]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * mm/rmap.c - physical to virtual reverse mappings | |
3 | * | |
4 | * Copyright 2001, Rik van Riel <riel@conectiva.com.br> | |
5 | * Released under the General Public License (GPL). | |
6 | * | |
7 | * Simple, low overhead reverse mapping scheme. | |
8 | * Please try to keep this thing as modular as possible. | |
9 | * | |
10 | * Provides methods for unmapping each kind of mapped page: | |
11 | * the anon methods track anonymous pages, and | |
12 | * the file methods track pages belonging to an inode. | |
13 | * | |
14 | * Original design by Rik van Riel <riel@conectiva.com.br> 2001 | |
15 | * File methods by Dave McCracken <dmccr@us.ibm.com> 2003, 2004 | |
16 | * Anonymous methods by Andrea Arcangeli <andrea@suse.de> 2004 | |
98f32602 | 17 | * Contributions by Hugh Dickins 2003, 2004 |
1da177e4 LT |
18 | */ |
19 | ||
20 | /* | |
21 | * Lock ordering in mm: | |
22 | * | |
1b1dcc1b | 23 | * inode->i_mutex (while writing or truncating, not reading or faulting) |
82591e6e NP |
24 | * inode->i_alloc_sem (vmtruncate_range) |
25 | * mm->mmap_sem | |
26 | * page->flags PG_locked (lock_page) | |
27 | * mapping->i_mmap_lock | |
28 | * anon_vma->lock | |
29 | * mm->page_table_lock or pte_lock | |
30 | * zone->lru_lock (in mark_page_accessed, isolate_lru_page) | |
31 | * swap_lock (in swap_duplicate, swap_info_get) | |
32 | * mmlist_lock (in mmput, drain_mmlist and others) | |
33 | * mapping->private_lock (in __set_page_dirty_buffers) | |
34 | * inode_lock (in set_page_dirty's __mark_inode_dirty) | |
35 | * sb_lock (within inode_lock in fs/fs-writeback.c) | |
36 | * mapping->tree_lock (widely used, in set_page_dirty, | |
37 | * in arch-dependent flush_dcache_mmap_lock, | |
38 | * within inode_lock in __sync_single_inode) | |
6a46079c AK |
39 | * |
40 | * (code doesn't rely on that order so it could be switched around) | |
41 | * ->tasklist_lock | |
42 | * anon_vma->lock (memory_failure, collect_procs_anon) | |
43 | * pte map lock | |
1da177e4 LT |
44 | */ |
45 | ||
46 | #include <linux/mm.h> | |
47 | #include <linux/pagemap.h> | |
48 | #include <linux/swap.h> | |
49 | #include <linux/swapops.h> | |
50 | #include <linux/slab.h> | |
51 | #include <linux/init.h> | |
5ad64688 | 52 | #include <linux/ksm.h> |
1da177e4 LT |
53 | #include <linux/rmap.h> |
54 | #include <linux/rcupdate.h> | |
a48d07af | 55 | #include <linux/module.h> |
8a9f3ccd | 56 | #include <linux/memcontrol.h> |
cddb8a5c | 57 | #include <linux/mmu_notifier.h> |
64cdd548 | 58 | #include <linux/migrate.h> |
1da177e4 LT |
59 | |
60 | #include <asm/tlbflush.h> | |
61 | ||
b291f000 NP |
62 | #include "internal.h" |
63 | ||
fdd2e5f8 | 64 | static struct kmem_cache *anon_vma_cachep; |
5beb4930 | 65 | static struct kmem_cache *anon_vma_chain_cachep; |
fdd2e5f8 AB |
66 | |
67 | static inline struct anon_vma *anon_vma_alloc(void) | |
68 | { | |
69 | return kmem_cache_alloc(anon_vma_cachep, GFP_KERNEL); | |
70 | } | |
71 | ||
db114b83 | 72 | void anon_vma_free(struct anon_vma *anon_vma) |
fdd2e5f8 AB |
73 | { |
74 | kmem_cache_free(anon_vma_cachep, anon_vma); | |
75 | } | |
1da177e4 | 76 | |
5beb4930 RR |
77 | static inline struct anon_vma_chain *anon_vma_chain_alloc(void) |
78 | { | |
79 | return kmem_cache_alloc(anon_vma_chain_cachep, GFP_KERNEL); | |
80 | } | |
81 | ||
82 | void anon_vma_chain_free(struct anon_vma_chain *anon_vma_chain) | |
83 | { | |
84 | kmem_cache_free(anon_vma_chain_cachep, anon_vma_chain); | |
85 | } | |
86 | ||
d9d332e0 LT |
87 | /** |
88 | * anon_vma_prepare - attach an anon_vma to a memory region | |
89 | * @vma: the memory region in question | |
90 | * | |
91 | * This makes sure the memory mapping described by 'vma' has | |
92 | * an 'anon_vma' attached to it, so that we can associate the | |
93 | * anonymous pages mapped into it with that anon_vma. | |
94 | * | |
95 | * The common case will be that we already have one, but if | |
96 | * if not we either need to find an adjacent mapping that we | |
97 | * can re-use the anon_vma from (very common when the only | |
98 | * reason for splitting a vma has been mprotect()), or we | |
99 | * allocate a new one. | |
100 | * | |
101 | * Anon-vma allocations are very subtle, because we may have | |
102 | * optimistically looked up an anon_vma in page_lock_anon_vma() | |
103 | * and that may actually touch the spinlock even in the newly | |
104 | * allocated vma (it depends on RCU to make sure that the | |
105 | * anon_vma isn't actually destroyed). | |
106 | * | |
107 | * As a result, we need to do proper anon_vma locking even | |
108 | * for the new allocation. At the same time, we do not want | |
109 | * to do any locking for the common case of already having | |
110 | * an anon_vma. | |
111 | * | |
112 | * This must be called with the mmap_sem held for reading. | |
113 | */ | |
1da177e4 LT |
114 | int anon_vma_prepare(struct vm_area_struct *vma) |
115 | { | |
116 | struct anon_vma *anon_vma = vma->anon_vma; | |
5beb4930 | 117 | struct anon_vma_chain *avc; |
1da177e4 LT |
118 | |
119 | might_sleep(); | |
120 | if (unlikely(!anon_vma)) { | |
121 | struct mm_struct *mm = vma->vm_mm; | |
d9d332e0 | 122 | struct anon_vma *allocated; |
1da177e4 | 123 | |
5beb4930 RR |
124 | avc = anon_vma_chain_alloc(); |
125 | if (!avc) | |
126 | goto out_enomem; | |
127 | ||
1da177e4 | 128 | anon_vma = find_mergeable_anon_vma(vma); |
d9d332e0 LT |
129 | allocated = NULL; |
130 | if (!anon_vma) { | |
1da177e4 LT |
131 | anon_vma = anon_vma_alloc(); |
132 | if (unlikely(!anon_vma)) | |
5beb4930 | 133 | goto out_enomem_free_avc; |
1da177e4 | 134 | allocated = anon_vma; |
5c341ee1 RR |
135 | /* |
136 | * This VMA had no anon_vma yet. This anon_vma is | |
137 | * the root of any anon_vma tree that might form. | |
138 | */ | |
139 | anon_vma->root = anon_vma; | |
1da177e4 LT |
140 | } |
141 | ||
cba48b98 | 142 | anon_vma_lock(anon_vma); |
1da177e4 LT |
143 | /* page_table_lock to protect against threads */ |
144 | spin_lock(&mm->page_table_lock); | |
145 | if (likely(!vma->anon_vma)) { | |
146 | vma->anon_vma = anon_vma; | |
5beb4930 RR |
147 | avc->anon_vma = anon_vma; |
148 | avc->vma = vma; | |
149 | list_add(&avc->same_vma, &vma->anon_vma_chain); | |
150 | list_add(&avc->same_anon_vma, &anon_vma->head); | |
1da177e4 | 151 | allocated = NULL; |
31f2b0eb | 152 | avc = NULL; |
1da177e4 LT |
153 | } |
154 | spin_unlock(&mm->page_table_lock); | |
cba48b98 | 155 | anon_vma_unlock(anon_vma); |
31f2b0eb ON |
156 | |
157 | if (unlikely(allocated)) | |
1da177e4 | 158 | anon_vma_free(allocated); |
31f2b0eb | 159 | if (unlikely(avc)) |
5beb4930 | 160 | anon_vma_chain_free(avc); |
1da177e4 LT |
161 | } |
162 | return 0; | |
5beb4930 RR |
163 | |
164 | out_enomem_free_avc: | |
165 | anon_vma_chain_free(avc); | |
166 | out_enomem: | |
167 | return -ENOMEM; | |
1da177e4 LT |
168 | } |
169 | ||
5beb4930 RR |
170 | static void anon_vma_chain_link(struct vm_area_struct *vma, |
171 | struct anon_vma_chain *avc, | |
172 | struct anon_vma *anon_vma) | |
1da177e4 | 173 | { |
5beb4930 RR |
174 | avc->vma = vma; |
175 | avc->anon_vma = anon_vma; | |
176 | list_add(&avc->same_vma, &vma->anon_vma_chain); | |
177 | ||
cba48b98 | 178 | anon_vma_lock(anon_vma); |
5beb4930 | 179 | list_add_tail(&avc->same_anon_vma, &anon_vma->head); |
cba48b98 | 180 | anon_vma_unlock(anon_vma); |
1da177e4 LT |
181 | } |
182 | ||
5beb4930 RR |
183 | /* |
184 | * Attach the anon_vmas from src to dst. | |
185 | * Returns 0 on success, -ENOMEM on failure. | |
186 | */ | |
187 | int anon_vma_clone(struct vm_area_struct *dst, struct vm_area_struct *src) | |
1da177e4 | 188 | { |
5beb4930 RR |
189 | struct anon_vma_chain *avc, *pavc; |
190 | ||
646d87b4 | 191 | list_for_each_entry_reverse(pavc, &src->anon_vma_chain, same_vma) { |
5beb4930 RR |
192 | avc = anon_vma_chain_alloc(); |
193 | if (!avc) | |
194 | goto enomem_failure; | |
195 | anon_vma_chain_link(dst, avc, pavc->anon_vma); | |
196 | } | |
197 | return 0; | |
1da177e4 | 198 | |
5beb4930 RR |
199 | enomem_failure: |
200 | unlink_anon_vmas(dst); | |
201 | return -ENOMEM; | |
1da177e4 LT |
202 | } |
203 | ||
5beb4930 RR |
204 | /* |
205 | * Attach vma to its own anon_vma, as well as to the anon_vmas that | |
206 | * the corresponding VMA in the parent process is attached to. | |
207 | * Returns 0 on success, non-zero on failure. | |
208 | */ | |
209 | int anon_vma_fork(struct vm_area_struct *vma, struct vm_area_struct *pvma) | |
1da177e4 | 210 | { |
5beb4930 RR |
211 | struct anon_vma_chain *avc; |
212 | struct anon_vma *anon_vma; | |
1da177e4 | 213 | |
5beb4930 RR |
214 | /* Don't bother if the parent process has no anon_vma here. */ |
215 | if (!pvma->anon_vma) | |
216 | return 0; | |
217 | ||
218 | /* | |
219 | * First, attach the new VMA to the parent VMA's anon_vmas, | |
220 | * so rmap can find non-COWed pages in child processes. | |
221 | */ | |
222 | if (anon_vma_clone(vma, pvma)) | |
223 | return -ENOMEM; | |
224 | ||
225 | /* Then add our own anon_vma. */ | |
226 | anon_vma = anon_vma_alloc(); | |
227 | if (!anon_vma) | |
228 | goto out_error; | |
229 | avc = anon_vma_chain_alloc(); | |
230 | if (!avc) | |
231 | goto out_error_free_anon_vma; | |
5c341ee1 RR |
232 | |
233 | /* | |
234 | * The root anon_vma's spinlock is the lock actually used when we | |
235 | * lock any of the anon_vmas in this anon_vma tree. | |
236 | */ | |
237 | anon_vma->root = pvma->anon_vma->root; | |
5beb4930 RR |
238 | /* Mark this anon_vma as the one where our new (COWed) pages go. */ |
239 | vma->anon_vma = anon_vma; | |
5c341ee1 | 240 | anon_vma_chain_link(vma, avc, anon_vma); |
5beb4930 RR |
241 | |
242 | return 0; | |
243 | ||
244 | out_error_free_anon_vma: | |
245 | anon_vma_free(anon_vma); | |
246 | out_error: | |
4946d54c | 247 | unlink_anon_vmas(vma); |
5beb4930 | 248 | return -ENOMEM; |
1da177e4 LT |
249 | } |
250 | ||
5beb4930 | 251 | static void anon_vma_unlink(struct anon_vma_chain *anon_vma_chain) |
1da177e4 | 252 | { |
5beb4930 | 253 | struct anon_vma *anon_vma = anon_vma_chain->anon_vma; |
1da177e4 LT |
254 | int empty; |
255 | ||
5beb4930 | 256 | /* If anon_vma_fork fails, we can get an empty anon_vma_chain. */ |
1da177e4 LT |
257 | if (!anon_vma) |
258 | return; | |
259 | ||
cba48b98 | 260 | anon_vma_lock(anon_vma); |
5beb4930 | 261 | list_del(&anon_vma_chain->same_anon_vma); |
1da177e4 LT |
262 | |
263 | /* We must garbage collect the anon_vma if it's empty */ | |
7f60c214 | 264 | empty = list_empty(&anon_vma->head) && !anonvma_external_refcount(anon_vma); |
cba48b98 | 265 | anon_vma_unlock(anon_vma); |
1da177e4 LT |
266 | |
267 | if (empty) | |
268 | anon_vma_free(anon_vma); | |
269 | } | |
270 | ||
5beb4930 RR |
271 | void unlink_anon_vmas(struct vm_area_struct *vma) |
272 | { | |
273 | struct anon_vma_chain *avc, *next; | |
274 | ||
5c341ee1 RR |
275 | /* |
276 | * Unlink each anon_vma chained to the VMA. This list is ordered | |
277 | * from newest to oldest, ensuring the root anon_vma gets freed last. | |
278 | */ | |
5beb4930 RR |
279 | list_for_each_entry_safe(avc, next, &vma->anon_vma_chain, same_vma) { |
280 | anon_vma_unlink(avc); | |
281 | list_del(&avc->same_vma); | |
282 | anon_vma_chain_free(avc); | |
283 | } | |
284 | } | |
285 | ||
51cc5068 | 286 | static void anon_vma_ctor(void *data) |
1da177e4 | 287 | { |
a35afb83 | 288 | struct anon_vma *anon_vma = data; |
1da177e4 | 289 | |
a35afb83 | 290 | spin_lock_init(&anon_vma->lock); |
7f60c214 | 291 | anonvma_external_refcount_init(anon_vma); |
a35afb83 | 292 | INIT_LIST_HEAD(&anon_vma->head); |
1da177e4 LT |
293 | } |
294 | ||
295 | void __init anon_vma_init(void) | |
296 | { | |
297 | anon_vma_cachep = kmem_cache_create("anon_vma", sizeof(struct anon_vma), | |
20c2df83 | 298 | 0, SLAB_DESTROY_BY_RCU|SLAB_PANIC, anon_vma_ctor); |
5beb4930 | 299 | anon_vma_chain_cachep = KMEM_CACHE(anon_vma_chain, SLAB_PANIC); |
1da177e4 LT |
300 | } |
301 | ||
302 | /* | |
303 | * Getting a lock on a stable anon_vma from a page off the LRU is | |
304 | * tricky: page_lock_anon_vma rely on RCU to guard against the races. | |
305 | */ | |
10be22df | 306 | struct anon_vma *page_lock_anon_vma(struct page *page) |
1da177e4 | 307 | { |
34bbd704 | 308 | struct anon_vma *anon_vma; |
1da177e4 LT |
309 | unsigned long anon_mapping; |
310 | ||
311 | rcu_read_lock(); | |
80e14822 | 312 | anon_mapping = (unsigned long) ACCESS_ONCE(page->mapping); |
3ca7b3c5 | 313 | if ((anon_mapping & PAGE_MAPPING_FLAGS) != PAGE_MAPPING_ANON) |
1da177e4 LT |
314 | goto out; |
315 | if (!page_mapped(page)) | |
316 | goto out; | |
317 | ||
318 | anon_vma = (struct anon_vma *) (anon_mapping - PAGE_MAPPING_ANON); | |
cba48b98 | 319 | anon_vma_lock(anon_vma); |
34bbd704 | 320 | return anon_vma; |
1da177e4 LT |
321 | out: |
322 | rcu_read_unlock(); | |
34bbd704 ON |
323 | return NULL; |
324 | } | |
325 | ||
10be22df | 326 | void page_unlock_anon_vma(struct anon_vma *anon_vma) |
34bbd704 | 327 | { |
cba48b98 | 328 | anon_vma_unlock(anon_vma); |
34bbd704 | 329 | rcu_read_unlock(); |
1da177e4 LT |
330 | } |
331 | ||
332 | /* | |
3ad33b24 LS |
333 | * At what user virtual address is page expected in @vma? |
334 | * Returns virtual address or -EFAULT if page's index/offset is not | |
335 | * within the range mapped the @vma. | |
1da177e4 LT |
336 | */ |
337 | static inline unsigned long | |
338 | vma_address(struct page *page, struct vm_area_struct *vma) | |
339 | { | |
340 | pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT); | |
341 | unsigned long address; | |
342 | ||
343 | address = vma->vm_start + ((pgoff - vma->vm_pgoff) << PAGE_SHIFT); | |
344 | if (unlikely(address < vma->vm_start || address >= vma->vm_end)) { | |
3ad33b24 | 345 | /* page should be within @vma mapping range */ |
1da177e4 LT |
346 | return -EFAULT; |
347 | } | |
348 | return address; | |
349 | } | |
350 | ||
351 | /* | |
bf89c8c8 | 352 | * At what user virtual address is page expected in vma? |
ab941e0f | 353 | * Caller should check the page is actually part of the vma. |
1da177e4 LT |
354 | */ |
355 | unsigned long page_address_in_vma(struct page *page, struct vm_area_struct *vma) | |
356 | { | |
ab941e0f NH |
357 | if (PageAnon(page)) |
358 | ; | |
359 | else if (page->mapping && !(vma->vm_flags & VM_NONLINEAR)) { | |
ee498ed7 HD |
360 | if (!vma->vm_file || |
361 | vma->vm_file->f_mapping != page->mapping) | |
1da177e4 LT |
362 | return -EFAULT; |
363 | } else | |
364 | return -EFAULT; | |
365 | return vma_address(page, vma); | |
366 | } | |
367 | ||
81b4082d ND |
368 | /* |
369 | * Check that @page is mapped at @address into @mm. | |
370 | * | |
479db0bf NP |
371 | * If @sync is false, page_check_address may perform a racy check to avoid |
372 | * the page table lock when the pte is not present (helpful when reclaiming | |
373 | * highly shared pages). | |
374 | * | |
b8072f09 | 375 | * On success returns with pte mapped and locked. |
81b4082d | 376 | */ |
ceffc078 | 377 | pte_t *page_check_address(struct page *page, struct mm_struct *mm, |
479db0bf | 378 | unsigned long address, spinlock_t **ptlp, int sync) |
81b4082d ND |
379 | { |
380 | pgd_t *pgd; | |
381 | pud_t *pud; | |
382 | pmd_t *pmd; | |
383 | pte_t *pte; | |
c0718806 | 384 | spinlock_t *ptl; |
81b4082d | 385 | |
81b4082d | 386 | pgd = pgd_offset(mm, address); |
c0718806 HD |
387 | if (!pgd_present(*pgd)) |
388 | return NULL; | |
389 | ||
390 | pud = pud_offset(pgd, address); | |
391 | if (!pud_present(*pud)) | |
392 | return NULL; | |
393 | ||
394 | pmd = pmd_offset(pud, address); | |
395 | if (!pmd_present(*pmd)) | |
396 | return NULL; | |
397 | ||
398 | pte = pte_offset_map(pmd, address); | |
399 | /* Make a quick check before getting the lock */ | |
479db0bf | 400 | if (!sync && !pte_present(*pte)) { |
c0718806 HD |
401 | pte_unmap(pte); |
402 | return NULL; | |
403 | } | |
404 | ||
4c21e2f2 | 405 | ptl = pte_lockptr(mm, pmd); |
c0718806 HD |
406 | spin_lock(ptl); |
407 | if (pte_present(*pte) && page_to_pfn(page) == pte_pfn(*pte)) { | |
408 | *ptlp = ptl; | |
409 | return pte; | |
81b4082d | 410 | } |
c0718806 HD |
411 | pte_unmap_unlock(pte, ptl); |
412 | return NULL; | |
81b4082d ND |
413 | } |
414 | ||
b291f000 NP |
415 | /** |
416 | * page_mapped_in_vma - check whether a page is really mapped in a VMA | |
417 | * @page: the page to test | |
418 | * @vma: the VMA to test | |
419 | * | |
420 | * Returns 1 if the page is mapped into the page tables of the VMA, 0 | |
421 | * if the page is not mapped into the page tables of this VMA. Only | |
422 | * valid for normal file or anonymous VMAs. | |
423 | */ | |
6a46079c | 424 | int page_mapped_in_vma(struct page *page, struct vm_area_struct *vma) |
b291f000 NP |
425 | { |
426 | unsigned long address; | |
427 | pte_t *pte; | |
428 | spinlock_t *ptl; | |
429 | ||
430 | address = vma_address(page, vma); | |
431 | if (address == -EFAULT) /* out of vma range */ | |
432 | return 0; | |
433 | pte = page_check_address(page, vma->vm_mm, address, &ptl, 1); | |
434 | if (!pte) /* the page is not in this mm */ | |
435 | return 0; | |
436 | pte_unmap_unlock(pte, ptl); | |
437 | ||
438 | return 1; | |
439 | } | |
440 | ||
1da177e4 LT |
441 | /* |
442 | * Subfunctions of page_referenced: page_referenced_one called | |
443 | * repeatedly from either page_referenced_anon or page_referenced_file. | |
444 | */ | |
5ad64688 HD |
445 | int page_referenced_one(struct page *page, struct vm_area_struct *vma, |
446 | unsigned long address, unsigned int *mapcount, | |
447 | unsigned long *vm_flags) | |
1da177e4 LT |
448 | { |
449 | struct mm_struct *mm = vma->vm_mm; | |
1da177e4 | 450 | pte_t *pte; |
c0718806 | 451 | spinlock_t *ptl; |
1da177e4 LT |
452 | int referenced = 0; |
453 | ||
479db0bf | 454 | pte = page_check_address(page, mm, address, &ptl, 0); |
c0718806 HD |
455 | if (!pte) |
456 | goto out; | |
1da177e4 | 457 | |
b291f000 NP |
458 | /* |
459 | * Don't want to elevate referenced for mlocked page that gets this far, | |
460 | * in order that it progresses to try_to_unmap and is moved to the | |
461 | * unevictable list. | |
462 | */ | |
5a9bbdcd | 463 | if (vma->vm_flags & VM_LOCKED) { |
5a9bbdcd | 464 | *mapcount = 1; /* break early from loop */ |
03ef83af | 465 | *vm_flags |= VM_LOCKED; |
b291f000 NP |
466 | goto out_unmap; |
467 | } | |
468 | ||
4917e5d0 JW |
469 | if (ptep_clear_flush_young_notify(vma, address, pte)) { |
470 | /* | |
471 | * Don't treat a reference through a sequentially read | |
472 | * mapping as such. If the page has been used in | |
473 | * another mapping, we will catch it; if this other | |
474 | * mapping is already gone, the unmap path will have | |
475 | * set PG_referenced or activated the page. | |
476 | */ | |
477 | if (likely(!VM_SequentialReadHint(vma))) | |
478 | referenced++; | |
479 | } | |
1da177e4 | 480 | |
c0718806 HD |
481 | /* Pretend the page is referenced if the task has the |
482 | swap token and is in the middle of a page fault. */ | |
f7b7fd8f | 483 | if (mm != current->mm && has_swap_token(mm) && |
c0718806 HD |
484 | rwsem_is_locked(&mm->mmap_sem)) |
485 | referenced++; | |
486 | ||
b291f000 | 487 | out_unmap: |
c0718806 HD |
488 | (*mapcount)--; |
489 | pte_unmap_unlock(pte, ptl); | |
273f047e | 490 | |
6fe6b7e3 WF |
491 | if (referenced) |
492 | *vm_flags |= vma->vm_flags; | |
273f047e | 493 | out: |
1da177e4 LT |
494 | return referenced; |
495 | } | |
496 | ||
bed7161a | 497 | static int page_referenced_anon(struct page *page, |
6fe6b7e3 WF |
498 | struct mem_cgroup *mem_cont, |
499 | unsigned long *vm_flags) | |
1da177e4 LT |
500 | { |
501 | unsigned int mapcount; | |
502 | struct anon_vma *anon_vma; | |
5beb4930 | 503 | struct anon_vma_chain *avc; |
1da177e4 LT |
504 | int referenced = 0; |
505 | ||
506 | anon_vma = page_lock_anon_vma(page); | |
507 | if (!anon_vma) | |
508 | return referenced; | |
509 | ||
510 | mapcount = page_mapcount(page); | |
5beb4930 RR |
511 | list_for_each_entry(avc, &anon_vma->head, same_anon_vma) { |
512 | struct vm_area_struct *vma = avc->vma; | |
1cb1729b HD |
513 | unsigned long address = vma_address(page, vma); |
514 | if (address == -EFAULT) | |
515 | continue; | |
bed7161a BS |
516 | /* |
517 | * If we are reclaiming on behalf of a cgroup, skip | |
518 | * counting on behalf of references from different | |
519 | * cgroups | |
520 | */ | |
bd845e38 | 521 | if (mem_cont && !mm_match_cgroup(vma->vm_mm, mem_cont)) |
bed7161a | 522 | continue; |
1cb1729b | 523 | referenced += page_referenced_one(page, vma, address, |
6fe6b7e3 | 524 | &mapcount, vm_flags); |
1da177e4 LT |
525 | if (!mapcount) |
526 | break; | |
527 | } | |
34bbd704 ON |
528 | |
529 | page_unlock_anon_vma(anon_vma); | |
1da177e4 LT |
530 | return referenced; |
531 | } | |
532 | ||
533 | /** | |
534 | * page_referenced_file - referenced check for object-based rmap | |
535 | * @page: the page we're checking references on. | |
43d8eac4 | 536 | * @mem_cont: target memory controller |
6fe6b7e3 | 537 | * @vm_flags: collect encountered vma->vm_flags who actually referenced the page |
1da177e4 LT |
538 | * |
539 | * For an object-based mapped page, find all the places it is mapped and | |
540 | * check/clear the referenced flag. This is done by following the page->mapping | |
541 | * pointer, then walking the chain of vmas it holds. It returns the number | |
542 | * of references it found. | |
543 | * | |
544 | * This function is only called from page_referenced for object-based pages. | |
545 | */ | |
bed7161a | 546 | static int page_referenced_file(struct page *page, |
6fe6b7e3 WF |
547 | struct mem_cgroup *mem_cont, |
548 | unsigned long *vm_flags) | |
1da177e4 LT |
549 | { |
550 | unsigned int mapcount; | |
551 | struct address_space *mapping = page->mapping; | |
552 | pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT); | |
553 | struct vm_area_struct *vma; | |
554 | struct prio_tree_iter iter; | |
555 | int referenced = 0; | |
556 | ||
557 | /* | |
558 | * The caller's checks on page->mapping and !PageAnon have made | |
559 | * sure that this is a file page: the check for page->mapping | |
560 | * excludes the case just before it gets set on an anon page. | |
561 | */ | |
562 | BUG_ON(PageAnon(page)); | |
563 | ||
564 | /* | |
565 | * The page lock not only makes sure that page->mapping cannot | |
566 | * suddenly be NULLified by truncation, it makes sure that the | |
567 | * structure at mapping cannot be freed and reused yet, | |
568 | * so we can safely take mapping->i_mmap_lock. | |
569 | */ | |
570 | BUG_ON(!PageLocked(page)); | |
571 | ||
572 | spin_lock(&mapping->i_mmap_lock); | |
573 | ||
574 | /* | |
575 | * i_mmap_lock does not stabilize mapcount at all, but mapcount | |
576 | * is more likely to be accurate if we note it after spinning. | |
577 | */ | |
578 | mapcount = page_mapcount(page); | |
579 | ||
580 | vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, pgoff, pgoff) { | |
1cb1729b HD |
581 | unsigned long address = vma_address(page, vma); |
582 | if (address == -EFAULT) | |
583 | continue; | |
bed7161a BS |
584 | /* |
585 | * If we are reclaiming on behalf of a cgroup, skip | |
586 | * counting on behalf of references from different | |
587 | * cgroups | |
588 | */ | |
bd845e38 | 589 | if (mem_cont && !mm_match_cgroup(vma->vm_mm, mem_cont)) |
bed7161a | 590 | continue; |
1cb1729b | 591 | referenced += page_referenced_one(page, vma, address, |
6fe6b7e3 | 592 | &mapcount, vm_flags); |
1da177e4 LT |
593 | if (!mapcount) |
594 | break; | |
595 | } | |
596 | ||
597 | spin_unlock(&mapping->i_mmap_lock); | |
598 | return referenced; | |
599 | } | |
600 | ||
601 | /** | |
602 | * page_referenced - test if the page was referenced | |
603 | * @page: the page to test | |
604 | * @is_locked: caller holds lock on the page | |
43d8eac4 | 605 | * @mem_cont: target memory controller |
6fe6b7e3 | 606 | * @vm_flags: collect encountered vma->vm_flags who actually referenced the page |
1da177e4 LT |
607 | * |
608 | * Quick test_and_clear_referenced for all mappings to a page, | |
609 | * returns the number of ptes which referenced the page. | |
610 | */ | |
6fe6b7e3 WF |
611 | int page_referenced(struct page *page, |
612 | int is_locked, | |
613 | struct mem_cgroup *mem_cont, | |
614 | unsigned long *vm_flags) | |
1da177e4 LT |
615 | { |
616 | int referenced = 0; | |
5ad64688 | 617 | int we_locked = 0; |
1da177e4 | 618 | |
6fe6b7e3 | 619 | *vm_flags = 0; |
3ca7b3c5 | 620 | if (page_mapped(page) && page_rmapping(page)) { |
5ad64688 HD |
621 | if (!is_locked && (!PageAnon(page) || PageKsm(page))) { |
622 | we_locked = trylock_page(page); | |
623 | if (!we_locked) { | |
624 | referenced++; | |
625 | goto out; | |
626 | } | |
627 | } | |
628 | if (unlikely(PageKsm(page))) | |
629 | referenced += page_referenced_ksm(page, mem_cont, | |
630 | vm_flags); | |
631 | else if (PageAnon(page)) | |
6fe6b7e3 WF |
632 | referenced += page_referenced_anon(page, mem_cont, |
633 | vm_flags); | |
5ad64688 | 634 | else if (page->mapping) |
6fe6b7e3 WF |
635 | referenced += page_referenced_file(page, mem_cont, |
636 | vm_flags); | |
5ad64688 | 637 | if (we_locked) |
1da177e4 | 638 | unlock_page(page); |
1da177e4 | 639 | } |
5ad64688 | 640 | out: |
5b7baf05 CB |
641 | if (page_test_and_clear_young(page)) |
642 | referenced++; | |
643 | ||
1da177e4 LT |
644 | return referenced; |
645 | } | |
646 | ||
1cb1729b HD |
647 | static int page_mkclean_one(struct page *page, struct vm_area_struct *vma, |
648 | unsigned long address) | |
d08b3851 PZ |
649 | { |
650 | struct mm_struct *mm = vma->vm_mm; | |
c2fda5fe | 651 | pte_t *pte; |
d08b3851 PZ |
652 | spinlock_t *ptl; |
653 | int ret = 0; | |
654 | ||
479db0bf | 655 | pte = page_check_address(page, mm, address, &ptl, 1); |
d08b3851 PZ |
656 | if (!pte) |
657 | goto out; | |
658 | ||
c2fda5fe PZ |
659 | if (pte_dirty(*pte) || pte_write(*pte)) { |
660 | pte_t entry; | |
d08b3851 | 661 | |
c2fda5fe | 662 | flush_cache_page(vma, address, pte_pfn(*pte)); |
cddb8a5c | 663 | entry = ptep_clear_flush_notify(vma, address, pte); |
c2fda5fe PZ |
664 | entry = pte_wrprotect(entry); |
665 | entry = pte_mkclean(entry); | |
d6e88e67 | 666 | set_pte_at(mm, address, pte, entry); |
c2fda5fe PZ |
667 | ret = 1; |
668 | } | |
d08b3851 | 669 | |
d08b3851 PZ |
670 | pte_unmap_unlock(pte, ptl); |
671 | out: | |
672 | return ret; | |
673 | } | |
674 | ||
675 | static int page_mkclean_file(struct address_space *mapping, struct page *page) | |
676 | { | |
677 | pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT); | |
678 | struct vm_area_struct *vma; | |
679 | struct prio_tree_iter iter; | |
680 | int ret = 0; | |
681 | ||
682 | BUG_ON(PageAnon(page)); | |
683 | ||
684 | spin_lock(&mapping->i_mmap_lock); | |
685 | vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, pgoff, pgoff) { | |
1cb1729b HD |
686 | if (vma->vm_flags & VM_SHARED) { |
687 | unsigned long address = vma_address(page, vma); | |
688 | if (address == -EFAULT) | |
689 | continue; | |
690 | ret += page_mkclean_one(page, vma, address); | |
691 | } | |
d08b3851 PZ |
692 | } |
693 | spin_unlock(&mapping->i_mmap_lock); | |
694 | return ret; | |
695 | } | |
696 | ||
697 | int page_mkclean(struct page *page) | |
698 | { | |
699 | int ret = 0; | |
700 | ||
701 | BUG_ON(!PageLocked(page)); | |
702 | ||
703 | if (page_mapped(page)) { | |
704 | struct address_space *mapping = page_mapping(page); | |
ce7e9fae | 705 | if (mapping) { |
d08b3851 | 706 | ret = page_mkclean_file(mapping, page); |
ce7e9fae CB |
707 | if (page_test_dirty(page)) { |
708 | page_clear_dirty(page); | |
709 | ret = 1; | |
710 | } | |
6c210482 | 711 | } |
d08b3851 PZ |
712 | } |
713 | ||
714 | return ret; | |
715 | } | |
60b59bea | 716 | EXPORT_SYMBOL_GPL(page_mkclean); |
d08b3851 | 717 | |
c44b6743 RR |
718 | /** |
719 | * page_move_anon_rmap - move a page to our anon_vma | |
720 | * @page: the page to move to our anon_vma | |
721 | * @vma: the vma the page belongs to | |
722 | * @address: the user virtual address mapped | |
723 | * | |
724 | * When a page belongs exclusively to one process after a COW event, | |
725 | * that page can be moved into the anon_vma that belongs to just that | |
726 | * process, so the rmap code will not search the parent or sibling | |
727 | * processes. | |
728 | */ | |
729 | void page_move_anon_rmap(struct page *page, | |
730 | struct vm_area_struct *vma, unsigned long address) | |
731 | { | |
732 | struct anon_vma *anon_vma = vma->anon_vma; | |
733 | ||
734 | VM_BUG_ON(!PageLocked(page)); | |
735 | VM_BUG_ON(!anon_vma); | |
736 | VM_BUG_ON(page->index != linear_page_index(vma, address)); | |
737 | ||
738 | anon_vma = (void *) anon_vma + PAGE_MAPPING_ANON; | |
739 | page->mapping = (struct address_space *) anon_vma; | |
740 | } | |
741 | ||
9617d95e | 742 | /** |
43d8eac4 | 743 | * __page_set_anon_rmap - setup new anonymous rmap |
9617d95e NP |
744 | * @page: the page to add the mapping to |
745 | * @vma: the vm area in which the mapping is added | |
746 | * @address: the user virtual address mapped | |
e8a03feb | 747 | * @exclusive: the page is exclusively owned by the current process |
9617d95e NP |
748 | */ |
749 | static void __page_set_anon_rmap(struct page *page, | |
e8a03feb | 750 | struct vm_area_struct *vma, unsigned long address, int exclusive) |
9617d95e | 751 | { |
e8a03feb | 752 | struct anon_vma *anon_vma = vma->anon_vma; |
ea90002b | 753 | |
e8a03feb | 754 | BUG_ON(!anon_vma); |
ea90002b LT |
755 | |
756 | /* | |
e8a03feb RR |
757 | * If the page isn't exclusively mapped into this vma, |
758 | * we must use the _oldest_ possible anon_vma for the | |
759 | * page mapping! | |
ea90002b | 760 | * |
e8a03feb RR |
761 | * So take the last AVC chain entry in the vma, which is |
762 | * the deepest ancestor, and use the anon_vma from that. | |
ea90002b | 763 | */ |
e8a03feb RR |
764 | if (!exclusive) { |
765 | struct anon_vma_chain *avc; | |
766 | avc = list_entry(vma->anon_vma_chain.prev, struct anon_vma_chain, same_vma); | |
767 | anon_vma = avc->anon_vma; | |
768 | } | |
9617d95e | 769 | |
9617d95e NP |
770 | anon_vma = (void *) anon_vma + PAGE_MAPPING_ANON; |
771 | page->mapping = (struct address_space *) anon_vma; | |
9617d95e | 772 | page->index = linear_page_index(vma, address); |
9617d95e NP |
773 | } |
774 | ||
c97a9e10 | 775 | /** |
43d8eac4 | 776 | * __page_check_anon_rmap - sanity check anonymous rmap addition |
c97a9e10 NP |
777 | * @page: the page to add the mapping to |
778 | * @vma: the vm area in which the mapping is added | |
779 | * @address: the user virtual address mapped | |
780 | */ | |
781 | static void __page_check_anon_rmap(struct page *page, | |
782 | struct vm_area_struct *vma, unsigned long address) | |
783 | { | |
784 | #ifdef CONFIG_DEBUG_VM | |
785 | /* | |
786 | * The page's anon-rmap details (mapping and index) are guaranteed to | |
787 | * be set up correctly at this point. | |
788 | * | |
789 | * We have exclusion against page_add_anon_rmap because the caller | |
790 | * always holds the page locked, except if called from page_dup_rmap, | |
791 | * in which case the page is already known to be setup. | |
792 | * | |
793 | * We have exclusion against page_add_new_anon_rmap because those pages | |
794 | * are initially only visible via the pagetables, and the pte is locked | |
795 | * over the call to page_add_new_anon_rmap. | |
796 | */ | |
c97a9e10 NP |
797 | BUG_ON(page->index != linear_page_index(vma, address)); |
798 | #endif | |
799 | } | |
800 | ||
1da177e4 LT |
801 | /** |
802 | * page_add_anon_rmap - add pte mapping to an anonymous page | |
803 | * @page: the page to add the mapping to | |
804 | * @vma: the vm area in which the mapping is added | |
805 | * @address: the user virtual address mapped | |
806 | * | |
5ad64688 | 807 | * The caller needs to hold the pte lock, and the page must be locked in |
80e14822 HD |
808 | * the anon_vma case: to serialize mapping,index checking after setting, |
809 | * and to ensure that PageAnon is not being upgraded racily to PageKsm | |
810 | * (but PageKsm is never downgraded to PageAnon). | |
1da177e4 LT |
811 | */ |
812 | void page_add_anon_rmap(struct page *page, | |
813 | struct vm_area_struct *vma, unsigned long address) | |
814 | { | |
5ad64688 HD |
815 | int first = atomic_inc_and_test(&page->_mapcount); |
816 | if (first) | |
817 | __inc_zone_page_state(page, NR_ANON_PAGES); | |
818 | if (unlikely(PageKsm(page))) | |
819 | return; | |
820 | ||
c97a9e10 NP |
821 | VM_BUG_ON(!PageLocked(page)); |
822 | VM_BUG_ON(address < vma->vm_start || address >= vma->vm_end); | |
5ad64688 | 823 | if (first) |
e8a03feb | 824 | __page_set_anon_rmap(page, vma, address, 0); |
69029cd5 | 825 | else |
c97a9e10 | 826 | __page_check_anon_rmap(page, vma, address); |
1da177e4 LT |
827 | } |
828 | ||
43d8eac4 | 829 | /** |
9617d95e NP |
830 | * page_add_new_anon_rmap - add pte mapping to a new anonymous page |
831 | * @page: the page to add the mapping to | |
832 | * @vma: the vm area in which the mapping is added | |
833 | * @address: the user virtual address mapped | |
834 | * | |
835 | * Same as page_add_anon_rmap but must only be called on *new* pages. | |
836 | * This means the inc-and-test can be bypassed. | |
c97a9e10 | 837 | * Page does not have to be locked. |
9617d95e NP |
838 | */ |
839 | void page_add_new_anon_rmap(struct page *page, | |
840 | struct vm_area_struct *vma, unsigned long address) | |
841 | { | |
b5934c53 | 842 | VM_BUG_ON(address < vma->vm_start || address >= vma->vm_end); |
cbf84b7a HD |
843 | SetPageSwapBacked(page); |
844 | atomic_set(&page->_mapcount, 0); /* increment count (starts at -1) */ | |
5ad64688 | 845 | __inc_zone_page_state(page, NR_ANON_PAGES); |
e8a03feb | 846 | __page_set_anon_rmap(page, vma, address, 1); |
b5934c53 | 847 | if (page_evictable(page, vma)) |
cbf84b7a | 848 | lru_cache_add_lru(page, LRU_ACTIVE_ANON); |
b5934c53 HD |
849 | else |
850 | add_page_to_unevictable_list(page); | |
9617d95e NP |
851 | } |
852 | ||
1da177e4 LT |
853 | /** |
854 | * page_add_file_rmap - add pte mapping to a file page | |
855 | * @page: the page to add the mapping to | |
856 | * | |
b8072f09 | 857 | * The caller needs to hold the pte lock. |
1da177e4 LT |
858 | */ |
859 | void page_add_file_rmap(struct page *page) | |
860 | { | |
d69b042f | 861 | if (atomic_inc_and_test(&page->_mapcount)) { |
65ba55f5 | 862 | __inc_zone_page_state(page, NR_FILE_MAPPED); |
d8046582 | 863 | mem_cgroup_update_file_mapped(page, 1); |
d69b042f | 864 | } |
1da177e4 LT |
865 | } |
866 | ||
867 | /** | |
868 | * page_remove_rmap - take down pte mapping from a page | |
869 | * @page: page to remove mapping from | |
870 | * | |
b8072f09 | 871 | * The caller needs to hold the pte lock. |
1da177e4 | 872 | */ |
edc315fd | 873 | void page_remove_rmap(struct page *page) |
1da177e4 | 874 | { |
b904dcfe KM |
875 | /* page still mapped by someone else? */ |
876 | if (!atomic_add_negative(-1, &page->_mapcount)) | |
877 | return; | |
878 | ||
879 | /* | |
880 | * Now that the last pte has gone, s390 must transfer dirty | |
881 | * flag from storage key to struct page. We can usually skip | |
882 | * this if the page is anon, so about to be freed; but perhaps | |
883 | * not if it's in swapcache - there might be another pte slot | |
884 | * containing the swap entry, but page not yet written to swap. | |
885 | */ | |
886 | if ((!PageAnon(page) || PageSwapCache(page)) && page_test_dirty(page)) { | |
887 | page_clear_dirty(page); | |
888 | set_page_dirty(page); | |
1da177e4 | 889 | } |
b904dcfe KM |
890 | if (PageAnon(page)) { |
891 | mem_cgroup_uncharge_page(page); | |
892 | __dec_zone_page_state(page, NR_ANON_PAGES); | |
893 | } else { | |
894 | __dec_zone_page_state(page, NR_FILE_MAPPED); | |
d8046582 | 895 | mem_cgroup_update_file_mapped(page, -1); |
b904dcfe | 896 | } |
b904dcfe KM |
897 | /* |
898 | * It would be tidy to reset the PageAnon mapping here, | |
899 | * but that might overwrite a racing page_add_anon_rmap | |
900 | * which increments mapcount after us but sets mapping | |
901 | * before us: so leave the reset to free_hot_cold_page, | |
902 | * and remember that it's only reliable while mapped. | |
903 | * Leaving it set also helps swapoff to reinstate ptes | |
904 | * faster for those pages still in swapcache. | |
905 | */ | |
1da177e4 LT |
906 | } |
907 | ||
908 | /* | |
909 | * Subfunctions of try_to_unmap: try_to_unmap_one called | |
910 | * repeatedly from either try_to_unmap_anon or try_to_unmap_file. | |
911 | */ | |
5ad64688 HD |
912 | int try_to_unmap_one(struct page *page, struct vm_area_struct *vma, |
913 | unsigned long address, enum ttu_flags flags) | |
1da177e4 LT |
914 | { |
915 | struct mm_struct *mm = vma->vm_mm; | |
1da177e4 LT |
916 | pte_t *pte; |
917 | pte_t pteval; | |
c0718806 | 918 | spinlock_t *ptl; |
1da177e4 LT |
919 | int ret = SWAP_AGAIN; |
920 | ||
479db0bf | 921 | pte = page_check_address(page, mm, address, &ptl, 0); |
c0718806 | 922 | if (!pte) |
81b4082d | 923 | goto out; |
1da177e4 LT |
924 | |
925 | /* | |
926 | * If the page is mlock()d, we cannot swap it out. | |
927 | * If it's recently referenced (perhaps page_referenced | |
928 | * skipped over this mm) then we should reactivate it. | |
929 | */ | |
14fa31b8 | 930 | if (!(flags & TTU_IGNORE_MLOCK)) { |
caed0f48 KM |
931 | if (vma->vm_flags & VM_LOCKED) |
932 | goto out_mlock; | |
933 | ||
af8e3354 | 934 | if (TTU_ACTION(flags) == TTU_MUNLOCK) |
53f79acb | 935 | goto out_unmap; |
14fa31b8 AK |
936 | } |
937 | if (!(flags & TTU_IGNORE_ACCESS)) { | |
b291f000 NP |
938 | if (ptep_clear_flush_young_notify(vma, address, pte)) { |
939 | ret = SWAP_FAIL; | |
940 | goto out_unmap; | |
941 | } | |
942 | } | |
1da177e4 | 943 | |
1da177e4 LT |
944 | /* Nuke the page table entry. */ |
945 | flush_cache_page(vma, address, page_to_pfn(page)); | |
cddb8a5c | 946 | pteval = ptep_clear_flush_notify(vma, address, pte); |
1da177e4 LT |
947 | |
948 | /* Move the dirty bit to the physical page now the pte is gone. */ | |
949 | if (pte_dirty(pteval)) | |
950 | set_page_dirty(page); | |
951 | ||
365e9c87 HD |
952 | /* Update high watermark before we lower rss */ |
953 | update_hiwater_rss(mm); | |
954 | ||
888b9f7c AK |
955 | if (PageHWPoison(page) && !(flags & TTU_IGNORE_HWPOISON)) { |
956 | if (PageAnon(page)) | |
d559db08 | 957 | dec_mm_counter(mm, MM_ANONPAGES); |
888b9f7c | 958 | else |
d559db08 | 959 | dec_mm_counter(mm, MM_FILEPAGES); |
888b9f7c AK |
960 | set_pte_at(mm, address, pte, |
961 | swp_entry_to_pte(make_hwpoison_entry(page))); | |
962 | } else if (PageAnon(page)) { | |
4c21e2f2 | 963 | swp_entry_t entry = { .val = page_private(page) }; |
0697212a CL |
964 | |
965 | if (PageSwapCache(page)) { | |
966 | /* | |
967 | * Store the swap location in the pte. | |
968 | * See handle_pte_fault() ... | |
969 | */ | |
570a335b HD |
970 | if (swap_duplicate(entry) < 0) { |
971 | set_pte_at(mm, address, pte, pteval); | |
972 | ret = SWAP_FAIL; | |
973 | goto out_unmap; | |
974 | } | |
0697212a CL |
975 | if (list_empty(&mm->mmlist)) { |
976 | spin_lock(&mmlist_lock); | |
977 | if (list_empty(&mm->mmlist)) | |
978 | list_add(&mm->mmlist, &init_mm.mmlist); | |
979 | spin_unlock(&mmlist_lock); | |
980 | } | |
d559db08 | 981 | dec_mm_counter(mm, MM_ANONPAGES); |
b084d435 | 982 | inc_mm_counter(mm, MM_SWAPENTS); |
64cdd548 | 983 | } else if (PAGE_MIGRATION) { |
0697212a CL |
984 | /* |
985 | * Store the pfn of the page in a special migration | |
986 | * pte. do_swap_page() will wait until the migration | |
987 | * pte is removed and then restart fault handling. | |
988 | */ | |
14fa31b8 | 989 | BUG_ON(TTU_ACTION(flags) != TTU_MIGRATION); |
0697212a | 990 | entry = make_migration_entry(page, pte_write(pteval)); |
1da177e4 LT |
991 | } |
992 | set_pte_at(mm, address, pte, swp_entry_to_pte(entry)); | |
993 | BUG_ON(pte_file(*pte)); | |
14fa31b8 | 994 | } else if (PAGE_MIGRATION && (TTU_ACTION(flags) == TTU_MIGRATION)) { |
04e62a29 CL |
995 | /* Establish migration entry for a file page */ |
996 | swp_entry_t entry; | |
997 | entry = make_migration_entry(page, pte_write(pteval)); | |
998 | set_pte_at(mm, address, pte, swp_entry_to_pte(entry)); | |
999 | } else | |
d559db08 | 1000 | dec_mm_counter(mm, MM_FILEPAGES); |
1da177e4 | 1001 | |
edc315fd | 1002 | page_remove_rmap(page); |
1da177e4 LT |
1003 | page_cache_release(page); |
1004 | ||
1005 | out_unmap: | |
c0718806 | 1006 | pte_unmap_unlock(pte, ptl); |
caed0f48 KM |
1007 | out: |
1008 | return ret; | |
53f79acb | 1009 | |
caed0f48 KM |
1010 | out_mlock: |
1011 | pte_unmap_unlock(pte, ptl); | |
1012 | ||
1013 | ||
1014 | /* | |
1015 | * We need mmap_sem locking, Otherwise VM_LOCKED check makes | |
1016 | * unstable result and race. Plus, We can't wait here because | |
1017 | * we now hold anon_vma->lock or mapping->i_mmap_lock. | |
1018 | * if trylock failed, the page remain in evictable lru and later | |
1019 | * vmscan could retry to move the page to unevictable lru if the | |
1020 | * page is actually mlocked. | |
1021 | */ | |
1022 | if (down_read_trylock(&vma->vm_mm->mmap_sem)) { | |
1023 | if (vma->vm_flags & VM_LOCKED) { | |
1024 | mlock_vma_page(page); | |
1025 | ret = SWAP_MLOCK; | |
53f79acb | 1026 | } |
caed0f48 | 1027 | up_read(&vma->vm_mm->mmap_sem); |
53f79acb | 1028 | } |
1da177e4 LT |
1029 | return ret; |
1030 | } | |
1031 | ||
1032 | /* | |
1033 | * objrmap doesn't work for nonlinear VMAs because the assumption that | |
1034 | * offset-into-file correlates with offset-into-virtual-addresses does not hold. | |
1035 | * Consequently, given a particular page and its ->index, we cannot locate the | |
1036 | * ptes which are mapping that page without an exhaustive linear search. | |
1037 | * | |
1038 | * So what this code does is a mini "virtual scan" of each nonlinear VMA which | |
1039 | * maps the file to which the target page belongs. The ->vm_private_data field | |
1040 | * holds the current cursor into that scan. Successive searches will circulate | |
1041 | * around the vma's virtual address space. | |
1042 | * | |
1043 | * So as more replacement pressure is applied to the pages in a nonlinear VMA, | |
1044 | * more scanning pressure is placed against them as well. Eventually pages | |
1045 | * will become fully unmapped and are eligible for eviction. | |
1046 | * | |
1047 | * For very sparsely populated VMAs this is a little inefficient - chances are | |
1048 | * there there won't be many ptes located within the scan cluster. In this case | |
1049 | * maybe we could scan further - to the end of the pte page, perhaps. | |
b291f000 NP |
1050 | * |
1051 | * Mlocked pages: check VM_LOCKED under mmap_sem held for read, if we can | |
1052 | * acquire it without blocking. If vma locked, mlock the pages in the cluster, | |
1053 | * rather than unmapping them. If we encounter the "check_page" that vmscan is | |
1054 | * trying to unmap, return SWAP_MLOCK, else default SWAP_AGAIN. | |
1da177e4 LT |
1055 | */ |
1056 | #define CLUSTER_SIZE min(32*PAGE_SIZE, PMD_SIZE) | |
1057 | #define CLUSTER_MASK (~(CLUSTER_SIZE - 1)) | |
1058 | ||
b291f000 NP |
1059 | static int try_to_unmap_cluster(unsigned long cursor, unsigned int *mapcount, |
1060 | struct vm_area_struct *vma, struct page *check_page) | |
1da177e4 LT |
1061 | { |
1062 | struct mm_struct *mm = vma->vm_mm; | |
1063 | pgd_t *pgd; | |
1064 | pud_t *pud; | |
1065 | pmd_t *pmd; | |
c0718806 | 1066 | pte_t *pte; |
1da177e4 | 1067 | pte_t pteval; |
c0718806 | 1068 | spinlock_t *ptl; |
1da177e4 LT |
1069 | struct page *page; |
1070 | unsigned long address; | |
1071 | unsigned long end; | |
b291f000 NP |
1072 | int ret = SWAP_AGAIN; |
1073 | int locked_vma = 0; | |
1da177e4 | 1074 | |
1da177e4 LT |
1075 | address = (vma->vm_start + cursor) & CLUSTER_MASK; |
1076 | end = address + CLUSTER_SIZE; | |
1077 | if (address < vma->vm_start) | |
1078 | address = vma->vm_start; | |
1079 | if (end > vma->vm_end) | |
1080 | end = vma->vm_end; | |
1081 | ||
1082 | pgd = pgd_offset(mm, address); | |
1083 | if (!pgd_present(*pgd)) | |
b291f000 | 1084 | return ret; |
1da177e4 LT |
1085 | |
1086 | pud = pud_offset(pgd, address); | |
1087 | if (!pud_present(*pud)) | |
b291f000 | 1088 | return ret; |
1da177e4 LT |
1089 | |
1090 | pmd = pmd_offset(pud, address); | |
1091 | if (!pmd_present(*pmd)) | |
b291f000 NP |
1092 | return ret; |
1093 | ||
1094 | /* | |
af8e3354 | 1095 | * If we can acquire the mmap_sem for read, and vma is VM_LOCKED, |
b291f000 NP |
1096 | * keep the sem while scanning the cluster for mlocking pages. |
1097 | */ | |
af8e3354 | 1098 | if (down_read_trylock(&vma->vm_mm->mmap_sem)) { |
b291f000 NP |
1099 | locked_vma = (vma->vm_flags & VM_LOCKED); |
1100 | if (!locked_vma) | |
1101 | up_read(&vma->vm_mm->mmap_sem); /* don't need it */ | |
1102 | } | |
c0718806 HD |
1103 | |
1104 | pte = pte_offset_map_lock(mm, pmd, address, &ptl); | |
1da177e4 | 1105 | |
365e9c87 HD |
1106 | /* Update high watermark before we lower rss */ |
1107 | update_hiwater_rss(mm); | |
1108 | ||
c0718806 | 1109 | for (; address < end; pte++, address += PAGE_SIZE) { |
1da177e4 LT |
1110 | if (!pte_present(*pte)) |
1111 | continue; | |
6aab341e LT |
1112 | page = vm_normal_page(vma, address, *pte); |
1113 | BUG_ON(!page || PageAnon(page)); | |
1da177e4 | 1114 | |
b291f000 NP |
1115 | if (locked_vma) { |
1116 | mlock_vma_page(page); /* no-op if already mlocked */ | |
1117 | if (page == check_page) | |
1118 | ret = SWAP_MLOCK; | |
1119 | continue; /* don't unmap */ | |
1120 | } | |
1121 | ||
cddb8a5c | 1122 | if (ptep_clear_flush_young_notify(vma, address, pte)) |
1da177e4 LT |
1123 | continue; |
1124 | ||
1125 | /* Nuke the page table entry. */ | |
eca35133 | 1126 | flush_cache_page(vma, address, pte_pfn(*pte)); |
cddb8a5c | 1127 | pteval = ptep_clear_flush_notify(vma, address, pte); |
1da177e4 LT |
1128 | |
1129 | /* If nonlinear, store the file page offset in the pte. */ | |
1130 | if (page->index != linear_page_index(vma, address)) | |
1131 | set_pte_at(mm, address, pte, pgoff_to_pte(page->index)); | |
1132 | ||
1133 | /* Move the dirty bit to the physical page now the pte is gone. */ | |
1134 | if (pte_dirty(pteval)) | |
1135 | set_page_dirty(page); | |
1136 | ||
edc315fd | 1137 | page_remove_rmap(page); |
1da177e4 | 1138 | page_cache_release(page); |
d559db08 | 1139 | dec_mm_counter(mm, MM_FILEPAGES); |
1da177e4 LT |
1140 | (*mapcount)--; |
1141 | } | |
c0718806 | 1142 | pte_unmap_unlock(pte - 1, ptl); |
b291f000 NP |
1143 | if (locked_vma) |
1144 | up_read(&vma->vm_mm->mmap_sem); | |
1145 | return ret; | |
1da177e4 LT |
1146 | } |
1147 | ||
a8bef8ff MG |
1148 | static bool is_vma_temporary_stack(struct vm_area_struct *vma) |
1149 | { | |
1150 | int maybe_stack = vma->vm_flags & (VM_GROWSDOWN | VM_GROWSUP); | |
1151 | ||
1152 | if (!maybe_stack) | |
1153 | return false; | |
1154 | ||
1155 | if ((vma->vm_flags & VM_STACK_INCOMPLETE_SETUP) == | |
1156 | VM_STACK_INCOMPLETE_SETUP) | |
1157 | return true; | |
1158 | ||
1159 | return false; | |
1160 | } | |
1161 | ||
b291f000 NP |
1162 | /** |
1163 | * try_to_unmap_anon - unmap or unlock anonymous page using the object-based | |
1164 | * rmap method | |
1165 | * @page: the page to unmap/unlock | |
8051be5e | 1166 | * @flags: action and flags |
b291f000 NP |
1167 | * |
1168 | * Find all the mappings of a page using the mapping pointer and the vma chains | |
1169 | * contained in the anon_vma struct it points to. | |
1170 | * | |
1171 | * This function is only called from try_to_unmap/try_to_munlock for | |
1172 | * anonymous pages. | |
1173 | * When called from try_to_munlock(), the mmap_sem of the mm containing the vma | |
1174 | * where the page was found will be held for write. So, we won't recheck | |
1175 | * vm_flags for that VMA. That should be OK, because that vma shouldn't be | |
1176 | * 'LOCKED. | |
1177 | */ | |
14fa31b8 | 1178 | static int try_to_unmap_anon(struct page *page, enum ttu_flags flags) |
1da177e4 LT |
1179 | { |
1180 | struct anon_vma *anon_vma; | |
5beb4930 | 1181 | struct anon_vma_chain *avc; |
1da177e4 | 1182 | int ret = SWAP_AGAIN; |
b291f000 | 1183 | |
1da177e4 LT |
1184 | anon_vma = page_lock_anon_vma(page); |
1185 | if (!anon_vma) | |
1186 | return ret; | |
1187 | ||
5beb4930 RR |
1188 | list_for_each_entry(avc, &anon_vma->head, same_anon_vma) { |
1189 | struct vm_area_struct *vma = avc->vma; | |
a8bef8ff MG |
1190 | unsigned long address; |
1191 | ||
1192 | /* | |
1193 | * During exec, a temporary VMA is setup and later moved. | |
1194 | * The VMA is moved under the anon_vma lock but not the | |
1195 | * page tables leading to a race where migration cannot | |
1196 | * find the migration ptes. Rather than increasing the | |
1197 | * locking requirements of exec(), migration skips | |
1198 | * temporary VMAs until after exec() completes. | |
1199 | */ | |
1200 | if (PAGE_MIGRATION && (flags & TTU_MIGRATION) && | |
1201 | is_vma_temporary_stack(vma)) | |
1202 | continue; | |
1203 | ||
1204 | address = vma_address(page, vma); | |
1cb1729b HD |
1205 | if (address == -EFAULT) |
1206 | continue; | |
1207 | ret = try_to_unmap_one(page, vma, address, flags); | |
53f79acb HD |
1208 | if (ret != SWAP_AGAIN || !page_mapped(page)) |
1209 | break; | |
1da177e4 | 1210 | } |
34bbd704 ON |
1211 | |
1212 | page_unlock_anon_vma(anon_vma); | |
1da177e4 LT |
1213 | return ret; |
1214 | } | |
1215 | ||
1216 | /** | |
b291f000 NP |
1217 | * try_to_unmap_file - unmap/unlock file page using the object-based rmap method |
1218 | * @page: the page to unmap/unlock | |
14fa31b8 | 1219 | * @flags: action and flags |
1da177e4 LT |
1220 | * |
1221 | * Find all the mappings of a page using the mapping pointer and the vma chains | |
1222 | * contained in the address_space struct it points to. | |
1223 | * | |
b291f000 NP |
1224 | * This function is only called from try_to_unmap/try_to_munlock for |
1225 | * object-based pages. | |
1226 | * When called from try_to_munlock(), the mmap_sem of the mm containing the vma | |
1227 | * where the page was found will be held for write. So, we won't recheck | |
1228 | * vm_flags for that VMA. That should be OK, because that vma shouldn't be | |
1229 | * 'LOCKED. | |
1da177e4 | 1230 | */ |
14fa31b8 | 1231 | static int try_to_unmap_file(struct page *page, enum ttu_flags flags) |
1da177e4 LT |
1232 | { |
1233 | struct address_space *mapping = page->mapping; | |
1234 | pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT); | |
1235 | struct vm_area_struct *vma; | |
1236 | struct prio_tree_iter iter; | |
1237 | int ret = SWAP_AGAIN; | |
1238 | unsigned long cursor; | |
1239 | unsigned long max_nl_cursor = 0; | |
1240 | unsigned long max_nl_size = 0; | |
1241 | unsigned int mapcount; | |
1242 | ||
1243 | spin_lock(&mapping->i_mmap_lock); | |
1244 | vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, pgoff, pgoff) { | |
1cb1729b HD |
1245 | unsigned long address = vma_address(page, vma); |
1246 | if (address == -EFAULT) | |
1247 | continue; | |
1248 | ret = try_to_unmap_one(page, vma, address, flags); | |
53f79acb HD |
1249 | if (ret != SWAP_AGAIN || !page_mapped(page)) |
1250 | goto out; | |
1da177e4 LT |
1251 | } |
1252 | ||
1253 | if (list_empty(&mapping->i_mmap_nonlinear)) | |
1254 | goto out; | |
1255 | ||
53f79acb HD |
1256 | /* |
1257 | * We don't bother to try to find the munlocked page in nonlinears. | |
1258 | * It's costly. Instead, later, page reclaim logic may call | |
1259 | * try_to_unmap(TTU_MUNLOCK) and recover PG_mlocked lazily. | |
1260 | */ | |
1261 | if (TTU_ACTION(flags) == TTU_MUNLOCK) | |
1262 | goto out; | |
1263 | ||
1da177e4 LT |
1264 | list_for_each_entry(vma, &mapping->i_mmap_nonlinear, |
1265 | shared.vm_set.list) { | |
1da177e4 LT |
1266 | cursor = (unsigned long) vma->vm_private_data; |
1267 | if (cursor > max_nl_cursor) | |
1268 | max_nl_cursor = cursor; | |
1269 | cursor = vma->vm_end - vma->vm_start; | |
1270 | if (cursor > max_nl_size) | |
1271 | max_nl_size = cursor; | |
1272 | } | |
1273 | ||
b291f000 | 1274 | if (max_nl_size == 0) { /* all nonlinears locked or reserved ? */ |
1da177e4 LT |
1275 | ret = SWAP_FAIL; |
1276 | goto out; | |
1277 | } | |
1278 | ||
1279 | /* | |
1280 | * We don't try to search for this page in the nonlinear vmas, | |
1281 | * and page_referenced wouldn't have found it anyway. Instead | |
1282 | * just walk the nonlinear vmas trying to age and unmap some. | |
1283 | * The mapcount of the page we came in with is irrelevant, | |
1284 | * but even so use it as a guide to how hard we should try? | |
1285 | */ | |
1286 | mapcount = page_mapcount(page); | |
1287 | if (!mapcount) | |
1288 | goto out; | |
1289 | cond_resched_lock(&mapping->i_mmap_lock); | |
1290 | ||
1291 | max_nl_size = (max_nl_size + CLUSTER_SIZE - 1) & CLUSTER_MASK; | |
1292 | if (max_nl_cursor == 0) | |
1293 | max_nl_cursor = CLUSTER_SIZE; | |
1294 | ||
1295 | do { | |
1296 | list_for_each_entry(vma, &mapping->i_mmap_nonlinear, | |
1297 | shared.vm_set.list) { | |
1da177e4 | 1298 | cursor = (unsigned long) vma->vm_private_data; |
839b9685 | 1299 | while ( cursor < max_nl_cursor && |
1da177e4 | 1300 | cursor < vma->vm_end - vma->vm_start) { |
53f79acb HD |
1301 | if (try_to_unmap_cluster(cursor, &mapcount, |
1302 | vma, page) == SWAP_MLOCK) | |
1303 | ret = SWAP_MLOCK; | |
1da177e4 LT |
1304 | cursor += CLUSTER_SIZE; |
1305 | vma->vm_private_data = (void *) cursor; | |
1306 | if ((int)mapcount <= 0) | |
1307 | goto out; | |
1308 | } | |
1309 | vma->vm_private_data = (void *) max_nl_cursor; | |
1310 | } | |
1311 | cond_resched_lock(&mapping->i_mmap_lock); | |
1312 | max_nl_cursor += CLUSTER_SIZE; | |
1313 | } while (max_nl_cursor <= max_nl_size); | |
1314 | ||
1315 | /* | |
1316 | * Don't loop forever (perhaps all the remaining pages are | |
1317 | * in locked vmas). Reset cursor on all unreserved nonlinear | |
1318 | * vmas, now forgetting on which ones it had fallen behind. | |
1319 | */ | |
101d2be7 HD |
1320 | list_for_each_entry(vma, &mapping->i_mmap_nonlinear, shared.vm_set.list) |
1321 | vma->vm_private_data = NULL; | |
1da177e4 LT |
1322 | out: |
1323 | spin_unlock(&mapping->i_mmap_lock); | |
1324 | return ret; | |
1325 | } | |
1326 | ||
1327 | /** | |
1328 | * try_to_unmap - try to remove all page table mappings to a page | |
1329 | * @page: the page to get unmapped | |
14fa31b8 | 1330 | * @flags: action and flags |
1da177e4 LT |
1331 | * |
1332 | * Tries to remove all the page table entries which are mapping this | |
1333 | * page, used in the pageout path. Caller must hold the page lock. | |
1334 | * Return values are: | |
1335 | * | |
1336 | * SWAP_SUCCESS - we succeeded in removing all mappings | |
1337 | * SWAP_AGAIN - we missed a mapping, try again later | |
1338 | * SWAP_FAIL - the page is unswappable | |
b291f000 | 1339 | * SWAP_MLOCK - page is mlocked. |
1da177e4 | 1340 | */ |
14fa31b8 | 1341 | int try_to_unmap(struct page *page, enum ttu_flags flags) |
1da177e4 LT |
1342 | { |
1343 | int ret; | |
1344 | ||
1da177e4 LT |
1345 | BUG_ON(!PageLocked(page)); |
1346 | ||
5ad64688 HD |
1347 | if (unlikely(PageKsm(page))) |
1348 | ret = try_to_unmap_ksm(page, flags); | |
1349 | else if (PageAnon(page)) | |
14fa31b8 | 1350 | ret = try_to_unmap_anon(page, flags); |
1da177e4 | 1351 | else |
14fa31b8 | 1352 | ret = try_to_unmap_file(page, flags); |
b291f000 | 1353 | if (ret != SWAP_MLOCK && !page_mapped(page)) |
1da177e4 LT |
1354 | ret = SWAP_SUCCESS; |
1355 | return ret; | |
1356 | } | |
81b4082d | 1357 | |
b291f000 NP |
1358 | /** |
1359 | * try_to_munlock - try to munlock a page | |
1360 | * @page: the page to be munlocked | |
1361 | * | |
1362 | * Called from munlock code. Checks all of the VMAs mapping the page | |
1363 | * to make sure nobody else has this page mlocked. The page will be | |
1364 | * returned with PG_mlocked cleared if no other vmas have it mlocked. | |
1365 | * | |
1366 | * Return values are: | |
1367 | * | |
53f79acb | 1368 | * SWAP_AGAIN - no vma is holding page mlocked, or, |
b291f000 | 1369 | * SWAP_AGAIN - page mapped in mlocked vma -- couldn't acquire mmap sem |
5ad64688 | 1370 | * SWAP_FAIL - page cannot be located at present |
b291f000 NP |
1371 | * SWAP_MLOCK - page is now mlocked. |
1372 | */ | |
1373 | int try_to_munlock(struct page *page) | |
1374 | { | |
1375 | VM_BUG_ON(!PageLocked(page) || PageLRU(page)); | |
1376 | ||
5ad64688 HD |
1377 | if (unlikely(PageKsm(page))) |
1378 | return try_to_unmap_ksm(page, TTU_MUNLOCK); | |
1379 | else if (PageAnon(page)) | |
14fa31b8 | 1380 | return try_to_unmap_anon(page, TTU_MUNLOCK); |
b291f000 | 1381 | else |
14fa31b8 | 1382 | return try_to_unmap_file(page, TTU_MUNLOCK); |
b291f000 | 1383 | } |
e9995ef9 HD |
1384 | |
1385 | #ifdef CONFIG_MIGRATION | |
1386 | /* | |
1387 | * rmap_walk() and its helpers rmap_walk_anon() and rmap_walk_file(): | |
1388 | * Called by migrate.c to remove migration ptes, but might be used more later. | |
1389 | */ | |
1390 | static int rmap_walk_anon(struct page *page, int (*rmap_one)(struct page *, | |
1391 | struct vm_area_struct *, unsigned long, void *), void *arg) | |
1392 | { | |
1393 | struct anon_vma *anon_vma; | |
5beb4930 | 1394 | struct anon_vma_chain *avc; |
e9995ef9 HD |
1395 | int ret = SWAP_AGAIN; |
1396 | ||
1397 | /* | |
1398 | * Note: remove_migration_ptes() cannot use page_lock_anon_vma() | |
1399 | * because that depends on page_mapped(); but not all its usages | |
3f6c8272 MG |
1400 | * are holding mmap_sem. Users without mmap_sem are required to |
1401 | * take a reference count to prevent the anon_vma disappearing | |
e9995ef9 HD |
1402 | */ |
1403 | anon_vma = page_anon_vma(page); | |
1404 | if (!anon_vma) | |
1405 | return ret; | |
cba48b98 | 1406 | anon_vma_lock(anon_vma); |
5beb4930 RR |
1407 | list_for_each_entry(avc, &anon_vma->head, same_anon_vma) { |
1408 | struct vm_area_struct *vma = avc->vma; | |
e9995ef9 HD |
1409 | unsigned long address = vma_address(page, vma); |
1410 | if (address == -EFAULT) | |
1411 | continue; | |
1412 | ret = rmap_one(page, vma, address, arg); | |
1413 | if (ret != SWAP_AGAIN) | |
1414 | break; | |
1415 | } | |
cba48b98 | 1416 | anon_vma_unlock(anon_vma); |
e9995ef9 HD |
1417 | return ret; |
1418 | } | |
1419 | ||
1420 | static int rmap_walk_file(struct page *page, int (*rmap_one)(struct page *, | |
1421 | struct vm_area_struct *, unsigned long, void *), void *arg) | |
1422 | { | |
1423 | struct address_space *mapping = page->mapping; | |
1424 | pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT); | |
1425 | struct vm_area_struct *vma; | |
1426 | struct prio_tree_iter iter; | |
1427 | int ret = SWAP_AGAIN; | |
1428 | ||
1429 | if (!mapping) | |
1430 | return ret; | |
1431 | spin_lock(&mapping->i_mmap_lock); | |
1432 | vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, pgoff, pgoff) { | |
1433 | unsigned long address = vma_address(page, vma); | |
1434 | if (address == -EFAULT) | |
1435 | continue; | |
1436 | ret = rmap_one(page, vma, address, arg); | |
1437 | if (ret != SWAP_AGAIN) | |
1438 | break; | |
1439 | } | |
1440 | /* | |
1441 | * No nonlinear handling: being always shared, nonlinear vmas | |
1442 | * never contain migration ptes. Decide what to do about this | |
1443 | * limitation to linear when we need rmap_walk() on nonlinear. | |
1444 | */ | |
1445 | spin_unlock(&mapping->i_mmap_lock); | |
1446 | return ret; | |
1447 | } | |
1448 | ||
1449 | int rmap_walk(struct page *page, int (*rmap_one)(struct page *, | |
1450 | struct vm_area_struct *, unsigned long, void *), void *arg) | |
1451 | { | |
1452 | VM_BUG_ON(!PageLocked(page)); | |
1453 | ||
1454 | if (unlikely(PageKsm(page))) | |
1455 | return rmap_walk_ksm(page, rmap_one, arg); | |
1456 | else if (PageAnon(page)) | |
1457 | return rmap_walk_anon(page, rmap_one, arg); | |
1458 | else | |
1459 | return rmap_walk_file(page, rmap_one, arg); | |
1460 | } | |
1461 | #endif /* CONFIG_MIGRATION */ |