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 * hugetlbfs_i_mmap_rwsem_key (in huge_pmd_share)
27 * mapping->i_mmap_rwsem
29 * mm->page_table_lock or pte_lock
30 * zone_lru_lock (in mark_page_accessed, isolate_lru_page)
31 * swap_lock (in swap_duplicate, swap_info_get)
32 * mmlist_lock (in mmput, drain_mmlist and others)
33 * mapping->private_lock (in __set_page_dirty_buffers)
34 * mem_cgroup_{begin,end}_page_stat (memcg->move_lock)
35 * mapping->tree_lock (widely used)
36 * inode->i_lock (in set_page_dirty's __mark_inode_dirty)
37 * bdi.wb->list_lock (in set_page_dirty's __mark_inode_dirty)
38 * sb_lock (within inode_lock in fs/fs-writeback.c)
39 * mapping->tree_lock (widely used, in set_page_dirty,
40 * in arch-dependent flush_dcache_mmap_lock,
41 * within bdi.wb->list_lock in __sync_single_inode)
43 * anon_vma->rwsem,mapping->i_mutex (memory_failure, collect_procs_anon)
49 #include <linux/pagemap.h>
50 #include <linux/swap.h>
51 #include <linux/swapops.h>
52 #include <linux/slab.h>
53 #include <linux/init.h>
54 #include <linux/ksm.h>
55 #include <linux/rmap.h>
56 #include <linux/rcupdate.h>
57 #include <linux/export.h>
58 #include <linux/memcontrol.h>
59 #include <linux/mmu_notifier.h>
60 #include <linux/migrate.h>
61 #include <linux/hugetlb.h>
62 #include <linux/backing-dev.h>
63 #include <linux/page_idle.h>
65 #include <asm/tlbflush.h>
67 #include <trace/events/tlb.h>
71 static struct kmem_cache
*anon_vma_cachep
;
72 static struct kmem_cache
*anon_vma_chain_cachep
;
74 static inline struct anon_vma
*anon_vma_alloc(void)
76 struct anon_vma
*anon_vma
;
78 anon_vma
= kmem_cache_alloc(anon_vma_cachep
, GFP_KERNEL
);
80 atomic_set(&anon_vma
->refcount
, 1);
81 anon_vma
->degree
= 1; /* Reference for first vma */
82 anon_vma
->parent
= anon_vma
;
84 * Initialise the anon_vma root to point to itself. If called
85 * from fork, the root will be reset to the parents anon_vma.
87 anon_vma
->root
= anon_vma
;
93 static inline void anon_vma_free(struct anon_vma
*anon_vma
)
95 VM_BUG_ON(atomic_read(&anon_vma
->refcount
));
98 * Synchronize against page_lock_anon_vma_read() such that
99 * we can safely hold the lock without the anon_vma getting
102 * Relies on the full mb implied by the atomic_dec_and_test() from
103 * put_anon_vma() against the acquire barrier implied by
104 * down_read_trylock() from page_lock_anon_vma_read(). This orders:
106 * page_lock_anon_vma_read() VS put_anon_vma()
107 * down_read_trylock() atomic_dec_and_test()
109 * atomic_read() rwsem_is_locked()
111 * LOCK should suffice since the actual taking of the lock must
112 * happen _before_ what follows.
115 if (rwsem_is_locked(&anon_vma
->root
->rwsem
)) {
116 anon_vma_lock_write(anon_vma
);
117 anon_vma_unlock_write(anon_vma
);
120 kmem_cache_free(anon_vma_cachep
, anon_vma
);
123 static inline struct anon_vma_chain
*anon_vma_chain_alloc(gfp_t gfp
)
125 return kmem_cache_alloc(anon_vma_chain_cachep
, gfp
);
128 static void anon_vma_chain_free(struct anon_vma_chain
*anon_vma_chain
)
130 kmem_cache_free(anon_vma_chain_cachep
, anon_vma_chain
);
133 static void anon_vma_chain_link(struct vm_area_struct
*vma
,
134 struct anon_vma_chain
*avc
,
135 struct anon_vma
*anon_vma
)
138 avc
->anon_vma
= anon_vma
;
139 list_add(&avc
->same_vma
, &vma
->anon_vma_chain
);
140 anon_vma_interval_tree_insert(avc
, &anon_vma
->rb_root
);
144 * __anon_vma_prepare - attach an anon_vma to a memory region
145 * @vma: the memory region in question
147 * This makes sure the memory mapping described by 'vma' has
148 * an 'anon_vma' attached to it, so that we can associate the
149 * anonymous pages mapped into it with that anon_vma.
151 * The common case will be that we already have one, which
152 * is handled inline by anon_vma_prepare(). But if
153 * not we either need to find an adjacent mapping that we
154 * can re-use the anon_vma from (very common when the only
155 * reason for splitting a vma has been mprotect()), or we
156 * allocate a new one.
158 * Anon-vma allocations are very subtle, because we may have
159 * optimistically looked up an anon_vma in page_lock_anon_vma_read()
160 * and that may actually touch the spinlock even in the newly
161 * allocated vma (it depends on RCU to make sure that the
162 * anon_vma isn't actually destroyed).
164 * As a result, we need to do proper anon_vma locking even
165 * for the new allocation. At the same time, we do not want
166 * to do any locking for the common case of already having
169 * This must be called with the mmap_sem held for reading.
171 int __anon_vma_prepare(struct vm_area_struct
*vma
)
173 struct mm_struct
*mm
= vma
->vm_mm
;
174 struct anon_vma
*anon_vma
, *allocated
;
175 struct anon_vma_chain
*avc
;
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 VM_WARN_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
|SLAB_ACCOUNT
,
433 anon_vma_chain_cachep
= KMEM_CACHE(anon_vma_chain
,
434 SLAB_PANIC
|SLAB_ACCOUNT
);
438 * Getting a lock on a stable anon_vma from a page off the LRU is tricky!
440 * Since there is no serialization what so ever against page_remove_rmap()
441 * the best this function can do is return a locked anon_vma that might
442 * have been relevant to this page.
444 * The page might have been remapped to a different anon_vma or the anon_vma
445 * returned may already be freed (and even reused).
447 * In case it was remapped to a different anon_vma, the new anon_vma will be a
448 * child of the old anon_vma, and the anon_vma lifetime rules will therefore
449 * ensure that any anon_vma obtained from the page will still be valid for as
450 * long as we observe page_mapped() [ hence all those page_mapped() tests ].
452 * All users of this function must be very careful when walking the anon_vma
453 * chain and verify that the page in question is indeed mapped in it
454 * [ something equivalent to page_mapped_in_vma() ].
456 * Since anon_vma's slab is DESTROY_BY_RCU and we know from page_remove_rmap()
457 * that the anon_vma pointer from page->mapping is valid if there is a
458 * mapcount, we can dereference the anon_vma after observing those.
460 struct anon_vma
*page_get_anon_vma(struct page
*page
)
462 struct anon_vma
*anon_vma
= NULL
;
463 unsigned long anon_mapping
;
466 anon_mapping
= (unsigned long)READ_ONCE(page
->mapping
);
467 if ((anon_mapping
& PAGE_MAPPING_FLAGS
) != PAGE_MAPPING_ANON
)
469 if (!page_mapped(page
))
472 anon_vma
= (struct anon_vma
*) (anon_mapping
- PAGE_MAPPING_ANON
);
473 if (!atomic_inc_not_zero(&anon_vma
->refcount
)) {
479 * If this page is still mapped, then its anon_vma cannot have been
480 * freed. But if it has been unmapped, we have no security against the
481 * anon_vma structure being freed and reused (for another anon_vma:
482 * SLAB_DESTROY_BY_RCU guarantees that - so the atomic_inc_not_zero()
483 * above cannot corrupt).
485 if (!page_mapped(page
)) {
487 put_anon_vma(anon_vma
);
497 * Similar to page_get_anon_vma() except it locks the anon_vma.
499 * Its a little more complex as it tries to keep the fast path to a single
500 * atomic op -- the trylock. If we fail the trylock, we fall back to getting a
501 * reference like with page_get_anon_vma() and then block on the mutex.
503 struct anon_vma
*page_lock_anon_vma_read(struct page
*page
)
505 struct anon_vma
*anon_vma
= NULL
;
506 struct anon_vma
*root_anon_vma
;
507 unsigned long anon_mapping
;
510 anon_mapping
= (unsigned long)READ_ONCE(page
->mapping
);
511 if ((anon_mapping
& PAGE_MAPPING_FLAGS
) != PAGE_MAPPING_ANON
)
513 if (!page_mapped(page
))
516 anon_vma
= (struct anon_vma
*) (anon_mapping
- PAGE_MAPPING_ANON
);
517 root_anon_vma
= READ_ONCE(anon_vma
->root
);
518 if (down_read_trylock(&root_anon_vma
->rwsem
)) {
520 * If the page is still mapped, then this anon_vma is still
521 * its anon_vma, and holding the mutex ensures that it will
522 * not go away, see anon_vma_free().
524 if (!page_mapped(page
)) {
525 up_read(&root_anon_vma
->rwsem
);
531 /* trylock failed, we got to sleep */
532 if (!atomic_inc_not_zero(&anon_vma
->refcount
)) {
537 if (!page_mapped(page
)) {
539 put_anon_vma(anon_vma
);
543 /* we pinned the anon_vma, its safe to sleep */
545 anon_vma_lock_read(anon_vma
);
547 if (atomic_dec_and_test(&anon_vma
->refcount
)) {
549 * Oops, we held the last refcount, release the lock
550 * and bail -- can't simply use put_anon_vma() because
551 * we'll deadlock on the anon_vma_lock_write() recursion.
553 anon_vma_unlock_read(anon_vma
);
554 __put_anon_vma(anon_vma
);
565 void page_unlock_anon_vma_read(struct anon_vma
*anon_vma
)
567 anon_vma_unlock_read(anon_vma
);
570 #ifdef CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH
572 * Flush TLB entries for recently unmapped pages from remote CPUs. It is
573 * important if a PTE was dirty when it was unmapped that it's flushed
574 * before any IO is initiated on the page to prevent lost writes. Similarly,
575 * it must be flushed before freeing to prevent data leakage.
577 void try_to_unmap_flush(void)
579 struct tlbflush_unmap_batch
*tlb_ubc
= ¤t
->tlb_ubc
;
582 if (!tlb_ubc
->flush_required
)
587 if (cpumask_test_cpu(cpu
, &tlb_ubc
->cpumask
)) {
588 count_vm_tlb_event(NR_TLB_LOCAL_FLUSH_ALL
);
590 trace_tlb_flush(TLB_LOCAL_SHOOTDOWN
, TLB_FLUSH_ALL
);
593 if (cpumask_any_but(&tlb_ubc
->cpumask
, cpu
) < nr_cpu_ids
)
594 flush_tlb_others(&tlb_ubc
->cpumask
, NULL
, 0, TLB_FLUSH_ALL
);
595 cpumask_clear(&tlb_ubc
->cpumask
);
596 tlb_ubc
->flush_required
= false;
597 tlb_ubc
->writable
= false;
601 /* Flush iff there are potentially writable TLB entries that can race with IO */
602 void try_to_unmap_flush_dirty(void)
604 struct tlbflush_unmap_batch
*tlb_ubc
= ¤t
->tlb_ubc
;
606 if (tlb_ubc
->writable
)
607 try_to_unmap_flush();
610 static void set_tlb_ubc_flush_pending(struct mm_struct
*mm
, bool writable
)
612 struct tlbflush_unmap_batch
*tlb_ubc
= ¤t
->tlb_ubc
;
614 cpumask_or(&tlb_ubc
->cpumask
, &tlb_ubc
->cpumask
, mm_cpumask(mm
));
615 tlb_ubc
->flush_required
= true;
618 * If the PTE was dirty then it's best to assume it's writable. The
619 * caller must use try_to_unmap_flush_dirty() or try_to_unmap_flush()
620 * before the page is queued for IO.
623 tlb_ubc
->writable
= true;
627 * Returns true if the TLB flush should be deferred to the end of a batch of
628 * unmap operations to reduce IPIs.
630 static bool should_defer_flush(struct mm_struct
*mm
, enum ttu_flags flags
)
632 bool should_defer
= false;
634 if (!(flags
& TTU_BATCH_FLUSH
))
637 /* If remote CPUs need to be flushed then defer batch the flush */
638 if (cpumask_any_but(mm_cpumask(mm
), get_cpu()) < nr_cpu_ids
)
645 static void set_tlb_ubc_flush_pending(struct mm_struct
*mm
, bool writable
)
649 static bool should_defer_flush(struct mm_struct
*mm
, enum ttu_flags flags
)
653 #endif /* CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH */
656 * At what user virtual address is page expected in vma?
657 * Caller should check the page is actually part of the vma.
659 unsigned long page_address_in_vma(struct page
*page
, struct vm_area_struct
*vma
)
661 unsigned long address
;
662 if (PageAnon(page
)) {
663 struct anon_vma
*page__anon_vma
= page_anon_vma(page
);
665 * Note: swapoff's unuse_vma() is more efficient with this
666 * check, and needs it to match anon_vma when KSM is active.
668 if (!vma
->anon_vma
|| !page__anon_vma
||
669 vma
->anon_vma
->root
!= page__anon_vma
->root
)
671 } else if (page
->mapping
) {
672 if (!vma
->vm_file
|| vma
->vm_file
->f_mapping
!= page
->mapping
)
676 address
= __vma_address(page
, vma
);
677 if (unlikely(address
< vma
->vm_start
|| address
>= vma
->vm_end
))
682 pmd_t
*mm_find_pmd(struct mm_struct
*mm
, unsigned long address
)
689 pgd
= pgd_offset(mm
, address
);
690 if (!pgd_present(*pgd
))
693 pud
= pud_offset(pgd
, address
);
694 if (!pud_present(*pud
))
697 pmd
= pmd_offset(pud
, address
);
699 * Some THP functions use the sequence pmdp_huge_clear_flush(), set_pmd_at()
700 * without holding anon_vma lock for write. So when looking for a
701 * genuine pmde (in which to find pte), test present and !THP together.
705 if (!pmd_present(pmde
) || pmd_trans_huge(pmde
))
711 struct page_referenced_arg
{
714 unsigned long vm_flags
;
715 struct mem_cgroup
*memcg
;
718 * arg: page_referenced_arg will be passed
720 static int page_referenced_one(struct page
*page
, struct vm_area_struct
*vma
,
721 unsigned long address
, void *arg
)
723 struct page_referenced_arg
*pra
= arg
;
724 struct page_vma_mapped_walk pvmw
= {
731 while (page_vma_mapped_walk(&pvmw
)) {
732 address
= pvmw
.address
;
734 if (vma
->vm_flags
& VM_LOCKED
) {
735 page_vma_mapped_walk_done(&pvmw
);
736 pra
->vm_flags
|= VM_LOCKED
;
737 return SWAP_FAIL
; /* To break the loop */
741 if (ptep_clear_flush_young_notify(vma
, address
,
744 * Don't treat a reference through
745 * a sequentially read mapping as such.
746 * If the page has been used in another mapping,
747 * we will catch it; if this other mapping is
748 * already gone, the unmap path will have set
749 * PG_referenced or activated the page.
751 if (likely(!(vma
->vm_flags
& VM_SEQ_READ
)))
754 } else if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE
)) {
755 if (pmdp_clear_flush_young_notify(vma
, address
,
759 /* unexpected pmd-mapped page? */
767 clear_page_idle(page
);
768 if (test_and_clear_page_young(page
))
773 pra
->vm_flags
|= vma
->vm_flags
;
777 return SWAP_SUCCESS
; /* To break the loop */
782 static bool invalid_page_referenced_vma(struct vm_area_struct
*vma
, void *arg
)
784 struct page_referenced_arg
*pra
= arg
;
785 struct mem_cgroup
*memcg
= pra
->memcg
;
787 if (!mm_match_cgroup(vma
->vm_mm
, memcg
))
794 * page_referenced - test if the page was referenced
795 * @page: the page to test
796 * @is_locked: caller holds lock on the page
797 * @memcg: target memory cgroup
798 * @vm_flags: collect encountered vma->vm_flags who actually referenced the page
800 * Quick test_and_clear_referenced for all mappings to a page,
801 * returns the number of ptes which referenced the page.
803 int page_referenced(struct page
*page
,
805 struct mem_cgroup
*memcg
,
806 unsigned long *vm_flags
)
810 struct page_referenced_arg pra
= {
811 .mapcount
= total_mapcount(page
),
814 struct rmap_walk_control rwc
= {
815 .rmap_one
= page_referenced_one
,
817 .anon_lock
= page_lock_anon_vma_read
,
821 if (!page_mapped(page
))
824 if (!page_rmapping(page
))
827 if (!is_locked
&& (!PageAnon(page
) || PageKsm(page
))) {
828 we_locked
= trylock_page(page
);
834 * If we are reclaiming on behalf of a cgroup, skip
835 * counting on behalf of references from different
839 rwc
.invalid_vma
= invalid_page_referenced_vma
;
842 ret
= rmap_walk(page
, &rwc
);
843 *vm_flags
= pra
.vm_flags
;
848 return pra
.referenced
;
851 static int page_mkclean_one(struct page
*page
, struct vm_area_struct
*vma
,
852 unsigned long address
, void *arg
)
854 struct page_vma_mapped_walk pvmw
= {
862 while (page_vma_mapped_walk(&pvmw
)) {
864 address
= pvmw
.address
;
867 pte_t
*pte
= pvmw
.pte
;
869 if (!pte_dirty(*pte
) && !pte_write(*pte
))
872 flush_cache_page(vma
, address
, pte_pfn(*pte
));
873 entry
= ptep_clear_flush(vma
, address
, pte
);
874 entry
= pte_wrprotect(entry
);
875 entry
= pte_mkclean(entry
);
876 set_pte_at(vma
->vm_mm
, address
, pte
, entry
);
879 #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
880 pmd_t
*pmd
= pvmw
.pmd
;
883 if (!pmd_dirty(*pmd
) && !pmd_write(*pmd
))
886 flush_cache_page(vma
, address
, page_to_pfn(page
));
887 entry
= pmdp_huge_clear_flush(vma
, address
, pmd
);
888 entry
= pmd_wrprotect(entry
);
889 entry
= pmd_mkclean(entry
);
890 set_pmd_at(vma
->vm_mm
, address
, pmd
, entry
);
893 /* unexpected pmd-mapped page? */
899 mmu_notifier_invalidate_page(vma
->vm_mm
, address
);
907 static bool invalid_mkclean_vma(struct vm_area_struct
*vma
, void *arg
)
909 if (vma
->vm_flags
& VM_SHARED
)
915 int page_mkclean(struct page
*page
)
918 struct address_space
*mapping
;
919 struct rmap_walk_control rwc
= {
920 .arg
= (void *)&cleaned
,
921 .rmap_one
= page_mkclean_one
,
922 .invalid_vma
= invalid_mkclean_vma
,
925 BUG_ON(!PageLocked(page
));
927 if (!page_mapped(page
))
930 mapping
= page_mapping(page
);
934 rmap_walk(page
, &rwc
);
938 EXPORT_SYMBOL_GPL(page_mkclean
);
941 * page_move_anon_rmap - move a page to our anon_vma
942 * @page: the page to move to our anon_vma
943 * @vma: the vma the page belongs to
945 * When a page belongs exclusively to one process after a COW event,
946 * that page can be moved into the anon_vma that belongs to just that
947 * process, so the rmap code will not search the parent or sibling
950 void page_move_anon_rmap(struct page
*page
, struct vm_area_struct
*vma
)
952 struct anon_vma
*anon_vma
= vma
->anon_vma
;
954 page
= compound_head(page
);
956 VM_BUG_ON_PAGE(!PageLocked(page
), page
);
957 VM_BUG_ON_VMA(!anon_vma
, vma
);
959 anon_vma
= (void *) anon_vma
+ PAGE_MAPPING_ANON
;
961 * Ensure that anon_vma and the PAGE_MAPPING_ANON bit are written
962 * simultaneously, so a concurrent reader (eg page_referenced()'s
963 * PageAnon()) will not see one without the other.
965 WRITE_ONCE(page
->mapping
, (struct address_space
*) anon_vma
);
969 * __page_set_anon_rmap - set up new anonymous rmap
970 * @page: Page to add to rmap
971 * @vma: VM area to add page to.
972 * @address: User virtual address of the mapping
973 * @exclusive: the page is exclusively owned by the current process
975 static void __page_set_anon_rmap(struct page
*page
,
976 struct vm_area_struct
*vma
, unsigned long address
, int exclusive
)
978 struct anon_vma
*anon_vma
= vma
->anon_vma
;
986 * If the page isn't exclusively mapped into this vma,
987 * we must use the _oldest_ possible anon_vma for the
991 anon_vma
= anon_vma
->root
;
993 anon_vma
= (void *) anon_vma
+ PAGE_MAPPING_ANON
;
994 page
->mapping
= (struct address_space
*) anon_vma
;
995 page
->index
= linear_page_index(vma
, address
);
999 * __page_check_anon_rmap - sanity check anonymous rmap addition
1000 * @page: the page to add the mapping to
1001 * @vma: the vm area in which the mapping is added
1002 * @address: the user virtual address mapped
1004 static void __page_check_anon_rmap(struct page
*page
,
1005 struct vm_area_struct
*vma
, unsigned long address
)
1007 #ifdef CONFIG_DEBUG_VM
1009 * The page's anon-rmap details (mapping and index) are guaranteed to
1010 * be set up correctly at this point.
1012 * We have exclusion against page_add_anon_rmap because the caller
1013 * always holds the page locked, except if called from page_dup_rmap,
1014 * in which case the page is already known to be setup.
1016 * We have exclusion against page_add_new_anon_rmap because those pages
1017 * are initially only visible via the pagetables, and the pte is locked
1018 * over the call to page_add_new_anon_rmap.
1020 BUG_ON(page_anon_vma(page
)->root
!= vma
->anon_vma
->root
);
1021 BUG_ON(page_to_pgoff(page
) != linear_page_index(vma
, address
));
1026 * page_add_anon_rmap - add pte mapping to an anonymous page
1027 * @page: the page to add the mapping to
1028 * @vma: the vm area in which the mapping is added
1029 * @address: the user virtual address mapped
1030 * @compound: charge the page as compound or small page
1032 * The caller needs to hold the pte lock, and the page must be locked in
1033 * the anon_vma case: to serialize mapping,index checking after setting,
1034 * and to ensure that PageAnon is not being upgraded racily to PageKsm
1035 * (but PageKsm is never downgraded to PageAnon).
1037 void page_add_anon_rmap(struct page
*page
,
1038 struct vm_area_struct
*vma
, unsigned long address
, bool compound
)
1040 do_page_add_anon_rmap(page
, vma
, address
, compound
? RMAP_COMPOUND
: 0);
1044 * Special version of the above for do_swap_page, which often runs
1045 * into pages that are exclusively owned by the current process.
1046 * Everybody else should continue to use page_add_anon_rmap above.
1048 void do_page_add_anon_rmap(struct page
*page
,
1049 struct vm_area_struct
*vma
, unsigned long address
, int flags
)
1051 bool compound
= flags
& RMAP_COMPOUND
;
1056 VM_BUG_ON_PAGE(!PageLocked(page
), page
);
1057 VM_BUG_ON_PAGE(!PageTransHuge(page
), page
);
1058 mapcount
= compound_mapcount_ptr(page
);
1059 first
= atomic_inc_and_test(mapcount
);
1061 first
= atomic_inc_and_test(&page
->_mapcount
);
1065 int nr
= compound
? hpage_nr_pages(page
) : 1;
1067 * We use the irq-unsafe __{inc|mod}_zone_page_stat because
1068 * these counters are not modified in interrupt context, and
1069 * pte lock(a spinlock) is held, which implies preemption
1073 __inc_node_page_state(page
, NR_ANON_THPS
);
1074 __mod_node_page_state(page_pgdat(page
), NR_ANON_MAPPED
, nr
);
1076 if (unlikely(PageKsm(page
)))
1079 VM_BUG_ON_PAGE(!PageLocked(page
), page
);
1081 /* address might be in next vma when migration races vma_adjust */
1083 __page_set_anon_rmap(page
, vma
, address
,
1084 flags
& RMAP_EXCLUSIVE
);
1086 __page_check_anon_rmap(page
, vma
, address
);
1090 * page_add_new_anon_rmap - add pte mapping to a new anonymous page
1091 * @page: the page to add the mapping to
1092 * @vma: the vm area in which the mapping is added
1093 * @address: the user virtual address mapped
1094 * @compound: charge the page as compound or small page
1096 * Same as page_add_anon_rmap but must only be called on *new* pages.
1097 * This means the inc-and-test can be bypassed.
1098 * Page does not have to be locked.
1100 void page_add_new_anon_rmap(struct page
*page
,
1101 struct vm_area_struct
*vma
, unsigned long address
, bool compound
)
1103 int nr
= compound
? hpage_nr_pages(page
) : 1;
1105 VM_BUG_ON_VMA(address
< vma
->vm_start
|| address
>= vma
->vm_end
, vma
);
1106 __SetPageSwapBacked(page
);
1108 VM_BUG_ON_PAGE(!PageTransHuge(page
), page
);
1109 /* increment count (starts at -1) */
1110 atomic_set(compound_mapcount_ptr(page
), 0);
1111 __inc_node_page_state(page
, NR_ANON_THPS
);
1113 /* Anon THP always mapped first with PMD */
1114 VM_BUG_ON_PAGE(PageTransCompound(page
), page
);
1115 /* increment count (starts at -1) */
1116 atomic_set(&page
->_mapcount
, 0);
1118 __mod_node_page_state(page_pgdat(page
), NR_ANON_MAPPED
, nr
);
1119 __page_set_anon_rmap(page
, vma
, address
, 1);
1123 * page_add_file_rmap - add pte mapping to a file page
1124 * @page: the page to add the mapping to
1126 * The caller needs to hold the pte lock.
1128 void page_add_file_rmap(struct page
*page
, bool compound
)
1132 VM_BUG_ON_PAGE(compound
&& !PageTransHuge(page
), page
);
1133 lock_page_memcg(page
);
1134 if (compound
&& PageTransHuge(page
)) {
1135 for (i
= 0, nr
= 0; i
< HPAGE_PMD_NR
; i
++) {
1136 if (atomic_inc_and_test(&page
[i
]._mapcount
))
1139 if (!atomic_inc_and_test(compound_mapcount_ptr(page
)))
1141 VM_BUG_ON_PAGE(!PageSwapBacked(page
), page
);
1142 __inc_node_page_state(page
, NR_SHMEM_PMDMAPPED
);
1144 if (PageTransCompound(page
) && page_mapping(page
)) {
1145 VM_WARN_ON_ONCE(!PageLocked(page
));
1147 SetPageDoubleMap(compound_head(page
));
1148 if (PageMlocked(page
))
1149 clear_page_mlock(compound_head(page
));
1151 if (!atomic_inc_and_test(&page
->_mapcount
))
1154 __mod_node_page_state(page_pgdat(page
), NR_FILE_MAPPED
, nr
);
1155 mem_cgroup_inc_page_stat(page
, MEM_CGROUP_STAT_FILE_MAPPED
);
1157 unlock_page_memcg(page
);
1160 static void page_remove_file_rmap(struct page
*page
, bool compound
)
1164 VM_BUG_ON_PAGE(compound
&& !PageHead(page
), page
);
1165 lock_page_memcg(page
);
1167 /* Hugepages are not counted in NR_FILE_MAPPED for now. */
1168 if (unlikely(PageHuge(page
))) {
1169 /* hugetlb pages are always mapped with pmds */
1170 atomic_dec(compound_mapcount_ptr(page
));
1174 /* page still mapped by someone else? */
1175 if (compound
&& PageTransHuge(page
)) {
1176 for (i
= 0, nr
= 0; i
< HPAGE_PMD_NR
; i
++) {
1177 if (atomic_add_negative(-1, &page
[i
]._mapcount
))
1180 if (!atomic_add_negative(-1, compound_mapcount_ptr(page
)))
1182 VM_BUG_ON_PAGE(!PageSwapBacked(page
), page
);
1183 __dec_node_page_state(page
, NR_SHMEM_PMDMAPPED
);
1185 if (!atomic_add_negative(-1, &page
->_mapcount
))
1190 * We use the irq-unsafe __{inc|mod}_zone_page_state because
1191 * these counters are not modified in interrupt context, and
1192 * pte lock(a spinlock) is held, which implies preemption disabled.
1194 __mod_node_page_state(page_pgdat(page
), NR_FILE_MAPPED
, -nr
);
1195 mem_cgroup_dec_page_stat(page
, MEM_CGROUP_STAT_FILE_MAPPED
);
1197 if (unlikely(PageMlocked(page
)))
1198 clear_page_mlock(page
);
1200 unlock_page_memcg(page
);
1203 static void page_remove_anon_compound_rmap(struct page
*page
)
1207 if (!atomic_add_negative(-1, compound_mapcount_ptr(page
)))
1210 /* Hugepages are not counted in NR_ANON_PAGES for now. */
1211 if (unlikely(PageHuge(page
)))
1214 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE
))
1217 __dec_node_page_state(page
, NR_ANON_THPS
);
1219 if (TestClearPageDoubleMap(page
)) {
1221 * Subpages can be mapped with PTEs too. Check how many of
1222 * themi are still mapped.
1224 for (i
= 0, nr
= 0; i
< HPAGE_PMD_NR
; i
++) {
1225 if (atomic_add_negative(-1, &page
[i
]._mapcount
))
1232 if (unlikely(PageMlocked(page
)))
1233 clear_page_mlock(page
);
1236 __mod_node_page_state(page_pgdat(page
), NR_ANON_MAPPED
, -nr
);
1237 deferred_split_huge_page(page
);
1242 * page_remove_rmap - take down pte mapping from a page
1243 * @page: page to remove mapping from
1244 * @compound: uncharge the page as compound or small page
1246 * The caller needs to hold the pte lock.
1248 void page_remove_rmap(struct page
*page
, bool compound
)
1250 if (!PageAnon(page
))
1251 return page_remove_file_rmap(page
, compound
);
1254 return page_remove_anon_compound_rmap(page
);
1256 /* page still mapped by someone else? */
1257 if (!atomic_add_negative(-1, &page
->_mapcount
))
1261 * We use the irq-unsafe __{inc|mod}_zone_page_stat because
1262 * these counters are not modified in interrupt context, and
1263 * pte lock(a spinlock) is held, which implies preemption disabled.
1265 __dec_node_page_state(page
, NR_ANON_MAPPED
);
1267 if (unlikely(PageMlocked(page
)))
1268 clear_page_mlock(page
);
1270 if (PageTransCompound(page
))
1271 deferred_split_huge_page(compound_head(page
));
1274 * It would be tidy to reset the PageAnon mapping here,
1275 * but that might overwrite a racing page_add_anon_rmap
1276 * which increments mapcount after us but sets mapping
1277 * before us: so leave the reset to free_hot_cold_page,
1278 * and remember that it's only reliable while mapped.
1279 * Leaving it set also helps swapoff to reinstate ptes
1280 * faster for those pages still in swapcache.
1284 struct rmap_private
{
1285 enum ttu_flags flags
;
1290 * @arg: enum ttu_flags will be passed to this argument
1292 static int try_to_unmap_one(struct page
*page
, struct vm_area_struct
*vma
,
1293 unsigned long address
, void *arg
)
1295 struct mm_struct
*mm
= vma
->vm_mm
;
1296 struct page_vma_mapped_walk pvmw
= {
1302 struct page
*subpage
;
1303 int ret
= SWAP_AGAIN
;
1304 struct rmap_private
*rp
= arg
;
1305 enum ttu_flags flags
= rp
->flags
;
1307 /* munlock has nothing to gain from examining un-locked vmas */
1308 if ((flags
& TTU_MUNLOCK
) && !(vma
->vm_flags
& VM_LOCKED
))
1311 if (flags
& TTU_SPLIT_HUGE_PMD
) {
1312 split_huge_pmd_address(vma
, address
,
1313 flags
& TTU_MIGRATION
, page
);
1316 while (page_vma_mapped_walk(&pvmw
)) {
1317 subpage
= page
- page_to_pfn(page
) + pte_pfn(*pvmw
.pte
);
1318 address
= pvmw
.address
;
1320 /* Unexpected PMD-mapped THP? */
1321 VM_BUG_ON_PAGE(!pvmw
.pte
, page
);
1324 * If the page is mlock()d, we cannot swap it out.
1325 * If it's recently referenced (perhaps page_referenced
1326 * skipped over this mm) then we should reactivate it.
1328 if (!(flags
& TTU_IGNORE_MLOCK
)) {
1329 if (vma
->vm_flags
& VM_LOCKED
) {
1330 /* PTE-mapped THP are never mlocked */
1331 if (!PageTransCompound(page
)) {
1333 * Holding pte lock, we do *not* need
1336 mlock_vma_page(page
);
1339 page_vma_mapped_walk_done(&pvmw
);
1342 if (flags
& TTU_MUNLOCK
)
1346 if (!(flags
& TTU_IGNORE_ACCESS
)) {
1347 if (ptep_clear_flush_young_notify(vma
, address
,
1350 page_vma_mapped_walk_done(&pvmw
);
1355 /* Nuke the page table entry. */
1356 flush_cache_page(vma
, address
, pte_pfn(*pvmw
.pte
));
1357 if (should_defer_flush(mm
, flags
)) {
1359 * We clear the PTE but do not flush so potentially
1360 * a remote CPU could still be writing to the page.
1361 * If the entry was previously clean then the
1362 * architecture must guarantee that a clear->dirty
1363 * transition on a cached TLB entry is written through
1364 * and traps if the PTE is unmapped.
1366 pteval
= ptep_get_and_clear(mm
, address
, pvmw
.pte
);
1368 set_tlb_ubc_flush_pending(mm
, pte_dirty(pteval
));
1370 pteval
= ptep_clear_flush(vma
, address
, pvmw
.pte
);
1373 /* Move the dirty bit to the page. Now the pte is gone. */
1374 if (pte_dirty(pteval
))
1375 set_page_dirty(page
);
1377 /* Update high watermark before we lower rss */
1378 update_hiwater_rss(mm
);
1380 if (PageHWPoison(page
) && !(flags
& TTU_IGNORE_HWPOISON
)) {
1381 if (PageHuge(page
)) {
1382 int nr
= 1 << compound_order(page
);
1383 hugetlb_count_sub(nr
, mm
);
1385 dec_mm_counter(mm
, mm_counter(page
));
1388 pteval
= swp_entry_to_pte(make_hwpoison_entry(subpage
));
1389 set_pte_at(mm
, address
, pvmw
.pte
, pteval
);
1390 } else if (pte_unused(pteval
)) {
1392 * The guest indicated that the page content is of no
1393 * interest anymore. Simply discard the pte, vmscan
1394 * will take care of the rest.
1396 dec_mm_counter(mm
, mm_counter(page
));
1397 } else if (IS_ENABLED(CONFIG_MIGRATION
) &&
1398 (flags
& TTU_MIGRATION
)) {
1402 * Store the pfn of the page in a special migration
1403 * pte. do_swap_page() will wait until the migration
1404 * pte is removed and then restart fault handling.
1406 entry
= make_migration_entry(subpage
,
1408 swp_pte
= swp_entry_to_pte(entry
);
1409 if (pte_soft_dirty(pteval
))
1410 swp_pte
= pte_swp_mksoft_dirty(swp_pte
);
1411 set_pte_at(mm
, address
, pvmw
.pte
, swp_pte
);
1412 } else if (PageAnon(page
)) {
1413 swp_entry_t entry
= { .val
= page_private(subpage
) };
1416 * Store the swap location in the pte.
1417 * See handle_pte_fault() ...
1419 VM_BUG_ON_PAGE(!PageSwapCache(page
), page
);
1421 if (!PageDirty(page
) && (flags
& TTU_LZFREE
)) {
1422 /* It's a freeable page by MADV_FREE */
1423 dec_mm_counter(mm
, MM_ANONPAGES
);
1428 if (swap_duplicate(entry
) < 0) {
1429 set_pte_at(mm
, address
, pvmw
.pte
, pteval
);
1431 page_vma_mapped_walk_done(&pvmw
);
1434 if (list_empty(&mm
->mmlist
)) {
1435 spin_lock(&mmlist_lock
);
1436 if (list_empty(&mm
->mmlist
))
1437 list_add(&mm
->mmlist
, &init_mm
.mmlist
);
1438 spin_unlock(&mmlist_lock
);
1440 dec_mm_counter(mm
, MM_ANONPAGES
);
1441 inc_mm_counter(mm
, MM_SWAPENTS
);
1442 swp_pte
= swp_entry_to_pte(entry
);
1443 if (pte_soft_dirty(pteval
))
1444 swp_pte
= pte_swp_mksoft_dirty(swp_pte
);
1445 set_pte_at(mm
, address
, pvmw
.pte
, swp_pte
);
1447 dec_mm_counter(mm
, mm_counter_file(page
));
1449 page_remove_rmap(subpage
, PageHuge(page
));
1451 mmu_notifier_invalidate_page(mm
, address
);
1456 bool is_vma_temporary_stack(struct vm_area_struct
*vma
)
1458 int maybe_stack
= vma
->vm_flags
& (VM_GROWSDOWN
| VM_GROWSUP
);
1463 if ((vma
->vm_flags
& VM_STACK_INCOMPLETE_SETUP
) ==
1464 VM_STACK_INCOMPLETE_SETUP
)
1470 static bool invalid_migration_vma(struct vm_area_struct
*vma
, void *arg
)
1472 return is_vma_temporary_stack(vma
);
1475 static int page_mapcount_is_zero(struct page
*page
)
1477 return !total_mapcount(page
);
1481 * try_to_unmap - try to remove all page table mappings to a page
1482 * @page: the page to get unmapped
1483 * @flags: action and flags
1485 * Tries to remove all the page table entries which are mapping this
1486 * page, used in the pageout path. Caller must hold the page lock.
1487 * Return values are:
1489 * SWAP_SUCCESS - we succeeded in removing all mappings
1490 * SWAP_AGAIN - we missed a mapping, try again later
1491 * SWAP_FAIL - the page is unswappable
1492 * SWAP_MLOCK - page is mlocked.
1494 int try_to_unmap(struct page
*page
, enum ttu_flags flags
)
1497 struct rmap_private rp
= {
1502 struct rmap_walk_control rwc
= {
1503 .rmap_one
= try_to_unmap_one
,
1505 .done
= page_mapcount_is_zero
,
1506 .anon_lock
= page_lock_anon_vma_read
,
1510 * During exec, a temporary VMA is setup and later moved.
1511 * The VMA is moved under the anon_vma lock but not the
1512 * page tables leading to a race where migration cannot
1513 * find the migration ptes. Rather than increasing the
1514 * locking requirements of exec(), migration skips
1515 * temporary VMAs until after exec() completes.
1517 if ((flags
& TTU_MIGRATION
) && !PageKsm(page
) && PageAnon(page
))
1518 rwc
.invalid_vma
= invalid_migration_vma
;
1520 if (flags
& TTU_RMAP_LOCKED
)
1521 ret
= rmap_walk_locked(page
, &rwc
);
1523 ret
= rmap_walk(page
, &rwc
);
1525 if (ret
!= SWAP_MLOCK
&& !page_mapcount(page
)) {
1527 if (rp
.lazyfreed
&& !PageDirty(page
))
1533 static int page_not_mapped(struct page
*page
)
1535 return !page_mapped(page
);
1539 * try_to_munlock - try to munlock a page
1540 * @page: the page to be munlocked
1542 * Called from munlock code. Checks all of the VMAs mapping the page
1543 * to make sure nobody else has this page mlocked. The page will be
1544 * returned with PG_mlocked cleared if no other vmas have it mlocked.
1546 * Return values are:
1548 * SWAP_AGAIN - no vma is holding page mlocked, or,
1549 * SWAP_AGAIN - page mapped in mlocked vma -- couldn't acquire mmap sem
1550 * SWAP_FAIL - page cannot be located at present
1551 * SWAP_MLOCK - page is now mlocked.
1553 int try_to_munlock(struct page
*page
)
1556 struct rmap_private rp
= {
1557 .flags
= TTU_MUNLOCK
,
1561 struct rmap_walk_control rwc
= {
1562 .rmap_one
= try_to_unmap_one
,
1564 .done
= page_not_mapped
,
1565 .anon_lock
= page_lock_anon_vma_read
,
1569 VM_BUG_ON_PAGE(!PageLocked(page
) || PageLRU(page
), page
);
1571 ret
= rmap_walk(page
, &rwc
);
1575 void __put_anon_vma(struct anon_vma
*anon_vma
)
1577 struct anon_vma
*root
= anon_vma
->root
;
1579 anon_vma_free(anon_vma
);
1580 if (root
!= anon_vma
&& atomic_dec_and_test(&root
->refcount
))
1581 anon_vma_free(root
);
1584 static struct anon_vma
*rmap_walk_anon_lock(struct page
*page
,
1585 struct rmap_walk_control
*rwc
)
1587 struct anon_vma
*anon_vma
;
1590 return rwc
->anon_lock(page
);
1593 * Note: remove_migration_ptes() cannot use page_lock_anon_vma_read()
1594 * because that depends on page_mapped(); but not all its usages
1595 * are holding mmap_sem. Users without mmap_sem are required to
1596 * take a reference count to prevent the anon_vma disappearing
1598 anon_vma
= page_anon_vma(page
);
1602 anon_vma_lock_read(anon_vma
);
1607 * rmap_walk_anon - do something to anonymous page using the object-based
1609 * @page: the page to be handled
1610 * @rwc: control variable according to each walk type
1612 * Find all the mappings of a page using the mapping pointer and the vma chains
1613 * contained in the anon_vma struct it points to.
1615 * When called from try_to_munlock(), the mmap_sem of the mm containing the vma
1616 * where the page was found will be held for write. So, we won't recheck
1617 * vm_flags for that VMA. That should be OK, because that vma shouldn't be
1620 static int rmap_walk_anon(struct page
*page
, struct rmap_walk_control
*rwc
,
1623 struct anon_vma
*anon_vma
;
1624 pgoff_t pgoff_start
, pgoff_end
;
1625 struct anon_vma_chain
*avc
;
1626 int ret
= SWAP_AGAIN
;
1629 anon_vma
= page_anon_vma(page
);
1630 /* anon_vma disappear under us? */
1631 VM_BUG_ON_PAGE(!anon_vma
, page
);
1633 anon_vma
= rmap_walk_anon_lock(page
, rwc
);
1638 pgoff_start
= page_to_pgoff(page
);
1639 pgoff_end
= pgoff_start
+ hpage_nr_pages(page
) - 1;
1640 anon_vma_interval_tree_foreach(avc
, &anon_vma
->rb_root
,
1641 pgoff_start
, pgoff_end
) {
1642 struct vm_area_struct
*vma
= avc
->vma
;
1643 unsigned long address
= vma_address(page
, vma
);
1647 if (rwc
->invalid_vma
&& rwc
->invalid_vma(vma
, rwc
->arg
))
1650 ret
= rwc
->rmap_one(page
, vma
, address
, rwc
->arg
);
1651 if (ret
!= SWAP_AGAIN
)
1653 if (rwc
->done
&& rwc
->done(page
))
1658 anon_vma_unlock_read(anon_vma
);
1663 * rmap_walk_file - do something to file page using the object-based rmap method
1664 * @page: the page to be handled
1665 * @rwc: control variable according to each walk type
1667 * Find all the mappings of a page using the mapping pointer and the vma chains
1668 * contained in the address_space struct it points to.
1670 * When called from try_to_munlock(), the mmap_sem of the mm containing the vma
1671 * where the page was found will be held for write. So, we won't recheck
1672 * vm_flags for that VMA. That should be OK, because that vma shouldn't be
1675 static int rmap_walk_file(struct page
*page
, struct rmap_walk_control
*rwc
,
1678 struct address_space
*mapping
= page_mapping(page
);
1679 pgoff_t pgoff_start
, pgoff_end
;
1680 struct vm_area_struct
*vma
;
1681 int ret
= SWAP_AGAIN
;
1684 * The page lock not only makes sure that page->mapping cannot
1685 * suddenly be NULLified by truncation, it makes sure that the
1686 * structure at mapping cannot be freed and reused yet,
1687 * so we can safely take mapping->i_mmap_rwsem.
1689 VM_BUG_ON_PAGE(!PageLocked(page
), page
);
1694 pgoff_start
= page_to_pgoff(page
);
1695 pgoff_end
= pgoff_start
+ hpage_nr_pages(page
) - 1;
1697 i_mmap_lock_read(mapping
);
1698 vma_interval_tree_foreach(vma
, &mapping
->i_mmap
,
1699 pgoff_start
, pgoff_end
) {
1700 unsigned long address
= vma_address(page
, vma
);
1704 if (rwc
->invalid_vma
&& rwc
->invalid_vma(vma
, rwc
->arg
))
1707 ret
= rwc
->rmap_one(page
, vma
, address
, rwc
->arg
);
1708 if (ret
!= SWAP_AGAIN
)
1710 if (rwc
->done
&& rwc
->done(page
))
1716 i_mmap_unlock_read(mapping
);
1720 int rmap_walk(struct page
*page
, struct rmap_walk_control
*rwc
)
1722 if (unlikely(PageKsm(page
)))
1723 return rmap_walk_ksm(page
, rwc
);
1724 else if (PageAnon(page
))
1725 return rmap_walk_anon(page
, rwc
, false);
1727 return rmap_walk_file(page
, rwc
, false);
1730 /* Like rmap_walk, but caller holds relevant rmap lock */
1731 int rmap_walk_locked(struct page
*page
, struct rmap_walk_control
*rwc
)
1733 /* no ksm support for now */
1734 VM_BUG_ON_PAGE(PageKsm(page
), page
);
1736 return rmap_walk_anon(page
, rwc
, true);
1738 return rmap_walk_file(page
, rwc
, true);
1741 #ifdef CONFIG_HUGETLB_PAGE
1743 * The following three functions are for anonymous (private mapped) hugepages.
1744 * Unlike common anonymous pages, anonymous hugepages have no accounting code
1745 * and no lru code, because we handle hugepages differently from common pages.
1747 static void __hugepage_set_anon_rmap(struct page
*page
,
1748 struct vm_area_struct
*vma
, unsigned long address
, int exclusive
)
1750 struct anon_vma
*anon_vma
= vma
->anon_vma
;
1757 anon_vma
= anon_vma
->root
;
1759 anon_vma
= (void *) anon_vma
+ PAGE_MAPPING_ANON
;
1760 page
->mapping
= (struct address_space
*) anon_vma
;
1761 page
->index
= linear_page_index(vma
, address
);
1764 void hugepage_add_anon_rmap(struct page
*page
,
1765 struct vm_area_struct
*vma
, unsigned long address
)
1767 struct anon_vma
*anon_vma
= vma
->anon_vma
;
1770 BUG_ON(!PageLocked(page
));
1772 /* address might be in next vma when migration races vma_adjust */
1773 first
= atomic_inc_and_test(compound_mapcount_ptr(page
));
1775 __hugepage_set_anon_rmap(page
, vma
, address
, 0);
1778 void hugepage_add_new_anon_rmap(struct page
*page
,
1779 struct vm_area_struct
*vma
, unsigned long address
)
1781 BUG_ON(address
< vma
->vm_start
|| address
>= vma
->vm_end
);
1782 atomic_set(compound_mapcount_ptr(page
), 0);
1783 __hugepage_set_anon_rmap(page
, vma
, address
, 1);
1785 #endif /* CONFIG_HUGETLB_PAGE */