blob: 09e2471afa0f7f4ab658adca783c9a2bd1eadc4f [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>
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>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090035#include <linux/gfp.h>
Christoph Lameterb20a3502006-03-22 00:09:12 -080036
37#include "internal.h"
38
Christoph Lameterb20a3502006-03-22 00:09:12 -080039#define lru_to_page(_head) (list_entry((_head)->prev, struct page, lru))
40
41/*
Christoph Lameter742755a2006-06-23 02:03:55 -070042 * migrate_prep() needs to be called before we start compiling a list of pages
Mel Gorman748446b2010-05-24 14:32:27 -070043 * to be migrated using isolate_lru_page(). If scheduling work on other CPUs is
44 * undesirable, use migrate_prep_local()
Christoph Lameterb20a3502006-03-22 00:09:12 -080045 */
46int migrate_prep(void)
47{
Christoph Lameterb20a3502006-03-22 00:09:12 -080048 /*
49 * Clear the LRU lists so pages can be isolated.
50 * Note that pages may be moved off the LRU after we have
51 * drained them. Those pages will fail to migrate like other
52 * pages that may be busy.
53 */
54 lru_add_drain_all();
55
56 return 0;
57}
58
Mel Gorman748446b2010-05-24 14:32:27 -070059/* Do the necessary work of migrate_prep but not if it involves other CPUs */
60int migrate_prep_local(void)
61{
62 lru_add_drain();
63
64 return 0;
65}
66
Christoph Lameterb20a3502006-03-22 00:09:12 -080067/*
Lee Schermerhorn894bc312008-10-18 20:26:39 -070068 * Add isolated pages on the list back to the LRU under page lock
69 * to avoid leaking evictable pages back onto unevictable list.
Christoph Lameterb20a3502006-03-22 00:09:12 -080070 */
Minchan Kime13861d2010-05-24 14:31:59 -070071void putback_lru_pages(struct list_head *l)
Christoph Lameterb20a3502006-03-22 00:09:12 -080072{
73 struct page *page;
74 struct page *page2;
Christoph Lameterb20a3502006-03-22 00:09:12 -080075
76 list_for_each_entry_safe(page, page2, l, lru) {
Christoph Lametere24f0b82006-06-23 02:03:51 -070077 list_del(&page->lru);
KOSAKI Motohiroa7312862009-09-21 17:01:37 -070078 dec_zone_page_state(page, NR_ISOLATED_ANON +
Johannes Weiner6c0b1352009-09-21 17:02:59 -070079 page_is_file_cache(page));
Lee Schermerhorn894bc312008-10-18 20:26:39 -070080 putback_lru_page(page);
Christoph Lameterb20a3502006-03-22 00:09:12 -080081 }
Christoph Lameterb20a3502006-03-22 00:09:12 -080082}
83
Christoph Lameter06972122006-06-23 02:03:35 -070084/*
85 * Restore a potential migration pte to a working pte entry
86 */
Hugh Dickinse9995ef2009-12-14 17:59:31 -080087static int remove_migration_pte(struct page *new, struct vm_area_struct *vma,
88 unsigned long addr, void *old)
Christoph Lameter06972122006-06-23 02:03:35 -070089{
90 struct mm_struct *mm = vma->vm_mm;
91 swp_entry_t entry;
92 pgd_t *pgd;
93 pud_t *pud;
94 pmd_t *pmd;
95 pte_t *ptep, pte;
96 spinlock_t *ptl;
97
98 pgd = pgd_offset(mm, addr);
99 if (!pgd_present(*pgd))
Hugh Dickinse9995ef2009-12-14 17:59:31 -0800100 goto out;
Christoph Lameter06972122006-06-23 02:03:35 -0700101
102 pud = pud_offset(pgd, addr);
103 if (!pud_present(*pud))
Hugh Dickinse9995ef2009-12-14 17:59:31 -0800104 goto out;
Christoph Lameter06972122006-06-23 02:03:35 -0700105
106 pmd = pmd_offset(pud, addr);
107 if (!pmd_present(*pmd))
Hugh Dickinse9995ef2009-12-14 17:59:31 -0800108 goto out;
Christoph Lameter06972122006-06-23 02:03:35 -0700109
110 ptep = pte_offset_map(pmd, addr);
111
112 if (!is_swap_pte(*ptep)) {
113 pte_unmap(ptep);
Hugh Dickinse9995ef2009-12-14 17:59:31 -0800114 goto out;
Christoph Lameter06972122006-06-23 02:03:35 -0700115 }
116
117 ptl = pte_lockptr(mm, pmd);
118 spin_lock(ptl);
119 pte = *ptep;
120 if (!is_swap_pte(pte))
Hugh Dickinse9995ef2009-12-14 17:59:31 -0800121 goto unlock;
Christoph Lameter06972122006-06-23 02:03:35 -0700122
123 entry = pte_to_swp_entry(pte);
124
Hugh Dickinse9995ef2009-12-14 17:59:31 -0800125 if (!is_migration_entry(entry) ||
126 migration_entry_to_page(entry) != old)
127 goto unlock;
Christoph Lameter06972122006-06-23 02:03:35 -0700128
Christoph Lameter06972122006-06-23 02:03:35 -0700129 get_page(new);
130 pte = pte_mkold(mk_pte(new, vma->vm_page_prot));
131 if (is_write_migration_entry(entry))
132 pte = pte_mkwrite(pte);
KAMEZAWA Hiroyuki97ee0522007-10-16 01:25:43 -0700133 flush_cache_page(vma, addr, pte_pfn(pte));
Christoph Lameter06972122006-06-23 02:03:35 -0700134 set_pte_at(mm, addr, ptep, pte);
Christoph Lameter04e62a22006-06-23 02:03:38 -0700135
136 if (PageAnon(new))
137 page_add_anon_rmap(new, vma, addr);
138 else
139 page_add_file_rmap(new);
140
141 /* No need to invalidate - it was non-present before */
Russell King4b3073e2009-12-18 16:40:18 +0000142 update_mmu_cache(vma, addr, ptep);
Hugh Dickinse9995ef2009-12-14 17:59:31 -0800143unlock:
Christoph Lameter06972122006-06-23 02:03:35 -0700144 pte_unmap_unlock(ptep, ptl);
Hugh Dickinse9995ef2009-12-14 17:59:31 -0800145out:
146 return SWAP_AGAIN;
Christoph Lameter06972122006-06-23 02:03:35 -0700147}
148
149/*
Christoph Lameter04e62a22006-06-23 02:03:38 -0700150 * Get rid of all migration entries and replace them by
151 * references to the indicated page.
152 */
153static void remove_migration_ptes(struct page *old, struct page *new)
154{
Hugh Dickinse9995ef2009-12-14 17:59:31 -0800155 rmap_walk(new, remove_migration_pte, old);
Christoph Lameter04e62a22006-06-23 02:03:38 -0700156}
157
158/*
Christoph Lameter06972122006-06-23 02:03:35 -0700159 * Something used the pte of a page under migration. We need to
160 * get to the page and wait until migration is finished.
161 * When we return from this function the fault will be retried.
162 *
163 * This function is called from do_swap_page().
164 */
165void migration_entry_wait(struct mm_struct *mm, pmd_t *pmd,
166 unsigned long address)
167{
168 pte_t *ptep, pte;
169 spinlock_t *ptl;
170 swp_entry_t entry;
171 struct page *page;
172
173 ptep = pte_offset_map_lock(mm, pmd, address, &ptl);
174 pte = *ptep;
175 if (!is_swap_pte(pte))
176 goto out;
177
178 entry = pte_to_swp_entry(pte);
179 if (!is_migration_entry(entry))
180 goto out;
181
182 page = migration_entry_to_page(entry);
183
Nick Piggine2867812008-07-25 19:45:30 -0700184 /*
185 * Once radix-tree replacement of page migration started, page_count
186 * *must* be zero. And, we don't want to call wait_on_page_locked()
187 * against a page without get_page().
188 * So, we use get_page_unless_zero(), here. Even failed, page fault
189 * will occur again.
190 */
191 if (!get_page_unless_zero(page))
192 goto out;
Christoph Lameter06972122006-06-23 02:03:35 -0700193 pte_unmap_unlock(ptep, ptl);
194 wait_on_page_locked(page);
195 put_page(page);
196 return;
197out:
198 pte_unmap_unlock(ptep, ptl);
199}
200
Christoph Lameterb20a3502006-03-22 00:09:12 -0800201/*
Christoph Lameterc3fcf8a2006-06-23 02:03:32 -0700202 * Replace the page in the mapping.
Christoph Lameter5b5c7122006-06-23 02:03:29 -0700203 *
204 * The number of remaining references must be:
205 * 1 for anonymous pages without a mapping
206 * 2 for pages with a mapping
David Howells266cf652009-04-03 16:42:36 +0100207 * 3 for pages with a mapping and PagePrivate/PagePrivate2 set.
Christoph Lameterb20a3502006-03-22 00:09:12 -0800208 */
Christoph Lameter2d1db3b2006-06-23 02:03:33 -0700209static int migrate_page_move_mapping(struct address_space *mapping,
210 struct page *newpage, struct page *page)
Christoph Lameterb20a3502006-03-22 00:09:12 -0800211{
Nick Piggine2867812008-07-25 19:45:30 -0700212 int expected_count;
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800213 void **pslot;
Christoph Lameterb20a3502006-03-22 00:09:12 -0800214
Christoph Lameter6c5240a2006-06-23 02:03:37 -0700215 if (!mapping) {
Christoph Lameter0e8c7d02007-04-23 14:41:09 -0700216 /* Anonymous page without mapping */
Christoph Lameter6c5240a2006-06-23 02:03:37 -0700217 if (page_count(page) != 1)
218 return -EAGAIN;
219 return 0;
220 }
221
Nick Piggin19fd6232008-07-25 19:45:32 -0700222 spin_lock_irq(&mapping->tree_lock);
Christoph Lameterb20a3502006-03-22 00:09:12 -0800223
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800224 pslot = radix_tree_lookup_slot(&mapping->page_tree,
225 page_index(page));
Christoph Lameterb20a3502006-03-22 00:09:12 -0800226
Johannes Weineredcf4742009-09-21 17:02:59 -0700227 expected_count = 2 + page_has_private(page);
Nick Piggine2867812008-07-25 19:45:30 -0700228 if (page_count(page) != expected_count ||
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800229 (struct page *)radix_tree_deref_slot(pslot) != page) {
Nick Piggin19fd6232008-07-25 19:45:32 -0700230 spin_unlock_irq(&mapping->tree_lock);
Christoph Lametere23ca002006-04-10 22:52:57 -0700231 return -EAGAIN;
Christoph Lameterb20a3502006-03-22 00:09:12 -0800232 }
233
Nick Piggine2867812008-07-25 19:45:30 -0700234 if (!page_freeze_refs(page, expected_count)) {
Nick Piggin19fd6232008-07-25 19:45:32 -0700235 spin_unlock_irq(&mapping->tree_lock);
Nick Piggine2867812008-07-25 19:45:30 -0700236 return -EAGAIN;
237 }
238
Christoph Lameterb20a3502006-03-22 00:09:12 -0800239 /*
240 * Now we know that no one else is looking at the page.
Christoph Lameterb20a3502006-03-22 00:09:12 -0800241 */
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800242 get_page(newpage); /* add cache reference */
Christoph Lameterb20a3502006-03-22 00:09:12 -0800243 if (PageSwapCache(page)) {
244 SetPageSwapCache(newpage);
245 set_page_private(newpage, page_private(page));
246 }
247
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800248 radix_tree_replace_slot(pslot, newpage);
249
Nick Piggine2867812008-07-25 19:45:30 -0700250 page_unfreeze_refs(page, expected_count);
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800251 /*
252 * Drop cache reference from old page.
253 * We know this isn't the last reference.
254 */
Christoph Lameterb20a3502006-03-22 00:09:12 -0800255 __put_page(page);
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800256
Christoph Lameter0e8c7d02007-04-23 14:41:09 -0700257 /*
258 * If moved to a different zone then also account
259 * the page for that zone. Other VM counters will be
260 * taken care of when we establish references to the
261 * new page and drop references to the old page.
262 *
263 * Note that anonymous pages are accounted for
264 * via NR_FILE_PAGES and NR_ANON_PAGES if they
265 * are mapped to swap space.
266 */
267 __dec_zone_page_state(page, NR_FILE_PAGES);
268 __inc_zone_page_state(newpage, NR_FILE_PAGES);
KOSAKI Motohiro4b021082009-09-21 17:01:33 -0700269 if (PageSwapBacked(page)) {
270 __dec_zone_page_state(page, NR_SHMEM);
271 __inc_zone_page_state(newpage, NR_SHMEM);
272 }
Nick Piggin19fd6232008-07-25 19:45:32 -0700273 spin_unlock_irq(&mapping->tree_lock);
Christoph Lameterb20a3502006-03-22 00:09:12 -0800274
275 return 0;
276}
Christoph Lameterb20a3502006-03-22 00:09:12 -0800277
278/*
279 * Copy the page to its new location
280 */
Christoph Lametere7340f72006-06-23 02:03:29 -0700281static void migrate_page_copy(struct page *newpage, struct page *page)
Christoph Lameterb20a3502006-03-22 00:09:12 -0800282{
283 copy_highpage(newpage, page);
284
285 if (PageError(page))
286 SetPageError(newpage);
287 if (PageReferenced(page))
288 SetPageReferenced(newpage);
289 if (PageUptodate(page))
290 SetPageUptodate(newpage);
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700291 if (TestClearPageActive(page)) {
292 VM_BUG_ON(PageUnevictable(page));
Christoph Lameterb20a3502006-03-22 00:09:12 -0800293 SetPageActive(newpage);
Lee Schermerhorn418b27e2009-12-14 17:59:54 -0800294 } else if (TestClearPageUnevictable(page))
295 SetPageUnevictable(newpage);
Christoph Lameterb20a3502006-03-22 00:09:12 -0800296 if (PageChecked(page))
297 SetPageChecked(newpage);
298 if (PageMappedToDisk(page))
299 SetPageMappedToDisk(newpage);
300
301 if (PageDirty(page)) {
302 clear_page_dirty_for_io(page);
Nick Piggin3a902c52008-04-30 00:55:16 -0700303 /*
304 * Want to mark the page and the radix tree as dirty, and
305 * redo the accounting that clear_page_dirty_for_io undid,
306 * but we can't use set_page_dirty because that function
307 * is actually a signal that all of the page has become dirty.
308 * Wheras only part of our page may be dirty.
309 */
310 __set_page_dirty_nobuffers(newpage);
Christoph Lameterb20a3502006-03-22 00:09:12 -0800311 }
312
Nick Pigginb291f002008-10-18 20:26:44 -0700313 mlock_migrate_page(newpage, page);
Hugh Dickinse9995ef2009-12-14 17:59:31 -0800314 ksm_migrate_page(newpage, page);
Nick Pigginb291f002008-10-18 20:26:44 -0700315
Christoph Lameterb20a3502006-03-22 00:09:12 -0800316 ClearPageSwapCache(page);
Christoph Lameterb20a3502006-03-22 00:09:12 -0800317 ClearPagePrivate(page);
318 set_page_private(page, 0);
319 page->mapping = NULL;
320
321 /*
322 * If any waiters have accumulated on the new page then
323 * wake them up.
324 */
325 if (PageWriteback(newpage))
326 end_page_writeback(newpage);
327}
Christoph Lameterb20a3502006-03-22 00:09:12 -0800328
Christoph Lameter1d8b85c2006-06-23 02:03:28 -0700329/************************************************************
330 * Migration functions
331 ***********************************************************/
332
333/* Always fail migration. Used for mappings that are not movable */
Christoph Lameter2d1db3b2006-06-23 02:03:33 -0700334int fail_migrate_page(struct address_space *mapping,
335 struct page *newpage, struct page *page)
Christoph Lameter1d8b85c2006-06-23 02:03:28 -0700336{
337 return -EIO;
338}
339EXPORT_SYMBOL(fail_migrate_page);
340
Christoph Lameterb20a3502006-03-22 00:09:12 -0800341/*
342 * Common logic to directly migrate a single page suitable for
David Howells266cf652009-04-03 16:42:36 +0100343 * pages that do not use PagePrivate/PagePrivate2.
Christoph Lameterb20a3502006-03-22 00:09:12 -0800344 *
345 * Pages are locked upon entry and exit.
346 */
Christoph Lameter2d1db3b2006-06-23 02:03:33 -0700347int migrate_page(struct address_space *mapping,
348 struct page *newpage, struct page *page)
Christoph Lameterb20a3502006-03-22 00:09:12 -0800349{
350 int rc;
351
352 BUG_ON(PageWriteback(page)); /* Writeback must be complete */
353
Christoph Lameter2d1db3b2006-06-23 02:03:33 -0700354 rc = migrate_page_move_mapping(mapping, newpage, page);
Christoph Lameterb20a3502006-03-22 00:09:12 -0800355
356 if (rc)
357 return rc;
358
359 migrate_page_copy(newpage, page);
Christoph Lameterb20a3502006-03-22 00:09:12 -0800360 return 0;
361}
362EXPORT_SYMBOL(migrate_page);
363
David Howells93614012006-09-30 20:45:40 +0200364#ifdef CONFIG_BLOCK
Christoph Lameterb20a3502006-03-22 00:09:12 -0800365/*
Christoph Lameter1d8b85c2006-06-23 02:03:28 -0700366 * Migration function for pages with buffers. This function can only be used
367 * if the underlying filesystem guarantees that no other references to "page"
368 * exist.
369 */
Christoph Lameter2d1db3b2006-06-23 02:03:33 -0700370int buffer_migrate_page(struct address_space *mapping,
371 struct page *newpage, struct page *page)
Christoph Lameter1d8b85c2006-06-23 02:03:28 -0700372{
Christoph Lameter1d8b85c2006-06-23 02:03:28 -0700373 struct buffer_head *bh, *head;
374 int rc;
375
Christoph Lameter1d8b85c2006-06-23 02:03:28 -0700376 if (!page_has_buffers(page))
Christoph Lameter2d1db3b2006-06-23 02:03:33 -0700377 return migrate_page(mapping, newpage, page);
Christoph Lameter1d8b85c2006-06-23 02:03:28 -0700378
379 head = page_buffers(page);
380
Christoph Lameter2d1db3b2006-06-23 02:03:33 -0700381 rc = migrate_page_move_mapping(mapping, newpage, page);
Christoph Lameter1d8b85c2006-06-23 02:03:28 -0700382
383 if (rc)
384 return rc;
385
386 bh = head;
387 do {
388 get_bh(bh);
389 lock_buffer(bh);
390 bh = bh->b_this_page;
391
392 } while (bh != head);
393
394 ClearPagePrivate(page);
395 set_page_private(newpage, page_private(page));
396 set_page_private(page, 0);
397 put_page(page);
398 get_page(newpage);
399
400 bh = head;
401 do {
402 set_bh_page(bh, newpage, bh_offset(bh));
403 bh = bh->b_this_page;
404
405 } while (bh != head);
406
407 SetPagePrivate(newpage);
408
409 migrate_page_copy(newpage, page);
410
411 bh = head;
412 do {
413 unlock_buffer(bh);
414 put_bh(bh);
415 bh = bh->b_this_page;
416
417 } while (bh != head);
418
419 return 0;
420}
421EXPORT_SYMBOL(buffer_migrate_page);
David Howells93614012006-09-30 20:45:40 +0200422#endif
Christoph Lameter1d8b85c2006-06-23 02:03:28 -0700423
Christoph Lameter04e62a22006-06-23 02:03:38 -0700424/*
425 * Writeback a page to clean the dirty state
426 */
427static int writeout(struct address_space *mapping, struct page *page)
428{
429 struct writeback_control wbc = {
430 .sync_mode = WB_SYNC_NONE,
431 .nr_to_write = 1,
432 .range_start = 0,
433 .range_end = LLONG_MAX,
434 .nonblocking = 1,
435 .for_reclaim = 1
436 };
437 int rc;
438
439 if (!mapping->a_ops->writepage)
440 /* No write method for the address space */
441 return -EINVAL;
442
443 if (!clear_page_dirty_for_io(page))
444 /* Someone else already triggered a write */
445 return -EAGAIN;
446
447 /*
448 * A dirty page may imply that the underlying filesystem has
449 * the page on some queue. So the page must be clean for
450 * migration. Writeout may mean we loose the lock and the
451 * page state is no longer what we checked for earlier.
452 * At this point we know that the migration attempt cannot
453 * be successful.
454 */
455 remove_migration_ptes(page, page);
456
457 rc = mapping->a_ops->writepage(page, &wbc);
Christoph Lameter04e62a22006-06-23 02:03:38 -0700458
459 if (rc != AOP_WRITEPAGE_ACTIVATE)
460 /* unlocked. Relock */
461 lock_page(page);
462
Hugh Dickinsbda85502008-11-19 15:36:36 -0800463 return (rc < 0) ? -EIO : -EAGAIN;
Christoph Lameter04e62a22006-06-23 02:03:38 -0700464}
465
466/*
467 * Default handling if a filesystem does not provide a migration function.
468 */
Christoph Lameter8351a6e2006-06-23 02:03:33 -0700469static int fallback_migrate_page(struct address_space *mapping,
470 struct page *newpage, struct page *page)
471{
Christoph Lameter04e62a22006-06-23 02:03:38 -0700472 if (PageDirty(page))
473 return writeout(mapping, page);
Christoph Lameter8351a6e2006-06-23 02:03:33 -0700474
475 /*
476 * Buffers may be managed in a filesystem specific way.
477 * We must have no buffers or drop them.
478 */
David Howells266cf652009-04-03 16:42:36 +0100479 if (page_has_private(page) &&
Christoph Lameter8351a6e2006-06-23 02:03:33 -0700480 !try_to_release_page(page, GFP_KERNEL))
481 return -EAGAIN;
482
483 return migrate_page(mapping, newpage, page);
484}
485
Christoph Lameter1d8b85c2006-06-23 02:03:28 -0700486/*
Christoph Lametere24f0b82006-06-23 02:03:51 -0700487 * Move a page to a newly allocated page
488 * The page is locked and all ptes have been successfully removed.
489 *
490 * The new page will have replaced the old page if this function
491 * is successful.
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700492 *
493 * Return value:
494 * < 0 - error code
495 * == 0 - success
Christoph Lametere24f0b82006-06-23 02:03:51 -0700496 */
Mel Gorman3fe20112010-05-24 14:32:20 -0700497static int move_to_new_page(struct page *newpage, struct page *page,
498 int remap_swapcache)
Christoph Lametere24f0b82006-06-23 02:03:51 -0700499{
500 struct address_space *mapping;
501 int rc;
502
503 /*
504 * Block others from accessing the page when we get around to
505 * establishing additional references. We are the only one
506 * holding a reference to the new page at this point.
507 */
Nick Piggin529ae9a2008-08-02 12:01:03 +0200508 if (!trylock_page(newpage))
Christoph Lametere24f0b82006-06-23 02:03:51 -0700509 BUG();
510
511 /* Prepare mapping for the new page.*/
512 newpage->index = page->index;
513 newpage->mapping = page->mapping;
Rik van Rielb2e18532008-10-18 20:26:30 -0700514 if (PageSwapBacked(page))
515 SetPageSwapBacked(newpage);
Christoph Lametere24f0b82006-06-23 02:03:51 -0700516
517 mapping = page_mapping(page);
518 if (!mapping)
519 rc = migrate_page(mapping, newpage, page);
520 else if (mapping->a_ops->migratepage)
521 /*
522 * Most pages have a mapping and most filesystems
523 * should provide a migration function. Anonymous
524 * pages are part of swap space which also has its
525 * own migration function. This is the most common
526 * path for page migration.
527 */
528 rc = mapping->a_ops->migratepage(mapping,
529 newpage, page);
530 else
531 rc = fallback_migrate_page(mapping, newpage, page);
532
Mel Gorman3fe20112010-05-24 14:32:20 -0700533 if (rc) {
Christoph Lametere24f0b82006-06-23 02:03:51 -0700534 newpage->mapping = NULL;
Mel Gorman3fe20112010-05-24 14:32:20 -0700535 } else {
536 if (remap_swapcache)
537 remove_migration_ptes(page, newpage);
538 }
Christoph Lametere24f0b82006-06-23 02:03:51 -0700539
540 unlock_page(newpage);
541
542 return rc;
543}
544
545/*
546 * Obtain the lock on page, remove all ptes and migrate the page
547 * to the newly allocated page in newpage.
548 */
Christoph Lameter95a402c2006-06-23 02:03:53 -0700549static int unmap_and_move(new_page_t get_new_page, unsigned long private,
Hugh Dickins62b61f62009-12-14 17:59:33 -0800550 struct page *page, int force, int offlining)
Christoph Lametere24f0b82006-06-23 02:03:51 -0700551{
552 int rc = 0;
Christoph Lameter742755a2006-06-23 02:03:55 -0700553 int *result = NULL;
554 struct page *newpage = get_new_page(page, private, &result);
Mel Gorman3fe20112010-05-24 14:32:20 -0700555 int remap_swapcache = 1;
KAMEZAWA Hiroyuki989f89c2007-08-30 23:56:21 -0700556 int rcu_locked = 0;
KAMEZAWA Hiroyukiae41be32008-02-07 00:14:10 -0800557 int charge = 0;
KAMEZAWA Hiroyukie00e4312009-11-11 14:26:26 -0800558 struct mem_cgroup *mem = NULL;
Mel Gorman3f6c8272010-05-24 14:32:17 -0700559 struct anon_vma *anon_vma = NULL;
Christoph Lameter95a402c2006-06-23 02:03:53 -0700560
561 if (!newpage)
562 return -ENOMEM;
Christoph Lametere24f0b82006-06-23 02:03:51 -0700563
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700564 if (page_count(page) == 1) {
Christoph Lametere24f0b82006-06-23 02:03:51 -0700565 /* page was freed from under us. So we are done. */
Christoph Lameter95a402c2006-06-23 02:03:53 -0700566 goto move_newpage;
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700567 }
Christoph Lametere24f0b82006-06-23 02:03:51 -0700568
KAMEZAWA Hiroyukie8589cc2008-07-25 01:47:10 -0700569 /* prepare cgroup just returns 0 or -ENOMEM */
Christoph Lametere24f0b82006-06-23 02:03:51 -0700570 rc = -EAGAIN;
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -0800571
Nick Piggin529ae9a2008-08-02 12:01:03 +0200572 if (!trylock_page(page)) {
Christoph Lametere24f0b82006-06-23 02:03:51 -0700573 if (!force)
Christoph Lameter95a402c2006-06-23 02:03:53 -0700574 goto move_newpage;
Christoph Lametere24f0b82006-06-23 02:03:51 -0700575 lock_page(page);
576 }
577
Hugh Dickins62b61f62009-12-14 17:59:33 -0800578 /*
579 * Only memory hotplug's offline_pages() caller has locked out KSM,
580 * and can safely migrate a KSM page. The other cases have skipped
581 * PageKsm along with PageReserved - but it is only now when we have
582 * the page lock that we can be certain it will not go KSM beneath us
583 * (KSM will not upgrade a page from PageAnon to PageKsm when it sees
584 * its pagecount raised, but only here do we take the page lock which
585 * serializes that).
586 */
587 if (PageKsm(page) && !offlining) {
588 rc = -EBUSY;
589 goto unlock;
590 }
591
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -0800592 /* charge against new page */
593 charge = mem_cgroup_prepare_migration(page, &mem);
594 if (charge == -ENOMEM) {
595 rc = -ENOMEM;
596 goto unlock;
597 }
598 BUG_ON(charge);
599
Christoph Lametere24f0b82006-06-23 02:03:51 -0700600 if (PageWriteback(page)) {
601 if (!force)
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -0800602 goto uncharge;
Christoph Lametere24f0b82006-06-23 02:03:51 -0700603 wait_on_page_writeback(page);
604 }
Christoph Lametere24f0b82006-06-23 02:03:51 -0700605 /*
KAMEZAWA Hiroyukidc386d42007-07-26 10:41:07 -0700606 * By try_to_unmap(), page->mapcount goes down to 0 here. In this case,
607 * we cannot notice that anon_vma is freed while we migrates a page.
608 * This rcu_read_lock() delays freeing anon_vma pointer until the end
609 * of migration. File cache pages are no problem because of page_lock()
KAMEZAWA Hiroyuki989f89c2007-08-30 23:56:21 -0700610 * File Caches may use write_page() or lock_page() in migration, then,
611 * just care Anon page here.
Christoph Lametere24f0b82006-06-23 02:03:51 -0700612 */
KAMEZAWA Hiroyuki989f89c2007-08-30 23:56:21 -0700613 if (PageAnon(page)) {
614 rcu_read_lock();
615 rcu_locked = 1;
Mel Gorman67b95092010-05-24 14:32:19 -0700616
Mel Gorman3fe20112010-05-24 14:32:20 -0700617 /* Determine how to safely use anon_vma */
618 if (!page_mapped(page)) {
619 if (!PageSwapCache(page))
620 goto rcu_unlock;
Mel Gorman67b95092010-05-24 14:32:19 -0700621
Mel Gorman3fe20112010-05-24 14:32:20 -0700622 /*
623 * We cannot be sure that the anon_vma of an unmapped
624 * swapcache page is safe to use because we don't
625 * know in advance if the VMA that this page belonged
626 * to still exists. If the VMA and others sharing the
627 * data have been freed, then the anon_vma could
628 * already be invalid.
629 *
630 * To avoid this possibility, swapcache pages get
631 * migrated but are not remapped when migration
632 * completes
633 */
634 remap_swapcache = 0;
635 } else {
636 /*
637 * Take a reference count on the anon_vma if the
638 * page is mapped so that it is guaranteed to
639 * exist when the page is remapped later
640 */
641 anon_vma = page_anon_vma(page);
642 atomic_inc(&anon_vma->external_refcount);
643 }
KAMEZAWA Hiroyuki989f89c2007-08-30 23:56:21 -0700644 }
Shaohua Li62e1c552008-02-04 22:29:33 -0800645
KAMEZAWA Hiroyukidc386d42007-07-26 10:41:07 -0700646 /*
Shaohua Li62e1c552008-02-04 22:29:33 -0800647 * Corner case handling:
648 * 1. When a new swap-cache page is read into, it is added to the LRU
649 * and treated as swapcache but it has no rmap yet.
650 * Calling try_to_unmap() against a page->mapping==NULL page will
651 * trigger a BUG. So handle it here.
652 * 2. An orphaned page (see truncate_complete_page) might have
653 * fs-private metadata. The page can be picked up due to memory
654 * offlining. Everywhere else except page reclaim, the page is
655 * invisible to the vm, so the page can not be migrated. So try to
656 * free the metadata, so the page can be freed.
KAMEZAWA Hiroyukidc386d42007-07-26 10:41:07 -0700657 */
Shaohua Li62e1c552008-02-04 22:29:33 -0800658 if (!page->mapping) {
David Howells266cf652009-04-03 16:42:36 +0100659 if (!PageAnon(page) && page_has_private(page)) {
Shaohua Li62e1c552008-02-04 22:29:33 -0800660 /*
661 * Go direct to try_to_free_buffers() here because
662 * a) that's what try_to_release_page() would do anyway
663 * b) we may be under rcu_read_lock() here, so we can't
664 * use GFP_KERNEL which is what try_to_release_page()
665 * needs to be effective.
666 */
667 try_to_free_buffers(page);
Shaohua Liabfc3482009-09-21 17:01:19 -0700668 goto rcu_unlock;
Shaohua Li62e1c552008-02-04 22:29:33 -0800669 }
Shaohua Liabfc3482009-09-21 17:01:19 -0700670 goto skip_unmap;
Shaohua Li62e1c552008-02-04 22:29:33 -0800671 }
672
KAMEZAWA Hiroyukidc386d42007-07-26 10:41:07 -0700673 /* Establish migration ptes or remove ptes */
Andi Kleen14fa31b2009-09-16 11:50:10 +0200674 try_to_unmap(page, TTU_MIGRATION|TTU_IGNORE_MLOCK|TTU_IGNORE_ACCESS);
KAMEZAWA Hiroyukidc386d42007-07-26 10:41:07 -0700675
Shaohua Liabfc3482009-09-21 17:01:19 -0700676skip_unmap:
Christoph Lametere6a15302006-06-25 05:46:49 -0700677 if (!page_mapped(page))
Mel Gorman3fe20112010-05-24 14:32:20 -0700678 rc = move_to_new_page(newpage, page, remap_swapcache);
Christoph Lametere24f0b82006-06-23 02:03:51 -0700679
Mel Gorman3fe20112010-05-24 14:32:20 -0700680 if (rc && remap_swapcache)
Christoph Lametere24f0b82006-06-23 02:03:51 -0700681 remove_migration_ptes(page, page);
KAMEZAWA Hiroyukidc386d42007-07-26 10:41:07 -0700682rcu_unlock:
Mel Gorman3f6c8272010-05-24 14:32:17 -0700683
684 /* Drop an anon_vma reference if we took one */
Mel Gorman7f60c212010-05-24 14:32:18 -0700685 if (anon_vma && atomic_dec_and_lock(&anon_vma->external_refcount, &anon_vma->lock)) {
Mel Gorman3f6c8272010-05-24 14:32:17 -0700686 int empty = list_empty(&anon_vma->head);
687 spin_unlock(&anon_vma->lock);
688 if (empty)
689 anon_vma_free(anon_vma);
690 }
691
KAMEZAWA Hiroyuki989f89c2007-08-30 23:56:21 -0700692 if (rcu_locked)
693 rcu_read_unlock();
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -0800694uncharge:
695 if (!charge)
696 mem_cgroup_end_migration(mem, page, newpage);
Christoph Lametere24f0b82006-06-23 02:03:51 -0700697unlock:
698 unlock_page(page);
Christoph Lameter95a402c2006-06-23 02:03:53 -0700699
Christoph Lametere24f0b82006-06-23 02:03:51 -0700700 if (rc != -EAGAIN) {
Christoph Lameteraaa994b2006-06-23 02:03:52 -0700701 /*
702 * A page that has been migrated has all references
703 * removed and will be freed. A page that has not been
704 * migrated will have kepts its references and be
705 * restored.
706 */
707 list_del(&page->lru);
KOSAKI Motohiroa7312862009-09-21 17:01:37 -0700708 dec_zone_page_state(page, NR_ISOLATED_ANON +
Johannes Weiner6c0b1352009-09-21 17:02:59 -0700709 page_is_file_cache(page));
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700710 putback_lru_page(page);
Christoph Lametere24f0b82006-06-23 02:03:51 -0700711 }
Christoph Lameter95a402c2006-06-23 02:03:53 -0700712
713move_newpage:
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700714
Christoph Lameter95a402c2006-06-23 02:03:53 -0700715 /*
716 * Move the new page to the LRU. If migration was not successful
717 * then this will free the page.
718 */
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700719 putback_lru_page(newpage);
720
Christoph Lameter742755a2006-06-23 02:03:55 -0700721 if (result) {
722 if (rc)
723 *result = rc;
724 else
725 *result = page_to_nid(newpage);
726 }
Christoph Lametere24f0b82006-06-23 02:03:51 -0700727 return rc;
728}
729
730/*
Christoph Lameterb20a3502006-03-22 00:09:12 -0800731 * migrate_pages
732 *
Christoph Lameter95a402c2006-06-23 02:03:53 -0700733 * The function takes one list of pages to migrate and a function
734 * that determines from the page to be migrated and the private data
735 * the target of the move and allocates the page.
Christoph Lameterb20a3502006-03-22 00:09:12 -0800736 *
737 * The function returns after 10 attempts or if no pages
738 * are movable anymore because to has become empty
Christoph Lameteraaa994b2006-06-23 02:03:52 -0700739 * or no retryable pages exist anymore. All pages will be
Gabriel Craciunescue9534b32007-10-20 02:13:26 +0200740 * returned to the LRU or freed.
Christoph Lameterb20a3502006-03-22 00:09:12 -0800741 *
Christoph Lameter95a402c2006-06-23 02:03:53 -0700742 * Return: Number of pages not migrated or error code.
Christoph Lameterb20a3502006-03-22 00:09:12 -0800743 */
Christoph Lameter95a402c2006-06-23 02:03:53 -0700744int migrate_pages(struct list_head *from,
Hugh Dickins62b61f62009-12-14 17:59:33 -0800745 new_page_t get_new_page, unsigned long private, int offlining)
Christoph Lameterb20a3502006-03-22 00:09:12 -0800746{
Christoph Lametere24f0b82006-06-23 02:03:51 -0700747 int retry = 1;
Christoph Lameterb20a3502006-03-22 00:09:12 -0800748 int nr_failed = 0;
749 int pass = 0;
750 struct page *page;
751 struct page *page2;
752 int swapwrite = current->flags & PF_SWAPWRITE;
753 int rc;
754
755 if (!swapwrite)
756 current->flags |= PF_SWAPWRITE;
757
Christoph Lametere24f0b82006-06-23 02:03:51 -0700758 for(pass = 0; pass < 10 && retry; pass++) {
759 retry = 0;
Christoph Lameterb20a3502006-03-22 00:09:12 -0800760
Christoph Lametere24f0b82006-06-23 02:03:51 -0700761 list_for_each_entry_safe(page, page2, from, lru) {
Christoph Lametere24f0b82006-06-23 02:03:51 -0700762 cond_resched();
Christoph Lameterb20a3502006-03-22 00:09:12 -0800763
Christoph Lameter95a402c2006-06-23 02:03:53 -0700764 rc = unmap_and_move(get_new_page, private,
Hugh Dickins62b61f62009-12-14 17:59:33 -0800765 page, pass > 2, offlining);
Christoph Lameterb20a3502006-03-22 00:09:12 -0800766
Christoph Lametere24f0b82006-06-23 02:03:51 -0700767 switch(rc) {
Christoph Lameter95a402c2006-06-23 02:03:53 -0700768 case -ENOMEM:
769 goto out;
Christoph Lametere24f0b82006-06-23 02:03:51 -0700770 case -EAGAIN:
Christoph Lameter2d1db3b2006-06-23 02:03:33 -0700771 retry++;
Christoph Lametere24f0b82006-06-23 02:03:51 -0700772 break;
773 case 0:
Christoph Lametere24f0b82006-06-23 02:03:51 -0700774 break;
775 default:
Christoph Lameter2d1db3b2006-06-23 02:03:33 -0700776 /* Permanent failure */
Christoph Lameter2d1db3b2006-06-23 02:03:33 -0700777 nr_failed++;
Christoph Lametere24f0b82006-06-23 02:03:51 -0700778 break;
Christoph Lameter2d1db3b2006-06-23 02:03:33 -0700779 }
Christoph Lameterb20a3502006-03-22 00:09:12 -0800780 }
781 }
Christoph Lameter95a402c2006-06-23 02:03:53 -0700782 rc = 0;
783out:
Christoph Lameterb20a3502006-03-22 00:09:12 -0800784 if (!swapwrite)
785 current->flags &= ~PF_SWAPWRITE;
786
Christoph Lameteraaa994b2006-06-23 02:03:52 -0700787 putback_lru_pages(from);
Christoph Lameter95a402c2006-06-23 02:03:53 -0700788
789 if (rc)
790 return rc;
791
Christoph Lameterb20a3502006-03-22 00:09:12 -0800792 return nr_failed + retry;
793}
794
Christoph Lameter742755a2006-06-23 02:03:55 -0700795#ifdef CONFIG_NUMA
796/*
797 * Move a list of individual pages
798 */
799struct page_to_node {
800 unsigned long addr;
801 struct page *page;
802 int node;
803 int status;
804};
805
806static struct page *new_page_node(struct page *p, unsigned long private,
807 int **result)
808{
809 struct page_to_node *pm = (struct page_to_node *)private;
810
811 while (pm->node != MAX_NUMNODES && pm->page != p)
812 pm++;
813
814 if (pm->node == MAX_NUMNODES)
815 return NULL;
816
817 *result = &pm->status;
818
Mel Gorman6484eb32009-06-16 15:31:54 -0700819 return alloc_pages_exact_node(pm->node,
Mel Gorman769848c2007-07-17 04:03:05 -0700820 GFP_HIGHUSER_MOVABLE | GFP_THISNODE, 0);
Christoph Lameter742755a2006-06-23 02:03:55 -0700821}
822
823/*
824 * Move a set of pages as indicated in the pm array. The addr
825 * field must be set to the virtual address of the page to be moved
826 * and the node number must contain a valid target node.
Brice Goglin5e9a0f02008-10-18 20:27:17 -0700827 * The pm array ends with node = MAX_NUMNODES.
Christoph Lameter742755a2006-06-23 02:03:55 -0700828 */
Brice Goglin5e9a0f02008-10-18 20:27:17 -0700829static int do_move_page_to_node_array(struct mm_struct *mm,
830 struct page_to_node *pm,
831 int migrate_all)
Christoph Lameter742755a2006-06-23 02:03:55 -0700832{
833 int err;
834 struct page_to_node *pp;
835 LIST_HEAD(pagelist);
836
837 down_read(&mm->mmap_sem);
838
839 /*
840 * Build a list of pages to migrate
841 */
Christoph Lameter742755a2006-06-23 02:03:55 -0700842 for (pp = pm; pp->node != MAX_NUMNODES; pp++) {
843 struct vm_area_struct *vma;
844 struct page *page;
845
Christoph Lameter742755a2006-06-23 02:03:55 -0700846 err = -EFAULT;
847 vma = find_vma(mm, pp->addr);
Christoph Lameter0dc952d2007-03-05 00:30:33 -0800848 if (!vma || !vma_migratable(vma))
Christoph Lameter742755a2006-06-23 02:03:55 -0700849 goto set_status;
850
851 page = follow_page(vma, pp->addr, FOLL_GET);
Linus Torvalds89f5b7d2008-06-20 11:18:25 -0700852
853 err = PTR_ERR(page);
854 if (IS_ERR(page))
855 goto set_status;
856
Christoph Lameter742755a2006-06-23 02:03:55 -0700857 err = -ENOENT;
858 if (!page)
859 goto set_status;
860
Hugh Dickins62b61f62009-12-14 17:59:33 -0800861 /* Use PageReserved to check for zero page */
862 if (PageReserved(page) || PageKsm(page))
Christoph Lameter742755a2006-06-23 02:03:55 -0700863 goto put_and_set;
864
865 pp->page = page;
866 err = page_to_nid(page);
867
868 if (err == pp->node)
869 /*
870 * Node already in the right place
871 */
872 goto put_and_set;
873
874 err = -EACCES;
875 if (page_mapcount(page) > 1 &&
876 !migrate_all)
877 goto put_and_set;
878
Nick Piggin62695a82008-10-18 20:26:09 -0700879 err = isolate_lru_page(page);
KOSAKI Motohiro6d9c2852009-12-14 17:58:11 -0800880 if (!err) {
Nick Piggin62695a82008-10-18 20:26:09 -0700881 list_add_tail(&page->lru, &pagelist);
KOSAKI Motohiro6d9c2852009-12-14 17:58:11 -0800882 inc_zone_page_state(page, NR_ISOLATED_ANON +
883 page_is_file_cache(page));
884 }
Christoph Lameter742755a2006-06-23 02:03:55 -0700885put_and_set:
886 /*
887 * Either remove the duplicate refcount from
888 * isolate_lru_page() or drop the page ref if it was
889 * not isolated.
890 */
891 put_page(page);
892set_status:
893 pp->status = err;
894 }
895
Brice Gogline78bbfa2008-10-18 20:27:15 -0700896 err = 0;
Christoph Lameter742755a2006-06-23 02:03:55 -0700897 if (!list_empty(&pagelist))
898 err = migrate_pages(&pagelist, new_page_node,
Hugh Dickins62b61f62009-12-14 17:59:33 -0800899 (unsigned long)pm, 0);
Christoph Lameter742755a2006-06-23 02:03:55 -0700900
901 up_read(&mm->mmap_sem);
902 return err;
903}
904
905/*
Brice Goglin5e9a0f02008-10-18 20:27:17 -0700906 * Migrate an array of page address onto an array of nodes and fill
907 * the corresponding array of status.
908 */
909static int do_pages_move(struct mm_struct *mm, struct task_struct *task,
910 unsigned long nr_pages,
911 const void __user * __user *pages,
912 const int __user *nodes,
913 int __user *status, int flags)
914{
Brice Goglin3140a222009-01-06 14:38:57 -0800915 struct page_to_node *pm;
Brice Goglin5e9a0f02008-10-18 20:27:17 -0700916 nodemask_t task_nodes;
Brice Goglin3140a222009-01-06 14:38:57 -0800917 unsigned long chunk_nr_pages;
918 unsigned long chunk_start;
919 int err;
Brice Goglin5e9a0f02008-10-18 20:27:17 -0700920
921 task_nodes = cpuset_mems_allowed(task);
922
Brice Goglin3140a222009-01-06 14:38:57 -0800923 err = -ENOMEM;
924 pm = (struct page_to_node *)__get_free_page(GFP_KERNEL);
925 if (!pm)
Brice Goglin5e9a0f02008-10-18 20:27:17 -0700926 goto out;
Brice Goglin35282a22009-06-16 15:32:43 -0700927
928 migrate_prep();
929
Brice Goglin5e9a0f02008-10-18 20:27:17 -0700930 /*
Brice Goglin3140a222009-01-06 14:38:57 -0800931 * Store a chunk of page_to_node array in a page,
932 * but keep the last one as a marker
Brice Goglin5e9a0f02008-10-18 20:27:17 -0700933 */
Brice Goglin3140a222009-01-06 14:38:57 -0800934 chunk_nr_pages = (PAGE_SIZE / sizeof(struct page_to_node)) - 1;
Brice Goglin5e9a0f02008-10-18 20:27:17 -0700935
Brice Goglin3140a222009-01-06 14:38:57 -0800936 for (chunk_start = 0;
937 chunk_start < nr_pages;
938 chunk_start += chunk_nr_pages) {
939 int j;
Brice Goglin5e9a0f02008-10-18 20:27:17 -0700940
Brice Goglin3140a222009-01-06 14:38:57 -0800941 if (chunk_start + chunk_nr_pages > nr_pages)
942 chunk_nr_pages = nr_pages - chunk_start;
943
944 /* fill the chunk pm with addrs and nodes from user-space */
945 for (j = 0; j < chunk_nr_pages; j++) {
946 const void __user *p;
Brice Goglin5e9a0f02008-10-18 20:27:17 -0700947 int node;
948
Brice Goglin3140a222009-01-06 14:38:57 -0800949 err = -EFAULT;
950 if (get_user(p, pages + j + chunk_start))
951 goto out_pm;
952 pm[j].addr = (unsigned long) p;
953
954 if (get_user(node, nodes + j + chunk_start))
Brice Goglin5e9a0f02008-10-18 20:27:17 -0700955 goto out_pm;
956
957 err = -ENODEV;
Linus Torvalds6f5a55f2010-02-05 16:16:50 -0800958 if (node < 0 || node >= MAX_NUMNODES)
959 goto out_pm;
960
Brice Goglin5e9a0f02008-10-18 20:27:17 -0700961 if (!node_state(node, N_HIGH_MEMORY))
962 goto out_pm;
963
964 err = -EACCES;
965 if (!node_isset(node, task_nodes))
966 goto out_pm;
967
Brice Goglin3140a222009-01-06 14:38:57 -0800968 pm[j].node = node;
969 }
Brice Goglin5e9a0f02008-10-18 20:27:17 -0700970
Brice Goglin3140a222009-01-06 14:38:57 -0800971 /* End marker for this chunk */
972 pm[chunk_nr_pages].node = MAX_NUMNODES;
973
974 /* Migrate this chunk */
975 err = do_move_page_to_node_array(mm, pm,
976 flags & MPOL_MF_MOVE_ALL);
977 if (err < 0)
978 goto out_pm;
979
Brice Goglin5e9a0f02008-10-18 20:27:17 -0700980 /* Return status information */
Brice Goglin3140a222009-01-06 14:38:57 -0800981 for (j = 0; j < chunk_nr_pages; j++)
982 if (put_user(pm[j].status, status + j + chunk_start)) {
Brice Goglin5e9a0f02008-10-18 20:27:17 -0700983 err = -EFAULT;
Brice Goglin3140a222009-01-06 14:38:57 -0800984 goto out_pm;
985 }
986 }
987 err = 0;
Brice Goglin5e9a0f02008-10-18 20:27:17 -0700988
989out_pm:
Brice Goglin3140a222009-01-06 14:38:57 -0800990 free_page((unsigned long)pm);
Brice Goglin5e9a0f02008-10-18 20:27:17 -0700991out:
992 return err;
993}
994
995/*
Brice Goglin2f007e72008-10-18 20:27:16 -0700996 * Determine the nodes of an array of pages and store it in an array of status.
Christoph Lameter742755a2006-06-23 02:03:55 -0700997 */
Brice Goglin80bba122008-12-09 13:14:23 -0800998static void do_pages_stat_array(struct mm_struct *mm, unsigned long nr_pages,
999 const void __user **pages, int *status)
Christoph Lameter742755a2006-06-23 02:03:55 -07001000{
Brice Goglin2f007e72008-10-18 20:27:16 -07001001 unsigned long i;
Brice Goglin2f007e72008-10-18 20:27:16 -07001002
Christoph Lameter742755a2006-06-23 02:03:55 -07001003 down_read(&mm->mmap_sem);
1004
Brice Goglin2f007e72008-10-18 20:27:16 -07001005 for (i = 0; i < nr_pages; i++) {
Brice Goglin80bba122008-12-09 13:14:23 -08001006 unsigned long addr = (unsigned long)(*pages);
Christoph Lameter742755a2006-06-23 02:03:55 -07001007 struct vm_area_struct *vma;
1008 struct page *page;
KOSAKI Motohiroc095adb2008-12-16 16:06:43 +09001009 int err = -EFAULT;
Brice Goglin2f007e72008-10-18 20:27:16 -07001010
1011 vma = find_vma(mm, addr);
Christoph Lameter742755a2006-06-23 02:03:55 -07001012 if (!vma)
1013 goto set_status;
1014
Brice Goglin2f007e72008-10-18 20:27:16 -07001015 page = follow_page(vma, addr, 0);
Linus Torvalds89f5b7d2008-06-20 11:18:25 -07001016
1017 err = PTR_ERR(page);
1018 if (IS_ERR(page))
1019 goto set_status;
1020
Christoph Lameter742755a2006-06-23 02:03:55 -07001021 err = -ENOENT;
1022 /* Use PageReserved to check for zero page */
Hugh Dickins62b61f62009-12-14 17:59:33 -08001023 if (!page || PageReserved(page) || PageKsm(page))
Christoph Lameter742755a2006-06-23 02:03:55 -07001024 goto set_status;
1025
1026 err = page_to_nid(page);
1027set_status:
Brice Goglin80bba122008-12-09 13:14:23 -08001028 *status = err;
1029
1030 pages++;
1031 status++;
1032 }
1033
1034 up_read(&mm->mmap_sem);
1035}
1036
1037/*
1038 * Determine the nodes of a user array of pages and store it in
1039 * a user array of status.
1040 */
1041static int do_pages_stat(struct mm_struct *mm, unsigned long nr_pages,
1042 const void __user * __user *pages,
1043 int __user *status)
1044{
1045#define DO_PAGES_STAT_CHUNK_NR 16
1046 const void __user *chunk_pages[DO_PAGES_STAT_CHUNK_NR];
1047 int chunk_status[DO_PAGES_STAT_CHUNK_NR];
Brice Goglin80bba122008-12-09 13:14:23 -08001048
H. Peter Anvin87b8d1a2010-02-18 16:13:40 -08001049 while (nr_pages) {
1050 unsigned long chunk_nr;
Brice Goglin80bba122008-12-09 13:14:23 -08001051
H. Peter Anvin87b8d1a2010-02-18 16:13:40 -08001052 chunk_nr = nr_pages;
1053 if (chunk_nr > DO_PAGES_STAT_CHUNK_NR)
1054 chunk_nr = DO_PAGES_STAT_CHUNK_NR;
1055
1056 if (copy_from_user(chunk_pages, pages, chunk_nr * sizeof(*chunk_pages)))
1057 break;
Brice Goglin80bba122008-12-09 13:14:23 -08001058
1059 do_pages_stat_array(mm, chunk_nr, chunk_pages, chunk_status);
1060
H. Peter Anvin87b8d1a2010-02-18 16:13:40 -08001061 if (copy_to_user(status, chunk_status, chunk_nr * sizeof(*status)))
1062 break;
Christoph Lameter742755a2006-06-23 02:03:55 -07001063
H. Peter Anvin87b8d1a2010-02-18 16:13:40 -08001064 pages += chunk_nr;
1065 status += chunk_nr;
1066 nr_pages -= chunk_nr;
1067 }
1068 return nr_pages ? -EFAULT : 0;
Christoph Lameter742755a2006-06-23 02:03:55 -07001069}
1070
1071/*
1072 * Move a list of pages in the address space of the currently executing
1073 * process.
1074 */
Heiko Carstens938bb9f2009-01-14 14:14:30 +01001075SYSCALL_DEFINE6(move_pages, pid_t, pid, unsigned long, nr_pages,
1076 const void __user * __user *, pages,
1077 const int __user *, nodes,
1078 int __user *, status, int, flags)
Christoph Lameter742755a2006-06-23 02:03:55 -07001079{
David Howellsc69e8d92008-11-14 10:39:19 +11001080 const struct cred *cred = current_cred(), *tcred;
Christoph Lameter742755a2006-06-23 02:03:55 -07001081 struct task_struct *task;
Christoph Lameter742755a2006-06-23 02:03:55 -07001082 struct mm_struct *mm;
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001083 int err;
Christoph Lameter742755a2006-06-23 02:03:55 -07001084
1085 /* Check flags */
1086 if (flags & ~(MPOL_MF_MOVE|MPOL_MF_MOVE_ALL))
1087 return -EINVAL;
1088
1089 if ((flags & MPOL_MF_MOVE_ALL) && !capable(CAP_SYS_NICE))
1090 return -EPERM;
1091
1092 /* Find the mm_struct */
1093 read_lock(&tasklist_lock);
Pavel Emelyanov228ebcb2007-10-18 23:40:16 -07001094 task = pid ? find_task_by_vpid(pid) : current;
Christoph Lameter742755a2006-06-23 02:03:55 -07001095 if (!task) {
1096 read_unlock(&tasklist_lock);
1097 return -ESRCH;
1098 }
1099 mm = get_task_mm(task);
1100 read_unlock(&tasklist_lock);
1101
1102 if (!mm)
1103 return -EINVAL;
1104
1105 /*
1106 * Check if this process has the right to modify the specified
1107 * process. The right exists if the process has administrative
1108 * capabilities, superuser privileges or the same
1109 * userid as the target process.
1110 */
David Howellsc69e8d92008-11-14 10:39:19 +11001111 rcu_read_lock();
1112 tcred = __task_cred(task);
David Howellsb6dff3e2008-11-14 10:39:16 +11001113 if (cred->euid != tcred->suid && cred->euid != tcred->uid &&
1114 cred->uid != tcred->suid && cred->uid != tcred->uid &&
Christoph Lameter742755a2006-06-23 02:03:55 -07001115 !capable(CAP_SYS_NICE)) {
David Howellsc69e8d92008-11-14 10:39:19 +11001116 rcu_read_unlock();
Christoph Lameter742755a2006-06-23 02:03:55 -07001117 err = -EPERM;
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001118 goto out;
Christoph Lameter742755a2006-06-23 02:03:55 -07001119 }
David Howellsc69e8d92008-11-14 10:39:19 +11001120 rcu_read_unlock();
Christoph Lameter742755a2006-06-23 02:03:55 -07001121
David Quigley86c3a762006-06-23 02:04:02 -07001122 err = security_task_movememory(task);
1123 if (err)
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001124 goto out;
David Quigley86c3a762006-06-23 02:04:02 -07001125
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001126 if (nodes) {
1127 err = do_pages_move(mm, task, nr_pages, pages, nodes, status,
1128 flags);
1129 } else {
Brice Goglin2f007e72008-10-18 20:27:16 -07001130 err = do_pages_stat(mm, nr_pages, pages, status);
Brice Goglin2f007e72008-10-18 20:27:16 -07001131 }
David Quigley86c3a762006-06-23 02:04:02 -07001132
Christoph Lameter742755a2006-06-23 02:03:55 -07001133out:
Christoph Lameter742755a2006-06-23 02:03:55 -07001134 mmput(mm);
1135 return err;
1136}
Christoph Lameter742755a2006-06-23 02:03:55 -07001137
Christoph Lameter7b2259b2006-06-25 05:46:48 -07001138/*
1139 * Call migration functions in the vma_ops that may prepare
1140 * memory in a vm for migration. migration functions may perform
1141 * the migration for vmas that do not have an underlying page struct.
1142 */
1143int migrate_vmas(struct mm_struct *mm, const nodemask_t *to,
1144 const nodemask_t *from, unsigned long flags)
1145{
1146 struct vm_area_struct *vma;
1147 int err = 0;
1148
Daisuke Nishimura1001c9f2009-02-11 13:04:18 -08001149 for (vma = mm->mmap; vma && !err; vma = vma->vm_next) {
Christoph Lameter7b2259b2006-06-25 05:46:48 -07001150 if (vma->vm_ops && vma->vm_ops->migrate) {
1151 err = vma->vm_ops->migrate(vma, to, from, flags);
1152 if (err)
1153 break;
1154 }
1155 }
1156 return err;
1157}
Gerald Schaefer83d16742008-07-23 21:28:22 -07001158#endif