Commit | Line | Data |
---|---|---|
b20a3503 CL |
1 | /* |
2 | * Memory Migration functionality - linux/mm/migration.c | |
3 | * | |
4 | * Copyright (C) 2006 Silicon Graphics, Inc., Christoph Lameter | |
5 | * | |
6 | * Page migration was first developed in the context of the memory hotplug | |
7 | * project. The main authors of the migration code are: | |
8 | * | |
9 | * IWAMOTO Toshihiro <iwamoto@valinux.co.jp> | |
10 | * Hirokazu Takahashi <taka@valinux.co.jp> | |
11 | * Dave Hansen <haveblue@us.ibm.com> | |
cde53535 | 12 | * Christoph Lameter |
b20a3503 CL |
13 | */ |
14 | ||
15 | #include <linux/migrate.h> | |
b95f1b31 | 16 | #include <linux/export.h> |
b20a3503 | 17 | #include <linux/swap.h> |
0697212a | 18 | #include <linux/swapops.h> |
b20a3503 | 19 | #include <linux/pagemap.h> |
e23ca00b | 20 | #include <linux/buffer_head.h> |
b20a3503 | 21 | #include <linux/mm_inline.h> |
b488893a | 22 | #include <linux/nsproxy.h> |
b20a3503 | 23 | #include <linux/pagevec.h> |
e9995ef9 | 24 | #include <linux/ksm.h> |
b20a3503 CL |
25 | #include <linux/rmap.h> |
26 | #include <linux/topology.h> | |
27 | #include <linux/cpu.h> | |
28 | #include <linux/cpuset.h> | |
04e62a29 | 29 | #include <linux/writeback.h> |
742755a1 CL |
30 | #include <linux/mempolicy.h> |
31 | #include <linux/vmalloc.h> | |
86c3a764 | 32 | #include <linux/security.h> |
8a9f3ccd | 33 | #include <linux/memcontrol.h> |
4f5ca265 | 34 | #include <linux/syscalls.h> |
290408d4 | 35 | #include <linux/hugetlb.h> |
5a0e3ad6 | 36 | #include <linux/gfp.h> |
b20a3503 | 37 | |
0d1836c3 MN |
38 | #include <asm/tlbflush.h> |
39 | ||
b20a3503 CL |
40 | #include "internal.h" |
41 | ||
b20a3503 | 42 | /* |
742755a1 | 43 | * migrate_prep() needs to be called before we start compiling a list of pages |
748446bb MG |
44 | * to be migrated using isolate_lru_page(). If scheduling work on other CPUs is |
45 | * undesirable, use migrate_prep_local() | |
b20a3503 CL |
46 | */ |
47 | int migrate_prep(void) | |
48 | { | |
b20a3503 CL |
49 | /* |
50 | * Clear the LRU lists so pages can be isolated. | |
51 | * Note that pages may be moved off the LRU after we have | |
52 | * drained them. Those pages will fail to migrate like other | |
53 | * pages that may be busy. | |
54 | */ | |
55 | lru_add_drain_all(); | |
56 | ||
57 | return 0; | |
58 | } | |
59 | ||
748446bb MG |
60 | /* Do the necessary work of migrate_prep but not if it involves other CPUs */ |
61 | int migrate_prep_local(void) | |
62 | { | |
63 | lru_add_drain(); | |
64 | ||
65 | return 0; | |
66 | } | |
67 | ||
b20a3503 | 68 | /* |
894bc310 LS |
69 | * Add isolated pages on the list back to the LRU under page lock |
70 | * to avoid leaking evictable pages back onto unevictable list. | |
b20a3503 | 71 | */ |
e13861d8 | 72 | void putback_lru_pages(struct list_head *l) |
b20a3503 CL |
73 | { |
74 | struct page *page; | |
75 | struct page *page2; | |
b20a3503 CL |
76 | |
77 | list_for_each_entry_safe(page, page2, l, lru) { | |
e24f0b8f | 78 | list_del(&page->lru); |
a731286d | 79 | dec_zone_page_state(page, NR_ISOLATED_ANON + |
6c0b1351 | 80 | page_is_file_cache(page)); |
894bc310 | 81 | putback_lru_page(page); |
b20a3503 | 82 | } |
b20a3503 CL |
83 | } |
84 | ||
0697212a CL |
85 | /* |
86 | * Restore a potential migration pte to a working pte entry | |
87 | */ | |
e9995ef9 HD |
88 | static int remove_migration_pte(struct page *new, struct vm_area_struct *vma, |
89 | unsigned long addr, void *old) | |
0697212a CL |
90 | { |
91 | struct mm_struct *mm = vma->vm_mm; | |
92 | swp_entry_t entry; | |
93 | pgd_t *pgd; | |
94 | pud_t *pud; | |
95 | pmd_t *pmd; | |
96 | pte_t *ptep, pte; | |
97 | spinlock_t *ptl; | |
98 | ||
290408d4 NH |
99 | if (unlikely(PageHuge(new))) { |
100 | ptep = huge_pte_offset(mm, addr); | |
101 | if (!ptep) | |
102 | goto out; | |
103 | ptl = &mm->page_table_lock; | |
104 | } else { | |
105 | pgd = pgd_offset(mm, addr); | |
106 | if (!pgd_present(*pgd)) | |
107 | goto out; | |
0697212a | 108 | |
290408d4 NH |
109 | pud = pud_offset(pgd, addr); |
110 | if (!pud_present(*pud)) | |
111 | goto out; | |
0697212a | 112 | |
290408d4 | 113 | pmd = pmd_offset(pud, addr); |
500d65d4 AA |
114 | if (pmd_trans_huge(*pmd)) |
115 | goto out; | |
290408d4 NH |
116 | if (!pmd_present(*pmd)) |
117 | goto out; | |
0697212a | 118 | |
290408d4 | 119 | ptep = pte_offset_map(pmd, addr); |
0697212a | 120 | |
486cf46f HD |
121 | /* |
122 | * Peek to check is_swap_pte() before taking ptlock? No, we | |
123 | * can race mremap's move_ptes(), which skips anon_vma lock. | |
124 | */ | |
290408d4 NH |
125 | |
126 | ptl = pte_lockptr(mm, pmd); | |
127 | } | |
0697212a | 128 | |
0697212a CL |
129 | spin_lock(ptl); |
130 | pte = *ptep; | |
131 | if (!is_swap_pte(pte)) | |
e9995ef9 | 132 | goto unlock; |
0697212a CL |
133 | |
134 | entry = pte_to_swp_entry(pte); | |
135 | ||
e9995ef9 HD |
136 | if (!is_migration_entry(entry) || |
137 | migration_entry_to_page(entry) != old) | |
138 | goto unlock; | |
0697212a | 139 | |
0697212a CL |
140 | get_page(new); |
141 | pte = pte_mkold(mk_pte(new, vma->vm_page_prot)); | |
142 | if (is_write_migration_entry(entry)) | |
143 | pte = pte_mkwrite(pte); | |
3ef8fd7f | 144 | #ifdef CONFIG_HUGETLB_PAGE |
290408d4 NH |
145 | if (PageHuge(new)) |
146 | pte = pte_mkhuge(pte); | |
3ef8fd7f | 147 | #endif |
97ee0524 | 148 | flush_cache_page(vma, addr, pte_pfn(pte)); |
0697212a | 149 | set_pte_at(mm, addr, ptep, pte); |
04e62a29 | 150 | |
290408d4 NH |
151 | if (PageHuge(new)) { |
152 | if (PageAnon(new)) | |
153 | hugepage_add_anon_rmap(new, vma, addr); | |
154 | else | |
155 | page_dup_rmap(new); | |
156 | } else if (PageAnon(new)) | |
04e62a29 CL |
157 | page_add_anon_rmap(new, vma, addr); |
158 | else | |
159 | page_add_file_rmap(new); | |
160 | ||
161 | /* No need to invalidate - it was non-present before */ | |
4b3073e1 | 162 | update_mmu_cache(vma, addr, ptep); |
e9995ef9 | 163 | unlock: |
0697212a | 164 | pte_unmap_unlock(ptep, ptl); |
e9995ef9 HD |
165 | out: |
166 | return SWAP_AGAIN; | |
0697212a CL |
167 | } |
168 | ||
04e62a29 CL |
169 | /* |
170 | * Get rid of all migration entries and replace them by | |
171 | * references to the indicated page. | |
172 | */ | |
173 | static void remove_migration_ptes(struct page *old, struct page *new) | |
174 | { | |
e9995ef9 | 175 | rmap_walk(new, remove_migration_pte, old); |
04e62a29 CL |
176 | } |
177 | ||
0697212a CL |
178 | /* |
179 | * Something used the pte of a page under migration. We need to | |
180 | * get to the page and wait until migration is finished. | |
181 | * When we return from this function the fault will be retried. | |
0697212a CL |
182 | */ |
183 | void migration_entry_wait(struct mm_struct *mm, pmd_t *pmd, | |
184 | unsigned long address) | |
185 | { | |
186 | pte_t *ptep, pte; | |
187 | spinlock_t *ptl; | |
188 | swp_entry_t entry; | |
189 | struct page *page; | |
190 | ||
191 | ptep = pte_offset_map_lock(mm, pmd, address, &ptl); | |
192 | pte = *ptep; | |
193 | if (!is_swap_pte(pte)) | |
194 | goto out; | |
195 | ||
196 | entry = pte_to_swp_entry(pte); | |
197 | if (!is_migration_entry(entry)) | |
198 | goto out; | |
199 | ||
200 | page = migration_entry_to_page(entry); | |
201 | ||
e286781d NP |
202 | /* |
203 | * Once radix-tree replacement of page migration started, page_count | |
204 | * *must* be zero. And, we don't want to call wait_on_page_locked() | |
205 | * against a page without get_page(). | |
206 | * So, we use get_page_unless_zero(), here. Even failed, page fault | |
207 | * will occur again. | |
208 | */ | |
209 | if (!get_page_unless_zero(page)) | |
210 | goto out; | |
0697212a CL |
211 | pte_unmap_unlock(ptep, ptl); |
212 | wait_on_page_locked(page); | |
213 | put_page(page); | |
214 | return; | |
215 | out: | |
216 | pte_unmap_unlock(ptep, ptl); | |
217 | } | |
218 | ||
b20a3503 | 219 | /* |
c3fcf8a5 | 220 | * Replace the page in the mapping. |
5b5c7120 CL |
221 | * |
222 | * The number of remaining references must be: | |
223 | * 1 for anonymous pages without a mapping | |
224 | * 2 for pages with a mapping | |
266cf658 | 225 | * 3 for pages with a mapping and PagePrivate/PagePrivate2 set. |
b20a3503 | 226 | */ |
2d1db3b1 CL |
227 | static int migrate_page_move_mapping(struct address_space *mapping, |
228 | struct page *newpage, struct page *page) | |
b20a3503 | 229 | { |
e286781d | 230 | int expected_count; |
7cf9c2c7 | 231 | void **pslot; |
b20a3503 | 232 | |
6c5240ae | 233 | if (!mapping) { |
0e8c7d0f | 234 | /* Anonymous page without mapping */ |
6c5240ae CL |
235 | if (page_count(page) != 1) |
236 | return -EAGAIN; | |
237 | return 0; | |
238 | } | |
239 | ||
19fd6231 | 240 | spin_lock_irq(&mapping->tree_lock); |
b20a3503 | 241 | |
7cf9c2c7 NP |
242 | pslot = radix_tree_lookup_slot(&mapping->page_tree, |
243 | page_index(page)); | |
b20a3503 | 244 | |
edcf4748 | 245 | expected_count = 2 + page_has_private(page); |
e286781d | 246 | if (page_count(page) != expected_count || |
29c1f677 | 247 | radix_tree_deref_slot_protected(pslot, &mapping->tree_lock) != page) { |
19fd6231 | 248 | spin_unlock_irq(&mapping->tree_lock); |
e23ca00b | 249 | return -EAGAIN; |
b20a3503 CL |
250 | } |
251 | ||
e286781d | 252 | if (!page_freeze_refs(page, expected_count)) { |
19fd6231 | 253 | spin_unlock_irq(&mapping->tree_lock); |
e286781d NP |
254 | return -EAGAIN; |
255 | } | |
256 | ||
b20a3503 CL |
257 | /* |
258 | * Now we know that no one else is looking at the page. | |
b20a3503 | 259 | */ |
7cf9c2c7 | 260 | get_page(newpage); /* add cache reference */ |
b20a3503 CL |
261 | if (PageSwapCache(page)) { |
262 | SetPageSwapCache(newpage); | |
263 | set_page_private(newpage, page_private(page)); | |
264 | } | |
265 | ||
7cf9c2c7 NP |
266 | radix_tree_replace_slot(pslot, newpage); |
267 | ||
268 | /* | |
937a94c9 JG |
269 | * Drop cache reference from old page by unfreezing |
270 | * to one less reference. | |
7cf9c2c7 NP |
271 | * We know this isn't the last reference. |
272 | */ | |
937a94c9 | 273 | page_unfreeze_refs(page, expected_count - 1); |
7cf9c2c7 | 274 | |
0e8c7d0f CL |
275 | /* |
276 | * If moved to a different zone then also account | |
277 | * the page for that zone. Other VM counters will be | |
278 | * taken care of when we establish references to the | |
279 | * new page and drop references to the old page. | |
280 | * | |
281 | * Note that anonymous pages are accounted for | |
282 | * via NR_FILE_PAGES and NR_ANON_PAGES if they | |
283 | * are mapped to swap space. | |
284 | */ | |
285 | __dec_zone_page_state(page, NR_FILE_PAGES); | |
286 | __inc_zone_page_state(newpage, NR_FILE_PAGES); | |
99a15e21 | 287 | if (!PageSwapCache(page) && PageSwapBacked(page)) { |
4b02108a KM |
288 | __dec_zone_page_state(page, NR_SHMEM); |
289 | __inc_zone_page_state(newpage, NR_SHMEM); | |
290 | } | |
19fd6231 | 291 | spin_unlock_irq(&mapping->tree_lock); |
b20a3503 CL |
292 | |
293 | return 0; | |
294 | } | |
b20a3503 | 295 | |
290408d4 NH |
296 | /* |
297 | * The expected number of remaining references is the same as that | |
298 | * of migrate_page_move_mapping(). | |
299 | */ | |
300 | int migrate_huge_page_move_mapping(struct address_space *mapping, | |
301 | struct page *newpage, struct page *page) | |
302 | { | |
303 | int expected_count; | |
304 | void **pslot; | |
305 | ||
306 | if (!mapping) { | |
307 | if (page_count(page) != 1) | |
308 | return -EAGAIN; | |
309 | return 0; | |
310 | } | |
311 | ||
312 | spin_lock_irq(&mapping->tree_lock); | |
313 | ||
314 | pslot = radix_tree_lookup_slot(&mapping->page_tree, | |
315 | page_index(page)); | |
316 | ||
317 | expected_count = 2 + page_has_private(page); | |
318 | if (page_count(page) != expected_count || | |
29c1f677 | 319 | radix_tree_deref_slot_protected(pslot, &mapping->tree_lock) != page) { |
290408d4 NH |
320 | spin_unlock_irq(&mapping->tree_lock); |
321 | return -EAGAIN; | |
322 | } | |
323 | ||
324 | if (!page_freeze_refs(page, expected_count)) { | |
325 | spin_unlock_irq(&mapping->tree_lock); | |
326 | return -EAGAIN; | |
327 | } | |
328 | ||
329 | get_page(newpage); | |
330 | ||
331 | radix_tree_replace_slot(pslot, newpage); | |
332 | ||
937a94c9 | 333 | page_unfreeze_refs(page, expected_count - 1); |
290408d4 NH |
334 | |
335 | spin_unlock_irq(&mapping->tree_lock); | |
336 | return 0; | |
337 | } | |
338 | ||
b20a3503 CL |
339 | /* |
340 | * Copy the page to its new location | |
341 | */ | |
290408d4 | 342 | void migrate_page_copy(struct page *newpage, struct page *page) |
b20a3503 | 343 | { |
290408d4 NH |
344 | if (PageHuge(page)) |
345 | copy_huge_page(newpage, page); | |
346 | else | |
347 | copy_highpage(newpage, page); | |
b20a3503 CL |
348 | |
349 | if (PageError(page)) | |
350 | SetPageError(newpage); | |
351 | if (PageReferenced(page)) | |
352 | SetPageReferenced(newpage); | |
353 | if (PageUptodate(page)) | |
354 | SetPageUptodate(newpage); | |
894bc310 LS |
355 | if (TestClearPageActive(page)) { |
356 | VM_BUG_ON(PageUnevictable(page)); | |
b20a3503 | 357 | SetPageActive(newpage); |
418b27ef LS |
358 | } else if (TestClearPageUnevictable(page)) |
359 | SetPageUnevictable(newpage); | |
b20a3503 CL |
360 | if (PageChecked(page)) |
361 | SetPageChecked(newpage); | |
362 | if (PageMappedToDisk(page)) | |
363 | SetPageMappedToDisk(newpage); | |
364 | ||
365 | if (PageDirty(page)) { | |
366 | clear_page_dirty_for_io(page); | |
3a902c5f NP |
367 | /* |
368 | * Want to mark the page and the radix tree as dirty, and | |
369 | * redo the accounting that clear_page_dirty_for_io undid, | |
370 | * but we can't use set_page_dirty because that function | |
371 | * is actually a signal that all of the page has become dirty. | |
25985edc | 372 | * Whereas only part of our page may be dirty. |
3a902c5f NP |
373 | */ |
374 | __set_page_dirty_nobuffers(newpage); | |
b20a3503 CL |
375 | } |
376 | ||
b291f000 | 377 | mlock_migrate_page(newpage, page); |
e9995ef9 | 378 | ksm_migrate_page(newpage, page); |
b291f000 | 379 | |
b20a3503 | 380 | ClearPageSwapCache(page); |
b20a3503 CL |
381 | ClearPagePrivate(page); |
382 | set_page_private(page, 0); | |
383 | page->mapping = NULL; | |
384 | ||
385 | /* | |
386 | * If any waiters have accumulated on the new page then | |
387 | * wake them up. | |
388 | */ | |
389 | if (PageWriteback(newpage)) | |
390 | end_page_writeback(newpage); | |
391 | } | |
b20a3503 | 392 | |
1d8b85cc CL |
393 | /************************************************************ |
394 | * Migration functions | |
395 | ***********************************************************/ | |
396 | ||
397 | /* Always fail migration. Used for mappings that are not movable */ | |
2d1db3b1 CL |
398 | int fail_migrate_page(struct address_space *mapping, |
399 | struct page *newpage, struct page *page) | |
1d8b85cc CL |
400 | { |
401 | return -EIO; | |
402 | } | |
403 | EXPORT_SYMBOL(fail_migrate_page); | |
404 | ||
b20a3503 CL |
405 | /* |
406 | * Common logic to directly migrate a single page suitable for | |
266cf658 | 407 | * pages that do not use PagePrivate/PagePrivate2. |
b20a3503 CL |
408 | * |
409 | * Pages are locked upon entry and exit. | |
410 | */ | |
2d1db3b1 CL |
411 | int migrate_page(struct address_space *mapping, |
412 | struct page *newpage, struct page *page) | |
b20a3503 CL |
413 | { |
414 | int rc; | |
415 | ||
416 | BUG_ON(PageWriteback(page)); /* Writeback must be complete */ | |
417 | ||
2d1db3b1 | 418 | rc = migrate_page_move_mapping(mapping, newpage, page); |
b20a3503 CL |
419 | |
420 | if (rc) | |
421 | return rc; | |
422 | ||
423 | migrate_page_copy(newpage, page); | |
b20a3503 CL |
424 | return 0; |
425 | } | |
426 | EXPORT_SYMBOL(migrate_page); | |
427 | ||
9361401e | 428 | #ifdef CONFIG_BLOCK |
1d8b85cc CL |
429 | /* |
430 | * Migration function for pages with buffers. This function can only be used | |
431 | * if the underlying filesystem guarantees that no other references to "page" | |
432 | * exist. | |
433 | */ | |
2d1db3b1 CL |
434 | int buffer_migrate_page(struct address_space *mapping, |
435 | struct page *newpage, struct page *page) | |
1d8b85cc | 436 | { |
1d8b85cc CL |
437 | struct buffer_head *bh, *head; |
438 | int rc; | |
439 | ||
1d8b85cc | 440 | if (!page_has_buffers(page)) |
2d1db3b1 | 441 | return migrate_page(mapping, newpage, page); |
1d8b85cc CL |
442 | |
443 | head = page_buffers(page); | |
444 | ||
2d1db3b1 | 445 | rc = migrate_page_move_mapping(mapping, newpage, page); |
1d8b85cc CL |
446 | |
447 | if (rc) | |
448 | return rc; | |
449 | ||
450 | bh = head; | |
451 | do { | |
452 | get_bh(bh); | |
453 | lock_buffer(bh); | |
454 | bh = bh->b_this_page; | |
455 | ||
456 | } while (bh != head); | |
457 | ||
458 | ClearPagePrivate(page); | |
459 | set_page_private(newpage, page_private(page)); | |
460 | set_page_private(page, 0); | |
461 | put_page(page); | |
462 | get_page(newpage); | |
463 | ||
464 | bh = head; | |
465 | do { | |
466 | set_bh_page(bh, newpage, bh_offset(bh)); | |
467 | bh = bh->b_this_page; | |
468 | ||
469 | } while (bh != head); | |
470 | ||
471 | SetPagePrivate(newpage); | |
472 | ||
473 | migrate_page_copy(newpage, page); | |
474 | ||
475 | bh = head; | |
476 | do { | |
477 | unlock_buffer(bh); | |
478 | put_bh(bh); | |
479 | bh = bh->b_this_page; | |
480 | ||
481 | } while (bh != head); | |
482 | ||
483 | return 0; | |
484 | } | |
485 | EXPORT_SYMBOL(buffer_migrate_page); | |
9361401e | 486 | #endif |
1d8b85cc | 487 | |
04e62a29 CL |
488 | /* |
489 | * Writeback a page to clean the dirty state | |
490 | */ | |
491 | static int writeout(struct address_space *mapping, struct page *page) | |
8351a6e4 | 492 | { |
04e62a29 CL |
493 | struct writeback_control wbc = { |
494 | .sync_mode = WB_SYNC_NONE, | |
495 | .nr_to_write = 1, | |
496 | .range_start = 0, | |
497 | .range_end = LLONG_MAX, | |
04e62a29 CL |
498 | .for_reclaim = 1 |
499 | }; | |
500 | int rc; | |
501 | ||
502 | if (!mapping->a_ops->writepage) | |
503 | /* No write method for the address space */ | |
504 | return -EINVAL; | |
505 | ||
506 | if (!clear_page_dirty_for_io(page)) | |
507 | /* Someone else already triggered a write */ | |
508 | return -EAGAIN; | |
509 | ||
8351a6e4 | 510 | /* |
04e62a29 CL |
511 | * A dirty page may imply that the underlying filesystem has |
512 | * the page on some queue. So the page must be clean for | |
513 | * migration. Writeout may mean we loose the lock and the | |
514 | * page state is no longer what we checked for earlier. | |
515 | * At this point we know that the migration attempt cannot | |
516 | * be successful. | |
8351a6e4 | 517 | */ |
04e62a29 | 518 | remove_migration_ptes(page, page); |
8351a6e4 | 519 | |
04e62a29 | 520 | rc = mapping->a_ops->writepage(page, &wbc); |
8351a6e4 | 521 | |
04e62a29 CL |
522 | if (rc != AOP_WRITEPAGE_ACTIVATE) |
523 | /* unlocked. Relock */ | |
524 | lock_page(page); | |
525 | ||
bda8550d | 526 | return (rc < 0) ? -EIO : -EAGAIN; |
04e62a29 CL |
527 | } |
528 | ||
529 | /* | |
530 | * Default handling if a filesystem does not provide a migration function. | |
531 | */ | |
532 | static int fallback_migrate_page(struct address_space *mapping, | |
533 | struct page *newpage, struct page *page) | |
534 | { | |
535 | if (PageDirty(page)) | |
536 | return writeout(mapping, page); | |
8351a6e4 CL |
537 | |
538 | /* | |
539 | * Buffers may be managed in a filesystem specific way. | |
540 | * We must have no buffers or drop them. | |
541 | */ | |
266cf658 | 542 | if (page_has_private(page) && |
8351a6e4 CL |
543 | !try_to_release_page(page, GFP_KERNEL)) |
544 | return -EAGAIN; | |
545 | ||
546 | return migrate_page(mapping, newpage, page); | |
547 | } | |
548 | ||
e24f0b8f CL |
549 | /* |
550 | * Move a page to a newly allocated page | |
551 | * The page is locked and all ptes have been successfully removed. | |
552 | * | |
553 | * The new page will have replaced the old page if this function | |
554 | * is successful. | |
894bc310 LS |
555 | * |
556 | * Return value: | |
557 | * < 0 - error code | |
558 | * == 0 - success | |
e24f0b8f | 559 | */ |
3fe2011f | 560 | static int move_to_new_page(struct page *newpage, struct page *page, |
11bc82d6 | 561 | int remap_swapcache, bool sync) |
e24f0b8f CL |
562 | { |
563 | struct address_space *mapping; | |
564 | int rc; | |
565 | ||
566 | /* | |
567 | * Block others from accessing the page when we get around to | |
568 | * establishing additional references. We are the only one | |
569 | * holding a reference to the new page at this point. | |
570 | */ | |
529ae9aa | 571 | if (!trylock_page(newpage)) |
e24f0b8f CL |
572 | BUG(); |
573 | ||
574 | /* Prepare mapping for the new page.*/ | |
575 | newpage->index = page->index; | |
576 | newpage->mapping = page->mapping; | |
b2e18538 RR |
577 | if (PageSwapBacked(page)) |
578 | SetPageSwapBacked(newpage); | |
e24f0b8f CL |
579 | |
580 | mapping = page_mapping(page); | |
581 | if (!mapping) | |
582 | rc = migrate_page(mapping, newpage, page); | |
11bc82d6 | 583 | else { |
e24f0b8f | 584 | /* |
11bc82d6 AA |
585 | * Do not writeback pages if !sync and migratepage is |
586 | * not pointing to migrate_page() which is nonblocking | |
587 | * (swapcache/tmpfs uses migratepage = migrate_page). | |
e24f0b8f | 588 | */ |
11bc82d6 AA |
589 | if (PageDirty(page) && !sync && |
590 | mapping->a_ops->migratepage != migrate_page) | |
591 | rc = -EBUSY; | |
592 | else if (mapping->a_ops->migratepage) | |
593 | /* | |
594 | * Most pages have a mapping and most filesystems | |
595 | * should provide a migration function. Anonymous | |
596 | * pages are part of swap space which also has its | |
597 | * own migration function. This is the most common | |
598 | * path for page migration. | |
599 | */ | |
600 | rc = mapping->a_ops->migratepage(mapping, | |
601 | newpage, page); | |
602 | else | |
603 | rc = fallback_migrate_page(mapping, newpage, page); | |
604 | } | |
e24f0b8f | 605 | |
3fe2011f | 606 | if (rc) { |
e24f0b8f | 607 | newpage->mapping = NULL; |
3fe2011f MG |
608 | } else { |
609 | if (remap_swapcache) | |
610 | remove_migration_ptes(page, newpage); | |
611 | } | |
e24f0b8f CL |
612 | |
613 | unlock_page(newpage); | |
614 | ||
615 | return rc; | |
616 | } | |
617 | ||
0dabec93 MK |
618 | static int __unmap_and_move(struct page *page, struct page *newpage, |
619 | int force, bool offlining, bool sync) | |
e24f0b8f | 620 | { |
0dabec93 | 621 | int rc = -EAGAIN; |
3fe2011f | 622 | int remap_swapcache = 1; |
ae41be37 | 623 | int charge = 0; |
56039efa | 624 | struct mem_cgroup *mem; |
3f6c8272 | 625 | struct anon_vma *anon_vma = NULL; |
95a402c3 | 626 | |
529ae9aa | 627 | if (!trylock_page(page)) { |
11bc82d6 | 628 | if (!force || !sync) |
0dabec93 | 629 | goto out; |
3e7d3449 MG |
630 | |
631 | /* | |
632 | * It's not safe for direct compaction to call lock_page. | |
633 | * For example, during page readahead pages are added locked | |
634 | * to the LRU. Later, when the IO completes the pages are | |
635 | * marked uptodate and unlocked. However, the queueing | |
636 | * could be merging multiple pages for one bio (e.g. | |
637 | * mpage_readpages). If an allocation happens for the | |
638 | * second or third page, the process can end up locking | |
639 | * the same page twice and deadlocking. Rather than | |
640 | * trying to be clever about what pages can be locked, | |
641 | * avoid the use of lock_page for direct compaction | |
642 | * altogether. | |
643 | */ | |
644 | if (current->flags & PF_MEMALLOC) | |
0dabec93 | 645 | goto out; |
3e7d3449 | 646 | |
e24f0b8f CL |
647 | lock_page(page); |
648 | } | |
649 | ||
62b61f61 HD |
650 | /* |
651 | * Only memory hotplug's offline_pages() caller has locked out KSM, | |
652 | * and can safely migrate a KSM page. The other cases have skipped | |
653 | * PageKsm along with PageReserved - but it is only now when we have | |
654 | * the page lock that we can be certain it will not go KSM beneath us | |
655 | * (KSM will not upgrade a page from PageAnon to PageKsm when it sees | |
656 | * its pagecount raised, but only here do we take the page lock which | |
657 | * serializes that). | |
658 | */ | |
659 | if (PageKsm(page) && !offlining) { | |
660 | rc = -EBUSY; | |
661 | goto unlock; | |
662 | } | |
663 | ||
01b1ae63 | 664 | /* charge against new page */ |
ef6a3c63 | 665 | charge = mem_cgroup_prepare_migration(page, newpage, &mem, GFP_KERNEL); |
01b1ae63 KH |
666 | if (charge == -ENOMEM) { |
667 | rc = -ENOMEM; | |
668 | goto unlock; | |
669 | } | |
670 | BUG_ON(charge); | |
671 | ||
e24f0b8f | 672 | if (PageWriteback(page)) { |
11bc82d6 AA |
673 | /* |
674 | * For !sync, there is no point retrying as the retry loop | |
675 | * is expected to be too short for PageWriteback to be cleared | |
676 | */ | |
677 | if (!sync) { | |
678 | rc = -EBUSY; | |
679 | goto uncharge; | |
680 | } | |
681 | if (!force) | |
01b1ae63 | 682 | goto uncharge; |
e24f0b8f CL |
683 | wait_on_page_writeback(page); |
684 | } | |
e24f0b8f | 685 | /* |
dc386d4d KH |
686 | * By try_to_unmap(), page->mapcount goes down to 0 here. In this case, |
687 | * we cannot notice that anon_vma is freed while we migrates a page. | |
1ce82b69 | 688 | * This get_anon_vma() delays freeing anon_vma pointer until the end |
dc386d4d | 689 | * of migration. File cache pages are no problem because of page_lock() |
989f89c5 KH |
690 | * File Caches may use write_page() or lock_page() in migration, then, |
691 | * just care Anon page here. | |
dc386d4d | 692 | */ |
989f89c5 | 693 | if (PageAnon(page)) { |
1ce82b69 HD |
694 | /* |
695 | * Only page_lock_anon_vma() understands the subtleties of | |
696 | * getting a hold on an anon_vma from outside one of its mms. | |
697 | */ | |
746b18d4 | 698 | anon_vma = page_get_anon_vma(page); |
1ce82b69 HD |
699 | if (anon_vma) { |
700 | /* | |
746b18d4 | 701 | * Anon page |
1ce82b69 | 702 | */ |
1ce82b69 | 703 | } else if (PageSwapCache(page)) { |
3fe2011f MG |
704 | /* |
705 | * We cannot be sure that the anon_vma of an unmapped | |
706 | * swapcache page is safe to use because we don't | |
707 | * know in advance if the VMA that this page belonged | |
708 | * to still exists. If the VMA and others sharing the | |
709 | * data have been freed, then the anon_vma could | |
710 | * already be invalid. | |
711 | * | |
712 | * To avoid this possibility, swapcache pages get | |
713 | * migrated but are not remapped when migration | |
714 | * completes | |
715 | */ | |
716 | remap_swapcache = 0; | |
717 | } else { | |
1ce82b69 | 718 | goto uncharge; |
3fe2011f | 719 | } |
989f89c5 | 720 | } |
62e1c553 | 721 | |
dc386d4d | 722 | /* |
62e1c553 SL |
723 | * Corner case handling: |
724 | * 1. When a new swap-cache page is read into, it is added to the LRU | |
725 | * and treated as swapcache but it has no rmap yet. | |
726 | * Calling try_to_unmap() against a page->mapping==NULL page will | |
727 | * trigger a BUG. So handle it here. | |
728 | * 2. An orphaned page (see truncate_complete_page) might have | |
729 | * fs-private metadata. The page can be picked up due to memory | |
730 | * offlining. Everywhere else except page reclaim, the page is | |
731 | * invisible to the vm, so the page can not be migrated. So try to | |
732 | * free the metadata, so the page can be freed. | |
e24f0b8f | 733 | */ |
62e1c553 | 734 | if (!page->mapping) { |
1ce82b69 HD |
735 | VM_BUG_ON(PageAnon(page)); |
736 | if (page_has_private(page)) { | |
62e1c553 | 737 | try_to_free_buffers(page); |
1ce82b69 | 738 | goto uncharge; |
62e1c553 | 739 | } |
abfc3488 | 740 | goto skip_unmap; |
62e1c553 SL |
741 | } |
742 | ||
dc386d4d | 743 | /* Establish migration ptes or remove ptes */ |
14fa31b8 | 744 | try_to_unmap(page, TTU_MIGRATION|TTU_IGNORE_MLOCK|TTU_IGNORE_ACCESS); |
dc386d4d | 745 | |
abfc3488 | 746 | skip_unmap: |
e6a1530d | 747 | if (!page_mapped(page)) |
11bc82d6 | 748 | rc = move_to_new_page(newpage, page, remap_swapcache, sync); |
e24f0b8f | 749 | |
3fe2011f | 750 | if (rc && remap_swapcache) |
e24f0b8f | 751 | remove_migration_ptes(page, page); |
3f6c8272 MG |
752 | |
753 | /* Drop an anon_vma reference if we took one */ | |
76545066 | 754 | if (anon_vma) |
9e60109f | 755 | put_anon_vma(anon_vma); |
3f6c8272 | 756 | |
01b1ae63 KH |
757 | uncharge: |
758 | if (!charge) | |
50de1dd9 | 759 | mem_cgroup_end_migration(mem, page, newpage, rc == 0); |
e24f0b8f CL |
760 | unlock: |
761 | unlock_page(page); | |
0dabec93 MK |
762 | out: |
763 | return rc; | |
764 | } | |
95a402c3 | 765 | |
0dabec93 MK |
766 | /* |
767 | * Obtain the lock on page, remove all ptes and migrate the page | |
768 | * to the newly allocated page in newpage. | |
769 | */ | |
770 | static int unmap_and_move(new_page_t get_new_page, unsigned long private, | |
771 | struct page *page, int force, bool offlining, bool sync) | |
772 | { | |
773 | int rc = 0; | |
774 | int *result = NULL; | |
775 | struct page *newpage = get_new_page(page, private, &result); | |
776 | ||
777 | if (!newpage) | |
778 | return -ENOMEM; | |
779 | ||
4e5f01c2 KH |
780 | mem_cgroup_reset_owner(newpage); |
781 | ||
0dabec93 MK |
782 | if (page_count(page) == 1) { |
783 | /* page was freed from under us. So we are done. */ | |
784 | goto out; | |
785 | } | |
786 | ||
787 | if (unlikely(PageTransHuge(page))) | |
788 | if (unlikely(split_huge_page(page))) | |
789 | goto out; | |
790 | ||
791 | rc = __unmap_and_move(page, newpage, force, offlining, sync); | |
792 | out: | |
e24f0b8f | 793 | if (rc != -EAGAIN) { |
0dabec93 MK |
794 | /* |
795 | * A page that has been migrated has all references | |
796 | * removed and will be freed. A page that has not been | |
797 | * migrated will have kepts its references and be | |
798 | * restored. | |
799 | */ | |
800 | list_del(&page->lru); | |
a731286d | 801 | dec_zone_page_state(page, NR_ISOLATED_ANON + |
6c0b1351 | 802 | page_is_file_cache(page)); |
894bc310 | 803 | putback_lru_page(page); |
e24f0b8f | 804 | } |
95a402c3 CL |
805 | /* |
806 | * Move the new page to the LRU. If migration was not successful | |
807 | * then this will free the page. | |
808 | */ | |
894bc310 | 809 | putback_lru_page(newpage); |
742755a1 CL |
810 | if (result) { |
811 | if (rc) | |
812 | *result = rc; | |
813 | else | |
814 | *result = page_to_nid(newpage); | |
815 | } | |
e24f0b8f CL |
816 | return rc; |
817 | } | |
818 | ||
290408d4 NH |
819 | /* |
820 | * Counterpart of unmap_and_move_page() for hugepage migration. | |
821 | * | |
822 | * This function doesn't wait the completion of hugepage I/O | |
823 | * because there is no race between I/O and migration for hugepage. | |
824 | * Note that currently hugepage I/O occurs only in direct I/O | |
825 | * where no lock is held and PG_writeback is irrelevant, | |
826 | * and writeback status of all subpages are counted in the reference | |
827 | * count of the head page (i.e. if all subpages of a 2MB hugepage are | |
828 | * under direct I/O, the reference of the head page is 512 and a bit more.) | |
829 | * This means that when we try to migrate hugepage whose subpages are | |
830 | * doing direct I/O, some references remain after try_to_unmap() and | |
831 | * hugepage migration fails without data corruption. | |
832 | * | |
833 | * There is also no race when direct I/O is issued on the page under migration, | |
834 | * because then pte is replaced with migration swap entry and direct I/O code | |
835 | * will wait in the page fault for migration to complete. | |
836 | */ | |
837 | static int unmap_and_move_huge_page(new_page_t get_new_page, | |
838 | unsigned long private, struct page *hpage, | |
7f0f2496 | 839 | int force, bool offlining, bool sync) |
290408d4 NH |
840 | { |
841 | int rc = 0; | |
842 | int *result = NULL; | |
843 | struct page *new_hpage = get_new_page(hpage, private, &result); | |
290408d4 NH |
844 | struct anon_vma *anon_vma = NULL; |
845 | ||
846 | if (!new_hpage) | |
847 | return -ENOMEM; | |
848 | ||
849 | rc = -EAGAIN; | |
850 | ||
851 | if (!trylock_page(hpage)) { | |
77f1fe6b | 852 | if (!force || !sync) |
290408d4 NH |
853 | goto out; |
854 | lock_page(hpage); | |
855 | } | |
856 | ||
746b18d4 PZ |
857 | if (PageAnon(hpage)) |
858 | anon_vma = page_get_anon_vma(hpage); | |
290408d4 NH |
859 | |
860 | try_to_unmap(hpage, TTU_MIGRATION|TTU_IGNORE_MLOCK|TTU_IGNORE_ACCESS); | |
861 | ||
862 | if (!page_mapped(hpage)) | |
11bc82d6 | 863 | rc = move_to_new_page(new_hpage, hpage, 1, sync); |
290408d4 NH |
864 | |
865 | if (rc) | |
866 | remove_migration_ptes(hpage, hpage); | |
867 | ||
fd4a4663 | 868 | if (anon_vma) |
9e60109f | 869 | put_anon_vma(anon_vma); |
290408d4 NH |
870 | unlock_page(hpage); |
871 | ||
09761333 | 872 | out: |
290408d4 NH |
873 | if (rc != -EAGAIN) { |
874 | list_del(&hpage->lru); | |
875 | put_page(hpage); | |
876 | } | |
877 | ||
878 | put_page(new_hpage); | |
879 | ||
880 | if (result) { | |
881 | if (rc) | |
882 | *result = rc; | |
883 | else | |
884 | *result = page_to_nid(new_hpage); | |
885 | } | |
886 | return rc; | |
887 | } | |
888 | ||
b20a3503 CL |
889 | /* |
890 | * migrate_pages | |
891 | * | |
95a402c3 CL |
892 | * The function takes one list of pages to migrate and a function |
893 | * that determines from the page to be migrated and the private data | |
894 | * the target of the move and allocates the page. | |
b20a3503 CL |
895 | * |
896 | * The function returns after 10 attempts or if no pages | |
897 | * are movable anymore because to has become empty | |
cf608ac1 MK |
898 | * or no retryable pages exist anymore. |
899 | * Caller should call putback_lru_pages to return pages to the LRU | |
28bd6578 | 900 | * or free list only if ret != 0. |
b20a3503 | 901 | * |
95a402c3 | 902 | * Return: Number of pages not migrated or error code. |
b20a3503 | 903 | */ |
95a402c3 | 904 | int migrate_pages(struct list_head *from, |
7f0f2496 | 905 | new_page_t get_new_page, unsigned long private, bool offlining, |
77f1fe6b | 906 | bool sync) |
b20a3503 | 907 | { |
e24f0b8f | 908 | int retry = 1; |
b20a3503 CL |
909 | int nr_failed = 0; |
910 | int pass = 0; | |
911 | struct page *page; | |
912 | struct page *page2; | |
913 | int swapwrite = current->flags & PF_SWAPWRITE; | |
914 | int rc; | |
915 | ||
916 | if (!swapwrite) | |
917 | current->flags |= PF_SWAPWRITE; | |
918 | ||
e24f0b8f CL |
919 | for(pass = 0; pass < 10 && retry; pass++) { |
920 | retry = 0; | |
b20a3503 | 921 | |
e24f0b8f | 922 | list_for_each_entry_safe(page, page2, from, lru) { |
e24f0b8f | 923 | cond_resched(); |
2d1db3b1 | 924 | |
95a402c3 | 925 | rc = unmap_and_move(get_new_page, private, |
77f1fe6b MG |
926 | page, pass > 2, offlining, |
927 | sync); | |
2d1db3b1 | 928 | |
e24f0b8f | 929 | switch(rc) { |
95a402c3 CL |
930 | case -ENOMEM: |
931 | goto out; | |
e24f0b8f | 932 | case -EAGAIN: |
2d1db3b1 | 933 | retry++; |
e24f0b8f CL |
934 | break; |
935 | case 0: | |
e24f0b8f CL |
936 | break; |
937 | default: | |
2d1db3b1 | 938 | /* Permanent failure */ |
2d1db3b1 | 939 | nr_failed++; |
e24f0b8f | 940 | break; |
2d1db3b1 | 941 | } |
b20a3503 CL |
942 | } |
943 | } | |
95a402c3 CL |
944 | rc = 0; |
945 | out: | |
b20a3503 CL |
946 | if (!swapwrite) |
947 | current->flags &= ~PF_SWAPWRITE; | |
948 | ||
95a402c3 CL |
949 | if (rc) |
950 | return rc; | |
b20a3503 | 951 | |
95a402c3 | 952 | return nr_failed + retry; |
b20a3503 | 953 | } |
95a402c3 | 954 | |
290408d4 | 955 | int migrate_huge_pages(struct list_head *from, |
7f0f2496 | 956 | new_page_t get_new_page, unsigned long private, bool offlining, |
77f1fe6b | 957 | bool sync) |
290408d4 NH |
958 | { |
959 | int retry = 1; | |
960 | int nr_failed = 0; | |
961 | int pass = 0; | |
962 | struct page *page; | |
963 | struct page *page2; | |
964 | int rc; | |
965 | ||
966 | for (pass = 0; pass < 10 && retry; pass++) { | |
967 | retry = 0; | |
968 | ||
969 | list_for_each_entry_safe(page, page2, from, lru) { | |
970 | cond_resched(); | |
971 | ||
972 | rc = unmap_and_move_huge_page(get_new_page, | |
77f1fe6b MG |
973 | private, page, pass > 2, offlining, |
974 | sync); | |
290408d4 NH |
975 | |
976 | switch(rc) { | |
977 | case -ENOMEM: | |
978 | goto out; | |
979 | case -EAGAIN: | |
980 | retry++; | |
981 | break; | |
982 | case 0: | |
983 | break; | |
984 | default: | |
985 | /* Permanent failure */ | |
986 | nr_failed++; | |
987 | break; | |
988 | } | |
989 | } | |
990 | } | |
991 | rc = 0; | |
992 | out: | |
290408d4 NH |
993 | if (rc) |
994 | return rc; | |
995 | ||
996 | return nr_failed + retry; | |
997 | } | |
998 | ||
742755a1 CL |
999 | #ifdef CONFIG_NUMA |
1000 | /* | |
1001 | * Move a list of individual pages | |
1002 | */ | |
1003 | struct page_to_node { | |
1004 | unsigned long addr; | |
1005 | struct page *page; | |
1006 | int node; | |
1007 | int status; | |
1008 | }; | |
1009 | ||
1010 | static struct page *new_page_node(struct page *p, unsigned long private, | |
1011 | int **result) | |
1012 | { | |
1013 | struct page_to_node *pm = (struct page_to_node *)private; | |
1014 | ||
1015 | while (pm->node != MAX_NUMNODES && pm->page != p) | |
1016 | pm++; | |
1017 | ||
1018 | if (pm->node == MAX_NUMNODES) | |
1019 | return NULL; | |
1020 | ||
1021 | *result = &pm->status; | |
1022 | ||
6484eb3e | 1023 | return alloc_pages_exact_node(pm->node, |
769848c0 | 1024 | GFP_HIGHUSER_MOVABLE | GFP_THISNODE, 0); |
742755a1 CL |
1025 | } |
1026 | ||
1027 | /* | |
1028 | * Move a set of pages as indicated in the pm array. The addr | |
1029 | * field must be set to the virtual address of the page to be moved | |
1030 | * and the node number must contain a valid target node. | |
5e9a0f02 | 1031 | * The pm array ends with node = MAX_NUMNODES. |
742755a1 | 1032 | */ |
5e9a0f02 BG |
1033 | static int do_move_page_to_node_array(struct mm_struct *mm, |
1034 | struct page_to_node *pm, | |
1035 | int migrate_all) | |
742755a1 CL |
1036 | { |
1037 | int err; | |
1038 | struct page_to_node *pp; | |
1039 | LIST_HEAD(pagelist); | |
1040 | ||
1041 | down_read(&mm->mmap_sem); | |
1042 | ||
1043 | /* | |
1044 | * Build a list of pages to migrate | |
1045 | */ | |
742755a1 CL |
1046 | for (pp = pm; pp->node != MAX_NUMNODES; pp++) { |
1047 | struct vm_area_struct *vma; | |
1048 | struct page *page; | |
1049 | ||
742755a1 CL |
1050 | err = -EFAULT; |
1051 | vma = find_vma(mm, pp->addr); | |
70384dc6 | 1052 | if (!vma || pp->addr < vma->vm_start || !vma_migratable(vma)) |
742755a1 CL |
1053 | goto set_status; |
1054 | ||
500d65d4 | 1055 | page = follow_page(vma, pp->addr, FOLL_GET|FOLL_SPLIT); |
89f5b7da LT |
1056 | |
1057 | err = PTR_ERR(page); | |
1058 | if (IS_ERR(page)) | |
1059 | goto set_status; | |
1060 | ||
742755a1 CL |
1061 | err = -ENOENT; |
1062 | if (!page) | |
1063 | goto set_status; | |
1064 | ||
62b61f61 HD |
1065 | /* Use PageReserved to check for zero page */ |
1066 | if (PageReserved(page) || PageKsm(page)) | |
742755a1 CL |
1067 | goto put_and_set; |
1068 | ||
1069 | pp->page = page; | |
1070 | err = page_to_nid(page); | |
1071 | ||
1072 | if (err == pp->node) | |
1073 | /* | |
1074 | * Node already in the right place | |
1075 | */ | |
1076 | goto put_and_set; | |
1077 | ||
1078 | err = -EACCES; | |
1079 | if (page_mapcount(page) > 1 && | |
1080 | !migrate_all) | |
1081 | goto put_and_set; | |
1082 | ||
62695a84 | 1083 | err = isolate_lru_page(page); |
6d9c285a | 1084 | if (!err) { |
62695a84 | 1085 | list_add_tail(&page->lru, &pagelist); |
6d9c285a KM |
1086 | inc_zone_page_state(page, NR_ISOLATED_ANON + |
1087 | page_is_file_cache(page)); | |
1088 | } | |
742755a1 CL |
1089 | put_and_set: |
1090 | /* | |
1091 | * Either remove the duplicate refcount from | |
1092 | * isolate_lru_page() or drop the page ref if it was | |
1093 | * not isolated. | |
1094 | */ | |
1095 | put_page(page); | |
1096 | set_status: | |
1097 | pp->status = err; | |
1098 | } | |
1099 | ||
e78bbfa8 | 1100 | err = 0; |
cf608ac1 | 1101 | if (!list_empty(&pagelist)) { |
742755a1 | 1102 | err = migrate_pages(&pagelist, new_page_node, |
77f1fe6b | 1103 | (unsigned long)pm, 0, true); |
cf608ac1 MK |
1104 | if (err) |
1105 | putback_lru_pages(&pagelist); | |
1106 | } | |
742755a1 CL |
1107 | |
1108 | up_read(&mm->mmap_sem); | |
1109 | return err; | |
1110 | } | |
1111 | ||
5e9a0f02 BG |
1112 | /* |
1113 | * Migrate an array of page address onto an array of nodes and fill | |
1114 | * the corresponding array of status. | |
1115 | */ | |
1116 | static int do_pages_move(struct mm_struct *mm, struct task_struct *task, | |
1117 | unsigned long nr_pages, | |
1118 | const void __user * __user *pages, | |
1119 | const int __user *nodes, | |
1120 | int __user *status, int flags) | |
1121 | { | |
3140a227 | 1122 | struct page_to_node *pm; |
5e9a0f02 | 1123 | nodemask_t task_nodes; |
3140a227 BG |
1124 | unsigned long chunk_nr_pages; |
1125 | unsigned long chunk_start; | |
1126 | int err; | |
5e9a0f02 BG |
1127 | |
1128 | task_nodes = cpuset_mems_allowed(task); | |
1129 | ||
3140a227 BG |
1130 | err = -ENOMEM; |
1131 | pm = (struct page_to_node *)__get_free_page(GFP_KERNEL); | |
1132 | if (!pm) | |
5e9a0f02 | 1133 | goto out; |
35282a2d BG |
1134 | |
1135 | migrate_prep(); | |
1136 | ||
5e9a0f02 | 1137 | /* |
3140a227 BG |
1138 | * Store a chunk of page_to_node array in a page, |
1139 | * but keep the last one as a marker | |
5e9a0f02 | 1140 | */ |
3140a227 | 1141 | chunk_nr_pages = (PAGE_SIZE / sizeof(struct page_to_node)) - 1; |
5e9a0f02 | 1142 | |
3140a227 BG |
1143 | for (chunk_start = 0; |
1144 | chunk_start < nr_pages; | |
1145 | chunk_start += chunk_nr_pages) { | |
1146 | int j; | |
5e9a0f02 | 1147 | |
3140a227 BG |
1148 | if (chunk_start + chunk_nr_pages > nr_pages) |
1149 | chunk_nr_pages = nr_pages - chunk_start; | |
1150 | ||
1151 | /* fill the chunk pm with addrs and nodes from user-space */ | |
1152 | for (j = 0; j < chunk_nr_pages; j++) { | |
1153 | const void __user *p; | |
5e9a0f02 BG |
1154 | int node; |
1155 | ||
3140a227 BG |
1156 | err = -EFAULT; |
1157 | if (get_user(p, pages + j + chunk_start)) | |
1158 | goto out_pm; | |
1159 | pm[j].addr = (unsigned long) p; | |
1160 | ||
1161 | if (get_user(node, nodes + j + chunk_start)) | |
5e9a0f02 BG |
1162 | goto out_pm; |
1163 | ||
1164 | err = -ENODEV; | |
6f5a55f1 LT |
1165 | if (node < 0 || node >= MAX_NUMNODES) |
1166 | goto out_pm; | |
1167 | ||
5e9a0f02 BG |
1168 | if (!node_state(node, N_HIGH_MEMORY)) |
1169 | goto out_pm; | |
1170 | ||
1171 | err = -EACCES; | |
1172 | if (!node_isset(node, task_nodes)) | |
1173 | goto out_pm; | |
1174 | ||
3140a227 BG |
1175 | pm[j].node = node; |
1176 | } | |
1177 | ||
1178 | /* End marker for this chunk */ | |
1179 | pm[chunk_nr_pages].node = MAX_NUMNODES; | |
1180 | ||
1181 | /* Migrate this chunk */ | |
1182 | err = do_move_page_to_node_array(mm, pm, | |
1183 | flags & MPOL_MF_MOVE_ALL); | |
1184 | if (err < 0) | |
1185 | goto out_pm; | |
5e9a0f02 | 1186 | |
5e9a0f02 | 1187 | /* Return status information */ |
3140a227 BG |
1188 | for (j = 0; j < chunk_nr_pages; j++) |
1189 | if (put_user(pm[j].status, status + j + chunk_start)) { | |
5e9a0f02 | 1190 | err = -EFAULT; |
3140a227 BG |
1191 | goto out_pm; |
1192 | } | |
1193 | } | |
1194 | err = 0; | |
5e9a0f02 BG |
1195 | |
1196 | out_pm: | |
3140a227 | 1197 | free_page((unsigned long)pm); |
5e9a0f02 BG |
1198 | out: |
1199 | return err; | |
1200 | } | |
1201 | ||
742755a1 | 1202 | /* |
2f007e74 | 1203 | * Determine the nodes of an array of pages and store it in an array of status. |
742755a1 | 1204 | */ |
80bba129 BG |
1205 | static void do_pages_stat_array(struct mm_struct *mm, unsigned long nr_pages, |
1206 | const void __user **pages, int *status) | |
742755a1 | 1207 | { |
2f007e74 | 1208 | unsigned long i; |
2f007e74 | 1209 | |
742755a1 CL |
1210 | down_read(&mm->mmap_sem); |
1211 | ||
2f007e74 | 1212 | for (i = 0; i < nr_pages; i++) { |
80bba129 | 1213 | unsigned long addr = (unsigned long)(*pages); |
742755a1 CL |
1214 | struct vm_area_struct *vma; |
1215 | struct page *page; | |
c095adbc | 1216 | int err = -EFAULT; |
2f007e74 BG |
1217 | |
1218 | vma = find_vma(mm, addr); | |
70384dc6 | 1219 | if (!vma || addr < vma->vm_start) |
742755a1 CL |
1220 | goto set_status; |
1221 | ||
2f007e74 | 1222 | page = follow_page(vma, addr, 0); |
89f5b7da LT |
1223 | |
1224 | err = PTR_ERR(page); | |
1225 | if (IS_ERR(page)) | |
1226 | goto set_status; | |
1227 | ||
742755a1 CL |
1228 | err = -ENOENT; |
1229 | /* Use PageReserved to check for zero page */ | |
62b61f61 | 1230 | if (!page || PageReserved(page) || PageKsm(page)) |
742755a1 CL |
1231 | goto set_status; |
1232 | ||
1233 | err = page_to_nid(page); | |
1234 | set_status: | |
80bba129 BG |
1235 | *status = err; |
1236 | ||
1237 | pages++; | |
1238 | status++; | |
1239 | } | |
1240 | ||
1241 | up_read(&mm->mmap_sem); | |
1242 | } | |
1243 | ||
1244 | /* | |
1245 | * Determine the nodes of a user array of pages and store it in | |
1246 | * a user array of status. | |
1247 | */ | |
1248 | static int do_pages_stat(struct mm_struct *mm, unsigned long nr_pages, | |
1249 | const void __user * __user *pages, | |
1250 | int __user *status) | |
1251 | { | |
1252 | #define DO_PAGES_STAT_CHUNK_NR 16 | |
1253 | const void __user *chunk_pages[DO_PAGES_STAT_CHUNK_NR]; | |
1254 | int chunk_status[DO_PAGES_STAT_CHUNK_NR]; | |
80bba129 | 1255 | |
87b8d1ad PA |
1256 | while (nr_pages) { |
1257 | unsigned long chunk_nr; | |
80bba129 | 1258 | |
87b8d1ad PA |
1259 | chunk_nr = nr_pages; |
1260 | if (chunk_nr > DO_PAGES_STAT_CHUNK_NR) | |
1261 | chunk_nr = DO_PAGES_STAT_CHUNK_NR; | |
1262 | ||
1263 | if (copy_from_user(chunk_pages, pages, chunk_nr * sizeof(*chunk_pages))) | |
1264 | break; | |
80bba129 BG |
1265 | |
1266 | do_pages_stat_array(mm, chunk_nr, chunk_pages, chunk_status); | |
1267 | ||
87b8d1ad PA |
1268 | if (copy_to_user(status, chunk_status, chunk_nr * sizeof(*status))) |
1269 | break; | |
742755a1 | 1270 | |
87b8d1ad PA |
1271 | pages += chunk_nr; |
1272 | status += chunk_nr; | |
1273 | nr_pages -= chunk_nr; | |
1274 | } | |
1275 | return nr_pages ? -EFAULT : 0; | |
742755a1 CL |
1276 | } |
1277 | ||
1278 | /* | |
1279 | * Move a list of pages in the address space of the currently executing | |
1280 | * process. | |
1281 | */ | |
938bb9f5 HC |
1282 | SYSCALL_DEFINE6(move_pages, pid_t, pid, unsigned long, nr_pages, |
1283 | const void __user * __user *, pages, | |
1284 | const int __user *, nodes, | |
1285 | int __user *, status, int, flags) | |
742755a1 | 1286 | { |
c69e8d9c | 1287 | const struct cred *cred = current_cred(), *tcred; |
742755a1 | 1288 | struct task_struct *task; |
742755a1 | 1289 | struct mm_struct *mm; |
5e9a0f02 | 1290 | int err; |
742755a1 CL |
1291 | |
1292 | /* Check flags */ | |
1293 | if (flags & ~(MPOL_MF_MOVE|MPOL_MF_MOVE_ALL)) | |
1294 | return -EINVAL; | |
1295 | ||
1296 | if ((flags & MPOL_MF_MOVE_ALL) && !capable(CAP_SYS_NICE)) | |
1297 | return -EPERM; | |
1298 | ||
1299 | /* Find the mm_struct */ | |
a879bf58 | 1300 | rcu_read_lock(); |
228ebcbe | 1301 | task = pid ? find_task_by_vpid(pid) : current; |
742755a1 | 1302 | if (!task) { |
a879bf58 | 1303 | rcu_read_unlock(); |
742755a1 CL |
1304 | return -ESRCH; |
1305 | } | |
1306 | mm = get_task_mm(task); | |
a879bf58 | 1307 | rcu_read_unlock(); |
742755a1 CL |
1308 | |
1309 | if (!mm) | |
1310 | return -EINVAL; | |
1311 | ||
1312 | /* | |
1313 | * Check if this process has the right to modify the specified | |
1314 | * process. The right exists if the process has administrative | |
1315 | * capabilities, superuser privileges or the same | |
1316 | * userid as the target process. | |
1317 | */ | |
c69e8d9c DH |
1318 | rcu_read_lock(); |
1319 | tcred = __task_cred(task); | |
b6dff3ec DH |
1320 | if (cred->euid != tcred->suid && cred->euid != tcred->uid && |
1321 | cred->uid != tcred->suid && cred->uid != tcred->uid && | |
742755a1 | 1322 | !capable(CAP_SYS_NICE)) { |
c69e8d9c | 1323 | rcu_read_unlock(); |
742755a1 | 1324 | err = -EPERM; |
5e9a0f02 | 1325 | goto out; |
742755a1 | 1326 | } |
c69e8d9c | 1327 | rcu_read_unlock(); |
742755a1 | 1328 | |
86c3a764 DQ |
1329 | err = security_task_movememory(task); |
1330 | if (err) | |
5e9a0f02 | 1331 | goto out; |
86c3a764 | 1332 | |
5e9a0f02 BG |
1333 | if (nodes) { |
1334 | err = do_pages_move(mm, task, nr_pages, pages, nodes, status, | |
1335 | flags); | |
1336 | } else { | |
2f007e74 | 1337 | err = do_pages_stat(mm, nr_pages, pages, status); |
742755a1 CL |
1338 | } |
1339 | ||
742755a1 | 1340 | out: |
742755a1 CL |
1341 | mmput(mm); |
1342 | return err; | |
1343 | } | |
742755a1 | 1344 | |
7b2259b3 CL |
1345 | /* |
1346 | * Call migration functions in the vma_ops that may prepare | |
1347 | * memory in a vm for migration. migration functions may perform | |
1348 | * the migration for vmas that do not have an underlying page struct. | |
1349 | */ | |
1350 | int migrate_vmas(struct mm_struct *mm, const nodemask_t *to, | |
1351 | const nodemask_t *from, unsigned long flags) | |
1352 | { | |
1353 | struct vm_area_struct *vma; | |
1354 | int err = 0; | |
1355 | ||
1001c9fb | 1356 | for (vma = mm->mmap; vma && !err; vma = vma->vm_next) { |
7b2259b3 CL |
1357 | if (vma->vm_ops && vma->vm_ops->migrate) { |
1358 | err = vma->vm_ops->migrate(vma, to, from, flags); | |
1359 | if (err) | |
1360 | break; | |
1361 | } | |
1362 | } | |
1363 | return err; | |
1364 | } | |
83d1674a | 1365 | #endif |