2 * Memory Migration functionality - linux/mm/migration.c
4 * Copyright (C) 2006 Silicon Graphics, Inc., Christoph Lameter
6 * Page migration was first developed in the context of the memory hotplug
7 * project. The main authors of the migration code are:
9 * IWAMOTO Toshihiro <iwamoto@valinux.co.jp>
10 * Hirokazu Takahashi <taka@valinux.co.jp>
11 * Dave Hansen <haveblue@us.ibm.com>
15 #include <linux/migrate.h>
16 #include <linux/export.h>
17 #include <linux/swap.h>
18 #include <linux/swapops.h>
19 #include <linux/pagemap.h>
20 #include <linux/buffer_head.h>
21 #include <linux/mm_inline.h>
22 #include <linux/nsproxy.h>
23 #include <linux/pagevec.h>
24 #include <linux/ksm.h>
25 #include <linux/rmap.h>
26 #include <linux/topology.h>
27 #include <linux/cpu.h>
28 #include <linux/cpuset.h>
29 #include <linux/writeback.h>
30 #include <linux/mempolicy.h>
31 #include <linux/vmalloc.h>
32 #include <linux/security.h>
33 #include <linux/memcontrol.h>
34 #include <linux/syscalls.h>
35 #include <linux/hugetlb.h>
36 #include <linux/gfp.h>
38 #include <asm/tlbflush.h>
43 * migrate_prep() needs to be called before we start compiling a list of pages
44 * to be migrated using isolate_lru_page(). If scheduling work on other CPUs is
45 * undesirable, use migrate_prep_local()
47 int migrate_prep(void)
50 * Clear the LRU lists so pages can be isolated.
51 * Note that pages may be moved off the LRU after we have
52 * drained them. Those pages will fail to migrate like other
53 * pages that may be busy.
60 /* Do the necessary work of migrate_prep but not if it involves other CPUs */
61 int migrate_prep_local(void)
69 * Add isolated pages on the list back to the LRU under page lock
70 * to avoid leaking evictable pages back onto unevictable list.
72 void putback_lru_pages(struct list_head
*l
)
77 list_for_each_entry_safe(page
, page2
, l
, lru
) {
79 dec_zone_page_state(page
, NR_ISOLATED_ANON
+
80 page_is_file_cache(page
));
81 putback_lru_page(page
);
86 * Restore a potential migration pte to a working pte entry
88 static int remove_migration_pte(struct page
*new, struct vm_area_struct
*vma
,
89 unsigned long addr
, void *old
)
91 struct mm_struct
*mm
= vma
->vm_mm
;
99 if (unlikely(PageHuge(new))) {
100 ptep
= huge_pte_offset(mm
, addr
);
103 ptl
= &mm
->page_table_lock
;
105 pgd
= pgd_offset(mm
, addr
);
106 if (!pgd_present(*pgd
))
109 pud
= pud_offset(pgd
, addr
);
110 if (!pud_present(*pud
))
113 pmd
= pmd_offset(pud
, addr
);
114 if (pmd_trans_huge(*pmd
))
116 if (!pmd_present(*pmd
))
119 ptep
= pte_offset_map(pmd
, addr
);
122 * Peek to check is_swap_pte() before taking ptlock? No, we
123 * can race mremap's move_ptes(), which skips anon_vma lock.
126 ptl
= pte_lockptr(mm
, pmd
);
131 if (!is_swap_pte(pte
))
134 entry
= pte_to_swp_entry(pte
);
136 if (!is_migration_entry(entry
) ||
137 migration_entry_to_page(entry
) != old
)
141 pte
= pte_mkold(mk_pte(new, vma
->vm_page_prot
));
142 if (is_write_migration_entry(entry
))
143 pte
= pte_mkwrite(pte
);
144 #ifdef CONFIG_HUGETLB_PAGE
146 pte
= pte_mkhuge(pte
);
148 flush_cache_page(vma
, addr
, pte_pfn(pte
));
149 set_pte_at(mm
, addr
, ptep
, pte
);
153 hugepage_add_anon_rmap(new, vma
, addr
);
156 } else if (PageAnon(new))
157 page_add_anon_rmap(new, vma
, addr
);
159 page_add_file_rmap(new);
161 /* No need to invalidate - it was non-present before */
162 update_mmu_cache(vma
, addr
, ptep
);
164 pte_unmap_unlock(ptep
, ptl
);
170 * Get rid of all migration entries and replace them by
171 * references to the indicated page.
173 static void remove_migration_ptes(struct page
*old
, struct page
*new)
175 rmap_walk(new, remove_migration_pte
, old
);
179 * Something used the pte of a page under migration. We need to
180 * get to the page and wait until migration is finished.
181 * When we return from this function the fault will be retried.
183 void migration_entry_wait(struct mm_struct
*mm
, pmd_t
*pmd
,
184 unsigned long address
)
191 ptep
= pte_offset_map_lock(mm
, pmd
, address
, &ptl
);
193 if (!is_swap_pte(pte
))
196 entry
= pte_to_swp_entry(pte
);
197 if (!is_migration_entry(entry
))
200 page
= migration_entry_to_page(entry
);
203 * Once radix-tree replacement of page migration started, page_count
204 * *must* be zero. And, we don't want to call wait_on_page_locked()
205 * against a page without get_page().
206 * So, we use get_page_unless_zero(), here. Even failed, page fault
209 if (!get_page_unless_zero(page
))
211 pte_unmap_unlock(ptep
, ptl
);
212 wait_on_page_locked(page
);
216 pte_unmap_unlock(ptep
, ptl
);
220 * Replace the page in the mapping.
222 * The number of remaining references must be:
223 * 1 for anonymous pages without a mapping
224 * 2 for pages with a mapping
225 * 3 for pages with a mapping and PagePrivate/PagePrivate2 set.
227 static int migrate_page_move_mapping(struct address_space
*mapping
,
228 struct page
*newpage
, struct page
*page
)
234 /* Anonymous page without mapping */
235 if (page_count(page
) != 1)
240 spin_lock_irq(&mapping
->tree_lock
);
242 pslot
= radix_tree_lookup_slot(&mapping
->page_tree
,
245 expected_count
= 2 + page_has_private(page
);
246 if (page_count(page
) != expected_count
||
247 radix_tree_deref_slot_protected(pslot
, &mapping
->tree_lock
) != page
) {
248 spin_unlock_irq(&mapping
->tree_lock
);
252 if (!page_freeze_refs(page
, expected_count
)) {
253 spin_unlock_irq(&mapping
->tree_lock
);
258 * Now we know that no one else is looking at the page.
260 get_page(newpage
); /* add cache reference */
261 if (PageSwapCache(page
)) {
262 SetPageSwapCache(newpage
);
263 set_page_private(newpage
, page_private(page
));
266 radix_tree_replace_slot(pslot
, newpage
);
269 * Drop cache reference from old page by unfreezing
270 * to one less reference.
271 * We know this isn't the last reference.
273 page_unfreeze_refs(page
, expected_count
- 1);
276 * If moved to a different zone then also account
277 * the page for that zone. Other VM counters will be
278 * taken care of when we establish references to the
279 * new page and drop references to the old page.
281 * Note that anonymous pages are accounted for
282 * via NR_FILE_PAGES and NR_ANON_PAGES if they
283 * are mapped to swap space.
285 __dec_zone_page_state(page
, NR_FILE_PAGES
);
286 __inc_zone_page_state(newpage
, NR_FILE_PAGES
);
287 if (!PageSwapCache(page
) && PageSwapBacked(page
)) {
288 __dec_zone_page_state(page
, NR_SHMEM
);
289 __inc_zone_page_state(newpage
, NR_SHMEM
);
291 spin_unlock_irq(&mapping
->tree_lock
);
297 * The expected number of remaining references is the same as that
298 * of migrate_page_move_mapping().
300 int migrate_huge_page_move_mapping(struct address_space
*mapping
,
301 struct page
*newpage
, struct page
*page
)
307 if (page_count(page
) != 1)
312 spin_lock_irq(&mapping
->tree_lock
);
314 pslot
= radix_tree_lookup_slot(&mapping
->page_tree
,
317 expected_count
= 2 + page_has_private(page
);
318 if (page_count(page
) != expected_count
||
319 radix_tree_deref_slot_protected(pslot
, &mapping
->tree_lock
) != page
) {
320 spin_unlock_irq(&mapping
->tree_lock
);
324 if (!page_freeze_refs(page
, expected_count
)) {
325 spin_unlock_irq(&mapping
->tree_lock
);
331 radix_tree_replace_slot(pslot
, newpage
);
333 page_unfreeze_refs(page
, expected_count
- 1);
335 spin_unlock_irq(&mapping
->tree_lock
);
340 * Copy the page to its new location
342 void migrate_page_copy(struct page
*newpage
, struct page
*page
)
345 copy_huge_page(newpage
, page
);
347 copy_highpage(newpage
, page
);
350 SetPageError(newpage
);
351 if (PageReferenced(page
))
352 SetPageReferenced(newpage
);
353 if (PageUptodate(page
))
354 SetPageUptodate(newpage
);
355 if (TestClearPageActive(page
)) {
356 VM_BUG_ON(PageUnevictable(page
));
357 SetPageActive(newpage
);
358 } else if (TestClearPageUnevictable(page
))
359 SetPageUnevictable(newpage
);
360 if (PageChecked(page
))
361 SetPageChecked(newpage
);
362 if (PageMappedToDisk(page
))
363 SetPageMappedToDisk(newpage
);
365 if (PageDirty(page
)) {
366 clear_page_dirty_for_io(page
);
368 * Want to mark the page and the radix tree as dirty, and
369 * redo the accounting that clear_page_dirty_for_io undid,
370 * but we can't use set_page_dirty because that function
371 * is actually a signal that all of the page has become dirty.
372 * Whereas only part of our page may be dirty.
374 __set_page_dirty_nobuffers(newpage
);
377 mlock_migrate_page(newpage
, page
);
378 ksm_migrate_page(newpage
, page
);
380 ClearPageSwapCache(page
);
381 ClearPagePrivate(page
);
382 set_page_private(page
, 0);
383 page
->mapping
= NULL
;
386 * If any waiters have accumulated on the new page then
389 if (PageWriteback(newpage
))
390 end_page_writeback(newpage
);
393 /************************************************************
394 * Migration functions
395 ***********************************************************/
397 /* Always fail migration. Used for mappings that are not movable */
398 int fail_migrate_page(struct address_space
*mapping
,
399 struct page
*newpage
, struct page
*page
)
403 EXPORT_SYMBOL(fail_migrate_page
);
406 * Common logic to directly migrate a single page suitable for
407 * pages that do not use PagePrivate/PagePrivate2.
409 * Pages are locked upon entry and exit.
411 int migrate_page(struct address_space
*mapping
,
412 struct page
*newpage
, struct page
*page
)
416 BUG_ON(PageWriteback(page
)); /* Writeback must be complete */
418 rc
= migrate_page_move_mapping(mapping
, newpage
, page
);
423 migrate_page_copy(newpage
, page
);
426 EXPORT_SYMBOL(migrate_page
);
430 * Migration function for pages with buffers. This function can only be used
431 * if the underlying filesystem guarantees that no other references to "page"
434 int buffer_migrate_page(struct address_space
*mapping
,
435 struct page
*newpage
, struct page
*page
)
437 struct buffer_head
*bh
, *head
;
440 if (!page_has_buffers(page
))
441 return migrate_page(mapping
, newpage
, page
);
443 head
= page_buffers(page
);
445 rc
= migrate_page_move_mapping(mapping
, newpage
, page
);
454 bh
= bh
->b_this_page
;
456 } while (bh
!= head
);
458 ClearPagePrivate(page
);
459 set_page_private(newpage
, page_private(page
));
460 set_page_private(page
, 0);
466 set_bh_page(bh
, newpage
, bh_offset(bh
));
467 bh
= bh
->b_this_page
;
469 } while (bh
!= head
);
471 SetPagePrivate(newpage
);
473 migrate_page_copy(newpage
, page
);
479 bh
= bh
->b_this_page
;
481 } while (bh
!= head
);
485 EXPORT_SYMBOL(buffer_migrate_page
);
489 * Writeback a page to clean the dirty state
491 static int writeout(struct address_space
*mapping
, struct page
*page
)
493 struct writeback_control wbc
= {
494 .sync_mode
= WB_SYNC_NONE
,
497 .range_end
= LLONG_MAX
,
502 if (!mapping
->a_ops
->writepage
)
503 /* No write method for the address space */
506 if (!clear_page_dirty_for_io(page
))
507 /* Someone else already triggered a write */
511 * A dirty page may imply that the underlying filesystem has
512 * the page on some queue. So the page must be clean for
513 * migration. Writeout may mean we loose the lock and the
514 * page state is no longer what we checked for earlier.
515 * At this point we know that the migration attempt cannot
518 remove_migration_ptes(page
, page
);
520 rc
= mapping
->a_ops
->writepage(page
, &wbc
);
522 if (rc
!= AOP_WRITEPAGE_ACTIVATE
)
523 /* unlocked. Relock */
526 return (rc
< 0) ? -EIO
: -EAGAIN
;
530 * Default handling if a filesystem does not provide a migration function.
532 static int fallback_migrate_page(struct address_space
*mapping
,
533 struct page
*newpage
, struct page
*page
)
536 return writeout(mapping
, page
);
539 * Buffers may be managed in a filesystem specific way.
540 * We must have no buffers or drop them.
542 if (page_has_private(page
) &&
543 !try_to_release_page(page
, GFP_KERNEL
))
546 return migrate_page(mapping
, newpage
, page
);
550 * Move a page to a newly allocated page
551 * The page is locked and all ptes have been successfully removed.
553 * The new page will have replaced the old page if this function
560 static int move_to_new_page(struct page
*newpage
, struct page
*page
,
561 int remap_swapcache
, bool sync
)
563 struct address_space
*mapping
;
567 * Block others from accessing the page when we get around to
568 * establishing additional references. We are the only one
569 * holding a reference to the new page at this point.
571 if (!trylock_page(newpage
))
574 /* Prepare mapping for the new page.*/
575 newpage
->index
= page
->index
;
576 newpage
->mapping
= page
->mapping
;
577 if (PageSwapBacked(page
))
578 SetPageSwapBacked(newpage
);
580 mapping
= page_mapping(page
);
582 rc
= migrate_page(mapping
, newpage
, page
);
585 * Do not writeback pages if !sync and migratepage is
586 * not pointing to migrate_page() which is nonblocking
587 * (swapcache/tmpfs uses migratepage = migrate_page).
589 if (PageDirty(page
) && !sync
&&
590 mapping
->a_ops
->migratepage
!= migrate_page
)
592 else if (mapping
->a_ops
->migratepage
)
594 * Most pages have a mapping and most filesystems
595 * should provide a migration function. Anonymous
596 * pages are part of swap space which also has its
597 * own migration function. This is the most common
598 * path for page migration.
600 rc
= mapping
->a_ops
->migratepage(mapping
,
603 rc
= fallback_migrate_page(mapping
, newpage
, page
);
607 newpage
->mapping
= NULL
;
610 remove_migration_ptes(page
, newpage
);
613 unlock_page(newpage
);
618 static int __unmap_and_move(struct page
*page
, struct page
*newpage
,
619 int force
, bool offlining
, bool sync
)
622 int remap_swapcache
= 1;
624 struct mem_cgroup
*mem
;
625 struct anon_vma
*anon_vma
= NULL
;
627 if (!trylock_page(page
)) {
632 * It's not safe for direct compaction to call lock_page.
633 * For example, during page readahead pages are added locked
634 * to the LRU. Later, when the IO completes the pages are
635 * marked uptodate and unlocked. However, the queueing
636 * could be merging multiple pages for one bio (e.g.
637 * mpage_readpages). If an allocation happens for the
638 * second or third page, the process can end up locking
639 * the same page twice and deadlocking. Rather than
640 * trying to be clever about what pages can be locked,
641 * avoid the use of lock_page for direct compaction
644 if (current
->flags
& PF_MEMALLOC
)
651 * Only memory hotplug's offline_pages() caller has locked out KSM,
652 * and can safely migrate a KSM page. The other cases have skipped
653 * PageKsm along with PageReserved - but it is only now when we have
654 * the page lock that we can be certain it will not go KSM beneath us
655 * (KSM will not upgrade a page from PageAnon to PageKsm when it sees
656 * its pagecount raised, but only here do we take the page lock which
659 if (PageKsm(page
) && !offlining
) {
664 /* charge against new page */
665 charge
= mem_cgroup_prepare_migration(page
, newpage
, &mem
, GFP_KERNEL
);
666 if (charge
== -ENOMEM
) {
672 if (PageWriteback(page
)) {
674 * For !sync, there is no point retrying as the retry loop
675 * is expected to be too short for PageWriteback to be cleared
683 wait_on_page_writeback(page
);
686 * By try_to_unmap(), page->mapcount goes down to 0 here. In this case,
687 * we cannot notice that anon_vma is freed while we migrates a page.
688 * This get_anon_vma() delays freeing anon_vma pointer until the end
689 * of migration. File cache pages are no problem because of page_lock()
690 * File Caches may use write_page() or lock_page() in migration, then,
691 * just care Anon page here.
693 if (PageAnon(page
)) {
695 * Only page_lock_anon_vma() understands the subtleties of
696 * getting a hold on an anon_vma from outside one of its mms.
698 anon_vma
= page_get_anon_vma(page
);
703 } else if (PageSwapCache(page
)) {
705 * We cannot be sure that the anon_vma of an unmapped
706 * swapcache page is safe to use because we don't
707 * know in advance if the VMA that this page belonged
708 * to still exists. If the VMA and others sharing the
709 * data have been freed, then the anon_vma could
710 * already be invalid.
712 * To avoid this possibility, swapcache pages get
713 * migrated but are not remapped when migration
723 * Corner case handling:
724 * 1. When a new swap-cache page is read into, it is added to the LRU
725 * and treated as swapcache but it has no rmap yet.
726 * Calling try_to_unmap() against a page->mapping==NULL page will
727 * trigger a BUG. So handle it here.
728 * 2. An orphaned page (see truncate_complete_page) might have
729 * fs-private metadata. The page can be picked up due to memory
730 * offlining. Everywhere else except page reclaim, the page is
731 * invisible to the vm, so the page can not be migrated. So try to
732 * free the metadata, so the page can be freed.
734 if (!page
->mapping
) {
735 VM_BUG_ON(PageAnon(page
));
736 if (page_has_private(page
)) {
737 try_to_free_buffers(page
);
743 /* Establish migration ptes or remove ptes */
744 try_to_unmap(page
, TTU_MIGRATION
|TTU_IGNORE_MLOCK
|TTU_IGNORE_ACCESS
);
747 if (!page_mapped(page
))
748 rc
= move_to_new_page(newpage
, page
, remap_swapcache
, sync
);
750 if (rc
&& remap_swapcache
)
751 remove_migration_ptes(page
, page
);
753 /* Drop an anon_vma reference if we took one */
755 put_anon_vma(anon_vma
);
759 mem_cgroup_end_migration(mem
, page
, newpage
, rc
== 0);
767 * Obtain the lock on page, remove all ptes and migrate the page
768 * to the newly allocated page in newpage.
770 static int unmap_and_move(new_page_t get_new_page
, unsigned long private,
771 struct page
*page
, int force
, bool offlining
, bool sync
)
775 struct page
*newpage
= get_new_page(page
, private, &result
);
780 if (page_count(page
) == 1) {
781 /* page was freed from under us. So we are done. */
785 if (unlikely(PageTransHuge(page
)))
786 if (unlikely(split_huge_page(page
)))
789 rc
= __unmap_and_move(page
, newpage
, force
, offlining
, sync
);
793 * A page that has been migrated has all references
794 * removed and will be freed. A page that has not been
795 * migrated will have kepts its references and be
798 list_del(&page
->lru
);
799 dec_zone_page_state(page
, NR_ISOLATED_ANON
+
800 page_is_file_cache(page
));
801 putback_lru_page(page
);
804 * Move the new page to the LRU. If migration was not successful
805 * then this will free the page.
807 putback_lru_page(newpage
);
812 *result
= page_to_nid(newpage
);
818 * Counterpart of unmap_and_move_page() for hugepage migration.
820 * This function doesn't wait the completion of hugepage I/O
821 * because there is no race between I/O and migration for hugepage.
822 * Note that currently hugepage I/O occurs only in direct I/O
823 * where no lock is held and PG_writeback is irrelevant,
824 * and writeback status of all subpages are counted in the reference
825 * count of the head page (i.e. if all subpages of a 2MB hugepage are
826 * under direct I/O, the reference of the head page is 512 and a bit more.)
827 * This means that when we try to migrate hugepage whose subpages are
828 * doing direct I/O, some references remain after try_to_unmap() and
829 * hugepage migration fails without data corruption.
831 * There is also no race when direct I/O is issued on the page under migration,
832 * because then pte is replaced with migration swap entry and direct I/O code
833 * will wait in the page fault for migration to complete.
835 static int unmap_and_move_huge_page(new_page_t get_new_page
,
836 unsigned long private, struct page
*hpage
,
837 int force
, bool offlining
, bool sync
)
841 struct page
*new_hpage
= get_new_page(hpage
, private, &result
);
842 struct anon_vma
*anon_vma
= NULL
;
849 if (!trylock_page(hpage
)) {
856 anon_vma
= page_get_anon_vma(hpage
);
858 try_to_unmap(hpage
, TTU_MIGRATION
|TTU_IGNORE_MLOCK
|TTU_IGNORE_ACCESS
);
860 if (!page_mapped(hpage
))
861 rc
= move_to_new_page(new_hpage
, hpage
, 1, sync
);
864 remove_migration_ptes(hpage
, hpage
);
867 put_anon_vma(anon_vma
);
872 list_del(&hpage
->lru
);
882 *result
= page_to_nid(new_hpage
);
890 * The function takes one list of pages to migrate and a function
891 * that determines from the page to be migrated and the private data
892 * the target of the move and allocates the page.
894 * The function returns after 10 attempts or if no pages
895 * are movable anymore because to has become empty
896 * or no retryable pages exist anymore.
897 * Caller should call putback_lru_pages to return pages to the LRU
898 * or free list only if ret != 0.
900 * Return: Number of pages not migrated or error code.
902 int migrate_pages(struct list_head
*from
,
903 new_page_t get_new_page
, unsigned long private, bool offlining
,
911 int swapwrite
= current
->flags
& PF_SWAPWRITE
;
915 current
->flags
|= PF_SWAPWRITE
;
917 for(pass
= 0; pass
< 10 && retry
; pass
++) {
920 list_for_each_entry_safe(page
, page2
, from
, lru
) {
923 rc
= unmap_and_move(get_new_page
, private,
924 page
, pass
> 2, offlining
,
936 /* Permanent failure */
945 current
->flags
&= ~PF_SWAPWRITE
;
950 return nr_failed
+ retry
;
953 int migrate_huge_pages(struct list_head
*from
,
954 new_page_t get_new_page
, unsigned long private, bool offlining
,
964 for (pass
= 0; pass
< 10 && retry
; pass
++) {
967 list_for_each_entry_safe(page
, page2
, from
, lru
) {
970 rc
= unmap_and_move_huge_page(get_new_page
,
971 private, page
, pass
> 2, offlining
,
983 /* Permanent failure */
994 return nr_failed
+ retry
;
999 * Move a list of individual pages
1001 struct page_to_node
{
1008 static struct page
*new_page_node(struct page
*p
, unsigned long private,
1011 struct page_to_node
*pm
= (struct page_to_node
*)private;
1013 while (pm
->node
!= MAX_NUMNODES
&& pm
->page
!= p
)
1016 if (pm
->node
== MAX_NUMNODES
)
1019 *result
= &pm
->status
;
1021 return alloc_pages_exact_node(pm
->node
,
1022 GFP_HIGHUSER_MOVABLE
| GFP_THISNODE
, 0);
1026 * Move a set of pages as indicated in the pm array. The addr
1027 * field must be set to the virtual address of the page to be moved
1028 * and the node number must contain a valid target node.
1029 * The pm array ends with node = MAX_NUMNODES.
1031 static int do_move_page_to_node_array(struct mm_struct
*mm
,
1032 struct page_to_node
*pm
,
1036 struct page_to_node
*pp
;
1037 LIST_HEAD(pagelist
);
1039 down_read(&mm
->mmap_sem
);
1042 * Build a list of pages to migrate
1044 for (pp
= pm
; pp
->node
!= MAX_NUMNODES
; pp
++) {
1045 struct vm_area_struct
*vma
;
1049 vma
= find_vma(mm
, pp
->addr
);
1050 if (!vma
|| pp
->addr
< vma
->vm_start
|| !vma_migratable(vma
))
1053 page
= follow_page(vma
, pp
->addr
, FOLL_GET
|FOLL_SPLIT
);
1055 err
= PTR_ERR(page
);
1063 /* Use PageReserved to check for zero page */
1064 if (PageReserved(page
) || PageKsm(page
))
1068 err
= page_to_nid(page
);
1070 if (err
== pp
->node
)
1072 * Node already in the right place
1077 if (page_mapcount(page
) > 1 &&
1081 err
= isolate_lru_page(page
);
1083 list_add_tail(&page
->lru
, &pagelist
);
1084 inc_zone_page_state(page
, NR_ISOLATED_ANON
+
1085 page_is_file_cache(page
));
1089 * Either remove the duplicate refcount from
1090 * isolate_lru_page() or drop the page ref if it was
1099 if (!list_empty(&pagelist
)) {
1100 err
= migrate_pages(&pagelist
, new_page_node
,
1101 (unsigned long)pm
, 0, true);
1103 putback_lru_pages(&pagelist
);
1106 up_read(&mm
->mmap_sem
);
1111 * Migrate an array of page address onto an array of nodes and fill
1112 * the corresponding array of status.
1114 static int do_pages_move(struct mm_struct
*mm
, struct task_struct
*task
,
1115 unsigned long nr_pages
,
1116 const void __user
* __user
*pages
,
1117 const int __user
*nodes
,
1118 int __user
*status
, int flags
)
1120 struct page_to_node
*pm
;
1121 nodemask_t task_nodes
;
1122 unsigned long chunk_nr_pages
;
1123 unsigned long chunk_start
;
1126 task_nodes
= cpuset_mems_allowed(task
);
1129 pm
= (struct page_to_node
*)__get_free_page(GFP_KERNEL
);
1136 * Store a chunk of page_to_node array in a page,
1137 * but keep the last one as a marker
1139 chunk_nr_pages
= (PAGE_SIZE
/ sizeof(struct page_to_node
)) - 1;
1141 for (chunk_start
= 0;
1142 chunk_start
< nr_pages
;
1143 chunk_start
+= chunk_nr_pages
) {
1146 if (chunk_start
+ chunk_nr_pages
> nr_pages
)
1147 chunk_nr_pages
= nr_pages
- chunk_start
;
1149 /* fill the chunk pm with addrs and nodes from user-space */
1150 for (j
= 0; j
< chunk_nr_pages
; j
++) {
1151 const void __user
*p
;
1155 if (get_user(p
, pages
+ j
+ chunk_start
))
1157 pm
[j
].addr
= (unsigned long) p
;
1159 if (get_user(node
, nodes
+ j
+ chunk_start
))
1163 if (node
< 0 || node
>= MAX_NUMNODES
)
1166 if (!node_state(node
, N_HIGH_MEMORY
))
1170 if (!node_isset(node
, task_nodes
))
1176 /* End marker for this chunk */
1177 pm
[chunk_nr_pages
].node
= MAX_NUMNODES
;
1179 /* Migrate this chunk */
1180 err
= do_move_page_to_node_array(mm
, pm
,
1181 flags
& MPOL_MF_MOVE_ALL
);
1185 /* Return status information */
1186 for (j
= 0; j
< chunk_nr_pages
; j
++)
1187 if (put_user(pm
[j
].status
, status
+ j
+ chunk_start
)) {
1195 free_page((unsigned long)pm
);
1201 * Determine the nodes of an array of pages and store it in an array of status.
1203 static void do_pages_stat_array(struct mm_struct
*mm
, unsigned long nr_pages
,
1204 const void __user
**pages
, int *status
)
1208 down_read(&mm
->mmap_sem
);
1210 for (i
= 0; i
< nr_pages
; i
++) {
1211 unsigned long addr
= (unsigned long)(*pages
);
1212 struct vm_area_struct
*vma
;
1216 vma
= find_vma(mm
, addr
);
1217 if (!vma
|| addr
< vma
->vm_start
)
1220 page
= follow_page(vma
, addr
, 0);
1222 err
= PTR_ERR(page
);
1227 /* Use PageReserved to check for zero page */
1228 if (!page
|| PageReserved(page
) || PageKsm(page
))
1231 err
= page_to_nid(page
);
1239 up_read(&mm
->mmap_sem
);
1243 * Determine the nodes of a user array of pages and store it in
1244 * a user array of status.
1246 static int do_pages_stat(struct mm_struct
*mm
, unsigned long nr_pages
,
1247 const void __user
* __user
*pages
,
1250 #define DO_PAGES_STAT_CHUNK_NR 16
1251 const void __user
*chunk_pages
[DO_PAGES_STAT_CHUNK_NR
];
1252 int chunk_status
[DO_PAGES_STAT_CHUNK_NR
];
1255 unsigned long chunk_nr
;
1257 chunk_nr
= nr_pages
;
1258 if (chunk_nr
> DO_PAGES_STAT_CHUNK_NR
)
1259 chunk_nr
= DO_PAGES_STAT_CHUNK_NR
;
1261 if (copy_from_user(chunk_pages
, pages
, chunk_nr
* sizeof(*chunk_pages
)))
1264 do_pages_stat_array(mm
, chunk_nr
, chunk_pages
, chunk_status
);
1266 if (copy_to_user(status
, chunk_status
, chunk_nr
* sizeof(*status
)))
1271 nr_pages
-= chunk_nr
;
1273 return nr_pages
? -EFAULT
: 0;
1277 * Move a list of pages in the address space of the currently executing
1280 SYSCALL_DEFINE6(move_pages
, pid_t
, pid
, unsigned long, nr_pages
,
1281 const void __user
* __user
*, pages
,
1282 const int __user
*, nodes
,
1283 int __user
*, status
, int, flags
)
1285 const struct cred
*cred
= current_cred(), *tcred
;
1286 struct task_struct
*task
;
1287 struct mm_struct
*mm
;
1291 if (flags
& ~(MPOL_MF_MOVE
|MPOL_MF_MOVE_ALL
))
1294 if ((flags
& MPOL_MF_MOVE_ALL
) && !capable(CAP_SYS_NICE
))
1297 /* Find the mm_struct */
1299 task
= pid
? find_task_by_vpid(pid
) : current
;
1304 mm
= get_task_mm(task
);
1311 * Check if this process has the right to modify the specified
1312 * process. The right exists if the process has administrative
1313 * capabilities, superuser privileges or the same
1314 * userid as the target process.
1317 tcred
= __task_cred(task
);
1318 if (cred
->euid
!= tcred
->suid
&& cred
->euid
!= tcred
->uid
&&
1319 cred
->uid
!= tcred
->suid
&& cred
->uid
!= tcred
->uid
&&
1320 !capable(CAP_SYS_NICE
)) {
1327 err
= security_task_movememory(task
);
1332 err
= do_pages_move(mm
, task
, nr_pages
, pages
, nodes
, status
,
1335 err
= do_pages_stat(mm
, nr_pages
, pages
, status
);
1344 * Call migration functions in the vma_ops that may prepare
1345 * memory in a vm for migration. migration functions may perform
1346 * the migration for vmas that do not have an underlying page struct.
1348 int migrate_vmas(struct mm_struct
*mm
, const nodemask_t
*to
,
1349 const nodemask_t
*from
, unsigned long flags
)
1351 struct vm_area_struct
*vma
;
1354 for (vma
= mm
->mmap
; vma
&& !err
; vma
= vma
->vm_next
) {
1355 if (vma
->vm_ops
&& vma
->vm_ops
->migrate
) {
1356 err
= vma
->vm_ops
->migrate(vma
, to
, from
, flags
);