blob: efddbf0926b283ae5ef292e076ff9900842c4dd5 [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>
Christoph Lameterb20a3502006-03-22 00:09:12 -080035
36#include "internal.h"
37
Christoph Lameterb20a3502006-03-22 00:09:12 -080038#define lru_to_page(_head) (list_entry((_head)->prev, struct page, lru))
39
40/*
Christoph Lameter742755a2006-06-23 02:03:55 -070041 * migrate_prep() needs to be called before we start compiling a list of pages
42 * to be migrated using isolate_lru_page().
Christoph Lameterb20a3502006-03-22 00:09:12 -080043 */
44int migrate_prep(void)
45{
Christoph Lameterb20a3502006-03-22 00:09:12 -080046 /*
47 * Clear the LRU lists so pages can be isolated.
48 * Note that pages may be moved off the LRU after we have
49 * drained them. Those pages will fail to migrate like other
50 * pages that may be busy.
51 */
52 lru_add_drain_all();
53
54 return 0;
55}
56
Christoph Lameterb20a3502006-03-22 00:09:12 -080057/*
Lee Schermerhorn894bc312008-10-18 20:26:39 -070058 * Add isolated pages on the list back to the LRU under page lock
59 * to avoid leaking evictable pages back onto unevictable list.
Christoph Lameterb20a3502006-03-22 00:09:12 -080060 *
61 * returns the number of pages put back.
62 */
63int putback_lru_pages(struct list_head *l)
64{
65 struct page *page;
66 struct page *page2;
67 int count = 0;
68
69 list_for_each_entry_safe(page, page2, l, lru) {
Christoph Lametere24f0b82006-06-23 02:03:51 -070070 list_del(&page->lru);
KOSAKI Motohiroa7312862009-09-21 17:01:37 -070071 dec_zone_page_state(page, NR_ISOLATED_ANON +
Johannes Weiner6c0b1352009-09-21 17:02:59 -070072 page_is_file_cache(page));
Lee Schermerhorn894bc312008-10-18 20:26:39 -070073 putback_lru_page(page);
Christoph Lameterb20a3502006-03-22 00:09:12 -080074 count++;
75 }
76 return count;
77}
78
Christoph Lameter06972122006-06-23 02:03:35 -070079/*
80 * Restore a potential migration pte to a working pte entry
81 */
Hugh Dickinse9995ef2009-12-14 17:59:31 -080082static int remove_migration_pte(struct page *new, struct vm_area_struct *vma,
83 unsigned long addr, void *old)
Christoph Lameter06972122006-06-23 02:03:35 -070084{
85 struct mm_struct *mm = vma->vm_mm;
86 swp_entry_t entry;
87 pgd_t *pgd;
88 pud_t *pud;
89 pmd_t *pmd;
90 pte_t *ptep, pte;
91 spinlock_t *ptl;
92
93 pgd = pgd_offset(mm, addr);
94 if (!pgd_present(*pgd))
Hugh Dickinse9995ef2009-12-14 17:59:31 -080095 goto out;
Christoph Lameter06972122006-06-23 02:03:35 -070096
97 pud = pud_offset(pgd, addr);
98 if (!pud_present(*pud))
Hugh Dickinse9995ef2009-12-14 17:59:31 -080099 goto out;
Christoph Lameter06972122006-06-23 02:03:35 -0700100
101 pmd = pmd_offset(pud, addr);
102 if (!pmd_present(*pmd))
Hugh Dickinse9995ef2009-12-14 17:59:31 -0800103 goto out;
Christoph Lameter06972122006-06-23 02:03:35 -0700104
105 ptep = pte_offset_map(pmd, addr);
106
107 if (!is_swap_pte(*ptep)) {
108 pte_unmap(ptep);
Hugh Dickinse9995ef2009-12-14 17:59:31 -0800109 goto out;
Christoph Lameter06972122006-06-23 02:03:35 -0700110 }
111
112 ptl = pte_lockptr(mm, pmd);
113 spin_lock(ptl);
114 pte = *ptep;
115 if (!is_swap_pte(pte))
Hugh Dickinse9995ef2009-12-14 17:59:31 -0800116 goto unlock;
Christoph Lameter06972122006-06-23 02:03:35 -0700117
118 entry = pte_to_swp_entry(pte);
119
Hugh Dickinse9995ef2009-12-14 17:59:31 -0800120 if (!is_migration_entry(entry) ||
121 migration_entry_to_page(entry) != old)
122 goto unlock;
Christoph Lameter06972122006-06-23 02:03:35 -0700123
Christoph Lameter06972122006-06-23 02:03:35 -0700124 get_page(new);
125 pte = pte_mkold(mk_pte(new, vma->vm_page_prot));
126 if (is_write_migration_entry(entry))
127 pte = pte_mkwrite(pte);
KAMEZAWA Hiroyuki97ee0522007-10-16 01:25:43 -0700128 flush_cache_page(vma, addr, pte_pfn(pte));
Christoph Lameter06972122006-06-23 02:03:35 -0700129 set_pte_at(mm, addr, ptep, pte);
Christoph Lameter04e62a22006-06-23 02:03:38 -0700130
131 if (PageAnon(new))
132 page_add_anon_rmap(new, vma, addr);
133 else
134 page_add_file_rmap(new);
135
136 /* No need to invalidate - it was non-present before */
137 update_mmu_cache(vma, addr, pte);
Hugh Dickinse9995ef2009-12-14 17:59:31 -0800138unlock:
Christoph Lameter06972122006-06-23 02:03:35 -0700139 pte_unmap_unlock(ptep, ptl);
Hugh Dickinse9995ef2009-12-14 17:59:31 -0800140out:
141 return SWAP_AGAIN;
Christoph Lameter06972122006-06-23 02:03:35 -0700142}
143
144/*
Christoph Lameter04e62a22006-06-23 02:03:38 -0700145 * Get rid of all migration entries and replace them by
146 * references to the indicated page.
147 */
148static void remove_migration_ptes(struct page *old, struct page *new)
149{
Hugh Dickinse9995ef2009-12-14 17:59:31 -0800150 rmap_walk(new, remove_migration_pte, old);
Christoph Lameter04e62a22006-06-23 02:03:38 -0700151}
152
153/*
Christoph Lameter06972122006-06-23 02:03:35 -0700154 * Something used the pte of a page under migration. We need to
155 * get to the page and wait until migration is finished.
156 * When we return from this function the fault will be retried.
157 *
158 * This function is called from do_swap_page().
159 */
160void migration_entry_wait(struct mm_struct *mm, pmd_t *pmd,
161 unsigned long address)
162{
163 pte_t *ptep, pte;
164 spinlock_t *ptl;
165 swp_entry_t entry;
166 struct page *page;
167
168 ptep = pte_offset_map_lock(mm, pmd, address, &ptl);
169 pte = *ptep;
170 if (!is_swap_pte(pte))
171 goto out;
172
173 entry = pte_to_swp_entry(pte);
174 if (!is_migration_entry(entry))
175 goto out;
176
177 page = migration_entry_to_page(entry);
178
Nick Piggine2867812008-07-25 19:45:30 -0700179 /*
180 * Once radix-tree replacement of page migration started, page_count
181 * *must* be zero. And, we don't want to call wait_on_page_locked()
182 * against a page without get_page().
183 * So, we use get_page_unless_zero(), here. Even failed, page fault
184 * will occur again.
185 */
186 if (!get_page_unless_zero(page))
187 goto out;
Christoph Lameter06972122006-06-23 02:03:35 -0700188 pte_unmap_unlock(ptep, ptl);
189 wait_on_page_locked(page);
190 put_page(page);
191 return;
192out:
193 pte_unmap_unlock(ptep, ptl);
194}
195
Christoph Lameterb20a3502006-03-22 00:09:12 -0800196/*
Christoph Lameterc3fcf8a2006-06-23 02:03:32 -0700197 * Replace the page in the mapping.
Christoph Lameter5b5c7122006-06-23 02:03:29 -0700198 *
199 * The number of remaining references must be:
200 * 1 for anonymous pages without a mapping
201 * 2 for pages with a mapping
David Howells266cf652009-04-03 16:42:36 +0100202 * 3 for pages with a mapping and PagePrivate/PagePrivate2 set.
Christoph Lameterb20a3502006-03-22 00:09:12 -0800203 */
Christoph Lameter2d1db3b2006-06-23 02:03:33 -0700204static int migrate_page_move_mapping(struct address_space *mapping,
205 struct page *newpage, struct page *page)
Christoph Lameterb20a3502006-03-22 00:09:12 -0800206{
Nick Piggine2867812008-07-25 19:45:30 -0700207 int expected_count;
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800208 void **pslot;
Christoph Lameterb20a3502006-03-22 00:09:12 -0800209
Christoph Lameter6c5240a2006-06-23 02:03:37 -0700210 if (!mapping) {
Christoph Lameter0e8c7d02007-04-23 14:41:09 -0700211 /* Anonymous page without mapping */
Christoph Lameter6c5240a2006-06-23 02:03:37 -0700212 if (page_count(page) != 1)
213 return -EAGAIN;
214 return 0;
215 }
216
Nick Piggin19fd6232008-07-25 19:45:32 -0700217 spin_lock_irq(&mapping->tree_lock);
Christoph Lameterb20a3502006-03-22 00:09:12 -0800218
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800219 pslot = radix_tree_lookup_slot(&mapping->page_tree,
220 page_index(page));
Christoph Lameterb20a3502006-03-22 00:09:12 -0800221
Johannes Weineredcf4742009-09-21 17:02:59 -0700222 expected_count = 2 + page_has_private(page);
Nick Piggine2867812008-07-25 19:45:30 -0700223 if (page_count(page) != expected_count ||
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800224 (struct page *)radix_tree_deref_slot(pslot) != page) {
Nick Piggin19fd6232008-07-25 19:45:32 -0700225 spin_unlock_irq(&mapping->tree_lock);
Christoph Lametere23ca002006-04-10 22:52:57 -0700226 return -EAGAIN;
Christoph Lameterb20a3502006-03-22 00:09:12 -0800227 }
228
Nick Piggine2867812008-07-25 19:45:30 -0700229 if (!page_freeze_refs(page, expected_count)) {
Nick Piggin19fd6232008-07-25 19:45:32 -0700230 spin_unlock_irq(&mapping->tree_lock);
Nick Piggine2867812008-07-25 19:45:30 -0700231 return -EAGAIN;
232 }
233
Christoph Lameterb20a3502006-03-22 00:09:12 -0800234 /*
235 * Now we know that no one else is looking at the page.
Christoph Lameterb20a3502006-03-22 00:09:12 -0800236 */
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800237 get_page(newpage); /* add cache reference */
Christoph Lameterb20a3502006-03-22 00:09:12 -0800238 if (PageSwapCache(page)) {
239 SetPageSwapCache(newpage);
240 set_page_private(newpage, page_private(page));
241 }
242
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800243 radix_tree_replace_slot(pslot, newpage);
244
Nick Piggine2867812008-07-25 19:45:30 -0700245 page_unfreeze_refs(page, expected_count);
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800246 /*
247 * Drop cache reference from old page.
248 * We know this isn't the last reference.
249 */
Christoph Lameterb20a3502006-03-22 00:09:12 -0800250 __put_page(page);
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800251
Christoph Lameter0e8c7d02007-04-23 14:41:09 -0700252 /*
253 * If moved to a different zone then also account
254 * the page for that zone. Other VM counters will be
255 * taken care of when we establish references to the
256 * new page and drop references to the old page.
257 *
258 * Note that anonymous pages are accounted for
259 * via NR_FILE_PAGES and NR_ANON_PAGES if they
260 * are mapped to swap space.
261 */
262 __dec_zone_page_state(page, NR_FILE_PAGES);
263 __inc_zone_page_state(newpage, NR_FILE_PAGES);
KOSAKI Motohiro4b021082009-09-21 17:01:33 -0700264 if (PageSwapBacked(page)) {
265 __dec_zone_page_state(page, NR_SHMEM);
266 __inc_zone_page_state(newpage, NR_SHMEM);
267 }
Nick Piggin19fd6232008-07-25 19:45:32 -0700268 spin_unlock_irq(&mapping->tree_lock);
Christoph Lameterb20a3502006-03-22 00:09:12 -0800269
270 return 0;
271}
Christoph Lameterb20a3502006-03-22 00:09:12 -0800272
273/*
274 * Copy the page to its new location
275 */
Christoph Lametere7340f72006-06-23 02:03:29 -0700276static void migrate_page_copy(struct page *newpage, struct page *page)
Christoph Lameterb20a3502006-03-22 00:09:12 -0800277{
KAMEZAWA Hiroyukib7abea92008-10-18 20:28:09 -0700278 int anon;
279
Christoph Lameterb20a3502006-03-22 00:09:12 -0800280 copy_highpage(newpage, page);
281
282 if (PageError(page))
283 SetPageError(newpage);
284 if (PageReferenced(page))
285 SetPageReferenced(newpage);
286 if (PageUptodate(page))
287 SetPageUptodate(newpage);
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700288 if (TestClearPageActive(page)) {
289 VM_BUG_ON(PageUnevictable(page));
Christoph Lameterb20a3502006-03-22 00:09:12 -0800290 SetPageActive(newpage);
Lee Schermerhorn418b27e2009-12-14 17:59:54 -0800291 } else if (TestClearPageUnevictable(page))
292 SetPageUnevictable(newpage);
Christoph Lameterb20a3502006-03-22 00:09:12 -0800293 if (PageChecked(page))
294 SetPageChecked(newpage);
295 if (PageMappedToDisk(page))
296 SetPageMappedToDisk(newpage);
297
298 if (PageDirty(page)) {
299 clear_page_dirty_for_io(page);
Nick Piggin3a902c52008-04-30 00:55:16 -0700300 /*
301 * Want to mark the page and the radix tree as dirty, and
302 * redo the accounting that clear_page_dirty_for_io undid,
303 * but we can't use set_page_dirty because that function
304 * is actually a signal that all of the page has become dirty.
305 * Wheras only part of our page may be dirty.
306 */
307 __set_page_dirty_nobuffers(newpage);
Christoph Lameterb20a3502006-03-22 00:09:12 -0800308 }
309
Nick Pigginb291f002008-10-18 20:26:44 -0700310 mlock_migrate_page(newpage, page);
Hugh Dickinse9995ef2009-12-14 17:59:31 -0800311 ksm_migrate_page(newpage, page);
Nick Pigginb291f002008-10-18 20:26:44 -0700312
Christoph Lameterb20a3502006-03-22 00:09:12 -0800313 ClearPageSwapCache(page);
Christoph Lameterb20a3502006-03-22 00:09:12 -0800314 ClearPagePrivate(page);
315 set_page_private(page, 0);
KAMEZAWA Hiroyukib7abea92008-10-18 20:28:09 -0700316 /* page->mapping contains a flag for PageAnon() */
317 anon = PageAnon(page);
Christoph Lameterb20a3502006-03-22 00:09:12 -0800318 page->mapping = NULL;
319
320 /*
321 * If any waiters have accumulated on the new page then
322 * wake them up.
323 */
324 if (PageWriteback(newpage))
325 end_page_writeback(newpage);
326}
Christoph Lameterb20a3502006-03-22 00:09:12 -0800327
Christoph Lameter1d8b85c2006-06-23 02:03:28 -0700328/************************************************************
329 * Migration functions
330 ***********************************************************/
331
332/* Always fail migration. Used for mappings that are not movable */
Christoph Lameter2d1db3b2006-06-23 02:03:33 -0700333int fail_migrate_page(struct address_space *mapping,
334 struct page *newpage, struct page *page)
Christoph Lameter1d8b85c2006-06-23 02:03:28 -0700335{
336 return -EIO;
337}
338EXPORT_SYMBOL(fail_migrate_page);
339
Christoph Lameterb20a3502006-03-22 00:09:12 -0800340/*
341 * Common logic to directly migrate a single page suitable for
David Howells266cf652009-04-03 16:42:36 +0100342 * pages that do not use PagePrivate/PagePrivate2.
Christoph Lameterb20a3502006-03-22 00:09:12 -0800343 *
344 * Pages are locked upon entry and exit.
345 */
Christoph Lameter2d1db3b2006-06-23 02:03:33 -0700346int migrate_page(struct address_space *mapping,
347 struct page *newpage, struct page *page)
Christoph Lameterb20a3502006-03-22 00:09:12 -0800348{
349 int rc;
350
351 BUG_ON(PageWriteback(page)); /* Writeback must be complete */
352
Christoph Lameter2d1db3b2006-06-23 02:03:33 -0700353 rc = migrate_page_move_mapping(mapping, newpage, page);
Christoph Lameterb20a3502006-03-22 00:09:12 -0800354
355 if (rc)
356 return rc;
357
358 migrate_page_copy(newpage, page);
Christoph Lameterb20a3502006-03-22 00:09:12 -0800359 return 0;
360}
361EXPORT_SYMBOL(migrate_page);
362
David Howells93614012006-09-30 20:45:40 +0200363#ifdef CONFIG_BLOCK
Christoph Lameterb20a3502006-03-22 00:09:12 -0800364/*
Christoph Lameter1d8b85c2006-06-23 02:03:28 -0700365 * Migration function for pages with buffers. This function can only be used
366 * if the underlying filesystem guarantees that no other references to "page"
367 * exist.
368 */
Christoph Lameter2d1db3b2006-06-23 02:03:33 -0700369int buffer_migrate_page(struct address_space *mapping,
370 struct page *newpage, struct page *page)
Christoph Lameter1d8b85c2006-06-23 02:03:28 -0700371{
Christoph Lameter1d8b85c2006-06-23 02:03:28 -0700372 struct buffer_head *bh, *head;
373 int rc;
374
Christoph Lameter1d8b85c2006-06-23 02:03:28 -0700375 if (!page_has_buffers(page))
Christoph Lameter2d1db3b2006-06-23 02:03:33 -0700376 return migrate_page(mapping, newpage, page);
Christoph Lameter1d8b85c2006-06-23 02:03:28 -0700377
378 head = page_buffers(page);
379
Christoph Lameter2d1db3b2006-06-23 02:03:33 -0700380 rc = migrate_page_move_mapping(mapping, newpage, page);
Christoph Lameter1d8b85c2006-06-23 02:03:28 -0700381
382 if (rc)
383 return rc;
384
385 bh = head;
386 do {
387 get_bh(bh);
388 lock_buffer(bh);
389 bh = bh->b_this_page;
390
391 } while (bh != head);
392
393 ClearPagePrivate(page);
394 set_page_private(newpage, page_private(page));
395 set_page_private(page, 0);
396 put_page(page);
397 get_page(newpage);
398
399 bh = head;
400 do {
401 set_bh_page(bh, newpage, bh_offset(bh));
402 bh = bh->b_this_page;
403
404 } while (bh != head);
405
406 SetPagePrivate(newpage);
407
408 migrate_page_copy(newpage, page);
409
410 bh = head;
411 do {
412 unlock_buffer(bh);
413 put_bh(bh);
414 bh = bh->b_this_page;
415
416 } while (bh != head);
417
418 return 0;
419}
420EXPORT_SYMBOL(buffer_migrate_page);
David Howells93614012006-09-30 20:45:40 +0200421#endif
Christoph Lameter1d8b85c2006-06-23 02:03:28 -0700422
Christoph Lameter04e62a22006-06-23 02:03:38 -0700423/*
424 * Writeback a page to clean the dirty state
425 */
426static int writeout(struct address_space *mapping, struct page *page)
427{
428 struct writeback_control wbc = {
429 .sync_mode = WB_SYNC_NONE,
430 .nr_to_write = 1,
431 .range_start = 0,
432 .range_end = LLONG_MAX,
433 .nonblocking = 1,
434 .for_reclaim = 1
435 };
436 int rc;
437
438 if (!mapping->a_ops->writepage)
439 /* No write method for the address space */
440 return -EINVAL;
441
442 if (!clear_page_dirty_for_io(page))
443 /* Someone else already triggered a write */
444 return -EAGAIN;
445
446 /*
447 * A dirty page may imply that the underlying filesystem has
448 * the page on some queue. So the page must be clean for
449 * migration. Writeout may mean we loose the lock and the
450 * page state is no longer what we checked for earlier.
451 * At this point we know that the migration attempt cannot
452 * be successful.
453 */
454 remove_migration_ptes(page, page);
455
456 rc = mapping->a_ops->writepage(page, &wbc);
Christoph Lameter04e62a22006-06-23 02:03:38 -0700457
458 if (rc != AOP_WRITEPAGE_ACTIVATE)
459 /* unlocked. Relock */
460 lock_page(page);
461
Hugh Dickinsbda85502008-11-19 15:36:36 -0800462 return (rc < 0) ? -EIO : -EAGAIN;
Christoph Lameter04e62a22006-06-23 02:03:38 -0700463}
464
465/*
466 * Default handling if a filesystem does not provide a migration function.
467 */
Christoph Lameter8351a6e2006-06-23 02:03:33 -0700468static int fallback_migrate_page(struct address_space *mapping,
469 struct page *newpage, struct page *page)
470{
Christoph Lameter04e62a22006-06-23 02:03:38 -0700471 if (PageDirty(page))
472 return writeout(mapping, page);
Christoph Lameter8351a6e2006-06-23 02:03:33 -0700473
474 /*
475 * Buffers may be managed in a filesystem specific way.
476 * We must have no buffers or drop them.
477 */
David Howells266cf652009-04-03 16:42:36 +0100478 if (page_has_private(page) &&
Christoph Lameter8351a6e2006-06-23 02:03:33 -0700479 !try_to_release_page(page, GFP_KERNEL))
480 return -EAGAIN;
481
482 return migrate_page(mapping, newpage, page);
483}
484
Christoph Lameter1d8b85c2006-06-23 02:03:28 -0700485/*
Christoph Lametere24f0b82006-06-23 02:03:51 -0700486 * Move a page to a newly allocated page
487 * The page is locked and all ptes have been successfully removed.
488 *
489 * The new page will have replaced the old page if this function
490 * is successful.
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700491 *
492 * Return value:
493 * < 0 - error code
494 * == 0 - success
Christoph Lametere24f0b82006-06-23 02:03:51 -0700495 */
496static int move_to_new_page(struct page *newpage, struct page *page)
497{
498 struct address_space *mapping;
499 int rc;
500
501 /*
502 * Block others from accessing the page when we get around to
503 * establishing additional references. We are the only one
504 * holding a reference to the new page at this point.
505 */
Nick Piggin529ae9a2008-08-02 12:01:03 +0200506 if (!trylock_page(newpage))
Christoph Lametere24f0b82006-06-23 02:03:51 -0700507 BUG();
508
509 /* Prepare mapping for the new page.*/
510 newpage->index = page->index;
511 newpage->mapping = page->mapping;
Rik van Rielb2e18532008-10-18 20:26:30 -0700512 if (PageSwapBacked(page))
513 SetPageSwapBacked(newpage);
Christoph Lametere24f0b82006-06-23 02:03:51 -0700514
515 mapping = page_mapping(page);
516 if (!mapping)
517 rc = migrate_page(mapping, newpage, page);
518 else if (mapping->a_ops->migratepage)
519 /*
520 * Most pages have a mapping and most filesystems
521 * should provide a migration function. Anonymous
522 * pages are part of swap space which also has its
523 * own migration function. This is the most common
524 * path for page migration.
525 */
526 rc = mapping->a_ops->migratepage(mapping,
527 newpage, page);
528 else
529 rc = fallback_migrate_page(mapping, newpage, page);
530
Hugh Dickinse9995ef2009-12-14 17:59:31 -0800531 if (!rc)
Christoph Lametere24f0b82006-06-23 02:03:51 -0700532 remove_migration_ptes(page, newpage);
Hugh Dickinse9995ef2009-12-14 17:59:31 -0800533 else
Christoph Lametere24f0b82006-06-23 02:03:51 -0700534 newpage->mapping = NULL;
535
536 unlock_page(newpage);
537
538 return rc;
539}
540
541/*
542 * Obtain the lock on page, remove all ptes and migrate the page
543 * to the newly allocated page in newpage.
544 */
Christoph Lameter95a402c2006-06-23 02:03:53 -0700545static int unmap_and_move(new_page_t get_new_page, unsigned long private,
Hugh Dickins62b61f62009-12-14 17:59:33 -0800546 struct page *page, int force, int offlining)
Christoph Lametere24f0b82006-06-23 02:03:51 -0700547{
548 int rc = 0;
Christoph Lameter742755a2006-06-23 02:03:55 -0700549 int *result = NULL;
550 struct page *newpage = get_new_page(page, private, &result);
KAMEZAWA Hiroyuki989f89c2007-08-30 23:56:21 -0700551 int rcu_locked = 0;
KAMEZAWA Hiroyukiae41be32008-02-07 00:14:10 -0800552 int charge = 0;
KAMEZAWA Hiroyukie00e4312009-11-11 14:26:26 -0800553 struct mem_cgroup *mem = NULL;
Christoph Lameter95a402c2006-06-23 02:03:53 -0700554
555 if (!newpage)
556 return -ENOMEM;
Christoph Lametere24f0b82006-06-23 02:03:51 -0700557
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700558 if (page_count(page) == 1) {
Christoph Lametere24f0b82006-06-23 02:03:51 -0700559 /* page was freed from under us. So we are done. */
Christoph Lameter95a402c2006-06-23 02:03:53 -0700560 goto move_newpage;
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700561 }
Christoph Lametere24f0b82006-06-23 02:03:51 -0700562
KAMEZAWA Hiroyukie8589cc2008-07-25 01:47:10 -0700563 /* prepare cgroup just returns 0 or -ENOMEM */
Christoph Lametere24f0b82006-06-23 02:03:51 -0700564 rc = -EAGAIN;
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -0800565
Nick Piggin529ae9a2008-08-02 12:01:03 +0200566 if (!trylock_page(page)) {
Christoph Lametere24f0b82006-06-23 02:03:51 -0700567 if (!force)
Christoph Lameter95a402c2006-06-23 02:03:53 -0700568 goto move_newpage;
Christoph Lametere24f0b82006-06-23 02:03:51 -0700569 lock_page(page);
570 }
571
Hugh Dickins62b61f62009-12-14 17:59:33 -0800572 /*
573 * Only memory hotplug's offline_pages() caller has locked out KSM,
574 * and can safely migrate a KSM page. The other cases have skipped
575 * PageKsm along with PageReserved - but it is only now when we have
576 * the page lock that we can be certain it will not go KSM beneath us
577 * (KSM will not upgrade a page from PageAnon to PageKsm when it sees
578 * its pagecount raised, but only here do we take the page lock which
579 * serializes that).
580 */
581 if (PageKsm(page) && !offlining) {
582 rc = -EBUSY;
583 goto unlock;
584 }
585
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -0800586 /* charge against new page */
587 charge = mem_cgroup_prepare_migration(page, &mem);
588 if (charge == -ENOMEM) {
589 rc = -ENOMEM;
590 goto unlock;
591 }
592 BUG_ON(charge);
593
Christoph Lametere24f0b82006-06-23 02:03:51 -0700594 if (PageWriteback(page)) {
595 if (!force)
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -0800596 goto uncharge;
Christoph Lametere24f0b82006-06-23 02:03:51 -0700597 wait_on_page_writeback(page);
598 }
Christoph Lametere24f0b82006-06-23 02:03:51 -0700599 /*
KAMEZAWA Hiroyukidc386d42007-07-26 10:41:07 -0700600 * By try_to_unmap(), page->mapcount goes down to 0 here. In this case,
601 * we cannot notice that anon_vma is freed while we migrates a page.
602 * This rcu_read_lock() delays freeing anon_vma pointer until the end
603 * of migration. File cache pages are no problem because of page_lock()
KAMEZAWA Hiroyuki989f89c2007-08-30 23:56:21 -0700604 * File Caches may use write_page() or lock_page() in migration, then,
605 * just care Anon page here.
Christoph Lametere24f0b82006-06-23 02:03:51 -0700606 */
KAMEZAWA Hiroyuki989f89c2007-08-30 23:56:21 -0700607 if (PageAnon(page)) {
608 rcu_read_lock();
609 rcu_locked = 1;
610 }
Shaohua Li62e1c552008-02-04 22:29:33 -0800611
KAMEZAWA Hiroyukidc386d42007-07-26 10:41:07 -0700612 /*
Shaohua Li62e1c552008-02-04 22:29:33 -0800613 * Corner case handling:
614 * 1. When a new swap-cache page is read into, it is added to the LRU
615 * and treated as swapcache but it has no rmap yet.
616 * Calling try_to_unmap() against a page->mapping==NULL page will
617 * trigger a BUG. So handle it here.
618 * 2. An orphaned page (see truncate_complete_page) might have
619 * fs-private metadata. The page can be picked up due to memory
620 * offlining. Everywhere else except page reclaim, the page is
621 * invisible to the vm, so the page can not be migrated. So try to
622 * free the metadata, so the page can be freed.
KAMEZAWA Hiroyukidc386d42007-07-26 10:41:07 -0700623 */
Shaohua Li62e1c552008-02-04 22:29:33 -0800624 if (!page->mapping) {
David Howells266cf652009-04-03 16:42:36 +0100625 if (!PageAnon(page) && page_has_private(page)) {
Shaohua Li62e1c552008-02-04 22:29:33 -0800626 /*
627 * Go direct to try_to_free_buffers() here because
628 * a) that's what try_to_release_page() would do anyway
629 * b) we may be under rcu_read_lock() here, so we can't
630 * use GFP_KERNEL which is what try_to_release_page()
631 * needs to be effective.
632 */
633 try_to_free_buffers(page);
Shaohua Liabfc3482009-09-21 17:01:19 -0700634 goto rcu_unlock;
Shaohua Li62e1c552008-02-04 22:29:33 -0800635 }
Shaohua Liabfc3482009-09-21 17:01:19 -0700636 goto skip_unmap;
Shaohua Li62e1c552008-02-04 22:29:33 -0800637 }
638
KAMEZAWA Hiroyukidc386d42007-07-26 10:41:07 -0700639 /* Establish migration ptes or remove ptes */
Andi Kleen14fa31b2009-09-16 11:50:10 +0200640 try_to_unmap(page, TTU_MIGRATION|TTU_IGNORE_MLOCK|TTU_IGNORE_ACCESS);
KAMEZAWA Hiroyukidc386d42007-07-26 10:41:07 -0700641
Shaohua Liabfc3482009-09-21 17:01:19 -0700642skip_unmap:
Christoph Lametere6a15302006-06-25 05:46:49 -0700643 if (!page_mapped(page))
644 rc = move_to_new_page(newpage, page);
Christoph Lametere24f0b82006-06-23 02:03:51 -0700645
KAMEZAWA Hiroyukie8589cc2008-07-25 01:47:10 -0700646 if (rc)
Christoph Lametere24f0b82006-06-23 02:03:51 -0700647 remove_migration_ptes(page, page);
KAMEZAWA Hiroyukidc386d42007-07-26 10:41:07 -0700648rcu_unlock:
KAMEZAWA Hiroyuki989f89c2007-08-30 23:56:21 -0700649 if (rcu_locked)
650 rcu_read_unlock();
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -0800651uncharge:
652 if (!charge)
653 mem_cgroup_end_migration(mem, page, newpage);
Christoph Lametere24f0b82006-06-23 02:03:51 -0700654unlock:
655 unlock_page(page);
Christoph Lameter95a402c2006-06-23 02:03:53 -0700656
Christoph Lametere24f0b82006-06-23 02:03:51 -0700657 if (rc != -EAGAIN) {
Christoph Lameteraaa994b2006-06-23 02:03:52 -0700658 /*
659 * A page that has been migrated has all references
660 * removed and will be freed. A page that has not been
661 * migrated will have kepts its references and be
662 * restored.
663 */
664 list_del(&page->lru);
KOSAKI Motohiroa7312862009-09-21 17:01:37 -0700665 dec_zone_page_state(page, NR_ISOLATED_ANON +
Johannes Weiner6c0b1352009-09-21 17:02:59 -0700666 page_is_file_cache(page));
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700667 putback_lru_page(page);
Christoph Lametere24f0b82006-06-23 02:03:51 -0700668 }
Christoph Lameter95a402c2006-06-23 02:03:53 -0700669
670move_newpage:
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700671
Christoph Lameter95a402c2006-06-23 02:03:53 -0700672 /*
673 * Move the new page to the LRU. If migration was not successful
674 * then this will free the page.
675 */
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700676 putback_lru_page(newpage);
677
Christoph Lameter742755a2006-06-23 02:03:55 -0700678 if (result) {
679 if (rc)
680 *result = rc;
681 else
682 *result = page_to_nid(newpage);
683 }
Christoph Lametere24f0b82006-06-23 02:03:51 -0700684 return rc;
685}
686
687/*
Christoph Lameterb20a3502006-03-22 00:09:12 -0800688 * migrate_pages
689 *
Christoph Lameter95a402c2006-06-23 02:03:53 -0700690 * The function takes one list of pages to migrate and a function
691 * that determines from the page to be migrated and the private data
692 * the target of the move and allocates the page.
Christoph Lameterb20a3502006-03-22 00:09:12 -0800693 *
694 * The function returns after 10 attempts or if no pages
695 * are movable anymore because to has become empty
Christoph Lameteraaa994b2006-06-23 02:03:52 -0700696 * or no retryable pages exist anymore. All pages will be
Gabriel Craciunescue9534b32007-10-20 02:13:26 +0200697 * returned to the LRU or freed.
Christoph Lameterb20a3502006-03-22 00:09:12 -0800698 *
Christoph Lameter95a402c2006-06-23 02:03:53 -0700699 * Return: Number of pages not migrated or error code.
Christoph Lameterb20a3502006-03-22 00:09:12 -0800700 */
Christoph Lameter95a402c2006-06-23 02:03:53 -0700701int migrate_pages(struct list_head *from,
Hugh Dickins62b61f62009-12-14 17:59:33 -0800702 new_page_t get_new_page, unsigned long private, int offlining)
Christoph Lameterb20a3502006-03-22 00:09:12 -0800703{
Christoph Lametere24f0b82006-06-23 02:03:51 -0700704 int retry = 1;
Christoph Lameterb20a3502006-03-22 00:09:12 -0800705 int nr_failed = 0;
706 int pass = 0;
707 struct page *page;
708 struct page *page2;
709 int swapwrite = current->flags & PF_SWAPWRITE;
710 int rc;
711
712 if (!swapwrite)
713 current->flags |= PF_SWAPWRITE;
714
Christoph Lametere24f0b82006-06-23 02:03:51 -0700715 for(pass = 0; pass < 10 && retry; pass++) {
716 retry = 0;
Christoph Lameterb20a3502006-03-22 00:09:12 -0800717
Christoph Lametere24f0b82006-06-23 02:03:51 -0700718 list_for_each_entry_safe(page, page2, from, lru) {
Christoph Lametere24f0b82006-06-23 02:03:51 -0700719 cond_resched();
Christoph Lameterb20a3502006-03-22 00:09:12 -0800720
Christoph Lameter95a402c2006-06-23 02:03:53 -0700721 rc = unmap_and_move(get_new_page, private,
Hugh Dickins62b61f62009-12-14 17:59:33 -0800722 page, pass > 2, offlining);
Christoph Lameterb20a3502006-03-22 00:09:12 -0800723
Christoph Lametere24f0b82006-06-23 02:03:51 -0700724 switch(rc) {
Christoph Lameter95a402c2006-06-23 02:03:53 -0700725 case -ENOMEM:
726 goto out;
Christoph Lametere24f0b82006-06-23 02:03:51 -0700727 case -EAGAIN:
Christoph Lameter2d1db3b2006-06-23 02:03:33 -0700728 retry++;
Christoph Lametere24f0b82006-06-23 02:03:51 -0700729 break;
730 case 0:
Christoph Lametere24f0b82006-06-23 02:03:51 -0700731 break;
732 default:
Christoph Lameter2d1db3b2006-06-23 02:03:33 -0700733 /* Permanent failure */
Christoph Lameter2d1db3b2006-06-23 02:03:33 -0700734 nr_failed++;
Christoph Lametere24f0b82006-06-23 02:03:51 -0700735 break;
Christoph Lameter2d1db3b2006-06-23 02:03:33 -0700736 }
Christoph Lameterb20a3502006-03-22 00:09:12 -0800737 }
738 }
Christoph Lameter95a402c2006-06-23 02:03:53 -0700739 rc = 0;
740out:
Christoph Lameterb20a3502006-03-22 00:09:12 -0800741 if (!swapwrite)
742 current->flags &= ~PF_SWAPWRITE;
743
Christoph Lameteraaa994b2006-06-23 02:03:52 -0700744 putback_lru_pages(from);
Christoph Lameter95a402c2006-06-23 02:03:53 -0700745
746 if (rc)
747 return rc;
748
Christoph Lameterb20a3502006-03-22 00:09:12 -0800749 return nr_failed + retry;
750}
751
Christoph Lameter742755a2006-06-23 02:03:55 -0700752#ifdef CONFIG_NUMA
753/*
754 * Move a list of individual pages
755 */
756struct page_to_node {
757 unsigned long addr;
758 struct page *page;
759 int node;
760 int status;
761};
762
763static struct page *new_page_node(struct page *p, unsigned long private,
764 int **result)
765{
766 struct page_to_node *pm = (struct page_to_node *)private;
767
768 while (pm->node != MAX_NUMNODES && pm->page != p)
769 pm++;
770
771 if (pm->node == MAX_NUMNODES)
772 return NULL;
773
774 *result = &pm->status;
775
Mel Gorman6484eb32009-06-16 15:31:54 -0700776 return alloc_pages_exact_node(pm->node,
Mel Gorman769848c2007-07-17 04:03:05 -0700777 GFP_HIGHUSER_MOVABLE | GFP_THISNODE, 0);
Christoph Lameter742755a2006-06-23 02:03:55 -0700778}
779
780/*
781 * Move a set of pages as indicated in the pm array. The addr
782 * field must be set to the virtual address of the page to be moved
783 * and the node number must contain a valid target node.
Brice Goglin5e9a0f02008-10-18 20:27:17 -0700784 * The pm array ends with node = MAX_NUMNODES.
Christoph Lameter742755a2006-06-23 02:03:55 -0700785 */
Brice Goglin5e9a0f02008-10-18 20:27:17 -0700786static int do_move_page_to_node_array(struct mm_struct *mm,
787 struct page_to_node *pm,
788 int migrate_all)
Christoph Lameter742755a2006-06-23 02:03:55 -0700789{
790 int err;
791 struct page_to_node *pp;
792 LIST_HEAD(pagelist);
793
794 down_read(&mm->mmap_sem);
795
796 /*
797 * Build a list of pages to migrate
798 */
Christoph Lameter742755a2006-06-23 02:03:55 -0700799 for (pp = pm; pp->node != MAX_NUMNODES; pp++) {
800 struct vm_area_struct *vma;
801 struct page *page;
802
Christoph Lameter742755a2006-06-23 02:03:55 -0700803 err = -EFAULT;
804 vma = find_vma(mm, pp->addr);
Christoph Lameter0dc952d2007-03-05 00:30:33 -0800805 if (!vma || !vma_migratable(vma))
Christoph Lameter742755a2006-06-23 02:03:55 -0700806 goto set_status;
807
808 page = follow_page(vma, pp->addr, FOLL_GET);
Linus Torvalds89f5b7d2008-06-20 11:18:25 -0700809
810 err = PTR_ERR(page);
811 if (IS_ERR(page))
812 goto set_status;
813
Christoph Lameter742755a2006-06-23 02:03:55 -0700814 err = -ENOENT;
815 if (!page)
816 goto set_status;
817
Hugh Dickins62b61f62009-12-14 17:59:33 -0800818 /* Use PageReserved to check for zero page */
819 if (PageReserved(page) || PageKsm(page))
Christoph Lameter742755a2006-06-23 02:03:55 -0700820 goto put_and_set;
821
822 pp->page = page;
823 err = page_to_nid(page);
824
825 if (err == pp->node)
826 /*
827 * Node already in the right place
828 */
829 goto put_and_set;
830
831 err = -EACCES;
832 if (page_mapcount(page) > 1 &&
833 !migrate_all)
834 goto put_and_set;
835
Nick Piggin62695a82008-10-18 20:26:09 -0700836 err = isolate_lru_page(page);
KOSAKI Motohiro6d9c2852009-12-14 17:58:11 -0800837 if (!err) {
Nick Piggin62695a82008-10-18 20:26:09 -0700838 list_add_tail(&page->lru, &pagelist);
KOSAKI Motohiro6d9c2852009-12-14 17:58:11 -0800839 inc_zone_page_state(page, NR_ISOLATED_ANON +
840 page_is_file_cache(page));
841 }
Christoph Lameter742755a2006-06-23 02:03:55 -0700842put_and_set:
843 /*
844 * Either remove the duplicate refcount from
845 * isolate_lru_page() or drop the page ref if it was
846 * not isolated.
847 */
848 put_page(page);
849set_status:
850 pp->status = err;
851 }
852
Brice Gogline78bbfa2008-10-18 20:27:15 -0700853 err = 0;
Christoph Lameter742755a2006-06-23 02:03:55 -0700854 if (!list_empty(&pagelist))
855 err = migrate_pages(&pagelist, new_page_node,
Hugh Dickins62b61f62009-12-14 17:59:33 -0800856 (unsigned long)pm, 0);
Christoph Lameter742755a2006-06-23 02:03:55 -0700857
858 up_read(&mm->mmap_sem);
859 return err;
860}
861
862/*
Brice Goglin5e9a0f02008-10-18 20:27:17 -0700863 * Migrate an array of page address onto an array of nodes and fill
864 * the corresponding array of status.
865 */
866static int do_pages_move(struct mm_struct *mm, struct task_struct *task,
867 unsigned long nr_pages,
868 const void __user * __user *pages,
869 const int __user *nodes,
870 int __user *status, int flags)
871{
Brice Goglin3140a222009-01-06 14:38:57 -0800872 struct page_to_node *pm;
Brice Goglin5e9a0f02008-10-18 20:27:17 -0700873 nodemask_t task_nodes;
Brice Goglin3140a222009-01-06 14:38:57 -0800874 unsigned long chunk_nr_pages;
875 unsigned long chunk_start;
876 int err;
Brice Goglin5e9a0f02008-10-18 20:27:17 -0700877
878 task_nodes = cpuset_mems_allowed(task);
879
Brice Goglin3140a222009-01-06 14:38:57 -0800880 err = -ENOMEM;
881 pm = (struct page_to_node *)__get_free_page(GFP_KERNEL);
882 if (!pm)
Brice Goglin5e9a0f02008-10-18 20:27:17 -0700883 goto out;
Brice Goglin35282a22009-06-16 15:32:43 -0700884
885 migrate_prep();
886
Brice Goglin5e9a0f02008-10-18 20:27:17 -0700887 /*
Brice Goglin3140a222009-01-06 14:38:57 -0800888 * Store a chunk of page_to_node array in a page,
889 * but keep the last one as a marker
Brice Goglin5e9a0f02008-10-18 20:27:17 -0700890 */
Brice Goglin3140a222009-01-06 14:38:57 -0800891 chunk_nr_pages = (PAGE_SIZE / sizeof(struct page_to_node)) - 1;
Brice Goglin5e9a0f02008-10-18 20:27:17 -0700892
Brice Goglin3140a222009-01-06 14:38:57 -0800893 for (chunk_start = 0;
894 chunk_start < nr_pages;
895 chunk_start += chunk_nr_pages) {
896 int j;
Brice Goglin5e9a0f02008-10-18 20:27:17 -0700897
Brice Goglin3140a222009-01-06 14:38:57 -0800898 if (chunk_start + chunk_nr_pages > nr_pages)
899 chunk_nr_pages = nr_pages - chunk_start;
900
901 /* fill the chunk pm with addrs and nodes from user-space */
902 for (j = 0; j < chunk_nr_pages; j++) {
903 const void __user *p;
Brice Goglin5e9a0f02008-10-18 20:27:17 -0700904 int node;
905
Brice Goglin3140a222009-01-06 14:38:57 -0800906 err = -EFAULT;
907 if (get_user(p, pages + j + chunk_start))
908 goto out_pm;
909 pm[j].addr = (unsigned long) p;
910
911 if (get_user(node, nodes + j + chunk_start))
Brice Goglin5e9a0f02008-10-18 20:27:17 -0700912 goto out_pm;
913
914 err = -ENODEV;
915 if (!node_state(node, N_HIGH_MEMORY))
916 goto out_pm;
917
918 err = -EACCES;
919 if (!node_isset(node, task_nodes))
920 goto out_pm;
921
Brice Goglin3140a222009-01-06 14:38:57 -0800922 pm[j].node = node;
923 }
Brice Goglin5e9a0f02008-10-18 20:27:17 -0700924
Brice Goglin3140a222009-01-06 14:38:57 -0800925 /* End marker for this chunk */
926 pm[chunk_nr_pages].node = MAX_NUMNODES;
927
928 /* Migrate this chunk */
929 err = do_move_page_to_node_array(mm, pm,
930 flags & MPOL_MF_MOVE_ALL);
931 if (err < 0)
932 goto out_pm;
933
Brice Goglin5e9a0f02008-10-18 20:27:17 -0700934 /* Return status information */
Brice Goglin3140a222009-01-06 14:38:57 -0800935 for (j = 0; j < chunk_nr_pages; j++)
936 if (put_user(pm[j].status, status + j + chunk_start)) {
Brice Goglin5e9a0f02008-10-18 20:27:17 -0700937 err = -EFAULT;
Brice Goglin3140a222009-01-06 14:38:57 -0800938 goto out_pm;
939 }
940 }
941 err = 0;
Brice Goglin5e9a0f02008-10-18 20:27:17 -0700942
943out_pm:
Brice Goglin3140a222009-01-06 14:38:57 -0800944 free_page((unsigned long)pm);
Brice Goglin5e9a0f02008-10-18 20:27:17 -0700945out:
946 return err;
947}
948
949/*
Brice Goglin2f007e72008-10-18 20:27:16 -0700950 * Determine the nodes of an array of pages and store it in an array of status.
Christoph Lameter742755a2006-06-23 02:03:55 -0700951 */
Brice Goglin80bba122008-12-09 13:14:23 -0800952static void do_pages_stat_array(struct mm_struct *mm, unsigned long nr_pages,
953 const void __user **pages, int *status)
Christoph Lameter742755a2006-06-23 02:03:55 -0700954{
Brice Goglin2f007e72008-10-18 20:27:16 -0700955 unsigned long i;
Brice Goglin2f007e72008-10-18 20:27:16 -0700956
Christoph Lameter742755a2006-06-23 02:03:55 -0700957 down_read(&mm->mmap_sem);
958
Brice Goglin2f007e72008-10-18 20:27:16 -0700959 for (i = 0; i < nr_pages; i++) {
Brice Goglin80bba122008-12-09 13:14:23 -0800960 unsigned long addr = (unsigned long)(*pages);
Christoph Lameter742755a2006-06-23 02:03:55 -0700961 struct vm_area_struct *vma;
962 struct page *page;
KOSAKI Motohiroc095adb2008-12-16 16:06:43 +0900963 int err = -EFAULT;
Brice Goglin2f007e72008-10-18 20:27:16 -0700964
965 vma = find_vma(mm, addr);
Christoph Lameter742755a2006-06-23 02:03:55 -0700966 if (!vma)
967 goto set_status;
968
Brice Goglin2f007e72008-10-18 20:27:16 -0700969 page = follow_page(vma, addr, 0);
Linus Torvalds89f5b7d2008-06-20 11:18:25 -0700970
971 err = PTR_ERR(page);
972 if (IS_ERR(page))
973 goto set_status;
974
Christoph Lameter742755a2006-06-23 02:03:55 -0700975 err = -ENOENT;
976 /* Use PageReserved to check for zero page */
Hugh Dickins62b61f62009-12-14 17:59:33 -0800977 if (!page || PageReserved(page) || PageKsm(page))
Christoph Lameter742755a2006-06-23 02:03:55 -0700978 goto set_status;
979
980 err = page_to_nid(page);
981set_status:
Brice Goglin80bba122008-12-09 13:14:23 -0800982 *status = err;
983
984 pages++;
985 status++;
986 }
987
988 up_read(&mm->mmap_sem);
989}
990
991/*
992 * Determine the nodes of a user array of pages and store it in
993 * a user array of status.
994 */
995static int do_pages_stat(struct mm_struct *mm, unsigned long nr_pages,
996 const void __user * __user *pages,
997 int __user *status)
998{
999#define DO_PAGES_STAT_CHUNK_NR 16
1000 const void __user *chunk_pages[DO_PAGES_STAT_CHUNK_NR];
1001 int chunk_status[DO_PAGES_STAT_CHUNK_NR];
1002 unsigned long i, chunk_nr = DO_PAGES_STAT_CHUNK_NR;
1003 int err;
1004
1005 for (i = 0; i < nr_pages; i += chunk_nr) {
H. Peter Anvinb92558502009-12-08 14:01:32 -08001006 if (chunk_nr > nr_pages - i)
Brice Goglin80bba122008-12-09 13:14:23 -08001007 chunk_nr = nr_pages - i;
1008
1009 err = copy_from_user(chunk_pages, &pages[i],
1010 chunk_nr * sizeof(*chunk_pages));
1011 if (err) {
1012 err = -EFAULT;
1013 goto out;
1014 }
1015
1016 do_pages_stat_array(mm, chunk_nr, chunk_pages, chunk_status);
1017
1018 err = copy_to_user(&status[i], chunk_status,
1019 chunk_nr * sizeof(*chunk_status));
1020 if (err) {
1021 err = -EFAULT;
1022 goto out;
1023 }
Christoph Lameter742755a2006-06-23 02:03:55 -07001024 }
Brice Goglin2f007e72008-10-18 20:27:16 -07001025 err = 0;
Christoph Lameter742755a2006-06-23 02:03:55 -07001026
Brice Goglin2f007e72008-10-18 20:27:16 -07001027out:
Brice Goglin2f007e72008-10-18 20:27:16 -07001028 return err;
Christoph Lameter742755a2006-06-23 02:03:55 -07001029}
1030
1031/*
1032 * Move a list of pages in the address space of the currently executing
1033 * process.
1034 */
Heiko Carstens938bb9f2009-01-14 14:14:30 +01001035SYSCALL_DEFINE6(move_pages, pid_t, pid, unsigned long, nr_pages,
1036 const void __user * __user *, pages,
1037 const int __user *, nodes,
1038 int __user *, status, int, flags)
Christoph Lameter742755a2006-06-23 02:03:55 -07001039{
David Howellsc69e8d92008-11-14 10:39:19 +11001040 const struct cred *cred = current_cred(), *tcred;
Christoph Lameter742755a2006-06-23 02:03:55 -07001041 struct task_struct *task;
Christoph Lameter742755a2006-06-23 02:03:55 -07001042 struct mm_struct *mm;
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001043 int err;
Christoph Lameter742755a2006-06-23 02:03:55 -07001044
1045 /* Check flags */
1046 if (flags & ~(MPOL_MF_MOVE|MPOL_MF_MOVE_ALL))
1047 return -EINVAL;
1048
1049 if ((flags & MPOL_MF_MOVE_ALL) && !capable(CAP_SYS_NICE))
1050 return -EPERM;
1051
1052 /* Find the mm_struct */
1053 read_lock(&tasklist_lock);
Pavel Emelyanov228ebcb2007-10-18 23:40:16 -07001054 task = pid ? find_task_by_vpid(pid) : current;
Christoph Lameter742755a2006-06-23 02:03:55 -07001055 if (!task) {
1056 read_unlock(&tasklist_lock);
1057 return -ESRCH;
1058 }
1059 mm = get_task_mm(task);
1060 read_unlock(&tasklist_lock);
1061
1062 if (!mm)
1063 return -EINVAL;
1064
1065 /*
1066 * Check if this process has the right to modify the specified
1067 * process. The right exists if the process has administrative
1068 * capabilities, superuser privileges or the same
1069 * userid as the target process.
1070 */
David Howellsc69e8d92008-11-14 10:39:19 +11001071 rcu_read_lock();
1072 tcred = __task_cred(task);
David Howellsb6dff3e2008-11-14 10:39:16 +11001073 if (cred->euid != tcred->suid && cred->euid != tcred->uid &&
1074 cred->uid != tcred->suid && cred->uid != tcred->uid &&
Christoph Lameter742755a2006-06-23 02:03:55 -07001075 !capable(CAP_SYS_NICE)) {
David Howellsc69e8d92008-11-14 10:39:19 +11001076 rcu_read_unlock();
Christoph Lameter742755a2006-06-23 02:03:55 -07001077 err = -EPERM;
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001078 goto out;
Christoph Lameter742755a2006-06-23 02:03:55 -07001079 }
David Howellsc69e8d92008-11-14 10:39:19 +11001080 rcu_read_unlock();
Christoph Lameter742755a2006-06-23 02:03:55 -07001081
David Quigley86c3a762006-06-23 02:04:02 -07001082 err = security_task_movememory(task);
1083 if (err)
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001084 goto out;
David Quigley86c3a762006-06-23 02:04:02 -07001085
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001086 if (nodes) {
1087 err = do_pages_move(mm, task, nr_pages, pages, nodes, status,
1088 flags);
1089 } else {
Brice Goglin2f007e72008-10-18 20:27:16 -07001090 err = do_pages_stat(mm, nr_pages, pages, status);
Brice Goglin2f007e72008-10-18 20:27:16 -07001091 }
David Quigley86c3a762006-06-23 02:04:02 -07001092
Christoph Lameter742755a2006-06-23 02:03:55 -07001093out:
Christoph Lameter742755a2006-06-23 02:03:55 -07001094 mmput(mm);
1095 return err;
1096}
Christoph Lameter742755a2006-06-23 02:03:55 -07001097
Christoph Lameter7b2259b2006-06-25 05:46:48 -07001098/*
1099 * Call migration functions in the vma_ops that may prepare
1100 * memory in a vm for migration. migration functions may perform
1101 * the migration for vmas that do not have an underlying page struct.
1102 */
1103int migrate_vmas(struct mm_struct *mm, const nodemask_t *to,
1104 const nodemask_t *from, unsigned long flags)
1105{
1106 struct vm_area_struct *vma;
1107 int err = 0;
1108
Daisuke Nishimura1001c9f2009-02-11 13:04:18 -08001109 for (vma = mm->mmap; vma && !err; vma = vma->vm_next) {
Christoph Lameter7b2259b2006-06-25 05:46:48 -07001110 if (vma->vm_ops && vma->vm_ops->migrate) {
1111 err = vma->vm_ops->migrate(vma, to, from, flags);
1112 if (err)
1113 break;
1114 }
1115 }
1116 return err;
1117}
Gerald Schaefer83d16742008-07-23 21:28:22 -07001118#endif