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_present(*pmd
))
119 ptep
= pte_offset_map(pmd
, addr
);
121 if (!is_swap_pte(*ptep
)) {
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 * This function is called from do_swap_page().
185 void migration_entry_wait(struct mm_struct
*mm
, pmd_t
*pmd
,
186 unsigned long address
)
193 ptep
= pte_offset_map_lock(mm
, pmd
, address
, &ptl
);
195 if (!is_swap_pte(pte
))
198 entry
= pte_to_swp_entry(pte
);
199 if (!is_migration_entry(entry
))
202 page
= migration_entry_to_page(entry
);
205 * Once radix-tree replacement of page migration started, page_count
206 * *must* be zero. And, we don't want to call wait_on_page_locked()
207 * against a page without get_page().
208 * So, we use get_page_unless_zero(), here. Even failed, page fault
211 if (!get_page_unless_zero(page
))
213 pte_unmap_unlock(ptep
, ptl
);
214 wait_on_page_locked(page
);
218 pte_unmap_unlock(ptep
, ptl
);
222 * Replace the page in the mapping.
224 * The number of remaining references must be:
225 * 1 for anonymous pages without a mapping
226 * 2 for pages with a mapping
227 * 3 for pages with a mapping and PagePrivate/PagePrivate2 set.
229 static int migrate_page_move_mapping(struct address_space
*mapping
,
230 struct page
*newpage
, struct page
*page
)
236 /* Anonymous page without mapping */
237 if (page_count(page
) != 1)
242 spin_lock_irq(&mapping
->tree_lock
);
244 pslot
= radix_tree_lookup_slot(&mapping
->page_tree
,
247 expected_count
= 2 + page_has_private(page
);
248 if (page_count(page
) != expected_count
||
249 radix_tree_deref_slot_protected(pslot
, &mapping
->tree_lock
) != page
) {
250 spin_unlock_irq(&mapping
->tree_lock
);
254 if (!page_freeze_refs(page
, expected_count
)) {
255 spin_unlock_irq(&mapping
->tree_lock
);
260 * Now we know that no one else is looking at the page.
262 get_page(newpage
); /* add cache reference */
263 if (PageSwapCache(page
)) {
264 SetPageSwapCache(newpage
);
265 set_page_private(newpage
, page_private(page
));
268 radix_tree_replace_slot(pslot
, newpage
);
270 page_unfreeze_refs(page
, expected_count
);
272 * Drop cache reference from old page.
273 * We know this isn't the last reference.
278 * If moved to a different zone then also account
279 * the page for that zone. Other VM counters will be
280 * taken care of when we establish references to the
281 * new page and drop references to the old page.
283 * Note that anonymous pages are accounted for
284 * via NR_FILE_PAGES and NR_ANON_PAGES if they
285 * are mapped to swap space.
287 __dec_zone_page_state(page
, NR_FILE_PAGES
);
288 __inc_zone_page_state(newpage
, NR_FILE_PAGES
);
289 if (PageSwapBacked(page
)) {
290 __dec_zone_page_state(page
, NR_SHMEM
);
291 __inc_zone_page_state(newpage
, NR_SHMEM
);
293 spin_unlock_irq(&mapping
->tree_lock
);
299 * The expected number of remaining references is the same as that
300 * of migrate_page_move_mapping().
302 int migrate_huge_page_move_mapping(struct address_space
*mapping
,
303 struct page
*newpage
, struct page
*page
)
309 if (page_count(page
) != 1)
314 spin_lock_irq(&mapping
->tree_lock
);
316 pslot
= radix_tree_lookup_slot(&mapping
->page_tree
,
319 expected_count
= 2 + page_has_private(page
);
320 if (page_count(page
) != expected_count
||
321 radix_tree_deref_slot_protected(pslot
, &mapping
->tree_lock
) != page
) {
322 spin_unlock_irq(&mapping
->tree_lock
);
326 if (!page_freeze_refs(page
, expected_count
)) {
327 spin_unlock_irq(&mapping
->tree_lock
);
333 radix_tree_replace_slot(pslot
, newpage
);
335 page_unfreeze_refs(page
, expected_count
);
339 spin_unlock_irq(&mapping
->tree_lock
);
344 * Copy the page to its new location
346 void migrate_page_copy(struct page
*newpage
, struct page
*page
)
349 copy_huge_page(newpage
, page
);
351 copy_highpage(newpage
, page
);
354 SetPageError(newpage
);
355 if (PageReferenced(page
))
356 SetPageReferenced(newpage
);
357 if (PageUptodate(page
))
358 SetPageUptodate(newpage
);
359 if (TestClearPageActive(page
)) {
360 VM_BUG_ON(PageUnevictable(page
));
361 SetPageActive(newpage
);
362 } else if (TestClearPageUnevictable(page
))
363 SetPageUnevictable(newpage
);
364 if (PageChecked(page
))
365 SetPageChecked(newpage
);
366 if (PageMappedToDisk(page
))
367 SetPageMappedToDisk(newpage
);
369 if (PageDirty(page
)) {
370 clear_page_dirty_for_io(page
);
372 * Want to mark the page and the radix tree as dirty, and
373 * redo the accounting that clear_page_dirty_for_io undid,
374 * but we can't use set_page_dirty because that function
375 * is actually a signal that all of the page has become dirty.
376 * Wheras only part of our page may be dirty.
378 __set_page_dirty_nobuffers(newpage
);
381 mlock_migrate_page(newpage
, page
);
382 ksm_migrate_page(newpage
, page
);
384 ClearPageSwapCache(page
);
385 ClearPagePrivate(page
);
386 set_page_private(page
, 0);
387 page
->mapping
= NULL
;
390 * If any waiters have accumulated on the new page then
393 if (PageWriteback(newpage
))
394 end_page_writeback(newpage
);
397 /************************************************************
398 * Migration functions
399 ***********************************************************/
401 /* Always fail migration. Used for mappings that are not movable */
402 int fail_migrate_page(struct address_space
*mapping
,
403 struct page
*newpage
, struct page
*page
)
407 EXPORT_SYMBOL(fail_migrate_page
);
410 * Common logic to directly migrate a single page suitable for
411 * pages that do not use PagePrivate/PagePrivate2.
413 * Pages are locked upon entry and exit.
415 int migrate_page(struct address_space
*mapping
,
416 struct page
*newpage
, struct page
*page
)
420 BUG_ON(PageWriteback(page
)); /* Writeback must be complete */
422 rc
= migrate_page_move_mapping(mapping
, newpage
, page
);
427 migrate_page_copy(newpage
, page
);
430 EXPORT_SYMBOL(migrate_page
);
434 * Migration function for pages with buffers. This function can only be used
435 * if the underlying filesystem guarantees that no other references to "page"
438 int buffer_migrate_page(struct address_space
*mapping
,
439 struct page
*newpage
, struct page
*page
)
441 struct buffer_head
*bh
, *head
;
444 if (!page_has_buffers(page
))
445 return migrate_page(mapping
, newpage
, page
);
447 head
= page_buffers(page
);
449 rc
= migrate_page_move_mapping(mapping
, newpage
, page
);
458 bh
= bh
->b_this_page
;
460 } while (bh
!= head
);
462 ClearPagePrivate(page
);
463 set_page_private(newpage
, page_private(page
));
464 set_page_private(page
, 0);
470 set_bh_page(bh
, newpage
, bh_offset(bh
));
471 bh
= bh
->b_this_page
;
473 } while (bh
!= head
);
475 SetPagePrivate(newpage
);
477 migrate_page_copy(newpage
, page
);
483 bh
= bh
->b_this_page
;
485 } while (bh
!= head
);
489 EXPORT_SYMBOL(buffer_migrate_page
);
493 * Writeback a page to clean the dirty state
495 static int writeout(struct address_space
*mapping
, struct page
*page
)
497 struct writeback_control wbc
= {
498 .sync_mode
= WB_SYNC_NONE
,
501 .range_end
= LLONG_MAX
,
506 if (!mapping
->a_ops
->writepage
)
507 /* No write method for the address space */
510 if (!clear_page_dirty_for_io(page
))
511 /* Someone else already triggered a write */
515 * A dirty page may imply that the underlying filesystem has
516 * the page on some queue. So the page must be clean for
517 * migration. Writeout may mean we loose the lock and the
518 * page state is no longer what we checked for earlier.
519 * At this point we know that the migration attempt cannot
522 remove_migration_ptes(page
, page
);
524 rc
= mapping
->a_ops
->writepage(page
, &wbc
);
526 if (rc
!= AOP_WRITEPAGE_ACTIVATE
)
527 /* unlocked. Relock */
530 return (rc
< 0) ? -EIO
: -EAGAIN
;
534 * Default handling if a filesystem does not provide a migration function.
536 static int fallback_migrate_page(struct address_space
*mapping
,
537 struct page
*newpage
, struct page
*page
)
540 return writeout(mapping
, page
);
543 * Buffers may be managed in a filesystem specific way.
544 * We must have no buffers or drop them.
546 if (page_has_private(page
) &&
547 !try_to_release_page(page
, GFP_KERNEL
))
550 return migrate_page(mapping
, newpage
, page
);
554 * Move a page to a newly allocated page
555 * The page is locked and all ptes have been successfully removed.
557 * The new page will have replaced the old page if this function
564 static int move_to_new_page(struct page
*newpage
, struct page
*page
,
567 struct address_space
*mapping
;
571 * Block others from accessing the page when we get around to
572 * establishing additional references. We are the only one
573 * holding a reference to the new page at this point.
575 if (!trylock_page(newpage
))
578 /* Prepare mapping for the new page.*/
579 newpage
->index
= page
->index
;
580 newpage
->mapping
= page
->mapping
;
581 if (PageSwapBacked(page
))
582 SetPageSwapBacked(newpage
);
584 mapping
= page_mapping(page
);
586 rc
= migrate_page(mapping
, newpage
, page
);
587 else if (mapping
->a_ops
->migratepage
)
589 * Most pages have a mapping and most filesystems
590 * should provide a migration function. Anonymous
591 * pages are part of swap space which also has its
592 * own migration function. This is the most common
593 * path for page migration.
595 rc
= mapping
->a_ops
->migratepage(mapping
,
598 rc
= fallback_migrate_page(mapping
, newpage
, page
);
601 newpage
->mapping
= NULL
;
604 remove_migration_ptes(page
, newpage
);
607 unlock_page(newpage
);
613 * Obtain the lock on page, remove all ptes and migrate the page
614 * to the newly allocated page in newpage.
616 static int unmap_and_move(new_page_t get_new_page
, unsigned long private,
617 struct page
*page
, int force
, int offlining
)
621 struct page
*newpage
= get_new_page(page
, private, &result
);
622 int remap_swapcache
= 1;
624 struct mem_cgroup
*mem
= NULL
;
625 struct anon_vma
*anon_vma
= NULL
;
630 if (page_count(page
) == 1) {
631 /* page was freed from under us. So we are done. */
635 /* prepare cgroup just returns 0 or -ENOMEM */
638 if (!trylock_page(page
)) {
645 * Only memory hotplug's offline_pages() caller has locked out KSM,
646 * and can safely migrate a KSM page. The other cases have skipped
647 * PageKsm along with PageReserved - but it is only now when we have
648 * the page lock that we can be certain it will not go KSM beneath us
649 * (KSM will not upgrade a page from PageAnon to PageKsm when it sees
650 * its pagecount raised, but only here do we take the page lock which
653 if (PageKsm(page
) && !offlining
) {
658 /* charge against new page */
659 charge
= mem_cgroup_prepare_migration(page
, newpage
, &mem
);
660 if (charge
== -ENOMEM
) {
666 if (PageWriteback(page
)) {
669 wait_on_page_writeback(page
);
672 * By try_to_unmap(), page->mapcount goes down to 0 here. In this case,
673 * we cannot notice that anon_vma is freed while we migrates a page.
674 * This get_anon_vma() delays freeing anon_vma pointer until the end
675 * of migration. File cache pages are no problem because of page_lock()
676 * File Caches may use write_page() or lock_page() in migration, then,
677 * just care Anon page here.
679 if (PageAnon(page
)) {
681 * Only page_lock_anon_vma() understands the subtleties of
682 * getting a hold on an anon_vma from outside one of its mms.
684 anon_vma
= page_lock_anon_vma(page
);
687 * Take a reference count on the anon_vma if the
688 * page is mapped so that it is guaranteed to
689 * exist when the page is remapped later
691 get_anon_vma(anon_vma
);
692 page_unlock_anon_vma(anon_vma
);
693 } else if (PageSwapCache(page
)) {
695 * We cannot be sure that the anon_vma of an unmapped
696 * swapcache page is safe to use because we don't
697 * know in advance if the VMA that this page belonged
698 * to still exists. If the VMA and others sharing the
699 * data have been freed, then the anon_vma could
700 * already be invalid.
702 * To avoid this possibility, swapcache pages get
703 * migrated but are not remapped when migration
713 * Corner case handling:
714 * 1. When a new swap-cache page is read into, it is added to the LRU
715 * and treated as swapcache but it has no rmap yet.
716 * Calling try_to_unmap() against a page->mapping==NULL page will
717 * trigger a BUG. So handle it here.
718 * 2. An orphaned page (see truncate_complete_page) might have
719 * fs-private metadata. The page can be picked up due to memory
720 * offlining. Everywhere else except page reclaim, the page is
721 * invisible to the vm, so the page can not be migrated. So try to
722 * free the metadata, so the page can be freed.
724 if (!page
->mapping
) {
725 VM_BUG_ON(PageAnon(page
));
726 if (page_has_private(page
)) {
727 try_to_free_buffers(page
);
733 /* Establish migration ptes or remove ptes */
734 try_to_unmap(page
, TTU_MIGRATION
|TTU_IGNORE_MLOCK
|TTU_IGNORE_ACCESS
);
737 if (!page_mapped(page
))
738 rc
= move_to_new_page(newpage
, page
, remap_swapcache
);
740 if (rc
&& remap_swapcache
)
741 remove_migration_ptes(page
, page
);
743 /* Drop an anon_vma reference if we took one */
745 drop_anon_vma(anon_vma
);
749 mem_cgroup_end_migration(mem
, page
, newpage
);
755 * A page that has been migrated has all references
756 * removed and will be freed. A page that has not been
757 * migrated will have kepts its references and be
760 list_del(&page
->lru
);
761 dec_zone_page_state(page
, NR_ISOLATED_ANON
+
762 page_is_file_cache(page
));
763 putback_lru_page(page
);
769 * Move the new page to the LRU. If migration was not successful
770 * then this will free the page.
772 putback_lru_page(newpage
);
778 *result
= page_to_nid(newpage
);
784 * Counterpart of unmap_and_move_page() for hugepage migration.
786 * This function doesn't wait the completion of hugepage I/O
787 * because there is no race between I/O and migration for hugepage.
788 * Note that currently hugepage I/O occurs only in direct I/O
789 * where no lock is held and PG_writeback is irrelevant,
790 * and writeback status of all subpages are counted in the reference
791 * count of the head page (i.e. if all subpages of a 2MB hugepage are
792 * under direct I/O, the reference of the head page is 512 and a bit more.)
793 * This means that when we try to migrate hugepage whose subpages are
794 * doing direct I/O, some references remain after try_to_unmap() and
795 * hugepage migration fails without data corruption.
797 * There is also no race when direct I/O is issued on the page under migration,
798 * because then pte is replaced with migration swap entry and direct I/O code
799 * will wait in the page fault for migration to complete.
801 static int unmap_and_move_huge_page(new_page_t get_new_page
,
802 unsigned long private, struct page
*hpage
,
803 int force
, int offlining
)
807 struct page
*new_hpage
= get_new_page(hpage
, private, &result
);
808 struct anon_vma
*anon_vma
= NULL
;
815 if (!trylock_page(hpage
)) {
821 if (PageAnon(hpage
)) {
822 anon_vma
= page_lock_anon_vma(hpage
);
824 get_anon_vma(anon_vma
);
825 page_unlock_anon_vma(anon_vma
);
829 try_to_unmap(hpage
, TTU_MIGRATION
|TTU_IGNORE_MLOCK
|TTU_IGNORE_ACCESS
);
831 if (!page_mapped(hpage
))
832 rc
= move_to_new_page(new_hpage
, hpage
, 1);
835 remove_migration_ptes(hpage
, hpage
);
838 drop_anon_vma(anon_vma
);
843 list_del(&hpage
->lru
);
853 *result
= page_to_nid(new_hpage
);
861 * The function takes one list of pages to migrate and a function
862 * that determines from the page to be migrated and the private data
863 * the target of the move and allocates the page.
865 * The function returns after 10 attempts or if no pages
866 * are movable anymore because to has become empty
867 * or no retryable pages exist anymore.
868 * Caller should call putback_lru_pages to return pages to the LRU
871 * Return: Number of pages not migrated or error code.
873 int migrate_pages(struct list_head
*from
,
874 new_page_t get_new_page
, unsigned long private, int offlining
)
881 int swapwrite
= current
->flags
& PF_SWAPWRITE
;
885 current
->flags
|= PF_SWAPWRITE
;
887 for(pass
= 0; pass
< 10 && retry
; pass
++) {
890 list_for_each_entry_safe(page
, page2
, from
, lru
) {
893 rc
= unmap_and_move(get_new_page
, private,
894 page
, pass
> 2, offlining
);
905 /* Permanent failure */
914 current
->flags
&= ~PF_SWAPWRITE
;
919 return nr_failed
+ retry
;
922 int migrate_huge_pages(struct list_head
*from
,
923 new_page_t get_new_page
, unsigned long private, int offlining
)
932 for (pass
= 0; pass
< 10 && retry
; pass
++) {
935 list_for_each_entry_safe(page
, page2
, from
, lru
) {
938 rc
= unmap_and_move_huge_page(get_new_page
,
939 private, page
, pass
> 2, offlining
);
950 /* Permanent failure */
959 list_for_each_entry_safe(page
, page2
, from
, lru
)
965 return nr_failed
+ retry
;
970 * Move a list of individual pages
972 struct page_to_node
{
979 static struct page
*new_page_node(struct page
*p
, unsigned long private,
982 struct page_to_node
*pm
= (struct page_to_node
*)private;
984 while (pm
->node
!= MAX_NUMNODES
&& pm
->page
!= p
)
987 if (pm
->node
== MAX_NUMNODES
)
990 *result
= &pm
->status
;
992 return alloc_pages_exact_node(pm
->node
,
993 GFP_HIGHUSER_MOVABLE
| GFP_THISNODE
, 0);
997 * Move a set of pages as indicated in the pm array. The addr
998 * field must be set to the virtual address of the page to be moved
999 * and the node number must contain a valid target node.
1000 * The pm array ends with node = MAX_NUMNODES.
1002 static int do_move_page_to_node_array(struct mm_struct
*mm
,
1003 struct page_to_node
*pm
,
1007 struct page_to_node
*pp
;
1008 LIST_HEAD(pagelist
);
1010 down_read(&mm
->mmap_sem
);
1013 * Build a list of pages to migrate
1015 for (pp
= pm
; pp
->node
!= MAX_NUMNODES
; pp
++) {
1016 struct vm_area_struct
*vma
;
1020 vma
= find_vma(mm
, pp
->addr
);
1021 if (!vma
|| pp
->addr
< vma
->vm_start
|| !vma_migratable(vma
))
1024 page
= follow_page(vma
, pp
->addr
, FOLL_GET
);
1026 err
= PTR_ERR(page
);
1034 /* Use PageReserved to check for zero page */
1035 if (PageReserved(page
) || PageKsm(page
))
1039 err
= page_to_nid(page
);
1041 if (err
== pp
->node
)
1043 * Node already in the right place
1048 if (page_mapcount(page
) > 1 &&
1052 err
= isolate_lru_page(page
);
1054 list_add_tail(&page
->lru
, &pagelist
);
1055 inc_zone_page_state(page
, NR_ISOLATED_ANON
+
1056 page_is_file_cache(page
));
1060 * Either remove the duplicate refcount from
1061 * isolate_lru_page() or drop the page ref if it was
1070 if (!list_empty(&pagelist
)) {
1071 err
= migrate_pages(&pagelist
, new_page_node
,
1072 (unsigned long)pm
, 0);
1074 putback_lru_pages(&pagelist
);
1077 up_read(&mm
->mmap_sem
);
1082 * Migrate an array of page address onto an array of nodes and fill
1083 * the corresponding array of status.
1085 static int do_pages_move(struct mm_struct
*mm
, struct task_struct
*task
,
1086 unsigned long nr_pages
,
1087 const void __user
* __user
*pages
,
1088 const int __user
*nodes
,
1089 int __user
*status
, int flags
)
1091 struct page_to_node
*pm
;
1092 nodemask_t task_nodes
;
1093 unsigned long chunk_nr_pages
;
1094 unsigned long chunk_start
;
1097 task_nodes
= cpuset_mems_allowed(task
);
1100 pm
= (struct page_to_node
*)__get_free_page(GFP_KERNEL
);
1107 * Store a chunk of page_to_node array in a page,
1108 * but keep the last one as a marker
1110 chunk_nr_pages
= (PAGE_SIZE
/ sizeof(struct page_to_node
)) - 1;
1112 for (chunk_start
= 0;
1113 chunk_start
< nr_pages
;
1114 chunk_start
+= chunk_nr_pages
) {
1117 if (chunk_start
+ chunk_nr_pages
> nr_pages
)
1118 chunk_nr_pages
= nr_pages
- chunk_start
;
1120 /* fill the chunk pm with addrs and nodes from user-space */
1121 for (j
= 0; j
< chunk_nr_pages
; j
++) {
1122 const void __user
*p
;
1126 if (get_user(p
, pages
+ j
+ chunk_start
))
1128 pm
[j
].addr
= (unsigned long) p
;
1130 if (get_user(node
, nodes
+ j
+ chunk_start
))
1134 if (node
< 0 || node
>= MAX_NUMNODES
)
1137 if (!node_state(node
, N_HIGH_MEMORY
))
1141 if (!node_isset(node
, task_nodes
))
1147 /* End marker for this chunk */
1148 pm
[chunk_nr_pages
].node
= MAX_NUMNODES
;
1150 /* Migrate this chunk */
1151 err
= do_move_page_to_node_array(mm
, pm
,
1152 flags
& MPOL_MF_MOVE_ALL
);
1156 /* Return status information */
1157 for (j
= 0; j
< chunk_nr_pages
; j
++)
1158 if (put_user(pm
[j
].status
, status
+ j
+ chunk_start
)) {
1166 free_page((unsigned long)pm
);
1172 * Determine the nodes of an array of pages and store it in an array of status.
1174 static void do_pages_stat_array(struct mm_struct
*mm
, unsigned long nr_pages
,
1175 const void __user
**pages
, int *status
)
1179 down_read(&mm
->mmap_sem
);
1181 for (i
= 0; i
< nr_pages
; i
++) {
1182 unsigned long addr
= (unsigned long)(*pages
);
1183 struct vm_area_struct
*vma
;
1187 vma
= find_vma(mm
, addr
);
1188 if (!vma
|| addr
< vma
->vm_start
)
1191 page
= follow_page(vma
, addr
, 0);
1193 err
= PTR_ERR(page
);
1198 /* Use PageReserved to check for zero page */
1199 if (!page
|| PageReserved(page
) || PageKsm(page
))
1202 err
= page_to_nid(page
);
1210 up_read(&mm
->mmap_sem
);
1214 * Determine the nodes of a user array of pages and store it in
1215 * a user array of status.
1217 static int do_pages_stat(struct mm_struct
*mm
, unsigned long nr_pages
,
1218 const void __user
* __user
*pages
,
1221 #define DO_PAGES_STAT_CHUNK_NR 16
1222 const void __user
*chunk_pages
[DO_PAGES_STAT_CHUNK_NR
];
1223 int chunk_status
[DO_PAGES_STAT_CHUNK_NR
];
1226 unsigned long chunk_nr
;
1228 chunk_nr
= nr_pages
;
1229 if (chunk_nr
> DO_PAGES_STAT_CHUNK_NR
)
1230 chunk_nr
= DO_PAGES_STAT_CHUNK_NR
;
1232 if (copy_from_user(chunk_pages
, pages
, chunk_nr
* sizeof(*chunk_pages
)))
1235 do_pages_stat_array(mm
, chunk_nr
, chunk_pages
, chunk_status
);
1237 if (copy_to_user(status
, chunk_status
, chunk_nr
* sizeof(*status
)))
1242 nr_pages
-= chunk_nr
;
1244 return nr_pages
? -EFAULT
: 0;
1248 * Move a list of pages in the address space of the currently executing
1251 SYSCALL_DEFINE6(move_pages
, pid_t
, pid
, unsigned long, nr_pages
,
1252 const void __user
* __user
*, pages
,
1253 const int __user
*, nodes
,
1254 int __user
*, status
, int, flags
)
1256 const struct cred
*cred
= current_cred(), *tcred
;
1257 struct task_struct
*task
;
1258 struct mm_struct
*mm
;
1262 if (flags
& ~(MPOL_MF_MOVE
|MPOL_MF_MOVE_ALL
))
1265 if ((flags
& MPOL_MF_MOVE_ALL
) && !capable(CAP_SYS_NICE
))
1268 /* Find the mm_struct */
1269 read_lock(&tasklist_lock
);
1270 task
= pid
? find_task_by_vpid(pid
) : current
;
1272 read_unlock(&tasklist_lock
);
1275 mm
= get_task_mm(task
);
1276 read_unlock(&tasklist_lock
);
1282 * Check if this process has the right to modify the specified
1283 * process. The right exists if the process has administrative
1284 * capabilities, superuser privileges or the same
1285 * userid as the target process.
1288 tcred
= __task_cred(task
);
1289 if (cred
->euid
!= tcred
->suid
&& cred
->euid
!= tcred
->uid
&&
1290 cred
->uid
!= tcred
->suid
&& cred
->uid
!= tcred
->uid
&&
1291 !capable(CAP_SYS_NICE
)) {
1298 err
= security_task_movememory(task
);
1303 err
= do_pages_move(mm
, task
, nr_pages
, pages
, nodes
, status
,
1306 err
= do_pages_stat(mm
, nr_pages
, pages
, status
);
1315 * Call migration functions in the vma_ops that may prepare
1316 * memory in a vm for migration. migration functions may perform
1317 * the migration for vmas that do not have an underlying page struct.
1319 int migrate_vmas(struct mm_struct
*mm
, const nodemask_t
*to
,
1320 const nodemask_t
*from
, unsigned long flags
)
1322 struct vm_area_struct
*vma
;
1325 for (vma
= mm
->mmap
; vma
&& !err
; vma
= vma
->vm_next
) {
1326 if (vma
->vm_ops
&& vma
->vm_ops
->migrate
) {
1327 err
= vma
->vm_ops
->migrate(vma
, to
, from
, flags
);