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>
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_trans_huge(*pmd))
116 if (!pmd_present(*pmd))
119 ptep = pte_offset_map(pmd, addr);
122 * Peek to check is_swap_pte() before taking ptlock? No, we
123 * can race mremap's move_ptes(), which skips anon_vma lock.
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 void migration_entry_wait(struct mm_struct *mm, pmd_t *pmd,
184 unsigned long address)
191 ptep = pte_offset_map_lock(mm, pmd, address, &ptl);
193 if (!is_swap_pte(pte))
196 entry = pte_to_swp_entry(pte);
197 if (!is_migration_entry(entry))
200 page = migration_entry_to_page(entry);
203 * Once radix-tree replacement of page migration started, page_count
204 * *must* be zero. And, we don't want to call wait_on_page_locked()
205 * against a page without get_page().
206 * So, we use get_page_unless_zero(), here. Even failed, page fault
209 if (!get_page_unless_zero(page))
211 pte_unmap_unlock(ptep, ptl);
212 wait_on_page_locked(page);
216 pte_unmap_unlock(ptep, ptl);
220 /* Returns true if all buffers are successfully locked */
221 static bool buffer_migrate_lock_buffers(struct buffer_head *head,
222 enum migrate_mode mode)
224 struct buffer_head *bh = head;
226 /* Simple case, sync compaction */
227 if (mode != MIGRATE_ASYNC) {
231 bh = bh->b_this_page;
233 } while (bh != head);
238 /* async case, we cannot block on lock_buffer so use trylock_buffer */
241 if (!trylock_buffer(bh)) {
243 * We failed to lock the buffer and cannot stall in
244 * async migration. Release the taken locks
246 struct buffer_head *failed_bh = bh;
249 while (bh != failed_bh) {
252 bh = bh->b_this_page;
257 bh = bh->b_this_page;
258 } while (bh != head);
262 static inline bool buffer_migrate_lock_buffers(struct buffer_head *head,
263 enum migrate_mode mode)
267 #endif /* CONFIG_BLOCK */
270 * Replace the page in the mapping.
272 * The number of remaining references must be:
273 * 1 for anonymous pages without a mapping
274 * 2 for pages with a mapping
275 * 3 for pages with a mapping and PagePrivate/PagePrivate2 set.
277 static int migrate_page_move_mapping(struct address_space *mapping,
278 struct page *newpage, struct page *page,
279 struct buffer_head *head, enum migrate_mode mode)
285 /* Anonymous page without mapping */
286 if (page_count(page) != 1)
291 spin_lock_irq(&mapping->tree_lock);
293 pslot = radix_tree_lookup_slot(&mapping->page_tree,
296 expected_count = 2 + page_has_private(page);
297 if (page_count(page) != expected_count ||
298 radix_tree_deref_slot_protected(pslot, &mapping->tree_lock) != page) {
299 spin_unlock_irq(&mapping->tree_lock);
303 if (!page_freeze_refs(page, expected_count)) {
304 spin_unlock_irq(&mapping->tree_lock);
309 * In the async migration case of moving a page with buffers, lock the
310 * buffers using trylock before the mapping is moved. If the mapping
311 * was moved, we later failed to lock the buffers and could not move
312 * the mapping back due to an elevated page count, we would have to
313 * block waiting on other references to be dropped.
315 if (mode == MIGRATE_ASYNC && head &&
316 !buffer_migrate_lock_buffers(head, mode)) {
317 page_unfreeze_refs(page, expected_count);
318 spin_unlock_irq(&mapping->tree_lock);
323 * Now we know that no one else is looking at the page.
325 get_page(newpage); /* add cache reference */
326 if (PageSwapCache(page)) {
327 SetPageSwapCache(newpage);
328 set_page_private(newpage, page_private(page));
331 radix_tree_replace_slot(pslot, newpage);
334 * Drop cache reference from old page by unfreezing
335 * to one less reference.
336 * We know this isn't the last reference.
338 page_unfreeze_refs(page, expected_count - 1);
341 * If moved to a different zone then also account
342 * the page for that zone. Other VM counters will be
343 * taken care of when we establish references to the
344 * new page and drop references to the old page.
346 * Note that anonymous pages are accounted for
347 * via NR_FILE_PAGES and NR_ANON_PAGES if they
348 * are mapped to swap space.
350 __dec_zone_page_state(page, NR_FILE_PAGES);
351 __inc_zone_page_state(newpage, NR_FILE_PAGES);
352 if (!PageSwapCache(page) && PageSwapBacked(page)) {
353 __dec_zone_page_state(page, NR_SHMEM);
354 __inc_zone_page_state(newpage, NR_SHMEM);
356 spin_unlock_irq(&mapping->tree_lock);
362 * The expected number of remaining references is the same as that
363 * of migrate_page_move_mapping().
365 int migrate_huge_page_move_mapping(struct address_space *mapping,
366 struct page *newpage, struct page *page)
372 if (page_count(page) != 1)
377 spin_lock_irq(&mapping->tree_lock);
379 pslot = radix_tree_lookup_slot(&mapping->page_tree,
382 expected_count = 2 + page_has_private(page);
383 if (page_count(page) != expected_count ||
384 radix_tree_deref_slot_protected(pslot, &mapping->tree_lock) != page) {
385 spin_unlock_irq(&mapping->tree_lock);
389 if (!page_freeze_refs(page, expected_count)) {
390 spin_unlock_irq(&mapping->tree_lock);
396 radix_tree_replace_slot(pslot, newpage);
398 page_unfreeze_refs(page, expected_count - 1);
400 spin_unlock_irq(&mapping->tree_lock);
405 * Copy the page to its new location
407 void migrate_page_copy(struct page *newpage, struct page *page)
410 copy_huge_page(newpage, page);
412 copy_highpage(newpage, page);
415 SetPageError(newpage);
416 if (PageReferenced(page))
417 SetPageReferenced(newpage);
418 if (PageUptodate(page))
419 SetPageUptodate(newpage);
420 if (TestClearPageActive(page)) {
421 VM_BUG_ON(PageUnevictable(page));
422 SetPageActive(newpage);
423 } else if (TestClearPageUnevictable(page))
424 SetPageUnevictable(newpage);
425 if (PageChecked(page))
426 SetPageChecked(newpage);
427 if (PageMappedToDisk(page))
428 SetPageMappedToDisk(newpage);
430 if (PageDirty(page)) {
431 clear_page_dirty_for_io(page);
433 * Want to mark the page and the radix tree as dirty, and
434 * redo the accounting that clear_page_dirty_for_io undid,
435 * but we can't use set_page_dirty because that function
436 * is actually a signal that all of the page has become dirty.
437 * Whereas only part of our page may be dirty.
439 __set_page_dirty_nobuffers(newpage);
442 mlock_migrate_page(newpage, page);
443 ksm_migrate_page(newpage, page);
445 ClearPageSwapCache(page);
446 ClearPagePrivate(page);
447 set_page_private(page, 0);
450 * If any waiters have accumulated on the new page then
453 if (PageWriteback(newpage))
454 end_page_writeback(newpage);
457 /************************************************************
458 * Migration functions
459 ***********************************************************/
461 /* Always fail migration. Used for mappings that are not movable */
462 int fail_migrate_page(struct address_space *mapping,
463 struct page *newpage, struct page *page)
467 EXPORT_SYMBOL(fail_migrate_page);
470 * Common logic to directly migrate a single page suitable for
471 * pages that do not use PagePrivate/PagePrivate2.
473 * Pages are locked upon entry and exit.
475 int migrate_page(struct address_space *mapping,
476 struct page *newpage, struct page *page,
477 enum migrate_mode mode)
481 BUG_ON(PageWriteback(page)); /* Writeback must be complete */
483 rc = migrate_page_move_mapping(mapping, newpage, page, NULL, mode);
488 migrate_page_copy(newpage, page);
491 EXPORT_SYMBOL(migrate_page);
495 * Migration function for pages with buffers. This function can only be used
496 * if the underlying filesystem guarantees that no other references to "page"
499 int buffer_migrate_page(struct address_space *mapping,
500 struct page *newpage, struct page *page, enum migrate_mode mode)
502 struct buffer_head *bh, *head;
505 if (!page_has_buffers(page))
506 return migrate_page(mapping, newpage, page, mode);
508 head = page_buffers(page);
510 rc = migrate_page_move_mapping(mapping, newpage, page, head, mode);
516 * In the async case, migrate_page_move_mapping locked the buffers
517 * with an IRQ-safe spinlock held. In the sync case, the buffers
518 * need to be locked now
520 if (mode != MIGRATE_ASYNC)
521 BUG_ON(!buffer_migrate_lock_buffers(head, mode));
523 ClearPagePrivate(page);
524 set_page_private(newpage, page_private(page));
525 set_page_private(page, 0);
531 set_bh_page(bh, newpage, bh_offset(bh));
532 bh = bh->b_this_page;
534 } while (bh != head);
536 SetPagePrivate(newpage);
538 migrate_page_copy(newpage, page);
544 bh = bh->b_this_page;
546 } while (bh != head);
550 EXPORT_SYMBOL(buffer_migrate_page);
554 * Writeback a page to clean the dirty state
556 static int writeout(struct address_space *mapping, struct page *page)
558 struct writeback_control wbc = {
559 .sync_mode = WB_SYNC_NONE,
562 .range_end = LLONG_MAX,
567 if (!mapping->a_ops->writepage)
568 /* No write method for the address space */
571 if (!clear_page_dirty_for_io(page))
572 /* Someone else already triggered a write */
576 * A dirty page may imply that the underlying filesystem has
577 * the page on some queue. So the page must be clean for
578 * migration. Writeout may mean we loose the lock and the
579 * page state is no longer what we checked for earlier.
580 * At this point we know that the migration attempt cannot
583 remove_migration_ptes(page, page);
585 rc = mapping->a_ops->writepage(page, &wbc);
587 if (rc != AOP_WRITEPAGE_ACTIVATE)
588 /* unlocked. Relock */
591 return (rc < 0) ? -EIO : -EAGAIN;
595 * Default handling if a filesystem does not provide a migration function.
597 static int fallback_migrate_page(struct address_space *mapping,
598 struct page *newpage, struct page *page, enum migrate_mode mode)
600 if (PageDirty(page)) {
601 /* Only writeback pages in full synchronous migration */
602 if (mode != MIGRATE_SYNC)
604 return writeout(mapping, page);
608 * Buffers may be managed in a filesystem specific way.
609 * We must have no buffers or drop them.
611 if (page_has_private(page) &&
612 !try_to_release_page(page, GFP_KERNEL))
615 return migrate_page(mapping, newpage, page, mode);
619 * Move a page to a newly allocated page
620 * The page is locked and all ptes have been successfully removed.
622 * The new page will have replaced the old page if this function
629 static int move_to_new_page(struct page *newpage, struct page *page,
630 int remap_swapcache, enum migrate_mode mode)
632 struct address_space *mapping;
636 * Block others from accessing the page when we get around to
637 * establishing additional references. We are the only one
638 * holding a reference to the new page at this point.
640 if (!trylock_page(newpage))
643 /* Prepare mapping for the new page.*/
644 newpage->index = page->index;
645 newpage->mapping = page->mapping;
646 if (PageSwapBacked(page))
647 SetPageSwapBacked(newpage);
649 mapping = page_mapping(page);
651 rc = migrate_page(mapping, newpage, page, mode);
652 else if (mapping->a_ops->migratepage)
654 * Most pages have a mapping and most filesystems provide a
655 * migratepage callback. Anonymous pages are part of swap
656 * space which also has its own migratepage callback. This
657 * is the most common path for page migration.
659 rc = mapping->a_ops->migratepage(mapping,
660 newpage, page, mode);
662 rc = fallback_migrate_page(mapping, newpage, page, mode);
665 newpage->mapping = NULL;
668 remove_migration_ptes(page, newpage);
669 page->mapping = NULL;
672 unlock_page(newpage);
677 static int __unmap_and_move(struct page *page, struct page *newpage,
678 int force, bool offlining, enum migrate_mode mode)
681 int remap_swapcache = 1;
683 struct mem_cgroup *mem;
684 struct anon_vma *anon_vma = NULL;
686 if (!trylock_page(page)) {
687 if (!force || mode == MIGRATE_ASYNC)
691 * It's not safe for direct compaction to call lock_page.
692 * For example, during page readahead pages are added locked
693 * to the LRU. Later, when the IO completes the pages are
694 * marked uptodate and unlocked. However, the queueing
695 * could be merging multiple pages for one bio (e.g.
696 * mpage_readpages). If an allocation happens for the
697 * second or third page, the process can end up locking
698 * the same page twice and deadlocking. Rather than
699 * trying to be clever about what pages can be locked,
700 * avoid the use of lock_page for direct compaction
703 if (current->flags & PF_MEMALLOC)
710 * Only memory hotplug's offline_pages() caller has locked out KSM,
711 * and can safely migrate a KSM page. The other cases have skipped
712 * PageKsm along with PageReserved - but it is only now when we have
713 * the page lock that we can be certain it will not go KSM beneath us
714 * (KSM will not upgrade a page from PageAnon to PageKsm when it sees
715 * its pagecount raised, but only here do we take the page lock which
718 if (PageKsm(page) && !offlining) {
723 /* charge against new page */
724 charge = mem_cgroup_prepare_migration(page, newpage, &mem, GFP_KERNEL);
725 if (charge == -ENOMEM) {
731 if (PageWriteback(page)) {
733 * Only in the case of a full syncronous migration is it
734 * necessary to wait for PageWriteback. In the async case,
735 * the retry loop is too short and in the sync-light case,
736 * the overhead of stalling is too much
738 if (mode != MIGRATE_SYNC) {
744 wait_on_page_writeback(page);
747 * By try_to_unmap(), page->mapcount goes down to 0 here. In this case,
748 * we cannot notice that anon_vma is freed while we migrates a page.
749 * This get_anon_vma() delays freeing anon_vma pointer until the end
750 * of migration. File cache pages are no problem because of page_lock()
751 * File Caches may use write_page() or lock_page() in migration, then,
752 * just care Anon page here.
754 if (PageAnon(page)) {
756 * Only page_lock_anon_vma() understands the subtleties of
757 * getting a hold on an anon_vma from outside one of its mms.
759 anon_vma = page_get_anon_vma(page);
764 } else if (PageSwapCache(page)) {
766 * We cannot be sure that the anon_vma of an unmapped
767 * swapcache page is safe to use because we don't
768 * know in advance if the VMA that this page belonged
769 * to still exists. If the VMA and others sharing the
770 * data have been freed, then the anon_vma could
771 * already be invalid.
773 * To avoid this possibility, swapcache pages get
774 * migrated but are not remapped when migration
784 * Corner case handling:
785 * 1. When a new swap-cache page is read into, it is added to the LRU
786 * and treated as swapcache but it has no rmap yet.
787 * Calling try_to_unmap() against a page->mapping==NULL page will
788 * trigger a BUG. So handle it here.
789 * 2. An orphaned page (see truncate_complete_page) might have
790 * fs-private metadata. The page can be picked up due to memory
791 * offlining. Everywhere else except page reclaim, the page is
792 * invisible to the vm, so the page can not be migrated. So try to
793 * free the metadata, so the page can be freed.
795 if (!page->mapping) {
796 VM_BUG_ON(PageAnon(page));
797 if (page_has_private(page)) {
798 try_to_free_buffers(page);
804 /* Establish migration ptes or remove ptes */
805 try_to_unmap(page, TTU_MIGRATION|TTU_IGNORE_MLOCK|TTU_IGNORE_ACCESS);
808 if (!page_mapped(page))
809 rc = move_to_new_page(newpage, page, remap_swapcache, mode);
811 if (rc && remap_swapcache)
812 remove_migration_ptes(page, page);
814 /* Drop an anon_vma reference if we took one */
816 put_anon_vma(anon_vma);
820 mem_cgroup_end_migration(mem, page, newpage, rc == 0);
828 * Obtain the lock on page, remove all ptes and migrate the page
829 * to the newly allocated page in newpage.
831 static int unmap_and_move(new_page_t get_new_page, unsigned long private,
832 struct page *page, int force, bool offlining,
833 enum migrate_mode mode)
837 struct page *newpage = get_new_page(page, private, &result);
842 if (page_count(page) == 1) {
843 /* page was freed from under us. So we are done. */
847 if (unlikely(PageTransHuge(page)))
848 if (unlikely(split_huge_page(page)))
851 rc = __unmap_and_move(page, newpage, force, offlining, mode);
855 * A page that has been migrated has all references
856 * removed and will be freed. A page that has not been
857 * migrated will have kepts its references and be
860 list_del(&page->lru);
861 dec_zone_page_state(page, NR_ISOLATED_ANON +
862 page_is_file_cache(page));
863 putback_lru_page(page);
866 * Move the new page to the LRU. If migration was not successful
867 * then this will free the page.
869 putback_lru_page(newpage);
874 *result = page_to_nid(newpage);
880 * Counterpart of unmap_and_move_page() for hugepage migration.
882 * This function doesn't wait the completion of hugepage I/O
883 * because there is no race between I/O and migration for hugepage.
884 * Note that currently hugepage I/O occurs only in direct I/O
885 * where no lock is held and PG_writeback is irrelevant,
886 * and writeback status of all subpages are counted in the reference
887 * count of the head page (i.e. if all subpages of a 2MB hugepage are
888 * under direct I/O, the reference of the head page is 512 and a bit more.)
889 * This means that when we try to migrate hugepage whose subpages are
890 * doing direct I/O, some references remain after try_to_unmap() and
891 * hugepage migration fails without data corruption.
893 * There is also no race when direct I/O is issued on the page under migration,
894 * because then pte is replaced with migration swap entry and direct I/O code
895 * will wait in the page fault for migration to complete.
897 static int unmap_and_move_huge_page(new_page_t get_new_page,
898 unsigned long private, struct page *hpage,
899 int force, bool offlining,
900 enum migrate_mode mode)
904 struct page *new_hpage = get_new_page(hpage, private, &result);
905 struct anon_vma *anon_vma = NULL;
912 if (!trylock_page(hpage)) {
913 if (!force || mode != MIGRATE_SYNC)
919 anon_vma = page_get_anon_vma(hpage);
921 try_to_unmap(hpage, TTU_MIGRATION|TTU_IGNORE_MLOCK|TTU_IGNORE_ACCESS);
923 if (!page_mapped(hpage))
924 rc = move_to_new_page(new_hpage, hpage, 1, mode);
927 remove_migration_ptes(hpage, hpage);
930 put_anon_vma(anon_vma);
935 list_del(&hpage->lru);
945 *result = page_to_nid(new_hpage);
953 * The function takes one list of pages to migrate and a function
954 * that determines from the page to be migrated and the private data
955 * the target of the move and allocates the page.
957 * The function returns after 10 attempts or if no pages
958 * are movable anymore because to has become empty
959 * or no retryable pages exist anymore.
960 * Caller should call putback_lru_pages to return pages to the LRU
961 * or free list only if ret != 0.
963 * Return: Number of pages not migrated or error code.
965 int migrate_pages(struct list_head *from,
966 new_page_t get_new_page, unsigned long private, bool offlining,
967 enum migrate_mode mode)
974 int swapwrite = current->flags & PF_SWAPWRITE;
978 current->flags |= PF_SWAPWRITE;
980 for(pass = 0; pass < 10 && retry; pass++) {
983 list_for_each_entry_safe(page, page2, from, lru) {
986 rc = unmap_and_move(get_new_page, private,
987 page, pass > 2, offlining,
999 /* Permanent failure */
1008 current->flags &= ~PF_SWAPWRITE;
1013 return nr_failed + retry;
1016 int migrate_huge_pages(struct list_head *from,
1017 new_page_t get_new_page, unsigned long private, bool offlining,
1018 enum migrate_mode mode)
1027 for (pass = 0; pass < 10 && retry; pass++) {
1030 list_for_each_entry_safe(page, page2, from, lru) {
1033 rc = unmap_and_move_huge_page(get_new_page,
1034 private, page, pass > 2, offlining,
1046 /* Permanent failure */
1057 return nr_failed + retry;
1062 * Move a list of individual pages
1064 struct page_to_node {
1071 static struct page *new_page_node(struct page *p, unsigned long private,
1074 struct page_to_node *pm = (struct page_to_node *)private;
1076 while (pm->node != MAX_NUMNODES && pm->page != p)
1079 if (pm->node == MAX_NUMNODES)
1082 *result = &pm->status;
1084 return alloc_pages_exact_node(pm->node,
1085 GFP_HIGHUSER_MOVABLE | GFP_THISNODE, 0);
1089 * Move a set of pages as indicated in the pm array. The addr
1090 * field must be set to the virtual address of the page to be moved
1091 * and the node number must contain a valid target node.
1092 * The pm array ends with node = MAX_NUMNODES.
1094 static int do_move_page_to_node_array(struct mm_struct *mm,
1095 struct page_to_node *pm,
1099 struct page_to_node *pp;
1100 LIST_HEAD(pagelist);
1102 down_read(&mm->mmap_sem);
1105 * Build a list of pages to migrate
1107 for (pp = pm; pp->node != MAX_NUMNODES; pp++) {
1108 struct vm_area_struct *vma;
1112 vma = find_vma(mm, pp->addr);
1113 if (!vma || pp->addr < vma->vm_start || !vma_migratable(vma))
1116 page = follow_page(vma, pp->addr, FOLL_GET|FOLL_SPLIT);
1118 err = PTR_ERR(page);
1126 /* Use PageReserved to check for zero page */
1127 if (PageReserved(page) || PageKsm(page))
1131 err = page_to_nid(page);
1133 if (err == pp->node)
1135 * Node already in the right place
1140 if (page_mapcount(page) > 1 &&
1144 err = isolate_lru_page(page);
1146 list_add_tail(&page->lru, &pagelist);
1147 inc_zone_page_state(page, NR_ISOLATED_ANON +
1148 page_is_file_cache(page));
1152 * Either remove the duplicate refcount from
1153 * isolate_lru_page() or drop the page ref if it was
1162 if (!list_empty(&pagelist)) {
1163 err = migrate_pages(&pagelist, new_page_node,
1164 (unsigned long)pm, 0, MIGRATE_SYNC);
1166 putback_lru_pages(&pagelist);
1169 up_read(&mm->mmap_sem);
1174 * Migrate an array of page address onto an array of nodes and fill
1175 * the corresponding array of status.
1177 static int do_pages_move(struct mm_struct *mm, nodemask_t task_nodes,
1178 unsigned long nr_pages,
1179 const void __user * __user *pages,
1180 const int __user *nodes,
1181 int __user *status, int flags)
1183 struct page_to_node *pm;
1184 unsigned long chunk_nr_pages;
1185 unsigned long chunk_start;
1189 pm = (struct page_to_node *)__get_free_page(GFP_KERNEL);
1196 * Store a chunk of page_to_node array in a page,
1197 * but keep the last one as a marker
1199 chunk_nr_pages = (PAGE_SIZE / sizeof(struct page_to_node)) - 1;
1201 for (chunk_start = 0;
1202 chunk_start < nr_pages;
1203 chunk_start += chunk_nr_pages) {
1206 if (chunk_start + chunk_nr_pages > nr_pages)
1207 chunk_nr_pages = nr_pages - chunk_start;
1209 /* fill the chunk pm with addrs and nodes from user-space */
1210 for (j = 0; j < chunk_nr_pages; j++) {
1211 const void __user *p;
1215 if (get_user(p, pages + j + chunk_start))
1217 pm[j].addr = (unsigned long) p;
1219 if (get_user(node, nodes + j + chunk_start))
1223 if (node < 0 || node >= MAX_NUMNODES)
1226 if (!node_state(node, N_HIGH_MEMORY))
1230 if (!node_isset(node, task_nodes))
1236 /* End marker for this chunk */
1237 pm[chunk_nr_pages].node = MAX_NUMNODES;
1239 /* Migrate this chunk */
1240 err = do_move_page_to_node_array(mm, pm,
1241 flags & MPOL_MF_MOVE_ALL);
1245 /* Return status information */
1246 for (j = 0; j < chunk_nr_pages; j++)
1247 if (put_user(pm[j].status, status + j + chunk_start)) {
1255 free_page((unsigned long)pm);
1261 * Determine the nodes of an array of pages and store it in an array of status.
1263 static void do_pages_stat_array(struct mm_struct *mm, unsigned long nr_pages,
1264 const void __user **pages, int *status)
1268 down_read(&mm->mmap_sem);
1270 for (i = 0; i < nr_pages; i++) {
1271 unsigned long addr = (unsigned long)(*pages);
1272 struct vm_area_struct *vma;
1276 vma = find_vma(mm, addr);
1277 if (!vma || addr < vma->vm_start)
1280 page = follow_page(vma, addr, 0);
1282 err = PTR_ERR(page);
1287 /* Use PageReserved to check for zero page */
1288 if (!page || PageReserved(page) || PageKsm(page))
1291 err = page_to_nid(page);
1299 up_read(&mm->mmap_sem);
1303 * Determine the nodes of a user array of pages and store it in
1304 * a user array of status.
1306 static int do_pages_stat(struct mm_struct *mm, unsigned long nr_pages,
1307 const void __user * __user *pages,
1310 #define DO_PAGES_STAT_CHUNK_NR 16
1311 const void __user *chunk_pages[DO_PAGES_STAT_CHUNK_NR];
1312 int chunk_status[DO_PAGES_STAT_CHUNK_NR];
1315 unsigned long chunk_nr;
1317 chunk_nr = nr_pages;
1318 if (chunk_nr > DO_PAGES_STAT_CHUNK_NR)
1319 chunk_nr = DO_PAGES_STAT_CHUNK_NR;
1321 if (copy_from_user(chunk_pages, pages, chunk_nr * sizeof(*chunk_pages)))
1324 do_pages_stat_array(mm, chunk_nr, chunk_pages, chunk_status);
1326 if (copy_to_user(status, chunk_status, chunk_nr * sizeof(*status)))
1331 nr_pages -= chunk_nr;
1333 return nr_pages ? -EFAULT : 0;
1337 * Move a list of pages in the address space of the currently executing
1340 SYSCALL_DEFINE6(move_pages, pid_t, pid, unsigned long, nr_pages,
1341 const void __user * __user *, pages,
1342 const int __user *, nodes,
1343 int __user *, status, int, flags)
1345 const struct cred *cred = current_cred(), *tcred;
1346 struct task_struct *task;
1347 struct mm_struct *mm;
1349 nodemask_t task_nodes;
1352 if (flags & ~(MPOL_MF_MOVE|MPOL_MF_MOVE_ALL))
1355 if ((flags & MPOL_MF_MOVE_ALL) && !capable(CAP_SYS_NICE))
1358 /* Find the mm_struct */
1360 task = pid ? find_task_by_vpid(pid) : current;
1365 get_task_struct(task);
1368 * Check if this process has the right to modify the specified
1369 * process. The right exists if the process has administrative
1370 * capabilities, superuser privileges or the same
1371 * userid as the target process.
1373 tcred = __task_cred(task);
1374 if (cred->euid != tcred->suid && cred->euid != tcred->uid &&
1375 cred->uid != tcred->suid && cred->uid != tcred->uid &&
1376 !capable(CAP_SYS_NICE)) {
1383 err = security_task_movememory(task);
1387 task_nodes = cpuset_mems_allowed(task);
1388 mm = get_task_mm(task);
1389 put_task_struct(task);
1395 err = do_pages_move(mm, task_nodes, nr_pages, pages,
1396 nodes, status, flags);
1398 err = do_pages_stat(mm, nr_pages, pages, status);
1404 put_task_struct(task);
1409 * Call migration functions in the vma_ops that may prepare
1410 * memory in a vm for migration. migration functions may perform
1411 * the migration for vmas that do not have an underlying page struct.
1413 int migrate_vmas(struct mm_struct *mm, const nodemask_t *to,
1414 const nodemask_t *from, unsigned long flags)
1416 struct vm_area_struct *vma;
1419 for (vma = mm->mmap; vma && !err; vma = vma->vm_next) {
1420 if (vma->vm_ops && vma->vm_ops->migrate) {
1421 err = vma->vm_ops->migrate(vma, to, from, flags);