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>
63 #include <asm/tlbflush.h>
65 #include <trace/events/tlb.h>
69 static struct kmem_cache
*anon_vma_cachep
;
70 static struct kmem_cache
*anon_vma_chain_cachep
;
72 static inline struct anon_vma
*anon_vma_alloc(void)
74 struct anon_vma
*anon_vma
;
76 anon_vma
= kmem_cache_alloc(anon_vma_cachep
, GFP_KERNEL
);
78 atomic_set(&anon_vma
->refcount
, 1);
79 anon_vma
->degree
= 1; /* Reference for first vma */
80 anon_vma
->parent
= anon_vma
;
82 * Initialise the anon_vma root to point to itself. If called
83 * from fork, the root will be reset to the parents anon_vma.
85 anon_vma
->root
= anon_vma
;
91 static inline void anon_vma_free(struct anon_vma
*anon_vma
)
93 VM_BUG_ON(atomic_read(&anon_vma
->refcount
));
96 * Synchronize against page_lock_anon_vma_read() such that
97 * we can safely hold the lock without the anon_vma getting
100 * Relies on the full mb implied by the atomic_dec_and_test() from
101 * put_anon_vma() against the acquire barrier implied by
102 * down_read_trylock() from page_lock_anon_vma_read(). This orders:
104 * page_lock_anon_vma_read() VS put_anon_vma()
105 * down_read_trylock() atomic_dec_and_test()
107 * atomic_read() rwsem_is_locked()
109 * LOCK should suffice since the actual taking of the lock must
110 * happen _before_ what follows.
113 if (rwsem_is_locked(&anon_vma
->root
->rwsem
)) {
114 anon_vma_lock_write(anon_vma
);
115 anon_vma_unlock_write(anon_vma
);
118 kmem_cache_free(anon_vma_cachep
, anon_vma
);
121 static inline struct anon_vma_chain
*anon_vma_chain_alloc(gfp_t gfp
)
123 return kmem_cache_alloc(anon_vma_chain_cachep
, gfp
);
126 static void anon_vma_chain_free(struct anon_vma_chain
*anon_vma_chain
)
128 kmem_cache_free(anon_vma_chain_cachep
, anon_vma_chain
);
131 static void anon_vma_chain_link(struct vm_area_struct
*vma
,
132 struct anon_vma_chain
*avc
,
133 struct anon_vma
*anon_vma
)
136 avc
->anon_vma
= anon_vma
;
137 list_add(&avc
->same_vma
, &vma
->anon_vma_chain
);
138 anon_vma_interval_tree_insert(avc
, &anon_vma
->rb_root
);
142 * anon_vma_prepare - attach an anon_vma to a memory region
143 * @vma: the memory region in question
145 * This makes sure the memory mapping described by 'vma' has
146 * an 'anon_vma' attached to it, so that we can associate the
147 * anonymous pages mapped into it with that anon_vma.
149 * The common case will be that we already have one, but if
150 * not we either need to find an adjacent mapping that we
151 * can re-use the anon_vma from (very common when the only
152 * reason for splitting a vma has been mprotect()), or we
153 * allocate a new one.
155 * Anon-vma allocations are very subtle, because we may have
156 * optimistically looked up an anon_vma in page_lock_anon_vma_read()
157 * and that may actually touch the spinlock even in the newly
158 * allocated vma (it depends on RCU to make sure that the
159 * anon_vma isn't actually destroyed).
161 * As a result, we need to do proper anon_vma locking even
162 * for the new allocation. At the same time, we do not want
163 * to do any locking for the common case of already having
166 * This must be called with the mmap_sem held for reading.
168 int anon_vma_prepare(struct vm_area_struct
*vma
)
170 struct anon_vma
*anon_vma
= vma
->anon_vma
;
171 struct anon_vma_chain
*avc
;
174 if (unlikely(!anon_vma
)) {
175 struct mm_struct
*mm
= vma
->vm_mm
;
176 struct anon_vma
*allocated
;
178 avc
= anon_vma_chain_alloc(GFP_KERNEL
);
182 anon_vma
= find_mergeable_anon_vma(vma
);
185 anon_vma
= anon_vma_alloc();
186 if (unlikely(!anon_vma
))
187 goto out_enomem_free_avc
;
188 allocated
= anon_vma
;
191 anon_vma_lock_write(anon_vma
);
192 /* page_table_lock to protect against threads */
193 spin_lock(&mm
->page_table_lock
);
194 if (likely(!vma
->anon_vma
)) {
195 vma
->anon_vma
= anon_vma
;
196 anon_vma_chain_link(vma
, avc
, anon_vma
);
197 /* vma reference or self-parent link for new root */
202 spin_unlock(&mm
->page_table_lock
);
203 anon_vma_unlock_write(anon_vma
);
205 if (unlikely(allocated
))
206 put_anon_vma(allocated
);
208 anon_vma_chain_free(avc
);
213 anon_vma_chain_free(avc
);
219 * This is a useful helper function for locking the anon_vma root as
220 * we traverse the vma->anon_vma_chain, looping over anon_vma's that
223 * Such anon_vma's should have the same root, so you'd expect to see
224 * just a single mutex_lock for the whole traversal.
226 static inline struct anon_vma
*lock_anon_vma_root(struct anon_vma
*root
, struct anon_vma
*anon_vma
)
228 struct anon_vma
*new_root
= anon_vma
->root
;
229 if (new_root
!= root
) {
230 if (WARN_ON_ONCE(root
))
231 up_write(&root
->rwsem
);
233 down_write(&root
->rwsem
);
238 static inline void unlock_anon_vma_root(struct anon_vma
*root
)
241 up_write(&root
->rwsem
);
245 * Attach the anon_vmas from src to dst.
246 * Returns 0 on success, -ENOMEM on failure.
248 * If dst->anon_vma is NULL this function tries to find and reuse existing
249 * anon_vma which has no vmas and only one child anon_vma. This prevents
250 * degradation of anon_vma hierarchy to endless linear chain in case of
251 * constantly forking task. On the other hand, an anon_vma with more than one
252 * child isn't reused even if there was no alive vma, thus rmap walker has a
253 * good chance of avoiding scanning the whole hierarchy when it searches where
256 int anon_vma_clone(struct vm_area_struct
*dst
, struct vm_area_struct
*src
)
258 struct anon_vma_chain
*avc
, *pavc
;
259 struct anon_vma
*root
= NULL
;
261 list_for_each_entry_reverse(pavc
, &src
->anon_vma_chain
, same_vma
) {
262 struct anon_vma
*anon_vma
;
264 avc
= anon_vma_chain_alloc(GFP_NOWAIT
| __GFP_NOWARN
);
265 if (unlikely(!avc
)) {
266 unlock_anon_vma_root(root
);
268 avc
= anon_vma_chain_alloc(GFP_KERNEL
);
272 anon_vma
= pavc
->anon_vma
;
273 root
= lock_anon_vma_root(root
, anon_vma
);
274 anon_vma_chain_link(dst
, avc
, anon_vma
);
277 * Reuse existing anon_vma if its degree lower than two,
278 * that means it has no vma and only one anon_vma child.
280 * Do not chose parent anon_vma, otherwise first child
281 * will always reuse it. Root anon_vma is never reused:
282 * it has self-parent reference and at least one child.
284 if (!dst
->anon_vma
&& anon_vma
!= src
->anon_vma
&&
285 anon_vma
->degree
< 2)
286 dst
->anon_vma
= anon_vma
;
289 dst
->anon_vma
->degree
++;
290 unlock_anon_vma_root(root
);
295 * dst->anon_vma is dropped here otherwise its degree can be incorrectly
296 * decremented in unlink_anon_vmas().
297 * We can safely do this because callers of anon_vma_clone() don't care
298 * about dst->anon_vma if anon_vma_clone() failed.
300 dst
->anon_vma
= NULL
;
301 unlink_anon_vmas(dst
);
306 * Attach vma to its own anon_vma, as well as to the anon_vmas that
307 * the corresponding VMA in the parent process is attached to.
308 * Returns 0 on success, non-zero on failure.
310 int anon_vma_fork(struct vm_area_struct
*vma
, struct vm_area_struct
*pvma
)
312 struct anon_vma_chain
*avc
;
313 struct anon_vma
*anon_vma
;
316 /* Don't bother if the parent process has no anon_vma here. */
320 /* Drop inherited anon_vma, we'll reuse existing or allocate new. */
321 vma
->anon_vma
= NULL
;
324 * First, attach the new VMA to the parent VMA's anon_vmas,
325 * so rmap can find non-COWed pages in child processes.
327 error
= anon_vma_clone(vma
, pvma
);
331 /* An existing anon_vma has been reused, all done then. */
335 /* Then add our own anon_vma. */
336 anon_vma
= anon_vma_alloc();
339 avc
= anon_vma_chain_alloc(GFP_KERNEL
);
341 goto out_error_free_anon_vma
;
344 * The root anon_vma's spinlock is the lock actually used when we
345 * lock any of the anon_vmas in this anon_vma tree.
347 anon_vma
->root
= pvma
->anon_vma
->root
;
348 anon_vma
->parent
= pvma
->anon_vma
;
350 * With refcounts, an anon_vma can stay around longer than the
351 * process it belongs to. The root anon_vma needs to be pinned until
352 * this anon_vma is freed, because the lock lives in the root.
354 get_anon_vma(anon_vma
->root
);
355 /* Mark this anon_vma as the one where our new (COWed) pages go. */
356 vma
->anon_vma
= anon_vma
;
357 anon_vma_lock_write(anon_vma
);
358 anon_vma_chain_link(vma
, avc
, anon_vma
);
359 anon_vma
->parent
->degree
++;
360 anon_vma_unlock_write(anon_vma
);
364 out_error_free_anon_vma
:
365 put_anon_vma(anon_vma
);
367 unlink_anon_vmas(vma
);
371 void unlink_anon_vmas(struct vm_area_struct
*vma
)
373 struct anon_vma_chain
*avc
, *next
;
374 struct anon_vma
*root
= NULL
;
377 * Unlink each anon_vma chained to the VMA. This list is ordered
378 * from newest to oldest, ensuring the root anon_vma gets freed last.
380 list_for_each_entry_safe(avc
, next
, &vma
->anon_vma_chain
, same_vma
) {
381 struct anon_vma
*anon_vma
= avc
->anon_vma
;
383 root
= lock_anon_vma_root(root
, anon_vma
);
384 anon_vma_interval_tree_remove(avc
, &anon_vma
->rb_root
);
387 * Leave empty anon_vmas on the list - we'll need
388 * to free them outside the lock.
390 if (RB_EMPTY_ROOT(&anon_vma
->rb_root
)) {
391 anon_vma
->parent
->degree
--;
395 list_del(&avc
->same_vma
);
396 anon_vma_chain_free(avc
);
399 vma
->anon_vma
->degree
--;
400 unlock_anon_vma_root(root
);
403 * Iterate the list once more, it now only contains empty and unlinked
404 * anon_vmas, destroy them. Could not do before due to __put_anon_vma()
405 * needing to write-acquire the anon_vma->root->rwsem.
407 list_for_each_entry_safe(avc
, next
, &vma
->anon_vma_chain
, same_vma
) {
408 struct anon_vma
*anon_vma
= avc
->anon_vma
;
410 BUG_ON(anon_vma
->degree
);
411 put_anon_vma(anon_vma
);
413 list_del(&avc
->same_vma
);
414 anon_vma_chain_free(avc
);
418 static void anon_vma_ctor(void *data
)
420 struct anon_vma
*anon_vma
= data
;
422 init_rwsem(&anon_vma
->rwsem
);
423 atomic_set(&anon_vma
->refcount
, 0);
424 anon_vma
->rb_root
= RB_ROOT
;
427 void __init
anon_vma_init(void)
429 anon_vma_cachep
= kmem_cache_create("anon_vma", sizeof(struct anon_vma
),
430 0, SLAB_DESTROY_BY_RCU
|SLAB_PANIC
, anon_vma_ctor
);
431 anon_vma_chain_cachep
= KMEM_CACHE(anon_vma_chain
, SLAB_PANIC
);
435 * Getting a lock on a stable anon_vma from a page off the LRU is tricky!
437 * Since there is no serialization what so ever against page_remove_rmap()
438 * the best this function can do is return a locked anon_vma that might
439 * have been relevant to this page.
441 * The page might have been remapped to a different anon_vma or the anon_vma
442 * returned may already be freed (and even reused).
444 * In case it was remapped to a different anon_vma, the new anon_vma will be a
445 * child of the old anon_vma, and the anon_vma lifetime rules will therefore
446 * ensure that any anon_vma obtained from the page will still be valid for as
447 * long as we observe page_mapped() [ hence all those page_mapped() tests ].
449 * All users of this function must be very careful when walking the anon_vma
450 * chain and verify that the page in question is indeed mapped in it
451 * [ something equivalent to page_mapped_in_vma() ].
453 * Since anon_vma's slab is DESTROY_BY_RCU and we know from page_remove_rmap()
454 * that the anon_vma pointer from page->mapping is valid if there is a
455 * mapcount, we can dereference the anon_vma after observing those.
457 struct anon_vma
*page_get_anon_vma(struct page
*page
)
459 struct anon_vma
*anon_vma
= NULL
;
460 unsigned long anon_mapping
;
463 anon_mapping
= (unsigned long)READ_ONCE(page
->mapping
);
464 if ((anon_mapping
& PAGE_MAPPING_FLAGS
) != PAGE_MAPPING_ANON
)
466 if (!page_mapped(page
))
469 anon_vma
= (struct anon_vma
*) (anon_mapping
- PAGE_MAPPING_ANON
);
470 if (!atomic_inc_not_zero(&anon_vma
->refcount
)) {
476 * If this page is still mapped, then its anon_vma cannot have been
477 * freed. But if it has been unmapped, we have no security against the
478 * anon_vma structure being freed and reused (for another anon_vma:
479 * SLAB_DESTROY_BY_RCU guarantees that - so the atomic_inc_not_zero()
480 * above cannot corrupt).
482 if (!page_mapped(page
)) {
484 put_anon_vma(anon_vma
);
494 * Similar to page_get_anon_vma() except it locks the anon_vma.
496 * Its a little more complex as it tries to keep the fast path to a single
497 * atomic op -- the trylock. If we fail the trylock, we fall back to getting a
498 * reference like with page_get_anon_vma() and then block on the mutex.
500 struct anon_vma
*page_lock_anon_vma_read(struct page
*page
)
502 struct anon_vma
*anon_vma
= NULL
;
503 struct anon_vma
*root_anon_vma
;
504 unsigned long anon_mapping
;
507 anon_mapping
= (unsigned long)READ_ONCE(page
->mapping
);
508 if ((anon_mapping
& PAGE_MAPPING_FLAGS
) != PAGE_MAPPING_ANON
)
510 if (!page_mapped(page
))
513 anon_vma
= (struct anon_vma
*) (anon_mapping
- PAGE_MAPPING_ANON
);
514 root_anon_vma
= READ_ONCE(anon_vma
->root
);
515 if (down_read_trylock(&root_anon_vma
->rwsem
)) {
517 * If the page is still mapped, then this anon_vma is still
518 * its anon_vma, and holding the mutex ensures that it will
519 * not go away, see anon_vma_free().
521 if (!page_mapped(page
)) {
522 up_read(&root_anon_vma
->rwsem
);
528 /* trylock failed, we got to sleep */
529 if (!atomic_inc_not_zero(&anon_vma
->refcount
)) {
534 if (!page_mapped(page
)) {
536 put_anon_vma(anon_vma
);
540 /* we pinned the anon_vma, its safe to sleep */
542 anon_vma_lock_read(anon_vma
);
544 if (atomic_dec_and_test(&anon_vma
->refcount
)) {
546 * Oops, we held the last refcount, release the lock
547 * and bail -- can't simply use put_anon_vma() because
548 * we'll deadlock on the anon_vma_lock_write() recursion.
550 anon_vma_unlock_read(anon_vma
);
551 __put_anon_vma(anon_vma
);
562 void page_unlock_anon_vma_read(struct anon_vma
*anon_vma
)
564 anon_vma_unlock_read(anon_vma
);
568 * At what user virtual address is page expected in @vma?
570 static inline unsigned long
571 __vma_address(struct page
*page
, struct vm_area_struct
*vma
)
573 pgoff_t pgoff
= page_to_pgoff(page
);
574 return vma
->vm_start
+ ((pgoff
- vma
->vm_pgoff
) << PAGE_SHIFT
);
578 vma_address(struct page
*page
, struct vm_area_struct
*vma
)
580 unsigned long address
= __vma_address(page
, vma
);
582 /* page should be within @vma mapping range */
583 VM_BUG_ON_VMA(address
< vma
->vm_start
|| address
>= vma
->vm_end
, vma
);
588 #ifdef CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH
589 static void percpu_flush_tlb_batch_pages(void *data
)
592 * All TLB entries are flushed on the assumption that it is
593 * cheaper to flush all TLBs and let them be refilled than
594 * flushing individual PFNs. Note that we do not track mm's
595 * to flush as that might simply be multiple full TLB flushes
598 count_vm_tlb_event(NR_TLB_REMOTE_FLUSH_RECEIVED
);
603 * Flush TLB entries for recently unmapped pages from remote CPUs. It is
604 * important if a PTE was dirty when it was unmapped that it's flushed
605 * before any IO is initiated on the page to prevent lost writes. Similarly,
606 * it must be flushed before freeing to prevent data leakage.
608 void try_to_unmap_flush(void)
610 struct tlbflush_unmap_batch
*tlb_ubc
= ¤t
->tlb_ubc
;
613 if (!tlb_ubc
->flush_required
)
618 trace_tlb_flush(TLB_REMOTE_SHOOTDOWN
, -1UL);
620 if (cpumask_test_cpu(cpu
, &tlb_ubc
->cpumask
))
621 percpu_flush_tlb_batch_pages(&tlb_ubc
->cpumask
);
623 if (cpumask_any_but(&tlb_ubc
->cpumask
, cpu
) < nr_cpu_ids
) {
624 smp_call_function_many(&tlb_ubc
->cpumask
,
625 percpu_flush_tlb_batch_pages
, (void *)tlb_ubc
, true);
627 cpumask_clear(&tlb_ubc
->cpumask
);
628 tlb_ubc
->flush_required
= false;
629 tlb_ubc
->writable
= false;
633 /* Flush iff there are potentially writable TLB entries that can race with IO */
634 void try_to_unmap_flush_dirty(void)
636 struct tlbflush_unmap_batch
*tlb_ubc
= ¤t
->tlb_ubc
;
638 if (tlb_ubc
->writable
)
639 try_to_unmap_flush();
642 static void set_tlb_ubc_flush_pending(struct mm_struct
*mm
,
643 struct page
*page
, bool writable
)
645 struct tlbflush_unmap_batch
*tlb_ubc
= ¤t
->tlb_ubc
;
647 cpumask_or(&tlb_ubc
->cpumask
, &tlb_ubc
->cpumask
, mm_cpumask(mm
));
648 tlb_ubc
->flush_required
= true;
651 * If the PTE was dirty then it's best to assume it's writable. The
652 * caller must use try_to_unmap_flush_dirty() or try_to_unmap_flush()
653 * before the page is queued for IO.
656 tlb_ubc
->writable
= true;
660 * Returns true if the TLB flush should be deferred to the end of a batch of
661 * unmap operations to reduce IPIs.
663 static bool should_defer_flush(struct mm_struct
*mm
, enum ttu_flags flags
)
665 bool should_defer
= false;
667 if (!(flags
& TTU_BATCH_FLUSH
))
670 /* If remote CPUs need to be flushed then defer batch the flush */
671 if (cpumask_any_but(mm_cpumask(mm
), get_cpu()) < nr_cpu_ids
)
678 static void set_tlb_ubc_flush_pending(struct mm_struct
*mm
,
679 struct page
*page
, bool writable
)
683 static bool should_defer_flush(struct mm_struct
*mm
, enum ttu_flags flags
)
687 #endif /* CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH */
690 * At what user virtual address is page expected in vma?
691 * Caller should check the page is actually part of the vma.
693 unsigned long page_address_in_vma(struct page
*page
, struct vm_area_struct
*vma
)
695 unsigned long address
;
696 if (PageAnon(page
)) {
697 struct anon_vma
*page__anon_vma
= page_anon_vma(page
);
699 * Note: swapoff's unuse_vma() is more efficient with this
700 * check, and needs it to match anon_vma when KSM is active.
702 if (!vma
->anon_vma
|| !page__anon_vma
||
703 vma
->anon_vma
->root
!= page__anon_vma
->root
)
705 } else if (page
->mapping
) {
706 if (!vma
->vm_file
|| vma
->vm_file
->f_mapping
!= page
->mapping
)
710 address
= __vma_address(page
, vma
);
711 if (unlikely(address
< vma
->vm_start
|| address
>= vma
->vm_end
))
716 pmd_t
*mm_find_pmd(struct mm_struct
*mm
, unsigned long address
)
723 pgd
= pgd_offset(mm
, address
);
724 if (!pgd_present(*pgd
))
727 pud
= pud_offset(pgd
, address
);
728 if (!pud_present(*pud
))
731 pmd
= pmd_offset(pud
, address
);
733 * Some THP functions use the sequence pmdp_huge_clear_flush(), set_pmd_at()
734 * without holding anon_vma lock for write. So when looking for a
735 * genuine pmde (in which to find pte), test present and !THP together.
739 if (!pmd_present(pmde
) || pmd_trans_huge(pmde
))
746 * Check that @page is mapped at @address into @mm.
748 * If @sync is false, page_check_address may perform a racy check to avoid
749 * the page table lock when the pte is not present (helpful when reclaiming
750 * highly shared pages).
752 * On success returns with pte mapped and locked.
754 pte_t
*__page_check_address(struct page
*page
, struct mm_struct
*mm
,
755 unsigned long address
, spinlock_t
**ptlp
, int sync
)
761 if (unlikely(PageHuge(page
))) {
762 /* when pud is not present, pte will be NULL */
763 pte
= huge_pte_offset(mm
, address
);
767 ptl
= huge_pte_lockptr(page_hstate(page
), mm
, pte
);
771 pmd
= mm_find_pmd(mm
, address
);
775 pte
= pte_offset_map(pmd
, address
);
776 /* Make a quick check before getting the lock */
777 if (!sync
&& !pte_present(*pte
)) {
782 ptl
= pte_lockptr(mm
, pmd
);
785 if (pte_present(*pte
) && page_to_pfn(page
) == pte_pfn(*pte
)) {
789 pte_unmap_unlock(pte
, ptl
);
794 * page_mapped_in_vma - check whether a page is really mapped in a VMA
795 * @page: the page to test
796 * @vma: the VMA to test
798 * Returns 1 if the page is mapped into the page tables of the VMA, 0
799 * if the page is not mapped into the page tables of this VMA. Only
800 * valid for normal file or anonymous VMAs.
802 int page_mapped_in_vma(struct page
*page
, struct vm_area_struct
*vma
)
804 unsigned long address
;
808 address
= __vma_address(page
, vma
);
809 if (unlikely(address
< vma
->vm_start
|| address
>= vma
->vm_end
))
811 pte
= page_check_address(page
, vma
->vm_mm
, address
, &ptl
, 1);
812 if (!pte
) /* the page is not in this mm */
814 pte_unmap_unlock(pte
, ptl
);
819 struct page_referenced_arg
{
822 unsigned long vm_flags
;
823 struct mem_cgroup
*memcg
;
826 * arg: page_referenced_arg will be passed
828 static int page_referenced_one(struct page
*page
, struct vm_area_struct
*vma
,
829 unsigned long address
, void *arg
)
831 struct mm_struct
*mm
= vma
->vm_mm
;
834 struct page_referenced_arg
*pra
= arg
;
836 if (unlikely(PageTransHuge(page
))) {
840 * rmap might return false positives; we must filter
841 * these out using page_check_address_pmd().
843 pmd
= page_check_address_pmd(page
, mm
, address
,
844 PAGE_CHECK_ADDRESS_PMD_FLAG
, &ptl
);
848 if (vma
->vm_flags
& VM_LOCKED
) {
850 pra
->vm_flags
|= VM_LOCKED
;
851 return SWAP_FAIL
; /* To break the loop */
854 /* go ahead even if the pmd is pmd_trans_splitting() */
855 if (pmdp_clear_flush_young_notify(vma
, address
, pmd
))
862 * rmap might return false positives; we must filter
863 * these out using page_check_address().
865 pte
= page_check_address(page
, mm
, address
, &ptl
, 0);
869 if (vma
->vm_flags
& VM_LOCKED
) {
870 pte_unmap_unlock(pte
, ptl
);
871 pra
->vm_flags
|= VM_LOCKED
;
872 return SWAP_FAIL
; /* To break the loop */
875 if (ptep_clear_flush_young_notify(vma
, address
, pte
)) {
877 * Don't treat a reference through a sequentially read
878 * mapping as such. If the page has been used in
879 * another mapping, we will catch it; if this other
880 * mapping is already gone, the unmap path will have
881 * set PG_referenced or activated the page.
883 if (likely(!(vma
->vm_flags
& VM_SEQ_READ
)))
886 pte_unmap_unlock(pte
, ptl
);
891 pra
->vm_flags
|= vma
->vm_flags
;
896 return SWAP_SUCCESS
; /* To break the loop */
901 static bool invalid_page_referenced_vma(struct vm_area_struct
*vma
, void *arg
)
903 struct page_referenced_arg
*pra
= arg
;
904 struct mem_cgroup
*memcg
= pra
->memcg
;
906 if (!mm_match_cgroup(vma
->vm_mm
, memcg
))
913 * page_referenced - test if the page was referenced
914 * @page: the page to test
915 * @is_locked: caller holds lock on the page
916 * @memcg: target memory cgroup
917 * @vm_flags: collect encountered vma->vm_flags who actually referenced the page
919 * Quick test_and_clear_referenced for all mappings to a page,
920 * returns the number of ptes which referenced the page.
922 int page_referenced(struct page
*page
,
924 struct mem_cgroup
*memcg
,
925 unsigned long *vm_flags
)
929 struct page_referenced_arg pra
= {
930 .mapcount
= page_mapcount(page
),
933 struct rmap_walk_control rwc
= {
934 .rmap_one
= page_referenced_one
,
936 .anon_lock
= page_lock_anon_vma_read
,
940 if (!page_mapped(page
))
943 if (!page_rmapping(page
))
946 if (!is_locked
&& (!PageAnon(page
) || PageKsm(page
))) {
947 we_locked
= trylock_page(page
);
953 * If we are reclaiming on behalf of a cgroup, skip
954 * counting on behalf of references from different
958 rwc
.invalid_vma
= invalid_page_referenced_vma
;
961 ret
= rmap_walk(page
, &rwc
);
962 *vm_flags
= pra
.vm_flags
;
967 return pra
.referenced
;
970 static int page_mkclean_one(struct page
*page
, struct vm_area_struct
*vma
,
971 unsigned long address
, void *arg
)
973 struct mm_struct
*mm
= vma
->vm_mm
;
979 pte
= page_check_address(page
, mm
, address
, &ptl
, 1);
983 if (pte_dirty(*pte
) || pte_write(*pte
)) {
986 flush_cache_page(vma
, address
, pte_pfn(*pte
));
987 entry
= ptep_clear_flush(vma
, address
, pte
);
988 entry
= pte_wrprotect(entry
);
989 entry
= pte_mkclean(entry
);
990 set_pte_at(mm
, address
, pte
, entry
);
994 pte_unmap_unlock(pte
, ptl
);
997 mmu_notifier_invalidate_page(mm
, address
);
1004 static bool invalid_mkclean_vma(struct vm_area_struct
*vma
, void *arg
)
1006 if (vma
->vm_flags
& VM_SHARED
)
1012 int page_mkclean(struct page
*page
)
1015 struct address_space
*mapping
;
1016 struct rmap_walk_control rwc
= {
1017 .arg
= (void *)&cleaned
,
1018 .rmap_one
= page_mkclean_one
,
1019 .invalid_vma
= invalid_mkclean_vma
,
1022 BUG_ON(!PageLocked(page
));
1024 if (!page_mapped(page
))
1027 mapping
= page_mapping(page
);
1031 rmap_walk(page
, &rwc
);
1035 EXPORT_SYMBOL_GPL(page_mkclean
);
1038 * page_move_anon_rmap - move a page to our anon_vma
1039 * @page: the page to move to our anon_vma
1040 * @vma: the vma the page belongs to
1041 * @address: the user virtual address mapped
1043 * When a page belongs exclusively to one process after a COW event,
1044 * that page can be moved into the anon_vma that belongs to just that
1045 * process, so the rmap code will not search the parent or sibling
1048 void page_move_anon_rmap(struct page
*page
,
1049 struct vm_area_struct
*vma
, unsigned long address
)
1051 struct anon_vma
*anon_vma
= vma
->anon_vma
;
1053 VM_BUG_ON_PAGE(!PageLocked(page
), page
);
1054 VM_BUG_ON_VMA(!anon_vma
, vma
);
1055 VM_BUG_ON_PAGE(page
->index
!= linear_page_index(vma
, address
), page
);
1057 anon_vma
= (void *) anon_vma
+ PAGE_MAPPING_ANON
;
1059 * Ensure that anon_vma and the PAGE_MAPPING_ANON bit are written
1060 * simultaneously, so a concurrent reader (eg page_referenced()'s
1061 * PageAnon()) will not see one without the other.
1063 WRITE_ONCE(page
->mapping
, (struct address_space
*) anon_vma
);
1067 * __page_set_anon_rmap - set up new anonymous rmap
1068 * @page: Page to add to rmap
1069 * @vma: VM area to add page to.
1070 * @address: User virtual address of the mapping
1071 * @exclusive: the page is exclusively owned by the current process
1073 static void __page_set_anon_rmap(struct page
*page
,
1074 struct vm_area_struct
*vma
, unsigned long address
, int exclusive
)
1076 struct anon_vma
*anon_vma
= vma
->anon_vma
;
1084 * If the page isn't exclusively mapped into this vma,
1085 * we must use the _oldest_ possible anon_vma for the
1089 anon_vma
= anon_vma
->root
;
1091 anon_vma
= (void *) anon_vma
+ PAGE_MAPPING_ANON
;
1092 page
->mapping
= (struct address_space
*) anon_vma
;
1093 page
->index
= linear_page_index(vma
, address
);
1097 * __page_check_anon_rmap - sanity check anonymous rmap addition
1098 * @page: the page to add the mapping to
1099 * @vma: the vm area in which the mapping is added
1100 * @address: the user virtual address mapped
1102 static void __page_check_anon_rmap(struct page
*page
,
1103 struct vm_area_struct
*vma
, unsigned long address
)
1105 #ifdef CONFIG_DEBUG_VM
1107 * The page's anon-rmap details (mapping and index) are guaranteed to
1108 * be set up correctly at this point.
1110 * We have exclusion against page_add_anon_rmap because the caller
1111 * always holds the page locked, except if called from page_dup_rmap,
1112 * in which case the page is already known to be setup.
1114 * We have exclusion against page_add_new_anon_rmap because those pages
1115 * are initially only visible via the pagetables, and the pte is locked
1116 * over the call to page_add_new_anon_rmap.
1118 BUG_ON(page_anon_vma(page
)->root
!= vma
->anon_vma
->root
);
1119 BUG_ON(page
->index
!= linear_page_index(vma
, address
));
1124 * page_add_anon_rmap - add pte mapping to an anonymous page
1125 * @page: the page to add the mapping to
1126 * @vma: the vm area in which the mapping is added
1127 * @address: the user virtual address mapped
1129 * The caller needs to hold the pte lock, and the page must be locked in
1130 * the anon_vma case: to serialize mapping,index checking after setting,
1131 * and to ensure that PageAnon is not being upgraded racily to PageKsm
1132 * (but PageKsm is never downgraded to PageAnon).
1134 void page_add_anon_rmap(struct page
*page
,
1135 struct vm_area_struct
*vma
, unsigned long address
)
1137 do_page_add_anon_rmap(page
, vma
, address
, 0);
1141 * Special version of the above for do_swap_page, which often runs
1142 * into pages that are exclusively owned by the current process.
1143 * Everybody else should continue to use page_add_anon_rmap above.
1145 void do_page_add_anon_rmap(struct page
*page
,
1146 struct vm_area_struct
*vma
, unsigned long address
, int exclusive
)
1148 int first
= atomic_inc_and_test(&page
->_mapcount
);
1151 * We use the irq-unsafe __{inc|mod}_zone_page_stat because
1152 * these counters are not modified in interrupt context, and
1153 * pte lock(a spinlock) is held, which implies preemption
1156 if (PageTransHuge(page
))
1157 __inc_zone_page_state(page
,
1158 NR_ANON_TRANSPARENT_HUGEPAGES
);
1159 __mod_zone_page_state(page_zone(page
), NR_ANON_PAGES
,
1160 hpage_nr_pages(page
));
1162 if (unlikely(PageKsm(page
)))
1165 VM_BUG_ON_PAGE(!PageLocked(page
), page
);
1166 /* address might be in next vma when migration races vma_adjust */
1168 __page_set_anon_rmap(page
, vma
, address
, exclusive
);
1170 __page_check_anon_rmap(page
, vma
, address
);
1174 * page_add_new_anon_rmap - add pte mapping to a new anonymous page
1175 * @page: the page to add the mapping to
1176 * @vma: the vm area in which the mapping is added
1177 * @address: the user virtual address mapped
1179 * Same as page_add_anon_rmap but must only be called on *new* pages.
1180 * This means the inc-and-test can be bypassed.
1181 * Page does not have to be locked.
1183 void page_add_new_anon_rmap(struct page
*page
,
1184 struct vm_area_struct
*vma
, unsigned long address
)
1186 VM_BUG_ON_VMA(address
< vma
->vm_start
|| address
>= vma
->vm_end
, vma
);
1187 SetPageSwapBacked(page
);
1188 atomic_set(&page
->_mapcount
, 0); /* increment count (starts at -1) */
1189 if (PageTransHuge(page
))
1190 __inc_zone_page_state(page
, NR_ANON_TRANSPARENT_HUGEPAGES
);
1191 __mod_zone_page_state(page_zone(page
), NR_ANON_PAGES
,
1192 hpage_nr_pages(page
));
1193 __page_set_anon_rmap(page
, vma
, address
, 1);
1197 * page_add_file_rmap - add pte mapping to a file page
1198 * @page: the page to add the mapping to
1200 * The caller needs to hold the pte lock.
1202 void page_add_file_rmap(struct page
*page
)
1204 struct mem_cgroup
*memcg
;
1206 memcg
= mem_cgroup_begin_page_stat(page
);
1207 if (atomic_inc_and_test(&page
->_mapcount
)) {
1208 __inc_zone_page_state(page
, NR_FILE_MAPPED
);
1209 mem_cgroup_inc_page_stat(memcg
, MEM_CGROUP_STAT_FILE_MAPPED
);
1211 mem_cgroup_end_page_stat(memcg
);
1214 static void page_remove_file_rmap(struct page
*page
)
1216 struct mem_cgroup
*memcg
;
1218 memcg
= mem_cgroup_begin_page_stat(page
);
1220 /* page still mapped by someone else? */
1221 if (!atomic_add_negative(-1, &page
->_mapcount
))
1224 /* Hugepages are not counted in NR_FILE_MAPPED for now. */
1225 if (unlikely(PageHuge(page
)))
1229 * We use the irq-unsafe __{inc|mod}_zone_page_stat because
1230 * these counters are not modified in interrupt context, and
1231 * pte lock(a spinlock) is held, which implies preemption disabled.
1233 __dec_zone_page_state(page
, NR_FILE_MAPPED
);
1234 mem_cgroup_dec_page_stat(memcg
, MEM_CGROUP_STAT_FILE_MAPPED
);
1236 if (unlikely(PageMlocked(page
)))
1237 clear_page_mlock(page
);
1239 mem_cgroup_end_page_stat(memcg
);
1243 * page_remove_rmap - take down pte mapping from a page
1244 * @page: page to remove mapping from
1246 * The caller needs to hold the pte lock.
1248 void page_remove_rmap(struct page
*page
)
1250 if (!PageAnon(page
)) {
1251 page_remove_file_rmap(page
);
1255 /* page still mapped by someone else? */
1256 if (!atomic_add_negative(-1, &page
->_mapcount
))
1259 /* Hugepages are not counted in NR_ANON_PAGES for now. */
1260 if (unlikely(PageHuge(page
)))
1264 * We use the irq-unsafe __{inc|mod}_zone_page_stat because
1265 * these counters are not modified in interrupt context, and
1266 * pte lock(a spinlock) is held, which implies preemption disabled.
1268 if (PageTransHuge(page
))
1269 __dec_zone_page_state(page
, NR_ANON_TRANSPARENT_HUGEPAGES
);
1271 __mod_zone_page_state(page_zone(page
), NR_ANON_PAGES
,
1272 -hpage_nr_pages(page
));
1274 if (unlikely(PageMlocked(page
)))
1275 clear_page_mlock(page
);
1278 * It would be tidy to reset the PageAnon mapping here,
1279 * but that might overwrite a racing page_add_anon_rmap
1280 * which increments mapcount after us but sets mapping
1281 * before us: so leave the reset to free_hot_cold_page,
1282 * and remember that it's only reliable while mapped.
1283 * Leaving it set also helps swapoff to reinstate ptes
1284 * faster for those pages still in swapcache.
1289 * @arg: enum ttu_flags will be passed to this argument
1291 static int try_to_unmap_one(struct page
*page
, struct vm_area_struct
*vma
,
1292 unsigned long address
, void *arg
)
1294 struct mm_struct
*mm
= vma
->vm_mm
;
1298 int ret
= SWAP_AGAIN
;
1299 enum ttu_flags flags
= (enum ttu_flags
)arg
;
1301 pte
= page_check_address(page
, mm
, address
, &ptl
, 0);
1306 * If the page is mlock()d, we cannot swap it out.
1307 * If it's recently referenced (perhaps page_referenced
1308 * skipped over this mm) then we should reactivate it.
1310 if (!(flags
& TTU_IGNORE_MLOCK
)) {
1311 if (vma
->vm_flags
& VM_LOCKED
)
1314 if (flags
& TTU_MUNLOCK
)
1317 if (!(flags
& TTU_IGNORE_ACCESS
)) {
1318 if (ptep_clear_flush_young_notify(vma
, address
, pte
)) {
1324 /* Nuke the page table entry. */
1325 flush_cache_page(vma
, address
, page_to_pfn(page
));
1326 if (should_defer_flush(mm
, flags
)) {
1328 * We clear the PTE but do not flush so potentially a remote
1329 * CPU could still be writing to the page. If the entry was
1330 * previously clean then the architecture must guarantee that
1331 * a clear->dirty transition on a cached TLB entry is written
1332 * through and traps if the PTE is unmapped.
1334 pteval
= ptep_get_and_clear(mm
, address
, pte
);
1336 set_tlb_ubc_flush_pending(mm
, page
, pte_dirty(pteval
));
1338 pteval
= ptep_clear_flush(vma
, address
, pte
);
1341 /* Move the dirty bit to the physical page now the pte is gone. */
1342 if (pte_dirty(pteval
))
1343 set_page_dirty(page
);
1345 /* Update high watermark before we lower rss */
1346 update_hiwater_rss(mm
);
1348 if (PageHWPoison(page
) && !(flags
& TTU_IGNORE_HWPOISON
)) {
1349 if (!PageHuge(page
)) {
1351 dec_mm_counter(mm
, MM_ANONPAGES
);
1353 dec_mm_counter(mm
, MM_FILEPAGES
);
1355 set_pte_at(mm
, address
, pte
,
1356 swp_entry_to_pte(make_hwpoison_entry(page
)));
1357 } else if (pte_unused(pteval
)) {
1359 * The guest indicated that the page content is of no
1360 * interest anymore. Simply discard the pte, vmscan
1361 * will take care of the rest.
1364 dec_mm_counter(mm
, MM_ANONPAGES
);
1366 dec_mm_counter(mm
, MM_FILEPAGES
);
1367 } else if (PageAnon(page
)) {
1368 swp_entry_t entry
= { .val
= page_private(page
) };
1371 if (PageSwapCache(page
)) {
1373 * Store the swap location in the pte.
1374 * See handle_pte_fault() ...
1376 if (swap_duplicate(entry
) < 0) {
1377 set_pte_at(mm
, address
, pte
, pteval
);
1381 if (list_empty(&mm
->mmlist
)) {
1382 spin_lock(&mmlist_lock
);
1383 if (list_empty(&mm
->mmlist
))
1384 list_add(&mm
->mmlist
, &init_mm
.mmlist
);
1385 spin_unlock(&mmlist_lock
);
1387 dec_mm_counter(mm
, MM_ANONPAGES
);
1388 inc_mm_counter(mm
, MM_SWAPENTS
);
1389 } else if (IS_ENABLED(CONFIG_MIGRATION
)) {
1391 * Store the pfn of the page in a special migration
1392 * pte. do_swap_page() will wait until the migration
1393 * pte is removed and then restart fault handling.
1395 BUG_ON(!(flags
& TTU_MIGRATION
));
1396 entry
= make_migration_entry(page
, pte_write(pteval
));
1398 swp_pte
= swp_entry_to_pte(entry
);
1399 if (pte_soft_dirty(pteval
))
1400 swp_pte
= pte_swp_mksoft_dirty(swp_pte
);
1401 set_pte_at(mm
, address
, pte
, swp_pte
);
1402 } else if (IS_ENABLED(CONFIG_MIGRATION
) &&
1403 (flags
& TTU_MIGRATION
)) {
1404 /* Establish migration entry for a file page */
1406 entry
= make_migration_entry(page
, pte_write(pteval
));
1407 set_pte_at(mm
, address
, pte
, swp_entry_to_pte(entry
));
1409 dec_mm_counter(mm
, MM_FILEPAGES
);
1411 page_remove_rmap(page
);
1412 page_cache_release(page
);
1415 pte_unmap_unlock(pte
, ptl
);
1416 if (ret
!= SWAP_FAIL
&& !(flags
& TTU_MUNLOCK
))
1417 mmu_notifier_invalidate_page(mm
, address
);
1422 pte_unmap_unlock(pte
, ptl
);
1426 * We need mmap_sem locking, Otherwise VM_LOCKED check makes
1427 * unstable result and race. Plus, We can't wait here because
1428 * we now hold anon_vma->rwsem or mapping->i_mmap_rwsem.
1429 * if trylock failed, the page remain in evictable lru and later
1430 * vmscan could retry to move the page to unevictable lru if the
1431 * page is actually mlocked.
1433 if (down_read_trylock(&vma
->vm_mm
->mmap_sem
)) {
1434 if (vma
->vm_flags
& VM_LOCKED
) {
1435 mlock_vma_page(page
);
1438 up_read(&vma
->vm_mm
->mmap_sem
);
1443 bool is_vma_temporary_stack(struct vm_area_struct
*vma
)
1445 int maybe_stack
= vma
->vm_flags
& (VM_GROWSDOWN
| VM_GROWSUP
);
1450 if ((vma
->vm_flags
& VM_STACK_INCOMPLETE_SETUP
) ==
1451 VM_STACK_INCOMPLETE_SETUP
)
1457 static bool invalid_migration_vma(struct vm_area_struct
*vma
, void *arg
)
1459 return is_vma_temporary_stack(vma
);
1462 static int page_not_mapped(struct page
*page
)
1464 return !page_mapped(page
);
1468 * try_to_unmap - try to remove all page table mappings to a page
1469 * @page: the page to get unmapped
1470 * @flags: action and flags
1472 * Tries to remove all the page table entries which are mapping this
1473 * page, used in the pageout path. Caller must hold the page lock.
1474 * Return values are:
1476 * SWAP_SUCCESS - we succeeded in removing all mappings
1477 * SWAP_AGAIN - we missed a mapping, try again later
1478 * SWAP_FAIL - the page is unswappable
1479 * SWAP_MLOCK - page is mlocked.
1481 int try_to_unmap(struct page
*page
, enum ttu_flags flags
)
1484 struct rmap_walk_control rwc
= {
1485 .rmap_one
= try_to_unmap_one
,
1486 .arg
= (void *)flags
,
1487 .done
= page_not_mapped
,
1488 .anon_lock
= page_lock_anon_vma_read
,
1491 VM_BUG_ON_PAGE(!PageHuge(page
) && PageTransHuge(page
), page
);
1494 * During exec, a temporary VMA is setup and later moved.
1495 * The VMA is moved under the anon_vma lock but not the
1496 * page tables leading to a race where migration cannot
1497 * find the migration ptes. Rather than increasing the
1498 * locking requirements of exec(), migration skips
1499 * temporary VMAs until after exec() completes.
1501 if ((flags
& TTU_MIGRATION
) && !PageKsm(page
) && PageAnon(page
))
1502 rwc
.invalid_vma
= invalid_migration_vma
;
1504 ret
= rmap_walk(page
, &rwc
);
1506 if (ret
!= SWAP_MLOCK
&& !page_mapped(page
))
1512 * try_to_munlock - try to munlock a page
1513 * @page: the page to be munlocked
1515 * Called from munlock code. Checks all of the VMAs mapping the page
1516 * to make sure nobody else has this page mlocked. The page will be
1517 * returned with PG_mlocked cleared if no other vmas have it mlocked.
1519 * Return values are:
1521 * SWAP_AGAIN - no vma is holding page mlocked, or,
1522 * SWAP_AGAIN - page mapped in mlocked vma -- couldn't acquire mmap sem
1523 * SWAP_FAIL - page cannot be located at present
1524 * SWAP_MLOCK - page is now mlocked.
1526 int try_to_munlock(struct page
*page
)
1529 struct rmap_walk_control rwc
= {
1530 .rmap_one
= try_to_unmap_one
,
1531 .arg
= (void *)TTU_MUNLOCK
,
1532 .done
= page_not_mapped
,
1533 .anon_lock
= page_lock_anon_vma_read
,
1537 VM_BUG_ON_PAGE(!PageLocked(page
) || PageLRU(page
), page
);
1539 ret
= rmap_walk(page
, &rwc
);
1543 void __put_anon_vma(struct anon_vma
*anon_vma
)
1545 struct anon_vma
*root
= anon_vma
->root
;
1547 anon_vma_free(anon_vma
);
1548 if (root
!= anon_vma
&& atomic_dec_and_test(&root
->refcount
))
1549 anon_vma_free(root
);
1552 static struct anon_vma
*rmap_walk_anon_lock(struct page
*page
,
1553 struct rmap_walk_control
*rwc
)
1555 struct anon_vma
*anon_vma
;
1558 return rwc
->anon_lock(page
);
1561 * Note: remove_migration_ptes() cannot use page_lock_anon_vma_read()
1562 * because that depends on page_mapped(); but not all its usages
1563 * are holding mmap_sem. Users without mmap_sem are required to
1564 * take a reference count to prevent the anon_vma disappearing
1566 anon_vma
= page_anon_vma(page
);
1570 anon_vma_lock_read(anon_vma
);
1575 * rmap_walk_anon - do something to anonymous page using the object-based
1577 * @page: the page to be handled
1578 * @rwc: control variable according to each walk type
1580 * Find all the mappings of a page using the mapping pointer and the vma chains
1581 * contained in the anon_vma struct it points to.
1583 * When called from try_to_munlock(), the mmap_sem of the mm containing the vma
1584 * where the page was found will be held for write. So, we won't recheck
1585 * vm_flags for that VMA. That should be OK, because that vma shouldn't be
1588 static int rmap_walk_anon(struct page
*page
, struct rmap_walk_control
*rwc
)
1590 struct anon_vma
*anon_vma
;
1592 struct anon_vma_chain
*avc
;
1593 int ret
= SWAP_AGAIN
;
1595 anon_vma
= rmap_walk_anon_lock(page
, rwc
);
1599 pgoff
= page_to_pgoff(page
);
1600 anon_vma_interval_tree_foreach(avc
, &anon_vma
->rb_root
, pgoff
, pgoff
) {
1601 struct vm_area_struct
*vma
= avc
->vma
;
1602 unsigned long address
= vma_address(page
, vma
);
1604 if (rwc
->invalid_vma
&& rwc
->invalid_vma(vma
, rwc
->arg
))
1607 ret
= rwc
->rmap_one(page
, vma
, address
, rwc
->arg
);
1608 if (ret
!= SWAP_AGAIN
)
1610 if (rwc
->done
&& rwc
->done(page
))
1613 anon_vma_unlock_read(anon_vma
);
1618 * rmap_walk_file - do something to file page using the object-based rmap method
1619 * @page: the page to be handled
1620 * @rwc: control variable according to each walk type
1622 * Find all the mappings of a page using the mapping pointer and the vma chains
1623 * contained in the address_space struct it points to.
1625 * When called from try_to_munlock(), the mmap_sem of the mm containing the vma
1626 * where the page was found will be held for write. So, we won't recheck
1627 * vm_flags for that VMA. That should be OK, because that vma shouldn't be
1630 static int rmap_walk_file(struct page
*page
, struct rmap_walk_control
*rwc
)
1632 struct address_space
*mapping
= page
->mapping
;
1634 struct vm_area_struct
*vma
;
1635 int ret
= SWAP_AGAIN
;
1638 * The page lock not only makes sure that page->mapping cannot
1639 * suddenly be NULLified by truncation, it makes sure that the
1640 * structure at mapping cannot be freed and reused yet,
1641 * so we can safely take mapping->i_mmap_rwsem.
1643 VM_BUG_ON_PAGE(!PageLocked(page
), page
);
1648 pgoff
= page_to_pgoff(page
);
1649 i_mmap_lock_read(mapping
);
1650 vma_interval_tree_foreach(vma
, &mapping
->i_mmap
, pgoff
, pgoff
) {
1651 unsigned long address
= vma_address(page
, vma
);
1653 if (rwc
->invalid_vma
&& rwc
->invalid_vma(vma
, rwc
->arg
))
1656 ret
= rwc
->rmap_one(page
, vma
, address
, rwc
->arg
);
1657 if (ret
!= SWAP_AGAIN
)
1659 if (rwc
->done
&& rwc
->done(page
))
1664 i_mmap_unlock_read(mapping
);
1668 int rmap_walk(struct page
*page
, struct rmap_walk_control
*rwc
)
1670 if (unlikely(PageKsm(page
)))
1671 return rmap_walk_ksm(page
, rwc
);
1672 else if (PageAnon(page
))
1673 return rmap_walk_anon(page
, rwc
);
1675 return rmap_walk_file(page
, rwc
);
1678 #ifdef CONFIG_HUGETLB_PAGE
1680 * The following three functions are for anonymous (private mapped) hugepages.
1681 * Unlike common anonymous pages, anonymous hugepages have no accounting code
1682 * and no lru code, because we handle hugepages differently from common pages.
1684 static void __hugepage_set_anon_rmap(struct page
*page
,
1685 struct vm_area_struct
*vma
, unsigned long address
, int exclusive
)
1687 struct anon_vma
*anon_vma
= vma
->anon_vma
;
1694 anon_vma
= anon_vma
->root
;
1696 anon_vma
= (void *) anon_vma
+ PAGE_MAPPING_ANON
;
1697 page
->mapping
= (struct address_space
*) anon_vma
;
1698 page
->index
= linear_page_index(vma
, address
);
1701 void hugepage_add_anon_rmap(struct page
*page
,
1702 struct vm_area_struct
*vma
, unsigned long address
)
1704 struct anon_vma
*anon_vma
= vma
->anon_vma
;
1707 BUG_ON(!PageLocked(page
));
1709 /* address might be in next vma when migration races vma_adjust */
1710 first
= atomic_inc_and_test(&page
->_mapcount
);
1712 __hugepage_set_anon_rmap(page
, vma
, address
, 0);
1715 void hugepage_add_new_anon_rmap(struct page
*page
,
1716 struct vm_area_struct
*vma
, unsigned long address
)
1718 BUG_ON(address
< vma
->vm_start
|| address
>= vma
->vm_end
);
1719 atomic_set(&page
->_mapcount
, 0);
1720 __hugepage_set_anon_rmap(page
, vma
, address
, 1);
1722 #endif /* CONFIG_HUGETLB_PAGE */