]>
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 | { | |
01d8b20d PZ |
70 | struct anon_vma *anon_vma; |
71 | ||
72 | anon_vma = kmem_cache_alloc(anon_vma_cachep, GFP_KERNEL); | |
73 | if (anon_vma) { | |
74 | atomic_set(&anon_vma->refcount, 1); | |
75 | /* | |
76 | * Initialise the anon_vma root to point to itself. If called | |
77 | * from fork, the root will be reset to the parents anon_vma. | |
78 | */ | |
79 | anon_vma->root = anon_vma; | |
80 | } | |
81 | ||
82 | return anon_vma; | |
fdd2e5f8 AB |
83 | } |
84 | ||
01d8b20d | 85 | static inline void anon_vma_free(struct anon_vma *anon_vma) |
fdd2e5f8 | 86 | { |
01d8b20d | 87 | VM_BUG_ON(atomic_read(&anon_vma->refcount)); |
fdd2e5f8 AB |
88 | kmem_cache_free(anon_vma_cachep, anon_vma); |
89 | } | |
1da177e4 | 90 | |
5beb4930 RR |
91 | static inline struct anon_vma_chain *anon_vma_chain_alloc(void) |
92 | { | |
93 | return kmem_cache_alloc(anon_vma_chain_cachep, GFP_KERNEL); | |
94 | } | |
95 | ||
e574b5fd | 96 | static void anon_vma_chain_free(struct anon_vma_chain *anon_vma_chain) |
5beb4930 RR |
97 | { |
98 | kmem_cache_free(anon_vma_chain_cachep, anon_vma_chain); | |
99 | } | |
100 | ||
d9d332e0 LT |
101 | /** |
102 | * anon_vma_prepare - attach an anon_vma to a memory region | |
103 | * @vma: the memory region in question | |
104 | * | |
105 | * This makes sure the memory mapping described by 'vma' has | |
106 | * an 'anon_vma' attached to it, so that we can associate the | |
107 | * anonymous pages mapped into it with that anon_vma. | |
108 | * | |
109 | * The common case will be that we already have one, but if | |
23a0790a | 110 | * not we either need to find an adjacent mapping that we |
d9d332e0 LT |
111 | * can re-use the anon_vma from (very common when the only |
112 | * reason for splitting a vma has been mprotect()), or we | |
113 | * allocate a new one. | |
114 | * | |
115 | * Anon-vma allocations are very subtle, because we may have | |
116 | * optimistically looked up an anon_vma in page_lock_anon_vma() | |
117 | * and that may actually touch the spinlock even in the newly | |
118 | * allocated vma (it depends on RCU to make sure that the | |
119 | * anon_vma isn't actually destroyed). | |
120 | * | |
121 | * As a result, we need to do proper anon_vma locking even | |
122 | * for the new allocation. At the same time, we do not want | |
123 | * to do any locking for the common case of already having | |
124 | * an anon_vma. | |
125 | * | |
126 | * This must be called with the mmap_sem held for reading. | |
127 | */ | |
1da177e4 LT |
128 | int anon_vma_prepare(struct vm_area_struct *vma) |
129 | { | |
130 | struct anon_vma *anon_vma = vma->anon_vma; | |
5beb4930 | 131 | struct anon_vma_chain *avc; |
1da177e4 LT |
132 | |
133 | might_sleep(); | |
134 | if (unlikely(!anon_vma)) { | |
135 | struct mm_struct *mm = vma->vm_mm; | |
d9d332e0 | 136 | struct anon_vma *allocated; |
1da177e4 | 137 | |
5beb4930 RR |
138 | avc = anon_vma_chain_alloc(); |
139 | if (!avc) | |
140 | goto out_enomem; | |
141 | ||
1da177e4 | 142 | anon_vma = find_mergeable_anon_vma(vma); |
d9d332e0 LT |
143 | allocated = NULL; |
144 | if (!anon_vma) { | |
1da177e4 LT |
145 | anon_vma = anon_vma_alloc(); |
146 | if (unlikely(!anon_vma)) | |
5beb4930 | 147 | goto out_enomem_free_avc; |
1da177e4 | 148 | allocated = anon_vma; |
1da177e4 LT |
149 | } |
150 | ||
cba48b98 | 151 | anon_vma_lock(anon_vma); |
1da177e4 LT |
152 | /* page_table_lock to protect against threads */ |
153 | spin_lock(&mm->page_table_lock); | |
154 | if (likely(!vma->anon_vma)) { | |
155 | vma->anon_vma = anon_vma; | |
5beb4930 RR |
156 | avc->anon_vma = anon_vma; |
157 | avc->vma = vma; | |
158 | list_add(&avc->same_vma, &vma->anon_vma_chain); | |
26ba0cb6 | 159 | list_add_tail(&avc->same_anon_vma, &anon_vma->head); |
1da177e4 | 160 | allocated = NULL; |
31f2b0eb | 161 | avc = NULL; |
1da177e4 LT |
162 | } |
163 | spin_unlock(&mm->page_table_lock); | |
cba48b98 | 164 | anon_vma_unlock(anon_vma); |
31f2b0eb ON |
165 | |
166 | if (unlikely(allocated)) | |
01d8b20d | 167 | put_anon_vma(allocated); |
31f2b0eb | 168 | if (unlikely(avc)) |
5beb4930 | 169 | anon_vma_chain_free(avc); |
1da177e4 LT |
170 | } |
171 | return 0; | |
5beb4930 RR |
172 | |
173 | out_enomem_free_avc: | |
174 | anon_vma_chain_free(avc); | |
175 | out_enomem: | |
176 | return -ENOMEM; | |
1da177e4 LT |
177 | } |
178 | ||
5beb4930 RR |
179 | static void anon_vma_chain_link(struct vm_area_struct *vma, |
180 | struct anon_vma_chain *avc, | |
181 | struct anon_vma *anon_vma) | |
1da177e4 | 182 | { |
5beb4930 RR |
183 | avc->vma = vma; |
184 | avc->anon_vma = anon_vma; | |
185 | list_add(&avc->same_vma, &vma->anon_vma_chain); | |
186 | ||
cba48b98 | 187 | anon_vma_lock(anon_vma); |
05759d38 AA |
188 | /* |
189 | * It's critical to add new vmas to the tail of the anon_vma, | |
190 | * see comment in huge_memory.c:__split_huge_page(). | |
191 | */ | |
5beb4930 | 192 | list_add_tail(&avc->same_anon_vma, &anon_vma->head); |
cba48b98 | 193 | anon_vma_unlock(anon_vma); |
1da177e4 LT |
194 | } |
195 | ||
5beb4930 RR |
196 | /* |
197 | * Attach the anon_vmas from src to dst. | |
198 | * Returns 0 on success, -ENOMEM on failure. | |
199 | */ | |
200 | int anon_vma_clone(struct vm_area_struct *dst, struct vm_area_struct *src) | |
1da177e4 | 201 | { |
5beb4930 RR |
202 | struct anon_vma_chain *avc, *pavc; |
203 | ||
646d87b4 | 204 | list_for_each_entry_reverse(pavc, &src->anon_vma_chain, same_vma) { |
5beb4930 RR |
205 | avc = anon_vma_chain_alloc(); |
206 | if (!avc) | |
207 | goto enomem_failure; | |
208 | anon_vma_chain_link(dst, avc, pavc->anon_vma); | |
209 | } | |
210 | return 0; | |
1da177e4 | 211 | |
5beb4930 RR |
212 | enomem_failure: |
213 | unlink_anon_vmas(dst); | |
214 | return -ENOMEM; | |
1da177e4 LT |
215 | } |
216 | ||
5beb4930 RR |
217 | /* |
218 | * Attach vma to its own anon_vma, as well as to the anon_vmas that | |
219 | * the corresponding VMA in the parent process is attached to. | |
220 | * Returns 0 on success, non-zero on failure. | |
221 | */ | |
222 | int anon_vma_fork(struct vm_area_struct *vma, struct vm_area_struct *pvma) | |
1da177e4 | 223 | { |
5beb4930 RR |
224 | struct anon_vma_chain *avc; |
225 | struct anon_vma *anon_vma; | |
1da177e4 | 226 | |
5beb4930 RR |
227 | /* Don't bother if the parent process has no anon_vma here. */ |
228 | if (!pvma->anon_vma) | |
229 | return 0; | |
230 | ||
231 | /* | |
232 | * First, attach the new VMA to the parent VMA's anon_vmas, | |
233 | * so rmap can find non-COWed pages in child processes. | |
234 | */ | |
235 | if (anon_vma_clone(vma, pvma)) | |
236 | return -ENOMEM; | |
237 | ||
238 | /* Then add our own anon_vma. */ | |
239 | anon_vma = anon_vma_alloc(); | |
240 | if (!anon_vma) | |
241 | goto out_error; | |
242 | avc = anon_vma_chain_alloc(); | |
243 | if (!avc) | |
244 | goto out_error_free_anon_vma; | |
5c341ee1 RR |
245 | |
246 | /* | |
247 | * The root anon_vma's spinlock is the lock actually used when we | |
248 | * lock any of the anon_vmas in this anon_vma tree. | |
249 | */ | |
250 | anon_vma->root = pvma->anon_vma->root; | |
76545066 | 251 | /* |
01d8b20d PZ |
252 | * With refcounts, an anon_vma can stay around longer than the |
253 | * process it belongs to. The root anon_vma needs to be pinned until | |
254 | * this anon_vma is freed, because the lock lives in the root. | |
76545066 RR |
255 | */ |
256 | get_anon_vma(anon_vma->root); | |
5beb4930 RR |
257 | /* Mark this anon_vma as the one where our new (COWed) pages go. */ |
258 | vma->anon_vma = anon_vma; | |
5c341ee1 | 259 | anon_vma_chain_link(vma, avc, anon_vma); |
5beb4930 RR |
260 | |
261 | return 0; | |
262 | ||
263 | out_error_free_anon_vma: | |
01d8b20d | 264 | put_anon_vma(anon_vma); |
5beb4930 | 265 | out_error: |
4946d54c | 266 | unlink_anon_vmas(vma); |
5beb4930 | 267 | return -ENOMEM; |
1da177e4 LT |
268 | } |
269 | ||
5beb4930 | 270 | static void anon_vma_unlink(struct anon_vma_chain *anon_vma_chain) |
1da177e4 | 271 | { |
5beb4930 | 272 | struct anon_vma *anon_vma = anon_vma_chain->anon_vma; |
1da177e4 LT |
273 | int empty; |
274 | ||
5beb4930 | 275 | /* If anon_vma_fork fails, we can get an empty anon_vma_chain. */ |
1da177e4 LT |
276 | if (!anon_vma) |
277 | return; | |
278 | ||
cba48b98 | 279 | anon_vma_lock(anon_vma); |
5beb4930 | 280 | list_del(&anon_vma_chain->same_anon_vma); |
1da177e4 LT |
281 | |
282 | /* We must garbage collect the anon_vma if it's empty */ | |
01d8b20d | 283 | empty = list_empty(&anon_vma->head); |
cba48b98 | 284 | anon_vma_unlock(anon_vma); |
1da177e4 | 285 | |
01d8b20d PZ |
286 | if (empty) |
287 | put_anon_vma(anon_vma); | |
1da177e4 LT |
288 | } |
289 | ||
5beb4930 RR |
290 | void unlink_anon_vmas(struct vm_area_struct *vma) |
291 | { | |
292 | struct anon_vma_chain *avc, *next; | |
293 | ||
5c341ee1 RR |
294 | /* |
295 | * Unlink each anon_vma chained to the VMA. This list is ordered | |
296 | * from newest to oldest, ensuring the root anon_vma gets freed last. | |
297 | */ | |
5beb4930 RR |
298 | list_for_each_entry_safe(avc, next, &vma->anon_vma_chain, same_vma) { |
299 | anon_vma_unlink(avc); | |
300 | list_del(&avc->same_vma); | |
301 | anon_vma_chain_free(avc); | |
302 | } | |
303 | } | |
304 | ||
51cc5068 | 305 | static void anon_vma_ctor(void *data) |
1da177e4 | 306 | { |
a35afb83 | 307 | struct anon_vma *anon_vma = data; |
1da177e4 | 308 | |
a35afb83 | 309 | spin_lock_init(&anon_vma->lock); |
83813267 | 310 | atomic_set(&anon_vma->refcount, 0); |
a35afb83 | 311 | INIT_LIST_HEAD(&anon_vma->head); |
1da177e4 LT |
312 | } |
313 | ||
314 | void __init anon_vma_init(void) | |
315 | { | |
316 | anon_vma_cachep = kmem_cache_create("anon_vma", sizeof(struct anon_vma), | |
20c2df83 | 317 | 0, SLAB_DESTROY_BY_RCU|SLAB_PANIC, anon_vma_ctor); |
5beb4930 | 318 | anon_vma_chain_cachep = KMEM_CACHE(anon_vma_chain, SLAB_PANIC); |
1da177e4 LT |
319 | } |
320 | ||
321 | /* | |
322 | * Getting a lock on a stable anon_vma from a page off the LRU is | |
323 | * tricky: page_lock_anon_vma rely on RCU to guard against the races. | |
324 | */ | |
ea4525b6 | 325 | struct anon_vma *__page_lock_anon_vma(struct page *page) |
1da177e4 | 326 | { |
f1819427 | 327 | struct anon_vma *anon_vma, *root_anon_vma; |
1da177e4 LT |
328 | unsigned long anon_mapping; |
329 | ||
330 | rcu_read_lock(); | |
80e14822 | 331 | anon_mapping = (unsigned long) ACCESS_ONCE(page->mapping); |
3ca7b3c5 | 332 | if ((anon_mapping & PAGE_MAPPING_FLAGS) != PAGE_MAPPING_ANON) |
1da177e4 LT |
333 | goto out; |
334 | if (!page_mapped(page)) | |
335 | goto out; | |
336 | ||
337 | anon_vma = (struct anon_vma *) (anon_mapping - PAGE_MAPPING_ANON); | |
f1819427 HD |
338 | root_anon_vma = ACCESS_ONCE(anon_vma->root); |
339 | spin_lock(&root_anon_vma->lock); | |
340 | ||
341 | /* | |
342 | * If this page is still mapped, then its anon_vma cannot have been | |
343 | * freed. But if it has been unmapped, we have no security against | |
344 | * the anon_vma structure being freed and reused (for another anon_vma: | |
345 | * SLAB_DESTROY_BY_RCU guarantees that - so the spin_lock above cannot | |
346 | * corrupt): with anon_vma_prepare() or anon_vma_fork() redirecting | |
347 | * anon_vma->root before page_unlock_anon_vma() is called to unlock. | |
348 | */ | |
349 | if (page_mapped(page)) | |
350 | return anon_vma; | |
351 | ||
352 | spin_unlock(&root_anon_vma->lock); | |
1da177e4 LT |
353 | out: |
354 | rcu_read_unlock(); | |
34bbd704 ON |
355 | return NULL; |
356 | } | |
357 | ||
10be22df | 358 | void page_unlock_anon_vma(struct anon_vma *anon_vma) |
ea4525b6 NK |
359 | __releases(&anon_vma->root->lock) |
360 | __releases(RCU) | |
34bbd704 | 361 | { |
cba48b98 | 362 | anon_vma_unlock(anon_vma); |
34bbd704 | 363 | rcu_read_unlock(); |
1da177e4 LT |
364 | } |
365 | ||
366 | /* | |
3ad33b24 LS |
367 | * At what user virtual address is page expected in @vma? |
368 | * Returns virtual address or -EFAULT if page's index/offset is not | |
369 | * within the range mapped the @vma. | |
1da177e4 | 370 | */ |
71e3aac0 | 371 | inline unsigned long |
1da177e4 LT |
372 | vma_address(struct page *page, struct vm_area_struct *vma) |
373 | { | |
374 | pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT); | |
375 | unsigned long address; | |
376 | ||
0fe6e20b NH |
377 | if (unlikely(is_vm_hugetlb_page(vma))) |
378 | pgoff = page->index << huge_page_order(page_hstate(page)); | |
1da177e4 LT |
379 | address = vma->vm_start + ((pgoff - vma->vm_pgoff) << PAGE_SHIFT); |
380 | if (unlikely(address < vma->vm_start || address >= vma->vm_end)) { | |
3ad33b24 | 381 | /* page should be within @vma mapping range */ |
1da177e4 LT |
382 | return -EFAULT; |
383 | } | |
384 | return address; | |
385 | } | |
386 | ||
387 | /* | |
bf89c8c8 | 388 | * At what user virtual address is page expected in vma? |
ab941e0f | 389 | * Caller should check the page is actually part of the vma. |
1da177e4 LT |
390 | */ |
391 | unsigned long page_address_in_vma(struct page *page, struct vm_area_struct *vma) | |
392 | { | |
21d0d443 | 393 | if (PageAnon(page)) { |
4829b906 HD |
394 | struct anon_vma *page__anon_vma = page_anon_vma(page); |
395 | /* | |
396 | * Note: swapoff's unuse_vma() is more efficient with this | |
397 | * check, and needs it to match anon_vma when KSM is active. | |
398 | */ | |
399 | if (!vma->anon_vma || !page__anon_vma || | |
400 | vma->anon_vma->root != page__anon_vma->root) | |
21d0d443 AA |
401 | return -EFAULT; |
402 | } else if (page->mapping && !(vma->vm_flags & VM_NONLINEAR)) { | |
ee498ed7 HD |
403 | if (!vma->vm_file || |
404 | vma->vm_file->f_mapping != page->mapping) | |
1da177e4 LT |
405 | return -EFAULT; |
406 | } else | |
407 | return -EFAULT; | |
408 | return vma_address(page, vma); | |
409 | } | |
410 | ||
81b4082d ND |
411 | /* |
412 | * Check that @page is mapped at @address into @mm. | |
413 | * | |
479db0bf NP |
414 | * If @sync is false, page_check_address may perform a racy check to avoid |
415 | * the page table lock when the pte is not present (helpful when reclaiming | |
416 | * highly shared pages). | |
417 | * | |
b8072f09 | 418 | * On success returns with pte mapped and locked. |
81b4082d | 419 | */ |
e9a81a82 | 420 | pte_t *__page_check_address(struct page *page, struct mm_struct *mm, |
479db0bf | 421 | unsigned long address, spinlock_t **ptlp, int sync) |
81b4082d ND |
422 | { |
423 | pgd_t *pgd; | |
424 | pud_t *pud; | |
425 | pmd_t *pmd; | |
426 | pte_t *pte; | |
c0718806 | 427 | spinlock_t *ptl; |
81b4082d | 428 | |
0fe6e20b NH |
429 | if (unlikely(PageHuge(page))) { |
430 | pte = huge_pte_offset(mm, address); | |
431 | ptl = &mm->page_table_lock; | |
432 | goto check; | |
433 | } | |
434 | ||
81b4082d | 435 | pgd = pgd_offset(mm, address); |
c0718806 HD |
436 | if (!pgd_present(*pgd)) |
437 | return NULL; | |
438 | ||
439 | pud = pud_offset(pgd, address); | |
440 | if (!pud_present(*pud)) | |
441 | return NULL; | |
442 | ||
443 | pmd = pmd_offset(pud, address); | |
444 | if (!pmd_present(*pmd)) | |
445 | return NULL; | |
71e3aac0 AA |
446 | if (pmd_trans_huge(*pmd)) |
447 | return NULL; | |
c0718806 HD |
448 | |
449 | pte = pte_offset_map(pmd, address); | |
450 | /* Make a quick check before getting the lock */ | |
479db0bf | 451 | if (!sync && !pte_present(*pte)) { |
c0718806 HD |
452 | pte_unmap(pte); |
453 | return NULL; | |
454 | } | |
455 | ||
4c21e2f2 | 456 | ptl = pte_lockptr(mm, pmd); |
0fe6e20b | 457 | check: |
c0718806 HD |
458 | spin_lock(ptl); |
459 | if (pte_present(*pte) && page_to_pfn(page) == pte_pfn(*pte)) { | |
460 | *ptlp = ptl; | |
461 | return pte; | |
81b4082d | 462 | } |
c0718806 HD |
463 | pte_unmap_unlock(pte, ptl); |
464 | return NULL; | |
81b4082d ND |
465 | } |
466 | ||
b291f000 NP |
467 | /** |
468 | * page_mapped_in_vma - check whether a page is really mapped in a VMA | |
469 | * @page: the page to test | |
470 | * @vma: the VMA to test | |
471 | * | |
472 | * Returns 1 if the page is mapped into the page tables of the VMA, 0 | |
473 | * if the page is not mapped into the page tables of this VMA. Only | |
474 | * valid for normal file or anonymous VMAs. | |
475 | */ | |
6a46079c | 476 | int page_mapped_in_vma(struct page *page, struct vm_area_struct *vma) |
b291f000 NP |
477 | { |
478 | unsigned long address; | |
479 | pte_t *pte; | |
480 | spinlock_t *ptl; | |
481 | ||
482 | address = vma_address(page, vma); | |
483 | if (address == -EFAULT) /* out of vma range */ | |
484 | return 0; | |
485 | pte = page_check_address(page, vma->vm_mm, address, &ptl, 1); | |
486 | if (!pte) /* the page is not in this mm */ | |
487 | return 0; | |
488 | pte_unmap_unlock(pte, ptl); | |
489 | ||
490 | return 1; | |
491 | } | |
492 | ||
1da177e4 LT |
493 | /* |
494 | * Subfunctions of page_referenced: page_referenced_one called | |
495 | * repeatedly from either page_referenced_anon or page_referenced_file. | |
496 | */ | |
5ad64688 HD |
497 | int page_referenced_one(struct page *page, struct vm_area_struct *vma, |
498 | unsigned long address, unsigned int *mapcount, | |
499 | unsigned long *vm_flags) | |
1da177e4 LT |
500 | { |
501 | struct mm_struct *mm = vma->vm_mm; | |
1da177e4 LT |
502 | int referenced = 0; |
503 | ||
71e3aac0 AA |
504 | if (unlikely(PageTransHuge(page))) { |
505 | pmd_t *pmd; | |
506 | ||
507 | spin_lock(&mm->page_table_lock); | |
2da28bfd AA |
508 | /* |
509 | * rmap might return false positives; we must filter | |
510 | * these out using page_check_address_pmd(). | |
511 | */ | |
71e3aac0 AA |
512 | pmd = page_check_address_pmd(page, mm, address, |
513 | PAGE_CHECK_ADDRESS_PMD_FLAG); | |
2da28bfd AA |
514 | if (!pmd) { |
515 | spin_unlock(&mm->page_table_lock); | |
516 | goto out; | |
517 | } | |
518 | ||
519 | if (vma->vm_flags & VM_LOCKED) { | |
520 | spin_unlock(&mm->page_table_lock); | |
521 | *mapcount = 0; /* break early from loop */ | |
522 | *vm_flags |= VM_LOCKED; | |
523 | goto out; | |
524 | } | |
525 | ||
526 | /* go ahead even if the pmd is pmd_trans_splitting() */ | |
527 | if (pmdp_clear_flush_young_notify(vma, address, pmd)) | |
71e3aac0 AA |
528 | referenced++; |
529 | spin_unlock(&mm->page_table_lock); | |
530 | } else { | |
531 | pte_t *pte; | |
532 | spinlock_t *ptl; | |
533 | ||
2da28bfd AA |
534 | /* |
535 | * rmap might return false positives; we must filter | |
536 | * these out using page_check_address(). | |
537 | */ | |
71e3aac0 AA |
538 | pte = page_check_address(page, mm, address, &ptl, 0); |
539 | if (!pte) | |
540 | goto out; | |
541 | ||
2da28bfd AA |
542 | if (vma->vm_flags & VM_LOCKED) { |
543 | pte_unmap_unlock(pte, ptl); | |
544 | *mapcount = 0; /* break early from loop */ | |
545 | *vm_flags |= VM_LOCKED; | |
546 | goto out; | |
547 | } | |
548 | ||
71e3aac0 AA |
549 | if (ptep_clear_flush_young_notify(vma, address, pte)) { |
550 | /* | |
551 | * Don't treat a reference through a sequentially read | |
552 | * mapping as such. If the page has been used in | |
553 | * another mapping, we will catch it; if this other | |
554 | * mapping is already gone, the unmap path will have | |
555 | * set PG_referenced or activated the page. | |
556 | */ | |
557 | if (likely(!VM_SequentialReadHint(vma))) | |
558 | referenced++; | |
559 | } | |
560 | pte_unmap_unlock(pte, ptl); | |
561 | } | |
562 | ||
2da28bfd AA |
563 | /* Pretend the page is referenced if the task has the |
564 | swap token and is in the middle of a page fault. */ | |
565 | if (mm != current->mm && has_swap_token(mm) && | |
566 | rwsem_is_locked(&mm->mmap_sem)) | |
567 | referenced++; | |
568 | ||
c0718806 | 569 | (*mapcount)--; |
273f047e | 570 | |
6fe6b7e3 WF |
571 | if (referenced) |
572 | *vm_flags |= vma->vm_flags; | |
273f047e | 573 | out: |
1da177e4 LT |
574 | return referenced; |
575 | } | |
576 | ||
bed7161a | 577 | static int page_referenced_anon(struct page *page, |
6fe6b7e3 WF |
578 | struct mem_cgroup *mem_cont, |
579 | unsigned long *vm_flags) | |
1da177e4 LT |
580 | { |
581 | unsigned int mapcount; | |
582 | struct anon_vma *anon_vma; | |
5beb4930 | 583 | struct anon_vma_chain *avc; |
1da177e4 LT |
584 | int referenced = 0; |
585 | ||
586 | anon_vma = page_lock_anon_vma(page); | |
587 | if (!anon_vma) | |
588 | return referenced; | |
589 | ||
590 | mapcount = page_mapcount(page); | |
5beb4930 RR |
591 | list_for_each_entry(avc, &anon_vma->head, same_anon_vma) { |
592 | struct vm_area_struct *vma = avc->vma; | |
1cb1729b HD |
593 | unsigned long address = vma_address(page, vma); |
594 | if (address == -EFAULT) | |
595 | continue; | |
bed7161a BS |
596 | /* |
597 | * If we are reclaiming on behalf of a cgroup, skip | |
598 | * counting on behalf of references from different | |
599 | * cgroups | |
600 | */ | |
bd845e38 | 601 | if (mem_cont && !mm_match_cgroup(vma->vm_mm, mem_cont)) |
bed7161a | 602 | continue; |
1cb1729b | 603 | referenced += page_referenced_one(page, vma, address, |
6fe6b7e3 | 604 | &mapcount, vm_flags); |
1da177e4 LT |
605 | if (!mapcount) |
606 | break; | |
607 | } | |
34bbd704 ON |
608 | |
609 | page_unlock_anon_vma(anon_vma); | |
1da177e4 LT |
610 | return referenced; |
611 | } | |
612 | ||
613 | /** | |
614 | * page_referenced_file - referenced check for object-based rmap | |
615 | * @page: the page we're checking references on. | |
43d8eac4 | 616 | * @mem_cont: target memory controller |
6fe6b7e3 | 617 | * @vm_flags: collect encountered vma->vm_flags who actually referenced the page |
1da177e4 LT |
618 | * |
619 | * For an object-based mapped page, find all the places it is mapped and | |
620 | * check/clear the referenced flag. This is done by following the page->mapping | |
621 | * pointer, then walking the chain of vmas it holds. It returns the number | |
622 | * of references it found. | |
623 | * | |
624 | * This function is only called from page_referenced for object-based pages. | |
625 | */ | |
bed7161a | 626 | static int page_referenced_file(struct page *page, |
6fe6b7e3 WF |
627 | struct mem_cgroup *mem_cont, |
628 | unsigned long *vm_flags) | |
1da177e4 LT |
629 | { |
630 | unsigned int mapcount; | |
631 | struct address_space *mapping = page->mapping; | |
632 | pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT); | |
633 | struct vm_area_struct *vma; | |
634 | struct prio_tree_iter iter; | |
635 | int referenced = 0; | |
636 | ||
637 | /* | |
638 | * The caller's checks on page->mapping and !PageAnon have made | |
639 | * sure that this is a file page: the check for page->mapping | |
640 | * excludes the case just before it gets set on an anon page. | |
641 | */ | |
642 | BUG_ON(PageAnon(page)); | |
643 | ||
644 | /* | |
645 | * The page lock not only makes sure that page->mapping cannot | |
646 | * suddenly be NULLified by truncation, it makes sure that the | |
647 | * structure at mapping cannot be freed and reused yet, | |
648 | * so we can safely take mapping->i_mmap_lock. | |
649 | */ | |
650 | BUG_ON(!PageLocked(page)); | |
651 | ||
652 | spin_lock(&mapping->i_mmap_lock); | |
653 | ||
654 | /* | |
655 | * i_mmap_lock does not stabilize mapcount at all, but mapcount | |
656 | * is more likely to be accurate if we note it after spinning. | |
657 | */ | |
658 | mapcount = page_mapcount(page); | |
659 | ||
660 | vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, pgoff, pgoff) { | |
1cb1729b HD |
661 | unsigned long address = vma_address(page, vma); |
662 | if (address == -EFAULT) | |
663 | continue; | |
bed7161a BS |
664 | /* |
665 | * If we are reclaiming on behalf of a cgroup, skip | |
666 | * counting on behalf of references from different | |
667 | * cgroups | |
668 | */ | |
bd845e38 | 669 | if (mem_cont && !mm_match_cgroup(vma->vm_mm, mem_cont)) |
bed7161a | 670 | continue; |
1cb1729b | 671 | referenced += page_referenced_one(page, vma, address, |
6fe6b7e3 | 672 | &mapcount, vm_flags); |
1da177e4 LT |
673 | if (!mapcount) |
674 | break; | |
675 | } | |
676 | ||
677 | spin_unlock(&mapping->i_mmap_lock); | |
678 | return referenced; | |
679 | } | |
680 | ||
681 | /** | |
682 | * page_referenced - test if the page was referenced | |
683 | * @page: the page to test | |
684 | * @is_locked: caller holds lock on the page | |
43d8eac4 | 685 | * @mem_cont: target memory controller |
6fe6b7e3 | 686 | * @vm_flags: collect encountered vma->vm_flags who actually referenced the page |
1da177e4 LT |
687 | * |
688 | * Quick test_and_clear_referenced for all mappings to a page, | |
689 | * returns the number of ptes which referenced the page. | |
690 | */ | |
6fe6b7e3 WF |
691 | int page_referenced(struct page *page, |
692 | int is_locked, | |
693 | struct mem_cgroup *mem_cont, | |
694 | unsigned long *vm_flags) | |
1da177e4 LT |
695 | { |
696 | int referenced = 0; | |
5ad64688 | 697 | int we_locked = 0; |
1da177e4 | 698 | |
6fe6b7e3 | 699 | *vm_flags = 0; |
3ca7b3c5 | 700 | if (page_mapped(page) && page_rmapping(page)) { |
5ad64688 HD |
701 | if (!is_locked && (!PageAnon(page) || PageKsm(page))) { |
702 | we_locked = trylock_page(page); | |
703 | if (!we_locked) { | |
704 | referenced++; | |
705 | goto out; | |
706 | } | |
707 | } | |
708 | if (unlikely(PageKsm(page))) | |
709 | referenced += page_referenced_ksm(page, mem_cont, | |
710 | vm_flags); | |
711 | else if (PageAnon(page)) | |
6fe6b7e3 WF |
712 | referenced += page_referenced_anon(page, mem_cont, |
713 | vm_flags); | |
5ad64688 | 714 | else if (page->mapping) |
6fe6b7e3 WF |
715 | referenced += page_referenced_file(page, mem_cont, |
716 | vm_flags); | |
5ad64688 | 717 | if (we_locked) |
1da177e4 | 718 | unlock_page(page); |
1da177e4 | 719 | } |
5ad64688 | 720 | out: |
5b7baf05 CB |
721 | if (page_test_and_clear_young(page)) |
722 | referenced++; | |
723 | ||
1da177e4 LT |
724 | return referenced; |
725 | } | |
726 | ||
1cb1729b HD |
727 | static int page_mkclean_one(struct page *page, struct vm_area_struct *vma, |
728 | unsigned long address) | |
d08b3851 PZ |
729 | { |
730 | struct mm_struct *mm = vma->vm_mm; | |
c2fda5fe | 731 | pte_t *pte; |
d08b3851 PZ |
732 | spinlock_t *ptl; |
733 | int ret = 0; | |
734 | ||
479db0bf | 735 | pte = page_check_address(page, mm, address, &ptl, 1); |
d08b3851 PZ |
736 | if (!pte) |
737 | goto out; | |
738 | ||
c2fda5fe PZ |
739 | if (pte_dirty(*pte) || pte_write(*pte)) { |
740 | pte_t entry; | |
d08b3851 | 741 | |
c2fda5fe | 742 | flush_cache_page(vma, address, pte_pfn(*pte)); |
cddb8a5c | 743 | entry = ptep_clear_flush_notify(vma, address, pte); |
c2fda5fe PZ |
744 | entry = pte_wrprotect(entry); |
745 | entry = pte_mkclean(entry); | |
d6e88e67 | 746 | set_pte_at(mm, address, pte, entry); |
c2fda5fe PZ |
747 | ret = 1; |
748 | } | |
d08b3851 | 749 | |
d08b3851 PZ |
750 | pte_unmap_unlock(pte, ptl); |
751 | out: | |
752 | return ret; | |
753 | } | |
754 | ||
755 | static int page_mkclean_file(struct address_space *mapping, struct page *page) | |
756 | { | |
757 | pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT); | |
758 | struct vm_area_struct *vma; | |
759 | struct prio_tree_iter iter; | |
760 | int ret = 0; | |
761 | ||
762 | BUG_ON(PageAnon(page)); | |
763 | ||
764 | spin_lock(&mapping->i_mmap_lock); | |
765 | vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, pgoff, pgoff) { | |
1cb1729b HD |
766 | if (vma->vm_flags & VM_SHARED) { |
767 | unsigned long address = vma_address(page, vma); | |
768 | if (address == -EFAULT) | |
769 | continue; | |
770 | ret += page_mkclean_one(page, vma, address); | |
771 | } | |
d08b3851 PZ |
772 | } |
773 | spin_unlock(&mapping->i_mmap_lock); | |
774 | return ret; | |
775 | } | |
776 | ||
777 | int page_mkclean(struct page *page) | |
778 | { | |
779 | int ret = 0; | |
780 | ||
781 | BUG_ON(!PageLocked(page)); | |
782 | ||
783 | if (page_mapped(page)) { | |
784 | struct address_space *mapping = page_mapping(page); | |
ce7e9fae | 785 | if (mapping) { |
d08b3851 | 786 | ret = page_mkclean_file(mapping, page); |
ce7e9fae | 787 | if (page_test_dirty(page)) { |
e2b8d7af | 788 | page_clear_dirty(page, 1); |
ce7e9fae CB |
789 | ret = 1; |
790 | } | |
6c210482 | 791 | } |
d08b3851 PZ |
792 | } |
793 | ||
794 | return ret; | |
795 | } | |
60b59bea | 796 | EXPORT_SYMBOL_GPL(page_mkclean); |
d08b3851 | 797 | |
c44b6743 RR |
798 | /** |
799 | * page_move_anon_rmap - move a page to our anon_vma | |
800 | * @page: the page to move to our anon_vma | |
801 | * @vma: the vma the page belongs to | |
802 | * @address: the user virtual address mapped | |
803 | * | |
804 | * When a page belongs exclusively to one process after a COW event, | |
805 | * that page can be moved into the anon_vma that belongs to just that | |
806 | * process, so the rmap code will not search the parent or sibling | |
807 | * processes. | |
808 | */ | |
809 | void page_move_anon_rmap(struct page *page, | |
810 | struct vm_area_struct *vma, unsigned long address) | |
811 | { | |
812 | struct anon_vma *anon_vma = vma->anon_vma; | |
813 | ||
814 | VM_BUG_ON(!PageLocked(page)); | |
815 | VM_BUG_ON(!anon_vma); | |
816 | VM_BUG_ON(page->index != linear_page_index(vma, address)); | |
817 | ||
818 | anon_vma = (void *) anon_vma + PAGE_MAPPING_ANON; | |
819 | page->mapping = (struct address_space *) anon_vma; | |
820 | } | |
821 | ||
9617d95e | 822 | /** |
4e1c1975 AK |
823 | * __page_set_anon_rmap - set up new anonymous rmap |
824 | * @page: Page to add to rmap | |
825 | * @vma: VM area to add page to. | |
826 | * @address: User virtual address of the mapping | |
e8a03feb | 827 | * @exclusive: the page is exclusively owned by the current process |
9617d95e NP |
828 | */ |
829 | static void __page_set_anon_rmap(struct page *page, | |
e8a03feb | 830 | struct vm_area_struct *vma, unsigned long address, int exclusive) |
9617d95e | 831 | { |
e8a03feb | 832 | struct anon_vma *anon_vma = vma->anon_vma; |
ea90002b | 833 | |
e8a03feb | 834 | BUG_ON(!anon_vma); |
ea90002b | 835 | |
4e1c1975 AK |
836 | if (PageAnon(page)) |
837 | return; | |
838 | ||
ea90002b | 839 | /* |
e8a03feb RR |
840 | * If the page isn't exclusively mapped into this vma, |
841 | * we must use the _oldest_ possible anon_vma for the | |
842 | * page mapping! | |
ea90002b | 843 | */ |
4e1c1975 | 844 | if (!exclusive) |
288468c3 | 845 | anon_vma = anon_vma->root; |
9617d95e | 846 | |
9617d95e NP |
847 | anon_vma = (void *) anon_vma + PAGE_MAPPING_ANON; |
848 | page->mapping = (struct address_space *) anon_vma; | |
9617d95e | 849 | page->index = linear_page_index(vma, address); |
9617d95e NP |
850 | } |
851 | ||
c97a9e10 | 852 | /** |
43d8eac4 | 853 | * __page_check_anon_rmap - sanity check anonymous rmap addition |
c97a9e10 NP |
854 | * @page: the page to add the mapping to |
855 | * @vma: the vm area in which the mapping is added | |
856 | * @address: the user virtual address mapped | |
857 | */ | |
858 | static void __page_check_anon_rmap(struct page *page, | |
859 | struct vm_area_struct *vma, unsigned long address) | |
860 | { | |
861 | #ifdef CONFIG_DEBUG_VM | |
862 | /* | |
863 | * The page's anon-rmap details (mapping and index) are guaranteed to | |
864 | * be set up correctly at this point. | |
865 | * | |
866 | * We have exclusion against page_add_anon_rmap because the caller | |
867 | * always holds the page locked, except if called from page_dup_rmap, | |
868 | * in which case the page is already known to be setup. | |
869 | * | |
870 | * We have exclusion against page_add_new_anon_rmap because those pages | |
871 | * are initially only visible via the pagetables, and the pte is locked | |
872 | * over the call to page_add_new_anon_rmap. | |
873 | */ | |
44ab57a0 | 874 | BUG_ON(page_anon_vma(page)->root != vma->anon_vma->root); |
c97a9e10 NP |
875 | BUG_ON(page->index != linear_page_index(vma, address)); |
876 | #endif | |
877 | } | |
878 | ||
1da177e4 LT |
879 | /** |
880 | * page_add_anon_rmap - add pte mapping to an anonymous page | |
881 | * @page: the page to add the mapping to | |
882 | * @vma: the vm area in which the mapping is added | |
883 | * @address: the user virtual address mapped | |
884 | * | |
5ad64688 | 885 | * The caller needs to hold the pte lock, and the page must be locked in |
80e14822 HD |
886 | * the anon_vma case: to serialize mapping,index checking after setting, |
887 | * and to ensure that PageAnon is not being upgraded racily to PageKsm | |
888 | * (but PageKsm is never downgraded to PageAnon). | |
1da177e4 LT |
889 | */ |
890 | void page_add_anon_rmap(struct page *page, | |
891 | struct vm_area_struct *vma, unsigned long address) | |
ad8c2ee8 RR |
892 | { |
893 | do_page_add_anon_rmap(page, vma, address, 0); | |
894 | } | |
895 | ||
896 | /* | |
897 | * Special version of the above for do_swap_page, which often runs | |
898 | * into pages that are exclusively owned by the current process. | |
899 | * Everybody else should continue to use page_add_anon_rmap above. | |
900 | */ | |
901 | void do_page_add_anon_rmap(struct page *page, | |
902 | struct vm_area_struct *vma, unsigned long address, int exclusive) | |
1da177e4 | 903 | { |
5ad64688 | 904 | int first = atomic_inc_and_test(&page->_mapcount); |
79134171 AA |
905 | if (first) { |
906 | if (!PageTransHuge(page)) | |
907 | __inc_zone_page_state(page, NR_ANON_PAGES); | |
908 | else | |
909 | __inc_zone_page_state(page, | |
910 | NR_ANON_TRANSPARENT_HUGEPAGES); | |
911 | } | |
5ad64688 HD |
912 | if (unlikely(PageKsm(page))) |
913 | return; | |
914 | ||
c97a9e10 NP |
915 | VM_BUG_ON(!PageLocked(page)); |
916 | VM_BUG_ON(address < vma->vm_start || address >= vma->vm_end); | |
5ad64688 | 917 | if (first) |
ad8c2ee8 | 918 | __page_set_anon_rmap(page, vma, address, exclusive); |
69029cd5 | 919 | else |
c97a9e10 | 920 | __page_check_anon_rmap(page, vma, address); |
1da177e4 LT |
921 | } |
922 | ||
43d8eac4 | 923 | /** |
9617d95e NP |
924 | * page_add_new_anon_rmap - add pte mapping to a new anonymous page |
925 | * @page: the page to add the mapping to | |
926 | * @vma: the vm area in which the mapping is added | |
927 | * @address: the user virtual address mapped | |
928 | * | |
929 | * Same as page_add_anon_rmap but must only be called on *new* pages. | |
930 | * This means the inc-and-test can be bypassed. | |
c97a9e10 | 931 | * Page does not have to be locked. |
9617d95e NP |
932 | */ |
933 | void page_add_new_anon_rmap(struct page *page, | |
934 | struct vm_area_struct *vma, unsigned long address) | |
935 | { | |
b5934c53 | 936 | VM_BUG_ON(address < vma->vm_start || address >= vma->vm_end); |
cbf84b7a HD |
937 | SetPageSwapBacked(page); |
938 | atomic_set(&page->_mapcount, 0); /* increment count (starts at -1) */ | |
79134171 AA |
939 | if (!PageTransHuge(page)) |
940 | __inc_zone_page_state(page, NR_ANON_PAGES); | |
941 | else | |
942 | __inc_zone_page_state(page, NR_ANON_TRANSPARENT_HUGEPAGES); | |
e8a03feb | 943 | __page_set_anon_rmap(page, vma, address, 1); |
b5934c53 | 944 | if (page_evictable(page, vma)) |
cbf84b7a | 945 | lru_cache_add_lru(page, LRU_ACTIVE_ANON); |
b5934c53 HD |
946 | else |
947 | add_page_to_unevictable_list(page); | |
9617d95e NP |
948 | } |
949 | ||
1da177e4 LT |
950 | /** |
951 | * page_add_file_rmap - add pte mapping to a file page | |
952 | * @page: the page to add the mapping to | |
953 | * | |
b8072f09 | 954 | * The caller needs to hold the pte lock. |
1da177e4 LT |
955 | */ |
956 | void page_add_file_rmap(struct page *page) | |
957 | { | |
d69b042f | 958 | if (atomic_inc_and_test(&page->_mapcount)) { |
65ba55f5 | 959 | __inc_zone_page_state(page, NR_FILE_MAPPED); |
2a7106f2 | 960 | mem_cgroup_inc_page_stat(page, MEMCG_NR_FILE_MAPPED); |
d69b042f | 961 | } |
1da177e4 LT |
962 | } |
963 | ||
964 | /** | |
965 | * page_remove_rmap - take down pte mapping from a page | |
966 | * @page: page to remove mapping from | |
967 | * | |
b8072f09 | 968 | * The caller needs to hold the pte lock. |
1da177e4 | 969 | */ |
edc315fd | 970 | void page_remove_rmap(struct page *page) |
1da177e4 | 971 | { |
b904dcfe KM |
972 | /* page still mapped by someone else? */ |
973 | if (!atomic_add_negative(-1, &page->_mapcount)) | |
974 | return; | |
975 | ||
976 | /* | |
977 | * Now that the last pte has gone, s390 must transfer dirty | |
978 | * flag from storage key to struct page. We can usually skip | |
979 | * this if the page is anon, so about to be freed; but perhaps | |
980 | * not if it's in swapcache - there might be another pte slot | |
981 | * containing the swap entry, but page not yet written to swap. | |
982 | */ | |
983 | if ((!PageAnon(page) || PageSwapCache(page)) && page_test_dirty(page)) { | |
e2b8d7af | 984 | page_clear_dirty(page, 1); |
b904dcfe | 985 | set_page_dirty(page); |
1da177e4 | 986 | } |
0fe6e20b NH |
987 | /* |
988 | * Hugepages are not counted in NR_ANON_PAGES nor NR_FILE_MAPPED | |
989 | * and not charged by memcg for now. | |
990 | */ | |
991 | if (unlikely(PageHuge(page))) | |
992 | return; | |
b904dcfe KM |
993 | if (PageAnon(page)) { |
994 | mem_cgroup_uncharge_page(page); | |
79134171 AA |
995 | if (!PageTransHuge(page)) |
996 | __dec_zone_page_state(page, NR_ANON_PAGES); | |
997 | else | |
998 | __dec_zone_page_state(page, | |
999 | NR_ANON_TRANSPARENT_HUGEPAGES); | |
b904dcfe KM |
1000 | } else { |
1001 | __dec_zone_page_state(page, NR_FILE_MAPPED); | |
2a7106f2 | 1002 | mem_cgroup_dec_page_stat(page, MEMCG_NR_FILE_MAPPED); |
b904dcfe | 1003 | } |
b904dcfe KM |
1004 | /* |
1005 | * It would be tidy to reset the PageAnon mapping here, | |
1006 | * but that might overwrite a racing page_add_anon_rmap | |
1007 | * which increments mapcount after us but sets mapping | |
1008 | * before us: so leave the reset to free_hot_cold_page, | |
1009 | * and remember that it's only reliable while mapped. | |
1010 | * Leaving it set also helps swapoff to reinstate ptes | |
1011 | * faster for those pages still in swapcache. | |
1012 | */ | |
1da177e4 LT |
1013 | } |
1014 | ||
1015 | /* | |
1016 | * Subfunctions of try_to_unmap: try_to_unmap_one called | |
1017 | * repeatedly from either try_to_unmap_anon or try_to_unmap_file. | |
1018 | */ | |
5ad64688 HD |
1019 | int try_to_unmap_one(struct page *page, struct vm_area_struct *vma, |
1020 | unsigned long address, enum ttu_flags flags) | |
1da177e4 LT |
1021 | { |
1022 | struct mm_struct *mm = vma->vm_mm; | |
1da177e4 LT |
1023 | pte_t *pte; |
1024 | pte_t pteval; | |
c0718806 | 1025 | spinlock_t *ptl; |
1da177e4 LT |
1026 | int ret = SWAP_AGAIN; |
1027 | ||
479db0bf | 1028 | pte = page_check_address(page, mm, address, &ptl, 0); |
c0718806 | 1029 | if (!pte) |
81b4082d | 1030 | goto out; |
1da177e4 LT |
1031 | |
1032 | /* | |
1033 | * If the page is mlock()d, we cannot swap it out. | |
1034 | * If it's recently referenced (perhaps page_referenced | |
1035 | * skipped over this mm) then we should reactivate it. | |
1036 | */ | |
14fa31b8 | 1037 | if (!(flags & TTU_IGNORE_MLOCK)) { |
caed0f48 KM |
1038 | if (vma->vm_flags & VM_LOCKED) |
1039 | goto out_mlock; | |
1040 | ||
af8e3354 | 1041 | if (TTU_ACTION(flags) == TTU_MUNLOCK) |
53f79acb | 1042 | goto out_unmap; |
14fa31b8 AK |
1043 | } |
1044 | if (!(flags & TTU_IGNORE_ACCESS)) { | |
b291f000 NP |
1045 | if (ptep_clear_flush_young_notify(vma, address, pte)) { |
1046 | ret = SWAP_FAIL; | |
1047 | goto out_unmap; | |
1048 | } | |
1049 | } | |
1da177e4 | 1050 | |
1da177e4 LT |
1051 | /* Nuke the page table entry. */ |
1052 | flush_cache_page(vma, address, page_to_pfn(page)); | |
cddb8a5c | 1053 | pteval = ptep_clear_flush_notify(vma, address, pte); |
1da177e4 LT |
1054 | |
1055 | /* Move the dirty bit to the physical page now the pte is gone. */ | |
1056 | if (pte_dirty(pteval)) | |
1057 | set_page_dirty(page); | |
1058 | ||
365e9c87 HD |
1059 | /* Update high watermark before we lower rss */ |
1060 | update_hiwater_rss(mm); | |
1061 | ||
888b9f7c AK |
1062 | if (PageHWPoison(page) && !(flags & TTU_IGNORE_HWPOISON)) { |
1063 | if (PageAnon(page)) | |
d559db08 | 1064 | dec_mm_counter(mm, MM_ANONPAGES); |
888b9f7c | 1065 | else |
d559db08 | 1066 | dec_mm_counter(mm, MM_FILEPAGES); |
888b9f7c AK |
1067 | set_pte_at(mm, address, pte, |
1068 | swp_entry_to_pte(make_hwpoison_entry(page))); | |
1069 | } else if (PageAnon(page)) { | |
4c21e2f2 | 1070 | swp_entry_t entry = { .val = page_private(page) }; |
0697212a CL |
1071 | |
1072 | if (PageSwapCache(page)) { | |
1073 | /* | |
1074 | * Store the swap location in the pte. | |
1075 | * See handle_pte_fault() ... | |
1076 | */ | |
570a335b HD |
1077 | if (swap_duplicate(entry) < 0) { |
1078 | set_pte_at(mm, address, pte, pteval); | |
1079 | ret = SWAP_FAIL; | |
1080 | goto out_unmap; | |
1081 | } | |
0697212a CL |
1082 | if (list_empty(&mm->mmlist)) { |
1083 | spin_lock(&mmlist_lock); | |
1084 | if (list_empty(&mm->mmlist)) | |
1085 | list_add(&mm->mmlist, &init_mm.mmlist); | |
1086 | spin_unlock(&mmlist_lock); | |
1087 | } | |
d559db08 | 1088 | dec_mm_counter(mm, MM_ANONPAGES); |
b084d435 | 1089 | inc_mm_counter(mm, MM_SWAPENTS); |
64cdd548 | 1090 | } else if (PAGE_MIGRATION) { |
0697212a CL |
1091 | /* |
1092 | * Store the pfn of the page in a special migration | |
1093 | * pte. do_swap_page() will wait until the migration | |
1094 | * pte is removed and then restart fault handling. | |
1095 | */ | |
14fa31b8 | 1096 | BUG_ON(TTU_ACTION(flags) != TTU_MIGRATION); |
0697212a | 1097 | entry = make_migration_entry(page, pte_write(pteval)); |
1da177e4 LT |
1098 | } |
1099 | set_pte_at(mm, address, pte, swp_entry_to_pte(entry)); | |
1100 | BUG_ON(pte_file(*pte)); | |
14fa31b8 | 1101 | } else if (PAGE_MIGRATION && (TTU_ACTION(flags) == TTU_MIGRATION)) { |
04e62a29 CL |
1102 | /* Establish migration entry for a file page */ |
1103 | swp_entry_t entry; | |
1104 | entry = make_migration_entry(page, pte_write(pteval)); | |
1105 | set_pte_at(mm, address, pte, swp_entry_to_pte(entry)); | |
1106 | } else | |
d559db08 | 1107 | dec_mm_counter(mm, MM_FILEPAGES); |
1da177e4 | 1108 | |
edc315fd | 1109 | page_remove_rmap(page); |
1da177e4 LT |
1110 | page_cache_release(page); |
1111 | ||
1112 | out_unmap: | |
c0718806 | 1113 | pte_unmap_unlock(pte, ptl); |
caed0f48 KM |
1114 | out: |
1115 | return ret; | |
53f79acb | 1116 | |
caed0f48 KM |
1117 | out_mlock: |
1118 | pte_unmap_unlock(pte, ptl); | |
1119 | ||
1120 | ||
1121 | /* | |
1122 | * We need mmap_sem locking, Otherwise VM_LOCKED check makes | |
1123 | * unstable result and race. Plus, We can't wait here because | |
1124 | * we now hold anon_vma->lock or mapping->i_mmap_lock. | |
1125 | * if trylock failed, the page remain in evictable lru and later | |
1126 | * vmscan could retry to move the page to unevictable lru if the | |
1127 | * page is actually mlocked. | |
1128 | */ | |
1129 | if (down_read_trylock(&vma->vm_mm->mmap_sem)) { | |
1130 | if (vma->vm_flags & VM_LOCKED) { | |
1131 | mlock_vma_page(page); | |
1132 | ret = SWAP_MLOCK; | |
53f79acb | 1133 | } |
caed0f48 | 1134 | up_read(&vma->vm_mm->mmap_sem); |
53f79acb | 1135 | } |
1da177e4 LT |
1136 | return ret; |
1137 | } | |
1138 | ||
1139 | /* | |
1140 | * objrmap doesn't work for nonlinear VMAs because the assumption that | |
1141 | * offset-into-file correlates with offset-into-virtual-addresses does not hold. | |
1142 | * Consequently, given a particular page and its ->index, we cannot locate the | |
1143 | * ptes which are mapping that page without an exhaustive linear search. | |
1144 | * | |
1145 | * So what this code does is a mini "virtual scan" of each nonlinear VMA which | |
1146 | * maps the file to which the target page belongs. The ->vm_private_data field | |
1147 | * holds the current cursor into that scan. Successive searches will circulate | |
1148 | * around the vma's virtual address space. | |
1149 | * | |
1150 | * So as more replacement pressure is applied to the pages in a nonlinear VMA, | |
1151 | * more scanning pressure is placed against them as well. Eventually pages | |
1152 | * will become fully unmapped and are eligible for eviction. | |
1153 | * | |
1154 | * For very sparsely populated VMAs this is a little inefficient - chances are | |
1155 | * there there won't be many ptes located within the scan cluster. In this case | |
1156 | * maybe we could scan further - to the end of the pte page, perhaps. | |
b291f000 NP |
1157 | * |
1158 | * Mlocked pages: check VM_LOCKED under mmap_sem held for read, if we can | |
1159 | * acquire it without blocking. If vma locked, mlock the pages in the cluster, | |
1160 | * rather than unmapping them. If we encounter the "check_page" that vmscan is | |
1161 | * trying to unmap, return SWAP_MLOCK, else default SWAP_AGAIN. | |
1da177e4 LT |
1162 | */ |
1163 | #define CLUSTER_SIZE min(32*PAGE_SIZE, PMD_SIZE) | |
1164 | #define CLUSTER_MASK (~(CLUSTER_SIZE - 1)) | |
1165 | ||
b291f000 NP |
1166 | static int try_to_unmap_cluster(unsigned long cursor, unsigned int *mapcount, |
1167 | struct vm_area_struct *vma, struct page *check_page) | |
1da177e4 LT |
1168 | { |
1169 | struct mm_struct *mm = vma->vm_mm; | |
1170 | pgd_t *pgd; | |
1171 | pud_t *pud; | |
1172 | pmd_t *pmd; | |
c0718806 | 1173 | pte_t *pte; |
1da177e4 | 1174 | pte_t pteval; |
c0718806 | 1175 | spinlock_t *ptl; |
1da177e4 LT |
1176 | struct page *page; |
1177 | unsigned long address; | |
1178 | unsigned long end; | |
b291f000 NP |
1179 | int ret = SWAP_AGAIN; |
1180 | int locked_vma = 0; | |
1da177e4 | 1181 | |
1da177e4 LT |
1182 | address = (vma->vm_start + cursor) & CLUSTER_MASK; |
1183 | end = address + CLUSTER_SIZE; | |
1184 | if (address < vma->vm_start) | |
1185 | address = vma->vm_start; | |
1186 | if (end > vma->vm_end) | |
1187 | end = vma->vm_end; | |
1188 | ||
1189 | pgd = pgd_offset(mm, address); | |
1190 | if (!pgd_present(*pgd)) | |
b291f000 | 1191 | return ret; |
1da177e4 LT |
1192 | |
1193 | pud = pud_offset(pgd, address); | |
1194 | if (!pud_present(*pud)) | |
b291f000 | 1195 | return ret; |
1da177e4 LT |
1196 | |
1197 | pmd = pmd_offset(pud, address); | |
1198 | if (!pmd_present(*pmd)) | |
b291f000 NP |
1199 | return ret; |
1200 | ||
1201 | /* | |
af8e3354 | 1202 | * If we can acquire the mmap_sem for read, and vma is VM_LOCKED, |
b291f000 NP |
1203 | * keep the sem while scanning the cluster for mlocking pages. |
1204 | */ | |
af8e3354 | 1205 | if (down_read_trylock(&vma->vm_mm->mmap_sem)) { |
b291f000 NP |
1206 | locked_vma = (vma->vm_flags & VM_LOCKED); |
1207 | if (!locked_vma) | |
1208 | up_read(&vma->vm_mm->mmap_sem); /* don't need it */ | |
1209 | } | |
c0718806 HD |
1210 | |
1211 | pte = pte_offset_map_lock(mm, pmd, address, &ptl); | |
1da177e4 | 1212 | |
365e9c87 HD |
1213 | /* Update high watermark before we lower rss */ |
1214 | update_hiwater_rss(mm); | |
1215 | ||
c0718806 | 1216 | for (; address < end; pte++, address += PAGE_SIZE) { |
1da177e4 LT |
1217 | if (!pte_present(*pte)) |
1218 | continue; | |
6aab341e LT |
1219 | page = vm_normal_page(vma, address, *pte); |
1220 | BUG_ON(!page || PageAnon(page)); | |
1da177e4 | 1221 | |
b291f000 NP |
1222 | if (locked_vma) { |
1223 | mlock_vma_page(page); /* no-op if already mlocked */ | |
1224 | if (page == check_page) | |
1225 | ret = SWAP_MLOCK; | |
1226 | continue; /* don't unmap */ | |
1227 | } | |
1228 | ||
cddb8a5c | 1229 | if (ptep_clear_flush_young_notify(vma, address, pte)) |
1da177e4 LT |
1230 | continue; |
1231 | ||
1232 | /* Nuke the page table entry. */ | |
eca35133 | 1233 | flush_cache_page(vma, address, pte_pfn(*pte)); |
cddb8a5c | 1234 | pteval = ptep_clear_flush_notify(vma, address, pte); |
1da177e4 LT |
1235 | |
1236 | /* If nonlinear, store the file page offset in the pte. */ | |
1237 | if (page->index != linear_page_index(vma, address)) | |
1238 | set_pte_at(mm, address, pte, pgoff_to_pte(page->index)); | |
1239 | ||
1240 | /* Move the dirty bit to the physical page now the pte is gone. */ | |
1241 | if (pte_dirty(pteval)) | |
1242 | set_page_dirty(page); | |
1243 | ||
edc315fd | 1244 | page_remove_rmap(page); |
1da177e4 | 1245 | page_cache_release(page); |
d559db08 | 1246 | dec_mm_counter(mm, MM_FILEPAGES); |
1da177e4 LT |
1247 | (*mapcount)--; |
1248 | } | |
c0718806 | 1249 | pte_unmap_unlock(pte - 1, ptl); |
b291f000 NP |
1250 | if (locked_vma) |
1251 | up_read(&vma->vm_mm->mmap_sem); | |
1252 | return ret; | |
1da177e4 LT |
1253 | } |
1254 | ||
71e3aac0 | 1255 | bool is_vma_temporary_stack(struct vm_area_struct *vma) |
a8bef8ff MG |
1256 | { |
1257 | int maybe_stack = vma->vm_flags & (VM_GROWSDOWN | VM_GROWSUP); | |
1258 | ||
1259 | if (!maybe_stack) | |
1260 | return false; | |
1261 | ||
1262 | if ((vma->vm_flags & VM_STACK_INCOMPLETE_SETUP) == | |
1263 | VM_STACK_INCOMPLETE_SETUP) | |
1264 | return true; | |
1265 | ||
1266 | return false; | |
1267 | } | |
1268 | ||
b291f000 NP |
1269 | /** |
1270 | * try_to_unmap_anon - unmap or unlock anonymous page using the object-based | |
1271 | * rmap method | |
1272 | * @page: the page to unmap/unlock | |
8051be5e | 1273 | * @flags: action and flags |
b291f000 NP |
1274 | * |
1275 | * Find all the mappings of a page using the mapping pointer and the vma chains | |
1276 | * contained in the anon_vma struct it points to. | |
1277 | * | |
1278 | * This function is only called from try_to_unmap/try_to_munlock for | |
1279 | * anonymous pages. | |
1280 | * When called from try_to_munlock(), the mmap_sem of the mm containing the vma | |
1281 | * where the page was found will be held for write. So, we won't recheck | |
1282 | * vm_flags for that VMA. That should be OK, because that vma shouldn't be | |
1283 | * 'LOCKED. | |
1284 | */ | |
14fa31b8 | 1285 | static int try_to_unmap_anon(struct page *page, enum ttu_flags flags) |
1da177e4 LT |
1286 | { |
1287 | struct anon_vma *anon_vma; | |
5beb4930 | 1288 | struct anon_vma_chain *avc; |
1da177e4 | 1289 | int ret = SWAP_AGAIN; |
b291f000 | 1290 | |
1da177e4 LT |
1291 | anon_vma = page_lock_anon_vma(page); |
1292 | if (!anon_vma) | |
1293 | return ret; | |
1294 | ||
5beb4930 RR |
1295 | list_for_each_entry(avc, &anon_vma->head, same_anon_vma) { |
1296 | struct vm_area_struct *vma = avc->vma; | |
a8bef8ff MG |
1297 | unsigned long address; |
1298 | ||
1299 | /* | |
1300 | * During exec, a temporary VMA is setup and later moved. | |
1301 | * The VMA is moved under the anon_vma lock but not the | |
1302 | * page tables leading to a race where migration cannot | |
1303 | * find the migration ptes. Rather than increasing the | |
1304 | * locking requirements of exec(), migration skips | |
1305 | * temporary VMAs until after exec() completes. | |
1306 | */ | |
1307 | if (PAGE_MIGRATION && (flags & TTU_MIGRATION) && | |
1308 | is_vma_temporary_stack(vma)) | |
1309 | continue; | |
1310 | ||
1311 | address = vma_address(page, vma); | |
1cb1729b HD |
1312 | if (address == -EFAULT) |
1313 | continue; | |
1314 | ret = try_to_unmap_one(page, vma, address, flags); | |
53f79acb HD |
1315 | if (ret != SWAP_AGAIN || !page_mapped(page)) |
1316 | break; | |
1da177e4 | 1317 | } |
34bbd704 ON |
1318 | |
1319 | page_unlock_anon_vma(anon_vma); | |
1da177e4 LT |
1320 | return ret; |
1321 | } | |
1322 | ||
1323 | /** | |
b291f000 NP |
1324 | * try_to_unmap_file - unmap/unlock file page using the object-based rmap method |
1325 | * @page: the page to unmap/unlock | |
14fa31b8 | 1326 | * @flags: action and flags |
1da177e4 LT |
1327 | * |
1328 | * Find all the mappings of a page using the mapping pointer and the vma chains | |
1329 | * contained in the address_space struct it points to. | |
1330 | * | |
b291f000 NP |
1331 | * This function is only called from try_to_unmap/try_to_munlock for |
1332 | * object-based pages. | |
1333 | * When called from try_to_munlock(), the mmap_sem of the mm containing the vma | |
1334 | * where the page was found will be held for write. So, we won't recheck | |
1335 | * vm_flags for that VMA. That should be OK, because that vma shouldn't be | |
1336 | * 'LOCKED. | |
1da177e4 | 1337 | */ |
14fa31b8 | 1338 | static int try_to_unmap_file(struct page *page, enum ttu_flags flags) |
1da177e4 LT |
1339 | { |
1340 | struct address_space *mapping = page->mapping; | |
1341 | pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT); | |
1342 | struct vm_area_struct *vma; | |
1343 | struct prio_tree_iter iter; | |
1344 | int ret = SWAP_AGAIN; | |
1345 | unsigned long cursor; | |
1346 | unsigned long max_nl_cursor = 0; | |
1347 | unsigned long max_nl_size = 0; | |
1348 | unsigned int mapcount; | |
1349 | ||
1350 | spin_lock(&mapping->i_mmap_lock); | |
1351 | vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, pgoff, pgoff) { | |
1cb1729b HD |
1352 | unsigned long address = vma_address(page, vma); |
1353 | if (address == -EFAULT) | |
1354 | continue; | |
1355 | ret = try_to_unmap_one(page, vma, address, flags); | |
53f79acb HD |
1356 | if (ret != SWAP_AGAIN || !page_mapped(page)) |
1357 | goto out; | |
1da177e4 LT |
1358 | } |
1359 | ||
1360 | if (list_empty(&mapping->i_mmap_nonlinear)) | |
1361 | goto out; | |
1362 | ||
53f79acb HD |
1363 | /* |
1364 | * We don't bother to try to find the munlocked page in nonlinears. | |
1365 | * It's costly. Instead, later, page reclaim logic may call | |
1366 | * try_to_unmap(TTU_MUNLOCK) and recover PG_mlocked lazily. | |
1367 | */ | |
1368 | if (TTU_ACTION(flags) == TTU_MUNLOCK) | |
1369 | goto out; | |
1370 | ||
1da177e4 LT |
1371 | list_for_each_entry(vma, &mapping->i_mmap_nonlinear, |
1372 | shared.vm_set.list) { | |
1da177e4 LT |
1373 | cursor = (unsigned long) vma->vm_private_data; |
1374 | if (cursor > max_nl_cursor) | |
1375 | max_nl_cursor = cursor; | |
1376 | cursor = vma->vm_end - vma->vm_start; | |
1377 | if (cursor > max_nl_size) | |
1378 | max_nl_size = cursor; | |
1379 | } | |
1380 | ||
b291f000 | 1381 | if (max_nl_size == 0) { /* all nonlinears locked or reserved ? */ |
1da177e4 LT |
1382 | ret = SWAP_FAIL; |
1383 | goto out; | |
1384 | } | |
1385 | ||
1386 | /* | |
1387 | * We don't try to search for this page in the nonlinear vmas, | |
1388 | * and page_referenced wouldn't have found it anyway. Instead | |
1389 | * just walk the nonlinear vmas trying to age and unmap some. | |
1390 | * The mapcount of the page we came in with is irrelevant, | |
1391 | * but even so use it as a guide to how hard we should try? | |
1392 | */ | |
1393 | mapcount = page_mapcount(page); | |
1394 | if (!mapcount) | |
1395 | goto out; | |
1396 | cond_resched_lock(&mapping->i_mmap_lock); | |
1397 | ||
1398 | max_nl_size = (max_nl_size + CLUSTER_SIZE - 1) & CLUSTER_MASK; | |
1399 | if (max_nl_cursor == 0) | |
1400 | max_nl_cursor = CLUSTER_SIZE; | |
1401 | ||
1402 | do { | |
1403 | list_for_each_entry(vma, &mapping->i_mmap_nonlinear, | |
1404 | shared.vm_set.list) { | |
1da177e4 | 1405 | cursor = (unsigned long) vma->vm_private_data; |
839b9685 | 1406 | while ( cursor < max_nl_cursor && |
1da177e4 | 1407 | cursor < vma->vm_end - vma->vm_start) { |
53f79acb HD |
1408 | if (try_to_unmap_cluster(cursor, &mapcount, |
1409 | vma, page) == SWAP_MLOCK) | |
1410 | ret = SWAP_MLOCK; | |
1da177e4 LT |
1411 | cursor += CLUSTER_SIZE; |
1412 | vma->vm_private_data = (void *) cursor; | |
1413 | if ((int)mapcount <= 0) | |
1414 | goto out; | |
1415 | } | |
1416 | vma->vm_private_data = (void *) max_nl_cursor; | |
1417 | } | |
1418 | cond_resched_lock(&mapping->i_mmap_lock); | |
1419 | max_nl_cursor += CLUSTER_SIZE; | |
1420 | } while (max_nl_cursor <= max_nl_size); | |
1421 | ||
1422 | /* | |
1423 | * Don't loop forever (perhaps all the remaining pages are | |
1424 | * in locked vmas). Reset cursor on all unreserved nonlinear | |
1425 | * vmas, now forgetting on which ones it had fallen behind. | |
1426 | */ | |
101d2be7 HD |
1427 | list_for_each_entry(vma, &mapping->i_mmap_nonlinear, shared.vm_set.list) |
1428 | vma->vm_private_data = NULL; | |
1da177e4 LT |
1429 | out: |
1430 | spin_unlock(&mapping->i_mmap_lock); | |
1431 | return ret; | |
1432 | } | |
1433 | ||
1434 | /** | |
1435 | * try_to_unmap - try to remove all page table mappings to a page | |
1436 | * @page: the page to get unmapped | |
14fa31b8 | 1437 | * @flags: action and flags |
1da177e4 LT |
1438 | * |
1439 | * Tries to remove all the page table entries which are mapping this | |
1440 | * page, used in the pageout path. Caller must hold the page lock. | |
1441 | * Return values are: | |
1442 | * | |
1443 | * SWAP_SUCCESS - we succeeded in removing all mappings | |
1444 | * SWAP_AGAIN - we missed a mapping, try again later | |
1445 | * SWAP_FAIL - the page is unswappable | |
b291f000 | 1446 | * SWAP_MLOCK - page is mlocked. |
1da177e4 | 1447 | */ |
14fa31b8 | 1448 | int try_to_unmap(struct page *page, enum ttu_flags flags) |
1da177e4 LT |
1449 | { |
1450 | int ret; | |
1451 | ||
1da177e4 | 1452 | BUG_ON(!PageLocked(page)); |
91600e9e | 1453 | VM_BUG_ON(!PageHuge(page) && PageTransHuge(page)); |
1da177e4 | 1454 | |
5ad64688 HD |
1455 | if (unlikely(PageKsm(page))) |
1456 | ret = try_to_unmap_ksm(page, flags); | |
1457 | else if (PageAnon(page)) | |
14fa31b8 | 1458 | ret = try_to_unmap_anon(page, flags); |
1da177e4 | 1459 | else |
14fa31b8 | 1460 | ret = try_to_unmap_file(page, flags); |
b291f000 | 1461 | if (ret != SWAP_MLOCK && !page_mapped(page)) |
1da177e4 LT |
1462 | ret = SWAP_SUCCESS; |
1463 | return ret; | |
1464 | } | |
81b4082d | 1465 | |
b291f000 NP |
1466 | /** |
1467 | * try_to_munlock - try to munlock a page | |
1468 | * @page: the page to be munlocked | |
1469 | * | |
1470 | * Called from munlock code. Checks all of the VMAs mapping the page | |
1471 | * to make sure nobody else has this page mlocked. The page will be | |
1472 | * returned with PG_mlocked cleared if no other vmas have it mlocked. | |
1473 | * | |
1474 | * Return values are: | |
1475 | * | |
53f79acb | 1476 | * SWAP_AGAIN - no vma is holding page mlocked, or, |
b291f000 | 1477 | * SWAP_AGAIN - page mapped in mlocked vma -- couldn't acquire mmap sem |
5ad64688 | 1478 | * SWAP_FAIL - page cannot be located at present |
b291f000 NP |
1479 | * SWAP_MLOCK - page is now mlocked. |
1480 | */ | |
1481 | int try_to_munlock(struct page *page) | |
1482 | { | |
1483 | VM_BUG_ON(!PageLocked(page) || PageLRU(page)); | |
1484 | ||
5ad64688 HD |
1485 | if (unlikely(PageKsm(page))) |
1486 | return try_to_unmap_ksm(page, TTU_MUNLOCK); | |
1487 | else if (PageAnon(page)) | |
14fa31b8 | 1488 | return try_to_unmap_anon(page, TTU_MUNLOCK); |
b291f000 | 1489 | else |
14fa31b8 | 1490 | return try_to_unmap_file(page, TTU_MUNLOCK); |
b291f000 | 1491 | } |
e9995ef9 | 1492 | |
01d8b20d | 1493 | void __put_anon_vma(struct anon_vma *anon_vma) |
76545066 | 1494 | { |
01d8b20d | 1495 | struct anon_vma *root = anon_vma->root; |
76545066 | 1496 | |
01d8b20d PZ |
1497 | if (root != anon_vma && atomic_dec_and_test(&root->refcount)) |
1498 | anon_vma_free(root); | |
76545066 | 1499 | |
01d8b20d | 1500 | anon_vma_free(anon_vma); |
76545066 | 1501 | } |
76545066 | 1502 | |
e9995ef9 HD |
1503 | #ifdef CONFIG_MIGRATION |
1504 | /* | |
1505 | * rmap_walk() and its helpers rmap_walk_anon() and rmap_walk_file(): | |
1506 | * Called by migrate.c to remove migration ptes, but might be used more later. | |
1507 | */ | |
1508 | static int rmap_walk_anon(struct page *page, int (*rmap_one)(struct page *, | |
1509 | struct vm_area_struct *, unsigned long, void *), void *arg) | |
1510 | { | |
1511 | struct anon_vma *anon_vma; | |
5beb4930 | 1512 | struct anon_vma_chain *avc; |
e9995ef9 HD |
1513 | int ret = SWAP_AGAIN; |
1514 | ||
1515 | /* | |
1516 | * Note: remove_migration_ptes() cannot use page_lock_anon_vma() | |
1517 | * because that depends on page_mapped(); but not all its usages | |
3f6c8272 MG |
1518 | * are holding mmap_sem. Users without mmap_sem are required to |
1519 | * take a reference count to prevent the anon_vma disappearing | |
e9995ef9 HD |
1520 | */ |
1521 | anon_vma = page_anon_vma(page); | |
1522 | if (!anon_vma) | |
1523 | return ret; | |
cba48b98 | 1524 | anon_vma_lock(anon_vma); |
5beb4930 RR |
1525 | list_for_each_entry(avc, &anon_vma->head, same_anon_vma) { |
1526 | struct vm_area_struct *vma = avc->vma; | |
e9995ef9 HD |
1527 | unsigned long address = vma_address(page, vma); |
1528 | if (address == -EFAULT) | |
1529 | continue; | |
1530 | ret = rmap_one(page, vma, address, arg); | |
1531 | if (ret != SWAP_AGAIN) | |
1532 | break; | |
1533 | } | |
cba48b98 | 1534 | anon_vma_unlock(anon_vma); |
e9995ef9 HD |
1535 | return ret; |
1536 | } | |
1537 | ||
1538 | static int rmap_walk_file(struct page *page, int (*rmap_one)(struct page *, | |
1539 | struct vm_area_struct *, unsigned long, void *), void *arg) | |
1540 | { | |
1541 | struct address_space *mapping = page->mapping; | |
1542 | pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT); | |
1543 | struct vm_area_struct *vma; | |
1544 | struct prio_tree_iter iter; | |
1545 | int ret = SWAP_AGAIN; | |
1546 | ||
1547 | if (!mapping) | |
1548 | return ret; | |
1549 | spin_lock(&mapping->i_mmap_lock); | |
1550 | vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, pgoff, pgoff) { | |
1551 | unsigned long address = vma_address(page, vma); | |
1552 | if (address == -EFAULT) | |
1553 | continue; | |
1554 | ret = rmap_one(page, vma, address, arg); | |
1555 | if (ret != SWAP_AGAIN) | |
1556 | break; | |
1557 | } | |
1558 | /* | |
1559 | * No nonlinear handling: being always shared, nonlinear vmas | |
1560 | * never contain migration ptes. Decide what to do about this | |
1561 | * limitation to linear when we need rmap_walk() on nonlinear. | |
1562 | */ | |
1563 | spin_unlock(&mapping->i_mmap_lock); | |
1564 | return ret; | |
1565 | } | |
1566 | ||
1567 | int rmap_walk(struct page *page, int (*rmap_one)(struct page *, | |
1568 | struct vm_area_struct *, unsigned long, void *), void *arg) | |
1569 | { | |
1570 | VM_BUG_ON(!PageLocked(page)); | |
1571 | ||
1572 | if (unlikely(PageKsm(page))) | |
1573 | return rmap_walk_ksm(page, rmap_one, arg); | |
1574 | else if (PageAnon(page)) | |
1575 | return rmap_walk_anon(page, rmap_one, arg); | |
1576 | else | |
1577 | return rmap_walk_file(page, rmap_one, arg); | |
1578 | } | |
1579 | #endif /* CONFIG_MIGRATION */ | |
0fe6e20b | 1580 | |
e3390f67 | 1581 | #ifdef CONFIG_HUGETLB_PAGE |
0fe6e20b NH |
1582 | /* |
1583 | * The following three functions are for anonymous (private mapped) hugepages. | |
1584 | * Unlike common anonymous pages, anonymous hugepages have no accounting code | |
1585 | * and no lru code, because we handle hugepages differently from common pages. | |
1586 | */ | |
1587 | static void __hugepage_set_anon_rmap(struct page *page, | |
1588 | struct vm_area_struct *vma, unsigned long address, int exclusive) | |
1589 | { | |
1590 | struct anon_vma *anon_vma = vma->anon_vma; | |
433abed6 | 1591 | |
0fe6e20b | 1592 | BUG_ON(!anon_vma); |
433abed6 NH |
1593 | |
1594 | if (PageAnon(page)) | |
1595 | return; | |
1596 | if (!exclusive) | |
1597 | anon_vma = anon_vma->root; | |
1598 | ||
0fe6e20b NH |
1599 | anon_vma = (void *) anon_vma + PAGE_MAPPING_ANON; |
1600 | page->mapping = (struct address_space *) anon_vma; | |
1601 | page->index = linear_page_index(vma, address); | |
1602 | } | |
1603 | ||
1604 | void hugepage_add_anon_rmap(struct page *page, | |
1605 | struct vm_area_struct *vma, unsigned long address) | |
1606 | { | |
1607 | struct anon_vma *anon_vma = vma->anon_vma; | |
1608 | int first; | |
a850ea30 NH |
1609 | |
1610 | BUG_ON(!PageLocked(page)); | |
0fe6e20b NH |
1611 | BUG_ON(!anon_vma); |
1612 | BUG_ON(address < vma->vm_start || address >= vma->vm_end); | |
1613 | first = atomic_inc_and_test(&page->_mapcount); | |
1614 | if (first) | |
1615 | __hugepage_set_anon_rmap(page, vma, address, 0); | |
1616 | } | |
1617 | ||
1618 | void hugepage_add_new_anon_rmap(struct page *page, | |
1619 | struct vm_area_struct *vma, unsigned long address) | |
1620 | { | |
1621 | BUG_ON(address < vma->vm_start || address >= vma->vm_end); | |
1622 | atomic_set(&page->_mapcount, 0); | |
1623 | __hugepage_set_anon_rmap(page, vma, address, 1); | |
1624 | } | |
e3390f67 | 1625 | #endif /* CONFIG_HUGETLB_PAGE */ |