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> |
0fe6e20b | 59 | #include <linux/hugetlb.h> |
1da177e4 LT |
60 | |
61 | #include <asm/tlbflush.h> | |
62 | ||
b291f000 NP |
63 | #include "internal.h" |
64 | ||
fdd2e5f8 | 65 | static struct kmem_cache *anon_vma_cachep; |
5beb4930 | 66 | static struct kmem_cache *anon_vma_chain_cachep; |
fdd2e5f8 AB |
67 | |
68 | static inline struct anon_vma *anon_vma_alloc(void) | |
69 | { | |
70 | return kmem_cache_alloc(anon_vma_cachep, GFP_KERNEL); | |
71 | } | |
72 | ||
db114b83 | 73 | void anon_vma_free(struct anon_vma *anon_vma) |
fdd2e5f8 AB |
74 | { |
75 | kmem_cache_free(anon_vma_cachep, anon_vma); | |
76 | } | |
1da177e4 | 77 | |
5beb4930 RR |
78 | static inline struct anon_vma_chain *anon_vma_chain_alloc(void) |
79 | { | |
80 | return kmem_cache_alloc(anon_vma_chain_cachep, GFP_KERNEL); | |
81 | } | |
82 | ||
e574b5fd | 83 | static void anon_vma_chain_free(struct anon_vma_chain *anon_vma_chain) |
5beb4930 RR |
84 | { |
85 | kmem_cache_free(anon_vma_chain_cachep, anon_vma_chain); | |
86 | } | |
87 | ||
d9d332e0 LT |
88 | /** |
89 | * anon_vma_prepare - attach an anon_vma to a memory region | |
90 | * @vma: the memory region in question | |
91 | * | |
92 | * This makes sure the memory mapping described by 'vma' has | |
93 | * an 'anon_vma' attached to it, so that we can associate the | |
94 | * anonymous pages mapped into it with that anon_vma. | |
95 | * | |
96 | * The common case will be that we already have one, but if | |
23a0790a | 97 | * not we either need to find an adjacent mapping that we |
d9d332e0 LT |
98 | * can re-use the anon_vma from (very common when the only |
99 | * reason for splitting a vma has been mprotect()), or we | |
100 | * allocate a new one. | |
101 | * | |
102 | * Anon-vma allocations are very subtle, because we may have | |
103 | * optimistically looked up an anon_vma in page_lock_anon_vma() | |
104 | * and that may actually touch the spinlock even in the newly | |
105 | * allocated vma (it depends on RCU to make sure that the | |
106 | * anon_vma isn't actually destroyed). | |
107 | * | |
108 | * As a result, we need to do proper anon_vma locking even | |
109 | * for the new allocation. At the same time, we do not want | |
110 | * to do any locking for the common case of already having | |
111 | * an anon_vma. | |
112 | * | |
113 | * This must be called with the mmap_sem held for reading. | |
114 | */ | |
1da177e4 LT |
115 | int anon_vma_prepare(struct vm_area_struct *vma) |
116 | { | |
117 | struct anon_vma *anon_vma = vma->anon_vma; | |
5beb4930 | 118 | struct anon_vma_chain *avc; |
1da177e4 LT |
119 | |
120 | might_sleep(); | |
121 | if (unlikely(!anon_vma)) { | |
122 | struct mm_struct *mm = vma->vm_mm; | |
d9d332e0 | 123 | struct anon_vma *allocated; |
1da177e4 | 124 | |
5beb4930 RR |
125 | avc = anon_vma_chain_alloc(); |
126 | if (!avc) | |
127 | goto out_enomem; | |
128 | ||
1da177e4 | 129 | anon_vma = find_mergeable_anon_vma(vma); |
d9d332e0 LT |
130 | allocated = NULL; |
131 | if (!anon_vma) { | |
1da177e4 LT |
132 | anon_vma = anon_vma_alloc(); |
133 | if (unlikely(!anon_vma)) | |
5beb4930 | 134 | goto out_enomem_free_avc; |
1da177e4 | 135 | allocated = anon_vma; |
5c341ee1 RR |
136 | /* |
137 | * This VMA had no anon_vma yet. This anon_vma is | |
138 | * the root of any anon_vma tree that might form. | |
139 | */ | |
140 | anon_vma->root = anon_vma; | |
1da177e4 LT |
141 | } |
142 | ||
cba48b98 | 143 | anon_vma_lock(anon_vma); |
1da177e4 LT |
144 | /* page_table_lock to protect against threads */ |
145 | spin_lock(&mm->page_table_lock); | |
146 | if (likely(!vma->anon_vma)) { | |
147 | vma->anon_vma = anon_vma; | |
5beb4930 RR |
148 | avc->anon_vma = anon_vma; |
149 | avc->vma = vma; | |
150 | list_add(&avc->same_vma, &vma->anon_vma_chain); | |
26ba0cb6 | 151 | list_add_tail(&avc->same_anon_vma, &anon_vma->head); |
1da177e4 | 152 | allocated = NULL; |
31f2b0eb | 153 | avc = NULL; |
1da177e4 LT |
154 | } |
155 | spin_unlock(&mm->page_table_lock); | |
cba48b98 | 156 | anon_vma_unlock(anon_vma); |
31f2b0eb ON |
157 | |
158 | if (unlikely(allocated)) | |
1da177e4 | 159 | anon_vma_free(allocated); |
31f2b0eb | 160 | if (unlikely(avc)) |
5beb4930 | 161 | anon_vma_chain_free(avc); |
1da177e4 LT |
162 | } |
163 | return 0; | |
5beb4930 RR |
164 | |
165 | out_enomem_free_avc: | |
166 | anon_vma_chain_free(avc); | |
167 | out_enomem: | |
168 | return -ENOMEM; | |
1da177e4 LT |
169 | } |
170 | ||
5beb4930 RR |
171 | static void anon_vma_chain_link(struct vm_area_struct *vma, |
172 | struct anon_vma_chain *avc, | |
173 | struct anon_vma *anon_vma) | |
1da177e4 | 174 | { |
5beb4930 RR |
175 | avc->vma = vma; |
176 | avc->anon_vma = anon_vma; | |
177 | list_add(&avc->same_vma, &vma->anon_vma_chain); | |
178 | ||
cba48b98 | 179 | anon_vma_lock(anon_vma); |
05759d38 AA |
180 | /* |
181 | * It's critical to add new vmas to the tail of the anon_vma, | |
182 | * see comment in huge_memory.c:__split_huge_page(). | |
183 | */ | |
5beb4930 | 184 | list_add_tail(&avc->same_anon_vma, &anon_vma->head); |
cba48b98 | 185 | anon_vma_unlock(anon_vma); |
1da177e4 LT |
186 | } |
187 | ||
5beb4930 RR |
188 | /* |
189 | * Attach the anon_vmas from src to dst. | |
190 | * Returns 0 on success, -ENOMEM on failure. | |
191 | */ | |
192 | int anon_vma_clone(struct vm_area_struct *dst, struct vm_area_struct *src) | |
1da177e4 | 193 | { |
5beb4930 RR |
194 | struct anon_vma_chain *avc, *pavc; |
195 | ||
646d87b4 | 196 | list_for_each_entry_reverse(pavc, &src->anon_vma_chain, same_vma) { |
5beb4930 RR |
197 | avc = anon_vma_chain_alloc(); |
198 | if (!avc) | |
199 | goto enomem_failure; | |
200 | anon_vma_chain_link(dst, avc, pavc->anon_vma); | |
201 | } | |
202 | return 0; | |
1da177e4 | 203 | |
5beb4930 RR |
204 | enomem_failure: |
205 | unlink_anon_vmas(dst); | |
206 | return -ENOMEM; | |
1da177e4 LT |
207 | } |
208 | ||
5beb4930 RR |
209 | /* |
210 | * Attach vma to its own anon_vma, as well as to the anon_vmas that | |
211 | * the corresponding VMA in the parent process is attached to. | |
212 | * Returns 0 on success, non-zero on failure. | |
213 | */ | |
214 | int anon_vma_fork(struct vm_area_struct *vma, struct vm_area_struct *pvma) | |
1da177e4 | 215 | { |
5beb4930 RR |
216 | struct anon_vma_chain *avc; |
217 | struct anon_vma *anon_vma; | |
1da177e4 | 218 | |
5beb4930 RR |
219 | /* Don't bother if the parent process has no anon_vma here. */ |
220 | if (!pvma->anon_vma) | |
221 | return 0; | |
222 | ||
223 | /* | |
224 | * First, attach the new VMA to the parent VMA's anon_vmas, | |
225 | * so rmap can find non-COWed pages in child processes. | |
226 | */ | |
227 | if (anon_vma_clone(vma, pvma)) | |
228 | return -ENOMEM; | |
229 | ||
230 | /* Then add our own anon_vma. */ | |
231 | anon_vma = anon_vma_alloc(); | |
232 | if (!anon_vma) | |
233 | goto out_error; | |
234 | avc = anon_vma_chain_alloc(); | |
235 | if (!avc) | |
236 | goto out_error_free_anon_vma; | |
5c341ee1 RR |
237 | |
238 | /* | |
239 | * The root anon_vma's spinlock is the lock actually used when we | |
240 | * lock any of the anon_vmas in this anon_vma tree. | |
241 | */ | |
242 | anon_vma->root = pvma->anon_vma->root; | |
76545066 RR |
243 | /* |
244 | * With KSM refcounts, an anon_vma can stay around longer than the | |
245 | * process it belongs to. The root anon_vma needs to be pinned | |
246 | * until this anon_vma is freed, because the lock lives in the root. | |
247 | */ | |
248 | get_anon_vma(anon_vma->root); | |
5beb4930 RR |
249 | /* Mark this anon_vma as the one where our new (COWed) pages go. */ |
250 | vma->anon_vma = anon_vma; | |
5c341ee1 | 251 | anon_vma_chain_link(vma, avc, anon_vma); |
5beb4930 RR |
252 | |
253 | return 0; | |
254 | ||
255 | out_error_free_anon_vma: | |
256 | anon_vma_free(anon_vma); | |
257 | out_error: | |
4946d54c | 258 | unlink_anon_vmas(vma); |
5beb4930 | 259 | return -ENOMEM; |
1da177e4 LT |
260 | } |
261 | ||
5beb4930 | 262 | static void anon_vma_unlink(struct anon_vma_chain *anon_vma_chain) |
1da177e4 | 263 | { |
5beb4930 | 264 | struct anon_vma *anon_vma = anon_vma_chain->anon_vma; |
1da177e4 LT |
265 | int empty; |
266 | ||
5beb4930 | 267 | /* If anon_vma_fork fails, we can get an empty anon_vma_chain. */ |
1da177e4 LT |
268 | if (!anon_vma) |
269 | return; | |
270 | ||
cba48b98 | 271 | anon_vma_lock(anon_vma); |
5beb4930 | 272 | list_del(&anon_vma_chain->same_anon_vma); |
1da177e4 LT |
273 | |
274 | /* We must garbage collect the anon_vma if it's empty */ | |
7f60c214 | 275 | empty = list_empty(&anon_vma->head) && !anonvma_external_refcount(anon_vma); |
cba48b98 | 276 | anon_vma_unlock(anon_vma); |
1da177e4 | 277 | |
76545066 RR |
278 | if (empty) { |
279 | /* We no longer need the root anon_vma */ | |
280 | if (anon_vma->root != anon_vma) | |
281 | drop_anon_vma(anon_vma->root); | |
1da177e4 | 282 | anon_vma_free(anon_vma); |
76545066 | 283 | } |
1da177e4 LT |
284 | } |
285 | ||
5beb4930 RR |
286 | void unlink_anon_vmas(struct vm_area_struct *vma) |
287 | { | |
288 | struct anon_vma_chain *avc, *next; | |
289 | ||
5c341ee1 RR |
290 | /* |
291 | * Unlink each anon_vma chained to the VMA. This list is ordered | |
292 | * from newest to oldest, ensuring the root anon_vma gets freed last. | |
293 | */ | |
5beb4930 RR |
294 | list_for_each_entry_safe(avc, next, &vma->anon_vma_chain, same_vma) { |
295 | anon_vma_unlink(avc); | |
296 | list_del(&avc->same_vma); | |
297 | anon_vma_chain_free(avc); | |
298 | } | |
299 | } | |
300 | ||
51cc5068 | 301 | static void anon_vma_ctor(void *data) |
1da177e4 | 302 | { |
a35afb83 | 303 | struct anon_vma *anon_vma = data; |
1da177e4 | 304 | |
a35afb83 | 305 | spin_lock_init(&anon_vma->lock); |
7f60c214 | 306 | anonvma_external_refcount_init(anon_vma); |
a35afb83 | 307 | INIT_LIST_HEAD(&anon_vma->head); |
1da177e4 LT |
308 | } |
309 | ||
310 | void __init anon_vma_init(void) | |
311 | { | |
312 | anon_vma_cachep = kmem_cache_create("anon_vma", sizeof(struct anon_vma), | |
20c2df83 | 313 | 0, SLAB_DESTROY_BY_RCU|SLAB_PANIC, anon_vma_ctor); |
5beb4930 | 314 | anon_vma_chain_cachep = KMEM_CACHE(anon_vma_chain, SLAB_PANIC); |
1da177e4 LT |
315 | } |
316 | ||
317 | /* | |
318 | * Getting a lock on a stable anon_vma from a page off the LRU is | |
319 | * tricky: page_lock_anon_vma rely on RCU to guard against the races. | |
320 | */ | |
ea4525b6 | 321 | struct anon_vma *__page_lock_anon_vma(struct page *page) |
1da177e4 | 322 | { |
f1819427 | 323 | struct anon_vma *anon_vma, *root_anon_vma; |
1da177e4 LT |
324 | unsigned long anon_mapping; |
325 | ||
326 | rcu_read_lock(); | |
80e14822 | 327 | anon_mapping = (unsigned long) ACCESS_ONCE(page->mapping); |
3ca7b3c5 | 328 | if ((anon_mapping & PAGE_MAPPING_FLAGS) != PAGE_MAPPING_ANON) |
1da177e4 LT |
329 | goto out; |
330 | if (!page_mapped(page)) | |
331 | goto out; | |
332 | ||
333 | anon_vma = (struct anon_vma *) (anon_mapping - PAGE_MAPPING_ANON); | |
f1819427 HD |
334 | root_anon_vma = ACCESS_ONCE(anon_vma->root); |
335 | spin_lock(&root_anon_vma->lock); | |
336 | ||
337 | /* | |
338 | * If this page is still mapped, then its anon_vma cannot have been | |
339 | * freed. But if it has been unmapped, we have no security against | |
340 | * the anon_vma structure being freed and reused (for another anon_vma: | |
341 | * SLAB_DESTROY_BY_RCU guarantees that - so the spin_lock above cannot | |
342 | * corrupt): with anon_vma_prepare() or anon_vma_fork() redirecting | |
343 | * anon_vma->root before page_unlock_anon_vma() is called to unlock. | |
344 | */ | |
345 | if (page_mapped(page)) | |
346 | return anon_vma; | |
347 | ||
348 | spin_unlock(&root_anon_vma->lock); | |
1da177e4 LT |
349 | out: |
350 | rcu_read_unlock(); | |
34bbd704 ON |
351 | return NULL; |
352 | } | |
353 | ||
10be22df | 354 | void page_unlock_anon_vma(struct anon_vma *anon_vma) |
ea4525b6 NK |
355 | __releases(&anon_vma->root->lock) |
356 | __releases(RCU) | |
34bbd704 | 357 | { |
cba48b98 | 358 | anon_vma_unlock(anon_vma); |
34bbd704 | 359 | rcu_read_unlock(); |
1da177e4 LT |
360 | } |
361 | ||
362 | /* | |
3ad33b24 LS |
363 | * At what user virtual address is page expected in @vma? |
364 | * Returns virtual address or -EFAULT if page's index/offset is not | |
365 | * within the range mapped the @vma. | |
1da177e4 | 366 | */ |
71e3aac0 | 367 | inline unsigned long |
1da177e4 LT |
368 | vma_address(struct page *page, struct vm_area_struct *vma) |
369 | { | |
370 | pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT); | |
371 | unsigned long address; | |
372 | ||
0fe6e20b NH |
373 | if (unlikely(is_vm_hugetlb_page(vma))) |
374 | pgoff = page->index << huge_page_order(page_hstate(page)); | |
1da177e4 LT |
375 | address = vma->vm_start + ((pgoff - vma->vm_pgoff) << PAGE_SHIFT); |
376 | if (unlikely(address < vma->vm_start || address >= vma->vm_end)) { | |
3ad33b24 | 377 | /* page should be within @vma mapping range */ |
1da177e4 LT |
378 | return -EFAULT; |
379 | } | |
380 | return address; | |
381 | } | |
382 | ||
383 | /* | |
bf89c8c8 | 384 | * At what user virtual address is page expected in vma? |
ab941e0f | 385 | * Caller should check the page is actually part of the vma. |
1da177e4 LT |
386 | */ |
387 | unsigned long page_address_in_vma(struct page *page, struct vm_area_struct *vma) | |
388 | { | |
21d0d443 | 389 | if (PageAnon(page)) { |
4829b906 HD |
390 | struct anon_vma *page__anon_vma = page_anon_vma(page); |
391 | /* | |
392 | * Note: swapoff's unuse_vma() is more efficient with this | |
393 | * check, and needs it to match anon_vma when KSM is active. | |
394 | */ | |
395 | if (!vma->anon_vma || !page__anon_vma || | |
396 | vma->anon_vma->root != page__anon_vma->root) | |
21d0d443 AA |
397 | return -EFAULT; |
398 | } else if (page->mapping && !(vma->vm_flags & VM_NONLINEAR)) { | |
ee498ed7 HD |
399 | if (!vma->vm_file || |
400 | vma->vm_file->f_mapping != page->mapping) | |
1da177e4 LT |
401 | return -EFAULT; |
402 | } else | |
403 | return -EFAULT; | |
404 | return vma_address(page, vma); | |
405 | } | |
406 | ||
81b4082d ND |
407 | /* |
408 | * Check that @page is mapped at @address into @mm. | |
409 | * | |
479db0bf NP |
410 | * If @sync is false, page_check_address may perform a racy check to avoid |
411 | * the page table lock when the pte is not present (helpful when reclaiming | |
412 | * highly shared pages). | |
413 | * | |
b8072f09 | 414 | * On success returns with pte mapped and locked. |
81b4082d | 415 | */ |
e9a81a82 | 416 | pte_t *__page_check_address(struct page *page, struct mm_struct *mm, |
479db0bf | 417 | unsigned long address, spinlock_t **ptlp, int sync) |
81b4082d ND |
418 | { |
419 | pgd_t *pgd; | |
420 | pud_t *pud; | |
421 | pmd_t *pmd; | |
422 | pte_t *pte; | |
c0718806 | 423 | spinlock_t *ptl; |
81b4082d | 424 | |
0fe6e20b NH |
425 | if (unlikely(PageHuge(page))) { |
426 | pte = huge_pte_offset(mm, address); | |
427 | ptl = &mm->page_table_lock; | |
428 | goto check; | |
429 | } | |
430 | ||
81b4082d | 431 | pgd = pgd_offset(mm, address); |
c0718806 HD |
432 | if (!pgd_present(*pgd)) |
433 | return NULL; | |
434 | ||
435 | pud = pud_offset(pgd, address); | |
436 | if (!pud_present(*pud)) | |
437 | return NULL; | |
438 | ||
439 | pmd = pmd_offset(pud, address); | |
440 | if (!pmd_present(*pmd)) | |
441 | return NULL; | |
71e3aac0 AA |
442 | if (pmd_trans_huge(*pmd)) |
443 | return NULL; | |
c0718806 HD |
444 | |
445 | pte = pte_offset_map(pmd, address); | |
446 | /* Make a quick check before getting the lock */ | |
479db0bf | 447 | if (!sync && !pte_present(*pte)) { |
c0718806 HD |
448 | pte_unmap(pte); |
449 | return NULL; | |
450 | } | |
451 | ||
4c21e2f2 | 452 | ptl = pte_lockptr(mm, pmd); |
0fe6e20b | 453 | check: |
c0718806 HD |
454 | spin_lock(ptl); |
455 | if (pte_present(*pte) && page_to_pfn(page) == pte_pfn(*pte)) { | |
456 | *ptlp = ptl; | |
457 | return pte; | |
81b4082d | 458 | } |
c0718806 HD |
459 | pte_unmap_unlock(pte, ptl); |
460 | return NULL; | |
81b4082d ND |
461 | } |
462 | ||
b291f000 NP |
463 | /** |
464 | * page_mapped_in_vma - check whether a page is really mapped in a VMA | |
465 | * @page: the page to test | |
466 | * @vma: the VMA to test | |
467 | * | |
468 | * Returns 1 if the page is mapped into the page tables of the VMA, 0 | |
469 | * if the page is not mapped into the page tables of this VMA. Only | |
470 | * valid for normal file or anonymous VMAs. | |
471 | */ | |
6a46079c | 472 | int page_mapped_in_vma(struct page *page, struct vm_area_struct *vma) |
b291f000 NP |
473 | { |
474 | unsigned long address; | |
475 | pte_t *pte; | |
476 | spinlock_t *ptl; | |
477 | ||
478 | address = vma_address(page, vma); | |
479 | if (address == -EFAULT) /* out of vma range */ | |
480 | return 0; | |
481 | pte = page_check_address(page, vma->vm_mm, address, &ptl, 1); | |
482 | if (!pte) /* the page is not in this mm */ | |
483 | return 0; | |
484 | pte_unmap_unlock(pte, ptl); | |
485 | ||
486 | return 1; | |
487 | } | |
488 | ||
1da177e4 LT |
489 | /* |
490 | * Subfunctions of page_referenced: page_referenced_one called | |
491 | * repeatedly from either page_referenced_anon or page_referenced_file. | |
492 | */ | |
5ad64688 HD |
493 | int page_referenced_one(struct page *page, struct vm_area_struct *vma, |
494 | unsigned long address, unsigned int *mapcount, | |
495 | unsigned long *vm_flags) | |
1da177e4 LT |
496 | { |
497 | struct mm_struct *mm = vma->vm_mm; | |
1da177e4 LT |
498 | int referenced = 0; |
499 | ||
b291f000 NP |
500 | /* |
501 | * Don't want to elevate referenced for mlocked page that gets this far, | |
502 | * in order that it progresses to try_to_unmap and is moved to the | |
503 | * unevictable list. | |
504 | */ | |
5a9bbdcd | 505 | if (vma->vm_flags & VM_LOCKED) { |
71e3aac0 | 506 | *mapcount = 0; /* break early from loop */ |
03ef83af | 507 | *vm_flags |= VM_LOCKED; |
71e3aac0 | 508 | goto out; |
4917e5d0 | 509 | } |
1da177e4 | 510 | |
c0718806 HD |
511 | /* Pretend the page is referenced if the task has the |
512 | swap token and is in the middle of a page fault. */ | |
f7b7fd8f | 513 | if (mm != current->mm && has_swap_token(mm) && |
c0718806 HD |
514 | rwsem_is_locked(&mm->mmap_sem)) |
515 | referenced++; | |
516 | ||
71e3aac0 AA |
517 | if (unlikely(PageTransHuge(page))) { |
518 | pmd_t *pmd; | |
519 | ||
520 | spin_lock(&mm->page_table_lock); | |
521 | pmd = page_check_address_pmd(page, mm, address, | |
522 | PAGE_CHECK_ADDRESS_PMD_FLAG); | |
523 | if (pmd && !pmd_trans_splitting(*pmd) && | |
524 | pmdp_clear_flush_young_notify(vma, address, pmd)) | |
525 | referenced++; | |
526 | spin_unlock(&mm->page_table_lock); | |
527 | } else { | |
528 | pte_t *pte; | |
529 | spinlock_t *ptl; | |
530 | ||
531 | pte = page_check_address(page, mm, address, &ptl, 0); | |
532 | if (!pte) | |
533 | goto out; | |
534 | ||
535 | if (ptep_clear_flush_young_notify(vma, address, pte)) { | |
536 | /* | |
537 | * Don't treat a reference through a sequentially read | |
538 | * mapping as such. If the page has been used in | |
539 | * another mapping, we will catch it; if this other | |
540 | * mapping is already gone, the unmap path will have | |
541 | * set PG_referenced or activated the page. | |
542 | */ | |
543 | if (likely(!VM_SequentialReadHint(vma))) | |
544 | referenced++; | |
545 | } | |
546 | pte_unmap_unlock(pte, ptl); | |
547 | } | |
548 | ||
c0718806 | 549 | (*mapcount)--; |
273f047e | 550 | |
6fe6b7e3 WF |
551 | if (referenced) |
552 | *vm_flags |= vma->vm_flags; | |
273f047e | 553 | out: |
1da177e4 LT |
554 | return referenced; |
555 | } | |
556 | ||
bed7161a | 557 | static int page_referenced_anon(struct page *page, |
6fe6b7e3 WF |
558 | struct mem_cgroup *mem_cont, |
559 | unsigned long *vm_flags) | |
1da177e4 LT |
560 | { |
561 | unsigned int mapcount; | |
562 | struct anon_vma *anon_vma; | |
5beb4930 | 563 | struct anon_vma_chain *avc; |
1da177e4 LT |
564 | int referenced = 0; |
565 | ||
566 | anon_vma = page_lock_anon_vma(page); | |
567 | if (!anon_vma) | |
568 | return referenced; | |
569 | ||
570 | mapcount = page_mapcount(page); | |
5beb4930 RR |
571 | list_for_each_entry(avc, &anon_vma->head, same_anon_vma) { |
572 | struct vm_area_struct *vma = avc->vma; | |
1cb1729b HD |
573 | unsigned long address = vma_address(page, vma); |
574 | if (address == -EFAULT) | |
575 | continue; | |
bed7161a BS |
576 | /* |
577 | * If we are reclaiming on behalf of a cgroup, skip | |
578 | * counting on behalf of references from different | |
579 | * cgroups | |
580 | */ | |
bd845e38 | 581 | if (mem_cont && !mm_match_cgroup(vma->vm_mm, mem_cont)) |
bed7161a | 582 | continue; |
1cb1729b | 583 | referenced += page_referenced_one(page, vma, address, |
6fe6b7e3 | 584 | &mapcount, vm_flags); |
1da177e4 LT |
585 | if (!mapcount) |
586 | break; | |
587 | } | |
34bbd704 ON |
588 | |
589 | page_unlock_anon_vma(anon_vma); | |
1da177e4 LT |
590 | return referenced; |
591 | } | |
592 | ||
593 | /** | |
594 | * page_referenced_file - referenced check for object-based rmap | |
595 | * @page: the page we're checking references on. | |
43d8eac4 | 596 | * @mem_cont: target memory controller |
6fe6b7e3 | 597 | * @vm_flags: collect encountered vma->vm_flags who actually referenced the page |
1da177e4 LT |
598 | * |
599 | * For an object-based mapped page, find all the places it is mapped and | |
600 | * check/clear the referenced flag. This is done by following the page->mapping | |
601 | * pointer, then walking the chain of vmas it holds. It returns the number | |
602 | * of references it found. | |
603 | * | |
604 | * This function is only called from page_referenced for object-based pages. | |
605 | */ | |
bed7161a | 606 | static int page_referenced_file(struct page *page, |
6fe6b7e3 WF |
607 | struct mem_cgroup *mem_cont, |
608 | unsigned long *vm_flags) | |
1da177e4 LT |
609 | { |
610 | unsigned int mapcount; | |
611 | struct address_space *mapping = page->mapping; | |
612 | pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT); | |
613 | struct vm_area_struct *vma; | |
614 | struct prio_tree_iter iter; | |
615 | int referenced = 0; | |
616 | ||
617 | /* | |
618 | * The caller's checks on page->mapping and !PageAnon have made | |
619 | * sure that this is a file page: the check for page->mapping | |
620 | * excludes the case just before it gets set on an anon page. | |
621 | */ | |
622 | BUG_ON(PageAnon(page)); | |
623 | ||
624 | /* | |
625 | * The page lock not only makes sure that page->mapping cannot | |
626 | * suddenly be NULLified by truncation, it makes sure that the | |
627 | * structure at mapping cannot be freed and reused yet, | |
628 | * so we can safely take mapping->i_mmap_lock. | |
629 | */ | |
630 | BUG_ON(!PageLocked(page)); | |
631 | ||
632 | spin_lock(&mapping->i_mmap_lock); | |
633 | ||
634 | /* | |
635 | * i_mmap_lock does not stabilize mapcount at all, but mapcount | |
636 | * is more likely to be accurate if we note it after spinning. | |
637 | */ | |
638 | mapcount = page_mapcount(page); | |
639 | ||
640 | vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, pgoff, pgoff) { | |
1cb1729b HD |
641 | unsigned long address = vma_address(page, vma); |
642 | if (address == -EFAULT) | |
643 | continue; | |
bed7161a BS |
644 | /* |
645 | * If we are reclaiming on behalf of a cgroup, skip | |
646 | * counting on behalf of references from different | |
647 | * cgroups | |
648 | */ | |
bd845e38 | 649 | if (mem_cont && !mm_match_cgroup(vma->vm_mm, mem_cont)) |
bed7161a | 650 | continue; |
1cb1729b | 651 | referenced += page_referenced_one(page, vma, address, |
6fe6b7e3 | 652 | &mapcount, vm_flags); |
1da177e4 LT |
653 | if (!mapcount) |
654 | break; | |
655 | } | |
656 | ||
657 | spin_unlock(&mapping->i_mmap_lock); | |
658 | return referenced; | |
659 | } | |
660 | ||
661 | /** | |
662 | * page_referenced - test if the page was referenced | |
663 | * @page: the page to test | |
664 | * @is_locked: caller holds lock on the page | |
43d8eac4 | 665 | * @mem_cont: target memory controller |
6fe6b7e3 | 666 | * @vm_flags: collect encountered vma->vm_flags who actually referenced the page |
1da177e4 LT |
667 | * |
668 | * Quick test_and_clear_referenced for all mappings to a page, | |
669 | * returns the number of ptes which referenced the page. | |
670 | */ | |
6fe6b7e3 WF |
671 | int page_referenced(struct page *page, |
672 | int is_locked, | |
673 | struct mem_cgroup *mem_cont, | |
674 | unsigned long *vm_flags) | |
1da177e4 LT |
675 | { |
676 | int referenced = 0; | |
5ad64688 | 677 | int we_locked = 0; |
1da177e4 | 678 | |
6fe6b7e3 | 679 | *vm_flags = 0; |
3ca7b3c5 | 680 | if (page_mapped(page) && page_rmapping(page)) { |
5ad64688 HD |
681 | if (!is_locked && (!PageAnon(page) || PageKsm(page))) { |
682 | we_locked = trylock_page(page); | |
683 | if (!we_locked) { | |
684 | referenced++; | |
685 | goto out; | |
686 | } | |
687 | } | |
688 | if (unlikely(PageKsm(page))) | |
689 | referenced += page_referenced_ksm(page, mem_cont, | |
690 | vm_flags); | |
691 | else if (PageAnon(page)) | |
6fe6b7e3 WF |
692 | referenced += page_referenced_anon(page, mem_cont, |
693 | vm_flags); | |
5ad64688 | 694 | else if (page->mapping) |
6fe6b7e3 WF |
695 | referenced += page_referenced_file(page, mem_cont, |
696 | vm_flags); | |
5ad64688 | 697 | if (we_locked) |
1da177e4 | 698 | unlock_page(page); |
1da177e4 | 699 | } |
5ad64688 | 700 | out: |
5b7baf05 CB |
701 | if (page_test_and_clear_young(page)) |
702 | referenced++; | |
703 | ||
1da177e4 LT |
704 | return referenced; |
705 | } | |
706 | ||
1cb1729b HD |
707 | static int page_mkclean_one(struct page *page, struct vm_area_struct *vma, |
708 | unsigned long address) | |
d08b3851 PZ |
709 | { |
710 | struct mm_struct *mm = vma->vm_mm; | |
c2fda5fe | 711 | pte_t *pte; |
d08b3851 PZ |
712 | spinlock_t *ptl; |
713 | int ret = 0; | |
714 | ||
479db0bf | 715 | pte = page_check_address(page, mm, address, &ptl, 1); |
d08b3851 PZ |
716 | if (!pte) |
717 | goto out; | |
718 | ||
c2fda5fe PZ |
719 | if (pte_dirty(*pte) || pte_write(*pte)) { |
720 | pte_t entry; | |
d08b3851 | 721 | |
c2fda5fe | 722 | flush_cache_page(vma, address, pte_pfn(*pte)); |
cddb8a5c | 723 | entry = ptep_clear_flush_notify(vma, address, pte); |
c2fda5fe PZ |
724 | entry = pte_wrprotect(entry); |
725 | entry = pte_mkclean(entry); | |
d6e88e67 | 726 | set_pte_at(mm, address, pte, entry); |
c2fda5fe PZ |
727 | ret = 1; |
728 | } | |
d08b3851 | 729 | |
d08b3851 PZ |
730 | pte_unmap_unlock(pte, ptl); |
731 | out: | |
732 | return ret; | |
733 | } | |
734 | ||
735 | static int page_mkclean_file(struct address_space *mapping, struct page *page) | |
736 | { | |
737 | pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT); | |
738 | struct vm_area_struct *vma; | |
739 | struct prio_tree_iter iter; | |
740 | int ret = 0; | |
741 | ||
742 | BUG_ON(PageAnon(page)); | |
743 | ||
744 | spin_lock(&mapping->i_mmap_lock); | |
745 | vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, pgoff, pgoff) { | |
1cb1729b HD |
746 | if (vma->vm_flags & VM_SHARED) { |
747 | unsigned long address = vma_address(page, vma); | |
748 | if (address == -EFAULT) | |
749 | continue; | |
750 | ret += page_mkclean_one(page, vma, address); | |
751 | } | |
d08b3851 PZ |
752 | } |
753 | spin_unlock(&mapping->i_mmap_lock); | |
754 | return ret; | |
755 | } | |
756 | ||
757 | int page_mkclean(struct page *page) | |
758 | { | |
759 | int ret = 0; | |
760 | ||
761 | BUG_ON(!PageLocked(page)); | |
762 | ||
763 | if (page_mapped(page)) { | |
764 | struct address_space *mapping = page_mapping(page); | |
ce7e9fae | 765 | if (mapping) { |
d08b3851 | 766 | ret = page_mkclean_file(mapping, page); |
ce7e9fae | 767 | if (page_test_dirty(page)) { |
e2b8d7af | 768 | page_clear_dirty(page, 1); |
ce7e9fae CB |
769 | ret = 1; |
770 | } | |
6c210482 | 771 | } |
d08b3851 PZ |
772 | } |
773 | ||
774 | return ret; | |
775 | } | |
60b59bea | 776 | EXPORT_SYMBOL_GPL(page_mkclean); |
d08b3851 | 777 | |
c44b6743 RR |
778 | /** |
779 | * page_move_anon_rmap - move a page to our anon_vma | |
780 | * @page: the page to move to our anon_vma | |
781 | * @vma: the vma the page belongs to | |
782 | * @address: the user virtual address mapped | |
783 | * | |
784 | * When a page belongs exclusively to one process after a COW event, | |
785 | * that page can be moved into the anon_vma that belongs to just that | |
786 | * process, so the rmap code will not search the parent or sibling | |
787 | * processes. | |
788 | */ | |
789 | void page_move_anon_rmap(struct page *page, | |
790 | struct vm_area_struct *vma, unsigned long address) | |
791 | { | |
792 | struct anon_vma *anon_vma = vma->anon_vma; | |
793 | ||
794 | VM_BUG_ON(!PageLocked(page)); | |
795 | VM_BUG_ON(!anon_vma); | |
796 | VM_BUG_ON(page->index != linear_page_index(vma, address)); | |
797 | ||
798 | anon_vma = (void *) anon_vma + PAGE_MAPPING_ANON; | |
799 | page->mapping = (struct address_space *) anon_vma; | |
800 | } | |
801 | ||
9617d95e | 802 | /** |
4e1c1975 AK |
803 | * __page_set_anon_rmap - set up new anonymous rmap |
804 | * @page: Page to add to rmap | |
805 | * @vma: VM area to add page to. | |
806 | * @address: User virtual address of the mapping | |
e8a03feb | 807 | * @exclusive: the page is exclusively owned by the current process |
9617d95e NP |
808 | */ |
809 | static void __page_set_anon_rmap(struct page *page, | |
e8a03feb | 810 | struct vm_area_struct *vma, unsigned long address, int exclusive) |
9617d95e | 811 | { |
e8a03feb | 812 | struct anon_vma *anon_vma = vma->anon_vma; |
ea90002b | 813 | |
e8a03feb | 814 | BUG_ON(!anon_vma); |
ea90002b | 815 | |
4e1c1975 AK |
816 | if (PageAnon(page)) |
817 | return; | |
818 | ||
ea90002b | 819 | /* |
e8a03feb RR |
820 | * If the page isn't exclusively mapped into this vma, |
821 | * we must use the _oldest_ possible anon_vma for the | |
822 | * page mapping! | |
ea90002b | 823 | */ |
4e1c1975 | 824 | if (!exclusive) |
288468c3 | 825 | anon_vma = anon_vma->root; |
9617d95e | 826 | |
9617d95e NP |
827 | anon_vma = (void *) anon_vma + PAGE_MAPPING_ANON; |
828 | page->mapping = (struct address_space *) anon_vma; | |
9617d95e | 829 | page->index = linear_page_index(vma, address); |
9617d95e NP |
830 | } |
831 | ||
c97a9e10 | 832 | /** |
43d8eac4 | 833 | * __page_check_anon_rmap - sanity check anonymous rmap addition |
c97a9e10 NP |
834 | * @page: the page to add the mapping to |
835 | * @vma: the vm area in which the mapping is added | |
836 | * @address: the user virtual address mapped | |
837 | */ | |
838 | static void __page_check_anon_rmap(struct page *page, | |
839 | struct vm_area_struct *vma, unsigned long address) | |
840 | { | |
841 | #ifdef CONFIG_DEBUG_VM | |
842 | /* | |
843 | * The page's anon-rmap details (mapping and index) are guaranteed to | |
844 | * be set up correctly at this point. | |
845 | * | |
846 | * We have exclusion against page_add_anon_rmap because the caller | |
847 | * always holds the page locked, except if called from page_dup_rmap, | |
848 | * in which case the page is already known to be setup. | |
849 | * | |
850 | * We have exclusion against page_add_new_anon_rmap because those pages | |
851 | * are initially only visible via the pagetables, and the pte is locked | |
852 | * over the call to page_add_new_anon_rmap. | |
853 | */ | |
44ab57a0 | 854 | BUG_ON(page_anon_vma(page)->root != vma->anon_vma->root); |
c97a9e10 NP |
855 | BUG_ON(page->index != linear_page_index(vma, address)); |
856 | #endif | |
857 | } | |
858 | ||
1da177e4 LT |
859 | /** |
860 | * page_add_anon_rmap - add pte mapping to an anonymous page | |
861 | * @page: the page to add the mapping to | |
862 | * @vma: the vm area in which the mapping is added | |
863 | * @address: the user virtual address mapped | |
864 | * | |
5ad64688 | 865 | * The caller needs to hold the pte lock, and the page must be locked in |
80e14822 HD |
866 | * the anon_vma case: to serialize mapping,index checking after setting, |
867 | * and to ensure that PageAnon is not being upgraded racily to PageKsm | |
868 | * (but PageKsm is never downgraded to PageAnon). | |
1da177e4 LT |
869 | */ |
870 | void page_add_anon_rmap(struct page *page, | |
871 | struct vm_area_struct *vma, unsigned long address) | |
ad8c2ee8 RR |
872 | { |
873 | do_page_add_anon_rmap(page, vma, address, 0); | |
874 | } | |
875 | ||
876 | /* | |
877 | * Special version of the above for do_swap_page, which often runs | |
878 | * into pages that are exclusively owned by the current process. | |
879 | * Everybody else should continue to use page_add_anon_rmap above. | |
880 | */ | |
881 | void do_page_add_anon_rmap(struct page *page, | |
882 | struct vm_area_struct *vma, unsigned long address, int exclusive) | |
1da177e4 | 883 | { |
5ad64688 | 884 | int first = atomic_inc_and_test(&page->_mapcount); |
79134171 AA |
885 | if (first) { |
886 | if (!PageTransHuge(page)) | |
887 | __inc_zone_page_state(page, NR_ANON_PAGES); | |
888 | else | |
889 | __inc_zone_page_state(page, | |
890 | NR_ANON_TRANSPARENT_HUGEPAGES); | |
891 | } | |
5ad64688 HD |
892 | if (unlikely(PageKsm(page))) |
893 | return; | |
894 | ||
c97a9e10 NP |
895 | VM_BUG_ON(!PageLocked(page)); |
896 | VM_BUG_ON(address < vma->vm_start || address >= vma->vm_end); | |
5ad64688 | 897 | if (first) |
ad8c2ee8 | 898 | __page_set_anon_rmap(page, vma, address, exclusive); |
69029cd5 | 899 | else |
c97a9e10 | 900 | __page_check_anon_rmap(page, vma, address); |
1da177e4 LT |
901 | } |
902 | ||
43d8eac4 | 903 | /** |
9617d95e NP |
904 | * page_add_new_anon_rmap - add pte mapping to a new anonymous page |
905 | * @page: the page to add the mapping to | |
906 | * @vma: the vm area in which the mapping is added | |
907 | * @address: the user virtual address mapped | |
908 | * | |
909 | * Same as page_add_anon_rmap but must only be called on *new* pages. | |
910 | * This means the inc-and-test can be bypassed. | |
c97a9e10 | 911 | * Page does not have to be locked. |
9617d95e NP |
912 | */ |
913 | void page_add_new_anon_rmap(struct page *page, | |
914 | struct vm_area_struct *vma, unsigned long address) | |
915 | { | |
b5934c53 | 916 | VM_BUG_ON(address < vma->vm_start || address >= vma->vm_end); |
cbf84b7a HD |
917 | SetPageSwapBacked(page); |
918 | atomic_set(&page->_mapcount, 0); /* increment count (starts at -1) */ | |
79134171 AA |
919 | if (!PageTransHuge(page)) |
920 | __inc_zone_page_state(page, NR_ANON_PAGES); | |
921 | else | |
922 | __inc_zone_page_state(page, NR_ANON_TRANSPARENT_HUGEPAGES); | |
e8a03feb | 923 | __page_set_anon_rmap(page, vma, address, 1); |
b5934c53 | 924 | if (page_evictable(page, vma)) |
cbf84b7a | 925 | lru_cache_add_lru(page, LRU_ACTIVE_ANON); |
b5934c53 HD |
926 | else |
927 | add_page_to_unevictable_list(page); | |
9617d95e NP |
928 | } |
929 | ||
1da177e4 LT |
930 | /** |
931 | * page_add_file_rmap - add pte mapping to a file page | |
932 | * @page: the page to add the mapping to | |
933 | * | |
b8072f09 | 934 | * The caller needs to hold the pte lock. |
1da177e4 LT |
935 | */ |
936 | void page_add_file_rmap(struct page *page) | |
937 | { | |
d69b042f | 938 | if (atomic_inc_and_test(&page->_mapcount)) { |
65ba55f5 | 939 | __inc_zone_page_state(page, NR_FILE_MAPPED); |
2a7106f2 | 940 | mem_cgroup_inc_page_stat(page, MEMCG_NR_FILE_MAPPED); |
d69b042f | 941 | } |
1da177e4 LT |
942 | } |
943 | ||
944 | /** | |
945 | * page_remove_rmap - take down pte mapping from a page | |
946 | * @page: page to remove mapping from | |
947 | * | |
b8072f09 | 948 | * The caller needs to hold the pte lock. |
1da177e4 | 949 | */ |
edc315fd | 950 | void page_remove_rmap(struct page *page) |
1da177e4 | 951 | { |
b904dcfe KM |
952 | /* page still mapped by someone else? */ |
953 | if (!atomic_add_negative(-1, &page->_mapcount)) | |
954 | return; | |
955 | ||
956 | /* | |
957 | * Now that the last pte has gone, s390 must transfer dirty | |
958 | * flag from storage key to struct page. We can usually skip | |
959 | * this if the page is anon, so about to be freed; but perhaps | |
960 | * not if it's in swapcache - there might be another pte slot | |
961 | * containing the swap entry, but page not yet written to swap. | |
962 | */ | |
963 | if ((!PageAnon(page) || PageSwapCache(page)) && page_test_dirty(page)) { | |
e2b8d7af | 964 | page_clear_dirty(page, 1); |
b904dcfe | 965 | set_page_dirty(page); |
1da177e4 | 966 | } |
0fe6e20b NH |
967 | /* |
968 | * Hugepages are not counted in NR_ANON_PAGES nor NR_FILE_MAPPED | |
969 | * and not charged by memcg for now. | |
970 | */ | |
971 | if (unlikely(PageHuge(page))) | |
972 | return; | |
b904dcfe KM |
973 | if (PageAnon(page)) { |
974 | mem_cgroup_uncharge_page(page); | |
79134171 AA |
975 | if (!PageTransHuge(page)) |
976 | __dec_zone_page_state(page, NR_ANON_PAGES); | |
977 | else | |
978 | __dec_zone_page_state(page, | |
979 | NR_ANON_TRANSPARENT_HUGEPAGES); | |
b904dcfe KM |
980 | } else { |
981 | __dec_zone_page_state(page, NR_FILE_MAPPED); | |
2a7106f2 | 982 | mem_cgroup_dec_page_stat(page, MEMCG_NR_FILE_MAPPED); |
b904dcfe | 983 | } |
b904dcfe KM |
984 | /* |
985 | * It would be tidy to reset the PageAnon mapping here, | |
986 | * but that might overwrite a racing page_add_anon_rmap | |
987 | * which increments mapcount after us but sets mapping | |
988 | * before us: so leave the reset to free_hot_cold_page, | |
989 | * and remember that it's only reliable while mapped. | |
990 | * Leaving it set also helps swapoff to reinstate ptes | |
991 | * faster for those pages still in swapcache. | |
992 | */ | |
1da177e4 LT |
993 | } |
994 | ||
995 | /* | |
996 | * Subfunctions of try_to_unmap: try_to_unmap_one called | |
997 | * repeatedly from either try_to_unmap_anon or try_to_unmap_file. | |
998 | */ | |
5ad64688 HD |
999 | int try_to_unmap_one(struct page *page, struct vm_area_struct *vma, |
1000 | unsigned long address, enum ttu_flags flags) | |
1da177e4 LT |
1001 | { |
1002 | struct mm_struct *mm = vma->vm_mm; | |
1da177e4 LT |
1003 | pte_t *pte; |
1004 | pte_t pteval; | |
c0718806 | 1005 | spinlock_t *ptl; |
1da177e4 LT |
1006 | int ret = SWAP_AGAIN; |
1007 | ||
479db0bf | 1008 | pte = page_check_address(page, mm, address, &ptl, 0); |
c0718806 | 1009 | if (!pte) |
81b4082d | 1010 | goto out; |
1da177e4 LT |
1011 | |
1012 | /* | |
1013 | * If the page is mlock()d, we cannot swap it out. | |
1014 | * If it's recently referenced (perhaps page_referenced | |
1015 | * skipped over this mm) then we should reactivate it. | |
1016 | */ | |
14fa31b8 | 1017 | if (!(flags & TTU_IGNORE_MLOCK)) { |
caed0f48 KM |
1018 | if (vma->vm_flags & VM_LOCKED) |
1019 | goto out_mlock; | |
1020 | ||
af8e3354 | 1021 | if (TTU_ACTION(flags) == TTU_MUNLOCK) |
53f79acb | 1022 | goto out_unmap; |
14fa31b8 AK |
1023 | } |
1024 | if (!(flags & TTU_IGNORE_ACCESS)) { | |
b291f000 NP |
1025 | if (ptep_clear_flush_young_notify(vma, address, pte)) { |
1026 | ret = SWAP_FAIL; | |
1027 | goto out_unmap; | |
1028 | } | |
1029 | } | |
1da177e4 | 1030 | |
1da177e4 LT |
1031 | /* Nuke the page table entry. */ |
1032 | flush_cache_page(vma, address, page_to_pfn(page)); | |
cddb8a5c | 1033 | pteval = ptep_clear_flush_notify(vma, address, pte); |
1da177e4 LT |
1034 | |
1035 | /* Move the dirty bit to the physical page now the pte is gone. */ | |
1036 | if (pte_dirty(pteval)) | |
1037 | set_page_dirty(page); | |
1038 | ||
365e9c87 HD |
1039 | /* Update high watermark before we lower rss */ |
1040 | update_hiwater_rss(mm); | |
1041 | ||
888b9f7c AK |
1042 | if (PageHWPoison(page) && !(flags & TTU_IGNORE_HWPOISON)) { |
1043 | if (PageAnon(page)) | |
d559db08 | 1044 | dec_mm_counter(mm, MM_ANONPAGES); |
888b9f7c | 1045 | else |
d559db08 | 1046 | dec_mm_counter(mm, MM_FILEPAGES); |
888b9f7c AK |
1047 | set_pte_at(mm, address, pte, |
1048 | swp_entry_to_pte(make_hwpoison_entry(page))); | |
1049 | } else if (PageAnon(page)) { | |
4c21e2f2 | 1050 | swp_entry_t entry = { .val = page_private(page) }; |
0697212a CL |
1051 | |
1052 | if (PageSwapCache(page)) { | |
1053 | /* | |
1054 | * Store the swap location in the pte. | |
1055 | * See handle_pte_fault() ... | |
1056 | */ | |
570a335b HD |
1057 | if (swap_duplicate(entry) < 0) { |
1058 | set_pte_at(mm, address, pte, pteval); | |
1059 | ret = SWAP_FAIL; | |
1060 | goto out_unmap; | |
1061 | } | |
0697212a CL |
1062 | if (list_empty(&mm->mmlist)) { |
1063 | spin_lock(&mmlist_lock); | |
1064 | if (list_empty(&mm->mmlist)) | |
1065 | list_add(&mm->mmlist, &init_mm.mmlist); | |
1066 | spin_unlock(&mmlist_lock); | |
1067 | } | |
d559db08 | 1068 | dec_mm_counter(mm, MM_ANONPAGES); |
b084d435 | 1069 | inc_mm_counter(mm, MM_SWAPENTS); |
64cdd548 | 1070 | } else if (PAGE_MIGRATION) { |
0697212a CL |
1071 | /* |
1072 | * Store the pfn of the page in a special migration | |
1073 | * pte. do_swap_page() will wait until the migration | |
1074 | * pte is removed and then restart fault handling. | |
1075 | */ | |
14fa31b8 | 1076 | BUG_ON(TTU_ACTION(flags) != TTU_MIGRATION); |
0697212a | 1077 | entry = make_migration_entry(page, pte_write(pteval)); |
1da177e4 LT |
1078 | } |
1079 | set_pte_at(mm, address, pte, swp_entry_to_pte(entry)); | |
1080 | BUG_ON(pte_file(*pte)); | |
14fa31b8 | 1081 | } else if (PAGE_MIGRATION && (TTU_ACTION(flags) == TTU_MIGRATION)) { |
04e62a29 CL |
1082 | /* Establish migration entry for a file page */ |
1083 | swp_entry_t entry; | |
1084 | entry = make_migration_entry(page, pte_write(pteval)); | |
1085 | set_pte_at(mm, address, pte, swp_entry_to_pte(entry)); | |
1086 | } else | |
d559db08 | 1087 | dec_mm_counter(mm, MM_FILEPAGES); |
1da177e4 | 1088 | |
edc315fd | 1089 | page_remove_rmap(page); |
1da177e4 LT |
1090 | page_cache_release(page); |
1091 | ||
1092 | out_unmap: | |
c0718806 | 1093 | pte_unmap_unlock(pte, ptl); |
caed0f48 KM |
1094 | out: |
1095 | return ret; | |
53f79acb | 1096 | |
caed0f48 KM |
1097 | out_mlock: |
1098 | pte_unmap_unlock(pte, ptl); | |
1099 | ||
1100 | ||
1101 | /* | |
1102 | * We need mmap_sem locking, Otherwise VM_LOCKED check makes | |
1103 | * unstable result and race. Plus, We can't wait here because | |
1104 | * we now hold anon_vma->lock or mapping->i_mmap_lock. | |
1105 | * if trylock failed, the page remain in evictable lru and later | |
1106 | * vmscan could retry to move the page to unevictable lru if the | |
1107 | * page is actually mlocked. | |
1108 | */ | |
1109 | if (down_read_trylock(&vma->vm_mm->mmap_sem)) { | |
1110 | if (vma->vm_flags & VM_LOCKED) { | |
1111 | mlock_vma_page(page); | |
1112 | ret = SWAP_MLOCK; | |
53f79acb | 1113 | } |
caed0f48 | 1114 | up_read(&vma->vm_mm->mmap_sem); |
53f79acb | 1115 | } |
1da177e4 LT |
1116 | return ret; |
1117 | } | |
1118 | ||
1119 | /* | |
1120 | * objrmap doesn't work for nonlinear VMAs because the assumption that | |
1121 | * offset-into-file correlates with offset-into-virtual-addresses does not hold. | |
1122 | * Consequently, given a particular page and its ->index, we cannot locate the | |
1123 | * ptes which are mapping that page without an exhaustive linear search. | |
1124 | * | |
1125 | * So what this code does is a mini "virtual scan" of each nonlinear VMA which | |
1126 | * maps the file to which the target page belongs. The ->vm_private_data field | |
1127 | * holds the current cursor into that scan. Successive searches will circulate | |
1128 | * around the vma's virtual address space. | |
1129 | * | |
1130 | * So as more replacement pressure is applied to the pages in a nonlinear VMA, | |
1131 | * more scanning pressure is placed against them as well. Eventually pages | |
1132 | * will become fully unmapped and are eligible for eviction. | |
1133 | * | |
1134 | * For very sparsely populated VMAs this is a little inefficient - chances are | |
1135 | * there there won't be many ptes located within the scan cluster. In this case | |
1136 | * maybe we could scan further - to the end of the pte page, perhaps. | |
b291f000 NP |
1137 | * |
1138 | * Mlocked pages: check VM_LOCKED under mmap_sem held for read, if we can | |
1139 | * acquire it without blocking. If vma locked, mlock the pages in the cluster, | |
1140 | * rather than unmapping them. If we encounter the "check_page" that vmscan is | |
1141 | * trying to unmap, return SWAP_MLOCK, else default SWAP_AGAIN. | |
1da177e4 LT |
1142 | */ |
1143 | #define CLUSTER_SIZE min(32*PAGE_SIZE, PMD_SIZE) | |
1144 | #define CLUSTER_MASK (~(CLUSTER_SIZE - 1)) | |
1145 | ||
b291f000 NP |
1146 | static int try_to_unmap_cluster(unsigned long cursor, unsigned int *mapcount, |
1147 | struct vm_area_struct *vma, struct page *check_page) | |
1da177e4 LT |
1148 | { |
1149 | struct mm_struct *mm = vma->vm_mm; | |
1150 | pgd_t *pgd; | |
1151 | pud_t *pud; | |
1152 | pmd_t *pmd; | |
c0718806 | 1153 | pte_t *pte; |
1da177e4 | 1154 | pte_t pteval; |
c0718806 | 1155 | spinlock_t *ptl; |
1da177e4 LT |
1156 | struct page *page; |
1157 | unsigned long address; | |
1158 | unsigned long end; | |
b291f000 NP |
1159 | int ret = SWAP_AGAIN; |
1160 | int locked_vma = 0; | |
1da177e4 | 1161 | |
1da177e4 LT |
1162 | address = (vma->vm_start + cursor) & CLUSTER_MASK; |
1163 | end = address + CLUSTER_SIZE; | |
1164 | if (address < vma->vm_start) | |
1165 | address = vma->vm_start; | |
1166 | if (end > vma->vm_end) | |
1167 | end = vma->vm_end; | |
1168 | ||
1169 | pgd = pgd_offset(mm, address); | |
1170 | if (!pgd_present(*pgd)) | |
b291f000 | 1171 | return ret; |
1da177e4 LT |
1172 | |
1173 | pud = pud_offset(pgd, address); | |
1174 | if (!pud_present(*pud)) | |
b291f000 | 1175 | return ret; |
1da177e4 LT |
1176 | |
1177 | pmd = pmd_offset(pud, address); | |
1178 | if (!pmd_present(*pmd)) | |
b291f000 NP |
1179 | return ret; |
1180 | ||
1181 | /* | |
af8e3354 | 1182 | * If we can acquire the mmap_sem for read, and vma is VM_LOCKED, |
b291f000 NP |
1183 | * keep the sem while scanning the cluster for mlocking pages. |
1184 | */ | |
af8e3354 | 1185 | if (down_read_trylock(&vma->vm_mm->mmap_sem)) { |
b291f000 NP |
1186 | locked_vma = (vma->vm_flags & VM_LOCKED); |
1187 | if (!locked_vma) | |
1188 | up_read(&vma->vm_mm->mmap_sem); /* don't need it */ | |
1189 | } | |
c0718806 HD |
1190 | |
1191 | pte = pte_offset_map_lock(mm, pmd, address, &ptl); | |
1da177e4 | 1192 | |
365e9c87 HD |
1193 | /* Update high watermark before we lower rss */ |
1194 | update_hiwater_rss(mm); | |
1195 | ||
c0718806 | 1196 | for (; address < end; pte++, address += PAGE_SIZE) { |
1da177e4 LT |
1197 | if (!pte_present(*pte)) |
1198 | continue; | |
6aab341e LT |
1199 | page = vm_normal_page(vma, address, *pte); |
1200 | BUG_ON(!page || PageAnon(page)); | |
1da177e4 | 1201 | |
b291f000 NP |
1202 | if (locked_vma) { |
1203 | mlock_vma_page(page); /* no-op if already mlocked */ | |
1204 | if (page == check_page) | |
1205 | ret = SWAP_MLOCK; | |
1206 | continue; /* don't unmap */ | |
1207 | } | |
1208 | ||
cddb8a5c | 1209 | if (ptep_clear_flush_young_notify(vma, address, pte)) |
1da177e4 LT |
1210 | continue; |
1211 | ||
1212 | /* Nuke the page table entry. */ | |
eca35133 | 1213 | flush_cache_page(vma, address, pte_pfn(*pte)); |
cddb8a5c | 1214 | pteval = ptep_clear_flush_notify(vma, address, pte); |
1da177e4 LT |
1215 | |
1216 | /* If nonlinear, store the file page offset in the pte. */ | |
1217 | if (page->index != linear_page_index(vma, address)) | |
1218 | set_pte_at(mm, address, pte, pgoff_to_pte(page->index)); | |
1219 | ||
1220 | /* Move the dirty bit to the physical page now the pte is gone. */ | |
1221 | if (pte_dirty(pteval)) | |
1222 | set_page_dirty(page); | |
1223 | ||
edc315fd | 1224 | page_remove_rmap(page); |
1da177e4 | 1225 | page_cache_release(page); |
d559db08 | 1226 | dec_mm_counter(mm, MM_FILEPAGES); |
1da177e4 LT |
1227 | (*mapcount)--; |
1228 | } | |
c0718806 | 1229 | pte_unmap_unlock(pte - 1, ptl); |
b291f000 NP |
1230 | if (locked_vma) |
1231 | up_read(&vma->vm_mm->mmap_sem); | |
1232 | return ret; | |
1da177e4 LT |
1233 | } |
1234 | ||
71e3aac0 | 1235 | bool is_vma_temporary_stack(struct vm_area_struct *vma) |
a8bef8ff MG |
1236 | { |
1237 | int maybe_stack = vma->vm_flags & (VM_GROWSDOWN | VM_GROWSUP); | |
1238 | ||
1239 | if (!maybe_stack) | |
1240 | return false; | |
1241 | ||
1242 | if ((vma->vm_flags & VM_STACK_INCOMPLETE_SETUP) == | |
1243 | VM_STACK_INCOMPLETE_SETUP) | |
1244 | return true; | |
1245 | ||
1246 | return false; | |
1247 | } | |
1248 | ||
b291f000 NP |
1249 | /** |
1250 | * try_to_unmap_anon - unmap or unlock anonymous page using the object-based | |
1251 | * rmap method | |
1252 | * @page: the page to unmap/unlock | |
8051be5e | 1253 | * @flags: action and flags |
b291f000 NP |
1254 | * |
1255 | * Find all the mappings of a page using the mapping pointer and the vma chains | |
1256 | * contained in the anon_vma struct it points to. | |
1257 | * | |
1258 | * This function is only called from try_to_unmap/try_to_munlock for | |
1259 | * anonymous pages. | |
1260 | * When called from try_to_munlock(), the mmap_sem of the mm containing the vma | |
1261 | * where the page was found will be held for write. So, we won't recheck | |
1262 | * vm_flags for that VMA. That should be OK, because that vma shouldn't be | |
1263 | * 'LOCKED. | |
1264 | */ | |
14fa31b8 | 1265 | static int try_to_unmap_anon(struct page *page, enum ttu_flags flags) |
1da177e4 LT |
1266 | { |
1267 | struct anon_vma *anon_vma; | |
5beb4930 | 1268 | struct anon_vma_chain *avc; |
1da177e4 | 1269 | int ret = SWAP_AGAIN; |
b291f000 | 1270 | |
1da177e4 LT |
1271 | anon_vma = page_lock_anon_vma(page); |
1272 | if (!anon_vma) | |
1273 | return ret; | |
1274 | ||
5beb4930 RR |
1275 | list_for_each_entry(avc, &anon_vma->head, same_anon_vma) { |
1276 | struct vm_area_struct *vma = avc->vma; | |
a8bef8ff MG |
1277 | unsigned long address; |
1278 | ||
1279 | /* | |
1280 | * During exec, a temporary VMA is setup and later moved. | |
1281 | * The VMA is moved under the anon_vma lock but not the | |
1282 | * page tables leading to a race where migration cannot | |
1283 | * find the migration ptes. Rather than increasing the | |
1284 | * locking requirements of exec(), migration skips | |
1285 | * temporary VMAs until after exec() completes. | |
1286 | */ | |
1287 | if (PAGE_MIGRATION && (flags & TTU_MIGRATION) && | |
1288 | is_vma_temporary_stack(vma)) | |
1289 | continue; | |
1290 | ||
1291 | address = vma_address(page, vma); | |
1cb1729b HD |
1292 | if (address == -EFAULT) |
1293 | continue; | |
1294 | ret = try_to_unmap_one(page, vma, address, flags); | |
53f79acb HD |
1295 | if (ret != SWAP_AGAIN || !page_mapped(page)) |
1296 | break; | |
1da177e4 | 1297 | } |
34bbd704 ON |
1298 | |
1299 | page_unlock_anon_vma(anon_vma); | |
1da177e4 LT |
1300 | return ret; |
1301 | } | |
1302 | ||
1303 | /** | |
b291f000 NP |
1304 | * try_to_unmap_file - unmap/unlock file page using the object-based rmap method |
1305 | * @page: the page to unmap/unlock | |
14fa31b8 | 1306 | * @flags: action and flags |
1da177e4 LT |
1307 | * |
1308 | * Find all the mappings of a page using the mapping pointer and the vma chains | |
1309 | * contained in the address_space struct it points to. | |
1310 | * | |
b291f000 NP |
1311 | * This function is only called from try_to_unmap/try_to_munlock for |
1312 | * object-based pages. | |
1313 | * When called from try_to_munlock(), the mmap_sem of the mm containing the vma | |
1314 | * where the page was found will be held for write. So, we won't recheck | |
1315 | * vm_flags for that VMA. That should be OK, because that vma shouldn't be | |
1316 | * 'LOCKED. | |
1da177e4 | 1317 | */ |
14fa31b8 | 1318 | static int try_to_unmap_file(struct page *page, enum ttu_flags flags) |
1da177e4 LT |
1319 | { |
1320 | struct address_space *mapping = page->mapping; | |
1321 | pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT); | |
1322 | struct vm_area_struct *vma; | |
1323 | struct prio_tree_iter iter; | |
1324 | int ret = SWAP_AGAIN; | |
1325 | unsigned long cursor; | |
1326 | unsigned long max_nl_cursor = 0; | |
1327 | unsigned long max_nl_size = 0; | |
1328 | unsigned int mapcount; | |
1329 | ||
1330 | spin_lock(&mapping->i_mmap_lock); | |
1331 | vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, pgoff, pgoff) { | |
1cb1729b HD |
1332 | unsigned long address = vma_address(page, vma); |
1333 | if (address == -EFAULT) | |
1334 | continue; | |
1335 | ret = try_to_unmap_one(page, vma, address, flags); | |
53f79acb HD |
1336 | if (ret != SWAP_AGAIN || !page_mapped(page)) |
1337 | goto out; | |
1da177e4 LT |
1338 | } |
1339 | ||
1340 | if (list_empty(&mapping->i_mmap_nonlinear)) | |
1341 | goto out; | |
1342 | ||
53f79acb HD |
1343 | /* |
1344 | * We don't bother to try to find the munlocked page in nonlinears. | |
1345 | * It's costly. Instead, later, page reclaim logic may call | |
1346 | * try_to_unmap(TTU_MUNLOCK) and recover PG_mlocked lazily. | |
1347 | */ | |
1348 | if (TTU_ACTION(flags) == TTU_MUNLOCK) | |
1349 | goto out; | |
1350 | ||
1da177e4 LT |
1351 | list_for_each_entry(vma, &mapping->i_mmap_nonlinear, |
1352 | shared.vm_set.list) { | |
1da177e4 LT |
1353 | cursor = (unsigned long) vma->vm_private_data; |
1354 | if (cursor > max_nl_cursor) | |
1355 | max_nl_cursor = cursor; | |
1356 | cursor = vma->vm_end - vma->vm_start; | |
1357 | if (cursor > max_nl_size) | |
1358 | max_nl_size = cursor; | |
1359 | } | |
1360 | ||
b291f000 | 1361 | if (max_nl_size == 0) { /* all nonlinears locked or reserved ? */ |
1da177e4 LT |
1362 | ret = SWAP_FAIL; |
1363 | goto out; | |
1364 | } | |
1365 | ||
1366 | /* | |
1367 | * We don't try to search for this page in the nonlinear vmas, | |
1368 | * and page_referenced wouldn't have found it anyway. Instead | |
1369 | * just walk the nonlinear vmas trying to age and unmap some. | |
1370 | * The mapcount of the page we came in with is irrelevant, | |
1371 | * but even so use it as a guide to how hard we should try? | |
1372 | */ | |
1373 | mapcount = page_mapcount(page); | |
1374 | if (!mapcount) | |
1375 | goto out; | |
1376 | cond_resched_lock(&mapping->i_mmap_lock); | |
1377 | ||
1378 | max_nl_size = (max_nl_size + CLUSTER_SIZE - 1) & CLUSTER_MASK; | |
1379 | if (max_nl_cursor == 0) | |
1380 | max_nl_cursor = CLUSTER_SIZE; | |
1381 | ||
1382 | do { | |
1383 | list_for_each_entry(vma, &mapping->i_mmap_nonlinear, | |
1384 | shared.vm_set.list) { | |
1da177e4 | 1385 | cursor = (unsigned long) vma->vm_private_data; |
839b9685 | 1386 | while ( cursor < max_nl_cursor && |
1da177e4 | 1387 | cursor < vma->vm_end - vma->vm_start) { |
53f79acb HD |
1388 | if (try_to_unmap_cluster(cursor, &mapcount, |
1389 | vma, page) == SWAP_MLOCK) | |
1390 | ret = SWAP_MLOCK; | |
1da177e4 LT |
1391 | cursor += CLUSTER_SIZE; |
1392 | vma->vm_private_data = (void *) cursor; | |
1393 | if ((int)mapcount <= 0) | |
1394 | goto out; | |
1395 | } | |
1396 | vma->vm_private_data = (void *) max_nl_cursor; | |
1397 | } | |
1398 | cond_resched_lock(&mapping->i_mmap_lock); | |
1399 | max_nl_cursor += CLUSTER_SIZE; | |
1400 | } while (max_nl_cursor <= max_nl_size); | |
1401 | ||
1402 | /* | |
1403 | * Don't loop forever (perhaps all the remaining pages are | |
1404 | * in locked vmas). Reset cursor on all unreserved nonlinear | |
1405 | * vmas, now forgetting on which ones it had fallen behind. | |
1406 | */ | |
101d2be7 HD |
1407 | list_for_each_entry(vma, &mapping->i_mmap_nonlinear, shared.vm_set.list) |
1408 | vma->vm_private_data = NULL; | |
1da177e4 LT |
1409 | out: |
1410 | spin_unlock(&mapping->i_mmap_lock); | |
1411 | return ret; | |
1412 | } | |
1413 | ||
1414 | /** | |
1415 | * try_to_unmap - try to remove all page table mappings to a page | |
1416 | * @page: the page to get unmapped | |
14fa31b8 | 1417 | * @flags: action and flags |
1da177e4 LT |
1418 | * |
1419 | * Tries to remove all the page table entries which are mapping this | |
1420 | * page, used in the pageout path. Caller must hold the page lock. | |
1421 | * Return values are: | |
1422 | * | |
1423 | * SWAP_SUCCESS - we succeeded in removing all mappings | |
1424 | * SWAP_AGAIN - we missed a mapping, try again later | |
1425 | * SWAP_FAIL - the page is unswappable | |
b291f000 | 1426 | * SWAP_MLOCK - page is mlocked. |
1da177e4 | 1427 | */ |
14fa31b8 | 1428 | int try_to_unmap(struct page *page, enum ttu_flags flags) |
1da177e4 LT |
1429 | { |
1430 | int ret; | |
1431 | ||
1da177e4 | 1432 | BUG_ON(!PageLocked(page)); |
91600e9e | 1433 | VM_BUG_ON(!PageHuge(page) && PageTransHuge(page)); |
1da177e4 | 1434 | |
5ad64688 HD |
1435 | if (unlikely(PageKsm(page))) |
1436 | ret = try_to_unmap_ksm(page, flags); | |
1437 | else if (PageAnon(page)) | |
14fa31b8 | 1438 | ret = try_to_unmap_anon(page, flags); |
1da177e4 | 1439 | else |
14fa31b8 | 1440 | ret = try_to_unmap_file(page, flags); |
b291f000 | 1441 | if (ret != SWAP_MLOCK && !page_mapped(page)) |
1da177e4 LT |
1442 | ret = SWAP_SUCCESS; |
1443 | return ret; | |
1444 | } | |
81b4082d | 1445 | |
b291f000 NP |
1446 | /** |
1447 | * try_to_munlock - try to munlock a page | |
1448 | * @page: the page to be munlocked | |
1449 | * | |
1450 | * Called from munlock code. Checks all of the VMAs mapping the page | |
1451 | * to make sure nobody else has this page mlocked. The page will be | |
1452 | * returned with PG_mlocked cleared if no other vmas have it mlocked. | |
1453 | * | |
1454 | * Return values are: | |
1455 | * | |
53f79acb | 1456 | * SWAP_AGAIN - no vma is holding page mlocked, or, |
b291f000 | 1457 | * SWAP_AGAIN - page mapped in mlocked vma -- couldn't acquire mmap sem |
5ad64688 | 1458 | * SWAP_FAIL - page cannot be located at present |
b291f000 NP |
1459 | * SWAP_MLOCK - page is now mlocked. |
1460 | */ | |
1461 | int try_to_munlock(struct page *page) | |
1462 | { | |
1463 | VM_BUG_ON(!PageLocked(page) || PageLRU(page)); | |
1464 | ||
5ad64688 HD |
1465 | if (unlikely(PageKsm(page))) |
1466 | return try_to_unmap_ksm(page, TTU_MUNLOCK); | |
1467 | else if (PageAnon(page)) | |
14fa31b8 | 1468 | return try_to_unmap_anon(page, TTU_MUNLOCK); |
b291f000 | 1469 | else |
14fa31b8 | 1470 | return try_to_unmap_file(page, TTU_MUNLOCK); |
b291f000 | 1471 | } |
e9995ef9 | 1472 | |
76545066 RR |
1473 | #if defined(CONFIG_KSM) || defined(CONFIG_MIGRATION) |
1474 | /* | |
1475 | * Drop an anon_vma refcount, freeing the anon_vma and anon_vma->root | |
1476 | * if necessary. Be careful to do all the tests under the lock. Once | |
1477 | * we know we are the last user, nobody else can get a reference and we | |
1478 | * can do the freeing without the lock. | |
1479 | */ | |
1480 | void drop_anon_vma(struct anon_vma *anon_vma) | |
1481 | { | |
44ab57a0 | 1482 | BUG_ON(atomic_read(&anon_vma->external_refcount) <= 0); |
76545066 RR |
1483 | if (atomic_dec_and_lock(&anon_vma->external_refcount, &anon_vma->root->lock)) { |
1484 | struct anon_vma *root = anon_vma->root; | |
1485 | int empty = list_empty(&anon_vma->head); | |
1486 | int last_root_user = 0; | |
1487 | int root_empty = 0; | |
1488 | ||
1489 | /* | |
1490 | * The refcount on a non-root anon_vma got dropped. Drop | |
1491 | * the refcount on the root and check if we need to free it. | |
1492 | */ | |
1493 | if (empty && anon_vma != root) { | |
44ab57a0 | 1494 | BUG_ON(atomic_read(&root->external_refcount) <= 0); |
76545066 RR |
1495 | last_root_user = atomic_dec_and_test(&root->external_refcount); |
1496 | root_empty = list_empty(&root->head); | |
1497 | } | |
1498 | anon_vma_unlock(anon_vma); | |
1499 | ||
1500 | if (empty) { | |
1501 | anon_vma_free(anon_vma); | |
1502 | if (root_empty && last_root_user) | |
1503 | anon_vma_free(root); | |
1504 | } | |
1505 | } | |
1506 | } | |
1507 | #endif | |
1508 | ||
e9995ef9 HD |
1509 | #ifdef CONFIG_MIGRATION |
1510 | /* | |
1511 | * rmap_walk() and its helpers rmap_walk_anon() and rmap_walk_file(): | |
1512 | * Called by migrate.c to remove migration ptes, but might be used more later. | |
1513 | */ | |
1514 | static int rmap_walk_anon(struct page *page, int (*rmap_one)(struct page *, | |
1515 | struct vm_area_struct *, unsigned long, void *), void *arg) | |
1516 | { | |
1517 | struct anon_vma *anon_vma; | |
5beb4930 | 1518 | struct anon_vma_chain *avc; |
e9995ef9 HD |
1519 | int ret = SWAP_AGAIN; |
1520 | ||
1521 | /* | |
1522 | * Note: remove_migration_ptes() cannot use page_lock_anon_vma() | |
1523 | * because that depends on page_mapped(); but not all its usages | |
3f6c8272 MG |
1524 | * are holding mmap_sem. Users without mmap_sem are required to |
1525 | * take a reference count to prevent the anon_vma disappearing | |
e9995ef9 HD |
1526 | */ |
1527 | anon_vma = page_anon_vma(page); | |
1528 | if (!anon_vma) | |
1529 | return ret; | |
cba48b98 | 1530 | anon_vma_lock(anon_vma); |
5beb4930 RR |
1531 | list_for_each_entry(avc, &anon_vma->head, same_anon_vma) { |
1532 | struct vm_area_struct *vma = avc->vma; | |
e9995ef9 HD |
1533 | unsigned long address = vma_address(page, vma); |
1534 | if (address == -EFAULT) | |
1535 | continue; | |
1536 | ret = rmap_one(page, vma, address, arg); | |
1537 | if (ret != SWAP_AGAIN) | |
1538 | break; | |
1539 | } | |
cba48b98 | 1540 | anon_vma_unlock(anon_vma); |
e9995ef9 HD |
1541 | return ret; |
1542 | } | |
1543 | ||
1544 | static int rmap_walk_file(struct page *page, int (*rmap_one)(struct page *, | |
1545 | struct vm_area_struct *, unsigned long, void *), void *arg) | |
1546 | { | |
1547 | struct address_space *mapping = page->mapping; | |
1548 | pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT); | |
1549 | struct vm_area_struct *vma; | |
1550 | struct prio_tree_iter iter; | |
1551 | int ret = SWAP_AGAIN; | |
1552 | ||
1553 | if (!mapping) | |
1554 | return ret; | |
1555 | spin_lock(&mapping->i_mmap_lock); | |
1556 | vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, pgoff, pgoff) { | |
1557 | unsigned long address = vma_address(page, vma); | |
1558 | if (address == -EFAULT) | |
1559 | continue; | |
1560 | ret = rmap_one(page, vma, address, arg); | |
1561 | if (ret != SWAP_AGAIN) | |
1562 | break; | |
1563 | } | |
1564 | /* | |
1565 | * No nonlinear handling: being always shared, nonlinear vmas | |
1566 | * never contain migration ptes. Decide what to do about this | |
1567 | * limitation to linear when we need rmap_walk() on nonlinear. | |
1568 | */ | |
1569 | spin_unlock(&mapping->i_mmap_lock); | |
1570 | return ret; | |
1571 | } | |
1572 | ||
1573 | int rmap_walk(struct page *page, int (*rmap_one)(struct page *, | |
1574 | struct vm_area_struct *, unsigned long, void *), void *arg) | |
1575 | { | |
1576 | VM_BUG_ON(!PageLocked(page)); | |
1577 | ||
1578 | if (unlikely(PageKsm(page))) | |
1579 | return rmap_walk_ksm(page, rmap_one, arg); | |
1580 | else if (PageAnon(page)) | |
1581 | return rmap_walk_anon(page, rmap_one, arg); | |
1582 | else | |
1583 | return rmap_walk_file(page, rmap_one, arg); | |
1584 | } | |
1585 | #endif /* CONFIG_MIGRATION */ | |
0fe6e20b | 1586 | |
e3390f67 | 1587 | #ifdef CONFIG_HUGETLB_PAGE |
0fe6e20b NH |
1588 | /* |
1589 | * The following three functions are for anonymous (private mapped) hugepages. | |
1590 | * Unlike common anonymous pages, anonymous hugepages have no accounting code | |
1591 | * and no lru code, because we handle hugepages differently from common pages. | |
1592 | */ | |
1593 | static void __hugepage_set_anon_rmap(struct page *page, | |
1594 | struct vm_area_struct *vma, unsigned long address, int exclusive) | |
1595 | { | |
1596 | struct anon_vma *anon_vma = vma->anon_vma; | |
433abed6 | 1597 | |
0fe6e20b | 1598 | BUG_ON(!anon_vma); |
433abed6 NH |
1599 | |
1600 | if (PageAnon(page)) | |
1601 | return; | |
1602 | if (!exclusive) | |
1603 | anon_vma = anon_vma->root; | |
1604 | ||
0fe6e20b NH |
1605 | anon_vma = (void *) anon_vma + PAGE_MAPPING_ANON; |
1606 | page->mapping = (struct address_space *) anon_vma; | |
1607 | page->index = linear_page_index(vma, address); | |
1608 | } | |
1609 | ||
1610 | void hugepage_add_anon_rmap(struct page *page, | |
1611 | struct vm_area_struct *vma, unsigned long address) | |
1612 | { | |
1613 | struct anon_vma *anon_vma = vma->anon_vma; | |
1614 | int first; | |
a850ea30 NH |
1615 | |
1616 | BUG_ON(!PageLocked(page)); | |
0fe6e20b NH |
1617 | BUG_ON(!anon_vma); |
1618 | BUG_ON(address < vma->vm_start || address >= vma->vm_end); | |
1619 | first = atomic_inc_and_test(&page->_mapcount); | |
1620 | if (first) | |
1621 | __hugepage_set_anon_rmap(page, vma, address, 0); | |
1622 | } | |
1623 | ||
1624 | void hugepage_add_new_anon_rmap(struct page *page, | |
1625 | struct vm_area_struct *vma, unsigned long address) | |
1626 | { | |
1627 | BUG_ON(address < vma->vm_start || address >= vma->vm_end); | |
1628 | atomic_set(&page->_mapcount, 0); | |
1629 | __hugepage_set_anon_rmap(page, vma, address, 1); | |
1630 | } | |
e3390f67 | 1631 | #endif /* CONFIG_HUGETLB_PAGE */ |