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);
143 pte = pte_mkhuge(pte);
144 flush_cache_page(vma, addr, pte_pfn(pte));
145 set_pte_at(mm, addr, ptep, pte);
149 hugepage_add_anon_rmap(new, vma, addr);
152 } else if (PageAnon(new))
153 page_add_anon_rmap(new, vma, addr);
155 page_add_file_rmap(new);
157 /* No need to invalidate - it was non-present before */
158 update_mmu_cache(vma, addr, ptep);
160 pte_unmap_unlock(ptep, ptl);
166 * Get rid of all migration entries and replace them by
167 * references to the indicated page.
169 static void remove_migration_ptes(struct page *old, struct page *new)
171 rmap_walk(new, remove_migration_pte, old);
175 * Something used the pte of a page under migration. We need to
176 * get to the page and wait until migration is finished.
177 * When we return from this function the fault will be retried.
179 * This function is called from do_swap_page().
181 void migration_entry_wait(struct mm_struct *mm, pmd_t *pmd,
182 unsigned long address)
189 ptep = pte_offset_map_lock(mm, pmd, address, &ptl);
191 if (!is_swap_pte(pte))
194 entry = pte_to_swp_entry(pte);
195 if (!is_migration_entry(entry))
198 page = migration_entry_to_page(entry);
201 * Once radix-tree replacement of page migration started, page_count
202 * *must* be zero. And, we don't want to call wait_on_page_locked()
203 * against a page without get_page().
204 * So, we use get_page_unless_zero(), here. Even failed, page fault
207 if (!get_page_unless_zero(page))
209 pte_unmap_unlock(ptep, ptl);
210 wait_on_page_locked(page);
214 pte_unmap_unlock(ptep, ptl);
218 * Replace the page in the mapping.
220 * The number of remaining references must be:
221 * 1 for anonymous pages without a mapping
222 * 2 for pages with a mapping
223 * 3 for pages with a mapping and PagePrivate/PagePrivate2 set.
225 static int migrate_page_move_mapping(struct address_space *mapping,
226 struct page *newpage, struct page *page)
232 /* Anonymous page without mapping */
233 if (page_count(page) != 1)
238 spin_lock_irq(&mapping->tree_lock);
240 pslot = radix_tree_lookup_slot(&mapping->page_tree,
243 expected_count = 2 + page_has_private(page);
244 if (page_count(page) != expected_count ||
245 (struct page *)radix_tree_deref_slot(pslot) != page) {
246 spin_unlock_irq(&mapping->tree_lock);
250 if (!page_freeze_refs(page, expected_count)) {
251 spin_unlock_irq(&mapping->tree_lock);
256 * Now we know that no one else is looking at the page.
258 get_page(newpage); /* add cache reference */
259 if (PageSwapCache(page)) {
260 SetPageSwapCache(newpage);
261 set_page_private(newpage, page_private(page));
264 radix_tree_replace_slot(pslot, newpage);
266 page_unfreeze_refs(page, expected_count);
268 * Drop cache reference from old page.
269 * We know this isn't the last reference.
274 * If moved to a different zone then also account
275 * the page for that zone. Other VM counters will be
276 * taken care of when we establish references to the
277 * new page and drop references to the old page.
279 * Note that anonymous pages are accounted for
280 * via NR_FILE_PAGES and NR_ANON_PAGES if they
281 * are mapped to swap space.
283 __dec_zone_page_state(page, NR_FILE_PAGES);
284 __inc_zone_page_state(newpage, NR_FILE_PAGES);
285 if (PageSwapBacked(page)) {
286 __dec_zone_page_state(page, NR_SHMEM);
287 __inc_zone_page_state(newpage, NR_SHMEM);
289 spin_unlock_irq(&mapping->tree_lock);
295 * The expected number of remaining references is the same as that
296 * of migrate_page_move_mapping().
298 int migrate_huge_page_move_mapping(struct address_space *mapping,
299 struct page *newpage, struct page *page)
305 if (page_count(page) != 1)
310 spin_lock_irq(&mapping->tree_lock);
312 pslot = radix_tree_lookup_slot(&mapping->page_tree,
315 expected_count = 2 + page_has_private(page);
316 if (page_count(page) != expected_count ||
317 (struct page *)radix_tree_deref_slot(pslot) != page) {
318 spin_unlock_irq(&mapping->tree_lock);
322 if (!page_freeze_refs(page, expected_count)) {
323 spin_unlock_irq(&mapping->tree_lock);
329 radix_tree_replace_slot(pslot, newpage);
331 page_unfreeze_refs(page, expected_count);
335 spin_unlock_irq(&mapping->tree_lock);
340 * Copy the page to its new location
342 void migrate_page_copy(struct page *newpage, struct page *page)
345 copy_huge_page(newpage, page);
347 copy_highpage(newpage, page);
350 SetPageError(newpage);
351 if (PageReferenced(page))
352 SetPageReferenced(newpage);
353 if (PageUptodate(page))
354 SetPageUptodate(newpage);
355 if (TestClearPageActive(page)) {
356 VM_BUG_ON(PageUnevictable(page));
357 SetPageActive(newpage);
358 } else if (TestClearPageUnevictable(page))
359 SetPageUnevictable(newpage);
360 if (PageChecked(page))
361 SetPageChecked(newpage);
362 if (PageMappedToDisk(page))
363 SetPageMappedToDisk(newpage);
365 if (PageDirty(page)) {
366 clear_page_dirty_for_io(page);
368 * Want to mark the page and the radix tree as dirty, and
369 * redo the accounting that clear_page_dirty_for_io undid,
370 * but we can't use set_page_dirty because that function
371 * is actually a signal that all of the page has become dirty.
372 * Wheras only part of our page may be dirty.
374 __set_page_dirty_nobuffers(newpage);
377 mlock_migrate_page(newpage, page);
378 ksm_migrate_page(newpage, page);
380 ClearPageSwapCache(page);
381 ClearPagePrivate(page);
382 set_page_private(page, 0);
383 page->mapping = NULL;
386 * If any waiters have accumulated on the new page then
389 if (PageWriteback(newpage))
390 end_page_writeback(newpage);
393 /************************************************************
394 * Migration functions
395 ***********************************************************/
397 /* Always fail migration. Used for mappings that are not movable */
398 int fail_migrate_page(struct address_space *mapping,
399 struct page *newpage, struct page *page)
403 EXPORT_SYMBOL(fail_migrate_page);
406 * Common logic to directly migrate a single page suitable for
407 * pages that do not use PagePrivate/PagePrivate2.
409 * Pages are locked upon entry and exit.
411 int migrate_page(struct address_space *mapping,
412 struct page *newpage, struct page *page)
416 BUG_ON(PageWriteback(page)); /* Writeback must be complete */
418 rc = migrate_page_move_mapping(mapping, newpage, page);
423 migrate_page_copy(newpage, page);
426 EXPORT_SYMBOL(migrate_page);
430 * Migration function for pages with buffers. This function can only be used
431 * if the underlying filesystem guarantees that no other references to "page"
434 int buffer_migrate_page(struct address_space *mapping,
435 struct page *newpage, struct page *page)
437 struct buffer_head *bh, *head;
440 if (!page_has_buffers(page))
441 return migrate_page(mapping, newpage, page);
443 head = page_buffers(page);
445 rc = migrate_page_move_mapping(mapping, newpage, page);
454 bh = bh->b_this_page;
456 } while (bh != head);
458 ClearPagePrivate(page);
459 set_page_private(newpage, page_private(page));
460 set_page_private(page, 0);
466 set_bh_page(bh, newpage, bh_offset(bh));
467 bh = bh->b_this_page;
469 } while (bh != head);
471 SetPagePrivate(newpage);
473 migrate_page_copy(newpage, page);
479 bh = bh->b_this_page;
481 } while (bh != head);
485 EXPORT_SYMBOL(buffer_migrate_page);
489 * Writeback a page to clean the dirty state
491 static int writeout(struct address_space *mapping, struct page *page)
493 struct writeback_control wbc = {
494 .sync_mode = WB_SYNC_NONE,
497 .range_end = LLONG_MAX,
503 if (!mapping->a_ops->writepage)
504 /* No write method for the address space */
507 if (!clear_page_dirty_for_io(page))
508 /* Someone else already triggered a write */
512 * A dirty page may imply that the underlying filesystem has
513 * the page on some queue. So the page must be clean for
514 * migration. Writeout may mean we loose the lock and the
515 * page state is no longer what we checked for earlier.
516 * At this point we know that the migration attempt cannot
519 remove_migration_ptes(page, page);
521 rc = mapping->a_ops->writepage(page, &wbc);
523 if (rc != AOP_WRITEPAGE_ACTIVATE)
524 /* unlocked. Relock */
527 return (rc < 0) ? -EIO : -EAGAIN;
531 * Default handling if a filesystem does not provide a migration function.
533 static int fallback_migrate_page(struct address_space *mapping,
534 struct page *newpage, struct page *page)
537 return writeout(mapping, page);
540 * Buffers may be managed in a filesystem specific way.
541 * We must have no buffers or drop them.
543 if (page_has_private(page) &&
544 !try_to_release_page(page, GFP_KERNEL))
547 return migrate_page(mapping, newpage, page);
551 * Move a page to a newly allocated page
552 * The page is locked and all ptes have been successfully removed.
554 * The new page will have replaced the old page if this function
561 static int move_to_new_page(struct page *newpage, struct page *page,
564 struct address_space *mapping;
568 * Block others from accessing the page when we get around to
569 * establishing additional references. We are the only one
570 * holding a reference to the new page at this point.
572 if (!trylock_page(newpage))
575 /* Prepare mapping for the new page.*/
576 newpage->index = page->index;
577 newpage->mapping = page->mapping;
578 if (PageSwapBacked(page))
579 SetPageSwapBacked(newpage);
581 mapping = page_mapping(page);
583 rc = migrate_page(mapping, newpage, page);
584 else if (mapping->a_ops->migratepage)
586 * Most pages have a mapping and most filesystems
587 * should provide a migration function. Anonymous
588 * pages are part of swap space which also has its
589 * own migration function. This is the most common
590 * path for page migration.
592 rc = mapping->a_ops->migratepage(mapping,
595 rc = fallback_migrate_page(mapping, newpage, page);
598 newpage->mapping = NULL;
601 remove_migration_ptes(page, newpage);
604 unlock_page(newpage);
610 * Obtain the lock on page, remove all ptes and migrate the page
611 * to the newly allocated page in newpage.
613 static int unmap_and_move(new_page_t get_new_page, unsigned long private,
614 struct page *page, int force, int offlining)
618 struct page *newpage = get_new_page(page, private, &result);
619 int remap_swapcache = 1;
622 struct mem_cgroup *mem = NULL;
623 struct anon_vma *anon_vma = NULL;
628 if (page_count(page) == 1) {
629 /* page was freed from under us. So we are done. */
633 /* prepare cgroup just returns 0 or -ENOMEM */
636 if (!trylock_page(page)) {
643 * Only memory hotplug's offline_pages() caller has locked out KSM,
644 * and can safely migrate a KSM page. The other cases have skipped
645 * PageKsm along with PageReserved - but it is only now when we have
646 * the page lock that we can be certain it will not go KSM beneath us
647 * (KSM will not upgrade a page from PageAnon to PageKsm when it sees
648 * its pagecount raised, but only here do we take the page lock which
651 if (PageKsm(page) && !offlining) {
656 /* charge against new page */
657 charge = mem_cgroup_prepare_migration(page, newpage, &mem);
658 if (charge == -ENOMEM) {
664 if (PageWriteback(page)) {
667 wait_on_page_writeback(page);
670 * By try_to_unmap(), page->mapcount goes down to 0 here. In this case,
671 * we cannot notice that anon_vma is freed while we migrates a page.
672 * This rcu_read_lock() delays freeing anon_vma pointer until the end
673 * of migration. File cache pages are no problem because of page_lock()
674 * File Caches may use write_page() or lock_page() in migration, then,
675 * just care Anon page here.
677 if (PageAnon(page)) {
681 /* Determine how to safely use anon_vma */
682 if (!page_mapped(page)) {
683 if (!PageSwapCache(page))
687 * We cannot be sure that the anon_vma of an unmapped
688 * swapcache page is safe to use because we don't
689 * know in advance if the VMA that this page belonged
690 * to still exists. If the VMA and others sharing the
691 * data have been freed, then the anon_vma could
692 * already be invalid.
694 * To avoid this possibility, swapcache pages get
695 * migrated but are not remapped when migration
701 * Take a reference count on the anon_vma if the
702 * page is mapped so that it is guaranteed to
703 * exist when the page is remapped later
705 anon_vma = page_anon_vma(page);
706 get_anon_vma(anon_vma);
711 * Corner case handling:
712 * 1. When a new swap-cache page is read into, it is added to the LRU
713 * and treated as swapcache but it has no rmap yet.
714 * Calling try_to_unmap() against a page->mapping==NULL page will
715 * trigger a BUG. So handle it here.
716 * 2. An orphaned page (see truncate_complete_page) might have
717 * fs-private metadata. The page can be picked up due to memory
718 * offlining. Everywhere else except page reclaim, the page is
719 * invisible to the vm, so the page can not be migrated. So try to
720 * free the metadata, so the page can be freed.
722 if (!page->mapping) {
723 if (!PageAnon(page) && page_has_private(page)) {
725 * Go direct to try_to_free_buffers() here because
726 * a) that's what try_to_release_page() would do anyway
727 * b) we may be under rcu_read_lock() here, so we can't
728 * use GFP_KERNEL which is what try_to_release_page()
729 * needs to be effective.
731 try_to_free_buffers(page);
737 /* Establish migration ptes or remove ptes */
738 try_to_unmap(page, TTU_MIGRATION|TTU_IGNORE_MLOCK|TTU_IGNORE_ACCESS);
741 if (!page_mapped(page))
742 rc = move_to_new_page(newpage, page, remap_swapcache);
744 if (rc && remap_swapcache)
745 remove_migration_ptes(page, page);
748 /* Drop an anon_vma reference if we took one */
750 drop_anon_vma(anon_vma);
756 mem_cgroup_end_migration(mem, page, newpage);
762 * A page that has been migrated has all references
763 * removed and will be freed. A page that has not been
764 * migrated will have kepts its references and be
767 list_del(&page->lru);
768 dec_zone_page_state(page, NR_ISOLATED_ANON +
769 page_is_file_cache(page));
770 putback_lru_page(page);
776 * Move the new page to the LRU. If migration was not successful
777 * then this will free the page.
779 putback_lru_page(newpage);
785 *result = page_to_nid(newpage);
791 * Counterpart of unmap_and_move_page() for hugepage migration.
793 * This function doesn't wait the completion of hugepage I/O
794 * because there is no race between I/O and migration for hugepage.
795 * Note that currently hugepage I/O occurs only in direct I/O
796 * where no lock is held and PG_writeback is irrelevant,
797 * and writeback status of all subpages are counted in the reference
798 * count of the head page (i.e. if all subpages of a 2MB hugepage are
799 * under direct I/O, the reference of the head page is 512 and a bit more.)
800 * This means that when we try to migrate hugepage whose subpages are
801 * doing direct I/O, some references remain after try_to_unmap() and
802 * hugepage migration fails without data corruption.
804 * There is also no race when direct I/O is issued on the page under migration,
805 * because then pte is replaced with migration swap entry and direct I/O code
806 * will wait in the page fault for migration to complete.
808 static int unmap_and_move_huge_page(new_page_t get_new_page,
809 unsigned long private, struct page *hpage,
810 int force, int offlining)
814 struct page *new_hpage = get_new_page(hpage, private, &result);
816 struct anon_vma *anon_vma = NULL;
823 if (!trylock_page(hpage)) {
829 if (PageAnon(hpage)) {
833 if (page_mapped(hpage)) {
834 anon_vma = page_anon_vma(hpage);
835 atomic_inc(&anon_vma->external_refcount);
839 try_to_unmap(hpage, TTU_MIGRATION|TTU_IGNORE_MLOCK|TTU_IGNORE_ACCESS);
841 if (!page_mapped(hpage))
842 rc = move_to_new_page(new_hpage, hpage, 1);
845 remove_migration_ptes(hpage, hpage);
847 if (anon_vma && atomic_dec_and_lock(&anon_vma->external_refcount,
849 int empty = list_empty(&anon_vma->head);
850 spin_unlock(&anon_vma->lock);
852 anon_vma_free(anon_vma);
861 list_del(&hpage->lru);
871 *result = page_to_nid(new_hpage);
879 * The function takes one list of pages to migrate and a function
880 * that determines from the page to be migrated and the private data
881 * the target of the move and allocates the page.
883 * The function returns after 10 attempts or if no pages
884 * are movable anymore because to has become empty
885 * or no retryable pages exist anymore. All pages will be
886 * returned to the LRU or freed.
888 * Return: Number of pages not migrated or error code.
890 int migrate_pages(struct list_head *from,
891 new_page_t get_new_page, unsigned long private, int offlining)
898 int swapwrite = current->flags & PF_SWAPWRITE;
902 current->flags |= PF_SWAPWRITE;
904 for(pass = 0; pass < 10 && retry; pass++) {
907 list_for_each_entry_safe(page, page2, from, lru) {
910 rc = unmap_and_move(get_new_page, private,
911 page, pass > 2, offlining);
922 /* Permanent failure */
931 current->flags &= ~PF_SWAPWRITE;
933 putback_lru_pages(from);
938 return nr_failed + retry;
941 int migrate_huge_pages(struct list_head *from,
942 new_page_t get_new_page, unsigned long private, int offlining)
951 for (pass = 0; pass < 10 && retry; pass++) {
954 list_for_each_entry_safe(page, page2, from, lru) {
957 rc = unmap_and_move_huge_page(get_new_page,
958 private, page, pass > 2, offlining);
969 /* Permanent failure */
978 list_for_each_entry_safe(page, page2, from, lru)
984 return nr_failed + retry;
989 * Move a list of individual pages
991 struct page_to_node {
998 static struct page *new_page_node(struct page *p, unsigned long private,
1001 struct page_to_node *pm = (struct page_to_node *)private;
1003 while (pm->node != MAX_NUMNODES && pm->page != p)
1006 if (pm->node == MAX_NUMNODES)
1009 *result = &pm->status;
1011 return alloc_pages_exact_node(pm->node,
1012 GFP_HIGHUSER_MOVABLE | GFP_THISNODE, 0);
1016 * Move a set of pages as indicated in the pm array. The addr
1017 * field must be set to the virtual address of the page to be moved
1018 * and the node number must contain a valid target node.
1019 * The pm array ends with node = MAX_NUMNODES.
1021 static int do_move_page_to_node_array(struct mm_struct *mm,
1022 struct page_to_node *pm,
1026 struct page_to_node *pp;
1027 LIST_HEAD(pagelist);
1029 down_read(&mm->mmap_sem);
1032 * Build a list of pages to migrate
1034 for (pp = pm; pp->node != MAX_NUMNODES; pp++) {
1035 struct vm_area_struct *vma;
1039 vma = find_vma(mm, pp->addr);
1040 if (!vma || !vma_migratable(vma))
1043 page = follow_page(vma, pp->addr, FOLL_GET);
1045 err = PTR_ERR(page);
1053 /* Use PageReserved to check for zero page */
1054 if (PageReserved(page) || PageKsm(page))
1058 err = page_to_nid(page);
1060 if (err == pp->node)
1062 * Node already in the right place
1067 if (page_mapcount(page) > 1 &&
1071 err = isolate_lru_page(page);
1073 list_add_tail(&page->lru, &pagelist);
1074 inc_zone_page_state(page, NR_ISOLATED_ANON +
1075 page_is_file_cache(page));
1079 * Either remove the duplicate refcount from
1080 * isolate_lru_page() or drop the page ref if it was
1089 if (!list_empty(&pagelist))
1090 err = migrate_pages(&pagelist, new_page_node,
1091 (unsigned long)pm, 0);
1093 up_read(&mm->mmap_sem);
1098 * Migrate an array of page address onto an array of nodes and fill
1099 * the corresponding array of status.
1101 static int do_pages_move(struct mm_struct *mm, struct task_struct *task,
1102 unsigned long nr_pages,
1103 const void __user * __user *pages,
1104 const int __user *nodes,
1105 int __user *status, int flags)
1107 struct page_to_node *pm;
1108 nodemask_t task_nodes;
1109 unsigned long chunk_nr_pages;
1110 unsigned long chunk_start;
1113 task_nodes = cpuset_mems_allowed(task);
1116 pm = (struct page_to_node *)__get_free_page(GFP_KERNEL);
1123 * Store a chunk of page_to_node array in a page,
1124 * but keep the last one as a marker
1126 chunk_nr_pages = (PAGE_SIZE / sizeof(struct page_to_node)) - 1;
1128 for (chunk_start = 0;
1129 chunk_start < nr_pages;
1130 chunk_start += chunk_nr_pages) {
1133 if (chunk_start + chunk_nr_pages > nr_pages)
1134 chunk_nr_pages = nr_pages - chunk_start;
1136 /* fill the chunk pm with addrs and nodes from user-space */
1137 for (j = 0; j < chunk_nr_pages; j++) {
1138 const void __user *p;
1142 if (get_user(p, pages + j + chunk_start))
1144 pm[j].addr = (unsigned long) p;
1146 if (get_user(node, nodes + j + chunk_start))
1150 if (node < 0 || node >= MAX_NUMNODES)
1153 if (!node_state(node, N_HIGH_MEMORY))
1157 if (!node_isset(node, task_nodes))
1163 /* End marker for this chunk */
1164 pm[chunk_nr_pages].node = MAX_NUMNODES;
1166 /* Migrate this chunk */
1167 err = do_move_page_to_node_array(mm, pm,
1168 flags & MPOL_MF_MOVE_ALL);
1172 /* Return status information */
1173 for (j = 0; j < chunk_nr_pages; j++)
1174 if (put_user(pm[j].status, status + j + chunk_start)) {
1182 free_page((unsigned long)pm);
1188 * Determine the nodes of an array of pages and store it in an array of status.
1190 static void do_pages_stat_array(struct mm_struct *mm, unsigned long nr_pages,
1191 const void __user **pages, int *status)
1195 down_read(&mm->mmap_sem);
1197 for (i = 0; i < nr_pages; i++) {
1198 unsigned long addr = (unsigned long)(*pages);
1199 struct vm_area_struct *vma;
1203 vma = find_vma(mm, addr);
1207 page = follow_page(vma, addr, 0);
1209 err = PTR_ERR(page);
1214 /* Use PageReserved to check for zero page */
1215 if (!page || PageReserved(page) || PageKsm(page))
1218 err = page_to_nid(page);
1226 up_read(&mm->mmap_sem);
1230 * Determine the nodes of a user array of pages and store it in
1231 * a user array of status.
1233 static int do_pages_stat(struct mm_struct *mm, unsigned long nr_pages,
1234 const void __user * __user *pages,
1237 #define DO_PAGES_STAT_CHUNK_NR 16
1238 const void __user *chunk_pages[DO_PAGES_STAT_CHUNK_NR];
1239 int chunk_status[DO_PAGES_STAT_CHUNK_NR];
1242 unsigned long chunk_nr;
1244 chunk_nr = nr_pages;
1245 if (chunk_nr > DO_PAGES_STAT_CHUNK_NR)
1246 chunk_nr = DO_PAGES_STAT_CHUNK_NR;
1248 if (copy_from_user(chunk_pages, pages, chunk_nr * sizeof(*chunk_pages)))
1251 do_pages_stat_array(mm, chunk_nr, chunk_pages, chunk_status);
1253 if (copy_to_user(status, chunk_status, chunk_nr * sizeof(*status)))
1258 nr_pages -= chunk_nr;
1260 return nr_pages ? -EFAULT : 0;
1264 * Move a list of pages in the address space of the currently executing
1267 SYSCALL_DEFINE6(move_pages, pid_t, pid, unsigned long, nr_pages,
1268 const void __user * __user *, pages,
1269 const int __user *, nodes,
1270 int __user *, status, int, flags)
1272 const struct cred *cred = current_cred(), *tcred;
1273 struct task_struct *task;
1274 struct mm_struct *mm;
1278 if (flags & ~(MPOL_MF_MOVE|MPOL_MF_MOVE_ALL))
1281 if ((flags & MPOL_MF_MOVE_ALL) && !capable(CAP_SYS_NICE))
1284 /* Find the mm_struct */
1285 read_lock(&tasklist_lock);
1286 task = pid ? find_task_by_vpid(pid) : current;
1288 read_unlock(&tasklist_lock);
1291 mm = get_task_mm(task);
1292 read_unlock(&tasklist_lock);
1298 * Check if this process has the right to modify the specified
1299 * process. The right exists if the process has administrative
1300 * capabilities, superuser privileges or the same
1301 * userid as the target process.
1304 tcred = __task_cred(task);
1305 if (cred->euid != tcred->suid && cred->euid != tcred->uid &&
1306 cred->uid != tcred->suid && cred->uid != tcred->uid &&
1307 !capable(CAP_SYS_NICE)) {
1314 err = security_task_movememory(task);
1319 err = do_pages_move(mm, task, nr_pages, pages, nodes, status,
1322 err = do_pages_stat(mm, nr_pages, pages, status);
1331 * Call migration functions in the vma_ops that may prepare
1332 * memory in a vm for migration. migration functions may perform
1333 * the migration for vmas that do not have an underlying page struct.
1335 int migrate_vmas(struct mm_struct *mm, const nodemask_t *to,
1336 const nodemask_t *from, unsigned long flags)
1338 struct vm_area_struct *vma;
1341 for (vma = mm->mmap; vma && !err; vma = vma->vm_next) {
1342 if (vma->vm_ops && vma->vm_ops->migrate) {
1343 err = vma->vm_ops->migrate(vma, to, from, flags);