blob: 1dc4598d2513c4f9d005512a4da7967e6b41a348 [file] [log] [blame]
Christoph Lameterb20a3502006-03-22 00:09:12 -08001/*
2 * Memory Migration functionality - linux/mm/migration.c
3 *
4 * Copyright (C) 2006 Silicon Graphics, Inc., Christoph Lameter
5 *
6 * Page migration was first developed in the context of the memory hotplug
7 * project. The main authors of the migration code are:
8 *
9 * IWAMOTO Toshihiro <iwamoto@valinux.co.jp>
10 * Hirokazu Takahashi <taka@valinux.co.jp>
11 * Dave Hansen <haveblue@us.ibm.com>
Christoph Lametercde53532008-07-04 09:59:22 -070012 * Christoph Lameter
Christoph Lameterb20a3502006-03-22 00:09:12 -080013 */
14
15#include <linux/migrate.h>
Paul Gortmakerb95f1b312011-10-16 02:01:52 -040016#include <linux/export.h>
Christoph Lameterb20a3502006-03-22 00:09:12 -080017#include <linux/swap.h>
Christoph Lameter06972122006-06-23 02:03:35 -070018#include <linux/swapops.h>
Christoph Lameterb20a3502006-03-22 00:09:12 -080019#include <linux/pagemap.h>
Christoph Lametere23ca002006-04-10 22:52:57 -070020#include <linux/buffer_head.h>
Christoph Lameterb20a3502006-03-22 00:09:12 -080021#include <linux/mm_inline.h>
Pavel Emelyanovb4888932007-10-18 23:40:14 -070022#include <linux/nsproxy.h>
Christoph Lameterb20a3502006-03-22 00:09:12 -080023#include <linux/pagevec.h>
Hugh Dickinse9995ef2009-12-14 17:59:31 -080024#include <linux/ksm.h>
Christoph Lameterb20a3502006-03-22 00:09:12 -080025#include <linux/rmap.h>
26#include <linux/topology.h>
27#include <linux/cpu.h>
28#include <linux/cpuset.h>
Christoph Lameter04e62a22006-06-23 02:03:38 -070029#include <linux/writeback.h>
Christoph Lameter742755a2006-06-23 02:03:55 -070030#include <linux/mempolicy.h>
31#include <linux/vmalloc.h>
David Quigley86c3a762006-06-23 02:04:02 -070032#include <linux/security.h>
Balbir Singh8a9f3cc2008-02-07 00:13:53 -080033#include <linux/memcontrol.h>
Adrian Bunk4f5ca262008-07-23 21:27:02 -070034#include <linux/syscalls.h>
Naoya Horiguchi290408d2010-09-08 10:19:35 +090035#include <linux/hugetlb.h>
Aneesh Kumar K.V8e6ac7f2012-07-31 16:42:27 -070036#include <linux/hugetlb_cgroup.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090037#include <linux/gfp.h>
Christoph Lameterb20a3502006-03-22 00:09:12 -080038
Michal Nazarewicz0d1836c2010-12-21 17:24:26 -080039#include <asm/tlbflush.h>
40
Christoph Lameterb20a3502006-03-22 00:09:12 -080041#include "internal.h"
42
Christoph Lameterb20a3502006-03-22 00:09:12 -080043/*
Christoph Lameter742755a2006-06-23 02:03:55 -070044 * migrate_prep() needs to be called before we start compiling a list of pages
Mel Gorman748446b2010-05-24 14:32:27 -070045 * to be migrated using isolate_lru_page(). If scheduling work on other CPUs is
46 * undesirable, use migrate_prep_local()
Christoph Lameterb20a3502006-03-22 00:09:12 -080047 */
48int migrate_prep(void)
49{
Christoph Lameterb20a3502006-03-22 00:09:12 -080050 /*
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.
55 */
56 lru_add_drain_all();
57
58 return 0;
59}
60
Mel Gorman748446b2010-05-24 14:32:27 -070061/* Do the necessary work of migrate_prep but not if it involves other CPUs */
62int migrate_prep_local(void)
63{
64 lru_add_drain();
65
66 return 0;
67}
68
Christoph Lameterb20a3502006-03-22 00:09:12 -080069/*
Lee Schermerhorn894bc312008-10-18 20:26:39 -070070 * Add isolated pages on the list back to the LRU under page lock
71 * to avoid leaking evictable pages back onto unevictable list.
Christoph Lameterb20a3502006-03-22 00:09:12 -080072 */
Minchan Kime13861d2010-05-24 14:31:59 -070073void putback_lru_pages(struct list_head *l)
Christoph Lameterb20a3502006-03-22 00:09:12 -080074{
75 struct page *page;
76 struct page *page2;
Christoph Lameterb20a3502006-03-22 00:09:12 -080077
78 list_for_each_entry_safe(page, page2, l, lru) {
Christoph Lametere24f0b82006-06-23 02:03:51 -070079 list_del(&page->lru);
KOSAKI Motohiroa7312862009-09-21 17:01:37 -070080 dec_zone_page_state(page, NR_ISOLATED_ANON +
Johannes Weiner6c0b1352009-09-21 17:02:59 -070081 page_is_file_cache(page));
Lee Schermerhorn894bc312008-10-18 20:26:39 -070082 putback_lru_page(page);
Christoph Lameterb20a3502006-03-22 00:09:12 -080083 }
Christoph Lameterb20a3502006-03-22 00:09:12 -080084}
85
Christoph Lameter06972122006-06-23 02:03:35 -070086/*
87 * Restore a potential migration pte to a working pte entry
88 */
Hugh Dickinse9995ef2009-12-14 17:59:31 -080089static int remove_migration_pte(struct page *new, struct vm_area_struct *vma,
90 unsigned long addr, void *old)
Christoph Lameter06972122006-06-23 02:03:35 -070091{
92 struct mm_struct *mm = vma->vm_mm;
93 swp_entry_t entry;
Christoph Lameter06972122006-06-23 02:03:35 -070094 pmd_t *pmd;
95 pte_t *ptep, pte;
96 spinlock_t *ptl;
97
Naoya Horiguchi290408d2010-09-08 10:19:35 +090098 if (unlikely(PageHuge(new))) {
99 ptep = huge_pte_offset(mm, addr);
100 if (!ptep)
101 goto out;
102 ptl = &mm->page_table_lock;
103 } else {
Bob Liu62190492012-12-11 16:00:37 -0800104 pmd = mm_find_pmd(mm, addr);
105 if (!pmd)
Naoya Horiguchi290408d2010-09-08 10:19:35 +0900106 goto out;
Andrea Arcangeli500d65d2011-01-13 15:46:55 -0800107 if (pmd_trans_huge(*pmd))
108 goto out;
Christoph Lameter06972122006-06-23 02:03:35 -0700109
Naoya Horiguchi290408d2010-09-08 10:19:35 +0900110 ptep = pte_offset_map(pmd, addr);
Christoph Lameter06972122006-06-23 02:03:35 -0700111
Hugh Dickins486cf462011-10-19 12:50:35 -0700112 /*
113 * Peek to check is_swap_pte() before taking ptlock? No, we
114 * can race mremap's move_ptes(), which skips anon_vma lock.
115 */
Christoph Lameter06972122006-06-23 02:03:35 -0700116
Naoya Horiguchi290408d2010-09-08 10:19:35 +0900117 ptl = pte_lockptr(mm, pmd);
118 }
119
Christoph Lameter06972122006-06-23 02:03:35 -0700120 spin_lock(ptl);
121 pte = *ptep;
122 if (!is_swap_pte(pte))
Hugh Dickinse9995ef2009-12-14 17:59:31 -0800123 goto unlock;
Christoph Lameter06972122006-06-23 02:03:35 -0700124
125 entry = pte_to_swp_entry(pte);
126
Hugh Dickinse9995ef2009-12-14 17:59:31 -0800127 if (!is_migration_entry(entry) ||
128 migration_entry_to_page(entry) != old)
129 goto unlock;
Christoph Lameter06972122006-06-23 02:03:35 -0700130
Christoph Lameter06972122006-06-23 02:03:35 -0700131 get_page(new);
132 pte = pte_mkold(mk_pte(new, vma->vm_page_prot));
133 if (is_write_migration_entry(entry))
134 pte = pte_mkwrite(pte);
Andi Kleen3ef8fd72010-10-11 16:03:21 +0200135#ifdef CONFIG_HUGETLB_PAGE
Naoya Horiguchi290408d2010-09-08 10:19:35 +0900136 if (PageHuge(new))
137 pte = pte_mkhuge(pte);
Andi Kleen3ef8fd72010-10-11 16:03:21 +0200138#endif
KAMEZAWA Hiroyuki97ee0522007-10-16 01:25:43 -0700139 flush_cache_page(vma, addr, pte_pfn(pte));
Christoph Lameter06972122006-06-23 02:03:35 -0700140 set_pte_at(mm, addr, ptep, pte);
Christoph Lameter04e62a22006-06-23 02:03:38 -0700141
Naoya Horiguchi290408d2010-09-08 10:19:35 +0900142 if (PageHuge(new)) {
143 if (PageAnon(new))
144 hugepage_add_anon_rmap(new, vma, addr);
145 else
146 page_dup_rmap(new);
147 } else if (PageAnon(new))
Christoph Lameter04e62a22006-06-23 02:03:38 -0700148 page_add_anon_rmap(new, vma, addr);
149 else
150 page_add_file_rmap(new);
151
152 /* No need to invalidate - it was non-present before */
Russell King4b3073e2009-12-18 16:40:18 +0000153 update_mmu_cache(vma, addr, ptep);
Hugh Dickinse9995ef2009-12-14 17:59:31 -0800154unlock:
Christoph Lameter06972122006-06-23 02:03:35 -0700155 pte_unmap_unlock(ptep, ptl);
Hugh Dickinse9995ef2009-12-14 17:59:31 -0800156out:
157 return SWAP_AGAIN;
Christoph Lameter06972122006-06-23 02:03:35 -0700158}
159
160/*
Christoph Lameter04e62a22006-06-23 02:03:38 -0700161 * Get rid of all migration entries and replace them by
162 * references to the indicated page.
163 */
164static void remove_migration_ptes(struct page *old, struct page *new)
165{
Hugh Dickinse9995ef2009-12-14 17:59:31 -0800166 rmap_walk(new, remove_migration_pte, old);
Christoph Lameter04e62a22006-06-23 02:03:38 -0700167}
168
169/*
Christoph Lameter06972122006-06-23 02:03:35 -0700170 * 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.
Christoph Lameter06972122006-06-23 02:03:35 -0700173 */
174void migration_entry_wait(struct mm_struct *mm, pmd_t *pmd,
175 unsigned long address)
176{
177 pte_t *ptep, pte;
178 spinlock_t *ptl;
179 swp_entry_t entry;
180 struct page *page;
181
182 ptep = pte_offset_map_lock(mm, pmd, address, &ptl);
183 pte = *ptep;
184 if (!is_swap_pte(pte))
185 goto out;
186
187 entry = pte_to_swp_entry(pte);
188 if (!is_migration_entry(entry))
189 goto out;
190
191 page = migration_entry_to_page(entry);
192
Nick Piggine2867812008-07-25 19:45:30 -0700193 /*
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
198 * will occur again.
199 */
200 if (!get_page_unless_zero(page))
201 goto out;
Christoph Lameter06972122006-06-23 02:03:35 -0700202 pte_unmap_unlock(ptep, ptl);
203 wait_on_page_locked(page);
204 put_page(page);
205 return;
206out:
207 pte_unmap_unlock(ptep, ptl);
208}
209
Mel Gormanb969c4a2012-01-12 17:19:34 -0800210#ifdef CONFIG_BLOCK
211/* Returns true if all buffers are successfully locked */
Mel Gormana6bc32b2012-01-12 17:19:43 -0800212static bool buffer_migrate_lock_buffers(struct buffer_head *head,
213 enum migrate_mode mode)
Mel Gormanb969c4a2012-01-12 17:19:34 -0800214{
215 struct buffer_head *bh = head;
216
217 /* Simple case, sync compaction */
Mel Gormana6bc32b2012-01-12 17:19:43 -0800218 if (mode != MIGRATE_ASYNC) {
Mel Gormanb969c4a2012-01-12 17:19:34 -0800219 do {
220 get_bh(bh);
221 lock_buffer(bh);
222 bh = bh->b_this_page;
223
224 } while (bh != head);
225
226 return true;
227 }
228
229 /* async case, we cannot block on lock_buffer so use trylock_buffer */
230 do {
231 get_bh(bh);
232 if (!trylock_buffer(bh)) {
233 /*
234 * We failed to lock the buffer and cannot stall in
235 * async migration. Release the taken locks
236 */
237 struct buffer_head *failed_bh = bh;
238 put_bh(failed_bh);
239 bh = head;
240 while (bh != failed_bh) {
241 unlock_buffer(bh);
242 put_bh(bh);
243 bh = bh->b_this_page;
244 }
245 return false;
246 }
247
248 bh = bh->b_this_page;
249 } while (bh != head);
250 return true;
251}
252#else
253static inline bool buffer_migrate_lock_buffers(struct buffer_head *head,
Mel Gormana6bc32b2012-01-12 17:19:43 -0800254 enum migrate_mode mode)
Mel Gormanb969c4a2012-01-12 17:19:34 -0800255{
256 return true;
257}
258#endif /* CONFIG_BLOCK */
259
Christoph Lameterb20a3502006-03-22 00:09:12 -0800260/*
Christoph Lameterc3fcf8a2006-06-23 02:03:32 -0700261 * Replace the page in the mapping.
Christoph Lameter5b5c7122006-06-23 02:03:29 -0700262 *
263 * The number of remaining references must be:
264 * 1 for anonymous pages without a mapping
265 * 2 for pages with a mapping
David Howells266cf652009-04-03 16:42:36 +0100266 * 3 for pages with a mapping and PagePrivate/PagePrivate2 set.
Christoph Lameterb20a3502006-03-22 00:09:12 -0800267 */
Christoph Lameter2d1db3b2006-06-23 02:03:33 -0700268static int migrate_page_move_mapping(struct address_space *mapping,
Mel Gormanb969c4a2012-01-12 17:19:34 -0800269 struct page *newpage, struct page *page,
Mel Gormana6bc32b2012-01-12 17:19:43 -0800270 struct buffer_head *head, enum migrate_mode mode)
Christoph Lameterb20a3502006-03-22 00:09:12 -0800271{
Nick Piggine2867812008-07-25 19:45:30 -0700272 int expected_count;
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800273 void **pslot;
Christoph Lameterb20a3502006-03-22 00:09:12 -0800274
Christoph Lameter6c5240a2006-06-23 02:03:37 -0700275 if (!mapping) {
Christoph Lameter0e8c7d02007-04-23 14:41:09 -0700276 /* Anonymous page without mapping */
Christoph Lameter6c5240a2006-06-23 02:03:37 -0700277 if (page_count(page) != 1)
278 return -EAGAIN;
279 return 0;
280 }
281
Nick Piggin19fd6232008-07-25 19:45:32 -0700282 spin_lock_irq(&mapping->tree_lock);
Christoph Lameterb20a3502006-03-22 00:09:12 -0800283
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800284 pslot = radix_tree_lookup_slot(&mapping->page_tree,
285 page_index(page));
Christoph Lameterb20a3502006-03-22 00:09:12 -0800286
Johannes Weineredcf4742009-09-21 17:02:59 -0700287 expected_count = 2 + page_has_private(page);
Nick Piggine2867812008-07-25 19:45:30 -0700288 if (page_count(page) != expected_count ||
Mel Gorman29c1f672011-01-13 15:47:21 -0800289 radix_tree_deref_slot_protected(pslot, &mapping->tree_lock) != page) {
Nick Piggin19fd6232008-07-25 19:45:32 -0700290 spin_unlock_irq(&mapping->tree_lock);
Christoph Lametere23ca002006-04-10 22:52:57 -0700291 return -EAGAIN;
Christoph Lameterb20a3502006-03-22 00:09:12 -0800292 }
293
Nick Piggine2867812008-07-25 19:45:30 -0700294 if (!page_freeze_refs(page, expected_count)) {
Nick Piggin19fd6232008-07-25 19:45:32 -0700295 spin_unlock_irq(&mapping->tree_lock);
Nick Piggine2867812008-07-25 19:45:30 -0700296 return -EAGAIN;
297 }
298
Christoph Lameterb20a3502006-03-22 00:09:12 -0800299 /*
Mel Gormanb969c4a2012-01-12 17:19:34 -0800300 * 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.
305 */
Mel Gormana6bc32b2012-01-12 17:19:43 -0800306 if (mode == MIGRATE_ASYNC && head &&
307 !buffer_migrate_lock_buffers(head, mode)) {
Mel Gormanb969c4a2012-01-12 17:19:34 -0800308 page_unfreeze_refs(page, expected_count);
309 spin_unlock_irq(&mapping->tree_lock);
310 return -EAGAIN;
311 }
312
313 /*
Christoph Lameterb20a3502006-03-22 00:09:12 -0800314 * Now we know that no one else is looking at the page.
Christoph Lameterb20a3502006-03-22 00:09:12 -0800315 */
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800316 get_page(newpage); /* add cache reference */
Christoph Lameterb20a3502006-03-22 00:09:12 -0800317 if (PageSwapCache(page)) {
318 SetPageSwapCache(newpage);
319 set_page_private(newpage, page_private(page));
320 }
321
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800322 radix_tree_replace_slot(pslot, newpage);
323
324 /*
Jacobo Giralt937a94c2012-01-10 15:07:11 -0800325 * Drop cache reference from old page by unfreezing
326 * to one less reference.
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800327 * We know this isn't the last reference.
328 */
Jacobo Giralt937a94c2012-01-10 15:07:11 -0800329 page_unfreeze_refs(page, expected_count - 1);
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800330
Christoph Lameter0e8c7d02007-04-23 14:41:09 -0700331 /*
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.
336 *
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.
340 */
341 __dec_zone_page_state(page, NR_FILE_PAGES);
342 __inc_zone_page_state(newpage, NR_FILE_PAGES);
Andrea Arcangeli99a15e22011-06-16 12:56:19 -0700343 if (!PageSwapCache(page) && PageSwapBacked(page)) {
KOSAKI Motohiro4b021082009-09-21 17:01:33 -0700344 __dec_zone_page_state(page, NR_SHMEM);
345 __inc_zone_page_state(newpage, NR_SHMEM);
346 }
Nick Piggin19fd6232008-07-25 19:45:32 -0700347 spin_unlock_irq(&mapping->tree_lock);
Christoph Lameterb20a3502006-03-22 00:09:12 -0800348
349 return 0;
350}
Christoph Lameterb20a3502006-03-22 00:09:12 -0800351
352/*
Naoya Horiguchi290408d2010-09-08 10:19:35 +0900353 * The expected number of remaining references is the same as that
354 * of migrate_page_move_mapping().
355 */
356int migrate_huge_page_move_mapping(struct address_space *mapping,
357 struct page *newpage, struct page *page)
358{
359 int expected_count;
360 void **pslot;
361
362 if (!mapping) {
363 if (page_count(page) != 1)
364 return -EAGAIN;
365 return 0;
366 }
367
368 spin_lock_irq(&mapping->tree_lock);
369
370 pslot = radix_tree_lookup_slot(&mapping->page_tree,
371 page_index(page));
372
373 expected_count = 2 + page_has_private(page);
374 if (page_count(page) != expected_count ||
Mel Gorman29c1f672011-01-13 15:47:21 -0800375 radix_tree_deref_slot_protected(pslot, &mapping->tree_lock) != page) {
Naoya Horiguchi290408d2010-09-08 10:19:35 +0900376 spin_unlock_irq(&mapping->tree_lock);
377 return -EAGAIN;
378 }
379
380 if (!page_freeze_refs(page, expected_count)) {
381 spin_unlock_irq(&mapping->tree_lock);
382 return -EAGAIN;
383 }
384
385 get_page(newpage);
386
387 radix_tree_replace_slot(pslot, newpage);
388
Jacobo Giralt937a94c2012-01-10 15:07:11 -0800389 page_unfreeze_refs(page, expected_count - 1);
Naoya Horiguchi290408d2010-09-08 10:19:35 +0900390
391 spin_unlock_irq(&mapping->tree_lock);
392 return 0;
393}
394
395/*
Christoph Lameterb20a3502006-03-22 00:09:12 -0800396 * Copy the page to its new location
397 */
Naoya Horiguchi290408d2010-09-08 10:19:35 +0900398void migrate_page_copy(struct page *newpage, struct page *page)
Christoph Lameterb20a3502006-03-22 00:09:12 -0800399{
Naoya Horiguchi290408d2010-09-08 10:19:35 +0900400 if (PageHuge(page))
401 copy_huge_page(newpage, page);
402 else
403 copy_highpage(newpage, page);
Christoph Lameterb20a3502006-03-22 00:09:12 -0800404
405 if (PageError(page))
406 SetPageError(newpage);
407 if (PageReferenced(page))
408 SetPageReferenced(newpage);
409 if (PageUptodate(page))
410 SetPageUptodate(newpage);
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700411 if (TestClearPageActive(page)) {
412 VM_BUG_ON(PageUnevictable(page));
Christoph Lameterb20a3502006-03-22 00:09:12 -0800413 SetPageActive(newpage);
Lee Schermerhorn418b27e2009-12-14 17:59:54 -0800414 } else if (TestClearPageUnevictable(page))
415 SetPageUnevictable(newpage);
Christoph Lameterb20a3502006-03-22 00:09:12 -0800416 if (PageChecked(page))
417 SetPageChecked(newpage);
418 if (PageMappedToDisk(page))
419 SetPageMappedToDisk(newpage);
420
421 if (PageDirty(page)) {
422 clear_page_dirty_for_io(page);
Nick Piggin3a902c52008-04-30 00:55:16 -0700423 /*
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.
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300428 * Whereas only part of our page may be dirty.
Nick Piggin3a902c52008-04-30 00:55:16 -0700429 */
Hugh Dickins752dc182012-06-02 00:27:47 -0700430 if (PageSwapBacked(page))
431 SetPageDirty(newpage);
432 else
433 __set_page_dirty_nobuffers(newpage);
Christoph Lameterb20a3502006-03-22 00:09:12 -0800434 }
435
Nick Pigginb291f002008-10-18 20:26:44 -0700436 mlock_migrate_page(newpage, page);
Hugh Dickinse9995ef2009-12-14 17:59:31 -0800437 ksm_migrate_page(newpage, page);
Nick Pigginb291f002008-10-18 20:26:44 -0700438
Christoph Lameterb20a3502006-03-22 00:09:12 -0800439 ClearPageSwapCache(page);
Christoph Lameterb20a3502006-03-22 00:09:12 -0800440 ClearPagePrivate(page);
441 set_page_private(page, 0);
Christoph Lameterb20a3502006-03-22 00:09:12 -0800442
443 /*
444 * If any waiters have accumulated on the new page then
445 * wake them up.
446 */
447 if (PageWriteback(newpage))
448 end_page_writeback(newpage);
449}
Christoph Lameterb20a3502006-03-22 00:09:12 -0800450
Christoph Lameter1d8b85c2006-06-23 02:03:28 -0700451/************************************************************
452 * Migration functions
453 ***********************************************************/
454
455/* Always fail migration. Used for mappings that are not movable */
Christoph Lameter2d1db3b2006-06-23 02:03:33 -0700456int fail_migrate_page(struct address_space *mapping,
457 struct page *newpage, struct page *page)
Christoph Lameter1d8b85c2006-06-23 02:03:28 -0700458{
459 return -EIO;
460}
461EXPORT_SYMBOL(fail_migrate_page);
462
Christoph Lameterb20a3502006-03-22 00:09:12 -0800463/*
464 * Common logic to directly migrate a single page suitable for
David Howells266cf652009-04-03 16:42:36 +0100465 * pages that do not use PagePrivate/PagePrivate2.
Christoph Lameterb20a3502006-03-22 00:09:12 -0800466 *
467 * Pages are locked upon entry and exit.
468 */
Christoph Lameter2d1db3b2006-06-23 02:03:33 -0700469int migrate_page(struct address_space *mapping,
Mel Gormana6bc32b2012-01-12 17:19:43 -0800470 struct page *newpage, struct page *page,
471 enum migrate_mode mode)
Christoph Lameterb20a3502006-03-22 00:09:12 -0800472{
473 int rc;
474
475 BUG_ON(PageWriteback(page)); /* Writeback must be complete */
476
Mel Gormana6bc32b2012-01-12 17:19:43 -0800477 rc = migrate_page_move_mapping(mapping, newpage, page, NULL, mode);
Christoph Lameterb20a3502006-03-22 00:09:12 -0800478
479 if (rc)
480 return rc;
481
482 migrate_page_copy(newpage, page);
Christoph Lameterb20a3502006-03-22 00:09:12 -0800483 return 0;
484}
485EXPORT_SYMBOL(migrate_page);
486
David Howells93614012006-09-30 20:45:40 +0200487#ifdef CONFIG_BLOCK
Christoph Lameterb20a3502006-03-22 00:09:12 -0800488/*
Christoph Lameter1d8b85c2006-06-23 02:03:28 -0700489 * Migration function for pages with buffers. This function can only be used
490 * if the underlying filesystem guarantees that no other references to "page"
491 * exist.
492 */
Christoph Lameter2d1db3b2006-06-23 02:03:33 -0700493int buffer_migrate_page(struct address_space *mapping,
Mel Gormana6bc32b2012-01-12 17:19:43 -0800494 struct page *newpage, struct page *page, enum migrate_mode mode)
Christoph Lameter1d8b85c2006-06-23 02:03:28 -0700495{
Christoph Lameter1d8b85c2006-06-23 02:03:28 -0700496 struct buffer_head *bh, *head;
497 int rc;
498
Christoph Lameter1d8b85c2006-06-23 02:03:28 -0700499 if (!page_has_buffers(page))
Mel Gormana6bc32b2012-01-12 17:19:43 -0800500 return migrate_page(mapping, newpage, page, mode);
Christoph Lameter1d8b85c2006-06-23 02:03:28 -0700501
502 head = page_buffers(page);
503
Mel Gormana6bc32b2012-01-12 17:19:43 -0800504 rc = migrate_page_move_mapping(mapping, newpage, page, head, mode);
Christoph Lameter1d8b85c2006-06-23 02:03:28 -0700505
506 if (rc)
507 return rc;
508
Mel Gormanb969c4a2012-01-12 17:19:34 -0800509 /*
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
513 */
Mel Gormana6bc32b2012-01-12 17:19:43 -0800514 if (mode != MIGRATE_ASYNC)
515 BUG_ON(!buffer_migrate_lock_buffers(head, mode));
Christoph Lameter1d8b85c2006-06-23 02:03:28 -0700516
517 ClearPagePrivate(page);
518 set_page_private(newpage, page_private(page));
519 set_page_private(page, 0);
520 put_page(page);
521 get_page(newpage);
522
523 bh = head;
524 do {
525 set_bh_page(bh, newpage, bh_offset(bh));
526 bh = bh->b_this_page;
527
528 } while (bh != head);
529
530 SetPagePrivate(newpage);
531
532 migrate_page_copy(newpage, page);
533
534 bh = head;
535 do {
536 unlock_buffer(bh);
537 put_bh(bh);
538 bh = bh->b_this_page;
539
540 } while (bh != head);
541
542 return 0;
543}
544EXPORT_SYMBOL(buffer_migrate_page);
David Howells93614012006-09-30 20:45:40 +0200545#endif
Christoph Lameter1d8b85c2006-06-23 02:03:28 -0700546
Christoph Lameter04e62a22006-06-23 02:03:38 -0700547/*
548 * Writeback a page to clean the dirty state
549 */
550static int writeout(struct address_space *mapping, struct page *page)
551{
552 struct writeback_control wbc = {
553 .sync_mode = WB_SYNC_NONE,
554 .nr_to_write = 1,
555 .range_start = 0,
556 .range_end = LLONG_MAX,
Christoph Lameter04e62a22006-06-23 02:03:38 -0700557 .for_reclaim = 1
558 };
559 int rc;
560
561 if (!mapping->a_ops->writepage)
562 /* No write method for the address space */
563 return -EINVAL;
564
565 if (!clear_page_dirty_for_io(page))
566 /* Someone else already triggered a write */
567 return -EAGAIN;
568
569 /*
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
575 * be successful.
576 */
577 remove_migration_ptes(page, page);
578
579 rc = mapping->a_ops->writepage(page, &wbc);
Christoph Lameter04e62a22006-06-23 02:03:38 -0700580
581 if (rc != AOP_WRITEPAGE_ACTIVATE)
582 /* unlocked. Relock */
583 lock_page(page);
584
Hugh Dickinsbda85502008-11-19 15:36:36 -0800585 return (rc < 0) ? -EIO : -EAGAIN;
Christoph Lameter04e62a22006-06-23 02:03:38 -0700586}
587
588/*
589 * Default handling if a filesystem does not provide a migration function.
590 */
Christoph Lameter8351a6e2006-06-23 02:03:33 -0700591static int fallback_migrate_page(struct address_space *mapping,
Mel Gormana6bc32b2012-01-12 17:19:43 -0800592 struct page *newpage, struct page *page, enum migrate_mode mode)
Christoph Lameter8351a6e2006-06-23 02:03:33 -0700593{
Mel Gormanb969c4a2012-01-12 17:19:34 -0800594 if (PageDirty(page)) {
Mel Gormana6bc32b2012-01-12 17:19:43 -0800595 /* Only writeback pages in full synchronous migration */
596 if (mode != MIGRATE_SYNC)
Mel Gormanb969c4a2012-01-12 17:19:34 -0800597 return -EBUSY;
Christoph Lameter04e62a22006-06-23 02:03:38 -0700598 return writeout(mapping, page);
Mel Gormanb969c4a2012-01-12 17:19:34 -0800599 }
Christoph Lameter8351a6e2006-06-23 02:03:33 -0700600
601 /*
602 * Buffers may be managed in a filesystem specific way.
603 * We must have no buffers or drop them.
604 */
David Howells266cf652009-04-03 16:42:36 +0100605 if (page_has_private(page) &&
Christoph Lameter8351a6e2006-06-23 02:03:33 -0700606 !try_to_release_page(page, GFP_KERNEL))
607 return -EAGAIN;
608
Mel Gormana6bc32b2012-01-12 17:19:43 -0800609 return migrate_page(mapping, newpage, page, mode);
Christoph Lameter8351a6e2006-06-23 02:03:33 -0700610}
611
Christoph Lameter1d8b85c2006-06-23 02:03:28 -0700612/*
Christoph Lametere24f0b82006-06-23 02:03:51 -0700613 * Move a page to a newly allocated page
614 * The page is locked and all ptes have been successfully removed.
615 *
616 * The new page will have replaced the old page if this function
617 * is successful.
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700618 *
619 * Return value:
620 * < 0 - error code
621 * == 0 - success
Christoph Lametere24f0b82006-06-23 02:03:51 -0700622 */
Mel Gorman3fe20112010-05-24 14:32:20 -0700623static int move_to_new_page(struct page *newpage, struct page *page,
Mel Gormana6bc32b2012-01-12 17:19:43 -0800624 int remap_swapcache, enum migrate_mode mode)
Christoph Lametere24f0b82006-06-23 02:03:51 -0700625{
626 struct address_space *mapping;
627 int rc;
628
629 /*
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.
633 */
Nick Piggin529ae9a2008-08-02 12:01:03 +0200634 if (!trylock_page(newpage))
Christoph Lametere24f0b82006-06-23 02:03:51 -0700635 BUG();
636
637 /* Prepare mapping for the new page.*/
638 newpage->index = page->index;
639 newpage->mapping = page->mapping;
Rik van Rielb2e18532008-10-18 20:26:30 -0700640 if (PageSwapBacked(page))
641 SetPageSwapBacked(newpage);
Christoph Lametere24f0b82006-06-23 02:03:51 -0700642
643 mapping = page_mapping(page);
644 if (!mapping)
Mel Gormana6bc32b2012-01-12 17:19:43 -0800645 rc = migrate_page(mapping, newpage, page, mode);
Mel Gormanb969c4a2012-01-12 17:19:34 -0800646 else if (mapping->a_ops->migratepage)
Christoph Lametere24f0b82006-06-23 02:03:51 -0700647 /*
Mel Gormanb969c4a2012-01-12 17:19:34 -0800648 * 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.
Christoph Lametere24f0b82006-06-23 02:03:51 -0700652 */
Mel Gormanb969c4a2012-01-12 17:19:34 -0800653 rc = mapping->a_ops->migratepage(mapping,
Mel Gormana6bc32b2012-01-12 17:19:43 -0800654 newpage, page, mode);
Mel Gormanb969c4a2012-01-12 17:19:34 -0800655 else
Mel Gormana6bc32b2012-01-12 17:19:43 -0800656 rc = fallback_migrate_page(mapping, newpage, page, mode);
Christoph Lametere24f0b82006-06-23 02:03:51 -0700657
Mel Gorman3fe20112010-05-24 14:32:20 -0700658 if (rc) {
Christoph Lametere24f0b82006-06-23 02:03:51 -0700659 newpage->mapping = NULL;
Mel Gorman3fe20112010-05-24 14:32:20 -0700660 } else {
661 if (remap_swapcache)
662 remove_migration_ptes(page, newpage);
Konstantin Khlebnikov35512ec2012-02-03 15:37:13 -0800663 page->mapping = NULL;
Mel Gorman3fe20112010-05-24 14:32:20 -0700664 }
Christoph Lametere24f0b82006-06-23 02:03:51 -0700665
666 unlock_page(newpage);
667
668 return rc;
669}
670
Minchan Kim0dabec92011-10-31 17:06:57 -0700671static int __unmap_and_move(struct page *page, struct page *newpage,
Mel Gormana6bc32b2012-01-12 17:19:43 -0800672 int force, bool offlining, enum migrate_mode mode)
Christoph Lametere24f0b82006-06-23 02:03:51 -0700673{
Minchan Kim0dabec92011-10-31 17:06:57 -0700674 int rc = -EAGAIN;
Mel Gorman3fe20112010-05-24 14:32:20 -0700675 int remap_swapcache = 1;
KAMEZAWA Hiroyuki56039ef2011-03-23 16:42:19 -0700676 struct mem_cgroup *mem;
Mel Gorman3f6c8272010-05-24 14:32:17 -0700677 struct anon_vma *anon_vma = NULL;
Christoph Lameter95a402c2006-06-23 02:03:53 -0700678
Nick Piggin529ae9a2008-08-02 12:01:03 +0200679 if (!trylock_page(page)) {
Mel Gormana6bc32b2012-01-12 17:19:43 -0800680 if (!force || mode == MIGRATE_ASYNC)
Minchan Kim0dabec92011-10-31 17:06:57 -0700681 goto out;
Mel Gorman3e7d3442011-01-13 15:45:56 -0800682
683 /*
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
694 * altogether.
695 */
696 if (current->flags & PF_MEMALLOC)
Minchan Kim0dabec92011-10-31 17:06:57 -0700697 goto out;
Mel Gorman3e7d3442011-01-13 15:45:56 -0800698
Christoph Lametere24f0b82006-06-23 02:03:51 -0700699 lock_page(page);
700 }
701
Hugh Dickins62b61f62009-12-14 17:59:33 -0800702 /*
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
709 * serializes that).
710 */
711 if (PageKsm(page) && !offlining) {
712 rc = -EBUSY;
713 goto unlock;
714 }
715
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -0800716 /* charge against new page */
Johannes Weiner0030f532012-07-31 16:45:25 -0700717 mem_cgroup_prepare_migration(page, newpage, &mem);
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -0800718
Christoph Lametere24f0b82006-06-23 02:03:51 -0700719 if (PageWriteback(page)) {
Andrea Arcangeli11bc82d2011-03-22 16:33:11 -0700720 /*
Mel Gormana6bc32b2012-01-12 17:19:43 -0800721 * 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
Andrea Arcangeli11bc82d2011-03-22 16:33:11 -0700725 */
Mel Gormana6bc32b2012-01-12 17:19:43 -0800726 if (mode != MIGRATE_SYNC) {
Andrea Arcangeli11bc82d2011-03-22 16:33:11 -0700727 rc = -EBUSY;
728 goto uncharge;
729 }
730 if (!force)
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -0800731 goto uncharge;
Christoph Lametere24f0b82006-06-23 02:03:51 -0700732 wait_on_page_writeback(page);
733 }
Christoph Lametere24f0b82006-06-23 02:03:51 -0700734 /*
KAMEZAWA Hiroyukidc386d42007-07-26 10:41:07 -0700735 * 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.
Hugh Dickins1ce82b62011-01-13 15:47:30 -0800737 * This get_anon_vma() delays freeing anon_vma pointer until the end
KAMEZAWA Hiroyukidc386d42007-07-26 10:41:07 -0700738 * of migration. File cache pages are no problem because of page_lock()
KAMEZAWA Hiroyuki989f89c2007-08-30 23:56:21 -0700739 * File Caches may use write_page() or lock_page() in migration, then,
740 * just care Anon page here.
Christoph Lametere24f0b82006-06-23 02:03:51 -0700741 */
KAMEZAWA Hiroyuki989f89c2007-08-30 23:56:21 -0700742 if (PageAnon(page)) {
Hugh Dickins1ce82b62011-01-13 15:47:30 -0800743 /*
744 * Only page_lock_anon_vma() understands the subtleties of
745 * getting a hold on an anon_vma from outside one of its mms.
746 */
Peter Zijlstra746b18d2011-05-24 17:12:10 -0700747 anon_vma = page_get_anon_vma(page);
Hugh Dickins1ce82b62011-01-13 15:47:30 -0800748 if (anon_vma) {
749 /*
Peter Zijlstra746b18d2011-05-24 17:12:10 -0700750 * Anon page
Hugh Dickins1ce82b62011-01-13 15:47:30 -0800751 */
Hugh Dickins1ce82b62011-01-13 15:47:30 -0800752 } else if (PageSwapCache(page)) {
Mel Gorman3fe20112010-05-24 14:32:20 -0700753 /*
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.
760 *
761 * To avoid this possibility, swapcache pages get
762 * migrated but are not remapped when migration
763 * completes
764 */
765 remap_swapcache = 0;
766 } else {
Hugh Dickins1ce82b62011-01-13 15:47:30 -0800767 goto uncharge;
Mel Gorman3fe20112010-05-24 14:32:20 -0700768 }
KAMEZAWA Hiroyuki989f89c2007-08-30 23:56:21 -0700769 }
Shaohua Li62e1c552008-02-04 22:29:33 -0800770
KAMEZAWA Hiroyukidc386d42007-07-26 10:41:07 -0700771 /*
Shaohua Li62e1c552008-02-04 22:29:33 -0800772 * 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.
KAMEZAWA Hiroyukidc386d42007-07-26 10:41:07 -0700782 */
Shaohua Li62e1c552008-02-04 22:29:33 -0800783 if (!page->mapping) {
Hugh Dickins1ce82b62011-01-13 15:47:30 -0800784 VM_BUG_ON(PageAnon(page));
785 if (page_has_private(page)) {
Shaohua Li62e1c552008-02-04 22:29:33 -0800786 try_to_free_buffers(page);
Hugh Dickins1ce82b62011-01-13 15:47:30 -0800787 goto uncharge;
Shaohua Li62e1c552008-02-04 22:29:33 -0800788 }
Shaohua Liabfc3482009-09-21 17:01:19 -0700789 goto skip_unmap;
Shaohua Li62e1c552008-02-04 22:29:33 -0800790 }
791
KAMEZAWA Hiroyukidc386d42007-07-26 10:41:07 -0700792 /* Establish migration ptes or remove ptes */
Andi Kleen14fa31b2009-09-16 11:50:10 +0200793 try_to_unmap(page, TTU_MIGRATION|TTU_IGNORE_MLOCK|TTU_IGNORE_ACCESS);
KAMEZAWA Hiroyukidc386d42007-07-26 10:41:07 -0700794
Shaohua Liabfc3482009-09-21 17:01:19 -0700795skip_unmap:
Christoph Lametere6a15302006-06-25 05:46:49 -0700796 if (!page_mapped(page))
Mel Gormana6bc32b2012-01-12 17:19:43 -0800797 rc = move_to_new_page(newpage, page, remap_swapcache, mode);
Christoph Lametere24f0b82006-06-23 02:03:51 -0700798
Mel Gorman3fe20112010-05-24 14:32:20 -0700799 if (rc && remap_swapcache)
Christoph Lametere24f0b82006-06-23 02:03:51 -0700800 remove_migration_ptes(page, page);
Mel Gorman3f6c8272010-05-24 14:32:17 -0700801
802 /* Drop an anon_vma reference if we took one */
Rik van Riel76545062010-08-09 17:18:41 -0700803 if (anon_vma)
Peter Zijlstra9e601092011-03-22 16:32:46 -0700804 put_anon_vma(anon_vma);
Mel Gorman3f6c8272010-05-24 14:32:17 -0700805
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -0800806uncharge:
Johannes Weiner0030f532012-07-31 16:45:25 -0700807 mem_cgroup_end_migration(mem, page, newpage, rc == 0);
Christoph Lametere24f0b82006-06-23 02:03:51 -0700808unlock:
809 unlock_page(page);
Minchan Kim0dabec92011-10-31 17:06:57 -0700810out:
811 return rc;
812}
Christoph Lameter95a402c2006-06-23 02:03:53 -0700813
Minchan Kim0dabec92011-10-31 17:06:57 -0700814/*
815 * Obtain the lock on page, remove all ptes and migrate the page
816 * to the newly allocated page in newpage.
817 */
818static int unmap_and_move(new_page_t get_new_page, unsigned long private,
Mel Gormana6bc32b2012-01-12 17:19:43 -0800819 struct page *page, int force, bool offlining,
820 enum migrate_mode mode)
Minchan Kim0dabec92011-10-31 17:06:57 -0700821{
822 int rc = 0;
823 int *result = NULL;
824 struct page *newpage = get_new_page(page, private, &result);
825
826 if (!newpage)
827 return -ENOMEM;
828
829 if (page_count(page) == 1) {
830 /* page was freed from under us. So we are done. */
831 goto out;
832 }
833
834 if (unlikely(PageTransHuge(page)))
835 if (unlikely(split_huge_page(page)))
836 goto out;
837
Mel Gormana6bc32b2012-01-12 17:19:43 -0800838 rc = __unmap_and_move(page, newpage, force, offlining, mode);
Minchan Kim0dabec92011-10-31 17:06:57 -0700839out:
Christoph Lametere24f0b82006-06-23 02:03:51 -0700840 if (rc != -EAGAIN) {
Minchan Kim0dabec92011-10-31 17:06:57 -0700841 /*
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
845 * restored.
846 */
847 list_del(&page->lru);
KOSAKI Motohiroa7312862009-09-21 17:01:37 -0700848 dec_zone_page_state(page, NR_ISOLATED_ANON +
Johannes Weiner6c0b1352009-09-21 17:02:59 -0700849 page_is_file_cache(page));
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700850 putback_lru_page(page);
Christoph Lametere24f0b82006-06-23 02:03:51 -0700851 }
Christoph Lameter95a402c2006-06-23 02:03:53 -0700852 /*
853 * Move the new page to the LRU. If migration was not successful
854 * then this will free the page.
855 */
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700856 putback_lru_page(newpage);
Christoph Lameter742755a2006-06-23 02:03:55 -0700857 if (result) {
858 if (rc)
859 *result = rc;
860 else
861 *result = page_to_nid(newpage);
862 }
Christoph Lametere24f0b82006-06-23 02:03:51 -0700863 return rc;
864}
865
866/*
Naoya Horiguchi290408d2010-09-08 10:19:35 +0900867 * Counterpart of unmap_and_move_page() for hugepage migration.
868 *
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.
879 *
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.
883 */
884static int unmap_and_move_huge_page(new_page_t get_new_page,
885 unsigned long private, struct page *hpage,
Mel Gormana6bc32b2012-01-12 17:19:43 -0800886 int force, bool offlining,
887 enum migrate_mode mode)
Naoya Horiguchi290408d2010-09-08 10:19:35 +0900888{
889 int rc = 0;
890 int *result = NULL;
891 struct page *new_hpage = get_new_page(hpage, private, &result);
Naoya Horiguchi290408d2010-09-08 10:19:35 +0900892 struct anon_vma *anon_vma = NULL;
893
894 if (!new_hpage)
895 return -ENOMEM;
896
897 rc = -EAGAIN;
898
899 if (!trylock_page(hpage)) {
Mel Gormana6bc32b2012-01-12 17:19:43 -0800900 if (!force || mode != MIGRATE_SYNC)
Naoya Horiguchi290408d2010-09-08 10:19:35 +0900901 goto out;
902 lock_page(hpage);
903 }
904
Peter Zijlstra746b18d2011-05-24 17:12:10 -0700905 if (PageAnon(hpage))
906 anon_vma = page_get_anon_vma(hpage);
Naoya Horiguchi290408d2010-09-08 10:19:35 +0900907
908 try_to_unmap(hpage, TTU_MIGRATION|TTU_IGNORE_MLOCK|TTU_IGNORE_ACCESS);
909
910 if (!page_mapped(hpage))
Mel Gormana6bc32b2012-01-12 17:19:43 -0800911 rc = move_to_new_page(new_hpage, hpage, 1, mode);
Naoya Horiguchi290408d2010-09-08 10:19:35 +0900912
913 if (rc)
914 remove_migration_ptes(hpage, hpage);
915
Hugh Dickinsfd4a4662011-01-13 15:47:31 -0800916 if (anon_vma)
Peter Zijlstra9e601092011-03-22 16:32:46 -0700917 put_anon_vma(anon_vma);
Aneesh Kumar K.V8e6ac7f2012-07-31 16:42:27 -0700918
919 if (!rc)
920 hugetlb_cgroup_migrate(hpage, new_hpage);
921
Naoya Horiguchi290408d2010-09-08 10:19:35 +0900922 unlock_page(hpage);
Hillf Danton09761332011-12-08 14:34:20 -0800923out:
Naoya Horiguchi290408d2010-09-08 10:19:35 +0900924 put_page(new_hpage);
Naoya Horiguchi290408d2010-09-08 10:19:35 +0900925 if (result) {
926 if (rc)
927 *result = rc;
928 else
929 *result = page_to_nid(new_hpage);
930 }
931 return rc;
932}
933
934/*
Christoph Lameterb20a3502006-03-22 00:09:12 -0800935 * migrate_pages
936 *
Christoph Lameter95a402c2006-06-23 02:03:53 -0700937 * 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.
Christoph Lameterb20a3502006-03-22 00:09:12 -0800940 *
941 * The function returns after 10 attempts or if no pages
942 * are movable anymore because to has become empty
Minchan Kimcf608ac2010-10-26 14:21:29 -0700943 * or no retryable pages exist anymore.
944 * Caller should call putback_lru_pages to return pages to the LRU
Minchan Kim28bd6572011-01-25 15:07:26 -0800945 * or free list only if ret != 0.
Christoph Lameterb20a3502006-03-22 00:09:12 -0800946 *
Christoph Lameter95a402c2006-06-23 02:03:53 -0700947 * Return: Number of pages not migrated or error code.
Christoph Lameterb20a3502006-03-22 00:09:12 -0800948 */
Christoph Lameter95a402c2006-06-23 02:03:53 -0700949int migrate_pages(struct list_head *from,
Mel Gorman7f0f2492011-01-13 15:45:58 -0800950 new_page_t get_new_page, unsigned long private, bool offlining,
Mel Gormana6bc32b2012-01-12 17:19:43 -0800951 enum migrate_mode mode)
Christoph Lameterb20a3502006-03-22 00:09:12 -0800952{
Christoph Lametere24f0b82006-06-23 02:03:51 -0700953 int retry = 1;
Christoph Lameterb20a3502006-03-22 00:09:12 -0800954 int nr_failed = 0;
955 int pass = 0;
956 struct page *page;
957 struct page *page2;
958 int swapwrite = current->flags & PF_SWAPWRITE;
959 int rc;
960
961 if (!swapwrite)
962 current->flags |= PF_SWAPWRITE;
963
Christoph Lametere24f0b82006-06-23 02:03:51 -0700964 for(pass = 0; pass < 10 && retry; pass++) {
965 retry = 0;
Christoph Lameterb20a3502006-03-22 00:09:12 -0800966
Christoph Lametere24f0b82006-06-23 02:03:51 -0700967 list_for_each_entry_safe(page, page2, from, lru) {
Christoph Lametere24f0b82006-06-23 02:03:51 -0700968 cond_resched();
Christoph Lameterb20a3502006-03-22 00:09:12 -0800969
Christoph Lameter95a402c2006-06-23 02:03:53 -0700970 rc = unmap_and_move(get_new_page, private,
Mel Gorman77f1fe62011-01-13 15:45:57 -0800971 page, pass > 2, offlining,
Mel Gormana6bc32b2012-01-12 17:19:43 -0800972 mode);
Christoph Lameterb20a3502006-03-22 00:09:12 -0800973
Christoph Lametere24f0b82006-06-23 02:03:51 -0700974 switch(rc) {
Christoph Lameter95a402c2006-06-23 02:03:53 -0700975 case -ENOMEM:
976 goto out;
Christoph Lametere24f0b82006-06-23 02:03:51 -0700977 case -EAGAIN:
Christoph Lameter2d1db3b2006-06-23 02:03:33 -0700978 retry++;
Christoph Lametere24f0b82006-06-23 02:03:51 -0700979 break;
980 case 0:
Christoph Lametere24f0b82006-06-23 02:03:51 -0700981 break;
982 default:
Christoph Lameter2d1db3b2006-06-23 02:03:33 -0700983 /* Permanent failure */
Christoph Lameter2d1db3b2006-06-23 02:03:33 -0700984 nr_failed++;
Christoph Lametere24f0b82006-06-23 02:03:51 -0700985 break;
Christoph Lameter2d1db3b2006-06-23 02:03:33 -0700986 }
Christoph Lameterb20a3502006-03-22 00:09:12 -0800987 }
988 }
Christoph Lameter95a402c2006-06-23 02:03:53 -0700989 rc = 0;
990out:
Christoph Lameterb20a3502006-03-22 00:09:12 -0800991 if (!swapwrite)
992 current->flags &= ~PF_SWAPWRITE;
993
Christoph Lameter95a402c2006-06-23 02:03:53 -0700994 if (rc)
995 return rc;
996
Christoph Lameterb20a3502006-03-22 00:09:12 -0800997 return nr_failed + retry;
998}
999
Aneesh Kumar K.V189ebff2012-07-31 16:42:06 -07001000int migrate_huge_page(struct page *hpage, new_page_t get_new_page,
1001 unsigned long private, bool offlining,
1002 enum migrate_mode mode)
Naoya Horiguchi290408d2010-09-08 10:19:35 +09001003{
Aneesh Kumar K.V189ebff2012-07-31 16:42:06 -07001004 int pass, rc;
Naoya Horiguchi290408d2010-09-08 10:19:35 +09001005
Aneesh Kumar K.V189ebff2012-07-31 16:42:06 -07001006 for (pass = 0; pass < 10; pass++) {
1007 rc = unmap_and_move_huge_page(get_new_page,
1008 private, hpage, pass > 2, offlining,
1009 mode);
1010 switch (rc) {
1011 case -ENOMEM:
1012 goto out;
1013 case -EAGAIN:
1014 /* try again */
Naoya Horiguchi290408d2010-09-08 10:19:35 +09001015 cond_resched();
Aneesh Kumar K.V189ebff2012-07-31 16:42:06 -07001016 break;
1017 case 0:
1018 goto out;
1019 default:
1020 rc = -EIO;
1021 goto out;
Naoya Horiguchi290408d2010-09-08 10:19:35 +09001022 }
1023 }
Naoya Horiguchi290408d2010-09-08 10:19:35 +09001024out:
Aneesh Kumar K.V189ebff2012-07-31 16:42:06 -07001025 return rc;
Naoya Horiguchi290408d2010-09-08 10:19:35 +09001026}
1027
Christoph Lameter742755a2006-06-23 02:03:55 -07001028#ifdef CONFIG_NUMA
1029/*
1030 * Move a list of individual pages
1031 */
1032struct page_to_node {
1033 unsigned long addr;
1034 struct page *page;
1035 int node;
1036 int status;
1037};
1038
1039static struct page *new_page_node(struct page *p, unsigned long private,
1040 int **result)
1041{
1042 struct page_to_node *pm = (struct page_to_node *)private;
1043
1044 while (pm->node != MAX_NUMNODES && pm->page != p)
1045 pm++;
1046
1047 if (pm->node == MAX_NUMNODES)
1048 return NULL;
1049
1050 *result = &pm->status;
1051
Mel Gorman6484eb32009-06-16 15:31:54 -07001052 return alloc_pages_exact_node(pm->node,
Mel Gorman769848c2007-07-17 04:03:05 -07001053 GFP_HIGHUSER_MOVABLE | GFP_THISNODE, 0);
Christoph Lameter742755a2006-06-23 02:03:55 -07001054}
1055
1056/*
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.
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001060 * The pm array ends with node = MAX_NUMNODES.
Christoph Lameter742755a2006-06-23 02:03:55 -07001061 */
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001062static int do_move_page_to_node_array(struct mm_struct *mm,
1063 struct page_to_node *pm,
1064 int migrate_all)
Christoph Lameter742755a2006-06-23 02:03:55 -07001065{
1066 int err;
1067 struct page_to_node *pp;
1068 LIST_HEAD(pagelist);
1069
1070 down_read(&mm->mmap_sem);
1071
1072 /*
1073 * Build a list of pages to migrate
1074 */
Christoph Lameter742755a2006-06-23 02:03:55 -07001075 for (pp = pm; pp->node != MAX_NUMNODES; pp++) {
1076 struct vm_area_struct *vma;
1077 struct page *page;
1078
Christoph Lameter742755a2006-06-23 02:03:55 -07001079 err = -EFAULT;
1080 vma = find_vma(mm, pp->addr);
Gleb Natapov70384dc2010-10-26 14:22:07 -07001081 if (!vma || pp->addr < vma->vm_start || !vma_migratable(vma))
Christoph Lameter742755a2006-06-23 02:03:55 -07001082 goto set_status;
1083
Andrea Arcangeli500d65d2011-01-13 15:46:55 -08001084 page = follow_page(vma, pp->addr, FOLL_GET|FOLL_SPLIT);
Linus Torvalds89f5b7d2008-06-20 11:18:25 -07001085
1086 err = PTR_ERR(page);
1087 if (IS_ERR(page))
1088 goto set_status;
1089
Christoph Lameter742755a2006-06-23 02:03:55 -07001090 err = -ENOENT;
1091 if (!page)
1092 goto set_status;
1093
Hugh Dickins62b61f62009-12-14 17:59:33 -08001094 /* Use PageReserved to check for zero page */
1095 if (PageReserved(page) || PageKsm(page))
Christoph Lameter742755a2006-06-23 02:03:55 -07001096 goto put_and_set;
1097
1098 pp->page = page;
1099 err = page_to_nid(page);
1100
1101 if (err == pp->node)
1102 /*
1103 * Node already in the right place
1104 */
1105 goto put_and_set;
1106
1107 err = -EACCES;
1108 if (page_mapcount(page) > 1 &&
1109 !migrate_all)
1110 goto put_and_set;
1111
Nick Piggin62695a82008-10-18 20:26:09 -07001112 err = isolate_lru_page(page);
KOSAKI Motohiro6d9c2852009-12-14 17:58:11 -08001113 if (!err) {
Nick Piggin62695a82008-10-18 20:26:09 -07001114 list_add_tail(&page->lru, &pagelist);
KOSAKI Motohiro6d9c2852009-12-14 17:58:11 -08001115 inc_zone_page_state(page, NR_ISOLATED_ANON +
1116 page_is_file_cache(page));
1117 }
Christoph Lameter742755a2006-06-23 02:03:55 -07001118put_and_set:
1119 /*
1120 * Either remove the duplicate refcount from
1121 * isolate_lru_page() or drop the page ref if it was
1122 * not isolated.
1123 */
1124 put_page(page);
1125set_status:
1126 pp->status = err;
1127 }
1128
Brice Gogline78bbfa2008-10-18 20:27:15 -07001129 err = 0;
Minchan Kimcf608ac2010-10-26 14:21:29 -07001130 if (!list_empty(&pagelist)) {
Christoph Lameter742755a2006-06-23 02:03:55 -07001131 err = migrate_pages(&pagelist, new_page_node,
Mel Gormana6bc32b2012-01-12 17:19:43 -08001132 (unsigned long)pm, 0, MIGRATE_SYNC);
Minchan Kimcf608ac2010-10-26 14:21:29 -07001133 if (err)
1134 putback_lru_pages(&pagelist);
1135 }
Christoph Lameter742755a2006-06-23 02:03:55 -07001136
1137 up_read(&mm->mmap_sem);
1138 return err;
1139}
1140
1141/*
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001142 * Migrate an array of page address onto an array of nodes and fill
1143 * the corresponding array of status.
1144 */
Christoph Lameter3268c632012-03-21 16:34:06 -07001145static int do_pages_move(struct mm_struct *mm, nodemask_t task_nodes,
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001146 unsigned long nr_pages,
1147 const void __user * __user *pages,
1148 const int __user *nodes,
1149 int __user *status, int flags)
1150{
Brice Goglin3140a222009-01-06 14:38:57 -08001151 struct page_to_node *pm;
Brice Goglin3140a222009-01-06 14:38:57 -08001152 unsigned long chunk_nr_pages;
1153 unsigned long chunk_start;
1154 int err;
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001155
Brice Goglin3140a222009-01-06 14:38:57 -08001156 err = -ENOMEM;
1157 pm = (struct page_to_node *)__get_free_page(GFP_KERNEL);
1158 if (!pm)
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001159 goto out;
Brice Goglin35282a22009-06-16 15:32:43 -07001160
1161 migrate_prep();
1162
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001163 /*
Brice Goglin3140a222009-01-06 14:38:57 -08001164 * Store a chunk of page_to_node array in a page,
1165 * but keep the last one as a marker
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001166 */
Brice Goglin3140a222009-01-06 14:38:57 -08001167 chunk_nr_pages = (PAGE_SIZE / sizeof(struct page_to_node)) - 1;
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001168
Brice Goglin3140a222009-01-06 14:38:57 -08001169 for (chunk_start = 0;
1170 chunk_start < nr_pages;
1171 chunk_start += chunk_nr_pages) {
1172 int j;
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001173
Brice Goglin3140a222009-01-06 14:38:57 -08001174 if (chunk_start + chunk_nr_pages > nr_pages)
1175 chunk_nr_pages = nr_pages - chunk_start;
1176
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;
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001180 int node;
1181
Brice Goglin3140a222009-01-06 14:38:57 -08001182 err = -EFAULT;
1183 if (get_user(p, pages + j + chunk_start))
1184 goto out_pm;
1185 pm[j].addr = (unsigned long) p;
1186
1187 if (get_user(node, nodes + j + chunk_start))
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001188 goto out_pm;
1189
1190 err = -ENODEV;
Linus Torvalds6f5a55f2010-02-05 16:16:50 -08001191 if (node < 0 || node >= MAX_NUMNODES)
1192 goto out_pm;
1193
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001194 if (!node_state(node, N_HIGH_MEMORY))
1195 goto out_pm;
1196
1197 err = -EACCES;
1198 if (!node_isset(node, task_nodes))
1199 goto out_pm;
1200
Brice Goglin3140a222009-01-06 14:38:57 -08001201 pm[j].node = node;
1202 }
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001203
Brice Goglin3140a222009-01-06 14:38:57 -08001204 /* End marker for this chunk */
1205 pm[chunk_nr_pages].node = MAX_NUMNODES;
1206
1207 /* Migrate this chunk */
1208 err = do_move_page_to_node_array(mm, pm,
1209 flags & MPOL_MF_MOVE_ALL);
1210 if (err < 0)
1211 goto out_pm;
1212
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001213 /* Return status information */
Brice Goglin3140a222009-01-06 14:38:57 -08001214 for (j = 0; j < chunk_nr_pages; j++)
1215 if (put_user(pm[j].status, status + j + chunk_start)) {
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001216 err = -EFAULT;
Brice Goglin3140a222009-01-06 14:38:57 -08001217 goto out_pm;
1218 }
1219 }
1220 err = 0;
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001221
1222out_pm:
Brice Goglin3140a222009-01-06 14:38:57 -08001223 free_page((unsigned long)pm);
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001224out:
1225 return err;
1226}
1227
1228/*
Brice Goglin2f007e72008-10-18 20:27:16 -07001229 * Determine the nodes of an array of pages and store it in an array of status.
Christoph Lameter742755a2006-06-23 02:03:55 -07001230 */
Brice Goglin80bba122008-12-09 13:14:23 -08001231static void do_pages_stat_array(struct mm_struct *mm, unsigned long nr_pages,
1232 const void __user **pages, int *status)
Christoph Lameter742755a2006-06-23 02:03:55 -07001233{
Brice Goglin2f007e72008-10-18 20:27:16 -07001234 unsigned long i;
Brice Goglin2f007e72008-10-18 20:27:16 -07001235
Christoph Lameter742755a2006-06-23 02:03:55 -07001236 down_read(&mm->mmap_sem);
1237
Brice Goglin2f007e72008-10-18 20:27:16 -07001238 for (i = 0; i < nr_pages; i++) {
Brice Goglin80bba122008-12-09 13:14:23 -08001239 unsigned long addr = (unsigned long)(*pages);
Christoph Lameter742755a2006-06-23 02:03:55 -07001240 struct vm_area_struct *vma;
1241 struct page *page;
KOSAKI Motohiroc095adb2008-12-16 16:06:43 +09001242 int err = -EFAULT;
Brice Goglin2f007e72008-10-18 20:27:16 -07001243
1244 vma = find_vma(mm, addr);
Gleb Natapov70384dc2010-10-26 14:22:07 -07001245 if (!vma || addr < vma->vm_start)
Christoph Lameter742755a2006-06-23 02:03:55 -07001246 goto set_status;
1247
Brice Goglin2f007e72008-10-18 20:27:16 -07001248 page = follow_page(vma, addr, 0);
Linus Torvalds89f5b7d2008-06-20 11:18:25 -07001249
1250 err = PTR_ERR(page);
1251 if (IS_ERR(page))
1252 goto set_status;
1253
Christoph Lameter742755a2006-06-23 02:03:55 -07001254 err = -ENOENT;
1255 /* Use PageReserved to check for zero page */
Hugh Dickins62b61f62009-12-14 17:59:33 -08001256 if (!page || PageReserved(page) || PageKsm(page))
Christoph Lameter742755a2006-06-23 02:03:55 -07001257 goto set_status;
1258
1259 err = page_to_nid(page);
1260set_status:
Brice Goglin80bba122008-12-09 13:14:23 -08001261 *status = err;
1262
1263 pages++;
1264 status++;
1265 }
1266
1267 up_read(&mm->mmap_sem);
1268}
1269
1270/*
1271 * Determine the nodes of a user array of pages and store it in
1272 * a user array of status.
1273 */
1274static int do_pages_stat(struct mm_struct *mm, unsigned long nr_pages,
1275 const void __user * __user *pages,
1276 int __user *status)
1277{
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];
Brice Goglin80bba122008-12-09 13:14:23 -08001281
H. Peter Anvin87b8d1a2010-02-18 16:13:40 -08001282 while (nr_pages) {
1283 unsigned long chunk_nr;
Brice Goglin80bba122008-12-09 13:14:23 -08001284
H. Peter Anvin87b8d1a2010-02-18 16:13:40 -08001285 chunk_nr = nr_pages;
1286 if (chunk_nr > DO_PAGES_STAT_CHUNK_NR)
1287 chunk_nr = DO_PAGES_STAT_CHUNK_NR;
1288
1289 if (copy_from_user(chunk_pages, pages, chunk_nr * sizeof(*chunk_pages)))
1290 break;
Brice Goglin80bba122008-12-09 13:14:23 -08001291
1292 do_pages_stat_array(mm, chunk_nr, chunk_pages, chunk_status);
1293
H. Peter Anvin87b8d1a2010-02-18 16:13:40 -08001294 if (copy_to_user(status, chunk_status, chunk_nr * sizeof(*status)))
1295 break;
Christoph Lameter742755a2006-06-23 02:03:55 -07001296
H. Peter Anvin87b8d1a2010-02-18 16:13:40 -08001297 pages += chunk_nr;
1298 status += chunk_nr;
1299 nr_pages -= chunk_nr;
1300 }
1301 return nr_pages ? -EFAULT : 0;
Christoph Lameter742755a2006-06-23 02:03:55 -07001302}
1303
1304/*
1305 * Move a list of pages in the address space of the currently executing
1306 * process.
1307 */
Heiko Carstens938bb9f2009-01-14 14:14:30 +01001308SYSCALL_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)
Christoph Lameter742755a2006-06-23 02:03:55 -07001312{
David Howellsc69e8d92008-11-14 10:39:19 +11001313 const struct cred *cred = current_cred(), *tcred;
Christoph Lameter742755a2006-06-23 02:03:55 -07001314 struct task_struct *task;
Christoph Lameter742755a2006-06-23 02:03:55 -07001315 struct mm_struct *mm;
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001316 int err;
Christoph Lameter3268c632012-03-21 16:34:06 -07001317 nodemask_t task_nodes;
Christoph Lameter742755a2006-06-23 02:03:55 -07001318
1319 /* Check flags */
1320 if (flags & ~(MPOL_MF_MOVE|MPOL_MF_MOVE_ALL))
1321 return -EINVAL;
1322
1323 if ((flags & MPOL_MF_MOVE_ALL) && !capable(CAP_SYS_NICE))
1324 return -EPERM;
1325
1326 /* Find the mm_struct */
Greg Thelena879bf52011-02-25 14:44:13 -08001327 rcu_read_lock();
Pavel Emelyanov228ebcb2007-10-18 23:40:16 -07001328 task = pid ? find_task_by_vpid(pid) : current;
Christoph Lameter742755a2006-06-23 02:03:55 -07001329 if (!task) {
Greg Thelena879bf52011-02-25 14:44:13 -08001330 rcu_read_unlock();
Christoph Lameter742755a2006-06-23 02:03:55 -07001331 return -ESRCH;
1332 }
Christoph Lameter3268c632012-03-21 16:34:06 -07001333 get_task_struct(task);
Christoph Lameter742755a2006-06-23 02:03:55 -07001334
1335 /*
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.
1340 */
David Howellsc69e8d92008-11-14 10:39:19 +11001341 tcred = __task_cred(task);
Eric W. Biedermanb38a86e2012-03-12 15:48:24 -07001342 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) &&
Christoph Lameter742755a2006-06-23 02:03:55 -07001344 !capable(CAP_SYS_NICE)) {
David Howellsc69e8d92008-11-14 10:39:19 +11001345 rcu_read_unlock();
Christoph Lameter742755a2006-06-23 02:03:55 -07001346 err = -EPERM;
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001347 goto out;
Christoph Lameter742755a2006-06-23 02:03:55 -07001348 }
David Howellsc69e8d92008-11-14 10:39:19 +11001349 rcu_read_unlock();
Christoph Lameter742755a2006-06-23 02:03:55 -07001350
David Quigley86c3a762006-06-23 02:04:02 -07001351 err = security_task_movememory(task);
1352 if (err)
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001353 goto out;
David Quigley86c3a762006-06-23 02:04:02 -07001354
Christoph Lameter3268c632012-03-21 16:34:06 -07001355 task_nodes = cpuset_mems_allowed(task);
1356 mm = get_task_mm(task);
1357 put_task_struct(task);
1358
Sasha Levin6e8b09e2012-04-25 16:01:53 -07001359 if (!mm)
1360 return -EINVAL;
1361
1362 if (nodes)
1363 err = do_pages_move(mm, task_nodes, nr_pages, pages,
1364 nodes, status, flags);
1365 else
1366 err = do_pages_stat(mm, nr_pages, pages, status);
Christoph Lameter3268c632012-03-21 16:34:06 -07001367
1368 mmput(mm);
1369 return err;
David Quigley86c3a762006-06-23 02:04:02 -07001370
Christoph Lameter742755a2006-06-23 02:03:55 -07001371out:
Christoph Lameter3268c632012-03-21 16:34:06 -07001372 put_task_struct(task);
Christoph Lameter742755a2006-06-23 02:03:55 -07001373 return err;
1374}
Christoph Lameter742755a2006-06-23 02:03:55 -07001375
Christoph Lameter7b2259b2006-06-25 05:46:48 -07001376/*
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.
1380 */
1381int migrate_vmas(struct mm_struct *mm, const nodemask_t *to,
1382 const nodemask_t *from, unsigned long flags)
1383{
1384 struct vm_area_struct *vma;
1385 int err = 0;
1386
Daisuke Nishimura1001c9f2009-02-11 13:04:18 -08001387 for (vma = mm->mmap; vma && !err; vma = vma->vm_next) {
Christoph Lameter7b2259b2006-06-25 05:46:48 -07001388 if (vma->vm_ops && vma->vm_ops->migrate) {
1389 err = vma->vm_ops->migrate(vma, to, from, flags);
1390 if (err)
1391 break;
1392 }
1393 }
1394 return err;
1395}
Gerald Schaefer83d16742008-07-23 21:28:22 -07001396#endif