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_rwsem
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 * mem_cgroup_{begin,end}_page_stat (memcg->move_lock)
34 * mapping->tree_lock (widely used)
35 * inode->i_lock (in set_page_dirty's __mark_inode_dirty)
36 * bdi.wb->list_lock (in set_page_dirty's __mark_inode_dirty)
37 * sb_lock (within inode_lock in fs/fs-writeback.c)
38 * mapping->tree_lock (widely used, in set_page_dirty,
39 * in arch-dependent flush_dcache_mmap_lock,
40 * within bdi.wb->list_lock in __sync_single_inode)
42 * anon_vma->rwsem,mapping->i_mutex (memory_failure, collect_procs_anon)
48 #include <linux/pagemap.h>
49 #include <linux/swap.h>
50 #include <linux/swapops.h>
51 #include <linux/slab.h>
52 #include <linux/init.h>
53 #include <linux/ksm.h>
54 #include <linux/rmap.h>
55 #include <linux/rcupdate.h>
56 #include <linux/export.h>
57 #include <linux/memcontrol.h>
58 #include <linux/mmu_notifier.h>
59 #include <linux/migrate.h>
60 #include <linux/hugetlb.h>
61 #include <linux/backing-dev.h>
62 #include <linux/page_idle.h>
64 #include <asm/tlbflush.h>
66 #include <trace/events/tlb.h>
70 static struct kmem_cache *anon_vma_cachep;
71 static struct kmem_cache *anon_vma_chain_cachep;
73 static inline struct anon_vma *anon_vma_alloc(void)
75 struct anon_vma *anon_vma;
77 anon_vma = kmem_cache_alloc(anon_vma_cachep, GFP_KERNEL);
79 atomic_set(&anon_vma->refcount, 1);
80 anon_vma->degree = 1; /* Reference for first vma */
81 anon_vma->parent = anon_vma;
83 * Initialise the anon_vma root to point to itself. If called
84 * from fork, the root will be reset to the parents anon_vma.
86 anon_vma->root = anon_vma;
92 static inline void anon_vma_free(struct anon_vma *anon_vma)
94 VM_BUG_ON(atomic_read(&anon_vma->refcount));
97 * Synchronize against page_lock_anon_vma_read() such that
98 * we can safely hold the lock without the anon_vma getting
101 * Relies on the full mb implied by the atomic_dec_and_test() from
102 * put_anon_vma() against the acquire barrier implied by
103 * down_read_trylock() from page_lock_anon_vma_read(). This orders:
105 * page_lock_anon_vma_read() VS put_anon_vma()
106 * down_read_trylock() atomic_dec_and_test()
108 * atomic_read() rwsem_is_locked()
110 * LOCK should suffice since the actual taking of the lock must
111 * happen _before_ what follows.
114 if (rwsem_is_locked(&anon_vma->root->rwsem)) {
115 anon_vma_lock_write(anon_vma);
116 anon_vma_unlock_write(anon_vma);
119 kmem_cache_free(anon_vma_cachep, anon_vma);
122 static inline struct anon_vma_chain *anon_vma_chain_alloc(gfp_t gfp)
124 return kmem_cache_alloc(anon_vma_chain_cachep, gfp);
127 static void anon_vma_chain_free(struct anon_vma_chain *anon_vma_chain)
129 kmem_cache_free(anon_vma_chain_cachep, anon_vma_chain);
132 static void anon_vma_chain_link(struct vm_area_struct *vma,
133 struct anon_vma_chain *avc,
134 struct anon_vma *anon_vma)
137 avc->anon_vma = anon_vma;
138 list_add(&avc->same_vma, &vma->anon_vma_chain);
139 anon_vma_interval_tree_insert(avc, &anon_vma->rb_root);
143 * anon_vma_prepare - attach an anon_vma to a memory region
144 * @vma: the memory region in question
146 * This makes sure the memory mapping described by 'vma' has
147 * an 'anon_vma' attached to it, so that we can associate the
148 * anonymous pages mapped into it with that anon_vma.
150 * The common case will be that we already have one, but if
151 * not we either need to find an adjacent mapping that we
152 * can re-use the anon_vma from (very common when the only
153 * reason for splitting a vma has been mprotect()), or we
154 * allocate a new one.
156 * Anon-vma allocations are very subtle, because we may have
157 * optimistically looked up an anon_vma in page_lock_anon_vma_read()
158 * and that may actually touch the spinlock even in the newly
159 * allocated vma (it depends on RCU to make sure that the
160 * anon_vma isn't actually destroyed).
162 * As a result, we need to do proper anon_vma locking even
163 * for the new allocation. At the same time, we do not want
164 * to do any locking for the common case of already having
167 * This must be called with the mmap_sem held for reading.
169 int anon_vma_prepare(struct vm_area_struct *vma)
171 struct anon_vma *anon_vma = vma->anon_vma;
172 struct anon_vma_chain *avc;
175 if (unlikely(!anon_vma)) {
176 struct mm_struct *mm = vma->vm_mm;
177 struct anon_vma *allocated;
179 avc = anon_vma_chain_alloc(GFP_KERNEL);
183 anon_vma = find_mergeable_anon_vma(vma);
186 anon_vma = anon_vma_alloc();
187 if (unlikely(!anon_vma))
188 goto out_enomem_free_avc;
189 allocated = anon_vma;
192 anon_vma_lock_write(anon_vma);
193 /* page_table_lock to protect against threads */
194 spin_lock(&mm->page_table_lock);
195 if (likely(!vma->anon_vma)) {
196 vma->anon_vma = anon_vma;
197 anon_vma_chain_link(vma, avc, anon_vma);
198 /* vma reference or self-parent link for new root */
203 spin_unlock(&mm->page_table_lock);
204 anon_vma_unlock_write(anon_vma);
206 if (unlikely(allocated))
207 put_anon_vma(allocated);
209 anon_vma_chain_free(avc);
214 anon_vma_chain_free(avc);
220 * This is a useful helper function for locking the anon_vma root as
221 * we traverse the vma->anon_vma_chain, looping over anon_vma's that
224 * Such anon_vma's should have the same root, so you'd expect to see
225 * just a single mutex_lock for the whole traversal.
227 static inline struct anon_vma *lock_anon_vma_root(struct anon_vma *root, struct anon_vma *anon_vma)
229 struct anon_vma *new_root = anon_vma->root;
230 if (new_root != root) {
231 if (WARN_ON_ONCE(root))
232 up_write(&root->rwsem);
234 down_write(&root->rwsem);
239 static inline void unlock_anon_vma_root(struct anon_vma *root)
242 up_write(&root->rwsem);
246 * Attach the anon_vmas from src to dst.
247 * Returns 0 on success, -ENOMEM on failure.
249 * If dst->anon_vma is NULL this function tries to find and reuse existing
250 * anon_vma which has no vmas and only one child anon_vma. This prevents
251 * degradation of anon_vma hierarchy to endless linear chain in case of
252 * constantly forking task. On the other hand, an anon_vma with more than one
253 * child isn't reused even if there was no alive vma, thus rmap walker has a
254 * good chance of avoiding scanning the whole hierarchy when it searches where
257 int anon_vma_clone(struct vm_area_struct *dst, struct vm_area_struct *src)
259 struct anon_vma_chain *avc, *pavc;
260 struct anon_vma *root = NULL;
262 list_for_each_entry_reverse(pavc, &src->anon_vma_chain, same_vma) {
263 struct anon_vma *anon_vma;
265 avc = anon_vma_chain_alloc(GFP_NOWAIT | __GFP_NOWARN);
266 if (unlikely(!avc)) {
267 unlock_anon_vma_root(root);
269 avc = anon_vma_chain_alloc(GFP_KERNEL);
273 anon_vma = pavc->anon_vma;
274 root = lock_anon_vma_root(root, anon_vma);
275 anon_vma_chain_link(dst, avc, anon_vma);
278 * Reuse existing anon_vma if its degree lower than two,
279 * that means it has no vma and only one anon_vma child.
281 * Do not chose parent anon_vma, otherwise first child
282 * will always reuse it. Root anon_vma is never reused:
283 * it has self-parent reference and at least one child.
285 if (!dst->anon_vma && anon_vma != src->anon_vma &&
286 anon_vma->degree < 2)
287 dst->anon_vma = anon_vma;
290 dst->anon_vma->degree++;
291 unlock_anon_vma_root(root);
296 * dst->anon_vma is dropped here otherwise its degree can be incorrectly
297 * decremented in unlink_anon_vmas().
298 * We can safely do this because callers of anon_vma_clone() don't care
299 * about dst->anon_vma if anon_vma_clone() failed.
301 dst->anon_vma = NULL;
302 unlink_anon_vmas(dst);
307 * Attach vma to its own anon_vma, as well as to the anon_vmas that
308 * the corresponding VMA in the parent process is attached to.
309 * Returns 0 on success, non-zero on failure.
311 int anon_vma_fork(struct vm_area_struct *vma, struct vm_area_struct *pvma)
313 struct anon_vma_chain *avc;
314 struct anon_vma *anon_vma;
317 /* Don't bother if the parent process has no anon_vma here. */
321 /* Drop inherited anon_vma, we'll reuse existing or allocate new. */
322 vma->anon_vma = NULL;
325 * First, attach the new VMA to the parent VMA's anon_vmas,
326 * so rmap can find non-COWed pages in child processes.
328 error = anon_vma_clone(vma, pvma);
332 /* An existing anon_vma has been reused, all done then. */
336 /* Then add our own anon_vma. */
337 anon_vma = anon_vma_alloc();
340 avc = anon_vma_chain_alloc(GFP_KERNEL);
342 goto out_error_free_anon_vma;
345 * The root anon_vma's spinlock is the lock actually used when we
346 * lock any of the anon_vmas in this anon_vma tree.
348 anon_vma->root = pvma->anon_vma->root;
349 anon_vma->parent = pvma->anon_vma;
351 * With refcounts, an anon_vma can stay around longer than the
352 * process it belongs to. The root anon_vma needs to be pinned until
353 * this anon_vma is freed, because the lock lives in the root.
355 get_anon_vma(anon_vma->root);
356 /* Mark this anon_vma as the one where our new (COWed) pages go. */
357 vma->anon_vma = anon_vma;
358 anon_vma_lock_write(anon_vma);
359 anon_vma_chain_link(vma, avc, anon_vma);
360 anon_vma->parent->degree++;
361 anon_vma_unlock_write(anon_vma);
365 out_error_free_anon_vma:
366 put_anon_vma(anon_vma);
368 unlink_anon_vmas(vma);
372 void unlink_anon_vmas(struct vm_area_struct *vma)
374 struct anon_vma_chain *avc, *next;
375 struct anon_vma *root = NULL;
378 * Unlink each anon_vma chained to the VMA. This list is ordered
379 * from newest to oldest, ensuring the root anon_vma gets freed last.
381 list_for_each_entry_safe(avc, next, &vma->anon_vma_chain, same_vma) {
382 struct anon_vma *anon_vma = avc->anon_vma;
384 root = lock_anon_vma_root(root, anon_vma);
385 anon_vma_interval_tree_remove(avc, &anon_vma->rb_root);
388 * Leave empty anon_vmas on the list - we'll need
389 * to free them outside the lock.
391 if (RB_EMPTY_ROOT(&anon_vma->rb_root)) {
392 anon_vma->parent->degree--;
396 list_del(&avc->same_vma);
397 anon_vma_chain_free(avc);
400 vma->anon_vma->degree--;
401 unlock_anon_vma_root(root);
404 * Iterate the list once more, it now only contains empty and unlinked
405 * anon_vmas, destroy them. Could not do before due to __put_anon_vma()
406 * needing to write-acquire the anon_vma->root->rwsem.
408 list_for_each_entry_safe(avc, next, &vma->anon_vma_chain, same_vma) {
409 struct anon_vma *anon_vma = avc->anon_vma;
411 BUG_ON(anon_vma->degree);
412 put_anon_vma(anon_vma);
414 list_del(&avc->same_vma);
415 anon_vma_chain_free(avc);
419 static void anon_vma_ctor(void *data)
421 struct anon_vma *anon_vma = data;
423 init_rwsem(&anon_vma->rwsem);
424 atomic_set(&anon_vma->refcount, 0);
425 anon_vma->rb_root = RB_ROOT;
428 void __init anon_vma_init(void)
430 anon_vma_cachep = kmem_cache_create("anon_vma", sizeof(struct anon_vma),
431 0, SLAB_DESTROY_BY_RCU|SLAB_PANIC, anon_vma_ctor);
432 anon_vma_chain_cachep = KMEM_CACHE(anon_vma_chain, SLAB_PANIC);
436 * Getting a lock on a stable anon_vma from a page off the LRU is tricky!
438 * Since there is no serialization what so ever against page_remove_rmap()
439 * the best this function can do is return a locked anon_vma that might
440 * have been relevant to this page.
442 * The page might have been remapped to a different anon_vma or the anon_vma
443 * returned may already be freed (and even reused).
445 * In case it was remapped to a different anon_vma, the new anon_vma will be a
446 * child of the old anon_vma, and the anon_vma lifetime rules will therefore
447 * ensure that any anon_vma obtained from the page will still be valid for as
448 * long as we observe page_mapped() [ hence all those page_mapped() tests ].
450 * All users of this function must be very careful when walking the anon_vma
451 * chain and verify that the page in question is indeed mapped in it
452 * [ something equivalent to page_mapped_in_vma() ].
454 * Since anon_vma's slab is DESTROY_BY_RCU and we know from page_remove_rmap()
455 * that the anon_vma pointer from page->mapping is valid if there is a
456 * mapcount, we can dereference the anon_vma after observing those.
458 struct anon_vma *page_get_anon_vma(struct page *page)
460 struct anon_vma *anon_vma = NULL;
461 unsigned long anon_mapping;
464 anon_mapping = (unsigned long)READ_ONCE(page->mapping);
465 if ((anon_mapping & PAGE_MAPPING_FLAGS) != PAGE_MAPPING_ANON)
467 if (!page_mapped(page))
470 anon_vma = (struct anon_vma *) (anon_mapping - PAGE_MAPPING_ANON);
471 if (!atomic_inc_not_zero(&anon_vma->refcount)) {
477 * If this page is still mapped, then its anon_vma cannot have been
478 * freed. But if it has been unmapped, we have no security against the
479 * anon_vma structure being freed and reused (for another anon_vma:
480 * SLAB_DESTROY_BY_RCU guarantees that - so the atomic_inc_not_zero()
481 * above cannot corrupt).
483 if (!page_mapped(page)) {
485 put_anon_vma(anon_vma);
495 * Similar to page_get_anon_vma() except it locks the anon_vma.
497 * Its a little more complex as it tries to keep the fast path to a single
498 * atomic op -- the trylock. If we fail the trylock, we fall back to getting a
499 * reference like with page_get_anon_vma() and then block on the mutex.
501 struct anon_vma *page_lock_anon_vma_read(struct page *page)
503 struct anon_vma *anon_vma = NULL;
504 struct anon_vma *root_anon_vma;
505 unsigned long anon_mapping;
508 anon_mapping = (unsigned long)READ_ONCE(page->mapping);
509 if ((anon_mapping & PAGE_MAPPING_FLAGS) != PAGE_MAPPING_ANON)
511 if (!page_mapped(page))
514 anon_vma = (struct anon_vma *) (anon_mapping - PAGE_MAPPING_ANON);
515 root_anon_vma = READ_ONCE(anon_vma->root);
516 if (down_read_trylock(&root_anon_vma->rwsem)) {
518 * If the page is still mapped, then this anon_vma is still
519 * its anon_vma, and holding the mutex ensures that it will
520 * not go away, see anon_vma_free().
522 if (!page_mapped(page)) {
523 up_read(&root_anon_vma->rwsem);
529 /* trylock failed, we got to sleep */
530 if (!atomic_inc_not_zero(&anon_vma->refcount)) {
535 if (!page_mapped(page)) {
537 put_anon_vma(anon_vma);
541 /* we pinned the anon_vma, its safe to sleep */
543 anon_vma_lock_read(anon_vma);
545 if (atomic_dec_and_test(&anon_vma->refcount)) {
547 * Oops, we held the last refcount, release the lock
548 * and bail -- can't simply use put_anon_vma() because
549 * we'll deadlock on the anon_vma_lock_write() recursion.
551 anon_vma_unlock_read(anon_vma);
552 __put_anon_vma(anon_vma);
563 void page_unlock_anon_vma_read(struct anon_vma *anon_vma)
565 anon_vma_unlock_read(anon_vma);
569 * At what user virtual address is page expected in @vma?
571 static inline unsigned long
572 __vma_address(struct page *page, struct vm_area_struct *vma)
574 pgoff_t pgoff = page_to_pgoff(page);
575 return vma->vm_start + ((pgoff - vma->vm_pgoff) << PAGE_SHIFT);
579 vma_address(struct page *page, struct vm_area_struct *vma)
581 unsigned long address = __vma_address(page, vma);
583 /* page should be within @vma mapping range */
584 VM_BUG_ON_VMA(address < vma->vm_start || address >= vma->vm_end, vma);
589 #ifdef CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH
590 static void percpu_flush_tlb_batch_pages(void *data)
593 * All TLB entries are flushed on the assumption that it is
594 * cheaper to flush all TLBs and let them be refilled than
595 * flushing individual PFNs. Note that we do not track mm's
596 * to flush as that might simply be multiple full TLB flushes
599 count_vm_tlb_event(NR_TLB_REMOTE_FLUSH_RECEIVED);
604 * Flush TLB entries for recently unmapped pages from remote CPUs. It is
605 * important if a PTE was dirty when it was unmapped that it's flushed
606 * before any IO is initiated on the page to prevent lost writes. Similarly,
607 * it must be flushed before freeing to prevent data leakage.
609 void try_to_unmap_flush(void)
611 struct tlbflush_unmap_batch *tlb_ubc = ¤t->tlb_ubc;
614 if (!tlb_ubc->flush_required)
619 trace_tlb_flush(TLB_REMOTE_SHOOTDOWN, -1UL);
621 if (cpumask_test_cpu(cpu, &tlb_ubc->cpumask))
622 percpu_flush_tlb_batch_pages(&tlb_ubc->cpumask);
624 if (cpumask_any_but(&tlb_ubc->cpumask, cpu) < nr_cpu_ids) {
625 smp_call_function_many(&tlb_ubc->cpumask,
626 percpu_flush_tlb_batch_pages, (void *)tlb_ubc, true);
628 cpumask_clear(&tlb_ubc->cpumask);
629 tlb_ubc->flush_required = false;
630 tlb_ubc->writable = false;
634 /* Flush iff there are potentially writable TLB entries that can race with IO */
635 void try_to_unmap_flush_dirty(void)
637 struct tlbflush_unmap_batch *tlb_ubc = ¤t->tlb_ubc;
639 if (tlb_ubc->writable)
640 try_to_unmap_flush();
643 static void set_tlb_ubc_flush_pending(struct mm_struct *mm,
644 struct page *page, bool writable)
646 struct tlbflush_unmap_batch *tlb_ubc = ¤t->tlb_ubc;
648 cpumask_or(&tlb_ubc->cpumask, &tlb_ubc->cpumask, mm_cpumask(mm));
649 tlb_ubc->flush_required = true;
652 * If the PTE was dirty then it's best to assume it's writable. The
653 * caller must use try_to_unmap_flush_dirty() or try_to_unmap_flush()
654 * before the page is queued for IO.
657 tlb_ubc->writable = true;
661 * Returns true if the TLB flush should be deferred to the end of a batch of
662 * unmap operations to reduce IPIs.
664 static bool should_defer_flush(struct mm_struct *mm, enum ttu_flags flags)
666 bool should_defer = false;
668 if (!(flags & TTU_BATCH_FLUSH))
671 /* If remote CPUs need to be flushed then defer batch the flush */
672 if (cpumask_any_but(mm_cpumask(mm), get_cpu()) < nr_cpu_ids)
679 static void set_tlb_ubc_flush_pending(struct mm_struct *mm,
680 struct page *page, bool writable)
684 static bool should_defer_flush(struct mm_struct *mm, enum ttu_flags flags)
688 #endif /* CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH */
691 * At what user virtual address is page expected in vma?
692 * Caller should check the page is actually part of the vma.
694 unsigned long page_address_in_vma(struct page *page, struct vm_area_struct *vma)
696 unsigned long address;
697 if (PageAnon(page)) {
698 struct anon_vma *page__anon_vma = page_anon_vma(page);
700 * Note: swapoff's unuse_vma() is more efficient with this
701 * check, and needs it to match anon_vma when KSM is active.
703 if (!vma->anon_vma || !page__anon_vma ||
704 vma->anon_vma->root != page__anon_vma->root)
706 } else if (page->mapping) {
707 if (!vma->vm_file || vma->vm_file->f_mapping != page->mapping)
711 address = __vma_address(page, vma);
712 if (unlikely(address < vma->vm_start || address >= vma->vm_end))
717 pmd_t *mm_find_pmd(struct mm_struct *mm, unsigned long address)
724 pgd = pgd_offset(mm, address);
725 if (!pgd_present(*pgd))
728 pud = pud_offset(pgd, address);
729 if (!pud_present(*pud))
732 pmd = pmd_offset(pud, address);
734 * Some THP functions use the sequence pmdp_huge_clear_flush(), set_pmd_at()
735 * without holding anon_vma lock for write. So when looking for a
736 * genuine pmde (in which to find pte), test present and !THP together.
740 if (!pmd_present(pmde) || pmd_trans_huge(pmde))
747 * Check that @page is mapped at @address into @mm.
749 * If @sync is false, page_check_address may perform a racy check to avoid
750 * the page table lock when the pte is not present (helpful when reclaiming
751 * highly shared pages).
753 * On success returns with pte mapped and locked.
755 pte_t *__page_check_address(struct page *page, struct mm_struct *mm,
756 unsigned long address, spinlock_t **ptlp, int sync)
762 if (unlikely(PageHuge(page))) {
763 /* when pud is not present, pte will be NULL */
764 pte = huge_pte_offset(mm, address);
768 ptl = huge_pte_lockptr(page_hstate(page), mm, pte);
772 pmd = mm_find_pmd(mm, address);
776 pte = pte_offset_map(pmd, address);
777 /* Make a quick check before getting the lock */
778 if (!sync && !pte_present(*pte)) {
783 ptl = pte_lockptr(mm, pmd);
786 if (pte_present(*pte) && page_to_pfn(page) == pte_pfn(*pte)) {
790 pte_unmap_unlock(pte, ptl);
795 * page_mapped_in_vma - check whether a page is really mapped in a VMA
796 * @page: the page to test
797 * @vma: the VMA to test
799 * Returns 1 if the page is mapped into the page tables of the VMA, 0
800 * if the page is not mapped into the page tables of this VMA. Only
801 * valid for normal file or anonymous VMAs.
803 int page_mapped_in_vma(struct page *page, struct vm_area_struct *vma)
805 unsigned long address;
809 address = __vma_address(page, vma);
810 if (unlikely(address < vma->vm_start || address >= vma->vm_end))
812 pte = page_check_address(page, vma->vm_mm, address, &ptl, 1);
813 if (!pte) /* the page is not in this mm */
815 pte_unmap_unlock(pte, ptl);
820 struct page_referenced_arg {
823 unsigned long vm_flags;
824 struct mem_cgroup *memcg;
827 * arg: page_referenced_arg will be passed
829 static int page_referenced_one(struct page *page, struct vm_area_struct *vma,
830 unsigned long address, void *arg)
832 struct mm_struct *mm = vma->vm_mm;
835 struct page_referenced_arg *pra = arg;
837 if (unlikely(PageTransHuge(page))) {
841 * rmap might return false positives; we must filter
842 * these out using page_check_address_pmd().
844 pmd = page_check_address_pmd(page, mm, address, &ptl);
848 if (vma->vm_flags & VM_LOCKED) {
850 pra->vm_flags |= VM_LOCKED;
851 return SWAP_FAIL; /* To break the loop */
854 if (pmdp_clear_flush_young_notify(vma, address, pmd))
861 * rmap might return false positives; we must filter
862 * these out using page_check_address().
864 pte = page_check_address(page, mm, address, &ptl, 0);
868 if (vma->vm_flags & VM_LOCKED) {
869 pte_unmap_unlock(pte, ptl);
870 pra->vm_flags |= VM_LOCKED;
871 return SWAP_FAIL; /* To break the loop */
874 if (ptep_clear_flush_young_notify(vma, address, pte)) {
876 * Don't treat a reference through a sequentially read
877 * mapping as such. If the page has been used in
878 * another mapping, we will catch it; if this other
879 * mapping is already gone, the unmap path will have
880 * set PG_referenced or activated the page.
882 if (likely(!(vma->vm_flags & VM_SEQ_READ)))
885 pte_unmap_unlock(pte, ptl);
889 clear_page_idle(page);
890 if (test_and_clear_page_young(page))
895 pra->vm_flags |= vma->vm_flags;
900 return SWAP_SUCCESS; /* To break the loop */
905 static bool invalid_page_referenced_vma(struct vm_area_struct *vma, void *arg)
907 struct page_referenced_arg *pra = arg;
908 struct mem_cgroup *memcg = pra->memcg;
910 if (!mm_match_cgroup(vma->vm_mm, memcg))
917 * page_referenced - test if the page was referenced
918 * @page: the page to test
919 * @is_locked: caller holds lock on the page
920 * @memcg: target memory cgroup
921 * @vm_flags: collect encountered vma->vm_flags who actually referenced the page
923 * Quick test_and_clear_referenced for all mappings to a page,
924 * returns the number of ptes which referenced the page.
926 int page_referenced(struct page *page,
928 struct mem_cgroup *memcg,
929 unsigned long *vm_flags)
933 struct page_referenced_arg pra = {
934 .mapcount = page_mapcount(page),
937 struct rmap_walk_control rwc = {
938 .rmap_one = page_referenced_one,
940 .anon_lock = page_lock_anon_vma_read,
944 if (!page_mapped(page))
947 if (!page_rmapping(page))
950 if (!is_locked && (!PageAnon(page) || PageKsm(page))) {
951 we_locked = trylock_page(page);
957 * If we are reclaiming on behalf of a cgroup, skip
958 * counting on behalf of references from different
962 rwc.invalid_vma = invalid_page_referenced_vma;
965 ret = rmap_walk(page, &rwc);
966 *vm_flags = pra.vm_flags;
971 return pra.referenced;
974 static int page_mkclean_one(struct page *page, struct vm_area_struct *vma,
975 unsigned long address, void *arg)
977 struct mm_struct *mm = vma->vm_mm;
983 pte = page_check_address(page, mm, address, &ptl, 1);
987 if (pte_dirty(*pte) || pte_write(*pte)) {
990 flush_cache_page(vma, address, pte_pfn(*pte));
991 entry = ptep_clear_flush(vma, address, pte);
992 entry = pte_wrprotect(entry);
993 entry = pte_mkclean(entry);
994 set_pte_at(mm, address, pte, entry);
998 pte_unmap_unlock(pte, ptl);
1001 mmu_notifier_invalidate_page(mm, address);
1008 static bool invalid_mkclean_vma(struct vm_area_struct *vma, void *arg)
1010 if (vma->vm_flags & VM_SHARED)
1016 int page_mkclean(struct page *page)
1019 struct address_space *mapping;
1020 struct rmap_walk_control rwc = {
1021 .arg = (void *)&cleaned,
1022 .rmap_one = page_mkclean_one,
1023 .invalid_vma = invalid_mkclean_vma,
1026 BUG_ON(!PageLocked(page));
1028 if (!page_mapped(page))
1031 mapping = page_mapping(page);
1035 rmap_walk(page, &rwc);
1039 EXPORT_SYMBOL_GPL(page_mkclean);
1042 * page_move_anon_rmap - move a page to our anon_vma
1043 * @page: the page to move to our anon_vma
1044 * @vma: the vma the page belongs to
1045 * @address: the user virtual address mapped
1047 * When a page belongs exclusively to one process after a COW event,
1048 * that page can be moved into the anon_vma that belongs to just that
1049 * process, so the rmap code will not search the parent or sibling
1052 void page_move_anon_rmap(struct page *page,
1053 struct vm_area_struct *vma, unsigned long address)
1055 struct anon_vma *anon_vma = vma->anon_vma;
1057 VM_BUG_ON_PAGE(!PageLocked(page), page);
1058 VM_BUG_ON_VMA(!anon_vma, vma);
1059 VM_BUG_ON_PAGE(page->index != linear_page_index(vma, address), page);
1061 anon_vma = (void *) anon_vma + PAGE_MAPPING_ANON;
1063 * Ensure that anon_vma and the PAGE_MAPPING_ANON bit are written
1064 * simultaneously, so a concurrent reader (eg page_referenced()'s
1065 * PageAnon()) will not see one without the other.
1067 WRITE_ONCE(page->mapping, (struct address_space *) anon_vma);
1071 * __page_set_anon_rmap - set up new anonymous rmap
1072 * @page: Page to add to rmap
1073 * @vma: VM area to add page to.
1074 * @address: User virtual address of the mapping
1075 * @exclusive: the page is exclusively owned by the current process
1077 static void __page_set_anon_rmap(struct page *page,
1078 struct vm_area_struct *vma, unsigned long address, int exclusive)
1080 struct anon_vma *anon_vma = vma->anon_vma;
1088 * If the page isn't exclusively mapped into this vma,
1089 * we must use the _oldest_ possible anon_vma for the
1093 anon_vma = anon_vma->root;
1095 anon_vma = (void *) anon_vma + PAGE_MAPPING_ANON;
1096 page->mapping = (struct address_space *) anon_vma;
1097 page->index = linear_page_index(vma, address);
1101 * __page_check_anon_rmap - sanity check anonymous rmap addition
1102 * @page: the page to add the mapping to
1103 * @vma: the vm area in which the mapping is added
1104 * @address: the user virtual address mapped
1106 static void __page_check_anon_rmap(struct page *page,
1107 struct vm_area_struct *vma, unsigned long address)
1109 #ifdef CONFIG_DEBUG_VM
1111 * The page's anon-rmap details (mapping and index) are guaranteed to
1112 * be set up correctly at this point.
1114 * We have exclusion against page_add_anon_rmap because the caller
1115 * always holds the page locked, except if called from page_dup_rmap,
1116 * in which case the page is already known to be setup.
1118 * We have exclusion against page_add_new_anon_rmap because those pages
1119 * are initially only visible via the pagetables, and the pte is locked
1120 * over the call to page_add_new_anon_rmap.
1122 BUG_ON(page_anon_vma(page)->root != vma->anon_vma->root);
1123 BUG_ON(page_to_pgoff(page) != linear_page_index(vma, address));
1128 * page_add_anon_rmap - add pte mapping to an anonymous page
1129 * @page: the page to add the mapping to
1130 * @vma: the vm area in which the mapping is added
1131 * @address: the user virtual address mapped
1132 * @compound: charge the page as compound or small page
1134 * The caller needs to hold the pte lock, and the page must be locked in
1135 * the anon_vma case: to serialize mapping,index checking after setting,
1136 * and to ensure that PageAnon is not being upgraded racily to PageKsm
1137 * (but PageKsm is never downgraded to PageAnon).
1139 void page_add_anon_rmap(struct page *page,
1140 struct vm_area_struct *vma, unsigned long address, bool compound)
1142 do_page_add_anon_rmap(page, vma, address, compound ? RMAP_COMPOUND : 0);
1146 * Special version of the above for do_swap_page, which often runs
1147 * into pages that are exclusively owned by the current process.
1148 * Everybody else should continue to use page_add_anon_rmap above.
1150 void do_page_add_anon_rmap(struct page *page,
1151 struct vm_area_struct *vma, unsigned long address, int flags)
1153 bool compound = flags & RMAP_COMPOUND;
1156 if (PageTransCompound(page)) {
1157 VM_BUG_ON_PAGE(!PageLocked(page), page);
1161 VM_BUG_ON_PAGE(!PageTransHuge(page), page);
1162 mapcount = compound_mapcount_ptr(page);
1163 first = atomic_inc_and_test(mapcount);
1165 /* Anon THP always mapped first with PMD */
1167 VM_BUG_ON_PAGE(!page_mapcount(page), page);
1168 atomic_inc(&page->_mapcount);
1171 VM_BUG_ON_PAGE(compound, page);
1172 first = atomic_inc_and_test(&page->_mapcount);
1176 int nr = compound ? hpage_nr_pages(page) : 1;
1178 * We use the irq-unsafe __{inc|mod}_zone_page_stat because
1179 * these counters are not modified in interrupt context, and
1180 * pte lock(a spinlock) is held, which implies preemption
1184 VM_BUG_ON_PAGE(!PageTransHuge(page), page);
1185 __inc_zone_page_state(page,
1186 NR_ANON_TRANSPARENT_HUGEPAGES);
1188 __mod_zone_page_state(page_zone(page), NR_ANON_PAGES, nr);
1190 if (unlikely(PageKsm(page)))
1193 VM_BUG_ON_PAGE(!PageLocked(page), page);
1195 /* address might be in next vma when migration races vma_adjust */
1197 __page_set_anon_rmap(page, vma, address,
1198 flags & RMAP_EXCLUSIVE);
1200 __page_check_anon_rmap(page, vma, address);
1204 * page_add_new_anon_rmap - add pte mapping to a new anonymous page
1205 * @page: the page to add the mapping to
1206 * @vma: the vm area in which the mapping is added
1207 * @address: the user virtual address mapped
1208 * @compound: charge the page as compound or small page
1210 * Same as page_add_anon_rmap but must only be called on *new* pages.
1211 * This means the inc-and-test can be bypassed.
1212 * Page does not have to be locked.
1214 void page_add_new_anon_rmap(struct page *page,
1215 struct vm_area_struct *vma, unsigned long address, bool compound)
1217 int nr = compound ? hpage_nr_pages(page) : 1;
1219 VM_BUG_ON_VMA(address < vma->vm_start || address >= vma->vm_end, vma);
1220 SetPageSwapBacked(page);
1222 VM_BUG_ON_PAGE(!PageTransHuge(page), page);
1223 /* increment count (starts at -1) */
1224 atomic_set(compound_mapcount_ptr(page), 0);
1225 __inc_zone_page_state(page, NR_ANON_TRANSPARENT_HUGEPAGES);
1227 /* Anon THP always mapped first with PMD */
1228 VM_BUG_ON_PAGE(PageTransCompound(page), page);
1229 /* increment count (starts at -1) */
1230 atomic_set(&page->_mapcount, 0);
1232 __mod_zone_page_state(page_zone(page), NR_ANON_PAGES, nr);
1233 __page_set_anon_rmap(page, vma, address, 1);
1237 * page_add_file_rmap - add pte mapping to a file page
1238 * @page: the page to add the mapping to
1240 * The caller needs to hold the pte lock.
1242 void page_add_file_rmap(struct page *page)
1244 struct mem_cgroup *memcg;
1246 memcg = mem_cgroup_begin_page_stat(page);
1247 if (atomic_inc_and_test(&page->_mapcount)) {
1248 __inc_zone_page_state(page, NR_FILE_MAPPED);
1249 mem_cgroup_inc_page_stat(memcg, MEM_CGROUP_STAT_FILE_MAPPED);
1251 mem_cgroup_end_page_stat(memcg);
1254 static void page_remove_file_rmap(struct page *page)
1256 struct mem_cgroup *memcg;
1258 memcg = mem_cgroup_begin_page_stat(page);
1260 /* Hugepages are not counted in NR_FILE_MAPPED for now. */
1261 if (unlikely(PageHuge(page))) {
1262 /* hugetlb pages are always mapped with pmds */
1263 atomic_dec(compound_mapcount_ptr(page));
1267 /* page still mapped by someone else? */
1268 if (!atomic_add_negative(-1, &page->_mapcount))
1272 * We use the irq-unsafe __{inc|mod}_zone_page_stat because
1273 * these counters are not modified in interrupt context, and
1274 * pte lock(a spinlock) is held, which implies preemption disabled.
1276 __dec_zone_page_state(page, NR_FILE_MAPPED);
1277 mem_cgroup_dec_page_stat(memcg, MEM_CGROUP_STAT_FILE_MAPPED);
1279 if (unlikely(PageMlocked(page)))
1280 clear_page_mlock(page);
1282 mem_cgroup_end_page_stat(memcg);
1285 static void page_remove_anon_compound_rmap(struct page *page)
1289 if (!atomic_add_negative(-1, compound_mapcount_ptr(page)))
1292 /* Hugepages are not counted in NR_ANON_PAGES for now. */
1293 if (unlikely(PageHuge(page)))
1296 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE))
1299 __dec_zone_page_state(page, NR_ANON_TRANSPARENT_HUGEPAGES);
1301 if (TestClearPageDoubleMap(page)) {
1303 * Subpages can be mapped with PTEs too. Check how many of
1304 * themi are still mapped.
1306 for (i = 0, nr = 0; i < HPAGE_PMD_NR; i++) {
1307 if (atomic_add_negative(-1, &page[i]._mapcount))
1315 __mod_zone_page_state(page_zone(page), NR_ANON_PAGES, -nr);
1319 * page_remove_rmap - take down pte mapping from a page
1320 * @page: page to remove mapping from
1321 * @compound: uncharge the page as compound or small page
1323 * The caller needs to hold the pte lock.
1325 void page_remove_rmap(struct page *page, bool compound)
1327 if (!PageAnon(page)) {
1328 VM_BUG_ON_PAGE(compound && !PageHuge(page), page);
1329 page_remove_file_rmap(page);
1334 return page_remove_anon_compound_rmap(page);
1336 /* page still mapped by someone else? */
1337 if (!atomic_add_negative(-1, &page->_mapcount))
1341 * We use the irq-unsafe __{inc|mod}_zone_page_stat because
1342 * these counters are not modified in interrupt context, and
1343 * pte lock(a spinlock) is held, which implies preemption disabled.
1345 __dec_zone_page_state(page, NR_ANON_PAGES);
1347 if (unlikely(PageMlocked(page)))
1348 clear_page_mlock(page);
1351 * It would be tidy to reset the PageAnon mapping here,
1352 * but that might overwrite a racing page_add_anon_rmap
1353 * which increments mapcount after us but sets mapping
1354 * before us: so leave the reset to free_hot_cold_page,
1355 * and remember that it's only reliable while mapped.
1356 * Leaving it set also helps swapoff to reinstate ptes
1357 * faster for those pages still in swapcache.
1362 * @arg: enum ttu_flags will be passed to this argument
1364 static int try_to_unmap_one(struct page *page, struct vm_area_struct *vma,
1365 unsigned long address, void *arg)
1367 struct mm_struct *mm = vma->vm_mm;
1371 int ret = SWAP_AGAIN;
1372 enum ttu_flags flags = (enum ttu_flags)arg;
1374 /* munlock has nothing to gain from examining un-locked vmas */
1375 if ((flags & TTU_MUNLOCK) && !(vma->vm_flags & VM_LOCKED))
1378 pte = page_check_address(page, mm, address, &ptl, 0);
1383 * If the page is mlock()d, we cannot swap it out.
1384 * If it's recently referenced (perhaps page_referenced
1385 * skipped over this mm) then we should reactivate it.
1387 if (!(flags & TTU_IGNORE_MLOCK)) {
1388 if (vma->vm_flags & VM_LOCKED) {
1389 /* Holding pte lock, we do *not* need mmap_sem here */
1390 mlock_vma_page(page);
1394 if (flags & TTU_MUNLOCK)
1397 if (!(flags & TTU_IGNORE_ACCESS)) {
1398 if (ptep_clear_flush_young_notify(vma, address, pte)) {
1404 /* Nuke the page table entry. */
1405 flush_cache_page(vma, address, page_to_pfn(page));
1406 if (should_defer_flush(mm, flags)) {
1408 * We clear the PTE but do not flush so potentially a remote
1409 * CPU could still be writing to the page. If the entry was
1410 * previously clean then the architecture must guarantee that
1411 * a clear->dirty transition on a cached TLB entry is written
1412 * through and traps if the PTE is unmapped.
1414 pteval = ptep_get_and_clear(mm, address, pte);
1416 set_tlb_ubc_flush_pending(mm, page, pte_dirty(pteval));
1418 pteval = ptep_clear_flush(vma, address, pte);
1421 /* Move the dirty bit to the physical page now the pte is gone. */
1422 if (pte_dirty(pteval))
1423 set_page_dirty(page);
1425 /* Update high watermark before we lower rss */
1426 update_hiwater_rss(mm);
1428 if (PageHWPoison(page) && !(flags & TTU_IGNORE_HWPOISON)) {
1429 if (PageHuge(page)) {
1430 hugetlb_count_sub(1 << compound_order(page), mm);
1433 dec_mm_counter(mm, MM_ANONPAGES);
1435 dec_mm_counter(mm, MM_FILEPAGES);
1437 set_pte_at(mm, address, pte,
1438 swp_entry_to_pte(make_hwpoison_entry(page)));
1439 } else if (pte_unused(pteval)) {
1441 * The guest indicated that the page content is of no
1442 * interest anymore. Simply discard the pte, vmscan
1443 * will take care of the rest.
1446 dec_mm_counter(mm, MM_ANONPAGES);
1448 dec_mm_counter(mm, MM_FILEPAGES);
1449 } else if (IS_ENABLED(CONFIG_MIGRATION) && (flags & TTU_MIGRATION)) {
1453 * Store the pfn of the page in a special migration
1454 * pte. do_swap_page() will wait until the migration
1455 * pte is removed and then restart fault handling.
1457 entry = make_migration_entry(page, pte_write(pteval));
1458 swp_pte = swp_entry_to_pte(entry);
1459 if (pte_soft_dirty(pteval))
1460 swp_pte = pte_swp_mksoft_dirty(swp_pte);
1461 set_pte_at(mm, address, pte, swp_pte);
1462 } else if (PageAnon(page)) {
1463 swp_entry_t entry = { .val = page_private(page) };
1466 * Store the swap location in the pte.
1467 * See handle_pte_fault() ...
1469 VM_BUG_ON_PAGE(!PageSwapCache(page), page);
1470 if (swap_duplicate(entry) < 0) {
1471 set_pte_at(mm, address, pte, pteval);
1475 if (list_empty(&mm->mmlist)) {
1476 spin_lock(&mmlist_lock);
1477 if (list_empty(&mm->mmlist))
1478 list_add(&mm->mmlist, &init_mm.mmlist);
1479 spin_unlock(&mmlist_lock);
1481 dec_mm_counter(mm, MM_ANONPAGES);
1482 inc_mm_counter(mm, MM_SWAPENTS);
1483 swp_pte = swp_entry_to_pte(entry);
1484 if (pte_soft_dirty(pteval))
1485 swp_pte = pte_swp_mksoft_dirty(swp_pte);
1486 set_pte_at(mm, address, pte, swp_pte);
1488 dec_mm_counter(mm, MM_FILEPAGES);
1490 page_remove_rmap(page, false);
1491 page_cache_release(page);
1494 pte_unmap_unlock(pte, ptl);
1495 if (ret != SWAP_FAIL && ret != SWAP_MLOCK && !(flags & TTU_MUNLOCK))
1496 mmu_notifier_invalidate_page(mm, address);
1501 bool is_vma_temporary_stack(struct vm_area_struct *vma)
1503 int maybe_stack = vma->vm_flags & (VM_GROWSDOWN | VM_GROWSUP);
1508 if ((vma->vm_flags & VM_STACK_INCOMPLETE_SETUP) ==
1509 VM_STACK_INCOMPLETE_SETUP)
1515 static bool invalid_migration_vma(struct vm_area_struct *vma, void *arg)
1517 return is_vma_temporary_stack(vma);
1520 static int page_not_mapped(struct page *page)
1522 return !page_mapped(page);
1526 * try_to_unmap - try to remove all page table mappings to a page
1527 * @page: the page to get unmapped
1528 * @flags: action and flags
1530 * Tries to remove all the page table entries which are mapping this
1531 * page, used in the pageout path. Caller must hold the page lock.
1532 * Return values are:
1534 * SWAP_SUCCESS - we succeeded in removing all mappings
1535 * SWAP_AGAIN - we missed a mapping, try again later
1536 * SWAP_FAIL - the page is unswappable
1537 * SWAP_MLOCK - page is mlocked.
1539 int try_to_unmap(struct page *page, enum ttu_flags flags)
1542 struct rmap_walk_control rwc = {
1543 .rmap_one = try_to_unmap_one,
1544 .arg = (void *)flags,
1545 .done = page_not_mapped,
1546 .anon_lock = page_lock_anon_vma_read,
1549 VM_BUG_ON_PAGE(!PageHuge(page) && PageTransHuge(page), page);
1552 * During exec, a temporary VMA is setup and later moved.
1553 * The VMA is moved under the anon_vma lock but not the
1554 * page tables leading to a race where migration cannot
1555 * find the migration ptes. Rather than increasing the
1556 * locking requirements of exec(), migration skips
1557 * temporary VMAs until after exec() completes.
1559 if ((flags & TTU_MIGRATION) && !PageKsm(page) && PageAnon(page))
1560 rwc.invalid_vma = invalid_migration_vma;
1562 ret = rmap_walk(page, &rwc);
1564 if (ret != SWAP_MLOCK && !page_mapped(page))
1570 * try_to_munlock - try to munlock a page
1571 * @page: the page to be munlocked
1573 * Called from munlock code. Checks all of the VMAs mapping the page
1574 * to make sure nobody else has this page mlocked. The page will be
1575 * returned with PG_mlocked cleared if no other vmas have it mlocked.
1577 * Return values are:
1579 * SWAP_AGAIN - no vma is holding page mlocked, or,
1580 * SWAP_AGAIN - page mapped in mlocked vma -- couldn't acquire mmap sem
1581 * SWAP_FAIL - page cannot be located at present
1582 * SWAP_MLOCK - page is now mlocked.
1584 int try_to_munlock(struct page *page)
1587 struct rmap_walk_control rwc = {
1588 .rmap_one = try_to_unmap_one,
1589 .arg = (void *)TTU_MUNLOCK,
1590 .done = page_not_mapped,
1591 .anon_lock = page_lock_anon_vma_read,
1595 VM_BUG_ON_PAGE(!PageLocked(page) || PageLRU(page), page);
1597 ret = rmap_walk(page, &rwc);
1601 void __put_anon_vma(struct anon_vma *anon_vma)
1603 struct anon_vma *root = anon_vma->root;
1605 anon_vma_free(anon_vma);
1606 if (root != anon_vma && atomic_dec_and_test(&root->refcount))
1607 anon_vma_free(root);
1610 static struct anon_vma *rmap_walk_anon_lock(struct page *page,
1611 struct rmap_walk_control *rwc)
1613 struct anon_vma *anon_vma;
1616 return rwc->anon_lock(page);
1619 * Note: remove_migration_ptes() cannot use page_lock_anon_vma_read()
1620 * because that depends on page_mapped(); but not all its usages
1621 * are holding mmap_sem. Users without mmap_sem are required to
1622 * take a reference count to prevent the anon_vma disappearing
1624 anon_vma = page_anon_vma(page);
1628 anon_vma_lock_read(anon_vma);
1633 * rmap_walk_anon - do something to anonymous page using the object-based
1635 * @page: the page to be handled
1636 * @rwc: control variable according to each walk type
1638 * Find all the mappings of a page using the mapping pointer and the vma chains
1639 * contained in the anon_vma struct it points to.
1641 * When called from try_to_munlock(), the mmap_sem of the mm containing the vma
1642 * where the page was found will be held for write. So, we won't recheck
1643 * vm_flags for that VMA. That should be OK, because that vma shouldn't be
1646 static int rmap_walk_anon(struct page *page, struct rmap_walk_control *rwc)
1648 struct anon_vma *anon_vma;
1650 struct anon_vma_chain *avc;
1651 int ret = SWAP_AGAIN;
1653 anon_vma = rmap_walk_anon_lock(page, rwc);
1657 pgoff = page_to_pgoff(page);
1658 anon_vma_interval_tree_foreach(avc, &anon_vma->rb_root, pgoff, pgoff) {
1659 struct vm_area_struct *vma = avc->vma;
1660 unsigned long address = vma_address(page, vma);
1664 if (rwc->invalid_vma && rwc->invalid_vma(vma, rwc->arg))
1667 ret = rwc->rmap_one(page, vma, address, rwc->arg);
1668 if (ret != SWAP_AGAIN)
1670 if (rwc->done && rwc->done(page))
1673 anon_vma_unlock_read(anon_vma);
1678 * rmap_walk_file - do something to file page using the object-based rmap method
1679 * @page: the page to be handled
1680 * @rwc: control variable according to each walk type
1682 * Find all the mappings of a page using the mapping pointer and the vma chains
1683 * contained in the address_space struct it points to.
1685 * When called from try_to_munlock(), the mmap_sem of the mm containing the vma
1686 * where the page was found will be held for write. So, we won't recheck
1687 * vm_flags for that VMA. That should be OK, because that vma shouldn't be
1690 static int rmap_walk_file(struct page *page, struct rmap_walk_control *rwc)
1692 struct address_space *mapping = page->mapping;
1694 struct vm_area_struct *vma;
1695 int ret = SWAP_AGAIN;
1698 * The page lock not only makes sure that page->mapping cannot
1699 * suddenly be NULLified by truncation, it makes sure that the
1700 * structure at mapping cannot be freed and reused yet,
1701 * so we can safely take mapping->i_mmap_rwsem.
1703 VM_BUG_ON_PAGE(!PageLocked(page), page);
1708 pgoff = page_to_pgoff(page);
1709 i_mmap_lock_read(mapping);
1710 vma_interval_tree_foreach(vma, &mapping->i_mmap, pgoff, pgoff) {
1711 unsigned long address = vma_address(page, vma);
1715 if (rwc->invalid_vma && rwc->invalid_vma(vma, rwc->arg))
1718 ret = rwc->rmap_one(page, vma, address, rwc->arg);
1719 if (ret != SWAP_AGAIN)
1721 if (rwc->done && rwc->done(page))
1726 i_mmap_unlock_read(mapping);
1730 int rmap_walk(struct page *page, struct rmap_walk_control *rwc)
1732 if (unlikely(PageKsm(page)))
1733 return rmap_walk_ksm(page, rwc);
1734 else if (PageAnon(page))
1735 return rmap_walk_anon(page, rwc);
1737 return rmap_walk_file(page, rwc);
1740 #ifdef CONFIG_HUGETLB_PAGE
1742 * The following three functions are for anonymous (private mapped) hugepages.
1743 * Unlike common anonymous pages, anonymous hugepages have no accounting code
1744 * and no lru code, because we handle hugepages differently from common pages.
1746 static void __hugepage_set_anon_rmap(struct page *page,
1747 struct vm_area_struct *vma, unsigned long address, int exclusive)
1749 struct anon_vma *anon_vma = vma->anon_vma;
1756 anon_vma = anon_vma->root;
1758 anon_vma = (void *) anon_vma + PAGE_MAPPING_ANON;
1759 page->mapping = (struct address_space *) anon_vma;
1760 page->index = linear_page_index(vma, address);
1763 void hugepage_add_anon_rmap(struct page *page,
1764 struct vm_area_struct *vma, unsigned long address)
1766 struct anon_vma *anon_vma = vma->anon_vma;
1769 BUG_ON(!PageLocked(page));
1771 /* address might be in next vma when migration races vma_adjust */
1772 first = atomic_inc_and_test(compound_mapcount_ptr(page));
1774 __hugepage_set_anon_rmap(page, vma, address, 0);
1777 void hugepage_add_new_anon_rmap(struct page *page,
1778 struct vm_area_struct *vma, unsigned long address)
1780 BUG_ON(address < vma->vm_start || address >= vma->vm_end);
1781 atomic_set(compound_mapcount_ptr(page), 0);
1782 __hugepage_set_anon_rmap(page, vma, address, 1);
1784 #endif /* CONFIG_HUGETLB_PAGE */