blob: be26d5cbe56b34d63f8c8ac8b799782f9ef6424b [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 */
Hugh Dickins752dc182012-06-02 00:27:47 -0700439 if (PageSwapBacked(page))
440 SetPageDirty(newpage);
441 else
442 __set_page_dirty_nobuffers(newpage);
Christoph Lameterb20a3502006-03-22 00:09:12 -0800443 }
444
Nick Pigginb291f002008-10-18 20:26:44 -0700445 mlock_migrate_page(newpage, page);
Hugh Dickinse9995ef2009-12-14 17:59:31 -0800446 ksm_migrate_page(newpage, page);
Nick Pigginb291f002008-10-18 20:26:44 -0700447
Christoph Lameterb20a3502006-03-22 00:09:12 -0800448 ClearPageSwapCache(page);
Christoph Lameterb20a3502006-03-22 00:09:12 -0800449 ClearPagePrivate(page);
450 set_page_private(page, 0);
Christoph Lameterb20a3502006-03-22 00:09:12 -0800451
452 /*
453 * If any waiters have accumulated on the new page then
454 * wake them up.
455 */
456 if (PageWriteback(newpage))
457 end_page_writeback(newpage);
458}
Christoph Lameterb20a3502006-03-22 00:09:12 -0800459
Christoph Lameter1d8b85c2006-06-23 02:03:28 -0700460/************************************************************
461 * Migration functions
462 ***********************************************************/
463
464/* Always fail migration. Used for mappings that are not movable */
Christoph Lameter2d1db3b2006-06-23 02:03:33 -0700465int fail_migrate_page(struct address_space *mapping,
466 struct page *newpage, struct page *page)
Christoph Lameter1d8b85c2006-06-23 02:03:28 -0700467{
468 return -EIO;
469}
470EXPORT_SYMBOL(fail_migrate_page);
471
Christoph Lameterb20a3502006-03-22 00:09:12 -0800472/*
473 * Common logic to directly migrate a single page suitable for
David Howells266cf652009-04-03 16:42:36 +0100474 * pages that do not use PagePrivate/PagePrivate2.
Christoph Lameterb20a3502006-03-22 00:09:12 -0800475 *
476 * Pages are locked upon entry and exit.
477 */
Christoph Lameter2d1db3b2006-06-23 02:03:33 -0700478int migrate_page(struct address_space *mapping,
Mel Gormana6bc32b2012-01-12 17:19:43 -0800479 struct page *newpage, struct page *page,
480 enum migrate_mode mode)
Christoph Lameterb20a3502006-03-22 00:09:12 -0800481{
482 int rc;
483
484 BUG_ON(PageWriteback(page)); /* Writeback must be complete */
485
Mel Gormana6bc32b2012-01-12 17:19:43 -0800486 rc = migrate_page_move_mapping(mapping, newpage, page, NULL, mode);
Christoph Lameterb20a3502006-03-22 00:09:12 -0800487
488 if (rc)
489 return rc;
490
491 migrate_page_copy(newpage, page);
Christoph Lameterb20a3502006-03-22 00:09:12 -0800492 return 0;
493}
494EXPORT_SYMBOL(migrate_page);
495
David Howells93614012006-09-30 20:45:40 +0200496#ifdef CONFIG_BLOCK
Christoph Lameterb20a3502006-03-22 00:09:12 -0800497/*
Christoph Lameter1d8b85c2006-06-23 02:03:28 -0700498 * Migration function for pages with buffers. This function can only be used
499 * if the underlying filesystem guarantees that no other references to "page"
500 * exist.
501 */
Christoph Lameter2d1db3b2006-06-23 02:03:33 -0700502int buffer_migrate_page(struct address_space *mapping,
Mel Gormana6bc32b2012-01-12 17:19:43 -0800503 struct page *newpage, struct page *page, enum migrate_mode mode)
Christoph Lameter1d8b85c2006-06-23 02:03:28 -0700504{
Christoph Lameter1d8b85c2006-06-23 02:03:28 -0700505 struct buffer_head *bh, *head;
506 int rc;
507
Christoph Lameter1d8b85c2006-06-23 02:03:28 -0700508 if (!page_has_buffers(page))
Mel Gormana6bc32b2012-01-12 17:19:43 -0800509 return migrate_page(mapping, newpage, page, mode);
Christoph Lameter1d8b85c2006-06-23 02:03:28 -0700510
511 head = page_buffers(page);
512
Mel Gormana6bc32b2012-01-12 17:19:43 -0800513 rc = migrate_page_move_mapping(mapping, newpage, page, head, mode);
Christoph Lameter1d8b85c2006-06-23 02:03:28 -0700514
515 if (rc)
516 return rc;
517
Mel Gormanb969c4a2012-01-12 17:19:34 -0800518 /*
519 * In the async case, migrate_page_move_mapping locked the buffers
520 * with an IRQ-safe spinlock held. In the sync case, the buffers
521 * need to be locked now
522 */
Mel Gormana6bc32b2012-01-12 17:19:43 -0800523 if (mode != MIGRATE_ASYNC)
524 BUG_ON(!buffer_migrate_lock_buffers(head, mode));
Christoph Lameter1d8b85c2006-06-23 02:03:28 -0700525
526 ClearPagePrivate(page);
527 set_page_private(newpage, page_private(page));
528 set_page_private(page, 0);
529 put_page(page);
530 get_page(newpage);
531
532 bh = head;
533 do {
534 set_bh_page(bh, newpage, bh_offset(bh));
535 bh = bh->b_this_page;
536
537 } while (bh != head);
538
539 SetPagePrivate(newpage);
540
541 migrate_page_copy(newpage, page);
542
543 bh = head;
544 do {
545 unlock_buffer(bh);
546 put_bh(bh);
547 bh = bh->b_this_page;
548
549 } while (bh != head);
550
551 return 0;
552}
553EXPORT_SYMBOL(buffer_migrate_page);
David Howells93614012006-09-30 20:45:40 +0200554#endif
Christoph Lameter1d8b85c2006-06-23 02:03:28 -0700555
Christoph Lameter04e62a22006-06-23 02:03:38 -0700556/*
557 * Writeback a page to clean the dirty state
558 */
559static int writeout(struct address_space *mapping, struct page *page)
560{
561 struct writeback_control wbc = {
562 .sync_mode = WB_SYNC_NONE,
563 .nr_to_write = 1,
564 .range_start = 0,
565 .range_end = LLONG_MAX,
Christoph Lameter04e62a22006-06-23 02:03:38 -0700566 .for_reclaim = 1
567 };
568 int rc;
569
570 if (!mapping->a_ops->writepage)
571 /* No write method for the address space */
572 return -EINVAL;
573
574 if (!clear_page_dirty_for_io(page))
575 /* Someone else already triggered a write */
576 return -EAGAIN;
577
578 /*
579 * A dirty page may imply that the underlying filesystem has
580 * the page on some queue. So the page must be clean for
581 * migration. Writeout may mean we loose the lock and the
582 * page state is no longer what we checked for earlier.
583 * At this point we know that the migration attempt cannot
584 * be successful.
585 */
586 remove_migration_ptes(page, page);
587
588 rc = mapping->a_ops->writepage(page, &wbc);
Christoph Lameter04e62a22006-06-23 02:03:38 -0700589
590 if (rc != AOP_WRITEPAGE_ACTIVATE)
591 /* unlocked. Relock */
592 lock_page(page);
593
Hugh Dickinsbda85502008-11-19 15:36:36 -0800594 return (rc < 0) ? -EIO : -EAGAIN;
Christoph Lameter04e62a22006-06-23 02:03:38 -0700595}
596
597/*
598 * Default handling if a filesystem does not provide a migration function.
599 */
Christoph Lameter8351a6e2006-06-23 02:03:33 -0700600static int fallback_migrate_page(struct address_space *mapping,
Mel Gormana6bc32b2012-01-12 17:19:43 -0800601 struct page *newpage, struct page *page, enum migrate_mode mode)
Christoph Lameter8351a6e2006-06-23 02:03:33 -0700602{
Mel Gormanb969c4a2012-01-12 17:19:34 -0800603 if (PageDirty(page)) {
Mel Gormana6bc32b2012-01-12 17:19:43 -0800604 /* Only writeback pages in full synchronous migration */
605 if (mode != MIGRATE_SYNC)
Mel Gormanb969c4a2012-01-12 17:19:34 -0800606 return -EBUSY;
Christoph Lameter04e62a22006-06-23 02:03:38 -0700607 return writeout(mapping, page);
Mel Gormanb969c4a2012-01-12 17:19:34 -0800608 }
Christoph Lameter8351a6e2006-06-23 02:03:33 -0700609
610 /*
611 * Buffers may be managed in a filesystem specific way.
612 * We must have no buffers or drop them.
613 */
David Howells266cf652009-04-03 16:42:36 +0100614 if (page_has_private(page) &&
Christoph Lameter8351a6e2006-06-23 02:03:33 -0700615 !try_to_release_page(page, GFP_KERNEL))
616 return -EAGAIN;
617
Mel Gormana6bc32b2012-01-12 17:19:43 -0800618 return migrate_page(mapping, newpage, page, mode);
Christoph Lameter8351a6e2006-06-23 02:03:33 -0700619}
620
Christoph Lameter1d8b85c2006-06-23 02:03:28 -0700621/*
Christoph Lametere24f0b82006-06-23 02:03:51 -0700622 * Move a page to a newly allocated page
623 * The page is locked and all ptes have been successfully removed.
624 *
625 * The new page will have replaced the old page if this function
626 * is successful.
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700627 *
628 * Return value:
629 * < 0 - error code
630 * == 0 - success
Christoph Lametere24f0b82006-06-23 02:03:51 -0700631 */
Mel Gorman3fe20112010-05-24 14:32:20 -0700632static int move_to_new_page(struct page *newpage, struct page *page,
Mel Gormana6bc32b2012-01-12 17:19:43 -0800633 int remap_swapcache, enum migrate_mode mode)
Christoph Lametere24f0b82006-06-23 02:03:51 -0700634{
635 struct address_space *mapping;
636 int rc;
637
638 /*
639 * Block others from accessing the page when we get around to
640 * establishing additional references. We are the only one
641 * holding a reference to the new page at this point.
642 */
Nick Piggin529ae9a2008-08-02 12:01:03 +0200643 if (!trylock_page(newpage))
Christoph Lametere24f0b82006-06-23 02:03:51 -0700644 BUG();
645
646 /* Prepare mapping for the new page.*/
647 newpage->index = page->index;
648 newpage->mapping = page->mapping;
Rik van Rielb2e18532008-10-18 20:26:30 -0700649 if (PageSwapBacked(page))
650 SetPageSwapBacked(newpage);
Christoph Lametere24f0b82006-06-23 02:03:51 -0700651
652 mapping = page_mapping(page);
653 if (!mapping)
Mel Gormana6bc32b2012-01-12 17:19:43 -0800654 rc = migrate_page(mapping, newpage, page, mode);
Mel Gormanb969c4a2012-01-12 17:19:34 -0800655 else if (mapping->a_ops->migratepage)
Christoph Lametere24f0b82006-06-23 02:03:51 -0700656 /*
Mel Gormanb969c4a2012-01-12 17:19:34 -0800657 * Most pages have a mapping and most filesystems provide a
658 * migratepage callback. Anonymous pages are part of swap
659 * space which also has its own migratepage callback. This
660 * is the most common path for page migration.
Christoph Lametere24f0b82006-06-23 02:03:51 -0700661 */
Mel Gormanb969c4a2012-01-12 17:19:34 -0800662 rc = mapping->a_ops->migratepage(mapping,
Mel Gormana6bc32b2012-01-12 17:19:43 -0800663 newpage, page, mode);
Mel Gormanb969c4a2012-01-12 17:19:34 -0800664 else
Mel Gormana6bc32b2012-01-12 17:19:43 -0800665 rc = fallback_migrate_page(mapping, newpage, page, mode);
Christoph Lametere24f0b82006-06-23 02:03:51 -0700666
Mel Gorman3fe20112010-05-24 14:32:20 -0700667 if (rc) {
Christoph Lametere24f0b82006-06-23 02:03:51 -0700668 newpage->mapping = NULL;
Mel Gorman3fe20112010-05-24 14:32:20 -0700669 } else {
670 if (remap_swapcache)
671 remove_migration_ptes(page, newpage);
Konstantin Khlebnikov35512ec2012-02-03 15:37:13 -0800672 page->mapping = NULL;
Mel Gorman3fe20112010-05-24 14:32:20 -0700673 }
Christoph Lametere24f0b82006-06-23 02:03:51 -0700674
675 unlock_page(newpage);
676
677 return rc;
678}
679
Minchan Kim0dabec92011-10-31 17:06:57 -0700680static int __unmap_and_move(struct page *page, struct page *newpage,
Mel Gormana6bc32b2012-01-12 17:19:43 -0800681 int force, bool offlining, enum migrate_mode mode)
Christoph Lametere24f0b82006-06-23 02:03:51 -0700682{
Minchan Kim0dabec92011-10-31 17:06:57 -0700683 int rc = -EAGAIN;
Mel Gorman3fe20112010-05-24 14:32:20 -0700684 int remap_swapcache = 1;
KAMEZAWA Hiroyukiae41be32008-02-07 00:14:10 -0800685 int charge = 0;
KAMEZAWA Hiroyuki56039ef2011-03-23 16:42:19 -0700686 struct mem_cgroup *mem;
Mel Gorman3f6c8272010-05-24 14:32:17 -0700687 struct anon_vma *anon_vma = NULL;
Christoph Lameter95a402c2006-06-23 02:03:53 -0700688
Nick Piggin529ae9a2008-08-02 12:01:03 +0200689 if (!trylock_page(page)) {
Mel Gormana6bc32b2012-01-12 17:19:43 -0800690 if (!force || mode == MIGRATE_ASYNC)
Minchan Kim0dabec92011-10-31 17:06:57 -0700691 goto out;
Mel Gorman3e7d3442011-01-13 15:45:56 -0800692
693 /*
694 * It's not safe for direct compaction to call lock_page.
695 * For example, during page readahead pages are added locked
696 * to the LRU. Later, when the IO completes the pages are
697 * marked uptodate and unlocked. However, the queueing
698 * could be merging multiple pages for one bio (e.g.
699 * mpage_readpages). If an allocation happens for the
700 * second or third page, the process can end up locking
701 * the same page twice and deadlocking. Rather than
702 * trying to be clever about what pages can be locked,
703 * avoid the use of lock_page for direct compaction
704 * altogether.
705 */
706 if (current->flags & PF_MEMALLOC)
Minchan Kim0dabec92011-10-31 17:06:57 -0700707 goto out;
Mel Gorman3e7d3442011-01-13 15:45:56 -0800708
Christoph Lametere24f0b82006-06-23 02:03:51 -0700709 lock_page(page);
710 }
711
Hugh Dickins62b61f62009-12-14 17:59:33 -0800712 /*
713 * Only memory hotplug's offline_pages() caller has locked out KSM,
714 * and can safely migrate a KSM page. The other cases have skipped
715 * PageKsm along with PageReserved - but it is only now when we have
716 * the page lock that we can be certain it will not go KSM beneath us
717 * (KSM will not upgrade a page from PageAnon to PageKsm when it sees
718 * its pagecount raised, but only here do we take the page lock which
719 * serializes that).
720 */
721 if (PageKsm(page) && !offlining) {
722 rc = -EBUSY;
723 goto unlock;
724 }
725
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -0800726 /* charge against new page */
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700727 charge = mem_cgroup_prepare_migration(page, newpage, &mem, GFP_KERNEL);
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -0800728 if (charge == -ENOMEM) {
729 rc = -ENOMEM;
730 goto unlock;
731 }
732 BUG_ON(charge);
733
Christoph Lametere24f0b82006-06-23 02:03:51 -0700734 if (PageWriteback(page)) {
Andrea Arcangeli11bc82d2011-03-22 16:33:11 -0700735 /*
Mel Gormana6bc32b2012-01-12 17:19:43 -0800736 * Only in the case of a full syncronous migration is it
737 * necessary to wait for PageWriteback. In the async case,
738 * the retry loop is too short and in the sync-light case,
739 * the overhead of stalling is too much
Andrea Arcangeli11bc82d2011-03-22 16:33:11 -0700740 */
Mel Gormana6bc32b2012-01-12 17:19:43 -0800741 if (mode != MIGRATE_SYNC) {
Andrea Arcangeli11bc82d2011-03-22 16:33:11 -0700742 rc = -EBUSY;
743 goto uncharge;
744 }
745 if (!force)
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -0800746 goto uncharge;
Christoph Lametere24f0b82006-06-23 02:03:51 -0700747 wait_on_page_writeback(page);
748 }
Christoph Lametere24f0b82006-06-23 02:03:51 -0700749 /*
KAMEZAWA Hiroyukidc386d42007-07-26 10:41:07 -0700750 * By try_to_unmap(), page->mapcount goes down to 0 here. In this case,
751 * we cannot notice that anon_vma is freed while we migrates a page.
Hugh Dickins1ce82b62011-01-13 15:47:30 -0800752 * This get_anon_vma() delays freeing anon_vma pointer until the end
KAMEZAWA Hiroyukidc386d42007-07-26 10:41:07 -0700753 * of migration. File cache pages are no problem because of page_lock()
KAMEZAWA Hiroyuki989f89c2007-08-30 23:56:21 -0700754 * File Caches may use write_page() or lock_page() in migration, then,
755 * just care Anon page here.
Christoph Lametere24f0b82006-06-23 02:03:51 -0700756 */
KAMEZAWA Hiroyuki989f89c2007-08-30 23:56:21 -0700757 if (PageAnon(page)) {
Hugh Dickins1ce82b62011-01-13 15:47:30 -0800758 /*
759 * Only page_lock_anon_vma() understands the subtleties of
760 * getting a hold on an anon_vma from outside one of its mms.
761 */
Peter Zijlstra746b18d2011-05-24 17:12:10 -0700762 anon_vma = page_get_anon_vma(page);
Hugh Dickins1ce82b62011-01-13 15:47:30 -0800763 if (anon_vma) {
764 /*
Peter Zijlstra746b18d2011-05-24 17:12:10 -0700765 * Anon page
Hugh Dickins1ce82b62011-01-13 15:47:30 -0800766 */
Hugh Dickins1ce82b62011-01-13 15:47:30 -0800767 } else if (PageSwapCache(page)) {
Mel Gorman3fe20112010-05-24 14:32:20 -0700768 /*
769 * We cannot be sure that the anon_vma of an unmapped
770 * swapcache page is safe to use because we don't
771 * know in advance if the VMA that this page belonged
772 * to still exists. If the VMA and others sharing the
773 * data have been freed, then the anon_vma could
774 * already be invalid.
775 *
776 * To avoid this possibility, swapcache pages get
777 * migrated but are not remapped when migration
778 * completes
779 */
780 remap_swapcache = 0;
781 } else {
Hugh Dickins1ce82b62011-01-13 15:47:30 -0800782 goto uncharge;
Mel Gorman3fe20112010-05-24 14:32:20 -0700783 }
KAMEZAWA Hiroyuki989f89c2007-08-30 23:56:21 -0700784 }
Shaohua Li62e1c552008-02-04 22:29:33 -0800785
KAMEZAWA Hiroyukidc386d42007-07-26 10:41:07 -0700786 /*
Shaohua Li62e1c552008-02-04 22:29:33 -0800787 * Corner case handling:
788 * 1. When a new swap-cache page is read into, it is added to the LRU
789 * and treated as swapcache but it has no rmap yet.
790 * Calling try_to_unmap() against a page->mapping==NULL page will
791 * trigger a BUG. So handle it here.
792 * 2. An orphaned page (see truncate_complete_page) might have
793 * fs-private metadata. The page can be picked up due to memory
794 * offlining. Everywhere else except page reclaim, the page is
795 * invisible to the vm, so the page can not be migrated. So try to
796 * free the metadata, so the page can be freed.
KAMEZAWA Hiroyukidc386d42007-07-26 10:41:07 -0700797 */
Shaohua Li62e1c552008-02-04 22:29:33 -0800798 if (!page->mapping) {
Hugh Dickins1ce82b62011-01-13 15:47:30 -0800799 VM_BUG_ON(PageAnon(page));
800 if (page_has_private(page)) {
Shaohua Li62e1c552008-02-04 22:29:33 -0800801 try_to_free_buffers(page);
Hugh Dickins1ce82b62011-01-13 15:47:30 -0800802 goto uncharge;
Shaohua Li62e1c552008-02-04 22:29:33 -0800803 }
Shaohua Liabfc3482009-09-21 17:01:19 -0700804 goto skip_unmap;
Shaohua Li62e1c552008-02-04 22:29:33 -0800805 }
806
KAMEZAWA Hiroyukidc386d42007-07-26 10:41:07 -0700807 /* Establish migration ptes or remove ptes */
Andi Kleen14fa31b2009-09-16 11:50:10 +0200808 try_to_unmap(page, TTU_MIGRATION|TTU_IGNORE_MLOCK|TTU_IGNORE_ACCESS);
KAMEZAWA Hiroyukidc386d42007-07-26 10:41:07 -0700809
Shaohua Liabfc3482009-09-21 17:01:19 -0700810skip_unmap:
Christoph Lametere6a15302006-06-25 05:46:49 -0700811 if (!page_mapped(page))
Mel Gormana6bc32b2012-01-12 17:19:43 -0800812 rc = move_to_new_page(newpage, page, remap_swapcache, mode);
Christoph Lametere24f0b82006-06-23 02:03:51 -0700813
Mel Gorman3fe20112010-05-24 14:32:20 -0700814 if (rc && remap_swapcache)
Christoph Lametere24f0b82006-06-23 02:03:51 -0700815 remove_migration_ptes(page, page);
Mel Gorman3f6c8272010-05-24 14:32:17 -0700816
817 /* Drop an anon_vma reference if we took one */
Rik van Riel76545062010-08-09 17:18:41 -0700818 if (anon_vma)
Peter Zijlstra9e601092011-03-22 16:32:46 -0700819 put_anon_vma(anon_vma);
Mel Gorman3f6c8272010-05-24 14:32:17 -0700820
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -0800821uncharge:
822 if (!charge)
Daisuke Nishimura50de1dd2011-01-13 15:47:43 -0800823 mem_cgroup_end_migration(mem, page, newpage, rc == 0);
Christoph Lametere24f0b82006-06-23 02:03:51 -0700824unlock:
825 unlock_page(page);
Minchan Kim0dabec92011-10-31 17:06:57 -0700826out:
827 return rc;
828}
Christoph Lameter95a402c2006-06-23 02:03:53 -0700829
Minchan Kim0dabec92011-10-31 17:06:57 -0700830/*
831 * Obtain the lock on page, remove all ptes and migrate the page
832 * to the newly allocated page in newpage.
833 */
834static int unmap_and_move(new_page_t get_new_page, unsigned long private,
Mel Gormana6bc32b2012-01-12 17:19:43 -0800835 struct page *page, int force, bool offlining,
836 enum migrate_mode mode)
Minchan Kim0dabec92011-10-31 17:06:57 -0700837{
838 int rc = 0;
839 int *result = NULL;
840 struct page *newpage = get_new_page(page, private, &result);
841
842 if (!newpage)
843 return -ENOMEM;
844
845 if (page_count(page) == 1) {
846 /* page was freed from under us. So we are done. */
847 goto out;
848 }
849
850 if (unlikely(PageTransHuge(page)))
851 if (unlikely(split_huge_page(page)))
852 goto out;
853
Mel Gormana6bc32b2012-01-12 17:19:43 -0800854 rc = __unmap_and_move(page, newpage, force, offlining, mode);
Minchan Kim0dabec92011-10-31 17:06:57 -0700855out:
Christoph Lametere24f0b82006-06-23 02:03:51 -0700856 if (rc != -EAGAIN) {
Minchan Kim0dabec92011-10-31 17:06:57 -0700857 /*
858 * A page that has been migrated has all references
859 * removed and will be freed. A page that has not been
860 * migrated will have kepts its references and be
861 * restored.
862 */
863 list_del(&page->lru);
KOSAKI Motohiroa7312862009-09-21 17:01:37 -0700864 dec_zone_page_state(page, NR_ISOLATED_ANON +
Johannes Weiner6c0b1352009-09-21 17:02:59 -0700865 page_is_file_cache(page));
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700866 putback_lru_page(page);
Christoph Lametere24f0b82006-06-23 02:03:51 -0700867 }
Christoph Lameter95a402c2006-06-23 02:03:53 -0700868 /*
869 * Move the new page to the LRU. If migration was not successful
870 * then this will free the page.
871 */
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700872 putback_lru_page(newpage);
Christoph Lameter742755a2006-06-23 02:03:55 -0700873 if (result) {
874 if (rc)
875 *result = rc;
876 else
877 *result = page_to_nid(newpage);
878 }
Christoph Lametere24f0b82006-06-23 02:03:51 -0700879 return rc;
880}
881
882/*
Naoya Horiguchi290408d2010-09-08 10:19:35 +0900883 * Counterpart of unmap_and_move_page() for hugepage migration.
884 *
885 * This function doesn't wait the completion of hugepage I/O
886 * because there is no race between I/O and migration for hugepage.
887 * Note that currently hugepage I/O occurs only in direct I/O
888 * where no lock is held and PG_writeback is irrelevant,
889 * and writeback status of all subpages are counted in the reference
890 * count of the head page (i.e. if all subpages of a 2MB hugepage are
891 * under direct I/O, the reference of the head page is 512 and a bit more.)
892 * This means that when we try to migrate hugepage whose subpages are
893 * doing direct I/O, some references remain after try_to_unmap() and
894 * hugepage migration fails without data corruption.
895 *
896 * There is also no race when direct I/O is issued on the page under migration,
897 * because then pte is replaced with migration swap entry and direct I/O code
898 * will wait in the page fault for migration to complete.
899 */
900static int unmap_and_move_huge_page(new_page_t get_new_page,
901 unsigned long private, struct page *hpage,
Mel Gormana6bc32b2012-01-12 17:19:43 -0800902 int force, bool offlining,
903 enum migrate_mode mode)
Naoya Horiguchi290408d2010-09-08 10:19:35 +0900904{
905 int rc = 0;
906 int *result = NULL;
907 struct page *new_hpage = get_new_page(hpage, private, &result);
Naoya Horiguchi290408d2010-09-08 10:19:35 +0900908 struct anon_vma *anon_vma = NULL;
909
910 if (!new_hpage)
911 return -ENOMEM;
912
913 rc = -EAGAIN;
914
915 if (!trylock_page(hpage)) {
Mel Gormana6bc32b2012-01-12 17:19:43 -0800916 if (!force || mode != MIGRATE_SYNC)
Naoya Horiguchi290408d2010-09-08 10:19:35 +0900917 goto out;
918 lock_page(hpage);
919 }
920
Peter Zijlstra746b18d2011-05-24 17:12:10 -0700921 if (PageAnon(hpage))
922 anon_vma = page_get_anon_vma(hpage);
Naoya Horiguchi290408d2010-09-08 10:19:35 +0900923
924 try_to_unmap(hpage, TTU_MIGRATION|TTU_IGNORE_MLOCK|TTU_IGNORE_ACCESS);
925
926 if (!page_mapped(hpage))
Mel Gormana6bc32b2012-01-12 17:19:43 -0800927 rc = move_to_new_page(new_hpage, hpage, 1, mode);
Naoya Horiguchi290408d2010-09-08 10:19:35 +0900928
929 if (rc)
930 remove_migration_ptes(hpage, hpage);
931
Hugh Dickinsfd4a4662011-01-13 15:47:31 -0800932 if (anon_vma)
Peter Zijlstra9e601092011-03-22 16:32:46 -0700933 put_anon_vma(anon_vma);
Naoya Horiguchi290408d2010-09-08 10:19:35 +0900934 unlock_page(hpage);
935
Hillf Danton09761332011-12-08 14:34:20 -0800936out:
Naoya Horiguchi290408d2010-09-08 10:19:35 +0900937 if (rc != -EAGAIN) {
938 list_del(&hpage->lru);
939 put_page(hpage);
940 }
941
942 put_page(new_hpage);
943
944 if (result) {
945 if (rc)
946 *result = rc;
947 else
948 *result = page_to_nid(new_hpage);
949 }
950 return rc;
951}
952
953/*
Christoph Lameterb20a3502006-03-22 00:09:12 -0800954 * migrate_pages
955 *
Christoph Lameter95a402c2006-06-23 02:03:53 -0700956 * The function takes one list of pages to migrate and a function
957 * that determines from the page to be migrated and the private data
958 * the target of the move and allocates the page.
Christoph Lameterb20a3502006-03-22 00:09:12 -0800959 *
960 * The function returns after 10 attempts or if no pages
961 * are movable anymore because to has become empty
Minchan Kimcf608ac2010-10-26 14:21:29 -0700962 * or no retryable pages exist anymore.
963 * Caller should call putback_lru_pages to return pages to the LRU
Minchan Kim28bd6572011-01-25 15:07:26 -0800964 * or free list only if ret != 0.
Christoph Lameterb20a3502006-03-22 00:09:12 -0800965 *
Christoph Lameter95a402c2006-06-23 02:03:53 -0700966 * Return: Number of pages not migrated or error code.
Christoph Lameterb20a3502006-03-22 00:09:12 -0800967 */
Christoph Lameter95a402c2006-06-23 02:03:53 -0700968int migrate_pages(struct list_head *from,
Mel Gorman7f0f2492011-01-13 15:45:58 -0800969 new_page_t get_new_page, unsigned long private, bool offlining,
Mel Gormana6bc32b2012-01-12 17:19:43 -0800970 enum migrate_mode mode)
Christoph Lameterb20a3502006-03-22 00:09:12 -0800971{
Christoph Lametere24f0b82006-06-23 02:03:51 -0700972 int retry = 1;
Christoph Lameterb20a3502006-03-22 00:09:12 -0800973 int nr_failed = 0;
974 int pass = 0;
975 struct page *page;
976 struct page *page2;
977 int swapwrite = current->flags & PF_SWAPWRITE;
978 int rc;
979
980 if (!swapwrite)
981 current->flags |= PF_SWAPWRITE;
982
Christoph Lametere24f0b82006-06-23 02:03:51 -0700983 for(pass = 0; pass < 10 && retry; pass++) {
984 retry = 0;
Christoph Lameterb20a3502006-03-22 00:09:12 -0800985
Christoph Lametere24f0b82006-06-23 02:03:51 -0700986 list_for_each_entry_safe(page, page2, from, lru) {
Christoph Lametere24f0b82006-06-23 02:03:51 -0700987 cond_resched();
Christoph Lameterb20a3502006-03-22 00:09:12 -0800988
Christoph Lameter95a402c2006-06-23 02:03:53 -0700989 rc = unmap_and_move(get_new_page, private,
Mel Gorman77f1fe62011-01-13 15:45:57 -0800990 page, pass > 2, offlining,
Mel Gormana6bc32b2012-01-12 17:19:43 -0800991 mode);
Christoph Lameterb20a3502006-03-22 00:09:12 -0800992
Christoph Lametere24f0b82006-06-23 02:03:51 -0700993 switch(rc) {
Christoph Lameter95a402c2006-06-23 02:03:53 -0700994 case -ENOMEM:
995 goto out;
Christoph Lametere24f0b82006-06-23 02:03:51 -0700996 case -EAGAIN:
Christoph Lameter2d1db3b2006-06-23 02:03:33 -0700997 retry++;
Christoph Lametere24f0b82006-06-23 02:03:51 -0700998 break;
999 case 0:
Christoph Lametere24f0b82006-06-23 02:03:51 -07001000 break;
1001 default:
Christoph Lameter2d1db3b2006-06-23 02:03:33 -07001002 /* Permanent failure */
Christoph Lameter2d1db3b2006-06-23 02:03:33 -07001003 nr_failed++;
Christoph Lametere24f0b82006-06-23 02:03:51 -07001004 break;
Christoph Lameter2d1db3b2006-06-23 02:03:33 -07001005 }
Christoph Lameterb20a3502006-03-22 00:09:12 -08001006 }
1007 }
Christoph Lameter95a402c2006-06-23 02:03:53 -07001008 rc = 0;
1009out:
Christoph Lameterb20a3502006-03-22 00:09:12 -08001010 if (!swapwrite)
1011 current->flags &= ~PF_SWAPWRITE;
1012
Christoph Lameter95a402c2006-06-23 02:03:53 -07001013 if (rc)
1014 return rc;
1015
Christoph Lameterb20a3502006-03-22 00:09:12 -08001016 return nr_failed + retry;
1017}
1018
Naoya Horiguchi290408d2010-09-08 10:19:35 +09001019int migrate_huge_pages(struct list_head *from,
Mel Gorman7f0f2492011-01-13 15:45:58 -08001020 new_page_t get_new_page, unsigned long private, bool offlining,
Mel Gormana6bc32b2012-01-12 17:19:43 -08001021 enum migrate_mode mode)
Naoya Horiguchi290408d2010-09-08 10:19:35 +09001022{
1023 int retry = 1;
1024 int nr_failed = 0;
1025 int pass = 0;
1026 struct page *page;
1027 struct page *page2;
1028 int rc;
1029
1030 for (pass = 0; pass < 10 && retry; pass++) {
1031 retry = 0;
1032
1033 list_for_each_entry_safe(page, page2, from, lru) {
1034 cond_resched();
1035
1036 rc = unmap_and_move_huge_page(get_new_page,
Mel Gorman77f1fe62011-01-13 15:45:57 -08001037 private, page, pass > 2, offlining,
Mel Gormana6bc32b2012-01-12 17:19:43 -08001038 mode);
Naoya Horiguchi290408d2010-09-08 10:19:35 +09001039
1040 switch(rc) {
1041 case -ENOMEM:
1042 goto out;
1043 case -EAGAIN:
1044 retry++;
1045 break;
1046 case 0:
1047 break;
1048 default:
1049 /* Permanent failure */
1050 nr_failed++;
1051 break;
1052 }
1053 }
1054 }
1055 rc = 0;
1056out:
Naoya Horiguchi290408d2010-09-08 10:19:35 +09001057 if (rc)
1058 return rc;
1059
1060 return nr_failed + retry;
1061}
1062
Christoph Lameter742755a2006-06-23 02:03:55 -07001063#ifdef CONFIG_NUMA
1064/*
1065 * Move a list of individual pages
1066 */
1067struct page_to_node {
1068 unsigned long addr;
1069 struct page *page;
1070 int node;
1071 int status;
1072};
1073
1074static struct page *new_page_node(struct page *p, unsigned long private,
1075 int **result)
1076{
1077 struct page_to_node *pm = (struct page_to_node *)private;
1078
1079 while (pm->node != MAX_NUMNODES && pm->page != p)
1080 pm++;
1081
1082 if (pm->node == MAX_NUMNODES)
1083 return NULL;
1084
1085 *result = &pm->status;
1086
Mel Gorman6484eb32009-06-16 15:31:54 -07001087 return alloc_pages_exact_node(pm->node,
Mel Gorman769848c2007-07-17 04:03:05 -07001088 GFP_HIGHUSER_MOVABLE | GFP_THISNODE, 0);
Christoph Lameter742755a2006-06-23 02:03:55 -07001089}
1090
1091/*
1092 * Move a set of pages as indicated in the pm array. The addr
1093 * field must be set to the virtual address of the page to be moved
1094 * and the node number must contain a valid target node.
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001095 * The pm array ends with node = MAX_NUMNODES.
Christoph Lameter742755a2006-06-23 02:03:55 -07001096 */
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001097static int do_move_page_to_node_array(struct mm_struct *mm,
1098 struct page_to_node *pm,
1099 int migrate_all)
Christoph Lameter742755a2006-06-23 02:03:55 -07001100{
1101 int err;
1102 struct page_to_node *pp;
1103 LIST_HEAD(pagelist);
1104
1105 down_read(&mm->mmap_sem);
1106
1107 /*
1108 * Build a list of pages to migrate
1109 */
Christoph Lameter742755a2006-06-23 02:03:55 -07001110 for (pp = pm; pp->node != MAX_NUMNODES; pp++) {
1111 struct vm_area_struct *vma;
1112 struct page *page;
1113
Christoph Lameter742755a2006-06-23 02:03:55 -07001114 err = -EFAULT;
1115 vma = find_vma(mm, pp->addr);
Gleb Natapov70384dc2010-10-26 14:22:07 -07001116 if (!vma || pp->addr < vma->vm_start || !vma_migratable(vma))
Christoph Lameter742755a2006-06-23 02:03:55 -07001117 goto set_status;
1118
Andrea Arcangeli500d65d2011-01-13 15:46:55 -08001119 page = follow_page(vma, pp->addr, FOLL_GET|FOLL_SPLIT);
Linus Torvalds89f5b7d2008-06-20 11:18:25 -07001120
1121 err = PTR_ERR(page);
1122 if (IS_ERR(page))
1123 goto set_status;
1124
Christoph Lameter742755a2006-06-23 02:03:55 -07001125 err = -ENOENT;
1126 if (!page)
1127 goto set_status;
1128
Hugh Dickins62b61f62009-12-14 17:59:33 -08001129 /* Use PageReserved to check for zero page */
1130 if (PageReserved(page) || PageKsm(page))
Christoph Lameter742755a2006-06-23 02:03:55 -07001131 goto put_and_set;
1132
1133 pp->page = page;
1134 err = page_to_nid(page);
1135
1136 if (err == pp->node)
1137 /*
1138 * Node already in the right place
1139 */
1140 goto put_and_set;
1141
1142 err = -EACCES;
1143 if (page_mapcount(page) > 1 &&
1144 !migrate_all)
1145 goto put_and_set;
1146
Nick Piggin62695a82008-10-18 20:26:09 -07001147 err = isolate_lru_page(page);
KOSAKI Motohiro6d9c2852009-12-14 17:58:11 -08001148 if (!err) {
Nick Piggin62695a82008-10-18 20:26:09 -07001149 list_add_tail(&page->lru, &pagelist);
KOSAKI Motohiro6d9c2852009-12-14 17:58:11 -08001150 inc_zone_page_state(page, NR_ISOLATED_ANON +
1151 page_is_file_cache(page));
1152 }
Christoph Lameter742755a2006-06-23 02:03:55 -07001153put_and_set:
1154 /*
1155 * Either remove the duplicate refcount from
1156 * isolate_lru_page() or drop the page ref if it was
1157 * not isolated.
1158 */
1159 put_page(page);
1160set_status:
1161 pp->status = err;
1162 }
1163
Brice Gogline78bbfa2008-10-18 20:27:15 -07001164 err = 0;
Minchan Kimcf608ac2010-10-26 14:21:29 -07001165 if (!list_empty(&pagelist)) {
Christoph Lameter742755a2006-06-23 02:03:55 -07001166 err = migrate_pages(&pagelist, new_page_node,
Mel Gormana6bc32b2012-01-12 17:19:43 -08001167 (unsigned long)pm, 0, MIGRATE_SYNC);
Minchan Kimcf608ac2010-10-26 14:21:29 -07001168 if (err)
1169 putback_lru_pages(&pagelist);
1170 }
Christoph Lameter742755a2006-06-23 02:03:55 -07001171
1172 up_read(&mm->mmap_sem);
1173 return err;
1174}
1175
1176/*
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001177 * Migrate an array of page address onto an array of nodes and fill
1178 * the corresponding array of status.
1179 */
Christoph Lameter3268c632012-03-21 16:34:06 -07001180static int do_pages_move(struct mm_struct *mm, nodemask_t task_nodes,
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001181 unsigned long nr_pages,
1182 const void __user * __user *pages,
1183 const int __user *nodes,
1184 int __user *status, int flags)
1185{
Brice Goglin3140a222009-01-06 14:38:57 -08001186 struct page_to_node *pm;
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
Brice Goglin3140a222009-01-06 14:38:57 -08001191 err = -ENOMEM;
1192 pm = (struct page_to_node *)__get_free_page(GFP_KERNEL);
1193 if (!pm)
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001194 goto out;
Brice Goglin35282a22009-06-16 15:32:43 -07001195
1196 migrate_prep();
1197
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001198 /*
Brice Goglin3140a222009-01-06 14:38:57 -08001199 * Store a chunk of page_to_node array in a page,
1200 * but keep the last one as a marker
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001201 */
Brice Goglin3140a222009-01-06 14:38:57 -08001202 chunk_nr_pages = (PAGE_SIZE / sizeof(struct page_to_node)) - 1;
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001203
Brice Goglin3140a222009-01-06 14:38:57 -08001204 for (chunk_start = 0;
1205 chunk_start < nr_pages;
1206 chunk_start += chunk_nr_pages) {
1207 int j;
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001208
Brice Goglin3140a222009-01-06 14:38:57 -08001209 if (chunk_start + chunk_nr_pages > nr_pages)
1210 chunk_nr_pages = nr_pages - chunk_start;
1211
1212 /* fill the chunk pm with addrs and nodes from user-space */
1213 for (j = 0; j < chunk_nr_pages; j++) {
1214 const void __user *p;
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001215 int node;
1216
Brice Goglin3140a222009-01-06 14:38:57 -08001217 err = -EFAULT;
1218 if (get_user(p, pages + j + chunk_start))
1219 goto out_pm;
1220 pm[j].addr = (unsigned long) p;
1221
1222 if (get_user(node, nodes + j + chunk_start))
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001223 goto out_pm;
1224
1225 err = -ENODEV;
Linus Torvalds6f5a55f2010-02-05 16:16:50 -08001226 if (node < 0 || node >= MAX_NUMNODES)
1227 goto out_pm;
1228
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001229 if (!node_state(node, N_HIGH_MEMORY))
1230 goto out_pm;
1231
1232 err = -EACCES;
1233 if (!node_isset(node, task_nodes))
1234 goto out_pm;
1235
Brice Goglin3140a222009-01-06 14:38:57 -08001236 pm[j].node = node;
1237 }
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001238
Brice Goglin3140a222009-01-06 14:38:57 -08001239 /* End marker for this chunk */
1240 pm[chunk_nr_pages].node = MAX_NUMNODES;
1241
1242 /* Migrate this chunk */
1243 err = do_move_page_to_node_array(mm, pm,
1244 flags & MPOL_MF_MOVE_ALL);
1245 if (err < 0)
1246 goto out_pm;
1247
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001248 /* Return status information */
Brice Goglin3140a222009-01-06 14:38:57 -08001249 for (j = 0; j < chunk_nr_pages; j++)
1250 if (put_user(pm[j].status, status + j + chunk_start)) {
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001251 err = -EFAULT;
Brice Goglin3140a222009-01-06 14:38:57 -08001252 goto out_pm;
1253 }
1254 }
1255 err = 0;
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001256
1257out_pm:
Brice Goglin3140a222009-01-06 14:38:57 -08001258 free_page((unsigned long)pm);
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001259out:
1260 return err;
1261}
1262
1263/*
Brice Goglin2f007e72008-10-18 20:27:16 -07001264 * Determine the nodes of an array of pages and store it in an array of status.
Christoph Lameter742755a2006-06-23 02:03:55 -07001265 */
Brice Goglin80bba122008-12-09 13:14:23 -08001266static void do_pages_stat_array(struct mm_struct *mm, unsigned long nr_pages,
1267 const void __user **pages, int *status)
Christoph Lameter742755a2006-06-23 02:03:55 -07001268{
Brice Goglin2f007e72008-10-18 20:27:16 -07001269 unsigned long i;
Brice Goglin2f007e72008-10-18 20:27:16 -07001270
Christoph Lameter742755a2006-06-23 02:03:55 -07001271 down_read(&mm->mmap_sem);
1272
Brice Goglin2f007e72008-10-18 20:27:16 -07001273 for (i = 0; i < nr_pages; i++) {
Brice Goglin80bba122008-12-09 13:14:23 -08001274 unsigned long addr = (unsigned long)(*pages);
Christoph Lameter742755a2006-06-23 02:03:55 -07001275 struct vm_area_struct *vma;
1276 struct page *page;
KOSAKI Motohiroc095adb2008-12-16 16:06:43 +09001277 int err = -EFAULT;
Brice Goglin2f007e72008-10-18 20:27:16 -07001278
1279 vma = find_vma(mm, addr);
Gleb Natapov70384dc2010-10-26 14:22:07 -07001280 if (!vma || addr < vma->vm_start)
Christoph Lameter742755a2006-06-23 02:03:55 -07001281 goto set_status;
1282
Brice Goglin2f007e72008-10-18 20:27:16 -07001283 page = follow_page(vma, addr, 0);
Linus Torvalds89f5b7d2008-06-20 11:18:25 -07001284
1285 err = PTR_ERR(page);
1286 if (IS_ERR(page))
1287 goto set_status;
1288
Christoph Lameter742755a2006-06-23 02:03:55 -07001289 err = -ENOENT;
1290 /* Use PageReserved to check for zero page */
Hugh Dickins62b61f62009-12-14 17:59:33 -08001291 if (!page || PageReserved(page) || PageKsm(page))
Christoph Lameter742755a2006-06-23 02:03:55 -07001292 goto set_status;
1293
1294 err = page_to_nid(page);
1295set_status:
Brice Goglin80bba122008-12-09 13:14:23 -08001296 *status = err;
1297
1298 pages++;
1299 status++;
1300 }
1301
1302 up_read(&mm->mmap_sem);
1303}
1304
1305/*
1306 * Determine the nodes of a user array of pages and store it in
1307 * a user array of status.
1308 */
1309static int do_pages_stat(struct mm_struct *mm, unsigned long nr_pages,
1310 const void __user * __user *pages,
1311 int __user *status)
1312{
1313#define DO_PAGES_STAT_CHUNK_NR 16
1314 const void __user *chunk_pages[DO_PAGES_STAT_CHUNK_NR];
1315 int chunk_status[DO_PAGES_STAT_CHUNK_NR];
Brice Goglin80bba122008-12-09 13:14:23 -08001316
H. Peter Anvin87b8d1a2010-02-18 16:13:40 -08001317 while (nr_pages) {
1318 unsigned long chunk_nr;
Brice Goglin80bba122008-12-09 13:14:23 -08001319
H. Peter Anvin87b8d1a2010-02-18 16:13:40 -08001320 chunk_nr = nr_pages;
1321 if (chunk_nr > DO_PAGES_STAT_CHUNK_NR)
1322 chunk_nr = DO_PAGES_STAT_CHUNK_NR;
1323
1324 if (copy_from_user(chunk_pages, pages, chunk_nr * sizeof(*chunk_pages)))
1325 break;
Brice Goglin80bba122008-12-09 13:14:23 -08001326
1327 do_pages_stat_array(mm, chunk_nr, chunk_pages, chunk_status);
1328
H. Peter Anvin87b8d1a2010-02-18 16:13:40 -08001329 if (copy_to_user(status, chunk_status, chunk_nr * sizeof(*status)))
1330 break;
Christoph Lameter742755a2006-06-23 02:03:55 -07001331
H. Peter Anvin87b8d1a2010-02-18 16:13:40 -08001332 pages += chunk_nr;
1333 status += chunk_nr;
1334 nr_pages -= chunk_nr;
1335 }
1336 return nr_pages ? -EFAULT : 0;
Christoph Lameter742755a2006-06-23 02:03:55 -07001337}
1338
1339/*
1340 * Move a list of pages in the address space of the currently executing
1341 * process.
1342 */
Heiko Carstens938bb9f2009-01-14 14:14:30 +01001343SYSCALL_DEFINE6(move_pages, pid_t, pid, unsigned long, nr_pages,
1344 const void __user * __user *, pages,
1345 const int __user *, nodes,
1346 int __user *, status, int, flags)
Christoph Lameter742755a2006-06-23 02:03:55 -07001347{
David Howellsc69e8d92008-11-14 10:39:19 +11001348 const struct cred *cred = current_cred(), *tcred;
Christoph Lameter742755a2006-06-23 02:03:55 -07001349 struct task_struct *task;
Christoph Lameter742755a2006-06-23 02:03:55 -07001350 struct mm_struct *mm;
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001351 int err;
Christoph Lameter3268c632012-03-21 16:34:06 -07001352 nodemask_t task_nodes;
Christoph Lameter742755a2006-06-23 02:03:55 -07001353
1354 /* Check flags */
1355 if (flags & ~(MPOL_MF_MOVE|MPOL_MF_MOVE_ALL))
1356 return -EINVAL;
1357
1358 if ((flags & MPOL_MF_MOVE_ALL) && !capable(CAP_SYS_NICE))
1359 return -EPERM;
1360
1361 /* Find the mm_struct */
Greg Thelena879bf52011-02-25 14:44:13 -08001362 rcu_read_lock();
Pavel Emelyanov228ebcb2007-10-18 23:40:16 -07001363 task = pid ? find_task_by_vpid(pid) : current;
Christoph Lameter742755a2006-06-23 02:03:55 -07001364 if (!task) {
Greg Thelena879bf52011-02-25 14:44:13 -08001365 rcu_read_unlock();
Christoph Lameter742755a2006-06-23 02:03:55 -07001366 return -ESRCH;
1367 }
Christoph Lameter3268c632012-03-21 16:34:06 -07001368 get_task_struct(task);
Christoph Lameter742755a2006-06-23 02:03:55 -07001369
1370 /*
1371 * Check if this process has the right to modify the specified
1372 * process. The right exists if the process has administrative
1373 * capabilities, superuser privileges or the same
1374 * userid as the target process.
1375 */
David Howellsc69e8d92008-11-14 10:39:19 +11001376 tcred = __task_cred(task);
Eric W. Biedermanb38a86e2012-03-12 15:48:24 -07001377 if (!uid_eq(cred->euid, tcred->suid) && !uid_eq(cred->euid, tcred->uid) &&
1378 !uid_eq(cred->uid, tcred->suid) && !uid_eq(cred->uid, tcred->uid) &&
Christoph Lameter742755a2006-06-23 02:03:55 -07001379 !capable(CAP_SYS_NICE)) {
David Howellsc69e8d92008-11-14 10:39:19 +11001380 rcu_read_unlock();
Christoph Lameter742755a2006-06-23 02:03:55 -07001381 err = -EPERM;
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001382 goto out;
Christoph Lameter742755a2006-06-23 02:03:55 -07001383 }
David Howellsc69e8d92008-11-14 10:39:19 +11001384 rcu_read_unlock();
Christoph Lameter742755a2006-06-23 02:03:55 -07001385
David Quigley86c3a762006-06-23 02:04:02 -07001386 err = security_task_movememory(task);
1387 if (err)
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001388 goto out;
David Quigley86c3a762006-06-23 02:04:02 -07001389
Christoph Lameter3268c632012-03-21 16:34:06 -07001390 task_nodes = cpuset_mems_allowed(task);
1391 mm = get_task_mm(task);
1392 put_task_struct(task);
1393
Sasha Levin6e8b09e2012-04-25 16:01:53 -07001394 if (!mm)
1395 return -EINVAL;
1396
1397 if (nodes)
1398 err = do_pages_move(mm, task_nodes, nr_pages, pages,
1399 nodes, status, flags);
1400 else
1401 err = do_pages_stat(mm, nr_pages, pages, status);
Christoph Lameter3268c632012-03-21 16:34:06 -07001402
1403 mmput(mm);
1404 return err;
David Quigley86c3a762006-06-23 02:04:02 -07001405
Christoph Lameter742755a2006-06-23 02:03:55 -07001406out:
Christoph Lameter3268c632012-03-21 16:34:06 -07001407 put_task_struct(task);
Christoph Lameter742755a2006-06-23 02:03:55 -07001408 return err;
1409}
Christoph Lameter742755a2006-06-23 02:03:55 -07001410
Christoph Lameter7b2259b2006-06-25 05:46:48 -07001411/*
1412 * Call migration functions in the vma_ops that may prepare
1413 * memory in a vm for migration. migration functions may perform
1414 * the migration for vmas that do not have an underlying page struct.
1415 */
1416int migrate_vmas(struct mm_struct *mm, const nodemask_t *to,
1417 const nodemask_t *from, unsigned long flags)
1418{
1419 struct vm_area_struct *vma;
1420 int err = 0;
1421
Daisuke Nishimura1001c9f2009-02-11 13:04:18 -08001422 for (vma = mm->mmap; vma && !err; vma = vma->vm_next) {
Christoph Lameter7b2259b2006-06-25 05:46:48 -07001423 if (vma->vm_ops && vma->vm_ops->migrate) {
1424 err = vma->vm_ops->migrate(vma, to, from, flags);
1425 if (err)
1426 break;
1427 }
1428 }
1429 return err;
1430}
Gerald Schaefer83d16742008-07-23 21:28:22 -07001431#endif