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; | |
76545066 RR |
238 | /* |
239 | * With KSM refcounts, an anon_vma can stay around longer than the | |
240 | * process it belongs to. The root anon_vma needs to be pinned | |
241 | * until this anon_vma is freed, because the lock lives in the root. | |
242 | */ | |
243 | get_anon_vma(anon_vma->root); | |
5beb4930 RR |
244 | /* Mark this anon_vma as the one where our new (COWed) pages go. */ |
245 | vma->anon_vma = anon_vma; | |
5c341ee1 | 246 | anon_vma_chain_link(vma, avc, anon_vma); |
5beb4930 RR |
247 | |
248 | return 0; | |
249 | ||
250 | out_error_free_anon_vma: | |
251 | anon_vma_free(anon_vma); | |
252 | out_error: | |
4946d54c | 253 | unlink_anon_vmas(vma); |
5beb4930 | 254 | return -ENOMEM; |
1da177e4 LT |
255 | } |
256 | ||
5beb4930 | 257 | static void anon_vma_unlink(struct anon_vma_chain *anon_vma_chain) |
1da177e4 | 258 | { |
5beb4930 | 259 | struct anon_vma *anon_vma = anon_vma_chain->anon_vma; |
1da177e4 LT |
260 | int empty; |
261 | ||
5beb4930 | 262 | /* If anon_vma_fork fails, we can get an empty anon_vma_chain. */ |
1da177e4 LT |
263 | if (!anon_vma) |
264 | return; | |
265 | ||
cba48b98 | 266 | anon_vma_lock(anon_vma); |
5beb4930 | 267 | list_del(&anon_vma_chain->same_anon_vma); |
1da177e4 LT |
268 | |
269 | /* We must garbage collect the anon_vma if it's empty */ | |
7f60c214 | 270 | empty = list_empty(&anon_vma->head) && !anonvma_external_refcount(anon_vma); |
cba48b98 | 271 | anon_vma_unlock(anon_vma); |
1da177e4 | 272 | |
76545066 RR |
273 | if (empty) { |
274 | /* We no longer need the root anon_vma */ | |
275 | if (anon_vma->root != anon_vma) | |
276 | drop_anon_vma(anon_vma->root); | |
1da177e4 | 277 | anon_vma_free(anon_vma); |
76545066 | 278 | } |
1da177e4 LT |
279 | } |
280 | ||
5beb4930 RR |
281 | void unlink_anon_vmas(struct vm_area_struct *vma) |
282 | { | |
283 | struct anon_vma_chain *avc, *next; | |
284 | ||
5c341ee1 RR |
285 | /* |
286 | * Unlink each anon_vma chained to the VMA. This list is ordered | |
287 | * from newest to oldest, ensuring the root anon_vma gets freed last. | |
288 | */ | |
5beb4930 RR |
289 | list_for_each_entry_safe(avc, next, &vma->anon_vma_chain, same_vma) { |
290 | anon_vma_unlink(avc); | |
291 | list_del(&avc->same_vma); | |
292 | anon_vma_chain_free(avc); | |
293 | } | |
294 | } | |
295 | ||
51cc5068 | 296 | static void anon_vma_ctor(void *data) |
1da177e4 | 297 | { |
a35afb83 | 298 | struct anon_vma *anon_vma = data; |
1da177e4 | 299 | |
a35afb83 | 300 | spin_lock_init(&anon_vma->lock); |
7f60c214 | 301 | anonvma_external_refcount_init(anon_vma); |
a35afb83 | 302 | INIT_LIST_HEAD(&anon_vma->head); |
1da177e4 LT |
303 | } |
304 | ||
305 | void __init anon_vma_init(void) | |
306 | { | |
307 | anon_vma_cachep = kmem_cache_create("anon_vma", sizeof(struct anon_vma), | |
20c2df83 | 308 | 0, SLAB_DESTROY_BY_RCU|SLAB_PANIC, anon_vma_ctor); |
5beb4930 | 309 | anon_vma_chain_cachep = KMEM_CACHE(anon_vma_chain, SLAB_PANIC); |
1da177e4 LT |
310 | } |
311 | ||
312 | /* | |
313 | * Getting a lock on a stable anon_vma from a page off the LRU is | |
314 | * tricky: page_lock_anon_vma rely on RCU to guard against the races. | |
315 | */ | |
10be22df | 316 | struct anon_vma *page_lock_anon_vma(struct page *page) |
1da177e4 | 317 | { |
34bbd704 | 318 | struct anon_vma *anon_vma; |
1da177e4 LT |
319 | unsigned long anon_mapping; |
320 | ||
321 | rcu_read_lock(); | |
80e14822 | 322 | anon_mapping = (unsigned long) ACCESS_ONCE(page->mapping); |
3ca7b3c5 | 323 | if ((anon_mapping & PAGE_MAPPING_FLAGS) != PAGE_MAPPING_ANON) |
1da177e4 LT |
324 | goto out; |
325 | if (!page_mapped(page)) | |
326 | goto out; | |
327 | ||
328 | anon_vma = (struct anon_vma *) (anon_mapping - PAGE_MAPPING_ANON); | |
cba48b98 | 329 | anon_vma_lock(anon_vma); |
34bbd704 | 330 | return anon_vma; |
1da177e4 LT |
331 | out: |
332 | rcu_read_unlock(); | |
34bbd704 ON |
333 | return NULL; |
334 | } | |
335 | ||
10be22df | 336 | void page_unlock_anon_vma(struct anon_vma *anon_vma) |
34bbd704 | 337 | { |
cba48b98 | 338 | anon_vma_unlock(anon_vma); |
34bbd704 | 339 | rcu_read_unlock(); |
1da177e4 LT |
340 | } |
341 | ||
342 | /* | |
3ad33b24 LS |
343 | * At what user virtual address is page expected in @vma? |
344 | * Returns virtual address or -EFAULT if page's index/offset is not | |
345 | * within the range mapped the @vma. | |
1da177e4 LT |
346 | */ |
347 | static inline unsigned long | |
348 | vma_address(struct page *page, struct vm_area_struct *vma) | |
349 | { | |
350 | pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT); | |
351 | unsigned long address; | |
352 | ||
353 | address = vma->vm_start + ((pgoff - vma->vm_pgoff) << PAGE_SHIFT); | |
354 | if (unlikely(address < vma->vm_start || address >= vma->vm_end)) { | |
3ad33b24 | 355 | /* page should be within @vma mapping range */ |
1da177e4 LT |
356 | return -EFAULT; |
357 | } | |
358 | return address; | |
359 | } | |
360 | ||
361 | /* | |
bf89c8c8 | 362 | * At what user virtual address is page expected in vma? |
ab941e0f | 363 | * Caller should check the page is actually part of the vma. |
1da177e4 LT |
364 | */ |
365 | unsigned long page_address_in_vma(struct page *page, struct vm_area_struct *vma) | |
366 | { | |
ab941e0f NH |
367 | if (PageAnon(page)) |
368 | ; | |
369 | else if (page->mapping && !(vma->vm_flags & VM_NONLINEAR)) { | |
ee498ed7 HD |
370 | if (!vma->vm_file || |
371 | vma->vm_file->f_mapping != page->mapping) | |
1da177e4 LT |
372 | return -EFAULT; |
373 | } else | |
374 | return -EFAULT; | |
375 | return vma_address(page, vma); | |
376 | } | |
377 | ||
81b4082d ND |
378 | /* |
379 | * Check that @page is mapped at @address into @mm. | |
380 | * | |
479db0bf NP |
381 | * If @sync is false, page_check_address may perform a racy check to avoid |
382 | * the page table lock when the pte is not present (helpful when reclaiming | |
383 | * highly shared pages). | |
384 | * | |
b8072f09 | 385 | * On success returns with pte mapped and locked. |
81b4082d | 386 | */ |
ceffc078 | 387 | pte_t *page_check_address(struct page *page, struct mm_struct *mm, |
479db0bf | 388 | unsigned long address, spinlock_t **ptlp, int sync) |
81b4082d ND |
389 | { |
390 | pgd_t *pgd; | |
391 | pud_t *pud; | |
392 | pmd_t *pmd; | |
393 | pte_t *pte; | |
c0718806 | 394 | spinlock_t *ptl; |
81b4082d | 395 | |
81b4082d | 396 | pgd = pgd_offset(mm, address); |
c0718806 HD |
397 | if (!pgd_present(*pgd)) |
398 | return NULL; | |
399 | ||
400 | pud = pud_offset(pgd, address); | |
401 | if (!pud_present(*pud)) | |
402 | return NULL; | |
403 | ||
404 | pmd = pmd_offset(pud, address); | |
405 | if (!pmd_present(*pmd)) | |
406 | return NULL; | |
407 | ||
408 | pte = pte_offset_map(pmd, address); | |
409 | /* Make a quick check before getting the lock */ | |
479db0bf | 410 | if (!sync && !pte_present(*pte)) { |
c0718806 HD |
411 | pte_unmap(pte); |
412 | return NULL; | |
413 | } | |
414 | ||
4c21e2f2 | 415 | ptl = pte_lockptr(mm, pmd); |
c0718806 HD |
416 | spin_lock(ptl); |
417 | if (pte_present(*pte) && page_to_pfn(page) == pte_pfn(*pte)) { | |
418 | *ptlp = ptl; | |
419 | return pte; | |
81b4082d | 420 | } |
c0718806 HD |
421 | pte_unmap_unlock(pte, ptl); |
422 | return NULL; | |
81b4082d ND |
423 | } |
424 | ||
b291f000 NP |
425 | /** |
426 | * page_mapped_in_vma - check whether a page is really mapped in a VMA | |
427 | * @page: the page to test | |
428 | * @vma: the VMA to test | |
429 | * | |
430 | * Returns 1 if the page is mapped into the page tables of the VMA, 0 | |
431 | * if the page is not mapped into the page tables of this VMA. Only | |
432 | * valid for normal file or anonymous VMAs. | |
433 | */ | |
6a46079c | 434 | int page_mapped_in_vma(struct page *page, struct vm_area_struct *vma) |
b291f000 NP |
435 | { |
436 | unsigned long address; | |
437 | pte_t *pte; | |
438 | spinlock_t *ptl; | |
439 | ||
440 | address = vma_address(page, vma); | |
441 | if (address == -EFAULT) /* out of vma range */ | |
442 | return 0; | |
443 | pte = page_check_address(page, vma->vm_mm, address, &ptl, 1); | |
444 | if (!pte) /* the page is not in this mm */ | |
445 | return 0; | |
446 | pte_unmap_unlock(pte, ptl); | |
447 | ||
448 | return 1; | |
449 | } | |
450 | ||
1da177e4 LT |
451 | /* |
452 | * Subfunctions of page_referenced: page_referenced_one called | |
453 | * repeatedly from either page_referenced_anon or page_referenced_file. | |
454 | */ | |
5ad64688 HD |
455 | int page_referenced_one(struct page *page, struct vm_area_struct *vma, |
456 | unsigned long address, unsigned int *mapcount, | |
457 | unsigned long *vm_flags) | |
1da177e4 LT |
458 | { |
459 | struct mm_struct *mm = vma->vm_mm; | |
1da177e4 | 460 | pte_t *pte; |
c0718806 | 461 | spinlock_t *ptl; |
1da177e4 LT |
462 | int referenced = 0; |
463 | ||
479db0bf | 464 | pte = page_check_address(page, mm, address, &ptl, 0); |
c0718806 HD |
465 | if (!pte) |
466 | goto out; | |
1da177e4 | 467 | |
b291f000 NP |
468 | /* |
469 | * Don't want to elevate referenced for mlocked page that gets this far, | |
470 | * in order that it progresses to try_to_unmap and is moved to the | |
471 | * unevictable list. | |
472 | */ | |
5a9bbdcd | 473 | if (vma->vm_flags & VM_LOCKED) { |
5a9bbdcd | 474 | *mapcount = 1; /* break early from loop */ |
03ef83af | 475 | *vm_flags |= VM_LOCKED; |
b291f000 NP |
476 | goto out_unmap; |
477 | } | |
478 | ||
4917e5d0 JW |
479 | if (ptep_clear_flush_young_notify(vma, address, pte)) { |
480 | /* | |
481 | * Don't treat a reference through a sequentially read | |
482 | * mapping as such. If the page has been used in | |
483 | * another mapping, we will catch it; if this other | |
484 | * mapping is already gone, the unmap path will have | |
485 | * set PG_referenced or activated the page. | |
486 | */ | |
487 | if (likely(!VM_SequentialReadHint(vma))) | |
488 | referenced++; | |
489 | } | |
1da177e4 | 490 | |
c0718806 HD |
491 | /* Pretend the page is referenced if the task has the |
492 | swap token and is in the middle of a page fault. */ | |
f7b7fd8f | 493 | if (mm != current->mm && has_swap_token(mm) && |
c0718806 HD |
494 | rwsem_is_locked(&mm->mmap_sem)) |
495 | referenced++; | |
496 | ||
b291f000 | 497 | out_unmap: |
c0718806 HD |
498 | (*mapcount)--; |
499 | pte_unmap_unlock(pte, ptl); | |
273f047e | 500 | |
6fe6b7e3 WF |
501 | if (referenced) |
502 | *vm_flags |= vma->vm_flags; | |
273f047e | 503 | out: |
1da177e4 LT |
504 | return referenced; |
505 | } | |
506 | ||
bed7161a | 507 | static int page_referenced_anon(struct page *page, |
6fe6b7e3 WF |
508 | struct mem_cgroup *mem_cont, |
509 | unsigned long *vm_flags) | |
1da177e4 LT |
510 | { |
511 | unsigned int mapcount; | |
512 | struct anon_vma *anon_vma; | |
5beb4930 | 513 | struct anon_vma_chain *avc; |
1da177e4 LT |
514 | int referenced = 0; |
515 | ||
516 | anon_vma = page_lock_anon_vma(page); | |
517 | if (!anon_vma) | |
518 | return referenced; | |
519 | ||
520 | mapcount = page_mapcount(page); | |
5beb4930 RR |
521 | list_for_each_entry(avc, &anon_vma->head, same_anon_vma) { |
522 | struct vm_area_struct *vma = avc->vma; | |
1cb1729b HD |
523 | unsigned long address = vma_address(page, vma); |
524 | if (address == -EFAULT) | |
525 | continue; | |
bed7161a BS |
526 | /* |
527 | * If we are reclaiming on behalf of a cgroup, skip | |
528 | * counting on behalf of references from different | |
529 | * cgroups | |
530 | */ | |
bd845e38 | 531 | if (mem_cont && !mm_match_cgroup(vma->vm_mm, mem_cont)) |
bed7161a | 532 | continue; |
1cb1729b | 533 | referenced += page_referenced_one(page, vma, address, |
6fe6b7e3 | 534 | &mapcount, vm_flags); |
1da177e4 LT |
535 | if (!mapcount) |
536 | break; | |
537 | } | |
34bbd704 ON |
538 | |
539 | page_unlock_anon_vma(anon_vma); | |
1da177e4 LT |
540 | return referenced; |
541 | } | |
542 | ||
543 | /** | |
544 | * page_referenced_file - referenced check for object-based rmap | |
545 | * @page: the page we're checking references on. | |
43d8eac4 | 546 | * @mem_cont: target memory controller |
6fe6b7e3 | 547 | * @vm_flags: collect encountered vma->vm_flags who actually referenced the page |
1da177e4 LT |
548 | * |
549 | * For an object-based mapped page, find all the places it is mapped and | |
550 | * check/clear the referenced flag. This is done by following the page->mapping | |
551 | * pointer, then walking the chain of vmas it holds. It returns the number | |
552 | * of references it found. | |
553 | * | |
554 | * This function is only called from page_referenced for object-based pages. | |
555 | */ | |
bed7161a | 556 | static int page_referenced_file(struct page *page, |
6fe6b7e3 WF |
557 | struct mem_cgroup *mem_cont, |
558 | unsigned long *vm_flags) | |
1da177e4 LT |
559 | { |
560 | unsigned int mapcount; | |
561 | struct address_space *mapping = page->mapping; | |
562 | pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT); | |
563 | struct vm_area_struct *vma; | |
564 | struct prio_tree_iter iter; | |
565 | int referenced = 0; | |
566 | ||
567 | /* | |
568 | * The caller's checks on page->mapping and !PageAnon have made | |
569 | * sure that this is a file page: the check for page->mapping | |
570 | * excludes the case just before it gets set on an anon page. | |
571 | */ | |
572 | BUG_ON(PageAnon(page)); | |
573 | ||
574 | /* | |
575 | * The page lock not only makes sure that page->mapping cannot | |
576 | * suddenly be NULLified by truncation, it makes sure that the | |
577 | * structure at mapping cannot be freed and reused yet, | |
578 | * so we can safely take mapping->i_mmap_lock. | |
579 | */ | |
580 | BUG_ON(!PageLocked(page)); | |
581 | ||
582 | spin_lock(&mapping->i_mmap_lock); | |
583 | ||
584 | /* | |
585 | * i_mmap_lock does not stabilize mapcount at all, but mapcount | |
586 | * is more likely to be accurate if we note it after spinning. | |
587 | */ | |
588 | mapcount = page_mapcount(page); | |
589 | ||
590 | vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, pgoff, pgoff) { | |
1cb1729b HD |
591 | unsigned long address = vma_address(page, vma); |
592 | if (address == -EFAULT) | |
593 | continue; | |
bed7161a BS |
594 | /* |
595 | * If we are reclaiming on behalf of a cgroup, skip | |
596 | * counting on behalf of references from different | |
597 | * cgroups | |
598 | */ | |
bd845e38 | 599 | if (mem_cont && !mm_match_cgroup(vma->vm_mm, mem_cont)) |
bed7161a | 600 | continue; |
1cb1729b | 601 | referenced += page_referenced_one(page, vma, address, |
6fe6b7e3 | 602 | &mapcount, vm_flags); |
1da177e4 LT |
603 | if (!mapcount) |
604 | break; | |
605 | } | |
606 | ||
607 | spin_unlock(&mapping->i_mmap_lock); | |
608 | return referenced; | |
609 | } | |
610 | ||
611 | /** | |
612 | * page_referenced - test if the page was referenced | |
613 | * @page: the page to test | |
614 | * @is_locked: caller holds lock on the page | |
43d8eac4 | 615 | * @mem_cont: target memory controller |
6fe6b7e3 | 616 | * @vm_flags: collect encountered vma->vm_flags who actually referenced the page |
1da177e4 LT |
617 | * |
618 | * Quick test_and_clear_referenced for all mappings to a page, | |
619 | * returns the number of ptes which referenced the page. | |
620 | */ | |
6fe6b7e3 WF |
621 | int page_referenced(struct page *page, |
622 | int is_locked, | |
623 | struct mem_cgroup *mem_cont, | |
624 | unsigned long *vm_flags) | |
1da177e4 LT |
625 | { |
626 | int referenced = 0; | |
5ad64688 | 627 | int we_locked = 0; |
1da177e4 | 628 | |
6fe6b7e3 | 629 | *vm_flags = 0; |
3ca7b3c5 | 630 | if (page_mapped(page) && page_rmapping(page)) { |
5ad64688 HD |
631 | if (!is_locked && (!PageAnon(page) || PageKsm(page))) { |
632 | we_locked = trylock_page(page); | |
633 | if (!we_locked) { | |
634 | referenced++; | |
635 | goto out; | |
636 | } | |
637 | } | |
638 | if (unlikely(PageKsm(page))) | |
639 | referenced += page_referenced_ksm(page, mem_cont, | |
640 | vm_flags); | |
641 | else if (PageAnon(page)) | |
6fe6b7e3 WF |
642 | referenced += page_referenced_anon(page, mem_cont, |
643 | vm_flags); | |
5ad64688 | 644 | else if (page->mapping) |
6fe6b7e3 WF |
645 | referenced += page_referenced_file(page, mem_cont, |
646 | vm_flags); | |
5ad64688 | 647 | if (we_locked) |
1da177e4 | 648 | unlock_page(page); |
1da177e4 | 649 | } |
5ad64688 | 650 | out: |
5b7baf05 CB |
651 | if (page_test_and_clear_young(page)) |
652 | referenced++; | |
653 | ||
1da177e4 LT |
654 | return referenced; |
655 | } | |
656 | ||
1cb1729b HD |
657 | static int page_mkclean_one(struct page *page, struct vm_area_struct *vma, |
658 | unsigned long address) | |
d08b3851 PZ |
659 | { |
660 | struct mm_struct *mm = vma->vm_mm; | |
c2fda5fe | 661 | pte_t *pte; |
d08b3851 PZ |
662 | spinlock_t *ptl; |
663 | int ret = 0; | |
664 | ||
479db0bf | 665 | pte = page_check_address(page, mm, address, &ptl, 1); |
d08b3851 PZ |
666 | if (!pte) |
667 | goto out; | |
668 | ||
c2fda5fe PZ |
669 | if (pte_dirty(*pte) || pte_write(*pte)) { |
670 | pte_t entry; | |
d08b3851 | 671 | |
c2fda5fe | 672 | flush_cache_page(vma, address, pte_pfn(*pte)); |
cddb8a5c | 673 | entry = ptep_clear_flush_notify(vma, address, pte); |
c2fda5fe PZ |
674 | entry = pte_wrprotect(entry); |
675 | entry = pte_mkclean(entry); | |
d6e88e67 | 676 | set_pte_at(mm, address, pte, entry); |
c2fda5fe PZ |
677 | ret = 1; |
678 | } | |
d08b3851 | 679 | |
d08b3851 PZ |
680 | pte_unmap_unlock(pte, ptl); |
681 | out: | |
682 | return ret; | |
683 | } | |
684 | ||
685 | static int page_mkclean_file(struct address_space *mapping, struct page *page) | |
686 | { | |
687 | pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT); | |
688 | struct vm_area_struct *vma; | |
689 | struct prio_tree_iter iter; | |
690 | int ret = 0; | |
691 | ||
692 | BUG_ON(PageAnon(page)); | |
693 | ||
694 | spin_lock(&mapping->i_mmap_lock); | |
695 | vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, pgoff, pgoff) { | |
1cb1729b HD |
696 | if (vma->vm_flags & VM_SHARED) { |
697 | unsigned long address = vma_address(page, vma); | |
698 | if (address == -EFAULT) | |
699 | continue; | |
700 | ret += page_mkclean_one(page, vma, address); | |
701 | } | |
d08b3851 PZ |
702 | } |
703 | spin_unlock(&mapping->i_mmap_lock); | |
704 | return ret; | |
705 | } | |
706 | ||
707 | int page_mkclean(struct page *page) | |
708 | { | |
709 | int ret = 0; | |
710 | ||
711 | BUG_ON(!PageLocked(page)); | |
712 | ||
713 | if (page_mapped(page)) { | |
714 | struct address_space *mapping = page_mapping(page); | |
ce7e9fae | 715 | if (mapping) { |
d08b3851 | 716 | ret = page_mkclean_file(mapping, page); |
ce7e9fae CB |
717 | if (page_test_dirty(page)) { |
718 | page_clear_dirty(page); | |
719 | ret = 1; | |
720 | } | |
6c210482 | 721 | } |
d08b3851 PZ |
722 | } |
723 | ||
724 | return ret; | |
725 | } | |
60b59bea | 726 | EXPORT_SYMBOL_GPL(page_mkclean); |
d08b3851 | 727 | |
c44b6743 RR |
728 | /** |
729 | * page_move_anon_rmap - move a page to our anon_vma | |
730 | * @page: the page to move to our anon_vma | |
731 | * @vma: the vma the page belongs to | |
732 | * @address: the user virtual address mapped | |
733 | * | |
734 | * When a page belongs exclusively to one process after a COW event, | |
735 | * that page can be moved into the anon_vma that belongs to just that | |
736 | * process, so the rmap code will not search the parent or sibling | |
737 | * processes. | |
738 | */ | |
739 | void page_move_anon_rmap(struct page *page, | |
740 | struct vm_area_struct *vma, unsigned long address) | |
741 | { | |
742 | struct anon_vma *anon_vma = vma->anon_vma; | |
743 | ||
744 | VM_BUG_ON(!PageLocked(page)); | |
745 | VM_BUG_ON(!anon_vma); | |
746 | VM_BUG_ON(page->index != linear_page_index(vma, address)); | |
747 | ||
748 | anon_vma = (void *) anon_vma + PAGE_MAPPING_ANON; | |
749 | page->mapping = (struct address_space *) anon_vma; | |
750 | } | |
751 | ||
9617d95e | 752 | /** |
43d8eac4 | 753 | * __page_set_anon_rmap - setup new anonymous rmap |
9617d95e NP |
754 | * @page: the page to add the mapping to |
755 | * @vma: the vm area in which the mapping is added | |
756 | * @address: the user virtual address mapped | |
e8a03feb | 757 | * @exclusive: the page is exclusively owned by the current process |
9617d95e NP |
758 | */ |
759 | static void __page_set_anon_rmap(struct page *page, | |
e8a03feb | 760 | struct vm_area_struct *vma, unsigned long address, int exclusive) |
9617d95e | 761 | { |
e8a03feb | 762 | struct anon_vma *anon_vma = vma->anon_vma; |
ea90002b | 763 | |
e8a03feb | 764 | BUG_ON(!anon_vma); |
ea90002b LT |
765 | |
766 | /* | |
e8a03feb RR |
767 | * If the page isn't exclusively mapped into this vma, |
768 | * we must use the _oldest_ possible anon_vma for the | |
769 | * page mapping! | |
ea90002b | 770 | * |
e8a03feb RR |
771 | * So take the last AVC chain entry in the vma, which is |
772 | * the deepest ancestor, and use the anon_vma from that. | |
ea90002b | 773 | */ |
e8a03feb RR |
774 | if (!exclusive) { |
775 | struct anon_vma_chain *avc; | |
776 | avc = list_entry(vma->anon_vma_chain.prev, struct anon_vma_chain, same_vma); | |
777 | anon_vma = avc->anon_vma; | |
778 | } | |
9617d95e | 779 | |
9617d95e NP |
780 | anon_vma = (void *) anon_vma + PAGE_MAPPING_ANON; |
781 | page->mapping = (struct address_space *) anon_vma; | |
9617d95e | 782 | page->index = linear_page_index(vma, address); |
9617d95e NP |
783 | } |
784 | ||
c97a9e10 | 785 | /** |
43d8eac4 | 786 | * __page_check_anon_rmap - sanity check anonymous rmap addition |
c97a9e10 NP |
787 | * @page: the page to add the mapping to |
788 | * @vma: the vm area in which the mapping is added | |
789 | * @address: the user virtual address mapped | |
790 | */ | |
791 | static void __page_check_anon_rmap(struct page *page, | |
792 | struct vm_area_struct *vma, unsigned long address) | |
793 | { | |
794 | #ifdef CONFIG_DEBUG_VM | |
795 | /* | |
796 | * The page's anon-rmap details (mapping and index) are guaranteed to | |
797 | * be set up correctly at this point. | |
798 | * | |
799 | * We have exclusion against page_add_anon_rmap because the caller | |
800 | * always holds the page locked, except if called from page_dup_rmap, | |
801 | * in which case the page is already known to be setup. | |
802 | * | |
803 | * We have exclusion against page_add_new_anon_rmap because those pages | |
804 | * are initially only visible via the pagetables, and the pte is locked | |
805 | * over the call to page_add_new_anon_rmap. | |
806 | */ | |
c97a9e10 NP |
807 | BUG_ON(page->index != linear_page_index(vma, address)); |
808 | #endif | |
809 | } | |
810 | ||
1da177e4 LT |
811 | /** |
812 | * page_add_anon_rmap - add pte mapping to an anonymous page | |
813 | * @page: the page to add the mapping to | |
814 | * @vma: the vm area in which the mapping is added | |
815 | * @address: the user virtual address mapped | |
816 | * | |
5ad64688 | 817 | * The caller needs to hold the pte lock, and the page must be locked in |
80e14822 HD |
818 | * the anon_vma case: to serialize mapping,index checking after setting, |
819 | * and to ensure that PageAnon is not being upgraded racily to PageKsm | |
820 | * (but PageKsm is never downgraded to PageAnon). | |
1da177e4 LT |
821 | */ |
822 | void page_add_anon_rmap(struct page *page, | |
823 | struct vm_area_struct *vma, unsigned long address) | |
824 | { | |
5ad64688 HD |
825 | int first = atomic_inc_and_test(&page->_mapcount); |
826 | if (first) | |
827 | __inc_zone_page_state(page, NR_ANON_PAGES); | |
828 | if (unlikely(PageKsm(page))) | |
829 | return; | |
830 | ||
c97a9e10 NP |
831 | VM_BUG_ON(!PageLocked(page)); |
832 | VM_BUG_ON(address < vma->vm_start || address >= vma->vm_end); | |
5ad64688 | 833 | if (first) |
e8a03feb | 834 | __page_set_anon_rmap(page, vma, address, 0); |
69029cd5 | 835 | else |
c97a9e10 | 836 | __page_check_anon_rmap(page, vma, address); |
1da177e4 LT |
837 | } |
838 | ||
43d8eac4 | 839 | /** |
9617d95e NP |
840 | * page_add_new_anon_rmap - add pte mapping to a new anonymous page |
841 | * @page: the page to add the mapping to | |
842 | * @vma: the vm area in which the mapping is added | |
843 | * @address: the user virtual address mapped | |
844 | * | |
845 | * Same as page_add_anon_rmap but must only be called on *new* pages. | |
846 | * This means the inc-and-test can be bypassed. | |
c97a9e10 | 847 | * Page does not have to be locked. |
9617d95e NP |
848 | */ |
849 | void page_add_new_anon_rmap(struct page *page, | |
850 | struct vm_area_struct *vma, unsigned long address) | |
851 | { | |
b5934c53 | 852 | VM_BUG_ON(address < vma->vm_start || address >= vma->vm_end); |
cbf84b7a HD |
853 | SetPageSwapBacked(page); |
854 | atomic_set(&page->_mapcount, 0); /* increment count (starts at -1) */ | |
5ad64688 | 855 | __inc_zone_page_state(page, NR_ANON_PAGES); |
e8a03feb | 856 | __page_set_anon_rmap(page, vma, address, 1); |
b5934c53 | 857 | if (page_evictable(page, vma)) |
cbf84b7a | 858 | lru_cache_add_lru(page, LRU_ACTIVE_ANON); |
b5934c53 HD |
859 | else |
860 | add_page_to_unevictable_list(page); | |
9617d95e NP |
861 | } |
862 | ||
1da177e4 LT |
863 | /** |
864 | * page_add_file_rmap - add pte mapping to a file page | |
865 | * @page: the page to add the mapping to | |
866 | * | |
b8072f09 | 867 | * The caller needs to hold the pte lock. |
1da177e4 LT |
868 | */ |
869 | void page_add_file_rmap(struct page *page) | |
870 | { | |
d69b042f | 871 | if (atomic_inc_and_test(&page->_mapcount)) { |
65ba55f5 | 872 | __inc_zone_page_state(page, NR_FILE_MAPPED); |
d8046582 | 873 | mem_cgroup_update_file_mapped(page, 1); |
d69b042f | 874 | } |
1da177e4 LT |
875 | } |
876 | ||
877 | /** | |
878 | * page_remove_rmap - take down pte mapping from a page | |
879 | * @page: page to remove mapping from | |
880 | * | |
b8072f09 | 881 | * The caller needs to hold the pte lock. |
1da177e4 | 882 | */ |
edc315fd | 883 | void page_remove_rmap(struct page *page) |
1da177e4 | 884 | { |
b904dcfe KM |
885 | /* page still mapped by someone else? */ |
886 | if (!atomic_add_negative(-1, &page->_mapcount)) | |
887 | return; | |
888 | ||
889 | /* | |
890 | * Now that the last pte has gone, s390 must transfer dirty | |
891 | * flag from storage key to struct page. We can usually skip | |
892 | * this if the page is anon, so about to be freed; but perhaps | |
893 | * not if it's in swapcache - there might be another pte slot | |
894 | * containing the swap entry, but page not yet written to swap. | |
895 | */ | |
896 | if ((!PageAnon(page) || PageSwapCache(page)) && page_test_dirty(page)) { | |
897 | page_clear_dirty(page); | |
898 | set_page_dirty(page); | |
1da177e4 | 899 | } |
b904dcfe KM |
900 | if (PageAnon(page)) { |
901 | mem_cgroup_uncharge_page(page); | |
902 | __dec_zone_page_state(page, NR_ANON_PAGES); | |
903 | } else { | |
904 | __dec_zone_page_state(page, NR_FILE_MAPPED); | |
d8046582 | 905 | mem_cgroup_update_file_mapped(page, -1); |
b904dcfe | 906 | } |
b904dcfe KM |
907 | /* |
908 | * It would be tidy to reset the PageAnon mapping here, | |
909 | * but that might overwrite a racing page_add_anon_rmap | |
910 | * which increments mapcount after us but sets mapping | |
911 | * before us: so leave the reset to free_hot_cold_page, | |
912 | * and remember that it's only reliable while mapped. | |
913 | * Leaving it set also helps swapoff to reinstate ptes | |
914 | * faster for those pages still in swapcache. | |
915 | */ | |
1da177e4 LT |
916 | } |
917 | ||
918 | /* | |
919 | * Subfunctions of try_to_unmap: try_to_unmap_one called | |
920 | * repeatedly from either try_to_unmap_anon or try_to_unmap_file. | |
921 | */ | |
5ad64688 HD |
922 | int try_to_unmap_one(struct page *page, struct vm_area_struct *vma, |
923 | unsigned long address, enum ttu_flags flags) | |
1da177e4 LT |
924 | { |
925 | struct mm_struct *mm = vma->vm_mm; | |
1da177e4 LT |
926 | pte_t *pte; |
927 | pte_t pteval; | |
c0718806 | 928 | spinlock_t *ptl; |
1da177e4 LT |
929 | int ret = SWAP_AGAIN; |
930 | ||
479db0bf | 931 | pte = page_check_address(page, mm, address, &ptl, 0); |
c0718806 | 932 | if (!pte) |
81b4082d | 933 | goto out; |
1da177e4 LT |
934 | |
935 | /* | |
936 | * If the page is mlock()d, we cannot swap it out. | |
937 | * If it's recently referenced (perhaps page_referenced | |
938 | * skipped over this mm) then we should reactivate it. | |
939 | */ | |
14fa31b8 | 940 | if (!(flags & TTU_IGNORE_MLOCK)) { |
caed0f48 KM |
941 | if (vma->vm_flags & VM_LOCKED) |
942 | goto out_mlock; | |
943 | ||
af8e3354 | 944 | if (TTU_ACTION(flags) == TTU_MUNLOCK) |
53f79acb | 945 | goto out_unmap; |
14fa31b8 AK |
946 | } |
947 | if (!(flags & TTU_IGNORE_ACCESS)) { | |
b291f000 NP |
948 | if (ptep_clear_flush_young_notify(vma, address, pte)) { |
949 | ret = SWAP_FAIL; | |
950 | goto out_unmap; | |
951 | } | |
952 | } | |
1da177e4 | 953 | |
1da177e4 LT |
954 | /* Nuke the page table entry. */ |
955 | flush_cache_page(vma, address, page_to_pfn(page)); | |
cddb8a5c | 956 | pteval = ptep_clear_flush_notify(vma, address, pte); |
1da177e4 LT |
957 | |
958 | /* Move the dirty bit to the physical page now the pte is gone. */ | |
959 | if (pte_dirty(pteval)) | |
960 | set_page_dirty(page); | |
961 | ||
365e9c87 HD |
962 | /* Update high watermark before we lower rss */ |
963 | update_hiwater_rss(mm); | |
964 | ||
888b9f7c AK |
965 | if (PageHWPoison(page) && !(flags & TTU_IGNORE_HWPOISON)) { |
966 | if (PageAnon(page)) | |
d559db08 | 967 | dec_mm_counter(mm, MM_ANONPAGES); |
888b9f7c | 968 | else |
d559db08 | 969 | dec_mm_counter(mm, MM_FILEPAGES); |
888b9f7c AK |
970 | set_pte_at(mm, address, pte, |
971 | swp_entry_to_pte(make_hwpoison_entry(page))); | |
972 | } else if (PageAnon(page)) { | |
4c21e2f2 | 973 | swp_entry_t entry = { .val = page_private(page) }; |
0697212a CL |
974 | |
975 | if (PageSwapCache(page)) { | |
976 | /* | |
977 | * Store the swap location in the pte. | |
978 | * See handle_pte_fault() ... | |
979 | */ | |
570a335b HD |
980 | if (swap_duplicate(entry) < 0) { |
981 | set_pte_at(mm, address, pte, pteval); | |
982 | ret = SWAP_FAIL; | |
983 | goto out_unmap; | |
984 | } | |
0697212a CL |
985 | if (list_empty(&mm->mmlist)) { |
986 | spin_lock(&mmlist_lock); | |
987 | if (list_empty(&mm->mmlist)) | |
988 | list_add(&mm->mmlist, &init_mm.mmlist); | |
989 | spin_unlock(&mmlist_lock); | |
990 | } | |
d559db08 | 991 | dec_mm_counter(mm, MM_ANONPAGES); |
b084d435 | 992 | inc_mm_counter(mm, MM_SWAPENTS); |
64cdd548 | 993 | } else if (PAGE_MIGRATION) { |
0697212a CL |
994 | /* |
995 | * Store the pfn of the page in a special migration | |
996 | * pte. do_swap_page() will wait until the migration | |
997 | * pte is removed and then restart fault handling. | |
998 | */ | |
14fa31b8 | 999 | BUG_ON(TTU_ACTION(flags) != TTU_MIGRATION); |
0697212a | 1000 | entry = make_migration_entry(page, pte_write(pteval)); |
1da177e4 LT |
1001 | } |
1002 | set_pte_at(mm, address, pte, swp_entry_to_pte(entry)); | |
1003 | BUG_ON(pte_file(*pte)); | |
14fa31b8 | 1004 | } else if (PAGE_MIGRATION && (TTU_ACTION(flags) == TTU_MIGRATION)) { |
04e62a29 CL |
1005 | /* Establish migration entry for a file page */ |
1006 | swp_entry_t entry; | |
1007 | entry = make_migration_entry(page, pte_write(pteval)); | |
1008 | set_pte_at(mm, address, pte, swp_entry_to_pte(entry)); | |
1009 | } else | |
d559db08 | 1010 | dec_mm_counter(mm, MM_FILEPAGES); |
1da177e4 | 1011 | |
edc315fd | 1012 | page_remove_rmap(page); |
1da177e4 LT |
1013 | page_cache_release(page); |
1014 | ||
1015 | out_unmap: | |
c0718806 | 1016 | pte_unmap_unlock(pte, ptl); |
caed0f48 KM |
1017 | out: |
1018 | return ret; | |
53f79acb | 1019 | |
caed0f48 KM |
1020 | out_mlock: |
1021 | pte_unmap_unlock(pte, ptl); | |
1022 | ||
1023 | ||
1024 | /* | |
1025 | * We need mmap_sem locking, Otherwise VM_LOCKED check makes | |
1026 | * unstable result and race. Plus, We can't wait here because | |
1027 | * we now hold anon_vma->lock or mapping->i_mmap_lock. | |
1028 | * if trylock failed, the page remain in evictable lru and later | |
1029 | * vmscan could retry to move the page to unevictable lru if the | |
1030 | * page is actually mlocked. | |
1031 | */ | |
1032 | if (down_read_trylock(&vma->vm_mm->mmap_sem)) { | |
1033 | if (vma->vm_flags & VM_LOCKED) { | |
1034 | mlock_vma_page(page); | |
1035 | ret = SWAP_MLOCK; | |
53f79acb | 1036 | } |
caed0f48 | 1037 | up_read(&vma->vm_mm->mmap_sem); |
53f79acb | 1038 | } |
1da177e4 LT |
1039 | return ret; |
1040 | } | |
1041 | ||
1042 | /* | |
1043 | * objrmap doesn't work for nonlinear VMAs because the assumption that | |
1044 | * offset-into-file correlates with offset-into-virtual-addresses does not hold. | |
1045 | * Consequently, given a particular page and its ->index, we cannot locate the | |
1046 | * ptes which are mapping that page without an exhaustive linear search. | |
1047 | * | |
1048 | * So what this code does is a mini "virtual scan" of each nonlinear VMA which | |
1049 | * maps the file to which the target page belongs. The ->vm_private_data field | |
1050 | * holds the current cursor into that scan. Successive searches will circulate | |
1051 | * around the vma's virtual address space. | |
1052 | * | |
1053 | * So as more replacement pressure is applied to the pages in a nonlinear VMA, | |
1054 | * more scanning pressure is placed against them as well. Eventually pages | |
1055 | * will become fully unmapped and are eligible for eviction. | |
1056 | * | |
1057 | * For very sparsely populated VMAs this is a little inefficient - chances are | |
1058 | * there there won't be many ptes located within the scan cluster. In this case | |
1059 | * maybe we could scan further - to the end of the pte page, perhaps. | |
b291f000 NP |
1060 | * |
1061 | * Mlocked pages: check VM_LOCKED under mmap_sem held for read, if we can | |
1062 | * acquire it without blocking. If vma locked, mlock the pages in the cluster, | |
1063 | * rather than unmapping them. If we encounter the "check_page" that vmscan is | |
1064 | * trying to unmap, return SWAP_MLOCK, else default SWAP_AGAIN. | |
1da177e4 LT |
1065 | */ |
1066 | #define CLUSTER_SIZE min(32*PAGE_SIZE, PMD_SIZE) | |
1067 | #define CLUSTER_MASK (~(CLUSTER_SIZE - 1)) | |
1068 | ||
b291f000 NP |
1069 | static int try_to_unmap_cluster(unsigned long cursor, unsigned int *mapcount, |
1070 | struct vm_area_struct *vma, struct page *check_page) | |
1da177e4 LT |
1071 | { |
1072 | struct mm_struct *mm = vma->vm_mm; | |
1073 | pgd_t *pgd; | |
1074 | pud_t *pud; | |
1075 | pmd_t *pmd; | |
c0718806 | 1076 | pte_t *pte; |
1da177e4 | 1077 | pte_t pteval; |
c0718806 | 1078 | spinlock_t *ptl; |
1da177e4 LT |
1079 | struct page *page; |
1080 | unsigned long address; | |
1081 | unsigned long end; | |
b291f000 NP |
1082 | int ret = SWAP_AGAIN; |
1083 | int locked_vma = 0; | |
1da177e4 | 1084 | |
1da177e4 LT |
1085 | address = (vma->vm_start + cursor) & CLUSTER_MASK; |
1086 | end = address + CLUSTER_SIZE; | |
1087 | if (address < vma->vm_start) | |
1088 | address = vma->vm_start; | |
1089 | if (end > vma->vm_end) | |
1090 | end = vma->vm_end; | |
1091 | ||
1092 | pgd = pgd_offset(mm, address); | |
1093 | if (!pgd_present(*pgd)) | |
b291f000 | 1094 | return ret; |
1da177e4 LT |
1095 | |
1096 | pud = pud_offset(pgd, address); | |
1097 | if (!pud_present(*pud)) | |
b291f000 | 1098 | return ret; |
1da177e4 LT |
1099 | |
1100 | pmd = pmd_offset(pud, address); | |
1101 | if (!pmd_present(*pmd)) | |
b291f000 NP |
1102 | return ret; |
1103 | ||
1104 | /* | |
af8e3354 | 1105 | * If we can acquire the mmap_sem for read, and vma is VM_LOCKED, |
b291f000 NP |
1106 | * keep the sem while scanning the cluster for mlocking pages. |
1107 | */ | |
af8e3354 | 1108 | if (down_read_trylock(&vma->vm_mm->mmap_sem)) { |
b291f000 NP |
1109 | locked_vma = (vma->vm_flags & VM_LOCKED); |
1110 | if (!locked_vma) | |
1111 | up_read(&vma->vm_mm->mmap_sem); /* don't need it */ | |
1112 | } | |
c0718806 HD |
1113 | |
1114 | pte = pte_offset_map_lock(mm, pmd, address, &ptl); | |
1da177e4 | 1115 | |
365e9c87 HD |
1116 | /* Update high watermark before we lower rss */ |
1117 | update_hiwater_rss(mm); | |
1118 | ||
c0718806 | 1119 | for (; address < end; pte++, address += PAGE_SIZE) { |
1da177e4 LT |
1120 | if (!pte_present(*pte)) |
1121 | continue; | |
6aab341e LT |
1122 | page = vm_normal_page(vma, address, *pte); |
1123 | BUG_ON(!page || PageAnon(page)); | |
1da177e4 | 1124 | |
b291f000 NP |
1125 | if (locked_vma) { |
1126 | mlock_vma_page(page); /* no-op if already mlocked */ | |
1127 | if (page == check_page) | |
1128 | ret = SWAP_MLOCK; | |
1129 | continue; /* don't unmap */ | |
1130 | } | |
1131 | ||
cddb8a5c | 1132 | if (ptep_clear_flush_young_notify(vma, address, pte)) |
1da177e4 LT |
1133 | continue; |
1134 | ||
1135 | /* Nuke the page table entry. */ | |
eca35133 | 1136 | flush_cache_page(vma, address, pte_pfn(*pte)); |
cddb8a5c | 1137 | pteval = ptep_clear_flush_notify(vma, address, pte); |
1da177e4 LT |
1138 | |
1139 | /* If nonlinear, store the file page offset in the pte. */ | |
1140 | if (page->index != linear_page_index(vma, address)) | |
1141 | set_pte_at(mm, address, pte, pgoff_to_pte(page->index)); | |
1142 | ||
1143 | /* Move the dirty bit to the physical page now the pte is gone. */ | |
1144 | if (pte_dirty(pteval)) | |
1145 | set_page_dirty(page); | |
1146 | ||
edc315fd | 1147 | page_remove_rmap(page); |
1da177e4 | 1148 | page_cache_release(page); |
d559db08 | 1149 | dec_mm_counter(mm, MM_FILEPAGES); |
1da177e4 LT |
1150 | (*mapcount)--; |
1151 | } | |
c0718806 | 1152 | pte_unmap_unlock(pte - 1, ptl); |
b291f000 NP |
1153 | if (locked_vma) |
1154 | up_read(&vma->vm_mm->mmap_sem); | |
1155 | return ret; | |
1da177e4 LT |
1156 | } |
1157 | ||
a8bef8ff MG |
1158 | static bool is_vma_temporary_stack(struct vm_area_struct *vma) |
1159 | { | |
1160 | int maybe_stack = vma->vm_flags & (VM_GROWSDOWN | VM_GROWSUP); | |
1161 | ||
1162 | if (!maybe_stack) | |
1163 | return false; | |
1164 | ||
1165 | if ((vma->vm_flags & VM_STACK_INCOMPLETE_SETUP) == | |
1166 | VM_STACK_INCOMPLETE_SETUP) | |
1167 | return true; | |
1168 | ||
1169 | return false; | |
1170 | } | |
1171 | ||
b291f000 NP |
1172 | /** |
1173 | * try_to_unmap_anon - unmap or unlock anonymous page using the object-based | |
1174 | * rmap method | |
1175 | * @page: the page to unmap/unlock | |
8051be5e | 1176 | * @flags: action and flags |
b291f000 NP |
1177 | * |
1178 | * Find all the mappings of a page using the mapping pointer and the vma chains | |
1179 | * contained in the anon_vma struct it points to. | |
1180 | * | |
1181 | * This function is only called from try_to_unmap/try_to_munlock for | |
1182 | * anonymous pages. | |
1183 | * When called from try_to_munlock(), the mmap_sem of the mm containing the vma | |
1184 | * where the page was found will be held for write. So, we won't recheck | |
1185 | * vm_flags for that VMA. That should be OK, because that vma shouldn't be | |
1186 | * 'LOCKED. | |
1187 | */ | |
14fa31b8 | 1188 | static int try_to_unmap_anon(struct page *page, enum ttu_flags flags) |
1da177e4 LT |
1189 | { |
1190 | struct anon_vma *anon_vma; | |
5beb4930 | 1191 | struct anon_vma_chain *avc; |
1da177e4 | 1192 | int ret = SWAP_AGAIN; |
b291f000 | 1193 | |
1da177e4 LT |
1194 | anon_vma = page_lock_anon_vma(page); |
1195 | if (!anon_vma) | |
1196 | return ret; | |
1197 | ||
5beb4930 RR |
1198 | list_for_each_entry(avc, &anon_vma->head, same_anon_vma) { |
1199 | struct vm_area_struct *vma = avc->vma; | |
a8bef8ff MG |
1200 | unsigned long address; |
1201 | ||
1202 | /* | |
1203 | * During exec, a temporary VMA is setup and later moved. | |
1204 | * The VMA is moved under the anon_vma lock but not the | |
1205 | * page tables leading to a race where migration cannot | |
1206 | * find the migration ptes. Rather than increasing the | |
1207 | * locking requirements of exec(), migration skips | |
1208 | * temporary VMAs until after exec() completes. | |
1209 | */ | |
1210 | if (PAGE_MIGRATION && (flags & TTU_MIGRATION) && | |
1211 | is_vma_temporary_stack(vma)) | |
1212 | continue; | |
1213 | ||
1214 | address = vma_address(page, vma); | |
1cb1729b HD |
1215 | if (address == -EFAULT) |
1216 | continue; | |
1217 | ret = try_to_unmap_one(page, vma, address, flags); | |
53f79acb HD |
1218 | if (ret != SWAP_AGAIN || !page_mapped(page)) |
1219 | break; | |
1da177e4 | 1220 | } |
34bbd704 ON |
1221 | |
1222 | page_unlock_anon_vma(anon_vma); | |
1da177e4 LT |
1223 | return ret; |
1224 | } | |
1225 | ||
1226 | /** | |
b291f000 NP |
1227 | * try_to_unmap_file - unmap/unlock file page using the object-based rmap method |
1228 | * @page: the page to unmap/unlock | |
14fa31b8 | 1229 | * @flags: action and flags |
1da177e4 LT |
1230 | * |
1231 | * Find all the mappings of a page using the mapping pointer and the vma chains | |
1232 | * contained in the address_space struct it points to. | |
1233 | * | |
b291f000 NP |
1234 | * This function is only called from try_to_unmap/try_to_munlock for |
1235 | * object-based pages. | |
1236 | * When called from try_to_munlock(), the mmap_sem of the mm containing the vma | |
1237 | * where the page was found will be held for write. So, we won't recheck | |
1238 | * vm_flags for that VMA. That should be OK, because that vma shouldn't be | |
1239 | * 'LOCKED. | |
1da177e4 | 1240 | */ |
14fa31b8 | 1241 | static int try_to_unmap_file(struct page *page, enum ttu_flags flags) |
1da177e4 LT |
1242 | { |
1243 | struct address_space *mapping = page->mapping; | |
1244 | pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT); | |
1245 | struct vm_area_struct *vma; | |
1246 | struct prio_tree_iter iter; | |
1247 | int ret = SWAP_AGAIN; | |
1248 | unsigned long cursor; | |
1249 | unsigned long max_nl_cursor = 0; | |
1250 | unsigned long max_nl_size = 0; | |
1251 | unsigned int mapcount; | |
1252 | ||
1253 | spin_lock(&mapping->i_mmap_lock); | |
1254 | vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, pgoff, pgoff) { | |
1cb1729b HD |
1255 | unsigned long address = vma_address(page, vma); |
1256 | if (address == -EFAULT) | |
1257 | continue; | |
1258 | ret = try_to_unmap_one(page, vma, address, flags); | |
53f79acb HD |
1259 | if (ret != SWAP_AGAIN || !page_mapped(page)) |
1260 | goto out; | |
1da177e4 LT |
1261 | } |
1262 | ||
1263 | if (list_empty(&mapping->i_mmap_nonlinear)) | |
1264 | goto out; | |
1265 | ||
53f79acb HD |
1266 | /* |
1267 | * We don't bother to try to find the munlocked page in nonlinears. | |
1268 | * It's costly. Instead, later, page reclaim logic may call | |
1269 | * try_to_unmap(TTU_MUNLOCK) and recover PG_mlocked lazily. | |
1270 | */ | |
1271 | if (TTU_ACTION(flags) == TTU_MUNLOCK) | |
1272 | goto out; | |
1273 | ||
1da177e4 LT |
1274 | list_for_each_entry(vma, &mapping->i_mmap_nonlinear, |
1275 | shared.vm_set.list) { | |
1da177e4 LT |
1276 | cursor = (unsigned long) vma->vm_private_data; |
1277 | if (cursor > max_nl_cursor) | |
1278 | max_nl_cursor = cursor; | |
1279 | cursor = vma->vm_end - vma->vm_start; | |
1280 | if (cursor > max_nl_size) | |
1281 | max_nl_size = cursor; | |
1282 | } | |
1283 | ||
b291f000 | 1284 | if (max_nl_size == 0) { /* all nonlinears locked or reserved ? */ |
1da177e4 LT |
1285 | ret = SWAP_FAIL; |
1286 | goto out; | |
1287 | } | |
1288 | ||
1289 | /* | |
1290 | * We don't try to search for this page in the nonlinear vmas, | |
1291 | * and page_referenced wouldn't have found it anyway. Instead | |
1292 | * just walk the nonlinear vmas trying to age and unmap some. | |
1293 | * The mapcount of the page we came in with is irrelevant, | |
1294 | * but even so use it as a guide to how hard we should try? | |
1295 | */ | |
1296 | mapcount = page_mapcount(page); | |
1297 | if (!mapcount) | |
1298 | goto out; | |
1299 | cond_resched_lock(&mapping->i_mmap_lock); | |
1300 | ||
1301 | max_nl_size = (max_nl_size + CLUSTER_SIZE - 1) & CLUSTER_MASK; | |
1302 | if (max_nl_cursor == 0) | |
1303 | max_nl_cursor = CLUSTER_SIZE; | |
1304 | ||
1305 | do { | |
1306 | list_for_each_entry(vma, &mapping->i_mmap_nonlinear, | |
1307 | shared.vm_set.list) { | |
1da177e4 | 1308 | cursor = (unsigned long) vma->vm_private_data; |
839b9685 | 1309 | while ( cursor < max_nl_cursor && |
1da177e4 | 1310 | cursor < vma->vm_end - vma->vm_start) { |
53f79acb HD |
1311 | if (try_to_unmap_cluster(cursor, &mapcount, |
1312 | vma, page) == SWAP_MLOCK) | |
1313 | ret = SWAP_MLOCK; | |
1da177e4 LT |
1314 | cursor += CLUSTER_SIZE; |
1315 | vma->vm_private_data = (void *) cursor; | |
1316 | if ((int)mapcount <= 0) | |
1317 | goto out; | |
1318 | } | |
1319 | vma->vm_private_data = (void *) max_nl_cursor; | |
1320 | } | |
1321 | cond_resched_lock(&mapping->i_mmap_lock); | |
1322 | max_nl_cursor += CLUSTER_SIZE; | |
1323 | } while (max_nl_cursor <= max_nl_size); | |
1324 | ||
1325 | /* | |
1326 | * Don't loop forever (perhaps all the remaining pages are | |
1327 | * in locked vmas). Reset cursor on all unreserved nonlinear | |
1328 | * vmas, now forgetting on which ones it had fallen behind. | |
1329 | */ | |
101d2be7 HD |
1330 | list_for_each_entry(vma, &mapping->i_mmap_nonlinear, shared.vm_set.list) |
1331 | vma->vm_private_data = NULL; | |
1da177e4 LT |
1332 | out: |
1333 | spin_unlock(&mapping->i_mmap_lock); | |
1334 | return ret; | |
1335 | } | |
1336 | ||
1337 | /** | |
1338 | * try_to_unmap - try to remove all page table mappings to a page | |
1339 | * @page: the page to get unmapped | |
14fa31b8 | 1340 | * @flags: action and flags |
1da177e4 LT |
1341 | * |
1342 | * Tries to remove all the page table entries which are mapping this | |
1343 | * page, used in the pageout path. Caller must hold the page lock. | |
1344 | * Return values are: | |
1345 | * | |
1346 | * SWAP_SUCCESS - we succeeded in removing all mappings | |
1347 | * SWAP_AGAIN - we missed a mapping, try again later | |
1348 | * SWAP_FAIL - the page is unswappable | |
b291f000 | 1349 | * SWAP_MLOCK - page is mlocked. |
1da177e4 | 1350 | */ |
14fa31b8 | 1351 | int try_to_unmap(struct page *page, enum ttu_flags flags) |
1da177e4 LT |
1352 | { |
1353 | int ret; | |
1354 | ||
1da177e4 LT |
1355 | BUG_ON(!PageLocked(page)); |
1356 | ||
5ad64688 HD |
1357 | if (unlikely(PageKsm(page))) |
1358 | ret = try_to_unmap_ksm(page, flags); | |
1359 | else if (PageAnon(page)) | |
14fa31b8 | 1360 | ret = try_to_unmap_anon(page, flags); |
1da177e4 | 1361 | else |
14fa31b8 | 1362 | ret = try_to_unmap_file(page, flags); |
b291f000 | 1363 | if (ret != SWAP_MLOCK && !page_mapped(page)) |
1da177e4 LT |
1364 | ret = SWAP_SUCCESS; |
1365 | return ret; | |
1366 | } | |
81b4082d | 1367 | |
b291f000 NP |
1368 | /** |
1369 | * try_to_munlock - try to munlock a page | |
1370 | * @page: the page to be munlocked | |
1371 | * | |
1372 | * Called from munlock code. Checks all of the VMAs mapping the page | |
1373 | * to make sure nobody else has this page mlocked. The page will be | |
1374 | * returned with PG_mlocked cleared if no other vmas have it mlocked. | |
1375 | * | |
1376 | * Return values are: | |
1377 | * | |
53f79acb | 1378 | * SWAP_AGAIN - no vma is holding page mlocked, or, |
b291f000 | 1379 | * SWAP_AGAIN - page mapped in mlocked vma -- couldn't acquire mmap sem |
5ad64688 | 1380 | * SWAP_FAIL - page cannot be located at present |
b291f000 NP |
1381 | * SWAP_MLOCK - page is now mlocked. |
1382 | */ | |
1383 | int try_to_munlock(struct page *page) | |
1384 | { | |
1385 | VM_BUG_ON(!PageLocked(page) || PageLRU(page)); | |
1386 | ||
5ad64688 HD |
1387 | if (unlikely(PageKsm(page))) |
1388 | return try_to_unmap_ksm(page, TTU_MUNLOCK); | |
1389 | else if (PageAnon(page)) | |
14fa31b8 | 1390 | return try_to_unmap_anon(page, TTU_MUNLOCK); |
b291f000 | 1391 | else |
14fa31b8 | 1392 | return try_to_unmap_file(page, TTU_MUNLOCK); |
b291f000 | 1393 | } |
e9995ef9 | 1394 | |
76545066 RR |
1395 | #if defined(CONFIG_KSM) || defined(CONFIG_MIGRATION) |
1396 | /* | |
1397 | * Drop an anon_vma refcount, freeing the anon_vma and anon_vma->root | |
1398 | * if necessary. Be careful to do all the tests under the lock. Once | |
1399 | * we know we are the last user, nobody else can get a reference and we | |
1400 | * can do the freeing without the lock. | |
1401 | */ | |
1402 | void drop_anon_vma(struct anon_vma *anon_vma) | |
1403 | { | |
1404 | if (atomic_dec_and_lock(&anon_vma->external_refcount, &anon_vma->root->lock)) { | |
1405 | struct anon_vma *root = anon_vma->root; | |
1406 | int empty = list_empty(&anon_vma->head); | |
1407 | int last_root_user = 0; | |
1408 | int root_empty = 0; | |
1409 | ||
1410 | /* | |
1411 | * The refcount on a non-root anon_vma got dropped. Drop | |
1412 | * the refcount on the root and check if we need to free it. | |
1413 | */ | |
1414 | if (empty && anon_vma != root) { | |
1415 | last_root_user = atomic_dec_and_test(&root->external_refcount); | |
1416 | root_empty = list_empty(&root->head); | |
1417 | } | |
1418 | anon_vma_unlock(anon_vma); | |
1419 | ||
1420 | if (empty) { | |
1421 | anon_vma_free(anon_vma); | |
1422 | if (root_empty && last_root_user) | |
1423 | anon_vma_free(root); | |
1424 | } | |
1425 | } | |
1426 | } | |
1427 | #endif | |
1428 | ||
e9995ef9 HD |
1429 | #ifdef CONFIG_MIGRATION |
1430 | /* | |
1431 | * rmap_walk() and its helpers rmap_walk_anon() and rmap_walk_file(): | |
1432 | * Called by migrate.c to remove migration ptes, but might be used more later. | |
1433 | */ | |
1434 | static int rmap_walk_anon(struct page *page, int (*rmap_one)(struct page *, | |
1435 | struct vm_area_struct *, unsigned long, void *), void *arg) | |
1436 | { | |
1437 | struct anon_vma *anon_vma; | |
5beb4930 | 1438 | struct anon_vma_chain *avc; |
e9995ef9 HD |
1439 | int ret = SWAP_AGAIN; |
1440 | ||
1441 | /* | |
1442 | * Note: remove_migration_ptes() cannot use page_lock_anon_vma() | |
1443 | * because that depends on page_mapped(); but not all its usages | |
3f6c8272 MG |
1444 | * are holding mmap_sem. Users without mmap_sem are required to |
1445 | * take a reference count to prevent the anon_vma disappearing | |
e9995ef9 HD |
1446 | */ |
1447 | anon_vma = page_anon_vma(page); | |
1448 | if (!anon_vma) | |
1449 | return ret; | |
cba48b98 | 1450 | anon_vma_lock(anon_vma); |
5beb4930 RR |
1451 | list_for_each_entry(avc, &anon_vma->head, same_anon_vma) { |
1452 | struct vm_area_struct *vma = avc->vma; | |
e9995ef9 HD |
1453 | unsigned long address = vma_address(page, vma); |
1454 | if (address == -EFAULT) | |
1455 | continue; | |
1456 | ret = rmap_one(page, vma, address, arg); | |
1457 | if (ret != SWAP_AGAIN) | |
1458 | break; | |
1459 | } | |
cba48b98 | 1460 | anon_vma_unlock(anon_vma); |
e9995ef9 HD |
1461 | return ret; |
1462 | } | |
1463 | ||
1464 | static int rmap_walk_file(struct page *page, int (*rmap_one)(struct page *, | |
1465 | struct vm_area_struct *, unsigned long, void *), void *arg) | |
1466 | { | |
1467 | struct address_space *mapping = page->mapping; | |
1468 | pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT); | |
1469 | struct vm_area_struct *vma; | |
1470 | struct prio_tree_iter iter; | |
1471 | int ret = SWAP_AGAIN; | |
1472 | ||
1473 | if (!mapping) | |
1474 | return ret; | |
1475 | spin_lock(&mapping->i_mmap_lock); | |
1476 | vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, pgoff, pgoff) { | |
1477 | unsigned long address = vma_address(page, vma); | |
1478 | if (address == -EFAULT) | |
1479 | continue; | |
1480 | ret = rmap_one(page, vma, address, arg); | |
1481 | if (ret != SWAP_AGAIN) | |
1482 | break; | |
1483 | } | |
1484 | /* | |
1485 | * No nonlinear handling: being always shared, nonlinear vmas | |
1486 | * never contain migration ptes. Decide what to do about this | |
1487 | * limitation to linear when we need rmap_walk() on nonlinear. | |
1488 | */ | |
1489 | spin_unlock(&mapping->i_mmap_lock); | |
1490 | return ret; | |
1491 | } | |
1492 | ||
1493 | int rmap_walk(struct page *page, int (*rmap_one)(struct page *, | |
1494 | struct vm_area_struct *, unsigned long, void *), void *arg) | |
1495 | { | |
1496 | VM_BUG_ON(!PageLocked(page)); | |
1497 | ||
1498 | if (unlikely(PageKsm(page))) | |
1499 | return rmap_walk_ksm(page, rmap_one, arg); | |
1500 | else if (PageAnon(page)) | |
1501 | return rmap_walk_anon(page, rmap_one, arg); | |
1502 | else | |
1503 | return rmap_walk_file(page, rmap_one, arg); | |
1504 | } | |
1505 | #endif /* CONFIG_MIGRATION */ |