698 struct anon_vma *anon_vma;
1699 pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT);
1700 struct anon_vma_chain *avc;
1701 int ret = SWAP_AGAIN;
1704 * Note: remove_migration_ptes() cannot use page_lock_anon_vma()
1705 * because that depends on page_mapped(); but not all its usages
1706 * are holding mmap_sem. Users without mmap_sem are required to
1707 * take a reference count to prevent the anon_vma disappearing
1709 anon_vma = page_anon_vma(page);
1712 anon_vma_lock(anon_vma);
1713 anon_vma_interval_tree_foreach(avc, &anon_vma->rb_root, pgoff, pgoff) {
1714 struct vm_area_struct *vma = avc->vma;
1715 unsigned long address = vma_address(page, vma);
1716 ret = rmap_one(page, vma, address, arg);
1717 if (ret != SWAP_AGAIN)
1720 anon_vma_unlock(anon_vma);
1724 static int rmap_walk_file(struct page *page, int (*rmap_one)(struct page *,
1725 struct vm_area_struct *, unsigned long, void *), void *arg)
1727 struct address_space *mapping = page->mapping;
1728 pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT);
1729 struct vm_area_struct *vma;
1730 int ret = SWAP_AGAIN;
1734 mutex_lock(&mapping->i_mmap_mutex);
1735 vma_interval_tree_foreach(vma, &mapping->i_mmap, pgoff, pgoff) {
1736 unsigned long address = vma_address(page, vma);
1737 ret = rmap_one(page, vma, address, arg);
1738 if (ret != SWAP_AGAIN)
1742 * No nonlinear handling: being always shared, nonlinear vmas
1743 * never contain migration ptes. Decide what to do about this
1744 * limitation to linear when we need rmap_walk() on nonlinear.
1746 mutex_unlock(&mapping->i_mmap_mutex);
1750 int rmap_walk(struct page *page, int (*rmap_one)(struct page *,
1751 struct vm_area_struct *, unsigned long, void *), void *arg)
1753 VM_BUG_ON(!PageLocked(page));
1755 if (unlikely(PageKsm(page)))
1756 return rmap_walk_ksm(page, rmap_one, arg);
1757 else if (PageAnon(page))
1758 return rmap_walk_anon(page, rmap_one, arg);
1760 return rmap_walk_file(page, rmap_one, arg);
1762 #endif /* CONFIG_MIGRATION */
1764 #ifdef CONFIG_HUGETLB_PAGE
1766 * The following three functions are for anonymous (private mapped) hugepages.
1767 * Unlike common anonymous pages, anonymous hugepages have no accounting code
1768 * and no lru code, because we handle hugepages differently from common pages.
1770 static void __hugepage_set_anon_rmap(struct page *page,
1771 struct vm_area_struct *vma, unsigned long address, int exclusive)
1773 struct anon_vma *anon_vma = vma->anon_vma;
1780 anon_vma = anon_vma->root;
1782 anon_vma = (void *) anon_vma + PAGE_MAPPING_ANON;
1783 page->mapping = (struct address_space *) anon_vma;
1784 page->index = linear_page_index(vma, address);
1787 void hugepage_add_anon_rmap(struct page *page,
1788 struct vm_area_struct *vma, unsigned long address)
1790 struct anon_vma *anon_vma = vma->anon_vma;
1793 BUG_ON(!PageLocked(page));
1795 /* address might be in next vma when migration races vma_adjust */
1796 first = atomic_inc_and_test(&page->_mapcount);
1798 __hugepage_set_anon_rmap(page, vma, address, 0);
1801 void hugepage_add_new_anon_rmap(struct page *page,
1802 struct vm_area_struct *vma, unsigned long address)
1804 BUG_ON(address < vma->vm_start || address >= vma->vm_end);
1805 atomic_set(&page->_mapcount, 0);
1806 __hugepage_set_anon_rmap(page, vma, address, 1);
1808 #endif /* CONFIG_HUGETLB_PAGE */