blob: 3b676b0c5c3ecca91d5e91d8ef2adc2982353f57 [file] [log] [blame]
Christoph Lameterb20a3502006-03-22 00:09:12 -08001/*
2 * Memory Migration functionality - linux/mm/migration.c
3 *
4 * Copyright (C) 2006 Silicon Graphics, Inc., Christoph Lameter
5 *
6 * Page migration was first developed in the context of the memory hotplug
7 * project. The main authors of the migration code are:
8 *
9 * IWAMOTO Toshihiro <iwamoto@valinux.co.jp>
10 * Hirokazu Takahashi <taka@valinux.co.jp>
11 * Dave Hansen <haveblue@us.ibm.com>
Christoph Lametercde53532008-07-04 09:59:22 -070012 * Christoph Lameter
Christoph Lameterb20a3502006-03-22 00:09:12 -080013 */
14
15#include <linux/migrate.h>
Paul Gortmakerb95f1b312011-10-16 02:01:52 -040016#include <linux/export.h>
Christoph Lameterb20a3502006-03-22 00:09:12 -080017#include <linux/swap.h>
Christoph Lameter06972122006-06-23 02:03:35 -070018#include <linux/swapops.h>
Christoph Lameterb20a3502006-03-22 00:09:12 -080019#include <linux/pagemap.h>
Christoph Lametere23ca002006-04-10 22:52:57 -070020#include <linux/buffer_head.h>
Christoph Lameterb20a3502006-03-22 00:09:12 -080021#include <linux/mm_inline.h>
Pavel Emelyanovb4888932007-10-18 23:40:14 -070022#include <linux/nsproxy.h>
Christoph Lameterb20a3502006-03-22 00:09:12 -080023#include <linux/pagevec.h>
Hugh Dickinse9995ef2009-12-14 17:59:31 -080024#include <linux/ksm.h>
Christoph Lameterb20a3502006-03-22 00:09:12 -080025#include <linux/rmap.h>
26#include <linux/topology.h>
27#include <linux/cpu.h>
28#include <linux/cpuset.h>
Christoph Lameter04e62a22006-06-23 02:03:38 -070029#include <linux/writeback.h>
Christoph Lameter742755a2006-06-23 02:03:55 -070030#include <linux/mempolicy.h>
31#include <linux/vmalloc.h>
David Quigley86c3a762006-06-23 02:04:02 -070032#include <linux/security.h>
Balbir Singh8a9f3cc2008-02-07 00:13:53 -080033#include <linux/memcontrol.h>
Adrian Bunk4f5ca262008-07-23 21:27:02 -070034#include <linux/syscalls.h>
Naoya Horiguchi290408d2010-09-08 10:19:35 +090035#include <linux/hugetlb.h>
Aneesh Kumar K.V8e6ac7f2012-07-31 16:42:27 -070036#include <linux/hugetlb_cgroup.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090037#include <linux/gfp.h>
Rafael Aquinibf6bddf2012-12-11 16:02:42 -080038#include <linux/balloon_compaction.h>
Christoph Lameterb20a3502006-03-22 00:09:12 -080039
Michal Nazarewicz0d1836c2010-12-21 17:24:26 -080040#include <asm/tlbflush.h>
41
Mel Gorman7b2a2d42012-10-19 14:07:31 +010042#define CREATE_TRACE_POINTS
43#include <trace/events/migrate.h>
44
Christoph Lameterb20a3502006-03-22 00:09:12 -080045#include "internal.h"
46
Christoph Lameterb20a3502006-03-22 00:09:12 -080047/*
Christoph Lameter742755a2006-06-23 02:03:55 -070048 * migrate_prep() needs to be called before we start compiling a list of pages
Mel Gorman748446b2010-05-24 14:32:27 -070049 * to be migrated using isolate_lru_page(). If scheduling work on other CPUs is
50 * undesirable, use migrate_prep_local()
Christoph Lameterb20a3502006-03-22 00:09:12 -080051 */
52int migrate_prep(void)
53{
Christoph Lameterb20a3502006-03-22 00:09:12 -080054 /*
55 * Clear the LRU lists so pages can be isolated.
56 * Note that pages may be moved off the LRU after we have
57 * drained them. Those pages will fail to migrate like other
58 * pages that may be busy.
59 */
60 lru_add_drain_all();
61
62 return 0;
63}
64
Mel Gorman748446b2010-05-24 14:32:27 -070065/* Do the necessary work of migrate_prep but not if it involves other CPUs */
66int migrate_prep_local(void)
67{
68 lru_add_drain();
69
70 return 0;
71}
72
Christoph Lameterb20a3502006-03-22 00:09:12 -080073/*
Lee Schermerhorn894bc312008-10-18 20:26:39 -070074 * Add isolated pages on the list back to the LRU under page lock
75 * to avoid leaking evictable pages back onto unevictable list.
Christoph Lameterb20a3502006-03-22 00:09:12 -080076 */
Minchan Kime13861d2010-05-24 14:31:59 -070077void putback_lru_pages(struct list_head *l)
Christoph Lameterb20a3502006-03-22 00:09:12 -080078{
79 struct page *page;
80 struct page *page2;
Christoph Lameterb20a3502006-03-22 00:09:12 -080081
82 list_for_each_entry_safe(page, page2, l, lru) {
Christoph Lametere24f0b82006-06-23 02:03:51 -070083 list_del(&page->lru);
KOSAKI Motohiroa7312862009-09-21 17:01:37 -070084 dec_zone_page_state(page, NR_ISOLATED_ANON +
Johannes Weiner6c0b1352009-09-21 17:02:59 -070085 page_is_file_cache(page));
Rafael Aquini5733c7d2012-12-11 16:02:47 -080086 putback_lru_page(page);
87 }
88}
89
90/*
91 * Put previously isolated pages back onto the appropriate lists
92 * from where they were once taken off for compaction/migration.
93 *
94 * This function shall be used instead of putback_lru_pages(),
95 * whenever the isolated pageset has been built by isolate_migratepages_range()
96 */
97void putback_movable_pages(struct list_head *l)
98{
99 struct page *page;
100 struct page *page2;
101
102 list_for_each_entry_safe(page, page2, l, lru) {
103 list_del(&page->lru);
104 dec_zone_page_state(page, NR_ISOLATED_ANON +
105 page_is_file_cache(page));
Rafael Aquinibf6bddf2012-12-11 16:02:42 -0800106 if (unlikely(balloon_page_movable(page)))
107 balloon_page_putback(page);
108 else
109 putback_lru_page(page);
Christoph Lameterb20a3502006-03-22 00:09:12 -0800110 }
Christoph Lameterb20a3502006-03-22 00:09:12 -0800111}
112
Christoph Lameter06972122006-06-23 02:03:35 -0700113/*
114 * Restore a potential migration pte to a working pte entry
115 */
Hugh Dickinse9995ef2009-12-14 17:59:31 -0800116static int remove_migration_pte(struct page *new, struct vm_area_struct *vma,
117 unsigned long addr, void *old)
Christoph Lameter06972122006-06-23 02:03:35 -0700118{
119 struct mm_struct *mm = vma->vm_mm;
120 swp_entry_t entry;
Christoph Lameter06972122006-06-23 02:03:35 -0700121 pmd_t *pmd;
122 pte_t *ptep, pte;
123 spinlock_t *ptl;
124
Naoya Horiguchi290408d2010-09-08 10:19:35 +0900125 if (unlikely(PageHuge(new))) {
126 ptep = huge_pte_offset(mm, addr);
127 if (!ptep)
128 goto out;
129 ptl = &mm->page_table_lock;
130 } else {
Bob Liu62190492012-12-11 16:00:37 -0800131 pmd = mm_find_pmd(mm, addr);
132 if (!pmd)
Naoya Horiguchi290408d2010-09-08 10:19:35 +0900133 goto out;
Andrea Arcangeli500d65d2011-01-13 15:46:55 -0800134 if (pmd_trans_huge(*pmd))
135 goto out;
Christoph Lameter06972122006-06-23 02:03:35 -0700136
Naoya Horiguchi290408d2010-09-08 10:19:35 +0900137 ptep = pte_offset_map(pmd, addr);
Christoph Lameter06972122006-06-23 02:03:35 -0700138
Hugh Dickins486cf462011-10-19 12:50:35 -0700139 /*
140 * Peek to check is_swap_pte() before taking ptlock? No, we
141 * can race mremap's move_ptes(), which skips anon_vma lock.
142 */
Christoph Lameter06972122006-06-23 02:03:35 -0700143
Naoya Horiguchi290408d2010-09-08 10:19:35 +0900144 ptl = pte_lockptr(mm, pmd);
145 }
146
Christoph Lameter06972122006-06-23 02:03:35 -0700147 spin_lock(ptl);
148 pte = *ptep;
149 if (!is_swap_pte(pte))
Hugh Dickinse9995ef2009-12-14 17:59:31 -0800150 goto unlock;
Christoph Lameter06972122006-06-23 02:03:35 -0700151
152 entry = pte_to_swp_entry(pte);
153
Hugh Dickinse9995ef2009-12-14 17:59:31 -0800154 if (!is_migration_entry(entry) ||
155 migration_entry_to_page(entry) != old)
156 goto unlock;
Christoph Lameter06972122006-06-23 02:03:35 -0700157
Christoph Lameter06972122006-06-23 02:03:35 -0700158 get_page(new);
159 pte = pte_mkold(mk_pte(new, vma->vm_page_prot));
160 if (is_write_migration_entry(entry))
161 pte = pte_mkwrite(pte);
Andi Kleen3ef8fd72010-10-11 16:03:21 +0200162#ifdef CONFIG_HUGETLB_PAGE
Naoya Horiguchi290408d2010-09-08 10:19:35 +0900163 if (PageHuge(new))
164 pte = pte_mkhuge(pte);
Andi Kleen3ef8fd72010-10-11 16:03:21 +0200165#endif
KAMEZAWA Hiroyuki97ee0522007-10-16 01:25:43 -0700166 flush_cache_page(vma, addr, pte_pfn(pte));
Christoph Lameter06972122006-06-23 02:03:35 -0700167 set_pte_at(mm, addr, ptep, pte);
Christoph Lameter04e62a22006-06-23 02:03:38 -0700168
Naoya Horiguchi290408d2010-09-08 10:19:35 +0900169 if (PageHuge(new)) {
170 if (PageAnon(new))
171 hugepage_add_anon_rmap(new, vma, addr);
172 else
173 page_dup_rmap(new);
174 } else if (PageAnon(new))
Christoph Lameter04e62a22006-06-23 02:03:38 -0700175 page_add_anon_rmap(new, vma, addr);
176 else
177 page_add_file_rmap(new);
178
179 /* No need to invalidate - it was non-present before */
Russell King4b3073e2009-12-18 16:40:18 +0000180 update_mmu_cache(vma, addr, ptep);
Hugh Dickinse9995ef2009-12-14 17:59:31 -0800181unlock:
Christoph Lameter06972122006-06-23 02:03:35 -0700182 pte_unmap_unlock(ptep, ptl);
Hugh Dickinse9995ef2009-12-14 17:59:31 -0800183out:
184 return SWAP_AGAIN;
Christoph Lameter06972122006-06-23 02:03:35 -0700185}
186
187/*
Christoph Lameter04e62a22006-06-23 02:03:38 -0700188 * Get rid of all migration entries and replace them by
189 * references to the indicated page.
190 */
191static void remove_migration_ptes(struct page *old, struct page *new)
192{
Hugh Dickinse9995ef2009-12-14 17:59:31 -0800193 rmap_walk(new, remove_migration_pte, old);
Christoph Lameter04e62a22006-06-23 02:03:38 -0700194}
195
196/*
Christoph Lameter06972122006-06-23 02:03:35 -0700197 * Something used the pte of a page under migration. We need to
198 * get to the page and wait until migration is finished.
199 * When we return from this function the fault will be retried.
Christoph Lameter06972122006-06-23 02:03:35 -0700200 */
201void migration_entry_wait(struct mm_struct *mm, pmd_t *pmd,
202 unsigned long address)
203{
204 pte_t *ptep, pte;
205 spinlock_t *ptl;
206 swp_entry_t entry;
207 struct page *page;
208
209 ptep = pte_offset_map_lock(mm, pmd, address, &ptl);
210 pte = *ptep;
211 if (!is_swap_pte(pte))
212 goto out;
213
214 entry = pte_to_swp_entry(pte);
215 if (!is_migration_entry(entry))
216 goto out;
217
218 page = migration_entry_to_page(entry);
219
Nick Piggine2867812008-07-25 19:45:30 -0700220 /*
221 * Once radix-tree replacement of page migration started, page_count
222 * *must* be zero. And, we don't want to call wait_on_page_locked()
223 * against a page without get_page().
224 * So, we use get_page_unless_zero(), here. Even failed, page fault
225 * will occur again.
226 */
227 if (!get_page_unless_zero(page))
228 goto out;
Christoph Lameter06972122006-06-23 02:03:35 -0700229 pte_unmap_unlock(ptep, ptl);
230 wait_on_page_locked(page);
231 put_page(page);
232 return;
233out:
234 pte_unmap_unlock(ptep, ptl);
235}
236
Mel Gormanb969c4a2012-01-12 17:19:34 -0800237#ifdef CONFIG_BLOCK
238/* Returns true if all buffers are successfully locked */
Mel Gormana6bc32b2012-01-12 17:19:43 -0800239static bool buffer_migrate_lock_buffers(struct buffer_head *head,
240 enum migrate_mode mode)
Mel Gormanb969c4a2012-01-12 17:19:34 -0800241{
242 struct buffer_head *bh = head;
243
244 /* Simple case, sync compaction */
Mel Gormana6bc32b2012-01-12 17:19:43 -0800245 if (mode != MIGRATE_ASYNC) {
Mel Gormanb969c4a2012-01-12 17:19:34 -0800246 do {
247 get_bh(bh);
248 lock_buffer(bh);
249 bh = bh->b_this_page;
250
251 } while (bh != head);
252
253 return true;
254 }
255
256 /* async case, we cannot block on lock_buffer so use trylock_buffer */
257 do {
258 get_bh(bh);
259 if (!trylock_buffer(bh)) {
260 /*
261 * We failed to lock the buffer and cannot stall in
262 * async migration. Release the taken locks
263 */
264 struct buffer_head *failed_bh = bh;
265 put_bh(failed_bh);
266 bh = head;
267 while (bh != failed_bh) {
268 unlock_buffer(bh);
269 put_bh(bh);
270 bh = bh->b_this_page;
271 }
272 return false;
273 }
274
275 bh = bh->b_this_page;
276 } while (bh != head);
277 return true;
278}
279#else
280static inline bool buffer_migrate_lock_buffers(struct buffer_head *head,
Mel Gormana6bc32b2012-01-12 17:19:43 -0800281 enum migrate_mode mode)
Mel Gormanb969c4a2012-01-12 17:19:34 -0800282{
283 return true;
284}
285#endif /* CONFIG_BLOCK */
286
Christoph Lameterb20a3502006-03-22 00:09:12 -0800287/*
Christoph Lameterc3fcf8a2006-06-23 02:03:32 -0700288 * Replace the page in the mapping.
Christoph Lameter5b5c7122006-06-23 02:03:29 -0700289 *
290 * The number of remaining references must be:
291 * 1 for anonymous pages without a mapping
292 * 2 for pages with a mapping
David Howells266cf652009-04-03 16:42:36 +0100293 * 3 for pages with a mapping and PagePrivate/PagePrivate2 set.
Christoph Lameterb20a3502006-03-22 00:09:12 -0800294 */
Christoph Lameter2d1db3b2006-06-23 02:03:33 -0700295static int migrate_page_move_mapping(struct address_space *mapping,
Mel Gormanb969c4a2012-01-12 17:19:34 -0800296 struct page *newpage, struct page *page,
Mel Gormana6bc32b2012-01-12 17:19:43 -0800297 struct buffer_head *head, enum migrate_mode mode)
Christoph Lameterb20a3502006-03-22 00:09:12 -0800298{
Peter Zijlstra7039e1d2012-10-25 14:16:34 +0200299 int expected_count = 0;
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800300 void **pslot;
Christoph Lameterb20a3502006-03-22 00:09:12 -0800301
Christoph Lameter6c5240a2006-06-23 02:03:37 -0700302 if (!mapping) {
Christoph Lameter0e8c7d02007-04-23 14:41:09 -0700303 /* Anonymous page without mapping */
Christoph Lameter6c5240a2006-06-23 02:03:37 -0700304 if (page_count(page) != 1)
305 return -EAGAIN;
Rafael Aquini78bd5202012-12-11 16:02:31 -0800306 return MIGRATEPAGE_SUCCESS;
Christoph Lameter6c5240a2006-06-23 02:03:37 -0700307 }
308
Nick Piggin19fd6232008-07-25 19:45:32 -0700309 spin_lock_irq(&mapping->tree_lock);
Christoph Lameterb20a3502006-03-22 00:09:12 -0800310
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800311 pslot = radix_tree_lookup_slot(&mapping->page_tree,
312 page_index(page));
Christoph Lameterb20a3502006-03-22 00:09:12 -0800313
Johannes Weineredcf4742009-09-21 17:02:59 -0700314 expected_count = 2 + page_has_private(page);
Nick Piggine2867812008-07-25 19:45:30 -0700315 if (page_count(page) != expected_count ||
Mel Gorman29c1f672011-01-13 15:47:21 -0800316 radix_tree_deref_slot_protected(pslot, &mapping->tree_lock) != page) {
Nick Piggin19fd6232008-07-25 19:45:32 -0700317 spin_unlock_irq(&mapping->tree_lock);
Christoph Lametere23ca002006-04-10 22:52:57 -0700318 return -EAGAIN;
Christoph Lameterb20a3502006-03-22 00:09:12 -0800319 }
320
Nick Piggine2867812008-07-25 19:45:30 -0700321 if (!page_freeze_refs(page, expected_count)) {
Nick Piggin19fd6232008-07-25 19:45:32 -0700322 spin_unlock_irq(&mapping->tree_lock);
Nick Piggine2867812008-07-25 19:45:30 -0700323 return -EAGAIN;
324 }
325
Christoph Lameterb20a3502006-03-22 00:09:12 -0800326 /*
Mel Gormanb969c4a2012-01-12 17:19:34 -0800327 * In the async migration case of moving a page with buffers, lock the
328 * buffers using trylock before the mapping is moved. If the mapping
329 * was moved, we later failed to lock the buffers and could not move
330 * the mapping back due to an elevated page count, we would have to
331 * block waiting on other references to be dropped.
332 */
Mel Gormana6bc32b2012-01-12 17:19:43 -0800333 if (mode == MIGRATE_ASYNC && head &&
334 !buffer_migrate_lock_buffers(head, mode)) {
Mel Gormanb969c4a2012-01-12 17:19:34 -0800335 page_unfreeze_refs(page, expected_count);
336 spin_unlock_irq(&mapping->tree_lock);
337 return -EAGAIN;
338 }
339
340 /*
Christoph Lameterb20a3502006-03-22 00:09:12 -0800341 * Now we know that no one else is looking at the page.
Christoph Lameterb20a3502006-03-22 00:09:12 -0800342 */
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800343 get_page(newpage); /* add cache reference */
Christoph Lameterb20a3502006-03-22 00:09:12 -0800344 if (PageSwapCache(page)) {
345 SetPageSwapCache(newpage);
346 set_page_private(newpage, page_private(page));
347 }
348
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800349 radix_tree_replace_slot(pslot, newpage);
350
351 /*
Jacobo Giralt937a94c2012-01-10 15:07:11 -0800352 * Drop cache reference from old page by unfreezing
353 * to one less reference.
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800354 * We know this isn't the last reference.
355 */
Jacobo Giralt937a94c2012-01-10 15:07:11 -0800356 page_unfreeze_refs(page, expected_count - 1);
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800357
Christoph Lameter0e8c7d02007-04-23 14:41:09 -0700358 /*
359 * If moved to a different zone then also account
360 * the page for that zone. Other VM counters will be
361 * taken care of when we establish references to the
362 * new page and drop references to the old page.
363 *
364 * Note that anonymous pages are accounted for
365 * via NR_FILE_PAGES and NR_ANON_PAGES if they
366 * are mapped to swap space.
367 */
368 __dec_zone_page_state(page, NR_FILE_PAGES);
369 __inc_zone_page_state(newpage, NR_FILE_PAGES);
Andrea Arcangeli99a15e22011-06-16 12:56:19 -0700370 if (!PageSwapCache(page) && PageSwapBacked(page)) {
KOSAKI Motohiro4b021082009-09-21 17:01:33 -0700371 __dec_zone_page_state(page, NR_SHMEM);
372 __inc_zone_page_state(newpage, NR_SHMEM);
373 }
Nick Piggin19fd6232008-07-25 19:45:32 -0700374 spin_unlock_irq(&mapping->tree_lock);
Christoph Lameterb20a3502006-03-22 00:09:12 -0800375
Rafael Aquini78bd5202012-12-11 16:02:31 -0800376 return MIGRATEPAGE_SUCCESS;
Christoph Lameterb20a3502006-03-22 00:09:12 -0800377}
Christoph Lameterb20a3502006-03-22 00:09:12 -0800378
379/*
Naoya Horiguchi290408d2010-09-08 10:19:35 +0900380 * The expected number of remaining references is the same as that
381 * of migrate_page_move_mapping().
382 */
383int migrate_huge_page_move_mapping(struct address_space *mapping,
384 struct page *newpage, struct page *page)
385{
386 int expected_count;
387 void **pslot;
388
389 if (!mapping) {
390 if (page_count(page) != 1)
391 return -EAGAIN;
Rafael Aquini78bd5202012-12-11 16:02:31 -0800392 return MIGRATEPAGE_SUCCESS;
Naoya Horiguchi290408d2010-09-08 10:19:35 +0900393 }
394
395 spin_lock_irq(&mapping->tree_lock);
396
397 pslot = radix_tree_lookup_slot(&mapping->page_tree,
398 page_index(page));
399
400 expected_count = 2 + page_has_private(page);
401 if (page_count(page) != expected_count ||
Mel Gorman29c1f672011-01-13 15:47:21 -0800402 radix_tree_deref_slot_protected(pslot, &mapping->tree_lock) != page) {
Naoya Horiguchi290408d2010-09-08 10:19:35 +0900403 spin_unlock_irq(&mapping->tree_lock);
404 return -EAGAIN;
405 }
406
407 if (!page_freeze_refs(page, expected_count)) {
408 spin_unlock_irq(&mapping->tree_lock);
409 return -EAGAIN;
410 }
411
412 get_page(newpage);
413
414 radix_tree_replace_slot(pslot, newpage);
415
Jacobo Giralt937a94c2012-01-10 15:07:11 -0800416 page_unfreeze_refs(page, expected_count - 1);
Naoya Horiguchi290408d2010-09-08 10:19:35 +0900417
418 spin_unlock_irq(&mapping->tree_lock);
Rafael Aquini78bd5202012-12-11 16:02:31 -0800419 return MIGRATEPAGE_SUCCESS;
Naoya Horiguchi290408d2010-09-08 10:19:35 +0900420}
421
422/*
Christoph Lameterb20a3502006-03-22 00:09:12 -0800423 * Copy the page to its new location
424 */
Naoya Horiguchi290408d2010-09-08 10:19:35 +0900425void migrate_page_copy(struct page *newpage, struct page *page)
Christoph Lameterb20a3502006-03-22 00:09:12 -0800426{
Mel Gormanb32967f2012-11-19 12:35:47 +0000427 if (PageHuge(page) || PageTransHuge(page))
Naoya Horiguchi290408d2010-09-08 10:19:35 +0900428 copy_huge_page(newpage, page);
429 else
430 copy_highpage(newpage, page);
Christoph Lameterb20a3502006-03-22 00:09:12 -0800431
432 if (PageError(page))
433 SetPageError(newpage);
434 if (PageReferenced(page))
435 SetPageReferenced(newpage);
436 if (PageUptodate(page))
437 SetPageUptodate(newpage);
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700438 if (TestClearPageActive(page)) {
439 VM_BUG_ON(PageUnevictable(page));
Christoph Lameterb20a3502006-03-22 00:09:12 -0800440 SetPageActive(newpage);
Lee Schermerhorn418b27e2009-12-14 17:59:54 -0800441 } else if (TestClearPageUnevictable(page))
442 SetPageUnevictable(newpage);
Christoph Lameterb20a3502006-03-22 00:09:12 -0800443 if (PageChecked(page))
444 SetPageChecked(newpage);
445 if (PageMappedToDisk(page))
446 SetPageMappedToDisk(newpage);
447
448 if (PageDirty(page)) {
449 clear_page_dirty_for_io(page);
Nick Piggin3a902c52008-04-30 00:55:16 -0700450 /*
451 * Want to mark the page and the radix tree as dirty, and
452 * redo the accounting that clear_page_dirty_for_io undid,
453 * but we can't use set_page_dirty because that function
454 * is actually a signal that all of the page has become dirty.
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300455 * Whereas only part of our page may be dirty.
Nick Piggin3a902c52008-04-30 00:55:16 -0700456 */
Hugh Dickins752dc182012-06-02 00:27:47 -0700457 if (PageSwapBacked(page))
458 SetPageDirty(newpage);
459 else
460 __set_page_dirty_nobuffers(newpage);
Christoph Lameterb20a3502006-03-22 00:09:12 -0800461 }
462
Nick Pigginb291f002008-10-18 20:26:44 -0700463 mlock_migrate_page(newpage, page);
Hugh Dickinse9995ef2009-12-14 17:59:31 -0800464 ksm_migrate_page(newpage, page);
Nick Pigginb291f002008-10-18 20:26:44 -0700465
Christoph Lameterb20a3502006-03-22 00:09:12 -0800466 ClearPageSwapCache(page);
Christoph Lameterb20a3502006-03-22 00:09:12 -0800467 ClearPagePrivate(page);
468 set_page_private(page, 0);
Christoph Lameterb20a3502006-03-22 00:09:12 -0800469
470 /*
471 * If any waiters have accumulated on the new page then
472 * wake them up.
473 */
474 if (PageWriteback(newpage))
475 end_page_writeback(newpage);
476}
Christoph Lameterb20a3502006-03-22 00:09:12 -0800477
Christoph Lameter1d8b85c2006-06-23 02:03:28 -0700478/************************************************************
479 * Migration functions
480 ***********************************************************/
481
482/* Always fail migration. Used for mappings that are not movable */
Christoph Lameter2d1db3b2006-06-23 02:03:33 -0700483int fail_migrate_page(struct address_space *mapping,
484 struct page *newpage, struct page *page)
Christoph Lameter1d8b85c2006-06-23 02:03:28 -0700485{
486 return -EIO;
487}
488EXPORT_SYMBOL(fail_migrate_page);
489
Christoph Lameterb20a3502006-03-22 00:09:12 -0800490/*
491 * Common logic to directly migrate a single page suitable for
David Howells266cf652009-04-03 16:42:36 +0100492 * pages that do not use PagePrivate/PagePrivate2.
Christoph Lameterb20a3502006-03-22 00:09:12 -0800493 *
494 * Pages are locked upon entry and exit.
495 */
Christoph Lameter2d1db3b2006-06-23 02:03:33 -0700496int migrate_page(struct address_space *mapping,
Mel Gormana6bc32b2012-01-12 17:19:43 -0800497 struct page *newpage, struct page *page,
498 enum migrate_mode mode)
Christoph Lameterb20a3502006-03-22 00:09:12 -0800499{
500 int rc;
501
502 BUG_ON(PageWriteback(page)); /* Writeback must be complete */
503
Mel Gormana6bc32b2012-01-12 17:19:43 -0800504 rc = migrate_page_move_mapping(mapping, newpage, page, NULL, mode);
Christoph Lameterb20a3502006-03-22 00:09:12 -0800505
Rafael Aquini78bd5202012-12-11 16:02:31 -0800506 if (rc != MIGRATEPAGE_SUCCESS)
Christoph Lameterb20a3502006-03-22 00:09:12 -0800507 return rc;
508
509 migrate_page_copy(newpage, page);
Rafael Aquini78bd5202012-12-11 16:02:31 -0800510 return MIGRATEPAGE_SUCCESS;
Christoph Lameterb20a3502006-03-22 00:09:12 -0800511}
512EXPORT_SYMBOL(migrate_page);
513
David Howells93614012006-09-30 20:45:40 +0200514#ifdef CONFIG_BLOCK
Christoph Lameterb20a3502006-03-22 00:09:12 -0800515/*
Christoph Lameter1d8b85c2006-06-23 02:03:28 -0700516 * Migration function for pages with buffers. This function can only be used
517 * if the underlying filesystem guarantees that no other references to "page"
518 * exist.
519 */
Christoph Lameter2d1db3b2006-06-23 02:03:33 -0700520int buffer_migrate_page(struct address_space *mapping,
Mel Gormana6bc32b2012-01-12 17:19:43 -0800521 struct page *newpage, struct page *page, enum migrate_mode mode)
Christoph Lameter1d8b85c2006-06-23 02:03:28 -0700522{
Christoph Lameter1d8b85c2006-06-23 02:03:28 -0700523 struct buffer_head *bh, *head;
524 int rc;
525
Christoph Lameter1d8b85c2006-06-23 02:03:28 -0700526 if (!page_has_buffers(page))
Mel Gormana6bc32b2012-01-12 17:19:43 -0800527 return migrate_page(mapping, newpage, page, mode);
Christoph Lameter1d8b85c2006-06-23 02:03:28 -0700528
529 head = page_buffers(page);
530
Mel Gormana6bc32b2012-01-12 17:19:43 -0800531 rc = migrate_page_move_mapping(mapping, newpage, page, head, mode);
Christoph Lameter1d8b85c2006-06-23 02:03:28 -0700532
Rafael Aquini78bd5202012-12-11 16:02:31 -0800533 if (rc != MIGRATEPAGE_SUCCESS)
Christoph Lameter1d8b85c2006-06-23 02:03:28 -0700534 return rc;
535
Mel Gormanb969c4a2012-01-12 17:19:34 -0800536 /*
537 * In the async case, migrate_page_move_mapping locked the buffers
538 * with an IRQ-safe spinlock held. In the sync case, the buffers
539 * need to be locked now
540 */
Mel Gormana6bc32b2012-01-12 17:19:43 -0800541 if (mode != MIGRATE_ASYNC)
542 BUG_ON(!buffer_migrate_lock_buffers(head, mode));
Christoph Lameter1d8b85c2006-06-23 02:03:28 -0700543
544 ClearPagePrivate(page);
545 set_page_private(newpage, page_private(page));
546 set_page_private(page, 0);
547 put_page(page);
548 get_page(newpage);
549
550 bh = head;
551 do {
552 set_bh_page(bh, newpage, bh_offset(bh));
553 bh = bh->b_this_page;
554
555 } while (bh != head);
556
557 SetPagePrivate(newpage);
558
559 migrate_page_copy(newpage, page);
560
561 bh = head;
562 do {
563 unlock_buffer(bh);
564 put_bh(bh);
565 bh = bh->b_this_page;
566
567 } while (bh != head);
568
Rafael Aquini78bd5202012-12-11 16:02:31 -0800569 return MIGRATEPAGE_SUCCESS;
Christoph Lameter1d8b85c2006-06-23 02:03:28 -0700570}
571EXPORT_SYMBOL(buffer_migrate_page);
David Howells93614012006-09-30 20:45:40 +0200572#endif
Christoph Lameter1d8b85c2006-06-23 02:03:28 -0700573
Christoph Lameter04e62a22006-06-23 02:03:38 -0700574/*
575 * Writeback a page to clean the dirty state
576 */
577static int writeout(struct address_space *mapping, struct page *page)
578{
579 struct writeback_control wbc = {
580 .sync_mode = WB_SYNC_NONE,
581 .nr_to_write = 1,
582 .range_start = 0,
583 .range_end = LLONG_MAX,
Christoph Lameter04e62a22006-06-23 02:03:38 -0700584 .for_reclaim = 1
585 };
586 int rc;
587
588 if (!mapping->a_ops->writepage)
589 /* No write method for the address space */
590 return -EINVAL;
591
592 if (!clear_page_dirty_for_io(page))
593 /* Someone else already triggered a write */
594 return -EAGAIN;
595
596 /*
597 * A dirty page may imply that the underlying filesystem has
598 * the page on some queue. So the page must be clean for
599 * migration. Writeout may mean we loose the lock and the
600 * page state is no longer what we checked for earlier.
601 * At this point we know that the migration attempt cannot
602 * be successful.
603 */
604 remove_migration_ptes(page, page);
605
606 rc = mapping->a_ops->writepage(page, &wbc);
Christoph Lameter04e62a22006-06-23 02:03:38 -0700607
608 if (rc != AOP_WRITEPAGE_ACTIVATE)
609 /* unlocked. Relock */
610 lock_page(page);
611
Hugh Dickinsbda85502008-11-19 15:36:36 -0800612 return (rc < 0) ? -EIO : -EAGAIN;
Christoph Lameter04e62a22006-06-23 02:03:38 -0700613}
614
615/*
616 * Default handling if a filesystem does not provide a migration function.
617 */
Christoph Lameter8351a6e2006-06-23 02:03:33 -0700618static int fallback_migrate_page(struct address_space *mapping,
Mel Gormana6bc32b2012-01-12 17:19:43 -0800619 struct page *newpage, struct page *page, enum migrate_mode mode)
Christoph Lameter8351a6e2006-06-23 02:03:33 -0700620{
Mel Gormanb969c4a2012-01-12 17:19:34 -0800621 if (PageDirty(page)) {
Mel Gormana6bc32b2012-01-12 17:19:43 -0800622 /* Only writeback pages in full synchronous migration */
623 if (mode != MIGRATE_SYNC)
Mel Gormanb969c4a2012-01-12 17:19:34 -0800624 return -EBUSY;
Christoph Lameter04e62a22006-06-23 02:03:38 -0700625 return writeout(mapping, page);
Mel Gormanb969c4a2012-01-12 17:19:34 -0800626 }
Christoph Lameter8351a6e2006-06-23 02:03:33 -0700627
628 /*
629 * Buffers may be managed in a filesystem specific way.
630 * We must have no buffers or drop them.
631 */
David Howells266cf652009-04-03 16:42:36 +0100632 if (page_has_private(page) &&
Christoph Lameter8351a6e2006-06-23 02:03:33 -0700633 !try_to_release_page(page, GFP_KERNEL))
634 return -EAGAIN;
635
Mel Gormana6bc32b2012-01-12 17:19:43 -0800636 return migrate_page(mapping, newpage, page, mode);
Christoph Lameter8351a6e2006-06-23 02:03:33 -0700637}
638
Christoph Lameter1d8b85c2006-06-23 02:03:28 -0700639/*
Christoph Lametere24f0b82006-06-23 02:03:51 -0700640 * Move a page to a newly allocated page
641 * The page is locked and all ptes have been successfully removed.
642 *
643 * The new page will have replaced the old page if this function
644 * is successful.
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700645 *
646 * Return value:
647 * < 0 - error code
Rafael Aquini78bd5202012-12-11 16:02:31 -0800648 * MIGRATEPAGE_SUCCESS - success
Christoph Lametere24f0b82006-06-23 02:03:51 -0700649 */
Mel Gorman3fe20112010-05-24 14:32:20 -0700650static int move_to_new_page(struct page *newpage, struct page *page,
Mel Gormana6bc32b2012-01-12 17:19:43 -0800651 int remap_swapcache, enum migrate_mode mode)
Christoph Lametere24f0b82006-06-23 02:03:51 -0700652{
653 struct address_space *mapping;
654 int rc;
655
656 /*
657 * Block others from accessing the page when we get around to
658 * establishing additional references. We are the only one
659 * holding a reference to the new page at this point.
660 */
Nick Piggin529ae9a2008-08-02 12:01:03 +0200661 if (!trylock_page(newpage))
Christoph Lametere24f0b82006-06-23 02:03:51 -0700662 BUG();
663
664 /* Prepare mapping for the new page.*/
665 newpage->index = page->index;
666 newpage->mapping = page->mapping;
Rik van Rielb2e18532008-10-18 20:26:30 -0700667 if (PageSwapBacked(page))
668 SetPageSwapBacked(newpage);
Christoph Lametere24f0b82006-06-23 02:03:51 -0700669
670 mapping = page_mapping(page);
671 if (!mapping)
Mel Gormana6bc32b2012-01-12 17:19:43 -0800672 rc = migrate_page(mapping, newpage, page, mode);
Mel Gormanb969c4a2012-01-12 17:19:34 -0800673 else if (mapping->a_ops->migratepage)
Christoph Lametere24f0b82006-06-23 02:03:51 -0700674 /*
Mel Gormanb969c4a2012-01-12 17:19:34 -0800675 * Most pages have a mapping and most filesystems provide a
676 * migratepage callback. Anonymous pages are part of swap
677 * space which also has its own migratepage callback. This
678 * is the most common path for page migration.
Christoph Lametere24f0b82006-06-23 02:03:51 -0700679 */
Mel Gormanb969c4a2012-01-12 17:19:34 -0800680 rc = mapping->a_ops->migratepage(mapping,
Mel Gormana6bc32b2012-01-12 17:19:43 -0800681 newpage, page, mode);
Mel Gormanb969c4a2012-01-12 17:19:34 -0800682 else
Mel Gormana6bc32b2012-01-12 17:19:43 -0800683 rc = fallback_migrate_page(mapping, newpage, page, mode);
Christoph Lametere24f0b82006-06-23 02:03:51 -0700684
Rafael Aquini78bd5202012-12-11 16:02:31 -0800685 if (rc != MIGRATEPAGE_SUCCESS) {
Christoph Lametere24f0b82006-06-23 02:03:51 -0700686 newpage->mapping = NULL;
Mel Gorman3fe20112010-05-24 14:32:20 -0700687 } else {
688 if (remap_swapcache)
689 remove_migration_ptes(page, newpage);
Konstantin Khlebnikov35512ec2012-02-03 15:37:13 -0800690 page->mapping = NULL;
Mel Gorman3fe20112010-05-24 14:32:20 -0700691 }
Christoph Lametere24f0b82006-06-23 02:03:51 -0700692
693 unlock_page(newpage);
694
695 return rc;
696}
697
Minchan Kim0dabec92011-10-31 17:06:57 -0700698static int __unmap_and_move(struct page *page, struct page *newpage,
Mel Gormana6bc32b2012-01-12 17:19:43 -0800699 int force, bool offlining, enum migrate_mode mode)
Christoph Lametere24f0b82006-06-23 02:03:51 -0700700{
Minchan Kim0dabec92011-10-31 17:06:57 -0700701 int rc = -EAGAIN;
Mel Gorman3fe20112010-05-24 14:32:20 -0700702 int remap_swapcache = 1;
KAMEZAWA Hiroyuki56039ef2011-03-23 16:42:19 -0700703 struct mem_cgroup *mem;
Mel Gorman3f6c8272010-05-24 14:32:17 -0700704 struct anon_vma *anon_vma = NULL;
Christoph Lameter95a402c2006-06-23 02:03:53 -0700705
Nick Piggin529ae9a2008-08-02 12:01:03 +0200706 if (!trylock_page(page)) {
Mel Gormana6bc32b2012-01-12 17:19:43 -0800707 if (!force || mode == MIGRATE_ASYNC)
Minchan Kim0dabec92011-10-31 17:06:57 -0700708 goto out;
Mel Gorman3e7d3442011-01-13 15:45:56 -0800709
710 /*
711 * It's not safe for direct compaction to call lock_page.
712 * For example, during page readahead pages are added locked
713 * to the LRU. Later, when the IO completes the pages are
714 * marked uptodate and unlocked. However, the queueing
715 * could be merging multiple pages for one bio (e.g.
716 * mpage_readpages). If an allocation happens for the
717 * second or third page, the process can end up locking
718 * the same page twice and deadlocking. Rather than
719 * trying to be clever about what pages can be locked,
720 * avoid the use of lock_page for direct compaction
721 * altogether.
722 */
723 if (current->flags & PF_MEMALLOC)
Minchan Kim0dabec92011-10-31 17:06:57 -0700724 goto out;
Mel Gorman3e7d3442011-01-13 15:45:56 -0800725
Christoph Lametere24f0b82006-06-23 02:03:51 -0700726 lock_page(page);
727 }
728
Hugh Dickins62b61f62009-12-14 17:59:33 -0800729 /*
730 * Only memory hotplug's offline_pages() caller has locked out KSM,
731 * and can safely migrate a KSM page. The other cases have skipped
732 * PageKsm along with PageReserved - but it is only now when we have
733 * the page lock that we can be certain it will not go KSM beneath us
734 * (KSM will not upgrade a page from PageAnon to PageKsm when it sees
735 * its pagecount raised, but only here do we take the page lock which
736 * serializes that).
737 */
738 if (PageKsm(page) && !offlining) {
739 rc = -EBUSY;
740 goto unlock;
741 }
742
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -0800743 /* charge against new page */
Johannes Weiner0030f532012-07-31 16:45:25 -0700744 mem_cgroup_prepare_migration(page, newpage, &mem);
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -0800745
Christoph Lametere24f0b82006-06-23 02:03:51 -0700746 if (PageWriteback(page)) {
Andrea Arcangeli11bc82d2011-03-22 16:33:11 -0700747 /*
Mel Gormana6bc32b2012-01-12 17:19:43 -0800748 * Only in the case of a full syncronous migration is it
749 * necessary to wait for PageWriteback. In the async case,
750 * the retry loop is too short and in the sync-light case,
751 * the overhead of stalling is too much
Andrea Arcangeli11bc82d2011-03-22 16:33:11 -0700752 */
Mel Gormana6bc32b2012-01-12 17:19:43 -0800753 if (mode != MIGRATE_SYNC) {
Andrea Arcangeli11bc82d2011-03-22 16:33:11 -0700754 rc = -EBUSY;
755 goto uncharge;
756 }
757 if (!force)
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -0800758 goto uncharge;
Christoph Lametere24f0b82006-06-23 02:03:51 -0700759 wait_on_page_writeback(page);
760 }
Christoph Lametere24f0b82006-06-23 02:03:51 -0700761 /*
KAMEZAWA Hiroyukidc386d42007-07-26 10:41:07 -0700762 * By try_to_unmap(), page->mapcount goes down to 0 here. In this case,
763 * we cannot notice that anon_vma is freed while we migrates a page.
Hugh Dickins1ce82b62011-01-13 15:47:30 -0800764 * This get_anon_vma() delays freeing anon_vma pointer until the end
KAMEZAWA Hiroyukidc386d42007-07-26 10:41:07 -0700765 * of migration. File cache pages are no problem because of page_lock()
KAMEZAWA Hiroyuki989f89c2007-08-30 23:56:21 -0700766 * File Caches may use write_page() or lock_page() in migration, then,
767 * just care Anon page here.
Christoph Lametere24f0b82006-06-23 02:03:51 -0700768 */
KAMEZAWA Hiroyuki989f89c2007-08-30 23:56:21 -0700769 if (PageAnon(page)) {
Hugh Dickins1ce82b62011-01-13 15:47:30 -0800770 /*
Ingo Molnar4fc3f1d2012-12-02 19:56:50 +0000771 * Only page_lock_anon_vma_read() understands the subtleties of
Hugh Dickins1ce82b62011-01-13 15:47:30 -0800772 * getting a hold on an anon_vma from outside one of its mms.
773 */
Peter Zijlstra746b18d2011-05-24 17:12:10 -0700774 anon_vma = page_get_anon_vma(page);
Hugh Dickins1ce82b62011-01-13 15:47:30 -0800775 if (anon_vma) {
776 /*
Peter Zijlstra746b18d2011-05-24 17:12:10 -0700777 * Anon page
Hugh Dickins1ce82b62011-01-13 15:47:30 -0800778 */
Hugh Dickins1ce82b62011-01-13 15:47:30 -0800779 } else if (PageSwapCache(page)) {
Mel Gorman3fe20112010-05-24 14:32:20 -0700780 /*
781 * We cannot be sure that the anon_vma of an unmapped
782 * swapcache page is safe to use because we don't
783 * know in advance if the VMA that this page belonged
784 * to still exists. If the VMA and others sharing the
785 * data have been freed, then the anon_vma could
786 * already be invalid.
787 *
788 * To avoid this possibility, swapcache pages get
789 * migrated but are not remapped when migration
790 * completes
791 */
792 remap_swapcache = 0;
793 } else {
Hugh Dickins1ce82b62011-01-13 15:47:30 -0800794 goto uncharge;
Mel Gorman3fe20112010-05-24 14:32:20 -0700795 }
KAMEZAWA Hiroyuki989f89c2007-08-30 23:56:21 -0700796 }
Shaohua Li62e1c552008-02-04 22:29:33 -0800797
Rafael Aquinibf6bddf2012-12-11 16:02:42 -0800798 if (unlikely(balloon_page_movable(page))) {
799 /*
800 * A ballooned page does not need any special attention from
801 * physical to virtual reverse mapping procedures.
802 * Skip any attempt to unmap PTEs or to remap swap cache,
803 * in order to avoid burning cycles at rmap level, and perform
804 * the page migration right away (proteced by page lock).
805 */
806 rc = balloon_page_migrate(newpage, page, mode);
807 goto uncharge;
808 }
809
KAMEZAWA Hiroyukidc386d42007-07-26 10:41:07 -0700810 /*
Shaohua Li62e1c552008-02-04 22:29:33 -0800811 * Corner case handling:
812 * 1. When a new swap-cache page is read into, it is added to the LRU
813 * and treated as swapcache but it has no rmap yet.
814 * Calling try_to_unmap() against a page->mapping==NULL page will
815 * trigger a BUG. So handle it here.
816 * 2. An orphaned page (see truncate_complete_page) might have
817 * fs-private metadata. The page can be picked up due to memory
818 * offlining. Everywhere else except page reclaim, the page is
819 * invisible to the vm, so the page can not be migrated. So try to
820 * free the metadata, so the page can be freed.
KAMEZAWA Hiroyukidc386d42007-07-26 10:41:07 -0700821 */
Shaohua Li62e1c552008-02-04 22:29:33 -0800822 if (!page->mapping) {
Hugh Dickins1ce82b62011-01-13 15:47:30 -0800823 VM_BUG_ON(PageAnon(page));
824 if (page_has_private(page)) {
Shaohua Li62e1c552008-02-04 22:29:33 -0800825 try_to_free_buffers(page);
Hugh Dickins1ce82b62011-01-13 15:47:30 -0800826 goto uncharge;
Shaohua Li62e1c552008-02-04 22:29:33 -0800827 }
Shaohua Liabfc3482009-09-21 17:01:19 -0700828 goto skip_unmap;
Shaohua Li62e1c552008-02-04 22:29:33 -0800829 }
830
KAMEZAWA Hiroyukidc386d42007-07-26 10:41:07 -0700831 /* Establish migration ptes or remove ptes */
Andi Kleen14fa31b2009-09-16 11:50:10 +0200832 try_to_unmap(page, TTU_MIGRATION|TTU_IGNORE_MLOCK|TTU_IGNORE_ACCESS);
KAMEZAWA Hiroyukidc386d42007-07-26 10:41:07 -0700833
Shaohua Liabfc3482009-09-21 17:01:19 -0700834skip_unmap:
Christoph Lametere6a15302006-06-25 05:46:49 -0700835 if (!page_mapped(page))
Mel Gormana6bc32b2012-01-12 17:19:43 -0800836 rc = move_to_new_page(newpage, page, remap_swapcache, mode);
Christoph Lametere24f0b82006-06-23 02:03:51 -0700837
Mel Gorman3fe20112010-05-24 14:32:20 -0700838 if (rc && remap_swapcache)
Christoph Lametere24f0b82006-06-23 02:03:51 -0700839 remove_migration_ptes(page, page);
Mel Gorman3f6c8272010-05-24 14:32:17 -0700840
841 /* Drop an anon_vma reference if we took one */
Rik van Riel76545062010-08-09 17:18:41 -0700842 if (anon_vma)
Peter Zijlstra9e601092011-03-22 16:32:46 -0700843 put_anon_vma(anon_vma);
Mel Gorman3f6c8272010-05-24 14:32:17 -0700844
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -0800845uncharge:
Rafael Aquinibf6bddf2012-12-11 16:02:42 -0800846 mem_cgroup_end_migration(mem, page, newpage,
847 (rc == MIGRATEPAGE_SUCCESS ||
848 rc == MIGRATEPAGE_BALLOON_SUCCESS));
Christoph Lametere24f0b82006-06-23 02:03:51 -0700849unlock:
850 unlock_page(page);
Minchan Kim0dabec92011-10-31 17:06:57 -0700851out:
852 return rc;
853}
Christoph Lameter95a402c2006-06-23 02:03:53 -0700854
Minchan Kim0dabec92011-10-31 17:06:57 -0700855/*
856 * Obtain the lock on page, remove all ptes and migrate the page
857 * to the newly allocated page in newpage.
858 */
859static int unmap_and_move(new_page_t get_new_page, unsigned long private,
Mel Gormana6bc32b2012-01-12 17:19:43 -0800860 struct page *page, int force, bool offlining,
861 enum migrate_mode mode)
Minchan Kim0dabec92011-10-31 17:06:57 -0700862{
863 int rc = 0;
864 int *result = NULL;
865 struct page *newpage = get_new_page(page, private, &result);
866
867 if (!newpage)
868 return -ENOMEM;
869
870 if (page_count(page) == 1) {
871 /* page was freed from under us. So we are done. */
872 goto out;
873 }
874
875 if (unlikely(PageTransHuge(page)))
876 if (unlikely(split_huge_page(page)))
877 goto out;
878
Mel Gormana6bc32b2012-01-12 17:19:43 -0800879 rc = __unmap_and_move(page, newpage, force, offlining, mode);
Rafael Aquinibf6bddf2012-12-11 16:02:42 -0800880
881 if (unlikely(rc == MIGRATEPAGE_BALLOON_SUCCESS)) {
882 /*
883 * A ballooned page has been migrated already.
884 * Now, it's the time to wrap-up counters,
885 * handle the page back to Buddy and return.
886 */
887 dec_zone_page_state(page, NR_ISOLATED_ANON +
888 page_is_file_cache(page));
889 balloon_page_free(page);
890 return MIGRATEPAGE_SUCCESS;
891 }
Minchan Kim0dabec92011-10-31 17:06:57 -0700892out:
Christoph Lametere24f0b82006-06-23 02:03:51 -0700893 if (rc != -EAGAIN) {
Minchan Kim0dabec92011-10-31 17:06:57 -0700894 /*
895 * A page that has been migrated has all references
896 * removed and will be freed. A page that has not been
897 * migrated will have kepts its references and be
898 * restored.
899 */
900 list_del(&page->lru);
KOSAKI Motohiroa7312862009-09-21 17:01:37 -0700901 dec_zone_page_state(page, NR_ISOLATED_ANON +
Johannes Weiner6c0b1352009-09-21 17:02:59 -0700902 page_is_file_cache(page));
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700903 putback_lru_page(page);
Christoph Lametere24f0b82006-06-23 02:03:51 -0700904 }
Christoph Lameter95a402c2006-06-23 02:03:53 -0700905 /*
906 * Move the new page to the LRU. If migration was not successful
907 * then this will free the page.
908 */
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700909 putback_lru_page(newpage);
Christoph Lameter742755a2006-06-23 02:03:55 -0700910 if (result) {
911 if (rc)
912 *result = rc;
913 else
914 *result = page_to_nid(newpage);
915 }
Christoph Lametere24f0b82006-06-23 02:03:51 -0700916 return rc;
917}
918
919/*
Naoya Horiguchi290408d2010-09-08 10:19:35 +0900920 * Counterpart of unmap_and_move_page() for hugepage migration.
921 *
922 * This function doesn't wait the completion of hugepage I/O
923 * because there is no race between I/O and migration for hugepage.
924 * Note that currently hugepage I/O occurs only in direct I/O
925 * where no lock is held and PG_writeback is irrelevant,
926 * and writeback status of all subpages are counted in the reference
927 * count of the head page (i.e. if all subpages of a 2MB hugepage are
928 * under direct I/O, the reference of the head page is 512 and a bit more.)
929 * This means that when we try to migrate hugepage whose subpages are
930 * doing direct I/O, some references remain after try_to_unmap() and
931 * hugepage migration fails without data corruption.
932 *
933 * There is also no race when direct I/O is issued on the page under migration,
934 * because then pte is replaced with migration swap entry and direct I/O code
935 * will wait in the page fault for migration to complete.
936 */
937static int unmap_and_move_huge_page(new_page_t get_new_page,
938 unsigned long private, struct page *hpage,
Mel Gormana6bc32b2012-01-12 17:19:43 -0800939 int force, bool offlining,
940 enum migrate_mode mode)
Naoya Horiguchi290408d2010-09-08 10:19:35 +0900941{
942 int rc = 0;
943 int *result = NULL;
944 struct page *new_hpage = get_new_page(hpage, private, &result);
Naoya Horiguchi290408d2010-09-08 10:19:35 +0900945 struct anon_vma *anon_vma = NULL;
946
947 if (!new_hpage)
948 return -ENOMEM;
949
950 rc = -EAGAIN;
951
952 if (!trylock_page(hpage)) {
Mel Gormana6bc32b2012-01-12 17:19:43 -0800953 if (!force || mode != MIGRATE_SYNC)
Naoya Horiguchi290408d2010-09-08 10:19:35 +0900954 goto out;
955 lock_page(hpage);
956 }
957
Peter Zijlstra746b18d2011-05-24 17:12:10 -0700958 if (PageAnon(hpage))
959 anon_vma = page_get_anon_vma(hpage);
Naoya Horiguchi290408d2010-09-08 10:19:35 +0900960
961 try_to_unmap(hpage, TTU_MIGRATION|TTU_IGNORE_MLOCK|TTU_IGNORE_ACCESS);
962
963 if (!page_mapped(hpage))
Mel Gormana6bc32b2012-01-12 17:19:43 -0800964 rc = move_to_new_page(new_hpage, hpage, 1, mode);
Naoya Horiguchi290408d2010-09-08 10:19:35 +0900965
966 if (rc)
967 remove_migration_ptes(hpage, hpage);
968
Hugh Dickinsfd4a4662011-01-13 15:47:31 -0800969 if (anon_vma)
Peter Zijlstra9e601092011-03-22 16:32:46 -0700970 put_anon_vma(anon_vma);
Aneesh Kumar K.V8e6ac7f2012-07-31 16:42:27 -0700971
972 if (!rc)
973 hugetlb_cgroup_migrate(hpage, new_hpage);
974
Naoya Horiguchi290408d2010-09-08 10:19:35 +0900975 unlock_page(hpage);
Hillf Danton09761332011-12-08 14:34:20 -0800976out:
Naoya Horiguchi290408d2010-09-08 10:19:35 +0900977 put_page(new_hpage);
Naoya Horiguchi290408d2010-09-08 10:19:35 +0900978 if (result) {
979 if (rc)
980 *result = rc;
981 else
982 *result = page_to_nid(new_hpage);
983 }
984 return rc;
985}
986
987/*
Christoph Lameterb20a3502006-03-22 00:09:12 -0800988 * migrate_pages
989 *
Christoph Lameter95a402c2006-06-23 02:03:53 -0700990 * The function takes one list of pages to migrate and a function
991 * that determines from the page to be migrated and the private data
992 * the target of the move and allocates the page.
Christoph Lameterb20a3502006-03-22 00:09:12 -0800993 *
994 * The function returns after 10 attempts or if no pages
995 * are movable anymore because to has become empty
Minchan Kimcf608ac2010-10-26 14:21:29 -0700996 * or no retryable pages exist anymore.
997 * Caller should call putback_lru_pages to return pages to the LRU
Minchan Kim28bd6572011-01-25 15:07:26 -0800998 * or free list only if ret != 0.
Christoph Lameterb20a3502006-03-22 00:09:12 -0800999 *
Christoph Lameter95a402c2006-06-23 02:03:53 -07001000 * Return: Number of pages not migrated or error code.
Christoph Lameterb20a3502006-03-22 00:09:12 -08001001 */
Christoph Lameter95a402c2006-06-23 02:03:53 -07001002int migrate_pages(struct list_head *from,
Mel Gorman7f0f2492011-01-13 15:45:58 -08001003 new_page_t get_new_page, unsigned long private, bool offlining,
Mel Gorman7b2a2d42012-10-19 14:07:31 +01001004 enum migrate_mode mode, int reason)
Christoph Lameterb20a3502006-03-22 00:09:12 -08001005{
Christoph Lametere24f0b82006-06-23 02:03:51 -07001006 int retry = 1;
Christoph Lameterb20a3502006-03-22 00:09:12 -08001007 int nr_failed = 0;
Mel Gorman5647bc22012-10-19 10:46:20 +01001008 int nr_succeeded = 0;
Christoph Lameterb20a3502006-03-22 00:09:12 -08001009 int pass = 0;
1010 struct page *page;
1011 struct page *page2;
1012 int swapwrite = current->flags & PF_SWAPWRITE;
1013 int rc;
1014
1015 if (!swapwrite)
1016 current->flags |= PF_SWAPWRITE;
1017
Christoph Lametere24f0b82006-06-23 02:03:51 -07001018 for(pass = 0; pass < 10 && retry; pass++) {
1019 retry = 0;
Christoph Lameterb20a3502006-03-22 00:09:12 -08001020
Christoph Lametere24f0b82006-06-23 02:03:51 -07001021 list_for_each_entry_safe(page, page2, from, lru) {
Christoph Lametere24f0b82006-06-23 02:03:51 -07001022 cond_resched();
Christoph Lameterb20a3502006-03-22 00:09:12 -08001023
Christoph Lameter95a402c2006-06-23 02:03:53 -07001024 rc = unmap_and_move(get_new_page, private,
Mel Gorman77f1fe62011-01-13 15:45:57 -08001025 page, pass > 2, offlining,
Mel Gormana6bc32b2012-01-12 17:19:43 -08001026 mode);
Christoph Lameterb20a3502006-03-22 00:09:12 -08001027
Christoph Lametere24f0b82006-06-23 02:03:51 -07001028 switch(rc) {
Christoph Lameter95a402c2006-06-23 02:03:53 -07001029 case -ENOMEM:
1030 goto out;
Christoph Lametere24f0b82006-06-23 02:03:51 -07001031 case -EAGAIN:
Christoph Lameter2d1db3b2006-06-23 02:03:33 -07001032 retry++;
Christoph Lametere24f0b82006-06-23 02:03:51 -07001033 break;
Rafael Aquini78bd5202012-12-11 16:02:31 -08001034 case MIGRATEPAGE_SUCCESS:
Mel Gorman5647bc22012-10-19 10:46:20 +01001035 nr_succeeded++;
Christoph Lametere24f0b82006-06-23 02:03:51 -07001036 break;
1037 default:
Christoph Lameter2d1db3b2006-06-23 02:03:33 -07001038 /* Permanent failure */
Christoph Lameter2d1db3b2006-06-23 02:03:33 -07001039 nr_failed++;
Christoph Lametere24f0b82006-06-23 02:03:51 -07001040 break;
Christoph Lameter2d1db3b2006-06-23 02:03:33 -07001041 }
Christoph Lameterb20a3502006-03-22 00:09:12 -08001042 }
1043 }
Rafael Aquini78bd5202012-12-11 16:02:31 -08001044 rc = nr_failed + retry;
Christoph Lameter95a402c2006-06-23 02:03:53 -07001045out:
Mel Gorman5647bc22012-10-19 10:46:20 +01001046 if (nr_succeeded)
1047 count_vm_events(PGMIGRATE_SUCCESS, nr_succeeded);
1048 if (nr_failed)
1049 count_vm_events(PGMIGRATE_FAIL, nr_failed);
Mel Gorman7b2a2d42012-10-19 14:07:31 +01001050 trace_mm_migrate_pages(nr_succeeded, nr_failed, mode, reason);
1051
Christoph Lameterb20a3502006-03-22 00:09:12 -08001052 if (!swapwrite)
1053 current->flags &= ~PF_SWAPWRITE;
1054
Rafael Aquini78bd5202012-12-11 16:02:31 -08001055 return rc;
Christoph Lameterb20a3502006-03-22 00:09:12 -08001056}
1057
Aneesh Kumar K.V189ebff2012-07-31 16:42:06 -07001058int migrate_huge_page(struct page *hpage, new_page_t get_new_page,
1059 unsigned long private, bool offlining,
1060 enum migrate_mode mode)
Naoya Horiguchi290408d2010-09-08 10:19:35 +09001061{
Aneesh Kumar K.V189ebff2012-07-31 16:42:06 -07001062 int pass, rc;
Naoya Horiguchi290408d2010-09-08 10:19:35 +09001063
Aneesh Kumar K.V189ebff2012-07-31 16:42:06 -07001064 for (pass = 0; pass < 10; pass++) {
1065 rc = unmap_and_move_huge_page(get_new_page,
1066 private, hpage, pass > 2, offlining,
1067 mode);
1068 switch (rc) {
1069 case -ENOMEM:
1070 goto out;
1071 case -EAGAIN:
1072 /* try again */
Naoya Horiguchi290408d2010-09-08 10:19:35 +09001073 cond_resched();
Aneesh Kumar K.V189ebff2012-07-31 16:42:06 -07001074 break;
Rafael Aquini78bd5202012-12-11 16:02:31 -08001075 case MIGRATEPAGE_SUCCESS:
Aneesh Kumar K.V189ebff2012-07-31 16:42:06 -07001076 goto out;
1077 default:
1078 rc = -EIO;
1079 goto out;
Naoya Horiguchi290408d2010-09-08 10:19:35 +09001080 }
1081 }
Naoya Horiguchi290408d2010-09-08 10:19:35 +09001082out:
Aneesh Kumar K.V189ebff2012-07-31 16:42:06 -07001083 return rc;
Naoya Horiguchi290408d2010-09-08 10:19:35 +09001084}
1085
Christoph Lameter742755a2006-06-23 02:03:55 -07001086#ifdef CONFIG_NUMA
1087/*
1088 * Move a list of individual pages
1089 */
1090struct page_to_node {
1091 unsigned long addr;
1092 struct page *page;
1093 int node;
1094 int status;
1095};
1096
1097static struct page *new_page_node(struct page *p, unsigned long private,
1098 int **result)
1099{
1100 struct page_to_node *pm = (struct page_to_node *)private;
1101
1102 while (pm->node != MAX_NUMNODES && pm->page != p)
1103 pm++;
1104
1105 if (pm->node == MAX_NUMNODES)
1106 return NULL;
1107
1108 *result = &pm->status;
1109
Mel Gorman6484eb32009-06-16 15:31:54 -07001110 return alloc_pages_exact_node(pm->node,
Mel Gorman769848c2007-07-17 04:03:05 -07001111 GFP_HIGHUSER_MOVABLE | GFP_THISNODE, 0);
Christoph Lameter742755a2006-06-23 02:03:55 -07001112}
1113
1114/*
1115 * Move a set of pages as indicated in the pm array. The addr
1116 * field must be set to the virtual address of the page to be moved
1117 * and the node number must contain a valid target node.
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001118 * The pm array ends with node = MAX_NUMNODES.
Christoph Lameter742755a2006-06-23 02:03:55 -07001119 */
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001120static int do_move_page_to_node_array(struct mm_struct *mm,
1121 struct page_to_node *pm,
1122 int migrate_all)
Christoph Lameter742755a2006-06-23 02:03:55 -07001123{
1124 int err;
1125 struct page_to_node *pp;
1126 LIST_HEAD(pagelist);
1127
1128 down_read(&mm->mmap_sem);
1129
1130 /*
1131 * Build a list of pages to migrate
1132 */
Christoph Lameter742755a2006-06-23 02:03:55 -07001133 for (pp = pm; pp->node != MAX_NUMNODES; pp++) {
1134 struct vm_area_struct *vma;
1135 struct page *page;
1136
Christoph Lameter742755a2006-06-23 02:03:55 -07001137 err = -EFAULT;
1138 vma = find_vma(mm, pp->addr);
Gleb Natapov70384dc2010-10-26 14:22:07 -07001139 if (!vma || pp->addr < vma->vm_start || !vma_migratable(vma))
Christoph Lameter742755a2006-06-23 02:03:55 -07001140 goto set_status;
1141
Andrea Arcangeli500d65d2011-01-13 15:46:55 -08001142 page = follow_page(vma, pp->addr, FOLL_GET|FOLL_SPLIT);
Linus Torvalds89f5b7d2008-06-20 11:18:25 -07001143
1144 err = PTR_ERR(page);
1145 if (IS_ERR(page))
1146 goto set_status;
1147
Christoph Lameter742755a2006-06-23 02:03:55 -07001148 err = -ENOENT;
1149 if (!page)
1150 goto set_status;
1151
Hugh Dickins62b61f62009-12-14 17:59:33 -08001152 /* Use PageReserved to check for zero page */
1153 if (PageReserved(page) || PageKsm(page))
Christoph Lameter742755a2006-06-23 02:03:55 -07001154 goto put_and_set;
1155
1156 pp->page = page;
1157 err = page_to_nid(page);
1158
1159 if (err == pp->node)
1160 /*
1161 * Node already in the right place
1162 */
1163 goto put_and_set;
1164
1165 err = -EACCES;
1166 if (page_mapcount(page) > 1 &&
1167 !migrate_all)
1168 goto put_and_set;
1169
Nick Piggin62695a82008-10-18 20:26:09 -07001170 err = isolate_lru_page(page);
KOSAKI Motohiro6d9c2852009-12-14 17:58:11 -08001171 if (!err) {
Nick Piggin62695a82008-10-18 20:26:09 -07001172 list_add_tail(&page->lru, &pagelist);
KOSAKI Motohiro6d9c2852009-12-14 17:58:11 -08001173 inc_zone_page_state(page, NR_ISOLATED_ANON +
1174 page_is_file_cache(page));
1175 }
Christoph Lameter742755a2006-06-23 02:03:55 -07001176put_and_set:
1177 /*
1178 * Either remove the duplicate refcount from
1179 * isolate_lru_page() or drop the page ref if it was
1180 * not isolated.
1181 */
1182 put_page(page);
1183set_status:
1184 pp->status = err;
1185 }
1186
Brice Gogline78bbfa2008-10-18 20:27:15 -07001187 err = 0;
Minchan Kimcf608ac2010-10-26 14:21:29 -07001188 if (!list_empty(&pagelist)) {
Christoph Lameter742755a2006-06-23 02:03:55 -07001189 err = migrate_pages(&pagelist, new_page_node,
Mel Gorman7b2a2d42012-10-19 14:07:31 +01001190 (unsigned long)pm, 0, MIGRATE_SYNC,
1191 MR_SYSCALL);
Minchan Kimcf608ac2010-10-26 14:21:29 -07001192 if (err)
1193 putback_lru_pages(&pagelist);
1194 }
Christoph Lameter742755a2006-06-23 02:03:55 -07001195
1196 up_read(&mm->mmap_sem);
1197 return err;
1198}
1199
1200/*
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001201 * Migrate an array of page address onto an array of nodes and fill
1202 * the corresponding array of status.
1203 */
Christoph Lameter3268c632012-03-21 16:34:06 -07001204static int do_pages_move(struct mm_struct *mm, nodemask_t task_nodes,
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001205 unsigned long nr_pages,
1206 const void __user * __user *pages,
1207 const int __user *nodes,
1208 int __user *status, int flags)
1209{
Brice Goglin3140a222009-01-06 14:38:57 -08001210 struct page_to_node *pm;
Brice Goglin3140a222009-01-06 14:38:57 -08001211 unsigned long chunk_nr_pages;
1212 unsigned long chunk_start;
1213 int err;
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001214
Brice Goglin3140a222009-01-06 14:38:57 -08001215 err = -ENOMEM;
1216 pm = (struct page_to_node *)__get_free_page(GFP_KERNEL);
1217 if (!pm)
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001218 goto out;
Brice Goglin35282a22009-06-16 15:32:43 -07001219
1220 migrate_prep();
1221
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001222 /*
Brice Goglin3140a222009-01-06 14:38:57 -08001223 * Store a chunk of page_to_node array in a page,
1224 * but keep the last one as a marker
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001225 */
Brice Goglin3140a222009-01-06 14:38:57 -08001226 chunk_nr_pages = (PAGE_SIZE / sizeof(struct page_to_node)) - 1;
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001227
Brice Goglin3140a222009-01-06 14:38:57 -08001228 for (chunk_start = 0;
1229 chunk_start < nr_pages;
1230 chunk_start += chunk_nr_pages) {
1231 int j;
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001232
Brice Goglin3140a222009-01-06 14:38:57 -08001233 if (chunk_start + chunk_nr_pages > nr_pages)
1234 chunk_nr_pages = nr_pages - chunk_start;
1235
1236 /* fill the chunk pm with addrs and nodes from user-space */
1237 for (j = 0; j < chunk_nr_pages; j++) {
1238 const void __user *p;
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001239 int node;
1240
Brice Goglin3140a222009-01-06 14:38:57 -08001241 err = -EFAULT;
1242 if (get_user(p, pages + j + chunk_start))
1243 goto out_pm;
1244 pm[j].addr = (unsigned long) p;
1245
1246 if (get_user(node, nodes + j + chunk_start))
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001247 goto out_pm;
1248
1249 err = -ENODEV;
Linus Torvalds6f5a55f2010-02-05 16:16:50 -08001250 if (node < 0 || node >= MAX_NUMNODES)
1251 goto out_pm;
1252
Lai Jiangshan389162c2012-12-12 13:51:30 -08001253 if (!node_state(node, N_MEMORY))
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001254 goto out_pm;
1255
1256 err = -EACCES;
1257 if (!node_isset(node, task_nodes))
1258 goto out_pm;
1259
Brice Goglin3140a222009-01-06 14:38:57 -08001260 pm[j].node = node;
1261 }
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001262
Brice Goglin3140a222009-01-06 14:38:57 -08001263 /* End marker for this chunk */
1264 pm[chunk_nr_pages].node = MAX_NUMNODES;
1265
1266 /* Migrate this chunk */
1267 err = do_move_page_to_node_array(mm, pm,
1268 flags & MPOL_MF_MOVE_ALL);
1269 if (err < 0)
1270 goto out_pm;
1271
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001272 /* Return status information */
Brice Goglin3140a222009-01-06 14:38:57 -08001273 for (j = 0; j < chunk_nr_pages; j++)
1274 if (put_user(pm[j].status, status + j + chunk_start)) {
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001275 err = -EFAULT;
Brice Goglin3140a222009-01-06 14:38:57 -08001276 goto out_pm;
1277 }
1278 }
1279 err = 0;
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001280
1281out_pm:
Brice Goglin3140a222009-01-06 14:38:57 -08001282 free_page((unsigned long)pm);
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001283out:
1284 return err;
1285}
1286
1287/*
Brice Goglin2f007e72008-10-18 20:27:16 -07001288 * Determine the nodes of an array of pages and store it in an array of status.
Christoph Lameter742755a2006-06-23 02:03:55 -07001289 */
Brice Goglin80bba122008-12-09 13:14:23 -08001290static void do_pages_stat_array(struct mm_struct *mm, unsigned long nr_pages,
1291 const void __user **pages, int *status)
Christoph Lameter742755a2006-06-23 02:03:55 -07001292{
Brice Goglin2f007e72008-10-18 20:27:16 -07001293 unsigned long i;
Brice Goglin2f007e72008-10-18 20:27:16 -07001294
Christoph Lameter742755a2006-06-23 02:03:55 -07001295 down_read(&mm->mmap_sem);
1296
Brice Goglin2f007e72008-10-18 20:27:16 -07001297 for (i = 0; i < nr_pages; i++) {
Brice Goglin80bba122008-12-09 13:14:23 -08001298 unsigned long addr = (unsigned long)(*pages);
Christoph Lameter742755a2006-06-23 02:03:55 -07001299 struct vm_area_struct *vma;
1300 struct page *page;
KOSAKI Motohiroc095adb2008-12-16 16:06:43 +09001301 int err = -EFAULT;
Brice Goglin2f007e72008-10-18 20:27:16 -07001302
1303 vma = find_vma(mm, addr);
Gleb Natapov70384dc2010-10-26 14:22:07 -07001304 if (!vma || addr < vma->vm_start)
Christoph Lameter742755a2006-06-23 02:03:55 -07001305 goto set_status;
1306
Brice Goglin2f007e72008-10-18 20:27:16 -07001307 page = follow_page(vma, addr, 0);
Linus Torvalds89f5b7d2008-06-20 11:18:25 -07001308
1309 err = PTR_ERR(page);
1310 if (IS_ERR(page))
1311 goto set_status;
1312
Christoph Lameter742755a2006-06-23 02:03:55 -07001313 err = -ENOENT;
1314 /* Use PageReserved to check for zero page */
Hugh Dickins62b61f62009-12-14 17:59:33 -08001315 if (!page || PageReserved(page) || PageKsm(page))
Christoph Lameter742755a2006-06-23 02:03:55 -07001316 goto set_status;
1317
1318 err = page_to_nid(page);
1319set_status:
Brice Goglin80bba122008-12-09 13:14:23 -08001320 *status = err;
1321
1322 pages++;
1323 status++;
1324 }
1325
1326 up_read(&mm->mmap_sem);
1327}
1328
1329/*
1330 * Determine the nodes of a user array of pages and store it in
1331 * a user array of status.
1332 */
1333static int do_pages_stat(struct mm_struct *mm, unsigned long nr_pages,
1334 const void __user * __user *pages,
1335 int __user *status)
1336{
1337#define DO_PAGES_STAT_CHUNK_NR 16
1338 const void __user *chunk_pages[DO_PAGES_STAT_CHUNK_NR];
1339 int chunk_status[DO_PAGES_STAT_CHUNK_NR];
Brice Goglin80bba122008-12-09 13:14:23 -08001340
H. Peter Anvin87b8d1a2010-02-18 16:13:40 -08001341 while (nr_pages) {
1342 unsigned long chunk_nr;
Brice Goglin80bba122008-12-09 13:14:23 -08001343
H. Peter Anvin87b8d1a2010-02-18 16:13:40 -08001344 chunk_nr = nr_pages;
1345 if (chunk_nr > DO_PAGES_STAT_CHUNK_NR)
1346 chunk_nr = DO_PAGES_STAT_CHUNK_NR;
1347
1348 if (copy_from_user(chunk_pages, pages, chunk_nr * sizeof(*chunk_pages)))
1349 break;
Brice Goglin80bba122008-12-09 13:14:23 -08001350
1351 do_pages_stat_array(mm, chunk_nr, chunk_pages, chunk_status);
1352
H. Peter Anvin87b8d1a2010-02-18 16:13:40 -08001353 if (copy_to_user(status, chunk_status, chunk_nr * sizeof(*status)))
1354 break;
Christoph Lameter742755a2006-06-23 02:03:55 -07001355
H. Peter Anvin87b8d1a2010-02-18 16:13:40 -08001356 pages += chunk_nr;
1357 status += chunk_nr;
1358 nr_pages -= chunk_nr;
1359 }
1360 return nr_pages ? -EFAULT : 0;
Christoph Lameter742755a2006-06-23 02:03:55 -07001361}
1362
1363/*
1364 * Move a list of pages in the address space of the currently executing
1365 * process.
1366 */
Heiko Carstens938bb9f2009-01-14 14:14:30 +01001367SYSCALL_DEFINE6(move_pages, pid_t, pid, unsigned long, nr_pages,
1368 const void __user * __user *, pages,
1369 const int __user *, nodes,
1370 int __user *, status, int, flags)
Christoph Lameter742755a2006-06-23 02:03:55 -07001371{
David Howellsc69e8d92008-11-14 10:39:19 +11001372 const struct cred *cred = current_cred(), *tcred;
Christoph Lameter742755a2006-06-23 02:03:55 -07001373 struct task_struct *task;
Christoph Lameter742755a2006-06-23 02:03:55 -07001374 struct mm_struct *mm;
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001375 int err;
Christoph Lameter3268c632012-03-21 16:34:06 -07001376 nodemask_t task_nodes;
Christoph Lameter742755a2006-06-23 02:03:55 -07001377
1378 /* Check flags */
1379 if (flags & ~(MPOL_MF_MOVE|MPOL_MF_MOVE_ALL))
1380 return -EINVAL;
1381
1382 if ((flags & MPOL_MF_MOVE_ALL) && !capable(CAP_SYS_NICE))
1383 return -EPERM;
1384
1385 /* Find the mm_struct */
Greg Thelena879bf52011-02-25 14:44:13 -08001386 rcu_read_lock();
Pavel Emelyanov228ebcb2007-10-18 23:40:16 -07001387 task = pid ? find_task_by_vpid(pid) : current;
Christoph Lameter742755a2006-06-23 02:03:55 -07001388 if (!task) {
Greg Thelena879bf52011-02-25 14:44:13 -08001389 rcu_read_unlock();
Christoph Lameter742755a2006-06-23 02:03:55 -07001390 return -ESRCH;
1391 }
Christoph Lameter3268c632012-03-21 16:34:06 -07001392 get_task_struct(task);
Christoph Lameter742755a2006-06-23 02:03:55 -07001393
1394 /*
1395 * Check if this process has the right to modify the specified
1396 * process. The right exists if the process has administrative
1397 * capabilities, superuser privileges or the same
1398 * userid as the target process.
1399 */
David Howellsc69e8d92008-11-14 10:39:19 +11001400 tcred = __task_cred(task);
Eric W. Biedermanb38a86e2012-03-12 15:48:24 -07001401 if (!uid_eq(cred->euid, tcred->suid) && !uid_eq(cred->euid, tcred->uid) &&
1402 !uid_eq(cred->uid, tcred->suid) && !uid_eq(cred->uid, tcred->uid) &&
Christoph Lameter742755a2006-06-23 02:03:55 -07001403 !capable(CAP_SYS_NICE)) {
David Howellsc69e8d92008-11-14 10:39:19 +11001404 rcu_read_unlock();
Christoph Lameter742755a2006-06-23 02:03:55 -07001405 err = -EPERM;
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001406 goto out;
Christoph Lameter742755a2006-06-23 02:03:55 -07001407 }
David Howellsc69e8d92008-11-14 10:39:19 +11001408 rcu_read_unlock();
Christoph Lameter742755a2006-06-23 02:03:55 -07001409
David Quigley86c3a762006-06-23 02:04:02 -07001410 err = security_task_movememory(task);
1411 if (err)
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001412 goto out;
David Quigley86c3a762006-06-23 02:04:02 -07001413
Christoph Lameter3268c632012-03-21 16:34:06 -07001414 task_nodes = cpuset_mems_allowed(task);
1415 mm = get_task_mm(task);
1416 put_task_struct(task);
1417
Sasha Levin6e8b09e2012-04-25 16:01:53 -07001418 if (!mm)
1419 return -EINVAL;
1420
1421 if (nodes)
1422 err = do_pages_move(mm, task_nodes, nr_pages, pages,
1423 nodes, status, flags);
1424 else
1425 err = do_pages_stat(mm, nr_pages, pages, status);
Christoph Lameter3268c632012-03-21 16:34:06 -07001426
1427 mmput(mm);
1428 return err;
David Quigley86c3a762006-06-23 02:04:02 -07001429
Christoph Lameter742755a2006-06-23 02:03:55 -07001430out:
Christoph Lameter3268c632012-03-21 16:34:06 -07001431 put_task_struct(task);
Christoph Lameter742755a2006-06-23 02:03:55 -07001432 return err;
1433}
Christoph Lameter742755a2006-06-23 02:03:55 -07001434
Christoph Lameter7b2259b2006-06-25 05:46:48 -07001435/*
1436 * Call migration functions in the vma_ops that may prepare
1437 * memory in a vm for migration. migration functions may perform
1438 * the migration for vmas that do not have an underlying page struct.
1439 */
1440int migrate_vmas(struct mm_struct *mm, const nodemask_t *to,
1441 const nodemask_t *from, unsigned long flags)
1442{
1443 struct vm_area_struct *vma;
1444 int err = 0;
1445
Daisuke Nishimura1001c9f2009-02-11 13:04:18 -08001446 for (vma = mm->mmap; vma && !err; vma = vma->vm_next) {
Christoph Lameter7b2259b2006-06-25 05:46:48 -07001447 if (vma->vm_ops && vma->vm_ops->migrate) {
1448 err = vma->vm_ops->migrate(vma, to, from, flags);
1449 if (err)
1450 break;
1451 }
1452 }
1453 return err;
1454}
Peter Zijlstra7039e1d2012-10-25 14:16:34 +02001455
1456#ifdef CONFIG_NUMA_BALANCING
1457/*
1458 * Returns true if this is a safe migration target node for misplaced NUMA
1459 * pages. Currently it only checks the watermarks which crude
1460 */
1461static bool migrate_balanced_pgdat(struct pglist_data *pgdat,
1462 int nr_migrate_pages)
1463{
1464 int z;
1465 for (z = pgdat->nr_zones - 1; z >= 0; z--) {
1466 struct zone *zone = pgdat->node_zones + z;
1467
1468 if (!populated_zone(zone))
1469 continue;
1470
1471 if (zone->all_unreclaimable)
1472 continue;
1473
1474 /* Avoid waking kswapd by allocating pages_to_migrate pages. */
1475 if (!zone_watermark_ok(zone, 0,
1476 high_wmark_pages(zone) +
1477 nr_migrate_pages,
1478 0, 0))
1479 continue;
1480 return true;
1481 }
1482 return false;
1483}
1484
1485static struct page *alloc_misplaced_dst_page(struct page *page,
1486 unsigned long data,
1487 int **result)
1488{
1489 int nid = (int) data;
1490 struct page *newpage;
1491
1492 newpage = alloc_pages_exact_node(nid,
1493 (GFP_HIGHUSER_MOVABLE | GFP_THISNODE |
1494 __GFP_NOMEMALLOC | __GFP_NORETRY |
1495 __GFP_NOWARN) &
1496 ~GFP_IOFS, 0);
Hillf Dantonbac03822012-11-27 14:46:24 +00001497 if (newpage)
1498 page_xchg_last_nid(newpage, page_last_nid(page));
1499
Peter Zijlstra7039e1d2012-10-25 14:16:34 +02001500 return newpage;
1501}
1502
1503/*
Mel Gormana8f60772012-11-14 21:41:46 +00001504 * page migration rate limiting control.
1505 * Do not migrate more than @pages_to_migrate in a @migrate_interval_millisecs
1506 * window of time. Default here says do not migrate more than 1280M per second.
Mel Gormane14808b2012-11-19 10:59:15 +00001507 * If a node is rate-limited then PTE NUMA updates are also rate-limited. However
1508 * as it is faults that reset the window, pte updates will happen unconditionally
1509 * if there has not been a fault since @pteupdate_interval_millisecs after the
1510 * throttle window closed.
Mel Gormana8f60772012-11-14 21:41:46 +00001511 */
1512static unsigned int migrate_interval_millisecs __read_mostly = 100;
Mel Gormane14808b2012-11-19 10:59:15 +00001513static unsigned int pteupdate_interval_millisecs __read_mostly = 1000;
Mel Gormana8f60772012-11-14 21:41:46 +00001514static unsigned int ratelimit_pages __read_mostly = 128 << (20 - PAGE_SHIFT);
1515
Mel Gormane14808b2012-11-19 10:59:15 +00001516/* Returns true if NUMA migration is currently rate limited */
1517bool migrate_ratelimited(int node)
1518{
1519 pg_data_t *pgdat = NODE_DATA(node);
1520
1521 if (time_after(jiffies, pgdat->numabalancing_migrate_next_window +
1522 msecs_to_jiffies(pteupdate_interval_millisecs)))
1523 return false;
1524
1525 if (pgdat->numabalancing_migrate_nr_pages < ratelimit_pages)
1526 return false;
1527
1528 return true;
1529}
1530
Mel Gormanb32967f2012-11-19 12:35:47 +00001531/* Returns true if the node is migrate rate-limited after the update */
Mel Gormand28d43352012-11-29 09:24:36 +00001532bool numamigrate_update_ratelimit(pg_data_t *pgdat, unsigned long nr_pages)
Mel Gormanb32967f2012-11-19 12:35:47 +00001533{
1534 bool rate_limited = false;
1535
1536 /*
1537 * Rate-limit the amount of data that is being migrated to a node.
1538 * Optimal placement is no good if the memory bus is saturated and
1539 * all the time is being spent migrating!
1540 */
1541 spin_lock(&pgdat->numabalancing_migrate_lock);
1542 if (time_after(jiffies, pgdat->numabalancing_migrate_next_window)) {
1543 pgdat->numabalancing_migrate_nr_pages = 0;
1544 pgdat->numabalancing_migrate_next_window = jiffies +
1545 msecs_to_jiffies(migrate_interval_millisecs);
1546 }
1547 if (pgdat->numabalancing_migrate_nr_pages > ratelimit_pages)
1548 rate_limited = true;
1549 else
Mel Gormand28d43352012-11-29 09:24:36 +00001550 pgdat->numabalancing_migrate_nr_pages += nr_pages;
Mel Gormanb32967f2012-11-19 12:35:47 +00001551 spin_unlock(&pgdat->numabalancing_migrate_lock);
1552
1553 return rate_limited;
1554}
1555
1556int numamigrate_isolate_page(pg_data_t *pgdat, struct page *page)
1557{
1558 int ret = 0;
1559
1560 /* Avoid migrating to a node that is nearly full */
1561 if (migrate_balanced_pgdat(pgdat, 1)) {
1562 int page_lru;
1563
1564 if (isolate_lru_page(page)) {
1565 put_page(page);
1566 return 0;
1567 }
1568
1569 /* Page is isolated */
1570 ret = 1;
1571 page_lru = page_is_file_cache(page);
1572 if (!PageTransHuge(page))
1573 inc_zone_page_state(page, NR_ISOLATED_ANON + page_lru);
1574 else
1575 mod_zone_page_state(page_zone(page),
1576 NR_ISOLATED_ANON + page_lru,
1577 HPAGE_PMD_NR);
1578 }
1579
1580 /*
1581 * Page is either isolated or there is not enough space on the target
1582 * node. If isolated, then it has taken a reference count and the
1583 * callers reference can be safely dropped without the page
1584 * disappearing underneath us during migration. Otherwise the page is
1585 * not to be migrated but the callers reference should still be
1586 * dropped so it does not leak.
1587 */
1588 put_page(page);
1589
1590 return ret;
1591}
1592
Mel Gormana8f60772012-11-14 21:41:46 +00001593/*
Peter Zijlstra7039e1d2012-10-25 14:16:34 +02001594 * Attempt to migrate a misplaced page to the specified destination
1595 * node. Caller is expected to have an elevated reference count on
1596 * the page that will be dropped by this function before returning.
1597 */
1598int migrate_misplaced_page(struct page *page, int node)
1599{
Mel Gormana8f60772012-11-14 21:41:46 +00001600 pg_data_t *pgdat = NODE_DATA(node);
Peter Zijlstra7039e1d2012-10-25 14:16:34 +02001601 int isolated = 0;
Mel Gormanb32967f2012-11-19 12:35:47 +00001602 int nr_remaining;
Peter Zijlstra7039e1d2012-10-25 14:16:34 +02001603 LIST_HEAD(migratepages);
1604
1605 /*
1606 * Don't migrate pages that are mapped in multiple processes.
1607 * TODO: Handle false sharing detection instead of this hammer
1608 */
1609 if (page_mapcount(page) != 1) {
1610 put_page(page);
1611 goto out;
1612 }
1613
Mel Gormana8f60772012-11-14 21:41:46 +00001614 /*
1615 * Rate-limit the amount of data that is being migrated to a node.
1616 * Optimal placement is no good if the memory bus is saturated and
1617 * all the time is being spent migrating!
1618 */
Mel Gormand28d43352012-11-29 09:24:36 +00001619 if (numamigrate_update_ratelimit(pgdat, 1)) {
Mel Gormana8f60772012-11-14 21:41:46 +00001620 put_page(page);
1621 goto out;
1622 }
Mel Gormana8f60772012-11-14 21:41:46 +00001623
Mel Gormanb32967f2012-11-19 12:35:47 +00001624 isolated = numamigrate_isolate_page(pgdat, page);
1625 if (!isolated)
1626 goto out;
Peter Zijlstra7039e1d2012-10-25 14:16:34 +02001627
Mel Gormanb32967f2012-11-19 12:35:47 +00001628 list_add(&page->lru, &migratepages);
1629 nr_remaining = migrate_pages(&migratepages,
1630 alloc_misplaced_dst_page,
1631 node, false, MIGRATE_ASYNC,
1632 MR_NUMA_MISPLACED);
1633 if (nr_remaining) {
1634 putback_lru_pages(&migratepages);
1635 isolated = 0;
1636 } else
1637 count_vm_numa_event(NUMA_PAGE_MIGRATE);
Peter Zijlstra7039e1d2012-10-25 14:16:34 +02001638 BUG_ON(!list_empty(&migratepages));
1639out:
1640 return isolated;
1641}
Mel Gorman220018d2012-12-05 09:32:56 +00001642#endif /* CONFIG_NUMA_BALANCING */
Mel Gormanb32967f2012-11-19 12:35:47 +00001643
Mel Gorman220018d2012-12-05 09:32:56 +00001644#if defined(CONFIG_NUMA_BALANCING) && defined(CONFIG_TRANSPARENT_HUGEPAGE)
Mel Gormanb32967f2012-11-19 12:35:47 +00001645int migrate_misplaced_transhuge_page(struct mm_struct *mm,
1646 struct vm_area_struct *vma,
1647 pmd_t *pmd, pmd_t entry,
1648 unsigned long address,
1649 struct page *page, int node)
1650{
1651 unsigned long haddr = address & HPAGE_PMD_MASK;
1652 pg_data_t *pgdat = NODE_DATA(node);
1653 int isolated = 0;
1654 struct page *new_page = NULL;
1655 struct mem_cgroup *memcg = NULL;
1656 int page_lru = page_is_file_cache(page);
1657
1658 /*
1659 * Don't migrate pages that are mapped in multiple processes.
1660 * TODO: Handle false sharing detection instead of this hammer
1661 */
1662 if (page_mapcount(page) != 1)
1663 goto out_dropref;
1664
1665 /*
1666 * Rate-limit the amount of data that is being migrated to a node.
1667 * Optimal placement is no good if the memory bus is saturated and
1668 * all the time is being spent migrating!
1669 */
Mel Gormand28d43352012-11-29 09:24:36 +00001670 if (numamigrate_update_ratelimit(pgdat, HPAGE_PMD_NR))
Mel Gormanb32967f2012-11-19 12:35:47 +00001671 goto out_dropref;
1672
1673 new_page = alloc_pages_node(node,
1674 (GFP_TRANSHUGE | GFP_THISNODE) & ~__GFP_WAIT, HPAGE_PMD_ORDER);
Mel Gorman75483412012-11-27 10:31:44 +00001675 if (!new_page) {
1676 count_vm_events(PGMIGRATE_FAIL, HPAGE_PMD_NR);
Mel Gormanb32967f2012-11-19 12:35:47 +00001677 goto out_dropref;
Mel Gorman75483412012-11-27 10:31:44 +00001678 }
Mel Gormanb32967f2012-11-19 12:35:47 +00001679 page_xchg_last_nid(new_page, page_last_nid(page));
1680
1681 isolated = numamigrate_isolate_page(pgdat, page);
1682 if (!isolated) {
Mel Gorman75483412012-11-27 10:31:44 +00001683 count_vm_events(PGMIGRATE_FAIL, HPAGE_PMD_NR);
Mel Gormanb32967f2012-11-19 12:35:47 +00001684 put_page(new_page);
1685 goto out_keep_locked;
1686 }
1687
1688 /* Prepare a page as a migration target */
1689 __set_page_locked(new_page);
1690 SetPageSwapBacked(new_page);
1691
1692 /* anon mapping, we can simply copy page->mapping to the new page: */
1693 new_page->mapping = page->mapping;
1694 new_page->index = page->index;
1695 migrate_page_copy(new_page, page);
1696 WARN_ON(PageLRU(new_page));
1697
1698 /* Recheck the target PMD */
1699 spin_lock(&mm->page_table_lock);
1700 if (unlikely(!pmd_same(*pmd, entry))) {
1701 spin_unlock(&mm->page_table_lock);
1702
1703 /* Reverse changes made by migrate_page_copy() */
1704 if (TestClearPageActive(new_page))
1705 SetPageActive(page);
1706 if (TestClearPageUnevictable(new_page))
1707 SetPageUnevictable(page);
1708 mlock_migrate_page(page, new_page);
1709
1710 unlock_page(new_page);
1711 put_page(new_page); /* Free it */
1712
1713 unlock_page(page);
1714 putback_lru_page(page);
1715
1716 count_vm_events(PGMIGRATE_FAIL, HPAGE_PMD_NR);
1717 goto out;
1718 }
1719
1720 /*
1721 * Traditional migration needs to prepare the memcg charge
1722 * transaction early to prevent the old page from being
1723 * uncharged when installing migration entries. Here we can
1724 * save the potential rollback and start the charge transfer
1725 * only when migration is already known to end successfully.
1726 */
1727 mem_cgroup_prepare_migration(page, new_page, &memcg);
1728
1729 entry = mk_pmd(new_page, vma->vm_page_prot);
1730 entry = pmd_mknonnuma(entry);
1731 entry = maybe_pmd_mkwrite(pmd_mkdirty(entry), vma);
1732 entry = pmd_mkhuge(entry);
1733
1734 page_add_new_anon_rmap(new_page, vma, haddr);
1735
1736 set_pmd_at(mm, haddr, pmd, entry);
Stephen Rothwellce4a9cc2012-12-10 19:50:57 +11001737 update_mmu_cache_pmd(vma, address, &entry);
Mel Gormanb32967f2012-11-19 12:35:47 +00001738 page_remove_rmap(page);
1739 /*
1740 * Finish the charge transaction under the page table lock to
1741 * prevent split_huge_page() from dividing up the charge
1742 * before it's fully transferred to the new page.
1743 */
1744 mem_cgroup_end_migration(memcg, page, new_page, true);
1745 spin_unlock(&mm->page_table_lock);
1746
1747 unlock_page(new_page);
1748 unlock_page(page);
1749 put_page(page); /* Drop the rmap reference */
1750 put_page(page); /* Drop the LRU isolation reference */
1751
1752 count_vm_events(PGMIGRATE_SUCCESS, HPAGE_PMD_NR);
1753 count_vm_numa_events(NUMA_PAGE_MIGRATE, HPAGE_PMD_NR);
1754
1755out:
1756 mod_zone_page_state(page_zone(page),
1757 NR_ISOLATED_ANON + page_lru,
1758 -HPAGE_PMD_NR);
1759 return isolated;
1760
1761out_dropref:
1762 put_page(page);
1763out_keep_locked:
1764 return 0;
1765}
Peter Zijlstra7039e1d2012-10-25 14:16:34 +02001766#endif /* CONFIG_NUMA_BALANCING */
1767
1768#endif /* CONFIG_NUMA */