blob: df141f60289eef61c7d278080e8290c3dfc717b9 [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>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090036#include <linux/gfp.h>
Christoph Lameterb20a3502006-03-22 00:09:12 -080037
Michal Nazarewicz0d1836c2010-12-21 17:24:26 -080038#include <asm/tlbflush.h>
39
Christoph Lameterb20a3502006-03-22 00:09:12 -080040#include "internal.h"
41
Christoph Lameterb20a3502006-03-22 00:09:12 -080042/*
Christoph Lameter742755a2006-06-23 02:03:55 -070043 * migrate_prep() needs to be called before we start compiling a list of pages
Mel Gorman748446b2010-05-24 14:32:27 -070044 * to be migrated using isolate_lru_page(). If scheduling work on other CPUs is
45 * undesirable, use migrate_prep_local()
Christoph Lameterb20a3502006-03-22 00:09:12 -080046 */
47int migrate_prep(void)
48{
Christoph Lameterb20a3502006-03-22 00:09:12 -080049 /*
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.
54 */
55 lru_add_drain_all();
56
57 return 0;
58}
59
Mel Gorman748446b2010-05-24 14:32:27 -070060/* Do the necessary work of migrate_prep but not if it involves other CPUs */
61int migrate_prep_local(void)
62{
63 lru_add_drain();
64
65 return 0;
66}
67
Christoph Lameterb20a3502006-03-22 00:09:12 -080068/*
Lee Schermerhorn894bc312008-10-18 20:26:39 -070069 * Add isolated pages on the list back to the LRU under page lock
70 * to avoid leaking evictable pages back onto unevictable list.
Christoph Lameterb20a3502006-03-22 00:09:12 -080071 */
Minchan Kime13861d2010-05-24 14:31:59 -070072void putback_lru_pages(struct list_head *l)
Christoph Lameterb20a3502006-03-22 00:09:12 -080073{
74 struct page *page;
75 struct page *page2;
Christoph Lameterb20a3502006-03-22 00:09:12 -080076
77 list_for_each_entry_safe(page, page2, l, lru) {
Christoph Lametere24f0b82006-06-23 02:03:51 -070078 list_del(&page->lru);
KOSAKI Motohiroa7312862009-09-21 17:01:37 -070079 dec_zone_page_state(page, NR_ISOLATED_ANON +
Johannes Weiner6c0b1352009-09-21 17:02:59 -070080 page_is_file_cache(page));
Lee Schermerhorn894bc312008-10-18 20:26:39 -070081 putback_lru_page(page);
Christoph Lameterb20a3502006-03-22 00:09:12 -080082 }
Christoph Lameterb20a3502006-03-22 00:09:12 -080083}
84
Christoph Lameter06972122006-06-23 02:03:35 -070085/*
86 * Restore a potential migration pte to a working pte entry
87 */
Hugh Dickinse9995ef2009-12-14 17:59:31 -080088static int remove_migration_pte(struct page *new, struct vm_area_struct *vma,
89 unsigned long addr, void *old)
Christoph Lameter06972122006-06-23 02:03:35 -070090{
91 struct mm_struct *mm = vma->vm_mm;
92 swp_entry_t entry;
93 pgd_t *pgd;
94 pud_t *pud;
95 pmd_t *pmd;
96 pte_t *ptep, pte;
97 spinlock_t *ptl;
98
Naoya Horiguchi290408d2010-09-08 10:19:35 +090099 if (unlikely(PageHuge(new))) {
100 ptep = huge_pte_offset(mm, addr);
101 if (!ptep)
102 goto out;
103 ptl = &mm->page_table_lock;
104 } else {
105 pgd = pgd_offset(mm, addr);
106 if (!pgd_present(*pgd))
107 goto out;
Christoph Lameter06972122006-06-23 02:03:35 -0700108
Naoya Horiguchi290408d2010-09-08 10:19:35 +0900109 pud = pud_offset(pgd, addr);
110 if (!pud_present(*pud))
111 goto out;
Christoph Lameter06972122006-06-23 02:03:35 -0700112
Naoya Horiguchi290408d2010-09-08 10:19:35 +0900113 pmd = pmd_offset(pud, addr);
Andrea Arcangeli500d65d2011-01-13 15:46:55 -0800114 if (pmd_trans_huge(*pmd))
115 goto out;
Naoya Horiguchi290408d2010-09-08 10:19:35 +0900116 if (!pmd_present(*pmd))
117 goto out;
Christoph Lameter06972122006-06-23 02:03:35 -0700118
Naoya Horiguchi290408d2010-09-08 10:19:35 +0900119 ptep = pte_offset_map(pmd, addr);
Christoph Lameter06972122006-06-23 02:03:35 -0700120
Hugh Dickins486cf462011-10-19 12:50:35 -0700121 /*
122 * Peek to check is_swap_pte() before taking ptlock? No, we
123 * can race mremap's move_ptes(), which skips anon_vma lock.
124 */
Christoph Lameter06972122006-06-23 02:03:35 -0700125
Naoya Horiguchi290408d2010-09-08 10:19:35 +0900126 ptl = pte_lockptr(mm, pmd);
127 }
128
Christoph Lameter06972122006-06-23 02:03:35 -0700129 spin_lock(ptl);
130 pte = *ptep;
131 if (!is_swap_pte(pte))
Hugh Dickinse9995ef2009-12-14 17:59:31 -0800132 goto unlock;
Christoph Lameter06972122006-06-23 02:03:35 -0700133
134 entry = pte_to_swp_entry(pte);
135
Hugh Dickinse9995ef2009-12-14 17:59:31 -0800136 if (!is_migration_entry(entry) ||
137 migration_entry_to_page(entry) != old)
138 goto unlock;
Christoph Lameter06972122006-06-23 02:03:35 -0700139
Christoph Lameter06972122006-06-23 02:03:35 -0700140 get_page(new);
141 pte = pte_mkold(mk_pte(new, vma->vm_page_prot));
142 if (is_write_migration_entry(entry))
143 pte = pte_mkwrite(pte);
Andi Kleen3ef8fd72010-10-11 16:03:21 +0200144#ifdef CONFIG_HUGETLB_PAGE
Naoya Horiguchi290408d2010-09-08 10:19:35 +0900145 if (PageHuge(new))
146 pte = pte_mkhuge(pte);
Andi Kleen3ef8fd72010-10-11 16:03:21 +0200147#endif
KAMEZAWA Hiroyuki97ee0522007-10-16 01:25:43 -0700148 flush_cache_page(vma, addr, pte_pfn(pte));
Christoph Lameter06972122006-06-23 02:03:35 -0700149 set_pte_at(mm, addr, ptep, pte);
Christoph Lameter04e62a22006-06-23 02:03:38 -0700150
Naoya Horiguchi290408d2010-09-08 10:19:35 +0900151 if (PageHuge(new)) {
152 if (PageAnon(new))
153 hugepage_add_anon_rmap(new, vma, addr);
154 else
155 page_dup_rmap(new);
156 } else if (PageAnon(new))
Christoph Lameter04e62a22006-06-23 02:03:38 -0700157 page_add_anon_rmap(new, vma, addr);
158 else
159 page_add_file_rmap(new);
160
161 /* No need to invalidate - it was non-present before */
Russell King4b3073e2009-12-18 16:40:18 +0000162 update_mmu_cache(vma, addr, ptep);
Hugh Dickinse9995ef2009-12-14 17:59:31 -0800163unlock:
Christoph Lameter06972122006-06-23 02:03:35 -0700164 pte_unmap_unlock(ptep, ptl);
Hugh Dickinse9995ef2009-12-14 17:59:31 -0800165out:
166 return SWAP_AGAIN;
Christoph Lameter06972122006-06-23 02:03:35 -0700167}
168
169/*
Christoph Lameter04e62a22006-06-23 02:03:38 -0700170 * Get rid of all migration entries and replace them by
171 * references to the indicated page.
172 */
173static void remove_migration_ptes(struct page *old, struct page *new)
174{
Hugh Dickinse9995ef2009-12-14 17:59:31 -0800175 rmap_walk(new, remove_migration_pte, old);
Christoph Lameter04e62a22006-06-23 02:03:38 -0700176}
177
178/*
Christoph Lameter06972122006-06-23 02:03:35 -0700179 * 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.
Christoph Lameter06972122006-06-23 02:03:35 -0700182 */
183void migration_entry_wait(struct mm_struct *mm, pmd_t *pmd,
184 unsigned long address)
185{
186 pte_t *ptep, pte;
187 spinlock_t *ptl;
188 swp_entry_t entry;
189 struct page *page;
190
191 ptep = pte_offset_map_lock(mm, pmd, address, &ptl);
192 pte = *ptep;
193 if (!is_swap_pte(pte))
194 goto out;
195
196 entry = pte_to_swp_entry(pte);
197 if (!is_migration_entry(entry))
198 goto out;
199
200 page = migration_entry_to_page(entry);
201
Nick Piggine2867812008-07-25 19:45:30 -0700202 /*
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
207 * will occur again.
208 */
209 if (!get_page_unless_zero(page))
210 goto out;
Christoph Lameter06972122006-06-23 02:03:35 -0700211 pte_unmap_unlock(ptep, ptl);
212 wait_on_page_locked(page);
213 put_page(page);
214 return;
215out:
216 pte_unmap_unlock(ptep, ptl);
217}
218
Mel Gormanb969c4a2012-01-12 17:19:34 -0800219#ifdef CONFIG_BLOCK
220/* Returns true if all buffers are successfully locked */
Mel Gormana6bc32b2012-01-12 17:19:43 -0800221static bool buffer_migrate_lock_buffers(struct buffer_head *head,
222 enum migrate_mode mode)
Mel Gormanb969c4a2012-01-12 17:19:34 -0800223{
224 struct buffer_head *bh = head;
225
226 /* Simple case, sync compaction */
Mel Gormana6bc32b2012-01-12 17:19:43 -0800227 if (mode != MIGRATE_ASYNC) {
Mel Gormanb969c4a2012-01-12 17:19:34 -0800228 do {
229 get_bh(bh);
230 lock_buffer(bh);
231 bh = bh->b_this_page;
232
233 } while (bh != head);
234
235 return true;
236 }
237
238 /* async case, we cannot block on lock_buffer so use trylock_buffer */
239 do {
240 get_bh(bh);
241 if (!trylock_buffer(bh)) {
242 /*
243 * We failed to lock the buffer and cannot stall in
244 * async migration. Release the taken locks
245 */
246 struct buffer_head *failed_bh = bh;
247 put_bh(failed_bh);
248 bh = head;
249 while (bh != failed_bh) {
250 unlock_buffer(bh);
251 put_bh(bh);
252 bh = bh->b_this_page;
253 }
254 return false;
255 }
256
257 bh = bh->b_this_page;
258 } while (bh != head);
259 return true;
260}
261#else
262static inline bool buffer_migrate_lock_buffers(struct buffer_head *head,
Mel Gormana6bc32b2012-01-12 17:19:43 -0800263 enum migrate_mode mode)
Mel Gormanb969c4a2012-01-12 17:19:34 -0800264{
265 return true;
266}
267#endif /* CONFIG_BLOCK */
268
Christoph Lameterb20a3502006-03-22 00:09:12 -0800269/*
Christoph Lameterc3fcf8a2006-06-23 02:03:32 -0700270 * Replace the page in the mapping.
Christoph Lameter5b5c7122006-06-23 02:03:29 -0700271 *
272 * The number of remaining references must be:
273 * 1 for anonymous pages without a mapping
274 * 2 for pages with a mapping
David Howells266cf652009-04-03 16:42:36 +0100275 * 3 for pages with a mapping and PagePrivate/PagePrivate2 set.
Christoph Lameterb20a3502006-03-22 00:09:12 -0800276 */
Christoph Lameter2d1db3b2006-06-23 02:03:33 -0700277static int migrate_page_move_mapping(struct address_space *mapping,
Mel Gormanb969c4a2012-01-12 17:19:34 -0800278 struct page *newpage, struct page *page,
Mel Gormana6bc32b2012-01-12 17:19:43 -0800279 struct buffer_head *head, enum migrate_mode mode)
Christoph Lameterb20a3502006-03-22 00:09:12 -0800280{
Nick Piggine2867812008-07-25 19:45:30 -0700281 int expected_count;
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800282 void **pslot;
Christoph Lameterb20a3502006-03-22 00:09:12 -0800283
Christoph Lameter6c5240a2006-06-23 02:03:37 -0700284 if (!mapping) {
Christoph Lameter0e8c7d02007-04-23 14:41:09 -0700285 /* Anonymous page without mapping */
Christoph Lameter6c5240a2006-06-23 02:03:37 -0700286 if (page_count(page) != 1)
287 return -EAGAIN;
288 return 0;
289 }
290
Nick Piggin19fd6232008-07-25 19:45:32 -0700291 spin_lock_irq(&mapping->tree_lock);
Christoph Lameterb20a3502006-03-22 00:09:12 -0800292
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800293 pslot = radix_tree_lookup_slot(&mapping->page_tree,
294 page_index(page));
Christoph Lameterb20a3502006-03-22 00:09:12 -0800295
Johannes Weineredcf4742009-09-21 17:02:59 -0700296 expected_count = 2 + page_has_private(page);
Nick Piggine2867812008-07-25 19:45:30 -0700297 if (page_count(page) != expected_count ||
Mel Gorman29c1f672011-01-13 15:47:21 -0800298 radix_tree_deref_slot_protected(pslot, &mapping->tree_lock) != page) {
Nick Piggin19fd6232008-07-25 19:45:32 -0700299 spin_unlock_irq(&mapping->tree_lock);
Christoph Lametere23ca002006-04-10 22:52:57 -0700300 return -EAGAIN;
Christoph Lameterb20a3502006-03-22 00:09:12 -0800301 }
302
Nick Piggine2867812008-07-25 19:45:30 -0700303 if (!page_freeze_refs(page, expected_count)) {
Nick Piggin19fd6232008-07-25 19:45:32 -0700304 spin_unlock_irq(&mapping->tree_lock);
Nick Piggine2867812008-07-25 19:45:30 -0700305 return -EAGAIN;
306 }
307
Christoph Lameterb20a3502006-03-22 00:09:12 -0800308 /*
Mel Gormanb969c4a2012-01-12 17:19:34 -0800309 * 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.
314 */
Mel Gormana6bc32b2012-01-12 17:19:43 -0800315 if (mode == MIGRATE_ASYNC && head &&
316 !buffer_migrate_lock_buffers(head, mode)) {
Mel Gormanb969c4a2012-01-12 17:19:34 -0800317 page_unfreeze_refs(page, expected_count);
318 spin_unlock_irq(&mapping->tree_lock);
319 return -EAGAIN;
320 }
321
322 /*
Christoph Lameterb20a3502006-03-22 00:09:12 -0800323 * Now we know that no one else is looking at the page.
Christoph Lameterb20a3502006-03-22 00:09:12 -0800324 */
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800325 get_page(newpage); /* add cache reference */
Christoph Lameterb20a3502006-03-22 00:09:12 -0800326 if (PageSwapCache(page)) {
327 SetPageSwapCache(newpage);
328 set_page_private(newpage, page_private(page));
329 }
330
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800331 radix_tree_replace_slot(pslot, newpage);
332
333 /*
Jacobo Giralt937a94c2012-01-10 15:07:11 -0800334 * Drop cache reference from old page by unfreezing
335 * to one less reference.
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800336 * We know this isn't the last reference.
337 */
Jacobo Giralt937a94c2012-01-10 15:07:11 -0800338 page_unfreeze_refs(page, expected_count - 1);
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800339
Christoph Lameter0e8c7d02007-04-23 14:41:09 -0700340 /*
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.
345 *
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.
349 */
350 __dec_zone_page_state(page, NR_FILE_PAGES);
351 __inc_zone_page_state(newpage, NR_FILE_PAGES);
Andrea Arcangeli99a15e22011-06-16 12:56:19 -0700352 if (!PageSwapCache(page) && PageSwapBacked(page)) {
KOSAKI Motohiro4b021082009-09-21 17:01:33 -0700353 __dec_zone_page_state(page, NR_SHMEM);
354 __inc_zone_page_state(newpage, NR_SHMEM);
355 }
Nick Piggin19fd6232008-07-25 19:45:32 -0700356 spin_unlock_irq(&mapping->tree_lock);
Christoph Lameterb20a3502006-03-22 00:09:12 -0800357
358 return 0;
359}
Christoph Lameterb20a3502006-03-22 00:09:12 -0800360
361/*
Naoya Horiguchi290408d2010-09-08 10:19:35 +0900362 * The expected number of remaining references is the same as that
363 * of migrate_page_move_mapping().
364 */
365int migrate_huge_page_move_mapping(struct address_space *mapping,
366 struct page *newpage, struct page *page)
367{
368 int expected_count;
369 void **pslot;
370
371 if (!mapping) {
372 if (page_count(page) != 1)
373 return -EAGAIN;
374 return 0;
375 }
376
377 spin_lock_irq(&mapping->tree_lock);
378
379 pslot = radix_tree_lookup_slot(&mapping->page_tree,
380 page_index(page));
381
382 expected_count = 2 + page_has_private(page);
383 if (page_count(page) != expected_count ||
Mel Gorman29c1f672011-01-13 15:47:21 -0800384 radix_tree_deref_slot_protected(pslot, &mapping->tree_lock) != page) {
Naoya Horiguchi290408d2010-09-08 10:19:35 +0900385 spin_unlock_irq(&mapping->tree_lock);
386 return -EAGAIN;
387 }
388
389 if (!page_freeze_refs(page, expected_count)) {
390 spin_unlock_irq(&mapping->tree_lock);
391 return -EAGAIN;
392 }
393
394 get_page(newpage);
395
396 radix_tree_replace_slot(pslot, newpage);
397
Jacobo Giralt937a94c2012-01-10 15:07:11 -0800398 page_unfreeze_refs(page, expected_count - 1);
Naoya Horiguchi290408d2010-09-08 10:19:35 +0900399
400 spin_unlock_irq(&mapping->tree_lock);
401 return 0;
402}
403
404/*
Christoph Lameterb20a3502006-03-22 00:09:12 -0800405 * Copy the page to its new location
406 */
Naoya Horiguchi290408d2010-09-08 10:19:35 +0900407void migrate_page_copy(struct page *newpage, struct page *page)
Christoph Lameterb20a3502006-03-22 00:09:12 -0800408{
Naoya Horiguchi290408d2010-09-08 10:19:35 +0900409 if (PageHuge(page))
410 copy_huge_page(newpage, page);
411 else
412 copy_highpage(newpage, page);
Christoph Lameterb20a3502006-03-22 00:09:12 -0800413
414 if (PageError(page))
415 SetPageError(newpage);
416 if (PageReferenced(page))
417 SetPageReferenced(newpage);
418 if (PageUptodate(page))
419 SetPageUptodate(newpage);
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700420 if (TestClearPageActive(page)) {
421 VM_BUG_ON(PageUnevictable(page));
Christoph Lameterb20a3502006-03-22 00:09:12 -0800422 SetPageActive(newpage);
Lee Schermerhorn418b27e2009-12-14 17:59:54 -0800423 } else if (TestClearPageUnevictable(page))
424 SetPageUnevictable(newpage);
Christoph Lameterb20a3502006-03-22 00:09:12 -0800425 if (PageChecked(page))
426 SetPageChecked(newpage);
427 if (PageMappedToDisk(page))
428 SetPageMappedToDisk(newpage);
429
430 if (PageDirty(page)) {
431 clear_page_dirty_for_io(page);
Nick Piggin3a902c52008-04-30 00:55:16 -0700432 /*
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.
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300437 * Whereas only part of our page may be dirty.
Nick Piggin3a902c52008-04-30 00:55:16 -0700438 */
439 __set_page_dirty_nobuffers(newpage);
Christoph Lameterb20a3502006-03-22 00:09:12 -0800440 }
441
Nick Pigginb291f002008-10-18 20:26:44 -0700442 mlock_migrate_page(newpage, page);
Hugh Dickinse9995ef2009-12-14 17:59:31 -0800443 ksm_migrate_page(newpage, page);
Nick Pigginb291f002008-10-18 20:26:44 -0700444
Christoph Lameterb20a3502006-03-22 00:09:12 -0800445 ClearPageSwapCache(page);
Christoph Lameterb20a3502006-03-22 00:09:12 -0800446 ClearPagePrivate(page);
447 set_page_private(page, 0);
Christoph Lameterb20a3502006-03-22 00:09:12 -0800448
449 /*
450 * If any waiters have accumulated on the new page then
451 * wake them up.
452 */
453 if (PageWriteback(newpage))
454 end_page_writeback(newpage);
455}
Christoph Lameterb20a3502006-03-22 00:09:12 -0800456
Christoph Lameter1d8b85c2006-06-23 02:03:28 -0700457/************************************************************
458 * Migration functions
459 ***********************************************************/
460
461/* Always fail migration. Used for mappings that are not movable */
Christoph Lameter2d1db3b2006-06-23 02:03:33 -0700462int fail_migrate_page(struct address_space *mapping,
463 struct page *newpage, struct page *page)
Christoph Lameter1d8b85c2006-06-23 02:03:28 -0700464{
465 return -EIO;
466}
467EXPORT_SYMBOL(fail_migrate_page);
468
Christoph Lameterb20a3502006-03-22 00:09:12 -0800469/*
470 * Common logic to directly migrate a single page suitable for
David Howells266cf652009-04-03 16:42:36 +0100471 * pages that do not use PagePrivate/PagePrivate2.
Christoph Lameterb20a3502006-03-22 00:09:12 -0800472 *
473 * Pages are locked upon entry and exit.
474 */
Christoph Lameter2d1db3b2006-06-23 02:03:33 -0700475int migrate_page(struct address_space *mapping,
Mel Gormana6bc32b2012-01-12 17:19:43 -0800476 struct page *newpage, struct page *page,
477 enum migrate_mode mode)
Christoph Lameterb20a3502006-03-22 00:09:12 -0800478{
479 int rc;
480
481 BUG_ON(PageWriteback(page)); /* Writeback must be complete */
482
Mel Gormana6bc32b2012-01-12 17:19:43 -0800483 rc = migrate_page_move_mapping(mapping, newpage, page, NULL, mode);
Christoph Lameterb20a3502006-03-22 00:09:12 -0800484
485 if (rc)
486 return rc;
487
488 migrate_page_copy(newpage, page);
Christoph Lameterb20a3502006-03-22 00:09:12 -0800489 return 0;
490}
491EXPORT_SYMBOL(migrate_page);
492
David Howells93614012006-09-30 20:45:40 +0200493#ifdef CONFIG_BLOCK
Christoph Lameterb20a3502006-03-22 00:09:12 -0800494/*
Christoph Lameter1d8b85c2006-06-23 02:03:28 -0700495 * Migration function for pages with buffers. This function can only be used
496 * if the underlying filesystem guarantees that no other references to "page"
497 * exist.
498 */
Christoph Lameter2d1db3b2006-06-23 02:03:33 -0700499int buffer_migrate_page(struct address_space *mapping,
Mel Gormana6bc32b2012-01-12 17:19:43 -0800500 struct page *newpage, struct page *page, enum migrate_mode mode)
Christoph Lameter1d8b85c2006-06-23 02:03:28 -0700501{
Christoph Lameter1d8b85c2006-06-23 02:03:28 -0700502 struct buffer_head *bh, *head;
503 int rc;
504
Christoph Lameter1d8b85c2006-06-23 02:03:28 -0700505 if (!page_has_buffers(page))
Mel Gormana6bc32b2012-01-12 17:19:43 -0800506 return migrate_page(mapping, newpage, page, mode);
Christoph Lameter1d8b85c2006-06-23 02:03:28 -0700507
508 head = page_buffers(page);
509
Mel Gormana6bc32b2012-01-12 17:19:43 -0800510 rc = migrate_page_move_mapping(mapping, newpage, page, head, mode);
Christoph Lameter1d8b85c2006-06-23 02:03:28 -0700511
512 if (rc)
513 return rc;
514
Mel Gormanb969c4a2012-01-12 17:19:34 -0800515 /*
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
519 */
Mel Gormana6bc32b2012-01-12 17:19:43 -0800520 if (mode != MIGRATE_ASYNC)
521 BUG_ON(!buffer_migrate_lock_buffers(head, mode));
Christoph Lameter1d8b85c2006-06-23 02:03:28 -0700522
523 ClearPagePrivate(page);
524 set_page_private(newpage, page_private(page));
525 set_page_private(page, 0);
526 put_page(page);
527 get_page(newpage);
528
529 bh = head;
530 do {
531 set_bh_page(bh, newpage, bh_offset(bh));
532 bh = bh->b_this_page;
533
534 } while (bh != head);
535
536 SetPagePrivate(newpage);
537
538 migrate_page_copy(newpage, page);
539
540 bh = head;
541 do {
542 unlock_buffer(bh);
543 put_bh(bh);
544 bh = bh->b_this_page;
545
546 } while (bh != head);
547
548 return 0;
549}
550EXPORT_SYMBOL(buffer_migrate_page);
David Howells93614012006-09-30 20:45:40 +0200551#endif
Christoph Lameter1d8b85c2006-06-23 02:03:28 -0700552
Christoph Lameter04e62a22006-06-23 02:03:38 -0700553/*
554 * Writeback a page to clean the dirty state
555 */
556static int writeout(struct address_space *mapping, struct page *page)
557{
558 struct writeback_control wbc = {
559 .sync_mode = WB_SYNC_NONE,
560 .nr_to_write = 1,
561 .range_start = 0,
562 .range_end = LLONG_MAX,
Christoph Lameter04e62a22006-06-23 02:03:38 -0700563 .for_reclaim = 1
564 };
565 int rc;
566
567 if (!mapping->a_ops->writepage)
568 /* No write method for the address space */
569 return -EINVAL;
570
571 if (!clear_page_dirty_for_io(page))
572 /* Someone else already triggered a write */
573 return -EAGAIN;
574
575 /*
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
581 * be successful.
582 */
583 remove_migration_ptes(page, page);
584
585 rc = mapping->a_ops->writepage(page, &wbc);
Christoph Lameter04e62a22006-06-23 02:03:38 -0700586
587 if (rc != AOP_WRITEPAGE_ACTIVATE)
588 /* unlocked. Relock */
589 lock_page(page);
590
Hugh Dickinsbda85502008-11-19 15:36:36 -0800591 return (rc < 0) ? -EIO : -EAGAIN;
Christoph Lameter04e62a22006-06-23 02:03:38 -0700592}
593
594/*
595 * Default handling if a filesystem does not provide a migration function.
596 */
Christoph Lameter8351a6e2006-06-23 02:03:33 -0700597static int fallback_migrate_page(struct address_space *mapping,
Mel Gormana6bc32b2012-01-12 17:19:43 -0800598 struct page *newpage, struct page *page, enum migrate_mode mode)
Christoph Lameter8351a6e2006-06-23 02:03:33 -0700599{
Mel Gormanb969c4a2012-01-12 17:19:34 -0800600 if (PageDirty(page)) {
Mel Gormana6bc32b2012-01-12 17:19:43 -0800601 /* Only writeback pages in full synchronous migration */
602 if (mode != MIGRATE_SYNC)
Mel Gormanb969c4a2012-01-12 17:19:34 -0800603 return -EBUSY;
Christoph Lameter04e62a22006-06-23 02:03:38 -0700604 return writeout(mapping, page);
Mel Gormanb969c4a2012-01-12 17:19:34 -0800605 }
Christoph Lameter8351a6e2006-06-23 02:03:33 -0700606
607 /*
608 * Buffers may be managed in a filesystem specific way.
609 * We must have no buffers or drop them.
610 */
David Howells266cf652009-04-03 16:42:36 +0100611 if (page_has_private(page) &&
Christoph Lameter8351a6e2006-06-23 02:03:33 -0700612 !try_to_release_page(page, GFP_KERNEL))
613 return -EAGAIN;
614
Mel Gormana6bc32b2012-01-12 17:19:43 -0800615 return migrate_page(mapping, newpage, page, mode);
Christoph Lameter8351a6e2006-06-23 02:03:33 -0700616}
617
Christoph Lameter1d8b85c2006-06-23 02:03:28 -0700618/*
Christoph Lametere24f0b82006-06-23 02:03:51 -0700619 * Move a page to a newly allocated page
620 * The page is locked and all ptes have been successfully removed.
621 *
622 * The new page will have replaced the old page if this function
623 * is successful.
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700624 *
625 * Return value:
626 * < 0 - error code
627 * == 0 - success
Christoph Lametere24f0b82006-06-23 02:03:51 -0700628 */
Mel Gorman3fe20112010-05-24 14:32:20 -0700629static int move_to_new_page(struct page *newpage, struct page *page,
Mel Gormana6bc32b2012-01-12 17:19:43 -0800630 int remap_swapcache, enum migrate_mode mode)
Christoph Lametere24f0b82006-06-23 02:03:51 -0700631{
632 struct address_space *mapping;
633 int rc;
634
635 /*
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.
639 */
Nick Piggin529ae9a2008-08-02 12:01:03 +0200640 if (!trylock_page(newpage))
Christoph Lametere24f0b82006-06-23 02:03:51 -0700641 BUG();
642
643 /* Prepare mapping for the new page.*/
644 newpage->index = page->index;
645 newpage->mapping = page->mapping;
Rik van Rielb2e18532008-10-18 20:26:30 -0700646 if (PageSwapBacked(page))
647 SetPageSwapBacked(newpage);
Christoph Lametere24f0b82006-06-23 02:03:51 -0700648
649 mapping = page_mapping(page);
650 if (!mapping)
Mel Gormana6bc32b2012-01-12 17:19:43 -0800651 rc = migrate_page(mapping, newpage, page, mode);
Mel Gormanb969c4a2012-01-12 17:19:34 -0800652 else if (mapping->a_ops->migratepage)
Christoph Lametere24f0b82006-06-23 02:03:51 -0700653 /*
Mel Gormanb969c4a2012-01-12 17:19:34 -0800654 * 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.
Christoph Lametere24f0b82006-06-23 02:03:51 -0700658 */
Mel Gormanb969c4a2012-01-12 17:19:34 -0800659 rc = mapping->a_ops->migratepage(mapping,
Mel Gormana6bc32b2012-01-12 17:19:43 -0800660 newpage, page, mode);
Mel Gormanb969c4a2012-01-12 17:19:34 -0800661 else
Mel Gormana6bc32b2012-01-12 17:19:43 -0800662 rc = fallback_migrate_page(mapping, newpage, page, mode);
Christoph Lametere24f0b82006-06-23 02:03:51 -0700663
Mel Gorman3fe20112010-05-24 14:32:20 -0700664 if (rc) {
Christoph Lametere24f0b82006-06-23 02:03:51 -0700665 newpage->mapping = NULL;
Mel Gorman3fe20112010-05-24 14:32:20 -0700666 } else {
667 if (remap_swapcache)
668 remove_migration_ptes(page, newpage);
Konstantin Khlebnikov35512ec2012-02-03 15:37:13 -0800669 page->mapping = NULL;
Mel Gorman3fe20112010-05-24 14:32:20 -0700670 }
Christoph Lametere24f0b82006-06-23 02:03:51 -0700671
672 unlock_page(newpage);
673
674 return rc;
675}
676
Minchan Kim0dabec92011-10-31 17:06:57 -0700677static int __unmap_and_move(struct page *page, struct page *newpage,
Mel Gormana6bc32b2012-01-12 17:19:43 -0800678 int force, bool offlining, enum migrate_mode mode)
Christoph Lametere24f0b82006-06-23 02:03:51 -0700679{
Minchan Kim0dabec92011-10-31 17:06:57 -0700680 int rc = -EAGAIN;
Mel Gorman3fe20112010-05-24 14:32:20 -0700681 int remap_swapcache = 1;
KAMEZAWA Hiroyukiae41be32008-02-07 00:14:10 -0800682 int charge = 0;
KAMEZAWA Hiroyuki56039ef2011-03-23 16:42:19 -0700683 struct mem_cgroup *mem;
Mel Gorman3f6c8272010-05-24 14:32:17 -0700684 struct anon_vma *anon_vma = NULL;
Christoph Lameter95a402c2006-06-23 02:03:53 -0700685
Nick Piggin529ae9a2008-08-02 12:01:03 +0200686 if (!trylock_page(page)) {
Mel Gormana6bc32b2012-01-12 17:19:43 -0800687 if (!force || mode == MIGRATE_ASYNC)
Minchan Kim0dabec92011-10-31 17:06:57 -0700688 goto out;
Mel Gorman3e7d3442011-01-13 15:45:56 -0800689
690 /*
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
701 * altogether.
702 */
703 if (current->flags & PF_MEMALLOC)
Minchan Kim0dabec92011-10-31 17:06:57 -0700704 goto out;
Mel Gorman3e7d3442011-01-13 15:45:56 -0800705
Christoph Lametere24f0b82006-06-23 02:03:51 -0700706 lock_page(page);
707 }
708
Hugh Dickins62b61f62009-12-14 17:59:33 -0800709 /*
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
716 * serializes that).
717 */
718 if (PageKsm(page) && !offlining) {
719 rc = -EBUSY;
720 goto unlock;
721 }
722
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -0800723 /* charge against new page */
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700724 charge = mem_cgroup_prepare_migration(page, newpage, &mem, GFP_KERNEL);
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -0800725 if (charge == -ENOMEM) {
726 rc = -ENOMEM;
727 goto unlock;
728 }
729 BUG_ON(charge);
730
Christoph Lametere24f0b82006-06-23 02:03:51 -0700731 if (PageWriteback(page)) {
Andrea Arcangeli11bc82d2011-03-22 16:33:11 -0700732 /*
Mel Gormana6bc32b2012-01-12 17:19:43 -0800733 * 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
Andrea Arcangeli11bc82d2011-03-22 16:33:11 -0700737 */
Mel Gormana6bc32b2012-01-12 17:19:43 -0800738 if (mode != MIGRATE_SYNC) {
Andrea Arcangeli11bc82d2011-03-22 16:33:11 -0700739 rc = -EBUSY;
740 goto uncharge;
741 }
742 if (!force)
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -0800743 goto uncharge;
Christoph Lametere24f0b82006-06-23 02:03:51 -0700744 wait_on_page_writeback(page);
745 }
Christoph Lametere24f0b82006-06-23 02:03:51 -0700746 /*
KAMEZAWA Hiroyukidc386d42007-07-26 10:41:07 -0700747 * 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.
Hugh Dickins1ce82b62011-01-13 15:47:30 -0800749 * This get_anon_vma() delays freeing anon_vma pointer until the end
KAMEZAWA Hiroyukidc386d42007-07-26 10:41:07 -0700750 * of migration. File cache pages are no problem because of page_lock()
KAMEZAWA Hiroyuki989f89c2007-08-30 23:56:21 -0700751 * File Caches may use write_page() or lock_page() in migration, then,
752 * just care Anon page here.
Christoph Lametere24f0b82006-06-23 02:03:51 -0700753 */
KAMEZAWA Hiroyuki989f89c2007-08-30 23:56:21 -0700754 if (PageAnon(page)) {
Hugh Dickins1ce82b62011-01-13 15:47:30 -0800755 /*
756 * Only page_lock_anon_vma() understands the subtleties of
757 * getting a hold on an anon_vma from outside one of its mms.
758 */
Peter Zijlstra746b18d2011-05-24 17:12:10 -0700759 anon_vma = page_get_anon_vma(page);
Hugh Dickins1ce82b62011-01-13 15:47:30 -0800760 if (anon_vma) {
761 /*
Peter Zijlstra746b18d2011-05-24 17:12:10 -0700762 * Anon page
Hugh Dickins1ce82b62011-01-13 15:47:30 -0800763 */
Hugh Dickins1ce82b62011-01-13 15:47:30 -0800764 } else if (PageSwapCache(page)) {
Mel Gorman3fe20112010-05-24 14:32:20 -0700765 /*
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.
772 *
773 * To avoid this possibility, swapcache pages get
774 * migrated but are not remapped when migration
775 * completes
776 */
777 remap_swapcache = 0;
778 } else {
Hugh Dickins1ce82b62011-01-13 15:47:30 -0800779 goto uncharge;
Mel Gorman3fe20112010-05-24 14:32:20 -0700780 }
KAMEZAWA Hiroyuki989f89c2007-08-30 23:56:21 -0700781 }
Shaohua Li62e1c552008-02-04 22:29:33 -0800782
KAMEZAWA Hiroyukidc386d42007-07-26 10:41:07 -0700783 /*
Shaohua Li62e1c552008-02-04 22:29:33 -0800784 * 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.
KAMEZAWA Hiroyukidc386d42007-07-26 10:41:07 -0700794 */
Shaohua Li62e1c552008-02-04 22:29:33 -0800795 if (!page->mapping) {
Hugh Dickins1ce82b62011-01-13 15:47:30 -0800796 VM_BUG_ON(PageAnon(page));
797 if (page_has_private(page)) {
Shaohua Li62e1c552008-02-04 22:29:33 -0800798 try_to_free_buffers(page);
Hugh Dickins1ce82b62011-01-13 15:47:30 -0800799 goto uncharge;
Shaohua Li62e1c552008-02-04 22:29:33 -0800800 }
Shaohua Liabfc3482009-09-21 17:01:19 -0700801 goto skip_unmap;
Shaohua Li62e1c552008-02-04 22:29:33 -0800802 }
803
KAMEZAWA Hiroyukidc386d42007-07-26 10:41:07 -0700804 /* Establish migration ptes or remove ptes */
Andi Kleen14fa31b2009-09-16 11:50:10 +0200805 try_to_unmap(page, TTU_MIGRATION|TTU_IGNORE_MLOCK|TTU_IGNORE_ACCESS);
KAMEZAWA Hiroyukidc386d42007-07-26 10:41:07 -0700806
Shaohua Liabfc3482009-09-21 17:01:19 -0700807skip_unmap:
Christoph Lametere6a15302006-06-25 05:46:49 -0700808 if (!page_mapped(page))
Mel Gormana6bc32b2012-01-12 17:19:43 -0800809 rc = move_to_new_page(newpage, page, remap_swapcache, mode);
Christoph Lametere24f0b82006-06-23 02:03:51 -0700810
Mel Gorman3fe20112010-05-24 14:32:20 -0700811 if (rc && remap_swapcache)
Christoph Lametere24f0b82006-06-23 02:03:51 -0700812 remove_migration_ptes(page, page);
Mel Gorman3f6c8272010-05-24 14:32:17 -0700813
814 /* Drop an anon_vma reference if we took one */
Rik van Riel76545062010-08-09 17:18:41 -0700815 if (anon_vma)
Peter Zijlstra9e601092011-03-22 16:32:46 -0700816 put_anon_vma(anon_vma);
Mel Gorman3f6c8272010-05-24 14:32:17 -0700817
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -0800818uncharge:
819 if (!charge)
Daisuke Nishimura50de1dd2011-01-13 15:47:43 -0800820 mem_cgroup_end_migration(mem, page, newpage, rc == 0);
Christoph Lametere24f0b82006-06-23 02:03:51 -0700821unlock:
822 unlock_page(page);
Minchan Kim0dabec92011-10-31 17:06:57 -0700823out:
824 return rc;
825}
Christoph Lameter95a402c2006-06-23 02:03:53 -0700826
Minchan Kim0dabec92011-10-31 17:06:57 -0700827/*
828 * Obtain the lock on page, remove all ptes and migrate the page
829 * to the newly allocated page in newpage.
830 */
831static int unmap_and_move(new_page_t get_new_page, unsigned long private,
Mel Gormana6bc32b2012-01-12 17:19:43 -0800832 struct page *page, int force, bool offlining,
833 enum migrate_mode mode)
Minchan Kim0dabec92011-10-31 17:06:57 -0700834{
835 int rc = 0;
836 int *result = NULL;
837 struct page *newpage = get_new_page(page, private, &result);
838
839 if (!newpage)
840 return -ENOMEM;
841
KAMEZAWA Hiroyuki4e5f01c2012-01-12 17:18:58 -0800842 mem_cgroup_reset_owner(newpage);
843
Minchan Kim0dabec92011-10-31 17:06:57 -0700844 if (page_count(page) == 1) {
845 /* page was freed from under us. So we are done. */
846 goto out;
847 }
848
849 if (unlikely(PageTransHuge(page)))
850 if (unlikely(split_huge_page(page)))
851 goto out;
852
Mel Gormana6bc32b2012-01-12 17:19:43 -0800853 rc = __unmap_and_move(page, newpage, force, offlining, mode);
Minchan Kim0dabec92011-10-31 17:06:57 -0700854out:
Christoph Lametere24f0b82006-06-23 02:03:51 -0700855 if (rc != -EAGAIN) {
Minchan Kim0dabec92011-10-31 17:06:57 -0700856 /*
857 * A page that has been migrated has all references
858 * removed and will be freed. A page that has not been
859 * migrated will have kepts its references and be
860 * restored.
861 */
862 list_del(&page->lru);
KOSAKI Motohiroa7312862009-09-21 17:01:37 -0700863 dec_zone_page_state(page, NR_ISOLATED_ANON +
Johannes Weiner6c0b1352009-09-21 17:02:59 -0700864 page_is_file_cache(page));
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700865 putback_lru_page(page);
Christoph Lametere24f0b82006-06-23 02:03:51 -0700866 }
Christoph Lameter95a402c2006-06-23 02:03:53 -0700867 /*
868 * Move the new page to the LRU. If migration was not successful
869 * then this will free the page.
870 */
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700871 putback_lru_page(newpage);
Christoph Lameter742755a2006-06-23 02:03:55 -0700872 if (result) {
873 if (rc)
874 *result = rc;
875 else
876 *result = page_to_nid(newpage);
877 }
Christoph Lametere24f0b82006-06-23 02:03:51 -0700878 return rc;
879}
880
881/*
Naoya Horiguchi290408d2010-09-08 10:19:35 +0900882 * Counterpart of unmap_and_move_page() for hugepage migration.
883 *
884 * This function doesn't wait the completion of hugepage I/O
885 * because there is no race between I/O and migration for hugepage.
886 * Note that currently hugepage I/O occurs only in direct I/O
887 * where no lock is held and PG_writeback is irrelevant,
888 * and writeback status of all subpages are counted in the reference
889 * count of the head page (i.e. if all subpages of a 2MB hugepage are
890 * under direct I/O, the reference of the head page is 512 and a bit more.)
891 * This means that when we try to migrate hugepage whose subpages are
892 * doing direct I/O, some references remain after try_to_unmap() and
893 * hugepage migration fails without data corruption.
894 *
895 * There is also no race when direct I/O is issued on the page under migration,
896 * because then pte is replaced with migration swap entry and direct I/O code
897 * will wait in the page fault for migration to complete.
898 */
899static int unmap_and_move_huge_page(new_page_t get_new_page,
900 unsigned long private, struct page *hpage,
Mel Gormana6bc32b2012-01-12 17:19:43 -0800901 int force, bool offlining,
902 enum migrate_mode mode)
Naoya Horiguchi290408d2010-09-08 10:19:35 +0900903{
904 int rc = 0;
905 int *result = NULL;
906 struct page *new_hpage = get_new_page(hpage, private, &result);
Naoya Horiguchi290408d2010-09-08 10:19:35 +0900907 struct anon_vma *anon_vma = NULL;
908
909 if (!new_hpage)
910 return -ENOMEM;
911
912 rc = -EAGAIN;
913
914 if (!trylock_page(hpage)) {
Mel Gormana6bc32b2012-01-12 17:19:43 -0800915 if (!force || mode != MIGRATE_SYNC)
Naoya Horiguchi290408d2010-09-08 10:19:35 +0900916 goto out;
917 lock_page(hpage);
918 }
919
Peter Zijlstra746b18d2011-05-24 17:12:10 -0700920 if (PageAnon(hpage))
921 anon_vma = page_get_anon_vma(hpage);
Naoya Horiguchi290408d2010-09-08 10:19:35 +0900922
923 try_to_unmap(hpage, TTU_MIGRATION|TTU_IGNORE_MLOCK|TTU_IGNORE_ACCESS);
924
925 if (!page_mapped(hpage))
Mel Gormana6bc32b2012-01-12 17:19:43 -0800926 rc = move_to_new_page(new_hpage, hpage, 1, mode);
Naoya Horiguchi290408d2010-09-08 10:19:35 +0900927
928 if (rc)
929 remove_migration_ptes(hpage, hpage);
930
Hugh Dickinsfd4a4662011-01-13 15:47:31 -0800931 if (anon_vma)
Peter Zijlstra9e601092011-03-22 16:32:46 -0700932 put_anon_vma(anon_vma);
Naoya Horiguchi290408d2010-09-08 10:19:35 +0900933 unlock_page(hpage);
934
Hillf Danton09761332011-12-08 14:34:20 -0800935out:
Naoya Horiguchi290408d2010-09-08 10:19:35 +0900936 if (rc != -EAGAIN) {
937 list_del(&hpage->lru);
938 put_page(hpage);
939 }
940
941 put_page(new_hpage);
942
943 if (result) {
944 if (rc)
945 *result = rc;
946 else
947 *result = page_to_nid(new_hpage);
948 }
949 return rc;
950}
951
952/*
Christoph Lameterb20a3502006-03-22 00:09:12 -0800953 * migrate_pages
954 *
Christoph Lameter95a402c2006-06-23 02:03:53 -0700955 * The function takes one list of pages to migrate and a function
956 * that determines from the page to be migrated and the private data
957 * the target of the move and allocates the page.
Christoph Lameterb20a3502006-03-22 00:09:12 -0800958 *
959 * The function returns after 10 attempts or if no pages
960 * are movable anymore because to has become empty
Minchan Kimcf608ac2010-10-26 14:21:29 -0700961 * or no retryable pages exist anymore.
962 * Caller should call putback_lru_pages to return pages to the LRU
Minchan Kim28bd6572011-01-25 15:07:26 -0800963 * or free list only if ret != 0.
Christoph Lameterb20a3502006-03-22 00:09:12 -0800964 *
Christoph Lameter95a402c2006-06-23 02:03:53 -0700965 * Return: Number of pages not migrated or error code.
Christoph Lameterb20a3502006-03-22 00:09:12 -0800966 */
Christoph Lameter95a402c2006-06-23 02:03:53 -0700967int migrate_pages(struct list_head *from,
Mel Gorman7f0f2492011-01-13 15:45:58 -0800968 new_page_t get_new_page, unsigned long private, bool offlining,
Mel Gormana6bc32b2012-01-12 17:19:43 -0800969 enum migrate_mode mode)
Christoph Lameterb20a3502006-03-22 00:09:12 -0800970{
Christoph Lametere24f0b82006-06-23 02:03:51 -0700971 int retry = 1;
Christoph Lameterb20a3502006-03-22 00:09:12 -0800972 int nr_failed = 0;
973 int pass = 0;
974 struct page *page;
975 struct page *page2;
976 int swapwrite = current->flags & PF_SWAPWRITE;
977 int rc;
978
979 if (!swapwrite)
980 current->flags |= PF_SWAPWRITE;
981
Christoph Lametere24f0b82006-06-23 02:03:51 -0700982 for(pass = 0; pass < 10 && retry; pass++) {
983 retry = 0;
Christoph Lameterb20a3502006-03-22 00:09:12 -0800984
Christoph Lametere24f0b82006-06-23 02:03:51 -0700985 list_for_each_entry_safe(page, page2, from, lru) {
Christoph Lametere24f0b82006-06-23 02:03:51 -0700986 cond_resched();
Christoph Lameterb20a3502006-03-22 00:09:12 -0800987
Christoph Lameter95a402c2006-06-23 02:03:53 -0700988 rc = unmap_and_move(get_new_page, private,
Mel Gorman77f1fe62011-01-13 15:45:57 -0800989 page, pass > 2, offlining,
Mel Gormana6bc32b2012-01-12 17:19:43 -0800990 mode);
Christoph Lameterb20a3502006-03-22 00:09:12 -0800991
Christoph Lametere24f0b82006-06-23 02:03:51 -0700992 switch(rc) {
Christoph Lameter95a402c2006-06-23 02:03:53 -0700993 case -ENOMEM:
994 goto out;
Christoph Lametere24f0b82006-06-23 02:03:51 -0700995 case -EAGAIN:
Christoph Lameter2d1db3b2006-06-23 02:03:33 -0700996 retry++;
Christoph Lametere24f0b82006-06-23 02:03:51 -0700997 break;
998 case 0:
Christoph Lametere24f0b82006-06-23 02:03:51 -0700999 break;
1000 default:
Christoph Lameter2d1db3b2006-06-23 02:03:33 -07001001 /* Permanent failure */
Christoph Lameter2d1db3b2006-06-23 02:03:33 -07001002 nr_failed++;
Christoph Lametere24f0b82006-06-23 02:03:51 -07001003 break;
Christoph Lameter2d1db3b2006-06-23 02:03:33 -07001004 }
Christoph Lameterb20a3502006-03-22 00:09:12 -08001005 }
1006 }
Christoph Lameter95a402c2006-06-23 02:03:53 -07001007 rc = 0;
1008out:
Christoph Lameterb20a3502006-03-22 00:09:12 -08001009 if (!swapwrite)
1010 current->flags &= ~PF_SWAPWRITE;
1011
Christoph Lameter95a402c2006-06-23 02:03:53 -07001012 if (rc)
1013 return rc;
1014
Christoph Lameterb20a3502006-03-22 00:09:12 -08001015 return nr_failed + retry;
1016}
1017
Naoya Horiguchi290408d2010-09-08 10:19:35 +09001018int migrate_huge_pages(struct list_head *from,
Mel Gorman7f0f2492011-01-13 15:45:58 -08001019 new_page_t get_new_page, unsigned long private, bool offlining,
Mel Gormana6bc32b2012-01-12 17:19:43 -08001020 enum migrate_mode mode)
Naoya Horiguchi290408d2010-09-08 10:19:35 +09001021{
1022 int retry = 1;
1023 int nr_failed = 0;
1024 int pass = 0;
1025 struct page *page;
1026 struct page *page2;
1027 int rc;
1028
1029 for (pass = 0; pass < 10 && retry; pass++) {
1030 retry = 0;
1031
1032 list_for_each_entry_safe(page, page2, from, lru) {
1033 cond_resched();
1034
1035 rc = unmap_and_move_huge_page(get_new_page,
Mel Gorman77f1fe62011-01-13 15:45:57 -08001036 private, page, pass > 2, offlining,
Mel Gormana6bc32b2012-01-12 17:19:43 -08001037 mode);
Naoya Horiguchi290408d2010-09-08 10:19:35 +09001038
1039 switch(rc) {
1040 case -ENOMEM:
1041 goto out;
1042 case -EAGAIN:
1043 retry++;
1044 break;
1045 case 0:
1046 break;
1047 default:
1048 /* Permanent failure */
1049 nr_failed++;
1050 break;
1051 }
1052 }
1053 }
1054 rc = 0;
1055out:
Naoya Horiguchi290408d2010-09-08 10:19:35 +09001056 if (rc)
1057 return rc;
1058
1059 return nr_failed + retry;
1060}
1061
Christoph Lameter742755a2006-06-23 02:03:55 -07001062#ifdef CONFIG_NUMA
1063/*
1064 * Move a list of individual pages
1065 */
1066struct page_to_node {
1067 unsigned long addr;
1068 struct page *page;
1069 int node;
1070 int status;
1071};
1072
1073static struct page *new_page_node(struct page *p, unsigned long private,
1074 int **result)
1075{
1076 struct page_to_node *pm = (struct page_to_node *)private;
1077
1078 while (pm->node != MAX_NUMNODES && pm->page != p)
1079 pm++;
1080
1081 if (pm->node == MAX_NUMNODES)
1082 return NULL;
1083
1084 *result = &pm->status;
1085
Mel Gorman6484eb32009-06-16 15:31:54 -07001086 return alloc_pages_exact_node(pm->node,
Mel Gorman769848c2007-07-17 04:03:05 -07001087 GFP_HIGHUSER_MOVABLE | GFP_THISNODE, 0);
Christoph Lameter742755a2006-06-23 02:03:55 -07001088}
1089
1090/*
1091 * Move a set of pages as indicated in the pm array. The addr
1092 * field must be set to the virtual address of the page to be moved
1093 * and the node number must contain a valid target node.
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001094 * The pm array ends with node = MAX_NUMNODES.
Christoph Lameter742755a2006-06-23 02:03:55 -07001095 */
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001096static int do_move_page_to_node_array(struct mm_struct *mm,
1097 struct page_to_node *pm,
1098 int migrate_all)
Christoph Lameter742755a2006-06-23 02:03:55 -07001099{
1100 int err;
1101 struct page_to_node *pp;
1102 LIST_HEAD(pagelist);
1103
1104 down_read(&mm->mmap_sem);
1105
1106 /*
1107 * Build a list of pages to migrate
1108 */
Christoph Lameter742755a2006-06-23 02:03:55 -07001109 for (pp = pm; pp->node != MAX_NUMNODES; pp++) {
1110 struct vm_area_struct *vma;
1111 struct page *page;
1112
Christoph Lameter742755a2006-06-23 02:03:55 -07001113 err = -EFAULT;
1114 vma = find_vma(mm, pp->addr);
Gleb Natapov70384dc2010-10-26 14:22:07 -07001115 if (!vma || pp->addr < vma->vm_start || !vma_migratable(vma))
Christoph Lameter742755a2006-06-23 02:03:55 -07001116 goto set_status;
1117
Andrea Arcangeli500d65d2011-01-13 15:46:55 -08001118 page = follow_page(vma, pp->addr, FOLL_GET|FOLL_SPLIT);
Linus Torvalds89f5b7d2008-06-20 11:18:25 -07001119
1120 err = PTR_ERR(page);
1121 if (IS_ERR(page))
1122 goto set_status;
1123
Christoph Lameter742755a2006-06-23 02:03:55 -07001124 err = -ENOENT;
1125 if (!page)
1126 goto set_status;
1127
Hugh Dickins62b61f62009-12-14 17:59:33 -08001128 /* Use PageReserved to check for zero page */
1129 if (PageReserved(page) || PageKsm(page))
Christoph Lameter742755a2006-06-23 02:03:55 -07001130 goto put_and_set;
1131
1132 pp->page = page;
1133 err = page_to_nid(page);
1134
1135 if (err == pp->node)
1136 /*
1137 * Node already in the right place
1138 */
1139 goto put_and_set;
1140
1141 err = -EACCES;
1142 if (page_mapcount(page) > 1 &&
1143 !migrate_all)
1144 goto put_and_set;
1145
Nick Piggin62695a82008-10-18 20:26:09 -07001146 err = isolate_lru_page(page);
KOSAKI Motohiro6d9c2852009-12-14 17:58:11 -08001147 if (!err) {
Nick Piggin62695a82008-10-18 20:26:09 -07001148 list_add_tail(&page->lru, &pagelist);
KOSAKI Motohiro6d9c2852009-12-14 17:58:11 -08001149 inc_zone_page_state(page, NR_ISOLATED_ANON +
1150 page_is_file_cache(page));
1151 }
Christoph Lameter742755a2006-06-23 02:03:55 -07001152put_and_set:
1153 /*
1154 * Either remove the duplicate refcount from
1155 * isolate_lru_page() or drop the page ref if it was
1156 * not isolated.
1157 */
1158 put_page(page);
1159set_status:
1160 pp->status = err;
1161 }
1162
Brice Gogline78bbfa2008-10-18 20:27:15 -07001163 err = 0;
Minchan Kimcf608ac2010-10-26 14:21:29 -07001164 if (!list_empty(&pagelist)) {
Christoph Lameter742755a2006-06-23 02:03:55 -07001165 err = migrate_pages(&pagelist, new_page_node,
Mel Gormana6bc32b2012-01-12 17:19:43 -08001166 (unsigned long)pm, 0, MIGRATE_SYNC);
Minchan Kimcf608ac2010-10-26 14:21:29 -07001167 if (err)
1168 putback_lru_pages(&pagelist);
1169 }
Christoph Lameter742755a2006-06-23 02:03:55 -07001170
1171 up_read(&mm->mmap_sem);
1172 return err;
1173}
1174
1175/*
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001176 * Migrate an array of page address onto an array of nodes and fill
1177 * the corresponding array of status.
1178 */
1179static int do_pages_move(struct mm_struct *mm, struct task_struct *task,
1180 unsigned long nr_pages,
1181 const void __user * __user *pages,
1182 const int __user *nodes,
1183 int __user *status, int flags)
1184{
Brice Goglin3140a222009-01-06 14:38:57 -08001185 struct page_to_node *pm;
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001186 nodemask_t task_nodes;
Brice Goglin3140a222009-01-06 14:38:57 -08001187 unsigned long chunk_nr_pages;
1188 unsigned long chunk_start;
1189 int err;
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001190
1191 task_nodes = cpuset_mems_allowed(task);
1192
Brice Goglin3140a222009-01-06 14:38:57 -08001193 err = -ENOMEM;
1194 pm = (struct page_to_node *)__get_free_page(GFP_KERNEL);
1195 if (!pm)
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001196 goto out;
Brice Goglin35282a22009-06-16 15:32:43 -07001197
1198 migrate_prep();
1199
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001200 /*
Brice Goglin3140a222009-01-06 14:38:57 -08001201 * Store a chunk of page_to_node array in a page,
1202 * but keep the last one as a marker
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001203 */
Brice Goglin3140a222009-01-06 14:38:57 -08001204 chunk_nr_pages = (PAGE_SIZE / sizeof(struct page_to_node)) - 1;
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001205
Brice Goglin3140a222009-01-06 14:38:57 -08001206 for (chunk_start = 0;
1207 chunk_start < nr_pages;
1208 chunk_start += chunk_nr_pages) {
1209 int j;
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001210
Brice Goglin3140a222009-01-06 14:38:57 -08001211 if (chunk_start + chunk_nr_pages > nr_pages)
1212 chunk_nr_pages = nr_pages - chunk_start;
1213
1214 /* fill the chunk pm with addrs and nodes from user-space */
1215 for (j = 0; j < chunk_nr_pages; j++) {
1216 const void __user *p;
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001217 int node;
1218
Brice Goglin3140a222009-01-06 14:38:57 -08001219 err = -EFAULT;
1220 if (get_user(p, pages + j + chunk_start))
1221 goto out_pm;
1222 pm[j].addr = (unsigned long) p;
1223
1224 if (get_user(node, nodes + j + chunk_start))
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001225 goto out_pm;
1226
1227 err = -ENODEV;
Linus Torvalds6f5a55f2010-02-05 16:16:50 -08001228 if (node < 0 || node >= MAX_NUMNODES)
1229 goto out_pm;
1230
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001231 if (!node_state(node, N_HIGH_MEMORY))
1232 goto out_pm;
1233
1234 err = -EACCES;
1235 if (!node_isset(node, task_nodes))
1236 goto out_pm;
1237
Brice Goglin3140a222009-01-06 14:38:57 -08001238 pm[j].node = node;
1239 }
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001240
Brice Goglin3140a222009-01-06 14:38:57 -08001241 /* End marker for this chunk */
1242 pm[chunk_nr_pages].node = MAX_NUMNODES;
1243
1244 /* Migrate this chunk */
1245 err = do_move_page_to_node_array(mm, pm,
1246 flags & MPOL_MF_MOVE_ALL);
1247 if (err < 0)
1248 goto out_pm;
1249
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001250 /* Return status information */
Brice Goglin3140a222009-01-06 14:38:57 -08001251 for (j = 0; j < chunk_nr_pages; j++)
1252 if (put_user(pm[j].status, status + j + chunk_start)) {
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001253 err = -EFAULT;
Brice Goglin3140a222009-01-06 14:38:57 -08001254 goto out_pm;
1255 }
1256 }
1257 err = 0;
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001258
1259out_pm:
Brice Goglin3140a222009-01-06 14:38:57 -08001260 free_page((unsigned long)pm);
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001261out:
1262 return err;
1263}
1264
1265/*
Brice Goglin2f007e72008-10-18 20:27:16 -07001266 * Determine the nodes of an array of pages and store it in an array of status.
Christoph Lameter742755a2006-06-23 02:03:55 -07001267 */
Brice Goglin80bba122008-12-09 13:14:23 -08001268static void do_pages_stat_array(struct mm_struct *mm, unsigned long nr_pages,
1269 const void __user **pages, int *status)
Christoph Lameter742755a2006-06-23 02:03:55 -07001270{
Brice Goglin2f007e72008-10-18 20:27:16 -07001271 unsigned long i;
Brice Goglin2f007e72008-10-18 20:27:16 -07001272
Christoph Lameter742755a2006-06-23 02:03:55 -07001273 down_read(&mm->mmap_sem);
1274
Brice Goglin2f007e72008-10-18 20:27:16 -07001275 for (i = 0; i < nr_pages; i++) {
Brice Goglin80bba122008-12-09 13:14:23 -08001276 unsigned long addr = (unsigned long)(*pages);
Christoph Lameter742755a2006-06-23 02:03:55 -07001277 struct vm_area_struct *vma;
1278 struct page *page;
KOSAKI Motohiroc095adb2008-12-16 16:06:43 +09001279 int err = -EFAULT;
Brice Goglin2f007e72008-10-18 20:27:16 -07001280
1281 vma = find_vma(mm, addr);
Gleb Natapov70384dc2010-10-26 14:22:07 -07001282 if (!vma || addr < vma->vm_start)
Christoph Lameter742755a2006-06-23 02:03:55 -07001283 goto set_status;
1284
Brice Goglin2f007e72008-10-18 20:27:16 -07001285 page = follow_page(vma, addr, 0);
Linus Torvalds89f5b7d2008-06-20 11:18:25 -07001286
1287 err = PTR_ERR(page);
1288 if (IS_ERR(page))
1289 goto set_status;
1290
Christoph Lameter742755a2006-06-23 02:03:55 -07001291 err = -ENOENT;
1292 /* Use PageReserved to check for zero page */
Hugh Dickins62b61f62009-12-14 17:59:33 -08001293 if (!page || PageReserved(page) || PageKsm(page))
Christoph Lameter742755a2006-06-23 02:03:55 -07001294 goto set_status;
1295
1296 err = page_to_nid(page);
1297set_status:
Brice Goglin80bba122008-12-09 13:14:23 -08001298 *status = err;
1299
1300 pages++;
1301 status++;
1302 }
1303
1304 up_read(&mm->mmap_sem);
1305}
1306
1307/*
1308 * Determine the nodes of a user array of pages and store it in
1309 * a user array of status.
1310 */
1311static int do_pages_stat(struct mm_struct *mm, unsigned long nr_pages,
1312 const void __user * __user *pages,
1313 int __user *status)
1314{
1315#define DO_PAGES_STAT_CHUNK_NR 16
1316 const void __user *chunk_pages[DO_PAGES_STAT_CHUNK_NR];
1317 int chunk_status[DO_PAGES_STAT_CHUNK_NR];
Brice Goglin80bba122008-12-09 13:14:23 -08001318
H. Peter Anvin87b8d1a2010-02-18 16:13:40 -08001319 while (nr_pages) {
1320 unsigned long chunk_nr;
Brice Goglin80bba122008-12-09 13:14:23 -08001321
H. Peter Anvin87b8d1a2010-02-18 16:13:40 -08001322 chunk_nr = nr_pages;
1323 if (chunk_nr > DO_PAGES_STAT_CHUNK_NR)
1324 chunk_nr = DO_PAGES_STAT_CHUNK_NR;
1325
1326 if (copy_from_user(chunk_pages, pages, chunk_nr * sizeof(*chunk_pages)))
1327 break;
Brice Goglin80bba122008-12-09 13:14:23 -08001328
1329 do_pages_stat_array(mm, chunk_nr, chunk_pages, chunk_status);
1330
H. Peter Anvin87b8d1a2010-02-18 16:13:40 -08001331 if (copy_to_user(status, chunk_status, chunk_nr * sizeof(*status)))
1332 break;
Christoph Lameter742755a2006-06-23 02:03:55 -07001333
H. Peter Anvin87b8d1a2010-02-18 16:13:40 -08001334 pages += chunk_nr;
1335 status += chunk_nr;
1336 nr_pages -= chunk_nr;
1337 }
1338 return nr_pages ? -EFAULT : 0;
Christoph Lameter742755a2006-06-23 02:03:55 -07001339}
1340
1341/*
1342 * Move a list of pages in the address space of the currently executing
1343 * process.
1344 */
Heiko Carstens938bb9f2009-01-14 14:14:30 +01001345SYSCALL_DEFINE6(move_pages, pid_t, pid, unsigned long, nr_pages,
1346 const void __user * __user *, pages,
1347 const int __user *, nodes,
1348 int __user *, status, int, flags)
Christoph Lameter742755a2006-06-23 02:03:55 -07001349{
David Howellsc69e8d92008-11-14 10:39:19 +11001350 const struct cred *cred = current_cred(), *tcred;
Christoph Lameter742755a2006-06-23 02:03:55 -07001351 struct task_struct *task;
Christoph Lameter742755a2006-06-23 02:03:55 -07001352 struct mm_struct *mm;
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001353 int err;
Christoph Lameter742755a2006-06-23 02:03:55 -07001354
1355 /* Check flags */
1356 if (flags & ~(MPOL_MF_MOVE|MPOL_MF_MOVE_ALL))
1357 return -EINVAL;
1358
1359 if ((flags & MPOL_MF_MOVE_ALL) && !capable(CAP_SYS_NICE))
1360 return -EPERM;
1361
1362 /* Find the mm_struct */
Greg Thelena879bf52011-02-25 14:44:13 -08001363 rcu_read_lock();
Pavel Emelyanov228ebcb2007-10-18 23:40:16 -07001364 task = pid ? find_task_by_vpid(pid) : current;
Christoph Lameter742755a2006-06-23 02:03:55 -07001365 if (!task) {
Greg Thelena879bf52011-02-25 14:44:13 -08001366 rcu_read_unlock();
Christoph Lameter742755a2006-06-23 02:03:55 -07001367 return -ESRCH;
1368 }
1369 mm = get_task_mm(task);
Greg Thelena879bf52011-02-25 14:44:13 -08001370 rcu_read_unlock();
Christoph Lameter742755a2006-06-23 02:03:55 -07001371
1372 if (!mm)
1373 return -EINVAL;
1374
1375 /*
1376 * Check if this process has the right to modify the specified
1377 * process. The right exists if the process has administrative
1378 * capabilities, superuser privileges or the same
1379 * userid as the target process.
1380 */
David Howellsc69e8d92008-11-14 10:39:19 +11001381 rcu_read_lock();
1382 tcred = __task_cred(task);
David Howellsb6dff3e2008-11-14 10:39:16 +11001383 if (cred->euid != tcred->suid && cred->euid != tcred->uid &&
1384 cred->uid != tcred->suid && cred->uid != tcred->uid &&
Christoph Lameter742755a2006-06-23 02:03:55 -07001385 !capable(CAP_SYS_NICE)) {
David Howellsc69e8d92008-11-14 10:39:19 +11001386 rcu_read_unlock();
Christoph Lameter742755a2006-06-23 02:03:55 -07001387 err = -EPERM;
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001388 goto out;
Christoph Lameter742755a2006-06-23 02:03:55 -07001389 }
David Howellsc69e8d92008-11-14 10:39:19 +11001390 rcu_read_unlock();
Christoph Lameter742755a2006-06-23 02:03:55 -07001391
David Quigley86c3a762006-06-23 02:04:02 -07001392 err = security_task_movememory(task);
1393 if (err)
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001394 goto out;
David Quigley86c3a762006-06-23 02:04:02 -07001395
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001396 if (nodes) {
1397 err = do_pages_move(mm, task, nr_pages, pages, nodes, status,
1398 flags);
1399 } else {
Brice Goglin2f007e72008-10-18 20:27:16 -07001400 err = do_pages_stat(mm, nr_pages, pages, status);
Brice Goglin2f007e72008-10-18 20:27:16 -07001401 }
David Quigley86c3a762006-06-23 02:04:02 -07001402
Christoph Lameter742755a2006-06-23 02:03:55 -07001403out:
Christoph Lameter742755a2006-06-23 02:03:55 -07001404 mmput(mm);
1405 return err;
1406}
Christoph Lameter742755a2006-06-23 02:03:55 -07001407
Christoph Lameter7b2259b2006-06-25 05:46:48 -07001408/*
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.
1412 */
1413int migrate_vmas(struct mm_struct *mm, const nodemask_t *to,
1414 const nodemask_t *from, unsigned long flags)
1415{
1416 struct vm_area_struct *vma;
1417 int err = 0;
1418
Daisuke Nishimura1001c9f2009-02-11 13:04:18 -08001419 for (vma = mm->mmap; vma && !err; vma = vma->vm_next) {
Christoph Lameter7b2259b2006-06-25 05:46:48 -07001420 if (vma->vm_ops && vma->vm_ops->migrate) {
1421 err = vma->vm_ops->migrate(vma, to, from, flags);
1422 if (err)
1423 break;
1424 }
1425 }
1426 return err;
1427}
Gerald Schaefer83d16742008-07-23 21:28:22 -07001428#endif