]>
Commit | Line | Data |
---|---|---|
b20a3503 CL |
1 | /* |
2 | * Memory Migration functionality - linux/mm/migration.c | |
3 | * | |
4 | * Copyright (C) 2006 Silicon Graphics, Inc., Christoph Lameter | |
5 | * | |
6 | * Page migration was first developed in the context of the memory hotplug | |
7 | * project. The main authors of the migration code are: | |
8 | * | |
9 | * IWAMOTO Toshihiro <iwamoto@valinux.co.jp> | |
10 | * Hirokazu Takahashi <taka@valinux.co.jp> | |
11 | * Dave Hansen <haveblue@us.ibm.com> | |
cde53535 | 12 | * Christoph Lameter |
b20a3503 CL |
13 | */ |
14 | ||
15 | #include <linux/migrate.h> | |
b95f1b31 | 16 | #include <linux/export.h> |
b20a3503 | 17 | #include <linux/swap.h> |
0697212a | 18 | #include <linux/swapops.h> |
b20a3503 | 19 | #include <linux/pagemap.h> |
e23ca00b | 20 | #include <linux/buffer_head.h> |
b20a3503 | 21 | #include <linux/mm_inline.h> |
b488893a | 22 | #include <linux/nsproxy.h> |
b20a3503 | 23 | #include <linux/pagevec.h> |
e9995ef9 | 24 | #include <linux/ksm.h> |
b20a3503 CL |
25 | #include <linux/rmap.h> |
26 | #include <linux/topology.h> | |
27 | #include <linux/cpu.h> | |
28 | #include <linux/cpuset.h> | |
04e62a29 | 29 | #include <linux/writeback.h> |
742755a1 CL |
30 | #include <linux/mempolicy.h> |
31 | #include <linux/vmalloc.h> | |
86c3a764 | 32 | #include <linux/security.h> |
8a9f3ccd | 33 | #include <linux/memcontrol.h> |
4f5ca265 | 34 | #include <linux/syscalls.h> |
290408d4 | 35 | #include <linux/hugetlb.h> |
8e6ac7fa | 36 | #include <linux/hugetlb_cgroup.h> |
5a0e3ad6 | 37 | #include <linux/gfp.h> |
bf6bddf1 | 38 | #include <linux/balloon_compaction.h> |
b20a3503 | 39 | |
0d1836c3 MN |
40 | #include <asm/tlbflush.h> |
41 | ||
b20a3503 CL |
42 | #include "internal.h" |
43 | ||
b20a3503 | 44 | /* |
742755a1 | 45 | * migrate_prep() needs to be called before we start compiling a list of pages |
748446bb MG |
46 | * to be migrated using isolate_lru_page(). If scheduling work on other CPUs is |
47 | * undesirable, use migrate_prep_local() | |
b20a3503 CL |
48 | */ |
49 | int migrate_prep(void) | |
50 | { | |
b20a3503 CL |
51 | /* |
52 | * Clear the LRU lists so pages can be isolated. | |
53 | * Note that pages may be moved off the LRU after we have | |
54 | * drained them. Those pages will fail to migrate like other | |
55 | * pages that may be busy. | |
56 | */ | |
57 | lru_add_drain_all(); | |
58 | ||
59 | return 0; | |
60 | } | |
61 | ||
748446bb MG |
62 | /* Do the necessary work of migrate_prep but not if it involves other CPUs */ |
63 | int migrate_prep_local(void) | |
64 | { | |
65 | lru_add_drain(); | |
66 | ||
67 | return 0; | |
68 | } | |
69 | ||
b20a3503 | 70 | /* |
894bc310 LS |
71 | * Add isolated pages on the list back to the LRU under page lock |
72 | * to avoid leaking evictable pages back onto unevictable list. | |
b20a3503 | 73 | */ |
e13861d8 | 74 | void putback_lru_pages(struct list_head *l) |
b20a3503 CL |
75 | { |
76 | struct page *page; | |
77 | struct page *page2; | |
b20a3503 | 78 | |
5733c7d1 RA |
79 | list_for_each_entry_safe(page, page2, l, lru) { |
80 | list_del(&page->lru); | |
81 | dec_zone_page_state(page, NR_ISOLATED_ANON + | |
82 | page_is_file_cache(page)); | |
83 | putback_lru_page(page); | |
84 | } | |
85 | } | |
86 | ||
87 | /* | |
88 | * Put previously isolated pages back onto the appropriate lists | |
89 | * from where they were once taken off for compaction/migration. | |
90 | * | |
91 | * This function shall be used instead of putback_lru_pages(), | |
92 | * whenever the isolated pageset has been built by isolate_migratepages_range() | |
93 | */ | |
94 | void putback_movable_pages(struct list_head *l) | |
95 | { | |
96 | struct page *page; | |
97 | struct page *page2; | |
98 | ||
b20a3503 | 99 | list_for_each_entry_safe(page, page2, l, lru) { |
e24f0b8f | 100 | list_del(&page->lru); |
a731286d | 101 | dec_zone_page_state(page, NR_ISOLATED_ANON + |
6c0b1351 | 102 | page_is_file_cache(page)); |
bf6bddf1 RA |
103 | if (unlikely(balloon_page_movable(page))) |
104 | balloon_page_putback(page); | |
105 | else | |
106 | putback_lru_page(page); | |
b20a3503 | 107 | } |
b20a3503 CL |
108 | } |
109 | ||
0697212a CL |
110 | /* |
111 | * Restore a potential migration pte to a working pte entry | |
112 | */ | |
e9995ef9 HD |
113 | static int remove_migration_pte(struct page *new, struct vm_area_struct *vma, |
114 | unsigned long addr, void *old) | |
0697212a CL |
115 | { |
116 | struct mm_struct *mm = vma->vm_mm; | |
117 | swp_entry_t entry; | |
0697212a CL |
118 | pmd_t *pmd; |
119 | pte_t *ptep, pte; | |
120 | spinlock_t *ptl; | |
121 | ||
290408d4 NH |
122 | if (unlikely(PageHuge(new))) { |
123 | ptep = huge_pte_offset(mm, addr); | |
124 | if (!ptep) | |
125 | goto out; | |
126 | ptl = &mm->page_table_lock; | |
127 | } else { | |
6219049a BL |
128 | pmd = mm_find_pmd(mm, addr); |
129 | if (!pmd) | |
290408d4 | 130 | goto out; |
500d65d4 AA |
131 | if (pmd_trans_huge(*pmd)) |
132 | goto out; | |
0697212a | 133 | |
290408d4 | 134 | ptep = pte_offset_map(pmd, addr); |
0697212a | 135 | |
486cf46f HD |
136 | /* |
137 | * Peek to check is_swap_pte() before taking ptlock? No, we | |
138 | * can race mremap's move_ptes(), which skips anon_vma lock. | |
139 | */ | |
290408d4 NH |
140 | |
141 | ptl = pte_lockptr(mm, pmd); | |
142 | } | |
0697212a | 143 | |
0697212a CL |
144 | spin_lock(ptl); |
145 | pte = *ptep; | |
146 | if (!is_swap_pte(pte)) | |
e9995ef9 | 147 | goto unlock; |
0697212a CL |
148 | |
149 | entry = pte_to_swp_entry(pte); | |
150 | ||
e9995ef9 HD |
151 | if (!is_migration_entry(entry) || |
152 | migration_entry_to_page(entry) != old) | |
153 | goto unlock; | |
0697212a | 154 | |
0697212a CL |
155 | get_page(new); |
156 | pte = pte_mkold(mk_pte(new, vma->vm_page_prot)); | |
157 | if (is_write_migration_entry(entry)) | |
158 | pte = pte_mkwrite(pte); | |
3ef8fd7f | 159 | #ifdef CONFIG_HUGETLB_PAGE |
290408d4 NH |
160 | if (PageHuge(new)) |
161 | pte = pte_mkhuge(pte); | |
3ef8fd7f | 162 | #endif |
97ee0524 | 163 | flush_cache_page(vma, addr, pte_pfn(pte)); |
0697212a | 164 | set_pte_at(mm, addr, ptep, pte); |
04e62a29 | 165 | |
290408d4 NH |
166 | if (PageHuge(new)) { |
167 | if (PageAnon(new)) | |
168 | hugepage_add_anon_rmap(new, vma, addr); | |
169 | else | |
170 | page_dup_rmap(new); | |
171 | } else if (PageAnon(new)) | |
04e62a29 CL |
172 | page_add_anon_rmap(new, vma, addr); |
173 | else | |
174 | page_add_file_rmap(new); | |
175 | ||
176 | /* No need to invalidate - it was non-present before */ | |
4b3073e1 | 177 | update_mmu_cache(vma, addr, ptep); |
e9995ef9 | 178 | unlock: |
0697212a | 179 | pte_unmap_unlock(ptep, ptl); |
e9995ef9 HD |
180 | out: |
181 | return SWAP_AGAIN; | |
0697212a CL |
182 | } |
183 | ||
04e62a29 CL |
184 | /* |
185 | * Get rid of all migration entries and replace them by | |
186 | * references to the indicated page. | |
187 | */ | |
188 | static void remove_migration_ptes(struct page *old, struct page *new) | |
189 | { | |
e9995ef9 | 190 | rmap_walk(new, remove_migration_pte, old); |
04e62a29 CL |
191 | } |
192 | ||
0697212a CL |
193 | /* |
194 | * Something used the pte of a page under migration. We need to | |
195 | * get to the page and wait until migration is finished. | |
196 | * When we return from this function the fault will be retried. | |
0697212a CL |
197 | */ |
198 | void migration_entry_wait(struct mm_struct *mm, pmd_t *pmd, | |
199 | unsigned long address) | |
200 | { | |
201 | pte_t *ptep, pte; | |
202 | spinlock_t *ptl; | |
203 | swp_entry_t entry; | |
204 | struct page *page; | |
205 | ||
206 | ptep = pte_offset_map_lock(mm, pmd, address, &ptl); | |
207 | pte = *ptep; | |
208 | if (!is_swap_pte(pte)) | |
209 | goto out; | |
210 | ||
211 | entry = pte_to_swp_entry(pte); | |
212 | if (!is_migration_entry(entry)) | |
213 | goto out; | |
214 | ||
215 | page = migration_entry_to_page(entry); | |
216 | ||
e286781d NP |
217 | /* |
218 | * Once radix-tree replacement of page migration started, page_count | |
219 | * *must* be zero. And, we don't want to call wait_on_page_locked() | |
220 | * against a page without get_page(). | |
221 | * So, we use get_page_unless_zero(), here. Even failed, page fault | |
222 | * will occur again. | |
223 | */ | |
224 | if (!get_page_unless_zero(page)) | |
225 | goto out; | |
0697212a CL |
226 | pte_unmap_unlock(ptep, ptl); |
227 | wait_on_page_locked(page); | |
228 | put_page(page); | |
229 | return; | |
230 | out: | |
231 | pte_unmap_unlock(ptep, ptl); | |
232 | } | |
233 | ||
b969c4ab MG |
234 | #ifdef CONFIG_BLOCK |
235 | /* Returns true if all buffers are successfully locked */ | |
a6bc32b8 MG |
236 | static bool buffer_migrate_lock_buffers(struct buffer_head *head, |
237 | enum migrate_mode mode) | |
b969c4ab MG |
238 | { |
239 | struct buffer_head *bh = head; | |
240 | ||
241 | /* Simple case, sync compaction */ | |
a6bc32b8 | 242 | if (mode != MIGRATE_ASYNC) { |
b969c4ab MG |
243 | do { |
244 | get_bh(bh); | |
245 | lock_buffer(bh); | |
246 | bh = bh->b_this_page; | |
247 | ||
248 | } while (bh != head); | |
249 | ||
250 | return true; | |
251 | } | |
252 | ||
253 | /* async case, we cannot block on lock_buffer so use trylock_buffer */ | |
254 | do { | |
255 | get_bh(bh); | |
256 | if (!trylock_buffer(bh)) { | |
257 | /* | |
258 | * We failed to lock the buffer and cannot stall in | |
259 | * async migration. Release the taken locks | |
260 | */ | |
261 | struct buffer_head *failed_bh = bh; | |
262 | put_bh(failed_bh); | |
263 | bh = head; | |
264 | while (bh != failed_bh) { | |
265 | unlock_buffer(bh); | |
266 | put_bh(bh); | |
267 | bh = bh->b_this_page; | |
268 | } | |
269 | return false; | |
270 | } | |
271 | ||
272 | bh = bh->b_this_page; | |
273 | } while (bh != head); | |
274 | return true; | |
275 | } | |
276 | #else | |
277 | static inline bool buffer_migrate_lock_buffers(struct buffer_head *head, | |
a6bc32b8 | 278 | enum migrate_mode mode) |
b969c4ab MG |
279 | { |
280 | return true; | |
281 | } | |
282 | #endif /* CONFIG_BLOCK */ | |
283 | ||
b20a3503 | 284 | /* |
c3fcf8a5 | 285 | * Replace the page in the mapping. |
5b5c7120 CL |
286 | * |
287 | * The number of remaining references must be: | |
288 | * 1 for anonymous pages without a mapping | |
289 | * 2 for pages with a mapping | |
266cf658 | 290 | * 3 for pages with a mapping and PagePrivate/PagePrivate2 set. |
b20a3503 | 291 | */ |
2d1db3b1 | 292 | static int migrate_page_move_mapping(struct address_space *mapping, |
b969c4ab | 293 | struct page *newpage, struct page *page, |
a6bc32b8 | 294 | struct buffer_head *head, enum migrate_mode mode) |
b20a3503 | 295 | { |
e286781d | 296 | int expected_count; |
7cf9c2c7 | 297 | void **pslot; |
b20a3503 | 298 | |
6c5240ae | 299 | if (!mapping) { |
0e8c7d0f | 300 | /* Anonymous page without mapping */ |
6c5240ae CL |
301 | if (page_count(page) != 1) |
302 | return -EAGAIN; | |
78bd5209 | 303 | return MIGRATEPAGE_SUCCESS; |
6c5240ae CL |
304 | } |
305 | ||
19fd6231 | 306 | spin_lock_irq(&mapping->tree_lock); |
b20a3503 | 307 | |
7cf9c2c7 NP |
308 | pslot = radix_tree_lookup_slot(&mapping->page_tree, |
309 | page_index(page)); | |
b20a3503 | 310 | |
edcf4748 | 311 | expected_count = 2 + page_has_private(page); |
e286781d | 312 | if (page_count(page) != expected_count || |
29c1f677 | 313 | radix_tree_deref_slot_protected(pslot, &mapping->tree_lock) != page) { |
19fd6231 | 314 | spin_unlock_irq(&mapping->tree_lock); |
e23ca00b | 315 | return -EAGAIN; |
b20a3503 CL |
316 | } |
317 | ||
e286781d | 318 | if (!page_freeze_refs(page, expected_count)) { |
19fd6231 | 319 | spin_unlock_irq(&mapping->tree_lock); |
e286781d NP |
320 | return -EAGAIN; |
321 | } | |
322 | ||
b969c4ab MG |
323 | /* |
324 | * In the async migration case of moving a page with buffers, lock the | |
325 | * buffers using trylock before the mapping is moved. If the mapping | |
326 | * was moved, we later failed to lock the buffers and could not move | |
327 | * the mapping back due to an elevated page count, we would have to | |
328 | * block waiting on other references to be dropped. | |
329 | */ | |
a6bc32b8 MG |
330 | if (mode == MIGRATE_ASYNC && head && |
331 | !buffer_migrate_lock_buffers(head, mode)) { | |
b969c4ab MG |
332 | page_unfreeze_refs(page, expected_count); |
333 | spin_unlock_irq(&mapping->tree_lock); | |
334 | return -EAGAIN; | |
335 | } | |
336 | ||
b20a3503 CL |
337 | /* |
338 | * Now we know that no one else is looking at the page. | |
b20a3503 | 339 | */ |
7cf9c2c7 | 340 | get_page(newpage); /* add cache reference */ |
b20a3503 CL |
341 | if (PageSwapCache(page)) { |
342 | SetPageSwapCache(newpage); | |
343 | set_page_private(newpage, page_private(page)); | |
344 | } | |
345 | ||
7cf9c2c7 NP |
346 | radix_tree_replace_slot(pslot, newpage); |
347 | ||
348 | /* | |
937a94c9 JG |
349 | * Drop cache reference from old page by unfreezing |
350 | * to one less reference. | |
7cf9c2c7 NP |
351 | * We know this isn't the last reference. |
352 | */ | |
937a94c9 | 353 | page_unfreeze_refs(page, expected_count - 1); |
7cf9c2c7 | 354 | |
0e8c7d0f CL |
355 | /* |
356 | * If moved to a different zone then also account | |
357 | * the page for that zone. Other VM counters will be | |
358 | * taken care of when we establish references to the | |
359 | * new page and drop references to the old page. | |
360 | * | |
361 | * Note that anonymous pages are accounted for | |
362 | * via NR_FILE_PAGES and NR_ANON_PAGES if they | |
363 | * are mapped to swap space. | |
364 | */ | |
365 | __dec_zone_page_state(page, NR_FILE_PAGES); | |
366 | __inc_zone_page_state(newpage, NR_FILE_PAGES); | |
99a15e21 | 367 | if (!PageSwapCache(page) && PageSwapBacked(page)) { |
4b02108a KM |
368 | __dec_zone_page_state(page, NR_SHMEM); |
369 | __inc_zone_page_state(newpage, NR_SHMEM); | |
370 | } | |
19fd6231 | 371 | spin_unlock_irq(&mapping->tree_lock); |
b20a3503 | 372 | |
78bd5209 | 373 | return MIGRATEPAGE_SUCCESS; |
b20a3503 | 374 | } |
b20a3503 | 375 | |
290408d4 NH |
376 | /* |
377 | * The expected number of remaining references is the same as that | |
378 | * of migrate_page_move_mapping(). | |
379 | */ | |
380 | int migrate_huge_page_move_mapping(struct address_space *mapping, | |
381 | struct page *newpage, struct page *page) | |
382 | { | |
383 | int expected_count; | |
384 | void **pslot; | |
385 | ||
386 | if (!mapping) { | |
387 | if (page_count(page) != 1) | |
388 | return -EAGAIN; | |
78bd5209 | 389 | return MIGRATEPAGE_SUCCESS; |
290408d4 NH |
390 | } |
391 | ||
392 | spin_lock_irq(&mapping->tree_lock); | |
393 | ||
394 | pslot = radix_tree_lookup_slot(&mapping->page_tree, | |
395 | page_index(page)); | |
396 | ||
397 | expected_count = 2 + page_has_private(page); | |
398 | if (page_count(page) != expected_count || | |
29c1f677 | 399 | radix_tree_deref_slot_protected(pslot, &mapping->tree_lock) != page) { |
290408d4 NH |
400 | spin_unlock_irq(&mapping->tree_lock); |
401 | return -EAGAIN; | |
402 | } | |
403 | ||
404 | if (!page_freeze_refs(page, expected_count)) { | |
405 | spin_unlock_irq(&mapping->tree_lock); | |
406 | return -EAGAIN; | |
407 | } | |
408 | ||
409 | get_page(newpage); | |
410 | ||
411 | radix_tree_replace_slot(pslot, newpage); | |
412 | ||
937a94c9 | 413 | page_unfreeze_refs(page, expected_count - 1); |
290408d4 NH |
414 | |
415 | spin_unlock_irq(&mapping->tree_lock); | |
78bd5209 | 416 | return MIGRATEPAGE_SUCCESS; |
290408d4 NH |
417 | } |
418 | ||
b20a3503 CL |
419 | /* |
420 | * Copy the page to its new location | |
421 | */ | |
290408d4 | 422 | void migrate_page_copy(struct page *newpage, struct page *page) |
b20a3503 | 423 | { |
290408d4 NH |
424 | if (PageHuge(page)) |
425 | copy_huge_page(newpage, page); | |
426 | else | |
427 | copy_highpage(newpage, page); | |
b20a3503 CL |
428 | |
429 | if (PageError(page)) | |
430 | SetPageError(newpage); | |
431 | if (PageReferenced(page)) | |
432 | SetPageReferenced(newpage); | |
433 | if (PageUptodate(page)) | |
434 | SetPageUptodate(newpage); | |
894bc310 LS |
435 | if (TestClearPageActive(page)) { |
436 | VM_BUG_ON(PageUnevictable(page)); | |
b20a3503 | 437 | SetPageActive(newpage); |
418b27ef LS |
438 | } else if (TestClearPageUnevictable(page)) |
439 | SetPageUnevictable(newpage); | |
b20a3503 CL |
440 | if (PageChecked(page)) |
441 | SetPageChecked(newpage); | |
442 | if (PageMappedToDisk(page)) | |
443 | SetPageMappedToDisk(newpage); | |
444 | ||
445 | if (PageDirty(page)) { | |
446 | clear_page_dirty_for_io(page); | |
3a902c5f NP |
447 | /* |
448 | * Want to mark the page and the radix tree as dirty, and | |
449 | * redo the accounting that clear_page_dirty_for_io undid, | |
450 | * but we can't use set_page_dirty because that function | |
451 | * is actually a signal that all of the page has become dirty. | |
25985edc | 452 | * Whereas only part of our page may be dirty. |
3a902c5f | 453 | */ |
752dc185 HD |
454 | if (PageSwapBacked(page)) |
455 | SetPageDirty(newpage); | |
456 | else | |
457 | __set_page_dirty_nobuffers(newpage); | |
b20a3503 CL |
458 | } |
459 | ||
b291f000 | 460 | mlock_migrate_page(newpage, page); |
e9995ef9 | 461 | ksm_migrate_page(newpage, page); |
b291f000 | 462 | |
b20a3503 | 463 | ClearPageSwapCache(page); |
b20a3503 CL |
464 | ClearPagePrivate(page); |
465 | set_page_private(page, 0); | |
b20a3503 CL |
466 | |
467 | /* | |
468 | * If any waiters have accumulated on the new page then | |
469 | * wake them up. | |
470 | */ | |
471 | if (PageWriteback(newpage)) | |
472 | end_page_writeback(newpage); | |
473 | } | |
b20a3503 | 474 | |
1d8b85cc CL |
475 | /************************************************************ |
476 | * Migration functions | |
477 | ***********************************************************/ | |
478 | ||
479 | /* Always fail migration. Used for mappings that are not movable */ | |
2d1db3b1 CL |
480 | int fail_migrate_page(struct address_space *mapping, |
481 | struct page *newpage, struct page *page) | |
1d8b85cc CL |
482 | { |
483 | return -EIO; | |
484 | } | |
485 | EXPORT_SYMBOL(fail_migrate_page); | |
486 | ||
b20a3503 CL |
487 | /* |
488 | * Common logic to directly migrate a single page suitable for | |
266cf658 | 489 | * pages that do not use PagePrivate/PagePrivate2. |
b20a3503 CL |
490 | * |
491 | * Pages are locked upon entry and exit. | |
492 | */ | |
2d1db3b1 | 493 | int migrate_page(struct address_space *mapping, |
a6bc32b8 MG |
494 | struct page *newpage, struct page *page, |
495 | enum migrate_mode mode) | |
b20a3503 CL |
496 | { |
497 | int rc; | |
498 | ||
499 | BUG_ON(PageWriteback(page)); /* Writeback must be complete */ | |
500 | ||
a6bc32b8 | 501 | rc = migrate_page_move_mapping(mapping, newpage, page, NULL, mode); |
b20a3503 | 502 | |
78bd5209 | 503 | if (rc != MIGRATEPAGE_SUCCESS) |
b20a3503 CL |
504 | return rc; |
505 | ||
506 | migrate_page_copy(newpage, page); | |
78bd5209 | 507 | return MIGRATEPAGE_SUCCESS; |
b20a3503 CL |
508 | } |
509 | EXPORT_SYMBOL(migrate_page); | |
510 | ||
9361401e | 511 | #ifdef CONFIG_BLOCK |
1d8b85cc CL |
512 | /* |
513 | * Migration function for pages with buffers. This function can only be used | |
514 | * if the underlying filesystem guarantees that no other references to "page" | |
515 | * exist. | |
516 | */ | |
2d1db3b1 | 517 | int buffer_migrate_page(struct address_space *mapping, |
a6bc32b8 | 518 | struct page *newpage, struct page *page, enum migrate_mode mode) |
1d8b85cc | 519 | { |
1d8b85cc CL |
520 | struct buffer_head *bh, *head; |
521 | int rc; | |
522 | ||
1d8b85cc | 523 | if (!page_has_buffers(page)) |
a6bc32b8 | 524 | return migrate_page(mapping, newpage, page, mode); |
1d8b85cc CL |
525 | |
526 | head = page_buffers(page); | |
527 | ||
a6bc32b8 | 528 | rc = migrate_page_move_mapping(mapping, newpage, page, head, mode); |
1d8b85cc | 529 | |
78bd5209 | 530 | if (rc != MIGRATEPAGE_SUCCESS) |
1d8b85cc CL |
531 | return rc; |
532 | ||
b969c4ab MG |
533 | /* |
534 | * In the async case, migrate_page_move_mapping locked the buffers | |
535 | * with an IRQ-safe spinlock held. In the sync case, the buffers | |
536 | * need to be locked now | |
537 | */ | |
a6bc32b8 MG |
538 | if (mode != MIGRATE_ASYNC) |
539 | BUG_ON(!buffer_migrate_lock_buffers(head, mode)); | |
1d8b85cc CL |
540 | |
541 | ClearPagePrivate(page); | |
542 | set_page_private(newpage, page_private(page)); | |
543 | set_page_private(page, 0); | |
544 | put_page(page); | |
545 | get_page(newpage); | |
546 | ||
547 | bh = head; | |
548 | do { | |
549 | set_bh_page(bh, newpage, bh_offset(bh)); | |
550 | bh = bh->b_this_page; | |
551 | ||
552 | } while (bh != head); | |
553 | ||
554 | SetPagePrivate(newpage); | |
555 | ||
556 | migrate_page_copy(newpage, page); | |
557 | ||
558 | bh = head; | |
559 | do { | |
560 | unlock_buffer(bh); | |
561 | put_bh(bh); | |
562 | bh = bh->b_this_page; | |
563 | ||
564 | } while (bh != head); | |
565 | ||
78bd5209 | 566 | return MIGRATEPAGE_SUCCESS; |
1d8b85cc CL |
567 | } |
568 | EXPORT_SYMBOL(buffer_migrate_page); | |
9361401e | 569 | #endif |
1d8b85cc | 570 | |
04e62a29 CL |
571 | /* |
572 | * Writeback a page to clean the dirty state | |
573 | */ | |
574 | static int writeout(struct address_space *mapping, struct page *page) | |
8351a6e4 | 575 | { |
04e62a29 CL |
576 | struct writeback_control wbc = { |
577 | .sync_mode = WB_SYNC_NONE, | |
578 | .nr_to_write = 1, | |
579 | .range_start = 0, | |
580 | .range_end = LLONG_MAX, | |
04e62a29 CL |
581 | .for_reclaim = 1 |
582 | }; | |
583 | int rc; | |
584 | ||
585 | if (!mapping->a_ops->writepage) | |
586 | /* No write method for the address space */ | |
587 | return -EINVAL; | |
588 | ||
589 | if (!clear_page_dirty_for_io(page)) | |
590 | /* Someone else already triggered a write */ | |
591 | return -EAGAIN; | |
592 | ||
8351a6e4 | 593 | /* |
04e62a29 CL |
594 | * A dirty page may imply that the underlying filesystem has |
595 | * the page on some queue. So the page must be clean for | |
596 | * migration. Writeout may mean we loose the lock and the | |
597 | * page state is no longer what we checked for earlier. | |
598 | * At this point we know that the migration attempt cannot | |
599 | * be successful. | |
8351a6e4 | 600 | */ |
04e62a29 | 601 | remove_migration_ptes(page, page); |
8351a6e4 | 602 | |
04e62a29 | 603 | rc = mapping->a_ops->writepage(page, &wbc); |
8351a6e4 | 604 | |
04e62a29 CL |
605 | if (rc != AOP_WRITEPAGE_ACTIVATE) |
606 | /* unlocked. Relock */ | |
607 | lock_page(page); | |
608 | ||
bda8550d | 609 | return (rc < 0) ? -EIO : -EAGAIN; |
04e62a29 CL |
610 | } |
611 | ||
612 | /* | |
613 | * Default handling if a filesystem does not provide a migration function. | |
614 | */ | |
615 | static int fallback_migrate_page(struct address_space *mapping, | |
a6bc32b8 | 616 | struct page *newpage, struct page *page, enum migrate_mode mode) |
04e62a29 | 617 | { |
b969c4ab | 618 | if (PageDirty(page)) { |
a6bc32b8 MG |
619 | /* Only writeback pages in full synchronous migration */ |
620 | if (mode != MIGRATE_SYNC) | |
b969c4ab | 621 | return -EBUSY; |
04e62a29 | 622 | return writeout(mapping, page); |
b969c4ab | 623 | } |
8351a6e4 CL |
624 | |
625 | /* | |
626 | * Buffers may be managed in a filesystem specific way. | |
627 | * We must have no buffers or drop them. | |
628 | */ | |
266cf658 | 629 | if (page_has_private(page) && |
8351a6e4 CL |
630 | !try_to_release_page(page, GFP_KERNEL)) |
631 | return -EAGAIN; | |
632 | ||
a6bc32b8 | 633 | return migrate_page(mapping, newpage, page, mode); |
8351a6e4 CL |
634 | } |
635 | ||
e24f0b8f CL |
636 | /* |
637 | * Move a page to a newly allocated page | |
638 | * The page is locked and all ptes have been successfully removed. | |
639 | * | |
640 | * The new page will have replaced the old page if this function | |
641 | * is successful. | |
894bc310 LS |
642 | * |
643 | * Return value: | |
644 | * < 0 - error code | |
78bd5209 | 645 | * MIGRATEPAGE_SUCCESS - success |
e24f0b8f | 646 | */ |
3fe2011f | 647 | static int move_to_new_page(struct page *newpage, struct page *page, |
a6bc32b8 | 648 | int remap_swapcache, enum migrate_mode mode) |
e24f0b8f CL |
649 | { |
650 | struct address_space *mapping; | |
651 | int rc; | |
652 | ||
653 | /* | |
654 | * Block others from accessing the page when we get around to | |
655 | * establishing additional references. We are the only one | |
656 | * holding a reference to the new page at this point. | |
657 | */ | |
529ae9aa | 658 | if (!trylock_page(newpage)) |
e24f0b8f CL |
659 | BUG(); |
660 | ||
661 | /* Prepare mapping for the new page.*/ | |
662 | newpage->index = page->index; | |
663 | newpage->mapping = page->mapping; | |
b2e18538 RR |
664 | if (PageSwapBacked(page)) |
665 | SetPageSwapBacked(newpage); | |
e24f0b8f CL |
666 | |
667 | mapping = page_mapping(page); | |
668 | if (!mapping) | |
a6bc32b8 | 669 | rc = migrate_page(mapping, newpage, page, mode); |
b969c4ab | 670 | else if (mapping->a_ops->migratepage) |
e24f0b8f | 671 | /* |
b969c4ab MG |
672 | * Most pages have a mapping and most filesystems provide a |
673 | * migratepage callback. Anonymous pages are part of swap | |
674 | * space which also has its own migratepage callback. This | |
675 | * is the most common path for page migration. | |
e24f0b8f | 676 | */ |
b969c4ab | 677 | rc = mapping->a_ops->migratepage(mapping, |
a6bc32b8 | 678 | newpage, page, mode); |
b969c4ab | 679 | else |
a6bc32b8 | 680 | rc = fallback_migrate_page(mapping, newpage, page, mode); |
e24f0b8f | 681 | |
78bd5209 | 682 | if (rc != MIGRATEPAGE_SUCCESS) { |
e24f0b8f | 683 | newpage->mapping = NULL; |
3fe2011f MG |
684 | } else { |
685 | if (remap_swapcache) | |
686 | remove_migration_ptes(page, newpage); | |
35512eca | 687 | page->mapping = NULL; |
3fe2011f | 688 | } |
e24f0b8f CL |
689 | |
690 | unlock_page(newpage); | |
691 | ||
692 | return rc; | |
693 | } | |
694 | ||
0dabec93 | 695 | static int __unmap_and_move(struct page *page, struct page *newpage, |
a6bc32b8 | 696 | int force, bool offlining, enum migrate_mode mode) |
e24f0b8f | 697 | { |
0dabec93 | 698 | int rc = -EAGAIN; |
3fe2011f | 699 | int remap_swapcache = 1; |
56039efa | 700 | struct mem_cgroup *mem; |
3f6c8272 | 701 | struct anon_vma *anon_vma = NULL; |
95a402c3 | 702 | |
529ae9aa | 703 | if (!trylock_page(page)) { |
a6bc32b8 | 704 | if (!force || mode == MIGRATE_ASYNC) |
0dabec93 | 705 | goto out; |
3e7d3449 MG |
706 | |
707 | /* | |
708 | * It's not safe for direct compaction to call lock_page. | |
709 | * For example, during page readahead pages are added locked | |
710 | * to the LRU. Later, when the IO completes the pages are | |
711 | * marked uptodate and unlocked. However, the queueing | |
712 | * could be merging multiple pages for one bio (e.g. | |
713 | * mpage_readpages). If an allocation happens for the | |
714 | * second or third page, the process can end up locking | |
715 | * the same page twice and deadlocking. Rather than | |
716 | * trying to be clever about what pages can be locked, | |
717 | * avoid the use of lock_page for direct compaction | |
718 | * altogether. | |
719 | */ | |
720 | if (current->flags & PF_MEMALLOC) | |
0dabec93 | 721 | goto out; |
3e7d3449 | 722 | |
e24f0b8f CL |
723 | lock_page(page); |
724 | } | |
725 | ||
62b61f61 HD |
726 | /* |
727 | * Only memory hotplug's offline_pages() caller has locked out KSM, | |
728 | * and can safely migrate a KSM page. The other cases have skipped | |
729 | * PageKsm along with PageReserved - but it is only now when we have | |
730 | * the page lock that we can be certain it will not go KSM beneath us | |
731 | * (KSM will not upgrade a page from PageAnon to PageKsm when it sees | |
732 | * its pagecount raised, but only here do we take the page lock which | |
733 | * serializes that). | |
734 | */ | |
735 | if (PageKsm(page) && !offlining) { | |
736 | rc = -EBUSY; | |
737 | goto unlock; | |
738 | } | |
739 | ||
01b1ae63 | 740 | /* charge against new page */ |
0030f535 | 741 | mem_cgroup_prepare_migration(page, newpage, &mem); |
01b1ae63 | 742 | |
e24f0b8f | 743 | if (PageWriteback(page)) { |
11bc82d6 | 744 | /* |
a6bc32b8 MG |
745 | * Only in the case of a full syncronous migration is it |
746 | * necessary to wait for PageWriteback. In the async case, | |
747 | * the retry loop is too short and in the sync-light case, | |
748 | * the overhead of stalling is too much | |
11bc82d6 | 749 | */ |
a6bc32b8 | 750 | if (mode != MIGRATE_SYNC) { |
11bc82d6 AA |
751 | rc = -EBUSY; |
752 | goto uncharge; | |
753 | } | |
754 | if (!force) | |
01b1ae63 | 755 | goto uncharge; |
e24f0b8f CL |
756 | wait_on_page_writeback(page); |
757 | } | |
e24f0b8f | 758 | /* |
dc386d4d KH |
759 | * By try_to_unmap(), page->mapcount goes down to 0 here. In this case, |
760 | * we cannot notice that anon_vma is freed while we migrates a page. | |
1ce82b69 | 761 | * This get_anon_vma() delays freeing anon_vma pointer until the end |
dc386d4d | 762 | * of migration. File cache pages are no problem because of page_lock() |
989f89c5 KH |
763 | * File Caches may use write_page() or lock_page() in migration, then, |
764 | * just care Anon page here. | |
dc386d4d | 765 | */ |
989f89c5 | 766 | if (PageAnon(page)) { |
1ce82b69 HD |
767 | /* |
768 | * Only page_lock_anon_vma() understands the subtleties of | |
769 | * getting a hold on an anon_vma from outside one of its mms. | |
770 | */ | |
746b18d4 | 771 | anon_vma = page_get_anon_vma(page); |
1ce82b69 HD |
772 | if (anon_vma) { |
773 | /* | |
746b18d4 | 774 | * Anon page |
1ce82b69 | 775 | */ |
1ce82b69 | 776 | } else if (PageSwapCache(page)) { |
3fe2011f MG |
777 | /* |
778 | * We cannot be sure that the anon_vma of an unmapped | |
779 | * swapcache page is safe to use because we don't | |
780 | * know in advance if the VMA that this page belonged | |
781 | * to still exists. If the VMA and others sharing the | |
782 | * data have been freed, then the anon_vma could | |
783 | * already be invalid. | |
784 | * | |
785 | * To avoid this possibility, swapcache pages get | |
786 | * migrated but are not remapped when migration | |
787 | * completes | |
788 | */ | |
789 | remap_swapcache = 0; | |
790 | } else { | |
1ce82b69 | 791 | goto uncharge; |
3fe2011f | 792 | } |
989f89c5 | 793 | } |
62e1c553 | 794 | |
bf6bddf1 RA |
795 | if (unlikely(balloon_page_movable(page))) { |
796 | /* | |
797 | * A ballooned page does not need any special attention from | |
798 | * physical to virtual reverse mapping procedures. | |
799 | * Skip any attempt to unmap PTEs or to remap swap cache, | |
800 | * in order to avoid burning cycles at rmap level, and perform | |
801 | * the page migration right away (proteced by page lock). | |
802 | */ | |
803 | rc = balloon_page_migrate(newpage, page, mode); | |
804 | goto uncharge; | |
805 | } | |
806 | ||
dc386d4d | 807 | /* |
62e1c553 SL |
808 | * Corner case handling: |
809 | * 1. When a new swap-cache page is read into, it is added to the LRU | |
810 | * and treated as swapcache but it has no rmap yet. | |
811 | * Calling try_to_unmap() against a page->mapping==NULL page will | |
812 | * trigger a BUG. So handle it here. | |
813 | * 2. An orphaned page (see truncate_complete_page) might have | |
814 | * fs-private metadata. The page can be picked up due to memory | |
815 | * offlining. Everywhere else except page reclaim, the page is | |
816 | * invisible to the vm, so the page can not be migrated. So try to | |
817 | * free the metadata, so the page can be freed. | |
e24f0b8f | 818 | */ |
62e1c553 | 819 | if (!page->mapping) { |
1ce82b69 HD |
820 | VM_BUG_ON(PageAnon(page)); |
821 | if (page_has_private(page)) { | |
62e1c553 | 822 | try_to_free_buffers(page); |
1ce82b69 | 823 | goto uncharge; |
62e1c553 | 824 | } |
abfc3488 | 825 | goto skip_unmap; |
62e1c553 SL |
826 | } |
827 | ||
dc386d4d | 828 | /* Establish migration ptes or remove ptes */ |
14fa31b8 | 829 | try_to_unmap(page, TTU_MIGRATION|TTU_IGNORE_MLOCK|TTU_IGNORE_ACCESS); |
dc386d4d | 830 | |
abfc3488 | 831 | skip_unmap: |
e6a1530d | 832 | if (!page_mapped(page)) |
a6bc32b8 | 833 | rc = move_to_new_page(newpage, page, remap_swapcache, mode); |
e24f0b8f | 834 | |
3fe2011f | 835 | if (rc && remap_swapcache) |
e24f0b8f | 836 | remove_migration_ptes(page, page); |
3f6c8272 MG |
837 | |
838 | /* Drop an anon_vma reference if we took one */ | |
76545066 | 839 | if (anon_vma) |
9e60109f | 840 | put_anon_vma(anon_vma); |
3f6c8272 | 841 | |
01b1ae63 | 842 | uncharge: |
bf6bddf1 RA |
843 | mem_cgroup_end_migration(mem, page, newpage, |
844 | (rc == MIGRATEPAGE_SUCCESS || | |
845 | rc == MIGRATEPAGE_BALLOON_SUCCESS)); | |
e24f0b8f CL |
846 | unlock: |
847 | unlock_page(page); | |
0dabec93 MK |
848 | out: |
849 | return rc; | |
850 | } | |
95a402c3 | 851 | |
0dabec93 MK |
852 | /* |
853 | * Obtain the lock on page, remove all ptes and migrate the page | |
854 | * to the newly allocated page in newpage. | |
855 | */ | |
856 | static int unmap_and_move(new_page_t get_new_page, unsigned long private, | |
a6bc32b8 MG |
857 | struct page *page, int force, bool offlining, |
858 | enum migrate_mode mode) | |
0dabec93 MK |
859 | { |
860 | int rc = 0; | |
861 | int *result = NULL; | |
862 | struct page *newpage = get_new_page(page, private, &result); | |
863 | ||
864 | if (!newpage) | |
865 | return -ENOMEM; | |
866 | ||
867 | if (page_count(page) == 1) { | |
868 | /* page was freed from under us. So we are done. */ | |
869 | goto out; | |
870 | } | |
871 | ||
872 | if (unlikely(PageTransHuge(page))) | |
873 | if (unlikely(split_huge_page(page))) | |
874 | goto out; | |
875 | ||
a6bc32b8 | 876 | rc = __unmap_and_move(page, newpage, force, offlining, mode); |
bf6bddf1 RA |
877 | |
878 | if (unlikely(rc == MIGRATEPAGE_BALLOON_SUCCESS)) { | |
879 | /* | |
880 | * A ballooned page has been migrated already. | |
881 | * Now, it's the time to wrap-up counters, | |
882 | * handle the page back to Buddy and return. | |
883 | */ | |
884 | dec_zone_page_state(page, NR_ISOLATED_ANON + | |
885 | page_is_file_cache(page)); | |
886 | balloon_page_free(page); | |
887 | return MIGRATEPAGE_SUCCESS; | |
888 | } | |
0dabec93 | 889 | out: |
e24f0b8f | 890 | if (rc != -EAGAIN) { |
0dabec93 MK |
891 | /* |
892 | * A page that has been migrated has all references | |
893 | * removed and will be freed. A page that has not been | |
894 | * migrated will have kepts its references and be | |
895 | * restored. | |
896 | */ | |
897 | list_del(&page->lru); | |
a731286d | 898 | dec_zone_page_state(page, NR_ISOLATED_ANON + |
6c0b1351 | 899 | page_is_file_cache(page)); |
894bc310 | 900 | putback_lru_page(page); |
e24f0b8f | 901 | } |
95a402c3 CL |
902 | /* |
903 | * Move the new page to the LRU. If migration was not successful | |
904 | * then this will free the page. | |
905 | */ | |
894bc310 | 906 | putback_lru_page(newpage); |
742755a1 CL |
907 | if (result) { |
908 | if (rc) | |
909 | *result = rc; | |
910 | else | |
911 | *result = page_to_nid(newpage); | |
912 | } | |
e24f0b8f CL |
913 | return rc; |
914 | } | |
915 | ||
290408d4 NH |
916 | /* |
917 | * Counterpart of unmap_and_move_page() for hugepage migration. | |
918 | * | |
919 | * This function doesn't wait the completion of hugepage I/O | |
920 | * because there is no race between I/O and migration for hugepage. | |
921 | * Note that currently hugepage I/O occurs only in direct I/O | |
922 | * where no lock is held and PG_writeback is irrelevant, | |
923 | * and writeback status of all subpages are counted in the reference | |
924 | * count of the head page (i.e. if all subpages of a 2MB hugepage are | |
925 | * under direct I/O, the reference of the head page is 512 and a bit more.) | |
926 | * This means that when we try to migrate hugepage whose subpages are | |
927 | * doing direct I/O, some references remain after try_to_unmap() and | |
928 | * hugepage migration fails without data corruption. | |
929 | * | |
930 | * There is also no race when direct I/O is issued on the page under migration, | |
931 | * because then pte is replaced with migration swap entry and direct I/O code | |
932 | * will wait in the page fault for migration to complete. | |
933 | */ | |
934 | static int unmap_and_move_huge_page(new_page_t get_new_page, | |
935 | unsigned long private, struct page *hpage, | |
a6bc32b8 MG |
936 | int force, bool offlining, |
937 | enum migrate_mode mode) | |
290408d4 NH |
938 | { |
939 | int rc = 0; | |
940 | int *result = NULL; | |
941 | struct page *new_hpage = get_new_page(hpage, private, &result); | |
290408d4 NH |
942 | struct anon_vma *anon_vma = NULL; |
943 | ||
944 | if (!new_hpage) | |
945 | return -ENOMEM; | |
946 | ||
947 | rc = -EAGAIN; | |
948 | ||
949 | if (!trylock_page(hpage)) { | |
a6bc32b8 | 950 | if (!force || mode != MIGRATE_SYNC) |
290408d4 NH |
951 | goto out; |
952 | lock_page(hpage); | |
953 | } | |
954 | ||
746b18d4 PZ |
955 | if (PageAnon(hpage)) |
956 | anon_vma = page_get_anon_vma(hpage); | |
290408d4 NH |
957 | |
958 | try_to_unmap(hpage, TTU_MIGRATION|TTU_IGNORE_MLOCK|TTU_IGNORE_ACCESS); | |
959 | ||
960 | if (!page_mapped(hpage)) | |
a6bc32b8 | 961 | rc = move_to_new_page(new_hpage, hpage, 1, mode); |
290408d4 NH |
962 | |
963 | if (rc) | |
964 | remove_migration_ptes(hpage, hpage); | |
965 | ||
fd4a4663 | 966 | if (anon_vma) |
9e60109f | 967 | put_anon_vma(anon_vma); |
8e6ac7fa AK |
968 | |
969 | if (!rc) | |
970 | hugetlb_cgroup_migrate(hpage, new_hpage); | |
971 | ||
290408d4 | 972 | unlock_page(hpage); |
09761333 | 973 | out: |
290408d4 | 974 | put_page(new_hpage); |
290408d4 NH |
975 | if (result) { |
976 | if (rc) | |
977 | *result = rc; | |
978 | else | |
979 | *result = page_to_nid(new_hpage); | |
980 | } | |
981 | return rc; | |
982 | } | |
983 | ||
b20a3503 CL |
984 | /* |
985 | * migrate_pages | |
986 | * | |
95a402c3 CL |
987 | * The function takes one list of pages to migrate and a function |
988 | * that determines from the page to be migrated and the private data | |
989 | * the target of the move and allocates the page. | |
b20a3503 CL |
990 | * |
991 | * The function returns after 10 attempts or if no pages | |
992 | * are movable anymore because to has become empty | |
cf608ac1 MK |
993 | * or no retryable pages exist anymore. |
994 | * Caller should call putback_lru_pages to return pages to the LRU | |
28bd6578 | 995 | * or free list only if ret != 0. |
b20a3503 | 996 | * |
95a402c3 | 997 | * Return: Number of pages not migrated or error code. |
b20a3503 | 998 | */ |
95a402c3 | 999 | int migrate_pages(struct list_head *from, |
7f0f2496 | 1000 | new_page_t get_new_page, unsigned long private, bool offlining, |
a6bc32b8 | 1001 | enum migrate_mode mode) |
b20a3503 | 1002 | { |
e24f0b8f | 1003 | int retry = 1; |
b20a3503 CL |
1004 | int nr_failed = 0; |
1005 | int pass = 0; | |
1006 | struct page *page; | |
1007 | struct page *page2; | |
1008 | int swapwrite = current->flags & PF_SWAPWRITE; | |
1009 | int rc; | |
1010 | ||
1011 | if (!swapwrite) | |
1012 | current->flags |= PF_SWAPWRITE; | |
1013 | ||
e24f0b8f CL |
1014 | for(pass = 0; pass < 10 && retry; pass++) { |
1015 | retry = 0; | |
b20a3503 | 1016 | |
e24f0b8f | 1017 | list_for_each_entry_safe(page, page2, from, lru) { |
e24f0b8f | 1018 | cond_resched(); |
2d1db3b1 | 1019 | |
95a402c3 | 1020 | rc = unmap_and_move(get_new_page, private, |
77f1fe6b | 1021 | page, pass > 2, offlining, |
a6bc32b8 | 1022 | mode); |
2d1db3b1 | 1023 | |
e24f0b8f | 1024 | switch(rc) { |
95a402c3 CL |
1025 | case -ENOMEM: |
1026 | goto out; | |
e24f0b8f | 1027 | case -EAGAIN: |
2d1db3b1 | 1028 | retry++; |
e24f0b8f | 1029 | break; |
78bd5209 | 1030 | case MIGRATEPAGE_SUCCESS: |
e24f0b8f CL |
1031 | break; |
1032 | default: | |
2d1db3b1 | 1033 | /* Permanent failure */ |
2d1db3b1 | 1034 | nr_failed++; |
e24f0b8f | 1035 | break; |
2d1db3b1 | 1036 | } |
b20a3503 CL |
1037 | } |
1038 | } | |
78bd5209 | 1039 | rc = nr_failed + retry; |
95a402c3 | 1040 | out: |
b20a3503 CL |
1041 | if (!swapwrite) |
1042 | current->flags &= ~PF_SWAPWRITE; | |
1043 | ||
78bd5209 | 1044 | return rc; |
b20a3503 | 1045 | } |
95a402c3 | 1046 | |
189ebff2 AK |
1047 | int migrate_huge_page(struct page *hpage, new_page_t get_new_page, |
1048 | unsigned long private, bool offlining, | |
1049 | enum migrate_mode mode) | |
290408d4 | 1050 | { |
189ebff2 AK |
1051 | int pass, rc; |
1052 | ||
1053 | for (pass = 0; pass < 10; pass++) { | |
1054 | rc = unmap_and_move_huge_page(get_new_page, | |
1055 | private, hpage, pass > 2, offlining, | |
1056 | mode); | |
1057 | switch (rc) { | |
1058 | case -ENOMEM: | |
1059 | goto out; | |
1060 | case -EAGAIN: | |
1061 | /* try again */ | |
290408d4 | 1062 | cond_resched(); |
189ebff2 | 1063 | break; |
78bd5209 | 1064 | case MIGRATEPAGE_SUCCESS: |
189ebff2 AK |
1065 | goto out; |
1066 | default: | |
1067 | rc = -EIO; | |
1068 | goto out; | |
290408d4 NH |
1069 | } |
1070 | } | |
290408d4 | 1071 | out: |
189ebff2 | 1072 | return rc; |
290408d4 NH |
1073 | } |
1074 | ||
742755a1 CL |
1075 | #ifdef CONFIG_NUMA |
1076 | /* | |
1077 | * Move a list of individual pages | |
1078 | */ | |
1079 | struct page_to_node { | |
1080 | unsigned long addr; | |
1081 | struct page *page; | |
1082 | int node; | |
1083 | int status; | |
1084 | }; | |
1085 | ||
1086 | static struct page *new_page_node(struct page *p, unsigned long private, | |
1087 | int **result) | |
1088 | { | |
1089 | struct page_to_node *pm = (struct page_to_node *)private; | |
1090 | ||
1091 | while (pm->node != MAX_NUMNODES && pm->page != p) | |
1092 | pm++; | |
1093 | ||
1094 | if (pm->node == MAX_NUMNODES) | |
1095 | return NULL; | |
1096 | ||
1097 | *result = &pm->status; | |
1098 | ||
6484eb3e | 1099 | return alloc_pages_exact_node(pm->node, |
769848c0 | 1100 | GFP_HIGHUSER_MOVABLE | GFP_THISNODE, 0); |
742755a1 CL |
1101 | } |
1102 | ||
1103 | /* | |
1104 | * Move a set of pages as indicated in the pm array. The addr | |
1105 | * field must be set to the virtual address of the page to be moved | |
1106 | * and the node number must contain a valid target node. | |
5e9a0f02 | 1107 | * The pm array ends with node = MAX_NUMNODES. |
742755a1 | 1108 | */ |
5e9a0f02 BG |
1109 | static int do_move_page_to_node_array(struct mm_struct *mm, |
1110 | struct page_to_node *pm, | |
1111 | int migrate_all) | |
742755a1 CL |
1112 | { |
1113 | int err; | |
1114 | struct page_to_node *pp; | |
1115 | LIST_HEAD(pagelist); | |
1116 | ||
1117 | down_read(&mm->mmap_sem); | |
1118 | ||
1119 | /* | |
1120 | * Build a list of pages to migrate | |
1121 | */ | |
742755a1 CL |
1122 | for (pp = pm; pp->node != MAX_NUMNODES; pp++) { |
1123 | struct vm_area_struct *vma; | |
1124 | struct page *page; | |
1125 | ||
742755a1 CL |
1126 | err = -EFAULT; |
1127 | vma = find_vma(mm, pp->addr); | |
70384dc6 | 1128 | if (!vma || pp->addr < vma->vm_start || !vma_migratable(vma)) |
742755a1 CL |
1129 | goto set_status; |
1130 | ||
500d65d4 | 1131 | page = follow_page(vma, pp->addr, FOLL_GET|FOLL_SPLIT); |
89f5b7da LT |
1132 | |
1133 | err = PTR_ERR(page); | |
1134 | if (IS_ERR(page)) | |
1135 | goto set_status; | |
1136 | ||
742755a1 CL |
1137 | err = -ENOENT; |
1138 | if (!page) | |
1139 | goto set_status; | |
1140 | ||
62b61f61 HD |
1141 | /* Use PageReserved to check for zero page */ |
1142 | if (PageReserved(page) || PageKsm(page)) | |
742755a1 CL |
1143 | goto put_and_set; |
1144 | ||
1145 | pp->page = page; | |
1146 | err = page_to_nid(page); | |
1147 | ||
1148 | if (err == pp->node) | |
1149 | /* | |
1150 | * Node already in the right place | |
1151 | */ | |
1152 | goto put_and_set; | |
1153 | ||
1154 | err = -EACCES; | |
1155 | if (page_mapcount(page) > 1 && | |
1156 | !migrate_all) | |
1157 | goto put_and_set; | |
1158 | ||
62695a84 | 1159 | err = isolate_lru_page(page); |
6d9c285a | 1160 | if (!err) { |
62695a84 | 1161 | list_add_tail(&page->lru, &pagelist); |
6d9c285a KM |
1162 | inc_zone_page_state(page, NR_ISOLATED_ANON + |
1163 | page_is_file_cache(page)); | |
1164 | } | |
742755a1 CL |
1165 | put_and_set: |
1166 | /* | |
1167 | * Either remove the duplicate refcount from | |
1168 | * isolate_lru_page() or drop the page ref if it was | |
1169 | * not isolated. | |
1170 | */ | |
1171 | put_page(page); | |
1172 | set_status: | |
1173 | pp->status = err; | |
1174 | } | |
1175 | ||
e78bbfa8 | 1176 | err = 0; |
cf608ac1 | 1177 | if (!list_empty(&pagelist)) { |
742755a1 | 1178 | err = migrate_pages(&pagelist, new_page_node, |
a6bc32b8 | 1179 | (unsigned long)pm, 0, MIGRATE_SYNC); |
cf608ac1 MK |
1180 | if (err) |
1181 | putback_lru_pages(&pagelist); | |
1182 | } | |
742755a1 CL |
1183 | |
1184 | up_read(&mm->mmap_sem); | |
1185 | return err; | |
1186 | } | |
1187 | ||
5e9a0f02 BG |
1188 | /* |
1189 | * Migrate an array of page address onto an array of nodes and fill | |
1190 | * the corresponding array of status. | |
1191 | */ | |
3268c63e | 1192 | static int do_pages_move(struct mm_struct *mm, nodemask_t task_nodes, |
5e9a0f02 BG |
1193 | unsigned long nr_pages, |
1194 | const void __user * __user *pages, | |
1195 | const int __user *nodes, | |
1196 | int __user *status, int flags) | |
1197 | { | |
3140a227 | 1198 | struct page_to_node *pm; |
3140a227 BG |
1199 | unsigned long chunk_nr_pages; |
1200 | unsigned long chunk_start; | |
1201 | int err; | |
5e9a0f02 | 1202 | |
3140a227 BG |
1203 | err = -ENOMEM; |
1204 | pm = (struct page_to_node *)__get_free_page(GFP_KERNEL); | |
1205 | if (!pm) | |
5e9a0f02 | 1206 | goto out; |
35282a2d BG |
1207 | |
1208 | migrate_prep(); | |
1209 | ||
5e9a0f02 | 1210 | /* |
3140a227 BG |
1211 | * Store a chunk of page_to_node array in a page, |
1212 | * but keep the last one as a marker | |
5e9a0f02 | 1213 | */ |
3140a227 | 1214 | chunk_nr_pages = (PAGE_SIZE / sizeof(struct page_to_node)) - 1; |
5e9a0f02 | 1215 | |
3140a227 BG |
1216 | for (chunk_start = 0; |
1217 | chunk_start < nr_pages; | |
1218 | chunk_start += chunk_nr_pages) { | |
1219 | int j; | |
5e9a0f02 | 1220 | |
3140a227 BG |
1221 | if (chunk_start + chunk_nr_pages > nr_pages) |
1222 | chunk_nr_pages = nr_pages - chunk_start; | |
1223 | ||
1224 | /* fill the chunk pm with addrs and nodes from user-space */ | |
1225 | for (j = 0; j < chunk_nr_pages; j++) { | |
1226 | const void __user *p; | |
5e9a0f02 BG |
1227 | int node; |
1228 | ||
3140a227 BG |
1229 | err = -EFAULT; |
1230 | if (get_user(p, pages + j + chunk_start)) | |
1231 | goto out_pm; | |
1232 | pm[j].addr = (unsigned long) p; | |
1233 | ||
1234 | if (get_user(node, nodes + j + chunk_start)) | |
5e9a0f02 BG |
1235 | goto out_pm; |
1236 | ||
1237 | err = -ENODEV; | |
6f5a55f1 LT |
1238 | if (node < 0 || node >= MAX_NUMNODES) |
1239 | goto out_pm; | |
1240 | ||
5e9a0f02 BG |
1241 | if (!node_state(node, N_HIGH_MEMORY)) |
1242 | goto out_pm; | |
1243 | ||
1244 | err = -EACCES; | |
1245 | if (!node_isset(node, task_nodes)) | |
1246 | goto out_pm; | |
1247 | ||
3140a227 BG |
1248 | pm[j].node = node; |
1249 | } | |
1250 | ||
1251 | /* End marker for this chunk */ | |
1252 | pm[chunk_nr_pages].node = MAX_NUMNODES; | |
1253 | ||
1254 | /* Migrate this chunk */ | |
1255 | err = do_move_page_to_node_array(mm, pm, | |
1256 | flags & MPOL_MF_MOVE_ALL); | |
1257 | if (err < 0) | |
1258 | goto out_pm; | |
5e9a0f02 | 1259 | |
5e9a0f02 | 1260 | /* Return status information */ |
3140a227 BG |
1261 | for (j = 0; j < chunk_nr_pages; j++) |
1262 | if (put_user(pm[j].status, status + j + chunk_start)) { | |
5e9a0f02 | 1263 | err = -EFAULT; |
3140a227 BG |
1264 | goto out_pm; |
1265 | } | |
1266 | } | |
1267 | err = 0; | |
5e9a0f02 BG |
1268 | |
1269 | out_pm: | |
3140a227 | 1270 | free_page((unsigned long)pm); |
5e9a0f02 BG |
1271 | out: |
1272 | return err; | |
1273 | } | |
1274 | ||
742755a1 | 1275 | /* |
2f007e74 | 1276 | * Determine the nodes of an array of pages and store it in an array of status. |
742755a1 | 1277 | */ |
80bba129 BG |
1278 | static void do_pages_stat_array(struct mm_struct *mm, unsigned long nr_pages, |
1279 | const void __user **pages, int *status) | |
742755a1 | 1280 | { |
2f007e74 | 1281 | unsigned long i; |
2f007e74 | 1282 | |
742755a1 CL |
1283 | down_read(&mm->mmap_sem); |
1284 | ||
2f007e74 | 1285 | for (i = 0; i < nr_pages; i++) { |
80bba129 | 1286 | unsigned long addr = (unsigned long)(*pages); |
742755a1 CL |
1287 | struct vm_area_struct *vma; |
1288 | struct page *page; | |
c095adbc | 1289 | int err = -EFAULT; |
2f007e74 BG |
1290 | |
1291 | vma = find_vma(mm, addr); | |
70384dc6 | 1292 | if (!vma || addr < vma->vm_start) |
742755a1 CL |
1293 | goto set_status; |
1294 | ||
2f007e74 | 1295 | page = follow_page(vma, addr, 0); |
89f5b7da LT |
1296 | |
1297 | err = PTR_ERR(page); | |
1298 | if (IS_ERR(page)) | |
1299 | goto set_status; | |
1300 | ||
742755a1 CL |
1301 | err = -ENOENT; |
1302 | /* Use PageReserved to check for zero page */ | |
62b61f61 | 1303 | if (!page || PageReserved(page) || PageKsm(page)) |
742755a1 CL |
1304 | goto set_status; |
1305 | ||
1306 | err = page_to_nid(page); | |
1307 | set_status: | |
80bba129 BG |
1308 | *status = err; |
1309 | ||
1310 | pages++; | |
1311 | status++; | |
1312 | } | |
1313 | ||
1314 | up_read(&mm->mmap_sem); | |
1315 | } | |
1316 | ||
1317 | /* | |
1318 | * Determine the nodes of a user array of pages and store it in | |
1319 | * a user array of status. | |
1320 | */ | |
1321 | static int do_pages_stat(struct mm_struct *mm, unsigned long nr_pages, | |
1322 | const void __user * __user *pages, | |
1323 | int __user *status) | |
1324 | { | |
1325 | #define DO_PAGES_STAT_CHUNK_NR 16 | |
1326 | const void __user *chunk_pages[DO_PAGES_STAT_CHUNK_NR]; | |
1327 | int chunk_status[DO_PAGES_STAT_CHUNK_NR]; | |
80bba129 | 1328 | |
87b8d1ad PA |
1329 | while (nr_pages) { |
1330 | unsigned long chunk_nr; | |
80bba129 | 1331 | |
87b8d1ad PA |
1332 | chunk_nr = nr_pages; |
1333 | if (chunk_nr > DO_PAGES_STAT_CHUNK_NR) | |
1334 | chunk_nr = DO_PAGES_STAT_CHUNK_NR; | |
1335 | ||
1336 | if (copy_from_user(chunk_pages, pages, chunk_nr * sizeof(*chunk_pages))) | |
1337 | break; | |
80bba129 BG |
1338 | |
1339 | do_pages_stat_array(mm, chunk_nr, chunk_pages, chunk_status); | |
1340 | ||
87b8d1ad PA |
1341 | if (copy_to_user(status, chunk_status, chunk_nr * sizeof(*status))) |
1342 | break; | |
742755a1 | 1343 | |
87b8d1ad PA |
1344 | pages += chunk_nr; |
1345 | status += chunk_nr; | |
1346 | nr_pages -= chunk_nr; | |
1347 | } | |
1348 | return nr_pages ? -EFAULT : 0; | |
742755a1 CL |
1349 | } |
1350 | ||
1351 | /* | |
1352 | * Move a list of pages in the address space of the currently executing | |
1353 | * process. | |
1354 | */ | |
938bb9f5 HC |
1355 | SYSCALL_DEFINE6(move_pages, pid_t, pid, unsigned long, nr_pages, |
1356 | const void __user * __user *, pages, | |
1357 | const int __user *, nodes, | |
1358 | int __user *, status, int, flags) | |
742755a1 | 1359 | { |
c69e8d9c | 1360 | const struct cred *cred = current_cred(), *tcred; |
742755a1 | 1361 | struct task_struct *task; |
742755a1 | 1362 | struct mm_struct *mm; |
5e9a0f02 | 1363 | int err; |
3268c63e | 1364 | nodemask_t task_nodes; |
742755a1 CL |
1365 | |
1366 | /* Check flags */ | |
1367 | if (flags & ~(MPOL_MF_MOVE|MPOL_MF_MOVE_ALL)) | |
1368 | return -EINVAL; | |
1369 | ||
1370 | if ((flags & MPOL_MF_MOVE_ALL) && !capable(CAP_SYS_NICE)) | |
1371 | return -EPERM; | |
1372 | ||
1373 | /* Find the mm_struct */ | |
a879bf58 | 1374 | rcu_read_lock(); |
228ebcbe | 1375 | task = pid ? find_task_by_vpid(pid) : current; |
742755a1 | 1376 | if (!task) { |
a879bf58 | 1377 | rcu_read_unlock(); |
742755a1 CL |
1378 | return -ESRCH; |
1379 | } | |
3268c63e | 1380 | get_task_struct(task); |
742755a1 CL |
1381 | |
1382 | /* | |
1383 | * Check if this process has the right to modify the specified | |
1384 | * process. The right exists if the process has administrative | |
1385 | * capabilities, superuser privileges or the same | |
1386 | * userid as the target process. | |
1387 | */ | |
c69e8d9c | 1388 | tcred = __task_cred(task); |
b38a86eb EB |
1389 | if (!uid_eq(cred->euid, tcred->suid) && !uid_eq(cred->euid, tcred->uid) && |
1390 | !uid_eq(cred->uid, tcred->suid) && !uid_eq(cred->uid, tcred->uid) && | |
742755a1 | 1391 | !capable(CAP_SYS_NICE)) { |
c69e8d9c | 1392 | rcu_read_unlock(); |
742755a1 | 1393 | err = -EPERM; |
5e9a0f02 | 1394 | goto out; |
742755a1 | 1395 | } |
c69e8d9c | 1396 | rcu_read_unlock(); |
742755a1 | 1397 | |
86c3a764 DQ |
1398 | err = security_task_movememory(task); |
1399 | if (err) | |
5e9a0f02 | 1400 | goto out; |
86c3a764 | 1401 | |
3268c63e CL |
1402 | task_nodes = cpuset_mems_allowed(task); |
1403 | mm = get_task_mm(task); | |
1404 | put_task_struct(task); | |
1405 | ||
6e8b09ea SL |
1406 | if (!mm) |
1407 | return -EINVAL; | |
1408 | ||
1409 | if (nodes) | |
1410 | err = do_pages_move(mm, task_nodes, nr_pages, pages, | |
1411 | nodes, status, flags); | |
1412 | else | |
1413 | err = do_pages_stat(mm, nr_pages, pages, status); | |
742755a1 | 1414 | |
742755a1 CL |
1415 | mmput(mm); |
1416 | return err; | |
3268c63e CL |
1417 | |
1418 | out: | |
1419 | put_task_struct(task); | |
1420 | return err; | |
742755a1 | 1421 | } |
742755a1 | 1422 | |
7b2259b3 CL |
1423 | /* |
1424 | * Call migration functions in the vma_ops that may prepare | |
1425 | * memory in a vm for migration. migration functions may perform | |
1426 | * the migration for vmas that do not have an underlying page struct. | |
1427 | */ | |
1428 | int migrate_vmas(struct mm_struct *mm, const nodemask_t *to, | |
1429 | const nodemask_t *from, unsigned long flags) | |
1430 | { | |
1431 | struct vm_area_struct *vma; | |
1432 | int err = 0; | |
1433 | ||
1001c9fb | 1434 | for (vma = mm->mmap; vma && !err; vma = vma->vm_next) { |
7b2259b3 CL |
1435 | if (vma->vm_ops && vma->vm_ops->migrate) { |
1436 | err = vma->vm_ops->migrate(vma, to, from, flags); | |
1437 | if (err) | |
1438 | break; | |
1439 | } | |
1440 | } | |
1441 | return err; | |
1442 | } | |
83d1674a | 1443 | #endif |