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> |
8e6ac7fa | 36 | #include <linux/hugetlb_cgroup.h> |
5a0e3ad6 | 37 | #include <linux/gfp.h> |
b20a3503 | 38 | |
0d1836c3 MN |
39 | #include <asm/tlbflush.h> |
40 | ||
b20a3503 CL |
41 | #include "internal.h" |
42 | ||
b20a3503 | 43 | /* |
742755a1 | 44 | * migrate_prep() needs to be called before we start compiling a list of pages |
748446bb MG |
45 | * to be migrated using isolate_lru_page(). If scheduling work on other CPUs is |
46 | * undesirable, use migrate_prep_local() | |
b20a3503 CL |
47 | */ |
48 | int migrate_prep(void) | |
49 | { | |
b20a3503 CL |
50 | /* |
51 | * Clear the LRU lists so pages can be isolated. | |
52 | * Note that pages may be moved off the LRU after we have | |
53 | * drained them. Those pages will fail to migrate like other | |
54 | * pages that may be busy. | |
55 | */ | |
56 | lru_add_drain_all(); | |
57 | ||
58 | return 0; | |
59 | } | |
60 | ||
748446bb MG |
61 | /* Do the necessary work of migrate_prep but not if it involves other CPUs */ |
62 | int migrate_prep_local(void) | |
63 | { | |
64 | lru_add_drain(); | |
65 | ||
66 | return 0; | |
67 | } | |
68 | ||
b20a3503 | 69 | /* |
894bc310 LS |
70 | * Add isolated pages on the list back to the LRU under page lock |
71 | * to avoid leaking evictable pages back onto unevictable list. | |
b20a3503 | 72 | */ |
e13861d8 | 73 | void putback_lru_pages(struct list_head *l) |
b20a3503 CL |
74 | { |
75 | struct page *page; | |
76 | struct page *page2; | |
b20a3503 CL |
77 | |
78 | list_for_each_entry_safe(page, page2, l, lru) { | |
e24f0b8f | 79 | list_del(&page->lru); |
a731286d | 80 | dec_zone_page_state(page, NR_ISOLATED_ANON + |
6c0b1351 | 81 | page_is_file_cache(page)); |
894bc310 | 82 | putback_lru_page(page); |
b20a3503 | 83 | } |
b20a3503 CL |
84 | } |
85 | ||
0697212a CL |
86 | /* |
87 | * Restore a potential migration pte to a working pte entry | |
88 | */ | |
e9995ef9 HD |
89 | static int remove_migration_pte(struct page *new, struct vm_area_struct *vma, |
90 | unsigned long addr, void *old) | |
0697212a CL |
91 | { |
92 | struct mm_struct *mm = vma->vm_mm; | |
93 | swp_entry_t entry; | |
0697212a CL |
94 | pmd_t *pmd; |
95 | pte_t *ptep, pte; | |
96 | spinlock_t *ptl; | |
97 | ||
290408d4 NH |
98 | if (unlikely(PageHuge(new))) { |
99 | ptep = huge_pte_offset(mm, addr); | |
100 | if (!ptep) | |
101 | goto out; | |
102 | ptl = &mm->page_table_lock; | |
103 | } else { | |
6219049a BL |
104 | pmd = mm_find_pmd(mm, addr); |
105 | if (!pmd) | |
290408d4 | 106 | goto out; |
500d65d4 AA |
107 | if (pmd_trans_huge(*pmd)) |
108 | goto out; | |
0697212a | 109 | |
290408d4 | 110 | ptep = pte_offset_map(pmd, addr); |
0697212a | 111 | |
486cf46f HD |
112 | /* |
113 | * Peek to check is_swap_pte() before taking ptlock? No, we | |
114 | * can race mremap's move_ptes(), which skips anon_vma lock. | |
115 | */ | |
290408d4 NH |
116 | |
117 | ptl = pte_lockptr(mm, pmd); | |
118 | } | |
0697212a | 119 | |
0697212a CL |
120 | spin_lock(ptl); |
121 | pte = *ptep; | |
122 | if (!is_swap_pte(pte)) | |
e9995ef9 | 123 | goto unlock; |
0697212a CL |
124 | |
125 | entry = pte_to_swp_entry(pte); | |
126 | ||
e9995ef9 HD |
127 | if (!is_migration_entry(entry) || |
128 | migration_entry_to_page(entry) != old) | |
129 | goto unlock; | |
0697212a | 130 | |
0697212a CL |
131 | get_page(new); |
132 | pte = pte_mkold(mk_pte(new, vma->vm_page_prot)); | |
133 | if (is_write_migration_entry(entry)) | |
134 | pte = pte_mkwrite(pte); | |
3ef8fd7f | 135 | #ifdef CONFIG_HUGETLB_PAGE |
290408d4 NH |
136 | if (PageHuge(new)) |
137 | pte = pte_mkhuge(pte); | |
3ef8fd7f | 138 | #endif |
97ee0524 | 139 | flush_cache_page(vma, addr, pte_pfn(pte)); |
0697212a | 140 | set_pte_at(mm, addr, ptep, pte); |
04e62a29 | 141 | |
290408d4 NH |
142 | if (PageHuge(new)) { |
143 | if (PageAnon(new)) | |
144 | hugepage_add_anon_rmap(new, vma, addr); | |
145 | else | |
146 | page_dup_rmap(new); | |
147 | } else if (PageAnon(new)) | |
04e62a29 CL |
148 | page_add_anon_rmap(new, vma, addr); |
149 | else | |
150 | page_add_file_rmap(new); | |
151 | ||
152 | /* No need to invalidate - it was non-present before */ | |
4b3073e1 | 153 | update_mmu_cache(vma, addr, ptep); |
e9995ef9 | 154 | unlock: |
0697212a | 155 | pte_unmap_unlock(ptep, ptl); |
e9995ef9 HD |
156 | out: |
157 | return SWAP_AGAIN; | |
0697212a CL |
158 | } |
159 | ||
04e62a29 CL |
160 | /* |
161 | * Get rid of all migration entries and replace them by | |
162 | * references to the indicated page. | |
163 | */ | |
164 | static void remove_migration_ptes(struct page *old, struct page *new) | |
165 | { | |
e9995ef9 | 166 | rmap_walk(new, remove_migration_pte, old); |
04e62a29 CL |
167 | } |
168 | ||
0697212a CL |
169 | /* |
170 | * Something used the pte of a page under migration. We need to | |
171 | * get to the page and wait until migration is finished. | |
172 | * When we return from this function the fault will be retried. | |
0697212a CL |
173 | */ |
174 | void migration_entry_wait(struct mm_struct *mm, pmd_t *pmd, | |
175 | unsigned long address) | |
176 | { | |
177 | pte_t *ptep, pte; | |
178 | spinlock_t *ptl; | |
179 | swp_entry_t entry; | |
180 | struct page *page; | |
181 | ||
182 | ptep = pte_offset_map_lock(mm, pmd, address, &ptl); | |
183 | pte = *ptep; | |
184 | if (!is_swap_pte(pte)) | |
185 | goto out; | |
186 | ||
187 | entry = pte_to_swp_entry(pte); | |
188 | if (!is_migration_entry(entry)) | |
189 | goto out; | |
190 | ||
191 | page = migration_entry_to_page(entry); | |
192 | ||
e286781d NP |
193 | /* |
194 | * Once radix-tree replacement of page migration started, page_count | |
195 | * *must* be zero. And, we don't want to call wait_on_page_locked() | |
196 | * against a page without get_page(). | |
197 | * So, we use get_page_unless_zero(), here. Even failed, page fault | |
198 | * will occur again. | |
199 | */ | |
200 | if (!get_page_unless_zero(page)) | |
201 | goto out; | |
0697212a CL |
202 | pte_unmap_unlock(ptep, ptl); |
203 | wait_on_page_locked(page); | |
204 | put_page(page); | |
205 | return; | |
206 | out: | |
207 | pte_unmap_unlock(ptep, ptl); | |
208 | } | |
209 | ||
b969c4ab MG |
210 | #ifdef CONFIG_BLOCK |
211 | /* Returns true if all buffers are successfully locked */ | |
a6bc32b8 MG |
212 | static bool buffer_migrate_lock_buffers(struct buffer_head *head, |
213 | enum migrate_mode mode) | |
b969c4ab MG |
214 | { |
215 | struct buffer_head *bh = head; | |
216 | ||
217 | /* Simple case, sync compaction */ | |
a6bc32b8 | 218 | if (mode != MIGRATE_ASYNC) { |
b969c4ab MG |
219 | do { |
220 | get_bh(bh); | |
221 | lock_buffer(bh); | |
222 | bh = bh->b_this_page; | |
223 | ||
224 | } while (bh != head); | |
225 | ||
226 | return true; | |
227 | } | |
228 | ||
229 | /* async case, we cannot block on lock_buffer so use trylock_buffer */ | |
230 | do { | |
231 | get_bh(bh); | |
232 | if (!trylock_buffer(bh)) { | |
233 | /* | |
234 | * We failed to lock the buffer and cannot stall in | |
235 | * async migration. Release the taken locks | |
236 | */ | |
237 | struct buffer_head *failed_bh = bh; | |
238 | put_bh(failed_bh); | |
239 | bh = head; | |
240 | while (bh != failed_bh) { | |
241 | unlock_buffer(bh); | |
242 | put_bh(bh); | |
243 | bh = bh->b_this_page; | |
244 | } | |
245 | return false; | |
246 | } | |
247 | ||
248 | bh = bh->b_this_page; | |
249 | } while (bh != head); | |
250 | return true; | |
251 | } | |
252 | #else | |
253 | static inline bool buffer_migrate_lock_buffers(struct buffer_head *head, | |
a6bc32b8 | 254 | enum migrate_mode mode) |
b969c4ab MG |
255 | { |
256 | return true; | |
257 | } | |
258 | #endif /* CONFIG_BLOCK */ | |
259 | ||
b20a3503 | 260 | /* |
c3fcf8a5 | 261 | * Replace the page in the mapping. |
5b5c7120 CL |
262 | * |
263 | * The number of remaining references must be: | |
264 | * 1 for anonymous pages without a mapping | |
265 | * 2 for pages with a mapping | |
266cf658 | 266 | * 3 for pages with a mapping and PagePrivate/PagePrivate2 set. |
b20a3503 | 267 | */ |
2d1db3b1 | 268 | static int migrate_page_move_mapping(struct address_space *mapping, |
b969c4ab | 269 | struct page *newpage, struct page *page, |
a6bc32b8 | 270 | struct buffer_head *head, enum migrate_mode mode) |
b20a3503 | 271 | { |
e286781d | 272 | int expected_count; |
7cf9c2c7 | 273 | void **pslot; |
b20a3503 | 274 | |
6c5240ae | 275 | if (!mapping) { |
0e8c7d0f | 276 | /* Anonymous page without mapping */ |
6c5240ae CL |
277 | if (page_count(page) != 1) |
278 | return -EAGAIN; | |
279 | return 0; | |
280 | } | |
281 | ||
19fd6231 | 282 | spin_lock_irq(&mapping->tree_lock); |
b20a3503 | 283 | |
7cf9c2c7 NP |
284 | pslot = radix_tree_lookup_slot(&mapping->page_tree, |
285 | page_index(page)); | |
b20a3503 | 286 | |
edcf4748 | 287 | expected_count = 2 + page_has_private(page); |
e286781d | 288 | if (page_count(page) != expected_count || |
29c1f677 | 289 | radix_tree_deref_slot_protected(pslot, &mapping->tree_lock) != page) { |
19fd6231 | 290 | spin_unlock_irq(&mapping->tree_lock); |
e23ca00b | 291 | return -EAGAIN; |
b20a3503 CL |
292 | } |
293 | ||
e286781d | 294 | if (!page_freeze_refs(page, expected_count)) { |
19fd6231 | 295 | spin_unlock_irq(&mapping->tree_lock); |
e286781d NP |
296 | return -EAGAIN; |
297 | } | |
298 | ||
b969c4ab MG |
299 | /* |
300 | * In the async migration case of moving a page with buffers, lock the | |
301 | * buffers using trylock before the mapping is moved. If the mapping | |
302 | * was moved, we later failed to lock the buffers and could not move | |
303 | * the mapping back due to an elevated page count, we would have to | |
304 | * block waiting on other references to be dropped. | |
305 | */ | |
a6bc32b8 MG |
306 | if (mode == MIGRATE_ASYNC && head && |
307 | !buffer_migrate_lock_buffers(head, mode)) { | |
b969c4ab MG |
308 | page_unfreeze_refs(page, expected_count); |
309 | spin_unlock_irq(&mapping->tree_lock); | |
310 | return -EAGAIN; | |
311 | } | |
312 | ||
b20a3503 CL |
313 | /* |
314 | * Now we know that no one else is looking at the page. | |
b20a3503 | 315 | */ |
7cf9c2c7 | 316 | get_page(newpage); /* add cache reference */ |
b20a3503 CL |
317 | if (PageSwapCache(page)) { |
318 | SetPageSwapCache(newpage); | |
319 | set_page_private(newpage, page_private(page)); | |
320 | } | |
321 | ||
7cf9c2c7 NP |
322 | radix_tree_replace_slot(pslot, newpage); |
323 | ||
324 | /* | |
937a94c9 JG |
325 | * Drop cache reference from old page by unfreezing |
326 | * to one less reference. | |
7cf9c2c7 NP |
327 | * We know this isn't the last reference. |
328 | */ | |
937a94c9 | 329 | page_unfreeze_refs(page, expected_count - 1); |
7cf9c2c7 | 330 | |
0e8c7d0f CL |
331 | /* |
332 | * If moved to a different zone then also account | |
333 | * the page for that zone. Other VM counters will be | |
334 | * taken care of when we establish references to the | |
335 | * new page and drop references to the old page. | |
336 | * | |
337 | * Note that anonymous pages are accounted for | |
338 | * via NR_FILE_PAGES and NR_ANON_PAGES if they | |
339 | * are mapped to swap space. | |
340 | */ | |
341 | __dec_zone_page_state(page, NR_FILE_PAGES); | |
342 | __inc_zone_page_state(newpage, NR_FILE_PAGES); | |
99a15e21 | 343 | if (!PageSwapCache(page) && PageSwapBacked(page)) { |
4b02108a KM |
344 | __dec_zone_page_state(page, NR_SHMEM); |
345 | __inc_zone_page_state(newpage, NR_SHMEM); | |
346 | } | |
19fd6231 | 347 | spin_unlock_irq(&mapping->tree_lock); |
b20a3503 CL |
348 | |
349 | return 0; | |
350 | } | |
b20a3503 | 351 | |
290408d4 NH |
352 | /* |
353 | * The expected number of remaining references is the same as that | |
354 | * of migrate_page_move_mapping(). | |
355 | */ | |
356 | int migrate_huge_page_move_mapping(struct address_space *mapping, | |
357 | struct page *newpage, struct page *page) | |
358 | { | |
359 | int expected_count; | |
360 | void **pslot; | |
361 | ||
362 | if (!mapping) { | |
363 | if (page_count(page) != 1) | |
364 | return -EAGAIN; | |
365 | return 0; | |
366 | } | |
367 | ||
368 | spin_lock_irq(&mapping->tree_lock); | |
369 | ||
370 | pslot = radix_tree_lookup_slot(&mapping->page_tree, | |
371 | page_index(page)); | |
372 | ||
373 | expected_count = 2 + page_has_private(page); | |
374 | if (page_count(page) != expected_count || | |
29c1f677 | 375 | radix_tree_deref_slot_protected(pslot, &mapping->tree_lock) != page) { |
290408d4 NH |
376 | spin_unlock_irq(&mapping->tree_lock); |
377 | return -EAGAIN; | |
378 | } | |
379 | ||
380 | if (!page_freeze_refs(page, expected_count)) { | |
381 | spin_unlock_irq(&mapping->tree_lock); | |
382 | return -EAGAIN; | |
383 | } | |
384 | ||
385 | get_page(newpage); | |
386 | ||
387 | radix_tree_replace_slot(pslot, newpage); | |
388 | ||
937a94c9 | 389 | page_unfreeze_refs(page, expected_count - 1); |
290408d4 NH |
390 | |
391 | spin_unlock_irq(&mapping->tree_lock); | |
392 | return 0; | |
393 | } | |
394 | ||
b20a3503 CL |
395 | /* |
396 | * Copy the page to its new location | |
397 | */ | |
290408d4 | 398 | void migrate_page_copy(struct page *newpage, struct page *page) |
b20a3503 | 399 | { |
290408d4 NH |
400 | if (PageHuge(page)) |
401 | copy_huge_page(newpage, page); | |
402 | else | |
403 | copy_highpage(newpage, page); | |
b20a3503 CL |
404 | |
405 | if (PageError(page)) | |
406 | SetPageError(newpage); | |
407 | if (PageReferenced(page)) | |
408 | SetPageReferenced(newpage); | |
409 | if (PageUptodate(page)) | |
410 | SetPageUptodate(newpage); | |
894bc310 LS |
411 | if (TestClearPageActive(page)) { |
412 | VM_BUG_ON(PageUnevictable(page)); | |
b20a3503 | 413 | SetPageActive(newpage); |
418b27ef LS |
414 | } else if (TestClearPageUnevictable(page)) |
415 | SetPageUnevictable(newpage); | |
b20a3503 CL |
416 | if (PageChecked(page)) |
417 | SetPageChecked(newpage); | |
418 | if (PageMappedToDisk(page)) | |
419 | SetPageMappedToDisk(newpage); | |
420 | ||
421 | if (PageDirty(page)) { | |
422 | clear_page_dirty_for_io(page); | |
3a902c5f NP |
423 | /* |
424 | * Want to mark the page and the radix tree as dirty, and | |
425 | * redo the accounting that clear_page_dirty_for_io undid, | |
426 | * but we can't use set_page_dirty because that function | |
427 | * is actually a signal that all of the page has become dirty. | |
25985edc | 428 | * Whereas only part of our page may be dirty. |
3a902c5f | 429 | */ |
752dc185 HD |
430 | if (PageSwapBacked(page)) |
431 | SetPageDirty(newpage); | |
432 | else | |
433 | __set_page_dirty_nobuffers(newpage); | |
b20a3503 CL |
434 | } |
435 | ||
b291f000 | 436 | mlock_migrate_page(newpage, page); |
e9995ef9 | 437 | ksm_migrate_page(newpage, page); |
b291f000 | 438 | |
b20a3503 | 439 | ClearPageSwapCache(page); |
b20a3503 CL |
440 | ClearPagePrivate(page); |
441 | set_page_private(page, 0); | |
b20a3503 CL |
442 | |
443 | /* | |
444 | * If any waiters have accumulated on the new page then | |
445 | * wake them up. | |
446 | */ | |
447 | if (PageWriteback(newpage)) | |
448 | end_page_writeback(newpage); | |
449 | } | |
b20a3503 | 450 | |
1d8b85cc CL |
451 | /************************************************************ |
452 | * Migration functions | |
453 | ***********************************************************/ | |
454 | ||
455 | /* Always fail migration. Used for mappings that are not movable */ | |
2d1db3b1 CL |
456 | int fail_migrate_page(struct address_space *mapping, |
457 | struct page *newpage, struct page *page) | |
1d8b85cc CL |
458 | { |
459 | return -EIO; | |
460 | } | |
461 | EXPORT_SYMBOL(fail_migrate_page); | |
462 | ||
b20a3503 CL |
463 | /* |
464 | * Common logic to directly migrate a single page suitable for | |
266cf658 | 465 | * pages that do not use PagePrivate/PagePrivate2. |
b20a3503 CL |
466 | * |
467 | * Pages are locked upon entry and exit. | |
468 | */ | |
2d1db3b1 | 469 | int migrate_page(struct address_space *mapping, |
a6bc32b8 MG |
470 | struct page *newpage, struct page *page, |
471 | enum migrate_mode mode) | |
b20a3503 CL |
472 | { |
473 | int rc; | |
474 | ||
475 | BUG_ON(PageWriteback(page)); /* Writeback must be complete */ | |
476 | ||
a6bc32b8 | 477 | rc = migrate_page_move_mapping(mapping, newpage, page, NULL, mode); |
b20a3503 CL |
478 | |
479 | if (rc) | |
480 | return rc; | |
481 | ||
482 | migrate_page_copy(newpage, page); | |
b20a3503 CL |
483 | return 0; |
484 | } | |
485 | EXPORT_SYMBOL(migrate_page); | |
486 | ||
9361401e | 487 | #ifdef CONFIG_BLOCK |
1d8b85cc CL |
488 | /* |
489 | * Migration function for pages with buffers. This function can only be used | |
490 | * if the underlying filesystem guarantees that no other references to "page" | |
491 | * exist. | |
492 | */ | |
2d1db3b1 | 493 | int buffer_migrate_page(struct address_space *mapping, |
a6bc32b8 | 494 | struct page *newpage, struct page *page, enum migrate_mode mode) |
1d8b85cc | 495 | { |
1d8b85cc CL |
496 | struct buffer_head *bh, *head; |
497 | int rc; | |
498 | ||
1d8b85cc | 499 | if (!page_has_buffers(page)) |
a6bc32b8 | 500 | return migrate_page(mapping, newpage, page, mode); |
1d8b85cc CL |
501 | |
502 | head = page_buffers(page); | |
503 | ||
a6bc32b8 | 504 | rc = migrate_page_move_mapping(mapping, newpage, page, head, mode); |
1d8b85cc CL |
505 | |
506 | if (rc) | |
507 | return rc; | |
508 | ||
b969c4ab MG |
509 | /* |
510 | * In the async case, migrate_page_move_mapping locked the buffers | |
511 | * with an IRQ-safe spinlock held. In the sync case, the buffers | |
512 | * need to be locked now | |
513 | */ | |
a6bc32b8 MG |
514 | if (mode != MIGRATE_ASYNC) |
515 | BUG_ON(!buffer_migrate_lock_buffers(head, mode)); | |
1d8b85cc CL |
516 | |
517 | ClearPagePrivate(page); | |
518 | set_page_private(newpage, page_private(page)); | |
519 | set_page_private(page, 0); | |
520 | put_page(page); | |
521 | get_page(newpage); | |
522 | ||
523 | bh = head; | |
524 | do { | |
525 | set_bh_page(bh, newpage, bh_offset(bh)); | |
526 | bh = bh->b_this_page; | |
527 | ||
528 | } while (bh != head); | |
529 | ||
530 | SetPagePrivate(newpage); | |
531 | ||
532 | migrate_page_copy(newpage, page); | |
533 | ||
534 | bh = head; | |
535 | do { | |
536 | unlock_buffer(bh); | |
537 | put_bh(bh); | |
538 | bh = bh->b_this_page; | |
539 | ||
540 | } while (bh != head); | |
541 | ||
542 | return 0; | |
543 | } | |
544 | EXPORT_SYMBOL(buffer_migrate_page); | |
9361401e | 545 | #endif |
1d8b85cc | 546 | |
04e62a29 CL |
547 | /* |
548 | * Writeback a page to clean the dirty state | |
549 | */ | |
550 | static int writeout(struct address_space *mapping, struct page *page) | |
8351a6e4 | 551 | { |
04e62a29 CL |
552 | struct writeback_control wbc = { |
553 | .sync_mode = WB_SYNC_NONE, | |
554 | .nr_to_write = 1, | |
555 | .range_start = 0, | |
556 | .range_end = LLONG_MAX, | |
04e62a29 CL |
557 | .for_reclaim = 1 |
558 | }; | |
559 | int rc; | |
560 | ||
561 | if (!mapping->a_ops->writepage) | |
562 | /* No write method for the address space */ | |
563 | return -EINVAL; | |
564 | ||
565 | if (!clear_page_dirty_for_io(page)) | |
566 | /* Someone else already triggered a write */ | |
567 | return -EAGAIN; | |
568 | ||
8351a6e4 | 569 | /* |
04e62a29 CL |
570 | * A dirty page may imply that the underlying filesystem has |
571 | * the page on some queue. So the page must be clean for | |
572 | * migration. Writeout may mean we loose the lock and the | |
573 | * page state is no longer what we checked for earlier. | |
574 | * At this point we know that the migration attempt cannot | |
575 | * be successful. | |
8351a6e4 | 576 | */ |
04e62a29 | 577 | remove_migration_ptes(page, page); |
8351a6e4 | 578 | |
04e62a29 | 579 | rc = mapping->a_ops->writepage(page, &wbc); |
8351a6e4 | 580 | |
04e62a29 CL |
581 | if (rc != AOP_WRITEPAGE_ACTIVATE) |
582 | /* unlocked. Relock */ | |
583 | lock_page(page); | |
584 | ||
bda8550d | 585 | return (rc < 0) ? -EIO : -EAGAIN; |
04e62a29 CL |
586 | } |
587 | ||
588 | /* | |
589 | * Default handling if a filesystem does not provide a migration function. | |
590 | */ | |
591 | static int fallback_migrate_page(struct address_space *mapping, | |
a6bc32b8 | 592 | struct page *newpage, struct page *page, enum migrate_mode mode) |
04e62a29 | 593 | { |
b969c4ab | 594 | if (PageDirty(page)) { |
a6bc32b8 MG |
595 | /* Only writeback pages in full synchronous migration */ |
596 | if (mode != MIGRATE_SYNC) | |
b969c4ab | 597 | return -EBUSY; |
04e62a29 | 598 | return writeout(mapping, page); |
b969c4ab | 599 | } |
8351a6e4 CL |
600 | |
601 | /* | |
602 | * Buffers may be managed in a filesystem specific way. | |
603 | * We must have no buffers or drop them. | |
604 | */ | |
266cf658 | 605 | if (page_has_private(page) && |
8351a6e4 CL |
606 | !try_to_release_page(page, GFP_KERNEL)) |
607 | return -EAGAIN; | |
608 | ||
a6bc32b8 | 609 | return migrate_page(mapping, newpage, page, mode); |
8351a6e4 CL |
610 | } |
611 | ||
e24f0b8f CL |
612 | /* |
613 | * Move a page to a newly allocated page | |
614 | * The page is locked and all ptes have been successfully removed. | |
615 | * | |
616 | * The new page will have replaced the old page if this function | |
617 | * is successful. | |
894bc310 LS |
618 | * |
619 | * Return value: | |
620 | * < 0 - error code | |
621 | * == 0 - success | |
e24f0b8f | 622 | */ |
3fe2011f | 623 | static int move_to_new_page(struct page *newpage, struct page *page, |
a6bc32b8 | 624 | int remap_swapcache, enum migrate_mode mode) |
e24f0b8f CL |
625 | { |
626 | struct address_space *mapping; | |
627 | int rc; | |
628 | ||
629 | /* | |
630 | * Block others from accessing the page when we get around to | |
631 | * establishing additional references. We are the only one | |
632 | * holding a reference to the new page at this point. | |
633 | */ | |
529ae9aa | 634 | if (!trylock_page(newpage)) |
e24f0b8f CL |
635 | BUG(); |
636 | ||
637 | /* Prepare mapping for the new page.*/ | |
638 | newpage->index = page->index; | |
639 | newpage->mapping = page->mapping; | |
b2e18538 RR |
640 | if (PageSwapBacked(page)) |
641 | SetPageSwapBacked(newpage); | |
e24f0b8f CL |
642 | |
643 | mapping = page_mapping(page); | |
644 | if (!mapping) | |
a6bc32b8 | 645 | rc = migrate_page(mapping, newpage, page, mode); |
b969c4ab | 646 | else if (mapping->a_ops->migratepage) |
e24f0b8f | 647 | /* |
b969c4ab MG |
648 | * Most pages have a mapping and most filesystems provide a |
649 | * migratepage callback. Anonymous pages are part of swap | |
650 | * space which also has its own migratepage callback. This | |
651 | * is the most common path for page migration. | |
e24f0b8f | 652 | */ |
b969c4ab | 653 | rc = mapping->a_ops->migratepage(mapping, |
a6bc32b8 | 654 | newpage, page, mode); |
b969c4ab | 655 | else |
a6bc32b8 | 656 | rc = fallback_migrate_page(mapping, newpage, page, mode); |
e24f0b8f | 657 | |
3fe2011f | 658 | if (rc) { |
e24f0b8f | 659 | newpage->mapping = NULL; |
3fe2011f MG |
660 | } else { |
661 | if (remap_swapcache) | |
662 | remove_migration_ptes(page, newpage); | |
35512eca | 663 | page->mapping = NULL; |
3fe2011f | 664 | } |
e24f0b8f CL |
665 | |
666 | unlock_page(newpage); | |
667 | ||
668 | return rc; | |
669 | } | |
670 | ||
0dabec93 | 671 | static int __unmap_and_move(struct page *page, struct page *newpage, |
a6bc32b8 | 672 | int force, bool offlining, enum migrate_mode mode) |
e24f0b8f | 673 | { |
0dabec93 | 674 | int rc = -EAGAIN; |
3fe2011f | 675 | int remap_swapcache = 1; |
56039efa | 676 | struct mem_cgroup *mem; |
3f6c8272 | 677 | struct anon_vma *anon_vma = NULL; |
95a402c3 | 678 | |
529ae9aa | 679 | if (!trylock_page(page)) { |
a6bc32b8 | 680 | if (!force || mode == MIGRATE_ASYNC) |
0dabec93 | 681 | goto out; |
3e7d3449 MG |
682 | |
683 | /* | |
684 | * It's not safe for direct compaction to call lock_page. | |
685 | * For example, during page readahead pages are added locked | |
686 | * to the LRU. Later, when the IO completes the pages are | |
687 | * marked uptodate and unlocked. However, the queueing | |
688 | * could be merging multiple pages for one bio (e.g. | |
689 | * mpage_readpages). If an allocation happens for the | |
690 | * second or third page, the process can end up locking | |
691 | * the same page twice and deadlocking. Rather than | |
692 | * trying to be clever about what pages can be locked, | |
693 | * avoid the use of lock_page for direct compaction | |
694 | * altogether. | |
695 | */ | |
696 | if (current->flags & PF_MEMALLOC) | |
0dabec93 | 697 | goto out; |
3e7d3449 | 698 | |
e24f0b8f CL |
699 | lock_page(page); |
700 | } | |
701 | ||
62b61f61 HD |
702 | /* |
703 | * Only memory hotplug's offline_pages() caller has locked out KSM, | |
704 | * and can safely migrate a KSM page. The other cases have skipped | |
705 | * PageKsm along with PageReserved - but it is only now when we have | |
706 | * the page lock that we can be certain it will not go KSM beneath us | |
707 | * (KSM will not upgrade a page from PageAnon to PageKsm when it sees | |
708 | * its pagecount raised, but only here do we take the page lock which | |
709 | * serializes that). | |
710 | */ | |
711 | if (PageKsm(page) && !offlining) { | |
712 | rc = -EBUSY; | |
713 | goto unlock; | |
714 | } | |
715 | ||
01b1ae63 | 716 | /* charge against new page */ |
0030f535 | 717 | mem_cgroup_prepare_migration(page, newpage, &mem); |
01b1ae63 | 718 | |
e24f0b8f | 719 | if (PageWriteback(page)) { |
11bc82d6 | 720 | /* |
a6bc32b8 MG |
721 | * Only in the case of a full syncronous migration is it |
722 | * necessary to wait for PageWriteback. In the async case, | |
723 | * the retry loop is too short and in the sync-light case, | |
724 | * the overhead of stalling is too much | |
11bc82d6 | 725 | */ |
a6bc32b8 | 726 | if (mode != MIGRATE_SYNC) { |
11bc82d6 AA |
727 | rc = -EBUSY; |
728 | goto uncharge; | |
729 | } | |
730 | if (!force) | |
01b1ae63 | 731 | goto uncharge; |
e24f0b8f CL |
732 | wait_on_page_writeback(page); |
733 | } | |
e24f0b8f | 734 | /* |
dc386d4d KH |
735 | * By try_to_unmap(), page->mapcount goes down to 0 here. In this case, |
736 | * we cannot notice that anon_vma is freed while we migrates a page. | |
1ce82b69 | 737 | * This get_anon_vma() delays freeing anon_vma pointer until the end |
dc386d4d | 738 | * of migration. File cache pages are no problem because of page_lock() |
989f89c5 KH |
739 | * File Caches may use write_page() or lock_page() in migration, then, |
740 | * just care Anon page here. | |
dc386d4d | 741 | */ |
989f89c5 | 742 | if (PageAnon(page)) { |
1ce82b69 HD |
743 | /* |
744 | * Only page_lock_anon_vma() understands the subtleties of | |
745 | * getting a hold on an anon_vma from outside one of its mms. | |
746 | */ | |
746b18d4 | 747 | anon_vma = page_get_anon_vma(page); |
1ce82b69 HD |
748 | if (anon_vma) { |
749 | /* | |
746b18d4 | 750 | * Anon page |
1ce82b69 | 751 | */ |
1ce82b69 | 752 | } else if (PageSwapCache(page)) { |
3fe2011f MG |
753 | /* |
754 | * We cannot be sure that the anon_vma of an unmapped | |
755 | * swapcache page is safe to use because we don't | |
756 | * know in advance if the VMA that this page belonged | |
757 | * to still exists. If the VMA and others sharing the | |
758 | * data have been freed, then the anon_vma could | |
759 | * already be invalid. | |
760 | * | |
761 | * To avoid this possibility, swapcache pages get | |
762 | * migrated but are not remapped when migration | |
763 | * completes | |
764 | */ | |
765 | remap_swapcache = 0; | |
766 | } else { | |
1ce82b69 | 767 | goto uncharge; |
3fe2011f | 768 | } |
989f89c5 | 769 | } |
62e1c553 | 770 | |
dc386d4d | 771 | /* |
62e1c553 SL |
772 | * Corner case handling: |
773 | * 1. When a new swap-cache page is read into, it is added to the LRU | |
774 | * and treated as swapcache but it has no rmap yet. | |
775 | * Calling try_to_unmap() against a page->mapping==NULL page will | |
776 | * trigger a BUG. So handle it here. | |
777 | * 2. An orphaned page (see truncate_complete_page) might have | |
778 | * fs-private metadata. The page can be picked up due to memory | |
779 | * offlining. Everywhere else except page reclaim, the page is | |
780 | * invisible to the vm, so the page can not be migrated. So try to | |
781 | * free the metadata, so the page can be freed. | |
e24f0b8f | 782 | */ |
62e1c553 | 783 | if (!page->mapping) { |
1ce82b69 HD |
784 | VM_BUG_ON(PageAnon(page)); |
785 | if (page_has_private(page)) { | |
62e1c553 | 786 | try_to_free_buffers(page); |
1ce82b69 | 787 | goto uncharge; |
62e1c553 | 788 | } |
abfc3488 | 789 | goto skip_unmap; |
62e1c553 SL |
790 | } |
791 | ||
dc386d4d | 792 | /* Establish migration ptes or remove ptes */ |
14fa31b8 | 793 | try_to_unmap(page, TTU_MIGRATION|TTU_IGNORE_MLOCK|TTU_IGNORE_ACCESS); |
dc386d4d | 794 | |
abfc3488 | 795 | skip_unmap: |
e6a1530d | 796 | if (!page_mapped(page)) |
a6bc32b8 | 797 | rc = move_to_new_page(newpage, page, remap_swapcache, mode); |
e24f0b8f | 798 | |
3fe2011f | 799 | if (rc && remap_swapcache) |
e24f0b8f | 800 | remove_migration_ptes(page, page); |
3f6c8272 MG |
801 | |
802 | /* Drop an anon_vma reference if we took one */ | |
76545066 | 803 | if (anon_vma) |
9e60109f | 804 | put_anon_vma(anon_vma); |
3f6c8272 | 805 | |
01b1ae63 | 806 | uncharge: |
0030f535 | 807 | mem_cgroup_end_migration(mem, page, newpage, rc == 0); |
e24f0b8f CL |
808 | unlock: |
809 | unlock_page(page); | |
0dabec93 MK |
810 | out: |
811 | return rc; | |
812 | } | |
95a402c3 | 813 | |
0dabec93 MK |
814 | /* |
815 | * Obtain the lock on page, remove all ptes and migrate the page | |
816 | * to the newly allocated page in newpage. | |
817 | */ | |
818 | static int unmap_and_move(new_page_t get_new_page, unsigned long private, | |
a6bc32b8 MG |
819 | struct page *page, int force, bool offlining, |
820 | enum migrate_mode mode) | |
0dabec93 MK |
821 | { |
822 | int rc = 0; | |
823 | int *result = NULL; | |
824 | struct page *newpage = get_new_page(page, private, &result); | |
825 | ||
826 | if (!newpage) | |
827 | return -ENOMEM; | |
828 | ||
829 | if (page_count(page) == 1) { | |
830 | /* page was freed from under us. So we are done. */ | |
831 | goto out; | |
832 | } | |
833 | ||
834 | if (unlikely(PageTransHuge(page))) | |
835 | if (unlikely(split_huge_page(page))) | |
836 | goto out; | |
837 | ||
a6bc32b8 | 838 | rc = __unmap_and_move(page, newpage, force, offlining, mode); |
0dabec93 | 839 | out: |
e24f0b8f | 840 | if (rc != -EAGAIN) { |
0dabec93 MK |
841 | /* |
842 | * A page that has been migrated has all references | |
843 | * removed and will be freed. A page that has not been | |
844 | * migrated will have kepts its references and be | |
845 | * restored. | |
846 | */ | |
847 | list_del(&page->lru); | |
a731286d | 848 | dec_zone_page_state(page, NR_ISOLATED_ANON + |
6c0b1351 | 849 | page_is_file_cache(page)); |
894bc310 | 850 | putback_lru_page(page); |
e24f0b8f | 851 | } |
95a402c3 CL |
852 | /* |
853 | * Move the new page to the LRU. If migration was not successful | |
854 | * then this will free the page. | |
855 | */ | |
894bc310 | 856 | putback_lru_page(newpage); |
742755a1 CL |
857 | if (result) { |
858 | if (rc) | |
859 | *result = rc; | |
860 | else | |
861 | *result = page_to_nid(newpage); | |
862 | } | |
e24f0b8f CL |
863 | return rc; |
864 | } | |
865 | ||
290408d4 NH |
866 | /* |
867 | * Counterpart of unmap_and_move_page() for hugepage migration. | |
868 | * | |
869 | * This function doesn't wait the completion of hugepage I/O | |
870 | * because there is no race between I/O and migration for hugepage. | |
871 | * Note that currently hugepage I/O occurs only in direct I/O | |
872 | * where no lock is held and PG_writeback is irrelevant, | |
873 | * and writeback status of all subpages are counted in the reference | |
874 | * count of the head page (i.e. if all subpages of a 2MB hugepage are | |
875 | * under direct I/O, the reference of the head page is 512 and a bit more.) | |
876 | * This means that when we try to migrate hugepage whose subpages are | |
877 | * doing direct I/O, some references remain after try_to_unmap() and | |
878 | * hugepage migration fails without data corruption. | |
879 | * | |
880 | * There is also no race when direct I/O is issued on the page under migration, | |
881 | * because then pte is replaced with migration swap entry and direct I/O code | |
882 | * will wait in the page fault for migration to complete. | |
883 | */ | |
884 | static int unmap_and_move_huge_page(new_page_t get_new_page, | |
885 | unsigned long private, struct page *hpage, | |
a6bc32b8 MG |
886 | int force, bool offlining, |
887 | enum migrate_mode mode) | |
290408d4 NH |
888 | { |
889 | int rc = 0; | |
890 | int *result = NULL; | |
891 | struct page *new_hpage = get_new_page(hpage, private, &result); | |
290408d4 NH |
892 | struct anon_vma *anon_vma = NULL; |
893 | ||
894 | if (!new_hpage) | |
895 | return -ENOMEM; | |
896 | ||
897 | rc = -EAGAIN; | |
898 | ||
899 | if (!trylock_page(hpage)) { | |
a6bc32b8 | 900 | if (!force || mode != MIGRATE_SYNC) |
290408d4 NH |
901 | goto out; |
902 | lock_page(hpage); | |
903 | } | |
904 | ||
746b18d4 PZ |
905 | if (PageAnon(hpage)) |
906 | anon_vma = page_get_anon_vma(hpage); | |
290408d4 NH |
907 | |
908 | try_to_unmap(hpage, TTU_MIGRATION|TTU_IGNORE_MLOCK|TTU_IGNORE_ACCESS); | |
909 | ||
910 | if (!page_mapped(hpage)) | |
a6bc32b8 | 911 | rc = move_to_new_page(new_hpage, hpage, 1, mode); |
290408d4 NH |
912 | |
913 | if (rc) | |
914 | remove_migration_ptes(hpage, hpage); | |
915 | ||
fd4a4663 | 916 | if (anon_vma) |
9e60109f | 917 | put_anon_vma(anon_vma); |
8e6ac7fa AK |
918 | |
919 | if (!rc) | |
920 | hugetlb_cgroup_migrate(hpage, new_hpage); | |
921 | ||
290408d4 | 922 | unlock_page(hpage); |
09761333 | 923 | out: |
290408d4 | 924 | put_page(new_hpage); |
290408d4 NH |
925 | if (result) { |
926 | if (rc) | |
927 | *result = rc; | |
928 | else | |
929 | *result = page_to_nid(new_hpage); | |
930 | } | |
931 | return rc; | |
932 | } | |
933 | ||
b20a3503 CL |
934 | /* |
935 | * migrate_pages | |
936 | * | |
95a402c3 CL |
937 | * The function takes one list of pages to migrate and a function |
938 | * that determines from the page to be migrated and the private data | |
939 | * the target of the move and allocates the page. | |
b20a3503 CL |
940 | * |
941 | * The function returns after 10 attempts or if no pages | |
942 | * are movable anymore because to has become empty | |
cf608ac1 MK |
943 | * or no retryable pages exist anymore. |
944 | * Caller should call putback_lru_pages to return pages to the LRU | |
28bd6578 | 945 | * or free list only if ret != 0. |
b20a3503 | 946 | * |
95a402c3 | 947 | * Return: Number of pages not migrated or error code. |
b20a3503 | 948 | */ |
95a402c3 | 949 | int migrate_pages(struct list_head *from, |
7f0f2496 | 950 | new_page_t get_new_page, unsigned long private, bool offlining, |
a6bc32b8 | 951 | enum migrate_mode mode) |
b20a3503 | 952 | { |
e24f0b8f | 953 | int retry = 1; |
b20a3503 CL |
954 | int nr_failed = 0; |
955 | int pass = 0; | |
956 | struct page *page; | |
957 | struct page *page2; | |
958 | int swapwrite = current->flags & PF_SWAPWRITE; | |
959 | int rc; | |
960 | ||
961 | if (!swapwrite) | |
962 | current->flags |= PF_SWAPWRITE; | |
963 | ||
e24f0b8f CL |
964 | for(pass = 0; pass < 10 && retry; pass++) { |
965 | retry = 0; | |
b20a3503 | 966 | |
e24f0b8f | 967 | list_for_each_entry_safe(page, page2, from, lru) { |
e24f0b8f | 968 | cond_resched(); |
2d1db3b1 | 969 | |
95a402c3 | 970 | rc = unmap_and_move(get_new_page, private, |
77f1fe6b | 971 | page, pass > 2, offlining, |
a6bc32b8 | 972 | mode); |
2d1db3b1 | 973 | |
e24f0b8f | 974 | switch(rc) { |
95a402c3 CL |
975 | case -ENOMEM: |
976 | goto out; | |
e24f0b8f | 977 | case -EAGAIN: |
2d1db3b1 | 978 | retry++; |
e24f0b8f CL |
979 | break; |
980 | case 0: | |
e24f0b8f CL |
981 | break; |
982 | default: | |
2d1db3b1 | 983 | /* Permanent failure */ |
2d1db3b1 | 984 | nr_failed++; |
e24f0b8f | 985 | break; |
2d1db3b1 | 986 | } |
b20a3503 CL |
987 | } |
988 | } | |
95a402c3 CL |
989 | rc = 0; |
990 | out: | |
b20a3503 CL |
991 | if (!swapwrite) |
992 | current->flags &= ~PF_SWAPWRITE; | |
993 | ||
95a402c3 CL |
994 | if (rc) |
995 | return rc; | |
b20a3503 | 996 | |
95a402c3 | 997 | return nr_failed + retry; |
b20a3503 | 998 | } |
95a402c3 | 999 | |
189ebff2 AK |
1000 | int migrate_huge_page(struct page *hpage, new_page_t get_new_page, |
1001 | unsigned long private, bool offlining, | |
1002 | enum migrate_mode mode) | |
290408d4 | 1003 | { |
189ebff2 AK |
1004 | int pass, rc; |
1005 | ||
1006 | for (pass = 0; pass < 10; pass++) { | |
1007 | rc = unmap_and_move_huge_page(get_new_page, | |
1008 | private, hpage, pass > 2, offlining, | |
1009 | mode); | |
1010 | switch (rc) { | |
1011 | case -ENOMEM: | |
1012 | goto out; | |
1013 | case -EAGAIN: | |
1014 | /* try again */ | |
290408d4 | 1015 | cond_resched(); |
189ebff2 AK |
1016 | break; |
1017 | case 0: | |
1018 | goto out; | |
1019 | default: | |
1020 | rc = -EIO; | |
1021 | goto out; | |
290408d4 NH |
1022 | } |
1023 | } | |
290408d4 | 1024 | out: |
189ebff2 | 1025 | return rc; |
290408d4 NH |
1026 | } |
1027 | ||
742755a1 CL |
1028 | #ifdef CONFIG_NUMA |
1029 | /* | |
1030 | * Move a list of individual pages | |
1031 | */ | |
1032 | struct page_to_node { | |
1033 | unsigned long addr; | |
1034 | struct page *page; | |
1035 | int node; | |
1036 | int status; | |
1037 | }; | |
1038 | ||
1039 | static struct page *new_page_node(struct page *p, unsigned long private, | |
1040 | int **result) | |
1041 | { | |
1042 | struct page_to_node *pm = (struct page_to_node *)private; | |
1043 | ||
1044 | while (pm->node != MAX_NUMNODES && pm->page != p) | |
1045 | pm++; | |
1046 | ||
1047 | if (pm->node == MAX_NUMNODES) | |
1048 | return NULL; | |
1049 | ||
1050 | *result = &pm->status; | |
1051 | ||
6484eb3e | 1052 | return alloc_pages_exact_node(pm->node, |
769848c0 | 1053 | GFP_HIGHUSER_MOVABLE | GFP_THISNODE, 0); |
742755a1 CL |
1054 | } |
1055 | ||
1056 | /* | |
1057 | * Move a set of pages as indicated in the pm array. The addr | |
1058 | * field must be set to the virtual address of the page to be moved | |
1059 | * and the node number must contain a valid target node. | |
5e9a0f02 | 1060 | * The pm array ends with node = MAX_NUMNODES. |
742755a1 | 1061 | */ |
5e9a0f02 BG |
1062 | static int do_move_page_to_node_array(struct mm_struct *mm, |
1063 | struct page_to_node *pm, | |
1064 | int migrate_all) | |
742755a1 CL |
1065 | { |
1066 | int err; | |
1067 | struct page_to_node *pp; | |
1068 | LIST_HEAD(pagelist); | |
1069 | ||
1070 | down_read(&mm->mmap_sem); | |
1071 | ||
1072 | /* | |
1073 | * Build a list of pages to migrate | |
1074 | */ | |
742755a1 CL |
1075 | for (pp = pm; pp->node != MAX_NUMNODES; pp++) { |
1076 | struct vm_area_struct *vma; | |
1077 | struct page *page; | |
1078 | ||
742755a1 CL |
1079 | err = -EFAULT; |
1080 | vma = find_vma(mm, pp->addr); | |
70384dc6 | 1081 | if (!vma || pp->addr < vma->vm_start || !vma_migratable(vma)) |
742755a1 CL |
1082 | goto set_status; |
1083 | ||
500d65d4 | 1084 | page = follow_page(vma, pp->addr, FOLL_GET|FOLL_SPLIT); |
89f5b7da LT |
1085 | |
1086 | err = PTR_ERR(page); | |
1087 | if (IS_ERR(page)) | |
1088 | goto set_status; | |
1089 | ||
742755a1 CL |
1090 | err = -ENOENT; |
1091 | if (!page) | |
1092 | goto set_status; | |
1093 | ||
62b61f61 HD |
1094 | /* Use PageReserved to check for zero page */ |
1095 | if (PageReserved(page) || PageKsm(page)) | |
742755a1 CL |
1096 | goto put_and_set; |
1097 | ||
1098 | pp->page = page; | |
1099 | err = page_to_nid(page); | |
1100 | ||
1101 | if (err == pp->node) | |
1102 | /* | |
1103 | * Node already in the right place | |
1104 | */ | |
1105 | goto put_and_set; | |
1106 | ||
1107 | err = -EACCES; | |
1108 | if (page_mapcount(page) > 1 && | |
1109 | !migrate_all) | |
1110 | goto put_and_set; | |
1111 | ||
62695a84 | 1112 | err = isolate_lru_page(page); |
6d9c285a | 1113 | if (!err) { |
62695a84 | 1114 | list_add_tail(&page->lru, &pagelist); |
6d9c285a KM |
1115 | inc_zone_page_state(page, NR_ISOLATED_ANON + |
1116 | page_is_file_cache(page)); | |
1117 | } | |
742755a1 CL |
1118 | put_and_set: |
1119 | /* | |
1120 | * Either remove the duplicate refcount from | |
1121 | * isolate_lru_page() or drop the page ref if it was | |
1122 | * not isolated. | |
1123 | */ | |
1124 | put_page(page); | |
1125 | set_status: | |
1126 | pp->status = err; | |
1127 | } | |
1128 | ||
e78bbfa8 | 1129 | err = 0; |
cf608ac1 | 1130 | if (!list_empty(&pagelist)) { |
742755a1 | 1131 | err = migrate_pages(&pagelist, new_page_node, |
a6bc32b8 | 1132 | (unsigned long)pm, 0, MIGRATE_SYNC); |
cf608ac1 MK |
1133 | if (err) |
1134 | putback_lru_pages(&pagelist); | |
1135 | } | |
742755a1 CL |
1136 | |
1137 | up_read(&mm->mmap_sem); | |
1138 | return err; | |
1139 | } | |
1140 | ||
5e9a0f02 BG |
1141 | /* |
1142 | * Migrate an array of page address onto an array of nodes and fill | |
1143 | * the corresponding array of status. | |
1144 | */ | |
3268c63e | 1145 | static int do_pages_move(struct mm_struct *mm, nodemask_t task_nodes, |
5e9a0f02 BG |
1146 | unsigned long nr_pages, |
1147 | const void __user * __user *pages, | |
1148 | const int __user *nodes, | |
1149 | int __user *status, int flags) | |
1150 | { | |
3140a227 | 1151 | struct page_to_node *pm; |
3140a227 BG |
1152 | unsigned long chunk_nr_pages; |
1153 | unsigned long chunk_start; | |
1154 | int err; | |
5e9a0f02 | 1155 | |
3140a227 BG |
1156 | err = -ENOMEM; |
1157 | pm = (struct page_to_node *)__get_free_page(GFP_KERNEL); | |
1158 | if (!pm) | |
5e9a0f02 | 1159 | goto out; |
35282a2d BG |
1160 | |
1161 | migrate_prep(); | |
1162 | ||
5e9a0f02 | 1163 | /* |
3140a227 BG |
1164 | * Store a chunk of page_to_node array in a page, |
1165 | * but keep the last one as a marker | |
5e9a0f02 | 1166 | */ |
3140a227 | 1167 | chunk_nr_pages = (PAGE_SIZE / sizeof(struct page_to_node)) - 1; |
5e9a0f02 | 1168 | |
3140a227 BG |
1169 | for (chunk_start = 0; |
1170 | chunk_start < nr_pages; | |
1171 | chunk_start += chunk_nr_pages) { | |
1172 | int j; | |
5e9a0f02 | 1173 | |
3140a227 BG |
1174 | if (chunk_start + chunk_nr_pages > nr_pages) |
1175 | chunk_nr_pages = nr_pages - chunk_start; | |
1176 | ||
1177 | /* fill the chunk pm with addrs and nodes from user-space */ | |
1178 | for (j = 0; j < chunk_nr_pages; j++) { | |
1179 | const void __user *p; | |
5e9a0f02 BG |
1180 | int node; |
1181 | ||
3140a227 BG |
1182 | err = -EFAULT; |
1183 | if (get_user(p, pages + j + chunk_start)) | |
1184 | goto out_pm; | |
1185 | pm[j].addr = (unsigned long) p; | |
1186 | ||
1187 | if (get_user(node, nodes + j + chunk_start)) | |
5e9a0f02 BG |
1188 | goto out_pm; |
1189 | ||
1190 | err = -ENODEV; | |
6f5a55f1 LT |
1191 | if (node < 0 || node >= MAX_NUMNODES) |
1192 | goto out_pm; | |
1193 | ||
5e9a0f02 BG |
1194 | if (!node_state(node, N_HIGH_MEMORY)) |
1195 | goto out_pm; | |
1196 | ||
1197 | err = -EACCES; | |
1198 | if (!node_isset(node, task_nodes)) | |
1199 | goto out_pm; | |
1200 | ||
3140a227 BG |
1201 | pm[j].node = node; |
1202 | } | |
1203 | ||
1204 | /* End marker for this chunk */ | |
1205 | pm[chunk_nr_pages].node = MAX_NUMNODES; | |
1206 | ||
1207 | /* Migrate this chunk */ | |
1208 | err = do_move_page_to_node_array(mm, pm, | |
1209 | flags & MPOL_MF_MOVE_ALL); | |
1210 | if (err < 0) | |
1211 | goto out_pm; | |
5e9a0f02 | 1212 | |
5e9a0f02 | 1213 | /* Return status information */ |
3140a227 BG |
1214 | for (j = 0; j < chunk_nr_pages; j++) |
1215 | if (put_user(pm[j].status, status + j + chunk_start)) { | |
5e9a0f02 | 1216 | err = -EFAULT; |
3140a227 BG |
1217 | goto out_pm; |
1218 | } | |
1219 | } | |
1220 | err = 0; | |
5e9a0f02 BG |
1221 | |
1222 | out_pm: | |
3140a227 | 1223 | free_page((unsigned long)pm); |
5e9a0f02 BG |
1224 | out: |
1225 | return err; | |
1226 | } | |
1227 | ||
742755a1 | 1228 | /* |
2f007e74 | 1229 | * Determine the nodes of an array of pages and store it in an array of status. |
742755a1 | 1230 | */ |
80bba129 BG |
1231 | static void do_pages_stat_array(struct mm_struct *mm, unsigned long nr_pages, |
1232 | const void __user **pages, int *status) | |
742755a1 | 1233 | { |
2f007e74 | 1234 | unsigned long i; |
2f007e74 | 1235 | |
742755a1 CL |
1236 | down_read(&mm->mmap_sem); |
1237 | ||
2f007e74 | 1238 | for (i = 0; i < nr_pages; i++) { |
80bba129 | 1239 | unsigned long addr = (unsigned long)(*pages); |
742755a1 CL |
1240 | struct vm_area_struct *vma; |
1241 | struct page *page; | |
c095adbc | 1242 | int err = -EFAULT; |
2f007e74 BG |
1243 | |
1244 | vma = find_vma(mm, addr); | |
70384dc6 | 1245 | if (!vma || addr < vma->vm_start) |
742755a1 CL |
1246 | goto set_status; |
1247 | ||
2f007e74 | 1248 | page = follow_page(vma, addr, 0); |
89f5b7da LT |
1249 | |
1250 | err = PTR_ERR(page); | |
1251 | if (IS_ERR(page)) | |
1252 | goto set_status; | |
1253 | ||
742755a1 CL |
1254 | err = -ENOENT; |
1255 | /* Use PageReserved to check for zero page */ | |
62b61f61 | 1256 | if (!page || PageReserved(page) || PageKsm(page)) |
742755a1 CL |
1257 | goto set_status; |
1258 | ||
1259 | err = page_to_nid(page); | |
1260 | set_status: | |
80bba129 BG |
1261 | *status = err; |
1262 | ||
1263 | pages++; | |
1264 | status++; | |
1265 | } | |
1266 | ||
1267 | up_read(&mm->mmap_sem); | |
1268 | } | |
1269 | ||
1270 | /* | |
1271 | * Determine the nodes of a user array of pages and store it in | |
1272 | * a user array of status. | |
1273 | */ | |
1274 | static int do_pages_stat(struct mm_struct *mm, unsigned long nr_pages, | |
1275 | const void __user * __user *pages, | |
1276 | int __user *status) | |
1277 | { | |
1278 | #define DO_PAGES_STAT_CHUNK_NR 16 | |
1279 | const void __user *chunk_pages[DO_PAGES_STAT_CHUNK_NR]; | |
1280 | int chunk_status[DO_PAGES_STAT_CHUNK_NR]; | |
80bba129 | 1281 | |
87b8d1ad PA |
1282 | while (nr_pages) { |
1283 | unsigned long chunk_nr; | |
80bba129 | 1284 | |
87b8d1ad PA |
1285 | chunk_nr = nr_pages; |
1286 | if (chunk_nr > DO_PAGES_STAT_CHUNK_NR) | |
1287 | chunk_nr = DO_PAGES_STAT_CHUNK_NR; | |
1288 | ||
1289 | if (copy_from_user(chunk_pages, pages, chunk_nr * sizeof(*chunk_pages))) | |
1290 | break; | |
80bba129 BG |
1291 | |
1292 | do_pages_stat_array(mm, chunk_nr, chunk_pages, chunk_status); | |
1293 | ||
87b8d1ad PA |
1294 | if (copy_to_user(status, chunk_status, chunk_nr * sizeof(*status))) |
1295 | break; | |
742755a1 | 1296 | |
87b8d1ad PA |
1297 | pages += chunk_nr; |
1298 | status += chunk_nr; | |
1299 | nr_pages -= chunk_nr; | |
1300 | } | |
1301 | return nr_pages ? -EFAULT : 0; | |
742755a1 CL |
1302 | } |
1303 | ||
1304 | /* | |
1305 | * Move a list of pages in the address space of the currently executing | |
1306 | * process. | |
1307 | */ | |
938bb9f5 HC |
1308 | SYSCALL_DEFINE6(move_pages, pid_t, pid, unsigned long, nr_pages, |
1309 | const void __user * __user *, pages, | |
1310 | const int __user *, nodes, | |
1311 | int __user *, status, int, flags) | |
742755a1 | 1312 | { |
c69e8d9c | 1313 | const struct cred *cred = current_cred(), *tcred; |
742755a1 | 1314 | struct task_struct *task; |
742755a1 | 1315 | struct mm_struct *mm; |
5e9a0f02 | 1316 | int err; |
3268c63e | 1317 | nodemask_t task_nodes; |
742755a1 CL |
1318 | |
1319 | /* Check flags */ | |
1320 | if (flags & ~(MPOL_MF_MOVE|MPOL_MF_MOVE_ALL)) | |
1321 | return -EINVAL; | |
1322 | ||
1323 | if ((flags & MPOL_MF_MOVE_ALL) && !capable(CAP_SYS_NICE)) | |
1324 | return -EPERM; | |
1325 | ||
1326 | /* Find the mm_struct */ | |
a879bf58 | 1327 | rcu_read_lock(); |
228ebcbe | 1328 | task = pid ? find_task_by_vpid(pid) : current; |
742755a1 | 1329 | if (!task) { |
a879bf58 | 1330 | rcu_read_unlock(); |
742755a1 CL |
1331 | return -ESRCH; |
1332 | } | |
3268c63e | 1333 | get_task_struct(task); |
742755a1 CL |
1334 | |
1335 | /* | |
1336 | * Check if this process has the right to modify the specified | |
1337 | * process. The right exists if the process has administrative | |
1338 | * capabilities, superuser privileges or the same | |
1339 | * userid as the target process. | |
1340 | */ | |
c69e8d9c | 1341 | tcred = __task_cred(task); |
b38a86eb EB |
1342 | if (!uid_eq(cred->euid, tcred->suid) && !uid_eq(cred->euid, tcred->uid) && |
1343 | !uid_eq(cred->uid, tcred->suid) && !uid_eq(cred->uid, tcred->uid) && | |
742755a1 | 1344 | !capable(CAP_SYS_NICE)) { |
c69e8d9c | 1345 | rcu_read_unlock(); |
742755a1 | 1346 | err = -EPERM; |
5e9a0f02 | 1347 | goto out; |
742755a1 | 1348 | } |
c69e8d9c | 1349 | rcu_read_unlock(); |
742755a1 | 1350 | |
86c3a764 DQ |
1351 | err = security_task_movememory(task); |
1352 | if (err) | |
5e9a0f02 | 1353 | goto out; |
86c3a764 | 1354 | |
3268c63e CL |
1355 | task_nodes = cpuset_mems_allowed(task); |
1356 | mm = get_task_mm(task); | |
1357 | put_task_struct(task); | |
1358 | ||
6e8b09ea SL |
1359 | if (!mm) |
1360 | return -EINVAL; | |
1361 | ||
1362 | if (nodes) | |
1363 | err = do_pages_move(mm, task_nodes, nr_pages, pages, | |
1364 | nodes, status, flags); | |
1365 | else | |
1366 | err = do_pages_stat(mm, nr_pages, pages, status); | |
742755a1 | 1367 | |
742755a1 CL |
1368 | mmput(mm); |
1369 | return err; | |
3268c63e CL |
1370 | |
1371 | out: | |
1372 | put_task_struct(task); | |
1373 | return err; | |
742755a1 | 1374 | } |
742755a1 | 1375 | |
7b2259b3 CL |
1376 | /* |
1377 | * Call migration functions in the vma_ops that may prepare | |
1378 | * memory in a vm for migration. migration functions may perform | |
1379 | * the migration for vmas that do not have an underlying page struct. | |
1380 | */ | |
1381 | int migrate_vmas(struct mm_struct *mm, const nodemask_t *to, | |
1382 | const nodemask_t *from, unsigned long flags) | |
1383 | { | |
1384 | struct vm_area_struct *vma; | |
1385 | int err = 0; | |
1386 | ||
1001c9fb | 1387 | for (vma = mm->mmap; vma && !err; vma = vma->vm_next) { |
7b2259b3 CL |
1388 | if (vma->vm_ops && vma->vm_ops->migrate) { |
1389 | err = vma->vm_ops->migrate(vma, to, from, flags); | |
1390 | if (err) | |
1391 | break; | |
1392 | } | |
1393 | } | |
1394 | return err; | |
1395 | } | |
83d1674a | 1396 | #endif |