]>
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> | |
b95f1b31 | 16 | #include <linux/export.h> |
b20a3503 | 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 | 42 | /* |
742755a1 | 43 | * migrate_prep() needs to be called before we start compiling a list of pages |
748446bb MG |
44 | * to be migrated using isolate_lru_page(). If scheduling work on other CPUs is |
45 | * undesirable, use migrate_prep_local() | |
b20a3503 CL |
46 | */ |
47 | int migrate_prep(void) | |
48 | { | |
b20a3503 CL |
49 | /* |
50 | * Clear the LRU lists so pages can be isolated. | |
51 | * Note that pages may be moved off the LRU after we have | |
52 | * drained them. Those pages will fail to migrate like other | |
53 | * pages that may be busy. | |
54 | */ | |
55 | lru_add_drain_all(); | |
56 | ||
57 | return 0; | |
58 | } | |
59 | ||
748446bb MG |
60 | /* Do the necessary work of migrate_prep but not if it involves other CPUs */ |
61 | int migrate_prep_local(void) | |
62 | { | |
63 | lru_add_drain(); | |
64 | ||
65 | return 0; | |
66 | } | |
67 | ||
b20a3503 | 68 | /* |
894bc310 LS |
69 | * Add isolated pages on the list back to the LRU under page lock |
70 | * to avoid leaking evictable pages back onto unevictable list. | |
b20a3503 | 71 | */ |
e13861d8 | 72 | void putback_lru_pages(struct list_head *l) |
b20a3503 CL |
73 | { |
74 | struct page *page; | |
75 | struct page *page2; | |
b20a3503 CL |
76 | |
77 | list_for_each_entry_safe(page, page2, l, lru) { | |
e24f0b8f | 78 | list_del(&page->lru); |
a731286d | 79 | dec_zone_page_state(page, NR_ISOLATED_ANON + |
6c0b1351 | 80 | page_is_file_cache(page)); |
894bc310 | 81 | putback_lru_page(page); |
b20a3503 | 82 | } |
b20a3503 CL |
83 | } |
84 | ||
0697212a CL |
85 | /* |
86 | * Restore a potential migration pte to a working pte entry | |
87 | */ | |
e9995ef9 HD |
88 | static int remove_migration_pte(struct page *new, struct vm_area_struct *vma, |
89 | unsigned long addr, void *old) | |
0697212a CL |
90 | { |
91 | struct mm_struct *mm = vma->vm_mm; | |
92 | swp_entry_t entry; | |
93 | pgd_t *pgd; | |
94 | pud_t *pud; | |
95 | pmd_t *pmd; | |
96 | pte_t *ptep, pte; | |
97 | spinlock_t *ptl; | |
98 | ||
290408d4 NH |
99 | if (unlikely(PageHuge(new))) { |
100 | ptep = huge_pte_offset(mm, addr); | |
101 | if (!ptep) | |
102 | goto out; | |
103 | ptl = &mm->page_table_lock; | |
104 | } else { | |
105 | pgd = pgd_offset(mm, addr); | |
106 | if (!pgd_present(*pgd)) | |
107 | goto out; | |
0697212a | 108 | |
290408d4 NH |
109 | pud = pud_offset(pgd, addr); |
110 | if (!pud_present(*pud)) | |
111 | goto out; | |
0697212a | 112 | |
290408d4 | 113 | pmd = pmd_offset(pud, addr); |
500d65d4 AA |
114 | if (pmd_trans_huge(*pmd)) |
115 | goto out; | |
290408d4 NH |
116 | if (!pmd_present(*pmd)) |
117 | goto out; | |
0697212a | 118 | |
290408d4 | 119 | ptep = pte_offset_map(pmd, addr); |
0697212a | 120 | |
486cf46f HD |
121 | /* |
122 | * Peek to check is_swap_pte() before taking ptlock? No, we | |
123 | * can race mremap's move_ptes(), which skips anon_vma lock. | |
124 | */ | |
290408d4 NH |
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. | |
0697212a CL |
182 | */ |
183 | void migration_entry_wait(struct mm_struct *mm, pmd_t *pmd, | |
184 | unsigned long address) | |
185 | { | |
186 | pte_t *ptep, pte; | |
187 | spinlock_t *ptl; | |
188 | swp_entry_t entry; | |
189 | struct page *page; | |
190 | ||
191 | ptep = pte_offset_map_lock(mm, pmd, address, &ptl); | |
192 | pte = *ptep; | |
193 | if (!is_swap_pte(pte)) | |
194 | goto out; | |
195 | ||
196 | entry = pte_to_swp_entry(pte); | |
197 | if (!is_migration_entry(entry)) | |
198 | goto out; | |
199 | ||
200 | page = migration_entry_to_page(entry); | |
201 | ||
e286781d NP |
202 | /* |
203 | * Once radix-tree replacement of page migration started, page_count | |
204 | * *must* be zero. And, we don't want to call wait_on_page_locked() | |
205 | * against a page without get_page(). | |
206 | * So, we use get_page_unless_zero(), here. Even failed, page fault | |
207 | * will occur again. | |
208 | */ | |
209 | if (!get_page_unless_zero(page)) | |
210 | goto out; | |
0697212a CL |
211 | pte_unmap_unlock(ptep, ptl); |
212 | wait_on_page_locked(page); | |
213 | put_page(page); | |
214 | return; | |
215 | out: | |
216 | pte_unmap_unlock(ptep, ptl); | |
217 | } | |
218 | ||
b969c4ab MG |
219 | #ifdef CONFIG_BLOCK |
220 | /* Returns true if all buffers are successfully locked */ | |
221 | static bool buffer_migrate_lock_buffers(struct buffer_head *head, bool sync) | |
222 | { | |
223 | struct buffer_head *bh = head; | |
224 | ||
225 | /* Simple case, sync compaction */ | |
226 | if (sync) { | |
227 | do { | |
228 | get_bh(bh); | |
229 | lock_buffer(bh); | |
230 | bh = bh->b_this_page; | |
231 | ||
232 | } while (bh != head); | |
233 | ||
234 | return true; | |
235 | } | |
236 | ||
237 | /* async case, we cannot block on lock_buffer so use trylock_buffer */ | |
238 | do { | |
239 | get_bh(bh); | |
240 | if (!trylock_buffer(bh)) { | |
241 | /* | |
242 | * We failed to lock the buffer and cannot stall in | |
243 | * async migration. Release the taken locks | |
244 | */ | |
245 | struct buffer_head *failed_bh = bh; | |
246 | put_bh(failed_bh); | |
247 | bh = head; | |
248 | while (bh != failed_bh) { | |
249 | unlock_buffer(bh); | |
250 | put_bh(bh); | |
251 | bh = bh->b_this_page; | |
252 | } | |
253 | return false; | |
254 | } | |
255 | ||
256 | bh = bh->b_this_page; | |
257 | } while (bh != head); | |
258 | return true; | |
259 | } | |
260 | #else | |
261 | static inline bool buffer_migrate_lock_buffers(struct buffer_head *head, | |
262 | bool sync) | |
263 | { | |
264 | return true; | |
265 | } | |
266 | #endif /* CONFIG_BLOCK */ | |
267 | ||
b20a3503 | 268 | /* |
c3fcf8a5 | 269 | * Replace the page in the mapping. |
5b5c7120 CL |
270 | * |
271 | * The number of remaining references must be: | |
272 | * 1 for anonymous pages without a mapping | |
273 | * 2 for pages with a mapping | |
266cf658 | 274 | * 3 for pages with a mapping and PagePrivate/PagePrivate2 set. |
b20a3503 | 275 | */ |
2d1db3b1 | 276 | static int migrate_page_move_mapping(struct address_space *mapping, |
b969c4ab MG |
277 | struct page *newpage, struct page *page, |
278 | struct buffer_head *head, bool sync) | |
b20a3503 | 279 | { |
e286781d | 280 | int expected_count; |
7cf9c2c7 | 281 | void **pslot; |
b20a3503 | 282 | |
6c5240ae | 283 | if (!mapping) { |
0e8c7d0f | 284 | /* Anonymous page without mapping */ |
6c5240ae CL |
285 | if (page_count(page) != 1) |
286 | return -EAGAIN; | |
287 | return 0; | |
288 | } | |
289 | ||
19fd6231 | 290 | spin_lock_irq(&mapping->tree_lock); |
b20a3503 | 291 | |
7cf9c2c7 NP |
292 | pslot = radix_tree_lookup_slot(&mapping->page_tree, |
293 | page_index(page)); | |
b20a3503 | 294 | |
edcf4748 | 295 | expected_count = 2 + page_has_private(page); |
e286781d | 296 | if (page_count(page) != expected_count || |
29c1f677 | 297 | radix_tree_deref_slot_protected(pslot, &mapping->tree_lock) != page) { |
19fd6231 | 298 | spin_unlock_irq(&mapping->tree_lock); |
e23ca00b | 299 | return -EAGAIN; |
b20a3503 CL |
300 | } |
301 | ||
e286781d | 302 | if (!page_freeze_refs(page, expected_count)) { |
19fd6231 | 303 | spin_unlock_irq(&mapping->tree_lock); |
e286781d NP |
304 | return -EAGAIN; |
305 | } | |
306 | ||
b969c4ab MG |
307 | /* |
308 | * In the async migration case of moving a page with buffers, lock the | |
309 | * buffers using trylock before the mapping is moved. If the mapping | |
310 | * was moved, we later failed to lock the buffers and could not move | |
311 | * the mapping back due to an elevated page count, we would have to | |
312 | * block waiting on other references to be dropped. | |
313 | */ | |
314 | if (!sync && head && !buffer_migrate_lock_buffers(head, sync)) { | |
315 | page_unfreeze_refs(page, expected_count); | |
316 | spin_unlock_irq(&mapping->tree_lock); | |
317 | return -EAGAIN; | |
318 | } | |
319 | ||
b20a3503 CL |
320 | /* |
321 | * Now we know that no one else is looking at the page. | |
b20a3503 | 322 | */ |
7cf9c2c7 | 323 | get_page(newpage); /* add cache reference */ |
b20a3503 CL |
324 | if (PageSwapCache(page)) { |
325 | SetPageSwapCache(newpage); | |
326 | set_page_private(newpage, page_private(page)); | |
327 | } | |
328 | ||
7cf9c2c7 NP |
329 | radix_tree_replace_slot(pslot, newpage); |
330 | ||
331 | /* | |
937a94c9 JG |
332 | * Drop cache reference from old page by unfreezing |
333 | * to one less reference. | |
7cf9c2c7 NP |
334 | * We know this isn't the last reference. |
335 | */ | |
937a94c9 | 336 | page_unfreeze_refs(page, expected_count - 1); |
7cf9c2c7 | 337 | |
0e8c7d0f CL |
338 | /* |
339 | * If moved to a different zone then also account | |
340 | * the page for that zone. Other VM counters will be | |
341 | * taken care of when we establish references to the | |
342 | * new page and drop references to the old page. | |
343 | * | |
344 | * Note that anonymous pages are accounted for | |
345 | * via NR_FILE_PAGES and NR_ANON_PAGES if they | |
346 | * are mapped to swap space. | |
347 | */ | |
348 | __dec_zone_page_state(page, NR_FILE_PAGES); | |
349 | __inc_zone_page_state(newpage, NR_FILE_PAGES); | |
99a15e21 | 350 | if (!PageSwapCache(page) && PageSwapBacked(page)) { |
4b02108a KM |
351 | __dec_zone_page_state(page, NR_SHMEM); |
352 | __inc_zone_page_state(newpage, NR_SHMEM); | |
353 | } | |
19fd6231 | 354 | spin_unlock_irq(&mapping->tree_lock); |
b20a3503 CL |
355 | |
356 | return 0; | |
357 | } | |
b20a3503 | 358 | |
290408d4 NH |
359 | /* |
360 | * The expected number of remaining references is the same as that | |
361 | * of migrate_page_move_mapping(). | |
362 | */ | |
363 | int migrate_huge_page_move_mapping(struct address_space *mapping, | |
364 | struct page *newpage, struct page *page) | |
365 | { | |
366 | int expected_count; | |
367 | void **pslot; | |
368 | ||
369 | if (!mapping) { | |
370 | if (page_count(page) != 1) | |
371 | return -EAGAIN; | |
372 | return 0; | |
373 | } | |
374 | ||
375 | spin_lock_irq(&mapping->tree_lock); | |
376 | ||
377 | pslot = radix_tree_lookup_slot(&mapping->page_tree, | |
378 | page_index(page)); | |
379 | ||
380 | expected_count = 2 + page_has_private(page); | |
381 | if (page_count(page) != expected_count || | |
29c1f677 | 382 | radix_tree_deref_slot_protected(pslot, &mapping->tree_lock) != page) { |
290408d4 NH |
383 | spin_unlock_irq(&mapping->tree_lock); |
384 | return -EAGAIN; | |
385 | } | |
386 | ||
387 | if (!page_freeze_refs(page, expected_count)) { | |
388 | spin_unlock_irq(&mapping->tree_lock); | |
389 | return -EAGAIN; | |
390 | } | |
391 | ||
392 | get_page(newpage); | |
393 | ||
394 | radix_tree_replace_slot(pslot, newpage); | |
395 | ||
937a94c9 | 396 | page_unfreeze_refs(page, expected_count - 1); |
290408d4 NH |
397 | |
398 | spin_unlock_irq(&mapping->tree_lock); | |
399 | return 0; | |
400 | } | |
401 | ||
b20a3503 CL |
402 | /* |
403 | * Copy the page to its new location | |
404 | */ | |
290408d4 | 405 | void migrate_page_copy(struct page *newpage, struct page *page) |
b20a3503 | 406 | { |
290408d4 NH |
407 | if (PageHuge(page)) |
408 | copy_huge_page(newpage, page); | |
409 | else | |
410 | copy_highpage(newpage, page); | |
b20a3503 CL |
411 | |
412 | if (PageError(page)) | |
413 | SetPageError(newpage); | |
414 | if (PageReferenced(page)) | |
415 | SetPageReferenced(newpage); | |
416 | if (PageUptodate(page)) | |
417 | SetPageUptodate(newpage); | |
894bc310 LS |
418 | if (TestClearPageActive(page)) { |
419 | VM_BUG_ON(PageUnevictable(page)); | |
b20a3503 | 420 | SetPageActive(newpage); |
418b27ef LS |
421 | } else if (TestClearPageUnevictable(page)) |
422 | SetPageUnevictable(newpage); | |
b20a3503 CL |
423 | if (PageChecked(page)) |
424 | SetPageChecked(newpage); | |
425 | if (PageMappedToDisk(page)) | |
426 | SetPageMappedToDisk(newpage); | |
427 | ||
428 | if (PageDirty(page)) { | |
429 | clear_page_dirty_for_io(page); | |
3a902c5f NP |
430 | /* |
431 | * Want to mark the page and the radix tree as dirty, and | |
432 | * redo the accounting that clear_page_dirty_for_io undid, | |
433 | * but we can't use set_page_dirty because that function | |
434 | * is actually a signal that all of the page has become dirty. | |
25985edc | 435 | * Whereas only part of our page may be dirty. |
3a902c5f NP |
436 | */ |
437 | __set_page_dirty_nobuffers(newpage); | |
b20a3503 CL |
438 | } |
439 | ||
b291f000 | 440 | mlock_migrate_page(newpage, page); |
e9995ef9 | 441 | ksm_migrate_page(newpage, page); |
b291f000 | 442 | |
b20a3503 | 443 | ClearPageSwapCache(page); |
b20a3503 CL |
444 | ClearPagePrivate(page); |
445 | set_page_private(page, 0); | |
446 | page->mapping = NULL; | |
447 | ||
448 | /* | |
449 | * If any waiters have accumulated on the new page then | |
450 | * wake them up. | |
451 | */ | |
452 | if (PageWriteback(newpage)) | |
453 | end_page_writeback(newpage); | |
454 | } | |
b20a3503 | 455 | |
1d8b85cc CL |
456 | /************************************************************ |
457 | * Migration functions | |
458 | ***********************************************************/ | |
459 | ||
460 | /* Always fail migration. Used for mappings that are not movable */ | |
2d1db3b1 CL |
461 | int fail_migrate_page(struct address_space *mapping, |
462 | struct page *newpage, struct page *page) | |
1d8b85cc CL |
463 | { |
464 | return -EIO; | |
465 | } | |
466 | EXPORT_SYMBOL(fail_migrate_page); | |
467 | ||
b20a3503 CL |
468 | /* |
469 | * Common logic to directly migrate a single page suitable for | |
266cf658 | 470 | * pages that do not use PagePrivate/PagePrivate2. |
b20a3503 CL |
471 | * |
472 | * Pages are locked upon entry and exit. | |
473 | */ | |
2d1db3b1 | 474 | int migrate_page(struct address_space *mapping, |
b969c4ab | 475 | struct page *newpage, struct page *page, bool sync) |
b20a3503 CL |
476 | { |
477 | int rc; | |
478 | ||
479 | BUG_ON(PageWriteback(page)); /* Writeback must be complete */ | |
480 | ||
b969c4ab | 481 | rc = migrate_page_move_mapping(mapping, newpage, page, NULL, sync); |
b20a3503 CL |
482 | |
483 | if (rc) | |
484 | return rc; | |
485 | ||
486 | migrate_page_copy(newpage, page); | |
b20a3503 CL |
487 | return 0; |
488 | } | |
489 | EXPORT_SYMBOL(migrate_page); | |
490 | ||
9361401e | 491 | #ifdef CONFIG_BLOCK |
1d8b85cc CL |
492 | /* |
493 | * Migration function for pages with buffers. This function can only be used | |
494 | * if the underlying filesystem guarantees that no other references to "page" | |
495 | * exist. | |
496 | */ | |
2d1db3b1 | 497 | int buffer_migrate_page(struct address_space *mapping, |
b969c4ab | 498 | struct page *newpage, struct page *page, bool sync) |
1d8b85cc | 499 | { |
1d8b85cc CL |
500 | struct buffer_head *bh, *head; |
501 | int rc; | |
502 | ||
1d8b85cc | 503 | if (!page_has_buffers(page)) |
b969c4ab | 504 | return migrate_page(mapping, newpage, page, sync); |
1d8b85cc CL |
505 | |
506 | head = page_buffers(page); | |
507 | ||
b969c4ab | 508 | rc = migrate_page_move_mapping(mapping, newpage, page, head, sync); |
1d8b85cc CL |
509 | |
510 | if (rc) | |
511 | return rc; | |
512 | ||
b969c4ab MG |
513 | /* |
514 | * In the async case, migrate_page_move_mapping locked the buffers | |
515 | * with an IRQ-safe spinlock held. In the sync case, the buffers | |
516 | * need to be locked now | |
517 | */ | |
518 | if (sync) | |
519 | BUG_ON(!buffer_migrate_lock_buffers(head, sync)); | |
1d8b85cc CL |
520 | |
521 | ClearPagePrivate(page); | |
522 | set_page_private(newpage, page_private(page)); | |
523 | set_page_private(page, 0); | |
524 | put_page(page); | |
525 | get_page(newpage); | |
526 | ||
527 | bh = head; | |
528 | do { | |
529 | set_bh_page(bh, newpage, bh_offset(bh)); | |
530 | bh = bh->b_this_page; | |
531 | ||
532 | } while (bh != head); | |
533 | ||
534 | SetPagePrivate(newpage); | |
535 | ||
536 | migrate_page_copy(newpage, page); | |
537 | ||
538 | bh = head; | |
539 | do { | |
540 | unlock_buffer(bh); | |
541 | put_bh(bh); | |
542 | bh = bh->b_this_page; | |
543 | ||
544 | } while (bh != head); | |
545 | ||
546 | return 0; | |
547 | } | |
548 | EXPORT_SYMBOL(buffer_migrate_page); | |
9361401e | 549 | #endif |
1d8b85cc | 550 | |
04e62a29 CL |
551 | /* |
552 | * Writeback a page to clean the dirty state | |
553 | */ | |
554 | static int writeout(struct address_space *mapping, struct page *page) | |
8351a6e4 | 555 | { |
04e62a29 CL |
556 | struct writeback_control wbc = { |
557 | .sync_mode = WB_SYNC_NONE, | |
558 | .nr_to_write = 1, | |
559 | .range_start = 0, | |
560 | .range_end = LLONG_MAX, | |
04e62a29 CL |
561 | .for_reclaim = 1 |
562 | }; | |
563 | int rc; | |
564 | ||
565 | if (!mapping->a_ops->writepage) | |
566 | /* No write method for the address space */ | |
567 | return -EINVAL; | |
568 | ||
569 | if (!clear_page_dirty_for_io(page)) | |
570 | /* Someone else already triggered a write */ | |
571 | return -EAGAIN; | |
572 | ||
8351a6e4 | 573 | /* |
04e62a29 CL |
574 | * A dirty page may imply that the underlying filesystem has |
575 | * the page on some queue. So the page must be clean for | |
576 | * migration. Writeout may mean we loose the lock and the | |
577 | * page state is no longer what we checked for earlier. | |
578 | * At this point we know that the migration attempt cannot | |
579 | * be successful. | |
8351a6e4 | 580 | */ |
04e62a29 | 581 | remove_migration_ptes(page, page); |
8351a6e4 | 582 | |
04e62a29 | 583 | rc = mapping->a_ops->writepage(page, &wbc); |
8351a6e4 | 584 | |
04e62a29 CL |
585 | if (rc != AOP_WRITEPAGE_ACTIVATE) |
586 | /* unlocked. Relock */ | |
587 | lock_page(page); | |
588 | ||
bda8550d | 589 | return (rc < 0) ? -EIO : -EAGAIN; |
04e62a29 CL |
590 | } |
591 | ||
592 | /* | |
593 | * Default handling if a filesystem does not provide a migration function. | |
594 | */ | |
595 | static int fallback_migrate_page(struct address_space *mapping, | |
b969c4ab | 596 | struct page *newpage, struct page *page, bool sync) |
04e62a29 | 597 | { |
b969c4ab MG |
598 | if (PageDirty(page)) { |
599 | if (!sync) | |
600 | return -EBUSY; | |
04e62a29 | 601 | return writeout(mapping, page); |
b969c4ab | 602 | } |
8351a6e4 CL |
603 | |
604 | /* | |
605 | * Buffers may be managed in a filesystem specific way. | |
606 | * We must have no buffers or drop them. | |
607 | */ | |
266cf658 | 608 | if (page_has_private(page) && |
8351a6e4 CL |
609 | !try_to_release_page(page, GFP_KERNEL)) |
610 | return -EAGAIN; | |
611 | ||
b969c4ab | 612 | return migrate_page(mapping, newpage, page, sync); |
8351a6e4 CL |
613 | } |
614 | ||
e24f0b8f CL |
615 | /* |
616 | * Move a page to a newly allocated page | |
617 | * The page is locked and all ptes have been successfully removed. | |
618 | * | |
619 | * The new page will have replaced the old page if this function | |
620 | * is successful. | |
894bc310 LS |
621 | * |
622 | * Return value: | |
623 | * < 0 - error code | |
624 | * == 0 - success | |
e24f0b8f | 625 | */ |
3fe2011f | 626 | static int move_to_new_page(struct page *newpage, struct page *page, |
11bc82d6 | 627 | int remap_swapcache, bool sync) |
e24f0b8f CL |
628 | { |
629 | struct address_space *mapping; | |
630 | int rc; | |
631 | ||
632 | /* | |
633 | * Block others from accessing the page when we get around to | |
634 | * establishing additional references. We are the only one | |
635 | * holding a reference to the new page at this point. | |
636 | */ | |
529ae9aa | 637 | if (!trylock_page(newpage)) |
e24f0b8f CL |
638 | BUG(); |
639 | ||
640 | /* Prepare mapping for the new page.*/ | |
641 | newpage->index = page->index; | |
642 | newpage->mapping = page->mapping; | |
b2e18538 RR |
643 | if (PageSwapBacked(page)) |
644 | SetPageSwapBacked(newpage); | |
e24f0b8f CL |
645 | |
646 | mapping = page_mapping(page); | |
647 | if (!mapping) | |
b969c4ab MG |
648 | rc = migrate_page(mapping, newpage, page, sync); |
649 | else if (mapping->a_ops->migratepage) | |
e24f0b8f | 650 | /* |
b969c4ab MG |
651 | * Most pages have a mapping and most filesystems provide a |
652 | * migratepage callback. Anonymous pages are part of swap | |
653 | * space which also has its own migratepage callback. This | |
654 | * is the most common path for page migration. | |
e24f0b8f | 655 | */ |
b969c4ab MG |
656 | rc = mapping->a_ops->migratepage(mapping, |
657 | newpage, page, sync); | |
658 | else | |
659 | rc = fallback_migrate_page(mapping, newpage, page, sync); | |
e24f0b8f | 660 | |
3fe2011f | 661 | if (rc) { |
e24f0b8f | 662 | newpage->mapping = NULL; |
3fe2011f MG |
663 | } else { |
664 | if (remap_swapcache) | |
665 | remove_migration_ptes(page, newpage); | |
666 | } | |
e24f0b8f CL |
667 | |
668 | unlock_page(newpage); | |
669 | ||
670 | return rc; | |
671 | } | |
672 | ||
0dabec93 MK |
673 | static int __unmap_and_move(struct page *page, struct page *newpage, |
674 | int force, bool offlining, bool sync) | |
e24f0b8f | 675 | { |
0dabec93 | 676 | int rc = -EAGAIN; |
3fe2011f | 677 | int remap_swapcache = 1; |
ae41be37 | 678 | int charge = 0; |
56039efa | 679 | struct mem_cgroup *mem; |
3f6c8272 | 680 | struct anon_vma *anon_vma = NULL; |
95a402c3 | 681 | |
529ae9aa | 682 | if (!trylock_page(page)) { |
11bc82d6 | 683 | if (!force || !sync) |
0dabec93 | 684 | goto out; |
3e7d3449 MG |
685 | |
686 | /* | |
687 | * It's not safe for direct compaction to call lock_page. | |
688 | * For example, during page readahead pages are added locked | |
689 | * to the LRU. Later, when the IO completes the pages are | |
690 | * marked uptodate and unlocked. However, the queueing | |
691 | * could be merging multiple pages for one bio (e.g. | |
692 | * mpage_readpages). If an allocation happens for the | |
693 | * second or third page, the process can end up locking | |
694 | * the same page twice and deadlocking. Rather than | |
695 | * trying to be clever about what pages can be locked, | |
696 | * avoid the use of lock_page for direct compaction | |
697 | * altogether. | |
698 | */ | |
699 | if (current->flags & PF_MEMALLOC) | |
0dabec93 | 700 | goto out; |
3e7d3449 | 701 | |
e24f0b8f CL |
702 | lock_page(page); |
703 | } | |
704 | ||
62b61f61 HD |
705 | /* |
706 | * Only memory hotplug's offline_pages() caller has locked out KSM, | |
707 | * and can safely migrate a KSM page. The other cases have skipped | |
708 | * PageKsm along with PageReserved - but it is only now when we have | |
709 | * the page lock that we can be certain it will not go KSM beneath us | |
710 | * (KSM will not upgrade a page from PageAnon to PageKsm when it sees | |
711 | * its pagecount raised, but only here do we take the page lock which | |
712 | * serializes that). | |
713 | */ | |
714 | if (PageKsm(page) && !offlining) { | |
715 | rc = -EBUSY; | |
716 | goto unlock; | |
717 | } | |
718 | ||
01b1ae63 | 719 | /* charge against new page */ |
ef6a3c63 | 720 | charge = mem_cgroup_prepare_migration(page, newpage, &mem, GFP_KERNEL); |
01b1ae63 KH |
721 | if (charge == -ENOMEM) { |
722 | rc = -ENOMEM; | |
723 | goto unlock; | |
724 | } | |
725 | BUG_ON(charge); | |
726 | ||
e24f0b8f | 727 | if (PageWriteback(page)) { |
11bc82d6 AA |
728 | /* |
729 | * For !sync, there is no point retrying as the retry loop | |
730 | * is expected to be too short for PageWriteback to be cleared | |
731 | */ | |
732 | if (!sync) { | |
733 | rc = -EBUSY; | |
734 | goto uncharge; | |
735 | } | |
736 | if (!force) | |
01b1ae63 | 737 | goto uncharge; |
e24f0b8f CL |
738 | wait_on_page_writeback(page); |
739 | } | |
e24f0b8f | 740 | /* |
dc386d4d KH |
741 | * By try_to_unmap(), page->mapcount goes down to 0 here. In this case, |
742 | * we cannot notice that anon_vma is freed while we migrates a page. | |
1ce82b69 | 743 | * This get_anon_vma() delays freeing anon_vma pointer until the end |
dc386d4d | 744 | * of migration. File cache pages are no problem because of page_lock() |
989f89c5 KH |
745 | * File Caches may use write_page() or lock_page() in migration, then, |
746 | * just care Anon page here. | |
dc386d4d | 747 | */ |
989f89c5 | 748 | if (PageAnon(page)) { |
1ce82b69 HD |
749 | /* |
750 | * Only page_lock_anon_vma() understands the subtleties of | |
751 | * getting a hold on an anon_vma from outside one of its mms. | |
752 | */ | |
746b18d4 | 753 | anon_vma = page_get_anon_vma(page); |
1ce82b69 HD |
754 | if (anon_vma) { |
755 | /* | |
746b18d4 | 756 | * Anon page |
1ce82b69 | 757 | */ |
1ce82b69 | 758 | } else if (PageSwapCache(page)) { |
3fe2011f MG |
759 | /* |
760 | * We cannot be sure that the anon_vma of an unmapped | |
761 | * swapcache page is safe to use because we don't | |
762 | * know in advance if the VMA that this page belonged | |
763 | * to still exists. If the VMA and others sharing the | |
764 | * data have been freed, then the anon_vma could | |
765 | * already be invalid. | |
766 | * | |
767 | * To avoid this possibility, swapcache pages get | |
768 | * migrated but are not remapped when migration | |
769 | * completes | |
770 | */ | |
771 | remap_swapcache = 0; | |
772 | } else { | |
1ce82b69 | 773 | goto uncharge; |
3fe2011f | 774 | } |
989f89c5 | 775 | } |
62e1c553 | 776 | |
dc386d4d | 777 | /* |
62e1c553 SL |
778 | * Corner case handling: |
779 | * 1. When a new swap-cache page is read into, it is added to the LRU | |
780 | * and treated as swapcache but it has no rmap yet. | |
781 | * Calling try_to_unmap() against a page->mapping==NULL page will | |
782 | * trigger a BUG. So handle it here. | |
783 | * 2. An orphaned page (see truncate_complete_page) might have | |
784 | * fs-private metadata. The page can be picked up due to memory | |
785 | * offlining. Everywhere else except page reclaim, the page is | |
786 | * invisible to the vm, so the page can not be migrated. So try to | |
787 | * free the metadata, so the page can be freed. | |
e24f0b8f | 788 | */ |
62e1c553 | 789 | if (!page->mapping) { |
1ce82b69 HD |
790 | VM_BUG_ON(PageAnon(page)); |
791 | if (page_has_private(page)) { | |
62e1c553 | 792 | try_to_free_buffers(page); |
1ce82b69 | 793 | goto uncharge; |
62e1c553 | 794 | } |
abfc3488 | 795 | goto skip_unmap; |
62e1c553 SL |
796 | } |
797 | ||
dc386d4d | 798 | /* Establish migration ptes or remove ptes */ |
14fa31b8 | 799 | try_to_unmap(page, TTU_MIGRATION|TTU_IGNORE_MLOCK|TTU_IGNORE_ACCESS); |
dc386d4d | 800 | |
abfc3488 | 801 | skip_unmap: |
e6a1530d | 802 | if (!page_mapped(page)) |
11bc82d6 | 803 | rc = move_to_new_page(newpage, page, remap_swapcache, sync); |
e24f0b8f | 804 | |
3fe2011f | 805 | if (rc && remap_swapcache) |
e24f0b8f | 806 | remove_migration_ptes(page, page); |
3f6c8272 MG |
807 | |
808 | /* Drop an anon_vma reference if we took one */ | |
76545066 | 809 | if (anon_vma) |
9e60109f | 810 | put_anon_vma(anon_vma); |
3f6c8272 | 811 | |
01b1ae63 KH |
812 | uncharge: |
813 | if (!charge) | |
50de1dd9 | 814 | mem_cgroup_end_migration(mem, page, newpage, rc == 0); |
e24f0b8f CL |
815 | unlock: |
816 | unlock_page(page); | |
0dabec93 MK |
817 | out: |
818 | return rc; | |
819 | } | |
95a402c3 | 820 | |
0dabec93 MK |
821 | /* |
822 | * Obtain the lock on page, remove all ptes and migrate the page | |
823 | * to the newly allocated page in newpage. | |
824 | */ | |
825 | static int unmap_and_move(new_page_t get_new_page, unsigned long private, | |
826 | struct page *page, int force, bool offlining, bool sync) | |
827 | { | |
828 | int rc = 0; | |
829 | int *result = NULL; | |
830 | struct page *newpage = get_new_page(page, private, &result); | |
831 | ||
832 | if (!newpage) | |
833 | return -ENOMEM; | |
834 | ||
4e5f01c2 KH |
835 | mem_cgroup_reset_owner(newpage); |
836 | ||
0dabec93 MK |
837 | if (page_count(page) == 1) { |
838 | /* page was freed from under us. So we are done. */ | |
839 | goto out; | |
840 | } | |
841 | ||
842 | if (unlikely(PageTransHuge(page))) | |
843 | if (unlikely(split_huge_page(page))) | |
844 | goto out; | |
845 | ||
846 | rc = __unmap_and_move(page, newpage, force, offlining, sync); | |
847 | out: | |
e24f0b8f | 848 | if (rc != -EAGAIN) { |
0dabec93 MK |
849 | /* |
850 | * A page that has been migrated has all references | |
851 | * removed and will be freed. A page that has not been | |
852 | * migrated will have kepts its references and be | |
853 | * restored. | |
854 | */ | |
855 | list_del(&page->lru); | |
a731286d | 856 | dec_zone_page_state(page, NR_ISOLATED_ANON + |
6c0b1351 | 857 | page_is_file_cache(page)); |
894bc310 | 858 | putback_lru_page(page); |
e24f0b8f | 859 | } |
95a402c3 CL |
860 | /* |
861 | * Move the new page to the LRU. If migration was not successful | |
862 | * then this will free the page. | |
863 | */ | |
894bc310 | 864 | putback_lru_page(newpage); |
742755a1 CL |
865 | if (result) { |
866 | if (rc) | |
867 | *result = rc; | |
868 | else | |
869 | *result = page_to_nid(newpage); | |
870 | } | |
e24f0b8f CL |
871 | return rc; |
872 | } | |
873 | ||
290408d4 NH |
874 | /* |
875 | * Counterpart of unmap_and_move_page() for hugepage migration. | |
876 | * | |
877 | * This function doesn't wait the completion of hugepage I/O | |
878 | * because there is no race between I/O and migration for hugepage. | |
879 | * Note that currently hugepage I/O occurs only in direct I/O | |
880 | * where no lock is held and PG_writeback is irrelevant, | |
881 | * and writeback status of all subpages are counted in the reference | |
882 | * count of the head page (i.e. if all subpages of a 2MB hugepage are | |
883 | * under direct I/O, the reference of the head page is 512 and a bit more.) | |
884 | * This means that when we try to migrate hugepage whose subpages are | |
885 | * doing direct I/O, some references remain after try_to_unmap() and | |
886 | * hugepage migration fails without data corruption. | |
887 | * | |
888 | * There is also no race when direct I/O is issued on the page under migration, | |
889 | * because then pte is replaced with migration swap entry and direct I/O code | |
890 | * will wait in the page fault for migration to complete. | |
891 | */ | |
892 | static int unmap_and_move_huge_page(new_page_t get_new_page, | |
893 | unsigned long private, struct page *hpage, | |
7f0f2496 | 894 | int force, bool offlining, bool sync) |
290408d4 NH |
895 | { |
896 | int rc = 0; | |
897 | int *result = NULL; | |
898 | struct page *new_hpage = get_new_page(hpage, private, &result); | |
290408d4 NH |
899 | struct anon_vma *anon_vma = NULL; |
900 | ||
901 | if (!new_hpage) | |
902 | return -ENOMEM; | |
903 | ||
904 | rc = -EAGAIN; | |
905 | ||
906 | if (!trylock_page(hpage)) { | |
77f1fe6b | 907 | if (!force || !sync) |
290408d4 NH |
908 | goto out; |
909 | lock_page(hpage); | |
910 | } | |
911 | ||
746b18d4 PZ |
912 | if (PageAnon(hpage)) |
913 | anon_vma = page_get_anon_vma(hpage); | |
290408d4 NH |
914 | |
915 | try_to_unmap(hpage, TTU_MIGRATION|TTU_IGNORE_MLOCK|TTU_IGNORE_ACCESS); | |
916 | ||
917 | if (!page_mapped(hpage)) | |
11bc82d6 | 918 | rc = move_to_new_page(new_hpage, hpage, 1, sync); |
290408d4 NH |
919 | |
920 | if (rc) | |
921 | remove_migration_ptes(hpage, hpage); | |
922 | ||
fd4a4663 | 923 | if (anon_vma) |
9e60109f | 924 | put_anon_vma(anon_vma); |
290408d4 NH |
925 | unlock_page(hpage); |
926 | ||
09761333 | 927 | out: |
290408d4 NH |
928 | if (rc != -EAGAIN) { |
929 | list_del(&hpage->lru); | |
930 | put_page(hpage); | |
931 | } | |
932 | ||
933 | put_page(new_hpage); | |
934 | ||
935 | if (result) { | |
936 | if (rc) | |
937 | *result = rc; | |
938 | else | |
939 | *result = page_to_nid(new_hpage); | |
940 | } | |
941 | return rc; | |
942 | } | |
943 | ||
b20a3503 CL |
944 | /* |
945 | * migrate_pages | |
946 | * | |
95a402c3 CL |
947 | * The function takes one list of pages to migrate and a function |
948 | * that determines from the page to be migrated and the private data | |
949 | * the target of the move and allocates the page. | |
b20a3503 CL |
950 | * |
951 | * The function returns after 10 attempts or if no pages | |
952 | * are movable anymore because to has become empty | |
cf608ac1 MK |
953 | * or no retryable pages exist anymore. |
954 | * Caller should call putback_lru_pages to return pages to the LRU | |
28bd6578 | 955 | * or free list only if ret != 0. |
b20a3503 | 956 | * |
95a402c3 | 957 | * Return: Number of pages not migrated or error code. |
b20a3503 | 958 | */ |
95a402c3 | 959 | int migrate_pages(struct list_head *from, |
7f0f2496 | 960 | new_page_t get_new_page, unsigned long private, bool offlining, |
77f1fe6b | 961 | bool sync) |
b20a3503 | 962 | { |
e24f0b8f | 963 | int retry = 1; |
b20a3503 CL |
964 | int nr_failed = 0; |
965 | int pass = 0; | |
966 | struct page *page; | |
967 | struct page *page2; | |
968 | int swapwrite = current->flags & PF_SWAPWRITE; | |
969 | int rc; | |
970 | ||
971 | if (!swapwrite) | |
972 | current->flags |= PF_SWAPWRITE; | |
973 | ||
e24f0b8f CL |
974 | for(pass = 0; pass < 10 && retry; pass++) { |
975 | retry = 0; | |
b20a3503 | 976 | |
e24f0b8f | 977 | list_for_each_entry_safe(page, page2, from, lru) { |
e24f0b8f | 978 | cond_resched(); |
2d1db3b1 | 979 | |
95a402c3 | 980 | rc = unmap_and_move(get_new_page, private, |
77f1fe6b MG |
981 | page, pass > 2, offlining, |
982 | sync); | |
2d1db3b1 | 983 | |
e24f0b8f | 984 | switch(rc) { |
95a402c3 CL |
985 | case -ENOMEM: |
986 | goto out; | |
e24f0b8f | 987 | case -EAGAIN: |
2d1db3b1 | 988 | retry++; |
e24f0b8f CL |
989 | break; |
990 | case 0: | |
e24f0b8f CL |
991 | break; |
992 | default: | |
2d1db3b1 | 993 | /* Permanent failure */ |
2d1db3b1 | 994 | nr_failed++; |
e24f0b8f | 995 | break; |
2d1db3b1 | 996 | } |
b20a3503 CL |
997 | } |
998 | } | |
95a402c3 CL |
999 | rc = 0; |
1000 | out: | |
b20a3503 CL |
1001 | if (!swapwrite) |
1002 | current->flags &= ~PF_SWAPWRITE; | |
1003 | ||
95a402c3 CL |
1004 | if (rc) |
1005 | return rc; | |
b20a3503 | 1006 | |
95a402c3 | 1007 | return nr_failed + retry; |
b20a3503 | 1008 | } |
95a402c3 | 1009 | |
290408d4 | 1010 | int migrate_huge_pages(struct list_head *from, |
7f0f2496 | 1011 | new_page_t get_new_page, unsigned long private, bool offlining, |
77f1fe6b | 1012 | bool sync) |
290408d4 NH |
1013 | { |
1014 | int retry = 1; | |
1015 | int nr_failed = 0; | |
1016 | int pass = 0; | |
1017 | struct page *page; | |
1018 | struct page *page2; | |
1019 | int rc; | |
1020 | ||
1021 | for (pass = 0; pass < 10 && retry; pass++) { | |
1022 | retry = 0; | |
1023 | ||
1024 | list_for_each_entry_safe(page, page2, from, lru) { | |
1025 | cond_resched(); | |
1026 | ||
1027 | rc = unmap_and_move_huge_page(get_new_page, | |
77f1fe6b MG |
1028 | private, page, pass > 2, offlining, |
1029 | sync); | |
290408d4 NH |
1030 | |
1031 | switch(rc) { | |
1032 | case -ENOMEM: | |
1033 | goto out; | |
1034 | case -EAGAIN: | |
1035 | retry++; | |
1036 | break; | |
1037 | case 0: | |
1038 | break; | |
1039 | default: | |
1040 | /* Permanent failure */ | |
1041 | nr_failed++; | |
1042 | break; | |
1043 | } | |
1044 | } | |
1045 | } | |
1046 | rc = 0; | |
1047 | out: | |
290408d4 NH |
1048 | if (rc) |
1049 | return rc; | |
1050 | ||
1051 | return nr_failed + retry; | |
1052 | } | |
1053 | ||
742755a1 CL |
1054 | #ifdef CONFIG_NUMA |
1055 | /* | |
1056 | * Move a list of individual pages | |
1057 | */ | |
1058 | struct page_to_node { | |
1059 | unsigned long addr; | |
1060 | struct page *page; | |
1061 | int node; | |
1062 | int status; | |
1063 | }; | |
1064 | ||
1065 | static struct page *new_page_node(struct page *p, unsigned long private, | |
1066 | int **result) | |
1067 | { | |
1068 | struct page_to_node *pm = (struct page_to_node *)private; | |
1069 | ||
1070 | while (pm->node != MAX_NUMNODES && pm->page != p) | |
1071 | pm++; | |
1072 | ||
1073 | if (pm->node == MAX_NUMNODES) | |
1074 | return NULL; | |
1075 | ||
1076 | *result = &pm->status; | |
1077 | ||
6484eb3e | 1078 | return alloc_pages_exact_node(pm->node, |
769848c0 | 1079 | GFP_HIGHUSER_MOVABLE | GFP_THISNODE, 0); |
742755a1 CL |
1080 | } |
1081 | ||
1082 | /* | |
1083 | * Move a set of pages as indicated in the pm array. The addr | |
1084 | * field must be set to the virtual address of the page to be moved | |
1085 | * and the node number must contain a valid target node. | |
5e9a0f02 | 1086 | * The pm array ends with node = MAX_NUMNODES. |
742755a1 | 1087 | */ |
5e9a0f02 BG |
1088 | static int do_move_page_to_node_array(struct mm_struct *mm, |
1089 | struct page_to_node *pm, | |
1090 | int migrate_all) | |
742755a1 CL |
1091 | { |
1092 | int err; | |
1093 | struct page_to_node *pp; | |
1094 | LIST_HEAD(pagelist); | |
1095 | ||
1096 | down_read(&mm->mmap_sem); | |
1097 | ||
1098 | /* | |
1099 | * Build a list of pages to migrate | |
1100 | */ | |
742755a1 CL |
1101 | for (pp = pm; pp->node != MAX_NUMNODES; pp++) { |
1102 | struct vm_area_struct *vma; | |
1103 | struct page *page; | |
1104 | ||
742755a1 CL |
1105 | err = -EFAULT; |
1106 | vma = find_vma(mm, pp->addr); | |
70384dc6 | 1107 | if (!vma || pp->addr < vma->vm_start || !vma_migratable(vma)) |
742755a1 CL |
1108 | goto set_status; |
1109 | ||
500d65d4 | 1110 | page = follow_page(vma, pp->addr, FOLL_GET|FOLL_SPLIT); |
89f5b7da LT |
1111 | |
1112 | err = PTR_ERR(page); | |
1113 | if (IS_ERR(page)) | |
1114 | goto set_status; | |
1115 | ||
742755a1 CL |
1116 | err = -ENOENT; |
1117 | if (!page) | |
1118 | goto set_status; | |
1119 | ||
62b61f61 HD |
1120 | /* Use PageReserved to check for zero page */ |
1121 | if (PageReserved(page) || PageKsm(page)) | |
742755a1 CL |
1122 | goto put_and_set; |
1123 | ||
1124 | pp->page = page; | |
1125 | err = page_to_nid(page); | |
1126 | ||
1127 | if (err == pp->node) | |
1128 | /* | |
1129 | * Node already in the right place | |
1130 | */ | |
1131 | goto put_and_set; | |
1132 | ||
1133 | err = -EACCES; | |
1134 | if (page_mapcount(page) > 1 && | |
1135 | !migrate_all) | |
1136 | goto put_and_set; | |
1137 | ||
62695a84 | 1138 | err = isolate_lru_page(page); |
6d9c285a | 1139 | if (!err) { |
62695a84 | 1140 | list_add_tail(&page->lru, &pagelist); |
6d9c285a KM |
1141 | inc_zone_page_state(page, NR_ISOLATED_ANON + |
1142 | page_is_file_cache(page)); | |
1143 | } | |
742755a1 CL |
1144 | put_and_set: |
1145 | /* | |
1146 | * Either remove the duplicate refcount from | |
1147 | * isolate_lru_page() or drop the page ref if it was | |
1148 | * not isolated. | |
1149 | */ | |
1150 | put_page(page); | |
1151 | set_status: | |
1152 | pp->status = err; | |
1153 | } | |
1154 | ||
e78bbfa8 | 1155 | err = 0; |
cf608ac1 | 1156 | if (!list_empty(&pagelist)) { |
742755a1 | 1157 | err = migrate_pages(&pagelist, new_page_node, |
77f1fe6b | 1158 | (unsigned long)pm, 0, true); |
cf608ac1 MK |
1159 | if (err) |
1160 | putback_lru_pages(&pagelist); | |
1161 | } | |
742755a1 CL |
1162 | |
1163 | up_read(&mm->mmap_sem); | |
1164 | return err; | |
1165 | } | |
1166 | ||
5e9a0f02 BG |
1167 | /* |
1168 | * Migrate an array of page address onto an array of nodes and fill | |
1169 | * the corresponding array of status. | |
1170 | */ | |
1171 | static int do_pages_move(struct mm_struct *mm, struct task_struct *task, | |
1172 | unsigned long nr_pages, | |
1173 | const void __user * __user *pages, | |
1174 | const int __user *nodes, | |
1175 | int __user *status, int flags) | |
1176 | { | |
3140a227 | 1177 | struct page_to_node *pm; |
5e9a0f02 | 1178 | nodemask_t task_nodes; |
3140a227 BG |
1179 | unsigned long chunk_nr_pages; |
1180 | unsigned long chunk_start; | |
1181 | int err; | |
5e9a0f02 BG |
1182 | |
1183 | task_nodes = cpuset_mems_allowed(task); | |
1184 | ||
3140a227 BG |
1185 | err = -ENOMEM; |
1186 | pm = (struct page_to_node *)__get_free_page(GFP_KERNEL); | |
1187 | if (!pm) | |
5e9a0f02 | 1188 | goto out; |
35282a2d BG |
1189 | |
1190 | migrate_prep(); | |
1191 | ||
5e9a0f02 | 1192 | /* |
3140a227 BG |
1193 | * Store a chunk of page_to_node array in a page, |
1194 | * but keep the last one as a marker | |
5e9a0f02 | 1195 | */ |
3140a227 | 1196 | chunk_nr_pages = (PAGE_SIZE / sizeof(struct page_to_node)) - 1; |
5e9a0f02 | 1197 | |
3140a227 BG |
1198 | for (chunk_start = 0; |
1199 | chunk_start < nr_pages; | |
1200 | chunk_start += chunk_nr_pages) { | |
1201 | int j; | |
5e9a0f02 | 1202 | |
3140a227 BG |
1203 | if (chunk_start + chunk_nr_pages > nr_pages) |
1204 | chunk_nr_pages = nr_pages - chunk_start; | |
1205 | ||
1206 | /* fill the chunk pm with addrs and nodes from user-space */ | |
1207 | for (j = 0; j < chunk_nr_pages; j++) { | |
1208 | const void __user *p; | |
5e9a0f02 BG |
1209 | int node; |
1210 | ||
3140a227 BG |
1211 | err = -EFAULT; |
1212 | if (get_user(p, pages + j + chunk_start)) | |
1213 | goto out_pm; | |
1214 | pm[j].addr = (unsigned long) p; | |
1215 | ||
1216 | if (get_user(node, nodes + j + chunk_start)) | |
5e9a0f02 BG |
1217 | goto out_pm; |
1218 | ||
1219 | err = -ENODEV; | |
6f5a55f1 LT |
1220 | if (node < 0 || node >= MAX_NUMNODES) |
1221 | goto out_pm; | |
1222 | ||
5e9a0f02 BG |
1223 | if (!node_state(node, N_HIGH_MEMORY)) |
1224 | goto out_pm; | |
1225 | ||
1226 | err = -EACCES; | |
1227 | if (!node_isset(node, task_nodes)) | |
1228 | goto out_pm; | |
1229 | ||
3140a227 BG |
1230 | pm[j].node = node; |
1231 | } | |
1232 | ||
1233 | /* End marker for this chunk */ | |
1234 | pm[chunk_nr_pages].node = MAX_NUMNODES; | |
1235 | ||
1236 | /* Migrate this chunk */ | |
1237 | err = do_move_page_to_node_array(mm, pm, | |
1238 | flags & MPOL_MF_MOVE_ALL); | |
1239 | if (err < 0) | |
1240 | goto out_pm; | |
5e9a0f02 | 1241 | |
5e9a0f02 | 1242 | /* Return status information */ |
3140a227 BG |
1243 | for (j = 0; j < chunk_nr_pages; j++) |
1244 | if (put_user(pm[j].status, status + j + chunk_start)) { | |
5e9a0f02 | 1245 | err = -EFAULT; |
3140a227 BG |
1246 | goto out_pm; |
1247 | } | |
1248 | } | |
1249 | err = 0; | |
5e9a0f02 BG |
1250 | |
1251 | out_pm: | |
3140a227 | 1252 | free_page((unsigned long)pm); |
5e9a0f02 BG |
1253 | out: |
1254 | return err; | |
1255 | } | |
1256 | ||
742755a1 | 1257 | /* |
2f007e74 | 1258 | * Determine the nodes of an array of pages and store it in an array of status. |
742755a1 | 1259 | */ |
80bba129 BG |
1260 | static void do_pages_stat_array(struct mm_struct *mm, unsigned long nr_pages, |
1261 | const void __user **pages, int *status) | |
742755a1 | 1262 | { |
2f007e74 | 1263 | unsigned long i; |
2f007e74 | 1264 | |
742755a1 CL |
1265 | down_read(&mm->mmap_sem); |
1266 | ||
2f007e74 | 1267 | for (i = 0; i < nr_pages; i++) { |
80bba129 | 1268 | unsigned long addr = (unsigned long)(*pages); |
742755a1 CL |
1269 | struct vm_area_struct *vma; |
1270 | struct page *page; | |
c095adbc | 1271 | int err = -EFAULT; |
2f007e74 BG |
1272 | |
1273 | vma = find_vma(mm, addr); | |
70384dc6 | 1274 | if (!vma || addr < vma->vm_start) |
742755a1 CL |
1275 | goto set_status; |
1276 | ||
2f007e74 | 1277 | page = follow_page(vma, addr, 0); |
89f5b7da LT |
1278 | |
1279 | err = PTR_ERR(page); | |
1280 | if (IS_ERR(page)) | |
1281 | goto set_status; | |
1282 | ||
742755a1 CL |
1283 | err = -ENOENT; |
1284 | /* Use PageReserved to check for zero page */ | |
62b61f61 | 1285 | if (!page || PageReserved(page) || PageKsm(page)) |
742755a1 CL |
1286 | goto set_status; |
1287 | ||
1288 | err = page_to_nid(page); | |
1289 | set_status: | |
80bba129 BG |
1290 | *status = err; |
1291 | ||
1292 | pages++; | |
1293 | status++; | |
1294 | } | |
1295 | ||
1296 | up_read(&mm->mmap_sem); | |
1297 | } | |
1298 | ||
1299 | /* | |
1300 | * Determine the nodes of a user array of pages and store it in | |
1301 | * a user array of status. | |
1302 | */ | |
1303 | static int do_pages_stat(struct mm_struct *mm, unsigned long nr_pages, | |
1304 | const void __user * __user *pages, | |
1305 | int __user *status) | |
1306 | { | |
1307 | #define DO_PAGES_STAT_CHUNK_NR 16 | |
1308 | const void __user *chunk_pages[DO_PAGES_STAT_CHUNK_NR]; | |
1309 | int chunk_status[DO_PAGES_STAT_CHUNK_NR]; | |
80bba129 | 1310 | |
87b8d1ad PA |
1311 | while (nr_pages) { |
1312 | unsigned long chunk_nr; | |
80bba129 | 1313 | |
87b8d1ad PA |
1314 | chunk_nr = nr_pages; |
1315 | if (chunk_nr > DO_PAGES_STAT_CHUNK_NR) | |
1316 | chunk_nr = DO_PAGES_STAT_CHUNK_NR; | |
1317 | ||
1318 | if (copy_from_user(chunk_pages, pages, chunk_nr * sizeof(*chunk_pages))) | |
1319 | break; | |
80bba129 BG |
1320 | |
1321 | do_pages_stat_array(mm, chunk_nr, chunk_pages, chunk_status); | |
1322 | ||
87b8d1ad PA |
1323 | if (copy_to_user(status, chunk_status, chunk_nr * sizeof(*status))) |
1324 | break; | |
742755a1 | 1325 | |
87b8d1ad PA |
1326 | pages += chunk_nr; |
1327 | status += chunk_nr; | |
1328 | nr_pages -= chunk_nr; | |
1329 | } | |
1330 | return nr_pages ? -EFAULT : 0; | |
742755a1 CL |
1331 | } |
1332 | ||
1333 | /* | |
1334 | * Move a list of pages in the address space of the currently executing | |
1335 | * process. | |
1336 | */ | |
938bb9f5 HC |
1337 | SYSCALL_DEFINE6(move_pages, pid_t, pid, unsigned long, nr_pages, |
1338 | const void __user * __user *, pages, | |
1339 | const int __user *, nodes, | |
1340 | int __user *, status, int, flags) | |
742755a1 | 1341 | { |
c69e8d9c | 1342 | const struct cred *cred = current_cred(), *tcred; |
742755a1 | 1343 | struct task_struct *task; |
742755a1 | 1344 | struct mm_struct *mm; |
5e9a0f02 | 1345 | int err; |
742755a1 CL |
1346 | |
1347 | /* Check flags */ | |
1348 | if (flags & ~(MPOL_MF_MOVE|MPOL_MF_MOVE_ALL)) | |
1349 | return -EINVAL; | |
1350 | ||
1351 | if ((flags & MPOL_MF_MOVE_ALL) && !capable(CAP_SYS_NICE)) | |
1352 | return -EPERM; | |
1353 | ||
1354 | /* Find the mm_struct */ | |
a879bf58 | 1355 | rcu_read_lock(); |
228ebcbe | 1356 | task = pid ? find_task_by_vpid(pid) : current; |
742755a1 | 1357 | if (!task) { |
a879bf58 | 1358 | rcu_read_unlock(); |
742755a1 CL |
1359 | return -ESRCH; |
1360 | } | |
1361 | mm = get_task_mm(task); | |
a879bf58 | 1362 | rcu_read_unlock(); |
742755a1 CL |
1363 | |
1364 | if (!mm) | |
1365 | return -EINVAL; | |
1366 | ||
1367 | /* | |
1368 | * Check if this process has the right to modify the specified | |
1369 | * process. The right exists if the process has administrative | |
1370 | * capabilities, superuser privileges or the same | |
1371 | * userid as the target process. | |
1372 | */ | |
c69e8d9c DH |
1373 | rcu_read_lock(); |
1374 | tcred = __task_cred(task); | |
b6dff3ec DH |
1375 | if (cred->euid != tcred->suid && cred->euid != tcred->uid && |
1376 | cred->uid != tcred->suid && cred->uid != tcred->uid && | |
742755a1 | 1377 | !capable(CAP_SYS_NICE)) { |
c69e8d9c | 1378 | rcu_read_unlock(); |
742755a1 | 1379 | err = -EPERM; |
5e9a0f02 | 1380 | goto out; |
742755a1 | 1381 | } |
c69e8d9c | 1382 | rcu_read_unlock(); |
742755a1 | 1383 | |
86c3a764 DQ |
1384 | err = security_task_movememory(task); |
1385 | if (err) | |
5e9a0f02 | 1386 | goto out; |
86c3a764 | 1387 | |
5e9a0f02 BG |
1388 | if (nodes) { |
1389 | err = do_pages_move(mm, task, nr_pages, pages, nodes, status, | |
1390 | flags); | |
1391 | } else { | |
2f007e74 | 1392 | err = do_pages_stat(mm, nr_pages, pages, status); |
742755a1 CL |
1393 | } |
1394 | ||
742755a1 | 1395 | out: |
742755a1 CL |
1396 | mmput(mm); |
1397 | return err; | |
1398 | } | |
742755a1 | 1399 | |
7b2259b3 CL |
1400 | /* |
1401 | * Call migration functions in the vma_ops that may prepare | |
1402 | * memory in a vm for migration. migration functions may perform | |
1403 | * the migration for vmas that do not have an underlying page struct. | |
1404 | */ | |
1405 | int migrate_vmas(struct mm_struct *mm, const nodemask_t *to, | |
1406 | const nodemask_t *from, unsigned long flags) | |
1407 | { | |
1408 | struct vm_area_struct *vma; | |
1409 | int err = 0; | |
1410 | ||
1001c9fb | 1411 | for (vma = mm->mmap; vma && !err; vma = vma->vm_next) { |
7b2259b3 CL |
1412 | if (vma->vm_ops && vma->vm_ops->migrate) { |
1413 | err = vma->vm_ops->migrate(vma, to, from, flags); | |
1414 | if (err) | |
1415 | break; | |
1416 | } | |
1417 | } | |
1418 | return err; | |
1419 | } | |
83d1674a | 1420 | #endif |