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>
40 #define lru_to_page(_head) (list_entry((_head)->prev, struct page, lru))
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_present(*pmd
))
117 ptep
= pte_offset_map(pmd
, addr
);
119 if (!is_swap_pte(*ptep
)) {
124 ptl
= pte_lockptr(mm
, pmd
);
129 if (!is_swap_pte(pte
))
132 entry
= pte_to_swp_entry(pte
);
134 if (!is_migration_entry(entry
) ||
135 migration_entry_to_page(entry
) != old
)
139 pte
= pte_mkold(mk_pte(new, vma
->vm_page_prot
));
140 if (is_write_migration_entry(entry
))
141 pte
= pte_mkwrite(pte
);
142 #ifdef CONFIG_HUGETLB_PAGE
144 pte
= pte_mkhuge(pte
);
146 flush_cache_page(vma
, addr
, pte_pfn(pte
));
147 set_pte_at(mm
, addr
, ptep
, pte
);
151 hugepage_add_anon_rmap(new, vma
, addr
);
154 } else if (PageAnon(new))
155 page_add_anon_rmap(new, vma
, addr
);
157 page_add_file_rmap(new);
159 /* No need to invalidate - it was non-present before */
160 update_mmu_cache(vma
, addr
, ptep
);
162 pte_unmap_unlock(ptep
, ptl
);
168 * Get rid of all migration entries and replace them by
169 * references to the indicated page.
171 static void remove_migration_ptes(struct page
*old
, struct page
*new)
173 rmap_walk(new, remove_migration_pte
, old
);
177 * Something used the pte of a page under migration. We need to
178 * get to the page and wait until migration is finished.
179 * When we return from this function the fault will be retried.
181 * This function is called from do_swap_page().
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 (struct page
*)radix_tree_deref_slot(pslot
) != 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
);
268 page_unfreeze_refs(page
, expected_count
);
270 * Drop cache reference from old page.
271 * We know this isn't the last reference.
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 (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 (struct page
*)radix_tree_deref_slot(pslot
) != 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
);
337 spin_unlock_irq(&mapping
->tree_lock
);
342 * Copy the page to its new location
344 void migrate_page_copy(struct page
*newpage
, struct page
*page
)
347 copy_huge_page(newpage
, page
);
349 copy_highpage(newpage
, page
);
352 SetPageError(newpage
);
353 if (PageReferenced(page
))
354 SetPageReferenced(newpage
);
355 if (PageUptodate(page
))
356 SetPageUptodate(newpage
);
357 if (TestClearPageActive(page
)) {
358 VM_BUG_ON(PageUnevictable(page
));
359 SetPageActive(newpage
);
360 } else if (TestClearPageUnevictable(page
))
361 SetPageUnevictable(newpage
);
362 if (PageChecked(page
))
363 SetPageChecked(newpage
);
364 if (PageMappedToDisk(page
))
365 SetPageMappedToDisk(newpage
);
367 if (PageDirty(page
)) {
368 clear_page_dirty_for_io(page
);
370 * Want to mark the page and the radix tree as dirty, and
371 * redo the accounting that clear_page_dirty_for_io undid,
372 * but we can't use set_page_dirty because that function
373 * is actually a signal that all of the page has become dirty.
374 * Wheras only part of our page may be dirty.
376 __set_page_dirty_nobuffers(newpage
);
379 mlock_migrate_page(newpage
, page
);
380 ksm_migrate_page(newpage
, page
);
382 ClearPageSwapCache(page
);
383 ClearPagePrivate(page
);
384 set_page_private(page
, 0);
385 page
->mapping
= NULL
;
388 * If any waiters have accumulated on the new page then
391 if (PageWriteback(newpage
))
392 end_page_writeback(newpage
);
395 /************************************************************
396 * Migration functions
397 ***********************************************************/
399 /* Always fail migration. Used for mappings that are not movable */
400 int fail_migrate_page(struct address_space
*mapping
,
401 struct page
*newpage
, struct page
*page
)
405 EXPORT_SYMBOL(fail_migrate_page
);
408 * Common logic to directly migrate a single page suitable for
409 * pages that do not use PagePrivate/PagePrivate2.
411 * Pages are locked upon entry and exit.
413 int migrate_page(struct address_space
*mapping
,
414 struct page
*newpage
, struct page
*page
)
418 BUG_ON(PageWriteback(page
)); /* Writeback must be complete */
420 rc
= migrate_page_move_mapping(mapping
, newpage
, page
);
425 migrate_page_copy(newpage
, page
);
428 EXPORT_SYMBOL(migrate_page
);
432 * Migration function for pages with buffers. This function can only be used
433 * if the underlying filesystem guarantees that no other references to "page"
436 int buffer_migrate_page(struct address_space
*mapping
,
437 struct page
*newpage
, struct page
*page
)
439 struct buffer_head
*bh
, *head
;
442 if (!page_has_buffers(page
))
443 return migrate_page(mapping
, newpage
, page
);
445 head
= page_buffers(page
);
447 rc
= migrate_page_move_mapping(mapping
, newpage
, page
);
456 bh
= bh
->b_this_page
;
458 } while (bh
!= head
);
460 ClearPagePrivate(page
);
461 set_page_private(newpage
, page_private(page
));
462 set_page_private(page
, 0);
468 set_bh_page(bh
, newpage
, bh_offset(bh
));
469 bh
= bh
->b_this_page
;
471 } while (bh
!= head
);
473 SetPagePrivate(newpage
);
475 migrate_page_copy(newpage
, page
);
481 bh
= bh
->b_this_page
;
483 } while (bh
!= head
);
487 EXPORT_SYMBOL(buffer_migrate_page
);
491 * Writeback a page to clean the dirty state
493 static int writeout(struct address_space
*mapping
, struct page
*page
)
495 struct writeback_control wbc
= {
496 .sync_mode
= WB_SYNC_NONE
,
499 .range_end
= LLONG_MAX
,
505 if (!mapping
->a_ops
->writepage
)
506 /* No write method for the address space */
509 if (!clear_page_dirty_for_io(page
))
510 /* Someone else already triggered a write */
514 * A dirty page may imply that the underlying filesystem has
515 * the page on some queue. So the page must be clean for
516 * migration. Writeout may mean we loose the lock and the
517 * page state is no longer what we checked for earlier.
518 * At this point we know that the migration attempt cannot
521 remove_migration_ptes(page
, page
);
523 rc
= mapping
->a_ops
->writepage(page
, &wbc
);
525 if (rc
!= AOP_WRITEPAGE_ACTIVATE
)
526 /* unlocked. Relock */
529 return (rc
< 0) ? -EIO
: -EAGAIN
;
533 * Default handling if a filesystem does not provide a migration function.
535 static int fallback_migrate_page(struct address_space
*mapping
,
536 struct page
*newpage
, struct page
*page
)
539 return writeout(mapping
, page
);
542 * Buffers may be managed in a filesystem specific way.
543 * We must have no buffers or drop them.
545 if (page_has_private(page
) &&
546 !try_to_release_page(page
, GFP_KERNEL
))
549 return migrate_page(mapping
, newpage
, page
);
553 * Move a page to a newly allocated page
554 * The page is locked and all ptes have been successfully removed.
556 * The new page will have replaced the old page if this function
563 static int move_to_new_page(struct page
*newpage
, struct page
*page
,
566 struct address_space
*mapping
;
570 * Block others from accessing the page when we get around to
571 * establishing additional references. We are the only one
572 * holding a reference to the new page at this point.
574 if (!trylock_page(newpage
))
577 /* Prepare mapping for the new page.*/
578 newpage
->index
= page
->index
;
579 newpage
->mapping
= page
->mapping
;
580 if (PageSwapBacked(page
))
581 SetPageSwapBacked(newpage
);
583 mapping
= page_mapping(page
);
585 rc
= migrate_page(mapping
, newpage
, page
);
586 else if (mapping
->a_ops
->migratepage
)
588 * Most pages have a mapping and most filesystems
589 * should provide a migration function. Anonymous
590 * pages are part of swap space which also has its
591 * own migration function. This is the most common
592 * path for page migration.
594 rc
= mapping
->a_ops
->migratepage(mapping
,
597 rc
= fallback_migrate_page(mapping
, newpage
, page
);
600 newpage
->mapping
= NULL
;
603 remove_migration_ptes(page
, newpage
);
606 unlock_page(newpage
);
612 * Obtain the lock on page, remove all ptes and migrate the page
613 * to the newly allocated page in newpage.
615 static int unmap_and_move(new_page_t get_new_page
, unsigned long private,
616 struct page
*page
, int force
, int offlining
)
620 struct page
*newpage
= get_new_page(page
, private, &result
);
621 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 rcu_read_lock() 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
)) {
683 /* Determine how to safely use anon_vma */
684 if (!page_mapped(page
)) {
685 if (!PageSwapCache(page
))
689 * We cannot be sure that the anon_vma of an unmapped
690 * swapcache page is safe to use because we don't
691 * know in advance if the VMA that this page belonged
692 * to still exists. If the VMA and others sharing the
693 * data have been freed, then the anon_vma could
694 * already be invalid.
696 * To avoid this possibility, swapcache pages get
697 * migrated but are not remapped when migration
703 * Take a reference count on the anon_vma if the
704 * page is mapped so that it is guaranteed to
705 * exist when the page is remapped later
707 anon_vma
= page_anon_vma(page
);
708 get_anon_vma(anon_vma
);
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 if (!PageAnon(page
) && page_has_private(page
)) {
727 * Go direct to try_to_free_buffers() here because
728 * a) that's what try_to_release_page() would do anyway
729 * b) we may be under rcu_read_lock() here, so we can't
730 * use GFP_KERNEL which is what try_to_release_page()
731 * needs to be effective.
733 try_to_free_buffers(page
);
739 /* Establish migration ptes or remove ptes */
740 try_to_unmap(page
, TTU_MIGRATION
|TTU_IGNORE_MLOCK
|TTU_IGNORE_ACCESS
);
743 if (!page_mapped(page
))
744 rc
= move_to_new_page(newpage
, page
, remap_swapcache
);
746 if (rc
&& remap_swapcache
)
747 remove_migration_ptes(page
, page
);
750 /* Drop an anon_vma reference if we took one */
752 drop_anon_vma(anon_vma
);
758 mem_cgroup_end_migration(mem
, page
, newpage
);
764 * A page that has been migrated has all references
765 * removed and will be freed. A page that has not been
766 * migrated will have kepts its references and be
769 list_del(&page
->lru
);
770 dec_zone_page_state(page
, NR_ISOLATED_ANON
+
771 page_is_file_cache(page
));
772 putback_lru_page(page
);
778 * Move the new page to the LRU. If migration was not successful
779 * then this will free the page.
781 putback_lru_page(newpage
);
787 *result
= page_to_nid(newpage
);
793 * Counterpart of unmap_and_move_page() for hugepage migration.
795 * This function doesn't wait the completion of hugepage I/O
796 * because there is no race between I/O and migration for hugepage.
797 * Note that currently hugepage I/O occurs only in direct I/O
798 * where no lock is held and PG_writeback is irrelevant,
799 * and writeback status of all subpages are counted in the reference
800 * count of the head page (i.e. if all subpages of a 2MB hugepage are
801 * under direct I/O, the reference of the head page is 512 and a bit more.)
802 * This means that when we try to migrate hugepage whose subpages are
803 * doing direct I/O, some references remain after try_to_unmap() and
804 * hugepage migration fails without data corruption.
806 * There is also no race when direct I/O is issued on the page under migration,
807 * because then pte is replaced with migration swap entry and direct I/O code
808 * will wait in the page fault for migration to complete.
810 static int unmap_and_move_huge_page(new_page_t get_new_page
,
811 unsigned long private, struct page
*hpage
,
812 int force
, int offlining
)
816 struct page
*new_hpage
= get_new_page(hpage
, private, &result
);
818 struct anon_vma
*anon_vma
= NULL
;
825 if (!trylock_page(hpage
)) {
831 if (PageAnon(hpage
)) {
835 if (page_mapped(hpage
)) {
836 anon_vma
= page_anon_vma(hpage
);
837 atomic_inc(&anon_vma
->external_refcount
);
841 try_to_unmap(hpage
, TTU_MIGRATION
|TTU_IGNORE_MLOCK
|TTU_IGNORE_ACCESS
);
843 if (!page_mapped(hpage
))
844 rc
= move_to_new_page(new_hpage
, hpage
, 1);
847 remove_migration_ptes(hpage
, hpage
);
849 if (anon_vma
&& atomic_dec_and_lock(&anon_vma
->external_refcount
,
851 int empty
= list_empty(&anon_vma
->head
);
852 spin_unlock(&anon_vma
->lock
);
854 anon_vma_free(anon_vma
);
863 list_del(&hpage
->lru
);
873 *result
= page_to_nid(new_hpage
);
881 * The function takes one list of pages to migrate and a function
882 * that determines from the page to be migrated and the private data
883 * the target of the move and allocates the page.
885 * The function returns after 10 attempts or if no pages
886 * are movable anymore because to has become empty
887 * or no retryable pages exist anymore. All pages will be
888 * returned to the LRU or freed.
890 * Return: Number of pages not migrated or error code.
892 int migrate_pages(struct list_head
*from
,
893 new_page_t get_new_page
, unsigned long private, int offlining
)
900 int swapwrite
= current
->flags
& PF_SWAPWRITE
;
904 current
->flags
|= PF_SWAPWRITE
;
906 for(pass
= 0; pass
< 10 && retry
; pass
++) {
909 list_for_each_entry_safe(page
, page2
, from
, lru
) {
912 rc
= unmap_and_move(get_new_page
, private,
913 page
, pass
> 2, offlining
);
924 /* Permanent failure */
933 current
->flags
&= ~PF_SWAPWRITE
;
935 putback_lru_pages(from
);
940 return nr_failed
+ retry
;
943 int migrate_huge_pages(struct list_head
*from
,
944 new_page_t get_new_page
, unsigned long private, int offlining
)
953 for (pass
= 0; pass
< 10 && retry
; pass
++) {
956 list_for_each_entry_safe(page
, page2
, from
, lru
) {
959 rc
= unmap_and_move_huge_page(get_new_page
,
960 private, page
, pass
> 2, offlining
);
971 /* Permanent failure */
980 list_for_each_entry_safe(page
, page2
, from
, lru
)
986 return nr_failed
+ retry
;
991 * Move a list of individual pages
993 struct page_to_node
{
1000 static struct page
*new_page_node(struct page
*p
, unsigned long private,
1003 struct page_to_node
*pm
= (struct page_to_node
*)private;
1005 while (pm
->node
!= MAX_NUMNODES
&& pm
->page
!= p
)
1008 if (pm
->node
== MAX_NUMNODES
)
1011 *result
= &pm
->status
;
1013 return alloc_pages_exact_node(pm
->node
,
1014 GFP_HIGHUSER_MOVABLE
| GFP_THISNODE
, 0);
1018 * Move a set of pages as indicated in the pm array. The addr
1019 * field must be set to the virtual address of the page to be moved
1020 * and the node number must contain a valid target node.
1021 * The pm array ends with node = MAX_NUMNODES.
1023 static int do_move_page_to_node_array(struct mm_struct
*mm
,
1024 struct page_to_node
*pm
,
1028 struct page_to_node
*pp
;
1029 LIST_HEAD(pagelist
);
1031 down_read(&mm
->mmap_sem
);
1034 * Build a list of pages to migrate
1036 for (pp
= pm
; pp
->node
!= MAX_NUMNODES
; pp
++) {
1037 struct vm_area_struct
*vma
;
1041 vma
= find_vma(mm
, pp
->addr
);
1042 if (!vma
|| !vma_migratable(vma
))
1045 page
= follow_page(vma
, pp
->addr
, FOLL_GET
);
1047 err
= PTR_ERR(page
);
1055 /* Use PageReserved to check for zero page */
1056 if (PageReserved(page
) || PageKsm(page
))
1060 err
= page_to_nid(page
);
1062 if (err
== pp
->node
)
1064 * Node already in the right place
1069 if (page_mapcount(page
) > 1 &&
1073 err
= isolate_lru_page(page
);
1075 list_add_tail(&page
->lru
, &pagelist
);
1076 inc_zone_page_state(page
, NR_ISOLATED_ANON
+
1077 page_is_file_cache(page
));
1081 * Either remove the duplicate refcount from
1082 * isolate_lru_page() or drop the page ref if it was
1091 if (!list_empty(&pagelist
))
1092 err
= migrate_pages(&pagelist
, new_page_node
,
1093 (unsigned long)pm
, 0);
1095 up_read(&mm
->mmap_sem
);
1100 * Migrate an array of page address onto an array of nodes and fill
1101 * the corresponding array of status.
1103 static int do_pages_move(struct mm_struct
*mm
, struct task_struct
*task
,
1104 unsigned long nr_pages
,
1105 const void __user
* __user
*pages
,
1106 const int __user
*nodes
,
1107 int __user
*status
, int flags
)
1109 struct page_to_node
*pm
;
1110 nodemask_t task_nodes
;
1111 unsigned long chunk_nr_pages
;
1112 unsigned long chunk_start
;
1115 task_nodes
= cpuset_mems_allowed(task
);
1118 pm
= (struct page_to_node
*)__get_free_page(GFP_KERNEL
);
1125 * Store a chunk of page_to_node array in a page,
1126 * but keep the last one as a marker
1128 chunk_nr_pages
= (PAGE_SIZE
/ sizeof(struct page_to_node
)) - 1;
1130 for (chunk_start
= 0;
1131 chunk_start
< nr_pages
;
1132 chunk_start
+= chunk_nr_pages
) {
1135 if (chunk_start
+ chunk_nr_pages
> nr_pages
)
1136 chunk_nr_pages
= nr_pages
- chunk_start
;
1138 /* fill the chunk pm with addrs and nodes from user-space */
1139 for (j
= 0; j
< chunk_nr_pages
; j
++) {
1140 const void __user
*p
;
1144 if (get_user(p
, pages
+ j
+ chunk_start
))
1146 pm
[j
].addr
= (unsigned long) p
;
1148 if (get_user(node
, nodes
+ j
+ chunk_start
))
1152 if (node
< 0 || node
>= MAX_NUMNODES
)
1155 if (!node_state(node
, N_HIGH_MEMORY
))
1159 if (!node_isset(node
, task_nodes
))
1165 /* End marker for this chunk */
1166 pm
[chunk_nr_pages
].node
= MAX_NUMNODES
;
1168 /* Migrate this chunk */
1169 err
= do_move_page_to_node_array(mm
, pm
,
1170 flags
& MPOL_MF_MOVE_ALL
);
1174 /* Return status information */
1175 for (j
= 0; j
< chunk_nr_pages
; j
++)
1176 if (put_user(pm
[j
].status
, status
+ j
+ chunk_start
)) {
1184 free_page((unsigned long)pm
);
1190 * Determine the nodes of an array of pages and store it in an array of status.
1192 static void do_pages_stat_array(struct mm_struct
*mm
, unsigned long nr_pages
,
1193 const void __user
**pages
, int *status
)
1197 down_read(&mm
->mmap_sem
);
1199 for (i
= 0; i
< nr_pages
; i
++) {
1200 unsigned long addr
= (unsigned long)(*pages
);
1201 struct vm_area_struct
*vma
;
1205 vma
= find_vma(mm
, addr
);
1209 page
= follow_page(vma
, addr
, 0);
1211 err
= PTR_ERR(page
);
1216 /* Use PageReserved to check for zero page */
1217 if (!page
|| PageReserved(page
) || PageKsm(page
))
1220 err
= page_to_nid(page
);
1228 up_read(&mm
->mmap_sem
);
1232 * Determine the nodes of a user array of pages and store it in
1233 * a user array of status.
1235 static int do_pages_stat(struct mm_struct
*mm
, unsigned long nr_pages
,
1236 const void __user
* __user
*pages
,
1239 #define DO_PAGES_STAT_CHUNK_NR 16
1240 const void __user
*chunk_pages
[DO_PAGES_STAT_CHUNK_NR
];
1241 int chunk_status
[DO_PAGES_STAT_CHUNK_NR
];
1244 unsigned long chunk_nr
;
1246 chunk_nr
= nr_pages
;
1247 if (chunk_nr
> DO_PAGES_STAT_CHUNK_NR
)
1248 chunk_nr
= DO_PAGES_STAT_CHUNK_NR
;
1250 if (copy_from_user(chunk_pages
, pages
, chunk_nr
* sizeof(*chunk_pages
)))
1253 do_pages_stat_array(mm
, chunk_nr
, chunk_pages
, chunk_status
);
1255 if (copy_to_user(status
, chunk_status
, chunk_nr
* sizeof(*status
)))
1260 nr_pages
-= chunk_nr
;
1262 return nr_pages
? -EFAULT
: 0;
1266 * Move a list of pages in the address space of the currently executing
1269 SYSCALL_DEFINE6(move_pages
, pid_t
, pid
, unsigned long, nr_pages
,
1270 const void __user
* __user
*, pages
,
1271 const int __user
*, nodes
,
1272 int __user
*, status
, int, flags
)
1274 const struct cred
*cred
= current_cred(), *tcred
;
1275 struct task_struct
*task
;
1276 struct mm_struct
*mm
;
1280 if (flags
& ~(MPOL_MF_MOVE
|MPOL_MF_MOVE_ALL
))
1283 if ((flags
& MPOL_MF_MOVE_ALL
) && !capable(CAP_SYS_NICE
))
1286 /* Find the mm_struct */
1287 read_lock(&tasklist_lock
);
1288 task
= pid
? find_task_by_vpid(pid
) : current
;
1290 read_unlock(&tasklist_lock
);
1293 mm
= get_task_mm(task
);
1294 read_unlock(&tasklist_lock
);
1300 * Check if this process has the right to modify the specified
1301 * process. The right exists if the process has administrative
1302 * capabilities, superuser privileges or the same
1303 * userid as the target process.
1306 tcred
= __task_cred(task
);
1307 if (cred
->euid
!= tcred
->suid
&& cred
->euid
!= tcred
->uid
&&
1308 cred
->uid
!= tcred
->suid
&& cred
->uid
!= tcred
->uid
&&
1309 !capable(CAP_SYS_NICE
)) {
1316 err
= security_task_movememory(task
);
1321 err
= do_pages_move(mm
, task
, nr_pages
, pages
, nodes
, status
,
1324 err
= do_pages_stat(mm
, nr_pages
, pages
, status
);
1333 * Call migration functions in the vma_ops that may prepare
1334 * memory in a vm for migration. migration functions may perform
1335 * the migration for vmas that do not have an underlying page struct.
1337 int migrate_vmas(struct mm_struct
*mm
, const nodemask_t
*to
,
1338 const nodemask_t
*from
, unsigned long flags
)
1340 struct vm_area_struct
*vma
;
1343 for (vma
= mm
->mmap
; vma
&& !err
; vma
= vma
->vm_next
) {
1344 if (vma
->vm_ops
&& vma
->vm_ops
->migrate
) {
1345 err
= vma
->vm_ops
->migrate(vma
, to
, from
, flags
);
This page took 0.057546 seconds and 6 git commands to generate.