]>
Commit | Line | Data |
---|---|---|
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> | |
12 | * Christoph Lameter | |
13 | */ | |
14 | ||
15 | #include <linux/migrate.h> | |
16 | #include <linux/export.h> | |
17 | #include <linux/swap.h> | |
18 | #include <linux/swapops.h> | |
19 | #include <linux/pagemap.h> | |
20 | #include <linux/buffer_head.h> | |
21 | #include <linux/mm_inline.h> | |
22 | #include <linux/nsproxy.h> | |
23 | #include <linux/pagevec.h> | |
24 | #include <linux/ksm.h> | |
25 | #include <linux/rmap.h> | |
26 | #include <linux/topology.h> | |
27 | #include <linux/cpu.h> | |
28 | #include <linux/cpuset.h> | |
29 | #include <linux/writeback.h> | |
30 | #include <linux/mempolicy.h> | |
31 | #include <linux/vmalloc.h> | |
32 | #include <linux/security.h> | |
33 | #include <linux/memcontrol.h> | |
34 | #include <linux/syscalls.h> | |
35 | #include <linux/hugetlb.h> | |
36 | #include <linux/hugetlb_cgroup.h> | |
37 | #include <linux/gfp.h> | |
38 | ||
39 | #include <asm/tlbflush.h> | |
40 | ||
41 | #define CREATE_TRACE_POINTS | |
42 | #include <trace/events/migrate.h> | |
43 | ||
44 | #include "internal.h" | |
45 | ||
46 | /* | |
47 | * migrate_prep() needs to be called before we start compiling a list of pages | |
48 | * to be migrated using isolate_lru_page(). If scheduling work on other CPUs is | |
49 | * undesirable, use migrate_prep_local() | |
50 | */ | |
51 | int migrate_prep(void) | |
52 | { | |
53 | /* | |
54 | * Clear the LRU lists so pages can be isolated. | |
55 | * Note that pages may be moved off the LRU after we have | |
56 | * drained them. Those pages will fail to migrate like other | |
57 | * pages that may be busy. | |
58 | */ | |
59 | lru_add_drain_all(); | |
60 | ||
61 | return 0; | |
62 | } | |
63 | ||
64 | /* Do the necessary work of migrate_prep but not if it involves other CPUs */ | |
65 | int migrate_prep_local(void) | |
66 | { | |
67 | lru_add_drain(); | |
68 | ||
69 | return 0; | |
70 | } | |
71 | ||
72 | /* | |
73 | * Add isolated pages on the list back to the LRU under page lock | |
74 | * to avoid leaking evictable pages back onto unevictable list. | |
75 | */ | |
76 | void putback_lru_pages(struct list_head *l) | |
77 | { | |
78 | struct page *page; | |
79 | struct page *page2; | |
80 | ||
81 | list_for_each_entry_safe(page, page2, l, lru) { | |
82 | list_del(&page->lru); | |
83 | dec_zone_page_state(page, NR_ISOLATED_ANON + | |
84 | page_is_file_cache(page)); | |
85 | putback_lru_page(page); | |
86 | } | |
87 | } | |
88 | ||
89 | /* | |
90 | * Restore a potential migration pte to a working pte entry | |
91 | */ | |
92 | static int remove_migration_pte(struct page *new, struct vm_area_struct *vma, | |
93 | unsigned long addr, void *old) | |
94 | { | |
95 | struct mm_struct *mm = vma->vm_mm; | |
96 | swp_entry_t entry; | |
97 | pgd_t *pgd; | |
98 | pud_t *pud; | |
99 | pmd_t *pmd; | |
100 | pte_t *ptep, pte; | |
101 | spinlock_t *ptl; | |
102 | ||
103 | if (unlikely(PageHuge(new))) { | |
104 | ptep = huge_pte_offset(mm, addr); | |
105 | if (!ptep) | |
106 | goto out; | |
107 | ptl = &mm->page_table_lock; | |
108 | } else { | |
109 | pgd = pgd_offset(mm, addr); | |
110 | if (!pgd_present(*pgd)) | |
111 | goto out; | |
112 | ||
113 | pud = pud_offset(pgd, addr); | |
114 | if (!pud_present(*pud)) | |
115 | goto out; | |
116 | ||
117 | pmd = pmd_offset(pud, addr); | |
118 | if (pmd_trans_huge(*pmd)) | |
119 | goto out; | |
120 | if (!pmd_present(*pmd)) | |
121 | goto out; | |
122 | ||
123 | ptep = pte_offset_map(pmd, addr); | |
124 | ||
125 | /* | |
126 | * Peek to check is_swap_pte() before taking ptlock? No, we | |
127 | * can race mremap's move_ptes(), which skips anon_vma lock. | |
128 | */ | |
129 | ||
130 | ptl = pte_lockptr(mm, pmd); | |
131 | } | |
132 | ||
133 | spin_lock(ptl); | |
134 | pte = *ptep; | |
135 | if (!is_swap_pte(pte)) | |
136 | goto unlock; | |
137 | ||
138 | entry = pte_to_swp_entry(pte); | |
139 | ||
140 | if (!is_migration_entry(entry) || | |
141 | migration_entry_to_page(entry) != old) | |
142 | goto unlock; | |
143 | ||
144 | get_page(new); | |
145 | pte = pte_mkold(mk_pte(new, vma->vm_page_prot)); | |
146 | if (is_write_migration_entry(entry)) | |
147 | pte = pte_mkwrite(pte); | |
148 | #ifdef CONFIG_HUGETLB_PAGE | |
149 | if (PageHuge(new)) | |
150 | pte = pte_mkhuge(pte); | |
151 | #endif | |
152 | flush_cache_page(vma, addr, pte_pfn(pte)); | |
153 | set_pte_at(mm, addr, ptep, pte); | |
154 | ||
155 | if (PageHuge(new)) { | |
156 | if (PageAnon(new)) | |
157 | hugepage_add_anon_rmap(new, vma, addr); | |
158 | else | |
159 | page_dup_rmap(new); | |
160 | } else if (PageAnon(new)) | |
161 | page_add_anon_rmap(new, vma, addr); | |
162 | else | |
163 | page_add_file_rmap(new); | |
164 | ||
165 | /* No need to invalidate - it was non-present before */ | |
166 | update_mmu_cache(vma, addr, ptep); | |
167 | unlock: | |
168 | pte_unmap_unlock(ptep, ptl); | |
169 | out: | |
170 | return SWAP_AGAIN; | |
171 | } | |
172 | ||
173 | /* | |
174 | * Get rid of all migration entries and replace them by | |
175 | * references to the indicated page. | |
176 | */ | |
177 | static void remove_migration_ptes(struct page *old, struct page *new) | |
178 | { | |
179 | rmap_walk(new, remove_migration_pte, old); | |
180 | } | |
181 | ||
182 | /* | |
183 | * Something used the pte of a page under migration. We need to | |
184 | * get to the page and wait until migration is finished. | |
185 | * When we return from this function the fault will be retried. | |
186 | */ | |
187 | void migration_entry_wait(struct mm_struct *mm, pmd_t *pmd, | |
188 | unsigned long address) | |
189 | { | |
190 | pte_t *ptep, pte; | |
191 | spinlock_t *ptl; | |
192 | swp_entry_t entry; | |
193 | struct page *page; | |
194 | ||
195 | ptep = pte_offset_map_lock(mm, pmd, address, &ptl); | |
196 | pte = *ptep; | |
197 | if (!is_swap_pte(pte)) | |
198 | goto out; | |
199 | ||
200 | entry = pte_to_swp_entry(pte); | |
201 | if (!is_migration_entry(entry)) | |
202 | goto out; | |
203 | ||
204 | page = migration_entry_to_page(entry); | |
205 | ||
206 | /* | |
207 | * Once radix-tree replacement of page migration started, page_count | |
208 | * *must* be zero. And, we don't want to call wait_on_page_locked() | |
209 | * against a page without get_page(). | |
210 | * So, we use get_page_unless_zero(), here. Even failed, page fault | |
211 | * will occur again. | |
212 | */ | |
213 | if (!get_page_unless_zero(page)) | |
214 | goto out; | |
215 | pte_unmap_unlock(ptep, ptl); | |
216 | wait_on_page_locked(page); | |
217 | put_page(page); | |
218 | return; | |
219 | out: | |
220 | pte_unmap_unlock(ptep, ptl); | |
221 | } | |
222 | ||
223 | #ifdef CONFIG_BLOCK | |
224 | /* Returns true if all buffers are successfully locked */ | |
225 | static bool buffer_migrate_lock_buffers(struct buffer_head *head, | |
226 | enum migrate_mode mode) | |
227 | { | |
228 | struct buffer_head *bh = head; | |
229 | ||
230 | /* Simple case, sync compaction */ | |
231 | if (mode != MIGRATE_ASYNC) { | |
232 | do { | |
233 | get_bh(bh); | |
234 | lock_buffer(bh); | |
235 | bh = bh->b_this_page; | |
236 | ||
237 | } while (bh != head); | |
238 | ||
239 | return true; | |
240 | } | |
241 | ||
242 | /* async case, we cannot block on lock_buffer so use trylock_buffer */ | |
243 | do { | |
244 | get_bh(bh); | |
245 | if (!trylock_buffer(bh)) { | |
246 | /* | |
247 | * We failed to lock the buffer and cannot stall in | |
248 | * async migration. Release the taken locks | |
249 | */ | |
250 | struct buffer_head *failed_bh = bh; | |
251 | put_bh(failed_bh); | |
252 | bh = head; | |
253 | while (bh != failed_bh) { | |
254 | unlock_buffer(bh); | |
255 | put_bh(bh); | |
256 | bh = bh->b_this_page; | |
257 | } | |
258 | return false; | |
259 | } | |
260 | ||
261 | bh = bh->b_this_page; | |
262 | } while (bh != head); | |
263 | return true; | |
264 | } | |
265 | #else | |
266 | static inline bool buffer_migrate_lock_buffers(struct buffer_head *head, | |
267 | enum migrate_mode mode) | |
268 | { | |
269 | return true; | |
270 | } | |
271 | #endif /* CONFIG_BLOCK */ | |
272 | ||
273 | /* | |
274 | * Replace the page in the mapping. | |
275 | * | |
276 | * The number of remaining references must be: | |
277 | * 1 for anonymous pages without a mapping | |
278 | * 2 for pages with a mapping | |
279 | * 3 for pages with a mapping and PagePrivate/PagePrivate2 set. | |
280 | */ | |
281 | static int migrate_page_move_mapping(struct address_space *mapping, | |
282 | struct page *newpage, struct page *page, | |
283 | struct buffer_head *head, enum migrate_mode mode) | |
284 | { | |
285 | int expected_count; | |
286 | void **pslot; | |
287 | ||
288 | if (!mapping) { | |
289 | /* Anonymous page without mapping */ | |
290 | if (page_count(page) != 1) | |
291 | return -EAGAIN; | |
292 | return 0; | |
293 | } | |
294 | ||
295 | spin_lock_irq(&mapping->tree_lock); | |
296 | ||
297 | pslot = radix_tree_lookup_slot(&mapping->page_tree, | |
298 | page_index(page)); | |
299 | ||
300 | expected_count = 2 + page_has_private(page); | |
301 | if (page_count(page) != expected_count || | |
302 | radix_tree_deref_slot_protected(pslot, &mapping->tree_lock) != page) { | |
303 | spin_unlock_irq(&mapping->tree_lock); | |
304 | return -EAGAIN; | |
305 | } | |
306 | ||
307 | if (!page_freeze_refs(page, expected_count)) { | |
308 | spin_unlock_irq(&mapping->tree_lock); | |
309 | return -EAGAIN; | |
310 | } | |
311 | ||
312 | /* | |
313 | * In the async migration case of moving a page with buffers, lock the | |
314 | * buffers using trylock before the mapping is moved. If the mapping | |
315 | * was moved, we later failed to lock the buffers and could not move | |
316 | * the mapping back due to an elevated page count, we would have to | |
317 | * block waiting on other references to be dropped. | |
318 | */ | |
319 | if (mode == MIGRATE_ASYNC && head && | |
320 | !buffer_migrate_lock_buffers(head, mode)) { | |
321 | page_unfreeze_refs(page, expected_count); | |
322 | spin_unlock_irq(&mapping->tree_lock); | |
323 | return -EAGAIN; | |
324 | } | |
325 | ||
326 | /* | |
327 | * Now we know that no one else is looking at the page. | |
328 | */ | |
329 | get_page(newpage); /* add cache reference */ | |
330 | if (PageSwapCache(page)) { | |
331 | SetPageSwapCache(newpage); | |
332 | set_page_private(newpage, page_private(page)); | |
333 | } | |
334 | ||
335 | radix_tree_replace_slot(pslot, newpage); | |
336 | ||
337 | /* | |
338 | * Drop cache reference from old page by unfreezing | |
339 | * to one less reference. | |
340 | * We know this isn't the last reference. | |
341 | */ | |
342 | page_unfreeze_refs(page, expected_count - 1); | |
343 | ||
344 | /* | |
345 | * If moved to a different zone then also account | |
346 | * the page for that zone. Other VM counters will be | |
347 | * taken care of when we establish references to the | |
348 | * new page and drop references to the old page. | |
349 | * | |
350 | * Note that anonymous pages are accounted for | |
351 | * via NR_FILE_PAGES and NR_ANON_PAGES if they | |
352 | * are mapped to swap space. | |
353 | */ | |
354 | __dec_zone_page_state(page, NR_FILE_PAGES); | |
355 | __inc_zone_page_state(newpage, NR_FILE_PAGES); | |
356 | if (!PageSwapCache(page) && PageSwapBacked(page)) { | |
357 | __dec_zone_page_state(page, NR_SHMEM); | |
358 | __inc_zone_page_state(newpage, NR_SHMEM); | |
359 | } | |
360 | spin_unlock_irq(&mapping->tree_lock); | |
361 | ||
362 | return 0; | |
363 | } | |
364 | ||
365 | /* | |
366 | * The expected number of remaining references is the same as that | |
367 | * of migrate_page_move_mapping(). | |
368 | */ | |
369 | int migrate_huge_page_move_mapping(struct address_space *mapping, | |
370 | struct page *newpage, struct page *page) | |
371 | { | |
372 | int expected_count; | |
373 | void **pslot; | |
374 | ||
375 | if (!mapping) { | |
376 | if (page_count(page) != 1) | |
377 | return -EAGAIN; | |
378 | return 0; | |
379 | } | |
380 | ||
381 | spin_lock_irq(&mapping->tree_lock); | |
382 | ||
383 | pslot = radix_tree_lookup_slot(&mapping->page_tree, | |
384 | page_index(page)); | |
385 | ||
386 | expected_count = 2 + page_has_private(page); | |
387 | if (page_count(page) != expected_count || | |
388 | radix_tree_deref_slot_protected(pslot, &mapping->tree_lock) != page) { | |
389 | spin_unlock_irq(&mapping->tree_lock); | |
390 | return -EAGAIN; | |
391 | } | |
392 | ||
393 | if (!page_freeze_refs(page, expected_count)) { | |
394 | spin_unlock_irq(&mapping->tree_lock); | |
395 | return -EAGAIN; | |
396 | } | |
397 | ||
398 | get_page(newpage); | |
399 | ||
400 | radix_tree_replace_slot(pslot, newpage); | |
401 | ||
402 | page_unfreeze_refs(page, expected_count - 1); | |
403 | ||
404 | spin_unlock_irq(&mapping->tree_lock); | |
405 | return 0; | |
406 | } | |
407 | ||
408 | /* | |
409 | * Copy the page to its new location | |
410 | */ | |
411 | void migrate_page_copy(struct page *newpage, struct page *page) | |
412 | { | |
413 | if (PageHuge(page)) | |
414 | copy_huge_page(newpage, page); | |
415 | else | |
416 | copy_highpage(newpage, page); | |
417 | ||
418 | if (PageError(page)) | |
419 | SetPageError(newpage); | |
420 | if (PageReferenced(page)) | |
421 | SetPageReferenced(newpage); | |
422 | if (PageUptodate(page)) | |
423 | SetPageUptodate(newpage); | |
424 | if (TestClearPageActive(page)) { | |
425 | VM_BUG_ON(PageUnevictable(page)); | |
426 | SetPageActive(newpage); | |
427 | } else if (TestClearPageUnevictable(page)) | |
428 | SetPageUnevictable(newpage); | |
429 | if (PageChecked(page)) | |
430 | SetPageChecked(newpage); | |
431 | if (PageMappedToDisk(page)) | |
432 | SetPageMappedToDisk(newpage); | |
433 | ||
434 | if (PageDirty(page)) { | |
435 | clear_page_dirty_for_io(page); | |
436 | /* | |
437 | * Want to mark the page and the radix tree as dirty, and | |
438 | * redo the accounting that clear_page_dirty_for_io undid, | |
439 | * but we can't use set_page_dirty because that function | |
440 | * is actually a signal that all of the page has become dirty. | |
441 | * Whereas only part of our page may be dirty. | |
442 | */ | |
443 | if (PageSwapBacked(page)) | |
444 | SetPageDirty(newpage); | |
445 | else | |
446 | __set_page_dirty_nobuffers(newpage); | |
447 | } | |
448 | ||
449 | mlock_migrate_page(newpage, page); | |
450 | ksm_migrate_page(newpage, page); | |
451 | ||
452 | ClearPageSwapCache(page); | |
453 | ClearPagePrivate(page); | |
454 | set_page_private(page, 0); | |
455 | ||
456 | /* | |
457 | * If any waiters have accumulated on the new page then | |
458 | * wake them up. | |
459 | */ | |
460 | if (PageWriteback(newpage)) | |
461 | end_page_writeback(newpage); | |
462 | } | |
463 | ||
464 | /************************************************************ | |
465 | * Migration functions | |
466 | ***********************************************************/ | |
467 | ||
468 | /* Always fail migration. Used for mappings that are not movable */ | |
469 | int fail_migrate_page(struct address_space *mapping, | |
470 | struct page *newpage, struct page *page) | |
471 | { | |
472 | return -EIO; | |
473 | } | |
474 | EXPORT_SYMBOL(fail_migrate_page); | |
475 | ||
476 | /* | |
477 | * Common logic to directly migrate a single page suitable for | |
478 | * pages that do not use PagePrivate/PagePrivate2. | |
479 | * | |
480 | * Pages are locked upon entry and exit. | |
481 | */ | |
482 | int migrate_page(struct address_space *mapping, | |
483 | struct page *newpage, struct page *page, | |
484 | enum migrate_mode mode) | |
485 | { | |
486 | int rc; | |
487 | ||
488 | BUG_ON(PageWriteback(page)); /* Writeback must be complete */ | |
489 | ||
490 | rc = migrate_page_move_mapping(mapping, newpage, page, NULL, mode); | |
491 | ||
492 | if (rc) | |
493 | return rc; | |
494 | ||
495 | migrate_page_copy(newpage, page); | |
496 | return 0; | |
497 | } | |
498 | EXPORT_SYMBOL(migrate_page); | |
499 | ||
500 | #ifdef CONFIG_BLOCK | |
501 | /* | |
502 | * Migration function for pages with buffers. This function can only be used | |
503 | * if the underlying filesystem guarantees that no other references to "page" | |
504 | * exist. | |
505 | */ | |
506 | int buffer_migrate_page(struct address_space *mapping, | |
507 | struct page *newpage, struct page *page, enum migrate_mode mode) | |
508 | { | |
509 | struct buffer_head *bh, *head; | |
510 | int rc; | |
511 | ||
512 | if (!page_has_buffers(page)) | |
513 | return migrate_page(mapping, newpage, page, mode); | |
514 | ||
515 | head = page_buffers(page); | |
516 | ||
517 | rc = migrate_page_move_mapping(mapping, newpage, page, head, mode); | |
518 | ||
519 | if (rc) | |
520 | return rc; | |
521 | ||
522 | /* | |
523 | * In the async case, migrate_page_move_mapping locked the buffers | |
524 | * with an IRQ-safe spinlock held. In the sync case, the buffers | |
525 | * need to be locked now | |
526 | */ | |
527 | if (mode != MIGRATE_ASYNC) | |
528 | BUG_ON(!buffer_migrate_lock_buffers(head, mode)); | |
529 | ||
530 | ClearPagePrivate(page); | |
531 | set_page_private(newpage, page_private(page)); | |
532 | set_page_private(page, 0); | |
533 | put_page(page); | |
534 | get_page(newpage); | |
535 | ||
536 | bh = head; | |
537 | do { | |
538 | set_bh_page(bh, newpage, bh_offset(bh)); | |
539 | bh = bh->b_this_page; | |
540 | ||
541 | } while (bh != head); | |
542 | ||
543 | SetPagePrivate(newpage); | |
544 | ||
545 | migrate_page_copy(newpage, page); | |
546 | ||
547 | bh = head; | |
548 | do { | |
549 | unlock_buffer(bh); | |
550 | put_bh(bh); | |
551 | bh = bh->b_this_page; | |
552 | ||
553 | } while (bh != head); | |
554 | ||
555 | return 0; | |
556 | } | |
557 | EXPORT_SYMBOL(buffer_migrate_page); | |
558 | #endif | |
559 | ||
560 | /* | |
561 | * Writeback a page to clean the dirty state | |
562 | */ | |
563 | static int writeout(struct address_space *mapping, struct page *page) | |
564 | { | |
565 | struct writeback_control wbc = { | |
566 | .sync_mode = WB_SYNC_NONE, | |
567 | .nr_to_write = 1, | |
568 | .range_start = 0, | |
569 | .range_end = LLONG_MAX, | |
570 | .for_reclaim = 1 | |
571 | }; | |
572 | int rc; | |
573 | ||
574 | if (!mapping->a_ops->writepage) | |
575 | /* No write method for the address space */ | |
576 | return -EINVAL; | |
577 | ||
578 | if (!clear_page_dirty_for_io(page)) | |
579 | /* Someone else already triggered a write */ | |
580 | return -EAGAIN; | |
581 | ||
582 | /* | |
583 | * A dirty page may imply that the underlying filesystem has | |
584 | * the page on some queue. So the page must be clean for | |
585 | * migration. Writeout may mean we loose the lock and the | |
586 | * page state is no longer what we checked for earlier. | |
587 | * At this point we know that the migration attempt cannot | |
588 | * be successful. | |
589 | */ | |
590 | remove_migration_ptes(page, page); | |
591 | ||
592 | rc = mapping->a_ops->writepage(page, &wbc); | |
593 | ||
594 | if (rc != AOP_WRITEPAGE_ACTIVATE) | |
595 | /* unlocked. Relock */ | |
596 | lock_page(page); | |
597 | ||
598 | return (rc < 0) ? -EIO : -EAGAIN; | |
599 | } | |
600 | ||
601 | /* | |
602 | * Default handling if a filesystem does not provide a migration function. | |
603 | */ | |
604 | static int fallback_migrate_page(struct address_space *mapping, | |
605 | struct page *newpage, struct page *page, enum migrate_mode mode) | |
606 | { | |
607 | if (PageDirty(page)) { | |
608 | /* Only writeback pages in full synchronous migration */ | |
609 | if (mode != MIGRATE_SYNC) | |
610 | return -EBUSY; | |
611 | return writeout(mapping, page); | |
612 | } | |
613 | ||
614 | /* | |
615 | * Buffers may be managed in a filesystem specific way. | |
616 | * We must have no buffers or drop them. | |
617 | */ | |
618 | if (page_has_private(page) && | |
619 | !try_to_release_page(page, GFP_KERNEL)) | |
620 | return -EAGAIN; | |
621 | ||
622 | return migrate_page(mapping, newpage, page, mode); | |
623 | } | |
624 | ||
625 | /* | |
626 | * Move a page to a newly allocated page | |
627 | * The page is locked and all ptes have been successfully removed. | |
628 | * | |
629 | * The new page will have replaced the old page if this function | |
630 | * is successful. | |
631 | * | |
632 | * Return value: | |
633 | * < 0 - error code | |
634 | * == 0 - success | |
635 | */ | |
636 | static int move_to_new_page(struct page *newpage, struct page *page, | |
637 | int remap_swapcache, enum migrate_mode mode) | |
638 | { | |
639 | struct address_space *mapping; | |
640 | int rc; | |
641 | ||
642 | /* | |
643 | * Block others from accessing the page when we get around to | |
644 | * establishing additional references. We are the only one | |
645 | * holding a reference to the new page at this point. | |
646 | */ | |
647 | if (!trylock_page(newpage)) | |
648 | BUG(); | |
649 | ||
650 | /* Prepare mapping for the new page.*/ | |
651 | newpage->index = page->index; | |
652 | newpage->mapping = page->mapping; | |
653 | if (PageSwapBacked(page)) | |
654 | SetPageSwapBacked(newpage); | |
655 | ||
656 | mapping = page_mapping(page); | |
657 | if (!mapping) | |
658 | rc = migrate_page(mapping, newpage, page, mode); | |
659 | else if (mapping->a_ops->migratepage) | |
660 | /* | |
661 | * Most pages have a mapping and most filesystems provide a | |
662 | * migratepage callback. Anonymous pages are part of swap | |
663 | * space which also has its own migratepage callback. This | |
664 | * is the most common path for page migration. | |
665 | */ | |
666 | rc = mapping->a_ops->migratepage(mapping, | |
667 | newpage, page, mode); | |
668 | else | |
669 | rc = fallback_migrate_page(mapping, newpage, page, mode); | |
670 | ||
671 | if (rc) { | |
672 | newpage->mapping = NULL; | |
673 | } else { | |
674 | if (remap_swapcache) | |
675 | remove_migration_ptes(page, newpage); | |
676 | page->mapping = NULL; | |
677 | } | |
678 | ||
679 | unlock_page(newpage); | |
680 | ||
681 | return rc; | |
682 | } | |
683 | ||
684 | static int __unmap_and_move(struct page *page, struct page *newpage, | |
685 | int force, bool offlining, enum migrate_mode mode) | |
686 | { | |
687 | int rc = -EAGAIN; | |
688 | int remap_swapcache = 1; | |
689 | struct mem_cgroup *mem; | |
690 | struct anon_vma *anon_vma = NULL; | |
691 | ||
692 | if (!trylock_page(page)) { | |
693 | if (!force || mode == MIGRATE_ASYNC) | |
694 | goto out; | |
695 | ||
696 | /* | |
697 | * It's not safe for direct compaction to call lock_page. | |
698 | * For example, during page readahead pages are added locked | |
699 | * to the LRU. Later, when the IO completes the pages are | |
700 | * marked uptodate and unlocked. However, the queueing | |
701 | * could be merging multiple pages for one bio (e.g. | |
702 | * mpage_readpages). If an allocation happens for the | |
703 | * second or third page, the process can end up locking | |
704 | * the same page twice and deadlocking. Rather than | |
705 | * trying to be clever about what pages can be locked, | |
706 | * avoid the use of lock_page for direct compaction | |
707 | * altogether. | |
708 | */ | |
709 | if (current->flags & PF_MEMALLOC) | |
710 | goto out; | |
711 | ||
712 | lock_page(page); | |
713 | } | |
714 | ||
715 | /* | |
716 | * Only memory hotplug's offline_pages() caller has locked out KSM, | |
717 | * and can safely migrate a KSM page. The other cases have skipped | |
718 | * PageKsm along with PageReserved - but it is only now when we have | |
719 | * the page lock that we can be certain it will not go KSM beneath us | |
720 | * (KSM will not upgrade a page from PageAnon to PageKsm when it sees | |
721 | * its pagecount raised, but only here do we take the page lock which | |
722 | * serializes that). | |
723 | */ | |
724 | if (PageKsm(page) && !offlining) { | |
725 | rc = -EBUSY; | |
726 | goto unlock; | |
727 | } | |
728 | ||
729 | /* charge against new page */ | |
730 | mem_cgroup_prepare_migration(page, newpage, &mem); | |
731 | ||
732 | if (PageWriteback(page)) { | |
733 | /* | |
734 | * Only in the case of a full syncronous migration is it | |
735 | * necessary to wait for PageWriteback. In the async case, | |
736 | * the retry loop is too short and in the sync-light case, | |
737 | * the overhead of stalling is too much | |
738 | */ | |
739 | if (mode != MIGRATE_SYNC) { | |
740 | rc = -EBUSY; | |
741 | goto uncharge; | |
742 | } | |
743 | if (!force) | |
744 | goto uncharge; | |
745 | wait_on_page_writeback(page); | |
746 | } | |
747 | /* | |
748 | * By try_to_unmap(), page->mapcount goes down to 0 here. In this case, | |
749 | * we cannot notice that anon_vma is freed while we migrates a page. | |
750 | * This get_anon_vma() delays freeing anon_vma pointer until the end | |
751 | * of migration. File cache pages are no problem because of page_lock() | |
752 | * File Caches may use write_page() or lock_page() in migration, then, | |
753 | * just care Anon page here. | |
754 | */ | |
755 | if (PageAnon(page)) { | |
756 | /* | |
757 | * Only page_lock_anon_vma() understands the subtleties of | |
758 | * getting a hold on an anon_vma from outside one of its mms. | |
759 | */ | |
760 | anon_vma = page_get_anon_vma(page); | |
761 | if (anon_vma) { | |
762 | /* | |
763 | * Anon page | |
764 | */ | |
765 | } else if (PageSwapCache(page)) { | |
766 | /* | |
767 | * We cannot be sure that the anon_vma of an unmapped | |
768 | * swapcache page is safe to use because we don't | |
769 | * know in advance if the VMA that this page belonged | |
770 | * to still exists. If the VMA and others sharing the | |
771 | * data have been freed, then the anon_vma could | |
772 | * already be invalid. | |
773 | * | |
774 | * To avoid this possibility, swapcache pages get | |
775 | * migrated but are not remapped when migration | |
776 | * completes | |
777 | */ | |
778 | remap_swapcache = 0; | |
779 | } else { | |
780 | goto uncharge; | |
781 | } | |
782 | } | |
783 | ||
784 | /* | |
785 | * Corner case handling: | |
786 | * 1. When a new swap-cache page is read into, it is added to the LRU | |
787 | * and treated as swapcache but it has no rmap yet. | |
788 | * Calling try_to_unmap() against a page->mapping==NULL page will | |
789 | * trigger a BUG. So handle it here. | |
790 | * 2. An orphaned page (see truncate_complete_page) might have | |
791 | * fs-private metadata. The page can be picked up due to memory | |
792 | * offlining. Everywhere else except page reclaim, the page is | |
793 | * invisible to the vm, so the page can not be migrated. So try to | |
794 | * free the metadata, so the page can be freed. | |
795 | */ | |
796 | if (!page->mapping) { | |
797 | VM_BUG_ON(PageAnon(page)); | |
798 | if (page_has_private(page)) { | |
799 | try_to_free_buffers(page); | |
800 | goto uncharge; | |
801 | } | |
802 | goto skip_unmap; | |
803 | } | |
804 | ||
805 | /* Establish migration ptes or remove ptes */ | |
806 | try_to_unmap(page, TTU_MIGRATION|TTU_IGNORE_MLOCK|TTU_IGNORE_ACCESS); | |
807 | ||
808 | skip_unmap: | |
809 | if (!page_mapped(page)) | |
810 | rc = move_to_new_page(newpage, page, remap_swapcache, mode); | |
811 | ||
812 | if (rc && remap_swapcache) | |
813 | remove_migration_ptes(page, page); | |
814 | ||
815 | /* Drop an anon_vma reference if we took one */ | |
816 | if (anon_vma) | |
817 | put_anon_vma(anon_vma); | |
818 | ||
819 | uncharge: | |
820 | mem_cgroup_end_migration(mem, page, newpage, rc == 0); | |
821 | unlock: | |
822 | unlock_page(page); | |
823 | out: | |
824 | return rc; | |
825 | } | |
826 | ||
827 | /* | |
828 | * Obtain the lock on page, remove all ptes and migrate the page | |
829 | * to the newly allocated page in newpage. | |
830 | */ | |
831 | static int unmap_and_move(new_page_t get_new_page, unsigned long private, | |
832 | struct page *page, int force, bool offlining, | |
833 | enum migrate_mode mode) | |
834 | { | |
835 | int rc = 0; | |
836 | int *result = NULL; | |
837 | struct page *newpage = get_new_page(page, private, &result); | |
838 | ||
839 | if (!newpage) | |
840 | return -ENOMEM; | |
841 | ||
842 | if (page_count(page) == 1) { | |
843 | /* page was freed from under us. So we are done. */ | |
844 | goto out; | |
845 | } | |
846 | ||
847 | if (unlikely(PageTransHuge(page))) | |
848 | if (unlikely(split_huge_page(page))) | |
849 | goto out; | |
850 | ||
851 | rc = __unmap_and_move(page, newpage, force, offlining, mode); | |
852 | out: | |
853 | if (rc != -EAGAIN) { | |
854 | /* | |
855 | * A page that has been migrated has all references | |
856 | * removed and will be freed. A page that has not been | |
857 | * migrated will have kepts its references and be | |
858 | * restored. | |
859 | */ | |
860 | list_del(&page->lru); | |
861 | dec_zone_page_state(page, NR_ISOLATED_ANON + | |
862 | page_is_file_cache(page)); | |
863 | putback_lru_page(page); | |
864 | } | |
865 | /* | |
866 | * Move the new page to the LRU. If migration was not successful | |
867 | * then this will free the page. | |
868 | */ | |
869 | putback_lru_page(newpage); | |
870 | if (result) { | |
871 | if (rc) | |
872 | *result = rc; | |
873 | else | |
874 | *result = page_to_nid(newpage); | |
875 | } | |
876 | return rc; | |
877 | } | |
878 | ||
879 | /* | |
880 | * Counterpart of unmap_and_move_page() for hugepage migration. | |
881 | * | |
882 | * This function doesn't wait the completion of hugepage I/O | |
883 | * because there is no race between I/O and migration for hugepage. | |
884 | * Note that currently hugepage I/O occurs only in direct I/O | |
885 | * where no lock is held and PG_writeback is irrelevant, | |
886 | * and writeback status of all subpages are counted in the reference | |
887 | * count of the head page (i.e. if all subpages of a 2MB hugepage are | |
888 | * under direct I/O, the reference of the head page is 512 and a bit more.) | |
889 | * This means that when we try to migrate hugepage whose subpages are | |
890 | * doing direct I/O, some references remain after try_to_unmap() and | |
891 | * hugepage migration fails without data corruption. | |
892 | * | |
893 | * There is also no race when direct I/O is issued on the page under migration, | |
894 | * because then pte is replaced with migration swap entry and direct I/O code | |
895 | * will wait in the page fault for migration to complete. | |
896 | */ | |
897 | static int unmap_and_move_huge_page(new_page_t get_new_page, | |
898 | unsigned long private, struct page *hpage, | |
899 | int force, bool offlining, | |
900 | enum migrate_mode mode) | |
901 | { | |
902 | int rc = 0; | |
903 | int *result = NULL; | |
904 | struct page *new_hpage = get_new_page(hpage, private, &result); | |
905 | struct anon_vma *anon_vma = NULL; | |
906 | ||
907 | if (!new_hpage) | |
908 | return -ENOMEM; | |
909 | ||
910 | rc = -EAGAIN; | |
911 | ||
912 | if (!trylock_page(hpage)) { | |
913 | if (!force || mode != MIGRATE_SYNC) | |
914 | goto out; | |
915 | lock_page(hpage); | |
916 | } | |
917 | ||
918 | if (PageAnon(hpage)) | |
919 | anon_vma = page_get_anon_vma(hpage); | |
920 | ||
921 | try_to_unmap(hpage, TTU_MIGRATION|TTU_IGNORE_MLOCK|TTU_IGNORE_ACCESS); | |
922 | ||
923 | if (!page_mapped(hpage)) | |
924 | rc = move_to_new_page(new_hpage, hpage, 1, mode); | |
925 | ||
926 | if (rc) | |
927 | remove_migration_ptes(hpage, hpage); | |
928 | ||
929 | if (anon_vma) | |
930 | put_anon_vma(anon_vma); | |
931 | ||
932 | if (!rc) | |
933 | hugetlb_cgroup_migrate(hpage, new_hpage); | |
934 | ||
935 | unlock_page(hpage); | |
936 | out: | |
937 | put_page(new_hpage); | |
938 | if (result) { | |
939 | if (rc) | |
940 | *result = rc; | |
941 | else | |
942 | *result = page_to_nid(new_hpage); | |
943 | } | |
944 | return rc; | |
945 | } | |
946 | ||
947 | /* | |
948 | * migrate_pages | |
949 | * | |
950 | * The function takes one list of pages to migrate and a function | |
951 | * that determines from the page to be migrated and the private data | |
952 | * the target of the move and allocates the page. | |
953 | * | |
954 | * The function returns after 10 attempts or if no pages | |
955 | * are movable anymore because to has become empty | |
956 | * or no retryable pages exist anymore. | |
957 | * Caller should call putback_lru_pages to return pages to the LRU | |
958 | * or free list only if ret != 0. | |
959 | * | |
960 | * Return: Number of pages not migrated or error code. | |
961 | */ | |
962 | int migrate_pages(struct list_head *from, | |
963 | new_page_t get_new_page, unsigned long private, bool offlining, | |
964 | enum migrate_mode mode, int reason) | |
965 | { | |
966 | int retry = 1; | |
967 | int nr_failed = 0; | |
968 | int nr_succeeded = 0; | |
969 | int pass = 0; | |
970 | struct page *page; | |
971 | struct page *page2; | |
972 | int swapwrite = current->flags & PF_SWAPWRITE; | |
973 | int rc; | |
974 | ||
975 | if (!swapwrite) | |
976 | current->flags |= PF_SWAPWRITE; | |
977 | ||
978 | for(pass = 0; pass < 10 && retry; pass++) { | |
979 | retry = 0; | |
980 | ||
981 | list_for_each_entry_safe(page, page2, from, lru) { | |
982 | cond_resched(); | |
983 | ||
984 | rc = unmap_and_move(get_new_page, private, | |
985 | page, pass > 2, offlining, | |
986 | mode); | |
987 | ||
988 | switch(rc) { | |
989 | case -ENOMEM: | |
990 | goto out; | |
991 | case -EAGAIN: | |
992 | retry++; | |
993 | break; | |
994 | case 0: | |
995 | nr_succeeded++; | |
996 | break; | |
997 | default: | |
998 | /* Permanent failure */ | |
999 | nr_failed++; | |
1000 | break; | |
1001 | } | |
1002 | } | |
1003 | } | |
1004 | rc = 0; | |
1005 | out: | |
1006 | if (nr_succeeded) | |
1007 | count_vm_events(PGMIGRATE_SUCCESS, nr_succeeded); | |
1008 | if (nr_failed) | |
1009 | count_vm_events(PGMIGRATE_FAIL, nr_failed); | |
1010 | trace_mm_migrate_pages(nr_succeeded, nr_failed, mode, reason); | |
1011 | ||
1012 | if (!swapwrite) | |
1013 | current->flags &= ~PF_SWAPWRITE; | |
1014 | ||
1015 | if (rc) | |
1016 | return rc; | |
1017 | ||
1018 | return nr_failed + retry; | |
1019 | } | |
1020 | ||
1021 | int migrate_huge_page(struct page *hpage, new_page_t get_new_page, | |
1022 | unsigned long private, bool offlining, | |
1023 | enum migrate_mode mode) | |
1024 | { | |
1025 | int pass, rc; | |
1026 | ||
1027 | for (pass = 0; pass < 10; pass++) { | |
1028 | rc = unmap_and_move_huge_page(get_new_page, | |
1029 | private, hpage, pass > 2, offlining, | |
1030 | mode); | |
1031 | switch (rc) { | |
1032 | case -ENOMEM: | |
1033 | goto out; | |
1034 | case -EAGAIN: | |
1035 | /* try again */ | |
1036 | cond_resched(); | |
1037 | break; | |
1038 | case 0: | |
1039 | goto out; | |
1040 | default: | |
1041 | rc = -EIO; | |
1042 | goto out; | |
1043 | } | |
1044 | } | |
1045 | out: | |
1046 | return rc; | |
1047 | } | |
1048 | ||
1049 | #ifdef CONFIG_NUMA | |
1050 | /* | |
1051 | * Move a list of individual pages | |
1052 | */ | |
1053 | struct page_to_node { | |
1054 | unsigned long addr; | |
1055 | struct page *page; | |
1056 | int node; | |
1057 | int status; | |
1058 | }; | |
1059 | ||
1060 | static struct page *new_page_node(struct page *p, unsigned long private, | |
1061 | int **result) | |
1062 | { | |
1063 | struct page_to_node *pm = (struct page_to_node *)private; | |
1064 | ||
1065 | while (pm->node != MAX_NUMNODES && pm->page != p) | |
1066 | pm++; | |
1067 | ||
1068 | if (pm->node == MAX_NUMNODES) | |
1069 | return NULL; | |
1070 | ||
1071 | *result = &pm->status; | |
1072 | ||
1073 | return alloc_pages_exact_node(pm->node, | |
1074 | GFP_HIGHUSER_MOVABLE | GFP_THISNODE, 0); | |
1075 | } | |
1076 | ||
1077 | /* | |
1078 | * Move a set of pages as indicated in the pm array. The addr | |
1079 | * field must be set to the virtual address of the page to be moved | |
1080 | * and the node number must contain a valid target node. | |
1081 | * The pm array ends with node = MAX_NUMNODES. | |
1082 | */ | |
1083 | static int do_move_page_to_node_array(struct mm_struct *mm, | |
1084 | struct page_to_node *pm, | |
1085 | int migrate_all) | |
1086 | { | |
1087 | int err; | |
1088 | struct page_to_node *pp; | |
1089 | LIST_HEAD(pagelist); | |
1090 | ||
1091 | down_read(&mm->mmap_sem); | |
1092 | ||
1093 | /* | |
1094 | * Build a list of pages to migrate | |
1095 | */ | |
1096 | for (pp = pm; pp->node != MAX_NUMNODES; pp++) { | |
1097 | struct vm_area_struct *vma; | |
1098 | struct page *page; | |
1099 | ||
1100 | err = -EFAULT; | |
1101 | vma = find_vma(mm, pp->addr); | |
1102 | if (!vma || pp->addr < vma->vm_start || !vma_migratable(vma)) | |
1103 | goto set_status; | |
1104 | ||
1105 | page = follow_page(vma, pp->addr, FOLL_GET|FOLL_SPLIT); | |
1106 | ||
1107 | err = PTR_ERR(page); | |
1108 | if (IS_ERR(page)) | |
1109 | goto set_status; | |
1110 | ||
1111 | err = -ENOENT; | |
1112 | if (!page) | |
1113 | goto set_status; | |
1114 | ||
1115 | /* Use PageReserved to check for zero page */ | |
1116 | if (PageReserved(page) || PageKsm(page)) | |
1117 | goto put_and_set; | |
1118 | ||
1119 | pp->page = page; | |
1120 | err = page_to_nid(page); | |
1121 | ||
1122 | if (err == pp->node) | |
1123 | /* | |
1124 | * Node already in the right place | |
1125 | */ | |
1126 | goto put_and_set; | |
1127 | ||
1128 | err = -EACCES; | |
1129 | if (page_mapcount(page) > 1 && | |
1130 | !migrate_all) | |
1131 | goto put_and_set; | |
1132 | ||
1133 | err = isolate_lru_page(page); | |
1134 | if (!err) { | |
1135 | list_add_tail(&page->lru, &pagelist); | |
1136 | inc_zone_page_state(page, NR_ISOLATED_ANON + | |
1137 | page_is_file_cache(page)); | |
1138 | } | |
1139 | put_and_set: | |
1140 | /* | |
1141 | * Either remove the duplicate refcount from | |
1142 | * isolate_lru_page() or drop the page ref if it was | |
1143 | * not isolated. | |
1144 | */ | |
1145 | put_page(page); | |
1146 | set_status: | |
1147 | pp->status = err; | |
1148 | } | |
1149 | ||
1150 | err = 0; | |
1151 | if (!list_empty(&pagelist)) { | |
1152 | err = migrate_pages(&pagelist, new_page_node, | |
1153 | (unsigned long)pm, 0, MIGRATE_SYNC, | |
1154 | MR_SYSCALL); | |
1155 | if (err) | |
1156 | putback_lru_pages(&pagelist); | |
1157 | } | |
1158 | ||
1159 | up_read(&mm->mmap_sem); | |
1160 | return err; | |
1161 | } | |
1162 | ||
1163 | /* | |
1164 | * Migrate an array of page address onto an array of nodes and fill | |
1165 | * the corresponding array of status. | |
1166 | */ | |
1167 | static int do_pages_move(struct mm_struct *mm, nodemask_t task_nodes, | |
1168 | unsigned long nr_pages, | |
1169 | const void __user * __user *pages, | |
1170 | const int __user *nodes, | |
1171 | int __user *status, int flags) | |
1172 | { | |
1173 | struct page_to_node *pm; | |
1174 | unsigned long chunk_nr_pages; | |
1175 | unsigned long chunk_start; | |
1176 | int err; | |
1177 | ||
1178 | err = -ENOMEM; | |
1179 | pm = (struct page_to_node *)__get_free_page(GFP_KERNEL); | |
1180 | if (!pm) | |
1181 | goto out; | |
1182 | ||
1183 | migrate_prep(); | |
1184 | ||
1185 | /* | |
1186 | * Store a chunk of page_to_node array in a page, | |
1187 | * but keep the last one as a marker | |
1188 | */ | |
1189 | chunk_nr_pages = (PAGE_SIZE / sizeof(struct page_to_node)) - 1; | |
1190 | ||
1191 | for (chunk_start = 0; | |
1192 | chunk_start < nr_pages; | |
1193 | chunk_start += chunk_nr_pages) { | |
1194 | int j; | |
1195 | ||
1196 | if (chunk_start + chunk_nr_pages > nr_pages) | |
1197 | chunk_nr_pages = nr_pages - chunk_start; | |
1198 | ||
1199 | /* fill the chunk pm with addrs and nodes from user-space */ | |
1200 | for (j = 0; j < chunk_nr_pages; j++) { | |
1201 | const void __user *p; | |
1202 | int node; | |
1203 | ||
1204 | err = -EFAULT; | |
1205 | if (get_user(p, pages + j + chunk_start)) | |
1206 | goto out_pm; | |
1207 | pm[j].addr = (unsigned long) p; | |
1208 | ||
1209 | if (get_user(node, nodes + j + chunk_start)) | |
1210 | goto out_pm; | |
1211 | ||
1212 | err = -ENODEV; | |
1213 | if (node < 0 || node >= MAX_NUMNODES) | |
1214 | goto out_pm; | |
1215 | ||
1216 | if (!node_state(node, N_HIGH_MEMORY)) | |
1217 | goto out_pm; | |
1218 | ||
1219 | err = -EACCES; | |
1220 | if (!node_isset(node, task_nodes)) | |
1221 | goto out_pm; | |
1222 | ||
1223 | pm[j].node = node; | |
1224 | } | |
1225 | ||
1226 | /* End marker for this chunk */ | |
1227 | pm[chunk_nr_pages].node = MAX_NUMNODES; | |
1228 | ||
1229 | /* Migrate this chunk */ | |
1230 | err = do_move_page_to_node_array(mm, pm, | |
1231 | flags & MPOL_MF_MOVE_ALL); | |
1232 | if (err < 0) | |
1233 | goto out_pm; | |
1234 | ||
1235 | /* Return status information */ | |
1236 | for (j = 0; j < chunk_nr_pages; j++) | |
1237 | if (put_user(pm[j].status, status + j + chunk_start)) { | |
1238 | err = -EFAULT; | |
1239 | goto out_pm; | |
1240 | } | |
1241 | } | |
1242 | err = 0; | |
1243 | ||
1244 | out_pm: | |
1245 | free_page((unsigned long)pm); | |
1246 | out: | |
1247 | return err; | |
1248 | } | |
1249 | ||
1250 | /* | |
1251 | * Determine the nodes of an array of pages and store it in an array of status. | |
1252 | */ | |
1253 | static void do_pages_stat_array(struct mm_struct *mm, unsigned long nr_pages, | |
1254 | const void __user **pages, int *status) | |
1255 | { | |
1256 | unsigned long i; | |
1257 | ||
1258 | down_read(&mm->mmap_sem); | |
1259 | ||
1260 | for (i = 0; i < nr_pages; i++) { | |
1261 | unsigned long addr = (unsigned long)(*pages); | |
1262 | struct vm_area_struct *vma; | |
1263 | struct page *page; | |
1264 | int err = -EFAULT; | |
1265 | ||
1266 | vma = find_vma(mm, addr); | |
1267 | if (!vma || addr < vma->vm_start) | |
1268 | goto set_status; | |
1269 | ||
1270 | page = follow_page(vma, addr, 0); | |
1271 | ||
1272 | err = PTR_ERR(page); | |
1273 | if (IS_ERR(page)) | |
1274 | goto set_status; | |
1275 | ||
1276 | err = -ENOENT; | |
1277 | /* Use PageReserved to check for zero page */ | |
1278 | if (!page || PageReserved(page) || PageKsm(page)) | |
1279 | goto set_status; | |
1280 | ||
1281 | err = page_to_nid(page); | |
1282 | set_status: | |
1283 | *status = err; | |
1284 | ||
1285 | pages++; | |
1286 | status++; | |
1287 | } | |
1288 | ||
1289 | up_read(&mm->mmap_sem); | |
1290 | } | |
1291 | ||
1292 | /* | |
1293 | * Determine the nodes of a user array of pages and store it in | |
1294 | * a user array of status. | |
1295 | */ | |
1296 | static int do_pages_stat(struct mm_struct *mm, unsigned long nr_pages, | |
1297 | const void __user * __user *pages, | |
1298 | int __user *status) | |
1299 | { | |
1300 | #define DO_PAGES_STAT_CHUNK_NR 16 | |
1301 | const void __user *chunk_pages[DO_PAGES_STAT_CHUNK_NR]; | |
1302 | int chunk_status[DO_PAGES_STAT_CHUNK_NR]; | |
1303 | ||
1304 | while (nr_pages) { | |
1305 | unsigned long chunk_nr; | |
1306 | ||
1307 | chunk_nr = nr_pages; | |
1308 | if (chunk_nr > DO_PAGES_STAT_CHUNK_NR) | |
1309 | chunk_nr = DO_PAGES_STAT_CHUNK_NR; | |
1310 | ||
1311 | if (copy_from_user(chunk_pages, pages, chunk_nr * sizeof(*chunk_pages))) | |
1312 | break; | |
1313 | ||
1314 | do_pages_stat_array(mm, chunk_nr, chunk_pages, chunk_status); | |
1315 | ||
1316 | if (copy_to_user(status, chunk_status, chunk_nr * sizeof(*status))) | |
1317 | break; | |
1318 | ||
1319 | pages += chunk_nr; | |
1320 | status += chunk_nr; | |
1321 | nr_pages -= chunk_nr; | |
1322 | } | |
1323 | return nr_pages ? -EFAULT : 0; | |
1324 | } | |
1325 | ||
1326 | /* | |
1327 | * Move a list of pages in the address space of the currently executing | |
1328 | * process. | |
1329 | */ | |
1330 | SYSCALL_DEFINE6(move_pages, pid_t, pid, unsigned long, nr_pages, | |
1331 | const void __user * __user *, pages, | |
1332 | const int __user *, nodes, | |
1333 | int __user *, status, int, flags) | |
1334 | { | |
1335 | const struct cred *cred = current_cred(), *tcred; | |
1336 | struct task_struct *task; | |
1337 | struct mm_struct *mm; | |
1338 | int err; | |
1339 | nodemask_t task_nodes; | |
1340 | ||
1341 | /* Check flags */ | |
1342 | if (flags & ~(MPOL_MF_MOVE|MPOL_MF_MOVE_ALL)) | |
1343 | return -EINVAL; | |
1344 | ||
1345 | if ((flags & MPOL_MF_MOVE_ALL) && !capable(CAP_SYS_NICE)) | |
1346 | return -EPERM; | |
1347 | ||
1348 | /* Find the mm_struct */ | |
1349 | rcu_read_lock(); | |
1350 | task = pid ? find_task_by_vpid(pid) : current; | |
1351 | if (!task) { | |
1352 | rcu_read_unlock(); | |
1353 | return -ESRCH; | |
1354 | } | |
1355 | get_task_struct(task); | |
1356 | ||
1357 | /* | |
1358 | * Check if this process has the right to modify the specified | |
1359 | * process. The right exists if the process has administrative | |
1360 | * capabilities, superuser privileges or the same | |
1361 | * userid as the target process. | |
1362 | */ | |
1363 | tcred = __task_cred(task); | |
1364 | if (!uid_eq(cred->euid, tcred->suid) && !uid_eq(cred->euid, tcred->uid) && | |
1365 | !uid_eq(cred->uid, tcred->suid) && !uid_eq(cred->uid, tcred->uid) && | |
1366 | !capable(CAP_SYS_NICE)) { | |
1367 | rcu_read_unlock(); | |
1368 | err = -EPERM; | |
1369 | goto out; | |
1370 | } | |
1371 | rcu_read_unlock(); | |
1372 | ||
1373 | err = security_task_movememory(task); | |
1374 | if (err) | |
1375 | goto out; | |
1376 | ||
1377 | task_nodes = cpuset_mems_allowed(task); | |
1378 | mm = get_task_mm(task); | |
1379 | put_task_struct(task); | |
1380 | ||
1381 | if (!mm) | |
1382 | return -EINVAL; | |
1383 | ||
1384 | if (nodes) | |
1385 | err = do_pages_move(mm, task_nodes, nr_pages, pages, | |
1386 | nodes, status, flags); | |
1387 | else | |
1388 | err = do_pages_stat(mm, nr_pages, pages, status); | |
1389 | ||
1390 | mmput(mm); | |
1391 | return err; | |
1392 | ||
1393 | out: | |
1394 | put_task_struct(task); | |
1395 | return err; | |
1396 | } | |
1397 | ||
1398 | /* | |
1399 | * Call migration functions in the vma_ops that may prepare | |
1400 | * memory in a vm for migration. migration functions may perform | |
1401 | * the migration for vmas that do not have an underlying page struct. | |
1402 | */ | |
1403 | int migrate_vmas(struct mm_struct *mm, const nodemask_t *to, | |
1404 | const nodemask_t *from, unsigned long flags) | |
1405 | { | |
1406 | struct vm_area_struct *vma; | |
1407 | int err = 0; | |
1408 | ||
1409 | for (vma = mm->mmap; vma && !err; vma = vma->vm_next) { | |
1410 | if (vma->vm_ops && vma->vm_ops->migrate) { | |
1411 | err = vma->vm_ops->migrate(vma, to, from, flags); | |
1412 | if (err) | |
1413 | break; | |
1414 | } | |
1415 | } | |
1416 | return err; | |
1417 | } | |
1418 | #endif |