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/module.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>
42 #define lru_to_page(_head) (list_entry((_head)->prev, struct page, lru))
45 * migrate_prep() needs to be called before we start compiling a list of pages
46 * to be migrated using isolate_lru_page(). If scheduling work on other CPUs is
47 * undesirable, use migrate_prep_local()
49 int migrate_prep(void)
52 * Clear the LRU lists so pages can be isolated.
53 * Note that pages may be moved off the LRU after we have
54 * drained them. Those pages will fail to migrate like other
55 * pages that may be busy.
62 /* Do the necessary work of migrate_prep but not if it involves other CPUs */
63 int migrate_prep_local(void)
71 * Add isolated pages on the list back to the LRU under page lock
72 * to avoid leaking evictable pages back onto unevictable list.
74 void putback_lru_pages(struct list_head
*l
)
79 list_for_each_entry_safe(page
, page2
, l
, lru
) {
81 dec_zone_page_state(page
, NR_ISOLATED_ANON
+
82 page_is_file_cache(page
));
83 putback_lru_page(page
);
88 * Restore a potential migration pte to a working pte entry
90 static int remove_migration_pte(struct page
*new, struct vm_area_struct
*vma
,
91 unsigned long addr
, void *old
)
93 struct mm_struct
*mm
= vma
->vm_mm
;
101 if (unlikely(PageHuge(new))) {
102 ptep
= huge_pte_offset(mm
, addr
);
105 ptl
= &mm
->page_table_lock
;
107 pgd
= pgd_offset(mm
, addr
);
108 if (!pgd_present(*pgd
))
111 pud
= pud_offset(pgd
, addr
);
112 if (!pud_present(*pud
))
115 pmd
= pmd_offset(pud
, addr
);
116 if (pmd_trans_huge(*pmd
))
118 if (!pmd_present(*pmd
))
121 ptep
= pte_offset_map(pmd
, addr
);
123 if (!is_swap_pte(*ptep
)) {
128 ptl
= pte_lockptr(mm
, pmd
);
133 if (!is_swap_pte(pte
))
136 entry
= pte_to_swp_entry(pte
);
138 if (!is_migration_entry(entry
) ||
139 migration_entry_to_page(entry
) != old
)
143 pte
= pte_mkold(mk_pte(new, vma
->vm_page_prot
));
144 if (is_write_migration_entry(entry
))
145 pte
= pte_mkwrite(pte
);
146 #ifdef CONFIG_HUGETLB_PAGE
148 pte
= pte_mkhuge(pte
);
150 flush_cache_page(vma
, addr
, pte_pfn(pte
));
151 set_pte_at(mm
, addr
, ptep
, pte
);
155 hugepage_add_anon_rmap(new, vma
, addr
);
158 } else if (PageAnon(new))
159 page_add_anon_rmap(new, vma
, addr
);
161 page_add_file_rmap(new);
163 /* No need to invalidate - it was non-present before */
164 update_mmu_cache(vma
, addr
, ptep
);
166 pte_unmap_unlock(ptep
, ptl
);
172 * Get rid of all migration entries and replace them by
173 * references to the indicated page.
175 static void remove_migration_ptes(struct page
*old
, struct page
*new)
177 rmap_walk(new, remove_migration_pte
, old
);
181 * Something used the pte of a page under migration. We need to
182 * get to the page and wait until migration is finished.
183 * When we return from this function the fault will be retried.
185 * This function is called from do_swap_page().
187 void migration_entry_wait(struct mm_struct
*mm
, pmd_t
*pmd
,
188 unsigned long address
)
195 ptep
= pte_offset_map_lock(mm
, pmd
, address
, &ptl
);
197 if (!is_swap_pte(pte
))
200 entry
= pte_to_swp_entry(pte
);
201 if (!is_migration_entry(entry
))
204 page
= migration_entry_to_page(entry
);
207 * Once radix-tree replacement of page migration started, page_count
208 * *must* be zero. And, we don't want to call wait_on_page_locked()
209 * against a page without get_page().
210 * So, we use get_page_unless_zero(), here. Even failed, page fault
213 if (!get_page_unless_zero(page
))
215 pte_unmap_unlock(ptep
, ptl
);
216 wait_on_page_locked(page
);
220 pte_unmap_unlock(ptep
, ptl
);
224 * Replace the page in the mapping.
226 * The number of remaining references must be:
227 * 1 for anonymous pages without a mapping
228 * 2 for pages with a mapping
229 * 3 for pages with a mapping and PagePrivate/PagePrivate2 set.
231 static int migrate_page_move_mapping(struct address_space
*mapping
,
232 struct page
*newpage
, struct page
*page
)
238 /* Anonymous page without mapping */
239 if (page_count(page
) != 1)
244 spin_lock_irq(&mapping
->tree_lock
);
246 pslot
= radix_tree_lookup_slot(&mapping
->page_tree
,
249 expected_count
= 2 + page_has_private(page
);
250 if (page_count(page
) != expected_count
||
251 radix_tree_deref_slot_protected(pslot
, &mapping
->tree_lock
) != page
) {
252 spin_unlock_irq(&mapping
->tree_lock
);
256 if (!page_freeze_refs(page
, expected_count
)) {
257 spin_unlock_irq(&mapping
->tree_lock
);
262 * Now we know that no one else is looking at the page.
264 get_page(newpage
); /* add cache reference */
265 if (PageSwapCache(page
)) {
266 SetPageSwapCache(newpage
);
267 set_page_private(newpage
, page_private(page
));
270 radix_tree_replace_slot(pslot
, newpage
);
272 page_unfreeze_refs(page
, expected_count
);
274 * Drop cache reference from old page.
275 * We know this isn't the last reference.
280 * If moved to a different zone then also account
281 * the page for that zone. Other VM counters will be
282 * taken care of when we establish references to the
283 * new page and drop references to the old page.
285 * Note that anonymous pages are accounted for
286 * via NR_FILE_PAGES and NR_ANON_PAGES if they
287 * are mapped to swap space.
289 __dec_zone_page_state(page
, NR_FILE_PAGES
);
290 __inc_zone_page_state(newpage
, NR_FILE_PAGES
);
291 if (PageSwapBacked(page
)) {
292 __dec_zone_page_state(page
, NR_SHMEM
);
293 __inc_zone_page_state(newpage
, NR_SHMEM
);
295 spin_unlock_irq(&mapping
->tree_lock
);
301 * The expected number of remaining references is the same as that
302 * of migrate_page_move_mapping().
304 int migrate_huge_page_move_mapping(struct address_space
*mapping
,
305 struct page
*newpage
, struct page
*page
)
311 if (page_count(page
) != 1)
316 spin_lock_irq(&mapping
->tree_lock
);
318 pslot
= radix_tree_lookup_slot(&mapping
->page_tree
,
321 expected_count
= 2 + page_has_private(page
);
322 if (page_count(page
) != expected_count
||
323 radix_tree_deref_slot_protected(pslot
, &mapping
->tree_lock
) != page
) {
324 spin_unlock_irq(&mapping
->tree_lock
);
328 if (!page_freeze_refs(page
, expected_count
)) {
329 spin_unlock_irq(&mapping
->tree_lock
);
335 radix_tree_replace_slot(pslot
, newpage
);
337 page_unfreeze_refs(page
, expected_count
);
341 spin_unlock_irq(&mapping
->tree_lock
);
346 * Copy the page to its new location
348 void migrate_page_copy(struct page
*newpage
, struct page
*page
)
351 copy_huge_page(newpage
, page
);
353 copy_highpage(newpage
, page
);
356 SetPageError(newpage
);
357 if (PageReferenced(page
))
358 SetPageReferenced(newpage
);
359 if (PageUptodate(page
))
360 SetPageUptodate(newpage
);
361 if (TestClearPageActive(page
)) {
362 VM_BUG_ON(PageUnevictable(page
));
363 SetPageActive(newpage
);
364 } else if (TestClearPageUnevictable(page
))
365 SetPageUnevictable(newpage
);
366 if (PageChecked(page
))
367 SetPageChecked(newpage
);
368 if (PageMappedToDisk(page
))
369 SetPageMappedToDisk(newpage
);
371 if (PageDirty(page
)) {
372 clear_page_dirty_for_io(page
);
374 * Want to mark the page and the radix tree as dirty, and
375 * redo the accounting that clear_page_dirty_for_io undid,
376 * but we can't use set_page_dirty because that function
377 * is actually a signal that all of the page has become dirty.
378 * Whereas only part of our page may be dirty.
380 __set_page_dirty_nobuffers(newpage
);
383 mlock_migrate_page(newpage
, page
);
384 ksm_migrate_page(newpage
, page
);
386 ClearPageSwapCache(page
);
387 ClearPagePrivate(page
);
388 set_page_private(page
, 0);
389 page
->mapping
= NULL
;
392 * If any waiters have accumulated on the new page then
395 if (PageWriteback(newpage
))
396 end_page_writeback(newpage
);
399 /************************************************************
400 * Migration functions
401 ***********************************************************/
403 /* Always fail migration. Used for mappings that are not movable */
404 int fail_migrate_page(struct address_space
*mapping
,
405 struct page
*newpage
, struct page
*page
)
409 EXPORT_SYMBOL(fail_migrate_page
);
412 * Common logic to directly migrate a single page suitable for
413 * pages that do not use PagePrivate/PagePrivate2.
415 * Pages are locked upon entry and exit.
417 int migrate_page(struct address_space
*mapping
,
418 struct page
*newpage
, struct page
*page
)
422 BUG_ON(PageWriteback(page
)); /* Writeback must be complete */
424 rc
= migrate_page_move_mapping(mapping
, newpage
, page
);
429 migrate_page_copy(newpage
, page
);
432 EXPORT_SYMBOL(migrate_page
);
436 * Migration function for pages with buffers. This function can only be used
437 * if the underlying filesystem guarantees that no other references to "page"
440 int buffer_migrate_page(struct address_space
*mapping
,
441 struct page
*newpage
, struct page
*page
)
443 struct buffer_head
*bh
, *head
;
446 if (!page_has_buffers(page
))
447 return migrate_page(mapping
, newpage
, page
);
449 head
= page_buffers(page
);
451 rc
= migrate_page_move_mapping(mapping
, newpage
, page
);
460 bh
= bh
->b_this_page
;
462 } while (bh
!= head
);
464 ClearPagePrivate(page
);
465 set_page_private(newpage
, page_private(page
));
466 set_page_private(page
, 0);
472 set_bh_page(bh
, newpage
, bh_offset(bh
));
473 bh
= bh
->b_this_page
;
475 } while (bh
!= head
);
477 SetPagePrivate(newpage
);
479 migrate_page_copy(newpage
, page
);
485 bh
= bh
->b_this_page
;
487 } while (bh
!= head
);
491 EXPORT_SYMBOL(buffer_migrate_page
);
495 * Writeback a page to clean the dirty state
497 static int writeout(struct address_space
*mapping
, struct page
*page
)
499 struct writeback_control wbc
= {
500 .sync_mode
= WB_SYNC_NONE
,
503 .range_end
= LLONG_MAX
,
508 if (!mapping
->a_ops
->writepage
)
509 /* No write method for the address space */
512 if (!clear_page_dirty_for_io(page
))
513 /* Someone else already triggered a write */
517 * A dirty page may imply that the underlying filesystem has
518 * the page on some queue. So the page must be clean for
519 * migration. Writeout may mean we loose the lock and the
520 * page state is no longer what we checked for earlier.
521 * At this point we know that the migration attempt cannot
524 remove_migration_ptes(page
, page
);
526 rc
= mapping
->a_ops
->writepage(page
, &wbc
);
528 if (rc
!= AOP_WRITEPAGE_ACTIVATE
)
529 /* unlocked. Relock */
532 return (rc
< 0) ? -EIO
: -EAGAIN
;
536 * Default handling if a filesystem does not provide a migration function.
538 static int fallback_migrate_page(struct address_space
*mapping
,
539 struct page
*newpage
, struct page
*page
)
542 return writeout(mapping
, page
);
545 * Buffers may be managed in a filesystem specific way.
546 * We must have no buffers or drop them.
548 if (page_has_private(page
) &&
549 !try_to_release_page(page
, GFP_KERNEL
))
552 return migrate_page(mapping
, newpage
, page
);
556 * Move a page to a newly allocated page
557 * The page is locked and all ptes have been successfully removed.
559 * The new page will have replaced the old page if this function
566 static int move_to_new_page(struct page
*newpage
, struct page
*page
,
567 int remap_swapcache
, bool sync
)
569 struct address_space
*mapping
;
573 * Block others from accessing the page when we get around to
574 * establishing additional references. We are the only one
575 * holding a reference to the new page at this point.
577 if (!trylock_page(newpage
))
580 /* Prepare mapping for the new page.*/
581 newpage
->index
= page
->index
;
582 newpage
->mapping
= page
->mapping
;
583 if (PageSwapBacked(page
))
584 SetPageSwapBacked(newpage
);
586 mapping
= page_mapping(page
);
588 rc
= migrate_page(mapping
, newpage
, page
);
591 * Do not writeback pages if !sync and migratepage is
592 * not pointing to migrate_page() which is nonblocking
593 * (swapcache/tmpfs uses migratepage = migrate_page).
595 if (PageDirty(page
) && !sync
&&
596 mapping
->a_ops
->migratepage
!= migrate_page
)
598 else if (mapping
->a_ops
->migratepage
)
600 * Most pages have a mapping and most filesystems
601 * should provide a migration function. Anonymous
602 * pages are part of swap space which also has its
603 * own migration function. This is the most common
604 * path for page migration.
606 rc
= mapping
->a_ops
->migratepage(mapping
,
609 rc
= fallback_migrate_page(mapping
, newpage
, page
);
613 newpage
->mapping
= NULL
;
616 remove_migration_ptes(page
, newpage
);
619 unlock_page(newpage
);
625 * Obtain the lock on page, remove all ptes and migrate the page
626 * to the newly allocated page in newpage.
628 static int unmap_and_move(new_page_t get_new_page
, unsigned long private,
629 struct page
*page
, int force
, bool offlining
, bool sync
)
633 struct page
*newpage
= get_new_page(page
, private, &result
);
634 int remap_swapcache
= 1;
636 struct mem_cgroup
*mem
;
637 struct anon_vma
*anon_vma
= NULL
;
642 if (page_count(page
) == 1) {
643 /* page was freed from under us. So we are done. */
646 if (unlikely(PageTransHuge(page
)))
647 if (unlikely(split_huge_page(page
)))
650 /* prepare cgroup just returns 0 or -ENOMEM */
653 if (!trylock_page(page
)) {
658 * It's not safe for direct compaction to call lock_page.
659 * For example, during page readahead pages are added locked
660 * to the LRU. Later, when the IO completes the pages are
661 * marked uptodate and unlocked. However, the queueing
662 * could be merging multiple pages for one bio (e.g.
663 * mpage_readpages). If an allocation happens for the
664 * second or third page, the process can end up locking
665 * the same page twice and deadlocking. Rather than
666 * trying to be clever about what pages can be locked,
667 * avoid the use of lock_page for direct compaction
670 if (current
->flags
& PF_MEMALLOC
)
677 * Only memory hotplug's offline_pages() caller has locked out KSM,
678 * and can safely migrate a KSM page. The other cases have skipped
679 * PageKsm along with PageReserved - but it is only now when we have
680 * the page lock that we can be certain it will not go KSM beneath us
681 * (KSM will not upgrade a page from PageAnon to PageKsm when it sees
682 * its pagecount raised, but only here do we take the page lock which
685 if (PageKsm(page
) && !offlining
) {
690 /* charge against new page */
691 charge
= mem_cgroup_prepare_migration(page
, newpage
, &mem
, GFP_KERNEL
);
692 if (charge
== -ENOMEM
) {
698 if (PageWriteback(page
)) {
700 * For !sync, there is no point retrying as the retry loop
701 * is expected to be too short for PageWriteback to be cleared
709 wait_on_page_writeback(page
);
712 * By try_to_unmap(), page->mapcount goes down to 0 here. In this case,
713 * we cannot notice that anon_vma is freed while we migrates a page.
714 * This get_anon_vma() delays freeing anon_vma pointer until the end
715 * of migration. File cache pages are no problem because of page_lock()
716 * File Caches may use write_page() or lock_page() in migration, then,
717 * just care Anon page here.
719 if (PageAnon(page
)) {
721 * Only page_lock_anon_vma() understands the subtleties of
722 * getting a hold on an anon_vma from outside one of its mms.
724 anon_vma
= page_lock_anon_vma(page
);
727 * Take a reference count on the anon_vma if the
728 * page is mapped so that it is guaranteed to
729 * exist when the page is remapped later
731 get_anon_vma(anon_vma
);
732 page_unlock_anon_vma(anon_vma
);
733 } else if (PageSwapCache(page
)) {
735 * We cannot be sure that the anon_vma of an unmapped
736 * swapcache page is safe to use because we don't
737 * know in advance if the VMA that this page belonged
738 * to still exists. If the VMA and others sharing the
739 * data have been freed, then the anon_vma could
740 * already be invalid.
742 * To avoid this possibility, swapcache pages get
743 * migrated but are not remapped when migration
753 * Corner case handling:
754 * 1. When a new swap-cache page is read into, it is added to the LRU
755 * and treated as swapcache but it has no rmap yet.
756 * Calling try_to_unmap() against a page->mapping==NULL page will
757 * trigger a BUG. So handle it here.
758 * 2. An orphaned page (see truncate_complete_page) might have
759 * fs-private metadata. The page can be picked up due to memory
760 * offlining. Everywhere else except page reclaim, the page is
761 * invisible to the vm, so the page can not be migrated. So try to
762 * free the metadata, so the page can be freed.
764 if (!page
->mapping
) {
765 VM_BUG_ON(PageAnon(page
));
766 if (page_has_private(page
)) {
767 try_to_free_buffers(page
);
773 /* Establish migration ptes or remove ptes */
774 try_to_unmap(page
, TTU_MIGRATION
|TTU_IGNORE_MLOCK
|TTU_IGNORE_ACCESS
);
777 if (!page_mapped(page
))
778 rc
= move_to_new_page(newpage
, page
, remap_swapcache
, sync
);
780 if (rc
&& remap_swapcache
)
781 remove_migration_ptes(page
, page
);
783 /* Drop an anon_vma reference if we took one */
785 put_anon_vma(anon_vma
);
789 mem_cgroup_end_migration(mem
, page
, newpage
, rc
== 0);
796 * A page that has been migrated has all references
797 * removed and will be freed. A page that has not been
798 * migrated will have kepts its references and be
801 list_del(&page
->lru
);
802 dec_zone_page_state(page
, NR_ISOLATED_ANON
+
803 page_is_file_cache(page
));
804 putback_lru_page(page
);
808 * Move the new page to the LRU. If migration was not successful
809 * then this will free the page.
811 putback_lru_page(newpage
);
817 *result
= page_to_nid(newpage
);
823 * Counterpart of unmap_and_move_page() for hugepage migration.
825 * This function doesn't wait the completion of hugepage I/O
826 * because there is no race between I/O and migration for hugepage.
827 * Note that currently hugepage I/O occurs only in direct I/O
828 * where no lock is held and PG_writeback is irrelevant,
829 * and writeback status of all subpages are counted in the reference
830 * count of the head page (i.e. if all subpages of a 2MB hugepage are
831 * under direct I/O, the reference of the head page is 512 and a bit more.)
832 * This means that when we try to migrate hugepage whose subpages are
833 * doing direct I/O, some references remain after try_to_unmap() and
834 * hugepage migration fails without data corruption.
836 * There is also no race when direct I/O is issued on the page under migration,
837 * because then pte is replaced with migration swap entry and direct I/O code
838 * will wait in the page fault for migration to complete.
840 static int unmap_and_move_huge_page(new_page_t get_new_page
,
841 unsigned long private, struct page
*hpage
,
842 int force
, bool offlining
, bool sync
)
846 struct page
*new_hpage
= get_new_page(hpage
, private, &result
);
847 struct anon_vma
*anon_vma
= NULL
;
854 if (!trylock_page(hpage
)) {
860 if (PageAnon(hpage
)) {
861 anon_vma
= page_lock_anon_vma(hpage
);
863 get_anon_vma(anon_vma
);
864 page_unlock_anon_vma(anon_vma
);
868 try_to_unmap(hpage
, TTU_MIGRATION
|TTU_IGNORE_MLOCK
|TTU_IGNORE_ACCESS
);
870 if (!page_mapped(hpage
))
871 rc
= move_to_new_page(new_hpage
, hpage
, 1, sync
);
874 remove_migration_ptes(hpage
, hpage
);
877 put_anon_vma(anon_vma
);
882 list_del(&hpage
->lru
);
892 *result
= page_to_nid(new_hpage
);
900 * The function takes one list of pages to migrate and a function
901 * that determines from the page to be migrated and the private data
902 * the target of the move and allocates the page.
904 * The function returns after 10 attempts or if no pages
905 * are movable anymore because to has become empty
906 * or no retryable pages exist anymore.
907 * Caller should call putback_lru_pages to return pages to the LRU
908 * or free list only if ret != 0.
910 * Return: Number of pages not migrated or error code.
912 int migrate_pages(struct list_head
*from
,
913 new_page_t get_new_page
, unsigned long private, bool offlining
,
921 int swapwrite
= current
->flags
& PF_SWAPWRITE
;
925 current
->flags
|= PF_SWAPWRITE
;
927 for(pass
= 0; pass
< 10 && retry
; pass
++) {
930 list_for_each_entry_safe(page
, page2
, from
, lru
) {
933 rc
= unmap_and_move(get_new_page
, private,
934 page
, pass
> 2, offlining
,
946 /* Permanent failure */
955 current
->flags
&= ~PF_SWAPWRITE
;
960 return nr_failed
+ retry
;
963 int migrate_huge_pages(struct list_head
*from
,
964 new_page_t get_new_page
, unsigned long private, bool offlining
,
974 for (pass
= 0; pass
< 10 && retry
; pass
++) {
977 list_for_each_entry_safe(page
, page2
, from
, lru
) {
980 rc
= unmap_and_move_huge_page(get_new_page
,
981 private, page
, pass
> 2, offlining
,
993 /* Permanent failure */
1004 return nr_failed
+ retry
;
1009 * Move a list of individual pages
1011 struct page_to_node
{
1018 static struct page
*new_page_node(struct page
*p
, unsigned long private,
1021 struct page_to_node
*pm
= (struct page_to_node
*)private;
1023 while (pm
->node
!= MAX_NUMNODES
&& pm
->page
!= p
)
1026 if (pm
->node
== MAX_NUMNODES
)
1029 *result
= &pm
->status
;
1031 return alloc_pages_exact_node(pm
->node
,
1032 GFP_HIGHUSER_MOVABLE
| GFP_THISNODE
, 0);
1036 * Move a set of pages as indicated in the pm array. The addr
1037 * field must be set to the virtual address of the page to be moved
1038 * and the node number must contain a valid target node.
1039 * The pm array ends with node = MAX_NUMNODES.
1041 static int do_move_page_to_node_array(struct mm_struct
*mm
,
1042 struct page_to_node
*pm
,
1046 struct page_to_node
*pp
;
1047 LIST_HEAD(pagelist
);
1049 down_read(&mm
->mmap_sem
);
1052 * Build a list of pages to migrate
1054 for (pp
= pm
; pp
->node
!= MAX_NUMNODES
; pp
++) {
1055 struct vm_area_struct
*vma
;
1059 vma
= find_vma(mm
, pp
->addr
);
1060 if (!vma
|| pp
->addr
< vma
->vm_start
|| !vma_migratable(vma
))
1063 page
= follow_page(vma
, pp
->addr
, FOLL_GET
|FOLL_SPLIT
);
1065 err
= PTR_ERR(page
);
1073 /* Use PageReserved to check for zero page */
1074 if (PageReserved(page
) || PageKsm(page
))
1078 err
= page_to_nid(page
);
1080 if (err
== pp
->node
)
1082 * Node already in the right place
1087 if (page_mapcount(page
) > 1 &&
1091 err
= isolate_lru_page(page
);
1093 list_add_tail(&page
->lru
, &pagelist
);
1094 inc_zone_page_state(page
, NR_ISOLATED_ANON
+
1095 page_is_file_cache(page
));
1099 * Either remove the duplicate refcount from
1100 * isolate_lru_page() or drop the page ref if it was
1109 if (!list_empty(&pagelist
)) {
1110 err
= migrate_pages(&pagelist
, new_page_node
,
1111 (unsigned long)pm
, 0, true);
1113 putback_lru_pages(&pagelist
);
1116 up_read(&mm
->mmap_sem
);
1121 * Migrate an array of page address onto an array of nodes and fill
1122 * the corresponding array of status.
1124 static int do_pages_move(struct mm_struct
*mm
, struct task_struct
*task
,
1125 unsigned long nr_pages
,
1126 const void __user
* __user
*pages
,
1127 const int __user
*nodes
,
1128 int __user
*status
, int flags
)
1130 struct page_to_node
*pm
;
1131 nodemask_t task_nodes
;
1132 unsigned long chunk_nr_pages
;
1133 unsigned long chunk_start
;
1136 task_nodes
= cpuset_mems_allowed(task
);
1139 pm
= (struct page_to_node
*)__get_free_page(GFP_KERNEL
);
1146 * Store a chunk of page_to_node array in a page,
1147 * but keep the last one as a marker
1149 chunk_nr_pages
= (PAGE_SIZE
/ sizeof(struct page_to_node
)) - 1;
1151 for (chunk_start
= 0;
1152 chunk_start
< nr_pages
;
1153 chunk_start
+= chunk_nr_pages
) {
1156 if (chunk_start
+ chunk_nr_pages
> nr_pages
)
1157 chunk_nr_pages
= nr_pages
- chunk_start
;
1159 /* fill the chunk pm with addrs and nodes from user-space */
1160 for (j
= 0; j
< chunk_nr_pages
; j
++) {
1161 const void __user
*p
;
1165 if (get_user(p
, pages
+ j
+ chunk_start
))
1167 pm
[j
].addr
= (unsigned long) p
;
1169 if (get_user(node
, nodes
+ j
+ chunk_start
))
1173 if (node
< 0 || node
>= MAX_NUMNODES
)
1176 if (!node_state(node
, N_HIGH_MEMORY
))
1180 if (!node_isset(node
, task_nodes
))
1186 /* End marker for this chunk */
1187 pm
[chunk_nr_pages
].node
= MAX_NUMNODES
;
1189 /* Migrate this chunk */
1190 err
= do_move_page_to_node_array(mm
, pm
,
1191 flags
& MPOL_MF_MOVE_ALL
);
1195 /* Return status information */
1196 for (j
= 0; j
< chunk_nr_pages
; j
++)
1197 if (put_user(pm
[j
].status
, status
+ j
+ chunk_start
)) {
1205 free_page((unsigned long)pm
);
1211 * Determine the nodes of an array of pages and store it in an array of status.
1213 static void do_pages_stat_array(struct mm_struct
*mm
, unsigned long nr_pages
,
1214 const void __user
**pages
, int *status
)
1218 down_read(&mm
->mmap_sem
);
1220 for (i
= 0; i
< nr_pages
; i
++) {
1221 unsigned long addr
= (unsigned long)(*pages
);
1222 struct vm_area_struct
*vma
;
1226 vma
= find_vma(mm
, addr
);
1227 if (!vma
|| addr
< vma
->vm_start
)
1230 page
= follow_page(vma
, addr
, 0);
1232 err
= PTR_ERR(page
);
1237 /* Use PageReserved to check for zero page */
1238 if (!page
|| PageReserved(page
) || PageKsm(page
))
1241 err
= page_to_nid(page
);
1249 up_read(&mm
->mmap_sem
);
1253 * Determine the nodes of a user array of pages and store it in
1254 * a user array of status.
1256 static int do_pages_stat(struct mm_struct
*mm
, unsigned long nr_pages
,
1257 const void __user
* __user
*pages
,
1260 #define DO_PAGES_STAT_CHUNK_NR 16
1261 const void __user
*chunk_pages
[DO_PAGES_STAT_CHUNK_NR
];
1262 int chunk_status
[DO_PAGES_STAT_CHUNK_NR
];
1265 unsigned long chunk_nr
;
1267 chunk_nr
= nr_pages
;
1268 if (chunk_nr
> DO_PAGES_STAT_CHUNK_NR
)
1269 chunk_nr
= DO_PAGES_STAT_CHUNK_NR
;
1271 if (copy_from_user(chunk_pages
, pages
, chunk_nr
* sizeof(*chunk_pages
)))
1274 do_pages_stat_array(mm
, chunk_nr
, chunk_pages
, chunk_status
);
1276 if (copy_to_user(status
, chunk_status
, chunk_nr
* sizeof(*status
)))
1281 nr_pages
-= chunk_nr
;
1283 return nr_pages
? -EFAULT
: 0;
1287 * Move a list of pages in the address space of the currently executing
1290 SYSCALL_DEFINE6(move_pages
, pid_t
, pid
, unsigned long, nr_pages
,
1291 const void __user
* __user
*, pages
,
1292 const int __user
*, nodes
,
1293 int __user
*, status
, int, flags
)
1295 const struct cred
*cred
= current_cred(), *tcred
;
1296 struct task_struct
*task
;
1297 struct mm_struct
*mm
;
1301 if (flags
& ~(MPOL_MF_MOVE
|MPOL_MF_MOVE_ALL
))
1304 if ((flags
& MPOL_MF_MOVE_ALL
) && !capable(CAP_SYS_NICE
))
1307 /* Find the mm_struct */
1309 task
= pid
? find_task_by_vpid(pid
) : current
;
1314 mm
= get_task_mm(task
);
1321 * Check if this process has the right to modify the specified
1322 * process. The right exists if the process has administrative
1323 * capabilities, superuser privileges or the same
1324 * userid as the target process.
1327 tcred
= __task_cred(task
);
1328 if (cred
->euid
!= tcred
->suid
&& cred
->euid
!= tcred
->uid
&&
1329 cred
->uid
!= tcred
->suid
&& cred
->uid
!= tcred
->uid
&&
1330 !capable(CAP_SYS_NICE
)) {
1337 err
= security_task_movememory(task
);
1342 err
= do_pages_move(mm
, task
, nr_pages
, pages
, nodes
, status
,
1345 err
= do_pages_stat(mm
, nr_pages
, pages
, status
);
1354 * Call migration functions in the vma_ops that may prepare
1355 * memory in a vm for migration. migration functions may perform
1356 * the migration for vmas that do not have an underlying page struct.
1358 int migrate_vmas(struct mm_struct
*mm
, const nodemask_t
*to
,
1359 const nodemask_t
*from
, unsigned long flags
)
1361 struct vm_area_struct
*vma
;
1364 for (vma
= mm
->mmap
; vma
&& !err
; vma
= vma
->vm_next
) {
1365 if (vma
->vm_ops
&& vma
->vm_ops
->migrate
) {
1366 err
= vma
->vm_ops
->migrate(vma
, to
, from
, flags
);