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)
67 EXPORT_SYMBOL(migrate_prep);
70 * Add isolated pages on the list back to the LRU under page lock
71 * to avoid leaking evictable pages back onto unevictable list.
73 void putback_lru_pages(struct list_head *l)
78 list_for_each_entry_safe(page, page2, l, lru) {
80 dec_zone_page_state(page, NR_ISOLATED_ANON +
81 page_is_file_cache(page));
82 putback_lru_page(page);
85 EXPORT_SYMBOL(putback_lru_pages);
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 (struct page *)radix_tree_deref_slot(pslot) != 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 (struct page *)radix_tree_deref_slot(pslot) != 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;
625 struct mem_cgroup *mem = NULL;
626 struct anon_vma *anon_vma = NULL;
631 if (page_count(page) == 1) {
632 /* page was freed from under us. So we are done. */
636 /* prepare cgroup just returns 0 or -ENOMEM */
639 if (!trylock_page(page)) {
646 * Only memory hotplug's offline_pages() caller has locked out KSM,
647 * and can safely migrate a KSM page. The other cases have skipped
648 * PageKsm along with PageReserved - but it is only now when we have
649 * the page lock that we can be certain it will not go KSM beneath us
650 * (KSM will not upgrade a page from PageAnon to PageKsm when it sees
651 * its pagecount raised, but only here do we take the page lock which
654 if (PageKsm(page) && !offlining) {
659 /* charge against new page */
660 charge = mem_cgroup_prepare_migration(page, newpage, &mem);
661 if (charge == -ENOMEM) {
667 if (PageWriteback(page)) {
670 wait_on_page_writeback(page);
673 * By try_to_unmap(), page->mapcount goes down to 0 here. In this case,
674 * we cannot notice that anon_vma is freed while we migrates a page.
675 * This rcu_read_lock() delays freeing anon_vma pointer until the end
676 * of migration. File cache pages are no problem because of page_lock()
677 * File Caches may use write_page() or lock_page() in migration, then,
678 * just care Anon page here.
680 if (PageAnon(page)) {
684 /* Determine how to safely use anon_vma */
685 if (!page_mapped(page)) {
686 if (!PageSwapCache(page))
690 * We cannot be sure that the anon_vma of an unmapped
691 * swapcache page is safe to use because we don't
692 * know in advance if the VMA that this page belonged
693 * to still exists. If the VMA and others sharing the
694 * data have been freed, then the anon_vma could
695 * already be invalid.
697 * To avoid this possibility, swapcache pages get
698 * migrated but are not remapped when migration
704 * Take a reference count on the anon_vma if the
705 * page is mapped so that it is guaranteed to
706 * exist when the page is remapped later
708 anon_vma = page_anon_vma(page);
709 get_anon_vma(anon_vma);
714 * Corner case handling:
715 * 1. When a new swap-cache page is read into, it is added to the LRU
716 * and treated as swapcache but it has no rmap yet.
717 * Calling try_to_unmap() against a page->mapping==NULL page will
718 * trigger a BUG. So handle it here.
719 * 2. An orphaned page (see truncate_complete_page) might have
720 * fs-private metadata. The page can be picked up due to memory
721 * offlining. Everywhere else except page reclaim, the page is
722 * invisible to the vm, so the page can not be migrated. So try to
723 * free the metadata, so the page can be freed.
725 if (!page->mapping) {
726 if (!PageAnon(page) && page_has_private(page)) {
728 * Go direct to try_to_free_buffers() here because
729 * a) that's what try_to_release_page() would do anyway
730 * b) we may be under rcu_read_lock() here, so we can't
731 * use GFP_KERNEL which is what try_to_release_page()
732 * needs to be effective.
734 try_to_free_buffers(page);
740 /* Establish migration ptes or remove ptes */
741 try_to_unmap(page, TTU_MIGRATION|TTU_IGNORE_MLOCK|TTU_IGNORE_ACCESS);
744 if (!page_mapped(page))
745 rc = move_to_new_page(newpage, page, remap_swapcache);
747 if (rc && remap_swapcache)
748 remove_migration_ptes(page, page);
751 /* Drop an anon_vma reference if we took one */
753 drop_anon_vma(anon_vma);
759 mem_cgroup_end_migration(mem, page, newpage);
765 * A page that has been migrated has all references
766 * removed and will be freed. A page that has not been
767 * migrated will have kepts its references and be
770 list_del(&page->lru);
771 if (PageMemError(page)) {
774 * A page with a memory error that has
775 * been migrated will not be moved to
781 * The page failed to migrate and will not
782 * be added to the bad page list. Clearing
783 * the error bit will allow another attempt
784 * to migrate if it gets another correctable
787 ClearPageMemError(page);
790 dec_zone_page_state(page, NR_ISOLATED_ANON +
791 page_is_file_cache(page));
792 putback_lru_page(page);
798 * Move the new page to the LRU. If migration was not successful
799 * then this will free the page.
801 putback_lru_page(newpage);
807 *result = page_to_nid(newpage);
813 * Counterpart of unmap_and_move_page() for hugepage migration.
815 * This function doesn't wait the completion of hugepage I/O
816 * because there is no race between I/O and migration for hugepage.
817 * Note that currently hugepage I/O occurs only in direct I/O
818 * where no lock is held and PG_writeback is irrelevant,
819 * and writeback status of all subpages are counted in the reference
820 * count of the head page (i.e. if all subpages of a 2MB hugepage are
821 * under direct I/O, the reference of the head page is 512 and a bit more.)
822 * This means that when we try to migrate hugepage whose subpages are
823 * doing direct I/O, some references remain after try_to_unmap() and
824 * hugepage migration fails without data corruption.
826 * There is also no race when direct I/O is issued on the page under migration,
827 * because then pte is replaced with migration swap entry and direct I/O code
828 * will wait in the page fault for migration to complete.
830 static int unmap_and_move_huge_page(new_page_t get_new_page,
831 unsigned long private, struct page *hpage,
832 int force, int offlining)
836 struct page *new_hpage = get_new_page(hpage, private, &result);
838 struct anon_vma *anon_vma = NULL;
845 if (!trylock_page(hpage)) {
851 if (PageAnon(hpage)) {
855 if (page_mapped(hpage)) {
856 anon_vma = page_anon_vma(hpage);
857 atomic_inc(&anon_vma->external_refcount);
861 try_to_unmap(hpage, TTU_MIGRATION|TTU_IGNORE_MLOCK|TTU_IGNORE_ACCESS);
863 if (!page_mapped(hpage))
864 rc = move_to_new_page(new_hpage, hpage, 1);
867 remove_migration_ptes(hpage, hpage);
869 if (anon_vma && atomic_dec_and_lock(&anon_vma->external_refcount,
871 int empty = list_empty(&anon_vma->head);
872 spin_unlock(&anon_vma->lock);
874 anon_vma_free(anon_vma);
883 list_del(&hpage->lru);
893 *result = page_to_nid(new_hpage);
901 * The function takes one list of pages to migrate and a function
902 * that determines from the page to be migrated and the private data
903 * the target of the move and allocates the page.
905 * The function returns after 10 attempts or if no pages
906 * are movable anymore because to has become empty
907 * or no retryable pages exist anymore.
908 * Caller should call putback_lru_pages to return pages to the LRU
911 * Return: Number of pages not migrated or error code.
913 int migrate_pages(struct list_head *from,
914 new_page_t get_new_page, unsigned long private, int 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);
945 /* Permanent failure */
953 list_for_each_entry_safe(page, page2, from, lru)
954 if (PageMemError(page))
956 * The page failed to migrate. Clearing
957 * the error bit will allow another attempt
958 * to migrate if it gets another correctable
961 ClearPageMemError(page);
965 current->flags &= ~PF_SWAPWRITE;
970 return nr_failed + retry;
973 int migrate_huge_pages(struct list_head *from,
974 new_page_t get_new_page, unsigned long private, int offlining)
983 for (pass = 0; pass < 10 && retry; pass++) {
986 list_for_each_entry_safe(page, page2, from, lru) {
989 rc = unmap_and_move_huge_page(get_new_page,
990 private, page, pass > 2, offlining);
1001 /* Permanent failure */
1010 list_for_each_entry_safe(page, page2, from, lru)
1016 return nr_failed + retry;
1018 EXPORT_SYMBOL(migrate_pages);
1022 * Move a list of individual pages
1024 struct page_to_node {
1031 static struct page *new_page_node(struct page *p, unsigned long private,
1034 struct page_to_node *pm = (struct page_to_node *)private;
1036 while (pm->node != MAX_NUMNODES && pm->page != p)
1039 if (pm->node == MAX_NUMNODES)
1042 *result = &pm->status;
1044 return alloc_pages_exact_node(pm->node,
1045 GFP_HIGHUSER_MOVABLE | GFP_THISNODE, 0);
1049 * Move a set of pages as indicated in the pm array. The addr
1050 * field must be set to the virtual address of the page to be moved
1051 * and the node number must contain a valid target node.
1052 * The pm array ends with node = MAX_NUMNODES.
1054 static int do_move_page_to_node_array(struct mm_struct *mm,
1055 struct page_to_node *pm,
1059 struct page_to_node *pp;
1060 LIST_HEAD(pagelist);
1062 down_read(&mm->mmap_sem);
1065 * Build a list of pages to migrate
1067 for (pp = pm; pp->node != MAX_NUMNODES; pp++) {
1068 struct vm_area_struct *vma;
1072 vma = find_vma(mm, pp->addr);
1073 if (!vma || pp->addr < vma->vm_start || !vma_migratable(vma))
1076 page = follow_page(vma, pp->addr, FOLL_GET);
1078 err = PTR_ERR(page);
1086 /* Use PageReserved to check for zero page */
1087 if (PageReserved(page) || PageKsm(page))
1091 err = page_to_nid(page);
1093 if (err == pp->node)
1095 * Node already in the right place
1100 if (page_mapcount(page) > 1 &&
1104 err = isolate_lru_page(page);
1106 list_add_tail(&page->lru, &pagelist);
1107 inc_zone_page_state(page, NR_ISOLATED_ANON +
1108 page_is_file_cache(page));
1112 * Either remove the duplicate refcount from
1113 * isolate_lru_page() or drop the page ref if it was
1122 if (!list_empty(&pagelist)) {
1123 err = migrate_pages(&pagelist, new_page_node,
1124 (unsigned long)pm, 0);
1126 putback_lru_pages(&pagelist);
1129 up_read(&mm->mmap_sem);
1134 * Migrate an array of page address onto an array of nodes and fill
1135 * the corresponding array of status.
1137 static int do_pages_move(struct mm_struct *mm, struct task_struct *task,
1138 unsigned long nr_pages,
1139 const void __user * __user *pages,
1140 const int __user *nodes,
1141 int __user *status, int flags)
1143 struct page_to_node *pm;
1144 nodemask_t task_nodes;
1145 unsigned long chunk_nr_pages;
1146 unsigned long chunk_start;
1149 task_nodes = cpuset_mems_allowed(task);
1152 pm = (struct page_to_node *)__get_free_page(GFP_KERNEL);
1159 * Store a chunk of page_to_node array in a page,
1160 * but keep the last one as a marker
1162 chunk_nr_pages = (PAGE_SIZE / sizeof(struct page_to_node)) - 1;
1164 for (chunk_start = 0;
1165 chunk_start < nr_pages;
1166 chunk_start += chunk_nr_pages) {
1169 if (chunk_start + chunk_nr_pages > nr_pages)
1170 chunk_nr_pages = nr_pages - chunk_start;
1172 /* fill the chunk pm with addrs and nodes from user-space */
1173 for (j = 0; j < chunk_nr_pages; j++) {
1174 const void __user *p;
1178 if (get_user(p, pages + j + chunk_start))
1180 pm[j].addr = (unsigned long) p;
1182 if (get_user(node, nodes + j + chunk_start))
1186 if (node < 0 || node >= MAX_NUMNODES)
1189 if (!node_state(node, N_HIGH_MEMORY))
1193 if (!node_isset(node, task_nodes))
1199 /* End marker for this chunk */
1200 pm[chunk_nr_pages].node = MAX_NUMNODES;
1202 /* Migrate this chunk */
1203 err = do_move_page_to_node_array(mm, pm,
1204 flags & MPOL_MF_MOVE_ALL);
1208 /* Return status information */
1209 for (j = 0; j < chunk_nr_pages; j++)
1210 if (put_user(pm[j].status, status + j + chunk_start)) {
1218 free_page((unsigned long)pm);
1224 * Determine the nodes of an array of pages and store it in an array of status.
1226 static void do_pages_stat_array(struct mm_struct *mm, unsigned long nr_pages,
1227 const void __user **pages, int *status)
1231 down_read(&mm->mmap_sem);
1233 for (i = 0; i < nr_pages; i++) {
1234 unsigned long addr = (unsigned long)(*pages);
1235 struct vm_area_struct *vma;
1239 vma = find_vma(mm, addr);
1240 if (!vma || addr < vma->vm_start)
1243 page = follow_page(vma, addr, 0);
1245 err = PTR_ERR(page);
1250 /* Use PageReserved to check for zero page */
1251 if (!page || PageReserved(page) || PageKsm(page))
1254 err = page_to_nid(page);
1262 up_read(&mm->mmap_sem);
1266 * Determine the nodes of a user array of pages and store it in
1267 * a user array of status.
1269 static int do_pages_stat(struct mm_struct *mm, unsigned long nr_pages,
1270 const void __user * __user *pages,
1273 #define DO_PAGES_STAT_CHUNK_NR 16
1274 const void __user *chunk_pages[DO_PAGES_STAT_CHUNK_NR];
1275 int chunk_status[DO_PAGES_STAT_CHUNK_NR];
1278 unsigned long chunk_nr;
1280 chunk_nr = nr_pages;
1281 if (chunk_nr > DO_PAGES_STAT_CHUNK_NR)
1282 chunk_nr = DO_PAGES_STAT_CHUNK_NR;
1284 if (copy_from_user(chunk_pages, pages, chunk_nr * sizeof(*chunk_pages)))
1287 do_pages_stat_array(mm, chunk_nr, chunk_pages, chunk_status);
1289 if (copy_to_user(status, chunk_status, chunk_nr * sizeof(*status)))
1294 nr_pages -= chunk_nr;
1296 return nr_pages ? -EFAULT : 0;
1300 * Move a list of pages in the address space of the currently executing
1303 SYSCALL_DEFINE6(move_pages, pid_t, pid, unsigned long, nr_pages,
1304 const void __user * __user *, pages,
1305 const int __user *, nodes,
1306 int __user *, status, int, flags)
1308 const struct cred *cred = current_cred(), *tcred;
1309 struct task_struct *task;
1310 struct mm_struct *mm;
1314 if (flags & ~(MPOL_MF_MOVE|MPOL_MF_MOVE_ALL))
1317 if ((flags & MPOL_MF_MOVE_ALL) && !capable(CAP_SYS_NICE))
1320 /* Find the mm_struct */
1321 read_lock(&tasklist_lock);
1322 task = pid ? find_task_by_vpid(pid) : current;
1324 read_unlock(&tasklist_lock);
1327 mm = get_task_mm(task);
1328 read_unlock(&tasklist_lock);
1334 * Check if this process has the right to modify the specified
1335 * process. The right exists if the process has administrative
1336 * capabilities, superuser privileges or the same
1337 * userid as the target process.
1340 tcred = __task_cred(task);
1341 if (cred->euid != tcred->suid && cred->euid != tcred->uid &&
1342 cred->uid != tcred->suid && cred->uid != tcred->uid &&
1343 !capable(CAP_SYS_NICE)) {
1350 err = security_task_movememory(task);
1355 err = do_pages_move(mm, task, nr_pages, pages, nodes, status,
1358 err = do_pages_stat(mm, nr_pages, pages, status);
1367 * Call migration functions in the vma_ops that may prepare
1368 * memory in a vm for migration. migration functions may perform
1369 * the migration for vmas that do not have an underlying page struct.
1371 int migrate_vmas(struct mm_struct *mm, const nodemask_t *to,
1372 const nodemask_t *from, unsigned long flags)
1374 struct vm_area_struct *vma;
1377 for (vma = mm->mmap; vma && !err; vma = vma->vm_next) {
1378 if (vma->vm_ops && vma->vm_ops->migrate) {
1379 err = vma->vm_ops->migrate(vma, to, from, flags);