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
|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
571 static void percpu_flush_tlb_batch_pages(void *data
)
574 * All TLB entries are flushed on the assumption that it is
575 * cheaper to flush all TLBs and let them be refilled than
576 * flushing individual PFNs. Note that we do not track mm's
577 * to flush as that might simply be multiple full TLB flushes
580 count_vm_tlb_event(NR_TLB_REMOTE_FLUSH_RECEIVED
);
585 * Flush TLB entries for recently unmapped pages from remote CPUs. It is
586 * important if a PTE was dirty when it was unmapped that it's flushed
587 * before any IO is initiated on the page to prevent lost writes. Similarly,
588 * it must be flushed before freeing to prevent data leakage.
590 void try_to_unmap_flush(void)
592 struct tlbflush_unmap_batch
*tlb_ubc
= ¤t
->tlb_ubc
;
595 if (!tlb_ubc
->flush_required
)
600 trace_tlb_flush(TLB_REMOTE_SHOOTDOWN
, -1UL);
602 if (cpumask_test_cpu(cpu
, &tlb_ubc
->cpumask
))
603 percpu_flush_tlb_batch_pages(&tlb_ubc
->cpumask
);
605 if (cpumask_any_but(&tlb_ubc
->cpumask
, cpu
) < nr_cpu_ids
) {
606 smp_call_function_many(&tlb_ubc
->cpumask
,
607 percpu_flush_tlb_batch_pages
, (void *)tlb_ubc
, true);
609 cpumask_clear(&tlb_ubc
->cpumask
);
610 tlb_ubc
->flush_required
= false;
611 tlb_ubc
->writable
= false;
615 /* Flush iff there are potentially writable TLB entries that can race with IO */
616 void try_to_unmap_flush_dirty(void)
618 struct tlbflush_unmap_batch
*tlb_ubc
= ¤t
->tlb_ubc
;
620 if (tlb_ubc
->writable
)
621 try_to_unmap_flush();
624 static void set_tlb_ubc_flush_pending(struct mm_struct
*mm
,
625 struct page
*page
, bool writable
)
627 struct tlbflush_unmap_batch
*tlb_ubc
= ¤t
->tlb_ubc
;
629 cpumask_or(&tlb_ubc
->cpumask
, &tlb_ubc
->cpumask
, mm_cpumask(mm
));
630 tlb_ubc
->flush_required
= true;
633 * If the PTE was dirty then it's best to assume it's writable. The
634 * caller must use try_to_unmap_flush_dirty() or try_to_unmap_flush()
635 * before the page is queued for IO.
638 tlb_ubc
->writable
= true;
642 * Returns true if the TLB flush should be deferred to the end of a batch of
643 * unmap operations to reduce IPIs.
645 static bool should_defer_flush(struct mm_struct
*mm
, enum ttu_flags flags
)
647 bool should_defer
= false;
649 if (!(flags
& TTU_BATCH_FLUSH
))
652 /* If remote CPUs need to be flushed then defer batch the flush */
653 if (cpumask_any_but(mm_cpumask(mm
), get_cpu()) < nr_cpu_ids
)
660 static void set_tlb_ubc_flush_pending(struct mm_struct
*mm
,
661 struct page
*page
, bool writable
)
665 static bool should_defer_flush(struct mm_struct
*mm
, enum ttu_flags flags
)
669 #endif /* CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH */
672 * At what user virtual address is page expected in vma?
673 * Caller should check the page is actually part of the vma.
675 unsigned long page_address_in_vma(struct page
*page
, struct vm_area_struct
*vma
)
677 unsigned long address
;
678 if (PageAnon(page
)) {
679 struct anon_vma
*page__anon_vma
= page_anon_vma(page
);
681 * Note: swapoff's unuse_vma() is more efficient with this
682 * check, and needs it to match anon_vma when KSM is active.
684 if (!vma
->anon_vma
|| !page__anon_vma
||
685 vma
->anon_vma
->root
!= page__anon_vma
->root
)
687 } else if (page
->mapping
) {
688 if (!vma
->vm_file
|| vma
->vm_file
->f_mapping
!= page
->mapping
)
692 address
= __vma_address(page
, vma
);
693 if (unlikely(address
< vma
->vm_start
|| address
>= vma
->vm_end
))
698 pmd_t
*mm_find_pmd(struct mm_struct
*mm
, unsigned long address
)
705 pgd
= pgd_offset(mm
, address
);
706 if (!pgd_present(*pgd
))
709 pud
= pud_offset(pgd
, address
);
710 if (!pud_present(*pud
))
713 pmd
= pmd_offset(pud
, address
);
715 * Some THP functions use the sequence pmdp_huge_clear_flush(), set_pmd_at()
716 * without holding anon_vma lock for write. So when looking for a
717 * genuine pmde (in which to find pte), test present and !THP together.
721 if (!pmd_present(pmde
) || pmd_trans_huge(pmde
))
728 * Check that @page is mapped at @address into @mm.
730 * If @sync is false, page_check_address may perform a racy check to avoid
731 * the page table lock when the pte is not present (helpful when reclaiming
732 * highly shared pages).
734 * On success returns with pte mapped and locked.
736 pte_t
*__page_check_address(struct page
*page
, struct mm_struct
*mm
,
737 unsigned long address
, spinlock_t
**ptlp
, int sync
)
743 if (unlikely(PageHuge(page
))) {
744 /* when pud is not present, pte will be NULL */
745 pte
= huge_pte_offset(mm
, address
);
749 ptl
= huge_pte_lockptr(page_hstate(page
), mm
, pte
);
753 pmd
= mm_find_pmd(mm
, address
);
757 pte
= pte_offset_map(pmd
, address
);
758 /* Make a quick check before getting the lock */
759 if (!sync
&& !pte_present(*pte
)) {
764 ptl
= pte_lockptr(mm
, pmd
);
767 if (pte_present(*pte
) && page_to_pfn(page
) == pte_pfn(*pte
)) {
771 pte_unmap_unlock(pte
, ptl
);
776 * page_mapped_in_vma - check whether a page is really mapped in a VMA
777 * @page: the page to test
778 * @vma: the VMA to test
780 * Returns 1 if the page is mapped into the page tables of the VMA, 0
781 * if the page is not mapped into the page tables of this VMA. Only
782 * valid for normal file or anonymous VMAs.
784 int page_mapped_in_vma(struct page
*page
, struct vm_area_struct
*vma
)
786 unsigned long address
;
790 address
= __vma_address(page
, vma
);
791 if (unlikely(address
< vma
->vm_start
|| address
>= vma
->vm_end
))
793 pte
= page_check_address(page
, vma
->vm_mm
, address
, &ptl
, 1);
794 if (!pte
) /* the page is not in this mm */
796 pte_unmap_unlock(pte
, ptl
);
801 struct page_referenced_arg
{
804 unsigned long vm_flags
;
805 struct mem_cgroup
*memcg
;
808 * arg: page_referenced_arg will be passed
810 static int page_referenced_one(struct page
*page
, struct vm_area_struct
*vma
,
811 unsigned long address
, void *arg
)
813 struct mm_struct
*mm
= vma
->vm_mm
;
816 struct page_referenced_arg
*pra
= arg
;
822 if (unlikely(PageHuge(page
))) {
823 /* when pud is not present, pte will be NULL */
824 pte
= huge_pte_offset(mm
, address
);
828 ptl
= huge_pte_lockptr(page_hstate(page
), mm
, pte
);
832 pgd
= pgd_offset(mm
, address
);
833 if (!pgd_present(*pgd
))
835 pud
= pud_offset(pgd
, address
);
836 if (!pud_present(*pud
))
838 pmd
= pmd_offset(pud
, address
);
840 if (pmd_trans_huge(*pmd
)) {
841 int ret
= SWAP_AGAIN
;
843 ptl
= pmd_lock(mm
, pmd
);
844 if (!pmd_present(*pmd
))
846 if (unlikely(!pmd_trans_huge(*pmd
))) {
851 if (pmd_page(*pmd
) != page
)
854 if (vma
->vm_flags
& VM_LOCKED
) {
855 pra
->vm_flags
|= VM_LOCKED
;
856 ret
= SWAP_FAIL
; /* To break the loop */
860 if (pmdp_clear_flush_young_notify(vma
, address
, pmd
))
871 if (!pmd_present(pmde
) || pmd_trans_huge(pmde
))
875 pte
= pte_offset_map(pmd
, address
);
876 if (!pte_present(*pte
)) {
881 ptl
= pte_lockptr(mm
, pmd
);
885 if (!pte_present(*pte
)) {
886 pte_unmap_unlock(pte
, ptl
);
890 /* THP can be referenced by any subpage */
891 if (pte_pfn(*pte
) - page_to_pfn(page
) >= hpage_nr_pages(page
)) {
892 pte_unmap_unlock(pte
, ptl
);
896 if (vma
->vm_flags
& VM_LOCKED
) {
897 pte_unmap_unlock(pte
, ptl
);
898 pra
->vm_flags
|= VM_LOCKED
;
899 return SWAP_FAIL
; /* To break the loop */
902 if (ptep_clear_flush_young_notify(vma
, address
, pte
)) {
904 * Don't treat a reference through a sequentially read
905 * mapping as such. If the page has been used in
906 * another mapping, we will catch it; if this other
907 * mapping is already gone, the unmap path will have
908 * set PG_referenced or activated the page.
910 if (likely(!(vma
->vm_flags
& VM_SEQ_READ
)))
913 pte_unmap_unlock(pte
, ptl
);
917 clear_page_idle(page
);
918 if (test_and_clear_page_young(page
))
923 pra
->vm_flags
|= vma
->vm_flags
;
928 return SWAP_SUCCESS
; /* To break the loop */
933 static bool invalid_page_referenced_vma(struct vm_area_struct
*vma
, void *arg
)
935 struct page_referenced_arg
*pra
= arg
;
936 struct mem_cgroup
*memcg
= pra
->memcg
;
938 if (!mm_match_cgroup(vma
->vm_mm
, memcg
))
945 * page_referenced - test if the page was referenced
946 * @page: the page to test
947 * @is_locked: caller holds lock on the page
948 * @memcg: target memory cgroup
949 * @vm_flags: collect encountered vma->vm_flags who actually referenced the page
951 * Quick test_and_clear_referenced for all mappings to a page,
952 * returns the number of ptes which referenced the page.
954 int page_referenced(struct page
*page
,
956 struct mem_cgroup
*memcg
,
957 unsigned long *vm_flags
)
961 struct page_referenced_arg pra
= {
962 .mapcount
= total_mapcount(page
),
965 struct rmap_walk_control rwc
= {
966 .rmap_one
= page_referenced_one
,
968 .anon_lock
= page_lock_anon_vma_read
,
972 if (!page_mapped(page
))
975 if (!page_rmapping(page
))
978 if (!is_locked
&& (!PageAnon(page
) || PageKsm(page
))) {
979 we_locked
= trylock_page(page
);
985 * If we are reclaiming on behalf of a cgroup, skip
986 * counting on behalf of references from different
990 rwc
.invalid_vma
= invalid_page_referenced_vma
;
993 ret
= rmap_walk(page
, &rwc
);
994 *vm_flags
= pra
.vm_flags
;
999 return pra
.referenced
;
1002 static int page_mkclean_one(struct page
*page
, struct vm_area_struct
*vma
,
1003 unsigned long address
, void *arg
)
1005 struct mm_struct
*mm
= vma
->vm_mm
;
1011 pte
= page_check_address(page
, mm
, address
, &ptl
, 1);
1015 if (pte_dirty(*pte
) || pte_write(*pte
)) {
1018 flush_cache_page(vma
, address
, pte_pfn(*pte
));
1019 entry
= ptep_clear_flush(vma
, address
, pte
);
1020 entry
= pte_wrprotect(entry
);
1021 entry
= pte_mkclean(entry
);
1022 set_pte_at(mm
, address
, pte
, entry
);
1026 pte_unmap_unlock(pte
, ptl
);
1029 mmu_notifier_invalidate_page(mm
, address
);
1036 static bool invalid_mkclean_vma(struct vm_area_struct
*vma
, void *arg
)
1038 if (vma
->vm_flags
& VM_SHARED
)
1044 int page_mkclean(struct page
*page
)
1047 struct address_space
*mapping
;
1048 struct rmap_walk_control rwc
= {
1049 .arg
= (void *)&cleaned
,
1050 .rmap_one
= page_mkclean_one
,
1051 .invalid_vma
= invalid_mkclean_vma
,
1054 BUG_ON(!PageLocked(page
));
1056 if (!page_mapped(page
))
1059 mapping
= page_mapping(page
);
1063 rmap_walk(page
, &rwc
);
1067 EXPORT_SYMBOL_GPL(page_mkclean
);
1070 * page_move_anon_rmap - move a page to our anon_vma
1071 * @page: the page to move to our anon_vma
1072 * @vma: the vma the page belongs to
1073 * @address: the user virtual address mapped
1075 * When a page belongs exclusively to one process after a COW event,
1076 * that page can be moved into the anon_vma that belongs to just that
1077 * process, so the rmap code will not search the parent or sibling
1080 void page_move_anon_rmap(struct page
*page
,
1081 struct vm_area_struct
*vma
, unsigned long address
)
1083 struct anon_vma
*anon_vma
= vma
->anon_vma
;
1085 VM_BUG_ON_PAGE(!PageLocked(page
), page
);
1086 VM_BUG_ON_VMA(!anon_vma
, vma
);
1087 VM_BUG_ON_PAGE(page
->index
!= linear_page_index(vma
, address
), page
);
1089 anon_vma
= (void *) anon_vma
+ PAGE_MAPPING_ANON
;
1091 * Ensure that anon_vma and the PAGE_MAPPING_ANON bit are written
1092 * simultaneously, so a concurrent reader (eg page_referenced()'s
1093 * PageAnon()) will not see one without the other.
1095 WRITE_ONCE(page
->mapping
, (struct address_space
*) anon_vma
);
1099 * __page_set_anon_rmap - set up new anonymous rmap
1100 * @page: Page to add to rmap
1101 * @vma: VM area to add page to.
1102 * @address: User virtual address of the mapping
1103 * @exclusive: the page is exclusively owned by the current process
1105 static void __page_set_anon_rmap(struct page
*page
,
1106 struct vm_area_struct
*vma
, unsigned long address
, int exclusive
)
1108 struct anon_vma
*anon_vma
= vma
->anon_vma
;
1116 * If the page isn't exclusively mapped into this vma,
1117 * we must use the _oldest_ possible anon_vma for the
1121 anon_vma
= anon_vma
->root
;
1123 anon_vma
= (void *) anon_vma
+ PAGE_MAPPING_ANON
;
1124 page
->mapping
= (struct address_space
*) anon_vma
;
1125 page
->index
= linear_page_index(vma
, address
);
1129 * __page_check_anon_rmap - sanity check anonymous rmap addition
1130 * @page: the page to add the mapping to
1131 * @vma: the vm area in which the mapping is added
1132 * @address: the user virtual address mapped
1134 static void __page_check_anon_rmap(struct page
*page
,
1135 struct vm_area_struct
*vma
, unsigned long address
)
1137 #ifdef CONFIG_DEBUG_VM
1139 * The page's anon-rmap details (mapping and index) are guaranteed to
1140 * be set up correctly at this point.
1142 * We have exclusion against page_add_anon_rmap because the caller
1143 * always holds the page locked, except if called from page_dup_rmap,
1144 * in which case the page is already known to be setup.
1146 * We have exclusion against page_add_new_anon_rmap because those pages
1147 * are initially only visible via the pagetables, and the pte is locked
1148 * over the call to page_add_new_anon_rmap.
1150 BUG_ON(page_anon_vma(page
)->root
!= vma
->anon_vma
->root
);
1151 BUG_ON(page_to_pgoff(page
) != linear_page_index(vma
, address
));
1156 * page_add_anon_rmap - add pte mapping to an anonymous page
1157 * @page: the page to add the mapping to
1158 * @vma: the vm area in which the mapping is added
1159 * @address: the user virtual address mapped
1160 * @compound: charge the page as compound or small page
1162 * The caller needs to hold the pte lock, and the page must be locked in
1163 * the anon_vma case: to serialize mapping,index checking after setting,
1164 * and to ensure that PageAnon is not being upgraded racily to PageKsm
1165 * (but PageKsm is never downgraded to PageAnon).
1167 void page_add_anon_rmap(struct page
*page
,
1168 struct vm_area_struct
*vma
, unsigned long address
, bool compound
)
1170 do_page_add_anon_rmap(page
, vma
, address
, compound
? RMAP_COMPOUND
: 0);
1174 * Special version of the above for do_swap_page, which often runs
1175 * into pages that are exclusively owned by the current process.
1176 * Everybody else should continue to use page_add_anon_rmap above.
1178 void do_page_add_anon_rmap(struct page
*page
,
1179 struct vm_area_struct
*vma
, unsigned long address
, int flags
)
1181 bool compound
= flags
& RMAP_COMPOUND
;
1186 VM_BUG_ON_PAGE(!PageLocked(page
), page
);
1187 VM_BUG_ON_PAGE(!PageTransHuge(page
), page
);
1188 mapcount
= compound_mapcount_ptr(page
);
1189 first
= atomic_inc_and_test(mapcount
);
1191 first
= atomic_inc_and_test(&page
->_mapcount
);
1195 int nr
= compound
? hpage_nr_pages(page
) : 1;
1197 * We use the irq-unsafe __{inc|mod}_zone_page_stat because
1198 * these counters are not modified in interrupt context, and
1199 * pte lock(a spinlock) is held, which implies preemption
1203 __inc_zone_page_state(page
,
1204 NR_ANON_TRANSPARENT_HUGEPAGES
);
1206 __mod_zone_page_state(page_zone(page
), NR_ANON_PAGES
, nr
);
1208 if (unlikely(PageKsm(page
)))
1211 VM_BUG_ON_PAGE(!PageLocked(page
), page
);
1213 /* address might be in next vma when migration races vma_adjust */
1215 __page_set_anon_rmap(page
, vma
, address
,
1216 flags
& RMAP_EXCLUSIVE
);
1218 __page_check_anon_rmap(page
, vma
, address
);
1222 * page_add_new_anon_rmap - add pte mapping to a new anonymous page
1223 * @page: the page to add the mapping to
1224 * @vma: the vm area in which the mapping is added
1225 * @address: the user virtual address mapped
1226 * @compound: charge the page as compound or small page
1228 * Same as page_add_anon_rmap but must only be called on *new* pages.
1229 * This means the inc-and-test can be bypassed.
1230 * Page does not have to be locked.
1232 void page_add_new_anon_rmap(struct page
*page
,
1233 struct vm_area_struct
*vma
, unsigned long address
, bool compound
)
1235 int nr
= compound
? hpage_nr_pages(page
) : 1;
1237 VM_BUG_ON_VMA(address
< vma
->vm_start
|| address
>= vma
->vm_end
, vma
);
1238 SetPageSwapBacked(page
);
1240 VM_BUG_ON_PAGE(!PageTransHuge(page
), page
);
1241 /* increment count (starts at -1) */
1242 atomic_set(compound_mapcount_ptr(page
), 0);
1243 __inc_zone_page_state(page
, NR_ANON_TRANSPARENT_HUGEPAGES
);
1245 /* Anon THP always mapped first with PMD */
1246 VM_BUG_ON_PAGE(PageTransCompound(page
), page
);
1247 /* increment count (starts at -1) */
1248 atomic_set(&page
->_mapcount
, 0);
1250 __mod_zone_page_state(page_zone(page
), NR_ANON_PAGES
, nr
);
1251 __page_set_anon_rmap(page
, vma
, address
, 1);
1255 * page_add_file_rmap - add pte mapping to a file page
1256 * @page: the page to add the mapping to
1258 * The caller needs to hold the pte lock.
1260 void page_add_file_rmap(struct page
*page
)
1262 struct mem_cgroup
*memcg
;
1264 memcg
= mem_cgroup_begin_page_stat(page
);
1265 if (atomic_inc_and_test(&page
->_mapcount
)) {
1266 __inc_zone_page_state(page
, NR_FILE_MAPPED
);
1267 mem_cgroup_inc_page_stat(memcg
, MEM_CGROUP_STAT_FILE_MAPPED
);
1269 mem_cgroup_end_page_stat(memcg
);
1272 static void page_remove_file_rmap(struct page
*page
)
1274 struct mem_cgroup
*memcg
;
1276 memcg
= mem_cgroup_begin_page_stat(page
);
1278 /* Hugepages are not counted in NR_FILE_MAPPED for now. */
1279 if (unlikely(PageHuge(page
))) {
1280 /* hugetlb pages are always mapped with pmds */
1281 atomic_dec(compound_mapcount_ptr(page
));
1285 /* page still mapped by someone else? */
1286 if (!atomic_add_negative(-1, &page
->_mapcount
))
1290 * We use the irq-unsafe __{inc|mod}_zone_page_stat because
1291 * these counters are not modified in interrupt context, and
1292 * pte lock(a spinlock) is held, which implies preemption disabled.
1294 __dec_zone_page_state(page
, NR_FILE_MAPPED
);
1295 mem_cgroup_dec_page_stat(memcg
, MEM_CGROUP_STAT_FILE_MAPPED
);
1297 if (unlikely(PageMlocked(page
)))
1298 clear_page_mlock(page
);
1300 mem_cgroup_end_page_stat(memcg
);
1303 static void page_remove_anon_compound_rmap(struct page
*page
)
1307 if (!atomic_add_negative(-1, compound_mapcount_ptr(page
)))
1310 /* Hugepages are not counted in NR_ANON_PAGES for now. */
1311 if (unlikely(PageHuge(page
)))
1314 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE
))
1317 __dec_zone_page_state(page
, NR_ANON_TRANSPARENT_HUGEPAGES
);
1319 if (TestClearPageDoubleMap(page
)) {
1321 * Subpages can be mapped with PTEs too. Check how many of
1322 * themi are still mapped.
1324 for (i
= 0, nr
= 0; i
< HPAGE_PMD_NR
; i
++) {
1325 if (atomic_add_negative(-1, &page
[i
]._mapcount
))
1332 if (unlikely(PageMlocked(page
)))
1333 clear_page_mlock(page
);
1336 __mod_zone_page_state(page_zone(page
), NR_ANON_PAGES
, -nr
);
1337 deferred_split_huge_page(page
);
1342 * page_remove_rmap - take down pte mapping from a page
1343 * @page: page to remove mapping from
1344 * @compound: uncharge the page as compound or small page
1346 * The caller needs to hold the pte lock.
1348 void page_remove_rmap(struct page
*page
, bool compound
)
1350 if (!PageAnon(page
)) {
1351 VM_BUG_ON_PAGE(compound
&& !PageHuge(page
), page
);
1352 page_remove_file_rmap(page
);
1357 return page_remove_anon_compound_rmap(page
);
1359 /* page still mapped by someone else? */
1360 if (!atomic_add_negative(-1, &page
->_mapcount
))
1364 * We use the irq-unsafe __{inc|mod}_zone_page_stat because
1365 * these counters are not modified in interrupt context, and
1366 * pte lock(a spinlock) is held, which implies preemption disabled.
1368 __dec_zone_page_state(page
, NR_ANON_PAGES
);
1370 if (unlikely(PageMlocked(page
)))
1371 clear_page_mlock(page
);
1373 if (PageTransCompound(page
))
1374 deferred_split_huge_page(compound_head(page
));
1377 * It would be tidy to reset the PageAnon mapping here,
1378 * but that might overwrite a racing page_add_anon_rmap
1379 * which increments mapcount after us but sets mapping
1380 * before us: so leave the reset to free_hot_cold_page,
1381 * and remember that it's only reliable while mapped.
1382 * Leaving it set also helps swapoff to reinstate ptes
1383 * faster for those pages still in swapcache.
1388 * @arg: enum ttu_flags will be passed to this argument
1390 static int try_to_unmap_one(struct page
*page
, struct vm_area_struct
*vma
,
1391 unsigned long address
, void *arg
)
1393 struct mm_struct
*mm
= vma
->vm_mm
;
1397 int ret
= SWAP_AGAIN
;
1398 enum ttu_flags flags
= (enum ttu_flags
)arg
;
1400 /* munlock has nothing to gain from examining un-locked vmas */
1401 if ((flags
& TTU_MUNLOCK
) && !(vma
->vm_flags
& VM_LOCKED
))
1404 pte
= page_check_address(page
, mm
, address
, &ptl
, 0);
1409 * If the page is mlock()d, we cannot swap it out.
1410 * If it's recently referenced (perhaps page_referenced
1411 * skipped over this mm) then we should reactivate it.
1413 if (!(flags
& TTU_IGNORE_MLOCK
)) {
1414 if (vma
->vm_flags
& VM_LOCKED
) {
1415 /* Holding pte lock, we do *not* need mmap_sem here */
1416 mlock_vma_page(page
);
1420 if (flags
& TTU_MUNLOCK
)
1423 if (!(flags
& TTU_IGNORE_ACCESS
)) {
1424 if (ptep_clear_flush_young_notify(vma
, address
, pte
)) {
1430 /* Nuke the page table entry. */
1431 flush_cache_page(vma
, address
, page_to_pfn(page
));
1432 if (should_defer_flush(mm
, flags
)) {
1434 * We clear the PTE but do not flush so potentially a remote
1435 * CPU could still be writing to the page. If the entry was
1436 * previously clean then the architecture must guarantee that
1437 * a clear->dirty transition on a cached TLB entry is written
1438 * through and traps if the PTE is unmapped.
1440 pteval
= ptep_get_and_clear(mm
, address
, pte
);
1442 set_tlb_ubc_flush_pending(mm
, page
, pte_dirty(pteval
));
1444 pteval
= ptep_clear_flush(vma
, address
, pte
);
1447 /* Move the dirty bit to the physical page now the pte is gone. */
1448 if (pte_dirty(pteval
))
1449 set_page_dirty(page
);
1451 /* Update high watermark before we lower rss */
1452 update_hiwater_rss(mm
);
1454 if (PageHWPoison(page
) && !(flags
& TTU_IGNORE_HWPOISON
)) {
1455 if (PageHuge(page
)) {
1456 hugetlb_count_sub(1 << compound_order(page
), mm
);
1458 dec_mm_counter(mm
, mm_counter(page
));
1460 set_pte_at(mm
, address
, pte
,
1461 swp_entry_to_pte(make_hwpoison_entry(page
)));
1462 } else if (pte_unused(pteval
)) {
1464 * The guest indicated that the page content is of no
1465 * interest anymore. Simply discard the pte, vmscan
1466 * will take care of the rest.
1468 dec_mm_counter(mm
, mm_counter(page
));
1469 } else if (IS_ENABLED(CONFIG_MIGRATION
) && (flags
& TTU_MIGRATION
)) {
1473 * Store the pfn of the page in a special migration
1474 * pte. do_swap_page() will wait until the migration
1475 * pte is removed and then restart fault handling.
1477 entry
= make_migration_entry(page
, pte_write(pteval
));
1478 swp_pte
= swp_entry_to_pte(entry
);
1479 if (pte_soft_dirty(pteval
))
1480 swp_pte
= pte_swp_mksoft_dirty(swp_pte
);
1481 set_pte_at(mm
, address
, pte
, swp_pte
);
1482 } else if (PageAnon(page
)) {
1483 swp_entry_t entry
= { .val
= page_private(page
) };
1486 * Store the swap location in the pte.
1487 * See handle_pte_fault() ...
1489 VM_BUG_ON_PAGE(!PageSwapCache(page
), page
);
1490 if (swap_duplicate(entry
) < 0) {
1491 set_pte_at(mm
, address
, pte
, pteval
);
1495 if (list_empty(&mm
->mmlist
)) {
1496 spin_lock(&mmlist_lock
);
1497 if (list_empty(&mm
->mmlist
))
1498 list_add(&mm
->mmlist
, &init_mm
.mmlist
);
1499 spin_unlock(&mmlist_lock
);
1501 dec_mm_counter(mm
, MM_ANONPAGES
);
1502 inc_mm_counter(mm
, MM_SWAPENTS
);
1503 swp_pte
= swp_entry_to_pte(entry
);
1504 if (pte_soft_dirty(pteval
))
1505 swp_pte
= pte_swp_mksoft_dirty(swp_pte
);
1506 set_pte_at(mm
, address
, pte
, swp_pte
);
1508 dec_mm_counter(mm
, mm_counter_file(page
));
1510 page_remove_rmap(page
, PageHuge(page
));
1511 page_cache_release(page
);
1514 pte_unmap_unlock(pte
, ptl
);
1515 if (ret
!= SWAP_FAIL
&& ret
!= SWAP_MLOCK
&& !(flags
& TTU_MUNLOCK
))
1516 mmu_notifier_invalidate_page(mm
, address
);
1521 bool is_vma_temporary_stack(struct vm_area_struct
*vma
)
1523 int maybe_stack
= vma
->vm_flags
& (VM_GROWSDOWN
| VM_GROWSUP
);
1528 if ((vma
->vm_flags
& VM_STACK_INCOMPLETE_SETUP
) ==
1529 VM_STACK_INCOMPLETE_SETUP
)
1535 static bool invalid_migration_vma(struct vm_area_struct
*vma
, void *arg
)
1537 return is_vma_temporary_stack(vma
);
1540 static int page_not_mapped(struct page
*page
)
1542 return !page_mapped(page
);
1546 * try_to_unmap - try to remove all page table mappings to a page
1547 * @page: the page to get unmapped
1548 * @flags: action and flags
1550 * Tries to remove all the page table entries which are mapping this
1551 * page, used in the pageout path. Caller must hold the page lock.
1552 * Return values are:
1554 * SWAP_SUCCESS - we succeeded in removing all mappings
1555 * SWAP_AGAIN - we missed a mapping, try again later
1556 * SWAP_FAIL - the page is unswappable
1557 * SWAP_MLOCK - page is mlocked.
1559 int try_to_unmap(struct page
*page
, enum ttu_flags flags
)
1562 struct rmap_walk_control rwc
= {
1563 .rmap_one
= try_to_unmap_one
,
1564 .arg
= (void *)flags
,
1565 .done
= page_not_mapped
,
1566 .anon_lock
= page_lock_anon_vma_read
,
1569 VM_BUG_ON_PAGE(!PageHuge(page
) && PageTransHuge(page
), page
);
1572 * During exec, a temporary VMA is setup and later moved.
1573 * The VMA is moved under the anon_vma lock but not the
1574 * page tables leading to a race where migration cannot
1575 * find the migration ptes. Rather than increasing the
1576 * locking requirements of exec(), migration skips
1577 * temporary VMAs until after exec() completes.
1579 if ((flags
& TTU_MIGRATION
) && !PageKsm(page
) && PageAnon(page
))
1580 rwc
.invalid_vma
= invalid_migration_vma
;
1582 ret
= rmap_walk(page
, &rwc
);
1584 if (ret
!= SWAP_MLOCK
&& !page_mapped(page
))
1590 * try_to_munlock - try to munlock a page
1591 * @page: the page to be munlocked
1593 * Called from munlock code. Checks all of the VMAs mapping the page
1594 * to make sure nobody else has this page mlocked. The page will be
1595 * returned with PG_mlocked cleared if no other vmas have it mlocked.
1597 * Return values are:
1599 * SWAP_AGAIN - no vma is holding page mlocked, or,
1600 * SWAP_AGAIN - page mapped in mlocked vma -- couldn't acquire mmap sem
1601 * SWAP_FAIL - page cannot be located at present
1602 * SWAP_MLOCK - page is now mlocked.
1604 int try_to_munlock(struct page
*page
)
1607 struct rmap_walk_control rwc
= {
1608 .rmap_one
= try_to_unmap_one
,
1609 .arg
= (void *)TTU_MUNLOCK
,
1610 .done
= page_not_mapped
,
1611 .anon_lock
= page_lock_anon_vma_read
,
1615 VM_BUG_ON_PAGE(!PageLocked(page
) || PageLRU(page
), page
);
1617 ret
= rmap_walk(page
, &rwc
);
1621 void __put_anon_vma(struct anon_vma
*anon_vma
)
1623 struct anon_vma
*root
= anon_vma
->root
;
1625 anon_vma_free(anon_vma
);
1626 if (root
!= anon_vma
&& atomic_dec_and_test(&root
->refcount
))
1627 anon_vma_free(root
);
1630 static struct anon_vma
*rmap_walk_anon_lock(struct page
*page
,
1631 struct rmap_walk_control
*rwc
)
1633 struct anon_vma
*anon_vma
;
1636 return rwc
->anon_lock(page
);
1639 * Note: remove_migration_ptes() cannot use page_lock_anon_vma_read()
1640 * because that depends on page_mapped(); but not all its usages
1641 * are holding mmap_sem. Users without mmap_sem are required to
1642 * take a reference count to prevent the anon_vma disappearing
1644 anon_vma
= page_anon_vma(page
);
1648 anon_vma_lock_read(anon_vma
);
1653 * rmap_walk_anon - do something to anonymous page using the object-based
1655 * @page: the page to be handled
1656 * @rwc: control variable according to each walk type
1658 * Find all the mappings of a page using the mapping pointer and the vma chains
1659 * contained in the anon_vma struct it points to.
1661 * When called from try_to_munlock(), the mmap_sem of the mm containing the vma
1662 * where the page was found will be held for write. So, we won't recheck
1663 * vm_flags for that VMA. That should be OK, because that vma shouldn't be
1666 static int rmap_walk_anon(struct page
*page
, struct rmap_walk_control
*rwc
)
1668 struct anon_vma
*anon_vma
;
1670 struct anon_vma_chain
*avc
;
1671 int ret
= SWAP_AGAIN
;
1673 anon_vma
= rmap_walk_anon_lock(page
, rwc
);
1677 pgoff
= page_to_pgoff(page
);
1678 anon_vma_interval_tree_foreach(avc
, &anon_vma
->rb_root
, pgoff
, pgoff
) {
1679 struct vm_area_struct
*vma
= avc
->vma
;
1680 unsigned long address
= vma_address(page
, vma
);
1684 if (rwc
->invalid_vma
&& rwc
->invalid_vma(vma
, rwc
->arg
))
1687 ret
= rwc
->rmap_one(page
, vma
, address
, rwc
->arg
);
1688 if (ret
!= SWAP_AGAIN
)
1690 if (rwc
->done
&& rwc
->done(page
))
1693 anon_vma_unlock_read(anon_vma
);
1698 * rmap_walk_file - do something to file page using the object-based rmap method
1699 * @page: the page to be handled
1700 * @rwc: control variable according to each walk type
1702 * Find all the mappings of a page using the mapping pointer and the vma chains
1703 * contained in the address_space struct it points to.
1705 * When called from try_to_munlock(), the mmap_sem of the mm containing the vma
1706 * where the page was found will be held for write. So, we won't recheck
1707 * vm_flags for that VMA. That should be OK, because that vma shouldn't be
1710 static int rmap_walk_file(struct page
*page
, struct rmap_walk_control
*rwc
)
1712 struct address_space
*mapping
= page
->mapping
;
1714 struct vm_area_struct
*vma
;
1715 int ret
= SWAP_AGAIN
;
1718 * The page lock not only makes sure that page->mapping cannot
1719 * suddenly be NULLified by truncation, it makes sure that the
1720 * structure at mapping cannot be freed and reused yet,
1721 * so we can safely take mapping->i_mmap_rwsem.
1723 VM_BUG_ON_PAGE(!PageLocked(page
), page
);
1728 pgoff
= page_to_pgoff(page
);
1729 i_mmap_lock_read(mapping
);
1730 vma_interval_tree_foreach(vma
, &mapping
->i_mmap
, pgoff
, pgoff
) {
1731 unsigned long address
= vma_address(page
, vma
);
1735 if (rwc
->invalid_vma
&& rwc
->invalid_vma(vma
, rwc
->arg
))
1738 ret
= rwc
->rmap_one(page
, vma
, address
, rwc
->arg
);
1739 if (ret
!= SWAP_AGAIN
)
1741 if (rwc
->done
&& rwc
->done(page
))
1746 i_mmap_unlock_read(mapping
);
1750 int rmap_walk(struct page
*page
, struct rmap_walk_control
*rwc
)
1752 if (unlikely(PageKsm(page
)))
1753 return rmap_walk_ksm(page
, rwc
);
1754 else if (PageAnon(page
))
1755 return rmap_walk_anon(page
, rwc
);
1757 return rmap_walk_file(page
, rwc
);
1760 #ifdef CONFIG_HUGETLB_PAGE
1762 * The following three functions are for anonymous (private mapped) hugepages.
1763 * Unlike common anonymous pages, anonymous hugepages have no accounting code
1764 * and no lru code, because we handle hugepages differently from common pages.
1766 static void __hugepage_set_anon_rmap(struct page
*page
,
1767 struct vm_area_struct
*vma
, unsigned long address
, int exclusive
)
1769 struct anon_vma
*anon_vma
= vma
->anon_vma
;
1776 anon_vma
= anon_vma
->root
;
1778 anon_vma
= (void *) anon_vma
+ PAGE_MAPPING_ANON
;
1779 page
->mapping
= (struct address_space
*) anon_vma
;
1780 page
->index
= linear_page_index(vma
, address
);
1783 void hugepage_add_anon_rmap(struct page
*page
,
1784 struct vm_area_struct
*vma
, unsigned long address
)
1786 struct anon_vma
*anon_vma
= vma
->anon_vma
;
1789 BUG_ON(!PageLocked(page
));
1791 /* address might be in next vma when migration races vma_adjust */
1792 first
= atomic_inc_and_test(compound_mapcount_ptr(page
));
1794 __hugepage_set_anon_rmap(page
, vma
, address
, 0);
1797 void hugepage_add_new_anon_rmap(struct page
*page
,
1798 struct vm_area_struct
*vma
, unsigned long address
)
1800 BUG_ON(address
< vma
->vm_start
|| address
>= vma
->vm_end
);
1801 atomic_set(compound_mapcount_ptr(page
), 0);
1802 __hugepage_set_anon_rmap(page
, vma
, address
, 1);
1804 #endif /* CONFIG_HUGETLB_PAGE */