]>
Commit | Line | Data |
---|---|---|
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 CL |
23 | #include <linux/pagevec.h> |
24 | #include <linux/rmap.h> | |
25 | #include <linux/topology.h> | |
26 | #include <linux/cpu.h> | |
27 | #include <linux/cpuset.h> | |
04e62a29 | 28 | #include <linux/writeback.h> |
742755a1 CL |
29 | #include <linux/mempolicy.h> |
30 | #include <linux/vmalloc.h> | |
86c3a764 | 31 | #include <linux/security.h> |
8a9f3ccd | 32 | #include <linux/memcontrol.h> |
4f5ca265 | 33 | #include <linux/syscalls.h> |
b20a3503 CL |
34 | |
35 | #include "internal.h" | |
36 | ||
b20a3503 CL |
37 | #define lru_to_page(_head) (list_entry((_head)->prev, struct page, lru)) |
38 | ||
b20a3503 | 39 | /* |
742755a1 CL |
40 | * migrate_prep() needs to be called before we start compiling a list of pages |
41 | * to be migrated using isolate_lru_page(). | |
b20a3503 CL |
42 | */ |
43 | int migrate_prep(void) | |
44 | { | |
b20a3503 CL |
45 | /* |
46 | * Clear the LRU lists so pages can be isolated. | |
47 | * Note that pages may be moved off the LRU after we have | |
48 | * drained them. Those pages will fail to migrate like other | |
49 | * pages that may be busy. | |
50 | */ | |
51 | lru_add_drain_all(); | |
52 | ||
53 | return 0; | |
54 | } | |
55 | ||
b20a3503 | 56 | /* |
894bc310 LS |
57 | * Add isolated pages on the list back to the LRU under page lock |
58 | * to avoid leaking evictable pages back onto unevictable list. | |
b20a3503 CL |
59 | * |
60 | * returns the number of pages put back. | |
61 | */ | |
62 | int putback_lru_pages(struct list_head *l) | |
63 | { | |
64 | struct page *page; | |
65 | struct page *page2; | |
66 | int count = 0; | |
67 | ||
68 | list_for_each_entry_safe(page, page2, l, lru) { | |
e24f0b8f | 69 | list_del(&page->lru); |
a731286d | 70 | dec_zone_page_state(page, NR_ISOLATED_ANON + |
6c0b1351 | 71 | page_is_file_cache(page)); |
894bc310 | 72 | putback_lru_page(page); |
b20a3503 CL |
73 | count++; |
74 | } | |
75 | return count; | |
76 | } | |
77 | ||
0697212a CL |
78 | /* |
79 | * Restore a potential migration pte to a working pte entry | |
80 | */ | |
04e62a29 | 81 | static void remove_migration_pte(struct vm_area_struct *vma, |
0697212a CL |
82 | struct page *old, struct page *new) |
83 | { | |
84 | struct mm_struct *mm = vma->vm_mm; | |
85 | swp_entry_t entry; | |
86 | pgd_t *pgd; | |
87 | pud_t *pud; | |
88 | pmd_t *pmd; | |
89 | pte_t *ptep, pte; | |
90 | spinlock_t *ptl; | |
04e62a29 CL |
91 | unsigned long addr = page_address_in_vma(new, vma); |
92 | ||
93 | if (addr == -EFAULT) | |
94 | return; | |
0697212a CL |
95 | |
96 | pgd = pgd_offset(mm, addr); | |
97 | if (!pgd_present(*pgd)) | |
98 | return; | |
99 | ||
100 | pud = pud_offset(pgd, addr); | |
101 | if (!pud_present(*pud)) | |
102 | return; | |
103 | ||
104 | pmd = pmd_offset(pud, addr); | |
105 | if (!pmd_present(*pmd)) | |
106 | return; | |
107 | ||
108 | ptep = pte_offset_map(pmd, addr); | |
109 | ||
110 | if (!is_swap_pte(*ptep)) { | |
111 | pte_unmap(ptep); | |
112 | return; | |
113 | } | |
114 | ||
115 | ptl = pte_lockptr(mm, pmd); | |
116 | spin_lock(ptl); | |
117 | pte = *ptep; | |
118 | if (!is_swap_pte(pte)) | |
119 | goto out; | |
120 | ||
121 | entry = pte_to_swp_entry(pte); | |
122 | ||
123 | if (!is_migration_entry(entry) || migration_entry_to_page(entry) != old) | |
124 | goto out; | |
125 | ||
0697212a CL |
126 | get_page(new); |
127 | pte = pte_mkold(mk_pte(new, vma->vm_page_prot)); | |
128 | if (is_write_migration_entry(entry)) | |
129 | pte = pte_mkwrite(pte); | |
97ee0524 | 130 | flush_cache_page(vma, addr, pte_pfn(pte)); |
0697212a | 131 | set_pte_at(mm, addr, ptep, pte); |
04e62a29 CL |
132 | |
133 | if (PageAnon(new)) | |
134 | page_add_anon_rmap(new, vma, addr); | |
135 | else | |
136 | page_add_file_rmap(new); | |
137 | ||
138 | /* No need to invalidate - it was non-present before */ | |
139 | update_mmu_cache(vma, addr, pte); | |
04e62a29 | 140 | |
0697212a CL |
141 | out: |
142 | pte_unmap_unlock(ptep, ptl); | |
143 | } | |
144 | ||
145 | /* | |
04e62a29 CL |
146 | * Note that remove_file_migration_ptes will only work on regular mappings, |
147 | * Nonlinear mappings do not use migration entries. | |
148 | */ | |
149 | static void remove_file_migration_ptes(struct page *old, struct page *new) | |
150 | { | |
151 | struct vm_area_struct *vma; | |
abfc3488 | 152 | struct address_space *mapping = new->mapping; |
04e62a29 CL |
153 | struct prio_tree_iter iter; |
154 | pgoff_t pgoff = new->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT); | |
155 | ||
156 | if (!mapping) | |
157 | return; | |
158 | ||
159 | spin_lock(&mapping->i_mmap_lock); | |
160 | ||
161 | vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, pgoff, pgoff) | |
162 | remove_migration_pte(vma, old, new); | |
163 | ||
164 | spin_unlock(&mapping->i_mmap_lock); | |
165 | } | |
166 | ||
167 | /* | |
0697212a CL |
168 | * Must hold mmap_sem lock on at least one of the vmas containing |
169 | * the page so that the anon_vma cannot vanish. | |
170 | */ | |
04e62a29 | 171 | static void remove_anon_migration_ptes(struct page *old, struct page *new) |
0697212a CL |
172 | { |
173 | struct anon_vma *anon_vma; | |
174 | struct vm_area_struct *vma; | |
0697212a CL |
175 | |
176 | /* | |
177 | * We hold the mmap_sem lock. So no need to call page_lock_anon_vma. | |
178 | */ | |
3ca7b3c5 HD |
179 | anon_vma = page_anon_vma(new); |
180 | if (!anon_vma) | |
181 | return; | |
182 | ||
0697212a CL |
183 | spin_lock(&anon_vma->lock); |
184 | ||
185 | list_for_each_entry(vma, &anon_vma->head, anon_vma_node) | |
04e62a29 | 186 | remove_migration_pte(vma, old, new); |
0697212a CL |
187 | |
188 | spin_unlock(&anon_vma->lock); | |
189 | } | |
190 | ||
04e62a29 CL |
191 | /* |
192 | * Get rid of all migration entries and replace them by | |
193 | * references to the indicated page. | |
194 | */ | |
195 | static void remove_migration_ptes(struct page *old, struct page *new) | |
196 | { | |
197 | if (PageAnon(new)) | |
198 | remove_anon_migration_ptes(old, new); | |
199 | else | |
200 | remove_file_migration_ptes(old, new); | |
201 | } | |
202 | ||
0697212a CL |
203 | /* |
204 | * Something used the pte of a page under migration. We need to | |
205 | * get to the page and wait until migration is finished. | |
206 | * When we return from this function the fault will be retried. | |
207 | * | |
208 | * This function is called from do_swap_page(). | |
209 | */ | |
210 | void migration_entry_wait(struct mm_struct *mm, pmd_t *pmd, | |
211 | unsigned long address) | |
212 | { | |
213 | pte_t *ptep, pte; | |
214 | spinlock_t *ptl; | |
215 | swp_entry_t entry; | |
216 | struct page *page; | |
217 | ||
218 | ptep = pte_offset_map_lock(mm, pmd, address, &ptl); | |
219 | pte = *ptep; | |
220 | if (!is_swap_pte(pte)) | |
221 | goto out; | |
222 | ||
223 | entry = pte_to_swp_entry(pte); | |
224 | if (!is_migration_entry(entry)) | |
225 | goto out; | |
226 | ||
227 | page = migration_entry_to_page(entry); | |
228 | ||
e286781d NP |
229 | /* |
230 | * Once radix-tree replacement of page migration started, page_count | |
231 | * *must* be zero. And, we don't want to call wait_on_page_locked() | |
232 | * against a page without get_page(). | |
233 | * So, we use get_page_unless_zero(), here. Even failed, page fault | |
234 | * will occur again. | |
235 | */ | |
236 | if (!get_page_unless_zero(page)) | |
237 | goto out; | |
0697212a CL |
238 | pte_unmap_unlock(ptep, ptl); |
239 | wait_on_page_locked(page); | |
240 | put_page(page); | |
241 | return; | |
242 | out: | |
243 | pte_unmap_unlock(ptep, ptl); | |
244 | } | |
245 | ||
b20a3503 | 246 | /* |
c3fcf8a5 | 247 | * Replace the page in the mapping. |
5b5c7120 CL |
248 | * |
249 | * The number of remaining references must be: | |
250 | * 1 for anonymous pages without a mapping | |
251 | * 2 for pages with a mapping | |
266cf658 | 252 | * 3 for pages with a mapping and PagePrivate/PagePrivate2 set. |
b20a3503 | 253 | */ |
2d1db3b1 CL |
254 | static int migrate_page_move_mapping(struct address_space *mapping, |
255 | struct page *newpage, struct page *page) | |
b20a3503 | 256 | { |
e286781d | 257 | int expected_count; |
7cf9c2c7 | 258 | void **pslot; |
b20a3503 | 259 | |
6c5240ae | 260 | if (!mapping) { |
0e8c7d0f | 261 | /* Anonymous page without mapping */ |
6c5240ae CL |
262 | if (page_count(page) != 1) |
263 | return -EAGAIN; | |
264 | return 0; | |
265 | } | |
266 | ||
19fd6231 | 267 | spin_lock_irq(&mapping->tree_lock); |
b20a3503 | 268 | |
7cf9c2c7 NP |
269 | pslot = radix_tree_lookup_slot(&mapping->page_tree, |
270 | page_index(page)); | |
b20a3503 | 271 | |
edcf4748 | 272 | expected_count = 2 + page_has_private(page); |
e286781d | 273 | if (page_count(page) != expected_count || |
7cf9c2c7 | 274 | (struct page *)radix_tree_deref_slot(pslot) != page) { |
19fd6231 | 275 | spin_unlock_irq(&mapping->tree_lock); |
e23ca00b | 276 | return -EAGAIN; |
b20a3503 CL |
277 | } |
278 | ||
e286781d | 279 | if (!page_freeze_refs(page, expected_count)) { |
19fd6231 | 280 | spin_unlock_irq(&mapping->tree_lock); |
e286781d NP |
281 | return -EAGAIN; |
282 | } | |
283 | ||
b20a3503 CL |
284 | /* |
285 | * Now we know that no one else is looking at the page. | |
b20a3503 | 286 | */ |
7cf9c2c7 | 287 | get_page(newpage); /* add cache reference */ |
b20a3503 CL |
288 | if (PageSwapCache(page)) { |
289 | SetPageSwapCache(newpage); | |
290 | set_page_private(newpage, page_private(page)); | |
291 | } | |
292 | ||
7cf9c2c7 NP |
293 | radix_tree_replace_slot(pslot, newpage); |
294 | ||
e286781d | 295 | page_unfreeze_refs(page, expected_count); |
7cf9c2c7 NP |
296 | /* |
297 | * Drop cache reference from old page. | |
298 | * We know this isn't the last reference. | |
299 | */ | |
b20a3503 | 300 | __put_page(page); |
7cf9c2c7 | 301 | |
0e8c7d0f CL |
302 | /* |
303 | * If moved to a different zone then also account | |
304 | * the page for that zone. Other VM counters will be | |
305 | * taken care of when we establish references to the | |
306 | * new page and drop references to the old page. | |
307 | * | |
308 | * Note that anonymous pages are accounted for | |
309 | * via NR_FILE_PAGES and NR_ANON_PAGES if they | |
310 | * are mapped to swap space. | |
311 | */ | |
312 | __dec_zone_page_state(page, NR_FILE_PAGES); | |
313 | __inc_zone_page_state(newpage, NR_FILE_PAGES); | |
4b02108a KM |
314 | if (PageSwapBacked(page)) { |
315 | __dec_zone_page_state(page, NR_SHMEM); | |
316 | __inc_zone_page_state(newpage, NR_SHMEM); | |
317 | } | |
19fd6231 | 318 | spin_unlock_irq(&mapping->tree_lock); |
b20a3503 CL |
319 | |
320 | return 0; | |
321 | } | |
b20a3503 CL |
322 | |
323 | /* | |
324 | * Copy the page to its new location | |
325 | */ | |
e7340f73 | 326 | static void migrate_page_copy(struct page *newpage, struct page *page) |
b20a3503 | 327 | { |
b7abea96 KH |
328 | int anon; |
329 | ||
b20a3503 CL |
330 | copy_highpage(newpage, page); |
331 | ||
332 | if (PageError(page)) | |
333 | SetPageError(newpage); | |
334 | if (PageReferenced(page)) | |
335 | SetPageReferenced(newpage); | |
336 | if (PageUptodate(page)) | |
337 | SetPageUptodate(newpage); | |
894bc310 LS |
338 | if (TestClearPageActive(page)) { |
339 | VM_BUG_ON(PageUnevictable(page)); | |
b20a3503 | 340 | SetPageActive(newpage); |
894bc310 LS |
341 | } else |
342 | unevictable_migrate_page(newpage, page); | |
b20a3503 CL |
343 | if (PageChecked(page)) |
344 | SetPageChecked(newpage); | |
345 | if (PageMappedToDisk(page)) | |
346 | SetPageMappedToDisk(newpage); | |
347 | ||
348 | if (PageDirty(page)) { | |
349 | clear_page_dirty_for_io(page); | |
3a902c5f NP |
350 | /* |
351 | * Want to mark the page and the radix tree as dirty, and | |
352 | * redo the accounting that clear_page_dirty_for_io undid, | |
353 | * but we can't use set_page_dirty because that function | |
354 | * is actually a signal that all of the page has become dirty. | |
355 | * Wheras only part of our page may be dirty. | |
356 | */ | |
357 | __set_page_dirty_nobuffers(newpage); | |
b20a3503 CL |
358 | } |
359 | ||
b291f000 NP |
360 | mlock_migrate_page(newpage, page); |
361 | ||
b20a3503 | 362 | ClearPageSwapCache(page); |
b20a3503 CL |
363 | ClearPagePrivate(page); |
364 | set_page_private(page, 0); | |
b7abea96 KH |
365 | /* page->mapping contains a flag for PageAnon() */ |
366 | anon = PageAnon(page); | |
b20a3503 CL |
367 | page->mapping = NULL; |
368 | ||
369 | /* | |
370 | * If any waiters have accumulated on the new page then | |
371 | * wake them up. | |
372 | */ | |
373 | if (PageWriteback(newpage)) | |
374 | end_page_writeback(newpage); | |
375 | } | |
b20a3503 | 376 | |
1d8b85cc CL |
377 | /************************************************************ |
378 | * Migration functions | |
379 | ***********************************************************/ | |
380 | ||
381 | /* Always fail migration. Used for mappings that are not movable */ | |
2d1db3b1 CL |
382 | int fail_migrate_page(struct address_space *mapping, |
383 | struct page *newpage, struct page *page) | |
1d8b85cc CL |
384 | { |
385 | return -EIO; | |
386 | } | |
387 | EXPORT_SYMBOL(fail_migrate_page); | |
388 | ||
b20a3503 CL |
389 | /* |
390 | * Common logic to directly migrate a single page suitable for | |
266cf658 | 391 | * pages that do not use PagePrivate/PagePrivate2. |
b20a3503 CL |
392 | * |
393 | * Pages are locked upon entry and exit. | |
394 | */ | |
2d1db3b1 CL |
395 | int migrate_page(struct address_space *mapping, |
396 | struct page *newpage, struct page *page) | |
b20a3503 CL |
397 | { |
398 | int rc; | |
399 | ||
400 | BUG_ON(PageWriteback(page)); /* Writeback must be complete */ | |
401 | ||
2d1db3b1 | 402 | rc = migrate_page_move_mapping(mapping, newpage, page); |
b20a3503 CL |
403 | |
404 | if (rc) | |
405 | return rc; | |
406 | ||
407 | migrate_page_copy(newpage, page); | |
b20a3503 CL |
408 | return 0; |
409 | } | |
410 | EXPORT_SYMBOL(migrate_page); | |
411 | ||
9361401e | 412 | #ifdef CONFIG_BLOCK |
1d8b85cc CL |
413 | /* |
414 | * Migration function for pages with buffers. This function can only be used | |
415 | * if the underlying filesystem guarantees that no other references to "page" | |
416 | * exist. | |
417 | */ | |
2d1db3b1 CL |
418 | int buffer_migrate_page(struct address_space *mapping, |
419 | struct page *newpage, struct page *page) | |
1d8b85cc | 420 | { |
1d8b85cc CL |
421 | struct buffer_head *bh, *head; |
422 | int rc; | |
423 | ||
1d8b85cc | 424 | if (!page_has_buffers(page)) |
2d1db3b1 | 425 | return migrate_page(mapping, newpage, page); |
1d8b85cc CL |
426 | |
427 | head = page_buffers(page); | |
428 | ||
2d1db3b1 | 429 | rc = migrate_page_move_mapping(mapping, newpage, page); |
1d8b85cc CL |
430 | |
431 | if (rc) | |
432 | return rc; | |
433 | ||
434 | bh = head; | |
435 | do { | |
436 | get_bh(bh); | |
437 | lock_buffer(bh); | |
438 | bh = bh->b_this_page; | |
439 | ||
440 | } while (bh != head); | |
441 | ||
442 | ClearPagePrivate(page); | |
443 | set_page_private(newpage, page_private(page)); | |
444 | set_page_private(page, 0); | |
445 | put_page(page); | |
446 | get_page(newpage); | |
447 | ||
448 | bh = head; | |
449 | do { | |
450 | set_bh_page(bh, newpage, bh_offset(bh)); | |
451 | bh = bh->b_this_page; | |
452 | ||
453 | } while (bh != head); | |
454 | ||
455 | SetPagePrivate(newpage); | |
456 | ||
457 | migrate_page_copy(newpage, page); | |
458 | ||
459 | bh = head; | |
460 | do { | |
461 | unlock_buffer(bh); | |
462 | put_bh(bh); | |
463 | bh = bh->b_this_page; | |
464 | ||
465 | } while (bh != head); | |
466 | ||
467 | return 0; | |
468 | } | |
469 | EXPORT_SYMBOL(buffer_migrate_page); | |
9361401e | 470 | #endif |
1d8b85cc | 471 | |
04e62a29 CL |
472 | /* |
473 | * Writeback a page to clean the dirty state | |
474 | */ | |
475 | static int writeout(struct address_space *mapping, struct page *page) | |
8351a6e4 | 476 | { |
04e62a29 CL |
477 | struct writeback_control wbc = { |
478 | .sync_mode = WB_SYNC_NONE, | |
479 | .nr_to_write = 1, | |
480 | .range_start = 0, | |
481 | .range_end = LLONG_MAX, | |
482 | .nonblocking = 1, | |
483 | .for_reclaim = 1 | |
484 | }; | |
485 | int rc; | |
486 | ||
487 | if (!mapping->a_ops->writepage) | |
488 | /* No write method for the address space */ | |
489 | return -EINVAL; | |
490 | ||
491 | if (!clear_page_dirty_for_io(page)) | |
492 | /* Someone else already triggered a write */ | |
493 | return -EAGAIN; | |
494 | ||
8351a6e4 | 495 | /* |
04e62a29 CL |
496 | * A dirty page may imply that the underlying filesystem has |
497 | * the page on some queue. So the page must be clean for | |
498 | * migration. Writeout may mean we loose the lock and the | |
499 | * page state is no longer what we checked for earlier. | |
500 | * At this point we know that the migration attempt cannot | |
501 | * be successful. | |
8351a6e4 | 502 | */ |
04e62a29 | 503 | remove_migration_ptes(page, page); |
8351a6e4 | 504 | |
04e62a29 | 505 | rc = mapping->a_ops->writepage(page, &wbc); |
8351a6e4 | 506 | |
04e62a29 CL |
507 | if (rc != AOP_WRITEPAGE_ACTIVATE) |
508 | /* unlocked. Relock */ | |
509 | lock_page(page); | |
510 | ||
bda8550d | 511 | return (rc < 0) ? -EIO : -EAGAIN; |
04e62a29 CL |
512 | } |
513 | ||
514 | /* | |
515 | * Default handling if a filesystem does not provide a migration function. | |
516 | */ | |
517 | static int fallback_migrate_page(struct address_space *mapping, | |
518 | struct page *newpage, struct page *page) | |
519 | { | |
520 | if (PageDirty(page)) | |
521 | return writeout(mapping, page); | |
8351a6e4 CL |
522 | |
523 | /* | |
524 | * Buffers may be managed in a filesystem specific way. | |
525 | * We must have no buffers or drop them. | |
526 | */ | |
266cf658 | 527 | if (page_has_private(page) && |
8351a6e4 CL |
528 | !try_to_release_page(page, GFP_KERNEL)) |
529 | return -EAGAIN; | |
530 | ||
531 | return migrate_page(mapping, newpage, page); | |
532 | } | |
533 | ||
e24f0b8f CL |
534 | /* |
535 | * Move a page to a newly allocated page | |
536 | * The page is locked and all ptes have been successfully removed. | |
537 | * | |
538 | * The new page will have replaced the old page if this function | |
539 | * is successful. | |
894bc310 LS |
540 | * |
541 | * Return value: | |
542 | * < 0 - error code | |
543 | * == 0 - success | |
e24f0b8f CL |
544 | */ |
545 | static int move_to_new_page(struct page *newpage, struct page *page) | |
546 | { | |
547 | struct address_space *mapping; | |
548 | int rc; | |
549 | ||
550 | /* | |
551 | * Block others from accessing the page when we get around to | |
552 | * establishing additional references. We are the only one | |
553 | * holding a reference to the new page at this point. | |
554 | */ | |
529ae9aa | 555 | if (!trylock_page(newpage)) |
e24f0b8f CL |
556 | BUG(); |
557 | ||
558 | /* Prepare mapping for the new page.*/ | |
559 | newpage->index = page->index; | |
560 | newpage->mapping = page->mapping; | |
b2e18538 RR |
561 | if (PageSwapBacked(page)) |
562 | SetPageSwapBacked(newpage); | |
e24f0b8f CL |
563 | |
564 | mapping = page_mapping(page); | |
565 | if (!mapping) | |
566 | rc = migrate_page(mapping, newpage, page); | |
567 | else if (mapping->a_ops->migratepage) | |
568 | /* | |
569 | * Most pages have a mapping and most filesystems | |
570 | * should provide a migration function. Anonymous | |
571 | * pages are part of swap space which also has its | |
572 | * own migration function. This is the most common | |
573 | * path for page migration. | |
574 | */ | |
575 | rc = mapping->a_ops->migratepage(mapping, | |
576 | newpage, page); | |
577 | else | |
578 | rc = fallback_migrate_page(mapping, newpage, page); | |
579 | ||
ae41be37 | 580 | if (!rc) { |
e24f0b8f | 581 | remove_migration_ptes(page, newpage); |
ae41be37 | 582 | } else |
e24f0b8f CL |
583 | newpage->mapping = NULL; |
584 | ||
585 | unlock_page(newpage); | |
586 | ||
587 | return rc; | |
588 | } | |
589 | ||
590 | /* | |
591 | * Obtain the lock on page, remove all ptes and migrate the page | |
592 | * to the newly allocated page in newpage. | |
593 | */ | |
95a402c3 CL |
594 | static int unmap_and_move(new_page_t get_new_page, unsigned long private, |
595 | struct page *page, int force) | |
e24f0b8f CL |
596 | { |
597 | int rc = 0; | |
742755a1 CL |
598 | int *result = NULL; |
599 | struct page *newpage = get_new_page(page, private, &result); | |
989f89c5 | 600 | int rcu_locked = 0; |
ae41be37 | 601 | int charge = 0; |
e00e4316 | 602 | struct mem_cgroup *mem = NULL; |
95a402c3 CL |
603 | |
604 | if (!newpage) | |
605 | return -ENOMEM; | |
e24f0b8f | 606 | |
894bc310 | 607 | if (page_count(page) == 1) { |
e24f0b8f | 608 | /* page was freed from under us. So we are done. */ |
95a402c3 | 609 | goto move_newpage; |
894bc310 | 610 | } |
e24f0b8f | 611 | |
e8589cc1 | 612 | /* prepare cgroup just returns 0 or -ENOMEM */ |
e24f0b8f | 613 | rc = -EAGAIN; |
01b1ae63 | 614 | |
529ae9aa | 615 | if (!trylock_page(page)) { |
e24f0b8f | 616 | if (!force) |
95a402c3 | 617 | goto move_newpage; |
e24f0b8f CL |
618 | lock_page(page); |
619 | } | |
620 | ||
01b1ae63 KH |
621 | /* charge against new page */ |
622 | charge = mem_cgroup_prepare_migration(page, &mem); | |
623 | if (charge == -ENOMEM) { | |
624 | rc = -ENOMEM; | |
625 | goto unlock; | |
626 | } | |
627 | BUG_ON(charge); | |
628 | ||
e24f0b8f CL |
629 | if (PageWriteback(page)) { |
630 | if (!force) | |
01b1ae63 | 631 | goto uncharge; |
e24f0b8f CL |
632 | wait_on_page_writeback(page); |
633 | } | |
e24f0b8f | 634 | /* |
dc386d4d KH |
635 | * By try_to_unmap(), page->mapcount goes down to 0 here. In this case, |
636 | * we cannot notice that anon_vma is freed while we migrates a page. | |
637 | * This rcu_read_lock() delays freeing anon_vma pointer until the end | |
638 | * of migration. File cache pages are no problem because of page_lock() | |
989f89c5 KH |
639 | * File Caches may use write_page() or lock_page() in migration, then, |
640 | * just care Anon page here. | |
dc386d4d | 641 | */ |
989f89c5 KH |
642 | if (PageAnon(page)) { |
643 | rcu_read_lock(); | |
644 | rcu_locked = 1; | |
645 | } | |
62e1c553 | 646 | |
dc386d4d | 647 | /* |
62e1c553 SL |
648 | * Corner case handling: |
649 | * 1. When a new swap-cache page is read into, it is added to the LRU | |
650 | * and treated as swapcache but it has no rmap yet. | |
651 | * Calling try_to_unmap() against a page->mapping==NULL page will | |
652 | * trigger a BUG. So handle it here. | |
653 | * 2. An orphaned page (see truncate_complete_page) might have | |
654 | * fs-private metadata. The page can be picked up due to memory | |
655 | * offlining. Everywhere else except page reclaim, the page is | |
656 | * invisible to the vm, so the page can not be migrated. So try to | |
657 | * free the metadata, so the page can be freed. | |
e24f0b8f | 658 | */ |
62e1c553 | 659 | if (!page->mapping) { |
266cf658 | 660 | if (!PageAnon(page) && page_has_private(page)) { |
62e1c553 SL |
661 | /* |
662 | * Go direct to try_to_free_buffers() here because | |
663 | * a) that's what try_to_release_page() would do anyway | |
664 | * b) we may be under rcu_read_lock() here, so we can't | |
665 | * use GFP_KERNEL which is what try_to_release_page() | |
666 | * needs to be effective. | |
667 | */ | |
668 | try_to_free_buffers(page); | |
abfc3488 | 669 | goto rcu_unlock; |
62e1c553 | 670 | } |
abfc3488 | 671 | goto skip_unmap; |
62e1c553 SL |
672 | } |
673 | ||
dc386d4d | 674 | /* Establish migration ptes or remove ptes */ |
14fa31b8 | 675 | try_to_unmap(page, TTU_MIGRATION|TTU_IGNORE_MLOCK|TTU_IGNORE_ACCESS); |
dc386d4d | 676 | |
abfc3488 | 677 | skip_unmap: |
e6a1530d CL |
678 | if (!page_mapped(page)) |
679 | rc = move_to_new_page(newpage, page); | |
e24f0b8f | 680 | |
e8589cc1 | 681 | if (rc) |
e24f0b8f | 682 | remove_migration_ptes(page, page); |
dc386d4d | 683 | rcu_unlock: |
989f89c5 KH |
684 | if (rcu_locked) |
685 | rcu_read_unlock(); | |
01b1ae63 KH |
686 | uncharge: |
687 | if (!charge) | |
688 | mem_cgroup_end_migration(mem, page, newpage); | |
e24f0b8f CL |
689 | unlock: |
690 | unlock_page(page); | |
95a402c3 | 691 | |
e24f0b8f | 692 | if (rc != -EAGAIN) { |
aaa994b3 CL |
693 | /* |
694 | * A page that has been migrated has all references | |
695 | * removed and will be freed. A page that has not been | |
696 | * migrated will have kepts its references and be | |
697 | * restored. | |
698 | */ | |
699 | list_del(&page->lru); | |
a731286d | 700 | dec_zone_page_state(page, NR_ISOLATED_ANON + |
6c0b1351 | 701 | page_is_file_cache(page)); |
894bc310 | 702 | putback_lru_page(page); |
e24f0b8f | 703 | } |
95a402c3 CL |
704 | |
705 | move_newpage: | |
894bc310 | 706 | |
95a402c3 CL |
707 | /* |
708 | * Move the new page to the LRU. If migration was not successful | |
709 | * then this will free the page. | |
710 | */ | |
894bc310 LS |
711 | putback_lru_page(newpage); |
712 | ||
742755a1 CL |
713 | if (result) { |
714 | if (rc) | |
715 | *result = rc; | |
716 | else | |
717 | *result = page_to_nid(newpage); | |
718 | } | |
e24f0b8f CL |
719 | return rc; |
720 | } | |
721 | ||
b20a3503 CL |
722 | /* |
723 | * migrate_pages | |
724 | * | |
95a402c3 CL |
725 | * The function takes one list of pages to migrate and a function |
726 | * that determines from the page to be migrated and the private data | |
727 | * the target of the move and allocates the page. | |
b20a3503 CL |
728 | * |
729 | * The function returns after 10 attempts or if no pages | |
730 | * are movable anymore because to has become empty | |
aaa994b3 | 731 | * or no retryable pages exist anymore. All pages will be |
e9534b3f | 732 | * returned to the LRU or freed. |
b20a3503 | 733 | * |
95a402c3 | 734 | * Return: Number of pages not migrated or error code. |
b20a3503 | 735 | */ |
95a402c3 CL |
736 | int migrate_pages(struct list_head *from, |
737 | new_page_t get_new_page, unsigned long private) | |
b20a3503 | 738 | { |
e24f0b8f | 739 | int retry = 1; |
b20a3503 CL |
740 | int nr_failed = 0; |
741 | int pass = 0; | |
742 | struct page *page; | |
743 | struct page *page2; | |
744 | int swapwrite = current->flags & PF_SWAPWRITE; | |
745 | int rc; | |
746 | ||
747 | if (!swapwrite) | |
748 | current->flags |= PF_SWAPWRITE; | |
749 | ||
e24f0b8f CL |
750 | for(pass = 0; pass < 10 && retry; pass++) { |
751 | retry = 0; | |
b20a3503 | 752 | |
e24f0b8f | 753 | list_for_each_entry_safe(page, page2, from, lru) { |
e24f0b8f | 754 | cond_resched(); |
2d1db3b1 | 755 | |
95a402c3 CL |
756 | rc = unmap_and_move(get_new_page, private, |
757 | page, pass > 2); | |
2d1db3b1 | 758 | |
e24f0b8f | 759 | switch(rc) { |
95a402c3 CL |
760 | case -ENOMEM: |
761 | goto out; | |
e24f0b8f | 762 | case -EAGAIN: |
2d1db3b1 | 763 | retry++; |
e24f0b8f CL |
764 | break; |
765 | case 0: | |
e24f0b8f CL |
766 | break; |
767 | default: | |
2d1db3b1 | 768 | /* Permanent failure */ |
2d1db3b1 | 769 | nr_failed++; |
e24f0b8f | 770 | break; |
2d1db3b1 | 771 | } |
b20a3503 CL |
772 | } |
773 | } | |
95a402c3 CL |
774 | rc = 0; |
775 | out: | |
b20a3503 CL |
776 | if (!swapwrite) |
777 | current->flags &= ~PF_SWAPWRITE; | |
778 | ||
aaa994b3 | 779 | putback_lru_pages(from); |
b20a3503 | 780 | |
95a402c3 CL |
781 | if (rc) |
782 | return rc; | |
b20a3503 | 783 | |
95a402c3 | 784 | return nr_failed + retry; |
b20a3503 | 785 | } |
95a402c3 | 786 | |
742755a1 CL |
787 | #ifdef CONFIG_NUMA |
788 | /* | |
789 | * Move a list of individual pages | |
790 | */ | |
791 | struct page_to_node { | |
792 | unsigned long addr; | |
793 | struct page *page; | |
794 | int node; | |
795 | int status; | |
796 | }; | |
797 | ||
798 | static struct page *new_page_node(struct page *p, unsigned long private, | |
799 | int **result) | |
800 | { | |
801 | struct page_to_node *pm = (struct page_to_node *)private; | |
802 | ||
803 | while (pm->node != MAX_NUMNODES && pm->page != p) | |
804 | pm++; | |
805 | ||
806 | if (pm->node == MAX_NUMNODES) | |
807 | return NULL; | |
808 | ||
809 | *result = &pm->status; | |
810 | ||
6484eb3e | 811 | return alloc_pages_exact_node(pm->node, |
769848c0 | 812 | GFP_HIGHUSER_MOVABLE | GFP_THISNODE, 0); |
742755a1 CL |
813 | } |
814 | ||
815 | /* | |
816 | * Move a set of pages as indicated in the pm array. The addr | |
817 | * field must be set to the virtual address of the page to be moved | |
818 | * and the node number must contain a valid target node. | |
5e9a0f02 | 819 | * The pm array ends with node = MAX_NUMNODES. |
742755a1 | 820 | */ |
5e9a0f02 BG |
821 | static int do_move_page_to_node_array(struct mm_struct *mm, |
822 | struct page_to_node *pm, | |
823 | int migrate_all) | |
742755a1 CL |
824 | { |
825 | int err; | |
826 | struct page_to_node *pp; | |
827 | LIST_HEAD(pagelist); | |
828 | ||
829 | down_read(&mm->mmap_sem); | |
830 | ||
831 | /* | |
832 | * Build a list of pages to migrate | |
833 | */ | |
742755a1 CL |
834 | for (pp = pm; pp->node != MAX_NUMNODES; pp++) { |
835 | struct vm_area_struct *vma; | |
836 | struct page *page; | |
837 | ||
742755a1 CL |
838 | err = -EFAULT; |
839 | vma = find_vma(mm, pp->addr); | |
0dc952dc | 840 | if (!vma || !vma_migratable(vma)) |
742755a1 CL |
841 | goto set_status; |
842 | ||
843 | page = follow_page(vma, pp->addr, FOLL_GET); | |
89f5b7da LT |
844 | |
845 | err = PTR_ERR(page); | |
846 | if (IS_ERR(page)) | |
847 | goto set_status; | |
848 | ||
742755a1 CL |
849 | err = -ENOENT; |
850 | if (!page) | |
851 | goto set_status; | |
852 | ||
853 | if (PageReserved(page)) /* Check for zero page */ | |
854 | goto put_and_set; | |
855 | ||
856 | pp->page = page; | |
857 | err = page_to_nid(page); | |
858 | ||
859 | if (err == pp->node) | |
860 | /* | |
861 | * Node already in the right place | |
862 | */ | |
863 | goto put_and_set; | |
864 | ||
865 | err = -EACCES; | |
866 | if (page_mapcount(page) > 1 && | |
867 | !migrate_all) | |
868 | goto put_and_set; | |
869 | ||
62695a84 | 870 | err = isolate_lru_page(page); |
6d9c285a | 871 | if (!err) { |
62695a84 | 872 | list_add_tail(&page->lru, &pagelist); |
6d9c285a KM |
873 | inc_zone_page_state(page, NR_ISOLATED_ANON + |
874 | page_is_file_cache(page)); | |
875 | } | |
742755a1 CL |
876 | put_and_set: |
877 | /* | |
878 | * Either remove the duplicate refcount from | |
879 | * isolate_lru_page() or drop the page ref if it was | |
880 | * not isolated. | |
881 | */ | |
882 | put_page(page); | |
883 | set_status: | |
884 | pp->status = err; | |
885 | } | |
886 | ||
e78bbfa8 | 887 | err = 0; |
742755a1 CL |
888 | if (!list_empty(&pagelist)) |
889 | err = migrate_pages(&pagelist, new_page_node, | |
890 | (unsigned long)pm); | |
742755a1 CL |
891 | |
892 | up_read(&mm->mmap_sem); | |
893 | return err; | |
894 | } | |
895 | ||
5e9a0f02 BG |
896 | /* |
897 | * Migrate an array of page address onto an array of nodes and fill | |
898 | * the corresponding array of status. | |
899 | */ | |
900 | static int do_pages_move(struct mm_struct *mm, struct task_struct *task, | |
901 | unsigned long nr_pages, | |
902 | const void __user * __user *pages, | |
903 | const int __user *nodes, | |
904 | int __user *status, int flags) | |
905 | { | |
3140a227 | 906 | struct page_to_node *pm; |
5e9a0f02 | 907 | nodemask_t task_nodes; |
3140a227 BG |
908 | unsigned long chunk_nr_pages; |
909 | unsigned long chunk_start; | |
910 | int err; | |
5e9a0f02 BG |
911 | |
912 | task_nodes = cpuset_mems_allowed(task); | |
913 | ||
3140a227 BG |
914 | err = -ENOMEM; |
915 | pm = (struct page_to_node *)__get_free_page(GFP_KERNEL); | |
916 | if (!pm) | |
5e9a0f02 | 917 | goto out; |
35282a2d BG |
918 | |
919 | migrate_prep(); | |
920 | ||
5e9a0f02 | 921 | /* |
3140a227 BG |
922 | * Store a chunk of page_to_node array in a page, |
923 | * but keep the last one as a marker | |
5e9a0f02 | 924 | */ |
3140a227 | 925 | chunk_nr_pages = (PAGE_SIZE / sizeof(struct page_to_node)) - 1; |
5e9a0f02 | 926 | |
3140a227 BG |
927 | for (chunk_start = 0; |
928 | chunk_start < nr_pages; | |
929 | chunk_start += chunk_nr_pages) { | |
930 | int j; | |
5e9a0f02 | 931 | |
3140a227 BG |
932 | if (chunk_start + chunk_nr_pages > nr_pages) |
933 | chunk_nr_pages = nr_pages - chunk_start; | |
934 | ||
935 | /* fill the chunk pm with addrs and nodes from user-space */ | |
936 | for (j = 0; j < chunk_nr_pages; j++) { | |
937 | const void __user *p; | |
5e9a0f02 BG |
938 | int node; |
939 | ||
3140a227 BG |
940 | err = -EFAULT; |
941 | if (get_user(p, pages + j + chunk_start)) | |
942 | goto out_pm; | |
943 | pm[j].addr = (unsigned long) p; | |
944 | ||
945 | if (get_user(node, nodes + j + chunk_start)) | |
5e9a0f02 BG |
946 | goto out_pm; |
947 | ||
948 | err = -ENODEV; | |
949 | if (!node_state(node, N_HIGH_MEMORY)) | |
950 | goto out_pm; | |
951 | ||
952 | err = -EACCES; | |
953 | if (!node_isset(node, task_nodes)) | |
954 | goto out_pm; | |
955 | ||
3140a227 BG |
956 | pm[j].node = node; |
957 | } | |
958 | ||
959 | /* End marker for this chunk */ | |
960 | pm[chunk_nr_pages].node = MAX_NUMNODES; | |
961 | ||
962 | /* Migrate this chunk */ | |
963 | err = do_move_page_to_node_array(mm, pm, | |
964 | flags & MPOL_MF_MOVE_ALL); | |
965 | if (err < 0) | |
966 | goto out_pm; | |
5e9a0f02 | 967 | |
5e9a0f02 | 968 | /* Return status information */ |
3140a227 BG |
969 | for (j = 0; j < chunk_nr_pages; j++) |
970 | if (put_user(pm[j].status, status + j + chunk_start)) { | |
5e9a0f02 | 971 | err = -EFAULT; |
3140a227 BG |
972 | goto out_pm; |
973 | } | |
974 | } | |
975 | err = 0; | |
5e9a0f02 BG |
976 | |
977 | out_pm: | |
3140a227 | 978 | free_page((unsigned long)pm); |
5e9a0f02 BG |
979 | out: |
980 | return err; | |
981 | } | |
982 | ||
742755a1 | 983 | /* |
2f007e74 | 984 | * Determine the nodes of an array of pages and store it in an array of status. |
742755a1 | 985 | */ |
80bba129 BG |
986 | static void do_pages_stat_array(struct mm_struct *mm, unsigned long nr_pages, |
987 | const void __user **pages, int *status) | |
742755a1 | 988 | { |
2f007e74 | 989 | unsigned long i; |
2f007e74 | 990 | |
742755a1 CL |
991 | down_read(&mm->mmap_sem); |
992 | ||
2f007e74 | 993 | for (i = 0; i < nr_pages; i++) { |
80bba129 | 994 | unsigned long addr = (unsigned long)(*pages); |
742755a1 CL |
995 | struct vm_area_struct *vma; |
996 | struct page *page; | |
c095adbc | 997 | int err = -EFAULT; |
2f007e74 BG |
998 | |
999 | vma = find_vma(mm, addr); | |
742755a1 CL |
1000 | if (!vma) |
1001 | goto set_status; | |
1002 | ||
2f007e74 | 1003 | page = follow_page(vma, addr, 0); |
89f5b7da LT |
1004 | |
1005 | err = PTR_ERR(page); | |
1006 | if (IS_ERR(page)) | |
1007 | goto set_status; | |
1008 | ||
742755a1 CL |
1009 | err = -ENOENT; |
1010 | /* Use PageReserved to check for zero page */ | |
1011 | if (!page || PageReserved(page)) | |
1012 | goto set_status; | |
1013 | ||
1014 | err = page_to_nid(page); | |
1015 | set_status: | |
80bba129 BG |
1016 | *status = err; |
1017 | ||
1018 | pages++; | |
1019 | status++; | |
1020 | } | |
1021 | ||
1022 | up_read(&mm->mmap_sem); | |
1023 | } | |
1024 | ||
1025 | /* | |
1026 | * Determine the nodes of a user array of pages and store it in | |
1027 | * a user array of status. | |
1028 | */ | |
1029 | static int do_pages_stat(struct mm_struct *mm, unsigned long nr_pages, | |
1030 | const void __user * __user *pages, | |
1031 | int __user *status) | |
1032 | { | |
1033 | #define DO_PAGES_STAT_CHUNK_NR 16 | |
1034 | const void __user *chunk_pages[DO_PAGES_STAT_CHUNK_NR]; | |
1035 | int chunk_status[DO_PAGES_STAT_CHUNK_NR]; | |
1036 | unsigned long i, chunk_nr = DO_PAGES_STAT_CHUNK_NR; | |
1037 | int err; | |
1038 | ||
1039 | for (i = 0; i < nr_pages; i += chunk_nr) { | |
b9255850 | 1040 | if (chunk_nr > nr_pages - i) |
80bba129 BG |
1041 | chunk_nr = nr_pages - i; |
1042 | ||
1043 | err = copy_from_user(chunk_pages, &pages[i], | |
1044 | chunk_nr * sizeof(*chunk_pages)); | |
1045 | if (err) { | |
1046 | err = -EFAULT; | |
1047 | goto out; | |
1048 | } | |
1049 | ||
1050 | do_pages_stat_array(mm, chunk_nr, chunk_pages, chunk_status); | |
1051 | ||
1052 | err = copy_to_user(&status[i], chunk_status, | |
1053 | chunk_nr * sizeof(*chunk_status)); | |
1054 | if (err) { | |
1055 | err = -EFAULT; | |
1056 | goto out; | |
1057 | } | |
742755a1 | 1058 | } |
2f007e74 | 1059 | err = 0; |
742755a1 | 1060 | |
2f007e74 | 1061 | out: |
2f007e74 | 1062 | return err; |
742755a1 CL |
1063 | } |
1064 | ||
1065 | /* | |
1066 | * Move a list of pages in the address space of the currently executing | |
1067 | * process. | |
1068 | */ | |
938bb9f5 HC |
1069 | SYSCALL_DEFINE6(move_pages, pid_t, pid, unsigned long, nr_pages, |
1070 | const void __user * __user *, pages, | |
1071 | const int __user *, nodes, | |
1072 | int __user *, status, int, flags) | |
742755a1 | 1073 | { |
c69e8d9c | 1074 | const struct cred *cred = current_cred(), *tcred; |
742755a1 | 1075 | struct task_struct *task; |
742755a1 | 1076 | struct mm_struct *mm; |
5e9a0f02 | 1077 | int err; |
742755a1 CL |
1078 | |
1079 | /* Check flags */ | |
1080 | if (flags & ~(MPOL_MF_MOVE|MPOL_MF_MOVE_ALL)) | |
1081 | return -EINVAL; | |
1082 | ||
1083 | if ((flags & MPOL_MF_MOVE_ALL) && !capable(CAP_SYS_NICE)) | |
1084 | return -EPERM; | |
1085 | ||
1086 | /* Find the mm_struct */ | |
1087 | read_lock(&tasklist_lock); | |
228ebcbe | 1088 | task = pid ? find_task_by_vpid(pid) : current; |
742755a1 CL |
1089 | if (!task) { |
1090 | read_unlock(&tasklist_lock); | |
1091 | return -ESRCH; | |
1092 | } | |
1093 | mm = get_task_mm(task); | |
1094 | read_unlock(&tasklist_lock); | |
1095 | ||
1096 | if (!mm) | |
1097 | return -EINVAL; | |
1098 | ||
1099 | /* | |
1100 | * Check if this process has the right to modify the specified | |
1101 | * process. The right exists if the process has administrative | |
1102 | * capabilities, superuser privileges or the same | |
1103 | * userid as the target process. | |
1104 | */ | |
c69e8d9c DH |
1105 | rcu_read_lock(); |
1106 | tcred = __task_cred(task); | |
b6dff3ec DH |
1107 | if (cred->euid != tcred->suid && cred->euid != tcred->uid && |
1108 | cred->uid != tcred->suid && cred->uid != tcred->uid && | |
742755a1 | 1109 | !capable(CAP_SYS_NICE)) { |
c69e8d9c | 1110 | rcu_read_unlock(); |
742755a1 | 1111 | err = -EPERM; |
5e9a0f02 | 1112 | goto out; |
742755a1 | 1113 | } |
c69e8d9c | 1114 | rcu_read_unlock(); |
742755a1 | 1115 | |
86c3a764 DQ |
1116 | err = security_task_movememory(task); |
1117 | if (err) | |
5e9a0f02 | 1118 | goto out; |
86c3a764 | 1119 | |
5e9a0f02 BG |
1120 | if (nodes) { |
1121 | err = do_pages_move(mm, task, nr_pages, pages, nodes, status, | |
1122 | flags); | |
1123 | } else { | |
2f007e74 | 1124 | err = do_pages_stat(mm, nr_pages, pages, status); |
742755a1 CL |
1125 | } |
1126 | ||
742755a1 | 1127 | out: |
742755a1 CL |
1128 | mmput(mm); |
1129 | return err; | |
1130 | } | |
742755a1 | 1131 | |
7b2259b3 CL |
1132 | /* |
1133 | * Call migration functions in the vma_ops that may prepare | |
1134 | * memory in a vm for migration. migration functions may perform | |
1135 | * the migration for vmas that do not have an underlying page struct. | |
1136 | */ | |
1137 | int migrate_vmas(struct mm_struct *mm, const nodemask_t *to, | |
1138 | const nodemask_t *from, unsigned long flags) | |
1139 | { | |
1140 | struct vm_area_struct *vma; | |
1141 | int err = 0; | |
1142 | ||
1001c9fb | 1143 | for (vma = mm->mmap; vma && !err; vma = vma->vm_next) { |
7b2259b3 CL |
1144 | if (vma->vm_ops && vma->vm_ops->migrate) { |
1145 | err = vma->vm_ops->migrate(vma, to, from, flags); | |
1146 | if (err) | |
1147 | break; | |
1148 | } | |
1149 | } | |
1150 | return err; | |
1151 | } | |
83d1674a | 1152 | #endif |