]> git.proxmox.com Git - mirror_ubuntu-jammy-kernel.git/blame - mm/migrate.c
Merge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/viro/vfs-2.6
[mirror_ubuntu-jammy-kernel.git] / mm / migrate.c
CommitLineData
b20a3503
CL
1/*
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>
cde53535 12 * Christoph Lameter
b20a3503
CL
13 */
14
15#include <linux/migrate.h>
16#include <linux/module.h>
17#include <linux/swap.h>
0697212a 18#include <linux/swapops.h>
b20a3503 19#include <linux/pagemap.h>
e23ca00b 20#include <linux/buffer_head.h>
b20a3503 21#include <linux/mm_inline.h>
b488893a 22#include <linux/nsproxy.h>
b20a3503 23#include <linux/pagevec.h>
e9995ef9 24#include <linux/ksm.h>
b20a3503
CL
25#include <linux/rmap.h>
26#include <linux/topology.h>
27#include <linux/cpu.h>
28#include <linux/cpuset.h>
04e62a29 29#include <linux/writeback.h>
742755a1
CL
30#include <linux/mempolicy.h>
31#include <linux/vmalloc.h>
86c3a764 32#include <linux/security.h>
8a9f3ccd 33#include <linux/memcontrol.h>
4f5ca265 34#include <linux/syscalls.h>
b20a3503
CL
35
36#include "internal.h"
37
b20a3503
CL
38#define lru_to_page(_head) (list_entry((_head)->prev, struct page, lru))
39
b20a3503 40/*
742755a1
CL
41 * migrate_prep() needs to be called before we start compiling a list of pages
42 * to be migrated using isolate_lru_page().
b20a3503
CL
43 */
44int migrate_prep(void)
45{
b20a3503
CL
46 /*
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
b20a3503 57/*
894bc310
LS
58 * Add isolated pages on the list back to the LRU under page lock
59 * to avoid leaking evictable pages back onto unevictable list.
b20a3503
CL
60 *
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) {
e24f0b8f 70 list_del(&page->lru);
a731286d 71 dec_zone_page_state(page, NR_ISOLATED_ANON +
6c0b1351 72 page_is_file_cache(page));
894bc310 73 putback_lru_page(page);
b20a3503
CL
74 count++;
75 }
76 return count;
77}
78
0697212a
CL
79/*
80 * Restore a potential migration pte to a working pte entry
81 */
e9995ef9
HD
82static int remove_migration_pte(struct page *new, struct vm_area_struct *vma,
83 unsigned long addr, void *old)
0697212a
CL
84{
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))
e9995ef9 95 goto out;
0697212a
CL
96
97 pud = pud_offset(pgd, addr);
98 if (!pud_present(*pud))
e9995ef9 99 goto out;
0697212a
CL
100
101 pmd = pmd_offset(pud, addr);
102 if (!pmd_present(*pmd))
e9995ef9 103 goto out;
0697212a
CL
104
105 ptep = pte_offset_map(pmd, addr);
106
107 if (!is_swap_pte(*ptep)) {
108 pte_unmap(ptep);
e9995ef9 109 goto out;
0697212a
CL
110 }
111
112 ptl = pte_lockptr(mm, pmd);
113 spin_lock(ptl);
114 pte = *ptep;
115 if (!is_swap_pte(pte))
e9995ef9 116 goto unlock;
0697212a
CL
117
118 entry = pte_to_swp_entry(pte);
119
e9995ef9
HD
120 if (!is_migration_entry(entry) ||
121 migration_entry_to_page(entry) != old)
122 goto unlock;
0697212a 123
0697212a
CL
124 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);
97ee0524 128 flush_cache_page(vma, addr, pte_pfn(pte));
0697212a 129 set_pte_at(mm, addr, ptep, pte);
04e62a29
CL
130
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);
e9995ef9 138unlock:
0697212a 139 pte_unmap_unlock(ptep, ptl);
e9995ef9
HD
140out:
141 return SWAP_AGAIN;
0697212a
CL
142}
143
04e62a29
CL
144/*
145 * 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{
e9995ef9 150 rmap_walk(new, remove_migration_pte, old);
04e62a29
CL
151}
152
0697212a
CL
153/*
154 * 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
e286781d
NP
179 /*
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;
0697212a
CL
188 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
b20a3503 196/*
c3fcf8a5 197 * Replace the page in the mapping.
5b5c7120
CL
198 *
199 * The number of remaining references must be:
200 * 1 for anonymous pages without a mapping
201 * 2 for pages with a mapping
266cf658 202 * 3 for pages with a mapping and PagePrivate/PagePrivate2 set.
b20a3503 203 */
2d1db3b1
CL
204static int migrate_page_move_mapping(struct address_space *mapping,
205 struct page *newpage, struct page *page)
b20a3503 206{
e286781d 207 int expected_count;
7cf9c2c7 208 void **pslot;
b20a3503 209
6c5240ae 210 if (!mapping) {
0e8c7d0f 211 /* Anonymous page without mapping */
6c5240ae
CL
212 if (page_count(page) != 1)
213 return -EAGAIN;
214 return 0;
215 }
216
19fd6231 217 spin_lock_irq(&mapping->tree_lock);
b20a3503 218
7cf9c2c7
NP
219 pslot = radix_tree_lookup_slot(&mapping->page_tree,
220 page_index(page));
b20a3503 221
edcf4748 222 expected_count = 2 + page_has_private(page);
e286781d 223 if (page_count(page) != expected_count ||
7cf9c2c7 224 (struct page *)radix_tree_deref_slot(pslot) != page) {
19fd6231 225 spin_unlock_irq(&mapping->tree_lock);
e23ca00b 226 return -EAGAIN;
b20a3503
CL
227 }
228
e286781d 229 if (!page_freeze_refs(page, expected_count)) {
19fd6231 230 spin_unlock_irq(&mapping->tree_lock);
e286781d
NP
231 return -EAGAIN;
232 }
233
b20a3503
CL
234 /*
235 * Now we know that no one else is looking at the page.
b20a3503 236 */
7cf9c2c7 237 get_page(newpage); /* add cache reference */
b20a3503
CL
238 if (PageSwapCache(page)) {
239 SetPageSwapCache(newpage);
240 set_page_private(newpage, page_private(page));
241 }
242
7cf9c2c7
NP
243 radix_tree_replace_slot(pslot, newpage);
244
e286781d 245 page_unfreeze_refs(page, expected_count);
7cf9c2c7
NP
246 /*
247 * Drop cache reference from old page.
248 * We know this isn't the last reference.
249 */
b20a3503 250 __put_page(page);
7cf9c2c7 251
0e8c7d0f
CL
252 /*
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);
4b02108a
KM
264 if (PageSwapBacked(page)) {
265 __dec_zone_page_state(page, NR_SHMEM);
266 __inc_zone_page_state(newpage, NR_SHMEM);
267 }
19fd6231 268 spin_unlock_irq(&mapping->tree_lock);
b20a3503
CL
269
270 return 0;
271}
b20a3503
CL
272
273/*
274 * Copy the page to its new location
275 */
e7340f73 276static void migrate_page_copy(struct page *newpage, struct page *page)
b20a3503 277{
b7abea96
KH
278 int anon;
279
b20a3503
CL
280 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);
894bc310
LS
288 if (TestClearPageActive(page)) {
289 VM_BUG_ON(PageUnevictable(page));
b20a3503 290 SetPageActive(newpage);
418b27ef
LS
291 } else if (TestClearPageUnevictable(page))
292 SetPageUnevictable(newpage);
b20a3503
CL
293 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);
3a902c5f
NP
300 /*
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);
b20a3503
CL
308 }
309
b291f000 310 mlock_migrate_page(newpage, page);
e9995ef9 311 ksm_migrate_page(newpage, page);
b291f000 312
b20a3503 313 ClearPageSwapCache(page);
b20a3503
CL
314 ClearPagePrivate(page);
315 set_page_private(page, 0);
b7abea96
KH
316 /* page->mapping contains a flag for PageAnon() */
317 anon = PageAnon(page);
b20a3503
CL
318 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}
b20a3503 327
1d8b85cc
CL
328/************************************************************
329 * Migration functions
330 ***********************************************************/
331
332/* Always fail migration. Used for mappings that are not movable */
2d1db3b1
CL
333int fail_migrate_page(struct address_space *mapping,
334 struct page *newpage, struct page *page)
1d8b85cc
CL
335{
336 return -EIO;
337}
338EXPORT_SYMBOL(fail_migrate_page);
339
b20a3503
CL
340/*
341 * Common logic to directly migrate a single page suitable for
266cf658 342 * pages that do not use PagePrivate/PagePrivate2.
b20a3503
CL
343 *
344 * Pages are locked upon entry and exit.
345 */
2d1db3b1
CL
346int migrate_page(struct address_space *mapping,
347 struct page *newpage, struct page *page)
b20a3503
CL
348{
349 int rc;
350
351 BUG_ON(PageWriteback(page)); /* Writeback must be complete */
352
2d1db3b1 353 rc = migrate_page_move_mapping(mapping, newpage, page);
b20a3503
CL
354
355 if (rc)
356 return rc;
357
358 migrate_page_copy(newpage, page);
b20a3503
CL
359 return 0;
360}
361EXPORT_SYMBOL(migrate_page);
362
9361401e 363#ifdef CONFIG_BLOCK
1d8b85cc
CL
364/*
365 * 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 */
2d1db3b1
CL
369int buffer_migrate_page(struct address_space *mapping,
370 struct page *newpage, struct page *page)
1d8b85cc 371{
1d8b85cc
CL
372 struct buffer_head *bh, *head;
373 int rc;
374
1d8b85cc 375 if (!page_has_buffers(page))
2d1db3b1 376 return migrate_page(mapping, newpage, page);
1d8b85cc
CL
377
378 head = page_buffers(page);
379
2d1db3b1 380 rc = migrate_page_move_mapping(mapping, newpage, page);
1d8b85cc
CL
381
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);
9361401e 421#endif
1d8b85cc 422
04e62a29
CL
423/*
424 * Writeback a page to clean the dirty state
425 */
426static int writeout(struct address_space *mapping, struct page *page)
8351a6e4 427{
04e62a29
CL
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
8351a6e4 446 /*
04e62a29
CL
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.
8351a6e4 453 */
04e62a29 454 remove_migration_ptes(page, page);
8351a6e4 455
04e62a29 456 rc = mapping->a_ops->writepage(page, &wbc);
8351a6e4 457
04e62a29
CL
458 if (rc != AOP_WRITEPAGE_ACTIVATE)
459 /* unlocked. Relock */
460 lock_page(page);
461
bda8550d 462 return (rc < 0) ? -EIO : -EAGAIN;
04e62a29
CL
463}
464
465/*
466 * Default handling if a filesystem does not provide a migration function.
467 */
468static int fallback_migrate_page(struct address_space *mapping,
469 struct page *newpage, struct page *page)
470{
471 if (PageDirty(page))
472 return writeout(mapping, page);
8351a6e4
CL
473
474 /*
475 * Buffers may be managed in a filesystem specific way.
476 * We must have no buffers or drop them.
477 */
266cf658 478 if (page_has_private(page) &&
8351a6e4
CL
479 !try_to_release_page(page, GFP_KERNEL))
480 return -EAGAIN;
481
482 return migrate_page(mapping, newpage, page);
483}
484
e24f0b8f
CL
485/*
486 * 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.
894bc310
LS
491 *
492 * Return value:
493 * < 0 - error code
494 * == 0 - success
e24f0b8f
CL
495 */
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 */
529ae9aa 506 if (!trylock_page(newpage))
e24f0b8f
CL
507 BUG();
508
509 /* Prepare mapping for the new page.*/
510 newpage->index = page->index;
511 newpage->mapping = page->mapping;
b2e18538
RR
512 if (PageSwapBacked(page))
513 SetPageSwapBacked(newpage);
e24f0b8f
CL
514
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
e9995ef9 531 if (!rc)
e24f0b8f 532 remove_migration_ptes(page, newpage);
e9995ef9 533 else
e24f0b8f
CL
534 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 */
95a402c3 545static int unmap_and_move(new_page_t get_new_page, unsigned long private,
62b61f61 546 struct page *page, int force, int offlining)
e24f0b8f
CL
547{
548 int rc = 0;
742755a1
CL
549 int *result = NULL;
550 struct page *newpage = get_new_page(page, private, &result);
989f89c5 551 int rcu_locked = 0;
ae41be37 552 int charge = 0;
e00e4316 553 struct mem_cgroup *mem = NULL;
95a402c3
CL
554
555 if (!newpage)
556 return -ENOMEM;
e24f0b8f 557
894bc310 558 if (page_count(page) == 1) {
e24f0b8f 559 /* page was freed from under us. So we are done. */
95a402c3 560 goto move_newpage;
894bc310 561 }
e24f0b8f 562
e8589cc1 563 /* prepare cgroup just returns 0 or -ENOMEM */
e24f0b8f 564 rc = -EAGAIN;
01b1ae63 565
529ae9aa 566 if (!trylock_page(page)) {
e24f0b8f 567 if (!force)
95a402c3 568 goto move_newpage;
e24f0b8f
CL
569 lock_page(page);
570 }
571
62b61f61
HD
572 /*
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
01b1ae63
KH
586 /* 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
e24f0b8f
CL
594 if (PageWriteback(page)) {
595 if (!force)
01b1ae63 596 goto uncharge;
e24f0b8f
CL
597 wait_on_page_writeback(page);
598 }
e24f0b8f 599 /*
dc386d4d
KH
600 * 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()
989f89c5
KH
604 * File Caches may use write_page() or lock_page() in migration, then,
605 * just care Anon page here.
dc386d4d 606 */
989f89c5
KH
607 if (PageAnon(page)) {
608 rcu_read_lock();
609 rcu_locked = 1;
610 }
62e1c553 611
dc386d4d 612 /*
62e1c553
SL
613 * 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.
e24f0b8f 623 */
62e1c553 624 if (!page->mapping) {
266cf658 625 if (!PageAnon(page) && page_has_private(page)) {
62e1c553
SL
626 /*
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);
abfc3488 634 goto rcu_unlock;
62e1c553 635 }
abfc3488 636 goto skip_unmap;
62e1c553
SL
637 }
638
dc386d4d 639 /* Establish migration ptes or remove ptes */
14fa31b8 640 try_to_unmap(page, TTU_MIGRATION|TTU_IGNORE_MLOCK|TTU_IGNORE_ACCESS);
dc386d4d 641
abfc3488 642skip_unmap:
e6a1530d
CL
643 if (!page_mapped(page))
644 rc = move_to_new_page(newpage, page);
e24f0b8f 645
e8589cc1 646 if (rc)
e24f0b8f 647 remove_migration_ptes(page, page);
dc386d4d 648rcu_unlock:
989f89c5
KH
649 if (rcu_locked)
650 rcu_read_unlock();
01b1ae63
KH
651uncharge:
652 if (!charge)
653 mem_cgroup_end_migration(mem, page, newpage);
e24f0b8f
CL
654unlock:
655 unlock_page(page);
95a402c3 656
e24f0b8f 657 if (rc != -EAGAIN) {
aaa994b3
CL
658 /*
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);
a731286d 665 dec_zone_page_state(page, NR_ISOLATED_ANON +
6c0b1351 666 page_is_file_cache(page));
894bc310 667 putback_lru_page(page);
e24f0b8f 668 }
95a402c3
CL
669
670move_newpage:
894bc310 671
95a402c3
CL
672 /*
673 * Move the new page to the LRU. If migration was not successful
674 * then this will free the page.
675 */
894bc310
LS
676 putback_lru_page(newpage);
677
742755a1
CL
678 if (result) {
679 if (rc)
680 *result = rc;
681 else
682 *result = page_to_nid(newpage);
683 }
e24f0b8f
CL
684 return rc;
685}
686
b20a3503
CL
687/*
688 * migrate_pages
689 *
95a402c3
CL
690 * 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.
b20a3503
CL
693 *
694 * The function returns after 10 attempts or if no pages
695 * are movable anymore because to has become empty
aaa994b3 696 * or no retryable pages exist anymore. All pages will be
e9534b3f 697 * returned to the LRU or freed.
b20a3503 698 *
95a402c3 699 * Return: Number of pages not migrated or error code.
b20a3503 700 */
95a402c3 701int migrate_pages(struct list_head *from,
62b61f61 702 new_page_t get_new_page, unsigned long private, int offlining)
b20a3503 703{
e24f0b8f 704 int retry = 1;
b20a3503
CL
705 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
e24f0b8f
CL
715 for(pass = 0; pass < 10 && retry; pass++) {
716 retry = 0;
b20a3503 717
e24f0b8f 718 list_for_each_entry_safe(page, page2, from, lru) {
e24f0b8f 719 cond_resched();
2d1db3b1 720
95a402c3 721 rc = unmap_and_move(get_new_page, private,
62b61f61 722 page, pass > 2, offlining);
2d1db3b1 723
e24f0b8f 724 switch(rc) {
95a402c3
CL
725 case -ENOMEM:
726 goto out;
e24f0b8f 727 case -EAGAIN:
2d1db3b1 728 retry++;
e24f0b8f
CL
729 break;
730 case 0:
e24f0b8f
CL
731 break;
732 default:
2d1db3b1 733 /* Permanent failure */
2d1db3b1 734 nr_failed++;
e24f0b8f 735 break;
2d1db3b1 736 }
b20a3503
CL
737 }
738 }
95a402c3
CL
739 rc = 0;
740out:
b20a3503
CL
741 if (!swapwrite)
742 current->flags &= ~PF_SWAPWRITE;
743
aaa994b3 744 putback_lru_pages(from);
b20a3503 745
95a402c3
CL
746 if (rc)
747 return rc;
b20a3503 748
95a402c3 749 return nr_failed + retry;
b20a3503 750}
95a402c3 751
742755a1
CL
752#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
6484eb3e 776 return alloc_pages_exact_node(pm->node,
769848c0 777 GFP_HIGHUSER_MOVABLE | GFP_THISNODE, 0);
742755a1
CL
778}
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.
5e9a0f02 784 * The pm array ends with node = MAX_NUMNODES.
742755a1 785 */
5e9a0f02
BG
786static int do_move_page_to_node_array(struct mm_struct *mm,
787 struct page_to_node *pm,
788 int migrate_all)
742755a1
CL
789{
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 */
742755a1
CL
799 for (pp = pm; pp->node != MAX_NUMNODES; pp++) {
800 struct vm_area_struct *vma;
801 struct page *page;
802
742755a1
CL
803 err = -EFAULT;
804 vma = find_vma(mm, pp->addr);
0dc952dc 805 if (!vma || !vma_migratable(vma))
742755a1
CL
806 goto set_status;
807
808 page = follow_page(vma, pp->addr, FOLL_GET);
89f5b7da
LT
809
810 err = PTR_ERR(page);
811 if (IS_ERR(page))
812 goto set_status;
813
742755a1
CL
814 err = -ENOENT;
815 if (!page)
816 goto set_status;
817
62b61f61
HD
818 /* Use PageReserved to check for zero page */
819 if (PageReserved(page) || PageKsm(page))
742755a1
CL
820 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
62695a84 836 err = isolate_lru_page(page);
6d9c285a 837 if (!err) {
62695a84 838 list_add_tail(&page->lru, &pagelist);
6d9c285a
KM
839 inc_zone_page_state(page, NR_ISOLATED_ANON +
840 page_is_file_cache(page));
841 }
742755a1
CL
842put_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
e78bbfa8 853 err = 0;
742755a1
CL
854 if (!list_empty(&pagelist))
855 err = migrate_pages(&pagelist, new_page_node,
62b61f61 856 (unsigned long)pm, 0);
742755a1
CL
857
858 up_read(&mm->mmap_sem);
859 return err;
860}
861
5e9a0f02
BG
862/*
863 * 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{
3140a227 872 struct page_to_node *pm;
5e9a0f02 873 nodemask_t task_nodes;
3140a227
BG
874 unsigned long chunk_nr_pages;
875 unsigned long chunk_start;
876 int err;
5e9a0f02
BG
877
878 task_nodes = cpuset_mems_allowed(task);
879
3140a227
BG
880 err = -ENOMEM;
881 pm = (struct page_to_node *)__get_free_page(GFP_KERNEL);
882 if (!pm)
5e9a0f02 883 goto out;
35282a2d
BG
884
885 migrate_prep();
886
5e9a0f02 887 /*
3140a227
BG
888 * Store a chunk of page_to_node array in a page,
889 * but keep the last one as a marker
5e9a0f02 890 */
3140a227 891 chunk_nr_pages = (PAGE_SIZE / sizeof(struct page_to_node)) - 1;
5e9a0f02 892
3140a227
BG
893 for (chunk_start = 0;
894 chunk_start < nr_pages;
895 chunk_start += chunk_nr_pages) {
896 int j;
5e9a0f02 897
3140a227
BG
898 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;
5e9a0f02
BG
904 int node;
905
3140a227
BG
906 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))
5e9a0f02
BG
912 goto out_pm;
913
914 err = -ENODEV;
6f5a55f1
LT
915 if (node < 0 || node >= MAX_NUMNODES)
916 goto out_pm;
917
5e9a0f02
BG
918 if (!node_state(node, N_HIGH_MEMORY))
919 goto out_pm;
920
921 err = -EACCES;
922 if (!node_isset(node, task_nodes))
923 goto out_pm;
924
3140a227
BG
925 pm[j].node = node;
926 }
927
928 /* End marker for this chunk */
929 pm[chunk_nr_pages].node = MAX_NUMNODES;
930
931 /* Migrate this chunk */
932 err = do_move_page_to_node_array(mm, pm,
933 flags & MPOL_MF_MOVE_ALL);
934 if (err < 0)
935 goto out_pm;
5e9a0f02 936
5e9a0f02 937 /* Return status information */
3140a227
BG
938 for (j = 0; j < chunk_nr_pages; j++)
939 if (put_user(pm[j].status, status + j + chunk_start)) {
5e9a0f02 940 err = -EFAULT;
3140a227
BG
941 goto out_pm;
942 }
943 }
944 err = 0;
5e9a0f02
BG
945
946out_pm:
3140a227 947 free_page((unsigned long)pm);
5e9a0f02
BG
948out:
949 return err;
950}
951
742755a1 952/*
2f007e74 953 * Determine the nodes of an array of pages and store it in an array of status.
742755a1 954 */
80bba129
BG
955static void do_pages_stat_array(struct mm_struct *mm, unsigned long nr_pages,
956 const void __user **pages, int *status)
742755a1 957{
2f007e74 958 unsigned long i;
2f007e74 959
742755a1
CL
960 down_read(&mm->mmap_sem);
961
2f007e74 962 for (i = 0; i < nr_pages; i++) {
80bba129 963 unsigned long addr = (unsigned long)(*pages);
742755a1
CL
964 struct vm_area_struct *vma;
965 struct page *page;
c095adbc 966 int err = -EFAULT;
2f007e74
BG
967
968 vma = find_vma(mm, addr);
742755a1
CL
969 if (!vma)
970 goto set_status;
971
2f007e74 972 page = follow_page(vma, addr, 0);
89f5b7da
LT
973
974 err = PTR_ERR(page);
975 if (IS_ERR(page))
976 goto set_status;
977
742755a1
CL
978 err = -ENOENT;
979 /* Use PageReserved to check for zero page */
62b61f61 980 if (!page || PageReserved(page) || PageKsm(page))
742755a1
CL
981 goto set_status;
982
983 err = page_to_nid(page);
984set_status:
80bba129
BG
985 *status = err;
986
987 pages++;
988 status++;
989 }
990
991 up_read(&mm->mmap_sem);
992}
993
994/*
995 * Determine the nodes of a user array of pages and store it in
996 * a user array of status.
997 */
998static int do_pages_stat(struct mm_struct *mm, unsigned long nr_pages,
999 const void __user * __user *pages,
1000 int __user *status)
1001{
1002#define DO_PAGES_STAT_CHUNK_NR 16
1003 const void __user *chunk_pages[DO_PAGES_STAT_CHUNK_NR];
1004 int chunk_status[DO_PAGES_STAT_CHUNK_NR];
1005 unsigned long i, chunk_nr = DO_PAGES_STAT_CHUNK_NR;
1006 int err;
1007
1008 for (i = 0; i < nr_pages; i += chunk_nr) {
b9255850 1009 if (chunk_nr > nr_pages - i)
80bba129
BG
1010 chunk_nr = nr_pages - i;
1011
1012 err = copy_from_user(chunk_pages, &pages[i],
1013 chunk_nr * sizeof(*chunk_pages));
1014 if (err) {
1015 err = -EFAULT;
1016 goto out;
1017 }
1018
1019 do_pages_stat_array(mm, chunk_nr, chunk_pages, chunk_status);
1020
1021 err = copy_to_user(&status[i], chunk_status,
1022 chunk_nr * sizeof(*chunk_status));
1023 if (err) {
1024 err = -EFAULT;
1025 goto out;
1026 }
742755a1 1027 }
2f007e74 1028 err = 0;
742755a1 1029
2f007e74 1030out:
2f007e74 1031 return err;
742755a1
CL
1032}
1033
1034/*
1035 * Move a list of pages in the address space of the currently executing
1036 * process.
1037 */
938bb9f5
HC
1038SYSCALL_DEFINE6(move_pages, pid_t, pid, unsigned long, nr_pages,
1039 const void __user * __user *, pages,
1040 const int __user *, nodes,
1041 int __user *, status, int, flags)
742755a1 1042{
c69e8d9c 1043 const struct cred *cred = current_cred(), *tcred;
742755a1 1044 struct task_struct *task;
742755a1 1045 struct mm_struct *mm;
5e9a0f02 1046 int err;
742755a1
CL
1047
1048 /* Check flags */
1049 if (flags & ~(MPOL_MF_MOVE|MPOL_MF_MOVE_ALL))
1050 return -EINVAL;
1051
1052 if ((flags & MPOL_MF_MOVE_ALL) && !capable(CAP_SYS_NICE))
1053 return -EPERM;
1054
1055 /* Find the mm_struct */
1056 read_lock(&tasklist_lock);
228ebcbe 1057 task = pid ? find_task_by_vpid(pid) : current;
742755a1
CL
1058 if (!task) {
1059 read_unlock(&tasklist_lock);
1060 return -ESRCH;
1061 }
1062 mm = get_task_mm(task);
1063 read_unlock(&tasklist_lock);
1064
1065 if (!mm)
1066 return -EINVAL;
1067
1068 /*
1069 * Check if this process has the right to modify the specified
1070 * process. The right exists if the process has administrative
1071 * capabilities, superuser privileges or the same
1072 * userid as the target process.
1073 */
c69e8d9c
DH
1074 rcu_read_lock();
1075 tcred = __task_cred(task);
b6dff3ec
DH
1076 if (cred->euid != tcred->suid && cred->euid != tcred->uid &&
1077 cred->uid != tcred->suid && cred->uid != tcred->uid &&
742755a1 1078 !capable(CAP_SYS_NICE)) {
c69e8d9c 1079 rcu_read_unlock();
742755a1 1080 err = -EPERM;
5e9a0f02 1081 goto out;
742755a1 1082 }
c69e8d9c 1083 rcu_read_unlock();
742755a1 1084
86c3a764
DQ
1085 err = security_task_movememory(task);
1086 if (err)
5e9a0f02 1087 goto out;
86c3a764 1088
5e9a0f02
BG
1089 if (nodes) {
1090 err = do_pages_move(mm, task, nr_pages, pages, nodes, status,
1091 flags);
1092 } else {
2f007e74 1093 err = do_pages_stat(mm, nr_pages, pages, status);
742755a1
CL
1094 }
1095
742755a1 1096out:
742755a1
CL
1097 mmput(mm);
1098 return err;
1099}
742755a1 1100
7b2259b3
CL
1101/*
1102 * Call migration functions in the vma_ops that may prepare
1103 * memory in a vm for migration. migration functions may perform
1104 * the migration for vmas that do not have an underlying page struct.
1105 */
1106int migrate_vmas(struct mm_struct *mm, const nodemask_t *to,
1107 const nodemask_t *from, unsigned long flags)
1108{
1109 struct vm_area_struct *vma;
1110 int err = 0;
1111
1001c9fb 1112 for (vma = mm->mmap; vma && !err; vma = vma->vm_next) {
7b2259b3
CL
1113 if (vma->vm_ops && vma->vm_ops->migrate) {
1114 err = vma->vm_ops->migrate(vma, to, from, flags);
1115 if (err)
1116 break;
1117 }
1118 }
1119 return err;
1120}
83d1674a 1121#endif