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/export.h>
17 #include <linux/swap.h>
18 #include <linux/swapops.h>
19 #include <linux/pagemap.h>
20 #include <linux/buffer_head.h>
21 #include <linux/mm_inline.h>
22 #include <linux/nsproxy.h>
23 #include <linux/pagevec.h>
24 #include <linux/ksm.h>
25 #include <linux/rmap.h>
26 #include <linux/topology.h>
27 #include <linux/cpu.h>
28 #include <linux/cpuset.h>
29 #include <linux/writeback.h>
30 #include <linux/mempolicy.h>
31 #include <linux/vmalloc.h>
32 #include <linux/security.h>
33 #include <linux/memcontrol.h>
34 #include <linux/syscalls.h>
35 #include <linux/hugetlb.h>
36 #include <linux/gfp.h>
38 #include <asm/tlbflush.h>
42 #define lru_to_page(_head) (list_entry((_head)->prev, struct page, lru))
45 * migrate_prep() needs to be called before we start compiling a list of pages
46 * to be migrated using isolate_lru_page(). If scheduling work on other CPUs is
47 * undesirable, use migrate_prep_local()
49 int migrate_prep(void)
52 * Clear the LRU lists so pages can be isolated.
53 * Note that pages may be moved off the LRU after we have
54 * drained them. Those pages will fail to migrate like other
55 * pages that may be busy.
62 /* Do the necessary work of migrate_prep but not if it involves other CPUs */
63 int migrate_prep_local(void)
71 * Add isolated pages on the list back to the LRU under page lock
72 * to avoid leaking evictable pages back onto unevictable list.
74 void putback_lru_pages(struct list_head *l)
79 list_for_each_entry_safe(page, page2, l, lru) {
81 dec_zone_page_state(page, NR_ISOLATED_ANON +
82 page_is_file_cache(page));
83 putback_lru_page(page);
88 * Restore a potential migration pte to a working pte entry
90 static int remove_migration_pte(struct page *new, struct vm_area_struct *vma,
91 unsigned long addr, void *old)
93 struct mm_struct *mm = vma->vm_mm;
101 if (unlikely(PageHuge(new))) {
102 ptep = huge_pte_offset(mm, addr);
105 ptl = &mm->page_table_lock;
107 pgd = pgd_offset(mm, addr);
108 if (!pgd_present(*pgd))
111 pud = pud_offset(pgd, addr);
112 if (!pud_present(*pud))
115 pmd = pmd_offset(pud, addr);
116 if (pmd_trans_huge(*pmd))
118 if (!pmd_present(*pmd))
121 ptep = pte_offset_map(pmd, addr);
124 * Peek to check is_swap_pte() before taking ptlock? No, we
125 * can race mremap's move_ptes(), which skips anon_vma lock.
128 ptl = pte_lockptr(mm, pmd);
133 if (!is_swap_pte(pte))
136 entry = pte_to_swp_entry(pte);
138 if (!is_migration_entry(entry) ||
139 migration_entry_to_page(entry) != old)
143 pte = pte_mkold(mk_pte(new, vma->vm_page_prot));
144 if (is_write_migration_entry(entry))
145 pte = pte_mkwrite(pte);
146 #ifdef CONFIG_HUGETLB_PAGE
148 pte = pte_mkhuge(pte);
150 flush_cache_page(vma, addr, pte_pfn(pte));
151 set_pte_at(mm, addr, ptep, pte);
155 hugepage_add_anon_rmap(new, vma, addr);
158 } else if (PageAnon(new))
159 page_add_anon_rmap(new, vma, addr);
161 page_add_file_rmap(new);
163 /* No need to invalidate - it was non-present before */
164 update_mmu_cache(vma, addr, ptep);
166 pte_unmap_unlock(ptep, ptl);
172 * Get rid of all migration entries and replace them by
173 * references to the indicated page.
175 static void remove_migration_ptes(struct page *old, struct page *new)
177 rmap_walk(new, remove_migration_pte, old);
181 * Something used the pte of a page under migration. We need to
182 * get to the page and wait until migration is finished.
183 * When we return from this function the fault will be retried.
185 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);
271 * Drop cache reference from old page by unfreezing
272 * to one less reference.
273 * We know this isn't the last reference.
275 page_unfreeze_refs(page, expected_count - 1);
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 (!PageSwapCache(page) && 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 - 1);
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 * Whereas 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,
504 if (!mapping->a_ops->writepage)
505 /* No write method for the address space */
508 if (!clear_page_dirty_for_io(page))
509 /* Someone else already triggered a write */
513 * A dirty page may imply that the underlying filesystem has
514 * the page on some queue. So the page must be clean for
515 * migration. Writeout may mean we loose the lock and the
516 * page state is no longer what we checked for earlier.
517 * At this point we know that the migration attempt cannot
520 remove_migration_ptes(page, page);
522 rc = mapping->a_ops->writepage(page, &wbc);
524 if (rc != AOP_WRITEPAGE_ACTIVATE)
525 /* unlocked. Relock */
528 return (rc < 0) ? -EIO : -EAGAIN;
532 * Default handling if a filesystem does not provide a migration function.
534 static int fallback_migrate_page(struct address_space *mapping,
535 struct page *newpage, struct page *page)
538 return writeout(mapping, page);
541 * Buffers may be managed in a filesystem specific way.
542 * We must have no buffers or drop them.
544 if (page_has_private(page) &&
545 !try_to_release_page(page, GFP_KERNEL))
548 return migrate_page(mapping, newpage, page);
552 * Move a page to a newly allocated page
553 * The page is locked and all ptes have been successfully removed.
555 * The new page will have replaced the old page if this function
562 static int move_to_new_page(struct page *newpage, struct page *page,
563 int remap_swapcache, bool sync)
565 struct address_space *mapping;
569 * Block others from accessing the page when we get around to
570 * establishing additional references. We are the only one
571 * holding a reference to the new page at this point.
573 if (!trylock_page(newpage))
576 /* Prepare mapping for the new page.*/
577 newpage->index = page->index;
578 newpage->mapping = page->mapping;
579 if (PageSwapBacked(page))
580 SetPageSwapBacked(newpage);
582 mapping = page_mapping(page);
584 rc = migrate_page(mapping, newpage, page);
587 * Do not writeback pages if !sync and migratepage is
588 * not pointing to migrate_page() which is nonblocking
589 * (swapcache/tmpfs uses migratepage = migrate_page).
591 if (PageDirty(page) && !sync &&
592 mapping->a_ops->migratepage != migrate_page)
594 else if (mapping->a_ops->migratepage)
596 * Most pages have a mapping and most filesystems
597 * should provide a migration function. Anonymous
598 * pages are part of swap space which also has its
599 * own migration function. This is the most common
600 * path for page migration.
602 rc = mapping->a_ops->migratepage(mapping,
605 rc = fallback_migrate_page(mapping, newpage, page);
609 newpage->mapping = NULL;
612 remove_migration_ptes(page, newpage);
615 unlock_page(newpage);
620 static int __unmap_and_move(struct page *page, struct page *newpage,
621 int force, bool offlining, bool sync)
624 int remap_swapcache = 1;
626 struct mem_cgroup *mem;
627 struct anon_vma *anon_vma = NULL;
629 if (!trylock_page(page)) {
634 * It's not safe for direct compaction to call lock_page.
635 * For example, during page readahead pages are added locked
636 * to the LRU. Later, when the IO completes the pages are
637 * marked uptodate and unlocked. However, the queueing
638 * could be merging multiple pages for one bio (e.g.
639 * mpage_readpages). If an allocation happens for the
640 * second or third page, the process can end up locking
641 * the same page twice and deadlocking. Rather than
642 * trying to be clever about what pages can be locked,
643 * avoid the use of lock_page for direct compaction
646 if (current->flags & PF_MEMALLOC)
653 * Only memory hotplug's offline_pages() caller has locked out KSM,
654 * and can safely migrate a KSM page. The other cases have skipped
655 * PageKsm along with PageReserved - but it is only now when we have
656 * the page lock that we can be certain it will not go KSM beneath us
657 * (KSM will not upgrade a page from PageAnon to PageKsm when it sees
658 * its pagecount raised, but only here do we take the page lock which
661 if (PageKsm(page) && !offlining) {
666 /* charge against new page */
667 charge = mem_cgroup_prepare_migration(page, newpage, &mem, GFP_KERNEL);
668 if (charge == -ENOMEM) {
674 if (PageWriteback(page)) {
676 * For !sync, there is no point retrying as the retry loop
677 * is expected to be too short for PageWriteback to be cleared
685 wait_on_page_writeback(page);
688 * By try_to_unmap(), page->mapcount goes down to 0 here. In this case,
689 * we cannot notice that anon_vma is freed while we migrates a page.
690 * This get_anon_vma() delays freeing anon_vma pointer until the end
691 * of migration. File cache pages are no problem because of page_lock()
692 * File Caches may use write_page() or lock_page() in migration, then,
693 * just care Anon page here.
695 if (PageAnon(page)) {
697 * Only page_lock_anon_vma() understands the subtleties of
698 * getting a hold on an anon_vma from outside one of its mms.
700 anon_vma = page_get_anon_vma(page);
705 } else if (PageSwapCache(page)) {
707 * We cannot be sure that the anon_vma of an unmapped
708 * swapcache page is safe to use because we don't
709 * know in advance if the VMA that this page belonged
710 * to still exists. If the VMA and others sharing the
711 * data have been freed, then the anon_vma could
712 * already be invalid.
714 * To avoid this possibility, swapcache pages get
715 * migrated but are not remapped when migration
725 * Corner case handling:
726 * 1. When a new swap-cache page is read into, it is added to the LRU
727 * and treated as swapcache but it has no rmap yet.
728 * Calling try_to_unmap() against a page->mapping==NULL page will
729 * trigger a BUG. So handle it here.
730 * 2. An orphaned page (see truncate_complete_page) might have
731 * fs-private metadata. The page can be picked up due to memory
732 * offlining. Everywhere else except page reclaim, the page is
733 * invisible to the vm, so the page can not be migrated. So try to
734 * free the metadata, so the page can be freed.
736 if (!page->mapping) {
737 VM_BUG_ON(PageAnon(page));
738 if (page_has_private(page)) {
739 try_to_free_buffers(page);
745 /* Establish migration ptes or remove ptes */
746 try_to_unmap(page, TTU_MIGRATION|TTU_IGNORE_MLOCK|TTU_IGNORE_ACCESS);
749 if (!page_mapped(page))
750 rc = move_to_new_page(newpage, page, remap_swapcache, sync);
752 if (rc && remap_swapcache)
753 remove_migration_ptes(page, page);
755 /* Drop an anon_vma reference if we took one */
757 put_anon_vma(anon_vma);
761 mem_cgroup_end_migration(mem, page, newpage, rc == 0);
769 * Obtain the lock on page, remove all ptes and migrate the page
770 * to the newly allocated page in newpage.
772 static int unmap_and_move(new_page_t get_new_page, unsigned long private,
773 struct page *page, int force, bool offlining, bool sync)
777 struct page *newpage = get_new_page(page, private, &result);
782 if (page_count(page) == 1) {
783 /* page was freed from under us. So we are done. */
787 if (unlikely(PageTransHuge(page)))
788 if (unlikely(split_huge_page(page)))
791 rc = __unmap_and_move(page, newpage, force, offlining, sync);
795 * A page that has been migrated has all references
796 * removed and will be freed. A page that has not been
797 * migrated will have kepts its references and be
800 list_del(&page->lru);
801 dec_zone_page_state(page, NR_ISOLATED_ANON +
802 page_is_file_cache(page));
803 putback_lru_page(page);
806 * Move the new page to the LRU. If migration was not successful
807 * then this will free the page.
809 putback_lru_page(newpage);
814 *result = page_to_nid(newpage);
820 * Counterpart of unmap_and_move_page() for hugepage migration.
822 * This function doesn't wait the completion of hugepage I/O
823 * because there is no race between I/O and migration for hugepage.
824 * Note that currently hugepage I/O occurs only in direct I/O
825 * where no lock is held and PG_writeback is irrelevant,
826 * and writeback status of all subpages are counted in the reference
827 * count of the head page (i.e. if all subpages of a 2MB hugepage are
828 * under direct I/O, the reference of the head page is 512 and a bit more.)
829 * This means that when we try to migrate hugepage whose subpages are
830 * doing direct I/O, some references remain after try_to_unmap() and
831 * hugepage migration fails without data corruption.
833 * There is also no race when direct I/O is issued on the page under migration,
834 * because then pte is replaced with migration swap entry and direct I/O code
835 * will wait in the page fault for migration to complete.
837 static int unmap_and_move_huge_page(new_page_t get_new_page,
838 unsigned long private, struct page *hpage,
839 int force, bool offlining, bool sync)
843 struct page *new_hpage = get_new_page(hpage, private, &result);
844 struct anon_vma *anon_vma = NULL;
851 if (!trylock_page(hpage)) {
858 anon_vma = page_get_anon_vma(hpage);
860 try_to_unmap(hpage, TTU_MIGRATION|TTU_IGNORE_MLOCK|TTU_IGNORE_ACCESS);
862 if (!page_mapped(hpage))
863 rc = move_to_new_page(new_hpage, hpage, 1, sync);
866 remove_migration_ptes(hpage, hpage);
869 put_anon_vma(anon_vma);
874 list_del(&hpage->lru);
884 *result = page_to_nid(new_hpage);
892 * The function takes one list of pages to migrate and a function
893 * that determines from the page to be migrated and the private data
894 * the target of the move and allocates the page.
896 * The function returns after 10 attempts or if no pages
897 * are movable anymore because to has become empty
898 * or no retryable pages exist anymore.
899 * Caller should call putback_lru_pages to return pages to the LRU
900 * or free list only if ret != 0.
902 * Return: Number of pages not migrated or error code.
904 int migrate_pages(struct list_head *from,
905 new_page_t get_new_page, unsigned long private, bool offlining,
913 int swapwrite = current->flags & PF_SWAPWRITE;
917 current->flags |= PF_SWAPWRITE;
919 for(pass = 0; pass < 10 && retry; pass++) {
922 list_for_each_entry_safe(page, page2, from, lru) {
925 rc = unmap_and_move(get_new_page, private,
926 page, pass > 2, offlining,
938 /* Permanent failure */
947 current->flags &= ~PF_SWAPWRITE;
952 return nr_failed + retry;
955 int migrate_huge_pages(struct list_head *from,
956 new_page_t get_new_page, unsigned long private, bool offlining,
966 for (pass = 0; pass < 10 && retry; pass++) {
969 list_for_each_entry_safe(page, page2, from, lru) {
972 rc = unmap_and_move_huge_page(get_new_page,
973 private, page, pass > 2, offlining,
985 /* Permanent failure */
996 return nr_failed + retry;
1001 * Move a list of individual pages
1003 struct page_to_node {
1010 static struct page *new_page_node(struct page *p, unsigned long private,
1013 struct page_to_node *pm = (struct page_to_node *)private;
1015 while (pm->node != MAX_NUMNODES && pm->page != p)
1018 if (pm->node == MAX_NUMNODES)
1021 *result = &pm->status;
1023 return alloc_pages_exact_node(pm->node,
1024 GFP_HIGHUSER_MOVABLE | GFP_THISNODE, 0);
1028 * Move a set of pages as indicated in the pm array. The addr
1029 * field must be set to the virtual address of the page to be moved
1030 * and the node number must contain a valid target node.
1031 * The pm array ends with node = MAX_NUMNODES.
1033 static int do_move_page_to_node_array(struct mm_struct *mm,
1034 struct page_to_node *pm,
1038 struct page_to_node *pp;
1039 LIST_HEAD(pagelist);
1041 down_read(&mm->mmap_sem);
1044 * Build a list of pages to migrate
1046 for (pp = pm; pp->node != MAX_NUMNODES; pp++) {
1047 struct vm_area_struct *vma;
1051 vma = find_vma(mm, pp->addr);
1052 if (!vma || pp->addr < vma->vm_start || !vma_migratable(vma))
1055 page = follow_page(vma, pp->addr, FOLL_GET|FOLL_SPLIT);
1057 err = PTR_ERR(page);
1065 /* Use PageReserved to check for zero page */
1066 if (PageReserved(page) || PageKsm(page))
1070 err = page_to_nid(page);
1072 if (err == pp->node)
1074 * Node already in the right place
1079 if (page_mapcount(page) > 1 &&
1083 err = isolate_lru_page(page);
1085 list_add_tail(&page->lru, &pagelist);
1086 inc_zone_page_state(page, NR_ISOLATED_ANON +
1087 page_is_file_cache(page));
1091 * Either remove the duplicate refcount from
1092 * isolate_lru_page() or drop the page ref if it was
1101 if (!list_empty(&pagelist)) {
1102 err = migrate_pages(&pagelist, new_page_node,
1103 (unsigned long)pm, 0, true);
1105 putback_lru_pages(&pagelist);
1108 up_read(&mm->mmap_sem);
1113 * Migrate an array of page address onto an array of nodes and fill
1114 * the corresponding array of status.
1116 static int do_pages_move(struct mm_struct *mm, struct task_struct *task,
1117 unsigned long nr_pages,
1118 const void __user * __user *pages,
1119 const int __user *nodes,
1120 int __user *status, int flags)
1122 struct page_to_node *pm;
1123 nodemask_t task_nodes;
1124 unsigned long chunk_nr_pages;
1125 unsigned long chunk_start;
1128 task_nodes = cpuset_mems_allowed(task);
1131 pm = (struct page_to_node *)__get_free_page(GFP_KERNEL);
1138 * Store a chunk of page_to_node array in a page,
1139 * but keep the last one as a marker
1141 chunk_nr_pages = (PAGE_SIZE / sizeof(struct page_to_node)) - 1;
1143 for (chunk_start = 0;
1144 chunk_start < nr_pages;
1145 chunk_start += chunk_nr_pages) {
1148 if (chunk_start + chunk_nr_pages > nr_pages)
1149 chunk_nr_pages = nr_pages - chunk_start;
1151 /* fill the chunk pm with addrs and nodes from user-space */
1152 for (j = 0; j < chunk_nr_pages; j++) {
1153 const void __user *p;
1157 if (get_user(p, pages + j + chunk_start))
1159 pm[j].addr = (unsigned long) p;
1161 if (get_user(node, nodes + j + chunk_start))
1165 if (node < 0 || node >= MAX_NUMNODES)
1168 if (!node_state(node, N_HIGH_MEMORY))
1172 if (!node_isset(node, task_nodes))
1178 /* End marker for this chunk */
1179 pm[chunk_nr_pages].node = MAX_NUMNODES;
1181 /* Migrate this chunk */
1182 err = do_move_page_to_node_array(mm, pm,
1183 flags & MPOL_MF_MOVE_ALL);
1187 /* Return status information */
1188 for (j = 0; j < chunk_nr_pages; j++)
1189 if (put_user(pm[j].status, status + j + chunk_start)) {
1197 free_page((unsigned long)pm);
1203 * Determine the nodes of an array of pages and store it in an array of status.
1205 static void do_pages_stat_array(struct mm_struct *mm, unsigned long nr_pages,
1206 const void __user **pages, int *status)
1210 down_read(&mm->mmap_sem);
1212 for (i = 0; i < nr_pages; i++) {
1213 unsigned long addr = (unsigned long)(*pages);
1214 struct vm_area_struct *vma;
1218 vma = find_vma(mm, addr);
1219 if (!vma || addr < vma->vm_start)
1222 page = follow_page(vma, addr, 0);
1224 err = PTR_ERR(page);
1229 /* Use PageReserved to check for zero page */
1230 if (!page || PageReserved(page) || PageKsm(page))
1233 err = page_to_nid(page);
1241 up_read(&mm->mmap_sem);
1245 * Determine the nodes of a user array of pages and store it in
1246 * a user array of status.
1248 static int do_pages_stat(struct mm_struct *mm, unsigned long nr_pages,
1249 const void __user * __user *pages,
1252 #define DO_PAGES_STAT_CHUNK_NR 16
1253 const void __user *chunk_pages[DO_PAGES_STAT_CHUNK_NR];
1254 int chunk_status[DO_PAGES_STAT_CHUNK_NR];
1257 unsigned long chunk_nr;
1259 chunk_nr = nr_pages;
1260 if (chunk_nr > DO_PAGES_STAT_CHUNK_NR)
1261 chunk_nr = DO_PAGES_STAT_CHUNK_NR;
1263 if (copy_from_user(chunk_pages, pages, chunk_nr * sizeof(*chunk_pages)))
1266 do_pages_stat_array(mm, chunk_nr, chunk_pages, chunk_status);
1268 if (copy_to_user(status, chunk_status, chunk_nr * sizeof(*status)))
1273 nr_pages -= chunk_nr;
1275 return nr_pages ? -EFAULT : 0;
1279 * Move a list of pages in the address space of the currently executing
1282 SYSCALL_DEFINE6(move_pages, pid_t, pid, unsigned long, nr_pages,
1283 const void __user * __user *, pages,
1284 const int __user *, nodes,
1285 int __user *, status, int, flags)
1287 const struct cred *cred = current_cred(), *tcred;
1288 struct task_struct *task;
1289 struct mm_struct *mm;
1293 if (flags & ~(MPOL_MF_MOVE|MPOL_MF_MOVE_ALL))
1296 if ((flags & MPOL_MF_MOVE_ALL) && !capable(CAP_SYS_NICE))
1299 /* Find the mm_struct */
1301 task = pid ? find_task_by_vpid(pid) : current;
1306 mm = get_task_mm(task);
1313 * Check if this process has the right to modify the specified
1314 * process. The right exists if the process has administrative
1315 * capabilities, superuser privileges or the same
1316 * userid as the target process.
1319 tcred = __task_cred(task);
1320 if (cred->euid != tcred->suid && cred->euid != tcred->uid &&
1321 cred->uid != tcred->suid && cred->uid != tcred->uid &&
1322 !capable(CAP_SYS_NICE)) {
1329 err = security_task_movememory(task);
1334 err = do_pages_move(mm, task, nr_pages, pages, nodes, status,
1337 err = do_pages_stat(mm, nr_pages, pages, status);
1346 * Call migration functions in the vma_ops that may prepare
1347 * memory in a vm for migration. migration functions may perform
1348 * the migration for vmas that do not have an underlying page struct.
1350 int migrate_vmas(struct mm_struct *mm, const nodemask_t *to,
1351 const nodemask_t *from, unsigned long flags)
1353 struct vm_area_struct *vma;
1356 for (vma = mm->mmap; vma && !err; vma = vma->vm_next) {
1357 if (vma->vm_ops && vma->vm_ops->migrate) {
1358 err = vma->vm_ops->migrate(vma, to, from, flags);