blob: 7dbcb22316d2b2e482de278a8023c6429fe7bfaf [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>
16#include <linux/module.h>
17#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>
24#include <linux/rmap.h>
25#include <linux/topology.h>
26#include <linux/cpu.h>
27#include <linux/cpuset.h>
Christoph Lameter04e62a22006-06-23 02:03:38 -070028#include <linux/writeback.h>
Christoph Lameter742755a2006-06-23 02:03:55 -070029#include <linux/mempolicy.h>
30#include <linux/vmalloc.h>
David Quigley86c3a762006-06-23 02:04:02 -070031#include <linux/security.h>
Balbir Singh8a9f3cc2008-02-07 00:13:53 -080032#include <linux/memcontrol.h>
Adrian Bunk4f5ca262008-07-23 21:27:02 -070033#include <linux/syscalls.h>
Christoph Lameterb20a3502006-03-22 00:09:12 -080034
35#include "internal.h"
36
Christoph Lameterb20a3502006-03-22 00:09:12 -080037#define lru_to_page(_head) (list_entry((_head)->prev, struct page, lru))
38
39/*
Christoph Lameter742755a2006-06-23 02:03:55 -070040 * migrate_prep() needs to be called before we start compiling a list of pages
41 * to be migrated using isolate_lru_page().
Christoph Lameterb20a3502006-03-22 00:09:12 -080042 */
43int migrate_prep(void)
44{
Christoph Lameterb20a3502006-03-22 00:09:12 -080045 /*
46 * Clear the LRU lists so pages can be isolated.
47 * Note that pages may be moved off the LRU after we have
48 * drained them. Those pages will fail to migrate like other
49 * pages that may be busy.
50 */
51 lru_add_drain_all();
52
53 return 0;
54}
55
Christoph Lameterb20a3502006-03-22 00:09:12 -080056/*
Lee Schermerhorn894bc312008-10-18 20:26:39 -070057 * Add isolated pages on the list back to the LRU under page lock
58 * to avoid leaking evictable pages back onto unevictable list.
Christoph Lameterb20a3502006-03-22 00:09:12 -080059 *
60 * returns the number of pages put back.
61 */
62int putback_lru_pages(struct list_head *l)
63{
64 struct page *page;
65 struct page *page2;
66 int count = 0;
67
68 list_for_each_entry_safe(page, page2, l, lru) {
Christoph Lametere24f0b82006-06-23 02:03:51 -070069 list_del(&page->lru);
KOSAKI Motohiroa7312862009-09-21 17:01:37 -070070 dec_zone_page_state(page, NR_ISOLATED_ANON +
Johannes Weiner6c0b1352009-09-21 17:02:59 -070071 page_is_file_cache(page));
Lee Schermerhorn894bc312008-10-18 20:26:39 -070072 putback_lru_page(page);
Christoph Lameterb20a3502006-03-22 00:09:12 -080073 count++;
74 }
75 return count;
76}
77
Christoph Lameter06972122006-06-23 02:03:35 -070078/*
79 * Restore a potential migration pte to a working pte entry
80 */
Christoph Lameter04e62a22006-06-23 02:03:38 -070081static void remove_migration_pte(struct vm_area_struct *vma,
Christoph Lameter06972122006-06-23 02:03:35 -070082 struct page *old, struct page *new)
83{
84 struct mm_struct *mm = vma->vm_mm;
85 swp_entry_t entry;
86 pgd_t *pgd;
87 pud_t *pud;
88 pmd_t *pmd;
89 pte_t *ptep, pte;
90 spinlock_t *ptl;
Christoph Lameter04e62a22006-06-23 02:03:38 -070091 unsigned long addr = page_address_in_vma(new, vma);
92
93 if (addr == -EFAULT)
94 return;
Christoph Lameter06972122006-06-23 02:03:35 -070095
96 pgd = pgd_offset(mm, addr);
97 if (!pgd_present(*pgd))
98 return;
99
100 pud = pud_offset(pgd, addr);
101 if (!pud_present(*pud))
102 return;
103
104 pmd = pmd_offset(pud, addr);
105 if (!pmd_present(*pmd))
106 return;
107
108 ptep = pte_offset_map(pmd, addr);
109
110 if (!is_swap_pte(*ptep)) {
111 pte_unmap(ptep);
112 return;
113 }
114
115 ptl = pte_lockptr(mm, pmd);
116 spin_lock(ptl);
117 pte = *ptep;
118 if (!is_swap_pte(pte))
119 goto out;
120
121 entry = pte_to_swp_entry(pte);
122
123 if (!is_migration_entry(entry) || migration_entry_to_page(entry) != old)
124 goto out;
125
Christoph Lameter06972122006-06-23 02:03:35 -0700126 get_page(new);
127 pte = pte_mkold(mk_pte(new, vma->vm_page_prot));
128 if (is_write_migration_entry(entry))
129 pte = pte_mkwrite(pte);
KAMEZAWA Hiroyuki97ee0522007-10-16 01:25:43 -0700130 flush_cache_page(vma, addr, pte_pfn(pte));
Christoph Lameter06972122006-06-23 02:03:35 -0700131 set_pte_at(mm, addr, ptep, pte);
Christoph Lameter04e62a22006-06-23 02:03:38 -0700132
133 if (PageAnon(new))
134 page_add_anon_rmap(new, vma, addr);
135 else
136 page_add_file_rmap(new);
137
138 /* No need to invalidate - it was non-present before */
139 update_mmu_cache(vma, addr, pte);
Christoph Lameter04e62a22006-06-23 02:03:38 -0700140
Christoph Lameter06972122006-06-23 02:03:35 -0700141out:
142 pte_unmap_unlock(ptep, ptl);
143}
144
145/*
Christoph Lameter04e62a22006-06-23 02:03:38 -0700146 * Note that remove_file_migration_ptes will only work on regular mappings,
147 * Nonlinear mappings do not use migration entries.
148 */
149static void remove_file_migration_ptes(struct page *old, struct page *new)
150{
151 struct vm_area_struct *vma;
Shaohua Liabfc3482009-09-21 17:01:19 -0700152 struct address_space *mapping = new->mapping;
Christoph Lameter04e62a22006-06-23 02:03:38 -0700153 struct prio_tree_iter iter;
154 pgoff_t pgoff = new->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT);
155
156 if (!mapping)
157 return;
158
159 spin_lock(&mapping->i_mmap_lock);
160
161 vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, pgoff, pgoff)
162 remove_migration_pte(vma, old, new);
163
164 spin_unlock(&mapping->i_mmap_lock);
165}
166
167/*
Christoph Lameter06972122006-06-23 02:03:35 -0700168 * Must hold mmap_sem lock on at least one of the vmas containing
169 * the page so that the anon_vma cannot vanish.
170 */
Christoph Lameter04e62a22006-06-23 02:03:38 -0700171static void remove_anon_migration_ptes(struct page *old, struct page *new)
Christoph Lameter06972122006-06-23 02:03:35 -0700172{
173 struct anon_vma *anon_vma;
174 struct vm_area_struct *vma;
175 unsigned long mapping;
176
177 mapping = (unsigned long)new->mapping;
178
179 if (!mapping || (mapping & PAGE_MAPPING_ANON) == 0)
180 return;
181
182 /*
183 * We hold the mmap_sem lock. So no need to call page_lock_anon_vma.
184 */
185 anon_vma = (struct anon_vma *) (mapping - PAGE_MAPPING_ANON);
186 spin_lock(&anon_vma->lock);
187
188 list_for_each_entry(vma, &anon_vma->head, anon_vma_node)
Christoph Lameter04e62a22006-06-23 02:03:38 -0700189 remove_migration_pte(vma, old, new);
Christoph Lameter06972122006-06-23 02:03:35 -0700190
191 spin_unlock(&anon_vma->lock);
192}
193
194/*
Christoph Lameter04e62a22006-06-23 02:03:38 -0700195 * Get rid of all migration entries and replace them by
196 * references to the indicated page.
197 */
198static void remove_migration_ptes(struct page *old, struct page *new)
199{
200 if (PageAnon(new))
201 remove_anon_migration_ptes(old, new);
202 else
203 remove_file_migration_ptes(old, new);
204}
205
206/*
Christoph Lameter06972122006-06-23 02:03:35 -0700207 * Something used the pte of a page under migration. We need to
208 * get to the page and wait until migration is finished.
209 * When we return from this function the fault will be retried.
210 *
211 * This function is called from do_swap_page().
212 */
213void migration_entry_wait(struct mm_struct *mm, pmd_t *pmd,
214 unsigned long address)
215{
216 pte_t *ptep, pte;
217 spinlock_t *ptl;
218 swp_entry_t entry;
219 struct page *page;
220
221 ptep = pte_offset_map_lock(mm, pmd, address, &ptl);
222 pte = *ptep;
223 if (!is_swap_pte(pte))
224 goto out;
225
226 entry = pte_to_swp_entry(pte);
227 if (!is_migration_entry(entry))
228 goto out;
229
230 page = migration_entry_to_page(entry);
231
Nick Piggine2867812008-07-25 19:45:30 -0700232 /*
233 * Once radix-tree replacement of page migration started, page_count
234 * *must* be zero. And, we don't want to call wait_on_page_locked()
235 * against a page without get_page().
236 * So, we use get_page_unless_zero(), here. Even failed, page fault
237 * will occur again.
238 */
239 if (!get_page_unless_zero(page))
240 goto out;
Christoph Lameter06972122006-06-23 02:03:35 -0700241 pte_unmap_unlock(ptep, ptl);
242 wait_on_page_locked(page);
243 put_page(page);
244 return;
245out:
246 pte_unmap_unlock(ptep, ptl);
247}
248
Christoph Lameterb20a3502006-03-22 00:09:12 -0800249/*
Christoph Lameterc3fcf8a2006-06-23 02:03:32 -0700250 * Replace the page in the mapping.
Christoph Lameter5b5c7122006-06-23 02:03:29 -0700251 *
252 * The number of remaining references must be:
253 * 1 for anonymous pages without a mapping
254 * 2 for pages with a mapping
David Howells266cf652009-04-03 16:42:36 +0100255 * 3 for pages with a mapping and PagePrivate/PagePrivate2 set.
Christoph Lameterb20a3502006-03-22 00:09:12 -0800256 */
Christoph Lameter2d1db3b2006-06-23 02:03:33 -0700257static int migrate_page_move_mapping(struct address_space *mapping,
258 struct page *newpage, struct page *page)
Christoph Lameterb20a3502006-03-22 00:09:12 -0800259{
Nick Piggine2867812008-07-25 19:45:30 -0700260 int expected_count;
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800261 void **pslot;
Christoph Lameterb20a3502006-03-22 00:09:12 -0800262
Christoph Lameter6c5240a2006-06-23 02:03:37 -0700263 if (!mapping) {
Christoph Lameter0e8c7d02007-04-23 14:41:09 -0700264 /* Anonymous page without mapping */
Christoph Lameter6c5240a2006-06-23 02:03:37 -0700265 if (page_count(page) != 1)
266 return -EAGAIN;
267 return 0;
268 }
269
Nick Piggin19fd6232008-07-25 19:45:32 -0700270 spin_lock_irq(&mapping->tree_lock);
Christoph Lameterb20a3502006-03-22 00:09:12 -0800271
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800272 pslot = radix_tree_lookup_slot(&mapping->page_tree,
273 page_index(page));
Christoph Lameterb20a3502006-03-22 00:09:12 -0800274
Johannes Weineredcf4742009-09-21 17:02:59 -0700275 expected_count = 2 + page_has_private(page);
Nick Piggine2867812008-07-25 19:45:30 -0700276 if (page_count(page) != expected_count ||
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800277 (struct page *)radix_tree_deref_slot(pslot) != page) {
Nick Piggin19fd6232008-07-25 19:45:32 -0700278 spin_unlock_irq(&mapping->tree_lock);
Christoph Lametere23ca002006-04-10 22:52:57 -0700279 return -EAGAIN;
Christoph Lameterb20a3502006-03-22 00:09:12 -0800280 }
281
Nick Piggine2867812008-07-25 19:45:30 -0700282 if (!page_freeze_refs(page, expected_count)) {
Nick Piggin19fd6232008-07-25 19:45:32 -0700283 spin_unlock_irq(&mapping->tree_lock);
Nick Piggine2867812008-07-25 19:45:30 -0700284 return -EAGAIN;
285 }
286
Christoph Lameterb20a3502006-03-22 00:09:12 -0800287 /*
288 * Now we know that no one else is looking at the page.
Christoph Lameterb20a3502006-03-22 00:09:12 -0800289 */
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800290 get_page(newpage); /* add cache reference */
Christoph Lameterb20a3502006-03-22 00:09:12 -0800291 if (PageSwapCache(page)) {
292 SetPageSwapCache(newpage);
293 set_page_private(newpage, page_private(page));
294 }
295
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800296 radix_tree_replace_slot(pslot, newpage);
297
Nick Piggine2867812008-07-25 19:45:30 -0700298 page_unfreeze_refs(page, expected_count);
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800299 /*
300 * Drop cache reference from old page.
301 * We know this isn't the last reference.
302 */
Christoph Lameterb20a3502006-03-22 00:09:12 -0800303 __put_page(page);
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800304
Christoph Lameter0e8c7d02007-04-23 14:41:09 -0700305 /*
306 * If moved to a different zone then also account
307 * the page for that zone. Other VM counters will be
308 * taken care of when we establish references to the
309 * new page and drop references to the old page.
310 *
311 * Note that anonymous pages are accounted for
312 * via NR_FILE_PAGES and NR_ANON_PAGES if they
313 * are mapped to swap space.
314 */
315 __dec_zone_page_state(page, NR_FILE_PAGES);
316 __inc_zone_page_state(newpage, NR_FILE_PAGES);
KOSAKI Motohiro4b021082009-09-21 17:01:33 -0700317 if (PageSwapBacked(page)) {
318 __dec_zone_page_state(page, NR_SHMEM);
319 __inc_zone_page_state(newpage, NR_SHMEM);
320 }
Nick Piggin19fd6232008-07-25 19:45:32 -0700321 spin_unlock_irq(&mapping->tree_lock);
Christoph Lameterb20a3502006-03-22 00:09:12 -0800322
323 return 0;
324}
Christoph Lameterb20a3502006-03-22 00:09:12 -0800325
326/*
327 * Copy the page to its new location
328 */
Christoph Lametere7340f72006-06-23 02:03:29 -0700329static void migrate_page_copy(struct page *newpage, struct page *page)
Christoph Lameterb20a3502006-03-22 00:09:12 -0800330{
KAMEZAWA Hiroyukib7abea92008-10-18 20:28:09 -0700331 int anon;
332
Christoph Lameterb20a3502006-03-22 00:09:12 -0800333 copy_highpage(newpage, page);
334
335 if (PageError(page))
336 SetPageError(newpage);
337 if (PageReferenced(page))
338 SetPageReferenced(newpage);
339 if (PageUptodate(page))
340 SetPageUptodate(newpage);
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700341 if (TestClearPageActive(page)) {
342 VM_BUG_ON(PageUnevictable(page));
Christoph Lameterb20a3502006-03-22 00:09:12 -0800343 SetPageActive(newpage);
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700344 } else
345 unevictable_migrate_page(newpage, page);
Christoph Lameterb20a3502006-03-22 00:09:12 -0800346 if (PageChecked(page))
347 SetPageChecked(newpage);
348 if (PageMappedToDisk(page))
349 SetPageMappedToDisk(newpage);
350
351 if (PageDirty(page)) {
352 clear_page_dirty_for_io(page);
Nick Piggin3a902c52008-04-30 00:55:16 -0700353 /*
354 * Want to mark the page and the radix tree as dirty, and
355 * redo the accounting that clear_page_dirty_for_io undid,
356 * but we can't use set_page_dirty because that function
357 * is actually a signal that all of the page has become dirty.
358 * Wheras only part of our page may be dirty.
359 */
360 __set_page_dirty_nobuffers(newpage);
Christoph Lameterb20a3502006-03-22 00:09:12 -0800361 }
362
Nick Pigginb291f002008-10-18 20:26:44 -0700363 mlock_migrate_page(newpage, page);
364
Christoph Lameterb20a3502006-03-22 00:09:12 -0800365 ClearPageSwapCache(page);
Christoph Lameterb20a3502006-03-22 00:09:12 -0800366 ClearPagePrivate(page);
367 set_page_private(page, 0);
KAMEZAWA Hiroyukib7abea92008-10-18 20:28:09 -0700368 /* page->mapping contains a flag for PageAnon() */
369 anon = PageAnon(page);
Christoph Lameterb20a3502006-03-22 00:09:12 -0800370 page->mapping = NULL;
371
372 /*
373 * If any waiters have accumulated on the new page then
374 * wake them up.
375 */
376 if (PageWriteback(newpage))
377 end_page_writeback(newpage);
378}
Christoph Lameterb20a3502006-03-22 00:09:12 -0800379
Christoph Lameter1d8b85c2006-06-23 02:03:28 -0700380/************************************************************
381 * Migration functions
382 ***********************************************************/
383
384/* Always fail migration. Used for mappings that are not movable */
Christoph Lameter2d1db3b2006-06-23 02:03:33 -0700385int fail_migrate_page(struct address_space *mapping,
386 struct page *newpage, struct page *page)
Christoph Lameter1d8b85c2006-06-23 02:03:28 -0700387{
388 return -EIO;
389}
390EXPORT_SYMBOL(fail_migrate_page);
391
Christoph Lameterb20a3502006-03-22 00:09:12 -0800392/*
393 * Common logic to directly migrate a single page suitable for
David Howells266cf652009-04-03 16:42:36 +0100394 * pages that do not use PagePrivate/PagePrivate2.
Christoph Lameterb20a3502006-03-22 00:09:12 -0800395 *
396 * Pages are locked upon entry and exit.
397 */
Christoph Lameter2d1db3b2006-06-23 02:03:33 -0700398int migrate_page(struct address_space *mapping,
399 struct page *newpage, struct page *page)
Christoph Lameterb20a3502006-03-22 00:09:12 -0800400{
401 int rc;
402
403 BUG_ON(PageWriteback(page)); /* Writeback must be complete */
404
Christoph Lameter2d1db3b2006-06-23 02:03:33 -0700405 rc = migrate_page_move_mapping(mapping, newpage, page);
Christoph Lameterb20a3502006-03-22 00:09:12 -0800406
407 if (rc)
408 return rc;
409
410 migrate_page_copy(newpage, page);
Christoph Lameterb20a3502006-03-22 00:09:12 -0800411 return 0;
412}
413EXPORT_SYMBOL(migrate_page);
414
David Howells93614012006-09-30 20:45:40 +0200415#ifdef CONFIG_BLOCK
Christoph Lameterb20a3502006-03-22 00:09:12 -0800416/*
Christoph Lameter1d8b85c2006-06-23 02:03:28 -0700417 * Migration function for pages with buffers. This function can only be used
418 * if the underlying filesystem guarantees that no other references to "page"
419 * exist.
420 */
Christoph Lameter2d1db3b2006-06-23 02:03:33 -0700421int buffer_migrate_page(struct address_space *mapping,
422 struct page *newpage, struct page *page)
Christoph Lameter1d8b85c2006-06-23 02:03:28 -0700423{
Christoph Lameter1d8b85c2006-06-23 02:03:28 -0700424 struct buffer_head *bh, *head;
425 int rc;
426
Christoph Lameter1d8b85c2006-06-23 02:03:28 -0700427 if (!page_has_buffers(page))
Christoph Lameter2d1db3b2006-06-23 02:03:33 -0700428 return migrate_page(mapping, newpage, page);
Christoph Lameter1d8b85c2006-06-23 02:03:28 -0700429
430 head = page_buffers(page);
431
Christoph Lameter2d1db3b2006-06-23 02:03:33 -0700432 rc = migrate_page_move_mapping(mapping, newpage, page);
Christoph Lameter1d8b85c2006-06-23 02:03:28 -0700433
434 if (rc)
435 return rc;
436
437 bh = head;
438 do {
439 get_bh(bh);
440 lock_buffer(bh);
441 bh = bh->b_this_page;
442
443 } while (bh != head);
444
445 ClearPagePrivate(page);
446 set_page_private(newpage, page_private(page));
447 set_page_private(page, 0);
448 put_page(page);
449 get_page(newpage);
450
451 bh = head;
452 do {
453 set_bh_page(bh, newpage, bh_offset(bh));
454 bh = bh->b_this_page;
455
456 } while (bh != head);
457
458 SetPagePrivate(newpage);
459
460 migrate_page_copy(newpage, page);
461
462 bh = head;
463 do {
464 unlock_buffer(bh);
465 put_bh(bh);
466 bh = bh->b_this_page;
467
468 } while (bh != head);
469
470 return 0;
471}
472EXPORT_SYMBOL(buffer_migrate_page);
David Howells93614012006-09-30 20:45:40 +0200473#endif
Christoph Lameter1d8b85c2006-06-23 02:03:28 -0700474
Christoph Lameter04e62a22006-06-23 02:03:38 -0700475/*
476 * Writeback a page to clean the dirty state
477 */
478static int writeout(struct address_space *mapping, struct page *page)
479{
480 struct writeback_control wbc = {
481 .sync_mode = WB_SYNC_NONE,
482 .nr_to_write = 1,
483 .range_start = 0,
484 .range_end = LLONG_MAX,
485 .nonblocking = 1,
486 .for_reclaim = 1
487 };
488 int rc;
489
490 if (!mapping->a_ops->writepage)
491 /* No write method for the address space */
492 return -EINVAL;
493
494 if (!clear_page_dirty_for_io(page))
495 /* Someone else already triggered a write */
496 return -EAGAIN;
497
498 /*
499 * A dirty page may imply that the underlying filesystem has
500 * the page on some queue. So the page must be clean for
501 * migration. Writeout may mean we loose the lock and the
502 * page state is no longer what we checked for earlier.
503 * At this point we know that the migration attempt cannot
504 * be successful.
505 */
506 remove_migration_ptes(page, page);
507
508 rc = mapping->a_ops->writepage(page, &wbc);
Christoph Lameter04e62a22006-06-23 02:03:38 -0700509
510 if (rc != AOP_WRITEPAGE_ACTIVATE)
511 /* unlocked. Relock */
512 lock_page(page);
513
Hugh Dickinsbda85502008-11-19 15:36:36 -0800514 return (rc < 0) ? -EIO : -EAGAIN;
Christoph Lameter04e62a22006-06-23 02:03:38 -0700515}
516
517/*
518 * Default handling if a filesystem does not provide a migration function.
519 */
Christoph Lameter8351a6e2006-06-23 02:03:33 -0700520static int fallback_migrate_page(struct address_space *mapping,
521 struct page *newpage, struct page *page)
522{
Christoph Lameter04e62a22006-06-23 02:03:38 -0700523 if (PageDirty(page))
524 return writeout(mapping, page);
Christoph Lameter8351a6e2006-06-23 02:03:33 -0700525
526 /*
527 * Buffers may be managed in a filesystem specific way.
528 * We must have no buffers or drop them.
529 */
David Howells266cf652009-04-03 16:42:36 +0100530 if (page_has_private(page) &&
Christoph Lameter8351a6e2006-06-23 02:03:33 -0700531 !try_to_release_page(page, GFP_KERNEL))
532 return -EAGAIN;
533
534 return migrate_page(mapping, newpage, page);
535}
536
Christoph Lameter1d8b85c2006-06-23 02:03:28 -0700537/*
Christoph Lametere24f0b82006-06-23 02:03:51 -0700538 * Move a page to a newly allocated page
539 * The page is locked and all ptes have been successfully removed.
540 *
541 * The new page will have replaced the old page if this function
542 * is successful.
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700543 *
544 * Return value:
545 * < 0 - error code
546 * == 0 - success
Christoph Lametere24f0b82006-06-23 02:03:51 -0700547 */
548static int move_to_new_page(struct page *newpage, struct page *page)
549{
550 struct address_space *mapping;
551 int rc;
552
553 /*
554 * Block others from accessing the page when we get around to
555 * establishing additional references. We are the only one
556 * holding a reference to the new page at this point.
557 */
Nick Piggin529ae9a2008-08-02 12:01:03 +0200558 if (!trylock_page(newpage))
Christoph Lametere24f0b82006-06-23 02:03:51 -0700559 BUG();
560
561 /* Prepare mapping for the new page.*/
562 newpage->index = page->index;
563 newpage->mapping = page->mapping;
Rik van Rielb2e18532008-10-18 20:26:30 -0700564 if (PageSwapBacked(page))
565 SetPageSwapBacked(newpage);
Christoph Lametere24f0b82006-06-23 02:03:51 -0700566
567 mapping = page_mapping(page);
568 if (!mapping)
569 rc = migrate_page(mapping, newpage, page);
570 else if (mapping->a_ops->migratepage)
571 /*
572 * Most pages have a mapping and most filesystems
573 * should provide a migration function. Anonymous
574 * pages are part of swap space which also has its
575 * own migration function. This is the most common
576 * path for page migration.
577 */
578 rc = mapping->a_ops->migratepage(mapping,
579 newpage, page);
580 else
581 rc = fallback_migrate_page(mapping, newpage, page);
582
KAMEZAWA Hiroyukiae41be32008-02-07 00:14:10 -0800583 if (!rc) {
Christoph Lametere24f0b82006-06-23 02:03:51 -0700584 remove_migration_ptes(page, newpage);
KAMEZAWA Hiroyukiae41be32008-02-07 00:14:10 -0800585 } else
Christoph Lametere24f0b82006-06-23 02:03:51 -0700586 newpage->mapping = NULL;
587
588 unlock_page(newpage);
589
590 return rc;
591}
592
593/*
594 * Obtain the lock on page, remove all ptes and migrate the page
595 * to the newly allocated page in newpage.
596 */
Christoph Lameter95a402c2006-06-23 02:03:53 -0700597static int unmap_and_move(new_page_t get_new_page, unsigned long private,
598 struct page *page, int force)
Christoph Lametere24f0b82006-06-23 02:03:51 -0700599{
600 int rc = 0;
Christoph Lameter742755a2006-06-23 02:03:55 -0700601 int *result = NULL;
602 struct page *newpage = get_new_page(page, private, &result);
KAMEZAWA Hiroyuki989f89c2007-08-30 23:56:21 -0700603 int rcu_locked = 0;
KAMEZAWA Hiroyukiae41be32008-02-07 00:14:10 -0800604 int charge = 0;
KAMEZAWA Hiroyukie00e4312009-11-11 14:26:26 -0800605 struct mem_cgroup *mem = NULL;
Christoph Lameter95a402c2006-06-23 02:03:53 -0700606
607 if (!newpage)
608 return -ENOMEM;
Christoph Lametere24f0b82006-06-23 02:03:51 -0700609
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700610 if (page_count(page) == 1) {
Christoph Lametere24f0b82006-06-23 02:03:51 -0700611 /* page was freed from under us. So we are done. */
Christoph Lameter95a402c2006-06-23 02:03:53 -0700612 goto move_newpage;
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700613 }
Christoph Lametere24f0b82006-06-23 02:03:51 -0700614
KAMEZAWA Hiroyukie8589cc2008-07-25 01:47:10 -0700615 /* prepare cgroup just returns 0 or -ENOMEM */
Christoph Lametere24f0b82006-06-23 02:03:51 -0700616 rc = -EAGAIN;
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -0800617
Nick Piggin529ae9a2008-08-02 12:01:03 +0200618 if (!trylock_page(page)) {
Christoph Lametere24f0b82006-06-23 02:03:51 -0700619 if (!force)
Christoph Lameter95a402c2006-06-23 02:03:53 -0700620 goto move_newpage;
Christoph Lametere24f0b82006-06-23 02:03:51 -0700621 lock_page(page);
622 }
623
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -0800624 /* charge against new page */
625 charge = mem_cgroup_prepare_migration(page, &mem);
626 if (charge == -ENOMEM) {
627 rc = -ENOMEM;
628 goto unlock;
629 }
630 BUG_ON(charge);
631
Christoph Lametere24f0b82006-06-23 02:03:51 -0700632 if (PageWriteback(page)) {
633 if (!force)
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -0800634 goto uncharge;
Christoph Lametere24f0b82006-06-23 02:03:51 -0700635 wait_on_page_writeback(page);
636 }
Christoph Lametere24f0b82006-06-23 02:03:51 -0700637 /*
KAMEZAWA Hiroyukidc386d42007-07-26 10:41:07 -0700638 * By try_to_unmap(), page->mapcount goes down to 0 here. In this case,
639 * we cannot notice that anon_vma is freed while we migrates a page.
640 * This rcu_read_lock() delays freeing anon_vma pointer until the end
641 * of migration. File cache pages are no problem because of page_lock()
KAMEZAWA Hiroyuki989f89c2007-08-30 23:56:21 -0700642 * File Caches may use write_page() or lock_page() in migration, then,
643 * just care Anon page here.
Christoph Lametere24f0b82006-06-23 02:03:51 -0700644 */
KAMEZAWA Hiroyuki989f89c2007-08-30 23:56:21 -0700645 if (PageAnon(page)) {
646 rcu_read_lock();
647 rcu_locked = 1;
648 }
Shaohua Li62e1c552008-02-04 22:29:33 -0800649
KAMEZAWA Hiroyukidc386d42007-07-26 10:41:07 -0700650 /*
Shaohua Li62e1c552008-02-04 22:29:33 -0800651 * Corner case handling:
652 * 1. When a new swap-cache page is read into, it is added to the LRU
653 * and treated as swapcache but it has no rmap yet.
654 * Calling try_to_unmap() against a page->mapping==NULL page will
655 * trigger a BUG. So handle it here.
656 * 2. An orphaned page (see truncate_complete_page) might have
657 * fs-private metadata. The page can be picked up due to memory
658 * offlining. Everywhere else except page reclaim, the page is
659 * invisible to the vm, so the page can not be migrated. So try to
660 * free the metadata, so the page can be freed.
KAMEZAWA Hiroyukidc386d42007-07-26 10:41:07 -0700661 */
Shaohua Li62e1c552008-02-04 22:29:33 -0800662 if (!page->mapping) {
David Howells266cf652009-04-03 16:42:36 +0100663 if (!PageAnon(page) && page_has_private(page)) {
Shaohua Li62e1c552008-02-04 22:29:33 -0800664 /*
665 * Go direct to try_to_free_buffers() here because
666 * a) that's what try_to_release_page() would do anyway
667 * b) we may be under rcu_read_lock() here, so we can't
668 * use GFP_KERNEL which is what try_to_release_page()
669 * needs to be effective.
670 */
671 try_to_free_buffers(page);
Shaohua Liabfc3482009-09-21 17:01:19 -0700672 goto rcu_unlock;
Shaohua Li62e1c552008-02-04 22:29:33 -0800673 }
Shaohua Liabfc3482009-09-21 17:01:19 -0700674 goto skip_unmap;
Shaohua Li62e1c552008-02-04 22:29:33 -0800675 }
676
KAMEZAWA Hiroyukidc386d42007-07-26 10:41:07 -0700677 /* Establish migration ptes or remove ptes */
Andi Kleen14fa31b2009-09-16 11:50:10 +0200678 try_to_unmap(page, TTU_MIGRATION|TTU_IGNORE_MLOCK|TTU_IGNORE_ACCESS);
KAMEZAWA Hiroyukidc386d42007-07-26 10:41:07 -0700679
Shaohua Liabfc3482009-09-21 17:01:19 -0700680skip_unmap:
Christoph Lametere6a15302006-06-25 05:46:49 -0700681 if (!page_mapped(page))
682 rc = move_to_new_page(newpage, page);
Christoph Lametere24f0b82006-06-23 02:03:51 -0700683
KAMEZAWA Hiroyukie8589cc2008-07-25 01:47:10 -0700684 if (rc)
Christoph Lametere24f0b82006-06-23 02:03:51 -0700685 remove_migration_ptes(page, page);
KAMEZAWA Hiroyukidc386d42007-07-26 10:41:07 -0700686rcu_unlock:
KAMEZAWA Hiroyuki989f89c2007-08-30 23:56:21 -0700687 if (rcu_locked)
688 rcu_read_unlock();
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -0800689uncharge:
690 if (!charge)
691 mem_cgroup_end_migration(mem, page, newpage);
Christoph Lametere24f0b82006-06-23 02:03:51 -0700692unlock:
693 unlock_page(page);
Christoph Lameter95a402c2006-06-23 02:03:53 -0700694
Christoph Lametere24f0b82006-06-23 02:03:51 -0700695 if (rc != -EAGAIN) {
Christoph Lameteraaa994b2006-06-23 02:03:52 -0700696 /*
697 * A page that has been migrated has all references
698 * removed and will be freed. A page that has not been
699 * migrated will have kepts its references and be
700 * restored.
701 */
702 list_del(&page->lru);
KOSAKI Motohiroa7312862009-09-21 17:01:37 -0700703 dec_zone_page_state(page, NR_ISOLATED_ANON +
Johannes Weiner6c0b1352009-09-21 17:02:59 -0700704 page_is_file_cache(page));
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700705 putback_lru_page(page);
Christoph Lametere24f0b82006-06-23 02:03:51 -0700706 }
Christoph Lameter95a402c2006-06-23 02:03:53 -0700707
708move_newpage:
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700709
Christoph Lameter95a402c2006-06-23 02:03:53 -0700710 /*
711 * Move the new page to the LRU. If migration was not successful
712 * then this will free the page.
713 */
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700714 putback_lru_page(newpage);
715
Christoph Lameter742755a2006-06-23 02:03:55 -0700716 if (result) {
717 if (rc)
718 *result = rc;
719 else
720 *result = page_to_nid(newpage);
721 }
Christoph Lametere24f0b82006-06-23 02:03:51 -0700722 return rc;
723}
724
725/*
Christoph Lameterb20a3502006-03-22 00:09:12 -0800726 * migrate_pages
727 *
Christoph Lameter95a402c2006-06-23 02:03:53 -0700728 * The function takes one list of pages to migrate and a function
729 * that determines from the page to be migrated and the private data
730 * the target of the move and allocates the page.
Christoph Lameterb20a3502006-03-22 00:09:12 -0800731 *
732 * The function returns after 10 attempts or if no pages
733 * are movable anymore because to has become empty
Christoph Lameteraaa994b2006-06-23 02:03:52 -0700734 * or no retryable pages exist anymore. All pages will be
Gabriel Craciunescue9534b32007-10-20 02:13:26 +0200735 * returned to the LRU or freed.
Christoph Lameterb20a3502006-03-22 00:09:12 -0800736 *
Christoph Lameter95a402c2006-06-23 02:03:53 -0700737 * Return: Number of pages not migrated or error code.
Christoph Lameterb20a3502006-03-22 00:09:12 -0800738 */
Christoph Lameter95a402c2006-06-23 02:03:53 -0700739int migrate_pages(struct list_head *from,
740 new_page_t get_new_page, unsigned long private)
Christoph Lameterb20a3502006-03-22 00:09:12 -0800741{
Christoph Lametere24f0b82006-06-23 02:03:51 -0700742 int retry = 1;
Christoph Lameterb20a3502006-03-22 00:09:12 -0800743 int nr_failed = 0;
744 int pass = 0;
745 struct page *page;
746 struct page *page2;
747 int swapwrite = current->flags & PF_SWAPWRITE;
748 int rc;
KOSAKI Motohiroa7312862009-09-21 17:01:37 -0700749 unsigned long flags;
750
751 local_irq_save(flags);
752 list_for_each_entry(page, from, lru)
753 __inc_zone_page_state(page, NR_ISOLATED_ANON +
Johannes Weiner6c0b1352009-09-21 17:02:59 -0700754 page_is_file_cache(page));
KOSAKI Motohiroa7312862009-09-21 17:01:37 -0700755 local_irq_restore(flags);
Christoph Lameterb20a3502006-03-22 00:09:12 -0800756
757 if (!swapwrite)
758 current->flags |= PF_SWAPWRITE;
759
Christoph Lametere24f0b82006-06-23 02:03:51 -0700760 for(pass = 0; pass < 10 && retry; pass++) {
761 retry = 0;
Christoph Lameterb20a3502006-03-22 00:09:12 -0800762
Christoph Lametere24f0b82006-06-23 02:03:51 -0700763 list_for_each_entry_safe(page, page2, from, lru) {
Christoph Lametere24f0b82006-06-23 02:03:51 -0700764 cond_resched();
Christoph Lameterb20a3502006-03-22 00:09:12 -0800765
Christoph Lameter95a402c2006-06-23 02:03:53 -0700766 rc = unmap_and_move(get_new_page, private,
767 page, pass > 2);
Christoph Lameterb20a3502006-03-22 00:09:12 -0800768
Christoph Lametere24f0b82006-06-23 02:03:51 -0700769 switch(rc) {
Christoph Lameter95a402c2006-06-23 02:03:53 -0700770 case -ENOMEM:
771 goto out;
Christoph Lametere24f0b82006-06-23 02:03:51 -0700772 case -EAGAIN:
Christoph Lameter2d1db3b2006-06-23 02:03:33 -0700773 retry++;
Christoph Lametere24f0b82006-06-23 02:03:51 -0700774 break;
775 case 0:
Christoph Lametere24f0b82006-06-23 02:03:51 -0700776 break;
777 default:
Christoph Lameter2d1db3b2006-06-23 02:03:33 -0700778 /* Permanent failure */
Christoph Lameter2d1db3b2006-06-23 02:03:33 -0700779 nr_failed++;
Christoph Lametere24f0b82006-06-23 02:03:51 -0700780 break;
Christoph Lameter2d1db3b2006-06-23 02:03:33 -0700781 }
Christoph Lameterb20a3502006-03-22 00:09:12 -0800782 }
783 }
Christoph Lameter95a402c2006-06-23 02:03:53 -0700784 rc = 0;
785out:
Christoph Lameterb20a3502006-03-22 00:09:12 -0800786 if (!swapwrite)
787 current->flags &= ~PF_SWAPWRITE;
788
Christoph Lameteraaa994b2006-06-23 02:03:52 -0700789 putback_lru_pages(from);
Christoph Lameter95a402c2006-06-23 02:03:53 -0700790
791 if (rc)
792 return rc;
793
Christoph Lameterb20a3502006-03-22 00:09:12 -0800794 return nr_failed + retry;
795}
796
Christoph Lameter742755a2006-06-23 02:03:55 -0700797#ifdef CONFIG_NUMA
798/*
799 * Move a list of individual pages
800 */
801struct page_to_node {
802 unsigned long addr;
803 struct page *page;
804 int node;
805 int status;
806};
807
808static struct page *new_page_node(struct page *p, unsigned long private,
809 int **result)
810{
811 struct page_to_node *pm = (struct page_to_node *)private;
812
813 while (pm->node != MAX_NUMNODES && pm->page != p)
814 pm++;
815
816 if (pm->node == MAX_NUMNODES)
817 return NULL;
818
819 *result = &pm->status;
820
Mel Gorman6484eb32009-06-16 15:31:54 -0700821 return alloc_pages_exact_node(pm->node,
Mel Gorman769848c2007-07-17 04:03:05 -0700822 GFP_HIGHUSER_MOVABLE | GFP_THISNODE, 0);
Christoph Lameter742755a2006-06-23 02:03:55 -0700823}
824
825/*
826 * Move a set of pages as indicated in the pm array. The addr
827 * field must be set to the virtual address of the page to be moved
828 * and the node number must contain a valid target node.
Brice Goglin5e9a0f02008-10-18 20:27:17 -0700829 * The pm array ends with node = MAX_NUMNODES.
Christoph Lameter742755a2006-06-23 02:03:55 -0700830 */
Brice Goglin5e9a0f02008-10-18 20:27:17 -0700831static int do_move_page_to_node_array(struct mm_struct *mm,
832 struct page_to_node *pm,
833 int migrate_all)
Christoph Lameter742755a2006-06-23 02:03:55 -0700834{
835 int err;
836 struct page_to_node *pp;
837 LIST_HEAD(pagelist);
838
839 down_read(&mm->mmap_sem);
840
841 /*
842 * Build a list of pages to migrate
843 */
Christoph Lameter742755a2006-06-23 02:03:55 -0700844 for (pp = pm; pp->node != MAX_NUMNODES; pp++) {
845 struct vm_area_struct *vma;
846 struct page *page;
847
Christoph Lameter742755a2006-06-23 02:03:55 -0700848 err = -EFAULT;
849 vma = find_vma(mm, pp->addr);
Christoph Lameter0dc952d2007-03-05 00:30:33 -0800850 if (!vma || !vma_migratable(vma))
Christoph Lameter742755a2006-06-23 02:03:55 -0700851 goto set_status;
852
853 page = follow_page(vma, pp->addr, FOLL_GET);
Linus Torvalds89f5b7d2008-06-20 11:18:25 -0700854
855 err = PTR_ERR(page);
856 if (IS_ERR(page))
857 goto set_status;
858
Christoph Lameter742755a2006-06-23 02:03:55 -0700859 err = -ENOENT;
860 if (!page)
861 goto set_status;
862
863 if (PageReserved(page)) /* Check for zero page */
864 goto put_and_set;
865
866 pp->page = page;
867 err = page_to_nid(page);
868
869 if (err == pp->node)
870 /*
871 * Node already in the right place
872 */
873 goto put_and_set;
874
875 err = -EACCES;
876 if (page_mapcount(page) > 1 &&
877 !migrate_all)
878 goto put_and_set;
879
Nick Piggin62695a82008-10-18 20:26:09 -0700880 err = isolate_lru_page(page);
881 if (!err)
882 list_add_tail(&page->lru, &pagelist);
Christoph Lameter742755a2006-06-23 02:03:55 -0700883put_and_set:
884 /*
885 * Either remove the duplicate refcount from
886 * isolate_lru_page() or drop the page ref if it was
887 * not isolated.
888 */
889 put_page(page);
890set_status:
891 pp->status = err;
892 }
893
Brice Gogline78bbfa2008-10-18 20:27:15 -0700894 err = 0;
Christoph Lameter742755a2006-06-23 02:03:55 -0700895 if (!list_empty(&pagelist))
896 err = migrate_pages(&pagelist, new_page_node,
897 (unsigned long)pm);
Christoph Lameter742755a2006-06-23 02:03:55 -0700898
899 up_read(&mm->mmap_sem);
900 return err;
901}
902
903/*
Brice Goglin5e9a0f02008-10-18 20:27:17 -0700904 * Migrate an array of page address onto an array of nodes and fill
905 * the corresponding array of status.
906 */
907static int do_pages_move(struct mm_struct *mm, struct task_struct *task,
908 unsigned long nr_pages,
909 const void __user * __user *pages,
910 const int __user *nodes,
911 int __user *status, int flags)
912{
Brice Goglin3140a222009-01-06 14:38:57 -0800913 struct page_to_node *pm;
Brice Goglin5e9a0f02008-10-18 20:27:17 -0700914 nodemask_t task_nodes;
Brice Goglin3140a222009-01-06 14:38:57 -0800915 unsigned long chunk_nr_pages;
916 unsigned long chunk_start;
917 int err;
Brice Goglin5e9a0f02008-10-18 20:27:17 -0700918
919 task_nodes = cpuset_mems_allowed(task);
920
Brice Goglin3140a222009-01-06 14:38:57 -0800921 err = -ENOMEM;
922 pm = (struct page_to_node *)__get_free_page(GFP_KERNEL);
923 if (!pm)
Brice Goglin5e9a0f02008-10-18 20:27:17 -0700924 goto out;
Brice Goglin35282a22009-06-16 15:32:43 -0700925
926 migrate_prep();
927
Brice Goglin5e9a0f02008-10-18 20:27:17 -0700928 /*
Brice Goglin3140a222009-01-06 14:38:57 -0800929 * Store a chunk of page_to_node array in a page,
930 * but keep the last one as a marker
Brice Goglin5e9a0f02008-10-18 20:27:17 -0700931 */
Brice Goglin3140a222009-01-06 14:38:57 -0800932 chunk_nr_pages = (PAGE_SIZE / sizeof(struct page_to_node)) - 1;
Brice Goglin5e9a0f02008-10-18 20:27:17 -0700933
Brice Goglin3140a222009-01-06 14:38:57 -0800934 for (chunk_start = 0;
935 chunk_start < nr_pages;
936 chunk_start += chunk_nr_pages) {
937 int j;
Brice Goglin5e9a0f02008-10-18 20:27:17 -0700938
Brice Goglin3140a222009-01-06 14:38:57 -0800939 if (chunk_start + chunk_nr_pages > nr_pages)
940 chunk_nr_pages = nr_pages - chunk_start;
941
942 /* fill the chunk pm with addrs and nodes from user-space */
943 for (j = 0; j < chunk_nr_pages; j++) {
944 const void __user *p;
Brice Goglin5e9a0f02008-10-18 20:27:17 -0700945 int node;
946
Brice Goglin3140a222009-01-06 14:38:57 -0800947 err = -EFAULT;
948 if (get_user(p, pages + j + chunk_start))
949 goto out_pm;
950 pm[j].addr = (unsigned long) p;
951
952 if (get_user(node, nodes + j + chunk_start))
Brice Goglin5e9a0f02008-10-18 20:27:17 -0700953 goto out_pm;
954
955 err = -ENODEV;
956 if (!node_state(node, N_HIGH_MEMORY))
957 goto out_pm;
958
959 err = -EACCES;
960 if (!node_isset(node, task_nodes))
961 goto out_pm;
962
Brice Goglin3140a222009-01-06 14:38:57 -0800963 pm[j].node = node;
964 }
Brice Goglin5e9a0f02008-10-18 20:27:17 -0700965
Brice Goglin3140a222009-01-06 14:38:57 -0800966 /* End marker for this chunk */
967 pm[chunk_nr_pages].node = MAX_NUMNODES;
968
969 /* Migrate this chunk */
970 err = do_move_page_to_node_array(mm, pm,
971 flags & MPOL_MF_MOVE_ALL);
972 if (err < 0)
973 goto out_pm;
974
Brice Goglin5e9a0f02008-10-18 20:27:17 -0700975 /* Return status information */
Brice Goglin3140a222009-01-06 14:38:57 -0800976 for (j = 0; j < chunk_nr_pages; j++)
977 if (put_user(pm[j].status, status + j + chunk_start)) {
Brice Goglin5e9a0f02008-10-18 20:27:17 -0700978 err = -EFAULT;
Brice Goglin3140a222009-01-06 14:38:57 -0800979 goto out_pm;
980 }
981 }
982 err = 0;
Brice Goglin5e9a0f02008-10-18 20:27:17 -0700983
984out_pm:
Brice Goglin3140a222009-01-06 14:38:57 -0800985 free_page((unsigned long)pm);
Brice Goglin5e9a0f02008-10-18 20:27:17 -0700986out:
987 return err;
988}
989
990/*
Brice Goglin2f007e72008-10-18 20:27:16 -0700991 * Determine the nodes of an array of pages and store it in an array of status.
Christoph Lameter742755a2006-06-23 02:03:55 -0700992 */
Brice Goglin80bba122008-12-09 13:14:23 -0800993static void do_pages_stat_array(struct mm_struct *mm, unsigned long nr_pages,
994 const void __user **pages, int *status)
Christoph Lameter742755a2006-06-23 02:03:55 -0700995{
Brice Goglin2f007e72008-10-18 20:27:16 -0700996 unsigned long i;
Brice Goglin2f007e72008-10-18 20:27:16 -0700997
Christoph Lameter742755a2006-06-23 02:03:55 -0700998 down_read(&mm->mmap_sem);
999
Brice Goglin2f007e72008-10-18 20:27:16 -07001000 for (i = 0; i < nr_pages; i++) {
Brice Goglin80bba122008-12-09 13:14:23 -08001001 unsigned long addr = (unsigned long)(*pages);
Christoph Lameter742755a2006-06-23 02:03:55 -07001002 struct vm_area_struct *vma;
1003 struct page *page;
KOSAKI Motohiroc095adb2008-12-16 16:06:43 +09001004 int err = -EFAULT;
Brice Goglin2f007e72008-10-18 20:27:16 -07001005
1006 vma = find_vma(mm, addr);
Christoph Lameter742755a2006-06-23 02:03:55 -07001007 if (!vma)
1008 goto set_status;
1009
Brice Goglin2f007e72008-10-18 20:27:16 -07001010 page = follow_page(vma, addr, 0);
Linus Torvalds89f5b7d2008-06-20 11:18:25 -07001011
1012 err = PTR_ERR(page);
1013 if (IS_ERR(page))
1014 goto set_status;
1015
Christoph Lameter742755a2006-06-23 02:03:55 -07001016 err = -ENOENT;
1017 /* Use PageReserved to check for zero page */
1018 if (!page || PageReserved(page))
1019 goto set_status;
1020
1021 err = page_to_nid(page);
1022set_status:
Brice Goglin80bba122008-12-09 13:14:23 -08001023 *status = err;
1024
1025 pages++;
1026 status++;
1027 }
1028
1029 up_read(&mm->mmap_sem);
1030}
1031
1032/*
1033 * Determine the nodes of a user array of pages and store it in
1034 * a user array of status.
1035 */
1036static int do_pages_stat(struct mm_struct *mm, unsigned long nr_pages,
1037 const void __user * __user *pages,
1038 int __user *status)
1039{
1040#define DO_PAGES_STAT_CHUNK_NR 16
1041 const void __user *chunk_pages[DO_PAGES_STAT_CHUNK_NR];
1042 int chunk_status[DO_PAGES_STAT_CHUNK_NR];
1043 unsigned long i, chunk_nr = DO_PAGES_STAT_CHUNK_NR;
1044 int err;
1045
1046 for (i = 0; i < nr_pages; i += chunk_nr) {
1047 if (chunk_nr + i > nr_pages)
1048 chunk_nr = nr_pages - i;
1049
1050 err = copy_from_user(chunk_pages, &pages[i],
1051 chunk_nr * sizeof(*chunk_pages));
1052 if (err) {
1053 err = -EFAULT;
1054 goto out;
1055 }
1056
1057 do_pages_stat_array(mm, chunk_nr, chunk_pages, chunk_status);
1058
1059 err = copy_to_user(&status[i], chunk_status,
1060 chunk_nr * sizeof(*chunk_status));
1061 if (err) {
1062 err = -EFAULT;
1063 goto out;
1064 }
Christoph Lameter742755a2006-06-23 02:03:55 -07001065 }
Brice Goglin2f007e72008-10-18 20:27:16 -07001066 err = 0;
Christoph Lameter742755a2006-06-23 02:03:55 -07001067
Brice Goglin2f007e72008-10-18 20:27:16 -07001068out:
Brice Goglin2f007e72008-10-18 20:27:16 -07001069 return err;
Christoph Lameter742755a2006-06-23 02:03:55 -07001070}
1071
1072/*
1073 * Move a list of pages in the address space of the currently executing
1074 * process.
1075 */
Heiko Carstens938bb9f2009-01-14 14:14:30 +01001076SYSCALL_DEFINE6(move_pages, pid_t, pid, unsigned long, nr_pages,
1077 const void __user * __user *, pages,
1078 const int __user *, nodes,
1079 int __user *, status, int, flags)
Christoph Lameter742755a2006-06-23 02:03:55 -07001080{
David Howellsc69e8d92008-11-14 10:39:19 +11001081 const struct cred *cred = current_cred(), *tcred;
Christoph Lameter742755a2006-06-23 02:03:55 -07001082 struct task_struct *task;
Christoph Lameter742755a2006-06-23 02:03:55 -07001083 struct mm_struct *mm;
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001084 int err;
Christoph Lameter742755a2006-06-23 02:03:55 -07001085
1086 /* Check flags */
1087 if (flags & ~(MPOL_MF_MOVE|MPOL_MF_MOVE_ALL))
1088 return -EINVAL;
1089
1090 if ((flags & MPOL_MF_MOVE_ALL) && !capable(CAP_SYS_NICE))
1091 return -EPERM;
1092
1093 /* Find the mm_struct */
1094 read_lock(&tasklist_lock);
Pavel Emelyanov228ebcb2007-10-18 23:40:16 -07001095 task = pid ? find_task_by_vpid(pid) : current;
Christoph Lameter742755a2006-06-23 02:03:55 -07001096 if (!task) {
1097 read_unlock(&tasklist_lock);
1098 return -ESRCH;
1099 }
1100 mm = get_task_mm(task);
1101 read_unlock(&tasklist_lock);
1102
1103 if (!mm)
1104 return -EINVAL;
1105
1106 /*
1107 * Check if this process has the right to modify the specified
1108 * process. The right exists if the process has administrative
1109 * capabilities, superuser privileges or the same
1110 * userid as the target process.
1111 */
David Howellsc69e8d92008-11-14 10:39:19 +11001112 rcu_read_lock();
1113 tcred = __task_cred(task);
David Howellsb6dff3e2008-11-14 10:39:16 +11001114 if (cred->euid != tcred->suid && cred->euid != tcred->uid &&
1115 cred->uid != tcred->suid && cred->uid != tcred->uid &&
Christoph Lameter742755a2006-06-23 02:03:55 -07001116 !capable(CAP_SYS_NICE)) {
David Howellsc69e8d92008-11-14 10:39:19 +11001117 rcu_read_unlock();
Christoph Lameter742755a2006-06-23 02:03:55 -07001118 err = -EPERM;
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001119 goto out;
Christoph Lameter742755a2006-06-23 02:03:55 -07001120 }
David Howellsc69e8d92008-11-14 10:39:19 +11001121 rcu_read_unlock();
Christoph Lameter742755a2006-06-23 02:03:55 -07001122
David Quigley86c3a762006-06-23 02:04:02 -07001123 err = security_task_movememory(task);
1124 if (err)
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001125 goto out;
David Quigley86c3a762006-06-23 02:04:02 -07001126
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001127 if (nodes) {
1128 err = do_pages_move(mm, task, nr_pages, pages, nodes, status,
1129 flags);
1130 } else {
Brice Goglin2f007e72008-10-18 20:27:16 -07001131 err = do_pages_stat(mm, nr_pages, pages, status);
Brice Goglin2f007e72008-10-18 20:27:16 -07001132 }
David Quigley86c3a762006-06-23 02:04:02 -07001133
Christoph Lameter742755a2006-06-23 02:03:55 -07001134out:
Christoph Lameter742755a2006-06-23 02:03:55 -07001135 mmput(mm);
1136 return err;
1137}
Christoph Lameter742755a2006-06-23 02:03:55 -07001138
Christoph Lameter7b2259b2006-06-25 05:46:48 -07001139/*
1140 * Call migration functions in the vma_ops that may prepare
1141 * memory in a vm for migration. migration functions may perform
1142 * the migration for vmas that do not have an underlying page struct.
1143 */
1144int migrate_vmas(struct mm_struct *mm, const nodemask_t *to,
1145 const nodemask_t *from, unsigned long flags)
1146{
1147 struct vm_area_struct *vma;
1148 int err = 0;
1149
Daisuke Nishimura1001c9f2009-02-11 13:04:18 -08001150 for (vma = mm->mmap; vma && !err; vma = vma->vm_next) {
Christoph Lameter7b2259b2006-06-25 05:46:48 -07001151 if (vma->vm_ops && vma->vm_ops->migrate) {
1152 err = vma->vm_ops->migrate(vma, to, from, flags);
1153 if (err)
1154 break;
1155 }
1156 }
1157 return err;
1158}
Gerald Schaefer83d16742008-07-23 21:28:22 -07001159#endif