]>
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 | 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> |
290408d4 | 35 | #include <linux/hugetlb.h> |
5a0e3ad6 | 36 | #include <linux/gfp.h> |
b20a3503 | 37 | |
0d1836c3 MN |
38 | #include <asm/tlbflush.h> |
39 | ||
b20a3503 CL |
40 | #include "internal.h" |
41 | ||
b20a3503 CL |
42 | #define lru_to_page(_head) (list_entry((_head)->prev, struct page, lru)) |
43 | ||
b20a3503 | 44 | /* |
742755a1 | 45 | * migrate_prep() needs to be called before we start compiling a list of pages |
748446bb MG |
46 | * to be migrated using isolate_lru_page(). If scheduling work on other CPUs is |
47 | * undesirable, use migrate_prep_local() | |
b20a3503 CL |
48 | */ |
49 | int migrate_prep(void) | |
50 | { | |
b20a3503 CL |
51 | /* |
52 | * Clear the LRU lists so pages can be isolated. | |
53 | * Note that pages may be moved off the LRU after we have | |
54 | * drained them. Those pages will fail to migrate like other | |
55 | * pages that may be busy. | |
56 | */ | |
57 | lru_add_drain_all(); | |
58 | ||
59 | return 0; | |
60 | } | |
61 | ||
748446bb MG |
62 | /* Do the necessary work of migrate_prep but not if it involves other CPUs */ |
63 | int migrate_prep_local(void) | |
64 | { | |
65 | lru_add_drain(); | |
66 | ||
67 | return 0; | |
68 | } | |
69 | ||
b20a3503 | 70 | /* |
894bc310 LS |
71 | * Add isolated pages on the list back to the LRU under page lock |
72 | * to avoid leaking evictable pages back onto unevictable list. | |
b20a3503 | 73 | */ |
e13861d8 | 74 | void putback_lru_pages(struct list_head *l) |
b20a3503 CL |
75 | { |
76 | struct page *page; | |
77 | struct page *page2; | |
b20a3503 CL |
78 | |
79 | list_for_each_entry_safe(page, page2, l, lru) { | |
e24f0b8f | 80 | list_del(&page->lru); |
a731286d | 81 | dec_zone_page_state(page, NR_ISOLATED_ANON + |
6c0b1351 | 82 | page_is_file_cache(page)); |
894bc310 | 83 | putback_lru_page(page); |
b20a3503 | 84 | } |
b20a3503 CL |
85 | } |
86 | ||
0697212a CL |
87 | /* |
88 | * Restore a potential migration pte to a working pte entry | |
89 | */ | |
e9995ef9 HD |
90 | static int remove_migration_pte(struct page *new, struct vm_area_struct *vma, |
91 | unsigned long addr, void *old) | |
0697212a CL |
92 | { |
93 | struct mm_struct *mm = vma->vm_mm; | |
94 | swp_entry_t entry; | |
95 | pgd_t *pgd; | |
96 | pud_t *pud; | |
97 | pmd_t *pmd; | |
98 | pte_t *ptep, pte; | |
99 | spinlock_t *ptl; | |
100 | ||
290408d4 NH |
101 | if (unlikely(PageHuge(new))) { |
102 | ptep = huge_pte_offset(mm, addr); | |
103 | if (!ptep) | |
104 | goto out; | |
105 | ptl = &mm->page_table_lock; | |
106 | } else { | |
107 | pgd = pgd_offset(mm, addr); | |
108 | if (!pgd_present(*pgd)) | |
109 | goto out; | |
0697212a | 110 | |
290408d4 NH |
111 | pud = pud_offset(pgd, addr); |
112 | if (!pud_present(*pud)) | |
113 | goto out; | |
0697212a | 114 | |
290408d4 NH |
115 | pmd = pmd_offset(pud, addr); |
116 | if (!pmd_present(*pmd)) | |
117 | goto out; | |
0697212a | 118 | |
290408d4 | 119 | ptep = pte_offset_map(pmd, addr); |
0697212a | 120 | |
290408d4 NH |
121 | if (!is_swap_pte(*ptep)) { |
122 | pte_unmap(ptep); | |
123 | goto out; | |
124 | } | |
125 | ||
126 | ptl = pte_lockptr(mm, pmd); | |
127 | } | |
0697212a | 128 | |
0697212a CL |
129 | spin_lock(ptl); |
130 | pte = *ptep; | |
131 | if (!is_swap_pte(pte)) | |
e9995ef9 | 132 | goto unlock; |
0697212a CL |
133 | |
134 | entry = pte_to_swp_entry(pte); | |
135 | ||
e9995ef9 HD |
136 | if (!is_migration_entry(entry) || |
137 | migration_entry_to_page(entry) != old) | |
138 | goto unlock; | |
0697212a | 139 | |
0697212a CL |
140 | get_page(new); |
141 | pte = pte_mkold(mk_pte(new, vma->vm_page_prot)); | |
142 | if (is_write_migration_entry(entry)) | |
143 | pte = pte_mkwrite(pte); | |
3ef8fd7f | 144 | #ifdef CONFIG_HUGETLB_PAGE |
290408d4 NH |
145 | if (PageHuge(new)) |
146 | pte = pte_mkhuge(pte); | |
3ef8fd7f | 147 | #endif |
97ee0524 | 148 | flush_cache_page(vma, addr, pte_pfn(pte)); |
0697212a | 149 | set_pte_at(mm, addr, ptep, pte); |
04e62a29 | 150 | |
290408d4 NH |
151 | if (PageHuge(new)) { |
152 | if (PageAnon(new)) | |
153 | hugepage_add_anon_rmap(new, vma, addr); | |
154 | else | |
155 | page_dup_rmap(new); | |
156 | } else if (PageAnon(new)) | |
04e62a29 CL |
157 | page_add_anon_rmap(new, vma, addr); |
158 | else | |
159 | page_add_file_rmap(new); | |
160 | ||
161 | /* No need to invalidate - it was non-present before */ | |
4b3073e1 | 162 | update_mmu_cache(vma, addr, ptep); |
e9995ef9 | 163 | unlock: |
0697212a | 164 | pte_unmap_unlock(ptep, ptl); |
e9995ef9 HD |
165 | out: |
166 | return SWAP_AGAIN; | |
0697212a CL |
167 | } |
168 | ||
04e62a29 CL |
169 | /* |
170 | * Get rid of all migration entries and replace them by | |
171 | * references to the indicated page. | |
172 | */ | |
173 | static void remove_migration_ptes(struct page *old, struct page *new) | |
174 | { | |
e9995ef9 | 175 | rmap_walk(new, remove_migration_pte, old); |
04e62a29 CL |
176 | } |
177 | ||
0697212a CL |
178 | /* |
179 | * Something used the pte of a page under migration. We need to | |
180 | * get to the page and wait until migration is finished. | |
181 | * When we return from this function the fault will be retried. | |
182 | * | |
183 | * This function is called from do_swap_page(). | |
184 | */ | |
185 | void migration_entry_wait(struct mm_struct *mm, pmd_t *pmd, | |
186 | unsigned long address) | |
187 | { | |
188 | pte_t *ptep, pte; | |
189 | spinlock_t *ptl; | |
190 | swp_entry_t entry; | |
191 | struct page *page; | |
192 | ||
193 | ptep = pte_offset_map_lock(mm, pmd, address, &ptl); | |
194 | pte = *ptep; | |
195 | if (!is_swap_pte(pte)) | |
196 | goto out; | |
197 | ||
198 | entry = pte_to_swp_entry(pte); | |
199 | if (!is_migration_entry(entry)) | |
200 | goto out; | |
201 | ||
202 | page = migration_entry_to_page(entry); | |
203 | ||
e286781d NP |
204 | /* |
205 | * Once radix-tree replacement of page migration started, page_count | |
206 | * *must* be zero. And, we don't want to call wait_on_page_locked() | |
207 | * against a page without get_page(). | |
208 | * So, we use get_page_unless_zero(), here. Even failed, page fault | |
209 | * will occur again. | |
210 | */ | |
211 | if (!get_page_unless_zero(page)) | |
212 | goto out; | |
0697212a CL |
213 | pte_unmap_unlock(ptep, ptl); |
214 | wait_on_page_locked(page); | |
215 | put_page(page); | |
216 | return; | |
217 | out: | |
218 | pte_unmap_unlock(ptep, ptl); | |
219 | } | |
220 | ||
b20a3503 | 221 | /* |
c3fcf8a5 | 222 | * Replace the page in the mapping. |
5b5c7120 CL |
223 | * |
224 | * The number of remaining references must be: | |
225 | * 1 for anonymous pages without a mapping | |
226 | * 2 for pages with a mapping | |
266cf658 | 227 | * 3 for pages with a mapping and PagePrivate/PagePrivate2 set. |
b20a3503 | 228 | */ |
2d1db3b1 CL |
229 | static int migrate_page_move_mapping(struct address_space *mapping, |
230 | struct page *newpage, struct page *page) | |
b20a3503 | 231 | { |
e286781d | 232 | int expected_count; |
7cf9c2c7 | 233 | void **pslot; |
b20a3503 | 234 | |
6c5240ae | 235 | if (!mapping) { |
0e8c7d0f | 236 | /* Anonymous page without mapping */ |
6c5240ae CL |
237 | if (page_count(page) != 1) |
238 | return -EAGAIN; | |
239 | return 0; | |
240 | } | |
241 | ||
19fd6231 | 242 | spin_lock_irq(&mapping->tree_lock); |
b20a3503 | 243 | |
7cf9c2c7 NP |
244 | pslot = radix_tree_lookup_slot(&mapping->page_tree, |
245 | page_index(page)); | |
b20a3503 | 246 | |
edcf4748 | 247 | expected_count = 2 + page_has_private(page); |
e286781d | 248 | if (page_count(page) != expected_count || |
7cf9c2c7 | 249 | (struct page *)radix_tree_deref_slot(pslot) != page) { |
19fd6231 | 250 | spin_unlock_irq(&mapping->tree_lock); |
e23ca00b | 251 | return -EAGAIN; |
b20a3503 CL |
252 | } |
253 | ||
e286781d | 254 | if (!page_freeze_refs(page, expected_count)) { |
19fd6231 | 255 | spin_unlock_irq(&mapping->tree_lock); |
e286781d NP |
256 | return -EAGAIN; |
257 | } | |
258 | ||
b20a3503 CL |
259 | /* |
260 | * Now we know that no one else is looking at the page. | |
b20a3503 | 261 | */ |
7cf9c2c7 | 262 | get_page(newpage); /* add cache reference */ |
b20a3503 CL |
263 | if (PageSwapCache(page)) { |
264 | SetPageSwapCache(newpage); | |
265 | set_page_private(newpage, page_private(page)); | |
266 | } | |
267 | ||
7cf9c2c7 NP |
268 | radix_tree_replace_slot(pslot, newpage); |
269 | ||
e286781d | 270 | page_unfreeze_refs(page, expected_count); |
7cf9c2c7 NP |
271 | /* |
272 | * Drop cache reference from old page. | |
273 | * We know this isn't the last reference. | |
274 | */ | |
b20a3503 | 275 | __put_page(page); |
7cf9c2c7 | 276 | |
0e8c7d0f CL |
277 | /* |
278 | * If moved to a different zone then also account | |
279 | * the page for that zone. Other VM counters will be | |
280 | * taken care of when we establish references to the | |
281 | * new page and drop references to the old page. | |
282 | * | |
283 | * Note that anonymous pages are accounted for | |
284 | * via NR_FILE_PAGES and NR_ANON_PAGES if they | |
285 | * are mapped to swap space. | |
286 | */ | |
287 | __dec_zone_page_state(page, NR_FILE_PAGES); | |
288 | __inc_zone_page_state(newpage, NR_FILE_PAGES); | |
4b02108a KM |
289 | if (PageSwapBacked(page)) { |
290 | __dec_zone_page_state(page, NR_SHMEM); | |
291 | __inc_zone_page_state(newpage, NR_SHMEM); | |
292 | } | |
19fd6231 | 293 | spin_unlock_irq(&mapping->tree_lock); |
b20a3503 CL |
294 | |
295 | return 0; | |
296 | } | |
b20a3503 | 297 | |
290408d4 NH |
298 | /* |
299 | * The expected number of remaining references is the same as that | |
300 | * of migrate_page_move_mapping(). | |
301 | */ | |
302 | int migrate_huge_page_move_mapping(struct address_space *mapping, | |
303 | struct page *newpage, struct page *page) | |
304 | { | |
305 | int expected_count; | |
306 | void **pslot; | |
307 | ||
308 | if (!mapping) { | |
309 | if (page_count(page) != 1) | |
310 | return -EAGAIN; | |
311 | return 0; | |
312 | } | |
313 | ||
314 | spin_lock_irq(&mapping->tree_lock); | |
315 | ||
316 | pslot = radix_tree_lookup_slot(&mapping->page_tree, | |
317 | page_index(page)); | |
318 | ||
319 | expected_count = 2 + page_has_private(page); | |
320 | if (page_count(page) != expected_count || | |
321 | (struct page *)radix_tree_deref_slot(pslot) != page) { | |
322 | spin_unlock_irq(&mapping->tree_lock); | |
323 | return -EAGAIN; | |
324 | } | |
325 | ||
326 | if (!page_freeze_refs(page, expected_count)) { | |
327 | spin_unlock_irq(&mapping->tree_lock); | |
328 | return -EAGAIN; | |
329 | } | |
330 | ||
331 | get_page(newpage); | |
332 | ||
333 | radix_tree_replace_slot(pslot, newpage); | |
334 | ||
335 | page_unfreeze_refs(page, expected_count); | |
336 | ||
337 | __put_page(page); | |
338 | ||
339 | spin_unlock_irq(&mapping->tree_lock); | |
340 | return 0; | |
341 | } | |
342 | ||
b20a3503 CL |
343 | /* |
344 | * Copy the page to its new location | |
345 | */ | |
290408d4 | 346 | void migrate_page_copy(struct page *newpage, struct page *page) |
b20a3503 | 347 | { |
290408d4 NH |
348 | if (PageHuge(page)) |
349 | copy_huge_page(newpage, page); | |
350 | else | |
351 | copy_highpage(newpage, page); | |
b20a3503 CL |
352 | |
353 | if (PageError(page)) | |
354 | SetPageError(newpage); | |
355 | if (PageReferenced(page)) | |
356 | SetPageReferenced(newpage); | |
357 | if (PageUptodate(page)) | |
358 | SetPageUptodate(newpage); | |
894bc310 LS |
359 | if (TestClearPageActive(page)) { |
360 | VM_BUG_ON(PageUnevictable(page)); | |
b20a3503 | 361 | SetPageActive(newpage); |
418b27ef LS |
362 | } else if (TestClearPageUnevictable(page)) |
363 | SetPageUnevictable(newpage); | |
b20a3503 CL |
364 | if (PageChecked(page)) |
365 | SetPageChecked(newpage); | |
366 | if (PageMappedToDisk(page)) | |
367 | SetPageMappedToDisk(newpage); | |
368 | ||
369 | if (PageDirty(page)) { | |
370 | clear_page_dirty_for_io(page); | |
3a902c5f NP |
371 | /* |
372 | * Want to mark the page and the radix tree as dirty, and | |
373 | * redo the accounting that clear_page_dirty_for_io undid, | |
374 | * but we can't use set_page_dirty because that function | |
375 | * is actually a signal that all of the page has become dirty. | |
376 | * Wheras only part of our page may be dirty. | |
377 | */ | |
378 | __set_page_dirty_nobuffers(newpage); | |
b20a3503 CL |
379 | } |
380 | ||
b291f000 | 381 | mlock_migrate_page(newpage, page); |
e9995ef9 | 382 | ksm_migrate_page(newpage, page); |
b291f000 | 383 | |
b20a3503 | 384 | ClearPageSwapCache(page); |
b20a3503 CL |
385 | ClearPagePrivate(page); |
386 | set_page_private(page, 0); | |
387 | page->mapping = NULL; | |
388 | ||
389 | /* | |
390 | * If any waiters have accumulated on the new page then | |
391 | * wake them up. | |
392 | */ | |
393 | if (PageWriteback(newpage)) | |
394 | end_page_writeback(newpage); | |
395 | } | |
b20a3503 | 396 | |
1d8b85cc CL |
397 | /************************************************************ |
398 | * Migration functions | |
399 | ***********************************************************/ | |
400 | ||
401 | /* Always fail migration. Used for mappings that are not movable */ | |
2d1db3b1 CL |
402 | int fail_migrate_page(struct address_space *mapping, |
403 | struct page *newpage, struct page *page) | |
1d8b85cc CL |
404 | { |
405 | return -EIO; | |
406 | } | |
407 | EXPORT_SYMBOL(fail_migrate_page); | |
408 | ||
b20a3503 CL |
409 | /* |
410 | * Common logic to directly migrate a single page suitable for | |
266cf658 | 411 | * pages that do not use PagePrivate/PagePrivate2. |
b20a3503 CL |
412 | * |
413 | * Pages are locked upon entry and exit. | |
414 | */ | |
2d1db3b1 CL |
415 | int migrate_page(struct address_space *mapping, |
416 | struct page *newpage, struct page *page) | |
b20a3503 CL |
417 | { |
418 | int rc; | |
419 | ||
420 | BUG_ON(PageWriteback(page)); /* Writeback must be complete */ | |
421 | ||
2d1db3b1 | 422 | rc = migrate_page_move_mapping(mapping, newpage, page); |
b20a3503 CL |
423 | |
424 | if (rc) | |
425 | return rc; | |
426 | ||
427 | migrate_page_copy(newpage, page); | |
b20a3503 CL |
428 | return 0; |
429 | } | |
430 | EXPORT_SYMBOL(migrate_page); | |
431 | ||
9361401e | 432 | #ifdef CONFIG_BLOCK |
1d8b85cc CL |
433 | /* |
434 | * Migration function for pages with buffers. This function can only be used | |
435 | * if the underlying filesystem guarantees that no other references to "page" | |
436 | * exist. | |
437 | */ | |
2d1db3b1 CL |
438 | int buffer_migrate_page(struct address_space *mapping, |
439 | struct page *newpage, struct page *page) | |
1d8b85cc | 440 | { |
1d8b85cc CL |
441 | struct buffer_head *bh, *head; |
442 | int rc; | |
443 | ||
1d8b85cc | 444 | if (!page_has_buffers(page)) |
2d1db3b1 | 445 | return migrate_page(mapping, newpage, page); |
1d8b85cc CL |
446 | |
447 | head = page_buffers(page); | |
448 | ||
2d1db3b1 | 449 | rc = migrate_page_move_mapping(mapping, newpage, page); |
1d8b85cc CL |
450 | |
451 | if (rc) | |
452 | return rc; | |
453 | ||
454 | bh = head; | |
455 | do { | |
456 | get_bh(bh); | |
457 | lock_buffer(bh); | |
458 | bh = bh->b_this_page; | |
459 | ||
460 | } while (bh != head); | |
461 | ||
462 | ClearPagePrivate(page); | |
463 | set_page_private(newpage, page_private(page)); | |
464 | set_page_private(page, 0); | |
465 | put_page(page); | |
466 | get_page(newpage); | |
467 | ||
468 | bh = head; | |
469 | do { | |
470 | set_bh_page(bh, newpage, bh_offset(bh)); | |
471 | bh = bh->b_this_page; | |
472 | ||
473 | } while (bh != head); | |
474 | ||
475 | SetPagePrivate(newpage); | |
476 | ||
477 | migrate_page_copy(newpage, page); | |
478 | ||
479 | bh = head; | |
480 | do { | |
481 | unlock_buffer(bh); | |
482 | put_bh(bh); | |
483 | bh = bh->b_this_page; | |
484 | ||
485 | } while (bh != head); | |
486 | ||
487 | return 0; | |
488 | } | |
489 | EXPORT_SYMBOL(buffer_migrate_page); | |
9361401e | 490 | #endif |
1d8b85cc | 491 | |
04e62a29 CL |
492 | /* |
493 | * Writeback a page to clean the dirty state | |
494 | */ | |
495 | static int writeout(struct address_space *mapping, struct page *page) | |
8351a6e4 | 496 | { |
04e62a29 CL |
497 | struct writeback_control wbc = { |
498 | .sync_mode = WB_SYNC_NONE, | |
499 | .nr_to_write = 1, | |
500 | .range_start = 0, | |
501 | .range_end = LLONG_MAX, | |
04e62a29 CL |
502 | .for_reclaim = 1 |
503 | }; | |
504 | int rc; | |
505 | ||
506 | if (!mapping->a_ops->writepage) | |
507 | /* No write method for the address space */ | |
508 | return -EINVAL; | |
509 | ||
510 | if (!clear_page_dirty_for_io(page)) | |
511 | /* Someone else already triggered a write */ | |
512 | return -EAGAIN; | |
513 | ||
8351a6e4 | 514 | /* |
04e62a29 CL |
515 | * A dirty page may imply that the underlying filesystem has |
516 | * the page on some queue. So the page must be clean for | |
517 | * migration. Writeout may mean we loose the lock and the | |
518 | * page state is no longer what we checked for earlier. | |
519 | * At this point we know that the migration attempt cannot | |
520 | * be successful. | |
8351a6e4 | 521 | */ |
04e62a29 | 522 | remove_migration_ptes(page, page); |
8351a6e4 | 523 | |
04e62a29 | 524 | rc = mapping->a_ops->writepage(page, &wbc); |
8351a6e4 | 525 | |
04e62a29 CL |
526 | if (rc != AOP_WRITEPAGE_ACTIVATE) |
527 | /* unlocked. Relock */ | |
528 | lock_page(page); | |
529 | ||
bda8550d | 530 | return (rc < 0) ? -EIO : -EAGAIN; |
04e62a29 CL |
531 | } |
532 | ||
533 | /* | |
534 | * Default handling if a filesystem does not provide a migration function. | |
535 | */ | |
536 | static int fallback_migrate_page(struct address_space *mapping, | |
537 | struct page *newpage, struct page *page) | |
538 | { | |
539 | if (PageDirty(page)) | |
540 | return writeout(mapping, page); | |
8351a6e4 CL |
541 | |
542 | /* | |
543 | * Buffers may be managed in a filesystem specific way. | |
544 | * We must have no buffers or drop them. | |
545 | */ | |
266cf658 | 546 | if (page_has_private(page) && |
8351a6e4 CL |
547 | !try_to_release_page(page, GFP_KERNEL)) |
548 | return -EAGAIN; | |
549 | ||
550 | return migrate_page(mapping, newpage, page); | |
551 | } | |
552 | ||
e24f0b8f CL |
553 | /* |
554 | * Move a page to a newly allocated page | |
555 | * The page is locked and all ptes have been successfully removed. | |
556 | * | |
557 | * The new page will have replaced the old page if this function | |
558 | * is successful. | |
894bc310 LS |
559 | * |
560 | * Return value: | |
561 | * < 0 - error code | |
562 | * == 0 - success | |
e24f0b8f | 563 | */ |
3fe2011f MG |
564 | static int move_to_new_page(struct page *newpage, struct page *page, |
565 | int remap_swapcache) | |
e24f0b8f CL |
566 | { |
567 | struct address_space *mapping; | |
568 | int rc; | |
569 | ||
570 | /* | |
571 | * Block others from accessing the page when we get around to | |
572 | * establishing additional references. We are the only one | |
573 | * holding a reference to the new page at this point. | |
574 | */ | |
529ae9aa | 575 | if (!trylock_page(newpage)) |
e24f0b8f CL |
576 | BUG(); |
577 | ||
578 | /* Prepare mapping for the new page.*/ | |
579 | newpage->index = page->index; | |
580 | newpage->mapping = page->mapping; | |
b2e18538 RR |
581 | if (PageSwapBacked(page)) |
582 | SetPageSwapBacked(newpage); | |
e24f0b8f CL |
583 | |
584 | mapping = page_mapping(page); | |
585 | if (!mapping) | |
586 | rc = migrate_page(mapping, newpage, page); | |
587 | else if (mapping->a_ops->migratepage) | |
588 | /* | |
589 | * Most pages have a mapping and most filesystems | |
590 | * should provide a migration function. Anonymous | |
591 | * pages are part of swap space which also has its | |
592 | * own migration function. This is the most common | |
593 | * path for page migration. | |
594 | */ | |
595 | rc = mapping->a_ops->migratepage(mapping, | |
596 | newpage, page); | |
597 | else | |
598 | rc = fallback_migrate_page(mapping, newpage, page); | |
599 | ||
3fe2011f | 600 | if (rc) { |
e24f0b8f | 601 | newpage->mapping = NULL; |
3fe2011f MG |
602 | } else { |
603 | if (remap_swapcache) | |
604 | remove_migration_ptes(page, newpage); | |
605 | } | |
e24f0b8f CL |
606 | |
607 | unlock_page(newpage); | |
608 | ||
609 | return rc; | |
610 | } | |
611 | ||
612 | /* | |
613 | * Obtain the lock on page, remove all ptes and migrate the page | |
614 | * to the newly allocated page in newpage. | |
615 | */ | |
95a402c3 | 616 | static int unmap_and_move(new_page_t get_new_page, unsigned long private, |
77f1fe6b | 617 | struct page *page, int force, int offlining, bool sync) |
e24f0b8f CL |
618 | { |
619 | int rc = 0; | |
742755a1 CL |
620 | int *result = NULL; |
621 | struct page *newpage = get_new_page(page, private, &result); | |
3fe2011f | 622 | int remap_swapcache = 1; |
989f89c5 | 623 | int rcu_locked = 0; |
ae41be37 | 624 | int charge = 0; |
e00e4316 | 625 | struct mem_cgroup *mem = NULL; |
3f6c8272 | 626 | struct anon_vma *anon_vma = NULL; |
95a402c3 CL |
627 | |
628 | if (!newpage) | |
629 | return -ENOMEM; | |
e24f0b8f | 630 | |
894bc310 | 631 | if (page_count(page) == 1) { |
e24f0b8f | 632 | /* page was freed from under us. So we are done. */ |
95a402c3 | 633 | goto move_newpage; |
894bc310 | 634 | } |
e24f0b8f | 635 | |
e8589cc1 | 636 | /* prepare cgroup just returns 0 or -ENOMEM */ |
e24f0b8f | 637 | rc = -EAGAIN; |
01b1ae63 | 638 | |
529ae9aa | 639 | if (!trylock_page(page)) { |
e24f0b8f | 640 | if (!force) |
95a402c3 | 641 | goto move_newpage; |
3e7d3449 MG |
642 | |
643 | /* | |
644 | * It's not safe for direct compaction to call lock_page. | |
645 | * For example, during page readahead pages are added locked | |
646 | * to the LRU. Later, when the IO completes the pages are | |
647 | * marked uptodate and unlocked. However, the queueing | |
648 | * could be merging multiple pages for one bio (e.g. | |
649 | * mpage_readpages). If an allocation happens for the | |
650 | * second or third page, the process can end up locking | |
651 | * the same page twice and deadlocking. Rather than | |
652 | * trying to be clever about what pages can be locked, | |
653 | * avoid the use of lock_page for direct compaction | |
654 | * altogether. | |
655 | */ | |
656 | if (current->flags & PF_MEMALLOC) | |
657 | goto move_newpage; | |
658 | ||
e24f0b8f CL |
659 | lock_page(page); |
660 | } | |
661 | ||
62b61f61 HD |
662 | /* |
663 | * Only memory hotplug's offline_pages() caller has locked out KSM, | |
664 | * and can safely migrate a KSM page. The other cases have skipped | |
665 | * PageKsm along with PageReserved - but it is only now when we have | |
666 | * the page lock that we can be certain it will not go KSM beneath us | |
667 | * (KSM will not upgrade a page from PageAnon to PageKsm when it sees | |
668 | * its pagecount raised, but only here do we take the page lock which | |
669 | * serializes that). | |
670 | */ | |
671 | if (PageKsm(page) && !offlining) { | |
672 | rc = -EBUSY; | |
673 | goto unlock; | |
674 | } | |
675 | ||
01b1ae63 | 676 | /* charge against new page */ |
ac39cf8c | 677 | charge = mem_cgroup_prepare_migration(page, newpage, &mem); |
01b1ae63 KH |
678 | if (charge == -ENOMEM) { |
679 | rc = -ENOMEM; | |
680 | goto unlock; | |
681 | } | |
682 | BUG_ON(charge); | |
683 | ||
e24f0b8f | 684 | if (PageWriteback(page)) { |
77f1fe6b | 685 | if (!force || !sync) |
01b1ae63 | 686 | goto uncharge; |
e24f0b8f CL |
687 | wait_on_page_writeback(page); |
688 | } | |
e24f0b8f | 689 | /* |
dc386d4d KH |
690 | * By try_to_unmap(), page->mapcount goes down to 0 here. In this case, |
691 | * we cannot notice that anon_vma is freed while we migrates a page. | |
692 | * This rcu_read_lock() delays freeing anon_vma pointer until the end | |
693 | * of migration. File cache pages are no problem because of page_lock() | |
989f89c5 KH |
694 | * File Caches may use write_page() or lock_page() in migration, then, |
695 | * just care Anon page here. | |
dc386d4d | 696 | */ |
989f89c5 KH |
697 | if (PageAnon(page)) { |
698 | rcu_read_lock(); | |
699 | rcu_locked = 1; | |
67b9509b | 700 | |
3fe2011f MG |
701 | /* Determine how to safely use anon_vma */ |
702 | if (!page_mapped(page)) { | |
703 | if (!PageSwapCache(page)) | |
704 | goto rcu_unlock; | |
67b9509b | 705 | |
3fe2011f MG |
706 | /* |
707 | * We cannot be sure that the anon_vma of an unmapped | |
708 | * swapcache page is safe to use because we don't | |
709 | * know in advance if the VMA that this page belonged | |
710 | * to still exists. If the VMA and others sharing the | |
711 | * data have been freed, then the anon_vma could | |
712 | * already be invalid. | |
713 | * | |
714 | * To avoid this possibility, swapcache pages get | |
715 | * migrated but are not remapped when migration | |
716 | * completes | |
717 | */ | |
718 | remap_swapcache = 0; | |
719 | } else { | |
720 | /* | |
721 | * Take a reference count on the anon_vma if the | |
722 | * page is mapped so that it is guaranteed to | |
723 | * exist when the page is remapped later | |
724 | */ | |
725 | anon_vma = page_anon_vma(page); | |
76545066 | 726 | get_anon_vma(anon_vma); |
3fe2011f | 727 | } |
989f89c5 | 728 | } |
62e1c553 | 729 | |
dc386d4d | 730 | /* |
62e1c553 SL |
731 | * Corner case handling: |
732 | * 1. When a new swap-cache page is read into, it is added to the LRU | |
733 | * and treated as swapcache but it has no rmap yet. | |
734 | * Calling try_to_unmap() against a page->mapping==NULL page will | |
735 | * trigger a BUG. So handle it here. | |
736 | * 2. An orphaned page (see truncate_complete_page) might have | |
737 | * fs-private metadata. The page can be picked up due to memory | |
738 | * offlining. Everywhere else except page reclaim, the page is | |
739 | * invisible to the vm, so the page can not be migrated. So try to | |
740 | * free the metadata, so the page can be freed. | |
e24f0b8f | 741 | */ |
62e1c553 | 742 | if (!page->mapping) { |
266cf658 | 743 | if (!PageAnon(page) && page_has_private(page)) { |
62e1c553 SL |
744 | /* |
745 | * Go direct to try_to_free_buffers() here because | |
746 | * a) that's what try_to_release_page() would do anyway | |
747 | * b) we may be under rcu_read_lock() here, so we can't | |
748 | * use GFP_KERNEL which is what try_to_release_page() | |
749 | * needs to be effective. | |
750 | */ | |
751 | try_to_free_buffers(page); | |
abfc3488 | 752 | goto rcu_unlock; |
62e1c553 | 753 | } |
abfc3488 | 754 | goto skip_unmap; |
62e1c553 SL |
755 | } |
756 | ||
dc386d4d | 757 | /* Establish migration ptes or remove ptes */ |
14fa31b8 | 758 | try_to_unmap(page, TTU_MIGRATION|TTU_IGNORE_MLOCK|TTU_IGNORE_ACCESS); |
dc386d4d | 759 | |
abfc3488 | 760 | skip_unmap: |
e6a1530d | 761 | if (!page_mapped(page)) |
3fe2011f | 762 | rc = move_to_new_page(newpage, page, remap_swapcache); |
e24f0b8f | 763 | |
3fe2011f | 764 | if (rc && remap_swapcache) |
e24f0b8f | 765 | remove_migration_ptes(page, page); |
dc386d4d | 766 | rcu_unlock: |
3f6c8272 MG |
767 | |
768 | /* Drop an anon_vma reference if we took one */ | |
76545066 RR |
769 | if (anon_vma) |
770 | drop_anon_vma(anon_vma); | |
3f6c8272 | 771 | |
989f89c5 KH |
772 | if (rcu_locked) |
773 | rcu_read_unlock(); | |
01b1ae63 KH |
774 | uncharge: |
775 | if (!charge) | |
776 | mem_cgroup_end_migration(mem, page, newpage); | |
e24f0b8f CL |
777 | unlock: |
778 | unlock_page(page); | |
95a402c3 | 779 | |
e24f0b8f | 780 | if (rc != -EAGAIN) { |
aaa994b3 CL |
781 | /* |
782 | * A page that has been migrated has all references | |
783 | * removed and will be freed. A page that has not been | |
784 | * migrated will have kepts its references and be | |
785 | * restored. | |
786 | */ | |
787 | list_del(&page->lru); | |
a731286d | 788 | dec_zone_page_state(page, NR_ISOLATED_ANON + |
6c0b1351 | 789 | page_is_file_cache(page)); |
894bc310 | 790 | putback_lru_page(page); |
e24f0b8f | 791 | } |
95a402c3 CL |
792 | |
793 | move_newpage: | |
894bc310 | 794 | |
95a402c3 CL |
795 | /* |
796 | * Move the new page to the LRU. If migration was not successful | |
797 | * then this will free the page. | |
798 | */ | |
894bc310 LS |
799 | putback_lru_page(newpage); |
800 | ||
742755a1 CL |
801 | if (result) { |
802 | if (rc) | |
803 | *result = rc; | |
804 | else | |
805 | *result = page_to_nid(newpage); | |
806 | } | |
e24f0b8f CL |
807 | return rc; |
808 | } | |
809 | ||
290408d4 NH |
810 | /* |
811 | * Counterpart of unmap_and_move_page() for hugepage migration. | |
812 | * | |
813 | * This function doesn't wait the completion of hugepage I/O | |
814 | * because there is no race between I/O and migration for hugepage. | |
815 | * Note that currently hugepage I/O occurs only in direct I/O | |
816 | * where no lock is held and PG_writeback is irrelevant, | |
817 | * and writeback status of all subpages are counted in the reference | |
818 | * count of the head page (i.e. if all subpages of a 2MB hugepage are | |
819 | * under direct I/O, the reference of the head page is 512 and a bit more.) | |
820 | * This means that when we try to migrate hugepage whose subpages are | |
821 | * doing direct I/O, some references remain after try_to_unmap() and | |
822 | * hugepage migration fails without data corruption. | |
823 | * | |
824 | * There is also no race when direct I/O is issued on the page under migration, | |
825 | * because then pte is replaced with migration swap entry and direct I/O code | |
826 | * will wait in the page fault for migration to complete. | |
827 | */ | |
828 | static int unmap_and_move_huge_page(new_page_t get_new_page, | |
829 | unsigned long private, struct page *hpage, | |
77f1fe6b | 830 | int force, int offlining, bool sync) |
290408d4 NH |
831 | { |
832 | int rc = 0; | |
833 | int *result = NULL; | |
834 | struct page *new_hpage = get_new_page(hpage, private, &result); | |
835 | int rcu_locked = 0; | |
836 | struct anon_vma *anon_vma = NULL; | |
837 | ||
838 | if (!new_hpage) | |
839 | return -ENOMEM; | |
840 | ||
841 | rc = -EAGAIN; | |
842 | ||
843 | if (!trylock_page(hpage)) { | |
77f1fe6b | 844 | if (!force || !sync) |
290408d4 NH |
845 | goto out; |
846 | lock_page(hpage); | |
847 | } | |
848 | ||
849 | if (PageAnon(hpage)) { | |
850 | rcu_read_lock(); | |
851 | rcu_locked = 1; | |
852 | ||
853 | if (page_mapped(hpage)) { | |
854 | anon_vma = page_anon_vma(hpage); | |
855 | atomic_inc(&anon_vma->external_refcount); | |
856 | } | |
857 | } | |
858 | ||
859 | try_to_unmap(hpage, TTU_MIGRATION|TTU_IGNORE_MLOCK|TTU_IGNORE_ACCESS); | |
860 | ||
861 | if (!page_mapped(hpage)) | |
862 | rc = move_to_new_page(new_hpage, hpage, 1); | |
863 | ||
864 | if (rc) | |
865 | remove_migration_ptes(hpage, hpage); | |
866 | ||
867 | if (anon_vma && atomic_dec_and_lock(&anon_vma->external_refcount, | |
868 | &anon_vma->lock)) { | |
869 | int empty = list_empty(&anon_vma->head); | |
870 | spin_unlock(&anon_vma->lock); | |
871 | if (empty) | |
872 | anon_vma_free(anon_vma); | |
873 | } | |
874 | ||
875 | if (rcu_locked) | |
876 | rcu_read_unlock(); | |
877 | out: | |
878 | unlock_page(hpage); | |
879 | ||
880 | if (rc != -EAGAIN) { | |
881 | list_del(&hpage->lru); | |
882 | put_page(hpage); | |
883 | } | |
884 | ||
885 | put_page(new_hpage); | |
886 | ||
887 | if (result) { | |
888 | if (rc) | |
889 | *result = rc; | |
890 | else | |
891 | *result = page_to_nid(new_hpage); | |
892 | } | |
893 | return rc; | |
894 | } | |
895 | ||
b20a3503 CL |
896 | /* |
897 | * migrate_pages | |
898 | * | |
95a402c3 CL |
899 | * The function takes one list of pages to migrate and a function |
900 | * that determines from the page to be migrated and the private data | |
901 | * the target of the move and allocates the page. | |
b20a3503 CL |
902 | * |
903 | * The function returns after 10 attempts or if no pages | |
904 | * are movable anymore because to has become empty | |
cf608ac1 MK |
905 | * or no retryable pages exist anymore. |
906 | * Caller should call putback_lru_pages to return pages to the LRU | |
907 | * or free list. | |
b20a3503 | 908 | * |
95a402c3 | 909 | * Return: Number of pages not migrated or error code. |
b20a3503 | 910 | */ |
95a402c3 | 911 | int migrate_pages(struct list_head *from, |
77f1fe6b MG |
912 | new_page_t get_new_page, unsigned long private, int offlining, |
913 | bool sync) | |
b20a3503 | 914 | { |
e24f0b8f | 915 | int retry = 1; |
b20a3503 CL |
916 | int nr_failed = 0; |
917 | int pass = 0; | |
918 | struct page *page; | |
919 | struct page *page2; | |
920 | int swapwrite = current->flags & PF_SWAPWRITE; | |
921 | int rc; | |
922 | ||
923 | if (!swapwrite) | |
924 | current->flags |= PF_SWAPWRITE; | |
925 | ||
e24f0b8f CL |
926 | for(pass = 0; pass < 10 && retry; pass++) { |
927 | retry = 0; | |
b20a3503 | 928 | |
e24f0b8f | 929 | list_for_each_entry_safe(page, page2, from, lru) { |
e24f0b8f | 930 | cond_resched(); |
2d1db3b1 | 931 | |
95a402c3 | 932 | rc = unmap_and_move(get_new_page, private, |
77f1fe6b MG |
933 | page, pass > 2, offlining, |
934 | sync); | |
2d1db3b1 | 935 | |
e24f0b8f | 936 | switch(rc) { |
95a402c3 CL |
937 | case -ENOMEM: |
938 | goto out; | |
e24f0b8f | 939 | case -EAGAIN: |
2d1db3b1 | 940 | retry++; |
e24f0b8f CL |
941 | break; |
942 | case 0: | |
e24f0b8f CL |
943 | break; |
944 | default: | |
2d1db3b1 | 945 | /* Permanent failure */ |
2d1db3b1 | 946 | nr_failed++; |
e24f0b8f | 947 | break; |
2d1db3b1 | 948 | } |
b20a3503 CL |
949 | } |
950 | } | |
95a402c3 CL |
951 | rc = 0; |
952 | out: | |
b20a3503 CL |
953 | if (!swapwrite) |
954 | current->flags &= ~PF_SWAPWRITE; | |
955 | ||
95a402c3 CL |
956 | if (rc) |
957 | return rc; | |
b20a3503 | 958 | |
95a402c3 | 959 | return nr_failed + retry; |
b20a3503 | 960 | } |
95a402c3 | 961 | |
290408d4 | 962 | int migrate_huge_pages(struct list_head *from, |
77f1fe6b MG |
963 | new_page_t get_new_page, unsigned long private, int offlining, |
964 | bool sync) | |
290408d4 NH |
965 | { |
966 | int retry = 1; | |
967 | int nr_failed = 0; | |
968 | int pass = 0; | |
969 | struct page *page; | |
970 | struct page *page2; | |
971 | int rc; | |
972 | ||
973 | for (pass = 0; pass < 10 && retry; pass++) { | |
974 | retry = 0; | |
975 | ||
976 | list_for_each_entry_safe(page, page2, from, lru) { | |
977 | cond_resched(); | |
978 | ||
979 | rc = unmap_and_move_huge_page(get_new_page, | |
77f1fe6b MG |
980 | private, page, pass > 2, offlining, |
981 | sync); | |
290408d4 NH |
982 | |
983 | switch(rc) { | |
984 | case -ENOMEM: | |
985 | goto out; | |
986 | case -EAGAIN: | |
987 | retry++; | |
988 | break; | |
989 | case 0: | |
990 | break; | |
991 | default: | |
992 | /* Permanent failure */ | |
993 | nr_failed++; | |
994 | break; | |
995 | } | |
996 | } | |
997 | } | |
998 | rc = 0; | |
999 | out: | |
1000 | ||
1001 | list_for_each_entry_safe(page, page2, from, lru) | |
1002 | put_page(page); | |
1003 | ||
1004 | if (rc) | |
1005 | return rc; | |
1006 | ||
1007 | return nr_failed + retry; | |
1008 | } | |
1009 | ||
742755a1 CL |
1010 | #ifdef CONFIG_NUMA |
1011 | /* | |
1012 | * Move a list of individual pages | |
1013 | */ | |
1014 | struct page_to_node { | |
1015 | unsigned long addr; | |
1016 | struct page *page; | |
1017 | int node; | |
1018 | int status; | |
1019 | }; | |
1020 | ||
1021 | static struct page *new_page_node(struct page *p, unsigned long private, | |
1022 | int **result) | |
1023 | { | |
1024 | struct page_to_node *pm = (struct page_to_node *)private; | |
1025 | ||
1026 | while (pm->node != MAX_NUMNODES && pm->page != p) | |
1027 | pm++; | |
1028 | ||
1029 | if (pm->node == MAX_NUMNODES) | |
1030 | return NULL; | |
1031 | ||
1032 | *result = &pm->status; | |
1033 | ||
6484eb3e | 1034 | return alloc_pages_exact_node(pm->node, |
769848c0 | 1035 | GFP_HIGHUSER_MOVABLE | GFP_THISNODE, 0); |
742755a1 CL |
1036 | } |
1037 | ||
1038 | /* | |
1039 | * Move a set of pages as indicated in the pm array. The addr | |
1040 | * field must be set to the virtual address of the page to be moved | |
1041 | * and the node number must contain a valid target node. | |
5e9a0f02 | 1042 | * The pm array ends with node = MAX_NUMNODES. |
742755a1 | 1043 | */ |
5e9a0f02 BG |
1044 | static int do_move_page_to_node_array(struct mm_struct *mm, |
1045 | struct page_to_node *pm, | |
1046 | int migrate_all) | |
742755a1 CL |
1047 | { |
1048 | int err; | |
1049 | struct page_to_node *pp; | |
1050 | LIST_HEAD(pagelist); | |
1051 | ||
1052 | down_read(&mm->mmap_sem); | |
1053 | ||
1054 | /* | |
1055 | * Build a list of pages to migrate | |
1056 | */ | |
742755a1 CL |
1057 | for (pp = pm; pp->node != MAX_NUMNODES; pp++) { |
1058 | struct vm_area_struct *vma; | |
1059 | struct page *page; | |
1060 | ||
742755a1 CL |
1061 | err = -EFAULT; |
1062 | vma = find_vma(mm, pp->addr); | |
70384dc6 | 1063 | if (!vma || pp->addr < vma->vm_start || !vma_migratable(vma)) |
742755a1 CL |
1064 | goto set_status; |
1065 | ||
1066 | page = follow_page(vma, pp->addr, FOLL_GET); | |
89f5b7da LT |
1067 | |
1068 | err = PTR_ERR(page); | |
1069 | if (IS_ERR(page)) | |
1070 | goto set_status; | |
1071 | ||
742755a1 CL |
1072 | err = -ENOENT; |
1073 | if (!page) | |
1074 | goto set_status; | |
1075 | ||
62b61f61 HD |
1076 | /* Use PageReserved to check for zero page */ |
1077 | if (PageReserved(page) || PageKsm(page)) | |
742755a1 CL |
1078 | goto put_and_set; |
1079 | ||
1080 | pp->page = page; | |
1081 | err = page_to_nid(page); | |
1082 | ||
1083 | if (err == pp->node) | |
1084 | /* | |
1085 | * Node already in the right place | |
1086 | */ | |
1087 | goto put_and_set; | |
1088 | ||
1089 | err = -EACCES; | |
1090 | if (page_mapcount(page) > 1 && | |
1091 | !migrate_all) | |
1092 | goto put_and_set; | |
1093 | ||
62695a84 | 1094 | err = isolate_lru_page(page); |
6d9c285a | 1095 | if (!err) { |
62695a84 | 1096 | list_add_tail(&page->lru, &pagelist); |
6d9c285a KM |
1097 | inc_zone_page_state(page, NR_ISOLATED_ANON + |
1098 | page_is_file_cache(page)); | |
1099 | } | |
742755a1 CL |
1100 | put_and_set: |
1101 | /* | |
1102 | * Either remove the duplicate refcount from | |
1103 | * isolate_lru_page() or drop the page ref if it was | |
1104 | * not isolated. | |
1105 | */ | |
1106 | put_page(page); | |
1107 | set_status: | |
1108 | pp->status = err; | |
1109 | } | |
1110 | ||
e78bbfa8 | 1111 | err = 0; |
cf608ac1 | 1112 | if (!list_empty(&pagelist)) { |
742755a1 | 1113 | err = migrate_pages(&pagelist, new_page_node, |
77f1fe6b | 1114 | (unsigned long)pm, 0, true); |
cf608ac1 MK |
1115 | if (err) |
1116 | putback_lru_pages(&pagelist); | |
1117 | } | |
742755a1 CL |
1118 | |
1119 | up_read(&mm->mmap_sem); | |
1120 | return err; | |
1121 | } | |
1122 | ||
5e9a0f02 BG |
1123 | /* |
1124 | * Migrate an array of page address onto an array of nodes and fill | |
1125 | * the corresponding array of status. | |
1126 | */ | |
1127 | static int do_pages_move(struct mm_struct *mm, struct task_struct *task, | |
1128 | unsigned long nr_pages, | |
1129 | const void __user * __user *pages, | |
1130 | const int __user *nodes, | |
1131 | int __user *status, int flags) | |
1132 | { | |
3140a227 | 1133 | struct page_to_node *pm; |
5e9a0f02 | 1134 | nodemask_t task_nodes; |
3140a227 BG |
1135 | unsigned long chunk_nr_pages; |
1136 | unsigned long chunk_start; | |
1137 | int err; | |
5e9a0f02 BG |
1138 | |
1139 | task_nodes = cpuset_mems_allowed(task); | |
1140 | ||
3140a227 BG |
1141 | err = -ENOMEM; |
1142 | pm = (struct page_to_node *)__get_free_page(GFP_KERNEL); | |
1143 | if (!pm) | |
5e9a0f02 | 1144 | goto out; |
35282a2d BG |
1145 | |
1146 | migrate_prep(); | |
1147 | ||
5e9a0f02 | 1148 | /* |
3140a227 BG |
1149 | * Store a chunk of page_to_node array in a page, |
1150 | * but keep the last one as a marker | |
5e9a0f02 | 1151 | */ |
3140a227 | 1152 | chunk_nr_pages = (PAGE_SIZE / sizeof(struct page_to_node)) - 1; |
5e9a0f02 | 1153 | |
3140a227 BG |
1154 | for (chunk_start = 0; |
1155 | chunk_start < nr_pages; | |
1156 | chunk_start += chunk_nr_pages) { | |
1157 | int j; | |
5e9a0f02 | 1158 | |
3140a227 BG |
1159 | if (chunk_start + chunk_nr_pages > nr_pages) |
1160 | chunk_nr_pages = nr_pages - chunk_start; | |
1161 | ||
1162 | /* fill the chunk pm with addrs and nodes from user-space */ | |
1163 | for (j = 0; j < chunk_nr_pages; j++) { | |
1164 | const void __user *p; | |
5e9a0f02 BG |
1165 | int node; |
1166 | ||
3140a227 BG |
1167 | err = -EFAULT; |
1168 | if (get_user(p, pages + j + chunk_start)) | |
1169 | goto out_pm; | |
1170 | pm[j].addr = (unsigned long) p; | |
1171 | ||
1172 | if (get_user(node, nodes + j + chunk_start)) | |
5e9a0f02 BG |
1173 | goto out_pm; |
1174 | ||
1175 | err = -ENODEV; | |
6f5a55f1 LT |
1176 | if (node < 0 || node >= MAX_NUMNODES) |
1177 | goto out_pm; | |
1178 | ||
5e9a0f02 BG |
1179 | if (!node_state(node, N_HIGH_MEMORY)) |
1180 | goto out_pm; | |
1181 | ||
1182 | err = -EACCES; | |
1183 | if (!node_isset(node, task_nodes)) | |
1184 | goto out_pm; | |
1185 | ||
3140a227 BG |
1186 | pm[j].node = node; |
1187 | } | |
1188 | ||
1189 | /* End marker for this chunk */ | |
1190 | pm[chunk_nr_pages].node = MAX_NUMNODES; | |
1191 | ||
1192 | /* Migrate this chunk */ | |
1193 | err = do_move_page_to_node_array(mm, pm, | |
1194 | flags & MPOL_MF_MOVE_ALL); | |
1195 | if (err < 0) | |
1196 | goto out_pm; | |
5e9a0f02 | 1197 | |
5e9a0f02 | 1198 | /* Return status information */ |
3140a227 BG |
1199 | for (j = 0; j < chunk_nr_pages; j++) |
1200 | if (put_user(pm[j].status, status + j + chunk_start)) { | |
5e9a0f02 | 1201 | err = -EFAULT; |
3140a227 BG |
1202 | goto out_pm; |
1203 | } | |
1204 | } | |
1205 | err = 0; | |
5e9a0f02 BG |
1206 | |
1207 | out_pm: | |
3140a227 | 1208 | free_page((unsigned long)pm); |
5e9a0f02 BG |
1209 | out: |
1210 | return err; | |
1211 | } | |
1212 | ||
742755a1 | 1213 | /* |
2f007e74 | 1214 | * Determine the nodes of an array of pages and store it in an array of status. |
742755a1 | 1215 | */ |
80bba129 BG |
1216 | static void do_pages_stat_array(struct mm_struct *mm, unsigned long nr_pages, |
1217 | const void __user **pages, int *status) | |
742755a1 | 1218 | { |
2f007e74 | 1219 | unsigned long i; |
2f007e74 | 1220 | |
742755a1 CL |
1221 | down_read(&mm->mmap_sem); |
1222 | ||
2f007e74 | 1223 | for (i = 0; i < nr_pages; i++) { |
80bba129 | 1224 | unsigned long addr = (unsigned long)(*pages); |
742755a1 CL |
1225 | struct vm_area_struct *vma; |
1226 | struct page *page; | |
c095adbc | 1227 | int err = -EFAULT; |
2f007e74 BG |
1228 | |
1229 | vma = find_vma(mm, addr); | |
70384dc6 | 1230 | if (!vma || addr < vma->vm_start) |
742755a1 CL |
1231 | goto set_status; |
1232 | ||
2f007e74 | 1233 | page = follow_page(vma, addr, 0); |
89f5b7da LT |
1234 | |
1235 | err = PTR_ERR(page); | |
1236 | if (IS_ERR(page)) | |
1237 | goto set_status; | |
1238 | ||
742755a1 CL |
1239 | err = -ENOENT; |
1240 | /* Use PageReserved to check for zero page */ | |
62b61f61 | 1241 | if (!page || PageReserved(page) || PageKsm(page)) |
742755a1 CL |
1242 | goto set_status; |
1243 | ||
1244 | err = page_to_nid(page); | |
1245 | set_status: | |
80bba129 BG |
1246 | *status = err; |
1247 | ||
1248 | pages++; | |
1249 | status++; | |
1250 | } | |
1251 | ||
1252 | up_read(&mm->mmap_sem); | |
1253 | } | |
1254 | ||
1255 | /* | |
1256 | * Determine the nodes of a user array of pages and store it in | |
1257 | * a user array of status. | |
1258 | */ | |
1259 | static int do_pages_stat(struct mm_struct *mm, unsigned long nr_pages, | |
1260 | const void __user * __user *pages, | |
1261 | int __user *status) | |
1262 | { | |
1263 | #define DO_PAGES_STAT_CHUNK_NR 16 | |
1264 | const void __user *chunk_pages[DO_PAGES_STAT_CHUNK_NR]; | |
1265 | int chunk_status[DO_PAGES_STAT_CHUNK_NR]; | |
80bba129 | 1266 | |
87b8d1ad PA |
1267 | while (nr_pages) { |
1268 | unsigned long chunk_nr; | |
80bba129 | 1269 | |
87b8d1ad PA |
1270 | chunk_nr = nr_pages; |
1271 | if (chunk_nr > DO_PAGES_STAT_CHUNK_NR) | |
1272 | chunk_nr = DO_PAGES_STAT_CHUNK_NR; | |
1273 | ||
1274 | if (copy_from_user(chunk_pages, pages, chunk_nr * sizeof(*chunk_pages))) | |
1275 | break; | |
80bba129 BG |
1276 | |
1277 | do_pages_stat_array(mm, chunk_nr, chunk_pages, chunk_status); | |
1278 | ||
87b8d1ad PA |
1279 | if (copy_to_user(status, chunk_status, chunk_nr * sizeof(*status))) |
1280 | break; | |
742755a1 | 1281 | |
87b8d1ad PA |
1282 | pages += chunk_nr; |
1283 | status += chunk_nr; | |
1284 | nr_pages -= chunk_nr; | |
1285 | } | |
1286 | return nr_pages ? -EFAULT : 0; | |
742755a1 CL |
1287 | } |
1288 | ||
1289 | /* | |
1290 | * Move a list of pages in the address space of the currently executing | |
1291 | * process. | |
1292 | */ | |
938bb9f5 HC |
1293 | SYSCALL_DEFINE6(move_pages, pid_t, pid, unsigned long, nr_pages, |
1294 | const void __user * __user *, pages, | |
1295 | const int __user *, nodes, | |
1296 | int __user *, status, int, flags) | |
742755a1 | 1297 | { |
c69e8d9c | 1298 | const struct cred *cred = current_cred(), *tcred; |
742755a1 | 1299 | struct task_struct *task; |
742755a1 | 1300 | struct mm_struct *mm; |
5e9a0f02 | 1301 | int err; |
742755a1 CL |
1302 | |
1303 | /* Check flags */ | |
1304 | if (flags & ~(MPOL_MF_MOVE|MPOL_MF_MOVE_ALL)) | |
1305 | return -EINVAL; | |
1306 | ||
1307 | if ((flags & MPOL_MF_MOVE_ALL) && !capable(CAP_SYS_NICE)) | |
1308 | return -EPERM; | |
1309 | ||
1310 | /* Find the mm_struct */ | |
1311 | read_lock(&tasklist_lock); | |
228ebcbe | 1312 | task = pid ? find_task_by_vpid(pid) : current; |
742755a1 CL |
1313 | if (!task) { |
1314 | read_unlock(&tasklist_lock); | |
1315 | return -ESRCH; | |
1316 | } | |
1317 | mm = get_task_mm(task); | |
1318 | read_unlock(&tasklist_lock); | |
1319 | ||
1320 | if (!mm) | |
1321 | return -EINVAL; | |
1322 | ||
1323 | /* | |
1324 | * Check if this process has the right to modify the specified | |
1325 | * process. The right exists if the process has administrative | |
1326 | * capabilities, superuser privileges or the same | |
1327 | * userid as the target process. | |
1328 | */ | |
c69e8d9c DH |
1329 | rcu_read_lock(); |
1330 | tcred = __task_cred(task); | |
b6dff3ec DH |
1331 | if (cred->euid != tcred->suid && cred->euid != tcred->uid && |
1332 | cred->uid != tcred->suid && cred->uid != tcred->uid && | |
742755a1 | 1333 | !capable(CAP_SYS_NICE)) { |
c69e8d9c | 1334 | rcu_read_unlock(); |
742755a1 | 1335 | err = -EPERM; |
5e9a0f02 | 1336 | goto out; |
742755a1 | 1337 | } |
c69e8d9c | 1338 | rcu_read_unlock(); |
742755a1 | 1339 | |
86c3a764 DQ |
1340 | err = security_task_movememory(task); |
1341 | if (err) | |
5e9a0f02 | 1342 | goto out; |
86c3a764 | 1343 | |
5e9a0f02 BG |
1344 | if (nodes) { |
1345 | err = do_pages_move(mm, task, nr_pages, pages, nodes, status, | |
1346 | flags); | |
1347 | } else { | |
2f007e74 | 1348 | err = do_pages_stat(mm, nr_pages, pages, status); |
742755a1 CL |
1349 | } |
1350 | ||
742755a1 | 1351 | out: |
742755a1 CL |
1352 | mmput(mm); |
1353 | return err; | |
1354 | } | |
742755a1 | 1355 | |
7b2259b3 CL |
1356 | /* |
1357 | * Call migration functions in the vma_ops that may prepare | |
1358 | * memory in a vm for migration. migration functions may perform | |
1359 | * the migration for vmas that do not have an underlying page struct. | |
1360 | */ | |
1361 | int migrate_vmas(struct mm_struct *mm, const nodemask_t *to, | |
1362 | const nodemask_t *from, unsigned long flags) | |
1363 | { | |
1364 | struct vm_area_struct *vma; | |
1365 | int err = 0; | |
1366 | ||
1001c9fb | 1367 | for (vma = mm->mmap; vma && !err; vma = vma->vm_next) { |
7b2259b3 CL |
1368 | if (vma->vm_ops && vma->vm_ops->migrate) { |
1369 | err = vma->vm_ops->migrate(vma, to, from, flags); | |
1370 | if (err) | |
1371 | break; | |
1372 | } | |
1373 | } | |
1374 | return err; | |
1375 | } | |
83d1674a | 1376 | #endif |