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/hugetlb_cgroup.h>
37 #include <linux/gfp.h>
39 #include <asm/tlbflush.h>
44 * migrate_prep() needs to be called before we start compiling a list of pages
45 * to be migrated using isolate_lru_page(). If scheduling work on other CPUs is
46 * undesirable, use migrate_prep_local()
48 int migrate_prep(void)
51 * Clear the LRU lists so pages can be isolated.
52 * Note that pages may be moved off the LRU after we have
53 * drained them. Those pages will fail to migrate like other
54 * pages that may be busy.
61 /* Do the necessary work of migrate_prep but not if it involves other CPUs */
62 int migrate_prep_local(void)
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);
87 * Restore a potential migration pte to a working pte entry
89 static int remove_migration_pte(struct page *new, struct vm_area_struct *vma,
90 unsigned long addr, void *old)
92 struct mm_struct *mm = vma->vm_mm;
98 if (unlikely(PageHuge(new))) {
99 ptep = huge_pte_offset(mm, addr);
102 ptl = &mm->page_table_lock;
104 pmd = mm_find_pmd(mm, addr);
107 if (pmd_trans_huge(*pmd))
110 ptep = pte_offset_map(pmd, addr);
113 * Peek to check is_swap_pte() before taking ptlock? No, we
114 * can race mremap's move_ptes(), which skips anon_vma lock.
117 ptl = pte_lockptr(mm, pmd);
122 if (!is_swap_pte(pte))
125 entry = pte_to_swp_entry(pte);
127 if (!is_migration_entry(entry) ||
128 migration_entry_to_page(entry) != old)
132 pte = pte_mkold(mk_pte(new, vma->vm_page_prot));
133 if (is_write_migration_entry(entry))
134 pte = pte_mkwrite(pte);
135 #ifdef CONFIG_HUGETLB_PAGE
137 pte = pte_mkhuge(pte);
139 flush_cache_page(vma, addr, pte_pfn(pte));
140 set_pte_at(mm, addr, ptep, pte);
144 hugepage_add_anon_rmap(new, vma, addr);
147 } else if (PageAnon(new))
148 page_add_anon_rmap(new, vma, addr);
150 page_add_file_rmap(new);
152 /* No need to invalidate - it was non-present before */
153 update_mmu_cache(vma, addr, ptep);
155 pte_unmap_unlock(ptep, ptl);
161 * Get rid of all migration entries and replace them by
162 * references to the indicated page.
164 static void remove_migration_ptes(struct page *old, struct page *new)
166 rmap_walk(new, remove_migration_pte, old);
170 * Something used the pte of a page under migration. We need to
171 * get to the page and wait until migration is finished.
172 * When we return from this function the fault will be retried.
174 void migration_entry_wait(struct mm_struct *mm, pmd_t *pmd,
175 unsigned long address)
182 ptep = pte_offset_map_lock(mm, pmd, address, &ptl);
184 if (!is_swap_pte(pte))
187 entry = pte_to_swp_entry(pte);
188 if (!is_migration_entry(entry))
191 page = migration_entry_to_page(entry);
194 * Once radix-tree replacement of page migration started, page_count
195 * *must* be zero. And, we don't want to call wait_on_page_locked()
196 * against a page without get_page().
197 * So, we use get_page_unless_zero(), here. Even failed, page fault
200 if (!get_page_unless_zero(page))
202 pte_unmap_unlock(ptep, ptl);
203 wait_on_page_locked(page);
207 pte_unmap_unlock(ptep, ptl);
211 /* Returns true if all buffers are successfully locked */
212 static bool buffer_migrate_lock_buffers(struct buffer_head *head,
213 enum migrate_mode mode)
215 struct buffer_head *bh = head;
217 /* Simple case, sync compaction */
218 if (mode != MIGRATE_ASYNC) {
222 bh = bh->b_this_page;
224 } while (bh != head);
229 /* async case, we cannot block on lock_buffer so use trylock_buffer */
232 if (!trylock_buffer(bh)) {
234 * We failed to lock the buffer and cannot stall in
235 * async migration. Release the taken locks
237 struct buffer_head *failed_bh = bh;
240 while (bh != failed_bh) {
243 bh = bh->b_this_page;
248 bh = bh->b_this_page;
249 } while (bh != head);
253 static inline bool buffer_migrate_lock_buffers(struct buffer_head *head,
254 enum migrate_mode mode)
258 #endif /* CONFIG_BLOCK */
261 * Replace the page in the mapping.
263 * The number of remaining references must be:
264 * 1 for anonymous pages without a mapping
265 * 2 for pages with a mapping
266 * 3 for pages with a mapping and PagePrivate/PagePrivate2 set.
268 static int migrate_page_move_mapping(struct address_space *mapping,
269 struct page *newpage, struct page *page,
270 struct buffer_head *head, enum migrate_mode mode)
276 /* Anonymous page without mapping */
277 if (page_count(page) != 1)
282 spin_lock_irq(&mapping->tree_lock);
284 pslot = radix_tree_lookup_slot(&mapping->page_tree,
287 expected_count = 2 + page_has_private(page);
288 if (page_count(page) != expected_count ||
289 radix_tree_deref_slot_protected(pslot, &mapping->tree_lock) != page) {
290 spin_unlock_irq(&mapping->tree_lock);
294 if (!page_freeze_refs(page, expected_count)) {
295 spin_unlock_irq(&mapping->tree_lock);
300 * In the async migration case of moving a page with buffers, lock the
301 * buffers using trylock before the mapping is moved. If the mapping
302 * was moved, we later failed to lock the buffers and could not move
303 * the mapping back due to an elevated page count, we would have to
304 * block waiting on other references to be dropped.
306 if (mode == MIGRATE_ASYNC && head &&
307 !buffer_migrate_lock_buffers(head, mode)) {
308 page_unfreeze_refs(page, expected_count);
309 spin_unlock_irq(&mapping->tree_lock);
314 * Now we know that no one else is looking at the page.
316 get_page(newpage); /* add cache reference */
317 if (PageSwapCache(page)) {
318 SetPageSwapCache(newpage);
319 set_page_private(newpage, page_private(page));
322 radix_tree_replace_slot(pslot, newpage);
325 * Drop cache reference from old page by unfreezing
326 * to one less reference.
327 * We know this isn't the last reference.
329 page_unfreeze_refs(page, expected_count - 1);
332 * If moved to a different zone then also account
333 * the page for that zone. Other VM counters will be
334 * taken care of when we establish references to the
335 * new page and drop references to the old page.
337 * Note that anonymous pages are accounted for
338 * via NR_FILE_PAGES and NR_ANON_PAGES if they
339 * are mapped to swap space.
341 __dec_zone_page_state(page, NR_FILE_PAGES);
342 __inc_zone_page_state(newpage, NR_FILE_PAGES);
343 if (!PageSwapCache(page) && PageSwapBacked(page)) {
344 __dec_zone_page_state(page, NR_SHMEM);
345 __inc_zone_page_state(newpage, NR_SHMEM);
347 spin_unlock_irq(&mapping->tree_lock);
353 * The expected number of remaining references is the same as that
354 * of migrate_page_move_mapping().
356 int migrate_huge_page_move_mapping(struct address_space *mapping,
357 struct page *newpage, struct page *page)
363 if (page_count(page) != 1)
368 spin_lock_irq(&mapping->tree_lock);
370 pslot = radix_tree_lookup_slot(&mapping->page_tree,
373 expected_count = 2 + page_has_private(page);
374 if (page_count(page) != expected_count ||
375 radix_tree_deref_slot_protected(pslot, &mapping->tree_lock) != page) {
376 spin_unlock_irq(&mapping->tree_lock);
380 if (!page_freeze_refs(page, expected_count)) {
381 spin_unlock_irq(&mapping->tree_lock);
387 radix_tree_replace_slot(pslot, newpage);
389 page_unfreeze_refs(page, expected_count - 1);
391 spin_unlock_irq(&mapping->tree_lock);
396 * Copy the page to its new location
398 void migrate_page_copy(struct page *newpage, struct page *page)
401 copy_huge_page(newpage, page);
403 copy_highpage(newpage, page);
406 SetPageError(newpage);
407 if (PageReferenced(page))
408 SetPageReferenced(newpage);
409 if (PageUptodate(page))
410 SetPageUptodate(newpage);
411 if (TestClearPageActive(page)) {
412 VM_BUG_ON(PageUnevictable(page));
413 SetPageActive(newpage);
414 } else if (TestClearPageUnevictable(page))
415 SetPageUnevictable(newpage);
416 if (PageChecked(page))
417 SetPageChecked(newpage);
418 if (PageMappedToDisk(page))
419 SetPageMappedToDisk(newpage);
421 if (PageDirty(page)) {
422 clear_page_dirty_for_io(page);
424 * Want to mark the page and the radix tree as dirty, and
425 * redo the accounting that clear_page_dirty_for_io undid,
426 * but we can't use set_page_dirty because that function
427 * is actually a signal that all of the page has become dirty.
428 * Whereas only part of our page may be dirty.
430 if (PageSwapBacked(page))
431 SetPageDirty(newpage);
433 __set_page_dirty_nobuffers(newpage);
436 mlock_migrate_page(newpage, page);
437 ksm_migrate_page(newpage, page);
439 ClearPageSwapCache(page);
440 ClearPagePrivate(page);
441 set_page_private(page, 0);
444 * If any waiters have accumulated on the new page then
447 if (PageWriteback(newpage))
448 end_page_writeback(newpage);
451 /************************************************************
452 * Migration functions
453 ***********************************************************/
455 /* Always fail migration. Used for mappings that are not movable */
456 int fail_migrate_page(struct address_space *mapping,
457 struct page *newpage, struct page *page)
461 EXPORT_SYMBOL(fail_migrate_page);
464 * Common logic to directly migrate a single page suitable for
465 * pages that do not use PagePrivate/PagePrivate2.
467 * Pages are locked upon entry and exit.
469 int migrate_page(struct address_space *mapping,
470 struct page *newpage, struct page *page,
471 enum migrate_mode mode)
475 BUG_ON(PageWriteback(page)); /* Writeback must be complete */
477 rc = migrate_page_move_mapping(mapping, newpage, page, NULL, mode);
482 migrate_page_copy(newpage, page);
485 EXPORT_SYMBOL(migrate_page);
489 * Migration function for pages with buffers. This function can only be used
490 * if the underlying filesystem guarantees that no other references to "page"
493 int buffer_migrate_page(struct address_space *mapping,
494 struct page *newpage, struct page *page, enum migrate_mode mode)
496 struct buffer_head *bh, *head;
499 if (!page_has_buffers(page))
500 return migrate_page(mapping, newpage, page, mode);
502 head = page_buffers(page);
504 rc = migrate_page_move_mapping(mapping, newpage, page, head, mode);
510 * In the async case, migrate_page_move_mapping locked the buffers
511 * with an IRQ-safe spinlock held. In the sync case, the buffers
512 * need to be locked now
514 if (mode != MIGRATE_ASYNC)
515 BUG_ON(!buffer_migrate_lock_buffers(head, mode));
517 ClearPagePrivate(page);
518 set_page_private(newpage, page_private(page));
519 set_page_private(page, 0);
525 set_bh_page(bh, newpage, bh_offset(bh));
526 bh = bh->b_this_page;
528 } while (bh != head);
530 SetPagePrivate(newpage);
532 migrate_page_copy(newpage, page);
538 bh = bh->b_this_page;
540 } while (bh != head);
544 EXPORT_SYMBOL(buffer_migrate_page);
548 * Writeback a page to clean the dirty state
550 static int writeout(struct address_space *mapping, struct page *page)
552 struct writeback_control wbc = {
553 .sync_mode = WB_SYNC_NONE,
556 .range_end = LLONG_MAX,
561 if (!mapping->a_ops->writepage)
562 /* No write method for the address space */
565 if (!clear_page_dirty_for_io(page))
566 /* Someone else already triggered a write */
570 * A dirty page may imply that the underlying filesystem has
571 * the page on some queue. So the page must be clean for
572 * migration. Writeout may mean we loose the lock and the
573 * page state is no longer what we checked for earlier.
574 * At this point we know that the migration attempt cannot
577 remove_migration_ptes(page, page);
579 rc = mapping->a_ops->writepage(page, &wbc);
581 if (rc != AOP_WRITEPAGE_ACTIVATE)
582 /* unlocked. Relock */
585 return (rc < 0) ? -EIO : -EAGAIN;
589 * Default handling if a filesystem does not provide a migration function.
591 static int fallback_migrate_page(struct address_space *mapping,
592 struct page *newpage, struct page *page, enum migrate_mode mode)
594 if (PageDirty(page)) {
595 /* Only writeback pages in full synchronous migration */
596 if (mode != MIGRATE_SYNC)
598 return writeout(mapping, page);
602 * Buffers may be managed in a filesystem specific way.
603 * We must have no buffers or drop them.
605 if (page_has_private(page) &&
606 !try_to_release_page(page, GFP_KERNEL))
609 return migrate_page(mapping, newpage, page, mode);
613 * Move a page to a newly allocated page
614 * The page is locked and all ptes have been successfully removed.
616 * The new page will have replaced the old page if this function
623 static int move_to_new_page(struct page *newpage, struct page *page,
624 int remap_swapcache, enum migrate_mode mode)
626 struct address_space *mapping;
630 * Block others from accessing the page when we get around to
631 * establishing additional references. We are the only one
632 * holding a reference to the new page at this point.
634 if (!trylock_page(newpage))
637 /* Prepare mapping for the new page.*/
638 newpage->index = page->index;
639 newpage->mapping = page->mapping;
640 if (PageSwapBacked(page))
641 SetPageSwapBacked(newpage);
643 mapping = page_mapping(page);
645 rc = migrate_page(mapping, newpage, page, mode);
646 else if (mapping->a_ops->migratepage)
648 * Most pages have a mapping and most filesystems provide a
649 * migratepage callback. Anonymous pages are part of swap
650 * space which also has its own migratepage callback. This
651 * is the most common path for page migration.
653 rc = mapping->a_ops->migratepage(mapping,
654 newpage, page, mode);
656 rc = fallback_migrate_page(mapping, newpage, page, mode);
659 newpage->mapping = NULL;
662 remove_migration_ptes(page, newpage);
663 page->mapping = NULL;
666 unlock_page(newpage);
671 static int __unmap_and_move(struct page *page, struct page *newpage,
672 int force, bool offlining, enum migrate_mode mode)
675 int remap_swapcache = 1;
676 struct mem_cgroup *mem;
677 struct anon_vma *anon_vma = NULL;
679 if (!trylock_page(page)) {
680 if (!force || mode == MIGRATE_ASYNC)
684 * It's not safe for direct compaction to call lock_page.
685 * For example, during page readahead pages are added locked
686 * to the LRU. Later, when the IO completes the pages are
687 * marked uptodate and unlocked. However, the queueing
688 * could be merging multiple pages for one bio (e.g.
689 * mpage_readpages). If an allocation happens for the
690 * second or third page, the process can end up locking
691 * the same page twice and deadlocking. Rather than
692 * trying to be clever about what pages can be locked,
693 * avoid the use of lock_page for direct compaction
696 if (current->flags & PF_MEMALLOC)
703 * Only memory hotplug's offline_pages() caller has locked out KSM,
704 * and can safely migrate a KSM page. The other cases have skipped
705 * PageKsm along with PageReserved - but it is only now when we have
706 * the page lock that we can be certain it will not go KSM beneath us
707 * (KSM will not upgrade a page from PageAnon to PageKsm when it sees
708 * its pagecount raised, but only here do we take the page lock which
711 if (PageKsm(page) && !offlining) {
716 /* charge against new page */
717 mem_cgroup_prepare_migration(page, newpage, &mem);
719 if (PageWriteback(page)) {
721 * Only in the case of a full syncronous migration is it
722 * necessary to wait for PageWriteback. In the async case,
723 * the retry loop is too short and in the sync-light case,
724 * the overhead of stalling is too much
726 if (mode != MIGRATE_SYNC) {
732 wait_on_page_writeback(page);
735 * By try_to_unmap(), page->mapcount goes down to 0 here. In this case,
736 * we cannot notice that anon_vma is freed while we migrates a page.
737 * This get_anon_vma() delays freeing anon_vma pointer until the end
738 * of migration. File cache pages are no problem because of page_lock()
739 * File Caches may use write_page() or lock_page() in migration, then,
740 * just care Anon page here.
742 if (PageAnon(page)) {
744 * Only page_lock_anon_vma() understands the subtleties of
745 * getting a hold on an anon_vma from outside one of its mms.
747 anon_vma = page_get_anon_vma(page);
752 } else if (PageSwapCache(page)) {
754 * We cannot be sure that the anon_vma of an unmapped
755 * swapcache page is safe to use because we don't
756 * know in advance if the VMA that this page belonged
757 * to still exists. If the VMA and others sharing the
758 * data have been freed, then the anon_vma could
759 * already be invalid.
761 * To avoid this possibility, swapcache pages get
762 * migrated but are not remapped when migration
772 * Corner case handling:
773 * 1. When a new swap-cache page is read into, it is added to the LRU
774 * and treated as swapcache but it has no rmap yet.
775 * Calling try_to_unmap() against a page->mapping==NULL page will
776 * trigger a BUG. So handle it here.
777 * 2. An orphaned page (see truncate_complete_page) might have
778 * fs-private metadata. The page can be picked up due to memory
779 * offlining. Everywhere else except page reclaim, the page is
780 * invisible to the vm, so the page can not be migrated. So try to
781 * free the metadata, so the page can be freed.
783 if (!page->mapping) {
784 VM_BUG_ON(PageAnon(page));
785 if (page_has_private(page)) {
786 try_to_free_buffers(page);
792 /* Establish migration ptes or remove ptes */
793 try_to_unmap(page, TTU_MIGRATION|TTU_IGNORE_MLOCK|TTU_IGNORE_ACCESS);
796 if (!page_mapped(page))
797 rc = move_to_new_page(newpage, page, remap_swapcache, mode);
799 if (rc && remap_swapcache)
800 remove_migration_ptes(page, page);
802 /* Drop an anon_vma reference if we took one */
804 put_anon_vma(anon_vma);
807 mem_cgroup_end_migration(mem, page, newpage, rc == 0);
815 * Obtain the lock on page, remove all ptes and migrate the page
816 * to the newly allocated page in newpage.
818 static int unmap_and_move(new_page_t get_new_page, unsigned long private,
819 struct page *page, int force, bool offlining,
820 enum migrate_mode mode)
824 struct page *newpage = get_new_page(page, private, &result);
829 if (page_count(page) == 1) {
830 /* page was freed from under us. So we are done. */
834 if (unlikely(PageTransHuge(page)))
835 if (unlikely(split_huge_page(page)))
838 rc = __unmap_and_move(page, newpage, force, offlining, mode);
842 * A page that has been migrated has all references
843 * removed and will be freed. A page that has not been
844 * migrated will have kepts its references and be
847 list_del(&page->lru);
848 dec_zone_page_state(page, NR_ISOLATED_ANON +
849 page_is_file_cache(page));
850 putback_lru_page(page);
853 * Move the new page to the LRU. If migration was not successful
854 * then this will free the page.
856 putback_lru_page(newpage);
861 *result = page_to_nid(newpage);
867 * Counterpart of unmap_and_move_page() for hugepage migration.
869 * This function doesn't wait the completion of hugepage I/O
870 * because there is no race between I/O and migration for hugepage.
871 * Note that currently hugepage I/O occurs only in direct I/O
872 * where no lock is held and PG_writeback is irrelevant,
873 * and writeback status of all subpages are counted in the reference
874 * count of the head page (i.e. if all subpages of a 2MB hugepage are
875 * under direct I/O, the reference of the head page is 512 and a bit more.)
876 * This means that when we try to migrate hugepage whose subpages are
877 * doing direct I/O, some references remain after try_to_unmap() and
878 * hugepage migration fails without data corruption.
880 * There is also no race when direct I/O is issued on the page under migration,
881 * because then pte is replaced with migration swap entry and direct I/O code
882 * will wait in the page fault for migration to complete.
884 static int unmap_and_move_huge_page(new_page_t get_new_page,
885 unsigned long private, struct page *hpage,
886 int force, bool offlining,
887 enum migrate_mode mode)
891 struct page *new_hpage = get_new_page(hpage, private, &result);
892 struct anon_vma *anon_vma = NULL;
899 if (!trylock_page(hpage)) {
900 if (!force || mode != MIGRATE_SYNC)
906 anon_vma = page_get_anon_vma(hpage);
908 try_to_unmap(hpage, TTU_MIGRATION|TTU_IGNORE_MLOCK|TTU_IGNORE_ACCESS);
910 if (!page_mapped(hpage))
911 rc = move_to_new_page(new_hpage, hpage, 1, mode);
914 remove_migration_ptes(hpage, hpage);
917 put_anon_vma(anon_vma);
920 hugetlb_cgroup_migrate(hpage, new_hpage);
929 *result = page_to_nid(new_hpage);
937 * The function takes one list of pages to migrate and a function
938 * that determines from the page to be migrated and the private data
939 * the target of the move and allocates the page.
941 * The function returns after 10 attempts or if no pages
942 * are movable anymore because to has become empty
943 * or no retryable pages exist anymore.
944 * Caller should call putback_lru_pages to return pages to the LRU
945 * or free list only if ret != 0.
947 * Return: Number of pages not migrated or error code.
949 int migrate_pages(struct list_head *from,
950 new_page_t get_new_page, unsigned long private, bool offlining,
951 enum migrate_mode mode)
958 int swapwrite = current->flags & PF_SWAPWRITE;
962 current->flags |= PF_SWAPWRITE;
964 for(pass = 0; pass < 10 && retry; pass++) {
967 list_for_each_entry_safe(page, page2, from, lru) {
970 rc = unmap_and_move(get_new_page, private,
971 page, pass > 2, offlining,
983 /* Permanent failure */
992 current->flags &= ~PF_SWAPWRITE;
997 return nr_failed + retry;
1000 int migrate_huge_page(struct page *hpage, new_page_t get_new_page,
1001 unsigned long private, bool offlining,
1002 enum migrate_mode mode)
1006 for (pass = 0; pass < 10; pass++) {
1007 rc = unmap_and_move_huge_page(get_new_page,
1008 private, hpage, pass > 2, offlining,
1030 * Move a list of individual pages
1032 struct page_to_node {
1039 static struct page *new_page_node(struct page *p, unsigned long private,
1042 struct page_to_node *pm = (struct page_to_node *)private;
1044 while (pm->node != MAX_NUMNODES && pm->page != p)
1047 if (pm->node == MAX_NUMNODES)
1050 *result = &pm->status;
1052 return alloc_pages_exact_node(pm->node,
1053 GFP_HIGHUSER_MOVABLE | GFP_THISNODE, 0);
1057 * Move a set of pages as indicated in the pm array. The addr
1058 * field must be set to the virtual address of the page to be moved
1059 * and the node number must contain a valid target node.
1060 * The pm array ends with node = MAX_NUMNODES.
1062 static int do_move_page_to_node_array(struct mm_struct *mm,
1063 struct page_to_node *pm,
1067 struct page_to_node *pp;
1068 LIST_HEAD(pagelist);
1070 down_read(&mm->mmap_sem);
1073 * Build a list of pages to migrate
1075 for (pp = pm; pp->node != MAX_NUMNODES; pp++) {
1076 struct vm_area_struct *vma;
1080 vma = find_vma(mm, pp->addr);
1081 if (!vma || pp->addr < vma->vm_start || !vma_migratable(vma))
1084 page = follow_page(vma, pp->addr, FOLL_GET|FOLL_SPLIT);
1086 err = PTR_ERR(page);
1094 /* Use PageReserved to check for zero page */
1095 if (PageReserved(page) || PageKsm(page))
1099 err = page_to_nid(page);
1101 if (err == pp->node)
1103 * Node already in the right place
1108 if (page_mapcount(page) > 1 &&
1112 err = isolate_lru_page(page);
1114 list_add_tail(&page->lru, &pagelist);
1115 inc_zone_page_state(page, NR_ISOLATED_ANON +
1116 page_is_file_cache(page));
1120 * Either remove the duplicate refcount from
1121 * isolate_lru_page() or drop the page ref if it was
1130 if (!list_empty(&pagelist)) {
1131 err = migrate_pages(&pagelist, new_page_node,
1132 (unsigned long)pm, 0, MIGRATE_SYNC);
1134 putback_lru_pages(&pagelist);
1137 up_read(&mm->mmap_sem);
1142 * Migrate an array of page address onto an array of nodes and fill
1143 * the corresponding array of status.
1145 static int do_pages_move(struct mm_struct *mm, nodemask_t task_nodes,
1146 unsigned long nr_pages,
1147 const void __user * __user *pages,
1148 const int __user *nodes,
1149 int __user *status, int flags)
1151 struct page_to_node *pm;
1152 unsigned long chunk_nr_pages;
1153 unsigned long chunk_start;
1157 pm = (struct page_to_node *)__get_free_page(GFP_KERNEL);
1164 * Store a chunk of page_to_node array in a page,
1165 * but keep the last one as a marker
1167 chunk_nr_pages = (PAGE_SIZE / sizeof(struct page_to_node)) - 1;
1169 for (chunk_start = 0;
1170 chunk_start < nr_pages;
1171 chunk_start += chunk_nr_pages) {
1174 if (chunk_start + chunk_nr_pages > nr_pages)
1175 chunk_nr_pages = nr_pages - chunk_start;
1177 /* fill the chunk pm with addrs and nodes from user-space */
1178 for (j = 0; j < chunk_nr_pages; j++) {
1179 const void __user *p;
1183 if (get_user(p, pages + j + chunk_start))
1185 pm[j].addr = (unsigned long) p;
1187 if (get_user(node, nodes + j + chunk_start))
1191 if (node < 0 || node >= MAX_NUMNODES)
1194 if (!node_state(node, N_HIGH_MEMORY))
1198 if (!node_isset(node, task_nodes))
1204 /* End marker for this chunk */
1205 pm[chunk_nr_pages].node = MAX_NUMNODES;
1207 /* Migrate this chunk */
1208 err = do_move_page_to_node_array(mm, pm,
1209 flags & MPOL_MF_MOVE_ALL);
1213 /* Return status information */
1214 for (j = 0; j < chunk_nr_pages; j++)
1215 if (put_user(pm[j].status, status + j + chunk_start)) {
1223 free_page((unsigned long)pm);
1229 * Determine the nodes of an array of pages and store it in an array of status.
1231 static void do_pages_stat_array(struct mm_struct *mm, unsigned long nr_pages,
1232 const void __user **pages, int *status)
1236 down_read(&mm->mmap_sem);
1238 for (i = 0; i < nr_pages; i++) {
1239 unsigned long addr = (unsigned long)(*pages);
1240 struct vm_area_struct *vma;
1244 vma = find_vma(mm, addr);
1245 if (!vma || addr < vma->vm_start)
1248 page = follow_page(vma, addr, 0);
1250 err = PTR_ERR(page);
1255 /* Use PageReserved to check for zero page */
1256 if (!page || PageReserved(page) || PageKsm(page))
1259 err = page_to_nid(page);
1267 up_read(&mm->mmap_sem);
1271 * Determine the nodes of a user array of pages and store it in
1272 * a user array of status.
1274 static int do_pages_stat(struct mm_struct *mm, unsigned long nr_pages,
1275 const void __user * __user *pages,
1278 #define DO_PAGES_STAT_CHUNK_NR 16
1279 const void __user *chunk_pages[DO_PAGES_STAT_CHUNK_NR];
1280 int chunk_status[DO_PAGES_STAT_CHUNK_NR];
1283 unsigned long chunk_nr;
1285 chunk_nr = nr_pages;
1286 if (chunk_nr > DO_PAGES_STAT_CHUNK_NR)
1287 chunk_nr = DO_PAGES_STAT_CHUNK_NR;
1289 if (copy_from_user(chunk_pages, pages, chunk_nr * sizeof(*chunk_pages)))
1292 do_pages_stat_array(mm, chunk_nr, chunk_pages, chunk_status);
1294 if (copy_to_user(status, chunk_status, chunk_nr * sizeof(*status)))
1299 nr_pages -= chunk_nr;
1301 return nr_pages ? -EFAULT : 0;
1305 * Move a list of pages in the address space of the currently executing
1308 SYSCALL_DEFINE6(move_pages, pid_t, pid, unsigned long, nr_pages,
1309 const void __user * __user *, pages,
1310 const int __user *, nodes,
1311 int __user *, status, int, flags)
1313 const struct cred *cred = current_cred(), *tcred;
1314 struct task_struct *task;
1315 struct mm_struct *mm;
1317 nodemask_t task_nodes;
1320 if (flags & ~(MPOL_MF_MOVE|MPOL_MF_MOVE_ALL))
1323 if ((flags & MPOL_MF_MOVE_ALL) && !capable(CAP_SYS_NICE))
1326 /* Find the mm_struct */
1328 task = pid ? find_task_by_vpid(pid) : current;
1333 get_task_struct(task);
1336 * Check if this process has the right to modify the specified
1337 * process. The right exists if the process has administrative
1338 * capabilities, superuser privileges or the same
1339 * userid as the target process.
1341 tcred = __task_cred(task);
1342 if (!uid_eq(cred->euid, tcred->suid) && !uid_eq(cred->euid, tcred->uid) &&
1343 !uid_eq(cred->uid, tcred->suid) && !uid_eq(cred->uid, tcred->uid) &&
1344 !capable(CAP_SYS_NICE)) {
1351 err = security_task_movememory(task);
1355 task_nodes = cpuset_mems_allowed(task);
1356 mm = get_task_mm(task);
1357 put_task_struct(task);
1363 err = do_pages_move(mm, task_nodes, nr_pages, pages,
1364 nodes, status, flags);
1366 err = do_pages_stat(mm, nr_pages, pages, status);
1372 put_task_struct(task);
1377 * Call migration functions in the vma_ops that may prepare
1378 * memory in a vm for migration. migration functions may perform
1379 * the migration for vmas that do not have an underlying page struct.
1381 int migrate_vmas(struct mm_struct *mm, const nodemask_t *to,
1382 const nodemask_t *from, unsigned long flags)
1384 struct vm_area_struct *vma;
1387 for (vma = mm->mmap; vma && !err; vma = vma->vm_next) {
1388 if (vma->vm_ops && vma->vm_ops->migrate) {
1389 err = vma->vm_ops->migrate(vma, to, from, flags);